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 <sys/cdefs.h> 35__FBSDID("$FreeBSD$"); 36 37#include "dev/drm/drmP.h" 38#include "dev/drm/drm.h" 39#include "dev/drm/radeon_drm.h" 40#include "dev/drm/radeon_drv.h" 41#include "dev/drm/r300_reg.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, 5); 240 ADD_RANGE(R300_ZB_HIZ_OFFSET, 5); 241 ADD_RANGE(R300_ZB_ZPASS_DATA, 1); 242 ADD_RANGE_MARK(R300_ZB_ZPASS_ADDR, 1, MARK_CHECK_OFFSET); /* check offset */ 243 ADD_RANGE(R300_ZB_DEPTHXY_OFFSET, 1) 244 245 ADD_RANGE(R300_TX_FILTER_0, 16); 246 ADD_RANGE(R300_TX_FILTER1_0, 16); 247 ADD_RANGE(R300_TX_SIZE_0, 16); 248 ADD_RANGE(R300_TX_FORMAT_0, 16); 249 ADD_RANGE(R300_TX_PITCH_0, 16); 250 /* Texture offset is dangerous and needs more checking */ 251 ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET); 252 ADD_RANGE(R300_TX_CHROMA_KEY_0, 16); 253 ADD_RANGE(R300_TX_BORDER_COLOR_0, 16); 254 255 /* Sporadic registers used as primitives are emitted */ 256 ADD_RANGE(R300_ZB_ZCACHE_CTLSTAT, 1); 257 ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1); 258 ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8); 259 ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8); 260 261 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) { 262 ADD_RANGE(R500_VAP_INDEX_OFFSET, 1); 263 ADD_RANGE(R500_US_CONFIG, 2); 264 ADD_RANGE(R500_US_CODE_ADDR, 3); 265 ADD_RANGE(R500_US_FC_CTRL, 1); 266 ADD_RANGE(R500_RS_IP_0, 16); 267 ADD_RANGE(R500_RS_INST_0, 16); 268 ADD_RANGE(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2); 269 ADD_RANGE(R500_RB3D_CONSTANT_COLOR_AR, 2); 270 ADD_RANGE(R500_ZB_FIFO_SIZE, 2); 271 } else { 272 ADD_RANGE(R300_PFS_CNTL_0, 3); 273 ADD_RANGE(R300_PFS_NODE_0, 4); 274 ADD_RANGE(R300_PFS_TEXI_0, 64); 275 ADD_RANGE(R300_PFS_INSTR0_0, 64); 276 ADD_RANGE(R300_PFS_INSTR1_0, 64); 277 ADD_RANGE(R300_PFS_INSTR2_0, 64); 278 ADD_RANGE(R300_PFS_INSTR3_0, 64); 279 ADD_RANGE(R300_RS_INTERP_0, 8); 280 ADD_RANGE(R300_RS_ROUTE_0, 8); 281 282 } 283} 284 285static __inline__ int r300_check_range(unsigned reg, int count) 286{ 287 int i; 288 if (reg & ~0xffff) 289 return -1; 290 for (i = (reg >> 2); i < (reg >> 2) + count; i++) 291 if (r300_reg_flags[i] != MARK_SAFE) 292 return 1; 293 return 0; 294} 295 296static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t * 297 dev_priv, 298 drm_radeon_kcmd_buffer_t 299 * cmdbuf, 300 drm_r300_cmd_header_t 301 header) 302{ 303 int reg; 304 int sz; 305 int i; 306 int values[64]; 307 RING_LOCALS; 308 309 sz = header.packet0.count; 310 reg = (header.packet0.reghi << 8) | header.packet0.reglo; 311 312 if ((sz > 64) || (sz < 0)) { 313 DRM_ERROR 314 ("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n", 315 reg, sz); 316 return -EINVAL; 317 } 318 for (i = 0; i < sz; i++) { 319 values[i] = ((int *)cmdbuf->buf)[i]; 320 switch (r300_reg_flags[(reg >> 2) + i]) { 321 case MARK_SAFE: 322 break; 323 case MARK_CHECK_OFFSET: 324 if (!radeon_check_offset(dev_priv, (u32) values[i])) { 325 DRM_ERROR 326 ("Offset failed range check (reg=%04x sz=%d)\n", 327 reg, sz); 328 return -EINVAL; 329 } 330 break; 331 default: 332 DRM_ERROR("Register %04x failed check as flag=%02x\n", 333 reg + i * 4, r300_reg_flags[(reg >> 2) + i]); 334 return -EINVAL; 335 } 336 } 337 338 BEGIN_RING(1 + sz); 339 OUT_RING(CP_PACKET0(reg, sz - 1)); 340 OUT_RING_TABLE(values, sz); 341 ADVANCE_RING(); 342 343 cmdbuf->buf += sz * 4; 344 cmdbuf->bufsz -= sz * 4; 345 346 return 0; 347} 348 349/** 350 * Emits a packet0 setting arbitrary registers. 351 * Called by r300_do_cp_cmdbuf. 352 * 353 * Note that checks are performed on contents and addresses of the registers 354 */ 355static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv, 356 drm_radeon_kcmd_buffer_t *cmdbuf, 357 drm_r300_cmd_header_t header) 358{ 359 int reg; 360 int sz; 361 RING_LOCALS; 362 363 sz = header.packet0.count; 364 reg = (header.packet0.reghi << 8) | header.packet0.reglo; 365 366 DRM_DEBUG("R300_CMD_PACKET0: reg %04x, sz %d\n", reg, sz); 367 if (!sz) 368 return 0; 369 370 if (sz * 4 > cmdbuf->bufsz) 371 return -EINVAL; 372 373 if (reg + sz * 4 >= 0x10000) { 374 DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg, 375 sz); 376 return -EINVAL; 377 } 378 379 if (r300_check_range(reg, sz)) { 380 /* go and check everything */ 381 return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf, 382 header); 383 } 384 /* the rest of the data is safe to emit, whatever the values the user passed */ 385 386 BEGIN_RING(1 + sz); 387 OUT_RING(CP_PACKET0(reg, sz - 1)); 388 OUT_RING_TABLE((int *)cmdbuf->buf, sz); 389 ADVANCE_RING(); 390 391 cmdbuf->buf += sz * 4; 392 cmdbuf->bufsz -= sz * 4; 393 394 return 0; 395} 396 397/** 398 * Uploads user-supplied vertex program instructions or parameters onto 399 * the graphics card. 400 * Called by r300_do_cp_cmdbuf. 401 */ 402static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv, 403 drm_radeon_kcmd_buffer_t *cmdbuf, 404 drm_r300_cmd_header_t header) 405{ 406 int sz; 407 int addr; 408 RING_LOCALS; 409 410 sz = header.vpu.count; 411 addr = (header.vpu.adrhi << 8) | header.vpu.adrlo; 412 413 if (!sz) 414 return 0; 415 if (sz * 16 > cmdbuf->bufsz) 416 return -EINVAL; 417 418 /* VAP is very sensitive so we purge cache before we program it 419 * and we also flush its state before & after */ 420 BEGIN_RING(6); 421 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0)); 422 OUT_RING(R300_RB3D_DC_FLUSH); 423 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0)); 424 OUT_RING(RADEON_WAIT_3D_IDLECLEAN); 425 OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0)); 426 OUT_RING(0); 427 ADVANCE_RING(); 428 /* set flush flag */ 429 dev_priv->track_flush |= RADEON_FLUSH_EMITED; 430 431 BEGIN_RING(3 + sz * 4); 432 OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr); 433 OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1)); 434 OUT_RING_TABLE((int *)cmdbuf->buf, sz * 4); 435 ADVANCE_RING(); 436 437 BEGIN_RING(2); 438 OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0)); 439 OUT_RING(0); 440 ADVANCE_RING(); 441 442 cmdbuf->buf += sz * 16; 443 cmdbuf->bufsz -= sz * 16; 444 445 return 0; 446} 447 448/** 449 * Emit a clear packet from userspace. 450 * Called by r300_emit_packet3. 451 */ 452static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv, 453 drm_radeon_kcmd_buffer_t *cmdbuf) 454{ 455 RING_LOCALS; 456 457 if (8 * 4 > cmdbuf->bufsz) 458 return -EINVAL; 459 460 BEGIN_RING(10); 461 OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8)); 462 OUT_RING(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING | 463 (1 << R300_PRIM_NUM_VERTICES_SHIFT)); 464 OUT_RING_TABLE((int *)cmdbuf->buf, 8); 465 ADVANCE_RING(); 466 467 BEGIN_RING(4); 468 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0)); 469 OUT_RING(R300_RB3D_DC_FLUSH); 470 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0)); 471 OUT_RING(RADEON_WAIT_3D_IDLECLEAN); 472 ADVANCE_RING(); 473 /* set flush flag */ 474 dev_priv->track_flush |= RADEON_FLUSH_EMITED; 475 476 cmdbuf->buf += 8 * 4; 477 cmdbuf->bufsz -= 8 * 4; 478 479 return 0; 480} 481 482static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv, 483 drm_radeon_kcmd_buffer_t *cmdbuf, 484 u32 header) 485{ 486 int count, i, k; 487#define MAX_ARRAY_PACKET 64 488 u32 payload[MAX_ARRAY_PACKET]; 489 u32 narrays; 490 RING_LOCALS; 491 492 count = (header >> 16) & 0x3fff; 493 494 if ((count + 1) > MAX_ARRAY_PACKET) { 495 DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n", 496 count); 497 return -EINVAL; 498 } 499 memset(payload, 0, MAX_ARRAY_PACKET * 4); 500 memcpy(payload, cmdbuf->buf + 4, (count + 1) * 4); 501 502 /* carefully check packet contents */ 503 504 narrays = payload[0]; 505 k = 0; 506 i = 1; 507 while ((k < narrays) && (i < (count + 1))) { 508 i++; /* skip attribute field */ 509 if (!radeon_check_offset(dev_priv, payload[i])) { 510 DRM_ERROR 511 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n", 512 k, i); 513 return -EINVAL; 514 } 515 k++; 516 i++; 517 if (k == narrays) 518 break; 519 /* have one more to process, they come in pairs */ 520 if (!radeon_check_offset(dev_priv, payload[i])) { 521 DRM_ERROR 522 ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n", 523 k, i); 524 return -EINVAL; 525 } 526 k++; 527 i++; 528 } 529 /* do the counts match what we expect ? */ 530 if ((k != narrays) || (i != (count + 1))) { 531 DRM_ERROR 532 ("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n", 533 k, i, narrays, count + 1); 534 return -EINVAL; 535 } 536 537 /* all clear, output packet */ 538 539 BEGIN_RING(count + 2); 540 OUT_RING(header); 541 OUT_RING_TABLE(payload, count + 1); 542 ADVANCE_RING(); 543 544 cmdbuf->buf += (count + 2) * 4; 545 cmdbuf->bufsz -= (count + 2) * 4; 546 547 return 0; 548} 549 550static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv, 551 drm_radeon_kcmd_buffer_t *cmdbuf) 552{ 553 u32 *cmd = (u32 *) cmdbuf->buf; 554 int count, ret; 555 RING_LOCALS; 556 557 count=(cmd[0]>>16) & 0x3fff; 558 559 if (cmd[0] & 0x8000) { 560 u32 offset; 561 562 if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL 563 | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) { 564 offset = cmd[2] << 10; 565 ret = !radeon_check_offset(dev_priv, offset); 566 if (ret) { 567 DRM_ERROR("Invalid bitblt first offset is %08X\n", offset); 568 return -EINVAL; 569 } 570 } 571 572 if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) && 573 (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) { 574 offset = cmd[3] << 10; 575 ret = !radeon_check_offset(dev_priv, offset); 576 if (ret) { 577 DRM_ERROR("Invalid bitblt second offset is %08X\n", offset); 578 return -EINVAL; 579 } 580 581 } 582 } 583 584 BEGIN_RING(count+2); 585 OUT_RING(cmd[0]); 586 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1); 587 ADVANCE_RING(); 588 589 cmdbuf->buf += (count+2)*4; 590 cmdbuf->bufsz -= (count+2)*4; 591 592 return 0; 593} 594 595static __inline__ int r300_emit_draw_indx_2(drm_radeon_private_t *dev_priv, 596 drm_radeon_kcmd_buffer_t *cmdbuf) 597{ 598 u32 *cmd; 599 int count; 600 int expected_count; 601 RING_LOCALS; 602 603 cmd = (u32 *) cmdbuf->buf; 604 count = (cmd[0]>>16) & 0x3fff; 605 expected_count = cmd[1] >> 16; 606 if (!(cmd[1] & R300_VAP_VF_CNTL__INDEX_SIZE_32bit)) 607 expected_count = (expected_count+1)/2; 608 609 if (count && count != expected_count) { 610 DRM_ERROR("3D_DRAW_INDX_2: packet size %i, expected %i\n", 611 count, expected_count); 612 return -EINVAL; 613 } 614 615 BEGIN_RING(count+2); 616 OUT_RING(cmd[0]); 617 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1); 618 ADVANCE_RING(); 619 620 cmdbuf->buf += (count+2)*4; 621 cmdbuf->bufsz -= (count+2)*4; 622 623 if (!count) { 624 drm_r300_cmd_header_t header; 625 626 if (cmdbuf->bufsz < 4*4 + sizeof(header)) { 627 DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER, but stream is too short.\n"); 628 return -EINVAL; 629 } 630 631 header.u = *(unsigned int *)cmdbuf->buf; 632 633 cmdbuf->buf += sizeof(header); 634 cmdbuf->bufsz -= sizeof(header); 635 cmd = (u32 *) cmdbuf->buf; 636 637 if (header.header.cmd_type != R300_CMD_PACKET3 || 638 header.packet3.packet != R300_CMD_PACKET3_RAW || 639 cmd[0] != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) { 640 DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER.\n"); 641 return -EINVAL; 642 } 643 644 if ((cmd[1] & 0x8000ffff) != 0x80000810) { 645 DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]); 646 return -EINVAL; 647 } 648 if (!radeon_check_offset(dev_priv, cmd[2])) { 649 DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]); 650 return -EINVAL; 651 } 652 if (cmd[3] != expected_count) { 653 DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n", 654 cmd[3], expected_count); 655 return -EINVAL; 656 } 657 658 BEGIN_RING(4); 659 OUT_RING(cmd[0]); 660 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), 3); 661 ADVANCE_RING(); 662 663 cmdbuf->buf += 4*4; 664 cmdbuf->bufsz -= 4*4; 665 } 666 667 return 0; 668} 669 670static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv, 671 drm_radeon_kcmd_buffer_t *cmdbuf) 672{ 673 u32 header; 674 int count; 675 RING_LOCALS; 676 677 if (4 > cmdbuf->bufsz) 678 return -EINVAL; 679 680 /* Fixme !! This simply emits a packet without much checking. 681 We need to be smarter. */ 682 683 /* obtain first word - actual packet3 header */ 684 header = *(u32 *) cmdbuf->buf; 685 686 /* Is it packet 3 ? */ 687 if ((header >> 30) != 0x3) { 688 DRM_ERROR("Not a packet3 header (0x%08x)\n", header); 689 return -EINVAL; 690 } 691 692 count = (header >> 16) & 0x3fff; 693 694 /* Check again now that we know how much data to expect */ 695 if ((count + 2) * 4 > cmdbuf->bufsz) { 696 DRM_ERROR 697 ("Expected packet3 of length %d but have only %d bytes left\n", 698 (count + 2) * 4, cmdbuf->bufsz); 699 return -EINVAL; 700 } 701 702 /* Is it a packet type we know about ? */ 703 switch (header & 0xff00) { 704 case RADEON_3D_LOAD_VBPNTR: /* load vertex array pointers */ 705 return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, header); 706 707 case RADEON_CNTL_BITBLT_MULTI: 708 return r300_emit_bitblt_multi(dev_priv, cmdbuf); 709 710 case RADEON_CP_INDX_BUFFER: 711 DRM_ERROR("packet3 INDX_BUFFER without preceding 3D_DRAW_INDX_2 is illegal.\n"); 712 return -EINVAL; 713 case RADEON_CP_3D_DRAW_IMMD_2: 714 /* triggers drawing using in-packet vertex data */ 715 case RADEON_CP_3D_DRAW_VBUF_2: 716 /* triggers drawing of vertex buffers setup elsewhere */ 717 dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED | 718 RADEON_PURGE_EMITED); 719 break; 720 case RADEON_CP_3D_DRAW_INDX_2: 721 /* triggers drawing using indices to vertex buffer */ 722 /* whenever we send vertex we clear flush & purge */ 723 dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED | 724 RADEON_PURGE_EMITED); 725 return r300_emit_draw_indx_2(dev_priv, cmdbuf); 726 case RADEON_WAIT_FOR_IDLE: 727 case RADEON_CP_NOP: 728 /* these packets are safe */ 729 break; 730 default: 731 DRM_ERROR("Unknown packet3 header (0x%08x)\n", header); 732 return -EINVAL; 733 } 734 735 BEGIN_RING(count + 2); 736 OUT_RING(header); 737 OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1); 738 ADVANCE_RING(); 739 740 cmdbuf->buf += (count + 2) * 4; 741 cmdbuf->bufsz -= (count + 2) * 4; 742 743 return 0; 744} 745 746/** 747 * Emit a rendering packet3 from userspace. 748 * Called by r300_do_cp_cmdbuf. 749 */ 750static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv, 751 drm_radeon_kcmd_buffer_t *cmdbuf, 752 drm_r300_cmd_header_t header) 753{ 754 int n; 755 int ret; 756 char *orig_buf = cmdbuf->buf; 757 int orig_bufsz = cmdbuf->bufsz; 758 759 /* This is a do-while-loop so that we run the interior at least once, 760 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale. 761 */ 762 n = 0; 763 do { 764 if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) { 765 ret = r300_emit_cliprects(dev_priv, cmdbuf, n); 766 if (ret) 767 return ret; 768 769 cmdbuf->buf = orig_buf; 770 cmdbuf->bufsz = orig_bufsz; 771 } 772 773 switch (header.packet3.packet) { 774 case R300_CMD_PACKET3_CLEAR: 775 DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n"); 776 ret = r300_emit_clear(dev_priv, cmdbuf); 777 if (ret) { 778 DRM_ERROR("r300_emit_clear failed\n"); 779 return ret; 780 } 781 break; 782 783 case R300_CMD_PACKET3_RAW: 784 DRM_DEBUG("R300_CMD_PACKET3_RAW\n"); 785 ret = r300_emit_raw_packet3(dev_priv, cmdbuf); 786 if (ret) { 787 DRM_ERROR("r300_emit_raw_packet3 failed\n"); 788 return ret; 789 } 790 break; 791 792 default: 793 DRM_ERROR("bad packet3 type %i at %p\n", 794 header.packet3.packet, 795 cmdbuf->buf - sizeof(header)); 796 return -EINVAL; 797 } 798 799 n += R300_SIMULTANEOUS_CLIPRECTS; 800 } while (n < cmdbuf->nbox); 801 802 return 0; 803} 804 805/* Some of the R300 chips seem to be extremely touchy about the two registers 806 * that are configured in r300_pacify. 807 * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace 808 * sends a command buffer that contains only state setting commands and a 809 * vertex program/parameter upload sequence, this will eventually lead to a 810 * lockup, unless the sequence is bracketed by calls to r300_pacify. 811 * So we should take great care to *always* call r300_pacify before 812 * *anything* 3D related, and again afterwards. This is what the 813 * call bracket in r300_do_cp_cmdbuf is for. 814 */ 815 816/** 817 * Emit the sequence to pacify R300. 818 */ 819static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv) 820{ 821 uint32_t cache_z, cache_3d, cache_2d; 822 RING_LOCALS; 823 824 cache_z = R300_ZC_FLUSH; 825 cache_2d = R300_RB2D_DC_FLUSH; 826 cache_3d = R300_RB3D_DC_FLUSH; 827 if (!(dev_priv->track_flush & RADEON_PURGE_EMITED)) { 828 /* we can purge, primitive where draw since last purge */ 829 cache_z |= R300_ZC_FREE; 830 cache_2d |= R300_RB2D_DC_FREE; 831 cache_3d |= R300_RB3D_DC_FREE; 832 } 833 834 /* flush & purge zbuffer */ 835 BEGIN_RING(2); 836 OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0)); 837 OUT_RING(cache_z); 838 ADVANCE_RING(); 839 /* flush & purge 3d */ 840 BEGIN_RING(2); 841 OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0)); 842 OUT_RING(cache_3d); 843 ADVANCE_RING(); 844 /* flush & purge texture */ 845 BEGIN_RING(2); 846 OUT_RING(CP_PACKET0(R300_TX_INVALTAGS, 0)); 847 OUT_RING(0); 848 ADVANCE_RING(); 849 /* FIXME: is this one really needed ? */ 850 BEGIN_RING(2); 851 OUT_RING(CP_PACKET0(R300_RB3D_AARESOLVE_CTL, 0)); 852 OUT_RING(0); 853 ADVANCE_RING(); 854 BEGIN_RING(2); 855 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0)); 856 OUT_RING(RADEON_WAIT_3D_IDLECLEAN); 857 ADVANCE_RING(); 858 /* flush & purge 2d through E2 as RB2D will trigger lockup */ 859 BEGIN_RING(4); 860 OUT_RING(CP_PACKET0(R300_DSTCACHE_CTLSTAT, 0)); 861 OUT_RING(cache_2d); 862 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0)); 863 OUT_RING(RADEON_WAIT_2D_IDLECLEAN | 864 RADEON_WAIT_HOST_IDLECLEAN); 865 ADVANCE_RING(); 866 /* set flush & purge flags */ 867 dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED; 868} 869 870/** 871 * Called by r300_do_cp_cmdbuf to update the internal buffer age and state. 872 * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must 873 * be careful about how this function is called. 874 */ 875static void r300_discard_buffer(struct drm_device * dev, struct drm_buf * buf) 876{ 877 drm_radeon_private_t *dev_priv = dev->dev_private; 878 drm_radeon_buf_priv_t *buf_priv = buf->dev_private; 879 880 buf_priv->age = ++dev_priv->sarea_priv->last_dispatch; 881 buf->pending = 1; 882 buf->used = 0; 883} 884 885static void r300_cmd_wait(drm_radeon_private_t * dev_priv, 886 drm_r300_cmd_header_t header) 887{ 888 u32 wait_until; 889 RING_LOCALS; 890 891 if (!header.wait.flags) 892 return; 893 894 wait_until = 0; 895 896 switch(header.wait.flags) { 897 case R300_WAIT_2D: 898 wait_until = RADEON_WAIT_2D_IDLE; 899 break; 900 case R300_WAIT_3D: 901 wait_until = RADEON_WAIT_3D_IDLE; 902 break; 903 case R300_NEW_WAIT_2D_3D: 904 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_3D_IDLE; 905 break; 906 case R300_NEW_WAIT_2D_2D_CLEAN: 907 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN; 908 break; 909 case R300_NEW_WAIT_3D_3D_CLEAN: 910 wait_until = RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN; 911 break; 912 case R300_NEW_WAIT_2D_2D_CLEAN_3D_3D_CLEAN: 913 wait_until = RADEON_WAIT_2D_IDLE|RADEON_WAIT_2D_IDLECLEAN; 914 wait_until |= RADEON_WAIT_3D_IDLE|RADEON_WAIT_3D_IDLECLEAN; 915 break; 916 default: 917 return; 918 } 919 920 BEGIN_RING(2); 921 OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0)); 922 OUT_RING(wait_until); 923 ADVANCE_RING(); 924} 925 926static int r300_scratch(drm_radeon_private_t *dev_priv, 927 drm_radeon_kcmd_buffer_t *cmdbuf, 928 drm_r300_cmd_header_t header) 929{ 930 u32 *ref_age_base; 931 u32 i, buf_idx, h_pending; 932 RING_LOCALS; 933 934 if (cmdbuf->bufsz < sizeof(uint64_t) + header.scratch.n_bufs * sizeof(buf_idx) ) { 935 return -EINVAL; 936 } 937 938 if (header.scratch.reg >= 5) { 939 return -EINVAL; 940 } 941 942 dev_priv->scratch_ages[header.scratch.reg] ++; 943 944 ref_age_base = (u32 *)(unsigned long)*((uint64_t *)cmdbuf->buf); 945 946 cmdbuf->buf += sizeof(uint64_t); 947 cmdbuf->bufsz -= sizeof(uint64_t); 948 949 for (i=0; i < header.scratch.n_bufs; i++) { 950 buf_idx = *(u32 *)cmdbuf->buf; 951 buf_idx *= 2; /* 8 bytes per buf */ 952 953 if (DRM_COPY_TO_USER(ref_age_base + buf_idx, &dev_priv->scratch_ages[header.scratch.reg], sizeof(u32))) { 954 return -EINVAL; 955 } 956 957 if (DRM_COPY_FROM_USER(&h_pending, ref_age_base + buf_idx + 1, sizeof(u32))) { 958 return -EINVAL; 959 } 960 961 if (h_pending == 0) { 962 return -EINVAL; 963 } 964 965 h_pending--; 966 967 if (DRM_COPY_TO_USER(ref_age_base + buf_idx + 1, &h_pending, sizeof(u32))) { 968 return -EINVAL; 969 } 970 971 cmdbuf->buf += sizeof(buf_idx); 972 cmdbuf->bufsz -= sizeof(buf_idx); 973 } 974 975 BEGIN_RING(2); 976 OUT_RING( CP_PACKET0( RADEON_SCRATCH_REG0 + header.scratch.reg * 4, 0 ) ); 977 OUT_RING( dev_priv->scratch_ages[header.scratch.reg] ); 978 ADVANCE_RING(); 979 980 return 0; 981} 982 983/** 984 * Uploads user-supplied vertex program instructions or parameters onto 985 * the graphics card. 986 * Called by r300_do_cp_cmdbuf. 987 */ 988static __inline__ int r300_emit_r500fp(drm_radeon_private_t *dev_priv, 989 drm_radeon_kcmd_buffer_t *cmdbuf, 990 drm_r300_cmd_header_t header) 991{ 992 int sz; 993 int addr; 994 int type; 995 int clamp; 996 int stride; 997 RING_LOCALS; 998 999 sz = header.r500fp.count; 1000 /* address is 9 bits 0 - 8, bit 1 of flags is part of address */ 1001 addr = ((header.r500fp.adrhi_flags & 1) << 8) | header.r500fp.adrlo; 1002 1003 type = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE); 1004 clamp = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP); 1005 1006 addr |= (type << 16); 1007 addr |= (clamp << 17); 1008 1009 stride = type ? 4 : 6; 1010 1011 DRM_DEBUG("r500fp %d %d type: %d\n", sz, addr, type); 1012 if (!sz) 1013 return 0; 1014 if (sz * stride * 4 > cmdbuf->bufsz) 1015 return -EINVAL; 1016 1017 BEGIN_RING(3 + sz * stride); 1018 OUT_RING_REG(R500_GA_US_VECTOR_INDEX, addr); 1019 OUT_RING(CP_PACKET0_TABLE(R500_GA_US_VECTOR_DATA, sz * stride - 1)); 1020 OUT_RING_TABLE((int *)cmdbuf->buf, sz * stride); 1021 1022 ADVANCE_RING(); 1023 1024 cmdbuf->buf += sz * stride * 4; 1025 cmdbuf->bufsz -= sz * stride * 4; 1026 1027 return 0; 1028} 1029 1030 1031/** 1032 * Parses and validates a user-supplied command buffer and emits appropriate 1033 * commands on the DMA ring buffer. 1034 * Called by the ioctl handler function radeon_cp_cmdbuf. 1035 */ 1036int r300_do_cp_cmdbuf(struct drm_device *dev, 1037 struct drm_file *file_priv, 1038 drm_radeon_kcmd_buffer_t *cmdbuf) 1039{ 1040 drm_radeon_private_t *dev_priv = dev->dev_private; 1041 struct drm_device_dma *dma = dev->dma; 1042 struct drm_buf *buf = NULL; 1043 int emit_dispatch_age = 0; 1044 int ret = 0; 1045 1046 DRM_DEBUG("\n"); 1047 1048 /* pacify */ 1049 r300_pacify(dev_priv); 1050 1051 if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) { 1052 ret = r300_emit_cliprects(dev_priv, cmdbuf, 0); 1053 if (ret) 1054 goto cleanup; 1055 } 1056 1057 while (cmdbuf->bufsz >= sizeof(drm_r300_cmd_header_t)) { 1058 int idx; 1059 drm_r300_cmd_header_t header; 1060 1061 header.u = *(unsigned int *)cmdbuf->buf; 1062 1063 cmdbuf->buf += sizeof(header); 1064 cmdbuf->bufsz -= sizeof(header); 1065 1066 switch (header.header.cmd_type) { 1067 case R300_CMD_PACKET0: 1068 ret = r300_emit_packet0(dev_priv, cmdbuf, header); 1069 if (ret) { 1070 DRM_ERROR("r300_emit_packet0 failed\n"); 1071 goto cleanup; 1072 } 1073 break; 1074 1075 case R300_CMD_VPU: 1076 DRM_DEBUG("R300_CMD_VPU\n"); 1077 ret = r300_emit_vpu(dev_priv, cmdbuf, header); 1078 if (ret) { 1079 DRM_ERROR("r300_emit_vpu failed\n"); 1080 goto cleanup; 1081 } 1082 break; 1083 1084 case R300_CMD_PACKET3: 1085 DRM_DEBUG("R300_CMD_PACKET3\n"); 1086 ret = r300_emit_packet3(dev_priv, cmdbuf, header); 1087 if (ret) { 1088 DRM_ERROR("r300_emit_packet3 failed\n"); 1089 goto cleanup; 1090 } 1091 break; 1092 1093 case R300_CMD_END3D: 1094 DRM_DEBUG("R300_CMD_END3D\n"); 1095 /* TODO: 1096 Ideally userspace driver should not need to issue this call, 1097 i.e. the drm driver should issue it automatically and prevent 1098 lockups. 1099 1100 In practice, we do not understand why this call is needed and what 1101 it does (except for some vague guesses that it has to do with cache 1102 coherence) and so the user space driver does it. 1103 1104 Once we are sure which uses prevent lockups the code could be moved 1105 into the kernel and the userspace driver will not 1106 need to use this command. 1107 1108 Note that issuing this command does not hurt anything 1109 except, possibly, performance */ 1110 r300_pacify(dev_priv); 1111 break; 1112 1113 case R300_CMD_CP_DELAY: 1114 /* simple enough, we can do it here */ 1115 DRM_DEBUG("R300_CMD_CP_DELAY\n"); 1116 { 1117 int i; 1118 RING_LOCALS; 1119 1120 BEGIN_RING(header.delay.count); 1121 for (i = 0; i < header.delay.count; i++) 1122 OUT_RING(RADEON_CP_PACKET2); 1123 ADVANCE_RING(); 1124 } 1125 break; 1126 1127 case R300_CMD_DMA_DISCARD: 1128 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n"); 1129 idx = header.dma.buf_idx; 1130 if (idx < 0 || idx >= dma->buf_count) { 1131 DRM_ERROR("buffer index %d (of %d max)\n", 1132 idx, dma->buf_count - 1); 1133 ret = -EINVAL; 1134 goto cleanup; 1135 } 1136 1137 buf = dma->buflist[idx]; 1138 if (buf->file_priv != file_priv || buf->pending) { 1139 DRM_ERROR("bad buffer %p %p %d\n", 1140 buf->file_priv, file_priv, 1141 buf->pending); 1142 ret = -EINVAL; 1143 goto cleanup; 1144 } 1145 1146 emit_dispatch_age = 1; 1147 r300_discard_buffer(dev, buf); 1148 break; 1149 1150 case R300_CMD_WAIT: 1151 DRM_DEBUG("R300_CMD_WAIT\n"); 1152 r300_cmd_wait(dev_priv, header); 1153 break; 1154 1155 case R300_CMD_SCRATCH: 1156 DRM_DEBUG("R300_CMD_SCRATCH\n"); 1157 ret = r300_scratch(dev_priv, cmdbuf, header); 1158 if (ret) { 1159 DRM_ERROR("r300_scratch failed\n"); 1160 goto cleanup; 1161 } 1162 break; 1163 1164 case R300_CMD_R500FP: 1165 if ((dev_priv->flags & RADEON_FAMILY_MASK) < CHIP_RV515) { 1166 DRM_ERROR("Calling r500 command on r300 card\n"); 1167 ret = -EINVAL; 1168 goto cleanup; 1169 } 1170 DRM_DEBUG("R300_CMD_R500FP\n"); 1171 ret = r300_emit_r500fp(dev_priv, cmdbuf, header); 1172 if (ret) { 1173 DRM_ERROR("r300_emit_r500fp failed\n"); 1174 goto cleanup; 1175 } 1176 break; 1177 default: 1178 DRM_ERROR("bad cmd_type %i at %p\n", 1179 header.header.cmd_type, 1180 cmdbuf->buf - sizeof(header)); 1181 ret = -EINVAL; 1182 goto cleanup; 1183 } 1184 } 1185 1186 DRM_DEBUG("END\n"); 1187 1188 cleanup: 1189 r300_pacify(dev_priv); 1190 1191 /* We emit the vertex buffer age here, outside the pacifier "brackets" 1192 * for two reasons: 1193 * (1) This may coalesce multiple age emissions into a single one and 1194 * (2) more importantly, some chips lock up hard when scratch registers 1195 * are written inside the pacifier bracket. 1196 */ 1197 if (emit_dispatch_age) { 1198 RING_LOCALS; 1199 1200 /* Emit the vertex buffer age */ 1201 BEGIN_RING(2); 1202 RADEON_DISPATCH_AGE(dev_priv->sarea_priv->last_dispatch); 1203 ADVANCE_RING(); 1204 } 1205 1206 COMMIT_RING(); 1207 1208 return ret; 1209} 1210