Loading...
1/*
2 * Copyright 2016 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#include <linux/firmware.h>
28#include <linux/module.h>
29#include <linux/pci.h>
30
31#include <drm/drm.h>
32
33#include "amdgpu.h"
34#include "amdgpu_pm.h"
35#include "amdgpu_vcn.h"
36#include "soc15d.h"
37#include "soc15_common.h"
38
39#include "vcn/vcn_1_0_offset.h"
40#include "vcn/vcn_1_0_sh_mask.h"
41
42/* 1 second timeout */
43#define VCN_IDLE_TIMEOUT msecs_to_jiffies(1000)
44
45/* Firmware Names */
46#define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
47#define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
48#define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
49#define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
50#define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
51#define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
52#define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
53#define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
54
55MODULE_FIRMWARE(FIRMWARE_RAVEN);
56MODULE_FIRMWARE(FIRMWARE_PICASSO);
57MODULE_FIRMWARE(FIRMWARE_RAVEN2);
58MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
59MODULE_FIRMWARE(FIRMWARE_RENOIR);
60MODULE_FIRMWARE(FIRMWARE_NAVI10);
61MODULE_FIRMWARE(FIRMWARE_NAVI14);
62MODULE_FIRMWARE(FIRMWARE_NAVI12);
63
64static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
65
66int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
67{
68 unsigned long bo_size;
69 const char *fw_name;
70 const struct common_firmware_header *hdr;
71 unsigned char fw_check;
72 int i, r;
73
74 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
75
76 switch (adev->asic_type) {
77 case CHIP_RAVEN:
78 if (adev->rev_id >= 8)
79 fw_name = FIRMWARE_RAVEN2;
80 else if (adev->pdev->device == 0x15d8)
81 fw_name = FIRMWARE_PICASSO;
82 else
83 fw_name = FIRMWARE_RAVEN;
84 break;
85 case CHIP_ARCTURUS:
86 fw_name = FIRMWARE_ARCTURUS;
87 break;
88 case CHIP_RENOIR:
89 fw_name = FIRMWARE_RENOIR;
90 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
91 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
92 adev->vcn.indirect_sram = true;
93 break;
94 case CHIP_NAVI10:
95 fw_name = FIRMWARE_NAVI10;
96 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
97 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
98 adev->vcn.indirect_sram = true;
99 break;
100 case CHIP_NAVI14:
101 fw_name = FIRMWARE_NAVI14;
102 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
103 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
104 adev->vcn.indirect_sram = true;
105 break;
106 case CHIP_NAVI12:
107 fw_name = FIRMWARE_NAVI12;
108 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
109 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
110 adev->vcn.indirect_sram = true;
111 break;
112 default:
113 return -EINVAL;
114 }
115
116 r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
117 if (r) {
118 dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
119 fw_name);
120 return r;
121 }
122
123 r = amdgpu_ucode_validate(adev->vcn.fw);
124 if (r) {
125 dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
126 fw_name);
127 release_firmware(adev->vcn.fw);
128 adev->vcn.fw = NULL;
129 return r;
130 }
131
132 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
133 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
134
135 /* Bit 20-23, it is encode major and non-zero for new naming convention.
136 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
137 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
138 * is zero in old naming convention, this field is always zero so far.
139 * These four bits are used to tell which naming convention is present.
140 */
141 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
142 if (fw_check) {
143 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
144
145 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
146 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
147 enc_major = fw_check;
148 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
149 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
150 DRM_INFO("Found VCN firmware Version ENC: %hu.%hu DEC: %hu VEP: %hu Revision: %hu\n",
151 enc_major, enc_minor, dec_ver, vep, fw_rev);
152 } else {
153 unsigned int version_major, version_minor, family_id;
154
155 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
156 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
157 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
158 DRM_INFO("Found VCN firmware Version: %hu.%hu Family ID: %hu\n",
159 version_major, version_minor, family_id);
160 }
161
162 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
163 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
164 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
165
166 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
167 if (adev->vcn.harvest_config & (1 << i))
168 continue;
169
170 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
171 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
172 &adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
173 if (r) {
174 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
175 return r;
176 }
177 }
178
179 if (adev->vcn.indirect_sram) {
180 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
181 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.dpg_sram_bo,
182 &adev->vcn.dpg_sram_gpu_addr, &adev->vcn.dpg_sram_cpu_addr);
183 if (r) {
184 dev_err(adev->dev, "(%d) failed to allocate DPG bo\n", r);
185 return r;
186 }
187 }
188
189 return 0;
190}
191
192int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
193{
194 int i, j;
195
196 if (adev->vcn.indirect_sram) {
197 amdgpu_bo_free_kernel(&adev->vcn.dpg_sram_bo,
198 &adev->vcn.dpg_sram_gpu_addr,
199 (void **)&adev->vcn.dpg_sram_cpu_addr);
200 }
201
202 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
203 if (adev->vcn.harvest_config & (1 << j))
204 continue;
205 kvfree(adev->vcn.inst[j].saved_bo);
206
207 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
208 &adev->vcn.inst[j].gpu_addr,
209 (void **)&adev->vcn.inst[j].cpu_addr);
210
211 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
212
213 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
214 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
215
216 amdgpu_ring_fini(&adev->vcn.inst[j].ring_jpeg);
217 }
218
219 release_firmware(adev->vcn.fw);
220
221 return 0;
222}
223
224int amdgpu_vcn_suspend(struct amdgpu_device *adev)
225{
226 unsigned size;
227 void *ptr;
228 int i;
229
230 cancel_delayed_work_sync(&adev->vcn.idle_work);
231
232 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
233 if (adev->vcn.harvest_config & (1 << i))
234 continue;
235 if (adev->vcn.inst[i].vcpu_bo == NULL)
236 return 0;
237
238 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
239 ptr = adev->vcn.inst[i].cpu_addr;
240
241 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
242 if (!adev->vcn.inst[i].saved_bo)
243 return -ENOMEM;
244
245 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
246 }
247 return 0;
248}
249
250int amdgpu_vcn_resume(struct amdgpu_device *adev)
251{
252 unsigned size;
253 void *ptr;
254 int i;
255
256 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
257 if (adev->vcn.harvest_config & (1 << i))
258 continue;
259 if (adev->vcn.inst[i].vcpu_bo == NULL)
260 return -EINVAL;
261
262 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
263 ptr = adev->vcn.inst[i].cpu_addr;
264
265 if (adev->vcn.inst[i].saved_bo != NULL) {
266 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
267 kvfree(adev->vcn.inst[i].saved_bo);
268 adev->vcn.inst[i].saved_bo = NULL;
269 } else {
270 const struct common_firmware_header *hdr;
271 unsigned offset;
272
273 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
274 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
275 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
276 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
277 le32_to_cpu(hdr->ucode_size_bytes));
278 size -= le32_to_cpu(hdr->ucode_size_bytes);
279 ptr += le32_to_cpu(hdr->ucode_size_bytes);
280 }
281 memset_io(ptr, 0, size);
282 }
283 }
284 return 0;
285}
286
287static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
288{
289 struct amdgpu_device *adev =
290 container_of(work, struct amdgpu_device, vcn.idle_work.work);
291 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
292 unsigned int i, j;
293
294 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
295 if (adev->vcn.harvest_config & (1 << j))
296 continue;
297 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
298 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
299 }
300
301 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
302 struct dpg_pause_state new_state;
303
304 if (fence[j])
305 new_state.fw_based = VCN_DPG_STATE__PAUSE;
306 else
307 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
308
309 if (amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_jpeg))
310 new_state.jpeg = VCN_DPG_STATE__PAUSE;
311 else
312 new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
313
314 adev->vcn.pause_dpg_mode(adev, &new_state);
315 }
316
317 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_jpeg);
318 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
319 fences += fence[j];
320 }
321
322 if (fences == 0) {
323 amdgpu_gfx_off_ctrl(adev, true);
324 if (adev->asic_type < CHIP_ARCTURUS && adev->pm.dpm_enabled)
325 amdgpu_dpm_enable_uvd(adev, false);
326 else
327 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
328 AMD_PG_STATE_GATE);
329 } else {
330 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
331 }
332}
333
334void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
335{
336 struct amdgpu_device *adev = ring->adev;
337 bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work);
338
339 if (set_clocks) {
340 amdgpu_gfx_off_ctrl(adev, false);
341 if (adev->asic_type < CHIP_ARCTURUS && adev->pm.dpm_enabled)
342 amdgpu_dpm_enable_uvd(adev, true);
343 else
344 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
345 AMD_PG_STATE_UNGATE);
346 }
347
348 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
349 struct dpg_pause_state new_state;
350 unsigned int fences = 0;
351 unsigned int i;
352
353 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
354 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
355 }
356 if (fences)
357 new_state.fw_based = VCN_DPG_STATE__PAUSE;
358 else
359 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
360
361 if (amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_jpeg))
362 new_state.jpeg = VCN_DPG_STATE__PAUSE;
363 else
364 new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
365
366 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
367 new_state.fw_based = VCN_DPG_STATE__PAUSE;
368 else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG)
369 new_state.jpeg = VCN_DPG_STATE__PAUSE;
370
371 adev->vcn.pause_dpg_mode(adev, &new_state);
372 }
373}
374
375void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
376{
377 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
378}
379
380int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
381{
382 struct amdgpu_device *adev = ring->adev;
383 uint32_t tmp = 0;
384 unsigned i;
385 int r;
386
387 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
388 r = amdgpu_ring_alloc(ring, 3);
389 if (r)
390 return r;
391 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
392 amdgpu_ring_write(ring, 0xDEADBEEF);
393 amdgpu_ring_commit(ring);
394 for (i = 0; i < adev->usec_timeout; i++) {
395 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
396 if (tmp == 0xDEADBEEF)
397 break;
398 udelay(1);
399 }
400
401 if (i >= adev->usec_timeout)
402 r = -ETIMEDOUT;
403
404 return r;
405}
406
407static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
408 struct amdgpu_bo *bo,
409 struct dma_fence **fence)
410{
411 struct amdgpu_device *adev = ring->adev;
412 struct dma_fence *f = NULL;
413 struct amdgpu_job *job;
414 struct amdgpu_ib *ib;
415 uint64_t addr;
416 int i, r;
417
418 r = amdgpu_job_alloc_with_ib(adev, 64, &job);
419 if (r)
420 goto err;
421
422 ib = &job->ibs[0];
423 addr = amdgpu_bo_gpu_offset(bo);
424 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
425 ib->ptr[1] = addr;
426 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
427 ib->ptr[3] = addr >> 32;
428 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
429 ib->ptr[5] = 0;
430 for (i = 6; i < 16; i += 2) {
431 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
432 ib->ptr[i+1] = 0;
433 }
434 ib->length_dw = 16;
435
436 r = amdgpu_job_submit_direct(job, ring, &f);
437 if (r)
438 goto err_free;
439
440 amdgpu_bo_fence(bo, f, false);
441 amdgpu_bo_unreserve(bo);
442 amdgpu_bo_unref(&bo);
443
444 if (fence)
445 *fence = dma_fence_get(f);
446 dma_fence_put(f);
447
448 return 0;
449
450err_free:
451 amdgpu_job_free(job);
452
453err:
454 amdgpu_bo_unreserve(bo);
455 amdgpu_bo_unref(&bo);
456 return r;
457}
458
459static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
460 struct dma_fence **fence)
461{
462 struct amdgpu_device *adev = ring->adev;
463 struct amdgpu_bo *bo = NULL;
464 uint32_t *msg;
465 int r, i;
466
467 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
468 AMDGPU_GEM_DOMAIN_VRAM,
469 &bo, NULL, (void **)&msg);
470 if (r)
471 return r;
472
473 msg[0] = cpu_to_le32(0x00000028);
474 msg[1] = cpu_to_le32(0x00000038);
475 msg[2] = cpu_to_le32(0x00000001);
476 msg[3] = cpu_to_le32(0x00000000);
477 msg[4] = cpu_to_le32(handle);
478 msg[5] = cpu_to_le32(0x00000000);
479 msg[6] = cpu_to_le32(0x00000001);
480 msg[7] = cpu_to_le32(0x00000028);
481 msg[8] = cpu_to_le32(0x00000010);
482 msg[9] = cpu_to_le32(0x00000000);
483 msg[10] = cpu_to_le32(0x00000007);
484 msg[11] = cpu_to_le32(0x00000000);
485 msg[12] = cpu_to_le32(0x00000780);
486 msg[13] = cpu_to_le32(0x00000440);
487 for (i = 14; i < 1024; ++i)
488 msg[i] = cpu_to_le32(0x0);
489
490 return amdgpu_vcn_dec_send_msg(ring, bo, fence);
491}
492
493static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
494 struct dma_fence **fence)
495{
496 struct amdgpu_device *adev = ring->adev;
497 struct amdgpu_bo *bo = NULL;
498 uint32_t *msg;
499 int r, i;
500
501 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
502 AMDGPU_GEM_DOMAIN_VRAM,
503 &bo, NULL, (void **)&msg);
504 if (r)
505 return r;
506
507 msg[0] = cpu_to_le32(0x00000028);
508 msg[1] = cpu_to_le32(0x00000018);
509 msg[2] = cpu_to_le32(0x00000000);
510 msg[3] = cpu_to_le32(0x00000002);
511 msg[4] = cpu_to_le32(handle);
512 msg[5] = cpu_to_le32(0x00000000);
513 for (i = 6; i < 1024; ++i)
514 msg[i] = cpu_to_le32(0x0);
515
516 return amdgpu_vcn_dec_send_msg(ring, bo, fence);
517}
518
519int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
520{
521 struct dma_fence *fence;
522 long r;
523
524 r = amdgpu_vcn_dec_get_create_msg(ring, 1, NULL);
525 if (r)
526 goto error;
527
528 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &fence);
529 if (r)
530 goto error;
531
532 r = dma_fence_wait_timeout(fence, false, timeout);
533 if (r == 0)
534 r = -ETIMEDOUT;
535 else if (r > 0)
536 r = 0;
537
538 dma_fence_put(fence);
539error:
540 return r;
541}
542
543int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
544{
545 struct amdgpu_device *adev = ring->adev;
546 uint32_t rptr;
547 unsigned i;
548 int r;
549
550 r = amdgpu_ring_alloc(ring, 16);
551 if (r)
552 return r;
553
554 rptr = amdgpu_ring_get_rptr(ring);
555
556 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
557 amdgpu_ring_commit(ring);
558
559 for (i = 0; i < adev->usec_timeout; i++) {
560 if (amdgpu_ring_get_rptr(ring) != rptr)
561 break;
562 udelay(1);
563 }
564
565 if (i >= adev->usec_timeout)
566 r = -ETIMEDOUT;
567
568 return r;
569}
570
571static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
572 struct amdgpu_bo *bo,
573 struct dma_fence **fence)
574{
575 const unsigned ib_size_dw = 16;
576 struct amdgpu_job *job;
577 struct amdgpu_ib *ib;
578 struct dma_fence *f = NULL;
579 uint64_t addr;
580 int i, r;
581
582 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
583 if (r)
584 return r;
585
586 ib = &job->ibs[0];
587 addr = amdgpu_bo_gpu_offset(bo);
588
589 ib->length_dw = 0;
590 ib->ptr[ib->length_dw++] = 0x00000018;
591 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
592 ib->ptr[ib->length_dw++] = handle;
593 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
594 ib->ptr[ib->length_dw++] = addr;
595 ib->ptr[ib->length_dw++] = 0x0000000b;
596
597 ib->ptr[ib->length_dw++] = 0x00000014;
598 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
599 ib->ptr[ib->length_dw++] = 0x0000001c;
600 ib->ptr[ib->length_dw++] = 0x00000000;
601 ib->ptr[ib->length_dw++] = 0x00000000;
602
603 ib->ptr[ib->length_dw++] = 0x00000008;
604 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
605
606 for (i = ib->length_dw; i < ib_size_dw; ++i)
607 ib->ptr[i] = 0x0;
608
609 r = amdgpu_job_submit_direct(job, ring, &f);
610 if (r)
611 goto err;
612
613 if (fence)
614 *fence = dma_fence_get(f);
615 dma_fence_put(f);
616
617 return 0;
618
619err:
620 amdgpu_job_free(job);
621 return r;
622}
623
624static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
625 struct amdgpu_bo *bo,
626 struct dma_fence **fence)
627{
628 const unsigned ib_size_dw = 16;
629 struct amdgpu_job *job;
630 struct amdgpu_ib *ib;
631 struct dma_fence *f = NULL;
632 uint64_t addr;
633 int i, r;
634
635 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
636 if (r)
637 return r;
638
639 ib = &job->ibs[0];
640 addr = amdgpu_bo_gpu_offset(bo);
641
642 ib->length_dw = 0;
643 ib->ptr[ib->length_dw++] = 0x00000018;
644 ib->ptr[ib->length_dw++] = 0x00000001;
645 ib->ptr[ib->length_dw++] = handle;
646 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
647 ib->ptr[ib->length_dw++] = addr;
648 ib->ptr[ib->length_dw++] = 0x0000000b;
649
650 ib->ptr[ib->length_dw++] = 0x00000014;
651 ib->ptr[ib->length_dw++] = 0x00000002;
652 ib->ptr[ib->length_dw++] = 0x0000001c;
653 ib->ptr[ib->length_dw++] = 0x00000000;
654 ib->ptr[ib->length_dw++] = 0x00000000;
655
656 ib->ptr[ib->length_dw++] = 0x00000008;
657 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
658
659 for (i = ib->length_dw; i < ib_size_dw; ++i)
660 ib->ptr[i] = 0x0;
661
662 r = amdgpu_job_submit_direct(job, ring, &f);
663 if (r)
664 goto err;
665
666 if (fence)
667 *fence = dma_fence_get(f);
668 dma_fence_put(f);
669
670 return 0;
671
672err:
673 amdgpu_job_free(job);
674 return r;
675}
676
677int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
678{
679 struct dma_fence *fence = NULL;
680 struct amdgpu_bo *bo = NULL;
681 long r;
682
683 r = amdgpu_bo_create_reserved(ring->adev, 128 * 1024, PAGE_SIZE,
684 AMDGPU_GEM_DOMAIN_VRAM,
685 &bo, NULL, NULL);
686 if (r)
687 return r;
688
689 r = amdgpu_vcn_enc_get_create_msg(ring, 1, bo, NULL);
690 if (r)
691 goto error;
692
693 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, bo, &fence);
694 if (r)
695 goto error;
696
697 r = dma_fence_wait_timeout(fence, false, timeout);
698 if (r == 0)
699 r = -ETIMEDOUT;
700 else if (r > 0)
701 r = 0;
702
703error:
704 dma_fence_put(fence);
705 amdgpu_bo_unreserve(bo);
706 amdgpu_bo_unref(&bo);
707 return r;
708}
709
710int amdgpu_vcn_jpeg_ring_test_ring(struct amdgpu_ring *ring)
711{
712 struct amdgpu_device *adev = ring->adev;
713 uint32_t tmp = 0;
714 unsigned i;
715 int r;
716
717 WREG32(adev->vcn.inst[ring->me].external.jpeg_pitch, 0xCAFEDEAD);
718 r = amdgpu_ring_alloc(ring, 3);
719 if (r)
720 return r;
721
722 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.jpeg_pitch, 0));
723 amdgpu_ring_write(ring, 0xDEADBEEF);
724 amdgpu_ring_commit(ring);
725
726 for (i = 0; i < adev->usec_timeout; i++) {
727 tmp = RREG32(adev->vcn.inst[ring->me].external.jpeg_pitch);
728 if (tmp == 0xDEADBEEF)
729 break;
730 udelay(1);
731 }
732
733 if (i >= adev->usec_timeout)
734 r = -ETIMEDOUT;
735
736 return r;
737}
738
739static int amdgpu_vcn_jpeg_set_reg(struct amdgpu_ring *ring, uint32_t handle,
740 struct dma_fence **fence)
741{
742 struct amdgpu_device *adev = ring->adev;
743 struct amdgpu_job *job;
744 struct amdgpu_ib *ib;
745 struct dma_fence *f = NULL;
746 const unsigned ib_size_dw = 16;
747 int i, r;
748
749 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
750 if (r)
751 return r;
752
753 ib = &job->ibs[0];
754
755 ib->ptr[0] = PACKETJ(adev->vcn.internal.jpeg_pitch, 0, 0, PACKETJ_TYPE0);
756 ib->ptr[1] = 0xDEADBEEF;
757 for (i = 2; i < 16; i += 2) {
758 ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6);
759 ib->ptr[i+1] = 0;
760 }
761 ib->length_dw = 16;
762
763 r = amdgpu_job_submit_direct(job, ring, &f);
764 if (r)
765 goto err;
766
767 if (fence)
768 *fence = dma_fence_get(f);
769 dma_fence_put(f);
770
771 return 0;
772
773err:
774 amdgpu_job_free(job);
775 return r;
776}
777
778int amdgpu_vcn_jpeg_ring_test_ib(struct amdgpu_ring *ring, long timeout)
779{
780 struct amdgpu_device *adev = ring->adev;
781 uint32_t tmp = 0;
782 unsigned i;
783 struct dma_fence *fence = NULL;
784 long r = 0;
785
786 r = amdgpu_vcn_jpeg_set_reg(ring, 1, &fence);
787 if (r)
788 goto error;
789
790 r = dma_fence_wait_timeout(fence, false, timeout);
791 if (r == 0) {
792 r = -ETIMEDOUT;
793 goto error;
794 } else if (r < 0) {
795 goto error;
796 } else {
797 r = 0;
798 }
799
800 for (i = 0; i < adev->usec_timeout; i++) {
801 tmp = RREG32(adev->vcn.inst[ring->me].external.jpeg_pitch);
802 if (tmp == 0xDEADBEEF)
803 break;
804 udelay(1);
805 }
806
807 if (i >= adev->usec_timeout)
808 r = -ETIMEDOUT;
809
810 dma_fence_put(fence);
811error:
812 return r;
813}
1/*
2 * Copyright 2016 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#include <linux/firmware.h>
28#include <linux/module.h>
29#include <linux/pci.h>
30
31#include "amdgpu.h"
32#include "amdgpu_pm.h"
33#include "amdgpu_vcn.h"
34#include "soc15d.h"
35
36/* Firmware Names */
37#define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
38#define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
39#define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
40#define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
41#define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
42#define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
43#define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
44#define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
45#define FIRMWARE_SIENNA_CICHLID "amdgpu/sienna_cichlid_vcn.bin"
46#define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin"
47
48MODULE_FIRMWARE(FIRMWARE_RAVEN);
49MODULE_FIRMWARE(FIRMWARE_PICASSO);
50MODULE_FIRMWARE(FIRMWARE_RAVEN2);
51MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
52MODULE_FIRMWARE(FIRMWARE_RENOIR);
53MODULE_FIRMWARE(FIRMWARE_NAVI10);
54MODULE_FIRMWARE(FIRMWARE_NAVI14);
55MODULE_FIRMWARE(FIRMWARE_NAVI12);
56MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
57MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
58
59static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
60
61int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
62{
63 unsigned long bo_size;
64 const char *fw_name;
65 const struct common_firmware_header *hdr;
66 unsigned char fw_check;
67 int i, r;
68
69 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
70 mutex_init(&adev->vcn.vcn_pg_lock);
71 atomic_set(&adev->vcn.total_submission_cnt, 0);
72 for (i = 0; i < adev->vcn.num_vcn_inst; i++)
73 atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
74
75 switch (adev->asic_type) {
76 case CHIP_RAVEN:
77 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
78 fw_name = FIRMWARE_RAVEN2;
79 else if (adev->apu_flags & AMD_APU_IS_PICASSO)
80 fw_name = FIRMWARE_PICASSO;
81 else
82 fw_name = FIRMWARE_RAVEN;
83 break;
84 case CHIP_ARCTURUS:
85 fw_name = FIRMWARE_ARCTURUS;
86 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
87 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
88 adev->vcn.indirect_sram = true;
89 break;
90 case CHIP_RENOIR:
91 fw_name = FIRMWARE_RENOIR;
92 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
93 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
94 adev->vcn.indirect_sram = true;
95 break;
96 case CHIP_NAVI10:
97 fw_name = FIRMWARE_NAVI10;
98 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
99 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
100 adev->vcn.indirect_sram = true;
101 break;
102 case CHIP_NAVI14:
103 fw_name = FIRMWARE_NAVI14;
104 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
105 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
106 adev->vcn.indirect_sram = true;
107 break;
108 case CHIP_NAVI12:
109 fw_name = FIRMWARE_NAVI12;
110 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
111 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
112 adev->vcn.indirect_sram = true;
113 break;
114 case CHIP_SIENNA_CICHLID:
115 fw_name = FIRMWARE_SIENNA_CICHLID;
116 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
117 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
118 adev->vcn.indirect_sram = true;
119 break;
120 case CHIP_NAVY_FLOUNDER:
121 fw_name = FIRMWARE_NAVY_FLOUNDER;
122 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
123 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
124 adev->vcn.indirect_sram = true;
125 break;
126 default:
127 return -EINVAL;
128 }
129
130 r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
131 if (r) {
132 dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
133 fw_name);
134 return r;
135 }
136
137 r = amdgpu_ucode_validate(adev->vcn.fw);
138 if (r) {
139 dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
140 fw_name);
141 release_firmware(adev->vcn.fw);
142 adev->vcn.fw = NULL;
143 return r;
144 }
145
146 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
147 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
148
149 /* Bit 20-23, it is encode major and non-zero for new naming convention.
150 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
151 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
152 * is zero in old naming convention, this field is always zero so far.
153 * These four bits are used to tell which naming convention is present.
154 */
155 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
156 if (fw_check) {
157 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
158
159 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
160 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
161 enc_major = fw_check;
162 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
163 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
164 DRM_INFO("Found VCN firmware Version ENC: %hu.%hu DEC: %hu VEP: %hu Revision: %hu\n",
165 enc_major, enc_minor, dec_ver, vep, fw_rev);
166 } else {
167 unsigned int version_major, version_minor, family_id;
168
169 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
170 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
171 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
172 DRM_INFO("Found VCN firmware Version: %hu.%hu Family ID: %hu\n",
173 version_major, version_minor, family_id);
174 }
175
176 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
177 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
178 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
179 bo_size += AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
180
181 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
182 if (adev->vcn.harvest_config & (1 << i))
183 continue;
184
185 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
186 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
187 &adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
188 if (r) {
189 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
190 return r;
191 }
192
193 adev->vcn.inst[i].fw_shared_cpu_addr = adev->vcn.inst[i].cpu_addr +
194 bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
195 adev->vcn.inst[i].fw_shared_gpu_addr = adev->vcn.inst[i].gpu_addr +
196 bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
197
198 if (adev->vcn.indirect_sram) {
199 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
200 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
201 &adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
202 if (r) {
203 dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
204 return r;
205 }
206 }
207 }
208
209 return 0;
210}
211
212int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
213{
214 int i, j;
215
216 cancel_delayed_work_sync(&adev->vcn.idle_work);
217
218 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
219 if (adev->vcn.harvest_config & (1 << j))
220 continue;
221
222 if (adev->vcn.indirect_sram) {
223 amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo,
224 &adev->vcn.inst[j].dpg_sram_gpu_addr,
225 (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
226 }
227 kvfree(adev->vcn.inst[j].saved_bo);
228
229 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
230 &adev->vcn.inst[j].gpu_addr,
231 (void **)&adev->vcn.inst[j].cpu_addr);
232
233 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
234
235 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
236 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
237 }
238
239 release_firmware(adev->vcn.fw);
240 mutex_destroy(&adev->vcn.vcn_pg_lock);
241
242 return 0;
243}
244
245int amdgpu_vcn_suspend(struct amdgpu_device *adev)
246{
247 unsigned size;
248 void *ptr;
249 int i;
250
251 cancel_delayed_work_sync(&adev->vcn.idle_work);
252
253 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
254 if (adev->vcn.harvest_config & (1 << i))
255 continue;
256 if (adev->vcn.inst[i].vcpu_bo == NULL)
257 return 0;
258
259 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
260 ptr = adev->vcn.inst[i].cpu_addr;
261
262 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
263 if (!adev->vcn.inst[i].saved_bo)
264 return -ENOMEM;
265
266 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
267 }
268 return 0;
269}
270
271int amdgpu_vcn_resume(struct amdgpu_device *adev)
272{
273 unsigned size;
274 void *ptr;
275 int i;
276
277 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
278 if (adev->vcn.harvest_config & (1 << i))
279 continue;
280 if (adev->vcn.inst[i].vcpu_bo == NULL)
281 return -EINVAL;
282
283 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
284 ptr = adev->vcn.inst[i].cpu_addr;
285
286 if (adev->vcn.inst[i].saved_bo != NULL) {
287 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
288 kvfree(adev->vcn.inst[i].saved_bo);
289 adev->vcn.inst[i].saved_bo = NULL;
290 } else {
291 const struct common_firmware_header *hdr;
292 unsigned offset;
293
294 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
295 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
296 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
297 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
298 le32_to_cpu(hdr->ucode_size_bytes));
299 size -= le32_to_cpu(hdr->ucode_size_bytes);
300 ptr += le32_to_cpu(hdr->ucode_size_bytes);
301 }
302 memset_io(ptr, 0, size);
303 }
304 }
305 return 0;
306}
307
308static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
309{
310 struct amdgpu_device *adev =
311 container_of(work, struct amdgpu_device, vcn.idle_work.work);
312 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
313 unsigned int i, j;
314
315 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
316 if (adev->vcn.harvest_config & (1 << j))
317 continue;
318
319 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
320 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
321 }
322
323 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
324 struct dpg_pause_state new_state;
325
326 if (fence[j] ||
327 unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
328 new_state.fw_based = VCN_DPG_STATE__PAUSE;
329 else
330 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
331
332 adev->vcn.pause_dpg_mode(adev, j, &new_state);
333 }
334
335 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
336 fences += fence[j];
337 }
338
339 if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
340 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
341 AMD_PG_STATE_GATE);
342 } else {
343 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
344 }
345}
346
347void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
348{
349 struct amdgpu_device *adev = ring->adev;
350
351 atomic_inc(&adev->vcn.total_submission_cnt);
352 cancel_delayed_work_sync(&adev->vcn.idle_work);
353
354 mutex_lock(&adev->vcn.vcn_pg_lock);
355 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
356 AMD_PG_STATE_UNGATE);
357
358 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
359 struct dpg_pause_state new_state;
360
361 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
362 atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
363 new_state.fw_based = VCN_DPG_STATE__PAUSE;
364 } else {
365 unsigned int fences = 0;
366 unsigned int i;
367
368 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
369 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
370
371 if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
372 new_state.fw_based = VCN_DPG_STATE__PAUSE;
373 else
374 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
375 }
376
377 adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
378 }
379 mutex_unlock(&adev->vcn.vcn_pg_lock);
380}
381
382void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
383{
384 if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
385 ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
386 atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
387
388 atomic_dec(&ring->adev->vcn.total_submission_cnt);
389
390 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
391}
392
393int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
394{
395 struct amdgpu_device *adev = ring->adev;
396 uint32_t tmp = 0;
397 unsigned i;
398 int r;
399
400 /* VCN in SRIOV does not support direct register read/write */
401 if (amdgpu_sriov_vf(adev))
402 return 0;
403
404 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
405 r = amdgpu_ring_alloc(ring, 3);
406 if (r)
407 return r;
408 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
409 amdgpu_ring_write(ring, 0xDEADBEEF);
410 amdgpu_ring_commit(ring);
411 for (i = 0; i < adev->usec_timeout; i++) {
412 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
413 if (tmp == 0xDEADBEEF)
414 break;
415 udelay(1);
416 }
417
418 if (i >= adev->usec_timeout)
419 r = -ETIMEDOUT;
420
421 return r;
422}
423
424static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
425 struct amdgpu_bo *bo,
426 struct dma_fence **fence)
427{
428 struct amdgpu_device *adev = ring->adev;
429 struct dma_fence *f = NULL;
430 struct amdgpu_job *job;
431 struct amdgpu_ib *ib;
432 uint64_t addr;
433 int i, r;
434
435 r = amdgpu_job_alloc_with_ib(adev, 64,
436 AMDGPU_IB_POOL_DIRECT, &job);
437 if (r)
438 goto err;
439
440 ib = &job->ibs[0];
441 addr = amdgpu_bo_gpu_offset(bo);
442 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
443 ib->ptr[1] = addr;
444 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
445 ib->ptr[3] = addr >> 32;
446 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
447 ib->ptr[5] = 0;
448 for (i = 6; i < 16; i += 2) {
449 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
450 ib->ptr[i+1] = 0;
451 }
452 ib->length_dw = 16;
453
454 r = amdgpu_job_submit_direct(job, ring, &f);
455 if (r)
456 goto err_free;
457
458 amdgpu_bo_fence(bo, f, false);
459 amdgpu_bo_unreserve(bo);
460 amdgpu_bo_unref(&bo);
461
462 if (fence)
463 *fence = dma_fence_get(f);
464 dma_fence_put(f);
465
466 return 0;
467
468err_free:
469 amdgpu_job_free(job);
470
471err:
472 amdgpu_bo_unreserve(bo);
473 amdgpu_bo_unref(&bo);
474 return r;
475}
476
477static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
478 struct dma_fence **fence)
479{
480 struct amdgpu_device *adev = ring->adev;
481 struct amdgpu_bo *bo = NULL;
482 uint32_t *msg;
483 int r, i;
484
485 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
486 AMDGPU_GEM_DOMAIN_VRAM,
487 &bo, NULL, (void **)&msg);
488 if (r)
489 return r;
490
491 msg[0] = cpu_to_le32(0x00000028);
492 msg[1] = cpu_to_le32(0x00000038);
493 msg[2] = cpu_to_le32(0x00000001);
494 msg[3] = cpu_to_le32(0x00000000);
495 msg[4] = cpu_to_le32(handle);
496 msg[5] = cpu_to_le32(0x00000000);
497 msg[6] = cpu_to_le32(0x00000001);
498 msg[7] = cpu_to_le32(0x00000028);
499 msg[8] = cpu_to_le32(0x00000010);
500 msg[9] = cpu_to_le32(0x00000000);
501 msg[10] = cpu_to_le32(0x00000007);
502 msg[11] = cpu_to_le32(0x00000000);
503 msg[12] = cpu_to_le32(0x00000780);
504 msg[13] = cpu_to_le32(0x00000440);
505 for (i = 14; i < 1024; ++i)
506 msg[i] = cpu_to_le32(0x0);
507
508 return amdgpu_vcn_dec_send_msg(ring, bo, fence);
509}
510
511static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
512 struct dma_fence **fence)
513{
514 struct amdgpu_device *adev = ring->adev;
515 struct amdgpu_bo *bo = NULL;
516 uint32_t *msg;
517 int r, i;
518
519 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
520 AMDGPU_GEM_DOMAIN_VRAM,
521 &bo, NULL, (void **)&msg);
522 if (r)
523 return r;
524
525 msg[0] = cpu_to_le32(0x00000028);
526 msg[1] = cpu_to_le32(0x00000018);
527 msg[2] = cpu_to_le32(0x00000000);
528 msg[3] = cpu_to_le32(0x00000002);
529 msg[4] = cpu_to_le32(handle);
530 msg[5] = cpu_to_le32(0x00000000);
531 for (i = 6; i < 1024; ++i)
532 msg[i] = cpu_to_le32(0x0);
533
534 return amdgpu_vcn_dec_send_msg(ring, bo, fence);
535}
536
537int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
538{
539 struct dma_fence *fence;
540 long r;
541
542 r = amdgpu_vcn_dec_get_create_msg(ring, 1, NULL);
543 if (r)
544 goto error;
545
546 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &fence);
547 if (r)
548 goto error;
549
550 r = dma_fence_wait_timeout(fence, false, timeout);
551 if (r == 0)
552 r = -ETIMEDOUT;
553 else if (r > 0)
554 r = 0;
555
556 dma_fence_put(fence);
557error:
558 return r;
559}
560
561int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
562{
563 struct amdgpu_device *adev = ring->adev;
564 uint32_t rptr;
565 unsigned i;
566 int r;
567
568 if (amdgpu_sriov_vf(adev))
569 return 0;
570
571 r = amdgpu_ring_alloc(ring, 16);
572 if (r)
573 return r;
574
575 rptr = amdgpu_ring_get_rptr(ring);
576
577 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
578 amdgpu_ring_commit(ring);
579
580 for (i = 0; i < adev->usec_timeout; i++) {
581 if (amdgpu_ring_get_rptr(ring) != rptr)
582 break;
583 udelay(1);
584 }
585
586 if (i >= adev->usec_timeout)
587 r = -ETIMEDOUT;
588
589 return r;
590}
591
592static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
593 struct amdgpu_bo *bo,
594 struct dma_fence **fence)
595{
596 const unsigned ib_size_dw = 16;
597 struct amdgpu_job *job;
598 struct amdgpu_ib *ib;
599 struct dma_fence *f = NULL;
600 uint64_t addr;
601 int i, r;
602
603 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
604 AMDGPU_IB_POOL_DIRECT, &job);
605 if (r)
606 return r;
607
608 ib = &job->ibs[0];
609 addr = amdgpu_bo_gpu_offset(bo);
610
611 ib->length_dw = 0;
612 ib->ptr[ib->length_dw++] = 0x00000018;
613 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
614 ib->ptr[ib->length_dw++] = handle;
615 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
616 ib->ptr[ib->length_dw++] = addr;
617 ib->ptr[ib->length_dw++] = 0x0000000b;
618
619 ib->ptr[ib->length_dw++] = 0x00000014;
620 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
621 ib->ptr[ib->length_dw++] = 0x0000001c;
622 ib->ptr[ib->length_dw++] = 0x00000000;
623 ib->ptr[ib->length_dw++] = 0x00000000;
624
625 ib->ptr[ib->length_dw++] = 0x00000008;
626 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
627
628 for (i = ib->length_dw; i < ib_size_dw; ++i)
629 ib->ptr[i] = 0x0;
630
631 r = amdgpu_job_submit_direct(job, ring, &f);
632 if (r)
633 goto err;
634
635 if (fence)
636 *fence = dma_fence_get(f);
637 dma_fence_put(f);
638
639 return 0;
640
641err:
642 amdgpu_job_free(job);
643 return r;
644}
645
646static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
647 struct amdgpu_bo *bo,
648 struct dma_fence **fence)
649{
650 const unsigned ib_size_dw = 16;
651 struct amdgpu_job *job;
652 struct amdgpu_ib *ib;
653 struct dma_fence *f = NULL;
654 uint64_t addr;
655 int i, r;
656
657 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
658 AMDGPU_IB_POOL_DIRECT, &job);
659 if (r)
660 return r;
661
662 ib = &job->ibs[0];
663 addr = amdgpu_bo_gpu_offset(bo);
664
665 ib->length_dw = 0;
666 ib->ptr[ib->length_dw++] = 0x00000018;
667 ib->ptr[ib->length_dw++] = 0x00000001;
668 ib->ptr[ib->length_dw++] = handle;
669 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
670 ib->ptr[ib->length_dw++] = addr;
671 ib->ptr[ib->length_dw++] = 0x0000000b;
672
673 ib->ptr[ib->length_dw++] = 0x00000014;
674 ib->ptr[ib->length_dw++] = 0x00000002;
675 ib->ptr[ib->length_dw++] = 0x0000001c;
676 ib->ptr[ib->length_dw++] = 0x00000000;
677 ib->ptr[ib->length_dw++] = 0x00000000;
678
679 ib->ptr[ib->length_dw++] = 0x00000008;
680 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
681
682 for (i = ib->length_dw; i < ib_size_dw; ++i)
683 ib->ptr[i] = 0x0;
684
685 r = amdgpu_job_submit_direct(job, ring, &f);
686 if (r)
687 goto err;
688
689 if (fence)
690 *fence = dma_fence_get(f);
691 dma_fence_put(f);
692
693 return 0;
694
695err:
696 amdgpu_job_free(job);
697 return r;
698}
699
700int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
701{
702 struct dma_fence *fence = NULL;
703 struct amdgpu_bo *bo = NULL;
704 long r;
705
706 r = amdgpu_bo_create_reserved(ring->adev, 128 * 1024, PAGE_SIZE,
707 AMDGPU_GEM_DOMAIN_VRAM,
708 &bo, NULL, NULL);
709 if (r)
710 return r;
711
712 r = amdgpu_vcn_enc_get_create_msg(ring, 1, bo, NULL);
713 if (r)
714 goto error;
715
716 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, bo, &fence);
717 if (r)
718 goto error;
719
720 r = dma_fence_wait_timeout(fence, false, timeout);
721 if (r == 0)
722 r = -ETIMEDOUT;
723 else if (r > 0)
724 r = 0;
725
726error:
727 dma_fence_put(fence);
728 amdgpu_bo_unreserve(bo);
729 amdgpu_bo_unref(&bo);
730 return r;
731}