xref: /linux/include/uapi/drm/amdgpu_drm.h (revision 1b78070aaef63512688aebfbc82365ef9d6660f1)
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 #define DRM_AMDGPU_PROC_OPTIONS		0x1A
62 
63 #define DRM_IOCTL_AMDGPU_GEM_CREATE	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_CREATE, union drm_amdgpu_gem_create)
64 #define DRM_IOCTL_AMDGPU_GEM_MMAP	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_MMAP, union drm_amdgpu_gem_mmap)
65 #define DRM_IOCTL_AMDGPU_CTX		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_CTX, union drm_amdgpu_ctx)
66 #define DRM_IOCTL_AMDGPU_BO_LIST	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_BO_LIST, union drm_amdgpu_bo_list)
67 #define DRM_IOCTL_AMDGPU_CS		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_CS, union drm_amdgpu_cs)
68 #define DRM_IOCTL_AMDGPU_INFO		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_INFO, struct drm_amdgpu_info)
69 #define DRM_IOCTL_AMDGPU_GEM_METADATA	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_METADATA, struct drm_amdgpu_gem_metadata)
70 #define DRM_IOCTL_AMDGPU_GEM_WAIT_IDLE	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_WAIT_IDLE, union drm_amdgpu_gem_wait_idle)
71 #define DRM_IOCTL_AMDGPU_GEM_VA		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_VA, struct drm_amdgpu_gem_va)
72 #define DRM_IOCTL_AMDGPU_WAIT_CS	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_CS, union drm_amdgpu_wait_cs)
73 #define DRM_IOCTL_AMDGPU_GEM_OP		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_OP, struct drm_amdgpu_gem_op)
74 #define DRM_IOCTL_AMDGPU_GEM_USERPTR	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_USERPTR, struct drm_amdgpu_gem_userptr)
75 #define DRM_IOCTL_AMDGPU_WAIT_FENCES	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_FENCES, union drm_amdgpu_wait_fences)
76 #define DRM_IOCTL_AMDGPU_VM		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_VM, union drm_amdgpu_vm)
77 #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
78 #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
79 #define DRM_IOCTL_AMDGPU_USERQ		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_USERQ, union drm_amdgpu_userq)
80 #define DRM_IOCTL_AMDGPU_USERQ_SIGNAL	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_USERQ_SIGNAL, struct drm_amdgpu_userq_signal)
81 #define DRM_IOCTL_AMDGPU_USERQ_WAIT	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_USERQ_WAIT, struct drm_amdgpu_userq_wait)
82 #define DRM_IOCTL_AMDGPU_GEM_LIST_HANDLES DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_LIST_HANDLES, struct drm_amdgpu_gem_list_handles)
83 #define DRM_IOCTL_AMDGPU_PROC_OPTIONS	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_PROC_OPTIONS, struct drm_amdgpu_proc_options)
84 
85 /**
86  * DOC: memory domains
87  *
88  * %AMDGPU_GEM_DOMAIN_CPU	System memory that is not GPU accessible.
89  * Memory in this pool could be swapped out to disk if there is pressure.
90  *
91  * %AMDGPU_GEM_DOMAIN_GTT	GPU accessible system memory, mapped into the
92  * GPU's virtual address space via gart. Gart memory linearizes non-contiguous
93  * pages of system memory, allows GPU access system memory in a linearized
94  * fashion.
95  *
96  * %AMDGPU_GEM_DOMAIN_VRAM	Local video memory. For APUs, it is memory
97  * carved out by the BIOS.
98  *
99  * %AMDGPU_GEM_DOMAIN_GDS	Global on-chip data storage used to share data
100  * across shader threads.
101  *
102  * %AMDGPU_GEM_DOMAIN_GWS	Global wave sync, used to synchronize the
103  * execution of all the waves on a device.
104  *
105  * %AMDGPU_GEM_DOMAIN_OA	Ordered append, used by 3D or Compute engines
106  * for appending data.
107  *
108  * %AMDGPU_GEM_DOMAIN_DOORBELL	Doorbell. It is an MMIO region for
109  * signalling user mode queues.
110  */
111 #define AMDGPU_GEM_DOMAIN_CPU		0x1
112 #define AMDGPU_GEM_DOMAIN_GTT		0x2
113 #define AMDGPU_GEM_DOMAIN_VRAM		0x4
114 #define AMDGPU_GEM_DOMAIN_GDS		0x8
115 #define AMDGPU_GEM_DOMAIN_GWS		0x10
116 #define AMDGPU_GEM_DOMAIN_OA		0x20
117 #define AMDGPU_GEM_DOMAIN_DOORBELL	0x40
118 #define AMDGPU_GEM_DOMAIN_MASK		(AMDGPU_GEM_DOMAIN_CPU | \
119 					 AMDGPU_GEM_DOMAIN_GTT | \
120 					 AMDGPU_GEM_DOMAIN_VRAM | \
121 					 AMDGPU_GEM_DOMAIN_GDS | \
122 					 AMDGPU_GEM_DOMAIN_GWS | \
123 					 AMDGPU_GEM_DOMAIN_OA |	\
124 					 AMDGPU_GEM_DOMAIN_DOORBELL)
125 
126 /* Flag that CPU access will be required for the case of VRAM domain */
127 #define AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED	(1 << 0)
128 /* Flag that CPU access will not work, this VRAM domain is invisible */
129 #define AMDGPU_GEM_CREATE_NO_CPU_ACCESS		(1 << 1)
130 /* Flag that USWC attributes should be used for GTT */
131 #define AMDGPU_GEM_CREATE_CPU_GTT_USWC		(1 << 2)
132 /* Flag that the memory should be in VRAM and cleared */
133 #define AMDGPU_GEM_CREATE_VRAM_CLEARED		(1 << 3)
134 /* Flag that allocating the BO should use linear VRAM */
135 #define AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS	(1 << 5)
136 /* Flag that BO is always valid in this VM */
137 #define AMDGPU_GEM_CREATE_VM_ALWAYS_VALID	(1 << 6)
138 /* Flag that BO sharing will be explicitly synchronized */
139 #define AMDGPU_GEM_CREATE_EXPLICIT_SYNC		(1 << 7)
140 /* Flag that indicates allocating MQD gart on GFX9, where the mtype
141  * for the second page onward should be set to NC. It should never
142  * be used by user space applications.
143  */
144 #define AMDGPU_GEM_CREATE_CP_MQD_GFX9		(1 << 8)
145 /* Flag that BO may contain sensitive data that must be wiped before
146  * releasing the memory
147  */
148 #define AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE	(1 << 9)
149 /* Flag that BO will be encrypted and that the TMZ bit should be
150  * set in the PTEs when mapping this buffer via GPUVM or
151  * accessing it with various hw blocks
152  */
153 #define AMDGPU_GEM_CREATE_ENCRYPTED		(1 << 10)
154 /* Flag that BO will be used only in preemptible context, which does
155  * not require GTT memory accounting
156  */
157 #define AMDGPU_GEM_CREATE_PREEMPTIBLE		(1 << 11)
158 /* Flag that BO can be discarded under memory pressure without keeping the
159  * content.
160  */
161 #define AMDGPU_GEM_CREATE_DISCARDABLE		(1 << 12)
162 /* Flag that BO is shared coherently between multiple devices or CPU threads.
163  * May depend on GPU instructions to flush caches to system scope explicitly.
164  *
165  * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
166  * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
167  */
168 #define AMDGPU_GEM_CREATE_COHERENT		(1 << 13)
169 /* Flag that BO should not be cached by GPU. Coherent without having to flush
170  * GPU caches explicitly
171  *
172  * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
173  * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
174  */
175 #define AMDGPU_GEM_CREATE_UNCACHED		(1 << 14)
176 /* Flag that BO should be coherent across devices when using device-level
177  * atomics. May depend on GPU instructions to flush caches to device scope
178  * explicitly, promoting them to system scope automatically.
179  *
180  * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and
181  * may override the MTYPE selected in AMDGPU_VA_OP_MAP.
182  */
183 #define AMDGPU_GEM_CREATE_EXT_COHERENT		(1 << 15)
184 /* Set PTE.D and recompress during GTT->VRAM moves according to TILING flags. */
185 #define AMDGPU_GEM_CREATE_GFX12_DCC		(1 << 16)
186 
187 struct drm_amdgpu_gem_create_in  {
188 	/** the requested memory size */
189 	__u64 bo_size;
190 	/** physical start_addr alignment in bytes for some HW requirements */
191 	__u64 alignment;
192 	/** the requested memory domains */
193 	__u64 domains;
194 	/** allocation flags */
195 	__u64 domain_flags;
196 };
197 
198 struct drm_amdgpu_gem_create_out  {
199 	/** returned GEM object handle */
200 	__u32 handle;
201 	__u32 _pad;
202 };
203 
204 union drm_amdgpu_gem_create {
205 	struct drm_amdgpu_gem_create_in		in;
206 	struct drm_amdgpu_gem_create_out	out;
207 };
208 
209 /** Opcode to create new residency list.  */
210 #define AMDGPU_BO_LIST_OP_CREATE	0
211 /** Opcode to destroy previously created residency list */
212 #define AMDGPU_BO_LIST_OP_DESTROY	1
213 /** Opcode to update resource information in the list */
214 #define AMDGPU_BO_LIST_OP_UPDATE	2
215 
216 struct drm_amdgpu_bo_list_in {
217 	/** Type of operation */
218 	__u32 operation;
219 	/** Handle of list or 0 if we want to create one */
220 	__u32 list_handle;
221 	/** Number of BOs in list  */
222 	__u32 bo_number;
223 	/** Size of each element describing BO */
224 	__u32 bo_info_size;
225 	/** Pointer to array describing BOs */
226 	__u64 bo_info_ptr;
227 };
228 
229 struct drm_amdgpu_bo_list_entry {
230 	/** Handle of BO */
231 	__u32 bo_handle;
232 	/** New (if specified) BO priority to be used during migration */
233 	__u32 bo_priority;
234 };
235 
236 struct drm_amdgpu_bo_list_out {
237 	/** Handle of resource list  */
238 	__u32 list_handle;
239 	__u32 _pad;
240 };
241 
242 union drm_amdgpu_bo_list {
243 	struct drm_amdgpu_bo_list_in in;
244 	struct drm_amdgpu_bo_list_out out;
245 };
246 
247 /* context related */
248 #define AMDGPU_CTX_OP_ALLOC_CTX	1
249 #define AMDGPU_CTX_OP_FREE_CTX	2
250 #define AMDGPU_CTX_OP_QUERY_STATE	3
251 #define AMDGPU_CTX_OP_QUERY_STATE2	4
252 #define AMDGPU_CTX_OP_GET_STABLE_PSTATE	5
253 #define AMDGPU_CTX_OP_SET_STABLE_PSTATE	6
254 
255 /* GPU reset status */
256 #define AMDGPU_CTX_NO_RESET		0
257 /* this the context caused it */
258 #define AMDGPU_CTX_GUILTY_RESET		1
259 /* some other context caused it */
260 #define AMDGPU_CTX_INNOCENT_RESET	2
261 /* unknown cause */
262 #define AMDGPU_CTX_UNKNOWN_RESET	3
263 
264 /* indicate gpu reset occurred after ctx created */
265 #define AMDGPU_CTX_QUERY2_FLAGS_RESET    (1<<0)
266 /* indicate vram lost occurred after ctx created */
267 #define AMDGPU_CTX_QUERY2_FLAGS_VRAMLOST (1<<1)
268 /* indicate some job from this context once cause gpu hang */
269 #define AMDGPU_CTX_QUERY2_FLAGS_GUILTY   (1<<2)
270 /* indicate some errors are detected by RAS */
271 #define AMDGPU_CTX_QUERY2_FLAGS_RAS_CE   (1<<3)
272 #define AMDGPU_CTX_QUERY2_FLAGS_RAS_UE   (1<<4)
273 /* indicate that the reset hasn't completed yet */
274 #define AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS (1<<5)
275 
276 /* Context priority level */
277 #define AMDGPU_CTX_PRIORITY_UNSET       -2048
278 #define AMDGPU_CTX_PRIORITY_VERY_LOW    -1023
279 #define AMDGPU_CTX_PRIORITY_LOW         -512
280 #define AMDGPU_CTX_PRIORITY_NORMAL      0
281 /*
282  * When used in struct drm_amdgpu_ctx_in, a priority above NORMAL requires
283  * CAP_SYS_NICE or DRM_MASTER
284 */
285 #define AMDGPU_CTX_PRIORITY_HIGH        512
286 #define AMDGPU_CTX_PRIORITY_VERY_HIGH   1023
287 
288 /* select a stable profiling pstate for perfmon tools */
289 #define AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK  0xf
290 #define AMDGPU_CTX_STABLE_PSTATE_NONE  0
291 #define AMDGPU_CTX_STABLE_PSTATE_STANDARD  1
292 #define AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK  2
293 #define AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK  3
294 #define AMDGPU_CTX_STABLE_PSTATE_PEAK  4
295 
296 struct drm_amdgpu_ctx_in {
297 	/** AMDGPU_CTX_OP_* */
298 	__u32	op;
299 	/** Flags */
300 	__u32	flags;
301 	__u32	ctx_id;
302 	/** AMDGPU_CTX_PRIORITY_* */
303 	__s32	priority;
304 };
305 
306 union drm_amdgpu_ctx_out {
307 		struct {
308 			__u32	ctx_id;
309 			__u32	_pad;
310 		} alloc;
311 
312 		struct {
313 			/** For future use, no flags defined so far */
314 			__u64	flags;
315 			/** Number of resets caused by this context so far. */
316 			__u32	hangs;
317 			/** Reset status since the last call of the ioctl. */
318 			__u32	reset_status;
319 		} state;
320 
321 		struct {
322 			__u32	flags;
323 			__u32	_pad;
324 		} pstate;
325 };
326 
327 union drm_amdgpu_ctx {
328 	struct drm_amdgpu_ctx_in in;
329 	union drm_amdgpu_ctx_out out;
330 };
331 
332 /* user queue IOCTL operations */
333 #define AMDGPU_USERQ_OP_CREATE	1
334 #define AMDGPU_USERQ_OP_FREE	2
335 
336 /* queue priority levels */
337 /* low < normal low < normal high < high */
338 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_MASK  0x3
339 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_SHIFT 0
340 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_NORMAL_LOW 0
341 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_LOW 1
342 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_NORMAL_HIGH 2
343 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_PRIORITY_HIGH 3 /* admin only */
344 /* for queues that need access to protected content */
345 #define AMDGPU_USERQ_CREATE_FLAGS_QUEUE_SECURE  (1 << 2)
346 
347 /*
348  * This structure is a container to pass input configuration
349  * info for all supported userqueue related operations.
350  * For operation AMDGPU_USERQ_OP_CREATE: user is expected
351  *  to set all fields, excep the parameter 'queue_id'.
352  * For operation AMDGPU_USERQ_OP_FREE: the only input parameter expected
353  *  to be set is 'queue_id', eveything else is ignored.
354  */
355 struct drm_amdgpu_userq_in {
356 	/** AMDGPU_USERQ_OP_* */
357 	__u32	op;
358 	/** Queue id passed for operation USERQ_OP_FREE */
359 	__u32	queue_id;
360 	/** the target GPU engine to execute workload (AMDGPU_HW_IP_*) */
361 	__u32   ip_type;
362 	/**
363 	 * @doorbell_handle: the handle of doorbell GEM object
364 	 * associated with this userqueue client.
365 	 */
366 	__u32   doorbell_handle;
367 	/**
368 	 * @doorbell_offset: 32-bit offset of the doorbell in the doorbell bo.
369 	 * Kernel will generate absolute doorbell offset using doorbell_handle
370 	 * and doorbell_offset in the doorbell bo.
371 	 */
372 	__u32   doorbell_offset;
373 	/**
374 	 * @flags: flags used for queue parameters
375 	 */
376 	__u32 flags;
377 	/**
378 	 * @queue_va: Virtual address of the GPU memory which holds the queue
379 	 * object. The queue holds the workload packets.
380 	 */
381 	__u64   queue_va;
382 	/**
383 	 * @queue_size: Size of the queue in bytes, this needs to be 256-byte
384 	 * aligned.
385 	 */
386 	__u64   queue_size;
387 	/**
388 	 * @rptr_va : Virtual address of the GPU memory which holds the ring RPTR.
389 	 * This object must be at least 8 byte in size and aligned to 8-byte offset.
390 	 */
391 	__u64   rptr_va;
392 	/**
393 	 * @wptr_va : Virtual address of the GPU memory which holds the ring WPTR.
394 	 * This object must be at least 8 byte in size and aligned to 8-byte offset.
395 	 *
396 	 * Queue, RPTR and WPTR can come from the same object, as long as the size
397 	 * and alignment related requirements are met.
398 	 */
399 	__u64   wptr_va;
400 	/**
401 	 * @mqd: MQD (memory queue descriptor) is a set of parameters which allow
402 	 * the GPU to uniquely define and identify a usermode queue.
403 	 *
404 	 * MQD data can be of different size for different GPU IP/engine and
405 	 * their respective versions/revisions, so this points to a __u64 *
406 	 * which holds IP specific MQD of this usermode queue.
407 	 */
408 	__u64 mqd;
409 	/**
410 	 * @size: size of MQD data in bytes, it must match the MQD structure
411 	 * size of the respective engine/revision defined in UAPI for ex, for
412 	 * gfx11 workloads, size = sizeof(drm_amdgpu_userq_mqd_gfx11).
413 	 */
414 	__u64 mqd_size;
415 };
416 
417 /* The structure to carry output of userqueue ops */
418 struct drm_amdgpu_userq_out {
419 	/**
420 	 * For operation AMDGPU_USERQ_OP_CREATE: This field contains a unique
421 	 * queue ID to represent the newly created userqueue in the system, otherwise
422 	 * it should be ignored.
423 	 */
424 	__u32	queue_id;
425 	__u32 _pad;
426 };
427 
428 union drm_amdgpu_userq {
429 	struct drm_amdgpu_userq_in in;
430 	struct drm_amdgpu_userq_out out;
431 };
432 
433 /* GFX V11 IP specific MQD parameters */
434 struct drm_amdgpu_userq_mqd_gfx11 {
435 	/**
436 	 * @shadow_va: Virtual address of the GPU memory to hold the shadow buffer.
437 	 * Use AMDGPU_INFO_IOCTL to find the exact size of the object.
438 	 */
439 	__u64   shadow_va;
440 	/**
441 	 * @csa_va: Virtual address of the GPU memory to hold the CSA buffer.
442 	 * Use AMDGPU_INFO_IOCTL to find the exact size of the object.
443 	 */
444 	__u64   csa_va;
445 };
446 
447 /* GFX V11 SDMA IP specific MQD parameters */
448 struct drm_amdgpu_userq_mqd_sdma_gfx11 {
449 	/**
450 	 * @csa_va: Virtual address of the GPU memory to hold the CSA buffer.
451 	 * This must be a from a separate GPU object, and use AMDGPU_INFO IOCTL
452 	 * to get the size.
453 	 */
454 	__u64   csa_va;
455 };
456 
457 /* GFX V11 Compute IP specific MQD parameters */
458 struct drm_amdgpu_userq_mqd_compute_gfx11 {
459 	/**
460 	 * @eop_va: Virtual address of the GPU memory to hold the EOP buffer.
461 	 * This must be a from a separate GPU object, and use AMDGPU_INFO IOCTL
462 	 * to get the size.
463 	 */
464 	__u64   eop_va;
465 };
466 
467 /* userq signal/wait ioctl */
468 struct drm_amdgpu_userq_signal {
469 	/**
470 	 * @queue_id: Queue handle used by the userq fence creation function
471 	 * to retrieve the WPTR.
472 	 */
473 	__u32	queue_id;
474 	__u32	pad;
475 	/**
476 	 * @syncobj_handles: The list of syncobj handles submitted by the user queue
477 	 * job to be signaled.
478 	 */
479 	__u64	syncobj_handles;
480 	/**
481 	 * @num_syncobj_handles: A count that represents the number of syncobj handles in
482 	 * @syncobj_handles.
483 	 */
484 	__u16	num_syncobj_handles;
485 	__u16	pad0;
486 	__u32	pad1;
487 	/**
488 	 * @bo_read_handles: The list of BO handles that the submitted user queue job
489 	 * is using for read only. This will update BO fences in the kernel.
490 	 */
491 	__u64	bo_read_handles;
492 	/**
493 	 * @bo_write_handles: The list of BO handles that the submitted user queue job
494 	 * is using for write only. This will update BO fences in the kernel.
495 	 */
496 	__u64	bo_write_handles;
497 	/**
498 	 * @num_bo_read_handles: A count that represents the number of read BO handles in
499 	 * @bo_read_handles.
500 	 */
501 	__u32	num_bo_read_handles;
502 	/**
503 	 * @num_bo_write_handles: A count that represents the number of write BO handles in
504 	 * @bo_write_handles.
505 	 */
506 	__u32	num_bo_write_handles;
507 };
508 
509 struct drm_amdgpu_userq_fence_info {
510 	/**
511 	 * @va: A gpu address allocated for each queue which stores the
512 	 * read pointer (RPTR) value.
513 	 */
514 	__u64	va;
515 	/**
516 	 * @value: A 64 bit value represents the write pointer (WPTR) of the
517 	 * queue commands which compared with the RPTR value to signal the
518 	 * fences.
519 	 */
520 	__u64	value;
521 };
522 
523 struct drm_amdgpu_userq_wait {
524 	/**
525 	 * @waitq_id: Queue handle used by the userq wait IOCTL to retrieve the
526 	 * wait queue and maintain the fence driver references in it.
527 	 */
528 	__u32	waitq_id;
529 	__u32	pad;
530 	/**
531 	 * @syncobj_handles: The list of syncobj handles submitted by the user queue
532 	 * job to get the va/value pairs.
533 	 */
534 	__u64	syncobj_handles;
535 	/**
536 	 * @syncobj_timeline_handles: The list of timeline syncobj handles submitted by
537 	 * the user queue job to get the va/value pairs at given @syncobj_timeline_points.
538 	 */
539 	__u64	syncobj_timeline_handles;
540 	/**
541 	 * @syncobj_timeline_points: The list of timeline syncobj points submitted by the
542 	 * user queue job for the corresponding @syncobj_timeline_handles.
543 	 */
544 	__u64	syncobj_timeline_points;
545 	/**
546 	 * @bo_read_handles: The list of read BO handles submitted by the user queue
547 	 * job to get the va/value pairs.
548 	 */
549 	__u64	bo_read_handles;
550 	/**
551 	 * @bo_write_handles: The list of write BO handles submitted by the user queue
552 	 * job to get the va/value pairs.
553 	 */
554 	__u64	bo_write_handles;
555 	/**
556 	 * @num_syncobj_timeline_handles: A count that represents the number of timeline
557 	 * syncobj handles in @syncobj_timeline_handles.
558 	 */
559 	__u16	num_syncobj_timeline_handles;
560 	/**
561 	 * @num_fences: This field can be used both as input and output. As input it defines
562 	 * the maximum number of fences that can be returned and as output it will specify
563 	 * how many fences were actually returned from the ioctl.
564 	 */
565 	__u16	num_fences;
566 	/**
567 	 * @num_syncobj_handles: A count that represents the number of syncobj handles in
568 	 * @syncobj_handles.
569 	 */
570 	__u16	num_syncobj_handles;
571 	__u16	pad0;
572 	/**
573 	 * @num_bo_read_handles: A count that represents the number of read BO handles in
574 	 * @bo_read_handles.
575 	 */
576 	__u32	num_bo_read_handles;
577 	/**
578 	 * @num_bo_write_handles: A count that represents the number of write BO handles in
579 	 * @bo_write_handles.
580 	 */
581 	__u32	num_bo_write_handles;
582 	/**
583 	 * @out_fences: The field is a return value from the ioctl containing the list of
584 	 * address/value pairs to wait for.
585 	 */
586 	__u64	out_fences;
587 };
588 
589 /* vm ioctl */
590 #define AMDGPU_VM_OP_RESERVE_VMID	1
591 #define AMDGPU_VM_OP_UNRESERVE_VMID	2
592 
593 struct drm_amdgpu_vm_in {
594 	/** AMDGPU_VM_OP_* */
595 	__u32	op;
596 	__u32	flags;
597 };
598 
599 struct drm_amdgpu_vm_out {
600 	/** For future use, no flags defined so far */
601 	__u64	flags;
602 };
603 
604 union drm_amdgpu_vm {
605 	struct drm_amdgpu_vm_in in;
606 	struct drm_amdgpu_vm_out out;
607 };
608 
609 /* sched ioctl */
610 #define AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE	1
611 #define AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE	2
612 
613 struct drm_amdgpu_sched_in {
614 	/* AMDGPU_SCHED_OP_* */
615 	__u32	op;
616 	__u32	fd;
617 	/** AMDGPU_CTX_PRIORITY_* */
618 	__s32	priority;
619 	__u32   ctx_id;
620 };
621 
622 union drm_amdgpu_sched {
623 	struct drm_amdgpu_sched_in in;
624 };
625 
626 /*
627  * This is not a reliable API and you should expect it to fail for any
628  * number of reasons and have fallback path that do not use userptr to
629  * perform any operation.
630  */
631 #define AMDGPU_GEM_USERPTR_READONLY	(1 << 0)
632 #define AMDGPU_GEM_USERPTR_ANONONLY	(1 << 1)
633 #define AMDGPU_GEM_USERPTR_VALIDATE	(1 << 2)
634 #define AMDGPU_GEM_USERPTR_REGISTER	(1 << 3)
635 
636 struct drm_amdgpu_gem_userptr {
637 	__u64		addr;
638 	__u64		size;
639 	/* AMDGPU_GEM_USERPTR_* */
640 	__u32		flags;
641 	/* Resulting GEM handle */
642 	__u32		handle;
643 };
644 
645 /* SI-CI-VI: */
646 /* same meaning as the GB_TILE_MODE and GL_MACRO_TILE_MODE fields */
647 #define AMDGPU_TILING_ARRAY_MODE_SHIFT			0
648 #define AMDGPU_TILING_ARRAY_MODE_MASK			0xf
649 #define AMDGPU_TILING_PIPE_CONFIG_SHIFT			4
650 #define AMDGPU_TILING_PIPE_CONFIG_MASK			0x1f
651 #define AMDGPU_TILING_TILE_SPLIT_SHIFT			9
652 #define AMDGPU_TILING_TILE_SPLIT_MASK			0x7
653 #define AMDGPU_TILING_MICRO_TILE_MODE_SHIFT		12
654 #define AMDGPU_TILING_MICRO_TILE_MODE_MASK		0x7
655 #define AMDGPU_TILING_BANK_WIDTH_SHIFT			15
656 #define AMDGPU_TILING_BANK_WIDTH_MASK			0x3
657 #define AMDGPU_TILING_BANK_HEIGHT_SHIFT			17
658 #define AMDGPU_TILING_BANK_HEIGHT_MASK			0x3
659 #define AMDGPU_TILING_MACRO_TILE_ASPECT_SHIFT		19
660 #define AMDGPU_TILING_MACRO_TILE_ASPECT_MASK		0x3
661 #define AMDGPU_TILING_NUM_BANKS_SHIFT			21
662 #define AMDGPU_TILING_NUM_BANKS_MASK			0x3
663 
664 /* GFX9 - GFX11: */
665 #define AMDGPU_TILING_SWIZZLE_MODE_SHIFT		0
666 #define AMDGPU_TILING_SWIZZLE_MODE_MASK			0x1f
667 #define AMDGPU_TILING_DCC_OFFSET_256B_SHIFT		5
668 #define AMDGPU_TILING_DCC_OFFSET_256B_MASK		0xFFFFFF
669 #define AMDGPU_TILING_DCC_PITCH_MAX_SHIFT		29
670 #define AMDGPU_TILING_DCC_PITCH_MAX_MASK		0x3FFF
671 #define AMDGPU_TILING_DCC_INDEPENDENT_64B_SHIFT		43
672 #define AMDGPU_TILING_DCC_INDEPENDENT_64B_MASK		0x1
673 #define AMDGPU_TILING_DCC_INDEPENDENT_128B_SHIFT	44
674 #define AMDGPU_TILING_DCC_INDEPENDENT_128B_MASK		0x1
675 #define AMDGPU_TILING_SCANOUT_SHIFT			63
676 #define AMDGPU_TILING_SCANOUT_MASK			0x1
677 
678 /* GFX12 and later: */
679 #define AMDGPU_TILING_GFX12_SWIZZLE_MODE_SHIFT			0
680 #define AMDGPU_TILING_GFX12_SWIZZLE_MODE_MASK			0x7
681 /* These are DCC recompression settings for memory management: */
682 #define AMDGPU_TILING_GFX12_DCC_MAX_COMPRESSED_BLOCK_SHIFT	3
683 #define AMDGPU_TILING_GFX12_DCC_MAX_COMPRESSED_BLOCK_MASK	0x3 /* 0:64B, 1:128B, 2:256B */
684 #define AMDGPU_TILING_GFX12_DCC_NUMBER_TYPE_SHIFT		5
685 #define AMDGPU_TILING_GFX12_DCC_NUMBER_TYPE_MASK		0x7 /* CB_COLOR0_INFO.NUMBER_TYPE */
686 #define AMDGPU_TILING_GFX12_DCC_DATA_FORMAT_SHIFT		8
687 #define AMDGPU_TILING_GFX12_DCC_DATA_FORMAT_MASK		0x3f /* [0:4]:CB_COLOR0_INFO.FORMAT, [5]:MM */
688 /* When clearing the buffer or moving it from VRAM to GTT, don't compress and set DCC metadata
689  * to uncompressed. Set when parts of an allocation bypass DCC and read raw data. */
690 #define AMDGPU_TILING_GFX12_DCC_WRITE_COMPRESS_DISABLE_SHIFT	14
691 #define AMDGPU_TILING_GFX12_DCC_WRITE_COMPRESS_DISABLE_MASK	0x1
692 /* bit gap */
693 #define AMDGPU_TILING_GFX12_SCANOUT_SHIFT			63
694 #define AMDGPU_TILING_GFX12_SCANOUT_MASK			0x1
695 
696 /* Set/Get helpers for tiling flags. */
697 #define AMDGPU_TILING_SET(field, value) \
698 	(((__u64)(value) & AMDGPU_TILING_##field##_MASK) << AMDGPU_TILING_##field##_SHIFT)
699 #define AMDGPU_TILING_GET(value, field) \
700 	(((__u64)(value) >> AMDGPU_TILING_##field##_SHIFT) & AMDGPU_TILING_##field##_MASK)
701 
702 #define AMDGPU_GEM_METADATA_OP_SET_METADATA                  1
703 #define AMDGPU_GEM_METADATA_OP_GET_METADATA                  2
704 
705 /** The same structure is shared for input/output */
706 struct drm_amdgpu_gem_metadata {
707 	/** GEM Object handle */
708 	__u32	handle;
709 	/** Do we want get or set metadata */
710 	__u32	op;
711 	struct {
712 		/** For future use, no flags defined so far */
713 		__u64	flags;
714 		/** family specific tiling info */
715 		__u64	tiling_info;
716 		__u32	data_size_bytes;
717 		__u32	data[64];
718 	} data;
719 };
720 
721 struct drm_amdgpu_gem_mmap_in {
722 	/** the GEM object handle */
723 	__u32 handle;
724 	__u32 _pad;
725 };
726 
727 struct drm_amdgpu_gem_mmap_out {
728 	/** mmap offset from the vma offset manager */
729 	__u64 addr_ptr;
730 };
731 
732 union drm_amdgpu_gem_mmap {
733 	struct drm_amdgpu_gem_mmap_in   in;
734 	struct drm_amdgpu_gem_mmap_out out;
735 };
736 
737 struct drm_amdgpu_gem_wait_idle_in {
738 	/** GEM object handle */
739 	__u32 handle;
740 	/** For future use, no flags defined so far */
741 	__u32 flags;
742 	/** Absolute timeout to wait */
743 	__u64 timeout;
744 };
745 
746 struct drm_amdgpu_gem_wait_idle_out {
747 	/** BO status:  0 - BO is idle, 1 - BO is busy */
748 	__u32 status;
749 	/** Returned current memory domain */
750 	__u32 domain;
751 };
752 
753 union drm_amdgpu_gem_wait_idle {
754 	struct drm_amdgpu_gem_wait_idle_in  in;
755 	struct drm_amdgpu_gem_wait_idle_out out;
756 };
757 
758 struct drm_amdgpu_wait_cs_in {
759 	/* Command submission handle
760          * handle equals 0 means none to wait for
761          * handle equals ~0ull means wait for the latest sequence number
762          */
763 	__u64 handle;
764 	/** Absolute timeout to wait */
765 	__u64 timeout;
766 	__u32 ip_type;
767 	__u32 ip_instance;
768 	__u32 ring;
769 	__u32 ctx_id;
770 };
771 
772 struct drm_amdgpu_wait_cs_out {
773 	/** CS status:  0 - CS completed, 1 - CS still busy */
774 	__u64 status;
775 };
776 
777 union drm_amdgpu_wait_cs {
778 	struct drm_amdgpu_wait_cs_in in;
779 	struct drm_amdgpu_wait_cs_out out;
780 };
781 
782 struct drm_amdgpu_fence {
783 	__u32 ctx_id;
784 	__u32 ip_type;
785 	__u32 ip_instance;
786 	__u32 ring;
787 	__u64 seq_no;
788 };
789 
790 struct drm_amdgpu_wait_fences_in {
791 	/** This points to uint64_t * which points to fences */
792 	__u64 fences;
793 	__u32 fence_count;
794 	__u32 wait_all;
795 	__u64 timeout_ns;
796 };
797 
798 struct drm_amdgpu_wait_fences_out {
799 	__u32 status;
800 	__u32 first_signaled;
801 };
802 
803 union drm_amdgpu_wait_fences {
804 	struct drm_amdgpu_wait_fences_in in;
805 	struct drm_amdgpu_wait_fences_out out;
806 };
807 
808 #define AMDGPU_GEM_OP_GET_GEM_CREATE_INFO	0
809 #define AMDGPU_GEM_OP_SET_PLACEMENT		1
810 #define AMDGPU_GEM_OP_GET_MAPPING_INFO		2
811 
812 struct drm_amdgpu_gem_vm_entry {
813 	/* Start of mapping (in bytes) */
814 	__u64 addr;
815 
816 	/* Size of mapping (in bytes) */
817 	__u64 size;
818 
819 	/* Mapping offset */
820 	__u64 offset;
821 
822 	/* flags needed to recreate mapping */
823 	__u64 flags;
824 };
825 
826 /* Sets or returns a value associated with a buffer. */
827 struct drm_amdgpu_gem_op {
828 	/** GEM object handle */
829 	__u32	handle;
830 	/** AMDGPU_GEM_OP_* */
831 	__u32	op;
832 	/** Input or return value. For MAPPING_INFO op: pointer to array of struct drm_amdgpu_gem_vm_entry */
833 	__u64	value;
834 	/** For MAPPING_INFO op: number of mappings (in/out) */
835 	__u32	num_entries;
836 
837 	__u32	padding;
838 };
839 
840 #define AMDGPU_GEM_LIST_HANDLES_FLAG_IS_IMPORT	(1 << 0)
841 
842 struct drm_amdgpu_gem_list_handles {
843 	/* User pointer to array of drm_amdgpu_gem_bo_info_entry */
844 	__u64   entries;
845 
846 	/* Size of entries buffer / Number of handles in process (if larger than size of buffer, must retry) */
847 	__u32   num_entries;
848 
849 	__u32 padding;
850 };
851 
852 struct drm_amdgpu_gem_list_handles_entry {
853 	/* gem handle of buffer object */
854 	__u32 gem_handle;
855 
856 	/* Currently just one flag: IS_IMPORT */
857 	__u32 flags;
858 
859 	/* Size of bo */
860 	__u64 size;
861 
862 	/* Preferred domains for GEM_CREATE */
863 	__u64 preferred_domains;
864 
865 	/* GEM_CREATE flags for re-creation of buffer */
866 	__u64 alloc_flags;
867 
868 	/* physical start_addr alignment in bytes for some HW requirements */
869 	__u64 alignment;
870 };
871 
872 #define AMDGPU_VA_OP_MAP			1
873 #define AMDGPU_VA_OP_UNMAP			2
874 #define AMDGPU_VA_OP_CLEAR			3
875 #define AMDGPU_VA_OP_REPLACE			4
876 
877 /* Delay the page table update till the next CS */
878 #define AMDGPU_VM_DELAY_UPDATE		(1 << 0)
879 
880 /* Mapping flags */
881 /* readable mapping */
882 #define AMDGPU_VM_PAGE_READABLE		(1 << 1)
883 /* writable mapping */
884 #define AMDGPU_VM_PAGE_WRITEABLE	(1 << 2)
885 /* executable mapping, new for VI */
886 #define AMDGPU_VM_PAGE_EXECUTABLE	(1 << 3)
887 /* unmapped page of partially resident textures */
888 #define AMDGPU_VM_PAGE_PRT		(1 << 4)
889 /* MTYPE flags use bit 5 to 8 */
890 #define AMDGPU_VM_MTYPE_MASK		(0xf << 5)
891 /* Default MTYPE. Pre-AI must use this.  Recommended for newer ASICs. */
892 #define AMDGPU_VM_MTYPE_DEFAULT		(0 << 5)
893 /* Use Non Coherent MTYPE instead of default MTYPE */
894 #define AMDGPU_VM_MTYPE_NC		(1 << 5)
895 /* Use Write Combine MTYPE instead of default MTYPE */
896 #define AMDGPU_VM_MTYPE_WC		(2 << 5)
897 /* Use Cache Coherent MTYPE instead of default MTYPE */
898 #define AMDGPU_VM_MTYPE_CC		(3 << 5)
899 /* Use UnCached MTYPE instead of default MTYPE */
900 #define AMDGPU_VM_MTYPE_UC		(4 << 5)
901 /* Use Read Write MTYPE instead of default MTYPE */
902 #define AMDGPU_VM_MTYPE_RW		(5 << 5)
903 /* don't allocate MALL */
904 #define AMDGPU_VM_PAGE_NOALLOC		(1 << 9)
905 
906 struct drm_amdgpu_gem_va {
907 	/** GEM object handle */
908 	__u32 handle;
909 	__u32 _pad;
910 	/** AMDGPU_VA_OP_* */
911 	__u32 operation;
912 	/** AMDGPU_VM_PAGE_* */
913 	__u32 flags;
914 	/** va address to assign . Must be correctly aligned.*/
915 	__u64 va_address;
916 	/** Specify offset inside of BO to assign. Must be correctly aligned.*/
917 	__u64 offset_in_bo;
918 	/** Specify mapping size. Must be correctly aligned. */
919 	__u64 map_size;
920 	/**
921 	 * vm_timeline_point is a sequence number used to add new timeline point.
922 	 */
923 	__u64 vm_timeline_point;
924 	/**
925 	 * The vm page table update fence is installed in given vm_timeline_syncobj_out
926 	 * at vm_timeline_point.
927 	 */
928 	__u32 vm_timeline_syncobj_out;
929 	/** the number of syncobj handles in @input_fence_syncobj_handles */
930 	__u32 num_syncobj_handles;
931 	/** Array of sync object handle to wait for given input fences */
932 	__u64 input_fence_syncobj_handles;
933 };
934 
935 #define AMDGPU_HW_IP_GFX          0
936 #define AMDGPU_HW_IP_COMPUTE      1
937 #define AMDGPU_HW_IP_DMA          2
938 #define AMDGPU_HW_IP_UVD          3
939 #define AMDGPU_HW_IP_VCE          4
940 #define AMDGPU_HW_IP_UVD_ENC      5
941 #define AMDGPU_HW_IP_VCN_DEC      6
942 /*
943  * From VCN4, AMDGPU_HW_IP_VCN_ENC is re-used to support
944  * both encoding and decoding jobs.
945  */
946 #define AMDGPU_HW_IP_VCN_ENC      7
947 #define AMDGPU_HW_IP_VCN_JPEG     8
948 #define AMDGPU_HW_IP_VPE          9
949 #define AMDGPU_HW_IP_NUM          10
950 
951 #define AMDGPU_HW_IP_INSTANCE_MAX_COUNT 1
952 
953 #define AMDGPU_CHUNK_ID_IB		0x01
954 #define AMDGPU_CHUNK_ID_FENCE		0x02
955 #define AMDGPU_CHUNK_ID_DEPENDENCIES	0x03
956 #define AMDGPU_CHUNK_ID_SYNCOBJ_IN      0x04
957 #define AMDGPU_CHUNK_ID_SYNCOBJ_OUT     0x05
958 #define AMDGPU_CHUNK_ID_BO_HANDLES      0x06
959 #define AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES	0x07
960 #define AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT    0x08
961 #define AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL  0x09
962 #define AMDGPU_CHUNK_ID_CP_GFX_SHADOW   0x0a
963 
964 struct drm_amdgpu_cs_chunk {
965 	__u32		chunk_id;
966 	__u32		length_dw;
967 	__u64		chunk_data;
968 };
969 
970 struct drm_amdgpu_cs_in {
971 	/** Rendering context id */
972 	__u32		ctx_id;
973 	/**  Handle of resource list associated with CS */
974 	__u32		bo_list_handle;
975 	__u32		num_chunks;
976 	__u32		flags;
977 	/** this points to __u64 * which point to cs chunks */
978 	__u64		chunks;
979 };
980 
981 struct drm_amdgpu_cs_out {
982 	__u64 handle;
983 };
984 
985 union drm_amdgpu_cs {
986 	struct drm_amdgpu_cs_in in;
987 	struct drm_amdgpu_cs_out out;
988 };
989 
990 /* Specify flags to be used for IB */
991 
992 /* This IB should be submitted to CE */
993 #define AMDGPU_IB_FLAG_CE	(1<<0)
994 
995 /* Preamble flag, which means the IB could be dropped if no context switch */
996 #define AMDGPU_IB_FLAG_PREAMBLE (1<<1)
997 
998 /* Preempt flag, IB should set Pre_enb bit if PREEMPT flag detected */
999 #define AMDGPU_IB_FLAG_PREEMPT (1<<2)
1000 
1001 /* The IB fence should do the L2 writeback but not invalidate any shader
1002  * caches (L2/vL1/sL1/I$). */
1003 #define AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE (1 << 3)
1004 
1005 /* Set GDS_COMPUTE_MAX_WAVE_ID = DEFAULT before PACKET3_INDIRECT_BUFFER.
1006  * This will reset wave ID counters for the IB.
1007  */
1008 #define AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID (1 << 4)
1009 
1010 /* Flag the IB as secure (TMZ)
1011  */
1012 #define AMDGPU_IB_FLAGS_SECURE  (1 << 5)
1013 
1014 /* Tell KMD to flush and invalidate caches
1015  */
1016 #define AMDGPU_IB_FLAG_EMIT_MEM_SYNC  (1 << 6)
1017 
1018 struct drm_amdgpu_cs_chunk_ib {
1019 	__u32 _pad;
1020 	/** AMDGPU_IB_FLAG_* */
1021 	__u32 flags;
1022 	/** Virtual address to begin IB execution */
1023 	__u64 va_start;
1024 	/** Size of submission */
1025 	__u32 ib_bytes;
1026 	/** HW IP to submit to */
1027 	__u32 ip_type;
1028 	/** HW IP index of the same type to submit to  */
1029 	__u32 ip_instance;
1030 	/** Ring index to submit to */
1031 	__u32 ring;
1032 };
1033 
1034 struct drm_amdgpu_cs_chunk_dep {
1035 	__u32 ip_type;
1036 	__u32 ip_instance;
1037 	__u32 ring;
1038 	__u32 ctx_id;
1039 	__u64 handle;
1040 };
1041 
1042 struct drm_amdgpu_cs_chunk_fence {
1043 	__u32 handle;
1044 	__u32 offset;
1045 };
1046 
1047 struct drm_amdgpu_cs_chunk_sem {
1048 	__u32 handle;
1049 };
1050 
1051 struct drm_amdgpu_cs_chunk_syncobj {
1052        __u32 handle;
1053        __u32 flags;
1054        __u64 point;
1055 };
1056 
1057 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ	0
1058 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ_FD	1
1059 #define AMDGPU_FENCE_TO_HANDLE_GET_SYNC_FILE_FD	2
1060 
1061 union drm_amdgpu_fence_to_handle {
1062 	struct {
1063 		struct drm_amdgpu_fence fence;
1064 		__u32 what;
1065 		__u32 pad;
1066 	} in;
1067 	struct {
1068 		__u32 handle;
1069 	} out;
1070 };
1071 
1072 struct drm_amdgpu_cs_chunk_data {
1073 	union {
1074 		struct drm_amdgpu_cs_chunk_ib		ib_data;
1075 		struct drm_amdgpu_cs_chunk_fence	fence_data;
1076 	};
1077 };
1078 
1079 #define AMDGPU_CS_CHUNK_CP_GFX_SHADOW_FLAGS_INIT_SHADOW         0x1
1080 
1081 struct drm_amdgpu_cs_chunk_cp_gfx_shadow {
1082 	__u64 shadow_va;
1083 	__u64 csa_va;
1084 	__u64 gds_va;
1085 	__u64 flags;
1086 };
1087 
1088 /*
1089  *  Query h/w info: Flag that this is integrated (a.h.a. fusion) GPU
1090  *
1091  */
1092 #define AMDGPU_IDS_FLAGS_FUSION			0x01
1093 #define AMDGPU_IDS_FLAGS_PREEMPTION		0x02
1094 #define AMDGPU_IDS_FLAGS_TMZ			0x04
1095 #define AMDGPU_IDS_FLAGS_CONFORMANT_TRUNC_COORD	0x08
1096 #define AMDGPU_IDS_FLAGS_GANG_SUBMIT		0x10
1097 
1098 /*
1099  *  Query h/w info: Flag identifying VF/PF/PT mode
1100  *
1101  */
1102 #define AMDGPU_IDS_FLAGS_MODE_MASK      0x300
1103 #define AMDGPU_IDS_FLAGS_MODE_SHIFT     0x8
1104 #define AMDGPU_IDS_FLAGS_MODE_PF        0x0
1105 #define AMDGPU_IDS_FLAGS_MODE_VF        0x1
1106 #define AMDGPU_IDS_FLAGS_MODE_PT        0x2
1107 
1108 /* indicate if acceleration can be working */
1109 #define AMDGPU_INFO_ACCEL_WORKING		0x00
1110 /* get the crtc_id from the mode object id? */
1111 #define AMDGPU_INFO_CRTC_FROM_ID		0x01
1112 /* query hw IP info */
1113 #define AMDGPU_INFO_HW_IP_INFO			0x02
1114 /* query hw IP instance count for the specified type */
1115 #define AMDGPU_INFO_HW_IP_COUNT			0x03
1116 /* timestamp for GL_ARB_timer_query */
1117 #define AMDGPU_INFO_TIMESTAMP			0x05
1118 /* Query the firmware version */
1119 #define AMDGPU_INFO_FW_VERSION			0x0e
1120 	/* Subquery id: Query VCE firmware version */
1121 	#define AMDGPU_INFO_FW_VCE		0x1
1122 	/* Subquery id: Query UVD firmware version */
1123 	#define AMDGPU_INFO_FW_UVD		0x2
1124 	/* Subquery id: Query GMC firmware version */
1125 	#define AMDGPU_INFO_FW_GMC		0x03
1126 	/* Subquery id: Query GFX ME firmware version */
1127 	#define AMDGPU_INFO_FW_GFX_ME		0x04
1128 	/* Subquery id: Query GFX PFP firmware version */
1129 	#define AMDGPU_INFO_FW_GFX_PFP		0x05
1130 	/* Subquery id: Query GFX CE firmware version */
1131 	#define AMDGPU_INFO_FW_GFX_CE		0x06
1132 	/* Subquery id: Query GFX RLC firmware version */
1133 	#define AMDGPU_INFO_FW_GFX_RLC		0x07
1134 	/* Subquery id: Query GFX MEC firmware version */
1135 	#define AMDGPU_INFO_FW_GFX_MEC		0x08
1136 	/* Subquery id: Query SMC firmware version */
1137 	#define AMDGPU_INFO_FW_SMC		0x0a
1138 	/* Subquery id: Query SDMA firmware version */
1139 	#define AMDGPU_INFO_FW_SDMA		0x0b
1140 	/* Subquery id: Query PSP SOS firmware version */
1141 	#define AMDGPU_INFO_FW_SOS		0x0c
1142 	/* Subquery id: Query PSP ASD firmware version */
1143 	#define AMDGPU_INFO_FW_ASD		0x0d
1144 	/* Subquery id: Query VCN firmware version */
1145 	#define AMDGPU_INFO_FW_VCN		0x0e
1146 	/* Subquery id: Query GFX RLC SRLC firmware version */
1147 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL 0x0f
1148 	/* Subquery id: Query GFX RLC SRLG firmware version */
1149 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM 0x10
1150 	/* Subquery id: Query GFX RLC SRLS firmware version */
1151 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM 0x11
1152 	/* Subquery id: Query DMCU firmware version */
1153 	#define AMDGPU_INFO_FW_DMCU		0x12
1154 	#define AMDGPU_INFO_FW_TA		0x13
1155 	/* Subquery id: Query DMCUB firmware version */
1156 	#define AMDGPU_INFO_FW_DMCUB		0x14
1157 	/* Subquery id: Query TOC firmware version */
1158 	#define AMDGPU_INFO_FW_TOC		0x15
1159 	/* Subquery id: Query CAP firmware version */
1160 	#define AMDGPU_INFO_FW_CAP		0x16
1161 	/* Subquery id: Query GFX RLCP firmware version */
1162 	#define AMDGPU_INFO_FW_GFX_RLCP		0x17
1163 	/* Subquery id: Query GFX RLCV firmware version */
1164 	#define AMDGPU_INFO_FW_GFX_RLCV		0x18
1165 	/* Subquery id: Query MES_KIQ firmware version */
1166 	#define AMDGPU_INFO_FW_MES_KIQ		0x19
1167 	/* Subquery id: Query MES firmware version */
1168 	#define AMDGPU_INFO_FW_MES		0x1a
1169 	/* Subquery id: Query IMU firmware version */
1170 	#define AMDGPU_INFO_FW_IMU		0x1b
1171 	/* Subquery id: Query VPE firmware version */
1172 	#define AMDGPU_INFO_FW_VPE		0x1c
1173 
1174 /* number of bytes moved for TTM migration */
1175 #define AMDGPU_INFO_NUM_BYTES_MOVED		0x0f
1176 /* the used VRAM size */
1177 #define AMDGPU_INFO_VRAM_USAGE			0x10
1178 /* the used GTT size */
1179 #define AMDGPU_INFO_GTT_USAGE			0x11
1180 /* Information about GDS, etc. resource configuration */
1181 #define AMDGPU_INFO_GDS_CONFIG			0x13
1182 /* Query information about VRAM and GTT domains */
1183 #define AMDGPU_INFO_VRAM_GTT			0x14
1184 /* Query information about register in MMR address space*/
1185 #define AMDGPU_INFO_READ_MMR_REG		0x15
1186 /* Query information about device: rev id, family, etc. */
1187 #define AMDGPU_INFO_DEV_INFO			0x16
1188 /* visible vram usage */
1189 #define AMDGPU_INFO_VIS_VRAM_USAGE		0x17
1190 /* number of TTM buffer evictions */
1191 #define AMDGPU_INFO_NUM_EVICTIONS		0x18
1192 /* Query memory about VRAM and GTT domains */
1193 #define AMDGPU_INFO_MEMORY			0x19
1194 /* Query vce clock table */
1195 #define AMDGPU_INFO_VCE_CLOCK_TABLE		0x1A
1196 /* Query vbios related information */
1197 #define AMDGPU_INFO_VBIOS			0x1B
1198 	/* Subquery id: Query vbios size */
1199 	#define AMDGPU_INFO_VBIOS_SIZE		0x1
1200 	/* Subquery id: Query vbios image */
1201 	#define AMDGPU_INFO_VBIOS_IMAGE		0x2
1202 	/* Subquery id: Query vbios info */
1203 	#define AMDGPU_INFO_VBIOS_INFO		0x3
1204 /* Query UVD handles */
1205 #define AMDGPU_INFO_NUM_HANDLES			0x1C
1206 /* Query sensor related information */
1207 #define AMDGPU_INFO_SENSOR			0x1D
1208 	/* Subquery id: Query GPU shader clock */
1209 	#define AMDGPU_INFO_SENSOR_GFX_SCLK		0x1
1210 	/* Subquery id: Query GPU memory clock */
1211 	#define AMDGPU_INFO_SENSOR_GFX_MCLK		0x2
1212 	/* Subquery id: Query GPU temperature */
1213 	#define AMDGPU_INFO_SENSOR_GPU_TEMP		0x3
1214 	/* Subquery id: Query GPU load */
1215 	#define AMDGPU_INFO_SENSOR_GPU_LOAD		0x4
1216 	/* Subquery id: Query average GPU power	*/
1217 	#define AMDGPU_INFO_SENSOR_GPU_AVG_POWER	0x5
1218 	/* Subquery id: Query northbridge voltage */
1219 	#define AMDGPU_INFO_SENSOR_VDDNB		0x6
1220 	/* Subquery id: Query graphics voltage */
1221 	#define AMDGPU_INFO_SENSOR_VDDGFX		0x7
1222 	/* Subquery id: Query GPU stable pstate shader clock */
1223 	#define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_SCLK		0x8
1224 	/* Subquery id: Query GPU stable pstate memory clock */
1225 	#define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_MCLK		0x9
1226 	/* Subquery id: Query GPU peak pstate shader clock */
1227 	#define AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_SCLK			0xa
1228 	/* Subquery id: Query GPU peak pstate memory clock */
1229 	#define AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_MCLK			0xb
1230 	/* Subquery id: Query input GPU power	*/
1231 	#define AMDGPU_INFO_SENSOR_GPU_INPUT_POWER	0xc
1232 /* Number of VRAM page faults on CPU access. */
1233 #define AMDGPU_INFO_NUM_VRAM_CPU_PAGE_FAULTS	0x1E
1234 #define AMDGPU_INFO_VRAM_LOST_COUNTER		0x1F
1235 /* query ras mask of enabled features*/
1236 #define AMDGPU_INFO_RAS_ENABLED_FEATURES	0x20
1237 /* RAS MASK: UMC (VRAM) */
1238 #define AMDGPU_INFO_RAS_ENABLED_UMC			(1 << 0)
1239 /* RAS MASK: SDMA */
1240 #define AMDGPU_INFO_RAS_ENABLED_SDMA			(1 << 1)
1241 /* RAS MASK: GFX */
1242 #define AMDGPU_INFO_RAS_ENABLED_GFX			(1 << 2)
1243 /* RAS MASK: MMHUB */
1244 #define AMDGPU_INFO_RAS_ENABLED_MMHUB			(1 << 3)
1245 /* RAS MASK: ATHUB */
1246 #define AMDGPU_INFO_RAS_ENABLED_ATHUB			(1 << 4)
1247 /* RAS MASK: PCIE */
1248 #define AMDGPU_INFO_RAS_ENABLED_PCIE			(1 << 5)
1249 /* RAS MASK: HDP */
1250 #define AMDGPU_INFO_RAS_ENABLED_HDP			(1 << 6)
1251 /* RAS MASK: XGMI */
1252 #define AMDGPU_INFO_RAS_ENABLED_XGMI			(1 << 7)
1253 /* RAS MASK: DF */
1254 #define AMDGPU_INFO_RAS_ENABLED_DF			(1 << 8)
1255 /* RAS MASK: SMN */
1256 #define AMDGPU_INFO_RAS_ENABLED_SMN			(1 << 9)
1257 /* RAS MASK: SEM */
1258 #define AMDGPU_INFO_RAS_ENABLED_SEM			(1 << 10)
1259 /* RAS MASK: MP0 */
1260 #define AMDGPU_INFO_RAS_ENABLED_MP0			(1 << 11)
1261 /* RAS MASK: MP1 */
1262 #define AMDGPU_INFO_RAS_ENABLED_MP1			(1 << 12)
1263 /* RAS MASK: FUSE */
1264 #define AMDGPU_INFO_RAS_ENABLED_FUSE			(1 << 13)
1265 /* query video encode/decode caps */
1266 #define AMDGPU_INFO_VIDEO_CAPS			0x21
1267 	/* Subquery id: Decode */
1268 	#define AMDGPU_INFO_VIDEO_CAPS_DECODE		0
1269 	/* Subquery id: Encode */
1270 	#define AMDGPU_INFO_VIDEO_CAPS_ENCODE		1
1271 /* Query the max number of IBs per gang per submission */
1272 #define AMDGPU_INFO_MAX_IBS			0x22
1273 /* query last page fault info */
1274 #define AMDGPU_INFO_GPUVM_FAULT			0x23
1275 /* query FW object size and alignment */
1276 #define AMDGPU_INFO_UQ_FW_AREAS			0x24
1277 
1278 #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT	0
1279 #define AMDGPU_INFO_MMR_SE_INDEX_MASK	0xff
1280 #define AMDGPU_INFO_MMR_SH_INDEX_SHIFT	8
1281 #define AMDGPU_INFO_MMR_SH_INDEX_MASK	0xff
1282 
1283 struct drm_amdgpu_query_fw {
1284 	/** AMDGPU_INFO_FW_* */
1285 	__u32 fw_type;
1286 	/**
1287 	 * Index of the IP if there are more IPs of
1288 	 * the same type.
1289 	 */
1290 	__u32 ip_instance;
1291 	/**
1292 	 * Index of the engine. Whether this is used depends
1293 	 * on the firmware type. (e.g. MEC, SDMA)
1294 	 */
1295 	__u32 index;
1296 	__u32 _pad;
1297 };
1298 
1299 /* Input structure for the INFO ioctl */
1300 struct drm_amdgpu_info {
1301 	/* Where the return value will be stored */
1302 	__u64 return_pointer;
1303 	/* The size of the return value. Just like "size" in "snprintf",
1304 	 * it limits how many bytes the kernel can write. */
1305 	__u32 return_size;
1306 	/* The query request id. */
1307 	__u32 query;
1308 
1309 	union {
1310 		struct {
1311 			__u32 id;
1312 			__u32 _pad;
1313 		} mode_crtc;
1314 
1315 		struct {
1316 			/** AMDGPU_HW_IP_* */
1317 			__u32 type;
1318 			/**
1319 			 * Index of the IP if there are more IPs of the same
1320 			 * type. Ignored by AMDGPU_INFO_HW_IP_COUNT.
1321 			 */
1322 			__u32 ip_instance;
1323 		} query_hw_ip;
1324 
1325 		struct {
1326 			__u32 dword_offset;
1327 			/** number of registers to read */
1328 			__u32 count;
1329 			__u32 instance;
1330 			/** For future use, no flags defined so far */
1331 			__u32 flags;
1332 		} read_mmr_reg;
1333 
1334 		struct drm_amdgpu_query_fw query_fw;
1335 
1336 		struct {
1337 			__u32 type;
1338 			__u32 offset;
1339 		} vbios_info;
1340 
1341 		struct {
1342 			__u32 type;
1343 		} sensor_info;
1344 
1345 		struct {
1346 			__u32 type;
1347 		} video_cap;
1348 	};
1349 };
1350 
1351 struct drm_amdgpu_info_gds {
1352 	/** GDS GFX partition size */
1353 	__u32 gds_gfx_partition_size;
1354 	/** GDS compute partition size */
1355 	__u32 compute_partition_size;
1356 	/** total GDS memory size */
1357 	__u32 gds_total_size;
1358 	/** GWS size per GFX partition */
1359 	__u32 gws_per_gfx_partition;
1360 	/** GSW size per compute partition */
1361 	__u32 gws_per_compute_partition;
1362 	/** OA size per GFX partition */
1363 	__u32 oa_per_gfx_partition;
1364 	/** OA size per compute partition */
1365 	__u32 oa_per_compute_partition;
1366 	__u32 _pad;
1367 };
1368 
1369 struct drm_amdgpu_info_vram_gtt {
1370 	__u64 vram_size;
1371 	__u64 vram_cpu_accessible_size;
1372 	__u64 gtt_size;
1373 };
1374 
1375 struct drm_amdgpu_heap_info {
1376 	/** max. physical memory */
1377 	__u64 total_heap_size;
1378 
1379 	/** Theoretical max. available memory in the given heap */
1380 	__u64 usable_heap_size;
1381 
1382 	/**
1383 	 * Number of bytes allocated in the heap. This includes all processes
1384 	 * and private allocations in the kernel. It changes when new buffers
1385 	 * are allocated, freed, and moved. It cannot be larger than
1386 	 * heap_size.
1387 	 */
1388 	__u64 heap_usage;
1389 
1390 	/**
1391 	 * Theoretical possible max. size of buffer which
1392 	 * could be allocated in the given heap
1393 	 */
1394 	__u64 max_allocation;
1395 };
1396 
1397 struct drm_amdgpu_memory_info {
1398 	struct drm_amdgpu_heap_info vram;
1399 	struct drm_amdgpu_heap_info cpu_accessible_vram;
1400 	struct drm_amdgpu_heap_info gtt;
1401 };
1402 
1403 struct drm_amdgpu_info_firmware {
1404 	__u32 ver;
1405 	__u32 feature;
1406 };
1407 
1408 struct drm_amdgpu_info_vbios {
1409 	__u8 name[64];
1410 	__u8 vbios_pn[64];
1411 	__u32 version;
1412 	__u32 pad;
1413 	__u8 vbios_ver_str[32];
1414 	__u8 date[32];
1415 };
1416 
1417 #define AMDGPU_VRAM_TYPE_UNKNOWN 0
1418 #define AMDGPU_VRAM_TYPE_GDDR1 1
1419 #define AMDGPU_VRAM_TYPE_DDR2  2
1420 #define AMDGPU_VRAM_TYPE_GDDR3 3
1421 #define AMDGPU_VRAM_TYPE_GDDR4 4
1422 #define AMDGPU_VRAM_TYPE_GDDR5 5
1423 #define AMDGPU_VRAM_TYPE_HBM   6
1424 #define AMDGPU_VRAM_TYPE_DDR3  7
1425 #define AMDGPU_VRAM_TYPE_DDR4  8
1426 #define AMDGPU_VRAM_TYPE_GDDR6 9
1427 #define AMDGPU_VRAM_TYPE_DDR5  10
1428 #define AMDGPU_VRAM_TYPE_LPDDR4 11
1429 #define AMDGPU_VRAM_TYPE_LPDDR5 12
1430 #define AMDGPU_VRAM_TYPE_HBM3E 13
1431 #define AMDGPU_VRAM_TYPE_HBM4 14
1432 
1433 struct drm_amdgpu_info_device {
1434 	/** PCI Device ID */
1435 	__u32 device_id;
1436 	/** Internal chip revision: A0, A1, etc.) */
1437 	__u32 chip_rev;
1438 	__u32 external_rev;
1439 	/** Revision id in PCI Config space */
1440 	__u32 pci_rev;
1441 	__u32 family;
1442 	__u32 num_shader_engines;
1443 	__u32 num_shader_arrays_per_engine;
1444 	/* in KHz */
1445 	__u32 gpu_counter_freq;
1446 	__u64 max_engine_clock;
1447 	__u64 max_memory_clock;
1448 	/* cu information */
1449 	__u32 cu_active_number;
1450 	/* NOTE: cu_ao_mask is INVALID, DON'T use it */
1451 	__u32 cu_ao_mask;
1452 	__u32 cu_bitmap[4][4];
1453 	/** Render backend pipe mask. One render backend is CB+DB. */
1454 	__u32 enabled_rb_pipes_mask;
1455 	__u32 num_rb_pipes;
1456 	__u32 num_hw_gfx_contexts;
1457 	/* PCIe version (the smaller of the GPU and the CPU/motherboard) */
1458 	__u32 pcie_gen;
1459 	__u64 ids_flags;
1460 	/** Starting virtual address for UMDs. */
1461 	__u64 virtual_address_offset;
1462 	/** The maximum virtual address */
1463 	__u64 virtual_address_max;
1464 	/** Required alignment of virtual addresses. */
1465 	__u32 virtual_address_alignment;
1466 	/** Page table entry - fragment size */
1467 	__u32 pte_fragment_size;
1468 	__u32 gart_page_size;
1469 	/** constant engine ram size*/
1470 	__u32 ce_ram_size;
1471 	/** video memory type info*/
1472 	__u32 vram_type;
1473 	/** video memory bit width*/
1474 	__u32 vram_bit_width;
1475 	/* vce harvesting instance */
1476 	__u32 vce_harvest_config;
1477 	/* gfx double offchip LDS buffers */
1478 	__u32 gc_double_offchip_lds_buf;
1479 	/* NGG Primitive Buffer */
1480 	__u64 prim_buf_gpu_addr;
1481 	/* NGG Position Buffer */
1482 	__u64 pos_buf_gpu_addr;
1483 	/* NGG Control Sideband */
1484 	__u64 cntl_sb_buf_gpu_addr;
1485 	/* NGG Parameter Cache */
1486 	__u64 param_buf_gpu_addr;
1487 	__u32 prim_buf_size;
1488 	__u32 pos_buf_size;
1489 	__u32 cntl_sb_buf_size;
1490 	__u32 param_buf_size;
1491 	/* wavefront size*/
1492 	__u32 wave_front_size;
1493 	/* shader visible vgprs*/
1494 	__u32 num_shader_visible_vgprs;
1495 	/* CU per shader array*/
1496 	__u32 num_cu_per_sh;
1497 	/* number of tcc blocks*/
1498 	__u32 num_tcc_blocks;
1499 	/* gs vgt table depth*/
1500 	__u32 gs_vgt_table_depth;
1501 	/* gs primitive buffer depth*/
1502 	__u32 gs_prim_buffer_depth;
1503 	/* max gs wavefront per vgt*/
1504 	__u32 max_gs_waves_per_vgt;
1505 	/* PCIe number of lanes (the smaller of the GPU and the CPU/motherboard) */
1506 	__u32 pcie_num_lanes;
1507 	/* always on cu bitmap */
1508 	__u32 cu_ao_bitmap[4][4];
1509 	/** Starting high virtual address for UMDs. */
1510 	__u64 high_va_offset;
1511 	/** The maximum high virtual address */
1512 	__u64 high_va_max;
1513 	/* gfx10 pa_sc_tile_steering_override */
1514 	__u32 pa_sc_tile_steering_override;
1515 	/* disabled TCCs */
1516 	__u64 tcc_disabled_mask;
1517 	__u64 min_engine_clock;
1518 	__u64 min_memory_clock;
1519 	/* The following fields are only set on gfx11+, older chips set 0. */
1520 	__u32 tcp_cache_size;       /* AKA GL0, VMEM cache */
1521 	__u32 num_sqc_per_wgp;
1522 	__u32 sqc_data_cache_size;  /* AKA SMEM cache */
1523 	__u32 sqc_inst_cache_size;
1524 	__u32 gl1c_cache_size;
1525 	__u32 gl2c_cache_size;
1526 	__u64 mall_size;            /* AKA infinity cache */
1527 	/* high 32 bits of the rb pipes mask */
1528 	__u32 enabled_rb_pipes_mask_hi;
1529 	/* shadow area size for gfx11 */
1530 	__u32 shadow_size;
1531 	/* shadow area base virtual alignment for gfx11 */
1532 	__u32 shadow_alignment;
1533 	/* context save area size for gfx11 */
1534 	__u32 csa_size;
1535 	/* context save area base virtual alignment for gfx11 */
1536 	__u32 csa_alignment;
1537 	/* Userq IP mask (1 << AMDGPU_HW_IP_*) */
1538 	__u32 userq_ip_mask;
1539 	__u32 pad;
1540 };
1541 
1542 struct drm_amdgpu_info_hw_ip {
1543 	/** Version of h/w IP */
1544 	__u32  hw_ip_version_major;
1545 	__u32  hw_ip_version_minor;
1546 	/** Capabilities */
1547 	__u64  capabilities_flags;
1548 	/** command buffer address start alignment*/
1549 	__u32  ib_start_alignment;
1550 	/** command buffer size alignment*/
1551 	__u32  ib_size_alignment;
1552 	/** Bitmask of available rings. Bit 0 means ring 0, etc. */
1553 	__u32  available_rings;
1554 	/** version info: bits 23:16 major, 15:8 minor, 7:0 revision */
1555 	__u32  ip_discovery_version;
1556 	/* Userq available slots */
1557 	__u32  userq_num_slots;
1558 };
1559 
1560 struct drm_amdgpu_info_num_handles {
1561 	/** Max handles as supported by firmware for UVD */
1562 	__u32  uvd_max_handles;
1563 	/** Handles currently in use for UVD */
1564 	__u32  uvd_used_handles;
1565 };
1566 
1567 #define AMDGPU_VCE_CLOCK_TABLE_ENTRIES		6
1568 
1569 struct drm_amdgpu_info_vce_clock_table_entry {
1570 	/** System clock */
1571 	__u32 sclk;
1572 	/** Memory clock */
1573 	__u32 mclk;
1574 	/** VCE clock */
1575 	__u32 eclk;
1576 	__u32 pad;
1577 };
1578 
1579 struct drm_amdgpu_info_vce_clock_table {
1580 	struct drm_amdgpu_info_vce_clock_table_entry entries[AMDGPU_VCE_CLOCK_TABLE_ENTRIES];
1581 	__u32 num_valid_entries;
1582 	__u32 pad;
1583 };
1584 
1585 /* query video encode/decode caps */
1586 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2			0
1587 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4			1
1588 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1			2
1589 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC		3
1590 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC			4
1591 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG			5
1592 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9			6
1593 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1			7
1594 #define AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT			8
1595 
1596 struct drm_amdgpu_info_video_codec_info {
1597 	__u32 valid;
1598 	__u32 max_width;
1599 	__u32 max_height;
1600 	__u32 max_pixels_per_frame;
1601 	__u32 max_level;
1602 	__u32 pad;
1603 };
1604 
1605 struct drm_amdgpu_info_video_caps {
1606 	struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
1607 };
1608 
1609 #define AMDGPU_VMHUB_TYPE_MASK			0xff
1610 #define AMDGPU_VMHUB_TYPE_SHIFT			0
1611 #define AMDGPU_VMHUB_TYPE_GFX			0
1612 #define AMDGPU_VMHUB_TYPE_MM0			1
1613 #define AMDGPU_VMHUB_TYPE_MM1			2
1614 #define AMDGPU_VMHUB_IDX_MASK			0xff00
1615 #define AMDGPU_VMHUB_IDX_SHIFT			8
1616 
1617 struct drm_amdgpu_info_gpuvm_fault {
1618 	__u64 addr;
1619 	__u32 status;
1620 	__u32 vmhub;
1621 };
1622 
1623 struct drm_amdgpu_info_uq_metadata_gfx {
1624 	/* shadow area size for gfx11 */
1625 	__u32 shadow_size;
1626 	/* shadow area base virtual alignment for gfx11 */
1627 	__u32 shadow_alignment;
1628 	/* context save area size for gfx11 */
1629 	__u32 csa_size;
1630 	/* context save area base virtual alignment for gfx11 */
1631 	__u32 csa_alignment;
1632 };
1633 
1634 struct drm_amdgpu_info_uq_metadata_compute {
1635 	/* EOP size for gfx11 */
1636 	__u32 eop_size;
1637 	/* EOP base virtual alignment for gfx11 */
1638 	__u32 eop_alignment;
1639 };
1640 
1641 struct drm_amdgpu_info_uq_metadata_sdma {
1642 	/* context save area size for sdma6 */
1643 	__u32 csa_size;
1644 	/* context save area base virtual alignment for sdma6 */
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 		struct drm_amdgpu_info_uq_metadata_compute compute;
1652 		struct drm_amdgpu_info_uq_metadata_sdma sdma;
1653 	};
1654 };
1655 
1656 /*
1657  * Supported GPU families
1658  */
1659 #define AMDGPU_FAMILY_UNKNOWN			0
1660 #define AMDGPU_FAMILY_SI			110 /* Hainan, Oland, Verde, Pitcairn, Tahiti */
1661 #define AMDGPU_FAMILY_CI			120 /* Bonaire, Hawaii */
1662 #define AMDGPU_FAMILY_KV			125 /* Kaveri, Kabini, Mullins */
1663 #define AMDGPU_FAMILY_VI			130 /* Iceland, Tonga */
1664 #define AMDGPU_FAMILY_CZ			135 /* Carrizo, Stoney */
1665 #define AMDGPU_FAMILY_AI			141 /* Vega10 */
1666 #define AMDGPU_FAMILY_RV			142 /* Raven */
1667 #define AMDGPU_FAMILY_NV			143 /* Navi10 */
1668 #define AMDGPU_FAMILY_VGH			144 /* Van Gogh */
1669 #define AMDGPU_FAMILY_GC_11_0_0			145 /* GC 11.0.0 */
1670 #define AMDGPU_FAMILY_YC			146 /* Yellow Carp */
1671 #define AMDGPU_FAMILY_GC_11_0_1			148 /* GC 11.0.1 */
1672 #define AMDGPU_FAMILY_GC_10_3_6			149 /* GC 10.3.6 */
1673 #define AMDGPU_FAMILY_GC_10_3_7			151 /* GC 10.3.7 */
1674 #define AMDGPU_FAMILY_GC_11_5_0			150 /* GC 11.5.0 */
1675 #define AMDGPU_FAMILY_GC_11_5_4			154 /* GC 11.5.4 */
1676 #define AMDGPU_FAMILY_GC_12_0_0			152 /* GC 12.0.0 */
1677 
1678 /*
1679  * Definition of user options
1680  *
1681  * option: AMDGPU_PROC_OPTIONS_OP_KFD_SIGBUS_DELAY
1682  *    0:          Disable sigbus delay - SIGBUS will be raised immediately
1683  *    0xFFFFFFFF: SIGBUS will not be raised
1684  *    other:      Set the sigbus delay in milliseconds
1685  */
1686 #define AMDGPU_PROC_OPTIONS_OP_KFD_SIGBUS_DELAY		0
1687 
1688 #define AMDGPU_PROC_OPTIONS_KFD_SIGBUS_DELAY_DISABLED	0xFFFFFFFFu
1689 
1690 struct drm_amdgpu_proc_options {
1691 	__u32 op;
1692 	struct {
1693 		__u32 value;
1694 	} kfd_sigbus_delay;
1695 };
1696 
1697 #if defined(__cplusplus)
1698 }
1699 #endif
1700 
1701 #endif
1702