| File: | dev/pci/drm/i915/gem/i915_gem_object.h |
| Warning: | line 130, column 17 Access to field 'contended' results in a dereference of a null pointer (loaded from variable 'ww') |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* | ||||
| 2 | * SPDX-License-Identifier: MIT | ||||
| 3 | * | ||||
| 4 | * Copyright © 2014-2016 Intel Corporation | ||||
| 5 | */ | ||||
| 6 | |||||
| 7 | #include "display/intel_frontbuffer.h" | ||||
| 8 | |||||
| 9 | #include "i915_drv.h" | ||||
| 10 | #include "i915_gem_clflush.h" | ||||
| 11 | #include "i915_gem_gtt.h" | ||||
| 12 | #include "i915_gem_ioctls.h" | ||||
| 13 | #include "i915_gem_object.h" | ||||
| 14 | #include "i915_vma.h" | ||||
| 15 | #include "i915_gem_lmem.h" | ||||
| 16 | #include "i915_gem_mman.h" | ||||
| 17 | |||||
| 18 | static void __i915_gem_object_flush_for_display(struct drm_i915_gem_object *obj) | ||||
| 19 | { | ||||
| 20 | /* | ||||
| 21 | * We manually flush the CPU domain so that we can override and | ||||
| 22 | * force the flush for the display, and perform it asyncrhonously. | ||||
| 23 | */ | ||||
| 24 | i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU0x00000001); | ||||
| 25 | if (obj->cache_dirty) | ||||
| 26 | i915_gem_clflush_object(obj, I915_CLFLUSH_FORCE(1UL << (0))); | ||||
| 27 | obj->write_domain = 0; | ||||
| 28 | } | ||||
| 29 | |||||
| 30 | void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj) | ||||
| 31 | { | ||||
| 32 | if (!i915_gem_object_is_framebuffer(obj)) | ||||
| 33 | return; | ||||
| 34 | |||||
| 35 | i915_gem_object_lock(obj, NULL((void *)0)); | ||||
| 36 | __i915_gem_object_flush_for_display(obj); | ||||
| 37 | i915_gem_object_unlock(obj); | ||||
| 38 | } | ||||
| 39 | |||||
| 40 | void i915_gem_object_flush_if_display_locked(struct drm_i915_gem_object *obj) | ||||
| 41 | { | ||||
| 42 | if (i915_gem_object_is_framebuffer(obj)) | ||||
| 43 | __i915_gem_object_flush_for_display(obj); | ||||
| 44 | } | ||||
| 45 | |||||
| 46 | /** | ||||
| 47 | * Moves a single object to the WC read, and possibly write domain. | ||||
| 48 | * @obj: object to act on | ||||
| 49 | * @write: ask for write access or read only | ||||
| 50 | * | ||||
| 51 | * This function returns when the move is complete, including waiting on | ||||
| 52 | * flushes to occur. | ||||
| 53 | */ | ||||
| 54 | int | ||||
| 55 | i915_gem_object_set_to_wc_domain(struct drm_i915_gem_object *obj, bool_Bool write) | ||||
| 56 | { | ||||
| 57 | int ret; | ||||
| 58 | |||||
| 59 | assert_object_held(obj)do { (void)(&((obj)->base.resv)->lock.base); } while (0); | ||||
| 60 | |||||
| 61 | ret = i915_gem_object_wait(obj, | ||||
| 62 | I915_WAIT_INTERRUPTIBLE(1UL << (0)) | | ||||
| 63 | (write ? I915_WAIT_ALL(1UL << (2)) : 0), | ||||
| 64 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 65 | if (ret) | ||||
| 66 | return ret; | ||||
| 67 | |||||
| 68 | if (obj->write_domain == I915_GEM_DOMAIN_WC0x00000080) | ||||
| 69 | return 0; | ||||
| 70 | |||||
| 71 | /* Flush and acquire obj->pages so that we are coherent through | ||||
| 72 | * direct access in memory with previous cached writes through | ||||
| 73 | * shmemfs and that our cache domain tracking remains valid. | ||||
| 74 | * For example, if the obj->filp was moved to swap without us | ||||
| 75 | * being notified and releasing the pages, we would mistakenly | ||||
| 76 | * continue to assume that the obj remained out of the CPU cached | ||||
| 77 | * domain. | ||||
| 78 | */ | ||||
| 79 | ret = i915_gem_object_pin_pages(obj); | ||||
| 80 | if (ret) | ||||
| 81 | return ret; | ||||
| 82 | |||||
| 83 | i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_WC0x00000080); | ||||
| 84 | |||||
| 85 | /* Serialise direct access to this object with the barriers for | ||||
| 86 | * coherent writes from the GPU, by effectively invalidating the | ||||
| 87 | * WC domain upon first access. | ||||
| 88 | */ | ||||
| 89 | if ((obj->read_domains & I915_GEM_DOMAIN_WC0x00000080) == 0) | ||||
| 90 | mb()do { __asm volatile("mfence" ::: "memory"); } while (0); | ||||
| 91 | |||||
| 92 | /* It should now be out of any other write domains, and we can update | ||||
| 93 | * the domain values for our changes. | ||||
| 94 | */ | ||||
| 95 | GEM_BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_WC) != 0)((void)0); | ||||
| 96 | obj->read_domains |= I915_GEM_DOMAIN_WC0x00000080; | ||||
| 97 | if (write) { | ||||
| 98 | obj->read_domains = I915_GEM_DOMAIN_WC0x00000080; | ||||
| 99 | obj->write_domain = I915_GEM_DOMAIN_WC0x00000080; | ||||
| 100 | obj->mm.dirty = true1; | ||||
| 101 | } | ||||
| 102 | |||||
| 103 | i915_gem_object_unpin_pages(obj); | ||||
| 104 | return 0; | ||||
| 105 | } | ||||
| 106 | |||||
| 107 | /** | ||||
| 108 | * Moves a single object to the GTT read, and possibly write domain. | ||||
| 109 | * @obj: object to act on | ||||
| 110 | * @write: ask for write access or read only | ||||
| 111 | * | ||||
| 112 | * This function returns when the move is complete, including waiting on | ||||
| 113 | * flushes to occur. | ||||
| 114 | */ | ||||
| 115 | int | ||||
| 116 | i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool_Bool write) | ||||
| 117 | { | ||||
| 118 | int ret; | ||||
| 119 | |||||
| 120 | assert_object_held(obj)do { (void)(&((obj)->base.resv)->lock.base); } while (0); | ||||
| 121 | |||||
| 122 | ret = i915_gem_object_wait(obj, | ||||
| 123 | I915_WAIT_INTERRUPTIBLE(1UL << (0)) | | ||||
| 124 | (write ? I915_WAIT_ALL(1UL << (2)) : 0), | ||||
| 125 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 126 | if (ret) | ||||
| 127 | return ret; | ||||
| 128 | |||||
| 129 | if (obj->write_domain == I915_GEM_DOMAIN_GTT0x00000040) | ||||
| 130 | return 0; | ||||
| 131 | |||||
| 132 | /* Flush and acquire obj->pages so that we are coherent through | ||||
| 133 | * direct access in memory with previous cached writes through | ||||
| 134 | * shmemfs and that our cache domain tracking remains valid. | ||||
| 135 | * For example, if the obj->filp was moved to swap without us | ||||
| 136 | * being notified and releasing the pages, we would mistakenly | ||||
| 137 | * continue to assume that the obj remained out of the CPU cached | ||||
| 138 | * domain. | ||||
| 139 | */ | ||||
| 140 | ret = i915_gem_object_pin_pages(obj); | ||||
| 141 | if (ret) | ||||
| 142 | return ret; | ||||
| 143 | |||||
| 144 | i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_GTT0x00000040); | ||||
| 145 | |||||
| 146 | /* Serialise direct access to this object with the barriers for | ||||
| 147 | * coherent writes from the GPU, by effectively invalidating the | ||||
| 148 | * GTT domain upon first access. | ||||
| 149 | */ | ||||
| 150 | if ((obj->read_domains & I915_GEM_DOMAIN_GTT0x00000040) == 0) | ||||
| 151 | mb()do { __asm volatile("mfence" ::: "memory"); } while (0); | ||||
| 152 | |||||
| 153 | /* It should now be out of any other write domains, and we can update | ||||
| 154 | * the domain values for our changes. | ||||
| 155 | */ | ||||
| 156 | GEM_BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_GTT) != 0)((void)0); | ||||
| 157 | obj->read_domains |= I915_GEM_DOMAIN_GTT0x00000040; | ||||
| 158 | if (write) { | ||||
| 159 | struct i915_vma *vma; | ||||
| 160 | |||||
| 161 | obj->read_domains = I915_GEM_DOMAIN_GTT0x00000040; | ||||
| 162 | obj->write_domain = I915_GEM_DOMAIN_GTT0x00000040; | ||||
| 163 | obj->mm.dirty = true1; | ||||
| 164 | |||||
| 165 | spin_lock(&obj->vma.lock)mtx_enter(&obj->vma.lock); | ||||
| 166 | for_each_ggtt_vma(vma, obj)for (vma = ({ const __typeof( ((__typeof(*vma) *)0)->obj_link ) *__mptr = ((&(obj)->vma.list)->next); (__typeof( *vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), obj_link) );}); &vma->obj_link != (&(obj)->vma .list); vma = ({ const __typeof( ((__typeof(*vma) *)0)->obj_link ) *__mptr = (vma->obj_link.next); (__typeof(*vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), obj_link) );}) ) if (!i915_vma_is_ggtt(vma)) break; else | ||||
| 167 | if (i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND((int)(1UL << (10))))) | ||||
| 168 | i915_vma_set_ggtt_write(vma); | ||||
| 169 | spin_unlock(&obj->vma.lock)mtx_leave(&obj->vma.lock); | ||||
| 170 | } | ||||
| 171 | |||||
| 172 | i915_gem_object_unpin_pages(obj); | ||||
| 173 | return 0; | ||||
| 174 | } | ||||
| 175 | |||||
| 176 | /** | ||||
| 177 | * Changes the cache-level of an object across all VMA. | ||||
| 178 | * @obj: object to act on | ||||
| 179 | * @cache_level: new cache level to set for the object | ||||
| 180 | * | ||||
| 181 | * After this function returns, the object will be in the new cache-level | ||||
| 182 | * across all GTT and the contents of the backing storage will be coherent, | ||||
| 183 | * with respect to the new cache-level. In order to keep the backing storage | ||||
| 184 | * coherent for all users, we only allow a single cache level to be set | ||||
| 185 | * globally on the object and prevent it from being changed whilst the | ||||
| 186 | * hardware is reading from the object. That is if the object is currently | ||||
| 187 | * on the scanout it will be set to uncached (or equivalent display | ||||
| 188 | * cache coherency) and all non-MOCS GPU access will also be uncached so | ||||
| 189 | * that all direct access to the scanout remains coherent. | ||||
| 190 | */ | ||||
| 191 | int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, | ||||
| 192 | enum i915_cache_level cache_level) | ||||
| 193 | { | ||||
| 194 | int ret; | ||||
| 195 | |||||
| 196 | if (obj->cache_level == cache_level) | ||||
| 197 | return 0; | ||||
| 198 | |||||
| 199 | ret = i915_gem_object_wait(obj, | ||||
| 200 | I915_WAIT_INTERRUPTIBLE(1UL << (0)) | | ||||
| 201 | I915_WAIT_ALL(1UL << (2)), | ||||
| 202 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 203 | if (ret) | ||||
| 204 | return ret; | ||||
| 205 | |||||
| 206 | /* Always invalidate stale cachelines */ | ||||
| 207 | if (obj->cache_level != cache_level) { | ||||
| 208 | i915_gem_object_set_cache_coherency(obj, cache_level); | ||||
| 209 | obj->cache_dirty = true1; | ||||
| 210 | } | ||||
| 211 | |||||
| 212 | /* The cache-level will be applied when each vma is rebound. */ | ||||
| 213 | return i915_gem_object_unbind(obj, | ||||
| 214 | I915_GEM_OBJECT_UNBIND_ACTIVE(1UL << (0)) | | ||||
| 215 | I915_GEM_OBJECT_UNBIND_BARRIER(1UL << (1))); | ||||
| 216 | } | ||||
| 217 | |||||
| 218 | int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, | ||||
| 219 | struct drm_file *file) | ||||
| 220 | { | ||||
| 221 | struct drm_i915_gem_caching *args = data; | ||||
| 222 | struct drm_i915_gem_object *obj; | ||||
| 223 | int err = 0; | ||||
| 224 | |||||
| 225 | rcu_read_lock(); | ||||
| 226 | obj = i915_gem_object_lookup_rcu(file, args->handle); | ||||
| 227 | if (!obj) { | ||||
| 228 | err = -ENOENT2; | ||||
| 229 | goto out; | ||||
| 230 | } | ||||
| 231 | |||||
| 232 | switch (obj->cache_level) { | ||||
| 233 | case I915_CACHE_LLC: | ||||
| 234 | case I915_CACHE_L3_LLC: | ||||
| 235 | args->caching = I915_CACHING_CACHED1; | ||||
| 236 | break; | ||||
| 237 | |||||
| 238 | case I915_CACHE_WT: | ||||
| 239 | args->caching = I915_CACHING_DISPLAY2; | ||||
| 240 | break; | ||||
| 241 | |||||
| 242 | default: | ||||
| 243 | args->caching = I915_CACHING_NONE0; | ||||
| 244 | break; | ||||
| 245 | } | ||||
| 246 | out: | ||||
| 247 | rcu_read_unlock(); | ||||
| 248 | return err; | ||||
| 249 | } | ||||
| 250 | |||||
| 251 | int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data, | ||||
| 252 | struct drm_file *file) | ||||
| 253 | { | ||||
| 254 | struct drm_i915_privateinteldrm_softc *i915 = to_i915(dev); | ||||
| 255 | struct drm_i915_gem_caching *args = data; | ||||
| 256 | struct drm_i915_gem_object *obj; | ||||
| 257 | enum i915_cache_level level; | ||||
| 258 | int ret = 0; | ||||
| 259 | |||||
| 260 | switch (args->caching) { | ||||
| 261 | case I915_CACHING_NONE0: | ||||
| 262 | level = I915_CACHE_NONE; | ||||
| 263 | break; | ||||
| 264 | case I915_CACHING_CACHED1: | ||||
| 265 | /* | ||||
| 266 | * Due to a HW issue on BXT A stepping, GPU stores via a | ||||
| 267 | * snooped mapping may leave stale data in a corresponding CPU | ||||
| 268 | * cacheline, whereas normally such cachelines would get | ||||
| 269 | * invalidated. | ||||
| 270 | */ | ||||
| 271 | if (!HAS_LLC(i915)((&(i915)->__info)->has_llc) && !HAS_SNOOP(i915)((&(i915)->__info)->has_snoop)) | ||||
| 272 | return -ENODEV19; | ||||
| 273 | |||||
| 274 | level = I915_CACHE_LLC; | ||||
| 275 | break; | ||||
| 276 | case I915_CACHING_DISPLAY2: | ||||
| 277 | level = HAS_WT(i915)((IS_PLATFORM(i915, INTEL_HASWELL) || IS_PLATFORM(i915, INTEL_BROADWELL )) && ((i915)->edram_size_mb)) ? I915_CACHE_WT : I915_CACHE_NONE; | ||||
| 278 | break; | ||||
| 279 | default: | ||||
| 280 | return -EINVAL22; | ||||
| 281 | } | ||||
| 282 | |||||
| 283 | obj = i915_gem_object_lookup(file, args->handle); | ||||
| 284 | if (!obj) | ||||
| 285 | return -ENOENT2; | ||||
| 286 | |||||
| 287 | /* | ||||
| 288 | * The caching mode of proxy object is handled by its generator, and | ||||
| 289 | * not allowed to be changed by userspace. | ||||
| 290 | */ | ||||
| 291 | if (i915_gem_object_is_proxy(obj)) { | ||||
| 292 | ret = -ENXIO6; | ||||
| 293 | goto out; | ||||
| 294 | } | ||||
| 295 | |||||
| 296 | ret = i915_gem_object_lock_interruptible(obj, NULL((void *)0)); | ||||
| 297 | if (ret) | ||||
| 298 | goto out; | ||||
| 299 | |||||
| 300 | ret = i915_gem_object_set_cache_level(obj, level); | ||||
| 301 | i915_gem_object_unlock(obj); | ||||
| 302 | |||||
| 303 | out: | ||||
| 304 | i915_gem_object_put(obj); | ||||
| 305 | return ret; | ||||
| 306 | } | ||||
| 307 | |||||
| 308 | /* | ||||
| 309 | * Prepare buffer for display plane (scanout, cursors, etc). Can be called from | ||||
| 310 | * an uninterruptible phase (modesetting) and allows any flushes to be pipelined | ||||
| 311 | * (for pageflips). We only flush the caches while preparing the buffer for | ||||
| 312 | * display, the callers are responsible for frontbuffer flush. | ||||
| 313 | */ | ||||
| 314 | struct i915_vma * | ||||
| 315 | i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, | ||||
| 316 | u32 alignment, | ||||
| 317 | const struct i915_ggtt_view *view, | ||||
| 318 | unsigned int flags) | ||||
| 319 | { | ||||
| 320 | struct drm_i915_privateinteldrm_softc *i915 = to_i915(obj->base.dev); | ||||
| 321 | struct i915_gem_ww_ctx ww; | ||||
| 322 | struct i915_vma *vma; | ||||
| 323 | int ret; | ||||
| 324 | |||||
| 325 | /* Frame buffer must be in LMEM (no migration yet) */ | ||||
| 326 | if (HAS_LMEM(i915)((&(i915)->__info)->memory_regions & ((1UL << (INTEL_REGION_LMEM)))) && !i915_gem_object_is_lmem(obj)) | ||||
| 327 | return ERR_PTR(-EINVAL22); | ||||
| 328 | |||||
| 329 | i915_gem_ww_ctx_init(&ww, true1); | ||||
| 330 | retry: | ||||
| 331 | ret = i915_gem_object_lock(obj, &ww); | ||||
| 332 | if (ret) | ||||
| 333 | goto err; | ||||
| 334 | /* | ||||
| 335 | * The display engine is not coherent with the LLC cache on gen6. As | ||||
| 336 | * a result, we make sure that the pinning that is about to occur is | ||||
| 337 | * done with uncached PTEs. This is lowest common denominator for all | ||||
| 338 | * chipsets. | ||||
| 339 | * | ||||
| 340 | * However for gen6+, we could do better by using the GFDT bit instead | ||||
| 341 | * of uncaching, which would allow us to flush all the LLC-cached data | ||||
| 342 | * with that bit in the PTE to main memory with just one PIPE_CONTROL. | ||||
| 343 | */ | ||||
| 344 | ret = i915_gem_object_set_cache_level(obj, | ||||
| 345 | HAS_WT(i915)((IS_PLATFORM(i915, INTEL_HASWELL) || IS_PLATFORM(i915, INTEL_BROADWELL )) && ((i915)->edram_size_mb)) ? | ||||
| 346 | I915_CACHE_WT : I915_CACHE_NONE); | ||||
| 347 | if (ret) | ||||
| 348 | goto err; | ||||
| 349 | |||||
| 350 | /* | ||||
| 351 | * As the user may map the buffer once pinned in the display plane | ||||
| 352 | * (e.g. libkms for the bootup splash), we have to ensure that we | ||||
| 353 | * always use map_and_fenceable for all scanout buffers. However, | ||||
| 354 | * it may simply be too big to fit into mappable, in which case | ||||
| 355 | * put it anyway and hope that userspace can cope (but always first | ||||
| 356 | * try to preserve the existing ABI). | ||||
| 357 | */ | ||||
| 358 | vma = ERR_PTR(-ENOSPC28); | ||||
| 359 | if ((flags & PIN_MAPPABLE(1ULL << (3))) == 0 && | ||||
| 360 | (!view || view->type == I915_GGTT_VIEW_NORMAL)) | ||||
| 361 | vma = i915_gem_object_ggtt_pin_ww(obj, &ww, view, 0, alignment, | ||||
| 362 | flags | PIN_MAPPABLE(1ULL << (3)) | | ||||
| 363 | PIN_NONBLOCK(1ULL << (2))); | ||||
| 364 | if (IS_ERR(vma) && vma != ERR_PTR(-EDEADLK11)) | ||||
| 365 | vma = i915_gem_object_ggtt_pin_ww(obj, &ww, view, 0, | ||||
| 366 | alignment, flags); | ||||
| 367 | if (IS_ERR(vma)) { | ||||
| 368 | ret = PTR_ERR(vma); | ||||
| 369 | goto err; | ||||
| 370 | } | ||||
| 371 | |||||
| 372 | vma->display_alignment = max_t(u64, vma->display_alignment, alignment)({ u64 __max_a = (vma->display_alignment); u64 __max_b = ( alignment); __max_a > __max_b ? __max_a : __max_b; }); | ||||
| 373 | |||||
| 374 | i915_gem_object_flush_if_display_locked(obj); | ||||
| 375 | |||||
| 376 | err: | ||||
| 377 | if (ret == -EDEADLK11) { | ||||
| 378 | ret = i915_gem_ww_ctx_backoff(&ww); | ||||
| 379 | if (!ret) | ||||
| 380 | goto retry; | ||||
| 381 | } | ||||
| 382 | i915_gem_ww_ctx_fini(&ww); | ||||
| 383 | |||||
| 384 | if (ret) | ||||
| 385 | return ERR_PTR(ret); | ||||
| 386 | |||||
| 387 | return vma; | ||||
| 388 | } | ||||
| 389 | |||||
| 390 | /** | ||||
| 391 | * Moves a single object to the CPU read, and possibly write domain. | ||||
| 392 | * @obj: object to act on | ||||
| 393 | * @write: requesting write or read-only access | ||||
| 394 | * | ||||
| 395 | * This function returns when the move is complete, including waiting on | ||||
| 396 | * flushes to occur. | ||||
| 397 | */ | ||||
| 398 | int | ||||
| 399 | i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool_Bool write) | ||||
| 400 | { | ||||
| 401 | int ret; | ||||
| 402 | |||||
| 403 | assert_object_held(obj)do { (void)(&((obj)->base.resv)->lock.base); } while (0); | ||||
| 404 | |||||
| 405 | ret = i915_gem_object_wait(obj, | ||||
| 406 | I915_WAIT_INTERRUPTIBLE(1UL << (0)) | | ||||
| 407 | (write ? I915_WAIT_ALL(1UL << (2)) : 0), | ||||
| 408 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 409 | if (ret) | ||||
| 410 | return ret; | ||||
| 411 | |||||
| 412 | i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU0x00000001); | ||||
| 413 | |||||
| 414 | /* Flush the CPU cache if it's still invalid. */ | ||||
| 415 | if ((obj->read_domains & I915_GEM_DOMAIN_CPU0x00000001) == 0) { | ||||
| 416 | i915_gem_clflush_object(obj, I915_CLFLUSH_SYNC(1UL << (1))); | ||||
| 417 | obj->read_domains |= I915_GEM_DOMAIN_CPU0x00000001; | ||||
| 418 | } | ||||
| 419 | |||||
| 420 | /* It should now be out of any other write domains, and we can update | ||||
| 421 | * the domain values for our changes. | ||||
| 422 | */ | ||||
| 423 | GEM_BUG_ON(obj->write_domain & ~I915_GEM_DOMAIN_CPU)((void)0); | ||||
| 424 | |||||
| 425 | /* If we're writing through the CPU, then the GPU read domains will | ||||
| 426 | * need to be invalidated at next use. | ||||
| 427 | */ | ||||
| 428 | if (write) | ||||
| 429 | __start_cpu_write(obj); | ||||
| 430 | |||||
| 431 | return 0; | ||||
| 432 | } | ||||
| 433 | |||||
| 434 | /** | ||||
| 435 | * Called when user space prepares to use an object with the CPU, either | ||||
| 436 | * through the mmap ioctl's mapping or a GTT mapping. | ||||
| 437 | * @dev: drm device | ||||
| 438 | * @data: ioctl data blob | ||||
| 439 | * @file: drm file | ||||
| 440 | */ | ||||
| 441 | int | ||||
| 442 | i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, | ||||
| 443 | struct drm_file *file) | ||||
| 444 | { | ||||
| 445 | struct drm_i915_gem_set_domain *args = data; | ||||
| 446 | struct drm_i915_gem_object *obj; | ||||
| 447 | u32 read_domains = args->read_domains; | ||||
| 448 | u32 write_domain = args->write_domain; | ||||
| 449 | int err; | ||||
| 450 | |||||
| 451 | /* Only handle setting domains to types used by the CPU. */ | ||||
| 452 | if ((write_domain | read_domains) & I915_GEM_GPU_DOMAINS(0x00000002 | 0x00000004 | 0x00000008 | 0x00000010 | 0x00000020 )) | ||||
| |||||
| 453 | return -EINVAL22; | ||||
| 454 | |||||
| 455 | /* | ||||
| 456 | * Having something in the write domain implies it's in the read | ||||
| 457 | * domain, and only that read domain. Enforce that in the request. | ||||
| 458 | */ | ||||
| 459 | if (write_domain && read_domains != write_domain) | ||||
| 460 | return -EINVAL22; | ||||
| 461 | |||||
| 462 | if (!read_domains) | ||||
| 463 | return 0; | ||||
| 464 | |||||
| 465 | obj = i915_gem_object_lookup(file, args->handle); | ||||
| 466 | if (!obj
| ||||
| 467 | return -ENOENT2; | ||||
| 468 | |||||
| 469 | /* | ||||
| 470 | * Try to flush the object off the GPU without holding the lock. | ||||
| 471 | * We will repeat the flush holding the lock in the normal manner | ||||
| 472 | * to catch cases where we are gazumped. | ||||
| 473 | */ | ||||
| 474 | err = i915_gem_object_wait(obj, | ||||
| 475 | I915_WAIT_INTERRUPTIBLE(1UL << (0)) | | ||||
| 476 | I915_WAIT_PRIORITY(1UL << (1)) | | ||||
| 477 | (write_domain
| ||||
| 478 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 479 | if (err) | ||||
| 480 | goto out; | ||||
| 481 | |||||
| 482 | /* | ||||
| 483 | * Proxy objects do not control access to the backing storage, ergo | ||||
| 484 | * they cannot be used as a means to manipulate the cache domain | ||||
| 485 | * tracking for that backing storage. The proxy object is always | ||||
| 486 | * considered to be outside of any cache domain. | ||||
| 487 | */ | ||||
| 488 | if (i915_gem_object_is_proxy(obj)) { | ||||
| 489 | err = -ENXIO6; | ||||
| 490 | goto out; | ||||
| 491 | } | ||||
| 492 | |||||
| 493 | /* | ||||
| 494 | * Flush and acquire obj->pages so that we are coherent through | ||||
| 495 | * direct access in memory with previous cached writes through | ||||
| 496 | * shmemfs and that our cache domain tracking remains valid. | ||||
| 497 | * For example, if the obj->filp was moved to swap without us | ||||
| 498 | * being notified and releasing the pages, we would mistakenly | ||||
| 499 | * continue to assume that the obj remained out of the CPU cached | ||||
| 500 | * domain. | ||||
| 501 | */ | ||||
| 502 | err = i915_gem_object_pin_pages(obj); | ||||
| 503 | if (err
| ||||
| 504 | goto out; | ||||
| 505 | |||||
| 506 | /* | ||||
| 507 | * Already in the desired write domain? Nothing for us to do! | ||||
| 508 | * | ||||
| 509 | * We apply a little bit of cunning here to catch a broader set of | ||||
| 510 | * no-ops. If obj->write_domain is set, we must be in the same | ||||
| 511 | * obj->read_domains, and only that domain. Therefore, if that | ||||
| 512 | * obj->write_domain matches the request read_domains, we are | ||||
| 513 | * already in the same read/write domain and can skip the operation, | ||||
| 514 | * without having to further check the requested write_domain. | ||||
| 515 | */ | ||||
| 516 | if (READ_ONCE(obj->write_domain)({ typeof(obj->write_domain) __tmp = *(volatile typeof(obj ->write_domain) *)&(obj->write_domain); membar_datadep_consumer (); __tmp; }) == read_domains) | ||||
| 517 | goto out_unpin; | ||||
| 518 | |||||
| 519 | err = i915_gem_object_lock_interruptible(obj, NULL((void *)0)); | ||||
| 520 | if (err) | ||||
| 521 | goto out_unpin; | ||||
| 522 | |||||
| 523 | if (read_domains & I915_GEM_DOMAIN_WC0x00000080) | ||||
| 524 | err = i915_gem_object_set_to_wc_domain(obj, write_domain); | ||||
| 525 | else if (read_domains & I915_GEM_DOMAIN_GTT0x00000040) | ||||
| 526 | err = i915_gem_object_set_to_gtt_domain(obj, write_domain); | ||||
| 527 | else | ||||
| 528 | err = i915_gem_object_set_to_cpu_domain(obj, write_domain); | ||||
| 529 | |||||
| 530 | i915_gem_object_unlock(obj); | ||||
| 531 | |||||
| 532 | if (write_domain) | ||||
| 533 | i915_gem_object_invalidate_frontbuffer(obj, ORIGIN_CPU); | ||||
| 534 | |||||
| 535 | out_unpin: | ||||
| 536 | i915_gem_object_unpin_pages(obj); | ||||
| 537 | out: | ||||
| 538 | i915_gem_object_put(obj); | ||||
| 539 | return err; | ||||
| 540 | } | ||||
| 541 | |||||
| 542 | /* | ||||
| 543 | * Pins the specified object's pages and synchronizes the object with | ||||
| 544 | * GPU accesses. Sets needs_clflush to non-zero if the caller should | ||||
| 545 | * flush the object from the CPU cache. | ||||
| 546 | */ | ||||
| 547 | int i915_gem_object_prepare_read(struct drm_i915_gem_object *obj, | ||||
| 548 | unsigned int *needs_clflush) | ||||
| 549 | { | ||||
| 550 | int ret; | ||||
| 551 | |||||
| 552 | *needs_clflush = 0; | ||||
| 553 | if (!i915_gem_object_has_struct_page(obj)) | ||||
| 554 | return -ENODEV19; | ||||
| 555 | |||||
| 556 | assert_object_held(obj)do { (void)(&((obj)->base.resv)->lock.base); } while (0); | ||||
| 557 | |||||
| 558 | ret = i915_gem_object_wait(obj, | ||||
| 559 | I915_WAIT_INTERRUPTIBLE(1UL << (0)), | ||||
| 560 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 561 | if (ret) | ||||
| 562 | return ret; | ||||
| 563 | |||||
| 564 | ret = i915_gem_object_pin_pages(obj); | ||||
| 565 | if (ret) | ||||
| 566 | return ret; | ||||
| 567 | |||||
| 568 | if (obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_READ(1UL << (0)) || | ||||
| 569 | !static_cpu_has(X86_FEATURE_CLFLUSH1)) { | ||||
| 570 | ret = i915_gem_object_set_to_cpu_domain(obj, false0); | ||||
| 571 | if (ret) | ||||
| 572 | goto err_unpin; | ||||
| 573 | else | ||||
| 574 | goto out; | ||||
| 575 | } | ||||
| 576 | |||||
| 577 | i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU0x00000001); | ||||
| 578 | |||||
| 579 | /* If we're not in the cpu read domain, set ourself into the gtt | ||||
| 580 | * read domain and manually flush cachelines (if required). This | ||||
| 581 | * optimizes for the case when the gpu will dirty the data | ||||
| 582 | * anyway again before the next pread happens. | ||||
| 583 | */ | ||||
| 584 | if (!obj->cache_dirty && | ||||
| 585 | !(obj->read_domains & I915_GEM_DOMAIN_CPU0x00000001)) | ||||
| 586 | *needs_clflush = CLFLUSH_BEFORE(1UL << (0)); | ||||
| 587 | |||||
| 588 | out: | ||||
| 589 | /* return with the pages pinned */ | ||||
| 590 | return 0; | ||||
| 591 | |||||
| 592 | err_unpin: | ||||
| 593 | i915_gem_object_unpin_pages(obj); | ||||
| 594 | return ret; | ||||
| 595 | } | ||||
| 596 | |||||
| 597 | int i915_gem_object_prepare_write(struct drm_i915_gem_object *obj, | ||||
| 598 | unsigned int *needs_clflush) | ||||
| 599 | { | ||||
| 600 | int ret; | ||||
| 601 | |||||
| 602 | *needs_clflush = 0; | ||||
| 603 | if (!i915_gem_object_has_struct_page(obj)) | ||||
| 604 | return -ENODEV19; | ||||
| 605 | |||||
| 606 | assert_object_held(obj)do { (void)(&((obj)->base.resv)->lock.base); } while (0); | ||||
| 607 | |||||
| 608 | ret = i915_gem_object_wait(obj, | ||||
| 609 | I915_WAIT_INTERRUPTIBLE(1UL << (0)) | | ||||
| 610 | I915_WAIT_ALL(1UL << (2)), | ||||
| 611 | MAX_SCHEDULE_TIMEOUT(0x7fffffff)); | ||||
| 612 | if (ret) | ||||
| 613 | return ret; | ||||
| 614 | |||||
| 615 | ret = i915_gem_object_pin_pages(obj); | ||||
| 616 | if (ret) | ||||
| 617 | return ret; | ||||
| 618 | |||||
| 619 | if (obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE(1UL << (1)) || | ||||
| 620 | !static_cpu_has(X86_FEATURE_CLFLUSH1)) { | ||||
| 621 | ret = i915_gem_object_set_to_cpu_domain(obj, true1); | ||||
| 622 | if (ret) | ||||
| 623 | goto err_unpin; | ||||
| 624 | else | ||||
| 625 | goto out; | ||||
| 626 | } | ||||
| 627 | |||||
| 628 | i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU0x00000001); | ||||
| 629 | |||||
| 630 | /* If we're not in the cpu write domain, set ourself into the | ||||
| 631 | * gtt write domain and manually flush cachelines (as required). | ||||
| 632 | * This optimizes for the case when the gpu will use the data | ||||
| 633 | * right away and we therefore have to clflush anyway. | ||||
| 634 | */ | ||||
| 635 | if (!obj->cache_dirty) { | ||||
| 636 | *needs_clflush |= CLFLUSH_AFTER(1UL << (1)); | ||||
| 637 | |||||
| 638 | /* | ||||
| 639 | * Same trick applies to invalidate partially written | ||||
| 640 | * cachelines read before writing. | ||||
| 641 | */ | ||||
| 642 | if (!(obj->read_domains & I915_GEM_DOMAIN_CPU0x00000001)) | ||||
| 643 | *needs_clflush |= CLFLUSH_BEFORE(1UL << (0)); | ||||
| 644 | } | ||||
| 645 | |||||
| 646 | out: | ||||
| 647 | i915_gem_object_invalidate_frontbuffer(obj, ORIGIN_CPU); | ||||
| 648 | obj->mm.dirty = true1; | ||||
| 649 | /* return with the pages pinned */ | ||||
| 650 | return 0; | ||||
| 651 | |||||
| 652 | err_unpin: | ||||
| 653 | i915_gem_object_unpin_pages(obj); | ||||
| 654 | return ret; | ||||
| 655 | } |
| 1 | /* | ||||
| 2 | * SPDX-License-Identifier: MIT | ||||
| 3 | * | ||||
| 4 | * Copyright © 2016 Intel Corporation | ||||
| 5 | */ | ||||
| 6 | |||||
| 7 | #ifndef __I915_GEM_OBJECT_H__ | ||||
| 8 | #define __I915_GEM_OBJECT_H__ | ||||
| 9 | |||||
| 10 | #include <drm/drm_gem.h> | ||||
| 11 | #include <drm/drm_file.h> | ||||
| 12 | #include <drm/drm_device.h> | ||||
| 13 | |||||
| 14 | #include "display/intel_frontbuffer.h" | ||||
| 15 | #include "i915_gem_object_types.h" | ||||
| 16 | #include "i915_gem_gtt.h" | ||||
| 17 | #include "i915_vma_types.h" | ||||
| 18 | |||||
| 19 | void i915_gem_init__objects(struct drm_i915_privateinteldrm_softc *i915); | ||||
| 20 | |||||
| 21 | struct drm_i915_gem_object *i915_gem_object_alloc(void); | ||||
| 22 | void i915_gem_object_free(struct drm_i915_gem_object *obj); | ||||
| 23 | |||||
| 24 | void i915_gem_object_init(struct drm_i915_gem_object *obj, | ||||
| 25 | const struct drm_i915_gem_object_ops *ops, | ||||
| 26 | struct lock_class_key *key); | ||||
| 27 | struct drm_i915_gem_object * | ||||
| 28 | i915_gem_object_create_shmem(struct drm_i915_privateinteldrm_softc *i915, | ||||
| 29 | resource_size_t size); | ||||
| 30 | struct drm_i915_gem_object * | ||||
| 31 | i915_gem_object_create_shmem_from_data(struct drm_i915_privateinteldrm_softc *i915, | ||||
| 32 | const void *data, resource_size_t size); | ||||
| 33 | |||||
| 34 | extern const struct drm_i915_gem_object_ops i915_gem_shmem_ops; | ||||
| 35 | void __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj, | ||||
| 36 | struct sg_table *pages, | ||||
| 37 | bool_Bool needs_clflush); | ||||
| 38 | |||||
| 39 | int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, int align); | ||||
| 40 | |||||
| 41 | void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file); | ||||
| 42 | void i915_gem_free_object(struct drm_gem_object *obj); | ||||
| 43 | |||||
| 44 | void i915_gem_flush_free_objects(struct drm_i915_privateinteldrm_softc *i915); | ||||
| 45 | |||||
| 46 | struct sg_table * | ||||
| 47 | __i915_gem_object_unset_pages(struct drm_i915_gem_object *obj); | ||||
| 48 | void i915_gem_object_truncate(struct drm_i915_gem_object *obj); | ||||
| 49 | |||||
| 50 | /** | ||||
| 51 | * i915_gem_object_lookup_rcu - look up a temporary GEM object from its handle | ||||
| 52 | * @filp: DRM file private date | ||||
| 53 | * @handle: userspace handle | ||||
| 54 | * | ||||
| 55 | * Returns: | ||||
| 56 | * | ||||
| 57 | * A pointer to the object named by the handle if such exists on @filp, NULL | ||||
| 58 | * otherwise. This object is only valid whilst under the RCU read lock, and | ||||
| 59 | * note carefully the object may be in the process of being destroyed. | ||||
| 60 | */ | ||||
| 61 | static inline struct drm_i915_gem_object * | ||||
| 62 | i915_gem_object_lookup_rcu(struct drm_file *file, u32 handle) | ||||
| 63 | { | ||||
| 64 | #ifdef CONFIG_LOCKDEP | ||||
| 65 | WARN_ON(debug_locks && !lock_is_held(&rcu_lock_map))({ int __ret = !!((debug_locks && !lock_is_held(& rcu_lock_map))); if (__ret) printf("%s", "WARN_ON(" "debug_locks && !lock_is_held(&rcu_lock_map)" ")"); __builtin_expect(!!(__ret), 0); }); | ||||
| 66 | #endif | ||||
| 67 | return idr_find(&file->object_idr, handle); | ||||
| 68 | } | ||||
| 69 | |||||
| 70 | static inline struct drm_i915_gem_object * | ||||
| 71 | i915_gem_object_get_rcu(struct drm_i915_gem_object *obj) | ||||
| 72 | { | ||||
| 73 | if (obj && !kref_get_unless_zero(&obj->base.refcount)) | ||||
| 74 | obj = NULL((void *)0); | ||||
| 75 | |||||
| 76 | return obj; | ||||
| 77 | } | ||||
| 78 | |||||
| 79 | static inline struct drm_i915_gem_object * | ||||
| 80 | i915_gem_object_lookup(struct drm_file *file, u32 handle) | ||||
| 81 | { | ||||
| 82 | struct drm_i915_gem_object *obj; | ||||
| 83 | |||||
| 84 | rcu_read_lock(); | ||||
| 85 | obj = i915_gem_object_lookup_rcu(file, handle); | ||||
| 86 | obj = i915_gem_object_get_rcu(obj); | ||||
| 87 | rcu_read_unlock(); | ||||
| 88 | |||||
| 89 | return obj; | ||||
| 90 | } | ||||
| 91 | |||||
| 92 | __deprecated | ||||
| 93 | struct drm_gem_object * | ||||
| 94 | drm_gem_object_lookup(struct drm_file *file, u32 handle); | ||||
| 95 | |||||
| 96 | __attribute__((nonnull)) | ||||
| 97 | static inline struct drm_i915_gem_object * | ||||
| 98 | i915_gem_object_get(struct drm_i915_gem_object *obj) | ||||
| 99 | { | ||||
| 100 | drm_gem_object_get(&obj->base); | ||||
| 101 | return obj; | ||||
| 102 | } | ||||
| 103 | |||||
| 104 | __attribute__((nonnull)) | ||||
| 105 | static inline void | ||||
| 106 | i915_gem_object_put(struct drm_i915_gem_object *obj) | ||||
| 107 | { | ||||
| 108 | __drm_gem_object_put(&obj->base); | ||||
| 109 | } | ||||
| 110 | |||||
| 111 | #define assert_object_held(obj)do { (void)(&((obj)->base.resv)->lock.base); } while (0) dma_resv_assert_held((obj)->base.resv)do { (void)(&((obj)->base.resv)->lock.base); } while (0) | ||||
| 112 | |||||
| 113 | static inline int __i915_gem_object_lock(struct drm_i915_gem_object *obj, | ||||
| 114 | struct i915_gem_ww_ctx *ww, | ||||
| 115 | bool_Bool intr) | ||||
| 116 | { | ||||
| 117 | int ret; | ||||
| 118 | |||||
| 119 | if (intr
| ||||
| 120 | ret = dma_resv_lock_interruptible(obj->base.resv, ww
| ||||
| 121 | else | ||||
| 122 | ret = dma_resv_lock(obj->base.resv, ww ? &ww->ctx : NULL((void *)0)); | ||||
| 123 | |||||
| 124 | if (!ret && ww) | ||||
| 125 | list_add_tail(&obj->obj_link, &ww->obj_list); | ||||
| 126 | if (ret == -EALREADY37) | ||||
| 127 | ret = 0; | ||||
| 128 | |||||
| 129 | if (ret == -EDEADLK11) | ||||
| 130 | ww->contended = obj; | ||||
| |||||
| 131 | |||||
| 132 | return ret; | ||||
| 133 | } | ||||
| 134 | |||||
| 135 | static inline int i915_gem_object_lock(struct drm_i915_gem_object *obj, | ||||
| 136 | struct i915_gem_ww_ctx *ww) | ||||
| 137 | { | ||||
| 138 | return __i915_gem_object_lock(obj, ww, ww && ww->intr); | ||||
| 139 | } | ||||
| 140 | |||||
| 141 | static inline int i915_gem_object_lock_interruptible(struct drm_i915_gem_object *obj, | ||||
| 142 | struct i915_gem_ww_ctx *ww) | ||||
| 143 | { | ||||
| 144 | WARN_ON(ww && !ww->intr)({ int __ret = !!((ww && !ww->intr)); if (__ret) printf ("%s", "WARN_ON(" "ww && !ww->intr" ")"); __builtin_expect (!!(__ret), 0); }); | ||||
| 145 | return __i915_gem_object_lock(obj, ww, true1); | ||||
| 146 | } | ||||
| 147 | |||||
| 148 | static inline bool_Bool i915_gem_object_trylock(struct drm_i915_gem_object *obj) | ||||
| 149 | { | ||||
| 150 | return dma_resv_trylock(obj->base.resv); | ||||
| 151 | } | ||||
| 152 | |||||
| 153 | static inline void i915_gem_object_unlock(struct drm_i915_gem_object *obj) | ||||
| 154 | { | ||||
| 155 | dma_resv_unlock(obj->base.resv); | ||||
| 156 | } | ||||
| 157 | |||||
| 158 | struct dma_fence * | ||||
| 159 | i915_gem_object_lock_fence(struct drm_i915_gem_object *obj); | ||||
| 160 | void i915_gem_object_unlock_fence(struct drm_i915_gem_object *obj, | ||||
| 161 | struct dma_fence *fence); | ||||
| 162 | |||||
| 163 | static inline void | ||||
| 164 | i915_gem_object_set_readonly(struct drm_i915_gem_object *obj) | ||||
| 165 | { | ||||
| 166 | obj->flags |= I915_BO_READONLY(1UL << (2)); | ||||
| 167 | } | ||||
| 168 | |||||
| 169 | static inline bool_Bool | ||||
| 170 | i915_gem_object_is_readonly(const struct drm_i915_gem_object *obj) | ||||
| 171 | { | ||||
| 172 | return obj->flags & I915_BO_READONLY(1UL << (2)); | ||||
| 173 | } | ||||
| 174 | |||||
| 175 | static inline bool_Bool | ||||
| 176 | i915_gem_object_is_contiguous(const struct drm_i915_gem_object *obj) | ||||
| 177 | { | ||||
| 178 | return obj->flags & I915_BO_ALLOC_CONTIGUOUS(1UL << (0)); | ||||
| 179 | } | ||||
| 180 | |||||
| 181 | static inline bool_Bool | ||||
| 182 | i915_gem_object_is_volatile(const struct drm_i915_gem_object *obj) | ||||
| 183 | { | ||||
| 184 | return obj->flags & I915_BO_ALLOC_VOLATILE(1UL << (1)); | ||||
| 185 | } | ||||
| 186 | |||||
| 187 | static inline void | ||||
| 188 | i915_gem_object_set_volatile(struct drm_i915_gem_object *obj) | ||||
| 189 | { | ||||
| 190 | obj->flags |= I915_BO_ALLOC_VOLATILE(1UL << (1)); | ||||
| 191 | } | ||||
| 192 | |||||
| 193 | static inline bool_Bool | ||||
| 194 | i915_gem_object_type_has(const struct drm_i915_gem_object *obj, | ||||
| 195 | unsigned long flags) | ||||
| 196 | { | ||||
| 197 | return obj->ops->flags & flags; | ||||
| 198 | } | ||||
| 199 | |||||
| 200 | static inline bool_Bool | ||||
| 201 | i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj) | ||||
| 202 | { | ||||
| 203 | return i915_gem_object_type_has(obj, I915_GEM_OBJECT_HAS_STRUCT_PAGE(1UL << (0))); | ||||
| 204 | } | ||||
| 205 | |||||
| 206 | static inline bool_Bool | ||||
| 207 | i915_gem_object_is_shrinkable(const struct drm_i915_gem_object *obj) | ||||
| 208 | { | ||||
| 209 | return i915_gem_object_type_has(obj, I915_GEM_OBJECT_IS_SHRINKABLE(1UL << (2))); | ||||
| 210 | } | ||||
| 211 | |||||
| 212 | static inline bool_Bool | ||||
| 213 | i915_gem_object_is_proxy(const struct drm_i915_gem_object *obj) | ||||
| 214 | { | ||||
| 215 | return i915_gem_object_type_has(obj, I915_GEM_OBJECT_IS_PROXY(1UL << (3))); | ||||
| 216 | } | ||||
| 217 | |||||
| 218 | static inline bool_Bool | ||||
| 219 | i915_gem_object_never_mmap(const struct drm_i915_gem_object *obj) | ||||
| 220 | { | ||||
| 221 | return i915_gem_object_type_has(obj, I915_GEM_OBJECT_NO_MMAP(1UL << (4))); | ||||
| 222 | } | ||||
| 223 | |||||
| 224 | static inline bool_Bool | ||||
| 225 | i915_gem_object_needs_async_cancel(const struct drm_i915_gem_object *obj) | ||||
| 226 | { | ||||
| 227 | return i915_gem_object_type_has(obj, I915_GEM_OBJECT_ASYNC_CANCEL(1UL << (5))); | ||||
| 228 | } | ||||
| 229 | |||||
| 230 | static inline bool_Bool | ||||
| 231 | i915_gem_object_is_framebuffer(const struct drm_i915_gem_object *obj) | ||||
| 232 | { | ||||
| 233 | return READ_ONCE(obj->frontbuffer)({ typeof(obj->frontbuffer) __tmp = *(volatile typeof(obj-> frontbuffer) *)&(obj->frontbuffer); membar_datadep_consumer (); __tmp; }); | ||||
| 234 | } | ||||
| 235 | |||||
| 236 | static inline unsigned int | ||||
| 237 | i915_gem_object_get_tiling(const struct drm_i915_gem_object *obj) | ||||
| 238 | { | ||||
| 239 | return obj->tiling_and_stride & TILING_MASK(128 - 1); | ||||
| 240 | } | ||||
| 241 | |||||
| 242 | static inline bool_Bool | ||||
| 243 | i915_gem_object_is_tiled(const struct drm_i915_gem_object *obj) | ||||
| 244 | { | ||||
| 245 | return i915_gem_object_get_tiling(obj) != I915_TILING_NONE0; | ||||
| 246 | } | ||||
| 247 | |||||
| 248 | static inline unsigned int | ||||
| 249 | i915_gem_object_get_stride(const struct drm_i915_gem_object *obj) | ||||
| 250 | { | ||||
| 251 | return obj->tiling_and_stride & STRIDE_MASK(~(128 - 1)); | ||||
| 252 | } | ||||
| 253 | |||||
| 254 | static inline unsigned int | ||||
| 255 | i915_gem_tile_height(unsigned int tiling) | ||||
| 256 | { | ||||
| 257 | GEM_BUG_ON(!tiling)((void)0); | ||||
| 258 | return tiling == I915_TILING_Y2 ? 32 : 8; | ||||
| 259 | } | ||||
| 260 | |||||
| 261 | static inline unsigned int | ||||
| 262 | i915_gem_object_get_tile_height(const struct drm_i915_gem_object *obj) | ||||
| 263 | { | ||||
| 264 | return i915_gem_tile_height(i915_gem_object_get_tiling(obj)); | ||||
| 265 | } | ||||
| 266 | |||||
| 267 | static inline unsigned int | ||||
| 268 | i915_gem_object_get_tile_row_size(const struct drm_i915_gem_object *obj) | ||||
| 269 | { | ||||
| 270 | return (i915_gem_object_get_stride(obj) * | ||||
| 271 | i915_gem_object_get_tile_height(obj)); | ||||
| 272 | } | ||||
| 273 | |||||
| 274 | int i915_gem_object_set_tiling(struct drm_i915_gem_object *obj, | ||||
| 275 | unsigned int tiling, unsigned int stride); | ||||
| 276 | |||||
| 277 | struct scatterlist * | ||||
| 278 | i915_gem_object_get_sg(struct drm_i915_gem_object *obj, | ||||
| 279 | unsigned int n, unsigned int *offset); | ||||
| 280 | |||||
| 281 | struct vm_page * | ||||
| 282 | i915_gem_object_get_page(struct drm_i915_gem_object *obj, | ||||
| 283 | unsigned int n); | ||||
| 284 | |||||
| 285 | struct vm_page * | ||||
| 286 | i915_gem_object_get_dirty_page(struct drm_i915_gem_object *obj, | ||||
| 287 | unsigned int n); | ||||
| 288 | |||||
| 289 | dma_addr_t | ||||
| 290 | i915_gem_object_get_dma_address_len(struct drm_i915_gem_object *obj, | ||||
| 291 | unsigned long n, | ||||
| 292 | unsigned int *len); | ||||
| 293 | |||||
| 294 | dma_addr_t | ||||
| 295 | i915_gem_object_get_dma_address(struct drm_i915_gem_object *obj, | ||||
| 296 | unsigned long n); | ||||
| 297 | |||||
| 298 | void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj, | ||||
| 299 | struct sg_table *pages, | ||||
| 300 | unsigned int sg_page_sizes); | ||||
| 301 | |||||
| 302 | int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj); | ||||
| 303 | int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj); | ||||
| 304 | |||||
| 305 | enum i915_mm_subclass { /* lockdep subclass for obj->mm.lock/struct_mutex */ | ||||
| 306 | I915_MM_NORMAL = 0, | ||||
| 307 | /* | ||||
| 308 | * Only used by struct_mutex, when called "recursively" from | ||||
| 309 | * direct-reclaim-esque. Safe because there is only every one | ||||
| 310 | * struct_mutex in the entire system. | ||||
| 311 | */ | ||||
| 312 | I915_MM_SHRINKER = 1, | ||||
| 313 | /* | ||||
| 314 | * Used for obj->mm.lock when allocating pages. Safe because the object | ||||
| 315 | * isn't yet on any LRU, and therefore the shrinker can't deadlock on | ||||
| 316 | * it. As soon as the object has pages, obj->mm.lock nests within | ||||
| 317 | * fs_reclaim. | ||||
| 318 | */ | ||||
| 319 | I915_MM_GET_PAGES = 1, | ||||
| 320 | }; | ||||
| 321 | |||||
| 322 | static inline int __must_check | ||||
| 323 | i915_gem_object_pin_pages(struct drm_i915_gem_object *obj) | ||||
| 324 | { | ||||
| 325 | might_lock_nested(&obj->mm.lock, I915_MM_GET_PAGES); | ||||
| 326 | |||||
| 327 | if (atomic_inc_not_zero(&obj->mm.pages_pin_count)atomic_add_unless((&obj->mm.pages_pin_count), 1, 0)) | ||||
| 328 | return 0; | ||||
| 329 | |||||
| 330 | return __i915_gem_object_get_pages(obj); | ||||
| 331 | } | ||||
| 332 | |||||
| 333 | static inline bool_Bool | ||||
| 334 | i915_gem_object_has_pages(struct drm_i915_gem_object *obj) | ||||
| 335 | { | ||||
| 336 | return !IS_ERR_OR_NULL(READ_ONCE(obj->mm.pages)({ typeof(obj->mm.pages) __tmp = *(volatile typeof(obj-> mm.pages) *)&(obj->mm.pages); membar_datadep_consumer( ); __tmp; })); | ||||
| 337 | } | ||||
| 338 | |||||
| 339 | static inline void | ||||
| 340 | __i915_gem_object_pin_pages(struct drm_i915_gem_object *obj) | ||||
| 341 | { | ||||
| 342 | GEM_BUG_ON(!i915_gem_object_has_pages(obj))((void)0); | ||||
| 343 | |||||
| 344 | atomic_inc(&obj->mm.pages_pin_count)__sync_fetch_and_add(&obj->mm.pages_pin_count, 1); | ||||
| 345 | } | ||||
| 346 | |||||
| 347 | static inline bool_Bool | ||||
| 348 | i915_gem_object_has_pinned_pages(struct drm_i915_gem_object *obj) | ||||
| 349 | { | ||||
| 350 | return atomic_read(&obj->mm.pages_pin_count)({ typeof(*(&obj->mm.pages_pin_count)) __tmp = *(volatile typeof(*(&obj->mm.pages_pin_count)) *)&(*(&obj ->mm.pages_pin_count)); membar_datadep_consumer(); __tmp; } ); | ||||
| 351 | } | ||||
| 352 | |||||
| 353 | static inline void | ||||
| 354 | __i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj) | ||||
| 355 | { | ||||
| 356 | GEM_BUG_ON(!i915_gem_object_has_pages(obj))((void)0); | ||||
| 357 | GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj))((void)0); | ||||
| 358 | |||||
| 359 | atomic_dec(&obj->mm.pages_pin_count)__sync_fetch_and_sub(&obj->mm.pages_pin_count, 1); | ||||
| 360 | } | ||||
| 361 | |||||
| 362 | static inline void | ||||
| 363 | i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj) | ||||
| 364 | { | ||||
| 365 | __i915_gem_object_unpin_pages(obj); | ||||
| 366 | } | ||||
| 367 | |||||
| 368 | int __i915_gem_object_put_pages(struct drm_i915_gem_object *obj); | ||||
| 369 | void i915_gem_object_truncate(struct drm_i915_gem_object *obj); | ||||
| 370 | void i915_gem_object_writeback(struct drm_i915_gem_object *obj); | ||||
| 371 | |||||
| 372 | enum i915_map_type { | ||||
| 373 | I915_MAP_WB = 0, | ||||
| 374 | I915_MAP_WC, | ||||
| 375 | #define I915_MAP_OVERRIDE(1UL << (31)) BIT(31)(1UL << (31)) | ||||
| 376 | I915_MAP_FORCE_WB = I915_MAP_WB | I915_MAP_OVERRIDE(1UL << (31)), | ||||
| 377 | I915_MAP_FORCE_WC = I915_MAP_WC | I915_MAP_OVERRIDE(1UL << (31)), | ||||
| 378 | }; | ||||
| 379 | |||||
| 380 | /** | ||||
| 381 | * i915_gem_object_pin_map - return a contiguous mapping of the entire object | ||||
| 382 | * @obj: the object to map into kernel address space | ||||
| 383 | * @type: the type of mapping, used to select pgprot_t | ||||
| 384 | * | ||||
| 385 | * Calls i915_gem_object_pin_pages() to prevent reaping of the object's | ||||
| 386 | * pages and then returns a contiguous mapping of the backing storage into | ||||
| 387 | * the kernel address space. Based on the @type of mapping, the PTE will be | ||||
| 388 | * set to either WriteBack or WriteCombine (via pgprot_t). | ||||
| 389 | * | ||||
| 390 | * The caller is responsible for calling i915_gem_object_unpin_map() when the | ||||
| 391 | * mapping is no longer required. | ||||
| 392 | * | ||||
| 393 | * Returns the pointer through which to access the mapped object, or an | ||||
| 394 | * ERR_PTR() on error. | ||||
| 395 | */ | ||||
| 396 | void *__must_check i915_gem_object_pin_map(struct drm_i915_gem_object *obj, | ||||
| 397 | enum i915_map_type type); | ||||
| 398 | |||||
| 399 | void __i915_gem_object_flush_map(struct drm_i915_gem_object *obj, | ||||
| 400 | unsigned long offset, | ||||
| 401 | unsigned long size); | ||||
| 402 | static inline void i915_gem_object_flush_map(struct drm_i915_gem_object *obj) | ||||
| 403 | { | ||||
| 404 | __i915_gem_object_flush_map(obj, 0, obj->base.size); | ||||
| 405 | } | ||||
| 406 | |||||
| 407 | /** | ||||
| 408 | * i915_gem_object_unpin_map - releases an earlier mapping | ||||
| 409 | * @obj: the object to unmap | ||||
| 410 | * | ||||
| 411 | * After pinning the object and mapping its pages, once you are finished | ||||
| 412 | * with your access, call i915_gem_object_unpin_map() to release the pin | ||||
| 413 | * upon the mapping. Once the pin count reaches zero, that mapping may be | ||||
| 414 | * removed. | ||||
| 415 | */ | ||||
| 416 | static inline void i915_gem_object_unpin_map(struct drm_i915_gem_object *obj) | ||||
| 417 | { | ||||
| 418 | i915_gem_object_unpin_pages(obj); | ||||
| 419 | } | ||||
| 420 | |||||
| 421 | void __i915_gem_object_release_map(struct drm_i915_gem_object *obj); | ||||
| 422 | |||||
| 423 | void | ||||
| 424 | i915_gem_object_flush_write_domain(struct drm_i915_gem_object *obj, | ||||
| 425 | unsigned int flush_domains); | ||||
| 426 | |||||
| 427 | int i915_gem_object_prepare_read(struct drm_i915_gem_object *obj, | ||||
| 428 | unsigned int *needs_clflush); | ||||
| 429 | int i915_gem_object_prepare_write(struct drm_i915_gem_object *obj, | ||||
| 430 | unsigned int *needs_clflush); | ||||
| 431 | #define CLFLUSH_BEFORE(1UL << (0)) BIT(0)(1UL << (0)) | ||||
| 432 | #define CLFLUSH_AFTER(1UL << (1)) BIT(1)(1UL << (1)) | ||||
| 433 | #define CLFLUSH_FLAGS((1UL << (0)) | (1UL << (1))) (CLFLUSH_BEFORE(1UL << (0)) | CLFLUSH_AFTER(1UL << (1))) | ||||
| 434 | |||||
| 435 | static inline void | ||||
| 436 | i915_gem_object_finish_access(struct drm_i915_gem_object *obj) | ||||
| 437 | { | ||||
| 438 | i915_gem_object_unpin_pages(obj); | ||||
| 439 | } | ||||
| 440 | |||||
| 441 | static inline struct intel_engine_cs * | ||||
| 442 | i915_gem_object_last_write_engine(struct drm_i915_gem_object *obj) | ||||
| 443 | { | ||||
| 444 | struct intel_engine_cs *engine = NULL((void *)0); | ||||
| 445 | struct dma_fence *fence; | ||||
| 446 | |||||
| 447 | rcu_read_lock(); | ||||
| 448 | fence = dma_resv_get_excl_rcu(obj->base.resv); | ||||
| 449 | rcu_read_unlock(); | ||||
| 450 | |||||
| 451 | if (fence && dma_fence_is_i915(fence) && !dma_fence_is_signaled(fence)) | ||||
| 452 | engine = to_request(fence)->engine; | ||||
| 453 | dma_fence_put(fence); | ||||
| 454 | |||||
| 455 | return engine; | ||||
| 456 | } | ||||
| 457 | |||||
| 458 | void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, | ||||
| 459 | unsigned int cache_level); | ||||
| 460 | void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj); | ||||
| 461 | void i915_gem_object_flush_if_display_locked(struct drm_i915_gem_object *obj); | ||||
| 462 | |||||
| 463 | int __must_check | ||||
| 464 | i915_gem_object_set_to_wc_domain(struct drm_i915_gem_object *obj, bool_Bool write); | ||||
| 465 | int __must_check | ||||
| 466 | i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool_Bool write); | ||||
| 467 | int __must_check | ||||
| 468 | i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool_Bool write); | ||||
| 469 | struct i915_vma * __must_check | ||||
| 470 | i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, | ||||
| 471 | u32 alignment, | ||||
| 472 | const struct i915_ggtt_view *view, | ||||
| 473 | unsigned int flags); | ||||
| 474 | |||||
| 475 | void i915_gem_object_make_unshrinkable(struct drm_i915_gem_object *obj); | ||||
| 476 | void i915_gem_object_make_shrinkable(struct drm_i915_gem_object *obj); | ||||
| 477 | void i915_gem_object_make_purgeable(struct drm_i915_gem_object *obj); | ||||
| 478 | |||||
| 479 | static inline bool_Bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj) | ||||
| 480 | { | ||||
| 481 | if (obj->cache_dirty) | ||||
| 482 | return false0; | ||||
| 483 | |||||
| 484 | if (!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE(1UL << (1)))) | ||||
| 485 | return true1; | ||||
| 486 | |||||
| 487 | /* Currently in use by HW (display engine)? Keep flushed. */ | ||||
| 488 | return i915_gem_object_is_framebuffer(obj); | ||||
| 489 | } | ||||
| 490 | |||||
| 491 | static inline void __start_cpu_write(struct drm_i915_gem_object *obj) | ||||
| 492 | { | ||||
| 493 | obj->read_domains = I915_GEM_DOMAIN_CPU0x00000001; | ||||
| 494 | obj->write_domain = I915_GEM_DOMAIN_CPU0x00000001; | ||||
| 495 | if (cpu_write_needs_clflush(obj)) | ||||
| 496 | obj->cache_dirty = true1; | ||||
| 497 | } | ||||
| 498 | |||||
| 499 | int i915_gem_object_wait(struct drm_i915_gem_object *obj, | ||||
| 500 | unsigned int flags, | ||||
| 501 | long timeout); | ||||
| 502 | int i915_gem_object_wait_priority(struct drm_i915_gem_object *obj, | ||||
| 503 | unsigned int flags, | ||||
| 504 | const struct i915_sched_attr *attr); | ||||
| 505 | |||||
| 506 | void __i915_gem_object_flush_frontbuffer(struct drm_i915_gem_object *obj, | ||||
| 507 | enum fb_op_origin origin); | ||||
| 508 | void __i915_gem_object_invalidate_frontbuffer(struct drm_i915_gem_object *obj, | ||||
| 509 | enum fb_op_origin origin); | ||||
| 510 | |||||
| 511 | static inline void | ||||
| 512 | i915_gem_object_flush_frontbuffer(struct drm_i915_gem_object *obj, | ||||
| 513 | enum fb_op_origin origin) | ||||
| 514 | { | ||||
| 515 | if (unlikely(rcu_access_pointer(obj->frontbuffer))__builtin_expect(!!((obj->frontbuffer)), 0)) | ||||
| 516 | __i915_gem_object_flush_frontbuffer(obj, origin); | ||||
| 517 | } | ||||
| 518 | |||||
| 519 | static inline void | ||||
| 520 | i915_gem_object_invalidate_frontbuffer(struct drm_i915_gem_object *obj, | ||||
| 521 | enum fb_op_origin origin) | ||||
| 522 | { | ||||
| 523 | if (unlikely(rcu_access_pointer(obj->frontbuffer))__builtin_expect(!!((obj->frontbuffer)), 0)) | ||||
| 524 | __i915_gem_object_invalidate_frontbuffer(obj, origin); | ||||
| 525 | } | ||||
| 526 | |||||
| 527 | #endif |