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 "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}
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/dmi.h>
30#include <linux/pci.h>
31#include <linux/debugfs.h>
32#include <drm/drm_drv.h>
33
34#include "amdgpu.h"
35#include "amdgpu_pm.h"
36#include "amdgpu_vcn.h"
37#include "soc15d.h"
38
39/* Firmware Names */
40#define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
41#define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
42#define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
43#define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
44#define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
45#define FIRMWARE_GREEN_SARDINE "amdgpu/green_sardine_vcn.bin"
46#define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
47#define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
48#define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
49#define FIRMWARE_SIENNA_CICHLID "amdgpu/sienna_cichlid_vcn.bin"
50#define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin"
51#define FIRMWARE_VANGOGH "amdgpu/vangogh_vcn.bin"
52#define FIRMWARE_DIMGREY_CAVEFISH "amdgpu/dimgrey_cavefish_vcn.bin"
53#define FIRMWARE_ALDEBARAN "amdgpu/aldebaran_vcn.bin"
54#define FIRMWARE_BEIGE_GOBY "amdgpu/beige_goby_vcn.bin"
55#define FIRMWARE_YELLOW_CARP "amdgpu/yellow_carp_vcn.bin"
56#define FIRMWARE_VCN_3_1_2 "amdgpu/vcn_3_1_2.bin"
57#define FIRMWARE_VCN4_0_0 "amdgpu/vcn_4_0_0.bin"
58#define FIRMWARE_VCN4_0_2 "amdgpu/vcn_4_0_2.bin"
59#define FIRMWARE_VCN4_0_3 "amdgpu/vcn_4_0_3.bin"
60#define FIRMWARE_VCN4_0_4 "amdgpu/vcn_4_0_4.bin"
61#define FIRMWARE_VCN4_0_5 "amdgpu/vcn_4_0_5.bin"
62#define FIRMWARE_VCN4_0_6 "amdgpu/vcn_4_0_6.bin"
63#define FIRMWARE_VCN4_0_6_1 "amdgpu/vcn_4_0_6_1.bin"
64#define FIRMWARE_VCN5_0_0 "amdgpu/vcn_5_0_0.bin"
65
66MODULE_FIRMWARE(FIRMWARE_RAVEN);
67MODULE_FIRMWARE(FIRMWARE_PICASSO);
68MODULE_FIRMWARE(FIRMWARE_RAVEN2);
69MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
70MODULE_FIRMWARE(FIRMWARE_RENOIR);
71MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
72MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
73MODULE_FIRMWARE(FIRMWARE_NAVI10);
74MODULE_FIRMWARE(FIRMWARE_NAVI14);
75MODULE_FIRMWARE(FIRMWARE_NAVI12);
76MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
77MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
78MODULE_FIRMWARE(FIRMWARE_VANGOGH);
79MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
80MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
81MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
82MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
83MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
84MODULE_FIRMWARE(FIRMWARE_VCN4_0_2);
85MODULE_FIRMWARE(FIRMWARE_VCN4_0_3);
86MODULE_FIRMWARE(FIRMWARE_VCN4_0_4);
87MODULE_FIRMWARE(FIRMWARE_VCN4_0_5);
88MODULE_FIRMWARE(FIRMWARE_VCN4_0_6);
89MODULE_FIRMWARE(FIRMWARE_VCN4_0_6_1);
90MODULE_FIRMWARE(FIRMWARE_VCN5_0_0);
91
92static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
93
94int amdgpu_vcn_early_init(struct amdgpu_device *adev)
95{
96 char ucode_prefix[25];
97 int r, i;
98
99 amdgpu_ucode_ip_version_decode(adev, UVD_HWIP, ucode_prefix, sizeof(ucode_prefix));
100 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
101 if (i == 1 && amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(4, 0, 6))
102 r = amdgpu_ucode_request(adev, &adev->vcn.fw[i], "amdgpu/%s_%d.bin", ucode_prefix, i);
103 else
104 r = amdgpu_ucode_request(adev, &adev->vcn.fw[i], "amdgpu/%s.bin", ucode_prefix);
105 if (r) {
106 amdgpu_ucode_release(&adev->vcn.fw[i]);
107 return r;
108 }
109 }
110 return r;
111}
112
113int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
114{
115 unsigned long bo_size;
116 const struct common_firmware_header *hdr;
117 unsigned char fw_check;
118 unsigned int fw_shared_size, log_offset;
119 int i, r;
120
121 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
122 mutex_init(&adev->vcn.vcn_pg_lock);
123 mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
124 atomic_set(&adev->vcn.total_submission_cnt, 0);
125 for (i = 0; i < adev->vcn.num_vcn_inst; i++)
126 atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
127
128 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
129 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
130 adev->vcn.indirect_sram = true;
131
132 /*
133 * Some Steam Deck's BIOS versions are incompatible with the
134 * indirect SRAM mode, leading to amdgpu being unable to get
135 * properly probed (and even potentially crashing the kernel).
136 * Hence, check for these versions here - notice this is
137 * restricted to Vangogh (Deck's APU).
138 */
139 if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(3, 0, 2)) {
140 const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION);
141
142 if (bios_ver && (!strncmp("F7A0113", bios_ver, 7) ||
143 !strncmp("F7A0114", bios_ver, 7))) {
144 adev->vcn.indirect_sram = false;
145 dev_info(adev->dev,
146 "Steam Deck quirk: indirect SRAM disabled on BIOS %s\n", bios_ver);
147 }
148 }
149
150 /* from vcn4 and above, only unified queue is used */
151 adev->vcn.using_unified_queue =
152 amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0);
153
154 hdr = (const struct common_firmware_header *)adev->vcn.fw[0]->data;
155 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
156
157 /* Bit 20-23, it is encode major and non-zero for new naming convention.
158 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
159 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
160 * is zero in old naming convention, this field is always zero so far.
161 * These four bits are used to tell which naming convention is present.
162 */
163 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
164 if (fw_check) {
165 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
166
167 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
168 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
169 enc_major = fw_check;
170 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
171 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
172 DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
173 enc_major, enc_minor, dec_ver, vep, fw_rev);
174 } else {
175 unsigned int version_major, version_minor, family_id;
176
177 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
178 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
179 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
180 DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
181 version_major, version_minor, family_id);
182 }
183
184 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
185 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
186 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
187
188 if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(5, 0, 0)) {
189 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn5_fw_shared));
190 log_offset = offsetof(struct amdgpu_vcn5_fw_shared, fw_log);
191 } else if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0)) {
192 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
193 log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
194 } else {
195 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
196 log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
197 }
198
199 bo_size += fw_shared_size;
200
201 if (amdgpu_vcnfw_log)
202 bo_size += AMDGPU_VCNFW_LOG_SIZE;
203
204 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
205 if (adev->vcn.harvest_config & (1 << i))
206 continue;
207
208 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
209 AMDGPU_GEM_DOMAIN_VRAM |
210 AMDGPU_GEM_DOMAIN_GTT,
211 &adev->vcn.inst[i].vcpu_bo,
212 &adev->vcn.inst[i].gpu_addr,
213 &adev->vcn.inst[i].cpu_addr);
214 if (r) {
215 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
216 return r;
217 }
218
219 adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr +
220 bo_size - fw_shared_size;
221 adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr +
222 bo_size - fw_shared_size;
223
224 adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size;
225
226 if (amdgpu_vcnfw_log) {
227 adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
228 adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
229 adev->vcn.inst[i].fw_shared.log_offset = log_offset;
230 }
231
232 if (adev->vcn.indirect_sram) {
233 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
234 AMDGPU_GEM_DOMAIN_VRAM |
235 AMDGPU_GEM_DOMAIN_GTT,
236 &adev->vcn.inst[i].dpg_sram_bo,
237 &adev->vcn.inst[i].dpg_sram_gpu_addr,
238 &adev->vcn.inst[i].dpg_sram_cpu_addr);
239 if (r) {
240 dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
241 return r;
242 }
243 }
244 }
245
246 return 0;
247}
248
249int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
250{
251 int i, j;
252
253 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
254 if (adev->vcn.harvest_config & (1 << j))
255 continue;
256
257 amdgpu_bo_free_kernel(
258 &adev->vcn.inst[j].dpg_sram_bo,
259 &adev->vcn.inst[j].dpg_sram_gpu_addr,
260 (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
261
262 kvfree(adev->vcn.inst[j].saved_bo);
263
264 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
265 &adev->vcn.inst[j].gpu_addr,
266 (void **)&adev->vcn.inst[j].cpu_addr);
267
268 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
269
270 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
271 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
272
273 amdgpu_ucode_release(&adev->vcn.fw[j]);
274 }
275
276 mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
277 mutex_destroy(&adev->vcn.vcn_pg_lock);
278
279 return 0;
280}
281
282bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
283{
284 bool ret = false;
285 int vcn_config = adev->vcn.vcn_config[vcn_instance];
286
287 if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK))
288 ret = true;
289 else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK))
290 ret = true;
291 else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK))
292 ret = true;
293
294 return ret;
295}
296
297int amdgpu_vcn_save_vcpu_bo(struct amdgpu_device *adev)
298{
299 unsigned int size;
300 void *ptr;
301 int i, idx;
302
303 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
304 if (adev->vcn.harvest_config & (1 << i))
305 continue;
306 if (adev->vcn.inst[i].vcpu_bo == NULL)
307 return 0;
308
309 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
310 ptr = adev->vcn.inst[i].cpu_addr;
311
312 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
313 if (!adev->vcn.inst[i].saved_bo)
314 return -ENOMEM;
315
316 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
317 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
318 drm_dev_exit(idx);
319 }
320 }
321
322 return 0;
323}
324
325int amdgpu_vcn_suspend(struct amdgpu_device *adev)
326{
327 bool in_ras_intr = amdgpu_ras_intr_triggered();
328
329 cancel_delayed_work_sync(&adev->vcn.idle_work);
330
331 /* err_event_athub will corrupt VCPU buffer, so we need to
332 * restore fw data and clear buffer in amdgpu_vcn_resume() */
333 if (in_ras_intr)
334 return 0;
335
336 return amdgpu_vcn_save_vcpu_bo(adev);
337}
338
339int amdgpu_vcn_resume(struct amdgpu_device *adev)
340{
341 unsigned int size;
342 void *ptr;
343 int i, idx;
344
345 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
346 if (adev->vcn.harvest_config & (1 << i))
347 continue;
348 if (adev->vcn.inst[i].vcpu_bo == NULL)
349 return -EINVAL;
350
351 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
352 ptr = adev->vcn.inst[i].cpu_addr;
353
354 if (adev->vcn.inst[i].saved_bo != NULL) {
355 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
356 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
357 drm_dev_exit(idx);
358 }
359 kvfree(adev->vcn.inst[i].saved_bo);
360 adev->vcn.inst[i].saved_bo = NULL;
361 } else {
362 const struct common_firmware_header *hdr;
363 unsigned int offset;
364
365 hdr = (const struct common_firmware_header *)adev->vcn.fw[i]->data;
366 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
367 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
368 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
369 memcpy_toio(adev->vcn.inst[i].cpu_addr,
370 adev->vcn.fw[i]->data + offset,
371 le32_to_cpu(hdr->ucode_size_bytes));
372 drm_dev_exit(idx);
373 }
374 size -= le32_to_cpu(hdr->ucode_size_bytes);
375 ptr += le32_to_cpu(hdr->ucode_size_bytes);
376 }
377 memset_io(ptr, 0, size);
378 }
379 }
380 return 0;
381}
382
383static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
384{
385 struct amdgpu_device *adev =
386 container_of(work, struct amdgpu_device, vcn.idle_work.work);
387 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
388 unsigned int i, j;
389 int r = 0;
390
391 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
392 if (adev->vcn.harvest_config & (1 << j))
393 continue;
394
395 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
396 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
397
398 /* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
399 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
400 !adev->vcn.using_unified_queue) {
401 struct dpg_pause_state new_state;
402
403 if (fence[j] ||
404 unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
405 new_state.fw_based = VCN_DPG_STATE__PAUSE;
406 else
407 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
408
409 adev->vcn.pause_dpg_mode(adev, j, &new_state);
410 }
411
412 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
413 fences += fence[j];
414 }
415
416 if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
417 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
418 AMD_PG_STATE_GATE);
419 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
420 false);
421 if (r)
422 dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
423 } else {
424 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
425 }
426}
427
428void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
429{
430 struct amdgpu_device *adev = ring->adev;
431 int r = 0;
432
433 atomic_inc(&adev->vcn.total_submission_cnt);
434
435 if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
436 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
437 true);
438 if (r)
439 dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
440 }
441
442 mutex_lock(&adev->vcn.vcn_pg_lock);
443 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
444 AMD_PG_STATE_UNGATE);
445
446 /* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
447 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
448 !adev->vcn.using_unified_queue) {
449 struct dpg_pause_state new_state;
450
451 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
452 atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
453 new_state.fw_based = VCN_DPG_STATE__PAUSE;
454 } else {
455 unsigned int fences = 0;
456 unsigned int i;
457
458 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
459 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
460
461 if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
462 new_state.fw_based = VCN_DPG_STATE__PAUSE;
463 else
464 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
465 }
466
467 adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
468 }
469 mutex_unlock(&adev->vcn.vcn_pg_lock);
470}
471
472void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
473{
474 struct amdgpu_device *adev = ring->adev;
475
476 /* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
477 if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
478 ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC &&
479 !adev->vcn.using_unified_queue)
480 atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
481
482 atomic_dec(&ring->adev->vcn.total_submission_cnt);
483
484 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
485}
486
487int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
488{
489 struct amdgpu_device *adev = ring->adev;
490 uint32_t tmp = 0;
491 unsigned int i;
492 int r;
493
494 /* VCN in SRIOV does not support direct register read/write */
495 if (amdgpu_sriov_vf(adev))
496 return 0;
497
498 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
499 r = amdgpu_ring_alloc(ring, 3);
500 if (r)
501 return r;
502 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
503 amdgpu_ring_write(ring, 0xDEADBEEF);
504 amdgpu_ring_commit(ring);
505 for (i = 0; i < adev->usec_timeout; i++) {
506 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
507 if (tmp == 0xDEADBEEF)
508 break;
509 udelay(1);
510 }
511
512 if (i >= adev->usec_timeout)
513 r = -ETIMEDOUT;
514
515 return r;
516}
517
518int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
519{
520 struct amdgpu_device *adev = ring->adev;
521 uint32_t rptr;
522 unsigned int i;
523 int r;
524
525 if (amdgpu_sriov_vf(adev))
526 return 0;
527
528 r = amdgpu_ring_alloc(ring, 16);
529 if (r)
530 return r;
531
532 rptr = amdgpu_ring_get_rptr(ring);
533
534 amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
535 amdgpu_ring_commit(ring);
536
537 for (i = 0; i < adev->usec_timeout; i++) {
538 if (amdgpu_ring_get_rptr(ring) != rptr)
539 break;
540 udelay(1);
541 }
542
543 if (i >= adev->usec_timeout)
544 r = -ETIMEDOUT;
545
546 return r;
547}
548
549static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
550 struct amdgpu_ib *ib_msg,
551 struct dma_fence **fence)
552{
553 u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
554 struct amdgpu_device *adev = ring->adev;
555 struct dma_fence *f = NULL;
556 struct amdgpu_job *job;
557 struct amdgpu_ib *ib;
558 int i, r;
559
560 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
561 64, AMDGPU_IB_POOL_DIRECT,
562 &job);
563 if (r)
564 goto err;
565
566 ib = &job->ibs[0];
567 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
568 ib->ptr[1] = addr;
569 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
570 ib->ptr[3] = addr >> 32;
571 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
572 ib->ptr[5] = 0;
573 for (i = 6; i < 16; i += 2) {
574 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
575 ib->ptr[i+1] = 0;
576 }
577 ib->length_dw = 16;
578
579 r = amdgpu_job_submit_direct(job, ring, &f);
580 if (r)
581 goto err_free;
582
583 amdgpu_ib_free(adev, ib_msg, f);
584
585 if (fence)
586 *fence = dma_fence_get(f);
587 dma_fence_put(f);
588
589 return 0;
590
591err_free:
592 amdgpu_job_free(job);
593err:
594 amdgpu_ib_free(adev, ib_msg, f);
595 return r;
596}
597
598static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
599 struct amdgpu_ib *ib)
600{
601 struct amdgpu_device *adev = ring->adev;
602 uint32_t *msg;
603 int r, i;
604
605 memset(ib, 0, sizeof(*ib));
606 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
607 AMDGPU_IB_POOL_DIRECT,
608 ib);
609 if (r)
610 return r;
611
612 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
613 msg[0] = cpu_to_le32(0x00000028);
614 msg[1] = cpu_to_le32(0x00000038);
615 msg[2] = cpu_to_le32(0x00000001);
616 msg[3] = cpu_to_le32(0x00000000);
617 msg[4] = cpu_to_le32(handle);
618 msg[5] = cpu_to_le32(0x00000000);
619 msg[6] = cpu_to_le32(0x00000001);
620 msg[7] = cpu_to_le32(0x00000028);
621 msg[8] = cpu_to_le32(0x00000010);
622 msg[9] = cpu_to_le32(0x00000000);
623 msg[10] = cpu_to_le32(0x00000007);
624 msg[11] = cpu_to_le32(0x00000000);
625 msg[12] = cpu_to_le32(0x00000780);
626 msg[13] = cpu_to_le32(0x00000440);
627 for (i = 14; i < 1024; ++i)
628 msg[i] = cpu_to_le32(0x0);
629
630 return 0;
631}
632
633static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
634 struct amdgpu_ib *ib)
635{
636 struct amdgpu_device *adev = ring->adev;
637 uint32_t *msg;
638 int r, i;
639
640 memset(ib, 0, sizeof(*ib));
641 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
642 AMDGPU_IB_POOL_DIRECT,
643 ib);
644 if (r)
645 return r;
646
647 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
648 msg[0] = cpu_to_le32(0x00000028);
649 msg[1] = cpu_to_le32(0x00000018);
650 msg[2] = cpu_to_le32(0x00000000);
651 msg[3] = cpu_to_le32(0x00000002);
652 msg[4] = cpu_to_le32(handle);
653 msg[5] = cpu_to_le32(0x00000000);
654 for (i = 6; i < 1024; ++i)
655 msg[i] = cpu_to_le32(0x0);
656
657 return 0;
658}
659
660int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
661{
662 struct dma_fence *fence = NULL;
663 struct amdgpu_ib ib;
664 long r;
665
666 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
667 if (r)
668 goto error;
669
670 r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
671 if (r)
672 goto error;
673 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
674 if (r)
675 goto error;
676
677 r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
678 if (r)
679 goto error;
680
681 r = dma_fence_wait_timeout(fence, false, timeout);
682 if (r == 0)
683 r = -ETIMEDOUT;
684 else if (r > 0)
685 r = 0;
686
687 dma_fence_put(fence);
688error:
689 return r;
690}
691
692static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib,
693 uint32_t ib_pack_in_dw, bool enc)
694{
695 uint32_t *ib_checksum;
696
697 ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */
698 ib->ptr[ib->length_dw++] = 0x30000002;
699 ib_checksum = &ib->ptr[ib->length_dw++];
700 ib->ptr[ib->length_dw++] = ib_pack_in_dw;
701
702 ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */
703 ib->ptr[ib->length_dw++] = 0x30000001;
704 ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3;
705 ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t);
706
707 return ib_checksum;
708}
709
710static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
711 uint32_t ib_pack_in_dw)
712{
713 uint32_t i;
714 uint32_t checksum = 0;
715
716 for (i = 0; i < ib_pack_in_dw; i++)
717 checksum += *(*ib_checksum + 2 + i);
718
719 **ib_checksum = checksum;
720}
721
722static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
723 struct amdgpu_ib *ib_msg,
724 struct dma_fence **fence)
725{
726 struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
727 unsigned int ib_size_dw = 64;
728 struct amdgpu_device *adev = ring->adev;
729 struct dma_fence *f = NULL;
730 struct amdgpu_job *job;
731 struct amdgpu_ib *ib;
732 uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
733 uint32_t *ib_checksum;
734 uint32_t ib_pack_in_dw;
735 int i, r;
736
737 if (adev->vcn.using_unified_queue)
738 ib_size_dw += 8;
739
740 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
741 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
742 &job);
743 if (r)
744 goto err;
745
746 ib = &job->ibs[0];
747 ib->length_dw = 0;
748
749 /* single queue headers */
750 if (adev->vcn.using_unified_queue) {
751 ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t)
752 + 4 + 2; /* engine info + decoding ib in dw */
753 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false);
754 }
755
756 ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
757 ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
758 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
759 ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
760 memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
761
762 decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
763 decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
764 decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
765
766 for (i = ib->length_dw; i < ib_size_dw; ++i)
767 ib->ptr[i] = 0x0;
768
769 if (adev->vcn.using_unified_queue)
770 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw);
771
772 r = amdgpu_job_submit_direct(job, ring, &f);
773 if (r)
774 goto err_free;
775
776 amdgpu_ib_free(adev, ib_msg, f);
777
778 if (fence)
779 *fence = dma_fence_get(f);
780 dma_fence_put(f);
781
782 return 0;
783
784err_free:
785 amdgpu_job_free(job);
786err:
787 amdgpu_ib_free(adev, ib_msg, f);
788 return r;
789}
790
791int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
792{
793 struct dma_fence *fence = NULL;
794 struct amdgpu_ib ib;
795 long r;
796
797 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
798 if (r)
799 goto error;
800
801 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
802 if (r)
803 goto error;
804 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
805 if (r)
806 goto error;
807
808 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
809 if (r)
810 goto error;
811
812 r = dma_fence_wait_timeout(fence, false, timeout);
813 if (r == 0)
814 r = -ETIMEDOUT;
815 else if (r > 0)
816 r = 0;
817
818 dma_fence_put(fence);
819error:
820 return r;
821}
822
823int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
824{
825 struct amdgpu_device *adev = ring->adev;
826 uint32_t rptr;
827 unsigned int i;
828 int r;
829
830 if (amdgpu_sriov_vf(adev))
831 return 0;
832
833 r = amdgpu_ring_alloc(ring, 16);
834 if (r)
835 return r;
836
837 rptr = amdgpu_ring_get_rptr(ring);
838
839 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
840 amdgpu_ring_commit(ring);
841
842 for (i = 0; i < adev->usec_timeout; i++) {
843 if (amdgpu_ring_get_rptr(ring) != rptr)
844 break;
845 udelay(1);
846 }
847
848 if (i >= adev->usec_timeout)
849 r = -ETIMEDOUT;
850
851 return r;
852}
853
854static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
855 struct amdgpu_ib *ib_msg,
856 struct dma_fence **fence)
857{
858 unsigned int ib_size_dw = 16;
859 struct amdgpu_device *adev = ring->adev;
860 struct amdgpu_job *job;
861 struct amdgpu_ib *ib;
862 struct dma_fence *f = NULL;
863 uint32_t *ib_checksum = NULL;
864 uint64_t addr;
865 int i, r;
866
867 if (adev->vcn.using_unified_queue)
868 ib_size_dw += 8;
869
870 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
871 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
872 &job);
873 if (r)
874 return r;
875
876 ib = &job->ibs[0];
877 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
878
879 ib->length_dw = 0;
880
881 if (adev->vcn.using_unified_queue)
882 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
883
884 ib->ptr[ib->length_dw++] = 0x00000018;
885 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
886 ib->ptr[ib->length_dw++] = handle;
887 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
888 ib->ptr[ib->length_dw++] = addr;
889 ib->ptr[ib->length_dw++] = 0x00000000;
890
891 ib->ptr[ib->length_dw++] = 0x00000014;
892 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
893 ib->ptr[ib->length_dw++] = 0x0000001c;
894 ib->ptr[ib->length_dw++] = 0x00000000;
895 ib->ptr[ib->length_dw++] = 0x00000000;
896
897 ib->ptr[ib->length_dw++] = 0x00000008;
898 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
899
900 for (i = ib->length_dw; i < ib_size_dw; ++i)
901 ib->ptr[i] = 0x0;
902
903 if (adev->vcn.using_unified_queue)
904 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
905
906 r = amdgpu_job_submit_direct(job, ring, &f);
907 if (r)
908 goto err;
909
910 if (fence)
911 *fence = dma_fence_get(f);
912 dma_fence_put(f);
913
914 return 0;
915
916err:
917 amdgpu_job_free(job);
918 return r;
919}
920
921static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
922 struct amdgpu_ib *ib_msg,
923 struct dma_fence **fence)
924{
925 unsigned int ib_size_dw = 16;
926 struct amdgpu_device *adev = ring->adev;
927 struct amdgpu_job *job;
928 struct amdgpu_ib *ib;
929 struct dma_fence *f = NULL;
930 uint32_t *ib_checksum = NULL;
931 uint64_t addr;
932 int i, r;
933
934 if (adev->vcn.using_unified_queue)
935 ib_size_dw += 8;
936
937 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
938 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
939 &job);
940 if (r)
941 return r;
942
943 ib = &job->ibs[0];
944 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
945
946 ib->length_dw = 0;
947
948 if (adev->vcn.using_unified_queue)
949 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
950
951 ib->ptr[ib->length_dw++] = 0x00000018;
952 ib->ptr[ib->length_dw++] = 0x00000001;
953 ib->ptr[ib->length_dw++] = handle;
954 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
955 ib->ptr[ib->length_dw++] = addr;
956 ib->ptr[ib->length_dw++] = 0x00000000;
957
958 ib->ptr[ib->length_dw++] = 0x00000014;
959 ib->ptr[ib->length_dw++] = 0x00000002;
960 ib->ptr[ib->length_dw++] = 0x0000001c;
961 ib->ptr[ib->length_dw++] = 0x00000000;
962 ib->ptr[ib->length_dw++] = 0x00000000;
963
964 ib->ptr[ib->length_dw++] = 0x00000008;
965 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
966
967 for (i = ib->length_dw; i < ib_size_dw; ++i)
968 ib->ptr[i] = 0x0;
969
970 if (adev->vcn.using_unified_queue)
971 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
972
973 r = amdgpu_job_submit_direct(job, ring, &f);
974 if (r)
975 goto err;
976
977 if (fence)
978 *fence = dma_fence_get(f);
979 dma_fence_put(f);
980
981 return 0;
982
983err:
984 amdgpu_job_free(job);
985 return r;
986}
987
988int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
989{
990 struct amdgpu_device *adev = ring->adev;
991 struct dma_fence *fence = NULL;
992 struct amdgpu_ib ib;
993 long r;
994
995 memset(&ib, 0, sizeof(ib));
996 r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
997 AMDGPU_IB_POOL_DIRECT,
998 &ib);
999 if (r)
1000 return r;
1001
1002 r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
1003 if (r)
1004 goto error;
1005
1006 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
1007 if (r)
1008 goto error;
1009
1010 r = dma_fence_wait_timeout(fence, false, timeout);
1011 if (r == 0)
1012 r = -ETIMEDOUT;
1013 else if (r > 0)
1014 r = 0;
1015
1016error:
1017 amdgpu_ib_free(adev, &ib, fence);
1018 dma_fence_put(fence);
1019
1020 return r;
1021}
1022
1023int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1024{
1025 struct amdgpu_device *adev = ring->adev;
1026 long r;
1027
1028 if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(4, 0, 3)) {
1029 r = amdgpu_vcn_enc_ring_test_ib(ring, timeout);
1030 if (r)
1031 goto error;
1032 }
1033
1034 r = amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout);
1035
1036error:
1037 return r;
1038}
1039
1040enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
1041{
1042 switch (ring) {
1043 case 0:
1044 return AMDGPU_RING_PRIO_0;
1045 case 1:
1046 return AMDGPU_RING_PRIO_1;
1047 case 2:
1048 return AMDGPU_RING_PRIO_2;
1049 default:
1050 return AMDGPU_RING_PRIO_0;
1051 }
1052}
1053
1054void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev)
1055{
1056 int i;
1057 unsigned int idx;
1058
1059 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1060 const struct common_firmware_header *hdr;
1061
1062 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1063 if (adev->vcn.harvest_config & (1 << i))
1064 continue;
1065
1066 hdr = (const struct common_firmware_header *)adev->vcn.fw[i]->data;
1067 /* currently only support 2 FW instances */
1068 if (i >= 2) {
1069 dev_info(adev->dev, "More then 2 VCN FW instances!\n");
1070 break;
1071 }
1072 idx = AMDGPU_UCODE_ID_VCN + i;
1073 adev->firmware.ucode[idx].ucode_id = idx;
1074 adev->firmware.ucode[idx].fw = adev->vcn.fw[i];
1075 adev->firmware.fw_size +=
1076 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
1077
1078 if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
1079 IP_VERSION(4, 0, 3))
1080 break;
1081 }
1082 }
1083}
1084
1085/*
1086 * debugfs for mapping vcn firmware log buffer.
1087 */
1088#if defined(CONFIG_DEBUG_FS)
1089static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf,
1090 size_t size, loff_t *pos)
1091{
1092 struct amdgpu_vcn_inst *vcn;
1093 void *log_buf;
1094 volatile struct amdgpu_vcn_fwlog *plog;
1095 unsigned int read_pos, write_pos, available, i, read_bytes = 0;
1096 unsigned int read_num[2] = {0};
1097
1098 vcn = file_inode(f)->i_private;
1099 if (!vcn)
1100 return -ENODEV;
1101
1102 if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log)
1103 return -EFAULT;
1104
1105 log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1106
1107 plog = (volatile struct amdgpu_vcn_fwlog *)log_buf;
1108 read_pos = plog->rptr;
1109 write_pos = plog->wptr;
1110
1111 if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE)
1112 return -EFAULT;
1113
1114 if (!size || (read_pos == write_pos))
1115 return 0;
1116
1117 if (write_pos > read_pos) {
1118 available = write_pos - read_pos;
1119 read_num[0] = min_t(size_t, size, available);
1120 } else {
1121 read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos;
1122 available = read_num[0] + write_pos - plog->header_size;
1123 if (size > available)
1124 read_num[1] = write_pos - plog->header_size;
1125 else if (size > read_num[0])
1126 read_num[1] = size - read_num[0];
1127 else
1128 read_num[0] = size;
1129 }
1130
1131 for (i = 0; i < 2; i++) {
1132 if (read_num[i]) {
1133 if (read_pos == AMDGPU_VCNFW_LOG_SIZE)
1134 read_pos = plog->header_size;
1135 if (read_num[i] == copy_to_user((buf + read_bytes),
1136 (log_buf + read_pos), read_num[i]))
1137 return -EFAULT;
1138
1139 read_bytes += read_num[i];
1140 read_pos += read_num[i];
1141 }
1142 }
1143
1144 plog->rptr = read_pos;
1145 *pos += read_bytes;
1146 return read_bytes;
1147}
1148
1149static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = {
1150 .owner = THIS_MODULE,
1151 .read = amdgpu_debugfs_vcn_fwlog_read,
1152 .llseek = default_llseek
1153};
1154#endif
1155
1156void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i,
1157 struct amdgpu_vcn_inst *vcn)
1158{
1159#if defined(CONFIG_DEBUG_FS)
1160 struct drm_minor *minor = adev_to_drm(adev)->primary;
1161 struct dentry *root = minor->debugfs_root;
1162 char name[32];
1163
1164 sprintf(name, "amdgpu_vcn_%d_fwlog", i);
1165 debugfs_create_file_size(name, S_IFREG | 0444, root, vcn,
1166 &amdgpu_debugfs_vcnfwlog_fops,
1167 AMDGPU_VCNFW_LOG_SIZE);
1168#endif
1169}
1170
1171void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn)
1172{
1173#if defined(CONFIG_DEBUG_FS)
1174 volatile uint32_t *flag = vcn->fw_shared.cpu_addr;
1175 void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1176 uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size;
1177 volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr;
1178 volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr
1179 + vcn->fw_shared.log_offset;
1180 *flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG);
1181 fw_log->is_enabled = 1;
1182 fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF);
1183 fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32);
1184 fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE);
1185
1186 log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog);
1187 log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE;
1188 log_buf->rptr = log_buf->header_size;
1189 log_buf->wptr = log_buf->header_size;
1190 log_buf->wrapped = 0;
1191#endif
1192}
1193
1194int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev,
1195 struct amdgpu_irq_src *source,
1196 struct amdgpu_iv_entry *entry)
1197{
1198 struct ras_common_if *ras_if = adev->vcn.ras_if;
1199 struct ras_dispatch_if ih_data = {
1200 .entry = entry,
1201 };
1202
1203 if (!ras_if)
1204 return 0;
1205
1206 if (!amdgpu_sriov_vf(adev)) {
1207 ih_data.head = *ras_if;
1208 amdgpu_ras_interrupt_dispatch(adev, &ih_data);
1209 } else {
1210 if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
1211 adev->virt.ops->ras_poison_handler(adev, ras_if->block);
1212 else
1213 dev_warn(adev->dev,
1214 "No ras_poison_handler interface in SRIOV for VCN!\n");
1215 }
1216
1217 return 0;
1218}
1219
1220int amdgpu_vcn_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
1221{
1222 int r, i;
1223
1224 r = amdgpu_ras_block_late_init(adev, ras_block);
1225 if (r)
1226 return r;
1227
1228 if (amdgpu_ras_is_supported(adev, ras_block->block)) {
1229 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1230 if (adev->vcn.harvest_config & (1 << i) ||
1231 !adev->vcn.inst[i].ras_poison_irq.funcs)
1232 continue;
1233
1234 r = amdgpu_irq_get(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
1235 if (r)
1236 goto late_fini;
1237 }
1238 }
1239 return 0;
1240
1241late_fini:
1242 amdgpu_ras_block_late_fini(adev, ras_block);
1243 return r;
1244}
1245
1246int amdgpu_vcn_ras_sw_init(struct amdgpu_device *adev)
1247{
1248 int err;
1249 struct amdgpu_vcn_ras *ras;
1250
1251 if (!adev->vcn.ras)
1252 return 0;
1253
1254 ras = adev->vcn.ras;
1255 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
1256 if (err) {
1257 dev_err(adev->dev, "Failed to register vcn ras block!\n");
1258 return err;
1259 }
1260
1261 strcpy(ras->ras_block.ras_comm.name, "vcn");
1262 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
1263 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
1264 adev->vcn.ras_if = &ras->ras_block.ras_comm;
1265
1266 if (!ras->ras_block.ras_late_init)
1267 ras->ras_block.ras_late_init = amdgpu_vcn_ras_late_init;
1268
1269 return 0;
1270}
1271
1272int amdgpu_vcn_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
1273 enum AMDGPU_UCODE_ID ucode_id)
1274{
1275 struct amdgpu_firmware_info ucode = {
1276 .ucode_id = (ucode_id ? ucode_id :
1277 (inst_idx ? AMDGPU_UCODE_ID_VCN1_RAM :
1278 AMDGPU_UCODE_ID_VCN0_RAM)),
1279 .mc_addr = adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1280 .ucode_size = ((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1281 (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr),
1282 };
1283
1284 return psp_execute_ip_fw_load(&adev->psp, &ucode);
1285}
1286
1287static ssize_t amdgpu_get_vcn_reset_mask(struct device *dev,
1288 struct device_attribute *attr,
1289 char *buf)
1290{
1291 struct drm_device *ddev = dev_get_drvdata(dev);
1292 struct amdgpu_device *adev = drm_to_adev(ddev);
1293
1294 if (!adev)
1295 return -ENODEV;
1296
1297 return amdgpu_show_reset_mask(buf, adev->vcn.supported_reset);
1298}
1299
1300static DEVICE_ATTR(vcn_reset_mask, 0444,
1301 amdgpu_get_vcn_reset_mask, NULL);
1302
1303int amdgpu_vcn_sysfs_reset_mask_init(struct amdgpu_device *adev)
1304{
1305 int r = 0;
1306
1307 if (adev->vcn.num_vcn_inst) {
1308 r = device_create_file(adev->dev, &dev_attr_vcn_reset_mask);
1309 if (r)
1310 return r;
1311 }
1312
1313 return r;
1314}
1315
1316void amdgpu_vcn_sysfs_reset_mask_fini(struct amdgpu_device *adev)
1317{
1318 if (adev->dev->kobj.sd) {
1319 if (adev->vcn.num_vcn_inst)
1320 device_remove_file(adev->dev, &dev_attr_vcn_reset_mask);
1321 }
1322}