]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/gpu/drm/radeon/evergreen_cs.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'acpica-validate-address-regression' into next
[mirror_ubuntu-jammy-kernel.git] / drivers / gpu / drm / radeon / evergreen_cs.c
1 /*
2 * Copyright 2010 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 *
24 * Authors: Dave Airlie
25 * Alex Deucher
26 * Jerome Glisse
27 */
28 #include "drmP.h"
29 #include "radeon.h"
30 #include "evergreend.h"
31 #include "evergreen_reg_safe.h"
32 #include "cayman_reg_safe.h"
33
34 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
35 struct radeon_cs_reloc **cs_reloc);
36
37 struct evergreen_cs_track {
38 u32 group_size;
39 u32 nbanks;
40 u32 npipes;
41 u32 row_size;
42 /* value we track */
43 u32 nsamples;
44 u32 cb_color_base_last[12];
45 struct radeon_bo *cb_color_bo[12];
46 u32 cb_color_bo_offset[12];
47 struct radeon_bo *cb_color_fmask_bo[8];
48 struct radeon_bo *cb_color_cmask_bo[8];
49 u32 cb_color_info[12];
50 u32 cb_color_view[12];
51 u32 cb_color_pitch_idx[12];
52 u32 cb_color_slice_idx[12];
53 u32 cb_color_dim_idx[12];
54 u32 cb_color_dim[12];
55 u32 cb_color_pitch[12];
56 u32 cb_color_slice[12];
57 u32 cb_color_cmask_slice[8];
58 u32 cb_color_fmask_slice[8];
59 u32 cb_target_mask;
60 u32 cb_shader_mask;
61 u32 vgt_strmout_config;
62 u32 vgt_strmout_buffer_config;
63 u32 db_depth_control;
64 u32 db_depth_view;
65 u32 db_depth_size;
66 u32 db_depth_size_idx;
67 u32 db_z_info;
68 u32 db_z_idx;
69 u32 db_z_read_offset;
70 u32 db_z_write_offset;
71 struct radeon_bo *db_z_read_bo;
72 struct radeon_bo *db_z_write_bo;
73 u32 db_s_info;
74 u32 db_s_idx;
75 u32 db_s_read_offset;
76 u32 db_s_write_offset;
77 struct radeon_bo *db_s_read_bo;
78 struct radeon_bo *db_s_write_bo;
79 };
80
81 static u32 evergreen_cs_get_aray_mode(u32 tiling_flags)
82 {
83 if (tiling_flags & RADEON_TILING_MACRO)
84 return ARRAY_2D_TILED_THIN1;
85 else if (tiling_flags & RADEON_TILING_MICRO)
86 return ARRAY_1D_TILED_THIN1;
87 else
88 return ARRAY_LINEAR_GENERAL;
89 }
90
91 static u32 evergreen_cs_get_num_banks(u32 nbanks)
92 {
93 switch (nbanks) {
94 case 2:
95 return ADDR_SURF_2_BANK;
96 case 4:
97 return ADDR_SURF_4_BANK;
98 case 8:
99 default:
100 return ADDR_SURF_8_BANK;
101 case 16:
102 return ADDR_SURF_16_BANK;
103 }
104 }
105
106 static u32 evergreen_cs_get_tile_split(u32 row_size)
107 {
108 switch (row_size) {
109 case 1:
110 default:
111 return ADDR_SURF_TILE_SPLIT_1KB;
112 case 2:
113 return ADDR_SURF_TILE_SPLIT_2KB;
114 case 4:
115 return ADDR_SURF_TILE_SPLIT_4KB;
116 }
117 }
118
119 static void evergreen_cs_track_init(struct evergreen_cs_track *track)
120 {
121 int i;
122
123 for (i = 0; i < 8; i++) {
124 track->cb_color_fmask_bo[i] = NULL;
125 track->cb_color_cmask_bo[i] = NULL;
126 track->cb_color_cmask_slice[i] = 0;
127 track->cb_color_fmask_slice[i] = 0;
128 }
129
130 for (i = 0; i < 12; i++) {
131 track->cb_color_base_last[i] = 0;
132 track->cb_color_bo[i] = NULL;
133 track->cb_color_bo_offset[i] = 0xFFFFFFFF;
134 track->cb_color_info[i] = 0;
135 track->cb_color_view[i] = 0;
136 track->cb_color_pitch_idx[i] = 0;
137 track->cb_color_slice_idx[i] = 0;
138 track->cb_color_dim[i] = 0;
139 track->cb_color_pitch[i] = 0;
140 track->cb_color_slice[i] = 0;
141 track->cb_color_dim[i] = 0;
142 }
143 track->cb_target_mask = 0xFFFFFFFF;
144 track->cb_shader_mask = 0xFFFFFFFF;
145
146 track->db_depth_view = 0xFFFFC000;
147 track->db_depth_size = 0xFFFFFFFF;
148 track->db_depth_size_idx = 0;
149 track->db_depth_control = 0xFFFFFFFF;
150 track->db_z_info = 0xFFFFFFFF;
151 track->db_z_idx = 0xFFFFFFFF;
152 track->db_z_read_offset = 0xFFFFFFFF;
153 track->db_z_write_offset = 0xFFFFFFFF;
154 track->db_z_read_bo = NULL;
155 track->db_z_write_bo = NULL;
156 track->db_s_info = 0xFFFFFFFF;
157 track->db_s_idx = 0xFFFFFFFF;
158 track->db_s_read_offset = 0xFFFFFFFF;
159 track->db_s_write_offset = 0xFFFFFFFF;
160 track->db_s_read_bo = NULL;
161 track->db_s_write_bo = NULL;
162 }
163
164 static int evergreen_cs_track_check(struct radeon_cs_parser *p)
165 {
166 struct evergreen_cs_track *track = p->track;
167
168 /* we don't support stream out buffer yet */
169 if (track->vgt_strmout_config || track->vgt_strmout_buffer_config) {
170 dev_warn(p->dev, "this kernel doesn't support SMX output buffer\n");
171 return -EINVAL;
172 }
173
174 /* XXX fill in */
175 return 0;
176 }
177
178 /**
179 * evergreen_cs_packet_parse() - parse cp packet and point ib index to next packet
180 * @parser: parser structure holding parsing context.
181 * @pkt: where to store packet informations
182 *
183 * Assume that chunk_ib_index is properly set. Will return -EINVAL
184 * if packet is bigger than remaining ib size. or if packets is unknown.
185 **/
186 int evergreen_cs_packet_parse(struct radeon_cs_parser *p,
187 struct radeon_cs_packet *pkt,
188 unsigned idx)
189 {
190 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
191 uint32_t header;
192
193 if (idx >= ib_chunk->length_dw) {
194 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
195 idx, ib_chunk->length_dw);
196 return -EINVAL;
197 }
198 header = radeon_get_ib_value(p, idx);
199 pkt->idx = idx;
200 pkt->type = CP_PACKET_GET_TYPE(header);
201 pkt->count = CP_PACKET_GET_COUNT(header);
202 pkt->one_reg_wr = 0;
203 switch (pkt->type) {
204 case PACKET_TYPE0:
205 pkt->reg = CP_PACKET0_GET_REG(header);
206 break;
207 case PACKET_TYPE3:
208 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
209 break;
210 case PACKET_TYPE2:
211 pkt->count = -1;
212 break;
213 default:
214 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
215 return -EINVAL;
216 }
217 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
218 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
219 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
220 return -EINVAL;
221 }
222 return 0;
223 }
224
225 /**
226 * evergreen_cs_packet_next_reloc() - parse next packet which should be reloc packet3
227 * @parser: parser structure holding parsing context.
228 * @data: pointer to relocation data
229 * @offset_start: starting offset
230 * @offset_mask: offset mask (to align start offset on)
231 * @reloc: reloc informations
232 *
233 * Check next packet is relocation packet3, do bo validation and compute
234 * GPU offset using the provided start.
235 **/
236 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
237 struct radeon_cs_reloc **cs_reloc)
238 {
239 struct radeon_cs_chunk *relocs_chunk;
240 struct radeon_cs_packet p3reloc;
241 unsigned idx;
242 int r;
243
244 if (p->chunk_relocs_idx == -1) {
245 DRM_ERROR("No relocation chunk !\n");
246 return -EINVAL;
247 }
248 *cs_reloc = NULL;
249 relocs_chunk = &p->chunks[p->chunk_relocs_idx];
250 r = evergreen_cs_packet_parse(p, &p3reloc, p->idx);
251 if (r) {
252 return r;
253 }
254 p->idx += p3reloc.count + 2;
255 if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
256 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
257 p3reloc.idx);
258 return -EINVAL;
259 }
260 idx = radeon_get_ib_value(p, p3reloc.idx + 1);
261 if (idx >= relocs_chunk->length_dw) {
262 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
263 idx, relocs_chunk->length_dw);
264 return -EINVAL;
265 }
266 /* FIXME: we assume reloc size is 4 dwords */
267 *cs_reloc = p->relocs_ptr[(idx / 4)];
268 return 0;
269 }
270
271 /**
272 * evergreen_cs_packet_next_vline() - parse userspace VLINE packet
273 * @parser: parser structure holding parsing context.
274 *
275 * Userspace sends a special sequence for VLINE waits.
276 * PACKET0 - VLINE_START_END + value
277 * PACKET3 - WAIT_REG_MEM poll vline status reg
278 * RELOC (P3) - crtc_id in reloc.
279 *
280 * This function parses this and relocates the VLINE START END
281 * and WAIT_REG_MEM packets to the correct crtc.
282 * It also detects a switched off crtc and nulls out the
283 * wait in that case.
284 */
285 static int evergreen_cs_packet_parse_vline(struct radeon_cs_parser *p)
286 {
287 struct drm_mode_object *obj;
288 struct drm_crtc *crtc;
289 struct radeon_crtc *radeon_crtc;
290 struct radeon_cs_packet p3reloc, wait_reg_mem;
291 int crtc_id;
292 int r;
293 uint32_t header, h_idx, reg, wait_reg_mem_info;
294 volatile uint32_t *ib;
295
296 ib = p->ib->ptr;
297
298 /* parse the WAIT_REG_MEM */
299 r = evergreen_cs_packet_parse(p, &wait_reg_mem, p->idx);
300 if (r)
301 return r;
302
303 /* check its a WAIT_REG_MEM */
304 if (wait_reg_mem.type != PACKET_TYPE3 ||
305 wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
306 DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
307 return -EINVAL;
308 }
309
310 wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
311 /* bit 4 is reg (0) or mem (1) */
312 if (wait_reg_mem_info & 0x10) {
313 DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
314 return -EINVAL;
315 }
316 /* waiting for value to be equal */
317 if ((wait_reg_mem_info & 0x7) != 0x3) {
318 DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
319 return -EINVAL;
320 }
321 if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != EVERGREEN_VLINE_STATUS) {
322 DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
323 return -EINVAL;
324 }
325
326 if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != EVERGREEN_VLINE_STAT) {
327 DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
328 return -EINVAL;
329 }
330
331 /* jump over the NOP */
332 r = evergreen_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
333 if (r)
334 return r;
335
336 h_idx = p->idx - 2;
337 p->idx += wait_reg_mem.count + 2;
338 p->idx += p3reloc.count + 2;
339
340 header = radeon_get_ib_value(p, h_idx);
341 crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
342 reg = CP_PACKET0_GET_REG(header);
343 obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
344 if (!obj) {
345 DRM_ERROR("cannot find crtc %d\n", crtc_id);
346 return -EINVAL;
347 }
348 crtc = obj_to_crtc(obj);
349 radeon_crtc = to_radeon_crtc(crtc);
350 crtc_id = radeon_crtc->crtc_id;
351
352 if (!crtc->enabled) {
353 /* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
354 ib[h_idx + 2] = PACKET2(0);
355 ib[h_idx + 3] = PACKET2(0);
356 ib[h_idx + 4] = PACKET2(0);
357 ib[h_idx + 5] = PACKET2(0);
358 ib[h_idx + 6] = PACKET2(0);
359 ib[h_idx + 7] = PACKET2(0);
360 ib[h_idx + 8] = PACKET2(0);
361 } else {
362 switch (reg) {
363 case EVERGREEN_VLINE_START_END:
364 header &= ~R600_CP_PACKET0_REG_MASK;
365 header |= (EVERGREEN_VLINE_START_END + radeon_crtc->crtc_offset) >> 2;
366 ib[h_idx] = header;
367 ib[h_idx + 4] = (EVERGREEN_VLINE_STATUS + radeon_crtc->crtc_offset) >> 2;
368 break;
369 default:
370 DRM_ERROR("unknown crtc reloc\n");
371 return -EINVAL;
372 }
373 }
374 return 0;
375 }
376
377 static int evergreen_packet0_check(struct radeon_cs_parser *p,
378 struct radeon_cs_packet *pkt,
379 unsigned idx, unsigned reg)
380 {
381 int r;
382
383 switch (reg) {
384 case EVERGREEN_VLINE_START_END:
385 r = evergreen_cs_packet_parse_vline(p);
386 if (r) {
387 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
388 idx, reg);
389 return r;
390 }
391 break;
392 default:
393 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
394 reg, idx);
395 return -EINVAL;
396 }
397 return 0;
398 }
399
400 static int evergreen_cs_parse_packet0(struct radeon_cs_parser *p,
401 struct radeon_cs_packet *pkt)
402 {
403 unsigned reg, i;
404 unsigned idx;
405 int r;
406
407 idx = pkt->idx + 1;
408 reg = pkt->reg;
409 for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
410 r = evergreen_packet0_check(p, pkt, idx, reg);
411 if (r) {
412 return r;
413 }
414 }
415 return 0;
416 }
417
418 /**
419 * evergreen_cs_check_reg() - check if register is authorized or not
420 * @parser: parser structure holding parsing context
421 * @reg: register we are testing
422 * @idx: index into the cs buffer
423 *
424 * This function will test against evergreen_reg_safe_bm and return 0
425 * if register is safe. If register is not flag as safe this function
426 * will test it against a list of register needind special handling.
427 */
428 static int evergreen_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
429 {
430 struct evergreen_cs_track *track = (struct evergreen_cs_track *)p->track;
431 struct radeon_cs_reloc *reloc;
432 u32 last_reg;
433 u32 m, i, tmp, *ib;
434 int r;
435
436 if (p->rdev->family >= CHIP_CAYMAN)
437 last_reg = ARRAY_SIZE(cayman_reg_safe_bm);
438 else
439 last_reg = ARRAY_SIZE(evergreen_reg_safe_bm);
440
441 i = (reg >> 7);
442 if (i >= last_reg) {
443 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
444 return -EINVAL;
445 }
446 m = 1 << ((reg >> 2) & 31);
447 if (p->rdev->family >= CHIP_CAYMAN) {
448 if (!(cayman_reg_safe_bm[i] & m))
449 return 0;
450 } else {
451 if (!(evergreen_reg_safe_bm[i] & m))
452 return 0;
453 }
454 ib = p->ib->ptr;
455 switch (reg) {
456 /* force following reg to 0 in an attempt to disable out buffer
457 * which will need us to better understand how it works to perform
458 * security check on it (Jerome)
459 */
460 case SQ_ESGS_RING_SIZE:
461 case SQ_GSVS_RING_SIZE:
462 case SQ_ESTMP_RING_SIZE:
463 case SQ_GSTMP_RING_SIZE:
464 case SQ_HSTMP_RING_SIZE:
465 case SQ_LSTMP_RING_SIZE:
466 case SQ_PSTMP_RING_SIZE:
467 case SQ_VSTMP_RING_SIZE:
468 case SQ_ESGS_RING_ITEMSIZE:
469 case SQ_ESTMP_RING_ITEMSIZE:
470 case SQ_GSTMP_RING_ITEMSIZE:
471 case SQ_GSVS_RING_ITEMSIZE:
472 case SQ_GS_VERT_ITEMSIZE:
473 case SQ_GS_VERT_ITEMSIZE_1:
474 case SQ_GS_VERT_ITEMSIZE_2:
475 case SQ_GS_VERT_ITEMSIZE_3:
476 case SQ_GSVS_RING_OFFSET_1:
477 case SQ_GSVS_RING_OFFSET_2:
478 case SQ_GSVS_RING_OFFSET_3:
479 case SQ_HSTMP_RING_ITEMSIZE:
480 case SQ_LSTMP_RING_ITEMSIZE:
481 case SQ_PSTMP_RING_ITEMSIZE:
482 case SQ_VSTMP_RING_ITEMSIZE:
483 case VGT_TF_RING_SIZE:
484 /* get value to populate the IB don't remove */
485 /*tmp =radeon_get_ib_value(p, idx);
486 ib[idx] = 0;*/
487 break;
488 case SQ_ESGS_RING_BASE:
489 case SQ_GSVS_RING_BASE:
490 case SQ_ESTMP_RING_BASE:
491 case SQ_GSTMP_RING_BASE:
492 case SQ_HSTMP_RING_BASE:
493 case SQ_LSTMP_RING_BASE:
494 case SQ_PSTMP_RING_BASE:
495 case SQ_VSTMP_RING_BASE:
496 r = evergreen_cs_packet_next_reloc(p, &reloc);
497 if (r) {
498 dev_warn(p->dev, "bad SET_CONTEXT_REG "
499 "0x%04X\n", reg);
500 return -EINVAL;
501 }
502 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
503 break;
504 case DB_DEPTH_CONTROL:
505 track->db_depth_control = radeon_get_ib_value(p, idx);
506 break;
507 case CAYMAN_DB_EQAA:
508 if (p->rdev->family < CHIP_CAYMAN) {
509 dev_warn(p->dev, "bad SET_CONTEXT_REG "
510 "0x%04X\n", reg);
511 return -EINVAL;
512 }
513 break;
514 case CAYMAN_DB_DEPTH_INFO:
515 if (p->rdev->family < CHIP_CAYMAN) {
516 dev_warn(p->dev, "bad SET_CONTEXT_REG "
517 "0x%04X\n", reg);
518 return -EINVAL;
519 }
520 break;
521 case DB_Z_INFO:
522 track->db_z_info = radeon_get_ib_value(p, idx);
523 if (!p->keep_tiling_flags) {
524 r = evergreen_cs_packet_next_reloc(p, &reloc);
525 if (r) {
526 dev_warn(p->dev, "bad SET_CONTEXT_REG "
527 "0x%04X\n", reg);
528 return -EINVAL;
529 }
530 ib[idx] &= ~Z_ARRAY_MODE(0xf);
531 track->db_z_info &= ~Z_ARRAY_MODE(0xf);
532 ib[idx] |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
533 track->db_z_info |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
534 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
535 ib[idx] |= DB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
536 ib[idx] |= DB_TILE_SPLIT(evergreen_cs_get_tile_split(track->row_size));
537 }
538 }
539 break;
540 case DB_STENCIL_INFO:
541 track->db_s_info = radeon_get_ib_value(p, idx);
542 break;
543 case DB_DEPTH_VIEW:
544 track->db_depth_view = radeon_get_ib_value(p, idx);
545 break;
546 case DB_DEPTH_SIZE:
547 track->db_depth_size = radeon_get_ib_value(p, idx);
548 track->db_depth_size_idx = idx;
549 break;
550 case DB_Z_READ_BASE:
551 r = evergreen_cs_packet_next_reloc(p, &reloc);
552 if (r) {
553 dev_warn(p->dev, "bad SET_CONTEXT_REG "
554 "0x%04X\n", reg);
555 return -EINVAL;
556 }
557 track->db_z_read_offset = radeon_get_ib_value(p, idx);
558 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
559 track->db_z_read_bo = reloc->robj;
560 break;
561 case DB_Z_WRITE_BASE:
562 r = evergreen_cs_packet_next_reloc(p, &reloc);
563 if (r) {
564 dev_warn(p->dev, "bad SET_CONTEXT_REG "
565 "0x%04X\n", reg);
566 return -EINVAL;
567 }
568 track->db_z_write_offset = radeon_get_ib_value(p, idx);
569 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
570 track->db_z_write_bo = reloc->robj;
571 break;
572 case DB_STENCIL_READ_BASE:
573 r = evergreen_cs_packet_next_reloc(p, &reloc);
574 if (r) {
575 dev_warn(p->dev, "bad SET_CONTEXT_REG "
576 "0x%04X\n", reg);
577 return -EINVAL;
578 }
579 track->db_s_read_offset = radeon_get_ib_value(p, idx);
580 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
581 track->db_s_read_bo = reloc->robj;
582 break;
583 case DB_STENCIL_WRITE_BASE:
584 r = evergreen_cs_packet_next_reloc(p, &reloc);
585 if (r) {
586 dev_warn(p->dev, "bad SET_CONTEXT_REG "
587 "0x%04X\n", reg);
588 return -EINVAL;
589 }
590 track->db_s_write_offset = radeon_get_ib_value(p, idx);
591 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
592 track->db_s_write_bo = reloc->robj;
593 break;
594 case VGT_STRMOUT_CONFIG:
595 track->vgt_strmout_config = radeon_get_ib_value(p, idx);
596 break;
597 case VGT_STRMOUT_BUFFER_CONFIG:
598 track->vgt_strmout_buffer_config = radeon_get_ib_value(p, idx);
599 break;
600 case CB_TARGET_MASK:
601 track->cb_target_mask = radeon_get_ib_value(p, idx);
602 break;
603 case CB_SHADER_MASK:
604 track->cb_shader_mask = radeon_get_ib_value(p, idx);
605 break;
606 case PA_SC_AA_CONFIG:
607 if (p->rdev->family >= CHIP_CAYMAN) {
608 dev_warn(p->dev, "bad SET_CONTEXT_REG "
609 "0x%04X\n", reg);
610 return -EINVAL;
611 }
612 tmp = radeon_get_ib_value(p, idx) & MSAA_NUM_SAMPLES_MASK;
613 track->nsamples = 1 << tmp;
614 break;
615 case CAYMAN_PA_SC_AA_CONFIG:
616 if (p->rdev->family < CHIP_CAYMAN) {
617 dev_warn(p->dev, "bad SET_CONTEXT_REG "
618 "0x%04X\n", reg);
619 return -EINVAL;
620 }
621 tmp = radeon_get_ib_value(p, idx) & CAYMAN_MSAA_NUM_SAMPLES_MASK;
622 track->nsamples = 1 << tmp;
623 break;
624 case CB_COLOR0_VIEW:
625 case CB_COLOR1_VIEW:
626 case CB_COLOR2_VIEW:
627 case CB_COLOR3_VIEW:
628 case CB_COLOR4_VIEW:
629 case CB_COLOR5_VIEW:
630 case CB_COLOR6_VIEW:
631 case CB_COLOR7_VIEW:
632 tmp = (reg - CB_COLOR0_VIEW) / 0x3c;
633 track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
634 break;
635 case CB_COLOR8_VIEW:
636 case CB_COLOR9_VIEW:
637 case CB_COLOR10_VIEW:
638 case CB_COLOR11_VIEW:
639 tmp = ((reg - CB_COLOR8_VIEW) / 0x1c) + 8;
640 track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
641 break;
642 case CB_COLOR0_INFO:
643 case CB_COLOR1_INFO:
644 case CB_COLOR2_INFO:
645 case CB_COLOR3_INFO:
646 case CB_COLOR4_INFO:
647 case CB_COLOR5_INFO:
648 case CB_COLOR6_INFO:
649 case CB_COLOR7_INFO:
650 tmp = (reg - CB_COLOR0_INFO) / 0x3c;
651 track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
652 if (!p->keep_tiling_flags) {
653 r = evergreen_cs_packet_next_reloc(p, &reloc);
654 if (r) {
655 dev_warn(p->dev, "bad SET_CONTEXT_REG "
656 "0x%04X\n", reg);
657 return -EINVAL;
658 }
659 ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
660 track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
661 }
662 break;
663 case CB_COLOR8_INFO:
664 case CB_COLOR9_INFO:
665 case CB_COLOR10_INFO:
666 case CB_COLOR11_INFO:
667 tmp = ((reg - CB_COLOR8_INFO) / 0x1c) + 8;
668 track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
669 if (!p->keep_tiling_flags) {
670 r = evergreen_cs_packet_next_reloc(p, &reloc);
671 if (r) {
672 dev_warn(p->dev, "bad SET_CONTEXT_REG "
673 "0x%04X\n", reg);
674 return -EINVAL;
675 }
676 ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
677 track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
678 }
679 break;
680 case CB_COLOR0_PITCH:
681 case CB_COLOR1_PITCH:
682 case CB_COLOR2_PITCH:
683 case CB_COLOR3_PITCH:
684 case CB_COLOR4_PITCH:
685 case CB_COLOR5_PITCH:
686 case CB_COLOR6_PITCH:
687 case CB_COLOR7_PITCH:
688 tmp = (reg - CB_COLOR0_PITCH) / 0x3c;
689 track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
690 track->cb_color_pitch_idx[tmp] = idx;
691 break;
692 case CB_COLOR8_PITCH:
693 case CB_COLOR9_PITCH:
694 case CB_COLOR10_PITCH:
695 case CB_COLOR11_PITCH:
696 tmp = ((reg - CB_COLOR8_PITCH) / 0x1c) + 8;
697 track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
698 track->cb_color_pitch_idx[tmp] = idx;
699 break;
700 case CB_COLOR0_SLICE:
701 case CB_COLOR1_SLICE:
702 case CB_COLOR2_SLICE:
703 case CB_COLOR3_SLICE:
704 case CB_COLOR4_SLICE:
705 case CB_COLOR5_SLICE:
706 case CB_COLOR6_SLICE:
707 case CB_COLOR7_SLICE:
708 tmp = (reg - CB_COLOR0_SLICE) / 0x3c;
709 track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
710 track->cb_color_slice_idx[tmp] = idx;
711 break;
712 case CB_COLOR8_SLICE:
713 case CB_COLOR9_SLICE:
714 case CB_COLOR10_SLICE:
715 case CB_COLOR11_SLICE:
716 tmp = ((reg - CB_COLOR8_SLICE) / 0x1c) + 8;
717 track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
718 track->cb_color_slice_idx[tmp] = idx;
719 break;
720 case CB_COLOR0_ATTRIB:
721 case CB_COLOR1_ATTRIB:
722 case CB_COLOR2_ATTRIB:
723 case CB_COLOR3_ATTRIB:
724 case CB_COLOR4_ATTRIB:
725 case CB_COLOR5_ATTRIB:
726 case CB_COLOR6_ATTRIB:
727 case CB_COLOR7_ATTRIB:
728 case CB_COLOR8_ATTRIB:
729 case CB_COLOR9_ATTRIB:
730 case CB_COLOR10_ATTRIB:
731 case CB_COLOR11_ATTRIB:
732 r = evergreen_cs_packet_next_reloc(p, &reloc);
733 if (r) {
734 dev_warn(p->dev, "bad SET_CONTEXT_REG "
735 "0x%04X\n", reg);
736 return -EINVAL;
737 }
738 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
739 ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
740 ib[idx] |= CB_TILE_SPLIT(evergreen_cs_get_tile_split(track->row_size));
741 }
742 break;
743 case CB_COLOR0_DIM:
744 case CB_COLOR1_DIM:
745 case CB_COLOR2_DIM:
746 case CB_COLOR3_DIM:
747 case CB_COLOR4_DIM:
748 case CB_COLOR5_DIM:
749 case CB_COLOR6_DIM:
750 case CB_COLOR7_DIM:
751 tmp = (reg - CB_COLOR0_DIM) / 0x3c;
752 track->cb_color_dim[tmp] = radeon_get_ib_value(p, idx);
753 track->cb_color_dim_idx[tmp] = idx;
754 break;
755 case CB_COLOR8_DIM:
756 case CB_COLOR9_DIM:
757 case CB_COLOR10_DIM:
758 case CB_COLOR11_DIM:
759 tmp = ((reg - CB_COLOR8_DIM) / 0x1c) + 8;
760 track->cb_color_dim[tmp] = radeon_get_ib_value(p, idx);
761 track->cb_color_dim_idx[tmp] = idx;
762 break;
763 case CB_COLOR0_FMASK:
764 case CB_COLOR1_FMASK:
765 case CB_COLOR2_FMASK:
766 case CB_COLOR3_FMASK:
767 case CB_COLOR4_FMASK:
768 case CB_COLOR5_FMASK:
769 case CB_COLOR6_FMASK:
770 case CB_COLOR7_FMASK:
771 tmp = (reg - CB_COLOR0_FMASK) / 0x3c;
772 r = evergreen_cs_packet_next_reloc(p, &reloc);
773 if (r) {
774 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
775 return -EINVAL;
776 }
777 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
778 track->cb_color_fmask_bo[tmp] = reloc->robj;
779 break;
780 case CB_COLOR0_CMASK:
781 case CB_COLOR1_CMASK:
782 case CB_COLOR2_CMASK:
783 case CB_COLOR3_CMASK:
784 case CB_COLOR4_CMASK:
785 case CB_COLOR5_CMASK:
786 case CB_COLOR6_CMASK:
787 case CB_COLOR7_CMASK:
788 tmp = (reg - CB_COLOR0_CMASK) / 0x3c;
789 r = evergreen_cs_packet_next_reloc(p, &reloc);
790 if (r) {
791 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
792 return -EINVAL;
793 }
794 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
795 track->cb_color_cmask_bo[tmp] = reloc->robj;
796 break;
797 case CB_COLOR0_FMASK_SLICE:
798 case CB_COLOR1_FMASK_SLICE:
799 case CB_COLOR2_FMASK_SLICE:
800 case CB_COLOR3_FMASK_SLICE:
801 case CB_COLOR4_FMASK_SLICE:
802 case CB_COLOR5_FMASK_SLICE:
803 case CB_COLOR6_FMASK_SLICE:
804 case CB_COLOR7_FMASK_SLICE:
805 tmp = (reg - CB_COLOR0_FMASK_SLICE) / 0x3c;
806 track->cb_color_fmask_slice[tmp] = radeon_get_ib_value(p, idx);
807 break;
808 case CB_COLOR0_CMASK_SLICE:
809 case CB_COLOR1_CMASK_SLICE:
810 case CB_COLOR2_CMASK_SLICE:
811 case CB_COLOR3_CMASK_SLICE:
812 case CB_COLOR4_CMASK_SLICE:
813 case CB_COLOR5_CMASK_SLICE:
814 case CB_COLOR6_CMASK_SLICE:
815 case CB_COLOR7_CMASK_SLICE:
816 tmp = (reg - CB_COLOR0_CMASK_SLICE) / 0x3c;
817 track->cb_color_cmask_slice[tmp] = radeon_get_ib_value(p, idx);
818 break;
819 case CB_COLOR0_BASE:
820 case CB_COLOR1_BASE:
821 case CB_COLOR2_BASE:
822 case CB_COLOR3_BASE:
823 case CB_COLOR4_BASE:
824 case CB_COLOR5_BASE:
825 case CB_COLOR6_BASE:
826 case CB_COLOR7_BASE:
827 r = evergreen_cs_packet_next_reloc(p, &reloc);
828 if (r) {
829 dev_warn(p->dev, "bad SET_CONTEXT_REG "
830 "0x%04X\n", reg);
831 return -EINVAL;
832 }
833 tmp = (reg - CB_COLOR0_BASE) / 0x3c;
834 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
835 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
836 track->cb_color_base_last[tmp] = ib[idx];
837 track->cb_color_bo[tmp] = reloc->robj;
838 break;
839 case CB_COLOR8_BASE:
840 case CB_COLOR9_BASE:
841 case CB_COLOR10_BASE:
842 case CB_COLOR11_BASE:
843 r = evergreen_cs_packet_next_reloc(p, &reloc);
844 if (r) {
845 dev_warn(p->dev, "bad SET_CONTEXT_REG "
846 "0x%04X\n", reg);
847 return -EINVAL;
848 }
849 tmp = ((reg - CB_COLOR8_BASE) / 0x1c) + 8;
850 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
851 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
852 track->cb_color_base_last[tmp] = ib[idx];
853 track->cb_color_bo[tmp] = reloc->robj;
854 break;
855 case CB_IMMED0_BASE:
856 case CB_IMMED1_BASE:
857 case CB_IMMED2_BASE:
858 case CB_IMMED3_BASE:
859 case CB_IMMED4_BASE:
860 case CB_IMMED5_BASE:
861 case CB_IMMED6_BASE:
862 case CB_IMMED7_BASE:
863 case CB_IMMED8_BASE:
864 case CB_IMMED9_BASE:
865 case CB_IMMED10_BASE:
866 case CB_IMMED11_BASE:
867 case DB_HTILE_DATA_BASE:
868 case SQ_PGM_START_FS:
869 case SQ_PGM_START_ES:
870 case SQ_PGM_START_VS:
871 case SQ_PGM_START_GS:
872 case SQ_PGM_START_PS:
873 case SQ_PGM_START_HS:
874 case SQ_PGM_START_LS:
875 case SQ_CONST_MEM_BASE:
876 case SQ_ALU_CONST_CACHE_GS_0:
877 case SQ_ALU_CONST_CACHE_GS_1:
878 case SQ_ALU_CONST_CACHE_GS_2:
879 case SQ_ALU_CONST_CACHE_GS_3:
880 case SQ_ALU_CONST_CACHE_GS_4:
881 case SQ_ALU_CONST_CACHE_GS_5:
882 case SQ_ALU_CONST_CACHE_GS_6:
883 case SQ_ALU_CONST_CACHE_GS_7:
884 case SQ_ALU_CONST_CACHE_GS_8:
885 case SQ_ALU_CONST_CACHE_GS_9:
886 case SQ_ALU_CONST_CACHE_GS_10:
887 case SQ_ALU_CONST_CACHE_GS_11:
888 case SQ_ALU_CONST_CACHE_GS_12:
889 case SQ_ALU_CONST_CACHE_GS_13:
890 case SQ_ALU_CONST_CACHE_GS_14:
891 case SQ_ALU_CONST_CACHE_GS_15:
892 case SQ_ALU_CONST_CACHE_PS_0:
893 case SQ_ALU_CONST_CACHE_PS_1:
894 case SQ_ALU_CONST_CACHE_PS_2:
895 case SQ_ALU_CONST_CACHE_PS_3:
896 case SQ_ALU_CONST_CACHE_PS_4:
897 case SQ_ALU_CONST_CACHE_PS_5:
898 case SQ_ALU_CONST_CACHE_PS_6:
899 case SQ_ALU_CONST_CACHE_PS_7:
900 case SQ_ALU_CONST_CACHE_PS_8:
901 case SQ_ALU_CONST_CACHE_PS_9:
902 case SQ_ALU_CONST_CACHE_PS_10:
903 case SQ_ALU_CONST_CACHE_PS_11:
904 case SQ_ALU_CONST_CACHE_PS_12:
905 case SQ_ALU_CONST_CACHE_PS_13:
906 case SQ_ALU_CONST_CACHE_PS_14:
907 case SQ_ALU_CONST_CACHE_PS_15:
908 case SQ_ALU_CONST_CACHE_VS_0:
909 case SQ_ALU_CONST_CACHE_VS_1:
910 case SQ_ALU_CONST_CACHE_VS_2:
911 case SQ_ALU_CONST_CACHE_VS_3:
912 case SQ_ALU_CONST_CACHE_VS_4:
913 case SQ_ALU_CONST_CACHE_VS_5:
914 case SQ_ALU_CONST_CACHE_VS_6:
915 case SQ_ALU_CONST_CACHE_VS_7:
916 case SQ_ALU_CONST_CACHE_VS_8:
917 case SQ_ALU_CONST_CACHE_VS_9:
918 case SQ_ALU_CONST_CACHE_VS_10:
919 case SQ_ALU_CONST_CACHE_VS_11:
920 case SQ_ALU_CONST_CACHE_VS_12:
921 case SQ_ALU_CONST_CACHE_VS_13:
922 case SQ_ALU_CONST_CACHE_VS_14:
923 case SQ_ALU_CONST_CACHE_VS_15:
924 case SQ_ALU_CONST_CACHE_HS_0:
925 case SQ_ALU_CONST_CACHE_HS_1:
926 case SQ_ALU_CONST_CACHE_HS_2:
927 case SQ_ALU_CONST_CACHE_HS_3:
928 case SQ_ALU_CONST_CACHE_HS_4:
929 case SQ_ALU_CONST_CACHE_HS_5:
930 case SQ_ALU_CONST_CACHE_HS_6:
931 case SQ_ALU_CONST_CACHE_HS_7:
932 case SQ_ALU_CONST_CACHE_HS_8:
933 case SQ_ALU_CONST_CACHE_HS_9:
934 case SQ_ALU_CONST_CACHE_HS_10:
935 case SQ_ALU_CONST_CACHE_HS_11:
936 case SQ_ALU_CONST_CACHE_HS_12:
937 case SQ_ALU_CONST_CACHE_HS_13:
938 case SQ_ALU_CONST_CACHE_HS_14:
939 case SQ_ALU_CONST_CACHE_HS_15:
940 case SQ_ALU_CONST_CACHE_LS_0:
941 case SQ_ALU_CONST_CACHE_LS_1:
942 case SQ_ALU_CONST_CACHE_LS_2:
943 case SQ_ALU_CONST_CACHE_LS_3:
944 case SQ_ALU_CONST_CACHE_LS_4:
945 case SQ_ALU_CONST_CACHE_LS_5:
946 case SQ_ALU_CONST_CACHE_LS_6:
947 case SQ_ALU_CONST_CACHE_LS_7:
948 case SQ_ALU_CONST_CACHE_LS_8:
949 case SQ_ALU_CONST_CACHE_LS_9:
950 case SQ_ALU_CONST_CACHE_LS_10:
951 case SQ_ALU_CONST_CACHE_LS_11:
952 case SQ_ALU_CONST_CACHE_LS_12:
953 case SQ_ALU_CONST_CACHE_LS_13:
954 case SQ_ALU_CONST_CACHE_LS_14:
955 case SQ_ALU_CONST_CACHE_LS_15:
956 r = evergreen_cs_packet_next_reloc(p, &reloc);
957 if (r) {
958 dev_warn(p->dev, "bad SET_CONTEXT_REG "
959 "0x%04X\n", reg);
960 return -EINVAL;
961 }
962 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
963 break;
964 case SX_MEMORY_EXPORT_BASE:
965 if (p->rdev->family >= CHIP_CAYMAN) {
966 dev_warn(p->dev, "bad SET_CONFIG_REG "
967 "0x%04X\n", reg);
968 return -EINVAL;
969 }
970 r = evergreen_cs_packet_next_reloc(p, &reloc);
971 if (r) {
972 dev_warn(p->dev, "bad SET_CONFIG_REG "
973 "0x%04X\n", reg);
974 return -EINVAL;
975 }
976 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
977 break;
978 case CAYMAN_SX_SCATTER_EXPORT_BASE:
979 if (p->rdev->family < CHIP_CAYMAN) {
980 dev_warn(p->dev, "bad SET_CONTEXT_REG "
981 "0x%04X\n", reg);
982 return -EINVAL;
983 }
984 r = evergreen_cs_packet_next_reloc(p, &reloc);
985 if (r) {
986 dev_warn(p->dev, "bad SET_CONTEXT_REG "
987 "0x%04X\n", reg);
988 return -EINVAL;
989 }
990 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
991 break;
992 default:
993 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
994 return -EINVAL;
995 }
996 return 0;
997 }
998
999 /**
1000 * evergreen_check_texture_resource() - check if register is authorized or not
1001 * @p: parser structure holding parsing context
1002 * @idx: index into the cs buffer
1003 * @texture: texture's bo structure
1004 * @mipmap: mipmap's bo structure
1005 *
1006 * This function will check that the resource has valid field and that
1007 * the texture and mipmap bo object are big enough to cover this resource.
1008 */
1009 static int evergreen_check_texture_resource(struct radeon_cs_parser *p, u32 idx,
1010 struct radeon_bo *texture,
1011 struct radeon_bo *mipmap)
1012 {
1013 /* XXX fill in */
1014 return 0;
1015 }
1016
1017 static int evergreen_packet3_check(struct radeon_cs_parser *p,
1018 struct radeon_cs_packet *pkt)
1019 {
1020 struct radeon_cs_reloc *reloc;
1021 struct evergreen_cs_track *track;
1022 volatile u32 *ib;
1023 unsigned idx;
1024 unsigned i;
1025 unsigned start_reg, end_reg, reg;
1026 int r;
1027 u32 idx_value;
1028
1029 track = (struct evergreen_cs_track *)p->track;
1030 ib = p->ib->ptr;
1031 idx = pkt->idx + 1;
1032 idx_value = radeon_get_ib_value(p, idx);
1033
1034 switch (pkt->opcode) {
1035 case PACKET3_SET_PREDICATION:
1036 {
1037 int pred_op;
1038 int tmp;
1039 if (pkt->count != 1) {
1040 DRM_ERROR("bad SET PREDICATION\n");
1041 return -EINVAL;
1042 }
1043
1044 tmp = radeon_get_ib_value(p, idx + 1);
1045 pred_op = (tmp >> 16) & 0x7;
1046
1047 /* for the clear predicate operation */
1048 if (pred_op == 0)
1049 return 0;
1050
1051 if (pred_op > 2) {
1052 DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1053 return -EINVAL;
1054 }
1055
1056 r = evergreen_cs_packet_next_reloc(p, &reloc);
1057 if (r) {
1058 DRM_ERROR("bad SET PREDICATION\n");
1059 return -EINVAL;
1060 }
1061
1062 ib[idx + 0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1063 ib[idx + 1] = tmp + (upper_32_bits(reloc->lobj.gpu_offset) & 0xff);
1064 }
1065 break;
1066 case PACKET3_CONTEXT_CONTROL:
1067 if (pkt->count != 1) {
1068 DRM_ERROR("bad CONTEXT_CONTROL\n");
1069 return -EINVAL;
1070 }
1071 break;
1072 case PACKET3_INDEX_TYPE:
1073 case PACKET3_NUM_INSTANCES:
1074 case PACKET3_CLEAR_STATE:
1075 if (pkt->count) {
1076 DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
1077 return -EINVAL;
1078 }
1079 break;
1080 case CAYMAN_PACKET3_DEALLOC_STATE:
1081 if (p->rdev->family < CHIP_CAYMAN) {
1082 DRM_ERROR("bad PACKET3_DEALLOC_STATE\n");
1083 return -EINVAL;
1084 }
1085 if (pkt->count) {
1086 DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
1087 return -EINVAL;
1088 }
1089 break;
1090 case PACKET3_INDEX_BASE:
1091 if (pkt->count != 1) {
1092 DRM_ERROR("bad INDEX_BASE\n");
1093 return -EINVAL;
1094 }
1095 r = evergreen_cs_packet_next_reloc(p, &reloc);
1096 if (r) {
1097 DRM_ERROR("bad INDEX_BASE\n");
1098 return -EINVAL;
1099 }
1100 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1101 ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1102 r = evergreen_cs_track_check(p);
1103 if (r) {
1104 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1105 return r;
1106 }
1107 break;
1108 case PACKET3_DRAW_INDEX:
1109 if (pkt->count != 3) {
1110 DRM_ERROR("bad DRAW_INDEX\n");
1111 return -EINVAL;
1112 }
1113 r = evergreen_cs_packet_next_reloc(p, &reloc);
1114 if (r) {
1115 DRM_ERROR("bad DRAW_INDEX\n");
1116 return -EINVAL;
1117 }
1118 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1119 ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1120 r = evergreen_cs_track_check(p);
1121 if (r) {
1122 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1123 return r;
1124 }
1125 break;
1126 case PACKET3_DRAW_INDEX_2:
1127 if (pkt->count != 4) {
1128 DRM_ERROR("bad DRAW_INDEX_2\n");
1129 return -EINVAL;
1130 }
1131 r = evergreen_cs_packet_next_reloc(p, &reloc);
1132 if (r) {
1133 DRM_ERROR("bad DRAW_INDEX_2\n");
1134 return -EINVAL;
1135 }
1136 ib[idx+1] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1137 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1138 r = evergreen_cs_track_check(p);
1139 if (r) {
1140 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1141 return r;
1142 }
1143 break;
1144 case PACKET3_DRAW_INDEX_AUTO:
1145 if (pkt->count != 1) {
1146 DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1147 return -EINVAL;
1148 }
1149 r = evergreen_cs_track_check(p);
1150 if (r) {
1151 dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1152 return r;
1153 }
1154 break;
1155 case PACKET3_DRAW_INDEX_MULTI_AUTO:
1156 if (pkt->count != 2) {
1157 DRM_ERROR("bad DRAW_INDEX_MULTI_AUTO\n");
1158 return -EINVAL;
1159 }
1160 r = evergreen_cs_track_check(p);
1161 if (r) {
1162 dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1163 return r;
1164 }
1165 break;
1166 case PACKET3_DRAW_INDEX_IMMD:
1167 if (pkt->count < 2) {
1168 DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1169 return -EINVAL;
1170 }
1171 r = evergreen_cs_track_check(p);
1172 if (r) {
1173 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1174 return r;
1175 }
1176 break;
1177 case PACKET3_DRAW_INDEX_OFFSET:
1178 if (pkt->count != 2) {
1179 DRM_ERROR("bad DRAW_INDEX_OFFSET\n");
1180 return -EINVAL;
1181 }
1182 r = evergreen_cs_track_check(p);
1183 if (r) {
1184 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1185 return r;
1186 }
1187 break;
1188 case PACKET3_DRAW_INDEX_OFFSET_2:
1189 if (pkt->count != 3) {
1190 DRM_ERROR("bad DRAW_INDEX_OFFSET_2\n");
1191 return -EINVAL;
1192 }
1193 r = evergreen_cs_track_check(p);
1194 if (r) {
1195 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1196 return r;
1197 }
1198 break;
1199 case PACKET3_DISPATCH_DIRECT:
1200 if (pkt->count != 3) {
1201 DRM_ERROR("bad DISPATCH_DIRECT\n");
1202 return -EINVAL;
1203 }
1204 r = evergreen_cs_track_check(p);
1205 if (r) {
1206 dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1207 return r;
1208 }
1209 break;
1210 case PACKET3_DISPATCH_INDIRECT:
1211 if (pkt->count != 1) {
1212 DRM_ERROR("bad DISPATCH_INDIRECT\n");
1213 return -EINVAL;
1214 }
1215 r = evergreen_cs_packet_next_reloc(p, &reloc);
1216 if (r) {
1217 DRM_ERROR("bad DISPATCH_INDIRECT\n");
1218 return -EINVAL;
1219 }
1220 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1221 r = evergreen_cs_track_check(p);
1222 if (r) {
1223 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1224 return r;
1225 }
1226 break;
1227 case PACKET3_WAIT_REG_MEM:
1228 if (pkt->count != 5) {
1229 DRM_ERROR("bad WAIT_REG_MEM\n");
1230 return -EINVAL;
1231 }
1232 /* bit 4 is reg (0) or mem (1) */
1233 if (idx_value & 0x10) {
1234 r = evergreen_cs_packet_next_reloc(p, &reloc);
1235 if (r) {
1236 DRM_ERROR("bad WAIT_REG_MEM\n");
1237 return -EINVAL;
1238 }
1239 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1240 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1241 }
1242 break;
1243 case PACKET3_SURFACE_SYNC:
1244 if (pkt->count != 3) {
1245 DRM_ERROR("bad SURFACE_SYNC\n");
1246 return -EINVAL;
1247 }
1248 /* 0xffffffff/0x0 is flush all cache flag */
1249 if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1250 radeon_get_ib_value(p, idx + 2) != 0) {
1251 r = evergreen_cs_packet_next_reloc(p, &reloc);
1252 if (r) {
1253 DRM_ERROR("bad SURFACE_SYNC\n");
1254 return -EINVAL;
1255 }
1256 ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1257 }
1258 break;
1259 case PACKET3_EVENT_WRITE:
1260 if (pkt->count != 2 && pkt->count != 0) {
1261 DRM_ERROR("bad EVENT_WRITE\n");
1262 return -EINVAL;
1263 }
1264 if (pkt->count) {
1265 r = evergreen_cs_packet_next_reloc(p, &reloc);
1266 if (r) {
1267 DRM_ERROR("bad EVENT_WRITE\n");
1268 return -EINVAL;
1269 }
1270 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1271 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1272 }
1273 break;
1274 case PACKET3_EVENT_WRITE_EOP:
1275 if (pkt->count != 4) {
1276 DRM_ERROR("bad EVENT_WRITE_EOP\n");
1277 return -EINVAL;
1278 }
1279 r = evergreen_cs_packet_next_reloc(p, &reloc);
1280 if (r) {
1281 DRM_ERROR("bad EVENT_WRITE_EOP\n");
1282 return -EINVAL;
1283 }
1284 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1285 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1286 break;
1287 case PACKET3_EVENT_WRITE_EOS:
1288 if (pkt->count != 3) {
1289 DRM_ERROR("bad EVENT_WRITE_EOS\n");
1290 return -EINVAL;
1291 }
1292 r = evergreen_cs_packet_next_reloc(p, &reloc);
1293 if (r) {
1294 DRM_ERROR("bad EVENT_WRITE_EOS\n");
1295 return -EINVAL;
1296 }
1297 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1298 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1299 break;
1300 case PACKET3_SET_CONFIG_REG:
1301 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
1302 end_reg = 4 * pkt->count + start_reg - 4;
1303 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
1304 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1305 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1306 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1307 return -EINVAL;
1308 }
1309 for (i = 0; i < pkt->count; i++) {
1310 reg = start_reg + (4 * i);
1311 r = evergreen_cs_check_reg(p, reg, idx+1+i);
1312 if (r)
1313 return r;
1314 }
1315 break;
1316 case PACKET3_SET_CONTEXT_REG:
1317 start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_START;
1318 end_reg = 4 * pkt->count + start_reg - 4;
1319 if ((start_reg < PACKET3_SET_CONTEXT_REG_START) ||
1320 (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1321 (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1322 DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1323 return -EINVAL;
1324 }
1325 for (i = 0; i < pkt->count; i++) {
1326 reg = start_reg + (4 * i);
1327 r = evergreen_cs_check_reg(p, reg, idx+1+i);
1328 if (r)
1329 return r;
1330 }
1331 break;
1332 case PACKET3_SET_RESOURCE:
1333 if (pkt->count % 8) {
1334 DRM_ERROR("bad SET_RESOURCE\n");
1335 return -EINVAL;
1336 }
1337 start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_START;
1338 end_reg = 4 * pkt->count + start_reg - 4;
1339 if ((start_reg < PACKET3_SET_RESOURCE_START) ||
1340 (start_reg >= PACKET3_SET_RESOURCE_END) ||
1341 (end_reg >= PACKET3_SET_RESOURCE_END)) {
1342 DRM_ERROR("bad SET_RESOURCE\n");
1343 return -EINVAL;
1344 }
1345 for (i = 0; i < (pkt->count / 8); i++) {
1346 struct radeon_bo *texture, *mipmap;
1347 u32 size, offset;
1348
1349 switch (G__SQ_CONSTANT_TYPE(radeon_get_ib_value(p, idx+1+(i*8)+7))) {
1350 case SQ_TEX_VTX_VALID_TEXTURE:
1351 /* tex base */
1352 r = evergreen_cs_packet_next_reloc(p, &reloc);
1353 if (r) {
1354 DRM_ERROR("bad SET_RESOURCE (tex)\n");
1355 return -EINVAL;
1356 }
1357 ib[idx+1+(i*8)+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1358 if (!p->keep_tiling_flags) {
1359 ib[idx+1+(i*8)+1] |=
1360 TEX_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1361 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1362 ib[idx+1+(i*8)+6] |=
1363 TEX_TILE_SPLIT(evergreen_cs_get_tile_split(track->row_size));
1364 ib[idx+1+(i*8)+7] |=
1365 TEX_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1366 }
1367 }
1368 texture = reloc->robj;
1369 /* tex mip base */
1370 r = evergreen_cs_packet_next_reloc(p, &reloc);
1371 if (r) {
1372 DRM_ERROR("bad SET_RESOURCE (tex)\n");
1373 return -EINVAL;
1374 }
1375 ib[idx+1+(i*8)+3] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1376 mipmap = reloc->robj;
1377 r = evergreen_check_texture_resource(p, idx+1+(i*8),
1378 texture, mipmap);
1379 if (r)
1380 return r;
1381 break;
1382 case SQ_TEX_VTX_VALID_BUFFER:
1383 /* vtx base */
1384 r = evergreen_cs_packet_next_reloc(p, &reloc);
1385 if (r) {
1386 DRM_ERROR("bad SET_RESOURCE (vtx)\n");
1387 return -EINVAL;
1388 }
1389 offset = radeon_get_ib_value(p, idx+1+(i*8)+0);
1390 size = radeon_get_ib_value(p, idx+1+(i*8)+1);
1391 if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
1392 /* force size to size of the buffer */
1393 dev_warn(p->dev, "vbo resource seems too big for the bo\n");
1394 ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj);
1395 }
1396 ib[idx+1+(i*8)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
1397 ib[idx+1+(i*8)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1398 break;
1399 case SQ_TEX_VTX_INVALID_TEXTURE:
1400 case SQ_TEX_VTX_INVALID_BUFFER:
1401 default:
1402 DRM_ERROR("bad SET_RESOURCE\n");
1403 return -EINVAL;
1404 }
1405 }
1406 break;
1407 case PACKET3_SET_ALU_CONST:
1408 /* XXX fix me ALU const buffers only */
1409 break;
1410 case PACKET3_SET_BOOL_CONST:
1411 start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_START;
1412 end_reg = 4 * pkt->count + start_reg - 4;
1413 if ((start_reg < PACKET3_SET_BOOL_CONST_START) ||
1414 (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
1415 (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
1416 DRM_ERROR("bad SET_BOOL_CONST\n");
1417 return -EINVAL;
1418 }
1419 break;
1420 case PACKET3_SET_LOOP_CONST:
1421 start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_START;
1422 end_reg = 4 * pkt->count + start_reg - 4;
1423 if ((start_reg < PACKET3_SET_LOOP_CONST_START) ||
1424 (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
1425 (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
1426 DRM_ERROR("bad SET_LOOP_CONST\n");
1427 return -EINVAL;
1428 }
1429 break;
1430 case PACKET3_SET_CTL_CONST:
1431 start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_START;
1432 end_reg = 4 * pkt->count + start_reg - 4;
1433 if ((start_reg < PACKET3_SET_CTL_CONST_START) ||
1434 (start_reg >= PACKET3_SET_CTL_CONST_END) ||
1435 (end_reg >= PACKET3_SET_CTL_CONST_END)) {
1436 DRM_ERROR("bad SET_CTL_CONST\n");
1437 return -EINVAL;
1438 }
1439 break;
1440 case PACKET3_SET_SAMPLER:
1441 if (pkt->count % 3) {
1442 DRM_ERROR("bad SET_SAMPLER\n");
1443 return -EINVAL;
1444 }
1445 start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_START;
1446 end_reg = 4 * pkt->count + start_reg - 4;
1447 if ((start_reg < PACKET3_SET_SAMPLER_START) ||
1448 (start_reg >= PACKET3_SET_SAMPLER_END) ||
1449 (end_reg >= PACKET3_SET_SAMPLER_END)) {
1450 DRM_ERROR("bad SET_SAMPLER\n");
1451 return -EINVAL;
1452 }
1453 break;
1454 case PACKET3_NOP:
1455 break;
1456 default:
1457 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1458 return -EINVAL;
1459 }
1460 return 0;
1461 }
1462
1463 int evergreen_cs_parse(struct radeon_cs_parser *p)
1464 {
1465 struct radeon_cs_packet pkt;
1466 struct evergreen_cs_track *track;
1467 u32 tmp;
1468 int r;
1469
1470 if (p->track == NULL) {
1471 /* initialize tracker, we are in kms */
1472 track = kzalloc(sizeof(*track), GFP_KERNEL);
1473 if (track == NULL)
1474 return -ENOMEM;
1475 evergreen_cs_track_init(track);
1476 if (p->rdev->family >= CHIP_CAYMAN)
1477 tmp = p->rdev->config.cayman.tile_config;
1478 else
1479 tmp = p->rdev->config.evergreen.tile_config;
1480
1481 switch (tmp & 0xf) {
1482 case 0:
1483 track->npipes = 1;
1484 break;
1485 case 1:
1486 default:
1487 track->npipes = 2;
1488 break;
1489 case 2:
1490 track->npipes = 4;
1491 break;
1492 case 3:
1493 track->npipes = 8;
1494 break;
1495 }
1496
1497 switch ((tmp & 0xf0) >> 4) {
1498 case 0:
1499 track->nbanks = 4;
1500 break;
1501 case 1:
1502 default:
1503 track->nbanks = 8;
1504 break;
1505 case 2:
1506 track->nbanks = 16;
1507 break;
1508 }
1509
1510 switch ((tmp & 0xf00) >> 8) {
1511 case 0:
1512 track->group_size = 256;
1513 break;
1514 case 1:
1515 default:
1516 track->group_size = 512;
1517 break;
1518 }
1519
1520 switch ((tmp & 0xf000) >> 12) {
1521 case 0:
1522 track->row_size = 1;
1523 break;
1524 case 1:
1525 default:
1526 track->row_size = 2;
1527 break;
1528 case 2:
1529 track->row_size = 4;
1530 break;
1531 }
1532
1533 p->track = track;
1534 }
1535 do {
1536 r = evergreen_cs_packet_parse(p, &pkt, p->idx);
1537 if (r) {
1538 kfree(p->track);
1539 p->track = NULL;
1540 return r;
1541 }
1542 p->idx += pkt.count + 2;
1543 switch (pkt.type) {
1544 case PACKET_TYPE0:
1545 r = evergreen_cs_parse_packet0(p, &pkt);
1546 break;
1547 case PACKET_TYPE2:
1548 break;
1549 case PACKET_TYPE3:
1550 r = evergreen_packet3_check(p, &pkt);
1551 break;
1552 default:
1553 DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1554 kfree(p->track);
1555 p->track = NULL;
1556 return -EINVAL;
1557 }
1558 if (r) {
1559 kfree(p->track);
1560 p->track = NULL;
1561 return r;
1562 }
1563 } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1564 #if 0
1565 for (r = 0; r < p->ib->length_dw; r++) {
1566 printk(KERN_INFO "%05d 0x%08X\n", r, p->ib->ptr[r]);
1567 mdelay(1);
1568 }
1569 #endif
1570 kfree(p->track);
1571 p->track = NULL;
1572 return 0;
1573 }
1574
Ubuntu Jammy Linux kernel mirror