1 /* amdgpu_drm.h -- Public header for the amdgpu driver -*- linux-c -*-
3 * Copyright 2000 Precision Insight, Inc., Cedar Park, Texas.
4 * Copyright 2000 VA Linux Systems, Inc., Fremont, California.
5 * Copyright 2002 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Copyright 2014 Advanced Micro Devices, Inc.
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
22 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
23 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
24 * OTHER DEALINGS IN THE SOFTWARE.
32 #ifndef __AMDGPU_DRM_H__
33 #define __AMDGPU_DRM_H__
37 #if defined(__cplusplus)
41 #define DRM_AMDGPU_GEM_CREATE 0x00
42 #define DRM_AMDGPU_GEM_MMAP 0x01
43 #define DRM_AMDGPU_CTX 0x02
44 #define DRM_AMDGPU_BO_LIST 0x03
45 #define DRM_AMDGPU_CS 0x04
46 #define DRM_AMDGPU_INFO 0x05
47 #define DRM_AMDGPU_GEM_METADATA 0x06
48 #define DRM_AMDGPU_GEM_WAIT_IDLE 0x07
49 #define DRM_AMDGPU_GEM_VA 0x08
50 #define DRM_AMDGPU_WAIT_CS 0x09
51 #define DRM_AMDGPU_GEM_OP 0x10
52 #define DRM_AMDGPU_GEM_USERPTR 0x11
53 #define DRM_AMDGPU_WAIT_FENCES 0x12
54 #define DRM_AMDGPU_VM 0x13
55 #define DRM_AMDGPU_FENCE_TO_HANDLE 0x14
56 #define DRM_AMDGPU_SCHED 0x15
58 #define DRM_IOCTL_AMDGPU_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_CREATE, union drm_amdgpu_gem_create)
59 #define DRM_IOCTL_AMDGPU_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_MMAP, union drm_amdgpu_gem_mmap)
60 #define DRM_IOCTL_AMDGPU_CTX DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_CTX, union drm_amdgpu_ctx)
61 #define DRM_IOCTL_AMDGPU_BO_LIST DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_BO_LIST, union drm_amdgpu_bo_list)
62 #define DRM_IOCTL_AMDGPU_CS DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_CS, union drm_amdgpu_cs)
63 #define DRM_IOCTL_AMDGPU_INFO DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_INFO, struct drm_amdgpu_info)
64 #define DRM_IOCTL_AMDGPU_GEM_METADATA DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_METADATA, struct drm_amdgpu_gem_metadata)
65 #define DRM_IOCTL_AMDGPU_GEM_WAIT_IDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_WAIT_IDLE, union drm_amdgpu_gem_wait_idle)
66 #define DRM_IOCTL_AMDGPU_GEM_VA DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_VA, struct drm_amdgpu_gem_va)
67 #define DRM_IOCTL_AMDGPU_WAIT_CS DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_CS, union drm_amdgpu_wait_cs)
68 #define DRM_IOCTL_AMDGPU_GEM_OP DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_OP, struct drm_amdgpu_gem_op)
69 #define DRM_IOCTL_AMDGPU_GEM_USERPTR DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_USERPTR, struct drm_amdgpu_gem_userptr)
70 #define DRM_IOCTL_AMDGPU_WAIT_FENCES DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_FENCES, union drm_amdgpu_wait_fences)
71 #define DRM_IOCTL_AMDGPU_VM DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_VM, union drm_amdgpu_vm)
72 #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
73 #define DRM_IOCTL_AMDGPU_SCHED DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
78 * %AMDGPU_GEM_DOMAIN_CPU System memory that is not GPU accessible.
79 * Memory in this pool could be swapped out to disk if there is pressure.
81 * %AMDGPU_GEM_DOMAIN_GTT GPU accessible system memory, mapped into the
82 * GPU's virtual address space via gart. Gart memory linearizes non-contiguous
83 * pages of system memory, allows GPU access system memory in a linearized
86 * %AMDGPU_GEM_DOMAIN_VRAM Local video memory. For APUs, it is memory
87 * carved out by the BIOS.
89 * %AMDGPU_GEM_DOMAIN_GDS Global on-chip data storage used to share data
90 * across shader threads.
92 * %AMDGPU_GEM_DOMAIN_GWS Global wave sync, used to synchronize the
93 * execution of all the waves on a device.
95 * %AMDGPU_GEM_DOMAIN_OA Ordered append, used by 3D or Compute engines
98 #define AMDGPU_GEM_DOMAIN_CPU 0x1
99 #define AMDGPU_GEM_DOMAIN_GTT 0x2
100 #define AMDGPU_GEM_DOMAIN_VRAM 0x4
101 #define AMDGPU_GEM_DOMAIN_GDS 0x8
102 #define AMDGPU_GEM_DOMAIN_GWS 0x10
103 #define AMDGPU_GEM_DOMAIN_OA 0x20
104 #define AMDGPU_GEM_DOMAIN_MASK (AMDGPU_GEM_DOMAIN_CPU | \
105 AMDGPU_GEM_DOMAIN_GTT | \
106 AMDGPU_GEM_DOMAIN_VRAM | \
107 AMDGPU_GEM_DOMAIN_GDS | \
108 AMDGPU_GEM_DOMAIN_GWS | \
109 AMDGPU_GEM_DOMAIN_OA)
111 /* Flag that CPU access will be required for the case of VRAM domain */
112 #define AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED (1 << 0)
113 /* Flag that CPU access will not work, this VRAM domain is invisible */
114 #define AMDGPU_GEM_CREATE_NO_CPU_ACCESS (1 << 1)
115 /* Flag that USWC attributes should be used for GTT */
116 #define AMDGPU_GEM_CREATE_CPU_GTT_USWC (1 << 2)
117 /* Flag that the memory should be in VRAM and cleared */
118 #define AMDGPU_GEM_CREATE_VRAM_CLEARED (1 << 3)
119 /* Flag that allocating the BO should use linear VRAM */
120 #define AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS (1 << 5)
121 /* Flag that BO is always valid in this VM */
122 #define AMDGPU_GEM_CREATE_VM_ALWAYS_VALID (1 << 6)
123 /* Flag that BO sharing will be explicitly synchronized */
124 #define AMDGPU_GEM_CREATE_EXPLICIT_SYNC (1 << 7)
125 /* Flag that indicates allocating MQD gart on GFX9, where the mtype
126 * for the second page onward should be set to NC. It should never
127 * be used by user space applications.
129 #define AMDGPU_GEM_CREATE_CP_MQD_GFX9 (1 << 8)
130 /* Flag that BO may contain sensitive data that must be wiped before
131 * releasing the memory
133 #define AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE (1 << 9)
134 /* Flag that BO will be encrypted and that the TMZ bit should be
135 * set in the PTEs when mapping this buffer via GPUVM or
136 * accessing it with various hw blocks
138 #define AMDGPU_GEM_CREATE_ENCRYPTED (1 << 10)
139 /* Flag that BO will be used only in preemptible context, which does
140 * not require GTT memory accounting
142 #define AMDGPU_GEM_CREATE_PREEMPTIBLE (1 << 11)
143 /* Flag that BO can be discarded under memory pressure without keeping the
146 #define AMDGPU_GEM_CREATE_DISCARDABLE (1 << 12)
147 /* Flag that BO is shared coherently between multiple devices or CPU threads.
148 * May depend on GPU instructions to flush caches explicitly
150 * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
151 * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
153 #define AMDGPU_GEM_CREATE_COHERENT (1 << 13)
154 /* Flag that BO should not be cached by GPU. Coherent without having to flush
155 * GPU caches explicitly
157 * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
158 * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
160 #define AMDGPU_GEM_CREATE_UNCACHED (1 << 14)
162 struct drm_amdgpu_gem_create_in {
163 /** the requested memory size */
165 /** physical start_addr alignment in bytes for some HW requirements */
167 /** the requested memory domains */
169 /** allocation flags */
173 struct drm_amdgpu_gem_create_out {
174 /** returned GEM object handle */
179 union drm_amdgpu_gem_create {
180 struct drm_amdgpu_gem_create_in in;
181 struct drm_amdgpu_gem_create_out out;
184 /** Opcode to create new residency list. */
185 #define AMDGPU_BO_LIST_OP_CREATE 0
186 /** Opcode to destroy previously created residency list */
187 #define AMDGPU_BO_LIST_OP_DESTROY 1
188 /** Opcode to update resource information in the list */
189 #define AMDGPU_BO_LIST_OP_UPDATE 2
191 struct drm_amdgpu_bo_list_in {
192 /** Type of operation */
194 /** Handle of list or 0 if we want to create one */
196 /** Number of BOs in list */
198 /** Size of each element describing BO */
200 /** Pointer to array describing BOs */
204 struct drm_amdgpu_bo_list_entry {
207 /** New (if specified) BO priority to be used during migration */
211 struct drm_amdgpu_bo_list_out {
212 /** Handle of resource list */
217 union drm_amdgpu_bo_list {
218 struct drm_amdgpu_bo_list_in in;
219 struct drm_amdgpu_bo_list_out out;
222 /* context related */
223 #define AMDGPU_CTX_OP_ALLOC_CTX 1
224 #define AMDGPU_CTX_OP_FREE_CTX 2
225 #define AMDGPU_CTX_OP_QUERY_STATE 3
226 #define AMDGPU_CTX_OP_QUERY_STATE2 4
227 #define AMDGPU_CTX_OP_GET_STABLE_PSTATE 5
228 #define AMDGPU_CTX_OP_SET_STABLE_PSTATE 6
230 /* GPU reset status */
231 #define AMDGPU_CTX_NO_RESET 0
232 /* this the context caused it */
233 #define AMDGPU_CTX_GUILTY_RESET 1
234 /* some other context caused it */
235 #define AMDGPU_CTX_INNOCENT_RESET 2
237 #define AMDGPU_CTX_UNKNOWN_RESET 3
239 /* indicate gpu reset occured after ctx created */
240 #define AMDGPU_CTX_QUERY2_FLAGS_RESET (1<<0)
241 /* indicate vram lost occured after ctx created */
242 #define AMDGPU_CTX_QUERY2_FLAGS_VRAMLOST (1<<1)
243 /* indicate some job from this context once cause gpu hang */
244 #define AMDGPU_CTX_QUERY2_FLAGS_GUILTY (1<<2)
245 /* indicate some errors are detected by RAS */
246 #define AMDGPU_CTX_QUERY2_FLAGS_RAS_CE (1<<3)
247 #define AMDGPU_CTX_QUERY2_FLAGS_RAS_UE (1<<4)
249 /* Context priority level */
250 #define AMDGPU_CTX_PRIORITY_UNSET -2048
251 #define AMDGPU_CTX_PRIORITY_VERY_LOW -1023
252 #define AMDGPU_CTX_PRIORITY_LOW -512
253 #define AMDGPU_CTX_PRIORITY_NORMAL 0
255 * When used in struct drm_amdgpu_ctx_in, a priority above NORMAL requires
256 * CAP_SYS_NICE or DRM_MASTER
258 #define AMDGPU_CTX_PRIORITY_HIGH 512
259 #define AMDGPU_CTX_PRIORITY_VERY_HIGH 1023
261 /* select a stable profiling pstate for perfmon tools */
262 #define AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK 0xf
263 #define AMDGPU_CTX_STABLE_PSTATE_NONE 0
264 #define AMDGPU_CTX_STABLE_PSTATE_STANDARD 1
265 #define AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK 2
266 #define AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK 3
267 #define AMDGPU_CTX_STABLE_PSTATE_PEAK 4
269 struct drm_amdgpu_ctx_in {
270 /** AMDGPU_CTX_OP_* */
275 /** AMDGPU_CTX_PRIORITY_* */
279 union drm_amdgpu_ctx_out {
286 /** For future use, no flags defined so far */
288 /** Number of resets caused by this context so far. */
290 /** Reset status since the last call of the ioctl. */
300 union drm_amdgpu_ctx {
301 struct drm_amdgpu_ctx_in in;
302 union drm_amdgpu_ctx_out out;
306 #define AMDGPU_VM_OP_RESERVE_VMID 1
307 #define AMDGPU_VM_OP_UNRESERVE_VMID 2
309 struct drm_amdgpu_vm_in {
310 /** AMDGPU_VM_OP_* */
315 struct drm_amdgpu_vm_out {
316 /** For future use, no flags defined so far */
320 union drm_amdgpu_vm {
321 struct drm_amdgpu_vm_in in;
322 struct drm_amdgpu_vm_out out;
326 #define AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE 1
327 #define AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE 2
329 struct drm_amdgpu_sched_in {
330 /* AMDGPU_SCHED_OP_* */
333 /** AMDGPU_CTX_PRIORITY_* */
338 union drm_amdgpu_sched {
339 struct drm_amdgpu_sched_in in;
343 * This is not a reliable API and you should expect it to fail for any
344 * number of reasons and have fallback path that do not use userptr to
345 * perform any operation.
347 #define AMDGPU_GEM_USERPTR_READONLY (1 << 0)
348 #define AMDGPU_GEM_USERPTR_ANONONLY (1 << 1)
349 #define AMDGPU_GEM_USERPTR_VALIDATE (1 << 2)
350 #define AMDGPU_GEM_USERPTR_REGISTER (1 << 3)
352 struct drm_amdgpu_gem_userptr {
355 /* AMDGPU_GEM_USERPTR_* */
357 /* Resulting GEM handle */
362 /* same meaning as the GB_TILE_MODE and GL_MACRO_TILE_MODE fields */
363 #define AMDGPU_TILING_ARRAY_MODE_SHIFT 0
364 #define AMDGPU_TILING_ARRAY_MODE_MASK 0xf
365 #define AMDGPU_TILING_PIPE_CONFIG_SHIFT 4
366 #define AMDGPU_TILING_PIPE_CONFIG_MASK 0x1f
367 #define AMDGPU_TILING_TILE_SPLIT_SHIFT 9
368 #define AMDGPU_TILING_TILE_SPLIT_MASK 0x7
369 #define AMDGPU_TILING_MICRO_TILE_MODE_SHIFT 12
370 #define AMDGPU_TILING_MICRO_TILE_MODE_MASK 0x7
371 #define AMDGPU_TILING_BANK_WIDTH_SHIFT 15
372 #define AMDGPU_TILING_BANK_WIDTH_MASK 0x3
373 #define AMDGPU_TILING_BANK_HEIGHT_SHIFT 17
374 #define AMDGPU_TILING_BANK_HEIGHT_MASK 0x3
375 #define AMDGPU_TILING_MACRO_TILE_ASPECT_SHIFT 19
376 #define AMDGPU_TILING_MACRO_TILE_ASPECT_MASK 0x3
377 #define AMDGPU_TILING_NUM_BANKS_SHIFT 21
378 #define AMDGPU_TILING_NUM_BANKS_MASK 0x3
380 /* GFX9 and later: */
381 #define AMDGPU_TILING_SWIZZLE_MODE_SHIFT 0
382 #define AMDGPU_TILING_SWIZZLE_MODE_MASK 0x1f
383 #define AMDGPU_TILING_DCC_OFFSET_256B_SHIFT 5
384 #define AMDGPU_TILING_DCC_OFFSET_256B_MASK 0xFFFFFF
385 #define AMDGPU_TILING_DCC_PITCH_MAX_SHIFT 29
386 #define AMDGPU_TILING_DCC_PITCH_MAX_MASK 0x3FFF
387 #define AMDGPU_TILING_DCC_INDEPENDENT_64B_SHIFT 43
388 #define AMDGPU_TILING_DCC_INDEPENDENT_64B_MASK 0x1
389 #define AMDGPU_TILING_DCC_INDEPENDENT_128B_SHIFT 44
390 #define AMDGPU_TILING_DCC_INDEPENDENT_128B_MASK 0x1
391 #define AMDGPU_TILING_SCANOUT_SHIFT 63
392 #define AMDGPU_TILING_SCANOUT_MASK 0x1
394 /* Set/Get helpers for tiling flags. */
395 #define AMDGPU_TILING_SET(field, value) \
396 (((__u64)(value) & AMDGPU_TILING_##field##_MASK) << AMDGPU_TILING_##field##_SHIFT)
397 #define AMDGPU_TILING_GET(value, field) \
398 (((__u64)(value) >> AMDGPU_TILING_##field##_SHIFT) & AMDGPU_TILING_##field##_MASK)
400 #define AMDGPU_GEM_METADATA_OP_SET_METADATA 1
401 #define AMDGPU_GEM_METADATA_OP_GET_METADATA 2
403 /** The same structure is shared for input/output */
404 struct drm_amdgpu_gem_metadata {
405 /** GEM Object handle */
407 /** Do we want get or set metadata */
410 /** For future use, no flags defined so far */
412 /** family specific tiling info */
414 __u32 data_size_bytes;
419 struct drm_amdgpu_gem_mmap_in {
420 /** the GEM object handle */
425 struct drm_amdgpu_gem_mmap_out {
426 /** mmap offset from the vma offset manager */
430 union drm_amdgpu_gem_mmap {
431 struct drm_amdgpu_gem_mmap_in in;
432 struct drm_amdgpu_gem_mmap_out out;
435 struct drm_amdgpu_gem_wait_idle_in {
436 /** GEM object handle */
438 /** For future use, no flags defined so far */
440 /** Absolute timeout to wait */
444 struct drm_amdgpu_gem_wait_idle_out {
445 /** BO status: 0 - BO is idle, 1 - BO is busy */
447 /** Returned current memory domain */
451 union drm_amdgpu_gem_wait_idle {
452 struct drm_amdgpu_gem_wait_idle_in in;
453 struct drm_amdgpu_gem_wait_idle_out out;
456 struct drm_amdgpu_wait_cs_in {
457 /* Command submission handle
458 * handle equals 0 means none to wait for
459 * handle equals ~0ull means wait for the latest sequence number
462 /** Absolute timeout to wait */
470 struct drm_amdgpu_wait_cs_out {
471 /** CS status: 0 - CS completed, 1 - CS still busy */
475 union drm_amdgpu_wait_cs {
476 struct drm_amdgpu_wait_cs_in in;
477 struct drm_amdgpu_wait_cs_out out;
480 struct drm_amdgpu_fence {
488 struct drm_amdgpu_wait_fences_in {
489 /** This points to uint64_t * which points to fences */
496 struct drm_amdgpu_wait_fences_out {
498 __u32 first_signaled;
501 union drm_amdgpu_wait_fences {
502 struct drm_amdgpu_wait_fences_in in;
503 struct drm_amdgpu_wait_fences_out out;
506 #define AMDGPU_GEM_OP_GET_GEM_CREATE_INFO 0
507 #define AMDGPU_GEM_OP_SET_PLACEMENT 1
509 /* Sets or returns a value associated with a buffer. */
510 struct drm_amdgpu_gem_op {
511 /** GEM object handle */
513 /** AMDGPU_GEM_OP_* */
515 /** Input or return value */
519 #define AMDGPU_VA_OP_MAP 1
520 #define AMDGPU_VA_OP_UNMAP 2
521 #define AMDGPU_VA_OP_CLEAR 3
522 #define AMDGPU_VA_OP_REPLACE 4
524 /* Delay the page table update till the next CS */
525 #define AMDGPU_VM_DELAY_UPDATE (1 << 0)
528 /* readable mapping */
529 #define AMDGPU_VM_PAGE_READABLE (1 << 1)
530 /* writable mapping */
531 #define AMDGPU_VM_PAGE_WRITEABLE (1 << 2)
532 /* executable mapping, new for VI */
533 #define AMDGPU_VM_PAGE_EXECUTABLE (1 << 3)
534 /* partially resident texture */
535 #define AMDGPU_VM_PAGE_PRT (1 << 4)
536 /* MTYPE flags use bit 5 to 8 */
537 #define AMDGPU_VM_MTYPE_MASK (0xf << 5)
538 /* Default MTYPE. Pre-AI must use this. Recommended for newer ASICs. */
539 #define AMDGPU_VM_MTYPE_DEFAULT (0 << 5)
540 /* Use Non Coherent MTYPE instead of default MTYPE */
541 #define AMDGPU_VM_MTYPE_NC (1 << 5)
542 /* Use Write Combine MTYPE instead of default MTYPE */
543 #define AMDGPU_VM_MTYPE_WC (2 << 5)
544 /* Use Cache Coherent MTYPE instead of default MTYPE */
545 #define AMDGPU_VM_MTYPE_CC (3 << 5)
546 /* Use UnCached MTYPE instead of default MTYPE */
547 #define AMDGPU_VM_MTYPE_UC (4 << 5)
548 /* Use Read Write MTYPE instead of default MTYPE */
549 #define AMDGPU_VM_MTYPE_RW (5 << 5)
550 /* don't allocate MALL */
551 #define AMDGPU_VM_PAGE_NOALLOC (1 << 9)
553 struct drm_amdgpu_gem_va {
554 /** GEM object handle */
557 /** AMDGPU_VA_OP_* */
559 /** AMDGPU_VM_PAGE_* */
561 /** va address to assign . Must be correctly aligned.*/
563 /** Specify offset inside of BO to assign. Must be correctly aligned.*/
565 /** Specify mapping size. Must be correctly aligned. */
569 #define AMDGPU_HW_IP_GFX 0
570 #define AMDGPU_HW_IP_COMPUTE 1
571 #define AMDGPU_HW_IP_DMA 2
572 #define AMDGPU_HW_IP_UVD 3
573 #define AMDGPU_HW_IP_VCE 4
574 #define AMDGPU_HW_IP_UVD_ENC 5
575 #define AMDGPU_HW_IP_VCN_DEC 6
577 * From VCN4, AMDGPU_HW_IP_VCN_ENC is re-used to support
578 * both encoding and decoding jobs.
580 #define AMDGPU_HW_IP_VCN_ENC 7
581 #define AMDGPU_HW_IP_VCN_JPEG 8
582 #define AMDGPU_HW_IP_NUM 9
584 #define AMDGPU_HW_IP_INSTANCE_MAX_COUNT 1
586 #define AMDGPU_CHUNK_ID_IB 0x01
587 #define AMDGPU_CHUNK_ID_FENCE 0x02
588 #define AMDGPU_CHUNK_ID_DEPENDENCIES 0x03
589 #define AMDGPU_CHUNK_ID_SYNCOBJ_IN 0x04
590 #define AMDGPU_CHUNK_ID_SYNCOBJ_OUT 0x05
591 #define AMDGPU_CHUNK_ID_BO_HANDLES 0x06
592 #define AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES 0x07
593 #define AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT 0x08
594 #define AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL 0x09
596 struct drm_amdgpu_cs_chunk {
602 struct drm_amdgpu_cs_in {
603 /** Rendering context id */
605 /** Handle of resource list associated with CS */
606 __u32 bo_list_handle;
609 /** this points to __u64 * which point to cs chunks */
613 struct drm_amdgpu_cs_out {
617 union drm_amdgpu_cs {
618 struct drm_amdgpu_cs_in in;
619 struct drm_amdgpu_cs_out out;
622 /* Specify flags to be used for IB */
624 /* This IB should be submitted to CE */
625 #define AMDGPU_IB_FLAG_CE (1<<0)
627 /* Preamble flag, which means the IB could be dropped if no context switch */
628 #define AMDGPU_IB_FLAG_PREAMBLE (1<<1)
630 /* Preempt flag, IB should set Pre_enb bit if PREEMPT flag detected */
631 #define AMDGPU_IB_FLAG_PREEMPT (1<<2)
633 /* The IB fence should do the L2 writeback but not invalidate any shader
634 * caches (L2/vL1/sL1/I$). */
635 #define AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE (1 << 3)
637 /* Set GDS_COMPUTE_MAX_WAVE_ID = DEFAULT before PACKET3_INDIRECT_BUFFER.
638 * This will reset wave ID counters for the IB.
640 #define AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID (1 << 4)
642 /* Flag the IB as secure (TMZ)
644 #define AMDGPU_IB_FLAGS_SECURE (1 << 5)
646 /* Tell KMD to flush and invalidate caches
648 #define AMDGPU_IB_FLAG_EMIT_MEM_SYNC (1 << 6)
650 struct drm_amdgpu_cs_chunk_ib {
652 /** AMDGPU_IB_FLAG_* */
654 /** Virtual address to begin IB execution */
656 /** Size of submission */
658 /** HW IP to submit to */
660 /** HW IP index of the same type to submit to */
662 /** Ring index to submit to */
666 struct drm_amdgpu_cs_chunk_dep {
674 struct drm_amdgpu_cs_chunk_fence {
679 struct drm_amdgpu_cs_chunk_sem {
683 struct drm_amdgpu_cs_chunk_syncobj {
689 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ 0
690 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ_FD 1
691 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNC_FILE_FD 2
693 union drm_amdgpu_fence_to_handle {
695 struct drm_amdgpu_fence fence;
704 struct drm_amdgpu_cs_chunk_data {
706 struct drm_amdgpu_cs_chunk_ib ib_data;
707 struct drm_amdgpu_cs_chunk_fence fence_data;
712 * Query h/w info: Flag that this is integrated (a.h.a. fusion) GPU
715 #define AMDGPU_IDS_FLAGS_FUSION 0x1
716 #define AMDGPU_IDS_FLAGS_PREEMPTION 0x2
717 #define AMDGPU_IDS_FLAGS_TMZ 0x4
719 /* indicate if acceleration can be working */
720 #define AMDGPU_INFO_ACCEL_WORKING 0x00
721 /* get the crtc_id from the mode object id? */
722 #define AMDGPU_INFO_CRTC_FROM_ID 0x01
723 /* query hw IP info */
724 #define AMDGPU_INFO_HW_IP_INFO 0x02
725 /* query hw IP instance count for the specified type */
726 #define AMDGPU_INFO_HW_IP_COUNT 0x03
727 /* timestamp for GL_ARB_timer_query */
728 #define AMDGPU_INFO_TIMESTAMP 0x05
729 /* Query the firmware version */
730 #define AMDGPU_INFO_FW_VERSION 0x0e
731 /* Subquery id: Query VCE firmware version */
732 #define AMDGPU_INFO_FW_VCE 0x1
733 /* Subquery id: Query UVD firmware version */
734 #define AMDGPU_INFO_FW_UVD 0x2
735 /* Subquery id: Query GMC firmware version */
736 #define AMDGPU_INFO_FW_GMC 0x03
737 /* Subquery id: Query GFX ME firmware version */
738 #define AMDGPU_INFO_FW_GFX_ME 0x04
739 /* Subquery id: Query GFX PFP firmware version */
740 #define AMDGPU_INFO_FW_GFX_PFP 0x05
741 /* Subquery id: Query GFX CE firmware version */
742 #define AMDGPU_INFO_FW_GFX_CE 0x06
743 /* Subquery id: Query GFX RLC firmware version */
744 #define AMDGPU_INFO_FW_GFX_RLC 0x07
745 /* Subquery id: Query GFX MEC firmware version */
746 #define AMDGPU_INFO_FW_GFX_MEC 0x08
747 /* Subquery id: Query SMC firmware version */
748 #define AMDGPU_INFO_FW_SMC 0x0a
749 /* Subquery id: Query SDMA firmware version */
750 #define AMDGPU_INFO_FW_SDMA 0x0b
751 /* Subquery id: Query PSP SOS firmware version */
752 #define AMDGPU_INFO_FW_SOS 0x0c
753 /* Subquery id: Query PSP ASD firmware version */
754 #define AMDGPU_INFO_FW_ASD 0x0d
755 /* Subquery id: Query VCN firmware version */
756 #define AMDGPU_INFO_FW_VCN 0x0e
757 /* Subquery id: Query GFX RLC SRLC firmware version */
758 #define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL 0x0f
759 /* Subquery id: Query GFX RLC SRLG firmware version */
760 #define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM 0x10
761 /* Subquery id: Query GFX RLC SRLS firmware version */
762 #define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM 0x11
763 /* Subquery id: Query DMCU firmware version */
764 #define AMDGPU_INFO_FW_DMCU 0x12
765 #define AMDGPU_INFO_FW_TA 0x13
766 /* Subquery id: Query DMCUB firmware version */
767 #define AMDGPU_INFO_FW_DMCUB 0x14
768 /* Subquery id: Query TOC firmware version */
769 #define AMDGPU_INFO_FW_TOC 0x15
770 /* Subquery id: Query CAP firmware version */
771 #define AMDGPU_INFO_FW_CAP 0x16
772 /* Subquery id: Query GFX RLCP firmware version */
773 #define AMDGPU_INFO_FW_GFX_RLCP 0x17
774 /* Subquery id: Query GFX RLCV firmware version */
775 #define AMDGPU_INFO_FW_GFX_RLCV 0x18
776 /* Subquery id: Query MES_KIQ firmware version */
777 #define AMDGPU_INFO_FW_MES_KIQ 0x19
778 /* Subquery id: Query MES firmware version */
779 #define AMDGPU_INFO_FW_MES 0x1a
780 /* Subquery id: Query IMU firmware version */
781 #define AMDGPU_INFO_FW_IMU 0x1b
783 /* number of bytes moved for TTM migration */
784 #define AMDGPU_INFO_NUM_BYTES_MOVED 0x0f
785 /* the used VRAM size */
786 #define AMDGPU_INFO_VRAM_USAGE 0x10
787 /* the used GTT size */
788 #define AMDGPU_INFO_GTT_USAGE 0x11
789 /* Information about GDS, etc. resource configuration */
790 #define AMDGPU_INFO_GDS_CONFIG 0x13
791 /* Query information about VRAM and GTT domains */
792 #define AMDGPU_INFO_VRAM_GTT 0x14
793 /* Query information about register in MMR address space*/
794 #define AMDGPU_INFO_READ_MMR_REG 0x15
795 /* Query information about device: rev id, family, etc. */
796 #define AMDGPU_INFO_DEV_INFO 0x16
797 /* visible vram usage */
798 #define AMDGPU_INFO_VIS_VRAM_USAGE 0x17
799 /* number of TTM buffer evictions */
800 #define AMDGPU_INFO_NUM_EVICTIONS 0x18
801 /* Query memory about VRAM and GTT domains */
802 #define AMDGPU_INFO_MEMORY 0x19
803 /* Query vce clock table */
804 #define AMDGPU_INFO_VCE_CLOCK_TABLE 0x1A
805 /* Query vbios related information */
806 #define AMDGPU_INFO_VBIOS 0x1B
807 /* Subquery id: Query vbios size */
808 #define AMDGPU_INFO_VBIOS_SIZE 0x1
809 /* Subquery id: Query vbios image */
810 #define AMDGPU_INFO_VBIOS_IMAGE 0x2
811 /* Subquery id: Query vbios info */
812 #define AMDGPU_INFO_VBIOS_INFO 0x3
813 /* Query UVD handles */
814 #define AMDGPU_INFO_NUM_HANDLES 0x1C
815 /* Query sensor related information */
816 #define AMDGPU_INFO_SENSOR 0x1D
817 /* Subquery id: Query GPU shader clock */
818 #define AMDGPU_INFO_SENSOR_GFX_SCLK 0x1
819 /* Subquery id: Query GPU memory clock */
820 #define AMDGPU_INFO_SENSOR_GFX_MCLK 0x2
821 /* Subquery id: Query GPU temperature */
822 #define AMDGPU_INFO_SENSOR_GPU_TEMP 0x3
823 /* Subquery id: Query GPU load */
824 #define AMDGPU_INFO_SENSOR_GPU_LOAD 0x4
825 /* Subquery id: Query average GPU power */
826 #define AMDGPU_INFO_SENSOR_GPU_AVG_POWER 0x5
827 /* Subquery id: Query northbridge voltage */
828 #define AMDGPU_INFO_SENSOR_VDDNB 0x6
829 /* Subquery id: Query graphics voltage */
830 #define AMDGPU_INFO_SENSOR_VDDGFX 0x7
831 /* Subquery id: Query GPU stable pstate shader clock */
832 #define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_SCLK 0x8
833 /* Subquery id: Query GPU stable pstate memory clock */
834 #define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_MCLK 0x9
835 /* Number of VRAM page faults on CPU access. */
836 #define AMDGPU_INFO_NUM_VRAM_CPU_PAGE_FAULTS 0x1E
837 #define AMDGPU_INFO_VRAM_LOST_COUNTER 0x1F
838 /* query ras mask of enabled features*/
839 #define AMDGPU_INFO_RAS_ENABLED_FEATURES 0x20
840 /* RAS MASK: UMC (VRAM) */
841 #define AMDGPU_INFO_RAS_ENABLED_UMC (1 << 0)
843 #define AMDGPU_INFO_RAS_ENABLED_SDMA (1 << 1)
845 #define AMDGPU_INFO_RAS_ENABLED_GFX (1 << 2)
846 /* RAS MASK: MMHUB */
847 #define AMDGPU_INFO_RAS_ENABLED_MMHUB (1 << 3)
848 /* RAS MASK: ATHUB */
849 #define AMDGPU_INFO_RAS_ENABLED_ATHUB (1 << 4)
851 #define AMDGPU_INFO_RAS_ENABLED_PCIE (1 << 5)
853 #define AMDGPU_INFO_RAS_ENABLED_HDP (1 << 6)
855 #define AMDGPU_INFO_RAS_ENABLED_XGMI (1 << 7)
857 #define AMDGPU_INFO_RAS_ENABLED_DF (1 << 8)
859 #define AMDGPU_INFO_RAS_ENABLED_SMN (1 << 9)
861 #define AMDGPU_INFO_RAS_ENABLED_SEM (1 << 10)
863 #define AMDGPU_INFO_RAS_ENABLED_MP0 (1 << 11)
865 #define AMDGPU_INFO_RAS_ENABLED_MP1 (1 << 12)
867 #define AMDGPU_INFO_RAS_ENABLED_FUSE (1 << 13)
868 /* query video encode/decode caps */
869 #define AMDGPU_INFO_VIDEO_CAPS 0x21
870 /* Subquery id: Decode */
871 #define AMDGPU_INFO_VIDEO_CAPS_DECODE 0
872 /* Subquery id: Encode */
873 #define AMDGPU_INFO_VIDEO_CAPS_ENCODE 1
875 #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT 0
876 #define AMDGPU_INFO_MMR_SE_INDEX_MASK 0xff
877 #define AMDGPU_INFO_MMR_SH_INDEX_SHIFT 8
878 #define AMDGPU_INFO_MMR_SH_INDEX_MASK 0xff
880 struct drm_amdgpu_query_fw {
881 /** AMDGPU_INFO_FW_* */
884 * Index of the IP if there are more IPs of
889 * Index of the engine. Whether this is used depends
890 * on the firmware type. (e.g. MEC, SDMA)
896 /* Input structure for the INFO ioctl */
897 struct drm_amdgpu_info {
898 /* Where the return value will be stored */
899 __u64 return_pointer;
900 /* The size of the return value. Just like "size" in "snprintf",
901 * it limits how many bytes the kernel can write. */
903 /* The query request id. */
913 /** AMDGPU_HW_IP_* */
916 * Index of the IP if there are more IPs of the same
917 * type. Ignored by AMDGPU_INFO_HW_IP_COUNT.
924 /** number of registers to read */
927 /** For future use, no flags defined so far */
931 struct drm_amdgpu_query_fw query_fw;
948 struct drm_amdgpu_info_gds {
949 /** GDS GFX partition size */
950 __u32 gds_gfx_partition_size;
951 /** GDS compute partition size */
952 __u32 compute_partition_size;
953 /** total GDS memory size */
954 __u32 gds_total_size;
955 /** GWS size per GFX partition */
956 __u32 gws_per_gfx_partition;
957 /** GSW size per compute partition */
958 __u32 gws_per_compute_partition;
959 /** OA size per GFX partition */
960 __u32 oa_per_gfx_partition;
961 /** OA size per compute partition */
962 __u32 oa_per_compute_partition;
966 struct drm_amdgpu_info_vram_gtt {
968 __u64 vram_cpu_accessible_size;
972 struct drm_amdgpu_heap_info {
973 /** max. physical memory */
974 __u64 total_heap_size;
976 /** Theoretical max. available memory in the given heap */
977 __u64 usable_heap_size;
980 * Number of bytes allocated in the heap. This includes all processes
981 * and private allocations in the kernel. It changes when new buffers
982 * are allocated, freed, and moved. It cannot be larger than
988 * Theoretical possible max. size of buffer which
989 * could be allocated in the given heap
991 __u64 max_allocation;
994 struct drm_amdgpu_memory_info {
995 struct drm_amdgpu_heap_info vram;
996 struct drm_amdgpu_heap_info cpu_accessible_vram;
997 struct drm_amdgpu_heap_info gtt;
1000 struct drm_amdgpu_info_firmware {
1005 struct drm_amdgpu_info_vbios {
1010 __u8 vbios_ver_str[32];
1014 #define AMDGPU_VRAM_TYPE_UNKNOWN 0
1015 #define AMDGPU_VRAM_TYPE_GDDR1 1
1016 #define AMDGPU_VRAM_TYPE_DDR2 2
1017 #define AMDGPU_VRAM_TYPE_GDDR3 3
1018 #define AMDGPU_VRAM_TYPE_GDDR4 4
1019 #define AMDGPU_VRAM_TYPE_GDDR5 5
1020 #define AMDGPU_VRAM_TYPE_HBM 6
1021 #define AMDGPU_VRAM_TYPE_DDR3 7
1022 #define AMDGPU_VRAM_TYPE_DDR4 8
1023 #define AMDGPU_VRAM_TYPE_GDDR6 9
1024 #define AMDGPU_VRAM_TYPE_DDR5 10
1025 #define AMDGPU_VRAM_TYPE_LPDDR4 11
1026 #define AMDGPU_VRAM_TYPE_LPDDR5 12
1028 struct drm_amdgpu_info_device {
1029 /** PCI Device ID */
1031 /** Internal chip revision: A0, A1, etc.) */
1034 /** Revision id in PCI Config space */
1037 __u32 num_shader_engines;
1038 __u32 num_shader_arrays_per_engine;
1040 __u32 gpu_counter_freq;
1041 __u64 max_engine_clock;
1042 __u64 max_memory_clock;
1043 /* cu information */
1044 __u32 cu_active_number;
1045 /* NOTE: cu_ao_mask is INVALID, DON'T use it */
1047 __u32 cu_bitmap[4][4];
1048 /** Render backend pipe mask. One render backend is CB+DB. */
1049 __u32 enabled_rb_pipes_mask;
1051 __u32 num_hw_gfx_contexts;
1054 /** Starting virtual address for UMDs. */
1055 __u64 virtual_address_offset;
1056 /** The maximum virtual address */
1057 __u64 virtual_address_max;
1058 /** Required alignment of virtual addresses. */
1059 __u32 virtual_address_alignment;
1060 /** Page table entry - fragment size */
1061 __u32 pte_fragment_size;
1062 __u32 gart_page_size;
1063 /** constant engine ram size*/
1065 /** video memory type info*/
1067 /** video memory bit width*/
1068 __u32 vram_bit_width;
1069 /* vce harvesting instance */
1070 __u32 vce_harvest_config;
1071 /* gfx double offchip LDS buffers */
1072 __u32 gc_double_offchip_lds_buf;
1073 /* NGG Primitive Buffer */
1074 __u64 prim_buf_gpu_addr;
1075 /* NGG Position Buffer */
1076 __u64 pos_buf_gpu_addr;
1077 /* NGG Control Sideband */
1078 __u64 cntl_sb_buf_gpu_addr;
1079 /* NGG Parameter Cache */
1080 __u64 param_buf_gpu_addr;
1081 __u32 prim_buf_size;
1083 __u32 cntl_sb_buf_size;
1084 __u32 param_buf_size;
1086 __u32 wave_front_size;
1087 /* shader visible vgprs*/
1088 __u32 num_shader_visible_vgprs;
1089 /* CU per shader array*/
1090 __u32 num_cu_per_sh;
1091 /* number of tcc blocks*/
1092 __u32 num_tcc_blocks;
1093 /* gs vgt table depth*/
1094 __u32 gs_vgt_table_depth;
1095 /* gs primitive buffer depth*/
1096 __u32 gs_prim_buffer_depth;
1097 /* max gs wavefront per vgt*/
1098 __u32 max_gs_waves_per_vgt;
1100 /* always on cu bitmap */
1101 __u32 cu_ao_bitmap[4][4];
1102 /** Starting high virtual address for UMDs. */
1103 __u64 high_va_offset;
1104 /** The maximum high virtual address */
1106 /* gfx10 pa_sc_tile_steering_override */
1107 __u32 pa_sc_tile_steering_override;
1109 __u64 tcc_disabled_mask;
1112 struct drm_amdgpu_info_hw_ip {
1113 /** Version of h/w IP */
1114 __u32 hw_ip_version_major;
1115 __u32 hw_ip_version_minor;
1117 __u64 capabilities_flags;
1118 /** command buffer address start alignment*/
1119 __u32 ib_start_alignment;
1120 /** command buffer size alignment*/
1121 __u32 ib_size_alignment;
1122 /** Bitmask of available rings. Bit 0 means ring 0, etc. */
1123 __u32 available_rings;
1124 /** version info: bits 23:16 major, 15:8 minor, 7:0 revision */
1125 __u32 ip_discovery_version;
1128 struct drm_amdgpu_info_num_handles {
1129 /** Max handles as supported by firmware for UVD */
1130 __u32 uvd_max_handles;
1131 /** Handles currently in use for UVD */
1132 __u32 uvd_used_handles;
1135 #define AMDGPU_VCE_CLOCK_TABLE_ENTRIES 6
1137 struct drm_amdgpu_info_vce_clock_table_entry {
1147 struct drm_amdgpu_info_vce_clock_table {
1148 struct drm_amdgpu_info_vce_clock_table_entry entries[AMDGPU_VCE_CLOCK_TABLE_ENTRIES];
1149 __u32 num_valid_entries;
1153 /* query video encode/decode caps */
1154 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2 0
1155 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4 1
1156 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1 2
1157 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC 3
1158 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC 4
1159 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG 5
1160 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9 6
1161 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1 7
1162 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT 8
1164 struct drm_amdgpu_info_video_codec_info {
1168 __u32 max_pixels_per_frame;
1173 struct drm_amdgpu_info_video_caps {
1174 struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
1178 * Supported GPU families
1180 #define AMDGPU_FAMILY_UNKNOWN 0
1181 #define AMDGPU_FAMILY_SI 110 /* Hainan, Oland, Verde, Pitcairn, Tahiti */
1182 #define AMDGPU_FAMILY_CI 120 /* Bonaire, Hawaii */
1183 #define AMDGPU_FAMILY_KV 125 /* Kaveri, Kabini, Mullins */
1184 #define AMDGPU_FAMILY_VI 130 /* Iceland, Tonga */
1185 #define AMDGPU_FAMILY_CZ 135 /* Carrizo, Stoney */
1186 #define AMDGPU_FAMILY_AI 141 /* Vega10 */
1187 #define AMDGPU_FAMILY_RV 142 /* Raven */
1188 #define AMDGPU_FAMILY_NV 143 /* Navi10 */
1189 #define AMDGPU_FAMILY_VGH 144 /* Van Gogh */
1190 #define AMDGPU_FAMILY_GC_11_0_0 145 /* GC 11.0.0 */
1191 #define AMDGPU_FAMILY_YC 146 /* Yellow Carp */
1192 #define AMDGPU_FAMILY_GC_11_0_1 148 /* GC 11.0.1 */
1193 #define AMDGPU_FAMILY_GC_10_3_6 149 /* GC 10.3.6 */
1194 #define AMDGPU_FAMILY_GC_10_3_7 151 /* GC 10.3.7 */
1196 #if defined(__cplusplus)