Loading...
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 "drm_buffer.h"
37#include "radeon_drm.h"
38#include "radeon_drv.h"
39#include "r300_reg.h"
40
41#include <asm/unaligned.h>
42
43#define R300_SIMULTANEOUS_CLIPRECTS 4
44
45/* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
46 */
47static const int r300_cliprect_cntl[4] = {
48 0xAAAA,
49 0xEEEE,
50 0xFEFE,
51 0xFFFE
52};
53
54/**
55 * Emit up to R300_SIMULTANEOUS_CLIPRECTS cliprects from the given command
56 * buffer, starting with index n.
57 */
58static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
59 drm_radeon_kcmd_buffer_t *cmdbuf, int n)
60{
61 struct drm_clip_rect box;
62 int nr;
63 int i;
64 RING_LOCALS;
65
66 nr = cmdbuf->nbox - n;
67 if (nr > R300_SIMULTANEOUS_CLIPRECTS)
68 nr = R300_SIMULTANEOUS_CLIPRECTS;
69
70 DRM_DEBUG("%i cliprects\n", nr);
71
72 if (nr) {
73 BEGIN_RING(6 + nr * 2);
74 OUT_RING(CP_PACKET0(R300_RE_CLIPRECT_TL_0, nr * 2 - 1));
75
76 for (i = 0; i < nr; ++i) {
77 if (DRM_COPY_FROM_USER_UNCHECKED
78 (&box, &cmdbuf->boxes[n + i], sizeof(box))) {
79 DRM_ERROR("copy cliprect faulted\n");
80 return -EFAULT;
81 }
82
83 box.x2--; /* Hardware expects inclusive bottom-right corner */
84 box.y2--;
85
86 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
87 box.x1 = (box.x1) &
88 R300_CLIPRECT_MASK;
89 box.y1 = (box.y1) &
90 R300_CLIPRECT_MASK;
91 box.x2 = (box.x2) &
92 R300_CLIPRECT_MASK;
93 box.y2 = (box.y2) &
94 R300_CLIPRECT_MASK;
95 } else {
96 box.x1 = (box.x1 + R300_CLIPRECT_OFFSET) &
97 R300_CLIPRECT_MASK;
98 box.y1 = (box.y1 + R300_CLIPRECT_OFFSET) &
99 R300_CLIPRECT_MASK;
100 box.x2 = (box.x2 + R300_CLIPRECT_OFFSET) &
101 R300_CLIPRECT_MASK;
102 box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) &
103 R300_CLIPRECT_MASK;
104 }
105
106 OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
107 (box.y1 << R300_CLIPRECT_Y_SHIFT));
108 OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
109 (box.y2 << R300_CLIPRECT_Y_SHIFT));
110
111 }
112
113 OUT_RING_REG(R300_RE_CLIPRECT_CNTL, r300_cliprect_cntl[nr - 1]);
114
115 /* TODO/SECURITY: Force scissors to a safe value, otherwise the
116 * client might be able to trample over memory.
117 * The impact should be very limited, but I'd rather be safe than
118 * sorry.
119 */
120 OUT_RING(CP_PACKET0(R300_RE_SCISSORS_TL, 1));
121 OUT_RING(0);
122 OUT_RING(R300_SCISSORS_X_MASK | R300_SCISSORS_Y_MASK);
123 ADVANCE_RING();
124 } else {
125 /* Why we allow zero cliprect rendering:
126 * There are some commands in a command buffer that must be submitted
127 * even when there are no cliprects, e.g. DMA buffer discard
128 * or state setting (though state setting could be avoided by
129 * simulating a loss of context).
130 *
131 * Now since the cmdbuf interface is so chaotic right now (and is
132 * bound to remain that way for a bit until things settle down),
133 * it is basically impossible to filter out the commands that are
134 * necessary and those that aren't.
135 *
136 * So I choose the safe way and don't do any filtering at all;
137 * instead, I simply set up the engine so that all rendering
138 * can't produce any fragments.
139 */
140 BEGIN_RING(2);
141 OUT_RING_REG(R300_RE_CLIPRECT_CNTL, 0);
142 ADVANCE_RING();
143 }
144
145 /* flus cache and wait idle clean after cliprect change */
146 BEGIN_RING(2);
147 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
148 OUT_RING(R300_RB3D_DC_FLUSH);
149 ADVANCE_RING();
150 BEGIN_RING(2);
151 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
152 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
153 ADVANCE_RING();
154 /* set flush flag */
155 dev_priv->track_flush |= RADEON_FLUSH_EMITED;
156
157 return 0;
158}
159
160static u8 r300_reg_flags[0x10000 >> 2];
161
162void r300_init_reg_flags(struct drm_device *dev)
163{
164 int i;
165 drm_radeon_private_t *dev_priv = dev->dev_private;
166
167 memset(r300_reg_flags, 0, 0x10000 >> 2);
168#define ADD_RANGE_MARK(reg, count,mark) \
169 for(i=((reg)>>2);i<((reg)>>2)+(count);i++)\
170 r300_reg_flags[i]|=(mark);
171
172#define MARK_SAFE 1
173#define MARK_CHECK_OFFSET 2
174
175#define ADD_RANGE(reg, count) ADD_RANGE_MARK(reg, count, MARK_SAFE)
176
177 /* these match cmducs() command in r300_driver/r300/r300_cmdbuf.c */
178 ADD_RANGE(R300_SE_VPORT_XSCALE, 6);
179 ADD_RANGE(R300_VAP_CNTL, 1);
180 ADD_RANGE(R300_SE_VTE_CNTL, 2);
181 ADD_RANGE(0x2134, 2);
182 ADD_RANGE(R300_VAP_CNTL_STATUS, 1);
183 ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
184 ADD_RANGE(0x21DC, 1);
185 ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
186 ADD_RANGE(R300_VAP_CLIP_X_0, 4);
187 ADD_RANGE(R300_VAP_PVS_STATE_FLUSH_REG, 1);
188 ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
189 ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
190 ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
191 ADD_RANGE(R300_GB_ENABLE, 1);
192 ADD_RANGE(R300_GB_MSPOS0, 5);
193 ADD_RANGE(R300_TX_INVALTAGS, 1);
194 ADD_RANGE(R300_TX_ENABLE, 1);
195 ADD_RANGE(0x4200, 4);
196 ADD_RANGE(0x4214, 1);
197 ADD_RANGE(R300_RE_POINTSIZE, 1);
198 ADD_RANGE(0x4230, 3);
199 ADD_RANGE(R300_RE_LINE_CNT, 1);
200 ADD_RANGE(R300_RE_UNK4238, 1);
201 ADD_RANGE(0x4260, 3);
202 ADD_RANGE(R300_RE_SHADE, 4);
203 ADD_RANGE(R300_RE_POLYGON_MODE, 5);
204 ADD_RANGE(R300_RE_ZBIAS_CNTL, 1);
205 ADD_RANGE(R300_RE_ZBIAS_T_FACTOR, 4);
206 ADD_RANGE(R300_RE_OCCLUSION_CNTL, 1);
207 ADD_RANGE(R300_RE_CULL_CNTL, 1);
208 ADD_RANGE(0x42C0, 2);
209 ADD_RANGE(R300_RS_CNTL_0, 2);
210
211 ADD_RANGE(R300_SU_REG_DEST, 1);
212 if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV530)
213 ADD_RANGE(RV530_FG_ZBREG_DEST, 1);
214
215 ADD_RANGE(R300_SC_HYPERZ, 2);
216 ADD_RANGE(0x43E8, 1);
217
218 ADD_RANGE(0x46A4, 5);
219
220 ADD_RANGE(R300_RE_FOG_STATE, 1);
221 ADD_RANGE(R300_FOG_COLOR_R, 3);
222 ADD_RANGE(R300_PP_ALPHA_TEST, 2);
223 ADD_RANGE(0x4BD8, 1);
224 ADD_RANGE(R300_PFS_PARAM_0_X, 64);
225 ADD_RANGE(0x4E00, 1);
226 ADD_RANGE(R300_RB3D_CBLEND, 2);
227 ADD_RANGE(R300_RB3D_COLORMASK, 1);
228 ADD_RANGE(R300_RB3D_BLEND_COLOR, 3);
229 ADD_RANGE_MARK(R300_RB3D_COLOROFFSET0, 1, MARK_CHECK_OFFSET); /* check offset */
230 ADD_RANGE(R300_RB3D_COLORPITCH0, 1);
231 ADD_RANGE(0x4E50, 9);
232 ADD_RANGE(0x4E88, 1);
233 ADD_RANGE(0x4EA0, 2);
234 ADD_RANGE(R300_ZB_CNTL, 3);
235 ADD_RANGE(R300_ZB_FORMAT, 4);
236 ADD_RANGE_MARK(R300_ZB_DEPTHOFFSET, 1, MARK_CHECK_OFFSET); /* check offset */
237 ADD_RANGE(R300_ZB_DEPTHPITCH, 1);
238 ADD_RANGE(R300_ZB_DEPTHCLEARVALUE, 1);
239 ADD_RANGE(R300_ZB_ZMASK_OFFSET, 13);
240 ADD_RANGE(R300_ZB_ZPASS_DATA, 2); /* ZB_ZPASS_DATA, ZB_ZPASS_ADDR */
241
242 ADD_RANGE(R300_TX_FILTER_0, 16);
243 ADD_RANGE(R300_TX_FILTER1_0, 16);
244 ADD_RANGE(R300_TX_SIZE_0, 16);
245 ADD_RANGE(R300_TX_FORMAT_0, 16);
246 ADD_RANGE(R300_TX_PITCH_0, 16);
247 /* Texture offset is dangerous and needs more checking */
248 ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET);
249 ADD_RANGE(R300_TX_CHROMA_KEY_0, 16);
250 ADD_RANGE(R300_TX_BORDER_COLOR_0, 16);
251
252 /* Sporadic registers used as primitives are emitted */
253 ADD_RANGE(R300_ZB_ZCACHE_CTLSTAT, 1);
254 ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1);
255 ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8);
256 ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8);
257
258 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
259 ADD_RANGE(R500_VAP_INDEX_OFFSET, 1);
260 ADD_RANGE(R500_US_CONFIG, 2);
261 ADD_RANGE(R500_US_CODE_ADDR, 3);
262 ADD_RANGE(R500_US_FC_CTRL, 1);
263 ADD_RANGE(R500_RS_IP_0, 16);
264 ADD_RANGE(R500_RS_INST_0, 16);
265 ADD_RANGE(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2);
266 ADD_RANGE(R500_RB3D_CONSTANT_COLOR_AR, 2);
267 ADD_RANGE(R500_ZB_FIFO_SIZE, 2);
268 } else {
269 ADD_RANGE(R300_PFS_CNTL_0, 3);
270 ADD_RANGE(R300_PFS_NODE_0, 4);
271 ADD_RANGE(R300_PFS_TEXI_0, 64);
272 ADD_RANGE(R300_PFS_INSTR0_0, 64);
273 ADD_RANGE(R300_PFS_INSTR1_0, 64);
274 ADD_RANGE(R300_PFS_INSTR2_0, 64);
275 ADD_RANGE(R300_PFS_INSTR3_0, 64);
276 ADD_RANGE(R300_RS_INTERP_0, 8);
277 ADD_RANGE(R300_RS_ROUTE_0, 8);
278
279 }
280}
281
282static __inline__ int r300_check_range(unsigned reg, int count)
283{
284 int i;
285 if (reg & ~0xffff)
286 return -1;
287 for (i = (reg >> 2); i < (reg >> 2) + count; i++)
288 if (r300_reg_flags[i] != MARK_SAFE)
289 return 1;
290 return 0;
291}
292
293static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t *
294 dev_priv,
295 drm_radeon_kcmd_buffer_t
296 * cmdbuf,
297 drm_r300_cmd_header_t
298 header)
299{
300 int reg;
301 int sz;
302 int i;
303 u32 *value;
304 RING_LOCALS;
305
306 sz = header.packet0.count;
307 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
308
309 if ((sz > 64) || (sz < 0)) {
310 DRM_ERROR("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n",
311 reg, sz);
312 return -EINVAL;
313 }
314
315 for (i = 0; i < sz; i++) {
316 switch (r300_reg_flags[(reg >> 2) + i]) {
317 case MARK_SAFE:
318 break;
319 case MARK_CHECK_OFFSET:
320 value = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
321 if (!radeon_check_offset(dev_priv, *value)) {
322 DRM_ERROR("Offset failed range check (reg=%04x sz=%d)\n",
323 reg, sz);
324 return -EINVAL;
325 }
326 break;
327 default:
328 DRM_ERROR("Register %04x failed check as flag=%02x\n",
329 reg + i * 4, r300_reg_flags[(reg >> 2) + i]);
330 return -EINVAL;
331 }
332 }
333
334 BEGIN_RING(1 + sz);
335 OUT_RING(CP_PACKET0(reg, sz - 1));
336 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
337 ADVANCE_RING();
338
339 return 0;
340}
341
342/**
343 * Emits a packet0 setting arbitrary registers.
344 * Called by r300_do_cp_cmdbuf.
345 *
346 * Note that checks are performed on contents and addresses of the registers
347 */
348static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv,
349 drm_radeon_kcmd_buffer_t *cmdbuf,
350 drm_r300_cmd_header_t header)
351{
352 int reg;
353 int sz;
354 RING_LOCALS;
355
356 sz = header.packet0.count;
357 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
358
359 if (!sz)
360 return 0;
361
362 if (sz * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
363 return -EINVAL;
364
365 if (reg + sz * 4 >= 0x10000) {
366 DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg,
367 sz);
368 return -EINVAL;
369 }
370
371 if (r300_check_range(reg, sz)) {
372 /* go and check everything */
373 return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf,
374 header);
375 }
376 /* the rest of the data is safe to emit, whatever the values the user passed */
377
378 BEGIN_RING(1 + sz);
379 OUT_RING(CP_PACKET0(reg, sz - 1));
380 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
381 ADVANCE_RING();
382
383 return 0;
384}
385
386/**
387 * Uploads user-supplied vertex program instructions or parameters onto
388 * the graphics card.
389 * Called by r300_do_cp_cmdbuf.
390 */
391static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
392 drm_radeon_kcmd_buffer_t *cmdbuf,
393 drm_r300_cmd_header_t header)
394{
395 int sz;
396 int addr;
397 RING_LOCALS;
398
399 sz = header.vpu.count;
400 addr = (header.vpu.adrhi << 8) | header.vpu.adrlo;
401
402 if (!sz)
403 return 0;
404 if (sz * 16 > drm_buffer_unprocessed(cmdbuf->buffer))
405 return -EINVAL;
406
407 /* VAP is very sensitive so we purge cache before we program it
408 * and we also flush its state before & after */
409 BEGIN_RING(6);
410 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
411 OUT_RING(R300_RB3D_DC_FLUSH);
412 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
413 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
414 OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
415 OUT_RING(0);
416 ADVANCE_RING();
417 /* set flush flag */
418 dev_priv->track_flush |= RADEON_FLUSH_EMITED;
419
420 BEGIN_RING(3 + sz * 4);
421 OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
422 OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
423 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz * 4);
424 ADVANCE_RING();
425
426 BEGIN_RING(2);
427 OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
428 OUT_RING(0);
429 ADVANCE_RING();
430
431 return 0;
432}
433
434/**
435 * Emit a clear packet from userspace.
436 * Called by r300_emit_packet3.
437 */
438static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
439 drm_radeon_kcmd_buffer_t *cmdbuf)
440{
441 RING_LOCALS;
442
443 if (8 * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
444 return -EINVAL;
445
446 BEGIN_RING(10);
447 OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8));
448 OUT_RING(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
449 (1 << R300_PRIM_NUM_VERTICES_SHIFT));
450 OUT_RING_DRM_BUFFER(cmdbuf->buffer, 8);
451 ADVANCE_RING();
452
453 BEGIN_RING(4);
454 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
455 OUT_RING(R300_RB3D_DC_FLUSH);
456 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
457 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
458 ADVANCE_RING();
459 /* set flush flag */
460 dev_priv->track_flush |= RADEON_FLUSH_EMITED;
461
462 return 0;
463}
464
465static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
466 drm_radeon_kcmd_buffer_t *cmdbuf,
467 u32 header)
468{
469 int count, i, k;
470#define MAX_ARRAY_PACKET 64
471 u32 *data;
472 u32 narrays;
473 RING_LOCALS;
474
475 count = (header & RADEON_CP_PACKET_COUNT_MASK) >> 16;
476
477 if ((count + 1) > MAX_ARRAY_PACKET) {
478 DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
479 count);
480 return -EINVAL;
481 }
482 /* carefully check packet contents */
483
484 /* We have already read the header so advance the buffer. */
485 drm_buffer_advance(cmdbuf->buffer, 4);
486
487 narrays = *(u32 *)drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
488 k = 0;
489 i = 1;
490 while ((k < narrays) && (i < (count + 1))) {
491 i++; /* skip attribute field */
492 data = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
493 if (!radeon_check_offset(dev_priv, *data)) {
494 DRM_ERROR
495 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
496 k, i);
497 return -EINVAL;
498 }
499 k++;
500 i++;
501 if (k == narrays)
502 break;
503 /* have one more to process, they come in pairs */
504 data = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
505 if (!radeon_check_offset(dev_priv, *data)) {
506 DRM_ERROR
507 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
508 k, i);
509 return -EINVAL;
510 }
511 k++;
512 i++;
513 }
514 /* do the counts match what we expect ? */
515 if ((k != narrays) || (i != (count + 1))) {
516 DRM_ERROR
517 ("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n",
518 k, i, narrays, count + 1);
519 return -EINVAL;
520 }
521
522 /* all clear, output packet */
523
524 BEGIN_RING(count + 2);
525 OUT_RING(header);
526 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 1);
527 ADVANCE_RING();
528
529 return 0;
530}
531
532static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
533 drm_radeon_kcmd_buffer_t *cmdbuf)
534{
535 u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
536 int count, ret;
537 RING_LOCALS;
538
539
540 count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
541
542 if (*cmd & 0x8000) {
543 u32 offset;
544 u32 *cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
545 if (*cmd1 & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
546 | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
547
548 u32 *cmd2 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
549 offset = *cmd2 << 10;
550 ret = !radeon_check_offset(dev_priv, offset);
551 if (ret) {
552 DRM_ERROR("Invalid bitblt first offset is %08X\n", offset);
553 return -EINVAL;
554 }
555 }
556
557 if ((*cmd1 & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
558 (*cmd1 & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
559 u32 *cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
560 offset = *cmd3 << 10;
561 ret = !radeon_check_offset(dev_priv, offset);
562 if (ret) {
563 DRM_ERROR("Invalid bitblt second offset is %08X\n", offset);
564 return -EINVAL;
565 }
566
567 }
568 }
569
570 BEGIN_RING(count+2);
571 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
572 ADVANCE_RING();
573
574 return 0;
575}
576
577static __inline__ int r300_emit_draw_indx_2(drm_radeon_private_t *dev_priv,
578 drm_radeon_kcmd_buffer_t *cmdbuf)
579{
580 u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
581 u32 *cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
582 int count;
583 int expected_count;
584 RING_LOCALS;
585
586 count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
587
588 expected_count = *cmd1 >> 16;
589 if (!(*cmd1 & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
590 expected_count = (expected_count+1)/2;
591
592 if (count && count != expected_count) {
593 DRM_ERROR("3D_DRAW_INDX_2: packet size %i, expected %i\n",
594 count, expected_count);
595 return -EINVAL;
596 }
597
598 BEGIN_RING(count+2);
599 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
600 ADVANCE_RING();
601
602 if (!count) {
603 drm_r300_cmd_header_t stack_header, *header;
604 u32 *cmd1, *cmd2, *cmd3;
605
606 if (drm_buffer_unprocessed(cmdbuf->buffer)
607 < 4*4 + sizeof(stack_header)) {
608 DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER, but stream is too short.\n");
609 return -EINVAL;
610 }
611
612 header = drm_buffer_read_object(cmdbuf->buffer,
613 sizeof(stack_header), &stack_header);
614
615 cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
616 cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
617 cmd2 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
618 cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
619
620 if (header->header.cmd_type != R300_CMD_PACKET3 ||
621 header->packet3.packet != R300_CMD_PACKET3_RAW ||
622 *cmd != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) {
623 DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER.\n");
624 return -EINVAL;
625 }
626
627 if ((*cmd1 & 0x8000ffff) != 0x80000810) {
628 DRM_ERROR("Invalid indx_buffer reg address %08X\n",
629 *cmd1);
630 return -EINVAL;
631 }
632 if (!radeon_check_offset(dev_priv, *cmd2)) {
633 DRM_ERROR("Invalid indx_buffer offset is %08X\n",
634 *cmd2);
635 return -EINVAL;
636 }
637 if (*cmd3 != expected_count) {
638 DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n",
639 *cmd3, expected_count);
640 return -EINVAL;
641 }
642
643 BEGIN_RING(4);
644 OUT_RING_DRM_BUFFER(cmdbuf->buffer, 4);
645 ADVANCE_RING();
646 }
647
648 return 0;
649}
650
651static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
652 drm_radeon_kcmd_buffer_t *cmdbuf)
653{
654 u32 *header;
655 int count;
656 RING_LOCALS;
657
658 if (4 > drm_buffer_unprocessed(cmdbuf->buffer))
659 return -EINVAL;
660
661 /* Fixme !! This simply emits a packet without much checking.
662 We need to be smarter. */
663
664 /* obtain first word - actual packet3 header */
665 header = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
666
667 /* Is it packet 3 ? */
668 if ((*header >> 30) != 0x3) {
669 DRM_ERROR("Not a packet3 header (0x%08x)\n", *header);
670 return -EINVAL;
671 }
672
673 count = (*header >> 16) & 0x3fff;
674
675 /* Check again now that we know how much data to expect */
676 if ((count + 2) * 4 > drm_buffer_unprocessed(cmdbuf->buffer)) {
677 DRM_ERROR
678 ("Expected packet3 of length %d but have only %d bytes left\n",
679 (count + 2) * 4, drm_buffer_unprocessed(cmdbuf->buffer));
680 return -EINVAL;
681 }
682
683 /* Is it a packet type we know about ? */
684 switch (*header & 0xff00) {
685 case RADEON_3D_LOAD_VBPNTR: /* load vertex array pointers */
686 return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, *header);
687
688 case RADEON_CNTL_BITBLT_MULTI:
689 return r300_emit_bitblt_multi(dev_priv, cmdbuf);
690
691 case RADEON_CP_INDX_BUFFER:
692 DRM_ERROR("packet3 INDX_BUFFER without preceding 3D_DRAW_INDX_2 is illegal.\n");
693 return -EINVAL;
694 case RADEON_CP_3D_DRAW_IMMD_2:
695 /* triggers drawing using in-packet vertex data */
696 case RADEON_CP_3D_DRAW_VBUF_2:
697 /* triggers drawing of vertex buffers setup elsewhere */
698 dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
699 RADEON_PURGE_EMITED);
700 break;
701 case RADEON_CP_3D_DRAW_INDX_2:
702 /* triggers drawing using indices to vertex buffer */
703 /* whenever we send vertex we clear flush & purge */
704 dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
705 RADEON_PURGE_EMITED);
706 return r300_emit_draw_indx_2(dev_priv, cmdbuf);
707 case RADEON_WAIT_FOR_IDLE:
708 case RADEON_CP_NOP:
709 /* these packets are safe */
710 break;
711 default:
712 DRM_ERROR("Unknown packet3 header (0x%08x)\n", *header);
713 return -EINVAL;
714 }
715
716 BEGIN_RING(count + 2);
717 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
718 ADVANCE_RING();
719
720 return 0;
721}
722
723/**
724 * Emit a rendering packet3 from userspace.
725 * Called by r300_do_cp_cmdbuf.
726 */
727static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
728 drm_radeon_kcmd_buffer_t *cmdbuf,
729 drm_r300_cmd_header_t header)
730{
731 int n;
732 int ret;
733 int orig_iter = cmdbuf->buffer->iterator;
734
735 /* This is a do-while-loop so that we run the interior at least once,
736 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
737 */
738 n = 0;
739 do {
740 if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) {
741 ret = r300_emit_cliprects(dev_priv, cmdbuf, n);
742 if (ret)
743 return ret;
744
745 cmdbuf->buffer->iterator = orig_iter;
746 }
747
748 switch (header.packet3.packet) {
749 case R300_CMD_PACKET3_CLEAR:
750 DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n");
751 ret = r300_emit_clear(dev_priv, cmdbuf);
752 if (ret) {
753 DRM_ERROR("r300_emit_clear failed\n");
754 return ret;
755 }
756 break;
757
758 case R300_CMD_PACKET3_RAW:
759 DRM_DEBUG("R300_CMD_PACKET3_RAW\n");
760 ret = r300_emit_raw_packet3(dev_priv, cmdbuf);
761 if (ret) {
762 DRM_ERROR("r300_emit_raw_packet3 failed\n");
763 return ret;
764 }
765 break;
766
767 default:
768 DRM_ERROR("bad packet3 type %i at byte %d\n",
769 header.packet3.packet,
770 cmdbuf->buffer->iterator - (int)sizeof(header));
771 return -EINVAL;
772 }
773
774 n += R300_SIMULTANEOUS_CLIPRECTS;
775 } while (n < cmdbuf->nbox);
776
777 return 0;
778}
779
780/* Some of the R300 chips seem to be extremely touchy about the two registers
781 * that are configured in r300_pacify.
782 * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace
783 * sends a command buffer that contains only state setting commands and a
784 * vertex program/parameter upload sequence, this will eventually lead to a
785 * lockup, unless the sequence is bracketed by calls to r300_pacify.
786 * So we should take great care to *always* call r300_pacify before
787 * *anything* 3D related, and again afterwards. This is what the
788 * call bracket in r300_do_cp_cmdbuf is for.
789 */
790
791/**
792 * Emit the sequence to pacify R300.
793 */
794static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
795{
796 uint32_t cache_z, cache_3d, cache_2d;
797 RING_LOCALS;
798
799 cache_z = R300_ZC_FLUSH;
800 cache_2d = R300_RB2D_DC_FLUSH;
801 cache_3d = R300_RB3D_DC_FLUSH;
802 if (!(dev_priv->track_flush & RADEON_PURGE_EMITED)) {
803 /* we can purge, primitive where draw since last purge */
804 cache_z |= R300_ZC_FREE;
805 cache_2d |= R300_RB2D_DC_FREE;
806 cache_3d |= R300_RB3D_DC_FREE;
807 }
808
809 /* flush & purge zbuffer */
810 BEGIN_RING(2);
811 OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));
812 OUT_RING(cache_z);
813 ADVANCE_RING();
814 /* flush & purge 3d */
815 BEGIN_RING(2);
816 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
817 OUT_RING(cache_3d);
818 ADVANCE_RING();
819 /* flush & purge texture */
820 BEGIN_RING(2);
821 OUT_RING(CP_PACKET0(R300_TX_INVALTAGS, 0));
822 OUT_RING(0);
823 ADVANCE_RING();
824 /* FIXME: is this one really needed ? */
825 BEGIN_RING(2);
826 OUT_RING(CP_PACKET0(R300_RB3D_AARESOLVE_CTL, 0));
827 OUT_RING(0);
828 ADVANCE_RING();
829 BEGIN_RING(2);
830 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
831 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
832 ADVANCE_RING();
833 /* flush & purge 2d through E2 as RB2D will trigger lockup */
834 BEGIN_RING(4);
835 OUT_RING(CP_PACKET0(R300_DSTCACHE_CTLSTAT, 0));
836 OUT_RING(cache_2d);
837 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
838 OUT_RING(RADEON_WAIT_2D_IDLECLEAN |
839 RADEON_WAIT_HOST_IDLECLEAN);
840 ADVANCE_RING();
841 /* set flush & purge flags */
842 dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED;
843}
844
845/**
846 * Called by r300_do_cp_cmdbuf to update the internal buffer age and state.
847 * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
848 * be careful about how this function is called.
849 */
850static void r300_discard_buffer(struct drm_device *dev, struct drm_master *master, struct drm_buf *buf)
851{
852 drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
853 struct drm_radeon_master_private *master_priv = master->driver_priv;
854
855 buf_priv->age = ++master_priv->sarea_priv->last_dispatch;
856 buf->pending = 1;
857 buf->used = 0;
858}
859
860static void r300_cmd_wait(drm_radeon_private_t * dev_priv,
861 drm_r300_cmd_header_t header)
862{
863 u32 wait_until;
864 RING_LOCALS;
865
866 if (!header.wait.flags)
867 return;
868
869 wait_until = 0;
870
871 switch(header.wait.flags) {
872 case R300_WAIT_2D:
873 wait_until = RADEON_WAIT_2D_IDLE;
874 break;
875 case R300_WAIT_3D:
876 wait_until = RADEON_WAIT_3D_IDLE;
877 break;
878 case R300_NEW_WAIT_2D_3D:
879 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_3D_IDLE;
880 break;
881 case R300_NEW_WAIT_2D_2D_CLEAN:
882 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN;
883 break;
884 case R300_NEW_WAIT_3D_3D_CLEAN:
885 wait_until = RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN;
886 break;
887 case R300_NEW_WAIT_2D_2D_CLEAN_3D_3D_CLEAN:
888 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN;
889 wait_until |= RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN;
890 break;
891 default:
892 return;
893 }
894
895 BEGIN_RING(2);
896 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
897 OUT_RING(wait_until);
898 ADVANCE_RING();
899}
900
901static int r300_scratch(drm_radeon_private_t *dev_priv,
902 drm_radeon_kcmd_buffer_t *cmdbuf,
903 drm_r300_cmd_header_t header)
904{
905 u32 *ref_age_base;
906 u32 i, *buf_idx, h_pending;
907 u64 *ptr_addr;
908 u64 stack_ptr_addr;
909 RING_LOCALS;
910
911 if (drm_buffer_unprocessed(cmdbuf->buffer) <
912 (sizeof(u64) + header.scratch.n_bufs * sizeof(*buf_idx))) {
913 return -EINVAL;
914 }
915
916 if (header.scratch.reg >= 5) {
917 return -EINVAL;
918 }
919
920 dev_priv->scratch_ages[header.scratch.reg]++;
921
922 ptr_addr = drm_buffer_read_object(cmdbuf->buffer,
923 sizeof(stack_ptr_addr), &stack_ptr_addr);
924 ref_age_base = (u32 *)(unsigned long)get_unaligned(ptr_addr);
925
926 for (i=0; i < header.scratch.n_bufs; i++) {
927 buf_idx = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
928 *buf_idx *= 2; /* 8 bytes per buf */
929
930 if (DRM_COPY_TO_USER(ref_age_base + *buf_idx,
931 &dev_priv->scratch_ages[header.scratch.reg],
932 sizeof(u32)))
933 return -EINVAL;
934
935 if (DRM_COPY_FROM_USER(&h_pending,
936 ref_age_base + *buf_idx + 1,
937 sizeof(u32)))
938 return -EINVAL;
939
940 if (h_pending == 0)
941 return -EINVAL;
942
943 h_pending--;
944
945 if (DRM_COPY_TO_USER(ref_age_base + *buf_idx + 1,
946 &h_pending,
947 sizeof(u32)))
948 return -EINVAL;
949
950 drm_buffer_advance(cmdbuf->buffer, sizeof(*buf_idx));
951 }
952
953 BEGIN_RING(2);
954 OUT_RING( CP_PACKET0( RADEON_SCRATCH_REG0 + header.scratch.reg * 4, 0 ) );
955 OUT_RING( dev_priv->scratch_ages[header.scratch.reg] );
956 ADVANCE_RING();
957
958 return 0;
959}
960
961/**
962 * Uploads user-supplied vertex program instructions or parameters onto
963 * the graphics card.
964 * Called by r300_do_cp_cmdbuf.
965 */
966static inline int r300_emit_r500fp(drm_radeon_private_t *dev_priv,
967 drm_radeon_kcmd_buffer_t *cmdbuf,
968 drm_r300_cmd_header_t header)
969{
970 int sz;
971 int addr;
972 int type;
973 int isclamp;
974 int stride;
975 RING_LOCALS;
976
977 sz = header.r500fp.count;
978 /* address is 9 bits 0 - 8, bit 1 of flags is part of address */
979 addr = ((header.r500fp.adrhi_flags & 1) << 8) | header.r500fp.adrlo;
980
981 type = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE);
982 isclamp = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP);
983
984 addr |= (type << 16);
985 addr |= (isclamp << 17);
986
987 stride = type ? 4 : 6;
988
989 DRM_DEBUG("r500fp %d %d type: %d\n", sz, addr, type);
990 if (!sz)
991 return 0;
992 if (sz * stride * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
993 return -EINVAL;
994
995 BEGIN_RING(3 + sz * stride);
996 OUT_RING_REG(R500_GA_US_VECTOR_INDEX, addr);
997 OUT_RING(CP_PACKET0_TABLE(R500_GA_US_VECTOR_DATA, sz * stride - 1));
998 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz * stride);
999
1000 ADVANCE_RING();
1001
1002 return 0;
1003}
1004
1005
1006/**
1007 * Parses and validates a user-supplied command buffer and emits appropriate
1008 * commands on the DMA ring buffer.
1009 * Called by the ioctl handler function radeon_cp_cmdbuf.
1010 */
1011int r300_do_cp_cmdbuf(struct drm_device *dev,
1012 struct drm_file *file_priv,
1013 drm_radeon_kcmd_buffer_t *cmdbuf)
1014{
1015 drm_radeon_private_t *dev_priv = dev->dev_private;
1016 struct drm_radeon_master_private *master_priv = file_priv->master->driver_priv;
1017 struct drm_device_dma *dma = dev->dma;
1018 struct drm_buf *buf = NULL;
1019 int emit_dispatch_age = 0;
1020 int ret = 0;
1021
1022 DRM_DEBUG("\n");
1023
1024 /* pacify */
1025 r300_pacify(dev_priv);
1026
1027 if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
1028 ret = r300_emit_cliprects(dev_priv, cmdbuf, 0);
1029 if (ret)
1030 goto cleanup;
1031 }
1032
1033 while (drm_buffer_unprocessed(cmdbuf->buffer)
1034 >= sizeof(drm_r300_cmd_header_t)) {
1035 int idx;
1036 drm_r300_cmd_header_t *header, stack_header;
1037
1038 header = drm_buffer_read_object(cmdbuf->buffer,
1039 sizeof(stack_header), &stack_header);
1040
1041 switch (header->header.cmd_type) {
1042 case R300_CMD_PACKET0:
1043 DRM_DEBUG("R300_CMD_PACKET0\n");
1044 ret = r300_emit_packet0(dev_priv, cmdbuf, *header);
1045 if (ret) {
1046 DRM_ERROR("r300_emit_packet0 failed\n");
1047 goto cleanup;
1048 }
1049 break;
1050
1051 case R300_CMD_VPU:
1052 DRM_DEBUG("R300_CMD_VPU\n");
1053 ret = r300_emit_vpu(dev_priv, cmdbuf, *header);
1054 if (ret) {
1055 DRM_ERROR("r300_emit_vpu failed\n");
1056 goto cleanup;
1057 }
1058 break;
1059
1060 case R300_CMD_PACKET3:
1061 DRM_DEBUG("R300_CMD_PACKET3\n");
1062 ret = r300_emit_packet3(dev_priv, cmdbuf, *header);
1063 if (ret) {
1064 DRM_ERROR("r300_emit_packet3 failed\n");
1065 goto cleanup;
1066 }
1067 break;
1068
1069 case R300_CMD_END3D:
1070 DRM_DEBUG("R300_CMD_END3D\n");
1071 /* TODO:
1072 Ideally userspace driver should not need to issue this call,
1073 i.e. the drm driver should issue it automatically and prevent
1074 lockups.
1075
1076 In practice, we do not understand why this call is needed and what
1077 it does (except for some vague guesses that it has to do with cache
1078 coherence) and so the user space driver does it.
1079
1080 Once we are sure which uses prevent lockups the code could be moved
1081 into the kernel and the userspace driver will not
1082 need to use this command.
1083
1084 Note that issuing this command does not hurt anything
1085 except, possibly, performance */
1086 r300_pacify(dev_priv);
1087 break;
1088
1089 case R300_CMD_CP_DELAY:
1090 /* simple enough, we can do it here */
1091 DRM_DEBUG("R300_CMD_CP_DELAY\n");
1092 {
1093 int i;
1094 RING_LOCALS;
1095
1096 BEGIN_RING(header->delay.count);
1097 for (i = 0; i < header->delay.count; i++)
1098 OUT_RING(RADEON_CP_PACKET2);
1099 ADVANCE_RING();
1100 }
1101 break;
1102
1103 case R300_CMD_DMA_DISCARD:
1104 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
1105 idx = header->dma.buf_idx;
1106 if (idx < 0 || idx >= dma->buf_count) {
1107 DRM_ERROR("buffer index %d (of %d max)\n",
1108 idx, dma->buf_count - 1);
1109 ret = -EINVAL;
1110 goto cleanup;
1111 }
1112
1113 buf = dma->buflist[idx];
1114 if (buf->file_priv != file_priv || buf->pending) {
1115 DRM_ERROR("bad buffer %p %p %d\n",
1116 buf->file_priv, file_priv,
1117 buf->pending);
1118 ret = -EINVAL;
1119 goto cleanup;
1120 }
1121
1122 emit_dispatch_age = 1;
1123 r300_discard_buffer(dev, file_priv->master, buf);
1124 break;
1125
1126 case R300_CMD_WAIT:
1127 DRM_DEBUG("R300_CMD_WAIT\n");
1128 r300_cmd_wait(dev_priv, *header);
1129 break;
1130
1131 case R300_CMD_SCRATCH:
1132 DRM_DEBUG("R300_CMD_SCRATCH\n");
1133 ret = r300_scratch(dev_priv, cmdbuf, *header);
1134 if (ret) {
1135 DRM_ERROR("r300_scratch failed\n");
1136 goto cleanup;
1137 }
1138 break;
1139
1140 case R300_CMD_R500FP:
1141 if ((dev_priv->flags & RADEON_FAMILY_MASK) < CHIP_RV515) {
1142 DRM_ERROR("Calling r500 command on r300 card\n");
1143 ret = -EINVAL;
1144 goto cleanup;
1145 }
1146 DRM_DEBUG("R300_CMD_R500FP\n");
1147 ret = r300_emit_r500fp(dev_priv, cmdbuf, *header);
1148 if (ret) {
1149 DRM_ERROR("r300_emit_r500fp failed\n");
1150 goto cleanup;
1151 }
1152 break;
1153 default:
1154 DRM_ERROR("bad cmd_type %i at byte %d\n",
1155 header->header.cmd_type,
1156 cmdbuf->buffer->iterator - (int)sizeof(*header));
1157 ret = -EINVAL;
1158 goto cleanup;
1159 }
1160 }
1161
1162 DRM_DEBUG("END\n");
1163
1164 cleanup:
1165 r300_pacify(dev_priv);
1166
1167 /* We emit the vertex buffer age here, outside the pacifier "brackets"
1168 * for two reasons:
1169 * (1) This may coalesce multiple age emissions into a single one and
1170 * (2) more importantly, some chips lock up hard when scratch registers
1171 * are written inside the pacifier bracket.
1172 */
1173 if (emit_dispatch_age) {
1174 RING_LOCALS;
1175
1176 /* Emit the vertex buffer age */
1177 BEGIN_RING(2);
1178 RADEON_DISPATCH_AGE(master_priv->sarea_priv->last_dispatch);
1179 ADVANCE_RING();
1180 }
1181
1182 COMMIT_RING();
1183
1184 return ret;
1185}
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 * ------------------------ This file is DEPRECATED! -------------------------
34 */
35
36#include <drm/drmP.h>
37#include <drm/drm_buffer.h>
38#include <drm/radeon_drm.h>
39#include "radeon_drv.h"
40#include "r300_reg.h"
41
42#include <asm/unaligned.h>
43
44#define R300_SIMULTANEOUS_CLIPRECTS 4
45
46/* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
47 */
48static const int r300_cliprect_cntl[4] = {
49 0xAAAA,
50 0xEEEE,
51 0xFEFE,
52 0xFFFE
53};
54
55/**
56 * Emit up to R300_SIMULTANEOUS_CLIPRECTS cliprects from the given command
57 * buffer, starting with index n.
58 */
59static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
60 drm_radeon_kcmd_buffer_t *cmdbuf, int n)
61{
62 struct drm_clip_rect box;
63 int nr;
64 int i;
65 RING_LOCALS;
66
67 nr = cmdbuf->nbox - n;
68 if (nr > R300_SIMULTANEOUS_CLIPRECTS)
69 nr = R300_SIMULTANEOUS_CLIPRECTS;
70
71 DRM_DEBUG("%i cliprects\n", nr);
72
73 if (nr) {
74 BEGIN_RING(6 + nr * 2);
75 OUT_RING(CP_PACKET0(R300_RE_CLIPRECT_TL_0, nr * 2 - 1));
76
77 for (i = 0; i < nr; ++i) {
78 if (copy_from_user
79 (&box, &cmdbuf->boxes[n + i], sizeof(box))) {
80 DRM_ERROR("copy cliprect faulted\n");
81 return -EFAULT;
82 }
83
84 box.x2--; /* Hardware expects inclusive bottom-right corner */
85 box.y2--;
86
87 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
88 box.x1 = (box.x1) &
89 R300_CLIPRECT_MASK;
90 box.y1 = (box.y1) &
91 R300_CLIPRECT_MASK;
92 box.x2 = (box.x2) &
93 R300_CLIPRECT_MASK;
94 box.y2 = (box.y2) &
95 R300_CLIPRECT_MASK;
96 } else {
97 box.x1 = (box.x1 + R300_CLIPRECT_OFFSET) &
98 R300_CLIPRECT_MASK;
99 box.y1 = (box.y1 + R300_CLIPRECT_OFFSET) &
100 R300_CLIPRECT_MASK;
101 box.x2 = (box.x2 + R300_CLIPRECT_OFFSET) &
102 R300_CLIPRECT_MASK;
103 box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) &
104 R300_CLIPRECT_MASK;
105 }
106
107 OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
108 (box.y1 << R300_CLIPRECT_Y_SHIFT));
109 OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
110 (box.y2 << R300_CLIPRECT_Y_SHIFT));
111
112 }
113
114 OUT_RING_REG(R300_RE_CLIPRECT_CNTL, r300_cliprect_cntl[nr - 1]);
115
116 /* TODO/SECURITY: Force scissors to a safe value, otherwise the
117 * client might be able to trample over memory.
118 * The impact should be very limited, but I'd rather be safe than
119 * sorry.
120 */
121 OUT_RING(CP_PACKET0(R300_RE_SCISSORS_TL, 1));
122 OUT_RING(0);
123 OUT_RING(R300_SCISSORS_X_MASK | R300_SCISSORS_Y_MASK);
124 ADVANCE_RING();
125 } else {
126 /* Why we allow zero cliprect rendering:
127 * There are some commands in a command buffer that must be submitted
128 * even when there are no cliprects, e.g. DMA buffer discard
129 * or state setting (though state setting could be avoided by
130 * simulating a loss of context).
131 *
132 * Now since the cmdbuf interface is so chaotic right now (and is
133 * bound to remain that way for a bit until things settle down),
134 * it is basically impossible to filter out the commands that are
135 * necessary and those that aren't.
136 *
137 * So I choose the safe way and don't do any filtering at all;
138 * instead, I simply set up the engine so that all rendering
139 * can't produce any fragments.
140 */
141 BEGIN_RING(2);
142 OUT_RING_REG(R300_RE_CLIPRECT_CNTL, 0);
143 ADVANCE_RING();
144 }
145
146 /* flus cache and wait idle clean after cliprect change */
147 BEGIN_RING(2);
148 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
149 OUT_RING(R300_RB3D_DC_FLUSH);
150 ADVANCE_RING();
151 BEGIN_RING(2);
152 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
153 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
154 ADVANCE_RING();
155 /* set flush flag */
156 dev_priv->track_flush |= RADEON_FLUSH_EMITED;
157
158 return 0;
159}
160
161static u8 r300_reg_flags[0x10000 >> 2];
162
163void r300_init_reg_flags(struct drm_device *dev)
164{
165 int i;
166 drm_radeon_private_t *dev_priv = dev->dev_private;
167
168 memset(r300_reg_flags, 0, 0x10000 >> 2);
169#define ADD_RANGE_MARK(reg, count,mark) \
170 for(i=((reg)>>2);i<((reg)>>2)+(count);i++)\
171 r300_reg_flags[i]|=(mark);
172
173#define MARK_SAFE 1
174#define MARK_CHECK_OFFSET 2
175
176#define ADD_RANGE(reg, count) ADD_RANGE_MARK(reg, count, MARK_SAFE)
177
178 /* these match cmducs() command in r300_driver/r300/r300_cmdbuf.c */
179 ADD_RANGE(R300_SE_VPORT_XSCALE, 6);
180 ADD_RANGE(R300_VAP_CNTL, 1);
181 ADD_RANGE(R300_SE_VTE_CNTL, 2);
182 ADD_RANGE(0x2134, 2);
183 ADD_RANGE(R300_VAP_CNTL_STATUS, 1);
184 ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
185 ADD_RANGE(0x21DC, 1);
186 ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
187 ADD_RANGE(R300_VAP_CLIP_X_0, 4);
188 ADD_RANGE(R300_VAP_PVS_STATE_FLUSH_REG, 1);
189 ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
190 ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
191 ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
192 ADD_RANGE(R300_GB_ENABLE, 1);
193 ADD_RANGE(R300_GB_MSPOS0, 5);
194 ADD_RANGE(R300_TX_INVALTAGS, 1);
195 ADD_RANGE(R300_TX_ENABLE, 1);
196 ADD_RANGE(0x4200, 4);
197 ADD_RANGE(0x4214, 1);
198 ADD_RANGE(R300_RE_POINTSIZE, 1);
199 ADD_RANGE(0x4230, 3);
200 ADD_RANGE(R300_RE_LINE_CNT, 1);
201 ADD_RANGE(R300_RE_UNK4238, 1);
202 ADD_RANGE(0x4260, 3);
203 ADD_RANGE(R300_RE_SHADE, 4);
204 ADD_RANGE(R300_RE_POLYGON_MODE, 5);
205 ADD_RANGE(R300_RE_ZBIAS_CNTL, 1);
206 ADD_RANGE(R300_RE_ZBIAS_T_FACTOR, 4);
207 ADD_RANGE(R300_RE_OCCLUSION_CNTL, 1);
208 ADD_RANGE(R300_RE_CULL_CNTL, 1);
209 ADD_RANGE(0x42C0, 2);
210 ADD_RANGE(R300_RS_CNTL_0, 2);
211
212 ADD_RANGE(R300_SU_REG_DEST, 1);
213 if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV530)
214 ADD_RANGE(RV530_FG_ZBREG_DEST, 1);
215
216 ADD_RANGE(R300_SC_HYPERZ, 2);
217 ADD_RANGE(0x43E8, 1);
218
219 ADD_RANGE(0x46A4, 5);
220
221 ADD_RANGE(R300_RE_FOG_STATE, 1);
222 ADD_RANGE(R300_FOG_COLOR_R, 3);
223 ADD_RANGE(R300_PP_ALPHA_TEST, 2);
224 ADD_RANGE(0x4BD8, 1);
225 ADD_RANGE(R300_PFS_PARAM_0_X, 64);
226 ADD_RANGE(0x4E00, 1);
227 ADD_RANGE(R300_RB3D_CBLEND, 2);
228 ADD_RANGE(R300_RB3D_COLORMASK, 1);
229 ADD_RANGE(R300_RB3D_BLEND_COLOR, 3);
230 ADD_RANGE_MARK(R300_RB3D_COLOROFFSET0, 1, MARK_CHECK_OFFSET); /* check offset */
231 ADD_RANGE(R300_RB3D_COLORPITCH0, 1);
232 ADD_RANGE(0x4E50, 9);
233 ADD_RANGE(0x4E88, 1);
234 ADD_RANGE(0x4EA0, 2);
235 ADD_RANGE(R300_ZB_CNTL, 3);
236 ADD_RANGE(R300_ZB_FORMAT, 4);
237 ADD_RANGE_MARK(R300_ZB_DEPTHOFFSET, 1, MARK_CHECK_OFFSET); /* check offset */
238 ADD_RANGE(R300_ZB_DEPTHPITCH, 1);
239 ADD_RANGE(R300_ZB_DEPTHCLEARVALUE, 1);
240 ADD_RANGE(R300_ZB_ZMASK_OFFSET, 13);
241 ADD_RANGE(R300_ZB_ZPASS_DATA, 2); /* ZB_ZPASS_DATA, ZB_ZPASS_ADDR */
242
243 ADD_RANGE(R300_TX_FILTER_0, 16);
244 ADD_RANGE(R300_TX_FILTER1_0, 16);
245 ADD_RANGE(R300_TX_SIZE_0, 16);
246 ADD_RANGE(R300_TX_FORMAT_0, 16);
247 ADD_RANGE(R300_TX_PITCH_0, 16);
248 /* Texture offset is dangerous and needs more checking */
249 ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET);
250 ADD_RANGE(R300_TX_CHROMA_KEY_0, 16);
251 ADD_RANGE(R300_TX_BORDER_COLOR_0, 16);
252
253 /* Sporadic registers used as primitives are emitted */
254 ADD_RANGE(R300_ZB_ZCACHE_CTLSTAT, 1);
255 ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1);
256 ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8);
257 ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8);
258
259 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
260 ADD_RANGE(R500_VAP_INDEX_OFFSET, 1);
261 ADD_RANGE(R500_US_CONFIG, 2);
262 ADD_RANGE(R500_US_CODE_ADDR, 3);
263 ADD_RANGE(R500_US_FC_CTRL, 1);
264 ADD_RANGE(R500_RS_IP_0, 16);
265 ADD_RANGE(R500_RS_INST_0, 16);
266 ADD_RANGE(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2);
267 ADD_RANGE(R500_RB3D_CONSTANT_COLOR_AR, 2);
268 ADD_RANGE(R500_ZB_FIFO_SIZE, 2);
269 } else {
270 ADD_RANGE(R300_PFS_CNTL_0, 3);
271 ADD_RANGE(R300_PFS_NODE_0, 4);
272 ADD_RANGE(R300_PFS_TEXI_0, 64);
273 ADD_RANGE(R300_PFS_INSTR0_0, 64);
274 ADD_RANGE(R300_PFS_INSTR1_0, 64);
275 ADD_RANGE(R300_PFS_INSTR2_0, 64);
276 ADD_RANGE(R300_PFS_INSTR3_0, 64);
277 ADD_RANGE(R300_RS_INTERP_0, 8);
278 ADD_RANGE(R300_RS_ROUTE_0, 8);
279
280 }
281}
282
283static __inline__ int r300_check_range(unsigned reg, int count)
284{
285 int i;
286 if (reg & ~0xffff)
287 return -1;
288 for (i = (reg >> 2); i < (reg >> 2) + count; i++)
289 if (r300_reg_flags[i] != MARK_SAFE)
290 return 1;
291 return 0;
292}
293
294static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t *
295 dev_priv,
296 drm_radeon_kcmd_buffer_t
297 * cmdbuf,
298 drm_r300_cmd_header_t
299 header)
300{
301 int reg;
302 int sz;
303 int i;
304 u32 *value;
305 RING_LOCALS;
306
307 sz = header.packet0.count;
308 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
309
310 if ((sz > 64) || (sz < 0)) {
311 DRM_ERROR("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n",
312 reg, sz);
313 return -EINVAL;
314 }
315
316 for (i = 0; i < sz; i++) {
317 switch (r300_reg_flags[(reg >> 2) + i]) {
318 case MARK_SAFE:
319 break;
320 case MARK_CHECK_OFFSET:
321 value = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
322 if (!radeon_check_offset(dev_priv, *value)) {
323 DRM_ERROR("Offset failed range check (reg=%04x sz=%d)\n",
324 reg, sz);
325 return -EINVAL;
326 }
327 break;
328 default:
329 DRM_ERROR("Register %04x failed check as flag=%02x\n",
330 reg + i * 4, r300_reg_flags[(reg >> 2) + i]);
331 return -EINVAL;
332 }
333 }
334
335 BEGIN_RING(1 + sz);
336 OUT_RING(CP_PACKET0(reg, sz - 1));
337 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
338 ADVANCE_RING();
339
340 return 0;
341}
342
343/**
344 * Emits a packet0 setting arbitrary registers.
345 * Called by r300_do_cp_cmdbuf.
346 *
347 * Note that checks are performed on contents and addresses of the registers
348 */
349static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv,
350 drm_radeon_kcmd_buffer_t *cmdbuf,
351 drm_r300_cmd_header_t header)
352{
353 int reg;
354 int sz;
355 RING_LOCALS;
356
357 sz = header.packet0.count;
358 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
359
360 if (!sz)
361 return 0;
362
363 if (sz * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
364 return -EINVAL;
365
366 if (reg + sz * 4 >= 0x10000) {
367 DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg,
368 sz);
369 return -EINVAL;
370 }
371
372 if (r300_check_range(reg, sz)) {
373 /* go and check everything */
374 return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf,
375 header);
376 }
377 /* the rest of the data is safe to emit, whatever the values the user passed */
378
379 BEGIN_RING(1 + sz);
380 OUT_RING(CP_PACKET0(reg, sz - 1));
381 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
382 ADVANCE_RING();
383
384 return 0;
385}
386
387/**
388 * Uploads user-supplied vertex program instructions or parameters onto
389 * the graphics card.
390 * Called by r300_do_cp_cmdbuf.
391 */
392static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
393 drm_radeon_kcmd_buffer_t *cmdbuf,
394 drm_r300_cmd_header_t header)
395{
396 int sz;
397 int addr;
398 RING_LOCALS;
399
400 sz = header.vpu.count;
401 addr = (header.vpu.adrhi << 8) | header.vpu.adrlo;
402
403 if (!sz)
404 return 0;
405 if (sz * 16 > drm_buffer_unprocessed(cmdbuf->buffer))
406 return -EINVAL;
407
408 /* VAP is very sensitive so we purge cache before we program it
409 * and we also flush its state before & after */
410 BEGIN_RING(6);
411 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
412 OUT_RING(R300_RB3D_DC_FLUSH);
413 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
414 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
415 OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
416 OUT_RING(0);
417 ADVANCE_RING();
418 /* set flush flag */
419 dev_priv->track_flush |= RADEON_FLUSH_EMITED;
420
421 BEGIN_RING(3 + sz * 4);
422 OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
423 OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
424 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz * 4);
425 ADVANCE_RING();
426
427 BEGIN_RING(2);
428 OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
429 OUT_RING(0);
430 ADVANCE_RING();
431
432 return 0;
433}
434
435/**
436 * Emit a clear packet from userspace.
437 * Called by r300_emit_packet3.
438 */
439static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
440 drm_radeon_kcmd_buffer_t *cmdbuf)
441{
442 RING_LOCALS;
443
444 if (8 * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
445 return -EINVAL;
446
447 BEGIN_RING(10);
448 OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8));
449 OUT_RING(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
450 (1 << R300_PRIM_NUM_VERTICES_SHIFT));
451 OUT_RING_DRM_BUFFER(cmdbuf->buffer, 8);
452 ADVANCE_RING();
453
454 BEGIN_RING(4);
455 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
456 OUT_RING(R300_RB3D_DC_FLUSH);
457 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
458 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
459 ADVANCE_RING();
460 /* set flush flag */
461 dev_priv->track_flush |= RADEON_FLUSH_EMITED;
462
463 return 0;
464}
465
466static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
467 drm_radeon_kcmd_buffer_t *cmdbuf,
468 u32 header)
469{
470 int count, i, k;
471#define MAX_ARRAY_PACKET 64
472 u32 *data;
473 u32 narrays;
474 RING_LOCALS;
475
476 count = (header & RADEON_CP_PACKET_COUNT_MASK) >> 16;
477
478 if ((count + 1) > MAX_ARRAY_PACKET) {
479 DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
480 count);
481 return -EINVAL;
482 }
483 /* carefully check packet contents */
484
485 /* We have already read the header so advance the buffer. */
486 drm_buffer_advance(cmdbuf->buffer, 4);
487
488 narrays = *(u32 *)drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
489 k = 0;
490 i = 1;
491 while ((k < narrays) && (i < (count + 1))) {
492 i++; /* skip attribute field */
493 data = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
494 if (!radeon_check_offset(dev_priv, *data)) {
495 DRM_ERROR
496 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
497 k, i);
498 return -EINVAL;
499 }
500 k++;
501 i++;
502 if (k == narrays)
503 break;
504 /* have one more to process, they come in pairs */
505 data = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
506 if (!radeon_check_offset(dev_priv, *data)) {
507 DRM_ERROR
508 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
509 k, i);
510 return -EINVAL;
511 }
512 k++;
513 i++;
514 }
515 /* do the counts match what we expect ? */
516 if ((k != narrays) || (i != (count + 1))) {
517 DRM_ERROR
518 ("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n",
519 k, i, narrays, count + 1);
520 return -EINVAL;
521 }
522
523 /* all clear, output packet */
524
525 BEGIN_RING(count + 2);
526 OUT_RING(header);
527 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 1);
528 ADVANCE_RING();
529
530 return 0;
531}
532
533static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
534 drm_radeon_kcmd_buffer_t *cmdbuf)
535{
536 u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
537 int count, ret;
538 RING_LOCALS;
539
540
541 count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
542
543 if (*cmd & 0x8000) {
544 u32 offset;
545 u32 *cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
546 if (*cmd1 & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
547 | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
548
549 u32 *cmd2 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
550 offset = *cmd2 << 10;
551 ret = !radeon_check_offset(dev_priv, offset);
552 if (ret) {
553 DRM_ERROR("Invalid bitblt first offset is %08X\n", offset);
554 return -EINVAL;
555 }
556 }
557
558 if ((*cmd1 & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
559 (*cmd1 & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
560 u32 *cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
561 offset = *cmd3 << 10;
562 ret = !radeon_check_offset(dev_priv, offset);
563 if (ret) {
564 DRM_ERROR("Invalid bitblt second offset is %08X\n", offset);
565 return -EINVAL;
566 }
567
568 }
569 }
570
571 BEGIN_RING(count+2);
572 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
573 ADVANCE_RING();
574
575 return 0;
576}
577
578static __inline__ int r300_emit_draw_indx_2(drm_radeon_private_t *dev_priv,
579 drm_radeon_kcmd_buffer_t *cmdbuf)
580{
581 u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
582 u32 *cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
583 int count;
584 int expected_count;
585 RING_LOCALS;
586
587 count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
588
589 expected_count = *cmd1 >> 16;
590 if (!(*cmd1 & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
591 expected_count = (expected_count+1)/2;
592
593 if (count && count != expected_count) {
594 DRM_ERROR("3D_DRAW_INDX_2: packet size %i, expected %i\n",
595 count, expected_count);
596 return -EINVAL;
597 }
598
599 BEGIN_RING(count+2);
600 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
601 ADVANCE_RING();
602
603 if (!count) {
604 drm_r300_cmd_header_t stack_header, *header;
605 u32 *cmd1, *cmd2, *cmd3;
606
607 if (drm_buffer_unprocessed(cmdbuf->buffer)
608 < 4*4 + sizeof(stack_header)) {
609 DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER, but stream is too short.\n");
610 return -EINVAL;
611 }
612
613 header = drm_buffer_read_object(cmdbuf->buffer,
614 sizeof(stack_header), &stack_header);
615
616 cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
617 cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
618 cmd2 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
619 cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
620
621 if (header->header.cmd_type != R300_CMD_PACKET3 ||
622 header->packet3.packet != R300_CMD_PACKET3_RAW ||
623 *cmd != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) {
624 DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER.\n");
625 return -EINVAL;
626 }
627
628 if ((*cmd1 & 0x8000ffff) != 0x80000810) {
629 DRM_ERROR("Invalid indx_buffer reg address %08X\n",
630 *cmd1);
631 return -EINVAL;
632 }
633 if (!radeon_check_offset(dev_priv, *cmd2)) {
634 DRM_ERROR("Invalid indx_buffer offset is %08X\n",
635 *cmd2);
636 return -EINVAL;
637 }
638 if (*cmd3 != expected_count) {
639 DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n",
640 *cmd3, expected_count);
641 return -EINVAL;
642 }
643
644 BEGIN_RING(4);
645 OUT_RING_DRM_BUFFER(cmdbuf->buffer, 4);
646 ADVANCE_RING();
647 }
648
649 return 0;
650}
651
652static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
653 drm_radeon_kcmd_buffer_t *cmdbuf)
654{
655 u32 *header;
656 int count;
657 RING_LOCALS;
658
659 if (4 > drm_buffer_unprocessed(cmdbuf->buffer))
660 return -EINVAL;
661
662 /* Fixme !! This simply emits a packet without much checking.
663 We need to be smarter. */
664
665 /* obtain first word - actual packet3 header */
666 header = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
667
668 /* Is it packet 3 ? */
669 if ((*header >> 30) != 0x3) {
670 DRM_ERROR("Not a packet3 header (0x%08x)\n", *header);
671 return -EINVAL;
672 }
673
674 count = (*header >> 16) & 0x3fff;
675
676 /* Check again now that we know how much data to expect */
677 if ((count + 2) * 4 > drm_buffer_unprocessed(cmdbuf->buffer)) {
678 DRM_ERROR
679 ("Expected packet3 of length %d but have only %d bytes left\n",
680 (count + 2) * 4, drm_buffer_unprocessed(cmdbuf->buffer));
681 return -EINVAL;
682 }
683
684 /* Is it a packet type we know about ? */
685 switch (*header & 0xff00) {
686 case RADEON_3D_LOAD_VBPNTR: /* load vertex array pointers */
687 return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, *header);
688
689 case RADEON_CNTL_BITBLT_MULTI:
690 return r300_emit_bitblt_multi(dev_priv, cmdbuf);
691
692 case RADEON_CP_INDX_BUFFER:
693 DRM_ERROR("packet3 INDX_BUFFER without preceding 3D_DRAW_INDX_2 is illegal.\n");
694 return -EINVAL;
695 case RADEON_CP_3D_DRAW_IMMD_2:
696 /* triggers drawing using in-packet vertex data */
697 case RADEON_CP_3D_DRAW_VBUF_2:
698 /* triggers drawing of vertex buffers setup elsewhere */
699 dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
700 RADEON_PURGE_EMITED);
701 break;
702 case RADEON_CP_3D_DRAW_INDX_2:
703 /* triggers drawing using indices to vertex buffer */
704 /* whenever we send vertex we clear flush & purge */
705 dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
706 RADEON_PURGE_EMITED);
707 return r300_emit_draw_indx_2(dev_priv, cmdbuf);
708 case RADEON_WAIT_FOR_IDLE:
709 case RADEON_CP_NOP:
710 /* these packets are safe */
711 break;
712 default:
713 DRM_ERROR("Unknown packet3 header (0x%08x)\n", *header);
714 return -EINVAL;
715 }
716
717 BEGIN_RING(count + 2);
718 OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
719 ADVANCE_RING();
720
721 return 0;
722}
723
724/**
725 * Emit a rendering packet3 from userspace.
726 * Called by r300_do_cp_cmdbuf.
727 */
728static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
729 drm_radeon_kcmd_buffer_t *cmdbuf,
730 drm_r300_cmd_header_t header)
731{
732 int n;
733 int ret;
734 int orig_iter = cmdbuf->buffer->iterator;
735
736 /* This is a do-while-loop so that we run the interior at least once,
737 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
738 */
739 n = 0;
740 do {
741 if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) {
742 ret = r300_emit_cliprects(dev_priv, cmdbuf, n);
743 if (ret)
744 return ret;
745
746 cmdbuf->buffer->iterator = orig_iter;
747 }
748
749 switch (header.packet3.packet) {
750 case R300_CMD_PACKET3_CLEAR:
751 DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n");
752 ret = r300_emit_clear(dev_priv, cmdbuf);
753 if (ret) {
754 DRM_ERROR("r300_emit_clear failed\n");
755 return ret;
756 }
757 break;
758
759 case R300_CMD_PACKET3_RAW:
760 DRM_DEBUG("R300_CMD_PACKET3_RAW\n");
761 ret = r300_emit_raw_packet3(dev_priv, cmdbuf);
762 if (ret) {
763 DRM_ERROR("r300_emit_raw_packet3 failed\n");
764 return ret;
765 }
766 break;
767
768 default:
769 DRM_ERROR("bad packet3 type %i at byte %d\n",
770 header.packet3.packet,
771 cmdbuf->buffer->iterator - (int)sizeof(header));
772 return -EINVAL;
773 }
774
775 n += R300_SIMULTANEOUS_CLIPRECTS;
776 } while (n < cmdbuf->nbox);
777
778 return 0;
779}
780
781/* Some of the R300 chips seem to be extremely touchy about the two registers
782 * that are configured in r300_pacify.
783 * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace
784 * sends a command buffer that contains only state setting commands and a
785 * vertex program/parameter upload sequence, this will eventually lead to a
786 * lockup, unless the sequence is bracketed by calls to r300_pacify.
787 * So we should take great care to *always* call r300_pacify before
788 * *anything* 3D related, and again afterwards. This is what the
789 * call bracket in r300_do_cp_cmdbuf is for.
790 */
791
792/**
793 * Emit the sequence to pacify R300.
794 */
795static void r300_pacify(drm_radeon_private_t *dev_priv)
796{
797 uint32_t cache_z, cache_3d, cache_2d;
798 RING_LOCALS;
799
800 cache_z = R300_ZC_FLUSH;
801 cache_2d = R300_RB2D_DC_FLUSH;
802 cache_3d = R300_RB3D_DC_FLUSH;
803 if (!(dev_priv->track_flush & RADEON_PURGE_EMITED)) {
804 /* we can purge, primitive where draw since last purge */
805 cache_z |= R300_ZC_FREE;
806 cache_2d |= R300_RB2D_DC_FREE;
807 cache_3d |= R300_RB3D_DC_FREE;
808 }
809
810 /* flush & purge zbuffer */
811 BEGIN_RING(2);
812 OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));
813 OUT_RING(cache_z);
814 ADVANCE_RING();
815 /* flush & purge 3d */
816 BEGIN_RING(2);
817 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
818 OUT_RING(cache_3d);
819 ADVANCE_RING();
820 /* flush & purge texture */
821 BEGIN_RING(2);
822 OUT_RING(CP_PACKET0(R300_TX_INVALTAGS, 0));
823 OUT_RING(0);
824 ADVANCE_RING();
825 /* FIXME: is this one really needed ? */
826 BEGIN_RING(2);
827 OUT_RING(CP_PACKET0(R300_RB3D_AARESOLVE_CTL, 0));
828 OUT_RING(0);
829 ADVANCE_RING();
830 BEGIN_RING(2);
831 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
832 OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
833 ADVANCE_RING();
834 /* flush & purge 2d through E2 as RB2D will trigger lockup */
835 BEGIN_RING(4);
836 OUT_RING(CP_PACKET0(R300_DSTCACHE_CTLSTAT, 0));
837 OUT_RING(cache_2d);
838 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
839 OUT_RING(RADEON_WAIT_2D_IDLECLEAN |
840 RADEON_WAIT_HOST_IDLECLEAN);
841 ADVANCE_RING();
842 /* set flush & purge flags */
843 dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED;
844}
845
846/**
847 * Called by r300_do_cp_cmdbuf to update the internal buffer age and state.
848 * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
849 * be careful about how this function is called.
850 */
851static void r300_discard_buffer(struct drm_device *dev, struct drm_master *master, struct drm_buf *buf)
852{
853 drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
854 struct drm_radeon_master_private *master_priv = master->driver_priv;
855
856 buf_priv->age = ++master_priv->sarea_priv->last_dispatch;
857 buf->pending = 1;
858 buf->used = 0;
859}
860
861static void r300_cmd_wait(drm_radeon_private_t * dev_priv,
862 drm_r300_cmd_header_t header)
863{
864 u32 wait_until;
865 RING_LOCALS;
866
867 if (!header.wait.flags)
868 return;
869
870 wait_until = 0;
871
872 switch(header.wait.flags) {
873 case R300_WAIT_2D:
874 wait_until = RADEON_WAIT_2D_IDLE;
875 break;
876 case R300_WAIT_3D:
877 wait_until = RADEON_WAIT_3D_IDLE;
878 break;
879 case R300_NEW_WAIT_2D_3D:
880 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_3D_IDLE;
881 break;
882 case R300_NEW_WAIT_2D_2D_CLEAN:
883 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN;
884 break;
885 case R300_NEW_WAIT_3D_3D_CLEAN:
886 wait_until = RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN;
887 break;
888 case R300_NEW_WAIT_2D_2D_CLEAN_3D_3D_CLEAN:
889 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN;
890 wait_until |= RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN;
891 break;
892 default:
893 return;
894 }
895
896 BEGIN_RING(2);
897 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
898 OUT_RING(wait_until);
899 ADVANCE_RING();
900}
901
902static int r300_scratch(drm_radeon_private_t *dev_priv,
903 drm_radeon_kcmd_buffer_t *cmdbuf,
904 drm_r300_cmd_header_t header)
905{
906 u32 *ref_age_base;
907 u32 i, *buf_idx, h_pending;
908 u64 *ptr_addr;
909 u64 stack_ptr_addr;
910 RING_LOCALS;
911
912 if (drm_buffer_unprocessed(cmdbuf->buffer) <
913 (sizeof(u64) + header.scratch.n_bufs * sizeof(*buf_idx))) {
914 return -EINVAL;
915 }
916
917 if (header.scratch.reg >= 5) {
918 return -EINVAL;
919 }
920
921 dev_priv->scratch_ages[header.scratch.reg]++;
922
923 ptr_addr = drm_buffer_read_object(cmdbuf->buffer,
924 sizeof(stack_ptr_addr), &stack_ptr_addr);
925 ref_age_base = (u32 *)(unsigned long)get_unaligned(ptr_addr);
926
927 for (i=0; i < header.scratch.n_bufs; i++) {
928 buf_idx = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
929 *buf_idx *= 2; /* 8 bytes per buf */
930
931 if (copy_to_user(ref_age_base + *buf_idx,
932 &dev_priv->scratch_ages[header.scratch.reg],
933 sizeof(u32)))
934 return -EINVAL;
935
936 if (copy_from_user(&h_pending,
937 ref_age_base + *buf_idx + 1,
938 sizeof(u32)))
939 return -EINVAL;
940
941 if (h_pending == 0)
942 return -EINVAL;
943
944 h_pending--;
945
946 if (copy_to_user(ref_age_base + *buf_idx + 1,
947 &h_pending,
948 sizeof(u32)))
949 return -EINVAL;
950
951 drm_buffer_advance(cmdbuf->buffer, sizeof(*buf_idx));
952 }
953
954 BEGIN_RING(2);
955 OUT_RING( CP_PACKET0( RADEON_SCRATCH_REG0 + header.scratch.reg * 4, 0 ) );
956 OUT_RING( dev_priv->scratch_ages[header.scratch.reg] );
957 ADVANCE_RING();
958
959 return 0;
960}
961
962/**
963 * Uploads user-supplied vertex program instructions or parameters onto
964 * the graphics card.
965 * Called by r300_do_cp_cmdbuf.
966 */
967static inline int r300_emit_r500fp(drm_radeon_private_t *dev_priv,
968 drm_radeon_kcmd_buffer_t *cmdbuf,
969 drm_r300_cmd_header_t header)
970{
971 int sz;
972 int addr;
973 int type;
974 int isclamp;
975 int stride;
976 RING_LOCALS;
977
978 sz = header.r500fp.count;
979 /* address is 9 bits 0 - 8, bit 1 of flags is part of address */
980 addr = ((header.r500fp.adrhi_flags & 1) << 8) | header.r500fp.adrlo;
981
982 type = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE);
983 isclamp = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP);
984
985 addr |= (type << 16);
986 addr |= (isclamp << 17);
987
988 stride = type ? 4 : 6;
989
990 DRM_DEBUG("r500fp %d %d type: %d\n", sz, addr, type);
991 if (!sz)
992 return 0;
993 if (sz * stride * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
994 return -EINVAL;
995
996 BEGIN_RING(3 + sz * stride);
997 OUT_RING_REG(R500_GA_US_VECTOR_INDEX, addr);
998 OUT_RING(CP_PACKET0_TABLE(R500_GA_US_VECTOR_DATA, sz * stride - 1));
999 OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz * stride);
1000
1001 ADVANCE_RING();
1002
1003 return 0;
1004}
1005
1006
1007/**
1008 * Parses and validates a user-supplied command buffer and emits appropriate
1009 * commands on the DMA ring buffer.
1010 * Called by the ioctl handler function radeon_cp_cmdbuf.
1011 */
1012int r300_do_cp_cmdbuf(struct drm_device *dev,
1013 struct drm_file *file_priv,
1014 drm_radeon_kcmd_buffer_t *cmdbuf)
1015{
1016 drm_radeon_private_t *dev_priv = dev->dev_private;
1017 struct drm_radeon_master_private *master_priv = file_priv->master->driver_priv;
1018 struct drm_device_dma *dma = dev->dma;
1019 struct drm_buf *buf = NULL;
1020 int emit_dispatch_age = 0;
1021 int ret = 0;
1022
1023 DRM_DEBUG("\n");
1024
1025 /* pacify */
1026 r300_pacify(dev_priv);
1027
1028 if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
1029 ret = r300_emit_cliprects(dev_priv, cmdbuf, 0);
1030 if (ret)
1031 goto cleanup;
1032 }
1033
1034 while (drm_buffer_unprocessed(cmdbuf->buffer)
1035 >= sizeof(drm_r300_cmd_header_t)) {
1036 int idx;
1037 drm_r300_cmd_header_t *header, stack_header;
1038
1039 header = drm_buffer_read_object(cmdbuf->buffer,
1040 sizeof(stack_header), &stack_header);
1041
1042 switch (header->header.cmd_type) {
1043 case R300_CMD_PACKET0:
1044 DRM_DEBUG("R300_CMD_PACKET0\n");
1045 ret = r300_emit_packet0(dev_priv, cmdbuf, *header);
1046 if (ret) {
1047 DRM_ERROR("r300_emit_packet0 failed\n");
1048 goto cleanup;
1049 }
1050 break;
1051
1052 case R300_CMD_VPU:
1053 DRM_DEBUG("R300_CMD_VPU\n");
1054 ret = r300_emit_vpu(dev_priv, cmdbuf, *header);
1055 if (ret) {
1056 DRM_ERROR("r300_emit_vpu failed\n");
1057 goto cleanup;
1058 }
1059 break;
1060
1061 case R300_CMD_PACKET3:
1062 DRM_DEBUG("R300_CMD_PACKET3\n");
1063 ret = r300_emit_packet3(dev_priv, cmdbuf, *header);
1064 if (ret) {
1065 DRM_ERROR("r300_emit_packet3 failed\n");
1066 goto cleanup;
1067 }
1068 break;
1069
1070 case R300_CMD_END3D:
1071 DRM_DEBUG("R300_CMD_END3D\n");
1072 /* TODO:
1073 Ideally userspace driver should not need to issue this call,
1074 i.e. the drm driver should issue it automatically and prevent
1075 lockups.
1076
1077 In practice, we do not understand why this call is needed and what
1078 it does (except for some vague guesses that it has to do with cache
1079 coherence) and so the user space driver does it.
1080
1081 Once we are sure which uses prevent lockups the code could be moved
1082 into the kernel and the userspace driver will not
1083 need to use this command.
1084
1085 Note that issuing this command does not hurt anything
1086 except, possibly, performance */
1087 r300_pacify(dev_priv);
1088 break;
1089
1090 case R300_CMD_CP_DELAY:
1091 /* simple enough, we can do it here */
1092 DRM_DEBUG("R300_CMD_CP_DELAY\n");
1093 {
1094 int i;
1095 RING_LOCALS;
1096
1097 BEGIN_RING(header->delay.count);
1098 for (i = 0; i < header->delay.count; i++)
1099 OUT_RING(RADEON_CP_PACKET2);
1100 ADVANCE_RING();
1101 }
1102 break;
1103
1104 case R300_CMD_DMA_DISCARD:
1105 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
1106 idx = header->dma.buf_idx;
1107 if (idx < 0 || idx >= dma->buf_count) {
1108 DRM_ERROR("buffer index %d (of %d max)\n",
1109 idx, dma->buf_count - 1);
1110 ret = -EINVAL;
1111 goto cleanup;
1112 }
1113
1114 buf = dma->buflist[idx];
1115 if (buf->file_priv != file_priv || buf->pending) {
1116 DRM_ERROR("bad buffer %p %p %d\n",
1117 buf->file_priv, file_priv,
1118 buf->pending);
1119 ret = -EINVAL;
1120 goto cleanup;
1121 }
1122
1123 emit_dispatch_age = 1;
1124 r300_discard_buffer(dev, file_priv->master, buf);
1125 break;
1126
1127 case R300_CMD_WAIT:
1128 DRM_DEBUG("R300_CMD_WAIT\n");
1129 r300_cmd_wait(dev_priv, *header);
1130 break;
1131
1132 case R300_CMD_SCRATCH:
1133 DRM_DEBUG("R300_CMD_SCRATCH\n");
1134 ret = r300_scratch(dev_priv, cmdbuf, *header);
1135 if (ret) {
1136 DRM_ERROR("r300_scratch failed\n");
1137 goto cleanup;
1138 }
1139 break;
1140
1141 case R300_CMD_R500FP:
1142 if ((dev_priv->flags & RADEON_FAMILY_MASK) < CHIP_RV515) {
1143 DRM_ERROR("Calling r500 command on r300 card\n");
1144 ret = -EINVAL;
1145 goto cleanup;
1146 }
1147 DRM_DEBUG("R300_CMD_R500FP\n");
1148 ret = r300_emit_r500fp(dev_priv, cmdbuf, *header);
1149 if (ret) {
1150 DRM_ERROR("r300_emit_r500fp failed\n");
1151 goto cleanup;
1152 }
1153 break;
1154 default:
1155 DRM_ERROR("bad cmd_type %i at byte %d\n",
1156 header->header.cmd_type,
1157 cmdbuf->buffer->iterator - (int)sizeof(*header));
1158 ret = -EINVAL;
1159 goto cleanup;
1160 }
1161 }
1162
1163 DRM_DEBUG("END\n");
1164
1165 cleanup:
1166 r300_pacify(dev_priv);
1167
1168 /* We emit the vertex buffer age here, outside the pacifier "brackets"
1169 * for two reasons:
1170 * (1) This may coalesce multiple age emissions into a single one and
1171 * (2) more importantly, some chips lock up hard when scratch registers
1172 * are written inside the pacifier bracket.
1173 */
1174 if (emit_dispatch_age) {
1175 RING_LOCALS;
1176
1177 /* Emit the vertex buffer age */
1178 BEGIN_RING(2);
1179 RADEON_DISPATCH_AGE(master_priv->sarea_priv->last_dispatch);
1180 ADVANCE_RING();
1181 }
1182
1183 COMMIT_RING();
1184
1185 return ret;
1186}