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
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_MS 1000
42 #ifdef CONFIG_DRM_AMDGPU_CIK
43 #define FIRMWARE_BONAIRE "radeon/bonaire_vce.bin"
44 #define FIRMWARE_KABINI "radeon/kabini_vce.bin"
45 #define FIRMWARE_KAVERI "radeon/kaveri_vce.bin"
46 #define FIRMWARE_HAWAII "radeon/hawaii_vce.bin"
47 #define FIRMWARE_MULLINS "radeon/mullins_vce.bin"
49 #define FIRMWARE_TONGA "amdgpu/tonga_vce.bin"
50 #define FIRMWARE_CARRIZO "amdgpu/carrizo_vce.bin"
52 #ifdef CONFIG_DRM_AMDGPU_CIK
53 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
54 MODULE_FIRMWARE(FIRMWARE_KABINI);
55 MODULE_FIRMWARE(FIRMWARE_KAVERI);
56 MODULE_FIRMWARE(FIRMWARE_HAWAII);
57 MODULE_FIRMWARE(FIRMWARE_MULLINS);
59 MODULE_FIRMWARE(FIRMWARE_TONGA);
60 MODULE_FIRMWARE(FIRMWARE_CARRIZO);
62 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
65 * amdgpu_vce_init - allocate memory, load vce firmware
67 * @adev: amdgpu_device pointer
69 * First step to get VCE online, allocate memory and load the firmware
71 int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
74 const struct common_firmware_header *hdr;
75 unsigned ucode_version, version_major, version_minor, binary_id;
78 INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler);
80 switch (adev->asic_type) {
81 #ifdef CONFIG_DRM_AMDGPU_CIK
83 fw_name = FIRMWARE_BONAIRE;
86 fw_name = FIRMWARE_KAVERI;
89 fw_name = FIRMWARE_KABINI;
92 fw_name = FIRMWARE_HAWAII;
95 fw_name = FIRMWARE_MULLINS;
99 fw_name = FIRMWARE_TONGA;
102 fw_name = FIRMWARE_CARRIZO;
109 r = request_firmware(&adev->vce.fw, fw_name, adev->dev);
111 dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n",
116 r = amdgpu_ucode_validate(adev->vce.fw);
118 dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n",
120 release_firmware(adev->vce.fw);
125 hdr = (const struct common_firmware_header *)adev->vce.fw->data;
127 ucode_version = le32_to_cpu(hdr->ucode_version);
128 version_major = (ucode_version >> 20) & 0xfff;
129 version_minor = (ucode_version >> 8) & 0xfff;
130 binary_id = ucode_version & 0xff;
131 DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n",
132 version_major, version_minor, binary_id);
133 adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) |
136 /* allocate firmware, stack and heap BO */
138 r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
139 AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->vce.vcpu_bo);
141 dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
145 r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
147 amdgpu_bo_unref(&adev->vce.vcpu_bo);
148 dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
152 r = amdgpu_bo_pin(adev->vce.vcpu_bo, AMDGPU_GEM_DOMAIN_VRAM,
153 &adev->vce.gpu_addr);
154 amdgpu_bo_unreserve(adev->vce.vcpu_bo);
156 amdgpu_bo_unref(&adev->vce.vcpu_bo);
157 dev_err(adev->dev, "(%d) VCE bo pin failed\n", r);
161 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
162 atomic_set(&adev->vce.handles[i], 0);
163 adev->vce.filp[i] = NULL;
170 * amdgpu_vce_fini - free memory
172 * @adev: amdgpu_device pointer
174 * Last step on VCE teardown, free firmware memory
176 int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
178 if (adev->vce.vcpu_bo == NULL)
181 amdgpu_bo_unref(&adev->vce.vcpu_bo);
183 amdgpu_ring_fini(&adev->vce.ring[0]);
184 amdgpu_ring_fini(&adev->vce.ring[1]);
186 release_firmware(adev->vce.fw);
192 * amdgpu_vce_suspend - unpin VCE fw memory
194 * @adev: amdgpu_device pointer
197 int amdgpu_vce_suspend(struct amdgpu_device *adev)
201 if (adev->vce.vcpu_bo == NULL)
204 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
205 if (atomic_read(&adev->vce.handles[i]))
208 if (i == AMDGPU_MAX_VCE_HANDLES)
211 /* TODO: suspending running encoding sessions isn't supported */
216 * amdgpu_vce_resume - pin VCE fw memory
218 * @adev: amdgpu_device pointer
221 int amdgpu_vce_resume(struct amdgpu_device *adev)
224 const struct common_firmware_header *hdr;
228 if (adev->vce.vcpu_bo == NULL)
231 r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
233 dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
237 r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr);
239 amdgpu_bo_unreserve(adev->vce.vcpu_bo);
240 dev_err(adev->dev, "(%d) VCE map failed\n", r);
244 hdr = (const struct common_firmware_header *)adev->vce.fw->data;
245 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
246 memcpy(cpu_addr, (adev->vce.fw->data) + offset,
247 (adev->vce.fw->size) - offset);
249 amdgpu_bo_kunmap(adev->vce.vcpu_bo);
251 amdgpu_bo_unreserve(adev->vce.vcpu_bo);
257 * amdgpu_vce_idle_work_handler - power off VCE
259 * @work: pointer to work structure
261 * power of VCE when it's not used any more
263 static void amdgpu_vce_idle_work_handler(struct work_struct *work)
265 struct amdgpu_device *adev =
266 container_of(work, struct amdgpu_device, vce.idle_work.work);
268 if ((amdgpu_fence_count_emitted(&adev->vce.ring[0]) == 0) &&
269 (amdgpu_fence_count_emitted(&adev->vce.ring[1]) == 0)) {
270 if (adev->pm.dpm_enabled) {
271 amdgpu_dpm_enable_vce(adev, false);
273 amdgpu_asic_set_vce_clocks(adev, 0, 0);
276 schedule_delayed_work(&adev->vce.idle_work,
277 msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
282 * amdgpu_vce_note_usage - power up VCE
284 * @adev: amdgpu_device pointer
286 * Make sure VCE is powerd up when we want to use it
288 static void amdgpu_vce_note_usage(struct amdgpu_device *adev)
290 bool streams_changed = false;
291 bool set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work);
292 set_clocks &= schedule_delayed_work(&adev->vce.idle_work,
293 msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
295 if (adev->pm.dpm_enabled) {
296 /* XXX figure out if the streams changed */
297 streams_changed = false;
300 if (set_clocks || streams_changed) {
301 if (adev->pm.dpm_enabled) {
302 amdgpu_dpm_enable_vce(adev, true);
304 amdgpu_asic_set_vce_clocks(adev, 53300, 40000);
310 * amdgpu_vce_free_handles - free still open VCE handles
312 * @adev: amdgpu_device pointer
313 * @filp: drm file pointer
315 * Close all VCE handles still open by this file pointer
317 void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
319 struct amdgpu_ring *ring = &adev->vce.ring[0];
321 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
322 uint32_t handle = atomic_read(&adev->vce.handles[i]);
323 if (!handle || adev->vce.filp[i] != filp)
326 amdgpu_vce_note_usage(adev);
328 r = amdgpu_vce_get_destroy_msg(ring, handle, NULL);
330 DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
332 adev->vce.filp[i] = NULL;
333 atomic_set(&adev->vce.handles[i], 0);
338 * amdgpu_vce_get_create_msg - generate a VCE create msg
340 * @adev: amdgpu_device pointer
341 * @ring: ring we should submit the msg to
342 * @handle: VCE session handle to use
343 * @fence: optional fence to return
345 * Open up a stream for HW test
347 int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
348 struct amdgpu_fence **fence)
350 const unsigned ib_size_dw = 1024;
355 r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
357 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
361 dummy = ib.gpu_addr + 1024;
363 /* stitch together an VCE create msg */
365 ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
366 ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
367 ib.ptr[ib.length_dw++] = handle;
369 ib.ptr[ib.length_dw++] = 0x00000030; /* len */
370 ib.ptr[ib.length_dw++] = 0x01000001; /* create cmd */
371 ib.ptr[ib.length_dw++] = 0x00000000;
372 ib.ptr[ib.length_dw++] = 0x00000042;
373 ib.ptr[ib.length_dw++] = 0x0000000a;
374 ib.ptr[ib.length_dw++] = 0x00000001;
375 ib.ptr[ib.length_dw++] = 0x00000080;
376 ib.ptr[ib.length_dw++] = 0x00000060;
377 ib.ptr[ib.length_dw++] = 0x00000100;
378 ib.ptr[ib.length_dw++] = 0x00000100;
379 ib.ptr[ib.length_dw++] = 0x0000000c;
380 ib.ptr[ib.length_dw++] = 0x00000000;
382 ib.ptr[ib.length_dw++] = 0x00000014; /* len */
383 ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
384 ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
385 ib.ptr[ib.length_dw++] = dummy;
386 ib.ptr[ib.length_dw++] = 0x00000001;
388 for (i = ib.length_dw; i < ib_size_dw; ++i)
391 r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
393 DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
397 *fence = amdgpu_fence_ref(ib.fence);
399 amdgpu_ib_free(ring->adev, &ib);
405 * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg
407 * @adev: amdgpu_device pointer
408 * @ring: ring we should submit the msg to
409 * @handle: VCE session handle to use
410 * @fence: optional fence to return
412 * Close up a stream for HW test or if userspace failed to do so
414 int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
415 struct amdgpu_fence **fence)
417 const unsigned ib_size_dw = 1024;
422 r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
424 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
428 dummy = ib.gpu_addr + 1024;
430 /* stitch together an VCE destroy msg */
432 ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
433 ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
434 ib.ptr[ib.length_dw++] = handle;
436 ib.ptr[ib.length_dw++] = 0x00000014; /* len */
437 ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
438 ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
439 ib.ptr[ib.length_dw++] = dummy;
440 ib.ptr[ib.length_dw++] = 0x00000001;
442 ib.ptr[ib.length_dw++] = 0x00000008; /* len */
443 ib.ptr[ib.length_dw++] = 0x02000001; /* destroy cmd */
445 for (i = ib.length_dw; i < ib_size_dw; ++i)
448 r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
450 DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
454 *fence = amdgpu_fence_ref(ib.fence);
456 amdgpu_ib_free(ring->adev, &ib);
462 * amdgpu_vce_cs_reloc - command submission relocation
465 * @lo: address of lower dword
466 * @hi: address of higher dword
468 * Patch relocation inside command stream with real buffer address
470 int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx, int lo, int hi)
472 struct amdgpu_bo_va_mapping *mapping;
473 struct amdgpu_ib *ib = &p->ibs[ib_idx];
474 struct amdgpu_bo *bo;
477 addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) |
478 ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32;
480 mapping = amdgpu_cs_find_mapping(p, addr, &bo);
481 if (mapping == NULL) {
482 DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d\n",
487 addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE;
488 addr += amdgpu_bo_gpu_offset(bo);
490 ib->ptr[lo] = addr & 0xFFFFFFFF;
491 ib->ptr[hi] = addr >> 32;
497 * amdgpu_vce_cs_parse - parse and validate the command stream
502 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
505 bool destroy = false;
507 struct amdgpu_ib *ib = &p->ibs[ib_idx];
509 amdgpu_vce_note_usage(p->adev);
511 while (idx < ib->length_dw) {
512 uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
513 uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
515 if ((len < 8) || (len & 3)) {
516 DRM_ERROR("invalid VCE command length (%d)!\n", len);
521 case 0x00000001: // session
522 handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
525 case 0x00000002: // task info
526 case 0x01000001: // create
527 case 0x04000001: // config extension
528 case 0x04000002: // pic control
529 case 0x04000005: // rate control
530 case 0x04000007: // motion estimation
531 case 0x04000008: // rdo
532 case 0x04000009: // vui
533 case 0x05000002: // auxiliary buffer
536 case 0x03000001: // encode
537 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9);
541 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11);
546 case 0x02000001: // destroy
550 case 0x05000001: // context buffer
551 case 0x05000004: // video bitstream buffer
552 case 0x05000005: // feedback buffer
553 r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2);
559 DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
567 /* IB contains a destroy msg, free the handle */
568 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
569 atomic_cmpxchg(&p->adev->vce.handles[i], handle, 0);
574 /* create or encode, validate the handle */
575 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
576 if (atomic_read(&p->adev->vce.handles[i]) == handle)
580 /* handle not found try to alloc a new one */
581 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
582 if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) {
583 p->adev->vce.filp[i] = p->filp;
588 DRM_ERROR("No more free VCE handles!\n");
594 * amdgpu_vce_ring_emit_semaphore - emit a semaphore command
596 * @ring: engine to use
597 * @semaphore: address of semaphore
598 * @emit_wait: true=emit wait, false=emit signal
601 bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring,
602 struct amdgpu_semaphore *semaphore,
605 uint64_t addr = semaphore->gpu_addr;
607 amdgpu_ring_write(ring, VCE_CMD_SEMAPHORE);
608 amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF);
609 amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF);
610 amdgpu_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0));
612 amdgpu_ring_write(ring, VCE_CMD_END);
618 * amdgpu_vce_ring_emit_ib - execute indirect buffer
620 * @ring: engine to use
621 * @ib: the IB to execute
624 void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
626 amdgpu_ring_write(ring, VCE_CMD_IB);
627 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
628 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
629 amdgpu_ring_write(ring, ib->length_dw);
633 * amdgpu_vce_ring_emit_fence - add a fence command to the ring
635 * @ring: engine to use
639 void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
642 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
644 amdgpu_ring_write(ring, VCE_CMD_FENCE);
645 amdgpu_ring_write(ring, addr);
646 amdgpu_ring_write(ring, upper_32_bits(addr));
647 amdgpu_ring_write(ring, seq);
648 amdgpu_ring_write(ring, VCE_CMD_TRAP);
649 amdgpu_ring_write(ring, VCE_CMD_END);
653 * amdgpu_vce_ring_test_ring - test if VCE ring is working
655 * @ring: the engine to test on
658 int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
660 struct amdgpu_device *adev = ring->adev;
661 uint32_t rptr = amdgpu_ring_get_rptr(ring);
665 r = amdgpu_ring_lock(ring, 16);
667 DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n",
671 amdgpu_ring_write(ring, VCE_CMD_END);
672 amdgpu_ring_unlock_commit(ring);
674 for (i = 0; i < adev->usec_timeout; i++) {
675 if (amdgpu_ring_get_rptr(ring) != rptr)
680 if (i < adev->usec_timeout) {
681 DRM_INFO("ring test on %d succeeded in %d usecs\n",
684 DRM_ERROR("amdgpu: ring %d test failed\n",
693 * amdgpu_vce_ring_test_ib - test if VCE IBs are working
695 * @ring: the engine to test on
698 int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
700 struct amdgpu_fence *fence = NULL;
703 r = amdgpu_vce_get_create_msg(ring, 1, NULL);
705 DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r);
709 r = amdgpu_vce_get_destroy_msg(ring, 1, &fence);
711 DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r);
715 r = amdgpu_fence_wait(fence, false);
717 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
719 DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
722 amdgpu_fence_unref(&fence);