Commit | Line | Data |
---|---|---|
b42fe9ca JL |
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> | |
2f80d7bd JN |
32 | #include <drm/drm_file.h> |
33 | #include <drm/drm_device.h> | |
b42fe9ca JL |
34 | |
35 | #include <drm/i915_drm.h> | |
36 | ||
e61e0f51 | 37 | #include "i915_request.h" |
8d28ba45 CW |
38 | #include "i915_selftest.h" |
39 | ||
b8f55be6 CW |
40 | struct drm_i915_gem_object; |
41 | ||
d1b48c1e CW |
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 | ||
b42fe9ca JL |
55 | struct drm_i915_gem_object_ops { |
56 | unsigned int flags; | |
a03f395a TZ |
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) | |
484d9a84 | 60 | #define I915_GEM_OBJECT_ASYNC_CANCEL BIT(3) |
b42fe9ca JL |
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 | */ | |
b91b09ee | 75 | int (*get_pages)(struct drm_i915_gem_object *); |
b42fe9ca JL |
76 | void (*put_pages)(struct drm_i915_gem_object *, struct sg_table *); |
77 | ||
7c55e2c5 CW |
78 | int (*pwrite)(struct drm_i915_gem_object *, |
79 | const struct drm_i915_gem_pwrite *); | |
80 | ||
b42fe9ca JL |
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 | ||
b5a82425 CW |
90 | /** |
91 | * @vma_list: List of VMAs backed by this object | |
92 | * | |
93 | * The VMA on this list are ordered by type, all GGTT vma are placed | |
94 | * at the head and all ppGTT vma are placed at the tail. The different | |
95 | * types of GGTT vma are unordered between themselves, use the | |
96 | * @vma_tree (which has a defined order between all VMA) to find an | |
97 | * exact match. | |
98 | */ | |
b42fe9ca | 99 | struct list_head vma_list; |
b5a82425 CW |
100 | /** |
101 | * @vma_tree: Ordered tree of VMAs backed by this object | |
102 | * | |
103 | * All VMA created for this object are placed in the @vma_tree for | |
104 | * fast retrieval via a binary search in i915_vma_instance(). | |
105 | * They are also added to @vma_list for easy iteration. | |
106 | */ | |
b42fe9ca | 107 | struct rb_root vma_tree; |
d1b48c1e CW |
108 | |
109 | /** | |
110 | * @lut_list: List of vma lookup entries in use for this object. | |
111 | * | |
112 | * If this object is closed, we need to remove all of its VMA from | |
113 | * the fast lookup index in associated contexts; @lut_list provides | |
114 | * this translation from object to context->handles_vma. | |
115 | */ | |
116 | struct list_head lut_list; | |
b42fe9ca JL |
117 | |
118 | /** Stolen memory for this object, instead of being backed by shmem. */ | |
119 | struct drm_mm_node *stolen; | |
b42fe9ca JL |
120 | union { |
121 | struct rcu_head rcu; | |
122 | struct llist_node freed; | |
123 | }; | |
124 | ||
125 | /** | |
126 | * Whether the object is currently in the GGTT mmap. | |
127 | */ | |
a65adaf8 | 128 | unsigned int userfault_count; |
b42fe9ca JL |
129 | struct list_head userfault_link; |
130 | ||
b42fe9ca | 131 | struct list_head batch_pool_link; |
8d28ba45 | 132 | I915_SELFTEST_DECLARE(struct list_head st_link); |
b42fe9ca JL |
133 | |
134 | unsigned long flags; | |
135 | ||
136 | /** | |
137 | * Have we taken a reference for the object for incomplete GPU | |
138 | * activity? | |
139 | */ | |
140 | #define I915_BO_ACTIVE_REF 0 | |
141 | ||
142 | /* | |
143 | * Is the object to be mapped as read-only to the GPU | |
144 | * Only honoured if hardware has relevant pte bit | |
145 | */ | |
b42fe9ca | 146 | unsigned int cache_level:3; |
b8f55be6 CW |
147 | unsigned int cache_coherent:2; |
148 | #define I915_BO_CACHE_COHERENT_FOR_READ BIT(0) | |
149 | #define I915_BO_CACHE_COHERENT_FOR_WRITE BIT(1) | |
b42fe9ca JL |
150 | unsigned int cache_dirty:1; |
151 | ||
c0a51fd0 CK |
152 | /** |
153 | * @read_domains: Read memory domains. | |
154 | * | |
155 | * These monitor which caches contain read/write data related to the | |
156 | * object. When transitioning from one set of domains to another, | |
157 | * the driver is called to ensure that caches are suitably flushed and | |
158 | * invalidated. | |
159 | */ | |
160 | u16 read_domains; | |
161 | ||
162 | /** | |
163 | * @write_domain: Corresponding unique write memory domain. | |
164 | */ | |
165 | u16 write_domain; | |
166 | ||
b42fe9ca JL |
167 | atomic_t frontbuffer_bits; |
168 | unsigned int frontbuffer_ggtt_origin; /* write once */ | |
5b8c8aec | 169 | struct i915_gem_active frontbuffer_write; |
b42fe9ca JL |
170 | |
171 | /** Current tiling stride for the object, if it's tiled. */ | |
172 | unsigned int tiling_and_stride; | |
173 | #define FENCE_MINIMUM_STRIDE 128 /* See i915_tiling_ok() */ | |
174 | #define TILING_MASK (FENCE_MINIMUM_STRIDE-1) | |
175 | #define STRIDE_MASK (~TILING_MASK) | |
176 | ||
177 | /** Count of VMA actually bound by this object */ | |
178 | unsigned int bind_count; | |
179 | unsigned int active_count; | |
bd3d2252 CW |
180 | /** Count of how many global VMA are currently pinned for use by HW */ |
181 | unsigned int pin_global; | |
b42fe9ca JL |
182 | |
183 | struct { | |
184 | struct mutex lock; /* protects the pages and their use */ | |
185 | atomic_t pages_pin_count; | |
186 | ||
187 | struct sg_table *pages; | |
188 | void *mapping; | |
189 | ||
d9ec12f8 | 190 | /* TODO: whack some of this into the error state */ |
a5c08166 MA |
191 | struct i915_page_sizes { |
192 | /** | |
193 | * The sg mask of the pages sg_table. i.e the mask of | |
194 | * of the lengths for each sg entry. | |
195 | */ | |
196 | unsigned int phys; | |
197 | ||
198 | /** | |
199 | * The gtt page sizes we are allowed to use given the | |
200 | * sg mask and the supported page sizes. This will | |
201 | * express the smallest unit we can use for the whole | |
202 | * object, as well as the larger sizes we may be able | |
203 | * to use opportunistically. | |
204 | */ | |
205 | unsigned int sg; | |
d9ec12f8 MA |
206 | |
207 | /** | |
208 | * The actual gtt page size usage. Since we can have | |
209 | * multiple vma associated with this object we need to | |
210 | * prevent any trampling of state, hence a copy of this | |
211 | * struct also lives in each vma, therefore the gtt | |
212 | * value here should only be read/write through the vma. | |
213 | */ | |
214 | unsigned int gtt; | |
a5c08166 MA |
215 | } page_sizes; |
216 | ||
4049866f MA |
217 | I915_SELFTEST_DECLARE(unsigned int page_mask); |
218 | ||
b42fe9ca JL |
219 | struct i915_gem_object_page_iter { |
220 | struct scatterlist *sg_pos; | |
221 | unsigned int sg_idx; /* in pages, but 32bit eek! */ | |
222 | ||
223 | struct radix_tree_root radix; | |
224 | struct mutex lock; /* protects this cache */ | |
225 | } get_page; | |
226 | ||
f2123818 CW |
227 | /** |
228 | * Element within i915->mm.unbound_list or i915->mm.bound_list, | |
229 | * locked by i915->mm.obj_lock. | |
230 | */ | |
231 | struct list_head link; | |
232 | ||
b42fe9ca JL |
233 | /** |
234 | * Advice: are the backing pages purgeable? | |
235 | */ | |
236 | unsigned int madv:2; | |
237 | ||
238 | /** | |
239 | * This is set if the object has been written to since the | |
240 | * pages were last acquired. | |
241 | */ | |
242 | bool dirty:1; | |
243 | ||
244 | /** | |
245 | * This is set if the object has been pinned due to unknown | |
246 | * swizzling. | |
247 | */ | |
248 | bool quirked:1; | |
249 | } mm; | |
250 | ||
251 | /** Breadcrumb of last rendering to the buffer. | |
252 | * There can only be one writer, but we allow for multiple readers. | |
253 | * If there is a writer that necessarily implies that all other | |
254 | * read requests are complete - but we may only be lazily clearing | |
255 | * the read requests. A read request is naturally the most recent | |
256 | * request on a ring, so we may have two different write and read | |
257 | * requests on one ring where the write request is older than the | |
258 | * read request. This allows for the CPU to read from an active | |
259 | * buffer by only waiting for the write to complete. | |
260 | */ | |
261 | struct reservation_object *resv; | |
262 | ||
263 | /** References from framebuffers, locks out tiling changes. */ | |
dd689287 | 264 | unsigned int framebuffer_references; |
b42fe9ca JL |
265 | |
266 | /** Record of address bit 17 of each page at last unbind. */ | |
267 | unsigned long *bit_17; | |
268 | ||
44653988 CW |
269 | union { |
270 | struct i915_gem_userptr { | |
271 | uintptr_t ptr; | |
44653988 CW |
272 | |
273 | struct i915_mm_struct *mm; | |
274 | struct i915_mmu_object *mmu_object; | |
275 | struct work_struct *work; | |
276 | } userptr; | |
277 | ||
278 | unsigned long scratch; | |
e546e281 TZ |
279 | |
280 | void *gvt_info; | |
44653988 | 281 | }; |
b42fe9ca JL |
282 | |
283 | /** for phys allocated objects */ | |
284 | struct drm_dma_handle *phys_handle; | |
285 | ||
286 | struct reservation_object __builtin_resv; | |
287 | }; | |
288 | ||
289 | static inline struct drm_i915_gem_object * | |
290 | to_intel_bo(struct drm_gem_object *gem) | |
291 | { | |
292 | /* Assert that to_intel_bo(NULL) == NULL */ | |
293 | BUILD_BUG_ON(offsetof(struct drm_i915_gem_object, base)); | |
294 | ||
295 | return container_of(gem, struct drm_i915_gem_object, base); | |
296 | } | |
297 | ||
298 | /** | |
299 | * i915_gem_object_lookup_rcu - look up a temporary GEM object from its handle | |
300 | * @filp: DRM file private date | |
301 | * @handle: userspace handle | |
302 | * | |
303 | * Returns: | |
304 | * | |
305 | * A pointer to the object named by the handle if such exists on @filp, NULL | |
306 | * otherwise. This object is only valid whilst under the RCU read lock, and | |
307 | * note carefully the object may be in the process of being destroyed. | |
308 | */ | |
309 | static inline struct drm_i915_gem_object * | |
310 | i915_gem_object_lookup_rcu(struct drm_file *file, u32 handle) | |
311 | { | |
312 | #ifdef CONFIG_LOCKDEP | |
313 | WARN_ON(debug_locks && !lock_is_held(&rcu_lock_map)); | |
314 | #endif | |
315 | return idr_find(&file->object_idr, handle); | |
316 | } | |
317 | ||
318 | static inline struct drm_i915_gem_object * | |
319 | i915_gem_object_lookup(struct drm_file *file, u32 handle) | |
320 | { | |
321 | struct drm_i915_gem_object *obj; | |
322 | ||
323 | rcu_read_lock(); | |
324 | obj = i915_gem_object_lookup_rcu(file, handle); | |
325 | if (obj && !kref_get_unless_zero(&obj->base.refcount)) | |
326 | obj = NULL; | |
327 | rcu_read_unlock(); | |
328 | ||
329 | return obj; | |
330 | } | |
331 | ||
332 | __deprecated | |
333 | extern struct drm_gem_object * | |
334 | drm_gem_object_lookup(struct drm_file *file, u32 handle); | |
335 | ||
336 | __attribute__((nonnull)) | |
337 | static inline struct drm_i915_gem_object * | |
338 | i915_gem_object_get(struct drm_i915_gem_object *obj) | |
339 | { | |
0f67706e | 340 | drm_gem_object_get(&obj->base); |
b42fe9ca JL |
341 | return obj; |
342 | } | |
343 | ||
b42fe9ca JL |
344 | __attribute__((nonnull)) |
345 | static inline void | |
346 | i915_gem_object_put(struct drm_i915_gem_object *obj) | |
347 | { | |
55f95c27 | 348 | __drm_gem_object_put(&obj->base); |
b42fe9ca JL |
349 | } |
350 | ||
dd689287 CW |
351 | static inline void i915_gem_object_lock(struct drm_i915_gem_object *obj) |
352 | { | |
353 | reservation_object_lock(obj->resv, NULL); | |
354 | } | |
355 | ||
356 | static inline void i915_gem_object_unlock(struct drm_i915_gem_object *obj) | |
357 | { | |
358 | reservation_object_unlock(obj->resv); | |
359 | } | |
360 | ||
3e977ac6 CW |
361 | static inline void |
362 | i915_gem_object_set_readonly(struct drm_i915_gem_object *obj) | |
363 | { | |
364 | obj->base.vma_node.readonly = true; | |
365 | } | |
366 | ||
367 | static inline bool | |
368 | i915_gem_object_is_readonly(const struct drm_i915_gem_object *obj) | |
369 | { | |
370 | return obj->base.vma_node.readonly; | |
371 | } | |
372 | ||
b42fe9ca JL |
373 | static inline bool |
374 | i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj) | |
375 | { | |
376 | return obj->ops->flags & I915_GEM_OBJECT_HAS_STRUCT_PAGE; | |
377 | } | |
378 | ||
379 | static inline bool | |
380 | i915_gem_object_is_shrinkable(const struct drm_i915_gem_object *obj) | |
381 | { | |
382 | return obj->ops->flags & I915_GEM_OBJECT_IS_SHRINKABLE; | |
383 | } | |
384 | ||
a03f395a TZ |
385 | static inline bool |
386 | i915_gem_object_is_proxy(const struct drm_i915_gem_object *obj) | |
387 | { | |
388 | return obj->ops->flags & I915_GEM_OBJECT_IS_PROXY; | |
389 | } | |
390 | ||
484d9a84 CW |
391 | static inline bool |
392 | i915_gem_object_needs_async_cancel(const struct drm_i915_gem_object *obj) | |
393 | { | |
394 | return obj->ops->flags & I915_GEM_OBJECT_ASYNC_CANCEL; | |
395 | } | |
396 | ||
b42fe9ca JL |
397 | static inline bool |
398 | i915_gem_object_is_active(const struct drm_i915_gem_object *obj) | |
399 | { | |
400 | return obj->active_count; | |
401 | } | |
402 | ||
403 | static inline bool | |
404 | i915_gem_object_has_active_reference(const struct drm_i915_gem_object *obj) | |
405 | { | |
406 | return test_bit(I915_BO_ACTIVE_REF, &obj->flags); | |
407 | } | |
408 | ||
409 | static inline void | |
410 | i915_gem_object_set_active_reference(struct drm_i915_gem_object *obj) | |
411 | { | |
412 | lockdep_assert_held(&obj->base.dev->struct_mutex); | |
413 | __set_bit(I915_BO_ACTIVE_REF, &obj->flags); | |
414 | } | |
415 | ||
416 | static inline void | |
417 | i915_gem_object_clear_active_reference(struct drm_i915_gem_object *obj) | |
418 | { | |
419 | lockdep_assert_held(&obj->base.dev->struct_mutex); | |
420 | __clear_bit(I915_BO_ACTIVE_REF, &obj->flags); | |
421 | } | |
422 | ||
423 | void __i915_gem_object_release_unless_active(struct drm_i915_gem_object *obj); | |
424 | ||
dd689287 CW |
425 | static inline bool |
426 | i915_gem_object_is_framebuffer(const struct drm_i915_gem_object *obj) | |
427 | { | |
428 | return READ_ONCE(obj->framebuffer_references); | |
429 | } | |
430 | ||
b42fe9ca | 431 | static inline unsigned int |
d899aceb | 432 | i915_gem_object_get_tiling(const struct drm_i915_gem_object *obj) |
b42fe9ca JL |
433 | { |
434 | return obj->tiling_and_stride & TILING_MASK; | |
435 | } | |
436 | ||
437 | static inline bool | |
d899aceb | 438 | i915_gem_object_is_tiled(const struct drm_i915_gem_object *obj) |
b42fe9ca JL |
439 | { |
440 | return i915_gem_object_get_tiling(obj) != I915_TILING_NONE; | |
441 | } | |
442 | ||
443 | static inline unsigned int | |
d899aceb | 444 | i915_gem_object_get_stride(const struct drm_i915_gem_object *obj) |
b42fe9ca JL |
445 | { |
446 | return obj->tiling_and_stride & STRIDE_MASK; | |
447 | } | |
448 | ||
6649a0b6 CW |
449 | static inline unsigned int |
450 | i915_gem_tile_height(unsigned int tiling) | |
451 | { | |
452 | GEM_BUG_ON(!tiling); | |
453 | return tiling == I915_TILING_Y ? 32 : 8; | |
454 | } | |
455 | ||
456 | static inline unsigned int | |
d899aceb | 457 | i915_gem_object_get_tile_height(const struct drm_i915_gem_object *obj) |
6649a0b6 CW |
458 | { |
459 | return i915_gem_tile_height(i915_gem_object_get_tiling(obj)); | |
460 | } | |
461 | ||
462 | static inline unsigned int | |
d899aceb | 463 | i915_gem_object_get_tile_row_size(const struct drm_i915_gem_object *obj) |
6649a0b6 CW |
464 | { |
465 | return (i915_gem_object_get_stride(obj) * | |
466 | i915_gem_object_get_tile_height(obj)); | |
467 | } | |
468 | ||
957870f9 CW |
469 | int i915_gem_object_set_tiling(struct drm_i915_gem_object *obj, |
470 | unsigned int tiling, unsigned int stride); | |
471 | ||
b42fe9ca JL |
472 | static inline struct intel_engine_cs * |
473 | i915_gem_object_last_write_engine(struct drm_i915_gem_object *obj) | |
474 | { | |
475 | struct intel_engine_cs *engine = NULL; | |
476 | struct dma_fence *fence; | |
477 | ||
478 | rcu_read_lock(); | |
479 | fence = reservation_object_get_excl_rcu(obj->resv); | |
480 | rcu_read_unlock(); | |
481 | ||
482 | if (fence && dma_fence_is_i915(fence) && !dma_fence_is_signaled(fence)) | |
483 | engine = to_request(fence)->engine; | |
484 | dma_fence_put(fence); | |
485 | ||
486 | return engine; | |
487 | } | |
488 | ||
b8f55be6 CW |
489 | void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj, |
490 | unsigned int cache_level); | |
5a97bcc6 CW |
491 | void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj); |
492 | ||
b42fe9ca JL |
493 | #endif |
494 |