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#include <drm/drm_drv.h>
31
32#include "amdgpu.h"
33#include "amdgpu_pm.h"
34#include "amdgpu_vcn.h"
35#include "soc15d.h"
36
37/* Firmware Names */
38#define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
39#define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
40#define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
41#define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
42#define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
43#define FIRMWARE_GREEN_SARDINE "amdgpu/green_sardine_vcn.bin"
44#define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
45#define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
46#define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
47#define FIRMWARE_SIENNA_CICHLID "amdgpu/sienna_cichlid_vcn.bin"
48#define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin"
49#define FIRMWARE_VANGOGH "amdgpu/vangogh_vcn.bin"
50#define FIRMWARE_DIMGREY_CAVEFISH "amdgpu/dimgrey_cavefish_vcn.bin"
51#define FIRMWARE_ALDEBARAN "amdgpu/aldebaran_vcn.bin"
52#define FIRMWARE_BEIGE_GOBY "amdgpu/beige_goby_vcn.bin"
53#define FIRMWARE_YELLOW_CARP "amdgpu/yellow_carp_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_GREEN_SARDINE);
61MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
62MODULE_FIRMWARE(FIRMWARE_NAVI10);
63MODULE_FIRMWARE(FIRMWARE_NAVI14);
64MODULE_FIRMWARE(FIRMWARE_NAVI12);
65MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
66MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
67MODULE_FIRMWARE(FIRMWARE_VANGOGH);
68MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
69MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
70MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
71
72static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
73
74int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
75{
76 unsigned long bo_size;
77 const char *fw_name;
78 const struct common_firmware_header *hdr;
79 unsigned char fw_check;
80 int i, r;
81
82 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
83 mutex_init(&adev->vcn.vcn_pg_lock);
84 mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
85 atomic_set(&adev->vcn.total_submission_cnt, 0);
86 for (i = 0; i < adev->vcn.num_vcn_inst; i++)
87 atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
88
89 switch (adev->asic_type) {
90 case CHIP_RAVEN:
91 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
92 fw_name = FIRMWARE_RAVEN2;
93 else if (adev->apu_flags & AMD_APU_IS_PICASSO)
94 fw_name = FIRMWARE_PICASSO;
95 else
96 fw_name = FIRMWARE_RAVEN;
97 break;
98 case CHIP_ARCTURUS:
99 fw_name = FIRMWARE_ARCTURUS;
100 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
101 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
102 adev->vcn.indirect_sram = true;
103 break;
104 case CHIP_RENOIR:
105 if (adev->apu_flags & AMD_APU_IS_RENOIR)
106 fw_name = FIRMWARE_RENOIR;
107 else
108 fw_name = FIRMWARE_GREEN_SARDINE;
109
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_ALDEBARAN:
115 fw_name = FIRMWARE_ALDEBARAN;
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_NAVI10:
121 fw_name = FIRMWARE_NAVI10;
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 case CHIP_NAVI14:
127 fw_name = FIRMWARE_NAVI14;
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 break;
132 case CHIP_NAVI12:
133 fw_name = FIRMWARE_NAVI12;
134 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
135 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
136 adev->vcn.indirect_sram = true;
137 break;
138 case CHIP_SIENNA_CICHLID:
139 fw_name = FIRMWARE_SIENNA_CICHLID;
140 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
141 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
142 adev->vcn.indirect_sram = true;
143 break;
144 case CHIP_NAVY_FLOUNDER:
145 fw_name = FIRMWARE_NAVY_FLOUNDER;
146 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
147 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
148 adev->vcn.indirect_sram = true;
149 break;
150 case CHIP_VANGOGH:
151 fw_name = FIRMWARE_VANGOGH;
152 break;
153 case CHIP_DIMGREY_CAVEFISH:
154 fw_name = FIRMWARE_DIMGREY_CAVEFISH;
155 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
156 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
157 adev->vcn.indirect_sram = true;
158 break;
159 case CHIP_BEIGE_GOBY:
160 fw_name = FIRMWARE_BEIGE_GOBY;
161 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
162 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
163 adev->vcn.indirect_sram = true;
164 break;
165 case CHIP_YELLOW_CARP:
166 fw_name = FIRMWARE_YELLOW_CARP;
167 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
168 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
169 adev->vcn.indirect_sram = true;
170 break;
171 default:
172 return -EINVAL;
173 }
174
175 r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
176 if (r) {
177 dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
178 fw_name);
179 return r;
180 }
181
182 r = amdgpu_ucode_validate(adev->vcn.fw);
183 if (r) {
184 dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
185 fw_name);
186 release_firmware(adev->vcn.fw);
187 adev->vcn.fw = NULL;
188 return r;
189 }
190
191 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
192 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
193
194 /* Bit 20-23, it is encode major and non-zero for new naming convention.
195 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
196 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
197 * is zero in old naming convention, this field is always zero so far.
198 * These four bits are used to tell which naming convention is present.
199 */
200 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
201 if (fw_check) {
202 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
203
204 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
205 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
206 enc_major = fw_check;
207 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
208 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
209 DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
210 enc_major, enc_minor, dec_ver, vep, fw_rev);
211 } else {
212 unsigned int version_major, version_minor, family_id;
213
214 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
215 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
216 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
217 DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
218 version_major, version_minor, family_id);
219 }
220
221 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
222 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
223 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
224 bo_size += AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
225
226 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
227 if (adev->vcn.harvest_config & (1 << i))
228 continue;
229
230 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
231 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
232 &adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
233 if (r) {
234 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
235 return r;
236 }
237
238 adev->vcn.inst[i].fw_shared_cpu_addr = adev->vcn.inst[i].cpu_addr +
239 bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
240 adev->vcn.inst[i].fw_shared_gpu_addr = adev->vcn.inst[i].gpu_addr +
241 bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
242
243 if (adev->vcn.indirect_sram) {
244 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
245 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
246 &adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
247 if (r) {
248 dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
249 return r;
250 }
251 }
252 }
253
254 return 0;
255}
256
257int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
258{
259 int i, j;
260
261 cancel_delayed_work_sync(&adev->vcn.idle_work);
262
263 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
264 if (adev->vcn.harvest_config & (1 << j))
265 continue;
266
267 if (adev->vcn.indirect_sram) {
268 amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo,
269 &adev->vcn.inst[j].dpg_sram_gpu_addr,
270 (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
271 }
272 kvfree(adev->vcn.inst[j].saved_bo);
273
274 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
275 &adev->vcn.inst[j].gpu_addr,
276 (void **)&adev->vcn.inst[j].cpu_addr);
277
278 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
279
280 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
281 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
282 }
283
284 release_firmware(adev->vcn.fw);
285 mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
286 mutex_destroy(&adev->vcn.vcn_pg_lock);
287
288 return 0;
289}
290
291bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
292{
293 bool ret = false;
294
295 int major;
296 int minor;
297 int revision;
298
299 /* if cannot find IP data, then this VCN does not exist */
300 if (amdgpu_discovery_get_vcn_version(adev, vcn_instance, &major, &minor, &revision) != 0)
301 return true;
302
303 if ((type == VCN_ENCODE_RING) && (revision & VCN_BLOCK_ENCODE_DISABLE_MASK)) {
304 ret = true;
305 } else if ((type == VCN_DECODE_RING) && (revision & VCN_BLOCK_DECODE_DISABLE_MASK)) {
306 ret = true;
307 } else if ((type == VCN_UNIFIED_RING) && (revision & VCN_BLOCK_QUEUE_DISABLE_MASK)) {
308 ret = true;
309 }
310
311 return ret;
312}
313
314int amdgpu_vcn_suspend(struct amdgpu_device *adev)
315{
316 unsigned size;
317 void *ptr;
318 int i, idx;
319
320 cancel_delayed_work_sync(&adev->vcn.idle_work);
321
322 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
323 if (adev->vcn.harvest_config & (1 << i))
324 continue;
325 if (adev->vcn.inst[i].vcpu_bo == NULL)
326 return 0;
327
328 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
329 ptr = adev->vcn.inst[i].cpu_addr;
330
331 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
332 if (!adev->vcn.inst[i].saved_bo)
333 return -ENOMEM;
334
335 if (drm_dev_enter(&adev->ddev, &idx)) {
336 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
337 drm_dev_exit(idx);
338 }
339 }
340 return 0;
341}
342
343int amdgpu_vcn_resume(struct amdgpu_device *adev)
344{
345 unsigned size;
346 void *ptr;
347 int i, idx;
348
349 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
350 if (adev->vcn.harvest_config & (1 << i))
351 continue;
352 if (adev->vcn.inst[i].vcpu_bo == NULL)
353 return -EINVAL;
354
355 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
356 ptr = adev->vcn.inst[i].cpu_addr;
357
358 if (adev->vcn.inst[i].saved_bo != NULL) {
359 if (drm_dev_enter(&adev->ddev, &idx)) {
360 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
361 drm_dev_exit(idx);
362 }
363 kvfree(adev->vcn.inst[i].saved_bo);
364 adev->vcn.inst[i].saved_bo = NULL;
365 } else {
366 const struct common_firmware_header *hdr;
367 unsigned offset;
368
369 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
370 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
371 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
372 if (drm_dev_enter(&adev->ddev, &idx)) {
373 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
374 le32_to_cpu(hdr->ucode_size_bytes));
375 drm_dev_exit(idx);
376 }
377 size -= le32_to_cpu(hdr->ucode_size_bytes);
378 ptr += le32_to_cpu(hdr->ucode_size_bytes);
379 }
380 memset_io(ptr, 0, size);
381 }
382 }
383 return 0;
384}
385
386static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
387{
388 struct amdgpu_device *adev =
389 container_of(work, struct amdgpu_device, vcn.idle_work.work);
390 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
391 unsigned int i, j;
392 int r = 0;
393
394 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
395 if (adev->vcn.harvest_config & (1 << j))
396 continue;
397
398 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
399 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
400 }
401
402 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
403 struct dpg_pause_state new_state;
404
405 if (fence[j] ||
406 unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
407 new_state.fw_based = VCN_DPG_STATE__PAUSE;
408 else
409 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
410
411 adev->vcn.pause_dpg_mode(adev, j, &new_state);
412 }
413
414 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
415 fences += fence[j];
416 }
417
418 if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
419 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
420 AMD_PG_STATE_GATE);
421 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
422 false);
423 if (r)
424 dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
425 } else {
426 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
427 }
428}
429
430void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
431{
432 struct amdgpu_device *adev = ring->adev;
433 int r = 0;
434
435 atomic_inc(&adev->vcn.total_submission_cnt);
436
437 if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
438 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
439 true);
440 if (r)
441 dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
442 }
443
444 mutex_lock(&adev->vcn.vcn_pg_lock);
445 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
446 AMD_PG_STATE_UNGATE);
447
448 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
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 if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
475 ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
476 atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
477
478 atomic_dec(&ring->adev->vcn.total_submission_cnt);
479
480 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
481}
482
483int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
484{
485 struct amdgpu_device *adev = ring->adev;
486 uint32_t tmp = 0;
487 unsigned i;
488 int r;
489
490 /* VCN in SRIOV does not support direct register read/write */
491 if (amdgpu_sriov_vf(adev))
492 return 0;
493
494 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
495 r = amdgpu_ring_alloc(ring, 3);
496 if (r)
497 return r;
498 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
499 amdgpu_ring_write(ring, 0xDEADBEEF);
500 amdgpu_ring_commit(ring);
501 for (i = 0; i < adev->usec_timeout; i++) {
502 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
503 if (tmp == 0xDEADBEEF)
504 break;
505 udelay(1);
506 }
507
508 if (i >= adev->usec_timeout)
509 r = -ETIMEDOUT;
510
511 return r;
512}
513
514int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
515{
516 struct amdgpu_device *adev = ring->adev;
517 uint32_t rptr;
518 unsigned int i;
519 int r;
520
521 if (amdgpu_sriov_vf(adev))
522 return 0;
523
524 r = amdgpu_ring_alloc(ring, 16);
525 if (r)
526 return r;
527
528 rptr = amdgpu_ring_get_rptr(ring);
529
530 amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
531 amdgpu_ring_commit(ring);
532
533 for (i = 0; i < adev->usec_timeout; i++) {
534 if (amdgpu_ring_get_rptr(ring) != rptr)
535 break;
536 udelay(1);
537 }
538
539 if (i >= adev->usec_timeout)
540 r = -ETIMEDOUT;
541
542 return r;
543}
544
545static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
546 struct amdgpu_bo *bo,
547 struct dma_fence **fence)
548{
549 struct amdgpu_device *adev = ring->adev;
550 struct dma_fence *f = NULL;
551 struct amdgpu_job *job;
552 struct amdgpu_ib *ib;
553 uint64_t addr;
554 void *msg = NULL;
555 int i, r;
556
557 r = amdgpu_job_alloc_with_ib(adev, 64,
558 AMDGPU_IB_POOL_DIRECT, &job);
559 if (r)
560 goto err;
561
562 ib = &job->ibs[0];
563 addr = amdgpu_bo_gpu_offset(bo);
564 msg = amdgpu_bo_kptr(bo);
565 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
566 ib->ptr[1] = addr;
567 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
568 ib->ptr[3] = addr >> 32;
569 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
570 ib->ptr[5] = 0;
571 for (i = 6; i < 16; i += 2) {
572 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
573 ib->ptr[i+1] = 0;
574 }
575 ib->length_dw = 16;
576
577 r = amdgpu_job_submit_direct(job, ring, &f);
578 if (r)
579 goto err_free;
580
581 amdgpu_bo_fence(bo, f, false);
582 amdgpu_bo_unreserve(bo);
583 amdgpu_bo_free_kernel(&bo, NULL, (void **)&msg);
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);
593
594err:
595 amdgpu_bo_unreserve(bo);
596 amdgpu_bo_free_kernel(&bo, NULL, (void **)&msg);
597 return r;
598}
599
600static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
601 struct amdgpu_bo **bo)
602{
603 struct amdgpu_device *adev = ring->adev;
604 uint32_t *msg;
605 int r, i;
606
607 *bo = NULL;
608 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
609 AMDGPU_GEM_DOMAIN_VRAM,
610 bo, NULL, (void **)&msg);
611 if (r)
612 return r;
613
614 msg[0] = cpu_to_le32(0x00000028);
615 msg[1] = cpu_to_le32(0x00000038);
616 msg[2] = cpu_to_le32(0x00000001);
617 msg[3] = cpu_to_le32(0x00000000);
618 msg[4] = cpu_to_le32(handle);
619 msg[5] = cpu_to_le32(0x00000000);
620 msg[6] = cpu_to_le32(0x00000001);
621 msg[7] = cpu_to_le32(0x00000028);
622 msg[8] = cpu_to_le32(0x00000010);
623 msg[9] = cpu_to_le32(0x00000000);
624 msg[10] = cpu_to_le32(0x00000007);
625 msg[11] = cpu_to_le32(0x00000000);
626 msg[12] = cpu_to_le32(0x00000780);
627 msg[13] = cpu_to_le32(0x00000440);
628 for (i = 14; i < 1024; ++i)
629 msg[i] = cpu_to_le32(0x0);
630
631 return 0;
632}
633
634static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
635 struct amdgpu_bo **bo)
636{
637 struct amdgpu_device *adev = ring->adev;
638 uint32_t *msg;
639 int r, i;
640
641 *bo = NULL;
642 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
643 AMDGPU_GEM_DOMAIN_VRAM,
644 bo, NULL, (void **)&msg);
645 if (r)
646 return r;
647
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_bo *bo;
664 long r;
665
666 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
667 if (r)
668 goto error;
669
670 r = amdgpu_vcn_dec_send_msg(ring, bo, NULL);
671 if (r)
672 goto error;
673 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &bo);
674 if (r)
675 goto error;
676
677 r = amdgpu_vcn_dec_send_msg(ring, bo, &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 int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
693 struct amdgpu_bo *bo,
694 struct dma_fence **fence)
695{
696 struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
697 const unsigned int ib_size_dw = 64;
698 struct amdgpu_device *adev = ring->adev;
699 struct dma_fence *f = NULL;
700 struct amdgpu_job *job;
701 struct amdgpu_ib *ib;
702 uint64_t addr;
703 int i, r;
704
705 r = amdgpu_job_alloc_with_ib(adev, ib_size_dw * 4,
706 AMDGPU_IB_POOL_DIRECT, &job);
707 if (r)
708 goto err;
709
710 ib = &job->ibs[0];
711 addr = amdgpu_bo_gpu_offset(bo);
712 ib->length_dw = 0;
713
714 ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
715 ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
716 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
717 ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
718 memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
719
720 decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
721 decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
722 decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
723
724 for (i = ib->length_dw; i < ib_size_dw; ++i)
725 ib->ptr[i] = 0x0;
726
727 r = amdgpu_job_submit_direct(job, ring, &f);
728 if (r)
729 goto err_free;
730
731 amdgpu_bo_fence(bo, f, false);
732 amdgpu_bo_unreserve(bo);
733 amdgpu_bo_unref(&bo);
734
735 if (fence)
736 *fence = dma_fence_get(f);
737 dma_fence_put(f);
738
739 return 0;
740
741err_free:
742 amdgpu_job_free(job);
743
744err:
745 amdgpu_bo_unreserve(bo);
746 amdgpu_bo_unref(&bo);
747 return r;
748}
749
750int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
751{
752 struct dma_fence *fence = NULL;
753 struct amdgpu_bo *bo;
754 long r;
755
756 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
757 if (r)
758 goto error;
759
760 r = amdgpu_vcn_dec_sw_send_msg(ring, bo, NULL);
761 if (r)
762 goto error;
763 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &bo);
764 if (r)
765 goto error;
766
767 r = amdgpu_vcn_dec_sw_send_msg(ring, bo, &fence);
768 if (r)
769 goto error;
770
771 r = dma_fence_wait_timeout(fence, false, timeout);
772 if (r == 0)
773 r = -ETIMEDOUT;
774 else if (r > 0)
775 r = 0;
776
777 dma_fence_put(fence);
778error:
779 return r;
780}
781
782int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
783{
784 struct amdgpu_device *adev = ring->adev;
785 uint32_t rptr;
786 unsigned i;
787 int r;
788
789 if (amdgpu_sriov_vf(adev))
790 return 0;
791
792 r = amdgpu_ring_alloc(ring, 16);
793 if (r)
794 return r;
795
796 rptr = amdgpu_ring_get_rptr(ring);
797
798 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
799 amdgpu_ring_commit(ring);
800
801 for (i = 0; i < adev->usec_timeout; i++) {
802 if (amdgpu_ring_get_rptr(ring) != rptr)
803 break;
804 udelay(1);
805 }
806
807 if (i >= adev->usec_timeout)
808 r = -ETIMEDOUT;
809
810 return r;
811}
812
813static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
814 struct amdgpu_bo *bo,
815 struct dma_fence **fence)
816{
817 const unsigned ib_size_dw = 16;
818 struct amdgpu_job *job;
819 struct amdgpu_ib *ib;
820 struct dma_fence *f = NULL;
821 uint64_t addr;
822 int i, r;
823
824 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
825 AMDGPU_IB_POOL_DIRECT, &job);
826 if (r)
827 return r;
828
829 ib = &job->ibs[0];
830 addr = amdgpu_bo_gpu_offset(bo);
831
832 ib->length_dw = 0;
833 ib->ptr[ib->length_dw++] = 0x00000018;
834 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
835 ib->ptr[ib->length_dw++] = handle;
836 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
837 ib->ptr[ib->length_dw++] = addr;
838 ib->ptr[ib->length_dw++] = 0x0000000b;
839
840 ib->ptr[ib->length_dw++] = 0x00000014;
841 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
842 ib->ptr[ib->length_dw++] = 0x0000001c;
843 ib->ptr[ib->length_dw++] = 0x00000000;
844 ib->ptr[ib->length_dw++] = 0x00000000;
845
846 ib->ptr[ib->length_dw++] = 0x00000008;
847 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
848
849 for (i = ib->length_dw; i < ib_size_dw; ++i)
850 ib->ptr[i] = 0x0;
851
852 r = amdgpu_job_submit_direct(job, ring, &f);
853 if (r)
854 goto err;
855
856 if (fence)
857 *fence = dma_fence_get(f);
858 dma_fence_put(f);
859
860 return 0;
861
862err:
863 amdgpu_job_free(job);
864 return r;
865}
866
867static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
868 struct amdgpu_bo *bo,
869 struct dma_fence **fence)
870{
871 const unsigned ib_size_dw = 16;
872 struct amdgpu_job *job;
873 struct amdgpu_ib *ib;
874 struct dma_fence *f = NULL;
875 uint64_t addr;
876 int i, r;
877
878 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
879 AMDGPU_IB_POOL_DIRECT, &job);
880 if (r)
881 return r;
882
883 ib = &job->ibs[0];
884 addr = amdgpu_bo_gpu_offset(bo);
885
886 ib->length_dw = 0;
887 ib->ptr[ib->length_dw++] = 0x00000018;
888 ib->ptr[ib->length_dw++] = 0x00000001;
889 ib->ptr[ib->length_dw++] = handle;
890 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
891 ib->ptr[ib->length_dw++] = addr;
892 ib->ptr[ib->length_dw++] = 0x0000000b;
893
894 ib->ptr[ib->length_dw++] = 0x00000014;
895 ib->ptr[ib->length_dw++] = 0x00000002;
896 ib->ptr[ib->length_dw++] = 0x0000001c;
897 ib->ptr[ib->length_dw++] = 0x00000000;
898 ib->ptr[ib->length_dw++] = 0x00000000;
899
900 ib->ptr[ib->length_dw++] = 0x00000008;
901 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
902
903 for (i = ib->length_dw; i < ib_size_dw; ++i)
904 ib->ptr[i] = 0x0;
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
921int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
922{
923 struct dma_fence *fence = NULL;
924 struct amdgpu_bo *bo = NULL;
925 long r;
926
927 r = amdgpu_bo_create_reserved(ring->adev, 128 * 1024, PAGE_SIZE,
928 AMDGPU_GEM_DOMAIN_VRAM,
929 &bo, NULL, NULL);
930 if (r)
931 return r;
932
933 r = amdgpu_vcn_enc_get_create_msg(ring, 1, bo, NULL);
934 if (r)
935 goto error;
936
937 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, bo, &fence);
938 if (r)
939 goto error;
940
941 r = dma_fence_wait_timeout(fence, false, timeout);
942 if (r == 0)
943 r = -ETIMEDOUT;
944 else if (r > 0)
945 r = 0;
946
947error:
948 dma_fence_put(fence);
949 amdgpu_bo_unreserve(bo);
950 amdgpu_bo_free_kernel(&bo, NULL, NULL);
951
952 return r;
953}
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}