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 its 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 loaded and fully authenticated, 678 * * 2 if HuC firmware is loaded and authenticated for clear media only 679 */ 680 #define I915_PARAM_HUC_STATUS 42 681 682 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to opt-out of 683 * synchronisation with implicit fencing on individual objects. 684 * See EXEC_OBJECT_ASYNC. 685 */ 686 #define I915_PARAM_HAS_EXEC_ASYNC 43 687 688 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports explicit fence support - 689 * both being able to pass in a sync_file fd to wait upon before executing, 690 * and being able to return a new sync_file fd that is signaled when the 691 * current request is complete. See I915_EXEC_FENCE_IN and I915_EXEC_FENCE_OUT. 692 */ 693 #define I915_PARAM_HAS_EXEC_FENCE 44 694 695 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to capture 696 * user specified bufffers for post-mortem debugging of GPU hangs. See 697 * EXEC_OBJECT_CAPTURE. 698 */ 699 #define I915_PARAM_HAS_EXEC_CAPTURE 45 700 701 #define I915_PARAM_SLICE_MASK 46 702 703 /* Assuming it's uniform for each slice, this queries the mask of subslices 704 * per-slice for this system. 705 */ 706 #define I915_PARAM_SUBSLICE_MASK 47 707 708 /* 709 * Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying the batch buffer 710 * as the first execobject as opposed to the last. See I915_EXEC_BATCH_FIRST. 711 */ 712 #define I915_PARAM_HAS_EXEC_BATCH_FIRST 48 713 714 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 715 * drm_i915_gem_exec_fence structures. See I915_EXEC_FENCE_ARRAY. 716 */ 717 #define I915_PARAM_HAS_EXEC_FENCE_ARRAY 49 718 719 /* 720 * Query whether every context (both per-file default and user created) is 721 * isolated (insofar as HW supports). If this parameter is not true, then 722 * freshly created contexts may inherit values from an existing context, 723 * rather than default HW values. If true, it also ensures (insofar as HW 724 * supports) that all state set by this context will not leak to any other 725 * context. 726 * 727 * As not every engine across every gen support contexts, the returned 728 * value reports the support of context isolation for individual engines by 729 * returning a bitmask of each engine class set to true if that class supports 730 * isolation. 731 */ 732 #define I915_PARAM_HAS_CONTEXT_ISOLATION 50 733 734 /* Frequency of the command streamer timestamps given by the *_TIMESTAMP 735 * registers. This used to be fixed per platform but from CNL onwards, this 736 * might vary depending on the parts. 737 */ 738 #define I915_PARAM_CS_TIMESTAMP_FREQUENCY 51 739 740 /* 741 * Once upon a time we supposed that writes through the GGTT would be 742 * immediately in physical memory (once flushed out of the CPU path). However, 743 * on a few different processors and chipsets, this is not necessarily the case 744 * as the writes appear to be buffered internally. Thus a read of the backing 745 * storage (physical memory) via a different path (with different physical tags 746 * to the indirect write via the GGTT) will see stale values from before 747 * the GGTT write. Inside the kernel, we can for the most part keep track of 748 * the different read/write domains in use (e.g. set-domain), but the assumption 749 * of coherency is baked into the ABI, hence reporting its true state in this 750 * parameter. 751 * 752 * Reports true when writes via mmap_gtt are immediately visible following an 753 * lfence to flush the WCB. 754 * 755 * Reports false when writes via mmap_gtt are indeterminately delayed in an in 756 * internal buffer and are _not_ immediately visible to third parties accessing 757 * directly via mmap_cpu/mmap_wc. Use of mmap_gtt as part of an IPC 758 * communications channel when reporting false is strongly disadvised. 759 */ 760 #define I915_PARAM_MMAP_GTT_COHERENT 52 761 762 /* 763 * Query whether DRM_I915_GEM_EXECBUFFER2 supports coordination of parallel 764 * execution through use of explicit fence support. 765 * See I915_EXEC_FENCE_OUT and I915_EXEC_FENCE_SUBMIT. 766 */ 767 #define I915_PARAM_HAS_EXEC_SUBMIT_FENCE 53 768 769 /* 770 * Revision of the i915-perf uAPI. The value returned helps determine what 771 * i915-perf features are available. See drm_i915_perf_property_id. 772 */ 773 #define I915_PARAM_PERF_REVISION 54 774 775 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports supplying an array of 776 * timeline syncobj through drm_i915_gem_execbuffer_ext_timeline_fences. See 777 * I915_EXEC_USE_EXTENSIONS. 778 */ 779 #define I915_PARAM_HAS_EXEC_TIMELINE_FENCES 55 780 781 /* Query if the kernel supports the I915_USERPTR_PROBE flag. */ 782 #define I915_PARAM_HAS_USERPTR_PROBE 56 783 784 /* 785 * Frequency of the timestamps in OA reports. This used to be the same as the CS 786 * timestamp frequency, but differs on some platforms. 787 */ 788 #define I915_PARAM_OA_TIMESTAMP_FREQUENCY 57 789 790 /* 791 * Query the status of PXP support in i915. 792 * 793 * The query can fail in the following scenarios with the listed error codes: 794 * -ENODEV = PXP support is not available on the GPU device or in the 795 * kernel due to missing component drivers or kernel configs. 796 * 797 * If the IOCTL is successful, the returned parameter will be set to one of 798 * the following values: 799 * 1 = PXP feature is supported and is ready for use. 800 * 2 = PXP feature is supported but should be ready soon (pending 801 * initialization of non-i915 system dependencies). 802 * 803 * NOTE: When param is supported (positive return values), user space should 804 * still refer to the GEM PXP context-creation UAPI header specs to be 805 * aware of possible failure due to system state machine at the time. 806 */ 807 #define I915_PARAM_PXP_STATUS 58 808 809 /* Must be kept compact -- no holes and well documented */ 810 811 /** 812 * struct drm_i915_getparam - Driver parameter query structure. 813 */ 814 struct drm_i915_getparam { 815 /** @param: Driver parameter to query. */ 816 __s32 param; 817 818 /** 819 * @value: Address of memory where queried value should be put. 820 * 821 * WARNING: Using pointers instead of fixed-size u64 means we need to write 822 * compat32 code. Don't repeat this mistake. 823 */ 824 int __user *value; 825 }; 826 827 /** 828 * typedef drm_i915_getparam_t - Driver parameter query structure. 829 * See struct drm_i915_getparam. 830 */ 831 typedef struct drm_i915_getparam drm_i915_getparam_t; 832 833 /* Ioctl to set kernel params: 834 */ 835 #define I915_SETPARAM_USE_MI_BATCHBUFFER_START 1 836 #define I915_SETPARAM_TEX_LRU_LOG_GRANULARITY 2 837 #define I915_SETPARAM_ALLOW_BATCHBUFFER 3 838 #define I915_SETPARAM_NUM_USED_FENCES 4 839 /* Must be kept compact -- no holes */ 840 841 typedef struct drm_i915_setparam { 842 int param; 843 int value; 844 } drm_i915_setparam_t; 845 846 /* A memory manager for regions of shared memory: 847 */ 848 #define I915_MEM_REGION_AGP 1 849 850 typedef struct drm_i915_mem_alloc { 851 int region; 852 int alignment; 853 int size; 854 int __user *region_offset; /* offset from start of fb or agp */ 855 } drm_i915_mem_alloc_t; 856 857 typedef struct drm_i915_mem_free { 858 int region; 859 int region_offset; 860 } drm_i915_mem_free_t; 861 862 typedef struct drm_i915_mem_init_heap { 863 int region; 864 int size; 865 int start; 866 } drm_i915_mem_init_heap_t; 867 868 /* Allow memory manager to be torn down and re-initialized (eg on 869 * rotate): 870 */ 871 typedef struct drm_i915_mem_destroy_heap { 872 int region; 873 } drm_i915_mem_destroy_heap_t; 874 875 /* Allow X server to configure which pipes to monitor for vblank signals 876 */ 877 #define DRM_I915_VBLANK_PIPE_A 1 878 #define DRM_I915_VBLANK_PIPE_B 2 879 880 typedef struct drm_i915_vblank_pipe { 881 int pipe; 882 } drm_i915_vblank_pipe_t; 883 884 /* Schedule buffer swap at given vertical blank: 885 */ 886 typedef struct drm_i915_vblank_swap { 887 drm_drawable_t drawable; 888 enum drm_vblank_seq_type seqtype; 889 unsigned int sequence; 890 } drm_i915_vblank_swap_t; 891 892 typedef struct drm_i915_hws_addr { 893 __u64 addr; 894 } drm_i915_hws_addr_t; 895 896 struct drm_i915_gem_init { 897 /** 898 * Beginning offset in the GTT to be managed by the DRM memory 899 * manager. 900 */ 901 __u64 gtt_start; 902 /** 903 * Ending offset in the GTT to be managed by the DRM memory 904 * manager. 905 */ 906 __u64 gtt_end; 907 }; 908 909 struct drm_i915_gem_create { 910 /** 911 * Requested size for the object. 912 * 913 * The (page-aligned) allocated size for the object will be returned. 914 */ 915 __u64 size; 916 /** 917 * Returned handle for the object. 918 * 919 * Object handles are nonzero. 920 */ 921 __u32 handle; 922 __u32 pad; 923 }; 924 925 struct drm_i915_gem_pread { 926 /** Handle for the object being read. */ 927 __u32 handle; 928 __u32 pad; 929 /** Offset into the object to read from */ 930 __u64 offset; 931 /** Length of data to read */ 932 __u64 size; 933 /** 934 * Pointer to write the data into. 935 * 936 * This is a fixed-size type for 32/64 compatibility. 937 */ 938 __u64 data_ptr; 939 }; 940 941 struct drm_i915_gem_pwrite { 942 /** Handle for the object being written to. */ 943 __u32 handle; 944 __u32 pad; 945 /** Offset into the object to write to */ 946 __u64 offset; 947 /** Length of data to write */ 948 __u64 size; 949 /** 950 * Pointer to read the data from. 951 * 952 * This is a fixed-size type for 32/64 compatibility. 953 */ 954 __u64 data_ptr; 955 }; 956 957 struct drm_i915_gem_mmap { 958 /** Handle for the object being mapped. */ 959 __u32 handle; 960 __u32 pad; 961 /** Offset in the object to map. */ 962 __u64 offset; 963 /** 964 * Length of data to map. 965 * 966 * The value will be page-aligned. 967 */ 968 __u64 size; 969 /** 970 * Returned pointer the data was mapped at. 971 * 972 * This is a fixed-size type for 32/64 compatibility. 973 */ 974 __u64 addr_ptr; 975 976 /** 977 * Flags for extended behaviour. 978 * 979 * Added in version 2. 980 */ 981 __u64 flags; 982 #define I915_MMAP_WC 0x1 983 }; 984 985 struct drm_i915_gem_mmap_gtt { 986 /** Handle for the object being mapped. */ 987 __u32 handle; 988 __u32 pad; 989 /** 990 * Fake offset to use for subsequent mmap call 991 * 992 * This is a fixed-size type for 32/64 compatibility. 993 */ 994 __u64 offset; 995 }; 996 997 /** 998 * struct drm_i915_gem_mmap_offset - Retrieve an offset so we can mmap this buffer object. 999 * 1000 * This struct is passed as argument to the `DRM_IOCTL_I915_GEM_MMAP_OFFSET` ioctl, 1001 * and is used to retrieve the fake offset to mmap an object specified by &handle. 1002 * 1003 * The legacy way of using `DRM_IOCTL_I915_GEM_MMAP` is removed on gen12+. 1004 * `DRM_IOCTL_I915_GEM_MMAP_GTT` is an older supported alias to this struct, but will behave 1005 * as setting the &extensions to 0, and &flags to `I915_MMAP_OFFSET_GTT`. 1006 */ 1007 struct drm_i915_gem_mmap_offset { 1008 /** @handle: Handle for the object being mapped. */ 1009 __u32 handle; 1010 /** @pad: Must be zero */ 1011 __u32 pad; 1012 /** 1013 * @offset: The fake offset to use for subsequent mmap call 1014 * 1015 * This is a fixed-size type for 32/64 compatibility. 1016 */ 1017 __u64 offset; 1018 1019 /** 1020 * @flags: Flags for extended behaviour. 1021 * 1022 * It is mandatory that one of the `MMAP_OFFSET` types 1023 * should be included: 1024 * 1025 * - `I915_MMAP_OFFSET_GTT`: Use mmap with the object bound to GTT. (Write-Combined) 1026 * - `I915_MMAP_OFFSET_WC`: Use Write-Combined caching. 1027 * - `I915_MMAP_OFFSET_WB`: Use Write-Back caching. 1028 * - `I915_MMAP_OFFSET_FIXED`: Use object placement to determine caching. 1029 * 1030 * On devices with local memory `I915_MMAP_OFFSET_FIXED` is the only valid 1031 * type. On devices without local memory, this caching mode is invalid. 1032 * 1033 * As caching mode when specifying `I915_MMAP_OFFSET_FIXED`, WC or WB will 1034 * be used, depending on the object placement on creation. WB will be used 1035 * when the object can only exist in system memory, WC otherwise. 1036 */ 1037 __u64 flags; 1038 1039 #define I915_MMAP_OFFSET_GTT 0 1040 #define I915_MMAP_OFFSET_WC 1 1041 #define I915_MMAP_OFFSET_WB 2 1042 #define I915_MMAP_OFFSET_UC 3 1043 #define I915_MMAP_OFFSET_FIXED 4 1044 1045 /** 1046 * @extensions: Zero-terminated chain of extensions. 1047 * 1048 * No current extensions defined; mbz. 1049 */ 1050 __u64 extensions; 1051 }; 1052 1053 /** 1054 * struct drm_i915_gem_set_domain - Adjust the objects write or read domain, in 1055 * preparation for accessing the pages via some CPU domain. 1056 * 1057 * Specifying a new write or read domain will flush the object out of the 1058 * previous domain(if required), before then updating the objects domain 1059 * tracking with the new domain. 1060 * 1061 * Note this might involve waiting for the object first if it is still active on 1062 * the GPU. 1063 * 1064 * Supported values for @read_domains and @write_domain: 1065 * 1066 * - I915_GEM_DOMAIN_WC: Uncached write-combined domain 1067 * - I915_GEM_DOMAIN_CPU: CPU cache domain 1068 * - I915_GEM_DOMAIN_GTT: Mappable aperture domain 1069 * 1070 * All other domains are rejected. 1071 * 1072 * Note that for discrete, starting from DG1, this is no longer supported, and 1073 * is instead rejected. On such platforms the CPU domain is effectively static, 1074 * where we also only support a single &drm_i915_gem_mmap_offset cache mode, 1075 * which can't be set explicitly and instead depends on the object placements, 1076 * as per the below. 1077 * 1078 * Implicit caching rules, starting from DG1: 1079 * 1080 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1081 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1082 * mapped as write-combined only. 1083 * 1084 * - Everything else is always allocated and mapped as write-back, with the 1085 * guarantee that everything is also coherent with the GPU. 1086 * 1087 * Note that this is likely to change in the future again, where we might need 1088 * more flexibility on future devices, so making this all explicit as part of a 1089 * new &drm_i915_gem_create_ext extension is probable. 1090 */ 1091 struct drm_i915_gem_set_domain { 1092 /** @handle: Handle for the object. */ 1093 __u32 handle; 1094 1095 /** @read_domains: New read domains. */ 1096 __u32 read_domains; 1097 1098 /** 1099 * @write_domain: New write domain. 1100 * 1101 * Note that having something in the write domain implies it's in the 1102 * read domain, and only that read domain. 1103 */ 1104 __u32 write_domain; 1105 }; 1106 1107 struct drm_i915_gem_sw_finish { 1108 /** Handle for the object */ 1109 __u32 handle; 1110 }; 1111 1112 struct drm_i915_gem_relocation_entry { 1113 /** 1114 * Handle of the buffer being pointed to by this relocation entry. 1115 * 1116 * It's appealing to make this be an index into the mm_validate_entry 1117 * list to refer to the buffer, but this allows the driver to create 1118 * a relocation list for state buffers and not re-write it per 1119 * exec using the buffer. 1120 */ 1121 __u32 target_handle; 1122 1123 /** 1124 * Value to be added to the offset of the target buffer to make up 1125 * the relocation entry. 1126 */ 1127 __u32 delta; 1128 1129 /** Offset in the buffer the relocation entry will be written into */ 1130 __u64 offset; 1131 1132 /** 1133 * Offset value of the target buffer that the relocation entry was last 1134 * written as. 1135 * 1136 * If the buffer has the same offset as last time, we can skip syncing 1137 * and writing the relocation. This value is written back out by 1138 * the execbuffer ioctl when the relocation is written. 1139 */ 1140 __u64 presumed_offset; 1141 1142 /** 1143 * Target memory domains read by this operation. 1144 */ 1145 __u32 read_domains; 1146 1147 /** 1148 * Target memory domains written by this operation. 1149 * 1150 * Note that only one domain may be written by the whole 1151 * execbuffer operation, so that where there are conflicts, 1152 * the application will get -EINVAL back. 1153 */ 1154 __u32 write_domain; 1155 }; 1156 1157 /** @{ 1158 * Intel memory domains 1159 * 1160 * Most of these just align with the various caches in 1161 * the system and are used to flush and invalidate as 1162 * objects end up cached in different domains. 1163 */ 1164 /** CPU cache */ 1165 #define I915_GEM_DOMAIN_CPU 0x00000001 1166 /** Render cache, used by 2D and 3D drawing */ 1167 #define I915_GEM_DOMAIN_RENDER 0x00000002 1168 /** Sampler cache, used by texture engine */ 1169 #define I915_GEM_DOMAIN_SAMPLER 0x00000004 1170 /** Command queue, used to load batch buffers */ 1171 #define I915_GEM_DOMAIN_COMMAND 0x00000008 1172 /** Instruction cache, used by shader programs */ 1173 #define I915_GEM_DOMAIN_INSTRUCTION 0x00000010 1174 /** Vertex address cache */ 1175 #define I915_GEM_DOMAIN_VERTEX 0x00000020 1176 /** GTT domain - aperture and scanout */ 1177 #define I915_GEM_DOMAIN_GTT 0x00000040 1178 /** WC domain - uncached access */ 1179 #define I915_GEM_DOMAIN_WC 0x00000080 1180 /** @} */ 1181 1182 struct drm_i915_gem_exec_object { 1183 /** 1184 * User's handle for a buffer to be bound into the GTT for this 1185 * operation. 1186 */ 1187 __u32 handle; 1188 1189 /** Number of relocations to be performed on this buffer */ 1190 __u32 relocation_count; 1191 /** 1192 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1193 * the relocations to be performed in this buffer. 1194 */ 1195 __u64 relocs_ptr; 1196 1197 /** Required alignment in graphics aperture */ 1198 __u64 alignment; 1199 1200 /** 1201 * Returned value of the updated offset of the object, for future 1202 * presumed_offset writes. 1203 */ 1204 __u64 offset; 1205 }; 1206 1207 /* DRM_IOCTL_I915_GEM_EXECBUFFER was removed in Linux 5.13 */ 1208 struct drm_i915_gem_execbuffer { 1209 /** 1210 * List of buffers to be validated with their relocations to be 1211 * performend on them. 1212 * 1213 * This is a pointer to an array of struct drm_i915_gem_validate_entry. 1214 * 1215 * These buffers must be listed in an order such that all relocations 1216 * a buffer is performing refer to buffers that have already appeared 1217 * in the validate list. 1218 */ 1219 __u64 buffers_ptr; 1220 __u32 buffer_count; 1221 1222 /** Offset in the batchbuffer to start execution from. */ 1223 __u32 batch_start_offset; 1224 /** Bytes used in batchbuffer from batch_start_offset */ 1225 __u32 batch_len; 1226 __u32 DR1; 1227 __u32 DR4; 1228 __u32 num_cliprects; 1229 /** This is a struct drm_clip_rect *cliprects */ 1230 __u64 cliprects_ptr; 1231 }; 1232 1233 struct drm_i915_gem_exec_object2 { 1234 /** 1235 * User's handle for a buffer to be bound into the GTT for this 1236 * operation. 1237 */ 1238 __u32 handle; 1239 1240 /** Number of relocations to be performed on this buffer */ 1241 __u32 relocation_count; 1242 /** 1243 * Pointer to array of struct drm_i915_gem_relocation_entry containing 1244 * the relocations to be performed in this buffer. 1245 */ 1246 __u64 relocs_ptr; 1247 1248 /** Required alignment in graphics aperture */ 1249 __u64 alignment; 1250 1251 /** 1252 * When the EXEC_OBJECT_PINNED flag is specified this is populated by 1253 * the user with the GTT offset at which this object will be pinned. 1254 * 1255 * When the I915_EXEC_NO_RELOC flag is specified this must contain the 1256 * presumed_offset of the object. 1257 * 1258 * During execbuffer2 the kernel populates it with the value of the 1259 * current GTT offset of the object, for future presumed_offset writes. 1260 * 1261 * See struct drm_i915_gem_create_ext for the rules when dealing with 1262 * alignment restrictions with I915_MEMORY_CLASS_DEVICE, on devices with 1263 * minimum page sizes, like DG2. 1264 */ 1265 __u64 offset; 1266 1267 #define EXEC_OBJECT_NEEDS_FENCE (1<<0) 1268 #define EXEC_OBJECT_NEEDS_GTT (1<<1) 1269 #define EXEC_OBJECT_WRITE (1<<2) 1270 #define EXEC_OBJECT_SUPPORTS_48B_ADDRESS (1<<3) 1271 #define EXEC_OBJECT_PINNED (1<<4) 1272 #define EXEC_OBJECT_PAD_TO_SIZE (1<<5) 1273 /* The kernel implicitly tracks GPU activity on all GEM objects, and 1274 * synchronises operations with outstanding rendering. This includes 1275 * rendering on other devices if exported via dma-buf. However, sometimes 1276 * this tracking is too coarse and the user knows better. For example, 1277 * if the object is split into non-overlapping ranges shared between different 1278 * clients or engines (i.e. suballocating objects), the implicit tracking 1279 * by kernel assumes that each operation affects the whole object rather 1280 * than an individual range, causing needless synchronisation between clients. 1281 * The kernel will also forgo any CPU cache flushes prior to rendering from 1282 * the object as the client is expected to be also handling such domain 1283 * tracking. 1284 * 1285 * The kernel maintains the implicit tracking in order to manage resources 1286 * used by the GPU - this flag only disables the synchronisation prior to 1287 * rendering with this object in this execbuf. 1288 * 1289 * Opting out of implicit synhronisation requires the user to do its own 1290 * explicit tracking to avoid rendering corruption. See, for example, 1291 * I915_PARAM_HAS_EXEC_FENCE to order execbufs and execute them asynchronously. 1292 */ 1293 #define EXEC_OBJECT_ASYNC (1<<6) 1294 /* Request that the contents of this execobject be copied into the error 1295 * state upon a GPU hang involving this batch for post-mortem debugging. 1296 * These buffers are recorded in no particular order as "user" in 1297 * /sys/class/drm/cardN/error. Query I915_PARAM_HAS_EXEC_CAPTURE to see 1298 * if the kernel supports this flag. 1299 */ 1300 #define EXEC_OBJECT_CAPTURE (1<<7) 1301 /* All remaining bits are MBZ and RESERVED FOR FUTURE USE */ 1302 #define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_CAPTURE<<1) 1303 __u64 flags; 1304 1305 union { 1306 __u64 rsvd1; 1307 __u64 pad_to_size; 1308 }; 1309 __u64 rsvd2; 1310 }; 1311 1312 /** 1313 * struct drm_i915_gem_exec_fence - An input or output fence for the execbuf 1314 * ioctl. 1315 * 1316 * The request will wait for input fence to signal before submission. 1317 * 1318 * The returned output fence will be signaled after the completion of the 1319 * request. 1320 */ 1321 struct drm_i915_gem_exec_fence { 1322 /** @handle: User's handle for a drm_syncobj to wait on or signal. */ 1323 __u32 handle; 1324 1325 /** 1326 * @flags: Supported flags are: 1327 * 1328 * I915_EXEC_FENCE_WAIT: 1329 * Wait for the input fence before request submission. 1330 * 1331 * I915_EXEC_FENCE_SIGNAL: 1332 * Return request completion fence as output 1333 */ 1334 __u32 flags; 1335 #define I915_EXEC_FENCE_WAIT (1<<0) 1336 #define I915_EXEC_FENCE_SIGNAL (1<<1) 1337 #define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1)) 1338 }; 1339 1340 /** 1341 * struct drm_i915_gem_execbuffer_ext_timeline_fences - Timeline fences 1342 * for execbuf ioctl. 1343 * 1344 * This structure describes an array of drm_syncobj and associated points for 1345 * timeline variants of drm_syncobj. It is invalid to append this structure to 1346 * the execbuf if I915_EXEC_FENCE_ARRAY is set. 1347 */ 1348 struct drm_i915_gem_execbuffer_ext_timeline_fences { 1349 #define DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES 0 1350 /** @base: Extension link. See struct i915_user_extension. */ 1351 struct i915_user_extension base; 1352 1353 /** 1354 * @fence_count: Number of elements in the @handles_ptr & @value_ptr 1355 * arrays. 1356 */ 1357 __u64 fence_count; 1358 1359 /** 1360 * @handles_ptr: Pointer to an array of struct drm_i915_gem_exec_fence 1361 * of length @fence_count. 1362 */ 1363 __u64 handles_ptr; 1364 1365 /** 1366 * @values_ptr: Pointer to an array of u64 values of length 1367 * @fence_count. 1368 * Values must be 0 for a binary drm_syncobj. A Value of 0 for a 1369 * timeline drm_syncobj is invalid as it turns a drm_syncobj into a 1370 * binary one. 1371 */ 1372 __u64 values_ptr; 1373 }; 1374 1375 /** 1376 * struct drm_i915_gem_execbuffer2 - Structure for DRM_I915_GEM_EXECBUFFER2 1377 * ioctl. 1378 */ 1379 struct drm_i915_gem_execbuffer2 { 1380 /** @buffers_ptr: Pointer to a list of gem_exec_object2 structs */ 1381 __u64 buffers_ptr; 1382 1383 /** @buffer_count: Number of elements in @buffers_ptr array */ 1384 __u32 buffer_count; 1385 1386 /** 1387 * @batch_start_offset: Offset in the batchbuffer to start execution 1388 * from. 1389 */ 1390 __u32 batch_start_offset; 1391 1392 /** 1393 * @batch_len: Length in bytes of the batch buffer, starting from the 1394 * @batch_start_offset. If 0, length is assumed to be the batch buffer 1395 * object size. 1396 */ 1397 __u32 batch_len; 1398 1399 /** @DR1: deprecated */ 1400 __u32 DR1; 1401 1402 /** @DR4: deprecated */ 1403 __u32 DR4; 1404 1405 /** @num_cliprects: See @cliprects_ptr */ 1406 __u32 num_cliprects; 1407 1408 /** 1409 * @cliprects_ptr: Kernel clipping was a DRI1 misfeature. 1410 * 1411 * It is invalid to use this field if I915_EXEC_FENCE_ARRAY or 1412 * I915_EXEC_USE_EXTENSIONS flags are not set. 1413 * 1414 * If I915_EXEC_FENCE_ARRAY is set, then this is a pointer to an array 1415 * of &drm_i915_gem_exec_fence and @num_cliprects is the length of the 1416 * array. 1417 * 1418 * If I915_EXEC_USE_EXTENSIONS is set, then this is a pointer to a 1419 * single &i915_user_extension and num_cliprects is 0. 1420 */ 1421 __u64 cliprects_ptr; 1422 1423 /** @flags: Execbuf flags */ 1424 __u64 flags; 1425 #define I915_EXEC_RING_MASK (0x3f) 1426 #define I915_EXEC_DEFAULT (0<<0) 1427 #define I915_EXEC_RENDER (1<<0) 1428 #define I915_EXEC_BSD (2<<0) 1429 #define I915_EXEC_BLT (3<<0) 1430 #define I915_EXEC_VEBOX (4<<0) 1431 1432 /* Used for switching the constants addressing mode on gen4+ RENDER ring. 1433 * Gen6+ only supports relative addressing to dynamic state (default) and 1434 * absolute addressing. 1435 * 1436 * These flags are ignored for the BSD and BLT rings. 1437 */ 1438 #define I915_EXEC_CONSTANTS_MASK (3<<6) 1439 #define I915_EXEC_CONSTANTS_REL_GENERAL (0<<6) /* default */ 1440 #define I915_EXEC_CONSTANTS_ABSOLUTE (1<<6) 1441 #define I915_EXEC_CONSTANTS_REL_SURFACE (2<<6) /* gen4/5 only */ 1442 1443 /** Resets the SO write offset registers for transform feedback on gen7. */ 1444 #define I915_EXEC_GEN7_SOL_RESET (1<<8) 1445 1446 /** Request a privileged ("secure") batch buffer. Note only available for 1447 * DRM_ROOT_ONLY | DRM_MASTER processes. 1448 */ 1449 #define I915_EXEC_SECURE (1<<9) 1450 1451 /** Inform the kernel that the batch is and will always be pinned. This 1452 * negates the requirement for a workaround to be performed to avoid 1453 * an incoherent CS (such as can be found on 830/845). If this flag is 1454 * not passed, the kernel will endeavour to make sure the batch is 1455 * coherent with the CS before execution. If this flag is passed, 1456 * userspace assumes the responsibility for ensuring the same. 1457 */ 1458 #define I915_EXEC_IS_PINNED (1<<10) 1459 1460 /** Provide a hint to the kernel that the command stream and auxiliary 1461 * state buffers already holds the correct presumed addresses and so the 1462 * relocation process may be skipped if no buffers need to be moved in 1463 * preparation for the execbuffer. 1464 */ 1465 #define I915_EXEC_NO_RELOC (1<<11) 1466 1467 /** Use the reloc.handle as an index into the exec object array rather 1468 * than as the per-file handle. 1469 */ 1470 #define I915_EXEC_HANDLE_LUT (1<<12) 1471 1472 /** Used for switching BSD rings on the platforms with two BSD rings */ 1473 #define I915_EXEC_BSD_SHIFT (13) 1474 #define I915_EXEC_BSD_MASK (3 << I915_EXEC_BSD_SHIFT) 1475 /* default ping-pong mode */ 1476 #define I915_EXEC_BSD_DEFAULT (0 << I915_EXEC_BSD_SHIFT) 1477 #define I915_EXEC_BSD_RING1 (1 << I915_EXEC_BSD_SHIFT) 1478 #define I915_EXEC_BSD_RING2 (2 << I915_EXEC_BSD_SHIFT) 1479 1480 /** Tell the kernel that the batchbuffer is processed by 1481 * the resource streamer. 1482 */ 1483 #define I915_EXEC_RESOURCE_STREAMER (1<<15) 1484 1485 /* Setting I915_EXEC_FENCE_IN implies that lower_32_bits(rsvd2) represent 1486 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1487 * the batch. 1488 * 1489 * Returns -EINVAL if the sync_file fd cannot be found. 1490 */ 1491 #define I915_EXEC_FENCE_IN (1<<16) 1492 1493 /* Setting I915_EXEC_FENCE_OUT causes the ioctl to return a sync_file fd 1494 * in the upper_32_bits(rsvd2) upon success. Ownership of the fd is given 1495 * to the caller, and it should be close() after use. (The fd is a regular 1496 * file descriptor and will be cleaned up on process termination. It holds 1497 * a reference to the request, but nothing else.) 1498 * 1499 * The sync_file fd can be combined with other sync_file and passed either 1500 * to execbuf using I915_EXEC_FENCE_IN, to atomic KMS ioctls (so that a flip 1501 * will only occur after this request completes), or to other devices. 1502 * 1503 * Using I915_EXEC_FENCE_OUT requires use of 1504 * DRM_IOCTL_I915_GEM_EXECBUFFER2_WR ioctl so that the result is written 1505 * back to userspace. Failure to do so will cause the out-fence to always 1506 * be reported as zero, and the real fence fd to be leaked. 1507 */ 1508 #define I915_EXEC_FENCE_OUT (1<<17) 1509 1510 /* 1511 * Traditionally the execbuf ioctl has only considered the final element in 1512 * the execobject[] to be the executable batch. Often though, the client 1513 * will known the batch object prior to construction and being able to place 1514 * it into the execobject[] array first can simplify the relocation tracking. 1515 * Setting I915_EXEC_BATCH_FIRST tells execbuf to use element 0 of the 1516 * execobject[] as the * batch instead (the default is to use the last 1517 * element). 1518 */ 1519 #define I915_EXEC_BATCH_FIRST (1<<18) 1520 1521 /* Setting I915_FENCE_ARRAY implies that num_cliprects and cliprects_ptr 1522 * define an array of i915_gem_exec_fence structures which specify a set of 1523 * dma fences to wait upon or signal. 1524 */ 1525 #define I915_EXEC_FENCE_ARRAY (1<<19) 1526 1527 /* 1528 * Setting I915_EXEC_FENCE_SUBMIT implies that lower_32_bits(rsvd2) represent 1529 * a sync_file fd to wait upon (in a nonblocking manner) prior to executing 1530 * the batch. 1531 * 1532 * Returns -EINVAL if the sync_file fd cannot be found. 1533 */ 1534 #define I915_EXEC_FENCE_SUBMIT (1 << 20) 1535 1536 /* 1537 * Setting I915_EXEC_USE_EXTENSIONS implies that 1538 * drm_i915_gem_execbuffer2.cliprects_ptr is treated as a pointer to an linked 1539 * list of i915_user_extension. Each i915_user_extension node is the base of a 1540 * larger structure. The list of supported structures are listed in the 1541 * drm_i915_gem_execbuffer_ext enum. 1542 */ 1543 #define I915_EXEC_USE_EXTENSIONS (1 << 21) 1544 #define __I915_EXEC_UNKNOWN_FLAGS (-(I915_EXEC_USE_EXTENSIONS << 1)) 1545 1546 /** @rsvd1: Context id */ 1547 __u64 rsvd1; 1548 1549 /** 1550 * @rsvd2: in and out sync_file file descriptors. 1551 * 1552 * When I915_EXEC_FENCE_IN or I915_EXEC_FENCE_SUBMIT flag is set, the 1553 * lower 32 bits of this field will have the in sync_file fd (input). 1554 * 1555 * When I915_EXEC_FENCE_OUT flag is set, the upper 32 bits of this 1556 * field will have the out sync_file fd (output). 1557 */ 1558 __u64 rsvd2; 1559 }; 1560 1561 #define I915_EXEC_CONTEXT_ID_MASK (0xffffffff) 1562 #define i915_execbuffer2_set_context_id(eb2, context) \ 1563 (eb2).rsvd1 = context & I915_EXEC_CONTEXT_ID_MASK 1564 #define i915_execbuffer2_get_context_id(eb2) \ 1565 ((eb2).rsvd1 & I915_EXEC_CONTEXT_ID_MASK) 1566 1567 struct drm_i915_gem_pin { 1568 /** Handle of the buffer to be pinned. */ 1569 __u32 handle; 1570 __u32 pad; 1571 1572 /** alignment required within the aperture */ 1573 __u64 alignment; 1574 1575 /** Returned GTT offset of the buffer. */ 1576 __u64 offset; 1577 }; 1578 1579 struct drm_i915_gem_unpin { 1580 /** Handle of the buffer to be unpinned. */ 1581 __u32 handle; 1582 __u32 pad; 1583 }; 1584 1585 struct drm_i915_gem_busy { 1586 /** Handle of the buffer to check for busy */ 1587 __u32 handle; 1588 1589 /** Return busy status 1590 * 1591 * A return of 0 implies that the object is idle (after 1592 * having flushed any pending activity), and a non-zero return that 1593 * the object is still in-flight on the GPU. (The GPU has not yet 1594 * signaled completion for all pending requests that reference the 1595 * object.) An object is guaranteed to become idle eventually (so 1596 * long as no new GPU commands are executed upon it). Due to the 1597 * asynchronous nature of the hardware, an object reported 1598 * as busy may become idle before the ioctl is completed. 1599 * 1600 * Furthermore, if the object is busy, which engine is busy is only 1601 * provided as a guide and only indirectly by reporting its class 1602 * (there may be more than one engine in each class). There are race 1603 * conditions which prevent the report of which engines are busy from 1604 * being always accurate. However, the converse is not true. If the 1605 * object is idle, the result of the ioctl, that all engines are idle, 1606 * is accurate. 1607 * 1608 * The returned dword is split into two fields to indicate both 1609 * the engine classess on which the object is being read, and the 1610 * engine class on which it is currently being written (if any). 1611 * 1612 * The low word (bits 0:15) indicate if the object is being written 1613 * to by any engine (there can only be one, as the GEM implicit 1614 * synchronisation rules force writes to be serialised). Only the 1615 * engine class (offset by 1, I915_ENGINE_CLASS_RENDER is reported as 1616 * 1 not 0 etc) for the last write is reported. 1617 * 1618 * The high word (bits 16:31) are a bitmask of which engines classes 1619 * are currently reading from the object. Multiple engines may be 1620 * reading from the object simultaneously. 1621 * 1622 * The value of each engine class is the same as specified in the 1623 * I915_CONTEXT_PARAM_ENGINES context parameter and via perf, i.e. 1624 * I915_ENGINE_CLASS_RENDER, I915_ENGINE_CLASS_COPY, etc. 1625 * Some hardware may have parallel execution engines, e.g. multiple 1626 * media engines, which are mapped to the same class identifier and so 1627 * are not separately reported for busyness. 1628 * 1629 * Caveat emptor: 1630 * Only the boolean result of this query is reliable; that is whether 1631 * the object is idle or busy. The report of which engines are busy 1632 * should be only used as a heuristic. 1633 */ 1634 __u32 busy; 1635 }; 1636 1637 /** 1638 * struct drm_i915_gem_caching - Set or get the caching for given object 1639 * handle. 1640 * 1641 * Allow userspace to control the GTT caching bits for a given object when the 1642 * object is later mapped through the ppGTT(or GGTT on older platforms lacking 1643 * ppGTT support, or if the object is used for scanout). Note that this might 1644 * require unbinding the object from the GTT first, if its current caching value 1645 * doesn't match. 1646 * 1647 * Note that this all changes on discrete platforms, starting from DG1, the 1648 * set/get caching is no longer supported, and is now rejected. Instead the CPU 1649 * caching attributes(WB vs WC) will become an immutable creation time property 1650 * for the object, along with the GTT caching level. For now we don't expose any 1651 * new uAPI for this, instead on DG1 this is all implicit, although this largely 1652 * shouldn't matter since DG1 is coherent by default(without any way of 1653 * controlling it). 1654 * 1655 * Implicit caching rules, starting from DG1: 1656 * 1657 * - If any of the object placements (see &drm_i915_gem_create_ext_memory_regions) 1658 * contain I915_MEMORY_CLASS_DEVICE then the object will be allocated and 1659 * mapped as write-combined only. 1660 * 1661 * - Everything else is always allocated and mapped as write-back, with the 1662 * guarantee that everything is also coherent with the GPU. 1663 * 1664 * Note that this is likely to change in the future again, where we might need 1665 * more flexibility on future devices, so making this all explicit as part of a 1666 * new &drm_i915_gem_create_ext extension is probable. 1667 * 1668 * Side note: Part of the reason for this is that changing the at-allocation-time CPU 1669 * caching attributes for the pages might be required(and is expensive) if we 1670 * need to then CPU map the pages later with different caching attributes. This 1671 * inconsistent caching behaviour, while supported on x86, is not universally 1672 * supported on other architectures. So for simplicity we opt for setting 1673 * everything at creation time, whilst also making it immutable, on discrete 1674 * platforms. 1675 */ 1676 struct drm_i915_gem_caching { 1677 /** 1678 * @handle: Handle of the buffer to set/get the caching level. 1679 */ 1680 __u32 handle; 1681 1682 /** 1683 * @caching: The GTT caching level to apply or possible return value. 1684 * 1685 * The supported @caching values: 1686 * 1687 * I915_CACHING_NONE: 1688 * 1689 * GPU access is not coherent with CPU caches. Default for machines 1690 * without an LLC. This means manual flushing might be needed, if we 1691 * want GPU access to be coherent. 1692 * 1693 * I915_CACHING_CACHED: 1694 * 1695 * GPU access is coherent with CPU caches and furthermore the data is 1696 * cached in last-level caches shared between CPU cores and the GPU GT. 1697 * 1698 * I915_CACHING_DISPLAY: 1699 * 1700 * Special GPU caching mode which is coherent with the scanout engines. 1701 * Transparently falls back to I915_CACHING_NONE on platforms where no 1702 * special cache mode (like write-through or gfdt flushing) is 1703 * available. The kernel automatically sets this mode when using a 1704 * buffer as a scanout target. Userspace can manually set this mode to 1705 * avoid a costly stall and clflush in the hotpath of drawing the first 1706 * frame. 1707 */ 1708 #define I915_CACHING_NONE 0 1709 #define I915_CACHING_CACHED 1 1710 #define I915_CACHING_DISPLAY 2 1711 __u32 caching; 1712 }; 1713 1714 #define I915_TILING_NONE 0 1715 #define I915_TILING_X 1 1716 #define I915_TILING_Y 2 1717 /* 1718 * Do not add new tiling types here. The I915_TILING_* values are for 1719 * de-tiling fence registers that no longer exist on modern platforms. Although 1720 * the hardware may support new types of tiling in general (e.g., Tile4), we 1721 * do not need to add them to the uapi that is specific to now-defunct ioctls. 1722 */ 1723 #define I915_TILING_LAST I915_TILING_Y 1724 1725 #define I915_BIT_6_SWIZZLE_NONE 0 1726 #define I915_BIT_6_SWIZZLE_9 1 1727 #define I915_BIT_6_SWIZZLE_9_10 2 1728 #define I915_BIT_6_SWIZZLE_9_11 3 1729 #define I915_BIT_6_SWIZZLE_9_10_11 4 1730 /* Not seen by userland */ 1731 #define I915_BIT_6_SWIZZLE_UNKNOWN 5 1732 /* Seen by userland. */ 1733 #define I915_BIT_6_SWIZZLE_9_17 6 1734 #define I915_BIT_6_SWIZZLE_9_10_17 7 1735 1736 struct drm_i915_gem_set_tiling { 1737 /** Handle of the buffer to have its tiling state updated */ 1738 __u32 handle; 1739 1740 /** 1741 * Tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1742 * I915_TILING_Y). 1743 * 1744 * This value is to be set on request, and will be updated by the 1745 * kernel on successful return with the actual chosen tiling layout. 1746 * 1747 * The tiling mode may be demoted to I915_TILING_NONE when the system 1748 * has bit 6 swizzling that can't be managed correctly by GEM. 1749 * 1750 * Buffer contents become undefined when changing tiling_mode. 1751 */ 1752 __u32 tiling_mode; 1753 1754 /** 1755 * Stride in bytes for the object when in I915_TILING_X or 1756 * I915_TILING_Y. 1757 */ 1758 __u32 stride; 1759 1760 /** 1761 * Returned address bit 6 swizzling required for CPU access through 1762 * mmap mapping. 1763 */ 1764 __u32 swizzle_mode; 1765 }; 1766 1767 struct drm_i915_gem_get_tiling { 1768 /** Handle of the buffer to get tiling state for. */ 1769 __u32 handle; 1770 1771 /** 1772 * Current tiling mode for the object (I915_TILING_NONE, I915_TILING_X, 1773 * I915_TILING_Y). 1774 */ 1775 __u32 tiling_mode; 1776 1777 /** 1778 * Returned address bit 6 swizzling required for CPU access through 1779 * mmap mapping. 1780 */ 1781 __u32 swizzle_mode; 1782 1783 /** 1784 * Returned address bit 6 swizzling required for CPU access through 1785 * mmap mapping whilst bound. 1786 */ 1787 __u32 phys_swizzle_mode; 1788 }; 1789 1790 struct drm_i915_gem_get_aperture { 1791 /** Total size of the aperture used by i915_gem_execbuffer, in bytes */ 1792 __u64 aper_size; 1793 1794 /** 1795 * Available space in the aperture used by i915_gem_execbuffer, in 1796 * bytes 1797 */ 1798 __u64 aper_available_size; 1799 }; 1800 1801 struct drm_i915_get_pipe_from_crtc_id { 1802 /** ID of CRTC being requested **/ 1803 __u32 crtc_id; 1804 1805 /** pipe of requested CRTC **/ 1806 __u32 pipe; 1807 }; 1808 1809 #define I915_MADV_WILLNEED 0 1810 #define I915_MADV_DONTNEED 1 1811 #define __I915_MADV_PURGED 2 /* internal state */ 1812 1813 struct drm_i915_gem_madvise { 1814 /** Handle of the buffer to change the backing store advice */ 1815 __u32 handle; 1816 1817 /* Advice: either the buffer will be needed again in the near future, 1818 * or wont be and could be discarded under memory pressure. 1819 */ 1820 __u32 madv; 1821 1822 /** Whether the backing store still exists. */ 1823 __u32 retained; 1824 }; 1825 1826 /* flags */ 1827 #define I915_OVERLAY_TYPE_MASK 0xff 1828 #define I915_OVERLAY_YUV_PLANAR 0x01 1829 #define I915_OVERLAY_YUV_PACKED 0x02 1830 #define I915_OVERLAY_RGB 0x03 1831 1832 #define I915_OVERLAY_DEPTH_MASK 0xff00 1833 #define I915_OVERLAY_RGB24 0x1000 1834 #define I915_OVERLAY_RGB16 0x2000 1835 #define I915_OVERLAY_RGB15 0x3000 1836 #define I915_OVERLAY_YUV422 0x0100 1837 #define I915_OVERLAY_YUV411 0x0200 1838 #define I915_OVERLAY_YUV420 0x0300 1839 #define I915_OVERLAY_YUV410 0x0400 1840 1841 #define I915_OVERLAY_SWAP_MASK 0xff0000 1842 #define I915_OVERLAY_NO_SWAP 0x000000 1843 #define I915_OVERLAY_UV_SWAP 0x010000 1844 #define I915_OVERLAY_Y_SWAP 0x020000 1845 #define I915_OVERLAY_Y_AND_UV_SWAP 0x030000 1846 1847 #define I915_OVERLAY_FLAGS_MASK 0xff000000 1848 #define I915_OVERLAY_ENABLE 0x01000000 1849 1850 struct drm_intel_overlay_put_image { 1851 /* various flags and src format description */ 1852 __u32 flags; 1853 /* source picture description */ 1854 __u32 bo_handle; 1855 /* stride values and offsets are in bytes, buffer relative */ 1856 __u16 stride_Y; /* stride for packed formats */ 1857 __u16 stride_UV; 1858 __u32 offset_Y; /* offset for packet formats */ 1859 __u32 offset_U; 1860 __u32 offset_V; 1861 /* in pixels */ 1862 __u16 src_width; 1863 __u16 src_height; 1864 /* to compensate the scaling factors for partially covered surfaces */ 1865 __u16 src_scan_width; 1866 __u16 src_scan_height; 1867 /* output crtc description */ 1868 __u32 crtc_id; 1869 __u16 dst_x; 1870 __u16 dst_y; 1871 __u16 dst_width; 1872 __u16 dst_height; 1873 }; 1874 1875 /* flags */ 1876 #define I915_OVERLAY_UPDATE_ATTRS (1<<0) 1877 #define I915_OVERLAY_UPDATE_GAMMA (1<<1) 1878 #define I915_OVERLAY_DISABLE_DEST_COLORKEY (1<<2) 1879 struct drm_intel_overlay_attrs { 1880 __u32 flags; 1881 __u32 color_key; 1882 __s32 brightness; 1883 __u32 contrast; 1884 __u32 saturation; 1885 __u32 gamma0; 1886 __u32 gamma1; 1887 __u32 gamma2; 1888 __u32 gamma3; 1889 __u32 gamma4; 1890 __u32 gamma5; 1891 }; 1892 1893 /* 1894 * Intel sprite handling 1895 * 1896 * Color keying works with a min/mask/max tuple. Both source and destination 1897 * color keying is allowed. 1898 * 1899 * Source keying: 1900 * Sprite pixels within the min & max values, masked against the color channels 1901 * specified in the mask field, will be transparent. All other pixels will 1902 * be displayed on top of the primary plane. For RGB surfaces, only the min 1903 * and mask fields will be used; ranged compares are not allowed. 1904 * 1905 * Destination keying: 1906 * Primary plane pixels that match the min value, masked against the color 1907 * channels specified in the mask field, will be replaced by corresponding 1908 * pixels from the sprite plane. 1909 * 1910 * Note that source & destination keying are exclusive; only one can be 1911 * active on a given plane. 1912 */ 1913 1914 #define I915_SET_COLORKEY_NONE (1<<0) /* Deprecated. Instead set 1915 * flags==0 to disable colorkeying. 1916 */ 1917 #define I915_SET_COLORKEY_DESTINATION (1<<1) 1918 #define I915_SET_COLORKEY_SOURCE (1<<2) 1919 struct drm_intel_sprite_colorkey { 1920 __u32 plane_id; 1921 __u32 min_value; 1922 __u32 channel_mask; 1923 __u32 max_value; 1924 __u32 flags; 1925 }; 1926 1927 struct drm_i915_gem_wait { 1928 /** Handle of BO we shall wait on */ 1929 __u32 bo_handle; 1930 __u32 flags; 1931 /** Number of nanoseconds to wait, Returns time remaining. */ 1932 __s64 timeout_ns; 1933 }; 1934 1935 struct drm_i915_gem_context_create { 1936 __u32 ctx_id; /* output: id of new context*/ 1937 __u32 pad; 1938 }; 1939 1940 /** 1941 * struct drm_i915_gem_context_create_ext - Structure for creating contexts. 1942 */ 1943 struct drm_i915_gem_context_create_ext { 1944 /** @ctx_id: Id of the created context (output) */ 1945 __u32 ctx_id; 1946 1947 /** 1948 * @flags: Supported flags are: 1949 * 1950 * I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS: 1951 * 1952 * Extensions may be appended to this structure and driver must check 1953 * for those. See @extensions. 1954 * 1955 * I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE 1956 * 1957 * Created context will have single timeline. 1958 */ 1959 __u32 flags; 1960 #define I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS (1u << 0) 1961 #define I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE (1u << 1) 1962 #define I915_CONTEXT_CREATE_FLAGS_UNKNOWN \ 1963 (-(I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE << 1)) 1964 1965 /** 1966 * @extensions: Zero-terminated chain of extensions. 1967 * 1968 * I915_CONTEXT_CREATE_EXT_SETPARAM: 1969 * Context parameter to set or query during context creation. 1970 * See struct drm_i915_gem_context_create_ext_setparam. 1971 * 1972 * I915_CONTEXT_CREATE_EXT_CLONE: 1973 * This extension has been removed. On the off chance someone somewhere 1974 * has attempted to use it, never re-use this extension number. 1975 */ 1976 __u64 extensions; 1977 #define I915_CONTEXT_CREATE_EXT_SETPARAM 0 1978 #define I915_CONTEXT_CREATE_EXT_CLONE 1 1979 }; 1980 1981 /** 1982 * struct drm_i915_gem_context_param - Context parameter to set or query. 1983 */ 1984 struct drm_i915_gem_context_param { 1985 /** @ctx_id: Context id */ 1986 __u32 ctx_id; 1987 1988 /** @size: Size of the parameter @value */ 1989 __u32 size; 1990 1991 /** @param: Parameter to set or query */ 1992 __u64 param; 1993 #define I915_CONTEXT_PARAM_BAN_PERIOD 0x1 1994 /* I915_CONTEXT_PARAM_NO_ZEROMAP has been removed. On the off chance 1995 * someone somewhere has attempted to use it, never re-use this context 1996 * param number. 1997 */ 1998 #define I915_CONTEXT_PARAM_NO_ZEROMAP 0x2 1999 #define I915_CONTEXT_PARAM_GTT_SIZE 0x3 2000 #define I915_CONTEXT_PARAM_NO_ERROR_CAPTURE 0x4 2001 #define I915_CONTEXT_PARAM_BANNABLE 0x5 2002 #define I915_CONTEXT_PARAM_PRIORITY 0x6 2003 #define I915_CONTEXT_MAX_USER_PRIORITY 1023 /* inclusive */ 2004 #define I915_CONTEXT_DEFAULT_PRIORITY 0 2005 #define I915_CONTEXT_MIN_USER_PRIORITY -1023 /* inclusive */ 2006 /* 2007 * When using the following param, value should be a pointer to 2008 * drm_i915_gem_context_param_sseu. 2009 */ 2010 #define I915_CONTEXT_PARAM_SSEU 0x7 2011 2012 /* 2013 * Not all clients may want to attempt automatic recover of a context after 2014 * a hang (for example, some clients may only submit very small incremental 2015 * batches relying on known logical state of previous batches which will never 2016 * recover correctly and each attempt will hang), and so would prefer that 2017 * the context is forever banned instead. 2018 * 2019 * If set to false (0), after a reset, subsequent (and in flight) rendering 2020 * from this context is discarded, and the client will need to create a new 2021 * context to use instead. 2022 * 2023 * If set to true (1), the kernel will automatically attempt to recover the 2024 * context by skipping the hanging batch and executing the next batch starting 2025 * from the default context state (discarding the incomplete logical context 2026 * state lost due to the reset). 2027 * 2028 * On creation, all new contexts are marked as recoverable. 2029 */ 2030 #define I915_CONTEXT_PARAM_RECOVERABLE 0x8 2031 2032 /* 2033 * The id of the associated virtual memory address space (ppGTT) of 2034 * this context. Can be retrieved and passed to another context 2035 * (on the same fd) for both to use the same ppGTT and so share 2036 * address layouts, and avoid reloading the page tables on context 2037 * switches between themselves. 2038 * 2039 * See DRM_I915_GEM_VM_CREATE and DRM_I915_GEM_VM_DESTROY. 2040 */ 2041 #define I915_CONTEXT_PARAM_VM 0x9 2042 2043 /* 2044 * I915_CONTEXT_PARAM_ENGINES: 2045 * 2046 * Bind this context to operate on this subset of available engines. Henceforth, 2047 * the I915_EXEC_RING selector for DRM_IOCTL_I915_GEM_EXECBUFFER2 operates as 2048 * an index into this array of engines; I915_EXEC_DEFAULT selecting engine[0] 2049 * and upwards. Slots 0...N are filled in using the specified (class, instance). 2050 * Use 2051 * engine_class: I915_ENGINE_CLASS_INVALID, 2052 * engine_instance: I915_ENGINE_CLASS_INVALID_NONE 2053 * to specify a gap in the array that can be filled in later, e.g. by a 2054 * virtual engine used for load balancing. 2055 * 2056 * Setting the number of engines bound to the context to 0, by passing a zero 2057 * sized argument, will revert back to default settings. 2058 * 2059 * See struct i915_context_param_engines. 2060 * 2061 * Extensions: 2062 * i915_context_engines_load_balance (I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE) 2063 * i915_context_engines_bond (I915_CONTEXT_ENGINES_EXT_BOND) 2064 * i915_context_engines_parallel_submit (I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT) 2065 */ 2066 #define I915_CONTEXT_PARAM_ENGINES 0xa 2067 2068 /* 2069 * I915_CONTEXT_PARAM_PERSISTENCE: 2070 * 2071 * Allow the context and active rendering to survive the process until 2072 * completion. Persistence allows fire-and-forget clients to queue up a 2073 * bunch of work, hand the output over to a display server and then quit. 2074 * If the context is marked as not persistent, upon closing (either via 2075 * an explicit DRM_I915_GEM_CONTEXT_DESTROY or implicitly from file closure 2076 * or process termination), the context and any outstanding requests will be 2077 * cancelled (and exported fences for cancelled requests marked as -EIO). 2078 * 2079 * By default, new contexts allow persistence. 2080 */ 2081 #define I915_CONTEXT_PARAM_PERSISTENCE 0xb 2082 2083 /* This API has been removed. On the off chance someone somewhere has 2084 * attempted to use it, never re-use this context param number. 2085 */ 2086 #define I915_CONTEXT_PARAM_RINGSIZE 0xc 2087 2088 /* 2089 * I915_CONTEXT_PARAM_PROTECTED_CONTENT: 2090 * 2091 * Mark that the context makes use of protected content, which will result 2092 * in the context being invalidated when the protected content session is. 2093 * Given that the protected content session is killed on suspend, the device 2094 * is kept awake for the lifetime of a protected context, so the user should 2095 * make sure to dispose of them once done. 2096 * This flag can only be set at context creation time and, when set to true, 2097 * must be preceded by an explicit setting of I915_CONTEXT_PARAM_RECOVERABLE 2098 * to false. This flag can't be set to true in conjunction with setting the 2099 * I915_CONTEXT_PARAM_BANNABLE flag to false. Creation example: 2100 * 2101 * .. code-block:: C 2102 * 2103 * struct drm_i915_gem_context_create_ext_setparam p_protected = { 2104 * .base = { 2105 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2106 * }, 2107 * .param = { 2108 * .param = I915_CONTEXT_PARAM_PROTECTED_CONTENT, 2109 * .value = 1, 2110 * } 2111 * }; 2112 * struct drm_i915_gem_context_create_ext_setparam p_norecover = { 2113 * .base = { 2114 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2115 * .next_extension = to_user_pointer(&p_protected), 2116 * }, 2117 * .param = { 2118 * .param = I915_CONTEXT_PARAM_RECOVERABLE, 2119 * .value = 0, 2120 * } 2121 * }; 2122 * struct drm_i915_gem_context_create_ext create = { 2123 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2124 * .extensions = to_user_pointer(&p_norecover); 2125 * }; 2126 * 2127 * ctx_id = gem_context_create_ext(drm_fd, &create); 2128 * 2129 * In addition to the normal failure cases, setting this flag during context 2130 * creation can result in the following errors: 2131 * 2132 * -ENODEV: feature not available 2133 * -EPERM: trying to mark a recoverable or not bannable context as protected 2134 * -ENXIO: A dependency such as a component driver or firmware is not yet 2135 * loaded so user space may need to attempt again. Depending on the 2136 * device, this error may be reported if protected context creation is 2137 * attempted very early after kernel start because the internal timeout 2138 * waiting for such dependencies is not guaranteed to be larger than 2139 * required (numbers differ depending on system and kernel config): 2140 * - ADL/RPL: dependencies may take up to 3 seconds from kernel start 2141 * while context creation internal timeout is 250 milisecs 2142 * - MTL: dependencies may take up to 8 seconds from kernel start 2143 * while context creation internal timeout is 250 milisecs 2144 * NOTE: such dependencies happen once, so a subsequent call to create a 2145 * protected context after a prior successful call will not experience 2146 * such timeouts and will not return -ENXIO (unless the driver is reloaded, 2147 * or, depending on the device, resumes from a suspended state). 2148 * -EIO: The firmware did not succeed in creating the protected context. 2149 */ 2150 #define I915_CONTEXT_PARAM_PROTECTED_CONTENT 0xd 2151 /* Must be kept compact -- no holes and well documented */ 2152 2153 /** @value: Context parameter value to be set or queried */ 2154 __u64 value; 2155 }; 2156 2157 /* 2158 * Context SSEU programming 2159 * 2160 * It may be necessary for either functional or performance reason to configure 2161 * a context to run with a reduced number of SSEU (where SSEU stands for Slice/ 2162 * Sub-slice/EU). 2163 * 2164 * This is done by configuring SSEU configuration using the below 2165 * @struct drm_i915_gem_context_param_sseu for every supported engine which 2166 * userspace intends to use. 2167 * 2168 * Not all GPUs or engines support this functionality in which case an error 2169 * code -ENODEV will be returned. 2170 * 2171 * Also, flexibility of possible SSEU configuration permutations varies between 2172 * GPU generations and software imposed limitations. Requesting such a 2173 * combination will return an error code of -EINVAL. 2174 * 2175 * NOTE: When perf/OA is active the context's SSEU configuration is ignored in 2176 * favour of a single global setting. 2177 */ 2178 struct drm_i915_gem_context_param_sseu { 2179 /* 2180 * Engine class & instance to be configured or queried. 2181 */ 2182 struct i915_engine_class_instance engine; 2183 2184 /* 2185 * Unknown flags must be cleared to zero. 2186 */ 2187 __u32 flags; 2188 #define I915_CONTEXT_SSEU_FLAG_ENGINE_INDEX (1u << 0) 2189 2190 /* 2191 * Mask of slices to enable for the context. Valid values are a subset 2192 * of the bitmask value returned for I915_PARAM_SLICE_MASK. 2193 */ 2194 __u64 slice_mask; 2195 2196 /* 2197 * Mask of subslices to enable for the context. Valid values are a 2198 * subset of the bitmask value return by I915_PARAM_SUBSLICE_MASK. 2199 */ 2200 __u64 subslice_mask; 2201 2202 /* 2203 * Minimum/Maximum number of EUs to enable per subslice for the 2204 * context. min_eus_per_subslice must be inferior or equal to 2205 * max_eus_per_subslice. 2206 */ 2207 __u16 min_eus_per_subslice; 2208 __u16 max_eus_per_subslice; 2209 2210 /* 2211 * Unused for now. Must be cleared to zero. 2212 */ 2213 __u32 rsvd; 2214 }; 2215 2216 /** 2217 * DOC: Virtual Engine uAPI 2218 * 2219 * Virtual engine is a concept where userspace is able to configure a set of 2220 * physical engines, submit a batch buffer, and let the driver execute it on any 2221 * engine from the set as it sees fit. 2222 * 2223 * This is primarily useful on parts which have multiple instances of a same 2224 * class engine, like for example GT3+ Skylake parts with their two VCS engines. 2225 * 2226 * For instance userspace can enumerate all engines of a certain class using the 2227 * previously described `Engine Discovery uAPI`_. After that userspace can 2228 * create a GEM context with a placeholder slot for the virtual engine (using 2229 * `I915_ENGINE_CLASS_INVALID` and `I915_ENGINE_CLASS_INVALID_NONE` for class 2230 * and instance respectively) and finally using the 2231 * `I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE` extension place a virtual engine in 2232 * the same reserved slot. 2233 * 2234 * Example of creating a virtual engine and submitting a batch buffer to it: 2235 * 2236 * .. code-block:: C 2237 * 2238 * I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(virtual, 2) = { 2239 * .base.name = I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE, 2240 * .engine_index = 0, // Place this virtual engine into engine map slot 0 2241 * .num_siblings = 2, 2242 * .engines = { { I915_ENGINE_CLASS_VIDEO, 0 }, 2243 * { I915_ENGINE_CLASS_VIDEO, 1 }, }, 2244 * }; 2245 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 1) = { 2246 * .engines = { { I915_ENGINE_CLASS_INVALID, 2247 * I915_ENGINE_CLASS_INVALID_NONE } }, 2248 * .extensions = to_user_pointer(&virtual), // Chains after load_balance extension 2249 * }; 2250 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2251 * .base = { 2252 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2253 * }, 2254 * .param = { 2255 * .param = I915_CONTEXT_PARAM_ENGINES, 2256 * .value = to_user_pointer(&engines), 2257 * .size = sizeof(engines), 2258 * }, 2259 * }; 2260 * struct drm_i915_gem_context_create_ext create = { 2261 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2262 * .extensions = to_user_pointer(&p_engines); 2263 * }; 2264 * 2265 * ctx_id = gem_context_create_ext(drm_fd, &create); 2266 * 2267 * // Now we have created a GEM context with its engine map containing a 2268 * // single virtual engine. Submissions to this slot can go either to 2269 * // vcs0 or vcs1, depending on the load balancing algorithm used inside 2270 * // the driver. The load balancing is dynamic from one batch buffer to 2271 * // another and transparent to userspace. 2272 * 2273 * ... 2274 * execbuf.rsvd1 = ctx_id; 2275 * execbuf.flags = 0; // Submits to index 0 which is the virtual engine 2276 * gem_execbuf(drm_fd, &execbuf); 2277 */ 2278 2279 /* 2280 * i915_context_engines_load_balance: 2281 * 2282 * Enable load balancing across this set of engines. 2283 * 2284 * Into the I915_EXEC_DEFAULT slot [0], a virtual engine is created that when 2285 * used will proxy the execbuffer request onto one of the set of engines 2286 * in such a way as to distribute the load evenly across the set. 2287 * 2288 * The set of engines must be compatible (e.g. the same HW class) as they 2289 * will share the same logical GPU context and ring. 2290 * 2291 * To intermix rendering with the virtual engine and direct rendering onto 2292 * the backing engines (bypassing the load balancing proxy), the context must 2293 * be defined to use a single timeline for all engines. 2294 */ 2295 struct i915_context_engines_load_balance { 2296 struct i915_user_extension base; 2297 2298 __u16 engine_index; 2299 __u16 num_siblings; 2300 __u32 flags; /* all undefined flags must be zero */ 2301 2302 __u64 mbz64; /* reserved for future use; must be zero */ 2303 2304 struct i915_engine_class_instance engines[]; 2305 } __attribute__((packed)); 2306 2307 #define I915_DEFINE_CONTEXT_ENGINES_LOAD_BALANCE(name__, N__) struct { \ 2308 struct i915_user_extension base; \ 2309 __u16 engine_index; \ 2310 __u16 num_siblings; \ 2311 __u32 flags; \ 2312 __u64 mbz64; \ 2313 struct i915_engine_class_instance engines[N__]; \ 2314 } __attribute__((packed)) name__ 2315 2316 /* 2317 * i915_context_engines_bond: 2318 * 2319 * Constructed bonded pairs for execution within a virtual engine. 2320 * 2321 * All engines are equal, but some are more equal than others. Given 2322 * the distribution of resources in the HW, it may be preferable to run 2323 * a request on a given subset of engines in parallel to a request on a 2324 * specific engine. We enable this selection of engines within a virtual 2325 * engine by specifying bonding pairs, for any given master engine we will 2326 * only execute on one of the corresponding siblings within the virtual engine. 2327 * 2328 * To execute a request in parallel on the master engine and a sibling requires 2329 * coordination with a I915_EXEC_FENCE_SUBMIT. 2330 */ 2331 struct i915_context_engines_bond { 2332 struct i915_user_extension base; 2333 2334 struct i915_engine_class_instance master; 2335 2336 __u16 virtual_index; /* index of virtual engine in ctx->engines[] */ 2337 __u16 num_bonds; 2338 2339 __u64 flags; /* all undefined flags must be zero */ 2340 __u64 mbz64[4]; /* reserved for future use; must be zero */ 2341 2342 struct i915_engine_class_instance engines[]; 2343 } __attribute__((packed)); 2344 2345 #define I915_DEFINE_CONTEXT_ENGINES_BOND(name__, N__) struct { \ 2346 struct i915_user_extension base; \ 2347 struct i915_engine_class_instance master; \ 2348 __u16 virtual_index; \ 2349 __u16 num_bonds; \ 2350 __u64 flags; \ 2351 __u64 mbz64[4]; \ 2352 struct i915_engine_class_instance engines[N__]; \ 2353 } __attribute__((packed)) name__ 2354 2355 /** 2356 * struct i915_context_engines_parallel_submit - Configure engine for 2357 * parallel submission. 2358 * 2359 * Setup a slot in the context engine map to allow multiple BBs to be submitted 2360 * in a single execbuf IOCTL. Those BBs will then be scheduled to run on the GPU 2361 * in parallel. Multiple hardware contexts are created internally in the i915 to 2362 * run these BBs. Once a slot is configured for N BBs only N BBs can be 2363 * submitted in each execbuf IOCTL and this is implicit behavior e.g. The user 2364 * doesn't tell the execbuf IOCTL there are N BBs, the execbuf IOCTL knows how 2365 * many BBs there are based on the slot's configuration. The N BBs are the last 2366 * N buffer objects or first N if I915_EXEC_BATCH_FIRST is set. 2367 * 2368 * The default placement behavior is to create implicit bonds between each 2369 * context if each context maps to more than 1 physical engine (e.g. context is 2370 * a virtual engine). Also we only allow contexts of same engine class and these 2371 * contexts must be in logically contiguous order. Examples of the placement 2372 * behavior are described below. Lastly, the default is to not allow BBs to be 2373 * preempted mid-batch. Rather insert coordinated preemption points on all 2374 * hardware contexts between each set of BBs. Flags could be added in the future 2375 * to change both of these default behaviors. 2376 * 2377 * Returns -EINVAL if hardware context placement configuration is invalid or if 2378 * the placement configuration isn't supported on the platform / submission 2379 * interface. 2380 * Returns -ENODEV if extension isn't supported on the platform / submission 2381 * interface. 2382 * 2383 * .. code-block:: none 2384 * 2385 * Examples syntax: 2386 * CS[X] = generic engine of same class, logical instance X 2387 * INVALID = I915_ENGINE_CLASS_INVALID, I915_ENGINE_CLASS_INVALID_NONE 2388 * 2389 * Example 1 pseudo code: 2390 * set_engines(INVALID) 2391 * set_parallel(engine_index=0, width=2, num_siblings=1, 2392 * engines=CS[0],CS[1]) 2393 * 2394 * Results in the following valid placement: 2395 * CS[0], CS[1] 2396 * 2397 * Example 2 pseudo code: 2398 * set_engines(INVALID) 2399 * set_parallel(engine_index=0, width=2, num_siblings=2, 2400 * engines=CS[0],CS[2],CS[1],CS[3]) 2401 * 2402 * Results in the following valid placements: 2403 * CS[0], CS[1] 2404 * CS[2], CS[3] 2405 * 2406 * This can be thought of as two virtual engines, each containing two 2407 * engines thereby making a 2D array. However, there are bonds tying the 2408 * entries together and placing restrictions on how they can be scheduled. 2409 * Specifically, the scheduler can choose only vertical columns from the 2D 2410 * array. That is, CS[0] is bonded to CS[1] and CS[2] to CS[3]. So if the 2411 * scheduler wants to submit to CS[0], it must also choose CS[1] and vice 2412 * versa. Same for CS[2] requires also using CS[3]. 2413 * VE[0] = CS[0], CS[2] 2414 * VE[1] = CS[1], CS[3] 2415 * 2416 * Example 3 pseudo code: 2417 * set_engines(INVALID) 2418 * set_parallel(engine_index=0, width=2, num_siblings=2, 2419 * engines=CS[0],CS[1],CS[1],CS[3]) 2420 * 2421 * Results in the following valid and invalid placements: 2422 * CS[0], CS[1] 2423 * CS[1], CS[3] - Not logically contiguous, return -EINVAL 2424 */ 2425 struct i915_context_engines_parallel_submit { 2426 /** 2427 * @base: base user extension. 2428 */ 2429 struct i915_user_extension base; 2430 2431 /** 2432 * @engine_index: slot for parallel engine 2433 */ 2434 __u16 engine_index; 2435 2436 /** 2437 * @width: number of contexts per parallel engine or in other words the 2438 * number of batches in each submission 2439 */ 2440 __u16 width; 2441 2442 /** 2443 * @num_siblings: number of siblings per context or in other words the 2444 * number of possible placements for each submission 2445 */ 2446 __u16 num_siblings; 2447 2448 /** 2449 * @mbz16: reserved for future use; must be zero 2450 */ 2451 __u16 mbz16; 2452 2453 /** 2454 * @flags: all undefined flags must be zero, currently not defined flags 2455 */ 2456 __u64 flags; 2457 2458 /** 2459 * @mbz64: reserved for future use; must be zero 2460 */ 2461 __u64 mbz64[3]; 2462 2463 /** 2464 * @engines: 2-d array of engine instances to configure parallel engine 2465 * 2466 * length = width (i) * num_siblings (j) 2467 * index = j + i * num_siblings 2468 */ 2469 struct i915_engine_class_instance engines[]; 2470 2471 } __packed; 2472 2473 #define I915_DEFINE_CONTEXT_ENGINES_PARALLEL_SUBMIT(name__, N__) struct { \ 2474 struct i915_user_extension base; \ 2475 __u16 engine_index; \ 2476 __u16 width; \ 2477 __u16 num_siblings; \ 2478 __u16 mbz16; \ 2479 __u64 flags; \ 2480 __u64 mbz64[3]; \ 2481 struct i915_engine_class_instance engines[N__]; \ 2482 } __attribute__((packed)) name__ 2483 2484 /** 2485 * DOC: Context Engine Map uAPI 2486 * 2487 * Context engine map is a new way of addressing engines when submitting batch- 2488 * buffers, replacing the existing way of using identifiers like `I915_EXEC_BLT` 2489 * inside the flags field of `struct drm_i915_gem_execbuffer2`. 2490 * 2491 * To use it created GEM contexts need to be configured with a list of engines 2492 * the user is intending to submit to. This is accomplished using the 2493 * `I915_CONTEXT_PARAM_ENGINES` parameter and `struct 2494 * i915_context_param_engines`. 2495 * 2496 * For such contexts the `I915_EXEC_RING_MASK` field becomes an index into the 2497 * configured map. 2498 * 2499 * Example of creating such context and submitting against it: 2500 * 2501 * .. code-block:: C 2502 * 2503 * I915_DEFINE_CONTEXT_PARAM_ENGINES(engines, 2) = { 2504 * .engines = { { I915_ENGINE_CLASS_RENDER, 0 }, 2505 * { I915_ENGINE_CLASS_COPY, 0 } } 2506 * }; 2507 * struct drm_i915_gem_context_create_ext_setparam p_engines = { 2508 * .base = { 2509 * .name = I915_CONTEXT_CREATE_EXT_SETPARAM, 2510 * }, 2511 * .param = { 2512 * .param = I915_CONTEXT_PARAM_ENGINES, 2513 * .value = to_user_pointer(&engines), 2514 * .size = sizeof(engines), 2515 * }, 2516 * }; 2517 * struct drm_i915_gem_context_create_ext create = { 2518 * .flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS, 2519 * .extensions = to_user_pointer(&p_engines); 2520 * }; 2521 * 2522 * ctx_id = gem_context_create_ext(drm_fd, &create); 2523 * 2524 * // We have now created a GEM context with two engines in the map: 2525 * // Index 0 points to rcs0 while index 1 points to bcs0. Other engines 2526 * // will not be accessible from this context. 2527 * 2528 * ... 2529 * execbuf.rsvd1 = ctx_id; 2530 * execbuf.flags = 0; // Submits to index 0, which is rcs0 for this context 2531 * gem_execbuf(drm_fd, &execbuf); 2532 * 2533 * ... 2534 * execbuf.rsvd1 = ctx_id; 2535 * execbuf.flags = 1; // Submits to index 0, which is bcs0 for this context 2536 * gem_execbuf(drm_fd, &execbuf); 2537 */ 2538 2539 struct i915_context_param_engines { 2540 __u64 extensions; /* linked chain of extension blocks, 0 terminates */ 2541 #define I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE 0 /* see i915_context_engines_load_balance */ 2542 #define I915_CONTEXT_ENGINES_EXT_BOND 1 /* see i915_context_engines_bond */ 2543 #define I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT 2 /* see i915_context_engines_parallel_submit */ 2544 struct i915_engine_class_instance engines[]; 2545 } __attribute__((packed)); 2546 2547 #define I915_DEFINE_CONTEXT_PARAM_ENGINES(name__, N__) struct { \ 2548 __u64 extensions; \ 2549 struct i915_engine_class_instance engines[N__]; \ 2550 } __attribute__((packed)) name__ 2551 2552 /** 2553 * struct drm_i915_gem_context_create_ext_setparam - Context parameter 2554 * to set or query during context creation. 2555 */ 2556 struct drm_i915_gem_context_create_ext_setparam { 2557 /** @base: Extension link. See struct i915_user_extension. */ 2558 struct i915_user_extension base; 2559 2560 /** 2561 * @param: Context parameter to set or query. 2562 * See struct drm_i915_gem_context_param. 2563 */ 2564 struct drm_i915_gem_context_param param; 2565 }; 2566 2567 struct drm_i915_gem_context_destroy { 2568 __u32 ctx_id; 2569 __u32 pad; 2570 }; 2571 2572 /** 2573 * struct drm_i915_gem_vm_control - Structure to create or destroy VM. 2574 * 2575 * DRM_I915_GEM_VM_CREATE - 2576 * 2577 * Create a new virtual memory address space (ppGTT) for use within a context 2578 * on the same file. Extensions can be provided to configure exactly how the 2579 * address space is setup upon creation. 2580 * 2581 * The id of new VM (bound to the fd) for use with I915_CONTEXT_PARAM_VM is 2582 * returned in the outparam @id. 2583 * 2584 * An extension chain maybe provided, starting with @extensions, and terminated 2585 * by the @next_extension being 0. Currently, no extensions are defined. 2586 * 2587 * DRM_I915_GEM_VM_DESTROY - 2588 * 2589 * Destroys a previously created VM id, specified in @vm_id. 2590 * 2591 * No extensions or flags are allowed currently, and so must be zero. 2592 */ 2593 struct drm_i915_gem_vm_control { 2594 /** @extensions: Zero-terminated chain of extensions. */ 2595 __u64 extensions; 2596 2597 /** @flags: reserved for future usage, currently MBZ */ 2598 __u32 flags; 2599 2600 /** @vm_id: Id of the VM created or to be destroyed */ 2601 __u32 vm_id; 2602 }; 2603 2604 struct drm_i915_reg_read { 2605 /* 2606 * Register offset. 2607 * For 64bit wide registers where the upper 32bits don't immediately 2608 * follow the lower 32bits, the offset of the lower 32bits must 2609 * be specified 2610 */ 2611 __u64 offset; 2612 #define I915_REG_READ_8B_WA (1ul << 0) 2613 2614 __u64 val; /* Return value */ 2615 }; 2616 2617 /* Known registers: 2618 * 2619 * Render engine timestamp - 0x2358 + 64bit - gen7+ 2620 * - Note this register returns an invalid value if using the default 2621 * single instruction 8byte read, in order to workaround that pass 2622 * flag I915_REG_READ_8B_WA in offset field. 2623 * 2624 */ 2625 2626 struct drm_i915_reset_stats { 2627 __u32 ctx_id; 2628 __u32 flags; 2629 2630 /* All resets since boot/module reload, for all contexts */ 2631 __u32 reset_count; 2632 2633 /* Number of batches lost when active in GPU, for this context */ 2634 __u32 batch_active; 2635 2636 /* Number of batches lost pending for execution, for this context */ 2637 __u32 batch_pending; 2638 2639 __u32 pad; 2640 }; 2641 2642 /** 2643 * struct drm_i915_gem_userptr - Create GEM object from user allocated memory. 2644 * 2645 * Userptr objects have several restrictions on what ioctls can be used with the 2646 * object handle. 2647 */ 2648 struct drm_i915_gem_userptr { 2649 /** 2650 * @user_ptr: The pointer to the allocated memory. 2651 * 2652 * Needs to be aligned to PAGE_SIZE. 2653 */ 2654 __u64 user_ptr; 2655 2656 /** 2657 * @user_size: 2658 * 2659 * The size in bytes for the allocated memory. This will also become the 2660 * object size. 2661 * 2662 * Needs to be aligned to PAGE_SIZE, and should be at least PAGE_SIZE, 2663 * or larger. 2664 */ 2665 __u64 user_size; 2666 2667 /** 2668 * @flags: 2669 * 2670 * Supported flags: 2671 * 2672 * I915_USERPTR_READ_ONLY: 2673 * 2674 * Mark the object as readonly, this also means GPU access can only be 2675 * readonly. This is only supported on HW which supports readonly access 2676 * through the GTT. If the HW can't support readonly access, an error is 2677 * returned. 2678 * 2679 * I915_USERPTR_PROBE: 2680 * 2681 * Probe the provided @user_ptr range and validate that the @user_ptr is 2682 * indeed pointing to normal memory and that the range is also valid. 2683 * For example if some garbage address is given to the kernel, then this 2684 * should complain. 2685 * 2686 * Returns -EFAULT if the probe failed. 2687 * 2688 * Note that this doesn't populate the backing pages, and also doesn't 2689 * guarantee that the object will remain valid when the object is 2690 * eventually used. 2691 * 2692 * The kernel supports this feature if I915_PARAM_HAS_USERPTR_PROBE 2693 * returns a non-zero value. 2694 * 2695 * I915_USERPTR_UNSYNCHRONIZED: 2696 * 2697 * NOT USED. Setting this flag will result in an error. 2698 */ 2699 __u32 flags; 2700 #define I915_USERPTR_READ_ONLY 0x1 2701 #define I915_USERPTR_PROBE 0x2 2702 #define I915_USERPTR_UNSYNCHRONIZED 0x80000000 2703 /** 2704 * @handle: Returned handle for the object. 2705 * 2706 * Object handles are nonzero. 2707 */ 2708 __u32 handle; 2709 }; 2710 2711 enum drm_i915_oa_format { 2712 I915_OA_FORMAT_A13 = 1, /* HSW only */ 2713 I915_OA_FORMAT_A29, /* HSW only */ 2714 I915_OA_FORMAT_A13_B8_C8, /* HSW only */ 2715 I915_OA_FORMAT_B4_C8, /* HSW only */ 2716 I915_OA_FORMAT_A45_B8_C8, /* HSW only */ 2717 I915_OA_FORMAT_B4_C8_A16, /* HSW only */ 2718 I915_OA_FORMAT_C4_B8, /* HSW+ */ 2719 2720 /* Gen8+ */ 2721 I915_OA_FORMAT_A12, 2722 I915_OA_FORMAT_A12_B8_C8, 2723 I915_OA_FORMAT_A32u40_A4u32_B8_C8, 2724 2725 /* DG2 */ 2726 I915_OAR_FORMAT_A32u40_A4u32_B8_C8, 2727 I915_OA_FORMAT_A24u40_A14u32_B8_C8, 2728 2729 /* MTL OAM */ 2730 I915_OAM_FORMAT_MPEC8u64_B8_C8, 2731 I915_OAM_FORMAT_MPEC8u32_B8_C8, 2732 2733 I915_OA_FORMAT_MAX /* non-ABI */ 2734 }; 2735 2736 enum drm_i915_perf_property_id { 2737 /** 2738 * Open the stream for a specific context handle (as used with 2739 * execbuffer2). A stream opened for a specific context this way 2740 * won't typically require root privileges. 2741 * 2742 * This property is available in perf revision 1. 2743 */ 2744 DRM_I915_PERF_PROP_CTX_HANDLE = 1, 2745 2746 /** 2747 * A value of 1 requests the inclusion of raw OA unit reports as 2748 * part of stream samples. 2749 * 2750 * This property is available in perf revision 1. 2751 */ 2752 DRM_I915_PERF_PROP_SAMPLE_OA, 2753 2754 /** 2755 * The value specifies which set of OA unit metrics should be 2756 * configured, defining the contents of any OA unit reports. 2757 * 2758 * This property is available in perf revision 1. 2759 */ 2760 DRM_I915_PERF_PROP_OA_METRICS_SET, 2761 2762 /** 2763 * The value specifies the size and layout of OA unit reports. 2764 * 2765 * This property is available in perf revision 1. 2766 */ 2767 DRM_I915_PERF_PROP_OA_FORMAT, 2768 2769 /** 2770 * Specifying this property implicitly requests periodic OA unit 2771 * sampling and (at least on Haswell) the sampling frequency is derived 2772 * from this exponent as follows: 2773 * 2774 * 80ns * 2^(period_exponent + 1) 2775 * 2776 * This property is available in perf revision 1. 2777 */ 2778 DRM_I915_PERF_PROP_OA_EXPONENT, 2779 2780 /** 2781 * Specifying this property is only valid when specify a context to 2782 * filter with DRM_I915_PERF_PROP_CTX_HANDLE. Specifying this property 2783 * will hold preemption of the particular context we want to gather 2784 * performance data about. The execbuf2 submissions must include a 2785 * drm_i915_gem_execbuffer_ext_perf parameter for this to apply. 2786 * 2787 * This property is available in perf revision 3. 2788 */ 2789 DRM_I915_PERF_PROP_HOLD_PREEMPTION, 2790 2791 /** 2792 * Specifying this pins all contexts to the specified SSEU power 2793 * configuration for the duration of the recording. 2794 * 2795 * This parameter's value is a pointer to a struct 2796 * drm_i915_gem_context_param_sseu. 2797 * 2798 * This property is available in perf revision 4. 2799 */ 2800 DRM_I915_PERF_PROP_GLOBAL_SSEU, 2801 2802 /** 2803 * This optional parameter specifies the timer interval in nanoseconds 2804 * at which the i915 driver will check the OA buffer for available data. 2805 * Minimum allowed value is 100 microseconds. A default value is used by 2806 * the driver if this parameter is not specified. Note that larger timer 2807 * values will reduce cpu consumption during OA perf captures. However, 2808 * excessively large values would potentially result in OA buffer 2809 * overwrites as captures reach end of the OA buffer. 2810 * 2811 * This property is available in perf revision 5. 2812 */ 2813 DRM_I915_PERF_PROP_POLL_OA_PERIOD, 2814 2815 /** 2816 * Multiple engines may be mapped to the same OA unit. The OA unit is 2817 * identified by class:instance of any engine mapped to it. 2818 * 2819 * This parameter specifies the engine class and must be passed along 2820 * with DRM_I915_PERF_PROP_OA_ENGINE_INSTANCE. 2821 * 2822 * This property is available in perf revision 6. 2823 */ 2824 DRM_I915_PERF_PROP_OA_ENGINE_CLASS, 2825 2826 /** 2827 * This parameter specifies the engine instance and must be passed along 2828 * with DRM_I915_PERF_PROP_OA_ENGINE_CLASS. 2829 * 2830 * This property is available in perf revision 6. 2831 */ 2832 DRM_I915_PERF_PROP_OA_ENGINE_INSTANCE, 2833 2834 DRM_I915_PERF_PROP_MAX /* non-ABI */ 2835 }; 2836 2837 struct drm_i915_perf_open_param { 2838 __u32 flags; 2839 #define I915_PERF_FLAG_FD_CLOEXEC (1<<0) 2840 #define I915_PERF_FLAG_FD_NONBLOCK (1<<1) 2841 #define I915_PERF_FLAG_DISABLED (1<<2) 2842 2843 /** The number of u64 (id, value) pairs */ 2844 __u32 num_properties; 2845 2846 /** 2847 * Pointer to array of u64 (id, value) pairs configuring the stream 2848 * to open. 2849 */ 2850 __u64 properties_ptr; 2851 }; 2852 2853 /* 2854 * Enable data capture for a stream that was either opened in a disabled state 2855 * via I915_PERF_FLAG_DISABLED or was later disabled via 2856 * I915_PERF_IOCTL_DISABLE. 2857 * 2858 * It is intended to be cheaper to disable and enable a stream than it may be 2859 * to close and re-open a stream with the same configuration. 2860 * 2861 * It's undefined whether any pending data for the stream will be lost. 2862 * 2863 * This ioctl is available in perf revision 1. 2864 */ 2865 #define I915_PERF_IOCTL_ENABLE _IO('i', 0x0) 2866 2867 /* 2868 * Disable data capture for a stream. 2869 * 2870 * It is an error to try and read a stream that is disabled. 2871 * 2872 * This ioctl is available in perf revision 1. 2873 */ 2874 #define I915_PERF_IOCTL_DISABLE _IO('i', 0x1) 2875 2876 /* 2877 * Change metrics_set captured by a stream. 2878 * 2879 * If the stream is bound to a specific context, the configuration change 2880 * will performed inline with that context such that it takes effect before 2881 * the next execbuf submission. 2882 * 2883 * Returns the previously bound metrics set id, or a negative error code. 2884 * 2885 * This ioctl is available in perf revision 2. 2886 */ 2887 #define I915_PERF_IOCTL_CONFIG _IO('i', 0x2) 2888 2889 /* 2890 * Common to all i915 perf records 2891 */ 2892 struct drm_i915_perf_record_header { 2893 __u32 type; 2894 __u16 pad; 2895 __u16 size; 2896 }; 2897 2898 enum drm_i915_perf_record_type { 2899 2900 /** 2901 * Samples are the work horse record type whose contents are extensible 2902 * and defined when opening an i915 perf stream based on the given 2903 * properties. 2904 * 2905 * Boolean properties following the naming convention 2906 * DRM_I915_PERF_SAMPLE_xyz_PROP request the inclusion of 'xyz' data in 2907 * every sample. 2908 * 2909 * The order of these sample properties given by userspace has no 2910 * affect on the ordering of data within a sample. The order is 2911 * documented here. 2912 * 2913 * struct { 2914 * struct drm_i915_perf_record_header header; 2915 * 2916 * { u32 oa_report[]; } && DRM_I915_PERF_PROP_SAMPLE_OA 2917 * }; 2918 */ 2919 DRM_I915_PERF_RECORD_SAMPLE = 1, 2920 2921 /* 2922 * Indicates that one or more OA reports were not written by the 2923 * hardware. This can happen for example if an MI_REPORT_PERF_COUNT 2924 * command collides with periodic sampling - which would be more likely 2925 * at higher sampling frequencies. 2926 */ 2927 DRM_I915_PERF_RECORD_OA_REPORT_LOST = 2, 2928 2929 /** 2930 * An error occurred that resulted in all pending OA reports being lost. 2931 */ 2932 DRM_I915_PERF_RECORD_OA_BUFFER_LOST = 3, 2933 2934 DRM_I915_PERF_RECORD_MAX /* non-ABI */ 2935 }; 2936 2937 /** 2938 * struct drm_i915_perf_oa_config 2939 * 2940 * Structure to upload perf dynamic configuration into the kernel. 2941 */ 2942 struct drm_i915_perf_oa_config { 2943 /** 2944 * @uuid: 2945 * 2946 * String formatted like "%\08x-%\04x-%\04x-%\04x-%\012x" 2947 */ 2948 char uuid[36]; 2949 2950 /** 2951 * @n_mux_regs: 2952 * 2953 * Number of mux regs in &mux_regs_ptr. 2954 */ 2955 __u32 n_mux_regs; 2956 2957 /** 2958 * @n_boolean_regs: 2959 * 2960 * Number of boolean regs in &boolean_regs_ptr. 2961 */ 2962 __u32 n_boolean_regs; 2963 2964 /** 2965 * @n_flex_regs: 2966 * 2967 * Number of flex regs in &flex_regs_ptr. 2968 */ 2969 __u32 n_flex_regs; 2970 2971 /** 2972 * @mux_regs_ptr: 2973 * 2974 * Pointer to tuples of u32 values (register address, value) for mux 2975 * registers. Expected length of buffer is (2 * sizeof(u32) * 2976 * &n_mux_regs). 2977 */ 2978 __u64 mux_regs_ptr; 2979 2980 /** 2981 * @boolean_regs_ptr: 2982 * 2983 * Pointer to tuples of u32 values (register address, value) for mux 2984 * registers. Expected length of buffer is (2 * sizeof(u32) * 2985 * &n_boolean_regs). 2986 */ 2987 __u64 boolean_regs_ptr; 2988 2989 /** 2990 * @flex_regs_ptr: 2991 * 2992 * Pointer to tuples of u32 values (register address, value) for mux 2993 * registers. Expected length of buffer is (2 * sizeof(u32) * 2994 * &n_flex_regs). 2995 */ 2996 __u64 flex_regs_ptr; 2997 }; 2998 2999 /** 3000 * struct drm_i915_query_item - An individual query for the kernel to process. 3001 * 3002 * The behaviour is determined by the @query_id. Note that exactly what 3003 * @data_ptr is also depends on the specific @query_id. 3004 */ 3005 struct drm_i915_query_item { 3006 /** 3007 * @query_id: 3008 * 3009 * The id for this query. Currently accepted query IDs are: 3010 * - %DRM_I915_QUERY_TOPOLOGY_INFO (see struct drm_i915_query_topology_info) 3011 * - %DRM_I915_QUERY_ENGINE_INFO (see struct drm_i915_engine_info) 3012 * - %DRM_I915_QUERY_PERF_CONFIG (see struct drm_i915_query_perf_config) 3013 * - %DRM_I915_QUERY_MEMORY_REGIONS (see struct drm_i915_query_memory_regions) 3014 * - %DRM_I915_QUERY_HWCONFIG_BLOB (see `GuC HWCONFIG blob uAPI`) 3015 * - %DRM_I915_QUERY_GEOMETRY_SUBSLICES (see struct drm_i915_query_topology_info) 3016 */ 3017 __u64 query_id; 3018 #define DRM_I915_QUERY_TOPOLOGY_INFO 1 3019 #define DRM_I915_QUERY_ENGINE_INFO 2 3020 #define DRM_I915_QUERY_PERF_CONFIG 3 3021 #define DRM_I915_QUERY_MEMORY_REGIONS 4 3022 #define DRM_I915_QUERY_HWCONFIG_BLOB 5 3023 #define DRM_I915_QUERY_GEOMETRY_SUBSLICES 6 3024 /* Must be kept compact -- no holes and well documented */ 3025 3026 /** 3027 * @length: 3028 * 3029 * When set to zero by userspace, this is filled with the size of the 3030 * data to be written at the @data_ptr pointer. The kernel sets this 3031 * value to a negative value to signal an error on a particular query 3032 * item. 3033 */ 3034 __s32 length; 3035 3036 /** 3037 * @flags: 3038 * 3039 * When &query_id == %DRM_I915_QUERY_TOPOLOGY_INFO, must be 0. 3040 * 3041 * When &query_id == %DRM_I915_QUERY_PERF_CONFIG, must be one of the 3042 * following: 3043 * 3044 * - %DRM_I915_QUERY_PERF_CONFIG_LIST 3045 * - %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 3046 * - %DRM_I915_QUERY_PERF_CONFIG_FOR_UUID 3047 * 3048 * When &query_id == %DRM_I915_QUERY_GEOMETRY_SUBSLICES must contain 3049 * a struct i915_engine_class_instance that references a render engine. 3050 */ 3051 __u32 flags; 3052 #define DRM_I915_QUERY_PERF_CONFIG_LIST 1 3053 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID 2 3054 #define DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID 3 3055 3056 /** 3057 * @data_ptr: 3058 * 3059 * Data will be written at the location pointed by @data_ptr when the 3060 * value of @length matches the length of the data to be written by the 3061 * kernel. 3062 */ 3063 __u64 data_ptr; 3064 }; 3065 3066 /** 3067 * struct drm_i915_query - Supply an array of struct drm_i915_query_item for the 3068 * kernel to fill out. 3069 * 3070 * Note that this is generally a two step process for each struct 3071 * drm_i915_query_item in the array: 3072 * 3073 * 1. Call the DRM_IOCTL_I915_QUERY, giving it our array of struct 3074 * drm_i915_query_item, with &drm_i915_query_item.length set to zero. The 3075 * kernel will then fill in the size, in bytes, which tells userspace how 3076 * memory it needs to allocate for the blob(say for an array of properties). 3077 * 3078 * 2. Next we call DRM_IOCTL_I915_QUERY again, this time with the 3079 * &drm_i915_query_item.data_ptr equal to our newly allocated blob. Note that 3080 * the &drm_i915_query_item.length should still be the same as what the 3081 * kernel previously set. At this point the kernel can fill in the blob. 3082 * 3083 * Note that for some query items it can make sense for userspace to just pass 3084 * in a buffer/blob equal to or larger than the required size. In this case only 3085 * a single ioctl call is needed. For some smaller query items this can work 3086 * quite well. 3087 * 3088 */ 3089 struct drm_i915_query { 3090 /** @num_items: The number of elements in the @items_ptr array */ 3091 __u32 num_items; 3092 3093 /** 3094 * @flags: Unused for now. Must be cleared to zero. 3095 */ 3096 __u32 flags; 3097 3098 /** 3099 * @items_ptr: 3100 * 3101 * Pointer to an array of struct drm_i915_query_item. The number of 3102 * array elements is @num_items. 3103 */ 3104 __u64 items_ptr; 3105 }; 3106 3107 /** 3108 * struct drm_i915_query_topology_info 3109 * 3110 * Describes slice/subslice/EU information queried by 3111 * %DRM_I915_QUERY_TOPOLOGY_INFO 3112 */ 3113 struct drm_i915_query_topology_info { 3114 /** 3115 * @flags: 3116 * 3117 * Unused for now. Must be cleared to zero. 3118 */ 3119 __u16 flags; 3120 3121 /** 3122 * @max_slices: 3123 * 3124 * The number of bits used to express the slice mask. 3125 */ 3126 __u16 max_slices; 3127 3128 /** 3129 * @max_subslices: 3130 * 3131 * The number of bits used to express the subslice mask. 3132 */ 3133 __u16 max_subslices; 3134 3135 /** 3136 * @max_eus_per_subslice: 3137 * 3138 * The number of bits in the EU mask that correspond to a single 3139 * subslice's EUs. 3140 */ 3141 __u16 max_eus_per_subslice; 3142 3143 /** 3144 * @subslice_offset: 3145 * 3146 * Offset in data[] at which the subslice masks are stored. 3147 */ 3148 __u16 subslice_offset; 3149 3150 /** 3151 * @subslice_stride: 3152 * 3153 * Stride at which each of the subslice masks for each slice are 3154 * stored. 3155 */ 3156 __u16 subslice_stride; 3157 3158 /** 3159 * @eu_offset: 3160 * 3161 * Offset in data[] at which the EU masks are stored. 3162 */ 3163 __u16 eu_offset; 3164 3165 /** 3166 * @eu_stride: 3167 * 3168 * Stride at which each of the EU masks for each subslice are stored. 3169 */ 3170 __u16 eu_stride; 3171 3172 /** 3173 * @data: 3174 * 3175 * Contains 3 pieces of information : 3176 * 3177 * - The slice mask with one bit per slice telling whether a slice is 3178 * available. The availability of slice X can be queried with the 3179 * following formula : 3180 * 3181 * .. code:: c 3182 * 3183 * (data[X / 8] >> (X % 8)) & 1 3184 * 3185 * Starting with Xe_HP platforms, Intel hardware no longer has 3186 * traditional slices so i915 will always report a single slice 3187 * (hardcoded slicemask = 0x1) which contains all of the platform's 3188 * subslices. I.e., the mask here does not reflect any of the newer 3189 * hardware concepts such as "gslices" or "cslices" since userspace 3190 * is capable of inferring those from the subslice mask. 3191 * 3192 * - The subslice mask for each slice with one bit per subslice telling 3193 * whether a subslice is available. Starting with Gen12 we use the 3194 * term "subslice" to refer to what the hardware documentation 3195 * describes as a "dual-subslices." The availability of subslice Y 3196 * in slice X can be queried with the following formula : 3197 * 3198 * .. code:: c 3199 * 3200 * (data[subslice_offset + X * subslice_stride + Y / 8] >> (Y % 8)) & 1 3201 * 3202 * - The EU mask for each subslice in each slice, with one bit per EU 3203 * telling whether an EU is available. The availability of EU Z in 3204 * subslice Y in slice X can be queried with the following formula : 3205 * 3206 * .. code:: c 3207 * 3208 * (data[eu_offset + 3209 * (X * max_subslices + Y) * eu_stride + 3210 * Z / 8 3211 * ] >> (Z % 8)) & 1 3212 */ 3213 __u8 data[]; 3214 }; 3215 3216 /** 3217 * DOC: Engine Discovery uAPI 3218 * 3219 * Engine discovery uAPI is a way of enumerating physical engines present in a 3220 * GPU associated with an open i915 DRM file descriptor. This supersedes the old 3221 * way of using `DRM_IOCTL_I915_GETPARAM` and engine identifiers like 3222 * `I915_PARAM_HAS_BLT`. 3223 * 3224 * The need for this interface came starting with Icelake and newer GPUs, which 3225 * started to establish a pattern of having multiple engines of a same class, 3226 * where not all instances were always completely functionally equivalent. 3227 * 3228 * Entry point for this uapi is `DRM_IOCTL_I915_QUERY` with the 3229 * `DRM_I915_QUERY_ENGINE_INFO` as the queried item id. 3230 * 3231 * Example for getting the list of engines: 3232 * 3233 * .. code-block:: C 3234 * 3235 * struct drm_i915_query_engine_info *info; 3236 * struct drm_i915_query_item item = { 3237 * .query_id = DRM_I915_QUERY_ENGINE_INFO; 3238 * }; 3239 * struct drm_i915_query query = { 3240 * .num_items = 1, 3241 * .items_ptr = (uintptr_t)&item, 3242 * }; 3243 * int err, i; 3244 * 3245 * // First query the size of the blob we need, this needs to be large 3246 * // enough to hold our array of engines. The kernel will fill out the 3247 * // item.length for us, which is the number of bytes we need. 3248 * // 3249 * // Alternatively a large buffer can be allocated straight away enabling 3250 * // querying in one pass, in which case item.length should contain the 3251 * // length of the provided buffer. 3252 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3253 * if (err) ... 3254 * 3255 * info = calloc(1, item.length); 3256 * // Now that we allocated the required number of bytes, we call the ioctl 3257 * // again, this time with the data_ptr pointing to our newly allocated 3258 * // blob, which the kernel can then populate with info on all engines. 3259 * item.data_ptr = (uintptr_t)&info, 3260 * 3261 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3262 * if (err) ... 3263 * 3264 * // We can now access each engine in the array 3265 * for (i = 0; i < info->num_engines; i++) { 3266 * struct drm_i915_engine_info einfo = info->engines[i]; 3267 * u16 class = einfo.engine.class; 3268 * u16 instance = einfo.engine.instance; 3269 * .... 3270 * } 3271 * 3272 * free(info); 3273 * 3274 * Each of the enumerated engines, apart from being defined by its class and 3275 * instance (see `struct i915_engine_class_instance`), also can have flags and 3276 * capabilities defined as documented in i915_drm.h. 3277 * 3278 * For instance video engines which support HEVC encoding will have the 3279 * `I915_VIDEO_CLASS_CAPABILITY_HEVC` capability bit set. 3280 * 3281 * Engine discovery only fully comes to its own when combined with the new way 3282 * of addressing engines when submitting batch buffers using contexts with 3283 * engine maps configured. 3284 */ 3285 3286 /** 3287 * struct drm_i915_engine_info 3288 * 3289 * Describes one engine and it's capabilities as known to the driver. 3290 */ 3291 struct drm_i915_engine_info { 3292 /** @engine: Engine class and instance. */ 3293 struct i915_engine_class_instance engine; 3294 3295 /** @rsvd0: Reserved field. */ 3296 __u32 rsvd0; 3297 3298 /** @flags: Engine flags. */ 3299 __u64 flags; 3300 #define I915_ENGINE_INFO_HAS_LOGICAL_INSTANCE (1 << 0) 3301 3302 /** @capabilities: Capabilities of this engine. */ 3303 __u64 capabilities; 3304 #define I915_VIDEO_CLASS_CAPABILITY_HEVC (1 << 0) 3305 #define I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC (1 << 1) 3306 3307 /** @logical_instance: Logical instance of engine */ 3308 __u16 logical_instance; 3309 3310 /** @rsvd1: Reserved fields. */ 3311 __u16 rsvd1[3]; 3312 /** @rsvd2: Reserved fields. */ 3313 __u64 rsvd2[3]; 3314 }; 3315 3316 /** 3317 * struct drm_i915_query_engine_info 3318 * 3319 * Engine info query enumerates all engines known to the driver by filling in 3320 * an array of struct drm_i915_engine_info structures. 3321 */ 3322 struct drm_i915_query_engine_info { 3323 /** @num_engines: Number of struct drm_i915_engine_info structs following. */ 3324 __u32 num_engines; 3325 3326 /** @rsvd: MBZ */ 3327 __u32 rsvd[3]; 3328 3329 /** @engines: Marker for drm_i915_engine_info structures. */ 3330 struct drm_i915_engine_info engines[]; 3331 }; 3332 3333 /** 3334 * struct drm_i915_query_perf_config 3335 * 3336 * Data written by the kernel with query %DRM_I915_QUERY_PERF_CONFIG and 3337 * %DRM_I915_QUERY_GEOMETRY_SUBSLICES. 3338 */ 3339 struct drm_i915_query_perf_config { 3340 union { 3341 /** 3342 * @n_configs: 3343 * 3344 * When &drm_i915_query_item.flags == 3345 * %DRM_I915_QUERY_PERF_CONFIG_LIST, i915 sets this fields to 3346 * the number of configurations available. 3347 */ 3348 __u64 n_configs; 3349 3350 /** 3351 * @config: 3352 * 3353 * When &drm_i915_query_item.flags == 3354 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_ID, i915 will use the 3355 * value in this field as configuration identifier to decide 3356 * what data to write into config_ptr. 3357 */ 3358 __u64 config; 3359 3360 /** 3361 * @uuid: 3362 * 3363 * When &drm_i915_query_item.flags == 3364 * %DRM_I915_QUERY_PERF_CONFIG_DATA_FOR_UUID, i915 will use the 3365 * value in this field as configuration identifier to decide 3366 * what data to write into config_ptr. 3367 * 3368 * String formatted like "%08x-%04x-%04x-%04x-%012x" 3369 */ 3370 char uuid[36]; 3371 }; 3372 3373 /** 3374 * @flags: 3375 * 3376 * Unused for now. Must be cleared to zero. 3377 */ 3378 __u32 flags; 3379 3380 /** 3381 * @data: 3382 * 3383 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_LIST, 3384 * i915 will write an array of __u64 of configuration identifiers. 3385 * 3386 * When &drm_i915_query_item.flags == %DRM_I915_QUERY_PERF_CONFIG_DATA, 3387 * i915 will write a struct drm_i915_perf_oa_config. If the following 3388 * fields of struct drm_i915_perf_oa_config are not set to 0, i915 will 3389 * write into the associated pointers the values of submitted when the 3390 * configuration was created : 3391 * 3392 * - &drm_i915_perf_oa_config.n_mux_regs 3393 * - &drm_i915_perf_oa_config.n_boolean_regs 3394 * - &drm_i915_perf_oa_config.n_flex_regs 3395 */ 3396 __u8 data[]; 3397 }; 3398 3399 /** 3400 * enum drm_i915_gem_memory_class - Supported memory classes 3401 */ 3402 enum drm_i915_gem_memory_class { 3403 /** @I915_MEMORY_CLASS_SYSTEM: System memory */ 3404 I915_MEMORY_CLASS_SYSTEM = 0, 3405 /** @I915_MEMORY_CLASS_DEVICE: Device local-memory */ 3406 I915_MEMORY_CLASS_DEVICE, 3407 }; 3408 3409 /** 3410 * struct drm_i915_gem_memory_class_instance - Identify particular memory region 3411 */ 3412 struct drm_i915_gem_memory_class_instance { 3413 /** @memory_class: See enum drm_i915_gem_memory_class */ 3414 __u16 memory_class; 3415 3416 /** @memory_instance: Which instance */ 3417 __u16 memory_instance; 3418 }; 3419 3420 /** 3421 * struct drm_i915_memory_region_info - Describes one region as known to the 3422 * driver. 3423 * 3424 * Note this is using both struct drm_i915_query_item and struct drm_i915_query. 3425 * For this new query we are adding the new query id DRM_I915_QUERY_MEMORY_REGIONS 3426 * at &drm_i915_query_item.query_id. 3427 */ 3428 struct drm_i915_memory_region_info { 3429 /** @region: The class:instance pair encoding */ 3430 struct drm_i915_gem_memory_class_instance region; 3431 3432 /** @rsvd0: MBZ */ 3433 __u32 rsvd0; 3434 3435 /** 3436 * @probed_size: Memory probed by the driver 3437 * 3438 * Note that it should not be possible to ever encounter a zero value 3439 * here, also note that no current region type will ever return -1 here. 3440 * Although for future region types, this might be a possibility. The 3441 * same applies to the other size fields. 3442 */ 3443 __u64 probed_size; 3444 3445 /** 3446 * @unallocated_size: Estimate of memory remaining 3447 * 3448 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable accounting. 3449 * Without this (or if this is an older kernel) the value here will 3450 * always equal the @probed_size. Note this is only currently tracked 3451 * for I915_MEMORY_CLASS_DEVICE regions (for other types the value here 3452 * will always equal the @probed_size). 3453 */ 3454 __u64 unallocated_size; 3455 3456 union { 3457 /** @rsvd1: MBZ */ 3458 __u64 rsvd1[8]; 3459 struct { 3460 /** 3461 * @probed_cpu_visible_size: Memory probed by the driver 3462 * that is CPU accessible. 3463 * 3464 * This will be always be <= @probed_size, and the 3465 * remainder (if there is any) will not be CPU 3466 * accessible. 3467 * 3468 * On systems without small BAR, the @probed_size will 3469 * always equal the @probed_cpu_visible_size, since all 3470 * of it will be CPU accessible. 3471 * 3472 * Note this is only tracked for 3473 * I915_MEMORY_CLASS_DEVICE regions (for other types the 3474 * value here will always equal the @probed_size). 3475 * 3476 * Note that if the value returned here is zero, then 3477 * this must be an old kernel which lacks the relevant 3478 * small-bar uAPI support (including 3479 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS), but on 3480 * such systems we should never actually end up with a 3481 * small BAR configuration, assuming we are able to load 3482 * the kernel module. Hence it should be safe to treat 3483 * this the same as when @probed_cpu_visible_size == 3484 * @probed_size. 3485 */ 3486 __u64 probed_cpu_visible_size; 3487 3488 /** 3489 * @unallocated_cpu_visible_size: Estimate of CPU 3490 * visible memory remaining. 3491 * 3492 * Note this is only tracked for 3493 * I915_MEMORY_CLASS_DEVICE regions (for other types the 3494 * value here will always equal the 3495 * @probed_cpu_visible_size). 3496 * 3497 * Requires CAP_PERFMON or CAP_SYS_ADMIN to get reliable 3498 * accounting. Without this the value here will always 3499 * equal the @probed_cpu_visible_size. Note this is only 3500 * currently tracked for I915_MEMORY_CLASS_DEVICE 3501 * regions (for other types the value here will also 3502 * always equal the @probed_cpu_visible_size). 3503 * 3504 * If this is an older kernel the value here will be 3505 * zero, see also @probed_cpu_visible_size. 3506 */ 3507 __u64 unallocated_cpu_visible_size; 3508 }; 3509 }; 3510 }; 3511 3512 /** 3513 * struct drm_i915_query_memory_regions 3514 * 3515 * The region info query enumerates all regions known to the driver by filling 3516 * in an array of struct drm_i915_memory_region_info structures. 3517 * 3518 * Example for getting the list of supported regions: 3519 * 3520 * .. code-block:: C 3521 * 3522 * struct drm_i915_query_memory_regions *info; 3523 * struct drm_i915_query_item item = { 3524 * .query_id = DRM_I915_QUERY_MEMORY_REGIONS; 3525 * }; 3526 * struct drm_i915_query query = { 3527 * .num_items = 1, 3528 * .items_ptr = (uintptr_t)&item, 3529 * }; 3530 * int err, i; 3531 * 3532 * // First query the size of the blob we need, this needs to be large 3533 * // enough to hold our array of regions. The kernel will fill out the 3534 * // item.length for us, which is the number of bytes we need. 3535 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3536 * if (err) ... 3537 * 3538 * info = calloc(1, item.length); 3539 * // Now that we allocated the required number of bytes, we call the ioctl 3540 * // again, this time with the data_ptr pointing to our newly allocated 3541 * // blob, which the kernel can then populate with the all the region info. 3542 * item.data_ptr = (uintptr_t)&info, 3543 * 3544 * err = ioctl(fd, DRM_IOCTL_I915_QUERY, &query); 3545 * if (err) ... 3546 * 3547 * // We can now access each region in the array 3548 * for (i = 0; i < info->num_regions; i++) { 3549 * struct drm_i915_memory_region_info mr = info->regions[i]; 3550 * u16 class = mr.region.class; 3551 * u16 instance = mr.region.instance; 3552 * 3553 * .... 3554 * } 3555 * 3556 * free(info); 3557 */ 3558 struct drm_i915_query_memory_regions { 3559 /** @num_regions: Number of supported regions */ 3560 __u32 num_regions; 3561 3562 /** @rsvd: MBZ */ 3563 __u32 rsvd[3]; 3564 3565 /** @regions: Info about each supported region */ 3566 struct drm_i915_memory_region_info regions[]; 3567 }; 3568 3569 /** 3570 * DOC: GuC HWCONFIG blob uAPI 3571 * 3572 * The GuC produces a blob with information about the current device. 3573 * i915 reads this blob from GuC and makes it available via this uAPI. 3574 * 3575 * The format and meaning of the blob content are documented in the 3576 * Programmer's Reference Manual. 3577 */ 3578 3579 /** 3580 * struct drm_i915_gem_create_ext - Existing gem_create behaviour, with added 3581 * extension support using struct i915_user_extension. 3582 * 3583 * Note that new buffer flags should be added here, at least for the stuff that 3584 * is immutable. Previously we would have two ioctls, one to create the object 3585 * with gem_create, and another to apply various parameters, however this 3586 * creates some ambiguity for the params which are considered immutable. Also in 3587 * general we're phasing out the various SET/GET ioctls. 3588 */ 3589 struct drm_i915_gem_create_ext { 3590 /** 3591 * @size: Requested size for the object. 3592 * 3593 * The (page-aligned) allocated size for the object will be returned. 3594 * 3595 * On platforms like DG2/ATS the kernel will always use 64K or larger 3596 * pages for I915_MEMORY_CLASS_DEVICE. The kernel also requires a 3597 * minimum of 64K GTT alignment for such objects. 3598 * 3599 * NOTE: Previously the ABI here required a minimum GTT alignment of 2M 3600 * on DG2/ATS, due to how the hardware implemented 64K GTT page support, 3601 * where we had the following complications: 3602 * 3603 * 1) The entire PDE (which covers a 2MB virtual address range), must 3604 * contain only 64K PTEs, i.e mixing 4K and 64K PTEs in the same 3605 * PDE is forbidden by the hardware. 3606 * 3607 * 2) We still need to support 4K PTEs for I915_MEMORY_CLASS_SYSTEM 3608 * objects. 3609 * 3610 * However on actual production HW this was completely changed to now 3611 * allow setting a TLB hint at the PTE level (see PS64), which is a lot 3612 * more flexible than the above. With this the 2M restriction was 3613 * dropped where we now only require 64K. 3614 */ 3615 __u64 size; 3616 3617 /** 3618 * @handle: Returned handle for the object. 3619 * 3620 * Object handles are nonzero. 3621 */ 3622 __u32 handle; 3623 3624 /** 3625 * @flags: Optional flags. 3626 * 3627 * Supported values: 3628 * 3629 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS - Signal to the kernel that 3630 * the object will need to be accessed via the CPU. 3631 * 3632 * Only valid when placing objects in I915_MEMORY_CLASS_DEVICE, and only 3633 * strictly required on configurations where some subset of the device 3634 * memory is directly visible/mappable through the CPU (which we also 3635 * call small BAR), like on some DG2+ systems. Note that this is quite 3636 * undesirable, but due to various factors like the client CPU, BIOS etc 3637 * it's something we can expect to see in the wild. See 3638 * &drm_i915_memory_region_info.probed_cpu_visible_size for how to 3639 * determine if this system applies. 3640 * 3641 * Note that one of the placements MUST be I915_MEMORY_CLASS_SYSTEM, to 3642 * ensure the kernel can always spill the allocation to system memory, 3643 * if the object can't be allocated in the mappable part of 3644 * I915_MEMORY_CLASS_DEVICE. 3645 * 3646 * Also note that since the kernel only supports flat-CCS on objects 3647 * that can *only* be placed in I915_MEMORY_CLASS_DEVICE, we therefore 3648 * don't support I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS together with 3649 * flat-CCS. 3650 * 3651 * Without this hint, the kernel will assume that non-mappable 3652 * I915_MEMORY_CLASS_DEVICE is preferred for this object. Note that the 3653 * kernel can still migrate the object to the mappable part, as a last 3654 * resort, if userspace ever CPU faults this object, but this might be 3655 * expensive, and so ideally should be avoided. 3656 * 3657 * On older kernels which lack the relevant small-bar uAPI support (see 3658 * also &drm_i915_memory_region_info.probed_cpu_visible_size), 3659 * usage of the flag will result in an error, but it should NEVER be 3660 * possible to end up with a small BAR configuration, assuming we can 3661 * also successfully load the i915 kernel module. In such cases the 3662 * entire I915_MEMORY_CLASS_DEVICE region will be CPU accessible, and as 3663 * such there are zero restrictions on where the object can be placed. 3664 */ 3665 #define I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS (1 << 0) 3666 __u32 flags; 3667 3668 /** 3669 * @extensions: The chain of extensions to apply to this object. 3670 * 3671 * This will be useful in the future when we need to support several 3672 * different extensions, and we need to apply more than one when 3673 * creating the object. See struct i915_user_extension. 3674 * 3675 * If we don't supply any extensions then we get the same old gem_create 3676 * behaviour. 3677 * 3678 * For I915_GEM_CREATE_EXT_MEMORY_REGIONS usage see 3679 * struct drm_i915_gem_create_ext_memory_regions. 3680 * 3681 * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see 3682 * struct drm_i915_gem_create_ext_protected_content. 3683 * 3684 * For I915_GEM_CREATE_EXT_SET_PAT usage see 3685 * struct drm_i915_gem_create_ext_set_pat. 3686 */ 3687 #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 3688 #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 3689 #define I915_GEM_CREATE_EXT_SET_PAT 2 3690 __u64 extensions; 3691 }; 3692 3693 /** 3694 * struct drm_i915_gem_create_ext_memory_regions - The 3695 * I915_GEM_CREATE_EXT_MEMORY_REGIONS extension. 3696 * 3697 * Set the object with the desired set of placements/regions in priority 3698 * order. Each entry must be unique and supported by the device. 3699 * 3700 * This is provided as an array of struct drm_i915_gem_memory_class_instance, or 3701 * an equivalent layout of class:instance pair encodings. See struct 3702 * drm_i915_query_memory_regions and DRM_I915_QUERY_MEMORY_REGIONS for how to 3703 * query the supported regions for a device. 3704 * 3705 * As an example, on discrete devices, if we wish to set the placement as 3706 * device local-memory we can do something like: 3707 * 3708 * .. code-block:: C 3709 * 3710 * struct drm_i915_gem_memory_class_instance region_lmem = { 3711 * .memory_class = I915_MEMORY_CLASS_DEVICE, 3712 * .memory_instance = 0, 3713 * }; 3714 * struct drm_i915_gem_create_ext_memory_regions regions = { 3715 * .base = { .name = I915_GEM_CREATE_EXT_MEMORY_REGIONS }, 3716 * .regions = (uintptr_t)®ion_lmem, 3717 * .num_regions = 1, 3718 * }; 3719 * struct drm_i915_gem_create_ext create_ext = { 3720 * .size = 16 * PAGE_SIZE, 3721 * .extensions = (uintptr_t)®ions, 3722 * }; 3723 * 3724 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3725 * if (err) ... 3726 * 3727 * At which point we get the object handle in &drm_i915_gem_create_ext.handle, 3728 * along with the final object size in &drm_i915_gem_create_ext.size, which 3729 * should account for any rounding up, if required. 3730 * 3731 * Note that userspace has no means of knowing the current backing region 3732 * for objects where @num_regions is larger than one. The kernel will only 3733 * ensure that the priority order of the @regions array is honoured, either 3734 * when initially placing the object, or when moving memory around due to 3735 * memory pressure 3736 * 3737 * On Flat-CCS capable HW, compression is supported for the objects residing 3738 * in I915_MEMORY_CLASS_DEVICE. When such objects (compressed) have other 3739 * memory class in @regions and migrated (by i915, due to memory 3740 * constraints) to the non I915_MEMORY_CLASS_DEVICE region, then i915 needs to 3741 * decompress the content. But i915 doesn't have the required information to 3742 * decompress the userspace compressed objects. 3743 * 3744 * So i915 supports Flat-CCS, on the objects which can reside only on 3745 * I915_MEMORY_CLASS_DEVICE regions. 3746 */ 3747 struct drm_i915_gem_create_ext_memory_regions { 3748 /** @base: Extension link. See struct i915_user_extension. */ 3749 struct i915_user_extension base; 3750 3751 /** @pad: MBZ */ 3752 __u32 pad; 3753 /** @num_regions: Number of elements in the @regions array. */ 3754 __u32 num_regions; 3755 /** 3756 * @regions: The regions/placements array. 3757 * 3758 * An array of struct drm_i915_gem_memory_class_instance. 3759 */ 3760 __u64 regions; 3761 }; 3762 3763 /** 3764 * struct drm_i915_gem_create_ext_protected_content - The 3765 * I915_OBJECT_PARAM_PROTECTED_CONTENT extension. 3766 * 3767 * If this extension is provided, buffer contents are expected to be protected 3768 * by PXP encryption and require decryption for scan out and processing. This 3769 * is only possible on platforms that have PXP enabled, on all other scenarios 3770 * using this extension will cause the ioctl to fail and return -ENODEV. The 3771 * flags parameter is reserved for future expansion and must currently be set 3772 * to zero. 3773 * 3774 * The buffer contents are considered invalid after a PXP session teardown. 3775 * 3776 * The encryption is guaranteed to be processed correctly only if the object 3777 * is submitted with a context created using the 3778 * I915_CONTEXT_PARAM_PROTECTED_CONTENT flag. This will also enable extra checks 3779 * at submission time on the validity of the objects involved. 3780 * 3781 * Below is an example on how to create a protected object: 3782 * 3783 * .. code-block:: C 3784 * 3785 * struct drm_i915_gem_create_ext_protected_content protected_ext = { 3786 * .base = { .name = I915_GEM_CREATE_EXT_PROTECTED_CONTENT }, 3787 * .flags = 0, 3788 * }; 3789 * struct drm_i915_gem_create_ext create_ext = { 3790 * .size = PAGE_SIZE, 3791 * .extensions = (uintptr_t)&protected_ext, 3792 * }; 3793 * 3794 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3795 * if (err) ... 3796 */ 3797 struct drm_i915_gem_create_ext_protected_content { 3798 /** @base: Extension link. See struct i915_user_extension. */ 3799 struct i915_user_extension base; 3800 /** @flags: reserved for future usage, currently MBZ */ 3801 __u32 flags; 3802 }; 3803 3804 /** 3805 * struct drm_i915_gem_create_ext_set_pat - The 3806 * I915_GEM_CREATE_EXT_SET_PAT extension. 3807 * 3808 * If this extension is provided, the specified caching policy (PAT index) is 3809 * applied to the buffer object. 3810 * 3811 * Below is an example on how to create an object with specific caching policy: 3812 * 3813 * .. code-block:: C 3814 * 3815 * struct drm_i915_gem_create_ext_set_pat set_pat_ext = { 3816 * .base = { .name = I915_GEM_CREATE_EXT_SET_PAT }, 3817 * .pat_index = 0, 3818 * }; 3819 * struct drm_i915_gem_create_ext create_ext = { 3820 * .size = PAGE_SIZE, 3821 * .extensions = (uintptr_t)&set_pat_ext, 3822 * }; 3823 * 3824 * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); 3825 * if (err) ... 3826 */ 3827 struct drm_i915_gem_create_ext_set_pat { 3828 /** @base: Extension link. See struct i915_user_extension. */ 3829 struct i915_user_extension base; 3830 /** 3831 * @pat_index: PAT index to be set 3832 * PAT index is a bit field in Page Table Entry to control caching 3833 * behaviors for GPU accesses. The definition of PAT index is 3834 * platform dependent and can be found in hardware specifications, 3835 */ 3836 __u32 pat_index; 3837 /** @rsvd: reserved for future use */ 3838 __u32 rsvd; 3839 }; 3840 3841 /* ID of the protected content session managed by i915 when PXP is active */ 3842 #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf 3843 3844 #if defined(__cplusplus) 3845 } 3846 #endif 3847 3848 #endif /* _UAPI_I915_DRM_H_ */ 3849