2 * Copyright 2011 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 * Christian König <deathsimple@vodafone.de>
31 #include <linux/firmware.h>
32 #include <linux/module.h>
37 #include "amdgpu_pm.h"
38 #include "amdgpu_uvd.h"
40 #include "uvd/uvd_4_2_d.h"
42 /* 1 second timeout */
43 #define UVD_IDLE_TIMEOUT_MS 1000
46 #ifdef CONFIG_DRM_AMDGPU_CIK
47 #define FIRMWARE_BONAIRE "radeon/bonaire_uvd.bin"
48 #define FIRMWARE_KABINI "radeon/kabini_uvd.bin"
49 #define FIRMWARE_KAVERI "radeon/kaveri_uvd.bin"
50 #define FIRMWARE_HAWAII "radeon/hawaii_uvd.bin"
51 #define FIRMWARE_MULLINS "radeon/mullins_uvd.bin"
53 #define FIRMWARE_TONGA "amdgpu/tonga_uvd.bin"
54 #define FIRMWARE_CARRIZO "amdgpu/carrizo_uvd.bin"
55 #define FIRMWARE_FIJI "amdgpu/fiji_uvd.bin"
58 * amdgpu_uvd_cs_ctx - Command submission parser context
60 * Used for emulating virtual memory support on UVD 4.2.
62 struct amdgpu_uvd_cs_ctx
{
63 struct amdgpu_cs_parser
*parser
;
65 unsigned data0
, data1
;
69 /* does the IB has a msg command */
72 /* minimum buffer sizes */
76 #ifdef CONFIG_DRM_AMDGPU_CIK
77 MODULE_FIRMWARE(FIRMWARE_BONAIRE
);
78 MODULE_FIRMWARE(FIRMWARE_KABINI
);
79 MODULE_FIRMWARE(FIRMWARE_KAVERI
);
80 MODULE_FIRMWARE(FIRMWARE_HAWAII
);
81 MODULE_FIRMWARE(FIRMWARE_MULLINS
);
83 MODULE_FIRMWARE(FIRMWARE_TONGA
);
84 MODULE_FIRMWARE(FIRMWARE_CARRIZO
);
85 MODULE_FIRMWARE(FIRMWARE_FIJI
);
87 static void amdgpu_uvd_note_usage(struct amdgpu_device
*adev
);
88 static void amdgpu_uvd_idle_work_handler(struct work_struct
*work
);
90 int amdgpu_uvd_sw_init(struct amdgpu_device
*adev
)
92 unsigned long bo_size
;
94 const struct common_firmware_header
*hdr
;
95 unsigned version_major
, version_minor
, family_id
;
98 INIT_DELAYED_WORK(&adev
->uvd
.idle_work
, amdgpu_uvd_idle_work_handler
);
100 switch (adev
->asic_type
) {
101 #ifdef CONFIG_DRM_AMDGPU_CIK
103 fw_name
= FIRMWARE_BONAIRE
;
106 fw_name
= FIRMWARE_KABINI
;
109 fw_name
= FIRMWARE_KAVERI
;
112 fw_name
= FIRMWARE_HAWAII
;
115 fw_name
= FIRMWARE_MULLINS
;
119 fw_name
= FIRMWARE_TONGA
;
122 fw_name
= FIRMWARE_FIJI
;
125 fw_name
= FIRMWARE_CARRIZO
;
131 r
= request_firmware(&adev
->uvd
.fw
, fw_name
, adev
->dev
);
133 dev_err(adev
->dev
, "amdgpu_uvd: Can't load firmware \"%s\"\n",
138 r
= amdgpu_ucode_validate(adev
->uvd
.fw
);
140 dev_err(adev
->dev
, "amdgpu_uvd: Can't validate firmware \"%s\"\n",
142 release_firmware(adev
->uvd
.fw
);
147 hdr
= (const struct common_firmware_header
*)adev
->uvd
.fw
->data
;
148 family_id
= le32_to_cpu(hdr
->ucode_version
) & 0xff;
149 version_major
= (le32_to_cpu(hdr
->ucode_version
) >> 24) & 0xff;
150 version_minor
= (le32_to_cpu(hdr
->ucode_version
) >> 8) & 0xff;
151 DRM_INFO("Found UVD firmware Version: %hu.%hu Family ID: %hu\n",
152 version_major
, version_minor
, family_id
);
154 bo_size
= AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr
->ucode_size_bytes
) + 8)
155 + AMDGPU_UVD_STACK_SIZE
+ AMDGPU_UVD_HEAP_SIZE
;
156 r
= amdgpu_bo_create(adev
, bo_size
, PAGE_SIZE
, true,
157 AMDGPU_GEM_DOMAIN_VRAM
, 0, NULL
, &adev
->uvd
.vcpu_bo
);
159 dev_err(adev
->dev
, "(%d) failed to allocate UVD bo\n", r
);
163 r
= amdgpu_bo_reserve(adev
->uvd
.vcpu_bo
, false);
165 amdgpu_bo_unref(&adev
->uvd
.vcpu_bo
);
166 dev_err(adev
->dev
, "(%d) failed to reserve UVD bo\n", r
);
170 r
= amdgpu_bo_pin(adev
->uvd
.vcpu_bo
, AMDGPU_GEM_DOMAIN_VRAM
,
171 &adev
->uvd
.gpu_addr
);
173 amdgpu_bo_unreserve(adev
->uvd
.vcpu_bo
);
174 amdgpu_bo_unref(&adev
->uvd
.vcpu_bo
);
175 dev_err(adev
->dev
, "(%d) UVD bo pin failed\n", r
);
179 r
= amdgpu_bo_kmap(adev
->uvd
.vcpu_bo
, &adev
->uvd
.cpu_addr
);
181 dev_err(adev
->dev
, "(%d) UVD map failed\n", r
);
185 amdgpu_bo_unreserve(adev
->uvd
.vcpu_bo
);
187 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
) {
188 atomic_set(&adev
->uvd
.handles
[i
], 0);
189 adev
->uvd
.filp
[i
] = NULL
;
192 /* from uvd v5.0 HW addressing capacity increased to 64 bits */
193 if (!amdgpu_ip_block_version_cmp(adev
, AMD_IP_BLOCK_TYPE_UVD
, 5, 0))
194 adev
->uvd
.address_64_bit
= true;
199 int amdgpu_uvd_sw_fini(struct amdgpu_device
*adev
)
203 if (adev
->uvd
.vcpu_bo
== NULL
)
206 r
= amdgpu_bo_reserve(adev
->uvd
.vcpu_bo
, false);
208 amdgpu_bo_kunmap(adev
->uvd
.vcpu_bo
);
209 amdgpu_bo_unpin(adev
->uvd
.vcpu_bo
);
210 amdgpu_bo_unreserve(adev
->uvd
.vcpu_bo
);
213 amdgpu_bo_unref(&adev
->uvd
.vcpu_bo
);
215 amdgpu_ring_fini(&adev
->uvd
.ring
);
217 release_firmware(adev
->uvd
.fw
);
222 int amdgpu_uvd_suspend(struct amdgpu_device
*adev
)
226 const struct common_firmware_header
*hdr
;
229 if (adev
->uvd
.vcpu_bo
== NULL
)
232 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
)
233 if (atomic_read(&adev
->uvd
.handles
[i
]))
236 if (i
== AMDGPU_MAX_UVD_HANDLES
)
239 hdr
= (const struct common_firmware_header
*)adev
->uvd
.fw
->data
;
241 size
= amdgpu_bo_size(adev
->uvd
.vcpu_bo
);
242 size
-= le32_to_cpu(hdr
->ucode_size_bytes
);
244 ptr
= adev
->uvd
.cpu_addr
;
245 ptr
+= le32_to_cpu(hdr
->ucode_size_bytes
);
247 adev
->uvd
.saved_bo
= kmalloc(size
, GFP_KERNEL
);
248 memcpy(adev
->uvd
.saved_bo
, ptr
, size
);
253 int amdgpu_uvd_resume(struct amdgpu_device
*adev
)
257 const struct common_firmware_header
*hdr
;
260 if (adev
->uvd
.vcpu_bo
== NULL
)
263 hdr
= (const struct common_firmware_header
*)adev
->uvd
.fw
->data
;
264 offset
= le32_to_cpu(hdr
->ucode_array_offset_bytes
);
265 memcpy(adev
->uvd
.cpu_addr
, (adev
->uvd
.fw
->data
) + offset
,
266 (adev
->uvd
.fw
->size
) - offset
);
268 size
= amdgpu_bo_size(adev
->uvd
.vcpu_bo
);
269 size
-= le32_to_cpu(hdr
->ucode_size_bytes
);
270 ptr
= adev
->uvd
.cpu_addr
;
271 ptr
+= le32_to_cpu(hdr
->ucode_size_bytes
);
273 if (adev
->uvd
.saved_bo
!= NULL
) {
274 memcpy(ptr
, adev
->uvd
.saved_bo
, size
);
275 kfree(adev
->uvd
.saved_bo
);
276 adev
->uvd
.saved_bo
= NULL
;
278 memset(ptr
, 0, size
);
283 void amdgpu_uvd_free_handles(struct amdgpu_device
*adev
, struct drm_file
*filp
)
285 struct amdgpu_ring
*ring
= &adev
->uvd
.ring
;
288 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
) {
289 uint32_t handle
= atomic_read(&adev
->uvd
.handles
[i
]);
290 if (handle
!= 0 && adev
->uvd
.filp
[i
] == filp
) {
293 amdgpu_uvd_note_usage(adev
);
295 r
= amdgpu_uvd_get_destroy_msg(ring
, handle
, &fence
);
297 DRM_ERROR("Error destroying UVD (%d)!\n", r
);
301 fence_wait(fence
, false);
304 adev
->uvd
.filp
[i
] = NULL
;
305 atomic_set(&adev
->uvd
.handles
[i
], 0);
310 static void amdgpu_uvd_force_into_uvd_segment(struct amdgpu_bo
*rbo
)
313 for (i
= 0; i
< rbo
->placement
.num_placement
; ++i
) {
314 rbo
->placements
[i
].fpfn
= 0 >> PAGE_SHIFT
;
315 rbo
->placements
[i
].lpfn
= (256 * 1024 * 1024) >> PAGE_SHIFT
;
320 * amdgpu_uvd_cs_pass1 - first parsing round
322 * @ctx: UVD parser context
324 * Make sure UVD message and feedback buffers are in VRAM and
325 * nobody is violating an 256MB boundary.
327 static int amdgpu_uvd_cs_pass1(struct amdgpu_uvd_cs_ctx
*ctx
)
329 struct amdgpu_bo_va_mapping
*mapping
;
330 struct amdgpu_bo
*bo
;
331 uint32_t cmd
, lo
, hi
;
335 lo
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->data0
);
336 hi
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->data1
);
337 addr
= ((uint64_t)lo
) | (((uint64_t)hi
) << 32);
339 mapping
= amdgpu_cs_find_mapping(ctx
->parser
, addr
, &bo
);
340 if (mapping
== NULL
) {
341 DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr
);
345 if (!ctx
->parser
->adev
->uvd
.address_64_bit
) {
346 /* check if it's a message or feedback command */
347 cmd
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->idx
) >> 1;
348 if (cmd
== 0x0 || cmd
== 0x3) {
349 /* yes, force it into VRAM */
350 uint32_t domain
= AMDGPU_GEM_DOMAIN_VRAM
;
351 amdgpu_ttm_placement_from_domain(bo
, domain
);
353 amdgpu_uvd_force_into_uvd_segment(bo
);
355 r
= ttm_bo_validate(&bo
->tbo
, &bo
->placement
, false, false);
362 * amdgpu_uvd_cs_msg_decode - handle UVD decode message
364 * @msg: pointer to message structure
365 * @buf_sizes: returned buffer sizes
367 * Peek into the decode message and calculate the necessary buffer sizes.
369 static int amdgpu_uvd_cs_msg_decode(uint32_t *msg
, unsigned buf_sizes
[])
371 unsigned stream_type
= msg
[4];
372 unsigned width
= msg
[6];
373 unsigned height
= msg
[7];
374 unsigned dpb_size
= msg
[9];
375 unsigned pitch
= msg
[28];
376 unsigned level
= msg
[57];
378 unsigned width_in_mb
= width
/ 16;
379 unsigned height_in_mb
= ALIGN(height
/ 16, 2);
380 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
382 unsigned image_size
, tmp
, min_dpb_size
, num_dpb_buffer
;
383 unsigned min_ctx_size
= 0;
385 image_size
= width
* height
;
386 image_size
+= image_size
/ 2;
387 image_size
= ALIGN(image_size
, 1024);
389 switch (stream_type
) {
391 case 7: /* H264 Perf */
394 num_dpb_buffer
= 8100 / fs_in_mb
;
397 num_dpb_buffer
= 18000 / fs_in_mb
;
400 num_dpb_buffer
= 20480 / fs_in_mb
;
403 num_dpb_buffer
= 32768 / fs_in_mb
;
406 num_dpb_buffer
= 34816 / fs_in_mb
;
409 num_dpb_buffer
= 110400 / fs_in_mb
;
412 num_dpb_buffer
= 184320 / fs_in_mb
;
415 num_dpb_buffer
= 184320 / fs_in_mb
;
419 if (num_dpb_buffer
> 17)
422 /* reference picture buffer */
423 min_dpb_size
= image_size
* num_dpb_buffer
;
425 /* macroblock context buffer */
426 min_dpb_size
+= width_in_mb
* height_in_mb
* num_dpb_buffer
* 192;
428 /* IT surface buffer */
429 min_dpb_size
+= width_in_mb
* height_in_mb
* 32;
434 /* reference picture buffer */
435 min_dpb_size
= image_size
* 3;
438 min_dpb_size
+= width_in_mb
* height_in_mb
* 128;
440 /* IT surface buffer */
441 min_dpb_size
+= width_in_mb
* 64;
443 /* DB surface buffer */
444 min_dpb_size
+= width_in_mb
* 128;
447 tmp
= max(width_in_mb
, height_in_mb
);
448 min_dpb_size
+= ALIGN(tmp
* 7 * 16, 64);
453 /* reference picture buffer */
454 min_dpb_size
= image_size
* 3;
459 /* reference picture buffer */
460 min_dpb_size
= image_size
* 3;
463 min_dpb_size
+= width_in_mb
* height_in_mb
* 64;
465 /* IT surface buffer */
466 min_dpb_size
+= ALIGN(width_in_mb
* height_in_mb
* 32, 64);
470 image_size
= (ALIGN(width
, 16) * ALIGN(height
, 16) * 3) / 2;
471 image_size
= ALIGN(image_size
, 256);
473 num_dpb_buffer
= (le32_to_cpu(msg
[59]) & 0xff) + 2;
474 min_dpb_size
= image_size
* num_dpb_buffer
;
475 min_ctx_size
= ((width
+ 255) / 16) * ((height
+ 255) / 16)
476 * 16 * num_dpb_buffer
+ 52 * 1024;
480 DRM_ERROR("UVD codec not handled %d!\n", stream_type
);
485 DRM_ERROR("Invalid UVD decoding target pitch!\n");
489 if (dpb_size
< min_dpb_size
) {
490 DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n",
491 dpb_size
, min_dpb_size
);
495 buf_sizes
[0x1] = dpb_size
;
496 buf_sizes
[0x2] = image_size
;
497 buf_sizes
[0x4] = min_ctx_size
;
502 * amdgpu_uvd_cs_msg - handle UVD message
504 * @ctx: UVD parser context
505 * @bo: buffer object containing the message
506 * @offset: offset into the buffer object
508 * Peek into the UVD message and extract the session id.
509 * Make sure that we don't open up to many sessions.
511 static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx
*ctx
,
512 struct amdgpu_bo
*bo
, unsigned offset
)
514 struct amdgpu_device
*adev
= ctx
->parser
->adev
;
515 int32_t *msg
, msg_type
, handle
;
521 DRM_ERROR("UVD messages must be 64 byte aligned!\n");
525 r
= reservation_object_wait_timeout_rcu(bo
->tbo
.resv
, true, false,
526 MAX_SCHEDULE_TIMEOUT
);
528 DRM_ERROR("Failed waiting for UVD message (%ld)!\n", r
);
532 r
= amdgpu_bo_kmap(bo
, &ptr
);
534 DRM_ERROR("Failed mapping the UVD message (%ld)!\n", r
);
544 DRM_ERROR("Invalid UVD handle!\n");
549 /* it's a decode msg, calc buffer sizes */
550 r
= amdgpu_uvd_cs_msg_decode(msg
, ctx
->buf_sizes
);
551 amdgpu_bo_kunmap(bo
);
555 } else if (msg_type
== 2) {
556 /* it's a destroy msg, free the handle */
557 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
)
558 atomic_cmpxchg(&adev
->uvd
.handles
[i
], handle
, 0);
559 amdgpu_bo_kunmap(bo
);
562 /* it's a create msg */
563 amdgpu_bo_kunmap(bo
);
566 DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type
);
570 /* it's a create msg, no special handling needed */
573 /* create or decode, validate the handle */
574 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
) {
575 if (atomic_read(&adev
->uvd
.handles
[i
]) == handle
)
579 /* handle not found try to alloc a new one */
580 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
) {
581 if (!atomic_cmpxchg(&adev
->uvd
.handles
[i
], 0, handle
)) {
582 adev
->uvd
.filp
[i
] = ctx
->parser
->filp
;
587 DRM_ERROR("No more free UVD handles!\n");
592 * amdgpu_uvd_cs_pass2 - second parsing round
594 * @ctx: UVD parser context
596 * Patch buffer addresses, make sure buffer sizes are correct.
598 static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx
*ctx
)
600 struct amdgpu_bo_va_mapping
*mapping
;
601 struct amdgpu_bo
*bo
;
602 struct amdgpu_ib
*ib
;
603 uint32_t cmd
, lo
, hi
;
608 lo
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->data0
);
609 hi
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->data1
);
610 addr
= ((uint64_t)lo
) | (((uint64_t)hi
) << 32);
612 mapping
= amdgpu_cs_find_mapping(ctx
->parser
, addr
, &bo
);
616 start
= amdgpu_bo_gpu_offset(bo
);
618 end
= (mapping
->it
.last
+ 1 - mapping
->it
.start
);
619 end
= end
* AMDGPU_GPU_PAGE_SIZE
+ start
;
621 addr
-= ((uint64_t)mapping
->it
.start
) * AMDGPU_GPU_PAGE_SIZE
;
624 ib
= &ctx
->parser
->ibs
[ctx
->ib_idx
];
625 ib
->ptr
[ctx
->data0
] = start
& 0xFFFFFFFF;
626 ib
->ptr
[ctx
->data1
] = start
>> 32;
628 cmd
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->idx
) >> 1;
630 if ((end
- start
) < ctx
->buf_sizes
[cmd
]) {
631 DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd
,
632 (unsigned)(end
- start
),
633 ctx
->buf_sizes
[cmd
]);
637 } else if (cmd
== 0x206) {
638 if ((end
- start
) < ctx
->buf_sizes
[4]) {
639 DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd
,
640 (unsigned)(end
- start
),
644 } else if ((cmd
!= 0x100) && (cmd
!= 0x204)) {
645 DRM_ERROR("invalid UVD command %X!\n", cmd
);
649 if (!ctx
->parser
->adev
->uvd
.address_64_bit
) {
650 if ((start
>> 28) != ((end
- 1) >> 28)) {
651 DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n",
656 if ((cmd
== 0 || cmd
== 0x3) &&
657 (start
>> 28) != (ctx
->parser
->adev
->uvd
.gpu_addr
>> 28)) {
658 DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
665 ctx
->has_msg_cmd
= true;
666 r
= amdgpu_uvd_cs_msg(ctx
, bo
, addr
);
669 } else if (!ctx
->has_msg_cmd
) {
670 DRM_ERROR("Message needed before other commands are send!\n");
678 * amdgpu_uvd_cs_reg - parse register writes
680 * @ctx: UVD parser context
681 * @cb: callback function
683 * Parse the register writes, call cb on each complete command.
685 static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx
*ctx
,
686 int (*cb
)(struct amdgpu_uvd_cs_ctx
*ctx
))
688 struct amdgpu_ib
*ib
= &ctx
->parser
->ibs
[ctx
->ib_idx
];
692 for (i
= 0; i
<= ctx
->count
; ++i
) {
693 unsigned reg
= ctx
->reg
+ i
;
695 if (ctx
->idx
>= ib
->length_dw
) {
696 DRM_ERROR("Register command after end of CS!\n");
701 case mmUVD_GPCOM_VCPU_DATA0
:
702 ctx
->data0
= ctx
->idx
;
704 case mmUVD_GPCOM_VCPU_DATA1
:
705 ctx
->data1
= ctx
->idx
;
707 case mmUVD_GPCOM_VCPU_CMD
:
712 case mmUVD_ENGINE_CNTL
:
715 DRM_ERROR("Invalid reg 0x%X!\n", reg
);
724 * amdgpu_uvd_cs_packets - parse UVD packets
726 * @ctx: UVD parser context
727 * @cb: callback function
729 * Parse the command stream packets.
731 static int amdgpu_uvd_cs_packets(struct amdgpu_uvd_cs_ctx
*ctx
,
732 int (*cb
)(struct amdgpu_uvd_cs_ctx
*ctx
))
734 struct amdgpu_ib
*ib
= &ctx
->parser
->ibs
[ctx
->ib_idx
];
737 for (ctx
->idx
= 0 ; ctx
->idx
< ib
->length_dw
; ) {
738 uint32_t cmd
= amdgpu_get_ib_value(ctx
->parser
, ctx
->ib_idx
, ctx
->idx
);
739 unsigned type
= CP_PACKET_GET_TYPE(cmd
);
742 ctx
->reg
= CP_PACKET0_GET_REG(cmd
);
743 ctx
->count
= CP_PACKET_GET_COUNT(cmd
);
744 r
= amdgpu_uvd_cs_reg(ctx
, cb
);
752 DRM_ERROR("Unknown packet type %d !\n", type
);
760 * amdgpu_uvd_ring_parse_cs - UVD command submission parser
762 * @parser: Command submission parser context
764 * Parse the command stream, patch in addresses as necessary.
766 int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser
*parser
, uint32_t ib_idx
)
768 struct amdgpu_uvd_cs_ctx ctx
= {};
769 unsigned buf_sizes
[] = {
771 [0x00000001] = 0xFFFFFFFF,
772 [0x00000002] = 0xFFFFFFFF,
774 [0x00000004] = 0xFFFFFFFF,
776 struct amdgpu_ib
*ib
= &parser
->ibs
[ib_idx
];
779 if (ib
->length_dw
% 16) {
780 DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n",
786 ctx
.buf_sizes
= buf_sizes
;
789 /* first round, make sure the buffers are actually in the UVD segment */
790 r
= amdgpu_uvd_cs_packets(&ctx
, amdgpu_uvd_cs_pass1
);
794 /* second round, patch buffer addresses into the command stream */
795 r
= amdgpu_uvd_cs_packets(&ctx
, amdgpu_uvd_cs_pass2
);
799 if (!ctx
.has_msg_cmd
) {
800 DRM_ERROR("UVD-IBs need a msg command!\n");
804 amdgpu_uvd_note_usage(ctx
.parser
->adev
);
809 static int amdgpu_uvd_free_job(
810 struct amdgpu_job
*sched_job
)
812 amdgpu_ib_free(sched_job
->adev
, sched_job
->ibs
);
813 kfree(sched_job
->ibs
);
817 static int amdgpu_uvd_send_msg(struct amdgpu_ring
*ring
,
818 struct amdgpu_bo
*bo
,
819 struct fence
**fence
)
821 struct ttm_validate_buffer tv
;
822 struct ww_acquire_ctx ticket
;
823 struct list_head head
;
824 struct amdgpu_ib
*ib
= NULL
;
825 struct fence
*f
= NULL
;
826 struct amdgpu_device
*adev
= ring
->adev
;
830 memset(&tv
, 0, sizeof(tv
));
833 INIT_LIST_HEAD(&head
);
834 list_add(&tv
.head
, &head
);
836 r
= ttm_eu_reserve_buffers(&ticket
, &head
, true, NULL
);
840 if (!bo
->adev
->uvd
.address_64_bit
) {
841 amdgpu_ttm_placement_from_domain(bo
, AMDGPU_GEM_DOMAIN_VRAM
);
842 amdgpu_uvd_force_into_uvd_segment(bo
);
845 r
= ttm_bo_validate(&bo
->tbo
, &bo
->placement
, true, false);
848 ib
= kzalloc(sizeof(struct amdgpu_ib
), GFP_KERNEL
);
853 r
= amdgpu_ib_get(ring
, NULL
, 64, ib
);
857 addr
= amdgpu_bo_gpu_offset(bo
);
858 ib
->ptr
[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0
, 0);
860 ib
->ptr
[2] = PACKET0(mmUVD_GPCOM_VCPU_DATA1
, 0);
861 ib
->ptr
[3] = addr
>> 32;
862 ib
->ptr
[4] = PACKET0(mmUVD_GPCOM_VCPU_CMD
, 0);
864 for (i
= 6; i
< 16; ++i
)
865 ib
->ptr
[i
] = PACKET2(0);
868 r
= amdgpu_sched_ib_submit_kernel_helper(adev
, ring
, ib
, 1,
869 &amdgpu_uvd_free_job
,
870 AMDGPU_FENCE_OWNER_UNDEFINED
,
875 ttm_eu_fence_buffer_objects(&ticket
, &head
, f
);
878 *fence
= fence_get(f
);
879 amdgpu_bo_unref(&bo
);
881 if (amdgpu_enable_scheduler
)
884 amdgpu_ib_free(ring
->adev
, ib
);
888 amdgpu_ib_free(ring
->adev
, ib
);
892 ttm_eu_backoff_reservation(&ticket
, &head
);
896 /* multiple fence commands without any stream commands in between can
897 crash the vcpu so just try to emmit a dummy create/destroy msg to
899 int amdgpu_uvd_get_create_msg(struct amdgpu_ring
*ring
, uint32_t handle
,
900 struct fence
**fence
)
902 struct amdgpu_device
*adev
= ring
->adev
;
903 struct amdgpu_bo
*bo
;
907 r
= amdgpu_bo_create(adev
, 1024, PAGE_SIZE
, true,
908 AMDGPU_GEM_DOMAIN_VRAM
, 0, NULL
, &bo
);
912 r
= amdgpu_bo_reserve(bo
, false);
914 amdgpu_bo_unref(&bo
);
918 r
= amdgpu_bo_kmap(bo
, (void **)&msg
);
920 amdgpu_bo_unreserve(bo
);
921 amdgpu_bo_unref(&bo
);
925 /* stitch together an UVD create msg */
926 msg
[0] = cpu_to_le32(0x00000de4);
927 msg
[1] = cpu_to_le32(0x00000000);
928 msg
[2] = cpu_to_le32(handle
);
929 msg
[3] = cpu_to_le32(0x00000000);
930 msg
[4] = cpu_to_le32(0x00000000);
931 msg
[5] = cpu_to_le32(0x00000000);
932 msg
[6] = cpu_to_le32(0x00000000);
933 msg
[7] = cpu_to_le32(0x00000780);
934 msg
[8] = cpu_to_le32(0x00000440);
935 msg
[9] = cpu_to_le32(0x00000000);
936 msg
[10] = cpu_to_le32(0x01b37000);
937 for (i
= 11; i
< 1024; ++i
)
938 msg
[i
] = cpu_to_le32(0x0);
940 amdgpu_bo_kunmap(bo
);
941 amdgpu_bo_unreserve(bo
);
943 return amdgpu_uvd_send_msg(ring
, bo
, fence
);
946 int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring
*ring
, uint32_t handle
,
947 struct fence
**fence
)
949 struct amdgpu_device
*adev
= ring
->adev
;
950 struct amdgpu_bo
*bo
;
954 r
= amdgpu_bo_create(adev
, 1024, PAGE_SIZE
, true,
955 AMDGPU_GEM_DOMAIN_VRAM
, 0, NULL
, &bo
);
959 r
= amdgpu_bo_reserve(bo
, false);
961 amdgpu_bo_unref(&bo
);
965 r
= amdgpu_bo_kmap(bo
, (void **)&msg
);
967 amdgpu_bo_unreserve(bo
);
968 amdgpu_bo_unref(&bo
);
972 /* stitch together an UVD destroy msg */
973 msg
[0] = cpu_to_le32(0x00000de4);
974 msg
[1] = cpu_to_le32(0x00000002);
975 msg
[2] = cpu_to_le32(handle
);
976 msg
[3] = cpu_to_le32(0x00000000);
977 for (i
= 4; i
< 1024; ++i
)
978 msg
[i
] = cpu_to_le32(0x0);
980 amdgpu_bo_kunmap(bo
);
981 amdgpu_bo_unreserve(bo
);
983 return amdgpu_uvd_send_msg(ring
, bo
, fence
);
986 static void amdgpu_uvd_idle_work_handler(struct work_struct
*work
)
988 struct amdgpu_device
*adev
=
989 container_of(work
, struct amdgpu_device
, uvd
.idle_work
.work
);
990 unsigned i
, fences
, handles
= 0;
992 fences
= amdgpu_fence_count_emitted(&adev
->uvd
.ring
);
994 for (i
= 0; i
< AMDGPU_MAX_UVD_HANDLES
; ++i
)
995 if (atomic_read(&adev
->uvd
.handles
[i
]))
998 if (fences
== 0 && handles
== 0) {
999 if (adev
->pm
.dpm_enabled
) {
1000 amdgpu_dpm_enable_uvd(adev
, false);
1002 amdgpu_asic_set_uvd_clocks(adev
, 0, 0);
1005 schedule_delayed_work(&adev
->uvd
.idle_work
,
1006 msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS
));
1010 static void amdgpu_uvd_note_usage(struct amdgpu_device
*adev
)
1012 bool set_clocks
= !cancel_delayed_work_sync(&adev
->uvd
.idle_work
);
1013 set_clocks
&= schedule_delayed_work(&adev
->uvd
.idle_work
,
1014 msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS
));
1017 if (adev
->pm
.dpm_enabled
) {
1018 amdgpu_dpm_enable_uvd(adev
, true);
1020 amdgpu_asic_set_uvd_clocks(adev
, 53300, 40000);