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_VMA_H__
26 #define __I915_VMA_H__
27
28 #include <linux/io-mapping.h>
29 #include <linux/rbtree.h>
30
31 #include <drm/drm_mm.h>
32
33 #include "gt/intel_ggtt_fencing.h"
34 #include "gem/i915_gem_object.h"
35
36 #include "i915_gem_gtt.h"
37
38 #include "i915_active.h"
39 #include "i915_request.h"
40 #include "i915_vma_resource.h"
41 #include "i915_vma_types.h"
42
43 struct i915_vma *
44 i915_vma_instance(struct drm_i915_gem_object *obj,
45 struct i915_address_space *vm,
46 const struct i915_gtt_view *view);
47
48 void i915_vma_unpin_and_release(struct i915_vma **p_vma, unsigned int flags);
49 #define I915_VMA_RELEASE_MAP BIT(0)
50
i915_vma_is_active(const struct i915_vma * vma)51 static inline bool i915_vma_is_active(const struct i915_vma *vma)
52 {
53 return !i915_active_is_idle(&vma->active);
54 }
55
56 /* do not reserve memory to prevent deadlocks */
57 #define __EXEC_OBJECT_NO_RESERVE BIT(31)
58 #define __EXEC_OBJECT_NO_REQUEST_AWAIT BIT(30)
59
60 int __must_check _i915_vma_move_to_active(struct i915_vma *vma,
61 struct i915_request *rq,
62 struct dma_fence *fence,
63 unsigned int flags);
64 static inline int __must_check
i915_vma_move_to_active(struct i915_vma * vma,struct i915_request * rq,unsigned int flags)65 i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq,
66 unsigned int flags)
67 {
68 return _i915_vma_move_to_active(vma, rq, &rq->fence, flags);
69 }
70
71 #ifdef __linux__
72 #define __i915_vma_flags(v) ((unsigned long *)&(v)->flags.counter)
73 #else
74 #define __i915_vma_flags(v) ((unsigned long *)&(v)->flags)
75 #endif
76
i915_vma_is_ggtt(const struct i915_vma * vma)77 static inline bool i915_vma_is_ggtt(const struct i915_vma *vma)
78 {
79 return test_bit(I915_VMA_GGTT_BIT, __i915_vma_flags(vma));
80 }
81
i915_vma_is_dpt(const struct i915_vma * vma)82 static inline bool i915_vma_is_dpt(const struct i915_vma *vma)
83 {
84 return i915_is_dpt(vma->vm);
85 }
86
i915_vma_has_ggtt_write(const struct i915_vma * vma)87 static inline bool i915_vma_has_ggtt_write(const struct i915_vma *vma)
88 {
89 return test_bit(I915_VMA_GGTT_WRITE_BIT, __i915_vma_flags(vma));
90 }
91
i915_vma_set_ggtt_write(struct i915_vma * vma)92 static inline void i915_vma_set_ggtt_write(struct i915_vma *vma)
93 {
94 GEM_BUG_ON(!i915_vma_is_ggtt(vma));
95 set_bit(I915_VMA_GGTT_WRITE_BIT, __i915_vma_flags(vma));
96 }
97
i915_vma_unset_ggtt_write(struct i915_vma * vma)98 static inline bool i915_vma_unset_ggtt_write(struct i915_vma *vma)
99 {
100 return test_and_clear_bit(I915_VMA_GGTT_WRITE_BIT,
101 __i915_vma_flags(vma));
102 }
103
104 void i915_vma_flush_writes(struct i915_vma *vma);
105
i915_vma_is_map_and_fenceable(const struct i915_vma * vma)106 static inline bool i915_vma_is_map_and_fenceable(const struct i915_vma *vma)
107 {
108 return test_bit(I915_VMA_CAN_FENCE_BIT, __i915_vma_flags(vma));
109 }
110
i915_vma_set_userfault(struct i915_vma * vma)111 static inline bool i915_vma_set_userfault(struct i915_vma *vma)
112 {
113 GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
114 return test_and_set_bit(I915_VMA_USERFAULT_BIT, __i915_vma_flags(vma));
115 }
116
i915_vma_unset_userfault(struct i915_vma * vma)117 static inline void i915_vma_unset_userfault(struct i915_vma *vma)
118 {
119 return clear_bit(I915_VMA_USERFAULT_BIT, __i915_vma_flags(vma));
120 }
121
i915_vma_has_userfault(const struct i915_vma * vma)122 static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
123 {
124 return test_bit(I915_VMA_USERFAULT_BIT, __i915_vma_flags(vma));
125 }
126
i915_vma_is_closed(const struct i915_vma * vma)127 static inline bool i915_vma_is_closed(const struct i915_vma *vma)
128 {
129 return !list_empty(&vma->closed_link);
130 }
131
132 /* Internal use only. */
__i915_vma_size(const struct i915_vma * vma)133 static inline u64 __i915_vma_size(const struct i915_vma *vma)
134 {
135 return vma->node.size - 2 * vma->guard;
136 }
137
138 /**
139 * i915_vma_size - Obtain the va range size of the vma
140 * @vma: The vma
141 *
142 * GPU virtual address space may be allocated with padding. This
143 * function returns the effective virtual address range size
144 * with padding subtracted.
145 *
146 * Return: The effective virtual address range size.
147 */
i915_vma_size(const struct i915_vma * vma)148 static inline u64 i915_vma_size(const struct i915_vma *vma)
149 {
150 GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
151 return __i915_vma_size(vma);
152 }
153
154 /* Internal use only. */
__i915_vma_offset(const struct i915_vma * vma)155 static inline u64 __i915_vma_offset(const struct i915_vma *vma)
156 {
157 /* The actual start of the vma->pages is after the guard pages. */
158 return vma->node.start + vma->guard;
159 }
160
161 /**
162 * i915_vma_offset - Obtain the va offset of the vma
163 * @vma: The vma
164 *
165 * GPU virtual address space may be allocated with padding. This
166 * function returns the effective virtual address offset the gpu
167 * should use to access the bound data.
168 *
169 * Return: The effective virtual address offset.
170 */
i915_vma_offset(const struct i915_vma * vma)171 static inline u64 i915_vma_offset(const struct i915_vma *vma)
172 {
173 GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
174 return __i915_vma_offset(vma);
175 }
176
i915_ggtt_offset(const struct i915_vma * vma)177 static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
178 {
179 GEM_BUG_ON(!i915_vma_is_ggtt(vma));
180 GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
181 GEM_BUG_ON(upper_32_bits(i915_vma_offset(vma)));
182 GEM_BUG_ON(upper_32_bits(i915_vma_offset(vma) +
183 i915_vma_size(vma) - 1));
184 return lower_32_bits(i915_vma_offset(vma));
185 }
186
i915_ggtt_pin_bias(struct i915_vma * vma)187 static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma)
188 {
189 return i915_vm_to_ggtt(vma->vm)->pin_bias;
190 }
191
i915_vma_get(struct i915_vma * vma)192 static inline struct i915_vma *i915_vma_get(struct i915_vma *vma)
193 {
194 i915_gem_object_get(vma->obj);
195 return vma;
196 }
197
i915_vma_tryget(struct i915_vma * vma)198 static inline struct i915_vma *i915_vma_tryget(struct i915_vma *vma)
199 {
200 if (likely(kref_get_unless_zero(&vma->obj->base.refcount)))
201 return vma;
202
203 return NULL;
204 }
205
i915_vma_put(struct i915_vma * vma)206 static inline void i915_vma_put(struct i915_vma *vma)
207 {
208 i915_gem_object_put(vma->obj);
209 }
210
211 static inline long
i915_vma_compare(struct i915_vma * vma,struct i915_address_space * vm,const struct i915_gtt_view * view)212 i915_vma_compare(struct i915_vma *vma,
213 struct i915_address_space *vm,
214 const struct i915_gtt_view *view)
215 {
216 ptrdiff_t cmp;
217
218 GEM_BUG_ON(view && !i915_is_ggtt_or_dpt(vm));
219
220 cmp = ptrdiff(vma->vm, vm);
221 if (cmp)
222 return cmp;
223
224 BUILD_BUG_ON(I915_GTT_VIEW_NORMAL != 0);
225 cmp = vma->gtt_view.type;
226 if (!view)
227 return cmp;
228
229 cmp -= view->type;
230 if (cmp)
231 return cmp;
232
233 assert_i915_gem_gtt_types();
234
235 /* gtt_view.type also encodes its size so that we both distinguish
236 * different views using it as a "type" and also use a compact (no
237 * accessing of uninitialised padding bytes) memcmp without storing
238 * an extra parameter or adding more code.
239 *
240 * To ensure that the memcmp is valid for all branches of the union,
241 * even though the code looks like it is just comparing one branch,
242 * we assert above that all branches have the same address, and that
243 * each branch has a unique type/size.
244 */
245 BUILD_BUG_ON(I915_GTT_VIEW_NORMAL >= I915_GTT_VIEW_PARTIAL);
246 BUILD_BUG_ON(I915_GTT_VIEW_PARTIAL >= I915_GTT_VIEW_ROTATED);
247 BUILD_BUG_ON(I915_GTT_VIEW_ROTATED >= I915_GTT_VIEW_REMAPPED);
248 BUILD_BUG_ON(offsetof(typeof(*view), rotated) !=
249 offsetof(typeof(*view), partial));
250 BUILD_BUG_ON(offsetof(typeof(*view), rotated) !=
251 offsetof(typeof(*view), remapped));
252 return memcmp(&vma->gtt_view.partial, &view->partial, view->type);
253 }
254
255 struct i915_vma_work *i915_vma_work(void);
256 int i915_vma_bind(struct i915_vma *vma,
257 unsigned int pat_index,
258 u32 flags,
259 struct i915_vma_work *work,
260 struct i915_vma_resource *vma_res);
261
262 bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long color);
263 bool i915_vma_misplaced(const struct i915_vma *vma,
264 u64 size, u64 alignment, u64 flags);
265 void __i915_vma_set_map_and_fenceable(struct i915_vma *vma);
266 void i915_vma_revoke_mmap(struct i915_vma *vma);
267 void vma_invalidate_tlb(struct i915_address_space *vm, u32 *tlb);
268 struct dma_fence *__i915_vma_evict(struct i915_vma *vma, bool async);
269 int __i915_vma_unbind(struct i915_vma *vma);
270 int __must_check i915_vma_unbind(struct i915_vma *vma);
271 int __must_check i915_vma_unbind_async(struct i915_vma *vma, bool trylock_vm);
272 int __must_check i915_vma_unbind_unlocked(struct i915_vma *vma);
273 void i915_vma_unlink_ctx(struct i915_vma *vma);
274 void i915_vma_close(struct i915_vma *vma);
275 void i915_vma_reopen(struct i915_vma *vma);
276
277 void i915_vma_destroy_locked(struct i915_vma *vma);
278 void i915_vma_destroy(struct i915_vma *vma);
279
280 #define assert_vma_held(vma) dma_resv_assert_held((vma)->obj->base.resv)
281
i915_vma_lock(struct i915_vma * vma)282 static inline void i915_vma_lock(struct i915_vma *vma)
283 {
284 dma_resv_lock(vma->obj->base.resv, NULL);
285 }
286
i915_vma_unlock(struct i915_vma * vma)287 static inline void i915_vma_unlock(struct i915_vma *vma)
288 {
289 dma_resv_unlock(vma->obj->base.resv);
290 }
291
292 int __must_check
293 i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
294 u64 size, u64 alignment, u64 flags);
295
296 static inline int __must_check
i915_vma_pin(struct i915_vma * vma,u64 size,u64 alignment,u64 flags)297 i915_vma_pin(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
298 {
299 struct i915_gem_ww_ctx ww;
300 int err;
301
302 i915_gem_ww_ctx_init(&ww, true);
303 retry:
304 err = i915_gem_object_lock(vma->obj, &ww);
305 if (!err)
306 err = i915_vma_pin_ww(vma, &ww, size, alignment, flags);
307 if (err == -EDEADLK) {
308 err = i915_gem_ww_ctx_backoff(&ww);
309 if (!err)
310 goto retry;
311 }
312 i915_gem_ww_ctx_fini(&ww);
313
314 return err;
315 }
316
317 int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
318 u32 align, unsigned int flags);
319
i915_vma_pin_count(const struct i915_vma * vma)320 static inline int i915_vma_pin_count(const struct i915_vma *vma)
321 {
322 return atomic_read(&vma->flags) & I915_VMA_PIN_MASK;
323 }
324
i915_vma_is_pinned(const struct i915_vma * vma)325 static inline bool i915_vma_is_pinned(const struct i915_vma *vma)
326 {
327 return i915_vma_pin_count(vma);
328 }
329
__i915_vma_pin(struct i915_vma * vma)330 static inline void __i915_vma_pin(struct i915_vma *vma)
331 {
332 atomic_inc(&vma->flags);
333 GEM_BUG_ON(!i915_vma_is_pinned(vma));
334 }
335
__i915_vma_unpin(struct i915_vma * vma)336 static inline void __i915_vma_unpin(struct i915_vma *vma)
337 {
338 GEM_BUG_ON(!i915_vma_is_pinned(vma));
339 atomic_dec(&vma->flags);
340 }
341
i915_vma_unpin(struct i915_vma * vma)342 static inline void i915_vma_unpin(struct i915_vma *vma)
343 {
344 GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
345 __i915_vma_unpin(vma);
346 }
347
i915_vma_is_bound(const struct i915_vma * vma,unsigned int where)348 static inline bool i915_vma_is_bound(const struct i915_vma *vma,
349 unsigned int where)
350 {
351 return atomic_read(&vma->flags) & where;
352 }
353
i915_node_color_differs(const struct drm_mm_node * node,unsigned long color)354 static inline bool i915_node_color_differs(const struct drm_mm_node *node,
355 unsigned long color)
356 {
357 return drm_mm_node_allocated(node) && node->color != color;
358 }
359
360 /**
361 * i915_vma_pin_iomap - calls ioremap_wc to map the GGTT VMA via the aperture
362 * @vma: VMA to iomap
363 *
364 * The passed in VMA has to be pinned in the global GTT mappable region.
365 * An extra pinning of the VMA is acquired for the return iomapping,
366 * the caller must call i915_vma_unpin_iomap to relinquish the pinning
367 * after the iomapping is no longer required.
368 *
369 * Returns a valid iomapped pointer or ERR_PTR.
370 */
371 void __iomem *i915_vma_pin_iomap(struct i915_vma *vma);
372
373 /**
374 * i915_vma_unpin_iomap - unpins the mapping returned from i915_vma_iomap
375 * @vma: VMA to unpin
376 *
377 * Unpins the previously iomapped VMA from i915_vma_pin_iomap().
378 *
379 * This function is only valid to be called on a VMA previously
380 * iomapped by the caller with i915_vma_pin_iomap().
381 */
382 void i915_vma_unpin_iomap(struct i915_vma *vma);
383
384 /**
385 * i915_vma_pin_fence - pin fencing state
386 * @vma: vma to pin fencing for
387 *
388 * This pins the fencing state (whether tiled or untiled) to make sure the
389 * vma (and its object) is ready to be used as a scanout target. Fencing
390 * status must be synchronize first by calling i915_vma_get_fence():
391 *
392 * The resulting fence pin reference must be released again with
393 * i915_vma_unpin_fence().
394 *
395 * Returns:
396 *
397 * True if the vma has a fence, false otherwise.
398 */
399 int __must_check i915_vma_pin_fence(struct i915_vma *vma);
400 void i915_vma_revoke_fence(struct i915_vma *vma);
401
402 int __i915_vma_pin_fence(struct i915_vma *vma);
403
__i915_vma_unpin_fence(struct i915_vma * vma)404 static inline void __i915_vma_unpin_fence(struct i915_vma *vma)
405 {
406 GEM_BUG_ON(atomic_read(&vma->fence->pin_count) <= 0);
407 atomic_dec(&vma->fence->pin_count);
408 }
409
410 /**
411 * i915_vma_unpin_fence - unpin fencing state
412 * @vma: vma to unpin fencing for
413 *
414 * This releases the fence pin reference acquired through
415 * i915_vma_pin_fence. It will handle both objects with and without an
416 * attached fence correctly, callers do not need to distinguish this.
417 */
418 static inline void
i915_vma_unpin_fence(struct i915_vma * vma)419 i915_vma_unpin_fence(struct i915_vma *vma)
420 {
421 if (vma->fence)
422 __i915_vma_unpin_fence(vma);
423 }
424
i915_vma_fence_id(const struct i915_vma * vma)425 static inline int i915_vma_fence_id(const struct i915_vma *vma)
426 {
427 return vma->fence ? vma->fence->id : -1;
428 }
429
430 void i915_vma_parked(struct intel_gt *gt);
431
i915_vma_is_scanout(const struct i915_vma * vma)432 static inline bool i915_vma_is_scanout(const struct i915_vma *vma)
433 {
434 return test_bit(I915_VMA_SCANOUT_BIT, __i915_vma_flags(vma));
435 }
436
i915_vma_mark_scanout(struct i915_vma * vma)437 static inline void i915_vma_mark_scanout(struct i915_vma *vma)
438 {
439 set_bit(I915_VMA_SCANOUT_BIT, __i915_vma_flags(vma));
440 }
441
i915_vma_clear_scanout(struct i915_vma * vma)442 static inline void i915_vma_clear_scanout(struct i915_vma *vma)
443 {
444 clear_bit(I915_VMA_SCANOUT_BIT, __i915_vma_flags(vma));
445 }
446
447 void i915_ggtt_clear_scanout(struct drm_i915_gem_object *obj);
448
449 #define for_each_until(cond) if (cond) break; else
450
451 /**
452 * for_each_ggtt_vma - Iterate over the GGTT VMA belonging to an object.
453 * @V: the #i915_vma iterator
454 * @OBJ: the #drm_i915_gem_object
455 *
456 * GGTT VMA are placed at the being of the object's vma_list, see
457 * vma_create(), so we can stop our walk as soon as we see a ppgtt VMA,
458 * or the list is empty ofc.
459 */
460 #define for_each_ggtt_vma(V, OBJ) \
461 list_for_each_entry(V, &(OBJ)->vma.list, obj_link) \
462 for_each_until(!i915_vma_is_ggtt(V))
463
464 struct i915_vma *i915_vma_make_unshrinkable(struct i915_vma *vma);
465 void i915_vma_make_shrinkable(struct i915_vma *vma);
466 void i915_vma_make_purgeable(struct i915_vma *vma);
467
468 int i915_vma_wait_for_bind(struct i915_vma *vma);
469
i915_vma_sync(struct i915_vma * vma)470 static inline int i915_vma_sync(struct i915_vma *vma)
471 {
472 /* Wait for the asynchronous bindings and pending GPU reads */
473 return i915_active_wait(&vma->active);
474 }
475
476 /**
477 * i915_vma_get_current_resource - Get the current resource of the vma
478 * @vma: The vma to get the current resource from.
479 *
480 * It's illegal to call this function if the vma is not bound.
481 *
482 * Return: A refcounted pointer to the current vma resource
483 * of the vma, assuming the vma is bound.
484 */
485 static inline struct i915_vma_resource *
i915_vma_get_current_resource(struct i915_vma * vma)486 i915_vma_get_current_resource(struct i915_vma *vma)
487 {
488 return i915_vma_resource_get(vma->resource);
489 }
490
491 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
492 void i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,
493 struct i915_vma *vma);
494 #endif
495
496 void i915_vma_module_exit(void);
497 int i915_vma_module_init(void);
498
499 I915_SELFTEST_DECLARE(int i915_vma_get_pages(struct i915_vma *vma));
500 I915_SELFTEST_DECLARE(void i915_vma_put_pages(struct i915_vma *vma));
501
502 #endif
503