summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/omapdrm/omap_gem.c
blob: ceba5459ceb75b33bded69b6b56ffb9c5919a03c (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
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
/*
 * drivers/gpu/drm/omapdrm/omap_gem.c
 *
 * Copyright (C) 2011 Texas Instruments
 * Author: Rob Clark <rob.clark@linaro.org>
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published by
 * the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program.  If not, see <http://www.gnu.org/licenses/>.
 */

#include <linux/shmem_fs.h>
#include <linux/spinlock.h>
#include <linux/pfn_t.h>

#include <drm/drm_vma_manager.h>

#include "omap_drv.h"
#include "omap_dmm_tiler.h"

/* remove these once drm core helpers are merged */
struct page **_drm_gem_get_pages(struct drm_gem_object *obj, gfp_t gfpmask);
void _drm_gem_put_pages(struct drm_gem_object *obj, struct page **pages,
		bool dirty, bool accessed);
int _drm_gem_create_mmap_offset_size(struct drm_gem_object *obj, size_t size);

/*
 * GEM buffer object implementation.
 */

#define to_omap_bo(x) container_of(x, struct omap_gem_object, base)

/* note: we use upper 8 bits of flags for driver-internal flags: */
#define OMAP_BO_DMA			0x01000000	/* actually is physically contiguous */
#define OMAP_BO_EXT_SYNC	0x02000000	/* externally allocated sync object */
#define OMAP_BO_EXT_MEM		0x04000000	/* externally allocated memory */


struct omap_gem_object {
	struct drm_gem_object base;

	struct list_head mm_list;

	uint32_t flags;

	/** width/height for tiled formats (rounded up to slot boundaries) */
	uint16_t width, height;

	/** roll applied when mapping to DMM */
	uint32_t roll;

	/**
	 * If buffer is allocated physically contiguous, the OMAP_BO_DMA flag
	 * is set and the paddr is valid.  Also if the buffer is remapped in
	 * TILER and paddr_cnt > 0, then paddr is valid.  But if you are using
	 * the physical address and OMAP_BO_DMA is not set, then you should
	 * be going thru omap_gem_{get,put}_paddr() to ensure the mapping is
	 * not removed from under your feet.
	 *
	 * Note that OMAP_BO_SCANOUT is a hint from userspace that DMA capable
	 * buffer is requested, but doesn't mean that it is.  Use the
	 * OMAP_BO_DMA flag to determine if the buffer has a DMA capable
	 * physical address.
	 */
	dma_addr_t paddr;

	/**
	 * # of users of paddr
	 */
	uint32_t paddr_cnt;

	/**
	 * tiler block used when buffer is remapped in DMM/TILER.
	 */
	struct tiler_block *block;

	/**
	 * Array of backing pages, if allocated.  Note that pages are never
	 * allocated for buffers originally allocated from contiguous memory
	 */
	struct page **pages;

	/** addresses corresponding to pages in above array */
	dma_addr_t *addrs;

	/**
	 * Virtual address, if mapped.
	 */
	void *vaddr;

	/**
	 * sync-object allocated on demand (if needed)
	 *
	 * Per-buffer sync-object for tracking pending and completed hw/dma
	 * read and write operations.  The layout in memory is dictated by
	 * the SGX firmware, which uses this information to stall the command
	 * stream if a surface is not ready yet.
	 *
	 * Note that when buffer is used by SGX, the sync-object needs to be
	 * allocated from a special heap of sync-objects.  This way many sync
	 * objects can be packed in a page, and not waste GPU virtual address
	 * space.  Because of this we have to have a omap_gem_set_sync_object()
	 * API to allow replacement of the syncobj after it has (potentially)
	 * already been allocated.  A bit ugly but I haven't thought of a
	 * better alternative.
	 */
	struct {
		uint32_t write_pending;
		uint32_t write_complete;
		uint32_t read_pending;
		uint32_t read_complete;
	} *sync;
};

static int get_pages(struct drm_gem_object *obj, struct page ***pages);
static uint64_t mmap_offset(struct drm_gem_object *obj);

/* To deal with userspace mmap'ings of 2d tiled buffers, which (a) are
 * not necessarily pinned in TILER all the time, and (b) when they are
 * they are not necessarily page aligned, we reserve one or more small
 * regions in each of the 2d containers to use as a user-GART where we
 * can create a second page-aligned mapping of parts of the buffer
 * being accessed from userspace.
 *
 * Note that we could optimize slightly when we know that multiple
 * tiler containers are backed by the same PAT.. but I'll leave that
 * for later..
 */
#define NUM_USERGART_ENTRIES 2
struct usergart_entry {
	struct tiler_block *block;	/* the reserved tiler block */
	dma_addr_t paddr;
	struct drm_gem_object *obj;	/* the current pinned obj */
	pgoff_t obj_pgoff;		/* page offset of obj currently
					   mapped in */
};
static struct {
	struct usergart_entry entry[NUM_USERGART_ENTRIES];
	int height;				/* height in rows */
	int height_shift;		/* ilog2(height in rows) */
	int slot_shift;			/* ilog2(width per slot) */
	int stride_pfn;			/* stride in pages */
	int last;				/* index of last used entry */
} *usergart;

static void evict_entry(struct drm_gem_object *obj,
		enum tiler_fmt fmt, struct usergart_entry *entry)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int n = usergart[fmt].height;
	size_t size = PAGE_SIZE * n;
	loff_t off = mmap_offset(obj) +
			(entry->obj_pgoff << PAGE_SHIFT);
	const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE);

	if (m > 1) {
		int i;
		/* if stride > than PAGE_SIZE then sparse mapping: */
		for (i = n; i > 0; i--) {
			unmap_mapping_range(obj->dev->anon_inode->i_mapping,
					    off, PAGE_SIZE, 1);
			off += PAGE_SIZE * m;
		}
	} else {
		unmap_mapping_range(obj->dev->anon_inode->i_mapping,
				    off, size, 1);
	}

	entry->obj = NULL;
}

/* Evict a buffer from usergart, if it is mapped there */
static void evict(struct drm_gem_object *obj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);

	if (omap_obj->flags & OMAP_BO_TILED) {
		enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
		int i;

		if (!usergart)
			return;

		for (i = 0; i < NUM_USERGART_ENTRIES; i++) {
			struct usergart_entry *entry = &usergart[fmt].entry[i];
			if (entry->obj == obj)
				evict_entry(obj, fmt, entry);
		}
	}
}

/* GEM objects can either be allocated from contiguous memory (in which
 * case obj->filp==NULL), or w/ shmem backing (obj->filp!=NULL).  But non
 * contiguous buffers can be remapped in TILER/DMM if they need to be
 * contiguous... but we don't do this all the time to reduce pressure
 * on TILER/DMM space when we know at allocation time that the buffer
 * will need to be scanned out.
 */
static inline bool is_shmem(struct drm_gem_object *obj)
{
	return obj->filp != NULL;
}

/**
 * shmem buffers that are mapped cached can simulate coherency via using
 * page faulting to keep track of dirty pages
 */
static inline bool is_cached_coherent(struct drm_gem_object *obj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	return is_shmem(obj) &&
		((omap_obj->flags & OMAP_BO_CACHE_MASK) == OMAP_BO_CACHED);
}

static DEFINE_SPINLOCK(sync_lock);

/** ensure backing pages are allocated */
static int omap_gem_attach_pages(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	struct page **pages;
	int npages = obj->size >> PAGE_SHIFT;
	int i, ret;
	dma_addr_t *addrs;

	WARN_ON(omap_obj->pages);

	pages = drm_gem_get_pages(obj);
	if (IS_ERR(pages)) {
		dev_err(obj->dev->dev, "could not get pages: %ld\n", PTR_ERR(pages));
		return PTR_ERR(pages);
	}

	/* for non-cached buffers, ensure the new pages are clean because
	 * DSS, GPU, etc. are not cache coherent:
	 */
	if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
		addrs = kmalloc(npages * sizeof(*addrs), GFP_KERNEL);
		if (!addrs) {
			ret = -ENOMEM;
			goto free_pages;
		}

		for (i = 0; i < npages; i++) {
			addrs[i] = dma_map_page(dev->dev, pages[i],
					0, PAGE_SIZE, DMA_BIDIRECTIONAL);
		}
	} else {
		addrs = kzalloc(npages * sizeof(*addrs), GFP_KERNEL);
		if (!addrs) {
			ret = -ENOMEM;
			goto free_pages;
		}
	}

	omap_obj->addrs = addrs;
	omap_obj->pages = pages;

	return 0;

free_pages:
	drm_gem_put_pages(obj, pages, true, false);

	return ret;
}

/** release backing pages */
static void omap_gem_detach_pages(struct drm_gem_object *obj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);

	/* for non-cached buffers, ensure the new pages are clean because
	 * DSS, GPU, etc. are not cache coherent:
	 */
	if (omap_obj->flags & (OMAP_BO_WC|OMAP_BO_UNCACHED)) {
		int i, npages = obj->size >> PAGE_SHIFT;
		for (i = 0; i < npages; i++) {
			dma_unmap_page(obj->dev->dev, omap_obj->addrs[i],
					PAGE_SIZE, DMA_BIDIRECTIONAL);
		}
	}

	kfree(omap_obj->addrs);
	omap_obj->addrs = NULL;

	drm_gem_put_pages(obj, omap_obj->pages, true, false);
	omap_obj->pages = NULL;
}

/* get buffer flags */
uint32_t omap_gem_flags(struct drm_gem_object *obj)
{
	return to_omap_bo(obj)->flags;
}

/** get mmap offset */
static uint64_t mmap_offset(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	int ret;
	size_t size;

	WARN_ON(!mutex_is_locked(&dev->struct_mutex));

	/* Make it mmapable */
	size = omap_gem_mmap_size(obj);
	ret = drm_gem_create_mmap_offset_size(obj, size);
	if (ret) {
		dev_err(dev->dev, "could not allocate mmap offset\n");
		return 0;
	}

	return drm_vma_node_offset_addr(&obj->vma_node);
}

uint64_t omap_gem_mmap_offset(struct drm_gem_object *obj)
{
	uint64_t offset;
	mutex_lock(&obj->dev->struct_mutex);
	offset = mmap_offset(obj);
	mutex_unlock(&obj->dev->struct_mutex);
	return offset;
}

/** get mmap size */
size_t omap_gem_mmap_size(struct drm_gem_object *obj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	size_t size = obj->size;

	if (omap_obj->flags & OMAP_BO_TILED) {
		/* for tiled buffers, the virtual size has stride rounded up
		 * to 4kb.. (to hide the fact that row n+1 might start 16kb or
		 * 32kb later!).  But we don't back the entire buffer with
		 * pages, only the valid picture part.. so need to adjust for
		 * this in the size used to mmap and generate mmap offset
		 */
		size = tiler_vsize(gem2fmt(omap_obj->flags),
				omap_obj->width, omap_obj->height);
	}

	return size;
}

/* get tiled size, returns -EINVAL if not tiled buffer */
int omap_gem_tiled_size(struct drm_gem_object *obj, uint16_t *w, uint16_t *h)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	if (omap_obj->flags & OMAP_BO_TILED) {
		*w = omap_obj->width;
		*h = omap_obj->height;
		return 0;
	}
	return -EINVAL;
}

/* Normal handling for the case of faulting in non-tiled buffers */
static int fault_1d(struct drm_gem_object *obj,
		struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	unsigned long pfn;
	pgoff_t pgoff;

	/* We don't use vmf->pgoff since that has the fake offset: */
	pgoff = ((unsigned long)vmf->virtual_address -
			vma->vm_start) >> PAGE_SHIFT;

	if (omap_obj->pages) {
		omap_gem_cpu_sync(obj, pgoff);
		pfn = page_to_pfn(omap_obj->pages[pgoff]);
	} else {
		BUG_ON(!(omap_obj->flags & OMAP_BO_DMA));
		pfn = (omap_obj->paddr >> PAGE_SHIFT) + pgoff;
	}

	VERB("Inserting %p pfn %lx, pa %lx", vmf->virtual_address,
			pfn, pfn << PAGE_SHIFT);

	return vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
			__pfn_to_pfn_t(pfn, PFN_DEV));
}

/* Special handling for the case of faulting in 2d tiled buffers */
static int fault_2d(struct drm_gem_object *obj,
		struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	struct usergart_entry *entry;
	enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
	struct page *pages[64];  /* XXX is this too much to have on stack? */
	unsigned long pfn;
	pgoff_t pgoff, base_pgoff;
	void __user *vaddr;
	int i, ret, slots;

	/*
	 * Note the height of the slot is also equal to the number of pages
	 * that need to be mapped in to fill 4kb wide CPU page.  If the slot
	 * height is 64, then 64 pages fill a 4kb wide by 64 row region.
	 */
	const int n = usergart[fmt].height;
	const int n_shift = usergart[fmt].height_shift;

	/*
	 * If buffer width in bytes > PAGE_SIZE then the virtual stride is
	 * rounded up to next multiple of PAGE_SIZE.. this need to be taken
	 * into account in some of the math, so figure out virtual stride
	 * in pages
	 */
	const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE);

	/* We don't use vmf->pgoff since that has the fake offset: */
	pgoff = ((unsigned long)vmf->virtual_address -
			vma->vm_start) >> PAGE_SHIFT;

	/*
	 * Actual address we start mapping at is rounded down to previous slot
	 * boundary in the y direction:
	 */
	base_pgoff = round_down(pgoff, m << n_shift);

	/* figure out buffer width in slots */
	slots = omap_obj->width >> usergart[fmt].slot_shift;

	vaddr = vmf->virtual_address - ((pgoff - base_pgoff) << PAGE_SHIFT);

	entry = &usergart[fmt].entry[usergart[fmt].last];

	/* evict previous buffer using this usergart entry, if any: */
	if (entry->obj)
		evict_entry(entry->obj, fmt, entry);

	entry->obj = obj;
	entry->obj_pgoff = base_pgoff;

	/* now convert base_pgoff to phys offset from virt offset: */
	base_pgoff = (base_pgoff >> n_shift) * slots;

	/* for wider-than 4k.. figure out which part of the slot-row we want: */
	if (m > 1) {
		int off = pgoff % m;
		entry->obj_pgoff += off;
		base_pgoff /= m;
		slots = min(slots - (off << n_shift), n);
		base_pgoff += off << n_shift;
		vaddr += off << PAGE_SHIFT;
	}

	/*
	 * Map in pages. Beyond the valid pixel part of the buffer, we set
	 * pages[i] to NULL to get a dummy page mapped in.. if someone
	 * reads/writes it they will get random/undefined content, but at
	 * least it won't be corrupting whatever other random page used to
	 * be mapped in, or other undefined behavior.
	 */
	memcpy(pages, &omap_obj->pages[base_pgoff],
			sizeof(struct page *) * slots);
	memset(pages + slots, 0,
			sizeof(struct page *) * (n - slots));

	ret = tiler_pin(entry->block, pages, ARRAY_SIZE(pages), 0, true);
	if (ret) {
		dev_err(obj->dev->dev, "failed to pin: %d\n", ret);
		return ret;
	}

	pfn = entry->paddr >> PAGE_SHIFT;

	VERB("Inserting %p pfn %lx, pa %lx", vmf->virtual_address,
			pfn, pfn << PAGE_SHIFT);

	for (i = n; i > 0; i--) {
		vm_insert_mixed(vma, (unsigned long)vaddr,
				__pfn_to_pfn_t(pfn, PFN_DEV));
		pfn += usergart[fmt].stride_pfn;
		vaddr += PAGE_SIZE * m;
	}

	/* simple round-robin: */
	usergart[fmt].last = (usergart[fmt].last + 1) % NUM_USERGART_ENTRIES;

	return 0;
}

/**
 * omap_gem_fault		-	pagefault handler for GEM objects
 * @vma: the VMA of the GEM object
 * @vmf: fault detail
 *
 * Invoked when a fault occurs on an mmap of a GEM managed area. GEM
 * does most of the work for us including the actual map/unmap calls
 * but we need to do the actual page work.
 *
 * The VMA was set up by GEM. In doing so it also ensured that the
 * vma->vm_private_data points to the GEM object that is backing this
 * mapping.
 */
int omap_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct drm_gem_object *obj = vma->vm_private_data;
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	struct drm_device *dev = obj->dev;
	struct page **pages;
	int ret;

	/* Make sure we don't parallel update on a fault, nor move or remove
	 * something from beneath our feet
	 */
	mutex_lock(&dev->struct_mutex);

	/* if a shmem backed object, make sure we have pages attached now */
	ret = get_pages(obj, &pages);
	if (ret)
		goto fail;

	/* where should we do corresponding put_pages().. we are mapping
	 * the original page, rather than thru a GART, so we can't rely
	 * on eviction to trigger this.  But munmap() or all mappings should
	 * probably trigger put_pages()?
	 */

	if (omap_obj->flags & OMAP_BO_TILED)
		ret = fault_2d(obj, vma, vmf);
	else
		ret = fault_1d(obj, vma, vmf);


fail:
	mutex_unlock(&dev->struct_mutex);
	switch (ret) {
	case 0:
	case -ERESTARTSYS:
	case -EINTR:
		return VM_FAULT_NOPAGE;
	case -ENOMEM:
		return VM_FAULT_OOM;
	default:
		return VM_FAULT_SIGBUS;
	}
}

/** We override mainly to fix up some of the vm mapping flags.. */
int omap_gem_mmap(struct file *filp, struct vm_area_struct *vma)
{
	int ret;

	ret = drm_gem_mmap(filp, vma);
	if (ret) {
		DBG("mmap failed: %d", ret);
		return ret;
	}

	return omap_gem_mmap_obj(vma->vm_private_data, vma);
}

int omap_gem_mmap_obj(struct drm_gem_object *obj,
		struct vm_area_struct *vma)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);

	vma->vm_flags &= ~VM_PFNMAP;
	vma->vm_flags |= VM_MIXEDMAP;

	if (omap_obj->flags & OMAP_BO_WC) {
		vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
	} else if (omap_obj->flags & OMAP_BO_UNCACHED) {
		vma->vm_page_prot = pgprot_noncached(vm_get_page_prot(vma->vm_flags));
	} else {
		/*
		 * We do have some private objects, at least for scanout buffers
		 * on hardware without DMM/TILER.  But these are allocated write-
		 * combine
		 */
		if (WARN_ON(!obj->filp))
			return -EINVAL;

		/*
		 * Shunt off cached objs to shmem file so they have their own
		 * address_space (so unmap_mapping_range does what we want,
		 * in particular in the case of mmap'd dmabufs)
		 */
		fput(vma->vm_file);
		vma->vm_pgoff = 0;
		vma->vm_file  = get_file(obj->filp);

		vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
	}

	return 0;
}


/**
 * omap_gem_dumb_create	-	create a dumb buffer
 * @drm_file: our client file
 * @dev: our device
 * @args: the requested arguments copied from userspace
 *
 * Allocate a buffer suitable for use for a frame buffer of the
 * form described by user space. Give userspace a handle by which
 * to reference it.
 */
int omap_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
		struct drm_mode_create_dumb *args)
{
	union omap_gem_size gsize;

	args->pitch = align_pitch(0, args->width, args->bpp);
	args->size = PAGE_ALIGN(args->pitch * args->height);

	gsize = (union omap_gem_size){
		.bytes = args->size,
	};

	return omap_gem_new_handle(dev, file, gsize,
			OMAP_BO_SCANOUT | OMAP_BO_WC, &args->handle);
}

/**
 * omap_gem_dumb_map	-	buffer mapping for dumb interface
 * @file: our drm client file
 * @dev: drm device
 * @handle: GEM handle to the object (from dumb_create)
 *
 * Do the necessary setup to allow the mapping of the frame buffer
 * into user memory. We don't have to do much here at the moment.
 */
int omap_gem_dumb_map_offset(struct drm_file *file, struct drm_device *dev,
		uint32_t handle, uint64_t *offset)
{
	struct drm_gem_object *obj;
	int ret = 0;

	/* GEM does all our handle to object mapping */
	obj = drm_gem_object_lookup(dev, file, handle);
	if (obj == NULL) {
		ret = -ENOENT;
		goto fail;
	}

	*offset = omap_gem_mmap_offset(obj);

	drm_gem_object_unreference_unlocked(obj);

fail:
	return ret;
}

/* Set scrolling position.  This allows us to implement fast scrolling
 * for console.
 *
 * Call only from non-atomic contexts.
 */
int omap_gem_roll(struct drm_gem_object *obj, uint32_t roll)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	uint32_t npages = obj->size >> PAGE_SHIFT;
	int ret = 0;

	if (roll > npages) {
		dev_err(obj->dev->dev, "invalid roll: %d\n", roll);
		return -EINVAL;
	}

	omap_obj->roll = roll;

	mutex_lock(&obj->dev->struct_mutex);

	/* if we aren't mapped yet, we don't need to do anything */
	if (omap_obj->block) {
		struct page **pages;
		ret = get_pages(obj, &pages);
		if (ret)
			goto fail;
		ret = tiler_pin(omap_obj->block, pages, npages, roll, true);
		if (ret)
			dev_err(obj->dev->dev, "could not repin: %d\n", ret);
	}

fail:
	mutex_unlock(&obj->dev->struct_mutex);

	return ret;
}

/* Sync the buffer for CPU access.. note pages should already be
 * attached, ie. omap_gem_get_pages()
 */
void omap_gem_cpu_sync(struct drm_gem_object *obj, int pgoff)
{
	struct drm_device *dev = obj->dev;
	struct omap_gem_object *omap_obj = to_omap_bo(obj);

	if (is_cached_coherent(obj) && omap_obj->addrs[pgoff]) {
		dma_unmap_page(dev->dev, omap_obj->addrs[pgoff],
				PAGE_SIZE, DMA_BIDIRECTIONAL);
		omap_obj->addrs[pgoff] = 0;
	}
}

/* sync the buffer for DMA access */
void omap_gem_dma_sync(struct drm_gem_object *obj,
		enum dma_data_direction dir)
{
	struct drm_device *dev = obj->dev;
	struct omap_gem_object *omap_obj = to_omap_bo(obj);

	if (is_cached_coherent(obj)) {
		int i, npages = obj->size >> PAGE_SHIFT;
		struct page **pages = omap_obj->pages;
		bool dirty = false;

		for (i = 0; i < npages; i++) {
			if (!omap_obj->addrs[i]) {
				omap_obj->addrs[i] = dma_map_page(dev->dev, pages[i], 0,
						PAGE_SIZE, DMA_BIDIRECTIONAL);
				dirty = true;
			}
		}

		if (dirty) {
			unmap_mapping_range(obj->filp->f_mapping, 0,
					omap_gem_mmap_size(obj), 1);
		}
	}
}

/* Get physical address for DMA.. if 'remap' is true, and the buffer is not
 * already contiguous, remap it to pin in physically contiguous memory.. (ie.
 * map in TILER)
 */
int omap_gem_get_paddr(struct drm_gem_object *obj,
		dma_addr_t *paddr, bool remap)
{
	struct omap_drm_private *priv = obj->dev->dev_private;
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = 0;

	mutex_lock(&obj->dev->struct_mutex);

	if (remap && is_shmem(obj) && priv->has_dmm) {
		if (omap_obj->paddr_cnt == 0) {
			struct page **pages;
			uint32_t npages = obj->size >> PAGE_SHIFT;
			enum tiler_fmt fmt = gem2fmt(omap_obj->flags);
			struct tiler_block *block;

			BUG_ON(omap_obj->block);

			ret = get_pages(obj, &pages);
			if (ret)
				goto fail;

			if (omap_obj->flags & OMAP_BO_TILED) {
				block = tiler_reserve_2d(fmt,
						omap_obj->width,
						omap_obj->height, 0);
			} else {
				block = tiler_reserve_1d(obj->size);
			}

			if (IS_ERR(block)) {
				ret = PTR_ERR(block);
				dev_err(obj->dev->dev,
					"could not remap: %d (%d)\n", ret, fmt);
				goto fail;
			}

			/* TODO: enable async refill.. */
			ret = tiler_pin(block, pages, npages,
					omap_obj->roll, true);
			if (ret) {
				tiler_release(block);
				dev_err(obj->dev->dev,
						"could not pin: %d\n", ret);
				goto fail;
			}

			omap_obj->paddr = tiler_ssptr(block);
			omap_obj->block = block;

			DBG("got paddr: %pad", &omap_obj->paddr);
		}

		omap_obj->paddr_cnt++;

		*paddr = omap_obj->paddr;
	} else if (omap_obj->flags & OMAP_BO_DMA) {
		*paddr = omap_obj->paddr;
	} else {
		ret = -EINVAL;
		goto fail;
	}

fail:
	mutex_unlock(&obj->dev->struct_mutex);

	return ret;
}

/* Release physical address, when DMA is no longer being performed.. this
 * could potentially unpin and unmap buffers from TILER
 */
void omap_gem_put_paddr(struct drm_gem_object *obj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret;

	mutex_lock(&obj->dev->struct_mutex);
	if (omap_obj->paddr_cnt > 0) {
		omap_obj->paddr_cnt--;
		if (omap_obj->paddr_cnt == 0) {
			ret = tiler_unpin(omap_obj->block);
			if (ret) {
				dev_err(obj->dev->dev,
					"could not unpin pages: %d\n", ret);
			}
			ret = tiler_release(omap_obj->block);
			if (ret) {
				dev_err(obj->dev->dev,
					"could not release unmap: %d\n", ret);
			}
			omap_obj->paddr = 0;
			omap_obj->block = NULL;
		}
	}

	mutex_unlock(&obj->dev->struct_mutex);
}

/* Get rotated scanout address (only valid if already pinned), at the
 * specified orientation and x,y offset from top-left corner of buffer
 * (only valid for tiled 2d buffers)
 */
int omap_gem_rotated_paddr(struct drm_gem_object *obj, uint32_t orient,
		int x, int y, dma_addr_t *paddr)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = -EINVAL;

	mutex_lock(&obj->dev->struct_mutex);
	if ((omap_obj->paddr_cnt > 0) && omap_obj->block &&
			(omap_obj->flags & OMAP_BO_TILED)) {
		*paddr = tiler_tsptr(omap_obj->block, orient, x, y);
		ret = 0;
	}
	mutex_unlock(&obj->dev->struct_mutex);
	return ret;
}

/* Get tiler stride for the buffer (only valid for 2d tiled buffers) */
int omap_gem_tiled_stride(struct drm_gem_object *obj, uint32_t orient)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = -EINVAL;
	if (omap_obj->flags & OMAP_BO_TILED)
		ret = tiler_stride(gem2fmt(omap_obj->flags), orient);
	return ret;
}

/* acquire pages when needed (for example, for DMA where physically
 * contiguous buffer is not required
 */
static int get_pages(struct drm_gem_object *obj, struct page ***pages)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = 0;

	if (is_shmem(obj) && !omap_obj->pages) {
		ret = omap_gem_attach_pages(obj);
		if (ret) {
			dev_err(obj->dev->dev, "could not attach pages\n");
			return ret;
		}
	}

	/* TODO: even phys-contig.. we should have a list of pages? */
	*pages = omap_obj->pages;

	return 0;
}

/* if !remap, and we don't have pages backing, then fail, rather than
 * increasing the pin count (which we don't really do yet anyways,
 * because we don't support swapping pages back out).  And 'remap'
 * might not be quite the right name, but I wanted to keep it working
 * similarly to omap_gem_get_paddr().  Note though that mutex is not
 * aquired if !remap (because this can be called in atomic ctxt),
 * but probably omap_gem_get_paddr() should be changed to work in the
 * same way.  If !remap, a matching omap_gem_put_pages() call is not
 * required (and should not be made).
 */
int omap_gem_get_pages(struct drm_gem_object *obj, struct page ***pages,
		bool remap)
{
	int ret;
	if (!remap) {
		struct omap_gem_object *omap_obj = to_omap_bo(obj);
		if (!omap_obj->pages)
			return -ENOMEM;
		*pages = omap_obj->pages;
		return 0;
	}
	mutex_lock(&obj->dev->struct_mutex);
	ret = get_pages(obj, pages);
	mutex_unlock(&obj->dev->struct_mutex);
	return ret;
}

/* release pages when DMA no longer being performed */
int omap_gem_put_pages(struct drm_gem_object *obj)
{
	/* do something here if we dynamically attach/detach pages.. at
	 * least they would no longer need to be pinned if everyone has
	 * released the pages..
	 */
	return 0;
}

/* Get kernel virtual address for CPU access.. this more or less only
 * exists for omap_fbdev.  This should be called with struct_mutex
 * held.
 */
void *omap_gem_vaddr(struct drm_gem_object *obj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
	if (!omap_obj->vaddr) {
		struct page **pages;
		int ret = get_pages(obj, &pages);
		if (ret)
			return ERR_PTR(ret);
		omap_obj->vaddr = vmap(pages, obj->size >> PAGE_SHIFT,
				VM_MAP, pgprot_writecombine(PAGE_KERNEL));
	}
	return omap_obj->vaddr;
}

#ifdef CONFIG_PM
/* re-pin objects in DMM in resume path: */
int omap_gem_resume(struct device *dev)
{
	struct drm_device *drm_dev = dev_get_drvdata(dev);
	struct omap_drm_private *priv = drm_dev->dev_private;
	struct omap_gem_object *omap_obj;
	int ret = 0;

	list_for_each_entry(omap_obj, &priv->obj_list, mm_list) {
		if (omap_obj->block) {
			struct drm_gem_object *obj = &omap_obj->base;
			uint32_t npages = obj->size >> PAGE_SHIFT;
			WARN_ON(!omap_obj->pages);  /* this can't happen */
			ret = tiler_pin(omap_obj->block,
					omap_obj->pages, npages,
					omap_obj->roll, true);
			if (ret) {
				dev_err(dev, "could not repin: %d\n", ret);
				return ret;
			}
		}
	}

	return 0;
}
#endif

#ifdef CONFIG_DEBUG_FS
void omap_gem_describe(struct drm_gem_object *obj, struct seq_file *m)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	uint64_t off;

	off = drm_vma_node_start(&obj->vma_node);

	seq_printf(m, "%08x: %2d (%2d) %08llx %pad (%2d) %p %4d",
			omap_obj->flags, obj->name, obj->refcount.refcount.counter,
			off, &omap_obj->paddr, omap_obj->paddr_cnt,
			omap_obj->vaddr, omap_obj->roll);

	if (omap_obj->flags & OMAP_BO_TILED) {
		seq_printf(m, " %dx%d", omap_obj->width, omap_obj->height);
		if (omap_obj->block) {
			struct tcm_area *area = &omap_obj->block->area;
			seq_printf(m, " (%dx%d, %dx%d)",
					area->p0.x, area->p0.y,
					area->p1.x, area->p1.y);
		}
	} else {
		seq_printf(m, " %d", obj->size);
	}

	seq_printf(m, "\n");
}

void omap_gem_describe_objects(struct list_head *list, struct seq_file *m)
{
	struct omap_gem_object *omap_obj;
	int count = 0;
	size_t size = 0;

	list_for_each_entry(omap_obj, list, mm_list) {
		struct drm_gem_object *obj = &omap_obj->base;
		seq_printf(m, "   ");
		omap_gem_describe(obj, m);
		count++;
		size += obj->size;
	}

	seq_printf(m, "Total %d objects, %zu bytes\n", count, size);
}
#endif

/* Buffer Synchronization:
 */

struct omap_gem_sync_waiter {
	struct list_head list;
	struct omap_gem_object *omap_obj;
	enum omap_gem_op op;
	uint32_t read_target, write_target;
	/* notify called w/ sync_lock held */
	void (*notify)(void *arg);
	void *arg;
};

/* list of omap_gem_sync_waiter.. the notify fxn gets called back when
 * the read and/or write target count is achieved which can call a user
 * callback (ex. to kick 3d and/or 2d), wakeup blocked task (prep for
 * cpu access), etc.
 */
static LIST_HEAD(waiters);

static inline bool is_waiting(struct omap_gem_sync_waiter *waiter)
{
	struct omap_gem_object *omap_obj = waiter->omap_obj;
	if ((waiter->op & OMAP_GEM_READ) &&
			(omap_obj->sync->write_complete < waiter->write_target))
		return true;
	if ((waiter->op & OMAP_GEM_WRITE) &&
			(omap_obj->sync->read_complete < waiter->read_target))
		return true;
	return false;
}

/* macro for sync debug.. */
#define SYNCDBG 0
#define SYNC(fmt, ...) do { if (SYNCDBG) \
		printk(KERN_ERR "%s:%d: "fmt"\n", \
				__func__, __LINE__, ##__VA_ARGS__); \
	} while (0)


static void sync_op_update(void)
{
	struct omap_gem_sync_waiter *waiter, *n;
	list_for_each_entry_safe(waiter, n, &waiters, list) {
		if (!is_waiting(waiter)) {
			list_del(&waiter->list);
			SYNC("notify: %p", waiter);
			waiter->notify(waiter->arg);
			kfree(waiter);
		}
	}
}

static inline int sync_op(struct drm_gem_object *obj,
		enum omap_gem_op op, bool start)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = 0;

	spin_lock(&sync_lock);

	if (!omap_obj->sync) {
		omap_obj->sync = kzalloc(sizeof(*omap_obj->sync), GFP_ATOMIC);
		if (!omap_obj->sync) {
			ret = -ENOMEM;
			goto unlock;
		}
	}

	if (start) {
		if (op & OMAP_GEM_READ)
			omap_obj->sync->read_pending++;
		if (op & OMAP_GEM_WRITE)
			omap_obj->sync->write_pending++;
	} else {
		if (op & OMAP_GEM_READ)
			omap_obj->sync->read_complete++;
		if (op & OMAP_GEM_WRITE)
			omap_obj->sync->write_complete++;
		sync_op_update();
	}

unlock:
	spin_unlock(&sync_lock);

	return ret;
}

/* it is a bit lame to handle updates in this sort of polling way, but
 * in case of PVR, the GPU can directly update read/write complete
 * values, and not really tell us which ones it updated.. this also
 * means that sync_lock is not quite sufficient.  So we'll need to
 * do something a bit better when it comes time to add support for
 * separate 2d hw..
 */
void omap_gem_op_update(void)
{
	spin_lock(&sync_lock);
	sync_op_update();
	spin_unlock(&sync_lock);
}

/* mark the start of read and/or write operation */
int omap_gem_op_start(struct drm_gem_object *obj, enum omap_gem_op op)
{
	return sync_op(obj, op, true);
}

int omap_gem_op_finish(struct drm_gem_object *obj, enum omap_gem_op op)
{
	return sync_op(obj, op, false);
}

static DECLARE_WAIT_QUEUE_HEAD(sync_event);

static void sync_notify(void *arg)
{
	struct task_struct **waiter_task = arg;
	*waiter_task = NULL;
	wake_up_all(&sync_event);
}

int omap_gem_op_sync(struct drm_gem_object *obj, enum omap_gem_op op)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = 0;
	if (omap_obj->sync) {
		struct task_struct *waiter_task = current;
		struct omap_gem_sync_waiter *waiter =
				kzalloc(sizeof(*waiter), GFP_KERNEL);

		if (!waiter)
			return -ENOMEM;

		waiter->omap_obj = omap_obj;
		waiter->op = op;
		waiter->read_target = omap_obj->sync->read_pending;
		waiter->write_target = omap_obj->sync->write_pending;
		waiter->notify = sync_notify;
		waiter->arg = &waiter_task;

		spin_lock(&sync_lock);
		if (is_waiting(waiter)) {
			SYNC("waited: %p", waiter);
			list_add_tail(&waiter->list, &waiters);
			spin_unlock(&sync_lock);
			ret = wait_event_interruptible(sync_event,
					(waiter_task == NULL));
			spin_lock(&sync_lock);
			if (waiter_task) {
				SYNC("interrupted: %p", waiter);
				/* we were interrupted */
				list_del(&waiter->list);
				waiter_task = NULL;
			} else {
				/* freed in sync_op_update() */
				waiter = NULL;
			}
		}
		spin_unlock(&sync_lock);
		kfree(waiter);
	}
	return ret;
}

/* call fxn(arg), either synchronously or asynchronously if the op
 * is currently blocked..  fxn() can be called from any context
 *
 * (TODO for now fxn is called back from whichever context calls
 * omap_gem_op_update().. but this could be better defined later
 * if needed)
 *
 * TODO more code in common w/ _sync()..
 */
int omap_gem_op_async(struct drm_gem_object *obj, enum omap_gem_op op,
		void (*fxn)(void *arg), void *arg)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	if (omap_obj->sync) {
		struct omap_gem_sync_waiter *waiter =
				kzalloc(sizeof(*waiter), GFP_ATOMIC);

		if (!waiter)
			return -ENOMEM;

		waiter->omap_obj = omap_obj;
		waiter->op = op;
		waiter->read_target = omap_obj->sync->read_pending;
		waiter->write_target = omap_obj->sync->write_pending;
		waiter->notify = fxn;
		waiter->arg = arg;

		spin_lock(&sync_lock);
		if (is_waiting(waiter)) {
			SYNC("waited: %p", waiter);
			list_add_tail(&waiter->list, &waiters);
			spin_unlock(&sync_lock);
			return 0;
		}

		spin_unlock(&sync_lock);

		kfree(waiter);
	}

	/* no waiting.. */
	fxn(arg);

	return 0;
}

/* special API so PVR can update the buffer to use a sync-object allocated
 * from it's sync-obj heap.  Only used for a newly allocated (from PVR's
 * perspective) sync-object, so we overwrite the new syncobj w/ values
 * from the already allocated syncobj (if there is one)
 */
int omap_gem_set_sync_object(struct drm_gem_object *obj, void *syncobj)
{
	struct omap_gem_object *omap_obj = to_omap_bo(obj);
	int ret = 0;

	spin_lock(&sync_lock);

	if ((omap_obj->flags & OMAP_BO_EXT_SYNC) && !syncobj) {
		/* clearing a previously set syncobj */
		syncobj = kmemdup(omap_obj->sync, sizeof(*omap_obj->sync),
				  GFP_ATOMIC);
		if (!syncobj) {
			ret = -ENOMEM;
			goto unlock;
		}
		omap_obj->flags &= ~OMAP_BO_EXT_SYNC;
		omap_obj->sync = syncobj;
	} else if (syncobj && !(omap_obj->flags & OMAP_BO_EXT_SYNC)) {
		/* replacing an existing syncobj */
		if (omap_obj->sync) {
			memcpy(syncobj, omap_obj->sync, sizeof(*omap_obj->sync));
			kfree(omap_obj->sync);
		}
		omap_obj->flags |= OMAP_BO_EXT_SYNC;
		omap_obj->sync = syncobj;
	}

unlock:
	spin_unlock(&sync_lock);
	return ret;
}

/* don't call directly.. called from GEM core when it is time to actually
 * free the object..
 */
void omap_gem_free_object(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct omap_drm_private *priv = dev->dev_private;
	struct omap_gem_object *omap_obj = to_omap_bo(obj);

	evict(obj);

	WARN_ON(!mutex_is_locked(&dev->struct_mutex));

	spin_lock(&priv->list_lock);
	list_del(&omap_obj->mm_list);
	spin_unlock(&priv->list_lock);

	drm_gem_free_mmap_offset(obj);

	/* this means the object is still pinned.. which really should
	 * not happen.  I think..
	 */
	WARN_ON(omap_obj->paddr_cnt > 0);

	/* don't free externally allocated backing memory */
	if (!(omap_obj->flags & OMAP_BO_EXT_MEM)) {
		if (omap_obj->pages)
			omap_gem_detach_pages(obj);

		if (!is_shmem(obj)) {
			dma_free_writecombine(dev->dev, obj->size,
					omap_obj->vaddr, omap_obj->paddr);
		} else if (omap_obj->vaddr) {
			vunmap(omap_obj->vaddr);
		}
	}

	/* don't free externally allocated syncobj */
	if (!(omap_obj->flags & OMAP_BO_EXT_SYNC))
		kfree(omap_obj->sync);

	drm_gem_object_release(obj);

	kfree(obj);
}

/* convenience method to construct a GEM buffer object, and userspace handle */
int omap_gem_new_handle(struct drm_device *dev, struct drm_file *file,
		union omap_gem_size gsize, uint32_t flags, uint32_t *handle)
{
	struct drm_gem_object *obj;
	int ret;

	obj = omap_gem_new(dev, gsize, flags);
	if (!obj)
		return -ENOMEM;

	ret = drm_gem_handle_create(file, obj, handle);
	if (ret) {
		drm_gem_object_release(obj);
		kfree(obj); /* TODO isn't there a dtor to call? just copying i915 */
		return ret;
	}

	/* drop reference from allocate - handle holds it now */
	drm_gem_object_unreference_unlocked(obj);

	return 0;
}

/* GEM buffer object constructor */
struct drm_gem_object *omap_gem_new(struct drm_device *dev,
		union omap_gem_size gsize, uint32_t flags)
{
	struct omap_drm_private *priv = dev->dev_private;
	struct omap_gem_object *omap_obj;
	struct drm_gem_object *obj = NULL;
	struct address_space *mapping;
	size_t size;
	int ret;

	if (flags & OMAP_BO_TILED) {
		if (!usergart) {
			dev_err(dev->dev, "Tiled buffers require DMM\n");
			goto fail;
		}

		/* tiled buffers are always shmem paged backed.. when they are
		 * scanned out, they are remapped into DMM/TILER
		 */
		flags &= ~OMAP_BO_SCANOUT;

		/* currently don't allow cached buffers.. there is some caching
		 * stuff that needs to be handled better
		 */
		flags &= ~(OMAP_BO_CACHED|OMAP_BO_WC|OMAP_BO_UNCACHED);
		flags |= tiler_get_cpu_cache_flags();

		/* align dimensions to slot boundaries... */
		tiler_align(gem2fmt(flags),
				&gsize.tiled.width, &gsize.tiled.height);

		/* ...and calculate size based on aligned dimensions */
		size = tiler_size(gem2fmt(flags),
				gsize.tiled.width, gsize.tiled.height);
	} else {
		size = PAGE_ALIGN(gsize.bytes);
	}

	omap_obj = kzalloc(sizeof(*omap_obj), GFP_KERNEL);
	if (!omap_obj)
		return NULL;

	obj = &omap_obj->base;

	if ((flags & OMAP_BO_SCANOUT) && !priv->has_dmm) {
		/* attempt to allocate contiguous memory if we don't
		 * have DMM for remappign discontiguous buffers
		 */
		omap_obj->vaddr =  dma_alloc_writecombine(dev->dev, size,
				&omap_obj->paddr, GFP_KERNEL);
		if (!omap_obj->vaddr) {
			kfree(omap_obj);

			return NULL;
		}

		flags |= OMAP_BO_DMA;
	}

	spin_lock(&priv->list_lock);
	list_add(&omap_obj->mm_list, &priv->obj_list);
	spin_unlock(&priv->list_lock);

	omap_obj->flags = flags;

	if (flags & OMAP_BO_TILED) {
		omap_obj->width = gsize.tiled.width;
		omap_obj->height = gsize.tiled.height;
	}

	if (flags & (OMAP_BO_DMA|OMAP_BO_EXT_MEM)) {
		drm_gem_private_object_init(dev, obj, size);
	} else {
		ret = drm_gem_object_init(dev, obj, size);
		if (ret)
			goto fail;

		mapping = file_inode(obj->filp)->i_mapping;
		mapping_set_gfp_mask(mapping, GFP_USER | __GFP_DMA32);
	}

	return obj;

fail:
	if (obj)
		omap_gem_free_object(obj);

	return NULL;
}

/* init/cleanup.. if DMM is used, we need to set some stuff up.. */
void omap_gem_init(struct drm_device *dev)
{
	struct omap_drm_private *priv = dev->dev_private;
	const enum tiler_fmt fmts[] = {
			TILFMT_8BIT, TILFMT_16BIT, TILFMT_32BIT
	};
	int i, j;

	if (!dmm_is_available()) {
		/* DMM only supported on OMAP4 and later, so this isn't fatal */
		dev_warn(dev->dev, "DMM not available, disable DMM support\n");
		return;
	}

	usergart = kcalloc(3, sizeof(*usergart), GFP_KERNEL);
	if (!usergart)
		return;

	/* reserve 4k aligned/wide regions for userspace mappings: */
	for (i = 0; i < ARRAY_SIZE(fmts); i++) {
		uint16_t h = 1, w = PAGE_SIZE >> i;
		tiler_align(fmts[i], &w, &h);
		/* note: since each region is 1 4kb page wide, and minimum
		 * number of rows, the height ends up being the same as the
		 * # of pages in the region
		 */
		usergart[i].height = h;
		usergart[i].height_shift = ilog2(h);
		usergart[i].stride_pfn = tiler_stride(fmts[i], 0) >> PAGE_SHIFT;
		usergart[i].slot_shift = ilog2((PAGE_SIZE / h) >> i);
		for (j = 0; j < NUM_USERGART_ENTRIES; j++) {
			struct usergart_entry *entry = &usergart[i].entry[j];
			struct tiler_block *block =
					tiler_reserve_2d(fmts[i], w, h,
							PAGE_SIZE);
			if (IS_ERR(block)) {
				dev_err(dev->dev,
						"reserve failed: %d, %d, %ld\n",
						i, j, PTR_ERR(block));
				return;
			}
			entry->paddr = tiler_ssptr(block);
			entry->block = block;

			DBG("%d:%d: %dx%d: paddr=%pad stride=%d", i, j, w, h,
					&entry->paddr,
					usergart[i].stride_pfn << PAGE_SHIFT);
		}
	}

	priv->has_dmm = true;
}

void omap_gem_deinit(struct drm_device *dev)
{
	/* I believe we can rely on there being no more outstanding GEM
	 * objects which could depend on usergart/dmm at this point.
	 */
	kfree(usergart);
}