Commit | Line | Data |
---|---|---|
95d0906f LL |
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 <drm/drmP.h> | |
30 | #include <drm/drm.h> | |
31 | ||
32 | #include "amdgpu.h" | |
33 | #include "amdgpu_pm.h" | |
34 | #include "amdgpu_vcn.h" | |
35 | #include "soc15d.h" | |
36 | #include "soc15_common.h" | |
37 | ||
b1ebd7c0 | 38 | #include "vcn/vcn_1_0_offset.h" |
95d0906f LL |
39 | |
40 | /* 1 second timeout */ | |
41 | #define VCN_IDLE_TIMEOUT msecs_to_jiffies(1000) | |
42 | ||
43 | /* Firmware Names */ | |
44 | #define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin" | |
45 | ||
46 | MODULE_FIRMWARE(FIRMWARE_RAVEN); | |
47 | ||
48 | static void amdgpu_vcn_idle_work_handler(struct work_struct *work); | |
49 | ||
50 | int amdgpu_vcn_sw_init(struct amdgpu_device *adev) | |
51 | { | |
52 | struct amdgpu_ring *ring; | |
1b1f42d8 | 53 | struct drm_sched_rq *rq; |
95d0906f LL |
54 | unsigned long bo_size; |
55 | const char *fw_name; | |
56 | const struct common_firmware_header *hdr; | |
57 | unsigned version_major, version_minor, family_id; | |
58 | int r; | |
59 | ||
60 | INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler); | |
61 | ||
62 | switch (adev->asic_type) { | |
63 | case CHIP_RAVEN: | |
64 | fw_name = FIRMWARE_RAVEN; | |
65 | break; | |
66 | default: | |
67 | return -EINVAL; | |
68 | } | |
69 | ||
70 | r = request_firmware(&adev->vcn.fw, fw_name, adev->dev); | |
71 | if (r) { | |
72 | dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n", | |
73 | fw_name); | |
74 | return r; | |
75 | } | |
76 | ||
77 | r = amdgpu_ucode_validate(adev->vcn.fw); | |
78 | if (r) { | |
79 | dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n", | |
80 | fw_name); | |
81 | release_firmware(adev->vcn.fw); | |
82 | adev->vcn.fw = NULL; | |
83 | return r; | |
84 | } | |
85 | ||
86 | hdr = (const struct common_firmware_header *)adev->vcn.fw->data; | |
87 | family_id = le32_to_cpu(hdr->ucode_version) & 0xff; | |
88 | version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff; | |
89 | version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff; | |
90 | DRM_INFO("Found VCN firmware Version: %hu.%hu Family ID: %hu\n", | |
91 | version_major, version_minor, family_id); | |
92 | ||
93 | ||
94 | bo_size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8) | |
95 | + AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_HEAP_SIZE | |
96 | + AMDGPU_VCN_SESSION_SIZE * 40; | |
97 | r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE, | |
98 | AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.vcpu_bo, | |
99 | &adev->vcn.gpu_addr, &adev->vcn.cpu_addr); | |
100 | if (r) { | |
101 | dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r); | |
102 | return r; | |
103 | } | |
104 | ||
105 | ring = &adev->vcn.ring_dec; | |
1b1f42d8 LS |
106 | rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; |
107 | r = drm_sched_entity_init(&ring->sched, &adev->vcn.entity_dec, | |
b3eebe3d | 108 | rq, amdgpu_sched_jobs, NULL); |
95d0906f LL |
109 | if (r != 0) { |
110 | DRM_ERROR("Failed setting up VCN dec run queue.\n"); | |
111 | return r; | |
112 | } | |
113 | ||
2d531d81 | 114 | ring = &adev->vcn.ring_enc[0]; |
1b1f42d8 LS |
115 | rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; |
116 | r = drm_sched_entity_init(&ring->sched, &adev->vcn.entity_enc, | |
b3eebe3d | 117 | rq, amdgpu_sched_jobs, NULL); |
2d531d81 LL |
118 | if (r != 0) { |
119 | DRM_ERROR("Failed setting up VCN enc run queue.\n"); | |
120 | return r; | |
121 | } | |
122 | ||
95d0906f LL |
123 | return 0; |
124 | } | |
125 | ||
126 | int amdgpu_vcn_sw_fini(struct amdgpu_device *adev) | |
127 | { | |
101c6fee LL |
128 | int i; |
129 | ||
95d0906f LL |
130 | kfree(adev->vcn.saved_bo); |
131 | ||
1b1f42d8 | 132 | drm_sched_entity_fini(&adev->vcn.ring_dec.sched, &adev->vcn.entity_dec); |
95d0906f | 133 | |
1b1f42d8 | 134 | drm_sched_entity_fini(&adev->vcn.ring_enc[0].sched, &adev->vcn.entity_enc); |
2d531d81 | 135 | |
95d0906f LL |
136 | amdgpu_bo_free_kernel(&adev->vcn.vcpu_bo, |
137 | &adev->vcn.gpu_addr, | |
138 | (void **)&adev->vcn.cpu_addr); | |
139 | ||
140 | amdgpu_ring_fini(&adev->vcn.ring_dec); | |
141 | ||
101c6fee LL |
142 | for (i = 0; i < adev->vcn.num_enc_rings; ++i) |
143 | amdgpu_ring_fini(&adev->vcn.ring_enc[i]); | |
144 | ||
95d0906f LL |
145 | release_firmware(adev->vcn.fw); |
146 | ||
147 | return 0; | |
148 | } | |
149 | ||
150 | int amdgpu_vcn_suspend(struct amdgpu_device *adev) | |
151 | { | |
152 | unsigned size; | |
153 | void *ptr; | |
154 | ||
155 | if (adev->vcn.vcpu_bo == NULL) | |
156 | return 0; | |
157 | ||
158 | cancel_delayed_work_sync(&adev->vcn.idle_work); | |
159 | ||
160 | size = amdgpu_bo_size(adev->vcn.vcpu_bo); | |
161 | ptr = adev->vcn.cpu_addr; | |
162 | ||
163 | adev->vcn.saved_bo = kmalloc(size, GFP_KERNEL); | |
164 | if (!adev->vcn.saved_bo) | |
165 | return -ENOMEM; | |
166 | ||
167 | memcpy_fromio(adev->vcn.saved_bo, ptr, size); | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
172 | int amdgpu_vcn_resume(struct amdgpu_device *adev) | |
173 | { | |
174 | unsigned size; | |
175 | void *ptr; | |
176 | ||
177 | if (adev->vcn.vcpu_bo == NULL) | |
178 | return -EINVAL; | |
179 | ||
180 | size = amdgpu_bo_size(adev->vcn.vcpu_bo); | |
181 | ptr = adev->vcn.cpu_addr; | |
182 | ||
183 | if (adev->vcn.saved_bo != NULL) { | |
184 | memcpy_toio(ptr, adev->vcn.saved_bo, size); | |
185 | kfree(adev->vcn.saved_bo); | |
186 | adev->vcn.saved_bo = NULL; | |
187 | } else { | |
188 | const struct common_firmware_header *hdr; | |
189 | unsigned offset; | |
190 | ||
191 | hdr = (const struct common_firmware_header *)adev->vcn.fw->data; | |
192 | offset = le32_to_cpu(hdr->ucode_array_offset_bytes); | |
193 | memcpy_toio(adev->vcn.cpu_addr, adev->vcn.fw->data + offset, | |
194 | le32_to_cpu(hdr->ucode_size_bytes)); | |
195 | size -= le32_to_cpu(hdr->ucode_size_bytes); | |
196 | ptr += le32_to_cpu(hdr->ucode_size_bytes); | |
197 | memset_io(ptr, 0, size); | |
198 | } | |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
3e1086cf LL |
203 | static void amdgpu_vcn_idle_work_handler(struct work_struct *work) |
204 | { | |
205 | struct amdgpu_device *adev = | |
206 | container_of(work, struct amdgpu_device, vcn.idle_work.work); | |
207 | unsigned fences = amdgpu_fence_count_emitted(&adev->vcn.ring_dec); | |
208 | ||
209 | if (fences == 0) { | |
210 | if (adev->pm.dpm_enabled) { | |
aef060e1 | 211 | /* might be used when with pg/cg |
3e1086cf | 212 | amdgpu_dpm_enable_uvd(adev, false); |
aef060e1 | 213 | */ |
3e1086cf LL |
214 | } |
215 | } else { | |
216 | schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT); | |
217 | } | |
218 | } | |
219 | ||
220 | void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring) | |
221 | { | |
222 | struct amdgpu_device *adev = ring->adev; | |
223 | bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work); | |
224 | ||
aef060e1 LL |
225 | if (set_clocks && adev->pm.dpm_enabled) { |
226 | /* might be used when with pg/cg | |
227 | amdgpu_dpm_enable_uvd(adev, true); | |
228 | */ | |
3e1086cf LL |
229 | } |
230 | } | |
231 | ||
232 | void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring) | |
233 | { | |
234 | schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT); | |
235 | } | |
236 | ||
8c303c01 LL |
237 | int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring) |
238 | { | |
239 | struct amdgpu_device *adev = ring->adev; | |
240 | uint32_t tmp = 0; | |
241 | unsigned i; | |
242 | int r; | |
243 | ||
244 | WREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0xCAFEDEAD); | |
245 | r = amdgpu_ring_alloc(ring, 3); | |
246 | if (r) { | |
247 | DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", | |
248 | ring->idx, r); | |
249 | return r; | |
250 | } | |
251 | amdgpu_ring_write(ring, | |
252 | PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0)); | |
253 | amdgpu_ring_write(ring, 0xDEADBEEF); | |
254 | amdgpu_ring_commit(ring); | |
255 | for (i = 0; i < adev->usec_timeout; i++) { | |
256 | tmp = RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID)); | |
257 | if (tmp == 0xDEADBEEF) | |
258 | break; | |
259 | DRM_UDELAY(1); | |
260 | } | |
261 | ||
262 | if (i < adev->usec_timeout) { | |
9953b72f | 263 | DRM_DEBUG("ring test on %d succeeded in %d usecs\n", |
8c303c01 LL |
264 | ring->idx, i); |
265 | } else { | |
266 | DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", | |
267 | ring->idx, tmp); | |
268 | r = -EINVAL; | |
269 | } | |
270 | return r; | |
271 | } | |
272 | ||
95d0906f LL |
273 | static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo, |
274 | bool direct, struct dma_fence **fence) | |
275 | { | |
19be5570 | 276 | struct ttm_operation_ctx ctx = { true, false }; |
95d0906f LL |
277 | struct ttm_validate_buffer tv; |
278 | struct ww_acquire_ctx ticket; | |
279 | struct list_head head; | |
280 | struct amdgpu_job *job; | |
281 | struct amdgpu_ib *ib; | |
282 | struct dma_fence *f = NULL; | |
283 | struct amdgpu_device *adev = ring->adev; | |
284 | uint64_t addr; | |
285 | int i, r; | |
286 | ||
287 | memset(&tv, 0, sizeof(tv)); | |
288 | tv.bo = &bo->tbo; | |
289 | ||
290 | INIT_LIST_HEAD(&head); | |
291 | list_add(&tv.head, &head); | |
292 | ||
293 | r = ttm_eu_reserve_buffers(&ticket, &head, true, NULL); | |
294 | if (r) | |
295 | return r; | |
296 | ||
19be5570 | 297 | r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); |
95d0906f LL |
298 | if (r) |
299 | goto err; | |
300 | ||
301 | r = amdgpu_job_alloc_with_ib(adev, 64, &job); | |
302 | if (r) | |
303 | goto err; | |
304 | ||
305 | ib = &job->ibs[0]; | |
306 | addr = amdgpu_bo_gpu_offset(bo); | |
307 | ib->ptr[0] = PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0); | |
308 | ib->ptr[1] = addr; | |
309 | ib->ptr[2] = PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0); | |
310 | ib->ptr[3] = addr >> 32; | |
311 | ib->ptr[4] = PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0); | |
312 | ib->ptr[5] = 0; | |
313 | for (i = 6; i < 16; i += 2) { | |
314 | ib->ptr[i] = PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0); | |
315 | ib->ptr[i+1] = 0; | |
316 | } | |
317 | ib->length_dw = 16; | |
318 | ||
319 | if (direct) { | |
320 | r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f); | |
321 | job->fence = dma_fence_get(f); | |
322 | if (r) | |
323 | goto err_free; | |
324 | ||
325 | amdgpu_job_free(job); | |
326 | } else { | |
327 | r = amdgpu_job_submit(job, ring, &adev->vcn.entity_dec, | |
328 | AMDGPU_FENCE_OWNER_UNDEFINED, &f); | |
329 | if (r) | |
330 | goto err_free; | |
331 | } | |
332 | ||
333 | ttm_eu_fence_buffer_objects(&ticket, &head, f); | |
334 | ||
335 | if (fence) | |
336 | *fence = dma_fence_get(f); | |
337 | amdgpu_bo_unref(&bo); | |
338 | dma_fence_put(f); | |
339 | ||
340 | return 0; | |
341 | ||
342 | err_free: | |
343 | amdgpu_job_free(job); | |
344 | ||
345 | err: | |
346 | ttm_eu_backoff_reservation(&ticket, &head); | |
347 | return r; | |
348 | } | |
349 | ||
350 | static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, | |
351 | struct dma_fence **fence) | |
352 | { | |
353 | struct amdgpu_device *adev = ring->adev; | |
354 | struct amdgpu_bo *bo; | |
355 | uint32_t *msg; | |
356 | int r, i; | |
357 | ||
358 | r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true, | |
359 | AMDGPU_GEM_DOMAIN_VRAM, | |
360 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | | |
361 | AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, | |
8febe617 | 362 | NULL, NULL, &bo); |
95d0906f LL |
363 | if (r) |
364 | return r; | |
365 | ||
366 | r = amdgpu_bo_reserve(bo, false); | |
367 | if (r) { | |
368 | amdgpu_bo_unref(&bo); | |
369 | return r; | |
370 | } | |
371 | ||
372 | r = amdgpu_bo_kmap(bo, (void **)&msg); | |
373 | if (r) { | |
374 | amdgpu_bo_unreserve(bo); | |
375 | amdgpu_bo_unref(&bo); | |
376 | return r; | |
377 | } | |
378 | ||
2d8a425b | 379 | msg[0] = cpu_to_le32(0x00000028); |
3b8f5ab3 | 380 | msg[1] = cpu_to_le32(0x00000038); |
2d8a425b | 381 | msg[2] = cpu_to_le32(0x00000001); |
95d0906f | 382 | msg[3] = cpu_to_le32(0x00000000); |
2d8a425b | 383 | msg[4] = cpu_to_le32(handle); |
95d0906f | 384 | msg[5] = cpu_to_le32(0x00000000); |
2d8a425b LL |
385 | msg[6] = cpu_to_le32(0x00000001); |
386 | msg[7] = cpu_to_le32(0x00000028); | |
3b8f5ab3 | 387 | msg[8] = cpu_to_le32(0x00000010); |
95d0906f | 388 | msg[9] = cpu_to_le32(0x00000000); |
2d8a425b LL |
389 | msg[10] = cpu_to_le32(0x00000007); |
390 | msg[11] = cpu_to_le32(0x00000000); | |
3b8f5ab3 LL |
391 | msg[12] = cpu_to_le32(0x00000780); |
392 | msg[13] = cpu_to_le32(0x00000440); | |
393 | for (i = 14; i < 1024; ++i) | |
95d0906f LL |
394 | msg[i] = cpu_to_le32(0x0); |
395 | ||
396 | amdgpu_bo_kunmap(bo); | |
397 | amdgpu_bo_unreserve(bo); | |
398 | ||
399 | return amdgpu_vcn_dec_send_msg(ring, bo, true, fence); | |
400 | } | |
401 | ||
402 | static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, | |
403 | bool direct, struct dma_fence **fence) | |
404 | { | |
405 | struct amdgpu_device *adev = ring->adev; | |
406 | struct amdgpu_bo *bo; | |
407 | uint32_t *msg; | |
408 | int r, i; | |
409 | ||
410 | r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true, | |
411 | AMDGPU_GEM_DOMAIN_VRAM, | |
412 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | | |
413 | AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, | |
8febe617 | 414 | NULL, NULL, &bo); |
95d0906f LL |
415 | if (r) |
416 | return r; | |
417 | ||
418 | r = amdgpu_bo_reserve(bo, false); | |
419 | if (r) { | |
420 | amdgpu_bo_unref(&bo); | |
421 | return r; | |
422 | } | |
423 | ||
424 | r = amdgpu_bo_kmap(bo, (void **)&msg); | |
425 | if (r) { | |
426 | amdgpu_bo_unreserve(bo); | |
427 | amdgpu_bo_unref(&bo); | |
428 | return r; | |
429 | } | |
430 | ||
2d8a425b LL |
431 | msg[0] = cpu_to_le32(0x00000028); |
432 | msg[1] = cpu_to_le32(0x00000018); | |
433 | msg[2] = cpu_to_le32(0x00000000); | |
434 | msg[3] = cpu_to_le32(0x00000002); | |
435 | msg[4] = cpu_to_le32(handle); | |
436 | msg[5] = cpu_to_le32(0x00000000); | |
437 | for (i = 6; i < 1024; ++i) | |
95d0906f LL |
438 | msg[i] = cpu_to_le32(0x0); |
439 | ||
440 | amdgpu_bo_kunmap(bo); | |
441 | amdgpu_bo_unreserve(bo); | |
442 | ||
443 | return amdgpu_vcn_dec_send_msg(ring, bo, direct, fence); | |
444 | } | |
445 | ||
95d0906f LL |
446 | int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout) |
447 | { | |
448 | struct dma_fence *fence; | |
449 | long r; | |
450 | ||
451 | r = amdgpu_vcn_dec_get_create_msg(ring, 1, NULL); | |
452 | if (r) { | |
453 | DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r); | |
454 | goto error; | |
455 | } | |
456 | ||
457 | r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, true, &fence); | |
458 | if (r) { | |
459 | DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r); | |
460 | goto error; | |
461 | } | |
462 | ||
463 | r = dma_fence_wait_timeout(fence, false, timeout); | |
464 | if (r == 0) { | |
465 | DRM_ERROR("amdgpu: IB test timed out.\n"); | |
466 | r = -ETIMEDOUT; | |
467 | } else if (r < 0) { | |
468 | DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); | |
469 | } else { | |
9953b72f | 470 | DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); |
95d0906f LL |
471 | r = 0; |
472 | } | |
473 | ||
474 | dma_fence_put(fence); | |
475 | ||
476 | error: | |
477 | return r; | |
478 | } | |
2d531d81 | 479 | |
3e1086cf LL |
480 | int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring) |
481 | { | |
482 | struct amdgpu_device *adev = ring->adev; | |
483 | uint32_t rptr = amdgpu_ring_get_rptr(ring); | |
484 | unsigned i; | |
485 | int r; | |
486 | ||
487 | r = amdgpu_ring_alloc(ring, 16); | |
488 | if (r) { | |
489 | DRM_ERROR("amdgpu: vcn enc failed to lock ring %d (%d).\n", | |
490 | ring->idx, r); | |
491 | return r; | |
492 | } | |
c3bd3040 | 493 | amdgpu_ring_write(ring, VCN_ENC_CMD_END); |
3e1086cf LL |
494 | amdgpu_ring_commit(ring); |
495 | ||
496 | for (i = 0; i < adev->usec_timeout; i++) { | |
497 | if (amdgpu_ring_get_rptr(ring) != rptr) | |
498 | break; | |
499 | DRM_UDELAY(1); | |
500 | } | |
501 | ||
502 | if (i < adev->usec_timeout) { | |
9953b72f | 503 | DRM_DEBUG("ring test on %d succeeded in %d usecs\n", |
3e1086cf LL |
504 | ring->idx, i); |
505 | } else { | |
506 | DRM_ERROR("amdgpu: ring %d test failed\n", | |
507 | ring->idx); | |
508 | r = -ETIMEDOUT; | |
509 | } | |
510 | ||
511 | return r; | |
512 | } | |
513 | ||
2d531d81 LL |
514 | static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, |
515 | struct dma_fence **fence) | |
516 | { | |
25547cfd | 517 | const unsigned ib_size_dw = 16; |
2d531d81 LL |
518 | struct amdgpu_job *job; |
519 | struct amdgpu_ib *ib; | |
520 | struct dma_fence *f = NULL; | |
521 | uint64_t dummy; | |
522 | int i, r; | |
523 | ||
524 | r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job); | |
525 | if (r) | |
526 | return r; | |
527 | ||
528 | ib = &job->ibs[0]; | |
2d531d81 LL |
529 | dummy = ib->gpu_addr + 1024; |
530 | ||
2d531d81 | 531 | ib->length_dw = 0; |
25547cfd LL |
532 | ib->ptr[ib->length_dw++] = 0x00000018; |
533 | ib->ptr[ib->length_dw++] = 0x00000001; /* session info */ | |
2d531d81 | 534 | ib->ptr[ib->length_dw++] = handle; |
25547cfd LL |
535 | ib->ptr[ib->length_dw++] = upper_32_bits(dummy); |
536 | ib->ptr[ib->length_dw++] = dummy; | |
537 | ib->ptr[ib->length_dw++] = 0x0000000b; | |
2d531d81 | 538 | |
25547cfd LL |
539 | ib->ptr[ib->length_dw++] = 0x00000014; |
540 | ib->ptr[ib->length_dw++] = 0x00000002; /* task info */ | |
541 | ib->ptr[ib->length_dw++] = 0x0000001c; | |
2d531d81 LL |
542 | ib->ptr[ib->length_dw++] = 0x00000000; |
543 | ib->ptr[ib->length_dw++] = 0x00000000; | |
544 | ||
25547cfd LL |
545 | ib->ptr[ib->length_dw++] = 0x00000008; |
546 | ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */ | |
2d531d81 LL |
547 | |
548 | for (i = ib->length_dw; i < ib_size_dw; ++i) | |
549 | ib->ptr[i] = 0x0; | |
550 | ||
551 | r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f); | |
552 | job->fence = dma_fence_get(f); | |
553 | if (r) | |
554 | goto err; | |
555 | ||
556 | amdgpu_job_free(job); | |
557 | if (fence) | |
558 | *fence = dma_fence_get(f); | |
559 | dma_fence_put(f); | |
25547cfd | 560 | |
2d531d81 LL |
561 | return 0; |
562 | ||
563 | err: | |
564 | amdgpu_job_free(job); | |
565 | return r; | |
566 | } | |
567 | ||
568 | static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, | |
25547cfd | 569 | struct dma_fence **fence) |
2d531d81 | 570 | { |
25547cfd | 571 | const unsigned ib_size_dw = 16; |
2d531d81 LL |
572 | struct amdgpu_job *job; |
573 | struct amdgpu_ib *ib; | |
574 | struct dma_fence *f = NULL; | |
25547cfd | 575 | uint64_t dummy; |
2d531d81 LL |
576 | int i, r; |
577 | ||
578 | r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job); | |
579 | if (r) | |
580 | return r; | |
581 | ||
582 | ib = &job->ibs[0]; | |
25547cfd | 583 | dummy = ib->gpu_addr + 1024; |
2d531d81 | 584 | |
2d531d81 | 585 | ib->length_dw = 0; |
25547cfd LL |
586 | ib->ptr[ib->length_dw++] = 0x00000018; |
587 | ib->ptr[ib->length_dw++] = 0x00000001; | |
2d531d81 | 588 | ib->ptr[ib->length_dw++] = handle; |
25547cfd LL |
589 | ib->ptr[ib->length_dw++] = upper_32_bits(dummy); |
590 | ib->ptr[ib->length_dw++] = dummy; | |
591 | ib->ptr[ib->length_dw++] = 0x0000000b; | |
2d531d81 | 592 | |
25547cfd LL |
593 | ib->ptr[ib->length_dw++] = 0x00000014; |
594 | ib->ptr[ib->length_dw++] = 0x00000002; | |
595 | ib->ptr[ib->length_dw++] = 0x0000001c; | |
2d531d81 | 596 | ib->ptr[ib->length_dw++] = 0x00000000; |
2d531d81 LL |
597 | ib->ptr[ib->length_dw++] = 0x00000000; |
598 | ||
25547cfd LL |
599 | ib->ptr[ib->length_dw++] = 0x00000008; |
600 | ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */ | |
2d531d81 LL |
601 | |
602 | for (i = ib->length_dw; i < ib_size_dw; ++i) | |
603 | ib->ptr[i] = 0x0; | |
604 | ||
25547cfd LL |
605 | r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f); |
606 | job->fence = dma_fence_get(f); | |
607 | if (r) | |
608 | goto err; | |
2d531d81 | 609 | |
25547cfd | 610 | amdgpu_job_free(job); |
2d531d81 LL |
611 | if (fence) |
612 | *fence = dma_fence_get(f); | |
613 | dma_fence_put(f); | |
25547cfd | 614 | |
2d531d81 LL |
615 | return 0; |
616 | ||
617 | err: | |
618 | amdgpu_job_free(job); | |
619 | return r; | |
620 | } | |
621 | ||
2d531d81 LL |
622 | int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) |
623 | { | |
624 | struct dma_fence *fence = NULL; | |
625 | long r; | |
626 | ||
627 | r = amdgpu_vcn_enc_get_create_msg(ring, 1, NULL); | |
628 | if (r) { | |
629 | DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r); | |
630 | goto error; | |
631 | } | |
632 | ||
25547cfd | 633 | r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &fence); |
2d531d81 LL |
634 | if (r) { |
635 | DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r); | |
636 | goto error; | |
637 | } | |
638 | ||
639 | r = dma_fence_wait_timeout(fence, false, timeout); | |
640 | if (r == 0) { | |
641 | DRM_ERROR("amdgpu: IB test timed out.\n"); | |
642 | r = -ETIMEDOUT; | |
643 | } else if (r < 0) { | |
644 | DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); | |
645 | } else { | |
9953b72f | 646 | DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); |
2d531d81 LL |
647 | r = 0; |
648 | } | |
649 | error: | |
650 | dma_fence_put(fence); | |
651 | return r; | |
652 | } |