1 /* 2 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sub license, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * The above copyright notice and this permission notice (including the 14 * next paragraph) shall be included in all copies or substantial portions 15 * of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 20 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 22 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 23 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 24 * 25 */ 26 27 #ifndef _UAPI_I915_DRM_H_ 28 #define _UAPI_I915_DRM_H_ 29 30 #include "drm.h" 31 32 #if defined(__cplusplus) 33 extern "C" { 34 #endif 35 36 /* Please note that modifications to all structs defined here are 37 * subject to backwards-compatibility constraints. 38 */ 39 40 /** 41 * DOC: uevents generated by i915 on it's device node 42 * 43 * I915_L3_PARITY_UEVENT - Generated when the driver receives a parity mismatch 44 * event from the gpu l3 cache. Additional information supplied is ROW, 45 * BANK, SUBBANK, SLICE of the affected cacheline. Userspace should keep 46 * track of these events and if a specific cache-line seems to have a 47 * persistent error remap it with the l3 remapping tool supplied in 48 * intel-gpu-tools. The value supplied with the event is always 1. 49 * 50 * I915_ERROR_UEVENT - Generated upon error detection, currently only via 51 * hangcheck. The error detection event is a good indicator of when things 52 * began to go badly. The value supplied with the event is a 1 upon error 53 * detection, and a 0 upon reset completion, signifying no more error 54 * exists. NOTE: Disabling hangcheck or reset via module parameter will 55 * cause the related events to not be seen. 56 * 57 * I915_RESET_UEVENT - Event is generated just before an attempt to reset the 58 * GPU. The value supplied with the event is always 1. NOTE: Disable 59 * reset via module parameter will cause this event to not be seen. 60 */ 61 #define I915_L3_PARITY_UEVENT "L3_PARITY_ERROR" 62 #define I915_ERROR_UEVENT "ERROR" 63 #define I915_RESET_UEVENT "RESET" 64 65 /** 66 * struct i915_user_extension - Base class for defining a chain of extensions 67 * 68 * Many interfaces need to grow over time. In most cases we can simply 69 * extend the struct and have userspace pass in more data. Another option, 70 * as demonstrated by Vulkan's approach to providing extensions for forward 71 * and backward compatibility, is to use a list of optional structs to 72 * provide those extra details. 73 * 74 * The key advantage to using an extension chain is that it allows us to 75 * redefine the interface more easily than an ever growing struct of 76 * increasing complexity, and for large parts of that interface to be 77 * entirely optional. The downside is more pointer chasing; chasing across 78 * the __user boundary with pointers encapsulated inside u64. 79 * 80 * Example chaining: 81 * 82 * .. code-block:: C 83 * 84 * struct i915_user_extension ext3 { 85 * .next_extension = 0, // end 86 * .name = ..., 87 * }; 88 * struct i915_user_extension ext2 { 89 * .next_extension = (uintptr_t)&ext3, 90 * .name = ..., 91 * }; 92 * struct i915_user_extension ext1 { 93 * .next_extension = (uintptr_t)&ext2, 94 * .name = ..., 95 * }; 96 * 97 * Typically the struct i915_user_extension would be embedded in some uAPI 98 * struct, and in this case we would feed it the head of the chain(i.e ext1), 99 * which would then apply all of the above extensions. 100 * 101 */ 102 struct i915_user_extension { 103 /** 104 * @next_extension: 105 * 106 * Pointer to the next struct i915_user_extension, or zero if the end. 107 */ 108 __u64 next_extension; 109 /** 110 * @name: Name of the extension. 111 * 112 * Note that the name here is just some integer. 113 * 114 * Also note that the name space for this is not global for the whole 115 * driver, but rather its scope/meaning is limited to the specific piece 116 * of uAPI which has embedded the struct i915_user_extension. 117 */ 118 __u32 name; 119 /** 120 * @flags: MBZ 121 * 122 * All undefined bits must be zero. 123 */ 124 __u32 flags; 125 /** 126 * @rsvd: MBZ 127 * 128 * Reserved for future use; must be zero. 129 */ 130 __u32 rsvd[4]; 131 }; 132 133 /* 134 * MOCS indexes used for GPU surfaces, defining the cacheability of the 135 * surface data and the coherency for this data wrt. CPU vs. GPU accesses. 136 */ 137 enum i915_mocs_table_index { 138 /* 139 * Not cached anywhere, coherency between CPU and GPU accesses is 140 * guaranteed. 141 */ 142 I915_MOCS_UNCACHED, 143 /* 144 * Cacheability and coherency controlled by the kernel automatically 145 * based on the DRM_I915_GEM_SET_CACHING IOCTL setting and the current 146 * usage of the surface (used for display scanout or not). 147 */ 148 I915_MOCS_PTE, 149 /* 150 * Cached in all GPU caches available on the platform. 151 * Coherency between CPU and GPU accesses to the surface is not 152 * guaranteed without extra synchronization. 153 */ 154 I915_MOCS_CACHED, 155 }; 156 157 /** 158 * enum drm_i915_gem_engine_class - uapi engine type enumeration 159 * 160 * Different engines serve different roles, and there may be more than one 161 * engine serving each role. This enum provides a classification of the role 162 * of the engine, which may be used when requesting operations to be performed 163 * on a certain subset of engines, or for providing information about that 164 * group. 165 */ 166 enum drm_i915_gem_engine_class { 167 /** 168 * @I915_ENGINE_CLASS_RENDER: 169 * 170 * Render engines support instructions used for 3D, Compute (GPGPU), 171 * and programmable media workloads. These instructions fetch data and 172 * dispatch individual work items to threads that operate in parallel. 173 * The threads run small programs (called "kernels" or "shaders") on 174 * the GPU's execution units (EUs). 175 */ 176 I915_ENGINE_CLASS_RENDER = 0, 177 178 /** 179 * @I915_ENGINE_CLASS_COPY: 180 * 181 * Copy engines (also referred to as "blitters") support instructions 182 * that move blocks of data from one location in memory to another, 183 * or that fill a specified location of memory with fixed data. 184 * Copy engines can perform pre-defined logical or bitwise operations 185 * on the source, destination, or pattern data. 186 */ 187 I915_ENGINE_CLASS_COPY = 1, 188 189 /** 190 * @I915_ENGINE_CLASS_VIDEO: 191 * 192 * Video engines (also referred to as "bit stream decode" (BSD) or 193 * "vdbox") support instructions that perform fixed-function media 194 * decode and encode. 195 */ 196 I915_ENGINE_CLASS_VIDEO = 2, 197 198 /** 199 * @I915_ENGINE_CLASS_VIDEO_ENHANCE: 200 * 201 * Video enhancement engines (also referred to as "vebox") support 202 * instructions related to image enhancement. 203 */ 204 I915_ENGINE_CLASS_VIDEO_ENHANCE = 3, 205 206 /** 207 * @I915_ENGINE_CLASS_COMPUTE: 208 * 209 * Compute engines support a subset of the instructions available 210 * on render engines: compute engines support Compute (GPGPU) and 211 * programmable media workloads, but do not support the 3D pipeline. 212 */ 213 I915_ENGINE_CLASS_COMPUTE = 4, 214 215 /* Values in this enum should be kept compact. */ 216 217 /** 218 * @I915_ENGINE_CLASS_INVALID: 219 * 220 * Placeholder value to represent an invalid engine class assignment. 221 */ 222 I915_ENGINE_CLASS_INVALID = -1 223 }; 224 225 /** 226 * struct i915_engine_class_instance - Engine class/instance identifier 227 * 228 * There may be more than one engine fulfilling any role within the system. 229 * Each engine of a class is given a unique instance number and therefore 230 * any engine can be specified by its class:instance tuplet. APIs that allow 231 * access to any engine in the system will use struct i915_engine_class_instance 232 * for this identification. 233 */ 234 struct i915_engine_class_instance { 235 /** 236 * @engine_class: 237 * 238 * Engine class from enum drm_i915_gem_engine_class 239 */ 240 __u16 engine_class; 241 #define I915_ENGINE_CLASS_INVALID_NONE -1 242 #define I915_ENGINE_CLASS_INVALID_VIRTUAL -2 243 244 /** 245 * @engine_instance: 246 * 247 * Engine instance. 248 */ 249 __u16 engine_instance; 250 }; 251 252 /** 253 * DOC: perf_events exposed by i915 through /sys/bus/event_sources/drivers/i915 254 * 255 */ 256 257 enum drm_i915_pmu_engine_sample { 258 I915_SAMPLE_BUSY = 0, 259 I915_SAMPLE_WAIT = 1, 260 I915_SAMPLE_SEMA = 2 261 }; 262 263 #define I915_PMU_SAMPLE_BITS (4) 264 #define I915_PMU_SAMPLE_MASK (0xf) 265 #define I915_PMU_SAMPLE_INSTANCE_BITS (8) 266 #define I915_PMU_CLASS_SHIFT \ 267 (I915_PMU_SAMPLE_BITS + I915_PMU_SAMPLE_INSTANCE_BITS) 268 269 #define __I915_PMU_ENGINE(class, instance, sample) \ 270 ((class) << I915_PMU_CLASS_SHIFT | \ 271 (instance) << I915_PMU_SAMPLE_BITS | \ 272 (sample)) 273 274 #define I915_PMU_ENGINE_BUSY(class, instance) \ 275 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_BUSY) 276 277 #define I915_PMU_ENGINE_WAIT(class, instance) \ 278 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_WAIT) 279 280 #define I915_PMU_ENGINE_SEMA(class, instance) \ 281 __I915_PMU_ENGINE(class, instance, I915_SAMPLE_SEMA) 282 283 /* 284 * Top 4 bits of every non-engine counter are GT id. 285 */ 286 #define __I915_PMU_GT_SHIFT (60) 287 288 #define ___I915_PMU_OTHER(gt, x) \ 289 (((__u64)__I915_PMU_ENGINE(0xff, 0xff, 0xf) + 1 + (x)) | \ 290 ((__u64)(gt) << __I915_PMU_GT_SHIFT)) 291 292 #define __I915_PMU_OTHER(x) ___I915_PMU_OTHER(0, x) 293 294 #define I915_PMU_ACTUAL_FREQUENCY __I915_PMU_OTHER(0) 295 #define I915_PMU_REQUESTED_FREQUENCY __I915_PMU_OTHER(1) 296 #define I915_PMU_INTERRUPTS __I915_PMU_OTHER(2) 297 #define I915_PMU_RC6_RESIDENCY __I915_PMU_OTHER(3) 298 #define I915_PMU_SOFTWARE_GT_AWAKE_TIME __I915_PMU_OTHER(4) 299 300 #define I915_PMU_LAST /* Deprecated - do not use */ I915_PMU_RC6_RESIDENCY 301 302 #define __I915_PMU_ACTUAL_FREQUENCY(gt) ___I915_PMU_OTHER(gt, 0) 303 #define __I915_PMU_REQUESTED_FREQUENCY(gt) ___I915_PMU_OTHER(gt, 1) 304 #define __I915_PMU_INTERRUPTS(gt) ___I915_PMU_OTHER(gt, 2) 305 #define __I915_PMU_RC6_RESIDENCY(gt) ___I915_PMU_OTHER(gt, 3) 306 #define __I915_PMU_SOFTWARE_GT_AWAKE_TIME(gt) ___I915_PMU_OTHER(gt, 4) 307 308 /* Each region is a minimum of 16k, and there are at most 255 of them. 309 */ 310 #define I915_NR_TEX_REGIONS 255 /* table size 2k - maximum due to use 311 * of chars for next/prev indices */ 312 #define I915_LOG_MIN_TEX_REGION_SIZE 14 313 314 typedef struct _drm_i915_init { 315 enum { 316 I915_INIT_DMA = 0x01, 317 I915_CLEANUP_DMA = 0x02, 318 I915_RESUME_DMA = 0x03 319 } func; 320 unsigned int mmio_offset; 321 int sarea_priv_offset; 322 unsigned int ring_start; 323 unsigned int ring_end; 324 unsigned int ring_size; 325 unsigned int front_offset; 326 unsigned int back_offset; 327 unsigned int depth_offset; 328 unsigned int w; 329 unsigned int h; 330 unsigned int pitch; 331 unsigned int pitch_bits; 332 unsigned int back_pitch; 333 unsigned int depth_pitch; 334 unsigned int cpp; 335 unsigned int chipset; 336 } drm_i915_init_t; 337 338 typedef struct _drm_i915_sarea { 339 struct drm_tex_region texList[I915_NR_TEX_REGIONS + 1]; 340 int last_upload; /* last time texture was uploaded */ 341 int last_enqueue; /* last time a buffer was enqueued */ 342 int last_dispatch; /* age of the most recently dispatched buffer */ 343 int ctxOwner; /* last context to upload state */ 344 int texAge; 345 int pf_enabled; /* is pageflipping allowed? */ 346 int pf_active; 347 int pf_current_page; /* which buffer is being displayed? */ 348 int perf_boxes; /* performance boxes to be displayed */ 349 int width, height; /* screen size in pixels */ 350 351 drm_handle_t front_handle; 352 int front_offset; 353 int front_size; 354 355 drm_handle_t back_handle; 356 int back_offset; 357 int back_size; 358 359 drm_handle_t depth_handle; 360 int depth_offset; 361 int depth_size; 362 363 drm_handle_t tex_handle; 364 int tex_offset; 365 int tex_size; 366 int log_tex_granularity; 367 int pitch; 368 int rotation; /* 0, 90, 180 or 270 */ 369 int rotated_offset; 370 int rotated_size; 371 int rotated_pitch; 372 int virtualX, virtualY; 373 374 unsigned int front_tiled; 375 unsigned int back_tiled; 376 unsigned int depth_tiled; 377 unsigned int rotated_tiled; 378 unsigned int rotated2_tiled; 379 380 int pipeA_x; 381 int pipeA_y; 382 int pipeA_w; 383 int pipeA_h; 384 int pipeB_x; 385 int pipeB_y; 386 int pipeB_w; 387 int pipeB_h; 388 389 /* fill out some space for old userspace triple buffer */ 390 drm_handle_t unused_handle; 391 __u32 unused1, unused2, unused3; 392 393 /* buffer object handles for static buffers. May change 394 * over the lifetime of the client. 395 */ 396 __u32 front_bo_handle; 397 __u32 back_bo_handle; 398 __u32 unused_bo_handle; 399 __u32 depth_bo_handle; 400 401 } drm_i915_sarea_t; 402 403 /* due to userspace building against these headers we need some compat here */ 404 #define planeA_x pipeA_x 405 #define planeA_y pipeA_y 406 #define planeA_w pipeA_w 407 #define planeA_h pipeA_h 408 #define planeB_x pipeB_x 409 #define planeB_y pipeB_y 410 #define planeB_w pipeB_w 411 #define planeB_h pipeB_h 412 413 /* Flags for perf_boxes 414 */ 415 #define I915_BOX_RING_EMPTY 0x1 416 #define I915_BOX_FLIP 0x2 417 #define I915_BOX_WAIT 0x4 418 #define I915_BOX_TEXTURE_LOAD 0x8 419 #define I915_BOX_LOST_CONTEXT 0x10 420 421 /* 422 * i915 specific ioctls. 423 * 424 * The device specific ioctl range is [DRM_COMMAND_BASE, DRM_COMMAND_END) ie 425 * [0x40, 0xa0) (a0 is excluded). The numbers below are defined as offset 426 * against DRM_COMMAND_BASE and should be between [0x0, 0x60). 427 */ 428 #define DRM_I915_INIT 0x00 429 #define DRM_I915_FLUSH 0x01 430 #define DRM_I915_FLIP 0x02 431 #define DRM_I915_BATCHBUFFER 0x03 432 #define DRM_I915_IRQ_EMIT 0x04 433 #define DRM_I915_IRQ_WAIT 0x05 434 #define DRM_I915_GETPARAM 0x06 435 #define DRM_I915_SETPARAM 0x07 436 #define DRM_I915_ALLOC 0x08 437 #define DRM_I915_FREE 0x09 438 #define DRM_I915_INIT_HEAP 0x0a 439 #define DRM_I915_CMDBUFFER 0x0b 440 #define DRM_I915_DESTROY_HEAP 0x0c 441 #define DRM_I915_SET_VBLANK_PIPE 0x0d 442 #define DRM_I915_GET_VBLANK_PIPE 0x0e 443 #define DRM_I915_VBLANK_SWAP 0x0f 444 #define DRM_I915_HWS_ADDR 0x11 445 #define DRM_I915_GEM_INIT 0x13 446 #define DRM_I915_GEM_EXECBUFFER 0x14 447 #define DRM_I915_GEM_PIN 0x15 448 #define DRM_I915_GEM_UNPIN 0x16 449 #define DRM_I915_GEM_BUSY 0x17 450 #define DRM_I915_GEM_THROTTLE 0x18 451 #define DRM_I915_GEM_ENTERVT 0x19 452 #define DRM_I915_GEM_LEAVEVT 0x1a 453 #define DRM_I915_GEM_CREATE 0x1b 454 #define DRM_I915_GEM_PREAD 0x1c 455 #define DRM_I915_GEM_PWRITE 0x1d 456 #define DRM_I915_GEM_MMAP 0x1e 457 #define DRM_I915_GEM_SET_DOMAIN 0x1f 458 #define DRM_I915_GEM_SW_FINISH 0x20 459 #define DRM_I915_GEM_SET_TILING 0x21 460 #define DRM_I915_GEM_GET_TILING 0x22 461 #define DRM_I915_GEM_GET_APERTURE 0x23 462 #define DRM_I915_GEM_MMAP_GTT 0x24 463 #define DRM_I915_GET_PIPE_FROM_CRTC_ID 0x25 464 #define DRM_I915_GEM_MADVISE 0x26 465 #define DRM_I915_OVERLAY_PUT_IMAGE 0x27 466 #define DRM_I915_OVERLAY_ATTRS 0x28 467 #define DRM_I915_GEM_EXECBUFFER2 0x29 468 #define DRM_I915_GEM_EXECBUFFER2_WR DRM_I915_GEM_EXECBUFFER2 469 #define DRM_I915_GET_SPRITE_COLORKEY 0x2a 470 #define DRM_I915_SET_SPRITE_COLORKEY 0x2b 471 #define DRM_I915_GEM_WAIT 0x2c 472 #define DRM_I915_GEM_CONTEXT_CREATE 0x2d 473 #define DRM_I915_GEM_CONTEXT_DESTROY 0x2e 474 #define DRM_I915_GEM_SET_CACHING 0x2f 475 #define DRM_I915_GEM_GET_CACHING 0x30 476 #define DRM_I915_REG_READ 0x31 477 #define DRM_I915_GET_RESET_STATS 0x32 478 #define DRM_I915_GEM_USERPTR 0x33 479 #define DRM_I915_GEM_CONTEXT_GETPARAM 0x34 480 #define DRM_I915_GEM_CONTEXT_SETPARAM 0x35 481 #define DRM_I915_PERF_OPEN 0x36 482 #define DRM_I915_PERF_ADD_CONFIG 0x37 483 #define DRM_I915_PERF_REMOVE_CONFIG 0x38 484 #define DRM_I915_QUERY 0x39 485 #define DRM_I915_GEM_VM_CREATE 0x3a 486 #define DRM_I915_GEM_VM_DESTROY 0x3b 487 #define DRM_I915_GEM_CREATE_EXT 0x3c 488 /* Must be kept compact -- no holes */ 489 490 #define DRM_IOCTL_I915_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t) 491 #define DRM_IOCTL_I915_FLUSH DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH) 492 #define DRM_IOCTL_I915_FLIP DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLIP) 493 #define DRM_IOCTL_I915_BATCHBUFFER DRM_IOW( DRM_COMMAND_BASE + DRM_I915_BATCHBUFFER, drm_i915_batchbuffer_t) 494 #define DRM_IOCTL_I915_IRQ_EMIT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_IRQ_EMIT, drm_i915_irq_emit_t) 495 #define DRM_IOCTL_I915_IRQ_WAIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_IRQ_WAIT, drm_i915_irq_wait_t) 496 #define DRM_IOCTL_I915_GETPARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GETPARAM, drm_i915_getparam_t) 497 #define DRM_IOCTL_I915_SETPARAM DRM_IOW( DRM_COMMAND_BASE + DRM_I915_SETPARAM, drm_i915_setparam_t) 498 #define DRM_IOCTL_I915_ALLOC DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_ALLOC, drm_i915_mem_alloc_t) 499 #define DRM_IOCTL_I915_FREE DRM_IOW( DRM_COMMAND_BASE + DRM_I915_FREE, drm_i915_mem_free_t) 500 #define DRM_IOCTL_I915_INIT_HEAP DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT_HEAP, drm_i915_mem_init_heap_t) 501 #define DRM_IOCTL_I915_CMDBUFFER DRM_IOW( DRM_COMMAND_BASE + DRM_I915_CMDBUFFER, drm_i915_cmdbuffer_t) 502 #define DRM_IOCTL_I915_DESTROY_HEAP DRM_IOW( DRM_COMMAND_BASE + DRM_I915_DESTROY_HEAP, drm_i915_mem_destroy_heap_t) 503 #define DRM_IOCTL_I915_SET_VBLANK_PIPE DRM_IOW( DRM_COMMAND_BASE + DRM_I915_SET_VBLANK_PIPE, drm_i915_vblank_pipe_t) 504 #define DRM_IOCTL_I915_GET_VBLANK_PIPE DRM_IOR( DRM_COMMAND_BASE + DRM_I915_GET_VBLANK_PIPE, drm_i915_vblank_pipe_t) 505 #define DRM_IOCTL_I915_VBLANK_SWAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_VBLANK_SWAP, drm_i915_vblank_swap_t) 506 #define DRM_IOCTL_I915_HWS_ADDR DRM_IOW(DRM_COMMAND_BASE + DRM_I915_HWS_ADDR, struct drm_i915_gem_init) 507 #define DRM_IOCTL_I915_GEM_INIT DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_INIT, struct drm_i915_gem_init) 508 #define DRM_IOCTL_I915_GEM_EXECBUFFER DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER, struct drm_i915_gem_execbuffer) 509 #define DRM_IOCTL_I915_GEM_EXECBUFFER2 DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER2, struct drm_i915_gem_execbuffer2) 510 #define DRM_IOCTL_I915_GEM_EXECBUFFER2_WR DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER2_WR, struct drm_i915_gem_execbuffer2) 511 #define DRM_IOCTL_I915_GEM_PIN DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_PIN, struct drm_i915_gem_pin) 512 #define DRM_IOCTL_I915_GEM_UNPIN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_UNPIN, struct drm_i915_gem_unpin) 513 #define DRM_IOCTL_I915_GEM_BUSY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_BUSY, struct drm_i915_gem_busy) 514 #define DRM_IOCTL_I915_GEM_SET_CACHING DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_SET_CACHING, struct drm_i915_gem_caching) 515 #define DRM_IOCTL_I915_GEM_GET_CACHING DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_GET_CACHING, struct drm_i915_gem_caching) 516 #define DRM_IOCTL_I915_GEM_THROTTLE DRM_IO ( DRM_COMMAND_BASE + DRM_I915_GEM_THROTTLE) 517 #define DRM_IOCTL_I915_GEM_ENTERVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_ENTERVT) 518 #define DRM_IOCTL_I915_GEM_LEAVEVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_LEAVEVT) 519 #define DRM_IOCTL_I915_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE, struct drm_i915_gem_create) 520 #define DRM_IOCTL_I915_GEM_CREATE_EXT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE_EXT, struct drm_i915_gem_create_ext) 521 #define DRM_IOCTL_I915_GEM_PREAD DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PREAD, struct drm_i915_gem_pread) 522 #define DRM_IOCTL_I915_GEM_PWRITE DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PWRITE, struct drm_i915_gem_pwrite) 523 #define DRM_IOCTL_I915_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP, struct drm_i915_gem_mmap) 524 #define DRM_IOCTL_I915_GEM_MMAP_GTT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_gtt) 525 #define DRM_IOCTL_I915_GEM_MMAP_OFFSET DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_offset) 526 #define DRM_IOCTL_I915_GEM_SET_DOMAIN DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SET_DOMAIN, struct drm_i915_gem_set_domain) 527 #define DRM_IOCTL_I915_GEM_SW_FINISH DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SW_FINISH, struct drm_i915_gem_sw_finish) 528 #define DRM_IOCTL_I915_GEM_SET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_SET_TILING, struct drm_i915_gem_set_tiling) 529 #define DRM_IOCTL_I915_GEM_GET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_TILING, struct drm_i915_gem_get_tiling) 530 #define DRM_IOCTL_I915_GEM_GET_APERTURE DRM_IOR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_APERTURE, struct drm_i915_gem_get_aperture) 531 #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) 532 #define DRM_IOCTL_I915_GEM_MADVISE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MADVISE, struct drm_i915_gem_madvise) 533 #define DRM_IOCTL_I915_OVERLAY_PUT_IMAGE DRM_IOW(DRM_COMMAND_BASE + DRM_I915_OVERLAY_PUT_IMAGE, struct drm_intel_overlay_put_image) 534 #define DRM_IOCTL_I915_OVERLAY_ATTRS DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_OVERLAY_ATTRS, struct drm_intel_overlay_attrs) 535 #define DRM_IOCTL_I915_SET_SPRITE_COLORKEY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_SET_SPRITE_COLORKEY, struct drm_intel_sprite_colorkey) 536 #define DRM_IOCTL_I915_GET_SPRITE_COLORKEY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GET_SPRITE_COLORKEY, struct drm_intel_sprite_colorkey) 537 #define DRM_IOCTL_I915_GEM_WAIT DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_WAIT, struct drm_i915_gem_wait) 538 #define DRM_IOCTL_I915_GEM_CONTEXT_CREATE DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_CREATE, struct drm_i915_gem_context_create) 539 #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) 540 #define DRM_IOCTL_I915_GEM_CONTEXT_DESTROY DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_DESTROY, struct drm_i915_gem_context_destroy) 541 #define DRM_IOCTL_I915_REG_READ DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_REG_READ, struct drm_i915_reg_read) 542 #define DRM_IOCTL_I915_GET_RESET_STATS DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GET_RESET_STATS, struct drm_i915_reset_stats) 543 #define DRM_IOCTL_I915_GEM_USERPTR DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_USERPTR, struct drm_i915_gem_userptr) 544 #define DRM_IOCTL_I915_GEM_CONTEXT_GETPARAM DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_GETPARAM, struct drm_i915_gem_context_param) 545 #define DRM_IOCTL_I915_GEM_CONTEXT_SETPARAM DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_SETPARAM, struct drm_i915_gem_context_param) 546 #define DRM_IOCTL_I915_PERF_OPEN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_OPEN, struct drm_i915_perf_open_param) 547 #define DRM_IOCTL_I915_PERF_ADD_CONFIG DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_ADD_CONFIG, struct drm_i915_perf_oa_config) 548 #define DRM_IOCTL_I915_PERF_REMOVE_CONFIG DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_REMOVE_CONFIG, __u64) 549 #define DRM_IOCTL_I915_QUERY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_QUERY, struct drm_i915_query) 550 #define DRM_IOCTL_I915_GEM_VM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_VM_CREATE, struct drm_i915_gem_vm_control) 551 #define DRM_IOCTL_I915_GEM_VM_DESTROY DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_VM_DESTROY, struct drm_i915_gem_vm_control) 552 553 /* Allow drivers to submit batchbuffers directly to hardware, relying 554 * on the security mechanisms provided by hardware. 555 */ 556 typedef struct drm_i915_batchbuffer { 557 int start; /* agp offset */ 558 int used; /* nr bytes in use */ 559 int DR1; /* hw flags for GFX_OP_DRAWRECT_INFO */ 560 int DR4; /* window origin for GFX_OP_DRAWRECT_INFO */ 561 int num_cliprects; /* mulitpass with multiple cliprects? */ 562 struct drm_clip_rect __user *cliprects; /* pointer to userspace cliprects */ 563 } drm_i915_batchbuffer_t; 564 565 /* As above, but pass a pointer to userspace buffer which can be 566 * validated by the kernel prior to sending to hardware. 567 */ 568 typedef struct _drm_i915_cmdbuffer { 569 char __user *buf; /* pointer to userspace command buffer */ 570 int sz; /* nr bytes in buf */ 571 int DR1; /* hw flags for GFX_OP_DRAWRECT_INFO */ 572 int DR4; /* window origin for GFX_OP_DRAWRECT_INFO */ 573 int num_cliprects; /* mulitpass with multiple cliprects? */ 574 struct drm_clip_rect __user *cliprects; /* pointer to userspace cliprects */ 575 } drm_i915_cmdbuffer_t; 576 577 /* Userspace can request & wait on irq's: 578 */ 579 typedef struct drm_i915_irq_emit { 580 int __user *irq_seq; 581 } drm_i915_irq_emit_t; 582 583 typedef struct drm_i915_irq_wait { 584 int irq_seq; 585 } drm_i915_irq_wait_t; 586 587 /* 588 * Different modes of per-process Graphics Translation Table, 589 * see I915_PARAM_HAS_ALIASING_PPGTT 590 */ 591 #define I915_GEM_PPGTT_NONE 0 592 #define I915_GEM_PPGTT_ALIASING 1 593 #define I915_GEM_PPGTT_FULL 2 594 595 /* Ioctl to query kernel params: 596 */ 597 #define I915_PARAM_IRQ_ACTIVE 1 598 #define I915_PARAM_ALLOW_BATCHBUFFER 2 599 #define I915_PARAM_LAST_DISPATCH 3 600 #define I915_PARAM_CHIPSET_ID 4 601 #define I915_PARAM_HAS_GEM 5 602 #define I915_PARAM_NUM_FENCES_AVAIL 6 603 #define I915_PARAM_HAS_OVERLAY 7 604 #define I915_PARAM_HAS_PAGEFLIPPING 8 605 #define I915_PARAM_HAS_EXECBUF2 9 606 #define I915_PARAM_HAS_BSD 10 607 #define I915_PARAM_HAS_BLT 11 608 #define I915_PARAM_HAS_RELAXED_FENCING 12 609 #define I915_PARAM_HAS_COHERENT_RINGS 13 610 #define I915_PARAM_HAS_EXEC_CONSTANTS 14 611 #define I915_PARAM_HAS_RELAXED_DELTA 15 612 #define I915_PARAM_HAS_GEN7_SOL_RESET 16 613 #define I915_PARAM_HAS_LLC 17 614 #define I915_PARAM_HAS_ALIASING_PPGTT 18 615 #define I915_PARAM_HAS_WAIT_TIMEOUT 19 616 #define I915_PARAM_HAS_SEMAPHORES 20 617 #define I915_PARAM_HAS_PRIME_VMAP_FLUSH 21 618 #define I915_PARAM_HAS_VEBOX 22 619 #define I915_PARAM_HAS_SECURE_BATCHES 23 620 #define I915_PARAM_HAS_PINNED_BATCHES 24 621 #define I915_PARAM_HAS_EXEC_NO_RELOC 25 622 #define I915_PARAM_HAS_EXEC_HANDLE_LUT 26 623 #define I915_PARAM_HAS_WT 27 624 #define I915_PARAM_CMD_PARSER_VERSION 28 625 #define I915_PARAM_HAS_COHERENT_PHYS_GTT 29 626 #define I915_PARAM_MMAP_VERSION 30 627 #define I915_PARAM_HAS_BSD2 31 628 #define I915_PARAM_REVISION 32 629 #define I915_PARAM_SUBSLICE_TOTAL 33 630 #define I915_PARAM_EU_TOTAL 34 631 #define I915_PARAM_HAS_GPU_RESET 35 632 #define I915_PARAM_HAS_RESOURCE_STREAMER 36 633 #define I915_PARAM_HAS_EXEC_SOFTPIN 37 634 #define I915_PARAM_HAS_POOLED_EU 38 635 #define I915_PARAM_MIN_EU_IN_POOL 39 636 #define I915_PARAM_MMAP_GTT_VERSION 40 637 638 /* 639 * Query whether DRM_I915_GEM_EXECBUFFER2 supports user defined execution 640 * priorities and the driver will attempt to execute batches in priority order. 641 * The param returns a capability bitmask, nonzero implies that the scheduler 642 * is enabled, with different features present according to the mask. 643 * 644 * The initial priority for each batch is supplied by the context and is 645 * controlled via I915_CONTEXT_PARAM_PRIORITY. 646 */ 647 #define I915_PARAM_HAS_SCHEDULER 41 648 #define I915_SCHEDULER_CAP_ENABLED (1ul << 0) 649 #define I915_SCHEDULER_CAP_PRIORITY (1ul << 1) 650 #define I915_SCHEDULER_CAP_PREEMPTION (1ul << 2) 651 #define I915_SCHEDULER_CAP_SEMAPHORES (1ul << 3) 652 #define I915_SCHEDULER_CAP_ENGINE_BUSY_STATS (1ul << 4) 653 /* 654 * Indicates the 2k user priority levels are statically mapped into 3 buckets as 655 * follows: 656 * 657 * -1k to -1 Low priority 658 * 0 Normal priority 659 * 1 to 1k Highest priority 660 */ 661 #define I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP (1ul << 5) 662 663 /* 664 * Query the status of HuC load. 665 * 666 * The query can fail in the following scenarios with the listed error codes: 667 * -ENODEV if HuC is not present on this platform, 668 * -EOPNOTSUPP if HuC firmware usage is disabled, 669 * -ENOPKG if HuC firmware fetch failed, 670 * -ENOEXEC if HuC firmware is invalid or mismatched, 671 * -ENOMEM if i915 failed to prepare the FW objects for transfer to the uC, 672 * -EIO if the FW transfer or the FW authentication failed. 673 * 674 * If the IOCTL is successful, the returned parameter will be set to one of the 675 * following values: 676 * * 0 if HuC firmware load is not complete, 677 * * 1 if HuC firmware is authenticated and running. 678 */ 679 #define I915_PARAM_HUC_STATUS 42 680 681 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to opt-out of 682 * synchronisation with implicit fencing on individual objects. 683 * See EXEC_OBJECT_ASYNC. 684 */ 685 #define I915_PARAM_HAS_EXEC_ASYNC 43 686 687 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports explicit fence support - 688 * both being able to pass in a sync_file fd to wait upon before executing, 689 * and being able to return a new sync_file fd that is signaled when the 690 * current request is complete. See I915_EXEC_FENCE_IN and I915_EXEC_FENCE_OUT. 691 */ 692 #define I915_PARAM_HAS_EXEC_FENCE 44 693 694 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to capture 695 * user specified bufffers for post-mortem debugging of GPU hangs. See 696 * EXEC_OBJECT_CAPTURE. 697 */ 698 #define I915_PARAM_HAS_EXEC_CAPTURE 45 699 700 #define I915_PARAM_SLICE_MASK 46 701 702 /* Assuming it's uniform for each slice, this queries the mask of subslices 703 * per-slice for this system. 704 */ 705 #define I915_PARAM_SUBSLICE_MASK 47 706 707 /* 708 * Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying the batch buffer 709 * as the first execobject as opposed to the last. See I915_EXEC_BATCH_FIRST. 710 */ 711 #define I915_PARAM_HAS_EXEC_BATCH_FIRST 48 712 713 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 714 * drm_i915_gem_exec_fence structures. See I915_EXEC_FENCE_ARRAY. 715 */ 716 #define I915_PARAM_HAS_EXEC_FENCE_ARRAY 49 717 718 /* 719 * Query whether every context (both per-file default and user created) is 720 * isolated (insofar as HW supports). If this parameter is not true, then 721 * freshly created contexts may inherit values from an existing context, 722 * rather than default HW values. If true, it also ensures (insofar as HW 723 * supports) that all state set by this context will not leak to any other 724 * context. 725 * 726 * As not every engine across every gen support contexts, the returned 727 * value reports the support of context isolation for individual engines by 728 * returning a bitmask of each engine class set to true if that class supports 729 * isolation. 730 */ 731 #define I915_PARAM_HAS_CONTEXT_ISOLATION 50 732 733 /* Frequency of the command streamer timestamps given by the *_TIMESTAMP 734 * registers. This used to be fixed per platform but from CNL onwards, this 735 * might vary depending on the parts. 736 */ 737 #define I915_PARAM_CS_TIMESTAMP_FREQUENCY 51 738 739 /* 740 * Once upon a time we supposed that writes through the GGTT would be 741 * immediately in physical memory (once flushed out of the CPU path). However, 742 * on a few different processors and chipsets, this is not necessarily the case 743 * as the writes appear to be buffered internally. Thus a read of the backing 744 * storage (physical memory) via a different path (with different physical tags 745 * to the indirect write via the GGTT) will see stale values from before 746 * the GGTT write. Inside the kernel, we can for the most part keep track of 747 * the different read/write domains in use (e.g. set-domain), but the assumption 748 * of coherency is baked into the ABI, hence reporting its true state in this 749 * parameter. 750 * 751 * Reports true when writes via mmap_gtt are immediately visible following an 752 * lfence to flush the WCB. 753 * 754 * Reports false when writes via mmap_gtt are indeterminately delayed in an in 755 * internal buffer and are _not_ immediately visible to third parties accessing 756 * directly via mmap_cpu/mmap_wc. Use of mmap_gtt as part of an IPC 757 * communications channel when reporting false is strongly disadvised. 758 */ 759 #define I915_PARAM_MMAP_GTT_COHERENT 52 760 761 /* 762 * Query whether DRM_I915_GEM_EXECBUFFER2 supports coordination of parallel 763 * execution through use of explicit fence support. 764 * See I915_EXEC_FENCE_OUT and I915_EXEC_FENCE_SUBMIT. 765 */ 766 #define I915_PARAM_HAS_EXEC_SUBMIT_FENCE 53 767 768 /* 769 * Revision of the i915-perf uAPI. The value returned helps determine what 770 * i915-perf features are available. See drm_i915_perf_property_id. 771 */ 772 #define I915_PARAM_PERF_REVISION 54 773 774 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 775 * timeline syncobj through drm_i915_gem_execbuffer_ext_timeline_fences. See 776 * I915_EXEC_USE_EXTENSIONS. 777 */ 778 #define I915_PARAM_HAS_EXEC_TIMELINE_FENCES 55 779 780 /* Query if the kernel supports the I915_USERPTR_PROBE flag. */ 781 #define I915_PARAM_HAS_USERPTR_PROBE 56 782 783 /* 784 * Frequency of the timestamps in OA reports. This used to be the same as the CS 785 * timestamp frequency, but differs on some platforms. 786 */ 787 #define I915_PARAM_OA_TIMESTAMP_FREQUENCY 57 788 789 /* 790 * Query the status of PXP support in i915. 791 * 792 * The query can fail in the following scenarios with the listed error codes: 793 * -ENODEV = PXP support is not available on the GPU device or in the 794 * kernel due to missing component drivers or kernel configs. 795 * 796 * If the IOCTL is successful, the returned parameter will be set to one of 797 * the following values: 798 * 1 = PXP feature is supported and is ready for use. 799 * 2 = PXP feature is supported but should be ready soon (pending 800 * initialization of non-i915 system dependencies). 801 * 802 * NOTE: When param is supported (positive return values), user space should 803 * still refer to the GEM PXP context-creation UAPI header specs to be 804 * aware of possible failure due to system state machine at the time. 805 */ 806 #define I915_PARAM_PXP_STATUS 58 807 808 /* Must be kept compact -- no holes and well documented */ 809 810 /** 811 * struct drm_i915_getparam - Driver parameter query structure. 812 */ 813 struct drm_i915_getparam { 814 /** @param: Driver parameter to query. */ 815 __s32 param; 816 817 /** 818 * @value: Address of memory where queried value should be put. 819 * 820 * WARNING: Using pointers instead of fixed-size u64 means we need to write 821 * compat32 code. Don't repeat this mistake. 822 */ 823 int __user *value; 824 }; 825 826 /** 827 * typedef drm_i915_getparam_t - Driver parameter query structure. 828 * See struct drm_i915_getparam. 829 */ 830 typedef struct drm_i915_getparam drm_i915_getparam_t; 831 832 /* Ioctl to set kernel params: 833 */ 834 #define I915_SETPARAM_USE_MI_BATCHBUFFER_START 1 835 #define I915_SETPARAM_TEX_LRU_LOG_GRANULARITY 2 836 #define I915_SETPARAM_ALLOW_BATCHBUFFER 3 837 #define I915_SETPARAM_NUM_USED_FENCES 4 838 /* Must be kept compact -- no holes */ 839 840 typedef struct drm_i915_setparam { 841 int param; 842 int value; 843 } drm_i915_setparam_t; 844 845 /* A memory manager for regions of shared memory: 846 */ 847 #define I915_MEM_REGION_AGP 1 848 849 typedef struct drm_i915_mem_alloc { 850 int region; 851 int alignment; 852 int size; 853 int __user *region_offset; /* offset from start of fb or agp */ 854 } drm_i915_mem_alloc_t; 855 856 typedef struct drm_i915_mem_free { 857 int region; 858 int region_offset; 859 } drm_i915_mem_free_t; 860 861 typedef struct drm_i915_mem_init_heap { 862 int region; 863 int size; 864 int start; 865 } drm_i915_mem_init_heap_t; 866 867 /* Allow memory manager to be torn down and re-initialized (eg on 868 * rotate): 869 */ 870 typedef struct drm_i915_mem_destroy_heap { 871 int region; 872 } drm_i915_mem_destroy_heap_t; 873 874 /* Allow X server to configure which pipes to monitor for vblank signals 875 */ 876 #define DRM_I915_VBLANK_PIPE_A 1 877 #define DRM_I915_VBLANK_PIPE_B 2 878 879 typedef struct drm_i915_vblank_pipe { 880 int pipe; 881 } drm_i915_vblank_pipe_t; 882 883 /* Schedule buffer swap at given vertical blank: 884 */ 885 typedef struct drm_i915_vblank_swap { 886 drm_drawable_t drawable; 887 enum drm_vblank_seq_type seqtype; 888 unsigned int sequence; 889 } drm_i915_vblank_swap_t; 890 891 typedef struct drm_i915_hws_addr { 892 __u64 addr; 893 } drm_i915_hws_addr_t; 894 895 struct drm_i915_gem_init { 896 /** 897 * Beginning offset in the GTT to be managed by the DRM memory 898 * manager. 899 */ 900 __u64 gtt_start; 901 /** 902 * Ending offset in the GTT to be managed by the DRM memory 903 * manager. 904 */ 905 __u64 gtt_end; 906 }; 907 908 struct drm_i915_gem_create { 909 /** 910 * Requested size for the object. 911 * 912 * The (page-aligned) allocated size for the object will be returned. 913 */ 914 __u64 size; 915 /** 916 * Returned handle for the object. 917 * 918 * Object handles are nonzero. 919 */ 920 __u32 handle; 921 __u32 pad; 922 }; 923 924 struct drm_i915_gem_pread { 925 /** Handle for the object being read. */ 926 __u32 handle; 927 __u32 pad; 928 /** Offset into the object to read from */ 929 __u64 offset; 930 /** Length of data to read */ 931 __u64 size; 932 /** 933 * Pointer to write the data into. 934 * 935 * This is a fixed-size type for 32/64 compatibility. 936 */ 937 __u64 data_ptr; 938 }; 939 940 struct drm_i915_gem_pwrite { 941 /** Handle for the object being written to. */ 942 __u32 handle; 943 __u32 pad; 944 /** Offset into the object to write to */ 945 __u64 offset; 946 /** Length of data to write */ 947 __u64 size; 948 /** 949 * Pointer to read the data from. 950 * 951 * This is a fixed-size type for 32/64 compatibility. 952 */ 953 __u64 data_ptr; 954 }; 955 956 struct drm_i915_gem_mmap { 957 /** Handle for the object being mapped. */ 958 __u32 handle; 959 __u32 pad; 960 /** Offset in the object to map. */ 961 __u64 offset; 962 /** 963 * Length of data to map. 964 * 965 * The value will be page-aligned. 966 */ 967 __u64 size; 968 /** 969 * Returned pointer the data was mapped at. 970 * 971 * This is a fixed-size type for 32/64 compatibility. 972 */ 973 __u64 addr_ptr; 974 975 /** 976 * Flags for extended behaviour. 977 * 978 * Added in version 2. 979 */ 980 __u64 flags; 981 #define I915_MMAP_WC 0x1 982 }; 983 984 struct drm_i915_gem_mmap_gtt { 985 /** Handle for the object being mapped. */ 986 __u32 handle; 987 __u32 pad; 988 /** 989 * Fake offset to use for subsequent mmap call 990 * 991 * This is a fixed-size type for 32/64 compatibility. 992 */ 993 __u64 offset; 994 }; 995 996 /** 997 * struct drm_i915_gem_mmap_offset - Retrieve an offset so we can mmap this buffer object. 998 * 999 * This struct is passed as argument to the `DRM_IOCTL_I915_GEM_MMAP_OFFSET` ioctl, 1000 * and is used to retrieve the fake offset to mmap an object specified by &handle. 1001 * 1002 * The legacy way of using `DRM_IOCTL_I915_GEM_MMAP` is removed on gen12+. 1003 * `DRM_IOCTL_I915_GEM_MMAP_GTT` is an older supported alias to this struct, but will behave 1004 * as setting the &extensions to 0, and &flags to `I915_MMAP_OFFSET_GTT`. 1005 */ 1006 struct drm_i915_gem_mmap_offset { 1007 /** @handle: Handle for the object being mapped. */ 1008 __u32 handle; 1009 /** @pad: Must be zero */ 1010 __u32 pad; 1011 /** 1012 * @offset: The fake offset to use for subsequent mmap call 1013 * 1014 * This is a fixed-size type for 32/64 compatibility. 1015 */ 1016 __u64 offset; 1017 1018 /** 1019 * @flags: Flags for extended behaviour. 1020 * 1021 * It is mandatory that one of the `MMAP_OFFSET` types 1022 * should be included: 1023 * 1024 * - `I915_MMAP_OFFSET_GTT`: Use mmap with the object bound to GTT. (Write-Combined) 1025 * - `I915_MMAP_OFFSET_WC`: Use Write-Combined caching. 1026 * - `I915_MMAP_OFFSET_WB`: Use Write-Back caching. 1027 * - `I915_MMAP_OFFSET_FIXED`: Use object placement to determine caching. 1028 * 1029 * On devices with local memory `I915_MMAP_OFFSET_FIXED` is the only valid 1030 * type. On devices without local memory, this caching mode is invalid. 1031 * 1032 * As caching mode when specifying `I915_MMAP_OFFSET_FIXED`, WC or WB will 1033 * be used, depending on the object placement on creation. WB will be used 1034 * when the object can only exist in system memory, WC otherwise. 1035 */ 1036 __u64 flags; 1037 1038 #define I915_MMAP_OFFSET_GTT 0 1039 #define I915_MMAP_OFFSET_WC 1 1040 #define I915_MMAP_OFFSET_WB 2 1041 #define I915_MMAP_OFFSET_UC 3 1042 #define I915_MMAP_OFFSET_FIXED 4 1043 1044 /** 1045 * @extensions: Zero-terminated chain of extensions. 1046 * 1047 * No current extensions defined; mbz. 1048 */ 1049 __u64 extensions; 1050 }; 1051 1052 /** 1053 * struct drm_i915_gem_set_domain - Adjust the objects write or read domain, in 1054 * preparation for accessing the pages via some CPU domain. 1055 * 1056 * Specifying a new write or read domain will flush the object out of the 1057 * previous domain(if required), before then updating the objects domain 1058 * tracking with the new domain. 1059 * 1060 * Note this might involve waiting for the object first if it is still active on 1061 * the GPU. 1062 * 1063 * Supported values for @read_domains and @write_domain: 1064 * 1065 * - I915_GEM_DOMAIN_WC: Uncached write-combined domain 1066 * - I915_GEM_DOMAIN_CPU: CPU cache domain 1067 * - I915_GEM_DOMAIN_GTT: Mappable aperture domain 1068 * 1069 * All other domains are rejected. 1070 * 1071 * Note that for discrete, starting from DG1, this is no longer supported, and 1072 * is instead rejected. On such platforms the CPU domain is effectively static, 1073 * where we also only support a single &drm_i915_gem_mmap_offset cache mode, 1074 * which can't be set explicitly and instead depends on the object placements, 1075 * as per the below. 1076 * 1077 * Implicit caching rules, starting from DG1: 1078 * 1079 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1080 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1081 * mapped as write-combined only. 1082 * 1083 * - Everything else is always allocated and mapped as write-back, with the 1084 * guarantee that everything is also coherent with the GPU. 1085 * 1086 * Note that this is likely to change in the future again, where we might need 1087 * more flexibility on future devices, so making this all explicit as part of a 1088 * new &drm_i915_gem_create_ext extension is probable. 1089 */ 1090 struct drm_i915_gem_set_domain { 1091 /** @handle: Handle for the object. */ 1092 __u32 handle; 1093 1094 /** @read_domains: New read domains. */ 1095 __u32 read_domains; 1096 1097 /** 1098 * @write_domain: New write domain. 1099 * 1100 * Note that having something in the write domain implies it's in the 1101 * read domain, and only that read domain. 1102 */ 1103 __u32 write_domain; 1104 }; 1105 1106 struct drm_i915_gem_sw_finish { 1107 /** Handle for the object */ 1108 __u32 handle; 1109 }; 1110 1111 struct drm_i915_gem_relocation_entry { 1112 /** 1113 * Handle of the buffer being pointed to by this relocation entry. 1114 * 1115 * It's appealing to make this be an index into the mm_validate_entry 1116 * list to refer to the buffer, but this allows the driver to create 1117 * a relocation list for state buffers and not re-write it per 1118 * exec using the buffer. 1119 */ 1120 __u32 target_handle; 1121 1122 /** 1123 * Value to be added to the offset of the target buffer to make up 1124 * the relocation entry. 1125 */ 1126 __u32 delta; 1127 1128 /** Offset in the buffer the relocation entry will be written into */ 1129 __u64 offset; 1130 1131 /** 1132 * Offset value of the target buffer that the relocation entry was last 1133 * written as. 1134 * 1135 * If the buffer has the same offset as last time, we can skip syncing 1136 * and writing the relocation. This value is written back out by 1137 * the execbuffer ioctl when the relocation is written. 1138 */ 1139 __u64 presumed_offset; 1140 1141 /** 1142 * Target memory domains read by this operation. 1143 */ 1144 __u32 read_domains; 1145 1146 /** 1147 * Target memory domains written by this operation. 1148 * 1149 * Note that only one domain may be written by the whole 1150 * execbuffer operation, so that where there are conflicts, 1151 * the application will get -EINVAL back. 1152 */ 1153 __u32 write_domain; 1154 }; 1155 1156 /** @{ 1157 * Intel memory domains 1158 * 1159 * Most of these just align with the various caches in 1160 * the system and are used to flush and invalidate as 1161 * objects end up cached in different domains. 1162 */ 1163 /** CPU cache */ 1164 #define I915_GEM_DOMAIN_CPU 0x00000001 1165 /** Render cache, used by 2D and 3D drawing */ 1166 #define I915_GEM_DOMAIN_RENDER 0x00000002 1167 /** Sampler cache, used by texture engine */ 1168 #define I915_GEM_DOMAIN_SAMPLER 0x00000004 1169 /** Command queue, used to load batch buffers */ 1170 #define I915_GEM_DOMAIN_COMMAND 0x00000008 1171 /** Instruction cache, used by shader programs */ 1172 #define I915_GEM_DOMAIN_INSTRUCTION 0x00000010 1173 /** Vertex address cache */ 1174 #define I915_GEM_DOMAIN_VERTEX 0x00000020 1175 /** GTT domain - aperture and scanout */ 1176 #define I915_GEM_DOMAIN_GTT 0x00000040 1177 /** WC domain - uncached access */ 1178 #define I915_GEM_DOMAIN_WC 0x00000080 1179 /** @} */ 1180 1181 struct drm_i915_gem_exec_object { 1182 /** 1183 * User's handle for a buffer to be bound into the GTT for this 1184 * operation. 1185 */ 1186 __u32 handle; 1187 1188 /** Number of relocations to be performed on this buffer */ 1189 __u32 relocation_count; 1190 /** 1191 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1192 * the relocations to be performed in this buffer. 1193 */ 1194 __u64 relocs_ptr; 1195 1196 /** Required alignment in graphics aperture */ 1197 __u64 alignment; 1198 1199 /** 1200 * Returned value of the updated offset of the object, for future 1201 * presumed_offset writes. 1202 */ 1203 __u64 offset; 1204 }; 1205 1206 /* DRM_IOCTL_I915_GEM_EXECBUFFER was removed in Linux 5.13 */ 1207 struct drm_i915_gem_execbuffer { 1208 /** 1209 * List of buffers to be validated with their relocations to be 1210 * performend on them. 1211 * 1212 * This is a pointer to an array of struct drm_i915_gem_validate_entry. 1213 * 1214 * These buffers must be listed in an order such that all relocations 1215 * a buffer is performing refer to buffers that have already appeared 1216 * in the validate list. 1217 */ 1218 __u64 buffers_ptr; 1219 __u32 buffer_count; 1220 1221 /** Offset in the batchbuffer to start execution from. */ 1222 __u32 batch_start_offset; 1223 /** Bytes used in batchbuffer from batch_start_offset */ 1224 __u32 batch_len; 1225 __u32 DR1; 1226 __u32 DR4; 1227 __u32 num_cliprects; 1228 /** This is a struct drm_clip_rect *cliprects */ 1229 __u64 cliprects_ptr; 1230 }; 1231 1232 struct drm_i915_gem_exec_object2 { 1233 /** 1234 * User's handle for a buffer to be bound into the GTT for this 1235 * operation. 1236 */ 1237 __u32 handle; 1238 1239 /** Number of relocations to be performed on this buffer */ 1240 __u32 relocation_count; 1241 /** 1242 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1243 * the relocations to be performed in this buffer. 1244 */ 1245 __u64 relocs_ptr; 1246 1247 /** Required alignment in graphics aperture */ 1248 __u64 alignment; 1249 1250 /** 1251 * When the EXEC_OBJECT_PINNED flag is specified this is populated by 1252 * the user with the GTT offset at which this object will be pinned. 1253 * 1254 * When the I915_EXEC_NO_RELOC flag is specified this must contain the 1255 * presumed_offset of the object. 1256 * 1257 * During execbuffer2 the kernel populates it with the value of the 1258 * current GTT offset of the object, for future presumed_offset writes. 1259 * 1260 * See struct drm_i915_gem_create_ext for the rules when dealing with 1261 * alignment restrictions with I915_MEMORY_CLASS_DEVICE, on devices with 1262 * minimum page sizes, like DG2. 1263 */ 1264 __u64 offset; 1265 1266 #define EXEC_OBJECT_NEEDS_FENCE (1<<0) 1267 #define EXEC_OBJECT_NEEDS_GTT (1<<1) 1268 #define EXEC_OBJECT_WRITE (1<<2) 1269 #define EXEC_OBJECT_SUPPORTS_48B_ADDRESS (1<<3) 1270 #define EXEC_OBJECT_PINNED (1<<4) 1271 #define EXEC_OBJECT_PAD_TO_SIZE (1<<5) 1272 /* The kernel implicitly tracks GPU activity on all GEM objects, and 1273 * synchronises operations with outstanding rendering. This includes 1274 * rendering on other devices if exported via dma-buf. However, sometimes 1275 * this tracking is too coarse and the user knows better. For example, 1276 * if the object is split into non-overlapping ranges shared between different 1277 * clients or engines (i.e. suballocating objects), the implicit tracking 1278 * by kernel assumes that each operation affects the whole object rather 1279 * than an individual range, causing needless synchronisation between clients. 1280 * The kernel will also forgo any CPU cache flushes prior to rendering from 1281 * the object as the client is expected to be also handling such domain 1282 * tracking. 1283 * 1284 * The kernel maintains the implicit tracking in order to manage resources 1285 * used by the GPU - this flag only disables the synchronisation prior to 1286 * rendering with this object in this execbuf. 1287 * 1288 * Opting out of implicit synhronisation requires the user to do its own 1289 * explicit tracking to avoid rendering corruption. See, for example, 1290 * I915_PARAM_HAS_EXEC_FENCE to order execbufs and execute them asynchronously. 1291 */ 1292 #define EXEC_OBJECT_ASYNC (1<<6) 1293 /* Request that the contents of this execobject be copied into the error 1294 * state upon a GPU hang involving this batch for post-mortem debugging. 1295 * These buffers are recorded in no particular order as "user" in 1296 * /sys/class/drm/cardN/error. Query I915_PARAM_HAS_EXEC_CAPTURE to see 1297 * if the kernel supports this flag. 1298 */ 1299 #define EXEC_OBJECT_CAPTURE (1<<7) 1300 /* All remaining bits are MBZ and RESERVED FOR FUTURE USE */ 1301 #define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_CAPTURE<<1) 1302 __u64 flags; 1303 1304 union { 1305 __u64 rsvd1; 1306 __u64 pad_to_size; 1307 }; 1308 __u64 rsvd2; 1309 }; 1310 1311 /** 1312 * struct drm_i915_gem_exec_fence - An input or output fence for the execbuf 1313 * ioctl. 1314 * 1315 * The request will wait for input fence to signal before submission. 1316 * 1317 * The returned output fence will be signaled after the completion of the 1318 * request. 1319 */ 1320 struct drm_i915_gem_exec_fence { 1321 /** @handle: User's handle for a drm_syncobj to wait on or signal. */ 1322 __u32 handle; 1323 1324 /** 1325 * @flags: Supported flags are: 1326 * 1327 * I915_EXEC_FENCE_WAIT: 1328 * Wait for the input fence before request submission. 1329 * 1330 * I915_EXEC_FENCE_SIGNAL: 1331 * Return request completion fence as output 1332 */ 1333 __u32 flags; 1334 #define I915_EXEC_FENCE_WAIT (1<<0) 1335 #define I915_EXEC_FENCE_SIGNAL (1<<1) 1336 #define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1)) 1337 }; 1338 1339 /** 1340 * struct drm_i915_gem_execbuffer_ext_timeline_fences - Timeline fences 1341 * for execbuf ioctl. 1342 * 1343 * This structure describes an array of drm_syncobj and associated points for 1344 * timeline variants of drm_syncobj. It is invalid to append this structure to 1345 * the execbuf if I915_EXEC_FENCE_ARRAY is set. 1346 */ 1347 struct drm_i915_gem_execbuffer_ext_timeline_fences { 1348 #define DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES 0 1349 /** @base: Extension link. See struct i915_user_extension. */ 1350 struct i915_user_extension base; 1351 1352 /** 1353 * @fence_count: Number of elements in the @handles_ptr & @value_ptr 1354 * arrays. 1355 */ 1356 __u64 fence_count; 1357 1358 /** 1359 * @handles_ptr: Pointer to an array of struct drm_i915_gem_exec_fence 1360 * of length @fence_count. 1361 */ 1362 __u64 handles_ptr; 1363 1364 /** 1365 * @values_ptr: Pointer to an array of u64 values of length 1366 * @fence_count. 1367 * Values must be 0 for a binary drm_syncobj. A Value of 0 for a 1368 * timeline drm_syncobj is invalid as it turns a drm_syncobj into a 1369 * binary one. 1370 */ 1371 __u64 values_ptr; 1372 }; 1373 1374 /** 1375 * struct drm_i915_gem_execbuffer2 - Structure for DRM_I915_GEM_EXECBUFFER2 1376 * ioctl. 1377 */ 1378 struct drm_i915_gem_execbuffer2 { 1379 /** @buffers_ptr: Pointer to a list of gem_exec_object2 structs */ 1380 __u64 buffers_ptr; 1381 1382 /** @buffer_count: Number of elements in @buffers_ptr array */ 1383 __u32 buffer_count; 1384 1385 /** 1386 * @batch_start_offset: Offset in the batchbuffer to start execution 1387 * from. 1388 */ 1389 __u32 batch_start_offset; 1390 1391 /** 1392 * @batch_len: Length in bytes of the batch buffer, starting from the 1393 * @batch_start_offset. If 0, length is assumed to be the batch buffer 1394 * object size. 1395 */ 1396 __u32 batch_len; 1397 1398 /** @DR1: deprecated */ 1399 __u32 DR1; 1400 1401 /** @DR4: deprecated */ 1402 __u32 DR4; 1403 1404 /** @num_cliprects: See @cliprects_ptr */ 1405 __u32 num_cliprects; 1406 1407 /** 1408 * @cliprects_ptr: Kernel clipping was a DRI1 misfeature. 1409 * 1410 * It is invalid to use this field if I915_EXEC_FENCE_ARRAY or 1411 * I915_EXEC_USE_EXTENSIONS flags are not set. 1412 * 1413 * If I915_EXEC_FENCE_ARRAY is set, then this is a pointer to an array 1414 * of &drm_i915_gem_exec_fence and @num_cliprects is the length of the 1415 * array. 1416 * 1417 * If I915_EXEC_USE_EXTENSIONS is set, then this is a pointer to a 1418 * single &i915_user_extension and num_cliprects is 0. 1419 */ 1420 __u64 cliprects_ptr; 1421 1422 /** @flags: Execbuf flags */ 1423 __u64 flags; 1424 #define I915_EXEC_RING_MASK (0x3f) 1425 #define I915_EXEC_DEFAULT (0<<0) 1426 #define I915_EXEC_RENDER (1<<0) 1427 #define I915_EXEC_BSD (2<<0) 1428 #define I915_EXEC_BLT (3<<0) 1429 #define I915_EXEC_VEBOX (4<<0) 1430 1431 /* Used for switching the constants addressing mode on gen4+ RENDER ring. 1432 * Gen6+ only supports relative addressing to dynamic state (default) and 1433 * absolute addressing. 1434 * 1435 * These flags are ignored for the BSD and BLT rings. 1436 */ 1437 #define I915_EXEC_CONSTANTS_MASK (3<<6) 1438 #define I915_EXEC_CONSTANTS_REL_GENERAL (0<<6) /* default */ 1439 #define I915_EXEC_CONSTANTS_ABSOLUTE (1<<6) 1440 #define I915_EXEC_CONSTANTS_REL_SURFACE (2<<6) /* gen4/5 only */ 1441 1442 /** Resets the SO write offset registers for transform feedback on gen7. */ 1443 #define I915_EXEC_GEN7_SOL_RESET (1<<8) 1444 1445 /** Request a privileged ("secure") batch buffer. Note only available for 1446 * DRM_ROOT_ONLY | DRM_MASTER processes. 1447 */ 1448 #define I915_EXEC_SECURE (1<<9) 1449 1450 /** Inform the kernel that the batch is and will always be pinned. This 1451 * negates the requirement for a workaround to be performed to avoid 1452 * an incoherent CS (such as can be found on 830/845). If this flag is 1453 * not passed, the kernel will endeavour to make sure the batch is 1454 * coherent with the CS before execution. If this flag is passed, 1455 * userspace assumes the responsibility for ensuring the same. 1456 */ 1457 #define I915_EXEC_IS_PINNED (1<<10) 1458 1459 /** Provide a hint to the kernel that the command stream and auxiliary 1460 * state buffers already holds the correct presumed addresses and so the 1461 * relocation process may be skipped if no buffers need to be moved in 1462 * preparation for the execbuffer. 1463 */ 1464 #define I915_EXEC_NO_RELOC (1<<11) 1465 1466 /** Use the reloc.handle as an index into the exec object array rather 1467 * than as the per-file handle. 1468 */ 1469 #define I915_EXEC_HANDLE_LUT (1<<12) 1470 1471 /** Used for switching BSD rings on the platforms with two BSD rings */ 1472 #define I915_EXEC_BSD_SHIFT (13) 1473 #define I915_EXEC_BSD_MASK (3 << I915_EXEC_BSD_SHIFT) 1474 /* default ping-pong mode */ 1475 #define I915_EXEC_BSD_DEFAULT (0 << I915_EXEC_BSD_SHIFT) 1476 #define I915_EXEC_BSD_RING1 (1 << I915_EXEC_BSD_SHIFT) 1477 #define I915_EXEC_BSD_RING2 (2 << I915_EXEC_BSD_SHIFT) 1478 1479 /** Tell the kernel that the batchbuffer is processed by 1480 * the resource streamer. 1481 */ 1482 #define I915_EXEC_RESOURCE_STREAMER (1<<15) 1483 1484 /* Setting I915_EXEC_FENCE_IN implies that lower_32_bits(rsvd2) represent 1485 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1486 * the batch. 1487 * 1488 * Returns -EINVAL if the sync_file fd cannot be found. 1489 */ 1490 #define I915_EXEC_FENCE_IN (1<<16) 1491 1492 /* Setting I915_EXEC_FENCE_OUT causes the ioctl to return a sync_file fd 1493 * in the upper_32_bits(rsvd2) upon success. Ownership of the fd is given 1494 * to the caller, and it should be close() after use. (The fd is a regular 1495 * file descriptor and will be cleaned up on process termination. It holds 1496 * a reference to the request, but nothing else.) 1497 * 1498 * The sync_file fd can be combined with other sync_file and passed either 1499 * to execbuf using I915_EXEC_FENCE_IN, to atomic KMS ioctls (so that a flip 1500 * will only occur after this request completes), or to other devices. 1501 * 1502 * Using I915_EXEC_FENCE_OUT requires use of 1503 * DRM_IOCTL_I915_GEM_EXECBUFFER2_WR ioctl so that the result is written 1504 * back to userspace. Failure to do so will cause the out-fence to always 1505 * be reported as zero, and the real fence fd to be leaked. 1506 */ 1507 #define I915_EXEC_FENCE_OUT (1<<17) 1508 1509 /* 1510 * Traditionally the execbuf ioctl has only considered the final element in 1511 * the execobject[] to be the executable batch. Often though, the client 1512 * will known the batch object prior to construction and being able to place 1513 * it into the execobject[] array first can simplify the relocation tracking. 1514 * Setting I915_EXEC_BATCH_FIRST tells execbuf to use element 0 of the 1515 * execobject[] as the * batch instead (the default is to use the last 1516 * element). 1517 */ 1518 #define I915_EXEC_BATCH_FIRST (1<<18) 1519 1520 /* Setting I915_FENCE_ARRAY implies that num_cliprects and cliprects_ptr 1521 * define an array of i915_gem_exec_fence structures which specify a set of 1522 * dma fences to wait upon or signal. 1523 */ 1524 #define I915_EXEC_FENCE_ARRAY (1<<19) 1525 1526 /* 1527 * Setting I915_EXEC_FENCE_SUBMIT implies that lower_32_bits(rsvd2) represent 1528 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1529 * the batch. 1530 * 1531 * Returns -EINVAL if the sync_file fd cannot be found. 1532 */ 1533 #define I915_EXEC_FENCE_SUBMIT (1 << 20) 1534 1535 /* 1536 * Setting I915_EXEC_USE_EXTENSIONS implies that 1537 * drm_i915_gem_execbuffer2.cliprects_ptr is treated as a pointer to an linked 1538 * list of i915_user_extension. Each i915_user_extension node is the base of a 1539 * larger structure. The list of supported structures are listed in the 1540 * drm_i915_gem_execbuffer_ext enum. 1541 */ 1542 #define I915_EXEC_USE_EXTENSIONS (1 << 21) 1543 #define __I915_EXEC_UNKNOWN_FLAGS (-(I915_EXEC_USE_EXTENSIONS << 1)) 1544 1545 /** @rsvd1: Context id */ 1546 __u64 rsvd1; 1547 1548 /** 1549 * @rsvd2: in and out sync_file file descriptors. 1550 * 1551 * When I915_EXEC_FENCE_IN or I915_EXEC_FENCE_SUBMIT flag is set, the 1552 * lower 32 bits of this field will have the in sync_file fd (input). 1553 * 1554 * When I915_EXEC_FENCE_OUT flag is set, the upper 32 bits of this 1555 * field will have the out sync_file fd (output). 1556 */ 1557 __u64 rsvd2; 1558 }; 1559 1560 #define I915_EXEC_CONTEXT_ID_MASK (0xffffffff) 1561 #define i915_execbuffer2_set_context_id(eb2, context) \ 1562 (eb2).rsvd1 = context & I915_EXEC_CONTEXT_ID_MASK 1563 #define i915_execbuffer2_get_context_id(eb2) \ 1564 ((eb2).rsvd1 & I915_EXEC_CONTEXT_ID_MASK) 1565 1566 struct drm_i915_gem_pin { 1567 /** Handle of the buffer to be pinned. */ 1568 __u32 handle; 1569 __u32 pad; 1570 1571 /** alignment required within the aperture */ 1572 __u64 alignment; 1573 1574 /** Returned GTT offset of the buffer. */ 1575 __u64 offset; 1576 }; 1577 1578 struct drm_i915_gem_unpin { 1579 /** Handle of the buffer to be unpinned. */ 1580 __u32 handle; 1581 __u32 pad; 1582 }; 1583 1584 struct drm_i915_gem_busy { 1585 /** Handle of the buffer to check for busy */ 1586 __u32 handle; 1587 1588 /** Return busy status 1589 * 1590 * A return of 0 implies that the object is idle (after 1591 * having flushed any pending activity), and a non-zero return that 1592 * the object is still in-flight on the GPU. (The GPU has not yet 1593 * signaled completion for all pending requests that reference the 1594 * object.) An object is guaranteed to become idle eventually (so 1595 * long as no new GPU commands are executed upon it). Due to the 1596 * asynchronous nature of the hardware, an object reported 1597 * as busy may become idle before the ioctl is completed. 1598 * 1599 * Furthermore, if the object is busy, which engine is busy is only 1600 * provided as a guide and only indirectly by reporting its class 1601 * (there may be more than one engine in each class). There are race 1602 * conditions which prevent the report of which engines are busy from 1603 * being always accurate. However, the converse is not true. If the 1604 * object is idle, the result of the ioctl, that all engines are idle, 1605 * is accurate. 1606 * 1607 * The returned dword is split into two fields to indicate both 1608 * the engine classess on which the object is being read, and the 1609 * engine class on which it is currently being written (if any). 1610 * 1611 * The low word (bits 0:15) indicate if the object is being written 1612 * to by any engine (there can only be one, as the GEM implicit 1613 * synchronisation rules force writes to be serialised). Only the 1614 * engine class (offset by 1, I915_ENGINE_CLASS_RENDER is reported as 1615 * 1 not 0 etc) for the last write is reported. 1616 * 1617 * The high word (bits 16:31) are a bitmask of which engines classes 1618 * are currently reading from the object. Multiple engines may be 1619 * reading from the object simultaneously. 1620 * 1621 * The value of each engine class is the same as specified in the 1622 * I915_CONTEXT_PARAM_ENGINES context parameter and via perf, i.e. 1623 * I915_ENGINE_CLASS_RENDER, I915_ENGINE_CLASS_COPY, etc. 1624 * Some hardware may have parallel execution engines, e.g. multiple 1625 * media engines, which are mapped to the same class identifier and so 1626 * are not separately reported for busyness. 1627 * 1628 * Caveat emptor: 1629 * Only the boolean result of this query is reliable; that is whether 1630 * the object is idle or busy. The report of which engines are busy 1631 * should be only used as a heuristic. 1632 */ 1633 __u32 busy; 1634 }; 1635 1636 /** 1637 * struct drm_i915_gem_caching - Set or get the caching for given object 1638 * handle. 1639 * 1640 * Allow userspace to control the GTT caching bits for a given object when the 1641 * object is later mapped through the ppGTT(or GGTT on older platforms lacking 1642 * ppGTT support, or if the object is used for scanout). Note that this might 1643 * require unbinding the object from the GTT first, if its current caching value 1644 * doesn't match. 1645 * 1646 * Note that this all changes on discrete platforms, starting from DG1, the 1647 * set/get caching is no longer supported, and is now rejected. Instead the CPU 1648 * caching attributes(WB vs WC) will become an immutable creation time property 1649 * for the object, along with the GTT caching level. For now we don't expose any 1650 * new uAPI for this, instead on DG1 this is all implicit, although this largely 1651 * shouldn't matter since DG1 is coherent by default(without any way of 1652 * controlling it). 1653 * 1654 * Implicit caching rules, starting from DG1: 1655 * 1656 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1657 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1658 * mapped as write-combined only. 1659 * 1660 * - Everything else is always allocated and mapped as write-back, with the 1661 * guarantee that everything is also coherent with the GPU. 1662 * 1663 * Note that this is likely to change in the future again, where we might need 1664 * more flexibility on future devices, so making this all explicit as part of a 1665 * new &drm_i915_gem_create_ext extension is probable. 1666 * 1667 * Side note: Part of the reason for this is that changing the at-allocation-time CPU 1668 * caching attributes for the pages might be required(and is expensive) if we 1669 * need to then CPU map the pages later with different caching attributes. This 1670 * inconsistent caching behaviour, while supported on x86, is not universally 1671 * supported on other architectures. So for simplicity we opt for setting 1672 * everything at creation time, whilst also making it immutable, on discrete 1673 * platforms. 1674 */ 1675 struct drm_i915_gem_caching { 1676 /** 1677 * @handle: Handle of the buffer to set/get the caching level. 1678 */ 1679 __u32 handle; 1680 1681 /** 1682 * @caching: The GTT caching level to apply or possible return value. 1683 * 1684 * The supported @caching values: 1685 * 1686 * I915_CACHING_NONE: 1687 * 1688 * GPU access is not coherent with CPU caches. Default for machines 1689 * without an LLC. This means manual flushing might be needed, if we 1690 * want GPU access to be coherent. 1691 * 1692 * I915_CACHING_CACHED: 1693 * 1694 * GPU access is coherent with CPU caches and furthermore the data is 1695 * cached in last-level caches shared between CPU cores and the GPU GT. 1696 * 1697 * I915_CACHING_DISPLAY: 1698 * 1699 * Special GPU caching mode which is coherent with the scanout engines. 1700 * Transparently falls back to I915_CACHING_NONE on platforms where no 1701 * special cache mode (like write-through or gfdt flushing) is 1702 * available. The kernel automatically sets this mode when using a 1703 * buffer as a scanout target. Userspace can manually set this mode to 1704 * avoid a costly stall and clflush in the hotpath of drawing the first 1705 * frame. 1706 */ 1707 #define I915_CACHING_NONE 0 1708 #define I915_CACHING_CACHED 1 1709 #define I915_CACHING_DISPLAY 2 1710 __u32 caching; 1711 }; 1712 1713 #define I915_TILING_NONE 0 1714 #define I915_TILING_X 1 1715 #define I915_TILING_Y 2 1716 /* 1717 * Do not add new tiling types here. The I915_TILING_* values are for 1718 * de-tiling fence registers that no longer exist on modern platforms. Although 1719 * the hardware may support new types of tiling in general (e.g., Tile4), we 1720 * do not need to add them to the uapi that is specific to now-defunct ioctls. 1721 */ 1722 #define I915_TILING_LAST I915_TILING_Y 1723 1724 #define I915_BIT_6_SWIZZLE_NONE 0 1725 #define I915_BIT_6_SWIZZLE_9 1 1726 #define I915_BIT_6_SWIZZLE_9_10 2 1727 #define I915_BIT_6_SWIZZLE_9_11 3 1728 #define I915_BIT_6_SWIZZLE_9_10_11 4 1729 /* Not seen by userland */ 1730 #define I915_BIT_6_SWIZZLE_UNKNOWN 5 1731 /* Seen by userland. */ 1732 #define I915_BIT_6_SWIZZLE_9_17 6 1733 #define I915_BIT_6_SWIZZLE_9_10_17 7 1734 1735 struct drm_i915_gem_set_tiling { 1736 /** Handle of the buffer to have its tiling state updated */ 1737 __u32 handle; 1738 1739 /** 1740 * Tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1741 * I915_TILING_Y). 1742 * 1743 * This value is to be set on request, and will be updated by the 1744 * kernel on successful return with the actual chosen tiling layout. 1745 * 1746 * The tiling mode may be demoted to I915_TILING_NONE when the system 1747 * has bit 6 swizzling that can't be managed correctly by GEM. 1748 * 1749 * Buffer contents become undefined when changing tiling_mode. 1750 */ 1751 __u32 tiling_mode; 1752 1753 /** 1754 * Stride in bytes for the object when in I915_TILING_X or 1755 * I915_TILING_Y. 1756 */ 1757 __u32 stride; 1758 1759 /** 1760 * Returned address bit 6 swizzling required for CPU access through 1761 * mmap mapping. 1762 */ 1763 __u32 swizzle_mode; 1764 }; 1765 1766 struct drm_i915_gem_get_tiling { 1767 /** Handle of the buffer to get tiling state for. */ 1768 __u32 handle; 1769 1770 /** 1771 * Current tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1772 * I915_TILING_Y). 1773 */ 1774 __u32 tiling_mode; 1775 1776 /** 1777 * Returned address bit 6 swizzling required for CPU access through 1778 * mmap mapping. 1779 */ 1780 __u32 swizzle_mode; 1781 1782 /** 1783 * Returned address bit 6 swizzling required for CPU access through 1784 * mmap mapping whilst bound. 1785 */ 1786 __u32 phys_swizzle_mode; 1787 }; 1788 1789 struct drm_i915_gem_get_aperture { 1790 /** Total size of the aperture used by i915_gem_execbuffer, in bytes */ 1791 __u64 aper_size; 1792 1793 /** 1794 * Available space in the aperture used by i915_gem_execbuffer, in 1795 * bytes 1796 */ 1797 __u64 aper_available_size; 1798 }; 1799 1800 struct drm_i915_get_pipe_from_crtc_id { 1801 /** ID of CRTC being requested **/ 1802 __u32 crtc_id; 1803 1804 /** pipe of requested CRTC **/ 1805 __u32 pipe; 1806 }; 1807 1808 #define I915_MADV_WILLNEED 0 1809 #define I915_MADV_DONTNEED 1 1810 #define __I915_MADV_PURGED 2 /* internal state */ 1811 1812 struct drm_i915_gem_madvise { 1813 /** Handle of the buffer to change the backing store advice */ 1814 __u32 handle; 1815 1816 /* Advice: either the buffer will be needed again in the near future, 1817 * or wont be and could be discarded under memory pressure. 1818 */ 1819 __u32 madv; 1820 1821 /** Whether the backing store still exists. */ 1822 __u32 retained; 1823 }; 1824 1825 /* flags */ 1826 #define I915_OVERLAY_TYPE_MASK 0xff 1827 #define I915_OVERLAY_YUV_PLANAR 0x01 1828 #define I915_OVERLAY_YUV_PACKED 0x02 1829 #define I915_OVERLAY_RGB 0x03 1830 1831 #define I915_OVERLAY_DEPTH_MASK 0xff00 1832 #define I915_OVERLAY_RGB24 0x1000 1833 #define I915_OVERLAY_RGB16 0x2000 1834 #define I915_OVERLAY_RGB15 0x3000 1835 #define I915_OVERLAY_YUV422 0x0100 1836 #define I915_OVERLAY_YUV411 0x0200 1837 #define I915_OVERLAY_YUV420 0x0300 1838 #define I915_OVERLAY_YUV410 0x0400 1839 1840 #define I915_OVERLAY_SWAP_MASK 0xff0000 1841 #define I915_OVERLAY_NO_SWAP 0x000000 1842 #define I915_OVERLAY_UV_SWAP 0x010000 1843 #define I915_OVERLAY_Y_SWAP 0x020000 1844 #define I915_OVERLAY_Y_AND_UV_SWAP 0x030000 1845 1846 #define I915_OVERLAY_FLAGS_MASK 0xff000000 1847 #define I915_OVERLAY_ENABLE 0x01000000 1848 1849 struct drm_intel_overlay_put_image { 1850 /* various flags and src format description */ 1851 __u32 flags; 1852 /* source picture description */ 1853 __u32 bo_handle; 1854 /* stride values and offsets are in bytes, buffer relative */ 1855 __u16 stride_Y; /* stride for packed formats */ 1856 __u16 stride_UV; 1857 __u32 offset_Y; /* offset for packet formats */ 1858 __u32 offset_U; 1859 __u32 offset_V; 1860 /* in pixels */ 1861 __u16 src_width; 1862 __u16 src_height; 1863 /* to compensate the scaling factors for partially covered surfaces */ 1864 __u16 src_scan_width; 1865 __u16 src_scan_height; 1866 /* output crtc description */ 1867 __u32 crtc_id; 1868 __u16 dst_x; 1869 __u16 dst_y; 1870 __u16 dst_width; 1871 __u16 dst_height; 1872 }; 1873 1874 /* flags */ 1875 #define I915_OVERLAY_UPDATE_ATTRS (1<<0) 1876 #define I915_OVERLAY_UPDATE_GAMMA (1<<1) 1877 #define I915_OVERLAY_DISABLE_DEST_COLORKEY (1<<2) 1878 struct drm_intel_overlay_attrs { 1879 __u32 flags; 1880 __u32 color_key; 1881 __s32 brightness; 1882 __u32 contrast; 1883 __u32 saturation; 1884 __u32 gamma0; 1885 __u32 gamma1; 1886 __u32 gamma2; 1887 __u32 gamma3; 1888 __u32 gamma4; 1889 __u32 gamma5; 1890 }; 1891 1892 /* 1893 * Intel sprite handling 1894 * 1895 * Color keying works with a min/mask/max tuple. Both source and destination 1896 * color keying is allowed. 1897 * 1898 * Source keying: 1899 * Sprite pixels within the min & max values, masked against the color channels 1900 * specified in the mask field, will be transparent. All other pixels will 1901 * be displayed on top of the primary plane. For RGB surfaces, only the min 1902 * and mask fields will be used; ranged compares are not allowed. 1903 * 1904 * Destination keying: 1905 * Primary plane pixels that match the min value, masked against the color 1906 * channels specified in the mask field, will be replaced by corresponding 1907 * pixels from the sprite plane. 1908 * 1909 * Note that source & destination keying are exclusive; only one can be 1910 * active on a given plane. 1911 */ 1912 1913 #define I915_SET_COLORKEY_NONE (1<<0) /* Deprecated. Instead set 1914 * flags==0 to disable colorkeying. 1915 */ 1916 #define I915_SET_COLORKEY_DESTINATION (1<<1) 1917 #define I915_SET_COLORKEY_SOURCE (1<<2) 1918 struct drm_intel_sprite_colorkey { 1919 __u32 plane_id; 1920 __u32 min_value; 1921 __u32 channel_mask; 1922 __u32 max_value; 1923 __u32 flags; 1924 }; 1925 1926 struct drm_i915_gem_wait { 1927 /** Handle of BO we shall wait on */ 1928 __u32 bo_handle; 1929 __u32 flags; 1930 /** Number of nanoseconds to wait, Returns time remaining. */ 1931 __s64 timeout_ns; 1932 }; 1933 1934 struct drm_i915_gem_context_create { 1935 __u32 ctx_id; /* output: id of new context*/ 1936 __u32 pad; 1937 }; 1938 1939 /** 1940 * struct drm_i915_gem_context_create_ext - Structure for creating contexts. 1941 */ 1942 struct drm_i915_gem_context_create_ext { 1943 /** @ctx_id: Id of the created context (output) */ 1944 __u32 ctx_id; 1945 1946 /** 1947 * @flags: Supported flags are: 1948 * 1949 * I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS: 1950 * 1951 * Extensions may be appended to this structure and driver must check 1952 * for those. See @extensions. 1953 * 1954 * I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE 1955 * 1956 * Created context will have single timeline. 1957 */ 1958 __u32 flags; 1959 #define I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS (1u << 0) 1960 #define I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE (1u << 1) 1961 #define I915_CONTEXT_CREATE_FLAGS_UNKNOWN \ 1962 (-(I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE << 1)) 1963 1964 /** 1965 * @extensions: Zero-terminated chain of extensions. 1966 * 1967 * I915_CONTEXT_CREATE_EXT_SETPARAM: 1968 * Context parameter to set or query during context creation. 1969 * See struct drm_i915_gem_context_create_ext_setparam. 1970 * 1971 * I915_CONTEXT_CREATE_EXT_CLONE: 1972 * This extension has been removed. On the off chance someone somewhere 1973 * has attempted to use it, never re-use this extension number. 1974 */ 1975 __u64 extensions; 1976 #define I915_CONTEXT_CREATE_EXT_SETPARAM 0 1977 #define I915_CONTEXT_CREATE_EXT_CLONE 1 1978 }; 1979 1980 /** 1981 * struct drm_i915_gem_context_param - Context parameter to set or query. 1982 */ 1983 struct drm_i915_gem_context_param { 1984 /** @ctx_id: Context id */ 1985 __u32 ctx_id; 1986 1987 /** @size: Size of the parameter @value */ 1988 __u32 size; 1989 1990 /** @param: Parameter to set or query */ 1991 __u64 param; 1992 #define I915_CONTEXT_PARAM_BAN_PERIOD 0x1 1993 /* I915_CONTEXT_PARAM_NO_ZEROMAP has been removed. On the off chance 1994 * someone somewhere has attempted to use it, never re-use this context 1995 * param number. 1996 */ 1997 #define I915_CONTEXT_PARAM_NO_ZEROMAP 0x2 1998 #define I915_CONTEXT_PARAM_GTT_SIZE 0x3 1999 #define I915_CONTEXT_PARAM_NO_ERROR_CAPTURE 0x4 2000 #define I915_CONTEXT_PARAM_BANNABLE 0x5 2001 #define I915_CONTEXT_PARAM_PRIORITY 0x6 2002 #define I915_CONTEXT_MAX_USER_PRIORITY 1023 /* inclusive */ 2003 #define I915_CONTEXT_DEFAULT_PRIORITY 0 2004 #define I915_CONTEXT_MIN_USER_PRIORITY -1023 /* inclusive */ 2005 /* 2006 * When using the following param, value should be a pointer to 2007 * drm_i915_gem_context_param_sseu. 2008 */ 2009 #define I915_CONTEXT_PARAM_SSEU 0x7 2010 2011 /* 2012 * Not all clients may want to attempt automatic recover of a context after 2013 * a hang (for example, some clients may only submit very small incremental 2014 * batches relying on known logical state of previous batches which will never 2015 * recover correctly and each attempt will hang), and so would prefer that 2016 * the context is forever banned instead. 2017 * 2018 * If set to false (0), after a reset, subsequent (and in flight) rendering 2019 * from this context is discarded, and the client will need to create a new 2020 * context to use instead. 2021 * 2022 * If set to true (1), the kernel will automatically attempt to recover the 2023 * context by skipping the hanging batch and executing the next batch starting 2024 * from the default context state (discarding the incomplete logical context 2025 * state lost due to the reset). 2026 * 2027 * On creation, all new contexts are marked as recoverable. 2028 */ 2029 #define I915_CONTEXT_PARAM_RECOVERABLE 0x8 2030 2031 /* 2032 * The id of the associated virtual memory address space (ppGTT) of 2033 * this context. Can be retrieved and passed to another context 2034 * (on the same fd) for both to use the same ppGTT and so share 2035 * address layouts, and avoid reloading the page tables on context 2036 * switches between themselves. 2037 * 2038 * See DRM_I915_GEM_VM_CREATE and DRM_I915_GEM_VM_DESTROY. 2039 */ 2040 #define I915_CONTEXT_PARAM_VM 0x9 2041 2042 /* 2043 * I915_CONTEXT_PARAM_ENGINES: 2044 * 2045 * Bind this context to operate on this subset of available engines. Henceforth, 2046 * the I915_EXEC_RING selector for DRM_IOCTL_I915_GEM_EXECBUFFER2 operates as 2047 * an index into this array of engines; I915_EXEC_DEFAULT selecting engine[0] 2048 * and upwards. Slots 0...N are filled in using the specified (class, instance). 2049 * Use 2050 * engine_class: I915_ENGINE_CLASS_INVALID, 2051 * engine_instance: I915_ENGINE_CLASS_INVALID_NONE 2052 * to specify a gap in the array that can be filled in later, e.g. by a 2053 * virtual engine used for load balancing. 2054 * 2055 * Setting the number of engines bound to the context to 0, by passing a zero 2056 * sized argument, will revert back to default settings. 2057 * 2058 * See struct i915_context_param_engines. 2059 * 2060 * Extensions: 2061 * i915_context_engines_load_balance (I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE) 2062 * i915_context_engines_bond (I915_CONTEXT_ENGINES_EXT_BOND) 2063 * i915_context_engines_parallel_submit (I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT) 2064 */ 2065 #define I915_CONTEXT_PARAM_ENGINES 0xa 2066 2067 /* 2068 * I915_CONTEXT_PARAM_PERSISTENCE: 2069 * 2070 * Allow the context and active rendering to survive the process until 2071 * completion. Persistence allows fire-and-forget clients to queue up a 2072 * bunch of work, hand the output over to a display server and then quit. 2073 * If the context is marked as not persistent, upon closing (either via 2074 * an explicit DRM_I915_GEM_CONTEXT_DESTROY or implicitly from file closure 2075 * or process termination), the context and any outstanding requests will be 2076 * cancelled (and exported fences for cancelled requests marked as -EIO). 2077 * 2078 * By default, new contexts allow persistence. 2079 */ 2080 #define I915_CONTEXT_PARAM_PERSISTENCE 0xb 2081 2082 /* This API has been removed. On the off chance someone somewhere has 2083 * attempted to use it, never re-use this context param number. 2084 */ 2085 #define I915_CONTEXT_PARAM_RINGSIZE 0xc 2086 2087 /* 2088 * I915_CONTEXT_PARAM_PROTECTED_CONTENT: 2089 * 2090 * Mark that the context makes use of protected content, which will result 2091 * in the context being invalidated when the protected content session is. 2092 * Given that the protected content session is killed on suspend, the device 2093 * is kept awake for the lifetime of a protected context, so the user should 2094 * make sure to dispose of them once done. 2095 * This flag can only be set at context creation time and, when set to true, 2096 * must be preceded by an explicit setting of I915_CONTEXT_PARAM_RECOVERABLE 2097 * to false. This flag can't be set to true in conjunction with setting the 2098 * I915_CONTEXT_PARAM_BANNABLE flag to false. Creation example: 2099 * 2100 * .. code-block:: C 2101 * 2102 * struct drm_i915_gem_context_create_ext_setparam p_protected = { 2103 * .base = { 2104 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2105 * }, 2106 * .param = { 2107 * .param = I915_CONTEXT_PARAM_PROTECTED_CONTENT, 2108 * .value = 1, 2109 * } 2110 * }; 2111 * struct drm_i915_gem_context_create_ext_setparam p_norecover = { 2112 * .base = { 2113 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2114 * .next_extension = to_user_pointer(&p_protected), 2115 * }, 2116 * .param = { 2117 * .param = I915_CONTEXT_PARAM_RECOVERABLE, 2118 * .value = 0, 2119 * } 2120 * }; 2121 * struct drm_i915_gem_context_create_ext create = { 2122 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2123 * .extensions = to_user_pointer(&p_norecover); 2124 * }; 2125 * 2126 * ctx_id = gem_context_create_ext(drm_fd, &create); 2127 * 2128 * In addition to the normal failure cases, setting this flag during context 2129 * creation can result in the following errors: 2130 * 2131 * -ENODEV: feature not available 2132 * -EPERM: trying to mark a recoverable or not bannable context as protected 2133 * -ENXIO: A dependency such as a component driver or firmware is not yet 2134 * loaded so user space may need to attempt again. Depending on the 2135 * device, this error may be reported if protected context creation is 2136 * attempted very early after kernel start because the internal timeout 2137 * waiting for such dependencies is not guaranteed to be larger than 2138 * required (numbers differ depending on system and kernel config): 2139 * - ADL/RPL: dependencies may take up to 3 seconds from kernel start 2140 * while context creation internal timeout is 250 milisecs 2141 * - MTL: dependencies may take up to 8 seconds from kernel start 2142 * while context creation internal timeout is 250 milisecs 2143 * NOTE: such dependencies happen once, so a subsequent call to create a 2144 * protected context after a prior successful call will not experience 2145 * such timeouts and will not return -ENXIO (unless the driver is reloaded, 2146 * or, depending on the device, resumes from a suspended state). 2147 * -EIO: The firmware did not succeed in creating the protected context. 2148 */ 2149 #define I915_CONTEXT_PARAM_PROTECTED_CONTENT 0xd 2150 /* Must be kept compact -- no holes and well documented */ 2151 2152 /** @value: Context parameter value to be set or queried */ 2153 __u64 value; 2154 }; 2155 2156 /* 2157 * Context SSEU programming 2158 * 2159 * It may be necessary for either functional or performance reason to configure 2160 * a context to run with a reduced number of SSEU (where SSEU stands for Slice/ 2161 * Sub-slice/EU). 2162 * 2163 * This is done by configuring SSEU configuration using the below 2164 * @struct drm_i915_gem_context_param_sseu for every supported engine which 2165 * userspace intends to use. 2166 * 2167 * Not all GPUs or engines support this functionality in which case an error 2168 * code -ENODEV will be returned. 2169 * 2170 * Also, flexibility of possible SSEU configuration permutations varies between 2171 * GPU generations and software imposed limitations. Requesting such a 2172 * combination will return an error code of -EINVAL. 2173 * 2174 * NOTE: When perf/OA is active the context's SSEU configuration is ignored in 2175 * favour of a single global setting. 2176 */ 2177 struct drm_i915_gem_context_param_sseu { 2178 /* 2179 * Engine class & instance to be configured or queried. 2180 */ 2181 struct i915_engine_class_instance engine; 2182 2183 /* 2184 * Unknown flags must be cleared to zero. 2185 */ 2186 __u32 flags; 2187 #define I915_CONTEXT_SSEU_FLAG_ENGINE_INDEX (1u << 0) 2188 2189 /* 2190 * Mask of slices to enable for the context. Valid values are a subset 2191 * of the bitmask value returned for I915_PARAM_SLICE_MASK. 2192 */ 2193 __u64 slice_mask; 2194 2195 /* 2196 * Mask of subslices to enable for the context. Valid values are a 2197 * subset of the bitmask value return by I915_PARAM_SUBSLICE_MASK. 2198 */ 2199 __u64 subslice_mask; 2200 2201 /* 2202 * Minimum/Maximum number of EUs to enable per subslice for the 2203 * context. min_eus_per_subslice must be inferior or equal to 2204 * max_eus_per_subslice. 2205 */ 2206 __u16 min_eus_per_subslice; 2207 __u16 max_eus_per_subslice; 2208 2209 /* 2210 * Unused for now. Must be cleared to zero. 2211 */ 2212 __u32 rsvd; 2213 }; 2214 2215 /** 2216 * DOC: Virtual Engine uAPI 2217 * 2218 * Virtual engine is a concept where userspace is able to configure a set of 2219 * physical engines, submit a batch buffer, and let the driver execute it on any 2220 * engine from the set as it sees fit. 2221 * 2222 * This is primarily useful on parts which have multiple instances of a same 2223 * class engine, like for example GT3+ Skylake parts with their two VCS engines. 2224 * 2225 * For instance userspace can enumerate all engines of a certain class using the 2226 * previously described `Engine Discovery uAPI`_. After that userspace can 2227 * create a GEM context with a placeholder slot for the virtual engine (using 2228 * `I915_ENGINE_CLASS_INVALID` and `I915_ENGINE_CLASS_INVALID_NONE` for class 2229 * and instance respectively) and finally using the 2230 * `I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE` extension place a virtual engine in 2231 * the same reserved slot. 2232 * 2233 * Example of creating a virtual engine and submitting a batch buffer to it: 2234 * 2235 * .. code-block:: C 2236 * 2237 * I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(virtual, 2) = { 2238 * .base.name = I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE, 2239 * .engine_index = 0, // Place this virtual engine into engine map slot 0 2240 * .num_siblings = 2, 2241 * .engines = { { I915_ENGINE_CLASS_VIDEO, 0 }, 2242 * { I915_ENGINE_CLASS_VIDEO, 1 }, }, 2243 * }; 2244 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 1) = { 2245 * .engines = { { I915_ENGINE_CLASS_INVALID, 2246 * I915_ENGINE_CLASS_INVALID_NONE } }, 2247 * .extensions = to_user_pointer(&virtual), // Chains after load_balance extension 2248 * }; 2249 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2250 * .base = { 2251 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2252 * }, 2253 * .param = { 2254 * .param = I915_CONTEXT_PARAM_ENGINES, 2255 * .value = to_user_pointer(&engines), 2256 * .size = sizeof(engines), 2257 * }, 2258 * }; 2259 * struct drm_i915_gem_context_create_ext create = { 2260 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2261 * .extensions = to_user_pointer(&p_engines); 2262 * }; 2263 * 2264 * ctx_id = gem_context_create_ext(drm_fd, &create); 2265 * 2266 * // Now we have created a GEM context with its engine map containing a 2267 * // single virtual engine. Submissions to this slot can go either to 2268 * // vcs0 or vcs1, depending on the load balancing algorithm used inside 2269 * // the driver. The load balancing is dynamic from one batch buffer to 2270 * // another and transparent to userspace. 2271 * 2272 * ... 2273 * execbuf.rsvd1 = ctx_id; 2274 * execbuf.flags = 0; // Submits to index 0 which is the virtual engine 2275 * gem_execbuf(drm_fd, &execbuf); 2276 */ 2277 2278 /* 2279 * i915_context_engines_load_balance: 2280 * 2281 * Enable load balancing across this set of engines. 2282 * 2283 * Into the I915_EXEC_DEFAULT slot [0], a virtual engine is created that when 2284 * used will proxy the execbuffer request onto one of the set of engines 2285 * in such a way as to distribute the load evenly across the set. 2286 * 2287 * The set of engines must be compatible (e.g. the same HW class) as they 2288 * will share the same logical GPU context and ring. 2289 * 2290 * To intermix rendering with the virtual engine and direct rendering onto 2291 * the backing engines (bypassing the load balancing proxy), the context must 2292 * be defined to use a single timeline for all engines. 2293 */ 2294 struct i915_context_engines_load_balance { 2295 struct i915_user_extension base; 2296 2297 __u16 engine_index; 2298 __u16 num_siblings; 2299 __u32 flags; /* all undefined flags must be zero */ 2300 2301 __u64 mbz64; /* reserved for future use; must be zero */ 2302 2303 struct i915_engine_class_instance engines[]; 2304 } __attribute__((packed)); 2305 2306 #define I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(name__, N__) struct { \ 2307 struct i915_user_extension base; \ 2308 __u16 engine_index; \ 2309 __u16 num_siblings; \ 2310 __u32 flags; \ 2311 __u64 mbz64; \ 2312 struct i915_engine_class_instance engines[N__]; \ 2313 } __attribute__((packed)) name__ 2314 2315 /* 2316 * i915_context_engines_bond: 2317 * 2318 * Constructed bonded pairs for execution within a virtual engine. 2319 * 2320 * All engines are equal, but some are more equal than others. Given 2321 * the distribution of resources in the HW, it may be preferable to run 2322 * a request on a given subset of engines in parallel to a request on a 2323 * specific engine. We enable this selection of engines within a virtual 2324 * engine by specifying bonding pairs, for any given master engine we will 2325 * only execute on one of the corresponding siblings within the virtual engine. 2326 * 2327 * To execute a request in parallel on the master engine and a sibling requires 2328 * coordination with a I915_EXEC_FENCE_SUBMIT. 2329 */ 2330 struct i915_context_engines_bond { 2331 struct i915_user_extension base; 2332 2333 struct i915_engine_class_instance master; 2334 2335 __u16 virtual_index; /* index of virtual engine in ctx->engines[] */ 2336 __u16 num_bonds; 2337 2338 __u64 flags; /* all undefined flags must be zero */ 2339 __u64 mbz64[4]; /* reserved for future use; must be zero */ 2340 2341 struct i915_engine_class_instance engines[]; 2342 } __attribute__((packed)); 2343 2344 #define I915_DEFINE_CONTEXT_ENGINES_BOND(name__, N__) struct { \ 2345 struct i915_user_extension base; \ 2346 struct i915_engine_class_instance master; \ 2347 __u16 virtual_index; \ 2348 __u16 num_bonds; \ 2349 __u64 flags; \ 2350 __u64 mbz64[4]; \ 2351 struct i915_engine_class_instance engines[N__]; \ 2352 } __attribute__((packed)) name__ 2353 2354 /** 2355 * struct i915_context_engines_parallel_submit - Configure engine for 2356 * parallel submission. 2357 * 2358 * Setup a slot in the context engine map to allow multiple BBs to be submitted 2359 * in a single execbuf IOCTL. Those BBs will then be scheduled to run on the GPU 2360 * in parallel. Multiple hardware contexts are created internally in the i915 to 2361 * run these BBs. Once a slot is configured for N BBs only N BBs can be 2362 * submitted in each execbuf IOCTL and this is implicit behavior e.g. The user 2363 * doesn't tell the execbuf IOCTL there are N BBs, the execbuf IOCTL knows how 2364 * many BBs there are based on the slot's configuration. The N BBs are the last 2365 * N buffer objects or first N if I915_EXEC_BATCH_FIRST is set. 2366 * 2367 * The default placement behavior is to create implicit bonds between each 2368 * context if each context maps to more than 1 physical engine (e.g. context is 2369 * a virtual engine). Also we only allow contexts of same engine class and these 2370 * contexts must be in logically contiguous order. Examples of the placement 2371 * behavior are described below. Lastly, the default is to not allow BBs to be 2372 * preempted mid-batch. Rather insert coordinated preemption points on all 2373 * hardware contexts between each set of BBs. Flags could be added in the future 2374 * to change both of these default behaviors. 2375 * 2376 * Returns -EINVAL if hardware context placement configuration is invalid or if 2377 * the placement configuration isn't supported on the platform / submission 2378 * interface. 2379 * Returns -ENODEV if extension isn't supported on the platform / submission 2380 * interface. 2381 * 2382 * .. code-block:: none 2383 * 2384 * Examples syntax: 2385 * CS[X] = generic engine of same class, logical instance X 2386 * INVALID = I915_ENGINE_CLASS_INVALID, I915_ENGINE_CLASS_INVALID_NONE 2387 * 2388 * Example 1 pseudo code: 2389 * set_engines(INVALID) 2390 * set_parallel(engine_index=0, width=2, num_siblings=1, 2391 * engines=CS[0],CS[1]) 2392 * 2393 * Results in the following valid placement: 2394 * CS[0], CS[1] 2395 * 2396 * Example 2 pseudo code: 2397 * set_engines(INVALID) 2398 * set_parallel(engine_index=0, width=2, num_siblings=2, 2399 * engines=CS[0],CS[2],CS[1],CS[3]) 2400 * 2401 * Results in the following valid placements: 2402 * CS[0], CS[1] 2403 * CS[2], CS[3] 2404 * 2405 * This can be thought of as two virtual engines, each containing two 2406 * engines thereby making a 2D array. However, there are bonds tying the 2407 * entries together and placing restrictions on how they can be scheduled. 2408 * Specifically, the scheduler can choose only vertical columns from the 2D 2409 * array. That is, CS[0] is bonded to CS[1] and CS[2] to CS[3]. So if the 2410 * scheduler wants to submit to CS[0], it must also choose CS[1] and vice 2411 * versa. Same for CS[2] requires also using CS[3]. 2412 * VE[0] = CS[0], CS[2] 2413 * VE[1] = CS[1], CS[3] 2414 * 2415 * Example 3 pseudo code: 2416 * set_engines(INVALID) 2417 * set_parallel(engine_index=0, width=2, num_siblings=2, 2418 * engines=CS[0],CS[1],CS[1],CS[3]) 2419 * 2420 * Results in the following valid and invalid placements: 2421 * CS[0], CS[1] 2422 * CS[1], CS[3] - Not logically contiguous, return -EINVAL 2423 */ 2424 struct i915_context_engines_parallel_submit { 2425 /** 2426 * @base: base user extension. 2427 */ 2428 struct i915_user_extension base; 2429 2430 /** 2431 * @engine_index: slot for parallel engine 2432 */ 2433 __u16 engine_index; 2434 2435 /** 2436 * @width: number of contexts per parallel engine or in other words the 2437 * number of batches in each submission 2438 */ 2439 __u16 width; 2440 2441 /** 2442 * @num_siblings: number of siblings per context or in other words the 2443 * number of possible placements for each submission 2444 */ 2445 __u16 num_siblings; 2446 2447 /** 2448 * @mbz16: reserved for future use; must be zero 2449 */ 2450 __u16 mbz16; 2451 2452 /** 2453 * @flags: all undefined flags must be zero, currently not defined flags 2454 */ 2455 __u64 flags; 2456 2457 /** 2458 * @mbz64: reserved for future use; must be zero 2459 */ 2460 __u64 mbz64[3]; 2461 2462 /** 2463 * @engines: 2-d array of engine instances to configure parallel engine 2464 * 2465 * length = width (i) * num_siblings (j) 2466 * index = j + i * num_siblings 2467 */ 2468 struct i915_engine_class_instance engines[]; 2469 2470 } __packed; 2471 2472 #define I915_DEFINE_CONTEXT_ENGINES_PARALLEL_SUBMIT(name__, N__) struct { \ 2473 struct i915_user_extension base; \ 2474 __u16 engine_index; \ 2475 __u16 width; \ 2476 __u16 num_siblings; \ 2477 __u16 mbz16; \ 2478 __u64 flags; \ 2479 __u64 mbz64[3]; \ 2480 struct i915_engine_class_instance engines[N__]; \ 2481 } __attribute__((packed)) name__ 2482 2483 /** 2484 * DOC: Context Engine Map uAPI 2485 * 2486 * Context engine map is a new way of addressing engines when submitting batch- 2487 * buffers, replacing the existing way of using identifiers like `I915_EXEC_BLT` 2488 * inside the flags field of `struct drm_i915_gem_execbuffer2`. 2489 * 2490 * To use it created GEM contexts need to be configured with a list of engines 2491 * the user is intending to submit to. This is accomplished using the 2492 * `I915_CONTEXT_PARAM_ENGINES` parameter and `struct 2493 * i915_context_param_engines`. 2494 * 2495 * For such contexts the `I915_EXEC_RING_MASK` field becomes an index into the 2496 * configured map. 2497 * 2498 * Example of creating such context and submitting against it: 2499 * 2500 * .. code-block:: C 2501 * 2502 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 2) = { 2503 * .engines = { { I915_ENGINE_CLASS_RENDER, 0 }, 2504 * { I915_ENGINE_CLASS_COPY, 0 } } 2505 * }; 2506 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2507 * .base = { 2508 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2509 * }, 2510 * .param = { 2511 * .param = I915_CONTEXT_PARAM_ENGINES, 2512 * .value = to_user_pointer(&engines), 2513 * .size = sizeof(engines), 2514 * }, 2515 * }; 2516 * struct drm_i915_gem_context_create_ext create = { 2517 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2518 * .extensions = to_user_pointer(&p_engines); 2519 * }; 2520 * 2521 * ctx_id = gem_context_create_ext(drm_fd, &create); 2522 * 2523 * // We have now created a GEM context with two engines in the map: 2524 * // Index 0 points to rcs0 while index 1 points to bcs0. Other engines 2525 * // will not be accessible from this context. 2526 * 2527 * ... 2528 * execbuf.rsvd1 = ctx_id; 2529 * execbuf.flags = 0; // Submits to index 0, which is rcs0 for this context 2530 * gem_execbuf(drm_fd, &execbuf); 2531 * 2532 * ... 2533 * execbuf.rsvd1 = ctx_id; 2534 * execbuf.flags = 1; // Submits to index 0, which is bcs0 for this context 2535 * gem_execbuf(drm_fd, &execbuf); 2536 */ 2537 2538 struct i915_context_param_engines { 2539 __u64 extensions; /* linked chain of extension blocks, 0 terminates */ 2540 #define I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE 0 /* see i915_context_engines_load_balance */ 2541 #define I915_CONTEXT_ENGINES_EXT_BOND 1 /* see i915_context_engines_bond */ 2542 #define I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT 2 /* see i915_context_engines_parallel_submit */ 2543 struct i915_engine_class_instance engines[]; 2544 } __attribute__((packed)); 2545 2546 #define I915_DEFINE_CONTEXT_PARAM_ENGINES(name__, N__) struct { \ 2547 __u64 extensions; \ 2548 struct i915_engine_class_instance engines[N__]; \ 2549 } __attribute__((packed)) name__ 2550 2551 /** 2552 * struct drm_i915_gem_context_create_ext_setparam - Context parameter 2553 * to set or query during context creation. 2554 */ 2555 struct drm_i915_gem_context_create_ext_setparam { 2556 /** @base: Extension link. See struct i915_user_extension. */ 2557 struct i915_user_extension base; 2558 2559 /** 2560 * @param: Context parameter to set or query. 2561 * See struct drm_i915_gem_context_param. 2562 */ 2563 struct drm_i915_gem_context_param param; 2564 }; 2565 2566 struct drm_i915_gem_context_destroy { 2567 __u32 ctx_id; 2568 __u32 pad; 2569 }; 2570 2571 /** 2572 * struct drm_i915_gem_vm_control - Structure to create or destroy VM. 2573 * 2574 * DRM_I915_GEM_VM_CREATE - 2575 * 2576 * Create a new virtual memory address space (ppGTT) for use within a context 2577 * on the same file. Extensions can be provided to configure exactly how the 2578 * address space is setup upon creation. 2579 * 2580 * The id of new VM (bound to the fd) for use with I915_CONTEXT_PARAM_VM is 2581 * returned in the outparam @id. 2582 * 2583 * An extension chain maybe provided, starting with @extensions, and terminated 2584 * by the @next_extension being 0. Currently, no extensions are defined. 2585 * 2586 * DRM_I915_GEM_VM_DESTROY - 2587 * 2588 * Destroys a previously created VM id, specified in @vm_id. 2589 * 2590 * No extensions or flags are allowed currently, and so must be zero. 2591 */ 2592 struct drm_i915_gem_vm_control { 2593 /** @extensions: Zero-terminated chain of extensions. */ 2594 __u64 extensions; 2595 2596 /** @flags: reserved for future usage, currently MBZ */ 2597 __u32 flags; 2598 2599 /** @vm_id: Id of the VM created or to be destroyed */ 2600 __u32 vm_id; 2601 }; 2602 2603 struct drm_i915_reg_read { 2604 /* 2605 * Register offset. 2606 * For 64bit wide registers where the upper 32bits don't immediately 2607 * follow the lower 32bits, the offset of the lower 32bits must 2608 * be specified 2609 */ 2610 __u64 offset; 2611 #define I915_REG_READ_8B_WA (1ul << 0) 2612 2613 __u64 val; /* Return value */ 2614 }; 2615 2616 /* Known registers: 2617 * 2618 * Render engine timestamp - 0x2358 + 64bit - gen7+ 2619 * - Note this register returns an invalid value if using the default 2620 * single instruction 8byte read, in order to workaround that pass 2621 * flag I915_REG_READ_8B_WA in offset field. 2622 * 2623 */ 2624 2625 struct drm_i915_reset_stats { 2626 __u32 ctx_id; 2627 __u32 flags; 2628 2629 /* All resets since boot/module reload, for all contexts */ 2630 __u32 reset_count; 2631 2632 /* Number of batches lost when active in GPU, for this context */ 2633 __u32 batch_active; 2634 2635 /* Number of batches lost pending for execution, for this context */ 2636 __u32 batch_pending; 2637 2638 __u32 pad; 2639 }; 2640 2641 /** 2642 * struct drm_i915_gem_userptr - Create GEM object from user allocated memory. 2643 * 2644 * Userptr objects have several restrictions on what ioctls can be used with the 2645 * object handle. 2646 */ 2647 struct drm_i915_gem_userptr { 2648 /** 2649 * @user_ptr: The pointer to the allocated memory. 2650 * 2651 * Needs to be aligned to PAGE_SIZE. 2652 */ 2653 __u64 user_ptr; 2654 2655 /** 2656 * @user_size: 2657 * 2658 * The size in bytes for the allocated memory. This will also become the 2659 * object size. 2660 * 2661 * Needs to be aligned to PAGE_SIZE, and should be at least PAGE_SIZE, 2662 * or larger. 2663 */ 2664 __u64 user_size; 2665 2666 /** 2667 * @flags: 2668 * 2669 * Supported flags: 2670 * 2671 * I915_USERPTR_READ_ONLY: 2672 * 2673 * Mark the object as readonly, this also means GPU access can only be 2674 * readonly. This is only supported on HW which supports readonly access 2675 * through the GTT. If the HW can't support readonly access, an error is 2676 * returned. 2677 * 2678 * I915_USERPTR_PROBE: 2679 * 2680 * Probe the provided @user_ptr range and validate that the @user_ptr is 2681 * indeed pointing to normal memory and that the range is also valid. 2682 * For example if some garbage address is given to the kernel, then this 2683 * should complain. 2684 * 2685 * Returns -EFAULT if the probe failed. 2686 * 2687 * Note that this doesn't populate the backing pages, and also doesn't 2688 * guarantee that the object will remain valid when the object is 2689 * eventually used. 2690 * 2691 * The kernel supports this feature if I915_PARAM_HAS_USERPTR_PROBE 2692 * returns a non-zero value. 2693 * 2694 * I915_USERPTR_UNSYNCHRONIZED: 2695 * 2696 * NOT USED. Setting this flag will result in an error. 2697 */ 2698 __u32 flags; 2699 #define I915_USERPTR_READ_ONLY 0x1 2700 #define I915_USERPTR_PROBE 0x2 2701 #define I915_USERPTR_UNSYNCHRONIZED 0x80000000 2702 /** 2703 * @handle: Returned handle for the object. 2704 * 2705 * Object handles are nonzero. 2706 */ 2707 __u32 handle; 2708 }; 2709 2710 enum drm_i915_oa_format { 2711 I915_OA_FORMAT_A13 = 1, /* HSW only */ 2712 I915_OA_FORMAT_A29, /* HSW only */ 2713 I915_OA_FORMAT_A13_B8_C8, /* HSW only */ 2714 I915_OA_FORMAT_B4_C8, /* HSW only */ 2715 I915_OA_FORMAT_A45_B8_C8, /* HSW only */ 2716 I915_OA_FORMAT_B4_C8_A16, /* HSW only */ 2717 I915_OA_FORMAT_C4_B8, /* HSW+ */ 2718 2719 /* Gen8+ */ 2720 I915_OA_FORMAT_A12, 2721 I915_OA_FORMAT_A12_B8_C8, 2722 I915_OA_FORMAT_A32u40_A4u32_B8_C8, 2723 2724 /* DG2 */ 2725 I915_OAR_FORMAT_A32u40_A4u32_B8_C8, 2726 I915_OA_FORMAT_A24u40_A14u32_B8_C8, 2727 2728 /* MTL OAM */ 2729 I915_OAM_FORMAT_MPEC8u64_B8_C8, 2730 I915_OAM_FORMAT_MPEC8u32_B8_C8, 2731 2732 I915_OA_FORMAT_MAX /* non-ABI */ 2733 }; 2734 2735 enum drm_i915_perf_property_id { 2736 /** 2737 * Open the stream for a specific context handle (as used with 2738 * execbuffer2). A stream opened for a specific context this way 2739 * won't typically require root privileges. 2740 * 2741 * This property is available in perf revision 1. 2742 */ 2743 DRM_I915_PERF_PROP_CTX_HANDLE = 1, 2744 2745 /** 2746 * A value of 1 requests the inclusion of raw OA unit reports as 2747 * part of stream samples. 2748 * 2749 * This property is available in perf revision 1. 2750 */ 2751 DRM_I915_PERF_PROP_SAMPLE_OA, 2752 2753 /** 2754 * The value specifies which set of OA unit metrics should be 2755 * configured, defining the contents of any OA unit reports. 2756 * 2757 * This property is available in perf revision 1. 2758 */ 2759 DRM_I915_PERF_PROP_OA_METRICS_SET, 2760 2761 /** 2762 * The value specifies the size and layout of OA unit reports. 2763 * 2764 * This property is available in perf revision 1. 2765 */ 2766 DRM_I915_PERF_PROP_OA_FORMAT, 2767 2768 /** 2769 * Specifying this property implicitly requests periodic OA unit 2770 * sampling and (at least on Haswell) the sampling frequency is derived 2771 * from this exponent as follows: 2772 * 2773 * 80ns * 2^(period_exponent + 1) 2774 * 2775 * This property is available in perf revision 1. 2776 */ 2777 DRM_I915_PERF_PROP_OA_EXPONENT, 2778 2779 /** 2780 * Specifying this property is only valid when specify a context to 2781 * filter with DRM_I915_PERF_PROP_CTX_HANDLE. Specifying this property 2782 * will hold preemption of the particular context we want to gather 2783 * performance data about. The execbuf2 submissions must include a 2784 * drm_i915_gem_execbuffer_ext_perf parameter for this to apply. 2785 * 2786 * This property is available in perf revision 3. 2787 */ 2788 DRM_I915_PERF_PROP_HOLD_PREEMPTION, 2789 2790 /** 2791 * Specifying this pins all contexts to the specified SSEU power 2792 * configuration for the duration of the recording. 2793 * 2794 * This parameter's value is a pointer to a struct 2795 * drm_i915_gem_context_param_sseu. 2796 * 2797 * This property is available in perf revision 4. 2798 */ 2799 DRM_I915_PERF_PROP_GLOBAL_SSEU, 2800 2801 /** 2802 * This optional parameter specifies the timer interval in nanoseconds 2803 * at which the i915 driver will check the OA buffer for available data. 2804 * Minimum allowed value is 100 microseconds. A default value is used by 2805 * the driver if this parameter is not specified. Note that larger timer 2806 * values will reduce cpu consumption during OA perf captures. However, 2807 * excessively large values would potentially result in OA buffer 2808 * overwrites as captures reach end of the OA buffer. 2809 * 2810 * This property is available in perf revision 5. 2811 */ 2812 DRM_I915_PERF_PROP_POLL_OA_PERIOD, 2813 2814 /** 2815 * Multiple engines may be mapped to the same OA unit. The OA unit is 2816 * identified by class:instance of any engine mapped to it. 2817 * 2818 * This parameter specifies the engine class and must be passed along 2819 * with DRM_I915_PERF_PROP_OA_ENGINE_INSTANCE. 2820 * 2821 * This property is available in perf revision 6. 2822 */ 2823 DRM_I915_PERF_PROP_OA_ENGINE_CLASS, 2824 2825 /** 2826 * This parameter specifies the engine instance and must be passed along 2827 * with DRM_I915_PERF_PROP_OA_ENGINE_CLASS. 2828 * 2829 * This property is available in perf revision 6. 2830 */ 2831 DRM_I915_PERF_PROP_OA_ENGINE_INSTANCE, 2832 2833 DRM_I915_PERF_PROP_MAX /* non-ABI */ 2834 }; 2835 2836 struct drm_i915_perf_open_param { 2837 __u32 flags; 2838 #define I915_PERF_FLAG_FD_CLOEXEC (1<<0) 2839 #define I915_PERF_FLAG_FD_NONBLOCK (1<<1) 2840 #define I915_PERF_FLAG_DISABLED (1<<2) 2841 2842 /** The number of u64 (id, value) pairs */ 2843 __u32 num_properties; 2844 2845 /** 2846 * Pointer to array of u64 (id, value) pairs configuring the stream 2847 * to open. 2848 */ 2849 __u64 properties_ptr; 2850 }; 2851 2852 /* 2853 * Enable data capture for a stream that was either opened in a disabled state 2854 * via I915_PERF_FLAG_DISABLED or was later disabled via 2855 * I915_PERF_IOCTL_DISABLE. 2856 * 2857 * It is intended to be cheaper to disable and enable a stream than it may be 2858 * to close and re-open a stream with the same configuration. 2859 * 2860 * It's undefined whether any pending data for the stream will be lost. 2861 * 2862 * This ioctl is available in perf revision 1. 2863 */ 2864 #define I915_PERF_IOCTL_ENABLE _IO('i', 0x0) 2865 2866 /* 2867 * Disable data capture for a stream. 2868 * 2869 * It is an error to try and read a stream that is disabled. 2870 * 2871 * This ioctl is available in perf revision 1. 2872 */ 2873 #define I915_PERF_IOCTL_DISABLE _IO('i', 0x1) 2874 2875 /* 2876 * Change metrics_set captured by a stream. 2877 * 2878 * If the stream is bound to a specific context, the configuration change 2879 * will performed inline with that context such that it takes effect before 2880 * the next execbuf submission. 2881 * 2882 * Returns the previously bound metrics set id, or a negative error code. 2883 * 2884 * This ioctl is available in perf revision 2. 2885 */ 2886 #define I915_PERF_IOCTL_CONFIG _IO('i', 0x2) 2887 2888 /* 2889 * Common to all i915 perf records 2890 */ 2891 struct drm_i915_perf_record_header { 2892 __u32 type; 2893 __u16 pad; 2894 __u16 size; 2895 }; 2896 2897 enum drm_i915_perf_record_type { 2898 2899 /** 2900 * Samples are the work horse record type whose contents are extensible 2901 * and defined when opening an i915 perf stream based on the given 2902 * properties. 2903 * 2904 * Boolean properties following the naming convention 2905 * DRM_I915_PERF_SAMPLE_xyz_PROP request the inclusion of 'xyz' data in 2906 * every sample. 2907 * 2908 * The order of these sample properties given by userspace has no 2909 * affect on the ordering of data within a sample. The order is 2910 * documented here. 2911 * 2912 * struct { 2913 * struct drm_i915_perf_record_header header; 2914 * 2915 * { u32 oa_report[]; } && DRM_I915_PERF_PROP_SAMPLE_OA 2916 * }; 2917 */ 2918 DRM_I915_PERF_RECORD_SAMPLE = 1, 2919 2920 /* 2921 * Indicates that one or more OA reports were not written by the 2922 * hardware. This can happen for example if an MI_REPORT_PERF_COUNT 2923 * command collides with periodic sampling - which would be more likely 2924 * at higher sampling frequencies. 2925 */ 2926 DRM_I915_PERF_RECORD_OA_REPORT_LOST = 2, 2927 2928 /** 2929 * An error occurred that resulted in all pending OA reports being lost. 2930 */ 2931 DRM_I915_PERF_RECORD_OA_BUFFER_LOST = 3, 2932 2933 DRM_I915_PERF_RECORD_MAX /* non-ABI */ 2934 }; 2935 2936 /** 2937 * struct drm_i915_perf_oa_config 2938 * 2939 * Structure to upload perf dynamic configuration into the kernel. 2940 */ 2941 struct drm_i915_perf_oa_config { 2942 /** 2943 * @uuid: 2944 * 2945 * String formatted like "%\08x-%\04x-%\04x-%\04x-%\012x" 2946 */ 2947 char uuid[36]; 2948 2949 /** 2950 * @n_mux_regs: 2951 * 2952 * Number of mux regs in &mux_regs_ptr. 2953 */ 2954 __u32 n_mux_regs; 2955 2956 /** 2957 * @n_boolean_regs: 2958 * 2959 * Number of boolean regs in &boolean_regs_ptr. 2960 */ 2961 __u32 n_boolean_regs; 2962 2963 /** 2964 * @n_flex_regs: 2965 * 2966 * Number of flex regs in &flex_regs_ptr. 2967 */ 2968 __u32 n_flex_regs; 2969 2970 /** 2971 * @mux_regs_ptr: 2972 * 2973 * Pointer to tuples of u32 values (register address, value) for mux 2974 * registers. Expected length of buffer is (2 * sizeof(u32) * 2975 * &n_mux_regs). 2976 */ 2977 __u64 mux_regs_ptr; 2978 2979 /** 2980 * @boolean_regs_ptr: 2981 * 2982 * Pointer to tuples of u32 values (register address, value) for mux 2983 * registers. Expected length of buffer is (2 * sizeof(u32) * 2984 * &n_boolean_regs). 2985 */ 2986 __u64 boolean_regs_ptr; 2987 2988 /** 2989 * @flex_regs_ptr: 2990 * 2991 * Pointer to tuples of u32 values (register address, value) for mux 2992 * registers. Expected length of buffer is (2 * sizeof(u32) * 2993 * &n_flex_regs). 2994 */ 2995 __u64 flex_regs_ptr; 2996 }; 2997 2998 /** 2999 * struct drm_i915_query_item - An individual query for the kernel to process. 3000 * 3001 * The behaviour is determined by the @query_id. Note that exactly what 3002 * @data_ptr is also depends on the specific @query_id. 3003 */ 3004 struct drm_i915_query_item { 3005 /** 3006 * @query_id: 3007 * 3008 * The id for this query. Currently accepted query IDs are: 3009 * - %DRM_I915_QUERY_TOPOLOGY_INFO (see struct drm_i915_query_topology_info) 3010 * - %DRM_I915_QUERY_ENGINE_INFO (see struct drm_i915_engine_info) 3011 * - %DRM_I915_QUERY_PERF_CONFIG (see struct drm_i915_query_perf_config) 3012 * - %DRM_I915_QUERY_MEMORY_REGIONS (see struct drm_i915_query_memory_regions) 3013 * - %DRM_I915_QUERY_HWCONFIG_BLOB (see `GuC HWCONFIG blob uAPI`) 3014 * - %DRM_I915_QUERY_GEOMETRY_SUBSLICES (see struct drm_i915_query_topology_info) 3015 */ 3016 __u64 query_id; 3017 #define DRM_I915_QUERY_TOPOLOGY_INFO 1 3018 #define DRM_I915_QUERY_ENGINE_INFO 2 3019 #define DRM_I915_QUERY_PERF_CONFIG 3 3020 #define DRM_I915_QUERY_MEMORY_REGIONS 4 3021 #define DRM_I915_QUERY_HWCONFIG_BLOB 5 3022 #define DRM_I915_QUERY_GEOMETRY_SUBSLICES 6 3023 /* Must be kept compact -- no holes and well documented */ 3024 3025 /** 3026 * @length: 3027 * 3028 * When set to zero by userspace, this is filled with the size of the 3029 * data to be written at the @data_ptr pointer. The kernel sets this 3030 * value to a negative value to signal an error on a particular query 3031 * item. 3032 */ 3033 __s32 length; 3034 3035 /** 3036 * @flags: 3037 * 3038 * When &query_id == %DRM_I915_QUERY_TOPOLOGY_INFO, must be 0. 3039 * 3040 * When &query_id == %DRM_I915_QUERY_PERF_CONFIG, must be one of the 3041 * following: 3042 * 3043 * - %DRM_I915_QUERY_PERF_CONFIG_LIST 3044 * - %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 3045 * - %DRM_I915_QUERY_PERF_CONFIG_FOR_UUID 3046 * 3047 * When &query_id == %DRM_I915_QUERY_GEOMETRY_SUBSLICES must contain 3048 * a struct i915_engine_class_instance that references a render engine. 3049 */ 3050 __u32 flags; 3051 #define DRM_I915_QUERY_PERF_CONFIG_LIST 1 3052 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2 3053 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID 3 3054 3055 /** 3056 * @data_ptr: 3057 * 3058 * Data will be written at the location pointed by @data_ptr when the 3059 * value of @length matches the length of the data to be written by the 3060 * kernel. 3061 */ 3062 __u64 data_ptr; 3063 }; 3064 3065 /** 3066 * struct drm_i915_query - Supply an array of struct drm_i915_query_item for the 3067 * kernel to fill out. 3068 * 3069 * Note that this is generally a two step process for each struct 3070 * drm_i915_query_item in the array: 3071 * 3072 * 1. Call the DRM_IOCTL_I915_QUERY, giving it our array of struct 3073 * drm_i915_query_item, with &drm_i915_query_item.length set to zero. The 3074 * kernel will then fill in the size, in bytes, which tells userspace how 3075 * memory it needs to allocate for the blob(say for an array of properties). 3076 * 3077 * 2. Next we call DRM_IOCTL_I915_QUERY again, this time with the 3078 * &drm_i915_query_item.data_ptr equal to our newly allocated blob. Note that 3079 * the &drm_i915_query_item.length should still be the same as what the 3080 * kernel previously set. At this point the kernel can fill in the blob. 3081 * 3082 * Note that for some query items it can make sense for userspace to just pass 3083 * in a buffer/blob equal to or larger than the required size. In this case only 3084 * a single ioctl call is needed. For some smaller query items this can work 3085 * quite well. 3086 * 3087 */ 3088 struct drm_i915_query { 3089 /** @num_items: The number of elements in the @items_ptr array */ 3090 __u32 num_items; 3091 3092 /** 3093 * @flags: Unused for now. Must be cleared to zero. 3094 */ 3095 __u32 flags; 3096 3097 /** 3098 * @items_ptr: 3099 * 3100 * Pointer to an array of struct drm_i915_query_item. The number of 3101 * array elements is @num_items. 3102 */ 3103 __u64 items_ptr; 3104 }; 3105 3106 /** 3107 * struct drm_i915_query_topology_info 3108 * 3109 * Describes slice/subslice/EU information queried by 3110 * %DRM_I915_QUERY_TOPOLOGY_INFO 3111 */ 3112 struct drm_i915_query_topology_info { 3113 /** 3114 * @flags: 3115 * 3116 * Unused for now. Must be cleared to zero. 3117 */ 3118 __u16 flags; 3119 3120 /** 3121 * @max_slices: 3122 * 3123 * The number of bits used to express the slice mask. 3124 */ 3125 __u16 max_slices; 3126 3127 /** 3128 * @max_subslices: 3129 * 3130 * The number of bits used to express the subslice mask. 3131 */ 3132 __u16 max_subslices; 3133 3134 /** 3135 * @max_eus_per_subslice: 3136 * 3137 * The number of bits in the EU mask that correspond to a single 3138 * subslice's EUs. 3139 */ 3140 __u16 max_eus_per_subslice; 3141 3142 /** 3143 * @subslice_offset: 3144 * 3145 * Offset in data[] at which the subslice masks are stored. 3146 */ 3147 __u16 subslice_offset; 3148 3149 /** 3150 * @subslice_stride: 3151 * 3152 * Stride at which each of the subslice masks for each slice are 3153 * stored. 3154 */ 3155 __u16 subslice_stride; 3156 3157 /** 3158 * @eu_offset: 3159 * 3160 * Offset in data[] at which the EU masks are stored. 3161 */ 3162 __u16 eu_offset; 3163 3164 /** 3165 * @eu_stride: 3166 * 3167 * Stride at which each of the EU masks for each subslice are stored. 3168 */ 3169 __u16 eu_stride; 3170 3171 /** 3172 * @data: 3173 * 3174 * Contains 3 pieces of information : 3175 * 3176 * - The slice mask with one bit per slice telling whether a slice is 3177 * available. The availability of slice X can be queried with the 3178 * following formula : 3179 * 3180 * .. code:: c 3181 * 3182 * (data[X / 8] >> (X % 8)) & 1 3183 * 3184 * Starting with Xe_HP platforms, Intel hardware no longer has 3185 * traditional slices so i915 will always report a single slice 3186 * (hardcoded slicemask = 0x1) which contains all of the platform's 3187 * subslices. I.e., the mask here does not reflect any of the newer 3188 * hardware concepts such as "gslices" or "cslices" since userspace 3189 * is capable of inferring those from the subslice mask. 3190 * 3191 * - The subslice mask for each slice with one bit per subslice telling 3192 * whether a subslice is available. Starting with Gen12 we use the 3193 * term "subslice" to refer to what the hardware documentation 3194 * describes as a "dual-subslices." The availability of subslice Y 3195 * in slice X can be queried with the following formula : 3196 * 3197 * .. code:: c 3198 * 3199 * (data[subslice_offset + X * subslice_stride + Y / 8] >> (Y % 8)) & 1 3200 * 3201 * - The EU mask for each subslice in each slice, with one bit per EU 3202 * telling whether an EU is available. The availability of EU Z in 3203 * subslice Y in slice X can be queried with the following formula : 3204 * 3205 * .. code:: c 3206 * 3207 * (data[eu_offset + 3208 * (X * max_subslices + Y) * eu_stride + 3209 * Z / 8 3210 * ] >> (Z % 8)) & 1 3211 */ 3212 __u8 data[]; 3213 }; 3214 3215 /** 3216 * DOC: Engine Discovery uAPI 3217 * 3218 * Engine discovery uAPI is a way of enumerating physical engines present in a 3219 * GPU associated with an open i915 DRM file descriptor. This supersedes the old 3220 * way of using `DRM_IOCTL_I915_GETPARAM` and engine identifiers like 3221 * `I915_PARAM_HAS_BLT`. 3222 * 3223 * The need for this interface came starting with Icelake and newer GPUs, which 3224 * started to establish a pattern of having multiple engines of a same class, 3225 * where not all instances were always completely functionally equivalent. 3226 * 3227 * Entry point for this uapi is `DRM_IOCTL_I915_QUERY` with the 3228 * `DRM_I915_QUERY_ENGINE_INFO` as the queried item id. 3229 * 3230 * Example for getting the list of engines: 3231 * 3232 * .. code-block:: C 3233 * 3234 * struct drm_i915_query_engine_info *info; 3235 * struct drm_i915_query_item item = { 3236 * .query_id = DRM_I915_QUERY_ENGINE_INFO; 3237 * }; 3238 * struct drm_i915_query query = { 3239 * .num_items = 1, 3240 * .items_ptr = (uintptr_t)&item, 3241 * }; 3242 * int err, i; 3243 * 3244 * // First query the size of the blob we need, this needs to be large 3245 * // enough to hold our array of engines. The kernel will fill out the 3246 * // item.length for us, which is the number of bytes we need. 3247 * // 3248 * // Alternatively a large buffer can be allocated straight away enabling 3249 * // querying in one pass, in which case item.length should contain the 3250 * // length of the provided buffer. 3251 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3252 * if (err) ... 3253 * 3254 * info = calloc(1, item.length); 3255 * // Now that we allocated the required number of bytes, we call the ioctl 3256 * // again, this time with the data_ptr pointing to our newly allocated 3257 * // blob, which the kernel can then populate with info on all engines. 3258 * item.data_ptr = (uintptr_t)&info, 3259 * 3260 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3261 * if (err) ... 3262 * 3263 * // We can now access each engine in the array 3264 * for (i = 0; i < info->num_engines; i++) { 3265 * struct drm_i915_engine_info einfo = info->engines[i]; 3266 * u16 class = einfo.engine.class; 3267 * u16 instance = einfo.engine.instance; 3268 * .... 3269 * } 3270 * 3271 * free(info); 3272 * 3273 * Each of the enumerated engines, apart from being defined by its class and 3274 * instance (see `struct i915_engine_class_instance`), also can have flags and 3275 * capabilities defined as documented in i915_drm.h. 3276 * 3277 * For instance video engines which support HEVC encoding will have the 3278 * `I915_VIDEO_CLASS_CAPABILITY_HEVC` capability bit set. 3279 * 3280 * Engine discovery only fully comes to its own when combined with the new way 3281 * of addressing engines when submitting batch buffers using contexts with 3282 * engine maps configured. 3283 */ 3284 3285 /** 3286 * struct drm_i915_engine_info 3287 * 3288 * Describes one engine and it's capabilities as known to the driver. 3289 */ 3290 struct drm_i915_engine_info { 3291 /** @engine: Engine class and instance. */ 3292 struct i915_engine_class_instance engine; 3293 3294 /** @rsvd0: Reserved field. */ 3295 __u32 rsvd0; 3296 3297 /** @flags: Engine flags. */ 3298 __u64 flags; 3299 #define I915_ENGINE_INFO_HAS_LOGICAL_INSTANCE (1 << 0) 3300 3301 /** @capabilities: Capabilities of this engine. */ 3302 __u64 capabilities; 3303 #define I915_VIDEO_CLASS_CAPABILITY_HEVC (1 << 0) 3304 #define I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC (1 << 1) 3305 3306 /** @logical_instance: Logical instance of engine */ 3307 __u16 logical_instance; 3308 3309 /** @rsvd1: Reserved fields. */ 3310 __u16 rsvd1[3]; 3311 /** @rsvd2: Reserved fields. */ 3312 __u64 rsvd2[3]; 3313 }; 3314 3315 /** 3316 * struct drm_i915_query_engine_info 3317 * 3318 * Engine info query enumerates all engines known to the driver by filling in 3319 * an array of struct drm_i915_engine_info structures. 3320 */ 3321 struct drm_i915_query_engine_info { 3322 /** @num_engines: Number of struct drm_i915_engine_info structs following. */ 3323 __u32 num_engines; 3324 3325 /** @rsvd: MBZ */ 3326 __u32 rsvd[3]; 3327 3328 /** @engines: Marker for drm_i915_engine_info structures. */ 3329 struct drm_i915_engine_info engines[]; 3330 }; 3331 3332 /** 3333 * struct drm_i915_query_perf_config 3334 * 3335 * Data written by the kernel with query %DRM_I915_QUERY_PERF_CONFIG and 3336 * %DRM_I915_QUERY_GEOMETRY_SUBSLICES. 3337 */ 3338 struct drm_i915_query_perf_config { 3339 union { 3340 /** 3341 * @n_configs: 3342 * 3343 * When &drm_i915_query_item.flags == 3344 * %DRM_I915_QUERY_PERF_CONFIG_LIST, i915 sets this fields to 3345 * the number of configurations available. 3346 */ 3347 __u64 n_configs; 3348 3349 /** 3350 * @config: 3351 * 3352 * When &drm_i915_query_item.flags == 3353 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID, i915 will use the 3354 * value in this field as configuration identifier to decide 3355 * what data to write into config_ptr. 3356 */ 3357 __u64 config; 3358 3359 /** 3360 * @uuid: 3361 * 3362 * When &drm_i915_query_item.flags == 3363 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID, i915 will use the 3364 * value in this field as configuration identifier to decide 3365 * what data to write into config_ptr. 3366 * 3367 * String formatted like "%08x-%04x-%04x-%04x-%012x" 3368 */ 3369 char uuid[36]; 3370 }; 3371 3372 /** 3373 * @flags: 3374 * 3375 * Unused for now. Must be cleared to zero. 3376 */ 3377 __u32 flags; 3378 3379 /** 3380 * @data: 3381 * 3382 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_LIST, 3383 * i915 will write an array of __u64 of configuration identifiers. 3384 * 3385 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_DATA, 3386 * i915 will write a struct drm_i915_perf_oa_config. If the following 3387 * fields of struct drm_i915_perf_oa_config are not set to 0, i915 will 3388 * write into the associated pointers the values of submitted when the 3389 * configuration was created : 3390 * 3391 * - &drm_i915_perf_oa_config.n_mux_regs 3392 * - &drm_i915_perf_oa_config.n_boolean_regs 3393 * - &drm_i915_perf_oa_config.n_flex_regs 3394 */ 3395 __u8 data[]; 3396 }; 3397 3398 /** 3399 * enum drm_i915_gem_memory_class - Supported memory classes 3400 */ 3401 enum drm_i915_gem_memory_class { 3402 /** @I915_MEMORY_CLASS_SYSTEM: System memory */ 3403 I915_MEMORY_CLASS_SYSTEM = 0, 3404 /** @I915_MEMORY_CLASS_DEVICE: Device local-memory */ 3405 I915_MEMORY_CLASS_DEVICE, 3406 }; 3407 3408 /** 3409 * struct drm_i915_gem_memory_class_instance - Identify particular memory region 3410 */ 3411 struct drm_i915_gem_memory_class_instance { 3412 /** @memory_class: See enum drm_i915_gem_memory_class */ 3413 __u16 memory_class; 3414 3415 /** @memory_instance: Which instance */ 3416 __u16 memory_instance; 3417 }; 3418 3419 /** 3420 * struct drm_i915_memory_region_info - Describes one region as known to the 3421 * driver. 3422 * 3423 * Note this is using both struct drm_i915_query_item and struct drm_i915_query. 3424 * For this new query we are adding the new query id DRM_I915_QUERY_MEMORY_REGIONS 3425 * at &drm_i915_query_item.query_id. 3426 */ 3427 struct drm_i915_memory_region_info { 3428 /** @region: The class:instance pair encoding */ 3429 struct drm_i915_gem_memory_class_instance region; 3430 3431 /** @rsvd0: MBZ */ 3432 __u32 rsvd0; 3433 3434 /** 3435 * @probed_size: Memory probed by the driver 3436 * 3437 * Note that it should not be possible to ever encounter a zero value 3438 * here, also note that no current region type will ever return -1 here. 3439 * Although for future region types, this might be a possibility. The 3440 * same applies to the other size fields. 3441 */ 3442 __u64 probed_size; 3443 3444 /** 3445 * @unallocated_size: Estimate of memory remaining 3446 * 3447 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable accounting. 3448 * Without this (or if this is an older kernel) the value here will 3449 * always equal the @probed_size. Note this is only currently tracked 3450 * for I915_MEMORY_CLASS_DEVICE regions (for other types the value here 3451 * will always equal the @probed_size). 3452 */ 3453 __u64 unallocated_size; 3454 3455 union { 3456 /** @rsvd1: MBZ */ 3457 __u64 rsvd1[8]; 3458 struct { 3459 /** 3460 * @probed_cpu_visible_size: Memory probed by the driver 3461 * that is CPU accessible. 3462 * 3463 * This will be always be <= @probed_size, and the 3464 * remainder (if there is any) will not be CPU 3465 * accessible. 3466 * 3467 * On systems without small BAR, the @probed_size will 3468 * always equal the @probed_cpu_visible_size, since all 3469 * of it will be CPU accessible. 3470 * 3471 * Note this is only tracked for 3472 * I915_MEMORY_CLASS_DEVICE regions (for other types the 3473 * value here will always equal the @probed_size). 3474 * 3475 * Note that if the value returned here is zero, then 3476 * this must be an old kernel which lacks the relevant 3477 * small-bar uAPI support (including 3478 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS), but on 3479 * such systems we should never actually end up with a 3480 * small BAR configuration, assuming we are able to load 3481 * the kernel module. Hence it should be safe to treat 3482 * this the same as when @probed_cpu_visible_size == 3483 * @probed_size. 3484 */ 3485 __u64 probed_cpu_visible_size; 3486 3487 /** 3488 * @unallocated_cpu_visible_size: Estimate of CPU 3489 * visible memory remaining. 3490 * 3491 * Note this is only tracked for 3492 * I915_MEMORY_CLASS_DEVICE regions (for other types the 3493 * value here will always equal the 3494 * @probed_cpu_visible_size). 3495 * 3496 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable 3497 * accounting. Without this the value here will always 3498 * equal the @probed_cpu_visible_size. Note this is only 3499 * currently tracked for I915_MEMORY_CLASS_DEVICE 3500 * regions (for other types the value here will also 3501 * always equal the @probed_cpu_visible_size). 3502 * 3503 * If this is an older kernel the value here will be 3504 * zero, see also @probed_cpu_visible_size. 3505 */ 3506 __u64 unallocated_cpu_visible_size; 3507 }; 3508 }; 3509 }; 3510 3511 /** 3512 * struct drm_i915_query_memory_regions 3513 * 3514 * The region info query enumerates all regions known to the driver by filling 3515 * in an array of struct drm_i915_memory_region_info structures. 3516 * 3517 * Example for getting the list of supported regions: 3518 * 3519 * .. code-block:: C 3520 * 3521 * struct drm_i915_query_memory_regions *info; 3522 * struct drm_i915_query_item item = { 3523 * .query_id = DRM_I915_QUERY_MEMORY_REGIONS; 3524 * }; 3525 * struct drm_i915_query query = { 3526 * .num_items = 1, 3527 * .items_ptr = (uintptr_t)&item, 3528 * }; 3529 * int err, i; 3530 * 3531 * // First query the size of the blob we need, this needs to be large 3532 * // enough to hold our array of regions. The kernel will fill out the 3533 * // item.length for us, which is the number of bytes we need. 3534 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3535 * if (err) ... 3536 * 3537 * info = calloc(1, item.length); 3538 * // Now that we allocated the required number of bytes, we call the ioctl 3539 * // again, this time with the data_ptr pointing to our newly allocated 3540 * // blob, which the kernel can then populate with the all the region info. 3541 * item.data_ptr = (uintptr_t)&info, 3542 * 3543 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3544 * if (err) ... 3545 * 3546 * // We can now access each region in the array 3547 * for (i = 0; i < info->num_regions; i++) { 3548 * struct drm_i915_memory_region_info mr = info->regions[i]; 3549 * u16 class = mr.region.class; 3550 * u16 instance = mr.region.instance; 3551 * 3552 * .... 3553 * } 3554 * 3555 * free(info); 3556 */ 3557 struct drm_i915_query_memory_regions { 3558 /** @num_regions: Number of supported regions */ 3559 __u32 num_regions; 3560 3561 /** @rsvd: MBZ */ 3562 __u32 rsvd[3]; 3563 3564 /** @regions: Info about each supported region */ 3565 struct drm_i915_memory_region_info regions[]; 3566 }; 3567 3568 /** 3569 * DOC: GuC HWCONFIG blob uAPI 3570 * 3571 * The GuC produces a blob with information about the current device. 3572 * i915 reads this blob from GuC and makes it available via this uAPI. 3573 * 3574 * The format and meaning of the blob content are documented in the 3575 * Programmer's Reference Manual. 3576 */ 3577 3578 /** 3579 * struct drm_i915_gem_create_ext - Existing gem_create behaviour, with added 3580 * extension support using struct i915_user_extension. 3581 * 3582 * Note that new buffer flags should be added here, at least for the stuff that 3583 * is immutable. Previously we would have two ioctls, one to create the object 3584 * with gem_create, and another to apply various parameters, however this 3585 * creates some ambiguity for the params which are considered immutable. Also in 3586 * general we're phasing out the various SET/GET ioctls. 3587 */ 3588 struct drm_i915_gem_create_ext { 3589 /** 3590 * @size: Requested size for the object. 3591 * 3592 * The (page-aligned) allocated size for the object will be returned. 3593 * 3594 * On platforms like DG2/ATS the kernel will always use 64K or larger 3595 * pages for I915_MEMORY_CLASS_DEVICE. The kernel also requires a 3596 * minimum of 64K GTT alignment for such objects. 3597 * 3598 * NOTE: Previously the ABI here required a minimum GTT alignment of 2M 3599 * on DG2/ATS, due to how the hardware implemented 64K GTT page support, 3600 * where we had the following complications: 3601 * 3602 * 1) The entire PDE (which covers a 2MB virtual address range), must 3603 * contain only 64K PTEs, i.e mixing 4K and 64K PTEs in the same 3604 * PDE is forbidden by the hardware. 3605 * 3606 * 2) We still need to support 4K PTEs for I915_MEMORY_CLASS_SYSTEM 3607 * objects. 3608 * 3609 * However on actual production HW this was completely changed to now 3610 * allow setting a TLB hint at the PTE level (see PS64), which is a lot 3611 * more flexible than the above. With this the 2M restriction was 3612 * dropped where we now only require 64K. 3613 */ 3614 __u64 size; 3615 3616 /** 3617 * @handle: Returned handle for the object. 3618 * 3619 * Object handles are nonzero. 3620 */ 3621 __u32 handle; 3622 3623 /** 3624 * @flags: Optional flags. 3625 * 3626 * Supported values: 3627 * 3628 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS - Signal to the kernel that 3629 * the object will need to be accessed via the CPU. 3630 * 3631 * Only valid when placing objects in I915_MEMORY_CLASS_DEVICE, and only 3632 * strictly required on configurations where some subset of the device 3633 * memory is directly visible/mappable through the CPU (which we also 3634 * call small BAR), like on some DG2+ systems. Note that this is quite 3635 * undesirable, but due to various factors like the client CPU, BIOS etc 3636 * it's something we can expect to see in the wild. See 3637 * &drm_i915_memory_region_info.probed_cpu_visible_size for how to 3638 * determine if this system applies. 3639 * 3640 * Note that one of the placements MUST be I915_MEMORY_CLASS_SYSTEM, to 3641 * ensure the kernel can always spill the allocation to system memory, 3642 * if the object can't be allocated in the mappable part of 3643 * I915_MEMORY_CLASS_DEVICE. 3644 * 3645 * Also note that since the kernel only supports flat-CCS on objects 3646 * that can *only* be placed in I915_MEMORY_CLASS_DEVICE, we therefore 3647 * don't support I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS together with 3648 * flat-CCS. 3649 * 3650 * Without this hint, the kernel will assume that non-mappable 3651 * I915_MEMORY_CLASS_DEVICE is preferred for this object. Note that the 3652 * kernel can still migrate the object to the mappable part, as a last 3653 * resort, if userspace ever CPU faults this object, but this might be 3654 * expensive, and so ideally should be avoided. 3655 * 3656 * On older kernels which lack the relevant small-bar uAPI support (see 3657 * also &drm_i915_memory_region_info.probed_cpu_visible_size), 3658 * usage of the flag will result in an error, but it should NEVER be 3659 * possible to end up with a small BAR configuration, assuming we can 3660 * also successfully load the i915 kernel module. In such cases the 3661 * entire I915_MEMORY_CLASS_DEVICE region will be CPU accessible, and as 3662 * such there are zero restrictions on where the object can be placed. 3663 */ 3664 #define I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS (1 << 0) 3665 __u32 flags; 3666 3667 /** 3668 * @extensions: The chain of extensions to apply to this object. 3669 * 3670 * This will be useful in the future when we need to support several 3671 * different extensions, and we need to apply more than one when 3672 * creating the object. See struct i915_user_extension. 3673 * 3674 * If we don't supply any extensions then we get the same old gem_create 3675 * behaviour. 3676 * 3677 * For I915_GEM_CREATE_EXT_MEMORY_REGIONS usage see 3678 * struct drm_i915_gem_create_ext_memory_regions. 3679 * 3680 * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see 3681 * struct drm_i915_gem_create_ext_protected_content. 3682 */ 3683 #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 3684 #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 3685 __u64 extensions; 3686 }; 3687 3688 /** 3689 * struct drm_i915_gem_create_ext_memory_regions - The 3690 * I915_GEM_CREATE_EXT_MEMORY_REGIONS extension. 3691 * 3692 * Set the object with the desired set of placements/regions in priority 3693 * order. Each entry must be unique and supported by the device. 3694 * 3695 * This is provided as an array of struct drm_i915_gem_memory_class_instance, or 3696 * an equivalent layout of class:instance pair encodings. See struct 3697 * drm_i915_query_memory_regions and DRM_I915_QUERY_MEMORY_REGIONS for how to 3698 * query the supported regions for a device. 3699 * 3700 * As an example, on discrete devices, if we wish to set the placement as 3701 * device local-memory we can do something like: 3702 * 3703 * .. code-block:: C 3704 * 3705 * struct drm_i915_gem_memory_class_instance region_lmem = { 3706 * .memory_class = I915_MEMORY_CLASS_DEVICE, 3707 * .memory_instance = 0, 3708 * }; 3709 * struct drm_i915_gem_create_ext_memory_regions regions = { 3710 * .base = { .name = I915_GEM_CREATE_EXT_MEMORY_REGIONS }, 3711 * .regions = (uintptr_t)®ion_lmem, 3712 * .num_regions = 1, 3713 * }; 3714 * struct drm_i915_gem_create_ext create_ext = { 3715 * .size = 16 * PAGE_SIZE, 3716 * .extensions = (uintptr_t)®ions, 3717 * }; 3718 * 3719 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3720 * if (err) ... 3721 * 3722 * At which point we get the object handle in &drm_i915_gem_create_ext.handle, 3723 * along with the final object size in &drm_i915_gem_create_ext.size, which 3724 * should account for any rounding up, if required. 3725 * 3726 * Note that userspace has no means of knowing the current backing region 3727 * for objects where @num_regions is larger than one. The kernel will only 3728 * ensure that the priority order of the @regions array is honoured, either 3729 * when initially placing the object, or when moving memory around due to 3730 * memory pressure 3731 * 3732 * On Flat-CCS capable HW, compression is supported for the objects residing 3733 * in I915_MEMORY_CLASS_DEVICE. When such objects (compressed) have other 3734 * memory class in @regions and migrated (by i915, due to memory 3735 * constraints) to the non I915_MEMORY_CLASS_DEVICE region, then i915 needs to 3736 * decompress the content. But i915 doesn't have the required information to 3737 * decompress the userspace compressed objects. 3738 * 3739 * So i915 supports Flat-CCS, on the objects which can reside only on 3740 * I915_MEMORY_CLASS_DEVICE regions. 3741 */ 3742 struct drm_i915_gem_create_ext_memory_regions { 3743 /** @base: Extension link. See struct i915_user_extension. */ 3744 struct i915_user_extension base; 3745 3746 /** @pad: MBZ */ 3747 __u32 pad; 3748 /** @num_regions: Number of elements in the @regions array. */ 3749 __u32 num_regions; 3750 /** 3751 * @regions: The regions/placements array. 3752 * 3753 * An array of struct drm_i915_gem_memory_class_instance. 3754 */ 3755 __u64 regions; 3756 }; 3757 3758 /** 3759 * struct drm_i915_gem_create_ext_protected_content - The 3760 * I915_OBJECT_PARAM_PROTECTED_CONTENT extension. 3761 * 3762 * If this extension is provided, buffer contents are expected to be protected 3763 * by PXP encryption and require decryption for scan out and processing. This 3764 * is only possible on platforms that have PXP enabled, on all other scenarios 3765 * using this extension will cause the ioctl to fail and return -ENODEV. The 3766 * flags parameter is reserved for future expansion and must currently be set 3767 * to zero. 3768 * 3769 * The buffer contents are considered invalid after a PXP session teardown. 3770 * 3771 * The encryption is guaranteed to be processed correctly only if the object 3772 * is submitted with a context created using the 3773 * I915_CONTEXT_PARAM_PROTECTED_CONTENT flag. This will also enable extra checks 3774 * at submission time on the validity of the objects involved. 3775 * 3776 * Below is an example on how to create a protected object: 3777 * 3778 * .. code-block:: C 3779 * 3780 * struct drm_i915_gem_create_ext_protected_content protected_ext = { 3781 * .base = { .name = I915_GEM_CREATE_EXT_PROTECTED_CONTENT }, 3782 * .flags = 0, 3783 * }; 3784 * struct drm_i915_gem_create_ext create_ext = { 3785 * .size = PAGE_SIZE, 3786 * .extensions = (uintptr_t)&protected_ext, 3787 * }; 3788 * 3789 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3790 * if (err) ... 3791 */ 3792 struct drm_i915_gem_create_ext_protected_content { 3793 /** @base: Extension link. See struct i915_user_extension. */ 3794 struct i915_user_extension base; 3795 /** @flags: reserved for future usage, currently MBZ */ 3796 __u32 flags; 3797 }; 3798 3799 /* ID of the protected content session managed by i915 when PXP is active */ 3800 #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf 3801 3802 #if defined(__cplusplus) 3803 } 3804 #endif 3805 3806 #endif /* _UAPI_I915_DRM_H_ */ 3807