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 _I915_DRM_H_ 28 #define _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 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 *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 *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 *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 *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 /** 755 * struct drm_i915_getparam - Driver parameter query structure. 756 */ 757 struct drm_i915_getparam { 758 /** @param: Driver parameter to query. */ 759 __s32 param; 760 761 /** 762 * @value: Address of memory where queried value should be put. 763 * 764 * WARNING: Using pointers instead of fixed-size u64 means we need to write 765 * compat32 code. Don't repeat this mistake. 766 */ 767 int *value; 768 }; 769 770 /** 771 * typedef drm_i915_getparam_t - Driver parameter query structure. 772 * See struct drm_i915_getparam. 773 */ 774 typedef struct drm_i915_getparam drm_i915_getparam_t; 775 776 /* Ioctl to set kernel params: 777 */ 778 #define I915_SETPARAM_USE_MI_BATCHBUFFER_START 1 779 #define I915_SETPARAM_TEX_LRU_LOG_GRANULARITY 2 780 #define I915_SETPARAM_ALLOW_BATCHBUFFER 3 781 #define I915_SETPARAM_NUM_USED_FENCES 4 782 /* Must be kept compact -- no holes */ 783 784 typedef struct drm_i915_setparam { 785 int param; 786 int value; 787 } drm_i915_setparam_t; 788 789 /* A memory manager for regions of shared memory: 790 */ 791 #define I915_MEM_REGION_AGP 1 792 793 typedef struct drm_i915_mem_alloc { 794 int region; 795 int alignment; 796 int size; 797 int *region_offset; /* offset from start of fb or agp */ 798 } drm_i915_mem_alloc_t; 799 800 typedef struct drm_i915_mem_free { 801 int region; 802 int region_offset; 803 } drm_i915_mem_free_t; 804 805 typedef struct drm_i915_mem_init_heap { 806 int region; 807 int size; 808 int start; 809 } drm_i915_mem_init_heap_t; 810 811 /* Allow memory manager to be torn down and re-initialized (eg on 812 * rotate): 813 */ 814 typedef struct drm_i915_mem_destroy_heap { 815 int region; 816 } drm_i915_mem_destroy_heap_t; 817 818 /* Allow X server to configure which pipes to monitor for vblank signals 819 */ 820 #define DRM_I915_VBLANK_PIPE_A 1 821 #define DRM_I915_VBLANK_PIPE_B 2 822 823 typedef struct drm_i915_vblank_pipe { 824 int pipe; 825 } drm_i915_vblank_pipe_t; 826 827 /* Schedule buffer swap at given vertical blank: 828 */ 829 typedef struct drm_i915_vblank_swap { 830 drm_drawable_t drawable; 831 enum drm_vblank_seq_type seqtype; 832 unsigned int sequence; 833 } drm_i915_vblank_swap_t; 834 835 typedef struct drm_i915_hws_addr { 836 __u64 addr; 837 } drm_i915_hws_addr_t; 838 839 struct drm_i915_gem_init { 840 /** 841 * Beginning offset in the GTT to be managed by the DRM memory 842 * manager. 843 */ 844 __u64 gtt_start; 845 /** 846 * Ending offset in the GTT to be managed by the DRM memory 847 * manager. 848 */ 849 __u64 gtt_end; 850 }; 851 852 struct drm_i915_gem_create { 853 /** 854 * Requested size for the object. 855 * 856 * The (page-aligned) allocated size for the object will be returned. 857 */ 858 __u64 size; 859 /** 860 * Returned handle for the object. 861 * 862 * Object handles are nonzero. 863 */ 864 __u32 handle; 865 __u32 pad; 866 }; 867 868 struct drm_i915_gem_pread { 869 /** Handle for the object being read. */ 870 __u32 handle; 871 __u32 pad; 872 /** Offset into the object to read from */ 873 __u64 offset; 874 /** Length of data to read */ 875 __u64 size; 876 /** 877 * Pointer to write the data into. 878 * 879 * This is a fixed-size type for 32/64 compatibility. 880 */ 881 __u64 data_ptr; 882 }; 883 884 struct drm_i915_gem_pwrite { 885 /** Handle for the object being written to. */ 886 __u32 handle; 887 __u32 pad; 888 /** Offset into the object to write to */ 889 __u64 offset; 890 /** Length of data to write */ 891 __u64 size; 892 /** 893 * Pointer to read the data from. 894 * 895 * This is a fixed-size type for 32/64 compatibility. 896 */ 897 __u64 data_ptr; 898 }; 899 900 struct drm_i915_gem_mmap { 901 /** Handle for the object being mapped. */ 902 __u32 handle; 903 __u32 pad; 904 /** Offset in the object to map. */ 905 __u64 offset; 906 /** 907 * Length of data to map. 908 * 909 * The value will be page-aligned. 910 */ 911 __u64 size; 912 /** 913 * Returned pointer the data was mapped at. 914 * 915 * This is a fixed-size type for 32/64 compatibility. 916 */ 917 __u64 addr_ptr; 918 919 /** 920 * Flags for extended behaviour. 921 * 922 * Added in version 2. 923 */ 924 __u64 flags; 925 #define I915_MMAP_WC 0x1 926 }; 927 928 struct drm_i915_gem_mmap_gtt { 929 /** Handle for the object being mapped. */ 930 __u32 handle; 931 __u32 pad; 932 /** 933 * Fake offset to use for subsequent mmap call 934 * 935 * This is a fixed-size type for 32/64 compatibility. 936 */ 937 __u64 offset; 938 }; 939 940 /** 941 * struct drm_i915_gem_mmap_offset - Retrieve an offset so we can mmap this buffer object. 942 * 943 * This struct is passed as argument to the `DRM_IOCTL_I915_GEM_MMAP_OFFSET` ioctl, 944 * and is used to retrieve the fake offset to mmap an object specified by &handle. 945 * 946 * The legacy way of using `DRM_IOCTL_I915_GEM_MMAP` is removed on gen12+. 947 * `DRM_IOCTL_I915_GEM_MMAP_GTT` is an older supported alias to this struct, but will behave 948 * as setting the &extensions to 0, and &flags to `I915_MMAP_OFFSET_GTT`. 949 */ 950 struct drm_i915_gem_mmap_offset { 951 /** @handle: Handle for the object being mapped. */ 952 __u32 handle; 953 /** @pad: Must be zero */ 954 __u32 pad; 955 /** 956 * @offset: The fake offset to use for subsequent mmap call 957 * 958 * This is a fixed-size type for 32/64 compatibility. 959 */ 960 __u64 offset; 961 962 /** 963 * @flags: Flags for extended behaviour. 964 * 965 * It is mandatory that one of the `MMAP_OFFSET` types 966 * should be included: 967 * 968 * - `I915_MMAP_OFFSET_GTT`: Use mmap with the object bound to GTT. (Write-Combined) 969 * - `I915_MMAP_OFFSET_WC`: Use Write-Combined caching. 970 * - `I915_MMAP_OFFSET_WB`: Use Write-Back caching. 971 * - `I915_MMAP_OFFSET_FIXED`: Use object placement to determine caching. 972 * 973 * On devices with local memory `I915_MMAP_OFFSET_FIXED` is the only valid 974 * type. On devices without local memory, this caching mode is invalid. 975 * 976 * As caching mode when specifying `I915_MMAP_OFFSET_FIXED`, WC or WB will 977 * be used, depending on the object placement on creation. WB will be used 978 * when the object can only exist in system memory, WC otherwise. 979 */ 980 __u64 flags; 981 982 #define I915_MMAP_OFFSET_GTT 0 983 #define I915_MMAP_OFFSET_WC 1 984 #define I915_MMAP_OFFSET_WB 2 985 #define I915_MMAP_OFFSET_UC 3 986 #define I915_MMAP_OFFSET_FIXED 4 987 988 /** 989 * @extensions: Zero-terminated chain of extensions. 990 * 991 * No current extensions defined; mbz. 992 */ 993 __u64 extensions; 994 }; 995 996 /** 997 * struct drm_i915_gem_set_domain - Adjust the objects write or read domain, in 998 * preparation for accessing the pages via some CPU domain. 999 * 1000 * Specifying a new write or read domain will flush the object out of the 1001 * previous domain(if required), before then updating the objects domain 1002 * tracking with the new domain. 1003 * 1004 * Note this might involve waiting for the object first if it is still active on 1005 * the GPU. 1006 * 1007 * Supported values for @read_domains and @write_domain: 1008 * 1009 * - I915_GEM_DOMAIN_WC: Uncached write-combined domain 1010 * - I915_GEM_DOMAIN_CPU: CPU cache domain 1011 * - I915_GEM_DOMAIN_GTT: Mappable aperture domain 1012 * 1013 * All other domains are rejected. 1014 * 1015 * Note that for discrete, starting from DG1, this is no longer supported, and 1016 * is instead rejected. On such platforms the CPU domain is effectively static, 1017 * where we also only support a single &drm_i915_gem_mmap_offset cache mode, 1018 * which can't be set explicitly and instead depends on the object placements, 1019 * as per the below. 1020 * 1021 * Implicit caching rules, starting from DG1: 1022 * 1023 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1024 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1025 * mapped as write-combined only. 1026 * 1027 * - Everything else is always allocated and mapped as write-back, with the 1028 * guarantee that everything is also coherent with the GPU. 1029 * 1030 * Note that this is likely to change in the future again, where we might need 1031 * more flexibility on future devices, so making this all explicit as part of a 1032 * new &drm_i915_gem_create_ext extension is probable. 1033 */ 1034 struct drm_i915_gem_set_domain { 1035 /** @handle: Handle for the object. */ 1036 __u32 handle; 1037 1038 /** @read_domains: New read domains. */ 1039 __u32 read_domains; 1040 1041 /** 1042 * @write_domain: New write domain. 1043 * 1044 * Note that having something in the write domain implies it's in the 1045 * read domain, and only that read domain. 1046 */ 1047 __u32 write_domain; 1048 }; 1049 1050 struct drm_i915_gem_sw_finish { 1051 /** Handle for the object */ 1052 __u32 handle; 1053 }; 1054 1055 struct drm_i915_gem_relocation_entry { 1056 /** 1057 * Handle of the buffer being pointed to by this relocation entry. 1058 * 1059 * It's appealing to make this be an index into the mm_validate_entry 1060 * list to refer to the buffer, but this allows the driver to create 1061 * a relocation list for state buffers and not re-write it per 1062 * exec using the buffer. 1063 */ 1064 __u32 target_handle; 1065 1066 /** 1067 * Value to be added to the offset of the target buffer to make up 1068 * the relocation entry. 1069 */ 1070 __u32 delta; 1071 1072 /** Offset in the buffer the relocation entry will be written into */ 1073 __u64 offset; 1074 1075 /** 1076 * Offset value of the target buffer that the relocation entry was last 1077 * written as. 1078 * 1079 * If the buffer has the same offset as last time, we can skip syncing 1080 * and writing the relocation. This value is written back out by 1081 * the execbuffer ioctl when the relocation is written. 1082 */ 1083 __u64 presumed_offset; 1084 1085 /** 1086 * Target memory domains read by this operation. 1087 */ 1088 __u32 read_domains; 1089 1090 /** 1091 * Target memory domains written by this operation. 1092 * 1093 * Note that only one domain may be written by the whole 1094 * execbuffer operation, so that where there are conflicts, 1095 * the application will get -EINVAL back. 1096 */ 1097 __u32 write_domain; 1098 }; 1099 1100 /** @{ 1101 * Intel memory domains 1102 * 1103 * Most of these just align with the various caches in 1104 * the system and are used to flush and invalidate as 1105 * objects end up cached in different domains. 1106 */ 1107 /** CPU cache */ 1108 #define I915_GEM_DOMAIN_CPU 0x00000001 1109 /** Render cache, used by 2D and 3D drawing */ 1110 #define I915_GEM_DOMAIN_RENDER 0x00000002 1111 /** Sampler cache, used by texture engine */ 1112 #define I915_GEM_DOMAIN_SAMPLER 0x00000004 1113 /** Command queue, used to load batch buffers */ 1114 #define I915_GEM_DOMAIN_COMMAND 0x00000008 1115 /** Instruction cache, used by shader programs */ 1116 #define I915_GEM_DOMAIN_INSTRUCTION 0x00000010 1117 /** Vertex address cache */ 1118 #define I915_GEM_DOMAIN_VERTEX 0x00000020 1119 /** GTT domain - aperture and scanout */ 1120 #define I915_GEM_DOMAIN_GTT 0x00000040 1121 /** WC domain - uncached access */ 1122 #define I915_GEM_DOMAIN_WC 0x00000080 1123 /** @} */ 1124 1125 struct drm_i915_gem_exec_object { 1126 /** 1127 * User's handle for a buffer to be bound into the GTT for this 1128 * operation. 1129 */ 1130 __u32 handle; 1131 1132 /** Number of relocations to be performed on this buffer */ 1133 __u32 relocation_count; 1134 /** 1135 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1136 * the relocations to be performed in this buffer. 1137 */ 1138 __u64 relocs_ptr; 1139 1140 /** Required alignment in graphics aperture */ 1141 __u64 alignment; 1142 1143 /** 1144 * Returned value of the updated offset of the object, for future 1145 * presumed_offset writes. 1146 */ 1147 __u64 offset; 1148 }; 1149 1150 /* DRM_IOCTL_I915_GEM_EXECBUFFER was removed in Linux 5.13 */ 1151 struct drm_i915_gem_execbuffer { 1152 /** 1153 * List of buffers to be validated with their relocations to be 1154 * performend on them. 1155 * 1156 * This is a pointer to an array of struct drm_i915_gem_validate_entry. 1157 * 1158 * These buffers must be listed in an order such that all relocations 1159 * a buffer is performing refer to buffers that have already appeared 1160 * in the validate list. 1161 */ 1162 __u64 buffers_ptr; 1163 __u32 buffer_count; 1164 1165 /** Offset in the batchbuffer to start execution from. */ 1166 __u32 batch_start_offset; 1167 /** Bytes used in batchbuffer from batch_start_offset */ 1168 __u32 batch_len; 1169 __u32 DR1; 1170 __u32 DR4; 1171 __u32 num_cliprects; 1172 /** This is a struct drm_clip_rect *cliprects */ 1173 __u64 cliprects_ptr; 1174 }; 1175 1176 struct drm_i915_gem_exec_object2 { 1177 /** 1178 * User's handle for a buffer to be bound into the GTT for this 1179 * operation. 1180 */ 1181 __u32 handle; 1182 1183 /** Number of relocations to be performed on this buffer */ 1184 __u32 relocation_count; 1185 /** 1186 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1187 * the relocations to be performed in this buffer. 1188 */ 1189 __u64 relocs_ptr; 1190 1191 /** Required alignment in graphics aperture */ 1192 __u64 alignment; 1193 1194 /** 1195 * When the EXEC_OBJECT_PINNED flag is specified this is populated by 1196 * the user with the GTT offset at which this object will be pinned. 1197 * 1198 * When the I915_EXEC_NO_RELOC flag is specified this must contain the 1199 * presumed_offset of the object. 1200 * 1201 * During execbuffer2 the kernel populates it with the value of the 1202 * current GTT offset of the object, for future presumed_offset writes. 1203 * 1204 * See struct drm_i915_gem_create_ext for the rules when dealing with 1205 * alignment restrictions with I915_MEMORY_CLASS_DEVICE, on devices with 1206 * minimum page sizes, like DG2. 1207 */ 1208 __u64 offset; 1209 1210 #define EXEC_OBJECT_NEEDS_FENCE (1<<0) 1211 #define EXEC_OBJECT_NEEDS_GTT (1<<1) 1212 #define EXEC_OBJECT_WRITE (1<<2) 1213 #define EXEC_OBJECT_SUPPORTS_48B_ADDRESS (1<<3) 1214 #define EXEC_OBJECT_PINNED (1<<4) 1215 #define EXEC_OBJECT_PAD_TO_SIZE (1<<5) 1216 /* The kernel implicitly tracks GPU activity on all GEM objects, and 1217 * synchronises operations with outstanding rendering. This includes 1218 * rendering on other devices if exported via dma-buf. However, sometimes 1219 * this tracking is too coarse and the user knows better. For example, 1220 * if the object is split into non-overlapping ranges shared between different 1221 * clients or engines (i.e. suballocating objects), the implicit tracking 1222 * by kernel assumes that each operation affects the whole object rather 1223 * than an individual range, causing needless synchronisation between clients. 1224 * The kernel will also forgo any CPU cache flushes prior to rendering from 1225 * the object as the client is expected to be also handling such domain 1226 * tracking. 1227 * 1228 * The kernel maintains the implicit tracking in order to manage resources 1229 * used by the GPU - this flag only disables the synchronisation prior to 1230 * rendering with this object in this execbuf. 1231 * 1232 * Opting out of implicit synhronisation requires the user to do its own 1233 * explicit tracking to avoid rendering corruption. See, for example, 1234 * I915_PARAM_HAS_EXEC_FENCE to order execbufs and execute them asynchronously. 1235 */ 1236 #define EXEC_OBJECT_ASYNC (1<<6) 1237 /* Request that the contents of this execobject be copied into the error 1238 * state upon a GPU hang involving this batch for post-mortem debugging. 1239 * These buffers are recorded in no particular order as "user" in 1240 * /sys/class/drm/cardN/error. Query I915_PARAM_HAS_EXEC_CAPTURE to see 1241 * if the kernel supports this flag. 1242 */ 1243 #define EXEC_OBJECT_CAPTURE (1<<7) 1244 /* All remaining bits are MBZ and RESERVED FOR FUTURE USE */ 1245 #define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_CAPTURE<<1) 1246 __u64 flags; 1247 1248 union { 1249 __u64 rsvd1; 1250 __u64 pad_to_size; 1251 }; 1252 __u64 rsvd2; 1253 }; 1254 1255 /** 1256 * struct drm_i915_gem_exec_fence - An input or output fence for the execbuf 1257 * ioctl. 1258 * 1259 * The request will wait for input fence to signal before submission. 1260 * 1261 * The returned output fence will be signaled after the completion of the 1262 * request. 1263 */ 1264 struct drm_i915_gem_exec_fence { 1265 /** @handle: User's handle for a drm_syncobj to wait on or signal. */ 1266 __u32 handle; 1267 1268 /** 1269 * @flags: Supported flags are: 1270 * 1271 * I915_EXEC_FENCE_WAIT: 1272 * Wait for the input fence before request submission. 1273 * 1274 * I915_EXEC_FENCE_SIGNAL: 1275 * Return request completion fence as output 1276 */ 1277 __u32 flags; 1278 #define I915_EXEC_FENCE_WAIT (1<<0) 1279 #define I915_EXEC_FENCE_SIGNAL (1<<1) 1280 #define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1)) 1281 }; 1282 1283 /** 1284 * struct drm_i915_gem_execbuffer_ext_timeline_fences - Timeline fences 1285 * for execbuf ioctl. 1286 * 1287 * This structure describes an array of drm_syncobj and associated points for 1288 * timeline variants of drm_syncobj. It is invalid to append this structure to 1289 * the execbuf if I915_EXEC_FENCE_ARRAY is set. 1290 */ 1291 struct drm_i915_gem_execbuffer_ext_timeline_fences { 1292 #define DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES 0 1293 /** @base: Extension link. See struct i915_user_extension. */ 1294 struct i915_user_extension base; 1295 1296 /** 1297 * @fence_count: Number of elements in the @handles_ptr & @value_ptr 1298 * arrays. 1299 */ 1300 __u64 fence_count; 1301 1302 /** 1303 * @handles_ptr: Pointer to an array of struct drm_i915_gem_exec_fence 1304 * of length @fence_count. 1305 */ 1306 __u64 handles_ptr; 1307 1308 /** 1309 * @values_ptr: Pointer to an array of u64 values of length 1310 * @fence_count. 1311 * Values must be 0 for a binary drm_syncobj. A Value of 0 for a 1312 * timeline drm_syncobj is invalid as it turns a drm_syncobj into a 1313 * binary one. 1314 */ 1315 __u64 values_ptr; 1316 }; 1317 1318 /** 1319 * struct drm_i915_gem_execbuffer2 - Structure for DRM_I915_GEM_EXECBUFFER2 1320 * ioctl. 1321 */ 1322 struct drm_i915_gem_execbuffer2 { 1323 /** @buffers_ptr: Pointer to a list of gem_exec_object2 structs */ 1324 __u64 buffers_ptr; 1325 1326 /** @buffer_count: Number of elements in @buffers_ptr array */ 1327 __u32 buffer_count; 1328 1329 /** 1330 * @batch_start_offset: Offset in the batchbuffer to start execution 1331 * from. 1332 */ 1333 __u32 batch_start_offset; 1334 1335 /** 1336 * @batch_len: Length in bytes of the batch buffer, starting from the 1337 * @batch_start_offset. If 0, length is assumed to be the batch buffer 1338 * object size. 1339 */ 1340 __u32 batch_len; 1341 1342 /** @DR1: deprecated */ 1343 __u32 DR1; 1344 1345 /** @DR4: deprecated */ 1346 __u32 DR4; 1347 1348 /** @num_cliprects: See @cliprects_ptr */ 1349 __u32 num_cliprects; 1350 1351 /** 1352 * @cliprects_ptr: Kernel clipping was a DRI1 misfeature. 1353 * 1354 * It is invalid to use this field if I915_EXEC_FENCE_ARRAY or 1355 * I915_EXEC_USE_EXTENSIONS flags are not set. 1356 * 1357 * If I915_EXEC_FENCE_ARRAY is set, then this is a pointer to an array 1358 * of &drm_i915_gem_exec_fence and @num_cliprects is the length of the 1359 * array. 1360 * 1361 * If I915_EXEC_USE_EXTENSIONS is set, then this is a pointer to a 1362 * single &i915_user_extension and num_cliprects is 0. 1363 */ 1364 __u64 cliprects_ptr; 1365 1366 /** @flags: Execbuf flags */ 1367 __u64 flags; 1368 #define I915_EXEC_RING_MASK (0x3f) 1369 #define I915_EXEC_DEFAULT (0<<0) 1370 #define I915_EXEC_RENDER (1<<0) 1371 #define I915_EXEC_BSD (2<<0) 1372 #define I915_EXEC_BLT (3<<0) 1373 #define I915_EXEC_VEBOX (4<<0) 1374 1375 /* Used for switching the constants addressing mode on gen4+ RENDER ring. 1376 * Gen6+ only supports relative addressing to dynamic state (default) and 1377 * absolute addressing. 1378 * 1379 * These flags are ignored for the BSD and BLT rings. 1380 */ 1381 #define I915_EXEC_CONSTANTS_MASK (3<<6) 1382 #define I915_EXEC_CONSTANTS_REL_GENERAL (0<<6) /* default */ 1383 #define I915_EXEC_CONSTANTS_ABSOLUTE (1<<6) 1384 #define I915_EXEC_CONSTANTS_REL_SURFACE (2<<6) /* gen4/5 only */ 1385 1386 /** Resets the SO write offset registers for transform feedback on gen7. */ 1387 #define I915_EXEC_GEN7_SOL_RESET (1<<8) 1388 1389 /** Request a privileged ("secure") batch buffer. Note only available for 1390 * DRM_ROOT_ONLY | DRM_MASTER processes. 1391 */ 1392 #define I915_EXEC_SECURE (1<<9) 1393 1394 /** Inform the kernel that the batch is and will always be pinned. This 1395 * negates the requirement for a workaround to be performed to avoid 1396 * an incoherent CS (such as can be found on 830/845). If this flag is 1397 * not passed, the kernel will endeavour to make sure the batch is 1398 * coherent with the CS before execution. If this flag is passed, 1399 * userspace assumes the responsibility for ensuring the same. 1400 */ 1401 #define I915_EXEC_IS_PINNED (1<<10) 1402 1403 /** Provide a hint to the kernel that the command stream and auxiliary 1404 * state buffers already holds the correct presumed addresses and so the 1405 * relocation process may be skipped if no buffers need to be moved in 1406 * preparation for the execbuffer. 1407 */ 1408 #define I915_EXEC_NO_RELOC (1<<11) 1409 1410 /** Use the reloc.handle as an index into the exec object array rather 1411 * than as the per-file handle. 1412 */ 1413 #define I915_EXEC_HANDLE_LUT (1<<12) 1414 1415 /** Used for switching BSD rings on the platforms with two BSD rings */ 1416 #define I915_EXEC_BSD_SHIFT (13) 1417 #define I915_EXEC_BSD_MASK (3 << I915_EXEC_BSD_SHIFT) 1418 /* default ping-pong mode */ 1419 #define I915_EXEC_BSD_DEFAULT (0 << I915_EXEC_BSD_SHIFT) 1420 #define I915_EXEC_BSD_RING1 (1 << I915_EXEC_BSD_SHIFT) 1421 #define I915_EXEC_BSD_RING2 (2 << I915_EXEC_BSD_SHIFT) 1422 1423 /** Tell the kernel that the batchbuffer is processed by 1424 * the resource streamer. 1425 */ 1426 #define I915_EXEC_RESOURCE_STREAMER (1<<15) 1427 1428 /* Setting I915_EXEC_FENCE_IN implies that lower_32_bits(rsvd2) represent 1429 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1430 * the batch. 1431 * 1432 * Returns -EINVAL if the sync_file fd cannot be found. 1433 */ 1434 #define I915_EXEC_FENCE_IN (1<<16) 1435 1436 /* Setting I915_EXEC_FENCE_OUT causes the ioctl to return a sync_file fd 1437 * in the upper_32_bits(rsvd2) upon success. Ownership of the fd is given 1438 * to the caller, and it should be close() after use. (The fd is a regular 1439 * file descriptor and will be cleaned up on process termination. It holds 1440 * a reference to the request, but nothing else.) 1441 * 1442 * The sync_file fd can be combined with other sync_file and passed either 1443 * to execbuf using I915_EXEC_FENCE_IN, to atomic KMS ioctls (so that a flip 1444 * will only occur after this request completes), or to other devices. 1445 * 1446 * Using I915_EXEC_FENCE_OUT requires use of 1447 * DRM_IOCTL_I915_GEM_EXECBUFFER2_WR ioctl so that the result is written 1448 * back to userspace. Failure to do so will cause the out-fence to always 1449 * be reported as zero, and the real fence fd to be leaked. 1450 */ 1451 #define I915_EXEC_FENCE_OUT (1<<17) 1452 1453 /* 1454 * Traditionally the execbuf ioctl has only considered the final element in 1455 * the execobject[] to be the executable batch. Often though, the client 1456 * will known the batch object prior to construction and being able to place 1457 * it into the execobject[] array first can simplify the relocation tracking. 1458 * Setting I915_EXEC_BATCH_FIRST tells execbuf to use element 0 of the 1459 * execobject[] as the * batch instead (the default is to use the last 1460 * element). 1461 */ 1462 #define I915_EXEC_BATCH_FIRST (1<<18) 1463 1464 /* Setting I915_FENCE_ARRAY implies that num_cliprects and cliprects_ptr 1465 * define an array of i915_gem_exec_fence structures which specify a set of 1466 * dma fences to wait upon or signal. 1467 */ 1468 #define I915_EXEC_FENCE_ARRAY (1<<19) 1469 1470 /* 1471 * Setting I915_EXEC_FENCE_SUBMIT implies that lower_32_bits(rsvd2) represent 1472 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1473 * the batch. 1474 * 1475 * Returns -EINVAL if the sync_file fd cannot be found. 1476 */ 1477 #define I915_EXEC_FENCE_SUBMIT (1 << 20) 1478 1479 /* 1480 * Setting I915_EXEC_USE_EXTENSIONS implies that 1481 * drm_i915_gem_execbuffer2.cliprects_ptr is treated as a pointer to an linked 1482 * list of i915_user_extension. Each i915_user_extension node is the base of a 1483 * larger structure. The list of supported structures are listed in the 1484 * drm_i915_gem_execbuffer_ext enum. 1485 */ 1486 #define I915_EXEC_USE_EXTENSIONS (1 << 21) 1487 #define __I915_EXEC_UNKNOWN_FLAGS (-(I915_EXEC_USE_EXTENSIONS << 1)) 1488 1489 /** @rsvd1: Context id */ 1490 __u64 rsvd1; 1491 1492 /** 1493 * @rsvd2: in and out sync_file file descriptors. 1494 * 1495 * When I915_EXEC_FENCE_IN or I915_EXEC_FENCE_SUBMIT flag is set, the 1496 * lower 32 bits of this field will have the in sync_file fd (input). 1497 * 1498 * When I915_EXEC_FENCE_OUT flag is set, the upper 32 bits of this 1499 * field will have the out sync_file fd (output). 1500 */ 1501 __u64 rsvd2; 1502 }; 1503 1504 #define I915_EXEC_CONTEXT_ID_MASK (0xffffffff) 1505 #define i915_execbuffer2_set_context_id(eb2, context) \ 1506 (eb2).rsvd1 = context & I915_EXEC_CONTEXT_ID_MASK 1507 #define i915_execbuffer2_get_context_id(eb2) \ 1508 ((eb2).rsvd1 & I915_EXEC_CONTEXT_ID_MASK) 1509 1510 struct drm_i915_gem_pin { 1511 /** Handle of the buffer to be pinned. */ 1512 __u32 handle; 1513 __u32 pad; 1514 1515 /** alignment required within the aperture */ 1516 __u64 alignment; 1517 1518 /** Returned GTT offset of the buffer. */ 1519 __u64 offset; 1520 }; 1521 1522 struct drm_i915_gem_unpin { 1523 /** Handle of the buffer to be unpinned. */ 1524 __u32 handle; 1525 __u32 pad; 1526 }; 1527 1528 struct drm_i915_gem_busy { 1529 /** Handle of the buffer to check for busy */ 1530 __u32 handle; 1531 1532 /** Return busy status 1533 * 1534 * A return of 0 implies that the object is idle (after 1535 * having flushed any pending activity), and a non-zero return that 1536 * the object is still in-flight on the GPU. (The GPU has not yet 1537 * signaled completion for all pending requests that reference the 1538 * object.) An object is guaranteed to become idle eventually (so 1539 * long as no new GPU commands are executed upon it). Due to the 1540 * asynchronous nature of the hardware, an object reported 1541 * as busy may become idle before the ioctl is completed. 1542 * 1543 * Furthermore, if the object is busy, which engine is busy is only 1544 * provided as a guide and only indirectly by reporting its class 1545 * (there may be more than one engine in each class). There are race 1546 * conditions which prevent the report of which engines are busy from 1547 * being always accurate. However, the converse is not true. If the 1548 * object is idle, the result of the ioctl, that all engines are idle, 1549 * is accurate. 1550 * 1551 * The returned dword is split into two fields to indicate both 1552 * the engine classess on which the object is being read, and the 1553 * engine class on which it is currently being written (if any). 1554 * 1555 * The low word (bits 0:15) indicate if the object is being written 1556 * to by any engine (there can only be one, as the GEM implicit 1557 * synchronisation rules force writes to be serialised). Only the 1558 * engine class (offset by 1, I915_ENGINE_CLASS_RENDER is reported as 1559 * 1 not 0 etc) for the last write is reported. 1560 * 1561 * The high word (bits 16:31) are a bitmask of which engines classes 1562 * are currently reading from the object. Multiple engines may be 1563 * reading from the object simultaneously. 1564 * 1565 * The value of each engine class is the same as specified in the 1566 * I915_CONTEXT_PARAM_ENGINES context parameter and via perf, i.e. 1567 * I915_ENGINE_CLASS_RENDER, I915_ENGINE_CLASS_COPY, etc. 1568 * Some hardware may have parallel execution engines, e.g. multiple 1569 * media engines, which are mapped to the same class identifier and so 1570 * are not separately reported for busyness. 1571 * 1572 * Caveat emptor: 1573 * Only the boolean result of this query is reliable; that is whether 1574 * the object is idle or busy. The report of which engines are busy 1575 * should be only used as a heuristic. 1576 */ 1577 __u32 busy; 1578 }; 1579 1580 /** 1581 * struct drm_i915_gem_caching - Set or get the caching for given object 1582 * handle. 1583 * 1584 * Allow userspace to control the GTT caching bits for a given object when the 1585 * object is later mapped through the ppGTT(or GGTT on older platforms lacking 1586 * ppGTT support, or if the object is used for scanout). Note that this might 1587 * require unbinding the object from the GTT first, if its current caching value 1588 * doesn't match. 1589 * 1590 * Note that this all changes on discrete platforms, starting from DG1, the 1591 * set/get caching is no longer supported, and is now rejected. Instead the CPU 1592 * caching attributes(WB vs WC) will become an immutable creation time property 1593 * for the object, along with the GTT caching level. For now we don't expose any 1594 * new uAPI for this, instead on DG1 this is all implicit, although this largely 1595 * shouldn't matter since DG1 is coherent by default(without any way of 1596 * controlling it). 1597 * 1598 * Implicit caching rules, starting from DG1: 1599 * 1600 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1601 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1602 * mapped as write-combined only. 1603 * 1604 * - Everything else is always allocated and mapped as write-back, with the 1605 * guarantee that everything is also coherent with the GPU. 1606 * 1607 * Note that this is likely to change in the future again, where we might need 1608 * more flexibility on future devices, so making this all explicit as part of a 1609 * new &drm_i915_gem_create_ext extension is probable. 1610 * 1611 * Side note: Part of the reason for this is that changing the at-allocation-time CPU 1612 * caching attributes for the pages might be required(and is expensive) if we 1613 * need to then CPU map the pages later with different caching attributes. This 1614 * inconsistent caching behaviour, while supported on x86, is not universally 1615 * supported on other architectures. So for simplicity we opt for setting 1616 * everything at creation time, whilst also making it immutable, on discrete 1617 * platforms. 1618 */ 1619 struct drm_i915_gem_caching { 1620 /** 1621 * @handle: Handle of the buffer to set/get the caching level. 1622 */ 1623 __u32 handle; 1624 1625 /** 1626 * @caching: The GTT caching level to apply or possible return value. 1627 * 1628 * The supported @caching values: 1629 * 1630 * I915_CACHING_NONE: 1631 * 1632 * GPU access is not coherent with CPU caches. Default for machines 1633 * without an LLC. This means manual flushing might be needed, if we 1634 * want GPU access to be coherent. 1635 * 1636 * I915_CACHING_CACHED: 1637 * 1638 * GPU access is coherent with CPU caches and furthermore the data is 1639 * cached in last-level caches shared between CPU cores and the GPU GT. 1640 * 1641 * I915_CACHING_DISPLAY: 1642 * 1643 * Special GPU caching mode which is coherent with the scanout engines. 1644 * Transparently falls back to I915_CACHING_NONE on platforms where no 1645 * special cache mode (like write-through or gfdt flushing) is 1646 * available. The kernel automatically sets this mode when using a 1647 * buffer as a scanout target. Userspace can manually set this mode to 1648 * avoid a costly stall and clflush in the hotpath of drawing the first 1649 * frame. 1650 */ 1651 #define I915_CACHING_NONE 0 1652 #define I915_CACHING_CACHED 1 1653 #define I915_CACHING_DISPLAY 2 1654 __u32 caching; 1655 }; 1656 1657 #define I915_TILING_NONE 0 1658 #define I915_TILING_X 1 1659 #define I915_TILING_Y 2 1660 /* 1661 * Do not add new tiling types here. The I915_TILING_* values are for 1662 * de-tiling fence registers that no longer exist on modern platforms. Although 1663 * the hardware may support new types of tiling in general (e.g., Tile4), we 1664 * do not need to add them to the uapi that is specific to now-defunct ioctls. 1665 */ 1666 #define I915_TILING_LAST I915_TILING_Y 1667 1668 #define I915_BIT_6_SWIZZLE_NONE 0 1669 #define I915_BIT_6_SWIZZLE_9 1 1670 #define I915_BIT_6_SWIZZLE_9_10 2 1671 #define I915_BIT_6_SWIZZLE_9_11 3 1672 #define I915_BIT_6_SWIZZLE_9_10_11 4 1673 /* Not seen by userland */ 1674 #define I915_BIT_6_SWIZZLE_UNKNOWN 5 1675 /* Seen by userland. */ 1676 #define I915_BIT_6_SWIZZLE_9_17 6 1677 #define I915_BIT_6_SWIZZLE_9_10_17 7 1678 1679 struct drm_i915_gem_set_tiling { 1680 /** Handle of the buffer to have its tiling state updated */ 1681 __u32 handle; 1682 1683 /** 1684 * Tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1685 * I915_TILING_Y). 1686 * 1687 * This value is to be set on request, and will be updated by the 1688 * kernel on successful return with the actual chosen tiling layout. 1689 * 1690 * The tiling mode may be demoted to I915_TILING_NONE when the system 1691 * has bit 6 swizzling that can't be managed correctly by GEM. 1692 * 1693 * Buffer contents become undefined when changing tiling_mode. 1694 */ 1695 __u32 tiling_mode; 1696 1697 /** 1698 * Stride in bytes for the object when in I915_TILING_X or 1699 * I915_TILING_Y. 1700 */ 1701 __u32 stride; 1702 1703 /** 1704 * Returned address bit 6 swizzling required for CPU access through 1705 * mmap mapping. 1706 */ 1707 __u32 swizzle_mode; 1708 }; 1709 1710 struct drm_i915_gem_get_tiling { 1711 /** Handle of the buffer to get tiling state for. */ 1712 __u32 handle; 1713 1714 /** 1715 * Current tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1716 * I915_TILING_Y). 1717 */ 1718 __u32 tiling_mode; 1719 1720 /** 1721 * Returned address bit 6 swizzling required for CPU access through 1722 * mmap mapping. 1723 */ 1724 __u32 swizzle_mode; 1725 1726 /** 1727 * Returned address bit 6 swizzling required for CPU access through 1728 * mmap mapping whilst bound. 1729 */ 1730 __u32 phys_swizzle_mode; 1731 }; 1732 1733 struct drm_i915_gem_get_aperture { 1734 /** Total size of the aperture used by i915_gem_execbuffer, in bytes */ 1735 __u64 aper_size; 1736 1737 /** 1738 * Available space in the aperture used by i915_gem_execbuffer, in 1739 * bytes 1740 */ 1741 __u64 aper_available_size; 1742 }; 1743 1744 struct drm_i915_get_pipe_from_crtc_id { 1745 /** ID of CRTC being requested **/ 1746 __u32 crtc_id; 1747 1748 /** pipe of requested CRTC **/ 1749 __u32 pipe; 1750 }; 1751 1752 #define I915_MADV_WILLNEED 0 1753 #define I915_MADV_DONTNEED 1 1754 #define __I915_MADV_PURGED 2 /* internal state */ 1755 1756 struct drm_i915_gem_madvise { 1757 /** Handle of the buffer to change the backing store advice */ 1758 __u32 handle; 1759 1760 /* Advice: either the buffer will be needed again in the near future, 1761 * or wont be and could be discarded under memory pressure. 1762 */ 1763 __u32 madv; 1764 1765 /** Whether the backing store still exists. */ 1766 __u32 retained; 1767 }; 1768 1769 /* flags */ 1770 #define I915_OVERLAY_TYPE_MASK 0xff 1771 #define I915_OVERLAY_YUV_PLANAR 0x01 1772 #define I915_OVERLAY_YUV_PACKED 0x02 1773 #define I915_OVERLAY_RGB 0x03 1774 1775 #define I915_OVERLAY_DEPTH_MASK 0xff00 1776 #define I915_OVERLAY_RGB24 0x1000 1777 #define I915_OVERLAY_RGB16 0x2000 1778 #define I915_OVERLAY_RGB15 0x3000 1779 #define I915_OVERLAY_YUV422 0x0100 1780 #define I915_OVERLAY_YUV411 0x0200 1781 #define I915_OVERLAY_YUV420 0x0300 1782 #define I915_OVERLAY_YUV410 0x0400 1783 1784 #define I915_OVERLAY_SWAP_MASK 0xff0000 1785 #define I915_OVERLAY_NO_SWAP 0x000000 1786 #define I915_OVERLAY_UV_SWAP 0x010000 1787 #define I915_OVERLAY_Y_SWAP 0x020000 1788 #define I915_OVERLAY_Y_AND_UV_SWAP 0x030000 1789 1790 #define I915_OVERLAY_FLAGS_MASK 0xff000000 1791 #define I915_OVERLAY_ENABLE 0x01000000 1792 1793 struct drm_intel_overlay_put_image { 1794 /* various flags and src format description */ 1795 __u32 flags; 1796 /* source picture description */ 1797 __u32 bo_handle; 1798 /* stride values and offsets are in bytes, buffer relative */ 1799 __u16 stride_Y; /* stride for packed formats */ 1800 __u16 stride_UV; 1801 __u32 offset_Y; /* offset for packet formats */ 1802 __u32 offset_U; 1803 __u32 offset_V; 1804 /* in pixels */ 1805 __u16 src_width; 1806 __u16 src_height; 1807 /* to compensate the scaling factors for partially covered surfaces */ 1808 __u16 src_scan_width; 1809 __u16 src_scan_height; 1810 /* output crtc description */ 1811 __u32 crtc_id; 1812 __u16 dst_x; 1813 __u16 dst_y; 1814 __u16 dst_width; 1815 __u16 dst_height; 1816 }; 1817 1818 /* flags */ 1819 #define I915_OVERLAY_UPDATE_ATTRS (1<<0) 1820 #define I915_OVERLAY_UPDATE_GAMMA (1<<1) 1821 #define I915_OVERLAY_DISABLE_DEST_COLORKEY (1<<2) 1822 struct drm_intel_overlay_attrs { 1823 __u32 flags; 1824 __u32 color_key; 1825 __s32 brightness; 1826 __u32 contrast; 1827 __u32 saturation; 1828 __u32 gamma0; 1829 __u32 gamma1; 1830 __u32 gamma2; 1831 __u32 gamma3; 1832 __u32 gamma4; 1833 __u32 gamma5; 1834 }; 1835 1836 /* 1837 * Intel sprite handling 1838 * 1839 * Color keying works with a min/mask/max tuple. Both source and destination 1840 * color keying is allowed. 1841 * 1842 * Source keying: 1843 * Sprite pixels within the min & max values, masked against the color channels 1844 * specified in the mask field, will be transparent. All other pixels will 1845 * be displayed on top of the primary plane. For RGB surfaces, only the min 1846 * and mask fields will be used; ranged compares are not allowed. 1847 * 1848 * Destination keying: 1849 * Primary plane pixels that match the min value, masked against the color 1850 * channels specified in the mask field, will be replaced by corresponding 1851 * pixels from the sprite plane. 1852 * 1853 * Note that source & destination keying are exclusive; only one can be 1854 * active on a given plane. 1855 */ 1856 1857 #define I915_SET_COLORKEY_NONE (1<<0) /* Deprecated. Instead set 1858 * flags==0 to disable colorkeying. 1859 */ 1860 #define I915_SET_COLORKEY_DESTINATION (1<<1) 1861 #define I915_SET_COLORKEY_SOURCE (1<<2) 1862 struct drm_intel_sprite_colorkey { 1863 __u32 plane_id; 1864 __u32 min_value; 1865 __u32 channel_mask; 1866 __u32 max_value; 1867 __u32 flags; 1868 }; 1869 1870 struct drm_i915_gem_wait { 1871 /** Handle of BO we shall wait on */ 1872 __u32 bo_handle; 1873 __u32 flags; 1874 /** Number of nanoseconds to wait, Returns time remaining. */ 1875 __s64 timeout_ns; 1876 }; 1877 1878 struct drm_i915_gem_context_create { 1879 __u32 ctx_id; /* output: id of new context*/ 1880 __u32 pad; 1881 }; 1882 1883 /** 1884 * struct drm_i915_gem_context_create_ext - Structure for creating contexts. 1885 */ 1886 struct drm_i915_gem_context_create_ext { 1887 /** @ctx_id: Id of the created context (output) */ 1888 __u32 ctx_id; 1889 1890 /** 1891 * @flags: Supported flags are: 1892 * 1893 * I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS: 1894 * 1895 * Extensions may be appended to this structure and driver must check 1896 * for those. See @extensions. 1897 * 1898 * I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE 1899 * 1900 * Created context will have single timeline. 1901 */ 1902 __u32 flags; 1903 #define I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS (1u << 0) 1904 #define I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE (1u << 1) 1905 #define I915_CONTEXT_CREATE_FLAGS_UNKNOWN \ 1906 (-(I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE << 1)) 1907 1908 /** 1909 * @extensions: Zero-terminated chain of extensions. 1910 * 1911 * I915_CONTEXT_CREATE_EXT_SETPARAM: 1912 * Context parameter to set or query during context creation. 1913 * See struct drm_i915_gem_context_create_ext_setparam. 1914 * 1915 * I915_CONTEXT_CREATE_EXT_CLONE: 1916 * This extension has been removed. On the off chance someone somewhere 1917 * has attempted to use it, never re-use this extension number. 1918 */ 1919 __u64 extensions; 1920 #define I915_CONTEXT_CREATE_EXT_SETPARAM 0 1921 #define I915_CONTEXT_CREATE_EXT_CLONE 1 1922 }; 1923 1924 /** 1925 * struct drm_i915_gem_context_param - Context parameter to set or query. 1926 */ 1927 struct drm_i915_gem_context_param { 1928 /** @ctx_id: Context id */ 1929 __u32 ctx_id; 1930 1931 /** @size: Size of the parameter @value */ 1932 __u32 size; 1933 1934 /** @param: Parameter to set or query */ 1935 __u64 param; 1936 #define I915_CONTEXT_PARAM_BAN_PERIOD 0x1 1937 /* I915_CONTEXT_PARAM_NO_ZEROMAP has been removed. On the off chance 1938 * someone somewhere has attempted to use it, never re-use this context 1939 * param number. 1940 */ 1941 #define I915_CONTEXT_PARAM_NO_ZEROMAP 0x2 1942 #define I915_CONTEXT_PARAM_GTT_SIZE 0x3 1943 #define I915_CONTEXT_PARAM_NO_ERROR_CAPTURE 0x4 1944 #define I915_CONTEXT_PARAM_BANNABLE 0x5 1945 #define I915_CONTEXT_PARAM_PRIORITY 0x6 1946 #define I915_CONTEXT_MAX_USER_PRIORITY 1023 /* inclusive */ 1947 #define I915_CONTEXT_DEFAULT_PRIORITY 0 1948 #define I915_CONTEXT_MIN_USER_PRIORITY -1023 /* inclusive */ 1949 /* 1950 * When using the following param, value should be a pointer to 1951 * drm_i915_gem_context_param_sseu. 1952 */ 1953 #define I915_CONTEXT_PARAM_SSEU 0x7 1954 1955 /* 1956 * Not all clients may want to attempt automatic recover of a context after 1957 * a hang (for example, some clients may only submit very small incremental 1958 * batches relying on known logical state of previous batches which will never 1959 * recover correctly and each attempt will hang), and so would prefer that 1960 * the context is forever banned instead. 1961 * 1962 * If set to false (0), after a reset, subsequent (and in flight) rendering 1963 * from this context is discarded, and the client will need to create a new 1964 * context to use instead. 1965 * 1966 * If set to true (1), the kernel will automatically attempt to recover the 1967 * context by skipping the hanging batch and executing the next batch starting 1968 * from the default context state (discarding the incomplete logical context 1969 * state lost due to the reset). 1970 * 1971 * On creation, all new contexts are marked as recoverable. 1972 */ 1973 #define I915_CONTEXT_PARAM_RECOVERABLE 0x8 1974 1975 /* 1976 * The id of the associated virtual memory address space (ppGTT) of 1977 * this context. Can be retrieved and passed to another context 1978 * (on the same fd) for both to use the same ppGTT and so share 1979 * address layouts, and avoid reloading the page tables on context 1980 * switches between themselves. 1981 * 1982 * See DRM_I915_GEM_VM_CREATE and DRM_I915_GEM_VM_DESTROY. 1983 */ 1984 #define I915_CONTEXT_PARAM_VM 0x9 1985 1986 /* 1987 * I915_CONTEXT_PARAM_ENGINES: 1988 * 1989 * Bind this context to operate on this subset of available engines. Henceforth, 1990 * the I915_EXEC_RING selector for DRM_IOCTL_I915_GEM_EXECBUFFER2 operates as 1991 * an index into this array of engines; I915_EXEC_DEFAULT selecting engine[0] 1992 * and upwards. Slots 0...N are filled in using the specified (class, instance). 1993 * Use 1994 * engine_class: I915_ENGINE_CLASS_INVALID, 1995 * engine_instance: I915_ENGINE_CLASS_INVALID_NONE 1996 * to specify a gap in the array that can be filled in later, e.g. by a 1997 * virtual engine used for load balancing. 1998 * 1999 * Setting the number of engines bound to the context to 0, by passing a zero 2000 * sized argument, will revert back to default settings. 2001 * 2002 * See struct i915_context_param_engines. 2003 * 2004 * Extensions: 2005 * i915_context_engines_load_balance (I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE) 2006 * i915_context_engines_bond (I915_CONTEXT_ENGINES_EXT_BOND) 2007 * i915_context_engines_parallel_submit (I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT) 2008 */ 2009 #define I915_CONTEXT_PARAM_ENGINES 0xa 2010 2011 /* 2012 * I915_CONTEXT_PARAM_PERSISTENCE: 2013 * 2014 * Allow the context and active rendering to survive the process until 2015 * completion. Persistence allows fire-and-forget clients to queue up a 2016 * bunch of work, hand the output over to a display server and then quit. 2017 * If the context is marked as not persistent, upon closing (either via 2018 * an explicit DRM_I915_GEM_CONTEXT_DESTROY or implicitly from file closure 2019 * or process termination), the context and any outstanding requests will be 2020 * cancelled (and exported fences for cancelled requests marked as -EIO). 2021 * 2022 * By default, new contexts allow persistence. 2023 */ 2024 #define I915_CONTEXT_PARAM_PERSISTENCE 0xb 2025 2026 /* This API has been removed. On the off chance someone somewhere has 2027 * attempted to use it, never re-use this context param number. 2028 */ 2029 #define I915_CONTEXT_PARAM_RINGSIZE 0xc 2030 2031 /* 2032 * I915_CONTEXT_PARAM_PROTECTED_CONTENT: 2033 * 2034 * Mark that the context makes use of protected content, which will result 2035 * in the context being invalidated when the protected content session is. 2036 * Given that the protected content session is killed on suspend, the device 2037 * is kept awake for the lifetime of a protected context, so the user should 2038 * make sure to dispose of them once done. 2039 * This flag can only be set at context creation time and, when set to true, 2040 * must be preceded by an explicit setting of I915_CONTEXT_PARAM_RECOVERABLE 2041 * to false. This flag can't be set to true in conjunction with setting the 2042 * I915_CONTEXT_PARAM_BANNABLE flag to false. Creation example: 2043 * 2044 * .. code-block:: C 2045 * 2046 * struct drm_i915_gem_context_create_ext_setparam p_protected = { 2047 * .base = { 2048 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2049 * }, 2050 * .param = { 2051 * .param = I915_CONTEXT_PARAM_PROTECTED_CONTENT, 2052 * .value = 1, 2053 * } 2054 * }; 2055 * struct drm_i915_gem_context_create_ext_setparam p_norecover = { 2056 * .base = { 2057 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2058 * .next_extension = to_user_pointer(&p_protected), 2059 * }, 2060 * .param = { 2061 * .param = I915_CONTEXT_PARAM_RECOVERABLE, 2062 * .value = 0, 2063 * } 2064 * }; 2065 * struct drm_i915_gem_context_create_ext create = { 2066 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2067 * .extensions = to_user_pointer(&p_norecover); 2068 * }; 2069 * 2070 * ctx_id = gem_context_create_ext(drm_fd, &create); 2071 * 2072 * In addition to the normal failure cases, setting this flag during context 2073 * creation can result in the following errors: 2074 * 2075 * -ENODEV: feature not available 2076 * -EPERM: trying to mark a recoverable or not bannable context as protected 2077 */ 2078 #define I915_CONTEXT_PARAM_PROTECTED_CONTENT 0xd 2079 /* Must be kept compact -- no holes and well documented */ 2080 2081 /** @value: Context parameter value to be set or queried */ 2082 __u64 value; 2083 }; 2084 2085 /* 2086 * Context SSEU programming 2087 * 2088 * It may be necessary for either functional or performance reason to configure 2089 * a context to run with a reduced number of SSEU (where SSEU stands for Slice/ 2090 * Sub-slice/EU). 2091 * 2092 * This is done by configuring SSEU configuration using the below 2093 * @struct drm_i915_gem_context_param_sseu for every supported engine which 2094 * userspace intends to use. 2095 * 2096 * Not all GPUs or engines support this functionality in which case an error 2097 * code -ENODEV will be returned. 2098 * 2099 * Also, flexibility of possible SSEU configuration permutations varies between 2100 * GPU generations and software imposed limitations. Requesting such a 2101 * combination will return an error code of -EINVAL. 2102 * 2103 * NOTE: When perf/OA is active the context's SSEU configuration is ignored in 2104 * favour of a single global setting. 2105 */ 2106 struct drm_i915_gem_context_param_sseu { 2107 /* 2108 * Engine class & instance to be configured or queried. 2109 */ 2110 struct i915_engine_class_instance engine; 2111 2112 /* 2113 * Unknown flags must be cleared to zero. 2114 */ 2115 __u32 flags; 2116 #define I915_CONTEXT_SSEU_FLAG_ENGINE_INDEX (1u << 0) 2117 2118 /* 2119 * Mask of slices to enable for the context. Valid values are a subset 2120 * of the bitmask value returned for I915_PARAM_SLICE_MASK. 2121 */ 2122 __u64 slice_mask; 2123 2124 /* 2125 * Mask of subslices to enable for the context. Valid values are a 2126 * subset of the bitmask value return by I915_PARAM_SUBSLICE_MASK. 2127 */ 2128 __u64 subslice_mask; 2129 2130 /* 2131 * Minimum/Maximum number of EUs to enable per subslice for the 2132 * context. min_eus_per_subslice must be inferior or equal to 2133 * max_eus_per_subslice. 2134 */ 2135 __u16 min_eus_per_subslice; 2136 __u16 max_eus_per_subslice; 2137 2138 /* 2139 * Unused for now. Must be cleared to zero. 2140 */ 2141 __u32 rsvd; 2142 }; 2143 2144 /** 2145 * DOC: Virtual Engine uAPI 2146 * 2147 * Virtual engine is a concept where userspace is able to configure a set of 2148 * physical engines, submit a batch buffer, and let the driver execute it on any 2149 * engine from the set as it sees fit. 2150 * 2151 * This is primarily useful on parts which have multiple instances of a same 2152 * class engine, like for example GT3+ Skylake parts with their two VCS engines. 2153 * 2154 * For instance userspace can enumerate all engines of a certain class using the 2155 * previously described `Engine Discovery uAPI`_. After that userspace can 2156 * create a GEM context with a placeholder slot for the virtual engine (using 2157 * `I915_ENGINE_CLASS_INVALID` and `I915_ENGINE_CLASS_INVALID_NONE` for class 2158 * and instance respectively) and finally using the 2159 * `I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE` extension place a virtual engine in 2160 * the same reserved slot. 2161 * 2162 * Example of creating a virtual engine and submitting a batch buffer to it: 2163 * 2164 * .. code-block:: C 2165 * 2166 * I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(virtual, 2) = { 2167 * .base.name = I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE, 2168 * .engine_index = 0, // Place this virtual engine into engine map slot 0 2169 * .num_siblings = 2, 2170 * .engines = { { I915_ENGINE_CLASS_VIDEO, 0 }, 2171 * { I915_ENGINE_CLASS_VIDEO, 1 }, }, 2172 * }; 2173 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 1) = { 2174 * .engines = { { I915_ENGINE_CLASS_INVALID, 2175 * I915_ENGINE_CLASS_INVALID_NONE } }, 2176 * .extensions = to_user_pointer(&virtual), // Chains after load_balance extension 2177 * }; 2178 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2179 * .base = { 2180 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2181 * }, 2182 * .param = { 2183 * .param = I915_CONTEXT_PARAM_ENGINES, 2184 * .value = to_user_pointer(&engines), 2185 * .size = sizeof(engines), 2186 * }, 2187 * }; 2188 * struct drm_i915_gem_context_create_ext create = { 2189 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2190 * .extensions = to_user_pointer(&p_engines); 2191 * }; 2192 * 2193 * ctx_id = gem_context_create_ext(drm_fd, &create); 2194 * 2195 * // Now we have created a GEM context with its engine map containing a 2196 * // single virtual engine. Submissions to this slot can go either to 2197 * // vcs0 or vcs1, depending on the load balancing algorithm used inside 2198 * // the driver. The load balancing is dynamic from one batch buffer to 2199 * // another and transparent to userspace. 2200 * 2201 * ... 2202 * execbuf.rsvd1 = ctx_id; 2203 * execbuf.flags = 0; // Submits to index 0 which is the virtual engine 2204 * gem_execbuf(drm_fd, &execbuf); 2205 */ 2206 2207 /* 2208 * i915_context_engines_load_balance: 2209 * 2210 * Enable load balancing across this set of engines. 2211 * 2212 * Into the I915_EXEC_DEFAULT slot [0], a virtual engine is created that when 2213 * used will proxy the execbuffer request onto one of the set of engines 2214 * in such a way as to distribute the load evenly across the set. 2215 * 2216 * The set of engines must be compatible (e.g. the same HW class) as they 2217 * will share the same logical GPU context and ring. 2218 * 2219 * To intermix rendering with the virtual engine and direct rendering onto 2220 * the backing engines (bypassing the load balancing proxy), the context must 2221 * be defined to use a single timeline for all engines. 2222 */ 2223 struct i915_context_engines_load_balance { 2224 struct i915_user_extension base; 2225 2226 __u16 engine_index; 2227 __u16 num_siblings; 2228 __u32 flags; /* all undefined flags must be zero */ 2229 2230 __u64 mbz64; /* reserved for future use; must be zero */ 2231 2232 struct i915_engine_class_instance engines[0]; 2233 } __attribute__((packed)); 2234 2235 #define I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(name__, N__) struct { \ 2236 struct i915_user_extension base; \ 2237 __u16 engine_index; \ 2238 __u16 num_siblings; \ 2239 __u32 flags; \ 2240 __u64 mbz64; \ 2241 struct i915_engine_class_instance engines[N__]; \ 2242 } __attribute__((packed)) name__ 2243 2244 /* 2245 * i915_context_engines_bond: 2246 * 2247 * Constructed bonded pairs for execution within a virtual engine. 2248 * 2249 * All engines are equal, but some are more equal than others. Given 2250 * the distribution of resources in the HW, it may be preferable to run 2251 * a request on a given subset of engines in parallel to a request on a 2252 * specific engine. We enable this selection of engines within a virtual 2253 * engine by specifying bonding pairs, for any given master engine we will 2254 * only execute on one of the corresponding siblings within the virtual engine. 2255 * 2256 * To execute a request in parallel on the master engine and a sibling requires 2257 * coordination with a I915_EXEC_FENCE_SUBMIT. 2258 */ 2259 struct i915_context_engines_bond { 2260 struct i915_user_extension base; 2261 2262 struct i915_engine_class_instance master; 2263 2264 __u16 virtual_index; /* index of virtual engine in ctx->engines[] */ 2265 __u16 num_bonds; 2266 2267 __u64 flags; /* all undefined flags must be zero */ 2268 __u64 mbz64[4]; /* reserved for future use; must be zero */ 2269 2270 struct i915_engine_class_instance engines[0]; 2271 } __attribute__((packed)); 2272 2273 #define I915_DEFINE_CONTEXT_ENGINES_BOND(name__, N__) struct { \ 2274 struct i915_user_extension base; \ 2275 struct i915_engine_class_instance master; \ 2276 __u16 virtual_index; \ 2277 __u16 num_bonds; \ 2278 __u64 flags; \ 2279 __u64 mbz64[4]; \ 2280 struct i915_engine_class_instance engines[N__]; \ 2281 } __attribute__((packed)) name__ 2282 2283 /** 2284 * struct i915_context_engines_parallel_submit - Configure engine for 2285 * parallel submission. 2286 * 2287 * Setup a slot in the context engine map to allow multiple BBs to be submitted 2288 * in a single execbuf IOCTL. Those BBs will then be scheduled to run on the GPU 2289 * in parallel. Multiple hardware contexts are created internally in the i915 to 2290 * run these BBs. Once a slot is configured for N BBs only N BBs can be 2291 * submitted in each execbuf IOCTL and this is implicit behavior e.g. The user 2292 * doesn't tell the execbuf IOCTL there are N BBs, the execbuf IOCTL knows how 2293 * many BBs there are based on the slot's configuration. The N BBs are the last 2294 * N buffer objects or first N if I915_EXEC_BATCH_FIRST is set. 2295 * 2296 * The default placement behavior is to create implicit bonds between each 2297 * context if each context maps to more than 1 physical engine (e.g. context is 2298 * a virtual engine). Also we only allow contexts of same engine class and these 2299 * contexts must be in logically contiguous order. Examples of the placement 2300 * behavior are described below. Lastly, the default is to not allow BBs to be 2301 * preempted mid-batch. Rather insert coordinated preemption points on all 2302 * hardware contexts between each set of BBs. Flags could be added in the future 2303 * to change both of these default behaviors. 2304 * 2305 * Returns -EINVAL if hardware context placement configuration is invalid or if 2306 * the placement configuration isn't supported on the platform / submission 2307 * interface. 2308 * Returns -ENODEV if extension isn't supported on the platform / submission 2309 * interface. 2310 * 2311 * .. code-block:: none 2312 * 2313 * Examples syntax: 2314 * CS[X] = generic engine of same class, logical instance X 2315 * INVALID = I915_ENGINE_CLASS_INVALID, I915_ENGINE_CLASS_INVALID_NONE 2316 * 2317 * Example 1 pseudo code: 2318 * set_engines(INVALID) 2319 * set_parallel(engine_index=0, width=2, num_siblings=1, 2320 * engines=CS[0],CS[1]) 2321 * 2322 * Results in the following valid placement: 2323 * CS[0], CS[1] 2324 * 2325 * Example 2 pseudo code: 2326 * set_engines(INVALID) 2327 * set_parallel(engine_index=0, width=2, num_siblings=2, 2328 * engines=CS[0],CS[2],CS[1],CS[3]) 2329 * 2330 * Results in the following valid placements: 2331 * CS[0], CS[1] 2332 * CS[2], CS[3] 2333 * 2334 * This can be thought of as two virtual engines, each containing two 2335 * engines thereby making a 2D array. However, there are bonds tying the 2336 * entries together and placing restrictions on how they can be scheduled. 2337 * Specifically, the scheduler can choose only vertical columns from the 2D 2338 * array. That is, CS[0] is bonded to CS[1] and CS[2] to CS[3]. So if the 2339 * scheduler wants to submit to CS[0], it must also choose CS[1] and vice 2340 * versa. Same for CS[2] requires also using CS[3]. 2341 * VE[0] = CS[0], CS[2] 2342 * VE[1] = CS[1], CS[3] 2343 * 2344 * Example 3 pseudo code: 2345 * set_engines(INVALID) 2346 * set_parallel(engine_index=0, width=2, num_siblings=2, 2347 * engines=CS[0],CS[1],CS[1],CS[3]) 2348 * 2349 * Results in the following valid and invalid placements: 2350 * CS[0], CS[1] 2351 * CS[1], CS[3] - Not logically contiguous, return -EINVAL 2352 */ 2353 struct i915_context_engines_parallel_submit { 2354 /** 2355 * @base: base user extension. 2356 */ 2357 struct i915_user_extension base; 2358 2359 /** 2360 * @engine_index: slot for parallel engine 2361 */ 2362 __u16 engine_index; 2363 2364 /** 2365 * @width: number of contexts per parallel engine or in other words the 2366 * number of batches in each submission 2367 */ 2368 __u16 width; 2369 2370 /** 2371 * @num_siblings: number of siblings per context or in other words the 2372 * number of possible placements for each submission 2373 */ 2374 __u16 num_siblings; 2375 2376 /** 2377 * @mbz16: reserved for future use; must be zero 2378 */ 2379 __u16 mbz16; 2380 2381 /** 2382 * @flags: all undefined flags must be zero, currently not defined flags 2383 */ 2384 __u64 flags; 2385 2386 /** 2387 * @mbz64: reserved for future use; must be zero 2388 */ 2389 __u64 mbz64[3]; 2390 2391 /** 2392 * @engines: 2-d array of engine instances to configure parallel engine 2393 * 2394 * length = width (i) * num_siblings (j) 2395 * index = j + i * num_siblings 2396 */ 2397 struct i915_engine_class_instance engines[0]; 2398 2399 } __attribute__((packed)); 2400 2401 #define I915_DEFINE_CONTEXT_ENGINES_PARALLEL_SUBMIT(name__, N__) struct { \ 2402 struct i915_user_extension base; \ 2403 __u16 engine_index; \ 2404 __u16 width; \ 2405 __u16 num_siblings; \ 2406 __u16 mbz16; \ 2407 __u64 flags; \ 2408 __u64 mbz64[3]; \ 2409 struct i915_engine_class_instance engines[N__]; \ 2410 } __attribute__((packed)) name__ 2411 2412 /** 2413 * DOC: Context Engine Map uAPI 2414 * 2415 * Context engine map is a new way of addressing engines when submitting batch- 2416 * buffers, replacing the existing way of using identifiers like `I915_EXEC_BLT` 2417 * inside the flags field of `struct drm_i915_gem_execbuffer2`. 2418 * 2419 * To use it created GEM contexts need to be configured with a list of engines 2420 * the user is intending to submit to. This is accomplished using the 2421 * `I915_CONTEXT_PARAM_ENGINES` parameter and `struct 2422 * i915_context_param_engines`. 2423 * 2424 * For such contexts the `I915_EXEC_RING_MASK` field becomes an index into the 2425 * configured map. 2426 * 2427 * Example of creating such context and submitting against it: 2428 * 2429 * .. code-block:: C 2430 * 2431 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 2) = { 2432 * .engines = { { I915_ENGINE_CLASS_RENDER, 0 }, 2433 * { I915_ENGINE_CLASS_COPY, 0 } } 2434 * }; 2435 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2436 * .base = { 2437 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2438 * }, 2439 * .param = { 2440 * .param = I915_CONTEXT_PARAM_ENGINES, 2441 * .value = to_user_pointer(&engines), 2442 * .size = sizeof(engines), 2443 * }, 2444 * }; 2445 * struct drm_i915_gem_context_create_ext create = { 2446 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2447 * .extensions = to_user_pointer(&p_engines); 2448 * }; 2449 * 2450 * ctx_id = gem_context_create_ext(drm_fd, &create); 2451 * 2452 * // We have now created a GEM context with two engines in the map: 2453 * // Index 0 points to rcs0 while index 1 points to bcs0. Other engines 2454 * // will not be accessible from this context. 2455 * 2456 * ... 2457 * execbuf.rsvd1 = ctx_id; 2458 * execbuf.flags = 0; // Submits to index 0, which is rcs0 for this context 2459 * gem_execbuf(drm_fd, &execbuf); 2460 * 2461 * ... 2462 * execbuf.rsvd1 = ctx_id; 2463 * execbuf.flags = 1; // Submits to index 0, which is bcs0 for this context 2464 * gem_execbuf(drm_fd, &execbuf); 2465 */ 2466 2467 struct i915_context_param_engines { 2468 __u64 extensions; /* linked chain of extension blocks, 0 terminates */ 2469 #define I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE 0 /* see i915_context_engines_load_balance */ 2470 #define I915_CONTEXT_ENGINES_EXT_BOND 1 /* see i915_context_engines_bond */ 2471 #define I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT 2 /* see i915_context_engines_parallel_submit */ 2472 struct i915_engine_class_instance engines[0]; 2473 } __attribute__((packed)); 2474 2475 #define I915_DEFINE_CONTEXT_PARAM_ENGINES(name__, N__) struct { \ 2476 __u64 extensions; \ 2477 struct i915_engine_class_instance engines[N__]; \ 2478 } __attribute__((packed)) name__ 2479 2480 /** 2481 * struct drm_i915_gem_context_create_ext_setparam - Context parameter 2482 * to set or query during context creation. 2483 */ 2484 struct drm_i915_gem_context_create_ext_setparam { 2485 /** @base: Extension link. See struct i915_user_extension. */ 2486 struct i915_user_extension base; 2487 2488 /** 2489 * @param: Context parameter to set or query. 2490 * See struct drm_i915_gem_context_param. 2491 */ 2492 struct drm_i915_gem_context_param param; 2493 }; 2494 2495 struct drm_i915_gem_context_destroy { 2496 __u32 ctx_id; 2497 __u32 pad; 2498 }; 2499 2500 /** 2501 * struct drm_i915_gem_vm_control - Structure to create or destroy VM. 2502 * 2503 * DRM_I915_GEM_VM_CREATE - 2504 * 2505 * Create a new virtual memory address space (ppGTT) for use within a context 2506 * on the same file. Extensions can be provided to configure exactly how the 2507 * address space is setup upon creation. 2508 * 2509 * The id of new VM (bound to the fd) for use with I915_CONTEXT_PARAM_VM is 2510 * returned in the outparam @id. 2511 * 2512 * An extension chain maybe provided, starting with @extensions, and terminated 2513 * by the @next_extension being 0. Currently, no extensions are defined. 2514 * 2515 * DRM_I915_GEM_VM_DESTROY - 2516 * 2517 * Destroys a previously created VM id, specified in @vm_id. 2518 * 2519 * No extensions or flags are allowed currently, and so must be zero. 2520 */ 2521 struct drm_i915_gem_vm_control { 2522 /** @extensions: Zero-terminated chain of extensions. */ 2523 __u64 extensions; 2524 2525 /** @flags: reserved for future usage, currently MBZ */ 2526 __u32 flags; 2527 2528 /** @vm_id: Id of the VM created or to be destroyed */ 2529 __u32 vm_id; 2530 }; 2531 2532 struct drm_i915_reg_read { 2533 /* 2534 * Register offset. 2535 * For 64bit wide registers where the upper 32bits don't immediately 2536 * follow the lower 32bits, the offset of the lower 32bits must 2537 * be specified 2538 */ 2539 __u64 offset; 2540 #define I915_REG_READ_8B_WA (1ul << 0) 2541 2542 __u64 val; /* Return value */ 2543 }; 2544 2545 /* Known registers: 2546 * 2547 * Render engine timestamp - 0x2358 + 64bit - gen7+ 2548 * - Note this register returns an invalid value if using the default 2549 * single instruction 8byte read, in order to workaround that pass 2550 * flag I915_REG_READ_8B_WA in offset field. 2551 * 2552 */ 2553 2554 struct drm_i915_reset_stats { 2555 __u32 ctx_id; 2556 __u32 flags; 2557 2558 /* All resets since boot/module reload, for all contexts */ 2559 __u32 reset_count; 2560 2561 /* Number of batches lost when active in GPU, for this context */ 2562 __u32 batch_active; 2563 2564 /* Number of batches lost pending for execution, for this context */ 2565 __u32 batch_pending; 2566 2567 __u32 pad; 2568 }; 2569 2570 /** 2571 * struct drm_i915_gem_userptr - Create GEM object from user allocated memory. 2572 * 2573 * Userptr objects have several restrictions on what ioctls can be used with the 2574 * object handle. 2575 */ 2576 struct drm_i915_gem_userptr { 2577 /** 2578 * @user_ptr: The pointer to the allocated memory. 2579 * 2580 * Needs to be aligned to PAGE_SIZE. 2581 */ 2582 __u64 user_ptr; 2583 2584 /** 2585 * @user_size: 2586 * 2587 * The size in bytes for the allocated memory. This will also become the 2588 * object size. 2589 * 2590 * Needs to be aligned to PAGE_SIZE, and should be at least PAGE_SIZE, 2591 * or larger. 2592 */ 2593 __u64 user_size; 2594 2595 /** 2596 * @flags: 2597 * 2598 * Supported flags: 2599 * 2600 * I915_USERPTR_READ_ONLY: 2601 * 2602 * Mark the object as readonly, this also means GPU access can only be 2603 * readonly. This is only supported on HW which supports readonly access 2604 * through the GTT. If the HW can't support readonly access, an error is 2605 * returned. 2606 * 2607 * I915_USERPTR_PROBE: 2608 * 2609 * Probe the provided @user_ptr range and validate that the @user_ptr is 2610 * indeed pointing to normal memory and that the range is also valid. 2611 * For example if some garbage address is given to the kernel, then this 2612 * should complain. 2613 * 2614 * Returns -EFAULT if the probe failed. 2615 * 2616 * Note that this doesn't populate the backing pages, and also doesn't 2617 * guarantee that the object will remain valid when the object is 2618 * eventually used. 2619 * 2620 * The kernel supports this feature if I915_PARAM_HAS_USERPTR_PROBE 2621 * returns a non-zero value. 2622 * 2623 * I915_USERPTR_UNSYNCHRONIZED: 2624 * 2625 * NOT USED. Setting this flag will result in an error. 2626 */ 2627 __u32 flags; 2628 #define I915_USERPTR_READ_ONLY 0x1 2629 #define I915_USERPTR_PROBE 0x2 2630 #define I915_USERPTR_UNSYNCHRONIZED 0x80000000 2631 /** 2632 * @handle: Returned handle for the object. 2633 * 2634 * Object handles are nonzero. 2635 */ 2636 __u32 handle; 2637 }; 2638 2639 enum drm_i915_oa_format { 2640 I915_OA_FORMAT_A13 = 1, /* HSW only */ 2641 I915_OA_FORMAT_A29, /* HSW only */ 2642 I915_OA_FORMAT_A13_B8_C8, /* HSW only */ 2643 I915_OA_FORMAT_B4_C8, /* HSW only */ 2644 I915_OA_FORMAT_A45_B8_C8, /* HSW only */ 2645 I915_OA_FORMAT_B4_C8_A16, /* HSW only */ 2646 I915_OA_FORMAT_C4_B8, /* HSW+ */ 2647 2648 /* Gen8+ */ 2649 I915_OA_FORMAT_A12, 2650 I915_OA_FORMAT_A12_B8_C8, 2651 I915_OA_FORMAT_A32u40_A4u32_B8_C8, 2652 2653 I915_OA_FORMAT_MAX /* non-ABI */ 2654 }; 2655 2656 enum drm_i915_perf_property_id { 2657 /** 2658 * Open the stream for a specific context handle (as used with 2659 * execbuffer2). A stream opened for a specific context this way 2660 * won't typically require root privileges. 2661 * 2662 * This property is available in perf revision 1. 2663 */ 2664 DRM_I915_PERF_PROP_CTX_HANDLE = 1, 2665 2666 /** 2667 * A value of 1 requests the inclusion of raw OA unit reports as 2668 * part of stream samples. 2669 * 2670 * This property is available in perf revision 1. 2671 */ 2672 DRM_I915_PERF_PROP_SAMPLE_OA, 2673 2674 /** 2675 * The value specifies which set of OA unit metrics should be 2676 * configured, defining the contents of any OA unit reports. 2677 * 2678 * This property is available in perf revision 1. 2679 */ 2680 DRM_I915_PERF_PROP_OA_METRICS_SET, 2681 2682 /** 2683 * The value specifies the size and layout of OA unit reports. 2684 * 2685 * This property is available in perf revision 1. 2686 */ 2687 DRM_I915_PERF_PROP_OA_FORMAT, 2688 2689 /** 2690 * Specifying this property implicitly requests periodic OA unit 2691 * sampling and (at least on Haswell) the sampling frequency is derived 2692 * from this exponent as follows: 2693 * 2694 * 80ns * 2^(period_exponent + 1) 2695 * 2696 * This property is available in perf revision 1. 2697 */ 2698 DRM_I915_PERF_PROP_OA_EXPONENT, 2699 2700 /** 2701 * Specifying this property is only valid when specify a context to 2702 * filter with DRM_I915_PERF_PROP_CTX_HANDLE. Specifying this property 2703 * will hold preemption of the particular context we want to gather 2704 * performance data about. The execbuf2 submissions must include a 2705 * drm_i915_gem_execbuffer_ext_perf parameter for this to apply. 2706 * 2707 * This property is available in perf revision 3. 2708 */ 2709 DRM_I915_PERF_PROP_HOLD_PREEMPTION, 2710 2711 /** 2712 * Specifying this pins all contexts to the specified SSEU power 2713 * configuration for the duration of the recording. 2714 * 2715 * This parameter's value is a pointer to a struct 2716 * drm_i915_gem_context_param_sseu. 2717 * 2718 * This property is available in perf revision 4. 2719 */ 2720 DRM_I915_PERF_PROP_GLOBAL_SSEU, 2721 2722 /** 2723 * This optional parameter specifies the timer interval in nanoseconds 2724 * at which the i915 driver will check the OA buffer for available data. 2725 * Minimum allowed value is 100 microseconds. A default value is used by 2726 * the driver if this parameter is not specified. Note that larger timer 2727 * values will reduce cpu consumption during OA perf captures. However, 2728 * excessively large values would potentially result in OA buffer 2729 * overwrites as captures reach end of the OA buffer. 2730 * 2731 * This property is available in perf revision 5. 2732 */ 2733 DRM_I915_PERF_PROP_POLL_OA_PERIOD, 2734 2735 DRM_I915_PERF_PROP_MAX /* non-ABI */ 2736 }; 2737 2738 struct drm_i915_perf_open_param { 2739 __u32 flags; 2740 #define I915_PERF_FLAG_FD_CLOEXEC (1<<0) 2741 #define I915_PERF_FLAG_FD_NONBLOCK (1<<1) 2742 #define I915_PERF_FLAG_DISABLED (1<<2) 2743 2744 /** The number of u64 (id, value) pairs */ 2745 __u32 num_properties; 2746 2747 /** 2748 * Pointer to array of u64 (id, value) pairs configuring the stream 2749 * to open. 2750 */ 2751 __u64 properties_ptr; 2752 }; 2753 2754 /* 2755 * Enable data capture for a stream that was either opened in a disabled state 2756 * via I915_PERF_FLAG_DISABLED or was later disabled via 2757 * I915_PERF_IOCTL_DISABLE. 2758 * 2759 * It is intended to be cheaper to disable and enable a stream than it may be 2760 * to close and re-open a stream with the same configuration. 2761 * 2762 * It's undefined whether any pending data for the stream will be lost. 2763 * 2764 * This ioctl is available in perf revision 1. 2765 */ 2766 #define I915_PERF_IOCTL_ENABLE _IO('i', 0x0) 2767 2768 /* 2769 * Disable data capture for a stream. 2770 * 2771 * It is an error to try and read a stream that is disabled. 2772 * 2773 * This ioctl is available in perf revision 1. 2774 */ 2775 #define I915_PERF_IOCTL_DISABLE _IO('i', 0x1) 2776 2777 /* 2778 * Change metrics_set captured by a stream. 2779 * 2780 * If the stream is bound to a specific context, the configuration change 2781 * will performed __inline__ with that context such that it takes effect before 2782 * the next execbuf submission. 2783 * 2784 * Returns the previously bound metrics set id, or a negative error code. 2785 * 2786 * This ioctl is available in perf revision 2. 2787 */ 2788 #define I915_PERF_IOCTL_CONFIG _IO('i', 0x2) 2789 2790 /* 2791 * Common to all i915 perf records 2792 */ 2793 struct drm_i915_perf_record_header { 2794 __u32 type; 2795 __u16 pad; 2796 __u16 size; 2797 }; 2798 2799 enum drm_i915_perf_record_type { 2800 2801 /** 2802 * Samples are the work horse record type whose contents are extensible 2803 * and defined when opening an i915 perf stream based on the given 2804 * properties. 2805 * 2806 * Boolean properties following the naming convention 2807 * DRM_I915_PERF_SAMPLE_xyz_PROP request the inclusion of 'xyz' data in 2808 * every sample. 2809 * 2810 * The order of these sample properties given by userspace has no 2811 * affect on the ordering of data within a sample. The order is 2812 * documented here. 2813 * 2814 * struct { 2815 * struct drm_i915_perf_record_header header; 2816 * 2817 * { u32 oa_report[]; } && DRM_I915_PERF_PROP_SAMPLE_OA 2818 * }; 2819 */ 2820 DRM_I915_PERF_RECORD_SAMPLE = 1, 2821 2822 /* 2823 * Indicates that one or more OA reports were not written by the 2824 * hardware. This can happen for example if an MI_REPORT_PERF_COUNT 2825 * command collides with periodic sampling - which would be more likely 2826 * at higher sampling frequencies. 2827 */ 2828 DRM_I915_PERF_RECORD_OA_REPORT_LOST = 2, 2829 2830 /** 2831 * An error occurred that resulted in all pending OA reports being lost. 2832 */ 2833 DRM_I915_PERF_RECORD_OA_BUFFER_LOST = 3, 2834 2835 DRM_I915_PERF_RECORD_MAX /* non-ABI */ 2836 }; 2837 2838 /** 2839 * struct drm_i915_perf_oa_config 2840 * 2841 * Structure to upload perf dynamic configuration into the kernel. 2842 */ 2843 struct drm_i915_perf_oa_config { 2844 /** 2845 * @uuid: 2846 * 2847 * String formatted like "%\08x-%\04x-%\04x-%\04x-%\012x" 2848 */ 2849 char uuid[36]; 2850 2851 /** 2852 * @n_mux_regs: 2853 * 2854 * Number of mux regs in &mux_regs_ptr. 2855 */ 2856 __u32 n_mux_regs; 2857 2858 /** 2859 * @n_boolean_regs: 2860 * 2861 * Number of boolean regs in &boolean_regs_ptr. 2862 */ 2863 __u32 n_boolean_regs; 2864 2865 /** 2866 * @n_flex_regs: 2867 * 2868 * Number of flex regs in &flex_regs_ptr. 2869 */ 2870 __u32 n_flex_regs; 2871 2872 /** 2873 * @mux_regs_ptr: 2874 * 2875 * Pointer to tuples of u32 values (register address, value) for mux 2876 * registers. Expected length of buffer is (2 * sizeof(u32) * 2877 * &n_mux_regs). 2878 */ 2879 __u64 mux_regs_ptr; 2880 2881 /** 2882 * @boolean_regs_ptr: 2883 * 2884 * Pointer to tuples of u32 values (register address, value) for mux 2885 * registers. Expected length of buffer is (2 * sizeof(u32) * 2886 * &n_boolean_regs). 2887 */ 2888 __u64 boolean_regs_ptr; 2889 2890 /** 2891 * @flex_regs_ptr: 2892 * 2893 * Pointer to tuples of u32 values (register address, value) for mux 2894 * registers. Expected length of buffer is (2 * sizeof(u32) * 2895 * &n_flex_regs). 2896 */ 2897 __u64 flex_regs_ptr; 2898 }; 2899 2900 /** 2901 * struct drm_i915_query_item - An individual query for the kernel to process. 2902 * 2903 * The behaviour is determined by the @query_id. Note that exactly what 2904 * @data_ptr is also depends on the specific @query_id. 2905 */ 2906 struct drm_i915_query_item { 2907 /** 2908 * @query_id: 2909 * 2910 * The id for this query. Currently accepted query IDs are: 2911 * - %DRM_I915_QUERY_TOPOLOGY_INFO (see struct drm_i915_query_topology_info) 2912 * - %DRM_I915_QUERY_ENGINE_INFO (see struct drm_i915_engine_info) 2913 * - %DRM_I915_QUERY_PERF_CONFIG (see struct drm_i915_query_perf_config) 2914 * - %DRM_I915_QUERY_MEMORY_REGIONS (see struct drm_i915_query_memory_regions) 2915 * - %DRM_I915_QUERY_HWCONFIG_BLOB (see `GuC HWCONFIG blob uAPI`) 2916 * - %DRM_I915_QUERY_GEOMETRY_SUBSLICES (see struct drm_i915_query_topology_info) 2917 */ 2918 __u64 query_id; 2919 #define DRM_I915_QUERY_TOPOLOGY_INFO 1 2920 #define DRM_I915_QUERY_ENGINE_INFO 2 2921 #define DRM_I915_QUERY_PERF_CONFIG 3 2922 #define DRM_I915_QUERY_MEMORY_REGIONS 4 2923 #define DRM_I915_QUERY_HWCONFIG_BLOB 5 2924 #define DRM_I915_QUERY_GEOMETRY_SUBSLICES 6 2925 /* Must be kept compact -- no holes and well documented */ 2926 2927 /** 2928 * @length: 2929 * 2930 * When set to zero by userspace, this is filled with the size of the 2931 * data to be written at the @data_ptr pointer. The kernel sets this 2932 * value to a negative value to signal an error on a particular query 2933 * item. 2934 */ 2935 __s32 length; 2936 2937 /** 2938 * @flags: 2939 * 2940 * When &query_id == %DRM_I915_QUERY_TOPOLOGY_INFO, must be 0. 2941 * 2942 * When &query_id == %DRM_I915_QUERY_PERF_CONFIG, must be one of the 2943 * following: 2944 * 2945 * - %DRM_I915_QUERY_PERF_CONFIG_LIST 2946 * - %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2947 * - %DRM_I915_QUERY_PERF_CONFIG_FOR_UUID 2948 * 2949 * When &query_id == %DRM_I915_QUERY_GEOMETRY_SUBSLICES must contain 2950 * a struct i915_engine_class_instance that references a render engine. 2951 */ 2952 __u32 flags; 2953 #define DRM_I915_QUERY_PERF_CONFIG_LIST 1 2954 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2 2955 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID 3 2956 2957 /** 2958 * @data_ptr: 2959 * 2960 * Data will be written at the location pointed by @data_ptr when the 2961 * value of @length matches the length of the data to be written by the 2962 * kernel. 2963 */ 2964 __u64 data_ptr; 2965 }; 2966 2967 /** 2968 * struct drm_i915_query - Supply an array of struct drm_i915_query_item for the 2969 * kernel to fill out. 2970 * 2971 * Note that this is generally a two step process for each struct 2972 * drm_i915_query_item in the array: 2973 * 2974 * 1. Call the DRM_IOCTL_I915_QUERY, giving it our array of struct 2975 * drm_i915_query_item, with &drm_i915_query_item.length set to zero. The 2976 * kernel will then fill in the size, in bytes, which tells userspace how 2977 * memory it needs to allocate for the blob(say for an array of properties). 2978 * 2979 * 2. Next we call DRM_IOCTL_I915_QUERY again, this time with the 2980 * &drm_i915_query_item.data_ptr equal to our newly allocated blob. Note that 2981 * the &drm_i915_query_item.length should still be the same as what the 2982 * kernel previously set. At this point the kernel can fill in the blob. 2983 * 2984 * Note that for some query items it can make sense for userspace to just pass 2985 * in a buffer/blob equal to or larger than the required size. In this case only 2986 * a single ioctl call is needed. For some smaller query items this can work 2987 * quite well. 2988 * 2989 */ 2990 struct drm_i915_query { 2991 /** @num_items: The number of elements in the @items_ptr array */ 2992 __u32 num_items; 2993 2994 /** 2995 * @flags: Unused for now. Must be cleared to zero. 2996 */ 2997 __u32 flags; 2998 2999 /** 3000 * @items_ptr: 3001 * 3002 * Pointer to an array of struct drm_i915_query_item. The number of 3003 * array elements is @num_items. 3004 */ 3005 __u64 items_ptr; 3006 }; 3007 3008 /** 3009 * struct drm_i915_query_topology_info 3010 * 3011 * Describes slice/subslice/EU information queried by 3012 * %DRM_I915_QUERY_TOPOLOGY_INFO 3013 */ 3014 struct drm_i915_query_topology_info { 3015 /** 3016 * @flags: 3017 * 3018 * Unused for now. Must be cleared to zero. 3019 */ 3020 __u16 flags; 3021 3022 /** 3023 * @max_slices: 3024 * 3025 * The number of bits used to express the slice mask. 3026 */ 3027 __u16 max_slices; 3028 3029 /** 3030 * @max_subslices: 3031 * 3032 * The number of bits used to express the subslice mask. 3033 */ 3034 __u16 max_subslices; 3035 3036 /** 3037 * @max_eus_per_subslice: 3038 * 3039 * The number of bits in the EU mask that correspond to a single 3040 * subslice's EUs. 3041 */ 3042 __u16 max_eus_per_subslice; 3043 3044 /** 3045 * @subslice_offset: 3046 * 3047 * Offset in data[] at which the subslice masks are stored. 3048 */ 3049 __u16 subslice_offset; 3050 3051 /** 3052 * @subslice_stride: 3053 * 3054 * Stride at which each of the subslice masks for each slice are 3055 * stored. 3056 */ 3057 __u16 subslice_stride; 3058 3059 /** 3060 * @eu_offset: 3061 * 3062 * Offset in data[] at which the EU masks are stored. 3063 */ 3064 __u16 eu_offset; 3065 3066 /** 3067 * @eu_stride: 3068 * 3069 * Stride at which each of the EU masks for each subslice are stored. 3070 */ 3071 __u16 eu_stride; 3072 3073 /** 3074 * @data: 3075 * 3076 * Contains 3 pieces of information : 3077 * 3078 * - The slice mask with one bit per slice telling whether a slice is 3079 * available. The availability of slice X can be queried with the 3080 * following formula : 3081 * 3082 * .. code:: c 3083 * 3084 * (data[X / 8] >> (X % 8)) & 1 3085 * 3086 * Starting with Xe_HP platforms, Intel hardware no longer has 3087 * traditional slices so i915 will always report a single slice 3088 * (hardcoded slicemask = 0x1) which contains all of the platform's 3089 * subslices. I.e., the mask here does not reflect any of the newer 3090 * hardware concepts such as "gslices" or "cslices" since userspace 3091 * is capable of inferring those from the subslice mask. 3092 * 3093 * - The subslice mask for each slice with one bit per subslice telling 3094 * whether a subslice is available. Starting with Gen12 we use the 3095 * term "subslice" to refer to what the hardware documentation 3096 * describes as a "dual-subslices." The availability of subslice Y 3097 * in slice X can be queried with the following formula : 3098 * 3099 * .. code:: c 3100 * 3101 * (data[subslice_offset + X * subslice_stride + Y / 8] >> (Y % 8)) & 1 3102 * 3103 * - The EU mask for each subslice in each slice, with one bit per EU 3104 * telling whether an EU is available. The availability of EU Z in 3105 * subslice Y in slice X can be queried with the following formula : 3106 * 3107 * .. code:: c 3108 * 3109 * (data[eu_offset + 3110 * (X * max_subslices + Y) * eu_stride + 3111 * Z / 8 3112 * ] >> (Z % 8)) & 1 3113 */ 3114 __u8 data[]; 3115 }; 3116 3117 /** 3118 * DOC: Engine Discovery uAPI 3119 * 3120 * Engine discovery uAPI is a way of enumerating physical engines present in a 3121 * GPU associated with an open i915 DRM file descriptor. This supersedes the old 3122 * way of using `DRM_IOCTL_I915_GETPARAM` and engine identifiers like 3123 * `I915_PARAM_HAS_BLT`. 3124 * 3125 * The need for this interface came starting with Icelake and newer GPUs, which 3126 * started to establish a pattern of having multiple engines of a same class, 3127 * where not all instances were always completely functionally equivalent. 3128 * 3129 * Entry point for this uapi is `DRM_IOCTL_I915_QUERY` with the 3130 * `DRM_I915_QUERY_ENGINE_INFO` as the queried item id. 3131 * 3132 * Example for getting the list of engines: 3133 * 3134 * .. code-block:: C 3135 * 3136 * struct drm_i915_query_engine_info *info; 3137 * struct drm_i915_query_item item = { 3138 * .query_id = DRM_I915_QUERY_ENGINE_INFO; 3139 * }; 3140 * struct drm_i915_query query = { 3141 * .num_items = 1, 3142 * .items_ptr = (uintptr_t)&item, 3143 * }; 3144 * int err, i; 3145 * 3146 * // First query the size of the blob we need, this needs to be large 3147 * // enough to hold our array of engines. The kernel will fill out the 3148 * // item.length for us, which is the number of bytes we need. 3149 * // 3150 * // Alternatively a large buffer can be allocated straight away enabling 3151 * // querying in one pass, in which case item.length should contain the 3152 * // length of the provided buffer. 3153 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3154 * if (err) ... 3155 * 3156 * info = calloc(1, item.length); 3157 * // Now that we allocated the required number of bytes, we call the ioctl 3158 * // again, this time with the data_ptr pointing to our newly allocated 3159 * // blob, which the kernel can then populate with info on all engines. 3160 * item.data_ptr = (uintptr_t)&info, 3161 * 3162 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3163 * if (err) ... 3164 * 3165 * // We can now access each engine in the array 3166 * for (i = 0; i < info->num_engines; i++) { 3167 * struct drm_i915_engine_info einfo = info->engines[i]; 3168 * u16 class = einfo.engine.class; 3169 * u16 instance = einfo.engine.instance; 3170 * .... 3171 * } 3172 * 3173 * free(info); 3174 * 3175 * Each of the enumerated engines, apart from being defined by its class and 3176 * instance (see `struct i915_engine_class_instance`), also can have flags and 3177 * capabilities defined as documented in i915_drm.h. 3178 * 3179 * For instance video engines which support HEVC encoding will have the 3180 * `I915_VIDEO_CLASS_CAPABILITY_HEVC` capability bit set. 3181 * 3182 * Engine discovery only fully comes to its own when combined with the new way 3183 * of addressing engines when submitting batch buffers using contexts with 3184 * engine maps configured. 3185 */ 3186 3187 /** 3188 * struct drm_i915_engine_info 3189 * 3190 * Describes one engine and it's capabilities as known to the driver. 3191 */ 3192 struct drm_i915_engine_info { 3193 /** @engine: Engine class and instance. */ 3194 struct i915_engine_class_instance engine; 3195 3196 /** @rsvd0: Reserved field. */ 3197 __u32 rsvd0; 3198 3199 /** @flags: Engine flags. */ 3200 __u64 flags; 3201 #define I915_ENGINE_INFO_HAS_LOGICAL_INSTANCE (1 << 0) 3202 3203 /** @capabilities: Capabilities of this engine. */ 3204 __u64 capabilities; 3205 #define I915_VIDEO_CLASS_CAPABILITY_HEVC (1 << 0) 3206 #define I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC (1 << 1) 3207 3208 /** @logical_instance: Logical instance of engine */ 3209 __u16 logical_instance; 3210 3211 /** @rsvd1: Reserved fields. */ 3212 __u16 rsvd1[3]; 3213 /** @rsvd2: Reserved fields. */ 3214 __u64 rsvd2[3]; 3215 }; 3216 3217 /** 3218 * struct drm_i915_query_engine_info 3219 * 3220 * Engine info query enumerates all engines known to the driver by filling in 3221 * an array of struct drm_i915_engine_info structures. 3222 */ 3223 struct drm_i915_query_engine_info { 3224 /** @num_engines: Number of struct drm_i915_engine_info structs following. */ 3225 __u32 num_engines; 3226 3227 /** @rsvd: MBZ */ 3228 __u32 rsvd[3]; 3229 3230 /** @engines: Marker for drm_i915_engine_info structures. */ 3231 struct drm_i915_engine_info engines[]; 3232 }; 3233 3234 /** 3235 * struct drm_i915_query_perf_config 3236 * 3237 * Data written by the kernel with query %DRM_I915_QUERY_PERF_CONFIG and 3238 * %DRM_I915_QUERY_GEOMETRY_SUBSLICES. 3239 */ 3240 struct drm_i915_query_perf_config { 3241 union { 3242 /** 3243 * @n_configs: 3244 * 3245 * When &drm_i915_query_item.flags == 3246 * %DRM_I915_QUERY_PERF_CONFIG_LIST, i915 sets this fields to 3247 * the number of configurations available. 3248 */ 3249 __u64 n_configs; 3250 3251 /** 3252 * @config: 3253 * 3254 * When &drm_i915_query_item.flags == 3255 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID, i915 will use the 3256 * value in this field as configuration identifier to decide 3257 * what data to write into config_ptr. 3258 */ 3259 __u64 config; 3260 3261 /** 3262 * @uuid: 3263 * 3264 * When &drm_i915_query_item.flags == 3265 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID, i915 will use the 3266 * value in this field as configuration identifier to decide 3267 * what data to write into config_ptr. 3268 * 3269 * String formatted like "%08x-%04x-%04x-%04x-%012x" 3270 */ 3271 char uuid[36]; 3272 }; 3273 3274 /** 3275 * @flags: 3276 * 3277 * Unused for now. Must be cleared to zero. 3278 */ 3279 __u32 flags; 3280 3281 /** 3282 * @data: 3283 * 3284 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_LIST, 3285 * i915 will write an array of __u64 of configuration identifiers. 3286 * 3287 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_DATA, 3288 * i915 will write a struct drm_i915_perf_oa_config. If the following 3289 * fields of struct drm_i915_perf_oa_config are not set to 0, i915 will 3290 * write into the associated pointers the values of submitted when the 3291 * configuration was created : 3292 * 3293 * - &drm_i915_perf_oa_config.n_mux_regs 3294 * - &drm_i915_perf_oa_config.n_boolean_regs 3295 * - &drm_i915_perf_oa_config.n_flex_regs 3296 */ 3297 __u8 data[]; 3298 }; 3299 3300 /** 3301 * enum drm_i915_gem_memory_class - Supported memory classes 3302 */ 3303 enum drm_i915_gem_memory_class { 3304 /** @I915_MEMORY_CLASS_SYSTEM: System memory */ 3305 I915_MEMORY_CLASS_SYSTEM = 0, 3306 /** @I915_MEMORY_CLASS_DEVICE: Device local-memory */ 3307 I915_MEMORY_CLASS_DEVICE, 3308 }; 3309 3310 /** 3311 * struct drm_i915_gem_memory_class_instance - Identify particular memory region 3312 */ 3313 struct drm_i915_gem_memory_class_instance { 3314 /** @memory_class: See enum drm_i915_gem_memory_class */ 3315 __u16 memory_class; 3316 3317 /** @memory_instance: Which instance */ 3318 __u16 memory_instance; 3319 }; 3320 3321 /** 3322 * struct drm_i915_memory_region_info - Describes one region as known to the 3323 * driver. 3324 * 3325 * Note this is using both struct drm_i915_query_item and struct drm_i915_query. 3326 * For this new query we are adding the new query id DRM_I915_QUERY_MEMORY_REGIONS 3327 * at &drm_i915_query_item.query_id. 3328 */ 3329 struct drm_i915_memory_region_info { 3330 /** @region: The class:instance pair encoding */ 3331 struct drm_i915_gem_memory_class_instance region; 3332 3333 /** @rsvd0: MBZ */ 3334 __u32 rsvd0; 3335 3336 /** 3337 * @probed_size: Memory probed by the driver 3338 * 3339 * Note that it should not be possible to ever encounter a zero value 3340 * here, also note that no current region type will ever return -1 here. 3341 * Although for future region types, this might be a possibility. The 3342 * same applies to the other size fields. 3343 */ 3344 __u64 probed_size; 3345 3346 /** 3347 * @unallocated_size: Estimate of memory remaining 3348 * 3349 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable accounting. 3350 * Without this (or if this is an older kernel) the value here will 3351 * always equal the @probed_size. Note this is only currently tracked 3352 * for I915_MEMORY_CLASS_DEVICE regions (for other types the value here 3353 * will always equal the @probed_size). 3354 */ 3355 __u64 unallocated_size; 3356 3357 union { 3358 /** @rsvd1: MBZ */ 3359 __u64 rsvd1[8]; 3360 struct { 3361 /** 3362 * @probed_cpu_visible_size: Memory probed by the driver 3363 * that is CPU accessible. 3364 * 3365 * This will be always be <= @probed_size, and the 3366 * remainder (if there is any) will not be CPU 3367 * accessible. 3368 * 3369 * On systems without small BAR, the @probed_size will 3370 * always equal the @probed_cpu_visible_size, since all 3371 * of it will be CPU accessible. 3372 * 3373 * Note this is only tracked for 3374 * I915_MEMORY_CLASS_DEVICE regions (for other types the 3375 * value here will always equal the @probed_size). 3376 * 3377 * Note that if the value returned here is zero, then 3378 * this must be an old kernel which lacks the relevant 3379 * small-bar uAPI support (including 3380 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS), but on 3381 * such systems we should never actually end up with a 3382 * small BAR configuration, assuming we are able to load 3383 * the kernel module. Hence it should be safe to treat 3384 * this the same as when @probed_cpu_visible_size == 3385 * @probed_size. 3386 */ 3387 __u64 probed_cpu_visible_size; 3388 3389 /** 3390 * @unallocated_cpu_visible_size: Estimate of CPU 3391 * visible memory remaining. 3392 * 3393 * Note this is only tracked for 3394 * I915_MEMORY_CLASS_DEVICE regions (for other types the 3395 * value here will always equal the 3396 * @probed_cpu_visible_size). 3397 * 3398 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable 3399 * accounting. Without this the value here will always 3400 * equal the @probed_cpu_visible_size. Note this is only 3401 * currently tracked for I915_MEMORY_CLASS_DEVICE 3402 * regions (for other types the value here will also 3403 * always equal the @probed_cpu_visible_size). 3404 * 3405 * If this is an older kernel the value here will be 3406 * zero, see also @probed_cpu_visible_size. 3407 */ 3408 __u64 unallocated_cpu_visible_size; 3409 }; 3410 }; 3411 }; 3412 3413 /** 3414 * struct drm_i915_query_memory_regions 3415 * 3416 * The region info query enumerates all regions known to the driver by filling 3417 * in an array of struct drm_i915_memory_region_info structures. 3418 * 3419 * Example for getting the list of supported regions: 3420 * 3421 * .. code-block:: C 3422 * 3423 * struct drm_i915_query_memory_regions *info; 3424 * struct drm_i915_query_item item = { 3425 * .query_id = DRM_I915_QUERY_MEMORY_REGIONS; 3426 * }; 3427 * struct drm_i915_query query = { 3428 * .num_items = 1, 3429 * .items_ptr = (uintptr_t)&item, 3430 * }; 3431 * int err, i; 3432 * 3433 * // First query the size of the blob we need, this needs to be large 3434 * // enough to hold our array of regions. The kernel will fill out the 3435 * // item.length for us, which is the number of bytes we need. 3436 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3437 * if (err) ... 3438 * 3439 * info = calloc(1, item.length); 3440 * // Now that we allocated the required number of bytes, we call the ioctl 3441 * // again, this time with the data_ptr pointing to our newly allocated 3442 * // blob, which the kernel can then populate with the all the region info. 3443 * item.data_ptr = (uintptr_t)&info, 3444 * 3445 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3446 * if (err) ... 3447 * 3448 * // We can now access each region in the array 3449 * for (i = 0; i < info->num_regions; i++) { 3450 * struct drm_i915_memory_region_info mr = info->regions[i]; 3451 * u16 class = mr.region.class; 3452 * u16 instance = mr.region.instance; 3453 * 3454 * .... 3455 * } 3456 * 3457 * free(info); 3458 */ 3459 struct drm_i915_query_memory_regions { 3460 /** @num_regions: Number of supported regions */ 3461 __u32 num_regions; 3462 3463 /** @rsvd: MBZ */ 3464 __u32 rsvd[3]; 3465 3466 /** @regions: Info about each supported region */ 3467 struct drm_i915_memory_region_info regions[]; 3468 }; 3469 3470 /** 3471 * DOC: GuC HWCONFIG blob uAPI 3472 * 3473 * The GuC produces a blob with information about the current device. 3474 * i915 reads this blob from GuC and makes it available via this uAPI. 3475 * 3476 * The format and meaning of the blob content are documented in the 3477 * Programmer's Reference Manual. 3478 */ 3479 3480 /** 3481 * struct drm_i915_gem_create_ext - Existing gem_create behaviour, with added 3482 * extension support using struct i915_user_extension. 3483 * 3484 * Note that new buffer flags should be added here, at least for the stuff that 3485 * is immutable. Previously we would have two ioctls, one to create the object 3486 * with gem_create, and another to apply various parameters, however this 3487 * creates some ambiguity for the params which are considered immutable. Also in 3488 * general we're phasing out the various SET/GET ioctls. 3489 */ 3490 struct drm_i915_gem_create_ext { 3491 /** 3492 * @size: Requested size for the object. 3493 * 3494 * The (page-aligned) allocated size for the object will be returned. 3495 * 3496 * DG2 64K min page size implications: 3497 * 3498 * On discrete platforms, starting from DG2, we have to contend with GTT 3499 * page size restrictions when dealing with I915_MEMORY_CLASS_DEVICE 3500 * objects. Specifically the hardware only supports 64K or larger GTT 3501 * page sizes for such memory. The kernel will already ensure that all 3502 * I915_MEMORY_CLASS_DEVICE memory is allocated using 64K or larger page 3503 * sizes underneath. 3504 * 3505 * Note that the returned size here will always reflect any required 3506 * rounding up done by the kernel, i.e 4K will now become 64K on devices 3507 * such as DG2. The kernel will always select the largest minimum 3508 * page-size for the set of possible placements as the value to use when 3509 * rounding up the @size. 3510 * 3511 * Special DG2 GTT address alignment requirement: 3512 * 3513 * The GTT alignment will also need to be at least 2M for such objects. 3514 * 3515 * Note that due to how the hardware implements 64K GTT page support, we 3516 * have some further complications: 3517 * 3518 * 1) The entire PDE (which covers a 2MB virtual address range), must 3519 * contain only 64K PTEs, i.e mixing 4K and 64K PTEs in the same 3520 * PDE is forbidden by the hardware. 3521 * 3522 * 2) We still need to support 4K PTEs for I915_MEMORY_CLASS_SYSTEM 3523 * objects. 3524 * 3525 * To keep things simple for userland, we mandate that any GTT mappings 3526 * must be aligned to and rounded up to 2MB. The kernel will internally 3527 * pad them out to the next 2MB boundary. As this only wastes virtual 3528 * address space and avoids userland having to copy any needlessly 3529 * complicated PDE sharing scheme (coloring) and only affects DG2, this 3530 * is deemed to be a good compromise. 3531 */ 3532 __u64 size; 3533 3534 /** 3535 * @handle: Returned handle for the object. 3536 * 3537 * Object handles are nonzero. 3538 */ 3539 __u32 handle; 3540 3541 /** 3542 * @flags: Optional flags. 3543 * 3544 * Supported values: 3545 * 3546 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS - Signal to the kernel that 3547 * the object will need to be accessed via the CPU. 3548 * 3549 * Only valid when placing objects in I915_MEMORY_CLASS_DEVICE, and only 3550 * strictly required on configurations where some subset of the device 3551 * memory is directly visible/mappable through the CPU (which we also 3552 * call small BAR), like on some DG2+ systems. Note that this is quite 3553 * undesirable, but due to various factors like the client CPU, BIOS etc 3554 * it's something we can expect to see in the wild. See 3555 * &drm_i915_memory_region_info.probed_cpu_visible_size for how to 3556 * determine if this system applies. 3557 * 3558 * Note that one of the placements MUST be I915_MEMORY_CLASS_SYSTEM, to 3559 * ensure the kernel can always spill the allocation to system memory, 3560 * if the object can't be allocated in the mappable part of 3561 * I915_MEMORY_CLASS_DEVICE. 3562 * 3563 * Also note that since the kernel only supports flat-CCS on objects 3564 * that can *only* be placed in I915_MEMORY_CLASS_DEVICE, we therefore 3565 * don't support I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS together with 3566 * flat-CCS. 3567 * 3568 * Without this hint, the kernel will assume that non-mappable 3569 * I915_MEMORY_CLASS_DEVICE is preferred for this object. Note that the 3570 * kernel can still migrate the object to the mappable part, as a last 3571 * resort, if userspace ever CPU faults this object, but this might be 3572 * expensive, and so ideally should be avoided. 3573 * 3574 * On older kernels which lack the relevant small-bar uAPI support (see 3575 * also &drm_i915_memory_region_info.probed_cpu_visible_size), 3576 * usage of the flag will result in an error, but it should NEVER be 3577 * possible to end up with a small BAR configuration, assuming we can 3578 * also successfully load the i915 kernel module. In such cases the 3579 * entire I915_MEMORY_CLASS_DEVICE region will be CPU accessible, and as 3580 * such there are zero restrictions on where the object can be placed. 3581 */ 3582 #define I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS (1 << 0) 3583 __u32 flags; 3584 3585 /** 3586 * @extensions: The chain of extensions to apply to this object. 3587 * 3588 * This will be useful in the future when we need to support several 3589 * different extensions, and we need to apply more than one when 3590 * creating the object. See struct i915_user_extension. 3591 * 3592 * If we don't supply any extensions then we get the same old gem_create 3593 * behaviour. 3594 * 3595 * For I915_GEM_CREATE_EXT_MEMORY_REGIONS usage see 3596 * struct drm_i915_gem_create_ext_memory_regions. 3597 * 3598 * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see 3599 * struct drm_i915_gem_create_ext_protected_content. 3600 */ 3601 #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 3602 #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 3603 __u64 extensions; 3604 }; 3605 3606 /** 3607 * struct drm_i915_gem_create_ext_memory_regions - The 3608 * I915_GEM_CREATE_EXT_MEMORY_REGIONS extension. 3609 * 3610 * Set the object with the desired set of placements/regions in priority 3611 * order. Each entry must be unique and supported by the device. 3612 * 3613 * This is provided as an array of struct drm_i915_gem_memory_class_instance, or 3614 * an equivalent layout of class:instance pair encodings. See struct 3615 * drm_i915_query_memory_regions and DRM_I915_QUERY_MEMORY_REGIONS for how to 3616 * query the supported regions for a device. 3617 * 3618 * As an example, on discrete devices, if we wish to set the placement as 3619 * device local-memory we can do something like: 3620 * 3621 * .. code-block:: C 3622 * 3623 * struct drm_i915_gem_memory_class_instance region_lmem = { 3624 * .memory_class = I915_MEMORY_CLASS_DEVICE, 3625 * .memory_instance = 0, 3626 * }; 3627 * struct drm_i915_gem_create_ext_memory_regions regions = { 3628 * .base = { .name = I915_GEM_CREATE_EXT_MEMORY_REGIONS }, 3629 * .regions = (uintptr_t)®ion_lmem, 3630 * .num_regions = 1, 3631 * }; 3632 * struct drm_i915_gem_create_ext create_ext = { 3633 * .size = 16 * PAGE_SIZE, 3634 * .extensions = (uintptr_t)®ions, 3635 * }; 3636 * 3637 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3638 * if (err) ... 3639 * 3640 * At which point we get the object handle in &drm_i915_gem_create_ext.handle, 3641 * along with the final object size in &drm_i915_gem_create_ext.size, which 3642 * should account for any rounding up, if required. 3643 * 3644 * Note that userspace has no means of knowing the current backing region 3645 * for objects where @num_regions is larger than one. The kernel will only 3646 * ensure that the priority order of the @regions array is honoured, either 3647 * when initially placing the object, or when moving memory around due to 3648 * memory pressure 3649 * 3650 * On Flat-CCS capable HW, compression is supported for the objects residing 3651 * in I915_MEMORY_CLASS_DEVICE. When such objects (compressed) have other 3652 * memory class in @regions and migrated (by i915, due to memory 3653 * constraints) to the non I915_MEMORY_CLASS_DEVICE region, then i915 needs to 3654 * decompress the content. But i915 doesn't have the required information to 3655 * decompress the userspace compressed objects. 3656 * 3657 * So i915 supports Flat-CCS, on the objects which can reside only on 3658 * I915_MEMORY_CLASS_DEVICE regions. 3659 */ 3660 struct drm_i915_gem_create_ext_memory_regions { 3661 /** @base: Extension link. See struct i915_user_extension. */ 3662 struct i915_user_extension base; 3663 3664 /** @pad: MBZ */ 3665 __u32 pad; 3666 /** @num_regions: Number of elements in the @regions array. */ 3667 __u32 num_regions; 3668 /** 3669 * @regions: The regions/placements array. 3670 * 3671 * An array of struct drm_i915_gem_memory_class_instance. 3672 */ 3673 __u64 regions; 3674 }; 3675 3676 /** 3677 * struct drm_i915_gem_create_ext_protected_content - The 3678 * I915_OBJECT_PARAM_PROTECTED_CONTENT extension. 3679 * 3680 * If this extension is provided, buffer contents are expected to be protected 3681 * by PXP encryption and require decryption for scan out and processing. This 3682 * is only possible on platforms that have PXP enabled, on all other scenarios 3683 * using this extension will cause the ioctl to fail and return -ENODEV. The 3684 * flags parameter is reserved for future expansion and must currently be set 3685 * to zero. 3686 * 3687 * The buffer contents are considered invalid after a PXP session teardown. 3688 * 3689 * The encryption is guaranteed to be processed correctly only if the object 3690 * is submitted with a context created using the 3691 * I915_CONTEXT_PARAM_PROTECTED_CONTENT flag. This will also enable extra checks 3692 * at submission time on the validity of the objects involved. 3693 * 3694 * Below is an example on how to create a protected object: 3695 * 3696 * .. code-block:: C 3697 * 3698 * struct drm_i915_gem_create_ext_protected_content protected_ext = { 3699 * .base = { .name = I915_GEM_CREATE_EXT_PROTECTED_CONTENT }, 3700 * .flags = 0, 3701 * }; 3702 * struct drm_i915_gem_create_ext create_ext = { 3703 * .size = PAGE_SIZE, 3704 * .extensions = (uintptr_t)&protected_ext, 3705 * }; 3706 * 3707 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3708 * if (err) ... 3709 */ 3710 struct drm_i915_gem_create_ext_protected_content { 3711 /** @base: Extension link. See struct i915_user_extension. */ 3712 struct i915_user_extension base; 3713 /** @flags: reserved for future usage, currently MBZ */ 3714 __u32 flags; 3715 }; 3716 3717 /* ID of the protected content session managed by i915 when PXP is active */ 3718 #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf 3719 3720 #if defined(__cplusplus) 3721 } 3722 #endif 3723 3724 #endif /* _I915_DRM_H_ */ 3725