summaryrefslogtreecommitdiff
path: root/shared-core/r300_cmdbuf.c
blob: dc866823323975332eb1347efa5f799034468d5a (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
/* r300_cmdbuf.c -- Command buffer emission for R300 -*- linux-c -*-
 *
 * Copyright (C) The Weather Channel, Inc.  2002.
 * Copyright (C) 2004 Nicolai Haehnle.
 * All Rights Reserved.
 *
 * The Weather Channel (TM) funded Tungsten Graphics to develop the
 * initial release of the Radeon 8500 driver under the XFree86 license.
 * This notice must be preserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * PRECISION INSIGHT 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.
 *
 * Authors:
 *    Nicolai Haehnle <prefect_@gmx.net>
 */

#include "drmP.h"
#include "drm.h"
#include "radeon_drm.h"
#include "radeon_drv.h"
#include "r300_reg.h"

#define R300_SIMULTANEOUS_CLIPRECTS		4

/* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
 */
static const int r300_cliprect_cntl[4] = {
	0xAAAA,
	0xEEEE,
	0xFEFE,
	0xFFFE
};

/**
 * Emit up to R300_SIMULTANEOUS_CLIPRECTS cliprects from the given command
 * buffer, starting with index n.
 */
static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
			       drm_radeon_kcmd_buffer_t *cmdbuf, int n)
{
	drm_clip_rect_t box;
	int nr;
	int i;
	RING_LOCALS;

	nr = cmdbuf->nbox - n;
	if (nr > R300_SIMULTANEOUS_CLIPRECTS)
		nr = R300_SIMULTANEOUS_CLIPRECTS;

	DRM_DEBUG("%i cliprects\n", nr);

	if (nr) {
		BEGIN_RING(6 + nr * 2);
		OUT_RING(CP_PACKET0(R300_RE_CLIPRECT_TL_0, nr * 2 - 1));

		for (i = 0; i < nr; ++i) {
			if (DRM_COPY_FROM_USER_UNCHECKED
			    (&box, &cmdbuf->boxes[n + i], sizeof(box))) {
				DRM_ERROR("copy cliprect faulted\n");
				return DRM_ERR(EFAULT);
			}

			box.x1 =
			    (box.x1 +
			     R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
			box.y1 =
			    (box.y1 +
			     R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
			box.x2 =
			    (box.x2 +
			     R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
			box.y2 =
			    (box.y2 +
			     R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;

			OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
				 (box.y1 << R300_CLIPRECT_Y_SHIFT));
			OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
				 (box.y2 << R300_CLIPRECT_Y_SHIFT));
		}

		OUT_RING_REG(R300_RE_CLIPRECT_CNTL, r300_cliprect_cntl[nr - 1]);

		/* TODO/SECURITY: Force scissors to a safe value, otherwise the
		 * client might be able to trample over memory.
		 * The impact should be very limited, but I'd rather be safe than
		 * sorry.
		 */
		OUT_RING(CP_PACKET0(R300_RE_SCISSORS_TL, 1));
		OUT_RING(0);
		OUT_RING(R300_SCISSORS_X_MASK | R300_SCISSORS_Y_MASK);
		ADVANCE_RING();
	} else {
		/* Why we allow zero cliprect rendering:
		 * There are some commands in a command buffer that must be submitted
		 * even when there are no cliprects, e.g. DMA buffer discard
		 * or state setting (though state setting could be avoided by
		 * simulating a loss of context).
		 *
		 * Now since the cmdbuf interface is so chaotic right now (and is
		 * bound to remain that way for a bit until things settle down),
		 * it is basically impossible to filter out the commands that are
		 * necessary and those that aren't.
		 *
		 * So I choose the safe way and don't do any filtering at all;
		 * instead, I simply set up the engine so that all rendering
		 * can't produce any fragments.
		 */
		BEGIN_RING(2);
		OUT_RING_REG(R300_RE_CLIPRECT_CNTL, 0);
		ADVANCE_RING();
	}

	return 0;
}

static u8 r300_reg_flags[0x10000 >> 2];

void r300_init_reg_flags(void)
{
	int i;
	memset(r300_reg_flags, 0, 0x10000 >> 2);
#define ADD_RANGE_MARK(reg, count,mark) \
		for(i=((reg)>>2);i<((reg)>>2)+(count);i++)\
			r300_reg_flags[i]|=(mark);

#define MARK_SAFE		1
#define MARK_CHECK_OFFSET	2

#define ADD_RANGE(reg, count)	ADD_RANGE_MARK(reg, count, MARK_SAFE)

	/* these match cmducs() command in r300_driver/r300/r300_cmdbuf.c */
	ADD_RANGE(R300_SE_VPORT_XSCALE, 6);
	ADD_RANGE(0x2080, 1);
	ADD_RANGE(R300_SE_VTE_CNTL, 2);
	ADD_RANGE(0x2134, 2);
	ADD_RANGE(0x2140, 1);
	ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
	ADD_RANGE(0x21DC, 1);
	ADD_RANGE(0x221C, 1);
	ADD_RANGE(0x2220, 4);
	ADD_RANGE(0x2288, 1);
	ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
	ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
	ADD_RANGE(R300_GB_ENABLE, 1);
	ADD_RANGE(R300_GB_MSPOS0, 5);
	ADD_RANGE(R300_TX_CNTL, 1);
	ADD_RANGE(R300_TX_ENABLE, 1);
	ADD_RANGE(0x4200, 4);
	ADD_RANGE(0x4214, 1);
	ADD_RANGE(R300_RE_POINTSIZE, 1);
	ADD_RANGE(0x4230, 3);
	ADD_RANGE(R300_RE_LINE_CNT, 1);
	ADD_RANGE(0x4238, 1);
	ADD_RANGE(0x4260, 3);
	ADD_RANGE(0x4274, 4);
	ADD_RANGE(0x4288, 5);
	ADD_RANGE(0x42A0, 1);
	ADD_RANGE(R300_RE_ZBIAS_T_FACTOR, 4);
	ADD_RANGE(0x42B4, 1);
	ADD_RANGE(R300_RE_CULL_CNTL, 1);
	ADD_RANGE(0x42C0, 2);
	ADD_RANGE(R300_RS_CNTL_0, 2);
	ADD_RANGE(R300_RS_INTERP_0, 8);
	ADD_RANGE(R300_RS_ROUTE_0, 8);
	ADD_RANGE(0x43A4, 2);
	ADD_RANGE(0x43E8, 1);
	ADD_RANGE(R300_PFS_CNTL_0, 3);
	ADD_RANGE(R300_PFS_NODE_0, 4);
	ADD_RANGE(R300_PFS_TEXI_0, 64);
	ADD_RANGE(0x46A4, 5);
	ADD_RANGE(R300_PFS_INSTR0_0, 64);
	ADD_RANGE(R300_PFS_INSTR1_0, 64);
	ADD_RANGE(R300_PFS_INSTR2_0, 64);
	ADD_RANGE(R300_PFS_INSTR3_0, 64);
	ADD_RANGE(0x4BC0, 1);
	ADD_RANGE(0x4BC8, 3);
	ADD_RANGE(R300_PP_ALPHA_TEST, 2);
	ADD_RANGE(0x4BD8, 1);
	ADD_RANGE(R300_PFS_PARAM_0_X, 64);
	ADD_RANGE(0x4E00, 1);
	ADD_RANGE(R300_RB3D_CBLEND, 2);
	ADD_RANGE(R300_RB3D_COLORMASK, 1);
	ADD_RANGE(0x4E10, 3);
	ADD_RANGE_MARK(R300_RB3D_COLOROFFSET0, 1, MARK_CHECK_OFFSET);	/* check offset */
	ADD_RANGE(R300_RB3D_COLORPITCH0, 1);
	ADD_RANGE(0x4E50, 9);
	ADD_RANGE(0x4E88, 1);
	ADD_RANGE(0x4EA0, 2);
	ADD_RANGE(R300_RB3D_ZSTENCIL_CNTL_0, 3);
	ADD_RANGE(0x4F10, 4);
	ADD_RANGE_MARK(R300_RB3D_DEPTHOFFSET, 1, MARK_CHECK_OFFSET);	/* check offset */
	ADD_RANGE(R300_RB3D_DEPTHPITCH, 1);
	ADD_RANGE(0x4F28, 1);
	ADD_RANGE(0x4F30, 2);
	ADD_RANGE(0x4F44, 1);
	ADD_RANGE(0x4F54, 1);

	ADD_RANGE(R300_TX_FILTER_0, 16);
	ADD_RANGE(R300_TX_FILTER1_0, 16);
	ADD_RANGE(R300_TX_SIZE_0, 16);
	ADD_RANGE(R300_TX_FORMAT_0, 16);
	ADD_RANGE(R300_TX_PITCH_0, 16);
	/* Texture offset is dangerous and needs more checking */
	ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET);
	ADD_RANGE(R300_TX_CHROMA_KEY_0, 16);
	ADD_RANGE(R300_TX_BORDER_COLOR_0, 16);

	/* Sporadic registers used as primitives are emitted */
	ADD_RANGE(0x4f18, 1);
	ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1);
	ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8);
	ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8);

}

static __inline__ int r300_check_range(unsigned reg, int count)
{
	int i;
	if (reg & ~0xffff)
		return -1;
	for (i = (reg >> 2); i < (reg >> 2) + count; i++)
		if (r300_reg_flags[i] != MARK_SAFE)
			return 1;
	return 0;
}

/*
 * we expect offsets passed to the framebuffer to be either within video 
 * memory or within AGP space 
 */
static __inline__ int r300_check_offset(drm_radeon_private_t *dev_priv,
					u32 offset)
{
	/* we realy want to check against end of video aperture
	   but this value is not being kept.
	   This code is correct for now (does the same thing as the
	   code that sets MC_FB_LOCATION) in radeon_cp.c */
	if (offset >= dev_priv->fb_location &&
	    offset < (dev_priv->fb_location + dev_priv->fb_size))
		return 0;
	if (offset >= dev_priv->gart_vm_start &&
	    offset < (dev_priv->gart_vm_start + dev_priv->gart_size))
		return 0;
	return 1;
}

static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t *
							  dev_priv,
							  drm_radeon_kcmd_buffer_t
							  * cmdbuf,
							  drm_r300_cmd_header_t
							  header)
{
	int reg;
	int sz;
	int i;
	int values[64];
	RING_LOCALS;

	sz = header.packet0.count;
	reg = (header.packet0.reghi << 8) | header.packet0.reglo;

	if ((sz > 64) || (sz < 0)) {
		DRM_ERROR
		    ("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n",
		     reg, sz);
		return DRM_ERR(EINVAL);
	}
	for (i = 0; i < sz; i++) {
		values[i] = ((int *)cmdbuf->buf)[i];
		switch (r300_reg_flags[(reg >> 2) + i]) {
		case MARK_SAFE:
			break;
		case MARK_CHECK_OFFSET:
			if (r300_check_offset(dev_priv, (u32) values[i])) {
				DRM_ERROR
				    ("Offset failed range check (reg=%04x sz=%d)\n",
				     reg, sz);
				return DRM_ERR(EINVAL);
			}
			break;
		default:
			DRM_ERROR("Register %04x failed check as flag=%02x\n",
				  reg + i * 4, r300_reg_flags[(reg >> 2) + i]);
			return DRM_ERR(EINVAL);
		}
	}

	BEGIN_RING(1 + sz);
	OUT_RING(CP_PACKET0(reg, sz - 1));
	OUT_RING_TABLE(values, sz);
	ADVANCE_RING();

	cmdbuf->buf += sz * 4;
	cmdbuf->bufsz -= sz * 4;

	return 0;
}

/**
 * Emits a packet0 setting arbitrary registers.
 * Called by r300_do_cp_cmdbuf.
 *
 * Note that checks are performed on contents and addresses of the registers
 */
static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv,
					drm_radeon_kcmd_buffer_t *cmdbuf,
					drm_r300_cmd_header_t header)
{
	int reg;
	int sz;
	RING_LOCALS;

	sz = header.packet0.count;
	reg = (header.packet0.reghi << 8) | header.packet0.reglo;

	if (!sz)
		return 0;

	if (sz * 4 > cmdbuf->bufsz)
		return DRM_ERR(EINVAL);

	if (reg + sz * 4 >= 0x10000) {
		DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg,
			  sz);
		return DRM_ERR(EINVAL);
	}

	if (r300_check_range(reg, sz)) {
		/* go and check everything */
		return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf,
							   header);
	}
	/* the rest of the data is safe to emit, whatever the values the user passed */

	BEGIN_RING(1 + sz);
	OUT_RING(CP_PACKET0(reg, sz - 1));
	OUT_RING_TABLE((int *)cmdbuf->buf, sz);
	ADVANCE_RING();

	cmdbuf->buf += sz * 4;
	cmdbuf->bufsz -= sz * 4;

	return 0;
}

/**
 * Uploads user-supplied vertex program instructions or parameters onto
 * the graphics card.
 * Called by r300_do_cp_cmdbuf.
 */
static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
				    drm_radeon_kcmd_buffer_t *cmdbuf,
				    drm_r300_cmd_header_t header)
{
	int sz;
	int addr;
	RING_LOCALS;

	sz = header.vpu.count;
	addr = (header.vpu.adrhi << 8) | header.vpu.adrlo;

	if (!sz)
		return 0;
	if (sz * 16 > cmdbuf->bufsz)
		return DRM_ERR(EINVAL);

	BEGIN_RING(5 + sz * 4);
	/* Wait for VAP to come to senses.. */
	/* there is no need to emit it multiple times, (only once before VAP is programmed,
	   but this optimization is for later */
	OUT_RING_REG(R300_VAP_PVS_WAITIDLE, 0);
	OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
	OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
	OUT_RING_TABLE((int *)cmdbuf->buf, sz * 4);

	ADVANCE_RING();

	cmdbuf->buf += sz * 16;
	cmdbuf->bufsz -= sz * 16;

	return 0;
}

/**
 * Emit a clear packet from userspace.
 * Called by r300_emit_packet3.
 */
static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
				      drm_radeon_kcmd_buffer_t *cmdbuf)
{
	RING_LOCALS;

	if (8 * 4 > cmdbuf->bufsz)
		return DRM_ERR(EINVAL);

	BEGIN_RING(10);
	OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8));
	OUT_RING(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
		 (1 << R300_PRIM_NUM_VERTICES_SHIFT));
	OUT_RING_TABLE((int *)cmdbuf->buf, 8);
	ADVANCE_RING();

	cmdbuf->buf += 8 * 4;
	cmdbuf->bufsz -= 8 * 4;

	return 0;
}

static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
					       drm_radeon_kcmd_buffer_t *cmdbuf,
					       u32 header)
{
	int count, i, k;
#define MAX_ARRAY_PACKET  64
	u32 payload[MAX_ARRAY_PACKET];
	u32 narrays;
	RING_LOCALS;

	count = (header >> 16) & 0x3fff;

	if ((count + 1) > MAX_ARRAY_PACKET) {
		DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
			  count);
		return DRM_ERR(EINVAL);
	}
	memset(payload, 0, MAX_ARRAY_PACKET * 4);
	memcpy(payload, cmdbuf->buf + 4, (count + 1) * 4);

	/* carefully check packet contents */

	narrays = payload[0];
	k = 0;
	i = 1;
	while ((k < narrays) && (i < (count + 1))) {
		i++;		/* skip attribute field */
		if (r300_check_offset(dev_priv, payload[i])) {
			DRM_ERROR
			    ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
			     k, i);
			return DRM_ERR(EINVAL);
		}
		k++;
		i++;
		if (k == narrays)
			break;
		/* have one more to process, they come in pairs */
		if (r300_check_offset(dev_priv, payload[i])) {
			DRM_ERROR
			    ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
			     k, i);
			return DRM_ERR(EINVAL);
		}
		k++;
		i++;
	}
	/* do the counts match what we expect ? */
	if ((k != narrays) || (i != (count + 1))) {
		DRM_ERROR
		    ("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n",
		     k, i, narrays, count + 1);
		return DRM_ERR(EINVAL);
	}

	/* all clear, output packet */

	BEGIN_RING(count + 2);
	OUT_RING(header);
	OUT_RING_TABLE(payload, count + 1);
	ADVANCE_RING();

	cmdbuf->buf += (count + 2) * 4;
	cmdbuf->bufsz -= (count + 2) * 4;

	return 0;
}

static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
					     drm_radeon_kcmd_buffer_t *cmdbuf)
{
	u32 *cmd = (u32 *) cmdbuf->buf;
	int count, ret;
	RING_LOCALS;

	count=(cmd[0]>>16) & 0x3fff;

	if (cmd[0] & 0x8000) {
		u32 offset;

		if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL 
			      | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
			offset = cmd[2] << 10;
			ret = r300_check_offset(dev_priv, offset);
			if (ret) {
				DRM_ERROR("Invalid bitblt first offset is %08X\n", offset);
				return DRM_ERR(EINVAL);
			}
		}

		if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
		    (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
			offset = cmd[3] << 10;
			ret = r300_check_offset(dev_priv, offset);
			if (ret) {
				DRM_ERROR("Invalid bitblt second offset is %08X\n", offset);
				return DRM_ERR(EINVAL);
			}
			
		}
	}

	BEGIN_RING(count+2);
	OUT_RING(cmd[0]);
	OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
	ADVANCE_RING();

	cmdbuf->buf += (count+2)*4;
	cmdbuf->bufsz -= (count+2)*4;

	return 0;
}

static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
					    drm_radeon_kcmd_buffer_t *cmdbuf)
{
	u32 header;
	int count;
	RING_LOCALS;

	if (4 > cmdbuf->bufsz)
		return DRM_ERR(EINVAL);

	/* Fixme !! This simply emits a packet without much checking.
	   We need to be smarter. */

	/* obtain first word - actual packet3 header */
	header = *(u32 *) cmdbuf->buf;

	/* Is it packet 3 ? */
	if ((header >> 30) != 0x3) {
		DRM_ERROR("Not a packet3 header (0x%08x)\n", header);
		return DRM_ERR(EINVAL);
	}

	count = (header >> 16) & 0x3fff;

	/* Check again now that we know how much data to expect */
	if ((count + 2) * 4 > cmdbuf->bufsz) {
		DRM_ERROR
		    ("Expected packet3 of length %d but have only %d bytes left\n",
		     (count + 2) * 4, cmdbuf->bufsz);
		return DRM_ERR(EINVAL);
	}

	/* Is it a packet type we know about ? */
	switch (header & 0xff00) {
	case RADEON_3D_LOAD_VBPNTR:	/* load vertex array pointers */
		return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, header);

	case RADEON_CNTL_BITBLT_MULTI:
		return r300_emit_bitblt_multi(dev_priv, cmdbuf);

	case RADEON_CP_3D_DRAW_IMMD_2:	/* triggers drawing using in-packet vertex data */
	case RADEON_CP_3D_DRAW_VBUF_2:	/* triggers drawing of vertex buffers setup elsewhere */
	case RADEON_CP_3D_DRAW_INDX_2:	/* triggers drawing using indices to vertex buffer */
	case RADEON_CP_INDX_BUFFER:	/* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
	case RADEON_WAIT_FOR_IDLE:
	case RADEON_CP_NOP:
		/* these packets are safe */
		break;
	default:
		DRM_ERROR("Unknown packet3 header (0x%08x)\n", header);
		return DRM_ERR(EINVAL);
	}

	BEGIN_RING(count + 2);
	OUT_RING(header);
	OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
	ADVANCE_RING();

	cmdbuf->buf += (count + 2) * 4;
	cmdbuf->bufsz -= (count + 2) * 4;

	return 0;
}

/**
 * Emit a rendering packet3 from userspace.
 * Called by r300_do_cp_cmdbuf.
 */
static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
					drm_radeon_kcmd_buffer_t *cmdbuf,
					drm_r300_cmd_header_t header)
{
	int n;
	int ret;
	char *orig_buf = cmdbuf->buf;
	int orig_bufsz = cmdbuf->bufsz;

	/* This is a do-while-loop so that we run the interior at least once,
	 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
	 */
	n = 0;
	do {
		if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) {
			ret = r300_emit_cliprects(dev_priv, cmdbuf, n);
			if (ret)
				return ret;

			cmdbuf->buf = orig_buf;
			cmdbuf->bufsz = orig_bufsz;
		}

		switch (header.packet3.packet) {
		case R300_CMD_PACKET3_CLEAR:
			DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n");
			ret = r300_emit_clear(dev_priv, cmdbuf);
			if (ret) {
				DRM_ERROR("r300_emit_clear failed\n");
				return ret;
			}
			break;

		case R300_CMD_PACKET3_RAW:
			DRM_DEBUG("R300_CMD_PACKET3_RAW\n");
			ret = r300_emit_raw_packet3(dev_priv, cmdbuf);
			if (ret) {
				DRM_ERROR("r300_emit_raw_packet3 failed\n");
				return ret;
			}
			break;

		default:
			DRM_ERROR("bad packet3 type %i at %p\n",
				  header.packet3.packet,
				  cmdbuf->buf - sizeof(header));
			return DRM_ERR(EINVAL);
		}

		n += R300_SIMULTANEOUS_CLIPRECTS;
	} while (n < cmdbuf->nbox);

	return 0;
}

/* Some of the R300 chips seem to be extremely touchy about the two registers
 * that are configured in r300_pacify.
 * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace
 * sends a command buffer that contains only state setting commands and a
 * vertex program/parameter upload sequence, this will eventually lead to a
 * lockup, unless the sequence is bracketed by calls to r300_pacify.
 * So we should take great care to *always* call r300_pacify before
 * *anything* 3D related, and again afterwards. This is what the
 * call bracket in r300_do_cp_cmdbuf is for.
 */

/**
 * Emit the sequence to pacify R300.
 */
static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
{
	RING_LOCALS;

	BEGIN_RING(6);
	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
	OUT_RING(0xa);
	OUT_RING(CP_PACKET0(0x4f18, 0));
	OUT_RING(0x3);
	OUT_RING(CP_PACKET3(RADEON_CP_NOP, 0));
	OUT_RING(0x0);
	ADVANCE_RING();
}

/**
 * Called by r300_do_cp_cmdbuf to update the internal buffer age and state.
 * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
 * be careful about how this function is called.
 */
static void r300_discard_buffer(drm_device_t * dev, drm_buf_t * buf)
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;

	buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
	buf->pending = 1;
	buf->used = 0;
}

static int r300_scratch(drm_radeon_private_t *dev_priv,
			drm_radeon_kcmd_buffer_t *cmdbuf,
			drm_r300_cmd_header_t header)
{
	u32 *ref_age_base;
	u32 i, buf_idx, h_pending;
	RING_LOCALS;
	
	if (cmdbuf->bufsz < sizeof(uint64_t) + header.scratch.n_bufs * sizeof(buf_idx) ) {
		return DRM_ERR(EINVAL);
	}
	
	if (header.scratch.reg >= 5) {
		return DRM_ERR(EINVAL);
	}
	
	dev_priv->scratch_ages[header.scratch.reg] ++;
	
	ref_age_base = (u32 *)(unsigned long)*((uint64_t *)cmdbuf->buf);
	
	cmdbuf->buf += sizeof(uint64_t);
	cmdbuf->bufsz -= sizeof(uint64_t);
	
	for (i=0; i < header.scratch.n_bufs; i++) {
		buf_idx = *(u32 *)cmdbuf->buf;
		buf_idx *= 2; /* 8 bytes per buf */
		
		if (DRM_COPY_TO_USER(ref_age_base + buf_idx, &dev_priv->scratch_ages[header.scratch.reg], sizeof(u32))) {
			return DRM_ERR(EINVAL);
		}
					
		if (DRM_COPY_FROM_USER(&h_pending, ref_age_base + buf_idx + 1, sizeof(u32))) {
			return DRM_ERR(EINVAL);
		}
					
		if (h_pending == 0) {
			return DRM_ERR(EINVAL);
		}
					
		h_pending--;
						
		if (DRM_COPY_TO_USER(ref_age_base + buf_idx + 1, &h_pending, sizeof(u32))) {
			return DRM_ERR(EINVAL);
		}
					
		cmdbuf->buf += sizeof(buf_idx);
		cmdbuf->bufsz -= sizeof(buf_idx);
	}
	
	BEGIN_RING(2);
	OUT_RING( CP_PACKET0( RADEON_SCRATCH_REG0 + header.scratch.reg * 4, 0 ) );
	OUT_RING( dev_priv->scratch_ages[header.scratch.reg] );
	ADVANCE_RING();
	
	return 0;
}

/**
 * Parses and validates a user-supplied command buffer and emits appropriate
 * commands on the DMA ring buffer.
 * Called by the ioctl handler function radeon_cp_cmdbuf.
 */
int r300_do_cp_cmdbuf(drm_device_t *dev,
		      DRMFILE filp,
		      drm_file_t *filp_priv,
		      drm_radeon_kcmd_buffer_t *cmdbuf)
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_device_dma_t *dma = dev->dma;
	drm_buf_t *buf = NULL;
	int emit_dispatch_age = 0;
	int ret = 0;

	DRM_DEBUG("\n");

	/* See the comment above r300_emit_begin3d for why this call must be here,
	 * and what the cleanup gotos are for. */
	r300_pacify(dev_priv);

	if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
		ret = r300_emit_cliprects(dev_priv, cmdbuf, 0);
		if (ret)
			goto cleanup;
	}

	while (cmdbuf->bufsz >= sizeof(drm_r300_cmd_header_t)) {
		int idx;
		drm_r300_cmd_header_t header;

		header.u = *(unsigned int *)cmdbuf->buf;

		cmdbuf->buf += sizeof(header);
		cmdbuf->bufsz -= sizeof(header);

		switch (header.header.cmd_type) {
		case R300_CMD_PACKET0:
			DRM_DEBUG("R300_CMD_PACKET0\n");
			ret = r300_emit_packet0(dev_priv, cmdbuf, header);
			if (ret) {
				DRM_ERROR("r300_emit_packet0 failed\n");
				goto cleanup;
			}
			break;

		case R300_CMD_VPU:
			DRM_DEBUG("R300_CMD_VPU\n");
			ret = r300_emit_vpu(dev_priv, cmdbuf, header);
			if (ret) {
				DRM_ERROR("r300_emit_vpu failed\n");
				goto cleanup;
			}
			break;

		case R300_CMD_PACKET3:
			DRM_DEBUG("R300_CMD_PACKET3\n");
			ret = r300_emit_packet3(dev_priv, cmdbuf, header);
			if (ret) {
				DRM_ERROR("r300_emit_packet3 failed\n");
				goto cleanup;
			}
			break;

		case R300_CMD_END3D:
			DRM_DEBUG("R300_CMD_END3D\n");
			/* TODO:
			   Ideally userspace driver should not need to issue this call,
			   i.e. the drm driver should issue it automatically and prevent
			   lockups.

			   In practice, we do not understand why this call is needed and what
			   it does (except for some vague guesses that it has to do with cache
			   coherence) and so the user space driver does it.

			   Once we are sure which uses prevent lockups the code could be moved
			   into the kernel and the userspace driver will not
			   need to use this command.

			   Note that issuing this command does not hurt anything
			   except, possibly, performance */
			r300_pacify(dev_priv);
			break;

		case R300_CMD_CP_DELAY:
			/* simple enough, we can do it here */
			DRM_DEBUG("R300_CMD_CP_DELAY\n");
			{
				int i;
				RING_LOCALS;

				BEGIN_RING(header.delay.count);
				for (i = 0; i < header.delay.count; i++)
					OUT_RING(RADEON_CP_PACKET2);
				ADVANCE_RING();
			}
			break;

		case R300_CMD_DMA_DISCARD:
			DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
			idx = header.dma.buf_idx;
			if (idx < 0 || idx >= dma->buf_count) {
				DRM_ERROR("buffer index %d (of %d max)\n",
					  idx, dma->buf_count - 1);
				ret = DRM_ERR(EINVAL);
				goto cleanup;
			}

			buf = dma->buflist[idx];
			if (buf->filp != filp || buf->pending) {
				DRM_ERROR("bad buffer %p %p %d\n",
					  buf->filp, filp, buf->pending);
				ret = DRM_ERR(EINVAL);
				goto cleanup;
			}

			emit_dispatch_age = 1;
			r300_discard_buffer(dev, buf);
			break;

		case R300_CMD_WAIT:
			/* simple enough, we can do it here */
			DRM_DEBUG("R300_CMD_WAIT\n");
			if (header.wait.flags == 0)
				break;	/* nothing to do */

			{
				RING_LOCALS;

				BEGIN_RING(2);
				OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
				OUT_RING((header.wait.flags & 0xf) << 14);
				ADVANCE_RING();
			}
			break;

		case R300_CMD_SCRATCH:
			DRM_DEBUG("R300_CMD_SCRATCH\n");
			ret = r300_scratch(dev_priv, cmdbuf, header);
			if (ret) {
				DRM_ERROR("r300_scratch failed\n");
				goto cleanup;
			}
			break;
			
		default:
			DRM_ERROR("bad cmd_type %i at %p\n",
				  header.header.cmd_type,
				  cmdbuf->buf - sizeof(header));
			ret = DRM_ERR(EINVAL);
			goto cleanup;
		}
	}

	DRM_DEBUG("END\n");

      cleanup:
	r300_pacify(dev_priv);

	/* We emit the vertex buffer age here, outside the pacifier "brackets"
	 * for two reasons:
	 *  (1) This may coalesce multiple age emissions into a single one and
	 *  (2) more importantly, some chips lock up hard when scratch registers
	 *      are written inside the pacifier bracket.
	 */
	if (emit_dispatch_age) {
		RING_LOCALS;

		/* Emit the vertex buffer age */
		BEGIN_RING(2);
		RADEON_DISPATCH_AGE(dev_priv->sarea_priv->last_dispatch);
		ADVANCE_RING();
	}

	COMMIT_RING();

	return ret;
}