2 * Copyright 2013 Advanced Micro Devices, Inc.
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:
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.
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
25 * Authors: Christian König <christian.koenig@amd.com>
28 #include <linux/firmware.h>
29 #include <linux/module.h>
34 #include "amdgpu_pm.h"
35 #include "amdgpu_vce.h"
38 /* 1 second timeout */
39 #define VCE_IDLE_TIMEOUT msecs_to_jiffies(1000)
42 #ifdef CONFIG_DRM_AMDGPU_CIK
43 #define FIRMWARE_BONAIRE "/*(DEBLOBBED)*/"
44 #define FIRMWARE_KABINI "/*(DEBLOBBED)*/"
45 #define FIRMWARE_KAVERI "/*(DEBLOBBED)*/"
46 #define FIRMWARE_HAWAII "/*(DEBLOBBED)*/"
47 #define FIRMWARE_MULLINS "/*(DEBLOBBED)*/"
49 #define FIRMWARE_TONGA "/*(DEBLOBBED)*/"
50 #define FIRMWARE_CARRIZO "/*(DEBLOBBED)*/"
51 #define FIRMWARE_FIJI "/*(DEBLOBBED)*/"
52 #define FIRMWARE_STONEY "/*(DEBLOBBED)*/"
53 #define FIRMWARE_POLARIS10 "/*(DEBLOBBED)*/"
54 #define FIRMWARE_POLARIS11 "/*(DEBLOBBED)*/"
55 #define FIRMWARE_POLARIS12 "/*(DEBLOBBED)*/"
57 #define FIRMWARE_VEGA10 "/*(DEBLOBBED)*/"
59 #ifdef CONFIG_DRM_AMDGPU_CIK
64 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
67 * amdgpu_vce_init - allocate memory, load vce firmware
69 * @adev: amdgpu_device pointer
71 * First step to get VCE online, allocate memory and load the firmware
73 int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
75 struct amdgpu_ring *ring;
76 struct amd_sched_rq *rq;
78 const struct common_firmware_header *hdr;
79 unsigned ucode_version, version_major, version_minor, binary_id;
82 switch (adev->asic_type) {
83 #ifdef CONFIG_DRM_AMDGPU_CIK
85 fw_name = FIRMWARE_BONAIRE;
88 fw_name = FIRMWARE_KAVERI;
91 fw_name = FIRMWARE_KABINI;
94 fw_name = FIRMWARE_HAWAII;
97 fw_name = FIRMWARE_MULLINS;
101 fw_name = FIRMWARE_TONGA;
104 fw_name = FIRMWARE_CARRIZO;
107 fw_name = FIRMWARE_FIJI;
110 fw_name = FIRMWARE_STONEY;
113 fw_name = FIRMWARE_POLARIS10;
116 fw_name = FIRMWARE_POLARIS11;
119 fw_name = FIRMWARE_VEGA10;
122 fw_name = FIRMWARE_POLARIS12;
129 r = reject_firmware(&adev->vce.fw, fw_name, adev->dev);
131 dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n",
136 r = amdgpu_ucode_validate(adev->vce.fw);
138 dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n",
140 release_firmware(adev->vce.fw);
145 hdr = (const struct common_firmware_header *)adev->vce.fw->data;
147 ucode_version = le32_to_cpu(hdr->ucode_version);
148 version_major = (ucode_version >> 20) & 0xfff;
149 version_minor = (ucode_version >> 8) & 0xfff;
150 binary_id = ucode_version & 0xff;
151 DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n",
152 version_major, version_minor, binary_id);
153 adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) |
156 r = amdgpu_bo_create_kernel(adev, size, PAGE_SIZE,
157 AMDGPU_GEM_DOMAIN_VRAM, &adev->vce.vcpu_bo,
158 &adev->vce.gpu_addr, &adev->vce.cpu_addr);
160 dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
164 ring = &adev->vce.ring[0];
165 rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL];
166 r = amd_sched_entity_init(&ring->sched, &adev->vce.entity,
167 rq, amdgpu_sched_jobs);
169 DRM_ERROR("Failed setting up VCE run queue.\n");
173 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
174 atomic_set(&adev->vce.handles[i], 0);
175 adev->vce.filp[i] = NULL;
178 INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler);
179 mutex_init(&adev->vce.idle_mutex);
185 * amdgpu_vce_fini - free memory
187 * @adev: amdgpu_device pointer
189 * Last step on VCE teardown, free firmware memory
191 int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
195 if (adev->vce.vcpu_bo == NULL)
198 amd_sched_entity_fini(&adev->vce.ring[0].sched, &adev->vce.entity);
200 amdgpu_bo_free_kernel(&adev->vce.vcpu_bo, &adev->vce.gpu_addr,
201 (void **)&adev->vce.cpu_addr);
203 for (i = 0; i < adev->vce.num_rings; i++)
204 amdgpu_ring_fini(&adev->vce.ring[i]);
206 release_firmware(adev->vce.fw);
207 mutex_destroy(&adev->vce.idle_mutex);
213 * amdgpu_vce_suspend - unpin VCE fw memory
215 * @adev: amdgpu_device pointer
218 int amdgpu_vce_suspend(struct amdgpu_device *adev)
222 cancel_delayed_work_sync(&adev->vce.idle_work);
224 if (adev->vce.vcpu_bo == NULL)
227 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
228 if (atomic_read(&adev->vce.handles[i]))
231 if (i == AMDGPU_MAX_VCE_HANDLES)
234 /* TODO: suspending running encoding sessions isn't supported */
239 * amdgpu_vce_resume - pin VCE fw memory
241 * @adev: amdgpu_device pointer
244 int amdgpu_vce_resume(struct amdgpu_device *adev)
247 const struct common_firmware_header *hdr;
251 if (adev->vce.vcpu_bo == NULL)
254 r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
256 dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
260 r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr);
262 amdgpu_bo_unreserve(adev->vce.vcpu_bo);
263 dev_err(adev->dev, "(%d) VCE map failed\n", r);
267 hdr = (const struct common_firmware_header *)adev->vce.fw->data;
268 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
269 memcpy_toio(cpu_addr, adev->vce.fw->data + offset,
270 adev->vce.fw->size - offset);
272 amdgpu_bo_kunmap(adev->vce.vcpu_bo);
274 amdgpu_bo_unreserve(adev->vce.vcpu_bo);
280 * amdgpu_vce_idle_work_handler - power off VCE
282 * @work: pointer to work structure
284 * power of VCE when it's not used any more
286 static void amdgpu_vce_idle_work_handler(struct work_struct *work)
288 struct amdgpu_device *adev =
289 container_of(work, struct amdgpu_device, vce.idle_work.work);
290 unsigned i, count = 0;
292 if (amdgpu_sriov_vf(adev))
295 for (i = 0; i < adev->vce.num_rings; i++)
296 count += amdgpu_fence_count_emitted(&adev->vce.ring[i]);
299 if (adev->pm.dpm_enabled) {
300 amdgpu_dpm_enable_vce(adev, false);
302 amdgpu_asic_set_vce_clocks(adev, 0, 0);
303 amdgpu_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
305 amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
309 schedule_delayed_work(&adev->vce.idle_work, VCE_IDLE_TIMEOUT);
314 * amdgpu_vce_ring_begin_use - power up VCE
318 * Make sure VCE is powerd up when we want to use it
320 void amdgpu_vce_ring_begin_use(struct amdgpu_ring *ring)
322 struct amdgpu_device *adev = ring->adev;
325 if (amdgpu_sriov_vf(adev))
328 mutex_lock(&adev->vce.idle_mutex);
329 set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work);
331 if (adev->pm.dpm_enabled) {
332 amdgpu_dpm_enable_vce(adev, true);
334 amdgpu_asic_set_vce_clocks(adev, 53300, 40000);
335 amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
336 AMD_CG_STATE_UNGATE);
337 amdgpu_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
338 AMD_PG_STATE_UNGATE);
342 mutex_unlock(&adev->vce.idle_mutex);
346 * amdgpu_vce_ring_end_use - power VCE down
350 * Schedule work to power VCE down again
352 void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring)
354 schedule_delayed_work(&ring->adev->vce.idle_work, VCE_IDLE_TIMEOUT);
358 * amdgpu_vce_free_handles - free still open VCE handles
360 * @adev: amdgpu_device pointer
361 * @filp: drm file pointer
363 * Close all VCE handles still open by this file pointer
365 void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
367 struct amdgpu_ring *ring = &adev->vce.ring[0];
369 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
370 uint32_t handle = atomic_read(&adev->vce.handles[i]);
372 if (!handle || adev->vce.filp[i] != filp)
375 r = amdgpu_vce_get_destroy_msg(ring, handle, false, NULL);
377 DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
379 adev->vce.filp[i] = NULL;
380 atomic_set(&adev->vce.handles[i], 0);
385 * amdgpu_vce_get_create_msg - generate a VCE create msg
387 * @adev: amdgpu_device pointer
388 * @ring: ring we should submit the msg to
389 * @handle: VCE session handle to use
390 * @fence: optional fence to return
392 * Open up a stream for HW test
394 int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
395 struct dma_fence **fence)
397 const unsigned ib_size_dw = 1024;
398 struct amdgpu_job *job;
399 struct amdgpu_ib *ib;
400 struct dma_fence *f = NULL;
404 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
410 dummy = ib->gpu_addr + 1024;
412 /* stitch together an VCE create msg */
414 ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
415 ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
416 ib->ptr[ib->length_dw++] = handle;
418 if ((ring->adev->vce.fw_version >> 24) >= 52)
419 ib->ptr[ib->length_dw++] = 0x00000040; /* len */
421 ib->ptr[ib->length_dw++] = 0x00000030; /* len */
422 ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
423 ib->ptr[ib->length_dw++] = 0x00000000;
424 ib->ptr[ib->length_dw++] = 0x00000042;
425 ib->ptr[ib->length_dw++] = 0x0000000a;
426 ib->ptr[ib->length_dw++] = 0x00000001;
427 ib->ptr[ib->length_dw++] = 0x00000080;
428 ib->ptr[ib->length_dw++] = 0x00000060;
429 ib->ptr[ib->length_dw++] = 0x00000100;
430 ib->ptr[ib->length_dw++] = 0x00000100;
431 ib->ptr[ib->length_dw++] = 0x0000000c;
432 ib->ptr[ib->length_dw++] = 0x00000000;
433 if ((ring->adev->vce.fw_version >> 24) >= 52) {
434 ib->ptr[ib->length_dw++] = 0x00000000;
435 ib->ptr[ib->length_dw++] = 0x00000000;
436 ib->ptr[ib->length_dw++] = 0x00000000;
437 ib->ptr[ib->length_dw++] = 0x00000000;
440 ib->ptr[ib->length_dw++] = 0x00000014; /* len */
441 ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
442 ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
443 ib->ptr[ib->length_dw++] = dummy;
444 ib->ptr[ib->length_dw++] = 0x00000001;
446 for (i = ib->length_dw; i < ib_size_dw; ++i)
449 r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f);
450 job->fence = dma_fence_get(f);
454 amdgpu_job_free(job);
456 *fence = dma_fence_get(f);
461 amdgpu_job_free(job);
466 * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg
468 * @adev: amdgpu_device pointer
469 * @ring: ring we should submit the msg to
470 * @handle: VCE session handle to use
471 * @fence: optional fence to return
473 * Close up a stream for HW test or if userspace failed to do so
475 int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
476 bool direct, struct dma_fence **fence)
478 const unsigned ib_size_dw = 1024;
479 struct amdgpu_job *job;
480 struct amdgpu_ib *ib;
481 struct dma_fence *f = NULL;
484 r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
490 /* stitch together an VCE destroy msg */
492 ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
493 ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
494 ib->ptr[ib->length_dw++] = handle;
496 ib->ptr[ib->length_dw++] = 0x00000020; /* len */
497 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
498 ib->ptr[ib->length_dw++] = 0xffffffff; /* next task info, set to 0xffffffff if no */
499 ib->ptr[ib->length_dw++] = 0x00000001; /* destroy session */
500 ib->ptr[ib->length_dw++] = 0x00000000;
501 ib->ptr[ib->length_dw++] = 0x00000000;
502 ib->ptr[ib->length_dw++] = 0xffffffff; /* feedback is not needed, set to 0xffffffff and firmware will not output feedback */
503 ib->ptr[ib->length_dw++] = 0x00000000;
505 ib->ptr[ib->length_dw++] = 0x00000008; /* len */
506 ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */
508 for (i = ib->length_dw; i < ib_size_dw; ++i)
512 r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f);
513 job->fence = dma_fence_get(f);
517 amdgpu_job_free(job);
519 r = amdgpu_job_submit(job, ring, &ring->adev->vce.entity,
520 AMDGPU_FENCE_OWNER_UNDEFINED, &f);
526 *fence = dma_fence_get(f);
531 amdgpu_job_free(job);
536 * amdgpu_vce_cs_reloc - command submission relocation
539 * @lo: address of lower dword
540 * @hi: address of higher dword
541 * @size: minimum size
543 * Patch relocation inside command stream with real buffer address
545 static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx,
546 int lo, int hi, unsigned size, uint32_t index)
548 struct amdgpu_bo_va_mapping *mapping;
549 struct amdgpu_bo *bo;
552 if (index == 0xffffffff)
555 addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) |
556 ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32;
557 addr += ((uint64_t)size) * ((uint64_t)index);
559 mapping = amdgpu_cs_find_mapping(p, addr, &bo);
560 if (mapping == NULL) {
561 DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d %d %d\n",
562 addr, lo, hi, size, index);
566 if ((addr + (uint64_t)size) >
567 (mapping->last + 1) * AMDGPU_GPU_PAGE_SIZE) {
568 DRM_ERROR("BO to small for addr 0x%010Lx %d %d\n",
573 addr -= mapping->start * AMDGPU_GPU_PAGE_SIZE;
574 addr += amdgpu_bo_gpu_offset(bo);
575 addr -= ((uint64_t)size) * ((uint64_t)index);
577 amdgpu_set_ib_value(p, ib_idx, lo, lower_32_bits(addr));
578 amdgpu_set_ib_value(p, ib_idx, hi, upper_32_bits(addr));
584 * amdgpu_vce_validate_handle - validate stream handle
587 * @handle: handle to validate
588 * @allocated: allocated a new handle?
590 * Validates the handle and return the found session index or -EINVAL
591 * we we don't have another free session index.
593 static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p,
594 uint32_t handle, uint32_t *allocated)
598 /* validate the handle */
599 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
600 if (atomic_read(&p->adev->vce.handles[i]) == handle) {
601 if (p->adev->vce.filp[i] != p->filp) {
602 DRM_ERROR("VCE handle collision detected!\n");
609 /* handle not found try to alloc a new one */
610 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
611 if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) {
612 p->adev->vce.filp[i] = p->filp;
613 p->adev->vce.img_size[i] = 0;
614 *allocated |= 1 << i;
619 DRM_ERROR("No more free VCE handles!\n");
624 * amdgpu_vce_cs_parse - parse and validate the command stream
629 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
631 struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
632 unsigned fb_idx = 0, bs_idx = 0;
633 int session_idx = -1;
634 uint32_t destroyed = 0;
635 uint32_t created = 0;
636 uint32_t allocated = 0;
637 uint32_t tmp, handle = 0;
638 uint32_t *size = &tmp;
639 int i, r = 0, idx = 0;
642 ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
644 r = amdgpu_cs_sysvm_access_required(p);
648 while (idx < ib->length_dw) {
649 uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
650 uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
652 if ((len < 8) || (len & 3)) {
653 DRM_ERROR("invalid VCE command length (%d)!\n", len);
659 case 0x00000001: /* session */
660 handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
661 session_idx = amdgpu_vce_validate_handle(p, handle,
663 if (session_idx < 0) {
667 size = &p->adev->vce.img_size[session_idx];
670 case 0x00000002: /* task info */
671 fb_idx = amdgpu_get_ib_value(p, ib_idx, idx + 6);
672 bs_idx = amdgpu_get_ib_value(p, ib_idx, idx + 7);
675 case 0x01000001: /* create */
676 created |= 1 << session_idx;
677 if (destroyed & (1 << session_idx)) {
678 destroyed &= ~(1 << session_idx);
679 allocated |= 1 << session_idx;
681 } else if (!(allocated & (1 << session_idx))) {
682 DRM_ERROR("Handle already in use!\n");
687 *size = amdgpu_get_ib_value(p, ib_idx, idx + 8) *
688 amdgpu_get_ib_value(p, ib_idx, idx + 10) *
692 case 0x04000001: /* config extension */
693 case 0x04000002: /* pic control */
694 case 0x04000005: /* rate control */
695 case 0x04000007: /* motion estimation */
696 case 0x04000008: /* rdo */
697 case 0x04000009: /* vui */
698 case 0x05000002: /* auxiliary buffer */
699 case 0x05000009: /* clock table */
702 case 0x0500000c: /* hw config */
703 switch (p->adev->asic_type) {
704 #ifdef CONFIG_DRM_AMDGPU_CIK
716 case 0x03000001: /* encode */
717 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9,
722 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11,
728 case 0x02000001: /* destroy */
729 destroyed |= 1 << session_idx;
732 case 0x05000001: /* context buffer */
733 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
739 case 0x05000004: /* video bitstream buffer */
740 tmp = amdgpu_get_ib_value(p, ib_idx, idx + 4);
741 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
747 case 0x05000005: /* feedback buffer */
748 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
755 DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
760 if (session_idx == -1) {
761 DRM_ERROR("no session command at start of IB\n");
769 if (allocated & ~created) {
770 DRM_ERROR("New session without create command!\n");
776 /* No error, free all destroyed handle slots */
779 /* Error during parsing, free all allocated handle slots */
783 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
785 atomic_set(&p->adev->vce.handles[i], 0);
791 * amdgpu_vce_cs_parse_vm - parse the command stream in VM mode
796 int amdgpu_vce_ring_parse_cs_vm(struct amdgpu_cs_parser *p, uint32_t ib_idx)
798 struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
799 int session_idx = -1;
800 uint32_t destroyed = 0;
801 uint32_t created = 0;
802 uint32_t allocated = 0;
803 uint32_t tmp, handle = 0;
804 int i, r = 0, idx = 0;
806 while (idx < ib->length_dw) {
807 uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
808 uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
810 if ((len < 8) || (len & 3)) {
811 DRM_ERROR("invalid VCE command length (%d)!\n", len);
817 case 0x00000001: /* session */
818 handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
819 session_idx = amdgpu_vce_validate_handle(p, handle,
821 if (session_idx < 0) {
827 case 0x01000001: /* create */
828 created |= 1 << session_idx;
829 if (destroyed & (1 << session_idx)) {
830 destroyed &= ~(1 << session_idx);
831 allocated |= 1 << session_idx;
833 } else if (!(allocated & (1 << session_idx))) {
834 DRM_ERROR("Handle already in use!\n");
841 case 0x02000001: /* destroy */
842 destroyed |= 1 << session_idx;
849 if (session_idx == -1) {
850 DRM_ERROR("no session command at start of IB\n");
858 if (allocated & ~created) {
859 DRM_ERROR("New session without create command!\n");
865 /* No error, free all destroyed handle slots */
867 amdgpu_ib_free(p->adev, ib, NULL);
869 /* Error during parsing, free all allocated handle slots */
873 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
875 atomic_set(&p->adev->vce.handles[i], 0);
881 * amdgpu_vce_ring_emit_ib - execute indirect buffer
883 * @ring: engine to use
884 * @ib: the IB to execute
887 void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib,
888 unsigned vm_id, bool ctx_switch)
890 amdgpu_ring_write(ring, VCE_CMD_IB);
891 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
892 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
893 amdgpu_ring_write(ring, ib->length_dw);
897 * amdgpu_vce_ring_emit_fence - add a fence command to the ring
899 * @ring: engine to use
903 void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
906 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
908 amdgpu_ring_write(ring, VCE_CMD_FENCE);
909 amdgpu_ring_write(ring, addr);
910 amdgpu_ring_write(ring, upper_32_bits(addr));
911 amdgpu_ring_write(ring, seq);
912 amdgpu_ring_write(ring, VCE_CMD_TRAP);
913 amdgpu_ring_write(ring, VCE_CMD_END);
917 * amdgpu_vce_ring_test_ring - test if VCE ring is working
919 * @ring: the engine to test on
922 int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
924 struct amdgpu_device *adev = ring->adev;
925 uint32_t rptr = amdgpu_ring_get_rptr(ring);
927 int r, timeout = adev->usec_timeout;
929 /* skip ring test for sriov*/
930 if (amdgpu_sriov_vf(adev))
933 r = amdgpu_ring_alloc(ring, 16);
935 DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n",
939 amdgpu_ring_write(ring, VCE_CMD_END);
940 amdgpu_ring_commit(ring);
942 for (i = 0; i < timeout; i++) {
943 if (amdgpu_ring_get_rptr(ring) != rptr)
949 DRM_INFO("ring test on %d succeeded in %d usecs\n",
952 DRM_ERROR("amdgpu: ring %d test failed\n",
961 * amdgpu_vce_ring_test_ib - test if VCE IBs are working
963 * @ring: the engine to test on
966 int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout)
968 struct dma_fence *fence = NULL;
971 /* skip vce ring1/2 ib test for now, since it's not reliable */
972 if (ring != &ring->adev->vce.ring[0])
975 r = amdgpu_vce_get_create_msg(ring, 1, NULL);
977 DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r);
981 r = amdgpu_vce_get_destroy_msg(ring, 1, true, &fence);
983 DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r);
987 r = dma_fence_wait_timeout(fence, false, timeout);
989 DRM_ERROR("amdgpu: IB test timed out.\n");
992 DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
994 DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
998 dma_fence_put(fence);