1 /* 2 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sub license, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * The above copyright notice and this permission notice (including the 14 * next paragraph) shall be included in all copies or substantial portions 15 * of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 20 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 22 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 23 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 24 * 25 */ 26 27 #ifndef _UAPI_I915_DRM_H_ 28 #define _UAPI_I915_DRM_H_ 29 30 #include "drm.h" 31 32 #if defined(__cplusplus) 33 extern "C" { 34 #endif 35 36 /* Please note that modifications to all structs defined here are 37 * subject to backwards-compatibility constraints. 38 */ 39 40 /** 41 * DOC: uevents generated by i915 on it's device node 42 * 43 * I915_L3_PARITY_UEVENT - Generated when the driver receives a parity mismatch 44 * event from the gpu l3 cache. Additional information supplied is ROW, 45 * BANK, SUBBANK, SLICE of the affected cacheline. Userspace should keep 46 * track of these events and if a specific cache-line seems to have a 47 * persistent error remap it with the l3 remapping tool supplied in 48 * intel-gpu-tools. The value supplied with the event is always 1. 49 * 50 * I915_ERROR_UEVENT - Generated upon error detection, currently only via 51 * hangcheck. The error detection event is a good indicator of when things 52 * began to go badly. The value supplied with the event is a 1 upon error 53 * detection, and a 0 upon reset completion, signifying no more error 54 * exists. NOTE: Disabling hangcheck or reset via module parameter will 55 * cause the related events to not be seen. 56 * 57 * I915_RESET_UEVENT - Event is generated just before an attempt to reset the 58 * GPU. The value supplied with the event is always 1. NOTE: Disable 59 * reset via module parameter will cause this event to not be seen. 60 */ 61 #define I915_L3_PARITY_UEVENT "L3_PARITY_ERROR" 62 #define I915_ERROR_UEVENT "ERROR" 63 #define I915_RESET_UEVENT "RESET" 64 65 /** 66 * struct i915_user_extension - Base class for defining a chain of extensions 67 * 68 * Many interfaces need to grow over time. In most cases we can simply 69 * extend the struct and have userspace pass in more data. Another option, 70 * as demonstrated by Vulkan's approach to providing extensions for forward 71 * and backward compatibility, is to use a list of optional structs to 72 * provide those extra details. 73 * 74 * The key advantage to using an extension chain is that it allows us to 75 * redefine the interface more easily than an ever growing struct of 76 * increasing complexity, and for large parts of that interface to be 77 * entirely optional. The downside is more pointer chasing; chasing across 78 * the __user boundary with pointers encapsulated inside u64. 79 * 80 * Example chaining: 81 * 82 * .. code-block:: C 83 * 84 * struct i915_user_extension ext3 { 85 * .next_extension = 0, // end 86 * .name = ..., 87 * }; 88 * struct i915_user_extension ext2 { 89 * .next_extension = (uintptr_t)&ext3, 90 * .name = ..., 91 * }; 92 * struct i915_user_extension ext1 { 93 * .next_extension = (uintptr_t)&ext2, 94 * .name = ..., 95 * }; 96 * 97 * Typically the struct i915_user_extension would be embedded in some uAPI 98 * struct, and in this case we would feed it the head of the chain(i.e ext1), 99 * which would then apply all of the above extensions. 100 * 101 */ 102 struct i915_user_extension { 103 /** 104 * @next_extension: 105 * 106 * Pointer to the next struct i915_user_extension, or zero if the end. 107 */ 108 __u64 next_extension; 109 /** 110 * @name: Name of the extension. 111 * 112 * Note that the name here is just some integer. 113 * 114 * Also note that the name space for this is not global for the whole 115 * driver, but rather its scope/meaning is limited to the specific piece 116 * of uAPI which has embedded the struct i915_user_extension. 117 */ 118 __u32 name; 119 /** 120 * @flags: MBZ 121 * 122 * All undefined bits must be zero. 123 */ 124 __u32 flags; 125 /** 126 * @rsvd: MBZ 127 * 128 * Reserved for future use; must be zero. 129 */ 130 __u32 rsvd[4]; 131 }; 132 133 /* 134 * MOCS indexes used for GPU surfaces, defining the cacheability of the 135 * surface data and the coherency for this data wrt. CPU vs. GPU accesses. 136 */ 137 enum i915_mocs_table_index { 138 /* 139 * Not cached anywhere, coherency between CPU and GPU accesses is 140 * guaranteed. 141 */ 142 I915_MOCS_UNCACHED, 143 /* 144 * Cacheability and coherency controlled by the kernel automatically 145 * based on the DRM_I915_GEM_SET_CACHING IOCTL setting and the current 146 * usage of the surface (used for display scanout or not). 147 */ 148 I915_MOCS_PTE, 149 /* 150 * Cached in all GPU caches available on the platform. 151 * Coherency between CPU and GPU accesses to the surface is not 152 * guaranteed without extra synchronization. 153 */ 154 I915_MOCS_CACHED, 155 }; 156 157 /* 158 * Different engines serve different roles, and there may be more than one 159 * engine serving each role. enum drm_i915_gem_engine_class provides a 160 * classification of the role of the engine, which may be used when requesting 161 * operations to be performed on a certain subset of engines, or for providing 162 * information about that group. 163 */ 164 enum drm_i915_gem_engine_class { 165 I915_ENGINE_CLASS_RENDER = 0, 166 I915_ENGINE_CLASS_COPY = 1, 167 I915_ENGINE_CLASS_VIDEO = 2, 168 I915_ENGINE_CLASS_VIDEO_ENHANCE = 3, 169 170 /* should be kept compact */ 171 172 I915_ENGINE_CLASS_INVALID = -1 173 }; 174 175 /* 176 * There may be more than one engine fulfilling any role within the system. 177 * Each engine of a class is given a unique instance number and therefore 178 * any engine can be specified by its class:instance tuplet. APIs that allow 179 * access to any engine in the system will use struct i915_engine_class_instance 180 * for this identification. 181 */ 182 struct i915_engine_class_instance { 183 __u16 engine_class; /* see enum drm_i915_gem_engine_class */ 184 __u16 engine_instance; 185 #define I915_ENGINE_CLASS_INVALID_NONE -1 186 #define I915_ENGINE_CLASS_INVALID_VIRTUAL -2 187 }; 188 189 /** 190 * DOC: perf_events exposed by i915 through /sys/bus/event_sources/drivers/i915 191 * 192 */ 193 194 enum drm_i915_pmu_engine_sample { 195 I915_SAMPLE_BUSY = 0, 196 I915_SAMPLE_WAIT = 1, 197 I915_SAMPLE_SEMA = 2 198 }; 199 200 #define I915_PMU_SAMPLE_BITS (4) 201 #define I915_PMU_SAMPLE_MASK (0xf) 202 #define I915_PMU_SAMPLE_INSTANCE_BITS (8) 203 #define I915_PMU_CLASS_SHIFT \ 204 (I915_PMU_SAMPLE_BITS + I915_PMU_SAMPLE_INSTANCE_BITS) 205 206 #define __I915_PMU_ENGINE(class, instance, sample) \ 207 ((class) << I915_PMU_CLASS_SHIFT | \ 208 (instance) << I915_PMU_SAMPLE_BITS | \ 209 (sample)) 210 211 #define I915_PMU_ENGINE_BUSY(class, instance) \ 212 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_BUSY) 213 214 #define I915_PMU_ENGINE_WAIT(class, instance) \ 215 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_WAIT) 216 217 #define I915_PMU_ENGINE_SEMA(class, instance) \ 218 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_SEMA) 219 220 #define __I915_PMU_OTHER(x) (__I915_PMU_ENGINE(0xff, 0xff, 0xf) + 1 + (x)) 221 222 #define I915_PMU_ACTUAL_FREQUENCY __I915_PMU_OTHER(0) 223 #define I915_PMU_REQUESTED_FREQUENCY __I915_PMU_OTHER(1) 224 #define I915_PMU_INTERRUPTS __I915_PMU_OTHER(2) 225 #define I915_PMU_RC6_RESIDENCY __I915_PMU_OTHER(3) 226 #define I915_PMU_SOFTWARE_GT_AWAKE_TIME __I915_PMU_OTHER(4) 227 228 #define I915_PMU_LAST /* Deprecated - do not use */ I915_PMU_RC6_RESIDENCY 229 230 /* Each region is a minimum of 16k, and there are at most 255 of them. 231 */ 232 #define I915_NR_TEX_REGIONS 255 /* table size 2k - maximum due to use 233 * of chars for next/prev indices */ 234 #define I915_LOG_MIN_TEX_REGION_SIZE 14 235 236 typedef struct _drm_i915_init { 237 enum { 238 I915_INIT_DMA = 0x01, 239 I915_CLEANUP_DMA = 0x02, 240 I915_RESUME_DMA = 0x03 241 } func; 242 unsigned int mmio_offset; 243 int sarea_priv_offset; 244 unsigned int ring_start; 245 unsigned int ring_end; 246 unsigned int ring_size; 247 unsigned int front_offset; 248 unsigned int back_offset; 249 unsigned int depth_offset; 250 unsigned int w; 251 unsigned int h; 252 unsigned int pitch; 253 unsigned int pitch_bits; 254 unsigned int back_pitch; 255 unsigned int depth_pitch; 256 unsigned int cpp; 257 unsigned int chipset; 258 } drm_i915_init_t; 259 260 typedef struct _drm_i915_sarea { 261 struct drm_tex_region texList[I915_NR_TEX_REGIONS + 1]; 262 int last_upload; /* last time texture was uploaded */ 263 int last_enqueue; /* last time a buffer was enqueued */ 264 int last_dispatch; /* age of the most recently dispatched buffer */ 265 int ctxOwner; /* last context to upload state */ 266 int texAge; 267 int pf_enabled; /* is pageflipping allowed? */ 268 int pf_active; 269 int pf_current_page; /* which buffer is being displayed? */ 270 int perf_boxes; /* performance boxes to be displayed */ 271 int width, height; /* screen size in pixels */ 272 273 drm_handle_t front_handle; 274 int front_offset; 275 int front_size; 276 277 drm_handle_t back_handle; 278 int back_offset; 279 int back_size; 280 281 drm_handle_t depth_handle; 282 int depth_offset; 283 int depth_size; 284 285 drm_handle_t tex_handle; 286 int tex_offset; 287 int tex_size; 288 int log_tex_granularity; 289 int pitch; 290 int rotation; /* 0, 90, 180 or 270 */ 291 int rotated_offset; 292 int rotated_size; 293 int rotated_pitch; 294 int virtualX, virtualY; 295 296 unsigned int front_tiled; 297 unsigned int back_tiled; 298 unsigned int depth_tiled; 299 unsigned int rotated_tiled; 300 unsigned int rotated2_tiled; 301 302 int pipeA_x; 303 int pipeA_y; 304 int pipeA_w; 305 int pipeA_h; 306 int pipeB_x; 307 int pipeB_y; 308 int pipeB_w; 309 int pipeB_h; 310 311 /* fill out some space for old userspace triple buffer */ 312 drm_handle_t unused_handle; 313 __u32 unused1, unused2, unused3; 314 315 /* buffer object handles for static buffers. May change 316 * over the lifetime of the client. 317 */ 318 __u32 front_bo_handle; 319 __u32 back_bo_handle; 320 __u32 unused_bo_handle; 321 __u32 depth_bo_handle; 322 323 } drm_i915_sarea_t; 324 325 /* due to userspace building against these headers we need some compat here */ 326 #define planeA_x pipeA_x 327 #define planeA_y pipeA_y 328 #define planeA_w pipeA_w 329 #define planeA_h pipeA_h 330 #define planeB_x pipeB_x 331 #define planeB_y pipeB_y 332 #define planeB_w pipeB_w 333 #define planeB_h pipeB_h 334 335 /* Flags for perf_boxes 336 */ 337 #define I915_BOX_RING_EMPTY 0x1 338 #define I915_BOX_FLIP 0x2 339 #define I915_BOX_WAIT 0x4 340 #define I915_BOX_TEXTURE_LOAD 0x8 341 #define I915_BOX_LOST_CONTEXT 0x10 342 343 /* 344 * i915 specific ioctls. 345 * 346 * The device specific ioctl range is [DRM_COMMAND_BASE, DRM_COMMAND_END) ie 347 * [0x40, 0xa0) (a0 is excluded). The numbers below are defined as offset 348 * against DRM_COMMAND_BASE and should be between [0x0, 0x60). 349 */ 350 #define DRM_I915_INIT 0x00 351 #define DRM_I915_FLUSH 0x01 352 #define DRM_I915_FLIP 0x02 353 #define DRM_I915_BATCHBUFFER 0x03 354 #define DRM_I915_IRQ_EMIT 0x04 355 #define DRM_I915_IRQ_WAIT 0x05 356 #define DRM_I915_GETPARAM 0x06 357 #define DRM_I915_SETPARAM 0x07 358 #define DRM_I915_ALLOC 0x08 359 #define DRM_I915_FREE 0x09 360 #define DRM_I915_INIT_HEAP 0x0a 361 #define DRM_I915_CMDBUFFER 0x0b 362 #define DRM_I915_DESTROY_HEAP 0x0c 363 #define DRM_I915_SET_VBLANK_PIPE 0x0d 364 #define DRM_I915_GET_VBLANK_PIPE 0x0e 365 #define DRM_I915_VBLANK_SWAP 0x0f 366 #define DRM_I915_HWS_ADDR 0x11 367 #define DRM_I915_GEM_INIT 0x13 368 #define DRM_I915_GEM_EXECBUFFER 0x14 369 #define DRM_I915_GEM_PIN 0x15 370 #define DRM_I915_GEM_UNPIN 0x16 371 #define DRM_I915_GEM_BUSY 0x17 372 #define DRM_I915_GEM_THROTTLE 0x18 373 #define DRM_I915_GEM_ENTERVT 0x19 374 #define DRM_I915_GEM_LEAVEVT 0x1a 375 #define DRM_I915_GEM_CREATE 0x1b 376 #define DRM_I915_GEM_PREAD 0x1c 377 #define DRM_I915_GEM_PWRITE 0x1d 378 #define DRM_I915_GEM_MMAP 0x1e 379 #define DRM_I915_GEM_SET_DOMAIN 0x1f 380 #define DRM_I915_GEM_SW_FINISH 0x20 381 #define DRM_I915_GEM_SET_TILING 0x21 382 #define DRM_I915_GEM_GET_TILING 0x22 383 #define DRM_I915_GEM_GET_APERTURE 0x23 384 #define DRM_I915_GEM_MMAP_GTT 0x24 385 #define DRM_I915_GET_PIPE_FROM_CRTC_ID 0x25 386 #define DRM_I915_GEM_MADVISE 0x26 387 #define DRM_I915_OVERLAY_PUT_IMAGE 0x27 388 #define DRM_I915_OVERLAY_ATTRS 0x28 389 #define DRM_I915_GEM_EXECBUFFER2 0x29 390 #define DRM_I915_GEM_EXECBUFFER2_WR DRM_I915_GEM_EXECBUFFER2 391 #define DRM_I915_GET_SPRITE_COLORKEY 0x2a 392 #define DRM_I915_SET_SPRITE_COLORKEY 0x2b 393 #define DRM_I915_GEM_WAIT 0x2c 394 #define DRM_I915_GEM_CONTEXT_CREATE 0x2d 395 #define DRM_I915_GEM_CONTEXT_DESTROY 0x2e 396 #define DRM_I915_GEM_SET_CACHING 0x2f 397 #define DRM_I915_GEM_GET_CACHING 0x30 398 #define DRM_I915_REG_READ 0x31 399 #define DRM_I915_GET_RESET_STATS 0x32 400 #define DRM_I915_GEM_USERPTR 0x33 401 #define DRM_I915_GEM_CONTEXT_GETPARAM 0x34 402 #define DRM_I915_GEM_CONTEXT_SETPARAM 0x35 403 #define DRM_I915_PERF_OPEN 0x36 404 #define DRM_I915_PERF_ADD_CONFIG 0x37 405 #define DRM_I915_PERF_REMOVE_CONFIG 0x38 406 #define DRM_I915_QUERY 0x39 407 #define DRM_I915_GEM_VM_CREATE 0x3a 408 #define DRM_I915_GEM_VM_DESTROY 0x3b 409 #define DRM_I915_GEM_CREATE_EXT 0x3c 410 /* Must be kept compact -- no holes */ 411 412 #define DRM_IOCTL_I915_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t) 413 #define DRM_IOCTL_I915_FLUSH DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH) 414 #define DRM_IOCTL_I915_FLIP DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLIP) 415 #define DRM_IOCTL_I915_BATCHBUFFER DRM_IOW( DRM_COMMAND_BASE + DRM_I915_BATCHBUFFER, drm_i915_batchbuffer_t) 416 #define DRM_IOCTL_I915_IRQ_EMIT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_IRQ_EMIT, drm_i915_irq_emit_t) 417 #define DRM_IOCTL_I915_IRQ_WAIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_IRQ_WAIT, drm_i915_irq_wait_t) 418 #define DRM_IOCTL_I915_GETPARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GETPARAM, drm_i915_getparam_t) 419 #define DRM_IOCTL_I915_SETPARAM DRM_IOW( DRM_COMMAND_BASE + DRM_I915_SETPARAM, drm_i915_setparam_t) 420 #define DRM_IOCTL_I915_ALLOC DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_ALLOC, drm_i915_mem_alloc_t) 421 #define DRM_IOCTL_I915_FREE DRM_IOW( DRM_COMMAND_BASE + DRM_I915_FREE, drm_i915_mem_free_t) 422 #define DRM_IOCTL_I915_INIT_HEAP DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT_HEAP, drm_i915_mem_init_heap_t) 423 #define DRM_IOCTL_I915_CMDBUFFER DRM_IOW( DRM_COMMAND_BASE + DRM_I915_CMDBUFFER, drm_i915_cmdbuffer_t) 424 #define DRM_IOCTL_I915_DESTROY_HEAP DRM_IOW( DRM_COMMAND_BASE + DRM_I915_DESTROY_HEAP, drm_i915_mem_destroy_heap_t) 425 #define DRM_IOCTL_I915_SET_VBLANK_PIPE DRM_IOW( DRM_COMMAND_BASE + DRM_I915_SET_VBLANK_PIPE, drm_i915_vblank_pipe_t) 426 #define DRM_IOCTL_I915_GET_VBLANK_PIPE DRM_IOR( DRM_COMMAND_BASE + DRM_I915_GET_VBLANK_PIPE, drm_i915_vblank_pipe_t) 427 #define DRM_IOCTL_I915_VBLANK_SWAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_VBLANK_SWAP, drm_i915_vblank_swap_t) 428 #define DRM_IOCTL_I915_HWS_ADDR DRM_IOW(DRM_COMMAND_BASE + DRM_I915_HWS_ADDR, struct drm_i915_gem_init) 429 #define DRM_IOCTL_I915_GEM_INIT DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_INIT, struct drm_i915_gem_init) 430 #define DRM_IOCTL_I915_GEM_EXECBUFFER DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER, struct drm_i915_gem_execbuffer) 431 #define DRM_IOCTL_I915_GEM_EXECBUFFER2 DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER2, struct drm_i915_gem_execbuffer2) 432 #define DRM_IOCTL_I915_GEM_EXECBUFFER2_WR DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER2_WR, struct drm_i915_gem_execbuffer2) 433 #define DRM_IOCTL_I915_GEM_PIN DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_PIN, struct drm_i915_gem_pin) 434 #define DRM_IOCTL_I915_GEM_UNPIN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_UNPIN, struct drm_i915_gem_unpin) 435 #define DRM_IOCTL_I915_GEM_BUSY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_BUSY, struct drm_i915_gem_busy) 436 #define DRM_IOCTL_I915_GEM_SET_CACHING DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_SET_CACHING, struct drm_i915_gem_caching) 437 #define DRM_IOCTL_I915_GEM_GET_CACHING DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_GET_CACHING, struct drm_i915_gem_caching) 438 #define DRM_IOCTL_I915_GEM_THROTTLE DRM_IO ( DRM_COMMAND_BASE + DRM_I915_GEM_THROTTLE) 439 #define DRM_IOCTL_I915_GEM_ENTERVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_ENTERVT) 440 #define DRM_IOCTL_I915_GEM_LEAVEVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_LEAVEVT) 441 #define DRM_IOCTL_I915_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE, struct drm_i915_gem_create) 442 #define DRM_IOCTL_I915_GEM_CREATE_EXT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE_EXT, struct drm_i915_gem_create_ext) 443 #define DRM_IOCTL_I915_GEM_PREAD DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PREAD, struct drm_i915_gem_pread) 444 #define DRM_IOCTL_I915_GEM_PWRITE DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PWRITE, struct drm_i915_gem_pwrite) 445 #define DRM_IOCTL_I915_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP, struct drm_i915_gem_mmap) 446 #define DRM_IOCTL_I915_GEM_MMAP_GTT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_gtt) 447 #define DRM_IOCTL_I915_GEM_MMAP_OFFSET DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_offset) 448 #define DRM_IOCTL_I915_GEM_SET_DOMAIN DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SET_DOMAIN, struct drm_i915_gem_set_domain) 449 #define DRM_IOCTL_I915_GEM_SW_FINISH DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SW_FINISH, struct drm_i915_gem_sw_finish) 450 #define DRM_IOCTL_I915_GEM_SET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_SET_TILING, struct drm_i915_gem_set_tiling) 451 #define DRM_IOCTL_I915_GEM_GET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_TILING, struct drm_i915_gem_get_tiling) 452 #define DRM_IOCTL_I915_GEM_GET_APERTURE DRM_IOR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_APERTURE, struct drm_i915_gem_get_aperture) 453 #define DRM_IOCTL_I915_GET_PIPE_FROM_CRTC_ID DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GET_PIPE_FROM_CRTC_ID, struct drm_i915_get_pipe_from_crtc_id) 454 #define DRM_IOCTL_I915_GEM_MADVISE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MADVISE, struct drm_i915_gem_madvise) 455 #define DRM_IOCTL_I915_OVERLAY_PUT_IMAGE DRM_IOW(DRM_COMMAND_BASE + DRM_I915_OVERLAY_PUT_IMAGE, struct drm_intel_overlay_put_image) 456 #define DRM_IOCTL_I915_OVERLAY_ATTRS DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_OVERLAY_ATTRS, struct drm_intel_overlay_attrs) 457 #define DRM_IOCTL_I915_SET_SPRITE_COLORKEY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_SET_SPRITE_COLORKEY, struct drm_intel_sprite_colorkey) 458 #define DRM_IOCTL_I915_GET_SPRITE_COLORKEY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GET_SPRITE_COLORKEY, struct drm_intel_sprite_colorkey) 459 #define DRM_IOCTL_I915_GEM_WAIT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_WAIT, struct drm_i915_gem_wait) 460 #define DRM_IOCTL_I915_GEM_CONTEXT_CREATE DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_CREATE, struct drm_i915_gem_context_create) 461 #define DRM_IOCTL_I915_GEM_CONTEXT_CREATE_EXT DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_CREATE, struct drm_i915_gem_context_create_ext) 462 #define DRM_IOCTL_I915_GEM_CONTEXT_DESTROY DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_DESTROY, struct drm_i915_gem_context_destroy) 463 #define DRM_IOCTL_I915_REG_READ DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_REG_READ, struct drm_i915_reg_read) 464 #define DRM_IOCTL_I915_GET_RESET_STATS DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GET_RESET_STATS, struct drm_i915_reset_stats) 465 #define DRM_IOCTL_I915_GEM_USERPTR DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_USERPTR, struct drm_i915_gem_userptr) 466 #define DRM_IOCTL_I915_GEM_CONTEXT_GETPARAM DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_GETPARAM, struct drm_i915_gem_context_param) 467 #define DRM_IOCTL_I915_GEM_CONTEXT_SETPARAM DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_SETPARAM, struct drm_i915_gem_context_param) 468 #define DRM_IOCTL_I915_PERF_OPEN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_OPEN, struct drm_i915_perf_open_param) 469 #define DRM_IOCTL_I915_PERF_ADD_CONFIG DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_ADD_CONFIG, struct drm_i915_perf_oa_config) 470 #define DRM_IOCTL_I915_PERF_REMOVE_CONFIG DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_REMOVE_CONFIG, __u64) 471 #define DRM_IOCTL_I915_QUERY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_QUERY, struct drm_i915_query) 472 #define DRM_IOCTL_I915_GEM_VM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_VM_CREATE, struct drm_i915_gem_vm_control) 473 #define DRM_IOCTL_I915_GEM_VM_DESTROY DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_VM_DESTROY, struct drm_i915_gem_vm_control) 474 475 /* Allow drivers to submit batchbuffers directly to hardware, relying 476 * on the security mechanisms provided by hardware. 477 */ 478 typedef struct drm_i915_batchbuffer { 479 int start; /* agp offset */ 480 int used; /* nr bytes in use */ 481 int DR1; /* hw flags for GFX_OP_DRAWRECT_INFO */ 482 int DR4; /* window origin for GFX_OP_DRAWRECT_INFO */ 483 int num_cliprects; /* mulitpass with multiple cliprects? */ 484 struct drm_clip_rect __user *cliprects; /* pointer to userspace cliprects */ 485 } drm_i915_batchbuffer_t; 486 487 /* As above, but pass a pointer to userspace buffer which can be 488 * validated by the kernel prior to sending to hardware. 489 */ 490 typedef struct _drm_i915_cmdbuffer { 491 char __user *buf; /* pointer to userspace command buffer */ 492 int sz; /* nr bytes in buf */ 493 int DR1; /* hw flags for GFX_OP_DRAWRECT_INFO */ 494 int DR4; /* window origin for GFX_OP_DRAWRECT_INFO */ 495 int num_cliprects; /* mulitpass with multiple cliprects? */ 496 struct drm_clip_rect __user *cliprects; /* pointer to userspace cliprects */ 497 } drm_i915_cmdbuffer_t; 498 499 /* Userspace can request & wait on irq's: 500 */ 501 typedef struct drm_i915_irq_emit { 502 int __user *irq_seq; 503 } drm_i915_irq_emit_t; 504 505 typedef struct drm_i915_irq_wait { 506 int irq_seq; 507 } drm_i915_irq_wait_t; 508 509 /* 510 * Different modes of per-process Graphics Translation Table, 511 * see I915_PARAM_HAS_ALIASING_PPGTT 512 */ 513 #define I915_GEM_PPGTT_NONE 0 514 #define I915_GEM_PPGTT_ALIASING 1 515 #define I915_GEM_PPGTT_FULL 2 516 517 /* Ioctl to query kernel params: 518 */ 519 #define I915_PARAM_IRQ_ACTIVE 1 520 #define I915_PARAM_ALLOW_BATCHBUFFER 2 521 #define I915_PARAM_LAST_DISPATCH 3 522 #define I915_PARAM_CHIPSET_ID 4 523 #define I915_PARAM_HAS_GEM 5 524 #define I915_PARAM_NUM_FENCES_AVAIL 6 525 #define I915_PARAM_HAS_OVERLAY 7 526 #define I915_PARAM_HAS_PAGEFLIPPING 8 527 #define I915_PARAM_HAS_EXECBUF2 9 528 #define I915_PARAM_HAS_BSD 10 529 #define I915_PARAM_HAS_BLT 11 530 #define I915_PARAM_HAS_RELAXED_FENCING 12 531 #define I915_PARAM_HAS_COHERENT_RINGS 13 532 #define I915_PARAM_HAS_EXEC_CONSTANTS 14 533 #define I915_PARAM_HAS_RELAXED_DELTA 15 534 #define I915_PARAM_HAS_GEN7_SOL_RESET 16 535 #define I915_PARAM_HAS_LLC 17 536 #define I915_PARAM_HAS_ALIASING_PPGTT 18 537 #define I915_PARAM_HAS_WAIT_TIMEOUT 19 538 #define I915_PARAM_HAS_SEMAPHORES 20 539 #define I915_PARAM_HAS_PRIME_VMAP_FLUSH 21 540 #define I915_PARAM_HAS_VEBOX 22 541 #define I915_PARAM_HAS_SECURE_BATCHES 23 542 #define I915_PARAM_HAS_PINNED_BATCHES 24 543 #define I915_PARAM_HAS_EXEC_NO_RELOC 25 544 #define I915_PARAM_HAS_EXEC_HANDLE_LUT 26 545 #define I915_PARAM_HAS_WT 27 546 #define I915_PARAM_CMD_PARSER_VERSION 28 547 #define I915_PARAM_HAS_COHERENT_PHYS_GTT 29 548 #define I915_PARAM_MMAP_VERSION 30 549 #define I915_PARAM_HAS_BSD2 31 550 #define I915_PARAM_REVISION 32 551 #define I915_PARAM_SUBSLICE_TOTAL 33 552 #define I915_PARAM_EU_TOTAL 34 553 #define I915_PARAM_HAS_GPU_RESET 35 554 #define I915_PARAM_HAS_RESOURCE_STREAMER 36 555 #define I915_PARAM_HAS_EXEC_SOFTPIN 37 556 #define I915_PARAM_HAS_POOLED_EU 38 557 #define I915_PARAM_MIN_EU_IN_POOL 39 558 #define I915_PARAM_MMAP_GTT_VERSION 40 559 560 /* 561 * Query whether DRM_I915_GEM_EXECBUFFER2 supports user defined execution 562 * priorities and the driver will attempt to execute batches in priority order. 563 * The param returns a capability bitmask, nonzero implies that the scheduler 564 * is enabled, with different features present according to the mask. 565 * 566 * The initial priority for each batch is supplied by the context and is 567 * controlled via I915_CONTEXT_PARAM_PRIORITY. 568 */ 569 #define I915_PARAM_HAS_SCHEDULER 41 570 #define I915_SCHEDULER_CAP_ENABLED (1ul << 0) 571 #define I915_SCHEDULER_CAP_PRIORITY (1ul << 1) 572 #define I915_SCHEDULER_CAP_PREEMPTION (1ul << 2) 573 #define I915_SCHEDULER_CAP_SEMAPHORES (1ul << 3) 574 #define I915_SCHEDULER_CAP_ENGINE_BUSY_STATS (1ul << 4) 575 /* 576 * Indicates the 2k user priority levels are statically mapped into 3 buckets as 577 * follows: 578 * 579 * -1k to -1 Low priority 580 * 0 Normal priority 581 * 1 to 1k Highest priority 582 */ 583 #define I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP (1ul << 5) 584 585 #define I915_PARAM_HUC_STATUS 42 586 587 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to opt-out of 588 * synchronisation with implicit fencing on individual objects. 589 * See EXEC_OBJECT_ASYNC. 590 */ 591 #define I915_PARAM_HAS_EXEC_ASYNC 43 592 593 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports explicit fence support - 594 * both being able to pass in a sync_file fd to wait upon before executing, 595 * and being able to return a new sync_file fd that is signaled when the 596 * current request is complete. See I915_EXEC_FENCE_IN and I915_EXEC_FENCE_OUT. 597 */ 598 #define I915_PARAM_HAS_EXEC_FENCE 44 599 600 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to capture 601 * user specified bufffers for post-mortem debugging of GPU hangs. See 602 * EXEC_OBJECT_CAPTURE. 603 */ 604 #define I915_PARAM_HAS_EXEC_CAPTURE 45 605 606 #define I915_PARAM_SLICE_MASK 46 607 608 /* Assuming it's uniform for each slice, this queries the mask of subslices 609 * per-slice for this system. 610 */ 611 #define I915_PARAM_SUBSLICE_MASK 47 612 613 /* 614 * Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying the batch buffer 615 * as the first execobject as opposed to the last. See I915_EXEC_BATCH_FIRST. 616 */ 617 #define I915_PARAM_HAS_EXEC_BATCH_FIRST 48 618 619 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 620 * drm_i915_gem_exec_fence structures. See I915_EXEC_FENCE_ARRAY. 621 */ 622 #define I915_PARAM_HAS_EXEC_FENCE_ARRAY 49 623 624 /* 625 * Query whether every context (both per-file default and user created) is 626 * isolated (insofar as HW supports). If this parameter is not true, then 627 * freshly created contexts may inherit values from an existing context, 628 * rather than default HW values. If true, it also ensures (insofar as HW 629 * supports) that all state set by this context will not leak to any other 630 * context. 631 * 632 * As not every engine across every gen support contexts, the returned 633 * value reports the support of context isolation for individual engines by 634 * returning a bitmask of each engine class set to true if that class supports 635 * isolation. 636 */ 637 #define I915_PARAM_HAS_CONTEXT_ISOLATION 50 638 639 /* Frequency of the command streamer timestamps given by the *_TIMESTAMP 640 * registers. This used to be fixed per platform but from CNL onwards, this 641 * might vary depending on the parts. 642 */ 643 #define I915_PARAM_CS_TIMESTAMP_FREQUENCY 51 644 645 /* 646 * Once upon a time we supposed that writes through the GGTT would be 647 * immediately in physical memory (once flushed out of the CPU path). However, 648 * on a few different processors and chipsets, this is not necessarily the case 649 * as the writes appear to be buffered internally. Thus a read of the backing 650 * storage (physical memory) via a different path (with different physical tags 651 * to the indirect write via the GGTT) will see stale values from before 652 * the GGTT write. Inside the kernel, we can for the most part keep track of 653 * the different read/write domains in use (e.g. set-domain), but the assumption 654 * of coherency is baked into the ABI, hence reporting its true state in this 655 * parameter. 656 * 657 * Reports true when writes via mmap_gtt are immediately visible following an 658 * lfence to flush the WCB. 659 * 660 * Reports false when writes via mmap_gtt are indeterminately delayed in an in 661 * internal buffer and are _not_ immediately visible to third parties accessing 662 * directly via mmap_cpu/mmap_wc. Use of mmap_gtt as part of an IPC 663 * communications channel when reporting false is strongly disadvised. 664 */ 665 #define I915_PARAM_MMAP_GTT_COHERENT 52 666 667 /* 668 * Query whether DRM_I915_GEM_EXECBUFFER2 supports coordination of parallel 669 * execution through use of explicit fence support. 670 * See I915_EXEC_FENCE_OUT and I915_EXEC_FENCE_SUBMIT. 671 */ 672 #define I915_PARAM_HAS_EXEC_SUBMIT_FENCE 53 673 674 /* 675 * Revision of the i915-perf uAPI. The value returned helps determine what 676 * i915-perf features are available. See drm_i915_perf_property_id. 677 */ 678 #define I915_PARAM_PERF_REVISION 54 679 680 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 681 * timeline syncobj through drm_i915_gem_execbuffer_ext_timeline_fences. See 682 * I915_EXEC_USE_EXTENSIONS. 683 */ 684 #define I915_PARAM_HAS_EXEC_TIMELINE_FENCES 55 685 686 /* Query if the kernel supports the I915_USERPTR_PROBE flag. */ 687 #define I915_PARAM_HAS_USERPTR_PROBE 56 688 689 /* Must be kept compact -- no holes and well documented */ 690 691 typedef struct drm_i915_getparam { 692 __s32 param; 693 /* 694 * WARNING: Using pointers instead of fixed-size u64 means we need to write 695 * compat32 code. Don't repeat this mistake. 696 */ 697 int __user *value; 698 } drm_i915_getparam_t; 699 700 /* Ioctl to set kernel params: 701 */ 702 #define I915_SETPARAM_USE_MI_BATCHBUFFER_START 1 703 #define I915_SETPARAM_TEX_LRU_LOG_GRANULARITY 2 704 #define I915_SETPARAM_ALLOW_BATCHBUFFER 3 705 #define I915_SETPARAM_NUM_USED_FENCES 4 706 /* Must be kept compact -- no holes */ 707 708 typedef struct drm_i915_setparam { 709 int param; 710 int value; 711 } drm_i915_setparam_t; 712 713 /* A memory manager for regions of shared memory: 714 */ 715 #define I915_MEM_REGION_AGP 1 716 717 typedef struct drm_i915_mem_alloc { 718 int region; 719 int alignment; 720 int size; 721 int __user *region_offset; /* offset from start of fb or agp */ 722 } drm_i915_mem_alloc_t; 723 724 typedef struct drm_i915_mem_free { 725 int region; 726 int region_offset; 727 } drm_i915_mem_free_t; 728 729 typedef struct drm_i915_mem_init_heap { 730 int region; 731 int size; 732 int start; 733 } drm_i915_mem_init_heap_t; 734 735 /* Allow memory manager to be torn down and re-initialized (eg on 736 * rotate): 737 */ 738 typedef struct drm_i915_mem_destroy_heap { 739 int region; 740 } drm_i915_mem_destroy_heap_t; 741 742 /* Allow X server to configure which pipes to monitor for vblank signals 743 */ 744 #define DRM_I915_VBLANK_PIPE_A 1 745 #define DRM_I915_VBLANK_PIPE_B 2 746 747 typedef struct drm_i915_vblank_pipe { 748 int pipe; 749 } drm_i915_vblank_pipe_t; 750 751 /* Schedule buffer swap at given vertical blank: 752 */ 753 typedef struct drm_i915_vblank_swap { 754 drm_drawable_t drawable; 755 enum drm_vblank_seq_type seqtype; 756 unsigned int sequence; 757 } drm_i915_vblank_swap_t; 758 759 typedef struct drm_i915_hws_addr { 760 __u64 addr; 761 } drm_i915_hws_addr_t; 762 763 struct drm_i915_gem_init { 764 /** 765 * Beginning offset in the GTT to be managed by the DRM memory 766 * manager. 767 */ 768 __u64 gtt_start; 769 /** 770 * Ending offset in the GTT to be managed by the DRM memory 771 * manager. 772 */ 773 __u64 gtt_end; 774 }; 775 776 struct drm_i915_gem_create { 777 /** 778 * Requested size for the object. 779 * 780 * The (page-aligned) allocated size for the object will be returned. 781 */ 782 __u64 size; 783 /** 784 * Returned handle for the object. 785 * 786 * Object handles are nonzero. 787 */ 788 __u32 handle; 789 __u32 pad; 790 }; 791 792 struct drm_i915_gem_pread { 793 /** Handle for the object being read. */ 794 __u32 handle; 795 __u32 pad; 796 /** Offset into the object to read from */ 797 __u64 offset; 798 /** Length of data to read */ 799 __u64 size; 800 /** 801 * Pointer to write the data into. 802 * 803 * This is a fixed-size type for 32/64 compatibility. 804 */ 805 __u64 data_ptr; 806 }; 807 808 struct drm_i915_gem_pwrite { 809 /** Handle for the object being written to. */ 810 __u32 handle; 811 __u32 pad; 812 /** Offset into the object to write to */ 813 __u64 offset; 814 /** Length of data to write */ 815 __u64 size; 816 /** 817 * Pointer to read the data from. 818 * 819 * This is a fixed-size type for 32/64 compatibility. 820 */ 821 __u64 data_ptr; 822 }; 823 824 struct drm_i915_gem_mmap { 825 /** Handle for the object being mapped. */ 826 __u32 handle; 827 __u32 pad; 828 /** Offset in the object to map. */ 829 __u64 offset; 830 /** 831 * Length of data to map. 832 * 833 * The value will be page-aligned. 834 */ 835 __u64 size; 836 /** 837 * Returned pointer the data was mapped at. 838 * 839 * This is a fixed-size type for 32/64 compatibility. 840 */ 841 __u64 addr_ptr; 842 843 /** 844 * Flags for extended behaviour. 845 * 846 * Added in version 2. 847 */ 848 __u64 flags; 849 #define I915_MMAP_WC 0x1 850 }; 851 852 struct drm_i915_gem_mmap_gtt { 853 /** Handle for the object being mapped. */ 854 __u32 handle; 855 __u32 pad; 856 /** 857 * Fake offset to use for subsequent mmap call 858 * 859 * This is a fixed-size type for 32/64 compatibility. 860 */ 861 __u64 offset; 862 }; 863 864 /** 865 * struct drm_i915_gem_mmap_offset - Retrieve an offset so we can mmap this buffer object. 866 * 867 * This struct is passed as argument to the `DRM_IOCTL_I915_GEM_MMAP_OFFSET` ioctl, 868 * and is used to retrieve the fake offset to mmap an object specified by &handle. 869 * 870 * The legacy way of using `DRM_IOCTL_I915_GEM_MMAP` is removed on gen12+. 871 * `DRM_IOCTL_I915_GEM_MMAP_GTT` is an older supported alias to this struct, but will behave 872 * as setting the &extensions to 0, and &flags to `I915_MMAP_OFFSET_GTT`. 873 */ 874 struct drm_i915_gem_mmap_offset { 875 /** @handle: Handle for the object being mapped. */ 876 __u32 handle; 877 /** @pad: Must be zero */ 878 __u32 pad; 879 /** 880 * @offset: The fake offset to use for subsequent mmap call 881 * 882 * This is a fixed-size type for 32/64 compatibility. 883 */ 884 __u64 offset; 885 886 /** 887 * @flags: Flags for extended behaviour. 888 * 889 * It is mandatory that one of the `MMAP_OFFSET` types 890 * should be included: 891 * 892 * - `I915_MMAP_OFFSET_GTT`: Use mmap with the object bound to GTT. (Write-Combined) 893 * - `I915_MMAP_OFFSET_WC`: Use Write-Combined caching. 894 * - `I915_MMAP_OFFSET_WB`: Use Write-Back caching. 895 * - `I915_MMAP_OFFSET_FIXED`: Use object placement to determine caching. 896 * 897 * On devices with local memory `I915_MMAP_OFFSET_FIXED` is the only valid 898 * type. On devices without local memory, this caching mode is invalid. 899 * 900 * As caching mode when specifying `I915_MMAP_OFFSET_FIXED`, WC or WB will 901 * be used, depending on the object placement on creation. WB will be used 902 * when the object can only exist in system memory, WC otherwise. 903 */ 904 __u64 flags; 905 906 #define I915_MMAP_OFFSET_GTT 0 907 #define I915_MMAP_OFFSET_WC 1 908 #define I915_MMAP_OFFSET_WB 2 909 #define I915_MMAP_OFFSET_UC 3 910 #define I915_MMAP_OFFSET_FIXED 4 911 912 /** 913 * @extensions: Zero-terminated chain of extensions. 914 * 915 * No current extensions defined; mbz. 916 */ 917 __u64 extensions; 918 }; 919 920 /** 921 * struct drm_i915_gem_set_domain - Adjust the objects write or read domain, in 922 * preparation for accessing the pages via some CPU domain. 923 * 924 * Specifying a new write or read domain will flush the object out of the 925 * previous domain(if required), before then updating the objects domain 926 * tracking with the new domain. 927 * 928 * Note this might involve waiting for the object first if it is still active on 929 * the GPU. 930 * 931 * Supported values for @read_domains and @write_domain: 932 * 933 * - I915_GEM_DOMAIN_WC: Uncached write-combined domain 934 * - I915_GEM_DOMAIN_CPU: CPU cache domain 935 * - I915_GEM_DOMAIN_GTT: Mappable aperture domain 936 * 937 * All other domains are rejected. 938 * 939 * Note that for discrete, starting from DG1, this is no longer supported, and 940 * is instead rejected. On such platforms the CPU domain is effectively static, 941 * where we also only support a single &drm_i915_gem_mmap_offset cache mode, 942 * which can't be set explicitly and instead depends on the object placements, 943 * as per the below. 944 * 945 * Implicit caching rules, starting from DG1: 946 * 947 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 948 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 949 * mapped as write-combined only. 950 * 951 * - Everything else is always allocated and mapped as write-back, with the 952 * guarantee that everything is also coherent with the GPU. 953 * 954 * Note that this is likely to change in the future again, where we might need 955 * more flexibility on future devices, so making this all explicit as part of a 956 * new &drm_i915_gem_create_ext extension is probable. 957 */ 958 struct drm_i915_gem_set_domain { 959 /** @handle: Handle for the object. */ 960 __u32 handle; 961 962 /** @read_domains: New read domains. */ 963 __u32 read_domains; 964 965 /** 966 * @write_domain: New write domain. 967 * 968 * Note that having something in the write domain implies it's in the 969 * read domain, and only that read domain. 970 */ 971 __u32 write_domain; 972 }; 973 974 struct drm_i915_gem_sw_finish { 975 /** Handle for the object */ 976 __u32 handle; 977 }; 978 979 struct drm_i915_gem_relocation_entry { 980 /** 981 * Handle of the buffer being pointed to by this relocation entry. 982 * 983 * It's appealing to make this be an index into the mm_validate_entry 984 * list to refer to the buffer, but this allows the driver to create 985 * a relocation list for state buffers and not re-write it per 986 * exec using the buffer. 987 */ 988 __u32 target_handle; 989 990 /** 991 * Value to be added to the offset of the target buffer to make up 992 * the relocation entry. 993 */ 994 __u32 delta; 995 996 /** Offset in the buffer the relocation entry will be written into */ 997 __u64 offset; 998 999 /** 1000 * Offset value of the target buffer that the relocation entry was last 1001 * written as. 1002 * 1003 * If the buffer has the same offset as last time, we can skip syncing 1004 * and writing the relocation. This value is written back out by 1005 * the execbuffer ioctl when the relocation is written. 1006 */ 1007 __u64 presumed_offset; 1008 1009 /** 1010 * Target memory domains read by this operation. 1011 */ 1012 __u32 read_domains; 1013 1014 /** 1015 * Target memory domains written by this operation. 1016 * 1017 * Note that only one domain may be written by the whole 1018 * execbuffer operation, so that where there are conflicts, 1019 * the application will get -EINVAL back. 1020 */ 1021 __u32 write_domain; 1022 }; 1023 1024 /** @{ 1025 * Intel memory domains 1026 * 1027 * Most of these just align with the various caches in 1028 * the system and are used to flush and invalidate as 1029 * objects end up cached in different domains. 1030 */ 1031 /** CPU cache */ 1032 #define I915_GEM_DOMAIN_CPU 0x00000001 1033 /** Render cache, used by 2D and 3D drawing */ 1034 #define I915_GEM_DOMAIN_RENDER 0x00000002 1035 /** Sampler cache, used by texture engine */ 1036 #define I915_GEM_DOMAIN_SAMPLER 0x00000004 1037 /** Command queue, used to load batch buffers */ 1038 #define I915_GEM_DOMAIN_COMMAND 0x00000008 1039 /** Instruction cache, used by shader programs */ 1040 #define I915_GEM_DOMAIN_INSTRUCTION 0x00000010 1041 /** Vertex address cache */ 1042 #define I915_GEM_DOMAIN_VERTEX 0x00000020 1043 /** GTT domain - aperture and scanout */ 1044 #define I915_GEM_DOMAIN_GTT 0x00000040 1045 /** WC domain - uncached access */ 1046 #define I915_GEM_DOMAIN_WC 0x00000080 1047 /** @} */ 1048 1049 struct drm_i915_gem_exec_object { 1050 /** 1051 * User's handle for a buffer to be bound into the GTT for this 1052 * operation. 1053 */ 1054 __u32 handle; 1055 1056 /** Number of relocations to be performed on this buffer */ 1057 __u32 relocation_count; 1058 /** 1059 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1060 * the relocations to be performed in this buffer. 1061 */ 1062 __u64 relocs_ptr; 1063 1064 /** Required alignment in graphics aperture */ 1065 __u64 alignment; 1066 1067 /** 1068 * Returned value of the updated offset of the object, for future 1069 * presumed_offset writes. 1070 */ 1071 __u64 offset; 1072 }; 1073 1074 /* DRM_IOCTL_I915_GEM_EXECBUFFER was removed in Linux 5.13 */ 1075 struct drm_i915_gem_execbuffer { 1076 /** 1077 * List of buffers to be validated with their relocations to be 1078 * performend on them. 1079 * 1080 * This is a pointer to an array of struct drm_i915_gem_validate_entry. 1081 * 1082 * These buffers must be listed in an order such that all relocations 1083 * a buffer is performing refer to buffers that have already appeared 1084 * in the validate list. 1085 */ 1086 __u64 buffers_ptr; 1087 __u32 buffer_count; 1088 1089 /** Offset in the batchbuffer to start execution from. */ 1090 __u32 batch_start_offset; 1091 /** Bytes used in batchbuffer from batch_start_offset */ 1092 __u32 batch_len; 1093 __u32 DR1; 1094 __u32 DR4; 1095 __u32 num_cliprects; 1096 /** This is a struct drm_clip_rect *cliprects */ 1097 __u64 cliprects_ptr; 1098 }; 1099 1100 struct drm_i915_gem_exec_object2 { 1101 /** 1102 * User's handle for a buffer to be bound into the GTT for this 1103 * operation. 1104 */ 1105 __u32 handle; 1106 1107 /** Number of relocations to be performed on this buffer */ 1108 __u32 relocation_count; 1109 /** 1110 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1111 * the relocations to be performed in this buffer. 1112 */ 1113 __u64 relocs_ptr; 1114 1115 /** Required alignment in graphics aperture */ 1116 __u64 alignment; 1117 1118 /** 1119 * When the EXEC_OBJECT_PINNED flag is specified this is populated by 1120 * the user with the GTT offset at which this object will be pinned. 1121 * 1122 * When the I915_EXEC_NO_RELOC flag is specified this must contain the 1123 * presumed_offset of the object. 1124 * 1125 * During execbuffer2 the kernel populates it with the value of the 1126 * current GTT offset of the object, for future presumed_offset writes. 1127 * 1128 * See struct drm_i915_gem_create_ext for the rules when dealing with 1129 * alignment restrictions with I915_MEMORY_CLASS_DEVICE, on devices with 1130 * minimum page sizes, like DG2. 1131 */ 1132 __u64 offset; 1133 1134 #define EXEC_OBJECT_NEEDS_FENCE (1<<0) 1135 #define EXEC_OBJECT_NEEDS_GTT (1<<1) 1136 #define EXEC_OBJECT_WRITE (1<<2) 1137 #define EXEC_OBJECT_SUPPORTS_48B_ADDRESS (1<<3) 1138 #define EXEC_OBJECT_PINNED (1<<4) 1139 #define EXEC_OBJECT_PAD_TO_SIZE (1<<5) 1140 /* The kernel implicitly tracks GPU activity on all GEM objects, and 1141 * synchronises operations with outstanding rendering. This includes 1142 * rendering on other devices if exported via dma-buf. However, sometimes 1143 * this tracking is too coarse and the user knows better. For example, 1144 * if the object is split into non-overlapping ranges shared between different 1145 * clients or engines (i.e. suballocating objects), the implicit tracking 1146 * by kernel assumes that each operation affects the whole object rather 1147 * than an individual range, causing needless synchronisation between clients. 1148 * The kernel will also forgo any CPU cache flushes prior to rendering from 1149 * the object as the client is expected to be also handling such domain 1150 * tracking. 1151 * 1152 * The kernel maintains the implicit tracking in order to manage resources 1153 * used by the GPU - this flag only disables the synchronisation prior to 1154 * rendering with this object in this execbuf. 1155 * 1156 * Opting out of implicit synhronisation requires the user to do its own 1157 * explicit tracking to avoid rendering corruption. See, for example, 1158 * I915_PARAM_HAS_EXEC_FENCE to order execbufs and execute them asynchronously. 1159 */ 1160 #define EXEC_OBJECT_ASYNC (1<<6) 1161 /* Request that the contents of this execobject be copied into the error 1162 * state upon a GPU hang involving this batch for post-mortem debugging. 1163 * These buffers are recorded in no particular order as "user" in 1164 * /sys/class/drm/cardN/error. Query I915_PARAM_HAS_EXEC_CAPTURE to see 1165 * if the kernel supports this flag. 1166 */ 1167 #define EXEC_OBJECT_CAPTURE (1<<7) 1168 /* All remaining bits are MBZ and RESERVED FOR FUTURE USE */ 1169 #define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_CAPTURE<<1) 1170 __u64 flags; 1171 1172 union { 1173 __u64 rsvd1; 1174 __u64 pad_to_size; 1175 }; 1176 __u64 rsvd2; 1177 }; 1178 1179 struct drm_i915_gem_exec_fence { 1180 /** 1181 * User's handle for a drm_syncobj to wait on or signal. 1182 */ 1183 __u32 handle; 1184 1185 #define I915_EXEC_FENCE_WAIT (1<<0) 1186 #define I915_EXEC_FENCE_SIGNAL (1<<1) 1187 #define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1)) 1188 __u32 flags; 1189 }; 1190 1191 /* 1192 * See drm_i915_gem_execbuffer_ext_timeline_fences. 1193 */ 1194 #define DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES 0 1195 1196 /* 1197 * This structure describes an array of drm_syncobj and associated points for 1198 * timeline variants of drm_syncobj. It is invalid to append this structure to 1199 * the execbuf if I915_EXEC_FENCE_ARRAY is set. 1200 */ 1201 struct drm_i915_gem_execbuffer_ext_timeline_fences { 1202 struct i915_user_extension base; 1203 1204 /** 1205 * Number of element in the handles_ptr & value_ptr arrays. 1206 */ 1207 __u64 fence_count; 1208 1209 /** 1210 * Pointer to an array of struct drm_i915_gem_exec_fence of length 1211 * fence_count. 1212 */ 1213 __u64 handles_ptr; 1214 1215 /** 1216 * Pointer to an array of u64 values of length fence_count. Values 1217 * must be 0 for a binary drm_syncobj. A Value of 0 for a timeline 1218 * drm_syncobj is invalid as it turns a drm_syncobj into a binary one. 1219 */ 1220 __u64 values_ptr; 1221 }; 1222 1223 struct drm_i915_gem_execbuffer2 { 1224 /** 1225 * List of gem_exec_object2 structs 1226 */ 1227 __u64 buffers_ptr; 1228 __u32 buffer_count; 1229 1230 /** Offset in the batchbuffer to start execution from. */ 1231 __u32 batch_start_offset; 1232 /** Bytes used in batchbuffer from batch_start_offset */ 1233 __u32 batch_len; 1234 __u32 DR1; 1235 __u32 DR4; 1236 __u32 num_cliprects; 1237 /** 1238 * This is a struct drm_clip_rect *cliprects if I915_EXEC_FENCE_ARRAY 1239 * & I915_EXEC_USE_EXTENSIONS are not set. 1240 * 1241 * If I915_EXEC_FENCE_ARRAY is set, then this is a pointer to an array 1242 * of struct drm_i915_gem_exec_fence and num_cliprects is the length 1243 * of the array. 1244 * 1245 * If I915_EXEC_USE_EXTENSIONS is set, then this is a pointer to a 1246 * single struct i915_user_extension and num_cliprects is 0. 1247 */ 1248 __u64 cliprects_ptr; 1249 #define I915_EXEC_RING_MASK (0x3f) 1250 #define I915_EXEC_DEFAULT (0<<0) 1251 #define I915_EXEC_RENDER (1<<0) 1252 #define I915_EXEC_BSD (2<<0) 1253 #define I915_EXEC_BLT (3<<0) 1254 #define I915_EXEC_VEBOX (4<<0) 1255 1256 /* Used for switching the constants addressing mode on gen4+ RENDER ring. 1257 * Gen6+ only supports relative addressing to dynamic state (default) and 1258 * absolute addressing. 1259 * 1260 * These flags are ignored for the BSD and BLT rings. 1261 */ 1262 #define I915_EXEC_CONSTANTS_MASK (3<<6) 1263 #define I915_EXEC_CONSTANTS_REL_GENERAL (0<<6) /* default */ 1264 #define I915_EXEC_CONSTANTS_ABSOLUTE (1<<6) 1265 #define I915_EXEC_CONSTANTS_REL_SURFACE (2<<6) /* gen4/5 only */ 1266 __u64 flags; 1267 __u64 rsvd1; /* now used for context info */ 1268 __u64 rsvd2; 1269 }; 1270 1271 /** Resets the SO write offset registers for transform feedback on gen7. */ 1272 #define I915_EXEC_GEN7_SOL_RESET (1<<8) 1273 1274 /** Request a privileged ("secure") batch buffer. Note only available for 1275 * DRM_ROOT_ONLY | DRM_MASTER processes. 1276 */ 1277 #define I915_EXEC_SECURE (1<<9) 1278 1279 /** Inform the kernel that the batch is and will always be pinned. This 1280 * negates the requirement for a workaround to be performed to avoid 1281 * an incoherent CS (such as can be found on 830/845). If this flag is 1282 * not passed, the kernel will endeavour to make sure the batch is 1283 * coherent with the CS before execution. If this flag is passed, 1284 * userspace assumes the responsibility for ensuring the same. 1285 */ 1286 #define I915_EXEC_IS_PINNED (1<<10) 1287 1288 /** Provide a hint to the kernel that the command stream and auxiliary 1289 * state buffers already holds the correct presumed addresses and so the 1290 * relocation process may be skipped if no buffers need to be moved in 1291 * preparation for the execbuffer. 1292 */ 1293 #define I915_EXEC_NO_RELOC (1<<11) 1294 1295 /** Use the reloc.handle as an index into the exec object array rather 1296 * than as the per-file handle. 1297 */ 1298 #define I915_EXEC_HANDLE_LUT (1<<12) 1299 1300 /** Used for switching BSD rings on the platforms with two BSD rings */ 1301 #define I915_EXEC_BSD_SHIFT (13) 1302 #define I915_EXEC_BSD_MASK (3 << I915_EXEC_BSD_SHIFT) 1303 /* default ping-pong mode */ 1304 #define I915_EXEC_BSD_DEFAULT (0 << I915_EXEC_BSD_SHIFT) 1305 #define I915_EXEC_BSD_RING1 (1 << I915_EXEC_BSD_SHIFT) 1306 #define I915_EXEC_BSD_RING2 (2 << I915_EXEC_BSD_SHIFT) 1307 1308 /** Tell the kernel that the batchbuffer is processed by 1309 * the resource streamer. 1310 */ 1311 #define I915_EXEC_RESOURCE_STREAMER (1<<15) 1312 1313 /* Setting I915_EXEC_FENCE_IN implies that lower_32_bits(rsvd2) represent 1314 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1315 * the batch. 1316 * 1317 * Returns -EINVAL if the sync_file fd cannot be found. 1318 */ 1319 #define I915_EXEC_FENCE_IN (1<<16) 1320 1321 /* Setting I915_EXEC_FENCE_OUT causes the ioctl to return a sync_file fd 1322 * in the upper_32_bits(rsvd2) upon success. Ownership of the fd is given 1323 * to the caller, and it should be close() after use. (The fd is a regular 1324 * file descriptor and will be cleaned up on process termination. It holds 1325 * a reference to the request, but nothing else.) 1326 * 1327 * The sync_file fd can be combined with other sync_file and passed either 1328 * to execbuf using I915_EXEC_FENCE_IN, to atomic KMS ioctls (so that a flip 1329 * will only occur after this request completes), or to other devices. 1330 * 1331 * Using I915_EXEC_FENCE_OUT requires use of 1332 * DRM_IOCTL_I915_GEM_EXECBUFFER2_WR ioctl so that the result is written 1333 * back to userspace. Failure to do so will cause the out-fence to always 1334 * be reported as zero, and the real fence fd to be leaked. 1335 */ 1336 #define I915_EXEC_FENCE_OUT (1<<17) 1337 1338 /* 1339 * Traditionally the execbuf ioctl has only considered the final element in 1340 * the execobject[] to be the executable batch. Often though, the client 1341 * will known the batch object prior to construction and being able to place 1342 * it into the execobject[] array first can simplify the relocation tracking. 1343 * Setting I915_EXEC_BATCH_FIRST tells execbuf to use element 0 of the 1344 * execobject[] as the * batch instead (the default is to use the last 1345 * element). 1346 */ 1347 #define I915_EXEC_BATCH_FIRST (1<<18) 1348 1349 /* Setting I915_FENCE_ARRAY implies that num_cliprects and cliprects_ptr 1350 * define an array of i915_gem_exec_fence structures which specify a set of 1351 * dma fences to wait upon or signal. 1352 */ 1353 #define I915_EXEC_FENCE_ARRAY (1<<19) 1354 1355 /* 1356 * Setting I915_EXEC_FENCE_SUBMIT implies that lower_32_bits(rsvd2) represent 1357 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1358 * the batch. 1359 * 1360 * Returns -EINVAL if the sync_file fd cannot be found. 1361 */ 1362 #define I915_EXEC_FENCE_SUBMIT (1 << 20) 1363 1364 /* 1365 * Setting I915_EXEC_USE_EXTENSIONS implies that 1366 * drm_i915_gem_execbuffer2.cliprects_ptr is treated as a pointer to an linked 1367 * list of i915_user_extension. Each i915_user_extension node is the base of a 1368 * larger structure. The list of supported structures are listed in the 1369 * drm_i915_gem_execbuffer_ext enum. 1370 */ 1371 #define I915_EXEC_USE_EXTENSIONS (1 << 21) 1372 1373 #define __I915_EXEC_UNKNOWN_FLAGS (-(I915_EXEC_USE_EXTENSIONS << 1)) 1374 1375 #define I915_EXEC_CONTEXT_ID_MASK (0xffffffff) 1376 #define i915_execbuffer2_set_context_id(eb2, context) \ 1377 (eb2).rsvd1 = context & I915_EXEC_CONTEXT_ID_MASK 1378 #define i915_execbuffer2_get_context_id(eb2) \ 1379 ((eb2).rsvd1 & I915_EXEC_CONTEXT_ID_MASK) 1380 1381 struct drm_i915_gem_pin { 1382 /** Handle of the buffer to be pinned. */ 1383 __u32 handle; 1384 __u32 pad; 1385 1386 /** alignment required within the aperture */ 1387 __u64 alignment; 1388 1389 /** Returned GTT offset of the buffer. */ 1390 __u64 offset; 1391 }; 1392 1393 struct drm_i915_gem_unpin { 1394 /** Handle of the buffer to be unpinned. */ 1395 __u32 handle; 1396 __u32 pad; 1397 }; 1398 1399 struct drm_i915_gem_busy { 1400 /** Handle of the buffer to check for busy */ 1401 __u32 handle; 1402 1403 /** Return busy status 1404 * 1405 * A return of 0 implies that the object is idle (after 1406 * having flushed any pending activity), and a non-zero return that 1407 * the object is still in-flight on the GPU. (The GPU has not yet 1408 * signaled completion for all pending requests that reference the 1409 * object.) An object is guaranteed to become idle eventually (so 1410 * long as no new GPU commands are executed upon it). Due to the 1411 * asynchronous nature of the hardware, an object reported 1412 * as busy may become idle before the ioctl is completed. 1413 * 1414 * Furthermore, if the object is busy, which engine is busy is only 1415 * provided as a guide and only indirectly by reporting its class 1416 * (there may be more than one engine in each class). There are race 1417 * conditions which prevent the report of which engines are busy from 1418 * being always accurate. However, the converse is not true. If the 1419 * object is idle, the result of the ioctl, that all engines are idle, 1420 * is accurate. 1421 * 1422 * The returned dword is split into two fields to indicate both 1423 * the engine classess on which the object is being read, and the 1424 * engine class on which it is currently being written (if any). 1425 * 1426 * The low word (bits 0:15) indicate if the object is being written 1427 * to by any engine (there can only be one, as the GEM implicit 1428 * synchronisation rules force writes to be serialised). Only the 1429 * engine class (offset by 1, I915_ENGINE_CLASS_RENDER is reported as 1430 * 1 not 0 etc) for the last write is reported. 1431 * 1432 * The high word (bits 16:31) are a bitmask of which engines classes 1433 * are currently reading from the object. Multiple engines may be 1434 * reading from the object simultaneously. 1435 * 1436 * The value of each engine class is the same as specified in the 1437 * I915_CONTEXT_PARAM_ENGINES context parameter and via perf, i.e. 1438 * I915_ENGINE_CLASS_RENDER, I915_ENGINE_CLASS_COPY, etc. 1439 * Some hardware may have parallel execution engines, e.g. multiple 1440 * media engines, which are mapped to the same class identifier and so 1441 * are not separately reported for busyness. 1442 * 1443 * Caveat emptor: 1444 * Only the boolean result of this query is reliable; that is whether 1445 * the object is idle or busy. The report of which engines are busy 1446 * should be only used as a heuristic. 1447 */ 1448 __u32 busy; 1449 }; 1450 1451 /** 1452 * struct drm_i915_gem_caching - Set or get the caching for given object 1453 * handle. 1454 * 1455 * Allow userspace to control the GTT caching bits for a given object when the 1456 * object is later mapped through the ppGTT(or GGTT on older platforms lacking 1457 * ppGTT support, or if the object is used for scanout). Note that this might 1458 * require unbinding the object from the GTT first, if its current caching value 1459 * doesn't match. 1460 * 1461 * Note that this all changes on discrete platforms, starting from DG1, the 1462 * set/get caching is no longer supported, and is now rejected. Instead the CPU 1463 * caching attributes(WB vs WC) will become an immutable creation time property 1464 * for the object, along with the GTT caching level. For now we don't expose any 1465 * new uAPI for this, instead on DG1 this is all implicit, although this largely 1466 * shouldn't matter since DG1 is coherent by default(without any way of 1467 * controlling it). 1468 * 1469 * Implicit caching rules, starting from DG1: 1470 * 1471 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1472 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1473 * mapped as write-combined only. 1474 * 1475 * - Everything else is always allocated and mapped as write-back, with the 1476 * guarantee that everything is also coherent with the GPU. 1477 * 1478 * Note that this is likely to change in the future again, where we might need 1479 * more flexibility on future devices, so making this all explicit as part of a 1480 * new &drm_i915_gem_create_ext extension is probable. 1481 * 1482 * Side note: Part of the reason for this is that changing the at-allocation-time CPU 1483 * caching attributes for the pages might be required(and is expensive) if we 1484 * need to then CPU map the pages later with different caching attributes. This 1485 * inconsistent caching behaviour, while supported on x86, is not universally 1486 * supported on other architectures. So for simplicity we opt for setting 1487 * everything at creation time, whilst also making it immutable, on discrete 1488 * platforms. 1489 */ 1490 struct drm_i915_gem_caching { 1491 /** 1492 * @handle: Handle of the buffer to set/get the caching level. 1493 */ 1494 __u32 handle; 1495 1496 /** 1497 * @caching: The GTT caching level to apply or possible return value. 1498 * 1499 * The supported @caching values: 1500 * 1501 * I915_CACHING_NONE: 1502 * 1503 * GPU access is not coherent with CPU caches. Default for machines 1504 * without an LLC. This means manual flushing might be needed, if we 1505 * want GPU access to be coherent. 1506 * 1507 * I915_CACHING_CACHED: 1508 * 1509 * GPU access is coherent with CPU caches and furthermore the data is 1510 * cached in last-level caches shared between CPU cores and the GPU GT. 1511 * 1512 * I915_CACHING_DISPLAY: 1513 * 1514 * Special GPU caching mode which is coherent with the scanout engines. 1515 * Transparently falls back to I915_CACHING_NONE on platforms where no 1516 * special cache mode (like write-through or gfdt flushing) is 1517 * available. The kernel automatically sets this mode when using a 1518 * buffer as a scanout target. Userspace can manually set this mode to 1519 * avoid a costly stall and clflush in the hotpath of drawing the first 1520 * frame. 1521 */ 1522 #define I915_CACHING_NONE 0 1523 #define I915_CACHING_CACHED 1 1524 #define I915_CACHING_DISPLAY 2 1525 __u32 caching; 1526 }; 1527 1528 #define I915_TILING_NONE 0 1529 #define I915_TILING_X 1 1530 #define I915_TILING_Y 2 1531 /* 1532 * Do not add new tiling types here. The I915_TILING_* values are for 1533 * de-tiling fence registers that no longer exist on modern platforms. Although 1534 * the hardware may support new types of tiling in general (e.g., Tile4), we 1535 * do not need to add them to the uapi that is specific to now-defunct ioctls. 1536 */ 1537 #define I915_TILING_LAST I915_TILING_Y 1538 1539 #define I915_BIT_6_SWIZZLE_NONE 0 1540 #define I915_BIT_6_SWIZZLE_9 1 1541 #define I915_BIT_6_SWIZZLE_9_10 2 1542 #define I915_BIT_6_SWIZZLE_9_11 3 1543 #define I915_BIT_6_SWIZZLE_9_10_11 4 1544 /* Not seen by userland */ 1545 #define I915_BIT_6_SWIZZLE_UNKNOWN 5 1546 /* Seen by userland. */ 1547 #define I915_BIT_6_SWIZZLE_9_17 6 1548 #define I915_BIT_6_SWIZZLE_9_10_17 7 1549 1550 struct drm_i915_gem_set_tiling { 1551 /** Handle of the buffer to have its tiling state updated */ 1552 __u32 handle; 1553 1554 /** 1555 * Tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1556 * I915_TILING_Y). 1557 * 1558 * This value is to be set on request, and will be updated by the 1559 * kernel on successful return with the actual chosen tiling layout. 1560 * 1561 * The tiling mode may be demoted to I915_TILING_NONE when the system 1562 * has bit 6 swizzling that can't be managed correctly by GEM. 1563 * 1564 * Buffer contents become undefined when changing tiling_mode. 1565 */ 1566 __u32 tiling_mode; 1567 1568 /** 1569 * Stride in bytes for the object when in I915_TILING_X or 1570 * I915_TILING_Y. 1571 */ 1572 __u32 stride; 1573 1574 /** 1575 * Returned address bit 6 swizzling required for CPU access through 1576 * mmap mapping. 1577 */ 1578 __u32 swizzle_mode; 1579 }; 1580 1581 struct drm_i915_gem_get_tiling { 1582 /** Handle of the buffer to get tiling state for. */ 1583 __u32 handle; 1584 1585 /** 1586 * Current tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1587 * I915_TILING_Y). 1588 */ 1589 __u32 tiling_mode; 1590 1591 /** 1592 * Returned address bit 6 swizzling required for CPU access through 1593 * mmap mapping. 1594 */ 1595 __u32 swizzle_mode; 1596 1597 /** 1598 * Returned address bit 6 swizzling required for CPU access through 1599 * mmap mapping whilst bound. 1600 */ 1601 __u32 phys_swizzle_mode; 1602 }; 1603 1604 struct drm_i915_gem_get_aperture { 1605 /** Total size of the aperture used by i915_gem_execbuffer, in bytes */ 1606 __u64 aper_size; 1607 1608 /** 1609 * Available space in the aperture used by i915_gem_execbuffer, in 1610 * bytes 1611 */ 1612 __u64 aper_available_size; 1613 }; 1614 1615 struct drm_i915_get_pipe_from_crtc_id { 1616 /** ID of CRTC being requested **/ 1617 __u32 crtc_id; 1618 1619 /** pipe of requested CRTC **/ 1620 __u32 pipe; 1621 }; 1622 1623 #define I915_MADV_WILLNEED 0 1624 #define I915_MADV_DONTNEED 1 1625 #define __I915_MADV_PURGED 2 /* internal state */ 1626 1627 struct drm_i915_gem_madvise { 1628 /** Handle of the buffer to change the backing store advice */ 1629 __u32 handle; 1630 1631 /* Advice: either the buffer will be needed again in the near future, 1632 * or wont be and could be discarded under memory pressure. 1633 */ 1634 __u32 madv; 1635 1636 /** Whether the backing store still exists. */ 1637 __u32 retained; 1638 }; 1639 1640 /* flags */ 1641 #define I915_OVERLAY_TYPE_MASK 0xff 1642 #define I915_OVERLAY_YUV_PLANAR 0x01 1643 #define I915_OVERLAY_YUV_PACKED 0x02 1644 #define I915_OVERLAY_RGB 0x03 1645 1646 #define I915_OVERLAY_DEPTH_MASK 0xff00 1647 #define I915_OVERLAY_RGB24 0x1000 1648 #define I915_OVERLAY_RGB16 0x2000 1649 #define I915_OVERLAY_RGB15 0x3000 1650 #define I915_OVERLAY_YUV422 0x0100 1651 #define I915_OVERLAY_YUV411 0x0200 1652 #define I915_OVERLAY_YUV420 0x0300 1653 #define I915_OVERLAY_YUV410 0x0400 1654 1655 #define I915_OVERLAY_SWAP_MASK 0xff0000 1656 #define I915_OVERLAY_NO_SWAP 0x000000 1657 #define I915_OVERLAY_UV_SWAP 0x010000 1658 #define I915_OVERLAY_Y_SWAP 0x020000 1659 #define I915_OVERLAY_Y_AND_UV_SWAP 0x030000 1660 1661 #define I915_OVERLAY_FLAGS_MASK 0xff000000 1662 #define I915_OVERLAY_ENABLE 0x01000000 1663 1664 struct drm_intel_overlay_put_image { 1665 /* various flags and src format description */ 1666 __u32 flags; 1667 /* source picture description */ 1668 __u32 bo_handle; 1669 /* stride values and offsets are in bytes, buffer relative */ 1670 __u16 stride_Y; /* stride for packed formats */ 1671 __u16 stride_UV; 1672 __u32 offset_Y; /* offset for packet formats */ 1673 __u32 offset_U; 1674 __u32 offset_V; 1675 /* in pixels */ 1676 __u16 src_width; 1677 __u16 src_height; 1678 /* to compensate the scaling factors for partially covered surfaces */ 1679 __u16 src_scan_width; 1680 __u16 src_scan_height; 1681 /* output crtc description */ 1682 __u32 crtc_id; 1683 __u16 dst_x; 1684 __u16 dst_y; 1685 __u16 dst_width; 1686 __u16 dst_height; 1687 }; 1688 1689 /* flags */ 1690 #define I915_OVERLAY_UPDATE_ATTRS (1<<0) 1691 #define I915_OVERLAY_UPDATE_GAMMA (1<<1) 1692 #define I915_OVERLAY_DISABLE_DEST_COLORKEY (1<<2) 1693 struct drm_intel_overlay_attrs { 1694 __u32 flags; 1695 __u32 color_key; 1696 __s32 brightness; 1697 __u32 contrast; 1698 __u32 saturation; 1699 __u32 gamma0; 1700 __u32 gamma1; 1701 __u32 gamma2; 1702 __u32 gamma3; 1703 __u32 gamma4; 1704 __u32 gamma5; 1705 }; 1706 1707 /* 1708 * Intel sprite handling 1709 * 1710 * Color keying works with a min/mask/max tuple. Both source and destination 1711 * color keying is allowed. 1712 * 1713 * Source keying: 1714 * Sprite pixels within the min & max values, masked against the color channels 1715 * specified in the mask field, will be transparent. All other pixels will 1716 * be displayed on top of the primary plane. For RGB surfaces, only the min 1717 * and mask fields will be used; ranged compares are not allowed. 1718 * 1719 * Destination keying: 1720 * Primary plane pixels that match the min value, masked against the color 1721 * channels specified in the mask field, will be replaced by corresponding 1722 * pixels from the sprite plane. 1723 * 1724 * Note that source & destination keying are exclusive; only one can be 1725 * active on a given plane. 1726 */ 1727 1728 #define I915_SET_COLORKEY_NONE (1<<0) /* Deprecated. Instead set 1729 * flags==0 to disable colorkeying. 1730 */ 1731 #define I915_SET_COLORKEY_DESTINATION (1<<1) 1732 #define I915_SET_COLORKEY_SOURCE (1<<2) 1733 struct drm_intel_sprite_colorkey { 1734 __u32 plane_id; 1735 __u32 min_value; 1736 __u32 channel_mask; 1737 __u32 max_value; 1738 __u32 flags; 1739 }; 1740 1741 struct drm_i915_gem_wait { 1742 /** Handle of BO we shall wait on */ 1743 __u32 bo_handle; 1744 __u32 flags; 1745 /** Number of nanoseconds to wait, Returns time remaining. */ 1746 __s64 timeout_ns; 1747 }; 1748 1749 struct drm_i915_gem_context_create { 1750 __u32 ctx_id; /* output: id of new context*/ 1751 __u32 pad; 1752 }; 1753 1754 struct drm_i915_gem_context_create_ext { 1755 __u32 ctx_id; /* output: id of new context*/ 1756 __u32 flags; 1757 #define I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS (1u << 0) 1758 #define I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE (1u << 1) 1759 #define I915_CONTEXT_CREATE_FLAGS_UNKNOWN \ 1760 (-(I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE << 1)) 1761 __u64 extensions; 1762 }; 1763 1764 struct drm_i915_gem_context_param { 1765 __u32 ctx_id; 1766 __u32 size; 1767 __u64 param; 1768 #define I915_CONTEXT_PARAM_BAN_PERIOD 0x1 1769 /* I915_CONTEXT_PARAM_NO_ZEROMAP has been removed. On the off chance 1770 * someone somewhere has attempted to use it, never re-use this context 1771 * param number. 1772 */ 1773 #define I915_CONTEXT_PARAM_NO_ZEROMAP 0x2 1774 #define I915_CONTEXT_PARAM_GTT_SIZE 0x3 1775 #define I915_CONTEXT_PARAM_NO_ERROR_CAPTURE 0x4 1776 #define I915_CONTEXT_PARAM_BANNABLE 0x5 1777 #define I915_CONTEXT_PARAM_PRIORITY 0x6 1778 #define I915_CONTEXT_MAX_USER_PRIORITY 1023 /* inclusive */ 1779 #define I915_CONTEXT_DEFAULT_PRIORITY 0 1780 #define I915_CONTEXT_MIN_USER_PRIORITY -1023 /* inclusive */ 1781 /* 1782 * When using the following param, value should be a pointer to 1783 * drm_i915_gem_context_param_sseu. 1784 */ 1785 #define I915_CONTEXT_PARAM_SSEU 0x7 1786 1787 /* 1788 * Not all clients may want to attempt automatic recover of a context after 1789 * a hang (for example, some clients may only submit very small incremental 1790 * batches relying on known logical state of previous batches which will never 1791 * recover correctly and each attempt will hang), and so would prefer that 1792 * the context is forever banned instead. 1793 * 1794 * If set to false (0), after a reset, subsequent (and in flight) rendering 1795 * from this context is discarded, and the client will need to create a new 1796 * context to use instead. 1797 * 1798 * If set to true (1), the kernel will automatically attempt to recover the 1799 * context by skipping the hanging batch and executing the next batch starting 1800 * from the default context state (discarding the incomplete logical context 1801 * state lost due to the reset). 1802 * 1803 * On creation, all new contexts are marked as recoverable. 1804 */ 1805 #define I915_CONTEXT_PARAM_RECOVERABLE 0x8 1806 1807 /* 1808 * The id of the associated virtual memory address space (ppGTT) of 1809 * this context. Can be retrieved and passed to another context 1810 * (on the same fd) for both to use the same ppGTT and so share 1811 * address layouts, and avoid reloading the page tables on context 1812 * switches between themselves. 1813 * 1814 * See DRM_I915_GEM_VM_CREATE and DRM_I915_GEM_VM_DESTROY. 1815 */ 1816 #define I915_CONTEXT_PARAM_VM 0x9 1817 1818 /* 1819 * I915_CONTEXT_PARAM_ENGINES: 1820 * 1821 * Bind this context to operate on this subset of available engines. Henceforth, 1822 * the I915_EXEC_RING selector for DRM_IOCTL_I915_GEM_EXECBUFFER2 operates as 1823 * an index into this array of engines; I915_EXEC_DEFAULT selecting engine[0] 1824 * and upwards. Slots 0...N are filled in using the specified (class, instance). 1825 * Use 1826 * engine_class: I915_ENGINE_CLASS_INVALID, 1827 * engine_instance: I915_ENGINE_CLASS_INVALID_NONE 1828 * to specify a gap in the array that can be filled in later, e.g. by a 1829 * virtual engine used for load balancing. 1830 * 1831 * Setting the number of engines bound to the context to 0, by passing a zero 1832 * sized argument, will revert back to default settings. 1833 * 1834 * See struct i915_context_param_engines. 1835 * 1836 * Extensions: 1837 * i915_context_engines_load_balance (I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE) 1838 * i915_context_engines_bond (I915_CONTEXT_ENGINES_EXT_BOND) 1839 * i915_context_engines_parallel_submit (I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT) 1840 */ 1841 #define I915_CONTEXT_PARAM_ENGINES 0xa 1842 1843 /* 1844 * I915_CONTEXT_PARAM_PERSISTENCE: 1845 * 1846 * Allow the context and active rendering to survive the process until 1847 * completion. Persistence allows fire-and-forget clients to queue up a 1848 * bunch of work, hand the output over to a display server and then quit. 1849 * If the context is marked as not persistent, upon closing (either via 1850 * an explicit DRM_I915_GEM_CONTEXT_DESTROY or implicitly from file closure 1851 * or process termination), the context and any outstanding requests will be 1852 * cancelled (and exported fences for cancelled requests marked as -EIO). 1853 * 1854 * By default, new contexts allow persistence. 1855 */ 1856 #define I915_CONTEXT_PARAM_PERSISTENCE 0xb 1857 1858 /* This API has been removed. On the off chance someone somewhere has 1859 * attempted to use it, never re-use this context param number. 1860 */ 1861 #define I915_CONTEXT_PARAM_RINGSIZE 0xc 1862 1863 /* 1864 * I915_CONTEXT_PARAM_PROTECTED_CONTENT: 1865 * 1866 * Mark that the context makes use of protected content, which will result 1867 * in the context being invalidated when the protected content session is. 1868 * Given that the protected content session is killed on suspend, the device 1869 * is kept awake for the lifetime of a protected context, so the user should 1870 * make sure to dispose of them once done. 1871 * This flag can only be set at context creation time and, when set to true, 1872 * must be preceded by an explicit setting of I915_CONTEXT_PARAM_RECOVERABLE 1873 * to false. This flag can't be set to true in conjunction with setting the 1874 * I915_CONTEXT_PARAM_BANNABLE flag to false. Creation example: 1875 * 1876 * .. code-block:: C 1877 * 1878 * struct drm_i915_gem_context_create_ext_setparam p_protected = { 1879 * .base = { 1880 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 1881 * }, 1882 * .param = { 1883 * .param = I915_CONTEXT_PARAM_PROTECTED_CONTENT, 1884 * .value = 1, 1885 * } 1886 * }; 1887 * struct drm_i915_gem_context_create_ext_setparam p_norecover = { 1888 * .base = { 1889 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 1890 * .next_extension = to_user_pointer(&p_protected), 1891 * }, 1892 * .param = { 1893 * .param = I915_CONTEXT_PARAM_RECOVERABLE, 1894 * .value = 0, 1895 * } 1896 * }; 1897 * struct drm_i915_gem_context_create_ext create = { 1898 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 1899 * .extensions = to_user_pointer(&p_norecover); 1900 * }; 1901 * 1902 * ctx_id = gem_context_create_ext(drm_fd, &create); 1903 * 1904 * In addition to the normal failure cases, setting this flag during context 1905 * creation can result in the following errors: 1906 * 1907 * -ENODEV: feature not available 1908 * -EPERM: trying to mark a recoverable or not bannable context as protected 1909 */ 1910 #define I915_CONTEXT_PARAM_PROTECTED_CONTENT 0xd 1911 /* Must be kept compact -- no holes and well documented */ 1912 1913 __u64 value; 1914 }; 1915 1916 /* 1917 * Context SSEU programming 1918 * 1919 * It may be necessary for either functional or performance reason to configure 1920 * a context to run with a reduced number of SSEU (where SSEU stands for Slice/ 1921 * Sub-slice/EU). 1922 * 1923 * This is done by configuring SSEU configuration using the below 1924 * @struct drm_i915_gem_context_param_sseu for every supported engine which 1925 * userspace intends to use. 1926 * 1927 * Not all GPUs or engines support this functionality in which case an error 1928 * code -ENODEV will be returned. 1929 * 1930 * Also, flexibility of possible SSEU configuration permutations varies between 1931 * GPU generations and software imposed limitations. Requesting such a 1932 * combination will return an error code of -EINVAL. 1933 * 1934 * NOTE: When perf/OA is active the context's SSEU configuration is ignored in 1935 * favour of a single global setting. 1936 */ 1937 struct drm_i915_gem_context_param_sseu { 1938 /* 1939 * Engine class & instance to be configured or queried. 1940 */ 1941 struct i915_engine_class_instance engine; 1942 1943 /* 1944 * Unknown flags must be cleared to zero. 1945 */ 1946 __u32 flags; 1947 #define I915_CONTEXT_SSEU_FLAG_ENGINE_INDEX (1u << 0) 1948 1949 /* 1950 * Mask of slices to enable for the context. Valid values are a subset 1951 * of the bitmask value returned for I915_PARAM_SLICE_MASK. 1952 */ 1953 __u64 slice_mask; 1954 1955 /* 1956 * Mask of subslices to enable for the context. Valid values are a 1957 * subset of the bitmask value return by I915_PARAM_SUBSLICE_MASK. 1958 */ 1959 __u64 subslice_mask; 1960 1961 /* 1962 * Minimum/Maximum number of EUs to enable per subslice for the 1963 * context. min_eus_per_subslice must be inferior or equal to 1964 * max_eus_per_subslice. 1965 */ 1966 __u16 min_eus_per_subslice; 1967 __u16 max_eus_per_subslice; 1968 1969 /* 1970 * Unused for now. Must be cleared to zero. 1971 */ 1972 __u32 rsvd; 1973 }; 1974 1975 /** 1976 * DOC: Virtual Engine uAPI 1977 * 1978 * Virtual engine is a concept where userspace is able to configure a set of 1979 * physical engines, submit a batch buffer, and let the driver execute it on any 1980 * engine from the set as it sees fit. 1981 * 1982 * This is primarily useful on parts which have multiple instances of a same 1983 * class engine, like for example GT3+ Skylake parts with their two VCS engines. 1984 * 1985 * For instance userspace can enumerate all engines of a certain class using the 1986 * previously described `Engine Discovery uAPI`_. After that userspace can 1987 * create a GEM context with a placeholder slot for the virtual engine (using 1988 * `I915_ENGINE_CLASS_INVALID` and `I915_ENGINE_CLASS_INVALID_NONE` for class 1989 * and instance respectively) and finally using the 1990 * `I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE` extension place a virtual engine in 1991 * the same reserved slot. 1992 * 1993 * Example of creating a virtual engine and submitting a batch buffer to it: 1994 * 1995 * .. code-block:: C 1996 * 1997 * I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(virtual, 2) = { 1998 * .base.name = I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE, 1999 * .engine_index = 0, // Place this virtual engine into engine map slot 0 2000 * .num_siblings = 2, 2001 * .engines = { { I915_ENGINE_CLASS_VIDEO, 0 }, 2002 * { I915_ENGINE_CLASS_VIDEO, 1 }, }, 2003 * }; 2004 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 1) = { 2005 * .engines = { { I915_ENGINE_CLASS_INVALID, 2006 * I915_ENGINE_CLASS_INVALID_NONE } }, 2007 * .extensions = to_user_pointer(&virtual), // Chains after load_balance extension 2008 * }; 2009 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2010 * .base = { 2011 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2012 * }, 2013 * .param = { 2014 * .param = I915_CONTEXT_PARAM_ENGINES, 2015 * .value = to_user_pointer(&engines), 2016 * .size = sizeof(engines), 2017 * }, 2018 * }; 2019 * struct drm_i915_gem_context_create_ext create = { 2020 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2021 * .extensions = to_user_pointer(&p_engines); 2022 * }; 2023 * 2024 * ctx_id = gem_context_create_ext(drm_fd, &create); 2025 * 2026 * // Now we have created a GEM context with its engine map containing a 2027 * // single virtual engine. Submissions to this slot can go either to 2028 * // vcs0 or vcs1, depending on the load balancing algorithm used inside 2029 * // the driver. The load balancing is dynamic from one batch buffer to 2030 * // another and transparent to userspace. 2031 * 2032 * ... 2033 * execbuf.rsvd1 = ctx_id; 2034 * execbuf.flags = 0; // Submits to index 0 which is the virtual engine 2035 * gem_execbuf(drm_fd, &execbuf); 2036 */ 2037 2038 /* 2039 * i915_context_engines_load_balance: 2040 * 2041 * Enable load balancing across this set of engines. 2042 * 2043 * Into the I915_EXEC_DEFAULT slot [0], a virtual engine is created that when 2044 * used will proxy the execbuffer request onto one of the set of engines 2045 * in such a way as to distribute the load evenly across the set. 2046 * 2047 * The set of engines must be compatible (e.g. the same HW class) as they 2048 * will share the same logical GPU context and ring. 2049 * 2050 * To intermix rendering with the virtual engine and direct rendering onto 2051 * the backing engines (bypassing the load balancing proxy), the context must 2052 * be defined to use a single timeline for all engines. 2053 */ 2054 struct i915_context_engines_load_balance { 2055 struct i915_user_extension base; 2056 2057 __u16 engine_index; 2058 __u16 num_siblings; 2059 __u32 flags; /* all undefined flags must be zero */ 2060 2061 __u64 mbz64; /* reserved for future use; must be zero */ 2062 2063 struct i915_engine_class_instance engines[0]; 2064 } __attribute__((packed)); 2065 2066 #define I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(name__, N__) struct { \ 2067 struct i915_user_extension base; \ 2068 __u16 engine_index; \ 2069 __u16 num_siblings; \ 2070 __u32 flags; \ 2071 __u64 mbz64; \ 2072 struct i915_engine_class_instance engines[N__]; \ 2073 } __attribute__((packed)) name__ 2074 2075 /* 2076 * i915_context_engines_bond: 2077 * 2078 * Constructed bonded pairs for execution within a virtual engine. 2079 * 2080 * All engines are equal, but some are more equal than others. Given 2081 * the distribution of resources in the HW, it may be preferable to run 2082 * a request on a given subset of engines in parallel to a request on a 2083 * specific engine. We enable this selection of engines within a virtual 2084 * engine by specifying bonding pairs, for any given master engine we will 2085 * only execute on one of the corresponding siblings within the virtual engine. 2086 * 2087 * To execute a request in parallel on the master engine and a sibling requires 2088 * coordination with a I915_EXEC_FENCE_SUBMIT. 2089 */ 2090 struct i915_context_engines_bond { 2091 struct i915_user_extension base; 2092 2093 struct i915_engine_class_instance master; 2094 2095 __u16 virtual_index; /* index of virtual engine in ctx->engines[] */ 2096 __u16 num_bonds; 2097 2098 __u64 flags; /* all undefined flags must be zero */ 2099 __u64 mbz64[4]; /* reserved for future use; must be zero */ 2100 2101 struct i915_engine_class_instance engines[0]; 2102 } __attribute__((packed)); 2103 2104 #define I915_DEFINE_CONTEXT_ENGINES_BOND(name__, N__) struct { \ 2105 struct i915_user_extension base; \ 2106 struct i915_engine_class_instance master; \ 2107 __u16 virtual_index; \ 2108 __u16 num_bonds; \ 2109 __u64 flags; \ 2110 __u64 mbz64[4]; \ 2111 struct i915_engine_class_instance engines[N__]; \ 2112 } __attribute__((packed)) name__ 2113 2114 /** 2115 * struct i915_context_engines_parallel_submit - Configure engine for 2116 * parallel submission. 2117 * 2118 * Setup a slot in the context engine map to allow multiple BBs to be submitted 2119 * in a single execbuf IOCTL. Those BBs will then be scheduled to run on the GPU 2120 * in parallel. Multiple hardware contexts are created internally in the i915 to 2121 * run these BBs. Once a slot is configured for N BBs only N BBs can be 2122 * submitted in each execbuf IOCTL and this is implicit behavior e.g. The user 2123 * doesn't tell the execbuf IOCTL there are N BBs, the execbuf IOCTL knows how 2124 * many BBs there are based on the slot's configuration. The N BBs are the last 2125 * N buffer objects or first N if I915_EXEC_BATCH_FIRST is set. 2126 * 2127 * The default placement behavior is to create implicit bonds between each 2128 * context if each context maps to more than 1 physical engine (e.g. context is 2129 * a virtual engine). Also we only allow contexts of same engine class and these 2130 * contexts must be in logically contiguous order. Examples of the placement 2131 * behavior are described below. Lastly, the default is to not allow BBs to be 2132 * preempted mid-batch. Rather insert coordinated preemption points on all 2133 * hardware contexts between each set of BBs. Flags could be added in the future 2134 * to change both of these default behaviors. 2135 * 2136 * Returns -EINVAL if hardware context placement configuration is invalid or if 2137 * the placement configuration isn't supported on the platform / submission 2138 * interface. 2139 * Returns -ENODEV if extension isn't supported on the platform / submission 2140 * interface. 2141 * 2142 * .. code-block:: none 2143 * 2144 * Examples syntax: 2145 * CS[X] = generic engine of same class, logical instance X 2146 * INVALID = I915_ENGINE_CLASS_INVALID, I915_ENGINE_CLASS_INVALID_NONE 2147 * 2148 * Example 1 pseudo code: 2149 * set_engines(INVALID) 2150 * set_parallel(engine_index=0, width=2, num_siblings=1, 2151 * engines=CS[0],CS[1]) 2152 * 2153 * Results in the following valid placement: 2154 * CS[0], CS[1] 2155 * 2156 * Example 2 pseudo code: 2157 * set_engines(INVALID) 2158 * set_parallel(engine_index=0, width=2, num_siblings=2, 2159 * engines=CS[0],CS[2],CS[1],CS[3]) 2160 * 2161 * Results in the following valid placements: 2162 * CS[0], CS[1] 2163 * CS[2], CS[3] 2164 * 2165 * This can be thought of as two virtual engines, each containing two 2166 * engines thereby making a 2D array. However, there are bonds tying the 2167 * entries together and placing restrictions on how they can be scheduled. 2168 * Specifically, the scheduler can choose only vertical columns from the 2D 2169 * array. That is, CS[0] is bonded to CS[1] and CS[2] to CS[3]. So if the 2170 * scheduler wants to submit to CS[0], it must also choose CS[1] and vice 2171 * versa. Same for CS[2] requires also using CS[3]. 2172 * VE[0] = CS[0], CS[2] 2173 * VE[1] = CS[1], CS[3] 2174 * 2175 * Example 3 pseudo code: 2176 * set_engines(INVALID) 2177 * set_parallel(engine_index=0, width=2, num_siblings=2, 2178 * engines=CS[0],CS[1],CS[1],CS[3]) 2179 * 2180 * Results in the following valid and invalid placements: 2181 * CS[0], CS[1] 2182 * CS[1], CS[3] - Not logically contiguous, return -EINVAL 2183 */ 2184 struct i915_context_engines_parallel_submit { 2185 /** 2186 * @base: base user extension. 2187 */ 2188 struct i915_user_extension base; 2189 2190 /** 2191 * @engine_index: slot for parallel engine 2192 */ 2193 __u16 engine_index; 2194 2195 /** 2196 * @width: number of contexts per parallel engine or in other words the 2197 * number of batches in each submission 2198 */ 2199 __u16 width; 2200 2201 /** 2202 * @num_siblings: number of siblings per context or in other words the 2203 * number of possible placements for each submission 2204 */ 2205 __u16 num_siblings; 2206 2207 /** 2208 * @mbz16: reserved for future use; must be zero 2209 */ 2210 __u16 mbz16; 2211 2212 /** 2213 * @flags: all undefined flags must be zero, currently not defined flags 2214 */ 2215 __u64 flags; 2216 2217 /** 2218 * @mbz64: reserved for future use; must be zero 2219 */ 2220 __u64 mbz64[3]; 2221 2222 /** 2223 * @engines: 2-d array of engine instances to configure parallel engine 2224 * 2225 * length = width (i) * num_siblings (j) 2226 * index = j + i * num_siblings 2227 */ 2228 struct i915_engine_class_instance engines[0]; 2229 2230 } __packed; 2231 2232 #define I915_DEFINE_CONTEXT_ENGINES_PARALLEL_SUBMIT(name__, N__) struct { \ 2233 struct i915_user_extension base; \ 2234 __u16 engine_index; \ 2235 __u16 width; \ 2236 __u16 num_siblings; \ 2237 __u16 mbz16; \ 2238 __u64 flags; \ 2239 __u64 mbz64[3]; \ 2240 struct i915_engine_class_instance engines[N__]; \ 2241 } __attribute__((packed)) name__ 2242 2243 /** 2244 * DOC: Context Engine Map uAPI 2245 * 2246 * Context engine map is a new way of addressing engines when submitting batch- 2247 * buffers, replacing the existing way of using identifiers like `I915_EXEC_BLT` 2248 * inside the flags field of `struct drm_i915_gem_execbuffer2`. 2249 * 2250 * To use it created GEM contexts need to be configured with a list of engines 2251 * the user is intending to submit to. This is accomplished using the 2252 * `I915_CONTEXT_PARAM_ENGINES` parameter and `struct 2253 * i915_context_param_engines`. 2254 * 2255 * For such contexts the `I915_EXEC_RING_MASK` field becomes an index into the 2256 * configured map. 2257 * 2258 * Example of creating such context and submitting against it: 2259 * 2260 * .. code-block:: C 2261 * 2262 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 2) = { 2263 * .engines = { { I915_ENGINE_CLASS_RENDER, 0 }, 2264 * { I915_ENGINE_CLASS_COPY, 0 } } 2265 * }; 2266 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2267 * .base = { 2268 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2269 * }, 2270 * .param = { 2271 * .param = I915_CONTEXT_PARAM_ENGINES, 2272 * .value = to_user_pointer(&engines), 2273 * .size = sizeof(engines), 2274 * }, 2275 * }; 2276 * struct drm_i915_gem_context_create_ext create = { 2277 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2278 * .extensions = to_user_pointer(&p_engines); 2279 * }; 2280 * 2281 * ctx_id = gem_context_create_ext(drm_fd, &create); 2282 * 2283 * // We have now created a GEM context with two engines in the map: 2284 * // Index 0 points to rcs0 while index 1 points to bcs0. Other engines 2285 * // will not be accessible from this context. 2286 * 2287 * ... 2288 * execbuf.rsvd1 = ctx_id; 2289 * execbuf.flags = 0; // Submits to index 0, which is rcs0 for this context 2290 * gem_execbuf(drm_fd, &execbuf); 2291 * 2292 * ... 2293 * execbuf.rsvd1 = ctx_id; 2294 * execbuf.flags = 1; // Submits to index 0, which is bcs0 for this context 2295 * gem_execbuf(drm_fd, &execbuf); 2296 */ 2297 2298 struct i915_context_param_engines { 2299 __u64 extensions; /* linked chain of extension blocks, 0 terminates */ 2300 #define I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE 0 /* see i915_context_engines_load_balance */ 2301 #define I915_CONTEXT_ENGINES_EXT_BOND 1 /* see i915_context_engines_bond */ 2302 #define I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT 2 /* see i915_context_engines_parallel_submit */ 2303 struct i915_engine_class_instance engines[0]; 2304 } __attribute__((packed)); 2305 2306 #define I915_DEFINE_CONTEXT_PARAM_ENGINES(name__, N__) struct { \ 2307 __u64 extensions; \ 2308 struct i915_engine_class_instance engines[N__]; \ 2309 } __attribute__((packed)) name__ 2310 2311 struct drm_i915_gem_context_create_ext_setparam { 2312 #define I915_CONTEXT_CREATE_EXT_SETPARAM 0 2313 struct i915_user_extension base; 2314 struct drm_i915_gem_context_param param; 2315 }; 2316 2317 /* This API has been removed. On the off chance someone somewhere has 2318 * attempted to use it, never re-use this extension number. 2319 */ 2320 #define I915_CONTEXT_CREATE_EXT_CLONE 1 2321 2322 struct drm_i915_gem_context_destroy { 2323 __u32 ctx_id; 2324 __u32 pad; 2325 }; 2326 2327 /* 2328 * DRM_I915_GEM_VM_CREATE - 2329 * 2330 * Create a new virtual memory address space (ppGTT) for use within a context 2331 * on the same file. Extensions can be provided to configure exactly how the 2332 * address space is setup upon creation. 2333 * 2334 * The id of new VM (bound to the fd) for use with I915_CONTEXT_PARAM_VM is 2335 * returned in the outparam @id. 2336 * 2337 * No flags are defined, with all bits reserved and must be zero. 2338 * 2339 * An extension chain maybe provided, starting with @extensions, and terminated 2340 * by the @next_extension being 0. Currently, no extensions are defined. 2341 * 2342 * DRM_I915_GEM_VM_DESTROY - 2343 * 2344 * Destroys a previously created VM id, specified in @id. 2345 * 2346 * No extensions or flags are allowed currently, and so must be zero. 2347 */ 2348 struct drm_i915_gem_vm_control { 2349 __u64 extensions; 2350 __u32 flags; 2351 __u32 vm_id; 2352 }; 2353 2354 struct drm_i915_reg_read { 2355 /* 2356 * Register offset. 2357 * For 64bit wide registers where the upper 32bits don't immediately 2358 * follow the lower 32bits, the offset of the lower 32bits must 2359 * be specified 2360 */ 2361 __u64 offset; 2362 #define I915_REG_READ_8B_WA (1ul << 0) 2363 2364 __u64 val; /* Return value */ 2365 }; 2366 2367 /* Known registers: 2368 * 2369 * Render engine timestamp - 0x2358 + 64bit - gen7+ 2370 * - Note this register returns an invalid value if using the default 2371 * single instruction 8byte read, in order to workaround that pass 2372 * flag I915_REG_READ_8B_WA in offset field. 2373 * 2374 */ 2375 2376 struct drm_i915_reset_stats { 2377 __u32 ctx_id; 2378 __u32 flags; 2379 2380 /* All resets since boot/module reload, for all contexts */ 2381 __u32 reset_count; 2382 2383 /* Number of batches lost when active in GPU, for this context */ 2384 __u32 batch_active; 2385 2386 /* Number of batches lost pending for execution, for this context */ 2387 __u32 batch_pending; 2388 2389 __u32 pad; 2390 }; 2391 2392 /** 2393 * struct drm_i915_gem_userptr - Create GEM object from user allocated memory. 2394 * 2395 * Userptr objects have several restrictions on what ioctls can be used with the 2396 * object handle. 2397 */ 2398 struct drm_i915_gem_userptr { 2399 /** 2400 * @user_ptr: The pointer to the allocated memory. 2401 * 2402 * Needs to be aligned to PAGE_SIZE. 2403 */ 2404 __u64 user_ptr; 2405 2406 /** 2407 * @user_size: 2408 * 2409 * The size in bytes for the allocated memory. This will also become the 2410 * object size. 2411 * 2412 * Needs to be aligned to PAGE_SIZE, and should be at least PAGE_SIZE, 2413 * or larger. 2414 */ 2415 __u64 user_size; 2416 2417 /** 2418 * @flags: 2419 * 2420 * Supported flags: 2421 * 2422 * I915_USERPTR_READ_ONLY: 2423 * 2424 * Mark the object as readonly, this also means GPU access can only be 2425 * readonly. This is only supported on HW which supports readonly access 2426 * through the GTT. If the HW can't support readonly access, an error is 2427 * returned. 2428 * 2429 * I915_USERPTR_PROBE: 2430 * 2431 * Probe the provided @user_ptr range and validate that the @user_ptr is 2432 * indeed pointing to normal memory and that the range is also valid. 2433 * For example if some garbage address is given to the kernel, then this 2434 * should complain. 2435 * 2436 * Returns -EFAULT if the probe failed. 2437 * 2438 * Note that this doesn't populate the backing pages, and also doesn't 2439 * guarantee that the object will remain valid when the object is 2440 * eventually used. 2441 * 2442 * The kernel supports this feature if I915_PARAM_HAS_USERPTR_PROBE 2443 * returns a non-zero value. 2444 * 2445 * I915_USERPTR_UNSYNCHRONIZED: 2446 * 2447 * NOT USED. Setting this flag will result in an error. 2448 */ 2449 __u32 flags; 2450 #define I915_USERPTR_READ_ONLY 0x1 2451 #define I915_USERPTR_PROBE 0x2 2452 #define I915_USERPTR_UNSYNCHRONIZED 0x80000000 2453 /** 2454 * @handle: Returned handle for the object. 2455 * 2456 * Object handles are nonzero. 2457 */ 2458 __u32 handle; 2459 }; 2460 2461 enum drm_i915_oa_format { 2462 I915_OA_FORMAT_A13 = 1, /* HSW only */ 2463 I915_OA_FORMAT_A29, /* HSW only */ 2464 I915_OA_FORMAT_A13_B8_C8, /* HSW only */ 2465 I915_OA_FORMAT_B4_C8, /* HSW only */ 2466 I915_OA_FORMAT_A45_B8_C8, /* HSW only */ 2467 I915_OA_FORMAT_B4_C8_A16, /* HSW only */ 2468 I915_OA_FORMAT_C4_B8, /* HSW+ */ 2469 2470 /* Gen8+ */ 2471 I915_OA_FORMAT_A12, 2472 I915_OA_FORMAT_A12_B8_C8, 2473 I915_OA_FORMAT_A32u40_A4u32_B8_C8, 2474 2475 I915_OA_FORMAT_MAX /* non-ABI */ 2476 }; 2477 2478 enum drm_i915_perf_property_id { 2479 /** 2480 * Open the stream for a specific context handle (as used with 2481 * execbuffer2). A stream opened for a specific context this way 2482 * won't typically require root privileges. 2483 * 2484 * This property is available in perf revision 1. 2485 */ 2486 DRM_I915_PERF_PROP_CTX_HANDLE = 1, 2487 2488 /** 2489 * A value of 1 requests the inclusion of raw OA unit reports as 2490 * part of stream samples. 2491 * 2492 * This property is available in perf revision 1. 2493 */ 2494 DRM_I915_PERF_PROP_SAMPLE_OA, 2495 2496 /** 2497 * The value specifies which set of OA unit metrics should be 2498 * configured, defining the contents of any OA unit reports. 2499 * 2500 * This property is available in perf revision 1. 2501 */ 2502 DRM_I915_PERF_PROP_OA_METRICS_SET, 2503 2504 /** 2505 * The value specifies the size and layout of OA unit reports. 2506 * 2507 * This property is available in perf revision 1. 2508 */ 2509 DRM_I915_PERF_PROP_OA_FORMAT, 2510 2511 /** 2512 * Specifying this property implicitly requests periodic OA unit 2513 * sampling and (at least on Haswell) the sampling frequency is derived 2514 * from this exponent as follows: 2515 * 2516 * 80ns * 2^(period_exponent + 1) 2517 * 2518 * This property is available in perf revision 1. 2519 */ 2520 DRM_I915_PERF_PROP_OA_EXPONENT, 2521 2522 /** 2523 * Specifying this property is only valid when specify a context to 2524 * filter with DRM_I915_PERF_PROP_CTX_HANDLE. Specifying this property 2525 * will hold preemption of the particular context we want to gather 2526 * performance data about. The execbuf2 submissions must include a 2527 * drm_i915_gem_execbuffer_ext_perf parameter for this to apply. 2528 * 2529 * This property is available in perf revision 3. 2530 */ 2531 DRM_I915_PERF_PROP_HOLD_PREEMPTION, 2532 2533 /** 2534 * Specifying this pins all contexts to the specified SSEU power 2535 * configuration for the duration of the recording. 2536 * 2537 * This parameter's value is a pointer to a struct 2538 * drm_i915_gem_context_param_sseu. 2539 * 2540 * This property is available in perf revision 4. 2541 */ 2542 DRM_I915_PERF_PROP_GLOBAL_SSEU, 2543 2544 /** 2545 * This optional parameter specifies the timer interval in nanoseconds 2546 * at which the i915 driver will check the OA buffer for available data. 2547 * Minimum allowed value is 100 microseconds. A default value is used by 2548 * the driver if this parameter is not specified. Note that larger timer 2549 * values will reduce cpu consumption during OA perf captures. However, 2550 * excessively large values would potentially result in OA buffer 2551 * overwrites as captures reach end of the OA buffer. 2552 * 2553 * This property is available in perf revision 5. 2554 */ 2555 DRM_I915_PERF_PROP_POLL_OA_PERIOD, 2556 2557 DRM_I915_PERF_PROP_MAX /* non-ABI */ 2558 }; 2559 2560 struct drm_i915_perf_open_param { 2561 __u32 flags; 2562 #define I915_PERF_FLAG_FD_CLOEXEC (1<<0) 2563 #define I915_PERF_FLAG_FD_NONBLOCK (1<<1) 2564 #define I915_PERF_FLAG_DISABLED (1<<2) 2565 2566 /** The number of u64 (id, value) pairs */ 2567 __u32 num_properties; 2568 2569 /** 2570 * Pointer to array of u64 (id, value) pairs configuring the stream 2571 * to open. 2572 */ 2573 __u64 properties_ptr; 2574 }; 2575 2576 /* 2577 * Enable data capture for a stream that was either opened in a disabled state 2578 * via I915_PERF_FLAG_DISABLED or was later disabled via 2579 * I915_PERF_IOCTL_DISABLE. 2580 * 2581 * It is intended to be cheaper to disable and enable a stream than it may be 2582 * to close and re-open a stream with the same configuration. 2583 * 2584 * It's undefined whether any pending data for the stream will be lost. 2585 * 2586 * This ioctl is available in perf revision 1. 2587 */ 2588 #define I915_PERF_IOCTL_ENABLE _IO('i', 0x0) 2589 2590 /* 2591 * Disable data capture for a stream. 2592 * 2593 * It is an error to try and read a stream that is disabled. 2594 * 2595 * This ioctl is available in perf revision 1. 2596 */ 2597 #define I915_PERF_IOCTL_DISABLE _IO('i', 0x1) 2598 2599 /* 2600 * Change metrics_set captured by a stream. 2601 * 2602 * If the stream is bound to a specific context, the configuration change 2603 * will performed inline with that context such that it takes effect before 2604 * the next execbuf submission. 2605 * 2606 * Returns the previously bound metrics set id, or a negative error code. 2607 * 2608 * This ioctl is available in perf revision 2. 2609 */ 2610 #define I915_PERF_IOCTL_CONFIG _IO('i', 0x2) 2611 2612 /* 2613 * Common to all i915 perf records 2614 */ 2615 struct drm_i915_perf_record_header { 2616 __u32 type; 2617 __u16 pad; 2618 __u16 size; 2619 }; 2620 2621 enum drm_i915_perf_record_type { 2622 2623 /** 2624 * Samples are the work horse record type whose contents are extensible 2625 * and defined when opening an i915 perf stream based on the given 2626 * properties. 2627 * 2628 * Boolean properties following the naming convention 2629 * DRM_I915_PERF_SAMPLE_xyz_PROP request the inclusion of 'xyz' data in 2630 * every sample. 2631 * 2632 * The order of these sample properties given by userspace has no 2633 * affect on the ordering of data within a sample. The order is 2634 * documented here. 2635 * 2636 * struct { 2637 * struct drm_i915_perf_record_header header; 2638 * 2639 * { u32 oa_report[]; } && DRM_I915_PERF_PROP_SAMPLE_OA 2640 * }; 2641 */ 2642 DRM_I915_PERF_RECORD_SAMPLE = 1, 2643 2644 /* 2645 * Indicates that one or more OA reports were not written by the 2646 * hardware. This can happen for example if an MI_REPORT_PERF_COUNT 2647 * command collides with periodic sampling - which would be more likely 2648 * at higher sampling frequencies. 2649 */ 2650 DRM_I915_PERF_RECORD_OA_REPORT_LOST = 2, 2651 2652 /** 2653 * An error occurred that resulted in all pending OA reports being lost. 2654 */ 2655 DRM_I915_PERF_RECORD_OA_BUFFER_LOST = 3, 2656 2657 DRM_I915_PERF_RECORD_MAX /* non-ABI */ 2658 }; 2659 2660 /* 2661 * Structure to upload perf dynamic configuration into the kernel. 2662 */ 2663 struct drm_i915_perf_oa_config { 2664 /** String formatted like "%08x-%04x-%04x-%04x-%012x" */ 2665 char uuid[36]; 2666 2667 __u32 n_mux_regs; 2668 __u32 n_boolean_regs; 2669 __u32 n_flex_regs; 2670 2671 /* 2672 * These fields are pointers to tuples of u32 values (register address, 2673 * value). For example the expected length of the buffer pointed by 2674 * mux_regs_ptr is (2 * sizeof(u32) * n_mux_regs). 2675 */ 2676 __u64 mux_regs_ptr; 2677 __u64 boolean_regs_ptr; 2678 __u64 flex_regs_ptr; 2679 }; 2680 2681 /** 2682 * struct drm_i915_query_item - An individual query for the kernel to process. 2683 * 2684 * The behaviour is determined by the @query_id. Note that exactly what 2685 * @data_ptr is also depends on the specific @query_id. 2686 */ 2687 struct drm_i915_query_item { 2688 /** @query_id: The id for this query */ 2689 __u64 query_id; 2690 #define DRM_I915_QUERY_TOPOLOGY_INFO 1 2691 #define DRM_I915_QUERY_ENGINE_INFO 2 2692 #define DRM_I915_QUERY_PERF_CONFIG 3 2693 #define DRM_I915_QUERY_MEMORY_REGIONS 4 2694 /* Must be kept compact -- no holes and well documented */ 2695 2696 /** 2697 * @length: 2698 * 2699 * When set to zero by userspace, this is filled with the size of the 2700 * data to be written at the @data_ptr pointer. The kernel sets this 2701 * value to a negative value to signal an error on a particular query 2702 * item. 2703 */ 2704 __s32 length; 2705 2706 /** 2707 * @flags: 2708 * 2709 * When query_id == DRM_I915_QUERY_TOPOLOGY_INFO, must be 0. 2710 * 2711 * When query_id == DRM_I915_QUERY_PERF_CONFIG, must be one of the 2712 * following: 2713 * 2714 * - DRM_I915_QUERY_PERF_CONFIG_LIST 2715 * - DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2716 * - DRM_I915_QUERY_PERF_CONFIG_FOR_UUID 2717 */ 2718 __u32 flags; 2719 #define DRM_I915_QUERY_PERF_CONFIG_LIST 1 2720 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2 2721 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID 3 2722 2723 /** 2724 * @data_ptr: 2725 * 2726 * Data will be written at the location pointed by @data_ptr when the 2727 * value of @length matches the length of the data to be written by the 2728 * kernel. 2729 */ 2730 __u64 data_ptr; 2731 }; 2732 2733 /** 2734 * struct drm_i915_query - Supply an array of struct drm_i915_query_item for the 2735 * kernel to fill out. 2736 * 2737 * Note that this is generally a two step process for each struct 2738 * drm_i915_query_item in the array: 2739 * 2740 * 1. Call the DRM_IOCTL_I915_QUERY, giving it our array of struct 2741 * drm_i915_query_item, with &drm_i915_query_item.length set to zero. The 2742 * kernel will then fill in the size, in bytes, which tells userspace how 2743 * memory it needs to allocate for the blob(say for an array of properties). 2744 * 2745 * 2. Next we call DRM_IOCTL_I915_QUERY again, this time with the 2746 * &drm_i915_query_item.data_ptr equal to our newly allocated blob. Note that 2747 * the &drm_i915_query_item.length should still be the same as what the 2748 * kernel previously set. At this point the kernel can fill in the blob. 2749 * 2750 * Note that for some query items it can make sense for userspace to just pass 2751 * in a buffer/blob equal to or larger than the required size. In this case only 2752 * a single ioctl call is needed. For some smaller query items this can work 2753 * quite well. 2754 * 2755 */ 2756 struct drm_i915_query { 2757 /** @num_items: The number of elements in the @items_ptr array */ 2758 __u32 num_items; 2759 2760 /** 2761 * @flags: Unused for now. Must be cleared to zero. 2762 */ 2763 __u32 flags; 2764 2765 /** 2766 * @items_ptr: 2767 * 2768 * Pointer to an array of struct drm_i915_query_item. The number of 2769 * array elements is @num_items. 2770 */ 2771 __u64 items_ptr; 2772 }; 2773 2774 /* 2775 * Data written by the kernel with query DRM_I915_QUERY_TOPOLOGY_INFO : 2776 * 2777 * data: contains the 3 pieces of information : 2778 * 2779 * - the slice mask with one bit per slice telling whether a slice is 2780 * available. The availability of slice X can be queried with the following 2781 * formula : 2782 * 2783 * (data[X / 8] >> (X % 8)) & 1 2784 * 2785 * - the subslice mask for each slice with one bit per subslice telling 2786 * whether a subslice is available. Gen12 has dual-subslices, which are 2787 * similar to two gen11 subslices. For gen12, this array represents dual- 2788 * subslices. The availability of subslice Y in slice X can be queried 2789 * with the following formula : 2790 * 2791 * (data[subslice_offset + 2792 * X * subslice_stride + 2793 * Y / 8] >> (Y % 8)) & 1 2794 * 2795 * - the EU mask for each subslice in each slice with one bit per EU telling 2796 * whether an EU is available. The availability of EU Z in subslice Y in 2797 * slice X can be queried with the following formula : 2798 * 2799 * (data[eu_offset + 2800 * (X * max_subslices + Y) * eu_stride + 2801 * Z / 8] >> (Z % 8)) & 1 2802 */ 2803 struct drm_i915_query_topology_info { 2804 /* 2805 * Unused for now. Must be cleared to zero. 2806 */ 2807 __u16 flags; 2808 2809 __u16 max_slices; 2810 __u16 max_subslices; 2811 __u16 max_eus_per_subslice; 2812 2813 /* 2814 * Offset in data[] at which the subslice masks are stored. 2815 */ 2816 __u16 subslice_offset; 2817 2818 /* 2819 * Stride at which each of the subslice masks for each slice are 2820 * stored. 2821 */ 2822 __u16 subslice_stride; 2823 2824 /* 2825 * Offset in data[] at which the EU masks are stored. 2826 */ 2827 __u16 eu_offset; 2828 2829 /* 2830 * Stride at which each of the EU masks for each subslice are stored. 2831 */ 2832 __u16 eu_stride; 2833 2834 __u8 data[]; 2835 }; 2836 2837 /** 2838 * DOC: Engine Discovery uAPI 2839 * 2840 * Engine discovery uAPI is a way of enumerating physical engines present in a 2841 * GPU associated with an open i915 DRM file descriptor. This supersedes the old 2842 * way of using `DRM_IOCTL_I915_GETPARAM` and engine identifiers like 2843 * `I915_PARAM_HAS_BLT`. 2844 * 2845 * The need for this interface came starting with Icelake and newer GPUs, which 2846 * started to establish a pattern of having multiple engines of a same class, 2847 * where not all instances were always completely functionally equivalent. 2848 * 2849 * Entry point for this uapi is `DRM_IOCTL_I915_QUERY` with the 2850 * `DRM_I915_QUERY_ENGINE_INFO` as the queried item id. 2851 * 2852 * Example for getting the list of engines: 2853 * 2854 * .. code-block:: C 2855 * 2856 * struct drm_i915_query_engine_info *info; 2857 * struct drm_i915_query_item item = { 2858 * .query_id = DRM_I915_QUERY_ENGINE_INFO; 2859 * }; 2860 * struct drm_i915_query query = { 2861 * .num_items = 1, 2862 * .items_ptr = (uintptr_t)&item, 2863 * }; 2864 * int err, i; 2865 * 2866 * // First query the size of the blob we need, this needs to be large 2867 * // enough to hold our array of engines. The kernel will fill out the 2868 * // item.length for us, which is the number of bytes we need. 2869 * // 2870 * // Alternatively a large buffer can be allocated straight away enabling 2871 * // querying in one pass, in which case item.length should contain the 2872 * // length of the provided buffer. 2873 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 2874 * if (err) ... 2875 * 2876 * info = calloc(1, item.length); 2877 * // Now that we allocated the required number of bytes, we call the ioctl 2878 * // again, this time with the data_ptr pointing to our newly allocated 2879 * // blob, which the kernel can then populate with info on all engines. 2880 * item.data_ptr = (uintptr_t)&info, 2881 * 2882 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 2883 * if (err) ... 2884 * 2885 * // We can now access each engine in the array 2886 * for (i = 0; i < info->num_engines; i++) { 2887 * struct drm_i915_engine_info einfo = info->engines[i]; 2888 * u16 class = einfo.engine.class; 2889 * u16 instance = einfo.engine.instance; 2890 * .... 2891 * } 2892 * 2893 * free(info); 2894 * 2895 * Each of the enumerated engines, apart from being defined by its class and 2896 * instance (see `struct i915_engine_class_instance`), also can have flags and 2897 * capabilities defined as documented in i915_drm.h. 2898 * 2899 * For instance video engines which support HEVC encoding will have the 2900 * `I915_VIDEO_CLASS_CAPABILITY_HEVC` capability bit set. 2901 * 2902 * Engine discovery only fully comes to its own when combined with the new way 2903 * of addressing engines when submitting batch buffers using contexts with 2904 * engine maps configured. 2905 */ 2906 2907 /** 2908 * struct drm_i915_engine_info 2909 * 2910 * Describes one engine and it's capabilities as known to the driver. 2911 */ 2912 struct drm_i915_engine_info { 2913 /** @engine: Engine class and instance. */ 2914 struct i915_engine_class_instance engine; 2915 2916 /** @rsvd0: Reserved field. */ 2917 __u32 rsvd0; 2918 2919 /** @flags: Engine flags. */ 2920 __u64 flags; 2921 #define I915_ENGINE_INFO_HAS_LOGICAL_INSTANCE (1 << 0) 2922 2923 /** @capabilities: Capabilities of this engine. */ 2924 __u64 capabilities; 2925 #define I915_VIDEO_CLASS_CAPABILITY_HEVC (1 << 0) 2926 #define I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC (1 << 1) 2927 2928 /** @logical_instance: Logical instance of engine */ 2929 __u16 logical_instance; 2930 2931 /** @rsvd1: Reserved fields. */ 2932 __u16 rsvd1[3]; 2933 /** @rsvd2: Reserved fields. */ 2934 __u64 rsvd2[3]; 2935 }; 2936 2937 /** 2938 * struct drm_i915_query_engine_info 2939 * 2940 * Engine info query enumerates all engines known to the driver by filling in 2941 * an array of struct drm_i915_engine_info structures. 2942 */ 2943 struct drm_i915_query_engine_info { 2944 /** @num_engines: Number of struct drm_i915_engine_info structs following. */ 2945 __u32 num_engines; 2946 2947 /** @rsvd: MBZ */ 2948 __u32 rsvd[3]; 2949 2950 /** @engines: Marker for drm_i915_engine_info structures. */ 2951 struct drm_i915_engine_info engines[]; 2952 }; 2953 2954 /* 2955 * Data written by the kernel with query DRM_I915_QUERY_PERF_CONFIG. 2956 */ 2957 struct drm_i915_query_perf_config { 2958 union { 2959 /* 2960 * When query_item.flags == DRM_I915_QUERY_PERF_CONFIG_LIST, i915 sets 2961 * this fields to the number of configurations available. 2962 */ 2963 __u64 n_configs; 2964 2965 /* 2966 * When query_id == DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID, 2967 * i915 will use the value in this field as configuration 2968 * identifier to decide what data to write into config_ptr. 2969 */ 2970 __u64 config; 2971 2972 /* 2973 * When query_id == DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID, 2974 * i915 will use the value in this field as configuration 2975 * identifier to decide what data to write into config_ptr. 2976 * 2977 * String formatted like "%08x-%04x-%04x-%04x-%012x" 2978 */ 2979 char uuid[36]; 2980 }; 2981 2982 /* 2983 * Unused for now. Must be cleared to zero. 2984 */ 2985 __u32 flags; 2986 2987 /* 2988 * When query_item.flags == DRM_I915_QUERY_PERF_CONFIG_LIST, i915 will 2989 * write an array of __u64 of configuration identifiers. 2990 * 2991 * When query_item.flags == DRM_I915_QUERY_PERF_CONFIG_DATA, i915 will 2992 * write a struct drm_i915_perf_oa_config. If the following fields of 2993 * drm_i915_perf_oa_config are set not set to 0, i915 will write into 2994 * the associated pointers the values of submitted when the 2995 * configuration was created : 2996 * 2997 * - n_mux_regs 2998 * - n_boolean_regs 2999 * - n_flex_regs 3000 */ 3001 __u8 data[]; 3002 }; 3003 3004 /** 3005 * enum drm_i915_gem_memory_class - Supported memory classes 3006 */ 3007 enum drm_i915_gem_memory_class { 3008 /** @I915_MEMORY_CLASS_SYSTEM: System memory */ 3009 I915_MEMORY_CLASS_SYSTEM = 0, 3010 /** @I915_MEMORY_CLASS_DEVICE: Device local-memory */ 3011 I915_MEMORY_CLASS_DEVICE, 3012 }; 3013 3014 /** 3015 * struct drm_i915_gem_memory_class_instance - Identify particular memory region 3016 */ 3017 struct drm_i915_gem_memory_class_instance { 3018 /** @memory_class: See enum drm_i915_gem_memory_class */ 3019 __u16 memory_class; 3020 3021 /** @memory_instance: Which instance */ 3022 __u16 memory_instance; 3023 }; 3024 3025 /** 3026 * struct drm_i915_memory_region_info - Describes one region as known to the 3027 * driver. 3028 * 3029 * Note that we reserve some stuff here for potential future work. As an example 3030 * we might want expose the capabilities for a given region, which could include 3031 * things like if the region is CPU mappable/accessible, what are the supported 3032 * mapping types etc. 3033 * 3034 * Note that to extend struct drm_i915_memory_region_info and struct 3035 * drm_i915_query_memory_regions in the future the plan is to do the following: 3036 * 3037 * .. code-block:: C 3038 * 3039 * struct drm_i915_memory_region_info { 3040 * struct drm_i915_gem_memory_class_instance region; 3041 * union { 3042 * __u32 rsvd0; 3043 * __u32 new_thing1; 3044 * }; 3045 * ... 3046 * union { 3047 * __u64 rsvd1[8]; 3048 * struct { 3049 * __u64 new_thing2; 3050 * __u64 new_thing3; 3051 * ... 3052 * }; 3053 * }; 3054 * }; 3055 * 3056 * With this things should remain source compatible between versions for 3057 * userspace, even as we add new fields. 3058 * 3059 * Note this is using both struct drm_i915_query_item and struct drm_i915_query. 3060 * For this new query we are adding the new query id DRM_I915_QUERY_MEMORY_REGIONS 3061 * at &drm_i915_query_item.query_id. 3062 */ 3063 struct drm_i915_memory_region_info { 3064 /** @region: The class:instance pair encoding */ 3065 struct drm_i915_gem_memory_class_instance region; 3066 3067 /** @rsvd0: MBZ */ 3068 __u32 rsvd0; 3069 3070 /** @probed_size: Memory probed by the driver (-1 = unknown) */ 3071 __u64 probed_size; 3072 3073 /** @unallocated_size: Estimate of memory remaining (-1 = unknown) */ 3074 __u64 unallocated_size; 3075 3076 /** @rsvd1: MBZ */ 3077 __u64 rsvd1[8]; 3078 }; 3079 3080 /** 3081 * struct drm_i915_query_memory_regions 3082 * 3083 * The region info query enumerates all regions known to the driver by filling 3084 * in an array of struct drm_i915_memory_region_info structures. 3085 * 3086 * Example for getting the list of supported regions: 3087 * 3088 * .. code-block:: C 3089 * 3090 * struct drm_i915_query_memory_regions *info; 3091 * struct drm_i915_query_item item = { 3092 * .query_id = DRM_I915_QUERY_MEMORY_REGIONS; 3093 * }; 3094 * struct drm_i915_query query = { 3095 * .num_items = 1, 3096 * .items_ptr = (uintptr_t)&item, 3097 * }; 3098 * int err, i; 3099 * 3100 * // First query the size of the blob we need, this needs to be large 3101 * // enough to hold our array of regions. The kernel will fill out the 3102 * // item.length for us, which is the number of bytes we need. 3103 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3104 * if (err) ... 3105 * 3106 * info = calloc(1, item.length); 3107 * // Now that we allocated the required number of bytes, we call the ioctl 3108 * // again, this time with the data_ptr pointing to our newly allocated 3109 * // blob, which the kernel can then populate with the all the region info. 3110 * item.data_ptr = (uintptr_t)&info, 3111 * 3112 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3113 * if (err) ... 3114 * 3115 * // We can now access each region in the array 3116 * for (i = 0; i < info->num_regions; i++) { 3117 * struct drm_i915_memory_region_info mr = info->regions[i]; 3118 * u16 class = mr.region.class; 3119 * u16 instance = mr.region.instance; 3120 * 3121 * .... 3122 * } 3123 * 3124 * free(info); 3125 */ 3126 struct drm_i915_query_memory_regions { 3127 /** @num_regions: Number of supported regions */ 3128 __u32 num_regions; 3129 3130 /** @rsvd: MBZ */ 3131 __u32 rsvd[3]; 3132 3133 /** @regions: Info about each supported region */ 3134 struct drm_i915_memory_region_info regions[]; 3135 }; 3136 3137 /** 3138 * struct drm_i915_gem_create_ext - Existing gem_create behaviour, with added 3139 * extension support using struct i915_user_extension. 3140 * 3141 * Note that in the future we want to have our buffer flags here, at least for 3142 * the stuff that is immutable. Previously we would have two ioctls, one to 3143 * create the object with gem_create, and another to apply various parameters, 3144 * however this creates some ambiguity for the params which are considered 3145 * immutable. Also in general we're phasing out the various SET/GET ioctls. 3146 */ 3147 struct drm_i915_gem_create_ext { 3148 /** 3149 * @size: Requested size for the object. 3150 * 3151 * The (page-aligned) allocated size for the object will be returned. 3152 * 3153 * 3154 * DG2 64K min page size implications: 3155 * 3156 * On discrete platforms, starting from DG2, we have to contend with GTT 3157 * page size restrictions when dealing with I915_MEMORY_CLASS_DEVICE 3158 * objects. Specifically the hardware only supports 64K or larger GTT 3159 * page sizes for such memory. The kernel will already ensure that all 3160 * I915_MEMORY_CLASS_DEVICE memory is allocated using 64K or larger page 3161 * sizes underneath. 3162 * 3163 * Note that the returned size here will always reflect any required 3164 * rounding up done by the kernel, i.e 4K will now become 64K on devices 3165 * such as DG2. 3166 * 3167 * Special DG2 GTT address alignment requirement: 3168 * 3169 * The GTT alignment will also need to be at least 2M for such objects. 3170 * 3171 * Note that due to how the hardware implements 64K GTT page support, we 3172 * have some further complications: 3173 * 3174 * 1) The entire PDE (which covers a 2MB virtual address range), must 3175 * contain only 64K PTEs, i.e mixing 4K and 64K PTEs in the same 3176 * PDE is forbidden by the hardware. 3177 * 3178 * 2) We still need to support 4K PTEs for I915_MEMORY_CLASS_SYSTEM 3179 * objects. 3180 * 3181 * To keep things simple for userland, we mandate that any GTT mappings 3182 * must be aligned to and rounded up to 2MB. The kernel will internally 3183 * pad them out to the next 2MB boundary. As this only wastes virtual 3184 * address space and avoids userland having to copy any needlessly 3185 * complicated PDE sharing scheme (coloring) and only affects DG2, this 3186 * is deemed to be a good compromise. 3187 */ 3188 __u64 size; 3189 /** 3190 * @handle: Returned handle for the object. 3191 * 3192 * Object handles are nonzero. 3193 */ 3194 __u32 handle; 3195 /** @flags: MBZ */ 3196 __u32 flags; 3197 /** 3198 * @extensions: The chain of extensions to apply to this object. 3199 * 3200 * This will be useful in the future when we need to support several 3201 * different extensions, and we need to apply more than one when 3202 * creating the object. See struct i915_user_extension. 3203 * 3204 * If we don't supply any extensions then we get the same old gem_create 3205 * behaviour. 3206 * 3207 * For I915_GEM_CREATE_EXT_MEMORY_REGIONS usage see 3208 * struct drm_i915_gem_create_ext_memory_regions. 3209 * 3210 * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see 3211 * struct drm_i915_gem_create_ext_protected_content. 3212 */ 3213 #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 3214 #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 3215 __u64 extensions; 3216 }; 3217 3218 /** 3219 * struct drm_i915_gem_create_ext_memory_regions - The 3220 * I915_GEM_CREATE_EXT_MEMORY_REGIONS extension. 3221 * 3222 * Set the object with the desired set of placements/regions in priority 3223 * order. Each entry must be unique and supported by the device. 3224 * 3225 * This is provided as an array of struct drm_i915_gem_memory_class_instance, or 3226 * an equivalent layout of class:instance pair encodings. See struct 3227 * drm_i915_query_memory_regions and DRM_I915_QUERY_MEMORY_REGIONS for how to 3228 * query the supported regions for a device. 3229 * 3230 * As an example, on discrete devices, if we wish to set the placement as 3231 * device local-memory we can do something like: 3232 * 3233 * .. code-block:: C 3234 * 3235 * struct drm_i915_gem_memory_class_instance region_lmem = { 3236 * .memory_class = I915_MEMORY_CLASS_DEVICE, 3237 * .memory_instance = 0, 3238 * }; 3239 * struct drm_i915_gem_create_ext_memory_regions regions = { 3240 * .base = { .name = I915_GEM_CREATE_EXT_MEMORY_REGIONS }, 3241 * .regions = (uintptr_t)®ion_lmem, 3242 * .num_regions = 1, 3243 * }; 3244 * struct drm_i915_gem_create_ext create_ext = { 3245 * .size = 16 * PAGE_SIZE, 3246 * .extensions = (uintptr_t)®ions, 3247 * }; 3248 * 3249 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3250 * if (err) ... 3251 * 3252 * At which point we get the object handle in &drm_i915_gem_create_ext.handle, 3253 * along with the final object size in &drm_i915_gem_create_ext.size, which 3254 * should account for any rounding up, if required. 3255 */ 3256 struct drm_i915_gem_create_ext_memory_regions { 3257 /** @base: Extension link. See struct i915_user_extension. */ 3258 struct i915_user_extension base; 3259 3260 /** @pad: MBZ */ 3261 __u32 pad; 3262 /** @num_regions: Number of elements in the @regions array. */ 3263 __u32 num_regions; 3264 /** 3265 * @regions: The regions/placements array. 3266 * 3267 * An array of struct drm_i915_gem_memory_class_instance. 3268 */ 3269 __u64 regions; 3270 }; 3271 3272 /** 3273 * struct drm_i915_gem_create_ext_protected_content - The 3274 * I915_OBJECT_PARAM_PROTECTED_CONTENT extension. 3275 * 3276 * If this extension is provided, buffer contents are expected to be protected 3277 * by PXP encryption and require decryption for scan out and processing. This 3278 * is only possible on platforms that have PXP enabled, on all other scenarios 3279 * using this extension will cause the ioctl to fail and return -ENODEV. The 3280 * flags parameter is reserved for future expansion and must currently be set 3281 * to zero. 3282 * 3283 * The buffer contents are considered invalid after a PXP session teardown. 3284 * 3285 * The encryption is guaranteed to be processed correctly only if the object 3286 * is submitted with a context created using the 3287 * I915_CONTEXT_PARAM_PROTECTED_CONTENT flag. This will also enable extra checks 3288 * at submission time on the validity of the objects involved. 3289 * 3290 * Below is an example on how to create a protected object: 3291 * 3292 * .. code-block:: C 3293 * 3294 * struct drm_i915_gem_create_ext_protected_content protected_ext = { 3295 * .base = { .name = I915_GEM_CREATE_EXT_PROTECTED_CONTENT }, 3296 * .flags = 0, 3297 * }; 3298 * struct drm_i915_gem_create_ext create_ext = { 3299 * .size = PAGE_SIZE, 3300 * .extensions = (uintptr_t)&protected_ext, 3301 * }; 3302 * 3303 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3304 * if (err) ... 3305 */ 3306 struct drm_i915_gem_create_ext_protected_content { 3307 /** @base: Extension link. See struct i915_user_extension. */ 3308 struct i915_user_extension base; 3309 /** @flags: reserved for future usage, currently MBZ */ 3310 __u32 flags; 3311 }; 3312 3313 /* ID of the protected content session managed by i915 when PXP is active */ 3314 #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf 3315 3316 #if defined(__cplusplus) 3317 } 3318 #endif 3319 3320 #endif /* _UAPI_I915_DRM_H_ */ 3321