xref: /linux/include/uapi/drm/xe_drm.h (revision 170aafe35cb98e0f3fbacb446ea86389fbce22ea)
1 /* SPDX-License-Identifier: MIT */
2 /*
3  * Copyright © 2023 Intel Corporation
4  */
5 
6 #ifndef _UAPI_XE_DRM_H_
7 #define _UAPI_XE_DRM_H_
8 
9 #include "drm.h"
10 
11 #if defined(__cplusplus)
12 extern "C" {
13 #endif
14 
15 /*
16  * Please note that modifications to all structs defined here are
17  * subject to backwards-compatibility constraints.
18  * Sections in this file are organized as follows:
19  *   1. IOCTL definition
20  *   2. Extension definition and helper structs
21  *   3. IOCTL's Query structs in the order of the Query's entries.
22  *   4. The rest of IOCTL structs in the order of IOCTL declaration.
23  */
24 
25 /**
26  * DOC: Xe Device Block Diagram
27  *
28  * The diagram below represents a high-level simplification of a discrete
29  * GPU supported by the Xe driver. It shows some device components which
30  * are necessary to understand this API, as well as how their relations
31  * to each other. This diagram does not represent real hardware::
32  *
33  *   ┌──────────────────────────────────────────────────────────────────┐
34  *   │ ┌──────────────────────────────────────────────────┐ ┌─────────┐ │
35  *   │ │        ┌───────────────────────┐   ┌─────┐       │ │ ┌─────┐ │ │
36  *   │ │        │         VRAM0         ├───┤ ... │       │ │ │VRAM1│ │ │
37  *   │ │        └───────────┬───────────┘   └─GT1─┘       │ │ └──┬──┘ │ │
38  *   │ │ ┌──────────────────┴───────────────────────────┐ │ │ ┌──┴──┐ │ │
39  *   │ │ │ ┌─────────────────────┐  ┌─────────────────┐ │ │ │ │     │ │ │
40  *   │ │ │ │ ┌──┐ ┌──┐ ┌──┐ ┌──┐ │  │ ┌─────┐ ┌─────┐ │ │ │ │ │     │ │ │
41  *   │ │ │ │ │EU│ │EU│ │EU│ │EU│ │  │ │RCS0 │ │BCS0 │ │ │ │ │ │     │ │ │
42  *   │ │ │ │ └──┘ └──┘ └──┘ └──┘ │  │ └─────┘ └─────┘ │ │ │ │ │     │ │ │
43  *   │ │ │ │ ┌──┐ ┌──┐ ┌──┐ ┌──┐ │  │ ┌─────┐ ┌─────┐ │ │ │ │ │     │ │ │
44  *   │ │ │ │ │EU│ │EU│ │EU│ │EU│ │  │ │VCS0 │ │VCS1 │ │ │ │ │ │     │ │ │
45  *   │ │ │ │ └──┘ └──┘ └──┘ └──┘ │  │ └─────┘ └─────┘ │ │ │ │ │     │ │ │
46  *   │ │ │ │ ┌──┐ ┌──┐ ┌──┐ ┌──┐ │  │ ┌─────┐ ┌─────┐ │ │ │ │ │     │ │ │
47  *   │ │ │ │ │EU│ │EU│ │EU│ │EU│ │  │ │VECS0│ │VECS1│ │ │ │ │ │ ... │ │ │
48  *   │ │ │ │ └──┘ └──┘ └──┘ └──┘ │  │ └─────┘ └─────┘ │ │ │ │ │     │ │ │
49  *   │ │ │ │ ┌──┐ ┌──┐ ┌──┐ ┌──┐ │  │ ┌─────┐ ┌─────┐ │ │ │ │ │     │ │ │
50  *   │ │ │ │ │EU│ │EU│ │EU│ │EU│ │  │ │CCS0 │ │CCS1 │ │ │ │ │ │     │ │ │
51  *   │ │ │ │ └──┘ └──┘ └──┘ └──┘ │  │ └─────┘ └─────┘ │ │ │ │ │     │ │ │
52  *   │ │ │ └─────────DSS─────────┘  │ ┌─────┐ ┌─────┐ │ │ │ │ │     │ │ │
53  *   │ │ │                          │ │CCS2 │ │CCS3 │ │ │ │ │ │     │ │ │
54  *   │ │ │ ┌─────┐ ┌─────┐ ┌─────┐  │ └─────┘ └─────┘ │ │ │ │ │     │ │ │
55  *   │ │ │ │ ... │ │ ... │ │ ... │  │                 │ │ │ │ │     │ │ │
56  *   │ │ │ └─DSS─┘ └─DSS─┘ └─DSS─┘  └─────Engines─────┘ │ │ │ │     │ │ │
57  *   │ │ └───────────────────────────GT0────────────────┘ │ │ └─GT2─┘ │ │
58  *   │ └────────────────────────────Tile0─────────────────┘ └─ Tile1──┘ │
59  *   └─────────────────────────────Device0───────┬──────────────────────┘
60  *                                               │
61  *                        ───────────────────────┴────────── PCI bus
62  */
63 
64 /**
65  * DOC: Xe uAPI Overview
66  *
67  * This section aims to describe the Xe's IOCTL entries, its structs, and other
68  * Xe related uAPI such as uevents and PMU (Platform Monitoring Unit) related
69  * entries and usage.
70  *
71  * List of supported IOCTLs:
72  *  - &DRM_IOCTL_XE_DEVICE_QUERY
73  *  - &DRM_IOCTL_XE_GEM_CREATE
74  *  - &DRM_IOCTL_XE_GEM_MMAP_OFFSET
75  *  - &DRM_IOCTL_XE_VM_CREATE
76  *  - &DRM_IOCTL_XE_VM_DESTROY
77  *  - &DRM_IOCTL_XE_VM_BIND
78  *  - &DRM_IOCTL_XE_EXEC_QUEUE_CREATE
79  *  - &DRM_IOCTL_XE_EXEC_QUEUE_DESTROY
80  *  - &DRM_IOCTL_XE_EXEC_QUEUE_GET_PROPERTY
81  *  - &DRM_IOCTL_XE_EXEC
82  *  - &DRM_IOCTL_XE_WAIT_USER_FENCE
83  *  - &DRM_IOCTL_XE_OBSERVATION
84  */
85 
86 /*
87  * xe specific ioctls.
88  *
89  * The device specific ioctl range is [DRM_COMMAND_BASE, DRM_COMMAND_END) ie
90  * [0x40, 0xa0) (a0 is excluded). The numbers below are defined as offset
91  * against DRM_COMMAND_BASE and should be between [0x0, 0x60).
92  */
93 #define DRM_XE_DEVICE_QUERY		0x00
94 #define DRM_XE_GEM_CREATE		0x01
95 #define DRM_XE_GEM_MMAP_OFFSET		0x02
96 #define DRM_XE_VM_CREATE		0x03
97 #define DRM_XE_VM_DESTROY		0x04
98 #define DRM_XE_VM_BIND			0x05
99 #define DRM_XE_EXEC_QUEUE_CREATE	0x06
100 #define DRM_XE_EXEC_QUEUE_DESTROY	0x07
101 #define DRM_XE_EXEC_QUEUE_GET_PROPERTY	0x08
102 #define DRM_XE_EXEC			0x09
103 #define DRM_XE_WAIT_USER_FENCE		0x0a
104 #define DRM_XE_OBSERVATION		0x0b
105 
106 /* Must be kept compact -- no holes */
107 
108 #define DRM_IOCTL_XE_DEVICE_QUERY		DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_DEVICE_QUERY, struct drm_xe_device_query)
109 #define DRM_IOCTL_XE_GEM_CREATE			DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_GEM_CREATE, struct drm_xe_gem_create)
110 #define DRM_IOCTL_XE_GEM_MMAP_OFFSET		DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_GEM_MMAP_OFFSET, struct drm_xe_gem_mmap_offset)
111 #define DRM_IOCTL_XE_VM_CREATE			DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_VM_CREATE, struct drm_xe_vm_create)
112 #define DRM_IOCTL_XE_VM_DESTROY			DRM_IOW(DRM_COMMAND_BASE + DRM_XE_VM_DESTROY, struct drm_xe_vm_destroy)
113 #define DRM_IOCTL_XE_VM_BIND			DRM_IOW(DRM_COMMAND_BASE + DRM_XE_VM_BIND, struct drm_xe_vm_bind)
114 #define DRM_IOCTL_XE_EXEC_QUEUE_CREATE		DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_EXEC_QUEUE_CREATE, struct drm_xe_exec_queue_create)
115 #define DRM_IOCTL_XE_EXEC_QUEUE_DESTROY		DRM_IOW(DRM_COMMAND_BASE + DRM_XE_EXEC_QUEUE_DESTROY, struct drm_xe_exec_queue_destroy)
116 #define DRM_IOCTL_XE_EXEC_QUEUE_GET_PROPERTY	DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_EXEC_QUEUE_GET_PROPERTY, struct drm_xe_exec_queue_get_property)
117 #define DRM_IOCTL_XE_EXEC			DRM_IOW(DRM_COMMAND_BASE + DRM_XE_EXEC, struct drm_xe_exec)
118 #define DRM_IOCTL_XE_WAIT_USER_FENCE		DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_WAIT_USER_FENCE, struct drm_xe_wait_user_fence)
119 #define DRM_IOCTL_XE_OBSERVATION		DRM_IOW(DRM_COMMAND_BASE + DRM_XE_OBSERVATION, struct drm_xe_observation_param)
120 
121 /**
122  * DOC: Xe IOCTL Extensions
123  *
124  * Before detailing the IOCTLs and its structs, it is important to highlight
125  * that every IOCTL in Xe is extensible.
126  *
127  * Many interfaces need to grow over time. In most cases we can simply
128  * extend the struct and have userspace pass in more data. Another option,
129  * as demonstrated by Vulkan's approach to providing extensions for forward
130  * and backward compatibility, is to use a list of optional structs to
131  * provide those extra details.
132  *
133  * The key advantage to using an extension chain is that it allows us to
134  * redefine the interface more easily than an ever growing struct of
135  * increasing complexity, and for large parts of that interface to be
136  * entirely optional. The downside is more pointer chasing; chasing across
137  * the __user boundary with pointers encapsulated inside u64.
138  *
139  * Example chaining:
140  *
141  * .. code-block:: C
142  *
143  *	struct drm_xe_user_extension ext3 {
144  *		.next_extension = 0, // end
145  *		.name = ...,
146  *	};
147  *	struct drm_xe_user_extension ext2 {
148  *		.next_extension = (uintptr_t)&ext3,
149  *		.name = ...,
150  *	};
151  *	struct drm_xe_user_extension ext1 {
152  *		.next_extension = (uintptr_t)&ext2,
153  *		.name = ...,
154  *	};
155  *
156  * Typically the struct drm_xe_user_extension would be embedded in some uAPI
157  * struct, and in this case we would feed it the head of the chain(i.e ext1),
158  * which would then apply all of the above extensions.
159 */
160 
161 /**
162  * struct drm_xe_user_extension - Base class for defining a chain of extensions
163  */
164 struct drm_xe_user_extension {
165 	/**
166 	 * @next_extension:
167 	 *
168 	 * Pointer to the next struct drm_xe_user_extension, or zero if the end.
169 	 */
170 	__u64 next_extension;
171 
172 	/**
173 	 * @name: Name of the extension.
174 	 *
175 	 * Note that the name here is just some integer.
176 	 *
177 	 * Also note that the name space for this is not global for the whole
178 	 * driver, but rather its scope/meaning is limited to the specific piece
179 	 * of uAPI which has embedded the struct drm_xe_user_extension.
180 	 */
181 	__u32 name;
182 
183 	/**
184 	 * @pad: MBZ
185 	 *
186 	 * All undefined bits must be zero.
187 	 */
188 	__u32 pad;
189 };
190 
191 /**
192  * struct drm_xe_ext_set_property - Generic set property extension
193  *
194  * A generic struct that allows any of the Xe's IOCTL to be extended
195  * with a set_property operation.
196  */
197 struct drm_xe_ext_set_property {
198 	/** @base: base user extension */
199 	struct drm_xe_user_extension base;
200 
201 	/** @property: property to set */
202 	__u32 property;
203 
204 	/** @pad: MBZ */
205 	__u32 pad;
206 
207 	/** @value: property value */
208 	__u64 value;
209 
210 	/** @reserved: Reserved */
211 	__u64 reserved[2];
212 };
213 
214 /**
215  * struct drm_xe_engine_class_instance - instance of an engine class
216  *
217  * It is returned as part of the @drm_xe_engine, but it also is used as
218  * the input of engine selection for both @drm_xe_exec_queue_create and
219  * @drm_xe_query_engine_cycles
220  *
221  * The @engine_class can be:
222  *  - %DRM_XE_ENGINE_CLASS_RENDER
223  *  - %DRM_XE_ENGINE_CLASS_COPY
224  *  - %DRM_XE_ENGINE_CLASS_VIDEO_DECODE
225  *  - %DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE
226  *  - %DRM_XE_ENGINE_CLASS_COMPUTE
227  *  - %DRM_XE_ENGINE_CLASS_VM_BIND - Kernel only classes (not actual
228  *    hardware engine class). Used for creating ordered queues of VM
229  *    bind operations.
230  */
231 struct drm_xe_engine_class_instance {
232 #define DRM_XE_ENGINE_CLASS_RENDER		0
233 #define DRM_XE_ENGINE_CLASS_COPY		1
234 #define DRM_XE_ENGINE_CLASS_VIDEO_DECODE	2
235 #define DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE	3
236 #define DRM_XE_ENGINE_CLASS_COMPUTE		4
237 #define DRM_XE_ENGINE_CLASS_VM_BIND		5
238 	/** @engine_class: engine class id */
239 	__u16 engine_class;
240 	/** @engine_instance: engine instance id */
241 	__u16 engine_instance;
242 	/** @gt_id: Unique ID of this GT within the PCI Device */
243 	__u16 gt_id;
244 	/** @pad: MBZ */
245 	__u16 pad;
246 };
247 
248 /**
249  * struct drm_xe_engine - describe hardware engine
250  */
251 struct drm_xe_engine {
252 	/** @instance: The @drm_xe_engine_class_instance */
253 	struct drm_xe_engine_class_instance instance;
254 
255 	/** @reserved: Reserved */
256 	__u64 reserved[3];
257 };
258 
259 /**
260  * struct drm_xe_query_engines - describe engines
261  *
262  * If a query is made with a struct @drm_xe_device_query where .query
263  * is equal to %DRM_XE_DEVICE_QUERY_ENGINES, then the reply uses an array of
264  * struct @drm_xe_query_engines in .data.
265  */
266 struct drm_xe_query_engines {
267 	/** @num_engines: number of engines returned in @engines */
268 	__u32 num_engines;
269 	/** @pad: MBZ */
270 	__u32 pad;
271 	/** @engines: The returned engines for this device */
272 	struct drm_xe_engine engines[];
273 };
274 
275 /**
276  * enum drm_xe_memory_class - Supported memory classes.
277  */
278 enum drm_xe_memory_class {
279 	/** @DRM_XE_MEM_REGION_CLASS_SYSMEM: Represents system memory. */
280 	DRM_XE_MEM_REGION_CLASS_SYSMEM = 0,
281 	/**
282 	 * @DRM_XE_MEM_REGION_CLASS_VRAM: On discrete platforms, this
283 	 * represents the memory that is local to the device, which we
284 	 * call VRAM. Not valid on integrated platforms.
285 	 */
286 	DRM_XE_MEM_REGION_CLASS_VRAM
287 };
288 
289 /**
290  * struct drm_xe_mem_region - Describes some region as known to
291  * the driver.
292  */
293 struct drm_xe_mem_region {
294 	/**
295 	 * @mem_class: The memory class describing this region.
296 	 *
297 	 * See enum drm_xe_memory_class for supported values.
298 	 */
299 	__u16 mem_class;
300 	/**
301 	 * @instance: The unique ID for this region, which serves as the
302 	 * index in the placement bitmask used as argument for
303 	 * &DRM_IOCTL_XE_GEM_CREATE
304 	 */
305 	__u16 instance;
306 	/**
307 	 * @min_page_size: Min page-size in bytes for this region.
308 	 *
309 	 * When the kernel allocates memory for this region, the
310 	 * underlying pages will be at least @min_page_size in size.
311 	 * Buffer objects with an allowable placement in this region must be
312 	 * created with a size aligned to this value.
313 	 * GPU virtual address mappings of (parts of) buffer objects that
314 	 * may be placed in this region must also have their GPU virtual
315 	 * address and range aligned to this value.
316 	 * Affected IOCTLS will return %-EINVAL if alignment restrictions are
317 	 * not met.
318 	 */
319 	__u32 min_page_size;
320 	/**
321 	 * @total_size: The usable size in bytes for this region.
322 	 */
323 	__u64 total_size;
324 	/**
325 	 * @used: Estimate of the memory used in bytes for this region.
326 	 *
327 	 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable
328 	 * accounting.  Without this the value here will always equal
329 	 * zero.
330 	 */
331 	__u64 used;
332 	/**
333 	 * @cpu_visible_size: How much of this region can be CPU
334 	 * accessed, in bytes.
335 	 *
336 	 * This will always be <= @total_size, and the remainder (if
337 	 * any) will not be CPU accessible. If the CPU accessible part
338 	 * is smaller than @total_size then this is referred to as a
339 	 * small BAR system.
340 	 *
341 	 * On systems without small BAR (full BAR), the probed_size will
342 	 * always equal the @total_size, since all of it will be CPU
343 	 * accessible.
344 	 *
345 	 * Note this is only tracked for DRM_XE_MEM_REGION_CLASS_VRAM
346 	 * regions (for other types the value here will always equal
347 	 * zero).
348 	 */
349 	__u64 cpu_visible_size;
350 	/**
351 	 * @cpu_visible_used: Estimate of CPU visible memory used, in
352 	 * bytes.
353 	 *
354 	 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable
355 	 * accounting. Without this the value here will always equal
356 	 * zero.  Note this is only currently tracked for
357 	 * DRM_XE_MEM_REGION_CLASS_VRAM regions (for other types the value
358 	 * here will always be zero).
359 	 */
360 	__u64 cpu_visible_used;
361 	/** @reserved: Reserved */
362 	__u64 reserved[6];
363 };
364 
365 /**
366  * struct drm_xe_query_mem_regions - describe memory regions
367  *
368  * If a query is made with a struct drm_xe_device_query where .query
369  * is equal to DRM_XE_DEVICE_QUERY_MEM_REGIONS, then the reply uses
370  * struct drm_xe_query_mem_regions in .data.
371  */
372 struct drm_xe_query_mem_regions {
373 	/** @num_mem_regions: number of memory regions returned in @mem_regions */
374 	__u32 num_mem_regions;
375 	/** @pad: MBZ */
376 	__u32 pad;
377 	/** @mem_regions: The returned memory regions for this device */
378 	struct drm_xe_mem_region mem_regions[];
379 };
380 
381 /**
382  * struct drm_xe_query_config - describe the device configuration
383  *
384  * If a query is made with a struct drm_xe_device_query where .query
385  * is equal to DRM_XE_DEVICE_QUERY_CONFIG, then the reply uses
386  * struct drm_xe_query_config in .data.
387  *
388  * The index in @info can be:
389  *  - %DRM_XE_QUERY_CONFIG_REV_AND_DEVICE_ID - Device ID (lower 16 bits)
390  *    and the device revision (next 8 bits)
391  *  - %DRM_XE_QUERY_CONFIG_FLAGS - Flags describing the device
392  *    configuration, see list below
393  *
394  *    - %DRM_XE_QUERY_CONFIG_FLAG_HAS_VRAM - Flag is set if the device
395  *      has usable VRAM
396  *  - %DRM_XE_QUERY_CONFIG_MIN_ALIGNMENT - Minimal memory alignment
397  *    required by this device, typically SZ_4K or SZ_64K
398  *  - %DRM_XE_QUERY_CONFIG_VA_BITS - Maximum bits of a virtual address
399  *  - %DRM_XE_QUERY_CONFIG_MAX_EXEC_QUEUE_PRIORITY - Value of the highest
400  *    available exec queue priority
401  */
402 struct drm_xe_query_config {
403 	/** @num_params: number of parameters returned in info */
404 	__u32 num_params;
405 
406 	/** @pad: MBZ */
407 	__u32 pad;
408 
409 #define DRM_XE_QUERY_CONFIG_REV_AND_DEVICE_ID	0
410 #define DRM_XE_QUERY_CONFIG_FLAGS			1
411 	#define DRM_XE_QUERY_CONFIG_FLAG_HAS_VRAM	(1 << 0)
412 #define DRM_XE_QUERY_CONFIG_MIN_ALIGNMENT		2
413 #define DRM_XE_QUERY_CONFIG_VA_BITS			3
414 #define DRM_XE_QUERY_CONFIG_MAX_EXEC_QUEUE_PRIORITY	4
415 	/** @info: array of elements containing the config info */
416 	__u64 info[];
417 };
418 
419 /**
420  * struct drm_xe_gt - describe an individual GT.
421  *
422  * To be used with drm_xe_query_gt_list, which will return a list with all the
423  * existing GT individual descriptions.
424  * Graphics Technology (GT) is a subset of a GPU/tile that is responsible for
425  * implementing graphics and/or media operations.
426  *
427  * The index in @type can be:
428  *  - %DRM_XE_QUERY_GT_TYPE_MAIN
429  *  - %DRM_XE_QUERY_GT_TYPE_MEDIA
430  */
431 struct drm_xe_gt {
432 #define DRM_XE_QUERY_GT_TYPE_MAIN		0
433 #define DRM_XE_QUERY_GT_TYPE_MEDIA		1
434 	/** @type: GT type: Main or Media */
435 	__u16 type;
436 	/** @tile_id: Tile ID where this GT lives (Information only) */
437 	__u16 tile_id;
438 	/** @gt_id: Unique ID of this GT within the PCI Device */
439 	__u16 gt_id;
440 	/** @pad: MBZ */
441 	__u16 pad[3];
442 	/** @reference_clock: A clock frequency for timestamp */
443 	__u32 reference_clock;
444 	/**
445 	 * @near_mem_regions: Bit mask of instances from
446 	 * drm_xe_query_mem_regions that are nearest to the current engines
447 	 * of this GT.
448 	 * Each index in this mask refers directly to the struct
449 	 * drm_xe_query_mem_regions' instance, no assumptions should
450 	 * be made about order. The type of each region is described
451 	 * by struct drm_xe_query_mem_regions' mem_class.
452 	 */
453 	__u64 near_mem_regions;
454 	/**
455 	 * @far_mem_regions: Bit mask of instances from
456 	 * drm_xe_query_mem_regions that are far from the engines of this GT.
457 	 * In general, they have extra indirections when compared to the
458 	 * @near_mem_regions. For a discrete device this could mean system
459 	 * memory and memory living in a different tile.
460 	 * Each index in this mask refers directly to the struct
461 	 * drm_xe_query_mem_regions' instance, no assumptions should
462 	 * be made about order. The type of each region is described
463 	 * by struct drm_xe_query_mem_regions' mem_class.
464 	 */
465 	__u64 far_mem_regions;
466 	/** @ip_ver_major: Graphics/media IP major version on GMD_ID platforms */
467 	__u16 ip_ver_major;
468 	/** @ip_ver_minor: Graphics/media IP minor version on GMD_ID platforms */
469 	__u16 ip_ver_minor;
470 	/** @ip_ver_rev: Graphics/media IP revision version on GMD_ID platforms */
471 	__u16 ip_ver_rev;
472 	/** @pad2: MBZ */
473 	__u16 pad2;
474 	/** @reserved: Reserved */
475 	__u64 reserved[7];
476 };
477 
478 /**
479  * struct drm_xe_query_gt_list - A list with GT description items.
480  *
481  * If a query is made with a struct drm_xe_device_query where .query
482  * is equal to DRM_XE_DEVICE_QUERY_GT_LIST, then the reply uses struct
483  * drm_xe_query_gt_list in .data.
484  */
485 struct drm_xe_query_gt_list {
486 	/** @num_gt: number of GT items returned in gt_list */
487 	__u32 num_gt;
488 	/** @pad: MBZ */
489 	__u32 pad;
490 	/** @gt_list: The GT list returned for this device */
491 	struct drm_xe_gt gt_list[];
492 };
493 
494 /**
495  * struct drm_xe_query_topology_mask - describe the topology mask of a GT
496  *
497  * This is the hardware topology which reflects the internal physical
498  * structure of the GPU.
499  *
500  * If a query is made with a struct drm_xe_device_query where .query
501  * is equal to DRM_XE_DEVICE_QUERY_GT_TOPOLOGY, then the reply uses
502  * struct drm_xe_query_topology_mask in .data.
503  *
504  * The @type can be:
505  *  - %DRM_XE_TOPO_DSS_GEOMETRY - To query the mask of Dual Sub Slices
506  *    (DSS) available for geometry operations. For example a query response
507  *    containing the following in mask:
508  *    ``DSS_GEOMETRY    ff ff ff ff 00 00 00 00``
509  *    means 32 DSS are available for geometry.
510  *  - %DRM_XE_TOPO_DSS_COMPUTE - To query the mask of Dual Sub Slices
511  *    (DSS) available for compute operations. For example a query response
512  *    containing the following in mask:
513  *    ``DSS_COMPUTE    ff ff ff ff 00 00 00 00``
514  *    means 32 DSS are available for compute.
515  *  - %DRM_XE_TOPO_L3_BANK - To query the mask of enabled L3 banks
516  *  - %DRM_XE_TOPO_EU_PER_DSS - To query the mask of Execution Units (EU)
517  *    available per Dual Sub Slices (DSS). For example a query response
518  *    containing the following in mask:
519  *    ``EU_PER_DSS    ff ff 00 00 00 00 00 00``
520  *    means each DSS has 16 EU.
521  */
522 struct drm_xe_query_topology_mask {
523 	/** @gt_id: GT ID the mask is associated with */
524 	__u16 gt_id;
525 
526 #define DRM_XE_TOPO_DSS_GEOMETRY	1
527 #define DRM_XE_TOPO_DSS_COMPUTE		2
528 #define DRM_XE_TOPO_L3_BANK		3
529 #define DRM_XE_TOPO_EU_PER_DSS		4
530 	/** @type: type of mask */
531 	__u16 type;
532 
533 	/** @num_bytes: number of bytes in requested mask */
534 	__u32 num_bytes;
535 
536 	/** @mask: little-endian mask of @num_bytes */
537 	__u8 mask[];
538 };
539 
540 /**
541  * struct drm_xe_query_engine_cycles - correlate CPU and GPU timestamps
542  *
543  * If a query is made with a struct drm_xe_device_query where .query is equal to
544  * DRM_XE_DEVICE_QUERY_ENGINE_CYCLES, then the reply uses struct drm_xe_query_engine_cycles
545  * in .data. struct drm_xe_query_engine_cycles is allocated by the user and
546  * .data points to this allocated structure.
547  *
548  * The query returns the engine cycles, which along with GT's @reference_clock,
549  * can be used to calculate the engine timestamp. In addition the
550  * query returns a set of cpu timestamps that indicate when the command
551  * streamer cycle count was captured.
552  */
553 struct drm_xe_query_engine_cycles {
554 	/**
555 	 * @eci: This is input by the user and is the engine for which command
556 	 * streamer cycles is queried.
557 	 */
558 	struct drm_xe_engine_class_instance eci;
559 
560 	/**
561 	 * @clockid: This is input by the user and is the reference clock id for
562 	 * CPU timestamp. For definition, see clock_gettime(2) and
563 	 * perf_event_open(2). Supported clock ids are CLOCK_MONOTONIC,
564 	 * CLOCK_MONOTONIC_RAW, CLOCK_REALTIME, CLOCK_BOOTTIME, CLOCK_TAI.
565 	 */
566 	__s32 clockid;
567 
568 	/** @width: Width of the engine cycle counter in bits. */
569 	__u32 width;
570 
571 	/**
572 	 * @engine_cycles: Engine cycles as read from its register
573 	 * at 0x358 offset.
574 	 */
575 	__u64 engine_cycles;
576 
577 	/**
578 	 * @cpu_timestamp: CPU timestamp in ns. The timestamp is captured before
579 	 * reading the engine_cycles register using the reference clockid set by the
580 	 * user.
581 	 */
582 	__u64 cpu_timestamp;
583 
584 	/**
585 	 * @cpu_delta: Time delta in ns captured around reading the lower dword
586 	 * of the engine_cycles register.
587 	 */
588 	__u64 cpu_delta;
589 };
590 
591 /**
592  * struct drm_xe_query_uc_fw_version - query a micro-controller firmware version
593  *
594  * Given a uc_type this will return the branch, major, minor and patch version
595  * of the micro-controller firmware.
596  */
597 struct drm_xe_query_uc_fw_version {
598 	/** @uc_type: The micro-controller type to query firmware version */
599 #define XE_QUERY_UC_TYPE_GUC_SUBMISSION 0
600 #define XE_QUERY_UC_TYPE_HUC 1
601 	__u16 uc_type;
602 
603 	/** @pad: MBZ */
604 	__u16 pad;
605 
606 	/** @branch_ver: branch uc fw version */
607 	__u32 branch_ver;
608 	/** @major_ver: major uc fw version */
609 	__u32 major_ver;
610 	/** @minor_ver: minor uc fw version */
611 	__u32 minor_ver;
612 	/** @patch_ver: patch uc fw version */
613 	__u32 patch_ver;
614 
615 	/** @pad2: MBZ */
616 	__u32 pad2;
617 
618 	/** @reserved: Reserved */
619 	__u64 reserved;
620 };
621 
622 /**
623  * struct drm_xe_device_query - Input of &DRM_IOCTL_XE_DEVICE_QUERY - main
624  * structure to query device information
625  *
626  * The user selects the type of data to query among DRM_XE_DEVICE_QUERY_*
627  * and sets the value in the query member. This determines the type of
628  * the structure provided by the driver in data, among struct drm_xe_query_*.
629  *
630  * The @query can be:
631  *  - %DRM_XE_DEVICE_QUERY_ENGINES
632  *  - %DRM_XE_DEVICE_QUERY_MEM_REGIONS
633  *  - %DRM_XE_DEVICE_QUERY_CONFIG
634  *  - %DRM_XE_DEVICE_QUERY_GT_LIST
635  *  - %DRM_XE_DEVICE_QUERY_HWCONFIG - Query type to retrieve the hardware
636  *    configuration of the device such as information on slices, memory,
637  *    caches, and so on. It is provided as a table of key / value
638  *    attributes.
639  *  - %DRM_XE_DEVICE_QUERY_GT_TOPOLOGY
640  *  - %DRM_XE_DEVICE_QUERY_ENGINE_CYCLES
641  *
642  * If size is set to 0, the driver fills it with the required size for
643  * the requested type of data to query. If size is equal to the required
644  * size, the queried information is copied into data. If size is set to
645  * a value different from 0 and different from the required size, the
646  * IOCTL call returns -EINVAL.
647  *
648  * For example the following code snippet allows retrieving and printing
649  * information about the device engines with DRM_XE_DEVICE_QUERY_ENGINES:
650  *
651  * .. code-block:: C
652  *
653  *     struct drm_xe_query_engines *engines;
654  *     struct drm_xe_device_query query = {
655  *         .extensions = 0,
656  *         .query = DRM_XE_DEVICE_QUERY_ENGINES,
657  *         .size = 0,
658  *         .data = 0,
659  *     };
660  *     ioctl(fd, DRM_IOCTL_XE_DEVICE_QUERY, &query);
661  *     engines = malloc(query.size);
662  *     query.data = (uintptr_t)engines;
663  *     ioctl(fd, DRM_IOCTL_XE_DEVICE_QUERY, &query);
664  *     for (int i = 0; i < engines->num_engines; i++) {
665  *         printf("Engine %d: %s\n", i,
666  *             engines->engines[i].instance.engine_class ==
667  *                 DRM_XE_ENGINE_CLASS_RENDER ? "RENDER":
668  *             engines->engines[i].instance.engine_class ==
669  *                 DRM_XE_ENGINE_CLASS_COPY ? "COPY":
670  *             engines->engines[i].instance.engine_class ==
671  *                 DRM_XE_ENGINE_CLASS_VIDEO_DECODE ? "VIDEO_DECODE":
672  *             engines->engines[i].instance.engine_class ==
673  *                 DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE ? "VIDEO_ENHANCE":
674  *             engines->engines[i].instance.engine_class ==
675  *                 DRM_XE_ENGINE_CLASS_COMPUTE ? "COMPUTE":
676  *             "UNKNOWN");
677  *     }
678  *     free(engines);
679  */
680 struct drm_xe_device_query {
681 	/** @extensions: Pointer to the first extension struct, if any */
682 	__u64 extensions;
683 
684 #define DRM_XE_DEVICE_QUERY_ENGINES		0
685 #define DRM_XE_DEVICE_QUERY_MEM_REGIONS		1
686 #define DRM_XE_DEVICE_QUERY_CONFIG		2
687 #define DRM_XE_DEVICE_QUERY_GT_LIST		3
688 #define DRM_XE_DEVICE_QUERY_HWCONFIG		4
689 #define DRM_XE_DEVICE_QUERY_GT_TOPOLOGY		5
690 #define DRM_XE_DEVICE_QUERY_ENGINE_CYCLES	6
691 #define DRM_XE_DEVICE_QUERY_UC_FW_VERSION	7
692 #define DRM_XE_DEVICE_QUERY_OA_UNITS		8
693 	/** @query: The type of data to query */
694 	__u32 query;
695 
696 	/** @size: Size of the queried data */
697 	__u32 size;
698 
699 	/** @data: Queried data is placed here */
700 	__u64 data;
701 
702 	/** @reserved: Reserved */
703 	__u64 reserved[2];
704 };
705 
706 /**
707  * struct drm_xe_gem_create - Input of &DRM_IOCTL_XE_GEM_CREATE - A structure for
708  * gem creation
709  *
710  * The @flags can be:
711  *  - %DRM_XE_GEM_CREATE_FLAG_DEFER_BACKING
712  *  - %DRM_XE_GEM_CREATE_FLAG_SCANOUT
713  *  - %DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM - When using VRAM as a
714  *    possible placement, ensure that the corresponding VRAM allocation
715  *    will always use the CPU accessible part of VRAM. This is important
716  *    for small-bar systems (on full-bar systems this gets turned into a
717  *    noop).
718  *    Note1: System memory can be used as an extra placement if the kernel
719  *    should spill the allocation to system memory, if space can't be made
720  *    available in the CPU accessible part of VRAM (giving the same
721  *    behaviour as the i915 interface, see
722  *    I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS).
723  *    Note2: For clear-color CCS surfaces the kernel needs to read the
724  *    clear-color value stored in the buffer, and on discrete platforms we
725  *    need to use VRAM for display surfaces, therefore the kernel requires
726  *    setting this flag for such objects, otherwise an error is thrown on
727  *    small-bar systems.
728  *
729  * @cpu_caching supports the following values:
730  *  - %DRM_XE_GEM_CPU_CACHING_WB - Allocate the pages with write-back
731  *    caching. On iGPU this can't be used for scanout surfaces. Currently
732  *    not allowed for objects placed in VRAM.
733  *  - %DRM_XE_GEM_CPU_CACHING_WC - Allocate the pages as write-combined. This
734  *    is uncached. Scanout surfaces should likely use this. All objects
735  *    that can be placed in VRAM must use this.
736  */
737 struct drm_xe_gem_create {
738 	/** @extensions: Pointer to the first extension struct, if any */
739 	__u64 extensions;
740 
741 	/**
742 	 * @size: Size of the object to be created, must match region
743 	 * (system or vram) minimum alignment (&min_page_size).
744 	 */
745 	__u64 size;
746 
747 	/**
748 	 * @placement: A mask of memory instances of where BO can be placed.
749 	 * Each index in this mask refers directly to the struct
750 	 * drm_xe_query_mem_regions' instance, no assumptions should
751 	 * be made about order. The type of each region is described
752 	 * by struct drm_xe_query_mem_regions' mem_class.
753 	 */
754 	__u32 placement;
755 
756 #define DRM_XE_GEM_CREATE_FLAG_DEFER_BACKING		(1 << 0)
757 #define DRM_XE_GEM_CREATE_FLAG_SCANOUT			(1 << 1)
758 #define DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM	(1 << 2)
759 	/**
760 	 * @flags: Flags, currently a mask of memory instances of where BO can
761 	 * be placed
762 	 */
763 	__u32 flags;
764 
765 	/**
766 	 * @vm_id: Attached VM, if any
767 	 *
768 	 * If a VM is specified, this BO must:
769 	 *
770 	 *  1. Only ever be bound to that VM.
771 	 *  2. Cannot be exported as a PRIME fd.
772 	 */
773 	__u32 vm_id;
774 
775 	/**
776 	 * @handle: Returned handle for the object.
777 	 *
778 	 * Object handles are nonzero.
779 	 */
780 	__u32 handle;
781 
782 #define DRM_XE_GEM_CPU_CACHING_WB                      1
783 #define DRM_XE_GEM_CPU_CACHING_WC                      2
784 	/**
785 	 * @cpu_caching: The CPU caching mode to select for this object. If
786 	 * mmaping the object the mode selected here will also be used. The
787 	 * exception is when mapping system memory (including data evicted
788 	 * to system) on discrete GPUs. The caching mode selected will
789 	 * then be overridden to DRM_XE_GEM_CPU_CACHING_WB, and coherency
790 	 * between GPU- and CPU is guaranteed. The caching mode of
791 	 * existing CPU-mappings will be updated transparently to
792 	 * user-space clients.
793 	 */
794 	__u16 cpu_caching;
795 	/** @pad: MBZ */
796 	__u16 pad[3];
797 
798 	/** @reserved: Reserved */
799 	__u64 reserved[2];
800 };
801 
802 /**
803  * struct drm_xe_gem_mmap_offset - Input of &DRM_IOCTL_XE_GEM_MMAP_OFFSET
804  */
805 struct drm_xe_gem_mmap_offset {
806 	/** @extensions: Pointer to the first extension struct, if any */
807 	__u64 extensions;
808 
809 	/** @handle: Handle for the object being mapped. */
810 	__u32 handle;
811 
812 	/** @flags: Must be zero */
813 	__u32 flags;
814 
815 	/** @offset: The fake offset to use for subsequent mmap call */
816 	__u64 offset;
817 
818 	/** @reserved: Reserved */
819 	__u64 reserved[2];
820 };
821 
822 /**
823  * struct drm_xe_vm_create - Input of &DRM_IOCTL_XE_VM_CREATE
824  *
825  * The @flags can be:
826  *  - %DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE
827  *  - %DRM_XE_VM_CREATE_FLAG_LR_MODE - An LR, or Long Running VM accepts
828  *    exec submissions to its exec_queues that don't have an upper time
829  *    limit on the job execution time. But exec submissions to these
830  *    don't allow any of the flags DRM_XE_SYNC_FLAG_SYNCOBJ,
831  *    DRM_XE_SYNC_FLAG_TIMELINE_SYNCOBJ, DRM_XE_SYNC_FLAG_DMA_BUF,
832  *    used as out-syncobjs, that is, together with DRM_XE_SYNC_FLAG_SIGNAL.
833  *    LR VMs can be created in recoverable page-fault mode using
834  *    DRM_XE_VM_CREATE_FLAG_FAULT_MODE, if the device supports it.
835  *    If that flag is omitted, the UMD can not rely on the slightly
836  *    different per-VM overcommit semantics that are enabled by
837  *    DRM_XE_VM_CREATE_FLAG_FAULT_MODE (see below), but KMD may
838  *    still enable recoverable pagefaults if supported by the device.
839  *  - %DRM_XE_VM_CREATE_FLAG_FAULT_MODE - Requires also
840  *    DRM_XE_VM_CREATE_FLAG_LR_MODE. It allows memory to be allocated on
841  *    demand when accessed, and also allows per-VM overcommit of memory.
842  *    The xe driver internally uses recoverable pagefaults to implement
843  *    this.
844  */
845 struct drm_xe_vm_create {
846 	/** @extensions: Pointer to the first extension struct, if any */
847 	__u64 extensions;
848 
849 #define DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE	(1 << 0)
850 #define DRM_XE_VM_CREATE_FLAG_LR_MODE	        (1 << 1)
851 #define DRM_XE_VM_CREATE_FLAG_FAULT_MODE	(1 << 2)
852 	/** @flags: Flags */
853 	__u32 flags;
854 
855 	/** @vm_id: Returned VM ID */
856 	__u32 vm_id;
857 
858 	/** @reserved: Reserved */
859 	__u64 reserved[2];
860 };
861 
862 /**
863  * struct drm_xe_vm_destroy - Input of &DRM_IOCTL_XE_VM_DESTROY
864  */
865 struct drm_xe_vm_destroy {
866 	/** @vm_id: VM ID */
867 	__u32 vm_id;
868 
869 	/** @pad: MBZ */
870 	__u32 pad;
871 
872 	/** @reserved: Reserved */
873 	__u64 reserved[2];
874 };
875 
876 /**
877  * struct drm_xe_vm_bind_op - run bind operations
878  *
879  * The @op can be:
880  *  - %DRM_XE_VM_BIND_OP_MAP
881  *  - %DRM_XE_VM_BIND_OP_UNMAP
882  *  - %DRM_XE_VM_BIND_OP_MAP_USERPTR
883  *  - %DRM_XE_VM_BIND_OP_UNMAP_ALL
884  *  - %DRM_XE_VM_BIND_OP_PREFETCH
885  *
886  * and the @flags can be:
887  *  - %DRM_XE_VM_BIND_FLAG_READONLY - Setup the page tables as read-only
888  *    to ensure write protection
889  *  - %DRM_XE_VM_BIND_FLAG_IMMEDIATE - On a faulting VM, do the
890  *    MAP operation immediately rather than deferring the MAP to the page
891  *    fault handler. This is implied on a non-faulting VM as there is no
892  *    fault handler to defer to.
893  *  - %DRM_XE_VM_BIND_FLAG_NULL - When the NULL flag is set, the page
894  *    tables are setup with a special bit which indicates writes are
895  *    dropped and all reads return zero. In the future, the NULL flags
896  *    will only be valid for DRM_XE_VM_BIND_OP_MAP operations, the BO
897  *    handle MBZ, and the BO offset MBZ. This flag is intended to
898  *    implement VK sparse bindings.
899  */
900 struct drm_xe_vm_bind_op {
901 	/** @extensions: Pointer to the first extension struct, if any */
902 	__u64 extensions;
903 
904 	/**
905 	 * @obj: GEM object to operate on, MBZ for MAP_USERPTR, MBZ for UNMAP
906 	 */
907 	__u32 obj;
908 
909 	/**
910 	 * @pat_index: The platform defined @pat_index to use for this mapping.
911 	 * The index basically maps to some predefined memory attributes,
912 	 * including things like caching, coherency, compression etc.  The exact
913 	 * meaning of the pat_index is platform specific and defined in the
914 	 * Bspec and PRMs.  When the KMD sets up the binding the index here is
915 	 * encoded into the ppGTT PTE.
916 	 *
917 	 * For coherency the @pat_index needs to be at least 1way coherent when
918 	 * drm_xe_gem_create.cpu_caching is DRM_XE_GEM_CPU_CACHING_WB. The KMD
919 	 * will extract the coherency mode from the @pat_index and reject if
920 	 * there is a mismatch (see note below for pre-MTL platforms).
921 	 *
922 	 * Note: On pre-MTL platforms there is only a caching mode and no
923 	 * explicit coherency mode, but on such hardware there is always a
924 	 * shared-LLC (or is dgpu) so all GT memory accesses are coherent with
925 	 * CPU caches even with the caching mode set as uncached.  It's only the
926 	 * display engine that is incoherent (on dgpu it must be in VRAM which
927 	 * is always mapped as WC on the CPU). However to keep the uapi somewhat
928 	 * consistent with newer platforms the KMD groups the different cache
929 	 * levels into the following coherency buckets on all pre-MTL platforms:
930 	 *
931 	 *	ppGTT UC -> COH_NONE
932 	 *	ppGTT WC -> COH_NONE
933 	 *	ppGTT WT -> COH_NONE
934 	 *	ppGTT WB -> COH_AT_LEAST_1WAY
935 	 *
936 	 * In practice UC/WC/WT should only ever used for scanout surfaces on
937 	 * such platforms (or perhaps in general for dma-buf if shared with
938 	 * another device) since it is only the display engine that is actually
939 	 * incoherent.  Everything else should typically use WB given that we
940 	 * have a shared-LLC.  On MTL+ this completely changes and the HW
941 	 * defines the coherency mode as part of the @pat_index, where
942 	 * incoherent GT access is possible.
943 	 *
944 	 * Note: For userptr and externally imported dma-buf the kernel expects
945 	 * either 1WAY or 2WAY for the @pat_index.
946 	 *
947 	 * For DRM_XE_VM_BIND_FLAG_NULL bindings there are no KMD restrictions
948 	 * on the @pat_index. For such mappings there is no actual memory being
949 	 * mapped (the address in the PTE is invalid), so the various PAT memory
950 	 * attributes likely do not apply.  Simply leaving as zero is one
951 	 * option (still a valid pat_index).
952 	 */
953 	__u16 pat_index;
954 
955 	/** @pad: MBZ */
956 	__u16 pad;
957 
958 	union {
959 		/**
960 		 * @obj_offset: Offset into the object, MBZ for CLEAR_RANGE,
961 		 * ignored for unbind
962 		 */
963 		__u64 obj_offset;
964 
965 		/** @userptr: user pointer to bind on */
966 		__u64 userptr;
967 	};
968 
969 	/**
970 	 * @range: Number of bytes from the object to bind to addr, MBZ for UNMAP_ALL
971 	 */
972 	__u64 range;
973 
974 	/** @addr: Address to operate on, MBZ for UNMAP_ALL */
975 	__u64 addr;
976 
977 #define DRM_XE_VM_BIND_OP_MAP		0x0
978 #define DRM_XE_VM_BIND_OP_UNMAP		0x1
979 #define DRM_XE_VM_BIND_OP_MAP_USERPTR	0x2
980 #define DRM_XE_VM_BIND_OP_UNMAP_ALL	0x3
981 #define DRM_XE_VM_BIND_OP_PREFETCH	0x4
982 	/** @op: Bind operation to perform */
983 	__u32 op;
984 
985 #define DRM_XE_VM_BIND_FLAG_READONLY	(1 << 0)
986 #define DRM_XE_VM_BIND_FLAG_IMMEDIATE	(1 << 1)
987 #define DRM_XE_VM_BIND_FLAG_NULL	(1 << 2)
988 #define DRM_XE_VM_BIND_FLAG_DUMPABLE	(1 << 3)
989 	/** @flags: Bind flags */
990 	__u32 flags;
991 
992 	/**
993 	 * @prefetch_mem_region_instance: Memory region to prefetch VMA to.
994 	 * It is a region instance, not a mask.
995 	 * To be used only with %DRM_XE_VM_BIND_OP_PREFETCH operation.
996 	 */
997 	__u32 prefetch_mem_region_instance;
998 
999 	/** @pad2: MBZ */
1000 	__u32 pad2;
1001 
1002 	/** @reserved: Reserved */
1003 	__u64 reserved[3];
1004 };
1005 
1006 /**
1007  * struct drm_xe_vm_bind - Input of &DRM_IOCTL_XE_VM_BIND
1008  *
1009  * Below is an example of a minimal use of @drm_xe_vm_bind to
1010  * asynchronously bind the buffer `data` at address `BIND_ADDRESS` to
1011  * illustrate `userptr`. It can be synchronized by using the example
1012  * provided for @drm_xe_sync.
1013  *
1014  * .. code-block:: C
1015  *
1016  *     data = aligned_alloc(ALIGNMENT, BO_SIZE);
1017  *     struct drm_xe_vm_bind bind = {
1018  *         .vm_id = vm,
1019  *         .num_binds = 1,
1020  *         .bind.obj = 0,
1021  *         .bind.obj_offset = to_user_pointer(data),
1022  *         .bind.range = BO_SIZE,
1023  *         .bind.addr = BIND_ADDRESS,
1024  *         .bind.op = DRM_XE_VM_BIND_OP_MAP_USERPTR,
1025  *         .bind.flags = 0,
1026  *         .num_syncs = 1,
1027  *         .syncs = &sync,
1028  *         .exec_queue_id = 0,
1029  *     };
1030  *     ioctl(fd, DRM_IOCTL_XE_VM_BIND, &bind);
1031  *
1032  */
1033 struct drm_xe_vm_bind {
1034 	/** @extensions: Pointer to the first extension struct, if any */
1035 	__u64 extensions;
1036 
1037 	/** @vm_id: The ID of the VM to bind to */
1038 	__u32 vm_id;
1039 
1040 	/**
1041 	 * @exec_queue_id: exec_queue_id, must be of class DRM_XE_ENGINE_CLASS_VM_BIND
1042 	 * and exec queue must have same vm_id. If zero, the default VM bind engine
1043 	 * is used.
1044 	 */
1045 	__u32 exec_queue_id;
1046 
1047 	/** @pad: MBZ */
1048 	__u32 pad;
1049 
1050 	/** @num_binds: number of binds in this IOCTL */
1051 	__u32 num_binds;
1052 
1053 	union {
1054 		/** @bind: used if num_binds == 1 */
1055 		struct drm_xe_vm_bind_op bind;
1056 
1057 		/**
1058 		 * @vector_of_binds: userptr to array of struct
1059 		 * drm_xe_vm_bind_op if num_binds > 1
1060 		 */
1061 		__u64 vector_of_binds;
1062 	};
1063 
1064 	/** @pad2: MBZ */
1065 	__u32 pad2;
1066 
1067 	/** @num_syncs: amount of syncs to wait on */
1068 	__u32 num_syncs;
1069 
1070 	/** @syncs: pointer to struct drm_xe_sync array */
1071 	__u64 syncs;
1072 
1073 	/** @reserved: Reserved */
1074 	__u64 reserved[2];
1075 };
1076 
1077 /**
1078  * struct drm_xe_exec_queue_create - Input of &DRM_IOCTL_XE_EXEC_QUEUE_CREATE
1079  *
1080  * The example below shows how to use @drm_xe_exec_queue_create to create
1081  * a simple exec_queue (no parallel submission) of class
1082  * &DRM_XE_ENGINE_CLASS_RENDER.
1083  *
1084  * .. code-block:: C
1085  *
1086  *     struct drm_xe_engine_class_instance instance = {
1087  *         .engine_class = DRM_XE_ENGINE_CLASS_RENDER,
1088  *     };
1089  *     struct drm_xe_exec_queue_create exec_queue_create = {
1090  *          .extensions = 0,
1091  *          .vm_id = vm,
1092  *          .num_bb_per_exec = 1,
1093  *          .num_eng_per_bb = 1,
1094  *          .instances = to_user_pointer(&instance),
1095  *     };
1096  *     ioctl(fd, DRM_IOCTL_XE_EXEC_QUEUE_CREATE, &exec_queue_create);
1097  *
1098  */
1099 struct drm_xe_exec_queue_create {
1100 #define DRM_XE_EXEC_QUEUE_EXTENSION_SET_PROPERTY		0
1101 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_PRIORITY		0
1102 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_TIMESLICE		1
1103 
1104 	/** @extensions: Pointer to the first extension struct, if any */
1105 	__u64 extensions;
1106 
1107 	/** @width: submission width (number BB per exec) for this exec queue */
1108 	__u16 width;
1109 
1110 	/** @num_placements: number of valid placements for this exec queue */
1111 	__u16 num_placements;
1112 
1113 	/** @vm_id: VM to use for this exec queue */
1114 	__u32 vm_id;
1115 
1116 	/** @flags: MBZ */
1117 	__u32 flags;
1118 
1119 	/** @exec_queue_id: Returned exec queue ID */
1120 	__u32 exec_queue_id;
1121 
1122 	/**
1123 	 * @instances: user pointer to a 2-d array of struct
1124 	 * drm_xe_engine_class_instance
1125 	 *
1126 	 * length = width (i) * num_placements (j)
1127 	 * index = j + i * width
1128 	 */
1129 	__u64 instances;
1130 
1131 	/** @reserved: Reserved */
1132 	__u64 reserved[2];
1133 };
1134 
1135 /**
1136  * struct drm_xe_exec_queue_destroy - Input of &DRM_IOCTL_XE_EXEC_QUEUE_DESTROY
1137  */
1138 struct drm_xe_exec_queue_destroy {
1139 	/** @exec_queue_id: Exec queue ID */
1140 	__u32 exec_queue_id;
1141 
1142 	/** @pad: MBZ */
1143 	__u32 pad;
1144 
1145 	/** @reserved: Reserved */
1146 	__u64 reserved[2];
1147 };
1148 
1149 /**
1150  * struct drm_xe_exec_queue_get_property - Input of &DRM_IOCTL_XE_EXEC_QUEUE_GET_PROPERTY
1151  *
1152  * The @property can be:
1153  *  - %DRM_XE_EXEC_QUEUE_GET_PROPERTY_BAN
1154  */
1155 struct drm_xe_exec_queue_get_property {
1156 	/** @extensions: Pointer to the first extension struct, if any */
1157 	__u64 extensions;
1158 
1159 	/** @exec_queue_id: Exec queue ID */
1160 	__u32 exec_queue_id;
1161 
1162 #define DRM_XE_EXEC_QUEUE_GET_PROPERTY_BAN	0
1163 	/** @property: property to get */
1164 	__u32 property;
1165 
1166 	/** @value: property value */
1167 	__u64 value;
1168 
1169 	/** @reserved: Reserved */
1170 	__u64 reserved[2];
1171 };
1172 
1173 /**
1174  * struct drm_xe_sync - sync object
1175  *
1176  * The @type can be:
1177  *  - %DRM_XE_SYNC_TYPE_SYNCOBJ
1178  *  - %DRM_XE_SYNC_TYPE_TIMELINE_SYNCOBJ
1179  *  - %DRM_XE_SYNC_TYPE_USER_FENCE
1180  *
1181  * and the @flags can be:
1182  *  - %DRM_XE_SYNC_FLAG_SIGNAL
1183  *
1184  * A minimal use of @drm_xe_sync looks like this:
1185  *
1186  * .. code-block:: C
1187  *
1188  *     struct drm_xe_sync sync = {
1189  *         .flags = DRM_XE_SYNC_FLAG_SIGNAL,
1190  *         .type = DRM_XE_SYNC_TYPE_SYNCOBJ,
1191  *     };
1192  *     struct drm_syncobj_create syncobj_create = { 0 };
1193  *     ioctl(fd, DRM_IOCTL_SYNCOBJ_CREATE, &syncobj_create);
1194  *     sync.handle = syncobj_create.handle;
1195  *         ...
1196  *         use of &sync in drm_xe_exec or drm_xe_vm_bind
1197  *         ...
1198  *     struct drm_syncobj_wait wait = {
1199  *         .handles = &sync.handle,
1200  *         .timeout_nsec = INT64_MAX,
1201  *         .count_handles = 1,
1202  *         .flags = 0,
1203  *         .first_signaled = 0,
1204  *         .pad = 0,
1205  *     };
1206  *     ioctl(fd, DRM_IOCTL_SYNCOBJ_WAIT, &wait);
1207  */
1208 struct drm_xe_sync {
1209 	/** @extensions: Pointer to the first extension struct, if any */
1210 	__u64 extensions;
1211 
1212 #define DRM_XE_SYNC_TYPE_SYNCOBJ		0x0
1213 #define DRM_XE_SYNC_TYPE_TIMELINE_SYNCOBJ	0x1
1214 #define DRM_XE_SYNC_TYPE_USER_FENCE		0x2
1215 	/** @type: Type of the this sync object */
1216 	__u32 type;
1217 
1218 #define DRM_XE_SYNC_FLAG_SIGNAL	(1 << 0)
1219 	/** @flags: Sync Flags */
1220 	__u32 flags;
1221 
1222 	union {
1223 		/** @handle: Handle for the object */
1224 		__u32 handle;
1225 
1226 		/**
1227 		 * @addr: Address of user fence. When sync is passed in via exec
1228 		 * IOCTL this is a GPU address in the VM. When sync passed in via
1229 		 * VM bind IOCTL this is a user pointer. In either case, it is
1230 		 * the users responsibility that this address is present and
1231 		 * mapped when the user fence is signalled. Must be qword
1232 		 * aligned.
1233 		 */
1234 		__u64 addr;
1235 	};
1236 
1237 	/**
1238 	 * @timeline_value: Input for the timeline sync object. Needs to be
1239 	 * different than 0 when used with %DRM_XE_SYNC_FLAG_TIMELINE_SYNCOBJ.
1240 	 */
1241 	__u64 timeline_value;
1242 
1243 	/** @reserved: Reserved */
1244 	__u64 reserved[2];
1245 };
1246 
1247 /**
1248  * struct drm_xe_exec - Input of &DRM_IOCTL_XE_EXEC
1249  *
1250  * This is an example to use @drm_xe_exec for execution of the object
1251  * at BIND_ADDRESS (see example in @drm_xe_vm_bind) by an exec_queue
1252  * (see example in @drm_xe_exec_queue_create). It can be synchronized
1253  * by using the example provided for @drm_xe_sync.
1254  *
1255  * .. code-block:: C
1256  *
1257  *     struct drm_xe_exec exec = {
1258  *         .exec_queue_id = exec_queue,
1259  *         .syncs = &sync,
1260  *         .num_syncs = 1,
1261  *         .address = BIND_ADDRESS,
1262  *         .num_batch_buffer = 1,
1263  *     };
1264  *     ioctl(fd, DRM_IOCTL_XE_EXEC, &exec);
1265  *
1266  */
1267 struct drm_xe_exec {
1268 	/** @extensions: Pointer to the first extension struct, if any */
1269 	__u64 extensions;
1270 
1271 	/** @exec_queue_id: Exec queue ID for the batch buffer */
1272 	__u32 exec_queue_id;
1273 
1274 	/** @num_syncs: Amount of struct drm_xe_sync in array. */
1275 	__u32 num_syncs;
1276 
1277 	/** @syncs: Pointer to struct drm_xe_sync array. */
1278 	__u64 syncs;
1279 
1280 	/**
1281 	 * @address: address of batch buffer if num_batch_buffer == 1 or an
1282 	 * array of batch buffer addresses
1283 	 */
1284 	__u64 address;
1285 
1286 	/**
1287 	 * @num_batch_buffer: number of batch buffer in this exec, must match
1288 	 * the width of the engine
1289 	 */
1290 	__u16 num_batch_buffer;
1291 
1292 	/** @pad: MBZ */
1293 	__u16 pad[3];
1294 
1295 	/** @reserved: Reserved */
1296 	__u64 reserved[2];
1297 };
1298 
1299 /**
1300  * struct drm_xe_wait_user_fence - Input of &DRM_IOCTL_XE_WAIT_USER_FENCE
1301  *
1302  * Wait on user fence, XE will wake-up on every HW engine interrupt in the
1303  * instances list and check if user fence is complete::
1304  *
1305  *	(*addr & MASK) OP (VALUE & MASK)
1306  *
1307  * Returns to user on user fence completion or timeout.
1308  *
1309  * The @op can be:
1310  *  - %DRM_XE_UFENCE_WAIT_OP_EQ
1311  *  - %DRM_XE_UFENCE_WAIT_OP_NEQ
1312  *  - %DRM_XE_UFENCE_WAIT_OP_GT
1313  *  - %DRM_XE_UFENCE_WAIT_OP_GTE
1314  *  - %DRM_XE_UFENCE_WAIT_OP_LT
1315  *  - %DRM_XE_UFENCE_WAIT_OP_LTE
1316  *
1317  * and the @flags can be:
1318  *  - %DRM_XE_UFENCE_WAIT_FLAG_ABSTIME
1319  *  - %DRM_XE_UFENCE_WAIT_FLAG_SOFT_OP
1320  *
1321  * The @mask values can be for example:
1322  *  - 0xffu for u8
1323  *  - 0xffffu for u16
1324  *  - 0xffffffffu for u32
1325  *  - 0xffffffffffffffffu for u64
1326  */
1327 struct drm_xe_wait_user_fence {
1328 	/** @extensions: Pointer to the first extension struct, if any */
1329 	__u64 extensions;
1330 
1331 	/**
1332 	 * @addr: user pointer address to wait on, must qword aligned
1333 	 */
1334 	__u64 addr;
1335 
1336 #define DRM_XE_UFENCE_WAIT_OP_EQ	0x0
1337 #define DRM_XE_UFENCE_WAIT_OP_NEQ	0x1
1338 #define DRM_XE_UFENCE_WAIT_OP_GT	0x2
1339 #define DRM_XE_UFENCE_WAIT_OP_GTE	0x3
1340 #define DRM_XE_UFENCE_WAIT_OP_LT	0x4
1341 #define DRM_XE_UFENCE_WAIT_OP_LTE	0x5
1342 	/** @op: wait operation (type of comparison) */
1343 	__u16 op;
1344 
1345 #define DRM_XE_UFENCE_WAIT_FLAG_ABSTIME	(1 << 0)
1346 	/** @flags: wait flags */
1347 	__u16 flags;
1348 
1349 	/** @pad: MBZ */
1350 	__u32 pad;
1351 
1352 	/** @value: compare value */
1353 	__u64 value;
1354 
1355 	/** @mask: comparison mask */
1356 	__u64 mask;
1357 
1358 	/**
1359 	 * @timeout: how long to wait before bailing, value in nanoseconds.
1360 	 * Without DRM_XE_UFENCE_WAIT_FLAG_ABSTIME flag set (relative timeout)
1361 	 * it contains timeout expressed in nanoseconds to wait (fence will
1362 	 * expire at now() + timeout).
1363 	 * When DRM_XE_UFENCE_WAIT_FLAG_ABSTIME flat is set (absolute timeout) wait
1364 	 * will end at timeout (uses system MONOTONIC_CLOCK).
1365 	 * Passing negative timeout leads to neverending wait.
1366 	 *
1367 	 * On relative timeout this value is updated with timeout left
1368 	 * (for restarting the call in case of signal delivery).
1369 	 * On absolute timeout this value stays intact (restarted call still
1370 	 * expire at the same point of time).
1371 	 */
1372 	__s64 timeout;
1373 
1374 	/** @exec_queue_id: exec_queue_id returned from xe_exec_queue_create_ioctl */
1375 	__u32 exec_queue_id;
1376 
1377 	/** @pad2: MBZ */
1378 	__u32 pad2;
1379 
1380 	/** @reserved: Reserved */
1381 	__u64 reserved[2];
1382 };
1383 
1384 /**
1385  * enum drm_xe_observation_type - Observation stream types
1386  */
1387 enum drm_xe_observation_type {
1388 	/** @DRM_XE_OBSERVATION_TYPE_OA: OA observation stream type */
1389 	DRM_XE_OBSERVATION_TYPE_OA,
1390 };
1391 
1392 /**
1393  * enum drm_xe_observation_op - Observation stream ops
1394  */
1395 enum drm_xe_observation_op {
1396 	/** @DRM_XE_OBSERVATION_OP_STREAM_OPEN: Open an observation stream */
1397 	DRM_XE_OBSERVATION_OP_STREAM_OPEN,
1398 
1399 	/** @DRM_XE_OBSERVATION_OP_ADD_CONFIG: Add observation stream config */
1400 	DRM_XE_OBSERVATION_OP_ADD_CONFIG,
1401 
1402 	/** @DRM_XE_OBSERVATION_OP_REMOVE_CONFIG: Remove observation stream config */
1403 	DRM_XE_OBSERVATION_OP_REMOVE_CONFIG,
1404 };
1405 
1406 /**
1407  * struct drm_xe_observation_param - Input of &DRM_XE_OBSERVATION
1408  *
1409  * The observation layer enables multiplexing observation streams of
1410  * multiple types. The actual params for a particular stream operation are
1411  * supplied via the @param pointer (use __copy_from_user to get these
1412  * params).
1413  */
1414 struct drm_xe_observation_param {
1415 	/** @extensions: Pointer to the first extension struct, if any */
1416 	__u64 extensions;
1417 	/** @observation_type: observation stream type, of enum @drm_xe_observation_type */
1418 	__u64 observation_type;
1419 	/** @observation_op: observation stream op, of enum @drm_xe_observation_op */
1420 	__u64 observation_op;
1421 	/** @param: Pointer to actual stream params */
1422 	__u64 param;
1423 };
1424 
1425 /**
1426  * enum drm_xe_observation_ioctls - Observation stream fd ioctl's
1427  *
1428  * Information exchanged between userspace and kernel for observation fd
1429  * ioctl's is stream type specific
1430  */
1431 enum drm_xe_observation_ioctls {
1432 	/** @DRM_XE_OBSERVATION_IOCTL_ENABLE: Enable data capture for an observation stream */
1433 	DRM_XE_OBSERVATION_IOCTL_ENABLE = _IO('i', 0x0),
1434 
1435 	/** @DRM_XE_OBSERVATION_IOCTL_DISABLE: Disable data capture for a observation stream */
1436 	DRM_XE_OBSERVATION_IOCTL_DISABLE = _IO('i', 0x1),
1437 
1438 	/** @DRM_XE_OBSERVATION_IOCTL_CONFIG: Change observation stream configuration */
1439 	DRM_XE_OBSERVATION_IOCTL_CONFIG = _IO('i', 0x2),
1440 
1441 	/** @DRM_XE_OBSERVATION_IOCTL_STATUS: Return observation stream status */
1442 	DRM_XE_OBSERVATION_IOCTL_STATUS = _IO('i', 0x3),
1443 
1444 	/** @DRM_XE_OBSERVATION_IOCTL_INFO: Return observation stream info */
1445 	DRM_XE_OBSERVATION_IOCTL_INFO = _IO('i', 0x4),
1446 };
1447 
1448 /**
1449  * enum drm_xe_oa_unit_type - OA unit types
1450  */
1451 enum drm_xe_oa_unit_type {
1452 	/**
1453 	 * @DRM_XE_OA_UNIT_TYPE_OAG: OAG OA unit. OAR/OAC are considered
1454 	 * sub-types of OAG. For OAR/OAC, use OAG.
1455 	 */
1456 	DRM_XE_OA_UNIT_TYPE_OAG,
1457 
1458 	/** @DRM_XE_OA_UNIT_TYPE_OAM: OAM OA unit */
1459 	DRM_XE_OA_UNIT_TYPE_OAM,
1460 };
1461 
1462 /**
1463  * struct drm_xe_oa_unit - describe OA unit
1464  */
1465 struct drm_xe_oa_unit {
1466 	/** @extensions: Pointer to the first extension struct, if any */
1467 	__u64 extensions;
1468 
1469 	/** @oa_unit_id: OA unit ID */
1470 	__u32 oa_unit_id;
1471 
1472 	/** @oa_unit_type: OA unit type of @drm_xe_oa_unit_type */
1473 	__u32 oa_unit_type;
1474 
1475 	/** @capabilities: OA capabilities bit-mask */
1476 	__u64 capabilities;
1477 #define DRM_XE_OA_CAPS_BASE		(1 << 0)
1478 
1479 	/** @oa_timestamp_freq: OA timestamp freq */
1480 	__u64 oa_timestamp_freq;
1481 
1482 	/** @reserved: MBZ */
1483 	__u64 reserved[4];
1484 
1485 	/** @num_engines: number of engines in @eci array */
1486 	__u64 num_engines;
1487 
1488 	/** @eci: engines attached to this OA unit */
1489 	struct drm_xe_engine_class_instance eci[];
1490 };
1491 
1492 /**
1493  * struct drm_xe_query_oa_units - describe OA units
1494  *
1495  * If a query is made with a struct drm_xe_device_query where .query
1496  * is equal to DRM_XE_DEVICE_QUERY_OA_UNITS, then the reply uses struct
1497  * drm_xe_query_oa_units in .data.
1498  *
1499  * OA unit properties for all OA units can be accessed using a code block
1500  * such as the one below:
1501  *
1502  * .. code-block:: C
1503  *
1504  *	struct drm_xe_query_oa_units *qoa;
1505  *	struct drm_xe_oa_unit *oau;
1506  *	u8 *poau;
1507  *
1508  *	// malloc qoa and issue DRM_XE_DEVICE_QUERY_OA_UNITS. Then:
1509  *	poau = (u8 *)&qoa->oa_units[0];
1510  *	for (int i = 0; i < qoa->num_oa_units; i++) {
1511  *		oau = (struct drm_xe_oa_unit *)poau;
1512  *		// Access 'struct drm_xe_oa_unit' fields here
1513  *		poau += sizeof(*oau) + oau->num_engines * sizeof(oau->eci[0]);
1514  *	}
1515  */
1516 struct drm_xe_query_oa_units {
1517 	/** @extensions: Pointer to the first extension struct, if any */
1518 	__u64 extensions;
1519 	/** @num_oa_units: number of OA units returned in oau[] */
1520 	__u32 num_oa_units;
1521 	/** @pad: MBZ */
1522 	__u32 pad;
1523 	/**
1524 	 * @oa_units: struct @drm_xe_oa_unit array returned for this device.
1525 	 * Written below as a u64 array to avoid problems with nested flexible
1526 	 * arrays with some compilers
1527 	 */
1528 	__u64 oa_units[];
1529 };
1530 
1531 /**
1532  * enum drm_xe_oa_format_type - OA format types as specified in PRM/Bspec
1533  * 52198/60942
1534  */
1535 enum drm_xe_oa_format_type {
1536 	/** @DRM_XE_OA_FMT_TYPE_OAG: OAG report format */
1537 	DRM_XE_OA_FMT_TYPE_OAG,
1538 	/** @DRM_XE_OA_FMT_TYPE_OAR: OAR report format */
1539 	DRM_XE_OA_FMT_TYPE_OAR,
1540 	/** @DRM_XE_OA_FMT_TYPE_OAM: OAM report format */
1541 	DRM_XE_OA_FMT_TYPE_OAM,
1542 	/** @DRM_XE_OA_FMT_TYPE_OAC: OAC report format */
1543 	DRM_XE_OA_FMT_TYPE_OAC,
1544 	/** @DRM_XE_OA_FMT_TYPE_OAM_MPEC: OAM SAMEDIA or OAM MPEC report format */
1545 	DRM_XE_OA_FMT_TYPE_OAM_MPEC,
1546 	/** @DRM_XE_OA_FMT_TYPE_PEC: PEC report format */
1547 	DRM_XE_OA_FMT_TYPE_PEC,
1548 };
1549 
1550 /**
1551  * enum drm_xe_oa_property_id - OA stream property id's
1552  *
1553  * Stream params are specified as a chain of @drm_xe_ext_set_property
1554  * struct's, with @property values from enum @drm_xe_oa_property_id and
1555  * @drm_xe_user_extension base.name set to @DRM_XE_OA_EXTENSION_SET_PROPERTY.
1556  * @param field in struct @drm_xe_observation_param points to the first
1557  * @drm_xe_ext_set_property struct.
1558  *
1559  * Exactly the same mechanism is also used for stream reconfiguration using the
1560  * @DRM_XE_OBSERVATION_IOCTL_CONFIG observation stream fd ioctl, though only a
1561  * subset of properties below can be specified for stream reconfiguration.
1562  */
1563 enum drm_xe_oa_property_id {
1564 #define DRM_XE_OA_EXTENSION_SET_PROPERTY	0
1565 	/**
1566 	 * @DRM_XE_OA_PROPERTY_OA_UNIT_ID: ID of the OA unit on which to open
1567 	 * the OA stream, see @oa_unit_id in 'struct
1568 	 * drm_xe_query_oa_units'. Defaults to 0 if not provided.
1569 	 */
1570 	DRM_XE_OA_PROPERTY_OA_UNIT_ID = 1,
1571 
1572 	/**
1573 	 * @DRM_XE_OA_PROPERTY_SAMPLE_OA: A value of 1 requests inclusion of raw
1574 	 * OA unit reports or stream samples in a global buffer attached to an
1575 	 * OA unit.
1576 	 */
1577 	DRM_XE_OA_PROPERTY_SAMPLE_OA,
1578 
1579 	/**
1580 	 * @DRM_XE_OA_PROPERTY_OA_METRIC_SET: OA metrics defining contents of OA
1581 	 * reports, previously added via @DRM_XE_OBSERVATION_OP_ADD_CONFIG.
1582 	 */
1583 	DRM_XE_OA_PROPERTY_OA_METRIC_SET,
1584 
1585 	/** @DRM_XE_OA_PROPERTY_OA_FORMAT: OA counter report format */
1586 	DRM_XE_OA_PROPERTY_OA_FORMAT,
1587 	/*
1588 	 * OA_FORMAT's are specified the same way as in PRM/Bspec 52198/60942,
1589 	 * in terms of the following quantities: a. enum @drm_xe_oa_format_type
1590 	 * b. Counter select c. Counter size and d. BC report. Also refer to the
1591 	 * oa_formats array in drivers/gpu/drm/xe/xe_oa.c.
1592 	 */
1593 #define DRM_XE_OA_FORMAT_MASK_FMT_TYPE		(0xffu << 0)
1594 #define DRM_XE_OA_FORMAT_MASK_COUNTER_SEL	(0xffu << 8)
1595 #define DRM_XE_OA_FORMAT_MASK_COUNTER_SIZE	(0xffu << 16)
1596 #define DRM_XE_OA_FORMAT_MASK_BC_REPORT		(0xffu << 24)
1597 
1598 	/**
1599 	 * @DRM_XE_OA_PROPERTY_OA_PERIOD_EXPONENT: Requests periodic OA unit
1600 	 * sampling with sampling frequency proportional to 2^(period_exponent + 1)
1601 	 */
1602 	DRM_XE_OA_PROPERTY_OA_PERIOD_EXPONENT,
1603 
1604 	/**
1605 	 * @DRM_XE_OA_PROPERTY_OA_DISABLED: A value of 1 will open the OA
1606 	 * stream in a DISABLED state (see @DRM_XE_OBSERVATION_IOCTL_ENABLE).
1607 	 */
1608 	DRM_XE_OA_PROPERTY_OA_DISABLED,
1609 
1610 	/**
1611 	 * @DRM_XE_OA_PROPERTY_EXEC_QUEUE_ID: Open the stream for a specific
1612 	 * @exec_queue_id. OA queries can be executed on this exec queue.
1613 	 */
1614 	DRM_XE_OA_PROPERTY_EXEC_QUEUE_ID,
1615 
1616 	/**
1617 	 * @DRM_XE_OA_PROPERTY_OA_ENGINE_INSTANCE: Optional engine instance to
1618 	 * pass along with @DRM_XE_OA_PROPERTY_EXEC_QUEUE_ID or will default to 0.
1619 	 */
1620 	DRM_XE_OA_PROPERTY_OA_ENGINE_INSTANCE,
1621 
1622 	/**
1623 	 * @DRM_XE_OA_PROPERTY_NO_PREEMPT: Allow preemption and timeslicing
1624 	 * to be disabled for the stream exec queue.
1625 	 */
1626 	DRM_XE_OA_PROPERTY_NO_PREEMPT,
1627 };
1628 
1629 /**
1630  * struct drm_xe_oa_config - OA metric configuration
1631  *
1632  * Multiple OA configs can be added using @DRM_XE_OBSERVATION_OP_ADD_CONFIG. A
1633  * particular config can be specified when opening an OA stream using
1634  * @DRM_XE_OA_PROPERTY_OA_METRIC_SET property.
1635  */
1636 struct drm_xe_oa_config {
1637 	/** @extensions: Pointer to the first extension struct, if any */
1638 	__u64 extensions;
1639 
1640 	/** @uuid: String formatted like "%\08x-%\04x-%\04x-%\04x-%\012x" */
1641 	char uuid[36];
1642 
1643 	/** @n_regs: Number of regs in @regs_ptr */
1644 	__u32 n_regs;
1645 
1646 	/**
1647 	 * @regs_ptr: Pointer to (register address, value) pairs for OA config
1648 	 * registers. Expected length of buffer is: (2 * sizeof(u32) * @n_regs).
1649 	 */
1650 	__u64 regs_ptr;
1651 };
1652 
1653 /**
1654  * struct drm_xe_oa_stream_status - OA stream status returned from
1655  * @DRM_XE_OBSERVATION_IOCTL_STATUS observation stream fd ioctl. Userspace can
1656  * call the ioctl to query stream status in response to EIO errno from
1657  * observation fd read().
1658  */
1659 struct drm_xe_oa_stream_status {
1660 	/** @extensions: Pointer to the first extension struct, if any */
1661 	__u64 extensions;
1662 
1663 	/** @oa_status: OA stream status (see Bspec 46717/61226) */
1664 	__u64 oa_status;
1665 #define DRM_XE_OASTATUS_MMIO_TRG_Q_FULL		(1 << 3)
1666 #define DRM_XE_OASTATUS_COUNTER_OVERFLOW	(1 << 2)
1667 #define DRM_XE_OASTATUS_BUFFER_OVERFLOW		(1 << 1)
1668 #define DRM_XE_OASTATUS_REPORT_LOST		(1 << 0)
1669 
1670 	/** @reserved: reserved for future use */
1671 	__u64 reserved[3];
1672 };
1673 
1674 /**
1675  * struct drm_xe_oa_stream_info - OA stream info returned from
1676  * @DRM_XE_OBSERVATION_IOCTL_INFO observation stream fd ioctl
1677  */
1678 struct drm_xe_oa_stream_info {
1679 	/** @extensions: Pointer to the first extension struct, if any */
1680 	__u64 extensions;
1681 
1682 	/** @oa_buf_size: OA buffer size */
1683 	__u64 oa_buf_size;
1684 
1685 	/** @reserved: reserved for future use */
1686 	__u64 reserved[3];
1687 };
1688 
1689 #if defined(__cplusplus)
1690 }
1691 #endif
1692 
1693 #endif /* _UAPI_XE_DRM_H_ */
1694