1 /* 2 * Copyright © 2016 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 * 23 */ 24 25 #ifndef __I915_GEM_OBJECT_H__ 26 #define __I915_GEM_OBJECT_H__ 27 28 #include <linux/reservation.h> 29 30 #include <drm/drm_vma_manager.h> 31 #include <drm/drm_gem.h> 32 #include <drm/drmP.h> 33 34 #include <drm/i915_drm.h> 35 36 #include "i915_gem_request.h" 37 #include "i915_selftest.h" 38 39 struct drm_i915_gem_object; 40 41 /* 42 * struct i915_lut_handle tracks the fast lookups from handle to vma used 43 * for execbuf. Although we use a radixtree for that mapping, in order to 44 * remove them as the object or context is closed, we need a secondary list 45 * and a translation entry (i915_lut_handle). 46 */ 47 struct i915_lut_handle { 48 struct list_head obj_link; 49 struct list_head ctx_link; 50 struct i915_gem_context *ctx; 51 u32 handle; 52 }; 53 54 struct drm_i915_gem_object_ops { 55 unsigned int flags; 56 #define I915_GEM_OBJECT_HAS_STRUCT_PAGE BIT(0) 57 #define I915_GEM_OBJECT_IS_SHRINKABLE BIT(1) 58 59 /* Interface between the GEM object and its backing storage. 60 * get_pages() is called once prior to the use of the associated set 61 * of pages before to binding them into the GTT, and put_pages() is 62 * called after we no longer need them. As we expect there to be 63 * associated cost with migrating pages between the backing storage 64 * and making them available for the GPU (e.g. clflush), we may hold 65 * onto the pages after they are no longer referenced by the GPU 66 * in case they may be used again shortly (for example migrating the 67 * pages to a different memory domain within the GTT). put_pages() 68 * will therefore most likely be called when the object itself is 69 * being released or under memory pressure (where we attempt to 70 * reap pages for the shrinker). 71 */ 72 int (*get_pages)(struct drm_i915_gem_object *); 73 void (*put_pages)(struct drm_i915_gem_object *, struct sg_table *); 74 75 int (*pwrite)(struct drm_i915_gem_object *, 76 const struct drm_i915_gem_pwrite *); 77 78 int (*dmabuf_export)(struct drm_i915_gem_object *); 79 void (*release)(struct drm_i915_gem_object *); 80 }; 81 82 struct drm_i915_gem_object { 83 struct drm_gem_object base; 84 85 const struct drm_i915_gem_object_ops *ops; 86 87 /** 88 * @vma_list: List of VMAs backed by this object 89 * 90 * The VMA on this list are ordered by type, all GGTT vma are placed 91 * at the head and all ppGTT vma are placed at the tail. The different 92 * types of GGTT vma are unordered between themselves, use the 93 * @vma_tree (which has a defined order between all VMA) to find an 94 * exact match. 95 */ 96 struct list_head vma_list; 97 /** 98 * @vma_tree: Ordered tree of VMAs backed by this object 99 * 100 * All VMA created for this object are placed in the @vma_tree for 101 * fast retrieval via a binary search in i915_vma_instance(). 102 * They are also added to @vma_list for easy iteration. 103 */ 104 struct rb_root vma_tree; 105 106 /** 107 * @lut_list: List of vma lookup entries in use for this object. 108 * 109 * If this object is closed, we need to remove all of its VMA from 110 * the fast lookup index in associated contexts; @lut_list provides 111 * this translation from object to context->handles_vma. 112 */ 113 struct list_head lut_list; 114 115 /** Stolen memory for this object, instead of being backed by shmem. */ 116 struct drm_mm_node *stolen; 117 union { 118 struct rcu_head rcu; 119 struct llist_node freed; 120 }; 121 122 /** 123 * Whether the object is currently in the GGTT mmap. 124 */ 125 unsigned int userfault_count; 126 struct list_head userfault_link; 127 128 struct list_head batch_pool_link; 129 I915_SELFTEST_DECLARE(struct list_head st_link); 130 131 unsigned long flags; 132 133 /** 134 * Have we taken a reference for the object for incomplete GPU 135 * activity? 136 */ 137 #define I915_BO_ACTIVE_REF 0 138 139 /* 140 * Is the object to be mapped as read-only to the GPU 141 * Only honoured if hardware has relevant pte bit 142 */ 143 unsigned long gt_ro:1; 144 unsigned int cache_level:3; 145 unsigned int cache_coherent:2; 146 #define I915_BO_CACHE_COHERENT_FOR_READ BIT(0) 147 #define I915_BO_CACHE_COHERENT_FOR_WRITE BIT(1) 148 unsigned int cache_dirty:1; 149 150 atomic_t frontbuffer_bits; 151 unsigned int frontbuffer_ggtt_origin; /* write once */ 152 struct i915_gem_active frontbuffer_write; 153 154 /** Current tiling stride for the object, if it's tiled. */ 155 unsigned int tiling_and_stride; 156 #define FENCE_MINIMUM_STRIDE 128 /* See i915_tiling_ok() */ 157 #define TILING_MASK (FENCE_MINIMUM_STRIDE-1) 158 #define STRIDE_MASK (~TILING_MASK) 159 160 /** Count of VMA actually bound by this object */ 161 unsigned int bind_count; 162 unsigned int active_count; 163 /** Count of how many global VMA are currently pinned for use by HW */ 164 unsigned int pin_global; 165 166 struct { 167 struct lock lock; /* protects the pages and their use */ 168 atomic_t pages_pin_count; 169 170 struct sg_table *pages; 171 void *mapping; 172 173 /* TODO: whack some of this into the error state */ 174 struct i915_page_sizes { 175 /** 176 * The sg mask of the pages sg_table. i.e the mask of 177 * of the lengths for each sg entry. 178 */ 179 unsigned int phys; 180 181 /** 182 * The gtt page sizes we are allowed to use given the 183 * sg mask and the supported page sizes. This will 184 * express the smallest unit we can use for the whole 185 * object, as well as the larger sizes we may be able 186 * to use opportunistically. 187 */ 188 unsigned int sg; 189 190 /** 191 * The actual gtt page size usage. Since we can have 192 * multiple vma associated with this object we need to 193 * prevent any trampling of state, hence a copy of this 194 * struct also lives in each vma, therefore the gtt 195 * value here should only be read/write through the vma. 196 */ 197 unsigned int gtt; 198 } page_sizes; 199 200 I915_SELFTEST_DECLARE(unsigned int page_mask); 201 202 struct i915_gem_object_page_iter { 203 struct scatterlist *sg_pos; 204 unsigned int sg_idx; /* in pages, but 32bit eek! */ 205 206 struct radix_tree_root radix; 207 struct lock lock; /* protects this cache */ 208 } get_page; 209 210 /** 211 * Element within i915->mm.unbound_list or i915->mm.bound_list, 212 * locked by i915->mm.obj_lock. 213 */ 214 struct list_head link; 215 216 /** 217 * Advice: are the backing pages purgeable? 218 */ 219 unsigned int madv:2; 220 221 /** 222 * This is set if the object has been written to since the 223 * pages were last acquired. 224 */ 225 bool dirty:1; 226 227 /** 228 * This is set if the object has been pinned due to unknown 229 * swizzling. 230 */ 231 bool quirked:1; 232 } mm; 233 234 /** Breadcrumb of last rendering to the buffer. 235 * There can only be one writer, but we allow for multiple readers. 236 * If there is a writer that necessarily implies that all other 237 * read requests are complete - but we may only be lazily clearing 238 * the read requests. A read request is naturally the most recent 239 * request on a ring, so we may have two different write and read 240 * requests on one ring where the write request is older than the 241 * read request. This allows for the CPU to read from an active 242 * buffer by only waiting for the write to complete. 243 */ 244 struct reservation_object *resv; 245 246 /** References from framebuffers, locks out tiling changes. */ 247 unsigned int framebuffer_references; 248 249 /** Record of address bit 17 of each page at last unbind. */ 250 unsigned long *bit_17; 251 252 union { 253 struct i915_gem_userptr { 254 uintptr_t ptr; 255 unsigned read_only :1; 256 257 struct i915_mm_struct *mm; 258 struct i915_mmu_object *mmu_object; 259 struct work_struct *work; 260 } userptr; 261 262 unsigned long scratch; 263 }; 264 265 /** for phys allocated objects */ 266 struct drm_dma_handle *phys_handle; 267 268 struct reservation_object __builtin_resv; 269 }; 270 271 static inline struct drm_i915_gem_object * 272 to_intel_bo(struct drm_gem_object *gem) 273 { 274 /* Assert that to_intel_bo(NULL) == NULL */ 275 BUILD_BUG_ON(offsetof(struct drm_i915_gem_object, base)); 276 277 return container_of(gem, struct drm_i915_gem_object, base); 278 } 279 280 /** 281 * i915_gem_object_lookup_rcu - look up a temporary GEM object from its handle 282 * @filp: DRM file private date 283 * @handle: userspace handle 284 * 285 * Returns: 286 * 287 * A pointer to the object named by the handle if such exists on @filp, NULL 288 * otherwise. This object is only valid whilst under the RCU read lock, and 289 * note carefully the object may be in the process of being destroyed. 290 */ 291 static inline struct drm_i915_gem_object * 292 i915_gem_object_lookup_rcu(struct drm_file *file, u32 handle) 293 { 294 #ifdef CONFIG_LOCKDEP 295 WARN_ON(debug_locks && !lock_is_held(&rcu_lock_map)); 296 #endif 297 return idr_find(&file->object_idr, handle); 298 } 299 300 static inline struct drm_i915_gem_object * 301 i915_gem_object_lookup(struct drm_file *file, u32 handle) 302 { 303 struct drm_i915_gem_object *obj; 304 305 rcu_read_lock(); 306 obj = i915_gem_object_lookup_rcu(file, handle); 307 if (obj && !kref_get_unless_zero(&obj->base.refcount)) 308 obj = NULL; 309 rcu_read_unlock(); 310 311 return obj; 312 } 313 314 __attribute__((nonnull)) 315 static inline struct drm_i915_gem_object * 316 i915_gem_object_get(struct drm_i915_gem_object *obj) 317 { 318 drm_gem_object_reference(&obj->base); 319 return obj; 320 } 321 322 __attribute__((nonnull)) 323 static inline void 324 i915_gem_object_put(struct drm_i915_gem_object *obj) 325 { 326 __drm_gem_object_unreference(&obj->base); 327 } 328 329 static inline void i915_gem_object_lock(struct drm_i915_gem_object *obj) 330 { 331 reservation_object_lock(obj->resv, NULL); 332 } 333 334 static inline void i915_gem_object_unlock(struct drm_i915_gem_object *obj) 335 { 336 reservation_object_unlock(obj->resv); 337 } 338 339 static inline bool 340 i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj) 341 { 342 return obj->ops->flags & I915_GEM_OBJECT_HAS_STRUCT_PAGE; 343 } 344 345 static inline bool 346 i915_gem_object_is_shrinkable(const struct drm_i915_gem_object *obj) 347 { 348 return obj->ops->flags & I915_GEM_OBJECT_IS_SHRINKABLE; 349 } 350 351 static inline bool 352 i915_gem_object_is_active(const struct drm_i915_gem_object *obj) 353 { 354 return obj->active_count; 355 } 356 357 static inline bool 358 i915_gem_object_has_active_reference(const struct drm_i915_gem_object *obj) 359 { 360 return test_bit(I915_BO_ACTIVE_REF, &obj->flags); 361 } 362 363 static inline void 364 i915_gem_object_set_active_reference(struct drm_i915_gem_object *obj) 365 { 366 lockdep_assert_held(&obj->base.dev->struct_mutex); 367 __set_bit(I915_BO_ACTIVE_REF, &obj->flags); 368 } 369 370 static inline void 371 i915_gem_object_clear_active_reference(struct drm_i915_gem_object *obj) 372 { 373 lockdep_assert_held(&obj->base.dev->struct_mutex); 374 __clear_bit(I915_BO_ACTIVE_REF, &obj->flags); 375 } 376 377 void __i915_gem_object_release_unless_active(struct drm_i915_gem_object *obj); 378 379 static inline bool 380 i915_gem_object_is_framebuffer(const struct drm_i915_gem_object *obj) 381 { 382 return READ_ONCE(obj->framebuffer_references); 383 } 384 385 static inline unsigned int 386 i915_gem_object_get_tiling(struct drm_i915_gem_object *obj) 387 { 388 return obj->tiling_and_stride & TILING_MASK; 389 } 390 391 static inline bool 392 i915_gem_object_is_tiled(struct drm_i915_gem_object *obj) 393 { 394 return i915_gem_object_get_tiling(obj) != I915_TILING_NONE; 395 } 396 397 static inline unsigned int 398 i915_gem_object_get_stride(struct drm_i915_gem_object *obj) 399 { 400 return obj->tiling_and_stride & STRIDE_MASK; 401 } 402 403 static inline unsigned int 404 i915_gem_tile_height(unsigned int tiling) 405 { 406 GEM_BUG_ON(!tiling); 407 return tiling == I915_TILING_Y ? 32 : 8; 408 } 409 410 static inline unsigned int 411 i915_gem_object_get_tile_height(struct drm_i915_gem_object *obj) 412 { 413 return i915_gem_tile_height(i915_gem_object_get_tiling(obj)); 414 } 415 416 static inline unsigned int 417 i915_gem_object_get_tile_row_size(struct drm_i915_gem_object *obj) 418 { 419 return (i915_gem_object_get_stride(obj) * 420 i915_gem_object_get_tile_height(obj)); 421 } 422 423 int i915_gem_object_set_tiling(struct drm_i915_gem_object *obj, 424 unsigned int tiling, unsigned int stride); 425 426 static inline struct intel_engine_cs * 427 i915_gem_object_last_write_engine(struct drm_i915_gem_object *obj) 428 { 429 struct intel_engine_cs *engine = NULL; 430 struct dma_fence *fence; 431 432 rcu_read_lock(); 433 fence = reservation_object_get_excl_rcu(obj->resv); 434 rcu_read_unlock(); 435 436 if (fence && dma_fence_is_i915(fence) && !dma_fence_is_signaled(fence)) 437 engine = to_request(fence)->engine; 438 dma_fence_put(fence); 439 440 return engine; 441 } 442 443 void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, 444 unsigned int cache_level); 445 void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj); 446 447 #endif 448 449