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