summaryrefslogtreecommitdiff
path: root/shared-core
AgeCommit message (Collapse)Author
2006-08-22Initial i915 buffer object driverThomas Hellstrom
2006-08-22Bring in stripped TTM functionality.Thomas Hellstrom
2006-08-21i915 fence object driver implementing 2 fence object types:Thomas Hellstrom
0x00 EXE fence. Signals when command stream interpreter has reached the point where the fence was emitted. 0x01 FLUSH fence. Signals when command stream interpreter has reached the point where the fence was emitted, and all previous drawing operations have been completed and flushed. Implements busy wait (for fastest response time / high CPU) and lazy wait (User interrupt or timer driven).
2006-08-21User / Kernel space fence objects (device-independent part).Thomas Hellstrom
2006-08-10i965 code and Linux coding style < 0Dave Airlie
smack my whitespace up.
2006-08-08Add support for Intel i965G chipsets.Alan Hourihane
This is a patch prepared by Guangdeng Liao based off of Tungsten Graphics's final code drop.
2006-07-26Revert "Make sure busmastering gets disabled on module unload."Michel Dänzer
This reverts af7b89d7246efbed7d05c38fcaa6a13c4b89db90 commit. It causes an oops on X server shutdown here, and for the reporter of bug #7629 as well.
2006-07-26Bug #7629: Fix for CHIP_IS_AGP getting 'restored' with non-AGP cardsMichel Dänzer
Commit 2a47f6bfecea5dabcbf79d5e1aaf271f50070b89 caused the CHIP_IS_AGP flag to get 'restored' with PCI(e) cards. I can't think of a way to fix this without introducing a (otherwise redundant) CHIP_IS_PCI flag.
2006-07-19Make sure busmastering gets disabled on module unload.Adam Jackson
2006-07-19Use RADEON_RB3D_DSTCACHE_CTLSTAT instead of RADEON_RB2D_DSTCACHE_CTLSTAT.Michel Dänzer
The latter seems to be a read-only mirror of the former.
2006-07-19Make sure CHIP_IS_AGP flag is set when not overriding to PCI mode.Michel Dänzer
This allows using AGP after overriding to PCI mode in a previous session without reloading the DRM.
2006-07-19When writeback isn't used, actually disable it in the hardware.Michel Dänzer
Not doing this might waste bus bandwidth or even cause memory corruption or system crashes on systems that check bus transfers. No such incident has been reported though.
2006-07-19Implement RADEON_PARAM_SCRATCH_OFFSET getparam.Michel Dänzer
When this succeeds, userspace can read the scratch register contents from the mapped writeback page directly.
2006-07-19Some debug output when the getparam ioctl is called with an unknown parameter.Michel Dänzer
2006-07-19.cvsignore -> .gitignoreMichel Dänzer
Sort the merged file, remove the redundant explicit .ko lines and add some generated symlinks.
2006-07-11Keep hashed user tokens, with the following changes:Thomas Hellstrom
32-bit physical device addresses are mapped directly to user-tokens. No duplicate maps are allowed, and the addresses are assumed to be outside of the range 0x10000000 through 0x30000000. The user-token is identical to the 32-bit physical start-address of the map. 64-bit physical device addressed are mapped to user-tokens in the range 0x10000000 to 0x30000000 with page-size increments. The user_token should not be interpreted as an address. Other map types, like upcoming TTM maps are mapped to user-tokens in the range 0x10000000 to 0x30000000 with page-size increments. The user_token should not be interpreted as an address. This keeps compatibility with buggy drivers, while still implementing a hashed map lookup. The SiS and via device driver major bumps are reverted.
2006-07-10Change drm Map handles to be arbitrary 32-bit hash tokens in the rangeThomas Hellstrom
0x10000000 to 0x90000000 in PAGE_SIZE increments. Implement hashed map lookups. This potentially breaks both 2D and 3D drivers. If so, the corresponding 2D and 3D driver should be fixed, and it's corresponding drm device driver should have its major bumped as soon as possible. Bump sis and via drm device driver majors. The SiS and Unichrome 3D drivers are fixed in Mesa CVS HEAD and mesa_6_4_branch.
2006-07-05SiS 315 Awareness.Thomas Hellstrom
2006-06-22Remove spurious debug messages from i915 vblank config pathsKeith Packard
2006-06-21i915: Save vblank pipe configuration to restore on resumeKeith Packard
2006-06-19Add i915 ioctls to configure pipes for vblank interrupt.Keith Packard
i915 vblanks can be generated from either pipe a or b, however a disabled pipe generates no interrupts. This change allows the X server to select which pipe generates vblank interrupts.
2006-06-19Fix buffer cleanup on close. Move memory manager reset from final_contextThomas Hellstrom
to lastclose.
2006-06-19via: Bump version number and date.Thomas Hellstrom
2006-06-15via:Thomas Hellstrom
-Remove out of memory error message. -Move sman cleanup from final_context to lastclose. -Add the P4VM800PRO (?) PCI ID.
2006-06-06Merge in the drm-sman-branchThomas Hellstrom
2006-05-24Add support for r200 vertex programs (R200_EMIT_VAP_PVS_CNTL, and newRoland Scheidegger
packet type for making it possible to address whole tcl vector space and have a larger count)
2006-05-20add forgotten register define for previous commitRoland Scheidegger
2006-05-20Do a tcl state flush before accessing tcl vector space. This fixes someRoland Scheidegger
more problems with flickering (bug #6637). drm may not be appropriate place for this, since doing that flush there might both be overkill and insufficient in some cases. However, it's hard to figure out when that flush is needed, so this has to suffice. There does not seem to be a performance penalty associated with it.
2006-05-18add consts to radeon microcode.Dave Airlie
From: tilman
2006-04-23Fix from Benh for ppc r300 scratchDave Airlie
2006-04-20check for __FreeBSD_kernel__ (bug 3810)Brian Paul
2006-04-18Err, use "ifndef" rather than "if !", to avoid compiler warning.Eric Anholt
2006-04-18Use __LP64__ instead of checking the linux-specific BITS_PER_LONG.Eric Anholt
2006-04-09Revert a change that accidentally went in with whitespace changes fromEric Anholt
Linux, which broke on FreeBSD. DRM_COPY_*_IOCTL checks for the size parameter matching the ioctl's command size there, since the copin/out happened earlier.
2006-04-08Compile fixes for FreeBSD.Eric Anholt
2006-04-05coverity bugfix from the kernelDave Airlie
2006-03-25radeon fix up the PCI ids for new memory map like the kernel one.. notDave Airlie
perfect but should be very safe... align some other kernel bits i810 align with kernel
2006-03-20Bump driver date to reflect airlied's last fix.Adam Jackson
2006-03-19make some functions static in via driverDave Airlie
2006-03-17Add missing pci ids for new radeons (most but not all are pcie, r420,Roland Scheidegger
rv380, rv410), with the exception of the rs400 igps. Hopefully they no longer lock up with new ddx, but no guarantees... (bug #5413)
2006-03-08Fix bug I reintroducedDave Airlie
2006-03-08fix some use before NULL checkDave Airlie
2006-03-07ia64 support for r300_scratch. (not tested)Aapo Tahkola
2006-03-06Add general-purpose packet for manipulating scratch registers (r300)Aapo Tahkola
2006-02-25Add all radeon pci ids known by ddx, but only r350/rv350 and below (newRoland Scheidegger
chips may be problematic). Leave the existing entries for new chips in though. Remove ids not known by ddx (secondary ids, non-existant,...). Correct some entries (name/family). Make the radeon family enum look more alike the ddx/dri versions. See #5413
2006-02-19missed a piece of benh patchDave Airlie
2006-02-18fix brace placementDave Airlie
2006-02-18clear i915 interrupts sources on server exitDave Airlie
2006-02-18add proper checking for bitblt multiDave Airlie
2006-02-18add benh's memory management patchDave Airlie
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 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633
/**************************************************************************
 * 
 * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
 * All Rights Reserved.
 * 
 * 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, sub license, 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 NON-INFRINGEMENT.
 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
 * 
 **************************************************************************/

/* Originally a fake version of the buffer manager so that we can
 * prototype the changes in a driver fairly quickly, has been fleshed
 * out to a fully functional interim solution.
 *
 * Basically wraps the old style memory management in the new
 * programming interface, but is more expressive and avoids many of
 * the bugs in the old texture manager.
 */

#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <errno.h>
#include <xf86drm.h>
#include <pthread.h>
#include "intel_bufmgr.h"
#include "intel_bufmgr_priv.h"
#include "drm.h"
#include "i915_drm.h"
#include "mm.h"
#include "libdrm_lists.h"

/* Support gcc's __FUNCTION__ for people using other compilers */
#if !defined(__GNUC__) && !defined(__FUNCTION__)
# define __FUNCTION__ __func__ /* C99 */
#endif

#define DBG(...) do {					\
	if (bufmgr_fake->bufmgr.debug)			\
		drmMsg(__VA_ARGS__);			\
} while (0)

/* Internal flags:
 */
#define BM_NO_BACKING_STORE			0x00000001
#define BM_NO_FENCE_SUBDATA			0x00000002
#define BM_PINNED				0x00000004

/* Wrapper around mm.c's mem_block, which understands that you must
 * wait for fences to expire before memory can be freed.  This is
 * specific to our use of memcpy for uploads - an upload that was
 * processed through the command queue wouldn't need to care about
 * fences.
 */
#define MAX_RELOCS 4096

struct fake_buffer_reloc {
	/** Buffer object that the relocation points at. */
	drm_intel_bo *target_buf;
	/** Offset of the relocation entry within reloc_buf. */
	uint32_t offset;
	/**
	 * Cached value of the offset when we last performed this relocation.
	 */
	uint32_t last_target_offset;
	/** Value added to target_buf's offset to get the relocation entry. */
	uint32_t delta;
	/** Cache domains the target buffer is read into. */
	uint32_t read_domains;
	/** Cache domain the target buffer will have dirty cachelines in. */
	uint32_t write_domain;
};

struct block {
	struct block *next, *prev;
	struct mem_block *mem;	/* BM_MEM_AGP */

	/**
	 * Marks that the block is currently in the aperture and has yet to be
	 * fenced.
	 */
	unsigned on_hardware:1;
	/**
	 * Marks that the block is currently fenced (being used by rendering)
	 * and can't be freed until @fence is passed.
	 */
	unsigned fenced:1;

	/** Fence cookie for the block. */
	unsigned fence;		/* Split to read_fence, write_fence */

	drm_intel_bo *bo;
	void *virtual;
};

typedef struct _bufmgr_fake {
	drm_intel_bufmgr bufmgr;

	pthread_mutex_t lock;

	unsigned long low_offset;
	unsigned long size;
	void *virtual;

	struct mem_block *heap;

	unsigned buf_nr;	/* for generating ids */

	/**
	 * List of blocks which are currently in the GART but haven't been
	 * fenced yet.
	 */
	struct block on_hardware;
	/**
	 * List of blocks which are in the GART and have an active fence on
	 * them.
	 */
	struct block fenced;
	/**
	 * List of blocks which have an expired fence and are ready to be
	 * evicted.
	 */
	struct block lru;

	unsigned int last_fence;

	unsigned fail:1;
	unsigned need_fence:1;
	int thrashing;

	/**
	 * Driver callback to emit a fence, returning the cookie.
	 *
	 * This allows the driver to hook in a replacement for the DRM usage in
	 * bufmgr_fake.
	 *
	 * Currently, this also requires that a write flush be emitted before
	 * emitting the fence, but this should change.
	 */
	unsigned int (*fence_emit) (void *private);
	/** Driver callback to wait for a fence cookie to have passed. */
	void (*fence_wait) (unsigned int fence, void *private);
	void *fence_priv;

	/**
	 * Driver callback to execute a buffer.
	 *
	 * This allows the driver to hook in a replacement for the DRM usage in
	 * bufmgr_fake.
	 */
	int (*exec) (drm_intel_bo *bo, unsigned int used, void *priv);
	void *exec_priv;

	/** Driver-supplied argument to driver callbacks */
	void *driver_priv;
	/**
	 * Pointer to kernel-updated sarea data for the last completed user irq
	 */
	volatile int *last_dispatch;

	int fd;

	int debug;

	int performed_rendering;
} drm_intel_bufmgr_fake;

typedef struct _drm_intel_bo_fake {
	drm_intel_bo bo;

	unsigned id;		/* debug only */
	const char *name;

	unsigned dirty:1;
	/**
	 * has the card written to this buffer - we make need to copy it back
	 */
	unsigned card_dirty:1;
	unsigned int refcount;
	/* Flags may consist of any of the DRM_BO flags, plus
	 * DRM_BO_NO_BACKING_STORE and BM_NO_FENCE_SUBDATA, which are the
	 * first two driver private flags.
	 */
	uint64_t flags;
	/** Cache domains the target buffer is read into. */
	uint32_t read_domains;
	/** Cache domain the target buffer will have dirty cachelines in. */
	uint32_t write_domain;

	unsigned int alignment;
	int is_static, validated;
	unsigned int map_count;

	/** relocation list */
	struct fake_buffer_reloc *relocs;
	int nr_relocs;
	/**
	 * Total size of the target_bos of this buffer.
	 *
	 * Used for estimation in check_aperture.
	 */
	unsigned int child_size;

	struct block *block;
	void *backing_store;
	void (*invalidate_cb) (drm_intel_bo *bo, void *ptr);
	void *invalidate_ptr;
} drm_intel_bo_fake;

static int clear_fenced(drm_intel_bufmgr_fake *bufmgr_fake,
			unsigned int fence_cookie);

#define MAXFENCE 0x7fffffff

static int
FENCE_LTE(unsigned a, unsigned b)
{
	if (a == b)
		return 1;

	if (a < b && b - a < (1 << 24))
		return 1;

	if (a > b && MAXFENCE - a + b < (1 << 24))
		return 1;

	return 0;
}

void
drm_intel_bufmgr_fake_set_fence_callback(drm_intel_bufmgr *bufmgr,
					 unsigned int (*emit) (void *priv),
					 void (*wait) (unsigned int fence,
						       void *priv),
					 void *priv)
{
	drm_intel_bufmgr_fake *bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;

	bufmgr_fake->fence_emit = emit;
	bufmgr_fake->fence_wait = wait;
	bufmgr_fake->fence_priv = priv;
}

static unsigned int
_fence_emit_internal(drm_intel_bufmgr_fake *bufmgr_fake)
{
	struct drm_i915_irq_emit ie;
	int ret, seq = 1;

	if (bufmgr_fake->fence_emit != NULL) {
		seq = bufmgr_fake->fence_emit(bufmgr_fake->fence_priv);
		return seq;
	}

	ie.irq_seq = &seq;
	ret = drmCommandWriteRead(bufmgr_fake->fd, DRM_I915_IRQ_EMIT,
				  &ie, sizeof(ie));
	if (ret) {
		drmMsg("%s: drm_i915_irq_emit: %d\n", __FUNCTION__, ret);
		abort();
	}

	DBG("emit 0x%08x\n", seq);
	return seq;
}

static void
_fence_wait_internal(drm_intel_bufmgr_fake *bufmgr_fake, int seq)
{
	struct drm_i915_irq_wait iw;
	int hw_seq, busy_count = 0;
	int ret;
	int kernel_lied;

	if (bufmgr_fake->fence_wait != NULL) {
		bufmgr_fake->fence_wait(seq, bufmgr_fake->fence_priv);
		clear_fenced(bufmgr_fake, seq);
		return;
	}

	iw.irq_seq = seq;

	DBG("wait 0x%08x\n", iw.irq_seq);

	/* The kernel IRQ_WAIT implementation is all sorts of broken.
	 * 1) It returns 1 to 0x7fffffff instead of using the full 32-bit
	 *    unsigned range.
	 * 2) It returns 0 if hw_seq >= seq, not seq - hw_seq < 0 on the 32-bit
	 *    signed range.
	 * 3) It waits if seq < hw_seq, not seq - hw_seq > 0 on the 32-bit
	 *    signed range.
	 * 4) It returns -EBUSY in 3 seconds even if the hardware is still
	 *    successfully chewing through buffers.
	 *
	 * Assume that in userland we treat sequence numbers as ints, which
	 * makes some of the comparisons convenient, since the sequence
	 * numbers are all postive signed integers.
	 *
	 * From this we get several cases we need to handle.  Here's a timeline.
	 * 0x2   0x7                                    0x7ffffff8   0x7ffffffd
	 *   |    |                                             |    |
	 * ------------------------------------------------------------
	 *
	 * A) Normal wait for hw to catch up
	 * hw_seq seq
	 *   |    |
	 * ------------------------------------------------------------
	 * seq - hw_seq = 5.  If we call IRQ_WAIT, it will wait for hw to
	 * catch up.
	 *
	 * B) Normal wait for a sequence number that's already passed.
	 * seq    hw_seq
	 *   |    |
	 * ------------------------------------------------------------
	 * seq - hw_seq = -5.  If we call IRQ_WAIT, it returns 0 quickly.
	 *
	 * C) Hardware has already wrapped around ahead of us
	 * hw_seq                                                    seq
	 *   |                                                       |
	 * ------------------------------------------------------------
	 * seq - hw_seq = 0x80000000 - 5.  If we called IRQ_WAIT, it would wait
	 * for hw_seq >= seq, which may never occur.  Thus, we want to catch
	 * this in userland and return 0.
	 *
	 * D) We've wrapped around ahead of the hardware.
	 * seq                                                      hw_seq
	 *   |                                                       |
	 * ------------------------------------------------------------
	 * seq - hw_seq = -(0x80000000 - 5).  If we called IRQ_WAIT, it would
	 * return 0 quickly because hw_seq >= seq, even though the hardware
	 * isn't caught up. Thus, we need to catch this early return in
	 * userland and bother the kernel until the hardware really does
	 * catch up.
	 *
	 * E) Hardware might wrap after we test in userland.
	 *                                                  hw_seq  seq
	 *                                                      |    |
	 * ------------------------------------------------------------
	 * seq - hw_seq = 5.  If we call IRQ_WAIT, it will likely see seq >=
	 * hw_seq and wait.  However, suppose hw_seq wraps before we make it
	 * into the kernel.  The kernel sees hw_seq >= seq and waits for 3
	 * seconds then returns -EBUSY.  This is case C).  We should catch
	 * this and then return successfully.
	 *
	 * F) Hardware might take a long time on a buffer.
	 * hw_seq seq
	 *   |    |
	 * -------------------------------------------------------------------
	 * seq - hw_seq = 5.  If we call IRQ_WAIT, if sequence 2 through 5
	 * take too long, it will return -EBUSY.  Batchbuffers in the
	 * gltestperf demo were seen to take up to 7 seconds.  We should
	 * catch early -EBUSY return and keep trying.
	 */

	do {
		/* Keep a copy of last_dispatch so that if the wait -EBUSYs
		 * because the hardware didn't catch up in 3 seconds, we can
		 * see if it at least made progress and retry.
		 */
		hw_seq = *bufmgr_fake->last_dispatch;

		/* Catch case C */
		if (seq - hw_seq > 0x40000000)
			return;

		ret = drmCommandWrite(bufmgr_fake->fd, DRM_I915_IRQ_WAIT,
				      &iw, sizeof(iw));
		/* Catch case D */
		kernel_lied = (ret == 0) && (seq - *bufmgr_fake->last_dispatch <
					     -0x40000000);

		/* Catch case E */
		if (ret == -EBUSY
		    && (seq - *bufmgr_fake->last_dispatch > 0x40000000))
			ret = 0;

		/* Catch case F: Allow up to 15 seconds chewing on one buffer. */
		if ((ret == -EBUSY) && (hw_seq != *bufmgr_fake->last_dispatch))
			busy_count = 0;
		else
			busy_count++;
	} while (kernel_lied || ret == -EAGAIN || ret == -EINTR ||
		 (ret == -EBUSY && busy_count < 5));

	if (ret != 0) {
		drmMsg("%s:%d: Error waiting for fence: %s.\n", __FILE__,
		       __LINE__, strerror(-ret));
		abort();
	}
	clear_fenced(bufmgr_fake, seq);
}

static int
_fence_test(drm_intel_bufmgr_fake *bufmgr_fake, unsigned fence)
{
	/* Slight problem with wrap-around:
	 */
	return fence == 0 || FENCE_LTE(fence, bufmgr_fake->last_fence);
}

/**
 * Allocate a memory manager block for the buffer.
 */
static int
alloc_block(drm_intel_bo *bo)
{
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	struct block *block = (struct block *)calloc(sizeof *block, 1);
	unsigned int align_log2 = ffs(bo_fake->alignment) - 1;
	unsigned int sz;

	if (!block)
		return 1;

	sz = (bo->size + bo_fake->alignment - 1) & ~(bo_fake->alignment - 1);

	block->mem = mmAllocMem(bufmgr_fake->heap, sz, align_log2, 0);
	if (!block->mem) {
		free(block);
		return 0;
	}

	DRMINITLISTHEAD(block);

	/* Insert at head or at tail??? */
	DRMLISTADDTAIL(block, &bufmgr_fake->lru);

	block->virtual = (uint8_t *) bufmgr_fake->virtual +
	    block->mem->ofs - bufmgr_fake->low_offset;
	block->bo = bo;

	bo_fake->block = block;

	return 1;
}

/* Release the card storage associated with buf:
 */
static void
free_block(drm_intel_bufmgr_fake *bufmgr_fake, struct block *block,
	   int skip_dirty_copy)
{
	drm_intel_bo_fake *bo_fake;
	DBG("free block %p %08x %d %d\n", block, block->mem->ofs,
	    block->on_hardware, block->fenced);

	if (!block)
		return;

	bo_fake = (drm_intel_bo_fake *) block->bo;

	if (bo_fake->flags & (BM_PINNED | BM_NO_BACKING_STORE))
		skip_dirty_copy = 1;

	if (!skip_dirty_copy && (bo_fake->card_dirty == 1)) {
		memcpy(bo_fake->backing_store, block->virtual, block->bo->size);
		bo_fake->card_dirty = 0;
		bo_fake->dirty = 1;
	}

	if (block->on_hardware) {
		block->bo = NULL;
	} else if (block->fenced) {
		block->bo = NULL;
	} else {
		DBG("    - free immediately\n");
		DRMLISTDEL(block);

		mmFreeMem(block->mem);
		free(block);
	}
}

static void
alloc_backing_store(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	assert(!bo_fake->backing_store);
	assert(!(bo_fake->flags & (BM_PINNED | BM_NO_BACKING_STORE)));

	bo_fake->backing_store = malloc(bo->size);

	DBG("alloc_backing - buf %d %p %d\n", bo_fake->id,
	    bo_fake->backing_store, bo->size);
	assert(bo_fake->backing_store);
}

static void
free_backing_store(drm_intel_bo *bo)
{
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	if (bo_fake->backing_store) {
		assert(!(bo_fake->flags & (BM_PINNED | BM_NO_BACKING_STORE)));
		free(bo_fake->backing_store);
		bo_fake->backing_store = NULL;
	}
}

static void
set_dirty(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	if (bo_fake->flags & BM_NO_BACKING_STORE
	    && bo_fake->invalidate_cb != NULL)
		bo_fake->invalidate_cb(bo, bo_fake->invalidate_ptr);

	assert(!(bo_fake->flags & BM_PINNED));

	DBG("set_dirty - buf %d\n", bo_fake->id);
	bo_fake->dirty = 1;
}

static int
evict_lru(drm_intel_bufmgr_fake *bufmgr_fake, unsigned int max_fence)
{
	struct block *block, *tmp;

	DBG("%s\n", __FUNCTION__);

	DRMLISTFOREACHSAFE(block, tmp, &bufmgr_fake->lru) {
		drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) block->bo;

		if (bo_fake != NULL && (bo_fake->flags & BM_NO_FENCE_SUBDATA))
			continue;

		if (block->fence && max_fence && !FENCE_LTE(block->fence,
							    max_fence))
			return 0;

		set_dirty(&bo_fake->bo);
		bo_fake->block = NULL;

		free_block(bufmgr_fake, block, 0);
		return 1;
	}

	return 0;
}

static int
evict_mru(drm_intel_bufmgr_fake *bufmgr_fake)
{
	struct block *block, *tmp;

	DBG("%s\n", __FUNCTION__);

	DRMLISTFOREACHSAFEREVERSE(block, tmp, &bufmgr_fake->lru) {
		drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) block->bo;

		if (bo_fake && (bo_fake->flags & BM_NO_FENCE_SUBDATA))
			continue;

		set_dirty(&bo_fake->bo);
		bo_fake->block = NULL;

		free_block(bufmgr_fake, block, 0);
		return 1;
	}

	return 0;
}

/**
 * Removes all objects from the fenced list older than the given fence.
 */
static int
clear_fenced(drm_intel_bufmgr_fake *bufmgr_fake, unsigned int fence_cookie)
{
	struct block *block, *tmp;
	int ret = 0;

	bufmgr_fake->last_fence = fence_cookie;
	DRMLISTFOREACHSAFE(block, tmp, &bufmgr_fake->fenced) {
		assert(block->fenced);

		if (_fence_test(bufmgr_fake, block->fence)) {

			block->fenced = 0;

			if (!block->bo) {
				DBG("delayed free: offset %x sz %x\n",
				    block->mem->ofs, block->mem->size);
				DRMLISTDEL(block);
				mmFreeMem(block->mem);
				free(block);
			} else {
				DBG("return to lru: offset %x sz %x\n",
				    block->mem->ofs, block->mem->size);
				DRMLISTDEL(block);
				DRMLISTADDTAIL(block, &bufmgr_fake->lru);
			}

			ret = 1;
		} else {
			/* Blocks are ordered by fence, so if one fails, all
			 * from here will fail also:
			 */
			DBG("fence not passed: offset %x sz %x %d %d \n",
			    block->mem->ofs, block->mem->size, block->fence,
			    bufmgr_fake->last_fence);
			break;
		}
	}

	DBG("%s: %d\n", __FUNCTION__, ret);
	return ret;
}

static void
fence_blocks(drm_intel_bufmgr_fake *bufmgr_fake, unsigned fence)
{
	struct block *block, *tmp;

	DRMLISTFOREACHSAFE(block, tmp, &bufmgr_fake->on_hardware) {
		DBG("Fence block %p (sz 0x%x ofs %x buf %p) with fence %d\n",
		    block, block->mem->size, block->mem->ofs, block->bo, fence);
		block->fence = fence;

		block->on_hardware = 0;
		block->fenced = 1;

		/* Move to tail of pending list here
		 */
		DRMLISTDEL(block);
		DRMLISTADDTAIL(block, &bufmgr_fake->fenced);
	}

	assert(DRMLISTEMPTY(&bufmgr_fake->on_hardware));
}

static int
evict_and_alloc_block(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	assert(bo_fake->block == NULL);

	/* Search for already free memory:
	 */
	if (alloc_block(bo))
		return 1;

	/* If we're not thrashing, allow lru eviction to dig deeper into
	 * recently used textures.  We'll probably be thrashing soon:
	 */
	if (!bufmgr_fake->thrashing) {
		while (evict_lru(bufmgr_fake, 0))
			if (alloc_block(bo))
				return 1;
	}

	/* Keep thrashing counter alive?
	 */
	if (bufmgr_fake->thrashing)
		bufmgr_fake->thrashing = 20;

	/* Wait on any already pending fences - here we are waiting for any
	 * freed memory that has been submitted to hardware and fenced to
	 * become available:
	 */
	while (!DRMLISTEMPTY(&bufmgr_fake->fenced)) {
		uint32_t fence = bufmgr_fake->fenced.next->fence;
		_fence_wait_internal(bufmgr_fake, fence);

		if (alloc_block(bo))
			return 1;
	}

	if (!DRMLISTEMPTY(&bufmgr_fake->on_hardware)) {
		while (!DRMLISTEMPTY(&bufmgr_fake->fenced)) {
			uint32_t fence = bufmgr_fake->fenced.next->fence;
			_fence_wait_internal(bufmgr_fake, fence);
		}

		if (!bufmgr_fake->thrashing) {
			DBG("thrashing\n");
		}
		bufmgr_fake->thrashing = 20;

		if (alloc_block(bo))
			return 1;
	}

	while (evict_mru(bufmgr_fake))
		if (alloc_block(bo))
			return 1;

	DBG("%s 0x%x bytes failed\n", __FUNCTION__, bo->size);

	return 0;
}

/***********************************************************************
 * Public functions
 */

/**
 * Wait for hardware idle by emitting a fence and waiting for it.
 */
static void
drm_intel_bufmgr_fake_wait_idle(drm_intel_bufmgr_fake *bufmgr_fake)
{
	unsigned int cookie;

	cookie = _fence_emit_internal(bufmgr_fake);
	_fence_wait_internal(bufmgr_fake, cookie);
}

/**
 * Wait for rendering to a buffer to complete.
 *
 * It is assumed that the bathcbuffer which performed the rendering included
 * the necessary flushing.
 */
static void
drm_intel_fake_bo_wait_rendering_locked(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	if (bo_fake->block == NULL || !bo_fake->block->fenced)
		return;

	_fence_wait_internal(bufmgr_fake, bo_fake->block->fence);
}

static void
drm_intel_fake_bo_wait_rendering(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;

	pthread_mutex_lock(&bufmgr_fake->lock);
	drm_intel_fake_bo_wait_rendering_locked(bo);
	pthread_mutex_unlock(&bufmgr_fake->lock);
}

/* Specifically ignore texture memory sharing.
 *  -- just evict everything
 *  -- and wait for idle
 */
void
drm_intel_bufmgr_fake_contended_lock_take(drm_intel_bufmgr *bufmgr)
{
	drm_intel_bufmgr_fake *bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;
	struct block *block, *tmp;

	pthread_mutex_lock(&bufmgr_fake->lock);

	bufmgr_fake->need_fence = 1;
	bufmgr_fake->fail = 0;

	/* Wait for hardware idle.  We don't know where acceleration has been
	 * happening, so we'll need to wait anyway before letting anything get
	 * put on the card again.
	 */
	drm_intel_bufmgr_fake_wait_idle(bufmgr_fake);

	/* Check that we hadn't released the lock without having fenced the last
	 * set of buffers.
	 */
	assert(DRMLISTEMPTY(&bufmgr_fake->fenced));
	assert(DRMLISTEMPTY(&bufmgr_fake->on_hardware));

	DRMLISTFOREACHSAFE(block, tmp, &bufmgr_fake->lru) {
		assert(_fence_test(bufmgr_fake, block->fence));
		set_dirty(block->bo);
	}

	pthread_mutex_unlock(&bufmgr_fake->lock);
}

static drm_intel_bo *
drm_intel_fake_bo_alloc(drm_intel_bufmgr *bufmgr,
			const char *name,
			unsigned long size,
			unsigned int alignment)
{
	drm_intel_bufmgr_fake *bufmgr_fake;
	drm_intel_bo_fake *bo_fake;

	bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;

	assert(size != 0);

	bo_fake = calloc(1, sizeof(*bo_fake));
	if (!bo_fake)
		return NULL;

	bo_fake->bo.size = size;
	bo_fake->bo.offset = -1;
	bo_fake->bo.virtual = NULL;
	bo_fake->bo.bufmgr = bufmgr;
	bo_fake->refcount = 1;

	/* Alignment must be a power of two */
	assert((alignment & (alignment - 1)) == 0);
	if (alignment == 0)
		alignment = 1;
	bo_fake->alignment = alignment;
	bo_fake->id = ++bufmgr_fake->buf_nr;
	bo_fake->name = name;
	bo_fake->flags = 0;
	bo_fake->is_static = 0;

	DBG("drm_bo_alloc: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
	    bo_fake->bo.size / 1024);

	return &bo_fake->bo;
}

static drm_intel_bo *
drm_intel_fake_bo_alloc_tiled(drm_intel_bufmgr * bufmgr,
			      const char *name,
			      int x, int y, int cpp,
			      uint32_t *tiling_mode,
			      unsigned long *pitch,
			      unsigned long flags)
{
	unsigned long stride, aligned_y;

	/* No runtime tiling support for fake. */
	*tiling_mode = I915_TILING_NONE;

	/* Align it for being a render target.  Shouldn't need anything else. */
	stride = x * cpp;
	stride = ROUND_UP_TO(stride, 64);

	/* 965 subspan loading alignment */
	aligned_y = ALIGN(y, 2);

	*pitch = stride;

	return drm_intel_fake_bo_alloc(bufmgr, name, stride * aligned_y,
				       4096);
}

drm_intel_bo *
drm_intel_bo_fake_alloc_static(drm_intel_bufmgr *bufmgr,
			       const char *name,
			       unsigned long offset,
			       unsigned long size, void *virtual)
{
	drm_intel_bufmgr_fake *bufmgr_fake;
	drm_intel_bo_fake *bo_fake;

	bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;

	assert(size != 0);

	bo_fake = calloc(1, sizeof(*bo_fake));
	if (!bo_fake)
		return NULL;

	bo_fake->bo.size = size;
	bo_fake->bo.offset = offset;
	bo_fake->bo.virtual = virtual;
	bo_fake->bo.bufmgr = bufmgr;
	bo_fake->refcount = 1;
	bo_fake->id = ++bufmgr_fake->buf_nr;
	bo_fake->name = name;
	bo_fake->flags = BM_PINNED;
	bo_fake->is_static = 1;

	DBG("drm_bo_alloc_static: (buf %d: %s, %d kb)\n", bo_fake->id,
	    bo_fake->name, bo_fake->bo.size / 1024);

	return &bo_fake->bo;
}

static void
drm_intel_fake_bo_reference(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	pthread_mutex_lock(&bufmgr_fake->lock);
	bo_fake->refcount++;
	pthread_mutex_unlock(&bufmgr_fake->lock);
}

static void
drm_intel_fake_bo_reference_locked(drm_intel_bo *bo)
{
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	bo_fake->refcount++;
}

static void
drm_intel_fake_bo_unreference_locked(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	int i;

	if (--bo_fake->refcount == 0) {
		assert(bo_fake->map_count == 0);
		/* No remaining references, so free it */
		if (bo_fake->block)
			free_block(bufmgr_fake, bo_fake->block, 1);
		free_backing_store(bo);

		for (i = 0; i < bo_fake->nr_relocs; i++)
			drm_intel_fake_bo_unreference_locked(bo_fake->relocs[i].
							     target_buf);

		DBG("drm_bo_unreference: free buf %d %s\n", bo_fake->id,
		    bo_fake->name);

		free(bo_fake->relocs);
		free(bo);
	}
}

static void
drm_intel_fake_bo_unreference(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;

	pthread_mutex_lock(&bufmgr_fake->lock);
	drm_intel_fake_bo_unreference_locked(bo);
	pthread_mutex_unlock(&bufmgr_fake->lock);
}

/**
 * Set the buffer as not requiring backing store, and instead get the callback
 * invoked whenever it would be set dirty.
 */
void
drm_intel_bo_fake_disable_backing_store(drm_intel_bo *bo,
					void (*invalidate_cb) (drm_intel_bo *bo,
							       void *ptr),
					void *ptr)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	pthread_mutex_lock(&bufmgr_fake->lock);

	if (bo_fake->backing_store)
		free_backing_store(bo);

	bo_fake->flags |= BM_NO_BACKING_STORE;

	DBG("disable_backing_store set buf %d dirty\n", bo_fake->id);
	bo_fake->dirty = 1;
	bo_fake->invalidate_cb = invalidate_cb;
	bo_fake->invalidate_ptr = ptr;

	/* Note that it is invalid right from the start.  Also note
	 * invalidate_cb is called with the bufmgr locked, so cannot
	 * itself make bufmgr calls.
	 */
	if (invalidate_cb != NULL)
		invalidate_cb(bo, ptr);

	pthread_mutex_unlock(&bufmgr_fake->lock);
}

/**
 * Map a buffer into bo->virtual, allocating either card memory space (If
 * BM_NO_BACKING_STORE or BM_PINNED) or backing store, as necessary.
 */
static int
 drm_intel_fake_bo_map_locked(drm_intel_bo *bo, int write_enable)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	/* Static buffers are always mapped. */
	if (bo_fake->is_static) {
		if (bo_fake->card_dirty) {
			drm_intel_bufmgr_fake_wait_idle(bufmgr_fake);
			bo_fake->card_dirty = 0;
		}
		return 0;
	}

	/* Allow recursive mapping.  Mesa may recursively map buffers with
	 * nested display loops, and it is used internally in bufmgr_fake
	 * for relocation.
	 */
	if (bo_fake->map_count++ != 0)
		return 0;

	{
		DBG("drm_bo_map: (buf %d: %s, %d kb)\n", bo_fake->id,
		    bo_fake->name, bo_fake->bo.size / 1024);

		if (bo->virtual != NULL) {
			drmMsg("%s: already mapped\n", __FUNCTION__);
			abort();
		} else if (bo_fake->flags & (BM_NO_BACKING_STORE | BM_PINNED)) {

			if (!bo_fake->block && !evict_and_alloc_block(bo)) {
				DBG("%s: alloc failed\n", __FUNCTION__);
				bufmgr_fake->fail = 1;
				return 1;
			} else {
				assert(bo_fake->block);
				bo_fake->dirty = 0;

				if (!(bo_fake->flags & BM_NO_FENCE_SUBDATA) &&
				    bo_fake->block->fenced) {
					drm_intel_fake_bo_wait_rendering_locked
					    (bo);
				}

				bo->virtual = bo_fake->block->virtual;
			}
		} else {
			if (write_enable)
				set_dirty(bo);

			if (bo_fake->backing_store == 0)
				alloc_backing_store(bo);

			if ((bo_fake->card_dirty == 1) && bo_fake->block) {
				if (bo_fake->block->fenced)
					drm_intel_fake_bo_wait_rendering_locked
					    (bo);

				memcpy(bo_fake->backing_store,
				       bo_fake->block->virtual,
				       bo_fake->block->bo->size);
				bo_fake->card_dirty = 0;
			}

			bo->virtual = bo_fake->backing_store;
		}
	}

	return 0;
}

static int
 drm_intel_fake_bo_map(drm_intel_bo *bo, int write_enable)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	int ret;

	pthread_mutex_lock(&bufmgr_fake->lock);
	ret = drm_intel_fake_bo_map_locked(bo, write_enable);
	pthread_mutex_unlock(&bufmgr_fake->lock);

	return ret;
}

static int
 drm_intel_fake_bo_unmap_locked(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	/* Static buffers are always mapped. */
	if (bo_fake->is_static)
		return 0;

	assert(bo_fake->map_count != 0);
	if (--bo_fake->map_count != 0)
		return 0;

	DBG("drm_bo_unmap: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
	    bo_fake->bo.size / 1024);

	bo->virtual = NULL;

	return 0;
}

static int drm_intel_fake_bo_unmap(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	int ret;

	pthread_mutex_lock(&bufmgr_fake->lock);
	ret = drm_intel_fake_bo_unmap_locked(bo);
	pthread_mutex_unlock(&bufmgr_fake->lock);

	return ret;
}

static int
drm_intel_fake_bo_subdata(drm_intel_bo *bo, unsigned long offset,
			  unsigned long size, const void *data)
{
	int ret;

	if (size == 0 || data == NULL)
		return 0;

	ret = drm_intel_bo_map(bo, 1);
	if (ret)
		return ret;
	memcpy((unsigned char *)bo->virtual + offset, data, size);
	drm_intel_bo_unmap(bo);
	return 0;
}

static void
 drm_intel_fake_kick_all_locked(drm_intel_bufmgr_fake *bufmgr_fake)
{
	struct block *block, *tmp;

	bufmgr_fake->performed_rendering = 0;
	/* okay for ever BO that is on the HW kick it off.
	   seriously not afraid of the POLICE right now */
	DRMLISTFOREACHSAFE(block, tmp, &bufmgr_fake->on_hardware) {
		drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) block->bo;

		block->on_hardware = 0;
		free_block(bufmgr_fake, block, 0);
		bo_fake->block = NULL;
		bo_fake->validated = 0;
		if (!(bo_fake->flags & BM_NO_BACKING_STORE))
			bo_fake->dirty = 1;
	}

}

static int
 drm_intel_fake_bo_validate(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;

	bufmgr_fake = (drm_intel_bufmgr_fake *) bo->bufmgr;

	DBG("drm_bo_validate: (buf %d: %s, %d kb)\n", bo_fake->id,
	    bo_fake->name, bo_fake->bo.size / 1024);

	/* Sanity check: Buffers should be unmapped before being validated.
	 * This is not so much of a problem for bufmgr_fake, but TTM refuses,
	 * and the problem is harder to debug there.
	 */
	assert(bo_fake->map_count == 0);

	if (bo_fake->is_static) {
		/* Add it to the needs-fence list */
		bufmgr_fake->need_fence = 1;
		return 0;
	}

	/* Allocate the card memory */
	if (!bo_fake->block && !evict_and_alloc_block(bo)) {
		bufmgr_fake->fail = 1;
		DBG("Failed to validate buf %d:%s\n", bo_fake->id,
		    bo_fake->name);
		return -1;
	}

	assert(bo_fake->block);
	assert(bo_fake->block->bo == &bo_fake->bo);

	bo->offset = bo_fake->block->mem->ofs;

	/* Upload the buffer contents if necessary */
	if (bo_fake->dirty) {
		DBG("Upload dirty buf %d:%s, sz %d offset 0x%x\n", bo_fake->id,
		    bo_fake->name, bo->size, bo_fake->block->mem->ofs);

		assert(!(bo_fake->flags & (BM_NO_BACKING_STORE | BM_PINNED)));

		/* Actually, should be able to just wait for a fence on the
		 * mmory, hich we would be tracking when we free it.  Waiting
		 * for idle is a sufficiently large hammer for now.
		 */
		drm_intel_bufmgr_fake_wait_idle(bufmgr_fake);

		/* we may never have mapped this BO so it might not have any
		 * backing store if this happens it should be rare, but 0 the
		 * card memory in any case */
		if (bo_fake->backing_store)
			memcpy(bo_fake->block->virtual, bo_fake->backing_store,
			       bo->size);
		else
			memset(bo_fake->block->virtual, 0, bo->size);

		bo_fake->dirty = 0;
	}

	bo_fake->block->fenced = 0;
	bo_fake->block->on_hardware = 1;
	DRMLISTDEL(bo_fake->block);
	DRMLISTADDTAIL(bo_fake->block, &bufmgr_fake->on_hardware);

	bo_fake->validated = 1;
	bufmgr_fake->need_fence = 1;

	return 0;
}

static void
drm_intel_fake_fence_validated(drm_intel_bufmgr *bufmgr)
{
	drm_intel_bufmgr_fake *bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;
	unsigned int cookie;

	cookie = _fence_emit_internal(bufmgr_fake);
	fence_blocks(bufmgr_fake, cookie);

	DBG("drm_fence_validated: 0x%08x cookie\n", cookie);
}

static void
drm_intel_fake_destroy(drm_intel_bufmgr *bufmgr)
{
	drm_intel_bufmgr_fake *bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;

	pthread_mutex_destroy(&bufmgr_fake->lock);
	mmDestroy(bufmgr_fake->heap);
	free(bufmgr);
}

static int
drm_intel_fake_emit_reloc(drm_intel_bo *bo, uint32_t offset,
			  drm_intel_bo *target_bo, uint32_t target_offset,
			  uint32_t read_domains, uint32_t write_domain)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	struct fake_buffer_reloc *r;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	drm_intel_bo_fake *target_fake = (drm_intel_bo_fake *) target_bo;
	int i;

	pthread_mutex_lock(&bufmgr_fake->lock);

	assert(bo);
	assert(target_bo);

	if (bo_fake->relocs == NULL) {
		bo_fake->relocs =
		    malloc(sizeof(struct fake_buffer_reloc) * MAX_RELOCS);
	}

	r = &bo_fake->relocs[bo_fake->nr_relocs++];

	assert(bo_fake->nr_relocs <= MAX_RELOCS);

	drm_intel_fake_bo_reference_locked(target_bo);

	if (!target_fake->is_static) {
		bo_fake->child_size +=
		    ALIGN(target_bo->size, target_fake->alignment);
		bo_fake->child_size += target_fake->child_size;
	}
	r->target_buf = target_bo;
	r->offset = offset;
	r->last_target_offset = target_bo->offset;
	r->delta = target_offset;
	r->read_domains = read_domains;
	r->write_domain = write_domain;

	if (bufmgr_fake->debug) {
		/* Check that a conflicting relocation hasn't already been
		 * emitted.
		 */
		for (i = 0; i < bo_fake->nr_relocs - 1; i++) {
			struct fake_buffer_reloc *r2 = &bo_fake->relocs[i];

			assert(r->offset != r2->offset);
		}
	}

	pthread_mutex_unlock(&bufmgr_fake->lock);

	return 0;
}

/**
 * Incorporates the validation flags associated with each relocation into
 * the combined validation flags for the buffer on this batchbuffer submission.
 */
static void
drm_intel_fake_calculate_domains(drm_intel_bo *bo)
{
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	int i;

	for (i = 0; i < bo_fake->nr_relocs; i++) {
		struct fake_buffer_reloc *r = &bo_fake->relocs[i];
		drm_intel_bo_fake *target_fake =
		    (drm_intel_bo_fake *) r->target_buf;

		/* Do the same for the tree of buffers we depend on */
		drm_intel_fake_calculate_domains(r->target_buf);

		target_fake->read_domains |= r->read_domains;
		target_fake->write_domain |= r->write_domain;
	}
}

static int
drm_intel_fake_reloc_and_validate_buffer(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	int i, ret;

	assert(bo_fake->map_count == 0);

	for (i = 0; i < bo_fake->nr_relocs; i++) {
		struct fake_buffer_reloc *r = &bo_fake->relocs[i];
		drm_intel_bo_fake *target_fake =
		    (drm_intel_bo_fake *) r->target_buf;
		uint32_t reloc_data;

		/* Validate the target buffer if that hasn't been done. */
		if (!target_fake->validated) {
			ret =
			    drm_intel_fake_reloc_and_validate_buffer(r->target_buf);
			if (ret != 0) {
				if (bo->virtual != NULL)
					drm_intel_fake_bo_unmap_locked(bo);
				return ret;
			}
		}

		/* Calculate the value of the relocation entry. */
		if (r->target_buf->offset != r->last_target_offset) {
			reloc_data = r->target_buf->offset + r->delta;

			if (bo->virtual == NULL)
				drm_intel_fake_bo_map_locked(bo, 1);

			*(uint32_t *) ((uint8_t *) bo->virtual + r->offset) =
			    reloc_data;

			r->last_target_offset = r->target_buf->offset;
		}
	}

	if (bo->virtual != NULL)
		drm_intel_fake_bo_unmap_locked(bo);

	if (bo_fake->write_domain != 0) {
		if (!(bo_fake->flags & (BM_NO_BACKING_STORE | BM_PINNED))) {
			if (bo_fake->backing_store == 0)
				alloc_backing_store(bo);
		}
		bo_fake->card_dirty = 1;
		bufmgr_fake->performed_rendering = 1;
	}

	return drm_intel_fake_bo_validate(bo);
}

static void
drm_intel_bo_fake_post_submit(drm_intel_bo *bo)
{
	drm_intel_bufmgr_fake *bufmgr_fake =
	    (drm_intel_bufmgr_fake *) bo->bufmgr;
	drm_intel_bo_fake *bo_fake = (drm_intel_bo_fake *) bo;
	int i;

	for (i = 0; i < bo_fake->nr_relocs; i++) {
		struct fake_buffer_reloc *r = &bo_fake->relocs[i];
		drm_intel_bo_fake *target_fake =
		    (drm_intel_bo_fake *) r->target_buf;

		if (target_fake->validated)
			drm_intel_bo_fake_post_submit(r->target_buf);

		DBG("%s@0x%08x + 0x%08x -> %s@0x%08x + 0x%08x\n",
		    bo_fake->name, (uint32_t) bo->offset, r->offset,
		    target_fake->name, (uint32_t) r->target_buf->offset,
		    r->delta);
	}

	assert(bo_fake->map_count == 0);
	bo_fake->validated = 0;
	bo_fake->read_domains = 0;
	bo_fake->write_domain = 0;
}

void
drm_intel_bufmgr_fake_set_exec_callback(drm_intel_bufmgr *bufmgr,
					     int (*exec) (drm_intel_bo *bo,
							  unsigned int used,
							  void *priv),
					     void *priv)
{
	drm_intel_bufmgr_fake *bufmgr_fake = (drm_intel_bufmgr_fake *) bufmgr;

	bufmgr_fake->exec = exec;
	bufmgr_fake->exec_priv = priv;
}

static int
drm_intel_fake_bo_exec(drm_intel_bo *bo, int used,
		       drm_clip_rect_t * cliprects, int num_cliprects, int DR4)
{
	drm_intel_bufmgr_fake *bufmgr_fake =