Rev 5271 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 5271 | Rev 6104 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright 2011 Advanced Micro Devices, Inc. |
2 | * Copyright 2011 Advanced Micro Devices, Inc. |
3 | * All Rights Reserved. |
3 | * All Rights Reserved. |
4 | * |
4 | * |
5 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * Permission is hereby granted, free of charge, to any person obtaining a |
6 | * copy of this software and associated documentation files (the |
6 | * copy of this software and associated documentation files (the |
7 | * "Software"), to deal in the Software without restriction, including |
7 | * "Software"), to deal in the Software without restriction, including |
8 | * without limitation the rights to use, copy, modify, merge, publish, |
8 | * without limitation the rights to use, copy, modify, merge, publish, |
9 | * distribute, sub license, and/or sell copies of the Software, and to |
9 | * distribute, sub license, and/or sell copies of the Software, and to |
10 | * permit persons to whom the Software is furnished to do so, subject to |
10 | * permit persons to whom the Software is furnished to do so, subject to |
11 | * the following conditions: |
11 | * the following conditions: |
12 | * |
12 | * |
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL |
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL |
16 | * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, |
16 | * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, |
17 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
17 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
18 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
18 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
19 | * USE OR OTHER DEALINGS IN THE SOFTWARE. |
19 | * USE OR OTHER DEALINGS IN THE SOFTWARE. |
20 | * |
20 | * |
21 | * The above copyright notice and this permission notice (including the |
21 | * The above copyright notice and this permission notice (including the |
22 | * next paragraph) shall be included in all copies or substantial portions |
22 | * next paragraph) shall be included in all copies or substantial portions |
23 | * of the Software. |
23 | * of the Software. |
24 | * |
24 | * |
25 | */ |
25 | */ |
26 | /* |
26 | /* |
27 | * Authors: |
27 | * Authors: |
28 | * Christian König |
28 | * Christian König |
29 | */ |
29 | */ |
30 | 30 | ||
31 | #include |
31 | #include |
32 | #include |
32 | #include |
33 | #include |
33 | #include |
34 | #include |
34 | #include |
35 | 35 | ||
36 | #include "radeon.h" |
36 | #include "radeon.h" |
37 | #include "r600d.h" |
37 | #include "r600d.h" |
38 | 38 | ||
39 | /* 1 second timeout */ |
39 | /* 1 second timeout */ |
40 | #define UVD_IDLE_TIMEOUT_MS 1000 |
40 | #define UVD_IDLE_TIMEOUT_MS 1000 |
41 | 41 | ||
42 | /* Firmware Names */ |
42 | /* Firmware Names */ |
- | 43 | #define FIRMWARE_R600 "radeon/R600_uvd.bin" |
|
- | 44 | #define FIRMWARE_RS780 "radeon/RS780_uvd.bin" |
|
- | 45 | #define FIRMWARE_RV770 "radeon/RV770_uvd.bin" |
|
43 | #define FIRMWARE_RV710 "radeon/RV710_uvd.bin" |
46 | #define FIRMWARE_RV710 "radeon/RV710_uvd.bin" |
44 | #define FIRMWARE_CYPRESS "radeon/CYPRESS_uvd.bin" |
47 | #define FIRMWARE_CYPRESS "radeon/CYPRESS_uvd.bin" |
45 | #define FIRMWARE_SUMO "radeon/SUMO_uvd.bin" |
48 | #define FIRMWARE_SUMO "radeon/SUMO_uvd.bin" |
46 | #define FIRMWARE_TAHITI "radeon/TAHITI_uvd.bin" |
49 | #define FIRMWARE_TAHITI "radeon/TAHITI_uvd.bin" |
47 | #define FIRMWARE_BONAIRE "radeon/BONAIRE_uvd.bin" |
50 | #define FIRMWARE_BONAIRE "radeon/BONAIRE_uvd.bin" |
48 | 51 | ||
49 | MODULE_FIRMWARE(FIRMWARE_R600); |
52 | MODULE_FIRMWARE(FIRMWARE_R600); |
50 | MODULE_FIRMWARE(FIRMWARE_RS780); |
53 | MODULE_FIRMWARE(FIRMWARE_RS780); |
51 | MODULE_FIRMWARE(FIRMWARE_RV770); |
54 | MODULE_FIRMWARE(FIRMWARE_RV770); |
52 | MODULE_FIRMWARE(FIRMWARE_RV710); |
55 | MODULE_FIRMWARE(FIRMWARE_RV710); |
53 | MODULE_FIRMWARE(FIRMWARE_CYPRESS); |
56 | MODULE_FIRMWARE(FIRMWARE_CYPRESS); |
54 | MODULE_FIRMWARE(FIRMWARE_SUMO); |
57 | MODULE_FIRMWARE(FIRMWARE_SUMO); |
55 | MODULE_FIRMWARE(FIRMWARE_TAHITI); |
58 | MODULE_FIRMWARE(FIRMWARE_TAHITI); |
56 | MODULE_FIRMWARE(FIRMWARE_BONAIRE); |
59 | MODULE_FIRMWARE(FIRMWARE_BONAIRE); |
57 | 60 | ||
58 | static void radeon_uvd_idle_work_handler(struct work_struct *work); |
61 | static void radeon_uvd_idle_work_handler(struct work_struct *work); |
59 | 62 | ||
60 | int radeon_uvd_init(struct radeon_device *rdev) |
63 | int radeon_uvd_init(struct radeon_device *rdev) |
61 | { |
64 | { |
62 | unsigned long bo_size; |
65 | unsigned long bo_size; |
63 | const char *fw_name; |
66 | const char *fw_name; |
64 | int i, r; |
67 | int i, r; |
65 | 68 | ||
66 | // INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler); |
69 | // INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler); |
67 | 70 | ||
68 | switch (rdev->family) { |
71 | switch (rdev->family) { |
- | 72 | case CHIP_RV610: |
|
- | 73 | case CHIP_RV630: |
|
- | 74 | case CHIP_RV670: |
|
- | 75 | case CHIP_RV620: |
|
- | 76 | case CHIP_RV635: |
|
- | 77 | fw_name = FIRMWARE_R600; |
|
- | 78 | break; |
|
- | 79 | ||
- | 80 | case CHIP_RS780: |
|
- | 81 | case CHIP_RS880: |
|
- | 82 | fw_name = FIRMWARE_RS780; |
|
- | 83 | break; |
|
- | 84 | ||
- | 85 | case CHIP_RV770: |
|
- | 86 | fw_name = FIRMWARE_RV770; |
|
- | 87 | break; |
|
- | 88 | ||
69 | case CHIP_RV710: |
89 | case CHIP_RV710: |
70 | case CHIP_RV730: |
90 | case CHIP_RV730: |
71 | case CHIP_RV740: |
91 | case CHIP_RV740: |
72 | fw_name = FIRMWARE_RV710; |
92 | fw_name = FIRMWARE_RV710; |
73 | break; |
93 | break; |
74 | 94 | ||
75 | case CHIP_CYPRESS: |
95 | case CHIP_CYPRESS: |
76 | case CHIP_HEMLOCK: |
96 | case CHIP_HEMLOCK: |
77 | case CHIP_JUNIPER: |
97 | case CHIP_JUNIPER: |
78 | case CHIP_REDWOOD: |
98 | case CHIP_REDWOOD: |
79 | case CHIP_CEDAR: |
99 | case CHIP_CEDAR: |
80 | fw_name = FIRMWARE_CYPRESS; |
100 | fw_name = FIRMWARE_CYPRESS; |
81 | break; |
101 | break; |
82 | 102 | ||
83 | case CHIP_SUMO: |
103 | case CHIP_SUMO: |
84 | case CHIP_SUMO2: |
104 | case CHIP_SUMO2: |
85 | case CHIP_PALM: |
105 | case CHIP_PALM: |
86 | case CHIP_CAYMAN: |
106 | case CHIP_CAYMAN: |
87 | case CHIP_BARTS: |
107 | case CHIP_BARTS: |
88 | case CHIP_TURKS: |
108 | case CHIP_TURKS: |
89 | case CHIP_CAICOS: |
109 | case CHIP_CAICOS: |
90 | fw_name = FIRMWARE_SUMO; |
110 | fw_name = FIRMWARE_SUMO; |
91 | break; |
111 | break; |
92 | 112 | ||
93 | case CHIP_TAHITI: |
113 | case CHIP_TAHITI: |
94 | case CHIP_VERDE: |
114 | case CHIP_VERDE: |
95 | case CHIP_PITCAIRN: |
115 | case CHIP_PITCAIRN: |
96 | case CHIP_ARUBA: |
116 | case CHIP_ARUBA: |
97 | case CHIP_OLAND: |
117 | case CHIP_OLAND: |
98 | fw_name = FIRMWARE_TAHITI; |
118 | fw_name = FIRMWARE_TAHITI; |
99 | break; |
119 | break; |
100 | 120 | ||
101 | case CHIP_BONAIRE: |
121 | case CHIP_BONAIRE: |
102 | case CHIP_KABINI: |
122 | case CHIP_KABINI: |
103 | case CHIP_KAVERI: |
123 | case CHIP_KAVERI: |
104 | case CHIP_HAWAII: |
124 | case CHIP_HAWAII: |
105 | case CHIP_MULLINS: |
125 | case CHIP_MULLINS: |
106 | fw_name = FIRMWARE_BONAIRE; |
126 | fw_name = FIRMWARE_BONAIRE; |
107 | break; |
127 | break; |
108 | 128 | ||
109 | default: |
129 | default: |
110 | return -EINVAL; |
130 | return -EINVAL; |
111 | } |
131 | } |
112 | 132 | ||
113 | r = request_firmware(&rdev->uvd_fw, fw_name, rdev->dev); |
133 | r = request_firmware(&rdev->uvd_fw, fw_name, rdev->dev); |
114 | if (r) { |
134 | if (r) { |
115 | dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n", |
135 | dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n", |
116 | fw_name); |
136 | fw_name); |
117 | return r; |
137 | return r; |
118 | } |
138 | } |
119 | 139 | ||
120 | bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) + |
140 | bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) + |
121 | RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE + |
141 | RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE + |
122 | RADEON_GPU_PAGE_SIZE; |
142 | RADEON_GPU_PAGE_SIZE; |
123 | r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true, |
143 | r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true, |
124 | RADEON_GEM_DOMAIN_VRAM, 0, NULL, |
144 | RADEON_GEM_DOMAIN_VRAM, 0, NULL, |
125 | NULL, &rdev->uvd.vcpu_bo); |
145 | NULL, &rdev->uvd.vcpu_bo); |
126 | if (r) { |
146 | if (r) { |
127 | dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r); |
147 | dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r); |
128 | return r; |
148 | return r; |
129 | } |
149 | } |
130 | 150 | ||
131 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false); |
151 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false); |
132 | if (r) { |
152 | if (r) { |
133 | radeon_bo_unref(&rdev->uvd.vcpu_bo); |
153 | radeon_bo_unref(&rdev->uvd.vcpu_bo); |
134 | dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r); |
154 | dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r); |
135 | return r; |
155 | return r; |
136 | } |
156 | } |
137 | 157 | ||
138 | r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM, |
158 | r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM, |
139 | &rdev->uvd.gpu_addr); |
159 | &rdev->uvd.gpu_addr); |
140 | if (r) { |
160 | if (r) { |
141 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
161 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
142 | radeon_bo_unref(&rdev->uvd.vcpu_bo); |
162 | radeon_bo_unref(&rdev->uvd.vcpu_bo); |
143 | dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r); |
163 | dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r); |
144 | return r; |
164 | return r; |
145 | } |
165 | } |
146 | 166 | ||
147 | r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr); |
167 | r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr); |
148 | if (r) { |
168 | if (r) { |
149 | dev_err(rdev->dev, "(%d) UVD map failed\n", r); |
169 | dev_err(rdev->dev, "(%d) UVD map failed\n", r); |
150 | return r; |
170 | return r; |
151 | } |
171 | } |
152 | 172 | ||
153 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
173 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
154 | 174 | ||
155 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
175 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
156 | atomic_set(&rdev->uvd.handles[i], 0); |
176 | atomic_set(&rdev->uvd.handles[i], 0); |
157 | rdev->uvd.filp[i] = NULL; |
177 | rdev->uvd.filp[i] = NULL; |
158 | rdev->uvd.img_size[i] = 0; |
178 | rdev->uvd.img_size[i] = 0; |
159 | } |
179 | } |
160 | 180 | ||
161 | return 0; |
181 | return 0; |
162 | } |
182 | } |
163 | 183 | ||
164 | void radeon_uvd_fini(struct radeon_device *rdev) |
184 | void radeon_uvd_fini(struct radeon_device *rdev) |
165 | { |
185 | { |
166 | int r; |
186 | int r; |
167 | 187 | ||
168 | if (rdev->uvd.vcpu_bo == NULL) |
188 | if (rdev->uvd.vcpu_bo == NULL) |
169 | return; |
189 | return; |
170 | 190 | ||
171 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false); |
191 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false); |
172 | if (!r) { |
192 | if (!r) { |
173 | radeon_bo_kunmap(rdev->uvd.vcpu_bo); |
193 | radeon_bo_kunmap(rdev->uvd.vcpu_bo); |
174 | radeon_bo_unpin(rdev->uvd.vcpu_bo); |
194 | radeon_bo_unpin(rdev->uvd.vcpu_bo); |
175 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
195 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
176 | } |
196 | } |
177 | 197 | ||
178 | radeon_bo_unref(&rdev->uvd.vcpu_bo); |
198 | radeon_bo_unref(&rdev->uvd.vcpu_bo); |
179 | 199 | ||
180 | radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX]); |
200 | radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX]); |
181 | 201 | ||
182 | release_firmware(rdev->uvd_fw); |
202 | release_firmware(rdev->uvd_fw); |
183 | } |
203 | } |
184 | 204 | ||
185 | int radeon_uvd_suspend(struct radeon_device *rdev) |
205 | int radeon_uvd_suspend(struct radeon_device *rdev) |
186 | { |
206 | { |
187 | unsigned size; |
- | |
188 | void *ptr; |
- | |
189 | int i; |
207 | int i, r; |
190 | 208 | ||
191 | if (rdev->uvd.vcpu_bo == NULL) |
209 | if (rdev->uvd.vcpu_bo == NULL) |
192 | return 0; |
210 | return 0; |
193 | 211 | ||
194 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) |
212 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
195 | if (atomic_read(&rdev->uvd.handles[i])) |
213 | uint32_t handle = atomic_read(&rdev->uvd.handles[i]); |
- | 214 | if (handle != 0) { |
|
196 | break; |
215 | struct radeon_fence *fence; |
197 | - | ||
198 | if (i == RADEON_MAX_UVD_HANDLES) |
216 | |
- | 217 | radeon_uvd_note_usage(rdev); |
|
- | 218 | ||
199 | return 0; |
219 | r = radeon_uvd_get_destroy_msg(rdev, |
- | 220 | R600_RING_TYPE_UVD_INDEX, handle, &fence); |
|
- | 221 | if (r) { |
|
200 | 222 | DRM_ERROR("Error destroying UVD (%d)!\n", r); |
|
201 | size = radeon_bo_size(rdev->uvd.vcpu_bo); |
223 | continue; |
202 | size -= rdev->uvd_fw->size; |
224 | } |
203 | 225 | ||
- | 226 | radeon_fence_wait(fence, false); |
|
- | 227 | radeon_fence_unref(&fence); |
|
204 | ptr = rdev->uvd.cpu_addr; |
228 | |
205 | ptr += rdev->uvd_fw->size; |
229 | rdev->uvd.filp[i] = NULL; |
206 | 230 | atomic_set(&rdev->uvd.handles[i], 0); |
|
207 | rdev->uvd.saved_bo = kmalloc(size, GFP_KERNEL); |
231 | } |
208 | memcpy(rdev->uvd.saved_bo, ptr, size); |
232 | } |
209 | 233 | ||
210 | return 0; |
234 | return 0; |
211 | } |
235 | } |
212 | 236 | ||
213 | int radeon_uvd_resume(struct radeon_device *rdev) |
237 | int radeon_uvd_resume(struct radeon_device *rdev) |
214 | { |
238 | { |
215 | unsigned size; |
239 | unsigned size; |
216 | void *ptr; |
240 | void *ptr; |
217 | 241 | ||
218 | if (rdev->uvd.vcpu_bo == NULL) |
242 | if (rdev->uvd.vcpu_bo == NULL) |
219 | return -EINVAL; |
243 | return -EINVAL; |
220 | 244 | ||
221 | memcpy(rdev->uvd.cpu_addr, rdev->uvd_fw->data, rdev->uvd_fw->size); |
245 | memcpy(rdev->uvd.cpu_addr, rdev->uvd_fw->data, rdev->uvd_fw->size); |
222 | 246 | ||
223 | size = radeon_bo_size(rdev->uvd.vcpu_bo); |
247 | size = radeon_bo_size(rdev->uvd.vcpu_bo); |
224 | size -= rdev->uvd_fw->size; |
248 | size -= rdev->uvd_fw->size; |
225 | 249 | ||
226 | ptr = rdev->uvd.cpu_addr; |
250 | ptr = rdev->uvd.cpu_addr; |
227 | ptr += rdev->uvd_fw->size; |
251 | ptr += rdev->uvd_fw->size; |
228 | - | ||
229 | if (rdev->uvd.saved_bo != NULL) { |
- | |
230 | memcpy(ptr, rdev->uvd.saved_bo, size); |
- | |
231 | kfree(rdev->uvd.saved_bo); |
- | |
232 | rdev->uvd.saved_bo = NULL; |
- | |
233 | } else |
252 | |
234 | memset(ptr, 0, size); |
253 | memset(ptr, 0, size); |
235 | 254 | ||
236 | return 0; |
255 | return 0; |
237 | } |
256 | } |
238 | 257 | ||
239 | void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo, |
258 | void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo, |
240 | uint32_t allowed_domains) |
259 | uint32_t allowed_domains) |
241 | { |
260 | { |
242 | int i; |
261 | int i; |
243 | 262 | ||
244 | for (i = 0; i < rbo->placement.num_placement; ++i) { |
263 | for (i = 0; i < rbo->placement.num_placement; ++i) { |
245 | rbo->placements[i].fpfn = 0 >> PAGE_SHIFT; |
264 | rbo->placements[i].fpfn = 0 >> PAGE_SHIFT; |
246 | rbo->placements[i].lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT; |
265 | rbo->placements[i].lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT; |
247 | } |
266 | } |
248 | 267 | ||
249 | /* If it must be in VRAM it must be in the first segment as well */ |
268 | /* If it must be in VRAM it must be in the first segment as well */ |
250 | if (allowed_domains == RADEON_GEM_DOMAIN_VRAM) |
269 | if (allowed_domains == RADEON_GEM_DOMAIN_VRAM) |
251 | return; |
270 | return; |
252 | 271 | ||
253 | /* abort if we already have more than one placement */ |
272 | /* abort if we already have more than one placement */ |
254 | if (rbo->placement.num_placement > 1) |
273 | if (rbo->placement.num_placement > 1) |
255 | return; |
274 | return; |
256 | 275 | ||
257 | /* add another 256MB segment */ |
276 | /* add another 256MB segment */ |
258 | rbo->placements[1] = rbo->placements[0]; |
277 | rbo->placements[1] = rbo->placements[0]; |
259 | rbo->placements[1].fpfn += (256 * 1024 * 1024) >> PAGE_SHIFT; |
278 | rbo->placements[1].fpfn += (256 * 1024 * 1024) >> PAGE_SHIFT; |
260 | rbo->placements[1].lpfn += (256 * 1024 * 1024) >> PAGE_SHIFT; |
279 | rbo->placements[1].lpfn += (256 * 1024 * 1024) >> PAGE_SHIFT; |
261 | rbo->placement.num_placement++; |
280 | rbo->placement.num_placement++; |
262 | rbo->placement.num_busy_placement++; |
281 | rbo->placement.num_busy_placement++; |
263 | } |
282 | } |
264 | 283 | ||
265 | void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp) |
284 | void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp) |
266 | { |
285 | { |
267 | int i, r; |
286 | int i, r; |
268 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
287 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
269 | uint32_t handle = atomic_read(&rdev->uvd.handles[i]); |
288 | uint32_t handle = atomic_read(&rdev->uvd.handles[i]); |
270 | if (handle != 0 && rdev->uvd.filp[i] == filp) { |
289 | if (handle != 0 && rdev->uvd.filp[i] == filp) { |
271 | struct radeon_fence *fence; |
290 | struct radeon_fence *fence; |
272 | 291 | ||
273 | radeon_uvd_note_usage(rdev); |
292 | radeon_uvd_note_usage(rdev); |
274 | 293 | ||
275 | r = radeon_uvd_get_destroy_msg(rdev, |
294 | r = radeon_uvd_get_destroy_msg(rdev, |
276 | R600_RING_TYPE_UVD_INDEX, handle, &fence); |
295 | R600_RING_TYPE_UVD_INDEX, handle, &fence); |
277 | if (r) { |
296 | if (r) { |
278 | DRM_ERROR("Error destroying UVD (%d)!\n", r); |
297 | DRM_ERROR("Error destroying UVD (%d)!\n", r); |
279 | continue; |
298 | continue; |
280 | } |
299 | } |
281 | 300 | ||
282 | radeon_fence_wait(fence, false); |
301 | radeon_fence_wait(fence, false); |
283 | radeon_fence_unref(&fence); |
302 | radeon_fence_unref(&fence); |
284 | 303 | ||
285 | rdev->uvd.filp[i] = NULL; |
304 | rdev->uvd.filp[i] = NULL; |
286 | atomic_set(&rdev->uvd.handles[i], 0); |
305 | atomic_set(&rdev->uvd.handles[i], 0); |
287 | } |
306 | } |
288 | } |
307 | } |
289 | } |
308 | } |
290 | 309 | ||
291 | static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[]) |
310 | static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[]) |
292 | { |
311 | { |
293 | unsigned stream_type = msg[4]; |
312 | unsigned stream_type = msg[4]; |
294 | unsigned width = msg[6]; |
313 | unsigned width = msg[6]; |
295 | unsigned height = msg[7]; |
314 | unsigned height = msg[7]; |
296 | unsigned dpb_size = msg[9]; |
315 | unsigned dpb_size = msg[9]; |
297 | unsigned pitch = msg[28]; |
316 | unsigned pitch = msg[28]; |
298 | 317 | ||
299 | unsigned width_in_mb = width / 16; |
318 | unsigned width_in_mb = width / 16; |
300 | unsigned height_in_mb = ALIGN(height / 16, 2); |
319 | unsigned height_in_mb = ALIGN(height / 16, 2); |
301 | 320 | ||
302 | unsigned image_size, tmp, min_dpb_size; |
321 | unsigned image_size, tmp, min_dpb_size; |
303 | 322 | ||
304 | image_size = width * height; |
323 | image_size = width * height; |
305 | image_size += image_size / 2; |
324 | image_size += image_size / 2; |
306 | image_size = ALIGN(image_size, 1024); |
325 | image_size = ALIGN(image_size, 1024); |
307 | 326 | ||
308 | switch (stream_type) { |
327 | switch (stream_type) { |
309 | case 0: /* H264 */ |
328 | case 0: /* H264 */ |
310 | 329 | ||
311 | /* reference picture buffer */ |
330 | /* reference picture buffer */ |
312 | min_dpb_size = image_size * 17; |
331 | min_dpb_size = image_size * 17; |
313 | 332 | ||
314 | /* macroblock context buffer */ |
333 | /* macroblock context buffer */ |
315 | min_dpb_size += width_in_mb * height_in_mb * 17 * 192; |
334 | min_dpb_size += width_in_mb * height_in_mb * 17 * 192; |
316 | 335 | ||
317 | /* IT surface buffer */ |
336 | /* IT surface buffer */ |
318 | min_dpb_size += width_in_mb * height_in_mb * 32; |
337 | min_dpb_size += width_in_mb * height_in_mb * 32; |
319 | break; |
338 | break; |
320 | 339 | ||
321 | case 1: /* VC1 */ |
340 | case 1: /* VC1 */ |
322 | 341 | ||
323 | /* reference picture buffer */ |
342 | /* reference picture buffer */ |
324 | min_dpb_size = image_size * 3; |
343 | min_dpb_size = image_size * 3; |
325 | 344 | ||
326 | /* CONTEXT_BUFFER */ |
345 | /* CONTEXT_BUFFER */ |
327 | min_dpb_size += width_in_mb * height_in_mb * 128; |
346 | min_dpb_size += width_in_mb * height_in_mb * 128; |
328 | 347 | ||
329 | /* IT surface buffer */ |
348 | /* IT surface buffer */ |
330 | min_dpb_size += width_in_mb * 64; |
349 | min_dpb_size += width_in_mb * 64; |
331 | 350 | ||
332 | /* DB surface buffer */ |
351 | /* DB surface buffer */ |
333 | min_dpb_size += width_in_mb * 128; |
352 | min_dpb_size += width_in_mb * 128; |
334 | 353 | ||
335 | /* BP */ |
354 | /* BP */ |
336 | tmp = max(width_in_mb, height_in_mb); |
355 | tmp = max(width_in_mb, height_in_mb); |
337 | min_dpb_size += ALIGN(tmp * 7 * 16, 64); |
356 | min_dpb_size += ALIGN(tmp * 7 * 16, 64); |
338 | break; |
357 | break; |
339 | 358 | ||
340 | case 3: /* MPEG2 */ |
359 | case 3: /* MPEG2 */ |
341 | 360 | ||
342 | /* reference picture buffer */ |
361 | /* reference picture buffer */ |
343 | min_dpb_size = image_size * 3; |
362 | min_dpb_size = image_size * 3; |
344 | break; |
363 | break; |
345 | 364 | ||
346 | case 4: /* MPEG4 */ |
365 | case 4: /* MPEG4 */ |
347 | 366 | ||
348 | /* reference picture buffer */ |
367 | /* reference picture buffer */ |
349 | min_dpb_size = image_size * 3; |
368 | min_dpb_size = image_size * 3; |
350 | 369 | ||
351 | /* CM */ |
370 | /* CM */ |
352 | min_dpb_size += width_in_mb * height_in_mb * 64; |
371 | min_dpb_size += width_in_mb * height_in_mb * 64; |
353 | 372 | ||
354 | /* IT surface buffer */ |
373 | /* IT surface buffer */ |
355 | min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64); |
374 | min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64); |
356 | break; |
375 | break; |
357 | 376 | ||
358 | default: |
377 | default: |
359 | DRM_ERROR("UVD codec not handled %d!\n", stream_type); |
378 | DRM_ERROR("UVD codec not handled %d!\n", stream_type); |
360 | return -EINVAL; |
379 | return -EINVAL; |
361 | } |
380 | } |
362 | 381 | ||
363 | if (width > pitch) { |
382 | if (width > pitch) { |
364 | DRM_ERROR("Invalid UVD decoding target pitch!\n"); |
383 | DRM_ERROR("Invalid UVD decoding target pitch!\n"); |
365 | return -EINVAL; |
384 | return -EINVAL; |
366 | } |
385 | } |
367 | 386 | ||
368 | if (dpb_size < min_dpb_size) { |
387 | if (dpb_size < min_dpb_size) { |
369 | DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n", |
388 | DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n", |
370 | dpb_size, min_dpb_size); |
389 | dpb_size, min_dpb_size); |
371 | return -EINVAL; |
390 | return -EINVAL; |
372 | } |
391 | } |
373 | 392 | ||
374 | buf_sizes[0x1] = dpb_size; |
393 | buf_sizes[0x1] = dpb_size; |
375 | buf_sizes[0x2] = image_size; |
394 | buf_sizes[0x2] = image_size; |
376 | return 0; |
395 | return 0; |
377 | } |
396 | } |
- | 397 | ||
- | 398 | static int radeon_uvd_validate_codec(struct radeon_cs_parser *p, |
|
- | 399 | unsigned stream_type) |
|
- | 400 | { |
|
- | 401 | switch (stream_type) { |
|
- | 402 | case 0: /* H264 */ |
|
- | 403 | case 1: /* VC1 */ |
|
- | 404 | /* always supported */ |
|
- | 405 | return 0; |
|
- | 406 | ||
- | 407 | case 3: /* MPEG2 */ |
|
- | 408 | case 4: /* MPEG4 */ |
|
- | 409 | /* only since UVD 3 */ |
|
- | 410 | if (p->rdev->family >= CHIP_PALM) |
|
- | 411 | return 0; |
|
- | 412 | ||
- | 413 | /* fall through */ |
|
- | 414 | default: |
|
- | 415 | DRM_ERROR("UVD codec not supported by hardware %d!\n", |
|
- | 416 | stream_type); |
|
- | 417 | return -EINVAL; |
|
- | 418 | } |
|
- | 419 | } |
|
378 | 420 | ||
379 | static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo, |
421 | static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo, |
380 | unsigned offset, unsigned buf_sizes[]) |
422 | unsigned offset, unsigned buf_sizes[]) |
381 | { |
423 | { |
382 | int32_t *msg, msg_type, handle; |
424 | int32_t *msg, msg_type, handle; |
383 | unsigned img_size = 0; |
425 | unsigned img_size = 0; |
384 | struct fence *f; |
426 | struct fence *f; |
385 | void *ptr; |
427 | void *ptr; |
386 | 428 | ||
387 | int i, r; |
429 | int i, r; |
388 | 430 | ||
389 | if (offset & 0x3F) { |
431 | if (offset & 0x3F) { |
390 | DRM_ERROR("UVD messages must be 64 byte aligned!\n"); |
432 | DRM_ERROR("UVD messages must be 64 byte aligned!\n"); |
391 | return -EINVAL; |
433 | return -EINVAL; |
392 | } |
434 | } |
393 | 435 | ||
394 | f = reservation_object_get_excl(bo->tbo.resv); |
436 | f = reservation_object_get_excl(bo->tbo.resv); |
395 | if (f) { |
437 | if (f) { |
396 | r = radeon_fence_wait((struct radeon_fence *)f, false); |
438 | r = radeon_fence_wait((struct radeon_fence *)f, false); |
397 | if (r) { |
439 | if (r) { |
398 | DRM_ERROR("Failed waiting for UVD message (%d)!\n", r); |
440 | DRM_ERROR("Failed waiting for UVD message (%d)!\n", r); |
399 | return r; |
441 | return r; |
400 | } |
442 | } |
401 | } |
443 | } |
402 | 444 | ||
403 | r = radeon_bo_kmap(bo, &ptr); |
445 | r = radeon_bo_kmap(bo, &ptr); |
404 | if (r) { |
446 | if (r) { |
405 | DRM_ERROR("Failed mapping the UVD message (%d)!\n", r); |
447 | DRM_ERROR("Failed mapping the UVD message (%d)!\n", r); |
406 | return r; |
448 | return r; |
407 | } |
449 | } |
408 | 450 | ||
409 | msg = ptr + offset; |
451 | msg = ptr + offset; |
410 | 452 | ||
411 | msg_type = msg[1]; |
453 | msg_type = msg[1]; |
412 | handle = msg[2]; |
454 | handle = msg[2]; |
413 | 455 | ||
414 | if (handle == 0) { |
456 | if (handle == 0) { |
415 | DRM_ERROR("Invalid UVD handle!\n"); |
457 | DRM_ERROR("Invalid UVD handle!\n"); |
416 | return -EINVAL; |
458 | return -EINVAL; |
417 | } |
459 | } |
418 | 460 | ||
419 | if (msg_type == 1) { |
- | |
420 | /* it's a decode msg, calc buffer sizes */ |
- | |
421 | r = radeon_uvd_cs_msg_decode(msg, buf_sizes); |
- | |
422 | /* calc image size (width * height) */ |
- | |
423 | img_size = msg[6] * msg[7]; |
- | |
424 | radeon_bo_kunmap(bo); |
- | |
425 | if (r) |
- | |
426 | return r; |
- | |
427 | - | ||
428 | } else if (msg_type == 2) { |
- | |
429 | /* it's a destroy msg, free the handle */ |
- | |
430 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) |
- | |
431 | atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0); |
- | |
432 | radeon_bo_kunmap(bo); |
- | |
433 | return 0; |
461 | switch (msg_type) { |
434 | } else { |
462 | case 0: |
435 | /* it's a create msg, calc image size (width * height) */ |
463 | /* it's a create msg, calc image size (width * height) */ |
- | 464 | img_size = msg[7] * msg[8]; |
|
- | 465 | ||
436 | img_size = msg[7] * msg[8]; |
466 | r = radeon_uvd_validate_codec(p, msg[4]); |
- | 467 | radeon_bo_kunmap(bo); |
|
- | 468 | if (r) |
|
437 | radeon_bo_kunmap(bo); |
469 | return r; |
- | 470 | ||
- | 471 | /* try to alloc a new handle */ |
|
438 | 472 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
|
439 | if (msg_type != 0) { |
473 | if (atomic_read(&p->rdev->uvd.handles[i]) == handle) { |
440 | DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type); |
474 | DRM_ERROR("Handle 0x%x already in use!\n", handle); |
441 | return -EINVAL; |
475 | return -EINVAL; |
442 | } |
476 | } |
443 | 477 | ||
- | 478 | if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) { |
|
- | 479 | p->rdev->uvd.filp[i] = p->filp; |
|
- | 480 | p->rdev->uvd.img_size[i] = img_size; |
|
- | 481 | return 0; |
|
444 | /* it's a create msg, no special handling needed */ |
482 | } |
- | 483 | } |
|
- | 484 | ||
- | 485 | DRM_ERROR("No more free UVD handles!\n"); |
|
- | 486 | return -EINVAL; |
|
- | 487 | ||
- | 488 | case 1: |
|
- | 489 | /* it's a decode msg, validate codec and calc buffer sizes */ |
|
- | 490 | r = radeon_uvd_validate_codec(p, msg[4]); |
|
- | 491 | if (!r) |
|
- | 492 | r = radeon_uvd_cs_msg_decode(msg, buf_sizes); |
|
- | 493 | radeon_bo_kunmap(bo); |
|
- | 494 | if (r) |
|
445 | } |
495 | return r; |
446 | 496 | ||
447 | /* create or decode, validate the handle */ |
497 | /* validate the handle */ |
- | 498 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
|
- | 499 | if (atomic_read(&p->rdev->uvd.handles[i]) == handle) { |
|
- | 500 | if (p->rdev->uvd.filp[i] != p->filp) { |
|
- | 501 | DRM_ERROR("UVD handle collision detected!\n"); |
|
448 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
502 | return -EINVAL; |
449 | if (atomic_read(&p->rdev->uvd.handles[i]) == handle) |
503 | } |
- | 504 | return 0; |
|
- | 505 | } |
|
- | 506 | } |
|
- | 507 | ||
- | 508 | DRM_ERROR("Invalid UVD handle 0x%x!\n", handle); |
|
450 | return 0; |
509 | return -ENOENT; |
451 | } |
510 | |
452 | 511 | case 2: |
|
453 | /* handle not found try to alloc a new one */ |
512 | /* it's a destroy msg, free the handle */ |
454 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
- | |
455 | if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) { |
513 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) |
- | 514 | atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0); |
|
- | 515 | radeon_bo_kunmap(bo); |
|
456 | p->rdev->uvd.filp[i] = p->filp; |
516 | return 0; |
- | 517 | ||
- | 518 | default: |
|
457 | p->rdev->uvd.img_size[i] = img_size; |
519 | |
458 | return 0; |
520 | DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type); |
459 | } |
521 | return -EINVAL; |
460 | } |
522 | } |
461 | 523 | ||
462 | DRM_ERROR("No more free UVD handles!\n"); |
524 | BUG(); |
463 | return -EINVAL; |
525 | return -EINVAL; |
464 | } |
526 | } |
465 | 527 | ||
466 | static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p, |
528 | static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p, |
467 | int data0, int data1, |
529 | int data0, int data1, |
468 | unsigned buf_sizes[], bool *has_msg_cmd) |
530 | unsigned buf_sizes[], bool *has_msg_cmd) |
469 | { |
531 | { |
470 | struct radeon_cs_chunk *relocs_chunk; |
532 | struct radeon_cs_chunk *relocs_chunk; |
471 | struct radeon_bo_list *reloc; |
533 | struct radeon_bo_list *reloc; |
472 | unsigned idx, cmd, offset; |
534 | unsigned idx, cmd, offset; |
473 | uint64_t start, end; |
535 | uint64_t start, end; |
474 | int r; |
536 | int r; |
475 | 537 | ||
476 | relocs_chunk = p->chunk_relocs; |
538 | relocs_chunk = p->chunk_relocs; |
477 | offset = radeon_get_ib_value(p, data0); |
539 | offset = radeon_get_ib_value(p, data0); |
478 | idx = radeon_get_ib_value(p, data1); |
540 | idx = radeon_get_ib_value(p, data1); |
479 | if (idx >= relocs_chunk->length_dw) { |
541 | if (idx >= relocs_chunk->length_dw) { |
480 | DRM_ERROR("Relocs at %d after relocations chunk end %d !\n", |
542 | DRM_ERROR("Relocs at %d after relocations chunk end %d !\n", |
481 | idx, relocs_chunk->length_dw); |
543 | idx, relocs_chunk->length_dw); |
482 | return -EINVAL; |
544 | return -EINVAL; |
483 | } |
545 | } |
484 | 546 | ||
485 | reloc = &p->relocs[(idx / 4)]; |
547 | reloc = &p->relocs[(idx / 4)]; |
486 | start = reloc->gpu_offset; |
548 | start = reloc->gpu_offset; |
487 | end = start + radeon_bo_size(reloc->robj); |
549 | end = start + radeon_bo_size(reloc->robj); |
488 | start += offset; |
550 | start += offset; |
489 | 551 | ||
490 | p->ib.ptr[data0] = start & 0xFFFFFFFF; |
552 | p->ib.ptr[data0] = start & 0xFFFFFFFF; |
491 | p->ib.ptr[data1] = start >> 32; |
553 | p->ib.ptr[data1] = start >> 32; |
492 | 554 | ||
493 | cmd = radeon_get_ib_value(p, p->idx) >> 1; |
555 | cmd = radeon_get_ib_value(p, p->idx) >> 1; |
494 | 556 | ||
495 | if (cmd < 0x4) { |
557 | if (cmd < 0x4) { |
496 | if (end <= start) { |
558 | if (end <= start) { |
497 | DRM_ERROR("invalid reloc offset %X!\n", offset); |
559 | DRM_ERROR("invalid reloc offset %X!\n", offset); |
498 | return -EINVAL; |
560 | return -EINVAL; |
499 | } |
561 | } |
500 | if ((end - start) < buf_sizes[cmd]) { |
562 | if ((end - start) < buf_sizes[cmd]) { |
501 | DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd, |
563 | DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd, |
502 | (unsigned)(end - start), buf_sizes[cmd]); |
564 | (unsigned)(end - start), buf_sizes[cmd]); |
503 | return -EINVAL; |
565 | return -EINVAL; |
504 | } |
566 | } |
505 | 567 | ||
506 | } else if (cmd != 0x100) { |
568 | } else if (cmd != 0x100) { |
507 | DRM_ERROR("invalid UVD command %X!\n", cmd); |
569 | DRM_ERROR("invalid UVD command %X!\n", cmd); |
508 | return -EINVAL; |
570 | return -EINVAL; |
509 | } |
571 | } |
510 | 572 | ||
511 | if ((start >> 28) != ((end - 1) >> 28)) { |
573 | if ((start >> 28) != ((end - 1) >> 28)) { |
512 | DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n", |
574 | DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n", |
513 | start, end); |
575 | start, end); |
514 | return -EINVAL; |
576 | return -EINVAL; |
515 | } |
577 | } |
516 | 578 | ||
517 | /* TODO: is this still necessary on NI+ ? */ |
579 | /* TODO: is this still necessary on NI+ ? */ |
518 | if ((cmd == 0 || cmd == 0x3) && |
580 | if ((cmd == 0 || cmd == 0x3) && |
519 | (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) { |
581 | (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) { |
520 | DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n", |
582 | DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n", |
521 | start, end); |
583 | start, end); |
522 | return -EINVAL; |
584 | return -EINVAL; |
523 | } |
585 | } |
524 | 586 | ||
525 | if (cmd == 0) { |
587 | if (cmd == 0) { |
526 | if (*has_msg_cmd) { |
588 | if (*has_msg_cmd) { |
527 | DRM_ERROR("More than one message in a UVD-IB!\n"); |
589 | DRM_ERROR("More than one message in a UVD-IB!\n"); |
528 | return -EINVAL; |
590 | return -EINVAL; |
529 | } |
591 | } |
530 | *has_msg_cmd = true; |
592 | *has_msg_cmd = true; |
531 | r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes); |
593 | r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes); |
532 | if (r) |
594 | if (r) |
533 | return r; |
595 | return r; |
534 | } else if (!*has_msg_cmd) { |
596 | } else if (!*has_msg_cmd) { |
535 | DRM_ERROR("Message needed before other commands are send!\n"); |
597 | DRM_ERROR("Message needed before other commands are send!\n"); |
536 | return -EINVAL; |
598 | return -EINVAL; |
537 | } |
599 | } |
538 | 600 | ||
539 | return 0; |
601 | return 0; |
540 | } |
602 | } |
541 | 603 | ||
542 | static int radeon_uvd_cs_reg(struct radeon_cs_parser *p, |
604 | static int radeon_uvd_cs_reg(struct radeon_cs_parser *p, |
543 | struct radeon_cs_packet *pkt, |
605 | struct radeon_cs_packet *pkt, |
544 | int *data0, int *data1, |
606 | int *data0, int *data1, |
545 | unsigned buf_sizes[], |
607 | unsigned buf_sizes[], |
546 | bool *has_msg_cmd) |
608 | bool *has_msg_cmd) |
547 | { |
609 | { |
548 | int i, r; |
610 | int i, r; |
549 | 611 | ||
550 | p->idx++; |
612 | p->idx++; |
551 | for (i = 0; i <= pkt->count; ++i) { |
613 | for (i = 0; i <= pkt->count; ++i) { |
552 | switch (pkt->reg + i*4) { |
614 | switch (pkt->reg + i*4) { |
553 | case UVD_GPCOM_VCPU_DATA0: |
615 | case UVD_GPCOM_VCPU_DATA0: |
554 | *data0 = p->idx; |
616 | *data0 = p->idx; |
555 | break; |
617 | break; |
556 | case UVD_GPCOM_VCPU_DATA1: |
618 | case UVD_GPCOM_VCPU_DATA1: |
557 | *data1 = p->idx; |
619 | *data1 = p->idx; |
558 | break; |
620 | break; |
559 | case UVD_GPCOM_VCPU_CMD: |
621 | case UVD_GPCOM_VCPU_CMD: |
560 | r = radeon_uvd_cs_reloc(p, *data0, *data1, |
622 | r = radeon_uvd_cs_reloc(p, *data0, *data1, |
561 | buf_sizes, has_msg_cmd); |
623 | buf_sizes, has_msg_cmd); |
562 | if (r) |
624 | if (r) |
563 | return r; |
625 | return r; |
564 | break; |
626 | break; |
565 | case UVD_ENGINE_CNTL: |
627 | case UVD_ENGINE_CNTL: |
566 | break; |
628 | break; |
567 | default: |
629 | default: |
568 | DRM_ERROR("Invalid reg 0x%X!\n", |
630 | DRM_ERROR("Invalid reg 0x%X!\n", |
569 | pkt->reg + i*4); |
631 | pkt->reg + i*4); |
570 | return -EINVAL; |
632 | return -EINVAL; |
571 | } |
633 | } |
572 | p->idx++; |
634 | p->idx++; |
573 | } |
635 | } |
574 | return 0; |
636 | return 0; |
575 | } |
637 | } |
576 | 638 | ||
577 | int radeon_uvd_cs_parse(struct radeon_cs_parser *p) |
639 | int radeon_uvd_cs_parse(struct radeon_cs_parser *p) |
578 | { |
640 | { |
579 | struct radeon_cs_packet pkt; |
641 | struct radeon_cs_packet pkt; |
580 | int r, data0 = 0, data1 = 0; |
642 | int r, data0 = 0, data1 = 0; |
581 | 643 | ||
582 | /* does the IB has a msg command */ |
644 | /* does the IB has a msg command */ |
583 | bool has_msg_cmd = false; |
645 | bool has_msg_cmd = false; |
584 | 646 | ||
585 | /* minimum buffer sizes */ |
647 | /* minimum buffer sizes */ |
586 | unsigned buf_sizes[] = { |
648 | unsigned buf_sizes[] = { |
587 | [0x00000000] = 2048, |
649 | [0x00000000] = 2048, |
588 | [0x00000001] = 32 * 1024 * 1024, |
650 | [0x00000001] = 32 * 1024 * 1024, |
589 | [0x00000002] = 2048 * 1152 * 3, |
651 | [0x00000002] = 2048 * 1152 * 3, |
590 | [0x00000003] = 2048, |
652 | [0x00000003] = 2048, |
591 | }; |
653 | }; |
592 | 654 | ||
593 | if (p->chunk_ib->length_dw % 16) { |
655 | if (p->chunk_ib->length_dw % 16) { |
594 | DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n", |
656 | DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n", |
595 | p->chunk_ib->length_dw); |
657 | p->chunk_ib->length_dw); |
596 | return -EINVAL; |
658 | return -EINVAL; |
597 | } |
659 | } |
598 | 660 | ||
599 | if (p->chunk_relocs == NULL) { |
661 | if (p->chunk_relocs == NULL) { |
600 | DRM_ERROR("No relocation chunk !\n"); |
662 | DRM_ERROR("No relocation chunk !\n"); |
601 | return -EINVAL; |
663 | return -EINVAL; |
602 | } |
664 | } |
603 | 665 | ||
604 | 666 | ||
605 | do { |
667 | do { |
606 | r = radeon_cs_packet_parse(p, &pkt, p->idx); |
668 | r = radeon_cs_packet_parse(p, &pkt, p->idx); |
607 | if (r) |
669 | if (r) |
608 | return r; |
670 | return r; |
609 | switch (pkt.type) { |
671 | switch (pkt.type) { |
610 | case RADEON_PACKET_TYPE0: |
672 | case RADEON_PACKET_TYPE0: |
611 | r = radeon_uvd_cs_reg(p, &pkt, &data0, &data1, |
673 | r = radeon_uvd_cs_reg(p, &pkt, &data0, &data1, |
612 | buf_sizes, &has_msg_cmd); |
674 | buf_sizes, &has_msg_cmd); |
613 | if (r) |
675 | if (r) |
614 | return r; |
676 | return r; |
615 | break; |
677 | break; |
616 | case RADEON_PACKET_TYPE2: |
678 | case RADEON_PACKET_TYPE2: |
617 | p->idx += pkt.count + 2; |
679 | p->idx += pkt.count + 2; |
618 | break; |
680 | break; |
619 | default: |
681 | default: |
620 | DRM_ERROR("Unknown packet type %d !\n", pkt.type); |
682 | DRM_ERROR("Unknown packet type %d !\n", pkt.type); |
621 | return -EINVAL; |
683 | return -EINVAL; |
622 | } |
684 | } |
623 | } while (p->idx < p->chunk_ib->length_dw); |
685 | } while (p->idx < p->chunk_ib->length_dw); |
624 | 686 | ||
625 | if (!has_msg_cmd) { |
687 | if (!has_msg_cmd) { |
626 | DRM_ERROR("UVD-IBs need a msg command!\n"); |
688 | DRM_ERROR("UVD-IBs need a msg command!\n"); |
627 | return -EINVAL; |
689 | return -EINVAL; |
628 | } |
690 | } |
629 | 691 | ||
630 | return 0; |
692 | return 0; |
631 | } |
693 | } |
632 | 694 | ||
633 | static int radeon_uvd_send_msg(struct radeon_device *rdev, |
695 | static int radeon_uvd_send_msg(struct radeon_device *rdev, |
634 | int ring, uint64_t addr, |
696 | int ring, uint64_t addr, |
635 | struct radeon_fence **fence) |
697 | struct radeon_fence **fence) |
636 | { |
698 | { |
637 | struct radeon_ib ib; |
699 | struct radeon_ib ib; |
638 | int i, r; |
700 | int i, r; |
639 | 701 | ||
640 | r = radeon_ib_get(rdev, ring, &ib, NULL, 64); |
702 | r = radeon_ib_get(rdev, ring, &ib, NULL, 64); |
641 | if (r) |
703 | if (r) |
642 | return r; |
704 | return r; |
643 | 705 | ||
644 | ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0); |
706 | ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0); |
645 | ib.ptr[1] = addr; |
707 | ib.ptr[1] = addr; |
646 | ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0); |
708 | ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0); |
647 | ib.ptr[3] = addr >> 32; |
709 | ib.ptr[3] = addr >> 32; |
648 | ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0); |
710 | ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0); |
649 | ib.ptr[5] = 0; |
711 | ib.ptr[5] = 0; |
650 | for (i = 6; i < 16; ++i) |
712 | for (i = 6; i < 16; ++i) |
651 | ib.ptr[i] = PACKET2(0); |
713 | ib.ptr[i] = PACKET2(0); |
652 | ib.length_dw = 16; |
714 | ib.length_dw = 16; |
653 | 715 | ||
654 | r = radeon_ib_schedule(rdev, &ib, NULL, false); |
716 | r = radeon_ib_schedule(rdev, &ib, NULL, false); |
655 | 717 | ||
656 | if (fence) |
718 | if (fence) |
657 | *fence = radeon_fence_ref(ib.fence); |
719 | *fence = radeon_fence_ref(ib.fence); |
658 | 720 | ||
659 | radeon_ib_free(rdev, &ib); |
721 | radeon_ib_free(rdev, &ib); |
660 | return r; |
722 | return r; |
661 | } |
723 | } |
662 | 724 | ||
663 | /* multiple fence commands without any stream commands in between can |
725 | /* multiple fence commands without any stream commands in between can |
664 | crash the vcpu so just try to emmit a dummy create/destroy msg to |
726 | crash the vcpu so just try to emmit a dummy create/destroy msg to |
665 | avoid this */ |
727 | avoid this */ |
666 | int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring, |
728 | int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring, |
667 | uint32_t handle, struct radeon_fence **fence) |
729 | uint32_t handle, struct radeon_fence **fence) |
668 | { |
730 | { |
669 | /* we use the last page of the vcpu bo for the UVD message */ |
731 | /* we use the last page of the vcpu bo for the UVD message */ |
670 | uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) - |
732 | uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) - |
671 | RADEON_GPU_PAGE_SIZE; |
733 | RADEON_GPU_PAGE_SIZE; |
672 | 734 | ||
673 | uint32_t *msg = rdev->uvd.cpu_addr + offs; |
735 | uint32_t *msg = rdev->uvd.cpu_addr + offs; |
674 | uint64_t addr = rdev->uvd.gpu_addr + offs; |
736 | uint64_t addr = rdev->uvd.gpu_addr + offs; |
675 | 737 | ||
676 | int r, i; |
738 | int r, i; |
677 | 739 | ||
678 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true); |
740 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true); |
679 | if (r) |
741 | if (r) |
680 | return r; |
742 | return r; |
681 | 743 | ||
682 | /* stitch together an UVD create msg */ |
744 | /* stitch together an UVD create msg */ |
683 | msg[0] = cpu_to_le32(0x00000de4); |
745 | msg[0] = cpu_to_le32(0x00000de4); |
684 | msg[1] = cpu_to_le32(0x00000000); |
746 | msg[1] = cpu_to_le32(0x00000000); |
685 | msg[2] = cpu_to_le32(handle); |
747 | msg[2] = cpu_to_le32(handle); |
686 | msg[3] = cpu_to_le32(0x00000000); |
748 | msg[3] = cpu_to_le32(0x00000000); |
687 | msg[4] = cpu_to_le32(0x00000000); |
749 | msg[4] = cpu_to_le32(0x00000000); |
688 | msg[5] = cpu_to_le32(0x00000000); |
750 | msg[5] = cpu_to_le32(0x00000000); |
689 | msg[6] = cpu_to_le32(0x00000000); |
751 | msg[6] = cpu_to_le32(0x00000000); |
690 | msg[7] = cpu_to_le32(0x00000780); |
752 | msg[7] = cpu_to_le32(0x00000780); |
691 | msg[8] = cpu_to_le32(0x00000440); |
753 | msg[8] = cpu_to_le32(0x00000440); |
692 | msg[9] = cpu_to_le32(0x00000000); |
754 | msg[9] = cpu_to_le32(0x00000000); |
693 | msg[10] = cpu_to_le32(0x01b37000); |
755 | msg[10] = cpu_to_le32(0x01b37000); |
694 | for (i = 11; i < 1024; ++i) |
756 | for (i = 11; i < 1024; ++i) |
695 | msg[i] = cpu_to_le32(0x0); |
757 | msg[i] = cpu_to_le32(0x0); |
696 | 758 | ||
697 | r = radeon_uvd_send_msg(rdev, ring, addr, fence); |
759 | r = radeon_uvd_send_msg(rdev, ring, addr, fence); |
698 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
760 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
699 | return r; |
761 | return r; |
700 | } |
762 | } |
701 | 763 | ||
702 | int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring, |
764 | int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring, |
703 | uint32_t handle, struct radeon_fence **fence) |
765 | uint32_t handle, struct radeon_fence **fence) |
704 | { |
766 | { |
705 | /* we use the last page of the vcpu bo for the UVD message */ |
767 | /* we use the last page of the vcpu bo for the UVD message */ |
706 | uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) - |
768 | uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) - |
707 | RADEON_GPU_PAGE_SIZE; |
769 | RADEON_GPU_PAGE_SIZE; |
708 | 770 | ||
709 | uint32_t *msg = rdev->uvd.cpu_addr + offs; |
771 | uint32_t *msg = rdev->uvd.cpu_addr + offs; |
710 | uint64_t addr = rdev->uvd.gpu_addr + offs; |
772 | uint64_t addr = rdev->uvd.gpu_addr + offs; |
711 | 773 | ||
712 | int r, i; |
774 | int r, i; |
713 | 775 | ||
714 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true); |
776 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true); |
715 | if (r) |
777 | if (r) |
716 | return r; |
778 | return r; |
717 | 779 | ||
718 | /* stitch together an UVD destroy msg */ |
780 | /* stitch together an UVD destroy msg */ |
719 | msg[0] = cpu_to_le32(0x00000de4); |
781 | msg[0] = cpu_to_le32(0x00000de4); |
720 | msg[1] = cpu_to_le32(0x00000002); |
782 | msg[1] = cpu_to_le32(0x00000002); |
721 | msg[2] = cpu_to_le32(handle); |
783 | msg[2] = cpu_to_le32(handle); |
722 | msg[3] = cpu_to_le32(0x00000000); |
784 | msg[3] = cpu_to_le32(0x00000000); |
723 | for (i = 4; i < 1024; ++i) |
785 | for (i = 4; i < 1024; ++i) |
724 | msg[i] = cpu_to_le32(0x0); |
786 | msg[i] = cpu_to_le32(0x0); |
725 | 787 | ||
726 | r = radeon_uvd_send_msg(rdev, ring, addr, fence); |
788 | r = radeon_uvd_send_msg(rdev, ring, addr, fence); |
727 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
789 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); |
728 | return r; |
790 | return r; |
729 | } |
791 | } |
730 | 792 | ||
731 | /** |
793 | /** |
732 | * radeon_uvd_count_handles - count number of open streams |
794 | * radeon_uvd_count_handles - count number of open streams |
733 | * |
795 | * |
734 | * @rdev: radeon_device pointer |
796 | * @rdev: radeon_device pointer |
735 | * @sd: number of SD streams |
797 | * @sd: number of SD streams |
736 | * @hd: number of HD streams |
798 | * @hd: number of HD streams |
737 | * |
799 | * |
738 | * Count the number of open SD/HD streams as a hint for power mangement |
800 | * Count the number of open SD/HD streams as a hint for power mangement |
739 | */ |
801 | */ |
740 | static void radeon_uvd_count_handles(struct radeon_device *rdev, |
802 | static void radeon_uvd_count_handles(struct radeon_device *rdev, |
741 | unsigned *sd, unsigned *hd) |
803 | unsigned *sd, unsigned *hd) |
742 | { |
804 | { |
743 | unsigned i; |
805 | unsigned i; |
744 | 806 | ||
745 | *sd = 0; |
807 | *sd = 0; |
746 | *hd = 0; |
808 | *hd = 0; |
747 | 809 | ||
748 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
810 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { |
749 | if (!atomic_read(&rdev->uvd.handles[i])) |
811 | if (!atomic_read(&rdev->uvd.handles[i])) |
750 | continue; |
812 | continue; |
751 | 813 | ||
752 | if (rdev->uvd.img_size[i] >= 720*576) |
814 | if (rdev->uvd.img_size[i] >= 720*576) |
753 | ++(*hd); |
815 | ++(*hd); |
754 | else |
816 | else |
755 | ++(*sd); |
817 | ++(*sd); |
756 | } |
818 | } |
757 | } |
819 | } |
758 | 820 | ||
759 | static void radeon_uvd_idle_work_handler(struct work_struct *work) |
821 | static void radeon_uvd_idle_work_handler(struct work_struct *work) |
760 | { |
822 | { |
761 | struct radeon_device *rdev = |
823 | struct radeon_device *rdev = |
762 | container_of(work, struct radeon_device, uvd.idle_work.work); |
824 | container_of(work, struct radeon_device, uvd.idle_work.work); |
763 | 825 | ||
764 | if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) { |
826 | if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) { |
765 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
827 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
766 | radeon_uvd_count_handles(rdev, &rdev->pm.dpm.sd, |
828 | radeon_uvd_count_handles(rdev, &rdev->pm.dpm.sd, |
767 | &rdev->pm.dpm.hd); |
829 | &rdev->pm.dpm.hd); |
768 | radeon_dpm_enable_uvd(rdev, false); |
830 | radeon_dpm_enable_uvd(rdev, false); |
769 | } else { |
831 | } else { |
770 | radeon_set_uvd_clocks(rdev, 0, 0); |
832 | radeon_set_uvd_clocks(rdev, 0, 0); |
771 | } |
833 | } |
772 | } else { |
834 | } else { |
773 | schedule_delayed_work(&rdev->uvd.idle_work, |
835 | schedule_delayed_work(&rdev->uvd.idle_work, |
774 | msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS)); |
836 | msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS)); |
775 | } |
837 | } |
776 | } |
838 | } |
777 | 839 | ||
778 | void radeon_uvd_note_usage(struct radeon_device *rdev) |
840 | void radeon_uvd_note_usage(struct radeon_device *rdev) |
779 | { |
841 | { |
780 | bool streams_changed = false; |
842 | bool streams_changed = false; |
781 | bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work); |
843 | bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work); |
782 | set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work, |
844 | set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work, |
783 | msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS)); |
845 | msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS)); |
784 | 846 | ||
785 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
847 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
786 | unsigned hd = 0, sd = 0; |
848 | unsigned hd = 0, sd = 0; |
787 | radeon_uvd_count_handles(rdev, &sd, &hd); |
849 | radeon_uvd_count_handles(rdev, &sd, &hd); |
788 | if ((rdev->pm.dpm.sd != sd) || |
850 | if ((rdev->pm.dpm.sd != sd) || |
789 | (rdev->pm.dpm.hd != hd)) { |
851 | (rdev->pm.dpm.hd != hd)) { |
790 | rdev->pm.dpm.sd = sd; |
852 | rdev->pm.dpm.sd = sd; |
791 | rdev->pm.dpm.hd = hd; |
853 | rdev->pm.dpm.hd = hd; |
792 | /* disable this for now */ |
854 | /* disable this for now */ |
793 | /*streams_changed = true;*/ |
855 | /*streams_changed = true;*/ |
794 | } |
856 | } |
795 | } |
857 | } |
796 | 858 | ||
797 | if (set_clocks || streams_changed) { |
859 | if (set_clocks || streams_changed) { |
798 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
860 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
799 | radeon_dpm_enable_uvd(rdev, true); |
861 | radeon_dpm_enable_uvd(rdev, true); |
800 | } else { |
862 | } else { |
801 | radeon_set_uvd_clocks(rdev, 53300, 40000); |
863 | radeon_set_uvd_clocks(rdev, 53300, 40000); |
802 | } |
864 | } |
803 | } |
865 | } |
804 | } |
866 | } |
805 | 867 | ||
806 | static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq, |
868 | static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq, |
807 | unsigned target_freq, |
869 | unsigned target_freq, |
808 | unsigned pd_min, |
870 | unsigned pd_min, |
809 | unsigned pd_even) |
871 | unsigned pd_even) |
810 | { |
872 | { |
811 | unsigned post_div = vco_freq / target_freq; |
873 | unsigned post_div = vco_freq / target_freq; |
812 | 874 | ||
813 | /* adjust to post divider minimum value */ |
875 | /* adjust to post divider minimum value */ |
814 | if (post_div < pd_min) |
876 | if (post_div < pd_min) |
815 | post_div = pd_min; |
877 | post_div = pd_min; |
816 | 878 | ||
817 | /* we alway need a frequency less than or equal the target */ |
879 | /* we alway need a frequency less than or equal the target */ |
818 | if ((vco_freq / post_div) > target_freq) |
880 | if ((vco_freq / post_div) > target_freq) |
819 | post_div += 1; |
881 | post_div += 1; |
820 | 882 | ||
821 | /* post dividers above a certain value must be even */ |
883 | /* post dividers above a certain value must be even */ |
822 | if (post_div > pd_even && post_div % 2) |
884 | if (post_div > pd_even && post_div % 2) |
823 | post_div += 1; |
885 | post_div += 1; |
824 | 886 | ||
825 | return post_div; |
887 | return post_div; |
826 | } |
888 | } |
827 | 889 | ||
828 | /** |
890 | /** |
829 | * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers |
891 | * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers |
830 | * |
892 | * |
831 | * @rdev: radeon_device pointer |
893 | * @rdev: radeon_device pointer |
832 | * @vclk: wanted VCLK |
894 | * @vclk: wanted VCLK |
833 | * @dclk: wanted DCLK |
895 | * @dclk: wanted DCLK |
834 | * @vco_min: minimum VCO frequency |
896 | * @vco_min: minimum VCO frequency |
835 | * @vco_max: maximum VCO frequency |
897 | * @vco_max: maximum VCO frequency |
836 | * @fb_factor: factor to multiply vco freq with |
898 | * @fb_factor: factor to multiply vco freq with |
837 | * @fb_mask: limit and bitmask for feedback divider |
899 | * @fb_mask: limit and bitmask for feedback divider |
838 | * @pd_min: post divider minimum |
900 | * @pd_min: post divider minimum |
839 | * @pd_max: post divider maximum |
901 | * @pd_max: post divider maximum |
840 | * @pd_even: post divider must be even above this value |
902 | * @pd_even: post divider must be even above this value |
841 | * @optimal_fb_div: resulting feedback divider |
903 | * @optimal_fb_div: resulting feedback divider |
842 | * @optimal_vclk_div: resulting vclk post divider |
904 | * @optimal_vclk_div: resulting vclk post divider |
843 | * @optimal_dclk_div: resulting dclk post divider |
905 | * @optimal_dclk_div: resulting dclk post divider |
844 | * |
906 | * |
845 | * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs). |
907 | * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs). |
846 | * Returns zero on success -EINVAL on error. |
908 | * Returns zero on success -EINVAL on error. |
847 | */ |
909 | */ |
848 | int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev, |
910 | int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev, |
849 | unsigned vclk, unsigned dclk, |
911 | unsigned vclk, unsigned dclk, |
850 | unsigned vco_min, unsigned vco_max, |
912 | unsigned vco_min, unsigned vco_max, |
851 | unsigned fb_factor, unsigned fb_mask, |
913 | unsigned fb_factor, unsigned fb_mask, |
852 | unsigned pd_min, unsigned pd_max, |
914 | unsigned pd_min, unsigned pd_max, |
853 | unsigned pd_even, |
915 | unsigned pd_even, |
854 | unsigned *optimal_fb_div, |
916 | unsigned *optimal_fb_div, |
855 | unsigned *optimal_vclk_div, |
917 | unsigned *optimal_vclk_div, |
856 | unsigned *optimal_dclk_div) |
918 | unsigned *optimal_dclk_div) |
857 | { |
919 | { |
858 | unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq; |
920 | unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq; |
859 | 921 | ||
860 | /* start off with something large */ |
922 | /* start off with something large */ |
861 | unsigned optimal_score = ~0; |
923 | unsigned optimal_score = ~0; |
862 | 924 | ||
863 | /* loop through vco from low to high */ |
925 | /* loop through vco from low to high */ |
864 | vco_min = max(max(vco_min, vclk), dclk); |
926 | vco_min = max(max(vco_min, vclk), dclk); |
865 | for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) { |
927 | for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) { |
866 | 928 | ||
867 | uint64_t fb_div = (uint64_t)vco_freq * fb_factor; |
929 | uint64_t fb_div = (uint64_t)vco_freq * fb_factor; |
868 | unsigned vclk_div, dclk_div, score; |
930 | unsigned vclk_div, dclk_div, score; |
869 | 931 | ||
870 | do_div(fb_div, ref_freq); |
932 | do_div(fb_div, ref_freq); |
871 | 933 | ||
872 | /* fb div out of range ? */ |
934 | /* fb div out of range ? */ |
873 | if (fb_div > fb_mask) |
935 | if (fb_div > fb_mask) |
874 | break; /* it can oly get worse */ |
936 | break; /* it can oly get worse */ |
875 | 937 | ||
876 | fb_div &= fb_mask; |
938 | fb_div &= fb_mask; |
877 | 939 | ||
878 | /* calc vclk divider with current vco freq */ |
940 | /* calc vclk divider with current vco freq */ |
879 | vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk, |
941 | vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk, |
880 | pd_min, pd_even); |
942 | pd_min, pd_even); |
881 | if (vclk_div > pd_max) |
943 | if (vclk_div > pd_max) |
882 | break; /* vco is too big, it has to stop */ |
944 | break; /* vco is too big, it has to stop */ |
883 | 945 | ||
884 | /* calc dclk divider with current vco freq */ |
946 | /* calc dclk divider with current vco freq */ |
885 | dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk, |
947 | dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk, |
886 | pd_min, pd_even); |
948 | pd_min, pd_even); |
887 | if (vclk_div > pd_max) |
949 | if (vclk_div > pd_max) |
888 | break; /* vco is too big, it has to stop */ |
950 | break; /* vco is too big, it has to stop */ |
889 | 951 | ||
890 | /* calc score with current vco freq */ |
952 | /* calc score with current vco freq */ |
891 | score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div); |
953 | score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div); |
892 | 954 | ||
893 | /* determine if this vco setting is better than current optimal settings */ |
955 | /* determine if this vco setting is better than current optimal settings */ |
894 | if (score < optimal_score) { |
956 | if (score < optimal_score) { |
895 | *optimal_fb_div = fb_div; |
957 | *optimal_fb_div = fb_div; |
896 | *optimal_vclk_div = vclk_div; |
958 | *optimal_vclk_div = vclk_div; |
897 | *optimal_dclk_div = dclk_div; |
959 | *optimal_dclk_div = dclk_div; |
898 | optimal_score = score; |
960 | optimal_score = score; |
899 | if (optimal_score == 0) |
961 | if (optimal_score == 0) |
900 | break; /* it can't get better than this */ |
962 | break; /* it can't get better than this */ |
901 | } |
963 | } |
902 | } |
964 | } |
903 | 965 | ||
904 | /* did we found a valid setup ? */ |
966 | /* did we found a valid setup ? */ |
905 | if (optimal_score == ~0) |
967 | if (optimal_score == ~0) |
906 | return -EINVAL; |
968 | return -EINVAL; |
907 | 969 | ||
908 | return 0; |
970 | return 0; |
909 | } |
971 | } |
910 | 972 | ||
911 | int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev, |
973 | int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev, |
912 | unsigned cg_upll_func_cntl) |
974 | unsigned cg_upll_func_cntl) |
913 | { |
975 | { |
914 | unsigned i; |
976 | unsigned i; |
915 | 977 | ||
916 | /* make sure UPLL_CTLREQ is deasserted */ |
978 | /* make sure UPLL_CTLREQ is deasserted */ |
917 | WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK); |
979 | WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK); |
918 | 980 | ||
919 | mdelay(10); |
981 | mdelay(10); |
920 | 982 | ||
921 | /* assert UPLL_CTLREQ */ |
983 | /* assert UPLL_CTLREQ */ |
922 | WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK); |
984 | WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK); |
923 | 985 | ||
924 | /* wait for CTLACK and CTLACK2 to get asserted */ |
986 | /* wait for CTLACK and CTLACK2 to get asserted */ |
925 | for (i = 0; i < 100; ++i) { |
987 | for (i = 0; i < 100; ++i) { |
926 | uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK; |
988 | uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK; |
927 | if ((RREG32(cg_upll_func_cntl) & mask) == mask) |
989 | if ((RREG32(cg_upll_func_cntl) & mask) == mask) |
928 | break; |
990 | break; |
929 | mdelay(10); |
991 | mdelay(10); |
930 | } |
992 | } |
931 | 993 | ||
932 | /* deassert UPLL_CTLREQ */ |
994 | /* deassert UPLL_CTLREQ */ |
933 | WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK); |
995 | WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK); |
934 | 996 | ||
935 | if (i == 100) { |
997 | if (i == 100) { |
936 | DRM_ERROR("Timeout setting UVD clocks!\n"); |
998 | DRM_ERROR("Timeout setting UVD clocks!\n"); |
937 | return -ETIMEDOUT; |
999 | return -ETIMEDOUT; |
938 | } |
1000 | } |
939 | 1001 | ||
940 | return 0; |
1002 | return 0; |
941 | }>>=>>>>>>>=>>=>>>>>>>>>> |
1003 | }>>=>>>>>>>=>>=>>>>>>>>>> |