drivers/gpu/drm/i915/i915_gem_object.h

   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/drm_file.h>
  33 #include <drm/drm_device.h>
  34
  35 #include <drm/i915_drm.h>
  36
  37 #include "i915_request.h"
  38 #include "i915_selftest.h"
  39
  40 struct drm_i915_gem_object;
  41
  42 /*
  43  * struct i915_lut_handle tracks the fast lookups from handle to vma used
  44  * for execbuf. Although we use a radixtree for that mapping, in order to
  45  * remove them as the object or context is closed, we need a secondary list
  46  * and a translation entry (i915_lut_handle).
  47  */
  48 struct i915_lut_handle {
  49         struct list_head obj_link;
  50         struct list_head ctx_link;
  51         struct i915_gem_context *ctx;
  52         u32 handle;
  53 };
  54
  55 struct drm_i915_gem_object_ops {
  56         unsigned int flags;
  57 #define I915_GEM_OBJECT_HAS_STRUCT_PAGE BIT(0)
  58 #define I915_GEM_OBJECT_IS_SHRINKABLE   BIT(1)
  59 #define I915_GEM_OBJECT_IS_PROXY        BIT(2)
  60 #define I915_GEM_OBJECT_ASYNC_CANCEL    BIT(3)
  61
  62         /* Interface between the GEM object and its backing storage.
  63          * get_pages() is called once prior to the use of the associated set
  64          * of pages before to binding them into the GTT, and put_pages() is
  65          * called after we no longer need them. As we expect there to be
  66          * associated cost with migrating pages between the backing storage
  67          * and making them available for the GPU (e.g. clflush), we may hold
  68          * onto the pages after they are no longer referenced by the GPU
  69          * in case they may be used again shortly (for example migrating the
  70          * pages to a different memory domain within the GTT). put_pages()
  71          * will therefore most likely be called when the object itself is
  72          * being released or under memory pressure (where we attempt to
  73          * reap pages for the shrinker).
  74          */
  75         int (*get_pages)(struct drm_i915_gem_object *);
  76         void (*put_pages)(struct drm_i915_gem_object *, struct sg_table *);
  77
  78         int (*pwrite)(struct drm_i915_gem_object *,
  79                       const struct drm_i915_gem_pwrite *);
  80
  81         int (*dmabuf_export)(struct drm_i915_gem_object *);
  82         void (*release)(struct drm_i915_gem_object *);
  83 };
  84
  85 struct drm_i915_gem_object {
  86         struct drm_gem_object base;
  87
  88         const struct drm_i915_gem_object_ops *ops;
  89
  90         struct {
  91                 /**
  92                  * @vma.lock: protect the list/tree of vmas
  93                  */
  94                 spinlock_t lock;
  95
  96                 /**
  97                  * @vma.list: List of VMAs backed by this object
  98                  *
  99                  * The VMA on this list are ordered by type, all GGTT vma are
 100                  * placed at the head and all ppGTT vma are placed at the tail.
 101                  * The different types of GGTT vma are unordered between
 102                  * themselves, use the @vma.tree (which has a defined order
 103                  * between all VMA) to quickly find an exact match.
 104                  */
 105                 struct list_head list;
 106
 107                 /**
 108                  * @vma.tree: Ordered tree of VMAs backed by this object
 109                  *
 110                  * All VMA created for this object are placed in the @vma.tree
 111                  * for fast retrieval via a binary search in
 112                  * i915_vma_instance(). They are also added to @vma.list for
 113                  * easy iteration.
 114                  */
 115                 struct rb_root tree;
 116         } vma;
 117
 118         /**
 119          * @lut_list: List of vma lookup entries in use for this object.
 120          *
 121          * If this object is closed, we need to remove all of its VMA from
 122          * the fast lookup index in associated contexts; @lut_list provides
 123          * this translation from object to context->handles_vma.
 124          */
 125         struct list_head lut_list;
 126
 127         /** Stolen memory for this object, instead of being backed by shmem. */
 128         struct drm_mm_node *stolen;
 129         union {
 130                 struct rcu_head rcu;
 131                 struct llist_node freed;
 132         };
 133
 134         /**
 135          * Whether the object is currently in the GGTT mmap.
 136          */
 137         unsigned int userfault_count;
 138         struct list_head userfault_link;
 139
 140         struct list_head batch_pool_link;
 141         I915_SELFTEST_DECLARE(struct list_head st_link);
 142
 143         unsigned long flags;
 144
 145         /**
 146          * Have we taken a reference for the object for incomplete GPU
 147          * activity?
 148          */
 149 #define I915_BO_ACTIVE_REF 0
 150
 151         /*
 152          * Is the object to be mapped as read-only to the GPU
 153          * Only honoured if hardware has relevant pte bit
 154          */
 155         unsigned int cache_level:3;
 156         unsigned int cache_coherent:2;
 157 #define I915_BO_CACHE_COHERENT_FOR_READ BIT(0)
 158 #define I915_BO_CACHE_COHERENT_FOR_WRITE BIT(1)
 159         unsigned int cache_dirty:1;
 160
 161         /**
 162          * @read_domains: Read memory domains.
 163          *
 164          * These monitor which caches contain read/write data related to the
 165          * object. When transitioning from one set of domains to another,
 166          * the driver is called to ensure that caches are suitably flushed and
 167          * invalidated.
 168          */
 169         u16 read_domains;
 170
 171         /**
 172          * @write_domain: Corresponding unique write memory domain.
 173          */
 174         u16 write_domain;
 175
 176         atomic_t frontbuffer_bits;
 177         unsigned int frontbuffer_ggtt_origin; /* write once */
 178         struct i915_active_request frontbuffer_write;
 179
 180         /** Current tiling stride for the object, if it's tiled. */
 181         unsigned int tiling_and_stride;
 182 #define FENCE_MINIMUM_STRIDE 128 /* See i915_tiling_ok() */
 183 #define TILING_MASK (FENCE_MINIMUM_STRIDE-1)
 184 #define STRIDE_MASK (~TILING_MASK)
 185
 186         /** Count of VMA actually bound by this object */
 187         unsigned int bind_count;
 188         unsigned int active_count;
 189         /** Count of how many global VMA are currently pinned for use by HW */
 190         unsigned int pin_global;
 191
 192         struct {
 193                 struct mutex lock; /* protects the pages and their use */
 194                 atomic_t pages_pin_count;
 195
 196                 struct sg_table *pages;
 197                 void *mapping;
 198
 199                 /* TODO: whack some of this into the error state */
 200                 struct i915_page_sizes {
 201                         /**
 202                          * The sg mask of the pages sg_table. i.e the mask of
 203                          * of the lengths for each sg entry.
 204                          */
 205                         unsigned int phys;
 206
 207                         /**
 208                          * The gtt page sizes we are allowed to use given the
 209                          * sg mask and the supported page sizes. This will
 210                          * express the smallest unit we can use for the whole
 211                          * object, as well as the larger sizes we may be able
 212                          * to use opportunistically.
 213                          */
 214                         unsigned int sg;
 215
 216                         /**
 217                          * The actual gtt page size usage. Since we can have
 218                          * multiple vma associated with this object we need to
 219                          * prevent any trampling of state, hence a copy of this
 220                          * struct also lives in each vma, therefore the gtt
 221                          * value here should only be read/write through the vma.
 222                          */
 223                         unsigned int gtt;
 224                 } page_sizes;
 225
 226                 I915_SELFTEST_DECLARE(unsigned int page_mask);
 227
 228                 struct i915_gem_object_page_iter {
 229                         struct scatterlist *sg_pos;
 230                         unsigned int sg_idx; /* in pages, but 32bit eek! */
 231
 232                         struct radix_tree_root radix;
 233                         struct mutex lock; /* protects this cache */
 234                 } get_page;
 235
 236                 /**
 237                  * Element within i915->mm.unbound_list or i915->mm.bound_list,
 238                  * locked by i915->mm.obj_lock.
 239                  */
 240                 struct list_head link;
 241
 242                 /**
 243                  * Advice: are the backing pages purgeable?
 244                  */
 245                 unsigned int madv:2;
 246
 247                 /**
 248                  * This is set if the object has been written to since the
 249                  * pages were last acquired.
 250                  */
 251                 bool dirty:1;
 252
 253                 /**
 254                  * This is set if the object has been pinned due to unknown
 255                  * swizzling.
 256                  */
 257                 bool quirked:1;
 258         } mm;
 259
 260         /** Breadcrumb of last rendering to the buffer.
 261          * There can only be one writer, but we allow for multiple readers.
 262          * If there is a writer that necessarily implies that all other
 263          * read requests are complete - but we may only be lazily clearing
 264          * the read requests. A read request is naturally the most recent
 265          * request on a ring, so we may have two different write and read
 266          * requests on one ring where the write request is older than the
 267          * read request. This allows for the CPU to read from an active
 268          * buffer by only waiting for the write to complete.
 269          */
 270         struct reservation_object *resv;
 271
 272         /** References from framebuffers, locks out tiling changes. */
 273         unsigned int framebuffer_references;
 274
 275         /** Record of address bit 17 of each page at last unbind. */
 276         unsigned long *bit_17;
 277
 278         union {
 279                 struct i915_gem_userptr {
 280                         uintptr_t ptr;
 281
 282                         struct i915_mm_struct *mm;
 283                         struct i915_mmu_object *mmu_object;
 284                         struct work_struct *work;
 285                 } userptr;
 286
 287                 unsigned long scratch;
 288
 289                 void *gvt_info;
 290         };
 291
 292         /** for phys allocated objects */
 293         struct drm_dma_handle *phys_handle;
 294
 295         struct reservation_object __builtin_resv;
 296 };
 297
 298 static inline struct drm_i915_gem_object *
 299 to_intel_bo(struct drm_gem_object *gem)
 300 {
 301         /* Assert that to_intel_bo(NULL) == NULL */
 302         BUILD_BUG_ON(offsetof(struct drm_i915_gem_object, base));
 303
 304         return container_of(gem, struct drm_i915_gem_object, base);
 305 }
 306
 307 struct drm_i915_gem_object *i915_gem_object_alloc(void);
 308 void i915_gem_object_free(struct drm_i915_gem_object *obj);
 309
 310 /**
 311  * i915_gem_object_lookup_rcu - look up a temporary GEM object from its handle
 312  * @filp: DRM file private date
 313  * @handle: userspace handle
 314  *
 315  * Returns:
 316  *
 317  * A pointer to the object named by the handle if such exists on @filp, NULL
 318  * otherwise. This object is only valid whilst under the RCU read lock, and
 319  * note carefully the object may be in the process of being destroyed.
 320  */
 321 static inline struct drm_i915_gem_object *
 322 i915_gem_object_lookup_rcu(struct drm_file *file, u32 handle)
 323 {
 324 #ifdef CONFIG_LOCKDEP
 325         WARN_ON(debug_locks && !lock_is_held(&rcu_lock_map));
 326 #endif
 327         return idr_find(&file->object_idr, handle);
 328 }
 329
 330 static inline struct drm_i915_gem_object *
 331 i915_gem_object_lookup(struct drm_file *file, u32 handle)
 332 {
 333         struct drm_i915_gem_object *obj;
 334
 335         rcu_read_lock();
 336         obj = i915_gem_object_lookup_rcu(file, handle);
 337         if (obj && !kref_get_unless_zero(&obj->base.refcount))
 338                 obj = NULL;
 339         rcu_read_unlock();
 340
 341         return obj;
 342 }
 343
 344 __deprecated
 345 extern struct drm_gem_object *
 346 drm_gem_object_lookup(struct drm_file *file, u32 handle);
 347
 348 __attribute__((nonnull))
 349 static inline struct drm_i915_gem_object *
 350 i915_gem_object_get(struct drm_i915_gem_object *obj)
 351 {
 352         drm_gem_object_get(&obj->base);
 353         return obj;
 354 }
 355
 356 __attribute__((nonnull))
 357 static inline void
 358 i915_gem_object_put(struct drm_i915_gem_object *obj)
 359 {
 360         __drm_gem_object_put(&obj->base);
 361 }
 362
 363 static inline void i915_gem_object_lock(struct drm_i915_gem_object *obj)
 364 {
 365         reservation_object_lock(obj->resv, NULL);
 366 }
 367
 368 static inline void i915_gem_object_unlock(struct drm_i915_gem_object *obj)
 369 {
 370         reservation_object_unlock(obj->resv);
 371 }
 372
 373 static inline void
 374 i915_gem_object_set_readonly(struct drm_i915_gem_object *obj)
 375 {
 376         obj->base.vma_node.readonly = true;
 377 }
 378
 379 static inline bool
 380 i915_gem_object_is_readonly(const struct drm_i915_gem_object *obj)
 381 {
 382         return obj->base.vma_node.readonly;
 383 }
 384
 385 static inline bool
 386 i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj)
 387 {
 388         return obj->ops->flags & I915_GEM_OBJECT_HAS_STRUCT_PAGE;
 389 }
 390
 391 static inline bool
 392 i915_gem_object_is_shrinkable(const struct drm_i915_gem_object *obj)
 393 {
 394         return obj->ops->flags & I915_GEM_OBJECT_IS_SHRINKABLE;
 395 }
 396
 397 static inline bool
 398 i915_gem_object_is_proxy(const struct drm_i915_gem_object *obj)
 399 {
 400         return obj->ops->flags & I915_GEM_OBJECT_IS_PROXY;
 401 }
 402
 403 static inline bool
 404 i915_gem_object_needs_async_cancel(const struct drm_i915_gem_object *obj)
 405 {
 406         return obj->ops->flags & I915_GEM_OBJECT_ASYNC_CANCEL;
 407 }
 408
 409 static inline bool
 410 i915_gem_object_is_active(const struct drm_i915_gem_object *obj)
 411 {
 412         return obj->active_count;
 413 }
 414
 415 static inline bool
 416 i915_gem_object_has_active_reference(const struct drm_i915_gem_object *obj)
 417 {
 418         return test_bit(I915_BO_ACTIVE_REF, &obj->flags);
 419 }
 420
 421 static inline void
 422 i915_gem_object_set_active_reference(struct drm_i915_gem_object *obj)
 423 {
 424         lockdep_assert_held(&obj->base.dev->struct_mutex);
 425         __set_bit(I915_BO_ACTIVE_REF, &obj->flags);
 426 }
 427
 428 static inline void
 429 i915_gem_object_clear_active_reference(struct drm_i915_gem_object *obj)
 430 {
 431         lockdep_assert_held(&obj->base.dev->struct_mutex);
 432         __clear_bit(I915_BO_ACTIVE_REF, &obj->flags);
 433 }
 434
 435 void __i915_gem_object_release_unless_active(struct drm_i915_gem_object *obj);
 436
 437 static inline bool
 438 i915_gem_object_is_framebuffer(const struct drm_i915_gem_object *obj)
 439 {
 440         return READ_ONCE(obj->framebuffer_references);
 441 }
 442
 443 static inline unsigned int
 444 i915_gem_object_get_tiling(const struct drm_i915_gem_object *obj)
 445 {
 446         return obj->tiling_and_stride & TILING_MASK;
 447 }
 448
 449 static inline bool
 450 i915_gem_object_is_tiled(const struct drm_i915_gem_object *obj)
 451 {
 452         return i915_gem_object_get_tiling(obj) != I915_TILING_NONE;
 453 }
 454
 455 static inline unsigned int
 456 i915_gem_object_get_stride(const struct drm_i915_gem_object *obj)
 457 {
 458         return obj->tiling_and_stride & STRIDE_MASK;
 459 }
 460
 461 static inline unsigned int
 462 i915_gem_tile_height(unsigned int tiling)
 463 {
 464         GEM_BUG_ON(!tiling);
 465         return tiling == I915_TILING_Y ? 32 : 8;
 466 }
 467
 468 static inline unsigned int
 469 i915_gem_object_get_tile_height(const struct drm_i915_gem_object *obj)
 470 {
 471         return i915_gem_tile_height(i915_gem_object_get_tiling(obj));
 472 }
 473
 474 static inline unsigned int
 475 i915_gem_object_get_tile_row_size(const struct drm_i915_gem_object *obj)
 476 {
 477         return (i915_gem_object_get_stride(obj) *
 478                 i915_gem_object_get_tile_height(obj));
 479 }
 480
 481 int i915_gem_object_set_tiling(struct drm_i915_gem_object *obj,
 482                                unsigned int tiling, unsigned int stride);
 483
 484 static inline struct intel_engine_cs *
 485 i915_gem_object_last_write_engine(struct drm_i915_gem_object *obj)
 486 {
 487         struct intel_engine_cs *engine = NULL;
 488         struct dma_fence *fence;
 489
 490         rcu_read_lock();
 491         fence = reservation_object_get_excl_rcu(obj->resv);
 492         rcu_read_unlock();
 493
 494         if (fence && dma_fence_is_i915(fence) && !dma_fence_is_signaled(fence))
 495                 engine = to_request(fence)->engine;
 496         dma_fence_put(fence);
 497
 498         return engine;
 499 }
 500
 501 void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 502                                          unsigned int cache_level);
 503 void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj);
 504
 505 void __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj,
 506                                      struct sg_table *pages,
 507                                      bool needs_clflush);
 508
 509 #endif