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 * enum drm_i915_gem_engine_class - uapi engine type enumeration 159 * 160 * Different engines serve different roles, and there may be more than one 161 * engine serving each role. This enum provides a classification of the role 162 * of the engine, which may be used when requesting operations to be performed 163 * on a certain subset of engines, or for providing information about that 164 * group. 165 */ 166 enum drm_i915_gem_engine_class { 167 /** 168 * @I915_ENGINE_CLASS_RENDER: 169 * 170 * Render engines support instructions used for 3D, Compute (GPGPU), 171 * and programmable media workloads. These instructions fetch data and 172 * dispatch individual work items to threads that operate in parallel. 173 * The threads run small programs (called "kernels" or "shaders") on 174 * the GPU's execution units (EUs). 175 */ 176 I915_ENGINE_CLASS_RENDER = 0, 177 178 /** 179 * @I915_ENGINE_CLASS_COPY: 180 * 181 * Copy engines (also referred to as "blitters") support instructions 182 * that move blocks of data from one location in memory to another, 183 * or that fill a specified location of memory with fixed data. 184 * Copy engines can perform pre-defined logical or bitwise operations 185 * on the source, destination, or pattern data. 186 */ 187 I915_ENGINE_CLASS_COPY = 1, 188 189 /** 190 * @I915_ENGINE_CLASS_VIDEO: 191 * 192 * Video engines (also referred to as "bit stream decode" (BSD) or 193 * "vdbox") support instructions that perform fixed-function media 194 * decode and encode. 195 */ 196 I915_ENGINE_CLASS_VIDEO = 2, 197 198 /** 199 * @I915_ENGINE_CLASS_VIDEO_ENHANCE: 200 * 201 * Video enhancement engines (also referred to as "vebox") support 202 * instructions related to image enhancement. 203 */ 204 I915_ENGINE_CLASS_VIDEO_ENHANCE = 3, 205 206 /** 207 * @I915_ENGINE_CLASS_COMPUTE: 208 * 209 * Compute engines support a subset of the instructions available 210 * on render engines: compute engines support Compute (GPGPU) and 211 * programmable media workloads, but do not support the 3D pipeline. 212 */ 213 I915_ENGINE_CLASS_COMPUTE = 4, 214 215 /* Values in this enum should be kept compact. */ 216 217 /** 218 * @I915_ENGINE_CLASS_INVALID: 219 * 220 * Placeholder value to represent an invalid engine class assignment. 221 */ 222 I915_ENGINE_CLASS_INVALID = -1 223 }; 224 225 /** 226 * struct i915_engine_class_instance - Engine class/instance identifier 227 * 228 * There may be more than one engine fulfilling any role within the system. 229 * Each engine of a class is given a unique instance number and therefore 230 * any engine can be specified by its class:instance tuplet. APIs that allow 231 * access to any engine in the system will use struct i915_engine_class_instance 232 * for this identification. 233 */ 234 struct i915_engine_class_instance { 235 /** 236 * @engine_class: 237 * 238 * Engine class from enum drm_i915_gem_engine_class 239 */ 240 __u16 engine_class; 241 #define I915_ENGINE_CLASS_INVALID_NONE -1 242 #define I915_ENGINE_CLASS_INVALID_VIRTUAL -2 243 244 /** 245 * @engine_instance: 246 * 247 * Engine instance. 248 */ 249 __u16 engine_instance; 250 }; 251 252 /** 253 * DOC: perf_events exposed by i915 through /sys/bus/event_sources/drivers/i915 254 * 255 */ 256 257 enum drm_i915_pmu_engine_sample { 258 I915_SAMPLE_BUSY = 0, 259 I915_SAMPLE_WAIT = 1, 260 I915_SAMPLE_SEMA = 2 261 }; 262 263 #define I915_PMU_SAMPLE_BITS (4) 264 #define I915_PMU_SAMPLE_MASK (0xf) 265 #define I915_PMU_SAMPLE_INSTANCE_BITS (8) 266 #define I915_PMU_CLASS_SHIFT \ 267 (I915_PMU_SAMPLE_BITS + I915_PMU_SAMPLE_INSTANCE_BITS) 268 269 #define __I915_PMU_ENGINE(class, instance, sample) \ 270 ((class) << I915_PMU_CLASS_SHIFT | \ 271 (instance) << I915_PMU_SAMPLE_BITS | \ 272 (sample)) 273 274 #define I915_PMU_ENGINE_BUSY(class, instance) \ 275 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_BUSY) 276 277 #define I915_PMU_ENGINE_WAIT(class, instance) \ 278 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_WAIT) 279 280 #define I915_PMU_ENGINE_SEMA(class, instance) \ 281 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_SEMA) 282 283 #define __I915_PMU_OTHER(x) (__I915_PMU_ENGINE(0xff, 0xff, 0xf) + 1 + (x)) 284 285 #define I915_PMU_ACTUAL_FREQUENCY __I915_PMU_OTHER(0) 286 #define I915_PMU_REQUESTED_FREQUENCY __I915_PMU_OTHER(1) 287 #define I915_PMU_INTERRUPTS __I915_PMU_OTHER(2) 288 #define I915_PMU_RC6_RESIDENCY __I915_PMU_OTHER(3) 289 #define I915_PMU_SOFTWARE_GT_AWAKE_TIME __I915_PMU_OTHER(4) 290 291 #define I915_PMU_LAST /* Deprecated - do not use */ I915_PMU_RC6_RESIDENCY 292 293 /* Each region is a minimum of 16k, and there are at most 255 of them. 294 */ 295 #define I915_NR_TEX_REGIONS 255 /* table size 2k - maximum due to use 296 * of chars for next/prev indices */ 297 #define I915_LOG_MIN_TEX_REGION_SIZE 14 298 299 typedef struct _drm_i915_init { 300 enum { 301 I915_INIT_DMA = 0x01, 302 I915_CLEANUP_DMA = 0x02, 303 I915_RESUME_DMA = 0x03 304 } func; 305 unsigned int mmio_offset; 306 int sarea_priv_offset; 307 unsigned int ring_start; 308 unsigned int ring_end; 309 unsigned int ring_size; 310 unsigned int front_offset; 311 unsigned int back_offset; 312 unsigned int depth_offset; 313 unsigned int w; 314 unsigned int h; 315 unsigned int pitch; 316 unsigned int pitch_bits; 317 unsigned int back_pitch; 318 unsigned int depth_pitch; 319 unsigned int cpp; 320 unsigned int chipset; 321 } drm_i915_init_t; 322 323 typedef struct _drm_i915_sarea { 324 struct drm_tex_region texList[I915_NR_TEX_REGIONS + 1]; 325 int last_upload; /* last time texture was uploaded */ 326 int last_enqueue; /* last time a buffer was enqueued */ 327 int last_dispatch; /* age of the most recently dispatched buffer */ 328 int ctxOwner; /* last context to upload state */ 329 int texAge; 330 int pf_enabled; /* is pageflipping allowed? */ 331 int pf_active; 332 int pf_current_page; /* which buffer is being displayed? */ 333 int perf_boxes; /* performance boxes to be displayed */ 334 int width, height; /* screen size in pixels */ 335 336 drm_handle_t front_handle; 337 int front_offset; 338 int front_size; 339 340 drm_handle_t back_handle; 341 int back_offset; 342 int back_size; 343 344 drm_handle_t depth_handle; 345 int depth_offset; 346 int depth_size; 347 348 drm_handle_t tex_handle; 349 int tex_offset; 350 int tex_size; 351 int log_tex_granularity; 352 int pitch; 353 int rotation; /* 0, 90, 180 or 270 */ 354 int rotated_offset; 355 int rotated_size; 356 int rotated_pitch; 357 int virtualX, virtualY; 358 359 unsigned int front_tiled; 360 unsigned int back_tiled; 361 unsigned int depth_tiled; 362 unsigned int rotated_tiled; 363 unsigned int rotated2_tiled; 364 365 int pipeA_x; 366 int pipeA_y; 367 int pipeA_w; 368 int pipeA_h; 369 int pipeB_x; 370 int pipeB_y; 371 int pipeB_w; 372 int pipeB_h; 373 374 /* fill out some space for old userspace triple buffer */ 375 drm_handle_t unused_handle; 376 __u32 unused1, unused2, unused3; 377 378 /* buffer object handles for static buffers. May change 379 * over the lifetime of the client. 380 */ 381 __u32 front_bo_handle; 382 __u32 back_bo_handle; 383 __u32 unused_bo_handle; 384 __u32 depth_bo_handle; 385 386 } drm_i915_sarea_t; 387 388 /* due to userspace building against these headers we need some compat here */ 389 #define planeA_x pipeA_x 390 #define planeA_y pipeA_y 391 #define planeA_w pipeA_w 392 #define planeA_h pipeA_h 393 #define planeB_x pipeB_x 394 #define planeB_y pipeB_y 395 #define planeB_w pipeB_w 396 #define planeB_h pipeB_h 397 398 /* Flags for perf_boxes 399 */ 400 #define I915_BOX_RING_EMPTY 0x1 401 #define I915_BOX_FLIP 0x2 402 #define I915_BOX_WAIT 0x4 403 #define I915_BOX_TEXTURE_LOAD 0x8 404 #define I915_BOX_LOST_CONTEXT 0x10 405 406 /* 407 * i915 specific ioctls. 408 * 409 * The device specific ioctl range is [DRM_COMMAND_BASE, DRM_COMMAND_END) ie 410 * [0x40, 0xa0) (a0 is excluded). The numbers below are defined as offset 411 * against DRM_COMMAND_BASE and should be between [0x0, 0x60). 412 */ 413 #define DRM_I915_INIT 0x00 414 #define DRM_I915_FLUSH 0x01 415 #define DRM_I915_FLIP 0x02 416 #define DRM_I915_BATCHBUFFER 0x03 417 #define DRM_I915_IRQ_EMIT 0x04 418 #define DRM_I915_IRQ_WAIT 0x05 419 #define DRM_I915_GETPARAM 0x06 420 #define DRM_I915_SETPARAM 0x07 421 #define DRM_I915_ALLOC 0x08 422 #define DRM_I915_FREE 0x09 423 #define DRM_I915_INIT_HEAP 0x0a 424 #define DRM_I915_CMDBUFFER 0x0b 425 #define DRM_I915_DESTROY_HEAP 0x0c 426 #define DRM_I915_SET_VBLANK_PIPE 0x0d 427 #define DRM_I915_GET_VBLANK_PIPE 0x0e 428 #define DRM_I915_VBLANK_SWAP 0x0f 429 #define DRM_I915_HWS_ADDR 0x11 430 #define DRM_I915_GEM_INIT 0x13 431 #define DRM_I915_GEM_EXECBUFFER 0x14 432 #define DRM_I915_GEM_PIN 0x15 433 #define DRM_I915_GEM_UNPIN 0x16 434 #define DRM_I915_GEM_BUSY 0x17 435 #define DRM_I915_GEM_THROTTLE 0x18 436 #define DRM_I915_GEM_ENTERVT 0x19 437 #define DRM_I915_GEM_LEAVEVT 0x1a 438 #define DRM_I915_GEM_CREATE 0x1b 439 #define DRM_I915_GEM_PREAD 0x1c 440 #define DRM_I915_GEM_PWRITE 0x1d 441 #define DRM_I915_GEM_MMAP 0x1e 442 #define DRM_I915_GEM_SET_DOMAIN 0x1f 443 #define DRM_I915_GEM_SW_FINISH 0x20 444 #define DRM_I915_GEM_SET_TILING 0x21 445 #define DRM_I915_GEM_GET_TILING 0x22 446 #define DRM_I915_GEM_GET_APERTURE 0x23 447 #define DRM_I915_GEM_MMAP_GTT 0x24 448 #define DRM_I915_GET_PIPE_FROM_CRTC_ID 0x25 449 #define DRM_I915_GEM_MADVISE 0x26 450 #define DRM_I915_OVERLAY_PUT_IMAGE 0x27 451 #define DRM_I915_OVERLAY_ATTRS 0x28 452 #define DRM_I915_GEM_EXECBUFFER2 0x29 453 #define DRM_I915_GEM_EXECBUFFER2_WR DRM_I915_GEM_EXECBUFFER2 454 #define DRM_I915_GET_SPRITE_COLORKEY 0x2a 455 #define DRM_I915_SET_SPRITE_COLORKEY 0x2b 456 #define DRM_I915_GEM_WAIT 0x2c 457 #define DRM_I915_GEM_CONTEXT_CREATE 0x2d 458 #define DRM_I915_GEM_CONTEXT_DESTROY 0x2e 459 #define DRM_I915_GEM_SET_CACHING 0x2f 460 #define DRM_I915_GEM_GET_CACHING 0x30 461 #define DRM_I915_REG_READ 0x31 462 #define DRM_I915_GET_RESET_STATS 0x32 463 #define DRM_I915_GEM_USERPTR 0x33 464 #define DRM_I915_GEM_CONTEXT_GETPARAM 0x34 465 #define DRM_I915_GEM_CONTEXT_SETPARAM 0x35 466 #define DRM_I915_PERF_OPEN 0x36 467 #define DRM_I915_PERF_ADD_CONFIG 0x37 468 #define DRM_I915_PERF_REMOVE_CONFIG 0x38 469 #define DRM_I915_QUERY 0x39 470 #define DRM_I915_GEM_VM_CREATE 0x3a 471 #define DRM_I915_GEM_VM_DESTROY 0x3b 472 #define DRM_I915_GEM_CREATE_EXT 0x3c 473 /* Must be kept compact -- no holes */ 474 475 #define DRM_IOCTL_I915_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t) 476 #define DRM_IOCTL_I915_FLUSH DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH) 477 #define DRM_IOCTL_I915_FLIP DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLIP) 478 #define DRM_IOCTL_I915_BATCHBUFFER DRM_IOW( DRM_COMMAND_BASE + DRM_I915_BATCHBUFFER, drm_i915_batchbuffer_t) 479 #define DRM_IOCTL_I915_IRQ_EMIT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_IRQ_EMIT, drm_i915_irq_emit_t) 480 #define DRM_IOCTL_I915_IRQ_WAIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_IRQ_WAIT, drm_i915_irq_wait_t) 481 #define DRM_IOCTL_I915_GETPARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GETPARAM, drm_i915_getparam_t) 482 #define DRM_IOCTL_I915_SETPARAM DRM_IOW( DRM_COMMAND_BASE + DRM_I915_SETPARAM, drm_i915_setparam_t) 483 #define DRM_IOCTL_I915_ALLOC DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_ALLOC, drm_i915_mem_alloc_t) 484 #define DRM_IOCTL_I915_FREE DRM_IOW( DRM_COMMAND_BASE + DRM_I915_FREE, drm_i915_mem_free_t) 485 #define DRM_IOCTL_I915_INIT_HEAP DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT_HEAP, drm_i915_mem_init_heap_t) 486 #define DRM_IOCTL_I915_CMDBUFFER DRM_IOW( DRM_COMMAND_BASE + DRM_I915_CMDBUFFER, drm_i915_cmdbuffer_t) 487 #define DRM_IOCTL_I915_DESTROY_HEAP DRM_IOW( DRM_COMMAND_BASE + DRM_I915_DESTROY_HEAP, drm_i915_mem_destroy_heap_t) 488 #define DRM_IOCTL_I915_SET_VBLANK_PIPE DRM_IOW( DRM_COMMAND_BASE + DRM_I915_SET_VBLANK_PIPE, drm_i915_vblank_pipe_t) 489 #define DRM_IOCTL_I915_GET_VBLANK_PIPE DRM_IOR( DRM_COMMAND_BASE + DRM_I915_GET_VBLANK_PIPE, drm_i915_vblank_pipe_t) 490 #define DRM_IOCTL_I915_VBLANK_SWAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_VBLANK_SWAP, drm_i915_vblank_swap_t) 491 #define DRM_IOCTL_I915_HWS_ADDR DRM_IOW(DRM_COMMAND_BASE + DRM_I915_HWS_ADDR, struct drm_i915_gem_init) 492 #define DRM_IOCTL_I915_GEM_INIT DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_INIT, struct drm_i915_gem_init) 493 #define DRM_IOCTL_I915_GEM_EXECBUFFER DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER, struct drm_i915_gem_execbuffer) 494 #define DRM_IOCTL_I915_GEM_EXECBUFFER2 DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER2, struct drm_i915_gem_execbuffer2) 495 #define DRM_IOCTL_I915_GEM_EXECBUFFER2_WR DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER2_WR, struct drm_i915_gem_execbuffer2) 496 #define DRM_IOCTL_I915_GEM_PIN DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_PIN, struct drm_i915_gem_pin) 497 #define DRM_IOCTL_I915_GEM_UNPIN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_UNPIN, struct drm_i915_gem_unpin) 498 #define DRM_IOCTL_I915_GEM_BUSY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_BUSY, struct drm_i915_gem_busy) 499 #define DRM_IOCTL_I915_GEM_SET_CACHING DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_SET_CACHING, struct drm_i915_gem_caching) 500 #define DRM_IOCTL_I915_GEM_GET_CACHING DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_GET_CACHING, struct drm_i915_gem_caching) 501 #define DRM_IOCTL_I915_GEM_THROTTLE DRM_IO ( DRM_COMMAND_BASE + DRM_I915_GEM_THROTTLE) 502 #define DRM_IOCTL_I915_GEM_ENTERVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_ENTERVT) 503 #define DRM_IOCTL_I915_GEM_LEAVEVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_LEAVEVT) 504 #define DRM_IOCTL_I915_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE, struct drm_i915_gem_create) 505 #define DRM_IOCTL_I915_GEM_CREATE_EXT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE_EXT, struct drm_i915_gem_create_ext) 506 #define DRM_IOCTL_I915_GEM_PREAD DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PREAD, struct drm_i915_gem_pread) 507 #define DRM_IOCTL_I915_GEM_PWRITE DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PWRITE, struct drm_i915_gem_pwrite) 508 #define DRM_IOCTL_I915_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP, struct drm_i915_gem_mmap) 509 #define DRM_IOCTL_I915_GEM_MMAP_GTT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_gtt) 510 #define DRM_IOCTL_I915_GEM_MMAP_OFFSET DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_offset) 511 #define DRM_IOCTL_I915_GEM_SET_DOMAIN DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SET_DOMAIN, struct drm_i915_gem_set_domain) 512 #define DRM_IOCTL_I915_GEM_SW_FINISH DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SW_FINISH, struct drm_i915_gem_sw_finish) 513 #define DRM_IOCTL_I915_GEM_SET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_SET_TILING, struct drm_i915_gem_set_tiling) 514 #define DRM_IOCTL_I915_GEM_GET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_TILING, struct drm_i915_gem_get_tiling) 515 #define DRM_IOCTL_I915_GEM_GET_APERTURE DRM_IOR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_APERTURE, struct drm_i915_gem_get_aperture) 516 #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) 517 #define DRM_IOCTL_I915_GEM_MADVISE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MADVISE, struct drm_i915_gem_madvise) 518 #define DRM_IOCTL_I915_OVERLAY_PUT_IMAGE DRM_IOW(DRM_COMMAND_BASE + DRM_I915_OVERLAY_PUT_IMAGE, struct drm_intel_overlay_put_image) 519 #define DRM_IOCTL_I915_OVERLAY_ATTRS DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_OVERLAY_ATTRS, struct drm_intel_overlay_attrs) 520 #define DRM_IOCTL_I915_SET_SPRITE_COLORKEY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_SET_SPRITE_COLORKEY, struct drm_intel_sprite_colorkey) 521 #define DRM_IOCTL_I915_GET_SPRITE_COLORKEY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GET_SPRITE_COLORKEY, struct drm_intel_sprite_colorkey) 522 #define DRM_IOCTL_I915_GEM_WAIT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_WAIT, struct drm_i915_gem_wait) 523 #define DRM_IOCTL_I915_GEM_CONTEXT_CREATE DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_CREATE, struct drm_i915_gem_context_create) 524 #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) 525 #define DRM_IOCTL_I915_GEM_CONTEXT_DESTROY DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_DESTROY, struct drm_i915_gem_context_destroy) 526 #define DRM_IOCTL_I915_REG_READ DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_REG_READ, struct drm_i915_reg_read) 527 #define DRM_IOCTL_I915_GET_RESET_STATS DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GET_RESET_STATS, struct drm_i915_reset_stats) 528 #define DRM_IOCTL_I915_GEM_USERPTR DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_USERPTR, struct drm_i915_gem_userptr) 529 #define DRM_IOCTL_I915_GEM_CONTEXT_GETPARAM DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_GETPARAM, struct drm_i915_gem_context_param) 530 #define DRM_IOCTL_I915_GEM_CONTEXT_SETPARAM DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_SETPARAM, struct drm_i915_gem_context_param) 531 #define DRM_IOCTL_I915_PERF_OPEN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_OPEN, struct drm_i915_perf_open_param) 532 #define DRM_IOCTL_I915_PERF_ADD_CONFIG DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_ADD_CONFIG, struct drm_i915_perf_oa_config) 533 #define DRM_IOCTL_I915_PERF_REMOVE_CONFIG DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_REMOVE_CONFIG, __u64) 534 #define DRM_IOCTL_I915_QUERY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_QUERY, struct drm_i915_query) 535 #define DRM_IOCTL_I915_GEM_VM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_VM_CREATE, struct drm_i915_gem_vm_control) 536 #define DRM_IOCTL_I915_GEM_VM_DESTROY DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_VM_DESTROY, struct drm_i915_gem_vm_control) 537 538 /* Allow drivers to submit batchbuffers directly to hardware, relying 539 * on the security mechanisms provided by hardware. 540 */ 541 typedef struct drm_i915_batchbuffer { 542 int start; /* agp offset */ 543 int used; /* nr bytes in use */ 544 int DR1; /* hw flags for GFX_OP_DRAWRECT_INFO */ 545 int DR4; /* window origin for GFX_OP_DRAWRECT_INFO */ 546 int num_cliprects; /* mulitpass with multiple cliprects? */ 547 struct drm_clip_rect __user *cliprects; /* pointer to userspace cliprects */ 548 } drm_i915_batchbuffer_t; 549 550 /* As above, but pass a pointer to userspace buffer which can be 551 * validated by the kernel prior to sending to hardware. 552 */ 553 typedef struct _drm_i915_cmdbuffer { 554 char __user *buf; /* pointer to userspace command buffer */ 555 int sz; /* nr bytes in buf */ 556 int DR1; /* hw flags for GFX_OP_DRAWRECT_INFO */ 557 int DR4; /* window origin for GFX_OP_DRAWRECT_INFO */ 558 int num_cliprects; /* mulitpass with multiple cliprects? */ 559 struct drm_clip_rect __user *cliprects; /* pointer to userspace cliprects */ 560 } drm_i915_cmdbuffer_t; 561 562 /* Userspace can request & wait on irq's: 563 */ 564 typedef struct drm_i915_irq_emit { 565 int __user *irq_seq; 566 } drm_i915_irq_emit_t; 567 568 typedef struct drm_i915_irq_wait { 569 int irq_seq; 570 } drm_i915_irq_wait_t; 571 572 /* 573 * Different modes of per-process Graphics Translation Table, 574 * see I915_PARAM_HAS_ALIASING_PPGTT 575 */ 576 #define I915_GEM_PPGTT_NONE 0 577 #define I915_GEM_PPGTT_ALIASING 1 578 #define I915_GEM_PPGTT_FULL 2 579 580 /* Ioctl to query kernel params: 581 */ 582 #define I915_PARAM_IRQ_ACTIVE 1 583 #define I915_PARAM_ALLOW_BATCHBUFFER 2 584 #define I915_PARAM_LAST_DISPATCH 3 585 #define I915_PARAM_CHIPSET_ID 4 586 #define I915_PARAM_HAS_GEM 5 587 #define I915_PARAM_NUM_FENCES_AVAIL 6 588 #define I915_PARAM_HAS_OVERLAY 7 589 #define I915_PARAM_HAS_PAGEFLIPPING 8 590 #define I915_PARAM_HAS_EXECBUF2 9 591 #define I915_PARAM_HAS_BSD 10 592 #define I915_PARAM_HAS_BLT 11 593 #define I915_PARAM_HAS_RELAXED_FENCING 12 594 #define I915_PARAM_HAS_COHERENT_RINGS 13 595 #define I915_PARAM_HAS_EXEC_CONSTANTS 14 596 #define I915_PARAM_HAS_RELAXED_DELTA 15 597 #define I915_PARAM_HAS_GEN7_SOL_RESET 16 598 #define I915_PARAM_HAS_LLC 17 599 #define I915_PARAM_HAS_ALIASING_PPGTT 18 600 #define I915_PARAM_HAS_WAIT_TIMEOUT 19 601 #define I915_PARAM_HAS_SEMAPHORES 20 602 #define I915_PARAM_HAS_PRIME_VMAP_FLUSH 21 603 #define I915_PARAM_HAS_VEBOX 22 604 #define I915_PARAM_HAS_SECURE_BATCHES 23 605 #define I915_PARAM_HAS_PINNED_BATCHES 24 606 #define I915_PARAM_HAS_EXEC_NO_RELOC 25 607 #define I915_PARAM_HAS_EXEC_HANDLE_LUT 26 608 #define I915_PARAM_HAS_WT 27 609 #define I915_PARAM_CMD_PARSER_VERSION 28 610 #define I915_PARAM_HAS_COHERENT_PHYS_GTT 29 611 #define I915_PARAM_MMAP_VERSION 30 612 #define I915_PARAM_HAS_BSD2 31 613 #define I915_PARAM_REVISION 32 614 #define I915_PARAM_SUBSLICE_TOTAL 33 615 #define I915_PARAM_EU_TOTAL 34 616 #define I915_PARAM_HAS_GPU_RESET 35 617 #define I915_PARAM_HAS_RESOURCE_STREAMER 36 618 #define I915_PARAM_HAS_EXEC_SOFTPIN 37 619 #define I915_PARAM_HAS_POOLED_EU 38 620 #define I915_PARAM_MIN_EU_IN_POOL 39 621 #define I915_PARAM_MMAP_GTT_VERSION 40 622 623 /* 624 * Query whether DRM_I915_GEM_EXECBUFFER2 supports user defined execution 625 * priorities and the driver will attempt to execute batches in priority order. 626 * The param returns a capability bitmask, nonzero implies that the scheduler 627 * is enabled, with different features present according to the mask. 628 * 629 * The initial priority for each batch is supplied by the context and is 630 * controlled via I915_CONTEXT_PARAM_PRIORITY. 631 */ 632 #define I915_PARAM_HAS_SCHEDULER 41 633 #define I915_SCHEDULER_CAP_ENABLED (1ul << 0) 634 #define I915_SCHEDULER_CAP_PRIORITY (1ul << 1) 635 #define I915_SCHEDULER_CAP_PREEMPTION (1ul << 2) 636 #define I915_SCHEDULER_CAP_SEMAPHORES (1ul << 3) 637 #define I915_SCHEDULER_CAP_ENGINE_BUSY_STATS (1ul << 4) 638 /* 639 * Indicates the 2k user priority levels are statically mapped into 3 buckets as 640 * follows: 641 * 642 * -1k to -1 Low priority 643 * 0 Normal priority 644 * 1 to 1k Highest priority 645 */ 646 #define I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP (1ul << 5) 647 648 #define I915_PARAM_HUC_STATUS 42 649 650 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to opt-out of 651 * synchronisation with implicit fencing on individual objects. 652 * See EXEC_OBJECT_ASYNC. 653 */ 654 #define I915_PARAM_HAS_EXEC_ASYNC 43 655 656 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports explicit fence support - 657 * both being able to pass in a sync_file fd to wait upon before executing, 658 * and being able to return a new sync_file fd that is signaled when the 659 * current request is complete. See I915_EXEC_FENCE_IN and I915_EXEC_FENCE_OUT. 660 */ 661 #define I915_PARAM_HAS_EXEC_FENCE 44 662 663 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to capture 664 * user specified bufffers for post-mortem debugging of GPU hangs. See 665 * EXEC_OBJECT_CAPTURE. 666 */ 667 #define I915_PARAM_HAS_EXEC_CAPTURE 45 668 669 #define I915_PARAM_SLICE_MASK 46 670 671 /* Assuming it's uniform for each slice, this queries the mask of subslices 672 * per-slice for this system. 673 */ 674 #define I915_PARAM_SUBSLICE_MASK 47 675 676 /* 677 * Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying the batch buffer 678 * as the first execobject as opposed to the last. See I915_EXEC_BATCH_FIRST. 679 */ 680 #define I915_PARAM_HAS_EXEC_BATCH_FIRST 48 681 682 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 683 * drm_i915_gem_exec_fence structures. See I915_EXEC_FENCE_ARRAY. 684 */ 685 #define I915_PARAM_HAS_EXEC_FENCE_ARRAY 49 686 687 /* 688 * Query whether every context (both per-file default and user created) is 689 * isolated (insofar as HW supports). If this parameter is not true, then 690 * freshly created contexts may inherit values from an existing context, 691 * rather than default HW values. If true, it also ensures (insofar as HW 692 * supports) that all state set by this context will not leak to any other 693 * context. 694 * 695 * As not every engine across every gen support contexts, the returned 696 * value reports the support of context isolation for individual engines by 697 * returning a bitmask of each engine class set to true if that class supports 698 * isolation. 699 */ 700 #define I915_PARAM_HAS_CONTEXT_ISOLATION 50 701 702 /* Frequency of the command streamer timestamps given by the *_TIMESTAMP 703 * registers. This used to be fixed per platform but from CNL onwards, this 704 * might vary depending on the parts. 705 */ 706 #define I915_PARAM_CS_TIMESTAMP_FREQUENCY 51 707 708 /* 709 * Once upon a time we supposed that writes through the GGTT would be 710 * immediately in physical memory (once flushed out of the CPU path). However, 711 * on a few different processors and chipsets, this is not necessarily the case 712 * as the writes appear to be buffered internally. Thus a read of the backing 713 * storage (physical memory) via a different path (with different physical tags 714 * to the indirect write via the GGTT) will see stale values from before 715 * the GGTT write. Inside the kernel, we can for the most part keep track of 716 * the different read/write domains in use (e.g. set-domain), but the assumption 717 * of coherency is baked into the ABI, hence reporting its true state in this 718 * parameter. 719 * 720 * Reports true when writes via mmap_gtt are immediately visible following an 721 * lfence to flush the WCB. 722 * 723 * Reports false when writes via mmap_gtt are indeterminately delayed in an in 724 * internal buffer and are _not_ immediately visible to third parties accessing 725 * directly via mmap_cpu/mmap_wc. Use of mmap_gtt as part of an IPC 726 * communications channel when reporting false is strongly disadvised. 727 */ 728 #define I915_PARAM_MMAP_GTT_COHERENT 52 729 730 /* 731 * Query whether DRM_I915_GEM_EXECBUFFER2 supports coordination of parallel 732 * execution through use of explicit fence support. 733 * See I915_EXEC_FENCE_OUT and I915_EXEC_FENCE_SUBMIT. 734 */ 735 #define I915_PARAM_HAS_EXEC_SUBMIT_FENCE 53 736 737 /* 738 * Revision of the i915-perf uAPI. The value returned helps determine what 739 * i915-perf features are available. See drm_i915_perf_property_id. 740 */ 741 #define I915_PARAM_PERF_REVISION 54 742 743 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 744 * timeline syncobj through drm_i915_gem_execbuffer_ext_timeline_fences. See 745 * I915_EXEC_USE_EXTENSIONS. 746 */ 747 #define I915_PARAM_HAS_EXEC_TIMELINE_FENCES 55 748 749 /* Query if the kernel supports the I915_USERPTR_PROBE flag. */ 750 #define I915_PARAM_HAS_USERPTR_PROBE 56 751 752 /* Must be kept compact -- no holes and well documented */ 753 754 typedef struct drm_i915_getparam { 755 __s32 param; 756 /* 757 * WARNING: Using pointers instead of fixed-size u64 means we need to write 758 * compat32 code. Don't repeat this mistake. 759 */ 760 int __user *value; 761 } drm_i915_getparam_t; 762 763 /* Ioctl to set kernel params: 764 */ 765 #define I915_SETPARAM_USE_MI_BATCHBUFFER_START 1 766 #define I915_SETPARAM_TEX_LRU_LOG_GRANULARITY 2 767 #define I915_SETPARAM_ALLOW_BATCHBUFFER 3 768 #define I915_SETPARAM_NUM_USED_FENCES 4 769 /* Must be kept compact -- no holes */ 770 771 typedef struct drm_i915_setparam { 772 int param; 773 int value; 774 } drm_i915_setparam_t; 775 776 /* A memory manager for regions of shared memory: 777 */ 778 #define I915_MEM_REGION_AGP 1 779 780 typedef struct drm_i915_mem_alloc { 781 int region; 782 int alignment; 783 int size; 784 int __user *region_offset; /* offset from start of fb or agp */ 785 } drm_i915_mem_alloc_t; 786 787 typedef struct drm_i915_mem_free { 788 int region; 789 int region_offset; 790 } drm_i915_mem_free_t; 791 792 typedef struct drm_i915_mem_init_heap { 793 int region; 794 int size; 795 int start; 796 } drm_i915_mem_init_heap_t; 797 798 /* Allow memory manager to be torn down and re-initialized (eg on 799 * rotate): 800 */ 801 typedef struct drm_i915_mem_destroy_heap { 802 int region; 803 } drm_i915_mem_destroy_heap_t; 804 805 /* Allow X server to configure which pipes to monitor for vblank signals 806 */ 807 #define DRM_I915_VBLANK_PIPE_A 1 808 #define DRM_I915_VBLANK_PIPE_B 2 809 810 typedef struct drm_i915_vblank_pipe { 811 int pipe; 812 } drm_i915_vblank_pipe_t; 813 814 /* Schedule buffer swap at given vertical blank: 815 */ 816 typedef struct drm_i915_vblank_swap { 817 drm_drawable_t drawable; 818 enum drm_vblank_seq_type seqtype; 819 unsigned int sequence; 820 } drm_i915_vblank_swap_t; 821 822 typedef struct drm_i915_hws_addr { 823 __u64 addr; 824 } drm_i915_hws_addr_t; 825 826 struct drm_i915_gem_init { 827 /** 828 * Beginning offset in the GTT to be managed by the DRM memory 829 * manager. 830 */ 831 __u64 gtt_start; 832 /** 833 * Ending offset in the GTT to be managed by the DRM memory 834 * manager. 835 */ 836 __u64 gtt_end; 837 }; 838 839 struct drm_i915_gem_create { 840 /** 841 * Requested size for the object. 842 * 843 * The (page-aligned) allocated size for the object will be returned. 844 */ 845 __u64 size; 846 /** 847 * Returned handle for the object. 848 * 849 * Object handles are nonzero. 850 */ 851 __u32 handle; 852 __u32 pad; 853 }; 854 855 struct drm_i915_gem_pread { 856 /** Handle for the object being read. */ 857 __u32 handle; 858 __u32 pad; 859 /** Offset into the object to read from */ 860 __u64 offset; 861 /** Length of data to read */ 862 __u64 size; 863 /** 864 * Pointer to write the data into. 865 * 866 * This is a fixed-size type for 32/64 compatibility. 867 */ 868 __u64 data_ptr; 869 }; 870 871 struct drm_i915_gem_pwrite { 872 /** Handle for the object being written to. */ 873 __u32 handle; 874 __u32 pad; 875 /** Offset into the object to write to */ 876 __u64 offset; 877 /** Length of data to write */ 878 __u64 size; 879 /** 880 * Pointer to read the data from. 881 * 882 * This is a fixed-size type for 32/64 compatibility. 883 */ 884 __u64 data_ptr; 885 }; 886 887 struct drm_i915_gem_mmap { 888 /** Handle for the object being mapped. */ 889 __u32 handle; 890 __u32 pad; 891 /** Offset in the object to map. */ 892 __u64 offset; 893 /** 894 * Length of data to map. 895 * 896 * The value will be page-aligned. 897 */ 898 __u64 size; 899 /** 900 * Returned pointer the data was mapped at. 901 * 902 * This is a fixed-size type for 32/64 compatibility. 903 */ 904 __u64 addr_ptr; 905 906 /** 907 * Flags for extended behaviour. 908 * 909 * Added in version 2. 910 */ 911 __u64 flags; 912 #define I915_MMAP_WC 0x1 913 }; 914 915 struct drm_i915_gem_mmap_gtt { 916 /** Handle for the object being mapped. */ 917 __u32 handle; 918 __u32 pad; 919 /** 920 * Fake offset to use for subsequent mmap call 921 * 922 * This is a fixed-size type for 32/64 compatibility. 923 */ 924 __u64 offset; 925 }; 926 927 /** 928 * struct drm_i915_gem_mmap_offset - Retrieve an offset so we can mmap this buffer object. 929 * 930 * This struct is passed as argument to the `DRM_IOCTL_I915_GEM_MMAP_OFFSET` ioctl, 931 * and is used to retrieve the fake offset to mmap an object specified by &handle. 932 * 933 * The legacy way of using `DRM_IOCTL_I915_GEM_MMAP` is removed on gen12+. 934 * `DRM_IOCTL_I915_GEM_MMAP_GTT` is an older supported alias to this struct, but will behave 935 * as setting the &extensions to 0, and &flags to `I915_MMAP_OFFSET_GTT`. 936 */ 937 struct drm_i915_gem_mmap_offset { 938 /** @handle: Handle for the object being mapped. */ 939 __u32 handle; 940 /** @pad: Must be zero */ 941 __u32 pad; 942 /** 943 * @offset: The fake offset to use for subsequent mmap call 944 * 945 * This is a fixed-size type for 32/64 compatibility. 946 */ 947 __u64 offset; 948 949 /** 950 * @flags: Flags for extended behaviour. 951 * 952 * It is mandatory that one of the `MMAP_OFFSET` types 953 * should be included: 954 * 955 * - `I915_MMAP_OFFSET_GTT`: Use mmap with the object bound to GTT. (Write-Combined) 956 * - `I915_MMAP_OFFSET_WC`: Use Write-Combined caching. 957 * - `I915_MMAP_OFFSET_WB`: Use Write-Back caching. 958 * - `I915_MMAP_OFFSET_FIXED`: Use object placement to determine caching. 959 * 960 * On devices with local memory `I915_MMAP_OFFSET_FIXED` is the only valid 961 * type. On devices without local memory, this caching mode is invalid. 962 * 963 * As caching mode when specifying `I915_MMAP_OFFSET_FIXED`, WC or WB will 964 * be used, depending on the object placement on creation. WB will be used 965 * when the object can only exist in system memory, WC otherwise. 966 */ 967 __u64 flags; 968 969 #define I915_MMAP_OFFSET_GTT 0 970 #define I915_MMAP_OFFSET_WC 1 971 #define I915_MMAP_OFFSET_WB 2 972 #define I915_MMAP_OFFSET_UC 3 973 #define I915_MMAP_OFFSET_FIXED 4 974 975 /** 976 * @extensions: Zero-terminated chain of extensions. 977 * 978 * No current extensions defined; mbz. 979 */ 980 __u64 extensions; 981 }; 982 983 /** 984 * struct drm_i915_gem_set_domain - Adjust the objects write or read domain, in 985 * preparation for accessing the pages via some CPU domain. 986 * 987 * Specifying a new write or read domain will flush the object out of the 988 * previous domain(if required), before then updating the objects domain 989 * tracking with the new domain. 990 * 991 * Note this might involve waiting for the object first if it is still active on 992 * the GPU. 993 * 994 * Supported values for @read_domains and @write_domain: 995 * 996 * - I915_GEM_DOMAIN_WC: Uncached write-combined domain 997 * - I915_GEM_DOMAIN_CPU: CPU cache domain 998 * - I915_GEM_DOMAIN_GTT: Mappable aperture domain 999 * 1000 * All other domains are rejected. 1001 * 1002 * Note that for discrete, starting from DG1, this is no longer supported, and 1003 * is instead rejected. On such platforms the CPU domain is effectively static, 1004 * where we also only support a single &drm_i915_gem_mmap_offset cache mode, 1005 * which can't be set explicitly and instead depends on the object placements, 1006 * as per the below. 1007 * 1008 * Implicit caching rules, starting from DG1: 1009 * 1010 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1011 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1012 * mapped as write-combined only. 1013 * 1014 * - Everything else is always allocated and mapped as write-back, with the 1015 * guarantee that everything is also coherent with the GPU. 1016 * 1017 * Note that this is likely to change in the future again, where we might need 1018 * more flexibility on future devices, so making this all explicit as part of a 1019 * new &drm_i915_gem_create_ext extension is probable. 1020 */ 1021 struct drm_i915_gem_set_domain { 1022 /** @handle: Handle for the object. */ 1023 __u32 handle; 1024 1025 /** @read_domains: New read domains. */ 1026 __u32 read_domains; 1027 1028 /** 1029 * @write_domain: New write domain. 1030 * 1031 * Note that having something in the write domain implies it's in the 1032 * read domain, and only that read domain. 1033 */ 1034 __u32 write_domain; 1035 }; 1036 1037 struct drm_i915_gem_sw_finish { 1038 /** Handle for the object */ 1039 __u32 handle; 1040 }; 1041 1042 struct drm_i915_gem_relocation_entry { 1043 /** 1044 * Handle of the buffer being pointed to by this relocation entry. 1045 * 1046 * It's appealing to make this be an index into the mm_validate_entry 1047 * list to refer to the buffer, but this allows the driver to create 1048 * a relocation list for state buffers and not re-write it per 1049 * exec using the buffer. 1050 */ 1051 __u32 target_handle; 1052 1053 /** 1054 * Value to be added to the offset of the target buffer to make up 1055 * the relocation entry. 1056 */ 1057 __u32 delta; 1058 1059 /** Offset in the buffer the relocation entry will be written into */ 1060 __u64 offset; 1061 1062 /** 1063 * Offset value of the target buffer that the relocation entry was last 1064 * written as. 1065 * 1066 * If the buffer has the same offset as last time, we can skip syncing 1067 * and writing the relocation. This value is written back out by 1068 * the execbuffer ioctl when the relocation is written. 1069 */ 1070 __u64 presumed_offset; 1071 1072 /** 1073 * Target memory domains read by this operation. 1074 */ 1075 __u32 read_domains; 1076 1077 /** 1078 * Target memory domains written by this operation. 1079 * 1080 * Note that only one domain may be written by the whole 1081 * execbuffer operation, so that where there are conflicts, 1082 * the application will get -EINVAL back. 1083 */ 1084 __u32 write_domain; 1085 }; 1086 1087 /** @{ 1088 * Intel memory domains 1089 * 1090 * Most of these just align with the various caches in 1091 * the system and are used to flush and invalidate as 1092 * objects end up cached in different domains. 1093 */ 1094 /** CPU cache */ 1095 #define I915_GEM_DOMAIN_CPU 0x00000001 1096 /** Render cache, used by 2D and 3D drawing */ 1097 #define I915_GEM_DOMAIN_RENDER 0x00000002 1098 /** Sampler cache, used by texture engine */ 1099 #define I915_GEM_DOMAIN_SAMPLER 0x00000004 1100 /** Command queue, used to load batch buffers */ 1101 #define I915_GEM_DOMAIN_COMMAND 0x00000008 1102 /** Instruction cache, used by shader programs */ 1103 #define I915_GEM_DOMAIN_INSTRUCTION 0x00000010 1104 /** Vertex address cache */ 1105 #define I915_GEM_DOMAIN_VERTEX 0x00000020 1106 /** GTT domain - aperture and scanout */ 1107 #define I915_GEM_DOMAIN_GTT 0x00000040 1108 /** WC domain - uncached access */ 1109 #define I915_GEM_DOMAIN_WC 0x00000080 1110 /** @} */ 1111 1112 struct drm_i915_gem_exec_object { 1113 /** 1114 * User's handle for a buffer to be bound into the GTT for this 1115 * operation. 1116 */ 1117 __u32 handle; 1118 1119 /** Number of relocations to be performed on this buffer */ 1120 __u32 relocation_count; 1121 /** 1122 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1123 * the relocations to be performed in this buffer. 1124 */ 1125 __u64 relocs_ptr; 1126 1127 /** Required alignment in graphics aperture */ 1128 __u64 alignment; 1129 1130 /** 1131 * Returned value of the updated offset of the object, for future 1132 * presumed_offset writes. 1133 */ 1134 __u64 offset; 1135 }; 1136 1137 /* DRM_IOCTL_I915_GEM_EXECBUFFER was removed in Linux 5.13 */ 1138 struct drm_i915_gem_execbuffer { 1139 /** 1140 * List of buffers to be validated with their relocations to be 1141 * performend on them. 1142 * 1143 * This is a pointer to an array of struct drm_i915_gem_validate_entry. 1144 * 1145 * These buffers must be listed in an order such that all relocations 1146 * a buffer is performing refer to buffers that have already appeared 1147 * in the validate list. 1148 */ 1149 __u64 buffers_ptr; 1150 __u32 buffer_count; 1151 1152 /** Offset in the batchbuffer to start execution from. */ 1153 __u32 batch_start_offset; 1154 /** Bytes used in batchbuffer from batch_start_offset */ 1155 __u32 batch_len; 1156 __u32 DR1; 1157 __u32 DR4; 1158 __u32 num_cliprects; 1159 /** This is a struct drm_clip_rect *cliprects */ 1160 __u64 cliprects_ptr; 1161 }; 1162 1163 struct drm_i915_gem_exec_object2 { 1164 /** 1165 * User's handle for a buffer to be bound into the GTT for this 1166 * operation. 1167 */ 1168 __u32 handle; 1169 1170 /** Number of relocations to be performed on this buffer */ 1171 __u32 relocation_count; 1172 /** 1173 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1174 * the relocations to be performed in this buffer. 1175 */ 1176 __u64 relocs_ptr; 1177 1178 /** Required alignment in graphics aperture */ 1179 __u64 alignment; 1180 1181 /** 1182 * When the EXEC_OBJECT_PINNED flag is specified this is populated by 1183 * the user with the GTT offset at which this object will be pinned. 1184 * 1185 * When the I915_EXEC_NO_RELOC flag is specified this must contain the 1186 * presumed_offset of the object. 1187 * 1188 * During execbuffer2 the kernel populates it with the value of the 1189 * current GTT offset of the object, for future presumed_offset writes. 1190 * 1191 * See struct drm_i915_gem_create_ext for the rules when dealing with 1192 * alignment restrictions with I915_MEMORY_CLASS_DEVICE, on devices with 1193 * minimum page sizes, like DG2. 1194 */ 1195 __u64 offset; 1196 1197 #define EXEC_OBJECT_NEEDS_FENCE (1<<0) 1198 #define EXEC_OBJECT_NEEDS_GTT (1<<1) 1199 #define EXEC_OBJECT_WRITE (1<<2) 1200 #define EXEC_OBJECT_SUPPORTS_48B_ADDRESS (1<<3) 1201 #define EXEC_OBJECT_PINNED (1<<4) 1202 #define EXEC_OBJECT_PAD_TO_SIZE (1<<5) 1203 /* The kernel implicitly tracks GPU activity on all GEM objects, and 1204 * synchronises operations with outstanding rendering. This includes 1205 * rendering on other devices if exported via dma-buf. However, sometimes 1206 * this tracking is too coarse and the user knows better. For example, 1207 * if the object is split into non-overlapping ranges shared between different 1208 * clients or engines (i.e. suballocating objects), the implicit tracking 1209 * by kernel assumes that each operation affects the whole object rather 1210 * than an individual range, causing needless synchronisation between clients. 1211 * The kernel will also forgo any CPU cache flushes prior to rendering from 1212 * the object as the client is expected to be also handling such domain 1213 * tracking. 1214 * 1215 * The kernel maintains the implicit tracking in order to manage resources 1216 * used by the GPU - this flag only disables the synchronisation prior to 1217 * rendering with this object in this execbuf. 1218 * 1219 * Opting out of implicit synhronisation requires the user to do its own 1220 * explicit tracking to avoid rendering corruption. See, for example, 1221 * I915_PARAM_HAS_EXEC_FENCE to order execbufs and execute them asynchronously. 1222 */ 1223 #define EXEC_OBJECT_ASYNC (1<<6) 1224 /* Request that the contents of this execobject be copied into the error 1225 * state upon a GPU hang involving this batch for post-mortem debugging. 1226 * These buffers are recorded in no particular order as "user" in 1227 * /sys/class/drm/cardN/error. Query I915_PARAM_HAS_EXEC_CAPTURE to see 1228 * if the kernel supports this flag. 1229 */ 1230 #define EXEC_OBJECT_CAPTURE (1<<7) 1231 /* All remaining bits are MBZ and RESERVED FOR FUTURE USE */ 1232 #define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_CAPTURE<<1) 1233 __u64 flags; 1234 1235 union { 1236 __u64 rsvd1; 1237 __u64 pad_to_size; 1238 }; 1239 __u64 rsvd2; 1240 }; 1241 1242 struct drm_i915_gem_exec_fence { 1243 /** 1244 * User's handle for a drm_syncobj to wait on or signal. 1245 */ 1246 __u32 handle; 1247 1248 #define I915_EXEC_FENCE_WAIT (1<<0) 1249 #define I915_EXEC_FENCE_SIGNAL (1<<1) 1250 #define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1)) 1251 __u32 flags; 1252 }; 1253 1254 /* 1255 * See drm_i915_gem_execbuffer_ext_timeline_fences. 1256 */ 1257 #define DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES 0 1258 1259 /* 1260 * This structure describes an array of drm_syncobj and associated points for 1261 * timeline variants of drm_syncobj. It is invalid to append this structure to 1262 * the execbuf if I915_EXEC_FENCE_ARRAY is set. 1263 */ 1264 struct drm_i915_gem_execbuffer_ext_timeline_fences { 1265 struct i915_user_extension base; 1266 1267 /** 1268 * Number of element in the handles_ptr & value_ptr arrays. 1269 */ 1270 __u64 fence_count; 1271 1272 /** 1273 * Pointer to an array of struct drm_i915_gem_exec_fence of length 1274 * fence_count. 1275 */ 1276 __u64 handles_ptr; 1277 1278 /** 1279 * Pointer to an array of u64 values of length fence_count. Values 1280 * must be 0 for a binary drm_syncobj. A Value of 0 for a timeline 1281 * drm_syncobj is invalid as it turns a drm_syncobj into a binary one. 1282 */ 1283 __u64 values_ptr; 1284 }; 1285 1286 struct drm_i915_gem_execbuffer2 { 1287 /** 1288 * List of gem_exec_object2 structs 1289 */ 1290 __u64 buffers_ptr; 1291 __u32 buffer_count; 1292 1293 /** Offset in the batchbuffer to start execution from. */ 1294 __u32 batch_start_offset; 1295 /** Bytes used in batchbuffer from batch_start_offset */ 1296 __u32 batch_len; 1297 __u32 DR1; 1298 __u32 DR4; 1299 __u32 num_cliprects; 1300 /** 1301 * This is a struct drm_clip_rect *cliprects if I915_EXEC_FENCE_ARRAY 1302 * & I915_EXEC_USE_EXTENSIONS are not set. 1303 * 1304 * If I915_EXEC_FENCE_ARRAY is set, then this is a pointer to an array 1305 * of struct drm_i915_gem_exec_fence and num_cliprects is the length 1306 * of the array. 1307 * 1308 * If I915_EXEC_USE_EXTENSIONS is set, then this is a pointer to a 1309 * single struct i915_user_extension and num_cliprects is 0. 1310 */ 1311 __u64 cliprects_ptr; 1312 #define I915_EXEC_RING_MASK (0x3f) 1313 #define I915_EXEC_DEFAULT (0<<0) 1314 #define I915_EXEC_RENDER (1<<0) 1315 #define I915_EXEC_BSD (2<<0) 1316 #define I915_EXEC_BLT (3<<0) 1317 #define I915_EXEC_VEBOX (4<<0) 1318 1319 /* Used for switching the constants addressing mode on gen4+ RENDER ring. 1320 * Gen6+ only supports relative addressing to dynamic state (default) and 1321 * absolute addressing. 1322 * 1323 * These flags are ignored for the BSD and BLT rings. 1324 */ 1325 #define I915_EXEC_CONSTANTS_MASK (3<<6) 1326 #define I915_EXEC_CONSTANTS_REL_GENERAL (0<<6) /* default */ 1327 #define I915_EXEC_CONSTANTS_ABSOLUTE (1<<6) 1328 #define I915_EXEC_CONSTANTS_REL_SURFACE (2<<6) /* gen4/5 only */ 1329 __u64 flags; 1330 __u64 rsvd1; /* now used for context info */ 1331 __u64 rsvd2; 1332 }; 1333 1334 /** Resets the SO write offset registers for transform feedback on gen7. */ 1335 #define I915_EXEC_GEN7_SOL_RESET (1<<8) 1336 1337 /** Request a privileged ("secure") batch buffer. Note only available for 1338 * DRM_ROOT_ONLY | DRM_MASTER processes. 1339 */ 1340 #define I915_EXEC_SECURE (1<<9) 1341 1342 /** Inform the kernel that the batch is and will always be pinned. This 1343 * negates the requirement for a workaround to be performed to avoid 1344 * an incoherent CS (such as can be found on 830/845). If this flag is 1345 * not passed, the kernel will endeavour to make sure the batch is 1346 * coherent with the CS before execution. If this flag is passed, 1347 * userspace assumes the responsibility for ensuring the same. 1348 */ 1349 #define I915_EXEC_IS_PINNED (1<<10) 1350 1351 /** Provide a hint to the kernel that the command stream and auxiliary 1352 * state buffers already holds the correct presumed addresses and so the 1353 * relocation process may be skipped if no buffers need to be moved in 1354 * preparation for the execbuffer. 1355 */ 1356 #define I915_EXEC_NO_RELOC (1<<11) 1357 1358 /** Use the reloc.handle as an index into the exec object array rather 1359 * than as the per-file handle. 1360 */ 1361 #define I915_EXEC_HANDLE_LUT (1<<12) 1362 1363 /** Used for switching BSD rings on the platforms with two BSD rings */ 1364 #define I915_EXEC_BSD_SHIFT (13) 1365 #define I915_EXEC_BSD_MASK (3 << I915_EXEC_BSD_SHIFT) 1366 /* default ping-pong mode */ 1367 #define I915_EXEC_BSD_DEFAULT (0 << I915_EXEC_BSD_SHIFT) 1368 #define I915_EXEC_BSD_RING1 (1 << I915_EXEC_BSD_SHIFT) 1369 #define I915_EXEC_BSD_RING2 (2 << I915_EXEC_BSD_SHIFT) 1370 1371 /** Tell the kernel that the batchbuffer is processed by 1372 * the resource streamer. 1373 */ 1374 #define I915_EXEC_RESOURCE_STREAMER (1<<15) 1375 1376 /* Setting I915_EXEC_FENCE_IN implies that lower_32_bits(rsvd2) represent 1377 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1378 * the batch. 1379 * 1380 * Returns -EINVAL if the sync_file fd cannot be found. 1381 */ 1382 #define I915_EXEC_FENCE_IN (1<<16) 1383 1384 /* Setting I915_EXEC_FENCE_OUT causes the ioctl to return a sync_file fd 1385 * in the upper_32_bits(rsvd2) upon success. Ownership of the fd is given 1386 * to the caller, and it should be close() after use. (The fd is a regular 1387 * file descriptor and will be cleaned up on process termination. It holds 1388 * a reference to the request, but nothing else.) 1389 * 1390 * The sync_file fd can be combined with other sync_file and passed either 1391 * to execbuf using I915_EXEC_FENCE_IN, to atomic KMS ioctls (so that a flip 1392 * will only occur after this request completes), or to other devices. 1393 * 1394 * Using I915_EXEC_FENCE_OUT requires use of 1395 * DRM_IOCTL_I915_GEM_EXECBUFFER2_WR ioctl so that the result is written 1396 * back to userspace. Failure to do so will cause the out-fence to always 1397 * be reported as zero, and the real fence fd to be leaked. 1398 */ 1399 #define I915_EXEC_FENCE_OUT (1<<17) 1400 1401 /* 1402 * Traditionally the execbuf ioctl has only considered the final element in 1403 * the execobject[] to be the executable batch. Often though, the client 1404 * will known the batch object prior to construction and being able to place 1405 * it into the execobject[] array first can simplify the relocation tracking. 1406 * Setting I915_EXEC_BATCH_FIRST tells execbuf to use element 0 of the 1407 * execobject[] as the * batch instead (the default is to use the last 1408 * element). 1409 */ 1410 #define I915_EXEC_BATCH_FIRST (1<<18) 1411 1412 /* Setting I915_FENCE_ARRAY implies that num_cliprects and cliprects_ptr 1413 * define an array of i915_gem_exec_fence structures which specify a set of 1414 * dma fences to wait upon or signal. 1415 */ 1416 #define I915_EXEC_FENCE_ARRAY (1<<19) 1417 1418 /* 1419 * Setting I915_EXEC_FENCE_SUBMIT implies that lower_32_bits(rsvd2) represent 1420 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1421 * the batch. 1422 * 1423 * Returns -EINVAL if the sync_file fd cannot be found. 1424 */ 1425 #define I915_EXEC_FENCE_SUBMIT (1 << 20) 1426 1427 /* 1428 * Setting I915_EXEC_USE_EXTENSIONS implies that 1429 * drm_i915_gem_execbuffer2.cliprects_ptr is treated as a pointer to an linked 1430 * list of i915_user_extension. Each i915_user_extension node is the base of a 1431 * larger structure. The list of supported structures are listed in the 1432 * drm_i915_gem_execbuffer_ext enum. 1433 */ 1434 #define I915_EXEC_USE_EXTENSIONS (1 << 21) 1435 1436 #define __I915_EXEC_UNKNOWN_FLAGS (-(I915_EXEC_USE_EXTENSIONS << 1)) 1437 1438 #define I915_EXEC_CONTEXT_ID_MASK (0xffffffff) 1439 #define i915_execbuffer2_set_context_id(eb2, context) \ 1440 (eb2).rsvd1 = context & I915_EXEC_CONTEXT_ID_MASK 1441 #define i915_execbuffer2_get_context_id(eb2) \ 1442 ((eb2).rsvd1 & I915_EXEC_CONTEXT_ID_MASK) 1443 1444 struct drm_i915_gem_pin { 1445 /** Handle of the buffer to be pinned. */ 1446 __u32 handle; 1447 __u32 pad; 1448 1449 /** alignment required within the aperture */ 1450 __u64 alignment; 1451 1452 /** Returned GTT offset of the buffer. */ 1453 __u64 offset; 1454 }; 1455 1456 struct drm_i915_gem_unpin { 1457 /** Handle of the buffer to be unpinned. */ 1458 __u32 handle; 1459 __u32 pad; 1460 }; 1461 1462 struct drm_i915_gem_busy { 1463 /** Handle of the buffer to check for busy */ 1464 __u32 handle; 1465 1466 /** Return busy status 1467 * 1468 * A return of 0 implies that the object is idle (after 1469 * having flushed any pending activity), and a non-zero return that 1470 * the object is still in-flight on the GPU. (The GPU has not yet 1471 * signaled completion for all pending requests that reference the 1472 * object.) An object is guaranteed to become idle eventually (so 1473 * long as no new GPU commands are executed upon it). Due to the 1474 * asynchronous nature of the hardware, an object reported 1475 * as busy may become idle before the ioctl is completed. 1476 * 1477 * Furthermore, if the object is busy, which engine is busy is only 1478 * provided as a guide and only indirectly by reporting its class 1479 * (there may be more than one engine in each class). There are race 1480 * conditions which prevent the report of which engines are busy from 1481 * being always accurate. However, the converse is not true. If the 1482 * object is idle, the result of the ioctl, that all engines are idle, 1483 * is accurate. 1484 * 1485 * The returned dword is split into two fields to indicate both 1486 * the engine classess on which the object is being read, and the 1487 * engine class on which it is currently being written (if any). 1488 * 1489 * The low word (bits 0:15) indicate if the object is being written 1490 * to by any engine (there can only be one, as the GEM implicit 1491 * synchronisation rules force writes to be serialised). Only the 1492 * engine class (offset by 1, I915_ENGINE_CLASS_RENDER is reported as 1493 * 1 not 0 etc) for the last write is reported. 1494 * 1495 * The high word (bits 16:31) are a bitmask of which engines classes 1496 * are currently reading from the object. Multiple engines may be 1497 * reading from the object simultaneously. 1498 * 1499 * The value of each engine class is the same as specified in the 1500 * I915_CONTEXT_PARAM_ENGINES context parameter and via perf, i.e. 1501 * I915_ENGINE_CLASS_RENDER, I915_ENGINE_CLASS_COPY, etc. 1502 * Some hardware may have parallel execution engines, e.g. multiple 1503 * media engines, which are mapped to the same class identifier and so 1504 * are not separately reported for busyness. 1505 * 1506 * Caveat emptor: 1507 * Only the boolean result of this query is reliable; that is whether 1508 * the object is idle or busy. The report of which engines are busy 1509 * should be only used as a heuristic. 1510 */ 1511 __u32 busy; 1512 }; 1513 1514 /** 1515 * struct drm_i915_gem_caching - Set or get the caching for given object 1516 * handle. 1517 * 1518 * Allow userspace to control the GTT caching bits for a given object when the 1519 * object is later mapped through the ppGTT(or GGTT on older platforms lacking 1520 * ppGTT support, or if the object is used for scanout). Note that this might 1521 * require unbinding the object from the GTT first, if its current caching value 1522 * doesn't match. 1523 * 1524 * Note that this all changes on discrete platforms, starting from DG1, the 1525 * set/get caching is no longer supported, and is now rejected. Instead the CPU 1526 * caching attributes(WB vs WC) will become an immutable creation time property 1527 * for the object, along with the GTT caching level. For now we don't expose any 1528 * new uAPI for this, instead on DG1 this is all implicit, although this largely 1529 * shouldn't matter since DG1 is coherent by default(without any way of 1530 * controlling it). 1531 * 1532 * Implicit caching rules, starting from DG1: 1533 * 1534 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1535 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1536 * mapped as write-combined only. 1537 * 1538 * - Everything else is always allocated and mapped as write-back, with the 1539 * guarantee that everything is also coherent with the GPU. 1540 * 1541 * Note that this is likely to change in the future again, where we might need 1542 * more flexibility on future devices, so making this all explicit as part of a 1543 * new &drm_i915_gem_create_ext extension is probable. 1544 * 1545 * Side note: Part of the reason for this is that changing the at-allocation-time CPU 1546 * caching attributes for the pages might be required(and is expensive) if we 1547 * need to then CPU map the pages later with different caching attributes. This 1548 * inconsistent caching behaviour, while supported on x86, is not universally 1549 * supported on other architectures. So for simplicity we opt for setting 1550 * everything at creation time, whilst also making it immutable, on discrete 1551 * platforms. 1552 */ 1553 struct drm_i915_gem_caching { 1554 /** 1555 * @handle: Handle of the buffer to set/get the caching level. 1556 */ 1557 __u32 handle; 1558 1559 /** 1560 * @caching: The GTT caching level to apply or possible return value. 1561 * 1562 * The supported @caching values: 1563 * 1564 * I915_CACHING_NONE: 1565 * 1566 * GPU access is not coherent with CPU caches. Default for machines 1567 * without an LLC. This means manual flushing might be needed, if we 1568 * want GPU access to be coherent. 1569 * 1570 * I915_CACHING_CACHED: 1571 * 1572 * GPU access is coherent with CPU caches and furthermore the data is 1573 * cached in last-level caches shared between CPU cores and the GPU GT. 1574 * 1575 * I915_CACHING_DISPLAY: 1576 * 1577 * Special GPU caching mode which is coherent with the scanout engines. 1578 * Transparently falls back to I915_CACHING_NONE on platforms where no 1579 * special cache mode (like write-through or gfdt flushing) is 1580 * available. The kernel automatically sets this mode when using a 1581 * buffer as a scanout target. Userspace can manually set this mode to 1582 * avoid a costly stall and clflush in the hotpath of drawing the first 1583 * frame. 1584 */ 1585 #define I915_CACHING_NONE 0 1586 #define I915_CACHING_CACHED 1 1587 #define I915_CACHING_DISPLAY 2 1588 __u32 caching; 1589 }; 1590 1591 #define I915_TILING_NONE 0 1592 #define I915_TILING_X 1 1593 #define I915_TILING_Y 2 1594 /* 1595 * Do not add new tiling types here. The I915_TILING_* values are for 1596 * de-tiling fence registers that no longer exist on modern platforms. Although 1597 * the hardware may support new types of tiling in general (e.g., Tile4), we 1598 * do not need to add them to the uapi that is specific to now-defunct ioctls. 1599 */ 1600 #define I915_TILING_LAST I915_TILING_Y 1601 1602 #define I915_BIT_6_SWIZZLE_NONE 0 1603 #define I915_BIT_6_SWIZZLE_9 1 1604 #define I915_BIT_6_SWIZZLE_9_10 2 1605 #define I915_BIT_6_SWIZZLE_9_11 3 1606 #define I915_BIT_6_SWIZZLE_9_10_11 4 1607 /* Not seen by userland */ 1608 #define I915_BIT_6_SWIZZLE_UNKNOWN 5 1609 /* Seen by userland. */ 1610 #define I915_BIT_6_SWIZZLE_9_17 6 1611 #define I915_BIT_6_SWIZZLE_9_10_17 7 1612 1613 struct drm_i915_gem_set_tiling { 1614 /** Handle of the buffer to have its tiling state updated */ 1615 __u32 handle; 1616 1617 /** 1618 * Tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1619 * I915_TILING_Y). 1620 * 1621 * This value is to be set on request, and will be updated by the 1622 * kernel on successful return with the actual chosen tiling layout. 1623 * 1624 * The tiling mode may be demoted to I915_TILING_NONE when the system 1625 * has bit 6 swizzling that can't be managed correctly by GEM. 1626 * 1627 * Buffer contents become undefined when changing tiling_mode. 1628 */ 1629 __u32 tiling_mode; 1630 1631 /** 1632 * Stride in bytes for the object when in I915_TILING_X or 1633 * I915_TILING_Y. 1634 */ 1635 __u32 stride; 1636 1637 /** 1638 * Returned address bit 6 swizzling required for CPU access through 1639 * mmap mapping. 1640 */ 1641 __u32 swizzle_mode; 1642 }; 1643 1644 struct drm_i915_gem_get_tiling { 1645 /** Handle of the buffer to get tiling state for. */ 1646 __u32 handle; 1647 1648 /** 1649 * Current tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1650 * I915_TILING_Y). 1651 */ 1652 __u32 tiling_mode; 1653 1654 /** 1655 * Returned address bit 6 swizzling required for CPU access through 1656 * mmap mapping. 1657 */ 1658 __u32 swizzle_mode; 1659 1660 /** 1661 * Returned address bit 6 swizzling required for CPU access through 1662 * mmap mapping whilst bound. 1663 */ 1664 __u32 phys_swizzle_mode; 1665 }; 1666 1667 struct drm_i915_gem_get_aperture { 1668 /** Total size of the aperture used by i915_gem_execbuffer, in bytes */ 1669 __u64 aper_size; 1670 1671 /** 1672 * Available space in the aperture used by i915_gem_execbuffer, in 1673 * bytes 1674 */ 1675 __u64 aper_available_size; 1676 }; 1677 1678 struct drm_i915_get_pipe_from_crtc_id { 1679 /** ID of CRTC being requested **/ 1680 __u32 crtc_id; 1681 1682 /** pipe of requested CRTC **/ 1683 __u32 pipe; 1684 }; 1685 1686 #define I915_MADV_WILLNEED 0 1687 #define I915_MADV_DONTNEED 1 1688 #define __I915_MADV_PURGED 2 /* internal state */ 1689 1690 struct drm_i915_gem_madvise { 1691 /** Handle of the buffer to change the backing store advice */ 1692 __u32 handle; 1693 1694 /* Advice: either the buffer will be needed again in the near future, 1695 * or wont be and could be discarded under memory pressure. 1696 */ 1697 __u32 madv; 1698 1699 /** Whether the backing store still exists. */ 1700 __u32 retained; 1701 }; 1702 1703 /* flags */ 1704 #define I915_OVERLAY_TYPE_MASK 0xff 1705 #define I915_OVERLAY_YUV_PLANAR 0x01 1706 #define I915_OVERLAY_YUV_PACKED 0x02 1707 #define I915_OVERLAY_RGB 0x03 1708 1709 #define I915_OVERLAY_DEPTH_MASK 0xff00 1710 #define I915_OVERLAY_RGB24 0x1000 1711 #define I915_OVERLAY_RGB16 0x2000 1712 #define I915_OVERLAY_RGB15 0x3000 1713 #define I915_OVERLAY_YUV422 0x0100 1714 #define I915_OVERLAY_YUV411 0x0200 1715 #define I915_OVERLAY_YUV420 0x0300 1716 #define I915_OVERLAY_YUV410 0x0400 1717 1718 #define I915_OVERLAY_SWAP_MASK 0xff0000 1719 #define I915_OVERLAY_NO_SWAP 0x000000 1720 #define I915_OVERLAY_UV_SWAP 0x010000 1721 #define I915_OVERLAY_Y_SWAP 0x020000 1722 #define I915_OVERLAY_Y_AND_UV_SWAP 0x030000 1723 1724 #define I915_OVERLAY_FLAGS_MASK 0xff000000 1725 #define I915_OVERLAY_ENABLE 0x01000000 1726 1727 struct drm_intel_overlay_put_image { 1728 /* various flags and src format description */ 1729 __u32 flags; 1730 /* source picture description */ 1731 __u32 bo_handle; 1732 /* stride values and offsets are in bytes, buffer relative */ 1733 __u16 stride_Y; /* stride for packed formats */ 1734 __u16 stride_UV; 1735 __u32 offset_Y; /* offset for packet formats */ 1736 __u32 offset_U; 1737 __u32 offset_V; 1738 /* in pixels */ 1739 __u16 src_width; 1740 __u16 src_height; 1741 /* to compensate the scaling factors for partially covered surfaces */ 1742 __u16 src_scan_width; 1743 __u16 src_scan_height; 1744 /* output crtc description */ 1745 __u32 crtc_id; 1746 __u16 dst_x; 1747 __u16 dst_y; 1748 __u16 dst_width; 1749 __u16 dst_height; 1750 }; 1751 1752 /* flags */ 1753 #define I915_OVERLAY_UPDATE_ATTRS (1<<0) 1754 #define I915_OVERLAY_UPDATE_GAMMA (1<<1) 1755 #define I915_OVERLAY_DISABLE_DEST_COLORKEY (1<<2) 1756 struct drm_intel_overlay_attrs { 1757 __u32 flags; 1758 __u32 color_key; 1759 __s32 brightness; 1760 __u32 contrast; 1761 __u32 saturation; 1762 __u32 gamma0; 1763 __u32 gamma1; 1764 __u32 gamma2; 1765 __u32 gamma3; 1766 __u32 gamma4; 1767 __u32 gamma5; 1768 }; 1769 1770 /* 1771 * Intel sprite handling 1772 * 1773 * Color keying works with a min/mask/max tuple. Both source and destination 1774 * color keying is allowed. 1775 * 1776 * Source keying: 1777 * Sprite pixels within the min & max values, masked against the color channels 1778 * specified in the mask field, will be transparent. All other pixels will 1779 * be displayed on top of the primary plane. For RGB surfaces, only the min 1780 * and mask fields will be used; ranged compares are not allowed. 1781 * 1782 * Destination keying: 1783 * Primary plane pixels that match the min value, masked against the color 1784 * channels specified in the mask field, will be replaced by corresponding 1785 * pixels from the sprite plane. 1786 * 1787 * Note that source & destination keying are exclusive; only one can be 1788 * active on a given plane. 1789 */ 1790 1791 #define I915_SET_COLORKEY_NONE (1<<0) /* Deprecated. Instead set 1792 * flags==0 to disable colorkeying. 1793 */ 1794 #define I915_SET_COLORKEY_DESTINATION (1<<1) 1795 #define I915_SET_COLORKEY_SOURCE (1<<2) 1796 struct drm_intel_sprite_colorkey { 1797 __u32 plane_id; 1798 __u32 min_value; 1799 __u32 channel_mask; 1800 __u32 max_value; 1801 __u32 flags; 1802 }; 1803 1804 struct drm_i915_gem_wait { 1805 /** Handle of BO we shall wait on */ 1806 __u32 bo_handle; 1807 __u32 flags; 1808 /** Number of nanoseconds to wait, Returns time remaining. */ 1809 __s64 timeout_ns; 1810 }; 1811 1812 struct drm_i915_gem_context_create { 1813 __u32 ctx_id; /* output: id of new context*/ 1814 __u32 pad; 1815 }; 1816 1817 struct drm_i915_gem_context_create_ext { 1818 __u32 ctx_id; /* output: id of new context*/ 1819 __u32 flags; 1820 #define I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS (1u << 0) 1821 #define I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE (1u << 1) 1822 #define I915_CONTEXT_CREATE_FLAGS_UNKNOWN \ 1823 (-(I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE << 1)) 1824 __u64 extensions; 1825 }; 1826 1827 struct drm_i915_gem_context_param { 1828 __u32 ctx_id; 1829 __u32 size; 1830 __u64 param; 1831 #define I915_CONTEXT_PARAM_BAN_PERIOD 0x1 1832 /* I915_CONTEXT_PARAM_NO_ZEROMAP has been removed. On the off chance 1833 * someone somewhere has attempted to use it, never re-use this context 1834 * param number. 1835 */ 1836 #define I915_CONTEXT_PARAM_NO_ZEROMAP 0x2 1837 #define I915_CONTEXT_PARAM_GTT_SIZE 0x3 1838 #define I915_CONTEXT_PARAM_NO_ERROR_CAPTURE 0x4 1839 #define I915_CONTEXT_PARAM_BANNABLE 0x5 1840 #define I915_CONTEXT_PARAM_PRIORITY 0x6 1841 #define I915_CONTEXT_MAX_USER_PRIORITY 1023 /* inclusive */ 1842 #define I915_CONTEXT_DEFAULT_PRIORITY 0 1843 #define I915_CONTEXT_MIN_USER_PRIORITY -1023 /* inclusive */ 1844 /* 1845 * When using the following param, value should be a pointer to 1846 * drm_i915_gem_context_param_sseu. 1847 */ 1848 #define I915_CONTEXT_PARAM_SSEU 0x7 1849 1850 /* 1851 * Not all clients may want to attempt automatic recover of a context after 1852 * a hang (for example, some clients may only submit very small incremental 1853 * batches relying on known logical state of previous batches which will never 1854 * recover correctly and each attempt will hang), and so would prefer that 1855 * the context is forever banned instead. 1856 * 1857 * If set to false (0), after a reset, subsequent (and in flight) rendering 1858 * from this context is discarded, and the client will need to create a new 1859 * context to use instead. 1860 * 1861 * If set to true (1), the kernel will automatically attempt to recover the 1862 * context by skipping the hanging batch and executing the next batch starting 1863 * from the default context state (discarding the incomplete logical context 1864 * state lost due to the reset). 1865 * 1866 * On creation, all new contexts are marked as recoverable. 1867 */ 1868 #define I915_CONTEXT_PARAM_RECOVERABLE 0x8 1869 1870 /* 1871 * The id of the associated virtual memory address space (ppGTT) of 1872 * this context. Can be retrieved and passed to another context 1873 * (on the same fd) for both to use the same ppGTT and so share 1874 * address layouts, and avoid reloading the page tables on context 1875 * switches between themselves. 1876 * 1877 * See DRM_I915_GEM_VM_CREATE and DRM_I915_GEM_VM_DESTROY. 1878 */ 1879 #define I915_CONTEXT_PARAM_VM 0x9 1880 1881 /* 1882 * I915_CONTEXT_PARAM_ENGINES: 1883 * 1884 * Bind this context to operate on this subset of available engines. Henceforth, 1885 * the I915_EXEC_RING selector for DRM_IOCTL_I915_GEM_EXECBUFFER2 operates as 1886 * an index into this array of engines; I915_EXEC_DEFAULT selecting engine[0] 1887 * and upwards. Slots 0...N are filled in using the specified (class, instance). 1888 * Use 1889 * engine_class: I915_ENGINE_CLASS_INVALID, 1890 * engine_instance: I915_ENGINE_CLASS_INVALID_NONE 1891 * to specify a gap in the array that can be filled in later, e.g. by a 1892 * virtual engine used for load balancing. 1893 * 1894 * Setting the number of engines bound to the context to 0, by passing a zero 1895 * sized argument, will revert back to default settings. 1896 * 1897 * See struct i915_context_param_engines. 1898 * 1899 * Extensions: 1900 * i915_context_engines_load_balance (I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE) 1901 * i915_context_engines_bond (I915_CONTEXT_ENGINES_EXT_BOND) 1902 * i915_context_engines_parallel_submit (I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT) 1903 */ 1904 #define I915_CONTEXT_PARAM_ENGINES 0xa 1905 1906 /* 1907 * I915_CONTEXT_PARAM_PERSISTENCE: 1908 * 1909 * Allow the context and active rendering to survive the process until 1910 * completion. Persistence allows fire-and-forget clients to queue up a 1911 * bunch of work, hand the output over to a display server and then quit. 1912 * If the context is marked as not persistent, upon closing (either via 1913 * an explicit DRM_I915_GEM_CONTEXT_DESTROY or implicitly from file closure 1914 * or process termination), the context and any outstanding requests will be 1915 * cancelled (and exported fences for cancelled requests marked as -EIO). 1916 * 1917 * By default, new contexts allow persistence. 1918 */ 1919 #define I915_CONTEXT_PARAM_PERSISTENCE 0xb 1920 1921 /* This API has been removed. On the off chance someone somewhere has 1922 * attempted to use it, never re-use this context param number. 1923 */ 1924 #define I915_CONTEXT_PARAM_RINGSIZE 0xc 1925 1926 /* 1927 * I915_CONTEXT_PARAM_PROTECTED_CONTENT: 1928 * 1929 * Mark that the context makes use of protected content, which will result 1930 * in the context being invalidated when the protected content session is. 1931 * Given that the protected content session is killed on suspend, the device 1932 * is kept awake for the lifetime of a protected context, so the user should 1933 * make sure to dispose of them once done. 1934 * This flag can only be set at context creation time and, when set to true, 1935 * must be preceded by an explicit setting of I915_CONTEXT_PARAM_RECOVERABLE 1936 * to false. This flag can't be set to true in conjunction with setting the 1937 * I915_CONTEXT_PARAM_BANNABLE flag to false. Creation example: 1938 * 1939 * .. code-block:: C 1940 * 1941 * struct drm_i915_gem_context_create_ext_setparam p_protected = { 1942 * .base = { 1943 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 1944 * }, 1945 * .param = { 1946 * .param = I915_CONTEXT_PARAM_PROTECTED_CONTENT, 1947 * .value = 1, 1948 * } 1949 * }; 1950 * struct drm_i915_gem_context_create_ext_setparam p_norecover = { 1951 * .base = { 1952 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 1953 * .next_extension = to_user_pointer(&p_protected), 1954 * }, 1955 * .param = { 1956 * .param = I915_CONTEXT_PARAM_RECOVERABLE, 1957 * .value = 0, 1958 * } 1959 * }; 1960 * struct drm_i915_gem_context_create_ext create = { 1961 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 1962 * .extensions = to_user_pointer(&p_norecover); 1963 * }; 1964 * 1965 * ctx_id = gem_context_create_ext(drm_fd, &create); 1966 * 1967 * In addition to the normal failure cases, setting this flag during context 1968 * creation can result in the following errors: 1969 * 1970 * -ENODEV: feature not available 1971 * -EPERM: trying to mark a recoverable or not bannable context as protected 1972 */ 1973 #define I915_CONTEXT_PARAM_PROTECTED_CONTENT 0xd 1974 /* Must be kept compact -- no holes and well documented */ 1975 1976 __u64 value; 1977 }; 1978 1979 /* 1980 * Context SSEU programming 1981 * 1982 * It may be necessary for either functional or performance reason to configure 1983 * a context to run with a reduced number of SSEU (where SSEU stands for Slice/ 1984 * Sub-slice/EU). 1985 * 1986 * This is done by configuring SSEU configuration using the below 1987 * @struct drm_i915_gem_context_param_sseu for every supported engine which 1988 * userspace intends to use. 1989 * 1990 * Not all GPUs or engines support this functionality in which case an error 1991 * code -ENODEV will be returned. 1992 * 1993 * Also, flexibility of possible SSEU configuration permutations varies between 1994 * GPU generations and software imposed limitations. Requesting such a 1995 * combination will return an error code of -EINVAL. 1996 * 1997 * NOTE: When perf/OA is active the context's SSEU configuration is ignored in 1998 * favour of a single global setting. 1999 */ 2000 struct drm_i915_gem_context_param_sseu { 2001 /* 2002 * Engine class & instance to be configured or queried. 2003 */ 2004 struct i915_engine_class_instance engine; 2005 2006 /* 2007 * Unknown flags must be cleared to zero. 2008 */ 2009 __u32 flags; 2010 #define I915_CONTEXT_SSEU_FLAG_ENGINE_INDEX (1u << 0) 2011 2012 /* 2013 * Mask of slices to enable for the context. Valid values are a subset 2014 * of the bitmask value returned for I915_PARAM_SLICE_MASK. 2015 */ 2016 __u64 slice_mask; 2017 2018 /* 2019 * Mask of subslices to enable for the context. Valid values are a 2020 * subset of the bitmask value return by I915_PARAM_SUBSLICE_MASK. 2021 */ 2022 __u64 subslice_mask; 2023 2024 /* 2025 * Minimum/Maximum number of EUs to enable per subslice for the 2026 * context. min_eus_per_subslice must be inferior or equal to 2027 * max_eus_per_subslice. 2028 */ 2029 __u16 min_eus_per_subslice; 2030 __u16 max_eus_per_subslice; 2031 2032 /* 2033 * Unused for now. Must be cleared to zero. 2034 */ 2035 __u32 rsvd; 2036 }; 2037 2038 /** 2039 * DOC: Virtual Engine uAPI 2040 * 2041 * Virtual engine is a concept where userspace is able to configure a set of 2042 * physical engines, submit a batch buffer, and let the driver execute it on any 2043 * engine from the set as it sees fit. 2044 * 2045 * This is primarily useful on parts which have multiple instances of a same 2046 * class engine, like for example GT3+ Skylake parts with their two VCS engines. 2047 * 2048 * For instance userspace can enumerate all engines of a certain class using the 2049 * previously described `Engine Discovery uAPI`_. After that userspace can 2050 * create a GEM context with a placeholder slot for the virtual engine (using 2051 * `I915_ENGINE_CLASS_INVALID` and `I915_ENGINE_CLASS_INVALID_NONE` for class 2052 * and instance respectively) and finally using the 2053 * `I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE` extension place a virtual engine in 2054 * the same reserved slot. 2055 * 2056 * Example of creating a virtual engine and submitting a batch buffer to it: 2057 * 2058 * .. code-block:: C 2059 * 2060 * I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(virtual, 2) = { 2061 * .base.name = I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE, 2062 * .engine_index = 0, // Place this virtual engine into engine map slot 0 2063 * .num_siblings = 2, 2064 * .engines = { { I915_ENGINE_CLASS_VIDEO, 0 }, 2065 * { I915_ENGINE_CLASS_VIDEO, 1 }, }, 2066 * }; 2067 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 1) = { 2068 * .engines = { { I915_ENGINE_CLASS_INVALID, 2069 * I915_ENGINE_CLASS_INVALID_NONE } }, 2070 * .extensions = to_user_pointer(&virtual), // Chains after load_balance extension 2071 * }; 2072 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2073 * .base = { 2074 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2075 * }, 2076 * .param = { 2077 * .param = I915_CONTEXT_PARAM_ENGINES, 2078 * .value = to_user_pointer(&engines), 2079 * .size = sizeof(engines), 2080 * }, 2081 * }; 2082 * struct drm_i915_gem_context_create_ext create = { 2083 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2084 * .extensions = to_user_pointer(&p_engines); 2085 * }; 2086 * 2087 * ctx_id = gem_context_create_ext(drm_fd, &create); 2088 * 2089 * // Now we have created a GEM context with its engine map containing a 2090 * // single virtual engine. Submissions to this slot can go either to 2091 * // vcs0 or vcs1, depending on the load balancing algorithm used inside 2092 * // the driver. The load balancing is dynamic from one batch buffer to 2093 * // another and transparent to userspace. 2094 * 2095 * ... 2096 * execbuf.rsvd1 = ctx_id; 2097 * execbuf.flags = 0; // Submits to index 0 which is the virtual engine 2098 * gem_execbuf(drm_fd, &execbuf); 2099 */ 2100 2101 /* 2102 * i915_context_engines_load_balance: 2103 * 2104 * Enable load balancing across this set of engines. 2105 * 2106 * Into the I915_EXEC_DEFAULT slot [0], a virtual engine is created that when 2107 * used will proxy the execbuffer request onto one of the set of engines 2108 * in such a way as to distribute the load evenly across the set. 2109 * 2110 * The set of engines must be compatible (e.g. the same HW class) as they 2111 * will share the same logical GPU context and ring. 2112 * 2113 * To intermix rendering with the virtual engine and direct rendering onto 2114 * the backing engines (bypassing the load balancing proxy), the context must 2115 * be defined to use a single timeline for all engines. 2116 */ 2117 struct i915_context_engines_load_balance { 2118 struct i915_user_extension base; 2119 2120 __u16 engine_index; 2121 __u16 num_siblings; 2122 __u32 flags; /* all undefined flags must be zero */ 2123 2124 __u64 mbz64; /* reserved for future use; must be zero */ 2125 2126 struct i915_engine_class_instance engines[0]; 2127 } __attribute__((packed)); 2128 2129 #define I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(name__, N__) struct { \ 2130 struct i915_user_extension base; \ 2131 __u16 engine_index; \ 2132 __u16 num_siblings; \ 2133 __u32 flags; \ 2134 __u64 mbz64; \ 2135 struct i915_engine_class_instance engines[N__]; \ 2136 } __attribute__((packed)) name__ 2137 2138 /* 2139 * i915_context_engines_bond: 2140 * 2141 * Constructed bonded pairs for execution within a virtual engine. 2142 * 2143 * All engines are equal, but some are more equal than others. Given 2144 * the distribution of resources in the HW, it may be preferable to run 2145 * a request on a given subset of engines in parallel to a request on a 2146 * specific engine. We enable this selection of engines within a virtual 2147 * engine by specifying bonding pairs, for any given master engine we will 2148 * only execute on one of the corresponding siblings within the virtual engine. 2149 * 2150 * To execute a request in parallel on the master engine and a sibling requires 2151 * coordination with a I915_EXEC_FENCE_SUBMIT. 2152 */ 2153 struct i915_context_engines_bond { 2154 struct i915_user_extension base; 2155 2156 struct i915_engine_class_instance master; 2157 2158 __u16 virtual_index; /* index of virtual engine in ctx->engines[] */ 2159 __u16 num_bonds; 2160 2161 __u64 flags; /* all undefined flags must be zero */ 2162 __u64 mbz64[4]; /* reserved for future use; must be zero */ 2163 2164 struct i915_engine_class_instance engines[0]; 2165 } __attribute__((packed)); 2166 2167 #define I915_DEFINE_CONTEXT_ENGINES_BOND(name__, N__) struct { \ 2168 struct i915_user_extension base; \ 2169 struct i915_engine_class_instance master; \ 2170 __u16 virtual_index; \ 2171 __u16 num_bonds; \ 2172 __u64 flags; \ 2173 __u64 mbz64[4]; \ 2174 struct i915_engine_class_instance engines[N__]; \ 2175 } __attribute__((packed)) name__ 2176 2177 /** 2178 * struct i915_context_engines_parallel_submit - Configure engine for 2179 * parallel submission. 2180 * 2181 * Setup a slot in the context engine map to allow multiple BBs to be submitted 2182 * in a single execbuf IOCTL. Those BBs will then be scheduled to run on the GPU 2183 * in parallel. Multiple hardware contexts are created internally in the i915 to 2184 * run these BBs. Once a slot is configured for N BBs only N BBs can be 2185 * submitted in each execbuf IOCTL and this is implicit behavior e.g. The user 2186 * doesn't tell the execbuf IOCTL there are N BBs, the execbuf IOCTL knows how 2187 * many BBs there are based on the slot's configuration. The N BBs are the last 2188 * N buffer objects or first N if I915_EXEC_BATCH_FIRST is set. 2189 * 2190 * The default placement behavior is to create implicit bonds between each 2191 * context if each context maps to more than 1 physical engine (e.g. context is 2192 * a virtual engine). Also we only allow contexts of same engine class and these 2193 * contexts must be in logically contiguous order. Examples of the placement 2194 * behavior are described below. Lastly, the default is to not allow BBs to be 2195 * preempted mid-batch. Rather insert coordinated preemption points on all 2196 * hardware contexts between each set of BBs. Flags could be added in the future 2197 * to change both of these default behaviors. 2198 * 2199 * Returns -EINVAL if hardware context placement configuration is invalid or if 2200 * the placement configuration isn't supported on the platform / submission 2201 * interface. 2202 * Returns -ENODEV if extension isn't supported on the platform / submission 2203 * interface. 2204 * 2205 * .. code-block:: none 2206 * 2207 * Examples syntax: 2208 * CS[X] = generic engine of same class, logical instance X 2209 * INVALID = I915_ENGINE_CLASS_INVALID, I915_ENGINE_CLASS_INVALID_NONE 2210 * 2211 * Example 1 pseudo code: 2212 * set_engines(INVALID) 2213 * set_parallel(engine_index=0, width=2, num_siblings=1, 2214 * engines=CS[0],CS[1]) 2215 * 2216 * Results in the following valid placement: 2217 * CS[0], CS[1] 2218 * 2219 * Example 2 pseudo code: 2220 * set_engines(INVALID) 2221 * set_parallel(engine_index=0, width=2, num_siblings=2, 2222 * engines=CS[0],CS[2],CS[1],CS[3]) 2223 * 2224 * Results in the following valid placements: 2225 * CS[0], CS[1] 2226 * CS[2], CS[3] 2227 * 2228 * This can be thought of as two virtual engines, each containing two 2229 * engines thereby making a 2D array. However, there are bonds tying the 2230 * entries together and placing restrictions on how they can be scheduled. 2231 * Specifically, the scheduler can choose only vertical columns from the 2D 2232 * array. That is, CS[0] is bonded to CS[1] and CS[2] to CS[3]. So if the 2233 * scheduler wants to submit to CS[0], it must also choose CS[1] and vice 2234 * versa. Same for CS[2] requires also using CS[3]. 2235 * VE[0] = CS[0], CS[2] 2236 * VE[1] = CS[1], CS[3] 2237 * 2238 * Example 3 pseudo code: 2239 * set_engines(INVALID) 2240 * set_parallel(engine_index=0, width=2, num_siblings=2, 2241 * engines=CS[0],CS[1],CS[1],CS[3]) 2242 * 2243 * Results in the following valid and invalid placements: 2244 * CS[0], CS[1] 2245 * CS[1], CS[3] - Not logically contiguous, return -EINVAL 2246 */ 2247 struct i915_context_engines_parallel_submit { 2248 /** 2249 * @base: base user extension. 2250 */ 2251 struct i915_user_extension base; 2252 2253 /** 2254 * @engine_index: slot for parallel engine 2255 */ 2256 __u16 engine_index; 2257 2258 /** 2259 * @width: number of contexts per parallel engine or in other words the 2260 * number of batches in each submission 2261 */ 2262 __u16 width; 2263 2264 /** 2265 * @num_siblings: number of siblings per context or in other words the 2266 * number of possible placements for each submission 2267 */ 2268 __u16 num_siblings; 2269 2270 /** 2271 * @mbz16: reserved for future use; must be zero 2272 */ 2273 __u16 mbz16; 2274 2275 /** 2276 * @flags: all undefined flags must be zero, currently not defined flags 2277 */ 2278 __u64 flags; 2279 2280 /** 2281 * @mbz64: reserved for future use; must be zero 2282 */ 2283 __u64 mbz64[3]; 2284 2285 /** 2286 * @engines: 2-d array of engine instances to configure parallel engine 2287 * 2288 * length = width (i) * num_siblings (j) 2289 * index = j + i * num_siblings 2290 */ 2291 struct i915_engine_class_instance engines[0]; 2292 2293 } __packed; 2294 2295 #define I915_DEFINE_CONTEXT_ENGINES_PARALLEL_SUBMIT(name__, N__) struct { \ 2296 struct i915_user_extension base; \ 2297 __u16 engine_index; \ 2298 __u16 width; \ 2299 __u16 num_siblings; \ 2300 __u16 mbz16; \ 2301 __u64 flags; \ 2302 __u64 mbz64[3]; \ 2303 struct i915_engine_class_instance engines[N__]; \ 2304 } __attribute__((packed)) name__ 2305 2306 /** 2307 * DOC: Context Engine Map uAPI 2308 * 2309 * Context engine map is a new way of addressing engines when submitting batch- 2310 * buffers, replacing the existing way of using identifiers like `I915_EXEC_BLT` 2311 * inside the flags field of `struct drm_i915_gem_execbuffer2`. 2312 * 2313 * To use it created GEM contexts need to be configured with a list of engines 2314 * the user is intending to submit to. This is accomplished using the 2315 * `I915_CONTEXT_PARAM_ENGINES` parameter and `struct 2316 * i915_context_param_engines`. 2317 * 2318 * For such contexts the `I915_EXEC_RING_MASK` field becomes an index into the 2319 * configured map. 2320 * 2321 * Example of creating such context and submitting against it: 2322 * 2323 * .. code-block:: C 2324 * 2325 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 2) = { 2326 * .engines = { { I915_ENGINE_CLASS_RENDER, 0 }, 2327 * { I915_ENGINE_CLASS_COPY, 0 } } 2328 * }; 2329 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2330 * .base = { 2331 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2332 * }, 2333 * .param = { 2334 * .param = I915_CONTEXT_PARAM_ENGINES, 2335 * .value = to_user_pointer(&engines), 2336 * .size = sizeof(engines), 2337 * }, 2338 * }; 2339 * struct drm_i915_gem_context_create_ext create = { 2340 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2341 * .extensions = to_user_pointer(&p_engines); 2342 * }; 2343 * 2344 * ctx_id = gem_context_create_ext(drm_fd, &create); 2345 * 2346 * // We have now created a GEM context with two engines in the map: 2347 * // Index 0 points to rcs0 while index 1 points to bcs0. Other engines 2348 * // will not be accessible from this context. 2349 * 2350 * ... 2351 * execbuf.rsvd1 = ctx_id; 2352 * execbuf.flags = 0; // Submits to index 0, which is rcs0 for this context 2353 * gem_execbuf(drm_fd, &execbuf); 2354 * 2355 * ... 2356 * execbuf.rsvd1 = ctx_id; 2357 * execbuf.flags = 1; // Submits to index 0, which is bcs0 for this context 2358 * gem_execbuf(drm_fd, &execbuf); 2359 */ 2360 2361 struct i915_context_param_engines { 2362 __u64 extensions; /* linked chain of extension blocks, 0 terminates */ 2363 #define I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE 0 /* see i915_context_engines_load_balance */ 2364 #define I915_CONTEXT_ENGINES_EXT_BOND 1 /* see i915_context_engines_bond */ 2365 #define I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT 2 /* see i915_context_engines_parallel_submit */ 2366 struct i915_engine_class_instance engines[0]; 2367 } __attribute__((packed)); 2368 2369 #define I915_DEFINE_CONTEXT_PARAM_ENGINES(name__, N__) struct { \ 2370 __u64 extensions; \ 2371 struct i915_engine_class_instance engines[N__]; \ 2372 } __attribute__((packed)) name__ 2373 2374 struct drm_i915_gem_context_create_ext_setparam { 2375 #define I915_CONTEXT_CREATE_EXT_SETPARAM 0 2376 struct i915_user_extension base; 2377 struct drm_i915_gem_context_param param; 2378 }; 2379 2380 /* This API has been removed. On the off chance someone somewhere has 2381 * attempted to use it, never re-use this extension number. 2382 */ 2383 #define I915_CONTEXT_CREATE_EXT_CLONE 1 2384 2385 struct drm_i915_gem_context_destroy { 2386 __u32 ctx_id; 2387 __u32 pad; 2388 }; 2389 2390 /* 2391 * DRM_I915_GEM_VM_CREATE - 2392 * 2393 * Create a new virtual memory address space (ppGTT) for use within a context 2394 * on the same file. Extensions can be provided to configure exactly how the 2395 * address space is setup upon creation. 2396 * 2397 * The id of new VM (bound to the fd) for use with I915_CONTEXT_PARAM_VM is 2398 * returned in the outparam @id. 2399 * 2400 * No flags are defined, with all bits reserved and must be zero. 2401 * 2402 * An extension chain maybe provided, starting with @extensions, and terminated 2403 * by the @next_extension being 0. Currently, no extensions are defined. 2404 * 2405 * DRM_I915_GEM_VM_DESTROY - 2406 * 2407 * Destroys a previously created VM id, specified in @id. 2408 * 2409 * No extensions or flags are allowed currently, and so must be zero. 2410 */ 2411 struct drm_i915_gem_vm_control { 2412 __u64 extensions; 2413 __u32 flags; 2414 __u32 vm_id; 2415 }; 2416 2417 struct drm_i915_reg_read { 2418 /* 2419 * Register offset. 2420 * For 64bit wide registers where the upper 32bits don't immediately 2421 * follow the lower 32bits, the offset of the lower 32bits must 2422 * be specified 2423 */ 2424 __u64 offset; 2425 #define I915_REG_READ_8B_WA (1ul << 0) 2426 2427 __u64 val; /* Return value */ 2428 }; 2429 2430 /* Known registers: 2431 * 2432 * Render engine timestamp - 0x2358 + 64bit - gen7+ 2433 * - Note this register returns an invalid value if using the default 2434 * single instruction 8byte read, in order to workaround that pass 2435 * flag I915_REG_READ_8B_WA in offset field. 2436 * 2437 */ 2438 2439 struct drm_i915_reset_stats { 2440 __u32 ctx_id; 2441 __u32 flags; 2442 2443 /* All resets since boot/module reload, for all contexts */ 2444 __u32 reset_count; 2445 2446 /* Number of batches lost when active in GPU, for this context */ 2447 __u32 batch_active; 2448 2449 /* Number of batches lost pending for execution, for this context */ 2450 __u32 batch_pending; 2451 2452 __u32 pad; 2453 }; 2454 2455 /** 2456 * struct drm_i915_gem_userptr - Create GEM object from user allocated memory. 2457 * 2458 * Userptr objects have several restrictions on what ioctls can be used with the 2459 * object handle. 2460 */ 2461 struct drm_i915_gem_userptr { 2462 /** 2463 * @user_ptr: The pointer to the allocated memory. 2464 * 2465 * Needs to be aligned to PAGE_SIZE. 2466 */ 2467 __u64 user_ptr; 2468 2469 /** 2470 * @user_size: 2471 * 2472 * The size in bytes for the allocated memory. This will also become the 2473 * object size. 2474 * 2475 * Needs to be aligned to PAGE_SIZE, and should be at least PAGE_SIZE, 2476 * or larger. 2477 */ 2478 __u64 user_size; 2479 2480 /** 2481 * @flags: 2482 * 2483 * Supported flags: 2484 * 2485 * I915_USERPTR_READ_ONLY: 2486 * 2487 * Mark the object as readonly, this also means GPU access can only be 2488 * readonly. This is only supported on HW which supports readonly access 2489 * through the GTT. If the HW can't support readonly access, an error is 2490 * returned. 2491 * 2492 * I915_USERPTR_PROBE: 2493 * 2494 * Probe the provided @user_ptr range and validate that the @user_ptr is 2495 * indeed pointing to normal memory and that the range is also valid. 2496 * For example if some garbage address is given to the kernel, then this 2497 * should complain. 2498 * 2499 * Returns -EFAULT if the probe failed. 2500 * 2501 * Note that this doesn't populate the backing pages, and also doesn't 2502 * guarantee that the object will remain valid when the object is 2503 * eventually used. 2504 * 2505 * The kernel supports this feature if I915_PARAM_HAS_USERPTR_PROBE 2506 * returns a non-zero value. 2507 * 2508 * I915_USERPTR_UNSYNCHRONIZED: 2509 * 2510 * NOT USED. Setting this flag will result in an error. 2511 */ 2512 __u32 flags; 2513 #define I915_USERPTR_READ_ONLY 0x1 2514 #define I915_USERPTR_PROBE 0x2 2515 #define I915_USERPTR_UNSYNCHRONIZED 0x80000000 2516 /** 2517 * @handle: Returned handle for the object. 2518 * 2519 * Object handles are nonzero. 2520 */ 2521 __u32 handle; 2522 }; 2523 2524 enum drm_i915_oa_format { 2525 I915_OA_FORMAT_A13 = 1, /* HSW only */ 2526 I915_OA_FORMAT_A29, /* HSW only */ 2527 I915_OA_FORMAT_A13_B8_C8, /* HSW only */ 2528 I915_OA_FORMAT_B4_C8, /* HSW only */ 2529 I915_OA_FORMAT_A45_B8_C8, /* HSW only */ 2530 I915_OA_FORMAT_B4_C8_A16, /* HSW only */ 2531 I915_OA_FORMAT_C4_B8, /* HSW+ */ 2532 2533 /* Gen8+ */ 2534 I915_OA_FORMAT_A12, 2535 I915_OA_FORMAT_A12_B8_C8, 2536 I915_OA_FORMAT_A32u40_A4u32_B8_C8, 2537 2538 I915_OA_FORMAT_MAX /* non-ABI */ 2539 }; 2540 2541 enum drm_i915_perf_property_id { 2542 /** 2543 * Open the stream for a specific context handle (as used with 2544 * execbuffer2). A stream opened for a specific context this way 2545 * won't typically require root privileges. 2546 * 2547 * This property is available in perf revision 1. 2548 */ 2549 DRM_I915_PERF_PROP_CTX_HANDLE = 1, 2550 2551 /** 2552 * A value of 1 requests the inclusion of raw OA unit reports as 2553 * part of stream samples. 2554 * 2555 * This property is available in perf revision 1. 2556 */ 2557 DRM_I915_PERF_PROP_SAMPLE_OA, 2558 2559 /** 2560 * The value specifies which set of OA unit metrics should be 2561 * configured, defining the contents of any OA unit reports. 2562 * 2563 * This property is available in perf revision 1. 2564 */ 2565 DRM_I915_PERF_PROP_OA_METRICS_SET, 2566 2567 /** 2568 * The value specifies the size and layout of OA unit reports. 2569 * 2570 * This property is available in perf revision 1. 2571 */ 2572 DRM_I915_PERF_PROP_OA_FORMAT, 2573 2574 /** 2575 * Specifying this property implicitly requests periodic OA unit 2576 * sampling and (at least on Haswell) the sampling frequency is derived 2577 * from this exponent as follows: 2578 * 2579 * 80ns * 2^(period_exponent + 1) 2580 * 2581 * This property is available in perf revision 1. 2582 */ 2583 DRM_I915_PERF_PROP_OA_EXPONENT, 2584 2585 /** 2586 * Specifying this property is only valid when specify a context to 2587 * filter with DRM_I915_PERF_PROP_CTX_HANDLE. Specifying this property 2588 * will hold preemption of the particular context we want to gather 2589 * performance data about. The execbuf2 submissions must include a 2590 * drm_i915_gem_execbuffer_ext_perf parameter for this to apply. 2591 * 2592 * This property is available in perf revision 3. 2593 */ 2594 DRM_I915_PERF_PROP_HOLD_PREEMPTION, 2595 2596 /** 2597 * Specifying this pins all contexts to the specified SSEU power 2598 * configuration for the duration of the recording. 2599 * 2600 * This parameter's value is a pointer to a struct 2601 * drm_i915_gem_context_param_sseu. 2602 * 2603 * This property is available in perf revision 4. 2604 */ 2605 DRM_I915_PERF_PROP_GLOBAL_SSEU, 2606 2607 /** 2608 * This optional parameter specifies the timer interval in nanoseconds 2609 * at which the i915 driver will check the OA buffer for available data. 2610 * Minimum allowed value is 100 microseconds. A default value is used by 2611 * the driver if this parameter is not specified. Note that larger timer 2612 * values will reduce cpu consumption during OA perf captures. However, 2613 * excessively large values would potentially result in OA buffer 2614 * overwrites as captures reach end of the OA buffer. 2615 * 2616 * This property is available in perf revision 5. 2617 */ 2618 DRM_I915_PERF_PROP_POLL_OA_PERIOD, 2619 2620 DRM_I915_PERF_PROP_MAX /* non-ABI */ 2621 }; 2622 2623 struct drm_i915_perf_open_param { 2624 __u32 flags; 2625 #define I915_PERF_FLAG_FD_CLOEXEC (1<<0) 2626 #define I915_PERF_FLAG_FD_NONBLOCK (1<<1) 2627 #define I915_PERF_FLAG_DISABLED (1<<2) 2628 2629 /** The number of u64 (id, value) pairs */ 2630 __u32 num_properties; 2631 2632 /** 2633 * Pointer to array of u64 (id, value) pairs configuring the stream 2634 * to open. 2635 */ 2636 __u64 properties_ptr; 2637 }; 2638 2639 /* 2640 * Enable data capture for a stream that was either opened in a disabled state 2641 * via I915_PERF_FLAG_DISABLED or was later disabled via 2642 * I915_PERF_IOCTL_DISABLE. 2643 * 2644 * It is intended to be cheaper to disable and enable a stream than it may be 2645 * to close and re-open a stream with the same configuration. 2646 * 2647 * It's undefined whether any pending data for the stream will be lost. 2648 * 2649 * This ioctl is available in perf revision 1. 2650 */ 2651 #define I915_PERF_IOCTL_ENABLE _IO('i', 0x0) 2652 2653 /* 2654 * Disable data capture for a stream. 2655 * 2656 * It is an error to try and read a stream that is disabled. 2657 * 2658 * This ioctl is available in perf revision 1. 2659 */ 2660 #define I915_PERF_IOCTL_DISABLE _IO('i', 0x1) 2661 2662 /* 2663 * Change metrics_set captured by a stream. 2664 * 2665 * If the stream is bound to a specific context, the configuration change 2666 * will performed inline with that context such that it takes effect before 2667 * the next execbuf submission. 2668 * 2669 * Returns the previously bound metrics set id, or a negative error code. 2670 * 2671 * This ioctl is available in perf revision 2. 2672 */ 2673 #define I915_PERF_IOCTL_CONFIG _IO('i', 0x2) 2674 2675 /* 2676 * Common to all i915 perf records 2677 */ 2678 struct drm_i915_perf_record_header { 2679 __u32 type; 2680 __u16 pad; 2681 __u16 size; 2682 }; 2683 2684 enum drm_i915_perf_record_type { 2685 2686 /** 2687 * Samples are the work horse record type whose contents are extensible 2688 * and defined when opening an i915 perf stream based on the given 2689 * properties. 2690 * 2691 * Boolean properties following the naming convention 2692 * DRM_I915_PERF_SAMPLE_xyz_PROP request the inclusion of 'xyz' data in 2693 * every sample. 2694 * 2695 * The order of these sample properties given by userspace has no 2696 * affect on the ordering of data within a sample. The order is 2697 * documented here. 2698 * 2699 * struct { 2700 * struct drm_i915_perf_record_header header; 2701 * 2702 * { u32 oa_report[]; } && DRM_I915_PERF_PROP_SAMPLE_OA 2703 * }; 2704 */ 2705 DRM_I915_PERF_RECORD_SAMPLE = 1, 2706 2707 /* 2708 * Indicates that one or more OA reports were not written by the 2709 * hardware. This can happen for example if an MI_REPORT_PERF_COUNT 2710 * command collides with periodic sampling - which would be more likely 2711 * at higher sampling frequencies. 2712 */ 2713 DRM_I915_PERF_RECORD_OA_REPORT_LOST = 2, 2714 2715 /** 2716 * An error occurred that resulted in all pending OA reports being lost. 2717 */ 2718 DRM_I915_PERF_RECORD_OA_BUFFER_LOST = 3, 2719 2720 DRM_I915_PERF_RECORD_MAX /* non-ABI */ 2721 }; 2722 2723 /** 2724 * struct drm_i915_perf_oa_config 2725 * 2726 * Structure to upload perf dynamic configuration into the kernel. 2727 */ 2728 struct drm_i915_perf_oa_config { 2729 /** 2730 * @uuid: 2731 * 2732 * String formatted like "%\08x-%\04x-%\04x-%\04x-%\012x" 2733 */ 2734 char uuid[36]; 2735 2736 /** 2737 * @n_mux_regs: 2738 * 2739 * Number of mux regs in &mux_regs_ptr. 2740 */ 2741 __u32 n_mux_regs; 2742 2743 /** 2744 * @n_boolean_regs: 2745 * 2746 * Number of boolean regs in &boolean_regs_ptr. 2747 */ 2748 __u32 n_boolean_regs; 2749 2750 /** 2751 * @n_flex_regs: 2752 * 2753 * Number of flex regs in &flex_regs_ptr. 2754 */ 2755 __u32 n_flex_regs; 2756 2757 /** 2758 * @mux_regs_ptr: 2759 * 2760 * Pointer to tuples of u32 values (register address, value) for mux 2761 * registers. Expected length of buffer is (2 * sizeof(u32) * 2762 * &n_mux_regs). 2763 */ 2764 __u64 mux_regs_ptr; 2765 2766 /** 2767 * @boolean_regs_ptr: 2768 * 2769 * Pointer to tuples of u32 values (register address, value) for mux 2770 * registers. Expected length of buffer is (2 * sizeof(u32) * 2771 * &n_boolean_regs). 2772 */ 2773 __u64 boolean_regs_ptr; 2774 2775 /** 2776 * @flex_regs_ptr: 2777 * 2778 * Pointer to tuples of u32 values (register address, value) for mux 2779 * registers. Expected length of buffer is (2 * sizeof(u32) * 2780 * &n_flex_regs). 2781 */ 2782 __u64 flex_regs_ptr; 2783 }; 2784 2785 /** 2786 * struct drm_i915_query_item - An individual query for the kernel to process. 2787 * 2788 * The behaviour is determined by the @query_id. Note that exactly what 2789 * @data_ptr is also depends on the specific @query_id. 2790 */ 2791 struct drm_i915_query_item { 2792 /** 2793 * @query_id: 2794 * 2795 * The id for this query. Currently accepted query IDs are: 2796 * - %DRM_I915_QUERY_TOPOLOGY_INFO (see struct drm_i915_query_topology_info) 2797 * - %DRM_I915_QUERY_ENGINE_INFO (see struct drm_i915_engine_info) 2798 * - %DRM_I915_QUERY_PERF_CONFIG (see struct drm_i915_query_perf_config) 2799 * - %DRM_I915_QUERY_MEMORY_REGIONS (see struct drm_i915_query_memory_regions) 2800 * - %DRM_I915_QUERY_HWCONFIG_BLOB (see `GuC HWCONFIG blob uAPI`) 2801 * - %DRM_I915_QUERY_GEOMETRY_SUBSLICES (see struct drm_i915_query_topology_info) 2802 */ 2803 __u64 query_id; 2804 #define DRM_I915_QUERY_TOPOLOGY_INFO 1 2805 #define DRM_I915_QUERY_ENGINE_INFO 2 2806 #define DRM_I915_QUERY_PERF_CONFIG 3 2807 #define DRM_I915_QUERY_MEMORY_REGIONS 4 2808 #define DRM_I915_QUERY_HWCONFIG_BLOB 5 2809 #define DRM_I915_QUERY_GEOMETRY_SUBSLICES 6 2810 /* Must be kept compact -- no holes and well documented */ 2811 2812 /** 2813 * @length: 2814 * 2815 * When set to zero by userspace, this is filled with the size of the 2816 * data to be written at the @data_ptr pointer. The kernel sets this 2817 * value to a negative value to signal an error on a particular query 2818 * item. 2819 */ 2820 __s32 length; 2821 2822 /** 2823 * @flags: 2824 * 2825 * When &query_id == %DRM_I915_QUERY_TOPOLOGY_INFO, must be 0. 2826 * 2827 * When &query_id == %DRM_I915_QUERY_PERF_CONFIG, must be one of the 2828 * following: 2829 * 2830 * - %DRM_I915_QUERY_PERF_CONFIG_LIST 2831 * - %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2832 * - %DRM_I915_QUERY_PERF_CONFIG_FOR_UUID 2833 * 2834 * When &query_id == %DRM_I915_QUERY_GEOMETRY_SUBSLICES must contain 2835 * a struct i915_engine_class_instance that references a render engine. 2836 */ 2837 __u32 flags; 2838 #define DRM_I915_QUERY_PERF_CONFIG_LIST 1 2839 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2 2840 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID 3 2841 2842 /** 2843 * @data_ptr: 2844 * 2845 * Data will be written at the location pointed by @data_ptr when the 2846 * value of @length matches the length of the data to be written by the 2847 * kernel. 2848 */ 2849 __u64 data_ptr; 2850 }; 2851 2852 /** 2853 * struct drm_i915_query - Supply an array of struct drm_i915_query_item for the 2854 * kernel to fill out. 2855 * 2856 * Note that this is generally a two step process for each struct 2857 * drm_i915_query_item in the array: 2858 * 2859 * 1. Call the DRM_IOCTL_I915_QUERY, giving it our array of struct 2860 * drm_i915_query_item, with &drm_i915_query_item.length set to zero. The 2861 * kernel will then fill in the size, in bytes, which tells userspace how 2862 * memory it needs to allocate for the blob(say for an array of properties). 2863 * 2864 * 2. Next we call DRM_IOCTL_I915_QUERY again, this time with the 2865 * &drm_i915_query_item.data_ptr equal to our newly allocated blob. Note that 2866 * the &drm_i915_query_item.length should still be the same as what the 2867 * kernel previously set. At this point the kernel can fill in the blob. 2868 * 2869 * Note that for some query items it can make sense for userspace to just pass 2870 * in a buffer/blob equal to or larger than the required size. In this case only 2871 * a single ioctl call is needed. For some smaller query items this can work 2872 * quite well. 2873 * 2874 */ 2875 struct drm_i915_query { 2876 /** @num_items: The number of elements in the @items_ptr array */ 2877 __u32 num_items; 2878 2879 /** 2880 * @flags: Unused for now. Must be cleared to zero. 2881 */ 2882 __u32 flags; 2883 2884 /** 2885 * @items_ptr: 2886 * 2887 * Pointer to an array of struct drm_i915_query_item. The number of 2888 * array elements is @num_items. 2889 */ 2890 __u64 items_ptr; 2891 }; 2892 2893 /** 2894 * struct drm_i915_query_topology_info 2895 * 2896 * Describes slice/subslice/EU information queried by 2897 * %DRM_I915_QUERY_TOPOLOGY_INFO 2898 */ 2899 struct drm_i915_query_topology_info { 2900 /** 2901 * @flags: 2902 * 2903 * Unused for now. Must be cleared to zero. 2904 */ 2905 __u16 flags; 2906 2907 /** 2908 * @max_slices: 2909 * 2910 * The number of bits used to express the slice mask. 2911 */ 2912 __u16 max_slices; 2913 2914 /** 2915 * @max_subslices: 2916 * 2917 * The number of bits used to express the subslice mask. 2918 */ 2919 __u16 max_subslices; 2920 2921 /** 2922 * @max_eus_per_subslice: 2923 * 2924 * The number of bits in the EU mask that correspond to a single 2925 * subslice's EUs. 2926 */ 2927 __u16 max_eus_per_subslice; 2928 2929 /** 2930 * @subslice_offset: 2931 * 2932 * Offset in data[] at which the subslice masks are stored. 2933 */ 2934 __u16 subslice_offset; 2935 2936 /** 2937 * @subslice_stride: 2938 * 2939 * Stride at which each of the subslice masks for each slice are 2940 * stored. 2941 */ 2942 __u16 subslice_stride; 2943 2944 /** 2945 * @eu_offset: 2946 * 2947 * Offset in data[] at which the EU masks are stored. 2948 */ 2949 __u16 eu_offset; 2950 2951 /** 2952 * @eu_stride: 2953 * 2954 * Stride at which each of the EU masks for each subslice are stored. 2955 */ 2956 __u16 eu_stride; 2957 2958 /** 2959 * @data: 2960 * 2961 * Contains 3 pieces of information : 2962 * 2963 * - The slice mask with one bit per slice telling whether a slice is 2964 * available. The availability of slice X can be queried with the 2965 * following formula : 2966 * 2967 * .. code:: c 2968 * 2969 * (data[X / 8] >> (X % 8)) & 1 2970 * 2971 * Starting with Xe_HP platforms, Intel hardware no longer has 2972 * traditional slices so i915 will always report a single slice 2973 * (hardcoded slicemask = 0x1) which contains all of the platform's 2974 * subslices. I.e., the mask here does not reflect any of the newer 2975 * hardware concepts such as "gslices" or "cslices" since userspace 2976 * is capable of inferring those from the subslice mask. 2977 * 2978 * - The subslice mask for each slice with one bit per subslice telling 2979 * whether a subslice is available. Starting with Gen12 we use the 2980 * term "subslice" to refer to what the hardware documentation 2981 * describes as a "dual-subslices." The availability of subslice Y 2982 * in slice X can be queried with the following formula : 2983 * 2984 * .. code:: c 2985 * 2986 * (data[subslice_offset + X * subslice_stride + Y / 8] >> (Y % 8)) & 1 2987 * 2988 * - The EU mask for each subslice in each slice, with one bit per EU 2989 * telling whether an EU is available. The availability of EU Z in 2990 * subslice Y in slice X can be queried with the following formula : 2991 * 2992 * .. code:: c 2993 * 2994 * (data[eu_offset + 2995 * (X * max_subslices + Y) * eu_stride + 2996 * Z / 8 2997 * ] >> (Z % 8)) & 1 2998 */ 2999 __u8 data[]; 3000 }; 3001 3002 /** 3003 * DOC: Engine Discovery uAPI 3004 * 3005 * Engine discovery uAPI is a way of enumerating physical engines present in a 3006 * GPU associated with an open i915 DRM file descriptor. This supersedes the old 3007 * way of using `DRM_IOCTL_I915_GETPARAM` and engine identifiers like 3008 * `I915_PARAM_HAS_BLT`. 3009 * 3010 * The need for this interface came starting with Icelake and newer GPUs, which 3011 * started to establish a pattern of having multiple engines of a same class, 3012 * where not all instances were always completely functionally equivalent. 3013 * 3014 * Entry point for this uapi is `DRM_IOCTL_I915_QUERY` with the 3015 * `DRM_I915_QUERY_ENGINE_INFO` as the queried item id. 3016 * 3017 * Example for getting the list of engines: 3018 * 3019 * .. code-block:: C 3020 * 3021 * struct drm_i915_query_engine_info *info; 3022 * struct drm_i915_query_item item = { 3023 * .query_id = DRM_I915_QUERY_ENGINE_INFO; 3024 * }; 3025 * struct drm_i915_query query = { 3026 * .num_items = 1, 3027 * .items_ptr = (uintptr_t)&item, 3028 * }; 3029 * int err, i; 3030 * 3031 * // First query the size of the blob we need, this needs to be large 3032 * // enough to hold our array of engines. The kernel will fill out the 3033 * // item.length for us, which is the number of bytes we need. 3034 * // 3035 * // Alternatively a large buffer can be allocated straight away enabling 3036 * // querying in one pass, in which case item.length should contain the 3037 * // length of the provided buffer. 3038 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3039 * if (err) ... 3040 * 3041 * info = calloc(1, item.length); 3042 * // Now that we allocated the required number of bytes, we call the ioctl 3043 * // again, this time with the data_ptr pointing to our newly allocated 3044 * // blob, which the kernel can then populate with info on all engines. 3045 * item.data_ptr = (uintptr_t)&info, 3046 * 3047 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3048 * if (err) ... 3049 * 3050 * // We can now access each engine in the array 3051 * for (i = 0; i < info->num_engines; i++) { 3052 * struct drm_i915_engine_info einfo = info->engines[i]; 3053 * u16 class = einfo.engine.class; 3054 * u16 instance = einfo.engine.instance; 3055 * .... 3056 * } 3057 * 3058 * free(info); 3059 * 3060 * Each of the enumerated engines, apart from being defined by its class and 3061 * instance (see `struct i915_engine_class_instance`), also can have flags and 3062 * capabilities defined as documented in i915_drm.h. 3063 * 3064 * For instance video engines which support HEVC encoding will have the 3065 * `I915_VIDEO_CLASS_CAPABILITY_HEVC` capability bit set. 3066 * 3067 * Engine discovery only fully comes to its own when combined with the new way 3068 * of addressing engines when submitting batch buffers using contexts with 3069 * engine maps configured. 3070 */ 3071 3072 /** 3073 * struct drm_i915_engine_info 3074 * 3075 * Describes one engine and it's capabilities as known to the driver. 3076 */ 3077 struct drm_i915_engine_info { 3078 /** @engine: Engine class and instance. */ 3079 struct i915_engine_class_instance engine; 3080 3081 /** @rsvd0: Reserved field. */ 3082 __u32 rsvd0; 3083 3084 /** @flags: Engine flags. */ 3085 __u64 flags; 3086 #define I915_ENGINE_INFO_HAS_LOGICAL_INSTANCE (1 << 0) 3087 3088 /** @capabilities: Capabilities of this engine. */ 3089 __u64 capabilities; 3090 #define I915_VIDEO_CLASS_CAPABILITY_HEVC (1 << 0) 3091 #define I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC (1 << 1) 3092 3093 /** @logical_instance: Logical instance of engine */ 3094 __u16 logical_instance; 3095 3096 /** @rsvd1: Reserved fields. */ 3097 __u16 rsvd1[3]; 3098 /** @rsvd2: Reserved fields. */ 3099 __u64 rsvd2[3]; 3100 }; 3101 3102 /** 3103 * struct drm_i915_query_engine_info 3104 * 3105 * Engine info query enumerates all engines known to the driver by filling in 3106 * an array of struct drm_i915_engine_info structures. 3107 */ 3108 struct drm_i915_query_engine_info { 3109 /** @num_engines: Number of struct drm_i915_engine_info structs following. */ 3110 __u32 num_engines; 3111 3112 /** @rsvd: MBZ */ 3113 __u32 rsvd[3]; 3114 3115 /** @engines: Marker for drm_i915_engine_info structures. */ 3116 struct drm_i915_engine_info engines[]; 3117 }; 3118 3119 /** 3120 * struct drm_i915_query_perf_config 3121 * 3122 * Data written by the kernel with query %DRM_I915_QUERY_PERF_CONFIG and 3123 * %DRM_I915_QUERY_GEOMETRY_SUBSLICES. 3124 */ 3125 struct drm_i915_query_perf_config { 3126 union { 3127 /** 3128 * @n_configs: 3129 * 3130 * When &drm_i915_query_item.flags == 3131 * %DRM_I915_QUERY_PERF_CONFIG_LIST, i915 sets this fields to 3132 * the number of configurations available. 3133 */ 3134 __u64 n_configs; 3135 3136 /** 3137 * @config: 3138 * 3139 * When &drm_i915_query_item.flags == 3140 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID, i915 will use the 3141 * value in this field as configuration identifier to decide 3142 * what data to write into config_ptr. 3143 */ 3144 __u64 config; 3145 3146 /** 3147 * @uuid: 3148 * 3149 * When &drm_i915_query_item.flags == 3150 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID, i915 will use the 3151 * value in this field as configuration identifier to decide 3152 * what data to write into config_ptr. 3153 * 3154 * String formatted like "%08x-%04x-%04x-%04x-%012x" 3155 */ 3156 char uuid[36]; 3157 }; 3158 3159 /** 3160 * @flags: 3161 * 3162 * Unused for now. Must be cleared to zero. 3163 */ 3164 __u32 flags; 3165 3166 /** 3167 * @data: 3168 * 3169 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_LIST, 3170 * i915 will write an array of __u64 of configuration identifiers. 3171 * 3172 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_DATA, 3173 * i915 will write a struct drm_i915_perf_oa_config. If the following 3174 * fields of struct drm_i915_perf_oa_config are not set to 0, i915 will 3175 * write into the associated pointers the values of submitted when the 3176 * configuration was created : 3177 * 3178 * - &drm_i915_perf_oa_config.n_mux_regs 3179 * - &drm_i915_perf_oa_config.n_boolean_regs 3180 * - &drm_i915_perf_oa_config.n_flex_regs 3181 */ 3182 __u8 data[]; 3183 }; 3184 3185 /** 3186 * enum drm_i915_gem_memory_class - Supported memory classes 3187 */ 3188 enum drm_i915_gem_memory_class { 3189 /** @I915_MEMORY_CLASS_SYSTEM: System memory */ 3190 I915_MEMORY_CLASS_SYSTEM = 0, 3191 /** @I915_MEMORY_CLASS_DEVICE: Device local-memory */ 3192 I915_MEMORY_CLASS_DEVICE, 3193 }; 3194 3195 /** 3196 * struct drm_i915_gem_memory_class_instance - Identify particular memory region 3197 */ 3198 struct drm_i915_gem_memory_class_instance { 3199 /** @memory_class: See enum drm_i915_gem_memory_class */ 3200 __u16 memory_class; 3201 3202 /** @memory_instance: Which instance */ 3203 __u16 memory_instance; 3204 }; 3205 3206 /** 3207 * struct drm_i915_memory_region_info - Describes one region as known to the 3208 * driver. 3209 * 3210 * Note that we reserve some stuff here for potential future work. As an example 3211 * we might want expose the capabilities for a given region, which could include 3212 * things like if the region is CPU mappable/accessible, what are the supported 3213 * mapping types etc. 3214 * 3215 * Note that to extend struct drm_i915_memory_region_info and struct 3216 * drm_i915_query_memory_regions in the future the plan is to do the following: 3217 * 3218 * .. code-block:: C 3219 * 3220 * struct drm_i915_memory_region_info { 3221 * struct drm_i915_gem_memory_class_instance region; 3222 * union { 3223 * __u32 rsvd0; 3224 * __u32 new_thing1; 3225 * }; 3226 * ... 3227 * union { 3228 * __u64 rsvd1[8]; 3229 * struct { 3230 * __u64 new_thing2; 3231 * __u64 new_thing3; 3232 * ... 3233 * }; 3234 * }; 3235 * }; 3236 * 3237 * With this things should remain source compatible between versions for 3238 * userspace, even as we add new fields. 3239 * 3240 * Note this is using both struct drm_i915_query_item and struct drm_i915_query. 3241 * For this new query we are adding the new query id DRM_I915_QUERY_MEMORY_REGIONS 3242 * at &drm_i915_query_item.query_id. 3243 */ 3244 struct drm_i915_memory_region_info { 3245 /** @region: The class:instance pair encoding */ 3246 struct drm_i915_gem_memory_class_instance region; 3247 3248 /** @rsvd0: MBZ */ 3249 __u32 rsvd0; 3250 3251 /** @probed_size: Memory probed by the driver (-1 = unknown) */ 3252 __u64 probed_size; 3253 3254 /** @unallocated_size: Estimate of memory remaining (-1 = unknown) */ 3255 __u64 unallocated_size; 3256 3257 /** @rsvd1: MBZ */ 3258 __u64 rsvd1[8]; 3259 }; 3260 3261 /** 3262 * struct drm_i915_query_memory_regions 3263 * 3264 * The region info query enumerates all regions known to the driver by filling 3265 * in an array of struct drm_i915_memory_region_info structures. 3266 * 3267 * Example for getting the list of supported regions: 3268 * 3269 * .. code-block:: C 3270 * 3271 * struct drm_i915_query_memory_regions *info; 3272 * struct drm_i915_query_item item = { 3273 * .query_id = DRM_I915_QUERY_MEMORY_REGIONS; 3274 * }; 3275 * struct drm_i915_query query = { 3276 * .num_items = 1, 3277 * .items_ptr = (uintptr_t)&item, 3278 * }; 3279 * int err, i; 3280 * 3281 * // First query the size of the blob we need, this needs to be large 3282 * // enough to hold our array of regions. The kernel will fill out the 3283 * // item.length for us, which is the number of bytes we need. 3284 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3285 * if (err) ... 3286 * 3287 * info = calloc(1, item.length); 3288 * // Now that we allocated the required number of bytes, we call the ioctl 3289 * // again, this time with the data_ptr pointing to our newly allocated 3290 * // blob, which the kernel can then populate with the all the region info. 3291 * item.data_ptr = (uintptr_t)&info, 3292 * 3293 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3294 * if (err) ... 3295 * 3296 * // We can now access each region in the array 3297 * for (i = 0; i < info->num_regions; i++) { 3298 * struct drm_i915_memory_region_info mr = info->regions[i]; 3299 * u16 class = mr.region.class; 3300 * u16 instance = mr.region.instance; 3301 * 3302 * .... 3303 * } 3304 * 3305 * free(info); 3306 */ 3307 struct drm_i915_query_memory_regions { 3308 /** @num_regions: Number of supported regions */ 3309 __u32 num_regions; 3310 3311 /** @rsvd: MBZ */ 3312 __u32 rsvd[3]; 3313 3314 /** @regions: Info about each supported region */ 3315 struct drm_i915_memory_region_info regions[]; 3316 }; 3317 3318 /** 3319 * DOC: GuC HWCONFIG blob uAPI 3320 * 3321 * The GuC produces a blob with information about the current device. 3322 * i915 reads this blob from GuC and makes it available via this uAPI. 3323 * 3324 * The format and meaning of the blob content are documented in the 3325 * Programmer's Reference Manual. 3326 */ 3327 3328 /** 3329 * struct drm_i915_gem_create_ext - Existing gem_create behaviour, with added 3330 * extension support using struct i915_user_extension. 3331 * 3332 * Note that in the future we want to have our buffer flags here, at least for 3333 * the stuff that is immutable. Previously we would have two ioctls, one to 3334 * create the object with gem_create, and another to apply various parameters, 3335 * however this creates some ambiguity for the params which are considered 3336 * immutable. Also in general we're phasing out the various SET/GET ioctls. 3337 */ 3338 struct drm_i915_gem_create_ext { 3339 /** 3340 * @size: Requested size for the object. 3341 * 3342 * The (page-aligned) allocated size for the object will be returned. 3343 * 3344 * 3345 * DG2 64K min page size implications: 3346 * 3347 * On discrete platforms, starting from DG2, we have to contend with GTT 3348 * page size restrictions when dealing with I915_MEMORY_CLASS_DEVICE 3349 * objects. Specifically the hardware only supports 64K or larger GTT 3350 * page sizes for such memory. The kernel will already ensure that all 3351 * I915_MEMORY_CLASS_DEVICE memory is allocated using 64K or larger page 3352 * sizes underneath. 3353 * 3354 * Note that the returned size here will always reflect any required 3355 * rounding up done by the kernel, i.e 4K will now become 64K on devices 3356 * such as DG2. 3357 * 3358 * Special DG2 GTT address alignment requirement: 3359 * 3360 * The GTT alignment will also need to be at least 2M for such objects. 3361 * 3362 * Note that due to how the hardware implements 64K GTT page support, we 3363 * have some further complications: 3364 * 3365 * 1) The entire PDE (which covers a 2MB virtual address range), must 3366 * contain only 64K PTEs, i.e mixing 4K and 64K PTEs in the same 3367 * PDE is forbidden by the hardware. 3368 * 3369 * 2) We still need to support 4K PTEs for I915_MEMORY_CLASS_SYSTEM 3370 * objects. 3371 * 3372 * To keep things simple for userland, we mandate that any GTT mappings 3373 * must be aligned to and rounded up to 2MB. The kernel will internally 3374 * pad them out to the next 2MB boundary. As this only wastes virtual 3375 * address space and avoids userland having to copy any needlessly 3376 * complicated PDE sharing scheme (coloring) and only affects DG2, this 3377 * is deemed to be a good compromise. 3378 */ 3379 __u64 size; 3380 /** 3381 * @handle: Returned handle for the object. 3382 * 3383 * Object handles are nonzero. 3384 */ 3385 __u32 handle; 3386 /** @flags: MBZ */ 3387 __u32 flags; 3388 /** 3389 * @extensions: The chain of extensions to apply to this object. 3390 * 3391 * This will be useful in the future when we need to support several 3392 * different extensions, and we need to apply more than one when 3393 * creating the object. See struct i915_user_extension. 3394 * 3395 * If we don't supply any extensions then we get the same old gem_create 3396 * behaviour. 3397 * 3398 * For I915_GEM_CREATE_EXT_MEMORY_REGIONS usage see 3399 * struct drm_i915_gem_create_ext_memory_regions. 3400 * 3401 * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see 3402 * struct drm_i915_gem_create_ext_protected_content. 3403 */ 3404 #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 3405 #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 3406 __u64 extensions; 3407 }; 3408 3409 /** 3410 * struct drm_i915_gem_create_ext_memory_regions - The 3411 * I915_GEM_CREATE_EXT_MEMORY_REGIONS extension. 3412 * 3413 * Set the object with the desired set of placements/regions in priority 3414 * order. Each entry must be unique and supported by the device. 3415 * 3416 * This is provided as an array of struct drm_i915_gem_memory_class_instance, or 3417 * an equivalent layout of class:instance pair encodings. See struct 3418 * drm_i915_query_memory_regions and DRM_I915_QUERY_MEMORY_REGIONS for how to 3419 * query the supported regions for a device. 3420 * 3421 * As an example, on discrete devices, if we wish to set the placement as 3422 * device local-memory we can do something like: 3423 * 3424 * .. code-block:: C 3425 * 3426 * struct drm_i915_gem_memory_class_instance region_lmem = { 3427 * .memory_class = I915_MEMORY_CLASS_DEVICE, 3428 * .memory_instance = 0, 3429 * }; 3430 * struct drm_i915_gem_create_ext_memory_regions regions = { 3431 * .base = { .name = I915_GEM_CREATE_EXT_MEMORY_REGIONS }, 3432 * .regions = (uintptr_t)®ion_lmem, 3433 * .num_regions = 1, 3434 * }; 3435 * struct drm_i915_gem_create_ext create_ext = { 3436 * .size = 16 * PAGE_SIZE, 3437 * .extensions = (uintptr_t)®ions, 3438 * }; 3439 * 3440 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3441 * if (err) ... 3442 * 3443 * At which point we get the object handle in &drm_i915_gem_create_ext.handle, 3444 * along with the final object size in &drm_i915_gem_create_ext.size, which 3445 * should account for any rounding up, if required. 3446 * 3447 * Note that userspace has no means of knowing the current backing region 3448 * for objects where @num_regions is larger than one. The kernel will only 3449 * ensure that the priority order of the @regions array is honoured, either 3450 * when initially placing the object, or when moving memory around due to 3451 * memory pressure 3452 * 3453 * On Flat-CCS capable HW, compression is supported for the objects residing 3454 * in I915_MEMORY_CLASS_DEVICE. When such objects (compressed) have other 3455 * memory class in @regions and migrated (by i915, due to memory 3456 * constraints) to the non I915_MEMORY_CLASS_DEVICE region, then i915 needs to 3457 * decompress the content. But i915 doesn't have the required information to 3458 * decompress the userspace compressed objects. 3459 * 3460 * So i915 supports Flat-CCS, on the objects which can reside only on 3461 * I915_MEMORY_CLASS_DEVICE regions. 3462 */ 3463 struct drm_i915_gem_create_ext_memory_regions { 3464 /** @base: Extension link. See struct i915_user_extension. */ 3465 struct i915_user_extension base; 3466 3467 /** @pad: MBZ */ 3468 __u32 pad; 3469 /** @num_regions: Number of elements in the @regions array. */ 3470 __u32 num_regions; 3471 /** 3472 * @regions: The regions/placements array. 3473 * 3474 * An array of struct drm_i915_gem_memory_class_instance. 3475 */ 3476 __u64 regions; 3477 }; 3478 3479 /** 3480 * struct drm_i915_gem_create_ext_protected_content - The 3481 * I915_OBJECT_PARAM_PROTECTED_CONTENT extension. 3482 * 3483 * If this extension is provided, buffer contents are expected to be protected 3484 * by PXP encryption and require decryption for scan out and processing. This 3485 * is only possible on platforms that have PXP enabled, on all other scenarios 3486 * using this extension will cause the ioctl to fail and return -ENODEV. The 3487 * flags parameter is reserved for future expansion and must currently be set 3488 * to zero. 3489 * 3490 * The buffer contents are considered invalid after a PXP session teardown. 3491 * 3492 * The encryption is guaranteed to be processed correctly only if the object 3493 * is submitted with a context created using the 3494 * I915_CONTEXT_PARAM_PROTECTED_CONTENT flag. This will also enable extra checks 3495 * at submission time on the validity of the objects involved. 3496 * 3497 * Below is an example on how to create a protected object: 3498 * 3499 * .. code-block:: C 3500 * 3501 * struct drm_i915_gem_create_ext_protected_content protected_ext = { 3502 * .base = { .name = I915_GEM_CREATE_EXT_PROTECTED_CONTENT }, 3503 * .flags = 0, 3504 * }; 3505 * struct drm_i915_gem_create_ext create_ext = { 3506 * .size = PAGE_SIZE, 3507 * .extensions = (uintptr_t)&protected_ext, 3508 * }; 3509 * 3510 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3511 * if (err) ... 3512 */ 3513 struct drm_i915_gem_create_ext_protected_content { 3514 /** @base: Extension link. See struct i915_user_extension. */ 3515 struct i915_user_extension base; 3516 /** @flags: reserved for future usage, currently MBZ */ 3517 __u32 flags; 3518 }; 3519 3520 /* ID of the protected content session managed by i915 when PXP is active */ 3521 #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf 3522 3523 #if defined(__cplusplus) 3524 } 3525 #endif 3526 3527 #endif /* _UAPI_I915_DRM_H_ */ 3528