2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
24 * Authors: Dave Airlie
33 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
34 struct radeon_cs_reloc **cs_reloc);
35 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
36 struct radeon_cs_reloc **cs_reloc);
37 typedef int (*next_reloc_t)(struct radeon_cs_parser*, struct radeon_cs_reloc**);
38 static next_reloc_t r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_mm;
41 * r600_cs_packet_parse() - parse cp packet and point ib index to next packet
42 * @parser: parser structure holding parsing context.
43 * @pkt: where to store packet informations
45 * Assume that chunk_ib_index is properly set. Will return -EINVAL
46 * if packet is bigger than remaining ib size. or if packets is unknown.
48 int r600_cs_packet_parse(struct radeon_cs_parser *p,
49 struct radeon_cs_packet *pkt,
52 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
55 if (idx >= ib_chunk->length_dw) {
56 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
57 idx, ib_chunk->length_dw);
60 header = ib_chunk->kdata[idx];
62 pkt->type = CP_PACKET_GET_TYPE(header);
63 pkt->count = CP_PACKET_GET_COUNT(header);
67 pkt->reg = CP_PACKET0_GET_REG(header);
70 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
76 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
79 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
80 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
81 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
88 * r600_cs_packet_next_reloc_mm() - parse next packet which should be reloc packet3
89 * @parser: parser structure holding parsing context.
90 * @data: pointer to relocation data
91 * @offset_start: starting offset
92 * @offset_mask: offset mask (to align start offset on)
93 * @reloc: reloc informations
95 * Check next packet is relocation packet3, do bo validation and compute
96 * GPU offset using the provided start.
98 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
99 struct radeon_cs_reloc **cs_reloc)
101 struct radeon_cs_chunk *ib_chunk;
102 struct radeon_cs_chunk *relocs_chunk;
103 struct radeon_cs_packet p3reloc;
107 if (p->chunk_relocs_idx == -1) {
108 DRM_ERROR("No relocation chunk !\n");
112 ib_chunk = &p->chunks[p->chunk_ib_idx];
113 relocs_chunk = &p->chunks[p->chunk_relocs_idx];
114 r = r600_cs_packet_parse(p, &p3reloc, p->idx);
118 p->idx += p3reloc.count + 2;
119 if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
120 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
124 idx = ib_chunk->kdata[p3reloc.idx + 1];
125 if (idx >= relocs_chunk->length_dw) {
126 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
127 idx, relocs_chunk->length_dw);
130 /* FIXME: we assume reloc size is 4 dwords */
131 *cs_reloc = p->relocs_ptr[(idx / 4)];
136 * r600_cs_packet_next_reloc_nomm() - parse next packet which should be reloc packet3
137 * @parser: parser structure holding parsing context.
138 * @data: pointer to relocation data
139 * @offset_start: starting offset
140 * @offset_mask: offset mask (to align start offset on)
141 * @reloc: reloc informations
143 * Check next packet is relocation packet3, do bo validation and compute
144 * GPU offset using the provided start.
146 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
147 struct radeon_cs_reloc **cs_reloc)
149 struct radeon_cs_chunk *ib_chunk;
150 struct radeon_cs_chunk *relocs_chunk;
151 struct radeon_cs_packet p3reloc;
155 if (p->chunk_relocs_idx == -1) {
156 DRM_ERROR("No relocation chunk !\n");
160 ib_chunk = &p->chunks[p->chunk_ib_idx];
161 relocs_chunk = &p->chunks[p->chunk_relocs_idx];
162 r = r600_cs_packet_parse(p, &p3reloc, p->idx);
166 p->idx += p3reloc.count + 2;
167 if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
168 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
172 idx = ib_chunk->kdata[p3reloc.idx + 1];
173 if (idx >= relocs_chunk->length_dw) {
174 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
175 idx, relocs_chunk->length_dw);
178 *cs_reloc = &p->relocs[0];
179 (*cs_reloc)->lobj.gpu_offset = (u64)relocs_chunk->kdata[idx + 3] << 32;
180 (*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
184 static int r600_packet0_check(struct radeon_cs_parser *p,
185 struct radeon_cs_packet *pkt,
186 unsigned idx, unsigned reg)
189 case AVIVO_D1MODE_VLINE_START_END:
190 case AVIVO_D2MODE_VLINE_START_END:
193 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
200 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
201 struct radeon_cs_packet *pkt)
209 for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
210 r = r600_packet0_check(p, pkt, idx, reg);
218 static int r600_packet3_check(struct radeon_cs_parser *p,
219 struct radeon_cs_packet *pkt)
221 struct radeon_cs_chunk *ib_chunk;
222 struct radeon_cs_reloc *reloc;
226 unsigned start_reg, end_reg, reg;
230 ib_chunk = &p->chunks[p->chunk_ib_idx];
232 switch (pkt->opcode) {
233 case PACKET3_START_3D_CMDBUF:
234 if (p->family >= CHIP_RV770 || pkt->count) {
235 DRM_ERROR("bad START_3D\n");
239 case PACKET3_CONTEXT_CONTROL:
240 if (pkt->count != 1) {
241 DRM_ERROR("bad CONTEXT_CONTROL\n");
245 case PACKET3_INDEX_TYPE:
246 case PACKET3_NUM_INSTANCES:
248 DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
252 case PACKET3_DRAW_INDEX:
253 if (pkt->count != 3) {
254 DRM_ERROR("bad DRAW_INDEX\n");
257 r = r600_cs_packet_next_reloc(p, &reloc);
259 DRM_ERROR("bad DRAW_INDEX\n");
262 ib[idx+0] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
263 ib[idx+1] = upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
265 case PACKET3_DRAW_INDEX_AUTO:
266 if (pkt->count != 1) {
267 DRM_ERROR("bad DRAW_INDEX_AUTO\n");
271 case PACKET3_DRAW_INDEX_IMMD_BE:
272 case PACKET3_DRAW_INDEX_IMMD:
273 if (pkt->count < 2) {
274 DRM_ERROR("bad DRAW_INDEX_IMMD\n");
278 case PACKET3_WAIT_REG_MEM:
279 if (pkt->count != 5) {
280 DRM_ERROR("bad WAIT_REG_MEM\n");
283 /* bit 4 is reg (0) or mem (1) */
284 if (ib_chunk->kdata[idx+0] & 0x10) {
285 r = r600_cs_packet_next_reloc(p, &reloc);
287 DRM_ERROR("bad WAIT_REG_MEM\n");
290 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
291 ib[idx+2] = upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
294 case PACKET3_SURFACE_SYNC:
295 if (pkt->count != 3) {
296 DRM_ERROR("bad SURFACE_SYNC\n");
299 /* 0xffffffff/0x0 is flush all cache flag */
300 if (ib_chunk->kdata[idx+1] != 0xffffffff ||
301 ib_chunk->kdata[idx+2] != 0) {
302 r = r600_cs_packet_next_reloc(p, &reloc);
304 DRM_ERROR("bad SURFACE_SYNC\n");
307 ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
310 case PACKET3_EVENT_WRITE:
311 if (pkt->count != 2 && pkt->count != 0) {
312 DRM_ERROR("bad EVENT_WRITE\n");
316 r = r600_cs_packet_next_reloc(p, &reloc);
318 DRM_ERROR("bad EVENT_WRITE\n");
321 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
322 ib[idx+2] |= upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
325 case PACKET3_EVENT_WRITE_EOP:
326 if (pkt->count != 4) {
327 DRM_ERROR("bad EVENT_WRITE_EOP\n");
330 r = r600_cs_packet_next_reloc(p, &reloc);
332 DRM_ERROR("bad EVENT_WRITE\n");
335 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
336 ib[idx+2] |= upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
338 case PACKET3_SET_CONFIG_REG:
339 start_reg = (ib[idx+0] << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
340 end_reg = 4 * pkt->count + start_reg - 4;
341 if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
342 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
343 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
344 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
347 for (i = 0; i < pkt->count; i++) {
348 reg = start_reg + (4 * i);
351 /* use PACKET3_SURFACE_SYNC */
358 case PACKET3_SET_CONTEXT_REG:
359 start_reg = (ib[idx+0] << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
360 end_reg = 4 * pkt->count + start_reg - 4;
361 if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
362 (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
363 (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
364 DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
367 for (i = 0; i < pkt->count; i++) {
368 reg = start_reg + (4 * i);
379 case SQ_PGM_START_FS:
380 case SQ_PGM_START_ES:
381 case SQ_PGM_START_VS:
382 case SQ_PGM_START_GS:
383 case SQ_PGM_START_PS:
384 r = r600_cs_packet_next_reloc(p, &reloc);
386 DRM_ERROR("bad SET_CONTEXT_REG "
390 ib[idx+1+i] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
393 case VGT_DMA_BASE_HI:
394 /* These should be handled by DRAW_INDEX packet 3 */
395 case VGT_STRMOUT_BASE_OFFSET_0:
396 case VGT_STRMOUT_BASE_OFFSET_1:
397 case VGT_STRMOUT_BASE_OFFSET_2:
398 case VGT_STRMOUT_BASE_OFFSET_3:
399 case VGT_STRMOUT_BASE_OFFSET_HI_0:
400 case VGT_STRMOUT_BASE_OFFSET_HI_1:
401 case VGT_STRMOUT_BASE_OFFSET_HI_2:
402 case VGT_STRMOUT_BASE_OFFSET_HI_3:
403 case VGT_STRMOUT_BUFFER_BASE_0:
404 case VGT_STRMOUT_BUFFER_BASE_1:
405 case VGT_STRMOUT_BUFFER_BASE_2:
406 case VGT_STRMOUT_BUFFER_BASE_3:
407 case VGT_STRMOUT_BUFFER_OFFSET_0:
408 case VGT_STRMOUT_BUFFER_OFFSET_1:
409 case VGT_STRMOUT_BUFFER_OFFSET_2:
410 case VGT_STRMOUT_BUFFER_OFFSET_3:
411 /* These should be handled by STRMOUT_BUFFER packet 3 */
412 DRM_ERROR("bad context reg: 0x%08x\n", reg);
419 case PACKET3_SET_RESOURCE:
420 if (pkt->count % 7) {
421 DRM_ERROR("bad SET_RESOURCE\n");
424 start_reg = (ib[idx+0] << 2) + PACKET3_SET_RESOURCE_OFFSET;
425 end_reg = 4 * pkt->count + start_reg - 4;
426 if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
427 (start_reg >= PACKET3_SET_RESOURCE_END) ||
428 (end_reg >= PACKET3_SET_RESOURCE_END)) {
429 DRM_ERROR("bad SET_RESOURCE\n");
432 for (i = 0; i < (pkt->count / 7); i++) {
433 switch (G__SQ_VTX_CONSTANT_TYPE(ib[idx+(i*7)+6+1])) {
434 case SQ_TEX_VTX_VALID_TEXTURE:
436 r = r600_cs_packet_next_reloc(p, &reloc);
438 DRM_ERROR("bad SET_RESOURCE\n");
441 ib[idx+1+(i*7)+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
443 r = r600_cs_packet_next_reloc(p, &reloc);
445 DRM_ERROR("bad SET_RESOURCE\n");
448 ib[idx+1+(i*7)+3] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
450 case SQ_TEX_VTX_VALID_BUFFER:
452 r = r600_cs_packet_next_reloc(p, &reloc);
454 DRM_ERROR("bad SET_RESOURCE\n");
457 ib[idx+1+(i*7)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
458 ib[idx+1+(i*7)+2] |= upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
460 case SQ_TEX_VTX_INVALID_TEXTURE:
461 case SQ_TEX_VTX_INVALID_BUFFER:
463 DRM_ERROR("bad SET_RESOURCE\n");
468 case PACKET3_SET_ALU_CONST:
469 start_reg = (ib[idx+0] << 2) + PACKET3_SET_ALU_CONST_OFFSET;
470 end_reg = 4 * pkt->count + start_reg - 4;
471 if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
472 (start_reg >= PACKET3_SET_ALU_CONST_END) ||
473 (end_reg >= PACKET3_SET_ALU_CONST_END)) {
474 DRM_ERROR("bad SET_ALU_CONST\n");
478 case PACKET3_SET_BOOL_CONST:
479 start_reg = (ib[idx+0] << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
480 end_reg = 4 * pkt->count + start_reg - 4;
481 if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
482 (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
483 (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
484 DRM_ERROR("bad SET_BOOL_CONST\n");
488 case PACKET3_SET_LOOP_CONST:
489 start_reg = (ib[idx+0] << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
490 end_reg = 4 * pkt->count + start_reg - 4;
491 if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
492 (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
493 (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
494 DRM_ERROR("bad SET_LOOP_CONST\n");
498 case PACKET3_SET_CTL_CONST:
499 start_reg = (ib[idx+0] << 2) + PACKET3_SET_CTL_CONST_OFFSET;
500 end_reg = 4 * pkt->count + start_reg - 4;
501 if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
502 (start_reg >= PACKET3_SET_CTL_CONST_END) ||
503 (end_reg >= PACKET3_SET_CTL_CONST_END)) {
504 DRM_ERROR("bad SET_CTL_CONST\n");
508 case PACKET3_SET_SAMPLER:
509 if (pkt->count % 3) {
510 DRM_ERROR("bad SET_SAMPLER\n");
513 start_reg = (ib[idx+0] << 2) + PACKET3_SET_SAMPLER_OFFSET;
514 end_reg = 4 * pkt->count + start_reg - 4;
515 if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
516 (start_reg >= PACKET3_SET_SAMPLER_END) ||
517 (end_reg >= PACKET3_SET_SAMPLER_END)) {
518 DRM_ERROR("bad SET_SAMPLER\n");
522 case PACKET3_SURFACE_BASE_UPDATE:
523 if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
524 DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
528 DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
535 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
541 int r600_cs_parse(struct radeon_cs_parser *p)
543 struct radeon_cs_packet pkt;
547 r = r600_cs_packet_parse(p, &pkt, p->idx);
551 p->idx += pkt.count + 2;
554 r = r600_cs_parse_packet0(p, &pkt);
559 r = r600_packet3_check(p, &pkt);
562 DRM_ERROR("Unknown packet type %d !\n", pkt.type);
568 } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
570 for (r = 0; r < p->ib->length_dw; r++) {
571 printk(KERN_INFO "%05d 0x%08X\n", r, p->ib->ptr[r]);
578 static int r600_cs_parser_relocs_legacy(struct radeon_cs_parser *p)
580 if (p->chunk_relocs_idx == -1) {
583 p->relocs = kcalloc(1, sizeof(struct radeon_cs_reloc), GFP_KERNEL);
584 if (p->relocs == NULL) {
591 * cs_parser_fini() - clean parser states
592 * @parser: parser structure holding parsing context.
593 * @error: error number
595 * If error is set than unvalidate buffer, otherwise just free memory
596 * used by parsing context.
598 static void r600_cs_parser_fini(struct radeon_cs_parser *parser, int error)
602 kfree(parser->relocs);
603 for (i = 0; i < parser->nchunks; i++) {
604 kfree(parser->chunks[i].kdata);
606 kfree(parser->chunks);
607 kfree(parser->chunks_array);
610 int r600_cs_legacy(struct drm_device *dev, void *data, struct drm_file *filp,
611 unsigned family, u32 *ib, int *l)
613 struct radeon_cs_parser parser;
614 struct radeon_cs_chunk *ib_chunk;
615 struct radeon_ib fake_ib;
618 /* initialize parser */
619 memset(&parser, 0, sizeof(struct radeon_cs_parser));
622 parser.family = family;
623 parser.ib = &fake_ib;
625 r = radeon_cs_parser_init(&parser, data);
627 DRM_ERROR("Failed to initialize parser !\n");
628 r600_cs_parser_fini(&parser, r);
631 r = r600_cs_parser_relocs_legacy(&parser);
633 DRM_ERROR("Failed to parse relocation !\n");
634 r600_cs_parser_fini(&parser, r);
637 /* Copy the packet into the IB, the parser will read from the
638 * input memory (cached) and write to the IB (which can be
640 ib_chunk = &parser.chunks[parser.chunk_ib_idx];
641 parser.ib->length_dw = ib_chunk->length_dw;
642 memcpy((void *)parser.ib->ptr, ib_chunk->kdata, ib_chunk->length_dw*4);
643 *l = parser.ib->length_dw;
644 r = r600_cs_parse(&parser);
646 DRM_ERROR("Invalid command stream !\n");
647 r600_cs_parser_fini(&parser, r);
650 r600_cs_parser_fini(&parser, r);
654 void r600_cs_legacy_init(void)
656 r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;