projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
drm/radeon: add hier-z registers for r300 and r500 chipsets
[deliverable/linux.git] / drivers / char / drm / r300_cmdbuf.c
1 /* r300_cmdbuf.c -- Command buffer emission for R300 -*- linux-c -*-
2 *
3 * Copyright (C) The Weather Channel, Inc. 2002.
4 * Copyright (C) 2004 Nicolai Haehnle.
5 * All Rights Reserved.
6 *
7 * The Weather Channel (TM) funded Tungsten Graphics to develop the
8 * initial release of the Radeon 8500 driver under the XFree86 license.
9 * This notice must be preserved.
10 *
11 * Permission is hereby granted, free of charge, to any person obtaining a
12 * copy of this software and associated documentation files (the "Software"),
13 * to deal in the Software without restriction, including without limitation
14 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
15 * and/or sell copies of the Software, and to permit persons to whom the
16 * Software is furnished to do so, subject to the following conditions:
17 *
18 * The above copyright notice and this permission notice (including the next
19 * paragraph) shall be included in all copies or substantial portions of the
20 * Software.
21 *
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
25 * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
26 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
27 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
28 * DEALINGS IN THE SOFTWARE.
29 *
30 * Authors:
31 * Nicolai Haehnle <prefect_@gmx.net>
32 */
33
34 #include "drmP.h"
35 #include "drm.h"
36 #include "radeon_drm.h"
37 #include "radeon_drv.h"
38 #include "r300_reg.h"
39
40 #define R300_SIMULTANEOUS_CLIPRECTS 4
41
42 /* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
43 */
44 static const int r300_cliprect_cntl[4] = {
45 0xAAAA,
46 0xEEEE,
47 0xFEFE,
48 0xFFFE
49 };
50
51 /**
52 * Emit up to R300_SIMULTANEOUS_CLIPRECTS cliprects from the given command
53 * buffer, starting with index n.
54 */
55 static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
56 drm_radeon_kcmd_buffer_t *cmdbuf, int n)
57 {
58 struct drm_clip_rect box;
59 int nr;
60 int i;
61 RING_LOCALS;
62
63 nr = cmdbuf->nbox - n;
64 if (nr > R300_SIMULTANEOUS_CLIPRECTS)
65 nr = R300_SIMULTANEOUS_CLIPRECTS;
66
67 DRM_DEBUG("%i cliprects\n", nr);
68
69 if (nr) {
70 BEGIN_RING(6 + nr * 2);
71 OUT_RING(CP_PACKET0(R300_RE_CLIPRECT_TL_0, nr * 2 - 1));
72
73 for (i = 0; i < nr; ++i) {
74 if (DRM_COPY_FROM_USER_UNCHECKED
75 (&box, &cmdbuf->boxes[n + i], sizeof(box))) {
76 DRM_ERROR("copy cliprect faulted\n");
77 return -EFAULT;
78 }
79
80 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
81 box.x1 = (box.x1) &
82 R300_CLIPRECT_MASK;
83 box.y1 = (box.y1) &
84 R300_CLIPRECT_MASK;
85 box.x2 = (box.x2) &
86 R300_CLIPRECT_MASK;
87 box.y2 = (box.y2) &
88 R300_CLIPRECT_MASK;
89 } else {
90 box.x1 = (box.x1 + R300_CLIPRECT_OFFSET) &
91 R300_CLIPRECT_MASK;
92 box.y1 = (box.y1 + R300_CLIPRECT_OFFSET) &
93 R300_CLIPRECT_MASK;
94 box.x2 = (box.x2 + R300_CLIPRECT_OFFSET) &
95 R300_CLIPRECT_MASK;
96 box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) &
97 R300_CLIPRECT_MASK;
98
99 }
100 OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
101 (box.y1 << R300_CLIPRECT_Y_SHIFT));
102 OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
103 (box.y2 << R300_CLIPRECT_Y_SHIFT));
104
105 }
106
107 OUT_RING_REG(R300_RE_CLIPRECT_CNTL, r300_cliprect_cntl[nr - 1]);
108
109 /* TODO/SECURITY: Force scissors to a safe value, otherwise the
110 * client might be able to trample over memory.
111 * The impact should be very limited, but I'd rather be safe than
112 * sorry.
113 */
114 OUT_RING(CP_PACKET0(R300_RE_SCISSORS_TL, 1));
115 OUT_RING(0);
116 OUT_RING(R300_SCISSORS_X_MASK | R300_SCISSORS_Y_MASK);
117 ADVANCE_RING();
118 } else {
119 /* Why we allow zero cliprect rendering:
120 * There are some commands in a command buffer that must be submitted
121 * even when there are no cliprects, e.g. DMA buffer discard
122 * or state setting (though state setting could be avoided by
123 * simulating a loss of context).
124 *
125 * Now since the cmdbuf interface is so chaotic right now (and is
126 * bound to remain that way for a bit until things settle down),
127 * it is basically impossible to filter out the commands that are
128 * necessary and those that aren't.
129 *
130 * So I choose the safe way and don't do any filtering at all;
131 * instead, I simply set up the engine so that all rendering
132 * can't produce any fragments.
133 */
134 BEGIN_RING(2);
135 OUT_RING_REG(R300_RE_CLIPRECT_CNTL, 0);
136 ADVANCE_RING();
137 }
138
139 return 0;
140 }
141
142 static u8 r300_reg_flags[0x10000 >> 2];
143
144 void r300_init_reg_flags(struct drm_device *dev)
145 {
146 int i;
147 drm_radeon_private_t *dev_priv = dev->dev_private;
148
149 memset(r300_reg_flags, 0, 0x10000 >> 2);
150 #define ADD_RANGE_MARK(reg, count,mark) \
151 for(i=((reg)>>2);i<((reg)>>2)+(count);i++)\
152 r300_reg_flags[i]|=(mark);
153
154 #define MARK_SAFE 1
155 #define MARK_CHECK_OFFSET 2
156
157 #define ADD_RANGE(reg, count) ADD_RANGE_MARK(reg, count, MARK_SAFE)
158
159 /* these match cmducs() command in r300_driver/r300/r300_cmdbuf.c */
160 ADD_RANGE(R300_SE_VPORT_XSCALE, 6);
161 ADD_RANGE(R300_VAP_CNTL, 1);
162 ADD_RANGE(R300_SE_VTE_CNTL, 2);
163 ADD_RANGE(0x2134, 2);
164 ADD_RANGE(R300_VAP_CNTL_STATUS, 1);
165 ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
166 ADD_RANGE(0x21DC, 1);
167 ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
168 ADD_RANGE(R300_VAP_CLIP_X_0, 4);
169 ADD_RANGE(R300_VAP_PVS_WAITIDLE, 1);
170 ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
171 ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
172 ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
173 ADD_RANGE(R300_GB_ENABLE, 1);
174 ADD_RANGE(R300_GB_MSPOS0, 5);
175 ADD_RANGE(R300_TX_CNTL, 1);
176 ADD_RANGE(R300_TX_ENABLE, 1);
177 ADD_RANGE(0x4200, 4);
178 ADD_RANGE(0x4214, 1);
179 ADD_RANGE(R300_RE_POINTSIZE, 1);
180 ADD_RANGE(0x4230, 3);
181 ADD_RANGE(R300_RE_LINE_CNT, 1);
182 ADD_RANGE(R300_RE_UNK4238, 1);
183 ADD_RANGE(0x4260, 3);
184 ADD_RANGE(R300_RE_SHADE, 4);
185 ADD_RANGE(R300_RE_POLYGON_MODE, 5);
186 ADD_RANGE(R300_RE_ZBIAS_CNTL, 1);
187 ADD_RANGE(R300_RE_ZBIAS_T_FACTOR, 4);
188 ADD_RANGE(R300_RE_OCCLUSION_CNTL, 1);
189 ADD_RANGE(R300_RE_CULL_CNTL, 1);
190 ADD_RANGE(0x42C0, 2);
191 ADD_RANGE(R300_RS_CNTL_0, 2);
192
193 ADD_RANGE(R300_SC_HYPERZ, 2);
194 ADD_RANGE(0x43E8, 1);
195
196 ADD_RANGE(0x46A4, 5);
197
198 ADD_RANGE(R300_RE_FOG_STATE, 1);
199 ADD_RANGE(R300_FOG_COLOR_R, 3);
200 ADD_RANGE(R300_PP_ALPHA_TEST, 2);
201 ADD_RANGE(0x4BD8, 1);
202 ADD_RANGE(R300_PFS_PARAM_0_X, 64);
203 ADD_RANGE(0x4E00, 1);
204 ADD_RANGE(R300_RB3D_CBLEND, 2);
205 ADD_RANGE(R300_RB3D_COLORMASK, 1);
206 ADD_RANGE(R300_RB3D_BLEND_COLOR, 3);
207 ADD_RANGE_MARK(R300_RB3D_COLOROFFSET0, 1, MARK_CHECK_OFFSET); /* check offset */
208 ADD_RANGE(R300_RB3D_COLORPITCH0, 1);
209 ADD_RANGE(0x4E50, 9);
210 ADD_RANGE(0x4E88, 1);
211 ADD_RANGE(0x4EA0, 2);
212 ADD_RANGE(R300_ZB_CNTL, 3);
213 ADD_RANGE(R300_ZB_FORMAT, 4);
214 ADD_RANGE_MARK(R300_ZB_DEPTHOFFSET, 1, MARK_CHECK_OFFSET); /* check offset */
215 ADD_RANGE(R300_ZB_DEPTHPITCH, 1);
216 ADD_RANGE(R300_ZB_DEPTHCLEARVALUE, 1);
217 ADD_RANGE(R300_ZB_ZMASK_OFFSET, 13);
218
219 ADD_RANGE(R300_TX_FILTER_0, 16);
220 ADD_RANGE(R300_TX_FILTER1_0, 16);
221 ADD_RANGE(R300_TX_SIZE_0, 16);
222 ADD_RANGE(R300_TX_FORMAT_0, 16);
223 ADD_RANGE(R300_TX_PITCH_0, 16);
224 /* Texture offset is dangerous and needs more checking */
225 ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET);
226 ADD_RANGE(R300_TX_CHROMA_KEY_0, 16);
227 ADD_RANGE(R300_TX_BORDER_COLOR_0, 16);
228
229 /* Sporadic registers used as primitives are emitted */
230 ADD_RANGE(R300_ZB_ZCACHE_CTLSTAT, 1);
231 ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1);
232 ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8);
233 ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8);
234
235 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
236 ADD_RANGE(R500_VAP_INDEX_OFFSET, 1);
237 ADD_RANGE(R500_US_CONFIG, 2);
238 ADD_RANGE(R500_US_CODE_ADDR, 3);
239 ADD_RANGE(R500_US_FC_CTRL, 1);
240 ADD_RANGE(R500_RS_IP_0, 16);
241 ADD_RANGE(R500_RS_INST_0, 16);
242 ADD_RANGE(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2);
243 ADD_RANGE(R500_RB3D_CONSTANT_COLOR_AR, 2);
244 ADD_RANGE(R500_ZB_FIFO_SIZE, 2);
245 } else {
246 ADD_RANGE(R300_PFS_CNTL_0, 3);
247 ADD_RANGE(R300_PFS_NODE_0, 4);
248 ADD_RANGE(R300_PFS_TEXI_0, 64);
249 ADD_RANGE(R300_PFS_INSTR0_0, 64);
250 ADD_RANGE(R300_PFS_INSTR1_0, 64);
251 ADD_RANGE(R300_PFS_INSTR2_0, 64);
252 ADD_RANGE(R300_PFS_INSTR3_0, 64);
253 ADD_RANGE(R300_RS_INTERP_0, 8);
254 ADD_RANGE(R300_RS_ROUTE_0, 8);
255
256 }
257 }
258
259 static __inline__ int r300_check_range(unsigned reg, int count)
260 {
261 int i;
262 if (reg & ~0xffff)
263 return -1;
264 for (i = (reg >> 2); i < (reg >> 2) + count; i++)
265 if (r300_reg_flags[i] != MARK_SAFE)
266 return 1;
267 return 0;
268 }
269
270 static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t *
271 dev_priv,
272 drm_radeon_kcmd_buffer_t
273 * cmdbuf,
274 drm_r300_cmd_header_t
275 header)
276 {
277 int reg;
278 int sz;
279 int i;
280 int values[64];
281 RING_LOCALS;
282
283 sz = header.packet0.count;
284 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
285
286 if ((sz > 64) || (sz < 0)) {
287 DRM_ERROR
288 ("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n",
289 reg, sz);
290 return -EINVAL;
291 }
292 for (i = 0; i < sz; i++) {
293 values[i] = ((int *)cmdbuf->buf)[i];
294 switch (r300_reg_flags[(reg >> 2) + i]) {
295 case MARK_SAFE:
296 break;
297 case MARK_CHECK_OFFSET:
298 if (!radeon_check_offset(dev_priv, (u32) values[i])) {
299 DRM_ERROR
300 ("Offset failed range check (reg=%04x sz=%d)\n",
301 reg, sz);
302 return -EINVAL;
303 }
304 break;
305 default:
306 DRM_ERROR("Register %04x failed check as flag=%02x\n",
307 reg + i * 4, r300_reg_flags[(reg >> 2) + i]);
308 return -EINVAL;
309 }
310 }
311
312 BEGIN_RING(1 + sz);
313 OUT_RING(CP_PACKET0(reg, sz - 1));
314 OUT_RING_TABLE(values, sz);
315 ADVANCE_RING();
316
317 cmdbuf->buf += sz * 4;
318 cmdbuf->bufsz -= sz * 4;
319
320 return 0;
321 }
322
323 /**
324 * Emits a packet0 setting arbitrary registers.
325 * Called by r300_do_cp_cmdbuf.
326 *
327 * Note that checks are performed on contents and addresses of the registers
328 */
329 static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv,
330 drm_radeon_kcmd_buffer_t *cmdbuf,
331 drm_r300_cmd_header_t header)
332 {
333 int reg;
334 int sz;
335 RING_LOCALS;
336
337 sz = header.packet0.count;
338 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
339
340 if (!sz)
341 return 0;
342
343 if (sz * 4 > cmdbuf->bufsz)
344 return -EINVAL;
345
346 if (reg + sz * 4 >= 0x10000) {
347 DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg,
348 sz);
349 return -EINVAL;
350 }
351
352 if (r300_check_range(reg, sz)) {
353 /* go and check everything */
354 return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf,
355 header);
356 }
357 /* the rest of the data is safe to emit, whatever the values the user passed */
358
359 BEGIN_RING(1 + sz);
360 OUT_RING(CP_PACKET0(reg, sz - 1));
361 OUT_RING_TABLE((int *)cmdbuf->buf, sz);
362 ADVANCE_RING();
363
364 cmdbuf->buf += sz * 4;
365 cmdbuf->bufsz -= sz * 4;
366
367 return 0;
368 }
369
370 /**
371 * Uploads user-supplied vertex program instructions or parameters onto
372 * the graphics card.
373 * Called by r300_do_cp_cmdbuf.
374 */
375 static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
376 drm_radeon_kcmd_buffer_t *cmdbuf,
377 drm_r300_cmd_header_t header)
378 {
379 int sz;
380 int addr;
381 RING_LOCALS;
382
383 sz = header.vpu.count;
384 addr = (header.vpu.adrhi << 8) | header.vpu.adrlo;
385
386 if (!sz)
387 return 0;
388 if (sz * 16 > cmdbuf->bufsz)
389 return -EINVAL;
390
391 BEGIN_RING(5 + sz * 4);
392 /* Wait for VAP to come to senses.. */
393 /* there is no need to emit it multiple times, (only once before VAP is programmed,
394 but this optimization is for later */
395 OUT_RING_REG(R300_VAP_PVS_WAITIDLE, 0);
396 OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
397 OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
398 OUT_RING_TABLE((int *)cmdbuf->buf, sz * 4);
399
400 ADVANCE_RING();
401
402 cmdbuf->buf += sz * 16;
403 cmdbuf->bufsz -= sz * 16;
404
405 return 0;
406 }
407
408 /**
409 * Emit a clear packet from userspace.
410 * Called by r300_emit_packet3.
411 */
412 static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
413 drm_radeon_kcmd_buffer_t *cmdbuf)
414 {
415 RING_LOCALS;
416
417 if (8 * 4 > cmdbuf->bufsz)
418 return -EINVAL;
419
420 BEGIN_RING(10);
421 OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8));
422 OUT_RING(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
423 (1 << R300_PRIM_NUM_VERTICES_SHIFT));
424 OUT_RING_TABLE((int *)cmdbuf->buf, 8);
425 ADVANCE_RING();
426
427 cmdbuf->buf += 8 * 4;
428 cmdbuf->bufsz -= 8 * 4;
429
430 return 0;
431 }
432
433 static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
434 drm_radeon_kcmd_buffer_t *cmdbuf,
435 u32 header)
436 {
437 int count, i, k;
438 #define MAX_ARRAY_PACKET 64
439 u32 payload[MAX_ARRAY_PACKET];
440 u32 narrays;
441 RING_LOCALS;
442
443 count = (header >> 16) & 0x3fff;
444
445 if ((count + 1) > MAX_ARRAY_PACKET) {
446 DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
447 count);
448 return -EINVAL;
449 }
450 memset(payload, 0, MAX_ARRAY_PACKET * 4);
451 memcpy(payload, cmdbuf->buf + 4, (count + 1) * 4);
452
453 /* carefully check packet contents */
454
455 narrays = payload[0];
456 k = 0;
457 i = 1;
458 while ((k < narrays) && (i < (count + 1))) {
459 i++; /* skip attribute field */
460 if (!radeon_check_offset(dev_priv, payload[i])) {
461 DRM_ERROR
462 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
463 k, i);
464 return -EINVAL;
465 }
466 k++;
467 i++;
468 if (k == narrays)
469 break;
470 /* have one more to process, they come in pairs */
471 if (!radeon_check_offset(dev_priv, payload[i])) {
472 DRM_ERROR
473 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
474 k, i);
475 return -EINVAL;
476 }
477 k++;
478 i++;
479 }
480 /* do the counts match what we expect ? */
481 if ((k != narrays) || (i != (count + 1))) {
482 DRM_ERROR
483 ("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n",
484 k, i, narrays, count + 1);
485 return -EINVAL;
486 }
487
488 /* all clear, output packet */
489
490 BEGIN_RING(count + 2);
491 OUT_RING(header);
492 OUT_RING_TABLE(payload, count + 1);
493 ADVANCE_RING();
494
495 cmdbuf->buf += (count + 2) * 4;
496 cmdbuf->bufsz -= (count + 2) * 4;
497
498 return 0;
499 }
500
501 static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
502 drm_radeon_kcmd_buffer_t *cmdbuf)
503 {
504 u32 *cmd = (u32 *) cmdbuf->buf;
505 int count, ret;
506 RING_LOCALS;
507
508 count=(cmd[0]>>16) & 0x3fff;
509
510 if (cmd[0] & 0x8000) {
511 u32 offset;
512
513 if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
514 | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
515 offset = cmd[2] << 10;
516 ret = !radeon_check_offset(dev_priv, offset);
517 if (ret) {
518 DRM_ERROR("Invalid bitblt first offset is %08X\n", offset);
519 return -EINVAL;
520 }
521 }
522
523 if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
524 (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
525 offset = cmd[3] << 10;
526 ret = !radeon_check_offset(dev_priv, offset);
527 if (ret) {
528 DRM_ERROR("Invalid bitblt second offset is %08X\n", offset);
529 return -EINVAL;
530 }
531
532 }
533 }
534
535 BEGIN_RING(count+2);
536 OUT_RING(cmd[0]);
537 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
538 ADVANCE_RING();
539
540 cmdbuf->buf += (count+2)*4;
541 cmdbuf->bufsz -= (count+2)*4;
542
543 return 0;
544 }
545
546 static __inline__ int r300_emit_indx_buffer(drm_radeon_private_t *dev_priv,
547 drm_radeon_kcmd_buffer_t *cmdbuf)
548 {
549 u32 *cmd = (u32 *) cmdbuf->buf;
550 int count, ret;
551 RING_LOCALS;
552
553 count=(cmd[0]>>16) & 0x3fff;
554
555 if ((cmd[1] & 0x8000ffff) != 0x80000810) {
556 DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
557 return -EINVAL;
558 }
559 ret = !radeon_check_offset(dev_priv, cmd[2]);
560 if (ret) {
561 DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
562 return -EINVAL;
563 }
564
565 BEGIN_RING(count+2);
566 OUT_RING(cmd[0]);
567 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
568 ADVANCE_RING();
569
570 cmdbuf->buf += (count+2)*4;
571 cmdbuf->bufsz -= (count+2)*4;
572
573 return 0;
574 }
575
576 static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
577 drm_radeon_kcmd_buffer_t *cmdbuf)
578 {
579 u32 header;
580 int count;
581 RING_LOCALS;
582
583 if (4 > cmdbuf->bufsz)
584 return -EINVAL;
585
586 /* Fixme !! This simply emits a packet without much checking.
587 We need to be smarter. */
588
589 /* obtain first word - actual packet3 header */
590 header = *(u32 *) cmdbuf->buf;
591
592 /* Is it packet 3 ? */
593 if ((header >> 30) != 0x3) {
594 DRM_ERROR("Not a packet3 header (0x%08x)\n", header);
595 return -EINVAL;
596 }
597
598 count = (header >> 16) & 0x3fff;
599
600 /* Check again now that we know how much data to expect */
601 if ((count + 2) * 4 > cmdbuf->bufsz) {
602 DRM_ERROR
603 ("Expected packet3 of length %d but have only %d bytes left\n",
604 (count + 2) * 4, cmdbuf->bufsz);
605 return -EINVAL;
606 }
607
608 /* Is it a packet type we know about ? */
609 switch (header & 0xff00) {
610 case RADEON_3D_LOAD_VBPNTR: /* load vertex array pointers */
611 return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, header);
612
613 case RADEON_CNTL_BITBLT_MULTI:
614 return r300_emit_bitblt_multi(dev_priv, cmdbuf);
615
616 case RADEON_CP_INDX_BUFFER: /* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
617 return r300_emit_indx_buffer(dev_priv, cmdbuf);
618 case RADEON_CP_3D_DRAW_IMMD_2: /* triggers drawing using in-packet vertex data */
619 case RADEON_CP_3D_DRAW_VBUF_2: /* triggers drawing of vertex buffers setup elsewhere */
620 case RADEON_CP_3D_DRAW_INDX_2: /* triggers drawing using indices to vertex buffer */
621 case RADEON_WAIT_FOR_IDLE:
622 case RADEON_CP_NOP:
623 /* these packets are safe */
624 break;
625 default:
626 DRM_ERROR("Unknown packet3 header (0x%08x)\n", header);
627 return -EINVAL;
628 }
629
630 BEGIN_RING(count + 2);
631 OUT_RING(header);
632 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
633 ADVANCE_RING();
634
635 cmdbuf->buf += (count + 2) * 4;
636 cmdbuf->bufsz -= (count + 2) * 4;
637
638 return 0;
639 }
640
641 /**
642 * Emit a rendering packet3 from userspace.
643 * Called by r300_do_cp_cmdbuf.
644 */
645 static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
646 drm_radeon_kcmd_buffer_t *cmdbuf,
647 drm_r300_cmd_header_t header)
648 {
649 int n;
650 int ret;
651 char *orig_buf = cmdbuf->buf;
652 int orig_bufsz = cmdbuf->bufsz;
653
654 /* This is a do-while-loop so that we run the interior at least once,
655 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
656 */
657 n = 0;
658 do {
659 if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) {
660 ret = r300_emit_cliprects(dev_priv, cmdbuf, n);
661 if (ret)
662 return ret;
663
664 cmdbuf->buf = orig_buf;
665 cmdbuf->bufsz = orig_bufsz;
666 }
667
668 switch (header.packet3.packet) {
669 case R300_CMD_PACKET3_CLEAR:
670 DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n");
671 ret = r300_emit_clear(dev_priv, cmdbuf);
672 if (ret) {
673 DRM_ERROR("r300_emit_clear failed\n");
674 return ret;
675 }
676 break;
677
678 case R300_CMD_PACKET3_RAW:
679 DRM_DEBUG("R300_CMD_PACKET3_RAW\n");
680 ret = r300_emit_raw_packet3(dev_priv, cmdbuf);
681 if (ret) {
682 DRM_ERROR("r300_emit_raw_packet3 failed\n");
683 return ret;
684 }
685 break;
686
687 default:
688 DRM_ERROR("bad packet3 type %i at %p\n",
689 header.packet3.packet,
690 cmdbuf->buf - sizeof(header));
691 return -EINVAL;
692 }
693
694 n += R300_SIMULTANEOUS_CLIPRECTS;
695 } while (n < cmdbuf->nbox);
696
697 return 0;
698 }
699
700 /* Some of the R300 chips seem to be extremely touchy about the two registers
701 * that are configured in r300_pacify.
702 * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace
703 * sends a command buffer that contains only state setting commands and a
704 * vertex program/parameter upload sequence, this will eventually lead to a
705 * lockup, unless the sequence is bracketed by calls to r300_pacify.
706 * So we should take great care to *always* call r300_pacify before
707 * *anything* 3D related, and again afterwards. This is what the
708 * call bracket in r300_do_cp_cmdbuf is for.
709 */
710
711 /**
712 * Emit the sequence to pacify R300.
713 */
714 static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
715 {
716 RING_LOCALS;
717
718 BEGIN_RING(6);
719 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
720 OUT_RING(R300_RB3D_DSTCACHE_UNKNOWN_0A);
721 OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));
722 OUT_RING(R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE|
723 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
724 OUT_RING(CP_PACKET3(RADEON_CP_NOP, 0));
725 OUT_RING(0x0);
726 ADVANCE_RING();
727 }
728
729 /**
730 * Called by r300_do_cp_cmdbuf to update the internal buffer age and state.
731 * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
732 * be careful about how this function is called.
733 */
734 static void r300_discard_buffer(struct drm_device * dev, struct drm_buf * buf)
735 {
736 drm_radeon_private_t *dev_priv = dev->dev_private;
737 drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
738
739 buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
740 buf->pending = 1;
741 buf->used = 0;
742 }
743
744 static void r300_cmd_wait(drm_radeon_private_t * dev_priv,
745 drm_r300_cmd_header_t header)
746 {
747 u32 wait_until;
748 RING_LOCALS;
749
750 if (!header.wait.flags)
751 return;
752
753 wait_until = 0;
754
755 switch(header.wait.flags) {
756 case R300_WAIT_2D:
757 wait_until = RADEON_WAIT_2D_IDLE;
758 break;
759 case R300_WAIT_3D:
760 wait_until = RADEON_WAIT_3D_IDLE;
761 break;
762 case R300_NEW_WAIT_2D_3D:
763 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_3D_IDLE;
764 break;
765 case R300_NEW_WAIT_2D_2D_CLEAN:
766 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN;
767 break;
768 case R300_NEW_WAIT_3D_3D_CLEAN:
769 wait_until = RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN;
770 break;
771 case R300_NEW_WAIT_2D_2D_CLEAN_3D_3D_CLEAN:
772 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN;
773 wait_until |= RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN;
774 break;
775 default:
776 return;
777 }
778
779 BEGIN_RING(2);
780 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
781 OUT_RING(wait_until);
782 ADVANCE_RING();
783 }
784
785 static int r300_scratch(drm_radeon_private_t *dev_priv,
786 drm_radeon_kcmd_buffer_t *cmdbuf,
787 drm_r300_cmd_header_t header)
788 {
789 u32 *ref_age_base;
790 u32 i, buf_idx, h_pending;
791 RING_LOCALS;
792
793 if (cmdbuf->bufsz <
794 (sizeof(u64) + header.scratch.n_bufs * sizeof(buf_idx))) {
795 return -EINVAL;
796 }
797
798 if (header.scratch.reg >= 5) {
799 return -EINVAL;
800 }
801
802 dev_priv->scratch_ages[header.scratch.reg]++;
803
804 ref_age_base = (u32 *)(unsigned long)*((uint64_t *)cmdbuf->buf);
805
806 cmdbuf->buf += sizeof(u64);
807 cmdbuf->bufsz -= sizeof(u64);
808
809 for (i=0; i < header.scratch.n_bufs; i++) {
810 buf_idx = *(u32 *)cmdbuf->buf;
811 buf_idx *= 2; /* 8 bytes per buf */
812
813 if (DRM_COPY_TO_USER(ref_age_base + buf_idx, &dev_priv->scratch_ages[header.scratch.reg], sizeof(u32))) {
814 return -EINVAL;
815 }
816
817 if (DRM_COPY_FROM_USER(&h_pending, ref_age_base + buf_idx + 1, sizeof(u32))) {
818 return -EINVAL;
819 }
820
821 if (h_pending == 0) {
822 return -EINVAL;
823 }
824
825 h_pending--;
826
827 if (DRM_COPY_TO_USER(ref_age_base + buf_idx + 1, &h_pending, sizeof(u32))) {
828 return -EINVAL;
829 }
830
831 cmdbuf->buf += sizeof(buf_idx);
832 cmdbuf->bufsz -= sizeof(buf_idx);
833 }
834
835 BEGIN_RING(2);
836 OUT_RING( CP_PACKET0( RADEON_SCRATCH_REG0 + header.scratch.reg * 4, 0 ) );
837 OUT_RING( dev_priv->scratch_ages[header.scratch.reg] );
838 ADVANCE_RING();
839
840 return 0;
841 }
842
843 /**
844 * Uploads user-supplied vertex program instructions or parameters onto
845 * the graphics card.
846 * Called by r300_do_cp_cmdbuf.
847 */
848 static inline int r300_emit_r500fp(drm_radeon_private_t *dev_priv,
849 drm_radeon_kcmd_buffer_t *cmdbuf,
850 drm_r300_cmd_header_t header)
851 {
852 int sz;
853 int addr;
854 int type;
855 int clamp;
856 int stride;
857 RING_LOCALS;
858
859 sz = header.r500fp.count;
860 /* address is 9 bits 0 - 8, bit 1 of flags is part of address */
861 addr = ((header.r500fp.adrhi_flags & 1) << 8) | header.r500fp.adrlo;
862
863 type = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE);
864 clamp = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP);
865
866 addr |= (type << 16);
867 addr |= (clamp << 17);
868
869 stride = type ? 4 : 6;
870
871 DRM_DEBUG("r500fp %d %d type: %d\n", sz, addr, type);
872 if (!sz)
873 return 0;
874 if (sz * stride * 4 > cmdbuf->bufsz)
875 return -EINVAL;
876
877 BEGIN_RING(3 + sz * stride);
878 OUT_RING_REG(R500_GA_US_VECTOR_INDEX, addr);
879 OUT_RING(CP_PACKET0_TABLE(R500_GA_US_VECTOR_DATA, sz * stride - 1));
880 OUT_RING_TABLE((int *)cmdbuf->buf, sz * stride);
881
882 ADVANCE_RING();
883
884 cmdbuf->buf += sz * stride * 4;
885 cmdbuf->bufsz -= sz * stride * 4;
886
887 return 0;
888 }
889
890
891 /**
892 * Parses and validates a user-supplied command buffer and emits appropriate
893 * commands on the DMA ring buffer.
894 * Called by the ioctl handler function radeon_cp_cmdbuf.
895 */
896 int r300_do_cp_cmdbuf(struct drm_device *dev,
897 struct drm_file *file_priv,
898 drm_radeon_kcmd_buffer_t *cmdbuf)
899 {
900 drm_radeon_private_t *dev_priv = dev->dev_private;
901 struct drm_device_dma *dma = dev->dma;
902 struct drm_buf *buf = NULL;
903 int emit_dispatch_age = 0;
904 int ret = 0;
905
906 DRM_DEBUG("\n");
907
908 /* See the comment above r300_emit_begin3d for why this call must be here,
909 * and what the cleanup gotos are for. */
910 r300_pacify(dev_priv);
911
912 if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
913 ret = r300_emit_cliprects(dev_priv, cmdbuf, 0);
914 if (ret)
915 goto cleanup;
916 }
917
918 while (cmdbuf->bufsz >= sizeof(drm_r300_cmd_header_t)) {
919 int idx;
920 drm_r300_cmd_header_t header;
921
922 header.u = *(unsigned int *)cmdbuf->buf;
923
924 cmdbuf->buf += sizeof(header);
925 cmdbuf->bufsz -= sizeof(header);
926
927 switch (header.header.cmd_type) {
928 case R300_CMD_PACKET0:
929 DRM_DEBUG("R300_CMD_PACKET0\n");
930 ret = r300_emit_packet0(dev_priv, cmdbuf, header);
931 if (ret) {
932 DRM_ERROR("r300_emit_packet0 failed\n");
933 goto cleanup;
934 }
935 break;
936
937 case R300_CMD_VPU:
938 DRM_DEBUG("R300_CMD_VPU\n");
939 ret = r300_emit_vpu(dev_priv, cmdbuf, header);
940 if (ret) {
941 DRM_ERROR("r300_emit_vpu failed\n");
942 goto cleanup;
943 }
944 break;
945
946 case R300_CMD_PACKET3:
947 DRM_DEBUG("R300_CMD_PACKET3\n");
948 ret = r300_emit_packet3(dev_priv, cmdbuf, header);
949 if (ret) {
950 DRM_ERROR("r300_emit_packet3 failed\n");
951 goto cleanup;
952 }
953 break;
954
955 case R300_CMD_END3D:
956 DRM_DEBUG("R300_CMD_END3D\n");
957 /* TODO:
958 Ideally userspace driver should not need to issue this call,
959 i.e. the drm driver should issue it automatically and prevent
960 lockups.
961
962 In practice, we do not understand why this call is needed and what
963 it does (except for some vague guesses that it has to do with cache
964 coherence) and so the user space driver does it.
965
966 Once we are sure which uses prevent lockups the code could be moved
967 into the kernel and the userspace driver will not
968 need to use this command.
969
970 Note that issuing this command does not hurt anything
971 except, possibly, performance */
972 r300_pacify(dev_priv);
973 break;
974
975 case R300_CMD_CP_DELAY:
976 /* simple enough, we can do it here */
977 DRM_DEBUG("R300_CMD_CP_DELAY\n");
978 {
979 int i;
980 RING_LOCALS;
981
982 BEGIN_RING(header.delay.count);
983 for (i = 0; i < header.delay.count; i++)
984 OUT_RING(RADEON_CP_PACKET2);
985 ADVANCE_RING();
986 }
987 break;
988
989 case R300_CMD_DMA_DISCARD:
990 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
991 idx = header.dma.buf_idx;
992 if (idx < 0 || idx >= dma->buf_count) {
993 DRM_ERROR("buffer index %d (of %d max)\n",
994 idx, dma->buf_count - 1);
995 ret = -EINVAL;
996 goto cleanup;
997 }
998
999 buf = dma->buflist[idx];
1000 if (buf->file_priv != file_priv || buf->pending) {
1001 DRM_ERROR("bad buffer %p %p %d\n",
1002 buf->file_priv, file_priv,
1003 buf->pending);
1004 ret = -EINVAL;
1005 goto cleanup;
1006 }
1007
1008 emit_dispatch_age = 1;
1009 r300_discard_buffer(dev, buf);
1010 break;
1011
1012 case R300_CMD_WAIT:
1013 DRM_DEBUG("R300_CMD_WAIT\n");
1014 r300_cmd_wait(dev_priv, header);
1015 break;
1016
1017 case R300_CMD_SCRATCH:
1018 DRM_DEBUG("R300_CMD_SCRATCH\n");
1019 ret = r300_scratch(dev_priv, cmdbuf, header);
1020 if (ret) {
1021 DRM_ERROR("r300_scratch failed\n");
1022 goto cleanup;
1023 }
1024 break;
1025
1026 case R300_CMD_R500FP:
1027 if ((dev_priv->flags & RADEON_FAMILY_MASK) < CHIP_RV515) {
1028 DRM_ERROR("Calling r500 command on r300 card\n");
1029 ret = -EINVAL;
1030 goto cleanup;
1031 }
1032 DRM_DEBUG("R300_CMD_R500FP\n");
1033 ret = r300_emit_r500fp(dev_priv, cmdbuf, header);
1034 if (ret) {
1035 DRM_ERROR("r300_emit_r500fp failed\n");
1036 goto cleanup;
1037 }
1038 break;
1039 default:
1040 DRM_ERROR("bad cmd_type %i at %p\n",
1041 header.header.cmd_type,
1042 cmdbuf->buf - sizeof(header));
1043 ret = -EINVAL;
1044 goto cleanup;
1045 }
1046 }
1047
1048 DRM_DEBUG("END\n");
1049
1050 cleanup:
1051 r300_pacify(dev_priv);
1052
1053 /* We emit the vertex buffer age here, outside the pacifier "brackets"
1054 * for two reasons:
1055 * (1) This may coalesce multiple age emissions into a single one and
1056 * (2) more importantly, some chips lock up hard when scratch registers
1057 * are written inside the pacifier bracket.
1058 */
1059 if (emit_dispatch_age) {
1060 RING_LOCALS;
1061
1062 /* Emit the vertex buffer age */
1063 BEGIN_RING(2);
1064 RADEON_DISPATCH_AGE(dev_priv->sarea_priv->last_dispatch);
1065 ADVANCE_RING();
1066 }
1067
1068 COMMIT_RING();
1069
1070 return ret;
1071 }
Linux kernel - Deliverables
This page took 0.054352 seconds and 6 git commands to generate.