1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31#include <linux/firmware.h>
32#include <linux/module.h>
33
34#include <drm/drm.h>
35
36#include "amdgpu.h"
37#include "amdgpu_pm.h"
38#include "amdgpu_uvd.h"
39#include "cikd.h"
40#include "uvd/uvd_4_2_d.h"
41
42#include "amdgpu_ras.h"
43
44
45#define UVD_IDLE_TIMEOUT msecs_to_jiffies(1000)
46
47
48#define FW_1_65_10 ((1 << 24) | (65 << 16) | (10 << 8))
49#define FW_1_87_11 ((1 << 24) | (87 << 16) | (11 << 8))
50#define FW_1_87_12 ((1 << 24) | (87 << 16) | (12 << 8))
51#define FW_1_37_15 ((1 << 24) | (37 << 16) | (15 << 8))
52
53
54#define FW_1_66_16 ((1 << 24) | (66 << 16) | (16 << 8))
55
56
57#ifdef CONFIG_DRM_AMDGPU_CIK
58#define FIRMWARE_BONAIRE "amdgpu/bonaire_uvd.bin"
59#define FIRMWARE_KABINI "amdgpu/kabini_uvd.bin"
60#define FIRMWARE_KAVERI "amdgpu/kaveri_uvd.bin"
61#define FIRMWARE_HAWAII "amdgpu/hawaii_uvd.bin"
62#define FIRMWARE_MULLINS "amdgpu/mullins_uvd.bin"
63#endif
64#define FIRMWARE_TONGA "amdgpu/tonga_uvd.bin"
65#define FIRMWARE_CARRIZO "amdgpu/carrizo_uvd.bin"
66#define FIRMWARE_FIJI "amdgpu/fiji_uvd.bin"
67#define FIRMWARE_STONEY "amdgpu/stoney_uvd.bin"
68#define FIRMWARE_POLARIS10 "amdgpu/polaris10_uvd.bin"
69#define FIRMWARE_POLARIS11 "amdgpu/polaris11_uvd.bin"
70#define FIRMWARE_POLARIS12 "amdgpu/polaris12_uvd.bin"
71#define FIRMWARE_VEGAM "amdgpu/vegam_uvd.bin"
72
73#define FIRMWARE_VEGA10 "amdgpu/vega10_uvd.bin"
74#define FIRMWARE_VEGA12 "amdgpu/vega12_uvd.bin"
75#define FIRMWARE_VEGA20 "amdgpu/vega20_uvd.bin"
76
77
78#define UVD_GPCOM_VCPU_CMD 0x03c3
79#define UVD_GPCOM_VCPU_DATA0 0x03c4
80#define UVD_GPCOM_VCPU_DATA1 0x03c5
81#define UVD_NO_OP 0x03ff
82#define UVD_BASE_SI 0x3800
83
84
85
86
87
88
89struct amdgpu_uvd_cs_ctx {
90 struct amdgpu_cs_parser *parser;
91 unsigned reg, count;
92 unsigned data0, data1;
93 unsigned idx;
94 unsigned ib_idx;
95
96
97 bool has_msg_cmd;
98
99
100 unsigned *buf_sizes;
101};
102
103#ifdef CONFIG_DRM_AMDGPU_CIK
104MODULE_FIRMWARE(FIRMWARE_BONAIRE);
105MODULE_FIRMWARE(FIRMWARE_KABINI);
106MODULE_FIRMWARE(FIRMWARE_KAVERI);
107MODULE_FIRMWARE(FIRMWARE_HAWAII);
108MODULE_FIRMWARE(FIRMWARE_MULLINS);
109#endif
110MODULE_FIRMWARE(FIRMWARE_TONGA);
111MODULE_FIRMWARE(FIRMWARE_CARRIZO);
112MODULE_FIRMWARE(FIRMWARE_FIJI);
113MODULE_FIRMWARE(FIRMWARE_STONEY);
114MODULE_FIRMWARE(FIRMWARE_POLARIS10);
115MODULE_FIRMWARE(FIRMWARE_POLARIS11);
116MODULE_FIRMWARE(FIRMWARE_POLARIS12);
117MODULE_FIRMWARE(FIRMWARE_VEGAM);
118
119MODULE_FIRMWARE(FIRMWARE_VEGA10);
120MODULE_FIRMWARE(FIRMWARE_VEGA12);
121MODULE_FIRMWARE(FIRMWARE_VEGA20);
122
123static void amdgpu_uvd_idle_work_handler(struct work_struct *work);
124
125int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
126{
127 unsigned long bo_size;
128 const char *fw_name;
129 const struct common_firmware_header *hdr;
130 unsigned family_id;
131 int i, j, r;
132
133 INIT_DELAYED_WORK(&adev->uvd.idle_work, amdgpu_uvd_idle_work_handler);
134
135 switch (adev->asic_type) {
136#ifdef CONFIG_DRM_AMDGPU_CIK
137 case CHIP_BONAIRE:
138 fw_name = FIRMWARE_BONAIRE;
139 break;
140 case CHIP_KABINI:
141 fw_name = FIRMWARE_KABINI;
142 break;
143 case CHIP_KAVERI:
144 fw_name = FIRMWARE_KAVERI;
145 break;
146 case CHIP_HAWAII:
147 fw_name = FIRMWARE_HAWAII;
148 break;
149 case CHIP_MULLINS:
150 fw_name = FIRMWARE_MULLINS;
151 break;
152#endif
153 case CHIP_TONGA:
154 fw_name = FIRMWARE_TONGA;
155 break;
156 case CHIP_FIJI:
157 fw_name = FIRMWARE_FIJI;
158 break;
159 case CHIP_CARRIZO:
160 fw_name = FIRMWARE_CARRIZO;
161 break;
162 case CHIP_STONEY:
163 fw_name = FIRMWARE_STONEY;
164 break;
165 case CHIP_POLARIS10:
166 fw_name = FIRMWARE_POLARIS10;
167 break;
168 case CHIP_POLARIS11:
169 fw_name = FIRMWARE_POLARIS11;
170 break;
171 case CHIP_POLARIS12:
172 fw_name = FIRMWARE_POLARIS12;
173 break;
174 case CHIP_VEGA10:
175 fw_name = FIRMWARE_VEGA10;
176 break;
177 case CHIP_VEGA12:
178 fw_name = FIRMWARE_VEGA12;
179 break;
180 case CHIP_VEGAM:
181 fw_name = FIRMWARE_VEGAM;
182 break;
183 case CHIP_VEGA20:
184 fw_name = FIRMWARE_VEGA20;
185 break;
186 default:
187 return -EINVAL;
188 }
189
190 r = request_firmware(&adev->uvd.fw, fw_name, adev->dev);
191 if (r) {
192 dev_err(adev->dev, "amdgpu_uvd: Can't load firmware \"%s\"\n",
193 fw_name);
194 return r;
195 }
196
197 r = amdgpu_ucode_validate(adev->uvd.fw);
198 if (r) {
199 dev_err(adev->dev, "amdgpu_uvd: Can't validate firmware \"%s\"\n",
200 fw_name);
201 release_firmware(adev->uvd.fw);
202 adev->uvd.fw = NULL;
203 return r;
204 }
205
206
207 adev->uvd.max_handles = AMDGPU_DEFAULT_UVD_HANDLES;
208
209 hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
210 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
211
212 if (adev->asic_type < CHIP_VEGA20) {
213 unsigned version_major, version_minor;
214
215 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
216 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
217 DRM_INFO("Found UVD firmware Version: %hu.%hu Family ID: %hu\n",
218 version_major, version_minor, family_id);
219
220
221
222
223
224
225
226 if ((version_major > 0x01) ||
227 ((version_major == 0x01) && (version_minor >= 0x50)))
228 adev->uvd.max_handles = AMDGPU_MAX_UVD_HANDLES;
229
230 adev->uvd.fw_version = ((version_major << 24) | (version_minor << 16) |
231 (family_id << 8));
232
233 if ((adev->asic_type == CHIP_POLARIS10 ||
234 adev->asic_type == CHIP_POLARIS11) &&
235 (adev->uvd.fw_version < FW_1_66_16))
236 DRM_ERROR("POLARIS10/11 UVD firmware version %hu.%hu is too old.\n",
237 version_major, version_minor);
238 } else {
239 unsigned int enc_major, enc_minor, dec_minor;
240
241 dec_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
242 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 24) & 0x3f;
243 enc_major = (le32_to_cpu(hdr->ucode_version) >> 30) & 0x3;
244 DRM_INFO("Found UVD firmware ENC: %hu.%hu DEC: .%hu Family ID: %hu\n",
245 enc_major, enc_minor, dec_minor, family_id);
246
247 adev->uvd.max_handles = AMDGPU_MAX_UVD_HANDLES;
248
249 adev->uvd.fw_version = le32_to_cpu(hdr->ucode_version);
250 }
251
252 bo_size = AMDGPU_UVD_STACK_SIZE + AMDGPU_UVD_HEAP_SIZE
253 + AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles;
254 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
255 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
256
257 for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
258 if (adev->uvd.harvest_config & (1 << j))
259 continue;
260 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
261 AMDGPU_GEM_DOMAIN_VRAM, &adev->uvd.inst[j].vcpu_bo,
262 &adev->uvd.inst[j].gpu_addr, &adev->uvd.inst[j].cpu_addr);
263 if (r) {
264 dev_err(adev->dev, "(%d) failed to allocate UVD bo\n", r);
265 return r;
266 }
267 }
268
269 for (i = 0; i < adev->uvd.max_handles; ++i) {
270 atomic_set(&adev->uvd.handles[i], 0);
271 adev->uvd.filp[i] = NULL;
272 }
273
274
275 if (!amdgpu_device_ip_block_version_cmp(adev, AMD_IP_BLOCK_TYPE_UVD, 5, 0))
276 adev->uvd.address_64_bit = true;
277
278 switch (adev->asic_type) {
279 case CHIP_TONGA:
280 adev->uvd.use_ctx_buf = adev->uvd.fw_version >= FW_1_65_10;
281 break;
282 case CHIP_CARRIZO:
283 adev->uvd.use_ctx_buf = adev->uvd.fw_version >= FW_1_87_11;
284 break;
285 case CHIP_FIJI:
286 adev->uvd.use_ctx_buf = adev->uvd.fw_version >= FW_1_87_12;
287 break;
288 case CHIP_STONEY:
289 adev->uvd.use_ctx_buf = adev->uvd.fw_version >= FW_1_37_15;
290 break;
291 default:
292 adev->uvd.use_ctx_buf = adev->asic_type >= CHIP_POLARIS10;
293 }
294
295 return 0;
296}
297
298int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
299{
300 int i, j;
301
302 cancel_delayed_work_sync(&adev->uvd.idle_work);
303 drm_sched_entity_destroy(&adev->uvd.entity);
304
305 for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
306 if (adev->uvd.harvest_config & (1 << j))
307 continue;
308 kvfree(adev->uvd.inst[j].saved_bo);
309
310 amdgpu_bo_free_kernel(&adev->uvd.inst[j].vcpu_bo,
311 &adev->uvd.inst[j].gpu_addr,
312 (void **)&adev->uvd.inst[j].cpu_addr);
313
314 amdgpu_ring_fini(&adev->uvd.inst[j].ring);
315
316 for (i = 0; i < AMDGPU_MAX_UVD_ENC_RINGS; ++i)
317 amdgpu_ring_fini(&adev->uvd.inst[j].ring_enc[i]);
318 }
319 release_firmware(adev->uvd.fw);
320
321 return 0;
322}
323
324
325
326
327
328
329
330int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
331{
332 struct amdgpu_ring *ring;
333 struct drm_gpu_scheduler *sched;
334 int r;
335
336 ring = &adev->uvd.inst[0].ring;
337 sched = &ring->sched;
338 r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
339 &sched, 1, NULL);
340 if (r) {
341 DRM_ERROR("Failed setting up UVD kernel entity.\n");
342 return r;
343 }
344
345 return 0;
346}
347
348int amdgpu_uvd_suspend(struct amdgpu_device *adev)
349{
350 unsigned size;
351 void *ptr;
352 int i, j;
353 bool in_ras_intr = amdgpu_ras_intr_triggered();
354
355 cancel_delayed_work_sync(&adev->uvd.idle_work);
356
357
358 if (adev->asic_type < CHIP_POLARIS10) {
359 for (i = 0; i < adev->uvd.max_handles; ++i)
360 if (atomic_read(&adev->uvd.handles[i]))
361 break;
362
363 if (i == adev->uvd.max_handles)
364 return 0;
365 }
366
367 for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
368 if (adev->uvd.harvest_config & (1 << j))
369 continue;
370 if (adev->uvd.inst[j].vcpu_bo == NULL)
371 continue;
372
373 size = amdgpu_bo_size(adev->uvd.inst[j].vcpu_bo);
374 ptr = adev->uvd.inst[j].cpu_addr;
375
376 adev->uvd.inst[j].saved_bo = kvmalloc(size, GFP_KERNEL);
377 if (!adev->uvd.inst[j].saved_bo)
378 return -ENOMEM;
379
380
381 if (in_ras_intr)
382 memset(adev->uvd.inst[j].saved_bo, 0, size);
383 else
384 memcpy_fromio(adev->uvd.inst[j].saved_bo, ptr, size);
385 }
386
387 if (in_ras_intr)
388 DRM_WARN("UVD VCPU state may lost due to RAS ERREVENT_ATHUB_INTERRUPT\n");
389
390 return 0;
391}
392
393int amdgpu_uvd_resume(struct amdgpu_device *adev)
394{
395 unsigned size;
396 void *ptr;
397 int i;
398
399 for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
400 if (adev->uvd.harvest_config & (1 << i))
401 continue;
402 if (adev->uvd.inst[i].vcpu_bo == NULL)
403 return -EINVAL;
404
405 size = amdgpu_bo_size(adev->uvd.inst[i].vcpu_bo);
406 ptr = adev->uvd.inst[i].cpu_addr;
407
408 if (adev->uvd.inst[i].saved_bo != NULL) {
409 memcpy_toio(ptr, adev->uvd.inst[i].saved_bo, size);
410 kvfree(adev->uvd.inst[i].saved_bo);
411 adev->uvd.inst[i].saved_bo = NULL;
412 } else {
413 const struct common_firmware_header *hdr;
414 unsigned offset;
415
416 hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
417 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
418 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
419 memcpy_toio(adev->uvd.inst[i].cpu_addr, adev->uvd.fw->data + offset,
420 le32_to_cpu(hdr->ucode_size_bytes));
421 size -= le32_to_cpu(hdr->ucode_size_bytes);
422 ptr += le32_to_cpu(hdr->ucode_size_bytes);
423 }
424 memset_io(ptr, 0, size);
425
426 amdgpu_fence_driver_force_completion(&adev->uvd.inst[i].ring);
427 }
428 }
429 return 0;
430}
431
432void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
433{
434 struct amdgpu_ring *ring = &adev->uvd.inst[0].ring;
435 int i, r;
436
437 for (i = 0; i < adev->uvd.max_handles; ++i) {
438 uint32_t handle = atomic_read(&adev->uvd.handles[i]);
439
440 if (handle != 0 && adev->uvd.filp[i] == filp) {
441 struct dma_fence *fence;
442
443 r = amdgpu_uvd_get_destroy_msg(ring, handle, false,
444 &fence);
445 if (r) {
446 DRM_ERROR("Error destroying UVD %d!\n", r);
447 continue;
448 }
449
450 dma_fence_wait(fence, false);
451 dma_fence_put(fence);
452
453 adev->uvd.filp[i] = NULL;
454 atomic_set(&adev->uvd.handles[i], 0);
455 }
456 }
457}
458
459static void amdgpu_uvd_force_into_uvd_segment(struct amdgpu_bo *abo)
460{
461 int i;
462 for (i = 0; i < abo->placement.num_placement; ++i) {
463 abo->placements[i].fpfn = 0 >> PAGE_SHIFT;
464 abo->placements[i].lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT;
465 }
466}
467
468static u64 amdgpu_uvd_get_addr_from_ctx(struct amdgpu_uvd_cs_ctx *ctx)
469{
470 uint32_t lo, hi;
471 uint64_t addr;
472
473 lo = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->data0);
474 hi = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->data1);
475 addr = ((uint64_t)lo) | (((uint64_t)hi) << 32);
476
477 return addr;
478}
479
480
481
482
483
484
485
486
487
488static int amdgpu_uvd_cs_pass1(struct amdgpu_uvd_cs_ctx *ctx)
489{
490 struct ttm_operation_ctx tctx = { false, false };
491 struct amdgpu_bo_va_mapping *mapping;
492 struct amdgpu_bo *bo;
493 uint32_t cmd;
494 uint64_t addr = amdgpu_uvd_get_addr_from_ctx(ctx);
495 int r = 0;
496
497 r = amdgpu_cs_find_mapping(ctx->parser, addr, &bo, &mapping);
498 if (r) {
499 DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
500 return r;
501 }
502
503 if (!ctx->parser->adev->uvd.address_64_bit) {
504
505 cmd = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->idx) >> 1;
506 if (cmd == 0x0 || cmd == 0x3) {
507
508 uint32_t domain = AMDGPU_GEM_DOMAIN_VRAM;
509 amdgpu_bo_placement_from_domain(bo, domain);
510 }
511 amdgpu_uvd_force_into_uvd_segment(bo);
512
513 r = ttm_bo_validate(&bo->tbo, &bo->placement, &tctx);
514 }
515
516 return r;
517}
518
519
520
521
522
523
524
525
526
527static int amdgpu_uvd_cs_msg_decode(struct amdgpu_device *adev, uint32_t *msg,
528 unsigned buf_sizes[])
529{
530 unsigned stream_type = msg[4];
531 unsigned width = msg[6];
532 unsigned height = msg[7];
533 unsigned dpb_size = msg[9];
534 unsigned pitch = msg[28];
535 unsigned level = msg[57];
536
537 unsigned width_in_mb = width / 16;
538 unsigned height_in_mb = ALIGN(height / 16, 2);
539 unsigned fs_in_mb = width_in_mb * height_in_mb;
540
541 unsigned image_size, tmp, min_dpb_size, num_dpb_buffer;
542 unsigned min_ctx_size = ~0;
543
544 image_size = width * height;
545 image_size += image_size / 2;
546 image_size = ALIGN(image_size, 1024);
547
548 switch (stream_type) {
549 case 0:
550 switch(level) {
551 case 30:
552 num_dpb_buffer = 8100 / fs_in_mb;
553 break;
554 case 31:
555 num_dpb_buffer = 18000 / fs_in_mb;
556 break;
557 case 32:
558 num_dpb_buffer = 20480 / fs_in_mb;
559 break;
560 case 41:
561 num_dpb_buffer = 32768 / fs_in_mb;
562 break;
563 case 42:
564 num_dpb_buffer = 34816 / fs_in_mb;
565 break;
566 case 50:
567 num_dpb_buffer = 110400 / fs_in_mb;
568 break;
569 case 51:
570 num_dpb_buffer = 184320 / fs_in_mb;
571 break;
572 default:
573 num_dpb_buffer = 184320 / fs_in_mb;
574 break;
575 }
576 num_dpb_buffer++;
577 if (num_dpb_buffer > 17)
578 num_dpb_buffer = 17;
579
580
581 min_dpb_size = image_size * num_dpb_buffer;
582
583
584 min_dpb_size += width_in_mb * height_in_mb * num_dpb_buffer * 192;
585
586
587 min_dpb_size += width_in_mb * height_in_mb * 32;
588 break;
589
590 case 1:
591
592
593 min_dpb_size = image_size * 3;
594
595
596 min_dpb_size += width_in_mb * height_in_mb * 128;
597
598
599 min_dpb_size += width_in_mb * 64;
600
601
602 min_dpb_size += width_in_mb * 128;
603
604
605 tmp = max(width_in_mb, height_in_mb);
606 min_dpb_size += ALIGN(tmp * 7 * 16, 64);
607 break;
608
609 case 3:
610
611
612 min_dpb_size = image_size * 3;
613 break;
614
615 case 4:
616
617
618 min_dpb_size = image_size * 3;
619
620
621 min_dpb_size += width_in_mb * height_in_mb * 64;
622
623
624 min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64);
625 break;
626
627 case 7:
628 switch(level) {
629 case 30:
630 num_dpb_buffer = 8100 / fs_in_mb;
631 break;
632 case 31:
633 num_dpb_buffer = 18000 / fs_in_mb;
634 break;
635 case 32:
636 num_dpb_buffer = 20480 / fs_in_mb;
637 break;
638 case 41:
639 num_dpb_buffer = 32768 / fs_in_mb;
640 break;
641 case 42:
642 num_dpb_buffer = 34816 / fs_in_mb;
643 break;
644 case 50:
645 num_dpb_buffer = 110400 / fs_in_mb;
646 break;
647 case 51:
648 num_dpb_buffer = 184320 / fs_in_mb;
649 break;
650 default:
651 num_dpb_buffer = 184320 / fs_in_mb;
652 break;
653 }
654 num_dpb_buffer++;
655 if (num_dpb_buffer > 17)
656 num_dpb_buffer = 17;
657
658
659 min_dpb_size = image_size * num_dpb_buffer;
660
661 if (!adev->uvd.use_ctx_buf){
662
663 min_dpb_size +=
664 width_in_mb * height_in_mb * num_dpb_buffer * 192;
665
666
667 min_dpb_size += width_in_mb * height_in_mb * 32;
668 } else {
669
670 min_ctx_size =
671 width_in_mb * height_in_mb * num_dpb_buffer * 192;
672 }
673 break;
674
675 case 8:
676 min_dpb_size = 0;
677 break;
678
679 case 16:
680 image_size = (ALIGN(width, 16) * ALIGN(height, 16) * 3) / 2;
681 image_size = ALIGN(image_size, 256);
682
683 num_dpb_buffer = (le32_to_cpu(msg[59]) & 0xff) + 2;
684 min_dpb_size = image_size * num_dpb_buffer;
685 min_ctx_size = ((width + 255) / 16) * ((height + 255) / 16)
686 * 16 * num_dpb_buffer + 52 * 1024;
687 break;
688
689 default:
690 DRM_ERROR("UVD codec not handled %d!\n", stream_type);
691 return -EINVAL;
692 }
693
694 if (width > pitch) {
695 DRM_ERROR("Invalid UVD decoding target pitch!\n");
696 return -EINVAL;
697 }
698
699 if (dpb_size < min_dpb_size) {
700 DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n",
701 dpb_size, min_dpb_size);
702 return -EINVAL;
703 }
704
705 buf_sizes[0x1] = dpb_size;
706 buf_sizes[0x2] = image_size;
707 buf_sizes[0x4] = min_ctx_size;
708
709 adev->uvd.decode_image_width = width;
710 return 0;
711}
712
713
714
715
716
717
718
719
720
721
722
723static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
724 struct amdgpu_bo *bo, unsigned offset)
725{
726 struct amdgpu_device *adev = ctx->parser->adev;
727 int32_t *msg, msg_type, handle;
728 void *ptr;
729 long r;
730 int i;
731
732 if (offset & 0x3F) {
733 DRM_ERROR("UVD messages must be 64 byte aligned!\n");
734 return -EINVAL;
735 }
736
737 r = amdgpu_bo_kmap(bo, &ptr);
738 if (r) {
739 DRM_ERROR("Failed mapping the UVD) message (%ld)!\n", r);
740 return r;
741 }
742
743 msg = ptr + offset;
744
745 msg_type = msg[1];
746 handle = msg[2];
747
748 if (handle == 0) {
749 DRM_ERROR("Invalid UVD handle!\n");
750 return -EINVAL;
751 }
752
753 switch (msg_type) {
754 case 0:
755
756 amdgpu_bo_kunmap(bo);
757
758
759 for (i = 0; i < adev->uvd.max_handles; ++i) {
760 if (atomic_read(&adev->uvd.handles[i]) == handle) {
761 DRM_ERROR(")Handle 0x%x already in use!\n",
762 handle);
763 return -EINVAL;
764 }
765
766 if (!atomic_cmpxchg(&adev->uvd.handles[i], 0, handle)) {
767 adev->uvd.filp[i] = ctx->parser->filp;
768 return 0;
769 }
770 }
771
772 DRM_ERROR("No more free UVD handles!\n");
773 return -ENOSPC;
774
775 case 1:
776
777 r = amdgpu_uvd_cs_msg_decode(adev, msg, ctx->buf_sizes);
778 amdgpu_bo_kunmap(bo);
779 if (r)
780 return r;
781
782
783 for (i = 0; i < adev->uvd.max_handles; ++i) {
784 if (atomic_read(&adev->uvd.handles[i]) == handle) {
785 if (adev->uvd.filp[i] != ctx->parser->filp) {
786 DRM_ERROR("UVD handle collision detected!\n");
787 return -EINVAL;
788 }
789 return 0;
790 }
791 }
792
793 DRM_ERROR("Invalid UVD handle 0x%x!\n", handle);
794 return -ENOENT;
795
796 case 2:
797
798 for (i = 0; i < adev->uvd.max_handles; ++i)
799 atomic_cmpxchg(&adev->uvd.handles[i], handle, 0);
800 amdgpu_bo_kunmap(bo);
801 return 0;
802
803 default:
804 DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type);
805 return -EINVAL;
806 }
807 BUG();
808 return -EINVAL;
809}
810
811
812
813
814
815
816
817
818static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx)
819{
820 struct amdgpu_bo_va_mapping *mapping;
821 struct amdgpu_bo *bo;
822 uint32_t cmd;
823 uint64_t start, end;
824 uint64_t addr = amdgpu_uvd_get_addr_from_ctx(ctx);
825 int r;
826
827 r = amdgpu_cs_find_mapping(ctx->parser, addr, &bo, &mapping);
828 if (r) {
829 DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
830 return r;
831 }
832
833 start = amdgpu_bo_gpu_offset(bo);
834
835 end = (mapping->last + 1 - mapping->start);
836 end = end * AMDGPU_GPU_PAGE_SIZE + start;
837
838 addr -= mapping->start * AMDGPU_GPU_PAGE_SIZE;
839 start += addr;
840
841 amdgpu_set_ib_value(ctx->parser, ctx->ib_idx, ctx->data0,
842 lower_32_bits(start));
843 amdgpu_set_ib_value(ctx->parser, ctx->ib_idx, ctx->data1,
844 upper_32_bits(start));
845
846 cmd = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->idx) >> 1;
847 if (cmd < 0x4) {
848 if ((end - start) < ctx->buf_sizes[cmd]) {
849 DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd,
850 (unsigned)(end - start),
851 ctx->buf_sizes[cmd]);
852 return -EINVAL;
853 }
854
855 } else if (cmd == 0x206) {
856 if ((end - start) < ctx->buf_sizes[4]) {
857 DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd,
858 (unsigned)(end - start),
859 ctx->buf_sizes[4]);
860 return -EINVAL;
861 }
862 } else if ((cmd != 0x100) && (cmd != 0x204)) {
863 DRM_ERROR("invalid UVD command %X!\n", cmd);
864 return -EINVAL;
865 }
866
867 if (!ctx->parser->adev->uvd.address_64_bit) {
868 if ((start >> 28) != ((end - 1) >> 28)) {
869 DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n",
870 start, end);
871 return -EINVAL;
872 }
873
874 if ((cmd == 0 || cmd == 0x3) &&
875 (start >> 28) != (ctx->parser->adev->uvd.inst->gpu_addr >> 28)) {
876 DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
877 start, end);
878 return -EINVAL;
879 }
880 }
881
882 if (cmd == 0) {
883 ctx->has_msg_cmd = true;
884 r = amdgpu_uvd_cs_msg(ctx, bo, addr);
885 if (r)
886 return r;
887 } else if (!ctx->has_msg_cmd) {
888 DRM_ERROR("Message needed before other commands are send!\n");
889 return -EINVAL;
890 }
891
892 return 0;
893}
894
895
896
897
898
899
900
901
902
903static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx,
904 int (*cb)(struct amdgpu_uvd_cs_ctx *ctx))
905{
906 struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx];
907 int i, r;
908
909 ctx->idx++;
910 for (i = 0; i <= ctx->count; ++i) {
911 unsigned reg = ctx->reg + i;
912
913 if (ctx->idx >= ib->length_dw) {
914 DRM_ERROR("Register command after end of CS!\n");
915 return -EINVAL;
916 }
917
918 switch (reg) {
919 case mmUVD_GPCOM_VCPU_DATA0:
920 ctx->data0 = ctx->idx;
921 break;
922 case mmUVD_GPCOM_VCPU_DATA1:
923 ctx->data1 = ctx->idx;
924 break;
925 case mmUVD_GPCOM_VCPU_CMD:
926 r = cb(ctx);
927 if (r)
928 return r;
929 break;
930 case mmUVD_ENGINE_CNTL:
931 case mmUVD_NO_OP:
932 break;
933 default:
934 DRM_ERROR("Invalid reg 0x%X!\n", reg);
935 return -EINVAL;
936 }
937 ctx->idx++;
938 }
939 return 0;
940}
941
942
943
944
945
946
947
948
949
950static int amdgpu_uvd_cs_packets(struct amdgpu_uvd_cs_ctx *ctx,
951 int (*cb)(struct amdgpu_uvd_cs_ctx *ctx))
952{
953 struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx];
954 int r;
955
956 for (ctx->idx = 0 ; ctx->idx < ib->length_dw; ) {
957 uint32_t cmd = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->idx);
958 unsigned type = CP_PACKET_GET_TYPE(cmd);
959 switch (type) {
960 case PACKET_TYPE0:
961 ctx->reg = CP_PACKET0_GET_REG(cmd);
962 ctx->count = CP_PACKET_GET_COUNT(cmd);
963 r = amdgpu_uvd_cs_reg(ctx, cb);
964 if (r)
965 return r;
966 break;
967 case PACKET_TYPE2:
968 ++ctx->idx;
969 break;
970 default:
971 DRM_ERROR("Unknown packet type %d !\n", type);
972 return -EINVAL;
973 }
974 }
975 return 0;
976}
977
978
979
980
981
982
983
984
985int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
986{
987 struct amdgpu_uvd_cs_ctx ctx = {};
988 unsigned buf_sizes[] = {
989 [0x00000000] = 2048,
990 [0x00000001] = 0xFFFFFFFF,
991 [0x00000002] = 0xFFFFFFFF,
992 [0x00000003] = 2048,
993 [0x00000004] = 0xFFFFFFFF,
994 };
995 struct amdgpu_ib *ib = &parser->job->ibs[ib_idx];
996 int r;
997
998 parser->job->vm = NULL;
999 ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
1000
1001 if (ib->length_dw % 16) {
1002 DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n",
1003 ib->length_dw);
1004 return -EINVAL;
1005 }
1006
1007 ctx.parser = parser;
1008 ctx.buf_sizes = buf_sizes;
1009 ctx.ib_idx = ib_idx;
1010
1011
1012 if (!parser->adev->uvd.address_64_bit) {
1013
1014 r = amdgpu_uvd_cs_packets(&ctx, amdgpu_uvd_cs_pass1);
1015 if (r)
1016 return r;
1017 }
1018
1019
1020 r = amdgpu_uvd_cs_packets(&ctx, amdgpu_uvd_cs_pass2);
1021 if (r)
1022 return r;
1023
1024 if (!ctx.has_msg_cmd) {
1025 DRM_ERROR("UVD-IBs need a msg command!\n");
1026 return -EINVAL;
1027 }
1028
1029 return 0;
1030}
1031
1032static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
1033 bool direct, struct dma_fence **fence)
1034{
1035 struct amdgpu_device *adev = ring->adev;
1036 struct dma_fence *f = NULL;
1037 struct amdgpu_job *job;
1038 struct amdgpu_ib *ib;
1039 uint32_t data[4];
1040 uint64_t addr;
1041 long r;
1042 int i;
1043 unsigned offset_idx = 0;
1044 unsigned offset[3] = { UVD_BASE_SI, 0, 0 };
1045
1046 amdgpu_bo_kunmap(bo);
1047 amdgpu_bo_unpin(bo);
1048
1049 if (!ring->adev->uvd.address_64_bit) {
1050 struct ttm_operation_ctx ctx = { true, false };
1051
1052 amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_VRAM);
1053 amdgpu_uvd_force_into_uvd_segment(bo);
1054 r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1055 if (r)
1056 goto err;
1057 }
1058
1059 r = amdgpu_job_alloc_with_ib(adev, 64, direct ? AMDGPU_IB_POOL_DIRECT :
1060 AMDGPU_IB_POOL_DELAYED, &job);
1061 if (r)
1062 goto err;
1063
1064 if (adev->asic_type >= CHIP_VEGA10) {
1065 offset_idx = 1 + ring->me;
1066 offset[1] = adev->reg_offset[UVD_HWIP][0][1];
1067 offset[2] = adev->reg_offset[UVD_HWIP][1][1];
1068 }
1069
1070 data[0] = PACKET0(offset[offset_idx] + UVD_GPCOM_VCPU_DATA0, 0);
1071 data[1] = PACKET0(offset[offset_idx] + UVD_GPCOM_VCPU_DATA1, 0);
1072 data[2] = PACKET0(offset[offset_idx] + UVD_GPCOM_VCPU_CMD, 0);
1073 data[3] = PACKET0(offset[offset_idx] + UVD_NO_OP, 0);
1074
1075 ib = &job->ibs[0];
1076 addr = amdgpu_bo_gpu_offset(bo);
1077 ib->ptr[0] = data[0];
1078 ib->ptr[1] = addr;
1079 ib->ptr[2] = data[1];
1080 ib->ptr[3] = addr >> 32;
1081 ib->ptr[4] = data[2];
1082 ib->ptr[5] = 0;
1083 for (i = 6; i < 16; i += 2) {
1084 ib->ptr[i] = data[3];
1085 ib->ptr[i+1] = 0;
1086 }
1087 ib->length_dw = 16;
1088
1089 if (direct) {
1090 r = dma_resv_wait_timeout_rcu(bo->tbo.base.resv,
1091 true, false,
1092 msecs_to_jiffies(10));
1093 if (r == 0)
1094 r = -ETIMEDOUT;
1095 if (r < 0)
1096 goto err_free;
1097
1098 r = amdgpu_job_submit_direct(job, ring, &f);
1099 if (r)
1100 goto err_free;
1101 } else {
1102 r = amdgpu_sync_resv(adev, &job->sync, bo->tbo.base.resv,
1103 AMDGPU_SYNC_ALWAYS,
1104 AMDGPU_FENCE_OWNER_UNDEFINED);
1105 if (r)
1106 goto err_free;
1107
1108 r = amdgpu_job_submit(job, &adev->uvd.entity,
1109 AMDGPU_FENCE_OWNER_UNDEFINED, &f);
1110 if (r)
1111 goto err_free;
1112 }
1113
1114 amdgpu_bo_fence(bo, f, false);
1115 amdgpu_bo_unreserve(bo);
1116 amdgpu_bo_unref(&bo);
1117
1118 if (fence)
1119 *fence = dma_fence_get(f);
1120 dma_fence_put(f);
1121
1122 return 0;
1123
1124err_free:
1125 amdgpu_job_free(job);
1126
1127err:
1128 amdgpu_bo_unreserve(bo);
1129 amdgpu_bo_unref(&bo);
1130 return r;
1131}
1132
1133
1134
1135
1136int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
1137 struct dma_fence **fence)
1138{
1139 struct amdgpu_device *adev = ring->adev;
1140 struct amdgpu_bo *bo = NULL;
1141 uint32_t *msg;
1142 int r, i;
1143
1144 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
1145 AMDGPU_GEM_DOMAIN_VRAM,
1146 &bo, NULL, (void **)&msg);
1147 if (r)
1148 return r;
1149
1150
1151 msg[0] = cpu_to_le32(0x00000de4);
1152 msg[1] = cpu_to_le32(0x00000000);
1153 msg[2] = cpu_to_le32(handle);
1154 msg[3] = cpu_to_le32(0x00000000);
1155 msg[4] = cpu_to_le32(0x00000000);
1156 msg[5] = cpu_to_le32(0x00000000);
1157 msg[6] = cpu_to_le32(0x00000000);
1158 msg[7] = cpu_to_le32(0x00000780);
1159 msg[8] = cpu_to_le32(0x00000440);
1160 msg[9] = cpu_to_le32(0x00000000);
1161 msg[10] = cpu_to_le32(0x01b37000);
1162 for (i = 11; i < 1024; ++i)
1163 msg[i] = cpu_to_le32(0x0);
1164
1165 return amdgpu_uvd_send_msg(ring, bo, true, fence);
1166}
1167
1168int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
1169 bool direct, struct dma_fence **fence)
1170{
1171 struct amdgpu_device *adev = ring->adev;
1172 struct amdgpu_bo *bo = NULL;
1173 uint32_t *msg;
1174 int r, i;
1175
1176 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
1177 AMDGPU_GEM_DOMAIN_VRAM,
1178 &bo, NULL, (void **)&msg);
1179 if (r)
1180 return r;
1181
1182
1183 msg[0] = cpu_to_le32(0x00000de4);
1184 msg[1] = cpu_to_le32(0x00000002);
1185 msg[2] = cpu_to_le32(handle);
1186 msg[3] = cpu_to_le32(0x00000000);
1187 for (i = 4; i < 1024; ++i)
1188 msg[i] = cpu_to_le32(0x0);
1189
1190 return amdgpu_uvd_send_msg(ring, bo, direct, fence);
1191}
1192
1193static void amdgpu_uvd_idle_work_handler(struct work_struct *work)
1194{
1195 struct amdgpu_device *adev =
1196 container_of(work, struct amdgpu_device, uvd.idle_work.work);
1197 unsigned fences = 0, i, j;
1198
1199 for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
1200 if (adev->uvd.harvest_config & (1 << i))
1201 continue;
1202 fences += amdgpu_fence_count_emitted(&adev->uvd.inst[i].ring);
1203 for (j = 0; j < adev->uvd.num_enc_rings; ++j) {
1204 fences += amdgpu_fence_count_emitted(&adev->uvd.inst[i].ring_enc[j]);
1205 }
1206 }
1207
1208 if (fences == 0) {
1209 if (adev->pm.dpm_enabled) {
1210 amdgpu_dpm_enable_uvd(adev, false);
1211 } else {
1212 amdgpu_asic_set_uvd_clocks(adev, 0, 0);
1213
1214 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
1215 AMD_PG_STATE_GATE);
1216 amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
1217 AMD_CG_STATE_GATE);
1218 }
1219 } else {
1220 schedule_delayed_work(&adev->uvd.idle_work, UVD_IDLE_TIMEOUT);
1221 }
1222}
1223
1224void amdgpu_uvd_ring_begin_use(struct amdgpu_ring *ring)
1225{
1226 struct amdgpu_device *adev = ring->adev;
1227 bool set_clocks;
1228
1229 if (amdgpu_sriov_vf(adev))
1230 return;
1231
1232 set_clocks = !cancel_delayed_work_sync(&adev->uvd.idle_work);
1233 if (set_clocks) {
1234 if (adev->pm.dpm_enabled) {
1235 amdgpu_dpm_enable_uvd(adev, true);
1236 } else {
1237 amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
1238 amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
1239 AMD_CG_STATE_UNGATE);
1240 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
1241 AMD_PG_STATE_UNGATE);
1242 }
1243 }
1244}
1245
1246void amdgpu_uvd_ring_end_use(struct amdgpu_ring *ring)
1247{
1248 if (!amdgpu_sriov_vf(ring->adev))
1249 schedule_delayed_work(&ring->adev->uvd.idle_work, UVD_IDLE_TIMEOUT);
1250}
1251
1252
1253
1254
1255
1256
1257
1258
1259int amdgpu_uvd_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1260{
1261 struct dma_fence *fence;
1262 long r;
1263
1264 r = amdgpu_uvd_get_create_msg(ring, 1, NULL);
1265 if (r)
1266 goto error;
1267
1268 r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence);
1269 if (r)
1270 goto error;
1271
1272 r = dma_fence_wait_timeout(fence, false, timeout);
1273 if (r == 0)
1274 r = -ETIMEDOUT;
1275 else if (r > 0)
1276 r = 0;
1277
1278 dma_fence_put(fence);
1279
1280error:
1281 return r;
1282}
1283
1284
1285
1286
1287
1288
1289
1290
1291uint32_t amdgpu_uvd_used_handles(struct amdgpu_device *adev)
1292{
1293 unsigned i;
1294 uint32_t used_handles = 0;
1295
1296 for (i = 0; i < adev->uvd.max_handles; ++i) {
1297
1298
1299
1300
1301
1302 if (atomic_read(&adev->uvd.handles[i]))
1303 used_handles++;
1304 }
1305
1306 return used_handles;
1307}
1308
