xref: /linux/include/uapi/drm/amdgpu_drm.h (revision 9fd2da71c301184d98fe37674ca8d017d1ce6600)
1 /* amdgpu_drm.h -- Public header for the amdgpu driver -*- linux-c -*-
2  *
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.
7  *
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:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
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.
25  *
26  * Authors:
27  *    Kevin E. Martin <martin@valinux.com>
28  *    Gareth Hughes <gareth@valinux.com>
29  *    Keith Whitwell <keith@tungstengraphics.com>
30  */
31 
32 #ifndef __AMDGPU_DRM_H__
33 #define __AMDGPU_DRM_H__
34 
35 #include "drm.h"
36 
37 #if defined(__cplusplus)
38 extern "C" {
39 #endif
40 
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
57 #define DRM_AMDGPU_USERQ		0x16
58 #define DRM_AMDGPU_USERQ_SIGNAL		0x17
59 #define DRM_AMDGPU_USERQ_WAIT		0x18
60 #define DRM_AMDGPU_GEM_LIST_HANDLES	0x19
61 
62 #define DRM_IOCTL_AMDGPU_GEM_CREATE	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_CREATE, union drm_amdgpu_gem_create)
63 #define DRM_IOCTL_AMDGPU_GEM_MMAP	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_MMAP, union drm_amdgpu_gem_mmap)
64 #define DRM_IOCTL_AMDGPU_CTX		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_CTX, union drm_amdgpu_ctx)
65 #define DRM_IOCTL_AMDGPU_BO_LIST	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_BO_LIST, union drm_amdgpu_bo_list)
66 #define DRM_IOCTL_AMDGPU_CS		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_CS, union drm_amdgpu_cs)
67 #define DRM_IOCTL_AMDGPU_INFO		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_INFO, struct drm_amdgpu_info)
68 #define DRM_IOCTL_AMDGPU_GEM_METADATA	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_METADATA, struct drm_amdgpu_gem_metadata)
69 #define DRM_IOCTL_AMDGPU_GEM_WAIT_IDLE	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_WAIT_IDLE, union drm_amdgpu_gem_wait_idle)
70 #define DRM_IOCTL_AMDGPU_GEM_VA		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_VA, struct drm_amdgpu_gem_va)
71 #define DRM_IOCTL_AMDGPU_WAIT_CS	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_CS, union drm_amdgpu_wait_cs)
72 #define DRM_IOCTL_AMDGPU_GEM_OP		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_OP, struct drm_amdgpu_gem_op)
73 #define DRM_IOCTL_AMDGPU_GEM_USERPTR	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_USERPTR, struct drm_amdgpu_gem_userptr)
74 #define DRM_IOCTL_AMDGPU_WAIT_FENCES	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_FENCES, union drm_amdgpu_wait_fences)
75 #define DRM_IOCTL_AMDGPU_VM		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_VM, union drm_amdgpu_vm)
76 #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
77 #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
78 #define DRM_IOCTL_AMDGPU_USERQ		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_USERQ, union drm_amdgpu_userq)
79 #define DRM_IOCTL_AMDGPU_USERQ_SIGNAL	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_USERQ_SIGNAL, struct drm_amdgpu_userq_signal)
80 #define DRM_IOCTL_AMDGPU_USERQ_WAIT	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_USERQ_WAIT, struct drm_amdgpu_userq_wait)
81 #define DRM_IOCTL_AMDGPU_GEM_LIST_HANDLES DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_LIST_HANDLES, struct drm_amdgpu_gem_list_handles)
82 
83 /**
84  * DOC: memory domains
85  *
86  * %AMDGPU_GEM_DOMAIN_CPU	System memory that is not GPU accessible.
87  * Memory in this pool could be swapped out to disk if there is pressure.
88  *
89  * %AMDGPU_GEM_DOMAIN_GTT	GPU accessible system memory, mapped into the
90  * GPU's virtual address space via gart. Gart memory linearizes non-contiguous
91  * pages of system memory, allows GPU access system memory in a linearized
92  * fashion.
93  *
94  * %AMDGPU_GEM_DOMAIN_VRAM	Local video memory. For APUs, it is memory
95  * carved out by the BIOS.
96  *
97  * %AMDGPU_GEM_DOMAIN_GDS	Global on-chip data storage used to share data
98  * across shader threads.
99  *
100  * %AMDGPU_GEM_DOMAIN_GWS	Global wave sync, used to synchronize the
101  * execution of all the waves on a device.
102  *
103  * %AMDGPU_GEM_DOMAIN_OA	Ordered append, used by 3D or Compute engines
104  * for appending data.
105  *
106  * %AMDGPU_GEM_DOMAIN_DOORBELL	Doorbell. It is an MMIO region for
107  * signalling user mode queues.
108  */
109 #define AMDGPU_GEM_DOMAIN_CPU		0x1
110 #define AMDGPU_GEM_DOMAIN_GTT		0x2
111 #define AMDGPU_GEM_DOMAIN_VRAM		0x4
112 #define AMDGPU_GEM_DOMAIN_GDS		0x8
113 #define AMDGPU_GEM_DOMAIN_GWS		0x10
114 #define AMDGPU_GEM_DOMAIN_OA		0x20
115 #define AMDGPU_GEM_DOMAIN_DOORBELL	0x40
116 #define AMDGPU_GEM_DOMAIN_MASK		(AMDGPU_GEM_DOMAIN_CPU | \
117 					 AMDGPU_GEM_DOMAIN_GTT | \
118 					 AMDGPU_GEM_DOMAIN_VRAM | \
119 					 AMDGPU_GEM_DOMAIN_GDS | \
120 					 AMDGPU_GEM_DOMAIN_GWS | \
121 					 AMDGPU_GEM_DOMAIN_OA | \
122 					 AMDGPU_GEM_DOMAIN_DOORBELL)
123 
124 /* Flag that CPU access will be required for the case of VRAM domain */
125 #define AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED	(1 << 0)
126 /* Flag that CPU access will not work, this VRAM domain is invisible */
127 #define AMDGPU_GEM_CREATE_NO_CPU_ACCESS		(1 << 1)
128 /* Flag that USWC attributes should be used for GTT */
129 #define AMDGPU_GEM_CREATE_CPU_GTT_USWC		(1 << 2)
130 /* Flag that the memory should be in VRAM and cleared */
131 #define AMDGPU_GEM_CREATE_VRAM_CLEARED		(1 << 3)
132 /* Flag that allocating the BO should use linear VRAM */
133 #define AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS	(1 << 5)
134 /* Flag that BO is always valid in this VM */
135 #define AMDGPU_GEM_CREATE_VM_ALWAYS_VALID	(1 << 6)
136 /* Flag that BO sharing will be explicitly synchronized */
137 #define AMDGPU_GEM_CREATE_EXPLICIT_SYNC		(1 << 7)
138 /* Flag that indicates allocating MQD gart on GFX9, where the mtype
139  * for the second page onward should be set to NC. It should never
140  * be used by user space applications.
141  */
142 #define AMDGPU_GEM_CREATE_CP_MQD_GFX9		(1 << 8)
143 /* Flag that BO may contain sensitive data that must be wiped before
144  * releasing the memory
145  */
146 #define AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE	(1 << 9)
147 /* Flag that BO will be encrypted and that the TMZ bit should be
148  * set in the PTEs when mapping this buffer via GPUVM or
149  * accessing it with various hw blocks
150  */
151 #define AMDGPU_GEM_CREATE_ENCRYPTED		(1 << 10)
152 /* Flag that BO will be used only in preemptible context, which does
153  * not require GTT memory accounting
154  */
155 #define AMDGPU_GEM_CREATE_PREEMPTIBLE		(1 << 11)
156 /* Flag that BO can be discarded under memory pressure without keeping the
157  * content.
158  */
159 #define AMDGPU_GEM_CREATE_DISCARDABLE		(1 << 12)
160 /* Flag that BO is shared coherently between multiple devices or CPU threads.
161  * May depend on GPU instructions to flush caches to system scope explicitly.
162  *
163  * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
164  * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
165  */
166 #define AMDGPU_GEM_CREATE_COHERENT		(1 << 13)
167 /* Flag that BO should not be cached by GPU. Coherent without having to flush
168  * GPU caches explicitly
169  *
170  * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
171  * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
172  */
173 #define AMDGPU_GEM_CREATE_UNCACHED		(1 << 14)
174 /* Flag that BO should be coherent across devices when using device-level
175  * atomics. May depend on GPU instructions to flush caches to device scope
176  * explicitly, promoting them to system scope automatically.
177  *
178  * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
179  * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
180  */
181 #define AMDGPU_GEM_CREATE_EXT_COHERENT		(1 << 15)
182 /* Set PTE.D and recompress during GTT->VRAM moves according to TILING flags. */
183 #define AMDGPU_GEM_CREATE_GFX12_DCC		(1 << 16)
184 
185 struct drm_amdgpu_gem_create_in  {
186 	/** the requested memory size */
187 	__u64 bo_size;
188 	/** physical start_addr alignment in bytes for some HW requirements */
189 	__u64 alignment;
190 	/** the requested memory domains */
191 	__u64 domains;
192 	/** allocation flags */
193 	__u64 domain_flags;
194 };
195 
196 struct drm_amdgpu_gem_create_out  {
197 	/** returned GEM object handle */
198 	__u32 handle;
199 	__u32 _pad;
200 };
201 
202 union drm_amdgpu_gem_create {
203 	struct drm_amdgpu_gem_create_in		in;
204 	struct drm_amdgpu_gem_create_out	out;
205 };
206 
207 /** Opcode to create new residency list.  */
208 #define AMDGPU_BO_LIST_OP_CREATE	0
209 /** Opcode to destroy previously created residency list */
210 #define AMDGPU_BO_LIST_OP_DESTROY	1
211 /** Opcode to update resource information in the list */
212 #define AMDGPU_BO_LIST_OP_UPDATE	2
213 
214 struct drm_amdgpu_bo_list_in {
215 	/** Type of operation */
216 	__u32 operation;
217 	/** Handle of list or 0 if we want to create one */
218 	__u32 list_handle;
219 	/** Number of BOs in list  */
220 	__u32 bo_number;
221 	/** Size of each element describing BO */
222 	__u32 bo_info_size;
223 	/** Pointer to array describing BOs */
224 	__u64 bo_info_ptr;
225 };
226 
227 struct drm_amdgpu_bo_list_entry {
228 	/** Handle of BO */
229 	__u32 bo_handle;
230 	/** New (if specified) BO priority to be used during migration */
231 	__u32 bo_priority;
232 };
233 
234 struct drm_amdgpu_bo_list_out {
235 	/** Handle of resource list  */
236 	__u32 list_handle;
237 	__u32 _pad;
238 };
239 
240 union drm_amdgpu_bo_list {
241 	struct drm_amdgpu_bo_list_in in;
242 	struct drm_amdgpu_bo_list_out out;
243 };
244 
245 /* context related */
246 #define AMDGPU_CTX_OP_ALLOC_CTX	1
247 #define AMDGPU_CTX_OP_FREE_CTX	2
248 #define AMDGPU_CTX_OP_QUERY_STATE	3
249 #define AMDGPU_CTX_OP_QUERY_STATE2	4
250 #define AMDGPU_CTX_OP_GET_STABLE_PSTATE	5
251 #define AMDGPU_CTX_OP_SET_STABLE_PSTATE	6
252 
253 /* GPU reset status */
254 #define AMDGPU_CTX_NO_RESET		0
255 /* this the context caused it */
256 #define AMDGPU_CTX_GUILTY_RESET		1
257 /* some other context caused it */
258 #define AMDGPU_CTX_INNOCENT_RESET	2
259 /* unknown cause */
260 #define AMDGPU_CTX_UNKNOWN_RESET	3
261 
262 /* indicate gpu reset occurred after ctx created */
263 #define AMDGPU_CTX_QUERY2_FLAGS_RESET    (1<<0)
264 /* indicate vram lost occurred after ctx created */
265 #define AMDGPU_CTX_QUERY2_FLAGS_VRAMLOST (1<<1)
266 /* indicate some job from this context once cause gpu hang */
267 #define AMDGPU_CTX_QUERY2_FLAGS_GUILTY   (1<<2)
268 /* indicate some errors are detected by RAS */
269 #define AMDGPU_CTX_QUERY2_FLAGS_RAS_CE   (1<<3)
270 #define AMDGPU_CTX_QUERY2_FLAGS_RAS_UE   (1<<4)
271 /* indicate that the reset hasn't completed yet */
272 #define AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS (1<<5)
273 
274 /* Context priority level */
275 #define AMDGPU_CTX_PRIORITY_UNSET       -2048
276 #define AMDGPU_CTX_PRIORITY_VERY_LOW    -1023
277 #define AMDGPU_CTX_PRIORITY_LOW         -512
278 #define AMDGPU_CTX_PRIORITY_NORMAL      0
279 /*
280  * When used in struct drm_amdgpu_ctx_in, a priority above NORMAL requires
281  * CAP_SYS_NICE or DRM_MASTER
282 */
283 #define AMDGPU_CTX_PRIORITY_HIGH        512
284 #define AMDGPU_CTX_PRIORITY_VERY_HIGH   1023
285 
286 /* select a stable profiling pstate for perfmon tools */
287 #define AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK  0xf
288 #define AMDGPU_CTX_STABLE_PSTATE_NONE  0
289 #define AMDGPU_CTX_STABLE_PSTATE_STANDARD  1
290 #define AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK  2
291 #define AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK  3
292 #define AMDGPU_CTX_STABLE_PSTATE_PEAK  4
293 
294 struct drm_amdgpu_ctx_in {
295 	/** AMDGPU_CTX_OP_* */
296 	__u32	op;
297 	/** Flags */
298 	__u32	flags;
299 	__u32	ctx_id;
300 	/** AMDGPU_CTX_PRIORITY_* */
301 	__s32	priority;
302 };
303 
304 union drm_amdgpu_ctx_out {
305 		struct {
306 			__u32	ctx_id;
307 			__u32	_pad;
308 		} alloc;
309 
310 		struct {
311 			/** For future use, no flags defined so far */
312 			__u64	flags;
313 			/** Number of resets caused by this context so far. */
314 			__u32	hangs;
315 			/** Reset status since the last call of the ioctl. */
316 			__u32	reset_status;
317 		} state;
318 
319 		struct {
320 			__u32	flags;
321 			__u32	_pad;
322 		} pstate;
323 };
324 
325 union drm_amdgpu_ctx {
326 	struct drm_amdgpu_ctx_in in;
327 	union drm_amdgpu_ctx_out out;
328 };
329 
330 /* user queue IOCTL operations */
331 #define AMDGPU_USERQ_OP_CREATE	1
332 #define AMDGPU_USERQ_OP_FREE	2
333 
334 /* queue priority levels */
335 /* low < normal low < normal high < high */
336 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_MASK  0x3
337 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_SHIFT 0
338 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_NORMAL_LOW 0
339 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_LOW 1
340 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_NORMAL_HIGH 2
341 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_HIGH 3 /* admin only */
342 /* for queues that need access to protected content */
343 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_SECURE  (1 << 2)
344 
345 /*
346  * This structure is a container to pass input configuration
347  * info for all supported userqueue related operations.
348  * For operation AMDGPU_USERQ_OP_CREATE: user is expected
349  *  to set all fields, excep the parameter 'queue_id'.
350  * For operation AMDGPU_USERQ_OP_FREE: the only input parameter expected
351  *  to be set is 'queue_id', eveything else is ignored.
352  */
353 struct drm_amdgpu_userq_in {
354 	/** AMDGPU_USERQ_OP_* */
355 	__u32	op;
356 	/** Queue id passed for operation USERQ_OP_FREE */
357 	__u32	queue_id;
358 	/** the target GPU engine to execute workload (AMDGPU_HW_IP_*) */
359 	__u32   ip_type;
360 	/**
361 	 * @doorbell_handle: the handle of doorbell GEM object
362 	 * associated with this userqueue client.
363 	 */
364 	__u32   doorbell_handle;
365 	/**
366 	 * @doorbell_offset: 32-bit offset of the doorbell in the doorbell bo.
367 	 * Kernel will generate absolute doorbell offset using doorbell_handle
368 	 * and doorbell_offset in the doorbell bo.
369 	 */
370 	__u32   doorbell_offset;
371 	/**
372 	 * @flags: flags used for queue parameters
373 	 */
374 	__u32 flags;
375 	/**
376 	 * @queue_va: Virtual address of the GPU memory which holds the queue
377 	 * object. The queue holds the workload packets.
378 	 */
379 	__u64   queue_va;
380 	/**
381 	 * @queue_size: Size of the queue in bytes, this needs to be 256-byte
382 	 * aligned.
383 	 */
384 	__u64   queue_size;
385 	/**
386 	 * @rptr_va : Virtual address of the GPU memory which holds the ring RPTR.
387 	 * This object must be at least 8 byte in size and aligned to 8-byte offset.
388 	 */
389 	__u64   rptr_va;
390 	/**
391 	 * @wptr_va : Virtual address of the GPU memory which holds the ring WPTR.
392 	 * This object must be at least 8 byte in size and aligned to 8-byte offset.
393 	 *
394 	 * Queue, RPTR and WPTR can come from the same object, as long as the size
395 	 * and alignment related requirements are met.
396 	 */
397 	__u64   wptr_va;
398 	/**
399 	 * @mqd: MQD (memory queue descriptor) is a set of parameters which allow
400 	 * the GPU to uniquely define and identify a usermode queue.
401 	 *
402 	 * MQD data can be of different size for different GPU IP/engine and
403 	 * their respective versions/revisions, so this points to a __u64 *
404 	 * which holds IP specific MQD of this usermode queue.
405 	 */
406 	__u64 mqd;
407 	/**
408 	 * @size: size of MQD data in bytes, it must match the MQD structure
409 	 * size of the respective engine/revision defined in UAPI for ex, for
410 	 * gfx11 workloads, size = sizeof(drm_amdgpu_userq_mqd_gfx11).
411 	 */
412 	__u64 mqd_size;
413 };
414 
415 /* The structure to carry output of userqueue ops */
416 struct drm_amdgpu_userq_out {
417 	/**
418 	 * For operation AMDGPU_USERQ_OP_CREATE: This field contains a unique
419 	 * queue ID to represent the newly created userqueue in the system, otherwise
420 	 * it should be ignored.
421 	 */
422 	__u32	queue_id;
423 	__u32 _pad;
424 };
425 
426 union drm_amdgpu_userq {
427 	struct drm_amdgpu_userq_in in;
428 	struct drm_amdgpu_userq_out out;
429 };
430 
431 /* GFX V11 IP specific MQD parameters */
432 struct drm_amdgpu_userq_mqd_gfx11 {
433 	/**
434 	 * @shadow_va: Virtual address of the GPU memory to hold the shadow buffer.
435 	 * Use AMDGPU_INFO_IOCTL to find the exact size of the object.
436 	 */
437 	__u64   shadow_va;
438 	/**
439 	 * @csa_va: Virtual address of the GPU memory to hold the CSA buffer.
440 	 * Use AMDGPU_INFO_IOCTL to find the exact size of the object.
441 	 */
442 	__u64   csa_va;
443 };
444 
445 /* GFX V11 SDMA IP specific MQD parameters */
446 struct drm_amdgpu_userq_mqd_sdma_gfx11 {
447 	/**
448 	 * @csa_va: Virtual address of the GPU memory to hold the CSA buffer.
449 	 * This must be a from a separate GPU object, and use AMDGPU_INFO IOCTL
450 	 * to get the size.
451 	 */
452 	__u64   csa_va;
453 };
454 
455 /* GFX V11 Compute IP specific MQD parameters */
456 struct drm_amdgpu_userq_mqd_compute_gfx11 {
457 	/**
458 	 * @eop_va: Virtual address of the GPU memory to hold the EOP buffer.
459 	 * This must be a from a separate GPU object, and use AMDGPU_INFO IOCTL
460 	 * to get the size.
461 	 */
462 	__u64   eop_va;
463 };
464 
465 /* userq signal/wait ioctl */
466 struct drm_amdgpu_userq_signal {
467 	/**
468 	 * @queue_id: Queue handle used by the userq fence creation function
469 	 * to retrieve the WPTR.
470 	 */
471 	__u32	queue_id;
472 	__u32	pad;
473 	/**
474 	 * @syncobj_handles: The list of syncobj handles submitted by the user queue
475 	 * job to be signaled.
476 	 */
477 	__u64	syncobj_handles;
478 	/**
479 	 * @num_syncobj_handles: A count that represents the number of syncobj handles in
480 	 * @syncobj_handles.
481 	 */
482 	__u64	num_syncobj_handles;
483 	/**
484 	 * @bo_read_handles: The list of BO handles that the submitted user queue job
485 	 * is using for read only. This will update BO fences in the kernel.
486 	 */
487 	__u64	bo_read_handles;
488 	/**
489 	 * @bo_write_handles: The list of BO handles that the submitted user queue job
490 	 * is using for write only. This will update BO fences in the kernel.
491 	 */
492 	__u64	bo_write_handles;
493 	/**
494 	 * @num_bo_read_handles: A count that represents the number of read BO handles in
495 	 * @bo_read_handles.
496 	 */
497 	__u32	num_bo_read_handles;
498 	/**
499 	 * @num_bo_write_handles: A count that represents the number of write BO handles in
500 	 * @bo_write_handles.
501 	 */
502 	__u32	num_bo_write_handles;
503 };
504 
505 struct drm_amdgpu_userq_fence_info {
506 	/**
507 	 * @va: A gpu address allocated for each queue which stores the
508 	 * read pointer (RPTR) value.
509 	 */
510 	__u64	va;
511 	/**
512 	 * @value: A 64 bit value represents the write pointer (WPTR) of the
513 	 * queue commands which compared with the RPTR value to signal the
514 	 * fences.
515 	 */
516 	__u64	value;
517 };
518 
519 struct drm_amdgpu_userq_wait {
520 	/**
521 	 * @waitq_id: Queue handle used by the userq wait IOCTL to retrieve the
522 	 * wait queue and maintain the fence driver references in it.
523 	 */
524 	__u32	waitq_id;
525 	__u32	pad;
526 	/**
527 	 * @syncobj_handles: The list of syncobj handles submitted by the user queue
528 	 * job to get the va/value pairs.
529 	 */
530 	__u64	syncobj_handles;
531 	/**
532 	 * @syncobj_timeline_handles: The list of timeline syncobj handles submitted by
533 	 * the user queue job to get the va/value pairs at given @syncobj_timeline_points.
534 	 */
535 	__u64	syncobj_timeline_handles;
536 	/**
537 	 * @syncobj_timeline_points: The list of timeline syncobj points submitted by the
538 	 * user queue job for the corresponding @syncobj_timeline_handles.
539 	 */
540 	__u64	syncobj_timeline_points;
541 	/**
542 	 * @bo_read_handles: The list of read BO handles submitted by the user queue
543 	 * job to get the va/value pairs.
544 	 */
545 	__u64	bo_read_handles;
546 	/**
547 	 * @bo_write_handles: The list of write BO handles submitted by the user queue
548 	 * job to get the va/value pairs.
549 	 */
550 	__u64	bo_write_handles;
551 	/**
552 	 * @num_syncobj_timeline_handles: A count that represents the number of timeline
553 	 * syncobj handles in @syncobj_timeline_handles.
554 	 */
555 	__u16	num_syncobj_timeline_handles;
556 	/**
557 	 * @num_fences: This field can be used both as input and output. As input it defines
558 	 * the maximum number of fences that can be returned and as output it will specify
559 	 * how many fences were actually returned from the ioctl.
560 	 */
561 	__u16	num_fences;
562 	/**
563 	 * @num_syncobj_handles: A count that represents the number of syncobj handles in
564 	 * @syncobj_handles.
565 	 */
566 	__u32	num_syncobj_handles;
567 	/**
568 	 * @num_bo_read_handles: A count that represents the number of read BO handles in
569 	 * @bo_read_handles.
570 	 */
571 	__u32	num_bo_read_handles;
572 	/**
573 	 * @num_bo_write_handles: A count that represents the number of write BO handles in
574 	 * @bo_write_handles.
575 	 */
576 	__u32	num_bo_write_handles;
577 	/**
578 	 * @out_fences: The field is a return value from the ioctl containing the list of
579 	 * address/value pairs to wait for.
580 	 */
581 	__u64	out_fences;
582 };
583 
584 /* vm ioctl */
585 #define AMDGPU_VM_OP_RESERVE_VMID	1
586 #define AMDGPU_VM_OP_UNRESERVE_VMID	2
587 
588 struct drm_amdgpu_vm_in {
589 	/** AMDGPU_VM_OP_* */
590 	__u32	op;
591 	__u32	flags;
592 };
593 
594 struct drm_amdgpu_vm_out {
595 	/** For future use, no flags defined so far */
596 	__u64	flags;
597 };
598 
599 union drm_amdgpu_vm {
600 	struct drm_amdgpu_vm_in in;
601 	struct drm_amdgpu_vm_out out;
602 };
603 
604 /* sched ioctl */
605 #define AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE	1
606 #define AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE	2
607 
608 struct drm_amdgpu_sched_in {
609 	/* AMDGPU_SCHED_OP_* */
610 	__u32	op;
611 	__u32	fd;
612 	/** AMDGPU_CTX_PRIORITY_* */
613 	__s32	priority;
614 	__u32   ctx_id;
615 };
616 
617 union drm_amdgpu_sched {
618 	struct drm_amdgpu_sched_in in;
619 };
620 
621 /*
622  * This is not a reliable API and you should expect it to fail for any
623  * number of reasons and have fallback path that do not use userptr to
624  * perform any operation.
625  */
626 #define AMDGPU_GEM_USERPTR_READONLY	(1 << 0)
627 #define AMDGPU_GEM_USERPTR_ANONONLY	(1 << 1)
628 #define AMDGPU_GEM_USERPTR_VALIDATE	(1 << 2)
629 #define AMDGPU_GEM_USERPTR_REGISTER	(1 << 3)
630 
631 struct drm_amdgpu_gem_userptr {
632 	__u64		addr;
633 	__u64		size;
634 	/* AMDGPU_GEM_USERPTR_* */
635 	__u32		flags;
636 	/* Resulting GEM handle */
637 	__u32		handle;
638 };
639 
640 /* SI-CI-VI: */
641 /* same meaning as the GB_TILE_MODE and GL_MACRO_TILE_MODE fields */
642 #define AMDGPU_TILING_ARRAY_MODE_SHIFT			0
643 #define AMDGPU_TILING_ARRAY_MODE_MASK			0xf
644 #define AMDGPU_TILING_PIPE_CONFIG_SHIFT			4
645 #define AMDGPU_TILING_PIPE_CONFIG_MASK			0x1f
646 #define AMDGPU_TILING_TILE_SPLIT_SHIFT			9
647 #define AMDGPU_TILING_TILE_SPLIT_MASK			0x7
648 #define AMDGPU_TILING_MICRO_TILE_MODE_SHIFT		12
649 #define AMDGPU_TILING_MICRO_TILE_MODE_MASK		0x7
650 #define AMDGPU_TILING_BANK_WIDTH_SHIFT			15
651 #define AMDGPU_TILING_BANK_WIDTH_MASK			0x3
652 #define AMDGPU_TILING_BANK_HEIGHT_SHIFT			17
653 #define AMDGPU_TILING_BANK_HEIGHT_MASK			0x3
654 #define AMDGPU_TILING_MACRO_TILE_ASPECT_SHIFT		19
655 #define AMDGPU_TILING_MACRO_TILE_ASPECT_MASK		0x3
656 #define AMDGPU_TILING_NUM_BANKS_SHIFT			21
657 #define AMDGPU_TILING_NUM_BANKS_MASK			0x3
658 
659 /* GFX9 - GFX11: */
660 #define AMDGPU_TILING_SWIZZLE_MODE_SHIFT		0
661 #define AMDGPU_TILING_SWIZZLE_MODE_MASK			0x1f
662 #define AMDGPU_TILING_DCC_OFFSET_256B_SHIFT		5
663 #define AMDGPU_TILING_DCC_OFFSET_256B_MASK		0xFFFFFF
664 #define AMDGPU_TILING_DCC_PITCH_MAX_SHIFT		29
665 #define AMDGPU_TILING_DCC_PITCH_MAX_MASK		0x3FFF
666 #define AMDGPU_TILING_DCC_INDEPENDENT_64B_SHIFT		43
667 #define AMDGPU_TILING_DCC_INDEPENDENT_64B_MASK		0x1
668 #define AMDGPU_TILING_DCC_INDEPENDENT_128B_SHIFT	44
669 #define AMDGPU_TILING_DCC_INDEPENDENT_128B_MASK		0x1
670 #define AMDGPU_TILING_SCANOUT_SHIFT			63
671 #define AMDGPU_TILING_SCANOUT_MASK			0x1
672 
673 /* GFX12 and later: */
674 #define AMDGPU_TILING_GFX12_SWIZZLE_MODE_SHIFT			0
675 #define AMDGPU_TILING_GFX12_SWIZZLE_MODE_MASK			0x7
676 /* These are DCC recompression settings for memory management: */
677 #define AMDGPU_TILING_GFX12_DCC_MAX_COMPRESSED_BLOCK_SHIFT	3
678 #define AMDGPU_TILING_GFX12_DCC_MAX_COMPRESSED_BLOCK_MASK	0x3 /* 0:64B, 1:128B, 2:256B */
679 #define AMDGPU_TILING_GFX12_DCC_NUMBER_TYPE_SHIFT		5
680 #define AMDGPU_TILING_GFX12_DCC_NUMBER_TYPE_MASK		0x7 /* CB_COLOR0_INFO.NUMBER_TYPE */
681 #define AMDGPU_TILING_GFX12_DCC_DATA_FORMAT_SHIFT		8
682 #define AMDGPU_TILING_GFX12_DCC_DATA_FORMAT_MASK		0x3f /* [0:4]:CB_COLOR0_INFO.FORMAT, [5]:MM */
683 /* When clearing the buffer or moving it from VRAM to GTT, don't compress and set DCC metadata
684  * to uncompressed. Set when parts of an allocation bypass DCC and read raw data. */
685 #define AMDGPU_TILING_GFX12_DCC_WRITE_COMPRESS_DISABLE_SHIFT	14
686 #define AMDGPU_TILING_GFX12_DCC_WRITE_COMPRESS_DISABLE_MASK	0x1
687 /* bit gap */
688 #define AMDGPU_TILING_GFX12_SCANOUT_SHIFT			63
689 #define AMDGPU_TILING_GFX12_SCANOUT_MASK			0x1
690 
691 /* Set/Get helpers for tiling flags. */
692 #define AMDGPU_TILING_SET(field, value) \
693 	(((__u64)(value) & AMDGPU_TILING_##field##_MASK) << AMDGPU_TILING_##field##_SHIFT)
694 #define AMDGPU_TILING_GET(value, field) \
695 	(((__u64)(value) >> AMDGPU_TILING_##field##_SHIFT) & AMDGPU_TILING_##field##_MASK)
696 
697 #define AMDGPU_GEM_METADATA_OP_SET_METADATA                  1
698 #define AMDGPU_GEM_METADATA_OP_GET_METADATA                  2
699 
700 /** The same structure is shared for input/output */
701 struct drm_amdgpu_gem_metadata {
702 	/** GEM Object handle */
703 	__u32	handle;
704 	/** Do we want get or set metadata */
705 	__u32	op;
706 	struct {
707 		/** For future use, no flags defined so far */
708 		__u64	flags;
709 		/** family specific tiling info */
710 		__u64	tiling_info;
711 		__u32	data_size_bytes;
712 		__u32	data[64];
713 	} data;
714 };
715 
716 struct drm_amdgpu_gem_mmap_in {
717 	/** the GEM object handle */
718 	__u32 handle;
719 	__u32 _pad;
720 };
721 
722 struct drm_amdgpu_gem_mmap_out {
723 	/** mmap offset from the vma offset manager */
724 	__u64 addr_ptr;
725 };
726 
727 union drm_amdgpu_gem_mmap {
728 	struct drm_amdgpu_gem_mmap_in   in;
729 	struct drm_amdgpu_gem_mmap_out out;
730 };
731 
732 struct drm_amdgpu_gem_wait_idle_in {
733 	/** GEM object handle */
734 	__u32 handle;
735 	/** For future use, no flags defined so far */
736 	__u32 flags;
737 	/** Absolute timeout to wait */
738 	__u64 timeout;
739 };
740 
741 struct drm_amdgpu_gem_wait_idle_out {
742 	/** BO status:  0 - BO is idle, 1 - BO is busy */
743 	__u32 status;
744 	/** Returned current memory domain */
745 	__u32 domain;
746 };
747 
748 union drm_amdgpu_gem_wait_idle {
749 	struct drm_amdgpu_gem_wait_idle_in  in;
750 	struct drm_amdgpu_gem_wait_idle_out out;
751 };
752 
753 struct drm_amdgpu_wait_cs_in {
754 	/* Command submission handle
755          * handle equals 0 means none to wait for
756          * handle equals ~0ull means wait for the latest sequence number
757          */
758 	__u64 handle;
759 	/** Absolute timeout to wait */
760 	__u64 timeout;
761 	__u32 ip_type;
762 	__u32 ip_instance;
763 	__u32 ring;
764 	__u32 ctx_id;
765 };
766 
767 struct drm_amdgpu_wait_cs_out {
768 	/** CS status:  0 - CS completed, 1 - CS still busy */
769 	__u64 status;
770 };
771 
772 union drm_amdgpu_wait_cs {
773 	struct drm_amdgpu_wait_cs_in in;
774 	struct drm_amdgpu_wait_cs_out out;
775 };
776 
777 struct drm_amdgpu_fence {
778 	__u32 ctx_id;
779 	__u32 ip_type;
780 	__u32 ip_instance;
781 	__u32 ring;
782 	__u64 seq_no;
783 };
784 
785 struct drm_amdgpu_wait_fences_in {
786 	/** This points to uint64_t * which points to fences */
787 	__u64 fences;
788 	__u32 fence_count;
789 	__u32 wait_all;
790 	__u64 timeout_ns;
791 };
792 
793 struct drm_amdgpu_wait_fences_out {
794 	__u32 status;
795 	__u32 first_signaled;
796 };
797 
798 union drm_amdgpu_wait_fences {
799 	struct drm_amdgpu_wait_fences_in in;
800 	struct drm_amdgpu_wait_fences_out out;
801 };
802 
803 #define AMDGPU_GEM_OP_GET_GEM_CREATE_INFO	0
804 #define AMDGPU_GEM_OP_SET_PLACEMENT		1
805 #define AMDGPU_GEM_OP_GET_MAPPING_INFO		2
806 
807 struct drm_amdgpu_gem_vm_entry {
808 	/* Start of mapping (in bytes) */
809 	__u64 addr;
810 
811 	/* Size of mapping (in bytes) */
812 	__u64 size;
813 
814 	/* Mapping offset */
815 	__u64 offset;
816 
817 	/* flags needed to recreate mapping */
818 	__u64 flags;
819 };
820 
821 /* Sets or returns a value associated with a buffer. */
822 struct drm_amdgpu_gem_op {
823 	/** GEM object handle */
824 	__u32	handle;
825 	/** AMDGPU_GEM_OP_* */
826 	__u32	op;
827 	/** Input or return value. For MAPPING_INFO op: pointer to array of struct drm_amdgpu_gem_vm_entry */
828 	__u64	value;
829 	/** For MAPPING_INFO op: number of mappings (in/out) */
830 	__u32	num_entries;
831 
832 	__u32	padding;
833 };
834 
835 #define AMDGPU_GEM_LIST_HANDLES_FLAG_IS_IMPORT	(1 << 0)
836 
837 struct drm_amdgpu_gem_list_handles {
838 	/* User pointer to array of drm_amdgpu_gem_bo_info_entry */
839 	__u64   entries;
840 
841 	/* Size of entries buffer / Number of handles in process (if larger than size of buffer, must retry) */
842 	__u32   num_entries;
843 
844 	__u32 padding;
845 };
846 
847 struct drm_amdgpu_gem_list_handles_entry {
848 	/* gem handle of buffer object */
849 	__u32 gem_handle;
850 
851 	/* Currently just one flag: IS_IMPORT */
852 	__u32 flags;
853 
854 	/* Size of bo */
855 	__u64 size;
856 
857 	/* Preferred domains for GEM_CREATE */
858 	__u64 preferred_domains;
859 
860 	/* GEM_CREATE flags for re-creation of buffer */
861 	__u64 alloc_flags;
862 
863 	/* physical start_addr alignment in bytes for some HW requirements */
864 	__u64 alignment;
865 };
866 
867 #define AMDGPU_VA_OP_MAP			1
868 #define AMDGPU_VA_OP_UNMAP			2
869 #define AMDGPU_VA_OP_CLEAR			3
870 #define AMDGPU_VA_OP_REPLACE			4
871 
872 /* Delay the page table update till the next CS */
873 #define AMDGPU_VM_DELAY_UPDATE		(1 << 0)
874 
875 /* Mapping flags */
876 /* readable mapping */
877 #define AMDGPU_VM_PAGE_READABLE		(1 << 1)
878 /* writable mapping */
879 #define AMDGPU_VM_PAGE_WRITEABLE	(1 << 2)
880 /* executable mapping, new for VI */
881 #define AMDGPU_VM_PAGE_EXECUTABLE	(1 << 3)
882 /* partially resident texture */
883 #define AMDGPU_VM_PAGE_PRT		(1 << 4)
884 /* MTYPE flags use bit 5 to 8 */
885 #define AMDGPU_VM_MTYPE_MASK		(0xf << 5)
886 /* Default MTYPE. Pre-AI must use this.  Recommended for newer ASICs. */
887 #define AMDGPU_VM_MTYPE_DEFAULT		(0 << 5)
888 /* Use Non Coherent MTYPE instead of default MTYPE */
889 #define AMDGPU_VM_MTYPE_NC		(1 << 5)
890 /* Use Write Combine MTYPE instead of default MTYPE */
891 #define AMDGPU_VM_MTYPE_WC		(2 << 5)
892 /* Use Cache Coherent MTYPE instead of default MTYPE */
893 #define AMDGPU_VM_MTYPE_CC		(3 << 5)
894 /* Use UnCached MTYPE instead of default MTYPE */
895 #define AMDGPU_VM_MTYPE_UC		(4 << 5)
896 /* Use Read Write MTYPE instead of default MTYPE */
897 #define AMDGPU_VM_MTYPE_RW		(5 << 5)
898 /* don't allocate MALL */
899 #define AMDGPU_VM_PAGE_NOALLOC		(1 << 9)
900 
901 struct drm_amdgpu_gem_va {
902 	/** GEM object handle */
903 	__u32 handle;
904 	__u32 _pad;
905 	/** AMDGPU_VA_OP_* */
906 	__u32 operation;
907 	/** AMDGPU_VM_PAGE_* */
908 	__u32 flags;
909 	/** va address to assign . Must be correctly aligned.*/
910 	__u64 va_address;
911 	/** Specify offset inside of BO to assign. Must be correctly aligned.*/
912 	__u64 offset_in_bo;
913 	/** Specify mapping size. Must be correctly aligned. */
914 	__u64 map_size;
915 	/**
916 	 * vm_timeline_point is a sequence number used to add new timeline point.
917 	 */
918 	__u64 vm_timeline_point;
919 	/**
920 	 * The vm page table update fence is installed in given vm_timeline_syncobj_out
921 	 * at vm_timeline_point.
922 	 */
923 	__u32 vm_timeline_syncobj_out;
924 	/** the number of syncobj handles in @input_fence_syncobj_handles */
925 	__u32 num_syncobj_handles;
926 	/** Array of sync object handle to wait for given input fences */
927 	__u64 input_fence_syncobj_handles;
928 };
929 
930 #define AMDGPU_HW_IP_GFX          0
931 #define AMDGPU_HW_IP_COMPUTE      1
932 #define AMDGPU_HW_IP_DMA          2
933 #define AMDGPU_HW_IP_UVD          3
934 #define AMDGPU_HW_IP_VCE          4
935 #define AMDGPU_HW_IP_UVD_ENC      5
936 #define AMDGPU_HW_IP_VCN_DEC      6
937 /*
938  * From VCN4, AMDGPU_HW_IP_VCN_ENC is re-used to support
939  * both encoding and decoding jobs.
940  */
941 #define AMDGPU_HW_IP_VCN_ENC      7
942 #define AMDGPU_HW_IP_VCN_JPEG     8
943 #define AMDGPU_HW_IP_VPE          9
944 #define AMDGPU_HW_IP_NUM          10
945 
946 #define AMDGPU_HW_IP_INSTANCE_MAX_COUNT 1
947 
948 #define AMDGPU_CHUNK_ID_IB		0x01
949 #define AMDGPU_CHUNK_ID_FENCE		0x02
950 #define AMDGPU_CHUNK_ID_DEPENDENCIES	0x03
951 #define AMDGPU_CHUNK_ID_SYNCOBJ_IN      0x04
952 #define AMDGPU_CHUNK_ID_SYNCOBJ_OUT     0x05
953 #define AMDGPU_CHUNK_ID_BO_HANDLES      0x06
954 #define AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES	0x07
955 #define AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT    0x08
956 #define AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL  0x09
957 #define AMDGPU_CHUNK_ID_CP_GFX_SHADOW   0x0a
958 
959 struct drm_amdgpu_cs_chunk {
960 	__u32		chunk_id;
961 	__u32		length_dw;
962 	__u64		chunk_data;
963 };
964 
965 struct drm_amdgpu_cs_in {
966 	/** Rendering context id */
967 	__u32		ctx_id;
968 	/**  Handle of resource list associated with CS */
969 	__u32		bo_list_handle;
970 	__u32		num_chunks;
971 	__u32		flags;
972 	/** this points to __u64 * which point to cs chunks */
973 	__u64		chunks;
974 };
975 
976 struct drm_amdgpu_cs_out {
977 	__u64 handle;
978 };
979 
980 union drm_amdgpu_cs {
981 	struct drm_amdgpu_cs_in in;
982 	struct drm_amdgpu_cs_out out;
983 };
984 
985 /* Specify flags to be used for IB */
986 
987 /* This IB should be submitted to CE */
988 #define AMDGPU_IB_FLAG_CE	(1<<0)
989 
990 /* Preamble flag, which means the IB could be dropped if no context switch */
991 #define AMDGPU_IB_FLAG_PREAMBLE (1<<1)
992 
993 /* Preempt flag, IB should set Pre_enb bit if PREEMPT flag detected */
994 #define AMDGPU_IB_FLAG_PREEMPT (1<<2)
995 
996 /* The IB fence should do the L2 writeback but not invalidate any shader
997  * caches (L2/vL1/sL1/I$). */
998 #define AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE (1 << 3)
999 
1000 /* Set GDS_COMPUTE_MAX_WAVE_ID = DEFAULT before PACKET3_INDIRECT_BUFFER.
1001  * This will reset wave ID counters for the IB.
1002  */
1003 #define AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID (1 << 4)
1004 
1005 /* Flag the IB as secure (TMZ)
1006  */
1007 #define AMDGPU_IB_FLAGS_SECURE  (1 << 5)
1008 
1009 /* Tell KMD to flush and invalidate caches
1010  */
1011 #define AMDGPU_IB_FLAG_EMIT_MEM_SYNC  (1 << 6)
1012 
1013 struct drm_amdgpu_cs_chunk_ib {
1014 	__u32 _pad;
1015 	/** AMDGPU_IB_FLAG_* */
1016 	__u32 flags;
1017 	/** Virtual address to begin IB execution */
1018 	__u64 va_start;
1019 	/** Size of submission */
1020 	__u32 ib_bytes;
1021 	/** HW IP to submit to */
1022 	__u32 ip_type;
1023 	/** HW IP index of the same type to submit to  */
1024 	__u32 ip_instance;
1025 	/** Ring index to submit to */
1026 	__u32 ring;
1027 };
1028 
1029 struct drm_amdgpu_cs_chunk_dep {
1030 	__u32 ip_type;
1031 	__u32 ip_instance;
1032 	__u32 ring;
1033 	__u32 ctx_id;
1034 	__u64 handle;
1035 };
1036 
1037 struct drm_amdgpu_cs_chunk_fence {
1038 	__u32 handle;
1039 	__u32 offset;
1040 };
1041 
1042 struct drm_amdgpu_cs_chunk_sem {
1043 	__u32 handle;
1044 };
1045 
1046 struct drm_amdgpu_cs_chunk_syncobj {
1047        __u32 handle;
1048        __u32 flags;
1049        __u64 point;
1050 };
1051 
1052 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ	0
1053 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ_FD	1
1054 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNC_FILE_FD	2
1055 
1056 union drm_amdgpu_fence_to_handle {
1057 	struct {
1058 		struct drm_amdgpu_fence fence;
1059 		__u32 what;
1060 		__u32 pad;
1061 	} in;
1062 	struct {
1063 		__u32 handle;
1064 	} out;
1065 };
1066 
1067 struct drm_amdgpu_cs_chunk_data {
1068 	union {
1069 		struct drm_amdgpu_cs_chunk_ib		ib_data;
1070 		struct drm_amdgpu_cs_chunk_fence	fence_data;
1071 	};
1072 };
1073 
1074 #define AMDGPU_CS_CHUNK_CP_GFX_SHADOW_FLAGS_INIT_SHADOW         0x1
1075 
1076 struct drm_amdgpu_cs_chunk_cp_gfx_shadow {
1077 	__u64 shadow_va;
1078 	__u64 csa_va;
1079 	__u64 gds_va;
1080 	__u64 flags;
1081 };
1082 
1083 /*
1084  *  Query h/w info: Flag that this is integrated (a.h.a. fusion) GPU
1085  *
1086  */
1087 #define AMDGPU_IDS_FLAGS_FUSION         0x1
1088 #define AMDGPU_IDS_FLAGS_PREEMPTION     0x2
1089 #define AMDGPU_IDS_FLAGS_TMZ            0x4
1090 #define AMDGPU_IDS_FLAGS_CONFORMANT_TRUNC_COORD 0x8
1091 
1092 /*
1093  *  Query h/w info: Flag identifying VF/PF/PT mode
1094  *
1095  */
1096 #define AMDGPU_IDS_FLAGS_MODE_MASK      0x300
1097 #define AMDGPU_IDS_FLAGS_MODE_SHIFT     0x8
1098 #define AMDGPU_IDS_FLAGS_MODE_PF        0x0
1099 #define AMDGPU_IDS_FLAGS_MODE_VF        0x1
1100 #define AMDGPU_IDS_FLAGS_MODE_PT        0x2
1101 
1102 /* indicate if acceleration can be working */
1103 #define AMDGPU_INFO_ACCEL_WORKING		0x00
1104 /* get the crtc_id from the mode object id? */
1105 #define AMDGPU_INFO_CRTC_FROM_ID		0x01
1106 /* query hw IP info */
1107 #define AMDGPU_INFO_HW_IP_INFO			0x02
1108 /* query hw IP instance count for the specified type */
1109 #define AMDGPU_INFO_HW_IP_COUNT			0x03
1110 /* timestamp for GL_ARB_timer_query */
1111 #define AMDGPU_INFO_TIMESTAMP			0x05
1112 /* Query the firmware version */
1113 #define AMDGPU_INFO_FW_VERSION			0x0e
1114 	/* Subquery id: Query VCE firmware version */
1115 	#define AMDGPU_INFO_FW_VCE		0x1
1116 	/* Subquery id: Query UVD firmware version */
1117 	#define AMDGPU_INFO_FW_UVD		0x2
1118 	/* Subquery id: Query GMC firmware version */
1119 	#define AMDGPU_INFO_FW_GMC		0x03
1120 	/* Subquery id: Query GFX ME firmware version */
1121 	#define AMDGPU_INFO_FW_GFX_ME		0x04
1122 	/* Subquery id: Query GFX PFP firmware version */
1123 	#define AMDGPU_INFO_FW_GFX_PFP		0x05
1124 	/* Subquery id: Query GFX CE firmware version */
1125 	#define AMDGPU_INFO_FW_GFX_CE		0x06
1126 	/* Subquery id: Query GFX RLC firmware version */
1127 	#define AMDGPU_INFO_FW_GFX_RLC		0x07
1128 	/* Subquery id: Query GFX MEC firmware version */
1129 	#define AMDGPU_INFO_FW_GFX_MEC		0x08
1130 	/* Subquery id: Query SMC firmware version */
1131 	#define AMDGPU_INFO_FW_SMC		0x0a
1132 	/* Subquery id: Query SDMA firmware version */
1133 	#define AMDGPU_INFO_FW_SDMA		0x0b
1134 	/* Subquery id: Query PSP SOS firmware version */
1135 	#define AMDGPU_INFO_FW_SOS		0x0c
1136 	/* Subquery id: Query PSP ASD firmware version */
1137 	#define AMDGPU_INFO_FW_ASD		0x0d
1138 	/* Subquery id: Query VCN firmware version */
1139 	#define AMDGPU_INFO_FW_VCN		0x0e
1140 	/* Subquery id: Query GFX RLC SRLC firmware version */
1141 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL 0x0f
1142 	/* Subquery id: Query GFX RLC SRLG firmware version */
1143 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM 0x10
1144 	/* Subquery id: Query GFX RLC SRLS firmware version */
1145 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM 0x11
1146 	/* Subquery id: Query DMCU firmware version */
1147 	#define AMDGPU_INFO_FW_DMCU		0x12
1148 	#define AMDGPU_INFO_FW_TA		0x13
1149 	/* Subquery id: Query DMCUB firmware version */
1150 	#define AMDGPU_INFO_FW_DMCUB		0x14
1151 	/* Subquery id: Query TOC firmware version */
1152 	#define AMDGPU_INFO_FW_TOC		0x15
1153 	/* Subquery id: Query CAP firmware version */
1154 	#define AMDGPU_INFO_FW_CAP		0x16
1155 	/* Subquery id: Query GFX RLCP firmware version */
1156 	#define AMDGPU_INFO_FW_GFX_RLCP		0x17
1157 	/* Subquery id: Query GFX RLCV firmware version */
1158 	#define AMDGPU_INFO_FW_GFX_RLCV		0x18
1159 	/* Subquery id: Query MES_KIQ firmware version */
1160 	#define AMDGPU_INFO_FW_MES_KIQ		0x19
1161 	/* Subquery id: Query MES firmware version */
1162 	#define AMDGPU_INFO_FW_MES		0x1a
1163 	/* Subquery id: Query IMU firmware version */
1164 	#define AMDGPU_INFO_FW_IMU		0x1b
1165 	/* Subquery id: Query VPE firmware version */
1166 	#define AMDGPU_INFO_FW_VPE		0x1c
1167 
1168 /* number of bytes moved for TTM migration */
1169 #define AMDGPU_INFO_NUM_BYTES_MOVED		0x0f
1170 /* the used VRAM size */
1171 #define AMDGPU_INFO_VRAM_USAGE			0x10
1172 /* the used GTT size */
1173 #define AMDGPU_INFO_GTT_USAGE			0x11
1174 /* Information about GDS, etc. resource configuration */
1175 #define AMDGPU_INFO_GDS_CONFIG			0x13
1176 /* Query information about VRAM and GTT domains */
1177 #define AMDGPU_INFO_VRAM_GTT			0x14
1178 /* Query information about register in MMR address space*/
1179 #define AMDGPU_INFO_READ_MMR_REG		0x15
1180 /* Query information about device: rev id, family, etc. */
1181 #define AMDGPU_INFO_DEV_INFO			0x16
1182 /* visible vram usage */
1183 #define AMDGPU_INFO_VIS_VRAM_USAGE		0x17
1184 /* number of TTM buffer evictions */
1185 #define AMDGPU_INFO_NUM_EVICTIONS		0x18
1186 /* Query memory about VRAM and GTT domains */
1187 #define AMDGPU_INFO_MEMORY			0x19
1188 /* Query vce clock table */
1189 #define AMDGPU_INFO_VCE_CLOCK_TABLE		0x1A
1190 /* Query vbios related information */
1191 #define AMDGPU_INFO_VBIOS			0x1B
1192 	/* Subquery id: Query vbios size */
1193 	#define AMDGPU_INFO_VBIOS_SIZE		0x1
1194 	/* Subquery id: Query vbios image */
1195 	#define AMDGPU_INFO_VBIOS_IMAGE		0x2
1196 	/* Subquery id: Query vbios info */
1197 	#define AMDGPU_INFO_VBIOS_INFO		0x3
1198 /* Query UVD handles */
1199 #define AMDGPU_INFO_NUM_HANDLES			0x1C
1200 /* Query sensor related information */
1201 #define AMDGPU_INFO_SENSOR			0x1D
1202 	/* Subquery id: Query GPU shader clock */
1203 	#define AMDGPU_INFO_SENSOR_GFX_SCLK		0x1
1204 	/* Subquery id: Query GPU memory clock */
1205 	#define AMDGPU_INFO_SENSOR_GFX_MCLK		0x2
1206 	/* Subquery id: Query GPU temperature */
1207 	#define AMDGPU_INFO_SENSOR_GPU_TEMP		0x3
1208 	/* Subquery id: Query GPU load */
1209 	#define AMDGPU_INFO_SENSOR_GPU_LOAD		0x4
1210 	/* Subquery id: Query average GPU power	*/
1211 	#define AMDGPU_INFO_SENSOR_GPU_AVG_POWER	0x5
1212 	/* Subquery id: Query northbridge voltage */
1213 	#define AMDGPU_INFO_SENSOR_VDDNB		0x6
1214 	/* Subquery id: Query graphics voltage */
1215 	#define AMDGPU_INFO_SENSOR_VDDGFX		0x7
1216 	/* Subquery id: Query GPU stable pstate shader clock */
1217 	#define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_SCLK		0x8
1218 	/* Subquery id: Query GPU stable pstate memory clock */
1219 	#define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_MCLK		0x9
1220 	/* Subquery id: Query GPU peak pstate shader clock */
1221 	#define AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_SCLK			0xa
1222 	/* Subquery id: Query GPU peak pstate memory clock */
1223 	#define AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_MCLK			0xb
1224 	/* Subquery id: Query input GPU power	*/
1225 	#define AMDGPU_INFO_SENSOR_GPU_INPUT_POWER	0xc
1226 /* Number of VRAM page faults on CPU access. */
1227 #define AMDGPU_INFO_NUM_VRAM_CPU_PAGE_FAULTS	0x1E
1228 #define AMDGPU_INFO_VRAM_LOST_COUNTER		0x1F
1229 /* query ras mask of enabled features*/
1230 #define AMDGPU_INFO_RAS_ENABLED_FEATURES	0x20
1231 /* RAS MASK: UMC (VRAM) */
1232 #define AMDGPU_INFO_RAS_ENABLED_UMC			(1 << 0)
1233 /* RAS MASK: SDMA */
1234 #define AMDGPU_INFO_RAS_ENABLED_SDMA			(1 << 1)
1235 /* RAS MASK: GFX */
1236 #define AMDGPU_INFO_RAS_ENABLED_GFX			(1 << 2)
1237 /* RAS MASK: MMHUB */
1238 #define AMDGPU_INFO_RAS_ENABLED_MMHUB			(1 << 3)
1239 /* RAS MASK: ATHUB */
1240 #define AMDGPU_INFO_RAS_ENABLED_ATHUB			(1 << 4)
1241 /* RAS MASK: PCIE */
1242 #define AMDGPU_INFO_RAS_ENABLED_PCIE			(1 << 5)
1243 /* RAS MASK: HDP */
1244 #define AMDGPU_INFO_RAS_ENABLED_HDP			(1 << 6)
1245 /* RAS MASK: XGMI */
1246 #define AMDGPU_INFO_RAS_ENABLED_XGMI			(1 << 7)
1247 /* RAS MASK: DF */
1248 #define AMDGPU_INFO_RAS_ENABLED_DF			(1 << 8)
1249 /* RAS MASK: SMN */
1250 #define AMDGPU_INFO_RAS_ENABLED_SMN			(1 << 9)
1251 /* RAS MASK: SEM */
1252 #define AMDGPU_INFO_RAS_ENABLED_SEM			(1 << 10)
1253 /* RAS MASK: MP0 */
1254 #define AMDGPU_INFO_RAS_ENABLED_MP0			(1 << 11)
1255 /* RAS MASK: MP1 */
1256 #define AMDGPU_INFO_RAS_ENABLED_MP1			(1 << 12)
1257 /* RAS MASK: FUSE */
1258 #define AMDGPU_INFO_RAS_ENABLED_FUSE			(1 << 13)
1259 /* query video encode/decode caps */
1260 #define AMDGPU_INFO_VIDEO_CAPS			0x21
1261 	/* Subquery id: Decode */
1262 	#define AMDGPU_INFO_VIDEO_CAPS_DECODE		0
1263 	/* Subquery id: Encode */
1264 	#define AMDGPU_INFO_VIDEO_CAPS_ENCODE		1
1265 /* Query the max number of IBs per gang per submission */
1266 #define AMDGPU_INFO_MAX_IBS			0x22
1267 /* query last page fault info */
1268 #define AMDGPU_INFO_GPUVM_FAULT			0x23
1269 /* query FW object size and alignment */
1270 #define AMDGPU_INFO_UQ_FW_AREAS			0x24
1271 
1272 #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT	0
1273 #define AMDGPU_INFO_MMR_SE_INDEX_MASK	0xff
1274 #define AMDGPU_INFO_MMR_SH_INDEX_SHIFT	8
1275 #define AMDGPU_INFO_MMR_SH_INDEX_MASK	0xff
1276 
1277 struct drm_amdgpu_query_fw {
1278 	/** AMDGPU_INFO_FW_* */
1279 	__u32 fw_type;
1280 	/**
1281 	 * Index of the IP if there are more IPs of
1282 	 * the same type.
1283 	 */
1284 	__u32 ip_instance;
1285 	/**
1286 	 * Index of the engine. Whether this is used depends
1287 	 * on the firmware type. (e.g. MEC, SDMA)
1288 	 */
1289 	__u32 index;
1290 	__u32 _pad;
1291 };
1292 
1293 /* Input structure for the INFO ioctl */
1294 struct drm_amdgpu_info {
1295 	/* Where the return value will be stored */
1296 	__u64 return_pointer;
1297 	/* The size of the return value. Just like "size" in "snprintf",
1298 	 * it limits how many bytes the kernel can write. */
1299 	__u32 return_size;
1300 	/* The query request id. */
1301 	__u32 query;
1302 
1303 	union {
1304 		struct {
1305 			__u32 id;
1306 			__u32 _pad;
1307 		} mode_crtc;
1308 
1309 		struct {
1310 			/** AMDGPU_HW_IP_* */
1311 			__u32 type;
1312 			/**
1313 			 * Index of the IP if there are more IPs of the same
1314 			 * type. Ignored by AMDGPU_INFO_HW_IP_COUNT.
1315 			 */
1316 			__u32 ip_instance;
1317 		} query_hw_ip;
1318 
1319 		struct {
1320 			__u32 dword_offset;
1321 			/** number of registers to read */
1322 			__u32 count;
1323 			__u32 instance;
1324 			/** For future use, no flags defined so far */
1325 			__u32 flags;
1326 		} read_mmr_reg;
1327 
1328 		struct drm_amdgpu_query_fw query_fw;
1329 
1330 		struct {
1331 			__u32 type;
1332 			__u32 offset;
1333 		} vbios_info;
1334 
1335 		struct {
1336 			__u32 type;
1337 		} sensor_info;
1338 
1339 		struct {
1340 			__u32 type;
1341 		} video_cap;
1342 	};
1343 };
1344 
1345 struct drm_amdgpu_info_gds {
1346 	/** GDS GFX partition size */
1347 	__u32 gds_gfx_partition_size;
1348 	/** GDS compute partition size */
1349 	__u32 compute_partition_size;
1350 	/** total GDS memory size */
1351 	__u32 gds_total_size;
1352 	/** GWS size per GFX partition */
1353 	__u32 gws_per_gfx_partition;
1354 	/** GSW size per compute partition */
1355 	__u32 gws_per_compute_partition;
1356 	/** OA size per GFX partition */
1357 	__u32 oa_per_gfx_partition;
1358 	/** OA size per compute partition */
1359 	__u32 oa_per_compute_partition;
1360 	__u32 _pad;
1361 };
1362 
1363 struct drm_amdgpu_info_vram_gtt {
1364 	__u64 vram_size;
1365 	__u64 vram_cpu_accessible_size;
1366 	__u64 gtt_size;
1367 };
1368 
1369 struct drm_amdgpu_heap_info {
1370 	/** max. physical memory */
1371 	__u64 total_heap_size;
1372 
1373 	/** Theoretical max. available memory in the given heap */
1374 	__u64 usable_heap_size;
1375 
1376 	/**
1377 	 * Number of bytes allocated in the heap. This includes all processes
1378 	 * and private allocations in the kernel. It changes when new buffers
1379 	 * are allocated, freed, and moved. It cannot be larger than
1380 	 * heap_size.
1381 	 */
1382 	__u64 heap_usage;
1383 
1384 	/**
1385 	 * Theoretical possible max. size of buffer which
1386 	 * could be allocated in the given heap
1387 	 */
1388 	__u64 max_allocation;
1389 };
1390 
1391 struct drm_amdgpu_memory_info {
1392 	struct drm_amdgpu_heap_info vram;
1393 	struct drm_amdgpu_heap_info cpu_accessible_vram;
1394 	struct drm_amdgpu_heap_info gtt;
1395 };
1396 
1397 struct drm_amdgpu_info_firmware {
1398 	__u32 ver;
1399 	__u32 feature;
1400 };
1401 
1402 struct drm_amdgpu_info_vbios {
1403 	__u8 name[64];
1404 	__u8 vbios_pn[64];
1405 	__u32 version;
1406 	__u32 pad;
1407 	__u8 vbios_ver_str[32];
1408 	__u8 date[32];
1409 };
1410 
1411 #define AMDGPU_VRAM_TYPE_UNKNOWN 0
1412 #define AMDGPU_VRAM_TYPE_GDDR1 1
1413 #define AMDGPU_VRAM_TYPE_DDR2  2
1414 #define AMDGPU_VRAM_TYPE_GDDR3 3
1415 #define AMDGPU_VRAM_TYPE_GDDR4 4
1416 #define AMDGPU_VRAM_TYPE_GDDR5 5
1417 #define AMDGPU_VRAM_TYPE_HBM   6
1418 #define AMDGPU_VRAM_TYPE_DDR3  7
1419 #define AMDGPU_VRAM_TYPE_DDR4  8
1420 #define AMDGPU_VRAM_TYPE_GDDR6 9
1421 #define AMDGPU_VRAM_TYPE_DDR5  10
1422 #define AMDGPU_VRAM_TYPE_LPDDR4 11
1423 #define AMDGPU_VRAM_TYPE_LPDDR5 12
1424 #define AMDGPU_VRAM_TYPE_HBM3E 13
1425 
1426 struct drm_amdgpu_info_device {
1427 	/** PCI Device ID */
1428 	__u32 device_id;
1429 	/** Internal chip revision: A0, A1, etc.) */
1430 	__u32 chip_rev;
1431 	__u32 external_rev;
1432 	/** Revision id in PCI Config space */
1433 	__u32 pci_rev;
1434 	__u32 family;
1435 	__u32 num_shader_engines;
1436 	__u32 num_shader_arrays_per_engine;
1437 	/* in KHz */
1438 	__u32 gpu_counter_freq;
1439 	__u64 max_engine_clock;
1440 	__u64 max_memory_clock;
1441 	/* cu information */
1442 	__u32 cu_active_number;
1443 	/* NOTE: cu_ao_mask is INVALID, DON'T use it */
1444 	__u32 cu_ao_mask;
1445 	__u32 cu_bitmap[4][4];
1446 	/** Render backend pipe mask. One render backend is CB+DB. */
1447 	__u32 enabled_rb_pipes_mask;
1448 	__u32 num_rb_pipes;
1449 	__u32 num_hw_gfx_contexts;
1450 	/* PCIe version (the smaller of the GPU and the CPU/motherboard) */
1451 	__u32 pcie_gen;
1452 	__u64 ids_flags;
1453 	/** Starting virtual address for UMDs. */
1454 	__u64 virtual_address_offset;
1455 	/** The maximum virtual address */
1456 	__u64 virtual_address_max;
1457 	/** Required alignment of virtual addresses. */
1458 	__u32 virtual_address_alignment;
1459 	/** Page table entry - fragment size */
1460 	__u32 pte_fragment_size;
1461 	__u32 gart_page_size;
1462 	/** constant engine ram size*/
1463 	__u32 ce_ram_size;
1464 	/** video memory type info*/
1465 	__u32 vram_type;
1466 	/** video memory bit width*/
1467 	__u32 vram_bit_width;
1468 	/* vce harvesting instance */
1469 	__u32 vce_harvest_config;
1470 	/* gfx double offchip LDS buffers */
1471 	__u32 gc_double_offchip_lds_buf;
1472 	/* NGG Primitive Buffer */
1473 	__u64 prim_buf_gpu_addr;
1474 	/* NGG Position Buffer */
1475 	__u64 pos_buf_gpu_addr;
1476 	/* NGG Control Sideband */
1477 	__u64 cntl_sb_buf_gpu_addr;
1478 	/* NGG Parameter Cache */
1479 	__u64 param_buf_gpu_addr;
1480 	__u32 prim_buf_size;
1481 	__u32 pos_buf_size;
1482 	__u32 cntl_sb_buf_size;
1483 	__u32 param_buf_size;
1484 	/* wavefront size*/
1485 	__u32 wave_front_size;
1486 	/* shader visible vgprs*/
1487 	__u32 num_shader_visible_vgprs;
1488 	/* CU per shader array*/
1489 	__u32 num_cu_per_sh;
1490 	/* number of tcc blocks*/
1491 	__u32 num_tcc_blocks;
1492 	/* gs vgt table depth*/
1493 	__u32 gs_vgt_table_depth;
1494 	/* gs primitive buffer depth*/
1495 	__u32 gs_prim_buffer_depth;
1496 	/* max gs wavefront per vgt*/
1497 	__u32 max_gs_waves_per_vgt;
1498 	/* PCIe number of lanes (the smaller of the GPU and the CPU/motherboard) */
1499 	__u32 pcie_num_lanes;
1500 	/* always on cu bitmap */
1501 	__u32 cu_ao_bitmap[4][4];
1502 	/** Starting high virtual address for UMDs. */
1503 	__u64 high_va_offset;
1504 	/** The maximum high virtual address */
1505 	__u64 high_va_max;
1506 	/* gfx10 pa_sc_tile_steering_override */
1507 	__u32 pa_sc_tile_steering_override;
1508 	/* disabled TCCs */
1509 	__u64 tcc_disabled_mask;
1510 	__u64 min_engine_clock;
1511 	__u64 min_memory_clock;
1512 	/* The following fields are only set on gfx11+, older chips set 0. */
1513 	__u32 tcp_cache_size;       /* AKA GL0, VMEM cache */
1514 	__u32 num_sqc_per_wgp;
1515 	__u32 sqc_data_cache_size;  /* AKA SMEM cache */
1516 	__u32 sqc_inst_cache_size;
1517 	__u32 gl1c_cache_size;
1518 	__u32 gl2c_cache_size;
1519 	__u64 mall_size;            /* AKA infinity cache */
1520 	/* high 32 bits of the rb pipes mask */
1521 	__u32 enabled_rb_pipes_mask_hi;
1522 	/* shadow area size for gfx11 */
1523 	__u32 shadow_size;
1524 	/* shadow area base virtual alignment for gfx11 */
1525 	__u32 shadow_alignment;
1526 	/* context save area size for gfx11 */
1527 	__u32 csa_size;
1528 	/* context save area base virtual alignment for gfx11 */
1529 	__u32 csa_alignment;
1530 	/* Userq IP mask (1 << AMDGPU_HW_IP_*) */
1531 	__u32 userq_ip_mask;
1532 	__u32 pad;
1533 };
1534 
1535 struct drm_amdgpu_info_hw_ip {
1536 	/** Version of h/w IP */
1537 	__u32  hw_ip_version_major;
1538 	__u32  hw_ip_version_minor;
1539 	/** Capabilities */
1540 	__u64  capabilities_flags;
1541 	/** command buffer address start alignment*/
1542 	__u32  ib_start_alignment;
1543 	/** command buffer size alignment*/
1544 	__u32  ib_size_alignment;
1545 	/** Bitmask of available rings. Bit 0 means ring 0, etc. */
1546 	__u32  available_rings;
1547 	/** version info: bits 23:16 major, 15:8 minor, 7:0 revision */
1548 	__u32  ip_discovery_version;
1549 	/* Userq available slots */
1550 	__u32  userq_num_slots;
1551 };
1552 
1553 /* GFX metadata BO sizes and alignment info (in bytes) */
1554 struct drm_amdgpu_info_uq_fw_areas_gfx {
1555 	/* shadow area size */
1556 	__u32 shadow_size;
1557 	/* shadow area base virtual mem alignment */
1558 	__u32 shadow_alignment;
1559 	/* context save area size */
1560 	__u32 csa_size;
1561 	/* context save area base virtual mem alignment */
1562 	__u32 csa_alignment;
1563 };
1564 
1565 /* IP specific fw related information used in the
1566  * subquery AMDGPU_INFO_UQ_FW_AREAS
1567  */
1568 struct drm_amdgpu_info_uq_fw_areas {
1569 	union {
1570 		struct drm_amdgpu_info_uq_fw_areas_gfx gfx;
1571 	};
1572 };
1573 
1574 struct drm_amdgpu_info_num_handles {
1575 	/** Max handles as supported by firmware for UVD */
1576 	__u32  uvd_max_handles;
1577 	/** Handles currently in use for UVD */
1578 	__u32  uvd_used_handles;
1579 };
1580 
1581 #define AMDGPU_VCE_CLOCK_TABLE_ENTRIES		6
1582 
1583 struct drm_amdgpu_info_vce_clock_table_entry {
1584 	/** System clock */
1585 	__u32 sclk;
1586 	/** Memory clock */
1587 	__u32 mclk;
1588 	/** VCE clock */
1589 	__u32 eclk;
1590 	__u32 pad;
1591 };
1592 
1593 struct drm_amdgpu_info_vce_clock_table {
1594 	struct drm_amdgpu_info_vce_clock_table_entry entries[AMDGPU_VCE_CLOCK_TABLE_ENTRIES];
1595 	__u32 num_valid_entries;
1596 	__u32 pad;
1597 };
1598 
1599 /* query video encode/decode caps */
1600 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2			0
1601 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4			1
1602 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1			2
1603 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC		3
1604 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC			4
1605 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG			5
1606 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9			6
1607 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1			7
1608 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT			8
1609 
1610 struct drm_amdgpu_info_video_codec_info {
1611 	__u32 valid;
1612 	__u32 max_width;
1613 	__u32 max_height;
1614 	__u32 max_pixels_per_frame;
1615 	__u32 max_level;
1616 	__u32 pad;
1617 };
1618 
1619 struct drm_amdgpu_info_video_caps {
1620 	struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
1621 };
1622 
1623 #define AMDGPU_VMHUB_TYPE_MASK			0xff
1624 #define AMDGPU_VMHUB_TYPE_SHIFT			0
1625 #define AMDGPU_VMHUB_TYPE_GFX			0
1626 #define AMDGPU_VMHUB_TYPE_MM0			1
1627 #define AMDGPU_VMHUB_TYPE_MM1			2
1628 #define AMDGPU_VMHUB_IDX_MASK			0xff00
1629 #define AMDGPU_VMHUB_IDX_SHIFT			8
1630 
1631 struct drm_amdgpu_info_gpuvm_fault {
1632 	__u64 addr;
1633 	__u32 status;
1634 	__u32 vmhub;
1635 };
1636 
1637 struct drm_amdgpu_info_uq_metadata_gfx {
1638 	/* shadow area size for gfx11 */
1639 	__u32 shadow_size;
1640 	/* shadow area base virtual alignment for gfx11 */
1641 	__u32 shadow_alignment;
1642 	/* context save area size for gfx11 */
1643 	__u32 csa_size;
1644 	/* context save area base virtual alignment for gfx11 */
1645 	__u32 csa_alignment;
1646 };
1647 
1648 struct drm_amdgpu_info_uq_metadata {
1649 	union {
1650 		struct drm_amdgpu_info_uq_metadata_gfx gfx;
1651 	};
1652 };
1653 
1654 /*
1655  * Supported GPU families
1656  */
1657 #define AMDGPU_FAMILY_UNKNOWN			0
1658 #define AMDGPU_FAMILY_SI			110 /* Hainan, Oland, Verde, Pitcairn, Tahiti */
1659 #define AMDGPU_FAMILY_CI			120 /* Bonaire, Hawaii */
1660 #define AMDGPU_FAMILY_KV			125 /* Kaveri, Kabini, Mullins */
1661 #define AMDGPU_FAMILY_VI			130 /* Iceland, Tonga */
1662 #define AMDGPU_FAMILY_CZ			135 /* Carrizo, Stoney */
1663 #define AMDGPU_FAMILY_AI			141 /* Vega10 */
1664 #define AMDGPU_FAMILY_RV			142 /* Raven */
1665 #define AMDGPU_FAMILY_NV			143 /* Navi10 */
1666 #define AMDGPU_FAMILY_VGH			144 /* Van Gogh */
1667 #define AMDGPU_FAMILY_GC_11_0_0			145 /* GC 11.0.0 */
1668 #define AMDGPU_FAMILY_YC			146 /* Yellow Carp */
1669 #define AMDGPU_FAMILY_GC_11_0_1			148 /* GC 11.0.1 */
1670 #define AMDGPU_FAMILY_GC_10_3_6			149 /* GC 10.3.6 */
1671 #define AMDGPU_FAMILY_GC_10_3_7			151 /* GC 10.3.7 */
1672 #define AMDGPU_FAMILY_GC_11_5_0			150 /* GC 11.5.0 */
1673 #define AMDGPU_FAMILY_GC_12_0_0			152 /* GC 12.0.0 */
1674 
1675 /* FIXME wrong namespace! */
1676 struct drm_color_ctm_3x4 {
1677 	/*
1678 	 * Conversion matrix with 3x4 dimensions in S31.32 sign-magnitude
1679 	 * (not two's complement!) format.
1680 	 */
1681 	__u64 matrix[12];
1682 };
1683 
1684 #if defined(__cplusplus)
1685 }
1686 #endif
1687 
1688 #endif
1689