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 <linux/debugfs.h>
31#include <drm/drm_drv.h>
32
33#include "amdgpu.h"
34#include "amdgpu_pm.h"
35#include "amdgpu_vcn.h"
36#include "soc15d.h"
37
38/* Firmware Names */
39#define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
40#define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
41#define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
42#define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
43#define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
44#define FIRMWARE_GREEN_SARDINE "amdgpu/green_sardine_vcn.bin"
45#define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
46#define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
47#define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
48#define FIRMWARE_SIENNA_CICHLID "amdgpu/sienna_cichlid_vcn.bin"
49#define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin"
50#define FIRMWARE_VANGOGH "amdgpu/vangogh_vcn.bin"
51#define FIRMWARE_DIMGREY_CAVEFISH "amdgpu/dimgrey_cavefish_vcn.bin"
52#define FIRMWARE_ALDEBARAN "amdgpu/aldebaran_vcn.bin"
53#define FIRMWARE_BEIGE_GOBY "amdgpu/beige_goby_vcn.bin"
54#define FIRMWARE_YELLOW_CARP "amdgpu/yellow_carp_vcn.bin"
55#define FIRMWARE_VCN_3_1_2 "amdgpu/vcn_3_1_2.bin"
56#define FIRMWARE_VCN4_0_0 "amdgpu/vcn_4_0_0.bin"
57#define FIRMWARE_VCN4_0_2 "amdgpu/vcn_4_0_2.bin"
58#define FIRMWARE_VCN4_0_4 "amdgpu/vcn_4_0_4.bin"
59
60MODULE_FIRMWARE(FIRMWARE_RAVEN);
61MODULE_FIRMWARE(FIRMWARE_PICASSO);
62MODULE_FIRMWARE(FIRMWARE_RAVEN2);
63MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
64MODULE_FIRMWARE(FIRMWARE_RENOIR);
65MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
66MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
67MODULE_FIRMWARE(FIRMWARE_NAVI10);
68MODULE_FIRMWARE(FIRMWARE_NAVI14);
69MODULE_FIRMWARE(FIRMWARE_NAVI12);
70MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
71MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
72MODULE_FIRMWARE(FIRMWARE_VANGOGH);
73MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
74MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
75MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
76MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
77MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
78MODULE_FIRMWARE(FIRMWARE_VCN4_0_2);
79MODULE_FIRMWARE(FIRMWARE_VCN4_0_4);
80
81static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
82
83int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
84{
85 unsigned long bo_size;
86 const char *fw_name;
87 const struct common_firmware_header *hdr;
88 unsigned char fw_check;
89 unsigned int fw_shared_size, log_offset;
90 int i, r;
91
92 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
93 mutex_init(&adev->vcn.vcn_pg_lock);
94 mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
95 atomic_set(&adev->vcn.total_submission_cnt, 0);
96 for (i = 0; i < adev->vcn.num_vcn_inst; i++)
97 atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
98
99 switch (adev->ip_versions[UVD_HWIP][0]) {
100 case IP_VERSION(1, 0, 0):
101 case IP_VERSION(1, 0, 1):
102 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
103 fw_name = FIRMWARE_RAVEN2;
104 else if (adev->apu_flags & AMD_APU_IS_PICASSO)
105 fw_name = FIRMWARE_PICASSO;
106 else
107 fw_name = FIRMWARE_RAVEN;
108 break;
109 case IP_VERSION(2, 5, 0):
110 fw_name = FIRMWARE_ARCTURUS;
111 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
112 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
113 adev->vcn.indirect_sram = true;
114 break;
115 case IP_VERSION(2, 2, 0):
116 if (adev->apu_flags & AMD_APU_IS_RENOIR)
117 fw_name = FIRMWARE_RENOIR;
118 else
119 fw_name = FIRMWARE_GREEN_SARDINE;
120
121 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
122 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
123 adev->vcn.indirect_sram = true;
124 break;
125 case IP_VERSION(2, 6, 0):
126 fw_name = FIRMWARE_ALDEBARAN;
127 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
128 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
129 adev->vcn.indirect_sram = true;
130 break;
131 case IP_VERSION(2, 0, 0):
132 fw_name = FIRMWARE_NAVI10;
133 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
134 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
135 adev->vcn.indirect_sram = true;
136 break;
137 case IP_VERSION(2, 0, 2):
138 if (adev->asic_type == CHIP_NAVI12)
139 fw_name = FIRMWARE_NAVI12;
140 else
141 fw_name = FIRMWARE_NAVI14;
142 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
143 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
144 adev->vcn.indirect_sram = true;
145 break;
146 case IP_VERSION(3, 0, 0):
147 case IP_VERSION(3, 0, 64):
148 case IP_VERSION(3, 0, 192):
149 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(10, 3, 0))
150 fw_name = FIRMWARE_SIENNA_CICHLID;
151 else
152 fw_name = FIRMWARE_NAVY_FLOUNDER;
153 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
154 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
155 adev->vcn.indirect_sram = true;
156 break;
157 case IP_VERSION(3, 0, 2):
158 fw_name = FIRMWARE_VANGOGH;
159 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
160 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
161 adev->vcn.indirect_sram = true;
162 break;
163 case IP_VERSION(3, 0, 16):
164 fw_name = FIRMWARE_DIMGREY_CAVEFISH;
165 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
166 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
167 adev->vcn.indirect_sram = true;
168 break;
169 case IP_VERSION(3, 0, 33):
170 fw_name = FIRMWARE_BEIGE_GOBY;
171 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
172 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
173 adev->vcn.indirect_sram = true;
174 break;
175 case IP_VERSION(3, 1, 1):
176 fw_name = FIRMWARE_YELLOW_CARP;
177 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
178 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
179 adev->vcn.indirect_sram = true;
180 break;
181 case IP_VERSION(3, 1, 2):
182 fw_name = FIRMWARE_VCN_3_1_2;
183 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
184 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
185 adev->vcn.indirect_sram = true;
186 break;
187 case IP_VERSION(4, 0, 0):
188 fw_name = FIRMWARE_VCN4_0_0;
189 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
190 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
191 adev->vcn.indirect_sram = true;
192 break;
193 case IP_VERSION(4, 0, 2):
194 fw_name = FIRMWARE_VCN4_0_2;
195 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
196 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
197 adev->vcn.indirect_sram = true;
198 break;
199 case IP_VERSION(4, 0, 4):
200 fw_name = FIRMWARE_VCN4_0_4;
201 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
202 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
203 adev->vcn.indirect_sram = true;
204 break;
205 default:
206 return -EINVAL;
207 }
208
209 r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
210 if (r) {
211 dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
212 fw_name);
213 return r;
214 }
215
216 r = amdgpu_ucode_validate(adev->vcn.fw);
217 if (r) {
218 dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
219 fw_name);
220 release_firmware(adev->vcn.fw);
221 adev->vcn.fw = NULL;
222 return r;
223 }
224
225 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
226 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
227
228 /* Bit 20-23, it is encode major and non-zero for new naming convention.
229 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
230 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
231 * is zero in old naming convention, this field is always zero so far.
232 * These four bits are used to tell which naming convention is present.
233 */
234 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
235 if (fw_check) {
236 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
237
238 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
239 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
240 enc_major = fw_check;
241 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
242 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
243 DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
244 enc_major, enc_minor, dec_ver, vep, fw_rev);
245 } else {
246 unsigned int version_major, version_minor, family_id;
247
248 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
249 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
250 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
251 DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
252 version_major, version_minor, family_id);
253 }
254
255 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
256 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
257 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
258
259 if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0)){
260 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
261 log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
262 } else {
263 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
264 log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
265 }
266
267 bo_size += fw_shared_size;
268
269 if (amdgpu_vcnfw_log)
270 bo_size += AMDGPU_VCNFW_LOG_SIZE;
271
272 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
273 if (adev->vcn.harvest_config & (1 << i))
274 continue;
275
276 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
277 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
278 &adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
279 if (r) {
280 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
281 return r;
282 }
283
284 adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr +
285 bo_size - fw_shared_size;
286 adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr +
287 bo_size - fw_shared_size;
288
289 adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size;
290
291 if (amdgpu_vcnfw_log) {
292 adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
293 adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
294 adev->vcn.inst[i].fw_shared.log_offset = log_offset;
295 }
296
297 if (adev->vcn.indirect_sram) {
298 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
299 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
300 &adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
301 if (r) {
302 dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
303 return r;
304 }
305 }
306 }
307
308 return 0;
309}
310
311int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
312{
313 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 if (adev->vcn.indirect_sram) {
320 amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo,
321 &adev->vcn.inst[j].dpg_sram_gpu_addr,
322 (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
323 }
324 kvfree(adev->vcn.inst[j].saved_bo);
325
326 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
327 &adev->vcn.inst[j].gpu_addr,
328 (void **)&adev->vcn.inst[j].cpu_addr);
329
330 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
331
332 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
333 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
334 }
335
336 release_firmware(adev->vcn.fw);
337 mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
338 mutex_destroy(&adev->vcn.vcn_pg_lock);
339
340 return 0;
341}
342
343/* from vcn4 and above, only unified queue is used */
344static bool amdgpu_vcn_using_unified_queue(struct amdgpu_ring *ring)
345{
346 struct amdgpu_device *adev = ring->adev;
347 bool ret = false;
348
349 if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0))
350 ret = true;
351
352 return ret;
353}
354
355bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
356{
357 bool ret = false;
358 int vcn_config = adev->vcn.vcn_config[vcn_instance];
359
360 if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK)) {
361 ret = true;
362 } else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK)) {
363 ret = true;
364 } else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK)) {
365 ret = true;
366 }
367
368 return ret;
369}
370
371int amdgpu_vcn_suspend(struct amdgpu_device *adev)
372{
373 unsigned size;
374 void *ptr;
375 int i, idx;
376
377 cancel_delayed_work_sync(&adev->vcn.idle_work);
378
379 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
380 if (adev->vcn.harvest_config & (1 << i))
381 continue;
382 if (adev->vcn.inst[i].vcpu_bo == NULL)
383 return 0;
384
385 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
386 ptr = adev->vcn.inst[i].cpu_addr;
387
388 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
389 if (!adev->vcn.inst[i].saved_bo)
390 return -ENOMEM;
391
392 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
393 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
394 drm_dev_exit(idx);
395 }
396 }
397 return 0;
398}
399
400int amdgpu_vcn_resume(struct amdgpu_device *adev)
401{
402 unsigned size;
403 void *ptr;
404 int i, idx;
405
406 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
407 if (adev->vcn.harvest_config & (1 << i))
408 continue;
409 if (adev->vcn.inst[i].vcpu_bo == NULL)
410 return -EINVAL;
411
412 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
413 ptr = adev->vcn.inst[i].cpu_addr;
414
415 if (adev->vcn.inst[i].saved_bo != NULL) {
416 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
417 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
418 drm_dev_exit(idx);
419 }
420 kvfree(adev->vcn.inst[i].saved_bo);
421 adev->vcn.inst[i].saved_bo = NULL;
422 } else {
423 const struct common_firmware_header *hdr;
424 unsigned offset;
425
426 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
427 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
428 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
429 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
430 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
431 le32_to_cpu(hdr->ucode_size_bytes));
432 drm_dev_exit(idx);
433 }
434 size -= le32_to_cpu(hdr->ucode_size_bytes);
435 ptr += le32_to_cpu(hdr->ucode_size_bytes);
436 }
437 memset_io(ptr, 0, size);
438 }
439 }
440 return 0;
441}
442
443static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
444{
445 struct amdgpu_device *adev =
446 container_of(work, struct amdgpu_device, vcn.idle_work.work);
447 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
448 unsigned int i, j;
449 int r = 0;
450
451 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
452 if (adev->vcn.harvest_config & (1 << j))
453 continue;
454
455 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
456 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
457 }
458
459 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
460 struct dpg_pause_state new_state;
461
462 if (fence[j] ||
463 unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
464 new_state.fw_based = VCN_DPG_STATE__PAUSE;
465 else
466 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
467
468 adev->vcn.pause_dpg_mode(adev, j, &new_state);
469 }
470
471 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
472 fences += fence[j];
473 }
474
475 if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
476 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
477 AMD_PG_STATE_GATE);
478 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
479 false);
480 if (r)
481 dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
482 } else {
483 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
484 }
485}
486
487void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
488{
489 struct amdgpu_device *adev = ring->adev;
490 int r = 0;
491
492 atomic_inc(&adev->vcn.total_submission_cnt);
493
494 if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
495 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
496 true);
497 if (r)
498 dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
499 }
500
501 mutex_lock(&adev->vcn.vcn_pg_lock);
502 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
503 AMD_PG_STATE_UNGATE);
504
505 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
506 struct dpg_pause_state new_state;
507
508 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
509 atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
510 new_state.fw_based = VCN_DPG_STATE__PAUSE;
511 } else {
512 unsigned int fences = 0;
513 unsigned int i;
514
515 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
516 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
517
518 if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
519 new_state.fw_based = VCN_DPG_STATE__PAUSE;
520 else
521 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
522 }
523
524 adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
525 }
526 mutex_unlock(&adev->vcn.vcn_pg_lock);
527}
528
529void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
530{
531 if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
532 ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
533 atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
534
535 atomic_dec(&ring->adev->vcn.total_submission_cnt);
536
537 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
538}
539
540int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
541{
542 struct amdgpu_device *adev = ring->adev;
543 uint32_t tmp = 0;
544 unsigned i;
545 int r;
546
547 /* VCN in SRIOV does not support direct register read/write */
548 if (amdgpu_sriov_vf(adev))
549 return 0;
550
551 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
552 r = amdgpu_ring_alloc(ring, 3);
553 if (r)
554 return r;
555 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
556 amdgpu_ring_write(ring, 0xDEADBEEF);
557 amdgpu_ring_commit(ring);
558 for (i = 0; i < adev->usec_timeout; i++) {
559 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
560 if (tmp == 0xDEADBEEF)
561 break;
562 udelay(1);
563 }
564
565 if (i >= adev->usec_timeout)
566 r = -ETIMEDOUT;
567
568 return r;
569}
570
571int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
572{
573 struct amdgpu_device *adev = ring->adev;
574 uint32_t rptr;
575 unsigned int i;
576 int r;
577
578 if (amdgpu_sriov_vf(adev))
579 return 0;
580
581 r = amdgpu_ring_alloc(ring, 16);
582 if (r)
583 return r;
584
585 rptr = amdgpu_ring_get_rptr(ring);
586
587 amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
588 amdgpu_ring_commit(ring);
589
590 for (i = 0; i < adev->usec_timeout; i++) {
591 if (amdgpu_ring_get_rptr(ring) != rptr)
592 break;
593 udelay(1);
594 }
595
596 if (i >= adev->usec_timeout)
597 r = -ETIMEDOUT;
598
599 return r;
600}
601
602static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
603 struct amdgpu_ib *ib_msg,
604 struct dma_fence **fence)
605{
606 u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
607 struct amdgpu_device *adev = ring->adev;
608 struct dma_fence *f = NULL;
609 struct amdgpu_job *job;
610 struct amdgpu_ib *ib;
611 int i, r;
612
613 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
614 64, AMDGPU_IB_POOL_DIRECT,
615 &job);
616 if (r)
617 goto err;
618
619 ib = &job->ibs[0];
620 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
621 ib->ptr[1] = addr;
622 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
623 ib->ptr[3] = addr >> 32;
624 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
625 ib->ptr[5] = 0;
626 for (i = 6; i < 16; i += 2) {
627 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
628 ib->ptr[i+1] = 0;
629 }
630 ib->length_dw = 16;
631
632 r = amdgpu_job_submit_direct(job, ring, &f);
633 if (r)
634 goto err_free;
635
636 amdgpu_ib_free(adev, ib_msg, f);
637
638 if (fence)
639 *fence = dma_fence_get(f);
640 dma_fence_put(f);
641
642 return 0;
643
644err_free:
645 amdgpu_job_free(job);
646err:
647 amdgpu_ib_free(adev, ib_msg, f);
648 return r;
649}
650
651static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
652 struct amdgpu_ib *ib)
653{
654 struct amdgpu_device *adev = ring->adev;
655 uint32_t *msg;
656 int r, i;
657
658 memset(ib, 0, sizeof(*ib));
659 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
660 AMDGPU_IB_POOL_DIRECT,
661 ib);
662 if (r)
663 return r;
664
665 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
666 msg[0] = cpu_to_le32(0x00000028);
667 msg[1] = cpu_to_le32(0x00000038);
668 msg[2] = cpu_to_le32(0x00000001);
669 msg[3] = cpu_to_le32(0x00000000);
670 msg[4] = cpu_to_le32(handle);
671 msg[5] = cpu_to_le32(0x00000000);
672 msg[6] = cpu_to_le32(0x00000001);
673 msg[7] = cpu_to_le32(0x00000028);
674 msg[8] = cpu_to_le32(0x00000010);
675 msg[9] = cpu_to_le32(0x00000000);
676 msg[10] = cpu_to_le32(0x00000007);
677 msg[11] = cpu_to_le32(0x00000000);
678 msg[12] = cpu_to_le32(0x00000780);
679 msg[13] = cpu_to_le32(0x00000440);
680 for (i = 14; i < 1024; ++i)
681 msg[i] = cpu_to_le32(0x0);
682
683 return 0;
684}
685
686static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
687 struct amdgpu_ib *ib)
688{
689 struct amdgpu_device *adev = ring->adev;
690 uint32_t *msg;
691 int r, i;
692
693 memset(ib, 0, sizeof(*ib));
694 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
695 AMDGPU_IB_POOL_DIRECT,
696 ib);
697 if (r)
698 return r;
699
700 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
701 msg[0] = cpu_to_le32(0x00000028);
702 msg[1] = cpu_to_le32(0x00000018);
703 msg[2] = cpu_to_le32(0x00000000);
704 msg[3] = cpu_to_le32(0x00000002);
705 msg[4] = cpu_to_le32(handle);
706 msg[5] = cpu_to_le32(0x00000000);
707 for (i = 6; i < 1024; ++i)
708 msg[i] = cpu_to_le32(0x0);
709
710 return 0;
711}
712
713int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
714{
715 struct dma_fence *fence = NULL;
716 struct amdgpu_ib ib;
717 long r;
718
719 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
720 if (r)
721 goto error;
722
723 r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
724 if (r)
725 goto error;
726 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
727 if (r)
728 goto error;
729
730 r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
731 if (r)
732 goto error;
733
734 r = dma_fence_wait_timeout(fence, false, timeout);
735 if (r == 0)
736 r = -ETIMEDOUT;
737 else if (r > 0)
738 r = 0;
739
740 dma_fence_put(fence);
741error:
742 return r;
743}
744
745static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib,
746 uint32_t ib_pack_in_dw, bool enc)
747{
748 uint32_t *ib_checksum;
749
750 ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */
751 ib->ptr[ib->length_dw++] = 0x30000002;
752 ib_checksum = &ib->ptr[ib->length_dw++];
753 ib->ptr[ib->length_dw++] = ib_pack_in_dw;
754
755 ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */
756 ib->ptr[ib->length_dw++] = 0x30000001;
757 ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3;
758 ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t);
759
760 return ib_checksum;
761}
762
763static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
764 uint32_t ib_pack_in_dw)
765{
766 uint32_t i;
767 uint32_t checksum = 0;
768
769 for (i = 0; i < ib_pack_in_dw; i++)
770 checksum += *(*ib_checksum + 2 + i);
771
772 **ib_checksum = checksum;
773}
774
775static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
776 struct amdgpu_ib *ib_msg,
777 struct dma_fence **fence)
778{
779 struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
780 unsigned int ib_size_dw = 64;
781 struct amdgpu_device *adev = ring->adev;
782 struct dma_fence *f = NULL;
783 struct amdgpu_job *job;
784 struct amdgpu_ib *ib;
785 uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
786 bool sq = amdgpu_vcn_using_unified_queue(ring);
787 uint32_t *ib_checksum;
788 uint32_t ib_pack_in_dw;
789 int i, r;
790
791 if (sq)
792 ib_size_dw += 8;
793
794 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
795 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
796 &job);
797 if (r)
798 goto err;
799
800 ib = &job->ibs[0];
801 ib->length_dw = 0;
802
803 /* single queue headers */
804 if (sq) {
805 ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t)
806 + 4 + 2; /* engine info + decoding ib in dw */
807 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false);
808 }
809
810 ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
811 ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
812 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
813 ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
814 memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
815
816 decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
817 decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
818 decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
819
820 for (i = ib->length_dw; i < ib_size_dw; ++i)
821 ib->ptr[i] = 0x0;
822
823 if (sq)
824 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw);
825
826 r = amdgpu_job_submit_direct(job, ring, &f);
827 if (r)
828 goto err_free;
829
830 amdgpu_ib_free(adev, ib_msg, f);
831
832 if (fence)
833 *fence = dma_fence_get(f);
834 dma_fence_put(f);
835
836 return 0;
837
838err_free:
839 amdgpu_job_free(job);
840err:
841 amdgpu_ib_free(adev, ib_msg, f);
842 return r;
843}
844
845int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
846{
847 struct dma_fence *fence = NULL;
848 struct amdgpu_ib ib;
849 long r;
850
851 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
852 if (r)
853 goto error;
854
855 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
856 if (r)
857 goto error;
858 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
859 if (r)
860 goto error;
861
862 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
863 if (r)
864 goto error;
865
866 r = dma_fence_wait_timeout(fence, false, timeout);
867 if (r == 0)
868 r = -ETIMEDOUT;
869 else if (r > 0)
870 r = 0;
871
872 dma_fence_put(fence);
873error:
874 return r;
875}
876
877int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
878{
879 struct amdgpu_device *adev = ring->adev;
880 uint32_t rptr;
881 unsigned i;
882 int r;
883
884 if (amdgpu_sriov_vf(adev))
885 return 0;
886
887 r = amdgpu_ring_alloc(ring, 16);
888 if (r)
889 return r;
890
891 rptr = amdgpu_ring_get_rptr(ring);
892
893 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
894 amdgpu_ring_commit(ring);
895
896 for (i = 0; i < adev->usec_timeout; i++) {
897 if (amdgpu_ring_get_rptr(ring) != rptr)
898 break;
899 udelay(1);
900 }
901
902 if (i >= adev->usec_timeout)
903 r = -ETIMEDOUT;
904
905 return r;
906}
907
908static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
909 struct amdgpu_ib *ib_msg,
910 struct dma_fence **fence)
911{
912 unsigned int ib_size_dw = 16;
913 struct amdgpu_job *job;
914 struct amdgpu_ib *ib;
915 struct dma_fence *f = NULL;
916 uint32_t *ib_checksum = NULL;
917 uint64_t addr;
918 bool sq = amdgpu_vcn_using_unified_queue(ring);
919 int i, r;
920
921 if (sq)
922 ib_size_dw += 8;
923
924 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
925 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
926 &job);
927 if (r)
928 return r;
929
930 ib = &job->ibs[0];
931 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
932
933 ib->length_dw = 0;
934
935 if (sq)
936 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
937
938 ib->ptr[ib->length_dw++] = 0x00000018;
939 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
940 ib->ptr[ib->length_dw++] = handle;
941 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
942 ib->ptr[ib->length_dw++] = addr;
943 ib->ptr[ib->length_dw++] = 0x0000000b;
944
945 ib->ptr[ib->length_dw++] = 0x00000014;
946 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
947 ib->ptr[ib->length_dw++] = 0x0000001c;
948 ib->ptr[ib->length_dw++] = 0x00000000;
949 ib->ptr[ib->length_dw++] = 0x00000000;
950
951 ib->ptr[ib->length_dw++] = 0x00000008;
952 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
953
954 for (i = ib->length_dw; i < ib_size_dw; ++i)
955 ib->ptr[i] = 0x0;
956
957 if (sq)
958 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
959
960 r = amdgpu_job_submit_direct(job, ring, &f);
961 if (r)
962 goto err;
963
964 if (fence)
965 *fence = dma_fence_get(f);
966 dma_fence_put(f);
967
968 return 0;
969
970err:
971 amdgpu_job_free(job);
972 return r;
973}
974
975static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
976 struct amdgpu_ib *ib_msg,
977 struct dma_fence **fence)
978{
979 unsigned int ib_size_dw = 16;
980 struct amdgpu_job *job;
981 struct amdgpu_ib *ib;
982 struct dma_fence *f = NULL;
983 uint32_t *ib_checksum = NULL;
984 uint64_t addr;
985 bool sq = amdgpu_vcn_using_unified_queue(ring);
986 int i, r;
987
988 if (sq)
989 ib_size_dw += 8;
990
991 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
992 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
993 &job);
994 if (r)
995 return r;
996
997 ib = &job->ibs[0];
998 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
999
1000 ib->length_dw = 0;
1001
1002 if (sq)
1003 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
1004
1005 ib->ptr[ib->length_dw++] = 0x00000018;
1006 ib->ptr[ib->length_dw++] = 0x00000001;
1007 ib->ptr[ib->length_dw++] = handle;
1008 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
1009 ib->ptr[ib->length_dw++] = addr;
1010 ib->ptr[ib->length_dw++] = 0x0000000b;
1011
1012 ib->ptr[ib->length_dw++] = 0x00000014;
1013 ib->ptr[ib->length_dw++] = 0x00000002;
1014 ib->ptr[ib->length_dw++] = 0x0000001c;
1015 ib->ptr[ib->length_dw++] = 0x00000000;
1016 ib->ptr[ib->length_dw++] = 0x00000000;
1017
1018 ib->ptr[ib->length_dw++] = 0x00000008;
1019 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
1020
1021 for (i = ib->length_dw; i < ib_size_dw; ++i)
1022 ib->ptr[i] = 0x0;
1023
1024 if (sq)
1025 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
1026
1027 r = amdgpu_job_submit_direct(job, ring, &f);
1028 if (r)
1029 goto err;
1030
1031 if (fence)
1032 *fence = dma_fence_get(f);
1033 dma_fence_put(f);
1034
1035 return 0;
1036
1037err:
1038 amdgpu_job_free(job);
1039 return r;
1040}
1041
1042int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1043{
1044 struct amdgpu_device *adev = ring->adev;
1045 struct dma_fence *fence = NULL;
1046 struct amdgpu_ib ib;
1047 long r;
1048
1049 memset(&ib, 0, sizeof(ib));
1050 r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
1051 AMDGPU_IB_POOL_DIRECT,
1052 &ib);
1053 if (r)
1054 return r;
1055
1056 r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
1057 if (r)
1058 goto error;
1059
1060 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
1061 if (r)
1062 goto error;
1063
1064 r = dma_fence_wait_timeout(fence, false, timeout);
1065 if (r == 0)
1066 r = -ETIMEDOUT;
1067 else if (r > 0)
1068 r = 0;
1069
1070error:
1071 amdgpu_ib_free(adev, &ib, fence);
1072 dma_fence_put(fence);
1073
1074 return r;
1075}
1076
1077int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1078{
1079 long r;
1080
1081 r = amdgpu_vcn_enc_ring_test_ib(ring, timeout);
1082 if (r)
1083 goto error;
1084
1085 r = amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout);
1086
1087error:
1088 return r;
1089}
1090
1091enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
1092{
1093 switch(ring) {
1094 case 0:
1095 return AMDGPU_RING_PRIO_0;
1096 case 1:
1097 return AMDGPU_RING_PRIO_1;
1098 case 2:
1099 return AMDGPU_RING_PRIO_2;
1100 default:
1101 return AMDGPU_RING_PRIO_0;
1102 }
1103}
1104
1105void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev)
1106{
1107 int i;
1108 unsigned int idx;
1109
1110 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1111 const struct common_firmware_header *hdr;
1112 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
1113
1114 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1115 if (adev->vcn.harvest_config & (1 << i))
1116 continue;
1117 /* currently only support 2 FW instances */
1118 if (i >= 2) {
1119 dev_info(adev->dev, "More then 2 VCN FW instances!\n");
1120 break;
1121 }
1122 idx = AMDGPU_UCODE_ID_VCN + i;
1123 adev->firmware.ucode[idx].ucode_id = idx;
1124 adev->firmware.ucode[idx].fw = adev->vcn.fw;
1125 adev->firmware.fw_size +=
1126 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
1127 }
1128 dev_info(adev->dev, "Will use PSP to load VCN firmware\n");
1129 }
1130}
1131
1132/*
1133 * debugfs for mapping vcn firmware log buffer.
1134 */
1135#if defined(CONFIG_DEBUG_FS)
1136static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf,
1137 size_t size, loff_t *pos)
1138{
1139 struct amdgpu_vcn_inst *vcn;
1140 void *log_buf;
1141 volatile struct amdgpu_vcn_fwlog *plog;
1142 unsigned int read_pos, write_pos, available, i, read_bytes = 0;
1143 unsigned int read_num[2] = {0};
1144
1145 vcn = file_inode(f)->i_private;
1146 if (!vcn)
1147 return -ENODEV;
1148
1149 if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log)
1150 return -EFAULT;
1151
1152 log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1153
1154 plog = (volatile struct amdgpu_vcn_fwlog *)log_buf;
1155 read_pos = plog->rptr;
1156 write_pos = plog->wptr;
1157
1158 if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE)
1159 return -EFAULT;
1160
1161 if (!size || (read_pos == write_pos))
1162 return 0;
1163
1164 if (write_pos > read_pos) {
1165 available = write_pos - read_pos;
1166 read_num[0] = min(size, (size_t)available);
1167 } else {
1168 read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos;
1169 available = read_num[0] + write_pos - plog->header_size;
1170 if (size > available)
1171 read_num[1] = write_pos - plog->header_size;
1172 else if (size > read_num[0])
1173 read_num[1] = size - read_num[0];
1174 else
1175 read_num[0] = size;
1176 }
1177
1178 for (i = 0; i < 2; i++) {
1179 if (read_num[i]) {
1180 if (read_pos == AMDGPU_VCNFW_LOG_SIZE)
1181 read_pos = plog->header_size;
1182 if (read_num[i] == copy_to_user((buf + read_bytes),
1183 (log_buf + read_pos), read_num[i]))
1184 return -EFAULT;
1185
1186 read_bytes += read_num[i];
1187 read_pos += read_num[i];
1188 }
1189 }
1190
1191 plog->rptr = read_pos;
1192 *pos += read_bytes;
1193 return read_bytes;
1194}
1195
1196static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = {
1197 .owner = THIS_MODULE,
1198 .read = amdgpu_debugfs_vcn_fwlog_read,
1199 .llseek = default_llseek
1200};
1201#endif
1202
1203void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i,
1204 struct amdgpu_vcn_inst *vcn)
1205{
1206#if defined(CONFIG_DEBUG_FS)
1207 struct drm_minor *minor = adev_to_drm(adev)->primary;
1208 struct dentry *root = minor->debugfs_root;
1209 char name[32];
1210
1211 sprintf(name, "amdgpu_vcn_%d_fwlog", i);
1212 debugfs_create_file_size(name, S_IFREG | S_IRUGO, root, vcn,
1213 &amdgpu_debugfs_vcnfwlog_fops,
1214 AMDGPU_VCNFW_LOG_SIZE);
1215#endif
1216}
1217
1218void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn)
1219{
1220#if defined(CONFIG_DEBUG_FS)
1221 volatile uint32_t *flag = vcn->fw_shared.cpu_addr;
1222 void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1223 uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size;
1224 volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr;
1225 volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr
1226 + vcn->fw_shared.log_offset;
1227 *flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG);
1228 fw_log->is_enabled = 1;
1229 fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF);
1230 fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32);
1231 fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE);
1232
1233 log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog);
1234 log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE;
1235 log_buf->rptr = log_buf->header_size;
1236 log_buf->wptr = log_buf->header_size;
1237 log_buf->wrapped = 0;
1238#endif
1239}
1240
1241int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev,
1242 struct amdgpu_irq_src *source,
1243 struct amdgpu_iv_entry *entry)
1244{
1245 struct ras_common_if *ras_if = adev->vcn.ras_if;
1246 struct ras_dispatch_if ih_data = {
1247 .entry = entry,
1248 };
1249
1250 if (!ras_if)
1251 return 0;
1252
1253 ih_data.head = *ras_if;
1254 amdgpu_ras_interrupt_dispatch(adev, &ih_data);
1255
1256 return 0;
1257}
1258
1259void amdgpu_vcn_set_ras_funcs(struct amdgpu_device *adev)
1260{
1261 if (!adev->vcn.ras)
1262 return;
1263
1264 amdgpu_ras_register_ras_block(adev, &adev->vcn.ras->ras_block);
1265
1266 strcpy(adev->vcn.ras->ras_block.ras_comm.name, "vcn");
1267 adev->vcn.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
1268 adev->vcn.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
1269 adev->vcn.ras_if = &adev->vcn.ras->ras_block.ras_comm;
1270
1271 /* If don't define special ras_late_init function, use default ras_late_init */
1272 if (!adev->vcn.ras->ras_block.ras_late_init)
1273 adev->vcn.ras->ras_block.ras_late_init = amdgpu_ras_block_late_init;
1274}
1/*
2 * Copyright 2016 Advanced Micro Devices, Inc.
3 * All Rights Reserved.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
20 *
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
23 * of the Software.
24 *
25 */
26
27#include <linux/firmware.h>
28#include <linux/module.h>
29#include <linux/pci.h>
30
31#include <drm/drm.h>
32
33#include "amdgpu.h"
34#include "amdgpu_pm.h"
35#include "amdgpu_vcn.h"
36#include "soc15d.h"
37#include "soc15_common.h"
38
39#include "vcn/vcn_1_0_offset.h"
40#include "vcn/vcn_1_0_sh_mask.h"
41
42/* 1 second timeout */
43#define VCN_IDLE_TIMEOUT msecs_to_jiffies(1000)
44
45/* Firmware Names */
46#define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
47#define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
48#define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
49#define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
50#define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
51#define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
52#define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
53#define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
54
55MODULE_FIRMWARE(FIRMWARE_RAVEN);
56MODULE_FIRMWARE(FIRMWARE_PICASSO);
57MODULE_FIRMWARE(FIRMWARE_RAVEN2);
58MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
59MODULE_FIRMWARE(FIRMWARE_RENOIR);
60MODULE_FIRMWARE(FIRMWARE_NAVI10);
61MODULE_FIRMWARE(FIRMWARE_NAVI14);
62MODULE_FIRMWARE(FIRMWARE_NAVI12);
63
64static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
65
66int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
67{
68 unsigned long bo_size;
69 const char *fw_name;
70 const struct common_firmware_header *hdr;
71 unsigned char fw_check;
72 int i, r;
73
74 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
75
76 switch (adev->asic_type) {
77 case CHIP_RAVEN:
78 if (adev->rev_id >= 8)
79 fw_name = FIRMWARE_RAVEN2;
80 else if (adev->pdev->device == 0x15d8)
81 fw_name = FIRMWARE_PICASSO;
82 else
83 fw_name = FIRMWARE_RAVEN;
84 break;
85 case CHIP_ARCTURUS:
86 fw_name = FIRMWARE_ARCTURUS;
87 break;
88 case CHIP_RENOIR:
89 fw_name = FIRMWARE_RENOIR;
90 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
91 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
92 adev->vcn.indirect_sram = true;
93 break;
94 case CHIP_NAVI10:
95 fw_name = FIRMWARE_NAVI10;
96 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
97 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
98 adev->vcn.indirect_sram = true;
99 break;
100 case CHIP_NAVI14:
101 fw_name = FIRMWARE_NAVI14;
102 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
103 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
104 adev->vcn.indirect_sram = true;
105 break;
106 case CHIP_NAVI12:
107 fw_name = FIRMWARE_NAVI12;
108 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
109 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
110 adev->vcn.indirect_sram = true;
111 break;
112 default:
113 return -EINVAL;
114 }
115
116 r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
117 if (r) {
118 dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
119 fw_name);
120 return r;
121 }
122
123 r = amdgpu_ucode_validate(adev->vcn.fw);
124 if (r) {
125 dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
126 fw_name);
127 release_firmware(adev->vcn.fw);
128 adev->vcn.fw = NULL;
129 return r;
130 }
131
132 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
133 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
134
135 /* Bit 20-23, it is encode major and non-zero for new naming convention.
136 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
137 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
138 * is zero in old naming convention, this field is always zero so far.
139 * These four bits are used to tell which naming convention is present.
140 */
141 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
142 if (fw_check) {
143 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
144
145 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
146 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
147 enc_major = fw_check;
148 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
149 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
150 DRM_INFO("Found VCN firmware Version ENC: %hu.%hu DEC: %hu VEP: %hu Revision: %hu\n",
151 enc_major, enc_minor, dec_ver, vep, fw_rev);
152 } else {
153 unsigned int version_major, version_minor, family_id;
154
155 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
156 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
157 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
158 DRM_INFO("Found VCN firmware Version: %hu.%hu Family ID: %hu\n",
159 version_major, version_minor, family_id);
160 }
161
162 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
163 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
164 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
165
166 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
167 if (adev->vcn.harvest_config & (1 << i))
168 continue;
169
170 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
171 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
172 &adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
173 if (r) {
174 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
175 return r;
176 }
177 }
178
179 if (adev->vcn.indirect_sram) {
180 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
181 AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.dpg_sram_bo,
182 &adev->vcn.dpg_sram_gpu_addr, &adev->vcn.dpg_sram_cpu_addr);
183 if (r) {
184 dev_err(adev->dev, "(%d) failed to allocate DPG bo\n", r);
185 return r;
186 }
187 }
188
189 return 0;
190}
191
192int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
193{
194 int i, j;
195
196 if (adev->vcn.indirect_sram) {
197 amdgpu_bo_free_kernel(&adev->vcn.dpg_sram_bo,
198 &adev->vcn.dpg_sram_gpu_addr,
199 (void **)&adev->vcn.dpg_sram_cpu_addr);
200 }
201
202 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
203 if (adev->vcn.harvest_config & (1 << j))
204 continue;
205 kvfree(adev->vcn.inst[j].saved_bo);
206
207 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
208 &adev->vcn.inst[j].gpu_addr,
209 (void **)&adev->vcn.inst[j].cpu_addr);
210
211 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
212
213 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
214 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
215
216 amdgpu_ring_fini(&adev->vcn.inst[j].ring_jpeg);
217 }
218
219 release_firmware(adev->vcn.fw);
220
221 return 0;
222}
223
224int amdgpu_vcn_suspend(struct amdgpu_device *adev)
225{
226 unsigned size;
227 void *ptr;
228 int i;
229
230 cancel_delayed_work_sync(&adev->vcn.idle_work);
231
232 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
233 if (adev->vcn.harvest_config & (1 << i))
234 continue;
235 if (adev->vcn.inst[i].vcpu_bo == NULL)
236 return 0;
237
238 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
239 ptr = adev->vcn.inst[i].cpu_addr;
240
241 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
242 if (!adev->vcn.inst[i].saved_bo)
243 return -ENOMEM;
244
245 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
246 }
247 return 0;
248}
249
250int amdgpu_vcn_resume(struct amdgpu_device *adev)
251{
252 unsigned size;
253 void *ptr;
254 int i;
255
256 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
257 if (adev->vcn.harvest_config & (1 << i))
258 continue;
259 if (adev->vcn.inst[i].vcpu_bo == NULL)
260 return -EINVAL;
261
262 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
263 ptr = adev->vcn.inst[i].cpu_addr;
264
265 if (adev->vcn.inst[i].saved_bo != NULL) {
266 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
267 kvfree(adev->vcn.inst[i].saved_bo);
268 adev->vcn.inst[i].saved_bo = NULL;
269 } else {
270 const struct common_firmware_header *hdr;
271 unsigned offset;
272
273 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
274 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
275 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
276 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
277 le32_to_cpu(hdr->ucode_size_bytes));
278 size -= le32_to_cpu(hdr->ucode_size_bytes);
279 ptr += le32_to_cpu(hdr->ucode_size_bytes);
280 }
281 memset_io(ptr, 0, size);
282 }
283 }
284 return 0;
285}
286
287static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
288{
289 struct amdgpu_device *adev =
290 container_of(work, struct amdgpu_device, vcn.idle_work.work);
291 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
292 unsigned int i, j;
293
294 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
295 if (adev->vcn.harvest_config & (1 << j))
296 continue;
297 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
298 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
299 }
300
301 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
302 struct dpg_pause_state new_state;
303
304 if (fence[j])
305 new_state.fw_based = VCN_DPG_STATE__PAUSE;
306 else
307 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
308
309 if (amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_jpeg))
310 new_state.jpeg = VCN_DPG_STATE__PAUSE;
311 else
312 new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
313
314 adev->vcn.pause_dpg_mode(adev, &new_state);
315 }
316
317 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_jpeg);
318 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
319 fences += fence[j];
320 }
321
322 if (fences == 0) {
323 amdgpu_gfx_off_ctrl(adev, true);
324 if (adev->asic_type < CHIP_ARCTURUS && adev->pm.dpm_enabled)
325 amdgpu_dpm_enable_uvd(adev, false);
326 else
327 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
328 AMD_PG_STATE_GATE);
329 } else {
330 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
331 }
332}
333
334void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
335{
336 struct amdgpu_device *adev = ring->adev;
337 bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work);
338
339 if (set_clocks) {
340 amdgpu_gfx_off_ctrl(adev, false);
341 if (adev->asic_type < CHIP_ARCTURUS && adev->pm.dpm_enabled)
342 amdgpu_dpm_enable_uvd(adev, true);
343 else
344 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
345 AMD_PG_STATE_UNGATE);
346 }
347
348 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
349 struct dpg_pause_state new_state;
350 unsigned int fences = 0;
351 unsigned int i;
352
353 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
354 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
355 }
356 if (fences)
357 new_state.fw_based = VCN_DPG_STATE__PAUSE;
358 else
359 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
360
361 if (amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_jpeg))
362 new_state.jpeg = VCN_DPG_STATE__PAUSE;
363 else
364 new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
365
366 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
367 new_state.fw_based = VCN_DPG_STATE__PAUSE;
368 else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG)
369 new_state.jpeg = VCN_DPG_STATE__PAUSE;
370
371 adev->vcn.pause_dpg_mode(adev, &new_state);
372 }
373}
374
375void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
376{
377 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
378}
379
380int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
381{
382 struct amdgpu_device *adev = ring->adev;
383 uint32_t tmp = 0;
384 unsigned i;
385 int r;
386
387 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
388 r = amdgpu_ring_alloc(ring, 3);
389 if (r)
390 return r;
391 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
392 amdgpu_ring_write(ring, 0xDEADBEEF);
393 amdgpu_ring_commit(ring);
394 for (i = 0; i < adev->usec_timeout; i++) {
395 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
396 if (tmp == 0xDEADBEEF)
397 break;
398 udelay(1);
399 }
400
401 if (i >= adev->usec_timeout)
402 r = -ETIMEDOUT;
403
404 return r;
405}
406
407static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
408 struct amdgpu_bo *bo,
409 struct dma_fence **fence)
410{
411 struct amdgpu_device *adev = ring->adev;
412 struct dma_fence *f = NULL;
413 struct amdgpu_job *job;
414 struct amdgpu_ib *ib;
415 uint64_t addr;
416 int i, r;
417
418 r = amdgpu_job_alloc_with_ib(adev, 64, &job);
419 if (r)
420 goto err;
421
422 ib = &job->ibs[0];
423 addr = amdgpu_bo_gpu_offset(bo);
424 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
425 ib->ptr[1] = addr;
426 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
427 ib->ptr[3] = addr >> 32;
428 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
429 ib->ptr[5] = 0;
430 for (i = 6; i < 16; i += 2) {
431 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
432 ib->ptr[i+1] = 0;
433 }
434 ib->length_dw = 16;
435
436 r = amdgpu_job_submit_direct(job, ring, &f);
437 if (r)
438 goto err_free;
439
440 amdgpu_bo_fence(bo, f, false);
441 amdgpu_bo_unreserve(bo);
442 amdgpu_bo_unref(&bo);
443
444 if (fence)
445 *fence = dma_fence_get(f);
446 dma_fence_put(f);
447
448 return 0;
449
450err_free:
451 amdgpu_job_free(job);
452
453err:
454 amdgpu_bo_unreserve(bo);
455 amdgpu_bo_unref(&bo);
456 return r;
457}
458
459static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
460 struct dma_fence **fence)
461{
462 struct amdgpu_device *adev = ring->adev;
463 struct amdgpu_bo *bo = NULL;
464 uint32_t *msg;
465 int r, i;
466
467 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
468 AMDGPU_GEM_DOMAIN_VRAM,
469 &bo, NULL, (void **)&msg);
470 if (r)
471 return r;
472
473 msg[0] = cpu_to_le32(0x00000028);
474 msg[1] = cpu_to_le32(0x00000038);
475 msg[2] = cpu_to_le32(0x00000001);
476 msg[3] = cpu_to_le32(0x00000000);
477 msg[4] = cpu_to_le32(handle);
478 msg[5] = cpu_to_le32(0x00000000);
479 msg[6] = cpu_to_le32(0x00000001);
480 msg[7] = cpu_to_le32(0x00000028);
481 msg[8] = cpu_to_le32(0x00000010);
482 msg[9] = cpu_to_le32(0x00000000);
483 msg[10] = cpu_to_le32(0x00000007);
484 msg[11] = cpu_to_le32(0x00000000);
485 msg[12] = cpu_to_le32(0x00000780);
486 msg[13] = cpu_to_le32(0x00000440);
487 for (i = 14; i < 1024; ++i)
488 msg[i] = cpu_to_le32(0x0);
489
490 return amdgpu_vcn_dec_send_msg(ring, bo, fence);
491}
492
493static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
494 struct dma_fence **fence)
495{
496 struct amdgpu_device *adev = ring->adev;
497 struct amdgpu_bo *bo = NULL;
498 uint32_t *msg;
499 int r, i;
500
501 r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
502 AMDGPU_GEM_DOMAIN_VRAM,
503 &bo, NULL, (void **)&msg);
504 if (r)
505 return r;
506
507 msg[0] = cpu_to_le32(0x00000028);
508 msg[1] = cpu_to_le32(0x00000018);
509 msg[2] = cpu_to_le32(0x00000000);
510 msg[3] = cpu_to_le32(0x00000002);
511 msg[4] = cpu_to_le32(handle);
512 msg[5] = cpu_to_le32(0x00000000);
513 for (i = 6; i < 1024; ++i)
514 msg[i] = cpu_to_le32(0x0);
515
516 return amdgpu_vcn_dec_send_msg(ring, bo, fence);
517}
518
519int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
520{
521 struct dma_fence *fence;
522 long r;
523
524 r = amdgpu_vcn_dec_get_create_msg(ring, 1, NULL);
525 if (r)
526 goto error;
527
528 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &fence);
529 if (r)
530 goto error;
531
532 r = dma_fence_wait_timeout(fence, false, timeout);
533 if (r == 0)
534 r = -ETIMEDOUT;
535 else if (r > 0)
536 r = 0;
537
538 dma_fence_put(fence);
539error:
540 return r;
541}
542
543int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
544{
545 struct amdgpu_device *adev = ring->adev;
546 uint32_t rptr;
547 unsigned i;
548 int r;
549
550 r = amdgpu_ring_alloc(ring, 16);
551 if (r)
552 return r;
553
554 rptr = amdgpu_ring_get_rptr(ring);
555
556 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
557 amdgpu_ring_commit(ring);
558
559 for (i = 0; i < adev->usec_timeout; i++) {
560 if (amdgpu_ring_get_rptr(ring) != rptr)
561 break;
562 udelay(1);
563 }
564
565 if (i >= adev->usec_timeout)
566 r = -ETIMEDOUT;
567
568 return r;
569}
570
571static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
572 struct amdgpu_bo *bo,
573 struct dma_fence **fence)
574{
575 const unsigned ib_size_dw = 16;
576 struct amdgpu_job *job;
577 struct amdgpu_ib *ib;
578 struct dma_fence *f = NULL;
579 uint64_t addr;
580 int i, r;
581
582 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
583 if (r)
584 return r;
585
586 ib = &job->ibs[0];
587 addr = amdgpu_bo_gpu_offset(bo);
588
589 ib->length_dw = 0;
590 ib->ptr[ib->length_dw++] = 0x00000018;
591 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
592 ib->ptr[ib->length_dw++] = handle;
593 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
594 ib->ptr[ib->length_dw++] = addr;
595 ib->ptr[ib->length_dw++] = 0x0000000b;
596
597 ib->ptr[ib->length_dw++] = 0x00000014;
598 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
599 ib->ptr[ib->length_dw++] = 0x0000001c;
600 ib->ptr[ib->length_dw++] = 0x00000000;
601 ib->ptr[ib->length_dw++] = 0x00000000;
602
603 ib->ptr[ib->length_dw++] = 0x00000008;
604 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
605
606 for (i = ib->length_dw; i < ib_size_dw; ++i)
607 ib->ptr[i] = 0x0;
608
609 r = amdgpu_job_submit_direct(job, ring, &f);
610 if (r)
611 goto err;
612
613 if (fence)
614 *fence = dma_fence_get(f);
615 dma_fence_put(f);
616
617 return 0;
618
619err:
620 amdgpu_job_free(job);
621 return r;
622}
623
624static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
625 struct amdgpu_bo *bo,
626 struct dma_fence **fence)
627{
628 const unsigned ib_size_dw = 16;
629 struct amdgpu_job *job;
630 struct amdgpu_ib *ib;
631 struct dma_fence *f = NULL;
632 uint64_t addr;
633 int i, r;
634
635 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
636 if (r)
637 return r;
638
639 ib = &job->ibs[0];
640 addr = amdgpu_bo_gpu_offset(bo);
641
642 ib->length_dw = 0;
643 ib->ptr[ib->length_dw++] = 0x00000018;
644 ib->ptr[ib->length_dw++] = 0x00000001;
645 ib->ptr[ib->length_dw++] = handle;
646 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
647 ib->ptr[ib->length_dw++] = addr;
648 ib->ptr[ib->length_dw++] = 0x0000000b;
649
650 ib->ptr[ib->length_dw++] = 0x00000014;
651 ib->ptr[ib->length_dw++] = 0x00000002;
652 ib->ptr[ib->length_dw++] = 0x0000001c;
653 ib->ptr[ib->length_dw++] = 0x00000000;
654 ib->ptr[ib->length_dw++] = 0x00000000;
655
656 ib->ptr[ib->length_dw++] = 0x00000008;
657 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
658
659 for (i = ib->length_dw; i < ib_size_dw; ++i)
660 ib->ptr[i] = 0x0;
661
662 r = amdgpu_job_submit_direct(job, ring, &f);
663 if (r)
664 goto err;
665
666 if (fence)
667 *fence = dma_fence_get(f);
668 dma_fence_put(f);
669
670 return 0;
671
672err:
673 amdgpu_job_free(job);
674 return r;
675}
676
677int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
678{
679 struct dma_fence *fence = NULL;
680 struct amdgpu_bo *bo = NULL;
681 long r;
682
683 r = amdgpu_bo_create_reserved(ring->adev, 128 * 1024, PAGE_SIZE,
684 AMDGPU_GEM_DOMAIN_VRAM,
685 &bo, NULL, NULL);
686 if (r)
687 return r;
688
689 r = amdgpu_vcn_enc_get_create_msg(ring, 1, bo, NULL);
690 if (r)
691 goto error;
692
693 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, bo, &fence);
694 if (r)
695 goto error;
696
697 r = dma_fence_wait_timeout(fence, false, timeout);
698 if (r == 0)
699 r = -ETIMEDOUT;
700 else if (r > 0)
701 r = 0;
702
703error:
704 dma_fence_put(fence);
705 amdgpu_bo_unreserve(bo);
706 amdgpu_bo_unref(&bo);
707 return r;
708}
709
710int amdgpu_vcn_jpeg_ring_test_ring(struct amdgpu_ring *ring)
711{
712 struct amdgpu_device *adev = ring->adev;
713 uint32_t tmp = 0;
714 unsigned i;
715 int r;
716
717 WREG32(adev->vcn.inst[ring->me].external.jpeg_pitch, 0xCAFEDEAD);
718 r = amdgpu_ring_alloc(ring, 3);
719 if (r)
720 return r;
721
722 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.jpeg_pitch, 0));
723 amdgpu_ring_write(ring, 0xDEADBEEF);
724 amdgpu_ring_commit(ring);
725
726 for (i = 0; i < adev->usec_timeout; i++) {
727 tmp = RREG32(adev->vcn.inst[ring->me].external.jpeg_pitch);
728 if (tmp == 0xDEADBEEF)
729 break;
730 udelay(1);
731 }
732
733 if (i >= adev->usec_timeout)
734 r = -ETIMEDOUT;
735
736 return r;
737}
738
739static int amdgpu_vcn_jpeg_set_reg(struct amdgpu_ring *ring, uint32_t handle,
740 struct dma_fence **fence)
741{
742 struct amdgpu_device *adev = ring->adev;
743 struct amdgpu_job *job;
744 struct amdgpu_ib *ib;
745 struct dma_fence *f = NULL;
746 const unsigned ib_size_dw = 16;
747 int i, r;
748
749 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
750 if (r)
751 return r;
752
753 ib = &job->ibs[0];
754
755 ib->ptr[0] = PACKETJ(adev->vcn.internal.jpeg_pitch, 0, 0, PACKETJ_TYPE0);
756 ib->ptr[1] = 0xDEADBEEF;
757 for (i = 2; i < 16; i += 2) {
758 ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6);
759 ib->ptr[i+1] = 0;
760 }
761 ib->length_dw = 16;
762
763 r = amdgpu_job_submit_direct(job, ring, &f);
764 if (r)
765 goto err;
766
767 if (fence)
768 *fence = dma_fence_get(f);
769 dma_fence_put(f);
770
771 return 0;
772
773err:
774 amdgpu_job_free(job);
775 return r;
776}
777
778int amdgpu_vcn_jpeg_ring_test_ib(struct amdgpu_ring *ring, long timeout)
779{
780 struct amdgpu_device *adev = ring->adev;
781 uint32_t tmp = 0;
782 unsigned i;
783 struct dma_fence *fence = NULL;
784 long r = 0;
785
786 r = amdgpu_vcn_jpeg_set_reg(ring, 1, &fence);
787 if (r)
788 goto error;
789
790 r = dma_fence_wait_timeout(fence, false, timeout);
791 if (r == 0) {
792 r = -ETIMEDOUT;
793 goto error;
794 } else if (r < 0) {
795 goto error;
796 } else {
797 r = 0;
798 }
799
800 for (i = 0; i < adev->usec_timeout; i++) {
801 tmp = RREG32(adev->vcn.inst[ring->me].external.jpeg_pitch);
802 if (tmp == 0xDEADBEEF)
803 break;
804 udelay(1);
805 }
806
807 if (i >= adev->usec_timeout)
808 r = -ETIMEDOUT;
809
810 dma_fence_put(fence);
811error:
812 return r;
813}