Commit | Line | Data |
---|---|---|
f2ba57b5 CK |
1 | /* |
2 | * Copyright 2011 Advanced Micro Devices, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
5 | * Permission is hereby granted, free of charge, to any person obtaining a | |
6 | * copy of this software and associated documentation files (the | |
7 | * "Software"), to deal in the Software without restriction, including | |
8 | * without limitation the rights to use, copy, modify, merge, publish, | |
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 | |
11 | * the following conditions: | |
12 | * | |
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, | |
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, | |
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 | |
19 | * USE OR OTHER DEALINGS IN THE SOFTWARE. | |
20 | * | |
21 | * The above copyright notice and this permission notice (including the | |
22 | * next paragraph) shall be included in all copies or substantial portions | |
23 | * of the Software. | |
24 | * | |
25 | */ | |
26 | /* | |
27 | * Authors: | |
28 | * Christian König <deathsimple@vodafone.de> | |
29 | */ | |
30 | ||
31 | #include <linux/firmware.h> | |
32 | #include <linux/module.h> | |
33 | #include <drm/drmP.h> | |
34 | #include <drm/drm.h> | |
35 | ||
36 | #include "radeon.h" | |
37 | #include "r600d.h" | |
38 | ||
55b51c88 CK |
39 | /* 1 second timeout */ |
40 | #define UVD_IDLE_TIMEOUT_MS 1000 | |
41 | ||
f2ba57b5 CK |
42 | /* Firmware Names */ |
43 | #define FIRMWARE_RV710 "radeon/RV710_uvd.bin" | |
44 | #define FIRMWARE_CYPRESS "radeon/CYPRESS_uvd.bin" | |
45 | #define FIRMWARE_SUMO "radeon/SUMO_uvd.bin" | |
46 | #define FIRMWARE_TAHITI "radeon/TAHITI_uvd.bin" | |
87167bb1 | 47 | #define FIRMWARE_BONAIRE "radeon/BONAIRE_uvd.bin" |
f2ba57b5 CK |
48 | |
49 | MODULE_FIRMWARE(FIRMWARE_RV710); | |
50 | MODULE_FIRMWARE(FIRMWARE_CYPRESS); | |
51 | MODULE_FIRMWARE(FIRMWARE_SUMO); | |
52 | MODULE_FIRMWARE(FIRMWARE_TAHITI); | |
87167bb1 | 53 | MODULE_FIRMWARE(FIRMWARE_BONAIRE); |
f2ba57b5 | 54 | |
55b51c88 CK |
55 | static void radeon_uvd_idle_work_handler(struct work_struct *work); |
56 | ||
f2ba57b5 CK |
57 | int radeon_uvd_init(struct radeon_device *rdev) |
58 | { | |
9cc2e0e9 | 59 | const struct firmware *fw; |
f2ba57b5 CK |
60 | unsigned long bo_size; |
61 | const char *fw_name; | |
62 | int i, r; | |
63 | ||
55b51c88 CK |
64 | INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler); |
65 | ||
f2ba57b5 CK |
66 | switch (rdev->family) { |
67 | case CHIP_RV710: | |
68 | case CHIP_RV730: | |
69 | case CHIP_RV740: | |
70 | fw_name = FIRMWARE_RV710; | |
71 | break; | |
72 | ||
73 | case CHIP_CYPRESS: | |
74 | case CHIP_HEMLOCK: | |
75 | case CHIP_JUNIPER: | |
76 | case CHIP_REDWOOD: | |
77 | case CHIP_CEDAR: | |
78 | fw_name = FIRMWARE_CYPRESS; | |
79 | break; | |
80 | ||
81 | case CHIP_SUMO: | |
82 | case CHIP_SUMO2: | |
83 | case CHIP_PALM: | |
84 | case CHIP_CAYMAN: | |
85 | case CHIP_BARTS: | |
86 | case CHIP_TURKS: | |
87 | case CHIP_CAICOS: | |
88 | fw_name = FIRMWARE_SUMO; | |
89 | break; | |
90 | ||
91 | case CHIP_TAHITI: | |
92 | case CHIP_VERDE: | |
93 | case CHIP_PITCAIRN: | |
94 | case CHIP_ARUBA: | |
95 | fw_name = FIRMWARE_TAHITI; | |
96 | break; | |
97 | ||
87167bb1 CK |
98 | case CHIP_BONAIRE: |
99 | case CHIP_KABINI: | |
100 | case CHIP_KAVERI: | |
101 | fw_name = FIRMWARE_BONAIRE; | |
102 | break; | |
103 | ||
f2ba57b5 CK |
104 | default: |
105 | return -EINVAL; | |
106 | } | |
107 | ||
9cc2e0e9 | 108 | r = request_firmware(&fw, fw_name, rdev->dev); |
f2ba57b5 CK |
109 | if (r) { |
110 | dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n", | |
111 | fw_name); | |
f2ba57b5 CK |
112 | return r; |
113 | } | |
114 | ||
9cc2e0e9 | 115 | bo_size = RADEON_GPU_PAGE_ALIGN(fw->size + 8) + |
f2ba57b5 CK |
116 | RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE; |
117 | r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true, | |
118 | RADEON_GEM_DOMAIN_VRAM, NULL, &rdev->uvd.vcpu_bo); | |
119 | if (r) { | |
120 | dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r); | |
121 | return r; | |
122 | } | |
123 | ||
9cc2e0e9 CK |
124 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false); |
125 | if (r) { | |
126 | radeon_bo_unref(&rdev->uvd.vcpu_bo); | |
127 | dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r); | |
f2ba57b5 | 128 | return r; |
9cc2e0e9 | 129 | } |
f2ba57b5 | 130 | |
9cc2e0e9 CK |
131 | r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM, |
132 | &rdev->uvd.gpu_addr); | |
133 | if (r) { | |
134 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); | |
135 | radeon_bo_unref(&rdev->uvd.vcpu_bo); | |
136 | dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r); | |
137 | return r; | |
138 | } | |
f2ba57b5 | 139 | |
9cc2e0e9 CK |
140 | r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr); |
141 | if (r) { | |
142 | dev_err(rdev->dev, "(%d) UVD map failed\n", r); | |
f2ba57b5 | 143 | return r; |
9cc2e0e9 CK |
144 | } |
145 | ||
146 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); | |
147 | ||
148 | rdev->uvd.fw_size = fw->size; | |
149 | memset(rdev->uvd.cpu_addr, 0, bo_size); | |
150 | memcpy(rdev->uvd.cpu_addr, fw->data, fw->size); | |
151 | ||
152 | release_firmware(fw); | |
f2ba57b5 CK |
153 | |
154 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { | |
155 | atomic_set(&rdev->uvd.handles[i], 0); | |
156 | rdev->uvd.filp[i] = NULL; | |
157 | } | |
158 | ||
159 | return 0; | |
160 | } | |
161 | ||
162 | void radeon_uvd_fini(struct radeon_device *rdev) | |
f2ba57b5 CK |
163 | { |
164 | int r; | |
165 | ||
166 | if (rdev->uvd.vcpu_bo == NULL) | |
9cc2e0e9 | 167 | return; |
f2ba57b5 CK |
168 | |
169 | r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false); | |
170 | if (!r) { | |
171 | radeon_bo_kunmap(rdev->uvd.vcpu_bo); | |
172 | radeon_bo_unpin(rdev->uvd.vcpu_bo); | |
173 | radeon_bo_unreserve(rdev->uvd.vcpu_bo); | |
174 | } | |
9cc2e0e9 CK |
175 | |
176 | radeon_bo_unref(&rdev->uvd.vcpu_bo); | |
f2ba57b5 CK |
177 | } |
178 | ||
9cc2e0e9 | 179 | int radeon_uvd_suspend(struct radeon_device *rdev) |
f2ba57b5 | 180 | { |
9cc2e0e9 | 181 | unsigned size; |
f2ba57b5 CK |
182 | |
183 | if (rdev->uvd.vcpu_bo == NULL) | |
9cc2e0e9 | 184 | return 0; |
f2ba57b5 | 185 | |
9cc2e0e9 CK |
186 | size = radeon_bo_size(rdev->uvd.vcpu_bo); |
187 | rdev->uvd.saved_bo = kmalloc(size, GFP_KERNEL); | |
188 | memcpy(rdev->uvd.saved_bo, rdev->uvd.cpu_addr, size); | |
f2ba57b5 | 189 | |
9cc2e0e9 CK |
190 | return 0; |
191 | } | |
089920f2 | 192 | |
9cc2e0e9 CK |
193 | int radeon_uvd_resume(struct radeon_device *rdev) |
194 | { | |
195 | if (rdev->uvd.vcpu_bo == NULL) | |
196 | return -EINVAL; | |
f2ba57b5 | 197 | |
9cc2e0e9 CK |
198 | if (rdev->uvd.saved_bo != NULL) { |
199 | unsigned size = radeon_bo_size(rdev->uvd.vcpu_bo); | |
200 | memcpy(rdev->uvd.cpu_addr, rdev->uvd.saved_bo, size); | |
201 | kfree(rdev->uvd.saved_bo); | |
202 | rdev->uvd.saved_bo = NULL; | |
f2ba57b5 CK |
203 | } |
204 | ||
f2ba57b5 CK |
205 | return 0; |
206 | } | |
207 | ||
208 | void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo) | |
209 | { | |
210 | rbo->placement.fpfn = 0 >> PAGE_SHIFT; | |
211 | rbo->placement.lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT; | |
212 | } | |
213 | ||
214 | void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp) | |
215 | { | |
216 | int i, r; | |
217 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { | |
218 | if (rdev->uvd.filp[i] == filp) { | |
219 | uint32_t handle = atomic_read(&rdev->uvd.handles[i]); | |
220 | struct radeon_fence *fence; | |
221 | ||
222 | r = radeon_uvd_get_destroy_msg(rdev, | |
223 | R600_RING_TYPE_UVD_INDEX, handle, &fence); | |
224 | if (r) { | |
225 | DRM_ERROR("Error destroying UVD (%d)!\n", r); | |
226 | continue; | |
227 | } | |
228 | ||
229 | radeon_fence_wait(fence, false); | |
230 | radeon_fence_unref(&fence); | |
231 | ||
232 | rdev->uvd.filp[i] = NULL; | |
233 | atomic_set(&rdev->uvd.handles[i], 0); | |
234 | } | |
235 | } | |
236 | } | |
237 | ||
238 | static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[]) | |
239 | { | |
240 | unsigned stream_type = msg[4]; | |
241 | unsigned width = msg[6]; | |
242 | unsigned height = msg[7]; | |
243 | unsigned dpb_size = msg[9]; | |
244 | unsigned pitch = msg[28]; | |
245 | ||
246 | unsigned width_in_mb = width / 16; | |
247 | unsigned height_in_mb = ALIGN(height / 16, 2); | |
248 | ||
249 | unsigned image_size, tmp, min_dpb_size; | |
250 | ||
251 | image_size = width * height; | |
252 | image_size += image_size / 2; | |
253 | image_size = ALIGN(image_size, 1024); | |
254 | ||
255 | switch (stream_type) { | |
256 | case 0: /* H264 */ | |
257 | ||
258 | /* reference picture buffer */ | |
259 | min_dpb_size = image_size * 17; | |
260 | ||
261 | /* macroblock context buffer */ | |
262 | min_dpb_size += width_in_mb * height_in_mb * 17 * 192; | |
263 | ||
264 | /* IT surface buffer */ | |
265 | min_dpb_size += width_in_mb * height_in_mb * 32; | |
266 | break; | |
267 | ||
268 | case 1: /* VC1 */ | |
269 | ||
270 | /* reference picture buffer */ | |
271 | min_dpb_size = image_size * 3; | |
272 | ||
273 | /* CONTEXT_BUFFER */ | |
274 | min_dpb_size += width_in_mb * height_in_mb * 128; | |
275 | ||
276 | /* IT surface buffer */ | |
277 | min_dpb_size += width_in_mb * 64; | |
278 | ||
279 | /* DB surface buffer */ | |
280 | min_dpb_size += width_in_mb * 128; | |
281 | ||
282 | /* BP */ | |
283 | tmp = max(width_in_mb, height_in_mb); | |
284 | min_dpb_size += ALIGN(tmp * 7 * 16, 64); | |
285 | break; | |
286 | ||
287 | case 3: /* MPEG2 */ | |
288 | ||
289 | /* reference picture buffer */ | |
290 | min_dpb_size = image_size * 3; | |
291 | break; | |
292 | ||
293 | case 4: /* MPEG4 */ | |
294 | ||
295 | /* reference picture buffer */ | |
296 | min_dpb_size = image_size * 3; | |
297 | ||
298 | /* CM */ | |
299 | min_dpb_size += width_in_mb * height_in_mb * 64; | |
300 | ||
301 | /* IT surface buffer */ | |
302 | min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64); | |
303 | break; | |
304 | ||
305 | default: | |
306 | DRM_ERROR("UVD codec not handled %d!\n", stream_type); | |
307 | return -EINVAL; | |
308 | } | |
309 | ||
310 | if (width > pitch) { | |
311 | DRM_ERROR("Invalid UVD decoding target pitch!\n"); | |
312 | return -EINVAL; | |
313 | } | |
314 | ||
315 | if (dpb_size < min_dpb_size) { | |
316 | DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n", | |
317 | dpb_size, min_dpb_size); | |
318 | return -EINVAL; | |
319 | } | |
320 | ||
321 | buf_sizes[0x1] = dpb_size; | |
322 | buf_sizes[0x2] = image_size; | |
323 | return 0; | |
324 | } | |
325 | ||
326 | static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo, | |
327 | unsigned offset, unsigned buf_sizes[]) | |
328 | { | |
329 | int32_t *msg, msg_type, handle; | |
330 | void *ptr; | |
331 | ||
332 | int i, r; | |
333 | ||
334 | if (offset & 0x3F) { | |
335 | DRM_ERROR("UVD messages must be 64 byte aligned!\n"); | |
336 | return -EINVAL; | |
337 | } | |
338 | ||
339 | r = radeon_bo_kmap(bo, &ptr); | |
340 | if (r) | |
341 | return r; | |
342 | ||
343 | msg = ptr + offset; | |
344 | ||
345 | msg_type = msg[1]; | |
346 | handle = msg[2]; | |
347 | ||
348 | if (handle == 0) { | |
349 | DRM_ERROR("Invalid UVD handle!\n"); | |
350 | return -EINVAL; | |
351 | } | |
352 | ||
353 | if (msg_type == 1) { | |
354 | /* it's a decode msg, calc buffer sizes */ | |
355 | r = radeon_uvd_cs_msg_decode(msg, buf_sizes); | |
356 | radeon_bo_kunmap(bo); | |
357 | if (r) | |
358 | return r; | |
359 | ||
360 | } else if (msg_type == 2) { | |
361 | /* it's a destroy msg, free the handle */ | |
362 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) | |
363 | atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0); | |
364 | radeon_bo_kunmap(bo); | |
365 | return 0; | |
366 | } else { | |
367 | /* it's a create msg, no special handling needed */ | |
368 | radeon_bo_kunmap(bo); | |
369 | } | |
370 | ||
371 | /* create or decode, validate the handle */ | |
372 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { | |
373 | if (atomic_read(&p->rdev->uvd.handles[i]) == handle) | |
374 | return 0; | |
375 | } | |
376 | ||
377 | /* handle not found try to alloc a new one */ | |
378 | for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) { | |
379 | if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) { | |
380 | p->rdev->uvd.filp[i] = p->filp; | |
381 | return 0; | |
382 | } | |
383 | } | |
384 | ||
385 | DRM_ERROR("No more free UVD handles!\n"); | |
386 | return -EINVAL; | |
387 | } | |
388 | ||
389 | static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p, | |
390 | int data0, int data1, | |
391 | unsigned buf_sizes[]) | |
392 | { | |
393 | struct radeon_cs_chunk *relocs_chunk; | |
394 | struct radeon_cs_reloc *reloc; | |
395 | unsigned idx, cmd, offset; | |
396 | uint64_t start, end; | |
397 | int r; | |
398 | ||
399 | relocs_chunk = &p->chunks[p->chunk_relocs_idx]; | |
400 | offset = radeon_get_ib_value(p, data0); | |
401 | idx = radeon_get_ib_value(p, data1); | |
402 | if (idx >= relocs_chunk->length_dw) { | |
403 | DRM_ERROR("Relocs at %d after relocations chunk end %d !\n", | |
404 | idx, relocs_chunk->length_dw); | |
405 | return -EINVAL; | |
406 | } | |
407 | ||
408 | reloc = p->relocs_ptr[(idx / 4)]; | |
409 | start = reloc->lobj.gpu_offset; | |
410 | end = start + radeon_bo_size(reloc->robj); | |
411 | start += offset; | |
412 | ||
413 | p->ib.ptr[data0] = start & 0xFFFFFFFF; | |
414 | p->ib.ptr[data1] = start >> 32; | |
415 | ||
416 | cmd = radeon_get_ib_value(p, p->idx) >> 1; | |
417 | ||
418 | if (cmd < 0x4) { | |
419 | if ((end - start) < buf_sizes[cmd]) { | |
420 | DRM_ERROR("buffer to small (%d / %d)!\n", | |
421 | (unsigned)(end - start), buf_sizes[cmd]); | |
422 | return -EINVAL; | |
423 | } | |
424 | ||
425 | } else if (cmd != 0x100) { | |
426 | DRM_ERROR("invalid UVD command %X!\n", cmd); | |
427 | return -EINVAL; | |
428 | } | |
429 | ||
a92c7d55 CK |
430 | if ((start >> 28) != (end >> 28)) { |
431 | DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n", | |
432 | start, end); | |
433 | return -EINVAL; | |
f2ba57b5 CK |
434 | } |
435 | ||
a92c7d55 CK |
436 | /* TODO: is this still necessary on NI+ ? */ |
437 | if ((cmd == 0 || cmd == 0x3) && | |
438 | (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) { | |
439 | DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n", | |
f2ba57b5 CK |
440 | start, end); |
441 | return -EINVAL; | |
442 | } | |
443 | ||
a92c7d55 CK |
444 | if (cmd == 0) { |
445 | r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes); | |
446 | if (r) | |
447 | return r; | |
448 | } | |
449 | ||
f2ba57b5 CK |
450 | return 0; |
451 | } | |
452 | ||
453 | static int radeon_uvd_cs_reg(struct radeon_cs_parser *p, | |
454 | struct radeon_cs_packet *pkt, | |
455 | int *data0, int *data1, | |
456 | unsigned buf_sizes[]) | |
457 | { | |
458 | int i, r; | |
459 | ||
460 | p->idx++; | |
461 | for (i = 0; i <= pkt->count; ++i) { | |
462 | switch (pkt->reg + i*4) { | |
463 | case UVD_GPCOM_VCPU_DATA0: | |
464 | *data0 = p->idx; | |
465 | break; | |
466 | case UVD_GPCOM_VCPU_DATA1: | |
467 | *data1 = p->idx; | |
468 | break; | |
469 | case UVD_GPCOM_VCPU_CMD: | |
470 | r = radeon_uvd_cs_reloc(p, *data0, *data1, buf_sizes); | |
471 | if (r) | |
472 | return r; | |
473 | break; | |
474 | case UVD_ENGINE_CNTL: | |
475 | break; | |
476 | default: | |
477 | DRM_ERROR("Invalid reg 0x%X!\n", | |
478 | pkt->reg + i*4); | |
479 | return -EINVAL; | |
480 | } | |
481 | p->idx++; | |
482 | } | |
483 | return 0; | |
484 | } | |
485 | ||
486 | int radeon_uvd_cs_parse(struct radeon_cs_parser *p) | |
487 | { | |
488 | struct radeon_cs_packet pkt; | |
489 | int r, data0 = 0, data1 = 0; | |
490 | ||
491 | /* minimum buffer sizes */ | |
492 | unsigned buf_sizes[] = { | |
493 | [0x00000000] = 2048, | |
494 | [0x00000001] = 32 * 1024 * 1024, | |
495 | [0x00000002] = 2048 * 1152 * 3, | |
496 | [0x00000003] = 2048, | |
497 | }; | |
498 | ||
499 | if (p->chunks[p->chunk_ib_idx].length_dw % 16) { | |
500 | DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n", | |
501 | p->chunks[p->chunk_ib_idx].length_dw); | |
502 | return -EINVAL; | |
503 | } | |
504 | ||
505 | if (p->chunk_relocs_idx == -1) { | |
506 | DRM_ERROR("No relocation chunk !\n"); | |
507 | return -EINVAL; | |
508 | } | |
509 | ||
510 | ||
511 | do { | |
512 | r = radeon_cs_packet_parse(p, &pkt, p->idx); | |
513 | if (r) | |
514 | return r; | |
515 | switch (pkt.type) { | |
516 | case RADEON_PACKET_TYPE0: | |
517 | r = radeon_uvd_cs_reg(p, &pkt, &data0, | |
518 | &data1, buf_sizes); | |
519 | if (r) | |
520 | return r; | |
521 | break; | |
522 | case RADEON_PACKET_TYPE2: | |
523 | p->idx += pkt.count + 2; | |
524 | break; | |
525 | default: | |
526 | DRM_ERROR("Unknown packet type %d !\n", pkt.type); | |
527 | return -EINVAL; | |
528 | } | |
529 | } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw); | |
530 | return 0; | |
531 | } | |
532 | ||
533 | static int radeon_uvd_send_msg(struct radeon_device *rdev, | |
534 | int ring, struct radeon_bo *bo, | |
535 | struct radeon_fence **fence) | |
536 | { | |
537 | struct ttm_validate_buffer tv; | |
ecff665f | 538 | struct ww_acquire_ctx ticket; |
f2ba57b5 CK |
539 | struct list_head head; |
540 | struct radeon_ib ib; | |
541 | uint64_t addr; | |
542 | int i, r; | |
543 | ||
544 | memset(&tv, 0, sizeof(tv)); | |
545 | tv.bo = &bo->tbo; | |
546 | ||
547 | INIT_LIST_HEAD(&head); | |
548 | list_add(&tv.head, &head); | |
549 | ||
ecff665f | 550 | r = ttm_eu_reserve_buffers(&ticket, &head); |
f2ba57b5 CK |
551 | if (r) |
552 | return r; | |
553 | ||
554 | radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_VRAM); | |
555 | radeon_uvd_force_into_uvd_segment(bo); | |
556 | ||
557 | r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); | |
ecff665f ML |
558 | if (r) |
559 | goto err; | |
f2ba57b5 CK |
560 | |
561 | r = radeon_ib_get(rdev, ring, &ib, NULL, 16); | |
ecff665f ML |
562 | if (r) |
563 | goto err; | |
f2ba57b5 CK |
564 | |
565 | addr = radeon_bo_gpu_offset(bo); | |
566 | ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0); | |
567 | ib.ptr[1] = addr; | |
568 | ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0); | |
569 | ib.ptr[3] = addr >> 32; | |
570 | ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0); | |
571 | ib.ptr[5] = 0; | |
572 | for (i = 6; i < 16; ++i) | |
573 | ib.ptr[i] = PACKET2(0); | |
574 | ib.length_dw = 16; | |
575 | ||
576 | r = radeon_ib_schedule(rdev, &ib, NULL); | |
ecff665f ML |
577 | if (r) |
578 | goto err; | |
579 | ttm_eu_fence_buffer_objects(&ticket, &head, ib.fence); | |
f2ba57b5 CK |
580 | |
581 | if (fence) | |
582 | *fence = radeon_fence_ref(ib.fence); | |
583 | ||
584 | radeon_ib_free(rdev, &ib); | |
585 | radeon_bo_unref(&bo); | |
586 | return 0; | |
ecff665f ML |
587 | |
588 | err: | |
589 | ttm_eu_backoff_reservation(&ticket, &head); | |
590 | return r; | |
f2ba57b5 CK |
591 | } |
592 | ||
593 | /* multiple fence commands without any stream commands in between can | |
594 | crash the vcpu so just try to emmit a dummy create/destroy msg to | |
595 | avoid this */ | |
596 | int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring, | |
597 | uint32_t handle, struct radeon_fence **fence) | |
598 | { | |
599 | struct radeon_bo *bo; | |
600 | uint32_t *msg; | |
601 | int r, i; | |
602 | ||
603 | r = radeon_bo_create(rdev, 1024, PAGE_SIZE, true, | |
604 | RADEON_GEM_DOMAIN_VRAM, NULL, &bo); | |
605 | if (r) | |
606 | return r; | |
607 | ||
608 | r = radeon_bo_reserve(bo, false); | |
609 | if (r) { | |
610 | radeon_bo_unref(&bo); | |
611 | return r; | |
612 | } | |
613 | ||
614 | r = radeon_bo_kmap(bo, (void **)&msg); | |
615 | if (r) { | |
616 | radeon_bo_unreserve(bo); | |
617 | radeon_bo_unref(&bo); | |
618 | return r; | |
619 | } | |
620 | ||
621 | /* stitch together an UVD create msg */ | |
9b1be4dc AD |
622 | msg[0] = cpu_to_le32(0x00000de4); |
623 | msg[1] = cpu_to_le32(0x00000000); | |
624 | msg[2] = cpu_to_le32(handle); | |
625 | msg[3] = cpu_to_le32(0x00000000); | |
626 | msg[4] = cpu_to_le32(0x00000000); | |
627 | msg[5] = cpu_to_le32(0x00000000); | |
628 | msg[6] = cpu_to_le32(0x00000000); | |
629 | msg[7] = cpu_to_le32(0x00000780); | |
630 | msg[8] = cpu_to_le32(0x00000440); | |
631 | msg[9] = cpu_to_le32(0x00000000); | |
632 | msg[10] = cpu_to_le32(0x01b37000); | |
f2ba57b5 | 633 | for (i = 11; i < 1024; ++i) |
9b1be4dc | 634 | msg[i] = cpu_to_le32(0x0); |
f2ba57b5 CK |
635 | |
636 | radeon_bo_kunmap(bo); | |
637 | radeon_bo_unreserve(bo); | |
638 | ||
639 | return radeon_uvd_send_msg(rdev, ring, bo, fence); | |
640 | } | |
641 | ||
642 | int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring, | |
643 | uint32_t handle, struct radeon_fence **fence) | |
644 | { | |
645 | struct radeon_bo *bo; | |
646 | uint32_t *msg; | |
647 | int r, i; | |
648 | ||
649 | r = radeon_bo_create(rdev, 1024, PAGE_SIZE, true, | |
650 | RADEON_GEM_DOMAIN_VRAM, NULL, &bo); | |
651 | if (r) | |
652 | return r; | |
653 | ||
654 | r = radeon_bo_reserve(bo, false); | |
655 | if (r) { | |
656 | radeon_bo_unref(&bo); | |
657 | return r; | |
658 | } | |
659 | ||
660 | r = radeon_bo_kmap(bo, (void **)&msg); | |
661 | if (r) { | |
662 | radeon_bo_unreserve(bo); | |
663 | radeon_bo_unref(&bo); | |
664 | return r; | |
665 | } | |
666 | ||
667 | /* stitch together an UVD destroy msg */ | |
9b1be4dc AD |
668 | msg[0] = cpu_to_le32(0x00000de4); |
669 | msg[1] = cpu_to_le32(0x00000002); | |
670 | msg[2] = cpu_to_le32(handle); | |
671 | msg[3] = cpu_to_le32(0x00000000); | |
f2ba57b5 | 672 | for (i = 4; i < 1024; ++i) |
9b1be4dc | 673 | msg[i] = cpu_to_le32(0x0); |
f2ba57b5 CK |
674 | |
675 | radeon_bo_kunmap(bo); | |
676 | radeon_bo_unreserve(bo); | |
677 | ||
678 | return radeon_uvd_send_msg(rdev, ring, bo, fence); | |
679 | } | |
55b51c88 CK |
680 | |
681 | static void radeon_uvd_idle_work_handler(struct work_struct *work) | |
682 | { | |
683 | struct radeon_device *rdev = | |
684 | container_of(work, struct radeon_device, uvd.idle_work.work); | |
685 | ||
8a227555 AD |
686 | if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) { |
687 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { | |
688 | mutex_lock(&rdev->pm.mutex); | |
689 | rdev->pm.dpm.uvd_active = false; | |
690 | mutex_unlock(&rdev->pm.mutex); | |
691 | radeon_pm_compute_clocks(rdev); | |
692 | } else { | |
693 | radeon_set_uvd_clocks(rdev, 0, 0); | |
694 | } | |
695 | } else { | |
55b51c88 CK |
696 | schedule_delayed_work(&rdev->uvd.idle_work, |
697 | msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS)); | |
8a227555 | 698 | } |
55b51c88 CK |
699 | } |
700 | ||
701 | void radeon_uvd_note_usage(struct radeon_device *rdev) | |
702 | { | |
703 | bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work); | |
704 | set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work, | |
705 | msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS)); | |
8a227555 AD |
706 | if (set_clocks) { |
707 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { | |
708 | /* XXX pick SD/HD/MVC */ | |
709 | radeon_dpm_enable_power_state(rdev, POWER_STATE_TYPE_INTERNAL_UVD); | |
710 | } else { | |
711 | radeon_set_uvd_clocks(rdev, 53300, 40000); | |
712 | } | |
713 | } | |
55b51c88 | 714 | } |
facd112d CK |
715 | |
716 | static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq, | |
717 | unsigned target_freq, | |
718 | unsigned pd_min, | |
719 | unsigned pd_even) | |
720 | { | |
721 | unsigned post_div = vco_freq / target_freq; | |
722 | ||
723 | /* adjust to post divider minimum value */ | |
724 | if (post_div < pd_min) | |
725 | post_div = pd_min; | |
726 | ||
727 | /* we alway need a frequency less than or equal the target */ | |
728 | if ((vco_freq / post_div) > target_freq) | |
729 | post_div += 1; | |
730 | ||
731 | /* post dividers above a certain value must be even */ | |
732 | if (post_div > pd_even && post_div % 2) | |
733 | post_div += 1; | |
734 | ||
735 | return post_div; | |
736 | } | |
737 | ||
738 | /** | |
739 | * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers | |
740 | * | |
741 | * @rdev: radeon_device pointer | |
742 | * @vclk: wanted VCLK | |
743 | * @dclk: wanted DCLK | |
744 | * @vco_min: minimum VCO frequency | |
745 | * @vco_max: maximum VCO frequency | |
746 | * @fb_factor: factor to multiply vco freq with | |
747 | * @fb_mask: limit and bitmask for feedback divider | |
748 | * @pd_min: post divider minimum | |
749 | * @pd_max: post divider maximum | |
750 | * @pd_even: post divider must be even above this value | |
751 | * @optimal_fb_div: resulting feedback divider | |
752 | * @optimal_vclk_div: resulting vclk post divider | |
753 | * @optimal_dclk_div: resulting dclk post divider | |
754 | * | |
755 | * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs). | |
756 | * Returns zero on success -EINVAL on error. | |
757 | */ | |
758 | int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev, | |
759 | unsigned vclk, unsigned dclk, | |
760 | unsigned vco_min, unsigned vco_max, | |
761 | unsigned fb_factor, unsigned fb_mask, | |
762 | unsigned pd_min, unsigned pd_max, | |
763 | unsigned pd_even, | |
764 | unsigned *optimal_fb_div, | |
765 | unsigned *optimal_vclk_div, | |
766 | unsigned *optimal_dclk_div) | |
767 | { | |
768 | unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq; | |
769 | ||
770 | /* start off with something large */ | |
771 | unsigned optimal_score = ~0; | |
772 | ||
773 | /* loop through vco from low to high */ | |
774 | vco_min = max(max(vco_min, vclk), dclk); | |
775 | for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) { | |
776 | ||
777 | uint64_t fb_div = (uint64_t)vco_freq * fb_factor; | |
778 | unsigned vclk_div, dclk_div, score; | |
779 | ||
780 | do_div(fb_div, ref_freq); | |
781 | ||
782 | /* fb div out of range ? */ | |
783 | if (fb_div > fb_mask) | |
784 | break; /* it can oly get worse */ | |
785 | ||
786 | fb_div &= fb_mask; | |
787 | ||
788 | /* calc vclk divider with current vco freq */ | |
789 | vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk, | |
790 | pd_min, pd_even); | |
791 | if (vclk_div > pd_max) | |
792 | break; /* vco is too big, it has to stop */ | |
793 | ||
794 | /* calc dclk divider with current vco freq */ | |
795 | dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk, | |
796 | pd_min, pd_even); | |
797 | if (vclk_div > pd_max) | |
798 | break; /* vco is too big, it has to stop */ | |
799 | ||
800 | /* calc score with current vco freq */ | |
801 | score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div); | |
802 | ||
803 | /* determine if this vco setting is better than current optimal settings */ | |
804 | if (score < optimal_score) { | |
805 | *optimal_fb_div = fb_div; | |
806 | *optimal_vclk_div = vclk_div; | |
807 | *optimal_dclk_div = dclk_div; | |
808 | optimal_score = score; | |
809 | if (optimal_score == 0) | |
810 | break; /* it can't get better than this */ | |
811 | } | |
812 | } | |
813 | ||
814 | /* did we found a valid setup ? */ | |
815 | if (optimal_score == ~0) | |
816 | return -EINVAL; | |
817 | ||
818 | return 0; | |
819 | } | |
820 | ||
821 | int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev, | |
822 | unsigned cg_upll_func_cntl) | |
823 | { | |
824 | unsigned i; | |
825 | ||
826 | /* make sure UPLL_CTLREQ is deasserted */ | |
827 | WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK); | |
828 | ||
829 | mdelay(10); | |
830 | ||
831 | /* assert UPLL_CTLREQ */ | |
832 | WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK); | |
833 | ||
834 | /* wait for CTLACK and CTLACK2 to get asserted */ | |
835 | for (i = 0; i < 100; ++i) { | |
836 | uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK; | |
837 | if ((RREG32(cg_upll_func_cntl) & mask) == mask) | |
838 | break; | |
839 | mdelay(10); | |
840 | } | |
841 | ||
842 | /* deassert UPLL_CTLREQ */ | |
843 | WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK); | |
844 | ||
845 | if (i == 100) { | |
846 | DRM_ERROR("Timeout setting UVD clocks!\n"); | |
847 | return -ETIMEDOUT; | |
848 | } | |
849 | ||
850 | return 0; | |
851 | } |