summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/i915/gem/i915_gem_mman.c
blob: 65fc6ff5f59dabd26976f1e159dbb9ff53ce7005 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
/*
 * SPDX-License-Identifier: MIT
 *
 * Copyright © 2014-2016 Intel Corporation
 */

#include <linux/anon_inodes.h>
#include <linux/mman.h>
#include <linux/pfn_t.h>
#include <linux/sizes.h>

#include "gt/intel_gt.h"
#include "gt/intel_gt_requests.h"

#include "i915_drv.h"
#include "i915_gem_gtt.h"
#include "i915_gem_ioctls.h"
#include "i915_gem_object.h"
#include "i915_gem_mman.h"
#include "i915_trace.h"
#include "i915_user_extensions.h"
#include "i915_gem_ttm.h"
#include "i915_vma.h"

static inline bool
__vma_matches(struct vm_area_struct *vma, struct file *filp,
	      unsigned long addr, unsigned long size)
{
	if (vma->vm_file != filp)
		return false;

	return vma->vm_start == addr &&
	       (vma->vm_end - vma->vm_start) == PAGE_ALIGN(size);
}

/**
 * i915_gem_mmap_ioctl - Maps the contents of an object, returning the address
 *			 it is mapped to.
 * @dev: drm device
 * @data: ioctl data blob
 * @file: drm file
 *
 * While the mapping holds a reference on the contents of the object, it doesn't
 * imply a ref on the object itself.
 *
 * IMPORTANT:
 *
 * DRM driver writers who look a this function as an example for how to do GEM
 * mmap support, please don't implement mmap support like here. The modern way
 * to implement DRM mmap support is with an mmap offset ioctl (like
 * i915_gem_mmap_gtt) and then using the mmap syscall on the DRM fd directly.
 * That way debug tooling like valgrind will understand what's going on, hiding
 * the mmap call in a driver private ioctl will break that. The i915 driver only
 * does cpu mmaps this way because we didn't know better.
 */
int
i915_gem_mmap_ioctl(struct drm_device *dev, void *data,
		    struct drm_file *file)
{
	struct drm_i915_private *i915 = to_i915(dev);
	struct drm_i915_gem_mmap *args = data;
	struct drm_i915_gem_object *obj;
	unsigned long addr;

	/*
	 * mmap ioctl is disallowed for all discrete platforms,
	 * and for all platforms with GRAPHICS_VER > 12.
	 */
	if (IS_DGFX(i915) || GRAPHICS_VER(i915) > 12)
		return -EOPNOTSUPP;

	if (args->flags & ~(I915_MMAP_WC))
		return -EINVAL;

	if (args->flags & I915_MMAP_WC && !boot_cpu_has(X86_FEATURE_PAT))
		return -ENODEV;

	obj = i915_gem_object_lookup(file, args->handle);
	if (!obj)
		return -ENOENT;

	/* prime objects have no backing filp to GEM mmap
	 * pages from.
	 */
	if (!obj->base.filp) {
		addr = -ENXIO;
		goto err;
	}

	if (range_overflows(args->offset, args->size, (u64)obj->base.size)) {
		addr = -EINVAL;
		goto err;
	}

	addr = vm_mmap(obj->base.filp, 0, args->size,
		       PROT_READ | PROT_WRITE, MAP_SHARED,
		       args->offset);
	if (IS_ERR_VALUE(addr))
		goto err;

	if (args->flags & I915_MMAP_WC) {
		struct mm_struct *mm = current->mm;
		struct vm_area_struct *vma;

		if (mmap_write_lock_killable(mm)) {
			addr = -EINTR;
			goto err;
		}
		vma = find_vma(mm, addr);
		if (vma && __vma_matches(vma, obj->base.filp, addr, args->size))
			vma->vm_page_prot =
				pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
		else
			addr = -ENOMEM;
		mmap_write_unlock(mm);
		if (IS_ERR_VALUE(addr))
			goto err;
	}
	i915_gem_object_put(obj);

	args->addr_ptr = (u64)addr;
	return 0;

err:
	i915_gem_object_put(obj);
	return addr;
}

static unsigned int tile_row_pages(const struct drm_i915_gem_object *obj)
{
	return i915_gem_object_get_tile_row_size(obj) >> PAGE_SHIFT;
}

/**
 * i915_gem_mmap_gtt_version - report the current feature set for GTT mmaps
 *
 * A history of the GTT mmap interface:
 *
 * 0 - Everything had to fit into the GTT. Both parties of a memcpy had to
 *     aligned and suitable for fencing, and still fit into the available
 *     mappable space left by the pinned display objects. A classic problem
 *     we called the page-fault-of-doom where we would ping-pong between
 *     two objects that could not fit inside the GTT and so the memcpy
 *     would page one object in at the expense of the other between every
 *     single byte.
 *
 * 1 - Objects can be any size, and have any compatible fencing (X Y, or none
 *     as set via i915_gem_set_tiling() [DRM_I915_GEM_SET_TILING]). If the
 *     object is too large for the available space (or simply too large
 *     for the mappable aperture!), a view is created instead and faulted
 *     into userspace. (This view is aligned and sized appropriately for
 *     fenced access.)
 *
 * 2 - Recognise WC as a separate cache domain so that we can flush the
 *     delayed writes via GTT before performing direct access via WC.
 *
 * 3 - Remove implicit set-domain(GTT) and synchronisation on initial
 *     pagefault; swapin remains transparent.
 *
 * 4 - Support multiple fault handlers per object depending on object's
 *     backing storage (a.k.a. MMAP_OFFSET).
 *
 * Restrictions:
 *
 *  * snoopable objects cannot be accessed via the GTT. It can cause machine
 *    hangs on some architectures, corruption on others. An attempt to service
 *    a GTT page fault from a snoopable object will generate a SIGBUS.
 *
 *  * the object must be able to fit into RAM (physical memory, though no
 *    limited to the mappable aperture).
 *
 *
 * Caveats:
 *
 *  * a new GTT page fault will synchronize rendering from the GPU and flush
 *    all data to system memory. Subsequent access will not be synchronized.
 *
 *  * all mappings are revoked on runtime device suspend.
 *
 *  * there are only 8, 16 or 32 fence registers to share between all users
 *    (older machines require fence register for display and blitter access
 *    as well). Contention of the fence registers will cause the previous users
 *    to be unmapped and any new access will generate new page faults.
 *
 *  * running out of memory while servicing a fault may generate a SIGBUS,
 *    rather than the expected SIGSEGV.
 */
int i915_gem_mmap_gtt_version(void)
{
	return 4;
}

static inline struct i915_ggtt_view
compute_partial_view(const struct drm_i915_gem_object *obj,
		     pgoff_t page_offset,
		     unsigned int chunk)
{
	struct i915_ggtt_view view;

	if (i915_gem_object_is_tiled(obj))
		chunk = roundup(chunk, tile_row_pages(obj) ?: 1);

	view.type = I915_GGTT_VIEW_PARTIAL;
	view.partial.offset = rounddown(page_offset, chunk);
	view.partial.size =
		min_t(unsigned int, chunk,
		      (obj->base.size >> PAGE_SHIFT) - view.partial.offset);

	/* If the partial covers the entire object, just create a normal VMA. */
	if (chunk >= obj->base.size >> PAGE_SHIFT)
		view.type = I915_GGTT_VIEW_NORMAL;

	return view;
}

static vm_fault_t i915_error_to_vmf_fault(int err)
{
	switch (err) {
	default:
		WARN_ONCE(err, "unhandled error in %s: %i\n", __func__, err);
		fallthrough;
	case -EIO: /* shmemfs failure from swap device */
	case -EFAULT: /* purged object */
	case -ENODEV: /* bad object, how did you get here! */
	case -ENXIO: /* unable to access backing store (on device) */
		return VM_FAULT_SIGBUS;

	case -ENOMEM: /* our allocation failure */
		return VM_FAULT_OOM;

	case 0:
	case -EAGAIN:
	case -ENOSPC: /* transient failure to evict? */
	case -ERESTARTSYS:
	case -EINTR:
	case -EBUSY:
		/*
		 * EBUSY is ok: this just means that another thread
		 * already did the job.
		 */
		return VM_FAULT_NOPAGE;
	}
}

static vm_fault_t vm_fault_cpu(struct vm_fault *vmf)
{
	struct vm_area_struct *area = vmf->vma;
	struct i915_mmap_offset *mmo = area->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;
	resource_size_t iomap;
	int err;

	/* Sanity check that we allow writing into this object */
	if (unlikely(i915_gem_object_is_readonly(obj) &&
		     area->vm_flags & VM_WRITE))
		return VM_FAULT_SIGBUS;

	if (i915_gem_object_lock_interruptible(obj, NULL))
		return VM_FAULT_NOPAGE;

	err = i915_gem_object_pin_pages(obj);
	if (err)
		goto out;

	iomap = -1;
	if (!i915_gem_object_has_struct_page(obj)) {
		iomap = obj->mm.region->iomap.base;
		iomap -= obj->mm.region->region.start;
	}

	/* PTEs are revoked in obj->ops->put_pages() */
	err = remap_io_sg(area,
			  area->vm_start, area->vm_end - area->vm_start,
			  obj->mm.pages->sgl, iomap);

	if (area->vm_flags & VM_WRITE) {
		GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
		obj->mm.dirty = true;
	}

	i915_gem_object_unpin_pages(obj);

out:
	i915_gem_object_unlock(obj);
	return i915_error_to_vmf_fault(err);
}

static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
{
#define MIN_CHUNK_PAGES (SZ_1M >> PAGE_SHIFT)
	struct vm_area_struct *area = vmf->vma;
	struct i915_mmap_offset *mmo = area->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;
	struct drm_device *dev = obj->base.dev;
	struct drm_i915_private *i915 = to_i915(dev);
	struct intel_runtime_pm *rpm = &i915->runtime_pm;
	struct i915_ggtt *ggtt = &i915->ggtt;
	bool write = area->vm_flags & VM_WRITE;
	struct i915_gem_ww_ctx ww;
	intel_wakeref_t wakeref;
	struct i915_vma *vma;
	pgoff_t page_offset;
	int srcu;
	int ret;

	/* We don't use vmf->pgoff since that has the fake offset */
	page_offset = (vmf->address - area->vm_start) >> PAGE_SHIFT;

	trace_i915_gem_object_fault(obj, page_offset, true, write);

	wakeref = intel_runtime_pm_get(rpm);

	i915_gem_ww_ctx_init(&ww, true);
retry:
	ret = i915_gem_object_lock(obj, &ww);
	if (ret)
		goto err_rpm;

	/* Sanity check that we allow writing into this object */
	if (i915_gem_object_is_readonly(obj) && write) {
		ret = -EFAULT;
		goto err_rpm;
	}

	ret = i915_gem_object_pin_pages(obj);
	if (ret)
		goto err_rpm;

	ret = intel_gt_reset_trylock(ggtt->vm.gt, &srcu);
	if (ret)
		goto err_pages;

	/* Now pin it into the GTT as needed */
	vma = i915_gem_object_ggtt_pin_ww(obj, &ww, NULL, 0, 0,
					  PIN_MAPPABLE |
					  PIN_NONBLOCK /* NOWARN */ |
					  PIN_NOEVICT);
	if (IS_ERR(vma) && vma != ERR_PTR(-EDEADLK)) {
		/* Use a partial view if it is bigger than available space */
		struct i915_ggtt_view view =
			compute_partial_view(obj, page_offset, MIN_CHUNK_PAGES);
		unsigned int flags;

		flags = PIN_MAPPABLE | PIN_NOSEARCH;
		if (view.type == I915_GGTT_VIEW_NORMAL)
			flags |= PIN_NONBLOCK; /* avoid warnings for pinned */

		/*
		 * Userspace is now writing through an untracked VMA, abandon
		 * all hope that the hardware is able to track future writes.
		 */

		vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
		if (IS_ERR(vma) && vma != ERR_PTR(-EDEADLK)) {
			flags = PIN_MAPPABLE;
			view.type = I915_GGTT_VIEW_PARTIAL;
			vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
		}

		/* The entire mappable GGTT is pinned? Unexpected! */
		GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
	}
	if (IS_ERR(vma)) {
		ret = PTR_ERR(vma);
		goto err_reset;
	}

	/* Access to snoopable pages through the GTT is incoherent. */
	if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(i915)) {
		ret = -EFAULT;
		goto err_unpin;
	}

	ret = i915_vma_pin_fence(vma);
	if (ret)
		goto err_unpin;

	/* Finally, remap it using the new GTT offset */
	ret = remap_io_mapping(area,
			       area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT),
			       (ggtt->gmadr.start + vma->node.start) >> PAGE_SHIFT,
			       min_t(u64, vma->size, area->vm_end - area->vm_start),
			       &ggtt->iomap);
	if (ret)
		goto err_fence;

	assert_rpm_wakelock_held(rpm);

	/* Mark as being mmapped into userspace for later revocation */
	mutex_lock(&i915->ggtt.vm.mutex);
	if (!i915_vma_set_userfault(vma) && !obj->userfault_count++)
		list_add(&obj->userfault_link, &i915->ggtt.userfault_list);
	mutex_unlock(&i915->ggtt.vm.mutex);

	/* Track the mmo associated with the fenced vma */
	vma->mmo = mmo;

	if (CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND)
		intel_wakeref_auto(&i915->ggtt.userfault_wakeref,
				   msecs_to_jiffies_timeout(CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND));

	if (write) {
		GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
		i915_vma_set_ggtt_write(vma);
		obj->mm.dirty = true;
	}

err_fence:
	i915_vma_unpin_fence(vma);
err_unpin:
	__i915_vma_unpin(vma);
err_reset:
	intel_gt_reset_unlock(ggtt->vm.gt, srcu);
err_pages:
	i915_gem_object_unpin_pages(obj);
err_rpm:
	if (ret == -EDEADLK) {
		ret = i915_gem_ww_ctx_backoff(&ww);
		if (!ret)
			goto retry;
	}
	i915_gem_ww_ctx_fini(&ww);
	intel_runtime_pm_put(rpm, wakeref);
	return i915_error_to_vmf_fault(ret);
}

static int
vm_access(struct vm_area_struct *area, unsigned long addr,
	  void *buf, int len, int write)
{
	struct i915_mmap_offset *mmo = area->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;
	struct i915_gem_ww_ctx ww;
	void *vaddr;
	int err = 0;

	if (i915_gem_object_is_readonly(obj) && write)
		return -EACCES;

	addr -= area->vm_start;
	if (addr >= obj->base.size)
		return -EINVAL;

	i915_gem_ww_ctx_init(&ww, true);
retry:
	err = i915_gem_object_lock(obj, &ww);
	if (err)
		goto out;

	/* As this is primarily for debugging, let's focus on simplicity */
	vaddr = i915_gem_object_pin_map(obj, I915_MAP_FORCE_WC);
	if (IS_ERR(vaddr)) {
		err = PTR_ERR(vaddr);
		goto out;
	}

	if (write) {
		memcpy(vaddr + addr, buf, len);
		__i915_gem_object_flush_map(obj, addr, len);
	} else {
		memcpy(buf, vaddr + addr, len);
	}

	i915_gem_object_unpin_map(obj);
out:
	if (err == -EDEADLK) {
		err = i915_gem_ww_ctx_backoff(&ww);
		if (!err)
			goto retry;
	}
	i915_gem_ww_ctx_fini(&ww);

	if (err)
		return err;

	return len;
}

void __i915_gem_object_release_mmap_gtt(struct drm_i915_gem_object *obj)
{
	struct i915_vma *vma;

	GEM_BUG_ON(!obj->userfault_count);

	for_each_ggtt_vma(vma, obj)
		i915_vma_revoke_mmap(vma);

	GEM_BUG_ON(obj->userfault_count);
}

/*
 * It is vital that we remove the page mapping if we have mapped a tiled
 * object through the GTT and then lose the fence register due to
 * resource pressure. Similarly if the object has been moved out of the
 * aperture, than pages mapped into userspace must be revoked. Removing the
 * mapping will then trigger a page fault on the next user access, allowing
 * fixup by vm_fault_gtt().
 */
void i915_gem_object_release_mmap_gtt(struct drm_i915_gem_object *obj)
{
	struct drm_i915_private *i915 = to_i915(obj->base.dev);
	intel_wakeref_t wakeref;

	/*
	 * Serialisation between user GTT access and our code depends upon
	 * revoking the CPU's PTE whilst the mutex is held. The next user
	 * pagefault then has to wait until we release the mutex.
	 *
	 * Note that RPM complicates somewhat by adding an additional
	 * requirement that operations to the GGTT be made holding the RPM
	 * wakeref.
	 */
	wakeref = intel_runtime_pm_get(&i915->runtime_pm);
	mutex_lock(&i915->ggtt.vm.mutex);

	if (!obj->userfault_count)
		goto out;

	__i915_gem_object_release_mmap_gtt(obj);

	/*
	 * Ensure that the CPU's PTE are revoked and there are not outstanding
	 * memory transactions from userspace before we return. The TLB
	 * flushing implied above by changing the PTE above *should* be
	 * sufficient, an extra barrier here just provides us with a bit
	 * of paranoid documentation about our requirement to serialise
	 * memory writes before touching registers / GSM.
	 */
	wmb();

out:
	mutex_unlock(&i915->ggtt.vm.mutex);
	intel_runtime_pm_put(&i915->runtime_pm, wakeref);
}

void i915_gem_object_release_mmap_offset(struct drm_i915_gem_object *obj)
{
	struct i915_mmap_offset *mmo, *mn;

	spin_lock(&obj->mmo.lock);
	rbtree_postorder_for_each_entry_safe(mmo, mn,
					     &obj->mmo.offsets, offset) {
		/*
		 * vma_node_unmap for GTT mmaps handled already in
		 * __i915_gem_object_release_mmap_gtt
		 */
		if (mmo->mmap_type == I915_MMAP_TYPE_GTT)
			continue;

		spin_unlock(&obj->mmo.lock);
		drm_vma_node_unmap(&mmo->vma_node,
				   obj->base.dev->anon_inode->i_mapping);
		spin_lock(&obj->mmo.lock);
	}
	spin_unlock(&obj->mmo.lock);
}

static struct i915_mmap_offset *
lookup_mmo(struct drm_i915_gem_object *obj,
	   enum i915_mmap_type mmap_type)
{
	struct rb_node *rb;

	spin_lock(&obj->mmo.lock);
	rb = obj->mmo.offsets.rb_node;
	while (rb) {
		struct i915_mmap_offset *mmo =
			rb_entry(rb, typeof(*mmo), offset);

		if (mmo->mmap_type == mmap_type) {
			spin_unlock(&obj->mmo.lock);
			return mmo;
		}

		if (mmo->mmap_type < mmap_type)
			rb = rb->rb_right;
		else
			rb = rb->rb_left;
	}
	spin_unlock(&obj->mmo.lock);

	return NULL;
}

static struct i915_mmap_offset *
insert_mmo(struct drm_i915_gem_object *obj, struct i915_mmap_offset *mmo)
{
	struct rb_node *rb, **p;

	spin_lock(&obj->mmo.lock);
	rb = NULL;
	p = &obj->mmo.offsets.rb_node;
	while (*p) {
		struct i915_mmap_offset *pos;

		rb = *p;
		pos = rb_entry(rb, typeof(*pos), offset);

		if (pos->mmap_type == mmo->mmap_type) {
			spin_unlock(&obj->mmo.lock);
			drm_vma_offset_remove(obj->base.dev->vma_offset_manager,
					      &mmo->vma_node);
			kfree(mmo);
			return pos;
		}

		if (pos->mmap_type < mmo->mmap_type)
			p = &rb->rb_right;
		else
			p = &rb->rb_left;
	}
	rb_link_node(&mmo->offset, rb, p);
	rb_insert_color(&mmo->offset, &obj->mmo.offsets);
	spin_unlock(&obj->mmo.lock);

	return mmo;
}

static struct i915_mmap_offset *
mmap_offset_attach(struct drm_i915_gem_object *obj,
		   enum i915_mmap_type mmap_type,
		   struct drm_file *file)
{
	struct drm_i915_private *i915 = to_i915(obj->base.dev);
	struct i915_mmap_offset *mmo;
	int err;

	GEM_BUG_ON(obj->ops->mmap_offset || obj->ops->mmap_ops);

	mmo = lookup_mmo(obj, mmap_type);
	if (mmo)
		goto out;

	mmo = kmalloc(sizeof(*mmo), GFP_KERNEL);
	if (!mmo)
		return ERR_PTR(-ENOMEM);

	mmo->obj = obj;
	mmo->mmap_type = mmap_type;
	drm_vma_node_reset(&mmo->vma_node);

	err = drm_vma_offset_add(obj->base.dev->vma_offset_manager,
				 &mmo->vma_node, obj->base.size / PAGE_SIZE);
	if (likely(!err))
		goto insert;

	/* Attempt to reap some mmap space from dead objects */
	err = intel_gt_retire_requests_timeout(&i915->gt, MAX_SCHEDULE_TIMEOUT,
					       NULL);
	if (err)
		goto err;

	i915_gem_drain_freed_objects(i915);
	err = drm_vma_offset_add(obj->base.dev->vma_offset_manager,
				 &mmo->vma_node, obj->base.size / PAGE_SIZE);
	if (err)
		goto err;

insert:
	mmo = insert_mmo(obj, mmo);
	GEM_BUG_ON(lookup_mmo(obj, mmap_type) != mmo);
out:
	if (file)
		drm_vma_node_allow(&mmo->vma_node, file);
	return mmo;

err:
	kfree(mmo);
	return ERR_PTR(err);
}

static int
__assign_mmap_offset(struct drm_i915_gem_object *obj,
		     enum i915_mmap_type mmap_type,
		     u64 *offset, struct drm_file *file)
{
	struct i915_mmap_offset *mmo;

	if (i915_gem_object_never_mmap(obj))
		return -ENODEV;

	if (obj->ops->mmap_offset)  {
		if (mmap_type != I915_MMAP_TYPE_FIXED)
			return -ENODEV;

		*offset = obj->ops->mmap_offset(obj);
		return 0;
	}

	if (mmap_type == I915_MMAP_TYPE_FIXED)
		return -ENODEV;

	if (mmap_type != I915_MMAP_TYPE_GTT &&
	    !i915_gem_object_has_struct_page(obj) &&
	    !i915_gem_object_has_iomem(obj))
		return -ENODEV;

	mmo = mmap_offset_attach(obj, mmap_type, file);
	if (IS_ERR(mmo))
		return PTR_ERR(mmo);

	*offset = drm_vma_node_offset_addr(&mmo->vma_node);
	return 0;
}

static int
__assign_mmap_offset_handle(struct drm_file *file,
			    u32 handle,
			    enum i915_mmap_type mmap_type,
			    u64 *offset)
{
	struct drm_i915_gem_object *obj;
	int err;

	obj = i915_gem_object_lookup(file, handle);
	if (!obj)
		return -ENOENT;

	err = i915_gem_object_lock_interruptible(obj, NULL);
	if (err)
		goto out_put;
	err = __assign_mmap_offset(obj, mmap_type, offset, file);
	i915_gem_object_unlock(obj);
out_put:
	i915_gem_object_put(obj);
	return err;
}

int
i915_gem_dumb_mmap_offset(struct drm_file *file,
			  struct drm_device *dev,
			  u32 handle,
			  u64 *offset)
{
	enum i915_mmap_type mmap_type;

	if (HAS_LMEM(to_i915(dev)))
		mmap_type = I915_MMAP_TYPE_FIXED;
	else if (boot_cpu_has(X86_FEATURE_PAT))
		mmap_type = I915_MMAP_TYPE_WC;
	else if (!i915_ggtt_has_aperture(&to_i915(dev)->ggtt))
		return -ENODEV;
	else
		mmap_type = I915_MMAP_TYPE_GTT;

	return __assign_mmap_offset_handle(file, handle, mmap_type, offset);
}

/**
 * i915_gem_mmap_offset_ioctl - prepare an object for GTT mmap'ing
 * @dev: DRM device
 * @data: GTT mapping ioctl data
 * @file: GEM object info
 *
 * Simply returns the fake offset to userspace so it can mmap it.
 * The mmap call will end up in drm_gem_mmap(), which will set things
 * up so we can get faults in the handler above.
 *
 * The fault handler will take care of binding the object into the GTT
 * (since it may have been evicted to make room for something), allocating
 * a fence register, and mapping the appropriate aperture address into
 * userspace.
 */
int
i915_gem_mmap_offset_ioctl(struct drm_device *dev, void *data,
			   struct drm_file *file)
{
	struct drm_i915_private *i915 = to_i915(dev);
	struct drm_i915_gem_mmap_offset *args = data;
	enum i915_mmap_type type;
	int err;

	/*
	 * Historically we failed to check args.pad and args.offset
	 * and so we cannot use those fields for user input and we cannot
	 * add -EINVAL for them as the ABI is fixed, i.e. old userspace
	 * may be feeding in garbage in those fields.
	 *
	 * if (args->pad) return -EINVAL; is verbotten!
	 */

	err = i915_user_extensions(u64_to_user_ptr(args->extensions),
				   NULL, 0, NULL);
	if (err)
		return err;

	switch (args->flags) {
	case I915_MMAP_OFFSET_GTT:
		if (!i915_ggtt_has_aperture(&i915->ggtt))
			return -ENODEV;
		type = I915_MMAP_TYPE_GTT;
		break;

	case I915_MMAP_OFFSET_WC:
		if (!boot_cpu_has(X86_FEATURE_PAT))
			return -ENODEV;
		type = I915_MMAP_TYPE_WC;
		break;

	case I915_MMAP_OFFSET_WB:
		type = I915_MMAP_TYPE_WB;
		break;

	case I915_MMAP_OFFSET_UC:
		if (!boot_cpu_has(X86_FEATURE_PAT))
			return -ENODEV;
		type = I915_MMAP_TYPE_UC;
		break;

	case I915_MMAP_OFFSET_FIXED:
		type = I915_MMAP_TYPE_FIXED;
		break;

	default:
		return -EINVAL;
	}

	return __assign_mmap_offset_handle(file, args->handle, type, &args->offset);
}

static void vm_open(struct vm_area_struct *vma)
{
	struct i915_mmap_offset *mmo = vma->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;

	GEM_BUG_ON(!obj);
	i915_gem_object_get(obj);
}

static void vm_close(struct vm_area_struct *vma)
{
	struct i915_mmap_offset *mmo = vma->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;

	GEM_BUG_ON(!obj);
	i915_gem_object_put(obj);
}

static const struct vm_operations_struct vm_ops_gtt = {
	.fault = vm_fault_gtt,
	.access = vm_access,
	.open = vm_open,
	.close = vm_close,
};

static const struct vm_operations_struct vm_ops_cpu = {
	.fault = vm_fault_cpu,
	.access = vm_access,
	.open = vm_open,
	.close = vm_close,
};

static int singleton_release(struct inode *inode, struct file *file)
{
	struct drm_i915_private *i915 = file->private_data;

	cmpxchg(&i915->gem.mmap_singleton, file, NULL);
	drm_dev_put(&i915->drm);

	return 0;
}

static const struct file_operations singleton_fops = {
	.owner = THIS_MODULE,
	.release = singleton_release,
};

static struct file *mmap_singleton(struct drm_i915_private *i915)
{
	struct file *file;

	rcu_read_lock();
	file = READ_ONCE(i915->gem.mmap_singleton);
	if (file && !get_file_rcu(file))
		file = NULL;
	rcu_read_unlock();
	if (file)
		return file;

	file = anon_inode_getfile("i915.gem", &singleton_fops, i915, O_RDWR);
	if (IS_ERR(file))
		return file;

	/* Everyone shares a single global address space */
	file->f_mapping = i915->drm.anon_inode->i_mapping;

	smp_store_mb(i915->gem.mmap_singleton, file);
	drm_dev_get(&i915->drm);

	return file;
}

/*
 * This overcomes the limitation in drm_gem_mmap's assignment of a
 * drm_gem_object as the vma->vm_private_data. Since we need to
 * be able to resolve multiple mmap offsets which could be tied
 * to a single gem object.
 */
int i915_gem_mmap(struct file *filp, struct vm_area_struct *vma)
{
	struct drm_vma_offset_node *node;
	struct drm_file *priv = filp->private_data;
	struct drm_device *dev = priv->minor->dev;
	struct drm_i915_gem_object *obj = NULL;
	struct i915_mmap_offset *mmo = NULL;
	struct file *anon;

	if (drm_dev_is_unplugged(dev))
		return -ENODEV;

	rcu_read_lock();
	drm_vma_offset_lock_lookup(dev->vma_offset_manager);
	node = drm_vma_offset_exact_lookup_locked(dev->vma_offset_manager,
						  vma->vm_pgoff,
						  vma_pages(vma));
	if (node && drm_vma_node_is_allowed(node, priv)) {
		/*
		 * Skip 0-refcnted objects as it is in the process of being
		 * destroyed and will be invalid when the vma manager lock
		 * is released.
		 */
		if (!node->driver_private) {
			mmo = container_of(node, struct i915_mmap_offset, vma_node);
			obj = i915_gem_object_get_rcu(mmo->obj);

			GEM_BUG_ON(obj && obj->ops->mmap_ops);
		} else {
			obj = i915_gem_object_get_rcu
				(container_of(node, struct drm_i915_gem_object,
					      base.vma_node));

			GEM_BUG_ON(obj && !obj->ops->mmap_ops);
		}
	}
	drm_vma_offset_unlock_lookup(dev->vma_offset_manager);
	rcu_read_unlock();
	if (!obj)
		return node ? -EACCES : -EINVAL;

	if (i915_gem_object_is_readonly(obj)) {
		if (vma->vm_flags & VM_WRITE) {
			i915_gem_object_put(obj);
			return -EINVAL;
		}
		vma->vm_flags &= ~VM_MAYWRITE;
	}

	anon = mmap_singleton(to_i915(dev));
	if (IS_ERR(anon)) {
		i915_gem_object_put(obj);
		return PTR_ERR(anon);
	}

	vma->vm_flags |= VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP | VM_IO;

	/*
	 * We keep the ref on mmo->obj, not vm_file, but we require
	 * vma->vm_file->f_mapping, see vma_link(), for later revocation.
	 * Our userspace is accustomed to having per-file resource cleanup
	 * (i.e. contexts, objects and requests) on their close(fd), which
	 * requires avoiding extraneous references to their filp, hence why
	 * we prefer to use an anonymous file for their mmaps.
	 */
	vma_set_file(vma, anon);
	/* Drop the initial creation reference, the vma is now holding one. */
	fput(anon);

	if (obj->ops->mmap_ops) {
		vma->vm_page_prot = pgprot_decrypted(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = obj->ops->mmap_ops;
		vma->vm_private_data = node->driver_private;
		return 0;
	}

	vma->vm_private_data = mmo;

	switch (mmo->mmap_type) {
	case I915_MMAP_TYPE_WC:
		vma->vm_page_prot =
			pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = &vm_ops_cpu;
		break;

	case I915_MMAP_TYPE_FIXED:
		GEM_WARN_ON(1);
		fallthrough;
	case I915_MMAP_TYPE_WB:
		vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
		vma->vm_ops = &vm_ops_cpu;
		break;

	case I915_MMAP_TYPE_UC:
		vma->vm_page_prot =
			pgprot_noncached(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = &vm_ops_cpu;
		break;

	case I915_MMAP_TYPE_GTT:
		vma->vm_page_prot =
			pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = &vm_ops_gtt;
		break;
	}
	vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot);

	return 0;
}

#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftests/i915_gem_mman.c"
#endif