Rev 5354 | Rev 6660 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 5354 | Rev 6084 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright © 2012 Intel Corporation |
2 | * Copyright © 2012 Intel Corporation |
3 | * |
3 | * |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * copy of this software and associated documentation files (the "Software"), |
5 | * copy of this software and associated documentation files (the "Software"), |
6 | * to deal in the Software without restriction, including without limitation |
6 | * to deal in the Software without restriction, including without limitation |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
8 | * and/or sell copies of the Software, and to permit persons to whom the |
8 | * and/or sell copies of the Software, and to permit persons to whom the |
9 | * Software is furnished to do so, subject to the following conditions: |
9 | * Software is furnished to do so, subject to the following conditions: |
10 | * |
10 | * |
11 | * The above copyright notice and this permission notice (including the next |
11 | * The above copyright notice and this permission notice (including the next |
12 | * paragraph) shall be included in all copies or substantial portions of the |
12 | * paragraph) shall be included in all copies or substantial portions of the |
13 | * Software. |
13 | * Software. |
14 | * |
14 | * |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
21 | * IN THE SOFTWARE. |
21 | * IN THE SOFTWARE. |
22 | * |
22 | * |
23 | * Authors: |
23 | * Authors: |
24 | * Eugeni Dodonov |
24 | * Eugeni Dodonov |
25 | * |
25 | * |
26 | */ |
26 | */ |
27 | 27 | ||
28 | //#include |
28 | //#include |
29 | #include "i915_drv.h" |
29 | #include "i915_drv.h" |
30 | #include "intel_drv.h" |
30 | #include "intel_drv.h" |
31 | #include |
- | |
32 | //#include "../../../platform/x86/intel_ips.h" |
31 | //#include "../../../platform/x86/intel_ips.h" |
33 | #include |
32 | #include |
34 | - | ||
35 | #include |
33 | |
36 | 34 | ||
37 | #define FORCEWAKE_ACK_TIMEOUT_MS 2 |
35 | #define FORCEWAKE_ACK_TIMEOUT_MS 2 |
38 | 36 | ||
39 | void getrawmonotonic(struct timespec *ts); |
37 | void getrawmonotonic(struct timespec *ts); |
40 | - | ||
41 | union ktime { |
- | |
42 | s64 tv64; |
- | |
43 | }; |
- | |
44 | - | ||
45 | typedef union ktime ktime_t; /* Kill this */ |
- | |
46 | - | ||
47 | #define ktime_to_ns(kt) ((kt).tv64) |
- | |
48 | - | ||
49 | static inline u64 ktime_get_raw_ns(void) |
- | |
50 | { |
- | |
51 | return 0; //ktime_to_ns(ktime_get_raw()); |
- | |
52 | } |
38 | |
53 | /** |
39 | /** |
54 | * RC6 is a special power stage which allows the GPU to enter an very |
40 | * RC6 is a special power stage which allows the GPU to enter an very |
55 | * low-voltage mode when idle, using down to 0V while at this stage. This |
41 | * low-voltage mode when idle, using down to 0V while at this stage. This |
56 | * stage is entered automatically when the GPU is idle when RC6 support is |
42 | * stage is entered automatically when the GPU is idle when RC6 support is |
57 | * enabled, and as soon as new workload arises GPU wakes up automatically as well. |
43 | * enabled, and as soon as new workload arises GPU wakes up automatically as well. |
58 | * |
44 | * |
59 | * There are different RC6 modes available in Intel GPU, which differentiate |
45 | * There are different RC6 modes available in Intel GPU, which differentiate |
60 | * among each other with the latency required to enter and leave RC6 and |
46 | * among each other with the latency required to enter and leave RC6 and |
61 | * voltage consumed by the GPU in different states. |
47 | * voltage consumed by the GPU in different states. |
62 | * |
48 | * |
63 | * The combination of the following flags define which states GPU is allowed |
49 | * The combination of the following flags define which states GPU is allowed |
64 | * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and |
50 | * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and |
65 | * RC6pp is deepest RC6. Their support by hardware varies according to the |
51 | * RC6pp is deepest RC6. Their support by hardware varies according to the |
66 | * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one |
52 | * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one |
67 | * which brings the most power savings; deeper states save more power, but |
53 | * which brings the most power savings; deeper states save more power, but |
68 | * require higher latency to switch to and wake up. |
54 | * require higher latency to switch to and wake up. |
69 | */ |
55 | */ |
70 | #define INTEL_RC6_ENABLE (1<<0) |
56 | #define INTEL_RC6_ENABLE (1<<0) |
71 | #define INTEL_RC6p_ENABLE (1<<1) |
57 | #define INTEL_RC6p_ENABLE (1<<1) |
72 | #define INTEL_RC6pp_ENABLE (1<<2) |
58 | #define INTEL_RC6pp_ENABLE (1<<2) |
73 | - | ||
74 | /* FBC, or Frame Buffer Compression, is a technique employed to compress the |
- | |
75 | * framebuffer contents in-memory, aiming at reducing the required bandwidth |
- | |
76 | * during in-memory transfers and, therefore, reduce the power packet. |
- | |
77 | * |
- | |
78 | * The benefits of FBC are mostly visible with solid backgrounds and |
- | |
79 | * variation-less patterns. |
- | |
80 | * |
- | |
81 | * FBC-related functionality can be enabled by the means of the |
- | |
82 | * i915.i915_enable_fbc parameter |
- | |
83 | */ |
- | |
84 | 59 | ||
85 | static void gen9_init_clock_gating(struct drm_device *dev) |
60 | static void bxt_init_clock_gating(struct drm_device *dev) |
86 | { |
61 | { |
87 | struct drm_i915_private *dev_priv = dev->dev_private; |
62 | struct drm_i915_private *dev_priv = dev->dev_private; |
88 | - | ||
89 | /* |
63 | |
90 | * WaDisableSDEUnitClockGating:skl |
- | |
91 | * This seems to be a pre-production w/a. |
- | |
92 | */ |
64 | /* WaDisableSDEUnitClockGating:bxt */ |
93 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
65 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
94 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
66 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
95 | 67 | ||
96 | /* |
68 | /* |
97 | * WaDisableDgMirrorFixInHalfSliceChicken5:skl |
69 | * FIXME: |
98 | * This is a pre-production w/a. |
70 | * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only. |
99 | */ |
71 | */ |
100 | I915_WRITE(GEN9_HALF_SLICE_CHICKEN5, |
72 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
101 | I915_READ(GEN9_HALF_SLICE_CHICKEN5) & |
- | |
102 | ~GEN9_DG_MIRROR_FIX_ENABLE); |
73 | GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ); |
103 | - | ||
104 | /* Wa4x4STCOptimizationDisable:skl */ |
- | |
105 | I915_WRITE(CACHE_MODE_1, |
- | |
106 | _MASKED_BIT_ENABLE(GEN8_4x4_STC_OPTIMIZATION_DISABLE)); |
- | |
107 | } |
- | |
108 | - | ||
109 | static void i8xx_disable_fbc(struct drm_device *dev) |
- | |
110 | { |
- | |
111 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
112 | u32 fbc_ctl; |
- | |
113 | - | ||
114 | dev_priv->fbc.enabled = false; |
- | |
115 | - | ||
116 | /* Disable compression */ |
- | |
117 | fbc_ctl = I915_READ(FBC_CONTROL); |
- | |
118 | if ((fbc_ctl & FBC_CTL_EN) == 0) |
- | |
119 | return; |
- | |
120 | - | ||
121 | fbc_ctl &= ~FBC_CTL_EN; |
- | |
122 | I915_WRITE(FBC_CONTROL, fbc_ctl); |
- | |
123 | - | ||
124 | /* Wait for compressing bit to clear */ |
- | |
125 | if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) { |
- | |
126 | DRM_DEBUG_KMS("FBC idle timed out\n"); |
- | |
127 | return; |
- | |
128 | } |
- | |
129 | - | ||
130 | DRM_DEBUG_KMS("disabled FBC\n"); |
- | |
131 | } |
- | |
132 | - | ||
133 | static void i8xx_enable_fbc(struct drm_crtc *crtc) |
- | |
134 | { |
- | |
135 | struct drm_device *dev = crtc->dev; |
- | |
136 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
137 | struct drm_framebuffer *fb = crtc->primary->fb; |
- | |
138 | struct drm_i915_gem_object *obj = intel_fb_obj(fb); |
- | |
139 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | |
140 | int cfb_pitch; |
- | |
141 | int i; |
- | |
142 | u32 fbc_ctl; |
- | |
143 | - | ||
144 | dev_priv->fbc.enabled = true; |
- | |
145 | - | ||
146 | cfb_pitch = dev_priv->fbc.size / FBC_LL_SIZE; |
- | |
147 | if (fb->pitches[0] < cfb_pitch) |
- | |
148 | cfb_pitch = fb->pitches[0]; |
- | |
149 | - | ||
150 | /* FBC_CTL wants 32B or 64B units */ |
- | |
151 | if (IS_GEN2(dev)) |
- | |
152 | cfb_pitch = (cfb_pitch / 32) - 1; |
- | |
153 | else |
- | |
154 | cfb_pitch = (cfb_pitch / 64) - 1; |
- | |
155 | - | ||
156 | /* Clear old tags */ |
- | |
157 | for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++) |
- | |
158 | I915_WRITE(FBC_TAG + (i * 4), 0); |
- | |
159 | - | ||
160 | if (IS_GEN4(dev)) { |
- | |
161 | u32 fbc_ctl2; |
- | |
162 | - | ||
163 | /* Set it up... */ |
- | |
164 | fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE; |
- | |
165 | fbc_ctl2 |= FBC_CTL_PLANE(intel_crtc->plane); |
- | |
166 | I915_WRITE(FBC_CONTROL2, fbc_ctl2); |
- | |
167 | I915_WRITE(FBC_FENCE_OFF, crtc->y); |
- | |
168 | } |
- | |
169 | - | ||
170 | /* enable it... */ |
- | |
171 | fbc_ctl = I915_READ(FBC_CONTROL); |
- | |
172 | fbc_ctl &= 0x3fff << FBC_CTL_INTERVAL_SHIFT; |
- | |
173 | fbc_ctl |= FBC_CTL_EN | FBC_CTL_PERIODIC; |
- | |
174 | if (IS_I945GM(dev)) |
- | |
175 | fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */ |
- | |
176 | fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT; |
- | |
177 | fbc_ctl |= obj->fence_reg; |
- | |
178 | I915_WRITE(FBC_CONTROL, fbc_ctl); |
- | |
179 | - | ||
180 | DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c\n", |
- | |
181 | cfb_pitch, crtc->y, plane_name(intel_crtc->plane)); |
- | |
182 | } |
- | |
183 | - | ||
184 | static bool i8xx_fbc_enabled(struct drm_device *dev) |
- | |
185 | { |
- | |
186 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
187 | - | ||
188 | return I915_READ(FBC_CONTROL) & FBC_CTL_EN; |
- | |
189 | } |
- | |
190 | - | ||
191 | static void g4x_enable_fbc(struct drm_crtc *crtc) |
- | |
192 | { |
- | |
193 | struct drm_device *dev = crtc->dev; |
- | |
194 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
195 | struct drm_framebuffer *fb = crtc->primary->fb; |
- | |
196 | struct drm_i915_gem_object *obj = intel_fb_obj(fb); |
- | |
197 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | |
198 | u32 dpfc_ctl; |
- | |
199 | - | ||
200 | dev_priv->fbc.enabled = true; |
- | |
201 | - | ||
202 | dpfc_ctl = DPFC_CTL_PLANE(intel_crtc->plane) | DPFC_SR_EN; |
- | |
203 | if (drm_format_plane_cpp(fb->pixel_format, 0) == 2) |
- | |
204 | dpfc_ctl |= DPFC_CTL_LIMIT_2X; |
- | |
205 | else |
- | |
206 | dpfc_ctl |= DPFC_CTL_LIMIT_1X; |
- | |
207 | dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg; |
- | |
208 | - | ||
209 | I915_WRITE(DPFC_FENCE_YOFF, crtc->y); |
- | |
210 | - | ||
211 | /* enable it... */ |
- | |
212 | I915_WRITE(DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN); |
- | |
213 | - | ||
214 | DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane)); |
- | |
215 | } |
- | |
216 | - | ||
217 | static void g4x_disable_fbc(struct drm_device *dev) |
- | |
218 | { |
- | |
219 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
220 | u32 dpfc_ctl; |
- | |
221 | - | ||
222 | dev_priv->fbc.enabled = false; |
- | |
223 | - | ||
224 | /* Disable compression */ |
- | |
225 | dpfc_ctl = I915_READ(DPFC_CONTROL); |
- | |
226 | if (dpfc_ctl & DPFC_CTL_EN) { |
- | |
227 | dpfc_ctl &= ~DPFC_CTL_EN; |
- | |
228 | I915_WRITE(DPFC_CONTROL, dpfc_ctl); |
- | |
229 | - | ||
230 | DRM_DEBUG_KMS("disabled FBC\n"); |
- | |
231 | } |
- | |
232 | } |
- | |
233 | - | ||
234 | static bool g4x_fbc_enabled(struct drm_device *dev) |
- | |
235 | { |
- | |
236 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
237 | - | ||
238 | return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN; |
- | |
239 | } |
- | |
240 | - | ||
241 | static void sandybridge_blit_fbc_update(struct drm_device *dev) |
- | |
242 | { |
- | |
243 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
244 | u32 blt_ecoskpd; |
- | |
245 | - | ||
246 | /* Make sure blitter notifies FBC of writes */ |
- | |
247 | - | ||
248 | /* Blitter is part of Media powerwell on VLV. No impact of |
- | |
249 | * his param in other platforms for now */ |
- | |
250 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_MEDIA); |
- | |
251 | - | ||
252 | blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD); |
- | |
253 | blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY << |
- | |
254 | GEN6_BLITTER_LOCK_SHIFT; |
- | |
255 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); |
- | |
256 | blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY; |
- | |
257 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); |
- | |
258 | blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY << |
- | |
259 | GEN6_BLITTER_LOCK_SHIFT); |
- | |
260 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); |
- | |
261 | POSTING_READ(GEN6_BLITTER_ECOSKPD); |
- | |
262 | - | ||
263 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_MEDIA); |
- | |
264 | } |
- | |
265 | - | ||
266 | static void ironlake_enable_fbc(struct drm_crtc *crtc) |
- | |
267 | { |
- | |
268 | struct drm_device *dev = crtc->dev; |
- | |
269 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
270 | struct drm_framebuffer *fb = crtc->primary->fb; |
- | |
271 | struct drm_i915_gem_object *obj = intel_fb_obj(fb); |
- | |
272 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | |
273 | u32 dpfc_ctl; |
- | |
274 | - | ||
275 | dev_priv->fbc.enabled = true; |
- | |
276 | - | ||
277 | dpfc_ctl = DPFC_CTL_PLANE(intel_crtc->plane); |
- | |
278 | if (drm_format_plane_cpp(fb->pixel_format, 0) == 2) |
- | |
279 | dev_priv->fbc.threshold++; |
- | |
280 | - | ||
281 | switch (dev_priv->fbc.threshold) { |
- | |
282 | case 4: |
- | |
283 | case 3: |
- | |
284 | dpfc_ctl |= DPFC_CTL_LIMIT_4X; |
- | |
285 | break; |
- | |
286 | case 2: |
- | |
287 | dpfc_ctl |= DPFC_CTL_LIMIT_2X; |
- | |
288 | break; |
- | |
289 | case 1: |
- | |
290 | dpfc_ctl |= DPFC_CTL_LIMIT_1X; |
- | |
291 | break; |
- | |
292 | } |
- | |
293 | dpfc_ctl |= DPFC_CTL_FENCE_EN; |
- | |
294 | if (IS_GEN5(dev)) |
- | |
295 | dpfc_ctl |= obj->fence_reg; |
- | |
296 | - | ||
297 | I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y); |
- | |
298 | I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID); |
- | |
299 | /* enable it... */ |
- | |
300 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN); |
- | |
301 | - | ||
302 | if (IS_GEN6(dev)) { |
- | |
303 | I915_WRITE(SNB_DPFC_CTL_SA, |
- | |
304 | SNB_CPU_FENCE_ENABLE | obj->fence_reg); |
- | |
305 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y); |
- | |
306 | sandybridge_blit_fbc_update(dev); |
- | |
307 | } |
- | |
308 | - | ||
309 | DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane)); |
- | |
310 | } |
- | |
311 | - | ||
312 | static void ironlake_disable_fbc(struct drm_device *dev) |
- | |
313 | { |
- | |
314 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
315 | u32 dpfc_ctl; |
- | |
316 | - | ||
317 | dev_priv->fbc.enabled = false; |
- | |
318 | - | ||
319 | /* Disable compression */ |
- | |
320 | dpfc_ctl = I915_READ(ILK_DPFC_CONTROL); |
- | |
321 | if (dpfc_ctl & DPFC_CTL_EN) { |
- | |
322 | dpfc_ctl &= ~DPFC_CTL_EN; |
- | |
323 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl); |
- | |
324 | - | ||
325 | DRM_DEBUG_KMS("disabled FBC\n"); |
- | |
326 | } |
- | |
327 | } |
- | |
328 | - | ||
329 | static bool ironlake_fbc_enabled(struct drm_device *dev) |
- | |
330 | { |
- | |
331 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
332 | - | ||
333 | return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN; |
- | |
334 | } |
- | |
335 | - | ||
336 | static void gen7_enable_fbc(struct drm_crtc *crtc) |
- | |
337 | { |
- | |
338 | struct drm_device *dev = crtc->dev; |
- | |
339 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
340 | struct drm_framebuffer *fb = crtc->primary->fb; |
- | |
341 | struct drm_i915_gem_object *obj = intel_fb_obj(fb); |
- | |
342 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | |
343 | u32 dpfc_ctl; |
- | |
344 | - | ||
345 | dev_priv->fbc.enabled = true; |
- | |
346 | - | ||
347 | dpfc_ctl = IVB_DPFC_CTL_PLANE(intel_crtc->plane); |
- | |
348 | if (drm_format_plane_cpp(fb->pixel_format, 0) == 2) |
- | |
349 | dev_priv->fbc.threshold++; |
- | |
350 | - | ||
351 | switch (dev_priv->fbc.threshold) { |
- | |
352 | case 4: |
- | |
353 | case 3: |
- | |
354 | dpfc_ctl |= DPFC_CTL_LIMIT_4X; |
- | |
355 | break; |
- | |
356 | case 2: |
- | |
357 | dpfc_ctl |= DPFC_CTL_LIMIT_2X; |
- | |
358 | break; |
- | |
359 | case 1: |
- | |
360 | dpfc_ctl |= DPFC_CTL_LIMIT_1X; |
- | |
361 | break; |
- | |
362 | } |
- | |
363 | - | ||
364 | dpfc_ctl |= IVB_DPFC_CTL_FENCE_EN; |
- | |
365 | - | ||
366 | if (dev_priv->fbc.false_color) |
- | |
367 | dpfc_ctl |= FBC_CTL_FALSE_COLOR; |
- | |
368 | - | ||
369 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN); |
- | |
370 | - | ||
371 | if (IS_IVYBRIDGE(dev)) { |
- | |
372 | /* WaFbcAsynchFlipDisableFbcQueue:ivb */ |
- | |
373 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
- | |
374 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
- | |
375 | ILK_FBCQ_DIS); |
- | |
376 | } else { |
- | |
377 | /* WaFbcAsynchFlipDisableFbcQueue:hsw,bdw */ |
- | |
378 | I915_WRITE(CHICKEN_PIPESL_1(intel_crtc->pipe), |
- | |
379 | I915_READ(CHICKEN_PIPESL_1(intel_crtc->pipe)) | |
- | |
380 | HSW_FBCQ_DIS); |
- | |
381 | } |
- | |
382 | - | ||
383 | I915_WRITE(SNB_DPFC_CTL_SA, |
- | |
384 | SNB_CPU_FENCE_ENABLE | obj->fence_reg); |
- | |
385 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y); |
- | |
386 | - | ||
387 | sandybridge_blit_fbc_update(dev); |
- | |
388 | - | ||
389 | DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane)); |
- | |
390 | } |
- | |
391 | - | ||
392 | bool intel_fbc_enabled(struct drm_device *dev) |
- | |
393 | { |
- | |
394 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
395 | - | ||
396 | return dev_priv->fbc.enabled; |
- | |
397 | } |
- | |
398 | - | ||
399 | void bdw_fbc_sw_flush(struct drm_device *dev, u32 value) |
- | |
400 | { |
- | |
401 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
402 | - | ||
403 | if (!IS_GEN8(dev)) |
- | |
404 | return; |
- | |
405 | - | ||
406 | if (!intel_fbc_enabled(dev)) |
- | |
407 | return; |
- | |
408 | - | ||
409 | I915_WRITE(MSG_FBC_REND_STATE, value); |
- | |
410 | } |
- | |
411 | - | ||
412 | static void intel_fbc_work_fn(struct work_struct *__work) |
- | |
413 | { |
- | |
414 | struct intel_fbc_work *work = |
- | |
415 | container_of(to_delayed_work(__work), |
- | |
416 | struct intel_fbc_work, work); |
- | |
417 | struct drm_device *dev = work->crtc->dev; |
- | |
418 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
419 | - | ||
420 | mutex_lock(&dev->struct_mutex); |
- | |
421 | if (work == dev_priv->fbc.fbc_work) { |
- | |
422 | /* Double check that we haven't switched fb without cancelling |
- | |
423 | * the prior work. |
- | |
424 | */ |
- | |
425 | if (work->crtc->primary->fb == work->fb) { |
- | |
426 | dev_priv->display.enable_fbc(work->crtc); |
- | |
427 | - | ||
428 | dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane; |
- | |
429 | dev_priv->fbc.fb_id = work->crtc->primary->fb->base.id; |
- | |
430 | dev_priv->fbc.y = work->crtc->y; |
- | |
431 | } |
- | |
432 | - | ||
433 | dev_priv->fbc.fbc_work = NULL; |
- | |
434 | } |
- | |
435 | mutex_unlock(&dev->struct_mutex); |
- | |
436 | - | ||
437 | kfree(work); |
- | |
438 | } |
- | |
439 | - | ||
440 | static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv) |
- | |
441 | { |
- | |
442 | if (dev_priv->fbc.fbc_work == NULL) |
- | |
443 | return; |
- | |
444 | - | ||
445 | DRM_DEBUG_KMS("cancelling pending FBC enable\n"); |
- | |
446 | - | ||
447 | /* Synchronisation is provided by struct_mutex and checking of |
- | |
448 | * dev_priv->fbc.fbc_work, so we can perform the cancellation |
- | |
449 | * entirely asynchronously. |
- | |
450 | */ |
- | |
451 | if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work)) |
- | |
452 | /* tasklet was killed before being run, clean up */ |
- | |
453 | kfree(dev_priv->fbc.fbc_work); |
- | |
454 | - | ||
455 | /* Mark the work as no longer wanted so that if it does |
- | |
456 | * wake-up (because the work was already running and waiting |
- | |
457 | * for our mutex), it will discover that is no longer |
- | |
458 | * necessary to run. |
- | |
459 | */ |
- | |
460 | dev_priv->fbc.fbc_work = NULL; |
- | |
461 | } |
- | |
462 | - | ||
463 | static void intel_enable_fbc(struct drm_crtc *crtc) |
- | |
464 | { |
- | |
465 | struct intel_fbc_work *work; |
- | |
466 | struct drm_device *dev = crtc->dev; |
- | |
467 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
468 | - | ||
469 | if (!dev_priv->display.enable_fbc) |
- | |
470 | return; |
- | |
471 | - | ||
472 | intel_cancel_fbc_work(dev_priv); |
- | |
473 | - | ||
474 | work = kzalloc(sizeof(*work), GFP_KERNEL); |
- | |
475 | if (work == NULL) { |
- | |
476 | DRM_ERROR("Failed to allocate FBC work structure\n"); |
- | |
477 | dev_priv->display.enable_fbc(crtc); |
- | |
478 | return; |
- | |
479 | } |
- | |
480 | - | ||
481 | work->crtc = crtc; |
- | |
482 | work->fb = crtc->primary->fb; |
- | |
483 | INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn); |
- | |
484 | - | ||
485 | dev_priv->fbc.fbc_work = work; |
- | |
486 | - | ||
487 | /* Delay the actual enabling to let pageflipping cease and the |
- | |
488 | * display to settle before starting the compression. Note that |
- | |
489 | * this delay also serves a second purpose: it allows for a |
- | |
490 | * vblank to pass after disabling the FBC before we attempt |
- | |
491 | * to modify the control registers. |
- | |
492 | * |
- | |
493 | * A more complicated solution would involve tracking vblanks |
- | |
494 | * following the termination of the page-flipping sequence |
- | |
495 | * and indeed performing the enable as a co-routine and not |
- | |
496 | * waiting synchronously upon the vblank. |
- | |
497 | * |
- | |
498 | * WaFbcWaitForVBlankBeforeEnable:ilk,snb |
- | |
499 | */ |
- | |
500 | schedule_delayed_work(&work->work, msecs_to_jiffies(50)); |
- | |
501 | } |
- | |
502 | - | ||
503 | void intel_disable_fbc(struct drm_device *dev) |
- | |
504 | { |
- | |
505 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
506 | - | ||
507 | intel_cancel_fbc_work(dev_priv); |
- | |
508 | - | ||
509 | if (!dev_priv->display.disable_fbc) |
- | |
510 | return; |
- | |
511 | - | ||
512 | dev_priv->display.disable_fbc(dev); |
- | |
513 | dev_priv->fbc.plane = -1; |
- | |
514 | } |
- | |
515 | - | ||
516 | static bool set_no_fbc_reason(struct drm_i915_private *dev_priv, |
- | |
517 | enum no_fbc_reason reason) |
- | |
518 | { |
- | |
519 | if (dev_priv->fbc.no_fbc_reason == reason) |
- | |
520 | return false; |
- | |
521 | - | ||
522 | dev_priv->fbc.no_fbc_reason = reason; |
- | |
523 | return true; |
- | |
524 | } |
- | |
525 | - | ||
526 | /** |
- | |
527 | * intel_update_fbc - enable/disable FBC as needed |
- | |
528 | * @dev: the drm_device |
- | |
529 | * |
- | |
530 | * Set up the framebuffer compression hardware at mode set time. We |
- | |
531 | * enable it if possible: |
- | |
532 | * - plane A only (on pre-965) |
- | |
533 | * - no pixel mulitply/line duplication |
- | |
534 | * - no alpha buffer discard |
- | |
535 | * - no dual wide |
- | |
536 | * - framebuffer <= max_hdisplay in width, max_vdisplay in height |
- | |
537 | * |
- | |
538 | * We can't assume that any compression will take place (worst case), |
- | |
539 | * so the compressed buffer has to be the same size as the uncompressed |
- | |
540 | * one. It also must reside (along with the line length buffer) in |
- | |
541 | * stolen memory. |
- | |
542 | * |
- | |
543 | * We need to enable/disable FBC on a global basis. |
- | |
544 | */ |
- | |
545 | void intel_update_fbc(struct drm_device *dev) |
- | |
546 | { |
- | |
547 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
548 | struct drm_crtc *crtc = NULL, *tmp_crtc; |
- | |
549 | struct intel_crtc *intel_crtc; |
- | |
550 | struct drm_framebuffer *fb; |
- | |
551 | struct drm_i915_gem_object *obj; |
- | |
552 | const struct drm_display_mode *adjusted_mode; |
- | |
553 | unsigned int max_width, max_height; |
- | |
554 | - | ||
555 | if (!HAS_FBC(dev)) { |
- | |
556 | set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED); |
- | |
557 | return; |
- | |
558 | } |
- | |
559 | - | ||
560 | if (!i915.powersave) { |
- | |
561 | if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM)) |
- | |
562 | DRM_DEBUG_KMS("fbc disabled per module param\n"); |
- | |
563 | return; |
- | |
564 | } |
- | |
565 | - | ||
566 | /* |
- | |
567 | * If FBC is already on, we just have to verify that we can |
- | |
568 | * keep it that way... |
- | |
569 | * Need to disable if: |
- | |
570 | * - more than one pipe is active |
- | |
571 | * - changing FBC params (stride, fence, mode) |
- | |
572 | * - new fb is too large to fit in compressed buffer |
- | |
573 | * - going to an unsupported config (interlace, pixel multiply, etc.) |
- | |
574 | */ |
- | |
575 | for_each_crtc(dev, tmp_crtc) { |
- | |
576 | if (intel_crtc_active(tmp_crtc) && |
- | |
577 | to_intel_crtc(tmp_crtc)->primary_enabled) { |
- | |
578 | if (crtc) { |
- | |
579 | if (set_no_fbc_reason(dev_priv, FBC_MULTIPLE_PIPES)) |
- | |
580 | DRM_DEBUG_KMS("more than one pipe active, disabling compression\n"); |
- | |
581 | goto out_disable; |
- | |
582 | } |
- | |
583 | crtc = tmp_crtc; |
- | |
584 | } |
- | |
585 | } |
- | |
586 | - | ||
587 | if (!crtc || crtc->primary->fb == NULL) { |
- | |
588 | if (set_no_fbc_reason(dev_priv, FBC_NO_OUTPUT)) |
- | |
589 | DRM_DEBUG_KMS("no output, disabling\n"); |
- | |
590 | goto out_disable; |
- | |
591 | } |
- | |
592 | - | ||
593 | intel_crtc = to_intel_crtc(crtc); |
- | |
594 | fb = crtc->primary->fb; |
- | |
595 | obj = intel_fb_obj(fb); |
- | |
596 | adjusted_mode = &intel_crtc->config.adjusted_mode; |
- | |
597 | - | ||
598 | if (i915.enable_fbc < 0) { |
- | |
599 | if (set_no_fbc_reason(dev_priv, FBC_CHIP_DEFAULT)) |
- | |
600 | DRM_DEBUG_KMS("disabled per chip default\n"); |
- | |
601 | goto out_disable; |
- | |
602 | } |
- | |
603 | if (!i915.enable_fbc) { |
- | |
604 | if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM)) |
- | |
605 | DRM_DEBUG_KMS("fbc disabled per module param\n"); |
- | |
606 | goto out_disable; |
- | |
607 | } |
- | |
608 | if ((adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) || |
- | |
609 | (adjusted_mode->flags & DRM_MODE_FLAG_DBLSCAN)) { |
- | |
610 | if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE)) |
- | |
611 | DRM_DEBUG_KMS("mode incompatible with compression, " |
- | |
612 | "disabling\n"); |
- | |
613 | goto out_disable; |
- | |
614 | } |
- | |
615 | - | ||
616 | if (INTEL_INFO(dev)->gen >= 8 || IS_HASWELL(dev)) { |
- | |
617 | max_width = 4096; |
- | |
618 | max_height = 4096; |
- | |
619 | } else if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) { |
- | |
620 | max_width = 4096; |
- | |
621 | max_height = 2048; |
- | |
622 | } else { |
- | |
623 | max_width = 2048; |
- | |
624 | max_height = 1536; |
- | |
625 | } |
- | |
626 | if (intel_crtc->config.pipe_src_w > max_width || |
- | |
627 | intel_crtc->config.pipe_src_h > max_height) { |
- | |
628 | if (set_no_fbc_reason(dev_priv, FBC_MODE_TOO_LARGE)) |
- | |
629 | DRM_DEBUG_KMS("mode too large for compression, disabling\n"); |
- | |
630 | goto out_disable; |
- | |
631 | } |
- | |
632 | if ((INTEL_INFO(dev)->gen < 4 || HAS_DDI(dev)) && |
- | |
633 | intel_crtc->plane != PLANE_A) { |
- | |
634 | if (set_no_fbc_reason(dev_priv, FBC_BAD_PLANE)) |
- | |
635 | DRM_DEBUG_KMS("plane not A, disabling compression\n"); |
- | |
636 | goto out_disable; |
- | |
637 | } |
- | |
638 | - | ||
639 | /* The use of a CPU fence is mandatory in order to detect writes |
- | |
640 | * by the CPU to the scanout and trigger updates to the FBC. |
- | |
641 | */ |
- | |
642 | if (obj->tiling_mode != I915_TILING_X || |
- | |
643 | obj->fence_reg == I915_FENCE_REG_NONE) { |
- | |
644 | if (set_no_fbc_reason(dev_priv, FBC_NOT_TILED)) |
- | |
645 | DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n"); |
- | |
646 | goto out_disable; |
- | |
647 | } |
- | |
648 | if (INTEL_INFO(dev)->gen <= 4 && !IS_G4X(dev) && |
- | |
649 | to_intel_plane(crtc->primary)->rotation != BIT(DRM_ROTATE_0)) { |
- | |
650 | if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE)) |
- | |
651 | DRM_DEBUG_KMS("Rotation unsupported, disabling\n"); |
- | |
652 | goto out_disable; |
- | |
653 | } |
- | |
654 | - | ||
655 | /* If the kernel debugger is active, always disable compression */ |
- | |
656 | if (in_dbg_master()) |
- | |
657 | goto out_disable; |
- | |
658 | - | ||
659 | if (i915_gem_stolen_setup_compression(dev, obj->base.size, |
- | |
660 | drm_format_plane_cpp(fb->pixel_format, 0))) { |
- | |
661 | if (set_no_fbc_reason(dev_priv, FBC_STOLEN_TOO_SMALL)) |
- | |
662 | DRM_DEBUG_KMS("framebuffer too large, disabling compression\n"); |
- | |
663 | goto out_disable; |
- | |
664 | } |
- | |
665 | - | ||
666 | /* If the scanout has not changed, don't modify the FBC settings. |
- | |
667 | * Note that we make the fundamental assumption that the fb->obj |
- | |
668 | * cannot be unpinned (and have its GTT offset and fence revoked) |
- | |
669 | * without first being decoupled from the scanout and FBC disabled. |
- | |
670 | */ |
- | |
671 | if (dev_priv->fbc.plane == intel_crtc->plane && |
- | |
672 | dev_priv->fbc.fb_id == fb->base.id && |
- | |
673 | dev_priv->fbc.y == crtc->y) |
- | |
674 | return; |
- | |
675 | - | ||
676 | if (intel_fbc_enabled(dev)) { |
- | |
677 | /* We update FBC along two paths, after changing fb/crtc |
- | |
678 | * configuration (modeswitching) and after page-flipping |
- | |
679 | * finishes. For the latter, we know that not only did |
- | |
680 | * we disable the FBC at the start of the page-flip |
- | |
681 | * sequence, but also more than one vblank has passed. |
- | |
682 | * |
- | |
683 | * For the former case of modeswitching, it is possible |
- | |
684 | * to switch between two FBC valid configurations |
- | |
685 | * instantaneously so we do need to disable the FBC |
- | |
686 | * before we can modify its control registers. We also |
- | |
687 | * have to wait for the next vblank for that to take |
- | |
688 | * effect. However, since we delay enabling FBC we can |
- | |
689 | * assume that a vblank has passed since disabling and |
- | |
690 | * that we can safely alter the registers in the deferred |
- | |
691 | * callback. |
- | |
692 | * |
- | |
693 | * In the scenario that we go from a valid to invalid |
- | |
694 | * and then back to valid FBC configuration we have |
- | |
695 | * no strict enforcement that a vblank occurred since |
- | |
696 | * disabling the FBC. However, along all current pipe |
- | |
697 | * disabling paths we do need to wait for a vblank at |
- | |
698 | * some point. And we wait before enabling FBC anyway. |
- | |
699 | */ |
- | |
700 | DRM_DEBUG_KMS("disabling active FBC for update\n"); |
- | |
701 | intel_disable_fbc(dev); |
- | |
702 | } |
- | |
703 | - | ||
704 | intel_enable_fbc(crtc); |
- | |
705 | dev_priv->fbc.no_fbc_reason = FBC_OK; |
- | |
706 | return; |
- | |
707 | - | ||
708 | out_disable: |
- | |
709 | /* Multiple disables should be harmless */ |
- | |
710 | if (intel_fbc_enabled(dev)) { |
- | |
711 | DRM_DEBUG_KMS("unsupported config, disabling FBC\n"); |
- | |
712 | intel_disable_fbc(dev); |
- | |
713 | } |
- | |
714 | i915_gem_stolen_cleanup_compression(dev); |
- | |
715 | } |
74 | } |
716 | 75 | ||
717 | static void i915_pineview_get_mem_freq(struct drm_device *dev) |
76 | static void i915_pineview_get_mem_freq(struct drm_device *dev) |
718 | { |
77 | { |
719 | struct drm_i915_private *dev_priv = dev->dev_private; |
78 | struct drm_i915_private *dev_priv = dev->dev_private; |
720 | u32 tmp; |
79 | u32 tmp; |
721 | 80 | ||
722 | tmp = I915_READ(CLKCFG); |
81 | tmp = I915_READ(CLKCFG); |
723 | 82 | ||
724 | switch (tmp & CLKCFG_FSB_MASK) { |
83 | switch (tmp & CLKCFG_FSB_MASK) { |
725 | case CLKCFG_FSB_533: |
84 | case CLKCFG_FSB_533: |
726 | dev_priv->fsb_freq = 533; /* 133*4 */ |
85 | dev_priv->fsb_freq = 533; /* 133*4 */ |
727 | break; |
86 | break; |
728 | case CLKCFG_FSB_800: |
87 | case CLKCFG_FSB_800: |
729 | dev_priv->fsb_freq = 800; /* 200*4 */ |
88 | dev_priv->fsb_freq = 800; /* 200*4 */ |
730 | break; |
89 | break; |
731 | case CLKCFG_FSB_667: |
90 | case CLKCFG_FSB_667: |
732 | dev_priv->fsb_freq = 667; /* 167*4 */ |
91 | dev_priv->fsb_freq = 667; /* 167*4 */ |
733 | break; |
92 | break; |
734 | case CLKCFG_FSB_400: |
93 | case CLKCFG_FSB_400: |
735 | dev_priv->fsb_freq = 400; /* 100*4 */ |
94 | dev_priv->fsb_freq = 400; /* 100*4 */ |
736 | break; |
95 | break; |
737 | } |
96 | } |
738 | 97 | ||
739 | switch (tmp & CLKCFG_MEM_MASK) { |
98 | switch (tmp & CLKCFG_MEM_MASK) { |
740 | case CLKCFG_MEM_533: |
99 | case CLKCFG_MEM_533: |
741 | dev_priv->mem_freq = 533; |
100 | dev_priv->mem_freq = 533; |
742 | break; |
101 | break; |
743 | case CLKCFG_MEM_667: |
102 | case CLKCFG_MEM_667: |
744 | dev_priv->mem_freq = 667; |
103 | dev_priv->mem_freq = 667; |
745 | break; |
104 | break; |
746 | case CLKCFG_MEM_800: |
105 | case CLKCFG_MEM_800: |
747 | dev_priv->mem_freq = 800; |
106 | dev_priv->mem_freq = 800; |
748 | break; |
107 | break; |
749 | } |
108 | } |
750 | 109 | ||
751 | /* detect pineview DDR3 setting */ |
110 | /* detect pineview DDR3 setting */ |
752 | tmp = I915_READ(CSHRDDR3CTL); |
111 | tmp = I915_READ(CSHRDDR3CTL); |
753 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; |
112 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; |
754 | } |
113 | } |
755 | 114 | ||
756 | static void i915_ironlake_get_mem_freq(struct drm_device *dev) |
115 | static void i915_ironlake_get_mem_freq(struct drm_device *dev) |
757 | { |
116 | { |
758 | struct drm_i915_private *dev_priv = dev->dev_private; |
117 | struct drm_i915_private *dev_priv = dev->dev_private; |
759 | u16 ddrpll, csipll; |
118 | u16 ddrpll, csipll; |
760 | 119 | ||
761 | ddrpll = I915_READ16(DDRMPLL1); |
120 | ddrpll = I915_READ16(DDRMPLL1); |
762 | csipll = I915_READ16(CSIPLL0); |
121 | csipll = I915_READ16(CSIPLL0); |
763 | 122 | ||
764 | switch (ddrpll & 0xff) { |
123 | switch (ddrpll & 0xff) { |
765 | case 0xc: |
124 | case 0xc: |
766 | dev_priv->mem_freq = 800; |
125 | dev_priv->mem_freq = 800; |
767 | break; |
126 | break; |
768 | case 0x10: |
127 | case 0x10: |
769 | dev_priv->mem_freq = 1066; |
128 | dev_priv->mem_freq = 1066; |
770 | break; |
129 | break; |
771 | case 0x14: |
130 | case 0x14: |
772 | dev_priv->mem_freq = 1333; |
131 | dev_priv->mem_freq = 1333; |
773 | break; |
132 | break; |
774 | case 0x18: |
133 | case 0x18: |
775 | dev_priv->mem_freq = 1600; |
134 | dev_priv->mem_freq = 1600; |
776 | break; |
135 | break; |
777 | default: |
136 | default: |
778 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", |
137 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", |
779 | ddrpll & 0xff); |
138 | ddrpll & 0xff); |
780 | dev_priv->mem_freq = 0; |
139 | dev_priv->mem_freq = 0; |
781 | break; |
140 | break; |
782 | } |
141 | } |
783 | 142 | ||
784 | dev_priv->ips.r_t = dev_priv->mem_freq; |
143 | dev_priv->ips.r_t = dev_priv->mem_freq; |
785 | 144 | ||
786 | switch (csipll & 0x3ff) { |
145 | switch (csipll & 0x3ff) { |
787 | case 0x00c: |
146 | case 0x00c: |
788 | dev_priv->fsb_freq = 3200; |
147 | dev_priv->fsb_freq = 3200; |
789 | break; |
148 | break; |
790 | case 0x00e: |
149 | case 0x00e: |
791 | dev_priv->fsb_freq = 3733; |
150 | dev_priv->fsb_freq = 3733; |
792 | break; |
151 | break; |
793 | case 0x010: |
152 | case 0x010: |
794 | dev_priv->fsb_freq = 4266; |
153 | dev_priv->fsb_freq = 4266; |
795 | break; |
154 | break; |
796 | case 0x012: |
155 | case 0x012: |
797 | dev_priv->fsb_freq = 4800; |
156 | dev_priv->fsb_freq = 4800; |
798 | break; |
157 | break; |
799 | case 0x014: |
158 | case 0x014: |
800 | dev_priv->fsb_freq = 5333; |
159 | dev_priv->fsb_freq = 5333; |
801 | break; |
160 | break; |
802 | case 0x016: |
161 | case 0x016: |
803 | dev_priv->fsb_freq = 5866; |
162 | dev_priv->fsb_freq = 5866; |
804 | break; |
163 | break; |
805 | case 0x018: |
164 | case 0x018: |
806 | dev_priv->fsb_freq = 6400; |
165 | dev_priv->fsb_freq = 6400; |
807 | break; |
166 | break; |
808 | default: |
167 | default: |
809 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", |
168 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", |
810 | csipll & 0x3ff); |
169 | csipll & 0x3ff); |
811 | dev_priv->fsb_freq = 0; |
170 | dev_priv->fsb_freq = 0; |
812 | break; |
171 | break; |
813 | } |
172 | } |
814 | 173 | ||
815 | if (dev_priv->fsb_freq == 3200) { |
174 | if (dev_priv->fsb_freq == 3200) { |
816 | dev_priv->ips.c_m = 0; |
175 | dev_priv->ips.c_m = 0; |
817 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
176 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
818 | dev_priv->ips.c_m = 1; |
177 | dev_priv->ips.c_m = 1; |
819 | } else { |
178 | } else { |
820 | dev_priv->ips.c_m = 2; |
179 | dev_priv->ips.c_m = 2; |
821 | } |
180 | } |
822 | } |
181 | } |
823 | 182 | ||
824 | static const struct cxsr_latency cxsr_latency_table[] = { |
183 | static const struct cxsr_latency cxsr_latency_table[] = { |
825 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ |
184 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ |
826 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ |
185 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ |
827 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ |
186 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ |
828 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ |
187 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ |
829 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ |
188 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ |
830 | 189 | ||
831 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ |
190 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ |
832 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ |
191 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ |
833 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ |
192 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ |
834 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ |
193 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ |
835 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ |
194 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ |
836 | 195 | ||
837 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ |
196 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ |
838 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ |
197 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ |
839 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ |
198 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ |
840 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ |
199 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ |
841 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ |
200 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ |
842 | 201 | ||
843 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ |
202 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ |
844 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ |
203 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ |
845 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ |
204 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ |
846 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ |
205 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ |
847 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ |
206 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ |
848 | 207 | ||
849 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ |
208 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ |
850 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ |
209 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ |
851 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ |
210 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ |
852 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ |
211 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ |
853 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ |
212 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ |
854 | 213 | ||
855 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ |
214 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ |
856 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ |
215 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ |
857 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ |
216 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ |
858 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ |
217 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ |
859 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ |
218 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ |
860 | }; |
219 | }; |
861 | 220 | ||
862 | static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop, |
221 | static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop, |
863 | int is_ddr3, |
222 | int is_ddr3, |
864 | int fsb, |
223 | int fsb, |
865 | int mem) |
224 | int mem) |
866 | { |
225 | { |
867 | const struct cxsr_latency *latency; |
226 | const struct cxsr_latency *latency; |
868 | int i; |
227 | int i; |
869 | 228 | ||
870 | if (fsb == 0 || mem == 0) |
229 | if (fsb == 0 || mem == 0) |
871 | return NULL; |
230 | return NULL; |
872 | 231 | ||
873 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { |
232 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { |
874 | latency = &cxsr_latency_table[i]; |
233 | latency = &cxsr_latency_table[i]; |
875 | if (is_desktop == latency->is_desktop && |
234 | if (is_desktop == latency->is_desktop && |
876 | is_ddr3 == latency->is_ddr3 && |
235 | is_ddr3 == latency->is_ddr3 && |
877 | fsb == latency->fsb_freq && mem == latency->mem_freq) |
236 | fsb == latency->fsb_freq && mem == latency->mem_freq) |
878 | return latency; |
237 | return latency; |
879 | } |
238 | } |
880 | 239 | ||
881 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); |
240 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); |
882 | 241 | ||
883 | return NULL; |
242 | return NULL; |
884 | } |
243 | } |
- | 244 | ||
- | 245 | static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable) |
|
- | 246 | { |
|
- | 247 | u32 val; |
|
- | 248 | ||
- | 249 | mutex_lock(&dev_priv->rps.hw_lock); |
|
- | 250 | ||
- | 251 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); |
|
- | 252 | if (enable) |
|
- | 253 | val &= ~FORCE_DDR_HIGH_FREQ; |
|
- | 254 | else |
|
- | 255 | val |= FORCE_DDR_HIGH_FREQ; |
|
- | 256 | val &= ~FORCE_DDR_LOW_FREQ; |
|
- | 257 | val |= FORCE_DDR_FREQ_REQ_ACK; |
|
- | 258 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); |
|
- | 259 | ||
- | 260 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & |
|
- | 261 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) |
|
- | 262 | DRM_ERROR("timed out waiting for Punit DDR DVFS request\n"); |
|
- | 263 | ||
- | 264 | mutex_unlock(&dev_priv->rps.hw_lock); |
|
- | 265 | } |
|
- | 266 | ||
- | 267 | static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable) |
|
- | 268 | { |
|
- | 269 | u32 val; |
|
- | 270 | ||
- | 271 | mutex_lock(&dev_priv->rps.hw_lock); |
|
- | 272 | ||
- | 273 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); |
|
- | 274 | if (enable) |
|
- | 275 | val |= DSP_MAXFIFO_PM5_ENABLE; |
|
- | 276 | else |
|
- | 277 | val &= ~DSP_MAXFIFO_PM5_ENABLE; |
|
- | 278 | vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val); |
|
- | 279 | ||
- | 280 | mutex_unlock(&dev_priv->rps.hw_lock); |
|
- | 281 | } |
|
- | 282 | ||
- | 283 | #define FW_WM(value, plane) \ |
|
- | 284 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK) |
|
885 | 285 | ||
886 | void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable) |
286 | void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable) |
887 | { |
287 | { |
888 | struct drm_device *dev = dev_priv->dev; |
288 | struct drm_device *dev = dev_priv->dev; |
889 | u32 val; |
289 | u32 val; |
890 | 290 | ||
891 | if (IS_VALLEYVIEW(dev)) { |
291 | if (IS_VALLEYVIEW(dev)) { |
892 | I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0); |
292 | I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0); |
- | 293 | POSTING_READ(FW_BLC_SELF_VLV); |
|
- | 294 | dev_priv->wm.vlv.cxsr = enable; |
|
893 | } else if (IS_G4X(dev) || IS_CRESTLINE(dev)) { |
295 | } else if (IS_G4X(dev) || IS_CRESTLINE(dev)) { |
894 | I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0); |
296 | I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0); |
- | 297 | POSTING_READ(FW_BLC_SELF); |
|
895 | } else if (IS_PINEVIEW(dev)) { |
298 | } else if (IS_PINEVIEW(dev)) { |
896 | val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN; |
299 | val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN; |
897 | val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0; |
300 | val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0; |
898 | I915_WRITE(DSPFW3, val); |
301 | I915_WRITE(DSPFW3, val); |
- | 302 | POSTING_READ(DSPFW3); |
|
899 | } else if (IS_I945G(dev) || IS_I945GM(dev)) { |
303 | } else if (IS_I945G(dev) || IS_I945GM(dev)) { |
900 | val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) : |
304 | val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) : |
901 | _MASKED_BIT_DISABLE(FW_BLC_SELF_EN); |
305 | _MASKED_BIT_DISABLE(FW_BLC_SELF_EN); |
902 | I915_WRITE(FW_BLC_SELF, val); |
306 | I915_WRITE(FW_BLC_SELF, val); |
- | 307 | POSTING_READ(FW_BLC_SELF); |
|
903 | } else if (IS_I915GM(dev)) { |
308 | } else if (IS_I915GM(dev)) { |
904 | val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) : |
309 | val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) : |
905 | _MASKED_BIT_DISABLE(INSTPM_SELF_EN); |
310 | _MASKED_BIT_DISABLE(INSTPM_SELF_EN); |
906 | I915_WRITE(INSTPM, val); |
311 | I915_WRITE(INSTPM, val); |
- | 312 | POSTING_READ(INSTPM); |
|
907 | } else { |
313 | } else { |
908 | return; |
314 | return; |
909 | } |
315 | } |
910 | 316 | ||
911 | DRM_DEBUG_KMS("memory self-refresh is %s\n", |
317 | DRM_DEBUG_KMS("memory self-refresh is %s\n", |
912 | enable ? "enabled" : "disabled"); |
318 | enable ? "enabled" : "disabled"); |
913 | } |
319 | } |
- | 320 | ||
914 | 321 | ||
915 | /* |
322 | /* |
916 | * Latency for FIFO fetches is dependent on several factors: |
323 | * Latency for FIFO fetches is dependent on several factors: |
917 | * - memory configuration (speed, channels) |
324 | * - memory configuration (speed, channels) |
918 | * - chipset |
325 | * - chipset |
919 | * - current MCH state |
326 | * - current MCH state |
920 | * It can be fairly high in some situations, so here we assume a fairly |
327 | * It can be fairly high in some situations, so here we assume a fairly |
921 | * pessimal value. It's a tradeoff between extra memory fetches (if we |
328 | * pessimal value. It's a tradeoff between extra memory fetches (if we |
922 | * set this value too high, the FIFO will fetch frequently to stay full) |
329 | * set this value too high, the FIFO will fetch frequently to stay full) |
923 | * and power consumption (set it too low to save power and we might see |
330 | * and power consumption (set it too low to save power and we might see |
924 | * FIFO underruns and display "flicker"). |
331 | * FIFO underruns and display "flicker"). |
925 | * |
332 | * |
926 | * A value of 5us seems to be a good balance; safe for very low end |
333 | * A value of 5us seems to be a good balance; safe for very low end |
927 | * platforms but not overly aggressive on lower latency configs. |
334 | * platforms but not overly aggressive on lower latency configs. |
928 | */ |
335 | */ |
929 | static const int pessimal_latency_ns = 5000; |
336 | static const int pessimal_latency_ns = 5000; |
- | 337 | ||
- | 338 | #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \ |
|
- | 339 | ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8)) |
|
- | 340 | ||
- | 341 | static int vlv_get_fifo_size(struct drm_device *dev, |
|
- | 342 | enum pipe pipe, int plane) |
|
- | 343 | { |
|
- | 344 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
- | 345 | int sprite0_start, sprite1_start, size; |
|
- | 346 | ||
- | 347 | switch (pipe) { |
|
- | 348 | uint32_t dsparb, dsparb2, dsparb3; |
|
- | 349 | case PIPE_A: |
|
- | 350 | dsparb = I915_READ(DSPARB); |
|
- | 351 | dsparb2 = I915_READ(DSPARB2); |
|
- | 352 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0); |
|
- | 353 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4); |
|
- | 354 | break; |
|
- | 355 | case PIPE_B: |
|
- | 356 | dsparb = I915_READ(DSPARB); |
|
- | 357 | dsparb2 = I915_READ(DSPARB2); |
|
- | 358 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8); |
|
- | 359 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12); |
|
- | 360 | break; |
|
- | 361 | case PIPE_C: |
|
- | 362 | dsparb2 = I915_READ(DSPARB2); |
|
- | 363 | dsparb3 = I915_READ(DSPARB3); |
|
- | 364 | sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16); |
|
- | 365 | sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20); |
|
- | 366 | break; |
|
- | 367 | default: |
|
- | 368 | return 0; |
|
- | 369 | } |
|
- | 370 | ||
- | 371 | switch (plane) { |
|
- | 372 | case 0: |
|
- | 373 | size = sprite0_start; |
|
- | 374 | break; |
|
- | 375 | case 1: |
|
- | 376 | size = sprite1_start - sprite0_start; |
|
- | 377 | break; |
|
- | 378 | case 2: |
|
- | 379 | size = 512 - 1 - sprite1_start; |
|
- | 380 | break; |
|
- | 381 | default: |
|
- | 382 | return 0; |
|
- | 383 | } |
|
- | 384 | ||
- | 385 | DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n", |
|
- | 386 | pipe_name(pipe), plane == 0 ? "primary" : "sprite", |
|
- | 387 | plane == 0 ? plane_name(pipe) : sprite_name(pipe, plane - 1), |
|
- | 388 | size); |
|
- | 389 | ||
- | 390 | return size; |
|
- | 391 | } |
|
930 | 392 | ||
931 | static int i9xx_get_fifo_size(struct drm_device *dev, int plane) |
393 | static int i9xx_get_fifo_size(struct drm_device *dev, int plane) |
932 | { |
394 | { |
933 | struct drm_i915_private *dev_priv = dev->dev_private; |
395 | struct drm_i915_private *dev_priv = dev->dev_private; |
934 | uint32_t dsparb = I915_READ(DSPARB); |
396 | uint32_t dsparb = I915_READ(DSPARB); |
935 | int size; |
397 | int size; |
936 | 398 | ||
937 | size = dsparb & 0x7f; |
399 | size = dsparb & 0x7f; |
938 | if (plane) |
400 | if (plane) |
939 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; |
401 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; |
940 | 402 | ||
941 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
403 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
942 | plane ? "B" : "A", size); |
404 | plane ? "B" : "A", size); |
943 | 405 | ||
944 | return size; |
406 | return size; |
945 | } |
407 | } |
946 | 408 | ||
947 | static int i830_get_fifo_size(struct drm_device *dev, int plane) |
409 | static int i830_get_fifo_size(struct drm_device *dev, int plane) |
948 | { |
410 | { |
949 | struct drm_i915_private *dev_priv = dev->dev_private; |
411 | struct drm_i915_private *dev_priv = dev->dev_private; |
950 | uint32_t dsparb = I915_READ(DSPARB); |
412 | uint32_t dsparb = I915_READ(DSPARB); |
951 | int size; |
413 | int size; |
952 | 414 | ||
953 | size = dsparb & 0x1ff; |
415 | size = dsparb & 0x1ff; |
954 | if (plane) |
416 | if (plane) |
955 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; |
417 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; |
956 | size >>= 1; /* Convert to cachelines */ |
418 | size >>= 1; /* Convert to cachelines */ |
957 | 419 | ||
958 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
420 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
959 | plane ? "B" : "A", size); |
421 | plane ? "B" : "A", size); |
960 | 422 | ||
961 | return size; |
423 | return size; |
962 | } |
424 | } |
963 | 425 | ||
964 | static int i845_get_fifo_size(struct drm_device *dev, int plane) |
426 | static int i845_get_fifo_size(struct drm_device *dev, int plane) |
965 | { |
427 | { |
966 | struct drm_i915_private *dev_priv = dev->dev_private; |
428 | struct drm_i915_private *dev_priv = dev->dev_private; |
967 | uint32_t dsparb = I915_READ(DSPARB); |
429 | uint32_t dsparb = I915_READ(DSPARB); |
968 | int size; |
430 | int size; |
969 | 431 | ||
970 | size = dsparb & 0x7f; |
432 | size = dsparb & 0x7f; |
971 | size >>= 2; /* Convert to cachelines */ |
433 | size >>= 2; /* Convert to cachelines */ |
972 | 434 | ||
973 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
435 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
974 | plane ? "B" : "A", |
436 | plane ? "B" : "A", |
975 | size); |
437 | size); |
976 | 438 | ||
977 | return size; |
439 | return size; |
978 | } |
440 | } |
979 | 441 | ||
980 | /* Pineview has different values for various configs */ |
442 | /* Pineview has different values for various configs */ |
981 | static const struct intel_watermark_params pineview_display_wm = { |
443 | static const struct intel_watermark_params pineview_display_wm = { |
982 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
444 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
983 | .max_wm = PINEVIEW_MAX_WM, |
445 | .max_wm = PINEVIEW_MAX_WM, |
984 | .default_wm = PINEVIEW_DFT_WM, |
446 | .default_wm = PINEVIEW_DFT_WM, |
985 | .guard_size = PINEVIEW_GUARD_WM, |
447 | .guard_size = PINEVIEW_GUARD_WM, |
986 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
448 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
987 | }; |
449 | }; |
988 | static const struct intel_watermark_params pineview_display_hplloff_wm = { |
450 | static const struct intel_watermark_params pineview_display_hplloff_wm = { |
989 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
451 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
990 | .max_wm = PINEVIEW_MAX_WM, |
452 | .max_wm = PINEVIEW_MAX_WM, |
991 | .default_wm = PINEVIEW_DFT_HPLLOFF_WM, |
453 | .default_wm = PINEVIEW_DFT_HPLLOFF_WM, |
992 | .guard_size = PINEVIEW_GUARD_WM, |
454 | .guard_size = PINEVIEW_GUARD_WM, |
993 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
455 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
994 | }; |
456 | }; |
995 | static const struct intel_watermark_params pineview_cursor_wm = { |
457 | static const struct intel_watermark_params pineview_cursor_wm = { |
996 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
458 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
997 | .max_wm = PINEVIEW_CURSOR_MAX_WM, |
459 | .max_wm = PINEVIEW_CURSOR_MAX_WM, |
998 | .default_wm = PINEVIEW_CURSOR_DFT_WM, |
460 | .default_wm = PINEVIEW_CURSOR_DFT_WM, |
999 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, |
461 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, |
1000 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
462 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
1001 | }; |
463 | }; |
1002 | static const struct intel_watermark_params pineview_cursor_hplloff_wm = { |
464 | static const struct intel_watermark_params pineview_cursor_hplloff_wm = { |
1003 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
465 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
1004 | .max_wm = PINEVIEW_CURSOR_MAX_WM, |
466 | .max_wm = PINEVIEW_CURSOR_MAX_WM, |
1005 | .default_wm = PINEVIEW_CURSOR_DFT_WM, |
467 | .default_wm = PINEVIEW_CURSOR_DFT_WM, |
1006 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, |
468 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, |
1007 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
469 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, |
1008 | }; |
470 | }; |
1009 | static const struct intel_watermark_params g4x_wm_info = { |
471 | static const struct intel_watermark_params g4x_wm_info = { |
1010 | .fifo_size = G4X_FIFO_SIZE, |
472 | .fifo_size = G4X_FIFO_SIZE, |
1011 | .max_wm = G4X_MAX_WM, |
473 | .max_wm = G4X_MAX_WM, |
1012 | .default_wm = G4X_MAX_WM, |
474 | .default_wm = G4X_MAX_WM, |
1013 | .guard_size = 2, |
475 | .guard_size = 2, |
1014 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
476 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
1015 | }; |
477 | }; |
1016 | static const struct intel_watermark_params g4x_cursor_wm_info = { |
478 | static const struct intel_watermark_params g4x_cursor_wm_info = { |
1017 | .fifo_size = I965_CURSOR_FIFO, |
479 | .fifo_size = I965_CURSOR_FIFO, |
1018 | .max_wm = I965_CURSOR_MAX_WM, |
480 | .max_wm = I965_CURSOR_MAX_WM, |
1019 | .default_wm = I965_CURSOR_DFT_WM, |
481 | .default_wm = I965_CURSOR_DFT_WM, |
1020 | .guard_size = 2, |
482 | .guard_size = 2, |
1021 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
483 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
1022 | }; |
484 | }; |
1023 | static const struct intel_watermark_params valleyview_wm_info = { |
485 | static const struct intel_watermark_params valleyview_wm_info = { |
1024 | .fifo_size = VALLEYVIEW_FIFO_SIZE, |
486 | .fifo_size = VALLEYVIEW_FIFO_SIZE, |
1025 | .max_wm = VALLEYVIEW_MAX_WM, |
487 | .max_wm = VALLEYVIEW_MAX_WM, |
1026 | .default_wm = VALLEYVIEW_MAX_WM, |
488 | .default_wm = VALLEYVIEW_MAX_WM, |
1027 | .guard_size = 2, |
489 | .guard_size = 2, |
1028 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
490 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
1029 | }; |
491 | }; |
1030 | static const struct intel_watermark_params valleyview_cursor_wm_info = { |
492 | static const struct intel_watermark_params valleyview_cursor_wm_info = { |
1031 | .fifo_size = I965_CURSOR_FIFO, |
493 | .fifo_size = I965_CURSOR_FIFO, |
1032 | .max_wm = VALLEYVIEW_CURSOR_MAX_WM, |
494 | .max_wm = VALLEYVIEW_CURSOR_MAX_WM, |
1033 | .default_wm = I965_CURSOR_DFT_WM, |
495 | .default_wm = I965_CURSOR_DFT_WM, |
1034 | .guard_size = 2, |
496 | .guard_size = 2, |
1035 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
497 | .cacheline_size = G4X_FIFO_LINE_SIZE, |
1036 | }; |
498 | }; |
1037 | static const struct intel_watermark_params i965_cursor_wm_info = { |
499 | static const struct intel_watermark_params i965_cursor_wm_info = { |
1038 | .fifo_size = I965_CURSOR_FIFO, |
500 | .fifo_size = I965_CURSOR_FIFO, |
1039 | .max_wm = I965_CURSOR_MAX_WM, |
501 | .max_wm = I965_CURSOR_MAX_WM, |
1040 | .default_wm = I965_CURSOR_DFT_WM, |
502 | .default_wm = I965_CURSOR_DFT_WM, |
1041 | .guard_size = 2, |
503 | .guard_size = 2, |
1042 | .cacheline_size = I915_FIFO_LINE_SIZE, |
504 | .cacheline_size = I915_FIFO_LINE_SIZE, |
1043 | }; |
505 | }; |
1044 | static const struct intel_watermark_params i945_wm_info = { |
506 | static const struct intel_watermark_params i945_wm_info = { |
1045 | .fifo_size = I945_FIFO_SIZE, |
507 | .fifo_size = I945_FIFO_SIZE, |
1046 | .max_wm = I915_MAX_WM, |
508 | .max_wm = I915_MAX_WM, |
1047 | .default_wm = 1, |
509 | .default_wm = 1, |
1048 | .guard_size = 2, |
510 | .guard_size = 2, |
1049 | .cacheline_size = I915_FIFO_LINE_SIZE, |
511 | .cacheline_size = I915_FIFO_LINE_SIZE, |
1050 | }; |
512 | }; |
1051 | static const struct intel_watermark_params i915_wm_info = { |
513 | static const struct intel_watermark_params i915_wm_info = { |
1052 | .fifo_size = I915_FIFO_SIZE, |
514 | .fifo_size = I915_FIFO_SIZE, |
1053 | .max_wm = I915_MAX_WM, |
515 | .max_wm = I915_MAX_WM, |
1054 | .default_wm = 1, |
516 | .default_wm = 1, |
1055 | .guard_size = 2, |
517 | .guard_size = 2, |
1056 | .cacheline_size = I915_FIFO_LINE_SIZE, |
518 | .cacheline_size = I915_FIFO_LINE_SIZE, |
1057 | }; |
519 | }; |
1058 | static const struct intel_watermark_params i830_a_wm_info = { |
520 | static const struct intel_watermark_params i830_a_wm_info = { |
1059 | .fifo_size = I855GM_FIFO_SIZE, |
521 | .fifo_size = I855GM_FIFO_SIZE, |
1060 | .max_wm = I915_MAX_WM, |
522 | .max_wm = I915_MAX_WM, |
1061 | .default_wm = 1, |
523 | .default_wm = 1, |
1062 | .guard_size = 2, |
524 | .guard_size = 2, |
1063 | .cacheline_size = I830_FIFO_LINE_SIZE, |
525 | .cacheline_size = I830_FIFO_LINE_SIZE, |
1064 | }; |
526 | }; |
1065 | static const struct intel_watermark_params i830_bc_wm_info = { |
527 | static const struct intel_watermark_params i830_bc_wm_info = { |
1066 | .fifo_size = I855GM_FIFO_SIZE, |
528 | .fifo_size = I855GM_FIFO_SIZE, |
1067 | .max_wm = I915_MAX_WM/2, |
529 | .max_wm = I915_MAX_WM/2, |
1068 | .default_wm = 1, |
530 | .default_wm = 1, |
1069 | .guard_size = 2, |
531 | .guard_size = 2, |
1070 | .cacheline_size = I830_FIFO_LINE_SIZE, |
532 | .cacheline_size = I830_FIFO_LINE_SIZE, |
1071 | }; |
533 | }; |
1072 | static const struct intel_watermark_params i845_wm_info = { |
534 | static const struct intel_watermark_params i845_wm_info = { |
1073 | .fifo_size = I830_FIFO_SIZE, |
535 | .fifo_size = I830_FIFO_SIZE, |
1074 | .max_wm = I915_MAX_WM, |
536 | .max_wm = I915_MAX_WM, |
1075 | .default_wm = 1, |
537 | .default_wm = 1, |
1076 | .guard_size = 2, |
538 | .guard_size = 2, |
1077 | .cacheline_size = I830_FIFO_LINE_SIZE, |
539 | .cacheline_size = I830_FIFO_LINE_SIZE, |
1078 | }; |
540 | }; |
1079 | 541 | ||
1080 | /** |
542 | /** |
1081 | * intel_calculate_wm - calculate watermark level |
543 | * intel_calculate_wm - calculate watermark level |
1082 | * @clock_in_khz: pixel clock |
544 | * @clock_in_khz: pixel clock |
1083 | * @wm: chip FIFO params |
545 | * @wm: chip FIFO params |
1084 | * @pixel_size: display pixel size |
546 | * @pixel_size: display pixel size |
1085 | * @latency_ns: memory latency for the platform |
547 | * @latency_ns: memory latency for the platform |
1086 | * |
548 | * |
1087 | * Calculate the watermark level (the level at which the display plane will |
549 | * Calculate the watermark level (the level at which the display plane will |
1088 | * start fetching from memory again). Each chip has a different display |
550 | * start fetching from memory again). Each chip has a different display |
1089 | * FIFO size and allocation, so the caller needs to figure that out and pass |
551 | * FIFO size and allocation, so the caller needs to figure that out and pass |
1090 | * in the correct intel_watermark_params structure. |
552 | * in the correct intel_watermark_params structure. |
1091 | * |
553 | * |
1092 | * As the pixel clock runs, the FIFO will be drained at a rate that depends |
554 | * As the pixel clock runs, the FIFO will be drained at a rate that depends |
1093 | * on the pixel size. When it reaches the watermark level, it'll start |
555 | * on the pixel size. When it reaches the watermark level, it'll start |
1094 | * fetching FIFO line sized based chunks from memory until the FIFO fills |
556 | * fetching FIFO line sized based chunks from memory until the FIFO fills |
1095 | * past the watermark point. If the FIFO drains completely, a FIFO underrun |
557 | * past the watermark point. If the FIFO drains completely, a FIFO underrun |
1096 | * will occur, and a display engine hang could result. |
558 | * will occur, and a display engine hang could result. |
1097 | */ |
559 | */ |
1098 | static unsigned long intel_calculate_wm(unsigned long clock_in_khz, |
560 | static unsigned long intel_calculate_wm(unsigned long clock_in_khz, |
1099 | const struct intel_watermark_params *wm, |
561 | const struct intel_watermark_params *wm, |
1100 | int fifo_size, |
562 | int fifo_size, |
1101 | int pixel_size, |
563 | int pixel_size, |
1102 | unsigned long latency_ns) |
564 | unsigned long latency_ns) |
1103 | { |
565 | { |
1104 | long entries_required, wm_size; |
566 | long entries_required, wm_size; |
1105 | 567 | ||
1106 | /* |
568 | /* |
1107 | * Note: we need to make sure we don't overflow for various clock & |
569 | * Note: we need to make sure we don't overflow for various clock & |
1108 | * latency values. |
570 | * latency values. |
1109 | * clocks go from a few thousand to several hundred thousand. |
571 | * clocks go from a few thousand to several hundred thousand. |
1110 | * latency is usually a few thousand |
572 | * latency is usually a few thousand |
1111 | */ |
573 | */ |
1112 | entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) / |
574 | entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) / |
1113 | 1000; |
575 | 1000; |
1114 | entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size); |
576 | entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size); |
1115 | 577 | ||
1116 | DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required); |
578 | DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required); |
1117 | 579 | ||
1118 | wm_size = fifo_size - (entries_required + wm->guard_size); |
580 | wm_size = fifo_size - (entries_required + wm->guard_size); |
1119 | 581 | ||
1120 | DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size); |
582 | DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size); |
1121 | 583 | ||
1122 | /* Don't promote wm_size to unsigned... */ |
584 | /* Don't promote wm_size to unsigned... */ |
1123 | if (wm_size > (long)wm->max_wm) |
585 | if (wm_size > (long)wm->max_wm) |
1124 | wm_size = wm->max_wm; |
586 | wm_size = wm->max_wm; |
1125 | if (wm_size <= 0) |
587 | if (wm_size <= 0) |
1126 | wm_size = wm->default_wm; |
588 | wm_size = wm->default_wm; |
1127 | 589 | ||
1128 | /* |
590 | /* |
1129 | * Bspec seems to indicate that the value shouldn't be lower than |
591 | * Bspec seems to indicate that the value shouldn't be lower than |
1130 | * 'burst size + 1'. Certainly 830 is quite unhappy with low values. |
592 | * 'burst size + 1'. Certainly 830 is quite unhappy with low values. |
1131 | * Lets go for 8 which is the burst size since certain platforms |
593 | * Lets go for 8 which is the burst size since certain platforms |
1132 | * already use a hardcoded 8 (which is what the spec says should be |
594 | * already use a hardcoded 8 (which is what the spec says should be |
1133 | * done). |
595 | * done). |
1134 | */ |
596 | */ |
1135 | if (wm_size <= 8) |
597 | if (wm_size <= 8) |
1136 | wm_size = 8; |
598 | wm_size = 8; |
1137 | 599 | ||
1138 | return wm_size; |
600 | return wm_size; |
1139 | } |
601 | } |
1140 | 602 | ||
1141 | static struct drm_crtc *single_enabled_crtc(struct drm_device *dev) |
603 | static struct drm_crtc *single_enabled_crtc(struct drm_device *dev) |
1142 | { |
604 | { |
1143 | struct drm_crtc *crtc, *enabled = NULL; |
605 | struct drm_crtc *crtc, *enabled = NULL; |
1144 | 606 | ||
1145 | for_each_crtc(dev, crtc) { |
607 | for_each_crtc(dev, crtc) { |
1146 | if (intel_crtc_active(crtc)) { |
608 | if (intel_crtc_active(crtc)) { |
1147 | if (enabled) |
609 | if (enabled) |
1148 | return NULL; |
610 | return NULL; |
1149 | enabled = crtc; |
611 | enabled = crtc; |
1150 | } |
612 | } |
1151 | } |
613 | } |
1152 | 614 | ||
1153 | return enabled; |
615 | return enabled; |
1154 | } |
616 | } |
1155 | 617 | ||
1156 | static void pineview_update_wm(struct drm_crtc *unused_crtc) |
618 | static void pineview_update_wm(struct drm_crtc *unused_crtc) |
1157 | { |
619 | { |
1158 | struct drm_device *dev = unused_crtc->dev; |
620 | struct drm_device *dev = unused_crtc->dev; |
1159 | struct drm_i915_private *dev_priv = dev->dev_private; |
621 | struct drm_i915_private *dev_priv = dev->dev_private; |
1160 | struct drm_crtc *crtc; |
622 | struct drm_crtc *crtc; |
1161 | const struct cxsr_latency *latency; |
623 | const struct cxsr_latency *latency; |
1162 | u32 reg; |
624 | u32 reg; |
1163 | unsigned long wm; |
625 | unsigned long wm; |
1164 | 626 | ||
1165 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3, |
627 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3, |
1166 | dev_priv->fsb_freq, dev_priv->mem_freq); |
628 | dev_priv->fsb_freq, dev_priv->mem_freq); |
1167 | if (!latency) { |
629 | if (!latency) { |
1168 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); |
630 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); |
1169 | intel_set_memory_cxsr(dev_priv, false); |
631 | intel_set_memory_cxsr(dev_priv, false); |
1170 | return; |
632 | return; |
1171 | } |
633 | } |
1172 | 634 | ||
1173 | crtc = single_enabled_crtc(dev); |
635 | crtc = single_enabled_crtc(dev); |
1174 | if (crtc) { |
636 | if (crtc) { |
1175 | const struct drm_display_mode *adjusted_mode; |
637 | const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1176 | int pixel_size = crtc->primary->fb->bits_per_pixel / 8; |
638 | int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
1177 | int clock; |
- | |
1178 | - | ||
1179 | adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode; |
- | |
1180 | clock = adjusted_mode->crtc_clock; |
639 | int clock = adjusted_mode->crtc_clock; |
1181 | 640 | ||
1182 | /* Display SR */ |
641 | /* Display SR */ |
1183 | wm = intel_calculate_wm(clock, &pineview_display_wm, |
642 | wm = intel_calculate_wm(clock, &pineview_display_wm, |
1184 | pineview_display_wm.fifo_size, |
643 | pineview_display_wm.fifo_size, |
1185 | pixel_size, latency->display_sr); |
644 | pixel_size, latency->display_sr); |
1186 | reg = I915_READ(DSPFW1); |
645 | reg = I915_READ(DSPFW1); |
1187 | reg &= ~DSPFW_SR_MASK; |
646 | reg &= ~DSPFW_SR_MASK; |
1188 | reg |= wm << DSPFW_SR_SHIFT; |
647 | reg |= FW_WM(wm, SR); |
1189 | I915_WRITE(DSPFW1, reg); |
648 | I915_WRITE(DSPFW1, reg); |
1190 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); |
649 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); |
1191 | 650 | ||
1192 | /* cursor SR */ |
651 | /* cursor SR */ |
1193 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, |
652 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, |
1194 | pineview_display_wm.fifo_size, |
653 | pineview_display_wm.fifo_size, |
1195 | pixel_size, latency->cursor_sr); |
654 | pixel_size, latency->cursor_sr); |
1196 | reg = I915_READ(DSPFW3); |
655 | reg = I915_READ(DSPFW3); |
1197 | reg &= ~DSPFW_CURSOR_SR_MASK; |
656 | reg &= ~DSPFW_CURSOR_SR_MASK; |
1198 | reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT; |
657 | reg |= FW_WM(wm, CURSOR_SR); |
1199 | I915_WRITE(DSPFW3, reg); |
658 | I915_WRITE(DSPFW3, reg); |
1200 | 659 | ||
1201 | /* Display HPLL off SR */ |
660 | /* Display HPLL off SR */ |
1202 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, |
661 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, |
1203 | pineview_display_hplloff_wm.fifo_size, |
662 | pineview_display_hplloff_wm.fifo_size, |
1204 | pixel_size, latency->display_hpll_disable); |
663 | pixel_size, latency->display_hpll_disable); |
1205 | reg = I915_READ(DSPFW3); |
664 | reg = I915_READ(DSPFW3); |
1206 | reg &= ~DSPFW_HPLL_SR_MASK; |
665 | reg &= ~DSPFW_HPLL_SR_MASK; |
1207 | reg |= wm & DSPFW_HPLL_SR_MASK; |
666 | reg |= FW_WM(wm, HPLL_SR); |
1208 | I915_WRITE(DSPFW3, reg); |
667 | I915_WRITE(DSPFW3, reg); |
1209 | 668 | ||
1210 | /* cursor HPLL off SR */ |
669 | /* cursor HPLL off SR */ |
1211 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, |
670 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, |
1212 | pineview_display_hplloff_wm.fifo_size, |
671 | pineview_display_hplloff_wm.fifo_size, |
1213 | pixel_size, latency->cursor_hpll_disable); |
672 | pixel_size, latency->cursor_hpll_disable); |
1214 | reg = I915_READ(DSPFW3); |
673 | reg = I915_READ(DSPFW3); |
1215 | reg &= ~DSPFW_HPLL_CURSOR_MASK; |
674 | reg &= ~DSPFW_HPLL_CURSOR_MASK; |
1216 | reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT; |
675 | reg |= FW_WM(wm, HPLL_CURSOR); |
1217 | I915_WRITE(DSPFW3, reg); |
676 | I915_WRITE(DSPFW3, reg); |
1218 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); |
677 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); |
1219 | 678 | ||
1220 | intel_set_memory_cxsr(dev_priv, true); |
679 | intel_set_memory_cxsr(dev_priv, true); |
1221 | } else { |
680 | } else { |
1222 | intel_set_memory_cxsr(dev_priv, false); |
681 | intel_set_memory_cxsr(dev_priv, false); |
1223 | } |
682 | } |
1224 | } |
683 | } |
1225 | 684 | ||
1226 | static bool g4x_compute_wm0(struct drm_device *dev, |
685 | static bool g4x_compute_wm0(struct drm_device *dev, |
1227 | int plane, |
686 | int plane, |
1228 | const struct intel_watermark_params *display, |
687 | const struct intel_watermark_params *display, |
1229 | int display_latency_ns, |
688 | int display_latency_ns, |
1230 | const struct intel_watermark_params *cursor, |
689 | const struct intel_watermark_params *cursor, |
1231 | int cursor_latency_ns, |
690 | int cursor_latency_ns, |
1232 | int *plane_wm, |
691 | int *plane_wm, |
1233 | int *cursor_wm) |
692 | int *cursor_wm) |
1234 | { |
693 | { |
1235 | struct drm_crtc *crtc; |
694 | struct drm_crtc *crtc; |
1236 | const struct drm_display_mode *adjusted_mode; |
695 | const struct drm_display_mode *adjusted_mode; |
1237 | int htotal, hdisplay, clock, pixel_size; |
696 | int htotal, hdisplay, clock, pixel_size; |
1238 | int line_time_us, line_count; |
697 | int line_time_us, line_count; |
1239 | int entries, tlb_miss; |
698 | int entries, tlb_miss; |
1240 | 699 | ||
1241 | crtc = intel_get_crtc_for_plane(dev, plane); |
700 | crtc = intel_get_crtc_for_plane(dev, plane); |
1242 | if (!intel_crtc_active(crtc)) { |
701 | if (!intel_crtc_active(crtc)) { |
1243 | *cursor_wm = cursor->guard_size; |
702 | *cursor_wm = cursor->guard_size; |
1244 | *plane_wm = display->guard_size; |
703 | *plane_wm = display->guard_size; |
1245 | return false; |
704 | return false; |
1246 | } |
705 | } |
1247 | 706 | ||
1248 | adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode; |
707 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1249 | clock = adjusted_mode->crtc_clock; |
708 | clock = adjusted_mode->crtc_clock; |
1250 | htotal = adjusted_mode->crtc_htotal; |
709 | htotal = adjusted_mode->crtc_htotal; |
1251 | hdisplay = to_intel_crtc(crtc)->config.pipe_src_w; |
710 | hdisplay = to_intel_crtc(crtc)->config->pipe_src_w; |
1252 | pixel_size = crtc->primary->fb->bits_per_pixel / 8; |
711 | pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
1253 | 712 | ||
1254 | /* Use the small buffer method to calculate plane watermark */ |
713 | /* Use the small buffer method to calculate plane watermark */ |
1255 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; |
714 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; |
1256 | tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8; |
715 | tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8; |
1257 | if (tlb_miss > 0) |
716 | if (tlb_miss > 0) |
1258 | entries += tlb_miss; |
717 | entries += tlb_miss; |
1259 | entries = DIV_ROUND_UP(entries, display->cacheline_size); |
718 | entries = DIV_ROUND_UP(entries, display->cacheline_size); |
1260 | *plane_wm = entries + display->guard_size; |
719 | *plane_wm = entries + display->guard_size; |
1261 | if (*plane_wm > (int)display->max_wm) |
720 | if (*plane_wm > (int)display->max_wm) |
1262 | *plane_wm = display->max_wm; |
721 | *plane_wm = display->max_wm; |
1263 | 722 | ||
1264 | /* Use the large buffer method to calculate cursor watermark */ |
723 | /* Use the large buffer method to calculate cursor watermark */ |
1265 | line_time_us = max(htotal * 1000 / clock, 1); |
724 | line_time_us = max(htotal * 1000 / clock, 1); |
1266 | line_count = (cursor_latency_ns / line_time_us + 1000) / 1000; |
725 | line_count = (cursor_latency_ns / line_time_us + 1000) / 1000; |
1267 | entries = line_count * to_intel_crtc(crtc)->cursor_width * pixel_size; |
726 | entries = line_count * crtc->cursor->state->crtc_w * pixel_size; |
1268 | tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8; |
727 | tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8; |
1269 | if (tlb_miss > 0) |
728 | if (tlb_miss > 0) |
1270 | entries += tlb_miss; |
729 | entries += tlb_miss; |
1271 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
730 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
1272 | *cursor_wm = entries + cursor->guard_size; |
731 | *cursor_wm = entries + cursor->guard_size; |
1273 | if (*cursor_wm > (int)cursor->max_wm) |
732 | if (*cursor_wm > (int)cursor->max_wm) |
1274 | *cursor_wm = (int)cursor->max_wm; |
733 | *cursor_wm = (int)cursor->max_wm; |
1275 | 734 | ||
1276 | return true; |
735 | return true; |
1277 | } |
736 | } |
1278 | 737 | ||
1279 | /* |
738 | /* |
1280 | * Check the wm result. |
739 | * Check the wm result. |
1281 | * |
740 | * |
1282 | * If any calculated watermark values is larger than the maximum value that |
741 | * If any calculated watermark values is larger than the maximum value that |
1283 | * can be programmed into the associated watermark register, that watermark |
742 | * can be programmed into the associated watermark register, that watermark |
1284 | * must be disabled. |
743 | * must be disabled. |
1285 | */ |
744 | */ |
1286 | static bool g4x_check_srwm(struct drm_device *dev, |
745 | static bool g4x_check_srwm(struct drm_device *dev, |
1287 | int display_wm, int cursor_wm, |
746 | int display_wm, int cursor_wm, |
1288 | const struct intel_watermark_params *display, |
747 | const struct intel_watermark_params *display, |
1289 | const struct intel_watermark_params *cursor) |
748 | const struct intel_watermark_params *cursor) |
1290 | { |
749 | { |
1291 | DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n", |
750 | DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n", |
1292 | display_wm, cursor_wm); |
751 | display_wm, cursor_wm); |
1293 | 752 | ||
1294 | if (display_wm > display->max_wm) { |
753 | if (display_wm > display->max_wm) { |
1295 | DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n", |
754 | DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n", |
1296 | display_wm, display->max_wm); |
755 | display_wm, display->max_wm); |
1297 | return false; |
756 | return false; |
1298 | } |
757 | } |
1299 | 758 | ||
1300 | if (cursor_wm > cursor->max_wm) { |
759 | if (cursor_wm > cursor->max_wm) { |
1301 | DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n", |
760 | DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n", |
1302 | cursor_wm, cursor->max_wm); |
761 | cursor_wm, cursor->max_wm); |
1303 | return false; |
762 | return false; |
1304 | } |
763 | } |
1305 | 764 | ||
1306 | if (!(display_wm || cursor_wm)) { |
765 | if (!(display_wm || cursor_wm)) { |
1307 | DRM_DEBUG_KMS("SR latency is 0, disabling\n"); |
766 | DRM_DEBUG_KMS("SR latency is 0, disabling\n"); |
1308 | return false; |
767 | return false; |
1309 | } |
768 | } |
1310 | 769 | ||
1311 | return true; |
770 | return true; |
1312 | } |
771 | } |
1313 | 772 | ||
1314 | static bool g4x_compute_srwm(struct drm_device *dev, |
773 | static bool g4x_compute_srwm(struct drm_device *dev, |
1315 | int plane, |
774 | int plane, |
1316 | int latency_ns, |
775 | int latency_ns, |
1317 | const struct intel_watermark_params *display, |
776 | const struct intel_watermark_params *display, |
1318 | const struct intel_watermark_params *cursor, |
777 | const struct intel_watermark_params *cursor, |
1319 | int *display_wm, int *cursor_wm) |
778 | int *display_wm, int *cursor_wm) |
1320 | { |
779 | { |
1321 | struct drm_crtc *crtc; |
780 | struct drm_crtc *crtc; |
1322 | const struct drm_display_mode *adjusted_mode; |
781 | const struct drm_display_mode *adjusted_mode; |
1323 | int hdisplay, htotal, pixel_size, clock; |
782 | int hdisplay, htotal, pixel_size, clock; |
1324 | unsigned long line_time_us; |
783 | unsigned long line_time_us; |
1325 | int line_count, line_size; |
784 | int line_count, line_size; |
1326 | int small, large; |
785 | int small, large; |
1327 | int entries; |
786 | int entries; |
1328 | 787 | ||
1329 | if (!latency_ns) { |
788 | if (!latency_ns) { |
1330 | *display_wm = *cursor_wm = 0; |
789 | *display_wm = *cursor_wm = 0; |
1331 | return false; |
790 | return false; |
1332 | } |
791 | } |
1333 | 792 | ||
1334 | crtc = intel_get_crtc_for_plane(dev, plane); |
793 | crtc = intel_get_crtc_for_plane(dev, plane); |
1335 | adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode; |
794 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1336 | clock = adjusted_mode->crtc_clock; |
795 | clock = adjusted_mode->crtc_clock; |
1337 | htotal = adjusted_mode->crtc_htotal; |
796 | htotal = adjusted_mode->crtc_htotal; |
1338 | hdisplay = to_intel_crtc(crtc)->config.pipe_src_w; |
797 | hdisplay = to_intel_crtc(crtc)->config->pipe_src_w; |
1339 | pixel_size = crtc->primary->fb->bits_per_pixel / 8; |
798 | pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
1340 | 799 | ||
1341 | line_time_us = max(htotal * 1000 / clock, 1); |
800 | line_time_us = max(htotal * 1000 / clock, 1); |
1342 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
801 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
1343 | line_size = hdisplay * pixel_size; |
802 | line_size = hdisplay * pixel_size; |
1344 | 803 | ||
1345 | /* Use the minimum of the small and large buffer method for primary */ |
804 | /* Use the minimum of the small and large buffer method for primary */ |
1346 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; |
805 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; |
1347 | large = line_count * line_size; |
806 | large = line_count * line_size; |
1348 | 807 | ||
1349 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); |
808 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); |
1350 | *display_wm = entries + display->guard_size; |
809 | *display_wm = entries + display->guard_size; |
1351 | 810 | ||
1352 | /* calculate the self-refresh watermark for display cursor */ |
811 | /* calculate the self-refresh watermark for display cursor */ |
1353 | entries = line_count * pixel_size * to_intel_crtc(crtc)->cursor_width; |
812 | entries = line_count * pixel_size * crtc->cursor->state->crtc_w; |
1354 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
813 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
1355 | *cursor_wm = entries + cursor->guard_size; |
814 | *cursor_wm = entries + cursor->guard_size; |
1356 | 815 | ||
1357 | return g4x_check_srwm(dev, |
816 | return g4x_check_srwm(dev, |
1358 | *display_wm, *cursor_wm, |
817 | *display_wm, *cursor_wm, |
1359 | display, cursor); |
818 | display, cursor); |
1360 | } |
819 | } |
1361 | 820 | ||
1362 | static bool vlv_compute_drain_latency(struct drm_crtc *crtc, |
821 | #define FW_WM_VLV(value, plane) \ |
- | 822 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV) |
|
1363 | int pixel_size, |
823 | |
1364 | int *prec_mult, |
824 | static void vlv_write_wm_values(struct intel_crtc *crtc, |
1365 | int *drain_latency) |
825 | const struct vlv_wm_values *wm) |
1366 | { |
826 | { |
1367 | struct drm_device *dev = crtc->dev; |
827 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); |
1368 | int entries; |
- | |
- | 828 | enum pipe pipe = crtc->pipe; |
|
1369 | int clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock; |
829 | |
- | 830 | I915_WRITE(VLV_DDL(pipe), |
|
- | 831 | (wm->ddl[pipe].cursor << DDL_CURSOR_SHIFT) | |
|
1370 | 832 | (wm->ddl[pipe].sprite[1] << DDL_SPRITE_SHIFT(1)) | |
|
- | 833 | (wm->ddl[pipe].sprite[0] << DDL_SPRITE_SHIFT(0)) | |
|
- | 834 | (wm->ddl[pipe].primary << DDL_PLANE_SHIFT)); |
|
- | 835 | ||
- | 836 | I915_WRITE(DSPFW1, |
|
- | 837 | FW_WM(wm->sr.plane, SR) | |
|
- | 838 | FW_WM(wm->pipe[PIPE_B].cursor, CURSORB) | |
|
- | 839 | FW_WM_VLV(wm->pipe[PIPE_B].primary, PLANEB) | |
|
- | 840 | FW_WM_VLV(wm->pipe[PIPE_A].primary, PLANEA)); |
|
1371 | if (WARN(clock == 0, "Pixel clock is zero!\n")) |
841 | I915_WRITE(DSPFW2, |
1372 | return false; |
842 | FW_WM_VLV(wm->pipe[PIPE_A].sprite[1], SPRITEB) | |
- | 843 | FW_WM(wm->pipe[PIPE_A].cursor, CURSORA) | |
|
- | 844 | FW_WM_VLV(wm->pipe[PIPE_A].sprite[0], SPRITEA)); |
|
- | 845 | I915_WRITE(DSPFW3, |
|
- | 846 | FW_WM(wm->sr.cursor, CURSOR_SR)); |
|
1373 | 847 | ||
1374 | if (WARN(pixel_size == 0, "Pixel size is zero!\n")) |
848 | if (IS_CHERRYVIEW(dev_priv)) { |
1375 | return false; |
849 | I915_WRITE(DSPFW7_CHV, |
- | 850 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) | |
|
- | 851 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC)); |
|
- | 852 | I915_WRITE(DSPFW8_CHV, |
|
- | 853 | FW_WM_VLV(wm->pipe[PIPE_C].sprite[1], SPRITEF) | |
|
- | 854 | FW_WM_VLV(wm->pipe[PIPE_C].sprite[0], SPRITEE)); |
|
1376 | 855 | I915_WRITE(DSPFW9_CHV, |
|
- | 856 | FW_WM_VLV(wm->pipe[PIPE_C].primary, PLANEC) | |
|
- | 857 | FW_WM(wm->pipe[PIPE_C].cursor, CURSORC)); |
|
- | 858 | I915_WRITE(DSPHOWM, |
|
- | 859 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
|
- | 860 | FW_WM(wm->pipe[PIPE_C].sprite[1] >> 8, SPRITEF_HI) | |
|
- | 861 | FW_WM(wm->pipe[PIPE_C].sprite[0] >> 8, SPRITEE_HI) | |
|
- | 862 | FW_WM(wm->pipe[PIPE_C].primary >> 8, PLANEC_HI) | |
|
- | 863 | FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) | |
|
- | 864 | FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) | |
|
1377 | entries = DIV_ROUND_UP(clock, 1000) * pixel_size; |
865 | FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) | |
- | 866 | FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) | |
|
1378 | if (IS_CHERRYVIEW(dev)) |
867 | FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) | |
- | 868 | FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI)); |
|
- | 869 | } else { |
|
1379 | *prec_mult = (entries > 128) ? DRAIN_LATENCY_PRECISION_32 : |
870 | I915_WRITE(DSPFW7, |
- | 871 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) | |
|
- | 872 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC)); |
|
- | 873 | I915_WRITE(DSPHOWM, |
|
- | 874 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
|
- | 875 | FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) | |
|
- | 876 | FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) | |
|
- | 877 | FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) | |
|
- | 878 | FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) | |
|
- | 879 | FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) | |
|
- | 880 | FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI)); |
|
- | 881 | } |
|
- | 882 | ||
- | 883 | /* zero (unused) WM1 watermarks */ |
|
- | 884 | I915_WRITE(DSPFW4, 0); |
|
- | 885 | I915_WRITE(DSPFW5, 0); |
|
- | 886 | I915_WRITE(DSPFW6, 0); |
|
- | 887 | I915_WRITE(DSPHOWM1, 0); |
|
- | 888 | ||
- | 889 | POSTING_READ(DSPFW1); |
|
- | 890 | } |
|
- | 891 | ||
- | 892 | #undef FW_WM_VLV |
|
- | 893 | ||
- | 894 | enum vlv_wm_level { |
|
- | 895 | VLV_WM_LEVEL_PM2, |
|
1380 | DRAIN_LATENCY_PRECISION_16; |
896 | VLV_WM_LEVEL_PM5, |
- | 897 | VLV_WM_LEVEL_DDR_DVFS, |
|
- | 898 | }; |
|
- | 899 | ||
- | 900 | /* latency must be in 0.1us units. */ |
|
- | 901 | static unsigned int vlv_wm_method2(unsigned int pixel_rate, |
|
- | 902 | unsigned int pipe_htotal, |
|
- | 903 | unsigned int horiz_pixels, |
|
1381 | else |
904 | unsigned int bytes_per_pixel, |
- | 905 | unsigned int latency) |
|
1382 | *prec_mult = (entries > 128) ? DRAIN_LATENCY_PRECISION_64 : |
906 | { |
1383 | DRAIN_LATENCY_PRECISION_32; |
907 | unsigned int ret; |
1384 | *drain_latency = (64 * (*prec_mult) * 4) / entries; |
908 | |
- | 909 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); |
|
1385 | 910 | ret = (ret + 1) * horiz_pixels * bytes_per_pixel; |
|
1386 | if (*drain_latency > DRAIN_LATENCY_MASK) |
911 | ret = DIV_ROUND_UP(ret, 64); |
- | 912 | ||
- | 913 | return ret; |
|
- | 914 | } |
|
- | 915 | ||
- | 916 | static void vlv_setup_wm_latency(struct drm_device *dev) |
|
- | 917 | { |
|
- | 918 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
- | 919 | ||
- | 920 | /* all latencies in usec */ |
|
- | 921 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3; |
|
- | 922 | ||
1387 | *drain_latency = DRAIN_LATENCY_MASK; |
923 | dev_priv->wm.max_level = VLV_WM_LEVEL_PM2; |
- | 924 | ||
- | 925 | if (IS_CHERRYVIEW(dev_priv)) { |
|
1388 | 926 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12; |
|
- | 927 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33; |
|
- | 928 | ||
- | 929 | dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS; |
|
- | 930 | } |
|
1389 | return true; |
931 | } |
- | 932 | ||
- | 933 | static uint16_t vlv_compute_wm_level(struct intel_plane *plane, |
|
- | 934 | struct intel_crtc *crtc, |
|
- | 935 | const struct intel_plane_state *state, |
|
- | 936 | int level) |
|
1390 | } |
937 | { |
- | 938 | struct drm_i915_private *dev_priv = to_i915(plane->base.dev); |
|
- | 939 | int clock, htotal, pixel_size, width, wm; |
|
- | 940 | ||
- | 941 | if (dev_priv->wm.pri_latency[level] == 0) |
|
- | 942 | return USHRT_MAX; |
|
- | 943 | ||
- | 944 | if (!state->visible) |
|
- | 945 | return 0; |
|
- | 946 | ||
- | 947 | pixel_size = drm_format_plane_cpp(state->base.fb->pixel_format, 0); |
|
- | 948 | clock = crtc->config->base.adjusted_mode.crtc_clock; |
|
- | 949 | htotal = crtc->config->base.adjusted_mode.crtc_htotal; |
|
- | 950 | width = crtc->config->pipe_src_w; |
|
- | 951 | if (WARN_ON(htotal == 0)) |
|
- | 952 | htotal = 1; |
|
1391 | 953 | ||
- | 954 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { |
|
- | 955 | /* |
|
- | 956 | * FIXME the formula gives values that are |
|
- | 957 | * too big for the cursor FIFO, and hence we |
|
- | 958 | * would never be able to use cursors. For |
|
- | 959 | * now just hardcode the watermark. |
|
- | 960 | */ |
|
- | 961 | wm = 63; |
|
1392 | /* |
962 | } else { |
1393 | * Update drain latency registers of memory arbiter |
963 | wm = vlv_wm_method2(clock, htotal, width, pixel_size, |
1394 | * |
964 | dev_priv->wm.pri_latency[level] * 10); |
1395 | * Valleyview SoC has a new memory arbiter and needs drain latency registers |
965 | } |
1396 | * to be programmed. Each plane has a drain latency multiplier and a drain |
966 | |
1397 | * latency value. |
- | |
1398 | */ |
967 | return min_t(int, wm, USHRT_MAX); |
1399 | 968 | } |
|
1400 | static void vlv_update_drain_latency(struct drm_crtc *crtc) |
969 | |
1401 | { |
- | |
1402 | struct drm_device *dev = crtc->dev; |
- | |
1403 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
1404 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | |
1405 | int pixel_size; |
- | |
1406 | int drain_latency; |
- | |
- | 970 | static void vlv_compute_fifo(struct intel_crtc *crtc) |
|
1407 | enum pipe pipe = intel_crtc->pipe; |
971 | { |
1408 | int plane_prec, prec_mult, plane_dl; |
972 | struct drm_device *dev = crtc->base.dev; |
- | 973 | struct vlv_wm_state *wm_state = &crtc->wm_state; |
|
- | 974 | struct intel_plane *plane; |
|
1409 | const int high_precision = IS_CHERRYVIEW(dev) ? |
975 | unsigned int total_rate = 0; |
- | 976 | const int fifo_size = 512 - 1; |
|
- | 977 | int fifo_extra, fifo_left = fifo_size; |
|
- | 978 | ||
- | 979 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
|
- | 980 | struct intel_plane_state *state = |
|
1410 | DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_64; |
981 | to_intel_plane_state(plane->base.state); |
1411 | - | ||
1412 | plane_dl = I915_READ(VLV_DDL(pipe)) & ~(DDL_PLANE_PRECISION_HIGH | |
- | |
1413 | DRAIN_LATENCY_MASK | DDL_CURSOR_PRECISION_HIGH | |
982 | |
1414 | (DRAIN_LATENCY_MASK << DDL_CURSOR_SHIFT)); |
983 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) |
1415 | - | ||
1416 | if (!intel_crtc_active(crtc)) { |
- | |
1417 | I915_WRITE(VLV_DDL(pipe), plane_dl); |
984 | continue; |
1418 | return; |
- | |
1419 | } |
- | |
1420 | 985 | ||
1421 | /* Primary plane Drain Latency */ |
- | |
1422 | pixel_size = crtc->primary->fb->bits_per_pixel / 8; /* BPP */ |
- | |
1423 | if (vlv_compute_drain_latency(crtc, pixel_size, &prec_mult, &drain_latency)) { |
- | |
1424 | plane_prec = (prec_mult == high_precision) ? |
986 | if (state->visible) { |
1425 | DDL_PLANE_PRECISION_HIGH : |
- | |
1426 | DDL_PLANE_PRECISION_LOW; |
- | |
1427 | plane_dl |= plane_prec | drain_latency; |
- | |
1428 | } |
987 | wm_state->num_active_planes++; |
1429 | 988 | total_rate += drm_format_plane_cpp(state->base.fb->pixel_format, 0); |
|
1430 | /* Cursor Drain Latency |
989 | } |
1431 | * BPP is always 4 for cursor |
- | |
1432 | */ |
990 | } |
- | 991 | ||
1433 | pixel_size = 4; |
992 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
- | 993 | struct intel_plane_state *state = |
|
1434 | 994 | to_intel_plane_state(plane->base.state); |
|
1435 | /* Program cursor DL only if it is enabled */ |
995 | unsigned int rate; |
- | 996 | ||
- | 997 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { |
|
- | 998 | plane->wm.fifo_size = 63; |
|
- | 999 | continue; |
|
- | 1000 | } |
|
- | 1001 | ||
- | 1002 | if (!state->visible) { |
|
- | 1003 | plane->wm.fifo_size = 0; |
|
- | 1004 | continue; |
|
- | 1005 | } |
|
- | 1006 | ||
- | 1007 | rate = drm_format_plane_cpp(state->base.fb->pixel_format, 0); |
|
- | 1008 | plane->wm.fifo_size = fifo_size * rate / total_rate; |
|
- | 1009 | fifo_left -= plane->wm.fifo_size; |
|
- | 1010 | } |
|
- | 1011 | ||
- | 1012 | fifo_extra = DIV_ROUND_UP(fifo_left, wm_state->num_active_planes ?: 1); |
|
- | 1013 | ||
- | 1014 | /* spread the remainder evenly */ |
|
- | 1015 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
|
- | 1016 | int plane_extra; |
|
- | 1017 | ||
- | 1018 | if (fifo_left == 0) |
|
- | 1019 | break; |
|
- | 1020 | ||
- | 1021 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) |
|
- | 1022 | continue; |
|
- | 1023 | ||
1436 | if (intel_crtc->cursor_base && |
1024 | /* give it all to the first plane if none are active */ |
1437 | vlv_compute_drain_latency(crtc, pixel_size, &prec_mult, &drain_latency)) { |
1025 | if (plane->wm.fifo_size == 0 && |
1438 | plane_prec = (prec_mult == high_precision) ? |
- | |
1439 | DDL_CURSOR_PRECISION_HIGH : |
- | |
1440 | DDL_CURSOR_PRECISION_LOW; |
1026 | wm_state->num_active_planes) |
1441 | plane_dl |= plane_prec | (drain_latency << DDL_CURSOR_SHIFT); |
- | |
1442 | } |
- | |
1443 | - | ||
1444 | I915_WRITE(VLV_DDL(pipe), plane_dl); |
- | |
1445 | } |
1027 | continue; |
- | 1028 | ||
- | 1029 | plane_extra = min(fifo_extra, fifo_left); |
|
- | 1030 | plane->wm.fifo_size += plane_extra; |
|
1446 | 1031 | fifo_left -= plane_extra; |
|
1447 | #define single_plane_enabled(mask) is_power_of_2(mask) |
- | |
1448 | 1032 | } |
|
1449 | static void valleyview_update_wm(struct drm_crtc *crtc) |
1033 | |
1450 | { |
- | |
1451 | struct drm_device *dev = crtc->dev; |
- | |
1452 | static const int sr_latency_ns = 12000; |
1034 | WARN_ON(fifo_left != 0); |
- | 1035 | } |
|
- | 1036 | ||
- | 1037 | static void vlv_invert_wms(struct intel_crtc *crtc) |
|
1453 | struct drm_i915_private *dev_priv = dev->dev_private; |
1038 | { |
- | 1039 | struct vlv_wm_state *wm_state = &crtc->wm_state; |
|
- | 1040 | int level; |
|
- | 1041 | ||
1454 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; |
1042 | for (level = 0; level < wm_state->num_levels; level++) { |
1455 | int plane_sr, cursor_sr; |
1043 | struct drm_device *dev = crtc->base.dev; |
- | 1044 | const int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1; |
|
- | 1045 | struct intel_plane *plane; |
|
1456 | int ignore_plane_sr, ignore_cursor_sr; |
1046 | |
- | 1047 | wm_state->sr[level].plane = sr_fifo_size - wm_state->sr[level].plane; |
|
- | 1048 | wm_state->sr[level].cursor = 63 - wm_state->sr[level].cursor; |
|
- | 1049 | ||
- | 1050 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
|
- | 1051 | switch (plane->base.type) { |
|
- | 1052 | int sprite; |
|
- | 1053 | case DRM_PLANE_TYPE_CURSOR: |
|
1457 | unsigned int enabled = 0; |
1054 | wm_state->wm[level].cursor = plane->wm.fifo_size - |
- | 1055 | wm_state->wm[level].cursor; |
|
1458 | bool cxsr_enabled; |
1056 | break; |
1459 | 1057 | case DRM_PLANE_TYPE_PRIMARY: |
|
1460 | vlv_update_drain_latency(crtc); |
1058 | wm_state->wm[level].primary = plane->wm.fifo_size - |
1461 | 1059 | wm_state->wm[level].primary; |
|
- | 1060 | break; |
|
- | 1061 | case DRM_PLANE_TYPE_OVERLAY: |
|
1462 | if (g4x_compute_wm0(dev, PIPE_A, |
1062 | sprite = plane->plane; |
- | 1063 | wm_state->wm[level].sprite[sprite] = plane->wm.fifo_size - |
|
1463 | &valleyview_wm_info, pessimal_latency_ns, |
1064 | wm_state->wm[level].sprite[sprite]; |
1464 | &valleyview_cursor_wm_info, pessimal_latency_ns, |
1065 | break; |
- | 1066 | } |
|
1465 | &planea_wm, &cursora_wm)) |
1067 | } |
- | 1068 | } |
|
1466 | enabled |= 1 << PIPE_A; |
1069 | } |
- | 1070 | ||
1467 | 1071 | static void vlv_compute_wm(struct intel_crtc *crtc) |
|
1468 | if (g4x_compute_wm0(dev, PIPE_B, |
1072 | { |
- | 1073 | struct drm_device *dev = crtc->base.dev; |
|
1469 | &valleyview_wm_info, pessimal_latency_ns, |
1074 | struct vlv_wm_state *wm_state = &crtc->wm_state; |
1470 | &valleyview_cursor_wm_info, pessimal_latency_ns, |
1075 | struct intel_plane *plane; |
1471 | &planeb_wm, &cursorb_wm)) |
1076 | int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1; |
1472 | enabled |= 1 << PIPE_B; |
1077 | int level; |
- | 1078 | ||
1473 | 1079 | memset(wm_state, 0, sizeof(*wm_state)); |
|
1474 | if (single_plane_enabled(enabled) && |
- | |
1475 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
1080 | |
1476 | sr_latency_ns, |
1081 | wm_state->cxsr = crtc->pipe != PIPE_C && crtc->wm.cxsr_allowed; |
1477 | &valleyview_wm_info, |
- | |
1478 | &valleyview_cursor_wm_info, |
1082 | wm_state->num_levels = to_i915(dev)->wm.max_level + 1; |
1479 | &plane_sr, &ignore_cursor_sr) && |
- | |
1480 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
- | |
1481 | 2*sr_latency_ns, |
- | |
1482 | &valleyview_wm_info, |
- | |
1483 | &valleyview_cursor_wm_info, |
- | |
1484 | &ignore_plane_sr, &cursor_sr)) { |
1083 | |
1485 | cxsr_enabled = true; |
- | |
1486 | } else { |
- | |
1487 | cxsr_enabled = false; |
1084 | wm_state->num_active_planes = 0; |
1488 | intel_set_memory_cxsr(dev_priv, false); |
- | |
1489 | plane_sr = cursor_sr = 0; |
- | |
- | 1085 | ||
- | 1086 | vlv_compute_fifo(crtc); |
|
- | 1087 | ||
- | 1088 | if (wm_state->num_active_planes != 1) |
|
- | 1089 | wm_state->cxsr = false; |
|
- | 1090 | ||
- | 1091 | if (wm_state->cxsr) { |
|
- | 1092 | for (level = 0; level < wm_state->num_levels; level++) { |
|
- | 1093 | wm_state->sr[level].plane = sr_fifo_size; |
|
- | 1094 | wm_state->sr[level].cursor = 63; |
|
- | 1095 | } |
|
- | 1096 | } |
|
- | 1097 | ||
- | 1098 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
|
- | 1099 | struct intel_plane_state *state = |
|
- | 1100 | to_intel_plane_state(plane->base.state); |
|
- | 1101 | ||
- | 1102 | if (!state->visible) |
|
- | 1103 | continue; |
|
- | 1104 | ||
- | 1105 | /* normal watermarks */ |
|
- | 1106 | for (level = 0; level < wm_state->num_levels; level++) { |
|
- | 1107 | int wm = vlv_compute_wm_level(plane, crtc, state, level); |
|
- | 1108 | int max_wm = plane->base.type == DRM_PLANE_TYPE_CURSOR ? 63 : 511; |
|
- | 1109 | ||
- | 1110 | /* hack */ |
|
- | 1111 | if (WARN_ON(level == 0 && wm > max_wm)) |
|
- | 1112 | wm = max_wm; |
|
- | 1113 | ||
1490 | } |
1114 | if (wm > plane->wm.fifo_size) |
- | 1115 | break; |
|
- | 1116 | ||
- | 1117 | switch (plane->base.type) { |
|
- | 1118 | int sprite; |
|
- | 1119 | case DRM_PLANE_TYPE_CURSOR: |
|
- | 1120 | wm_state->wm[level].cursor = wm; |
|
- | 1121 | break; |
|
- | 1122 | case DRM_PLANE_TYPE_PRIMARY: |
|
- | 1123 | wm_state->wm[level].primary = wm; |
|
- | 1124 | break; |
|
- | 1125 | case DRM_PLANE_TYPE_OVERLAY: |
|
- | 1126 | sprite = plane->plane; |
|
- | 1127 | wm_state->wm[level].sprite[sprite] = wm; |
|
- | 1128 | break; |
|
1491 | 1129 | } |
|
- | 1130 | } |
|
- | 1131 | ||
- | 1132 | wm_state->num_levels = level; |
|
- | 1133 | ||
- | 1134 | if (!wm_state->cxsr) |
|
- | 1135 | continue; |
|
- | 1136 | ||
- | 1137 | /* maxfifo watermarks */ |
|
- | 1138 | switch (plane->base.type) { |
|
- | 1139 | int sprite, level; |
|
- | 1140 | case DRM_PLANE_TYPE_CURSOR: |
|
- | 1141 | for (level = 0; level < wm_state->num_levels; level++) |
|
- | 1142 | wm_state->sr[level].cursor = |
|
- | 1143 | wm_state->wm[level].cursor; |
|
- | 1144 | break; |
|
1492 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, " |
1145 | case DRM_PLANE_TYPE_PRIMARY: |
- | 1146 | for (level = 0; level < wm_state->num_levels; level++) |
|
- | 1147 | wm_state->sr[level].plane = |
|
- | 1148 | min(wm_state->sr[level].plane, |
|
- | 1149 | wm_state->wm[level].primary); |
|
- | 1150 | break; |
|
- | 1151 | case DRM_PLANE_TYPE_OVERLAY: |
|
1493 | "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", |
1152 | sprite = plane->plane; |
1494 | planea_wm, cursora_wm, |
1153 | for (level = 0; level < wm_state->num_levels; level++) |
1495 | planeb_wm, cursorb_wm, |
1154 | wm_state->sr[level].plane = |
1496 | plane_sr, cursor_sr); |
- | |
1497 | 1155 | min(wm_state->sr[level].plane, |
|
1498 | I915_WRITE(DSPFW1, |
1156 | wm_state->wm[level].sprite[sprite]); |
1499 | (plane_sr << DSPFW_SR_SHIFT) | |
1157 | break; |
1500 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | |
- | |
1501 | (planeb_wm << DSPFW_PLANEB_SHIFT) | |
- | |
1502 | (planea_wm << DSPFW_PLANEA_SHIFT)); |
- | |
1503 | I915_WRITE(DSPFW2, |
- | |
- | 1158 | } |
|
- | 1159 | } |
|
1504 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
1160 | |
- | 1161 | /* clear any (partially) filled invalid levels */ |
|
- | 1162 | for (level = wm_state->num_levels; level < to_i915(dev)->wm.max_level + 1; level++) { |
|
1505 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1163 | memset(&wm_state->wm[level], 0, sizeof(wm_state->wm[level])); |
1506 | I915_WRITE(DSPFW3, |
1164 | memset(&wm_state->sr[level], 0, sizeof(wm_state->sr[level])); |
1507 | (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) | |
1165 | } |
1508 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
1166 | |
- | 1167 | vlv_invert_wms(crtc); |
|
1509 | 1168 | } |
|
- | 1169 | ||
1510 | if (cxsr_enabled) |
- | |
1511 | intel_set_memory_cxsr(dev_priv, true); |
- | |
1512 | } |
- | |
1513 | 1170 | #define VLV_FIFO(plane, value) \ |
|
1514 | static void cherryview_update_wm(struct drm_crtc *crtc) |
- | |
- | 1171 | (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV) |
|
- | 1172 | ||
- | 1173 | static void vlv_pipe_set_fifo_size(struct intel_crtc *crtc) |
|
- | 1174 | { |
|
- | 1175 | struct drm_device *dev = crtc->base.dev; |
|
- | 1176 | struct drm_i915_private *dev_priv = to_i915(dev); |
|
- | 1177 | struct intel_plane *plane; |
|
- | 1178 | int sprite0_start = 0, sprite1_start = 0, fifo_size = 0; |
|
- | 1179 | ||
- | 1180 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
|
- | 1181 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { |
|
1515 | { |
1182 | WARN_ON(plane->wm.fifo_size != 63); |
- | 1183 | continue; |
|
1516 | struct drm_device *dev = crtc->dev; |
1184 | } |
- | 1185 | ||
- | 1186 | if (plane->base.type == DRM_PLANE_TYPE_PRIMARY) |
|
- | 1187 | sprite0_start = plane->wm.fifo_size; |
|
- | 1188 | else if (plane->plane == 0) |
|
1517 | static const int sr_latency_ns = 12000; |
1189 | sprite1_start = sprite0_start + plane->wm.fifo_size; |
- | 1190 | else |
|
- | 1191 | fifo_size = sprite1_start + plane->wm.fifo_size; |
|
- | 1192 | } |
|
- | 1193 | ||
- | 1194 | WARN_ON(fifo_size != 512 - 1); |
|
- | 1195 | ||
- | 1196 | DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n", |
|
- | 1197 | pipe_name(crtc->pipe), sprite0_start, |
|
- | 1198 | sprite1_start, fifo_size); |
|
1518 | struct drm_i915_private *dev_priv = dev->dev_private; |
1199 | |
- | 1200 | switch (crtc->pipe) { |
|
- | 1201 | uint32_t dsparb, dsparb2, dsparb3; |
|
- | 1202 | case PIPE_A: |
|
- | 1203 | dsparb = I915_READ(DSPARB); |
|
- | 1204 | dsparb2 = I915_READ(DSPARB2); |
|
- | 1205 | ||
- | 1206 | dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) | |
|
- | 1207 | VLV_FIFO(SPRITEB, 0xff)); |
|
- | 1208 | dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) | |
|
- | 1209 | VLV_FIFO(SPRITEB, sprite1_start)); |
|
- | 1210 | ||
1519 | int planea_wm, planeb_wm, planec_wm; |
1211 | dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) | |
- | 1212 | VLV_FIFO(SPRITEB_HI, 0x1)); |
|
- | 1213 | dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) | |
|
- | 1214 | VLV_FIFO(SPRITEB_HI, sprite1_start >> 8)); |
|
- | 1215 | ||
- | 1216 | I915_WRITE(DSPARB, dsparb); |
|
- | 1217 | I915_WRITE(DSPARB2, dsparb2); |
|
- | 1218 | break; |
|
- | 1219 | case PIPE_B: |
|
- | 1220 | dsparb = I915_READ(DSPARB); |
|
- | 1221 | dsparb2 = I915_READ(DSPARB2); |
|
- | 1222 | ||
- | 1223 | dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) | |
|
1520 | int cursora_wm, cursorb_wm, cursorc_wm; |
1224 | VLV_FIFO(SPRITED, 0xff)); |
1521 | int plane_sr, cursor_sr; |
1225 | dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) | |
1522 | int ignore_plane_sr, ignore_cursor_sr; |
1226 | VLV_FIFO(SPRITED, sprite1_start)); |
1523 | unsigned int enabled = 0; |
- | |
1524 | bool cxsr_enabled; |
- | |
1525 | - | ||
1526 | vlv_update_drain_latency(crtc); |
- | |
1527 | 1227 | ||
1528 | if (g4x_compute_wm0(dev, PIPE_A, |
- | |
1529 | &valleyview_wm_info, pessimal_latency_ns, |
- | |
1530 | &valleyview_cursor_wm_info, pessimal_latency_ns, |
- | |
1531 | &planea_wm, &cursora_wm)) |
1228 | dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) | |
1532 | enabled |= 1 << PIPE_A; |
1229 | VLV_FIFO(SPRITED_HI, 0xff)); |
1533 | - | ||
1534 | if (g4x_compute_wm0(dev, PIPE_B, |
- | |
1535 | &valleyview_wm_info, pessimal_latency_ns, |
- | |
1536 | &valleyview_cursor_wm_info, pessimal_latency_ns, |
1230 | dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) | |
1537 | &planeb_wm, &cursorb_wm)) |
- | |
1538 | enabled |= 1 << PIPE_B; |
- | |
1539 | - | ||
1540 | if (g4x_compute_wm0(dev, PIPE_C, |
- | |
1541 | &valleyview_wm_info, pessimal_latency_ns, |
- | |
1542 | &valleyview_cursor_wm_info, pessimal_latency_ns, |
- | |
1543 | &planec_wm, &cursorc_wm)) |
1231 | VLV_FIFO(SPRITED_HI, sprite1_start >> 8)); |
1544 | enabled |= 1 << PIPE_C; |
- | |
1545 | - | ||
1546 | if (single_plane_enabled(enabled) && |
- | |
1547 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
- | |
1548 | sr_latency_ns, |
1232 | |
1549 | &valleyview_wm_info, |
- | |
1550 | &valleyview_cursor_wm_info, |
- | |
1551 | &plane_sr, &ignore_cursor_sr) && |
1233 | I915_WRITE(DSPARB, dsparb); |
1552 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
1234 | I915_WRITE(DSPARB2, dsparb2); |
1553 | 2*sr_latency_ns, |
- | |
1554 | &valleyview_wm_info, |
- | |
1555 | &valleyview_cursor_wm_info, |
1235 | break; |
1556 | &ignore_plane_sr, &cursor_sr)) { |
- | |
1557 | cxsr_enabled = true; |
1236 | case PIPE_C: |
1558 | } else { |
- | |
1559 | cxsr_enabled = false; |
- | |
- | 1237 | dsparb3 = I915_READ(DSPARB3); |
|
1560 | intel_set_memory_cxsr(dev_priv, false); |
1238 | dsparb2 = I915_READ(DSPARB2); |
- | 1239 | ||
1561 | plane_sr = cursor_sr = 0; |
1240 | dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) | |
- | 1241 | VLV_FIFO(SPRITEF, 0xff)); |
|
- | 1242 | dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) | |
|
- | 1243 | VLV_FIFO(SPRITEF, sprite1_start)); |
|
- | 1244 | ||
- | 1245 | dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) | |
|
- | 1246 | VLV_FIFO(SPRITEF_HI, 0xff)); |
|
- | 1247 | dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) | |
|
- | 1248 | VLV_FIFO(SPRITEF_HI, sprite1_start >> 8)); |
|
- | 1249 | ||
- | 1250 | I915_WRITE(DSPARB3, dsparb3); |
|
1562 | } |
1251 | I915_WRITE(DSPARB2, dsparb2); |
1563 | 1252 | break; |
|
- | 1253 | default: |
|
- | 1254 | break; |
|
1564 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, " |
1255 | } |
- | 1256 | } |
|
- | 1257 | ||
1565 | "B: plane=%d, cursor=%d, C: plane=%d, cursor=%d, " |
1258 | #undef VLV_FIFO |
- | 1259 | ||
1566 | "SR: plane=%d, cursor=%d\n", |
1260 | static void vlv_merge_wm(struct drm_device *dev, |
- | 1261 | struct vlv_wm_values *wm) |
|
- | 1262 | { |
|
1567 | planea_wm, cursora_wm, |
1263 | struct intel_crtc *crtc; |
- | 1264 | int num_active_crtcs = 0; |
|
- | 1265 | ||
- | 1266 | wm->level = to_i915(dev)->wm.max_level; |
|
1568 | planeb_wm, cursorb_wm, |
1267 | wm->cxsr = true; |
- | 1268 | ||
- | 1269 | for_each_intel_crtc(dev, crtc) { |
|
- | 1270 | const struct vlv_wm_state *wm_state = &crtc->wm_state; |
|
- | 1271 | ||
- | 1272 | if (!crtc->active) |
|
- | 1273 | continue; |
|
- | 1274 | ||
- | 1275 | if (!wm_state->cxsr) |
|
- | 1276 | wm->cxsr = false; |
|
1569 | planec_wm, cursorc_wm, |
1277 | |
1570 | plane_sr, cursor_sr); |
1278 | num_active_crtcs++; |
1571 | 1279 | wm->level = min_t(int, wm->level, wm_state->num_levels - 1); |
|
1572 | I915_WRITE(DSPFW1, |
1280 | } |
1573 | (plane_sr << DSPFW_SR_SHIFT) | |
1281 | |
1574 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | |
1282 | if (num_active_crtcs != 1) |
1575 | (planeb_wm << DSPFW_PLANEB_SHIFT) | |
1283 | wm->cxsr = false; |
1576 | (planea_wm << DSPFW_PLANEA_SHIFT)); |
1284 | |
1577 | I915_WRITE(DSPFW2, |
1285 | if (num_active_crtcs > 1) |
1578 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
1286 | wm->level = VLV_WM_LEVEL_PM2; |
1579 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1287 | |
1580 | I915_WRITE(DSPFW3, |
1288 | for_each_intel_crtc(dev, crtc) { |
1581 | (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) | |
1289 | struct vlv_wm_state *wm_state = &crtc->wm_state; |
1582 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
1290 | enum pipe pipe = crtc->pipe; |
1583 | I915_WRITE(DSPFW9_CHV, |
1291 | |
1584 | (I915_READ(DSPFW9_CHV) & ~(DSPFW_PLANEC_MASK | |
1292 | if (!crtc->active) |
1585 | DSPFW_CURSORC_MASK)) | |
1293 | continue; |
1586 | (planec_wm << DSPFW_PLANEC_SHIFT) | |
1294 | |
1587 | (cursorc_wm << DSPFW_CURSORC_SHIFT)); |
1295 | wm->pipe[pipe] = wm_state->wm[wm->level]; |
1588 | 1296 | if (wm->cxsr) |
|
1589 | if (cxsr_enabled) |
1297 | wm->sr = wm_state->sr[wm->level]; |
1590 | intel_set_memory_cxsr(dev_priv, true); |
1298 | |
1591 | } |
1299 | wm->ddl[pipe].primary = DDL_PRECISION_HIGH | 2; |
1592 | 1300 | wm->ddl[pipe].sprite[0] = DDL_PRECISION_HIGH | 2; |
|
1593 | static void valleyview_update_sprite_wm(struct drm_plane *plane, |
1301 | wm->ddl[pipe].sprite[1] = DDL_PRECISION_HIGH | 2; |
1594 | struct drm_crtc *crtc, |
1302 | wm->ddl[pipe].cursor = DDL_PRECISION_HIGH | 2; |
1595 | uint32_t sprite_width, |
1303 | } |
1596 | uint32_t sprite_height, |
1304 | } |
1597 | int pixel_size, |
1305 | |
1598 | bool enabled, bool scaled) |
1306 | static void vlv_update_wm(struct drm_crtc *crtc) |
1599 | { |
1307 | { |
1600 | struct drm_device *dev = crtc->dev; |
1308 | struct drm_device *dev = crtc->dev; |
1601 | struct drm_i915_private *dev_priv = dev->dev_private; |
1309 | struct drm_i915_private *dev_priv = dev->dev_private; |
1602 | int pipe = to_intel_plane(plane)->pipe; |
1310 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
1603 | int sprite = to_intel_plane(plane)->plane; |
1311 | enum pipe pipe = intel_crtc->pipe; |
1604 | int drain_latency; |
1312 | struct vlv_wm_values wm = {}; |
1605 | int plane_prec; |
- | |
1606 | int sprite_dl; |
- | |
1607 | int prec_mult; |
- | |
1608 | const int high_precision = IS_CHERRYVIEW(dev) ? |
- | |
1609 | DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_64; |
- | |
1610 | 1313 | ||
1611 | sprite_dl = I915_READ(VLV_DDL(pipe)) & ~(DDL_SPRITE_PRECISION_HIGH(sprite) | |
1314 | vlv_compute_wm(intel_crtc); |
1612 | (DRAIN_LATENCY_MASK << DDL_SPRITE_SHIFT(sprite))); |
1315 | vlv_merge_wm(dev, &wm); |
1613 | 1316 | ||
1614 | if (enabled && vlv_compute_drain_latency(crtc, pixel_size, &prec_mult, |
1317 | if (memcmp(&dev_priv->wm.vlv, &wm, sizeof(wm)) == 0) { |
1615 | &drain_latency)) { |
- | |
1616 | plane_prec = (prec_mult == high_precision) ? |
1318 | /* FIXME should be part of crtc atomic commit */ |
1617 | DDL_SPRITE_PRECISION_HIGH(sprite) : |
- | |
1618 | DDL_SPRITE_PRECISION_LOW(sprite); |
1319 | vlv_pipe_set_fifo_size(intel_crtc); |
1619 | sprite_dl |= plane_prec | |
1320 | return; |
1620 | (drain_latency << DDL_SPRITE_SHIFT(sprite)); |
- | |
1621 | } |
1321 | } |
- | 1322 | ||
- | 1323 | if (wm.level < VLV_WM_LEVEL_DDR_DVFS && |
|
- | 1324 | dev_priv->wm.vlv.level >= VLV_WM_LEVEL_DDR_DVFS) |
|
- | 1325 | chv_set_memory_dvfs(dev_priv, false); |
|
- | 1326 | ||
- | 1327 | if (wm.level < VLV_WM_LEVEL_PM5 && |
|
1622 | 1328 | dev_priv->wm.vlv.level >= VLV_WM_LEVEL_PM5) |
|
- | 1329 | chv_set_memory_pm5(dev_priv, false); |
|
- | 1330 | ||
- | 1331 | if (!wm.cxsr && dev_priv->wm.vlv.cxsr) |
|
- | 1332 | intel_set_memory_cxsr(dev_priv, false); |
|
- | 1333 | ||
- | 1334 | /* FIXME should be part of crtc atomic commit */ |
|
- | 1335 | vlv_pipe_set_fifo_size(intel_crtc); |
|
- | 1336 | ||
- | 1337 | vlv_write_wm_values(intel_crtc, &wm); |
|
- | 1338 | ||
- | 1339 | DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, " |
|
- | 1340 | "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n", |
|
- | 1341 | pipe_name(pipe), wm.pipe[pipe].primary, wm.pipe[pipe].cursor, |
|
- | 1342 | wm.pipe[pipe].sprite[0], wm.pipe[pipe].sprite[1], |
|
- | 1343 | wm.sr.plane, wm.sr.cursor, wm.level, wm.cxsr); |
|
- | 1344 | ||
- | 1345 | if (wm.cxsr && !dev_priv->wm.vlv.cxsr) |
|
- | 1346 | intel_set_memory_cxsr(dev_priv, true); |
|
- | 1347 | ||
- | 1348 | if (wm.level >= VLV_WM_LEVEL_PM5 && |
|
- | 1349 | dev_priv->wm.vlv.level < VLV_WM_LEVEL_PM5) |
|
- | 1350 | chv_set_memory_pm5(dev_priv, true); |
|
- | 1351 | ||
- | 1352 | if (wm.level >= VLV_WM_LEVEL_DDR_DVFS && |
|
- | 1353 | dev_priv->wm.vlv.level < VLV_WM_LEVEL_DDR_DVFS) |
|
- | 1354 | chv_set_memory_dvfs(dev_priv, true); |
|
- | 1355 | ||
1623 | I915_WRITE(VLV_DDL(pipe), sprite_dl); |
1356 | dev_priv->wm.vlv = wm; |
- | 1357 | } |
|
- | 1358 | ||
1624 | } |
1359 | #define single_plane_enabled(mask) is_power_of_2(mask) |
1625 | 1360 | ||
1626 | static void g4x_update_wm(struct drm_crtc *crtc) |
1361 | static void g4x_update_wm(struct drm_crtc *crtc) |
1627 | { |
1362 | { |
1628 | struct drm_device *dev = crtc->dev; |
1363 | struct drm_device *dev = crtc->dev; |
1629 | static const int sr_latency_ns = 12000; |
1364 | static const int sr_latency_ns = 12000; |
1630 | struct drm_i915_private *dev_priv = dev->dev_private; |
1365 | struct drm_i915_private *dev_priv = dev->dev_private; |
1631 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; |
1366 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; |
1632 | int plane_sr, cursor_sr; |
1367 | int plane_sr, cursor_sr; |
1633 | unsigned int enabled = 0; |
1368 | unsigned int enabled = 0; |
1634 | bool cxsr_enabled; |
1369 | bool cxsr_enabled; |
1635 | 1370 | ||
1636 | if (g4x_compute_wm0(dev, PIPE_A, |
1371 | if (g4x_compute_wm0(dev, PIPE_A, |
1637 | &g4x_wm_info, pessimal_latency_ns, |
1372 | &g4x_wm_info, pessimal_latency_ns, |
1638 | &g4x_cursor_wm_info, pessimal_latency_ns, |
1373 | &g4x_cursor_wm_info, pessimal_latency_ns, |
1639 | &planea_wm, &cursora_wm)) |
1374 | &planea_wm, &cursora_wm)) |
1640 | enabled |= 1 << PIPE_A; |
1375 | enabled |= 1 << PIPE_A; |
1641 | 1376 | ||
1642 | if (g4x_compute_wm0(dev, PIPE_B, |
1377 | if (g4x_compute_wm0(dev, PIPE_B, |
1643 | &g4x_wm_info, pessimal_latency_ns, |
1378 | &g4x_wm_info, pessimal_latency_ns, |
1644 | &g4x_cursor_wm_info, pessimal_latency_ns, |
1379 | &g4x_cursor_wm_info, pessimal_latency_ns, |
1645 | &planeb_wm, &cursorb_wm)) |
1380 | &planeb_wm, &cursorb_wm)) |
1646 | enabled |= 1 << PIPE_B; |
1381 | enabled |= 1 << PIPE_B; |
1647 | 1382 | ||
1648 | if (single_plane_enabled(enabled) && |
1383 | if (single_plane_enabled(enabled) && |
1649 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
1384 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
1650 | sr_latency_ns, |
1385 | sr_latency_ns, |
1651 | &g4x_wm_info, |
1386 | &g4x_wm_info, |
1652 | &g4x_cursor_wm_info, |
1387 | &g4x_cursor_wm_info, |
1653 | &plane_sr, &cursor_sr)) { |
1388 | &plane_sr, &cursor_sr)) { |
1654 | cxsr_enabled = true; |
1389 | cxsr_enabled = true; |
1655 | } else { |
1390 | } else { |
1656 | cxsr_enabled = false; |
1391 | cxsr_enabled = false; |
1657 | intel_set_memory_cxsr(dev_priv, false); |
1392 | intel_set_memory_cxsr(dev_priv, false); |
1658 | plane_sr = cursor_sr = 0; |
1393 | plane_sr = cursor_sr = 0; |
1659 | } |
1394 | } |
1660 | 1395 | ||
1661 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, " |
1396 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, " |
1662 | "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", |
1397 | "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", |
1663 | planea_wm, cursora_wm, |
1398 | planea_wm, cursora_wm, |
1664 | planeb_wm, cursorb_wm, |
1399 | planeb_wm, cursorb_wm, |
1665 | plane_sr, cursor_sr); |
1400 | plane_sr, cursor_sr); |
1666 | 1401 | ||
1667 | I915_WRITE(DSPFW1, |
1402 | I915_WRITE(DSPFW1, |
1668 | (plane_sr << DSPFW_SR_SHIFT) | |
1403 | FW_WM(plane_sr, SR) | |
1669 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | |
1404 | FW_WM(cursorb_wm, CURSORB) | |
1670 | (planeb_wm << DSPFW_PLANEB_SHIFT) | |
1405 | FW_WM(planeb_wm, PLANEB) | |
1671 | (planea_wm << DSPFW_PLANEA_SHIFT)); |
1406 | FW_WM(planea_wm, PLANEA)); |
1672 | I915_WRITE(DSPFW2, |
1407 | I915_WRITE(DSPFW2, |
1673 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
1408 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
1674 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1409 | FW_WM(cursora_wm, CURSORA)); |
1675 | /* HPLL off in SR has some issues on G4x... disable it */ |
1410 | /* HPLL off in SR has some issues on G4x... disable it */ |
1676 | I915_WRITE(DSPFW3, |
1411 | I915_WRITE(DSPFW3, |
1677 | (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) | |
1412 | (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) | |
1678 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
1413 | FW_WM(cursor_sr, CURSOR_SR)); |
1679 | 1414 | ||
1680 | if (cxsr_enabled) |
1415 | if (cxsr_enabled) |
1681 | intel_set_memory_cxsr(dev_priv, true); |
1416 | intel_set_memory_cxsr(dev_priv, true); |
1682 | } |
1417 | } |
1683 | 1418 | ||
1684 | static void i965_update_wm(struct drm_crtc *unused_crtc) |
1419 | static void i965_update_wm(struct drm_crtc *unused_crtc) |
1685 | { |
1420 | { |
1686 | struct drm_device *dev = unused_crtc->dev; |
1421 | struct drm_device *dev = unused_crtc->dev; |
1687 | struct drm_i915_private *dev_priv = dev->dev_private; |
1422 | struct drm_i915_private *dev_priv = dev->dev_private; |
1688 | struct drm_crtc *crtc; |
1423 | struct drm_crtc *crtc; |
1689 | int srwm = 1; |
1424 | int srwm = 1; |
1690 | int cursor_sr = 16; |
1425 | int cursor_sr = 16; |
1691 | bool cxsr_enabled; |
1426 | bool cxsr_enabled; |
1692 | 1427 | ||
1693 | /* Calc sr entries for one plane configs */ |
1428 | /* Calc sr entries for one plane configs */ |
1694 | crtc = single_enabled_crtc(dev); |
1429 | crtc = single_enabled_crtc(dev); |
1695 | if (crtc) { |
1430 | if (crtc) { |
1696 | /* self-refresh has much higher latency */ |
1431 | /* self-refresh has much higher latency */ |
1697 | static const int sr_latency_ns = 12000; |
1432 | static const int sr_latency_ns = 12000; |
1698 | const struct drm_display_mode *adjusted_mode = |
1433 | const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1699 | &to_intel_crtc(crtc)->config.adjusted_mode; |
- | |
1700 | int clock = adjusted_mode->crtc_clock; |
1434 | int clock = adjusted_mode->crtc_clock; |
1701 | int htotal = adjusted_mode->crtc_htotal; |
1435 | int htotal = adjusted_mode->crtc_htotal; |
1702 | int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w; |
1436 | int hdisplay = to_intel_crtc(crtc)->config->pipe_src_w; |
1703 | int pixel_size = crtc->primary->fb->bits_per_pixel / 8; |
1437 | int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
1704 | unsigned long line_time_us; |
1438 | unsigned long line_time_us; |
1705 | int entries; |
1439 | int entries; |
1706 | 1440 | ||
1707 | line_time_us = max(htotal * 1000 / clock, 1); |
1441 | line_time_us = max(htotal * 1000 / clock, 1); |
1708 | 1442 | ||
1709 | /* Use ns/us then divide to preserve precision */ |
1443 | /* Use ns/us then divide to preserve precision */ |
1710 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
1444 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
1711 | pixel_size * hdisplay; |
1445 | pixel_size * hdisplay; |
1712 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); |
1446 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); |
1713 | srwm = I965_FIFO_SIZE - entries; |
1447 | srwm = I965_FIFO_SIZE - entries; |
1714 | if (srwm < 0) |
1448 | if (srwm < 0) |
1715 | srwm = 1; |
1449 | srwm = 1; |
1716 | srwm &= 0x1ff; |
1450 | srwm &= 0x1ff; |
1717 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", |
1451 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", |
1718 | entries, srwm); |
1452 | entries, srwm); |
1719 | 1453 | ||
1720 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
1454 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
1721 | pixel_size * to_intel_crtc(crtc)->cursor_width; |
1455 | pixel_size * crtc->cursor->state->crtc_w; |
1722 | entries = DIV_ROUND_UP(entries, |
1456 | entries = DIV_ROUND_UP(entries, |
1723 | i965_cursor_wm_info.cacheline_size); |
1457 | i965_cursor_wm_info.cacheline_size); |
1724 | cursor_sr = i965_cursor_wm_info.fifo_size - |
1458 | cursor_sr = i965_cursor_wm_info.fifo_size - |
1725 | (entries + i965_cursor_wm_info.guard_size); |
1459 | (entries + i965_cursor_wm_info.guard_size); |
1726 | 1460 | ||
1727 | if (cursor_sr > i965_cursor_wm_info.max_wm) |
1461 | if (cursor_sr > i965_cursor_wm_info.max_wm) |
1728 | cursor_sr = i965_cursor_wm_info.max_wm; |
1462 | cursor_sr = i965_cursor_wm_info.max_wm; |
1729 | 1463 | ||
1730 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " |
1464 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " |
1731 | "cursor %d\n", srwm, cursor_sr); |
1465 | "cursor %d\n", srwm, cursor_sr); |
1732 | 1466 | ||
1733 | cxsr_enabled = true; |
1467 | cxsr_enabled = true; |
1734 | } else { |
1468 | } else { |
1735 | cxsr_enabled = false; |
1469 | cxsr_enabled = false; |
1736 | /* Turn off self refresh if both pipes are enabled */ |
1470 | /* Turn off self refresh if both pipes are enabled */ |
1737 | intel_set_memory_cxsr(dev_priv, false); |
1471 | intel_set_memory_cxsr(dev_priv, false); |
1738 | } |
1472 | } |
1739 | 1473 | ||
1740 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", |
1474 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", |
1741 | srwm); |
1475 | srwm); |
1742 | 1476 | ||
1743 | /* 965 has limitations... */ |
1477 | /* 965 has limitations... */ |
1744 | I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) | |
1478 | I915_WRITE(DSPFW1, FW_WM(srwm, SR) | |
1745 | (8 << DSPFW_CURSORB_SHIFT) | |
1479 | FW_WM(8, CURSORB) | |
1746 | (8 << DSPFW_PLANEB_SHIFT) | |
1480 | FW_WM(8, PLANEB) | |
1747 | (8 << DSPFW_PLANEA_SHIFT)); |
1481 | FW_WM(8, PLANEA)); |
1748 | I915_WRITE(DSPFW2, (8 << DSPFW_CURSORA_SHIFT) | |
1482 | I915_WRITE(DSPFW2, FW_WM(8, CURSORA) | |
1749 | (8 << DSPFW_PLANEC_SHIFT_OLD)); |
1483 | FW_WM(8, PLANEC_OLD)); |
1750 | /* update cursor SR watermark */ |
1484 | /* update cursor SR watermark */ |
1751 | I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
1485 | I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR)); |
1752 | 1486 | ||
1753 | if (cxsr_enabled) |
1487 | if (cxsr_enabled) |
1754 | intel_set_memory_cxsr(dev_priv, true); |
1488 | intel_set_memory_cxsr(dev_priv, true); |
1755 | } |
1489 | } |
- | 1490 | ||
- | 1491 | #undef FW_WM |
|
1756 | 1492 | ||
1757 | static void i9xx_update_wm(struct drm_crtc *unused_crtc) |
1493 | static void i9xx_update_wm(struct drm_crtc *unused_crtc) |
1758 | { |
1494 | { |
1759 | struct drm_device *dev = unused_crtc->dev; |
1495 | struct drm_device *dev = unused_crtc->dev; |
1760 | struct drm_i915_private *dev_priv = dev->dev_private; |
1496 | struct drm_i915_private *dev_priv = dev->dev_private; |
1761 | const struct intel_watermark_params *wm_info; |
1497 | const struct intel_watermark_params *wm_info; |
1762 | uint32_t fwater_lo; |
1498 | uint32_t fwater_lo; |
1763 | uint32_t fwater_hi; |
1499 | uint32_t fwater_hi; |
1764 | int cwm, srwm = 1; |
1500 | int cwm, srwm = 1; |
1765 | int fifo_size; |
1501 | int fifo_size; |
1766 | int planea_wm, planeb_wm; |
1502 | int planea_wm, planeb_wm; |
1767 | struct drm_crtc *crtc, *enabled = NULL; |
1503 | struct drm_crtc *crtc, *enabled = NULL; |
1768 | 1504 | ||
1769 | if (IS_I945GM(dev)) |
1505 | if (IS_I945GM(dev)) |
1770 | wm_info = &i945_wm_info; |
1506 | wm_info = &i945_wm_info; |
1771 | else if (!IS_GEN2(dev)) |
1507 | else if (!IS_GEN2(dev)) |
1772 | wm_info = &i915_wm_info; |
1508 | wm_info = &i915_wm_info; |
1773 | else |
1509 | else |
1774 | wm_info = &i830_a_wm_info; |
1510 | wm_info = &i830_a_wm_info; |
1775 | 1511 | ||
1776 | fifo_size = dev_priv->display.get_fifo_size(dev, 0); |
1512 | fifo_size = dev_priv->display.get_fifo_size(dev, 0); |
1777 | crtc = intel_get_crtc_for_plane(dev, 0); |
1513 | crtc = intel_get_crtc_for_plane(dev, 0); |
1778 | if (intel_crtc_active(crtc)) { |
1514 | if (intel_crtc_active(crtc)) { |
1779 | const struct drm_display_mode *adjusted_mode; |
1515 | const struct drm_display_mode *adjusted_mode; |
1780 | int cpp = crtc->primary->fb->bits_per_pixel / 8; |
1516 | int cpp = crtc->primary->state->fb->bits_per_pixel / 8; |
1781 | if (IS_GEN2(dev)) |
1517 | if (IS_GEN2(dev)) |
1782 | cpp = 4; |
1518 | cpp = 4; |
1783 | 1519 | ||
1784 | adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode; |
1520 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1785 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
1521 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
1786 | wm_info, fifo_size, cpp, |
1522 | wm_info, fifo_size, cpp, |
1787 | pessimal_latency_ns); |
1523 | pessimal_latency_ns); |
1788 | enabled = crtc; |
1524 | enabled = crtc; |
1789 | } else { |
1525 | } else { |
1790 | planea_wm = fifo_size - wm_info->guard_size; |
1526 | planea_wm = fifo_size - wm_info->guard_size; |
1791 | if (planea_wm > (long)wm_info->max_wm) |
1527 | if (planea_wm > (long)wm_info->max_wm) |
1792 | planea_wm = wm_info->max_wm; |
1528 | planea_wm = wm_info->max_wm; |
1793 | } |
1529 | } |
1794 | 1530 | ||
1795 | if (IS_GEN2(dev)) |
1531 | if (IS_GEN2(dev)) |
1796 | wm_info = &i830_bc_wm_info; |
1532 | wm_info = &i830_bc_wm_info; |
1797 | 1533 | ||
1798 | fifo_size = dev_priv->display.get_fifo_size(dev, 1); |
1534 | fifo_size = dev_priv->display.get_fifo_size(dev, 1); |
1799 | crtc = intel_get_crtc_for_plane(dev, 1); |
1535 | crtc = intel_get_crtc_for_plane(dev, 1); |
1800 | if (intel_crtc_active(crtc)) { |
1536 | if (intel_crtc_active(crtc)) { |
1801 | const struct drm_display_mode *adjusted_mode; |
1537 | const struct drm_display_mode *adjusted_mode; |
1802 | int cpp = crtc->primary->fb->bits_per_pixel / 8; |
1538 | int cpp = crtc->primary->state->fb->bits_per_pixel / 8; |
1803 | if (IS_GEN2(dev)) |
1539 | if (IS_GEN2(dev)) |
1804 | cpp = 4; |
1540 | cpp = 4; |
1805 | 1541 | ||
1806 | adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode; |
1542 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1807 | planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
1543 | planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
1808 | wm_info, fifo_size, cpp, |
1544 | wm_info, fifo_size, cpp, |
1809 | pessimal_latency_ns); |
1545 | pessimal_latency_ns); |
1810 | if (enabled == NULL) |
1546 | if (enabled == NULL) |
1811 | enabled = crtc; |
1547 | enabled = crtc; |
1812 | else |
1548 | else |
1813 | enabled = NULL; |
1549 | enabled = NULL; |
1814 | } else { |
1550 | } else { |
1815 | planeb_wm = fifo_size - wm_info->guard_size; |
1551 | planeb_wm = fifo_size - wm_info->guard_size; |
1816 | if (planeb_wm > (long)wm_info->max_wm) |
1552 | if (planeb_wm > (long)wm_info->max_wm) |
1817 | planeb_wm = wm_info->max_wm; |
1553 | planeb_wm = wm_info->max_wm; |
1818 | } |
1554 | } |
1819 | 1555 | ||
1820 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); |
1556 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); |
1821 | 1557 | ||
1822 | if (IS_I915GM(dev) && enabled) { |
1558 | if (IS_I915GM(dev) && enabled) { |
1823 | struct drm_i915_gem_object *obj; |
1559 | struct drm_i915_gem_object *obj; |
1824 | 1560 | ||
1825 | obj = intel_fb_obj(enabled->primary->fb); |
1561 | obj = intel_fb_obj(enabled->primary->state->fb); |
1826 | 1562 | ||
1827 | /* self-refresh seems busted with untiled */ |
1563 | /* self-refresh seems busted with untiled */ |
1828 | if (obj->tiling_mode == I915_TILING_NONE) |
1564 | if (obj->tiling_mode == I915_TILING_NONE) |
1829 | enabled = NULL; |
1565 | enabled = NULL; |
1830 | } |
1566 | } |
1831 | 1567 | ||
1832 | /* |
1568 | /* |
1833 | * Overlay gets an aggressive default since video jitter is bad. |
1569 | * Overlay gets an aggressive default since video jitter is bad. |
1834 | */ |
1570 | */ |
1835 | cwm = 2; |
1571 | cwm = 2; |
1836 | 1572 | ||
1837 | /* Play safe and disable self-refresh before adjusting watermarks. */ |
1573 | /* Play safe and disable self-refresh before adjusting watermarks. */ |
1838 | intel_set_memory_cxsr(dev_priv, false); |
1574 | intel_set_memory_cxsr(dev_priv, false); |
1839 | 1575 | ||
1840 | /* Calc sr entries for one plane configs */ |
1576 | /* Calc sr entries for one plane configs */ |
1841 | if (HAS_FW_BLC(dev) && enabled) { |
1577 | if (HAS_FW_BLC(dev) && enabled) { |
1842 | /* self-refresh has much higher latency */ |
1578 | /* self-refresh has much higher latency */ |
1843 | static const int sr_latency_ns = 6000; |
1579 | static const int sr_latency_ns = 6000; |
1844 | const struct drm_display_mode *adjusted_mode = |
- | |
1845 | &to_intel_crtc(enabled)->config.adjusted_mode; |
1580 | const struct drm_display_mode *adjusted_mode = &to_intel_crtc(enabled)->config->base.adjusted_mode; |
1846 | int clock = adjusted_mode->crtc_clock; |
1581 | int clock = adjusted_mode->crtc_clock; |
1847 | int htotal = adjusted_mode->crtc_htotal; |
1582 | int htotal = adjusted_mode->crtc_htotal; |
1848 | int hdisplay = to_intel_crtc(enabled)->config.pipe_src_w; |
1583 | int hdisplay = to_intel_crtc(enabled)->config->pipe_src_w; |
1849 | int pixel_size = enabled->primary->fb->bits_per_pixel / 8; |
1584 | int pixel_size = enabled->primary->state->fb->bits_per_pixel / 8; |
1850 | unsigned long line_time_us; |
1585 | unsigned long line_time_us; |
1851 | int entries; |
1586 | int entries; |
1852 | 1587 | ||
1853 | line_time_us = max(htotal * 1000 / clock, 1); |
1588 | line_time_us = max(htotal * 1000 / clock, 1); |
1854 | 1589 | ||
1855 | /* Use ns/us then divide to preserve precision */ |
1590 | /* Use ns/us then divide to preserve precision */ |
1856 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
1591 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
1857 | pixel_size * hdisplay; |
1592 | pixel_size * hdisplay; |
1858 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); |
1593 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); |
1859 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); |
1594 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); |
1860 | srwm = wm_info->fifo_size - entries; |
1595 | srwm = wm_info->fifo_size - entries; |
1861 | if (srwm < 0) |
1596 | if (srwm < 0) |
1862 | srwm = 1; |
1597 | srwm = 1; |
1863 | 1598 | ||
1864 | if (IS_I945G(dev) || IS_I945GM(dev)) |
1599 | if (IS_I945G(dev) || IS_I945GM(dev)) |
1865 | I915_WRITE(FW_BLC_SELF, |
1600 | I915_WRITE(FW_BLC_SELF, |
1866 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); |
1601 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); |
1867 | else if (IS_I915GM(dev)) |
1602 | else if (IS_I915GM(dev)) |
1868 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); |
1603 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); |
1869 | } |
1604 | } |
1870 | 1605 | ||
1871 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", |
1606 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", |
1872 | planea_wm, planeb_wm, cwm, srwm); |
1607 | planea_wm, planeb_wm, cwm, srwm); |
1873 | 1608 | ||
1874 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); |
1609 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); |
1875 | fwater_hi = (cwm & 0x1f); |
1610 | fwater_hi = (cwm & 0x1f); |
1876 | 1611 | ||
1877 | /* Set request length to 8 cachelines per fetch */ |
1612 | /* Set request length to 8 cachelines per fetch */ |
1878 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); |
1613 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); |
1879 | fwater_hi = fwater_hi | (1 << 8); |
1614 | fwater_hi = fwater_hi | (1 << 8); |
1880 | 1615 | ||
1881 | I915_WRITE(FW_BLC, fwater_lo); |
1616 | I915_WRITE(FW_BLC, fwater_lo); |
1882 | I915_WRITE(FW_BLC2, fwater_hi); |
1617 | I915_WRITE(FW_BLC2, fwater_hi); |
1883 | 1618 | ||
1884 | if (enabled) |
1619 | if (enabled) |
1885 | intel_set_memory_cxsr(dev_priv, true); |
1620 | intel_set_memory_cxsr(dev_priv, true); |
1886 | } |
1621 | } |
1887 | 1622 | ||
1888 | static void i845_update_wm(struct drm_crtc *unused_crtc) |
1623 | static void i845_update_wm(struct drm_crtc *unused_crtc) |
1889 | { |
1624 | { |
1890 | struct drm_device *dev = unused_crtc->dev; |
1625 | struct drm_device *dev = unused_crtc->dev; |
1891 | struct drm_i915_private *dev_priv = dev->dev_private; |
1626 | struct drm_i915_private *dev_priv = dev->dev_private; |
1892 | struct drm_crtc *crtc; |
1627 | struct drm_crtc *crtc; |
1893 | const struct drm_display_mode *adjusted_mode; |
1628 | const struct drm_display_mode *adjusted_mode; |
1894 | uint32_t fwater_lo; |
1629 | uint32_t fwater_lo; |
1895 | int planea_wm; |
1630 | int planea_wm; |
1896 | 1631 | ||
1897 | crtc = single_enabled_crtc(dev); |
1632 | crtc = single_enabled_crtc(dev); |
1898 | if (crtc == NULL) |
1633 | if (crtc == NULL) |
1899 | return; |
1634 | return; |
1900 | 1635 | ||
1901 | adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode; |
1636 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
1902 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
1637 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
1903 | &i845_wm_info, |
1638 | &i845_wm_info, |
1904 | dev_priv->display.get_fifo_size(dev, 0), |
1639 | dev_priv->display.get_fifo_size(dev, 0), |
1905 | 4, pessimal_latency_ns); |
1640 | 4, pessimal_latency_ns); |
1906 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
1641 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
1907 | fwater_lo |= (3<<8) | planea_wm; |
1642 | fwater_lo |= (3<<8) | planea_wm; |
1908 | 1643 | ||
1909 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); |
1644 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); |
1910 | 1645 | ||
1911 | I915_WRITE(FW_BLC, fwater_lo); |
1646 | I915_WRITE(FW_BLC, fwater_lo); |
1912 | } |
1647 | } |
1913 | 1648 | ||
1914 | static uint32_t ilk_pipe_pixel_rate(struct drm_device *dev, |
- | |
1915 | struct drm_crtc *crtc) |
1649 | uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state *pipe_config) |
1916 | { |
- | |
1917 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
1650 | { |
1918 | uint32_t pixel_rate; |
1651 | uint32_t pixel_rate; |
1919 | 1652 | ||
1920 | pixel_rate = intel_crtc->config.adjusted_mode.crtc_clock; |
1653 | pixel_rate = pipe_config->base.adjusted_mode.crtc_clock; |
1921 | 1654 | ||
1922 | /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to |
1655 | /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to |
1923 | * adjust the pixel_rate here. */ |
1656 | * adjust the pixel_rate here. */ |
1924 | 1657 | ||
1925 | if (intel_crtc->config.pch_pfit.enabled) { |
1658 | if (pipe_config->pch_pfit.enabled) { |
1926 | uint64_t pipe_w, pipe_h, pfit_w, pfit_h; |
1659 | uint64_t pipe_w, pipe_h, pfit_w, pfit_h; |
- | 1660 | uint32_t pfit_size = pipe_config->pch_pfit.size; |
|
- | 1661 | ||
- | 1662 | pipe_w = pipe_config->pipe_src_w; |
|
1927 | uint32_t pfit_size = intel_crtc->config.pch_pfit.size; |
- | |
1928 | - | ||
1929 | pipe_w = intel_crtc->config.pipe_src_w; |
1663 | pipe_h = pipe_config->pipe_src_h; |
1930 | pipe_h = intel_crtc->config.pipe_src_h; |
1664 | |
1931 | pfit_w = (pfit_size >> 16) & 0xFFFF; |
1665 | pfit_w = (pfit_size >> 16) & 0xFFFF; |
1932 | pfit_h = pfit_size & 0xFFFF; |
1666 | pfit_h = pfit_size & 0xFFFF; |
1933 | if (pipe_w < pfit_w) |
1667 | if (pipe_w < pfit_w) |
1934 | pipe_w = pfit_w; |
1668 | pipe_w = pfit_w; |
1935 | if (pipe_h < pfit_h) |
1669 | if (pipe_h < pfit_h) |
1936 | pipe_h = pfit_h; |
1670 | pipe_h = pfit_h; |
1937 | 1671 | ||
1938 | pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h, |
1672 | pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h, |
1939 | pfit_w * pfit_h); |
1673 | pfit_w * pfit_h); |
1940 | } |
1674 | } |
1941 | 1675 | ||
1942 | return pixel_rate; |
1676 | return pixel_rate; |
1943 | } |
1677 | } |
1944 | 1678 | ||
1945 | /* latency must be in 0.1us units. */ |
1679 | /* latency must be in 0.1us units. */ |
1946 | static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, |
1680 | static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, |
1947 | uint32_t latency) |
1681 | uint32_t latency) |
1948 | { |
1682 | { |
1949 | uint64_t ret; |
1683 | uint64_t ret; |
1950 | 1684 | ||
1951 | if (WARN(latency == 0, "Latency value missing\n")) |
1685 | if (WARN(latency == 0, "Latency value missing\n")) |
1952 | return UINT_MAX; |
1686 | return UINT_MAX; |
1953 | 1687 | ||
1954 | ret = (uint64_t) pixel_rate * bytes_per_pixel * latency; |
1688 | ret = (uint64_t) pixel_rate * bytes_per_pixel * latency; |
1955 | ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2; |
1689 | ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2; |
1956 | 1690 | ||
1957 | return ret; |
1691 | return ret; |
1958 | } |
1692 | } |
1959 | 1693 | ||
1960 | /* latency must be in 0.1us units. */ |
1694 | /* latency must be in 0.1us units. */ |
1961 | static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, |
1695 | static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, |
1962 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, |
1696 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, |
1963 | uint32_t latency) |
1697 | uint32_t latency) |
1964 | { |
1698 | { |
1965 | uint32_t ret; |
1699 | uint32_t ret; |
1966 | 1700 | ||
1967 | if (WARN(latency == 0, "Latency value missing\n")) |
1701 | if (WARN(latency == 0, "Latency value missing\n")) |
1968 | return UINT_MAX; |
1702 | return UINT_MAX; |
1969 | 1703 | ||
1970 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); |
1704 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); |
1971 | ret = (ret + 1) * horiz_pixels * bytes_per_pixel; |
1705 | ret = (ret + 1) * horiz_pixels * bytes_per_pixel; |
1972 | ret = DIV_ROUND_UP(ret, 64) + 2; |
1706 | ret = DIV_ROUND_UP(ret, 64) + 2; |
1973 | return ret; |
1707 | return ret; |
1974 | } |
1708 | } |
1975 | 1709 | ||
1976 | static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels, |
1710 | static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels, |
1977 | uint8_t bytes_per_pixel) |
1711 | uint8_t bytes_per_pixel) |
1978 | { |
1712 | { |
1979 | return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2; |
1713 | return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2; |
1980 | } |
1714 | } |
1981 | 1715 | ||
1982 | struct skl_pipe_wm_parameters { |
1716 | struct skl_pipe_wm_parameters { |
1983 | bool active; |
1717 | bool active; |
1984 | uint32_t pipe_htotal; |
1718 | uint32_t pipe_htotal; |
1985 | uint32_t pixel_rate; /* in KHz */ |
1719 | uint32_t pixel_rate; /* in KHz */ |
1986 | struct intel_plane_wm_parameters plane[I915_MAX_PLANES]; |
1720 | struct intel_plane_wm_parameters plane[I915_MAX_PLANES]; |
1987 | struct intel_plane_wm_parameters cursor; |
- | |
1988 | }; |
- | |
1989 | - | ||
1990 | struct ilk_pipe_wm_parameters { |
- | |
1991 | bool active; |
- | |
1992 | uint32_t pipe_htotal; |
- | |
1993 | uint32_t pixel_rate; |
- | |
1994 | struct intel_plane_wm_parameters pri; |
- | |
1995 | struct intel_plane_wm_parameters spr; |
- | |
1996 | struct intel_plane_wm_parameters cur; |
- | |
1997 | }; |
1721 | }; |
1998 | 1722 | ||
1999 | struct ilk_wm_maximums { |
1723 | struct ilk_wm_maximums { |
2000 | uint16_t pri; |
1724 | uint16_t pri; |
2001 | uint16_t spr; |
1725 | uint16_t spr; |
2002 | uint16_t cur; |
1726 | uint16_t cur; |
2003 | uint16_t fbc; |
1727 | uint16_t fbc; |
2004 | }; |
1728 | }; |
2005 | 1729 | ||
2006 | /* used in computing the new watermarks state */ |
1730 | /* used in computing the new watermarks state */ |
2007 | struct intel_wm_config { |
1731 | struct intel_wm_config { |
2008 | unsigned int num_pipes_active; |
1732 | unsigned int num_pipes_active; |
2009 | bool sprites_enabled; |
1733 | bool sprites_enabled; |
2010 | bool sprites_scaled; |
1734 | bool sprites_scaled; |
2011 | }; |
1735 | }; |
2012 | 1736 | ||
2013 | /* |
1737 | /* |
2014 | * For both WM_PIPE and WM_LP. |
1738 | * For both WM_PIPE and WM_LP. |
2015 | * mem_value must be in 0.1us units. |
1739 | * mem_value must be in 0.1us units. |
2016 | */ |
1740 | */ |
2017 | static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters *params, |
1741 | static uint32_t ilk_compute_pri_wm(const struct intel_crtc_state *cstate, |
- | 1742 | const struct intel_plane_state *pstate, |
|
2018 | uint32_t mem_value, |
1743 | uint32_t mem_value, |
2019 | bool is_lp) |
1744 | bool is_lp) |
2020 | { |
1745 | { |
- | 1746 | int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0; |
|
2021 | uint32_t method1, method2; |
1747 | uint32_t method1, method2; |
2022 | 1748 | ||
2023 | if (!params->active || !params->pri.enabled) |
1749 | if (!cstate->base.active || !pstate->visible) |
2024 | return 0; |
1750 | return 0; |
2025 | - | ||
2026 | method1 = ilk_wm_method1(params->pixel_rate, |
- | |
2027 | params->pri.bytes_per_pixel, |
1751 | |
2028 | mem_value); |
1752 | method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), bpp, mem_value); |
2029 | 1753 | ||
2030 | if (!is_lp) |
1754 | if (!is_lp) |
2031 | return method1; |
1755 | return method1; |
2032 | 1756 | ||
2033 | method2 = ilk_wm_method2(params->pixel_rate, |
1757 | method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate), |
2034 | params->pipe_htotal, |
1758 | cstate->base.adjusted_mode.crtc_htotal, |
2035 | params->pri.horiz_pixels, |
1759 | drm_rect_width(&pstate->dst), |
2036 | params->pri.bytes_per_pixel, |
1760 | bpp, |
2037 | mem_value); |
1761 | mem_value); |
2038 | 1762 | ||
2039 | return min(method1, method2); |
1763 | return min(method1, method2); |
2040 | } |
1764 | } |
2041 | 1765 | ||
2042 | /* |
1766 | /* |
2043 | * For both WM_PIPE and WM_LP. |
1767 | * For both WM_PIPE and WM_LP. |
2044 | * mem_value must be in 0.1us units. |
1768 | * mem_value must be in 0.1us units. |
2045 | */ |
1769 | */ |
2046 | static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters *params, |
1770 | static uint32_t ilk_compute_spr_wm(const struct intel_crtc_state *cstate, |
- | 1771 | const struct intel_plane_state *pstate, |
|
2047 | uint32_t mem_value) |
1772 | uint32_t mem_value) |
2048 | { |
1773 | { |
- | 1774 | int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0; |
|
2049 | uint32_t method1, method2; |
1775 | uint32_t method1, method2; |
2050 | 1776 | ||
2051 | if (!params->active || !params->spr.enabled) |
1777 | if (!cstate->base.active || !pstate->visible) |
2052 | return 0; |
1778 | return 0; |
2053 | - | ||
2054 | method1 = ilk_wm_method1(params->pixel_rate, |
- | |
2055 | params->spr.bytes_per_pixel, |
1779 | |
2056 | mem_value); |
1780 | method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), bpp, mem_value); |
2057 | method2 = ilk_wm_method2(params->pixel_rate, |
1781 | method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate), |
2058 | params->pipe_htotal, |
1782 | cstate->base.adjusted_mode.crtc_htotal, |
2059 | params->spr.horiz_pixels, |
1783 | drm_rect_width(&pstate->dst), |
2060 | params->spr.bytes_per_pixel, |
1784 | bpp, |
2061 | mem_value); |
1785 | mem_value); |
2062 | return min(method1, method2); |
1786 | return min(method1, method2); |
2063 | } |
1787 | } |
2064 | 1788 | ||
2065 | /* |
1789 | /* |
2066 | * For both WM_PIPE and WM_LP. |
1790 | * For both WM_PIPE and WM_LP. |
2067 | * mem_value must be in 0.1us units. |
1791 | * mem_value must be in 0.1us units. |
2068 | */ |
1792 | */ |
2069 | static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters *params, |
1793 | static uint32_t ilk_compute_cur_wm(const struct intel_crtc_state *cstate, |
- | 1794 | const struct intel_plane_state *pstate, |
|
2070 | uint32_t mem_value) |
1795 | uint32_t mem_value) |
2071 | { |
1796 | { |
- | 1797 | int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0; |
|
- | 1798 | ||
2072 | if (!params->active || !params->cur.enabled) |
1799 | if (!cstate->base.active || !pstate->visible) |
2073 | return 0; |
1800 | return 0; |
2074 | 1801 | ||
2075 | return ilk_wm_method2(params->pixel_rate, |
1802 | return ilk_wm_method2(ilk_pipe_pixel_rate(cstate), |
2076 | params->pipe_htotal, |
1803 | cstate->base.adjusted_mode.crtc_htotal, |
2077 | params->cur.horiz_pixels, |
1804 | drm_rect_width(&pstate->dst), |
2078 | params->cur.bytes_per_pixel, |
1805 | bpp, |
2079 | mem_value); |
1806 | mem_value); |
2080 | } |
1807 | } |
2081 | 1808 | ||
2082 | /* Only for WM_LP. */ |
1809 | /* Only for WM_LP. */ |
2083 | static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters *params, |
1810 | static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate, |
- | 1811 | const struct intel_plane_state *pstate, |
|
2084 | uint32_t pri_val) |
1812 | uint32_t pri_val) |
2085 | { |
1813 | { |
- | 1814 | int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0; |
|
- | 1815 | ||
2086 | if (!params->active || !params->pri.enabled) |
1816 | if (!cstate->base.active || !pstate->visible) |
2087 | return 0; |
1817 | return 0; |
2088 | 1818 | ||
2089 | return ilk_wm_fbc(pri_val, |
- | |
2090 | params->pri.horiz_pixels, |
- | |
2091 | params->pri.bytes_per_pixel); |
1819 | return ilk_wm_fbc(pri_val, drm_rect_width(&pstate->dst), bpp); |
2092 | } |
1820 | } |
2093 | 1821 | ||
2094 | static unsigned int ilk_display_fifo_size(const struct drm_device *dev) |
1822 | static unsigned int ilk_display_fifo_size(const struct drm_device *dev) |
2095 | { |
1823 | { |
2096 | if (INTEL_INFO(dev)->gen >= 8) |
1824 | if (INTEL_INFO(dev)->gen >= 8) |
2097 | return 3072; |
1825 | return 3072; |
2098 | else if (INTEL_INFO(dev)->gen >= 7) |
1826 | else if (INTEL_INFO(dev)->gen >= 7) |
2099 | return 768; |
1827 | return 768; |
2100 | else |
1828 | else |
2101 | return 512; |
1829 | return 512; |
2102 | } |
1830 | } |
2103 | 1831 | ||
2104 | static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev, |
1832 | static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev, |
2105 | int level, bool is_sprite) |
1833 | int level, bool is_sprite) |
2106 | { |
1834 | { |
2107 | if (INTEL_INFO(dev)->gen >= 8) |
1835 | if (INTEL_INFO(dev)->gen >= 8) |
2108 | /* BDW primary/sprite plane watermarks */ |
1836 | /* BDW primary/sprite plane watermarks */ |
2109 | return level == 0 ? 255 : 2047; |
1837 | return level == 0 ? 255 : 2047; |
2110 | else if (INTEL_INFO(dev)->gen >= 7) |
1838 | else if (INTEL_INFO(dev)->gen >= 7) |
2111 | /* IVB/HSW primary/sprite plane watermarks */ |
1839 | /* IVB/HSW primary/sprite plane watermarks */ |
2112 | return level == 0 ? 127 : 1023; |
1840 | return level == 0 ? 127 : 1023; |
2113 | else if (!is_sprite) |
1841 | else if (!is_sprite) |
2114 | /* ILK/SNB primary plane watermarks */ |
1842 | /* ILK/SNB primary plane watermarks */ |
2115 | return level == 0 ? 127 : 511; |
1843 | return level == 0 ? 127 : 511; |
2116 | else |
1844 | else |
2117 | /* ILK/SNB sprite plane watermarks */ |
1845 | /* ILK/SNB sprite plane watermarks */ |
2118 | return level == 0 ? 63 : 255; |
1846 | return level == 0 ? 63 : 255; |
2119 | } |
1847 | } |
2120 | 1848 | ||
2121 | static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev, |
1849 | static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev, |
2122 | int level) |
1850 | int level) |
2123 | { |
1851 | { |
2124 | if (INTEL_INFO(dev)->gen >= 7) |
1852 | if (INTEL_INFO(dev)->gen >= 7) |
2125 | return level == 0 ? 63 : 255; |
1853 | return level == 0 ? 63 : 255; |
2126 | else |
1854 | else |
2127 | return level == 0 ? 31 : 63; |
1855 | return level == 0 ? 31 : 63; |
2128 | } |
1856 | } |
2129 | 1857 | ||
2130 | static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev) |
1858 | static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev) |
2131 | { |
1859 | { |
2132 | if (INTEL_INFO(dev)->gen >= 8) |
1860 | if (INTEL_INFO(dev)->gen >= 8) |
2133 | return 31; |
1861 | return 31; |
2134 | else |
1862 | else |
2135 | return 15; |
1863 | return 15; |
2136 | } |
1864 | } |
2137 | 1865 | ||
2138 | /* Calculate the maximum primary/sprite plane watermark */ |
1866 | /* Calculate the maximum primary/sprite plane watermark */ |
2139 | static unsigned int ilk_plane_wm_max(const struct drm_device *dev, |
1867 | static unsigned int ilk_plane_wm_max(const struct drm_device *dev, |
2140 | int level, |
1868 | int level, |
2141 | const struct intel_wm_config *config, |
1869 | const struct intel_wm_config *config, |
2142 | enum intel_ddb_partitioning ddb_partitioning, |
1870 | enum intel_ddb_partitioning ddb_partitioning, |
2143 | bool is_sprite) |
1871 | bool is_sprite) |
2144 | { |
1872 | { |
2145 | unsigned int fifo_size = ilk_display_fifo_size(dev); |
1873 | unsigned int fifo_size = ilk_display_fifo_size(dev); |
2146 | 1874 | ||
2147 | /* if sprites aren't enabled, sprites get nothing */ |
1875 | /* if sprites aren't enabled, sprites get nothing */ |
2148 | if (is_sprite && !config->sprites_enabled) |
1876 | if (is_sprite && !config->sprites_enabled) |
2149 | return 0; |
1877 | return 0; |
2150 | 1878 | ||
2151 | /* HSW allows LP1+ watermarks even with multiple pipes */ |
1879 | /* HSW allows LP1+ watermarks even with multiple pipes */ |
2152 | if (level == 0 || config->num_pipes_active > 1) { |
1880 | if (level == 0 || config->num_pipes_active > 1) { |
2153 | fifo_size /= INTEL_INFO(dev)->num_pipes; |
1881 | fifo_size /= INTEL_INFO(dev)->num_pipes; |
2154 | 1882 | ||
2155 | /* |
1883 | /* |
2156 | * For some reason the non self refresh |
1884 | * For some reason the non self refresh |
2157 | * FIFO size is only half of the self |
1885 | * FIFO size is only half of the self |
2158 | * refresh FIFO size on ILK/SNB. |
1886 | * refresh FIFO size on ILK/SNB. |
2159 | */ |
1887 | */ |
2160 | if (INTEL_INFO(dev)->gen <= 6) |
1888 | if (INTEL_INFO(dev)->gen <= 6) |
2161 | fifo_size /= 2; |
1889 | fifo_size /= 2; |
2162 | } |
1890 | } |
2163 | 1891 | ||
2164 | if (config->sprites_enabled) { |
1892 | if (config->sprites_enabled) { |
2165 | /* level 0 is always calculated with 1:1 split */ |
1893 | /* level 0 is always calculated with 1:1 split */ |
2166 | if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) { |
1894 | if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) { |
2167 | if (is_sprite) |
1895 | if (is_sprite) |
2168 | fifo_size *= 5; |
1896 | fifo_size *= 5; |
2169 | fifo_size /= 6; |
1897 | fifo_size /= 6; |
2170 | } else { |
1898 | } else { |
2171 | fifo_size /= 2; |
1899 | fifo_size /= 2; |
2172 | } |
1900 | } |
2173 | } |
1901 | } |
2174 | 1902 | ||
2175 | /* clamp to max that the registers can hold */ |
1903 | /* clamp to max that the registers can hold */ |
2176 | return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite)); |
1904 | return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite)); |
2177 | } |
1905 | } |
2178 | 1906 | ||
2179 | /* Calculate the maximum cursor plane watermark */ |
1907 | /* Calculate the maximum cursor plane watermark */ |
2180 | static unsigned int ilk_cursor_wm_max(const struct drm_device *dev, |
1908 | static unsigned int ilk_cursor_wm_max(const struct drm_device *dev, |
2181 | int level, |
1909 | int level, |
2182 | const struct intel_wm_config *config) |
1910 | const struct intel_wm_config *config) |
2183 | { |
1911 | { |
2184 | /* HSW LP1+ watermarks w/ multiple pipes */ |
1912 | /* HSW LP1+ watermarks w/ multiple pipes */ |
2185 | if (level > 0 && config->num_pipes_active > 1) |
1913 | if (level > 0 && config->num_pipes_active > 1) |
2186 | return 64; |
1914 | return 64; |
2187 | 1915 | ||
2188 | /* otherwise just report max that registers can hold */ |
1916 | /* otherwise just report max that registers can hold */ |
2189 | return ilk_cursor_wm_reg_max(dev, level); |
1917 | return ilk_cursor_wm_reg_max(dev, level); |
2190 | } |
1918 | } |
2191 | 1919 | ||
2192 | static void ilk_compute_wm_maximums(const struct drm_device *dev, |
1920 | static void ilk_compute_wm_maximums(const struct drm_device *dev, |
2193 | int level, |
1921 | int level, |
2194 | const struct intel_wm_config *config, |
1922 | const struct intel_wm_config *config, |
2195 | enum intel_ddb_partitioning ddb_partitioning, |
1923 | enum intel_ddb_partitioning ddb_partitioning, |
2196 | struct ilk_wm_maximums *max) |
1924 | struct ilk_wm_maximums *max) |
2197 | { |
1925 | { |
2198 | max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false); |
1926 | max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false); |
2199 | max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true); |
1927 | max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true); |
2200 | max->cur = ilk_cursor_wm_max(dev, level, config); |
1928 | max->cur = ilk_cursor_wm_max(dev, level, config); |
2201 | max->fbc = ilk_fbc_wm_reg_max(dev); |
1929 | max->fbc = ilk_fbc_wm_reg_max(dev); |
2202 | } |
1930 | } |
2203 | 1931 | ||
2204 | static void ilk_compute_wm_reg_maximums(struct drm_device *dev, |
1932 | static void ilk_compute_wm_reg_maximums(struct drm_device *dev, |
2205 | int level, |
1933 | int level, |
2206 | struct ilk_wm_maximums *max) |
1934 | struct ilk_wm_maximums *max) |
2207 | { |
1935 | { |
2208 | max->pri = ilk_plane_wm_reg_max(dev, level, false); |
1936 | max->pri = ilk_plane_wm_reg_max(dev, level, false); |
2209 | max->spr = ilk_plane_wm_reg_max(dev, level, true); |
1937 | max->spr = ilk_plane_wm_reg_max(dev, level, true); |
2210 | max->cur = ilk_cursor_wm_reg_max(dev, level); |
1938 | max->cur = ilk_cursor_wm_reg_max(dev, level); |
2211 | max->fbc = ilk_fbc_wm_reg_max(dev); |
1939 | max->fbc = ilk_fbc_wm_reg_max(dev); |
2212 | } |
1940 | } |
2213 | 1941 | ||
2214 | static bool ilk_validate_wm_level(int level, |
1942 | static bool ilk_validate_wm_level(int level, |
2215 | const struct ilk_wm_maximums *max, |
1943 | const struct ilk_wm_maximums *max, |
2216 | struct intel_wm_level *result) |
1944 | struct intel_wm_level *result) |
2217 | { |
1945 | { |
2218 | bool ret; |
1946 | bool ret; |
2219 | 1947 | ||
2220 | /* already determined to be invalid? */ |
1948 | /* already determined to be invalid? */ |
2221 | if (!result->enable) |
1949 | if (!result->enable) |
2222 | return false; |
1950 | return false; |
2223 | 1951 | ||
2224 | result->enable = result->pri_val <= max->pri && |
1952 | result->enable = result->pri_val <= max->pri && |
2225 | result->spr_val <= max->spr && |
1953 | result->spr_val <= max->spr && |
2226 | result->cur_val <= max->cur; |
1954 | result->cur_val <= max->cur; |
2227 | 1955 | ||
2228 | ret = result->enable; |
1956 | ret = result->enable; |
2229 | 1957 | ||
2230 | /* |
1958 | /* |
2231 | * HACK until we can pre-compute everything, |
1959 | * HACK until we can pre-compute everything, |
2232 | * and thus fail gracefully if LP0 watermarks |
1960 | * and thus fail gracefully if LP0 watermarks |
2233 | * are exceeded... |
1961 | * are exceeded... |
2234 | */ |
1962 | */ |
2235 | if (level == 0 && !result->enable) { |
1963 | if (level == 0 && !result->enable) { |
2236 | if (result->pri_val > max->pri) |
1964 | if (result->pri_val > max->pri) |
2237 | DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n", |
1965 | DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n", |
2238 | level, result->pri_val, max->pri); |
1966 | level, result->pri_val, max->pri); |
2239 | if (result->spr_val > max->spr) |
1967 | if (result->spr_val > max->spr) |
2240 | DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n", |
1968 | DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n", |
2241 | level, result->spr_val, max->spr); |
1969 | level, result->spr_val, max->spr); |
2242 | if (result->cur_val > max->cur) |
1970 | if (result->cur_val > max->cur) |
2243 | DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n", |
1971 | DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n", |
2244 | level, result->cur_val, max->cur); |
1972 | level, result->cur_val, max->cur); |
2245 | 1973 | ||
2246 | result->pri_val = min_t(uint32_t, result->pri_val, max->pri); |
1974 | result->pri_val = min_t(uint32_t, result->pri_val, max->pri); |
2247 | result->spr_val = min_t(uint32_t, result->spr_val, max->spr); |
1975 | result->spr_val = min_t(uint32_t, result->spr_val, max->spr); |
2248 | result->cur_val = min_t(uint32_t, result->cur_val, max->cur); |
1976 | result->cur_val = min_t(uint32_t, result->cur_val, max->cur); |
2249 | result->enable = true; |
1977 | result->enable = true; |
2250 | } |
1978 | } |
2251 | 1979 | ||
2252 | return ret; |
1980 | return ret; |
2253 | } |
1981 | } |
2254 | 1982 | ||
2255 | static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv, |
1983 | static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv, |
- | 1984 | const struct intel_crtc *intel_crtc, |
|
2256 | int level, |
1985 | int level, |
2257 | const struct ilk_pipe_wm_parameters *p, |
1986 | struct intel_crtc_state *cstate, |
2258 | struct intel_wm_level *result) |
1987 | struct intel_wm_level *result) |
2259 | { |
1988 | { |
- | 1989 | struct intel_plane *intel_plane; |
|
2260 | uint16_t pri_latency = dev_priv->wm.pri_latency[level]; |
1990 | uint16_t pri_latency = dev_priv->wm.pri_latency[level]; |
2261 | uint16_t spr_latency = dev_priv->wm.spr_latency[level]; |
1991 | uint16_t spr_latency = dev_priv->wm.spr_latency[level]; |
2262 | uint16_t cur_latency = dev_priv->wm.cur_latency[level]; |
1992 | uint16_t cur_latency = dev_priv->wm.cur_latency[level]; |
2263 | 1993 | ||
2264 | /* WM1+ latency values stored in 0.5us units */ |
1994 | /* WM1+ latency values stored in 0.5us units */ |
2265 | if (level > 0) { |
1995 | if (level > 0) { |
2266 | pri_latency *= 5; |
1996 | pri_latency *= 5; |
2267 | spr_latency *= 5; |
1997 | spr_latency *= 5; |
2268 | cur_latency *= 5; |
1998 | cur_latency *= 5; |
2269 | } |
1999 | } |
- | 2000 | ||
- | 2001 | for_each_intel_plane_on_crtc(dev_priv->dev, intel_crtc, intel_plane) { |
|
- | 2002 | struct intel_plane_state *pstate = |
|
- | 2003 | to_intel_plane_state(intel_plane->base.state); |
|
- | 2004 | ||
- | 2005 | switch (intel_plane->base.type) { |
|
2270 | 2006 | case DRM_PLANE_TYPE_PRIMARY: |
|
- | 2007 | result->pri_val = ilk_compute_pri_wm(cstate, pstate, |
|
- | 2008 | pri_latency, |
|
2271 | result->pri_val = ilk_compute_pri_wm(p, pri_latency, level); |
2009 | level); |
- | 2010 | result->fbc_val = ilk_compute_fbc_wm(cstate, pstate, |
|
- | 2011 | result->pri_val); |
|
- | 2012 | break; |
|
2272 | result->spr_val = ilk_compute_spr_wm(p, spr_latency); |
2013 | case DRM_PLANE_TYPE_OVERLAY: |
- | 2014 | result->spr_val = ilk_compute_spr_wm(cstate, pstate, |
|
- | 2015 | spr_latency); |
|
- | 2016 | break; |
|
2273 | result->cur_val = ilk_compute_cur_wm(p, cur_latency); |
2017 | case DRM_PLANE_TYPE_CURSOR: |
- | 2018 | result->cur_val = ilk_compute_cur_wm(cstate, pstate, |
|
- | 2019 | cur_latency); |
|
- | 2020 | break; |
|
- | 2021 | } |
|
- | 2022 | } |
|
2274 | result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val); |
2023 | |
2275 | result->enable = true; |
2024 | result->enable = true; |
2276 | } |
2025 | } |
2277 | 2026 | ||
2278 | static uint32_t |
2027 | static uint32_t |
2279 | hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc) |
2028 | hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc) |
2280 | { |
2029 | { |
2281 | struct drm_i915_private *dev_priv = dev->dev_private; |
2030 | struct drm_i915_private *dev_priv = dev->dev_private; |
2282 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
2031 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
2283 | struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode; |
2032 | const struct drm_display_mode *adjusted_mode = &intel_crtc->config->base.adjusted_mode; |
2284 | u32 linetime, ips_linetime; |
2033 | u32 linetime, ips_linetime; |
2285 | 2034 | ||
2286 | if (!intel_crtc_active(crtc)) |
2035 | if (!intel_crtc->active) |
2287 | return 0; |
2036 | return 0; |
2288 | 2037 | ||
2289 | /* The WM are computed with base on how long it takes to fill a single |
2038 | /* The WM are computed with base on how long it takes to fill a single |
2290 | * row at the given clock rate, multiplied by 8. |
2039 | * row at the given clock rate, multiplied by 8. |
2291 | * */ |
2040 | * */ |
2292 | linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8, |
2041 | linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8, |
2293 | mode->crtc_clock); |
2042 | adjusted_mode->crtc_clock); |
2294 | ips_linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8, |
2043 | ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8, |
2295 | intel_ddi_get_cdclk_freq(dev_priv)); |
2044 | dev_priv->cdclk_freq); |
2296 | 2045 | ||
2297 | return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) | |
2046 | return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) | |
2298 | PIPE_WM_LINETIME_TIME(linetime); |
2047 | PIPE_WM_LINETIME_TIME(linetime); |
2299 | } |
2048 | } |
2300 | 2049 | ||
2301 | static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[8]) |
2050 | static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[8]) |
2302 | { |
2051 | { |
2303 | struct drm_i915_private *dev_priv = dev->dev_private; |
2052 | struct drm_i915_private *dev_priv = dev->dev_private; |
2304 | 2053 | ||
2305 | if (IS_GEN9(dev)) { |
2054 | if (IS_GEN9(dev)) { |
2306 | uint32_t val; |
2055 | uint32_t val; |
2307 | int ret, i; |
2056 | int ret, i; |
2308 | int level, max_level = ilk_wm_max_level(dev); |
2057 | int level, max_level = ilk_wm_max_level(dev); |
2309 | 2058 | ||
2310 | /* read the first set of memory latencies[0:3] */ |
2059 | /* read the first set of memory latencies[0:3] */ |
2311 | val = 0; /* data0 to be programmed to 0 for first set */ |
2060 | val = 0; /* data0 to be programmed to 0 for first set */ |
2312 | mutex_lock(&dev_priv->rps.hw_lock); |
2061 | mutex_lock(&dev_priv->rps.hw_lock); |
2313 | ret = sandybridge_pcode_read(dev_priv, |
2062 | ret = sandybridge_pcode_read(dev_priv, |
2314 | GEN9_PCODE_READ_MEM_LATENCY, |
2063 | GEN9_PCODE_READ_MEM_LATENCY, |
2315 | &val); |
2064 | &val); |
2316 | mutex_unlock(&dev_priv->rps.hw_lock); |
2065 | mutex_unlock(&dev_priv->rps.hw_lock); |
2317 | 2066 | ||
2318 | if (ret) { |
2067 | if (ret) { |
2319 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); |
2068 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); |
2320 | return; |
2069 | return; |
2321 | } |
2070 | } |
2322 | 2071 | ||
2323 | wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK; |
2072 | wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK; |
2324 | wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & |
2073 | wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & |
2325 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2074 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2326 | wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & |
2075 | wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & |
2327 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2076 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2328 | wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & |
2077 | wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & |
2329 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2078 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2330 | 2079 | ||
2331 | /* read the second set of memory latencies[4:7] */ |
2080 | /* read the second set of memory latencies[4:7] */ |
2332 | val = 1; /* data0 to be programmed to 1 for second set */ |
2081 | val = 1; /* data0 to be programmed to 1 for second set */ |
2333 | mutex_lock(&dev_priv->rps.hw_lock); |
2082 | mutex_lock(&dev_priv->rps.hw_lock); |
2334 | ret = sandybridge_pcode_read(dev_priv, |
2083 | ret = sandybridge_pcode_read(dev_priv, |
2335 | GEN9_PCODE_READ_MEM_LATENCY, |
2084 | GEN9_PCODE_READ_MEM_LATENCY, |
2336 | &val); |
2085 | &val); |
2337 | mutex_unlock(&dev_priv->rps.hw_lock); |
2086 | mutex_unlock(&dev_priv->rps.hw_lock); |
2338 | if (ret) { |
2087 | if (ret) { |
2339 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); |
2088 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); |
2340 | return; |
2089 | return; |
2341 | } |
2090 | } |
2342 | 2091 | ||
2343 | wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK; |
2092 | wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK; |
2344 | wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & |
2093 | wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & |
2345 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2094 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2346 | wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & |
2095 | wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & |
2347 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2096 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2348 | wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & |
2097 | wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & |
2349 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2098 | GEN9_MEM_LATENCY_LEVEL_MASK; |
2350 | 2099 | ||
2351 | /* |
2100 | /* |
- | 2101 | * WaWmMemoryReadLatency:skl |
|
- | 2102 | * |
|
2352 | * punit doesn't take into account the read latency so we need |
2103 | * punit doesn't take into account the read latency so we need |
2353 | * to add 2us to the various latency levels we retrieve from |
2104 | * to add 2us to the various latency levels we retrieve from |
2354 | * the punit. |
2105 | * the punit. |
2355 | * - W0 is a bit special in that it's the only level that |
2106 | * - W0 is a bit special in that it's the only level that |
2356 | * can't be disabled if we want to have display working, so |
2107 | * can't be disabled if we want to have display working, so |
2357 | * we always add 2us there. |
2108 | * we always add 2us there. |
2358 | * - For levels >=1, punit returns 0us latency when they are |
2109 | * - For levels >=1, punit returns 0us latency when they are |
2359 | * disabled, so we respect that and don't add 2us then |
2110 | * disabled, so we respect that and don't add 2us then |
2360 | * |
2111 | * |
2361 | * Additionally, if a level n (n > 1) has a 0us latency, all |
2112 | * Additionally, if a level n (n > 1) has a 0us latency, all |
2362 | * levels m (m >= n) need to be disabled. We make sure to |
2113 | * levels m (m >= n) need to be disabled. We make sure to |
2363 | * sanitize the values out of the punit to satisfy this |
2114 | * sanitize the values out of the punit to satisfy this |
2364 | * requirement. |
2115 | * requirement. |
2365 | */ |
2116 | */ |
2366 | wm[0] += 2; |
2117 | wm[0] += 2; |
2367 | for (level = 1; level <= max_level; level++) |
2118 | for (level = 1; level <= max_level; level++) |
2368 | if (wm[level] != 0) |
2119 | if (wm[level] != 0) |
2369 | wm[level] += 2; |
2120 | wm[level] += 2; |
2370 | else { |
2121 | else { |
2371 | for (i = level + 1; i <= max_level; i++) |
2122 | for (i = level + 1; i <= max_level; i++) |
2372 | wm[i] = 0; |
2123 | wm[i] = 0; |
2373 | 2124 | ||
2374 | break; |
2125 | break; |
2375 | } |
2126 | } |
2376 | } else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
2127 | } else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
2377 | uint64_t sskpd = I915_READ64(MCH_SSKPD); |
2128 | uint64_t sskpd = I915_READ64(MCH_SSKPD); |
2378 | 2129 | ||
2379 | wm[0] = (sskpd >> 56) & 0xFF; |
2130 | wm[0] = (sskpd >> 56) & 0xFF; |
2380 | if (wm[0] == 0) |
2131 | if (wm[0] == 0) |
2381 | wm[0] = sskpd & 0xF; |
2132 | wm[0] = sskpd & 0xF; |
2382 | wm[1] = (sskpd >> 4) & 0xFF; |
2133 | wm[1] = (sskpd >> 4) & 0xFF; |
2383 | wm[2] = (sskpd >> 12) & 0xFF; |
2134 | wm[2] = (sskpd >> 12) & 0xFF; |
2384 | wm[3] = (sskpd >> 20) & 0x1FF; |
2135 | wm[3] = (sskpd >> 20) & 0x1FF; |
2385 | wm[4] = (sskpd >> 32) & 0x1FF; |
2136 | wm[4] = (sskpd >> 32) & 0x1FF; |
2386 | } else if (INTEL_INFO(dev)->gen >= 6) { |
2137 | } else if (INTEL_INFO(dev)->gen >= 6) { |
2387 | uint32_t sskpd = I915_READ(MCH_SSKPD); |
2138 | uint32_t sskpd = I915_READ(MCH_SSKPD); |
2388 | 2139 | ||
2389 | wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK; |
2140 | wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK; |
2390 | wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK; |
2141 | wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK; |
2391 | wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK; |
2142 | wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK; |
2392 | wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK; |
2143 | wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK; |
2393 | } else if (INTEL_INFO(dev)->gen >= 5) { |
2144 | } else if (INTEL_INFO(dev)->gen >= 5) { |
2394 | uint32_t mltr = I915_READ(MLTR_ILK); |
2145 | uint32_t mltr = I915_READ(MLTR_ILK); |
2395 | 2146 | ||
2396 | /* ILK primary LP0 latency is 700 ns */ |
2147 | /* ILK primary LP0 latency is 700 ns */ |
2397 | wm[0] = 7; |
2148 | wm[0] = 7; |
2398 | wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK; |
2149 | wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK; |
2399 | wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK; |
2150 | wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK; |
2400 | } |
2151 | } |
2401 | } |
2152 | } |
2402 | 2153 | ||
2403 | static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5]) |
2154 | static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5]) |
2404 | { |
2155 | { |
2405 | /* ILK sprite LP0 latency is 1300 ns */ |
2156 | /* ILK sprite LP0 latency is 1300 ns */ |
2406 | if (INTEL_INFO(dev)->gen == 5) |
2157 | if (INTEL_INFO(dev)->gen == 5) |
2407 | wm[0] = 13; |
2158 | wm[0] = 13; |
2408 | } |
2159 | } |
2409 | 2160 | ||
2410 | static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5]) |
2161 | static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5]) |
2411 | { |
2162 | { |
2412 | /* ILK cursor LP0 latency is 1300 ns */ |
2163 | /* ILK cursor LP0 latency is 1300 ns */ |
2413 | if (INTEL_INFO(dev)->gen == 5) |
2164 | if (INTEL_INFO(dev)->gen == 5) |
2414 | wm[0] = 13; |
2165 | wm[0] = 13; |
2415 | 2166 | ||
2416 | /* WaDoubleCursorLP3Latency:ivb */ |
2167 | /* WaDoubleCursorLP3Latency:ivb */ |
2417 | if (IS_IVYBRIDGE(dev)) |
2168 | if (IS_IVYBRIDGE(dev)) |
2418 | wm[3] *= 2; |
2169 | wm[3] *= 2; |
2419 | } |
2170 | } |
2420 | 2171 | ||
2421 | int ilk_wm_max_level(const struct drm_device *dev) |
2172 | int ilk_wm_max_level(const struct drm_device *dev) |
2422 | { |
2173 | { |
2423 | /* how many WM levels are we expecting */ |
2174 | /* how many WM levels are we expecting */ |
2424 | if (IS_GEN9(dev)) |
2175 | if (INTEL_INFO(dev)->gen >= 9) |
2425 | return 7; |
2176 | return 7; |
2426 | else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
2177 | else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
2427 | return 4; |
2178 | return 4; |
2428 | else if (INTEL_INFO(dev)->gen >= 6) |
2179 | else if (INTEL_INFO(dev)->gen >= 6) |
2429 | return 3; |
2180 | return 3; |
2430 | else |
2181 | else |
2431 | return 2; |
2182 | return 2; |
2432 | } |
2183 | } |
2433 | 2184 | ||
2434 | static void intel_print_wm_latency(struct drm_device *dev, |
2185 | static void intel_print_wm_latency(struct drm_device *dev, |
2435 | const char *name, |
2186 | const char *name, |
2436 | const uint16_t wm[8]) |
2187 | const uint16_t wm[8]) |
2437 | { |
2188 | { |
2438 | int level, max_level = ilk_wm_max_level(dev); |
2189 | int level, max_level = ilk_wm_max_level(dev); |
2439 | 2190 | ||
2440 | for (level = 0; level <= max_level; level++) { |
2191 | for (level = 0; level <= max_level; level++) { |
2441 | unsigned int latency = wm[level]; |
2192 | unsigned int latency = wm[level]; |
2442 | 2193 | ||
2443 | if (latency == 0) { |
2194 | if (latency == 0) { |
2444 | DRM_ERROR("%s WM%d latency not provided\n", |
2195 | DRM_ERROR("%s WM%d latency not provided\n", |
2445 | name, level); |
2196 | name, level); |
2446 | continue; |
2197 | continue; |
2447 | } |
2198 | } |
2448 | 2199 | ||
2449 | /* |
2200 | /* |
2450 | * - latencies are in us on gen9. |
2201 | * - latencies are in us on gen9. |
2451 | * - before then, WM1+ latency values are in 0.5us units |
2202 | * - before then, WM1+ latency values are in 0.5us units |
2452 | */ |
2203 | */ |
2453 | if (IS_GEN9(dev)) |
2204 | if (IS_GEN9(dev)) |
2454 | latency *= 10; |
2205 | latency *= 10; |
2455 | else if (level > 0) |
2206 | else if (level > 0) |
2456 | latency *= 5; |
2207 | latency *= 5; |
2457 | 2208 | ||
2458 | DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n", |
2209 | DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n", |
2459 | name, level, wm[level], |
2210 | name, level, wm[level], |
2460 | latency / 10, latency % 10); |
2211 | latency / 10, latency % 10); |
2461 | } |
2212 | } |
2462 | } |
2213 | } |
2463 | 2214 | ||
2464 | static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv, |
2215 | static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv, |
2465 | uint16_t wm[5], uint16_t min) |
2216 | uint16_t wm[5], uint16_t min) |
2466 | { |
2217 | { |
2467 | int level, max_level = ilk_wm_max_level(dev_priv->dev); |
2218 | int level, max_level = ilk_wm_max_level(dev_priv->dev); |
2468 | 2219 | ||
2469 | if (wm[0] >= min) |
2220 | if (wm[0] >= min) |
2470 | return false; |
2221 | return false; |
2471 | 2222 | ||
2472 | wm[0] = max(wm[0], min); |
2223 | wm[0] = max(wm[0], min); |
2473 | for (level = 1; level <= max_level; level++) |
2224 | for (level = 1; level <= max_level; level++) |
2474 | wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5)); |
2225 | wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5)); |
2475 | 2226 | ||
2476 | return true; |
2227 | return true; |
2477 | } |
2228 | } |
2478 | 2229 | ||
2479 | static void snb_wm_latency_quirk(struct drm_device *dev) |
2230 | static void snb_wm_latency_quirk(struct drm_device *dev) |
2480 | { |
2231 | { |
2481 | struct drm_i915_private *dev_priv = dev->dev_private; |
2232 | struct drm_i915_private *dev_priv = dev->dev_private; |
2482 | bool changed; |
2233 | bool changed; |
2483 | 2234 | ||
2484 | /* |
2235 | /* |
2485 | * The BIOS provided WM memory latency values are often |
2236 | * The BIOS provided WM memory latency values are often |
2486 | * inadequate for high resolution displays. Adjust them. |
2237 | * inadequate for high resolution displays. Adjust them. |
2487 | */ |
2238 | */ |
2488 | changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) | |
2239 | changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) | |
2489 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) | |
2240 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) | |
2490 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12); |
2241 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12); |
2491 | 2242 | ||
2492 | if (!changed) |
2243 | if (!changed) |
2493 | return; |
2244 | return; |
2494 | 2245 | ||
2495 | DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n"); |
2246 | DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n"); |
2496 | intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency); |
2247 | intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency); |
2497 | intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency); |
2248 | intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency); |
2498 | intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency); |
2249 | intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency); |
2499 | } |
2250 | } |
2500 | 2251 | ||
2501 | static void ilk_setup_wm_latency(struct drm_device *dev) |
2252 | static void ilk_setup_wm_latency(struct drm_device *dev) |
2502 | { |
2253 | { |
2503 | struct drm_i915_private *dev_priv = dev->dev_private; |
2254 | struct drm_i915_private *dev_priv = dev->dev_private; |
2504 | 2255 | ||
2505 | intel_read_wm_latency(dev, dev_priv->wm.pri_latency); |
2256 | intel_read_wm_latency(dev, dev_priv->wm.pri_latency); |
2506 | 2257 | ||
2507 | memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency, |
2258 | memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency, |
2508 | sizeof(dev_priv->wm.pri_latency)); |
2259 | sizeof(dev_priv->wm.pri_latency)); |
2509 | memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency, |
2260 | memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency, |
2510 | sizeof(dev_priv->wm.pri_latency)); |
2261 | sizeof(dev_priv->wm.pri_latency)); |
2511 | 2262 | ||
2512 | intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency); |
2263 | intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency); |
2513 | intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency); |
2264 | intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency); |
2514 | 2265 | ||
2515 | intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency); |
2266 | intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency); |
2516 | intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency); |
2267 | intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency); |
2517 | intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency); |
2268 | intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency); |
2518 | 2269 | ||
2519 | if (IS_GEN6(dev)) |
2270 | if (IS_GEN6(dev)) |
2520 | snb_wm_latency_quirk(dev); |
2271 | snb_wm_latency_quirk(dev); |
2521 | } |
2272 | } |
2522 | 2273 | ||
2523 | static void skl_setup_wm_latency(struct drm_device *dev) |
2274 | static void skl_setup_wm_latency(struct drm_device *dev) |
2524 | { |
2275 | { |
2525 | struct drm_i915_private *dev_priv = dev->dev_private; |
2276 | struct drm_i915_private *dev_priv = dev->dev_private; |
2526 | 2277 | ||
2527 | intel_read_wm_latency(dev, dev_priv->wm.skl_latency); |
2278 | intel_read_wm_latency(dev, dev_priv->wm.skl_latency); |
2528 | intel_print_wm_latency(dev, "Gen9 Plane", dev_priv->wm.skl_latency); |
2279 | intel_print_wm_latency(dev, "Gen9 Plane", dev_priv->wm.skl_latency); |
2529 | } |
2280 | } |
2530 | - | ||
2531 | static void ilk_compute_wm_parameters(struct drm_crtc *crtc, |
- | |
2532 | struct ilk_pipe_wm_parameters *p) |
- | |
2533 | { |
- | |
2534 | struct drm_device *dev = crtc->dev; |
- | |
2535 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | |
2536 | enum pipe pipe = intel_crtc->pipe; |
- | |
2537 | struct drm_plane *plane; |
- | |
2538 | - | ||
2539 | if (!intel_crtc_active(crtc)) |
- | |
2540 | return; |
- | |
2541 | - | ||
2542 | p->active = true; |
- | |
2543 | p->pipe_htotal = intel_crtc->config.adjusted_mode.crtc_htotal; |
- | |
2544 | p->pixel_rate = ilk_pipe_pixel_rate(dev, crtc); |
- | |
2545 | p->pri.bytes_per_pixel = crtc->primary->fb->bits_per_pixel / 8; |
- | |
2546 | p->cur.bytes_per_pixel = 4; |
- | |
2547 | p->pri.horiz_pixels = intel_crtc->config.pipe_src_w; |
- | |
2548 | p->cur.horiz_pixels = intel_crtc->cursor_width; |
- | |
2549 | /* TODO: for now, assume primary and cursor planes are always enabled. */ |
- | |
2550 | p->pri.enabled = true; |
- | |
2551 | p->cur.enabled = true; |
- | |
2552 | - | ||
2553 | drm_for_each_legacy_plane(plane, &dev->mode_config.plane_list) { |
- | |
2554 | struct intel_plane *intel_plane = to_intel_plane(plane); |
- | |
2555 | - | ||
2556 | if (intel_plane->pipe == pipe) { |
- | |
2557 | p->spr = intel_plane->wm; |
- | |
2558 | break; |
- | |
2559 | } |
- | |
2560 | } |
- | |
2561 | } |
- | |
2562 | 2281 | ||
2563 | static void ilk_compute_wm_config(struct drm_device *dev, |
2282 | static void ilk_compute_wm_config(struct drm_device *dev, |
2564 | struct intel_wm_config *config) |
2283 | struct intel_wm_config *config) |
2565 | { |
2284 | { |
2566 | struct intel_crtc *intel_crtc; |
2285 | struct intel_crtc *intel_crtc; |
2567 | 2286 | ||
2568 | /* Compute the currently _active_ config */ |
2287 | /* Compute the currently _active_ config */ |
2569 | for_each_intel_crtc(dev, intel_crtc) { |
2288 | for_each_intel_crtc(dev, intel_crtc) { |
2570 | const struct intel_pipe_wm *wm = &intel_crtc->wm.active; |
2289 | const struct intel_pipe_wm *wm = &intel_crtc->wm.active; |
2571 | 2290 | ||
2572 | if (!wm->pipe_enabled) |
2291 | if (!wm->pipe_enabled) |
2573 | continue; |
2292 | continue; |
2574 | 2293 | ||
2575 | config->sprites_enabled |= wm->sprites_enabled; |
2294 | config->sprites_enabled |= wm->sprites_enabled; |
2576 | config->sprites_scaled |= wm->sprites_scaled; |
2295 | config->sprites_scaled |= wm->sprites_scaled; |
2577 | config->num_pipes_active++; |
2296 | config->num_pipes_active++; |
2578 | } |
2297 | } |
2579 | } |
2298 | } |
2580 | 2299 | ||
2581 | /* Compute new watermarks for the pipe */ |
2300 | /* Compute new watermarks for the pipe */ |
2582 | static bool intel_compute_pipe_wm(struct drm_crtc *crtc, |
2301 | static bool intel_compute_pipe_wm(struct intel_crtc_state *cstate, |
2583 | const struct ilk_pipe_wm_parameters *params, |
- | |
2584 | struct intel_pipe_wm *pipe_wm) |
2302 | struct intel_pipe_wm *pipe_wm) |
2585 | { |
2303 | { |
- | 2304 | struct drm_crtc *crtc = cstate->base.crtc; |
|
2586 | struct drm_device *dev = crtc->dev; |
2305 | struct drm_device *dev = crtc->dev; |
2587 | const struct drm_i915_private *dev_priv = dev->dev_private; |
2306 | const struct drm_i915_private *dev_priv = dev->dev_private; |
- | 2307 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
|
- | 2308 | struct intel_plane *intel_plane; |
|
- | 2309 | struct intel_plane_state *sprstate = NULL; |
|
2588 | int level, max_level = ilk_wm_max_level(dev); |
2310 | int level, max_level = ilk_wm_max_level(dev); |
2589 | /* LP0 watermark maximums depend on this pipe alone */ |
2311 | /* LP0 watermark maximums depend on this pipe alone */ |
2590 | struct intel_wm_config config = { |
2312 | struct intel_wm_config config = { |
2591 | .num_pipes_active = 1, |
2313 | .num_pipes_active = 1, |
2592 | .sprites_enabled = params->spr.enabled, |
- | |
2593 | .sprites_scaled = params->spr.scaled, |
- | |
2594 | }; |
2314 | }; |
2595 | struct ilk_wm_maximums max; |
2315 | struct ilk_wm_maximums max; |
- | 2316 | ||
- | 2317 | for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) { |
|
- | 2318 | if (intel_plane->base.type == DRM_PLANE_TYPE_OVERLAY) { |
|
- | 2319 | sprstate = to_intel_plane_state(intel_plane->base.state); |
|
- | 2320 | break; |
|
- | 2321 | } |
|
- | 2322 | } |
|
- | 2323 | ||
- | 2324 | config.sprites_enabled = sprstate->visible; |
|
- | 2325 | config.sprites_scaled = sprstate->visible && |
|
- | 2326 | (drm_rect_width(&sprstate->dst) != drm_rect_width(&sprstate->src) >> 16 || |
|
- | 2327 | drm_rect_height(&sprstate->dst) != drm_rect_height(&sprstate->src) >> 16); |
|
2596 | 2328 | ||
2597 | pipe_wm->pipe_enabled = params->active; |
2329 | pipe_wm->pipe_enabled = cstate->base.active; |
2598 | pipe_wm->sprites_enabled = params->spr.enabled; |
2330 | pipe_wm->sprites_enabled = sprstate->visible; |
2599 | pipe_wm->sprites_scaled = params->spr.scaled; |
2331 | pipe_wm->sprites_scaled = config.sprites_scaled; |
2600 | 2332 | ||
2601 | /* ILK/SNB: LP2+ watermarks only w/o sprites */ |
2333 | /* ILK/SNB: LP2+ watermarks only w/o sprites */ |
2602 | if (INTEL_INFO(dev)->gen <= 6 && params->spr.enabled) |
2334 | if (INTEL_INFO(dev)->gen <= 6 && sprstate->visible) |
2603 | max_level = 1; |
2335 | max_level = 1; |
2604 | 2336 | ||
2605 | /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */ |
2337 | /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */ |
2606 | if (params->spr.scaled) |
2338 | if (config.sprites_scaled) |
2607 | max_level = 0; |
2339 | max_level = 0; |
2608 | 2340 | ||
2609 | ilk_compute_wm_level(dev_priv, 0, params, &pipe_wm->wm[0]); |
2341 | ilk_compute_wm_level(dev_priv, intel_crtc, 0, cstate, &pipe_wm->wm[0]); |
2610 | 2342 | ||
2611 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
2343 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
2612 | pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc); |
2344 | pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc); |
2613 | 2345 | ||
2614 | /* LP0 watermarks always use 1/2 DDB partitioning */ |
2346 | /* LP0 watermarks always use 1/2 DDB partitioning */ |
2615 | ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max); |
2347 | ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max); |
2616 | 2348 | ||
2617 | /* At least LP0 must be valid */ |
2349 | /* At least LP0 must be valid */ |
2618 | if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) |
2350 | if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) |
2619 | return false; |
2351 | return false; |
2620 | 2352 | ||
2621 | ilk_compute_wm_reg_maximums(dev, 1, &max); |
2353 | ilk_compute_wm_reg_maximums(dev, 1, &max); |
2622 | 2354 | ||
2623 | for (level = 1; level <= max_level; level++) { |
2355 | for (level = 1; level <= max_level; level++) { |
2624 | struct intel_wm_level wm = {}; |
2356 | struct intel_wm_level wm = {}; |
2625 | 2357 | ||
2626 | ilk_compute_wm_level(dev_priv, level, params, &wm); |
2358 | ilk_compute_wm_level(dev_priv, intel_crtc, level, cstate, &wm); |
2627 | 2359 | ||
2628 | /* |
2360 | /* |
2629 | * Disable any watermark level that exceeds the |
2361 | * Disable any watermark level that exceeds the |
2630 | * register maximums since such watermarks are |
2362 | * register maximums since such watermarks are |
2631 | * always invalid. |
2363 | * always invalid. |
2632 | */ |
2364 | */ |
2633 | if (!ilk_validate_wm_level(level, &max, &wm)) |
2365 | if (!ilk_validate_wm_level(level, &max, &wm)) |
2634 | break; |
2366 | break; |
2635 | 2367 | ||
2636 | pipe_wm->wm[level] = wm; |
2368 | pipe_wm->wm[level] = wm; |
2637 | } |
2369 | } |
2638 | 2370 | ||
2639 | return true; |
2371 | return true; |
2640 | } |
2372 | } |
2641 | 2373 | ||
2642 | /* |
2374 | /* |
2643 | * Merge the watermarks from all active pipes for a specific level. |
2375 | * Merge the watermarks from all active pipes for a specific level. |
2644 | */ |
2376 | */ |
2645 | static void ilk_merge_wm_level(struct drm_device *dev, |
2377 | static void ilk_merge_wm_level(struct drm_device *dev, |
2646 | int level, |
2378 | int level, |
2647 | struct intel_wm_level *ret_wm) |
2379 | struct intel_wm_level *ret_wm) |
2648 | { |
2380 | { |
2649 | const struct intel_crtc *intel_crtc; |
2381 | const struct intel_crtc *intel_crtc; |
2650 | 2382 | ||
2651 | ret_wm->enable = true; |
2383 | ret_wm->enable = true; |
2652 | 2384 | ||
2653 | for_each_intel_crtc(dev, intel_crtc) { |
2385 | for_each_intel_crtc(dev, intel_crtc) { |
2654 | const struct intel_pipe_wm *active = &intel_crtc->wm.active; |
2386 | const struct intel_pipe_wm *active = &intel_crtc->wm.active; |
2655 | const struct intel_wm_level *wm = &active->wm[level]; |
2387 | const struct intel_wm_level *wm = &active->wm[level]; |
2656 | 2388 | ||
2657 | if (!active->pipe_enabled) |
2389 | if (!active->pipe_enabled) |
2658 | continue; |
2390 | continue; |
2659 | 2391 | ||
2660 | /* |
2392 | /* |
2661 | * The watermark values may have been used in the past, |
2393 | * The watermark values may have been used in the past, |
2662 | * so we must maintain them in the registers for some |
2394 | * so we must maintain them in the registers for some |
2663 | * time even if the level is now disabled. |
2395 | * time even if the level is now disabled. |
2664 | */ |
2396 | */ |
2665 | if (!wm->enable) |
2397 | if (!wm->enable) |
2666 | ret_wm->enable = false; |
2398 | ret_wm->enable = false; |
2667 | 2399 | ||
2668 | ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val); |
2400 | ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val); |
2669 | ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val); |
2401 | ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val); |
2670 | ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val); |
2402 | ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val); |
2671 | ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val); |
2403 | ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val); |
2672 | } |
2404 | } |
2673 | } |
2405 | } |
2674 | 2406 | ||
2675 | /* |
2407 | /* |
2676 | * Merge all low power watermarks for all active pipes. |
2408 | * Merge all low power watermarks for all active pipes. |
2677 | */ |
2409 | */ |
2678 | static void ilk_wm_merge(struct drm_device *dev, |
2410 | static void ilk_wm_merge(struct drm_device *dev, |
2679 | const struct intel_wm_config *config, |
2411 | const struct intel_wm_config *config, |
2680 | const struct ilk_wm_maximums *max, |
2412 | const struct ilk_wm_maximums *max, |
2681 | struct intel_pipe_wm *merged) |
2413 | struct intel_pipe_wm *merged) |
2682 | { |
2414 | { |
- | 2415 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
2683 | int level, max_level = ilk_wm_max_level(dev); |
2416 | int level, max_level = ilk_wm_max_level(dev); |
2684 | int last_enabled_level = max_level; |
2417 | int last_enabled_level = max_level; |
2685 | 2418 | ||
2686 | /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */ |
2419 | /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */ |
2687 | if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) && |
2420 | if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) && |
2688 | config->num_pipes_active > 1) |
2421 | config->num_pipes_active > 1) |
2689 | return; |
2422 | return; |
2690 | 2423 | ||
2691 | /* ILK: FBC WM must be disabled always */ |
2424 | /* ILK: FBC WM must be disabled always */ |
2692 | merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6; |
2425 | merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6; |
2693 | 2426 | ||
2694 | /* merge each WM1+ level */ |
2427 | /* merge each WM1+ level */ |
2695 | for (level = 1; level <= max_level; level++) { |
2428 | for (level = 1; level <= max_level; level++) { |
2696 | struct intel_wm_level *wm = &merged->wm[level]; |
2429 | struct intel_wm_level *wm = &merged->wm[level]; |
2697 | 2430 | ||
2698 | ilk_merge_wm_level(dev, level, wm); |
2431 | ilk_merge_wm_level(dev, level, wm); |
2699 | 2432 | ||
2700 | if (level > last_enabled_level) |
2433 | if (level > last_enabled_level) |
2701 | wm->enable = false; |
2434 | wm->enable = false; |
2702 | else if (!ilk_validate_wm_level(level, max, wm)) |
2435 | else if (!ilk_validate_wm_level(level, max, wm)) |
2703 | /* make sure all following levels get disabled */ |
2436 | /* make sure all following levels get disabled */ |
2704 | last_enabled_level = level - 1; |
2437 | last_enabled_level = level - 1; |
2705 | 2438 | ||
2706 | /* |
2439 | /* |
2707 | * The spec says it is preferred to disable |
2440 | * The spec says it is preferred to disable |
2708 | * FBC WMs instead of disabling a WM level. |
2441 | * FBC WMs instead of disabling a WM level. |
2709 | */ |
2442 | */ |
2710 | if (wm->fbc_val > max->fbc) { |
2443 | if (wm->fbc_val > max->fbc) { |
2711 | if (wm->enable) |
2444 | if (wm->enable) |
2712 | merged->fbc_wm_enabled = false; |
2445 | merged->fbc_wm_enabled = false; |
2713 | wm->fbc_val = 0; |
2446 | wm->fbc_val = 0; |
2714 | } |
2447 | } |
2715 | } |
2448 | } |
2716 | 2449 | ||
2717 | /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */ |
2450 | /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */ |
2718 | /* |
2451 | /* |
2719 | * FIXME this is racy. FBC might get enabled later. |
2452 | * FIXME this is racy. FBC might get enabled later. |
2720 | * What we should check here is whether FBC can be |
2453 | * What we should check here is whether FBC can be |
2721 | * enabled sometime later. |
2454 | * enabled sometime later. |
2722 | */ |
2455 | */ |
2723 | if (IS_GEN5(dev) && !merged->fbc_wm_enabled && intel_fbc_enabled(dev)) { |
2456 | if (IS_GEN5(dev) && !merged->fbc_wm_enabled && |
- | 2457 | intel_fbc_enabled(dev_priv)) { |
|
2724 | for (level = 2; level <= max_level; level++) { |
2458 | for (level = 2; level <= max_level; level++) { |
2725 | struct intel_wm_level *wm = &merged->wm[level]; |
2459 | struct intel_wm_level *wm = &merged->wm[level]; |
2726 | 2460 | ||
2727 | wm->enable = false; |
2461 | wm->enable = false; |
2728 | } |
2462 | } |
2729 | } |
2463 | } |
2730 | } |
2464 | } |
2731 | 2465 | ||
2732 | static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm) |
2466 | static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm) |
2733 | { |
2467 | { |
2734 | /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */ |
2468 | /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */ |
2735 | return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable); |
2469 | return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable); |
2736 | } |
2470 | } |
2737 | 2471 | ||
2738 | /* The value we need to program into the WM_LPx latency field */ |
2472 | /* The value we need to program into the WM_LPx latency field */ |
2739 | static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level) |
2473 | static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level) |
2740 | { |
2474 | { |
2741 | struct drm_i915_private *dev_priv = dev->dev_private; |
2475 | struct drm_i915_private *dev_priv = dev->dev_private; |
2742 | 2476 | ||
2743 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
2477 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
2744 | return 2 * level; |
2478 | return 2 * level; |
2745 | else |
2479 | else |
2746 | return dev_priv->wm.pri_latency[level]; |
2480 | return dev_priv->wm.pri_latency[level]; |
2747 | } |
2481 | } |
2748 | 2482 | ||
2749 | static void ilk_compute_wm_results(struct drm_device *dev, |
2483 | static void ilk_compute_wm_results(struct drm_device *dev, |
2750 | const struct intel_pipe_wm *merged, |
2484 | const struct intel_pipe_wm *merged, |
2751 | enum intel_ddb_partitioning partitioning, |
2485 | enum intel_ddb_partitioning partitioning, |
2752 | struct ilk_wm_values *results) |
2486 | struct ilk_wm_values *results) |
2753 | { |
2487 | { |
2754 | struct intel_crtc *intel_crtc; |
2488 | struct intel_crtc *intel_crtc; |
2755 | int level, wm_lp; |
2489 | int level, wm_lp; |
2756 | 2490 | ||
2757 | results->enable_fbc_wm = merged->fbc_wm_enabled; |
2491 | results->enable_fbc_wm = merged->fbc_wm_enabled; |
2758 | results->partitioning = partitioning; |
2492 | results->partitioning = partitioning; |
2759 | 2493 | ||
2760 | /* LP1+ register values */ |
2494 | /* LP1+ register values */ |
2761 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
2495 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
2762 | const struct intel_wm_level *r; |
2496 | const struct intel_wm_level *r; |
2763 | 2497 | ||
2764 | level = ilk_wm_lp_to_level(wm_lp, merged); |
2498 | level = ilk_wm_lp_to_level(wm_lp, merged); |
2765 | 2499 | ||
2766 | r = &merged->wm[level]; |
2500 | r = &merged->wm[level]; |
2767 | 2501 | ||
2768 | /* |
2502 | /* |
2769 | * Maintain the watermark values even if the level is |
2503 | * Maintain the watermark values even if the level is |
2770 | * disabled. Doing otherwise could cause underruns. |
2504 | * disabled. Doing otherwise could cause underruns. |
2771 | */ |
2505 | */ |
2772 | results->wm_lp[wm_lp - 1] = |
2506 | results->wm_lp[wm_lp - 1] = |
2773 | (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) | |
2507 | (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) | |
2774 | (r->pri_val << WM1_LP_SR_SHIFT) | |
2508 | (r->pri_val << WM1_LP_SR_SHIFT) | |
2775 | r->cur_val; |
2509 | r->cur_val; |
2776 | 2510 | ||
2777 | if (r->enable) |
2511 | if (r->enable) |
2778 | results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN; |
2512 | results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN; |
2779 | 2513 | ||
2780 | if (INTEL_INFO(dev)->gen >= 8) |
2514 | if (INTEL_INFO(dev)->gen >= 8) |
2781 | results->wm_lp[wm_lp - 1] |= |
2515 | results->wm_lp[wm_lp - 1] |= |
2782 | r->fbc_val << WM1_LP_FBC_SHIFT_BDW; |
2516 | r->fbc_val << WM1_LP_FBC_SHIFT_BDW; |
2783 | else |
2517 | else |
2784 | results->wm_lp[wm_lp - 1] |= |
2518 | results->wm_lp[wm_lp - 1] |= |
2785 | r->fbc_val << WM1_LP_FBC_SHIFT; |
2519 | r->fbc_val << WM1_LP_FBC_SHIFT; |
2786 | 2520 | ||
2787 | /* |
2521 | /* |
2788 | * Always set WM1S_LP_EN when spr_val != 0, even if the |
2522 | * Always set WM1S_LP_EN when spr_val != 0, even if the |
2789 | * level is disabled. Doing otherwise could cause underruns. |
2523 | * level is disabled. Doing otherwise could cause underruns. |
2790 | */ |
2524 | */ |
2791 | if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) { |
2525 | if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) { |
2792 | WARN_ON(wm_lp != 1); |
2526 | WARN_ON(wm_lp != 1); |
2793 | results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val; |
2527 | results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val; |
2794 | } else |
2528 | } else |
2795 | results->wm_lp_spr[wm_lp - 1] = r->spr_val; |
2529 | results->wm_lp_spr[wm_lp - 1] = r->spr_val; |
2796 | } |
2530 | } |
2797 | 2531 | ||
2798 | /* LP0 register values */ |
2532 | /* LP0 register values */ |
2799 | for_each_intel_crtc(dev, intel_crtc) { |
2533 | for_each_intel_crtc(dev, intel_crtc) { |
2800 | enum pipe pipe = intel_crtc->pipe; |
2534 | enum pipe pipe = intel_crtc->pipe; |
2801 | const struct intel_wm_level *r = |
2535 | const struct intel_wm_level *r = |
2802 | &intel_crtc->wm.active.wm[0]; |
2536 | &intel_crtc->wm.active.wm[0]; |
2803 | 2537 | ||
2804 | if (WARN_ON(!r->enable)) |
2538 | if (WARN_ON(!r->enable)) |
2805 | continue; |
2539 | continue; |
2806 | 2540 | ||
2807 | results->wm_linetime[pipe] = intel_crtc->wm.active.linetime; |
2541 | results->wm_linetime[pipe] = intel_crtc->wm.active.linetime; |
2808 | 2542 | ||
2809 | results->wm_pipe[pipe] = |
2543 | results->wm_pipe[pipe] = |
2810 | (r->pri_val << WM0_PIPE_PLANE_SHIFT) | |
2544 | (r->pri_val << WM0_PIPE_PLANE_SHIFT) | |
2811 | (r->spr_val << WM0_PIPE_SPRITE_SHIFT) | |
2545 | (r->spr_val << WM0_PIPE_SPRITE_SHIFT) | |
2812 | r->cur_val; |
2546 | r->cur_val; |
2813 | } |
2547 | } |
2814 | } |
2548 | } |
2815 | 2549 | ||
2816 | /* Find the result with the highest level enabled. Check for enable_fbc_wm in |
2550 | /* Find the result with the highest level enabled. Check for enable_fbc_wm in |
2817 | * case both are at the same level. Prefer r1 in case they're the same. */ |
2551 | * case both are at the same level. Prefer r1 in case they're the same. */ |
2818 | static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev, |
2552 | static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev, |
2819 | struct intel_pipe_wm *r1, |
2553 | struct intel_pipe_wm *r1, |
2820 | struct intel_pipe_wm *r2) |
2554 | struct intel_pipe_wm *r2) |
2821 | { |
2555 | { |
2822 | int level, max_level = ilk_wm_max_level(dev); |
2556 | int level, max_level = ilk_wm_max_level(dev); |
2823 | int level1 = 0, level2 = 0; |
2557 | int level1 = 0, level2 = 0; |
2824 | 2558 | ||
2825 | for (level = 1; level <= max_level; level++) { |
2559 | for (level = 1; level <= max_level; level++) { |
2826 | if (r1->wm[level].enable) |
2560 | if (r1->wm[level].enable) |
2827 | level1 = level; |
2561 | level1 = level; |
2828 | if (r2->wm[level].enable) |
2562 | if (r2->wm[level].enable) |
2829 | level2 = level; |
2563 | level2 = level; |
2830 | } |
2564 | } |
2831 | 2565 | ||
2832 | if (level1 == level2) { |
2566 | if (level1 == level2) { |
2833 | if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled) |
2567 | if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled) |
2834 | return r2; |
2568 | return r2; |
2835 | else |
2569 | else |
2836 | return r1; |
2570 | return r1; |
2837 | } else if (level1 > level2) { |
2571 | } else if (level1 > level2) { |
2838 | return r1; |
2572 | return r1; |
2839 | } else { |
2573 | } else { |
2840 | return r2; |
2574 | return r2; |
2841 | } |
2575 | } |
2842 | } |
2576 | } |
2843 | 2577 | ||
2844 | /* dirty bits used to track which watermarks need changes */ |
2578 | /* dirty bits used to track which watermarks need changes */ |
2845 | #define WM_DIRTY_PIPE(pipe) (1 << (pipe)) |
2579 | #define WM_DIRTY_PIPE(pipe) (1 << (pipe)) |
2846 | #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe))) |
2580 | #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe))) |
2847 | #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp))) |
2581 | #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp))) |
2848 | #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3)) |
2582 | #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3)) |
2849 | #define WM_DIRTY_FBC (1 << 24) |
2583 | #define WM_DIRTY_FBC (1 << 24) |
2850 | #define WM_DIRTY_DDB (1 << 25) |
2584 | #define WM_DIRTY_DDB (1 << 25) |
2851 | 2585 | ||
2852 | static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv, |
2586 | static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv, |
2853 | const struct ilk_wm_values *old, |
2587 | const struct ilk_wm_values *old, |
2854 | const struct ilk_wm_values *new) |
2588 | const struct ilk_wm_values *new) |
2855 | { |
2589 | { |
2856 | unsigned int dirty = 0; |
2590 | unsigned int dirty = 0; |
2857 | enum pipe pipe; |
2591 | enum pipe pipe; |
2858 | int wm_lp; |
2592 | int wm_lp; |
2859 | 2593 | ||
2860 | for_each_pipe(dev_priv, pipe) { |
2594 | for_each_pipe(dev_priv, pipe) { |
2861 | if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) { |
2595 | if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) { |
2862 | dirty |= WM_DIRTY_LINETIME(pipe); |
2596 | dirty |= WM_DIRTY_LINETIME(pipe); |
2863 | /* Must disable LP1+ watermarks too */ |
2597 | /* Must disable LP1+ watermarks too */ |
2864 | dirty |= WM_DIRTY_LP_ALL; |
2598 | dirty |= WM_DIRTY_LP_ALL; |
2865 | } |
2599 | } |
2866 | 2600 | ||
2867 | if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) { |
2601 | if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) { |
2868 | dirty |= WM_DIRTY_PIPE(pipe); |
2602 | dirty |= WM_DIRTY_PIPE(pipe); |
2869 | /* Must disable LP1+ watermarks too */ |
2603 | /* Must disable LP1+ watermarks too */ |
2870 | dirty |= WM_DIRTY_LP_ALL; |
2604 | dirty |= WM_DIRTY_LP_ALL; |
2871 | } |
2605 | } |
2872 | } |
2606 | } |
2873 | 2607 | ||
2874 | if (old->enable_fbc_wm != new->enable_fbc_wm) { |
2608 | if (old->enable_fbc_wm != new->enable_fbc_wm) { |
2875 | dirty |= WM_DIRTY_FBC; |
2609 | dirty |= WM_DIRTY_FBC; |
2876 | /* Must disable LP1+ watermarks too */ |
2610 | /* Must disable LP1+ watermarks too */ |
2877 | dirty |= WM_DIRTY_LP_ALL; |
2611 | dirty |= WM_DIRTY_LP_ALL; |
2878 | } |
2612 | } |
2879 | 2613 | ||
2880 | if (old->partitioning != new->partitioning) { |
2614 | if (old->partitioning != new->partitioning) { |
2881 | dirty |= WM_DIRTY_DDB; |
2615 | dirty |= WM_DIRTY_DDB; |
2882 | /* Must disable LP1+ watermarks too */ |
2616 | /* Must disable LP1+ watermarks too */ |
2883 | dirty |= WM_DIRTY_LP_ALL; |
2617 | dirty |= WM_DIRTY_LP_ALL; |
2884 | } |
2618 | } |
2885 | 2619 | ||
2886 | /* LP1+ watermarks already deemed dirty, no need to continue */ |
2620 | /* LP1+ watermarks already deemed dirty, no need to continue */ |
2887 | if (dirty & WM_DIRTY_LP_ALL) |
2621 | if (dirty & WM_DIRTY_LP_ALL) |
2888 | return dirty; |
2622 | return dirty; |
2889 | 2623 | ||
2890 | /* Find the lowest numbered LP1+ watermark in need of an update... */ |
2624 | /* Find the lowest numbered LP1+ watermark in need of an update... */ |
2891 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
2625 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
2892 | if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] || |
2626 | if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] || |
2893 | old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1]) |
2627 | old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1]) |
2894 | break; |
2628 | break; |
2895 | } |
2629 | } |
2896 | 2630 | ||
2897 | /* ...and mark it and all higher numbered LP1+ watermarks as dirty */ |
2631 | /* ...and mark it and all higher numbered LP1+ watermarks as dirty */ |
2898 | for (; wm_lp <= 3; wm_lp++) |
2632 | for (; wm_lp <= 3; wm_lp++) |
2899 | dirty |= WM_DIRTY_LP(wm_lp); |
2633 | dirty |= WM_DIRTY_LP(wm_lp); |
2900 | 2634 | ||
2901 | return dirty; |
2635 | return dirty; |
2902 | } |
2636 | } |
2903 | 2637 | ||
2904 | static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv, |
2638 | static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv, |
2905 | unsigned int dirty) |
2639 | unsigned int dirty) |
2906 | { |
2640 | { |
2907 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
2641 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
2908 | bool changed = false; |
2642 | bool changed = false; |
2909 | 2643 | ||
2910 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) { |
2644 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) { |
2911 | previous->wm_lp[2] &= ~WM1_LP_SR_EN; |
2645 | previous->wm_lp[2] &= ~WM1_LP_SR_EN; |
2912 | I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]); |
2646 | I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]); |
2913 | changed = true; |
2647 | changed = true; |
2914 | } |
2648 | } |
2915 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) { |
2649 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) { |
2916 | previous->wm_lp[1] &= ~WM1_LP_SR_EN; |
2650 | previous->wm_lp[1] &= ~WM1_LP_SR_EN; |
2917 | I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]); |
2651 | I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]); |
2918 | changed = true; |
2652 | changed = true; |
2919 | } |
2653 | } |
2920 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) { |
2654 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) { |
2921 | previous->wm_lp[0] &= ~WM1_LP_SR_EN; |
2655 | previous->wm_lp[0] &= ~WM1_LP_SR_EN; |
2922 | I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]); |
2656 | I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]); |
2923 | changed = true; |
2657 | changed = true; |
2924 | } |
2658 | } |
2925 | 2659 | ||
2926 | /* |
2660 | /* |
2927 | * Don't touch WM1S_LP_EN here. |
2661 | * Don't touch WM1S_LP_EN here. |
2928 | * Doing so could cause underruns. |
2662 | * Doing so could cause underruns. |
2929 | */ |
2663 | */ |
2930 | 2664 | ||
2931 | return changed; |
2665 | return changed; |
2932 | } |
2666 | } |
2933 | 2667 | ||
2934 | /* |
2668 | /* |
2935 | * The spec says we shouldn't write when we don't need, because every write |
2669 | * The spec says we shouldn't write when we don't need, because every write |
2936 | * causes WMs to be re-evaluated, expending some power. |
2670 | * causes WMs to be re-evaluated, expending some power. |
2937 | */ |
2671 | */ |
2938 | static void ilk_write_wm_values(struct drm_i915_private *dev_priv, |
2672 | static void ilk_write_wm_values(struct drm_i915_private *dev_priv, |
2939 | struct ilk_wm_values *results) |
2673 | struct ilk_wm_values *results) |
2940 | { |
2674 | { |
2941 | struct drm_device *dev = dev_priv->dev; |
2675 | struct drm_device *dev = dev_priv->dev; |
2942 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
2676 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
2943 | unsigned int dirty; |
2677 | unsigned int dirty; |
2944 | uint32_t val; |
2678 | uint32_t val; |
2945 | 2679 | ||
2946 | dirty = ilk_compute_wm_dirty(dev_priv, previous, results); |
2680 | dirty = ilk_compute_wm_dirty(dev_priv, previous, results); |
2947 | if (!dirty) |
2681 | if (!dirty) |
2948 | return; |
2682 | return; |
2949 | 2683 | ||
2950 | _ilk_disable_lp_wm(dev_priv, dirty); |
2684 | _ilk_disable_lp_wm(dev_priv, dirty); |
2951 | 2685 | ||
2952 | if (dirty & WM_DIRTY_PIPE(PIPE_A)) |
2686 | if (dirty & WM_DIRTY_PIPE(PIPE_A)) |
2953 | I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]); |
2687 | I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]); |
2954 | if (dirty & WM_DIRTY_PIPE(PIPE_B)) |
2688 | if (dirty & WM_DIRTY_PIPE(PIPE_B)) |
2955 | I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]); |
2689 | I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]); |
2956 | if (dirty & WM_DIRTY_PIPE(PIPE_C)) |
2690 | if (dirty & WM_DIRTY_PIPE(PIPE_C)) |
2957 | I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]); |
2691 | I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]); |
2958 | 2692 | ||
2959 | if (dirty & WM_DIRTY_LINETIME(PIPE_A)) |
2693 | if (dirty & WM_DIRTY_LINETIME(PIPE_A)) |
2960 | I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]); |
2694 | I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]); |
2961 | if (dirty & WM_DIRTY_LINETIME(PIPE_B)) |
2695 | if (dirty & WM_DIRTY_LINETIME(PIPE_B)) |
2962 | I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]); |
2696 | I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]); |
2963 | if (dirty & WM_DIRTY_LINETIME(PIPE_C)) |
2697 | if (dirty & WM_DIRTY_LINETIME(PIPE_C)) |
2964 | I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]); |
2698 | I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]); |
2965 | 2699 | ||
2966 | if (dirty & WM_DIRTY_DDB) { |
2700 | if (dirty & WM_DIRTY_DDB) { |
2967 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
2701 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
2968 | val = I915_READ(WM_MISC); |
2702 | val = I915_READ(WM_MISC); |
2969 | if (results->partitioning == INTEL_DDB_PART_1_2) |
2703 | if (results->partitioning == INTEL_DDB_PART_1_2) |
2970 | val &= ~WM_MISC_DATA_PARTITION_5_6; |
2704 | val &= ~WM_MISC_DATA_PARTITION_5_6; |
2971 | else |
2705 | else |
2972 | val |= WM_MISC_DATA_PARTITION_5_6; |
2706 | val |= WM_MISC_DATA_PARTITION_5_6; |
2973 | I915_WRITE(WM_MISC, val); |
2707 | I915_WRITE(WM_MISC, val); |
2974 | } else { |
2708 | } else { |
2975 | val = I915_READ(DISP_ARB_CTL2); |
2709 | val = I915_READ(DISP_ARB_CTL2); |
2976 | if (results->partitioning == INTEL_DDB_PART_1_2) |
2710 | if (results->partitioning == INTEL_DDB_PART_1_2) |
2977 | val &= ~DISP_DATA_PARTITION_5_6; |
2711 | val &= ~DISP_DATA_PARTITION_5_6; |
2978 | else |
2712 | else |
2979 | val |= DISP_DATA_PARTITION_5_6; |
2713 | val |= DISP_DATA_PARTITION_5_6; |
2980 | I915_WRITE(DISP_ARB_CTL2, val); |
2714 | I915_WRITE(DISP_ARB_CTL2, val); |
2981 | } |
2715 | } |
2982 | } |
2716 | } |
2983 | 2717 | ||
2984 | if (dirty & WM_DIRTY_FBC) { |
2718 | if (dirty & WM_DIRTY_FBC) { |
2985 | val = I915_READ(DISP_ARB_CTL); |
2719 | val = I915_READ(DISP_ARB_CTL); |
2986 | if (results->enable_fbc_wm) |
2720 | if (results->enable_fbc_wm) |
2987 | val &= ~DISP_FBC_WM_DIS; |
2721 | val &= ~DISP_FBC_WM_DIS; |
2988 | else |
2722 | else |
2989 | val |= DISP_FBC_WM_DIS; |
2723 | val |= DISP_FBC_WM_DIS; |
2990 | I915_WRITE(DISP_ARB_CTL, val); |
2724 | I915_WRITE(DISP_ARB_CTL, val); |
2991 | } |
2725 | } |
2992 | 2726 | ||
2993 | if (dirty & WM_DIRTY_LP(1) && |
2727 | if (dirty & WM_DIRTY_LP(1) && |
2994 | previous->wm_lp_spr[0] != results->wm_lp_spr[0]) |
2728 | previous->wm_lp_spr[0] != results->wm_lp_spr[0]) |
2995 | I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]); |
2729 | I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]); |
2996 | 2730 | ||
2997 | if (INTEL_INFO(dev)->gen >= 7) { |
2731 | if (INTEL_INFO(dev)->gen >= 7) { |
2998 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1]) |
2732 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1]) |
2999 | I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]); |
2733 | I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]); |
3000 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2]) |
2734 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2]) |
3001 | I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]); |
2735 | I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]); |
3002 | } |
2736 | } |
3003 | 2737 | ||
3004 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0]) |
2738 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0]) |
3005 | I915_WRITE(WM1_LP_ILK, results->wm_lp[0]); |
2739 | I915_WRITE(WM1_LP_ILK, results->wm_lp[0]); |
3006 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1]) |
2740 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1]) |
3007 | I915_WRITE(WM2_LP_ILK, results->wm_lp[1]); |
2741 | I915_WRITE(WM2_LP_ILK, results->wm_lp[1]); |
3008 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2]) |
2742 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2]) |
3009 | I915_WRITE(WM3_LP_ILK, results->wm_lp[2]); |
2743 | I915_WRITE(WM3_LP_ILK, results->wm_lp[2]); |
3010 | 2744 | ||
3011 | dev_priv->wm.hw = *results; |
2745 | dev_priv->wm.hw = *results; |
3012 | } |
2746 | } |
3013 | 2747 | ||
3014 | static bool ilk_disable_lp_wm(struct drm_device *dev) |
2748 | static bool ilk_disable_lp_wm(struct drm_device *dev) |
3015 | { |
2749 | { |
3016 | struct drm_i915_private *dev_priv = dev->dev_private; |
2750 | struct drm_i915_private *dev_priv = dev->dev_private; |
3017 | 2751 | ||
3018 | return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL); |
2752 | return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL); |
3019 | } |
2753 | } |
3020 | 2754 | ||
3021 | /* |
2755 | /* |
3022 | * On gen9, we need to allocate Display Data Buffer (DDB) portions to the |
2756 | * On gen9, we need to allocate Display Data Buffer (DDB) portions to the |
3023 | * different active planes. |
2757 | * different active planes. |
3024 | */ |
2758 | */ |
3025 | 2759 | ||
3026 | #define SKL_DDB_SIZE 896 /* in blocks */ |
2760 | #define SKL_DDB_SIZE 896 /* in blocks */ |
- | 2761 | #define BXT_DDB_SIZE 512 |
|
3027 | 2762 | ||
3028 | static void |
2763 | static void |
3029 | skl_ddb_get_pipe_allocation_limits(struct drm_device *dev, |
2764 | skl_ddb_get_pipe_allocation_limits(struct drm_device *dev, |
3030 | struct drm_crtc *for_crtc, |
2765 | struct drm_crtc *for_crtc, |
3031 | const struct intel_wm_config *config, |
2766 | const struct intel_wm_config *config, |
3032 | const struct skl_pipe_wm_parameters *params, |
2767 | const struct skl_pipe_wm_parameters *params, |
3033 | struct skl_ddb_entry *alloc /* out */) |
2768 | struct skl_ddb_entry *alloc /* out */) |
3034 | { |
2769 | { |
3035 | struct drm_crtc *crtc; |
2770 | struct drm_crtc *crtc; |
3036 | unsigned int pipe_size, ddb_size; |
2771 | unsigned int pipe_size, ddb_size; |
3037 | int nth_active_pipe; |
2772 | int nth_active_pipe; |
3038 | 2773 | ||
3039 | if (!params->active) { |
2774 | if (!params->active) { |
3040 | alloc->start = 0; |
2775 | alloc->start = 0; |
3041 | alloc->end = 0; |
2776 | alloc->end = 0; |
3042 | return; |
2777 | return; |
3043 | } |
2778 | } |
- | 2779 | ||
- | 2780 | if (IS_BROXTON(dev)) |
|
- | 2781 | ddb_size = BXT_DDB_SIZE; |
|
3044 | 2782 | else |
|
3045 | ddb_size = SKL_DDB_SIZE; |
2783 | ddb_size = SKL_DDB_SIZE; |
3046 | 2784 | ||
3047 | ddb_size -= 4; /* 4 blocks for bypass path allocation */ |
2785 | ddb_size -= 4; /* 4 blocks for bypass path allocation */ |
3048 | 2786 | ||
3049 | nth_active_pipe = 0; |
2787 | nth_active_pipe = 0; |
3050 | for_each_crtc(dev, crtc) { |
2788 | for_each_crtc(dev, crtc) { |
3051 | if (!intel_crtc_active(crtc)) |
2789 | if (!to_intel_crtc(crtc)->active) |
3052 | continue; |
2790 | continue; |
3053 | 2791 | ||
3054 | if (crtc == for_crtc) |
2792 | if (crtc == for_crtc) |
3055 | break; |
2793 | break; |
3056 | 2794 | ||
3057 | nth_active_pipe++; |
2795 | nth_active_pipe++; |
3058 | } |
2796 | } |
3059 | 2797 | ||
3060 | pipe_size = ddb_size / config->num_pipes_active; |
2798 | pipe_size = ddb_size / config->num_pipes_active; |
3061 | alloc->start = nth_active_pipe * ddb_size / config->num_pipes_active; |
2799 | alloc->start = nth_active_pipe * ddb_size / config->num_pipes_active; |
3062 | alloc->end = alloc->start + pipe_size; |
2800 | alloc->end = alloc->start + pipe_size; |
3063 | } |
2801 | } |
3064 | 2802 | ||
3065 | static unsigned int skl_cursor_allocation(const struct intel_wm_config *config) |
2803 | static unsigned int skl_cursor_allocation(const struct intel_wm_config *config) |
3066 | { |
2804 | { |
3067 | if (config->num_pipes_active == 1) |
2805 | if (config->num_pipes_active == 1) |
3068 | return 32; |
2806 | return 32; |
3069 | 2807 | ||
3070 | return 8; |
2808 | return 8; |
3071 | } |
2809 | } |
3072 | 2810 | ||
3073 | static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg) |
2811 | static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg) |
3074 | { |
2812 | { |
3075 | entry->start = reg & 0x3ff; |
2813 | entry->start = reg & 0x3ff; |
3076 | entry->end = (reg >> 16) & 0x3ff; |
2814 | entry->end = (reg >> 16) & 0x3ff; |
3077 | if (entry->end) |
2815 | if (entry->end) |
3078 | entry->end += 1; |
2816 | entry->end += 1; |
3079 | } |
2817 | } |
3080 | 2818 | ||
3081 | void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv, |
2819 | void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv, |
3082 | struct skl_ddb_allocation *ddb /* out */) |
2820 | struct skl_ddb_allocation *ddb /* out */) |
3083 | { |
2821 | { |
3084 | struct drm_device *dev = dev_priv->dev; |
- | |
3085 | enum pipe pipe; |
2822 | enum pipe pipe; |
3086 | int plane; |
2823 | int plane; |
3087 | u32 val; |
2824 | u32 val; |
- | 2825 | ||
- | 2826 | memset(ddb, 0, sizeof(*ddb)); |
|
3088 | 2827 | ||
- | 2828 | for_each_pipe(dev_priv, pipe) { |
|
- | 2829 | if (!intel_display_power_is_enabled(dev_priv, POWER_DOMAIN_PIPE(pipe))) |
|
- | 2830 | continue; |
|
3089 | for_each_pipe(dev_priv, pipe) { |
2831 | |
3090 | for_each_plane(pipe, plane) { |
2832 | for_each_plane(dev_priv, pipe, plane) { |
3091 | val = I915_READ(PLANE_BUF_CFG(pipe, plane)); |
2833 | val = I915_READ(PLANE_BUF_CFG(pipe, plane)); |
3092 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane], |
2834 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane], |
3093 | val); |
2835 | val); |
3094 | } |
2836 | } |
3095 | 2837 | ||
3096 | val = I915_READ(CUR_BUF_CFG(pipe)); |
2838 | val = I915_READ(CUR_BUF_CFG(pipe)); |
3097 | skl_ddb_entry_init_from_hw(&ddb->cursor[pipe], val); |
2839 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][PLANE_CURSOR], |
- | 2840 | val); |
|
3098 | } |
2841 | } |
3099 | } |
2842 | } |
3100 | 2843 | ||
3101 | static unsigned int |
2844 | static unsigned int |
3102 | skl_plane_relative_data_rate(const struct intel_plane_wm_parameters *p) |
2845 | skl_plane_relative_data_rate(const struct intel_plane_wm_parameters *p, int y) |
3103 | { |
2846 | { |
- | 2847 | ||
- | 2848 | /* for planar format */ |
|
- | 2849 | if (p->y_bytes_per_pixel) { |
|
- | 2850 | if (y) /* y-plane data rate */ |
|
- | 2851 | return p->horiz_pixels * p->vert_pixels * p->y_bytes_per_pixel; |
|
- | 2852 | else /* uv-plane data rate */ |
|
- | 2853 | return (p->horiz_pixels/2) * (p->vert_pixels/2) * p->bytes_per_pixel; |
|
- | 2854 | } |
|
- | 2855 | ||
- | 2856 | /* for packed formats */ |
|
3104 | return p->horiz_pixels * p->vert_pixels * p->bytes_per_pixel; |
2857 | return p->horiz_pixels * p->vert_pixels * p->bytes_per_pixel; |
3105 | } |
2858 | } |
3106 | 2859 | ||
3107 | /* |
2860 | /* |
3108 | * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching |
2861 | * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching |
3109 | * a 8192x4096@32bpp framebuffer: |
2862 | * a 8192x4096@32bpp framebuffer: |
3110 | * 3 * 4096 * 8192 * 4 < 2^32 |
2863 | * 3 * 4096 * 8192 * 4 < 2^32 |
3111 | */ |
2864 | */ |
3112 | static unsigned int |
2865 | static unsigned int |
3113 | skl_get_total_relative_data_rate(struct intel_crtc *intel_crtc, |
2866 | skl_get_total_relative_data_rate(struct intel_crtc *intel_crtc, |
3114 | const struct skl_pipe_wm_parameters *params) |
2867 | const struct skl_pipe_wm_parameters *params) |
3115 | { |
2868 | { |
3116 | unsigned int total_data_rate = 0; |
2869 | unsigned int total_data_rate = 0; |
3117 | int plane; |
2870 | int plane; |
3118 | 2871 | ||
3119 | for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) { |
2872 | for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) { |
3120 | const struct intel_plane_wm_parameters *p; |
2873 | const struct intel_plane_wm_parameters *p; |
3121 | 2874 | ||
3122 | p = ¶ms->plane[plane]; |
2875 | p = ¶ms->plane[plane]; |
3123 | if (!p->enabled) |
2876 | if (!p->enabled) |
3124 | continue; |
2877 | continue; |
- | 2878 | ||
- | 2879 | total_data_rate += skl_plane_relative_data_rate(p, 0); /* packed/uv */ |
|
3125 | 2880 | if (p->y_bytes_per_pixel) { |
|
- | 2881 | total_data_rate += skl_plane_relative_data_rate(p, 1); /* y-plane */ |
|
3126 | total_data_rate += skl_plane_relative_data_rate(p); |
2882 | } |
3127 | } |
2883 | } |
3128 | 2884 | ||
3129 | return total_data_rate; |
2885 | return total_data_rate; |
3130 | } |
2886 | } |
3131 | 2887 | ||
3132 | static void |
2888 | static void |
3133 | skl_allocate_pipe_ddb(struct drm_crtc *crtc, |
2889 | skl_allocate_pipe_ddb(struct drm_crtc *crtc, |
3134 | const struct intel_wm_config *config, |
2890 | const struct intel_wm_config *config, |
3135 | const struct skl_pipe_wm_parameters *params, |
2891 | const struct skl_pipe_wm_parameters *params, |
3136 | struct skl_ddb_allocation *ddb /* out */) |
2892 | struct skl_ddb_allocation *ddb /* out */) |
3137 | { |
2893 | { |
3138 | struct drm_device *dev = crtc->dev; |
2894 | struct drm_device *dev = crtc->dev; |
- | 2895 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
3139 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
2896 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3140 | enum pipe pipe = intel_crtc->pipe; |
2897 | enum pipe pipe = intel_crtc->pipe; |
3141 | struct skl_ddb_entry *alloc = &ddb->pipe[pipe]; |
2898 | struct skl_ddb_entry *alloc = &ddb->pipe[pipe]; |
3142 | uint16_t alloc_size, start, cursor_blocks; |
2899 | uint16_t alloc_size, start, cursor_blocks; |
- | 2900 | uint16_t minimum[I915_MAX_PLANES]; |
|
- | 2901 | uint16_t y_minimum[I915_MAX_PLANES]; |
|
3143 | unsigned int total_data_rate; |
2902 | unsigned int total_data_rate; |
3144 | int plane; |
2903 | int plane; |
3145 | 2904 | ||
3146 | skl_ddb_get_pipe_allocation_limits(dev, crtc, config, params, alloc); |
2905 | skl_ddb_get_pipe_allocation_limits(dev, crtc, config, params, alloc); |
3147 | alloc_size = skl_ddb_entry_size(alloc); |
2906 | alloc_size = skl_ddb_entry_size(alloc); |
3148 | if (alloc_size == 0) { |
2907 | if (alloc_size == 0) { |
3149 | memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe])); |
2908 | memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe])); |
3150 | memset(&ddb->cursor[pipe], 0, sizeof(ddb->cursor[pipe])); |
2909 | memset(&ddb->plane[pipe][PLANE_CURSOR], 0, |
- | 2910 | sizeof(ddb->plane[pipe][PLANE_CURSOR])); |
|
3151 | return; |
2911 | return; |
3152 | } |
2912 | } |
3153 | 2913 | ||
3154 | cursor_blocks = skl_cursor_allocation(config); |
2914 | cursor_blocks = skl_cursor_allocation(config); |
3155 | ddb->cursor[pipe].start = alloc->end - cursor_blocks; |
2915 | ddb->plane[pipe][PLANE_CURSOR].start = alloc->end - cursor_blocks; |
3156 | ddb->cursor[pipe].end = alloc->end; |
2916 | ddb->plane[pipe][PLANE_CURSOR].end = alloc->end; |
3157 | 2917 | ||
3158 | alloc_size -= cursor_blocks; |
2918 | alloc_size -= cursor_blocks; |
3159 | alloc->end -= cursor_blocks; |
2919 | alloc->end -= cursor_blocks; |
- | 2920 | ||
- | 2921 | /* 1. Allocate the mininum required blocks for each active plane */ |
|
- | 2922 | for_each_plane(dev_priv, pipe, plane) { |
|
- | 2923 | const struct intel_plane_wm_parameters *p; |
|
- | 2924 | ||
- | 2925 | p = ¶ms->plane[plane]; |
|
- | 2926 | if (!p->enabled) |
|
- | 2927 | continue; |
|
- | 2928 | ||
- | 2929 | minimum[plane] = 8; |
|
- | 2930 | alloc_size -= minimum[plane]; |
|
- | 2931 | y_minimum[plane] = p->y_bytes_per_pixel ? 8 : 0; |
|
- | 2932 | alloc_size -= y_minimum[plane]; |
|
- | 2933 | } |
|
3160 | 2934 | ||
3161 | /* |
2935 | /* |
3162 | * Each active plane get a portion of the remaining space, in |
2936 | * 2. Distribute the remaining space in proportion to the amount of |
3163 | * proportion to the amount of data they need to fetch from memory. |
2937 | * data each plane needs to fetch from memory. |
3164 | * |
2938 | * |
3165 | * FIXME: we may not allocate every single block here. |
2939 | * FIXME: we may not allocate every single block here. |
3166 | */ |
2940 | */ |
3167 | total_data_rate = skl_get_total_relative_data_rate(intel_crtc, params); |
2941 | total_data_rate = skl_get_total_relative_data_rate(intel_crtc, params); |
3168 | 2942 | ||
3169 | start = alloc->start; |
2943 | start = alloc->start; |
3170 | for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) { |
2944 | for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) { |
3171 | const struct intel_plane_wm_parameters *p; |
2945 | const struct intel_plane_wm_parameters *p; |
3172 | unsigned int data_rate; |
2946 | unsigned int data_rate, y_data_rate; |
3173 | uint16_t plane_blocks; |
2947 | uint16_t plane_blocks, y_plane_blocks = 0; |
3174 | 2948 | ||
3175 | p = ¶ms->plane[plane]; |
2949 | p = ¶ms->plane[plane]; |
3176 | if (!p->enabled) |
2950 | if (!p->enabled) |
3177 | continue; |
2951 | continue; |
3178 | 2952 | ||
3179 | data_rate = skl_plane_relative_data_rate(p); |
2953 | data_rate = skl_plane_relative_data_rate(p, 0); |
- | 2954 | ||
3180 | 2955 | /* |
|
3181 | /* |
2956 | * allocation for (packed formats) or (uv-plane part of planar format): |
3182 | * promote the expression to 64 bits to avoid overflowing, the |
2957 | * promote the expression to 64 bits to avoid overflowing, the |
3183 | * result is < available as data_rate / total_data_rate < 1 |
2958 | * result is < available as data_rate / total_data_rate < 1 |
3184 | */ |
2959 | */ |
- | 2960 | plane_blocks = minimum[plane]; |
|
3185 | plane_blocks = div_u64((uint64_t)alloc_size * data_rate, |
2961 | plane_blocks += div_u64((uint64_t)alloc_size * data_rate, |
3186 | total_data_rate); |
2962 | total_data_rate); |
3187 | 2963 | ||
3188 | ddb->plane[pipe][plane].start = start; |
2964 | ddb->plane[pipe][plane].start = start; |
3189 | ddb->plane[pipe][plane].end = start + plane_blocks; |
2965 | ddb->plane[pipe][plane].end = start + plane_blocks; |
3190 | 2966 | ||
3191 | start += plane_blocks; |
2967 | start += plane_blocks; |
- | 2968 | ||
- | 2969 | /* |
|
- | 2970 | * allocation for y_plane part of planar format: |
|
- | 2971 | */ |
|
- | 2972 | if (p->y_bytes_per_pixel) { |
|
- | 2973 | y_data_rate = skl_plane_relative_data_rate(p, 1); |
|
- | 2974 | y_plane_blocks = y_minimum[plane]; |
|
- | 2975 | y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate, |
|
- | 2976 | total_data_rate); |
|
- | 2977 | ||
- | 2978 | ddb->y_plane[pipe][plane].start = start; |
|
- | 2979 | ddb->y_plane[pipe][plane].end = start + y_plane_blocks; |
|
- | 2980 | ||
- | 2981 | start += y_plane_blocks; |
|
3192 | } |
2982 | } |
3193 | 2983 | ||
3194 | } |
2984 | } |
- | 2985 | ||
- | 2986 | } |
|
3195 | 2987 | ||
3196 | static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_config *config) |
2988 | static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state *config) |
3197 | { |
2989 | { |
3198 | /* TODO: Take into account the scalers once we support them */ |
2990 | /* TODO: Take into account the scalers once we support them */ |
3199 | return config->adjusted_mode.crtc_clock; |
2991 | return config->base.adjusted_mode.crtc_clock; |
3200 | } |
2992 | } |
3201 | 2993 | ||
3202 | /* |
2994 | /* |
3203 | * The max latency should be 257 (max the punit can code is 255 and we add 2us |
2995 | * The max latency should be 257 (max the punit can code is 255 and we add 2us |
3204 | * for the read latency) and bytes_per_pixel should always be <= 8, so that |
2996 | * for the read latency) and bytes_per_pixel should always be <= 8, so that |
3205 | * should allow pixel_rate up to ~2 GHz which seems sufficient since max |
2997 | * should allow pixel_rate up to ~2 GHz which seems sufficient since max |
3206 | * 2xcdclk is 1350 MHz and the pixel rate should never exceed that. |
2998 | * 2xcdclk is 1350 MHz and the pixel rate should never exceed that. |
3207 | */ |
2999 | */ |
3208 | static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, |
3000 | static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, |
3209 | uint32_t latency) |
3001 | uint32_t latency) |
3210 | { |
3002 | { |
3211 | uint32_t wm_intermediate_val, ret; |
3003 | uint32_t wm_intermediate_val, ret; |
3212 | 3004 | ||
3213 | if (latency == 0) |
3005 | if (latency == 0) |
3214 | return UINT_MAX; |
3006 | return UINT_MAX; |
3215 | 3007 | ||
3216 | wm_intermediate_val = latency * pixel_rate * bytes_per_pixel; |
3008 | wm_intermediate_val = latency * pixel_rate * bytes_per_pixel / 512; |
3217 | ret = DIV_ROUND_UP(wm_intermediate_val, 1000); |
3009 | ret = DIV_ROUND_UP(wm_intermediate_val, 1000); |
3218 | 3010 | ||
3219 | return ret; |
3011 | return ret; |
3220 | } |
3012 | } |
3221 | 3013 | ||
3222 | static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, |
3014 | static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, |
3223 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, |
3015 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, |
3224 | uint32_t latency) |
3016 | uint64_t tiling, uint32_t latency) |
3225 | { |
3017 | { |
- | 3018 | uint32_t ret; |
|
3226 | uint32_t ret, plane_bytes_per_line, wm_intermediate_val; |
3019 | uint32_t plane_bytes_per_line, plane_blocks_per_line; |
- | 3020 | uint32_t wm_intermediate_val; |
|
3227 | 3021 | ||
3228 | if (latency == 0) |
3022 | if (latency == 0) |
3229 | return UINT_MAX; |
3023 | return UINT_MAX; |
3230 | 3024 | ||
3231 | plane_bytes_per_line = horiz_pixels * bytes_per_pixel; |
3025 | plane_bytes_per_line = horiz_pixels * bytes_per_pixel; |
- | 3026 | ||
- | 3027 | if (tiling == I915_FORMAT_MOD_Y_TILED || |
|
- | 3028 | tiling == I915_FORMAT_MOD_Yf_TILED) { |
|
- | 3029 | plane_bytes_per_line *= 4; |
|
- | 3030 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); |
|
- | 3031 | plane_blocks_per_line /= 4; |
|
- | 3032 | } else { |
|
- | 3033 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); |
|
- | 3034 | } |
|
- | 3035 | ||
3232 | wm_intermediate_val = latency * pixel_rate; |
3036 | wm_intermediate_val = latency * pixel_rate; |
3233 | ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) * |
3037 | ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) * |
3234 | plane_bytes_per_line; |
3038 | plane_blocks_per_line; |
3235 | 3039 | ||
3236 | return ret; |
3040 | return ret; |
3237 | } |
3041 | } |
3238 | 3042 | ||
3239 | static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation *new_ddb, |
3043 | static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation *new_ddb, |
3240 | const struct intel_crtc *intel_crtc) |
3044 | const struct intel_crtc *intel_crtc) |
3241 | { |
3045 | { |
3242 | struct drm_device *dev = intel_crtc->base.dev; |
3046 | struct drm_device *dev = intel_crtc->base.dev; |
3243 | struct drm_i915_private *dev_priv = dev->dev_private; |
3047 | struct drm_i915_private *dev_priv = dev->dev_private; |
3244 | const struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb; |
3048 | const struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb; |
3245 | enum pipe pipe = intel_crtc->pipe; |
3049 | enum pipe pipe = intel_crtc->pipe; |
3246 | 3050 | ||
3247 | if (memcmp(new_ddb->plane[pipe], cur_ddb->plane[pipe], |
3051 | if (memcmp(new_ddb->plane[pipe], cur_ddb->plane[pipe], |
3248 | sizeof(new_ddb->plane[pipe]))) |
3052 | sizeof(new_ddb->plane[pipe]))) |
3249 | return true; |
3053 | return true; |
3250 | 3054 | ||
3251 | if (memcmp(&new_ddb->cursor[pipe], &cur_ddb->cursor[pipe], |
3055 | if (memcmp(&new_ddb->plane[pipe][PLANE_CURSOR], &cur_ddb->plane[pipe][PLANE_CURSOR], |
3252 | sizeof(new_ddb->cursor[pipe]))) |
3056 | sizeof(new_ddb->plane[pipe][PLANE_CURSOR]))) |
3253 | return true; |
3057 | return true; |
3254 | 3058 | ||
3255 | return false; |
3059 | return false; |
3256 | } |
3060 | } |
3257 | 3061 | ||
3258 | static void skl_compute_wm_global_parameters(struct drm_device *dev, |
3062 | static void skl_compute_wm_global_parameters(struct drm_device *dev, |
3259 | struct intel_wm_config *config) |
3063 | struct intel_wm_config *config) |
3260 | { |
3064 | { |
3261 | struct drm_crtc *crtc; |
3065 | struct drm_crtc *crtc; |
3262 | struct drm_plane *plane; |
3066 | struct drm_plane *plane; |
3263 | 3067 | ||
3264 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) |
3068 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) |
3265 | config->num_pipes_active += intel_crtc_active(crtc); |
3069 | config->num_pipes_active += to_intel_crtc(crtc)->active; |
3266 | 3070 | ||
3267 | /* FIXME: I don't think we need those two global parameters on SKL */ |
3071 | /* FIXME: I don't think we need those two global parameters on SKL */ |
3268 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { |
3072 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { |
3269 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3073 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3270 | 3074 | ||
3271 | config->sprites_enabled |= intel_plane->wm.enabled; |
3075 | config->sprites_enabled |= intel_plane->wm.enabled; |
3272 | config->sprites_scaled |= intel_plane->wm.scaled; |
3076 | config->sprites_scaled |= intel_plane->wm.scaled; |
3273 | } |
3077 | } |
3274 | } |
3078 | } |
3275 | 3079 | ||
3276 | static void skl_compute_wm_pipe_parameters(struct drm_crtc *crtc, |
3080 | static void skl_compute_wm_pipe_parameters(struct drm_crtc *crtc, |
3277 | struct skl_pipe_wm_parameters *p) |
3081 | struct skl_pipe_wm_parameters *p) |
3278 | { |
3082 | { |
3279 | struct drm_device *dev = crtc->dev; |
3083 | struct drm_device *dev = crtc->dev; |
3280 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3084 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3281 | enum pipe pipe = intel_crtc->pipe; |
3085 | enum pipe pipe = intel_crtc->pipe; |
3282 | struct drm_plane *plane; |
3086 | struct drm_plane *plane; |
- | 3087 | struct drm_framebuffer *fb; |
|
3283 | int i = 1; /* Index for sprite planes start */ |
3088 | int i = 1; /* Index for sprite planes start */ |
3284 | 3089 | ||
3285 | p->active = intel_crtc_active(crtc); |
3090 | p->active = intel_crtc->active; |
3286 | if (p->active) { |
3091 | if (p->active) { |
3287 | p->pipe_htotal = intel_crtc->config.adjusted_mode.crtc_htotal; |
3092 | p->pipe_htotal = intel_crtc->config->base.adjusted_mode.crtc_htotal; |
3288 | p->pixel_rate = skl_pipe_pixel_rate(&intel_crtc->config); |
3093 | p->pixel_rate = skl_pipe_pixel_rate(intel_crtc->config); |
3289 | 3094 | ||
3290 | /* |
3095 | fb = crtc->primary->state->fb; |
3291 | * For now, assume primary and cursor planes are always enabled. |
3096 | /* For planar: Bpp is for uv plane, y_Bpp is for y plane */ |
- | 3097 | if (fb) { |
|
- | 3098 | p->plane[0].enabled = true; |
|
- | 3099 | p->plane[0].bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ? |
|
- | 3100 | drm_format_plane_cpp(fb->pixel_format, 1) : |
|
- | 3101 | drm_format_plane_cpp(fb->pixel_format, 0); |
|
- | 3102 | p->plane[0].y_bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ? |
|
- | 3103 | drm_format_plane_cpp(fb->pixel_format, 0) : 0; |
|
- | 3104 | p->plane[0].tiling = fb->modifier[0]; |
|
3292 | */ |
3105 | } else { |
3293 | p->plane[0].enabled = true; |
3106 | p->plane[0].enabled = false; |
- | 3107 | p->plane[0].bytes_per_pixel = 0; |
|
- | 3108 | p->plane[0].y_bytes_per_pixel = 0; |
|
3294 | p->plane[0].bytes_per_pixel = |
3109 | p->plane[0].tiling = DRM_FORMAT_MOD_NONE; |
3295 | crtc->primary->fb->bits_per_pixel / 8; |
3110 | } |
- | 3111 | p->plane[0].horiz_pixels = intel_crtc->config->pipe_src_w; |
|
3296 | p->plane[0].horiz_pixels = intel_crtc->config.pipe_src_w; |
3112 | p->plane[0].vert_pixels = intel_crtc->config->pipe_src_h; |
3297 | p->plane[0].vert_pixels = intel_crtc->config.pipe_src_h; |
3113 | p->plane[0].rotation = crtc->primary->state->rotation; |
3298 | 3114 | ||
- | 3115 | fb = crtc->cursor->state->fb; |
|
- | 3116 | p->plane[PLANE_CURSOR].y_bytes_per_pixel = 0; |
|
- | 3117 | if (fb) { |
|
3299 | p->cursor.enabled = true; |
3118 | p->plane[PLANE_CURSOR].enabled = true; |
- | 3119 | p->plane[PLANE_CURSOR].bytes_per_pixel = fb->bits_per_pixel / 8; |
|
- | 3120 | p->plane[PLANE_CURSOR].horiz_pixels = crtc->cursor->state->crtc_w; |
|
- | 3121 | p->plane[PLANE_CURSOR].vert_pixels = crtc->cursor->state->crtc_h; |
|
- | 3122 | } else { |
|
3300 | p->cursor.bytes_per_pixel = 4; |
3123 | p->plane[PLANE_CURSOR].enabled = false; |
- | 3124 | p->plane[PLANE_CURSOR].bytes_per_pixel = 0; |
|
- | 3125 | p->plane[PLANE_CURSOR].horiz_pixels = 64; |
|
3301 | p->cursor.horiz_pixels = intel_crtc->cursor_width ? |
3126 | p->plane[PLANE_CURSOR].vert_pixels = 64; |
3302 | intel_crtc->cursor_width : 64; |
3127 | } |
3303 | } |
3128 | } |
3304 | 3129 | ||
3305 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { |
3130 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { |
3306 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3131 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3307 | 3132 | ||
- | 3133 | if (intel_plane->pipe == pipe && |
|
3308 | if (intel_plane->pipe == pipe) |
3134 | plane->type == DRM_PLANE_TYPE_OVERLAY) |
3309 | p->plane[i++] = intel_plane->wm; |
3135 | p->plane[i++] = intel_plane->wm; |
3310 | } |
3136 | } |
3311 | } |
3137 | } |
3312 | 3138 | ||
- | 3139 | static bool skl_compute_plane_wm(const struct drm_i915_private *dev_priv, |
|
3313 | static bool skl_compute_plane_wm(struct skl_pipe_wm_parameters *p, |
3140 | struct skl_pipe_wm_parameters *p, |
3314 | struct intel_plane_wm_parameters *p_params, |
3141 | struct intel_plane_wm_parameters *p_params, |
3315 | uint16_t ddb_allocation, |
3142 | uint16_t ddb_allocation, |
3316 | uint32_t mem_value, |
3143 | int level, |
3317 | uint16_t *out_blocks, /* out */ |
3144 | uint16_t *out_blocks, /* out */ |
3318 | uint8_t *out_lines /* out */) |
3145 | uint8_t *out_lines /* out */) |
3319 | { |
3146 | { |
- | 3147 | uint32_t latency = dev_priv->wm.skl_latency[level]; |
|
- | 3148 | uint32_t method1, method2; |
|
3320 | uint32_t method1, method2, plane_bytes_per_line, res_blocks, res_lines; |
3149 | uint32_t plane_bytes_per_line, plane_blocks_per_line; |
- | 3150 | uint32_t res_blocks, res_lines; |
|
3321 | uint32_t result_bytes; |
3151 | uint32_t selected_result; |
- | 3152 | uint8_t bytes_per_pixel; |
|
3322 | 3153 | ||
3323 | if (mem_value == 0 || !p->active || !p_params->enabled) |
3154 | if (latency == 0 || !p->active || !p_params->enabled) |
- | 3155 | return false; |
|
- | 3156 | ||
- | 3157 | bytes_per_pixel = p_params->y_bytes_per_pixel ? |
|
3324 | return false; |
3158 | p_params->y_bytes_per_pixel : |
3325 | 3159 | p_params->bytes_per_pixel; |
|
3326 | method1 = skl_wm_method1(p->pixel_rate, |
3160 | method1 = skl_wm_method1(p->pixel_rate, |
3327 | p_params->bytes_per_pixel, |
3161 | bytes_per_pixel, |
3328 | mem_value); |
3162 | latency); |
3329 | method2 = skl_wm_method2(p->pixel_rate, |
3163 | method2 = skl_wm_method2(p->pixel_rate, |
3330 | p->pipe_htotal, |
3164 | p->pipe_htotal, |
3331 | p_params->horiz_pixels, |
3165 | p_params->horiz_pixels, |
- | 3166 | bytes_per_pixel, |
|
- | 3167 | p_params->tiling, |
|
- | 3168 | latency); |
|
- | 3169 | ||
- | 3170 | plane_bytes_per_line = p_params->horiz_pixels * bytes_per_pixel; |
|
- | 3171 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); |
|
- | 3172 | ||
- | 3173 | if (p_params->tiling == I915_FORMAT_MOD_Y_TILED || |
|
- | 3174 | p_params->tiling == I915_FORMAT_MOD_Yf_TILED) { |
|
- | 3175 | uint32_t min_scanlines = 4; |
|
- | 3176 | uint32_t y_tile_minimum; |
|
- | 3177 | if (intel_rotation_90_or_270(p_params->rotation)) { |
|
3332 | p_params->bytes_per_pixel, |
3178 | switch (p_params->bytes_per_pixel) { |
- | 3179 | case 1: |
|
- | 3180 | min_scanlines = 16; |
|
3333 | mem_value); |
3181 | break; |
- | 3182 | case 2: |
|
- | 3183 | min_scanlines = 8; |
|
- | 3184 | break; |
|
- | 3185 | case 8: |
|
- | 3186 | WARN(1, "Unsupported pixel depth for rotation"); |
|
- | 3187 | } |
|
- | 3188 | } |
|
- | 3189 | y_tile_minimum = plane_blocks_per_line * min_scanlines; |
|
- | 3190 | selected_result = max(method2, y_tile_minimum); |
|
- | 3191 | } else { |
|
- | 3192 | if ((ddb_allocation / plane_blocks_per_line) >= 1) |
|
- | 3193 | selected_result = min(method1, method2); |
|
- | 3194 | else |
|
- | 3195 | selected_result = method1; |
|
- | 3196 | } |
|
3334 | 3197 | ||
3335 | plane_bytes_per_line = p_params->horiz_pixels * |
3198 | res_blocks = selected_result + 1; |
3336 | p_params->bytes_per_pixel; |
3199 | res_lines = DIV_ROUND_UP(selected_result, plane_blocks_per_line); |
3337 | 3200 | ||
3338 | /* For now xtile and linear */ |
3201 | if (level >= 1 && level <= 7) { |
- | 3202 | if (p_params->tiling == I915_FORMAT_MOD_Y_TILED || |
|
3339 | if (((ddb_allocation * 512) / plane_bytes_per_line) >= 1) |
3203 | p_params->tiling == I915_FORMAT_MOD_Yf_TILED) |
3340 | result_bytes = min(method1, method2); |
3204 | res_lines += 4; |
3341 | else |
3205 | else |
3342 | result_bytes = method1; |
- | |
3343 | - | ||
3344 | res_blocks = DIV_ROUND_UP(result_bytes, 512) + 1; |
3206 | res_blocks++; |
3345 | res_lines = DIV_ROUND_UP(result_bytes, plane_bytes_per_line); |
3207 | } |
3346 | 3208 | ||
3347 | if (res_blocks > ddb_allocation || res_lines > 31) |
3209 | if (res_blocks >= ddb_allocation || res_lines > 31) |
3348 | return false; |
3210 | return false; |
3349 | 3211 | ||
3350 | *out_blocks = res_blocks; |
3212 | *out_blocks = res_blocks; |
3351 | *out_lines = res_lines; |
3213 | *out_lines = res_lines; |
3352 | 3214 | ||
3353 | return true; |
3215 | return true; |
3354 | } |
3216 | } |
3355 | 3217 | ||
3356 | static void skl_compute_wm_level(const struct drm_i915_private *dev_priv, |
3218 | static void skl_compute_wm_level(const struct drm_i915_private *dev_priv, |
3357 | struct skl_ddb_allocation *ddb, |
3219 | struct skl_ddb_allocation *ddb, |
3358 | struct skl_pipe_wm_parameters *p, |
3220 | struct skl_pipe_wm_parameters *p, |
3359 | enum pipe pipe, |
3221 | enum pipe pipe, |
3360 | int level, |
3222 | int level, |
3361 | int num_planes, |
3223 | int num_planes, |
3362 | struct skl_wm_level *result) |
3224 | struct skl_wm_level *result) |
3363 | { |
3225 | { |
3364 | uint16_t latency = dev_priv->wm.skl_latency[level]; |
- | |
3365 | uint16_t ddb_blocks; |
3226 | uint16_t ddb_blocks; |
3366 | int i; |
3227 | int i; |
3367 | 3228 | ||
3368 | for (i = 0; i < num_planes; i++) { |
3229 | for (i = 0; i < num_planes; i++) { |
3369 | ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]); |
3230 | ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]); |
3370 | 3231 | ||
- | 3232 | result->plane_en[i] = skl_compute_plane_wm(dev_priv, |
|
3371 | result->plane_en[i] = skl_compute_plane_wm(p, &p->plane[i], |
3233 | p, &p->plane[i], |
3372 | ddb_blocks, |
3234 | ddb_blocks, |
3373 | latency, |
3235 | level, |
3374 | &result->plane_res_b[i], |
3236 | &result->plane_res_b[i], |
3375 | &result->plane_res_l[i]); |
3237 | &result->plane_res_l[i]); |
3376 | } |
3238 | } |
3377 | 3239 | ||
3378 | ddb_blocks = skl_ddb_entry_size(&ddb->cursor[pipe]); |
3240 | ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][PLANE_CURSOR]); |
- | 3241 | result->plane_en[PLANE_CURSOR] = skl_compute_plane_wm(dev_priv, p, |
|
- | 3242 | &p->plane[PLANE_CURSOR], |
|
3379 | result->cursor_en = skl_compute_plane_wm(p, &p->cursor, ddb_blocks, |
3243 | ddb_blocks, level, |
3380 | latency, &result->cursor_res_b, |
3244 | &result->plane_res_b[PLANE_CURSOR], |
3381 | &result->cursor_res_l); |
3245 | &result->plane_res_l[PLANE_CURSOR]); |
3382 | } |
3246 | } |
3383 | 3247 | ||
3384 | static uint32_t |
3248 | static uint32_t |
3385 | skl_compute_linetime_wm(struct drm_crtc *crtc, struct skl_pipe_wm_parameters *p) |
3249 | skl_compute_linetime_wm(struct drm_crtc *crtc, struct skl_pipe_wm_parameters *p) |
3386 | { |
3250 | { |
3387 | if (!intel_crtc_active(crtc)) |
3251 | if (!to_intel_crtc(crtc)->active) |
3388 | return 0; |
3252 | return 0; |
3389 | 3253 | ||
- | 3254 | if (WARN_ON(p->pixel_rate == 0)) |
|
- | 3255 | return 0; |
|
3390 | return DIV_ROUND_UP(8 * p->pipe_htotal * 1000, p->pixel_rate); |
3256 | |
3391 | 3257 | return DIV_ROUND_UP(8 * p->pipe_htotal * 1000, p->pixel_rate); |
|
3392 | } |
3258 | } |
3393 | 3259 | ||
3394 | static void skl_compute_transition_wm(struct drm_crtc *crtc, |
3260 | static void skl_compute_transition_wm(struct drm_crtc *crtc, |
3395 | struct skl_pipe_wm_parameters *params, |
3261 | struct skl_pipe_wm_parameters *params, |
3396 | struct skl_wm_level *trans_wm /* out */) |
3262 | struct skl_wm_level *trans_wm /* out */) |
3397 | { |
3263 | { |
3398 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3264 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3399 | int i; |
3265 | int i; |
3400 | 3266 | ||
3401 | if (!params->active) |
3267 | if (!params->active) |
3402 | return; |
3268 | return; |
3403 | 3269 | ||
3404 | /* Until we know more, just disable transition WMs */ |
3270 | /* Until we know more, just disable transition WMs */ |
3405 | for (i = 0; i < intel_num_planes(intel_crtc); i++) |
3271 | for (i = 0; i < intel_num_planes(intel_crtc); i++) |
3406 | trans_wm->plane_en[i] = false; |
3272 | trans_wm->plane_en[i] = false; |
3407 | trans_wm->cursor_en = false; |
3273 | trans_wm->plane_en[PLANE_CURSOR] = false; |
3408 | } |
3274 | } |
3409 | 3275 | ||
3410 | static void skl_compute_pipe_wm(struct drm_crtc *crtc, |
3276 | static void skl_compute_pipe_wm(struct drm_crtc *crtc, |
3411 | struct skl_ddb_allocation *ddb, |
3277 | struct skl_ddb_allocation *ddb, |
3412 | struct skl_pipe_wm_parameters *params, |
3278 | struct skl_pipe_wm_parameters *params, |
3413 | struct skl_pipe_wm *pipe_wm) |
3279 | struct skl_pipe_wm *pipe_wm) |
3414 | { |
3280 | { |
3415 | struct drm_device *dev = crtc->dev; |
3281 | struct drm_device *dev = crtc->dev; |
3416 | const struct drm_i915_private *dev_priv = dev->dev_private; |
3282 | const struct drm_i915_private *dev_priv = dev->dev_private; |
3417 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3283 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3418 | int level, max_level = ilk_wm_max_level(dev); |
3284 | int level, max_level = ilk_wm_max_level(dev); |
3419 | 3285 | ||
3420 | for (level = 0; level <= max_level; level++) { |
3286 | for (level = 0; level <= max_level; level++) { |
3421 | skl_compute_wm_level(dev_priv, ddb, params, intel_crtc->pipe, |
3287 | skl_compute_wm_level(dev_priv, ddb, params, intel_crtc->pipe, |
3422 | level, intel_num_planes(intel_crtc), |
3288 | level, intel_num_planes(intel_crtc), |
3423 | &pipe_wm->wm[level]); |
3289 | &pipe_wm->wm[level]); |
3424 | } |
3290 | } |
3425 | pipe_wm->linetime = skl_compute_linetime_wm(crtc, params); |
3291 | pipe_wm->linetime = skl_compute_linetime_wm(crtc, params); |
3426 | 3292 | ||
3427 | skl_compute_transition_wm(crtc, params, &pipe_wm->trans_wm); |
3293 | skl_compute_transition_wm(crtc, params, &pipe_wm->trans_wm); |
3428 | } |
3294 | } |
3429 | 3295 | ||
3430 | static void skl_compute_wm_results(struct drm_device *dev, |
3296 | static void skl_compute_wm_results(struct drm_device *dev, |
3431 | struct skl_pipe_wm_parameters *p, |
3297 | struct skl_pipe_wm_parameters *p, |
3432 | struct skl_pipe_wm *p_wm, |
3298 | struct skl_pipe_wm *p_wm, |
3433 | struct skl_wm_values *r, |
3299 | struct skl_wm_values *r, |
3434 | struct intel_crtc *intel_crtc) |
3300 | struct intel_crtc *intel_crtc) |
3435 | { |
3301 | { |
3436 | int level, max_level = ilk_wm_max_level(dev); |
3302 | int level, max_level = ilk_wm_max_level(dev); |
3437 | enum pipe pipe = intel_crtc->pipe; |
3303 | enum pipe pipe = intel_crtc->pipe; |
3438 | uint32_t temp; |
3304 | uint32_t temp; |
3439 | int i; |
3305 | int i; |
3440 | 3306 | ||
3441 | for (level = 0; level <= max_level; level++) { |
3307 | for (level = 0; level <= max_level; level++) { |
3442 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3308 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3443 | temp = 0; |
3309 | temp = 0; |
3444 | 3310 | ||
3445 | temp |= p_wm->wm[level].plane_res_l[i] << |
3311 | temp |= p_wm->wm[level].plane_res_l[i] << |
3446 | PLANE_WM_LINES_SHIFT; |
3312 | PLANE_WM_LINES_SHIFT; |
3447 | temp |= p_wm->wm[level].plane_res_b[i]; |
3313 | temp |= p_wm->wm[level].plane_res_b[i]; |
3448 | if (p_wm->wm[level].plane_en[i]) |
3314 | if (p_wm->wm[level].plane_en[i]) |
3449 | temp |= PLANE_WM_EN; |
3315 | temp |= PLANE_WM_EN; |
3450 | 3316 | ||
3451 | r->plane[pipe][i][level] = temp; |
3317 | r->plane[pipe][i][level] = temp; |
3452 | } |
3318 | } |
3453 | 3319 | ||
3454 | temp = 0; |
3320 | temp = 0; |
3455 | 3321 | ||
3456 | temp |= p_wm->wm[level].cursor_res_l << PLANE_WM_LINES_SHIFT; |
3322 | temp |= p_wm->wm[level].plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT; |
3457 | temp |= p_wm->wm[level].cursor_res_b; |
3323 | temp |= p_wm->wm[level].plane_res_b[PLANE_CURSOR]; |
3458 | 3324 | ||
3459 | if (p_wm->wm[level].cursor_en) |
3325 | if (p_wm->wm[level].plane_en[PLANE_CURSOR]) |
3460 | temp |= PLANE_WM_EN; |
3326 | temp |= PLANE_WM_EN; |
3461 | 3327 | ||
3462 | r->cursor[pipe][level] = temp; |
3328 | r->plane[pipe][PLANE_CURSOR][level] = temp; |
3463 | 3329 | ||
3464 | } |
3330 | } |
3465 | 3331 | ||
3466 | /* transition WMs */ |
3332 | /* transition WMs */ |
3467 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3333 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3468 | temp = 0; |
3334 | temp = 0; |
3469 | temp |= p_wm->trans_wm.plane_res_l[i] << PLANE_WM_LINES_SHIFT; |
3335 | temp |= p_wm->trans_wm.plane_res_l[i] << PLANE_WM_LINES_SHIFT; |
3470 | temp |= p_wm->trans_wm.plane_res_b[i]; |
3336 | temp |= p_wm->trans_wm.plane_res_b[i]; |
3471 | if (p_wm->trans_wm.plane_en[i]) |
3337 | if (p_wm->trans_wm.plane_en[i]) |
3472 | temp |= PLANE_WM_EN; |
3338 | temp |= PLANE_WM_EN; |
3473 | 3339 | ||
3474 | r->plane_trans[pipe][i] = temp; |
3340 | r->plane_trans[pipe][i] = temp; |
3475 | } |
3341 | } |
3476 | 3342 | ||
3477 | temp = 0; |
3343 | temp = 0; |
3478 | temp |= p_wm->trans_wm.cursor_res_l << PLANE_WM_LINES_SHIFT; |
3344 | temp |= p_wm->trans_wm.plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT; |
3479 | temp |= p_wm->trans_wm.cursor_res_b; |
3345 | temp |= p_wm->trans_wm.plane_res_b[PLANE_CURSOR]; |
3480 | if (p_wm->trans_wm.cursor_en) |
3346 | if (p_wm->trans_wm.plane_en[PLANE_CURSOR]) |
3481 | temp |= PLANE_WM_EN; |
3347 | temp |= PLANE_WM_EN; |
3482 | 3348 | ||
3483 | r->cursor_trans[pipe] = temp; |
3349 | r->plane_trans[pipe][PLANE_CURSOR] = temp; |
3484 | 3350 | ||
3485 | r->wm_linetime[pipe] = p_wm->linetime; |
3351 | r->wm_linetime[pipe] = p_wm->linetime; |
3486 | } |
3352 | } |
3487 | 3353 | ||
3488 | static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, uint32_t reg, |
3354 | static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, uint32_t reg, |
3489 | const struct skl_ddb_entry *entry) |
3355 | const struct skl_ddb_entry *entry) |
3490 | { |
3356 | { |
3491 | if (entry->end) |
3357 | if (entry->end) |
3492 | I915_WRITE(reg, (entry->end - 1) << 16 | entry->start); |
3358 | I915_WRITE(reg, (entry->end - 1) << 16 | entry->start); |
3493 | else |
3359 | else |
3494 | I915_WRITE(reg, 0); |
3360 | I915_WRITE(reg, 0); |
3495 | } |
3361 | } |
3496 | 3362 | ||
3497 | static void skl_write_wm_values(struct drm_i915_private *dev_priv, |
3363 | static void skl_write_wm_values(struct drm_i915_private *dev_priv, |
3498 | const struct skl_wm_values *new) |
3364 | const struct skl_wm_values *new) |
3499 | { |
3365 | { |
3500 | struct drm_device *dev = dev_priv->dev; |
3366 | struct drm_device *dev = dev_priv->dev; |
3501 | struct intel_crtc *crtc; |
3367 | struct intel_crtc *crtc; |
3502 | 3368 | ||
3503 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, base.head) { |
3369 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, base.head) { |
3504 | int i, level, max_level = ilk_wm_max_level(dev); |
3370 | int i, level, max_level = ilk_wm_max_level(dev); |
3505 | enum pipe pipe = crtc->pipe; |
3371 | enum pipe pipe = crtc->pipe; |
3506 | 3372 | ||
3507 | if (!new->dirty[pipe]) |
3373 | if (!new->dirty[pipe]) |
3508 | continue; |
3374 | continue; |
3509 | 3375 | ||
3510 | I915_WRITE(PIPE_WM_LINETIME(pipe), new->wm_linetime[pipe]); |
3376 | I915_WRITE(PIPE_WM_LINETIME(pipe), new->wm_linetime[pipe]); |
3511 | 3377 | ||
3512 | for (level = 0; level <= max_level; level++) { |
3378 | for (level = 0; level <= max_level; level++) { |
3513 | for (i = 0; i < intel_num_planes(crtc); i++) |
3379 | for (i = 0; i < intel_num_planes(crtc); i++) |
3514 | I915_WRITE(PLANE_WM(pipe, i, level), |
3380 | I915_WRITE(PLANE_WM(pipe, i, level), |
3515 | new->plane[pipe][i][level]); |
3381 | new->plane[pipe][i][level]); |
3516 | I915_WRITE(CUR_WM(pipe, level), |
3382 | I915_WRITE(CUR_WM(pipe, level), |
3517 | new->cursor[pipe][level]); |
3383 | new->plane[pipe][PLANE_CURSOR][level]); |
3518 | } |
3384 | } |
3519 | for (i = 0; i < intel_num_planes(crtc); i++) |
3385 | for (i = 0; i < intel_num_planes(crtc); i++) |
3520 | I915_WRITE(PLANE_WM_TRANS(pipe, i), |
3386 | I915_WRITE(PLANE_WM_TRANS(pipe, i), |
3521 | new->plane_trans[pipe][i]); |
3387 | new->plane_trans[pipe][i]); |
3522 | I915_WRITE(CUR_WM_TRANS(pipe), new->cursor_trans[pipe]); |
3388 | I915_WRITE(CUR_WM_TRANS(pipe), |
- | 3389 | new->plane_trans[pipe][PLANE_CURSOR]); |
|
3523 | 3390 | ||
3524 | for (i = 0; i < intel_num_planes(crtc); i++) |
3391 | for (i = 0; i < intel_num_planes(crtc); i++) { |
3525 | skl_ddb_entry_write(dev_priv, |
3392 | skl_ddb_entry_write(dev_priv, |
3526 | PLANE_BUF_CFG(pipe, i), |
3393 | PLANE_BUF_CFG(pipe, i), |
3527 | &new->ddb.plane[pipe][i]); |
3394 | &new->ddb.plane[pipe][i]); |
- | 3395 | skl_ddb_entry_write(dev_priv, |
|
- | 3396 | PLANE_NV12_BUF_CFG(pipe, i), |
|
- | 3397 | &new->ddb.y_plane[pipe][i]); |
|
- | 3398 | } |
|
3528 | 3399 | ||
3529 | skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe), |
3400 | skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe), |
3530 | &new->ddb.cursor[pipe]); |
3401 | &new->ddb.plane[pipe][PLANE_CURSOR]); |
3531 | } |
3402 | } |
3532 | } |
3403 | } |
3533 | 3404 | ||
3534 | /* |
3405 | /* |
3535 | * When setting up a new DDB allocation arrangement, we need to correctly |
3406 | * When setting up a new DDB allocation arrangement, we need to correctly |
3536 | * sequence the times at which the new allocations for the pipes are taken into |
3407 | * sequence the times at which the new allocations for the pipes are taken into |
3537 | * account or we'll have pipes fetching from space previously allocated to |
3408 | * account or we'll have pipes fetching from space previously allocated to |
3538 | * another pipe. |
3409 | * another pipe. |
3539 | * |
3410 | * |
3540 | * Roughly the sequence looks like: |
3411 | * Roughly the sequence looks like: |
3541 | * 1. re-allocate the pipe(s) with the allocation being reduced and not |
3412 | * 1. re-allocate the pipe(s) with the allocation being reduced and not |
3542 | * overlapping with a previous light-up pipe (another way to put it is: |
3413 | * overlapping with a previous light-up pipe (another way to put it is: |
3543 | * pipes with their new allocation strickly included into their old ones). |
3414 | * pipes with their new allocation strickly included into their old ones). |
3544 | * 2. re-allocate the other pipes that get their allocation reduced |
3415 | * 2. re-allocate the other pipes that get their allocation reduced |
3545 | * 3. allocate the pipes having their allocation increased |
3416 | * 3. allocate the pipes having their allocation increased |
3546 | * |
3417 | * |
3547 | * Steps 1. and 2. are here to take care of the following case: |
3418 | * Steps 1. and 2. are here to take care of the following case: |
3548 | * - Initially DDB looks like this: |
3419 | * - Initially DDB looks like this: |
3549 | * | B | C | |
3420 | * | B | C | |
3550 | * - enable pipe A. |
3421 | * - enable pipe A. |
3551 | * - pipe B has a reduced DDB allocation that overlaps with the old pipe C |
3422 | * - pipe B has a reduced DDB allocation that overlaps with the old pipe C |
3552 | * allocation |
3423 | * allocation |
3553 | * | A | B | C | |
3424 | * | A | B | C | |
3554 | * |
3425 | * |
3555 | * We need to sequence the re-allocation: C, B, A (and not B, C, A). |
3426 | * We need to sequence the re-allocation: C, B, A (and not B, C, A). |
3556 | */ |
3427 | */ |
3557 | 3428 | ||
3558 | static void |
3429 | static void |
3559 | skl_wm_flush_pipe(struct drm_i915_private *dev_priv, enum pipe pipe, int pass) |
3430 | skl_wm_flush_pipe(struct drm_i915_private *dev_priv, enum pipe pipe, int pass) |
3560 | { |
3431 | { |
3561 | struct drm_device *dev = dev_priv->dev; |
- | |
3562 | int plane; |
3432 | int plane; |
3563 | 3433 | ||
3564 | DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe), pass); |
3434 | DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe), pass); |
3565 | 3435 | ||
3566 | for_each_plane(pipe, plane) { |
3436 | for_each_plane(dev_priv, pipe, plane) { |
3567 | I915_WRITE(PLANE_SURF(pipe, plane), |
3437 | I915_WRITE(PLANE_SURF(pipe, plane), |
3568 | I915_READ(PLANE_SURF(pipe, plane))); |
3438 | I915_READ(PLANE_SURF(pipe, plane))); |
3569 | } |
3439 | } |
3570 | I915_WRITE(CURBASE(pipe), I915_READ(CURBASE(pipe))); |
3440 | I915_WRITE(CURBASE(pipe), I915_READ(CURBASE(pipe))); |
3571 | } |
3441 | } |
3572 | 3442 | ||
3573 | static bool |
3443 | static bool |
3574 | skl_ddb_allocation_included(const struct skl_ddb_allocation *old, |
3444 | skl_ddb_allocation_included(const struct skl_ddb_allocation *old, |
3575 | const struct skl_ddb_allocation *new, |
3445 | const struct skl_ddb_allocation *new, |
3576 | enum pipe pipe) |
3446 | enum pipe pipe) |
3577 | { |
3447 | { |
3578 | uint16_t old_size, new_size; |
3448 | uint16_t old_size, new_size; |
3579 | 3449 | ||
3580 | old_size = skl_ddb_entry_size(&old->pipe[pipe]); |
3450 | old_size = skl_ddb_entry_size(&old->pipe[pipe]); |
3581 | new_size = skl_ddb_entry_size(&new->pipe[pipe]); |
3451 | new_size = skl_ddb_entry_size(&new->pipe[pipe]); |
3582 | 3452 | ||
3583 | return old_size != new_size && |
3453 | return old_size != new_size && |
3584 | new->pipe[pipe].start >= old->pipe[pipe].start && |
3454 | new->pipe[pipe].start >= old->pipe[pipe].start && |
3585 | new->pipe[pipe].end <= old->pipe[pipe].end; |
3455 | new->pipe[pipe].end <= old->pipe[pipe].end; |
3586 | } |
3456 | } |
3587 | 3457 | ||
3588 | static void skl_flush_wm_values(struct drm_i915_private *dev_priv, |
3458 | static void skl_flush_wm_values(struct drm_i915_private *dev_priv, |
3589 | struct skl_wm_values *new_values) |
3459 | struct skl_wm_values *new_values) |
3590 | { |
3460 | { |
3591 | struct drm_device *dev = dev_priv->dev; |
3461 | struct drm_device *dev = dev_priv->dev; |
3592 | struct skl_ddb_allocation *cur_ddb, *new_ddb; |
3462 | struct skl_ddb_allocation *cur_ddb, *new_ddb; |
3593 | bool reallocated[I915_MAX_PIPES] = {false, false, false}; |
3463 | bool reallocated[I915_MAX_PIPES] = {}; |
3594 | struct intel_crtc *crtc; |
3464 | struct intel_crtc *crtc; |
3595 | enum pipe pipe; |
3465 | enum pipe pipe; |
3596 | 3466 | ||
3597 | new_ddb = &new_values->ddb; |
3467 | new_ddb = &new_values->ddb; |
3598 | cur_ddb = &dev_priv->wm.skl_hw.ddb; |
3468 | cur_ddb = &dev_priv->wm.skl_hw.ddb; |
3599 | 3469 | ||
3600 | /* |
3470 | /* |
3601 | * First pass: flush the pipes with the new allocation contained into |
3471 | * First pass: flush the pipes with the new allocation contained into |
3602 | * the old space. |
3472 | * the old space. |
3603 | * |
3473 | * |
3604 | * We'll wait for the vblank on those pipes to ensure we can safely |
3474 | * We'll wait for the vblank on those pipes to ensure we can safely |
3605 | * re-allocate the freed space without this pipe fetching from it. |
3475 | * re-allocate the freed space without this pipe fetching from it. |
3606 | */ |
3476 | */ |
3607 | for_each_intel_crtc(dev, crtc) { |
3477 | for_each_intel_crtc(dev, crtc) { |
3608 | if (!crtc->active) |
3478 | if (!crtc->active) |
3609 | continue; |
3479 | continue; |
3610 | 3480 | ||
3611 | pipe = crtc->pipe; |
3481 | pipe = crtc->pipe; |
3612 | 3482 | ||
3613 | if (!skl_ddb_allocation_included(cur_ddb, new_ddb, pipe)) |
3483 | if (!skl_ddb_allocation_included(cur_ddb, new_ddb, pipe)) |
3614 | continue; |
3484 | continue; |
3615 | 3485 | ||
3616 | skl_wm_flush_pipe(dev_priv, pipe, 1); |
3486 | skl_wm_flush_pipe(dev_priv, pipe, 1); |
3617 | intel_wait_for_vblank(dev, pipe); |
3487 | intel_wait_for_vblank(dev, pipe); |
3618 | 3488 | ||
3619 | reallocated[pipe] = true; |
3489 | reallocated[pipe] = true; |
3620 | } |
3490 | } |
3621 | 3491 | ||
3622 | 3492 | ||
3623 | /* |
3493 | /* |
3624 | * Second pass: flush the pipes that are having their allocation |
3494 | * Second pass: flush the pipes that are having their allocation |
3625 | * reduced, but overlapping with a previous allocation. |
3495 | * reduced, but overlapping with a previous allocation. |
3626 | * |
3496 | * |
3627 | * Here as well we need to wait for the vblank to make sure the freed |
3497 | * Here as well we need to wait for the vblank to make sure the freed |
3628 | * space is not used anymore. |
3498 | * space is not used anymore. |
3629 | */ |
3499 | */ |
3630 | for_each_intel_crtc(dev, crtc) { |
3500 | for_each_intel_crtc(dev, crtc) { |
3631 | if (!crtc->active) |
3501 | if (!crtc->active) |
3632 | continue; |
3502 | continue; |
3633 | 3503 | ||
3634 | pipe = crtc->pipe; |
3504 | pipe = crtc->pipe; |
3635 | 3505 | ||
3636 | if (reallocated[pipe]) |
3506 | if (reallocated[pipe]) |
3637 | continue; |
3507 | continue; |
3638 | 3508 | ||
3639 | if (skl_ddb_entry_size(&new_ddb->pipe[pipe]) < |
3509 | if (skl_ddb_entry_size(&new_ddb->pipe[pipe]) < |
3640 | skl_ddb_entry_size(&cur_ddb->pipe[pipe])) { |
3510 | skl_ddb_entry_size(&cur_ddb->pipe[pipe])) { |
3641 | skl_wm_flush_pipe(dev_priv, pipe, 2); |
3511 | skl_wm_flush_pipe(dev_priv, pipe, 2); |
3642 | intel_wait_for_vblank(dev, pipe); |
3512 | intel_wait_for_vblank(dev, pipe); |
3643 | } |
- | |
3644 | - | ||
3645 | reallocated[pipe] = true; |
3513 | reallocated[pipe] = true; |
3646 | } |
3514 | } |
- | 3515 | } |
|
3647 | 3516 | ||
3648 | /* |
3517 | /* |
3649 | * Third pass: flush the pipes that got more space allocated. |
3518 | * Third pass: flush the pipes that got more space allocated. |
3650 | * |
3519 | * |
3651 | * We don't need to actively wait for the update here, next vblank |
3520 | * We don't need to actively wait for the update here, next vblank |
3652 | * will just get more DDB space with the correct WM values. |
3521 | * will just get more DDB space with the correct WM values. |
3653 | */ |
3522 | */ |
3654 | for_each_intel_crtc(dev, crtc) { |
3523 | for_each_intel_crtc(dev, crtc) { |
3655 | if (!crtc->active) |
3524 | if (!crtc->active) |
3656 | continue; |
3525 | continue; |
3657 | 3526 | ||
3658 | pipe = crtc->pipe; |
3527 | pipe = crtc->pipe; |
3659 | 3528 | ||
3660 | /* |
3529 | /* |
3661 | * At this point, only the pipes more space than before are |
3530 | * At this point, only the pipes more space than before are |
3662 | * left to re-allocate. |
3531 | * left to re-allocate. |
3663 | */ |
3532 | */ |
3664 | if (reallocated[pipe]) |
3533 | if (reallocated[pipe]) |
3665 | continue; |
3534 | continue; |
3666 | 3535 | ||
3667 | skl_wm_flush_pipe(dev_priv, pipe, 3); |
3536 | skl_wm_flush_pipe(dev_priv, pipe, 3); |
3668 | } |
3537 | } |
3669 | } |
3538 | } |
3670 | 3539 | ||
3671 | static bool skl_update_pipe_wm(struct drm_crtc *crtc, |
3540 | static bool skl_update_pipe_wm(struct drm_crtc *crtc, |
3672 | struct skl_pipe_wm_parameters *params, |
3541 | struct skl_pipe_wm_parameters *params, |
3673 | struct intel_wm_config *config, |
3542 | struct intel_wm_config *config, |
3674 | struct skl_ddb_allocation *ddb, /* out */ |
3543 | struct skl_ddb_allocation *ddb, /* out */ |
3675 | struct skl_pipe_wm *pipe_wm /* out */) |
3544 | struct skl_pipe_wm *pipe_wm /* out */) |
3676 | { |
3545 | { |
3677 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3546 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3678 | 3547 | ||
3679 | skl_compute_wm_pipe_parameters(crtc, params); |
3548 | skl_compute_wm_pipe_parameters(crtc, params); |
3680 | skl_allocate_pipe_ddb(crtc, config, params, ddb); |
3549 | skl_allocate_pipe_ddb(crtc, config, params, ddb); |
3681 | skl_compute_pipe_wm(crtc, ddb, params, pipe_wm); |
3550 | skl_compute_pipe_wm(crtc, ddb, params, pipe_wm); |
3682 | 3551 | ||
3683 | if (!memcmp(&intel_crtc->wm.skl_active, pipe_wm, sizeof(*pipe_wm))) |
3552 | if (!memcmp(&intel_crtc->wm.skl_active, pipe_wm, sizeof(*pipe_wm))) |
3684 | return false; |
3553 | return false; |
3685 | 3554 | ||
3686 | intel_crtc->wm.skl_active = *pipe_wm; |
3555 | intel_crtc->wm.skl_active = *pipe_wm; |
- | 3556 | ||
3687 | return true; |
3557 | return true; |
3688 | } |
3558 | } |
3689 | 3559 | ||
3690 | static void skl_update_other_pipe_wm(struct drm_device *dev, |
3560 | static void skl_update_other_pipe_wm(struct drm_device *dev, |
3691 | struct drm_crtc *crtc, |
3561 | struct drm_crtc *crtc, |
3692 | struct intel_wm_config *config, |
3562 | struct intel_wm_config *config, |
3693 | struct skl_wm_values *r) |
3563 | struct skl_wm_values *r) |
3694 | { |
3564 | { |
3695 | struct intel_crtc *intel_crtc; |
3565 | struct intel_crtc *intel_crtc; |
3696 | struct intel_crtc *this_crtc = to_intel_crtc(crtc); |
3566 | struct intel_crtc *this_crtc = to_intel_crtc(crtc); |
3697 | 3567 | ||
3698 | /* |
3568 | /* |
3699 | * If the WM update hasn't changed the allocation for this_crtc (the |
3569 | * If the WM update hasn't changed the allocation for this_crtc (the |
3700 | * crtc we are currently computing the new WM values for), other |
3570 | * crtc we are currently computing the new WM values for), other |
3701 | * enabled crtcs will keep the same allocation and we don't need to |
3571 | * enabled crtcs will keep the same allocation and we don't need to |
3702 | * recompute anything for them. |
3572 | * recompute anything for them. |
3703 | */ |
3573 | */ |
3704 | if (!skl_ddb_allocation_changed(&r->ddb, this_crtc)) |
3574 | if (!skl_ddb_allocation_changed(&r->ddb, this_crtc)) |
3705 | return; |
3575 | return; |
3706 | 3576 | ||
3707 | /* |
3577 | /* |
3708 | * Otherwise, because of this_crtc being freshly enabled/disabled, the |
3578 | * Otherwise, because of this_crtc being freshly enabled/disabled, the |
3709 | * other active pipes need new DDB allocation and WM values. |
3579 | * other active pipes need new DDB allocation and WM values. |
3710 | */ |
3580 | */ |
3711 | list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, |
3581 | list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, |
3712 | base.head) { |
3582 | base.head) { |
3713 | struct skl_pipe_wm_parameters params = {}; |
3583 | struct skl_pipe_wm_parameters params = {}; |
3714 | struct skl_pipe_wm pipe_wm = {}; |
3584 | struct skl_pipe_wm pipe_wm = {}; |
3715 | bool wm_changed; |
3585 | bool wm_changed; |
3716 | 3586 | ||
3717 | if (this_crtc->pipe == intel_crtc->pipe) |
3587 | if (this_crtc->pipe == intel_crtc->pipe) |
3718 | continue; |
3588 | continue; |
3719 | 3589 | ||
3720 | if (!intel_crtc->active) |
3590 | if (!intel_crtc->active) |
3721 | continue; |
3591 | continue; |
3722 | 3592 | ||
3723 | wm_changed = skl_update_pipe_wm(&intel_crtc->base, |
3593 | wm_changed = skl_update_pipe_wm(&intel_crtc->base, |
3724 | ¶ms, config, |
3594 | ¶ms, config, |
3725 | &r->ddb, &pipe_wm); |
3595 | &r->ddb, &pipe_wm); |
3726 | 3596 | ||
3727 | /* |
3597 | /* |
3728 | * If we end up re-computing the other pipe WM values, it's |
3598 | * If we end up re-computing the other pipe WM values, it's |
3729 | * because it was really needed, so we expect the WM values to |
3599 | * because it was really needed, so we expect the WM values to |
3730 | * be different. |
3600 | * be different. |
3731 | */ |
3601 | */ |
3732 | WARN_ON(!wm_changed); |
3602 | WARN_ON(!wm_changed); |
3733 | 3603 | ||
3734 | skl_compute_wm_results(dev, ¶ms, &pipe_wm, r, intel_crtc); |
3604 | skl_compute_wm_results(dev, ¶ms, &pipe_wm, r, intel_crtc); |
3735 | r->dirty[intel_crtc->pipe] = true; |
3605 | r->dirty[intel_crtc->pipe] = true; |
3736 | } |
3606 | } |
3737 | } |
3607 | } |
- | 3608 | ||
- | 3609 | static void skl_clear_wm(struct skl_wm_values *watermarks, enum pipe pipe) |
|
- | 3610 | { |
|
- | 3611 | watermarks->wm_linetime[pipe] = 0; |
|
- | 3612 | memset(watermarks->plane[pipe], 0, |
|
- | 3613 | sizeof(uint32_t) * 8 * I915_MAX_PLANES); |
|
- | 3614 | memset(watermarks->plane_trans[pipe], |
|
- | 3615 | 0, sizeof(uint32_t) * I915_MAX_PLANES); |
|
- | 3616 | watermarks->plane_trans[pipe][PLANE_CURSOR] = 0; |
|
- | 3617 | ||
- | 3618 | /* Clear ddb entries for pipe */ |
|
- | 3619 | memset(&watermarks->ddb.pipe[pipe], 0, sizeof(struct skl_ddb_entry)); |
|
- | 3620 | memset(&watermarks->ddb.plane[pipe], 0, |
|
- | 3621 | sizeof(struct skl_ddb_entry) * I915_MAX_PLANES); |
|
- | 3622 | memset(&watermarks->ddb.y_plane[pipe], 0, |
|
- | 3623 | sizeof(struct skl_ddb_entry) * I915_MAX_PLANES); |
|
- | 3624 | memset(&watermarks->ddb.plane[pipe][PLANE_CURSOR], 0, |
|
- | 3625 | sizeof(struct skl_ddb_entry)); |
|
- | 3626 | ||
- | 3627 | } |
|
3738 | 3628 | ||
3739 | static void skl_update_wm(struct drm_crtc *crtc) |
3629 | static void skl_update_wm(struct drm_crtc *crtc) |
3740 | { |
3630 | { |
3741 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3631 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3742 | struct drm_device *dev = crtc->dev; |
3632 | struct drm_device *dev = crtc->dev; |
3743 | struct drm_i915_private *dev_priv = dev->dev_private; |
3633 | struct drm_i915_private *dev_priv = dev->dev_private; |
3744 | struct skl_pipe_wm_parameters params = {}; |
3634 | struct skl_pipe_wm_parameters params = {}; |
3745 | struct skl_wm_values *results = &dev_priv->wm.skl_results; |
3635 | struct skl_wm_values *results = &dev_priv->wm.skl_results; |
3746 | struct skl_pipe_wm pipe_wm = {}; |
3636 | struct skl_pipe_wm pipe_wm = {}; |
3747 | struct intel_wm_config config = {}; |
3637 | struct intel_wm_config config = {}; |
- | 3638 | ||
- | 3639 | ||
3748 | 3640 | /* Clear all dirty flags */ |
|
- | 3641 | memset(results->dirty, 0, sizeof(bool) * I915_MAX_PIPES); |
|
- | 3642 | ||
3749 | memset(results, 0, sizeof(*results)); |
3643 | skl_clear_wm(results, intel_crtc->pipe); |
3750 | 3644 | ||
3751 | skl_compute_wm_global_parameters(dev, &config); |
3645 | skl_compute_wm_global_parameters(dev, &config); |
3752 | 3646 | ||
3753 | if (!skl_update_pipe_wm(crtc, ¶ms, &config, |
3647 | if (!skl_update_pipe_wm(crtc, ¶ms, &config, |
3754 | &results->ddb, &pipe_wm)) |
3648 | &results->ddb, &pipe_wm)) |
3755 | return; |
3649 | return; |
3756 | 3650 | ||
3757 | skl_compute_wm_results(dev, ¶ms, &pipe_wm, results, intel_crtc); |
3651 | skl_compute_wm_results(dev, ¶ms, &pipe_wm, results, intel_crtc); |
3758 | results->dirty[intel_crtc->pipe] = true; |
3652 | results->dirty[intel_crtc->pipe] = true; |
3759 | 3653 | ||
3760 | skl_update_other_pipe_wm(dev, crtc, &config, results); |
3654 | skl_update_other_pipe_wm(dev, crtc, &config, results); |
3761 | skl_write_wm_values(dev_priv, results); |
3655 | skl_write_wm_values(dev_priv, results); |
3762 | skl_flush_wm_values(dev_priv, results); |
3656 | skl_flush_wm_values(dev_priv, results); |
3763 | 3657 | ||
3764 | /* store the new configuration */ |
3658 | /* store the new configuration */ |
3765 | dev_priv->wm.skl_hw = *results; |
3659 | dev_priv->wm.skl_hw = *results; |
3766 | } |
3660 | } |
3767 | 3661 | ||
3768 | static void |
3662 | static void |
3769 | skl_update_sprite_wm(struct drm_plane *plane, struct drm_crtc *crtc, |
3663 | skl_update_sprite_wm(struct drm_plane *plane, struct drm_crtc *crtc, |
3770 | uint32_t sprite_width, uint32_t sprite_height, |
3664 | uint32_t sprite_width, uint32_t sprite_height, |
3771 | int pixel_size, bool enabled, bool scaled) |
3665 | int pixel_size, bool enabled, bool scaled) |
3772 | { |
3666 | { |
3773 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3667 | struct intel_plane *intel_plane = to_intel_plane(plane); |
- | 3668 | struct drm_framebuffer *fb = plane->state->fb; |
|
3774 | 3669 | ||
3775 | intel_plane->wm.enabled = enabled; |
3670 | intel_plane->wm.enabled = enabled; |
3776 | intel_plane->wm.scaled = scaled; |
3671 | intel_plane->wm.scaled = scaled; |
3777 | intel_plane->wm.horiz_pixels = sprite_width; |
3672 | intel_plane->wm.horiz_pixels = sprite_width; |
3778 | intel_plane->wm.vert_pixels = sprite_height; |
3673 | intel_plane->wm.vert_pixels = sprite_height; |
- | 3674 | intel_plane->wm.tiling = DRM_FORMAT_MOD_NONE; |
|
- | 3675 | ||
- | 3676 | /* For planar: Bpp is for UV plane, y_Bpp is for Y plane */ |
|
3779 | intel_plane->wm.bytes_per_pixel = pixel_size; |
3677 | intel_plane->wm.bytes_per_pixel = |
- | 3678 | (fb && fb->pixel_format == DRM_FORMAT_NV12) ? |
|
- | 3679 | drm_format_plane_cpp(plane->state->fb->pixel_format, 1) : pixel_size; |
|
- | 3680 | intel_plane->wm.y_bytes_per_pixel = |
|
- | 3681 | (fb && fb->pixel_format == DRM_FORMAT_NV12) ? |
|
- | 3682 | drm_format_plane_cpp(plane->state->fb->pixel_format, 0) : 0; |
|
- | 3683 | ||
- | 3684 | /* |
|
- | 3685 | * Framebuffer can be NULL on plane disable, but it does not |
|
- | 3686 | * matter for watermarks if we assume no tiling in that case. |
|
- | 3687 | */ |
|
- | 3688 | if (fb) |
|
- | 3689 | intel_plane->wm.tiling = fb->modifier[0]; |
|
- | 3690 | intel_plane->wm.rotation = plane->state->rotation; |
|
3780 | 3691 | ||
3781 | skl_update_wm(crtc); |
3692 | skl_update_wm(crtc); |
3782 | } |
3693 | } |
3783 | 3694 | ||
3784 | static void ilk_update_wm(struct drm_crtc *crtc) |
3695 | static void ilk_update_wm(struct drm_crtc *crtc) |
3785 | { |
3696 | { |
3786 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3697 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
- | 3698 | struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state); |
|
3787 | struct drm_device *dev = crtc->dev; |
3699 | struct drm_device *dev = crtc->dev; |
3788 | struct drm_i915_private *dev_priv = dev->dev_private; |
3700 | struct drm_i915_private *dev_priv = dev->dev_private; |
3789 | struct ilk_wm_maximums max; |
3701 | struct ilk_wm_maximums max; |
3790 | struct ilk_pipe_wm_parameters params = {}; |
- | |
3791 | struct ilk_wm_values results = {}; |
3702 | struct ilk_wm_values results = {}; |
3792 | enum intel_ddb_partitioning partitioning; |
3703 | enum intel_ddb_partitioning partitioning; |
3793 | struct intel_pipe_wm pipe_wm = {}; |
3704 | struct intel_pipe_wm pipe_wm = {}; |
3794 | struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm; |
3705 | struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm; |
3795 | struct intel_wm_config config = {}; |
3706 | struct intel_wm_config config = {}; |
3796 | 3707 | ||
3797 | ilk_compute_wm_parameters(crtc, ¶ms); |
3708 | WARN_ON(cstate->base.active != intel_crtc->active); |
3798 | 3709 | ||
3799 | intel_compute_pipe_wm(crtc, ¶ms, &pipe_wm); |
3710 | intel_compute_pipe_wm(cstate, &pipe_wm); |
3800 | 3711 | ||
3801 | if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm))) |
3712 | if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm))) |
3802 | return; |
3713 | return; |
3803 | 3714 | ||
3804 | intel_crtc->wm.active = pipe_wm; |
3715 | intel_crtc->wm.active = pipe_wm; |
3805 | 3716 | ||
3806 | ilk_compute_wm_config(dev, &config); |
3717 | ilk_compute_wm_config(dev, &config); |
3807 | 3718 | ||
3808 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max); |
3719 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max); |
3809 | ilk_wm_merge(dev, &config, &max, &lp_wm_1_2); |
3720 | ilk_wm_merge(dev, &config, &max, &lp_wm_1_2); |
3810 | 3721 | ||
3811 | /* 5/6 split only in single pipe config on IVB+ */ |
3722 | /* 5/6 split only in single pipe config on IVB+ */ |
3812 | if (INTEL_INFO(dev)->gen >= 7 && |
3723 | if (INTEL_INFO(dev)->gen >= 7 && |
3813 | config.num_pipes_active == 1 && config.sprites_enabled) { |
3724 | config.num_pipes_active == 1 && config.sprites_enabled) { |
3814 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max); |
3725 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max); |
3815 | ilk_wm_merge(dev, &config, &max, &lp_wm_5_6); |
3726 | ilk_wm_merge(dev, &config, &max, &lp_wm_5_6); |
3816 | 3727 | ||
3817 | best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6); |
3728 | best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6); |
3818 | } else { |
3729 | } else { |
3819 | best_lp_wm = &lp_wm_1_2; |
3730 | best_lp_wm = &lp_wm_1_2; |
3820 | } |
3731 | } |
3821 | 3732 | ||
3822 | partitioning = (best_lp_wm == &lp_wm_1_2) ? |
3733 | partitioning = (best_lp_wm == &lp_wm_1_2) ? |
3823 | INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6; |
3734 | INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6; |
3824 | 3735 | ||
3825 | ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results); |
3736 | ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results); |
3826 | 3737 | ||
3827 | ilk_write_wm_values(dev_priv, &results); |
3738 | ilk_write_wm_values(dev_priv, &results); |
3828 | } |
3739 | } |
3829 | 3740 | ||
3830 | static void |
3741 | static void |
3831 | ilk_update_sprite_wm(struct drm_plane *plane, |
3742 | ilk_update_sprite_wm(struct drm_plane *plane, |
3832 | struct drm_crtc *crtc, |
3743 | struct drm_crtc *crtc, |
3833 | uint32_t sprite_width, uint32_t sprite_height, |
3744 | uint32_t sprite_width, uint32_t sprite_height, |
3834 | int pixel_size, bool enabled, bool scaled) |
3745 | int pixel_size, bool enabled, bool scaled) |
3835 | { |
3746 | { |
3836 | struct drm_device *dev = plane->dev; |
3747 | struct drm_device *dev = plane->dev; |
3837 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3748 | struct intel_plane *intel_plane = to_intel_plane(plane); |
3838 | - | ||
3839 | intel_plane->wm.enabled = enabled; |
- | |
3840 | intel_plane->wm.scaled = scaled; |
- | |
3841 | intel_plane->wm.horiz_pixels = sprite_width; |
- | |
3842 | intel_plane->wm.vert_pixels = sprite_width; |
- | |
3843 | intel_plane->wm.bytes_per_pixel = pixel_size; |
- | |
3844 | 3749 | ||
3845 | /* |
3750 | /* |
3846 | * IVB workaround: must disable low power watermarks for at least |
3751 | * IVB workaround: must disable low power watermarks for at least |
3847 | * one frame before enabling scaling. LP watermarks can be re-enabled |
3752 | * one frame before enabling scaling. LP watermarks can be re-enabled |
3848 | * when scaling is disabled. |
3753 | * when scaling is disabled. |
3849 | * |
3754 | * |
3850 | * WaCxSRDisabledForSpriteScaling:ivb |
3755 | * WaCxSRDisabledForSpriteScaling:ivb |
3851 | */ |
3756 | */ |
3852 | if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev)) |
3757 | if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev)) |
3853 | intel_wait_for_vblank(dev, intel_plane->pipe); |
3758 | intel_wait_for_vblank(dev, intel_plane->pipe); |
3854 | 3759 | ||
3855 | ilk_update_wm(crtc); |
3760 | ilk_update_wm(crtc); |
3856 | } |
3761 | } |
3857 | 3762 | ||
3858 | static void skl_pipe_wm_active_state(uint32_t val, |
3763 | static void skl_pipe_wm_active_state(uint32_t val, |
3859 | struct skl_pipe_wm *active, |
3764 | struct skl_pipe_wm *active, |
3860 | bool is_transwm, |
3765 | bool is_transwm, |
3861 | bool is_cursor, |
3766 | bool is_cursor, |
3862 | int i, |
3767 | int i, |
3863 | int level) |
3768 | int level) |
3864 | { |
3769 | { |
3865 | bool is_enabled = (val & PLANE_WM_EN) != 0; |
3770 | bool is_enabled = (val & PLANE_WM_EN) != 0; |
3866 | 3771 | ||
3867 | if (!is_transwm) { |
3772 | if (!is_transwm) { |
3868 | if (!is_cursor) { |
3773 | if (!is_cursor) { |
3869 | active->wm[level].plane_en[i] = is_enabled; |
3774 | active->wm[level].plane_en[i] = is_enabled; |
3870 | active->wm[level].plane_res_b[i] = |
3775 | active->wm[level].plane_res_b[i] = |
3871 | val & PLANE_WM_BLOCKS_MASK; |
3776 | val & PLANE_WM_BLOCKS_MASK; |
3872 | active->wm[level].plane_res_l[i] = |
3777 | active->wm[level].plane_res_l[i] = |
3873 | (val >> PLANE_WM_LINES_SHIFT) & |
3778 | (val >> PLANE_WM_LINES_SHIFT) & |
3874 | PLANE_WM_LINES_MASK; |
3779 | PLANE_WM_LINES_MASK; |
3875 | } else { |
3780 | } else { |
3876 | active->wm[level].cursor_en = is_enabled; |
3781 | active->wm[level].plane_en[PLANE_CURSOR] = is_enabled; |
3877 | active->wm[level].cursor_res_b = |
3782 | active->wm[level].plane_res_b[PLANE_CURSOR] = |
3878 | val & PLANE_WM_BLOCKS_MASK; |
3783 | val & PLANE_WM_BLOCKS_MASK; |
3879 | active->wm[level].cursor_res_l = |
3784 | active->wm[level].plane_res_l[PLANE_CURSOR] = |
3880 | (val >> PLANE_WM_LINES_SHIFT) & |
3785 | (val >> PLANE_WM_LINES_SHIFT) & |
3881 | PLANE_WM_LINES_MASK; |
3786 | PLANE_WM_LINES_MASK; |
3882 | } |
3787 | } |
3883 | } else { |
3788 | } else { |
3884 | if (!is_cursor) { |
3789 | if (!is_cursor) { |
3885 | active->trans_wm.plane_en[i] = is_enabled; |
3790 | active->trans_wm.plane_en[i] = is_enabled; |
3886 | active->trans_wm.plane_res_b[i] = |
3791 | active->trans_wm.plane_res_b[i] = |
3887 | val & PLANE_WM_BLOCKS_MASK; |
3792 | val & PLANE_WM_BLOCKS_MASK; |
3888 | active->trans_wm.plane_res_l[i] = |
3793 | active->trans_wm.plane_res_l[i] = |
3889 | (val >> PLANE_WM_LINES_SHIFT) & |
3794 | (val >> PLANE_WM_LINES_SHIFT) & |
3890 | PLANE_WM_LINES_MASK; |
3795 | PLANE_WM_LINES_MASK; |
3891 | } else { |
3796 | } else { |
3892 | active->trans_wm.cursor_en = is_enabled; |
3797 | active->trans_wm.plane_en[PLANE_CURSOR] = is_enabled; |
3893 | active->trans_wm.cursor_res_b = |
3798 | active->trans_wm.plane_res_b[PLANE_CURSOR] = |
3894 | val & PLANE_WM_BLOCKS_MASK; |
3799 | val & PLANE_WM_BLOCKS_MASK; |
3895 | active->trans_wm.cursor_res_l = |
3800 | active->trans_wm.plane_res_l[PLANE_CURSOR] = |
3896 | (val >> PLANE_WM_LINES_SHIFT) & |
3801 | (val >> PLANE_WM_LINES_SHIFT) & |
3897 | PLANE_WM_LINES_MASK; |
3802 | PLANE_WM_LINES_MASK; |
3898 | } |
3803 | } |
3899 | } |
3804 | } |
3900 | } |
3805 | } |
3901 | 3806 | ||
3902 | static void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
3807 | static void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
3903 | { |
3808 | { |
3904 | struct drm_device *dev = crtc->dev; |
3809 | struct drm_device *dev = crtc->dev; |
3905 | struct drm_i915_private *dev_priv = dev->dev_private; |
3810 | struct drm_i915_private *dev_priv = dev->dev_private; |
3906 | struct skl_wm_values *hw = &dev_priv->wm.skl_hw; |
3811 | struct skl_wm_values *hw = &dev_priv->wm.skl_hw; |
3907 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3812 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3908 | struct skl_pipe_wm *active = &intel_crtc->wm.skl_active; |
3813 | struct skl_pipe_wm *active = &intel_crtc->wm.skl_active; |
3909 | enum pipe pipe = intel_crtc->pipe; |
3814 | enum pipe pipe = intel_crtc->pipe; |
3910 | int level, i, max_level; |
3815 | int level, i, max_level; |
3911 | uint32_t temp; |
3816 | uint32_t temp; |
3912 | 3817 | ||
3913 | max_level = ilk_wm_max_level(dev); |
3818 | max_level = ilk_wm_max_level(dev); |
3914 | 3819 | ||
3915 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
3820 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
3916 | 3821 | ||
3917 | for (level = 0; level <= max_level; level++) { |
3822 | for (level = 0; level <= max_level; level++) { |
3918 | for (i = 0; i < intel_num_planes(intel_crtc); i++) |
3823 | for (i = 0; i < intel_num_planes(intel_crtc); i++) |
3919 | hw->plane[pipe][i][level] = |
3824 | hw->plane[pipe][i][level] = |
3920 | I915_READ(PLANE_WM(pipe, i, level)); |
3825 | I915_READ(PLANE_WM(pipe, i, level)); |
3921 | hw->cursor[pipe][level] = I915_READ(CUR_WM(pipe, level)); |
3826 | hw->plane[pipe][PLANE_CURSOR][level] = I915_READ(CUR_WM(pipe, level)); |
3922 | } |
3827 | } |
3923 | 3828 | ||
3924 | for (i = 0; i < intel_num_planes(intel_crtc); i++) |
3829 | for (i = 0; i < intel_num_planes(intel_crtc); i++) |
3925 | hw->plane_trans[pipe][i] = I915_READ(PLANE_WM_TRANS(pipe, i)); |
3830 | hw->plane_trans[pipe][i] = I915_READ(PLANE_WM_TRANS(pipe, i)); |
3926 | hw->cursor_trans[pipe] = I915_READ(CUR_WM_TRANS(pipe)); |
3831 | hw->plane_trans[pipe][PLANE_CURSOR] = I915_READ(CUR_WM_TRANS(pipe)); |
3927 | 3832 | ||
3928 | if (!intel_crtc_active(crtc)) |
3833 | if (!intel_crtc->active) |
3929 | return; |
3834 | return; |
3930 | 3835 | ||
3931 | hw->dirty[pipe] = true; |
3836 | hw->dirty[pipe] = true; |
3932 | 3837 | ||
3933 | active->linetime = hw->wm_linetime[pipe]; |
3838 | active->linetime = hw->wm_linetime[pipe]; |
3934 | 3839 | ||
3935 | for (level = 0; level <= max_level; level++) { |
3840 | for (level = 0; level <= max_level; level++) { |
3936 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3841 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3937 | temp = hw->plane[pipe][i][level]; |
3842 | temp = hw->plane[pipe][i][level]; |
3938 | skl_pipe_wm_active_state(temp, active, false, |
3843 | skl_pipe_wm_active_state(temp, active, false, |
3939 | false, i, level); |
3844 | false, i, level); |
3940 | } |
3845 | } |
3941 | temp = hw->cursor[pipe][level]; |
3846 | temp = hw->plane[pipe][PLANE_CURSOR][level]; |
3942 | skl_pipe_wm_active_state(temp, active, false, true, i, level); |
3847 | skl_pipe_wm_active_state(temp, active, false, true, i, level); |
3943 | } |
3848 | } |
3944 | 3849 | ||
3945 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3850 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3946 | temp = hw->plane_trans[pipe][i]; |
3851 | temp = hw->plane_trans[pipe][i]; |
3947 | skl_pipe_wm_active_state(temp, active, true, false, i, 0); |
3852 | skl_pipe_wm_active_state(temp, active, true, false, i, 0); |
3948 | } |
3853 | } |
3949 | 3854 | ||
3950 | temp = hw->cursor_trans[pipe]; |
3855 | temp = hw->plane_trans[pipe][PLANE_CURSOR]; |
3951 | skl_pipe_wm_active_state(temp, active, true, true, i, 0); |
3856 | skl_pipe_wm_active_state(temp, active, true, true, i, 0); |
3952 | } |
3857 | } |
3953 | 3858 | ||
3954 | void skl_wm_get_hw_state(struct drm_device *dev) |
3859 | void skl_wm_get_hw_state(struct drm_device *dev) |
3955 | { |
3860 | { |
3956 | struct drm_i915_private *dev_priv = dev->dev_private; |
3861 | struct drm_i915_private *dev_priv = dev->dev_private; |
3957 | struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb; |
3862 | struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb; |
3958 | struct drm_crtc *crtc; |
3863 | struct drm_crtc *crtc; |
3959 | 3864 | ||
3960 | skl_ddb_get_hw_state(dev_priv, ddb); |
3865 | skl_ddb_get_hw_state(dev_priv, ddb); |
3961 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) |
3866 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) |
3962 | skl_pipe_wm_get_hw_state(crtc); |
3867 | skl_pipe_wm_get_hw_state(crtc); |
3963 | } |
3868 | } |
3964 | 3869 | ||
3965 | static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
3870 | static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
3966 | { |
3871 | { |
3967 | struct drm_device *dev = crtc->dev; |
3872 | struct drm_device *dev = crtc->dev; |
3968 | struct drm_i915_private *dev_priv = dev->dev_private; |
3873 | struct drm_i915_private *dev_priv = dev->dev_private; |
3969 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
3874 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
3970 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3875 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3971 | struct intel_pipe_wm *active = &intel_crtc->wm.active; |
3876 | struct intel_pipe_wm *active = &intel_crtc->wm.active; |
3972 | enum pipe pipe = intel_crtc->pipe; |
3877 | enum pipe pipe = intel_crtc->pipe; |
3973 | static const unsigned int wm0_pipe_reg[] = { |
3878 | static const unsigned int wm0_pipe_reg[] = { |
3974 | [PIPE_A] = WM0_PIPEA_ILK, |
3879 | [PIPE_A] = WM0_PIPEA_ILK, |
3975 | [PIPE_B] = WM0_PIPEB_ILK, |
3880 | [PIPE_B] = WM0_PIPEB_ILK, |
3976 | [PIPE_C] = WM0_PIPEC_IVB, |
3881 | [PIPE_C] = WM0_PIPEC_IVB, |
3977 | }; |
3882 | }; |
3978 | 3883 | ||
3979 | hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]); |
3884 | hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]); |
3980 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
3885 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
3981 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
3886 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
3982 | 3887 | ||
3983 | active->pipe_enabled = intel_crtc_active(crtc); |
3888 | active->pipe_enabled = intel_crtc->active; |
3984 | 3889 | ||
3985 | if (active->pipe_enabled) { |
3890 | if (active->pipe_enabled) { |
3986 | u32 tmp = hw->wm_pipe[pipe]; |
3891 | u32 tmp = hw->wm_pipe[pipe]; |
3987 | 3892 | ||
3988 | /* |
3893 | /* |
3989 | * For active pipes LP0 watermark is marked as |
3894 | * For active pipes LP0 watermark is marked as |
3990 | * enabled, and LP1+ watermaks as disabled since |
3895 | * enabled, and LP1+ watermaks as disabled since |
3991 | * we can't really reverse compute them in case |
3896 | * we can't really reverse compute them in case |
3992 | * multiple pipes are active. |
3897 | * multiple pipes are active. |
3993 | */ |
3898 | */ |
3994 | active->wm[0].enable = true; |
3899 | active->wm[0].enable = true; |
3995 | active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT; |
3900 | active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT; |
3996 | active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT; |
3901 | active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT; |
3997 | active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK; |
3902 | active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK; |
3998 | active->linetime = hw->wm_linetime[pipe]; |
3903 | active->linetime = hw->wm_linetime[pipe]; |
3999 | } else { |
3904 | } else { |
4000 | int level, max_level = ilk_wm_max_level(dev); |
3905 | int level, max_level = ilk_wm_max_level(dev); |
4001 | 3906 | ||
4002 | /* |
3907 | /* |
4003 | * For inactive pipes, all watermark levels |
3908 | * For inactive pipes, all watermark levels |
4004 | * should be marked as enabled but zeroed, |
3909 | * should be marked as enabled but zeroed, |
4005 | * which is what we'd compute them to. |
3910 | * which is what we'd compute them to. |
4006 | */ |
3911 | */ |
4007 | for (level = 0; level <= max_level; level++) |
3912 | for (level = 0; level <= max_level; level++) |
4008 | active->wm[level].enable = true; |
3913 | active->wm[level].enable = true; |
4009 | } |
3914 | } |
4010 | } |
3915 | } |
- | 3916 | ||
- | 3917 | #define _FW_WM(value, plane) \ |
|
- | 3918 | (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT) |
|
- | 3919 | #define _FW_WM_VLV(value, plane) \ |
|
- | 3920 | (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT) |
|
- | 3921 | ||
- | 3922 | static void vlv_read_wm_values(struct drm_i915_private *dev_priv, |
|
- | 3923 | struct vlv_wm_values *wm) |
|
- | 3924 | { |
|
- | 3925 | enum pipe pipe; |
|
- | 3926 | uint32_t tmp; |
|
- | 3927 | ||
- | 3928 | for_each_pipe(dev_priv, pipe) { |
|
- | 3929 | tmp = I915_READ(VLV_DDL(pipe)); |
|
- | 3930 | ||
- | 3931 | wm->ddl[pipe].primary = |
|
- | 3932 | (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
|
- | 3933 | wm->ddl[pipe].cursor = |
|
- | 3934 | (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
|
- | 3935 | wm->ddl[pipe].sprite[0] = |
|
- | 3936 | (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
|
- | 3937 | wm->ddl[pipe].sprite[1] = |
|
- | 3938 | (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
|
- | 3939 | } |
|
- | 3940 | ||
- | 3941 | tmp = I915_READ(DSPFW1); |
|
- | 3942 | wm->sr.plane = _FW_WM(tmp, SR); |
|
- | 3943 | wm->pipe[PIPE_B].cursor = _FW_WM(tmp, CURSORB); |
|
- | 3944 | wm->pipe[PIPE_B].primary = _FW_WM_VLV(tmp, PLANEB); |
|
- | 3945 | wm->pipe[PIPE_A].primary = _FW_WM_VLV(tmp, PLANEA); |
|
- | 3946 | ||
- | 3947 | tmp = I915_READ(DSPFW2); |
|
- | 3948 | wm->pipe[PIPE_A].sprite[1] = _FW_WM_VLV(tmp, SPRITEB); |
|
- | 3949 | wm->pipe[PIPE_A].cursor = _FW_WM(tmp, CURSORA); |
|
- | 3950 | wm->pipe[PIPE_A].sprite[0] = _FW_WM_VLV(tmp, SPRITEA); |
|
- | 3951 | ||
- | 3952 | tmp = I915_READ(DSPFW3); |
|
- | 3953 | wm->sr.cursor = _FW_WM(tmp, CURSOR_SR); |
|
- | 3954 | ||
- | 3955 | if (IS_CHERRYVIEW(dev_priv)) { |
|
- | 3956 | tmp = I915_READ(DSPFW7_CHV); |
|
- | 3957 | wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED); |
|
- | 3958 | wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC); |
|
- | 3959 | ||
- | 3960 | tmp = I915_READ(DSPFW8_CHV); |
|
- | 3961 | wm->pipe[PIPE_C].sprite[1] = _FW_WM_VLV(tmp, SPRITEF); |
|
- | 3962 | wm->pipe[PIPE_C].sprite[0] = _FW_WM_VLV(tmp, SPRITEE); |
|
- | 3963 | ||
- | 3964 | tmp = I915_READ(DSPFW9_CHV); |
|
- | 3965 | wm->pipe[PIPE_C].primary = _FW_WM_VLV(tmp, PLANEC); |
|
- | 3966 | wm->pipe[PIPE_C].cursor = _FW_WM(tmp, CURSORC); |
|
- | 3967 | ||
- | 3968 | tmp = I915_READ(DSPHOWM); |
|
- | 3969 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; |
|
- | 3970 | wm->pipe[PIPE_C].sprite[1] |= _FW_WM(tmp, SPRITEF_HI) << 8; |
|
- | 3971 | wm->pipe[PIPE_C].sprite[0] |= _FW_WM(tmp, SPRITEE_HI) << 8; |
|
- | 3972 | wm->pipe[PIPE_C].primary |= _FW_WM(tmp, PLANEC_HI) << 8; |
|
- | 3973 | wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8; |
|
- | 3974 | wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8; |
|
- | 3975 | wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8; |
|
- | 3976 | wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8; |
|
- | 3977 | wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8; |
|
- | 3978 | wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8; |
|
- | 3979 | } else { |
|
- | 3980 | tmp = I915_READ(DSPFW7); |
|
- | 3981 | wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED); |
|
- | 3982 | wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC); |
|
- | 3983 | ||
- | 3984 | tmp = I915_READ(DSPHOWM); |
|
- | 3985 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; |
|
- | 3986 | wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8; |
|
- | 3987 | wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8; |
|
- | 3988 | wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8; |
|
- | 3989 | wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8; |
|
- | 3990 | wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8; |
|
- | 3991 | wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8; |
|
- | 3992 | } |
|
- | 3993 | } |
|
- | 3994 | ||
- | 3995 | #undef _FW_WM |
|
- | 3996 | #undef _FW_WM_VLV |
|
- | 3997 | ||
- | 3998 | void vlv_wm_get_hw_state(struct drm_device *dev) |
|
- | 3999 | { |
|
- | 4000 | struct drm_i915_private *dev_priv = to_i915(dev); |
|
- | 4001 | struct vlv_wm_values *wm = &dev_priv->wm.vlv; |
|
- | 4002 | struct intel_plane *plane; |
|
- | 4003 | enum pipe pipe; |
|
- | 4004 | u32 val; |
|
- | 4005 | ||
- | 4006 | vlv_read_wm_values(dev_priv, wm); |
|
- | 4007 | ||
- | 4008 | for_each_intel_plane(dev, plane) { |
|
- | 4009 | switch (plane->base.type) { |
|
- | 4010 | int sprite; |
|
- | 4011 | case DRM_PLANE_TYPE_CURSOR: |
|
- | 4012 | plane->wm.fifo_size = 63; |
|
- | 4013 | break; |
|
- | 4014 | case DRM_PLANE_TYPE_PRIMARY: |
|
- | 4015 | plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, 0); |
|
- | 4016 | break; |
|
- | 4017 | case DRM_PLANE_TYPE_OVERLAY: |
|
- | 4018 | sprite = plane->plane; |
|
- | 4019 | plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, sprite + 1); |
|
- | 4020 | break; |
|
- | 4021 | } |
|
- | 4022 | } |
|
- | 4023 | ||
- | 4024 | wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN; |
|
- | 4025 | wm->level = VLV_WM_LEVEL_PM2; |
|
- | 4026 | ||
- | 4027 | if (IS_CHERRYVIEW(dev_priv)) { |
|
- | 4028 | mutex_lock(&dev_priv->rps.hw_lock); |
|
- | 4029 | ||
- | 4030 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); |
|
- | 4031 | if (val & DSP_MAXFIFO_PM5_ENABLE) |
|
- | 4032 | wm->level = VLV_WM_LEVEL_PM5; |
|
- | 4033 | ||
- | 4034 | /* |
|
- | 4035 | * If DDR DVFS is disabled in the BIOS, Punit |
|
- | 4036 | * will never ack the request. So if that happens |
|
- | 4037 | * assume we don't have to enable/disable DDR DVFS |
|
- | 4038 | * dynamically. To test that just set the REQ_ACK |
|
- | 4039 | * bit to poke the Punit, but don't change the |
|
- | 4040 | * HIGH/LOW bits so that we don't actually change |
|
- | 4041 | * the current state. |
|
- | 4042 | */ |
|
- | 4043 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); |
|
- | 4044 | val |= FORCE_DDR_FREQ_REQ_ACK; |
|
- | 4045 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); |
|
- | 4046 | ||
- | 4047 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & |
|
- | 4048 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) { |
|
- | 4049 | DRM_DEBUG_KMS("Punit not acking DDR DVFS request, " |
|
- | 4050 | "assuming DDR DVFS is disabled\n"); |
|
- | 4051 | dev_priv->wm.max_level = VLV_WM_LEVEL_PM5; |
|
- | 4052 | } else { |
|
- | 4053 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); |
|
- | 4054 | if ((val & FORCE_DDR_HIGH_FREQ) == 0) |
|
- | 4055 | wm->level = VLV_WM_LEVEL_DDR_DVFS; |
|
- | 4056 | } |
|
- | 4057 | ||
- | 4058 | mutex_unlock(&dev_priv->rps.hw_lock); |
|
- | 4059 | } |
|
- | 4060 | ||
- | 4061 | for_each_pipe(dev_priv, pipe) |
|
- | 4062 | DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n", |
|
- | 4063 | pipe_name(pipe), wm->pipe[pipe].primary, wm->pipe[pipe].cursor, |
|
- | 4064 | wm->pipe[pipe].sprite[0], wm->pipe[pipe].sprite[1]); |
|
- | 4065 | ||
- | 4066 | DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n", |
|
- | 4067 | wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr); |
|
- | 4068 | } |
|
4011 | 4069 | ||
4012 | void ilk_wm_get_hw_state(struct drm_device *dev) |
4070 | void ilk_wm_get_hw_state(struct drm_device *dev) |
4013 | { |
4071 | { |
4014 | struct drm_i915_private *dev_priv = dev->dev_private; |
4072 | struct drm_i915_private *dev_priv = dev->dev_private; |
4015 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
4073 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
4016 | struct drm_crtc *crtc; |
4074 | struct drm_crtc *crtc; |
4017 | 4075 | ||
4018 | for_each_crtc(dev, crtc) |
4076 | for_each_crtc(dev, crtc) |
4019 | ilk_pipe_wm_get_hw_state(crtc); |
4077 | ilk_pipe_wm_get_hw_state(crtc); |
4020 | 4078 | ||
4021 | hw->wm_lp[0] = I915_READ(WM1_LP_ILK); |
4079 | hw->wm_lp[0] = I915_READ(WM1_LP_ILK); |
4022 | hw->wm_lp[1] = I915_READ(WM2_LP_ILK); |
4080 | hw->wm_lp[1] = I915_READ(WM2_LP_ILK); |
4023 | hw->wm_lp[2] = I915_READ(WM3_LP_ILK); |
4081 | hw->wm_lp[2] = I915_READ(WM3_LP_ILK); |
4024 | 4082 | ||
4025 | hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK); |
4083 | hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK); |
4026 | if (INTEL_INFO(dev)->gen >= 7) { |
4084 | if (INTEL_INFO(dev)->gen >= 7) { |
4027 | hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB); |
4085 | hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB); |
4028 | hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB); |
4086 | hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB); |
4029 | } |
4087 | } |
4030 | 4088 | ||
4031 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
4089 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
4032 | hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ? |
4090 | hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ? |
4033 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; |
4091 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; |
4034 | else if (IS_IVYBRIDGE(dev)) |
4092 | else if (IS_IVYBRIDGE(dev)) |
4035 | hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ? |
4093 | hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ? |
4036 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; |
4094 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; |
4037 | 4095 | ||
4038 | hw->enable_fbc_wm = |
4096 | hw->enable_fbc_wm = |
4039 | !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS); |
4097 | !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS); |
4040 | } |
4098 | } |
4041 | 4099 | ||
4042 | /** |
4100 | /** |
4043 | * intel_update_watermarks - update FIFO watermark values based on current modes |
4101 | * intel_update_watermarks - update FIFO watermark values based on current modes |
4044 | * |
4102 | * |
4045 | * Calculate watermark values for the various WM regs based on current mode |
4103 | * Calculate watermark values for the various WM regs based on current mode |
4046 | * and plane configuration. |
4104 | * and plane configuration. |
4047 | * |
4105 | * |
4048 | * There are several cases to deal with here: |
4106 | * There are several cases to deal with here: |
4049 | * - normal (i.e. non-self-refresh) |
4107 | * - normal (i.e. non-self-refresh) |
4050 | * - self-refresh (SR) mode |
4108 | * - self-refresh (SR) mode |
4051 | * - lines are large relative to FIFO size (buffer can hold up to 2) |
4109 | * - lines are large relative to FIFO size (buffer can hold up to 2) |
4052 | * - lines are small relative to FIFO size (buffer can hold more than 2 |
4110 | * - lines are small relative to FIFO size (buffer can hold more than 2 |
4053 | * lines), so need to account for TLB latency |
4111 | * lines), so need to account for TLB latency |
4054 | * |
4112 | * |
4055 | * The normal calculation is: |
4113 | * The normal calculation is: |
4056 | * watermark = dotclock * bytes per pixel * latency |
4114 | * watermark = dotclock * bytes per pixel * latency |
4057 | * where latency is platform & configuration dependent (we assume pessimal |
4115 | * where latency is platform & configuration dependent (we assume pessimal |
4058 | * values here). |
4116 | * values here). |
4059 | * |
4117 | * |
4060 | * The SR calculation is: |
4118 | * The SR calculation is: |
4061 | * watermark = (trunc(latency/line time)+1) * surface width * |
4119 | * watermark = (trunc(latency/line time)+1) * surface width * |
4062 | * bytes per pixel |
4120 | * bytes per pixel |
4063 | * where |
4121 | * where |
4064 | * line time = htotal / dotclock |
4122 | * line time = htotal / dotclock |
4065 | * surface width = hdisplay for normal plane and 64 for cursor |
4123 | * surface width = hdisplay for normal plane and 64 for cursor |
4066 | * and latency is assumed to be high, as above. |
4124 | * and latency is assumed to be high, as above. |
4067 | * |
4125 | * |
4068 | * The final value programmed to the register should always be rounded up, |
4126 | * The final value programmed to the register should always be rounded up, |
4069 | * and include an extra 2 entries to account for clock crossings. |
4127 | * and include an extra 2 entries to account for clock crossings. |
4070 | * |
4128 | * |
4071 | * We don't use the sprite, so we can ignore that. And on Crestline we have |
4129 | * We don't use the sprite, so we can ignore that. And on Crestline we have |
4072 | * to set the non-SR watermarks to 8. |
4130 | * to set the non-SR watermarks to 8. |
4073 | */ |
4131 | */ |
4074 | void intel_update_watermarks(struct drm_crtc *crtc) |
4132 | void intel_update_watermarks(struct drm_crtc *crtc) |
4075 | { |
4133 | { |
4076 | struct drm_i915_private *dev_priv = crtc->dev->dev_private; |
4134 | struct drm_i915_private *dev_priv = crtc->dev->dev_private; |
4077 | 4135 | ||
4078 | if (dev_priv->display.update_wm) |
4136 | if (dev_priv->display.update_wm) |
4079 | dev_priv->display.update_wm(crtc); |
4137 | dev_priv->display.update_wm(crtc); |
4080 | } |
4138 | } |
4081 | 4139 | ||
4082 | void intel_update_sprite_watermarks(struct drm_plane *plane, |
4140 | void intel_update_sprite_watermarks(struct drm_plane *plane, |
4083 | struct drm_crtc *crtc, |
4141 | struct drm_crtc *crtc, |
4084 | uint32_t sprite_width, |
4142 | uint32_t sprite_width, |
4085 | uint32_t sprite_height, |
4143 | uint32_t sprite_height, |
4086 | int pixel_size, |
4144 | int pixel_size, |
4087 | bool enabled, bool scaled) |
4145 | bool enabled, bool scaled) |
4088 | { |
4146 | { |
4089 | struct drm_i915_private *dev_priv = plane->dev->dev_private; |
4147 | struct drm_i915_private *dev_priv = plane->dev->dev_private; |
4090 | 4148 | ||
4091 | if (dev_priv->display.update_sprite_wm) |
4149 | if (dev_priv->display.update_sprite_wm) |
4092 | dev_priv->display.update_sprite_wm(plane, crtc, |
4150 | dev_priv->display.update_sprite_wm(plane, crtc, |
4093 | sprite_width, sprite_height, |
4151 | sprite_width, sprite_height, |
4094 | pixel_size, enabled, scaled); |
4152 | pixel_size, enabled, scaled); |
4095 | } |
4153 | } |
4096 | - | ||
4097 | static struct drm_i915_gem_object * |
- | |
4098 | intel_alloc_context_page(struct drm_device *dev) |
- | |
4099 | { |
- | |
4100 | struct drm_i915_gem_object *ctx; |
- | |
4101 | int ret; |
- | |
4102 | - | ||
4103 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
- | |
4104 | - | ||
4105 | ctx = i915_gem_alloc_object(dev, 4096); |
- | |
4106 | if (!ctx) { |
- | |
4107 | DRM_DEBUG("failed to alloc power context, RC6 disabled\n"); |
- | |
4108 | return NULL; |
- | |
4109 | } |
- | |
4110 | - | ||
4111 | ret = i915_gem_obj_ggtt_pin(ctx, 4096, 0); |
- | |
4112 | if (ret) { |
- | |
4113 | DRM_ERROR("failed to pin power context: %d\n", ret); |
- | |
4114 | goto err_unref; |
- | |
4115 | } |
- | |
4116 | - | ||
4117 | ret = i915_gem_object_set_to_gtt_domain(ctx, 1); |
- | |
4118 | if (ret) { |
- | |
4119 | DRM_ERROR("failed to set-domain on power context: %d\n", ret); |
- | |
4120 | goto err_unpin; |
- | |
4121 | } |
- | |
4122 | - | ||
4123 | return ctx; |
- | |
4124 | - | ||
4125 | err_unpin: |
- | |
4126 | i915_gem_object_ggtt_unpin(ctx); |
- | |
4127 | err_unref: |
- | |
4128 | drm_gem_object_unreference(&ctx->base); |
- | |
4129 | return NULL; |
- | |
4130 | } |
- | |
4131 | 4154 | ||
4132 | /** |
4155 | /** |
4133 | * Lock protecting IPS related data structures |
4156 | * Lock protecting IPS related data structures |
4134 | */ |
4157 | */ |
4135 | DEFINE_SPINLOCK(mchdev_lock); |
4158 | DEFINE_SPINLOCK(mchdev_lock); |
4136 | 4159 | ||
4137 | /* Global for IPS driver to get at the current i915 device. Protected by |
4160 | /* Global for IPS driver to get at the current i915 device. Protected by |
4138 | * mchdev_lock. */ |
4161 | * mchdev_lock. */ |
4139 | static struct drm_i915_private *i915_mch_dev; |
4162 | static struct drm_i915_private *i915_mch_dev; |
4140 | 4163 | ||
4141 | bool ironlake_set_drps(struct drm_device *dev, u8 val) |
4164 | bool ironlake_set_drps(struct drm_device *dev, u8 val) |
4142 | { |
4165 | { |
4143 | struct drm_i915_private *dev_priv = dev->dev_private; |
4166 | struct drm_i915_private *dev_priv = dev->dev_private; |
4144 | u16 rgvswctl; |
4167 | u16 rgvswctl; |
4145 | 4168 | ||
4146 | assert_spin_locked(&mchdev_lock); |
4169 | assert_spin_locked(&mchdev_lock); |
4147 | 4170 | ||
4148 | rgvswctl = I915_READ16(MEMSWCTL); |
4171 | rgvswctl = I915_READ16(MEMSWCTL); |
4149 | if (rgvswctl & MEMCTL_CMD_STS) { |
4172 | if (rgvswctl & MEMCTL_CMD_STS) { |
4150 | DRM_DEBUG("gpu busy, RCS change rejected\n"); |
4173 | DRM_DEBUG("gpu busy, RCS change rejected\n"); |
4151 | return false; /* still busy with another command */ |
4174 | return false; /* still busy with another command */ |
4152 | } |
4175 | } |
4153 | 4176 | ||
4154 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | |
4177 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | |
4155 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; |
4178 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; |
4156 | I915_WRITE16(MEMSWCTL, rgvswctl); |
4179 | I915_WRITE16(MEMSWCTL, rgvswctl); |
4157 | POSTING_READ16(MEMSWCTL); |
4180 | POSTING_READ16(MEMSWCTL); |
4158 | 4181 | ||
4159 | rgvswctl |= MEMCTL_CMD_STS; |
4182 | rgvswctl |= MEMCTL_CMD_STS; |
4160 | I915_WRITE16(MEMSWCTL, rgvswctl); |
4183 | I915_WRITE16(MEMSWCTL, rgvswctl); |
4161 | 4184 | ||
4162 | return true; |
4185 | return true; |
4163 | } |
4186 | } |
4164 | 4187 | ||
4165 | static void ironlake_enable_drps(struct drm_device *dev) |
4188 | static void ironlake_enable_drps(struct drm_device *dev) |
4166 | { |
4189 | { |
4167 | struct drm_i915_private *dev_priv = dev->dev_private; |
4190 | struct drm_i915_private *dev_priv = dev->dev_private; |
4168 | u32 rgvmodectl = I915_READ(MEMMODECTL); |
4191 | u32 rgvmodectl = I915_READ(MEMMODECTL); |
4169 | u8 fmax, fmin, fstart, vstart; |
4192 | u8 fmax, fmin, fstart, vstart; |
4170 | 4193 | ||
4171 | spin_lock_irq(&mchdev_lock); |
4194 | spin_lock_irq(&mchdev_lock); |
4172 | 4195 | ||
4173 | /* Enable temp reporting */ |
4196 | /* Enable temp reporting */ |
4174 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); |
4197 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); |
4175 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); |
4198 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); |
4176 | 4199 | ||
4177 | /* 100ms RC evaluation intervals */ |
4200 | /* 100ms RC evaluation intervals */ |
4178 | I915_WRITE(RCUPEI, 100000); |
4201 | I915_WRITE(RCUPEI, 100000); |
4179 | I915_WRITE(RCDNEI, 100000); |
4202 | I915_WRITE(RCDNEI, 100000); |
4180 | 4203 | ||
4181 | /* Set max/min thresholds to 90ms and 80ms respectively */ |
4204 | /* Set max/min thresholds to 90ms and 80ms respectively */ |
4182 | I915_WRITE(RCBMAXAVG, 90000); |
4205 | I915_WRITE(RCBMAXAVG, 90000); |
4183 | I915_WRITE(RCBMINAVG, 80000); |
4206 | I915_WRITE(RCBMINAVG, 80000); |
4184 | 4207 | ||
4185 | I915_WRITE(MEMIHYST, 1); |
4208 | I915_WRITE(MEMIHYST, 1); |
4186 | 4209 | ||
4187 | /* Set up min, max, and cur for interrupt handling */ |
4210 | /* Set up min, max, and cur for interrupt handling */ |
4188 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; |
4211 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; |
4189 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); |
4212 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); |
4190 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> |
4213 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> |
4191 | MEMMODE_FSTART_SHIFT; |
4214 | MEMMODE_FSTART_SHIFT; |
4192 | 4215 | ||
4193 | vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >> |
4216 | vstart = (I915_READ(PXVFREQ(fstart)) & PXVFREQ_PX_MASK) >> |
4194 | PXVFREQ_PX_SHIFT; |
4217 | PXVFREQ_PX_SHIFT; |
4195 | 4218 | ||
4196 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
4219 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
4197 | dev_priv->ips.fstart = fstart; |
4220 | dev_priv->ips.fstart = fstart; |
4198 | 4221 | ||
4199 | dev_priv->ips.max_delay = fstart; |
4222 | dev_priv->ips.max_delay = fstart; |
4200 | dev_priv->ips.min_delay = fmin; |
4223 | dev_priv->ips.min_delay = fmin; |
4201 | dev_priv->ips.cur_delay = fstart; |
4224 | dev_priv->ips.cur_delay = fstart; |
4202 | 4225 | ||
4203 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", |
4226 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", |
4204 | fmax, fmin, fstart); |
4227 | fmax, fmin, fstart); |
4205 | 4228 | ||
4206 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); |
4229 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); |
4207 | 4230 | ||
4208 | /* |
4231 | /* |
4209 | * Interrupts will be enabled in ironlake_irq_postinstall |
4232 | * Interrupts will be enabled in ironlake_irq_postinstall |
4210 | */ |
4233 | */ |
4211 | 4234 | ||
4212 | I915_WRITE(VIDSTART, vstart); |
4235 | I915_WRITE(VIDSTART, vstart); |
4213 | POSTING_READ(VIDSTART); |
4236 | POSTING_READ(VIDSTART); |
4214 | 4237 | ||
4215 | rgvmodectl |= MEMMODE_SWMODE_EN; |
4238 | rgvmodectl |= MEMMODE_SWMODE_EN; |
4216 | I915_WRITE(MEMMODECTL, rgvmodectl); |
4239 | I915_WRITE(MEMMODECTL, rgvmodectl); |
4217 | 4240 | ||
4218 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
4241 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
4219 | DRM_ERROR("stuck trying to change perf mode\n"); |
4242 | DRM_ERROR("stuck trying to change perf mode\n"); |
4220 | mdelay(1); |
4243 | mdelay(1); |
4221 | 4244 | ||
4222 | ironlake_set_drps(dev, fstart); |
4245 | ironlake_set_drps(dev, fstart); |
4223 | 4246 | ||
4224 | dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) + |
4247 | dev_priv->ips.last_count1 = I915_READ(DMIEC) + |
4225 | I915_READ(0x112e0); |
4248 | I915_READ(DDREC) + I915_READ(CSIEC); |
4226 | dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies); |
4249 | dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies); |
4227 | dev_priv->ips.last_count2 = I915_READ(0x112f4); |
4250 | dev_priv->ips.last_count2 = I915_READ(GFXEC); |
4228 | dev_priv->ips.last_time2 = ktime_get_raw_ns(); |
4251 | dev_priv->ips.last_time2 = ktime_get_raw_ns(); |
4229 | 4252 | ||
4230 | spin_unlock_irq(&mchdev_lock); |
4253 | spin_unlock_irq(&mchdev_lock); |
4231 | } |
4254 | } |
4232 | 4255 | ||
4233 | static void ironlake_disable_drps(struct drm_device *dev) |
4256 | static void ironlake_disable_drps(struct drm_device *dev) |
4234 | { |
4257 | { |
4235 | struct drm_i915_private *dev_priv = dev->dev_private; |
4258 | struct drm_i915_private *dev_priv = dev->dev_private; |
4236 | u16 rgvswctl; |
4259 | u16 rgvswctl; |
4237 | 4260 | ||
4238 | spin_lock_irq(&mchdev_lock); |
4261 | spin_lock_irq(&mchdev_lock); |
4239 | 4262 | ||
4240 | rgvswctl = I915_READ16(MEMSWCTL); |
4263 | rgvswctl = I915_READ16(MEMSWCTL); |
4241 | 4264 | ||
4242 | /* Ack interrupts, disable EFC interrupt */ |
4265 | /* Ack interrupts, disable EFC interrupt */ |
4243 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); |
4266 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); |
4244 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); |
4267 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); |
4245 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); |
4268 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); |
4246 | I915_WRITE(DEIIR, DE_PCU_EVENT); |
4269 | I915_WRITE(DEIIR, DE_PCU_EVENT); |
4247 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); |
4270 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); |
4248 | 4271 | ||
4249 | /* Go back to the starting frequency */ |
4272 | /* Go back to the starting frequency */ |
4250 | ironlake_set_drps(dev, dev_priv->ips.fstart); |
4273 | ironlake_set_drps(dev, dev_priv->ips.fstart); |
4251 | mdelay(1); |
4274 | mdelay(1); |
4252 | rgvswctl |= MEMCTL_CMD_STS; |
4275 | rgvswctl |= MEMCTL_CMD_STS; |
4253 | I915_WRITE(MEMSWCTL, rgvswctl); |
4276 | I915_WRITE(MEMSWCTL, rgvswctl); |
4254 | mdelay(1); |
4277 | mdelay(1); |
4255 | 4278 | ||
4256 | spin_unlock_irq(&mchdev_lock); |
4279 | spin_unlock_irq(&mchdev_lock); |
4257 | } |
4280 | } |
4258 | 4281 | ||
4259 | /* There's a funny hw issue where the hw returns all 0 when reading from |
4282 | /* There's a funny hw issue where the hw returns all 0 when reading from |
4260 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value |
4283 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value |
4261 | * ourselves, instead of doing a rmw cycle (which might result in us clearing |
4284 | * ourselves, instead of doing a rmw cycle (which might result in us clearing |
4262 | * all limits and the gpu stuck at whatever frequency it is at atm). |
4285 | * all limits and the gpu stuck at whatever frequency it is at atm). |
4263 | */ |
4286 | */ |
4264 | static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 val) |
4287 | static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val) |
4265 | { |
4288 | { |
4266 | u32 limits; |
4289 | u32 limits; |
4267 | 4290 | ||
4268 | /* Only set the down limit when we've reached the lowest level to avoid |
4291 | /* Only set the down limit when we've reached the lowest level to avoid |
4269 | * getting more interrupts, otherwise leave this clear. This prevents a |
4292 | * getting more interrupts, otherwise leave this clear. This prevents a |
4270 | * race in the hw when coming out of rc6: There's a tiny window where |
4293 | * race in the hw when coming out of rc6: There's a tiny window where |
4271 | * the hw runs at the minimal clock before selecting the desired |
4294 | * the hw runs at the minimal clock before selecting the desired |
4272 | * frequency, if the down threshold expires in that window we will not |
4295 | * frequency, if the down threshold expires in that window we will not |
4273 | * receive a down interrupt. */ |
4296 | * receive a down interrupt. */ |
- | 4297 | if (IS_GEN9(dev_priv->dev)) { |
|
- | 4298 | limits = (dev_priv->rps.max_freq_softlimit) << 23; |
|
- | 4299 | if (val <= dev_priv->rps.min_freq_softlimit) |
|
- | 4300 | limits |= (dev_priv->rps.min_freq_softlimit) << 14; |
|
- | 4301 | } else { |
|
4274 | limits = dev_priv->rps.max_freq_softlimit << 24; |
4302 | limits = dev_priv->rps.max_freq_softlimit << 24; |
4275 | if (val <= dev_priv->rps.min_freq_softlimit) |
4303 | if (val <= dev_priv->rps.min_freq_softlimit) |
4276 | limits |= dev_priv->rps.min_freq_softlimit << 16; |
4304 | limits |= dev_priv->rps.min_freq_softlimit << 16; |
- | 4305 | } |
|
4277 | 4306 | ||
4278 | return limits; |
4307 | return limits; |
4279 | } |
4308 | } |
4280 | 4309 | ||
4281 | static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val) |
4310 | static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val) |
4282 | { |
4311 | { |
4283 | int new_power; |
4312 | int new_power; |
- | 4313 | u32 threshold_up = 0, threshold_down = 0; /* in % */ |
|
- | 4314 | u32 ei_up = 0, ei_down = 0; |
|
4284 | 4315 | ||
4285 | new_power = dev_priv->rps.power; |
4316 | new_power = dev_priv->rps.power; |
4286 | switch (dev_priv->rps.power) { |
4317 | switch (dev_priv->rps.power) { |
4287 | case LOW_POWER: |
4318 | case LOW_POWER: |
4288 | if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq) |
4319 | if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq) |
4289 | new_power = BETWEEN; |
4320 | new_power = BETWEEN; |
4290 | break; |
4321 | break; |
4291 | 4322 | ||
4292 | case BETWEEN: |
4323 | case BETWEEN: |
4293 | if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq) |
4324 | if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq) |
4294 | new_power = LOW_POWER; |
4325 | new_power = LOW_POWER; |
4295 | else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq) |
4326 | else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq) |
4296 | new_power = HIGH_POWER; |
4327 | new_power = HIGH_POWER; |
4297 | break; |
4328 | break; |
4298 | 4329 | ||
4299 | case HIGH_POWER: |
4330 | case HIGH_POWER: |
4300 | if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq) |
4331 | if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq) |
4301 | new_power = BETWEEN; |
4332 | new_power = BETWEEN; |
4302 | break; |
4333 | break; |
4303 | } |
4334 | } |
4304 | /* Max/min bins are special */ |
4335 | /* Max/min bins are special */ |
4305 | if (val == dev_priv->rps.min_freq_softlimit) |
4336 | if (val <= dev_priv->rps.min_freq_softlimit) |
4306 | new_power = LOW_POWER; |
4337 | new_power = LOW_POWER; |
4307 | if (val == dev_priv->rps.max_freq_softlimit) |
4338 | if (val >= dev_priv->rps.max_freq_softlimit) |
4308 | new_power = HIGH_POWER; |
4339 | new_power = HIGH_POWER; |
4309 | if (new_power == dev_priv->rps.power) |
4340 | if (new_power == dev_priv->rps.power) |
4310 | return; |
4341 | return; |
4311 | 4342 | ||
4312 | /* Note the units here are not exactly 1us, but 1280ns. */ |
4343 | /* Note the units here are not exactly 1us, but 1280ns. */ |
4313 | switch (new_power) { |
4344 | switch (new_power) { |
4314 | case LOW_POWER: |
4345 | case LOW_POWER: |
4315 | /* Upclock if more than 95% busy over 16ms */ |
4346 | /* Upclock if more than 95% busy over 16ms */ |
4316 | I915_WRITE(GEN6_RP_UP_EI, 12500); |
4347 | ei_up = 16000; |
4317 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 11800); |
4348 | threshold_up = 95; |
4318 | 4349 | ||
4319 | /* Downclock if less than 85% busy over 32ms */ |
- | |
4320 | I915_WRITE(GEN6_RP_DOWN_EI, 25000); |
- | |
4321 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 21250); |
- | |
4322 | - | ||
4323 | I915_WRITE(GEN6_RP_CONTROL, |
4350 | /* Downclock if less than 85% busy over 32ms */ |
4324 | GEN6_RP_MEDIA_TURBO | |
- | |
4325 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
- | |
4326 | GEN6_RP_MEDIA_IS_GFX | |
4351 | ei_down = 32000; |
4327 | GEN6_RP_ENABLE | |
- | |
4328 | GEN6_RP_UP_BUSY_AVG | |
- | |
4329 | GEN6_RP_DOWN_IDLE_AVG); |
4352 | threshold_down = 85; |
4330 | break; |
4353 | break; |
4331 | 4354 | ||
4332 | case BETWEEN: |
4355 | case BETWEEN: |
4333 | /* Upclock if more than 90% busy over 13ms */ |
4356 | /* Upclock if more than 90% busy over 13ms */ |
4334 | I915_WRITE(GEN6_RP_UP_EI, 10250); |
4357 | ei_up = 13000; |
4335 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 9225); |
4358 | threshold_up = 90; |
4336 | 4359 | ||
4337 | /* Downclock if less than 75% busy over 32ms */ |
- | |
4338 | I915_WRITE(GEN6_RP_DOWN_EI, 25000); |
- | |
4339 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 18750); |
- | |
4340 | - | ||
4341 | I915_WRITE(GEN6_RP_CONTROL, |
4360 | /* Downclock if less than 75% busy over 32ms */ |
4342 | GEN6_RP_MEDIA_TURBO | |
- | |
4343 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
- | |
4344 | GEN6_RP_MEDIA_IS_GFX | |
4361 | ei_down = 32000; |
4345 | GEN6_RP_ENABLE | |
- | |
4346 | GEN6_RP_UP_BUSY_AVG | |
- | |
4347 | GEN6_RP_DOWN_IDLE_AVG); |
4362 | threshold_down = 75; |
4348 | break; |
4363 | break; |
4349 | 4364 | ||
4350 | case HIGH_POWER: |
4365 | case HIGH_POWER: |
4351 | /* Upclock if more than 85% busy over 10ms */ |
4366 | /* Upclock if more than 85% busy over 10ms */ |
4352 | I915_WRITE(GEN6_RP_UP_EI, 8000); |
4367 | ei_up = 10000; |
4353 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 6800); |
4368 | threshold_up = 85; |
4354 | 4369 | ||
- | 4370 | /* Downclock if less than 60% busy over 32ms */ |
|
- | 4371 | ei_down = 32000; |
|
- | 4372 | threshold_down = 60; |
|
- | 4373 | break; |
|
- | 4374 | } |
|
- | 4375 | ||
- | 4376 | I915_WRITE(GEN6_RP_UP_EI, |
|
- | 4377 | GT_INTERVAL_FROM_US(dev_priv, ei_up)); |
|
- | 4378 | I915_WRITE(GEN6_RP_UP_THRESHOLD, |
|
- | 4379 | GT_INTERVAL_FROM_US(dev_priv, (ei_up * threshold_up / 100))); |
|
4355 | /* Downclock if less than 60% busy over 32ms */ |
4380 | |
- | 4381 | I915_WRITE(GEN6_RP_DOWN_EI, |
|
4356 | I915_WRITE(GEN6_RP_DOWN_EI, 25000); |
4382 | GT_INTERVAL_FROM_US(dev_priv, ei_down)); |
- | 4383 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, |
|
4357 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 15000); |
4384 | GT_INTERVAL_FROM_US(dev_priv, (ei_down * threshold_down / 100))); |
4358 | 4385 | ||
4359 | I915_WRITE(GEN6_RP_CONTROL, |
4386 | I915_WRITE(GEN6_RP_CONTROL, |
4360 | GEN6_RP_MEDIA_TURBO | |
4387 | GEN6_RP_MEDIA_TURBO | |
4361 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
4388 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
4362 | GEN6_RP_MEDIA_IS_GFX | |
4389 | GEN6_RP_MEDIA_IS_GFX | |
4363 | GEN6_RP_ENABLE | |
4390 | GEN6_RP_ENABLE | |
4364 | GEN6_RP_UP_BUSY_AVG | |
4391 | GEN6_RP_UP_BUSY_AVG | |
4365 | GEN6_RP_DOWN_IDLE_AVG); |
4392 | GEN6_RP_DOWN_IDLE_AVG); |
4366 | break; |
- | |
4367 | } |
- | |
4368 | 4393 | ||
- | 4394 | dev_priv->rps.power = new_power; |
|
- | 4395 | dev_priv->rps.up_threshold = threshold_up; |
|
4369 | dev_priv->rps.power = new_power; |
4396 | dev_priv->rps.down_threshold = threshold_down; |
4370 | dev_priv->rps.last_adj = 0; |
4397 | dev_priv->rps.last_adj = 0; |
4371 | } |
4398 | } |
4372 | 4399 | ||
4373 | static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val) |
4400 | static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val) |
4374 | { |
4401 | { |
4375 | u32 mask = 0; |
4402 | u32 mask = 0; |
4376 | 4403 | ||
4377 | if (val > dev_priv->rps.min_freq_softlimit) |
4404 | if (val > dev_priv->rps.min_freq_softlimit) |
4378 | mask |= GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT; |
4405 | mask |= GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT; |
4379 | if (val < dev_priv->rps.max_freq_softlimit) |
4406 | if (val < dev_priv->rps.max_freq_softlimit) |
4380 | mask |= GEN6_PM_RP_UP_THRESHOLD; |
4407 | mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD; |
4381 | - | ||
4382 | mask |= dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED); |
4408 | |
4383 | mask &= dev_priv->pm_rps_events; |
- | |
4384 | 4409 | mask &= dev_priv->pm_rps_events; |
|
4385 | /* IVB and SNB hard hangs on looping batchbuffer |
- | |
4386 | * if GEN6_PM_UP_EI_EXPIRED is masked. |
- | |
4387 | */ |
- | |
4388 | if (INTEL_INFO(dev_priv->dev)->gen <= 7 && !IS_HASWELL(dev_priv->dev)) |
- | |
4389 | mask |= GEN6_PM_RP_UP_EI_EXPIRED; |
- | |
4390 | - | ||
4391 | if (IS_GEN8(dev_priv->dev)) |
- | |
4392 | mask |= GEN8_PMINTR_REDIRECT_TO_NON_DISP; |
- | |
4393 | 4410 | ||
4394 | return ~mask; |
4411 | return gen6_sanitize_rps_pm_mask(dev_priv, ~mask); |
4395 | } |
4412 | } |
4396 | 4413 | ||
4397 | /* gen6_set_rps is called to update the frequency request, but should also be |
4414 | /* gen6_set_rps is called to update the frequency request, but should also be |
4398 | * called when the range (min_delay and max_delay) is modified so that we can |
4415 | * called when the range (min_delay and max_delay) is modified so that we can |
4399 | * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */ |
4416 | * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */ |
4400 | void gen6_set_rps(struct drm_device *dev, u8 val) |
4417 | static void gen6_set_rps(struct drm_device *dev, u8 val) |
4401 | { |
4418 | { |
4402 | struct drm_i915_private *dev_priv = dev->dev_private; |
4419 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | 4420 | ||
- | 4421 | /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */ |
|
- | 4422 | if (IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) |
|
- | 4423 | return; |
|
4403 | 4424 | ||
4404 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
4425 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
4405 | WARN_ON(val > dev_priv->rps.max_freq_softlimit); |
4426 | WARN_ON(val > dev_priv->rps.max_freq); |
4406 | WARN_ON(val < dev_priv->rps.min_freq_softlimit); |
4427 | WARN_ON(val < dev_priv->rps.min_freq); |
4407 | 4428 | ||
4408 | /* min/max delay may still have been modified so be sure to |
4429 | /* min/max delay may still have been modified so be sure to |
4409 | * write the limits value. |
4430 | * write the limits value. |
4410 | */ |
4431 | */ |
4411 | if (val != dev_priv->rps.cur_freq) { |
4432 | if (val != dev_priv->rps.cur_freq) { |
4412 | gen6_set_rps_thresholds(dev_priv, val); |
4433 | gen6_set_rps_thresholds(dev_priv, val); |
- | 4434 | ||
- | 4435 | if (IS_GEN9(dev)) |
|
- | 4436 | I915_WRITE(GEN6_RPNSWREQ, |
|
4413 | 4437 | GEN9_FREQUENCY(val)); |
|
4414 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
4438 | else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
4415 | I915_WRITE(GEN6_RPNSWREQ, |
4439 | I915_WRITE(GEN6_RPNSWREQ, |
4416 | HSW_FREQUENCY(val)); |
4440 | HSW_FREQUENCY(val)); |
4417 | else |
4441 | else |
4418 | I915_WRITE(GEN6_RPNSWREQ, |
4442 | I915_WRITE(GEN6_RPNSWREQ, |
4419 | GEN6_FREQUENCY(val) | |
4443 | GEN6_FREQUENCY(val) | |
4420 | GEN6_OFFSET(0) | |
4444 | GEN6_OFFSET(0) | |
4421 | GEN6_AGGRESSIVE_TURBO); |
4445 | GEN6_AGGRESSIVE_TURBO); |
4422 | } |
4446 | } |
4423 | 4447 | ||
4424 | /* Make sure we continue to get interrupts |
4448 | /* Make sure we continue to get interrupts |
4425 | * until we hit the minimum or maximum frequencies. |
4449 | * until we hit the minimum or maximum frequencies. |
4426 | */ |
4450 | */ |
4427 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, gen6_rps_limits(dev_priv, val)); |
4451 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val)); |
4428 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
4452 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
4429 | 4453 | ||
4430 | POSTING_READ(GEN6_RPNSWREQ); |
4454 | POSTING_READ(GEN6_RPNSWREQ); |
4431 | 4455 | ||
4432 | dev_priv->rps.cur_freq = val; |
4456 | dev_priv->rps.cur_freq = val; |
4433 | trace_intel_gpu_freq_change(val * 50); |
4457 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); |
4434 | } |
4458 | } |
4435 | - | ||
4436 | /* vlv_set_rps_idle: Set the frequency to Rpn if Gfx clocks are down |
- | |
4437 | * |
- | |
4438 | * * If Gfx is Idle, then |
- | |
4439 | * 1. Mask Turbo interrupts |
- | |
4440 | * 2. Bring up Gfx clock |
- | |
4441 | * 3. Change the freq to Rpn and wait till P-Unit updates freq |
- | |
4442 | * 4. Clear the Force GFX CLK ON bit so that Gfx can down |
- | |
4443 | * 5. Unmask Turbo interrupts |
- | |
4444 | */ |
4459 | |
4445 | static void vlv_set_rps_idle(struct drm_i915_private *dev_priv) |
4460 | static void valleyview_set_rps(struct drm_device *dev, u8 val) |
4446 | { |
4461 | { |
4447 | struct drm_device *dev = dev_priv->dev; |
- | |
4448 | - | ||
4449 | /* Latest VLV doesn't need to force the gfx clock */ |
- | |
4450 | if (dev->pdev->revision >= 0xd) { |
- | |
4451 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); |
- | |
4452 | return; |
- | |
4453 | } |
- | |
4454 | - | ||
4455 | /* |
- | |
4456 | * When we are idle. Drop to min voltage state. |
- | |
- | 4462 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
4457 | */ |
4463 | |
4458 | 4464 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
|
4459 | if (dev_priv->rps.cur_freq <= dev_priv->rps.min_freq_softlimit) |
4465 | WARN_ON(val > dev_priv->rps.max_freq); |
4460 | return; |
4466 | WARN_ON(val < dev_priv->rps.min_freq); |
- | 4467 | ||
4461 | 4468 | if (WARN_ONCE(IS_CHERRYVIEW(dev) && (val & 1), |
|
4462 | /* Mask turbo interrupt so that they will not come in between */ |
4469 | "Odd GPU freq value\n")) |
- | 4470 | val &= ~1; |
|
- | 4471 | ||
- | 4472 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
|
- | 4473 | ||
4463 | I915_WRITE(GEN6_PMINTRMSK, 0xffffffff); |
4474 | if (val != dev_priv->rps.cur_freq) { |
4464 | 4475 | vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val); |
|
- | 4476 | if (!IS_CHERRYVIEW(dev_priv)) |
|
4465 | vlv_force_gfx_clock(dev_priv, true); |
4477 | gen6_set_rps_thresholds(dev_priv, val); |
- | 4478 | } |
|
- | 4479 | ||
4466 | 4480 | dev_priv->rps.cur_freq = val; |
|
- | 4481 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); |
|
4467 | dev_priv->rps.cur_freq = dev_priv->rps.min_freq_softlimit; |
4482 | } |
- | 4483 | ||
- | 4484 | /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down |
|
- | 4485 | * |
|
- | 4486 | * * If Gfx is Idle, then |
|
4468 | 4487 | * 1. Forcewake Media well. |
|
- | 4488 | * 2. Request idle freq. |
|
- | 4489 | * 3. Release Forcewake of Media well. |
|
- | 4490 | */ |
|
- | 4491 | static void vlv_set_rps_idle(struct drm_i915_private *dev_priv) |
|
4469 | vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, |
4492 | { |
4470 | dev_priv->rps.min_freq_softlimit); |
4493 | u32 val = dev_priv->rps.idle_freq; |
4471 | 4494 | ||
4472 | if (wait_for(((vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS)) |
4495 | if (dev_priv->rps.cur_freq <= val) |
4473 | & GENFREQSTATUS) == 0, 100)) |
4496 | return; |
4474 | DRM_ERROR("timed out waiting for Punit\n"); |
- | |
4475 | - | ||
4476 | vlv_force_gfx_clock(dev_priv, false); |
4497 | |
4477 | 4498 | /* Wake up the media well, as that takes a lot less |
|
4478 | I915_WRITE(GEN6_PMINTRMSK, |
- | |
4479 | gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq)); |
4499 | * power than the Render well. */ |
4480 | } |
4500 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA); |
4481 | 4501 | valleyview_set_rps(dev_priv->dev, val); |
|
4482 | void gen6_rps_idle(struct drm_i915_private *dev_priv) |
- | |
4483 | { |
4502 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA); |
4484 | struct drm_device *dev = dev_priv->dev; |
- | |
4485 | 4503 | } |
|
4486 | mutex_lock(&dev_priv->rps.hw_lock); |
4504 | |
4487 | if (dev_priv->rps.enabled) { |
4505 | void gen6_rps_busy(struct drm_i915_private *dev_priv) |
4488 | if (IS_CHERRYVIEW(dev)) |
4506 | { |
4489 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); |
4507 | mutex_lock(&dev_priv->rps.hw_lock); |
4490 | else if (IS_VALLEYVIEW(dev)) |
4508 | if (dev_priv->rps.enabled) { |
4491 | vlv_set_rps_idle(dev_priv); |
4509 | if (dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED)) |
4492 | else |
4510 | gen6_rps_reset_ei(dev_priv); |
4493 | gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); |
4511 | I915_WRITE(GEN6_PMINTRMSK, |
4494 | dev_priv->rps.last_adj = 0; |
4512 | gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq)); |
4495 | } |
4513 | } |
4496 | mutex_unlock(&dev_priv->rps.hw_lock); |
4514 | mutex_unlock(&dev_priv->rps.hw_lock); |
4497 | } |
4515 | } |
4498 | 4516 | ||
4499 | void gen6_rps_boost(struct drm_i915_private *dev_priv) |
4517 | void gen6_rps_idle(struct drm_i915_private *dev_priv) |
4500 | { |
4518 | { |
4501 | struct drm_device *dev = dev_priv->dev; |
4519 | struct drm_device *dev = dev_priv->dev; |
4502 | 4520 | ||
4503 | mutex_lock(&dev_priv->rps.hw_lock); |
4521 | mutex_lock(&dev_priv->rps.hw_lock); |
4504 | if (dev_priv->rps.enabled) { |
4522 | if (dev_priv->rps.enabled) { |
4505 | if (IS_VALLEYVIEW(dev)) |
4523 | if (IS_VALLEYVIEW(dev)) |
4506 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.max_freq_softlimit); |
4524 | vlv_set_rps_idle(dev_priv); |
4507 | else |
4525 | else |
4508 | gen6_set_rps(dev_priv->dev, dev_priv->rps.max_freq_softlimit); |
4526 | gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq); |
4509 | dev_priv->rps.last_adj = 0; |
4527 | dev_priv->rps.last_adj = 0; |
- | 4528 | I915_WRITE(GEN6_PMINTRMSK, 0xffffffff); |
|
4510 | } |
4529 | } |
4511 | mutex_unlock(&dev_priv->rps.hw_lock); |
4530 | mutex_unlock(&dev_priv->rps.hw_lock); |
- | 4531 | ||
- | 4532 | spin_lock(&dev_priv->rps.client_lock); |
|
- | 4533 | while (!list_empty(&dev_priv->rps.clients)) |
|
- | 4534 | list_del_init(dev_priv->rps.clients.next); |
|
- | 4535 | spin_unlock(&dev_priv->rps.client_lock); |
|
4512 | } |
4536 | } |
4513 | 4537 | ||
- | 4538 | void gen6_rps_boost(struct drm_i915_private *dev_priv, |
|
- | 4539 | struct intel_rps_client *rps, |
|
4514 | void valleyview_set_rps(struct drm_device *dev, u8 val) |
4540 | unsigned long submitted) |
4515 | { |
4541 | { |
- | 4542 | /* This is intentionally racy! We peek at the state here, then |
|
4516 | struct drm_i915_private *dev_priv = dev->dev_private; |
4543 | * validate inside the RPS worker. |
4517 | 4544 | */ |
|
4518 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
4545 | if (!(dev_priv->mm.busy && |
4519 | WARN_ON(val > dev_priv->rps.max_freq_softlimit); |
4546 | dev_priv->rps.enabled && |
- | 4547 | dev_priv->rps.cur_freq < dev_priv->rps.max_freq_softlimit)) |
|
4520 | WARN_ON(val < dev_priv->rps.min_freq_softlimit); |
4548 | return; |
4521 | 4549 | ||
- | 4550 | /* Force a RPS boost (and don't count it against the client) if |
|
- | 4551 | * the GPU is severely congested. |
|
4522 | if (WARN_ONCE(IS_CHERRYVIEW(dev) && (val & 1), |
4552 | */ |
4523 | "Odd GPU freq value\n")) |
4553 | if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES)) |
- | 4554 | rps = NULL; |
|
- | 4555 | ||
- | 4556 | spin_lock(&dev_priv->rps.client_lock); |
|
- | 4557 | if (rps == NULL || list_empty(&rps->link)) { |
|
4524 | val &= ~1; |
4558 | spin_lock_irq(&dev_priv->irq_lock); |
- | 4559 | if (dev_priv->rps.interrupts_enabled) { |
|
- | 4560 | dev_priv->rps.client_boost = true; |
|
- | 4561 | queue_work(dev_priv->wq, &dev_priv->rps.work); |
|
4525 | 4562 | } |
|
- | 4563 | spin_unlock_irq(&dev_priv->irq_lock); |
|
- | 4564 | ||
- | 4565 | if (rps != NULL) { |
|
- | 4566 | list_add(&rps->link, &dev_priv->rps.clients); |
|
- | 4567 | rps->boosts++; |
|
- | 4568 | } else |
|
- | 4569 | dev_priv->rps.boosts++; |
|
- | 4570 | } |
|
- | 4571 | spin_unlock(&dev_priv->rps.client_lock); |
|
4526 | if (val != dev_priv->rps.cur_freq) |
4572 | } |
- | 4573 | ||
4527 | vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val); |
4574 | void intel_set_rps(struct drm_device *dev, u8 val) |
4528 | 4575 | { |
|
4529 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
4576 | if (IS_VALLEYVIEW(dev)) |
4530 | 4577 | valleyview_set_rps(dev, val); |
|
4531 | dev_priv->rps.cur_freq = val; |
4578 | else |
4532 | trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv, val)); |
4579 | gen6_set_rps(dev, val); |
4533 | } |
4580 | } |
4534 | 4581 | ||
4535 | static void gen9_disable_rps(struct drm_device *dev) |
4582 | static void gen9_disable_rps(struct drm_device *dev) |
4536 | { |
4583 | { |
4537 | struct drm_i915_private *dev_priv = dev->dev_private; |
4584 | struct drm_i915_private *dev_priv = dev->dev_private; |
4538 | 4585 | ||
4539 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4586 | I915_WRITE(GEN6_RC_CONTROL, 0); |
- | 4587 | I915_WRITE(GEN9_PG_ENABLE, 0); |
|
4540 | } |
4588 | } |
4541 | 4589 | ||
4542 | static void gen6_disable_rps(struct drm_device *dev) |
4590 | static void gen6_disable_rps(struct drm_device *dev) |
4543 | { |
4591 | { |
4544 | struct drm_i915_private *dev_priv = dev->dev_private; |
4592 | struct drm_i915_private *dev_priv = dev->dev_private; |
4545 | 4593 | ||
4546 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4594 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4547 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
4595 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
4548 | } |
4596 | } |
4549 | 4597 | ||
4550 | static void cherryview_disable_rps(struct drm_device *dev) |
4598 | static void cherryview_disable_rps(struct drm_device *dev) |
4551 | { |
4599 | { |
4552 | struct drm_i915_private *dev_priv = dev->dev_private; |
4600 | struct drm_i915_private *dev_priv = dev->dev_private; |
4553 | 4601 | ||
4554 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4602 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4555 | } |
4603 | } |
4556 | 4604 | ||
4557 | static void valleyview_disable_rps(struct drm_device *dev) |
4605 | static void valleyview_disable_rps(struct drm_device *dev) |
4558 | { |
4606 | { |
4559 | struct drm_i915_private *dev_priv = dev->dev_private; |
4607 | struct drm_i915_private *dev_priv = dev->dev_private; |
4560 | 4608 | ||
4561 | /* we're doing forcewake before Disabling RC6, |
4609 | /* we're doing forcewake before Disabling RC6, |
4562 | * This what the BIOS expects when going into suspend */ |
4610 | * This what the BIOS expects when going into suspend */ |
4563 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL); |
4611 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
4564 | 4612 | ||
4565 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4613 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4566 | 4614 | ||
4567 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL); |
4615 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
4568 | } |
4616 | } |
4569 | 4617 | ||
4570 | static void intel_print_rc6_info(struct drm_device *dev, u32 mode) |
4618 | static void intel_print_rc6_info(struct drm_device *dev, u32 mode) |
4571 | { |
4619 | { |
4572 | if (IS_VALLEYVIEW(dev)) { |
4620 | if (IS_VALLEYVIEW(dev)) { |
4573 | if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1))) |
4621 | if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1))) |
4574 | mode = GEN6_RC_CTL_RC6_ENABLE; |
4622 | mode = GEN6_RC_CTL_RC6_ENABLE; |
4575 | else |
4623 | else |
4576 | mode = 0; |
4624 | mode = 0; |
4577 | } |
4625 | } |
4578 | if (HAS_RC6p(dev)) |
4626 | if (HAS_RC6p(dev)) |
4579 | DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n", |
4627 | DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n", |
4580 | (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off", |
4628 | (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off", |
4581 | (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off", |
4629 | (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off", |
4582 | (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off"); |
4630 | (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off"); |
4583 | 4631 | ||
4584 | else |
4632 | else |
4585 | DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n", |
4633 | DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n", |
4586 | (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off"); |
4634 | (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off"); |
4587 | } |
4635 | } |
4588 | 4636 | ||
4589 | static int sanitize_rc6_option(const struct drm_device *dev, int enable_rc6) |
4637 | static int sanitize_rc6_option(const struct drm_device *dev, int enable_rc6) |
4590 | { |
4638 | { |
4591 | /* No RC6 before Ironlake */ |
4639 | /* No RC6 before Ironlake and code is gone for ilk. */ |
4592 | if (INTEL_INFO(dev)->gen < 5) |
4640 | if (INTEL_INFO(dev)->gen < 6) |
4593 | return 0; |
- | |
4594 | - | ||
4595 | /* RC6 is only on Ironlake mobile not on desktop */ |
- | |
4596 | if (INTEL_INFO(dev)->gen == 5 && !IS_IRONLAKE_M(dev)) |
- | |
4597 | return 0; |
4641 | return 0; |
4598 | 4642 | ||
4599 | /* Respect the kernel parameter if it is set */ |
4643 | /* Respect the kernel parameter if it is set */ |
4600 | if (enable_rc6 >= 0) { |
4644 | if (enable_rc6 >= 0) { |
4601 | int mask; |
4645 | int mask; |
4602 | 4646 | ||
4603 | if (HAS_RC6p(dev)) |
4647 | if (HAS_RC6p(dev)) |
4604 | mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE | |
4648 | mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE | |
4605 | INTEL_RC6pp_ENABLE; |
4649 | INTEL_RC6pp_ENABLE; |
4606 | else |
4650 | else |
4607 | mask = INTEL_RC6_ENABLE; |
4651 | mask = INTEL_RC6_ENABLE; |
4608 | 4652 | ||
4609 | if ((enable_rc6 & mask) != enable_rc6) |
4653 | if ((enable_rc6 & mask) != enable_rc6) |
4610 | DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n", |
4654 | DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n", |
4611 | enable_rc6 & mask, enable_rc6, mask); |
4655 | enable_rc6 & mask, enable_rc6, mask); |
4612 | 4656 | ||
4613 | return enable_rc6 & mask; |
4657 | return enable_rc6 & mask; |
4614 | } |
4658 | } |
4615 | - | ||
4616 | /* Disable RC6 on Ironlake */ |
- | |
4617 | if (INTEL_INFO(dev)->gen == 5) |
- | |
4618 | return 0; |
- | |
4619 | 4659 | ||
4620 | if (IS_IVYBRIDGE(dev)) |
4660 | if (IS_IVYBRIDGE(dev)) |
4621 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); |
4661 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); |
4622 | 4662 | ||
4623 | return INTEL_RC6_ENABLE; |
4663 | return INTEL_RC6_ENABLE; |
4624 | } |
4664 | } |
4625 | 4665 | ||
4626 | int intel_enable_rc6(const struct drm_device *dev) |
4666 | int intel_enable_rc6(const struct drm_device *dev) |
4627 | { |
4667 | { |
4628 | return i915.enable_rc6; |
4668 | return i915.enable_rc6; |
4629 | } |
4669 | } |
4630 | 4670 | ||
4631 | static void gen6_init_rps_frequencies(struct drm_device *dev) |
4671 | static void gen6_init_rps_frequencies(struct drm_device *dev) |
4632 | { |
4672 | { |
4633 | struct drm_i915_private *dev_priv = dev->dev_private; |
4673 | struct drm_i915_private *dev_priv = dev->dev_private; |
4634 | uint32_t rp_state_cap; |
4674 | uint32_t rp_state_cap; |
4635 | u32 ddcc_status = 0; |
4675 | u32 ddcc_status = 0; |
4636 | int ret; |
4676 | int ret; |
4637 | - | ||
4638 | rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); |
4677 | |
4639 | /* All of these values are in units of 50MHz */ |
4678 | /* All of these values are in units of 50MHz */ |
4640 | dev_priv->rps.cur_freq = 0; |
4679 | dev_priv->rps.cur_freq = 0; |
4641 | /* static values from HW: RP0 > RP1 > RPn (min_freq) */ |
4680 | /* static values from HW: RP0 > RP1 > RPn (min_freq) */ |
- | 4681 | if (IS_BROXTON(dev)) { |
|
- | 4682 | rp_state_cap = I915_READ(BXT_RP_STATE_CAP); |
|
- | 4683 | dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff; |
|
- | 4684 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; |
|
- | 4685 | dev_priv->rps.min_freq = (rp_state_cap >> 0) & 0xff; |
|
- | 4686 | } else { |
|
- | 4687 | rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); |
|
4642 | dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff; |
4688 | dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff; |
4643 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; |
4689 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; |
4644 | dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff; |
4690 | dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff; |
- | 4691 | } |
|
- | 4692 | ||
4645 | /* hw_max = RP0 until we check for overclocking */ |
4693 | /* hw_max = RP0 until we check for overclocking */ |
4646 | dev_priv->rps.max_freq = dev_priv->rps.rp0_freq; |
4694 | dev_priv->rps.max_freq = dev_priv->rps.rp0_freq; |
4647 | 4695 | ||
4648 | dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq; |
4696 | dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq; |
4649 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
4697 | if (IS_HASWELL(dev) || IS_BROADWELL(dev) || IS_SKYLAKE(dev)) { |
4650 | ret = sandybridge_pcode_read(dev_priv, |
4698 | ret = sandybridge_pcode_read(dev_priv, |
4651 | HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL, |
4699 | HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL, |
4652 | &ddcc_status); |
4700 | &ddcc_status); |
4653 | if (0 == ret) |
4701 | if (0 == ret) |
4654 | dev_priv->rps.efficient_freq = |
4702 | dev_priv->rps.efficient_freq = |
- | 4703 | clamp_t(u8, |
|
4655 | (ddcc_status >> 8) & 0xff; |
4704 | ((ddcc_status >> 8) & 0xff), |
- | 4705 | dev_priv->rps.min_freq, |
|
- | 4706 | dev_priv->rps.max_freq); |
|
- | 4707 | } |
|
- | 4708 | ||
- | 4709 | if (IS_SKYLAKE(dev)) { |
|
- | 4710 | /* Store the frequency values in 16.66 MHZ units, which is |
|
- | 4711 | the natural hardware unit for SKL */ |
|
- | 4712 | dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER; |
|
- | 4713 | dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER; |
|
- | 4714 | dev_priv->rps.min_freq *= GEN9_FREQ_SCALER; |
|
- | 4715 | dev_priv->rps.max_freq *= GEN9_FREQ_SCALER; |
|
- | 4716 | dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER; |
|
4656 | } |
4717 | } |
- | 4718 | ||
- | 4719 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; |
|
4657 | 4720 | ||
4658 | /* Preserve min/max settings in case of re-init */ |
4721 | /* Preserve min/max settings in case of re-init */ |
4659 | if (dev_priv->rps.max_freq_softlimit == 0) |
4722 | if (dev_priv->rps.max_freq_softlimit == 0) |
4660 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; |
4723 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; |
4661 | 4724 | ||
4662 | if (dev_priv->rps.min_freq_softlimit == 0) { |
4725 | if (dev_priv->rps.min_freq_softlimit == 0) { |
4663 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
4726 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
4664 | dev_priv->rps.min_freq_softlimit = |
4727 | dev_priv->rps.min_freq_softlimit = |
4665 | /* max(RPe, 450 MHz) */ |
4728 | max_t(int, dev_priv->rps.efficient_freq, |
4666 | max(dev_priv->rps.efficient_freq, (u8) 9); |
4729 | intel_freq_opcode(dev_priv, 450)); |
4667 | else |
4730 | else |
4668 | dev_priv->rps.min_freq_softlimit = |
4731 | dev_priv->rps.min_freq_softlimit = |
4669 | dev_priv->rps.min_freq; |
4732 | dev_priv->rps.min_freq; |
4670 | } |
4733 | } |
4671 | } |
4734 | } |
- | 4735 | ||
4672 | 4736 | /* See the Gen9_GT_PM_Programming_Guide doc for the below */ |
|
4673 | static void gen9_enable_rps(struct drm_device *dev) |
4737 | static void gen9_enable_rps(struct drm_device *dev) |
4674 | { |
4738 | { |
4675 | struct drm_i915_private *dev_priv = dev->dev_private; |
4739 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | 4740 | ||
- | 4741 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
|
- | 4742 | ||
- | 4743 | gen6_init_rps_frequencies(dev); |
|
- | 4744 | ||
- | 4745 | /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */ |
|
- | 4746 | if (IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) { |
|
- | 4747 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
|
- | 4748 | return; |
|
- | 4749 | } |
|
- | 4750 | ||
- | 4751 | /* Program defaults and thresholds for RPS*/ |
|
- | 4752 | I915_WRITE(GEN6_RC_VIDEO_FREQ, |
|
- | 4753 | GEN9_FREQUENCY(dev_priv->rps.rp1_freq)); |
|
- | 4754 | ||
- | 4755 | /* 1 second timeout*/ |
|
- | 4756 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, |
|
- | 4757 | GT_INTERVAL_FROM_US(dev_priv, 1000000)); |
|
- | 4758 | ||
- | 4759 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa); |
|
- | 4760 | ||
- | 4761 | /* Leaning on the below call to gen6_set_rps to program/setup the |
|
- | 4762 | * Up/Down EI & threshold registers, as well as the RP_CONTROL, |
|
- | 4763 | * RP_INTERRUPT_LIMITS & RPNSWREQ registers */ |
|
- | 4764 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
|
- | 4765 | gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); |
|
- | 4766 | ||
- | 4767 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
|
- | 4768 | } |
|
- | 4769 | ||
- | 4770 | static void gen9_enable_rc6(struct drm_device *dev) |
|
- | 4771 | { |
|
- | 4772 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
4676 | struct intel_engine_cs *ring; |
4773 | struct intel_engine_cs *ring; |
4677 | uint32_t rc6_mask = 0; |
4774 | uint32_t rc6_mask = 0; |
4678 | int unused; |
4775 | int unused; |
4679 | 4776 | ||
4680 | /* 1a: Software RC state - RC0 */ |
4777 | /* 1a: Software RC state - RC0 */ |
4681 | I915_WRITE(GEN6_RC_STATE, 0); |
4778 | I915_WRITE(GEN6_RC_STATE, 0); |
4682 | 4779 | ||
4683 | /* 1b: Get forcewake during program sequence. Although the driver |
4780 | /* 1b: Get forcewake during program sequence. Although the driver |
4684 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ |
4781 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ |
4685 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL); |
4782 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
4686 | 4783 | ||
4687 | /* 2a: Disable RC states. */ |
4784 | /* 2a: Disable RC states. */ |
4688 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4785 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4689 | 4786 | ||
4690 | /* 2b: Program RC6 thresholds.*/ |
4787 | /* 2b: Program RC6 thresholds.*/ |
- | 4788 | ||
- | 4789 | /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */ |
|
- | 4790 | if (IS_SKYLAKE(dev)) |
|
- | 4791 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16); |
|
- | 4792 | else |
|
4691 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16); |
4793 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16); |
4692 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
4794 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
4693 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ |
4795 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ |
4694 | for_each_ring(ring, dev_priv, unused) |
4796 | for_each_ring(ring, dev_priv, unused) |
4695 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
4797 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
- | 4798 | ||
- | 4799 | if (HAS_GUC_UCODE(dev)) |
|
- | 4800 | I915_WRITE(GUC_MAX_IDLE_COUNT, 0xA); |
|
- | 4801 | ||
4696 | I915_WRITE(GEN6_RC_SLEEP, 0); |
4802 | I915_WRITE(GEN6_RC_SLEEP, 0); |
- | 4803 | ||
- | 4804 | /* 2c: Program Coarse Power Gating Policies. */ |
|
- | 4805 | I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25); |
|
4697 | I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */ |
4806 | I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25); |
4698 | 4807 | ||
4699 | /* 3a: Enable RC6 */ |
4808 | /* 3a: Enable RC6 */ |
4700 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
4809 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
4701 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
4810 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
4702 | DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? |
4811 | DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? |
4703 | "on" : "off"); |
4812 | "on" : "off"); |
- | 4813 | /* WaRsUseTimeoutMode */ |
|
- | 4814 | if ((IS_SKYLAKE(dev) && INTEL_REVID(dev) <= SKL_REVID_D0) || |
|
- | 4815 | (IS_BROXTON(dev) && INTEL_REVID(dev) <= BXT_REVID_A0)) { |
|
- | 4816 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us */ |
|
- | 4817 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
|
- | 4818 | GEN7_RC_CTL_TO_MODE | |
|
- | 4819 | rc6_mask); |
|
- | 4820 | } else { |
|
- | 4821 | I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */ |
|
4704 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
4822 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
4705 | GEN6_RC_CTL_EI_MODE(1) | |
4823 | GEN6_RC_CTL_EI_MODE(1) | |
4706 | rc6_mask); |
4824 | rc6_mask); |
- | 4825 | } |
|
- | 4826 | ||
- | 4827 | /* |
|
- | 4828 | * 3b: Enable Coarse Power Gating only when RC6 is enabled. |
|
- | 4829 | * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6. |
|
- | 4830 | */ |
|
- | 4831 | if ((IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) || |
|
- | 4832 | ((IS_SKL_GT3(dev) || IS_SKL_GT4(dev)) && (INTEL_REVID(dev) <= SKL_REVID_F0))) |
|
- | 4833 | I915_WRITE(GEN9_PG_ENABLE, 0); |
|
- | 4834 | else |
|
- | 4835 | I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? |
|
- | 4836 | (GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE) : 0); |
|
4707 | 4837 | ||
4708 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL); |
4838 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
4709 | 4839 | ||
4710 | } |
4840 | } |
4711 | 4841 | ||
4712 | static void gen8_enable_rps(struct drm_device *dev) |
4842 | static void gen8_enable_rps(struct drm_device *dev) |
4713 | { |
4843 | { |
4714 | struct drm_i915_private *dev_priv = dev->dev_private; |
4844 | struct drm_i915_private *dev_priv = dev->dev_private; |
4715 | struct intel_engine_cs *ring; |
4845 | struct intel_engine_cs *ring; |
4716 | uint32_t rc6_mask = 0; |
4846 | uint32_t rc6_mask = 0; |
4717 | int unused; |
4847 | int unused; |
4718 | 4848 | ||
4719 | /* 1a: Software RC state - RC0 */ |
4849 | /* 1a: Software RC state - RC0 */ |
4720 | I915_WRITE(GEN6_RC_STATE, 0); |
4850 | I915_WRITE(GEN6_RC_STATE, 0); |
4721 | 4851 | ||
4722 | /* 1c & 1d: Get forcewake during program sequence. Although the driver |
4852 | /* 1c & 1d: Get forcewake during program sequence. Although the driver |
4723 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ |
4853 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ |
4724 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL); |
4854 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
4725 | 4855 | ||
4726 | /* 2a: Disable RC states. */ |
4856 | /* 2a: Disable RC states. */ |
4727 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4857 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4728 | 4858 | ||
4729 | /* Initialize rps frequencies */ |
4859 | /* Initialize rps frequencies */ |
4730 | gen6_init_rps_frequencies(dev); |
4860 | gen6_init_rps_frequencies(dev); |
4731 | 4861 | ||
4732 | /* 2b: Program RC6 thresholds.*/ |
4862 | /* 2b: Program RC6 thresholds.*/ |
4733 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); |
4863 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); |
4734 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
4864 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
4735 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ |
4865 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ |
4736 | for_each_ring(ring, dev_priv, unused) |
4866 | for_each_ring(ring, dev_priv, unused) |
4737 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
4867 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
4738 | I915_WRITE(GEN6_RC_SLEEP, 0); |
4868 | I915_WRITE(GEN6_RC_SLEEP, 0); |
4739 | if (IS_BROADWELL(dev)) |
4869 | if (IS_BROADWELL(dev)) |
4740 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */ |
4870 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */ |
4741 | else |
4871 | else |
4742 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */ |
4872 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */ |
4743 | 4873 | ||
4744 | /* 3: Enable RC6 */ |
4874 | /* 3: Enable RC6 */ |
4745 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
4875 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
4746 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
4876 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
4747 | intel_print_rc6_info(dev, rc6_mask); |
4877 | intel_print_rc6_info(dev, rc6_mask); |
4748 | if (IS_BROADWELL(dev)) |
4878 | if (IS_BROADWELL(dev)) |
4749 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
4879 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
4750 | GEN7_RC_CTL_TO_MODE | |
4880 | GEN7_RC_CTL_TO_MODE | |
4751 | rc6_mask); |
4881 | rc6_mask); |
4752 | else |
4882 | else |
4753 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
4883 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
4754 | GEN6_RC_CTL_EI_MODE(1) | |
4884 | GEN6_RC_CTL_EI_MODE(1) | |
4755 | rc6_mask); |
4885 | rc6_mask); |
4756 | 4886 | ||
4757 | /* 4 Program defaults and thresholds for RPS*/ |
4887 | /* 4 Program defaults and thresholds for RPS*/ |
4758 | I915_WRITE(GEN6_RPNSWREQ, |
4888 | I915_WRITE(GEN6_RPNSWREQ, |
4759 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); |
4889 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); |
4760 | I915_WRITE(GEN6_RC_VIDEO_FREQ, |
4890 | I915_WRITE(GEN6_RC_VIDEO_FREQ, |
4761 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); |
4891 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); |
4762 | /* NB: Docs say 1s, and 1000000 - which aren't equivalent */ |
4892 | /* NB: Docs say 1s, and 1000000 - which aren't equivalent */ |
4763 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */ |
4893 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */ |
4764 | 4894 | ||
4765 | /* Docs recommend 900MHz, and 300 MHz respectively */ |
4895 | /* Docs recommend 900MHz, and 300 MHz respectively */ |
4766 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, |
4896 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, |
4767 | dev_priv->rps.max_freq_softlimit << 24 | |
4897 | dev_priv->rps.max_freq_softlimit << 24 | |
4768 | dev_priv->rps.min_freq_softlimit << 16); |
4898 | dev_priv->rps.min_freq_softlimit << 16); |
4769 | 4899 | ||
4770 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */ |
4900 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */ |
4771 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/ |
4901 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/ |
4772 | I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */ |
4902 | I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */ |
4773 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */ |
4903 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */ |
4774 | 4904 | ||
4775 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
4905 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
4776 | 4906 | ||
4777 | /* 5: Enable RPS */ |
4907 | /* 5: Enable RPS */ |
4778 | I915_WRITE(GEN6_RP_CONTROL, |
4908 | I915_WRITE(GEN6_RP_CONTROL, |
4779 | GEN6_RP_MEDIA_TURBO | |
4909 | GEN6_RP_MEDIA_TURBO | |
4780 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
4910 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
4781 | GEN6_RP_MEDIA_IS_GFX | |
4911 | GEN6_RP_MEDIA_IS_GFX | |
4782 | GEN6_RP_ENABLE | |
4912 | GEN6_RP_ENABLE | |
4783 | GEN6_RP_UP_BUSY_AVG | |
4913 | GEN6_RP_UP_BUSY_AVG | |
4784 | GEN6_RP_DOWN_IDLE_AVG); |
4914 | GEN6_RP_DOWN_IDLE_AVG); |
4785 | 4915 | ||
4786 | /* 6: Ring frequency + overclocking (our driver does this later */ |
4916 | /* 6: Ring frequency + overclocking (our driver does this later */ |
4787 | 4917 | ||
4788 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
4918 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
4789 | gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); |
4919 | gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq); |
4790 | 4920 | ||
4791 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL); |
4921 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
4792 | } |
4922 | } |
4793 | 4923 | ||
4794 | static void gen6_enable_rps(struct drm_device *dev) |
4924 | static void gen6_enable_rps(struct drm_device *dev) |
4795 | { |
4925 | { |
4796 | struct drm_i915_private *dev_priv = dev->dev_private; |
4926 | struct drm_i915_private *dev_priv = dev->dev_private; |
4797 | struct intel_engine_cs *ring; |
4927 | struct intel_engine_cs *ring; |
4798 | u32 rc6vids, pcu_mbox = 0, rc6_mask = 0; |
4928 | u32 rc6vids, pcu_mbox = 0, rc6_mask = 0; |
4799 | u32 gtfifodbg; |
4929 | u32 gtfifodbg; |
4800 | int rc6_mode; |
4930 | int rc6_mode; |
4801 | int i, ret; |
4931 | int i, ret; |
4802 | 4932 | ||
4803 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
4933 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
4804 | 4934 | ||
4805 | /* Here begins a magic sequence of register writes to enable |
4935 | /* Here begins a magic sequence of register writes to enable |
4806 | * auto-downclocking. |
4936 | * auto-downclocking. |
4807 | * |
4937 | * |
4808 | * Perhaps there might be some value in exposing these to |
4938 | * Perhaps there might be some value in exposing these to |
4809 | * userspace... |
4939 | * userspace... |
4810 | */ |
4940 | */ |
4811 | I915_WRITE(GEN6_RC_STATE, 0); |
4941 | I915_WRITE(GEN6_RC_STATE, 0); |
4812 | 4942 | ||
4813 | /* Clear the DBG now so we don't confuse earlier errors */ |
4943 | /* Clear the DBG now so we don't confuse earlier errors */ |
4814 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { |
4944 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { |
4815 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); |
4945 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); |
4816 | I915_WRITE(GTFIFODBG, gtfifodbg); |
4946 | I915_WRITE(GTFIFODBG, gtfifodbg); |
4817 | } |
4947 | } |
4818 | 4948 | ||
4819 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL); |
4949 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
4820 | 4950 | ||
4821 | /* Initialize rps frequencies */ |
4951 | /* Initialize rps frequencies */ |
4822 | gen6_init_rps_frequencies(dev); |
4952 | gen6_init_rps_frequencies(dev); |
4823 | 4953 | ||
4824 | /* disable the counters and set deterministic thresholds */ |
4954 | /* disable the counters and set deterministic thresholds */ |
4825 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4955 | I915_WRITE(GEN6_RC_CONTROL, 0); |
4826 | 4956 | ||
4827 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); |
4957 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); |
4828 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); |
4958 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); |
4829 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); |
4959 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); |
4830 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); |
4960 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); |
4831 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); |
4961 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); |
4832 | 4962 | ||
4833 | for_each_ring(ring, dev_priv, i) |
4963 | for_each_ring(ring, dev_priv, i) |
4834 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
4964 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
4835 | 4965 | ||
4836 | I915_WRITE(GEN6_RC_SLEEP, 0); |
4966 | I915_WRITE(GEN6_RC_SLEEP, 0); |
4837 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); |
4967 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); |
4838 | if (IS_IVYBRIDGE(dev)) |
4968 | if (IS_IVYBRIDGE(dev)) |
4839 | I915_WRITE(GEN6_RC6_THRESHOLD, 125000); |
4969 | I915_WRITE(GEN6_RC6_THRESHOLD, 125000); |
4840 | else |
4970 | else |
4841 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); |
4971 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); |
4842 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
4972 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
4843 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
4973 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
4844 | 4974 | ||
4845 | /* Check if we are enabling RC6 */ |
4975 | /* Check if we are enabling RC6 */ |
4846 | rc6_mode = intel_enable_rc6(dev_priv->dev); |
4976 | rc6_mode = intel_enable_rc6(dev_priv->dev); |
4847 | if (rc6_mode & INTEL_RC6_ENABLE) |
4977 | if (rc6_mode & INTEL_RC6_ENABLE) |
4848 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; |
4978 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; |
4849 | 4979 | ||
4850 | /* We don't use those on Haswell */ |
4980 | /* We don't use those on Haswell */ |
4851 | if (!IS_HASWELL(dev)) { |
4981 | if (!IS_HASWELL(dev)) { |
4852 | if (rc6_mode & INTEL_RC6p_ENABLE) |
4982 | if (rc6_mode & INTEL_RC6p_ENABLE) |
4853 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; |
4983 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; |
4854 | 4984 | ||
4855 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
4985 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
4856 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; |
4986 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; |
4857 | } |
4987 | } |
4858 | 4988 | ||
4859 | intel_print_rc6_info(dev, rc6_mask); |
4989 | intel_print_rc6_info(dev, rc6_mask); |
4860 | 4990 | ||
4861 | I915_WRITE(GEN6_RC_CONTROL, |
4991 | I915_WRITE(GEN6_RC_CONTROL, |
4862 | rc6_mask | |
4992 | rc6_mask | |
4863 | GEN6_RC_CTL_EI_MODE(1) | |
4993 | GEN6_RC_CTL_EI_MODE(1) | |
4864 | GEN6_RC_CTL_HW_ENABLE); |
4994 | GEN6_RC_CTL_HW_ENABLE); |
4865 | 4995 | ||
4866 | /* Power down if completely idle for over 50ms */ |
4996 | /* Power down if completely idle for over 50ms */ |
4867 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000); |
4997 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000); |
4868 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
4998 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
4869 | 4999 | ||
4870 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0); |
5000 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0); |
4871 | if (ret) |
5001 | if (ret) |
4872 | DRM_DEBUG_DRIVER("Failed to set the min frequency\n"); |
5002 | DRM_DEBUG_DRIVER("Failed to set the min frequency\n"); |
4873 | 5003 | ||
4874 | ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox); |
5004 | ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox); |
4875 | if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */ |
5005 | if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */ |
4876 | DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n", |
5006 | DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n", |
4877 | (dev_priv->rps.max_freq_softlimit & 0xff) * 50, |
5007 | (dev_priv->rps.max_freq_softlimit & 0xff) * 50, |
4878 | (pcu_mbox & 0xff) * 50); |
5008 | (pcu_mbox & 0xff) * 50); |
4879 | dev_priv->rps.max_freq = pcu_mbox & 0xff; |
5009 | dev_priv->rps.max_freq = pcu_mbox & 0xff; |
4880 | } |
5010 | } |
4881 | 5011 | ||
4882 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
5012 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
4883 | gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); |
5013 | gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq); |
4884 | 5014 | ||
4885 | rc6vids = 0; |
5015 | rc6vids = 0; |
4886 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); |
5016 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); |
4887 | if (IS_GEN6(dev) && ret) { |
5017 | if (IS_GEN6(dev) && ret) { |
4888 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); |
5018 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); |
4889 | } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { |
5019 | } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { |
4890 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", |
5020 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", |
4891 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); |
5021 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); |
4892 | rc6vids &= 0xffff00; |
5022 | rc6vids &= 0xffff00; |
4893 | rc6vids |= GEN6_ENCODE_RC6_VID(450); |
5023 | rc6vids |= GEN6_ENCODE_RC6_VID(450); |
4894 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); |
5024 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); |
4895 | if (ret) |
5025 | if (ret) |
4896 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); |
5026 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); |
4897 | } |
5027 | } |
4898 | 5028 | ||
4899 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL); |
5029 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
4900 | } |
5030 | } |
4901 | 5031 | ||
4902 | static void __gen6_update_ring_freq(struct drm_device *dev) |
5032 | static void __gen6_update_ring_freq(struct drm_device *dev) |
4903 | { |
5033 | { |
4904 | struct drm_i915_private *dev_priv = dev->dev_private; |
5034 | struct drm_i915_private *dev_priv = dev->dev_private; |
4905 | int min_freq = 15; |
5035 | int min_freq = 15; |
4906 | unsigned int gpu_freq; |
5036 | unsigned int gpu_freq; |
4907 | unsigned int max_ia_freq, min_ring_freq; |
5037 | unsigned int max_ia_freq, min_ring_freq; |
- | 5038 | unsigned int max_gpu_freq, min_gpu_freq; |
|
4908 | int scaling_factor = 180; |
5039 | int scaling_factor = 180; |
4909 | struct cpufreq_policy *policy; |
5040 | struct cpufreq_policy *policy; |
4910 | 5041 | ||
4911 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
5042 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
4912 | 5043 | ||
4913 | max_ia_freq = cpufreq_quick_get_max(0); |
5044 | max_ia_freq = cpufreq_quick_get_max(0); |
4914 | /* |
5045 | /* |
4915 | * Default to measured freq if none found, PCU will ensure we |
5046 | * Default to measured freq if none found, PCU will ensure we |
4916 | * don't go over |
5047 | * don't go over |
4917 | */ |
5048 | */ |
4918 | max_ia_freq = tsc_khz; |
5049 | max_ia_freq = tsc_khz; |
4919 | 5050 | ||
4920 | /* Convert from kHz to MHz */ |
5051 | /* Convert from kHz to MHz */ |
4921 | max_ia_freq /= 1000; |
5052 | max_ia_freq /= 1000; |
4922 | 5053 | ||
4923 | min_ring_freq = I915_READ(DCLK) & 0xf; |
5054 | min_ring_freq = I915_READ(DCLK) & 0xf; |
4924 | /* convert DDR frequency from units of 266.6MHz to bandwidth */ |
5055 | /* convert DDR frequency from units of 266.6MHz to bandwidth */ |
4925 | min_ring_freq = mult_frac(min_ring_freq, 8, 3); |
5056 | min_ring_freq = mult_frac(min_ring_freq, 8, 3); |
- | 5057 | ||
- | 5058 | if (IS_SKYLAKE(dev)) { |
|
- | 5059 | /* Convert GT frequency to 50 HZ units */ |
|
- | 5060 | min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER; |
|
- | 5061 | max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER; |
|
- | 5062 | } else { |
|
- | 5063 | min_gpu_freq = dev_priv->rps.min_freq; |
|
- | 5064 | max_gpu_freq = dev_priv->rps.max_freq; |
|
- | 5065 | } |
|
4926 | 5066 | ||
4927 | /* |
5067 | /* |
4928 | * For each potential GPU frequency, load a ring frequency we'd like |
5068 | * For each potential GPU frequency, load a ring frequency we'd like |
4929 | * to use for memory access. We do this by specifying the IA frequency |
5069 | * to use for memory access. We do this by specifying the IA frequency |
4930 | * the PCU should use as a reference to determine the ring frequency. |
5070 | * the PCU should use as a reference to determine the ring frequency. |
4931 | */ |
5071 | */ |
4932 | for (gpu_freq = dev_priv->rps.max_freq; gpu_freq >= dev_priv->rps.min_freq; |
5072 | for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) { |
4933 | gpu_freq--) { |
- | |
4934 | int diff = dev_priv->rps.max_freq - gpu_freq; |
5073 | int diff = max_gpu_freq - gpu_freq; |
4935 | unsigned int ia_freq = 0, ring_freq = 0; |
5074 | unsigned int ia_freq = 0, ring_freq = 0; |
- | 5075 | ||
- | 5076 | if (IS_SKYLAKE(dev)) { |
|
- | 5077 | /* |
|
- | 5078 | * ring_freq = 2 * GT. ring_freq is in 100MHz units |
|
- | 5079 | * No floor required for ring frequency on SKL. |
|
- | 5080 | */ |
|
4936 | 5081 | ring_freq = gpu_freq; |
|
4937 | if (INTEL_INFO(dev)->gen >= 8) { |
5082 | } else if (INTEL_INFO(dev)->gen >= 8) { |
4938 | /* max(2 * GT, DDR). NB: GT is 50MHz units */ |
5083 | /* max(2 * GT, DDR). NB: GT is 50MHz units */ |
4939 | ring_freq = max(min_ring_freq, gpu_freq); |
5084 | ring_freq = max(min_ring_freq, gpu_freq); |
4940 | } else if (IS_HASWELL(dev)) { |
5085 | } else if (IS_HASWELL(dev)) { |
4941 | ring_freq = mult_frac(gpu_freq, 5, 4); |
5086 | ring_freq = mult_frac(gpu_freq, 5, 4); |
4942 | ring_freq = max(min_ring_freq, ring_freq); |
5087 | ring_freq = max(min_ring_freq, ring_freq); |
4943 | /* leave ia_freq as the default, chosen by cpufreq */ |
5088 | /* leave ia_freq as the default, chosen by cpufreq */ |
4944 | } else { |
5089 | } else { |
4945 | /* On older processors, there is no separate ring |
5090 | /* On older processors, there is no separate ring |
4946 | * clock domain, so in order to boost the bandwidth |
5091 | * clock domain, so in order to boost the bandwidth |
4947 | * of the ring, we need to upclock the CPU (ia_freq). |
5092 | * of the ring, we need to upclock the CPU (ia_freq). |
4948 | * |
5093 | * |
4949 | * For GPU frequencies less than 750MHz, |
5094 | * For GPU frequencies less than 750MHz, |
4950 | * just use the lowest ring freq. |
5095 | * just use the lowest ring freq. |
4951 | */ |
5096 | */ |
4952 | if (gpu_freq < min_freq) |
5097 | if (gpu_freq < min_freq) |
4953 | ia_freq = 800; |
5098 | ia_freq = 800; |
4954 | else |
5099 | else |
4955 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); |
5100 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); |
4956 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); |
5101 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); |
4957 | } |
5102 | } |
4958 | 5103 | ||
4959 | sandybridge_pcode_write(dev_priv, |
5104 | sandybridge_pcode_write(dev_priv, |
4960 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, |
5105 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, |
4961 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
5106 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
4962 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | |
5107 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | |
4963 | gpu_freq); |
5108 | gpu_freq); |
4964 | } |
5109 | } |
4965 | } |
5110 | } |
4966 | 5111 | ||
4967 | void gen6_update_ring_freq(struct drm_device *dev) |
5112 | void gen6_update_ring_freq(struct drm_device *dev) |
4968 | { |
5113 | { |
4969 | struct drm_i915_private *dev_priv = dev->dev_private; |
5114 | struct drm_i915_private *dev_priv = dev->dev_private; |
4970 | 5115 | ||
4971 | if (INTEL_INFO(dev)->gen < 6 || IS_VALLEYVIEW(dev)) |
5116 | if (!HAS_CORE_RING_FREQ(dev)) |
4972 | return; |
5117 | return; |
4973 | 5118 | ||
4974 | mutex_lock(&dev_priv->rps.hw_lock); |
5119 | mutex_lock(&dev_priv->rps.hw_lock); |
4975 | __gen6_update_ring_freq(dev); |
5120 | __gen6_update_ring_freq(dev); |
4976 | mutex_unlock(&dev_priv->rps.hw_lock); |
5121 | mutex_unlock(&dev_priv->rps.hw_lock); |
4977 | } |
5122 | } |
4978 | 5123 | ||
4979 | static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv) |
5124 | static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv) |
4980 | { |
5125 | { |
- | 5126 | struct drm_device *dev = dev_priv->dev; |
|
4981 | u32 val, rp0; |
5127 | u32 val, rp0; |
4982 | 5128 | ||
- | 5129 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); |
|
- | 5130 | ||
- | 5131 | switch (INTEL_INFO(dev)->eu_total) { |
|
- | 5132 | case 8: |
|
- | 5133 | /* (2 * 4) config */ |
|
- | 5134 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT); |
|
- | 5135 | break; |
|
- | 5136 | case 12: |
|
- | 5137 | /* (2 * 6) config */ |
|
- | 5138 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT); |
|
- | 5139 | break; |
|
- | 5140 | case 16: |
|
- | 5141 | /* (2 * 8) config */ |
|
- | 5142 | default: |
|
4983 | val = vlv_punit_read(dev_priv, PUNIT_GPU_STATUS_REG); |
5143 | /* Setting (2 * 8) Min RP0 for any other combination */ |
- | 5144 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT); |
|
- | 5145 | break; |
|
- | 5146 | } |
|
- | 5147 | ||
4984 | rp0 = (val >> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT) & PUNIT_GPU_STATUS_MAX_FREQ_MASK; |
5148 | rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK); |
4985 | 5149 | ||
4986 | return rp0; |
5150 | return rp0; |
4987 | } |
5151 | } |
4988 | 5152 | ||
4989 | static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv) |
5153 | static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv) |
4990 | { |
5154 | { |
4991 | u32 val, rpe; |
5155 | u32 val, rpe; |
4992 | 5156 | ||
4993 | val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG); |
5157 | val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG); |
4994 | rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK; |
5158 | rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK; |
4995 | 5159 | ||
4996 | return rpe; |
5160 | return rpe; |
4997 | } |
5161 | } |
4998 | 5162 | ||
4999 | static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5163 | static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5000 | { |
5164 | { |
5001 | u32 val, rp1; |
5165 | u32 val, rp1; |
5002 | 5166 | ||
5003 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5167 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); |
5004 | rp1 = (val >> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT) & PUNIT_GPU_STATUS_MAX_FREQ_MASK; |
5168 | rp1 = (val & FB_GFX_FREQ_FUSE_MASK); |
5005 | 5169 | ||
5006 | return rp1; |
5170 | return rp1; |
5007 | } |
5171 | } |
5008 | - | ||
5009 | static int cherryview_rps_min_freq(struct drm_i915_private *dev_priv) |
- | |
5010 | { |
- | |
5011 | u32 val, rpn; |
- | |
5012 | - | ||
5013 | val = vlv_punit_read(dev_priv, PUNIT_GPU_STATUS_REG); |
- | |
5014 | rpn = (val >> PUNIT_GPU_STATIS_GFX_MIN_FREQ_SHIFT) & PUNIT_GPU_STATUS_GFX_MIN_FREQ_MASK; |
- | |
5015 | return rpn; |
- | |
5016 | } |
- | |
5017 | 5172 | ||
5018 | static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5173 | static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5019 | { |
5174 | { |
5020 | u32 val, rp1; |
5175 | u32 val, rp1; |
5021 | 5176 | ||
5022 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
5177 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
5023 | 5178 | ||
5024 | rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT; |
5179 | rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT; |
5025 | 5180 | ||
5026 | return rp1; |
5181 | return rp1; |
5027 | } |
5182 | } |
5028 | 5183 | ||
5029 | static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv) |
5184 | static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv) |
5030 | { |
5185 | { |
5031 | u32 val, rp0; |
5186 | u32 val, rp0; |
5032 | 5187 | ||
5033 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
5188 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
5034 | 5189 | ||
5035 | rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT; |
5190 | rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT; |
5036 | /* Clamp to max */ |
5191 | /* Clamp to max */ |
5037 | rp0 = min_t(u32, rp0, 0xea); |
5192 | rp0 = min_t(u32, rp0, 0xea); |
5038 | 5193 | ||
5039 | return rp0; |
5194 | return rp0; |
5040 | } |
5195 | } |
5041 | 5196 | ||
5042 | static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv) |
5197 | static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv) |
5043 | { |
5198 | { |
5044 | u32 val, rpe; |
5199 | u32 val, rpe; |
5045 | 5200 | ||
5046 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO); |
5201 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO); |
5047 | rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT; |
5202 | rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT; |
5048 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI); |
5203 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI); |
5049 | rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5; |
5204 | rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5; |
5050 | 5205 | ||
5051 | return rpe; |
5206 | return rpe; |
5052 | } |
5207 | } |
5053 | 5208 | ||
5054 | static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv) |
5209 | static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv) |
5055 | { |
5210 | { |
5056 | return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff; |
5211 | return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff; |
5057 | } |
5212 | } |
5058 | 5213 | ||
5059 | /* Check that the pctx buffer wasn't move under us. */ |
5214 | /* Check that the pctx buffer wasn't move under us. */ |
5060 | static void valleyview_check_pctx(struct drm_i915_private *dev_priv) |
5215 | static void valleyview_check_pctx(struct drm_i915_private *dev_priv) |
5061 | { |
5216 | { |
5062 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; |
5217 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; |
5063 | 5218 | ||
5064 | WARN_ON(pctx_addr != dev_priv->mm.stolen_base + |
5219 | WARN_ON(pctx_addr != dev_priv->mm.stolen_base + |
5065 | dev_priv->vlv_pctx->stolen->start); |
5220 | dev_priv->vlv_pctx->stolen->start); |
5066 | } |
5221 | } |
5067 | 5222 | ||
5068 | 5223 | ||
5069 | /* Check that the pcbr address is not empty. */ |
5224 | /* Check that the pcbr address is not empty. */ |
5070 | static void cherryview_check_pctx(struct drm_i915_private *dev_priv) |
5225 | static void cherryview_check_pctx(struct drm_i915_private *dev_priv) |
5071 | { |
5226 | { |
5072 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; |
5227 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; |
5073 | 5228 | ||
5074 | WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0); |
5229 | WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0); |
5075 | } |
5230 | } |
5076 | 5231 | ||
5077 | static void cherryview_setup_pctx(struct drm_device *dev) |
5232 | static void cherryview_setup_pctx(struct drm_device *dev) |
5078 | { |
5233 | { |
5079 | struct drm_i915_private *dev_priv = dev->dev_private; |
5234 | struct drm_i915_private *dev_priv = dev->dev_private; |
5080 | unsigned long pctx_paddr, paddr; |
5235 | unsigned long pctx_paddr, paddr; |
5081 | struct i915_gtt *gtt = &dev_priv->gtt; |
5236 | struct i915_gtt *gtt = &dev_priv->gtt; |
5082 | u32 pcbr; |
5237 | u32 pcbr; |
5083 | int pctx_size = 32*1024; |
5238 | int pctx_size = 32*1024; |
5084 | 5239 | ||
5085 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
5240 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
5086 | 5241 | ||
5087 | pcbr = I915_READ(VLV_PCBR); |
5242 | pcbr = I915_READ(VLV_PCBR); |
5088 | if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) { |
5243 | if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) { |
5089 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
5244 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
5090 | paddr = (dev_priv->mm.stolen_base + |
5245 | paddr = (dev_priv->mm.stolen_base + |
5091 | (gtt->stolen_size - pctx_size)); |
5246 | (gtt->stolen_size - pctx_size)); |
5092 | 5247 | ||
5093 | pctx_paddr = (paddr & (~4095)); |
5248 | pctx_paddr = (paddr & (~4095)); |
5094 | I915_WRITE(VLV_PCBR, pctx_paddr); |
5249 | I915_WRITE(VLV_PCBR, pctx_paddr); |
5095 | } |
5250 | } |
5096 | 5251 | ||
5097 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
5252 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
5098 | } |
5253 | } |
5099 | 5254 | ||
5100 | static void valleyview_setup_pctx(struct drm_device *dev) |
5255 | static void valleyview_setup_pctx(struct drm_device *dev) |
5101 | { |
5256 | { |
5102 | struct drm_i915_private *dev_priv = dev->dev_private; |
5257 | struct drm_i915_private *dev_priv = dev->dev_private; |
5103 | struct drm_i915_gem_object *pctx; |
5258 | struct drm_i915_gem_object *pctx; |
5104 | unsigned long pctx_paddr; |
5259 | unsigned long pctx_paddr; |
5105 | u32 pcbr; |
5260 | u32 pcbr; |
5106 | int pctx_size = 24*1024; |
5261 | int pctx_size = 24*1024; |
5107 | 5262 | ||
5108 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
5263 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
5109 | 5264 | ||
5110 | pcbr = I915_READ(VLV_PCBR); |
5265 | pcbr = I915_READ(VLV_PCBR); |
5111 | if (pcbr) { |
5266 | if (pcbr) { |
5112 | /* BIOS set it up already, grab the pre-alloc'd space */ |
5267 | /* BIOS set it up already, grab the pre-alloc'd space */ |
5113 | int pcbr_offset; |
5268 | int pcbr_offset; |
5114 | 5269 | ||
5115 | pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base; |
5270 | pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base; |
5116 | pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev, |
5271 | pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev, |
5117 | pcbr_offset, |
5272 | pcbr_offset, |
5118 | I915_GTT_OFFSET_NONE, |
5273 | I915_GTT_OFFSET_NONE, |
5119 | pctx_size); |
5274 | pctx_size); |
5120 | goto out; |
5275 | goto out; |
5121 | } |
5276 | } |
5122 | 5277 | ||
5123 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
5278 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
5124 | 5279 | ||
5125 | /* |
5280 | /* |
5126 | * From the Gunit register HAS: |
5281 | * From the Gunit register HAS: |
5127 | * The Gfx driver is expected to program this register and ensure |
5282 | * The Gfx driver is expected to program this register and ensure |
5128 | * proper allocation within Gfx stolen memory. For example, this |
5283 | * proper allocation within Gfx stolen memory. For example, this |
5129 | * register should be programmed such than the PCBR range does not |
5284 | * register should be programmed such than the PCBR range does not |
5130 | * overlap with other ranges, such as the frame buffer, protected |
5285 | * overlap with other ranges, such as the frame buffer, protected |
5131 | * memory, or any other relevant ranges. |
5286 | * memory, or any other relevant ranges. |
5132 | */ |
5287 | */ |
5133 | pctx = i915_gem_object_create_stolen(dev, pctx_size); |
5288 | pctx = i915_gem_object_create_stolen(dev, pctx_size); |
5134 | if (!pctx) { |
5289 | if (!pctx) { |
5135 | DRM_DEBUG("not enough stolen space for PCTX, disabling\n"); |
5290 | DRM_DEBUG("not enough stolen space for PCTX, disabling\n"); |
5136 | return; |
5291 | return; |
5137 | } |
5292 | } |
5138 | 5293 | ||
5139 | pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start; |
5294 | pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start; |
5140 | I915_WRITE(VLV_PCBR, pctx_paddr); |
5295 | I915_WRITE(VLV_PCBR, pctx_paddr); |
5141 | 5296 | ||
5142 | out: |
5297 | out: |
5143 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
5298 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
5144 | dev_priv->vlv_pctx = pctx; |
5299 | dev_priv->vlv_pctx = pctx; |
5145 | } |
5300 | } |
5146 | 5301 | ||
5147 | static void valleyview_cleanup_pctx(struct drm_device *dev) |
5302 | static void valleyview_cleanup_pctx(struct drm_device *dev) |
5148 | { |
5303 | { |
5149 | struct drm_i915_private *dev_priv = dev->dev_private; |
5304 | struct drm_i915_private *dev_priv = dev->dev_private; |
5150 | 5305 | ||
5151 | if (WARN_ON(!dev_priv->vlv_pctx)) |
5306 | if (WARN_ON(!dev_priv->vlv_pctx)) |
5152 | return; |
5307 | return; |
5153 | 5308 | ||
5154 | drm_gem_object_unreference(&dev_priv->vlv_pctx->base); |
5309 | drm_gem_object_unreference(&dev_priv->vlv_pctx->base); |
5155 | dev_priv->vlv_pctx = NULL; |
5310 | dev_priv->vlv_pctx = NULL; |
5156 | } |
5311 | } |
5157 | 5312 | ||
5158 | static void valleyview_init_gt_powersave(struct drm_device *dev) |
5313 | static void valleyview_init_gt_powersave(struct drm_device *dev) |
5159 | { |
5314 | { |
5160 | struct drm_i915_private *dev_priv = dev->dev_private; |
5315 | struct drm_i915_private *dev_priv = dev->dev_private; |
5161 | u32 val; |
5316 | u32 val; |
5162 | 5317 | ||
5163 | valleyview_setup_pctx(dev); |
5318 | valleyview_setup_pctx(dev); |
5164 | 5319 | ||
5165 | mutex_lock(&dev_priv->rps.hw_lock); |
5320 | mutex_lock(&dev_priv->rps.hw_lock); |
5166 | 5321 | ||
5167 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5322 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5168 | switch ((val >> 6) & 3) { |
5323 | switch ((val >> 6) & 3) { |
5169 | case 0: |
5324 | case 0: |
5170 | case 1: |
5325 | case 1: |
5171 | dev_priv->mem_freq = 800; |
5326 | dev_priv->mem_freq = 800; |
5172 | break; |
5327 | break; |
5173 | case 2: |
5328 | case 2: |
5174 | dev_priv->mem_freq = 1066; |
5329 | dev_priv->mem_freq = 1066; |
5175 | break; |
5330 | break; |
5176 | case 3: |
5331 | case 3: |
5177 | dev_priv->mem_freq = 1333; |
5332 | dev_priv->mem_freq = 1333; |
5178 | break; |
5333 | break; |
5179 | } |
5334 | } |
5180 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
5335 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
5181 | 5336 | ||
5182 | dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv); |
5337 | dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv); |
5183 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; |
5338 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; |
5184 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", |
5339 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", |
5185 | vlv_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
5340 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
5186 | dev_priv->rps.max_freq); |
5341 | dev_priv->rps.max_freq); |
5187 | 5342 | ||
5188 | dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv); |
5343 | dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv); |
5189 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", |
5344 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", |
5190 | vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5345 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5191 | dev_priv->rps.efficient_freq); |
5346 | dev_priv->rps.efficient_freq); |
5192 | 5347 | ||
5193 | dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv); |
5348 | dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv); |
5194 | DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n", |
5349 | DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n", |
5195 | vlv_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
5350 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
5196 | dev_priv->rps.rp1_freq); |
5351 | dev_priv->rps.rp1_freq); |
5197 | 5352 | ||
5198 | dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv); |
5353 | dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv); |
5199 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
5354 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
5200 | vlv_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
5355 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
5201 | dev_priv->rps.min_freq); |
5356 | dev_priv->rps.min_freq); |
- | 5357 | ||
- | 5358 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; |
|
5202 | 5359 | ||
5203 | /* Preserve min/max settings in case of re-init */ |
5360 | /* Preserve min/max settings in case of re-init */ |
5204 | if (dev_priv->rps.max_freq_softlimit == 0) |
5361 | if (dev_priv->rps.max_freq_softlimit == 0) |
5205 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; |
5362 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; |
5206 | 5363 | ||
5207 | if (dev_priv->rps.min_freq_softlimit == 0) |
5364 | if (dev_priv->rps.min_freq_softlimit == 0) |
5208 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; |
5365 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; |
5209 | 5366 | ||
5210 | mutex_unlock(&dev_priv->rps.hw_lock); |
5367 | mutex_unlock(&dev_priv->rps.hw_lock); |
5211 | } |
5368 | } |
5212 | 5369 | ||
5213 | static void cherryview_init_gt_powersave(struct drm_device *dev) |
5370 | static void cherryview_init_gt_powersave(struct drm_device *dev) |
5214 | { |
5371 | { |
5215 | struct drm_i915_private *dev_priv = dev->dev_private; |
5372 | struct drm_i915_private *dev_priv = dev->dev_private; |
5216 | u32 val; |
5373 | u32 val; |
5217 | 5374 | ||
5218 | cherryview_setup_pctx(dev); |
5375 | cherryview_setup_pctx(dev); |
5219 | 5376 | ||
5220 | mutex_lock(&dev_priv->rps.hw_lock); |
5377 | mutex_lock(&dev_priv->rps.hw_lock); |
5221 | 5378 | ||
5222 | mutex_lock(&dev_priv->dpio_lock); |
5379 | mutex_lock(&dev_priv->sb_lock); |
5223 | val = vlv_cck_read(dev_priv, CCK_FUSE_REG); |
5380 | val = vlv_cck_read(dev_priv, CCK_FUSE_REG); |
5224 | mutex_unlock(&dev_priv->dpio_lock); |
5381 | mutex_unlock(&dev_priv->sb_lock); |
5225 | - | ||
5226 | switch ((val >> 2) & 0x7) { |
- | |
5227 | case 0: |
- | |
5228 | case 1: |
- | |
5229 | dev_priv->rps.cz_freq = 200; |
- | |
5230 | dev_priv->mem_freq = 1600; |
- | |
5231 | break; |
- | |
5232 | case 2: |
- | |
5233 | dev_priv->rps.cz_freq = 267; |
- | |
5234 | dev_priv->mem_freq = 1600; |
5382 | |
5235 | break; |
- | |
5236 | case 3: |
5383 | switch ((val >> 2) & 0x7) { |
5237 | dev_priv->rps.cz_freq = 333; |
5384 | case 3: |
5238 | dev_priv->mem_freq = 2000; |
- | |
5239 | break; |
- | |
5240 | case 4: |
- | |
5241 | dev_priv->rps.cz_freq = 320; |
- | |
5242 | dev_priv->mem_freq = 1600; |
5385 | dev_priv->mem_freq = 2000; |
5243 | break; |
- | |
5244 | case 5: |
5386 | break; |
5245 | dev_priv->rps.cz_freq = 400; |
5387 | default: |
5246 | dev_priv->mem_freq = 1600; |
5388 | dev_priv->mem_freq = 1600; |
5247 | break; |
5389 | break; |
5248 | } |
5390 | } |
5249 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
5391 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
5250 | 5392 | ||
5251 | dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv); |
5393 | dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv); |
5252 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; |
5394 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; |
5253 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", |
5395 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", |
5254 | vlv_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
5396 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
5255 | dev_priv->rps.max_freq); |
5397 | dev_priv->rps.max_freq); |
5256 | 5398 | ||
5257 | dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv); |
5399 | dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv); |
5258 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", |
5400 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", |
5259 | vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5401 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5260 | dev_priv->rps.efficient_freq); |
5402 | dev_priv->rps.efficient_freq); |
5261 | 5403 | ||
5262 | dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv); |
5404 | dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv); |
5263 | DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n", |
5405 | DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n", |
5264 | vlv_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
5406 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
5265 | dev_priv->rps.rp1_freq); |
5407 | dev_priv->rps.rp1_freq); |
- | 5408 | ||
5266 | 5409 | /* PUnit validated range is only [RPe, RP0] */ |
|
5267 | dev_priv->rps.min_freq = cherryview_rps_min_freq(dev_priv); |
5410 | dev_priv->rps.min_freq = dev_priv->rps.efficient_freq; |
5268 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
5411 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
5269 | vlv_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
5412 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
5270 | dev_priv->rps.min_freq); |
5413 | dev_priv->rps.min_freq); |
5271 | 5414 | ||
5272 | WARN_ONCE((dev_priv->rps.max_freq | |
5415 | WARN_ONCE((dev_priv->rps.max_freq | |
5273 | dev_priv->rps.efficient_freq | |
5416 | dev_priv->rps.efficient_freq | |
5274 | dev_priv->rps.rp1_freq | |
5417 | dev_priv->rps.rp1_freq | |
5275 | dev_priv->rps.min_freq) & 1, |
5418 | dev_priv->rps.min_freq) & 1, |
5276 | "Odd GPU freq values\n"); |
5419 | "Odd GPU freq values\n"); |
- | 5420 | ||
- | 5421 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; |
|
5277 | 5422 | ||
5278 | /* Preserve min/max settings in case of re-init */ |
5423 | /* Preserve min/max settings in case of re-init */ |
5279 | if (dev_priv->rps.max_freq_softlimit == 0) |
5424 | if (dev_priv->rps.max_freq_softlimit == 0) |
5280 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; |
5425 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; |
5281 | 5426 | ||
5282 | if (dev_priv->rps.min_freq_softlimit == 0) |
5427 | if (dev_priv->rps.min_freq_softlimit == 0) |
5283 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; |
5428 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; |
5284 | 5429 | ||
5285 | mutex_unlock(&dev_priv->rps.hw_lock); |
5430 | mutex_unlock(&dev_priv->rps.hw_lock); |
5286 | } |
5431 | } |
5287 | 5432 | ||
5288 | static void valleyview_cleanup_gt_powersave(struct drm_device *dev) |
5433 | static void valleyview_cleanup_gt_powersave(struct drm_device *dev) |
5289 | { |
5434 | { |
5290 | valleyview_cleanup_pctx(dev); |
5435 | valleyview_cleanup_pctx(dev); |
5291 | } |
5436 | } |
5292 | 5437 | ||
5293 | static void cherryview_enable_rps(struct drm_device *dev) |
5438 | static void cherryview_enable_rps(struct drm_device *dev) |
5294 | { |
5439 | { |
5295 | struct drm_i915_private *dev_priv = dev->dev_private; |
5440 | struct drm_i915_private *dev_priv = dev->dev_private; |
5296 | struct intel_engine_cs *ring; |
5441 | struct intel_engine_cs *ring; |
5297 | u32 gtfifodbg, val, rc6_mode = 0, pcbr; |
5442 | u32 gtfifodbg, val, rc6_mode = 0, pcbr; |
5298 | int i; |
5443 | int i; |
5299 | 5444 | ||
5300 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
5445 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
5301 | 5446 | ||
5302 | gtfifodbg = I915_READ(GTFIFODBG); |
5447 | gtfifodbg = I915_READ(GTFIFODBG); |
5303 | if (gtfifodbg) { |
5448 | if (gtfifodbg) { |
5304 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
5449 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
5305 | gtfifodbg); |
5450 | gtfifodbg); |
5306 | I915_WRITE(GTFIFODBG, gtfifodbg); |
5451 | I915_WRITE(GTFIFODBG, gtfifodbg); |
5307 | } |
5452 | } |
5308 | 5453 | ||
5309 | cherryview_check_pctx(dev_priv); |
5454 | cherryview_check_pctx(dev_priv); |
5310 | 5455 | ||
5311 | /* 1a & 1b: Get forcewake during program sequence. Although the driver |
5456 | /* 1a & 1b: Get forcewake during program sequence. Although the driver |
5312 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ |
5457 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ |
5313 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL); |
5458 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
- | 5459 | ||
- | 5460 | /* Disable RC states. */ |
|
- | 5461 | I915_WRITE(GEN6_RC_CONTROL, 0); |
|
5314 | 5462 | ||
5315 | /* 2a: Program RC6 thresholds.*/ |
5463 | /* 2a: Program RC6 thresholds.*/ |
5316 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); |
5464 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); |
5317 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
5465 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
5318 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ |
5466 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ |
5319 | 5467 | ||
5320 | for_each_ring(ring, dev_priv, i) |
5468 | for_each_ring(ring, dev_priv, i) |
5321 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
5469 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
5322 | I915_WRITE(GEN6_RC_SLEEP, 0); |
5470 | I915_WRITE(GEN6_RC_SLEEP, 0); |
- | 5471 | ||
5323 | 5472 | /* TO threshold set to 500 us ( 0x186 * 1.28 us) */ |
|
5324 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */ |
5473 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x186); |
5325 | 5474 | ||
5326 | /* allows RC6 residency counter to work */ |
5475 | /* allows RC6 residency counter to work */ |
5327 | I915_WRITE(VLV_COUNTER_CONTROL, |
5476 | I915_WRITE(VLV_COUNTER_CONTROL, |
5328 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH | |
5477 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH | |
5329 | VLV_MEDIA_RC6_COUNT_EN | |
5478 | VLV_MEDIA_RC6_COUNT_EN | |
5330 | VLV_RENDER_RC6_COUNT_EN)); |
5479 | VLV_RENDER_RC6_COUNT_EN)); |
5331 | 5480 | ||
5332 | /* For now we assume BIOS is allocating and populating the PCBR */ |
5481 | /* For now we assume BIOS is allocating and populating the PCBR */ |
5333 | pcbr = I915_READ(VLV_PCBR); |
5482 | pcbr = I915_READ(VLV_PCBR); |
5334 | 5483 | ||
5335 | /* 3: Enable RC6 */ |
5484 | /* 3: Enable RC6 */ |
5336 | if ((intel_enable_rc6(dev) & INTEL_RC6_ENABLE) && |
5485 | if ((intel_enable_rc6(dev) & INTEL_RC6_ENABLE) && |
5337 | (pcbr >> VLV_PCBR_ADDR_SHIFT)) |
5486 | (pcbr >> VLV_PCBR_ADDR_SHIFT)) |
5338 | rc6_mode = GEN6_RC_CTL_EI_MODE(1); |
5487 | rc6_mode = GEN7_RC_CTL_TO_MODE; |
5339 | 5488 | ||
5340 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
5489 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
5341 | 5490 | ||
5342 | /* 4 Program defaults and thresholds for RPS*/ |
5491 | /* 4 Program defaults and thresholds for RPS*/ |
- | 5492 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
|
5343 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
5493 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
5344 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); |
5494 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); |
5345 | I915_WRITE(GEN6_RP_UP_EI, 66000); |
5495 | I915_WRITE(GEN6_RP_UP_EI, 66000); |
5346 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); |
5496 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); |
5347 | 5497 | ||
5348 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
5498 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
5349 | - | ||
5350 | /* WaDisablePwrmtrEvent:chv (pre-production hw) */ |
- | |
5351 | I915_WRITE(0xA80C, I915_READ(0xA80C) & 0x00ffffff); |
- | |
5352 | I915_WRITE(0xA810, I915_READ(0xA810) & 0xffffff00); |
- | |
5353 | 5499 | ||
5354 | /* 5: Enable RPS */ |
5500 | /* 5: Enable RPS */ |
5355 | I915_WRITE(GEN6_RP_CONTROL, |
5501 | I915_WRITE(GEN6_RP_CONTROL, |
5356 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
5502 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
5357 | GEN6_RP_MEDIA_IS_GFX | /* WaSetMaskForGfxBusyness:chv (pre-production hw ?) */ |
5503 | GEN6_RP_MEDIA_IS_GFX | |
5358 | GEN6_RP_ENABLE | |
5504 | GEN6_RP_ENABLE | |
5359 | GEN6_RP_UP_BUSY_AVG | |
5505 | GEN6_RP_UP_BUSY_AVG | |
5360 | GEN6_RP_DOWN_IDLE_AVG); |
5506 | GEN6_RP_DOWN_IDLE_AVG); |
- | 5507 | ||
- | 5508 | /* Setting Fixed Bias */ |
|
- | 5509 | val = VLV_OVERRIDE_EN | |
|
- | 5510 | VLV_SOC_TDP_EN | |
|
- | 5511 | CHV_BIAS_CPU_50_SOC_50; |
|
- | 5512 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); |
|
5361 | 5513 | ||
5362 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5514 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5363 | 5515 | ||
5364 | /* RPS code assumes GPLL is used */ |
5516 | /* RPS code assumes GPLL is used */ |
5365 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); |
5517 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); |
5366 | 5518 | ||
5367 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & GPLLENABLE ? "yes" : "no"); |
5519 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE)); |
5368 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
5520 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
5369 | 5521 | ||
5370 | dev_priv->rps.cur_freq = (val >> 8) & 0xff; |
5522 | dev_priv->rps.cur_freq = (val >> 8) & 0xff; |
5371 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", |
5523 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", |
5372 | vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq), |
5524 | intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq), |
5373 | dev_priv->rps.cur_freq); |
5525 | dev_priv->rps.cur_freq); |
5374 | 5526 | ||
5375 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", |
5527 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", |
5376 | vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5528 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5377 | dev_priv->rps.efficient_freq); |
5529 | dev_priv->rps.efficient_freq); |
5378 | 5530 | ||
5379 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq); |
5531 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq); |
5380 | 5532 | ||
5381 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL); |
5533 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
5382 | } |
5534 | } |
5383 | 5535 | ||
5384 | static void valleyview_enable_rps(struct drm_device *dev) |
5536 | static void valleyview_enable_rps(struct drm_device *dev) |
5385 | { |
5537 | { |
5386 | struct drm_i915_private *dev_priv = dev->dev_private; |
5538 | struct drm_i915_private *dev_priv = dev->dev_private; |
5387 | struct intel_engine_cs *ring; |
5539 | struct intel_engine_cs *ring; |
5388 | u32 gtfifodbg, val, rc6_mode = 0; |
5540 | u32 gtfifodbg, val, rc6_mode = 0; |
5389 | int i; |
5541 | int i; |
5390 | 5542 | ||
5391 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
5543 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
5392 | 5544 | ||
5393 | valleyview_check_pctx(dev_priv); |
5545 | valleyview_check_pctx(dev_priv); |
5394 | 5546 | ||
5395 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { |
5547 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { |
5396 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
5548 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
5397 | gtfifodbg); |
5549 | gtfifodbg); |
5398 | I915_WRITE(GTFIFODBG, gtfifodbg); |
5550 | I915_WRITE(GTFIFODBG, gtfifodbg); |
5399 | } |
5551 | } |
5400 | 5552 | ||
5401 | /* If VLV, Forcewake all wells, else re-direct to regular path */ |
5553 | /* If VLV, Forcewake all wells, else re-direct to regular path */ |
5402 | gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL); |
5554 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
- | 5555 | ||
- | 5556 | /* Disable RC states. */ |
|
- | 5557 | I915_WRITE(GEN6_RC_CONTROL, 0); |
|
- | 5558 | ||
5403 | 5559 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
|
5404 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
5560 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
5405 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); |
5561 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); |
5406 | I915_WRITE(GEN6_RP_UP_EI, 66000); |
5562 | I915_WRITE(GEN6_RP_UP_EI, 66000); |
5407 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); |
5563 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); |
5408 | 5564 | ||
5409 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
5565 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
5410 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 0xf4240); |
- | |
5411 | 5566 | ||
5412 | I915_WRITE(GEN6_RP_CONTROL, |
5567 | I915_WRITE(GEN6_RP_CONTROL, |
5413 | GEN6_RP_MEDIA_TURBO | |
5568 | GEN6_RP_MEDIA_TURBO | |
5414 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
5569 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
5415 | GEN6_RP_MEDIA_IS_GFX | |
5570 | GEN6_RP_MEDIA_IS_GFX | |
5416 | GEN6_RP_ENABLE | |
5571 | GEN6_RP_ENABLE | |
5417 | GEN6_RP_UP_BUSY_AVG | |
5572 | GEN6_RP_UP_BUSY_AVG | |
5418 | GEN6_RP_DOWN_IDLE_CONT); |
5573 | GEN6_RP_DOWN_IDLE_CONT); |
5419 | 5574 | ||
5420 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000); |
5575 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000); |
5421 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); |
5576 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); |
5422 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); |
5577 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); |
5423 | 5578 | ||
5424 | for_each_ring(ring, dev_priv, i) |
5579 | for_each_ring(ring, dev_priv, i) |
5425 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
5580 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
5426 | 5581 | ||
5427 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x557); |
5582 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x557); |
5428 | 5583 | ||
5429 | /* allows RC6 residency counter to work */ |
5584 | /* allows RC6 residency counter to work */ |
5430 | I915_WRITE(VLV_COUNTER_CONTROL, |
5585 | I915_WRITE(VLV_COUNTER_CONTROL, |
5431 | _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN | |
5586 | _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN | |
5432 | VLV_RENDER_RC0_COUNT_EN | |
5587 | VLV_RENDER_RC0_COUNT_EN | |
5433 | VLV_MEDIA_RC6_COUNT_EN | |
5588 | VLV_MEDIA_RC6_COUNT_EN | |
5434 | VLV_RENDER_RC6_COUNT_EN)); |
5589 | VLV_RENDER_RC6_COUNT_EN)); |
5435 | 5590 | ||
5436 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
5591 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
5437 | rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL; |
5592 | rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL; |
5438 | 5593 | ||
5439 | intel_print_rc6_info(dev, rc6_mode); |
5594 | intel_print_rc6_info(dev, rc6_mode); |
5440 | 5595 | ||
5441 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
5596 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
- | 5597 | ||
- | 5598 | /* Setting Fixed Bias */ |
|
- | 5599 | val = VLV_OVERRIDE_EN | |
|
- | 5600 | VLV_SOC_TDP_EN | |
|
- | 5601 | VLV_BIAS_CPU_125_SOC_875; |
|
- | 5602 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); |
|
5442 | 5603 | ||
5443 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5604 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5444 | 5605 | ||
5445 | /* RPS code assumes GPLL is used */ |
5606 | /* RPS code assumes GPLL is used */ |
5446 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); |
5607 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); |
5447 | 5608 | ||
5448 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & GPLLENABLE ? "yes" : "no"); |
5609 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE)); |
5449 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
5610 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
5450 | 5611 | ||
5451 | dev_priv->rps.cur_freq = (val >> 8) & 0xff; |
5612 | dev_priv->rps.cur_freq = (val >> 8) & 0xff; |
5452 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", |
5613 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", |
5453 | vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq), |
5614 | intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq), |
5454 | dev_priv->rps.cur_freq); |
5615 | dev_priv->rps.cur_freq); |
5455 | 5616 | ||
5456 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", |
5617 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", |
5457 | vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5618 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
5458 | dev_priv->rps.efficient_freq); |
5619 | dev_priv->rps.efficient_freq); |
5459 | 5620 | ||
5460 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq); |
5621 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq); |
5461 | 5622 | ||
5462 | gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL); |
- | |
5463 | } |
- | |
5464 | - | ||
5465 | void ironlake_teardown_rc6(struct drm_device *dev) |
- | |
5466 | { |
- | |
5467 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
5468 | - | ||
5469 | if (dev_priv->ips.renderctx) { |
- | |
5470 | i915_gem_object_ggtt_unpin(dev_priv->ips.renderctx); |
- | |
5471 | drm_gem_object_unreference(&dev_priv->ips.renderctx->base); |
- | |
5472 | dev_priv->ips.renderctx = NULL; |
- | |
5473 | } |
- | |
5474 | - | ||
5475 | if (dev_priv->ips.pwrctx) { |
- | |
5476 | i915_gem_object_ggtt_unpin(dev_priv->ips.pwrctx); |
- | |
5477 | drm_gem_object_unreference(&dev_priv->ips.pwrctx->base); |
- | |
5478 | dev_priv->ips.pwrctx = NULL; |
- | |
5479 | } |
- | |
5480 | } |
- | |
5481 | - | ||
5482 | static void ironlake_disable_rc6(struct drm_device *dev) |
- | |
5483 | { |
- | |
5484 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
5485 | - | ||
5486 | if (I915_READ(PWRCTXA)) { |
- | |
5487 | /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */ |
- | |
5488 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT); |
- | |
5489 | wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON), |
- | |
5490 | 50); |
- | |
5491 | - | ||
5492 | I915_WRITE(PWRCTXA, 0); |
- | |
5493 | POSTING_READ(PWRCTXA); |
- | |
5494 | - | ||
5495 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT); |
- | |
5496 | POSTING_READ(RSTDBYCTL); |
- | |
5497 | } |
- | |
5498 | } |
- | |
5499 | - | ||
5500 | static int ironlake_setup_rc6(struct drm_device *dev) |
- | |
5501 | { |
- | |
5502 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
5503 | - | ||
5504 | if (dev_priv->ips.renderctx == NULL) |
- | |
5505 | dev_priv->ips.renderctx = intel_alloc_context_page(dev); |
- | |
5506 | if (!dev_priv->ips.renderctx) |
- | |
5507 | return -ENOMEM; |
- | |
5508 | - | ||
5509 | if (dev_priv->ips.pwrctx == NULL) |
- | |
5510 | dev_priv->ips.pwrctx = intel_alloc_context_page(dev); |
- | |
5511 | if (!dev_priv->ips.pwrctx) { |
- | |
5512 | ironlake_teardown_rc6(dev); |
- | |
5513 | return -ENOMEM; |
- | |
5514 | } |
- | |
5515 | - | ||
5516 | return 0; |
- | |
5517 | } |
- | |
5518 | - | ||
5519 | static void ironlake_enable_rc6(struct drm_device *dev) |
- | |
5520 | { |
- | |
5521 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
5522 | struct intel_engine_cs *ring = &dev_priv->ring[RCS]; |
- | |
5523 | bool was_interruptible; |
- | |
5524 | int ret; |
- | |
5525 | - | ||
5526 | /* rc6 disabled by default due to repeated reports of hanging during |
- | |
5527 | * boot and resume. |
- | |
5528 | */ |
- | |
5529 | if (!intel_enable_rc6(dev)) |
- | |
5530 | return; |
- | |
5531 | - | ||
5532 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
- | |
5533 | - | ||
5534 | ret = ironlake_setup_rc6(dev); |
- | |
5535 | if (ret) |
- | |
5536 | return; |
- | |
5537 | - | ||
5538 | was_interruptible = dev_priv->mm.interruptible; |
- | |
5539 | dev_priv->mm.interruptible = false; |
- | |
5540 | - | ||
5541 | /* |
- | |
5542 | * GPU can automatically power down the render unit if given a page |
- | |
5543 | * to save state. |
- | |
5544 | */ |
- | |
5545 | ret = intel_ring_begin(ring, 6); |
- | |
5546 | if (ret) { |
- | |
5547 | ironlake_teardown_rc6(dev); |
- | |
5548 | dev_priv->mm.interruptible = was_interruptible; |
- | |
5549 | return; |
- | |
5550 | } |
- | |
5551 | - | ||
5552 | intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN); |
- | |
5553 | intel_ring_emit(ring, MI_SET_CONTEXT); |
- | |
5554 | intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) | |
- | |
5555 | MI_MM_SPACE_GTT | |
- | |
5556 | MI_SAVE_EXT_STATE_EN | |
- | |
5557 | MI_RESTORE_EXT_STATE_EN | |
- | |
5558 | MI_RESTORE_INHIBIT); |
- | |
5559 | intel_ring_emit(ring, MI_SUSPEND_FLUSH); |
- | |
5560 | intel_ring_emit(ring, MI_NOOP); |
- | |
5561 | intel_ring_emit(ring, MI_FLUSH); |
- | |
5562 | intel_ring_advance(ring); |
- | |
5563 | - | ||
5564 | /* |
- | |
5565 | * Wait for the command parser to advance past MI_SET_CONTEXT. The HW |
- | |
5566 | * does an implicit flush, combined with MI_FLUSH above, it should be |
- | |
5567 | * safe to assume that renderctx is valid |
- | |
5568 | */ |
- | |
5569 | ret = intel_ring_idle(ring); |
- | |
5570 | dev_priv->mm.interruptible = was_interruptible; |
- | |
5571 | if (ret) { |
- | |
5572 | DRM_ERROR("failed to enable ironlake power savings\n"); |
- | |
5573 | ironlake_teardown_rc6(dev); |
- | |
5574 | return; |
- | |
5575 | } |
- | |
5576 | - | ||
5577 | I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN); |
- | |
5578 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT); |
- | |
5579 | - | ||
5580 | intel_print_rc6_info(dev, GEN6_RC_CTL_RC6_ENABLE); |
5623 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
5581 | } |
5624 | } |
5582 | 5625 | ||
5583 | static unsigned long intel_pxfreq(u32 vidfreq) |
5626 | static unsigned long intel_pxfreq(u32 vidfreq) |
5584 | { |
5627 | { |
5585 | unsigned long freq; |
5628 | unsigned long freq; |
5586 | int div = (vidfreq & 0x3f0000) >> 16; |
5629 | int div = (vidfreq & 0x3f0000) >> 16; |
5587 | int post = (vidfreq & 0x3000) >> 12; |
5630 | int post = (vidfreq & 0x3000) >> 12; |
5588 | int pre = (vidfreq & 0x7); |
5631 | int pre = (vidfreq & 0x7); |
5589 | 5632 | ||
5590 | if (!pre) |
5633 | if (!pre) |
5591 | return 0; |
5634 | return 0; |
5592 | 5635 | ||
5593 | freq = ((div * 133333) / ((1< |
5636 | freq = ((div * 133333) / ((1< |
5594 | 5637 | ||
5595 | return freq; |
5638 | return freq; |
5596 | } |
5639 | } |
5597 | 5640 | ||
5598 | static const struct cparams { |
5641 | static const struct cparams { |
5599 | u16 i; |
5642 | u16 i; |
5600 | u16 t; |
5643 | u16 t; |
5601 | u16 m; |
5644 | u16 m; |
5602 | u16 c; |
5645 | u16 c; |
5603 | } cparams[] = { |
5646 | } cparams[] = { |
5604 | { 1, 1333, 301, 28664 }, |
5647 | { 1, 1333, 301, 28664 }, |
5605 | { 1, 1066, 294, 24460 }, |
5648 | { 1, 1066, 294, 24460 }, |
5606 | { 1, 800, 294, 25192 }, |
5649 | { 1, 800, 294, 25192 }, |
5607 | { 0, 1333, 276, 27605 }, |
5650 | { 0, 1333, 276, 27605 }, |
5608 | { 0, 1066, 276, 27605 }, |
5651 | { 0, 1066, 276, 27605 }, |
5609 | { 0, 800, 231, 23784 }, |
5652 | { 0, 800, 231, 23784 }, |
5610 | }; |
5653 | }; |
5611 | 5654 | ||
5612 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
5655 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
5613 | { |
5656 | { |
5614 | u64 total_count, diff, ret; |
5657 | u64 total_count, diff, ret; |
5615 | u32 count1, count2, count3, m = 0, c = 0; |
5658 | u32 count1, count2, count3, m = 0, c = 0; |
5616 | unsigned long now = jiffies_to_msecs(jiffies), diff1; |
5659 | unsigned long now = jiffies_to_msecs(jiffies), diff1; |
5617 | int i; |
5660 | int i; |
5618 | 5661 | ||
5619 | assert_spin_locked(&mchdev_lock); |
5662 | assert_spin_locked(&mchdev_lock); |
5620 | 5663 | ||
5621 | diff1 = now - dev_priv->ips.last_time1; |
5664 | diff1 = now - dev_priv->ips.last_time1; |
5622 | 5665 | ||
5623 | /* Prevent division-by-zero if we are asking too fast. |
5666 | /* Prevent division-by-zero if we are asking too fast. |
5624 | * Also, we don't get interesting results if we are polling |
5667 | * Also, we don't get interesting results if we are polling |
5625 | * faster than once in 10ms, so just return the saved value |
5668 | * faster than once in 10ms, so just return the saved value |
5626 | * in such cases. |
5669 | * in such cases. |
5627 | */ |
5670 | */ |
5628 | if (diff1 <= 10) |
5671 | if (diff1 <= 10) |
5629 | return dev_priv->ips.chipset_power; |
5672 | return dev_priv->ips.chipset_power; |
5630 | 5673 | ||
5631 | count1 = I915_READ(DMIEC); |
5674 | count1 = I915_READ(DMIEC); |
5632 | count2 = I915_READ(DDREC); |
5675 | count2 = I915_READ(DDREC); |
5633 | count3 = I915_READ(CSIEC); |
5676 | count3 = I915_READ(CSIEC); |
5634 | 5677 | ||
5635 | total_count = count1 + count2 + count3; |
5678 | total_count = count1 + count2 + count3; |
5636 | 5679 | ||
5637 | /* FIXME: handle per-counter overflow */ |
5680 | /* FIXME: handle per-counter overflow */ |
5638 | if (total_count < dev_priv->ips.last_count1) { |
5681 | if (total_count < dev_priv->ips.last_count1) { |
5639 | diff = ~0UL - dev_priv->ips.last_count1; |
5682 | diff = ~0UL - dev_priv->ips.last_count1; |
5640 | diff += total_count; |
5683 | diff += total_count; |
5641 | } else { |
5684 | } else { |
5642 | diff = total_count - dev_priv->ips.last_count1; |
5685 | diff = total_count - dev_priv->ips.last_count1; |
5643 | } |
5686 | } |
5644 | 5687 | ||
5645 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { |
5688 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { |
5646 | if (cparams[i].i == dev_priv->ips.c_m && |
5689 | if (cparams[i].i == dev_priv->ips.c_m && |
5647 | cparams[i].t == dev_priv->ips.r_t) { |
5690 | cparams[i].t == dev_priv->ips.r_t) { |
5648 | m = cparams[i].m; |
5691 | m = cparams[i].m; |
5649 | c = cparams[i].c; |
5692 | c = cparams[i].c; |
5650 | break; |
5693 | break; |
5651 | } |
5694 | } |
5652 | } |
5695 | } |
5653 | 5696 | ||
5654 | diff = div_u64(diff, diff1); |
5697 | diff = div_u64(diff, diff1); |
5655 | ret = ((m * diff) + c); |
5698 | ret = ((m * diff) + c); |
5656 | ret = div_u64(ret, 10); |
5699 | ret = div_u64(ret, 10); |
5657 | 5700 | ||
5658 | dev_priv->ips.last_count1 = total_count; |
5701 | dev_priv->ips.last_count1 = total_count; |
5659 | dev_priv->ips.last_time1 = now; |
5702 | dev_priv->ips.last_time1 = now; |
5660 | 5703 | ||
5661 | dev_priv->ips.chipset_power = ret; |
5704 | dev_priv->ips.chipset_power = ret; |
5662 | 5705 | ||
5663 | return ret; |
5706 | return ret; |
5664 | } |
5707 | } |
5665 | 5708 | ||
5666 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
5709 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
5667 | { |
5710 | { |
5668 | struct drm_device *dev = dev_priv->dev; |
5711 | struct drm_device *dev = dev_priv->dev; |
5669 | unsigned long val; |
5712 | unsigned long val; |
5670 | 5713 | ||
5671 | if (INTEL_INFO(dev)->gen != 5) |
5714 | if (INTEL_INFO(dev)->gen != 5) |
5672 | return 0; |
5715 | return 0; |
5673 | 5716 | ||
5674 | spin_lock_irq(&mchdev_lock); |
5717 | spin_lock_irq(&mchdev_lock); |
5675 | 5718 | ||
5676 | val = __i915_chipset_val(dev_priv); |
5719 | val = __i915_chipset_val(dev_priv); |
5677 | 5720 | ||
5678 | spin_unlock_irq(&mchdev_lock); |
5721 | spin_unlock_irq(&mchdev_lock); |
5679 | 5722 | ||
5680 | return val; |
5723 | return val; |
5681 | } |
5724 | } |
5682 | 5725 | ||
5683 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
5726 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
5684 | { |
5727 | { |
5685 | unsigned long m, x, b; |
5728 | unsigned long m, x, b; |
5686 | u32 tsfs; |
5729 | u32 tsfs; |
5687 | 5730 | ||
5688 | tsfs = I915_READ(TSFS); |
5731 | tsfs = I915_READ(TSFS); |
5689 | 5732 | ||
5690 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); |
5733 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); |
5691 | x = I915_READ8(TR1); |
5734 | x = I915_READ8(TR1); |
5692 | 5735 | ||
5693 | b = tsfs & TSFS_INTR_MASK; |
5736 | b = tsfs & TSFS_INTR_MASK; |
5694 | 5737 | ||
5695 | return ((m * x) / 127) - b; |
5738 | return ((m * x) / 127) - b; |
5696 | } |
5739 | } |
- | 5740 | ||
- | 5741 | static int _pxvid_to_vd(u8 pxvid) |
|
- | 5742 | { |
|
- | 5743 | if (pxvid == 0) |
|
- | 5744 | return 0; |
|
- | 5745 | ||
- | 5746 | if (pxvid >= 8 && pxvid < 31) |
|
- | 5747 | pxvid = 31; |
|
- | 5748 | ||
- | 5749 | return (pxvid + 2) * 125; |
|
- | 5750 | } |
|
5697 | 5751 | ||
5698 | static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) |
5752 | static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) |
5699 | { |
5753 | { |
5700 | struct drm_device *dev = dev_priv->dev; |
5754 | struct drm_device *dev = dev_priv->dev; |
5701 | static const struct v_table { |
5755 | const int vd = _pxvid_to_vd(pxvid); |
5702 | u16 vd; /* in .1 mil */ |
- | |
5703 | u16 vm; /* in .1 mil */ |
- | |
5704 | } v_table[] = { |
- | |
5705 | { 0, 0, }, |
- | |
5706 | { 375, 0, }, |
- | |
5707 | { 500, 0, }, |
- | |
5708 | { 625, 0, }, |
- | |
5709 | { 750, 0, }, |
- | |
5710 | { 875, 0, }, |
- | |
5711 | { 1000, 0, }, |
- | |
5712 | { 1125, 0, }, |
- | |
5713 | { 4125, 3000, }, |
- | |
5714 | { 4125, 3000, }, |
- | |
5715 | { 4125, 3000, }, |
- | |
5716 | { 4125, 3000, }, |
- | |
5717 | { 4125, 3000, }, |
- | |
5718 | { 4125, 3000, }, |
- | |
5719 | { 4125, 3000, }, |
- | |
5720 | { 4125, 3000, }, |
- | |
5721 | { 4125, 3000, }, |
- | |
5722 | { 4125, 3000, }, |
- | |
5723 | { 4125, 3000, }, |
- | |
5724 | { 4125, 3000, }, |
- | |
5725 | { 4125, 3000, }, |
- | |
5726 | { 4125, 3000, }, |
- | |
5727 | { 4125, 3000, }, |
- | |
5728 | { 4125, 3000, }, |
- | |
5729 | { 4125, 3000, }, |
- | |
5730 | { 4125, 3000, }, |
- | |
5731 | { 4125, 3000, }, |
- | |
5732 | { 4125, 3000, }, |
- | |
5733 | { 4125, 3000, }, |
- | |
5734 | { 4125, 3000, }, |
- | |
5735 | { 4125, 3000, }, |
- | |
5736 | { 4125, 3000, }, |
- | |
5737 | { 4250, 3125, }, |
- | |
5738 | { 4375, 3250, }, |
- | |
5739 | { 4500, 3375, }, |
- | |
5740 | { 4625, 3500, }, |
- | |
5741 | { 4750, 3625, }, |
- | |
5742 | { 4875, 3750, }, |
- | |
5743 | { 5000, 3875, }, |
- | |
5744 | { 5125, 4000, }, |
- | |
5745 | { 5250, 4125, }, |
- | |
5746 | { 5375, 4250, }, |
- | |
5747 | { 5500, 4375, }, |
- | |
5748 | { 5625, 4500, }, |
- | |
5749 | { 5750, 4625, }, |
- | |
5750 | { 5875, 4750, }, |
- | |
5751 | { 6000, 4875, }, |
- | |
5752 | { 6125, 5000, }, |
- | |
5753 | { 6250, 5125, }, |
- | |
5754 | { 6375, 5250, }, |
- | |
5755 | { 6500, 5375, }, |
- | |
5756 | { 6625, 5500, }, |
- | |
5757 | { 6750, 5625, }, |
- | |
5758 | { 6875, 5750, }, |
- | |
5759 | { 7000, 5875, }, |
- | |
5760 | { 7125, 6000, }, |
- | |
5761 | { 7250, 6125, }, |
- | |
5762 | { 7375, 6250, }, |
- | |
5763 | { 7500, 6375, }, |
- | |
5764 | { 7625, 6500, }, |
- | |
5765 | { 7750, 6625, }, |
- | |
5766 | { 7875, 6750, }, |
- | |
5767 | { 8000, 6875, }, |
- | |
5768 | { 8125, 7000, }, |
- | |
5769 | { 8250, 7125, }, |
- | |
5770 | { 8375, 7250, }, |
- | |
5771 | { 8500, 7375, }, |
- | |
5772 | { 8625, 7500, }, |
- | |
5773 | { 8750, 7625, }, |
- | |
5774 | { 8875, 7750, }, |
- | |
5775 | { 9000, 7875, }, |
- | |
5776 | { 9125, 8000, }, |
- | |
5777 | { 9250, 8125, }, |
- | |
5778 | { 9375, 8250, }, |
- | |
5779 | { 9500, 8375, }, |
- | |
5780 | { 9625, 8500, }, |
- | |
5781 | { 9750, 8625, }, |
- | |
5782 | { 9875, 8750, }, |
- | |
5783 | { 10000, 8875, }, |
- | |
5784 | { 10125, 9000, }, |
- | |
5785 | { 10250, 9125, }, |
- | |
5786 | { 10375, 9250, }, |
- | |
5787 | { 10500, 9375, }, |
- | |
5788 | { 10625, 9500, }, |
- | |
5789 | { 10750, 9625, }, |
- | |
5790 | { 10875, 9750, }, |
- | |
5791 | { 11000, 9875, }, |
- | |
5792 | { 11125, 10000, }, |
- | |
5793 | { 11250, 10125, }, |
- | |
5794 | { 11375, 10250, }, |
- | |
5795 | { 11500, 10375, }, |
- | |
5796 | { 11625, 10500, }, |
- | |
5797 | { 11750, 10625, }, |
- | |
5798 | { 11875, 10750, }, |
- | |
5799 | { 12000, 10875, }, |
- | |
5800 | { 12125, 11000, }, |
- | |
5801 | { 12250, 11125, }, |
- | |
5802 | { 12375, 11250, }, |
- | |
5803 | { 12500, 11375, }, |
- | |
5804 | { 12625, 11500, }, |
- | |
5805 | { 12750, 11625, }, |
- | |
5806 | { 12875, 11750, }, |
- | |
5807 | { 13000, 11875, }, |
- | |
5808 | { 13125, 12000, }, |
- | |
5809 | { 13250, 12125, }, |
- | |
5810 | { 13375, 12250, }, |
- | |
5811 | { 13500, 12375, }, |
- | |
5812 | { 13625, 12500, }, |
- | |
5813 | { 13750, 12625, }, |
- | |
5814 | { 13875, 12750, }, |
- | |
5815 | { 14000, 12875, }, |
- | |
5816 | { 14125, 13000, }, |
- | |
5817 | { 14250, 13125, }, |
- | |
5818 | { 14375, 13250, }, |
- | |
5819 | { 14500, 13375, }, |
- | |
5820 | { 14625, 13500, }, |
- | |
5821 | { 14750, 13625, }, |
- | |
5822 | { 14875, 13750, }, |
- | |
5823 | { 15000, 13875, }, |
- | |
5824 | { 15125, 14000, }, |
- | |
5825 | { 15250, 14125, }, |
- | |
5826 | { 15375, 14250, }, |
- | |
5827 | { 15500, 14375, }, |
- | |
5828 | { 15625, 14500, }, |
- | |
5829 | { 15750, 14625, }, |
- | |
5830 | { 15875, 14750, }, |
- | |
5831 | { 16000, 14875, }, |
- | |
5832 | { 16125, 15000, }, |
5756 | const int vm = vd - 1125; |
5833 | }; |
5757 | |
5834 | if (INTEL_INFO(dev)->is_mobile) |
5758 | if (INTEL_INFO(dev)->is_mobile) |
5835 | return v_table[pxvid].vm; |
5759 | return vm > 0 ? vm : 0; |
5836 | else |
5760 | |
5837 | return v_table[pxvid].vd; |
5761 | return vd; |
5838 | } |
5762 | } |
5839 | 5763 | ||
5840 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
5764 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
5841 | { |
5765 | { |
5842 | u64 now, diff, diffms; |
5766 | u64 now, diff, diffms; |
5843 | u32 count; |
5767 | u32 count; |
5844 | 5768 | ||
5845 | assert_spin_locked(&mchdev_lock); |
5769 | assert_spin_locked(&mchdev_lock); |
5846 | 5770 | ||
5847 | now = ktime_get_raw_ns(); |
5771 | now = ktime_get_raw_ns(); |
5848 | diffms = now - dev_priv->ips.last_time2; |
5772 | diffms = now - dev_priv->ips.last_time2; |
5849 | do_div(diffms, NSEC_PER_MSEC); |
5773 | do_div(diffms, NSEC_PER_MSEC); |
5850 | 5774 | ||
5851 | /* Don't divide by 0 */ |
5775 | /* Don't divide by 0 */ |
5852 | if (!diffms) |
5776 | if (!diffms) |
5853 | return; |
5777 | return; |
5854 | 5778 | ||
5855 | count = I915_READ(GFXEC); |
5779 | count = I915_READ(GFXEC); |
5856 | 5780 | ||
5857 | if (count < dev_priv->ips.last_count2) { |
5781 | if (count < dev_priv->ips.last_count2) { |
5858 | diff = ~0UL - dev_priv->ips.last_count2; |
5782 | diff = ~0UL - dev_priv->ips.last_count2; |
5859 | diff += count; |
5783 | diff += count; |
5860 | } else { |
5784 | } else { |
5861 | diff = count - dev_priv->ips.last_count2; |
5785 | diff = count - dev_priv->ips.last_count2; |
5862 | } |
5786 | } |
5863 | 5787 | ||
5864 | dev_priv->ips.last_count2 = count; |
5788 | dev_priv->ips.last_count2 = count; |
5865 | dev_priv->ips.last_time2 = now; |
5789 | dev_priv->ips.last_time2 = now; |
5866 | 5790 | ||
5867 | /* More magic constants... */ |
5791 | /* More magic constants... */ |
5868 | diff = diff * 1181; |
5792 | diff = diff * 1181; |
5869 | diff = div_u64(diff, diffms * 10); |
5793 | diff = div_u64(diff, diffms * 10); |
5870 | dev_priv->ips.gfx_power = diff; |
5794 | dev_priv->ips.gfx_power = diff; |
5871 | } |
5795 | } |
5872 | 5796 | ||
5873 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
5797 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
5874 | { |
5798 | { |
5875 | struct drm_device *dev = dev_priv->dev; |
5799 | struct drm_device *dev = dev_priv->dev; |
5876 | 5800 | ||
5877 | if (INTEL_INFO(dev)->gen != 5) |
5801 | if (INTEL_INFO(dev)->gen != 5) |
5878 | return; |
5802 | return; |
5879 | 5803 | ||
5880 | spin_lock_irq(&mchdev_lock); |
5804 | spin_lock_irq(&mchdev_lock); |
5881 | 5805 | ||
5882 | __i915_update_gfx_val(dev_priv); |
5806 | __i915_update_gfx_val(dev_priv); |
5883 | 5807 | ||
5884 | spin_unlock_irq(&mchdev_lock); |
5808 | spin_unlock_irq(&mchdev_lock); |
5885 | } |
5809 | } |
5886 | 5810 | ||
5887 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
5811 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
5888 | { |
5812 | { |
5889 | unsigned long t, corr, state1, corr2, state2; |
5813 | unsigned long t, corr, state1, corr2, state2; |
5890 | u32 pxvid, ext_v; |
5814 | u32 pxvid, ext_v; |
5891 | 5815 | ||
5892 | assert_spin_locked(&mchdev_lock); |
5816 | assert_spin_locked(&mchdev_lock); |
5893 | 5817 | ||
5894 | pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_freq * 4)); |
5818 | pxvid = I915_READ(PXVFREQ(dev_priv->rps.cur_freq)); |
5895 | pxvid = (pxvid >> 24) & 0x7f; |
5819 | pxvid = (pxvid >> 24) & 0x7f; |
5896 | ext_v = pvid_to_extvid(dev_priv, pxvid); |
5820 | ext_v = pvid_to_extvid(dev_priv, pxvid); |
5897 | 5821 | ||
5898 | state1 = ext_v; |
5822 | state1 = ext_v; |
5899 | 5823 | ||
5900 | t = i915_mch_val(dev_priv); |
5824 | t = i915_mch_val(dev_priv); |
5901 | 5825 | ||
5902 | /* Revel in the empirically derived constants */ |
5826 | /* Revel in the empirically derived constants */ |
5903 | 5827 | ||
5904 | /* Correction factor in 1/100000 units */ |
5828 | /* Correction factor in 1/100000 units */ |
5905 | if (t > 80) |
5829 | if (t > 80) |
5906 | corr = ((t * 2349) + 135940); |
5830 | corr = ((t * 2349) + 135940); |
5907 | else if (t >= 50) |
5831 | else if (t >= 50) |
5908 | corr = ((t * 964) + 29317); |
5832 | corr = ((t * 964) + 29317); |
5909 | else /* < 50 */ |
5833 | else /* < 50 */ |
5910 | corr = ((t * 301) + 1004); |
5834 | corr = ((t * 301) + 1004); |
5911 | 5835 | ||
5912 | corr = corr * ((150142 * state1) / 10000 - 78642); |
5836 | corr = corr * ((150142 * state1) / 10000 - 78642); |
5913 | corr /= 100000; |
5837 | corr /= 100000; |
5914 | corr2 = (corr * dev_priv->ips.corr); |
5838 | corr2 = (corr * dev_priv->ips.corr); |
5915 | 5839 | ||
5916 | state2 = (corr2 * state1) / 10000; |
5840 | state2 = (corr2 * state1) / 10000; |
5917 | state2 /= 100; /* convert to mW */ |
5841 | state2 /= 100; /* convert to mW */ |
5918 | 5842 | ||
5919 | __i915_update_gfx_val(dev_priv); |
5843 | __i915_update_gfx_val(dev_priv); |
5920 | 5844 | ||
5921 | return dev_priv->ips.gfx_power + state2; |
5845 | return dev_priv->ips.gfx_power + state2; |
5922 | } |
5846 | } |
5923 | 5847 | ||
5924 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
5848 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
5925 | { |
5849 | { |
5926 | struct drm_device *dev = dev_priv->dev; |
5850 | struct drm_device *dev = dev_priv->dev; |
5927 | unsigned long val; |
5851 | unsigned long val; |
5928 | 5852 | ||
5929 | if (INTEL_INFO(dev)->gen != 5) |
5853 | if (INTEL_INFO(dev)->gen != 5) |
5930 | return 0; |
5854 | return 0; |
5931 | 5855 | ||
5932 | spin_lock_irq(&mchdev_lock); |
5856 | spin_lock_irq(&mchdev_lock); |
5933 | 5857 | ||
5934 | val = __i915_gfx_val(dev_priv); |
5858 | val = __i915_gfx_val(dev_priv); |
5935 | 5859 | ||
5936 | spin_unlock_irq(&mchdev_lock); |
5860 | spin_unlock_irq(&mchdev_lock); |
5937 | 5861 | ||
5938 | return val; |
5862 | return val; |
5939 | } |
5863 | } |
5940 | 5864 | ||
5941 | /** |
5865 | /** |
5942 | * i915_read_mch_val - return value for IPS use |
5866 | * i915_read_mch_val - return value for IPS use |
5943 | * |
5867 | * |
5944 | * Calculate and return a value for the IPS driver to use when deciding whether |
5868 | * Calculate and return a value for the IPS driver to use when deciding whether |
5945 | * we have thermal and power headroom to increase CPU or GPU power budget. |
5869 | * we have thermal and power headroom to increase CPU or GPU power budget. |
5946 | */ |
5870 | */ |
5947 | unsigned long i915_read_mch_val(void) |
5871 | unsigned long i915_read_mch_val(void) |
5948 | { |
5872 | { |
5949 | struct drm_i915_private *dev_priv; |
5873 | struct drm_i915_private *dev_priv; |
5950 | unsigned long chipset_val, graphics_val, ret = 0; |
5874 | unsigned long chipset_val, graphics_val, ret = 0; |
5951 | 5875 | ||
5952 | spin_lock_irq(&mchdev_lock); |
5876 | spin_lock_irq(&mchdev_lock); |
5953 | if (!i915_mch_dev) |
5877 | if (!i915_mch_dev) |
5954 | goto out_unlock; |
5878 | goto out_unlock; |
5955 | dev_priv = i915_mch_dev; |
5879 | dev_priv = i915_mch_dev; |
5956 | 5880 | ||
5957 | chipset_val = __i915_chipset_val(dev_priv); |
5881 | chipset_val = __i915_chipset_val(dev_priv); |
5958 | graphics_val = __i915_gfx_val(dev_priv); |
5882 | graphics_val = __i915_gfx_val(dev_priv); |
5959 | 5883 | ||
5960 | ret = chipset_val + graphics_val; |
5884 | ret = chipset_val + graphics_val; |
5961 | 5885 | ||
5962 | out_unlock: |
5886 | out_unlock: |
5963 | spin_unlock_irq(&mchdev_lock); |
5887 | spin_unlock_irq(&mchdev_lock); |
5964 | 5888 | ||
5965 | return ret; |
5889 | return ret; |
5966 | } |
5890 | } |
5967 | EXPORT_SYMBOL_GPL(i915_read_mch_val); |
5891 | EXPORT_SYMBOL_GPL(i915_read_mch_val); |
5968 | 5892 | ||
5969 | /** |
5893 | /** |
5970 | * i915_gpu_raise - raise GPU frequency limit |
5894 | * i915_gpu_raise - raise GPU frequency limit |
5971 | * |
5895 | * |
5972 | * Raise the limit; IPS indicates we have thermal headroom. |
5896 | * Raise the limit; IPS indicates we have thermal headroom. |
5973 | */ |
5897 | */ |
5974 | bool i915_gpu_raise(void) |
5898 | bool i915_gpu_raise(void) |
5975 | { |
5899 | { |
5976 | struct drm_i915_private *dev_priv; |
5900 | struct drm_i915_private *dev_priv; |
5977 | bool ret = true; |
5901 | bool ret = true; |
5978 | 5902 | ||
5979 | spin_lock_irq(&mchdev_lock); |
5903 | spin_lock_irq(&mchdev_lock); |
5980 | if (!i915_mch_dev) { |
5904 | if (!i915_mch_dev) { |
5981 | ret = false; |
5905 | ret = false; |
5982 | goto out_unlock; |
5906 | goto out_unlock; |
5983 | } |
5907 | } |
5984 | dev_priv = i915_mch_dev; |
5908 | dev_priv = i915_mch_dev; |
5985 | 5909 | ||
5986 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
5910 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
5987 | dev_priv->ips.max_delay--; |
5911 | dev_priv->ips.max_delay--; |
5988 | 5912 | ||
5989 | out_unlock: |
5913 | out_unlock: |
5990 | spin_unlock_irq(&mchdev_lock); |
5914 | spin_unlock_irq(&mchdev_lock); |
5991 | 5915 | ||
5992 | return ret; |
5916 | return ret; |
5993 | } |
5917 | } |
5994 | EXPORT_SYMBOL_GPL(i915_gpu_raise); |
5918 | EXPORT_SYMBOL_GPL(i915_gpu_raise); |
5995 | 5919 | ||
5996 | /** |
5920 | /** |
5997 | * i915_gpu_lower - lower GPU frequency limit |
5921 | * i915_gpu_lower - lower GPU frequency limit |
5998 | * |
5922 | * |
5999 | * IPS indicates we're close to a thermal limit, so throttle back the GPU |
5923 | * IPS indicates we're close to a thermal limit, so throttle back the GPU |
6000 | * frequency maximum. |
5924 | * frequency maximum. |
6001 | */ |
5925 | */ |
6002 | bool i915_gpu_lower(void) |
5926 | bool i915_gpu_lower(void) |
6003 | { |
5927 | { |
6004 | struct drm_i915_private *dev_priv; |
5928 | struct drm_i915_private *dev_priv; |
6005 | bool ret = true; |
5929 | bool ret = true; |
6006 | 5930 | ||
6007 | spin_lock_irq(&mchdev_lock); |
5931 | spin_lock_irq(&mchdev_lock); |
6008 | if (!i915_mch_dev) { |
5932 | if (!i915_mch_dev) { |
6009 | ret = false; |
5933 | ret = false; |
6010 | goto out_unlock; |
5934 | goto out_unlock; |
6011 | } |
5935 | } |
6012 | dev_priv = i915_mch_dev; |
5936 | dev_priv = i915_mch_dev; |
6013 | 5937 | ||
6014 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
5938 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
6015 | dev_priv->ips.max_delay++; |
5939 | dev_priv->ips.max_delay++; |
6016 | 5940 | ||
6017 | out_unlock: |
5941 | out_unlock: |
6018 | spin_unlock_irq(&mchdev_lock); |
5942 | spin_unlock_irq(&mchdev_lock); |
6019 | 5943 | ||
6020 | return ret; |
5944 | return ret; |
6021 | } |
5945 | } |
6022 | EXPORT_SYMBOL_GPL(i915_gpu_lower); |
5946 | EXPORT_SYMBOL_GPL(i915_gpu_lower); |
6023 | 5947 | ||
6024 | /** |
5948 | /** |
6025 | * i915_gpu_busy - indicate GPU business to IPS |
5949 | * i915_gpu_busy - indicate GPU business to IPS |
6026 | * |
5950 | * |
6027 | * Tell the IPS driver whether or not the GPU is busy. |
5951 | * Tell the IPS driver whether or not the GPU is busy. |
6028 | */ |
5952 | */ |
6029 | bool i915_gpu_busy(void) |
5953 | bool i915_gpu_busy(void) |
6030 | { |
5954 | { |
6031 | struct drm_i915_private *dev_priv; |
5955 | struct drm_i915_private *dev_priv; |
6032 | struct intel_engine_cs *ring; |
5956 | struct intel_engine_cs *ring; |
6033 | bool ret = false; |
5957 | bool ret = false; |
6034 | int i; |
5958 | int i; |
6035 | 5959 | ||
6036 | spin_lock_irq(&mchdev_lock); |
5960 | spin_lock_irq(&mchdev_lock); |
6037 | if (!i915_mch_dev) |
5961 | if (!i915_mch_dev) |
6038 | goto out_unlock; |
5962 | goto out_unlock; |
6039 | dev_priv = i915_mch_dev; |
5963 | dev_priv = i915_mch_dev; |
6040 | 5964 | ||
6041 | for_each_ring(ring, dev_priv, i) |
5965 | for_each_ring(ring, dev_priv, i) |
6042 | ret |= !list_empty(&ring->request_list); |
5966 | ret |= !list_empty(&ring->request_list); |
6043 | 5967 | ||
6044 | out_unlock: |
5968 | out_unlock: |
6045 | spin_unlock_irq(&mchdev_lock); |
5969 | spin_unlock_irq(&mchdev_lock); |
6046 | 5970 | ||
6047 | return ret; |
5971 | return ret; |
6048 | } |
5972 | } |
6049 | EXPORT_SYMBOL_GPL(i915_gpu_busy); |
5973 | EXPORT_SYMBOL_GPL(i915_gpu_busy); |
6050 | 5974 | ||
6051 | /** |
5975 | /** |
6052 | * i915_gpu_turbo_disable - disable graphics turbo |
5976 | * i915_gpu_turbo_disable - disable graphics turbo |
6053 | * |
5977 | * |
6054 | * Disable graphics turbo by resetting the max frequency and setting the |
5978 | * Disable graphics turbo by resetting the max frequency and setting the |
6055 | * current frequency to the default. |
5979 | * current frequency to the default. |
6056 | */ |
5980 | */ |
6057 | bool i915_gpu_turbo_disable(void) |
5981 | bool i915_gpu_turbo_disable(void) |
6058 | { |
5982 | { |
6059 | struct drm_i915_private *dev_priv; |
5983 | struct drm_i915_private *dev_priv; |
6060 | bool ret = true; |
5984 | bool ret = true; |
6061 | 5985 | ||
6062 | spin_lock_irq(&mchdev_lock); |
5986 | spin_lock_irq(&mchdev_lock); |
6063 | if (!i915_mch_dev) { |
5987 | if (!i915_mch_dev) { |
6064 | ret = false; |
5988 | ret = false; |
6065 | goto out_unlock; |
5989 | goto out_unlock; |
6066 | } |
5990 | } |
6067 | dev_priv = i915_mch_dev; |
5991 | dev_priv = i915_mch_dev; |
6068 | 5992 | ||
6069 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
5993 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
6070 | 5994 | ||
6071 | if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart)) |
5995 | if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart)) |
6072 | ret = false; |
5996 | ret = false; |
6073 | 5997 | ||
6074 | out_unlock: |
5998 | out_unlock: |
6075 | spin_unlock_irq(&mchdev_lock); |
5999 | spin_unlock_irq(&mchdev_lock); |
6076 | 6000 | ||
6077 | return ret; |
6001 | return ret; |
6078 | } |
6002 | } |
6079 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); |
6003 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); |
6080 | 6004 | ||
6081 | /** |
6005 | /** |
6082 | * Tells the intel_ips driver that the i915 driver is now loaded, if |
6006 | * Tells the intel_ips driver that the i915 driver is now loaded, if |
6083 | * IPS got loaded first. |
6007 | * IPS got loaded first. |
6084 | * |
6008 | * |
6085 | * This awkward dance is so that neither module has to depend on the |
6009 | * This awkward dance is so that neither module has to depend on the |
6086 | * other in order for IPS to do the appropriate communication of |
6010 | * other in order for IPS to do the appropriate communication of |
6087 | * GPU turbo limits to i915. |
6011 | * GPU turbo limits to i915. |
6088 | */ |
6012 | */ |
6089 | static void |
6013 | static void |
6090 | ips_ping_for_i915_load(void) |
6014 | ips_ping_for_i915_load(void) |
6091 | { |
6015 | { |
6092 | void (*link)(void); |
6016 | void (*link)(void); |
6093 | 6017 | ||
6094 | // link = symbol_get(ips_link_to_i915_driver); |
6018 | // link = symbol_get(ips_link_to_i915_driver); |
6095 | // if (link) { |
6019 | // if (link) { |
6096 | // link(); |
6020 | // link(); |
6097 | // symbol_put(ips_link_to_i915_driver); |
6021 | // symbol_put(ips_link_to_i915_driver); |
6098 | // } |
6022 | // } |
6099 | } |
6023 | } |
6100 | 6024 | ||
6101 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) |
6025 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) |
6102 | { |
6026 | { |
6103 | /* We only register the i915 ips part with intel-ips once everything is |
6027 | /* We only register the i915 ips part with intel-ips once everything is |
6104 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ |
6028 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ |
6105 | spin_lock_irq(&mchdev_lock); |
6029 | spin_lock_irq(&mchdev_lock); |
6106 | i915_mch_dev = dev_priv; |
6030 | i915_mch_dev = dev_priv; |
6107 | spin_unlock_irq(&mchdev_lock); |
6031 | spin_unlock_irq(&mchdev_lock); |
6108 | 6032 | ||
6109 | ips_ping_for_i915_load(); |
6033 | ips_ping_for_i915_load(); |
6110 | } |
6034 | } |
6111 | 6035 | ||
6112 | void intel_gpu_ips_teardown(void) |
6036 | void intel_gpu_ips_teardown(void) |
6113 | { |
6037 | { |
6114 | spin_lock_irq(&mchdev_lock); |
6038 | spin_lock_irq(&mchdev_lock); |
6115 | i915_mch_dev = NULL; |
6039 | i915_mch_dev = NULL; |
6116 | spin_unlock_irq(&mchdev_lock); |
6040 | spin_unlock_irq(&mchdev_lock); |
6117 | } |
6041 | } |
6118 | 6042 | ||
6119 | static void intel_init_emon(struct drm_device *dev) |
6043 | static void intel_init_emon(struct drm_device *dev) |
6120 | { |
6044 | { |
6121 | struct drm_i915_private *dev_priv = dev->dev_private; |
6045 | struct drm_i915_private *dev_priv = dev->dev_private; |
6122 | u32 lcfuse; |
6046 | u32 lcfuse; |
6123 | u8 pxw[16]; |
6047 | u8 pxw[16]; |
6124 | int i; |
6048 | int i; |
6125 | 6049 | ||
6126 | /* Disable to program */ |
6050 | /* Disable to program */ |
6127 | I915_WRITE(ECR, 0); |
6051 | I915_WRITE(ECR, 0); |
6128 | POSTING_READ(ECR); |
6052 | POSTING_READ(ECR); |
6129 | 6053 | ||
6130 | /* Program energy weights for various events */ |
6054 | /* Program energy weights for various events */ |
6131 | I915_WRITE(SDEW, 0x15040d00); |
6055 | I915_WRITE(SDEW, 0x15040d00); |
6132 | I915_WRITE(CSIEW0, 0x007f0000); |
6056 | I915_WRITE(CSIEW0, 0x007f0000); |
6133 | I915_WRITE(CSIEW1, 0x1e220004); |
6057 | I915_WRITE(CSIEW1, 0x1e220004); |
6134 | I915_WRITE(CSIEW2, 0x04000004); |
6058 | I915_WRITE(CSIEW2, 0x04000004); |
6135 | 6059 | ||
6136 | for (i = 0; i < 5; i++) |
6060 | for (i = 0; i < 5; i++) |
6137 | I915_WRITE(PEW + (i * 4), 0); |
6061 | I915_WRITE(PEW(i), 0); |
6138 | for (i = 0; i < 3; i++) |
6062 | for (i = 0; i < 3; i++) |
6139 | I915_WRITE(DEW + (i * 4), 0); |
6063 | I915_WRITE(DEW(i), 0); |
6140 | 6064 | ||
6141 | /* Program P-state weights to account for frequency power adjustment */ |
6065 | /* Program P-state weights to account for frequency power adjustment */ |
6142 | for (i = 0; i < 16; i++) { |
6066 | for (i = 0; i < 16; i++) { |
6143 | u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4)); |
6067 | u32 pxvidfreq = I915_READ(PXVFREQ(i)); |
6144 | unsigned long freq = intel_pxfreq(pxvidfreq); |
6068 | unsigned long freq = intel_pxfreq(pxvidfreq); |
6145 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> |
6069 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> |
6146 | PXVFREQ_PX_SHIFT; |
6070 | PXVFREQ_PX_SHIFT; |
6147 | unsigned long val; |
6071 | unsigned long val; |
6148 | 6072 | ||
6149 | val = vid * vid; |
6073 | val = vid * vid; |
6150 | val *= (freq / 1000); |
6074 | val *= (freq / 1000); |
6151 | val *= 255; |
6075 | val *= 255; |
6152 | val /= (127*127*900); |
6076 | val /= (127*127*900); |
6153 | if (val > 0xff) |
6077 | if (val > 0xff) |
6154 | DRM_ERROR("bad pxval: %ld\n", val); |
6078 | DRM_ERROR("bad pxval: %ld\n", val); |
6155 | pxw[i] = val; |
6079 | pxw[i] = val; |
6156 | } |
6080 | } |
6157 | /* Render standby states get 0 weight */ |
6081 | /* Render standby states get 0 weight */ |
6158 | pxw[14] = 0; |
6082 | pxw[14] = 0; |
6159 | pxw[15] = 0; |
6083 | pxw[15] = 0; |
6160 | 6084 | ||
6161 | for (i = 0; i < 4; i++) { |
6085 | for (i = 0; i < 4; i++) { |
6162 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | |
6086 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | |
6163 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); |
6087 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); |
6164 | I915_WRITE(PXW + (i * 4), val); |
6088 | I915_WRITE(PXW(i), val); |
6165 | } |
6089 | } |
6166 | 6090 | ||
6167 | /* Adjust magic regs to magic values (more experimental results) */ |
6091 | /* Adjust magic regs to magic values (more experimental results) */ |
6168 | I915_WRITE(OGW0, 0); |
6092 | I915_WRITE(OGW0, 0); |
6169 | I915_WRITE(OGW1, 0); |
6093 | I915_WRITE(OGW1, 0); |
6170 | I915_WRITE(EG0, 0x00007f00); |
6094 | I915_WRITE(EG0, 0x00007f00); |
6171 | I915_WRITE(EG1, 0x0000000e); |
6095 | I915_WRITE(EG1, 0x0000000e); |
6172 | I915_WRITE(EG2, 0x000e0000); |
6096 | I915_WRITE(EG2, 0x000e0000); |
6173 | I915_WRITE(EG3, 0x68000300); |
6097 | I915_WRITE(EG3, 0x68000300); |
6174 | I915_WRITE(EG4, 0x42000000); |
6098 | I915_WRITE(EG4, 0x42000000); |
6175 | I915_WRITE(EG5, 0x00140031); |
6099 | I915_WRITE(EG5, 0x00140031); |
6176 | I915_WRITE(EG6, 0); |
6100 | I915_WRITE(EG6, 0); |
6177 | I915_WRITE(EG7, 0); |
6101 | I915_WRITE(EG7, 0); |
6178 | 6102 | ||
6179 | for (i = 0; i < 8; i++) |
6103 | for (i = 0; i < 8; i++) |
6180 | I915_WRITE(PXWL + (i * 4), 0); |
6104 | I915_WRITE(PXWL(i), 0); |
6181 | 6105 | ||
6182 | /* Enable PMON + select events */ |
6106 | /* Enable PMON + select events */ |
6183 | I915_WRITE(ECR, 0x80000019); |
6107 | I915_WRITE(ECR, 0x80000019); |
6184 | 6108 | ||
6185 | lcfuse = I915_READ(LCFUSE02); |
6109 | lcfuse = I915_READ(LCFUSE02); |
6186 | 6110 | ||
6187 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
6111 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
6188 | } |
6112 | } |
6189 | 6113 | ||
6190 | void intel_init_gt_powersave(struct drm_device *dev) |
6114 | void intel_init_gt_powersave(struct drm_device *dev) |
6191 | { |
6115 | { |
6192 | i915.enable_rc6 = sanitize_rc6_option(dev, i915.enable_rc6); |
6116 | i915.enable_rc6 = sanitize_rc6_option(dev, i915.enable_rc6); |
6193 | 6117 | ||
6194 | if (IS_CHERRYVIEW(dev)) |
6118 | if (IS_CHERRYVIEW(dev)) |
6195 | cherryview_init_gt_powersave(dev); |
6119 | cherryview_init_gt_powersave(dev); |
6196 | else if (IS_VALLEYVIEW(dev)) |
6120 | else if (IS_VALLEYVIEW(dev)) |
6197 | valleyview_init_gt_powersave(dev); |
6121 | valleyview_init_gt_powersave(dev); |
6198 | } |
6122 | } |
6199 | 6123 | ||
6200 | void intel_cleanup_gt_powersave(struct drm_device *dev) |
6124 | void intel_cleanup_gt_powersave(struct drm_device *dev) |
6201 | { |
6125 | { |
6202 | if (IS_CHERRYVIEW(dev)) |
6126 | if (IS_CHERRYVIEW(dev)) |
6203 | return; |
6127 | return; |
6204 | else if (IS_VALLEYVIEW(dev)) |
6128 | else if (IS_VALLEYVIEW(dev)) |
6205 | valleyview_cleanup_gt_powersave(dev); |
6129 | valleyview_cleanup_gt_powersave(dev); |
6206 | } |
6130 | } |
6207 | 6131 | ||
6208 | static void gen6_suspend_rps(struct drm_device *dev) |
6132 | static void gen6_suspend_rps(struct drm_device *dev) |
6209 | { |
6133 | { |
6210 | struct drm_i915_private *dev_priv = dev->dev_private; |
6134 | struct drm_i915_private *dev_priv = dev->dev_private; |
6211 | 6135 | ||
6212 | // flush_delayed_work(&dev_priv->rps.delayed_resume_work); |
6136 | // flush_delayed_work(&dev_priv->rps.delayed_resume_work); |
6213 | - | ||
6214 | /* |
- | |
6215 | * TODO: disable RPS interrupts on GEN9+ too once RPS support |
- | |
6216 | * is added for it. |
- | |
6217 | */ |
- | |
6218 | if (INTEL_INFO(dev)->gen < 9) |
6137 | |
6219 | gen6_disable_rps_interrupts(dev); |
6138 | gen6_disable_rps_interrupts(dev); |
6220 | } |
6139 | } |
6221 | 6140 | ||
6222 | /** |
6141 | /** |
6223 | * intel_suspend_gt_powersave - suspend PM work and helper threads |
6142 | * intel_suspend_gt_powersave - suspend PM work and helper threads |
6224 | * @dev: drm device |
6143 | * @dev: drm device |
6225 | * |
6144 | * |
6226 | * We don't want to disable RC6 or other features here, we just want |
6145 | * We don't want to disable RC6 or other features here, we just want |
6227 | * to make sure any work we've queued has finished and won't bother |
6146 | * to make sure any work we've queued has finished and won't bother |
6228 | * us while we're suspended. |
6147 | * us while we're suspended. |
6229 | */ |
6148 | */ |
6230 | void intel_suspend_gt_powersave(struct drm_device *dev) |
6149 | void intel_suspend_gt_powersave(struct drm_device *dev) |
6231 | { |
6150 | { |
6232 | struct drm_i915_private *dev_priv = dev->dev_private; |
6151 | struct drm_i915_private *dev_priv = dev->dev_private; |
6233 | 6152 | ||
6234 | if (INTEL_INFO(dev)->gen < 6) |
6153 | if (INTEL_INFO(dev)->gen < 6) |
6235 | return; |
6154 | return; |
6236 | 6155 | ||
6237 | gen6_suspend_rps(dev); |
6156 | gen6_suspend_rps(dev); |
6238 | 6157 | ||
6239 | /* Force GPU to min freq during suspend */ |
6158 | /* Force GPU to min freq during suspend */ |
6240 | gen6_rps_idle(dev_priv); |
6159 | gen6_rps_idle(dev_priv); |
6241 | } |
6160 | } |
6242 | 6161 | ||
6243 | void intel_disable_gt_powersave(struct drm_device *dev) |
6162 | void intel_disable_gt_powersave(struct drm_device *dev) |
6244 | { |
6163 | { |
6245 | struct drm_i915_private *dev_priv = dev->dev_private; |
6164 | struct drm_i915_private *dev_priv = dev->dev_private; |
6246 | 6165 | ||
6247 | if (IS_IRONLAKE_M(dev)) { |
6166 | if (IS_IRONLAKE_M(dev)) { |
6248 | ironlake_disable_drps(dev); |
6167 | ironlake_disable_drps(dev); |
6249 | ironlake_disable_rc6(dev); |
- | |
6250 | } else if (INTEL_INFO(dev)->gen >= 6) { |
6168 | } else if (INTEL_INFO(dev)->gen >= 6) { |
6251 | intel_suspend_gt_powersave(dev); |
6169 | intel_suspend_gt_powersave(dev); |
6252 | 6170 | ||
6253 | mutex_lock(&dev_priv->rps.hw_lock); |
6171 | mutex_lock(&dev_priv->rps.hw_lock); |
6254 | if (INTEL_INFO(dev)->gen >= 9) |
6172 | if (INTEL_INFO(dev)->gen >= 9) |
6255 | gen9_disable_rps(dev); |
6173 | gen9_disable_rps(dev); |
6256 | else if (IS_CHERRYVIEW(dev)) |
6174 | else if (IS_CHERRYVIEW(dev)) |
6257 | cherryview_disable_rps(dev); |
6175 | cherryview_disable_rps(dev); |
6258 | else if (IS_VALLEYVIEW(dev)) |
6176 | else if (IS_VALLEYVIEW(dev)) |
6259 | valleyview_disable_rps(dev); |
6177 | valleyview_disable_rps(dev); |
6260 | else |
6178 | else |
6261 | gen6_disable_rps(dev); |
6179 | gen6_disable_rps(dev); |
6262 | 6180 | ||
6263 | dev_priv->rps.enabled = false; |
6181 | dev_priv->rps.enabled = false; |
6264 | mutex_unlock(&dev_priv->rps.hw_lock); |
6182 | mutex_unlock(&dev_priv->rps.hw_lock); |
6265 | } |
6183 | } |
6266 | } |
6184 | } |
6267 | 6185 | ||
6268 | static void intel_gen6_powersave_work(struct work_struct *work) |
6186 | static void intel_gen6_powersave_work(struct work_struct *work) |
6269 | { |
6187 | { |
6270 | struct drm_i915_private *dev_priv = |
6188 | struct drm_i915_private *dev_priv = |
6271 | container_of(work, struct drm_i915_private, |
6189 | container_of(work, struct drm_i915_private, |
6272 | rps.delayed_resume_work.work); |
6190 | rps.delayed_resume_work.work); |
6273 | struct drm_device *dev = dev_priv->dev; |
6191 | struct drm_device *dev = dev_priv->dev; |
6274 | 6192 | ||
6275 | mutex_lock(&dev_priv->rps.hw_lock); |
6193 | mutex_lock(&dev_priv->rps.hw_lock); |
6276 | - | ||
6277 | /* |
- | |
6278 | * TODO: reset/enable RPS interrupts on GEN9+ too, once RPS support is |
- | |
6279 | * added for it. |
- | |
6280 | */ |
- | |
6281 | if (INTEL_INFO(dev)->gen < 9) |
6194 | |
6282 | gen6_reset_rps_interrupts(dev); |
6195 | gen6_reset_rps_interrupts(dev); |
6283 | 6196 | ||
6284 | if (IS_CHERRYVIEW(dev)) { |
6197 | if (IS_CHERRYVIEW(dev)) { |
6285 | cherryview_enable_rps(dev); |
6198 | cherryview_enable_rps(dev); |
6286 | } else if (IS_VALLEYVIEW(dev)) { |
6199 | } else if (IS_VALLEYVIEW(dev)) { |
6287 | valleyview_enable_rps(dev); |
6200 | valleyview_enable_rps(dev); |
6288 | } else if (INTEL_INFO(dev)->gen >= 9) { |
6201 | } else if (INTEL_INFO(dev)->gen >= 9) { |
- | 6202 | gen9_enable_rc6(dev); |
|
6289 | gen9_enable_rps(dev); |
6203 | gen9_enable_rps(dev); |
- | 6204 | if (IS_SKYLAKE(dev)) |
|
- | 6205 | __gen6_update_ring_freq(dev); |
|
6290 | } else if (IS_BROADWELL(dev)) { |
6206 | } else if (IS_BROADWELL(dev)) { |
6291 | gen8_enable_rps(dev); |
6207 | gen8_enable_rps(dev); |
6292 | __gen6_update_ring_freq(dev); |
6208 | __gen6_update_ring_freq(dev); |
6293 | } else { |
6209 | } else { |
6294 | gen6_enable_rps(dev); |
6210 | gen6_enable_rps(dev); |
6295 | __gen6_update_ring_freq(dev); |
6211 | __gen6_update_ring_freq(dev); |
6296 | } |
6212 | } |
- | 6213 | ||
- | 6214 | WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq); |
|
- | 6215 | WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq); |
|
- | 6216 | ||
- | 6217 | WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq); |
|
- | 6218 | WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq); |
|
- | 6219 | ||
6297 | dev_priv->rps.enabled = true; |
6220 | dev_priv->rps.enabled = true; |
6298 | - | ||
6299 | if (INTEL_INFO(dev)->gen < 9) |
6221 | |
6300 | gen6_enable_rps_interrupts(dev); |
6222 | gen6_enable_rps_interrupts(dev); |
6301 | 6223 | ||
6302 | mutex_unlock(&dev_priv->rps.hw_lock); |
6224 | mutex_unlock(&dev_priv->rps.hw_lock); |
6303 | 6225 | ||
6304 | intel_runtime_pm_put(dev_priv); |
6226 | intel_runtime_pm_put(dev_priv); |
6305 | } |
6227 | } |
6306 | 6228 | ||
6307 | void intel_enable_gt_powersave(struct drm_device *dev) |
6229 | void intel_enable_gt_powersave(struct drm_device *dev) |
6308 | { |
6230 | { |
6309 | struct drm_i915_private *dev_priv = dev->dev_private; |
6231 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | 6232 | ||
- | 6233 | /* Powersaving is controlled by the host when inside a VM */ |
|
- | 6234 | if (intel_vgpu_active(dev)) |
|
- | 6235 | return; |
|
6310 | 6236 | ||
6311 | if (IS_IRONLAKE_M(dev)) { |
6237 | if (IS_IRONLAKE_M(dev)) { |
6312 | mutex_lock(&dev->struct_mutex); |
6238 | mutex_lock(&dev->struct_mutex); |
6313 | ironlake_enable_drps(dev); |
6239 | ironlake_enable_drps(dev); |
6314 | ironlake_enable_rc6(dev); |
- | |
6315 | intel_init_emon(dev); |
6240 | intel_init_emon(dev); |
6316 | mutex_unlock(&dev->struct_mutex); |
6241 | mutex_unlock(&dev->struct_mutex); |
6317 | } else if (INTEL_INFO(dev)->gen >= 6) { |
6242 | } else if (INTEL_INFO(dev)->gen >= 6) { |
6318 | /* |
6243 | /* |
6319 | * PCU communication is slow and this doesn't need to be |
6244 | * PCU communication is slow and this doesn't need to be |
6320 | * done at any specific time, so do this out of our fast path |
6245 | * done at any specific time, so do this out of our fast path |
6321 | * to make resume and init faster. |
6246 | * to make resume and init faster. |
6322 | * |
6247 | * |
6323 | * We depend on the HW RC6 power context save/restore |
6248 | * We depend on the HW RC6 power context save/restore |
6324 | * mechanism when entering D3 through runtime PM suspend. So |
6249 | * mechanism when entering D3 through runtime PM suspend. So |
6325 | * disable RPM until RPS/RC6 is properly setup. We can only |
6250 | * disable RPM until RPS/RC6 is properly setup. We can only |
6326 | * get here via the driver load/system resume/runtime resume |
6251 | * get here via the driver load/system resume/runtime resume |
6327 | * paths, so the _noresume version is enough (and in case of |
6252 | * paths, so the _noresume version is enough (and in case of |
6328 | * runtime resume it's necessary). |
6253 | * runtime resume it's necessary). |
6329 | */ |
6254 | */ |
6330 | if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work, |
6255 | if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work, |
6331 | round_jiffies_up_relative(HZ))) |
6256 | round_jiffies_up_relative(HZ))) |
6332 | intel_runtime_pm_get_noresume(dev_priv); |
6257 | intel_runtime_pm_get_noresume(dev_priv); |
6333 | } |
6258 | } |
6334 | } |
6259 | } |
6335 | 6260 | ||
6336 | void intel_reset_gt_powersave(struct drm_device *dev) |
6261 | void intel_reset_gt_powersave(struct drm_device *dev) |
6337 | { |
6262 | { |
6338 | struct drm_i915_private *dev_priv = dev->dev_private; |
6263 | struct drm_i915_private *dev_priv = dev->dev_private; |
6339 | 6264 | ||
6340 | if (INTEL_INFO(dev)->gen < 6) |
6265 | if (INTEL_INFO(dev)->gen < 6) |
6341 | return; |
6266 | return; |
6342 | 6267 | ||
6343 | gen6_suspend_rps(dev); |
6268 | gen6_suspend_rps(dev); |
6344 | dev_priv->rps.enabled = false; |
6269 | dev_priv->rps.enabled = false; |
6345 | } |
6270 | } |
6346 | 6271 | ||
6347 | static void ibx_init_clock_gating(struct drm_device *dev) |
6272 | static void ibx_init_clock_gating(struct drm_device *dev) |
6348 | { |
6273 | { |
6349 | struct drm_i915_private *dev_priv = dev->dev_private; |
6274 | struct drm_i915_private *dev_priv = dev->dev_private; |
6350 | 6275 | ||
6351 | /* |
6276 | /* |
6352 | * On Ibex Peak and Cougar Point, we need to disable clock |
6277 | * On Ibex Peak and Cougar Point, we need to disable clock |
6353 | * gating for the panel power sequencer or it will fail to |
6278 | * gating for the panel power sequencer or it will fail to |
6354 | * start up when no ports are active. |
6279 | * start up when no ports are active. |
6355 | */ |
6280 | */ |
6356 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); |
6281 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); |
6357 | } |
6282 | } |
6358 | 6283 | ||
6359 | static void g4x_disable_trickle_feed(struct drm_device *dev) |
6284 | static void g4x_disable_trickle_feed(struct drm_device *dev) |
6360 | { |
6285 | { |
6361 | struct drm_i915_private *dev_priv = dev->dev_private; |
6286 | struct drm_i915_private *dev_priv = dev->dev_private; |
6362 | int pipe; |
6287 | enum pipe pipe; |
6363 | 6288 | ||
6364 | for_each_pipe(dev_priv, pipe) { |
6289 | for_each_pipe(dev_priv, pipe) { |
6365 | I915_WRITE(DSPCNTR(pipe), |
6290 | I915_WRITE(DSPCNTR(pipe), |
6366 | I915_READ(DSPCNTR(pipe)) | |
6291 | I915_READ(DSPCNTR(pipe)) | |
6367 | DISPPLANE_TRICKLE_FEED_DISABLE); |
6292 | DISPPLANE_TRICKLE_FEED_DISABLE); |
- | 6293 | ||
6368 | intel_flush_primary_plane(dev_priv, pipe); |
6294 | I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe))); |
- | 6295 | POSTING_READ(DSPSURF(pipe)); |
|
6369 | } |
6296 | } |
6370 | } |
6297 | } |
6371 | 6298 | ||
6372 | static void ilk_init_lp_watermarks(struct drm_device *dev) |
6299 | static void ilk_init_lp_watermarks(struct drm_device *dev) |
6373 | { |
6300 | { |
6374 | struct drm_i915_private *dev_priv = dev->dev_private; |
6301 | struct drm_i915_private *dev_priv = dev->dev_private; |
6375 | 6302 | ||
6376 | I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN); |
6303 | I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN); |
6377 | I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN); |
6304 | I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN); |
6378 | I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN); |
6305 | I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN); |
6379 | 6306 | ||
6380 | /* |
6307 | /* |
6381 | * Don't touch WM1S_LP_EN here. |
6308 | * Don't touch WM1S_LP_EN here. |
6382 | * Doing so could cause underruns. |
6309 | * Doing so could cause underruns. |
6383 | */ |
6310 | */ |
6384 | } |
6311 | } |
6385 | 6312 | ||
6386 | static void ironlake_init_clock_gating(struct drm_device *dev) |
6313 | static void ironlake_init_clock_gating(struct drm_device *dev) |
6387 | { |
6314 | { |
6388 | struct drm_i915_private *dev_priv = dev->dev_private; |
6315 | struct drm_i915_private *dev_priv = dev->dev_private; |
6389 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6316 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6390 | 6317 | ||
6391 | /* |
6318 | /* |
6392 | * Required for FBC |
6319 | * Required for FBC |
6393 | * WaFbcDisableDpfcClockGating:ilk |
6320 | * WaFbcDisableDpfcClockGating:ilk |
6394 | */ |
6321 | */ |
6395 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
6322 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
6396 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | |
6323 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | |
6397 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; |
6324 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; |
6398 | 6325 | ||
6399 | I915_WRITE(PCH_3DCGDIS0, |
6326 | I915_WRITE(PCH_3DCGDIS0, |
6400 | MARIUNIT_CLOCK_GATE_DISABLE | |
6327 | MARIUNIT_CLOCK_GATE_DISABLE | |
6401 | SVSMUNIT_CLOCK_GATE_DISABLE); |
6328 | SVSMUNIT_CLOCK_GATE_DISABLE); |
6402 | I915_WRITE(PCH_3DCGDIS1, |
6329 | I915_WRITE(PCH_3DCGDIS1, |
6403 | VFMUNIT_CLOCK_GATE_DISABLE); |
6330 | VFMUNIT_CLOCK_GATE_DISABLE); |
6404 | 6331 | ||
6405 | /* |
6332 | /* |
6406 | * According to the spec the following bits should be set in |
6333 | * According to the spec the following bits should be set in |
6407 | * order to enable memory self-refresh |
6334 | * order to enable memory self-refresh |
6408 | * The bit 22/21 of 0x42004 |
6335 | * The bit 22/21 of 0x42004 |
6409 | * The bit 5 of 0x42020 |
6336 | * The bit 5 of 0x42020 |
6410 | * The bit 15 of 0x45000 |
6337 | * The bit 15 of 0x45000 |
6411 | */ |
6338 | */ |
6412 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6339 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6413 | (I915_READ(ILK_DISPLAY_CHICKEN2) | |
6340 | (I915_READ(ILK_DISPLAY_CHICKEN2) | |
6414 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); |
6341 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); |
6415 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
6342 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
6416 | I915_WRITE(DISP_ARB_CTL, |
6343 | I915_WRITE(DISP_ARB_CTL, |
6417 | (I915_READ(DISP_ARB_CTL) | |
6344 | (I915_READ(DISP_ARB_CTL) | |
6418 | DISP_FBC_WM_DIS)); |
6345 | DISP_FBC_WM_DIS)); |
6419 | 6346 | ||
6420 | ilk_init_lp_watermarks(dev); |
6347 | ilk_init_lp_watermarks(dev); |
6421 | 6348 | ||
6422 | /* |
6349 | /* |
6423 | * Based on the document from hardware guys the following bits |
6350 | * Based on the document from hardware guys the following bits |
6424 | * should be set unconditionally in order to enable FBC. |
6351 | * should be set unconditionally in order to enable FBC. |
6425 | * The bit 22 of 0x42000 |
6352 | * The bit 22 of 0x42000 |
6426 | * The bit 22 of 0x42004 |
6353 | * The bit 22 of 0x42004 |
6427 | * The bit 7,8,9 of 0x42020. |
6354 | * The bit 7,8,9 of 0x42020. |
6428 | */ |
6355 | */ |
6429 | if (IS_IRONLAKE_M(dev)) { |
6356 | if (IS_IRONLAKE_M(dev)) { |
6430 | /* WaFbcAsynchFlipDisableFbcQueue:ilk */ |
6357 | /* WaFbcAsynchFlipDisableFbcQueue:ilk */ |
6431 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
6358 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
6432 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
6359 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
6433 | ILK_FBCQ_DIS); |
6360 | ILK_FBCQ_DIS); |
6434 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6361 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6435 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6362 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6436 | ILK_DPARB_GATE); |
6363 | ILK_DPARB_GATE); |
6437 | } |
6364 | } |
6438 | 6365 | ||
6439 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6366 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6440 | 6367 | ||
6441 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6368 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6442 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6369 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6443 | ILK_ELPIN_409_SELECT); |
6370 | ILK_ELPIN_409_SELECT); |
6444 | I915_WRITE(_3D_CHICKEN2, |
6371 | I915_WRITE(_3D_CHICKEN2, |
6445 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | |
6372 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | |
6446 | _3D_CHICKEN2_WM_READ_PIPELINED); |
6373 | _3D_CHICKEN2_WM_READ_PIPELINED); |
6447 | 6374 | ||
6448 | /* WaDisableRenderCachePipelinedFlush:ilk */ |
6375 | /* WaDisableRenderCachePipelinedFlush:ilk */ |
6449 | I915_WRITE(CACHE_MODE_0, |
6376 | I915_WRITE(CACHE_MODE_0, |
6450 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); |
6377 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); |
6451 | 6378 | ||
6452 | /* WaDisable_RenderCache_OperationalFlush:ilk */ |
6379 | /* WaDisable_RenderCache_OperationalFlush:ilk */ |
6453 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6380 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6454 | 6381 | ||
6455 | g4x_disable_trickle_feed(dev); |
6382 | g4x_disable_trickle_feed(dev); |
6456 | 6383 | ||
6457 | ibx_init_clock_gating(dev); |
6384 | ibx_init_clock_gating(dev); |
6458 | } |
6385 | } |
6459 | 6386 | ||
6460 | static void cpt_init_clock_gating(struct drm_device *dev) |
6387 | static void cpt_init_clock_gating(struct drm_device *dev) |
6461 | { |
6388 | { |
6462 | struct drm_i915_private *dev_priv = dev->dev_private; |
6389 | struct drm_i915_private *dev_priv = dev->dev_private; |
6463 | int pipe; |
6390 | int pipe; |
6464 | uint32_t val; |
6391 | uint32_t val; |
6465 | 6392 | ||
6466 | /* |
6393 | /* |
6467 | * On Ibex Peak and Cougar Point, we need to disable clock |
6394 | * On Ibex Peak and Cougar Point, we need to disable clock |
6468 | * gating for the panel power sequencer or it will fail to |
6395 | * gating for the panel power sequencer or it will fail to |
6469 | * start up when no ports are active. |
6396 | * start up when no ports are active. |
6470 | */ |
6397 | */ |
6471 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE | |
6398 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE | |
6472 | PCH_DPLUNIT_CLOCK_GATE_DISABLE | |
6399 | PCH_DPLUNIT_CLOCK_GATE_DISABLE | |
6473 | PCH_CPUNIT_CLOCK_GATE_DISABLE); |
6400 | PCH_CPUNIT_CLOCK_GATE_DISABLE); |
6474 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | |
6401 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | |
6475 | DPLS_EDP_PPS_FIX_DIS); |
6402 | DPLS_EDP_PPS_FIX_DIS); |
6476 | /* The below fixes the weird display corruption, a few pixels shifted |
6403 | /* The below fixes the weird display corruption, a few pixels shifted |
6477 | * downward, on (only) LVDS of some HP laptops with IVY. |
6404 | * downward, on (only) LVDS of some HP laptops with IVY. |
6478 | */ |
6405 | */ |
6479 | for_each_pipe(dev_priv, pipe) { |
6406 | for_each_pipe(dev_priv, pipe) { |
6480 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
6407 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
6481 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; |
6408 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; |
6482 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
6409 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
6483 | if (dev_priv->vbt.fdi_rx_polarity_inverted) |
6410 | if (dev_priv->vbt.fdi_rx_polarity_inverted) |
6484 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
6411 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
6485 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
6412 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
6486 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; |
6413 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; |
6487 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; |
6414 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; |
6488 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
6415 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
6489 | } |
6416 | } |
6490 | /* WADP0ClockGatingDisable */ |
6417 | /* WADP0ClockGatingDisable */ |
6491 | for_each_pipe(dev_priv, pipe) { |
6418 | for_each_pipe(dev_priv, pipe) { |
6492 | I915_WRITE(TRANS_CHICKEN1(pipe), |
6419 | I915_WRITE(TRANS_CHICKEN1(pipe), |
6493 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
6420 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
6494 | } |
6421 | } |
6495 | } |
6422 | } |
6496 | 6423 | ||
6497 | static void gen6_check_mch_setup(struct drm_device *dev) |
6424 | static void gen6_check_mch_setup(struct drm_device *dev) |
6498 | { |
6425 | { |
6499 | struct drm_i915_private *dev_priv = dev->dev_private; |
6426 | struct drm_i915_private *dev_priv = dev->dev_private; |
6500 | uint32_t tmp; |
6427 | uint32_t tmp; |
6501 | 6428 | ||
6502 | tmp = I915_READ(MCH_SSKPD); |
6429 | tmp = I915_READ(MCH_SSKPD); |
6503 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) |
6430 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) |
6504 | DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n", |
6431 | DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n", |
6505 | tmp); |
6432 | tmp); |
6506 | } |
6433 | } |
6507 | 6434 | ||
6508 | static void gen6_init_clock_gating(struct drm_device *dev) |
6435 | static void gen6_init_clock_gating(struct drm_device *dev) |
6509 | { |
6436 | { |
6510 | struct drm_i915_private *dev_priv = dev->dev_private; |
6437 | struct drm_i915_private *dev_priv = dev->dev_private; |
6511 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6438 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6512 | 6439 | ||
6513 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6440 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6514 | 6441 | ||
6515 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6442 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6516 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6443 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6517 | ILK_ELPIN_409_SELECT); |
6444 | ILK_ELPIN_409_SELECT); |
6518 | 6445 | ||
6519 | /* WaDisableHiZPlanesWhenMSAAEnabled:snb */ |
6446 | /* WaDisableHiZPlanesWhenMSAAEnabled:snb */ |
6520 | I915_WRITE(_3D_CHICKEN, |
6447 | I915_WRITE(_3D_CHICKEN, |
6521 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); |
6448 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); |
6522 | 6449 | ||
6523 | /* WaDisable_RenderCache_OperationalFlush:snb */ |
6450 | /* WaDisable_RenderCache_OperationalFlush:snb */ |
6524 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6451 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6525 | 6452 | ||
6526 | /* |
6453 | /* |
6527 | * BSpec recoomends 8x4 when MSAA is used, |
6454 | * BSpec recoomends 8x4 when MSAA is used, |
6528 | * however in practice 16x4 seems fastest. |
6455 | * however in practice 16x4 seems fastest. |
6529 | * |
6456 | * |
6530 | * Note that PS/WM thread counts depend on the WIZ hashing |
6457 | * Note that PS/WM thread counts depend on the WIZ hashing |
6531 | * disable bit, which we don't touch here, but it's good |
6458 | * disable bit, which we don't touch here, but it's good |
6532 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
6459 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
6533 | */ |
6460 | */ |
6534 | I915_WRITE(GEN6_GT_MODE, |
6461 | I915_WRITE(GEN6_GT_MODE, |
6535 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
6462 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
6536 | 6463 | ||
6537 | ilk_init_lp_watermarks(dev); |
6464 | ilk_init_lp_watermarks(dev); |
6538 | 6465 | ||
6539 | I915_WRITE(CACHE_MODE_0, |
6466 | I915_WRITE(CACHE_MODE_0, |
6540 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
6467 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
6541 | 6468 | ||
6542 | I915_WRITE(GEN6_UCGCTL1, |
6469 | I915_WRITE(GEN6_UCGCTL1, |
6543 | I915_READ(GEN6_UCGCTL1) | |
6470 | I915_READ(GEN6_UCGCTL1) | |
6544 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | |
6471 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | |
6545 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); |
6472 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); |
6546 | 6473 | ||
6547 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock |
6474 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock |
6548 | * gating disable must be set. Failure to set it results in |
6475 | * gating disable must be set. Failure to set it results in |
6549 | * flickering pixels due to Z write ordering failures after |
6476 | * flickering pixels due to Z write ordering failures after |
6550 | * some amount of runtime in the Mesa "fire" demo, and Unigine |
6477 | * some amount of runtime in the Mesa "fire" demo, and Unigine |
6551 | * Sanctuary and Tropics, and apparently anything else with |
6478 | * Sanctuary and Tropics, and apparently anything else with |
6552 | * alpha test or pixel discard. |
6479 | * alpha test or pixel discard. |
6553 | * |
6480 | * |
6554 | * According to the spec, bit 11 (RCCUNIT) must also be set, |
6481 | * According to the spec, bit 11 (RCCUNIT) must also be set, |
6555 | * but we didn't debug actual testcases to find it out. |
6482 | * but we didn't debug actual testcases to find it out. |
6556 | * |
6483 | * |
6557 | * WaDisableRCCUnitClockGating:snb |
6484 | * WaDisableRCCUnitClockGating:snb |
6558 | * WaDisableRCPBUnitClockGating:snb |
6485 | * WaDisableRCPBUnitClockGating:snb |
6559 | */ |
6486 | */ |
6560 | I915_WRITE(GEN6_UCGCTL2, |
6487 | I915_WRITE(GEN6_UCGCTL2, |
6561 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | |
6488 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | |
6562 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); |
6489 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); |
6563 | 6490 | ||
6564 | /* WaStripsFansDisableFastClipPerformanceFix:snb */ |
6491 | /* WaStripsFansDisableFastClipPerformanceFix:snb */ |
6565 | I915_WRITE(_3D_CHICKEN3, |
6492 | I915_WRITE(_3D_CHICKEN3, |
6566 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL)); |
6493 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL)); |
6567 | 6494 | ||
6568 | /* |
6495 | /* |
6569 | * Bspec says: |
6496 | * Bspec says: |
6570 | * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and |
6497 | * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and |
6571 | * 3DSTATE_SF number of SF output attributes is more than 16." |
6498 | * 3DSTATE_SF number of SF output attributes is more than 16." |
6572 | */ |
6499 | */ |
6573 | I915_WRITE(_3D_CHICKEN3, |
6500 | I915_WRITE(_3D_CHICKEN3, |
6574 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH)); |
6501 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH)); |
6575 | 6502 | ||
6576 | /* |
6503 | /* |
6577 | * According to the spec the following bits should be |
6504 | * According to the spec the following bits should be |
6578 | * set in order to enable memory self-refresh and fbc: |
6505 | * set in order to enable memory self-refresh and fbc: |
6579 | * The bit21 and bit22 of 0x42000 |
6506 | * The bit21 and bit22 of 0x42000 |
6580 | * The bit21 and bit22 of 0x42004 |
6507 | * The bit21 and bit22 of 0x42004 |
6581 | * The bit5 and bit7 of 0x42020 |
6508 | * The bit5 and bit7 of 0x42020 |
6582 | * The bit14 of 0x70180 |
6509 | * The bit14 of 0x70180 |
6583 | * The bit14 of 0x71180 |
6510 | * The bit14 of 0x71180 |
6584 | * |
6511 | * |
6585 | * WaFbcAsynchFlipDisableFbcQueue:snb |
6512 | * WaFbcAsynchFlipDisableFbcQueue:snb |
6586 | */ |
6513 | */ |
6587 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
6514 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
6588 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
6515 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
6589 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); |
6516 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); |
6590 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6517 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6591 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6518 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
6592 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); |
6519 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); |
6593 | I915_WRITE(ILK_DSPCLK_GATE_D, |
6520 | I915_WRITE(ILK_DSPCLK_GATE_D, |
6594 | I915_READ(ILK_DSPCLK_GATE_D) | |
6521 | I915_READ(ILK_DSPCLK_GATE_D) | |
6595 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | |
6522 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | |
6596 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); |
6523 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); |
6597 | 6524 | ||
6598 | g4x_disable_trickle_feed(dev); |
6525 | g4x_disable_trickle_feed(dev); |
6599 | 6526 | ||
6600 | cpt_init_clock_gating(dev); |
6527 | cpt_init_clock_gating(dev); |
6601 | 6528 | ||
6602 | gen6_check_mch_setup(dev); |
6529 | gen6_check_mch_setup(dev); |
6603 | } |
6530 | } |
6604 | 6531 | ||
6605 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) |
6532 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) |
6606 | { |
6533 | { |
6607 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); |
6534 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); |
6608 | 6535 | ||
6609 | /* |
6536 | /* |
6610 | * WaVSThreadDispatchOverride:ivb,vlv |
6537 | * WaVSThreadDispatchOverride:ivb,vlv |
6611 | * |
6538 | * |
6612 | * This actually overrides the dispatch |
6539 | * This actually overrides the dispatch |
6613 | * mode for all thread types. |
6540 | * mode for all thread types. |
6614 | */ |
6541 | */ |
6615 | reg &= ~GEN7_FF_SCHED_MASK; |
6542 | reg &= ~GEN7_FF_SCHED_MASK; |
6616 | reg |= GEN7_FF_TS_SCHED_HW; |
6543 | reg |= GEN7_FF_TS_SCHED_HW; |
6617 | reg |= GEN7_FF_VS_SCHED_HW; |
6544 | reg |= GEN7_FF_VS_SCHED_HW; |
6618 | reg |= GEN7_FF_DS_SCHED_HW; |
6545 | reg |= GEN7_FF_DS_SCHED_HW; |
6619 | 6546 | ||
6620 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); |
6547 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); |
6621 | } |
6548 | } |
6622 | 6549 | ||
6623 | static void lpt_init_clock_gating(struct drm_device *dev) |
6550 | static void lpt_init_clock_gating(struct drm_device *dev) |
6624 | { |
6551 | { |
6625 | struct drm_i915_private *dev_priv = dev->dev_private; |
6552 | struct drm_i915_private *dev_priv = dev->dev_private; |
6626 | 6553 | ||
6627 | /* |
6554 | /* |
6628 | * TODO: this bit should only be enabled when really needed, then |
6555 | * TODO: this bit should only be enabled when really needed, then |
6629 | * disabled when not needed anymore in order to save power. |
6556 | * disabled when not needed anymore in order to save power. |
6630 | */ |
6557 | */ |
6631 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) |
6558 | if (HAS_PCH_LPT_LP(dev)) |
6632 | I915_WRITE(SOUTH_DSPCLK_GATE_D, |
6559 | I915_WRITE(SOUTH_DSPCLK_GATE_D, |
6633 | I915_READ(SOUTH_DSPCLK_GATE_D) | |
6560 | I915_READ(SOUTH_DSPCLK_GATE_D) | |
6634 | PCH_LP_PARTITION_LEVEL_DISABLE); |
6561 | PCH_LP_PARTITION_LEVEL_DISABLE); |
6635 | 6562 | ||
6636 | /* WADPOClockGatingDisable:hsw */ |
6563 | /* WADPOClockGatingDisable:hsw */ |
6637 | I915_WRITE(_TRANSA_CHICKEN1, |
6564 | I915_WRITE(TRANS_CHICKEN1(PIPE_A), |
6638 | I915_READ(_TRANSA_CHICKEN1) | |
6565 | I915_READ(TRANS_CHICKEN1(PIPE_A)) | |
6639 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
6566 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
6640 | } |
6567 | } |
6641 | 6568 | ||
6642 | static void lpt_suspend_hw(struct drm_device *dev) |
6569 | static void lpt_suspend_hw(struct drm_device *dev) |
6643 | { |
6570 | { |
6644 | struct drm_i915_private *dev_priv = dev->dev_private; |
6571 | struct drm_i915_private *dev_priv = dev->dev_private; |
6645 | 6572 | ||
6646 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) { |
6573 | if (HAS_PCH_LPT_LP(dev)) { |
6647 | uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D); |
6574 | uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D); |
6648 | 6575 | ||
6649 | val &= ~PCH_LP_PARTITION_LEVEL_DISABLE; |
6576 | val &= ~PCH_LP_PARTITION_LEVEL_DISABLE; |
6650 | I915_WRITE(SOUTH_DSPCLK_GATE_D, val); |
6577 | I915_WRITE(SOUTH_DSPCLK_GATE_D, val); |
6651 | } |
6578 | } |
6652 | } |
6579 | } |
6653 | 6580 | ||
6654 | static void broadwell_init_clock_gating(struct drm_device *dev) |
6581 | static void broadwell_init_clock_gating(struct drm_device *dev) |
6655 | { |
6582 | { |
6656 | struct drm_i915_private *dev_priv = dev->dev_private; |
6583 | struct drm_i915_private *dev_priv = dev->dev_private; |
6657 | enum pipe pipe; |
6584 | enum pipe pipe; |
- | 6585 | uint32_t misccpctl; |
|
6658 | - | ||
6659 | I915_WRITE(WM3_LP_ILK, 0); |
- | |
6660 | I915_WRITE(WM2_LP_ILK, 0); |
6586 | |
6661 | I915_WRITE(WM1_LP_ILK, 0); |
6587 | ilk_init_lp_watermarks(dev); |
6662 | 6588 | ||
6663 | /* WaSwitchSolVfFArbitrationPriority:bdw */ |
6589 | /* WaSwitchSolVfFArbitrationPriority:bdw */ |
6664 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
6590 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
6665 | 6591 | ||
6666 | /* WaPsrDPAMaskVBlankInSRD:bdw */ |
6592 | /* WaPsrDPAMaskVBlankInSRD:bdw */ |
6667 | I915_WRITE(CHICKEN_PAR1_1, |
6593 | I915_WRITE(CHICKEN_PAR1_1, |
6668 | I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD); |
6594 | I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD); |
6669 | 6595 | ||
6670 | /* WaPsrDPRSUnmaskVBlankInSRD:bdw */ |
6596 | /* WaPsrDPRSUnmaskVBlankInSRD:bdw */ |
6671 | for_each_pipe(dev_priv, pipe) { |
6597 | for_each_pipe(dev_priv, pipe) { |
6672 | I915_WRITE(CHICKEN_PIPESL_1(pipe), |
6598 | I915_WRITE(CHICKEN_PIPESL_1(pipe), |
6673 | I915_READ(CHICKEN_PIPESL_1(pipe)) | |
6599 | I915_READ(CHICKEN_PIPESL_1(pipe)) | |
6674 | BDW_DPRS_MASK_VBLANK_SRD); |
6600 | BDW_DPRS_MASK_VBLANK_SRD); |
6675 | } |
6601 | } |
6676 | 6602 | ||
6677 | /* WaVSRefCountFullforceMissDisable:bdw */ |
6603 | /* WaVSRefCountFullforceMissDisable:bdw */ |
6678 | /* WaDSRefCountFullforceMissDisable:bdw */ |
6604 | /* WaDSRefCountFullforceMissDisable:bdw */ |
6679 | I915_WRITE(GEN7_FF_THREAD_MODE, |
6605 | I915_WRITE(GEN7_FF_THREAD_MODE, |
6680 | I915_READ(GEN7_FF_THREAD_MODE) & |
6606 | I915_READ(GEN7_FF_THREAD_MODE) & |
6681 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); |
6607 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); |
6682 | 6608 | ||
6683 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
6609 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
6684 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); |
6610 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); |
6685 | 6611 | ||
6686 | /* WaDisableSDEUnitClockGating:bdw */ |
6612 | /* WaDisableSDEUnitClockGating:bdw */ |
6687 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
6613 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
6688 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
6614 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
- | 6615 | ||
- | 6616 | /* |
|
- | 6617 | * WaProgramL3SqcReg1Default:bdw |
|
- | 6618 | * WaTempDisableDOPClkGating:bdw |
|
- | 6619 | */ |
|
- | 6620 | misccpctl = I915_READ(GEN7_MISCCPCTL); |
|
- | 6621 | I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE); |
|
- | 6622 | I915_WRITE(GEN8_L3SQCREG1, BDW_WA_L3SQCREG1_DEFAULT); |
|
- | 6623 | I915_WRITE(GEN7_MISCCPCTL, misccpctl); |
|
- | 6624 | ||
- | 6625 | /* |
|
- | 6626 | * WaGttCachingOffByDefault:bdw |
|
- | 6627 | * GTT cache may not work with big pages, so if those |
|
- | 6628 | * are ever enabled GTT cache may need to be disabled. |
|
- | 6629 | */ |
|
- | 6630 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); |
|
6689 | 6631 | ||
6690 | lpt_init_clock_gating(dev); |
6632 | lpt_init_clock_gating(dev); |
6691 | } |
6633 | } |
6692 | 6634 | ||
6693 | static void haswell_init_clock_gating(struct drm_device *dev) |
6635 | static void haswell_init_clock_gating(struct drm_device *dev) |
6694 | { |
6636 | { |
6695 | struct drm_i915_private *dev_priv = dev->dev_private; |
6637 | struct drm_i915_private *dev_priv = dev->dev_private; |
6696 | 6638 | ||
6697 | ilk_init_lp_watermarks(dev); |
6639 | ilk_init_lp_watermarks(dev); |
6698 | 6640 | ||
6699 | /* L3 caching of data atomics doesn't work -- disable it. */ |
6641 | /* L3 caching of data atomics doesn't work -- disable it. */ |
6700 | I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE); |
6642 | I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE); |
6701 | I915_WRITE(HSW_ROW_CHICKEN3, |
6643 | I915_WRITE(HSW_ROW_CHICKEN3, |
6702 | _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE)); |
6644 | _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE)); |
6703 | 6645 | ||
6704 | /* This is required by WaCatErrorRejectionIssue:hsw */ |
6646 | /* This is required by WaCatErrorRejectionIssue:hsw */ |
6705 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6647 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6706 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
6648 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
6707 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
6649 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
6708 | 6650 | ||
6709 | /* WaVSRefCountFullforceMissDisable:hsw */ |
6651 | /* WaVSRefCountFullforceMissDisable:hsw */ |
6710 | I915_WRITE(GEN7_FF_THREAD_MODE, |
6652 | I915_WRITE(GEN7_FF_THREAD_MODE, |
6711 | I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME); |
6653 | I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME); |
6712 | 6654 | ||
6713 | /* WaDisable_RenderCache_OperationalFlush:hsw */ |
6655 | /* WaDisable_RenderCache_OperationalFlush:hsw */ |
6714 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6656 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6715 | 6657 | ||
6716 | /* enable HiZ Raw Stall Optimization */ |
6658 | /* enable HiZ Raw Stall Optimization */ |
6717 | I915_WRITE(CACHE_MODE_0_GEN7, |
6659 | I915_WRITE(CACHE_MODE_0_GEN7, |
6718 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); |
6660 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); |
6719 | 6661 | ||
6720 | /* WaDisable4x2SubspanOptimization:hsw */ |
6662 | /* WaDisable4x2SubspanOptimization:hsw */ |
6721 | I915_WRITE(CACHE_MODE_1, |
6663 | I915_WRITE(CACHE_MODE_1, |
6722 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
6664 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
6723 | 6665 | ||
6724 | /* |
6666 | /* |
6725 | * BSpec recommends 8x4 when MSAA is used, |
6667 | * BSpec recommends 8x4 when MSAA is used, |
6726 | * however in practice 16x4 seems fastest. |
6668 | * however in practice 16x4 seems fastest. |
6727 | * |
6669 | * |
6728 | * Note that PS/WM thread counts depend on the WIZ hashing |
6670 | * Note that PS/WM thread counts depend on the WIZ hashing |
6729 | * disable bit, which we don't touch here, but it's good |
6671 | * disable bit, which we don't touch here, but it's good |
6730 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
6672 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
6731 | */ |
6673 | */ |
6732 | I915_WRITE(GEN7_GT_MODE, |
6674 | I915_WRITE(GEN7_GT_MODE, |
6733 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
6675 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
- | 6676 | ||
- | 6677 | /* WaSampleCChickenBitEnable:hsw */ |
|
- | 6678 | I915_WRITE(HALF_SLICE_CHICKEN3, |
|
- | 6679 | _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE)); |
|
6734 | 6680 | ||
6735 | /* WaSwitchSolVfFArbitrationPriority:hsw */ |
6681 | /* WaSwitchSolVfFArbitrationPriority:hsw */ |
6736 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
6682 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
6737 | 6683 | ||
6738 | /* WaRsPkgCStateDisplayPMReq:hsw */ |
6684 | /* WaRsPkgCStateDisplayPMReq:hsw */ |
6739 | I915_WRITE(CHICKEN_PAR1_1, |
6685 | I915_WRITE(CHICKEN_PAR1_1, |
6740 | I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES); |
6686 | I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES); |
6741 | 6687 | ||
6742 | lpt_init_clock_gating(dev); |
6688 | lpt_init_clock_gating(dev); |
6743 | } |
6689 | } |
6744 | 6690 | ||
6745 | static void ivybridge_init_clock_gating(struct drm_device *dev) |
6691 | static void ivybridge_init_clock_gating(struct drm_device *dev) |
6746 | { |
6692 | { |
6747 | struct drm_i915_private *dev_priv = dev->dev_private; |
6693 | struct drm_i915_private *dev_priv = dev->dev_private; |
6748 | uint32_t snpcr; |
6694 | uint32_t snpcr; |
6749 | 6695 | ||
6750 | ilk_init_lp_watermarks(dev); |
6696 | ilk_init_lp_watermarks(dev); |
6751 | 6697 | ||
6752 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
6698 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
6753 | 6699 | ||
6754 | /* WaDisableEarlyCull:ivb */ |
6700 | /* WaDisableEarlyCull:ivb */ |
6755 | I915_WRITE(_3D_CHICKEN3, |
6701 | I915_WRITE(_3D_CHICKEN3, |
6756 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); |
6702 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); |
6757 | 6703 | ||
6758 | /* WaDisableBackToBackFlipFix:ivb */ |
6704 | /* WaDisableBackToBackFlipFix:ivb */ |
6759 | I915_WRITE(IVB_CHICKEN3, |
6705 | I915_WRITE(IVB_CHICKEN3, |
6760 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | |
6706 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | |
6761 | CHICKEN3_DGMG_DONE_FIX_DISABLE); |
6707 | CHICKEN3_DGMG_DONE_FIX_DISABLE); |
6762 | 6708 | ||
6763 | /* WaDisablePSDDualDispatchEnable:ivb */ |
6709 | /* WaDisablePSDDualDispatchEnable:ivb */ |
6764 | if (IS_IVB_GT1(dev)) |
6710 | if (IS_IVB_GT1(dev)) |
6765 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
6711 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
6766 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
6712 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
6767 | 6713 | ||
6768 | /* WaDisable_RenderCache_OperationalFlush:ivb */ |
6714 | /* WaDisable_RenderCache_OperationalFlush:ivb */ |
6769 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6715 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6770 | 6716 | ||
6771 | /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */ |
6717 | /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */ |
6772 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
6718 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
6773 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); |
6719 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); |
6774 | 6720 | ||
6775 | /* WaApplyL3ControlAndL3ChickenMode:ivb */ |
6721 | /* WaApplyL3ControlAndL3ChickenMode:ivb */ |
6776 | I915_WRITE(GEN7_L3CNTLREG1, |
6722 | I915_WRITE(GEN7_L3CNTLREG1, |
6777 | GEN7_WA_FOR_GEN7_L3_CONTROL); |
6723 | GEN7_WA_FOR_GEN7_L3_CONTROL); |
6778 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, |
6724 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, |
6779 | GEN7_WA_L3_CHICKEN_MODE); |
6725 | GEN7_WA_L3_CHICKEN_MODE); |
6780 | if (IS_IVB_GT1(dev)) |
6726 | if (IS_IVB_GT1(dev)) |
6781 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6727 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6782 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6728 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6783 | else { |
6729 | else { |
6784 | /* must write both registers */ |
6730 | /* must write both registers */ |
6785 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6731 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6786 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6732 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6787 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, |
6733 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, |
6788 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6734 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6789 | } |
6735 | } |
6790 | 6736 | ||
6791 | /* WaForceL3Serialization:ivb */ |
6737 | /* WaForceL3Serialization:ivb */ |
6792 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
6738 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
6793 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
6739 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
6794 | 6740 | ||
6795 | /* |
6741 | /* |
6796 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
6742 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
6797 | * This implements the WaDisableRCZUnitClockGating:ivb workaround. |
6743 | * This implements the WaDisableRCZUnitClockGating:ivb workaround. |
6798 | */ |
6744 | */ |
6799 | I915_WRITE(GEN6_UCGCTL2, |
6745 | I915_WRITE(GEN6_UCGCTL2, |
6800 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
6746 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
6801 | 6747 | ||
6802 | /* This is required by WaCatErrorRejectionIssue:ivb */ |
6748 | /* This is required by WaCatErrorRejectionIssue:ivb */ |
6803 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6749 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6804 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
6750 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
6805 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
6751 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
6806 | 6752 | ||
6807 | g4x_disable_trickle_feed(dev); |
6753 | g4x_disable_trickle_feed(dev); |
6808 | 6754 | ||
6809 | gen7_setup_fixed_func_scheduler(dev_priv); |
6755 | gen7_setup_fixed_func_scheduler(dev_priv); |
6810 | 6756 | ||
6811 | if (0) { /* causes HiZ corruption on ivb:gt1 */ |
6757 | if (0) { /* causes HiZ corruption on ivb:gt1 */ |
6812 | /* enable HiZ Raw Stall Optimization */ |
6758 | /* enable HiZ Raw Stall Optimization */ |
6813 | I915_WRITE(CACHE_MODE_0_GEN7, |
6759 | I915_WRITE(CACHE_MODE_0_GEN7, |
6814 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); |
6760 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); |
6815 | } |
6761 | } |
6816 | 6762 | ||
6817 | /* WaDisable4x2SubspanOptimization:ivb */ |
6763 | /* WaDisable4x2SubspanOptimization:ivb */ |
6818 | I915_WRITE(CACHE_MODE_1, |
6764 | I915_WRITE(CACHE_MODE_1, |
6819 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
6765 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
6820 | 6766 | ||
6821 | /* |
6767 | /* |
6822 | * BSpec recommends 8x4 when MSAA is used, |
6768 | * BSpec recommends 8x4 when MSAA is used, |
6823 | * however in practice 16x4 seems fastest. |
6769 | * however in practice 16x4 seems fastest. |
6824 | * |
6770 | * |
6825 | * Note that PS/WM thread counts depend on the WIZ hashing |
6771 | * Note that PS/WM thread counts depend on the WIZ hashing |
6826 | * disable bit, which we don't touch here, but it's good |
6772 | * disable bit, which we don't touch here, but it's good |
6827 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
6773 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
6828 | */ |
6774 | */ |
6829 | I915_WRITE(GEN7_GT_MODE, |
6775 | I915_WRITE(GEN7_GT_MODE, |
6830 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
6776 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
6831 | 6777 | ||
6832 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); |
6778 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); |
6833 | snpcr &= ~GEN6_MBC_SNPCR_MASK; |
6779 | snpcr &= ~GEN6_MBC_SNPCR_MASK; |
6834 | snpcr |= GEN6_MBC_SNPCR_MED; |
6780 | snpcr |= GEN6_MBC_SNPCR_MED; |
6835 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); |
6781 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); |
6836 | 6782 | ||
6837 | if (!HAS_PCH_NOP(dev)) |
6783 | if (!HAS_PCH_NOP(dev)) |
6838 | cpt_init_clock_gating(dev); |
6784 | cpt_init_clock_gating(dev); |
6839 | 6785 | ||
6840 | gen6_check_mch_setup(dev); |
6786 | gen6_check_mch_setup(dev); |
6841 | } |
6787 | } |
- | 6788 | ||
- | 6789 | static void vlv_init_display_clock_gating(struct drm_i915_private *dev_priv) |
|
- | 6790 | { |
|
- | 6791 | I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE); |
|
- | 6792 | ||
- | 6793 | /* |
|
- | 6794 | * Disable trickle feed and enable pnd deadline calculation |
|
- | 6795 | */ |
|
- | 6796 | I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE); |
|
- | 6797 | I915_WRITE(CBR1_VLV, 0); |
|
- | 6798 | } |
|
6842 | 6799 | ||
6843 | static void valleyview_init_clock_gating(struct drm_device *dev) |
6800 | static void valleyview_init_clock_gating(struct drm_device *dev) |
6844 | { |
6801 | { |
6845 | struct drm_i915_private *dev_priv = dev->dev_private; |
6802 | struct drm_i915_private *dev_priv = dev->dev_private; |
6846 | 6803 | ||
6847 | I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE); |
6804 | vlv_init_display_clock_gating(dev_priv); |
6848 | 6805 | ||
6849 | /* WaDisableEarlyCull:vlv */ |
6806 | /* WaDisableEarlyCull:vlv */ |
6850 | I915_WRITE(_3D_CHICKEN3, |
6807 | I915_WRITE(_3D_CHICKEN3, |
6851 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); |
6808 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); |
6852 | 6809 | ||
6853 | /* WaDisableBackToBackFlipFix:vlv */ |
6810 | /* WaDisableBackToBackFlipFix:vlv */ |
6854 | I915_WRITE(IVB_CHICKEN3, |
6811 | I915_WRITE(IVB_CHICKEN3, |
6855 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | |
6812 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | |
6856 | CHICKEN3_DGMG_DONE_FIX_DISABLE); |
6813 | CHICKEN3_DGMG_DONE_FIX_DISABLE); |
6857 | 6814 | ||
6858 | /* WaPsdDispatchEnable:vlv */ |
6815 | /* WaPsdDispatchEnable:vlv */ |
6859 | /* WaDisablePSDDualDispatchEnable:vlv */ |
6816 | /* WaDisablePSDDualDispatchEnable:vlv */ |
6860 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
6817 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
6861 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
6818 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
6862 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
6819 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
6863 | 6820 | ||
6864 | /* WaDisable_RenderCache_OperationalFlush:vlv */ |
6821 | /* WaDisable_RenderCache_OperationalFlush:vlv */ |
6865 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6822 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6866 | 6823 | ||
6867 | /* WaForceL3Serialization:vlv */ |
6824 | /* WaForceL3Serialization:vlv */ |
6868 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
6825 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
6869 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
6826 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
6870 | 6827 | ||
6871 | /* WaDisableDopClockGating:vlv */ |
6828 | /* WaDisableDopClockGating:vlv */ |
6872 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6829 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6873 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6830 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
6874 | 6831 | ||
6875 | /* This is required by WaCatErrorRejectionIssue:vlv */ |
6832 | /* This is required by WaCatErrorRejectionIssue:vlv */ |
6876 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6833 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6877 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
6834 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
6878 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
6835 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
6879 | 6836 | ||
6880 | gen7_setup_fixed_func_scheduler(dev_priv); |
6837 | gen7_setup_fixed_func_scheduler(dev_priv); |
6881 | 6838 | ||
6882 | /* |
6839 | /* |
6883 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
6840 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
6884 | * This implements the WaDisableRCZUnitClockGating:vlv workaround. |
6841 | * This implements the WaDisableRCZUnitClockGating:vlv workaround. |
6885 | */ |
6842 | */ |
6886 | I915_WRITE(GEN6_UCGCTL2, |
6843 | I915_WRITE(GEN6_UCGCTL2, |
6887 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
6844 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
6888 | 6845 | ||
6889 | /* WaDisableL3Bank2xClockGate:vlv |
6846 | /* WaDisableL3Bank2xClockGate:vlv |
6890 | * Disabling L3 clock gating- MMIO 940c[25] = 1 |
6847 | * Disabling L3 clock gating- MMIO 940c[25] = 1 |
6891 | * Set bit 25, to disable L3_BANK_2x_CLK_GATING */ |
6848 | * Set bit 25, to disable L3_BANK_2x_CLK_GATING */ |
6892 | I915_WRITE(GEN7_UCGCTL4, |
6849 | I915_WRITE(GEN7_UCGCTL4, |
6893 | I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE); |
6850 | I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE); |
6894 | - | ||
6895 | I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE); |
- | |
6896 | 6851 | ||
6897 | /* |
6852 | /* |
6898 | * BSpec says this must be set, even though |
6853 | * BSpec says this must be set, even though |
6899 | * WaDisable4x2SubspanOptimization isn't listed for VLV. |
6854 | * WaDisable4x2SubspanOptimization isn't listed for VLV. |
6900 | */ |
6855 | */ |
6901 | I915_WRITE(CACHE_MODE_1, |
6856 | I915_WRITE(CACHE_MODE_1, |
6902 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
6857 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
6903 | 6858 | ||
6904 | /* |
6859 | /* |
- | 6860 | * BSpec recommends 8x4 when MSAA is used, |
|
- | 6861 | * however in practice 16x4 seems fastest. |
|
- | 6862 | * |
|
- | 6863 | * Note that PS/WM thread counts depend on the WIZ hashing |
|
- | 6864 | * disable bit, which we don't touch here, but it's good |
|
- | 6865 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). |
|
- | 6866 | */ |
|
- | 6867 | I915_WRITE(GEN7_GT_MODE, |
|
- | 6868 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
|
- | 6869 | ||
- | 6870 | /* |
|
6905 | * WaIncreaseL3CreditsForVLVB0:vlv |
6871 | * WaIncreaseL3CreditsForVLVB0:vlv |
6906 | * This is the hardware default actually. |
6872 | * This is the hardware default actually. |
6907 | */ |
6873 | */ |
6908 | I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE); |
6874 | I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE); |
6909 | 6875 | ||
6910 | /* |
6876 | /* |
6911 | * WaDisableVLVClockGating_VBIIssue:vlv |
6877 | * WaDisableVLVClockGating_VBIIssue:vlv |
6912 | * Disable clock gating on th GCFG unit to prevent a delay |
6878 | * Disable clock gating on th GCFG unit to prevent a delay |
6913 | * in the reporting of vblank events. |
6879 | * in the reporting of vblank events. |
6914 | */ |
6880 | */ |
6915 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS); |
6881 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS); |
6916 | } |
6882 | } |
6917 | 6883 | ||
6918 | static void cherryview_init_clock_gating(struct drm_device *dev) |
6884 | static void cherryview_init_clock_gating(struct drm_device *dev) |
6919 | { |
6885 | { |
6920 | struct drm_i915_private *dev_priv = dev->dev_private; |
6886 | struct drm_i915_private *dev_priv = dev->dev_private; |
6921 | 6887 | ||
6922 | I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE); |
- | |
6923 | - | ||
6924 | I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE); |
6888 | vlv_init_display_clock_gating(dev_priv); |
6925 | 6889 | ||
6926 | /* WaVSRefCountFullforceMissDisable:chv */ |
6890 | /* WaVSRefCountFullforceMissDisable:chv */ |
6927 | /* WaDSRefCountFullforceMissDisable:chv */ |
6891 | /* WaDSRefCountFullforceMissDisable:chv */ |
6928 | I915_WRITE(GEN7_FF_THREAD_MODE, |
6892 | I915_WRITE(GEN7_FF_THREAD_MODE, |
6929 | I915_READ(GEN7_FF_THREAD_MODE) & |
6893 | I915_READ(GEN7_FF_THREAD_MODE) & |
6930 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); |
6894 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); |
6931 | 6895 | ||
6932 | /* WaDisableSemaphoreAndSyncFlipWait:chv */ |
6896 | /* WaDisableSemaphoreAndSyncFlipWait:chv */ |
6933 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
6897 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
6934 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); |
6898 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); |
6935 | 6899 | ||
6936 | /* WaDisableCSUnitClockGating:chv */ |
6900 | /* WaDisableCSUnitClockGating:chv */ |
6937 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | |
6901 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | |
6938 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); |
6902 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); |
6939 | 6903 | ||
6940 | /* WaDisableSDEUnitClockGating:chv */ |
6904 | /* WaDisableSDEUnitClockGating:chv */ |
6941 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
6905 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
6942 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
6906 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
- | 6907 | ||
- | 6908 | /* |
|
- | 6909 | * GTT cache may not work with big pages, so if those |
|
- | 6910 | * are ever enabled GTT cache may need to be disabled. |
|
- | 6911 | */ |
|
- | 6912 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); |
|
6943 | } |
6913 | } |
6944 | 6914 | ||
6945 | static void g4x_init_clock_gating(struct drm_device *dev) |
6915 | static void g4x_init_clock_gating(struct drm_device *dev) |
6946 | { |
6916 | { |
6947 | struct drm_i915_private *dev_priv = dev->dev_private; |
6917 | struct drm_i915_private *dev_priv = dev->dev_private; |
6948 | uint32_t dspclk_gate; |
6918 | uint32_t dspclk_gate; |
6949 | 6919 | ||
6950 | I915_WRITE(RENCLK_GATE_D1, 0); |
6920 | I915_WRITE(RENCLK_GATE_D1, 0); |
6951 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | |
6921 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | |
6952 | GS_UNIT_CLOCK_GATE_DISABLE | |
6922 | GS_UNIT_CLOCK_GATE_DISABLE | |
6953 | CL_UNIT_CLOCK_GATE_DISABLE); |
6923 | CL_UNIT_CLOCK_GATE_DISABLE); |
6954 | I915_WRITE(RAMCLK_GATE_D, 0); |
6924 | I915_WRITE(RAMCLK_GATE_D, 0); |
6955 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | |
6925 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | |
6956 | OVRUNIT_CLOCK_GATE_DISABLE | |
6926 | OVRUNIT_CLOCK_GATE_DISABLE | |
6957 | OVCUNIT_CLOCK_GATE_DISABLE; |
6927 | OVCUNIT_CLOCK_GATE_DISABLE; |
6958 | if (IS_GM45(dev)) |
6928 | if (IS_GM45(dev)) |
6959 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; |
6929 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; |
6960 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); |
6930 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); |
6961 | 6931 | ||
6962 | /* WaDisableRenderCachePipelinedFlush */ |
6932 | /* WaDisableRenderCachePipelinedFlush */ |
6963 | I915_WRITE(CACHE_MODE_0, |
6933 | I915_WRITE(CACHE_MODE_0, |
6964 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); |
6934 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); |
6965 | 6935 | ||
6966 | /* WaDisable_RenderCache_OperationalFlush:g4x */ |
6936 | /* WaDisable_RenderCache_OperationalFlush:g4x */ |
6967 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6937 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6968 | 6938 | ||
6969 | g4x_disable_trickle_feed(dev); |
6939 | g4x_disable_trickle_feed(dev); |
6970 | } |
6940 | } |
6971 | 6941 | ||
6972 | static void crestline_init_clock_gating(struct drm_device *dev) |
6942 | static void crestline_init_clock_gating(struct drm_device *dev) |
6973 | { |
6943 | { |
6974 | struct drm_i915_private *dev_priv = dev->dev_private; |
6944 | struct drm_i915_private *dev_priv = dev->dev_private; |
6975 | 6945 | ||
6976 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); |
6946 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); |
6977 | I915_WRITE(RENCLK_GATE_D2, 0); |
6947 | I915_WRITE(RENCLK_GATE_D2, 0); |
6978 | I915_WRITE(DSPCLK_GATE_D, 0); |
6948 | I915_WRITE(DSPCLK_GATE_D, 0); |
6979 | I915_WRITE(RAMCLK_GATE_D, 0); |
6949 | I915_WRITE(RAMCLK_GATE_D, 0); |
6980 | I915_WRITE16(DEUC, 0); |
6950 | I915_WRITE16(DEUC, 0); |
6981 | I915_WRITE(MI_ARB_STATE, |
6951 | I915_WRITE(MI_ARB_STATE, |
6982 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); |
6952 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); |
6983 | 6953 | ||
6984 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ |
6954 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ |
6985 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6955 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6986 | } |
6956 | } |
6987 | 6957 | ||
6988 | static void broadwater_init_clock_gating(struct drm_device *dev) |
6958 | static void broadwater_init_clock_gating(struct drm_device *dev) |
6989 | { |
6959 | { |
6990 | struct drm_i915_private *dev_priv = dev->dev_private; |
6960 | struct drm_i915_private *dev_priv = dev->dev_private; |
6991 | 6961 | ||
6992 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | |
6962 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | |
6993 | I965_RCC_CLOCK_GATE_DISABLE | |
6963 | I965_RCC_CLOCK_GATE_DISABLE | |
6994 | I965_RCPB_CLOCK_GATE_DISABLE | |
6964 | I965_RCPB_CLOCK_GATE_DISABLE | |
6995 | I965_ISC_CLOCK_GATE_DISABLE | |
6965 | I965_ISC_CLOCK_GATE_DISABLE | |
6996 | I965_FBC_CLOCK_GATE_DISABLE); |
6966 | I965_FBC_CLOCK_GATE_DISABLE); |
6997 | I915_WRITE(RENCLK_GATE_D2, 0); |
6967 | I915_WRITE(RENCLK_GATE_D2, 0); |
6998 | I915_WRITE(MI_ARB_STATE, |
6968 | I915_WRITE(MI_ARB_STATE, |
6999 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); |
6969 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); |
7000 | 6970 | ||
7001 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ |
6971 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ |
7002 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
6972 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); |
7003 | } |
6973 | } |
7004 | 6974 | ||
7005 | static void gen3_init_clock_gating(struct drm_device *dev) |
6975 | static void gen3_init_clock_gating(struct drm_device *dev) |
7006 | { |
6976 | { |
7007 | struct drm_i915_private *dev_priv = dev->dev_private; |
6977 | struct drm_i915_private *dev_priv = dev->dev_private; |
7008 | u32 dstate = I915_READ(D_STATE); |
6978 | u32 dstate = I915_READ(D_STATE); |
7009 | 6979 | ||
7010 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | |
6980 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | |
7011 | DSTATE_DOT_CLOCK_GATING; |
6981 | DSTATE_DOT_CLOCK_GATING; |
7012 | I915_WRITE(D_STATE, dstate); |
6982 | I915_WRITE(D_STATE, dstate); |
7013 | 6983 | ||
7014 | if (IS_PINEVIEW(dev)) |
6984 | if (IS_PINEVIEW(dev)) |
7015 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); |
6985 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); |
7016 | 6986 | ||
7017 | /* IIR "flip pending" means done if this bit is set */ |
6987 | /* IIR "flip pending" means done if this bit is set */ |
7018 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); |
6988 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); |
7019 | 6989 | ||
7020 | /* interrupts should cause a wake up from C3 */ |
6990 | /* interrupts should cause a wake up from C3 */ |
7021 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN)); |
6991 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN)); |
7022 | 6992 | ||
7023 | /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ |
6993 | /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ |
7024 | I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); |
6994 | I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); |
7025 | 6995 | ||
7026 | I915_WRITE(MI_ARB_STATE, |
6996 | I915_WRITE(MI_ARB_STATE, |
7027 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); |
6997 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); |
7028 | } |
6998 | } |
7029 | 6999 | ||
7030 | static void i85x_init_clock_gating(struct drm_device *dev) |
7000 | static void i85x_init_clock_gating(struct drm_device *dev) |
7031 | { |
7001 | { |
7032 | struct drm_i915_private *dev_priv = dev->dev_private; |
7002 | struct drm_i915_private *dev_priv = dev->dev_private; |
7033 | 7003 | ||
7034 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); |
7004 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); |
7035 | 7005 | ||
7036 | /* interrupts should cause a wake up from C3 */ |
7006 | /* interrupts should cause a wake up from C3 */ |
7037 | I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) | |
7007 | I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) | |
7038 | _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE)); |
7008 | _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE)); |
7039 | 7009 | ||
7040 | I915_WRITE(MEM_MODE, |
7010 | I915_WRITE(MEM_MODE, |
7041 | _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE)); |
7011 | _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE)); |
7042 | } |
7012 | } |
7043 | 7013 | ||
7044 | static void i830_init_clock_gating(struct drm_device *dev) |
7014 | static void i830_init_clock_gating(struct drm_device *dev) |
7045 | { |
7015 | { |
7046 | struct drm_i915_private *dev_priv = dev->dev_private; |
7016 | struct drm_i915_private *dev_priv = dev->dev_private; |
7047 | 7017 | ||
7048 | I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE); |
7018 | I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE); |
7049 | 7019 | ||
7050 | I915_WRITE(MEM_MODE, |
7020 | I915_WRITE(MEM_MODE, |
7051 | _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) | |
7021 | _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) | |
7052 | _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE)); |
7022 | _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE)); |
7053 | } |
7023 | } |
7054 | 7024 | ||
7055 | void intel_init_clock_gating(struct drm_device *dev) |
7025 | void intel_init_clock_gating(struct drm_device *dev) |
7056 | { |
7026 | { |
7057 | struct drm_i915_private *dev_priv = dev->dev_private; |
7027 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | 7028 | ||
7058 | 7029 | if (dev_priv->display.init_clock_gating) |
|
7059 | dev_priv->display.init_clock_gating(dev); |
7030 | dev_priv->display.init_clock_gating(dev); |
7060 | } |
7031 | } |
7061 | 7032 | ||
7062 | void intel_suspend_hw(struct drm_device *dev) |
7033 | void intel_suspend_hw(struct drm_device *dev) |
7063 | { |
7034 | { |
7064 | if (HAS_PCH_LPT(dev)) |
7035 | if (HAS_PCH_LPT(dev)) |
7065 | lpt_suspend_hw(dev); |
7036 | lpt_suspend_hw(dev); |
7066 | } |
7037 | } |
7067 | - | ||
7068 | static void intel_init_fbc(struct drm_i915_private *dev_priv) |
- | |
7069 | { |
- | |
7070 | if (!HAS_FBC(dev_priv)) { |
- | |
7071 | dev_priv->fbc.enabled = false; |
- | |
7072 | return; |
- | |
7073 | } |
- | |
7074 | - | ||
7075 | if (INTEL_INFO(dev_priv)->gen >= 7) { |
- | |
7076 | dev_priv->display.fbc_enabled = ironlake_fbc_enabled; |
- | |
7077 | dev_priv->display.enable_fbc = gen7_enable_fbc; |
- | |
7078 | dev_priv->display.disable_fbc = ironlake_disable_fbc; |
- | |
7079 | } else if (INTEL_INFO(dev_priv)->gen >= 5) { |
- | |
7080 | dev_priv->display.fbc_enabled = ironlake_fbc_enabled; |
- | |
7081 | dev_priv->display.enable_fbc = ironlake_enable_fbc; |
- | |
7082 | dev_priv->display.disable_fbc = ironlake_disable_fbc; |
- | |
7083 | } else if (IS_GM45(dev_priv)) { |
- | |
7084 | dev_priv->display.fbc_enabled = g4x_fbc_enabled; |
- | |
7085 | dev_priv->display.enable_fbc = g4x_enable_fbc; |
- | |
7086 | dev_priv->display.disable_fbc = g4x_disable_fbc; |
- | |
7087 | } else { |
- | |
7088 | dev_priv->display.fbc_enabled = i8xx_fbc_enabled; |
- | |
7089 | dev_priv->display.enable_fbc = i8xx_enable_fbc; |
- | |
7090 | dev_priv->display.disable_fbc = i8xx_disable_fbc; |
- | |
7091 | - | ||
7092 | /* This value was pulled out of someone's hat */ |
- | |
7093 | I915_WRITE(FBC_CONTROL, 500 << FBC_CTL_INTERVAL_SHIFT); |
- | |
7094 | } |
- | |
7095 | - | ||
7096 | dev_priv->fbc.enabled = dev_priv->display.fbc_enabled(dev_priv->dev); |
- | |
7097 | } |
- | |
7098 | 7038 | ||
7099 | /* Set up chip specific power management-related functions */ |
7039 | /* Set up chip specific power management-related functions */ |
7100 | void intel_init_pm(struct drm_device *dev) |
7040 | void intel_init_pm(struct drm_device *dev) |
7101 | { |
7041 | { |
7102 | struct drm_i915_private *dev_priv = dev->dev_private; |
7042 | struct drm_i915_private *dev_priv = dev->dev_private; |
7103 | 7043 | ||
7104 | intel_init_fbc(dev_priv); |
7044 | intel_fbc_init(dev_priv); |
7105 | 7045 | ||
7106 | /* For cxsr */ |
7046 | /* For cxsr */ |
7107 | if (IS_PINEVIEW(dev)) |
7047 | if (IS_PINEVIEW(dev)) |
7108 | i915_pineview_get_mem_freq(dev); |
7048 | i915_pineview_get_mem_freq(dev); |
7109 | else if (IS_GEN5(dev)) |
7049 | else if (IS_GEN5(dev)) |
7110 | i915_ironlake_get_mem_freq(dev); |
7050 | i915_ironlake_get_mem_freq(dev); |
7111 | 7051 | ||
7112 | /* For FIFO watermark updates */ |
7052 | /* For FIFO watermark updates */ |
7113 | if (INTEL_INFO(dev)->gen >= 9) { |
7053 | if (INTEL_INFO(dev)->gen >= 9) { |
7114 | skl_setup_wm_latency(dev); |
7054 | skl_setup_wm_latency(dev); |
- | 7055 | ||
7115 | 7056 | if (IS_BROXTON(dev)) |
|
- | 7057 | dev_priv->display.init_clock_gating = |
|
7116 | dev_priv->display.init_clock_gating = gen9_init_clock_gating; |
7058 | bxt_init_clock_gating; |
7117 | dev_priv->display.update_wm = skl_update_wm; |
7059 | dev_priv->display.update_wm = skl_update_wm; |
7118 | dev_priv->display.update_sprite_wm = skl_update_sprite_wm; |
7060 | dev_priv->display.update_sprite_wm = skl_update_sprite_wm; |
7119 | } else if (HAS_PCH_SPLIT(dev)) { |
7061 | } else if (HAS_PCH_SPLIT(dev)) { |
7120 | ilk_setup_wm_latency(dev); |
7062 | ilk_setup_wm_latency(dev); |
7121 | 7063 | ||
7122 | if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] && |
7064 | if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] && |
7123 | dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) || |
7065 | dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) || |
7124 | (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] && |
7066 | (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] && |
7125 | dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) { |
7067 | dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) { |
7126 | dev_priv->display.update_wm = ilk_update_wm; |
7068 | dev_priv->display.update_wm = ilk_update_wm; |
7127 | dev_priv->display.update_sprite_wm = ilk_update_sprite_wm; |
7069 | dev_priv->display.update_sprite_wm = ilk_update_sprite_wm; |
7128 | } else { |
7070 | } else { |
7129 | DRM_DEBUG_KMS("Failed to read display plane latency. " |
7071 | DRM_DEBUG_KMS("Failed to read display plane latency. " |
7130 | "Disable CxSR\n"); |
7072 | "Disable CxSR\n"); |
7131 | } |
7073 | } |
7132 | 7074 | ||
7133 | if (IS_GEN5(dev)) |
7075 | if (IS_GEN5(dev)) |
7134 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; |
7076 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; |
7135 | else if (IS_GEN6(dev)) |
7077 | else if (IS_GEN6(dev)) |
7136 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; |
7078 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; |
7137 | else if (IS_IVYBRIDGE(dev)) |
7079 | else if (IS_IVYBRIDGE(dev)) |
7138 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; |
7080 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; |
7139 | else if (IS_HASWELL(dev)) |
7081 | else if (IS_HASWELL(dev)) |
7140 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; |
7082 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; |
7141 | else if (INTEL_INFO(dev)->gen == 8) |
7083 | else if (INTEL_INFO(dev)->gen == 8) |
7142 | dev_priv->display.init_clock_gating = broadwell_init_clock_gating; |
7084 | dev_priv->display.init_clock_gating = broadwell_init_clock_gating; |
7143 | } else if (IS_CHERRYVIEW(dev)) { |
7085 | } else if (IS_CHERRYVIEW(dev)) { |
7144 | dev_priv->display.update_wm = cherryview_update_wm; |
7086 | vlv_setup_wm_latency(dev); |
- | 7087 | ||
7145 | dev_priv->display.update_sprite_wm = valleyview_update_sprite_wm; |
7088 | dev_priv->display.update_wm = vlv_update_wm; |
7146 | dev_priv->display.init_clock_gating = |
7089 | dev_priv->display.init_clock_gating = |
7147 | cherryview_init_clock_gating; |
7090 | cherryview_init_clock_gating; |
7148 | } else if (IS_VALLEYVIEW(dev)) { |
7091 | } else if (IS_VALLEYVIEW(dev)) { |
7149 | dev_priv->display.update_wm = valleyview_update_wm; |
7092 | vlv_setup_wm_latency(dev); |
- | 7093 | ||
7150 | dev_priv->display.update_sprite_wm = valleyview_update_sprite_wm; |
7094 | dev_priv->display.update_wm = vlv_update_wm; |
7151 | dev_priv->display.init_clock_gating = |
7095 | dev_priv->display.init_clock_gating = |
7152 | valleyview_init_clock_gating; |
7096 | valleyview_init_clock_gating; |
7153 | } else if (IS_PINEVIEW(dev)) { |
7097 | } else if (IS_PINEVIEW(dev)) { |
7154 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev), |
7098 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev), |
7155 | dev_priv->is_ddr3, |
7099 | dev_priv->is_ddr3, |
7156 | dev_priv->fsb_freq, |
7100 | dev_priv->fsb_freq, |
7157 | dev_priv->mem_freq)) { |
7101 | dev_priv->mem_freq)) { |
7158 | DRM_INFO("failed to find known CxSR latency " |
7102 | DRM_INFO("failed to find known CxSR latency " |
7159 | "(found ddr%s fsb freq %d, mem freq %d), " |
7103 | "(found ddr%s fsb freq %d, mem freq %d), " |
7160 | "disabling CxSR\n", |
7104 | "disabling CxSR\n", |
7161 | (dev_priv->is_ddr3 == 1) ? "3" : "2", |
7105 | (dev_priv->is_ddr3 == 1) ? "3" : "2", |
7162 | dev_priv->fsb_freq, dev_priv->mem_freq); |
7106 | dev_priv->fsb_freq, dev_priv->mem_freq); |
7163 | /* Disable CxSR and never update its watermark again */ |
7107 | /* Disable CxSR and never update its watermark again */ |
7164 | intel_set_memory_cxsr(dev_priv, false); |
7108 | intel_set_memory_cxsr(dev_priv, false); |
7165 | dev_priv->display.update_wm = NULL; |
7109 | dev_priv->display.update_wm = NULL; |
7166 | } else |
7110 | } else |
7167 | dev_priv->display.update_wm = pineview_update_wm; |
7111 | dev_priv->display.update_wm = pineview_update_wm; |
7168 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; |
7112 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; |
7169 | } else if (IS_G4X(dev)) { |
7113 | } else if (IS_G4X(dev)) { |
7170 | dev_priv->display.update_wm = g4x_update_wm; |
7114 | dev_priv->display.update_wm = g4x_update_wm; |
7171 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; |
7115 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; |
7172 | } else if (IS_GEN4(dev)) { |
7116 | } else if (IS_GEN4(dev)) { |
7173 | dev_priv->display.update_wm = i965_update_wm; |
7117 | dev_priv->display.update_wm = i965_update_wm; |
7174 | if (IS_CRESTLINE(dev)) |
7118 | if (IS_CRESTLINE(dev)) |
7175 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; |
7119 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; |
7176 | else if (IS_BROADWATER(dev)) |
7120 | else if (IS_BROADWATER(dev)) |
7177 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; |
7121 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; |
7178 | } else if (IS_GEN3(dev)) { |
7122 | } else if (IS_GEN3(dev)) { |
7179 | dev_priv->display.update_wm = i9xx_update_wm; |
7123 | dev_priv->display.update_wm = i9xx_update_wm; |
7180 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; |
7124 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; |
7181 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; |
7125 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; |
7182 | } else if (IS_GEN2(dev)) { |
7126 | } else if (IS_GEN2(dev)) { |
7183 | if (INTEL_INFO(dev)->num_pipes == 1) { |
7127 | if (INTEL_INFO(dev)->num_pipes == 1) { |
7184 | dev_priv->display.update_wm = i845_update_wm; |
7128 | dev_priv->display.update_wm = i845_update_wm; |
7185 | dev_priv->display.get_fifo_size = i845_get_fifo_size; |
7129 | dev_priv->display.get_fifo_size = i845_get_fifo_size; |
7186 | } else { |
7130 | } else { |
7187 | dev_priv->display.update_wm = i9xx_update_wm; |
7131 | dev_priv->display.update_wm = i9xx_update_wm; |
7188 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
7132 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
7189 | } |
7133 | } |
7190 | 7134 | ||
7191 | if (IS_I85X(dev) || IS_I865G(dev)) |
7135 | if (IS_I85X(dev) || IS_I865G(dev)) |
7192 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; |
7136 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; |
7193 | else |
7137 | else |
7194 | dev_priv->display.init_clock_gating = i830_init_clock_gating; |
7138 | dev_priv->display.init_clock_gating = i830_init_clock_gating; |
7195 | } else { |
7139 | } else { |
7196 | DRM_ERROR("unexpected fall-through in intel_init_pm\n"); |
7140 | DRM_ERROR("unexpected fall-through in intel_init_pm\n"); |
7197 | } |
7141 | } |
7198 | } |
7142 | } |
7199 | 7143 | ||
7200 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val) |
7144 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val) |
7201 | { |
7145 | { |
7202 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
7146 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
7203 | 7147 | ||
7204 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { |
7148 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { |
7205 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); |
7149 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); |
7206 | return -EAGAIN; |
7150 | return -EAGAIN; |
7207 | } |
7151 | } |
7208 | 7152 | ||
7209 | I915_WRITE(GEN6_PCODE_DATA, *val); |
7153 | I915_WRITE(GEN6_PCODE_DATA, *val); |
7210 | I915_WRITE(GEN6_PCODE_DATA1, 0); |
7154 | I915_WRITE(GEN6_PCODE_DATA1, 0); |
7211 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
7155 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
7212 | 7156 | ||
7213 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, |
7157 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, |
7214 | 500)) { |
7158 | 500)) { |
7215 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); |
7159 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); |
7216 | return -ETIMEDOUT; |
7160 | return -ETIMEDOUT; |
7217 | } |
7161 | } |
7218 | 7162 | ||
7219 | *val = I915_READ(GEN6_PCODE_DATA); |
7163 | *val = I915_READ(GEN6_PCODE_DATA); |
7220 | I915_WRITE(GEN6_PCODE_DATA, 0); |
7164 | I915_WRITE(GEN6_PCODE_DATA, 0); |
7221 | 7165 | ||
7222 | return 0; |
7166 | return 0; |
7223 | } |
7167 | } |
7224 | 7168 | ||
7225 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u32 mbox, u32 val) |
7169 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u32 mbox, u32 val) |
7226 | { |
7170 | { |
7227 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
7171 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
7228 | 7172 | ||
7229 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { |
7173 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { |
7230 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); |
7174 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); |
7231 | return -EAGAIN; |
7175 | return -EAGAIN; |
7232 | } |
7176 | } |
7233 | 7177 | ||
7234 | I915_WRITE(GEN6_PCODE_DATA, val); |
7178 | I915_WRITE(GEN6_PCODE_DATA, val); |
7235 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
7179 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
7236 | 7180 | ||
7237 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, |
7181 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, |
7238 | 500)) { |
7182 | 500)) { |
7239 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); |
7183 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); |
7240 | return -ETIMEDOUT; |
7184 | return -ETIMEDOUT; |
7241 | } |
7185 | } |
7242 | 7186 | ||
7243 | I915_WRITE(GEN6_PCODE_DATA, 0); |
7187 | I915_WRITE(GEN6_PCODE_DATA, 0); |
7244 | 7188 | ||
7245 | return 0; |
7189 | return 0; |
7246 | } |
7190 | } |
7247 | 7191 | ||
7248 | static int vlv_gpu_freq_div(unsigned int czclk_freq) |
7192 | static int vlv_gpu_freq_div(unsigned int czclk_freq) |
7249 | { |
7193 | { |
7250 | switch (czclk_freq) { |
7194 | switch (czclk_freq) { |
7251 | case 200: |
7195 | case 200: |
7252 | return 10; |
7196 | return 10; |
7253 | case 267: |
7197 | case 267: |
7254 | return 12; |
7198 | return 12; |
7255 | case 320: |
7199 | case 320: |
7256 | case 333: |
7200 | case 333: |
7257 | return 16; |
7201 | return 16; |
7258 | case 400: |
7202 | case 400: |
7259 | return 20; |
7203 | return 20; |
7260 | default: |
7204 | default: |
7261 | return -1; |
7205 | return -1; |
7262 | } |
7206 | } |
7263 | } |
7207 | } |
7264 | 7208 | ||
7265 | static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7209 | static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7266 | { |
7210 | { |
7267 | int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->mem_freq, 4); |
7211 | int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000); |
7268 | 7212 | ||
7269 | div = vlv_gpu_freq_div(czclk_freq); |
7213 | div = vlv_gpu_freq_div(czclk_freq); |
7270 | if (div < 0) |
7214 | if (div < 0) |
7271 | return div; |
7215 | return div; |
7272 | 7216 | ||
7273 | return DIV_ROUND_CLOSEST(czclk_freq * (val + 6 - 0xbd), div); |
7217 | return DIV_ROUND_CLOSEST(czclk_freq * (val + 6 - 0xbd), div); |
7274 | } |
7218 | } |
7275 | 7219 | ||
7276 | static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val) |
7220 | static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val) |
7277 | { |
7221 | { |
7278 | int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->mem_freq, 4); |
7222 | int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000); |
7279 | 7223 | ||
7280 | mul = vlv_gpu_freq_div(czclk_freq); |
7224 | mul = vlv_gpu_freq_div(czclk_freq); |
7281 | if (mul < 0) |
7225 | if (mul < 0) |
7282 | return mul; |
7226 | return mul; |
7283 | 7227 | ||
7284 | return DIV_ROUND_CLOSEST(mul * val, czclk_freq) + 0xbd - 6; |
7228 | return DIV_ROUND_CLOSEST(mul * val, czclk_freq) + 0xbd - 6; |
7285 | } |
7229 | } |
7286 | 7230 | ||
7287 | static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7231 | static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7288 | { |
7232 | { |
7289 | int div, czclk_freq = dev_priv->rps.cz_freq; |
7233 | int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000); |
7290 | 7234 | ||
7291 | div = vlv_gpu_freq_div(czclk_freq) / 2; |
7235 | div = vlv_gpu_freq_div(czclk_freq) / 2; |
7292 | if (div < 0) |
7236 | if (div < 0) |
7293 | return div; |
7237 | return div; |
7294 | 7238 | ||
7295 | return DIV_ROUND_CLOSEST(czclk_freq * val, 2 * div) / 2; |
7239 | return DIV_ROUND_CLOSEST(czclk_freq * val, 2 * div) / 2; |
7296 | } |
7240 | } |
7297 | 7241 | ||
7298 | static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val) |
7242 | static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val) |
7299 | { |
7243 | { |
7300 | int mul, czclk_freq = dev_priv->rps.cz_freq; |
7244 | int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000); |
7301 | 7245 | ||
7302 | mul = vlv_gpu_freq_div(czclk_freq) / 2; |
7246 | mul = vlv_gpu_freq_div(czclk_freq) / 2; |
7303 | if (mul < 0) |
7247 | if (mul < 0) |
7304 | return mul; |
7248 | return mul; |
7305 | 7249 | ||
7306 | /* CHV needs even values */ |
7250 | /* CHV needs even values */ |
7307 | return DIV_ROUND_CLOSEST(val * 2 * mul, czclk_freq) * 2; |
7251 | return DIV_ROUND_CLOSEST(val * 2 * mul, czclk_freq) * 2; |
7308 | } |
7252 | } |
7309 | 7253 | ||
7310 | int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7254 | int intel_gpu_freq(struct drm_i915_private *dev_priv, int val) |
- | 7255 | { |
|
- | 7256 | if (IS_GEN9(dev_priv->dev)) |
|
7311 | { |
7257 | return DIV_ROUND_CLOSEST(val * GT_FREQUENCY_MULTIPLIER, |
- | 7258 | GEN9_FREQ_SCALER); |
|
- | 7259 | else if (IS_CHERRYVIEW(dev_priv->dev)) |
|
- | 7260 | return chv_gpu_freq(dev_priv, val); |
|
- | 7261 | else if (IS_VALLEYVIEW(dev_priv->dev)) |
|
- | 7262 | return byt_gpu_freq(dev_priv, val); |
|
- | 7263 | else |
|
- | 7264 | return val * GT_FREQUENCY_MULTIPLIER; |
|
- | 7265 | } |
|
- | 7266 | ||
- | 7267 | int intel_freq_opcode(struct drm_i915_private *dev_priv, int val) |
|
- | 7268 | { |
|
- | 7269 | if (IS_GEN9(dev_priv->dev)) |
|
7312 | int ret = -1; |
7270 | return DIV_ROUND_CLOSEST(val * GEN9_FREQ_SCALER, |
7313 | 7271 | GT_FREQUENCY_MULTIPLIER); |
|
7314 | if (IS_CHERRYVIEW(dev_priv->dev)) |
7272 | else if (IS_CHERRYVIEW(dev_priv->dev)) |
7315 | ret = chv_gpu_freq(dev_priv, val); |
7273 | return chv_freq_opcode(dev_priv, val); |
- | 7274 | else if (IS_VALLEYVIEW(dev_priv->dev)) |
|
- | 7275 | return byt_freq_opcode(dev_priv, val); |
|
- | 7276 | else |
|
- | 7277 | return DIV_ROUND_CLOSEST(val, GT_FREQUENCY_MULTIPLIER); |
|
- | 7278 | } |
|
- | 7279 | ||
- | 7280 | struct request_boost { |
|
- | 7281 | struct work_struct work; |
|
- | 7282 | struct drm_i915_gem_request *req; |
|
- | 7283 | }; |
|
- | 7284 | ||
- | 7285 | static void __intel_rps_boost_work(struct work_struct *work) |
|
- | 7286 | { |
|
- | 7287 | struct request_boost *boost = container_of(work, struct request_boost, work); |
|
- | 7288 | struct drm_i915_gem_request *req = boost->req; |
|
- | 7289 | ||
- | 7290 | if (!i915_gem_request_completed(req, true)) |
|
- | 7291 | gen6_rps_boost(to_i915(req->ring->dev), NULL, |
|
7316 | else if (IS_VALLEYVIEW(dev_priv->dev)) |
7292 | req->emitted_jiffies); |
7317 | ret = byt_gpu_freq(dev_priv, val); |
7293 | |
- | 7294 | i915_gem_request_unreference__unlocked(req); |
|
7318 | 7295 | kfree(boost); |
|
7319 | return ret; |
7296 | } |
7320 | } |
7297 | |
7321 | - | ||
7322 | int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val) |
7298 | void intel_queue_rps_boost_for_request(struct drm_device *dev, |
7323 | { |
- | |
7324 | int ret = -1; |
7299 | struct drm_i915_gem_request *req) |
- | 7300 | { |
|
7325 | 7301 | struct request_boost *boost; |
|
- | 7302 | ||
- | 7303 | if (req == NULL || INTEL_INFO(dev)->gen < 6) |
|
- | 7304 | return; |
|
- | 7305 | ||
- | 7306 | if (i915_gem_request_completed(req, true)) |
|
- | 7307 | return; |
|
- | 7308 | ||
- | 7309 | boost = kmalloc(sizeof(*boost), GFP_ATOMIC); |
|
- | 7310 | if (boost == NULL) |
|
- | 7311 | return; |
|
7326 | if (IS_CHERRYVIEW(dev_priv->dev)) |
7312 | |
7327 | ret = chv_freq_opcode(dev_priv, val); |
7313 | i915_gem_request_reference(req); |
7328 | else if (IS_VALLEYVIEW(dev_priv->dev)) |
7314 | boost->req = req; |
7329 | ret = byt_freq_opcode(dev_priv, val); |
7315 | |
7330 | 7316 | INIT_WORK(&boost->work, __intel_rps_boost_work); |
|
7331 | return ret; |
7317 | queue_work(to_i915(dev)->wq, &boost->work); |
7332 | } |
7318 | } |
7333 | 7319 | ||
7334 | void intel_pm_setup(struct drm_device *dev) |
7320 | void intel_pm_setup(struct drm_device *dev) |
7335 | { |
7321 | { |
7336 | struct drm_i915_private *dev_priv = dev->dev_private; |
7322 | struct drm_i915_private *dev_priv = dev->dev_private; |
7337 | 7323 | ||
7338 | mutex_init(&dev_priv->rps.hw_lock); |
7324 | mutex_init(&dev_priv->rps.hw_lock); |
- | 7325 | spin_lock_init(&dev_priv->rps.client_lock); |
|
7339 | 7326 | ||
7340 | INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work, |
7327 | INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work, |
7341 | intel_gen6_powersave_work); |
7328 | intel_gen6_powersave_work); |
- | 7329 | INIT_LIST_HEAD(&dev_priv->rps.clients); |
|
- | 7330 | INIT_LIST_HEAD(&dev_priv->rps.semaphores.link); |
|
- | 7331 | INIT_LIST_HEAD(&dev_priv->rps.mmioflips.link); |
|
7342 | 7332 | ||
7343 | dev_priv->pm.suspended = false; |
7333 | dev_priv->pm.suspended = false; |
7344 | }>>>>><>><>>>>>>>><>><>><>>>>>>>>>>=> |
7334 | }>>>>>><>>>>>>><>><>><>>>>>>>>>>>=> |
7345 | >=>>>>=>><>><>><>>><> |
7335 | >=>>>>=>><>><>><>>><> |
7346 | >< |
7336 | >< |
7347 | >>=>=>>>=>>>>>>=>=>><>><>><>><>><>=>><>><>=>><>><>><>><>=>=>=>=>=>=>=>=>=>=>=>=>=>=>>>8)><8)>><>><>><>><>>><>><>><>><>><>><>><>>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>=>=>>=>=>>>=> |
7337 | >>=>=>>>=>=>>>>>>=>=>><>><>><>><>><>=>><>><>=>><>><>><>><>=>=>=>=>=>=>=>=>=>=>=>=>=>=>>>8)><8)>><>><>><>><>>>><>><>>>>>><>>>>>>>>><>><>><>><>><>=>=>><>><>>=>2) |
7348 | >< |
7338 | |
7349 | > |
7339 | static><2) |
7350 | >< |
7340 | |
7351 | >><>><>>>2) |
7341 | static>1) |
7352 | 7342 | #define><1) |
|
7353 | /*><2) |
7343 | #define>0) |
7354 | 7344 | #define><0) |
|
7355 | /*>1) |
7345 | #define> |
7356 | #define><1) |
- | |
7357 | #define>0) |
- | |
7358 | #define><0) |
- | |
7359 | #define> |
- |