Rev 3482 | Rev 4104 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
3031 | serge | 1 | /* |
2 | * Copyright © 2012 Intel Corporation |
||
3 | * |
||
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
5 | * copy of this software and associated documentation files (the "Software"), |
||
6 | * to deal in the Software without restriction, including without limitation |
||
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 |
||
9 | * Software is furnished to do so, subject to the following conditions: |
||
10 | * |
||
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 |
||
13 | * Software. |
||
14 | * |
||
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, |
||
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 |
||
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 |
||
21 | * IN THE SOFTWARE. |
||
22 | * |
||
23 | * Authors: |
||
24 | * Eugeni Dodonov |
||
25 | * |
||
26 | */ |
||
27 | |||
28 | //#include |
||
29 | #include "i915_drv.h" |
||
30 | #include "intel_drv.h" |
||
31 | #include |
||
32 | //#include "../../../platform/x86/intel_ips.h" |
||
33 | #include |
||
34 | |||
35 | #define FORCEWAKE_ACK_TIMEOUT_MS 2 |
||
36 | |||
37 | #define assert_spin_locked(x) |
||
38 | |||
39 | void getrawmonotonic(struct timespec *ts); |
||
40 | void set_normalized_timespec(struct timespec *ts, time_t sec, long nsec); |
||
41 | |||
42 | static inline struct timespec timespec_sub(struct timespec lhs, |
||
43 | struct timespec rhs) |
||
44 | { |
||
45 | struct timespec ts_delta; |
||
46 | set_normalized_timespec(&ts_delta, lhs.tv_sec - rhs.tv_sec, |
||
47 | lhs.tv_nsec - rhs.tv_nsec); |
||
48 | return ts_delta; |
||
49 | } |
||
50 | |||
51 | |||
52 | /* FBC, or Frame Buffer Compression, is a technique employed to compress the |
||
53 | * framebuffer contents in-memory, aiming at reducing the required bandwidth |
||
54 | * during in-memory transfers and, therefore, reduce the power packet. |
||
55 | * |
||
56 | * The benefits of FBC are mostly visible with solid backgrounds and |
||
57 | * variation-less patterns. |
||
58 | * |
||
59 | * FBC-related functionality can be enabled by the means of the |
||
60 | * i915.i915_enable_fbc parameter |
||
61 | */ |
||
62 | |||
3243 | Serge | 63 | static bool intel_crtc_active(struct drm_crtc *crtc) |
64 | { |
||
65 | /* Be paranoid as we can arrive here with only partial |
||
66 | * state retrieved from the hardware during setup. |
||
67 | */ |
||
68 | return to_intel_crtc(crtc)->active && crtc->fb && crtc->mode.clock; |
||
69 | } |
||
70 | |||
3031 | serge | 71 | static void i8xx_disable_fbc(struct drm_device *dev) |
72 | { |
||
73 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
74 | u32 fbc_ctl; |
||
75 | |||
76 | /* Disable compression */ |
||
77 | fbc_ctl = I915_READ(FBC_CONTROL); |
||
78 | if ((fbc_ctl & FBC_CTL_EN) == 0) |
||
79 | return; |
||
80 | |||
81 | fbc_ctl &= ~FBC_CTL_EN; |
||
82 | I915_WRITE(FBC_CONTROL, fbc_ctl); |
||
83 | |||
84 | /* Wait for compressing bit to clear */ |
||
85 | if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) { |
||
86 | DRM_DEBUG_KMS("FBC idle timed out\n"); |
||
87 | return; |
||
88 | } |
||
89 | |||
90 | DRM_DEBUG_KMS("disabled FBC\n"); |
||
91 | } |
||
92 | |||
93 | static void i8xx_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
||
94 | { |
||
95 | struct drm_device *dev = crtc->dev; |
||
96 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
97 | struct drm_framebuffer *fb = crtc->fb; |
||
98 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); |
||
99 | struct drm_i915_gem_object *obj = intel_fb->obj; |
||
100 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
||
101 | int cfb_pitch; |
||
102 | int plane, i; |
||
103 | u32 fbc_ctl, fbc_ctl2; |
||
104 | |||
105 | cfb_pitch = dev_priv->cfb_size / FBC_LL_SIZE; |
||
106 | if (fb->pitches[0] < cfb_pitch) |
||
107 | cfb_pitch = fb->pitches[0]; |
||
108 | |||
109 | /* FBC_CTL wants 64B units */ |
||
110 | cfb_pitch = (cfb_pitch / 64) - 1; |
||
111 | plane = intel_crtc->plane == 0 ? FBC_CTL_PLANEA : FBC_CTL_PLANEB; |
||
112 | |||
113 | /* Clear old tags */ |
||
114 | for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++) |
||
115 | I915_WRITE(FBC_TAG + (i * 4), 0); |
||
116 | |||
117 | /* Set it up... */ |
||
118 | fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE; |
||
119 | fbc_ctl2 |= plane; |
||
120 | I915_WRITE(FBC_CONTROL2, fbc_ctl2); |
||
121 | I915_WRITE(FBC_FENCE_OFF, crtc->y); |
||
122 | |||
123 | /* enable it... */ |
||
124 | fbc_ctl = FBC_CTL_EN | FBC_CTL_PERIODIC; |
||
125 | if (IS_I945GM(dev)) |
||
126 | fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */ |
||
127 | fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT; |
||
128 | fbc_ctl |= (interval & 0x2fff) << FBC_CTL_INTERVAL_SHIFT; |
||
129 | fbc_ctl |= obj->fence_reg; |
||
130 | I915_WRITE(FBC_CONTROL, fbc_ctl); |
||
131 | |||
132 | DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %d, ", |
||
133 | cfb_pitch, crtc->y, intel_crtc->plane); |
||
134 | } |
||
135 | |||
136 | static bool i8xx_fbc_enabled(struct drm_device *dev) |
||
137 | { |
||
138 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
139 | |||
140 | return I915_READ(FBC_CONTROL) & FBC_CTL_EN; |
||
141 | } |
||
142 | |||
143 | static void g4x_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
||
144 | { |
||
145 | struct drm_device *dev = crtc->dev; |
||
146 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
147 | struct drm_framebuffer *fb = crtc->fb; |
||
148 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); |
||
149 | struct drm_i915_gem_object *obj = intel_fb->obj; |
||
150 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
||
151 | int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB; |
||
152 | unsigned long stall_watermark = 200; |
||
153 | u32 dpfc_ctl; |
||
154 | |||
155 | dpfc_ctl = plane | DPFC_SR_EN | DPFC_CTL_LIMIT_1X; |
||
156 | dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg; |
||
157 | I915_WRITE(DPFC_CHICKEN, DPFC_HT_MODIFY); |
||
158 | |||
159 | I915_WRITE(DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN | |
||
160 | (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) | |
||
161 | (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT)); |
||
162 | I915_WRITE(DPFC_FENCE_YOFF, crtc->y); |
||
163 | |||
164 | /* enable it... */ |
||
165 | I915_WRITE(DPFC_CONTROL, I915_READ(DPFC_CONTROL) | DPFC_CTL_EN); |
||
166 | |||
167 | DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc->plane); |
||
168 | } |
||
169 | |||
170 | static void g4x_disable_fbc(struct drm_device *dev) |
||
171 | { |
||
172 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
173 | u32 dpfc_ctl; |
||
174 | |||
175 | /* Disable compression */ |
||
176 | dpfc_ctl = I915_READ(DPFC_CONTROL); |
||
177 | if (dpfc_ctl & DPFC_CTL_EN) { |
||
178 | dpfc_ctl &= ~DPFC_CTL_EN; |
||
179 | I915_WRITE(DPFC_CONTROL, dpfc_ctl); |
||
180 | |||
181 | DRM_DEBUG_KMS("disabled FBC\n"); |
||
182 | } |
||
183 | } |
||
184 | |||
185 | static bool g4x_fbc_enabled(struct drm_device *dev) |
||
186 | { |
||
187 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
188 | |||
189 | return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN; |
||
190 | } |
||
191 | |||
192 | static void sandybridge_blit_fbc_update(struct drm_device *dev) |
||
193 | { |
||
194 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
195 | u32 blt_ecoskpd; |
||
196 | |||
197 | /* Make sure blitter notifies FBC of writes */ |
||
198 | gen6_gt_force_wake_get(dev_priv); |
||
199 | blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD); |
||
200 | blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY << |
||
201 | GEN6_BLITTER_LOCK_SHIFT; |
||
202 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); |
||
203 | blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY; |
||
204 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); |
||
205 | blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY << |
||
206 | GEN6_BLITTER_LOCK_SHIFT); |
||
207 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); |
||
208 | POSTING_READ(GEN6_BLITTER_ECOSKPD); |
||
209 | gen6_gt_force_wake_put(dev_priv); |
||
210 | } |
||
211 | |||
212 | static void ironlake_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
||
213 | { |
||
214 | struct drm_device *dev = crtc->dev; |
||
215 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
216 | struct drm_framebuffer *fb = crtc->fb; |
||
217 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); |
||
218 | struct drm_i915_gem_object *obj = intel_fb->obj; |
||
219 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
||
220 | int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB; |
||
221 | unsigned long stall_watermark = 200; |
||
222 | u32 dpfc_ctl; |
||
223 | |||
224 | dpfc_ctl = I915_READ(ILK_DPFC_CONTROL); |
||
225 | dpfc_ctl &= DPFC_RESERVED; |
||
226 | dpfc_ctl |= (plane | DPFC_CTL_LIMIT_1X); |
||
227 | /* Set persistent mode for front-buffer rendering, ala X. */ |
||
228 | dpfc_ctl |= DPFC_CTL_PERSISTENT_MODE; |
||
229 | dpfc_ctl |= (DPFC_CTL_FENCE_EN | obj->fence_reg); |
||
230 | I915_WRITE(ILK_DPFC_CHICKEN, DPFC_HT_MODIFY); |
||
231 | |||
232 | I915_WRITE(ILK_DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN | |
||
233 | (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) | |
||
234 | (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT)); |
||
235 | I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y); |
||
236 | I915_WRITE(ILK_FBC_RT_BASE, obj->gtt_offset | ILK_FBC_RT_VALID); |
||
237 | /* enable it... */ |
||
238 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN); |
||
239 | |||
240 | if (IS_GEN6(dev)) { |
||
241 | I915_WRITE(SNB_DPFC_CTL_SA, |
||
242 | SNB_CPU_FENCE_ENABLE | obj->fence_reg); |
||
243 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y); |
||
244 | sandybridge_blit_fbc_update(dev); |
||
245 | } |
||
246 | |||
247 | DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc->plane); |
||
248 | } |
||
249 | |||
250 | static void ironlake_disable_fbc(struct drm_device *dev) |
||
251 | { |
||
252 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
253 | u32 dpfc_ctl; |
||
254 | |||
255 | /* Disable compression */ |
||
256 | dpfc_ctl = I915_READ(ILK_DPFC_CONTROL); |
||
257 | if (dpfc_ctl & DPFC_CTL_EN) { |
||
258 | dpfc_ctl &= ~DPFC_CTL_EN; |
||
259 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl); |
||
260 | |||
261 | DRM_DEBUG_KMS("disabled FBC\n"); |
||
262 | } |
||
263 | } |
||
264 | |||
265 | static bool ironlake_fbc_enabled(struct drm_device *dev) |
||
266 | { |
||
267 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
268 | |||
269 | return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN; |
||
270 | } |
||
271 | |||
272 | bool intel_fbc_enabled(struct drm_device *dev) |
||
273 | { |
||
274 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
275 | |||
276 | if (!dev_priv->display.fbc_enabled) |
||
277 | return false; |
||
278 | |||
279 | return dev_priv->display.fbc_enabled(dev); |
||
280 | } |
||
281 | |||
282 | static void intel_fbc_work_fn(struct work_struct *__work) |
||
283 | { |
||
284 | struct intel_fbc_work *work = |
||
285 | container_of(to_delayed_work(__work), |
||
286 | struct intel_fbc_work, work); |
||
287 | struct drm_device *dev = work->crtc->dev; |
||
288 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
289 | |||
290 | mutex_lock(&dev->struct_mutex); |
||
291 | if (work == dev_priv->fbc_work) { |
||
292 | /* Double check that we haven't switched fb without cancelling |
||
293 | * the prior work. |
||
294 | */ |
||
295 | if (work->crtc->fb == work->fb) { |
||
296 | dev_priv->display.enable_fbc(work->crtc, |
||
297 | work->interval); |
||
298 | |||
299 | dev_priv->cfb_plane = to_intel_crtc(work->crtc)->plane; |
||
300 | dev_priv->cfb_fb = work->crtc->fb->base.id; |
||
301 | dev_priv->cfb_y = work->crtc->y; |
||
302 | } |
||
303 | |||
304 | dev_priv->fbc_work = NULL; |
||
305 | } |
||
306 | mutex_unlock(&dev->struct_mutex); |
||
307 | |||
308 | kfree(work); |
||
309 | } |
||
310 | |||
311 | static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv) |
||
312 | { |
||
313 | if (dev_priv->fbc_work == NULL) |
||
314 | return; |
||
315 | |||
316 | DRM_DEBUG_KMS("cancelling pending FBC enable\n"); |
||
317 | |||
318 | /* Synchronisation is provided by struct_mutex and checking of |
||
319 | * dev_priv->fbc_work, so we can perform the cancellation |
||
320 | * entirely asynchronously. |
||
321 | */ |
||
3482 | Serge | 322 | // if (cancel_delayed_work(&dev_priv->fbc_work->work)) |
3031 | serge | 323 | /* tasklet was killed before being run, clean up */ |
3482 | Serge | 324 | // kfree(dev_priv->fbc_work); |
3031 | serge | 325 | |
326 | /* Mark the work as no longer wanted so that if it does |
||
327 | * wake-up (because the work was already running and waiting |
||
328 | * for our mutex), it will discover that is no longer |
||
329 | * necessary to run. |
||
330 | */ |
||
331 | dev_priv->fbc_work = NULL; |
||
332 | } |
||
333 | |||
334 | void intel_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
||
335 | { |
||
336 | struct intel_fbc_work *work; |
||
337 | struct drm_device *dev = crtc->dev; |
||
338 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
339 | |||
3482 | Serge | 340 | if (!dev_priv->display.enable_fbc) |
3031 | serge | 341 | return; |
3482 | Serge | 342 | |
3031 | serge | 343 | intel_cancel_fbc_work(dev_priv); |
344 | |||
345 | work = kzalloc(sizeof *work, GFP_KERNEL); |
||
346 | if (work == NULL) { |
||
347 | dev_priv->display.enable_fbc(crtc, interval); |
||
348 | return; |
||
349 | } |
||
350 | |||
351 | work->crtc = crtc; |
||
352 | work->fb = crtc->fb; |
||
353 | work->interval = interval; |
||
354 | INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn); |
||
355 | |||
356 | dev_priv->fbc_work = work; |
||
357 | |||
358 | DRM_DEBUG_KMS("scheduling delayed FBC enable\n"); |
||
359 | |||
360 | /* Delay the actual enabling to let pageflipping cease and the |
||
361 | * display to settle before starting the compression. Note that |
||
362 | * this delay also serves a second purpose: it allows for a |
||
363 | * vblank to pass after disabling the FBC before we attempt |
||
364 | * to modify the control registers. |
||
365 | * |
||
366 | * A more complicated solution would involve tracking vblanks |
||
367 | * following the termination of the page-flipping sequence |
||
368 | * and indeed performing the enable as a co-routine and not |
||
369 | * waiting synchronously upon the vblank. |
||
370 | */ |
||
371 | schedule_delayed_work(&work->work, msecs_to_jiffies(50)); |
||
372 | } |
||
373 | |||
374 | void intel_disable_fbc(struct drm_device *dev) |
||
375 | { |
||
376 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
377 | |||
3482 | Serge | 378 | intel_cancel_fbc_work(dev_priv); |
3031 | serge | 379 | |
3482 | Serge | 380 | if (!dev_priv->display.disable_fbc) |
381 | return; |
||
3031 | serge | 382 | |
3482 | Serge | 383 | dev_priv->display.disable_fbc(dev); |
3031 | serge | 384 | dev_priv->cfb_plane = -1; |
385 | } |
||
386 | |||
387 | /** |
||
388 | * intel_update_fbc - enable/disable FBC as needed |
||
389 | * @dev: the drm_device |
||
390 | * |
||
391 | * Set up the framebuffer compression hardware at mode set time. We |
||
392 | * enable it if possible: |
||
393 | * - plane A only (on pre-965) |
||
394 | * - no pixel mulitply/line duplication |
||
395 | * - no alpha buffer discard |
||
396 | * - no dual wide |
||
397 | * - framebuffer <= 2048 in width, 1536 in height |
||
398 | * |
||
399 | * We can't assume that any compression will take place (worst case), |
||
400 | * so the compressed buffer has to be the same size as the uncompressed |
||
401 | * one. It also must reside (along with the line length buffer) in |
||
402 | * stolen memory. |
||
403 | * |
||
404 | * We need to enable/disable FBC on a global basis. |
||
405 | */ |
||
406 | void intel_update_fbc(struct drm_device *dev) |
||
407 | { |
||
408 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
409 | struct drm_crtc *crtc = NULL, *tmp_crtc; |
||
410 | struct intel_crtc *intel_crtc; |
||
411 | struct drm_framebuffer *fb; |
||
412 | struct intel_framebuffer *intel_fb; |
||
413 | struct drm_i915_gem_object *obj; |
||
414 | int enable_fbc; |
||
415 | |||
3482 | Serge | 416 | ENTER(); |
417 | |||
3031 | serge | 418 | if (!i915_powersave) |
419 | return; |
||
420 | |||
421 | if (!I915_HAS_FBC(dev)) |
||
422 | return; |
||
423 | |||
424 | /* |
||
425 | * If FBC is already on, we just have to verify that we can |
||
426 | * keep it that way... |
||
427 | * Need to disable if: |
||
428 | * - more than one pipe is active |
||
429 | * - changing FBC params (stride, fence, mode) |
||
430 | * - new fb is too large to fit in compressed buffer |
||
431 | * - going to an unsupported config (interlace, pixel multiply, etc.) |
||
432 | */ |
||
433 | list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) { |
||
3243 | Serge | 434 | if (intel_crtc_active(tmp_crtc) && |
435 | !to_intel_crtc(tmp_crtc)->primary_disabled) { |
||
3031 | serge | 436 | if (crtc) { |
437 | DRM_DEBUG_KMS("more than one pipe active, disabling compression\n"); |
||
438 | dev_priv->no_fbc_reason = FBC_MULTIPLE_PIPES; |
||
439 | goto out_disable; |
||
440 | } |
||
441 | crtc = tmp_crtc; |
||
442 | } |
||
443 | } |
||
444 | |||
445 | if (!crtc || crtc->fb == NULL) { |
||
446 | DRM_DEBUG_KMS("no output, disabling\n"); |
||
447 | dev_priv->no_fbc_reason = FBC_NO_OUTPUT; |
||
448 | goto out_disable; |
||
449 | } |
||
450 | |||
451 | intel_crtc = to_intel_crtc(crtc); |
||
452 | fb = crtc->fb; |
||
453 | intel_fb = to_intel_framebuffer(fb); |
||
454 | obj = intel_fb->obj; |
||
455 | |||
456 | enable_fbc = i915_enable_fbc; |
||
457 | if (enable_fbc < 0) { |
||
458 | DRM_DEBUG_KMS("fbc set to per-chip default\n"); |
||
459 | enable_fbc = 1; |
||
460 | if (INTEL_INFO(dev)->gen <= 6) |
||
461 | enable_fbc = 0; |
||
462 | } |
||
463 | if (!enable_fbc) { |
||
464 | DRM_DEBUG_KMS("fbc disabled per module param\n"); |
||
465 | dev_priv->no_fbc_reason = FBC_MODULE_PARAM; |
||
466 | goto out_disable; |
||
467 | } |
||
468 | if ((crtc->mode.flags & DRM_MODE_FLAG_INTERLACE) || |
||
469 | (crtc->mode.flags & DRM_MODE_FLAG_DBLSCAN)) { |
||
470 | DRM_DEBUG_KMS("mode incompatible with compression, " |
||
471 | "disabling\n"); |
||
472 | dev_priv->no_fbc_reason = FBC_UNSUPPORTED_MODE; |
||
473 | goto out_disable; |
||
474 | } |
||
475 | if ((crtc->mode.hdisplay > 2048) || |
||
476 | (crtc->mode.vdisplay > 1536)) { |
||
477 | DRM_DEBUG_KMS("mode too large for compression, disabling\n"); |
||
478 | dev_priv->no_fbc_reason = FBC_MODE_TOO_LARGE; |
||
479 | goto out_disable; |
||
480 | } |
||
481 | if ((IS_I915GM(dev) || IS_I945GM(dev)) && intel_crtc->plane != 0) { |
||
482 | DRM_DEBUG_KMS("plane not 0, disabling compression\n"); |
||
483 | dev_priv->no_fbc_reason = FBC_BAD_PLANE; |
||
484 | goto out_disable; |
||
485 | } |
||
486 | |||
487 | /* The use of a CPU fence is mandatory in order to detect writes |
||
488 | * by the CPU to the scanout and trigger updates to the FBC. |
||
489 | */ |
||
490 | if (obj->tiling_mode != I915_TILING_X || |
||
491 | obj->fence_reg == I915_FENCE_REG_NONE) { |
||
492 | DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n"); |
||
493 | dev_priv->no_fbc_reason = FBC_NOT_TILED; |
||
494 | goto out_disable; |
||
495 | } |
||
496 | |||
497 | /* If the kernel debugger is active, always disable compression */ |
||
498 | if (in_dbg_master()) |
||
499 | goto out_disable; |
||
500 | |||
3480 | Serge | 501 | if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) { |
502 | DRM_INFO("not enough stolen space for compressed buffer (need %zd bytes), disabling\n", intel_fb->obj->base.size); |
||
503 | DRM_INFO("hint: you may be able to increase stolen memory size in the BIOS to avoid this\n"); |
||
504 | DRM_DEBUG_KMS("framebuffer too large, disabling compression\n"); |
||
505 | dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL; |
||
506 | goto out_disable; |
||
507 | } |
||
508 | |||
3031 | serge | 509 | /* If the scanout has not changed, don't modify the FBC settings. |
510 | * Note that we make the fundamental assumption that the fb->obj |
||
511 | * cannot be unpinned (and have its GTT offset and fence revoked) |
||
512 | * without first being decoupled from the scanout and FBC disabled. |
||
513 | */ |
||
514 | if (dev_priv->cfb_plane == intel_crtc->plane && |
||
515 | dev_priv->cfb_fb == fb->base.id && |
||
516 | dev_priv->cfb_y == crtc->y) |
||
517 | return; |
||
518 | |||
519 | if (intel_fbc_enabled(dev)) { |
||
520 | /* We update FBC along two paths, after changing fb/crtc |
||
521 | * configuration (modeswitching) and after page-flipping |
||
522 | * finishes. For the latter, we know that not only did |
||
523 | * we disable the FBC at the start of the page-flip |
||
524 | * sequence, but also more than one vblank has passed. |
||
525 | * |
||
526 | * For the former case of modeswitching, it is possible |
||
527 | * to switch between two FBC valid configurations |
||
528 | * instantaneously so we do need to disable the FBC |
||
529 | * before we can modify its control registers. We also |
||
530 | * have to wait for the next vblank for that to take |
||
531 | * effect. However, since we delay enabling FBC we can |
||
532 | * assume that a vblank has passed since disabling and |
||
533 | * that we can safely alter the registers in the deferred |
||
534 | * callback. |
||
535 | * |
||
536 | * In the scenario that we go from a valid to invalid |
||
537 | * and then back to valid FBC configuration we have |
||
538 | * no strict enforcement that a vblank occurred since |
||
539 | * disabling the FBC. However, along all current pipe |
||
540 | * disabling paths we do need to wait for a vblank at |
||
541 | * some point. And we wait before enabling FBC anyway. |
||
542 | */ |
||
543 | DRM_DEBUG_KMS("disabling active FBC for update\n"); |
||
544 | intel_disable_fbc(dev); |
||
545 | } |
||
546 | |||
547 | intel_enable_fbc(crtc, 500); |
||
3482 | Serge | 548 | LEAVE(); |
549 | |||
3031 | serge | 550 | return; |
551 | |||
552 | out_disable: |
||
553 | /* Multiple disables should be harmless */ |
||
554 | if (intel_fbc_enabled(dev)) { |
||
555 | DRM_DEBUG_KMS("unsupported config, disabling FBC\n"); |
||
556 | intel_disable_fbc(dev); |
||
557 | } |
||
3480 | Serge | 558 | i915_gem_stolen_cleanup_compression(dev); |
3482 | Serge | 559 | LEAVE(); |
3031 | serge | 560 | } |
561 | |||
562 | static void i915_pineview_get_mem_freq(struct drm_device *dev) |
||
563 | { |
||
564 | drm_i915_private_t *dev_priv = dev->dev_private; |
||
565 | u32 tmp; |
||
566 | |||
567 | tmp = I915_READ(CLKCFG); |
||
568 | |||
569 | switch (tmp & CLKCFG_FSB_MASK) { |
||
570 | case CLKCFG_FSB_533: |
||
571 | dev_priv->fsb_freq = 533; /* 133*4 */ |
||
572 | break; |
||
573 | case CLKCFG_FSB_800: |
||
574 | dev_priv->fsb_freq = 800; /* 200*4 */ |
||
575 | break; |
||
576 | case CLKCFG_FSB_667: |
||
577 | dev_priv->fsb_freq = 667; /* 167*4 */ |
||
578 | break; |
||
579 | case CLKCFG_FSB_400: |
||
580 | dev_priv->fsb_freq = 400; /* 100*4 */ |
||
581 | break; |
||
582 | } |
||
583 | |||
584 | switch (tmp & CLKCFG_MEM_MASK) { |
||
585 | case CLKCFG_MEM_533: |
||
586 | dev_priv->mem_freq = 533; |
||
587 | break; |
||
588 | case CLKCFG_MEM_667: |
||
589 | dev_priv->mem_freq = 667; |
||
590 | break; |
||
591 | case CLKCFG_MEM_800: |
||
592 | dev_priv->mem_freq = 800; |
||
593 | break; |
||
594 | } |
||
595 | |||
596 | /* detect pineview DDR3 setting */ |
||
597 | tmp = I915_READ(CSHRDDR3CTL); |
||
598 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; |
||
599 | } |
||
600 | |||
601 | static void i915_ironlake_get_mem_freq(struct drm_device *dev) |
||
602 | { |
||
603 | drm_i915_private_t *dev_priv = dev->dev_private; |
||
604 | u16 ddrpll, csipll; |
||
605 | |||
606 | ddrpll = I915_READ16(DDRMPLL1); |
||
607 | csipll = I915_READ16(CSIPLL0); |
||
608 | |||
609 | switch (ddrpll & 0xff) { |
||
610 | case 0xc: |
||
611 | dev_priv->mem_freq = 800; |
||
612 | break; |
||
613 | case 0x10: |
||
614 | dev_priv->mem_freq = 1066; |
||
615 | break; |
||
616 | case 0x14: |
||
617 | dev_priv->mem_freq = 1333; |
||
618 | break; |
||
619 | case 0x18: |
||
620 | dev_priv->mem_freq = 1600; |
||
621 | break; |
||
622 | default: |
||
623 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", |
||
624 | ddrpll & 0xff); |
||
625 | dev_priv->mem_freq = 0; |
||
626 | break; |
||
627 | } |
||
628 | |||
629 | dev_priv->ips.r_t = dev_priv->mem_freq; |
||
630 | |||
631 | switch (csipll & 0x3ff) { |
||
632 | case 0x00c: |
||
633 | dev_priv->fsb_freq = 3200; |
||
634 | break; |
||
635 | case 0x00e: |
||
636 | dev_priv->fsb_freq = 3733; |
||
637 | break; |
||
638 | case 0x010: |
||
639 | dev_priv->fsb_freq = 4266; |
||
640 | break; |
||
641 | case 0x012: |
||
642 | dev_priv->fsb_freq = 4800; |
||
643 | break; |
||
644 | case 0x014: |
||
645 | dev_priv->fsb_freq = 5333; |
||
646 | break; |
||
647 | case 0x016: |
||
648 | dev_priv->fsb_freq = 5866; |
||
649 | break; |
||
650 | case 0x018: |
||
651 | dev_priv->fsb_freq = 6400; |
||
652 | break; |
||
653 | default: |
||
654 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", |
||
655 | csipll & 0x3ff); |
||
656 | dev_priv->fsb_freq = 0; |
||
657 | break; |
||
658 | } |
||
659 | |||
660 | if (dev_priv->fsb_freq == 3200) { |
||
661 | dev_priv->ips.c_m = 0; |
||
662 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
||
663 | dev_priv->ips.c_m = 1; |
||
664 | } else { |
||
665 | dev_priv->ips.c_m = 2; |
||
666 | } |
||
667 | } |
||
668 | |||
669 | static const struct cxsr_latency cxsr_latency_table[] = { |
||
670 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ |
||
671 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ |
||
672 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ |
||
673 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ |
||
674 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ |
||
675 | |||
676 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ |
||
677 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ |
||
678 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ |
||
679 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ |
||
680 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ |
||
681 | |||
682 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ |
||
683 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ |
||
684 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ |
||
685 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ |
||
686 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ |
||
687 | |||
688 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ |
||
689 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ |
||
690 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ |
||
691 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ |
||
692 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ |
||
693 | |||
694 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ |
||
695 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ |
||
696 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ |
||
697 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ |
||
698 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ |
||
699 | |||
700 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ |
||
701 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ |
||
702 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ |
||
703 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ |
||
704 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ |
||
705 | }; |
||
706 | |||
707 | static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop, |
||
708 | int is_ddr3, |
||
709 | int fsb, |
||
710 | int mem) |
||
711 | { |
||
712 | const struct cxsr_latency *latency; |
||
713 | int i; |
||
714 | |||
715 | if (fsb == 0 || mem == 0) |
||
716 | return NULL; |
||
717 | |||
718 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { |
||
719 | latency = &cxsr_latency_table[i]; |
||
720 | if (is_desktop == latency->is_desktop && |
||
721 | is_ddr3 == latency->is_ddr3 && |
||
722 | fsb == latency->fsb_freq && mem == latency->mem_freq) |
||
723 | return latency; |
||
724 | } |
||
725 | |||
726 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); |
||
727 | |||
728 | return NULL; |
||
729 | } |
||
730 | |||
731 | static void pineview_disable_cxsr(struct drm_device *dev) |
||
732 | { |
||
733 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
734 | |||
735 | /* deactivate cxsr */ |
||
736 | I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN); |
||
737 | } |
||
738 | |||
739 | /* |
||
740 | * Latency for FIFO fetches is dependent on several factors: |
||
741 | * - memory configuration (speed, channels) |
||
742 | * - chipset |
||
743 | * - current MCH state |
||
744 | * It can be fairly high in some situations, so here we assume a fairly |
||
745 | * pessimal value. It's a tradeoff between extra memory fetches (if we |
||
746 | * set this value too high, the FIFO will fetch frequently to stay full) |
||
747 | * and power consumption (set it too low to save power and we might see |
||
748 | * FIFO underruns and display "flicker"). |
||
749 | * |
||
750 | * A value of 5us seems to be a good balance; safe for very low end |
||
751 | * platforms but not overly aggressive on lower latency configs. |
||
752 | */ |
||
753 | static const int latency_ns = 5000; |
||
754 | |||
755 | static int i9xx_get_fifo_size(struct drm_device *dev, int plane) |
||
756 | { |
||
757 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
758 | uint32_t dsparb = I915_READ(DSPARB); |
||
759 | int size; |
||
760 | |||
761 | size = dsparb & 0x7f; |
||
762 | if (plane) |
||
763 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; |
||
764 | |||
765 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
||
766 | plane ? "B" : "A", size); |
||
767 | |||
768 | return size; |
||
769 | } |
||
770 | |||
771 | static int i85x_get_fifo_size(struct drm_device *dev, int plane) |
||
772 | { |
||
773 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
774 | uint32_t dsparb = I915_READ(DSPARB); |
||
775 | int size; |
||
776 | |||
777 | size = dsparb & 0x1ff; |
||
778 | if (plane) |
||
779 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; |
||
780 | size >>= 1; /* Convert to cachelines */ |
||
781 | |||
782 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
||
783 | plane ? "B" : "A", size); |
||
784 | |||
785 | return size; |
||
786 | } |
||
787 | |||
788 | static int i845_get_fifo_size(struct drm_device *dev, int plane) |
||
789 | { |
||
790 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
791 | uint32_t dsparb = I915_READ(DSPARB); |
||
792 | int size; |
||
793 | |||
794 | size = dsparb & 0x7f; |
||
795 | size >>= 2; /* Convert to cachelines */ |
||
796 | |||
797 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
||
798 | plane ? "B" : "A", |
||
799 | size); |
||
800 | |||
801 | return size; |
||
802 | } |
||
803 | |||
804 | static int i830_get_fifo_size(struct drm_device *dev, int plane) |
||
805 | { |
||
806 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
807 | uint32_t dsparb = I915_READ(DSPARB); |
||
808 | int size; |
||
809 | |||
810 | size = dsparb & 0x7f; |
||
811 | size >>= 1; /* Convert to cachelines */ |
||
812 | |||
813 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, |
||
814 | plane ? "B" : "A", size); |
||
815 | |||
816 | return size; |
||
817 | } |
||
818 | |||
819 | /* Pineview has different values for various configs */ |
||
820 | static const struct intel_watermark_params pineview_display_wm = { |
||
821 | PINEVIEW_DISPLAY_FIFO, |
||
822 | PINEVIEW_MAX_WM, |
||
823 | PINEVIEW_DFT_WM, |
||
824 | PINEVIEW_GUARD_WM, |
||
825 | PINEVIEW_FIFO_LINE_SIZE |
||
826 | }; |
||
827 | static const struct intel_watermark_params pineview_display_hplloff_wm = { |
||
828 | PINEVIEW_DISPLAY_FIFO, |
||
829 | PINEVIEW_MAX_WM, |
||
830 | PINEVIEW_DFT_HPLLOFF_WM, |
||
831 | PINEVIEW_GUARD_WM, |
||
832 | PINEVIEW_FIFO_LINE_SIZE |
||
833 | }; |
||
834 | static const struct intel_watermark_params pineview_cursor_wm = { |
||
835 | PINEVIEW_CURSOR_FIFO, |
||
836 | PINEVIEW_CURSOR_MAX_WM, |
||
837 | PINEVIEW_CURSOR_DFT_WM, |
||
838 | PINEVIEW_CURSOR_GUARD_WM, |
||
839 | PINEVIEW_FIFO_LINE_SIZE, |
||
840 | }; |
||
841 | static const struct intel_watermark_params pineview_cursor_hplloff_wm = { |
||
842 | PINEVIEW_CURSOR_FIFO, |
||
843 | PINEVIEW_CURSOR_MAX_WM, |
||
844 | PINEVIEW_CURSOR_DFT_WM, |
||
845 | PINEVIEW_CURSOR_GUARD_WM, |
||
846 | PINEVIEW_FIFO_LINE_SIZE |
||
847 | }; |
||
848 | static const struct intel_watermark_params g4x_wm_info = { |
||
849 | G4X_FIFO_SIZE, |
||
850 | G4X_MAX_WM, |
||
851 | G4X_MAX_WM, |
||
852 | 2, |
||
853 | G4X_FIFO_LINE_SIZE, |
||
854 | }; |
||
855 | static const struct intel_watermark_params g4x_cursor_wm_info = { |
||
856 | I965_CURSOR_FIFO, |
||
857 | I965_CURSOR_MAX_WM, |
||
858 | I965_CURSOR_DFT_WM, |
||
859 | 2, |
||
860 | G4X_FIFO_LINE_SIZE, |
||
861 | }; |
||
862 | static const struct intel_watermark_params valleyview_wm_info = { |
||
863 | VALLEYVIEW_FIFO_SIZE, |
||
864 | VALLEYVIEW_MAX_WM, |
||
865 | VALLEYVIEW_MAX_WM, |
||
866 | 2, |
||
867 | G4X_FIFO_LINE_SIZE, |
||
868 | }; |
||
869 | static const struct intel_watermark_params valleyview_cursor_wm_info = { |
||
870 | I965_CURSOR_FIFO, |
||
871 | VALLEYVIEW_CURSOR_MAX_WM, |
||
872 | I965_CURSOR_DFT_WM, |
||
873 | 2, |
||
874 | G4X_FIFO_LINE_SIZE, |
||
875 | }; |
||
876 | static const struct intel_watermark_params i965_cursor_wm_info = { |
||
877 | I965_CURSOR_FIFO, |
||
878 | I965_CURSOR_MAX_WM, |
||
879 | I965_CURSOR_DFT_WM, |
||
880 | 2, |
||
881 | I915_FIFO_LINE_SIZE, |
||
882 | }; |
||
883 | static const struct intel_watermark_params i945_wm_info = { |
||
884 | I945_FIFO_SIZE, |
||
885 | I915_MAX_WM, |
||
886 | 1, |
||
887 | 2, |
||
888 | I915_FIFO_LINE_SIZE |
||
889 | }; |
||
890 | static const struct intel_watermark_params i915_wm_info = { |
||
891 | I915_FIFO_SIZE, |
||
892 | I915_MAX_WM, |
||
893 | 1, |
||
894 | 2, |
||
895 | I915_FIFO_LINE_SIZE |
||
896 | }; |
||
897 | static const struct intel_watermark_params i855_wm_info = { |
||
898 | I855GM_FIFO_SIZE, |
||
899 | I915_MAX_WM, |
||
900 | 1, |
||
901 | 2, |
||
902 | I830_FIFO_LINE_SIZE |
||
903 | }; |
||
904 | static const struct intel_watermark_params i830_wm_info = { |
||
905 | I830_FIFO_SIZE, |
||
906 | I915_MAX_WM, |
||
907 | 1, |
||
908 | 2, |
||
909 | I830_FIFO_LINE_SIZE |
||
910 | }; |
||
911 | |||
912 | static const struct intel_watermark_params ironlake_display_wm_info = { |
||
913 | ILK_DISPLAY_FIFO, |
||
914 | ILK_DISPLAY_MAXWM, |
||
915 | ILK_DISPLAY_DFTWM, |
||
916 | 2, |
||
917 | ILK_FIFO_LINE_SIZE |
||
918 | }; |
||
919 | static const struct intel_watermark_params ironlake_cursor_wm_info = { |
||
920 | ILK_CURSOR_FIFO, |
||
921 | ILK_CURSOR_MAXWM, |
||
922 | ILK_CURSOR_DFTWM, |
||
923 | 2, |
||
924 | ILK_FIFO_LINE_SIZE |
||
925 | }; |
||
926 | static const struct intel_watermark_params ironlake_display_srwm_info = { |
||
927 | ILK_DISPLAY_SR_FIFO, |
||
928 | ILK_DISPLAY_MAX_SRWM, |
||
929 | ILK_DISPLAY_DFT_SRWM, |
||
930 | 2, |
||
931 | ILK_FIFO_LINE_SIZE |
||
932 | }; |
||
933 | static const struct intel_watermark_params ironlake_cursor_srwm_info = { |
||
934 | ILK_CURSOR_SR_FIFO, |
||
935 | ILK_CURSOR_MAX_SRWM, |
||
936 | ILK_CURSOR_DFT_SRWM, |
||
937 | 2, |
||
938 | ILK_FIFO_LINE_SIZE |
||
939 | }; |
||
940 | |||
941 | static const struct intel_watermark_params sandybridge_display_wm_info = { |
||
942 | SNB_DISPLAY_FIFO, |
||
943 | SNB_DISPLAY_MAXWM, |
||
944 | SNB_DISPLAY_DFTWM, |
||
945 | 2, |
||
946 | SNB_FIFO_LINE_SIZE |
||
947 | }; |
||
948 | static const struct intel_watermark_params sandybridge_cursor_wm_info = { |
||
949 | SNB_CURSOR_FIFO, |
||
950 | SNB_CURSOR_MAXWM, |
||
951 | SNB_CURSOR_DFTWM, |
||
952 | 2, |
||
953 | SNB_FIFO_LINE_SIZE |
||
954 | }; |
||
955 | static const struct intel_watermark_params sandybridge_display_srwm_info = { |
||
956 | SNB_DISPLAY_SR_FIFO, |
||
957 | SNB_DISPLAY_MAX_SRWM, |
||
958 | SNB_DISPLAY_DFT_SRWM, |
||
959 | 2, |
||
960 | SNB_FIFO_LINE_SIZE |
||
961 | }; |
||
962 | static const struct intel_watermark_params sandybridge_cursor_srwm_info = { |
||
963 | SNB_CURSOR_SR_FIFO, |
||
964 | SNB_CURSOR_MAX_SRWM, |
||
965 | SNB_CURSOR_DFT_SRWM, |
||
966 | 2, |
||
967 | SNB_FIFO_LINE_SIZE |
||
968 | }; |
||
969 | |||
970 | |||
971 | /** |
||
972 | * intel_calculate_wm - calculate watermark level |
||
973 | * @clock_in_khz: pixel clock |
||
974 | * @wm: chip FIFO params |
||
975 | * @pixel_size: display pixel size |
||
976 | * @latency_ns: memory latency for the platform |
||
977 | * |
||
978 | * Calculate the watermark level (the level at which the display plane will |
||
979 | * start fetching from memory again). Each chip has a different display |
||
980 | * FIFO size and allocation, so the caller needs to figure that out and pass |
||
981 | * in the correct intel_watermark_params structure. |
||
982 | * |
||
983 | * As the pixel clock runs, the FIFO will be drained at a rate that depends |
||
984 | * on the pixel size. When it reaches the watermark level, it'll start |
||
985 | * fetching FIFO line sized based chunks from memory until the FIFO fills |
||
986 | * past the watermark point. If the FIFO drains completely, a FIFO underrun |
||
987 | * will occur, and a display engine hang could result. |
||
988 | */ |
||
989 | static unsigned long intel_calculate_wm(unsigned long clock_in_khz, |
||
990 | const struct intel_watermark_params *wm, |
||
991 | int fifo_size, |
||
992 | int pixel_size, |
||
993 | unsigned long latency_ns) |
||
994 | { |
||
995 | long entries_required, wm_size; |
||
996 | |||
997 | /* |
||
998 | * Note: we need to make sure we don't overflow for various clock & |
||
999 | * latency values. |
||
1000 | * clocks go from a few thousand to several hundred thousand. |
||
1001 | * latency is usually a few thousand |
||
1002 | */ |
||
1003 | entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) / |
||
1004 | 1000; |
||
1005 | entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size); |
||
1006 | |||
1007 | DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required); |
||
1008 | |||
1009 | wm_size = fifo_size - (entries_required + wm->guard_size); |
||
1010 | |||
1011 | DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size); |
||
1012 | |||
1013 | /* Don't promote wm_size to unsigned... */ |
||
1014 | if (wm_size > (long)wm->max_wm) |
||
1015 | wm_size = wm->max_wm; |
||
1016 | if (wm_size <= 0) |
||
1017 | wm_size = wm->default_wm; |
||
1018 | return wm_size; |
||
1019 | } |
||
1020 | |||
1021 | static struct drm_crtc *single_enabled_crtc(struct drm_device *dev) |
||
1022 | { |
||
1023 | struct drm_crtc *crtc, *enabled = NULL; |
||
1024 | |||
1025 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { |
||
3243 | Serge | 1026 | if (intel_crtc_active(crtc)) { |
3031 | serge | 1027 | if (enabled) |
1028 | return NULL; |
||
1029 | enabled = crtc; |
||
1030 | } |
||
1031 | } |
||
1032 | |||
1033 | return enabled; |
||
1034 | } |
||
1035 | |||
1036 | static void pineview_update_wm(struct drm_device *dev) |
||
1037 | { |
||
1038 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1039 | struct drm_crtc *crtc; |
||
1040 | const struct cxsr_latency *latency; |
||
1041 | u32 reg; |
||
1042 | unsigned long wm; |
||
1043 | |||
1044 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3, |
||
1045 | dev_priv->fsb_freq, dev_priv->mem_freq); |
||
1046 | if (!latency) { |
||
1047 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); |
||
1048 | pineview_disable_cxsr(dev); |
||
1049 | return; |
||
1050 | } |
||
1051 | |||
1052 | crtc = single_enabled_crtc(dev); |
||
1053 | if (crtc) { |
||
1054 | int clock = crtc->mode.clock; |
||
1055 | int pixel_size = crtc->fb->bits_per_pixel / 8; |
||
1056 | |||
1057 | /* Display SR */ |
||
1058 | wm = intel_calculate_wm(clock, &pineview_display_wm, |
||
1059 | pineview_display_wm.fifo_size, |
||
1060 | pixel_size, latency->display_sr); |
||
1061 | reg = I915_READ(DSPFW1); |
||
1062 | reg &= ~DSPFW_SR_MASK; |
||
1063 | reg |= wm << DSPFW_SR_SHIFT; |
||
1064 | I915_WRITE(DSPFW1, reg); |
||
1065 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); |
||
1066 | |||
1067 | /* cursor SR */ |
||
1068 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, |
||
1069 | pineview_display_wm.fifo_size, |
||
1070 | pixel_size, latency->cursor_sr); |
||
1071 | reg = I915_READ(DSPFW3); |
||
1072 | reg &= ~DSPFW_CURSOR_SR_MASK; |
||
1073 | reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT; |
||
1074 | I915_WRITE(DSPFW3, reg); |
||
1075 | |||
1076 | /* Display HPLL off SR */ |
||
1077 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, |
||
1078 | pineview_display_hplloff_wm.fifo_size, |
||
1079 | pixel_size, latency->display_hpll_disable); |
||
1080 | reg = I915_READ(DSPFW3); |
||
1081 | reg &= ~DSPFW_HPLL_SR_MASK; |
||
1082 | reg |= wm & DSPFW_HPLL_SR_MASK; |
||
1083 | I915_WRITE(DSPFW3, reg); |
||
1084 | |||
1085 | /* cursor HPLL off SR */ |
||
1086 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, |
||
1087 | pineview_display_hplloff_wm.fifo_size, |
||
1088 | pixel_size, latency->cursor_hpll_disable); |
||
1089 | reg = I915_READ(DSPFW3); |
||
1090 | reg &= ~DSPFW_HPLL_CURSOR_MASK; |
||
1091 | reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT; |
||
1092 | I915_WRITE(DSPFW3, reg); |
||
1093 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); |
||
1094 | |||
1095 | /* activate cxsr */ |
||
1096 | I915_WRITE(DSPFW3, |
||
1097 | I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN); |
||
1098 | DRM_DEBUG_KMS("Self-refresh is enabled\n"); |
||
1099 | } else { |
||
1100 | pineview_disable_cxsr(dev); |
||
1101 | DRM_DEBUG_KMS("Self-refresh is disabled\n"); |
||
1102 | } |
||
1103 | } |
||
1104 | |||
1105 | static bool g4x_compute_wm0(struct drm_device *dev, |
||
1106 | int plane, |
||
1107 | const struct intel_watermark_params *display, |
||
1108 | int display_latency_ns, |
||
1109 | const struct intel_watermark_params *cursor, |
||
1110 | int cursor_latency_ns, |
||
1111 | int *plane_wm, |
||
1112 | int *cursor_wm) |
||
1113 | { |
||
1114 | struct drm_crtc *crtc; |
||
1115 | int htotal, hdisplay, clock, pixel_size; |
||
1116 | int line_time_us, line_count; |
||
1117 | int entries, tlb_miss; |
||
1118 | |||
1119 | crtc = intel_get_crtc_for_plane(dev, plane); |
||
3243 | Serge | 1120 | if (!intel_crtc_active(crtc)) { |
3031 | serge | 1121 | *cursor_wm = cursor->guard_size; |
1122 | *plane_wm = display->guard_size; |
||
1123 | return false; |
||
1124 | } |
||
1125 | |||
1126 | htotal = crtc->mode.htotal; |
||
1127 | hdisplay = crtc->mode.hdisplay; |
||
1128 | clock = crtc->mode.clock; |
||
1129 | pixel_size = crtc->fb->bits_per_pixel / 8; |
||
1130 | |||
1131 | /* Use the small buffer method to calculate plane watermark */ |
||
1132 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; |
||
1133 | tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8; |
||
1134 | if (tlb_miss > 0) |
||
1135 | entries += tlb_miss; |
||
1136 | entries = DIV_ROUND_UP(entries, display->cacheline_size); |
||
1137 | *plane_wm = entries + display->guard_size; |
||
1138 | if (*plane_wm > (int)display->max_wm) |
||
1139 | *plane_wm = display->max_wm; |
||
1140 | |||
1141 | /* Use the large buffer method to calculate cursor watermark */ |
||
1142 | line_time_us = ((htotal * 1000) / clock); |
||
1143 | line_count = (cursor_latency_ns / line_time_us + 1000) / 1000; |
||
1144 | entries = line_count * 64 * pixel_size; |
||
1145 | tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8; |
||
1146 | if (tlb_miss > 0) |
||
1147 | entries += tlb_miss; |
||
1148 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
||
1149 | *cursor_wm = entries + cursor->guard_size; |
||
1150 | if (*cursor_wm > (int)cursor->max_wm) |
||
1151 | *cursor_wm = (int)cursor->max_wm; |
||
1152 | |||
1153 | return true; |
||
1154 | } |
||
1155 | |||
1156 | /* |
||
1157 | * Check the wm result. |
||
1158 | * |
||
1159 | * If any calculated watermark values is larger than the maximum value that |
||
1160 | * can be programmed into the associated watermark register, that watermark |
||
1161 | * must be disabled. |
||
1162 | */ |
||
1163 | static bool g4x_check_srwm(struct drm_device *dev, |
||
1164 | int display_wm, int cursor_wm, |
||
1165 | const struct intel_watermark_params *display, |
||
1166 | const struct intel_watermark_params *cursor) |
||
1167 | { |
||
1168 | DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n", |
||
1169 | display_wm, cursor_wm); |
||
1170 | |||
1171 | if (display_wm > display->max_wm) { |
||
1172 | DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n", |
||
1173 | display_wm, display->max_wm); |
||
1174 | return false; |
||
1175 | } |
||
1176 | |||
1177 | if (cursor_wm > cursor->max_wm) { |
||
1178 | DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n", |
||
1179 | cursor_wm, cursor->max_wm); |
||
1180 | return false; |
||
1181 | } |
||
1182 | |||
1183 | if (!(display_wm || cursor_wm)) { |
||
1184 | DRM_DEBUG_KMS("SR latency is 0, disabling\n"); |
||
1185 | return false; |
||
1186 | } |
||
1187 | |||
1188 | return true; |
||
1189 | } |
||
1190 | |||
1191 | static bool g4x_compute_srwm(struct drm_device *dev, |
||
1192 | int plane, |
||
1193 | int latency_ns, |
||
1194 | const struct intel_watermark_params *display, |
||
1195 | const struct intel_watermark_params *cursor, |
||
1196 | int *display_wm, int *cursor_wm) |
||
1197 | { |
||
1198 | struct drm_crtc *crtc; |
||
1199 | int hdisplay, htotal, pixel_size, clock; |
||
1200 | unsigned long line_time_us; |
||
1201 | int line_count, line_size; |
||
1202 | int small, large; |
||
1203 | int entries; |
||
1204 | |||
1205 | if (!latency_ns) { |
||
1206 | *display_wm = *cursor_wm = 0; |
||
1207 | return false; |
||
1208 | } |
||
1209 | |||
1210 | crtc = intel_get_crtc_for_plane(dev, plane); |
||
1211 | hdisplay = crtc->mode.hdisplay; |
||
1212 | htotal = crtc->mode.htotal; |
||
1213 | clock = crtc->mode.clock; |
||
1214 | pixel_size = crtc->fb->bits_per_pixel / 8; |
||
1215 | |||
1216 | line_time_us = (htotal * 1000) / clock; |
||
1217 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
||
1218 | line_size = hdisplay * pixel_size; |
||
1219 | |||
1220 | /* Use the minimum of the small and large buffer method for primary */ |
||
1221 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; |
||
1222 | large = line_count * line_size; |
||
1223 | |||
1224 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); |
||
1225 | *display_wm = entries + display->guard_size; |
||
1226 | |||
1227 | /* calculate the self-refresh watermark for display cursor */ |
||
1228 | entries = line_count * pixel_size * 64; |
||
1229 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
||
1230 | *cursor_wm = entries + cursor->guard_size; |
||
1231 | |||
1232 | return g4x_check_srwm(dev, |
||
1233 | *display_wm, *cursor_wm, |
||
1234 | display, cursor); |
||
1235 | } |
||
1236 | |||
1237 | static bool vlv_compute_drain_latency(struct drm_device *dev, |
||
1238 | int plane, |
||
1239 | int *plane_prec_mult, |
||
1240 | int *plane_dl, |
||
1241 | int *cursor_prec_mult, |
||
1242 | int *cursor_dl) |
||
1243 | { |
||
1244 | struct drm_crtc *crtc; |
||
1245 | int clock, pixel_size; |
||
1246 | int entries; |
||
1247 | |||
1248 | crtc = intel_get_crtc_for_plane(dev, plane); |
||
3243 | Serge | 1249 | if (!intel_crtc_active(crtc)) |
3031 | serge | 1250 | return false; |
1251 | |||
1252 | clock = crtc->mode.clock; /* VESA DOT Clock */ |
||
1253 | pixel_size = crtc->fb->bits_per_pixel / 8; /* BPP */ |
||
1254 | |||
1255 | entries = (clock / 1000) * pixel_size; |
||
1256 | *plane_prec_mult = (entries > 256) ? |
||
1257 | DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16; |
||
1258 | *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) * |
||
1259 | pixel_size); |
||
1260 | |||
1261 | entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */ |
||
1262 | *cursor_prec_mult = (entries > 256) ? |
||
1263 | DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16; |
||
1264 | *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4); |
||
1265 | |||
1266 | return true; |
||
1267 | } |
||
1268 | |||
1269 | /* |
||
1270 | * Update drain latency registers of memory arbiter |
||
1271 | * |
||
1272 | * Valleyview SoC has a new memory arbiter and needs drain latency registers |
||
1273 | * to be programmed. Each plane has a drain latency multiplier and a drain |
||
1274 | * latency value. |
||
1275 | */ |
||
1276 | |||
1277 | static void vlv_update_drain_latency(struct drm_device *dev) |
||
1278 | { |
||
1279 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1280 | int planea_prec, planea_dl, planeb_prec, planeb_dl; |
||
1281 | int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl; |
||
1282 | int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is |
||
1283 | either 16 or 32 */ |
||
1284 | |||
1285 | /* For plane A, Cursor A */ |
||
1286 | if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl, |
||
1287 | &cursor_prec_mult, &cursora_dl)) { |
||
1288 | cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ? |
||
1289 | DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16; |
||
1290 | planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ? |
||
1291 | DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16; |
||
1292 | |||
1293 | I915_WRITE(VLV_DDL1, cursora_prec | |
||
1294 | (cursora_dl << DDL_CURSORA_SHIFT) | |
||
1295 | planea_prec | planea_dl); |
||
1296 | } |
||
1297 | |||
1298 | /* For plane B, Cursor B */ |
||
1299 | if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl, |
||
1300 | &cursor_prec_mult, &cursorb_dl)) { |
||
1301 | cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ? |
||
1302 | DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16; |
||
1303 | planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ? |
||
1304 | DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16; |
||
1305 | |||
1306 | I915_WRITE(VLV_DDL2, cursorb_prec | |
||
1307 | (cursorb_dl << DDL_CURSORB_SHIFT) | |
||
1308 | planeb_prec | planeb_dl); |
||
1309 | } |
||
1310 | } |
||
1311 | |||
1312 | #define single_plane_enabled(mask) is_power_of_2(mask) |
||
1313 | |||
1314 | static void valleyview_update_wm(struct drm_device *dev) |
||
1315 | { |
||
1316 | static const int sr_latency_ns = 12000; |
||
1317 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1318 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; |
||
1319 | int plane_sr, cursor_sr; |
||
3243 | Serge | 1320 | int ignore_plane_sr, ignore_cursor_sr; |
3031 | serge | 1321 | unsigned int enabled = 0; |
1322 | |||
1323 | vlv_update_drain_latency(dev); |
||
1324 | |||
3746 | Serge | 1325 | if (g4x_compute_wm0(dev, PIPE_A, |
3031 | serge | 1326 | &valleyview_wm_info, latency_ns, |
1327 | &valleyview_cursor_wm_info, latency_ns, |
||
1328 | &planea_wm, &cursora_wm)) |
||
3746 | Serge | 1329 | enabled |= 1 << PIPE_A; |
3031 | serge | 1330 | |
3746 | Serge | 1331 | if (g4x_compute_wm0(dev, PIPE_B, |
3031 | serge | 1332 | &valleyview_wm_info, latency_ns, |
1333 | &valleyview_cursor_wm_info, latency_ns, |
||
1334 | &planeb_wm, &cursorb_wm)) |
||
3746 | Serge | 1335 | enabled |= 1 << PIPE_B; |
3031 | serge | 1336 | |
1337 | if (single_plane_enabled(enabled) && |
||
1338 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
||
1339 | sr_latency_ns, |
||
1340 | &valleyview_wm_info, |
||
1341 | &valleyview_cursor_wm_info, |
||
3243 | Serge | 1342 | &plane_sr, &ignore_cursor_sr) && |
1343 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
||
1344 | 2*sr_latency_ns, |
||
1345 | &valleyview_wm_info, |
||
1346 | &valleyview_cursor_wm_info, |
||
1347 | &ignore_plane_sr, &cursor_sr)) { |
||
3031 | serge | 1348 | I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN); |
3243 | Serge | 1349 | } else { |
3031 | serge | 1350 | I915_WRITE(FW_BLC_SELF_VLV, |
1351 | I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN); |
||
3243 | Serge | 1352 | plane_sr = cursor_sr = 0; |
1353 | } |
||
3031 | serge | 1354 | |
1355 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", |
||
1356 | planea_wm, cursora_wm, |
||
1357 | planeb_wm, cursorb_wm, |
||
1358 | plane_sr, cursor_sr); |
||
1359 | |||
1360 | I915_WRITE(DSPFW1, |
||
1361 | (plane_sr << DSPFW_SR_SHIFT) | |
||
1362 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | |
||
1363 | (planeb_wm << DSPFW_PLANEB_SHIFT) | |
||
1364 | planea_wm); |
||
1365 | I915_WRITE(DSPFW2, |
||
3243 | Serge | 1366 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
3031 | serge | 1367 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1368 | I915_WRITE(DSPFW3, |
||
3243 | Serge | 1369 | (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) | |
1370 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
||
3031 | serge | 1371 | } |
1372 | |||
1373 | static void g4x_update_wm(struct drm_device *dev) |
||
1374 | { |
||
1375 | static const int sr_latency_ns = 12000; |
||
1376 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1377 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; |
||
1378 | int plane_sr, cursor_sr; |
||
1379 | unsigned int enabled = 0; |
||
1380 | |||
3746 | Serge | 1381 | if (g4x_compute_wm0(dev, PIPE_A, |
3031 | serge | 1382 | &g4x_wm_info, latency_ns, |
1383 | &g4x_cursor_wm_info, latency_ns, |
||
1384 | &planea_wm, &cursora_wm)) |
||
3746 | Serge | 1385 | enabled |= 1 << PIPE_A; |
3031 | serge | 1386 | |
3746 | Serge | 1387 | if (g4x_compute_wm0(dev, PIPE_B, |
3031 | serge | 1388 | &g4x_wm_info, latency_ns, |
1389 | &g4x_cursor_wm_info, latency_ns, |
||
1390 | &planeb_wm, &cursorb_wm)) |
||
3746 | Serge | 1391 | enabled |= 1 << PIPE_B; |
3031 | serge | 1392 | |
1393 | if (single_plane_enabled(enabled) && |
||
1394 | g4x_compute_srwm(dev, ffs(enabled) - 1, |
||
1395 | sr_latency_ns, |
||
1396 | &g4x_wm_info, |
||
1397 | &g4x_cursor_wm_info, |
||
3243 | Serge | 1398 | &plane_sr, &cursor_sr)) { |
3031 | serge | 1399 | I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN); |
3243 | Serge | 1400 | } else { |
3031 | serge | 1401 | I915_WRITE(FW_BLC_SELF, |
1402 | I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN); |
||
3243 | Serge | 1403 | plane_sr = cursor_sr = 0; |
1404 | } |
||
3031 | serge | 1405 | |
1406 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", |
||
1407 | planea_wm, cursora_wm, |
||
1408 | planeb_wm, cursorb_wm, |
||
1409 | plane_sr, cursor_sr); |
||
1410 | |||
1411 | I915_WRITE(DSPFW1, |
||
1412 | (plane_sr << DSPFW_SR_SHIFT) | |
||
1413 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | |
||
1414 | (planeb_wm << DSPFW_PLANEB_SHIFT) | |
||
1415 | planea_wm); |
||
1416 | I915_WRITE(DSPFW2, |
||
3243 | Serge | 1417 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
3031 | serge | 1418 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1419 | /* HPLL off in SR has some issues on G4x... disable it */ |
||
1420 | I915_WRITE(DSPFW3, |
||
3243 | Serge | 1421 | (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) | |
3031 | serge | 1422 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
1423 | } |
||
1424 | |||
1425 | static void i965_update_wm(struct drm_device *dev) |
||
1426 | { |
||
1427 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1428 | struct drm_crtc *crtc; |
||
1429 | int srwm = 1; |
||
1430 | int cursor_sr = 16; |
||
1431 | |||
1432 | /* Calc sr entries for one plane configs */ |
||
1433 | crtc = single_enabled_crtc(dev); |
||
1434 | if (crtc) { |
||
1435 | /* self-refresh has much higher latency */ |
||
1436 | static const int sr_latency_ns = 12000; |
||
1437 | int clock = crtc->mode.clock; |
||
1438 | int htotal = crtc->mode.htotal; |
||
1439 | int hdisplay = crtc->mode.hdisplay; |
||
1440 | int pixel_size = crtc->fb->bits_per_pixel / 8; |
||
1441 | unsigned long line_time_us; |
||
1442 | int entries; |
||
1443 | |||
1444 | line_time_us = ((htotal * 1000) / clock); |
||
1445 | |||
1446 | /* Use ns/us then divide to preserve precision */ |
||
1447 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
||
1448 | pixel_size * hdisplay; |
||
1449 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); |
||
1450 | srwm = I965_FIFO_SIZE - entries; |
||
1451 | if (srwm < 0) |
||
1452 | srwm = 1; |
||
1453 | srwm &= 0x1ff; |
||
1454 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", |
||
1455 | entries, srwm); |
||
1456 | |||
1457 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
||
1458 | pixel_size * 64; |
||
1459 | entries = DIV_ROUND_UP(entries, |
||
1460 | i965_cursor_wm_info.cacheline_size); |
||
1461 | cursor_sr = i965_cursor_wm_info.fifo_size - |
||
1462 | (entries + i965_cursor_wm_info.guard_size); |
||
1463 | |||
1464 | if (cursor_sr > i965_cursor_wm_info.max_wm) |
||
1465 | cursor_sr = i965_cursor_wm_info.max_wm; |
||
1466 | |||
1467 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " |
||
1468 | "cursor %d\n", srwm, cursor_sr); |
||
1469 | |||
1470 | if (IS_CRESTLINE(dev)) |
||
1471 | I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN); |
||
1472 | } else { |
||
1473 | /* Turn off self refresh if both pipes are enabled */ |
||
1474 | if (IS_CRESTLINE(dev)) |
||
1475 | I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF) |
||
1476 | & ~FW_BLC_SELF_EN); |
||
1477 | } |
||
1478 | |||
1479 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", |
||
1480 | srwm); |
||
1481 | |||
1482 | /* 965 has limitations... */ |
||
1483 | I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) | |
||
1484 | (8 << 16) | (8 << 8) | (8 << 0)); |
||
1485 | I915_WRITE(DSPFW2, (8 << 8) | (8 << 0)); |
||
1486 | /* update cursor SR watermark */ |
||
1487 | I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
||
1488 | } |
||
1489 | |||
1490 | static void i9xx_update_wm(struct drm_device *dev) |
||
1491 | { |
||
1492 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1493 | const struct intel_watermark_params *wm_info; |
||
1494 | uint32_t fwater_lo; |
||
1495 | uint32_t fwater_hi; |
||
1496 | int cwm, srwm = 1; |
||
1497 | int fifo_size; |
||
1498 | int planea_wm, planeb_wm; |
||
1499 | struct drm_crtc *crtc, *enabled = NULL; |
||
1500 | |||
1501 | if (IS_I945GM(dev)) |
||
1502 | wm_info = &i945_wm_info; |
||
1503 | else if (!IS_GEN2(dev)) |
||
1504 | wm_info = &i915_wm_info; |
||
1505 | else |
||
1506 | wm_info = &i855_wm_info; |
||
1507 | |||
1508 | fifo_size = dev_priv->display.get_fifo_size(dev, 0); |
||
1509 | crtc = intel_get_crtc_for_plane(dev, 0); |
||
3243 | Serge | 1510 | if (intel_crtc_active(crtc)) { |
1511 | int cpp = crtc->fb->bits_per_pixel / 8; |
||
1512 | if (IS_GEN2(dev)) |
||
1513 | cpp = 4; |
||
1514 | |||
3031 | serge | 1515 | planea_wm = intel_calculate_wm(crtc->mode.clock, |
3243 | Serge | 1516 | wm_info, fifo_size, cpp, |
3031 | serge | 1517 | latency_ns); |
1518 | enabled = crtc; |
||
1519 | } else |
||
1520 | planea_wm = fifo_size - wm_info->guard_size; |
||
1521 | |||
1522 | fifo_size = dev_priv->display.get_fifo_size(dev, 1); |
||
1523 | crtc = intel_get_crtc_for_plane(dev, 1); |
||
3243 | Serge | 1524 | if (intel_crtc_active(crtc)) { |
1525 | int cpp = crtc->fb->bits_per_pixel / 8; |
||
1526 | if (IS_GEN2(dev)) |
||
1527 | cpp = 4; |
||
1528 | |||
3031 | serge | 1529 | planeb_wm = intel_calculate_wm(crtc->mode.clock, |
3243 | Serge | 1530 | wm_info, fifo_size, cpp, |
3031 | serge | 1531 | latency_ns); |
1532 | if (enabled == NULL) |
||
1533 | enabled = crtc; |
||
1534 | else |
||
1535 | enabled = NULL; |
||
1536 | } else |
||
1537 | planeb_wm = fifo_size - wm_info->guard_size; |
||
1538 | |||
1539 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); |
||
1540 | |||
1541 | /* |
||
1542 | * Overlay gets an aggressive default since video jitter is bad. |
||
1543 | */ |
||
1544 | cwm = 2; |
||
1545 | |||
1546 | /* Play safe and disable self-refresh before adjusting watermarks. */ |
||
1547 | if (IS_I945G(dev) || IS_I945GM(dev)) |
||
1548 | I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0); |
||
1549 | else if (IS_I915GM(dev)) |
||
1550 | I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN); |
||
1551 | |||
1552 | /* Calc sr entries for one plane configs */ |
||
1553 | if (HAS_FW_BLC(dev) && enabled) { |
||
1554 | /* self-refresh has much higher latency */ |
||
1555 | static const int sr_latency_ns = 6000; |
||
1556 | int clock = enabled->mode.clock; |
||
1557 | int htotal = enabled->mode.htotal; |
||
1558 | int hdisplay = enabled->mode.hdisplay; |
||
1559 | int pixel_size = enabled->fb->bits_per_pixel / 8; |
||
1560 | unsigned long line_time_us; |
||
1561 | int entries; |
||
1562 | |||
1563 | line_time_us = (htotal * 1000) / clock; |
||
1564 | |||
1565 | /* Use ns/us then divide to preserve precision */ |
||
1566 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * |
||
1567 | pixel_size * hdisplay; |
||
1568 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); |
||
1569 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); |
||
1570 | srwm = wm_info->fifo_size - entries; |
||
1571 | if (srwm < 0) |
||
1572 | srwm = 1; |
||
1573 | |||
1574 | if (IS_I945G(dev) || IS_I945GM(dev)) |
||
1575 | I915_WRITE(FW_BLC_SELF, |
||
1576 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); |
||
1577 | else if (IS_I915GM(dev)) |
||
1578 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); |
||
1579 | } |
||
1580 | |||
1581 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", |
||
1582 | planea_wm, planeb_wm, cwm, srwm); |
||
1583 | |||
1584 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); |
||
1585 | fwater_hi = (cwm & 0x1f); |
||
1586 | |||
1587 | /* Set request length to 8 cachelines per fetch */ |
||
1588 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); |
||
1589 | fwater_hi = fwater_hi | (1 << 8); |
||
1590 | |||
1591 | I915_WRITE(FW_BLC, fwater_lo); |
||
1592 | I915_WRITE(FW_BLC2, fwater_hi); |
||
1593 | |||
1594 | if (HAS_FW_BLC(dev)) { |
||
1595 | if (enabled) { |
||
1596 | if (IS_I945G(dev) || IS_I945GM(dev)) |
||
1597 | I915_WRITE(FW_BLC_SELF, |
||
1598 | FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN); |
||
1599 | else if (IS_I915GM(dev)) |
||
1600 | I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN); |
||
1601 | DRM_DEBUG_KMS("memory self refresh enabled\n"); |
||
1602 | } else |
||
1603 | DRM_DEBUG_KMS("memory self refresh disabled\n"); |
||
1604 | } |
||
1605 | } |
||
1606 | |||
1607 | static void i830_update_wm(struct drm_device *dev) |
||
1608 | { |
||
1609 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1610 | struct drm_crtc *crtc; |
||
1611 | uint32_t fwater_lo; |
||
1612 | int planea_wm; |
||
1613 | |||
1614 | crtc = single_enabled_crtc(dev); |
||
1615 | if (crtc == NULL) |
||
1616 | return; |
||
1617 | |||
1618 | planea_wm = intel_calculate_wm(crtc->mode.clock, &i830_wm_info, |
||
1619 | dev_priv->display.get_fifo_size(dev, 0), |
||
3243 | Serge | 1620 | 4, latency_ns); |
3031 | serge | 1621 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
1622 | fwater_lo |= (3<<8) | planea_wm; |
||
1623 | |||
1624 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); |
||
1625 | |||
1626 | I915_WRITE(FW_BLC, fwater_lo); |
||
1627 | } |
||
1628 | |||
1629 | #define ILK_LP0_PLANE_LATENCY 700 |
||
1630 | #define ILK_LP0_CURSOR_LATENCY 1300 |
||
1631 | |||
1632 | /* |
||
1633 | * Check the wm result. |
||
1634 | * |
||
1635 | * If any calculated watermark values is larger than the maximum value that |
||
1636 | * can be programmed into the associated watermark register, that watermark |
||
1637 | * must be disabled. |
||
1638 | */ |
||
1639 | static bool ironlake_check_srwm(struct drm_device *dev, int level, |
||
1640 | int fbc_wm, int display_wm, int cursor_wm, |
||
1641 | const struct intel_watermark_params *display, |
||
1642 | const struct intel_watermark_params *cursor) |
||
1643 | { |
||
1644 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1645 | |||
1646 | DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d," |
||
1647 | " cursor %d\n", level, display_wm, fbc_wm, cursor_wm); |
||
1648 | |||
1649 | if (fbc_wm > SNB_FBC_MAX_SRWM) { |
||
1650 | DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n", |
||
1651 | fbc_wm, SNB_FBC_MAX_SRWM, level); |
||
1652 | |||
1653 | /* fbc has it's own way to disable FBC WM */ |
||
1654 | I915_WRITE(DISP_ARB_CTL, |
||
1655 | I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS); |
||
1656 | return false; |
||
1657 | } |
||
1658 | |||
1659 | if (display_wm > display->max_wm) { |
||
1660 | DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n", |
||
1661 | display_wm, SNB_DISPLAY_MAX_SRWM, level); |
||
1662 | return false; |
||
1663 | } |
||
1664 | |||
1665 | if (cursor_wm > cursor->max_wm) { |
||
1666 | DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n", |
||
1667 | cursor_wm, SNB_CURSOR_MAX_SRWM, level); |
||
1668 | return false; |
||
1669 | } |
||
1670 | |||
1671 | if (!(fbc_wm || display_wm || cursor_wm)) { |
||
1672 | DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level); |
||
1673 | return false; |
||
1674 | } |
||
1675 | |||
1676 | return true; |
||
1677 | } |
||
1678 | |||
1679 | /* |
||
1680 | * Compute watermark values of WM[1-3], |
||
1681 | */ |
||
1682 | static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane, |
||
1683 | int latency_ns, |
||
1684 | const struct intel_watermark_params *display, |
||
1685 | const struct intel_watermark_params *cursor, |
||
1686 | int *fbc_wm, int *display_wm, int *cursor_wm) |
||
1687 | { |
||
1688 | struct drm_crtc *crtc; |
||
1689 | unsigned long line_time_us; |
||
1690 | int hdisplay, htotal, pixel_size, clock; |
||
1691 | int line_count, line_size; |
||
1692 | int small, large; |
||
1693 | int entries; |
||
1694 | |||
1695 | if (!latency_ns) { |
||
1696 | *fbc_wm = *display_wm = *cursor_wm = 0; |
||
1697 | return false; |
||
1698 | } |
||
1699 | |||
1700 | crtc = intel_get_crtc_for_plane(dev, plane); |
||
1701 | hdisplay = crtc->mode.hdisplay; |
||
1702 | htotal = crtc->mode.htotal; |
||
1703 | clock = crtc->mode.clock; |
||
1704 | pixel_size = crtc->fb->bits_per_pixel / 8; |
||
1705 | |||
1706 | line_time_us = (htotal * 1000) / clock; |
||
1707 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
||
1708 | line_size = hdisplay * pixel_size; |
||
1709 | |||
1710 | /* Use the minimum of the small and large buffer method for primary */ |
||
1711 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; |
||
1712 | large = line_count * line_size; |
||
1713 | |||
1714 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); |
||
1715 | *display_wm = entries + display->guard_size; |
||
1716 | |||
1717 | /* |
||
1718 | * Spec says: |
||
1719 | * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2 |
||
1720 | */ |
||
1721 | *fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2; |
||
1722 | |||
1723 | /* calculate the self-refresh watermark for display cursor */ |
||
1724 | entries = line_count * pixel_size * 64; |
||
1725 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
||
1726 | *cursor_wm = entries + cursor->guard_size; |
||
1727 | |||
1728 | return ironlake_check_srwm(dev, level, |
||
1729 | *fbc_wm, *display_wm, *cursor_wm, |
||
1730 | display, cursor); |
||
1731 | } |
||
1732 | |||
1733 | static void ironlake_update_wm(struct drm_device *dev) |
||
1734 | { |
||
1735 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1736 | int fbc_wm, plane_wm, cursor_wm; |
||
1737 | unsigned int enabled; |
||
1738 | |||
1739 | enabled = 0; |
||
3746 | Serge | 1740 | if (g4x_compute_wm0(dev, PIPE_A, |
3031 | serge | 1741 | &ironlake_display_wm_info, |
1742 | ILK_LP0_PLANE_LATENCY, |
||
1743 | &ironlake_cursor_wm_info, |
||
1744 | ILK_LP0_CURSOR_LATENCY, |
||
1745 | &plane_wm, &cursor_wm)) { |
||
1746 | I915_WRITE(WM0_PIPEA_ILK, |
||
1747 | (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm); |
||
1748 | DRM_DEBUG_KMS("FIFO watermarks For pipe A -" |
||
1749 | " plane %d, " "cursor: %d\n", |
||
1750 | plane_wm, cursor_wm); |
||
3746 | Serge | 1751 | enabled |= 1 << PIPE_A; |
3031 | serge | 1752 | } |
1753 | |||
3746 | Serge | 1754 | if (g4x_compute_wm0(dev, PIPE_B, |
3031 | serge | 1755 | &ironlake_display_wm_info, |
1756 | ILK_LP0_PLANE_LATENCY, |
||
1757 | &ironlake_cursor_wm_info, |
||
1758 | ILK_LP0_CURSOR_LATENCY, |
||
1759 | &plane_wm, &cursor_wm)) { |
||
1760 | I915_WRITE(WM0_PIPEB_ILK, |
||
1761 | (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm); |
||
1762 | DRM_DEBUG_KMS("FIFO watermarks For pipe B -" |
||
1763 | " plane %d, cursor: %d\n", |
||
1764 | plane_wm, cursor_wm); |
||
3746 | Serge | 1765 | enabled |= 1 << PIPE_B; |
3031 | serge | 1766 | } |
1767 | |||
1768 | /* |
||
1769 | * Calculate and update the self-refresh watermark only when one |
||
1770 | * display plane is used. |
||
1771 | */ |
||
1772 | I915_WRITE(WM3_LP_ILK, 0); |
||
1773 | I915_WRITE(WM2_LP_ILK, 0); |
||
1774 | I915_WRITE(WM1_LP_ILK, 0); |
||
1775 | |||
1776 | if (!single_plane_enabled(enabled)) |
||
1777 | return; |
||
1778 | enabled = ffs(enabled) - 1; |
||
1779 | |||
1780 | /* WM1 */ |
||
1781 | if (!ironlake_compute_srwm(dev, 1, enabled, |
||
1782 | ILK_READ_WM1_LATENCY() * 500, |
||
1783 | &ironlake_display_srwm_info, |
||
1784 | &ironlake_cursor_srwm_info, |
||
1785 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
1786 | return; |
||
1787 | |||
1788 | I915_WRITE(WM1_LP_ILK, |
||
1789 | WM1_LP_SR_EN | |
||
1790 | (ILK_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
1791 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
1792 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
1793 | cursor_wm); |
||
1794 | |||
1795 | /* WM2 */ |
||
1796 | if (!ironlake_compute_srwm(dev, 2, enabled, |
||
1797 | ILK_READ_WM2_LATENCY() * 500, |
||
1798 | &ironlake_display_srwm_info, |
||
1799 | &ironlake_cursor_srwm_info, |
||
1800 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
1801 | return; |
||
1802 | |||
1803 | I915_WRITE(WM2_LP_ILK, |
||
1804 | WM2_LP_EN | |
||
1805 | (ILK_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
1806 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
1807 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
1808 | cursor_wm); |
||
1809 | |||
1810 | /* |
||
1811 | * WM3 is unsupported on ILK, probably because we don't have latency |
||
1812 | * data for that power state |
||
1813 | */ |
||
1814 | } |
||
1815 | |||
1816 | static void sandybridge_update_wm(struct drm_device *dev) |
||
1817 | { |
||
1818 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1819 | int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */ |
||
1820 | u32 val; |
||
1821 | int fbc_wm, plane_wm, cursor_wm; |
||
1822 | unsigned int enabled; |
||
1823 | |||
1824 | enabled = 0; |
||
3746 | Serge | 1825 | if (g4x_compute_wm0(dev, PIPE_A, |
3031 | serge | 1826 | &sandybridge_display_wm_info, latency, |
1827 | &sandybridge_cursor_wm_info, latency, |
||
1828 | &plane_wm, &cursor_wm)) { |
||
1829 | val = I915_READ(WM0_PIPEA_ILK); |
||
1830 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); |
||
1831 | I915_WRITE(WM0_PIPEA_ILK, val | |
||
1832 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); |
||
1833 | DRM_DEBUG_KMS("FIFO watermarks For pipe A -" |
||
1834 | " plane %d, " "cursor: %d\n", |
||
1835 | plane_wm, cursor_wm); |
||
3746 | Serge | 1836 | enabled |= 1 << PIPE_A; |
3031 | serge | 1837 | } |
1838 | |||
3746 | Serge | 1839 | if (g4x_compute_wm0(dev, PIPE_B, |
3031 | serge | 1840 | &sandybridge_display_wm_info, latency, |
1841 | &sandybridge_cursor_wm_info, latency, |
||
1842 | &plane_wm, &cursor_wm)) { |
||
1843 | val = I915_READ(WM0_PIPEB_ILK); |
||
1844 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); |
||
1845 | I915_WRITE(WM0_PIPEB_ILK, val | |
||
1846 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); |
||
1847 | DRM_DEBUG_KMS("FIFO watermarks For pipe B -" |
||
1848 | " plane %d, cursor: %d\n", |
||
1849 | plane_wm, cursor_wm); |
||
3746 | Serge | 1850 | enabled |= 1 << PIPE_B; |
3031 | serge | 1851 | } |
1852 | |||
3243 | Serge | 1853 | /* |
1854 | * Calculate and update the self-refresh watermark only when one |
||
1855 | * display plane is used. |
||
1856 | * |
||
1857 | * SNB support 3 levels of watermark. |
||
1858 | * |
||
1859 | * WM1/WM2/WM2 watermarks have to be enabled in the ascending order, |
||
1860 | * and disabled in the descending order |
||
1861 | * |
||
1862 | */ |
||
1863 | I915_WRITE(WM3_LP_ILK, 0); |
||
1864 | I915_WRITE(WM2_LP_ILK, 0); |
||
1865 | I915_WRITE(WM1_LP_ILK, 0); |
||
1866 | |||
1867 | if (!single_plane_enabled(enabled) || |
||
1868 | dev_priv->sprite_scaling_enabled) |
||
1869 | return; |
||
1870 | enabled = ffs(enabled) - 1; |
||
1871 | |||
1872 | /* WM1 */ |
||
1873 | if (!ironlake_compute_srwm(dev, 1, enabled, |
||
1874 | SNB_READ_WM1_LATENCY() * 500, |
||
1875 | &sandybridge_display_srwm_info, |
||
1876 | &sandybridge_cursor_srwm_info, |
||
1877 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
1878 | return; |
||
1879 | |||
1880 | I915_WRITE(WM1_LP_ILK, |
||
1881 | WM1_LP_SR_EN | |
||
1882 | (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
1883 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
1884 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
1885 | cursor_wm); |
||
1886 | |||
1887 | /* WM2 */ |
||
1888 | if (!ironlake_compute_srwm(dev, 2, enabled, |
||
1889 | SNB_READ_WM2_LATENCY() * 500, |
||
1890 | &sandybridge_display_srwm_info, |
||
1891 | &sandybridge_cursor_srwm_info, |
||
1892 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
1893 | return; |
||
1894 | |||
1895 | I915_WRITE(WM2_LP_ILK, |
||
1896 | WM2_LP_EN | |
||
1897 | (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
1898 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
1899 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
1900 | cursor_wm); |
||
1901 | |||
1902 | /* WM3 */ |
||
1903 | if (!ironlake_compute_srwm(dev, 3, enabled, |
||
1904 | SNB_READ_WM3_LATENCY() * 500, |
||
1905 | &sandybridge_display_srwm_info, |
||
1906 | &sandybridge_cursor_srwm_info, |
||
1907 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
1908 | return; |
||
1909 | |||
1910 | I915_WRITE(WM3_LP_ILK, |
||
1911 | WM3_LP_EN | |
||
1912 | (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
1913 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
1914 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
1915 | cursor_wm); |
||
1916 | } |
||
1917 | |||
1918 | static void ivybridge_update_wm(struct drm_device *dev) |
||
1919 | { |
||
1920 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1921 | int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */ |
||
1922 | u32 val; |
||
1923 | int fbc_wm, plane_wm, cursor_wm; |
||
1924 | int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm; |
||
1925 | unsigned int enabled; |
||
1926 | |||
1927 | enabled = 0; |
||
3746 | Serge | 1928 | if (g4x_compute_wm0(dev, PIPE_A, |
3031 | serge | 1929 | &sandybridge_display_wm_info, latency, |
1930 | &sandybridge_cursor_wm_info, latency, |
||
1931 | &plane_wm, &cursor_wm)) { |
||
3243 | Serge | 1932 | val = I915_READ(WM0_PIPEA_ILK); |
1933 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); |
||
1934 | I915_WRITE(WM0_PIPEA_ILK, val | |
||
1935 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); |
||
1936 | DRM_DEBUG_KMS("FIFO watermarks For pipe A -" |
||
1937 | " plane %d, " "cursor: %d\n", |
||
1938 | plane_wm, cursor_wm); |
||
3746 | Serge | 1939 | enabled |= 1 << PIPE_A; |
3243 | Serge | 1940 | } |
1941 | |||
3746 | Serge | 1942 | if (g4x_compute_wm0(dev, PIPE_B, |
3243 | Serge | 1943 | &sandybridge_display_wm_info, latency, |
1944 | &sandybridge_cursor_wm_info, latency, |
||
1945 | &plane_wm, &cursor_wm)) { |
||
1946 | val = I915_READ(WM0_PIPEB_ILK); |
||
1947 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); |
||
1948 | I915_WRITE(WM0_PIPEB_ILK, val | |
||
1949 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); |
||
1950 | DRM_DEBUG_KMS("FIFO watermarks For pipe B -" |
||
1951 | " plane %d, cursor: %d\n", |
||
1952 | plane_wm, cursor_wm); |
||
3746 | Serge | 1953 | enabled |= 1 << PIPE_B; |
3243 | Serge | 1954 | } |
1955 | |||
3746 | Serge | 1956 | if (g4x_compute_wm0(dev, PIPE_C, |
3243 | Serge | 1957 | &sandybridge_display_wm_info, latency, |
1958 | &sandybridge_cursor_wm_info, latency, |
||
1959 | &plane_wm, &cursor_wm)) { |
||
3031 | serge | 1960 | val = I915_READ(WM0_PIPEC_IVB); |
1961 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); |
||
1962 | I915_WRITE(WM0_PIPEC_IVB, val | |
||
1963 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); |
||
1964 | DRM_DEBUG_KMS("FIFO watermarks For pipe C -" |
||
1965 | " plane %d, cursor: %d\n", |
||
1966 | plane_wm, cursor_wm); |
||
3746 | Serge | 1967 | enabled |= 1 << PIPE_C; |
3031 | serge | 1968 | } |
1969 | |||
1970 | /* |
||
1971 | * Calculate and update the self-refresh watermark only when one |
||
1972 | * display plane is used. |
||
1973 | * |
||
1974 | * SNB support 3 levels of watermark. |
||
1975 | * |
||
1976 | * WM1/WM2/WM2 watermarks have to be enabled in the ascending order, |
||
1977 | * and disabled in the descending order |
||
1978 | * |
||
1979 | */ |
||
1980 | I915_WRITE(WM3_LP_ILK, 0); |
||
1981 | I915_WRITE(WM2_LP_ILK, 0); |
||
1982 | I915_WRITE(WM1_LP_ILK, 0); |
||
1983 | |||
1984 | if (!single_plane_enabled(enabled) || |
||
1985 | dev_priv->sprite_scaling_enabled) |
||
1986 | return; |
||
1987 | enabled = ffs(enabled) - 1; |
||
1988 | |||
1989 | /* WM1 */ |
||
1990 | if (!ironlake_compute_srwm(dev, 1, enabled, |
||
1991 | SNB_READ_WM1_LATENCY() * 500, |
||
1992 | &sandybridge_display_srwm_info, |
||
1993 | &sandybridge_cursor_srwm_info, |
||
1994 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
1995 | return; |
||
1996 | |||
1997 | I915_WRITE(WM1_LP_ILK, |
||
1998 | WM1_LP_SR_EN | |
||
1999 | (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
2000 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
2001 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
2002 | cursor_wm); |
||
2003 | |||
2004 | /* WM2 */ |
||
2005 | if (!ironlake_compute_srwm(dev, 2, enabled, |
||
2006 | SNB_READ_WM2_LATENCY() * 500, |
||
2007 | &sandybridge_display_srwm_info, |
||
2008 | &sandybridge_cursor_srwm_info, |
||
2009 | &fbc_wm, &plane_wm, &cursor_wm)) |
||
2010 | return; |
||
2011 | |||
2012 | I915_WRITE(WM2_LP_ILK, |
||
2013 | WM2_LP_EN | |
||
2014 | (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
2015 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
2016 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
2017 | cursor_wm); |
||
2018 | |||
3243 | Serge | 2019 | /* WM3, note we have to correct the cursor latency */ |
3031 | serge | 2020 | if (!ironlake_compute_srwm(dev, 3, enabled, |
2021 | SNB_READ_WM3_LATENCY() * 500, |
||
2022 | &sandybridge_display_srwm_info, |
||
2023 | &sandybridge_cursor_srwm_info, |
||
3243 | Serge | 2024 | &fbc_wm, &plane_wm, &ignore_cursor_wm) || |
2025 | !ironlake_compute_srwm(dev, 3, enabled, |
||
2026 | 2 * SNB_READ_WM3_LATENCY() * 500, |
||
2027 | &sandybridge_display_srwm_info, |
||
2028 | &sandybridge_cursor_srwm_info, |
||
2029 | &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm)) |
||
3031 | serge | 2030 | return; |
2031 | |||
2032 | I915_WRITE(WM3_LP_ILK, |
||
2033 | WM3_LP_EN | |
||
2034 | (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) | |
||
2035 | (fbc_wm << WM1_LP_FBC_SHIFT) | |
||
2036 | (plane_wm << WM1_LP_SR_SHIFT) | |
||
2037 | cursor_wm); |
||
2038 | } |
||
2039 | |||
2040 | static void |
||
2041 | haswell_update_linetime_wm(struct drm_device *dev, int pipe, |
||
2042 | struct drm_display_mode *mode) |
||
2043 | { |
||
2044 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2045 | u32 temp; |
||
2046 | |||
2047 | temp = I915_READ(PIPE_WM_LINETIME(pipe)); |
||
2048 | temp &= ~PIPE_WM_LINETIME_MASK; |
||
2049 | |||
2050 | /* The WM are computed with base on how long it takes to fill a single |
||
2051 | * row at the given clock rate, multiplied by 8. |
||
2052 | * */ |
||
2053 | temp |= PIPE_WM_LINETIME_TIME( |
||
2054 | ((mode->crtc_hdisplay * 1000) / mode->clock) * 8); |
||
2055 | |||
2056 | /* IPS watermarks are only used by pipe A, and are ignored by |
||
2057 | * pipes B and C. They are calculated similarly to the common |
||
2058 | * linetime values, except that we are using CD clock frequency |
||
2059 | * in MHz instead of pixel rate for the division. |
||
2060 | * |
||
2061 | * This is a placeholder for the IPS watermark calculation code. |
||
2062 | */ |
||
2063 | |||
2064 | I915_WRITE(PIPE_WM_LINETIME(pipe), temp); |
||
2065 | } |
||
2066 | |||
2067 | static bool |
||
2068 | sandybridge_compute_sprite_wm(struct drm_device *dev, int plane, |
||
2069 | uint32_t sprite_width, int pixel_size, |
||
2070 | const struct intel_watermark_params *display, |
||
2071 | int display_latency_ns, int *sprite_wm) |
||
2072 | { |
||
2073 | struct drm_crtc *crtc; |
||
2074 | int clock; |
||
2075 | int entries, tlb_miss; |
||
2076 | |||
2077 | crtc = intel_get_crtc_for_plane(dev, plane); |
||
3243 | Serge | 2078 | if (!intel_crtc_active(crtc)) { |
3031 | serge | 2079 | *sprite_wm = display->guard_size; |
2080 | return false; |
||
2081 | } |
||
2082 | |||
2083 | clock = crtc->mode.clock; |
||
2084 | |||
2085 | /* Use the small buffer method to calculate the sprite watermark */ |
||
2086 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; |
||
2087 | tlb_miss = display->fifo_size*display->cacheline_size - |
||
2088 | sprite_width * 8; |
||
2089 | if (tlb_miss > 0) |
||
2090 | entries += tlb_miss; |
||
2091 | entries = DIV_ROUND_UP(entries, display->cacheline_size); |
||
2092 | *sprite_wm = entries + display->guard_size; |
||
2093 | if (*sprite_wm > (int)display->max_wm) |
||
2094 | *sprite_wm = display->max_wm; |
||
2095 | |||
2096 | return true; |
||
2097 | } |
||
2098 | |||
2099 | static bool |
||
2100 | sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane, |
||
2101 | uint32_t sprite_width, int pixel_size, |
||
2102 | const struct intel_watermark_params *display, |
||
2103 | int latency_ns, int *sprite_wm) |
||
2104 | { |
||
2105 | struct drm_crtc *crtc; |
||
2106 | unsigned long line_time_us; |
||
2107 | int clock; |
||
2108 | int line_count, line_size; |
||
2109 | int small, large; |
||
2110 | int entries; |
||
2111 | |||
2112 | if (!latency_ns) { |
||
2113 | *sprite_wm = 0; |
||
2114 | return false; |
||
2115 | } |
||
2116 | |||
2117 | crtc = intel_get_crtc_for_plane(dev, plane); |
||
2118 | clock = crtc->mode.clock; |
||
2119 | if (!clock) { |
||
2120 | *sprite_wm = 0; |
||
2121 | return false; |
||
2122 | } |
||
2123 | |||
2124 | line_time_us = (sprite_width * 1000) / clock; |
||
2125 | if (!line_time_us) { |
||
2126 | *sprite_wm = 0; |
||
2127 | return false; |
||
2128 | } |
||
2129 | |||
2130 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
||
2131 | line_size = sprite_width * pixel_size; |
||
2132 | |||
2133 | /* Use the minimum of the small and large buffer method for primary */ |
||
2134 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; |
||
2135 | large = line_count * line_size; |
||
2136 | |||
2137 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); |
||
2138 | *sprite_wm = entries + display->guard_size; |
||
2139 | |||
2140 | return *sprite_wm > 0x3ff ? false : true; |
||
2141 | } |
||
2142 | |||
2143 | static void sandybridge_update_sprite_wm(struct drm_device *dev, int pipe, |
||
2144 | uint32_t sprite_width, int pixel_size) |
||
2145 | { |
||
2146 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2147 | int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */ |
||
2148 | u32 val; |
||
2149 | int sprite_wm, reg; |
||
2150 | int ret; |
||
2151 | |||
2152 | switch (pipe) { |
||
2153 | case 0: |
||
2154 | reg = WM0_PIPEA_ILK; |
||
2155 | break; |
||
2156 | case 1: |
||
2157 | reg = WM0_PIPEB_ILK; |
||
2158 | break; |
||
2159 | case 2: |
||
2160 | reg = WM0_PIPEC_IVB; |
||
2161 | break; |
||
2162 | default: |
||
2163 | return; /* bad pipe */ |
||
2164 | } |
||
2165 | |||
2166 | ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size, |
||
2167 | &sandybridge_display_wm_info, |
||
2168 | latency, &sprite_wm); |
||
2169 | if (!ret) { |
||
2170 | DRM_DEBUG_KMS("failed to compute sprite wm for pipe %d\n", |
||
2171 | pipe); |
||
2172 | return; |
||
2173 | } |
||
2174 | |||
2175 | val = I915_READ(reg); |
||
2176 | val &= ~WM0_PIPE_SPRITE_MASK; |
||
2177 | I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT)); |
||
2178 | DRM_DEBUG_KMS("sprite watermarks For pipe %d - %d\n", pipe, sprite_wm); |
||
2179 | |||
2180 | |||
2181 | ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width, |
||
2182 | pixel_size, |
||
2183 | &sandybridge_display_srwm_info, |
||
2184 | SNB_READ_WM1_LATENCY() * 500, |
||
2185 | &sprite_wm); |
||
2186 | if (!ret) { |
||
2187 | DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %d\n", |
||
2188 | pipe); |
||
2189 | return; |
||
2190 | } |
||
2191 | I915_WRITE(WM1S_LP_ILK, sprite_wm); |
||
2192 | |||
2193 | /* Only IVB has two more LP watermarks for sprite */ |
||
2194 | if (!IS_IVYBRIDGE(dev)) |
||
2195 | return; |
||
2196 | |||
2197 | ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width, |
||
2198 | pixel_size, |
||
2199 | &sandybridge_display_srwm_info, |
||
2200 | SNB_READ_WM2_LATENCY() * 500, |
||
2201 | &sprite_wm); |
||
2202 | if (!ret) { |
||
2203 | DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %d\n", |
||
2204 | pipe); |
||
2205 | return; |
||
2206 | } |
||
2207 | I915_WRITE(WM2S_LP_IVB, sprite_wm); |
||
2208 | |||
2209 | ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width, |
||
2210 | pixel_size, |
||
2211 | &sandybridge_display_srwm_info, |
||
2212 | SNB_READ_WM3_LATENCY() * 500, |
||
2213 | &sprite_wm); |
||
2214 | if (!ret) { |
||
2215 | DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %d\n", |
||
2216 | pipe); |
||
2217 | return; |
||
2218 | } |
||
2219 | I915_WRITE(WM3S_LP_IVB, sprite_wm); |
||
2220 | } |
||
2221 | |||
2222 | /** |
||
2223 | * intel_update_watermarks - update FIFO watermark values based on current modes |
||
2224 | * |
||
2225 | * Calculate watermark values for the various WM regs based on current mode |
||
2226 | * and plane configuration. |
||
2227 | * |
||
2228 | * There are several cases to deal with here: |
||
2229 | * - normal (i.e. non-self-refresh) |
||
2230 | * - self-refresh (SR) mode |
||
2231 | * - lines are large relative to FIFO size (buffer can hold up to 2) |
||
2232 | * - lines are small relative to FIFO size (buffer can hold more than 2 |
||
2233 | * lines), so need to account for TLB latency |
||
2234 | * |
||
2235 | * The normal calculation is: |
||
2236 | * watermark = dotclock * bytes per pixel * latency |
||
2237 | * where latency is platform & configuration dependent (we assume pessimal |
||
2238 | * values here). |
||
2239 | * |
||
2240 | * The SR calculation is: |
||
2241 | * watermark = (trunc(latency/line time)+1) * surface width * |
||
2242 | * bytes per pixel |
||
2243 | * where |
||
2244 | * line time = htotal / dotclock |
||
2245 | * surface width = hdisplay for normal plane and 64 for cursor |
||
2246 | * and latency is assumed to be high, as above. |
||
2247 | * |
||
2248 | * The final value programmed to the register should always be rounded up, |
||
2249 | * and include an extra 2 entries to account for clock crossings. |
||
2250 | * |
||
2251 | * We don't use the sprite, so we can ignore that. And on Crestline we have |
||
2252 | * to set the non-SR watermarks to 8. |
||
2253 | */ |
||
2254 | void intel_update_watermarks(struct drm_device *dev) |
||
2255 | { |
||
2256 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2257 | |||
2258 | if (dev_priv->display.update_wm) |
||
2259 | dev_priv->display.update_wm(dev); |
||
2260 | } |
||
2261 | |||
2262 | void intel_update_linetime_watermarks(struct drm_device *dev, |
||
2263 | int pipe, struct drm_display_mode *mode) |
||
2264 | { |
||
2265 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2266 | |||
2267 | if (dev_priv->display.update_linetime_wm) |
||
2268 | dev_priv->display.update_linetime_wm(dev, pipe, mode); |
||
2269 | } |
||
2270 | |||
2271 | void intel_update_sprite_watermarks(struct drm_device *dev, int pipe, |
||
2272 | uint32_t sprite_width, int pixel_size) |
||
2273 | { |
||
2274 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2275 | |||
2276 | if (dev_priv->display.update_sprite_wm) |
||
2277 | dev_priv->display.update_sprite_wm(dev, pipe, sprite_width, |
||
2278 | pixel_size); |
||
2279 | } |
||
2280 | |||
2281 | static struct drm_i915_gem_object * |
||
2282 | intel_alloc_context_page(struct drm_device *dev) |
||
2283 | { |
||
2284 | struct drm_i915_gem_object *ctx; |
||
2285 | int ret; |
||
2286 | |||
2287 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
||
2288 | |||
2289 | ctx = i915_gem_alloc_object(dev, 4096); |
||
2290 | if (!ctx) { |
||
2291 | DRM_DEBUG("failed to alloc power context, RC6 disabled\n"); |
||
2292 | return NULL; |
||
2293 | } |
||
2294 | |||
2295 | ret = i915_gem_object_pin(ctx, 4096, true, false); |
||
2296 | if (ret) { |
||
2297 | DRM_ERROR("failed to pin power context: %d\n", ret); |
||
2298 | goto err_unref; |
||
2299 | } |
||
2300 | |||
2301 | ret = i915_gem_object_set_to_gtt_domain(ctx, 1); |
||
2302 | if (ret) { |
||
2303 | DRM_ERROR("failed to set-domain on power context: %d\n", ret); |
||
2304 | goto err_unpin; |
||
2305 | } |
||
2306 | |||
2307 | return ctx; |
||
2308 | |||
2309 | err_unpin: |
||
2310 | i915_gem_object_unpin(ctx); |
||
2311 | err_unref: |
||
2312 | drm_gem_object_unreference(&ctx->base); |
||
2313 | return NULL; |
||
2314 | } |
||
2315 | |||
2316 | /** |
||
2317 | * Lock protecting IPS related data structures |
||
2318 | */ |
||
2319 | DEFINE_SPINLOCK(mchdev_lock); |
||
2320 | |||
2321 | /* Global for IPS driver to get at the current i915 device. Protected by |
||
2322 | * mchdev_lock. */ |
||
2323 | static struct drm_i915_private *i915_mch_dev; |
||
2324 | |||
2325 | bool ironlake_set_drps(struct drm_device *dev, u8 val) |
||
2326 | { |
||
2327 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2328 | u16 rgvswctl; |
||
2329 | |||
2330 | assert_spin_locked(&mchdev_lock); |
||
2331 | |||
2332 | rgvswctl = I915_READ16(MEMSWCTL); |
||
2333 | if (rgvswctl & MEMCTL_CMD_STS) { |
||
2334 | DRM_DEBUG("gpu busy, RCS change rejected\n"); |
||
2335 | return false; /* still busy with another command */ |
||
2336 | } |
||
2337 | |||
2338 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | |
||
2339 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; |
||
2340 | I915_WRITE16(MEMSWCTL, rgvswctl); |
||
2341 | POSTING_READ16(MEMSWCTL); |
||
2342 | |||
2343 | rgvswctl |= MEMCTL_CMD_STS; |
||
2344 | I915_WRITE16(MEMSWCTL, rgvswctl); |
||
2345 | |||
2346 | return true; |
||
2347 | } |
||
2348 | |||
2349 | static void ironlake_enable_drps(struct drm_device *dev) |
||
2350 | { |
||
2351 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2352 | u32 rgvmodectl = I915_READ(MEMMODECTL); |
||
2353 | u8 fmax, fmin, fstart, vstart; |
||
2354 | |||
2355 | spin_lock_irq(&mchdev_lock); |
||
2356 | |||
2357 | /* Enable temp reporting */ |
||
2358 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); |
||
2359 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); |
||
2360 | |||
2361 | /* 100ms RC evaluation intervals */ |
||
2362 | I915_WRITE(RCUPEI, 100000); |
||
2363 | I915_WRITE(RCDNEI, 100000); |
||
2364 | |||
2365 | /* Set max/min thresholds to 90ms and 80ms respectively */ |
||
2366 | I915_WRITE(RCBMAXAVG, 90000); |
||
2367 | I915_WRITE(RCBMINAVG, 80000); |
||
2368 | |||
2369 | I915_WRITE(MEMIHYST, 1); |
||
2370 | |||
2371 | /* Set up min, max, and cur for interrupt handling */ |
||
2372 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; |
||
2373 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); |
||
2374 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> |
||
2375 | MEMMODE_FSTART_SHIFT; |
||
2376 | |||
2377 | vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >> |
||
2378 | PXVFREQ_PX_SHIFT; |
||
2379 | |||
2380 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
||
2381 | dev_priv->ips.fstart = fstart; |
||
2382 | |||
2383 | dev_priv->ips.max_delay = fstart; |
||
2384 | dev_priv->ips.min_delay = fmin; |
||
2385 | dev_priv->ips.cur_delay = fstart; |
||
2386 | |||
2387 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", |
||
2388 | fmax, fmin, fstart); |
||
2389 | |||
2390 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); |
||
2391 | |||
2392 | /* |
||
2393 | * Interrupts will be enabled in ironlake_irq_postinstall |
||
2394 | */ |
||
2395 | |||
2396 | I915_WRITE(VIDSTART, vstart); |
||
2397 | POSTING_READ(VIDSTART); |
||
2398 | |||
2399 | rgvmodectl |= MEMMODE_SWMODE_EN; |
||
2400 | I915_WRITE(MEMMODECTL, rgvmodectl); |
||
2401 | |||
2402 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
||
2403 | DRM_ERROR("stuck trying to change perf mode\n"); |
||
2404 | mdelay(1); |
||
2405 | |||
2406 | ironlake_set_drps(dev, fstart); |
||
2407 | |||
2408 | dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) + |
||
2409 | I915_READ(0x112e0); |
||
2410 | dev_priv->ips.last_time1 = jiffies_to_msecs(GetTimerTicks()); |
||
2411 | dev_priv->ips.last_count2 = I915_READ(0x112f4); |
||
3482 | Serge | 2412 | getrawmonotonic(&dev_priv->ips.last_time2); |
3031 | serge | 2413 | |
2414 | spin_unlock_irq(&mchdev_lock); |
||
2415 | } |
||
2416 | |||
2417 | static void ironlake_disable_drps(struct drm_device *dev) |
||
2418 | { |
||
2419 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2420 | u16 rgvswctl; |
||
2421 | |||
2422 | spin_lock_irq(&mchdev_lock); |
||
2423 | |||
2424 | rgvswctl = I915_READ16(MEMSWCTL); |
||
2425 | |||
2426 | /* Ack interrupts, disable EFC interrupt */ |
||
2427 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); |
||
2428 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); |
||
2429 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); |
||
2430 | I915_WRITE(DEIIR, DE_PCU_EVENT); |
||
2431 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); |
||
2432 | |||
2433 | /* Go back to the starting frequency */ |
||
2434 | ironlake_set_drps(dev, dev_priv->ips.fstart); |
||
2435 | mdelay(1); |
||
2436 | rgvswctl |= MEMCTL_CMD_STS; |
||
2437 | I915_WRITE(MEMSWCTL, rgvswctl); |
||
2438 | mdelay(1); |
||
2439 | |||
2440 | spin_unlock_irq(&mchdev_lock); |
||
2441 | } |
||
2442 | |||
2443 | /* There's a funny hw issue where the hw returns all 0 when reading from |
||
2444 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value |
||
2445 | * ourselves, instead of doing a rmw cycle (which might result in us clearing |
||
2446 | * all limits and the gpu stuck at whatever frequency it is at atm). |
||
2447 | */ |
||
2448 | static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 *val) |
||
2449 | { |
||
2450 | u32 limits; |
||
2451 | |||
2452 | limits = 0; |
||
2453 | |||
2454 | if (*val >= dev_priv->rps.max_delay) |
||
2455 | *val = dev_priv->rps.max_delay; |
||
2456 | limits |= dev_priv->rps.max_delay << 24; |
||
2457 | |||
2458 | /* Only set the down limit when we've reached the lowest level to avoid |
||
2459 | * getting more interrupts, otherwise leave this clear. This prevents a |
||
2460 | * race in the hw when coming out of rc6: There's a tiny window where |
||
2461 | * the hw runs at the minimal clock before selecting the desired |
||
2462 | * frequency, if the down threshold expires in that window we will not |
||
2463 | * receive a down interrupt. */ |
||
2464 | if (*val <= dev_priv->rps.min_delay) { |
||
2465 | *val = dev_priv->rps.min_delay; |
||
2466 | limits |= dev_priv->rps.min_delay << 16; |
||
2467 | } |
||
2468 | |||
2469 | return limits; |
||
2470 | } |
||
2471 | |||
2472 | void gen6_set_rps(struct drm_device *dev, u8 val) |
||
2473 | { |
||
2474 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2475 | u32 limits = gen6_rps_limits(dev_priv, &val); |
||
2476 | |||
3243 | Serge | 2477 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
3031 | serge | 2478 | WARN_ON(val > dev_priv->rps.max_delay); |
2479 | WARN_ON(val < dev_priv->rps.min_delay); |
||
2480 | |||
2481 | if (val == dev_priv->rps.cur_delay) |
||
2482 | return; |
||
2483 | |||
3746 | Serge | 2484 | if (IS_HASWELL(dev)) |
2485 | I915_WRITE(GEN6_RPNSWREQ, |
||
2486 | HSW_FREQUENCY(val)); |
||
2487 | else |
||
3031 | serge | 2488 | I915_WRITE(GEN6_RPNSWREQ, |
2489 | GEN6_FREQUENCY(val) | |
||
2490 | GEN6_OFFSET(0) | |
||
2491 | GEN6_AGGRESSIVE_TURBO); |
||
2492 | |||
2493 | /* Make sure we continue to get interrupts |
||
2494 | * until we hit the minimum or maximum frequencies. |
||
2495 | */ |
||
2496 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits); |
||
2497 | |||
2498 | POSTING_READ(GEN6_RPNSWREQ); |
||
2499 | |||
2500 | dev_priv->rps.cur_delay = val; |
||
2501 | |||
2502 | trace_intel_gpu_freq_change(val * 50); |
||
2503 | } |
||
2504 | |||
2505 | static void gen6_disable_rps(struct drm_device *dev) |
||
2506 | { |
||
2507 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2508 | |||
2509 | I915_WRITE(GEN6_RC_CONTROL, 0); |
||
2510 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
||
2511 | I915_WRITE(GEN6_PMINTRMSK, 0xffffffff); |
||
2512 | I915_WRITE(GEN6_PMIER, 0); |
||
2513 | /* Complete PM interrupt masking here doesn't race with the rps work |
||
2514 | * item again unmasking PM interrupts because that is using a different |
||
2515 | * register (PMIMR) to mask PM interrupts. The only risk is in leaving |
||
2516 | * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */ |
||
2517 | |||
2518 | spin_lock_irq(&dev_priv->rps.lock); |
||
2519 | dev_priv->rps.pm_iir = 0; |
||
2520 | spin_unlock_irq(&dev_priv->rps.lock); |
||
2521 | |||
2522 | I915_WRITE(GEN6_PMIIR, I915_READ(GEN6_PMIIR)); |
||
2523 | } |
||
2524 | |||
2525 | int intel_enable_rc6(const struct drm_device *dev) |
||
2526 | { |
||
2527 | /* Respect the kernel parameter if it is set */ |
||
2528 | if (i915_enable_rc6 >= 0) |
||
2529 | return i915_enable_rc6; |
||
2530 | |||
3120 | serge | 2531 | /* Disable RC6 on Ironlake */ |
2532 | if (INTEL_INFO(dev)->gen == 5) |
||
2533 | return 0; |
||
3031 | serge | 2534 | |
2535 | if (IS_HASWELL(dev)) { |
||
2536 | DRM_DEBUG_DRIVER("Haswell: only RC6 available\n"); |
||
2537 | return INTEL_RC6_ENABLE; |
||
2538 | } |
||
2539 | |||
2540 | /* snb/ivb have more than one rc6 state. */ |
||
2541 | if (INTEL_INFO(dev)->gen == 6) { |
||
2542 | DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n"); |
||
2543 | return INTEL_RC6_ENABLE; |
||
2544 | } |
||
2545 | |||
2546 | DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n"); |
||
2547 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); |
||
2548 | } |
||
2549 | |||
2550 | static void gen6_enable_rps(struct drm_device *dev) |
||
2551 | { |
||
2552 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2553 | struct intel_ring_buffer *ring; |
||
2554 | u32 rp_state_cap; |
||
2555 | u32 gt_perf_status; |
||
3243 | Serge | 2556 | u32 rc6vids, pcu_mbox, rc6_mask = 0; |
3031 | serge | 2557 | u32 gtfifodbg; |
2558 | int rc6_mode; |
||
3243 | Serge | 2559 | int i, ret; |
3031 | serge | 2560 | |
3243 | Serge | 2561 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
3031 | serge | 2562 | |
2563 | /* Here begins a magic sequence of register writes to enable |
||
2564 | * auto-downclocking. |
||
2565 | * |
||
2566 | * Perhaps there might be some value in exposing these to |
||
2567 | * userspace... |
||
2568 | */ |
||
2569 | I915_WRITE(GEN6_RC_STATE, 0); |
||
2570 | |||
2571 | /* Clear the DBG now so we don't confuse earlier errors */ |
||
2572 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { |
||
2573 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); |
||
2574 | I915_WRITE(GTFIFODBG, gtfifodbg); |
||
2575 | } |
||
2576 | |||
2577 | gen6_gt_force_wake_get(dev_priv); |
||
2578 | |||
2579 | rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); |
||
2580 | gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS); |
||
2581 | |||
3746 | Serge | 2582 | /* In units of 50MHz */ |
2583 | dev_priv->rps.hw_max = dev_priv->rps.max_delay = rp_state_cap & 0xff; |
||
3031 | serge | 2584 | dev_priv->rps.min_delay = (rp_state_cap & 0xff0000) >> 16; |
2585 | dev_priv->rps.cur_delay = 0; |
||
2586 | |||
2587 | /* disable the counters and set deterministic thresholds */ |
||
2588 | I915_WRITE(GEN6_RC_CONTROL, 0); |
||
2589 | |||
2590 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); |
||
2591 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); |
||
2592 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); |
||
2593 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); |
||
2594 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); |
||
2595 | |||
2596 | for_each_ring(ring, dev_priv, i) |
||
2597 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); |
||
2598 | |||
2599 | I915_WRITE(GEN6_RC_SLEEP, 0); |
||
2600 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); |
||
2601 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); |
||
3480 | Serge | 2602 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
3031 | serge | 2603 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
2604 | |||
2605 | /* Check if we are enabling RC6 */ |
||
2606 | rc6_mode = intel_enable_rc6(dev_priv->dev); |
||
2607 | if (rc6_mode & INTEL_RC6_ENABLE) |
||
2608 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; |
||
2609 | |||
2610 | /* We don't use those on Haswell */ |
||
2611 | if (!IS_HASWELL(dev)) { |
||
2612 | if (rc6_mode & INTEL_RC6p_ENABLE) |
||
2613 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; |
||
2614 | |||
2615 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
||
2616 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; |
||
2617 | } |
||
2618 | |||
2619 | DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n", |
||
2620 | (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off", |
||
2621 | (rc6_mask & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off", |
||
2622 | (rc6_mask & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off"); |
||
2623 | |||
2624 | I915_WRITE(GEN6_RC_CONTROL, |
||
2625 | rc6_mask | |
||
2626 | GEN6_RC_CTL_EI_MODE(1) | |
||
2627 | GEN6_RC_CTL_HW_ENABLE); |
||
2628 | |||
3746 | Serge | 2629 | if (IS_HASWELL(dev)) { |
2630 | I915_WRITE(GEN6_RPNSWREQ, |
||
2631 | HSW_FREQUENCY(10)); |
||
2632 | I915_WRITE(GEN6_RC_VIDEO_FREQ, |
||
2633 | HSW_FREQUENCY(12)); |
||
2634 | } else { |
||
3031 | serge | 2635 | I915_WRITE(GEN6_RPNSWREQ, |
2636 | GEN6_FREQUENCY(10) | |
||
2637 | GEN6_OFFSET(0) | |
||
2638 | GEN6_AGGRESSIVE_TURBO); |
||
2639 | I915_WRITE(GEN6_RC_VIDEO_FREQ, |
||
2640 | GEN6_FREQUENCY(12)); |
||
3746 | Serge | 2641 | } |
3031 | serge | 2642 | |
2643 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
||
2644 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, |
||
2645 | dev_priv->rps.max_delay << 24 | |
||
2646 | dev_priv->rps.min_delay << 16); |
||
2647 | |||
2648 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
||
2649 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); |
||
2650 | I915_WRITE(GEN6_RP_UP_EI, 66000); |
||
2651 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); |
||
2652 | |||
2653 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
||
2654 | I915_WRITE(GEN6_RP_CONTROL, |
||
2655 | GEN6_RP_MEDIA_TURBO | |
||
2656 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
||
2657 | GEN6_RP_MEDIA_IS_GFX | |
||
2658 | GEN6_RP_ENABLE | |
||
2659 | GEN6_RP_UP_BUSY_AVG | |
||
2660 | (IS_HASWELL(dev) ? GEN7_RP_DOWN_IDLE_AVG : GEN6_RP_DOWN_IDLE_CONT)); |
||
2661 | |||
3243 | Serge | 2662 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0); |
2663 | if (!ret) { |
||
2664 | pcu_mbox = 0; |
||
2665 | ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox); |
||
3746 | Serge | 2666 | if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */ |
2667 | DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n", |
||
2668 | (dev_priv->rps.max_delay & 0xff) * 50, |
||
2669 | (pcu_mbox & 0xff) * 50); |
||
2670 | dev_priv->rps.hw_max = pcu_mbox & 0xff; |
||
3031 | serge | 2671 | } |
3243 | Serge | 2672 | } else { |
2673 | DRM_DEBUG_DRIVER("Failed to set the min frequency\n"); |
||
2674 | } |
||
3031 | serge | 2675 | |
2676 | gen6_set_rps(dev_priv->dev, (gt_perf_status & 0xff00) >> 8); |
||
2677 | |||
2678 | /* requires MSI enabled */ |
||
2679 | I915_WRITE(GEN6_PMIER, GEN6_PM_DEFERRED_EVENTS); |
||
2680 | spin_lock_irq(&dev_priv->rps.lock); |
||
2681 | WARN_ON(dev_priv->rps.pm_iir != 0); |
||
2682 | I915_WRITE(GEN6_PMIMR, 0); |
||
2683 | spin_unlock_irq(&dev_priv->rps.lock); |
||
2684 | /* enable all PM interrupts */ |
||
2685 | I915_WRITE(GEN6_PMINTRMSK, 0); |
||
2686 | |||
3243 | Serge | 2687 | rc6vids = 0; |
2688 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); |
||
2689 | if (IS_GEN6(dev) && ret) { |
||
2690 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); |
||
2691 | } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { |
||
2692 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", |
||
2693 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); |
||
2694 | rc6vids &= 0xffff00; |
||
2695 | rc6vids |= GEN6_ENCODE_RC6_VID(450); |
||
2696 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); |
||
2697 | if (ret) |
||
2698 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); |
||
2699 | } |
||
2700 | |||
3031 | serge | 2701 | gen6_gt_force_wake_put(dev_priv); |
2702 | } |
||
2703 | |||
2704 | static void gen6_update_ring_freq(struct drm_device *dev) |
||
2705 | { |
||
2706 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2707 | int min_freq = 15; |
||
3746 | Serge | 2708 | unsigned int gpu_freq; |
2709 | unsigned int max_ia_freq, min_ring_freq; |
||
3031 | serge | 2710 | int scaling_factor = 180; |
2711 | |||
3243 | Serge | 2712 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
3031 | serge | 2713 | |
2714 | max_ia_freq = cpufreq_quick_get_max(0); |
||
2715 | /* |
||
2716 | * Default to measured freq if none found, PCU will ensure we don't go |
||
2717 | * over |
||
2718 | */ |
||
2719 | if (!max_ia_freq) |
||
2720 | max_ia_freq = tsc_khz; |
||
2721 | |||
2722 | /* Convert from kHz to MHz */ |
||
2723 | max_ia_freq /= 1000; |
||
2724 | |||
3746 | Serge | 2725 | min_ring_freq = I915_READ(MCHBAR_MIRROR_BASE_SNB + DCLK); |
2726 | /* convert DDR frequency from units of 133.3MHz to bandwidth */ |
||
2727 | min_ring_freq = (2 * 4 * min_ring_freq + 2) / 3; |
||
2728 | |||
3031 | serge | 2729 | /* |
2730 | * For each potential GPU frequency, load a ring frequency we'd like |
||
2731 | * to use for memory access. We do this by specifying the IA frequency |
||
2732 | * the PCU should use as a reference to determine the ring frequency. |
||
2733 | */ |
||
2734 | for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay; |
||
2735 | gpu_freq--) { |
||
2736 | int diff = dev_priv->rps.max_delay - gpu_freq; |
||
3746 | Serge | 2737 | unsigned int ia_freq = 0, ring_freq = 0; |
3031 | serge | 2738 | |
3746 | Serge | 2739 | if (IS_HASWELL(dev)) { |
2740 | ring_freq = (gpu_freq * 5 + 3) / 4; |
||
2741 | ring_freq = max(min_ring_freq, ring_freq); |
||
2742 | /* leave ia_freq as the default, chosen by cpufreq */ |
||
2743 | } else { |
||
2744 | /* On older processors, there is no separate ring |
||
2745 | * clock domain, so in order to boost the bandwidth |
||
2746 | * of the ring, we need to upclock the CPU (ia_freq). |
||
2747 | * |
||
2748 | * For GPU frequencies less than 750MHz, |
||
2749 | * just use the lowest ring freq. |
||
3031 | serge | 2750 | */ |
2751 | if (gpu_freq < min_freq) |
||
2752 | ia_freq = 800; |
||
2753 | else |
||
2754 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); |
||
2755 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); |
||
3746 | Serge | 2756 | } |
3031 | serge | 2757 | |
3243 | Serge | 2758 | sandybridge_pcode_write(dev_priv, |
2759 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, |
||
3746 | Serge | 2760 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
2761 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | |
||
2762 | gpu_freq); |
||
3031 | serge | 2763 | } |
2764 | } |
||
2765 | |||
2766 | void ironlake_teardown_rc6(struct drm_device *dev) |
||
2767 | { |
||
2768 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2769 | |||
3243 | Serge | 2770 | if (dev_priv->ips.renderctx) { |
2771 | i915_gem_object_unpin(dev_priv->ips.renderctx); |
||
2772 | drm_gem_object_unreference(&dev_priv->ips.renderctx->base); |
||
2773 | dev_priv->ips.renderctx = NULL; |
||
3031 | serge | 2774 | } |
2775 | |||
3243 | Serge | 2776 | if (dev_priv->ips.pwrctx) { |
2777 | i915_gem_object_unpin(dev_priv->ips.pwrctx); |
||
2778 | drm_gem_object_unreference(&dev_priv->ips.pwrctx->base); |
||
2779 | dev_priv->ips.pwrctx = NULL; |
||
3031 | serge | 2780 | } |
2781 | } |
||
2782 | |||
2783 | static void ironlake_disable_rc6(struct drm_device *dev) |
||
2784 | { |
||
2785 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2786 | |||
2787 | if (I915_READ(PWRCTXA)) { |
||
2788 | /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */ |
||
2789 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT); |
||
2790 | wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON), |
||
2791 | 50); |
||
2792 | |||
2793 | I915_WRITE(PWRCTXA, 0); |
||
2794 | POSTING_READ(PWRCTXA); |
||
2795 | |||
2796 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT); |
||
2797 | POSTING_READ(RSTDBYCTL); |
||
2798 | } |
||
2799 | } |
||
2800 | |||
2801 | static int ironlake_setup_rc6(struct drm_device *dev) |
||
2802 | { |
||
2803 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2804 | |||
3243 | Serge | 2805 | if (dev_priv->ips.renderctx == NULL) |
2806 | dev_priv->ips.renderctx = intel_alloc_context_page(dev); |
||
2807 | if (!dev_priv->ips.renderctx) |
||
3031 | serge | 2808 | return -ENOMEM; |
2809 | |||
3243 | Serge | 2810 | if (dev_priv->ips.pwrctx == NULL) |
2811 | dev_priv->ips.pwrctx = intel_alloc_context_page(dev); |
||
2812 | if (!dev_priv->ips.pwrctx) { |
||
3031 | serge | 2813 | ironlake_teardown_rc6(dev); |
2814 | return -ENOMEM; |
||
2815 | } |
||
2816 | |||
2817 | return 0; |
||
2818 | } |
||
2819 | |||
2820 | static void ironlake_enable_rc6(struct drm_device *dev) |
||
2821 | { |
||
2822 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
2823 | struct intel_ring_buffer *ring = &dev_priv->ring[RCS]; |
||
3243 | Serge | 2824 | bool was_interruptible; |
3031 | serge | 2825 | int ret; |
2826 | |||
2827 | /* rc6 disabled by default due to repeated reports of hanging during |
||
2828 | * boot and resume. |
||
2829 | */ |
||
2830 | if (!intel_enable_rc6(dev)) |
||
2831 | return; |
||
2832 | |||
2833 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
||
2834 | |||
2835 | ret = ironlake_setup_rc6(dev); |
||
2836 | if (ret) |
||
2837 | return; |
||
2838 | |||
3243 | Serge | 2839 | was_interruptible = dev_priv->mm.interruptible; |
2840 | dev_priv->mm.interruptible = false; |
||
2841 | |||
3031 | serge | 2842 | /* |
2843 | * GPU can automatically power down the render unit if given a page |
||
2844 | * to save state. |
||
2845 | */ |
||
2846 | ret = intel_ring_begin(ring, 6); |
||
2847 | if (ret) { |
||
2848 | ironlake_teardown_rc6(dev); |
||
3243 | Serge | 2849 | dev_priv->mm.interruptible = was_interruptible; |
3031 | serge | 2850 | return; |
2851 | } |
||
2852 | |||
2853 | intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN); |
||
2854 | intel_ring_emit(ring, MI_SET_CONTEXT); |
||
3243 | Serge | 2855 | intel_ring_emit(ring, dev_priv->ips.renderctx->gtt_offset | |
3031 | serge | 2856 | MI_MM_SPACE_GTT | |
2857 | MI_SAVE_EXT_STATE_EN | |
||
2858 | MI_RESTORE_EXT_STATE_EN | |
||
2859 | MI_RESTORE_INHIBIT); |
||
2860 | intel_ring_emit(ring, MI_SUSPEND_FLUSH); |
||
2861 | intel_ring_emit(ring, MI_NOOP); |
||
2862 | intel_ring_emit(ring, MI_FLUSH); |
||
2863 | intel_ring_advance(ring); |
||
2864 | |||
2865 | /* |
||
2866 | * Wait for the command parser to advance past MI_SET_CONTEXT. The HW |
||
2867 | * does an implicit flush, combined with MI_FLUSH above, it should be |
||
2868 | * safe to assume that renderctx is valid |
||
2869 | */ |
||
3243 | Serge | 2870 | ret = intel_ring_idle(ring); |
2871 | dev_priv->mm.interruptible = was_interruptible; |
||
3031 | serge | 2872 | if (ret) { |
3746 | Serge | 2873 | DRM_ERROR("failed to enable ironlake power savings\n"); |
3031 | serge | 2874 | ironlake_teardown_rc6(dev); |
2875 | return; |
||
2876 | } |
||
2877 | |||
3243 | Serge | 2878 | I915_WRITE(PWRCTXA, dev_priv->ips.pwrctx->gtt_offset | PWRCTX_EN); |
3031 | serge | 2879 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT); |
2880 | } |
||
2881 | |||
2882 | static unsigned long intel_pxfreq(u32 vidfreq) |
||
2883 | { |
||
2884 | unsigned long freq; |
||
2885 | int div = (vidfreq & 0x3f0000) >> 16; |
||
2886 | int post = (vidfreq & 0x3000) >> 12; |
||
2887 | int pre = (vidfreq & 0x7); |
||
2888 | |||
2889 | if (!pre) |
||
2890 | return 0; |
||
2891 | |||
2892 | freq = ((div * 133333) / ((1< |
||
2893 | |||
2894 | return freq; |
||
2895 | } |
||
2896 | |||
2897 | static const struct cparams { |
||
2898 | u16 i; |
||
2899 | u16 t; |
||
2900 | u16 m; |
||
2901 | u16 c; |
||
2902 | } cparams[] = { |
||
2903 | { 1, 1333, 301, 28664 }, |
||
2904 | { 1, 1066, 294, 24460 }, |
||
2905 | { 1, 800, 294, 25192 }, |
||
2906 | { 0, 1333, 276, 27605 }, |
||
2907 | { 0, 1066, 276, 27605 }, |
||
2908 | { 0, 800, 231, 23784 }, |
||
2909 | }; |
||
2910 | |||
2911 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
||
2912 | { |
||
2913 | u64 total_count, diff, ret; |
||
2914 | u32 count1, count2, count3, m = 0, c = 0; |
||
2915 | unsigned long now = jiffies_to_msecs(GetTimerTicks()), diff1; |
||
2916 | int i; |
||
2917 | |||
2918 | assert_spin_locked(&mchdev_lock); |
||
2919 | |||
2920 | diff1 = now - dev_priv->ips.last_time1; |
||
2921 | |||
2922 | /* Prevent division-by-zero if we are asking too fast. |
||
2923 | * Also, we don't get interesting results if we are polling |
||
2924 | * faster than once in 10ms, so just return the saved value |
||
2925 | * in such cases. |
||
2926 | */ |
||
2927 | if (diff1 <= 10) |
||
2928 | return dev_priv->ips.chipset_power; |
||
2929 | |||
2930 | count1 = I915_READ(DMIEC); |
||
2931 | count2 = I915_READ(DDREC); |
||
2932 | count3 = I915_READ(CSIEC); |
||
2933 | |||
2934 | total_count = count1 + count2 + count3; |
||
2935 | |||
2936 | /* FIXME: handle per-counter overflow */ |
||
2937 | if (total_count < dev_priv->ips.last_count1) { |
||
2938 | diff = ~0UL - dev_priv->ips.last_count1; |
||
2939 | diff += total_count; |
||
2940 | } else { |
||
2941 | diff = total_count - dev_priv->ips.last_count1; |
||
2942 | } |
||
2943 | |||
2944 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { |
||
2945 | if (cparams[i].i == dev_priv->ips.c_m && |
||
2946 | cparams[i].t == dev_priv->ips.r_t) { |
||
2947 | m = cparams[i].m; |
||
2948 | c = cparams[i].c; |
||
2949 | break; |
||
2950 | } |
||
2951 | } |
||
2952 | |||
2953 | diff = div_u64(diff, diff1); |
||
2954 | ret = ((m * diff) + c); |
||
2955 | ret = div_u64(ret, 10); |
||
2956 | |||
2957 | dev_priv->ips.last_count1 = total_count; |
||
2958 | dev_priv->ips.last_time1 = now; |
||
2959 | |||
2960 | dev_priv->ips.chipset_power = ret; |
||
2961 | |||
2962 | return ret; |
||
2963 | } |
||
2964 | |||
2965 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
||
2966 | { |
||
2967 | unsigned long val; |
||
2968 | |||
2969 | if (dev_priv->info->gen != 5) |
||
2970 | return 0; |
||
2971 | |||
2972 | spin_lock_irq(&mchdev_lock); |
||
2973 | |||
2974 | val = __i915_chipset_val(dev_priv); |
||
2975 | |||
2976 | spin_unlock_irq(&mchdev_lock); |
||
2977 | |||
2978 | return val; |
||
2979 | } |
||
2980 | |||
2981 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
||
2982 | { |
||
2983 | unsigned long m, x, b; |
||
2984 | u32 tsfs; |
||
2985 | |||
2986 | tsfs = I915_READ(TSFS); |
||
2987 | |||
2988 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); |
||
2989 | x = I915_READ8(TR1); |
||
2990 | |||
2991 | b = tsfs & TSFS_INTR_MASK; |
||
2992 | |||
2993 | return ((m * x) / 127) - b; |
||
2994 | } |
||
2995 | |||
2996 | static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) |
||
2997 | { |
||
2998 | static const struct v_table { |
||
2999 | u16 vd; /* in .1 mil */ |
||
3000 | u16 vm; /* in .1 mil */ |
||
3001 | } v_table[] = { |
||
3002 | { 0, 0, }, |
||
3003 | { 375, 0, }, |
||
3004 | { 500, 0, }, |
||
3005 | { 625, 0, }, |
||
3006 | { 750, 0, }, |
||
3007 | { 875, 0, }, |
||
3008 | { 1000, 0, }, |
||
3009 | { 1125, 0, }, |
||
3010 | { 4125, 3000, }, |
||
3011 | { 4125, 3000, }, |
||
3012 | { 4125, 3000, }, |
||
3013 | { 4125, 3000, }, |
||
3014 | { 4125, 3000, }, |
||
3015 | { 4125, 3000, }, |
||
3016 | { 4125, 3000, }, |
||
3017 | { 4125, 3000, }, |
||
3018 | { 4125, 3000, }, |
||
3019 | { 4125, 3000, }, |
||
3020 | { 4125, 3000, }, |
||
3021 | { 4125, 3000, }, |
||
3022 | { 4125, 3000, }, |
||
3023 | { 4125, 3000, }, |
||
3024 | { 4125, 3000, }, |
||
3025 | { 4125, 3000, }, |
||
3026 | { 4125, 3000, }, |
||
3027 | { 4125, 3000, }, |
||
3028 | { 4125, 3000, }, |
||
3029 | { 4125, 3000, }, |
||
3030 | { 4125, 3000, }, |
||
3031 | { 4125, 3000, }, |
||
3032 | { 4125, 3000, }, |
||
3033 | { 4125, 3000, }, |
||
3034 | { 4250, 3125, }, |
||
3035 | { 4375, 3250, }, |
||
3036 | { 4500, 3375, }, |
||
3037 | { 4625, 3500, }, |
||
3038 | { 4750, 3625, }, |
||
3039 | { 4875, 3750, }, |
||
3040 | { 5000, 3875, }, |
||
3041 | { 5125, 4000, }, |
||
3042 | { 5250, 4125, }, |
||
3043 | { 5375, 4250, }, |
||
3044 | { 5500, 4375, }, |
||
3045 | { 5625, 4500, }, |
||
3046 | { 5750, 4625, }, |
||
3047 | { 5875, 4750, }, |
||
3048 | { 6000, 4875, }, |
||
3049 | { 6125, 5000, }, |
||
3050 | { 6250, 5125, }, |
||
3051 | { 6375, 5250, }, |
||
3052 | { 6500, 5375, }, |
||
3053 | { 6625, 5500, }, |
||
3054 | { 6750, 5625, }, |
||
3055 | { 6875, 5750, }, |
||
3056 | { 7000, 5875, }, |
||
3057 | { 7125, 6000, }, |
||
3058 | { 7250, 6125, }, |
||
3059 | { 7375, 6250, }, |
||
3060 | { 7500, 6375, }, |
||
3061 | { 7625, 6500, }, |
||
3062 | { 7750, 6625, }, |
||
3063 | { 7875, 6750, }, |
||
3064 | { 8000, 6875, }, |
||
3065 | { 8125, 7000, }, |
||
3066 | { 8250, 7125, }, |
||
3067 | { 8375, 7250, }, |
||
3068 | { 8500, 7375, }, |
||
3069 | { 8625, 7500, }, |
||
3070 | { 8750, 7625, }, |
||
3071 | { 8875, 7750, }, |
||
3072 | { 9000, 7875, }, |
||
3073 | { 9125, 8000, }, |
||
3074 | { 9250, 8125, }, |
||
3075 | { 9375, 8250, }, |
||
3076 | { 9500, 8375, }, |
||
3077 | { 9625, 8500, }, |
||
3078 | { 9750, 8625, }, |
||
3079 | { 9875, 8750, }, |
||
3080 | { 10000, 8875, }, |
||
3081 | { 10125, 9000, }, |
||
3082 | { 10250, 9125, }, |
||
3083 | { 10375, 9250, }, |
||
3084 | { 10500, 9375, }, |
||
3085 | { 10625, 9500, }, |
||
3086 | { 10750, 9625, }, |
||
3087 | { 10875, 9750, }, |
||
3088 | { 11000, 9875, }, |
||
3089 | { 11125, 10000, }, |
||
3090 | { 11250, 10125, }, |
||
3091 | { 11375, 10250, }, |
||
3092 | { 11500, 10375, }, |
||
3093 | { 11625, 10500, }, |
||
3094 | { 11750, 10625, }, |
||
3095 | { 11875, 10750, }, |
||
3096 | { 12000, 10875, }, |
||
3097 | { 12125, 11000, }, |
||
3098 | { 12250, 11125, }, |
||
3099 | { 12375, 11250, }, |
||
3100 | { 12500, 11375, }, |
||
3101 | { 12625, 11500, }, |
||
3102 | { 12750, 11625, }, |
||
3103 | { 12875, 11750, }, |
||
3104 | { 13000, 11875, }, |
||
3105 | { 13125, 12000, }, |
||
3106 | { 13250, 12125, }, |
||
3107 | { 13375, 12250, }, |
||
3108 | { 13500, 12375, }, |
||
3109 | { 13625, 12500, }, |
||
3110 | { 13750, 12625, }, |
||
3111 | { 13875, 12750, }, |
||
3112 | { 14000, 12875, }, |
||
3113 | { 14125, 13000, }, |
||
3114 | { 14250, 13125, }, |
||
3115 | { 14375, 13250, }, |
||
3116 | { 14500, 13375, }, |
||
3117 | { 14625, 13500, }, |
||
3118 | { 14750, 13625, }, |
||
3119 | { 14875, 13750, }, |
||
3120 | { 15000, 13875, }, |
||
3121 | { 15125, 14000, }, |
||
3122 | { 15250, 14125, }, |
||
3123 | { 15375, 14250, }, |
||
3124 | { 15500, 14375, }, |
||
3125 | { 15625, 14500, }, |
||
3126 | { 15750, 14625, }, |
||
3127 | { 15875, 14750, }, |
||
3128 | { 16000, 14875, }, |
||
3129 | { 16125, 15000, }, |
||
3130 | }; |
||
3131 | if (dev_priv->info->is_mobile) |
||
3132 | return v_table[pxvid].vm; |
||
3133 | else |
||
3134 | return v_table[pxvid].vd; |
||
3135 | } |
||
3136 | |||
3137 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
||
3138 | { |
||
3139 | struct timespec now, diff1; |
||
3140 | u64 diff; |
||
3141 | unsigned long diffms; |
||
3142 | u32 count; |
||
3143 | |||
3144 | assert_spin_locked(&mchdev_lock); |
||
3145 | |||
3146 | getrawmonotonic(&now); |
||
3147 | diff1 = timespec_sub(now, dev_priv->ips.last_time2); |
||
3148 | |||
3149 | /* Don't divide by 0 */ |
||
3150 | diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000; |
||
3151 | if (!diffms) |
||
3152 | return; |
||
3153 | |||
3154 | count = I915_READ(GFXEC); |
||
3155 | |||
3156 | if (count < dev_priv->ips.last_count2) { |
||
3157 | diff = ~0UL - dev_priv->ips.last_count2; |
||
3158 | diff += count; |
||
3159 | } else { |
||
3160 | diff = count - dev_priv->ips.last_count2; |
||
3161 | } |
||
3162 | |||
3163 | dev_priv->ips.last_count2 = count; |
||
3164 | dev_priv->ips.last_time2 = now; |
||
3165 | |||
3166 | /* More magic constants... */ |
||
3167 | diff = diff * 1181; |
||
3168 | diff = div_u64(diff, diffms * 10); |
||
3169 | dev_priv->ips.gfx_power = diff; |
||
3170 | } |
||
3171 | |||
3172 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
||
3173 | { |
||
3174 | if (dev_priv->info->gen != 5) |
||
3175 | return; |
||
3176 | |||
3177 | spin_lock_irq(&mchdev_lock); |
||
3178 | |||
3179 | __i915_update_gfx_val(dev_priv); |
||
3180 | |||
3181 | spin_unlock_irq(&mchdev_lock); |
||
3182 | } |
||
3183 | |||
3184 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
||
3185 | { |
||
3186 | unsigned long t, corr, state1, corr2, state2; |
||
3187 | u32 pxvid, ext_v; |
||
3188 | |||
3189 | assert_spin_locked(&mchdev_lock); |
||
3190 | |||
3191 | pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4)); |
||
3192 | pxvid = (pxvid >> 24) & 0x7f; |
||
3193 | ext_v = pvid_to_extvid(dev_priv, pxvid); |
||
3194 | |||
3195 | state1 = ext_v; |
||
3196 | |||
3197 | t = i915_mch_val(dev_priv); |
||
3198 | |||
3199 | /* Revel in the empirically derived constants */ |
||
3200 | |||
3201 | /* Correction factor in 1/100000 units */ |
||
3202 | if (t > 80) |
||
3203 | corr = ((t * 2349) + 135940); |
||
3204 | else if (t >= 50) |
||
3205 | corr = ((t * 964) + 29317); |
||
3206 | else /* < 50 */ |
||
3207 | corr = ((t * 301) + 1004); |
||
3208 | |||
3209 | corr = corr * ((150142 * state1) / 10000 - 78642); |
||
3210 | corr /= 100000; |
||
3211 | corr2 = (corr * dev_priv->ips.corr); |
||
3212 | |||
3213 | state2 = (corr2 * state1) / 10000; |
||
3214 | state2 /= 100; /* convert to mW */ |
||
3215 | |||
3216 | __i915_update_gfx_val(dev_priv); |
||
3217 | |||
3218 | return dev_priv->ips.gfx_power + state2; |
||
3219 | } |
||
3220 | |||
3221 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
||
3222 | { |
||
3223 | unsigned long val; |
||
3224 | |||
3225 | if (dev_priv->info->gen != 5) |
||
3226 | return 0; |
||
3227 | |||
3228 | spin_lock_irq(&mchdev_lock); |
||
3229 | |||
3230 | val = __i915_gfx_val(dev_priv); |
||
3231 | |||
3232 | spin_unlock_irq(&mchdev_lock); |
||
3233 | |||
3234 | return val; |
||
3235 | } |
||
3236 | |||
3237 | /** |
||
3238 | * i915_read_mch_val - return value for IPS use |
||
3239 | * |
||
3240 | * Calculate and return a value for the IPS driver to use when deciding whether |
||
3241 | * we have thermal and power headroom to increase CPU or GPU power budget. |
||
3242 | */ |
||
3243 | unsigned long i915_read_mch_val(void) |
||
3244 | { |
||
3245 | struct drm_i915_private *dev_priv; |
||
3246 | unsigned long chipset_val, graphics_val, ret = 0; |
||
3247 | |||
3248 | spin_lock_irq(&mchdev_lock); |
||
3249 | if (!i915_mch_dev) |
||
3250 | goto out_unlock; |
||
3251 | dev_priv = i915_mch_dev; |
||
3252 | |||
3253 | chipset_val = __i915_chipset_val(dev_priv); |
||
3254 | graphics_val = __i915_gfx_val(dev_priv); |
||
3255 | |||
3256 | ret = chipset_val + graphics_val; |
||
3257 | |||
3258 | out_unlock: |
||
3259 | spin_unlock_irq(&mchdev_lock); |
||
3260 | |||
3261 | return ret; |
||
3262 | } |
||
3263 | EXPORT_SYMBOL_GPL(i915_read_mch_val); |
||
3264 | |||
3265 | /** |
||
3266 | * i915_gpu_raise - raise GPU frequency limit |
||
3267 | * |
||
3268 | * Raise the limit; IPS indicates we have thermal headroom. |
||
3269 | */ |
||
3270 | bool i915_gpu_raise(void) |
||
3271 | { |
||
3272 | struct drm_i915_private *dev_priv; |
||
3273 | bool ret = true; |
||
3274 | |||
3275 | spin_lock_irq(&mchdev_lock); |
||
3276 | if (!i915_mch_dev) { |
||
3277 | ret = false; |
||
3278 | goto out_unlock; |
||
3279 | } |
||
3280 | dev_priv = i915_mch_dev; |
||
3281 | |||
3282 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
||
3283 | dev_priv->ips.max_delay--; |
||
3284 | |||
3285 | out_unlock: |
||
3286 | spin_unlock_irq(&mchdev_lock); |
||
3287 | |||
3288 | return ret; |
||
3289 | } |
||
3290 | EXPORT_SYMBOL_GPL(i915_gpu_raise); |
||
3291 | |||
3292 | /** |
||
3293 | * i915_gpu_lower - lower GPU frequency limit |
||
3294 | * |
||
3295 | * IPS indicates we're close to a thermal limit, so throttle back the GPU |
||
3296 | * frequency maximum. |
||
3297 | */ |
||
3298 | bool i915_gpu_lower(void) |
||
3299 | { |
||
3300 | struct drm_i915_private *dev_priv; |
||
3301 | bool ret = true; |
||
3302 | |||
3303 | spin_lock_irq(&mchdev_lock); |
||
3304 | if (!i915_mch_dev) { |
||
3305 | ret = false; |
||
3306 | goto out_unlock; |
||
3307 | } |
||
3308 | dev_priv = i915_mch_dev; |
||
3309 | |||
3310 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
||
3311 | dev_priv->ips.max_delay++; |
||
3312 | |||
3313 | out_unlock: |
||
3314 | spin_unlock_irq(&mchdev_lock); |
||
3315 | |||
3316 | return ret; |
||
3317 | } |
||
3318 | EXPORT_SYMBOL_GPL(i915_gpu_lower); |
||
3319 | |||
3320 | /** |
||
3321 | * i915_gpu_busy - indicate GPU business to IPS |
||
3322 | * |
||
3323 | * Tell the IPS driver whether or not the GPU is busy. |
||
3324 | */ |
||
3325 | bool i915_gpu_busy(void) |
||
3326 | { |
||
3327 | struct drm_i915_private *dev_priv; |
||
3328 | struct intel_ring_buffer *ring; |
||
3329 | bool ret = false; |
||
3330 | int i; |
||
3331 | |||
3332 | spin_lock_irq(&mchdev_lock); |
||
3333 | if (!i915_mch_dev) |
||
3334 | goto out_unlock; |
||
3335 | dev_priv = i915_mch_dev; |
||
3336 | |||
3337 | for_each_ring(ring, dev_priv, i) |
||
3338 | ret |= !list_empty(&ring->request_list); |
||
3339 | |||
3340 | out_unlock: |
||
3341 | spin_unlock_irq(&mchdev_lock); |
||
3342 | |||
3343 | return ret; |
||
3344 | } |
||
3345 | EXPORT_SYMBOL_GPL(i915_gpu_busy); |
||
3346 | |||
3347 | /** |
||
3348 | * i915_gpu_turbo_disable - disable graphics turbo |
||
3349 | * |
||
3350 | * Disable graphics turbo by resetting the max frequency and setting the |
||
3351 | * current frequency to the default. |
||
3352 | */ |
||
3353 | bool i915_gpu_turbo_disable(void) |
||
3354 | { |
||
3355 | struct drm_i915_private *dev_priv; |
||
3356 | bool ret = true; |
||
3357 | |||
3358 | spin_lock_irq(&mchdev_lock); |
||
3359 | if (!i915_mch_dev) { |
||
3360 | ret = false; |
||
3361 | goto out_unlock; |
||
3362 | } |
||
3363 | dev_priv = i915_mch_dev; |
||
3364 | |||
3365 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
||
3366 | |||
3367 | if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart)) |
||
3368 | ret = false; |
||
3369 | |||
3370 | out_unlock: |
||
3371 | spin_unlock_irq(&mchdev_lock); |
||
3372 | |||
3373 | return ret; |
||
3374 | } |
||
3375 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); |
||
3376 | |||
3377 | /** |
||
3378 | * Tells the intel_ips driver that the i915 driver is now loaded, if |
||
3379 | * IPS got loaded first. |
||
3380 | * |
||
3381 | * This awkward dance is so that neither module has to depend on the |
||
3382 | * other in order for IPS to do the appropriate communication of |
||
3383 | * GPU turbo limits to i915. |
||
3384 | */ |
||
3385 | static void |
||
3386 | ips_ping_for_i915_load(void) |
||
3387 | { |
||
3388 | void (*link)(void); |
||
3389 | |||
3390 | // link = symbol_get(ips_link_to_i915_driver); |
||
3391 | // if (link) { |
||
3392 | // link(); |
||
3393 | // symbol_put(ips_link_to_i915_driver); |
||
3394 | // } |
||
3395 | } |
||
3396 | |||
3397 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) |
||
3398 | { |
||
3399 | /* We only register the i915 ips part with intel-ips once everything is |
||
3400 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ |
||
3401 | spin_lock_irq(&mchdev_lock); |
||
3402 | i915_mch_dev = dev_priv; |
||
3403 | spin_unlock_irq(&mchdev_lock); |
||
3404 | |||
3405 | ips_ping_for_i915_load(); |
||
3406 | } |
||
3407 | |||
3408 | void intel_gpu_ips_teardown(void) |
||
3409 | { |
||
3410 | spin_lock_irq(&mchdev_lock); |
||
3411 | i915_mch_dev = NULL; |
||
3412 | spin_unlock_irq(&mchdev_lock); |
||
3413 | } |
||
3414 | static void intel_init_emon(struct drm_device *dev) |
||
3415 | { |
||
3416 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3417 | u32 lcfuse; |
||
3418 | u8 pxw[16]; |
||
3419 | int i; |
||
3420 | |||
3421 | /* Disable to program */ |
||
3422 | I915_WRITE(ECR, 0); |
||
3423 | POSTING_READ(ECR); |
||
3424 | |||
3425 | /* Program energy weights for various events */ |
||
3426 | I915_WRITE(SDEW, 0x15040d00); |
||
3427 | I915_WRITE(CSIEW0, 0x007f0000); |
||
3428 | I915_WRITE(CSIEW1, 0x1e220004); |
||
3429 | I915_WRITE(CSIEW2, 0x04000004); |
||
3430 | |||
3431 | for (i = 0; i < 5; i++) |
||
3432 | I915_WRITE(PEW + (i * 4), 0); |
||
3433 | for (i = 0; i < 3; i++) |
||
3434 | I915_WRITE(DEW + (i * 4), 0); |
||
3435 | |||
3436 | /* Program P-state weights to account for frequency power adjustment */ |
||
3437 | for (i = 0; i < 16; i++) { |
||
3438 | u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4)); |
||
3439 | unsigned long freq = intel_pxfreq(pxvidfreq); |
||
3440 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> |
||
3441 | PXVFREQ_PX_SHIFT; |
||
3442 | unsigned long val; |
||
3443 | |||
3444 | val = vid * vid; |
||
3445 | val *= (freq / 1000); |
||
3446 | val *= 255; |
||
3447 | val /= (127*127*900); |
||
3448 | if (val > 0xff) |
||
3449 | DRM_ERROR("bad pxval: %ld\n", val); |
||
3450 | pxw[i] = val; |
||
3451 | } |
||
3452 | /* Render standby states get 0 weight */ |
||
3453 | pxw[14] = 0; |
||
3454 | pxw[15] = 0; |
||
3455 | |||
3456 | for (i = 0; i < 4; i++) { |
||
3457 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | |
||
3458 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); |
||
3459 | I915_WRITE(PXW + (i * 4), val); |
||
3460 | } |
||
3461 | |||
3462 | /* Adjust magic regs to magic values (more experimental results) */ |
||
3463 | I915_WRITE(OGW0, 0); |
||
3464 | I915_WRITE(OGW1, 0); |
||
3465 | I915_WRITE(EG0, 0x00007f00); |
||
3466 | I915_WRITE(EG1, 0x0000000e); |
||
3467 | I915_WRITE(EG2, 0x000e0000); |
||
3468 | I915_WRITE(EG3, 0x68000300); |
||
3469 | I915_WRITE(EG4, 0x42000000); |
||
3470 | I915_WRITE(EG5, 0x00140031); |
||
3471 | I915_WRITE(EG6, 0); |
||
3472 | I915_WRITE(EG7, 0); |
||
3473 | |||
3474 | for (i = 0; i < 8; i++) |
||
3475 | I915_WRITE(PXWL + (i * 4), 0); |
||
3476 | |||
3477 | /* Enable PMON + select events */ |
||
3478 | I915_WRITE(ECR, 0x80000019); |
||
3479 | |||
3480 | lcfuse = I915_READ(LCFUSE02); |
||
3481 | |||
3482 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
||
3483 | } |
||
3484 | |||
3485 | void intel_disable_gt_powersave(struct drm_device *dev) |
||
3486 | { |
||
3243 | Serge | 3487 | struct drm_i915_private *dev_priv = dev->dev_private; |
3488 | |||
3031 | serge | 3489 | if (IS_IRONLAKE_M(dev)) { |
3490 | ironlake_disable_drps(dev); |
||
3491 | ironlake_disable_rc6(dev); |
||
3492 | } else if (INTEL_INFO(dev)->gen >= 6 && !IS_VALLEYVIEW(dev)) { |
||
3482 | Serge | 3493 | // cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work); |
3494 | mutex_lock(&dev_priv->rps.hw_lock); |
||
3031 | serge | 3495 | gen6_disable_rps(dev); |
3480 | Serge | 3496 | mutex_unlock(&dev_priv->rps.hw_lock); |
3031 | serge | 3497 | } |
3498 | } |
||
3499 | |||
3482 | Serge | 3500 | static void intel_gen6_powersave_work(struct work_struct *work) |
3501 | { |
||
3502 | struct drm_i915_private *dev_priv = |
||
3503 | container_of(work, struct drm_i915_private, |
||
3504 | rps.delayed_resume_work.work); |
||
3505 | struct drm_device *dev = dev_priv->dev; |
||
3506 | |||
3507 | ENTER(); |
||
3508 | |||
3509 | mutex_lock(&dev_priv->rps.hw_lock); |
||
3510 | gen6_enable_rps(dev); |
||
3511 | gen6_update_ring_freq(dev); |
||
3512 | mutex_unlock(&dev_priv->rps.hw_lock); |
||
3513 | |||
3514 | LEAVE(); |
||
3515 | } |
||
3516 | |||
3031 | serge | 3517 | void intel_enable_gt_powersave(struct drm_device *dev) |
3518 | { |
||
3243 | Serge | 3519 | struct drm_i915_private *dev_priv = dev->dev_private; |
3520 | |||
3031 | serge | 3521 | if (IS_IRONLAKE_M(dev)) { |
3522 | ironlake_enable_drps(dev); |
||
3523 | ironlake_enable_rc6(dev); |
||
3524 | intel_init_emon(dev); |
||
3525 | } else if ((IS_GEN6(dev) || IS_GEN7(dev)) && !IS_VALLEYVIEW(dev)) { |
||
3243 | Serge | 3526 | /* |
3527 | * PCU communication is slow and this doesn't need to be |
||
3528 | * done at any specific time, so do this out of our fast path |
||
3529 | * to make resume and init faster. |
||
3530 | */ |
||
3482 | Serge | 3531 | schedule_delayed_work(&dev_priv->rps.delayed_resume_work, |
3532 | round_jiffies_up_relative(HZ)); |
||
3031 | serge | 3533 | } |
3534 | } |
||
3535 | |||
3243 | Serge | 3536 | static void ibx_init_clock_gating(struct drm_device *dev) |
3537 | { |
||
3538 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3539 | |||
3540 | /* |
||
3541 | * On Ibex Peak and Cougar Point, we need to disable clock |
||
3542 | * gating for the panel power sequencer or it will fail to |
||
3543 | * start up when no ports are active. |
||
3544 | */ |
||
3545 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); |
||
3546 | } |
||
3547 | |||
3031 | serge | 3548 | static void ironlake_init_clock_gating(struct drm_device *dev) |
3549 | { |
||
3550 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3243 | Serge | 3551 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
3031 | serge | 3552 | |
3553 | /* Required for FBC */ |
||
3243 | Serge | 3554 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
3555 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | |
||
3556 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; |
||
3031 | serge | 3557 | |
3558 | I915_WRITE(PCH_3DCGDIS0, |
||
3559 | MARIUNIT_CLOCK_GATE_DISABLE | |
||
3560 | SVSMUNIT_CLOCK_GATE_DISABLE); |
||
3561 | I915_WRITE(PCH_3DCGDIS1, |
||
3562 | VFMUNIT_CLOCK_GATE_DISABLE); |
||
3563 | |||
3564 | /* |
||
3565 | * According to the spec the following bits should be set in |
||
3566 | * order to enable memory self-refresh |
||
3567 | * The bit 22/21 of 0x42004 |
||
3568 | * The bit 5 of 0x42020 |
||
3569 | * The bit 15 of 0x45000 |
||
3570 | */ |
||
3571 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
||
3572 | (I915_READ(ILK_DISPLAY_CHICKEN2) | |
||
3573 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); |
||
3243 | Serge | 3574 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
3031 | serge | 3575 | I915_WRITE(DISP_ARB_CTL, |
3576 | (I915_READ(DISP_ARB_CTL) | |
||
3577 | DISP_FBC_WM_DIS)); |
||
3578 | I915_WRITE(WM3_LP_ILK, 0); |
||
3579 | I915_WRITE(WM2_LP_ILK, 0); |
||
3580 | I915_WRITE(WM1_LP_ILK, 0); |
||
3581 | |||
3582 | /* |
||
3583 | * Based on the document from hardware guys the following bits |
||
3584 | * should be set unconditionally in order to enable FBC. |
||
3585 | * The bit 22 of 0x42000 |
||
3586 | * The bit 22 of 0x42004 |
||
3587 | * The bit 7,8,9 of 0x42020. |
||
3588 | */ |
||
3589 | if (IS_IRONLAKE_M(dev)) { |
||
3590 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
||
3591 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
||
3592 | ILK_FBCQ_DIS); |
||
3593 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
||
3594 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
||
3595 | ILK_DPARB_GATE); |
||
3596 | } |
||
3597 | |||
3243 | Serge | 3598 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
3599 | |||
3031 | serge | 3600 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
3601 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
||
3602 | ILK_ELPIN_409_SELECT); |
||
3603 | I915_WRITE(_3D_CHICKEN2, |
||
3604 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | |
||
3605 | _3D_CHICKEN2_WM_READ_PIPELINED); |
||
3243 | Serge | 3606 | |
3607 | /* WaDisableRenderCachePipelinedFlush */ |
||
3608 | I915_WRITE(CACHE_MODE_0, |
||
3609 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); |
||
3610 | |||
3611 | ibx_init_clock_gating(dev); |
||
3031 | serge | 3612 | } |
3613 | |||
3243 | Serge | 3614 | static void cpt_init_clock_gating(struct drm_device *dev) |
3615 | { |
||
3616 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3617 | int pipe; |
||
3746 | Serge | 3618 | uint32_t val; |
3243 | Serge | 3619 | |
3620 | /* |
||
3621 | * On Ibex Peak and Cougar Point, we need to disable clock |
||
3622 | * gating for the panel power sequencer or it will fail to |
||
3623 | * start up when no ports are active. |
||
3624 | */ |
||
3625 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); |
||
3626 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | |
||
3627 | DPLS_EDP_PPS_FIX_DIS); |
||
3628 | /* The below fixes the weird display corruption, a few pixels shifted |
||
3629 | * downward, on (only) LVDS of some HP laptops with IVY. |
||
3630 | */ |
||
3746 | Serge | 3631 | for_each_pipe(pipe) { |
3632 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
||
3633 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; |
||
3634 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
||
3635 | if (dev_priv->fdi_rx_polarity_inverted) |
||
3636 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
||
3637 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
||
3638 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; |
||
3639 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; |
||
3640 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
||
3641 | } |
||
3243 | Serge | 3642 | /* WADP0ClockGatingDisable */ |
3643 | for_each_pipe(pipe) { |
||
3644 | I915_WRITE(TRANS_CHICKEN1(pipe), |
||
3645 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
||
3646 | } |
||
3647 | } |
||
3648 | |||
3480 | Serge | 3649 | static void gen6_check_mch_setup(struct drm_device *dev) |
3650 | { |
||
3651 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3652 | uint32_t tmp; |
||
3653 | |||
3654 | tmp = I915_READ(MCH_SSKPD); |
||
3655 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) { |
||
3656 | DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp); |
||
3657 | DRM_INFO("This can cause pipe underruns and display issues.\n"); |
||
3658 | DRM_INFO("Please upgrade your BIOS to fix this.\n"); |
||
3659 | } |
||
3660 | } |
||
3661 | |||
3031 | serge | 3662 | static void gen6_init_clock_gating(struct drm_device *dev) |
3663 | { |
||
3664 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3665 | int pipe; |
||
3243 | Serge | 3666 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
3031 | serge | 3667 | |
3243 | Serge | 3668 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
3031 | serge | 3669 | |
3670 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
||
3671 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
||
3672 | ILK_ELPIN_409_SELECT); |
||
3673 | |||
3243 | Serge | 3674 | /* WaDisableHiZPlanesWhenMSAAEnabled */ |
3675 | I915_WRITE(_3D_CHICKEN, |
||
3676 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); |
||
3677 | |||
3678 | /* WaSetupGtModeTdRowDispatch */ |
||
3679 | if (IS_SNB_GT1(dev)) |
||
3680 | I915_WRITE(GEN6_GT_MODE, |
||
3681 | _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE)); |
||
3682 | |||
3031 | serge | 3683 | I915_WRITE(WM3_LP_ILK, 0); |
3684 | I915_WRITE(WM2_LP_ILK, 0); |
||
3685 | I915_WRITE(WM1_LP_ILK, 0); |
||
3686 | |||
3687 | I915_WRITE(CACHE_MODE_0, |
||
3688 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
||
3689 | |||
3690 | I915_WRITE(GEN6_UCGCTL1, |
||
3691 | I915_READ(GEN6_UCGCTL1) | |
||
3692 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | |
||
3693 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); |
||
3694 | |||
3695 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock |
||
3696 | * gating disable must be set. Failure to set it results in |
||
3697 | * flickering pixels due to Z write ordering failures after |
||
3698 | * some amount of runtime in the Mesa "fire" demo, and Unigine |
||
3699 | * Sanctuary and Tropics, and apparently anything else with |
||
3700 | * alpha test or pixel discard. |
||
3701 | * |
||
3702 | * According to the spec, bit 11 (RCCUNIT) must also be set, |
||
3703 | * but we didn't debug actual testcases to find it out. |
||
3704 | * |
||
3705 | * Also apply WaDisableVDSUnitClockGating and |
||
3706 | * WaDisableRCPBUnitClockGating. |
||
3707 | */ |
||
3708 | I915_WRITE(GEN6_UCGCTL2, |
||
3709 | GEN7_VDSUNIT_CLOCK_GATE_DISABLE | |
||
3710 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | |
||
3711 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); |
||
3712 | |||
3713 | /* Bspec says we need to always set all mask bits. */ |
||
3714 | I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) | |
||
3715 | _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL); |
||
3716 | |||
3717 | /* |
||
3718 | * According to the spec the following bits should be |
||
3719 | * set in order to enable memory self-refresh and fbc: |
||
3720 | * The bit21 and bit22 of 0x42000 |
||
3721 | * The bit21 and bit22 of 0x42004 |
||
3722 | * The bit5 and bit7 of 0x42020 |
||
3723 | * The bit14 of 0x70180 |
||
3724 | * The bit14 of 0x71180 |
||
3725 | */ |
||
3726 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
||
3727 | I915_READ(ILK_DISPLAY_CHICKEN1) | |
||
3728 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); |
||
3729 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
||
3730 | I915_READ(ILK_DISPLAY_CHICKEN2) | |
||
3731 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); |
||
3243 | Serge | 3732 | I915_WRITE(ILK_DSPCLK_GATE_D, |
3733 | I915_READ(ILK_DSPCLK_GATE_D) | |
||
3734 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | |
||
3735 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); |
||
3031 | serge | 3736 | |
3243 | Serge | 3737 | /* WaMbcDriverBootEnable */ |
3031 | serge | 3738 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
3739 | GEN6_MBCTL_ENABLE_BOOT_FETCH); |
||
3740 | |||
3741 | for_each_pipe(pipe) { |
||
3742 | I915_WRITE(DSPCNTR(pipe), |
||
3743 | I915_READ(DSPCNTR(pipe)) | |
||
3744 | DISPPLANE_TRICKLE_FEED_DISABLE); |
||
3745 | intel_flush_display_plane(dev_priv, pipe); |
||
3746 | } |
||
3747 | |||
3748 | /* The default value should be 0x200 according to docs, but the two |
||
3749 | * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */ |
||
3750 | I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff)); |
||
3751 | I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI)); |
||
3243 | Serge | 3752 | |
3753 | cpt_init_clock_gating(dev); |
||
3480 | Serge | 3754 | |
3755 | gen6_check_mch_setup(dev); |
||
3031 | serge | 3756 | } |
3757 | |||
3758 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) |
||
3759 | { |
||
3760 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); |
||
3761 | |||
3762 | reg &= ~GEN7_FF_SCHED_MASK; |
||
3763 | reg |= GEN7_FF_TS_SCHED_HW; |
||
3764 | reg |= GEN7_FF_VS_SCHED_HW; |
||
3765 | reg |= GEN7_FF_DS_SCHED_HW; |
||
3766 | |||
3480 | Serge | 3767 | /* WaVSRefCountFullforceMissDisable */ |
3768 | if (IS_HASWELL(dev_priv->dev)) |
||
3769 | reg &= ~GEN7_FF_VS_REF_CNT_FFME; |
||
3770 | |||
3031 | serge | 3771 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); |
3772 | } |
||
3773 | |||
3243 | Serge | 3774 | static void lpt_init_clock_gating(struct drm_device *dev) |
3775 | { |
||
3776 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3777 | |||
3778 | /* |
||
3779 | * TODO: this bit should only be enabled when really needed, then |
||
3780 | * disabled when not needed anymore in order to save power. |
||
3781 | */ |
||
3782 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) |
||
3783 | I915_WRITE(SOUTH_DSPCLK_GATE_D, |
||
3784 | I915_READ(SOUTH_DSPCLK_GATE_D) | |
||
3785 | PCH_LP_PARTITION_LEVEL_DISABLE); |
||
3786 | } |
||
3787 | |||
3031 | serge | 3788 | static void haswell_init_clock_gating(struct drm_device *dev) |
3789 | { |
||
3790 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3791 | int pipe; |
||
3792 | |||
3793 | I915_WRITE(WM3_LP_ILK, 0); |
||
3794 | I915_WRITE(WM2_LP_ILK, 0); |
||
3795 | I915_WRITE(WM1_LP_ILK, 0); |
||
3796 | |||
3797 | /* According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
||
3798 | * This implements the WaDisableRCZUnitClockGating workaround. |
||
3799 | */ |
||
3800 | I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
||
3801 | |||
3802 | /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */ |
||
3803 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
||
3804 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); |
||
3805 | |||
3806 | /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */ |
||
3807 | I915_WRITE(GEN7_L3CNTLREG1, |
||
3808 | GEN7_WA_FOR_GEN7_L3_CONTROL); |
||
3809 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, |
||
3810 | GEN7_WA_L3_CHICKEN_MODE); |
||
3811 | |||
3812 | /* This is required by WaCatErrorRejectionIssue */ |
||
3813 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
||
3814 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
||
3815 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
||
3816 | |||
3817 | for_each_pipe(pipe) { |
||
3818 | I915_WRITE(DSPCNTR(pipe), |
||
3819 | I915_READ(DSPCNTR(pipe)) | |
||
3820 | DISPPLANE_TRICKLE_FEED_DISABLE); |
||
3821 | intel_flush_display_plane(dev_priv, pipe); |
||
3822 | } |
||
3823 | |||
3824 | gen7_setup_fixed_func_scheduler(dev_priv); |
||
3825 | |||
3826 | /* WaDisable4x2SubspanOptimization */ |
||
3827 | I915_WRITE(CACHE_MODE_1, |
||
3828 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
||
3829 | |||
3243 | Serge | 3830 | /* WaMbcDriverBootEnable */ |
3831 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
||
3832 | GEN6_MBCTL_ENABLE_BOOT_FETCH); |
||
3833 | |||
3746 | Serge | 3834 | /* WaSwitchSolVfFArbitrationPriority */ |
3835 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
||
3836 | |||
3031 | serge | 3837 | /* XXX: This is a workaround for early silicon revisions and should be |
3838 | * removed later. |
||
3839 | */ |
||
3840 | I915_WRITE(WM_DBG, |
||
3841 | I915_READ(WM_DBG) | |
||
3842 | WM_DBG_DISALLOW_MULTIPLE_LP | |
||
3843 | WM_DBG_DISALLOW_SPRITE | |
||
3844 | WM_DBG_DISALLOW_MAXFIFO); |
||
3845 | |||
3243 | Serge | 3846 | lpt_init_clock_gating(dev); |
3031 | serge | 3847 | } |
3848 | |||
3849 | static void ivybridge_init_clock_gating(struct drm_device *dev) |
||
3850 | { |
||
3851 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3852 | int pipe; |
||
3853 | uint32_t snpcr; |
||
3854 | |||
3855 | I915_WRITE(WM3_LP_ILK, 0); |
||
3856 | I915_WRITE(WM2_LP_ILK, 0); |
||
3857 | I915_WRITE(WM1_LP_ILK, 0); |
||
3858 | |||
3243 | Serge | 3859 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
3031 | serge | 3860 | |
3243 | Serge | 3861 | /* WaDisableEarlyCull */ |
3862 | I915_WRITE(_3D_CHICKEN3, |
||
3863 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); |
||
3864 | |||
3865 | /* WaDisableBackToBackFlipFix */ |
||
3031 | serge | 3866 | I915_WRITE(IVB_CHICKEN3, |
3867 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | |
||
3868 | CHICKEN3_DGMG_DONE_FIX_DISABLE); |
||
3869 | |||
3243 | Serge | 3870 | /* WaDisablePSDDualDispatchEnable */ |
3871 | if (IS_IVB_GT1(dev)) |
||
3872 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
||
3873 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
||
3874 | else |
||
3875 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2, |
||
3876 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
||
3877 | |||
3031 | serge | 3878 | /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */ |
3879 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
||
3880 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); |
||
3881 | |||
3882 | /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */ |
||
3883 | I915_WRITE(GEN7_L3CNTLREG1, |
||
3884 | GEN7_WA_FOR_GEN7_L3_CONTROL); |
||
3885 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, |
||
3886 | GEN7_WA_L3_CHICKEN_MODE); |
||
3243 | Serge | 3887 | if (IS_IVB_GT1(dev)) |
3888 | I915_WRITE(GEN7_ROW_CHICKEN2, |
||
3889 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
||
3890 | else |
||
3891 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, |
||
3892 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
||
3031 | serge | 3893 | |
3243 | Serge | 3894 | |
3895 | /* WaForceL3Serialization */ |
||
3896 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
||
3897 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
||
3898 | |||
3031 | serge | 3899 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock |
3900 | * gating disable must be set. Failure to set it results in |
||
3901 | * flickering pixels due to Z write ordering failures after |
||
3902 | * some amount of runtime in the Mesa "fire" demo, and Unigine |
||
3903 | * Sanctuary and Tropics, and apparently anything else with |
||
3904 | * alpha test or pixel discard. |
||
3905 | * |
||
3906 | * According to the spec, bit 11 (RCCUNIT) must also be set, |
||
3907 | * but we didn't debug actual testcases to find it out. |
||
3908 | * |
||
3909 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
||
3910 | * This implements the WaDisableRCZUnitClockGating workaround. |
||
3911 | */ |
||
3912 | I915_WRITE(GEN6_UCGCTL2, |
||
3913 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE | |
||
3914 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); |
||
3915 | |||
3916 | /* This is required by WaCatErrorRejectionIssue */ |
||
3917 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
||
3918 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
||
3919 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
||
3920 | |||
3921 | for_each_pipe(pipe) { |
||
3922 | I915_WRITE(DSPCNTR(pipe), |
||
3923 | I915_READ(DSPCNTR(pipe)) | |
||
3924 | DISPPLANE_TRICKLE_FEED_DISABLE); |
||
3925 | intel_flush_display_plane(dev_priv, pipe); |
||
3926 | } |
||
3927 | |||
3243 | Serge | 3928 | /* WaMbcDriverBootEnable */ |
3031 | serge | 3929 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
3930 | GEN6_MBCTL_ENABLE_BOOT_FETCH); |
||
3931 | |||
3932 | gen7_setup_fixed_func_scheduler(dev_priv); |
||
3933 | |||
3934 | /* WaDisable4x2SubspanOptimization */ |
||
3935 | I915_WRITE(CACHE_MODE_1, |
||
3936 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
||
3937 | |||
3938 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); |
||
3939 | snpcr &= ~GEN6_MBC_SNPCR_MASK; |
||
3940 | snpcr |= GEN6_MBC_SNPCR_MED; |
||
3941 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); |
||
3243 | Serge | 3942 | |
3746 | Serge | 3943 | if (!HAS_PCH_NOP(dev)) |
3243 | Serge | 3944 | cpt_init_clock_gating(dev); |
3480 | Serge | 3945 | |
3946 | gen6_check_mch_setup(dev); |
||
3031 | serge | 3947 | } |
3948 | |||
3949 | static void valleyview_init_clock_gating(struct drm_device *dev) |
||
3950 | { |
||
3951 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3952 | int pipe; |
||
3953 | |||
3954 | I915_WRITE(WM3_LP_ILK, 0); |
||
3955 | I915_WRITE(WM2_LP_ILK, 0); |
||
3956 | I915_WRITE(WM1_LP_ILK, 0); |
||
3957 | |||
3243 | Serge | 3958 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
3031 | serge | 3959 | |
3243 | Serge | 3960 | /* WaDisableEarlyCull */ |
3961 | I915_WRITE(_3D_CHICKEN3, |
||
3962 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); |
||
3963 | |||
3964 | /* WaDisableBackToBackFlipFix */ |
||
3031 | serge | 3965 | I915_WRITE(IVB_CHICKEN3, |
3966 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | |
||
3967 | CHICKEN3_DGMG_DONE_FIX_DISABLE); |
||
3968 | |||
3746 | Serge | 3969 | /* WaDisablePSDDualDispatchEnable */ |
3243 | Serge | 3970 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
3746 | Serge | 3971 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
3972 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); |
||
3243 | Serge | 3973 | |
3031 | serge | 3974 | /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */ |
3975 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
||
3976 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); |
||
3977 | |||
3978 | /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */ |
||
3243 | Serge | 3979 | I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS); |
3031 | serge | 3980 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE); |
3981 | |||
3243 | Serge | 3982 | /* WaForceL3Serialization */ |
3983 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
||
3984 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
||
3985 | |||
3986 | /* WaDisableDopClockGating */ |
||
3987 | I915_WRITE(GEN7_ROW_CHICKEN2, |
||
3988 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); |
||
3989 | |||
3990 | /* WaForceL3Serialization */ |
||
3991 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
||
3992 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); |
||
3993 | |||
3031 | serge | 3994 | /* This is required by WaCatErrorRejectionIssue */ |
3995 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
||
3996 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | |
||
3997 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); |
||
3998 | |||
3243 | Serge | 3999 | /* WaMbcDriverBootEnable */ |
3031 | serge | 4000 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
4001 | GEN6_MBCTL_ENABLE_BOOT_FETCH); |
||
4002 | |||
4003 | |||
4004 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock |
||
4005 | * gating disable must be set. Failure to set it results in |
||
4006 | * flickering pixels due to Z write ordering failures after |
||
4007 | * some amount of runtime in the Mesa "fire" demo, and Unigine |
||
4008 | * Sanctuary and Tropics, and apparently anything else with |
||
4009 | * alpha test or pixel discard. |
||
4010 | * |
||
4011 | * According to the spec, bit 11 (RCCUNIT) must also be set, |
||
4012 | * but we didn't debug actual testcases to find it out. |
||
4013 | * |
||
4014 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
||
4015 | * This implements the WaDisableRCZUnitClockGating workaround. |
||
4016 | * |
||
4017 | * Also apply WaDisableVDSUnitClockGating and |
||
4018 | * WaDisableRCPBUnitClockGating. |
||
4019 | */ |
||
4020 | I915_WRITE(GEN6_UCGCTL2, |
||
4021 | GEN7_VDSUNIT_CLOCK_GATE_DISABLE | |
||
4022 | GEN7_TDLUNIT_CLOCK_GATE_DISABLE | |
||
4023 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE | |
||
4024 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | |
||
4025 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); |
||
4026 | |||
4027 | I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE); |
||
4028 | |||
4029 | for_each_pipe(pipe) { |
||
4030 | I915_WRITE(DSPCNTR(pipe), |
||
4031 | I915_READ(DSPCNTR(pipe)) | |
||
4032 | DISPPLANE_TRICKLE_FEED_DISABLE); |
||
4033 | intel_flush_display_plane(dev_priv, pipe); |
||
4034 | } |
||
4035 | |||
4036 | I915_WRITE(CACHE_MODE_1, |
||
4037 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); |
||
4038 | |||
4039 | /* |
||
3243 | Serge | 4040 | * WaDisableVLVClockGating_VBIIssue |
4041 | * Disable clock gating on th GCFG unit to prevent a delay |
||
4042 | * in the reporting of vblank events. |
||
4043 | */ |
||
3746 | Serge | 4044 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, 0xffffffff); |
4045 | |||
4046 | /* Conservative clock gating settings for now */ |
||
4047 | I915_WRITE(0x9400, 0xffffffff); |
||
4048 | I915_WRITE(0x9404, 0xffffffff); |
||
4049 | I915_WRITE(0x9408, 0xffffffff); |
||
4050 | I915_WRITE(0x940c, 0xffffffff); |
||
4051 | I915_WRITE(0x9410, 0xffffffff); |
||
4052 | I915_WRITE(0x9414, 0xffffffff); |
||
4053 | I915_WRITE(0x9418, 0xffffffff); |
||
3031 | serge | 4054 | } |
4055 | |||
4056 | static void g4x_init_clock_gating(struct drm_device *dev) |
||
4057 | { |
||
4058 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4059 | uint32_t dspclk_gate; |
||
4060 | |||
4061 | I915_WRITE(RENCLK_GATE_D1, 0); |
||
4062 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | |
||
4063 | GS_UNIT_CLOCK_GATE_DISABLE | |
||
4064 | CL_UNIT_CLOCK_GATE_DISABLE); |
||
4065 | I915_WRITE(RAMCLK_GATE_D, 0); |
||
4066 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | |
||
4067 | OVRUNIT_CLOCK_GATE_DISABLE | |
||
4068 | OVCUNIT_CLOCK_GATE_DISABLE; |
||
4069 | if (IS_GM45(dev)) |
||
4070 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; |
||
4071 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); |
||
3243 | Serge | 4072 | |
4073 | /* WaDisableRenderCachePipelinedFlush */ |
||
4074 | I915_WRITE(CACHE_MODE_0, |
||
4075 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); |
||
3031 | serge | 4076 | } |
4077 | |||
4078 | static void crestline_init_clock_gating(struct drm_device *dev) |
||
4079 | { |
||
4080 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4081 | |||
4082 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); |
||
4083 | I915_WRITE(RENCLK_GATE_D2, 0); |
||
4084 | I915_WRITE(DSPCLK_GATE_D, 0); |
||
4085 | I915_WRITE(RAMCLK_GATE_D, 0); |
||
4086 | I915_WRITE16(DEUC, 0); |
||
4087 | } |
||
4088 | |||
4089 | static void broadwater_init_clock_gating(struct drm_device *dev) |
||
4090 | { |
||
4091 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4092 | |||
4093 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | |
||
4094 | I965_RCC_CLOCK_GATE_DISABLE | |
||
4095 | I965_RCPB_CLOCK_GATE_DISABLE | |
||
4096 | I965_ISC_CLOCK_GATE_DISABLE | |
||
4097 | I965_FBC_CLOCK_GATE_DISABLE); |
||
4098 | I915_WRITE(RENCLK_GATE_D2, 0); |
||
4099 | } |
||
4100 | |||
4101 | static void gen3_init_clock_gating(struct drm_device *dev) |
||
4102 | { |
||
4103 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4104 | u32 dstate = I915_READ(D_STATE); |
||
4105 | |||
4106 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | |
||
4107 | DSTATE_DOT_CLOCK_GATING; |
||
4108 | I915_WRITE(D_STATE, dstate); |
||
4109 | |||
4110 | if (IS_PINEVIEW(dev)) |
||
4111 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); |
||
4112 | |||
4113 | /* IIR "flip pending" means done if this bit is set */ |
||
4114 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); |
||
4115 | } |
||
4116 | |||
4117 | static void i85x_init_clock_gating(struct drm_device *dev) |
||
4118 | { |
||
4119 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4120 | |||
4121 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); |
||
4122 | } |
||
4123 | |||
4124 | static void i830_init_clock_gating(struct drm_device *dev) |
||
4125 | { |
||
4126 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4127 | |||
4128 | I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE); |
||
4129 | } |
||
4130 | |||
4131 | void intel_init_clock_gating(struct drm_device *dev) |
||
4132 | { |
||
4133 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4134 | |||
4135 | dev_priv->display.init_clock_gating(dev); |
||
4136 | } |
||
4137 | |||
3746 | Serge | 4138 | /** |
4139 | * We should only use the power well if we explicitly asked the hardware to |
||
4140 | * enable it, so check if it's enabled and also check if we've requested it to |
||
4141 | * be enabled. |
||
4142 | */ |
||
4143 | bool intel_using_power_well(struct drm_device *dev) |
||
4144 | { |
||
4145 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4146 | |||
4147 | if (IS_HASWELL(dev)) |
||
4148 | return I915_READ(HSW_PWR_WELL_DRIVER) == |
||
4149 | (HSW_PWR_WELL_ENABLE | HSW_PWR_WELL_STATE); |
||
4150 | else |
||
4151 | return true; |
||
4152 | } |
||
4153 | |||
3480 | Serge | 4154 | void intel_set_power_well(struct drm_device *dev, bool enable) |
3031 | serge | 4155 | { |
4156 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
3480 | Serge | 4157 | bool is_enabled, enable_requested; |
4158 | uint32_t tmp; |
||
3031 | serge | 4159 | |
3746 | Serge | 4160 | if (!HAS_POWER_WELL(dev)) |
3031 | serge | 4161 | return; |
4162 | |||
3482 | Serge | 4163 | if (!i915_disable_power_well && !enable) |
4164 | return; |
||
4165 | |||
3480 | Serge | 4166 | tmp = I915_READ(HSW_PWR_WELL_DRIVER); |
4167 | is_enabled = tmp & HSW_PWR_WELL_STATE; |
||
4168 | enable_requested = tmp & HSW_PWR_WELL_ENABLE; |
||
3031 | serge | 4169 | |
3480 | Serge | 4170 | if (enable) { |
4171 | if (!enable_requested) |
||
4172 | I915_WRITE(HSW_PWR_WELL_DRIVER, HSW_PWR_WELL_ENABLE); |
||
3031 | serge | 4173 | |
3480 | Serge | 4174 | if (!is_enabled) { |
4175 | DRM_DEBUG_KMS("Enabling power well\n"); |
||
4176 | if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) & |
||
4177 | HSW_PWR_WELL_STATE), 20)) |
||
4178 | DRM_ERROR("Timeout enabling power well\n"); |
||
3031 | serge | 4179 | } |
3480 | Serge | 4180 | } else { |
4181 | if (enable_requested) { |
||
4182 | I915_WRITE(HSW_PWR_WELL_DRIVER, 0); |
||
4183 | DRM_DEBUG_KMS("Requesting to disable the power well\n"); |
||
3031 | serge | 4184 | } |
3480 | Serge | 4185 | } |
4186 | } |
||
3031 | serge | 4187 | |
3480 | Serge | 4188 | /* |
4189 | * Starting with Haswell, we have a "Power Down Well" that can be turned off |
||
4190 | * when not needed anymore. We have 4 registers that can request the power well |
||
4191 | * to be enabled, and it will only be disabled if none of the registers is |
||
4192 | * requesting it to be enabled. |
||
4193 | */ |
||
4194 | void intel_init_power_well(struct drm_device *dev) |
||
4195 | { |
||
4196 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4197 | |||
3746 | Serge | 4198 | if (!HAS_POWER_WELL(dev)) |
3480 | Serge | 4199 | return; |
4200 | |||
4201 | /* For now, we need the power well to be always enabled. */ |
||
4202 | intel_set_power_well(dev, true); |
||
4203 | |||
4204 | /* We're taking over the BIOS, so clear any requests made by it since |
||
4205 | * the driver is in charge now. */ |
||
4206 | if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE) |
||
4207 | I915_WRITE(HSW_PWR_WELL_BIOS, 0); |
||
3031 | serge | 4208 | } |
4209 | |||
4210 | /* Set up chip specific power management-related functions */ |
||
4211 | void intel_init_pm(struct drm_device *dev) |
||
4212 | { |
||
4213 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4214 | |||
4215 | if (I915_HAS_FBC(dev)) { |
||
4216 | if (HAS_PCH_SPLIT(dev)) { |
||
4217 | dev_priv->display.fbc_enabled = ironlake_fbc_enabled; |
||
4218 | dev_priv->display.enable_fbc = ironlake_enable_fbc; |
||
4219 | dev_priv->display.disable_fbc = ironlake_disable_fbc; |
||
4220 | } else if (IS_GM45(dev)) { |
||
4221 | dev_priv->display.fbc_enabled = g4x_fbc_enabled; |
||
4222 | dev_priv->display.enable_fbc = g4x_enable_fbc; |
||
4223 | dev_priv->display.disable_fbc = g4x_disable_fbc; |
||
4224 | } else if (IS_CRESTLINE(dev)) { |
||
4225 | dev_priv->display.fbc_enabled = i8xx_fbc_enabled; |
||
4226 | dev_priv->display.enable_fbc = i8xx_enable_fbc; |
||
4227 | dev_priv->display.disable_fbc = i8xx_disable_fbc; |
||
4228 | } |
||
4229 | /* 855GM needs testing */ |
||
4230 | } |
||
4231 | |||
4232 | /* For cxsr */ |
||
4233 | if (IS_PINEVIEW(dev)) |
||
4234 | i915_pineview_get_mem_freq(dev); |
||
4235 | else if (IS_GEN5(dev)) |
||
4236 | i915_ironlake_get_mem_freq(dev); |
||
4237 | |||
4238 | /* For FIFO watermark updates */ |
||
4239 | if (HAS_PCH_SPLIT(dev)) { |
||
4240 | if (IS_GEN5(dev)) { |
||
4241 | if (I915_READ(MLTR_ILK) & ILK_SRLT_MASK) |
||
4242 | dev_priv->display.update_wm = ironlake_update_wm; |
||
4243 | else { |
||
4244 | DRM_DEBUG_KMS("Failed to get proper latency. " |
||
4245 | "Disable CxSR\n"); |
||
4246 | dev_priv->display.update_wm = NULL; |
||
4247 | } |
||
4248 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; |
||
4249 | } else if (IS_GEN6(dev)) { |
||
4250 | if (SNB_READ_WM0_LATENCY()) { |
||
4251 | dev_priv->display.update_wm = sandybridge_update_wm; |
||
4252 | dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm; |
||
4253 | } else { |
||
4254 | DRM_DEBUG_KMS("Failed to read display plane latency. " |
||
4255 | "Disable CxSR\n"); |
||
4256 | dev_priv->display.update_wm = NULL; |
||
4257 | } |
||
4258 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; |
||
4259 | } else if (IS_IVYBRIDGE(dev)) { |
||
4260 | if (SNB_READ_WM0_LATENCY()) { |
||
3243 | Serge | 4261 | dev_priv->display.update_wm = ivybridge_update_wm; |
3031 | serge | 4262 | dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm; |
4263 | } else { |
||
4264 | DRM_DEBUG_KMS("Failed to read display plane latency. " |
||
4265 | "Disable CxSR\n"); |
||
4266 | dev_priv->display.update_wm = NULL; |
||
4267 | } |
||
4268 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; |
||
4269 | } else if (IS_HASWELL(dev)) { |
||
4270 | if (SNB_READ_WM0_LATENCY()) { |
||
4271 | dev_priv->display.update_wm = sandybridge_update_wm; |
||
4272 | dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm; |
||
4273 | dev_priv->display.update_linetime_wm = haswell_update_linetime_wm; |
||
4274 | } else { |
||
4275 | DRM_DEBUG_KMS("Failed to read display plane latency. " |
||
4276 | "Disable CxSR\n"); |
||
4277 | dev_priv->display.update_wm = NULL; |
||
4278 | } |
||
4279 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; |
||
4280 | } else |
||
4281 | dev_priv->display.update_wm = NULL; |
||
4282 | } else if (IS_VALLEYVIEW(dev)) { |
||
4283 | dev_priv->display.update_wm = valleyview_update_wm; |
||
4284 | dev_priv->display.init_clock_gating = |
||
4285 | valleyview_init_clock_gating; |
||
4286 | } else if (IS_PINEVIEW(dev)) { |
||
4287 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev), |
||
4288 | dev_priv->is_ddr3, |
||
4289 | dev_priv->fsb_freq, |
||
4290 | dev_priv->mem_freq)) { |
||
4291 | DRM_INFO("failed to find known CxSR latency " |
||
4292 | "(found ddr%s fsb freq %d, mem freq %d), " |
||
4293 | "disabling CxSR\n", |
||
4294 | (dev_priv->is_ddr3 == 1) ? "3" : "2", |
||
4295 | dev_priv->fsb_freq, dev_priv->mem_freq); |
||
4296 | /* Disable CxSR and never update its watermark again */ |
||
4297 | pineview_disable_cxsr(dev); |
||
4298 | dev_priv->display.update_wm = NULL; |
||
4299 | } else |
||
4300 | dev_priv->display.update_wm = pineview_update_wm; |
||
4301 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; |
||
4302 | } else if (IS_G4X(dev)) { |
||
4303 | dev_priv->display.update_wm = g4x_update_wm; |
||
4304 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; |
||
4305 | } else if (IS_GEN4(dev)) { |
||
4306 | dev_priv->display.update_wm = i965_update_wm; |
||
4307 | if (IS_CRESTLINE(dev)) |
||
4308 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; |
||
4309 | else if (IS_BROADWATER(dev)) |
||
4310 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; |
||
4311 | } else if (IS_GEN3(dev)) { |
||
4312 | dev_priv->display.update_wm = i9xx_update_wm; |
||
4313 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; |
||
4314 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; |
||
4315 | } else if (IS_I865G(dev)) { |
||
4316 | dev_priv->display.update_wm = i830_update_wm; |
||
4317 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; |
||
4318 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
||
4319 | } else if (IS_I85X(dev)) { |
||
4320 | dev_priv->display.update_wm = i9xx_update_wm; |
||
4321 | dev_priv->display.get_fifo_size = i85x_get_fifo_size; |
||
4322 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; |
||
4323 | } else { |
||
4324 | dev_priv->display.update_wm = i830_update_wm; |
||
4325 | dev_priv->display.init_clock_gating = i830_init_clock_gating; |
||
4326 | if (IS_845G(dev)) |
||
4327 | dev_priv->display.get_fifo_size = i845_get_fifo_size; |
||
4328 | else |
||
4329 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
||
4330 | } |
||
4331 | } |
||
4332 | |||
4333 | static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv) |
||
4334 | { |
||
4335 | u32 gt_thread_status_mask; |
||
4336 | |||
4337 | if (IS_HASWELL(dev_priv->dev)) |
||
4338 | gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK_HSW; |
||
4339 | else |
||
4340 | gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK; |
||
4341 | |||
4342 | /* w/a for a sporadic read returning 0 by waiting for the GT |
||
4343 | * thread to wake up. |
||
4344 | */ |
||
4345 | if (wait_for_atomic_us((I915_READ_NOTRACE(GEN6_GT_THREAD_STATUS_REG) & gt_thread_status_mask) == 0, 500)) |
||
4346 | DRM_ERROR("GT thread status wait timed out\n"); |
||
4347 | } |
||
4348 | |||
3243 | Serge | 4349 | static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv) |
4350 | { |
||
4351 | I915_WRITE_NOTRACE(FORCEWAKE, 0); |
||
4352 | POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */ |
||
4353 | } |
||
4354 | |||
3031 | serge | 4355 | static void __gen6_gt_force_wake_get(struct drm_i915_private *dev_priv) |
4356 | { |
||
3746 | Serge | 4357 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1) == 0, |
3031 | serge | 4358 | FORCEWAKE_ACK_TIMEOUT_MS)) |
4359 | DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n"); |
||
4360 | |||
3746 | Serge | 4361 | I915_WRITE_NOTRACE(FORCEWAKE, 1); |
3031 | serge | 4362 | POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */ |
4363 | |||
3746 | Serge | 4364 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1), |
3031 | serge | 4365 | FORCEWAKE_ACK_TIMEOUT_MS)) |
4366 | DRM_ERROR("Timed out waiting for forcewake to ack request.\n"); |
||
4367 | |||
4368 | __gen6_gt_wait_for_thread_c0(dev_priv); |
||
4369 | } |
||
4370 | |||
3243 | Serge | 4371 | static void __gen6_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv) |
4372 | { |
||
4373 | I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(0xffff)); |
||
4374 | /* something from same cacheline, but !FORCEWAKE_MT */ |
||
4375 | POSTING_READ(ECOBUS); |
||
4376 | } |
||
4377 | |||
3031 | serge | 4378 | static void __gen6_gt_force_wake_mt_get(struct drm_i915_private *dev_priv) |
4379 | { |
||
4380 | u32 forcewake_ack; |
||
4381 | |||
4382 | if (IS_HASWELL(dev_priv->dev)) |
||
4383 | forcewake_ack = FORCEWAKE_ACK_HSW; |
||
4384 | else |
||
4385 | forcewake_ack = FORCEWAKE_MT_ACK; |
||
4386 | |||
3746 | Serge | 4387 | if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL) == 0, |
3031 | serge | 4388 | FORCEWAKE_ACK_TIMEOUT_MS)) |
4389 | DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n"); |
||
4390 | |||
3243 | Serge | 4391 | I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL)); |
4392 | /* something from same cacheline, but !FORCEWAKE_MT */ |
||
4393 | POSTING_READ(ECOBUS); |
||
3031 | serge | 4394 | |
3746 | Serge | 4395 | if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL), |
3031 | serge | 4396 | FORCEWAKE_ACK_TIMEOUT_MS)) |
4397 | DRM_ERROR("Timed out waiting for forcewake to ack request.\n"); |
||
4398 | |||
4399 | __gen6_gt_wait_for_thread_c0(dev_priv); |
||
4400 | } |
||
4401 | |||
4402 | /* |
||
4403 | * Generally this is called implicitly by the register read function. However, |
||
4404 | * if some sequence requires the GT to not power down then this function should |
||
4405 | * be called at the beginning of the sequence followed by a call to |
||
4406 | * gen6_gt_force_wake_put() at the end of the sequence. |
||
4407 | */ |
||
4408 | void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv) |
||
4409 | { |
||
4410 | unsigned long irqflags; |
||
4411 | |||
4412 | spin_lock_irqsave(&dev_priv->gt_lock, irqflags); |
||
4413 | if (dev_priv->forcewake_count++ == 0) |
||
4414 | dev_priv->gt.force_wake_get(dev_priv); |
||
4415 | spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags); |
||
4416 | } |
||
4417 | |||
4418 | void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv) |
||
4419 | { |
||
4420 | u32 gtfifodbg; |
||
4421 | gtfifodbg = I915_READ_NOTRACE(GTFIFODBG); |
||
4422 | if (WARN(gtfifodbg & GT_FIFO_CPU_ERROR_MASK, |
||
4423 | "MMIO read or write has been dropped %x\n", gtfifodbg)) |
||
4424 | I915_WRITE_NOTRACE(GTFIFODBG, GT_FIFO_CPU_ERROR_MASK); |
||
4425 | } |
||
4426 | |||
4427 | static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv) |
||
4428 | { |
||
4429 | I915_WRITE_NOTRACE(FORCEWAKE, 0); |
||
3243 | Serge | 4430 | /* something from same cacheline, but !FORCEWAKE */ |
4431 | POSTING_READ(ECOBUS); |
||
3031 | serge | 4432 | gen6_gt_check_fifodbg(dev_priv); |
4433 | } |
||
4434 | |||
4435 | static void __gen6_gt_force_wake_mt_put(struct drm_i915_private *dev_priv) |
||
4436 | { |
||
3243 | Serge | 4437 | I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL)); |
4438 | /* something from same cacheline, but !FORCEWAKE_MT */ |
||
4439 | POSTING_READ(ECOBUS); |
||
3031 | serge | 4440 | gen6_gt_check_fifodbg(dev_priv); |
4441 | } |
||
4442 | |||
4443 | /* |
||
4444 | * see gen6_gt_force_wake_get() |
||
4445 | */ |
||
4446 | void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv) |
||
4447 | { |
||
4448 | unsigned long irqflags; |
||
4449 | |||
4450 | spin_lock_irqsave(&dev_priv->gt_lock, irqflags); |
||
4451 | if (--dev_priv->forcewake_count == 0) |
||
4452 | dev_priv->gt.force_wake_put(dev_priv); |
||
4453 | spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags); |
||
4454 | } |
||
4455 | |||
4456 | int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv) |
||
4457 | { |
||
4458 | int ret = 0; |
||
4459 | |||
4460 | if (dev_priv->gt_fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) { |
||
4461 | int loop = 500; |
||
4462 | u32 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES); |
||
4463 | while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) { |
||
4464 | udelay(10); |
||
4465 | fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES); |
||
4466 | } |
||
4467 | if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES)) |
||
4468 | ++ret; |
||
4469 | dev_priv->gt_fifo_count = fifo; |
||
4470 | } |
||
4471 | dev_priv->gt_fifo_count--; |
||
4472 | |||
4473 | return ret; |
||
4474 | } |
||
4475 | |||
3243 | Serge | 4476 | static void vlv_force_wake_reset(struct drm_i915_private *dev_priv) |
4477 | { |
||
4478 | I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(0xffff)); |
||
4479 | /* something from same cacheline, but !FORCEWAKE_VLV */ |
||
4480 | POSTING_READ(FORCEWAKE_ACK_VLV); |
||
4481 | } |
||
4482 | |||
3031 | serge | 4483 | static void vlv_force_wake_get(struct drm_i915_private *dev_priv) |
4484 | { |
||
3746 | Serge | 4485 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL) == 0, |
3031 | serge | 4486 | FORCEWAKE_ACK_TIMEOUT_MS)) |
4487 | DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n"); |
||
4488 | |||
3243 | Serge | 4489 | I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL)); |
3746 | Serge | 4490 | I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV, |
4491 | _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL)); |
||
3031 | serge | 4492 | |
3746 | Serge | 4493 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL), |
3031 | serge | 4494 | FORCEWAKE_ACK_TIMEOUT_MS)) |
3746 | Serge | 4495 | DRM_ERROR("Timed out waiting for GT to ack forcewake request.\n"); |
3031 | serge | 4496 | |
3746 | Serge | 4497 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_MEDIA_VLV) & |
4498 | FORCEWAKE_KERNEL), |
||
4499 | FORCEWAKE_ACK_TIMEOUT_MS)) |
||
4500 | DRM_ERROR("Timed out waiting for media to ack forcewake request.\n"); |
||
4501 | |||
3031 | serge | 4502 | __gen6_gt_wait_for_thread_c0(dev_priv); |
4503 | } |
||
4504 | |||
4505 | static void vlv_force_wake_put(struct drm_i915_private *dev_priv) |
||
4506 | { |
||
3243 | Serge | 4507 | I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL)); |
3746 | Serge | 4508 | I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV, |
4509 | _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL)); |
||
4510 | /* The below doubles as a POSTING_READ */ |
||
3031 | serge | 4511 | gen6_gt_check_fifodbg(dev_priv); |
4512 | } |
||
4513 | |||
3243 | Serge | 4514 | void intel_gt_reset(struct drm_device *dev) |
4515 | { |
||
4516 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4517 | |||
4518 | if (IS_VALLEYVIEW(dev)) { |
||
4519 | vlv_force_wake_reset(dev_priv); |
||
4520 | } else if (INTEL_INFO(dev)->gen >= 6) { |
||
4521 | __gen6_gt_force_wake_reset(dev_priv); |
||
4522 | if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) |
||
4523 | __gen6_gt_force_wake_mt_reset(dev_priv); |
||
4524 | } |
||
4525 | } |
||
4526 | |||
3031 | serge | 4527 | void intel_gt_init(struct drm_device *dev) |
4528 | { |
||
4529 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
4530 | |||
4531 | spin_lock_init(&dev_priv->gt_lock); |
||
4532 | |||
3243 | Serge | 4533 | intel_gt_reset(dev); |
4534 | |||
3031 | serge | 4535 | if (IS_VALLEYVIEW(dev)) { |
4536 | dev_priv->gt.force_wake_get = vlv_force_wake_get; |
||
4537 | dev_priv->gt.force_wake_put = vlv_force_wake_put; |
||
3243 | Serge | 4538 | } else if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) { |
4539 | dev_priv->gt.force_wake_get = __gen6_gt_force_wake_mt_get; |
||
4540 | dev_priv->gt.force_wake_put = __gen6_gt_force_wake_mt_put; |
||
4541 | } else if (IS_GEN6(dev)) { |
||
3031 | serge | 4542 | dev_priv->gt.force_wake_get = __gen6_gt_force_wake_get; |
4543 | dev_priv->gt.force_wake_put = __gen6_gt_force_wake_put; |
||
3243 | Serge | 4544 | } |
3482 | Serge | 4545 | INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work, |
4546 | intel_gen6_powersave_work); |
||
3243 | Serge | 4547 | } |
3031 | serge | 4548 | |
3243 | Serge | 4549 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val) |
4550 | { |
||
4551 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
||
3031 | serge | 4552 | |
3243 | Serge | 4553 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { |
4554 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); |
||
4555 | return -EAGAIN; |
||
4556 | } |
||
3031 | serge | 4557 | |
3243 | Serge | 4558 | I915_WRITE(GEN6_PCODE_DATA, *val); |
4559 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
||
4560 | |||
4561 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, |
||
4562 | 500)) { |
||
4563 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); |
||
4564 | return -ETIMEDOUT; |
||
3031 | serge | 4565 | } |
3243 | Serge | 4566 | |
4567 | *val = I915_READ(GEN6_PCODE_DATA); |
||
4568 | I915_WRITE(GEN6_PCODE_DATA, 0); |
||
4569 | |||
4570 | return 0; |
||
4571 | } |
||
4572 | |||
4573 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val) |
||
4574 | { |
||
4575 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
||
4576 | |||
4577 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { |
||
4578 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); |
||
4579 | return -EAGAIN; |
||
3031 | serge | 4580 | } |
3243 | Serge | 4581 | |
4582 | I915_WRITE(GEN6_PCODE_DATA, val); |
||
4583 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
||
4584 | |||
4585 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, |
||
4586 | 500)) { |
||
4587 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); |
||
4588 | return -ETIMEDOUT; |
||
3031 | serge | 4589 | } |
3243 | Serge | 4590 | |
4591 | I915_WRITE(GEN6_PCODE_DATA, 0); |
||
4592 | |||
4593 | return 0; |
||
3031 | serge | 4594 | } |
3746 | Serge | 4595 | |
4596 | static int vlv_punit_rw(struct drm_i915_private *dev_priv, u8 opcode, |
||
4597 | u8 addr, u32 *val) |
||
4598 | { |
||
4599 | u32 cmd, devfn, port, be, bar; |
||
4600 | |||
4601 | bar = 0; |
||
4602 | be = 0xf; |
||
4603 | port = IOSF_PORT_PUNIT; |
||
4604 | devfn = PCI_DEVFN(2, 0); |
||
4605 | |||
4606 | cmd = (devfn << IOSF_DEVFN_SHIFT) | (opcode << IOSF_OPCODE_SHIFT) | |
||
4607 | (port << IOSF_PORT_SHIFT) | (be << IOSF_BYTE_ENABLES_SHIFT) | |
||
4608 | (bar << IOSF_BAR_SHIFT); |
||
4609 | |||
4610 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
||
4611 | |||
4612 | if (I915_READ(VLV_IOSF_DOORBELL_REQ) & IOSF_SB_BUSY) { |
||
4613 | DRM_DEBUG_DRIVER("warning: pcode (%s) mailbox access failed\n", |
||
4614 | opcode == PUNIT_OPCODE_REG_READ ? |
||
4615 | "read" : "write"); |
||
4616 | return -EAGAIN; |
||
4617 | } |
||
4618 | |||
4619 | I915_WRITE(VLV_IOSF_ADDR, addr); |
||
4620 | if (opcode == PUNIT_OPCODE_REG_WRITE) |
||
4621 | I915_WRITE(VLV_IOSF_DATA, *val); |
||
4622 | I915_WRITE(VLV_IOSF_DOORBELL_REQ, cmd); |
||
4623 | |||
4624 | if (wait_for((I915_READ(VLV_IOSF_DOORBELL_REQ) & IOSF_SB_BUSY) == 0, |
||
4625 | 500)) { |
||
4626 | DRM_ERROR("timeout waiting for pcode %s (%d) to finish\n", |
||
4627 | opcode == PUNIT_OPCODE_REG_READ ? "read" : "write", |
||
4628 | addr); |
||
4629 | return -ETIMEDOUT; |
||
4630 | } |
||
4631 | |||
4632 | if (opcode == PUNIT_OPCODE_REG_READ) |
||
4633 | *val = I915_READ(VLV_IOSF_DATA); |
||
4634 | I915_WRITE(VLV_IOSF_DATA, 0); |
||
4635 | |||
4636 | return 0; |
||
4637 | } |
||
4638 | |||
4639 | int valleyview_punit_read(struct drm_i915_private *dev_priv, u8 addr, u32 *val) |
||
4640 | { |
||
4641 | return vlv_punit_rw(dev_priv, PUNIT_OPCODE_REG_READ, addr, val); |
||
4642 | } |
||
4643 | |||
4644 | int valleyview_punit_write(struct drm_i915_private *dev_priv, u8 addr, u32 val) |
||
4645 | { |
||
4646 | return vlv_punit_rw(dev_priv, PUNIT_OPCODE_REG_WRITE, addr, &val); |
||
4647 | }><>><>><>><>><>=>>=>>><>><>>><>><>><>>>>>>>>>>=> |