summaryrefslogtreecommitdiff
path: root/linux-core/i915_gem.c
blob: 9d701ba14d1e1fae5b191773b643ba5ad6a88d12 (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
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
/*
 * Copyright © 2008 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * Authors:
 *    Eric Anholt <eric@anholt.net>
 *
 */

#include "drmP.h"
#include "drm.h"
#include "i915_drm.h"
#include "i915_drv.h"

#define WATCH_BUF	0
#define WATCH_EXEC	0
#define WATCH_LRU	0
#define WATCH_RELOC	0

int
i915_gem_init_ioctl(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_init *args = data;

	if (args->gtt_start >= args->gtt_end ||
	    (args->gtt_start & (PAGE_SIZE - 1)) != 0 ||
	    (args->gtt_end & (PAGE_SIZE - 1)) != 0)
		return -EINVAL;

	drm_memrange_init(&dev_priv->mm.gtt_space, args->gtt_start,
	    args->gtt_end - args->gtt_start);

	return 0;
}

static void
i915_gem_object_free_page_list(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	int page_count = obj->size / PAGE_SIZE;
	int i;

	if (obj_priv->page_list == NULL)
		return;


	for (i = 0; i < page_count; i++)
		if (obj_priv->page_list[i] != NULL)
			page_cache_release(obj_priv->page_list[i]);

	drm_free(obj_priv->page_list,
		 page_count * sizeof(struct page *),
		 DRM_MEM_DRIVER);
	obj_priv->page_list = NULL;
}

static void
i915_gem_flush(struct drm_device *dev,
	       uint32_t invalidate_domains,
	       uint32_t flush_domains)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	uint32_t cmd;
	RING_LOCALS;

#if WATCH_EXEC
	DRM_INFO("%s: invalidate %08x flush %08x\n", __func__,
		  invalidate_domains, flush_domains);
#endif

	if (flush_domains & DRM_GEM_DOMAIN_CPU)
		drm_agp_chipset_flush(dev);

	if ((invalidate_domains|flush_domains) & ~DRM_GEM_DOMAIN_CPU) {
		/*
		 * read/write caches:
		 *
		 * DRM_GEM_DOMAIN_I915_RENDER is always invalidated, but is
		 * only flushed if MI_NO_WRITE_FLUSH is unset.  On 965, it is
		 * also flushed at 2d versus 3d pipeline switches.
		 *
		 * read-only caches:
		 *
		 * DRM_GEM_DOMAIN_I915_SAMPLER is flushed on pre-965 if
		 * MI_READ_FLUSH is set, and is always flushed on 965.
		 *
		 * DRM_GEM_DOMAIN_I915_COMMAND may not exist?
		 *
		 * DRM_GEM_DOMAIN_I915_INSTRUCTION, which exists on 965, is
		 * invalidated when MI_EXE_FLUSH is set.
		 *
		 * DRM_GEM_DOMAIN_I915_VERTEX, which exists on 965, is
		 * invalidated with every MI_FLUSH.
		 *
		 * TLBs:
		 *
		 * On 965, TLBs associated with DRM_GEM_DOMAIN_I915_COMMAND
		 * and DRM_GEM_DOMAIN_CPU in are invalidated at PTE write and
		 * DRM_GEM_DOMAIN_I915_RENDER and DRM_GEM_DOMAIN_I915_SAMPLER
		 * are flushed at any MI_FLUSH.
		 */

		cmd = CMD_MI_FLUSH | MI_NO_WRITE_FLUSH;
		if ((invalidate_domains|flush_domains) &
		    DRM_GEM_DOMAIN_I915_RENDER)
			cmd &= ~MI_NO_WRITE_FLUSH;
		if (!IS_I965G(dev)) {
			/*
			 * On the 965, the sampler cache always gets flushed
			 * and this bit is reserved.
			 */
			if (invalidate_domains & DRM_GEM_DOMAIN_I915_SAMPLER)
				cmd |= MI_READ_FLUSH;
		}
		if (invalidate_domains & DRM_GEM_DOMAIN_I915_INSTRUCTION)
			cmd |= MI_EXE_FLUSH;

		BEGIN_LP_RING(2);
		OUT_RING(cmd);
		OUT_RING(0); /* noop */
		ADVANCE_LP_RING();
	}
}

/**
 * Ensures that all rendering to the object has completed and the object is
 * safe to unbind from the GTT or access from the CPU.
 */
static int
i915_gem_object_wait_rendering(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	int ret;

	/* If there are writes queued to the buffer, flush and
	 * create a new cookie to wait for.
	 */
	if (obj->write_domain & ~(DRM_GEM_DOMAIN_CPU)) {
#if WATCH_BUF
		DRM_INFO("%s: flushing object %p from write domain %08x\n",
			  __func__, obj, obj->write_domain);
#endif
		i915_gem_flush(dev, 0, obj->write_domain);
		obj->write_domain = 0;

		/* Add a reference since we're gaining a cookie. */
		if (obj_priv->last_rendering_cookie == 0)
			drm_gem_object_reference(obj);
		/* Move from whatever list we were on to the tail of execution.
		 */
		list_move_tail(&obj_priv->gtt_lru_entry,
			       &dev_priv->mm.execution_list);
		obj_priv->last_rendering_cookie = i915_emit_irq(dev);
		BUG_ON(obj_priv->last_rendering_cookie == 0);
#if WATCH_LRU
		DRM_INFO("%s: flush moves to exec list %p\n", __func__, obj);
#endif
	}
	/* If there is rendering queued on the buffer being evicted, wait for
	 * it.
	 */
	if (obj_priv->last_rendering_cookie != 0) {
#if WATCH_BUF
		DRM_INFO("%s: object %p wait for cookie %08x\n",
			  __func__, obj, obj_priv->last_rendering_cookie);
#endif
		ret = i915_wait_irq(dev, obj_priv->last_rendering_cookie);
		if (ret != 0)
			return ret;

		/* Clear it now that we know it's passed. */
		obj_priv->last_rendering_cookie = 0;

		/* We were on the execution list since we had a cookie.
		 * Move to the tail of the LRU list now since we're done.
		 */
		if (obj_priv->pin_count == 0)
			list_move_tail(&obj_priv->gtt_lru_entry,
				       &dev_priv->mm.gtt_lru);

#if WATCH_LRU
		DRM_INFO("%s: wait moves to lru list %p\n", __func__, obj);
#endif
		/* The cookie held a reference to the object, release that
		 * now
		 */
		drm_gem_object_unreference(obj);
	}

	return 0;
}

/**
 * Unbinds an object from the GTT aperture.
 */
static void
i915_gem_object_unbind(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv = obj->driver_private;

#if WATCH_BUF
	DRM_INFO("%s:%d %p\n", __func__, __LINE__, obj);
	DRM_INFO("gtt_space %p\n", obj_priv->gtt_space);
#endif
	if (obj_priv->gtt_space == NULL)
		return;

	/* Ignore the return value of wait_rendering.  If we're here but
	 * a wait_rendering hasn't completed, we're in the freeing process,
	 * and we want the buffer to go away even if the command queue is hung.
	 */
	(void)i915_gem_object_wait_rendering(obj);

	if (obj_priv->agp_mem != NULL) {
		drm_unbind_agp(obj_priv->agp_mem);
		drm_free_agp(obj_priv->agp_mem, obj->size / PAGE_SIZE);
		obj_priv->agp_mem = NULL;
	}

	i915_gem_object_free_page_list(obj);

	drm_memrange_put_block(obj_priv->gtt_space);
	obj_priv->gtt_space = NULL;

	/* Remove ourselves from the LRU list if present. */
	if (!list_empty(&obj_priv->gtt_lru_entry)) {
		list_del_init(&obj_priv->gtt_lru_entry);
		if (obj_priv->last_rendering_cookie) {
			DRM_ERROR("Failed to wait on buffer when unbinding, "
				  "continued anyway.\n");
			obj_priv->last_rendering_cookie = 0;
			drm_gem_object_unreference(obj);
		}
	}
}

#if WATCH_BUF | WATCH_EXEC
static void
i915_gem_dump_page(struct page *page, uint32_t start, uint32_t end,
		   uint32_t bias, uint32_t mark)
{
	uint32_t *mem = kmap_atomic(page, KM_USER0);
	int i;
	for (i = start; i < end; i += 4)
		DRM_INFO("%08x: %08x%s\n",
			  (int) (bias + i), mem[i / 4],
			  (bias + i == mark) ? " ********" : "");
	kunmap_atomic(mem, KM_USER0);
	/* give syslog time to catch up */
	msleep(1);
}

static void
i915_gem_dump_object(struct drm_gem_object *obj, int len,
		     const char *where, uint32_t mark)
{
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	int page;

	DRM_INFO("%s: object at offset %08x\n", where, obj_priv->gtt_offset);
	for (page = 0; page < (len + PAGE_SIZE-1) / PAGE_SIZE; page++) {
		int page_len, chunk, chunk_len;

		page_len = len - page * PAGE_SIZE;
		if (page_len > PAGE_SIZE)
			page_len = PAGE_SIZE;

		for (chunk = 0; chunk < page_len; chunk += 128) {
			chunk_len = page_len - chunk;
			if (chunk_len > 128)
				chunk_len = 128;
			i915_gem_dump_page(obj_priv->page_list[page],
					   chunk, chunk + chunk_len,
					   obj_priv->gtt_offset +
					   page * PAGE_SIZE,
					   mark);
		}
	}
}
#endif

#if WATCH_LRU
static void
i915_dump_lru(struct drm_device *dev, const char *where)
{
	drm_i915_private_t		*dev_priv = dev->dev_private;
	struct drm_i915_gem_object	*obj_priv;

	DRM_INFO("GTT execution list %s {\n", where);
	list_for_each_entry(obj_priv, &dev_priv->mm.execution_list,
			    gtt_lru_entry)
	{
		DRM_INFO("    %p: %08x\n", obj_priv,
			 obj_priv->last_rendering_cookie);
	}
	DRM_INFO("GTT LRU %s {\n", where);
	list_for_each_entry(obj_priv, &dev_priv->mm.gtt_lru, gtt_lru_entry) {
		DRM_INFO("    %p: %08x\n", obj_priv,
			 obj_priv->last_rendering_cookie);
	}
	DRM_INFO("}\n");
}
#endif

static int
i915_gem_evict_something(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret;

	/* Find the LRU buffer. */
	if (!list_empty(&dev_priv->mm.gtt_lru)) {
		obj_priv = list_first_entry(&dev_priv->mm.gtt_lru,
					    struct drm_i915_gem_object,
					    gtt_lru_entry);
	} else if (!list_empty(&dev_priv->mm.execution_list)) {
		/* If there's nothing unused and ready, grab the first
		 * unpinned object from the currently executing list.
		 */
		list_for_each_entry(obj_priv, &dev_priv->mm.execution_list,
				    gtt_lru_entry)
			if (obj_priv->pin_count == 0)
				break;
		if (!obj_priv)
			return -ENOMEM;
	} else {
		return -ENOMEM;
	}
	obj = obj_priv->obj;
	drm_gem_object_reference(obj);
#if WATCH_LRU
	DRM_INFO("%s: evicting %p\n", __func__, obj);
#endif

	/* Only unpinned buffers should be on this list. */
	BUG_ON(obj_priv->pin_count != 0);

	/* Do this separately from the wait_rendering in
	 * i915_gem_object_unbind() because we want to catch interrupts and
	 * return.
	 */
	ret = i915_gem_object_wait_rendering(obj);
	if (ret != 0)
		return ret;

	/* Wait on the rendering and unbind the buffer. */
	i915_gem_object_unbind(obj);
#if WATCH_LRU
	DRM_INFO("%s: evicted %p\n", __func__, obj);
#endif
	drm_gem_object_unreference(obj);

	return 0;
}

static int
i915_gem_object_get_page_list(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	int page_count, i;
	if (obj_priv->page_list)
		return 0;

	/* Get the list of pages out of our struct file.  They'll be pinned
	 * at this point until we release them.
	 */
	page_count = obj->size / PAGE_SIZE;
	BUG_ON(obj_priv->page_list != NULL);
	obj_priv->page_list = drm_calloc(page_count, sizeof(struct page *),
					 DRM_MEM_DRIVER);
	if (obj_priv->page_list == NULL)
		return -ENOMEM;

	for (i = 0; i < page_count; i++) {
		obj_priv->page_list[i] =
		    find_or_create_page(obj->filp->f_mapping, i, GFP_HIGHUSER);

		if (obj_priv->page_list[i] == NULL) {
			i915_gem_object_free_page_list(obj);
			return -ENOMEM;
		}
		unlock_page(obj_priv->page_list[i]);
	}
	return 0;
}

/**
 * Finds free space in the GTT aperture and binds the object there.
 */
static int
i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	struct drm_memrange_node *free_space;
	int page_count, ret;

	if (alignment == 0)
		alignment = PAGE_SIZE;
	if (alignment & (PAGE_SIZE - 1)) {
		DRM_ERROR("Invalid object alignment requested %u\n", alignment);
		return -EINVAL;
	}

 search_free:
	free_space = drm_memrange_search_free(&dev_priv->mm.gtt_space,
					      obj->size,
					      alignment, 0);
	if (free_space != NULL) {
		obj_priv->gtt_space =
			drm_memrange_get_block(free_space, obj->size,
					       alignment);
		if (obj_priv->gtt_space != NULL) {
			obj_priv->gtt_space->private = obj;
			obj_priv->gtt_offset = obj_priv->gtt_space->start;
		}
	}
	if (obj_priv->gtt_space == NULL) {
		/* If the gtt is empty and we're still having trouble
		 * fitting our object in, we're out of memory.
		 */
#if WATCH_LRU
		DRM_INFO("%s: GTT full, evicting something\n", __func__);
#endif
		if (list_empty(&dev_priv->mm.gtt_lru) &&
		    list_empty(&dev_priv->mm.execution_list)) {
			DRM_ERROR("GTT full, but LRU list empty\n");
			return -ENOMEM;
		}

		ret = i915_gem_evict_something(dev);
		if (ret != 0)
			return ret;
		goto search_free;
	}

#if WATCH_BUF
	DRM_INFO("Binding object of size %d at 0x%08x\n",
		 obj->size, obj_priv->gtt_offset);
#endif
	ret = i915_gem_object_get_page_list (obj);
	if (ret) {
		drm_memrange_put_block (obj_priv->gtt_space);
		obj_priv->gtt_space = NULL;
		return ret;
	}

	page_count = obj->size / PAGE_SIZE;
	/* Create an AGP memory structure pointing at our pages, and bind it
	 * into the GTT.
	 */
	obj_priv->agp_mem = drm_agp_bind_pages(dev,
					       obj_priv->page_list,
					       page_count,
					       obj_priv->gtt_offset);
	if (obj_priv->agp_mem == NULL) {
		i915_gem_object_free_page_list(obj);
		drm_memrange_put_block(obj_priv->gtt_space);
		obj_priv->gtt_space = NULL;
		return -ENOMEM;
	}

	/* When we have just bound an object, we have no valid read
	 * caches on it, regardless of where it was before.  We also need
	 * an MI_FLUSH to occur so that the render and sampler TLBs
	 * get flushed and pick up our binding change above.
	 */
	obj->read_domains = 0;

	return 0;
}

static void
i915_gem_clflush_object(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object	*obj_priv = obj->driver_private;

	/* If we don't have a page list set up, then we're not pinned
	 * to GPU, and we can ignore the cache flush because it'll happen
	 * again at bind time.
	 */
	if (obj_priv->page_list == NULL)
		return;

	drm_ttm_cache_flush(obj_priv->page_list, obj->size / PAGE_SIZE);
}

/*
 * Set the next domain for the specified object. This
 * may not actually perform the necessary flushing/invaliding though,
 * as that may want to be batched with other set_domain operations
 */
static void
i915_gem_object_set_domain(struct drm_gem_object *obj,
			    uint32_t read_domains,
			    uint32_t write_domain)
{
	struct drm_device		*dev = obj->dev;
	uint32_t			invalidate_domains = 0;
	uint32_t			flush_domains = 0;

#if WATCH_BUF
	DRM_INFO("%s: object %p read %08x write %08x\n",
		 __func__, obj, read_domains, write_domain);
#endif
	/*
	 * Flush the current write domain if
	 * the new read domains don't match. Invalidate
	 * any read domains which differ from the old
	 * write domain
	 */
	if (obj->write_domain && obj->write_domain != read_domains) {
		flush_domains |= obj->write_domain;
		invalidate_domains |= read_domains & ~obj->write_domain;
	}
	/*
	 * Invalidate any read caches which may have
	 * stale data. That is, any new read domains.
	 */
	invalidate_domains |= read_domains & ~obj->read_domains;
	if ((flush_domains | invalidate_domains) & DRM_GEM_DOMAIN_CPU) {
#if WATCH_BUF
		DRM_INFO("%s: CPU domain flush %08x invalidate %08x\n",
			 __func__, flush_domains, invalidate_domains);
#endif
		/*
		 * If we're invaliding the CPU cache and flushing a GPU cache,
		 * then pause for rendering so that the GPU caches will be
		 * flushed before the cpu cache is invalidated
		 */
		if ((invalidate_domains & DRM_GEM_DOMAIN_CPU) &&
		    (flush_domains & ~DRM_GEM_DOMAIN_CPU))
			i915_gem_object_wait_rendering(obj);
		i915_gem_clflush_object(obj);
	}

	obj->write_domain = write_domain;
	obj->read_domains = read_domains;
	dev->invalidate_domains |= invalidate_domains;
	dev->flush_domains |= flush_domains;
}

/**
 * Once all of the objects have been set in the proper domain,
 * perform the necessary flush and invalidate operations
 */

static void
i915_gem_dev_set_domain(struct drm_device *dev)
{
	/*
	 * Now that all the buffers are synced to the proper domains,
	 * flush and invalidate the collected domains
	 */
	if (dev->invalidate_domains | dev->flush_domains) {
#if WATCH_EXEC
		DRM_INFO("%s: invalidate_domains %08x flush_domains %08x\n",
			  __func__,
			 dev->invalidate_domains,
			 dev->flush_domains);
#endif
		i915_gem_flush(dev,
			       dev->invalidate_domains,
			       dev->flush_domains);
		dev->invalidate_domains = 0;
		dev->flush_domains = 0;
	}
}

static int
i915_gem_reloc_and_validate_object(struct drm_gem_object *obj,
				   struct drm_file *file_priv,
				   struct drm_i915_gem_exec_object *entry)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_relocation_entry reloc;
	struct drm_i915_gem_relocation_entry __user *relocs;
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	int i;
	uint32_t last_reloc_offset = -1;
	void *reloc_page = NULL;

	/* Choose the GTT offset for our buffer and put it there. */
	if (obj_priv->gtt_space == NULL) {
		i915_gem_object_bind_to_gtt(obj, (unsigned) entry->alignment);
		if (obj_priv->gtt_space == NULL)
			return -ENOMEM;
	}

	entry->offset = obj_priv->gtt_offset;

	relocs = (struct drm_i915_gem_relocation_entry __user *)
		 (uintptr_t) entry->relocs_ptr;
	/* Apply the relocations, using the GTT aperture to avoid cache
	 * flushing requirements.
	 */
	for (i = 0; i < entry->relocation_count; i++) {
		struct drm_gem_object *target_obj;
		struct drm_i915_gem_object *target_obj_priv;
		uint32_t reloc_val, reloc_offset, *reloc_entry;
		int ret;

		ret = copy_from_user(&reloc, relocs + i, sizeof(reloc));
		if (ret != 0)
			return ret;

		target_obj = drm_gem_object_lookup(obj->dev, file_priv,
						   reloc.target_handle);
		if (target_obj == NULL)
			return -EINVAL;
		target_obj_priv = target_obj->driver_private;

		/* The target buffer should have appeared before us in the
		 * validate list, so it should have a GTT space bound by now.
		 */
		if (target_obj_priv->gtt_space == NULL) {
			DRM_ERROR("No GTT space found for object %d\n",
				  reloc.target_handle);
			drm_gem_object_unreference(target_obj);
			return -EINVAL;
		}

		if (reloc.offset > obj->size - 4) {
			DRM_ERROR("Relocation beyond object bounds: "
				  "obj %p target %d offset %d size %d.\n",
				  obj, reloc.target_handle,
				  (int) reloc.offset, (int) obj->size);
			drm_gem_object_unreference(target_obj);
			return -EINVAL;
		}
		if (reloc.offset & 3) {
			DRM_ERROR("Relocation not 4-byte aligned: "
				  "obj %p target %d offset %d.\n",
				  obj, reloc.target_handle,
				  (int) reloc.offset);
			drm_gem_object_unreference(target_obj);
			return -EINVAL;
		}

		if (reloc.write_domain && target_obj->pending_write_domain &&
		    reloc.write_domain != target_obj->pending_write_domain) {
			DRM_ERROR("Write domain conflict: "
				  "obj %p target %d offset %d "
				  "new %08x old %08x\n",
				  obj, reloc.target_handle,
				  (int) reloc.offset,
				  reloc.write_domain,
				  target_obj->pending_write_domain);
			drm_gem_object_unreference(target_obj);
			return -EINVAL;
		}

#if WATCH_RELOC
		DRM_INFO("%s: obj %p offset %08x target %d "
			 "read %08x write %08x gtt %08x "
			 "presumed %08x delta %08x\n",
			 __func__,
			 obj,
			 (int) reloc.offset,
			 (int) reloc.target_handle,
			 (int) reloc.read_domains,
			 (int) reloc.write_domain,
			 (int) target_obj_priv->gtt_offset,
			 (int) reloc.presumed_offset,
			 reloc.delta);
#endif

		target_obj->pending_read_domains |= reloc.read_domains;
		target_obj->pending_write_domain |= reloc.write_domain;

		/* If the relocation already has the right value in it, no
		 * more work needs to be done.
		 */
		if (target_obj_priv->gtt_offset == reloc.presumed_offset) {
			drm_gem_object_unreference(target_obj);
			continue;
		}

		/* Now that we're going to actually write some data in,
		 * make sure that any rendering using this buffer's contents
		 * is completed.
		 */
		i915_gem_object_wait_rendering(obj);

		/* As we're writing through the gtt, flush
		 * any CPU writes before we write the relocations
		 */
		if (obj->write_domain & DRM_GEM_DOMAIN_CPU) {
			i915_gem_clflush_object(obj);
			drm_agp_chipset_flush(dev);
			obj->write_domain = 0;
		}

		/* Map the page containing the relocation we're going to
		 * perform.
		 */
		reloc_offset = obj_priv->gtt_offset + reloc.offset;
		if (reloc_page == NULL ||
		    (last_reloc_offset & ~(PAGE_SIZE - 1)) !=
		    (reloc_offset & ~(PAGE_SIZE - 1))) {
			if (reloc_page != NULL)
				iounmap(reloc_page);

			reloc_page = ioremap(dev->agp->base +
					     (reloc_offset & ~(PAGE_SIZE - 1)),
					     PAGE_SIZE);
			last_reloc_offset = reloc_offset;
			if (reloc_page == NULL) {
				drm_gem_object_unreference(target_obj);
				return -ENOMEM;
			}
		}

		reloc_entry = (uint32_t *)((char *)reloc_page +
					   (reloc_offset & (PAGE_SIZE - 1)));
		reloc_val = target_obj_priv->gtt_offset + reloc.delta;

#if WATCH_BUF
		DRM_INFO("Applied relocation: %p@0x%08x %08x -> %08x\n",
			  obj, (unsigned int) reloc.offset,
			  readl(reloc_entry), reloc_val);
#endif
		writel(reloc_val, reloc_entry);

		drm_gem_object_unreference(target_obj);
	}

	if (reloc_page != NULL)
		iounmap(reloc_page);

#if WATCH_BUF
	i915_gem_dump_object(obj, 128, __func__, ~0);
#endif
	return 0;
}

static int
i915_dispatch_gem_execbuffer(struct drm_device *dev,
			      struct drm_i915_gem_execbuffer *exec,
			      uint64_t exec_offset)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_clip_rect __user *boxes = (struct drm_clip_rect __user *)
					     (uintptr_t) exec->cliprects_ptr;
	int nbox = exec->num_cliprects;
	int i = 0, count;
	uint32_t	exec_start, exec_len;
	RING_LOCALS;

	exec_start = (uint32_t) exec_offset + exec->batch_start_offset;
	exec_len = (uint32_t) exec->batch_len;

	if ((exec_start | exec_len) & 0x7) {
		DRM_ERROR("alignment\n");
		return -EINVAL;
	}

	if (!exec_start)
		return -EINVAL;

	count = nbox ? nbox : 1;

	for (i = 0; i < count; i++) {
		if (i < nbox) {
			int ret = i915_emit_box(dev, boxes, i,
						exec->DR1, exec->DR4);
			if (ret)
				return ret;
		}

		if (dev_priv->use_mi_batchbuffer_start) {
			BEGIN_LP_RING(2);
			if (IS_I965G(dev)) {
				OUT_RING(MI_BATCH_BUFFER_START |
					 (2 << 6) |
					 MI_BATCH_NON_SECURE_I965);
				OUT_RING(exec_start);
			} else {
				OUT_RING(MI_BATCH_BUFFER_START |
					 (2 << 6));
				OUT_RING(exec_start | MI_BATCH_NON_SECURE);
			}
			ADVANCE_LP_RING();

		} else {
			BEGIN_LP_RING(4);
			OUT_RING(MI_BATCH_BUFFER);
			OUT_RING(exec_start | MI_BATCH_NON_SECURE);
			OUT_RING(exec_start + exec_len - 4);
			OUT_RING(0);
			ADVANCE_LP_RING();
		}
	}

	/* XXX breadcrumb */
	return 0;
}

int
i915_gem_execbuffer(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_execbuffer *args = data;
	struct drm_i915_gem_exec_object *validate_list;
	struct drm_gem_object **object_list;
	struct drm_gem_object *batch_obj;
	int ret, i;
	uint64_t exec_offset;
	uint32_t cookie;

	LOCK_TEST_WITH_RETURN(dev, file_priv);

#if WATCH_EXEC
	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
#endif
	i915_kernel_lost_context(dev);

	/* Copy in the validate list from userland */
	validate_list = drm_calloc(sizeof(*validate_list), args->buffer_count,
				   DRM_MEM_DRIVER);
	object_list = drm_calloc(sizeof(*object_list), args->buffer_count,
				 DRM_MEM_DRIVER);
	if (validate_list == NULL || object_list == NULL) {
		DRM_ERROR("Failed to allocate validate or object list "
			  "for %d buffers\n",
			  args->buffer_count);
		ret = -ENOMEM;
		goto err;
	}
	ret = copy_from_user(validate_list,
			     (struct drm_i915_relocation_entry __user *)
			     (uintptr_t) args->buffers_ptr,
			     sizeof(*validate_list) * args->buffer_count);
	if (ret != 0) {
		DRM_ERROR("copy %d validate entries failed %d\n",
			  args->buffer_count, ret);
		goto err;
	}

	/* Look up object handles and perform the relocations */
	for (i = 0; i < args->buffer_count; i++) {
		object_list[i] = drm_gem_object_lookup(dev, file_priv,
						       validate_list[i].handle);
		if (object_list[i] == NULL) {
			DRM_ERROR("Invalid object handle %d at index %d\n",
				   validate_list[i].handle, i);
			ret = -EINVAL;
			goto err;
		}

		ret = i915_gem_reloc_and_validate_object(object_list[i],
							 file_priv,
							 &validate_list[i]);
		if (ret) {
			DRM_ERROR("reloc and validate failed %d\n", ret);
			goto err;
		}
	}

	/* Set the pending read domains for the batch buffer to COMMAND */
	batch_obj = object_list[args->buffer_count-1];
	batch_obj->pending_read_domains = DRM_GEM_DOMAIN_I915_COMMAND;
	batch_obj->pending_write_domain = 0;

	for (i = 0; i < args->buffer_count; i++) {
		struct drm_gem_object *obj = object_list[i];
		struct drm_i915_gem_object *obj_priv = obj->driver_private;

		if (obj_priv->gtt_space == NULL) {
			/* We evicted the buffer in the process of validating
			 * our set of buffers in.  We could try to recover by
			 * kicking them everything out and trying again from
			 * the start.
			 */
			ret = -ENOMEM;
			goto err;
		}

		/* make sure all previous memory operations have passed */
		i915_gem_object_set_domain(obj,
					    obj->pending_read_domains,
					    obj->pending_write_domain);
		obj->pending_read_domains = 0;
		obj->pending_write_domain = 0;
	}

	/* Flush/invalidate caches and chipset buffer */
	i915_gem_dev_set_domain(dev);

	exec_offset = validate_list[args->buffer_count - 1].offset;

#if WATCH_EXEC
	i915_gem_dump_object(object_list[args->buffer_count - 1],
			      args->batch_len,
			      __func__,
			      ~0);
#endif

	/* Exec the batchbuffer */
	ret = i915_dispatch_gem_execbuffer(dev, args, exec_offset);
	if (ret) {
		DRM_ERROR("dispatch failed %d\n", ret);
		goto err;
	}

	/*
	 * Get a cookie representing the execution of the current buffer,
	 * which we can wait on.  We would like to mitigate these interrupts,
	 * likely by only creating cookies occasionally (so that we have
	 * *some* interrupts representing completion of buffers that we can
	 * wait on when trying to clear up gtt space).
	 */
	cookie = i915_emit_irq(dev);
	for (i = 0; i < args->buffer_count; i++) {
		struct drm_gem_object *obj = object_list[i];
		struct drm_i915_gem_object *obj_priv = obj->driver_private;

		/*
		 * Have the cookie hold a reference to this object
		 * which is freed when the object is waited for
		 */
		if (obj_priv->last_rendering_cookie == 0)
			drm_gem_object_reference(obj);
		obj_priv->last_rendering_cookie = cookie;
		BUG_ON(obj_priv->last_rendering_cookie == 0);
		/* Move our buffer to the tail of the execution list. */
		list_move_tail(&obj_priv->gtt_lru_entry,
			       &dev_priv->mm.execution_list);
#if WATCH_LRU
		DRM_INFO("%s: move to exec list %p\n", __func__, obj);
#endif
	}
#if WATCH_LRU && 0
	i915_dump_lru(dev, __func__);
#endif

	/* Copy the new buffer offsets back to the user's validate list. */
	ret = copy_to_user((struct drm_i915_relocation_entry __user *)
			   (uintptr_t) args->buffers_ptr,
			   validate_list,
			   sizeof(*validate_list) * args->buffer_count);
	if (ret)
		DRM_ERROR("failed to copy %d validate entries "
			  "back to user (%d)\n",
			   args->buffer_count, ret);
err:
	if (object_list != NULL) {
		for (i = 0; i < args->buffer_count; i++)
			drm_gem_object_unreference(object_list[i]);
	}

	drm_free(object_list, sizeof(*object_list) * args->buffer_count,
		 DRM_MEM_DRIVER);
	drm_free(validate_list, sizeof(*validate_list) * args->buffer_count,
		 DRM_MEM_DRIVER);

	return ret;
}

int
i915_gem_pin_ioctl(struct drm_device *dev, void *data,
		   struct drm_file *file_priv)
{
	struct drm_i915_gem_pin *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		DRM_ERROR("Bad handle in i915_gem_pin_ioctl(): %d\n",
			  args->handle);
		return -EINVAL;
	}

	obj_priv = obj->driver_private;
	if (obj_priv->gtt_space == NULL) {
		ret = i915_gem_object_bind_to_gtt(obj,
						  (unsigned) args->alignment);
		if (ret != 0) {
			DRM_ERROR("Failure to bind in "
				  "i915_gem_pin_ioctl(): %d\n",
				  ret);
			drm_gem_object_unreference(obj);
			return ret;
		}
	}

	obj_priv->pin_count++;
	args->offset = obj_priv->gtt_offset;
	drm_gem_object_unreference(obj);

	return 0;
}

int
i915_gem_unpin_ioctl(struct drm_device *dev, void *data,
		     struct drm_file *file_priv)
{
	struct drm_i915_gem_pin *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		DRM_ERROR("Bad handle in i915_gem_unpin_ioctl(): %d\n",
			  args->handle);
		return -EINVAL;
	}

	obj_priv = obj->driver_private;
	obj_priv->pin_count--;
	drm_gem_object_unreference(obj);
	return 0;
}

int i915_gem_init_object(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv;

	obj_priv = drm_calloc(1, sizeof(*obj_priv), DRM_MEM_DRIVER);
	if (obj_priv == NULL)
		return -ENOMEM;

	obj->driver_private = obj_priv;
	obj_priv->obj = obj;
	INIT_LIST_HEAD(&obj_priv->gtt_lru_entry);
	return 0;
}

void i915_gem_free_object(struct drm_gem_object *obj)
{
	i915_gem_object_unbind(obj);

	drm_free(obj->driver_private, 1, DRM_MEM_DRIVER);
}

int
i915_gem_set_domain_ioctl(struct drm_gem_object *obj,
			  uint32_t read_domains,
			  uint32_t write_domain)
{
	i915_gem_object_set_domain(obj, read_domains, write_domain);
	i915_gem_dev_set_domain(obj->dev);
	return 0;
}

int
i915_gem_flush_pwrite(struct drm_gem_object *obj,
		      uint64_t offset, uint64_t size)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = obj->driver_private;
	/*
	 * As far as I can tell, writes of > 64 bytes will use non-temporal
	 * stores which should obviate the need for this clflush.
	 * It doesn't work for me though...
	 */
/*	if (size <= 64)  */{
		if (obj_priv->gtt_space == NULL) {
			int ret = i915_gem_object_get_page_list (obj);
			if (ret)
				 return ret;
		}
		i915_gem_clflush_object(obj);
		if (obj_priv->gtt_space == NULL)
			i915_gem_object_free_page_list (obj);
	}
	drm_agp_chipset_flush(dev);
	return 0;
}

void
i915_gem_lastclose(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

	/* Assume that the chip has been idled at this point. Just pull them
	 * off the execution list and unref them.  Since this is the last
	 * close, this is also the last ref and they'll go away.
	 */

	while (!list_empty(&dev_priv->mm.execution_list)) {
		struct drm_i915_gem_object *obj_priv;

		obj_priv = list_first_entry(&dev_priv->mm.execution_list,
					    struct drm_i915_gem_object,
					    gtt_lru_entry);

		list_del_init(&obj_priv->gtt_lru_entry);
		obj_priv->last_rendering_cookie = 0;
		obj_priv->obj->write_domain = 0;
		drm_gem_object_unreference(obj_priv->obj);
	}
}