2 * Copyright 2010 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
28 * $FreeBSD: head/sys/dev/drm2/radeon/evergreen_cs.c 254885 2013-08-25 19:37:15Z dumbbell $
33 #include "radeon_asic.h"
34 #include "evergreend.h"
35 #include "evergreen_reg_safe.h"
36 #include "cayman_reg_safe.h"
39 #define MAX(a,b) (((a)>(b))?(a):(b))
40 #define MIN(a,b) (((a)<(b))?(a):(b))
42 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
43 struct radeon_cs_reloc **cs_reloc);
45 struct evergreen_cs_track {
51 u32 nsamples; /* unused */
52 struct radeon_bo *cb_color_bo[12];
53 u32 cb_color_bo_offset[12];
54 struct radeon_bo *cb_color_fmask_bo[8]; /* unused */
55 struct radeon_bo *cb_color_cmask_bo[8]; /* unused */
56 u32 cb_color_info[12];
57 u32 cb_color_view[12];
58 u32 cb_color_pitch[12];
59 u32 cb_color_slice[12];
60 u32 cb_color_slice_idx[12];
61 u32 cb_color_attrib[12];
62 u32 cb_color_cmask_slice[8];/* unused */
63 u32 cb_color_fmask_slice[8];/* unused */
65 u32 cb_shader_mask; /* unused */
66 u32 vgt_strmout_config;
67 u32 vgt_strmout_buffer_config;
68 struct radeon_bo *vgt_strmout_bo[4];
69 u32 vgt_strmout_bo_offset[4];
70 u32 vgt_strmout_size[4];
77 u32 db_z_write_offset;
78 struct radeon_bo *db_z_read_bo;
79 struct radeon_bo *db_z_write_bo;
82 u32 db_s_write_offset;
83 struct radeon_bo *db_s_read_bo;
84 struct radeon_bo *db_s_write_bo;
85 bool sx_misc_kill_all_prims;
91 struct radeon_bo *htile_bo;
94 static u32 evergreen_cs_get_aray_mode(u32 tiling_flags)
96 if (tiling_flags & RADEON_TILING_MACRO)
97 return ARRAY_2D_TILED_THIN1;
98 else if (tiling_flags & RADEON_TILING_MICRO)
99 return ARRAY_1D_TILED_THIN1;
101 return ARRAY_LINEAR_GENERAL;
104 static u32 evergreen_cs_get_num_banks(u32 nbanks)
108 return ADDR_SURF_2_BANK;
110 return ADDR_SURF_4_BANK;
113 return ADDR_SURF_8_BANK;
115 return ADDR_SURF_16_BANK;
119 static void evergreen_cs_track_init(struct evergreen_cs_track *track)
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;
130 for (i = 0; i < 12; i++) {
131 track->cb_color_bo[i] = NULL;
132 track->cb_color_bo_offset[i] = 0xFFFFFFFF;
133 track->cb_color_info[i] = 0;
134 track->cb_color_view[i] = 0xFFFFFFFF;
135 track->cb_color_pitch[i] = 0;
136 track->cb_color_slice[i] = 0xfffffff;
137 track->cb_color_slice_idx[i] = 0;
139 track->cb_target_mask = 0xFFFFFFFF;
140 track->cb_shader_mask = 0xFFFFFFFF;
141 track->cb_dirty = true;
143 track->db_depth_slice = 0xffffffff;
144 track->db_depth_view = 0xFFFFC000;
145 track->db_depth_size = 0xFFFFFFFF;
146 track->db_depth_control = 0xFFFFFFFF;
147 track->db_z_info = 0xFFFFFFFF;
148 track->db_z_read_offset = 0xFFFFFFFF;
149 track->db_z_write_offset = 0xFFFFFFFF;
150 track->db_z_read_bo = NULL;
151 track->db_z_write_bo = NULL;
152 track->db_s_info = 0xFFFFFFFF;
153 track->db_s_read_offset = 0xFFFFFFFF;
154 track->db_s_write_offset = 0xFFFFFFFF;
155 track->db_s_read_bo = NULL;
156 track->db_s_write_bo = NULL;
157 track->db_dirty = true;
158 track->htile_bo = NULL;
159 track->htile_offset = 0xFFFFFFFF;
160 track->htile_surface = 0;
162 for (i = 0; i < 4; i++) {
163 track->vgt_strmout_size[i] = 0;
164 track->vgt_strmout_bo[i] = NULL;
165 track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
167 track->streamout_dirty = true;
168 track->sx_misc_kill_all_prims = false;
172 /* value gathered from cs */
188 unsigned long base_align;
191 static int evergreen_surface_check_linear(struct radeon_cs_parser *p,
192 struct eg_surface *surf,
195 surf->layer_size = surf->nbx * surf->nby * surf->bpe * surf->nsamples;
196 surf->base_align = surf->bpe;
202 static int evergreen_surface_check_linear_aligned(struct radeon_cs_parser *p,
203 struct eg_surface *surf,
206 struct evergreen_cs_track *track = p->track;
209 palign = MAX(64, track->group_size / surf->bpe);
210 surf->layer_size = surf->nbx * surf->nby * surf->bpe * surf->nsamples;
211 surf->base_align = track->group_size;
212 surf->palign = palign;
214 if (surf->nbx & (palign - 1)) {
216 dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d\n",
217 __func__, __LINE__, prefix, surf->nbx, palign);
224 static int evergreen_surface_check_1d(struct radeon_cs_parser *p,
225 struct eg_surface *surf,
228 struct evergreen_cs_track *track = p->track;
231 palign = track->group_size / (8 * surf->bpe * surf->nsamples);
232 palign = MAX(8, palign);
233 surf->layer_size = surf->nbx * surf->nby * surf->bpe;
234 surf->base_align = track->group_size;
235 surf->palign = palign;
237 if ((surf->nbx & (palign - 1))) {
239 dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d (%d %d %d)\n",
240 __func__, __LINE__, prefix, surf->nbx, palign,
241 track->group_size, surf->bpe, surf->nsamples);
245 if ((surf->nby & (8 - 1))) {
247 dev_warn(p->dev, "%s:%d %s height %d invalid must be aligned with 8\n",
248 __func__, __LINE__, prefix, surf->nby);
255 static int evergreen_surface_check_2d(struct radeon_cs_parser *p,
256 struct eg_surface *surf,
259 struct evergreen_cs_track *track = p->track;
260 unsigned palign, halign, tileb, slice_pt;
261 unsigned mtile_pr, mtile_ps, mtileb;
263 tileb = 64 * surf->bpe * surf->nsamples;
265 if (tileb > surf->tsplit) {
266 slice_pt = tileb / surf->tsplit;
268 tileb = tileb / slice_pt;
269 /* macro tile width & height */
270 palign = (8 * surf->bankw * track->npipes) * surf->mtilea;
271 halign = (8 * surf->bankh * surf->nbanks) / surf->mtilea;
272 mtileb = (palign / 8) * (halign / 8) * tileb;
273 mtile_pr = surf->nbx / palign;
274 mtile_ps = (mtile_pr * surf->nby) / halign;
275 surf->layer_size = mtile_ps * mtileb * slice_pt;
276 surf->base_align = (palign / 8) * (halign / 8) * tileb;
277 surf->palign = palign;
278 surf->halign = halign;
280 if ((surf->nbx & (palign - 1))) {
282 dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d\n",
283 __func__, __LINE__, prefix, surf->nbx, palign);
287 if ((surf->nby & (halign - 1))) {
289 dev_warn(p->dev, "%s:%d %s height %d invalid must be aligned with %d\n",
290 __func__, __LINE__, prefix, surf->nby, halign);
298 static int evergreen_surface_check(struct radeon_cs_parser *p,
299 struct eg_surface *surf,
302 /* some common value computed here */
303 surf->bpe = r600_fmt_get_blocksize(surf->format);
305 switch (surf->mode) {
306 case ARRAY_LINEAR_GENERAL:
307 return evergreen_surface_check_linear(p, surf, prefix);
308 case ARRAY_LINEAR_ALIGNED:
309 return evergreen_surface_check_linear_aligned(p, surf, prefix);
310 case ARRAY_1D_TILED_THIN1:
311 return evergreen_surface_check_1d(p, surf, prefix);
312 case ARRAY_2D_TILED_THIN1:
313 return evergreen_surface_check_2d(p, surf, prefix);
315 dev_warn(p->dev, "%s:%d %s invalid array mode %d\n",
316 __func__, __LINE__, prefix, surf->mode);
322 static int evergreen_surface_value_conv_check(struct radeon_cs_parser *p,
323 struct eg_surface *surf,
326 switch (surf->mode) {
327 case ARRAY_2D_TILED_THIN1:
329 case ARRAY_LINEAR_GENERAL:
330 case ARRAY_LINEAR_ALIGNED:
331 case ARRAY_1D_TILED_THIN1:
334 dev_warn(p->dev, "%s:%d %s invalid array mode %d\n",
335 __func__, __LINE__, prefix, surf->mode);
339 switch (surf->nbanks) {
340 case 0: surf->nbanks = 2; break;
341 case 1: surf->nbanks = 4; break;
342 case 2: surf->nbanks = 8; break;
343 case 3: surf->nbanks = 16; break;
345 dev_warn(p->dev, "%s:%d %s invalid number of banks %d\n",
346 __func__, __LINE__, prefix, surf->nbanks);
349 switch (surf->bankw) {
350 case 0: surf->bankw = 1; break;
351 case 1: surf->bankw = 2; break;
352 case 2: surf->bankw = 4; break;
353 case 3: surf->bankw = 8; break;
355 dev_warn(p->dev, "%s:%d %s invalid bankw %d\n",
356 __func__, __LINE__, prefix, surf->bankw);
359 switch (surf->bankh) {
360 case 0: surf->bankh = 1; break;
361 case 1: surf->bankh = 2; break;
362 case 2: surf->bankh = 4; break;
363 case 3: surf->bankh = 8; break;
365 dev_warn(p->dev, "%s:%d %s invalid bankh %d\n",
366 __func__, __LINE__, prefix, surf->bankh);
369 switch (surf->mtilea) {
370 case 0: surf->mtilea = 1; break;
371 case 1: surf->mtilea = 2; break;
372 case 2: surf->mtilea = 4; break;
373 case 3: surf->mtilea = 8; break;
375 dev_warn(p->dev, "%s:%d %s invalid macro tile aspect %d\n",
376 __func__, __LINE__, prefix, surf->mtilea);
379 switch (surf->tsplit) {
380 case 0: surf->tsplit = 64; break;
381 case 1: surf->tsplit = 128; break;
382 case 2: surf->tsplit = 256; break;
383 case 3: surf->tsplit = 512; break;
384 case 4: surf->tsplit = 1024; break;
385 case 5: surf->tsplit = 2048; break;
386 case 6: surf->tsplit = 4096; break;
388 dev_warn(p->dev, "%s:%d %s invalid tile split %d\n",
389 __func__, __LINE__, prefix, surf->tsplit);
395 static int evergreen_cs_track_validate_cb(struct radeon_cs_parser *p, unsigned id)
397 struct evergreen_cs_track *track = p->track;
398 struct eg_surface surf;
399 unsigned pitch, slice, mslice;
400 unsigned long offset;
403 mslice = G_028C6C_SLICE_MAX(track->cb_color_view[id]) + 1;
404 pitch = track->cb_color_pitch[id];
405 slice = track->cb_color_slice[id];
406 surf.nbx = (pitch + 1) * 8;
407 surf.nby = ((slice + 1) * 64) / surf.nbx;
408 surf.mode = G_028C70_ARRAY_MODE(track->cb_color_info[id]);
409 surf.format = G_028C70_FORMAT(track->cb_color_info[id]);
410 surf.tsplit = G_028C74_TILE_SPLIT(track->cb_color_attrib[id]);
411 surf.nbanks = G_028C74_NUM_BANKS(track->cb_color_attrib[id]);
412 surf.bankw = G_028C74_BANK_WIDTH(track->cb_color_attrib[id]);
413 surf.bankh = G_028C74_BANK_HEIGHT(track->cb_color_attrib[id]);
414 surf.mtilea = G_028C74_MACRO_TILE_ASPECT(track->cb_color_attrib[id]);
417 if (!r600_fmt_is_valid_color(surf.format)) {
418 dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08x)\n",
419 __func__, __LINE__, surf.format,
420 id, track->cb_color_info[id]);
424 r = evergreen_surface_value_conv_check(p, &surf, "cb");
429 r = evergreen_surface_check(p, &surf, "cb");
431 dev_warn(p->dev, "%s:%d cb[%d] invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
432 __func__, __LINE__, id, track->cb_color_pitch[id],
433 track->cb_color_slice[id], track->cb_color_attrib[id],
434 track->cb_color_info[id]);
438 offset = track->cb_color_bo_offset[id] << 8;
439 if (offset & (surf.base_align - 1)) {
440 dev_warn(p->dev, "%s:%d cb[%d] bo base %ld not aligned with %ld\n",
441 __func__, __LINE__, id, offset, surf.base_align);
445 offset += surf.layer_size * mslice;
446 if (offset > radeon_bo_size(track->cb_color_bo[id])) {
447 /* old ddx are broken they allocate bo with w*h*bpp but
448 * program slice with ALIGN(h, 8), catch this and patch
452 volatile u32 *ib = p->ib.ptr;
453 unsigned long tmp, nby, bsize, size, min = 0;
455 /* find the height the ddx wants */
459 bsize = radeon_bo_size(track->cb_color_bo[id]);
460 tmp = track->cb_color_bo_offset[id] << 8;
461 for (nby = surf.nby; nby > min; nby--) {
462 size = nby * surf.nbx * surf.bpe * surf.nsamples;
463 if ((tmp + size * mslice) <= bsize) {
469 slice = ((nby * surf.nbx) / 64) - 1;
470 if (!evergreen_surface_check(p, &surf, "cb")) {
471 /* check if this one works */
472 tmp += surf.layer_size * mslice;
474 ib[track->cb_color_slice_idx[id]] = slice;
480 dev_warn(p->dev, "%s:%d cb[%d] bo too small (layer size %d, "
481 "offset %d, max layer %d, bo size %ld, slice %d)\n",
482 __func__, __LINE__, id, surf.layer_size,
483 track->cb_color_bo_offset[id] << 8, mslice,
484 radeon_bo_size(track->cb_color_bo[id]), slice);
485 dev_warn(p->dev, "%s:%d problematic surf: (%d %d) (%d %d %d %d %d %d %d)\n",
486 __func__, __LINE__, surf.nbx, surf.nby,
487 surf.mode, surf.bpe, surf.nsamples,
488 surf.bankw, surf.bankh,
489 surf.tsplit, surf.mtilea);
497 static int evergreen_cs_track_validate_htile(struct radeon_cs_parser *p,
498 unsigned nbx, unsigned nby)
500 struct evergreen_cs_track *track = p->track;
503 if (track->htile_bo == NULL) {
504 dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
505 __func__, __LINE__, track->db_z_info);
509 if (G_028ABC_LINEAR(track->htile_surface)) {
510 /* pitch must be 16 htiles aligned == 16 * 8 pixel aligned */
511 nbx = roundup(nbx, 16 * 8);
512 /* height is npipes htiles aligned == npipes * 8 pixel aligned */
513 nby = roundup(nby, track->npipes * 8);
515 /* always assume 8x8 htile */
516 /* align is htile align * 8, htile align vary according to
517 * number of pipe and tile width and nby
519 switch (track->npipes) {
521 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
522 nbx = roundup(nbx, 64 * 8);
523 nby = roundup(nby, 64 * 8);
526 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
527 nbx = roundup(nbx, 64 * 8);
528 nby = roundup(nby, 32 * 8);
531 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
532 nbx = roundup(nbx, 32 * 8);
533 nby = roundup(nby, 32 * 8);
536 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
537 nbx = roundup(nbx, 32 * 8);
538 nby = roundup(nby, 16 * 8);
541 dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
542 __func__, __LINE__, track->npipes);
546 /* compute number of htile */
549 /* size must be aligned on npipes * 2K boundary */
550 size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
551 size += track->htile_offset;
553 if (size > radeon_bo_size(track->htile_bo)) {
554 dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
555 __func__, __LINE__, radeon_bo_size(track->htile_bo),
562 static int evergreen_cs_track_validate_stencil(struct radeon_cs_parser *p)
564 struct evergreen_cs_track *track = p->track;
565 struct eg_surface surf;
566 unsigned pitch, slice, mslice;
567 unsigned long offset;
570 mslice = G_028008_SLICE_MAX(track->db_depth_view) + 1;
571 pitch = G_028058_PITCH_TILE_MAX(track->db_depth_size);
572 slice = track->db_depth_slice;
573 surf.nbx = (pitch + 1) * 8;
574 surf.nby = ((slice + 1) * 64) / surf.nbx;
575 surf.mode = G_028040_ARRAY_MODE(track->db_z_info);
576 surf.format = G_028044_FORMAT(track->db_s_info);
577 surf.tsplit = G_028044_TILE_SPLIT(track->db_s_info);
578 surf.nbanks = G_028040_NUM_BANKS(track->db_z_info);
579 surf.bankw = G_028040_BANK_WIDTH(track->db_z_info);
580 surf.bankh = G_028040_BANK_HEIGHT(track->db_z_info);
581 surf.mtilea = G_028040_MACRO_TILE_ASPECT(track->db_z_info);
584 if (surf.format != 1) {
585 dev_warn(p->dev, "%s:%d stencil invalid format %d\n",
586 __func__, __LINE__, surf.format);
589 /* replace by color format so we can use same code */
590 surf.format = V_028C70_COLOR_8;
592 r = evergreen_surface_value_conv_check(p, &surf, "stencil");
597 r = evergreen_surface_check(p, &surf, NULL);
599 /* old userspace doesn't compute proper depth/stencil alignment
600 * check that alignment against a bigger byte per elements and
601 * only report if that alignment is wrong too.
603 surf.format = V_028C70_COLOR_8_8_8_8;
604 r = evergreen_surface_check(p, &surf, "stencil");
606 dev_warn(p->dev, "%s:%d stencil invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
607 __func__, __LINE__, track->db_depth_size,
608 track->db_depth_slice, track->db_s_info, track->db_z_info);
613 offset = track->db_s_read_offset << 8;
614 if (offset & (surf.base_align - 1)) {
615 dev_warn(p->dev, "%s:%d stencil read bo base %ld not aligned with %ld\n",
616 __func__, __LINE__, offset, surf.base_align);
619 offset += surf.layer_size * mslice;
620 if (offset > radeon_bo_size(track->db_s_read_bo)) {
621 dev_warn(p->dev, "%s:%d stencil read bo too small (layer size %d, "
622 "offset %ld, max layer %d, bo size %ld)\n",
623 __func__, __LINE__, surf.layer_size,
624 (unsigned long)track->db_s_read_offset << 8, mslice,
625 radeon_bo_size(track->db_s_read_bo));
626 dev_warn(p->dev, "%s:%d stencil invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
627 __func__, __LINE__, track->db_depth_size,
628 track->db_depth_slice, track->db_s_info, track->db_z_info);
632 offset = track->db_s_write_offset << 8;
633 if (offset & (surf.base_align - 1)) {
634 dev_warn(p->dev, "%s:%d stencil write bo base %ld not aligned with %ld\n",
635 __func__, __LINE__, offset, surf.base_align);
638 offset += surf.layer_size * mslice;
639 if (offset > radeon_bo_size(track->db_s_write_bo)) {
640 dev_warn(p->dev, "%s:%d stencil write bo too small (layer size %d, "
641 "offset %ld, max layer %d, bo size %ld)\n",
642 __func__, __LINE__, surf.layer_size,
643 (unsigned long)track->db_s_write_offset << 8, mslice,
644 radeon_bo_size(track->db_s_write_bo));
649 if (G_028040_TILE_SURFACE_ENABLE(track->db_z_info)) {
650 r = evergreen_cs_track_validate_htile(p, surf.nbx, surf.nby);
659 static int evergreen_cs_track_validate_depth(struct radeon_cs_parser *p)
661 struct evergreen_cs_track *track = p->track;
662 struct eg_surface surf;
663 unsigned pitch, slice, mslice;
664 unsigned long offset;
667 mslice = G_028008_SLICE_MAX(track->db_depth_view) + 1;
668 pitch = G_028058_PITCH_TILE_MAX(track->db_depth_size);
669 slice = track->db_depth_slice;
670 surf.nbx = (pitch + 1) * 8;
671 surf.nby = ((slice + 1) * 64) / surf.nbx;
672 surf.mode = G_028040_ARRAY_MODE(track->db_z_info);
673 surf.format = G_028040_FORMAT(track->db_z_info);
674 surf.tsplit = G_028040_TILE_SPLIT(track->db_z_info);
675 surf.nbanks = G_028040_NUM_BANKS(track->db_z_info);
676 surf.bankw = G_028040_BANK_WIDTH(track->db_z_info);
677 surf.bankh = G_028040_BANK_HEIGHT(track->db_z_info);
678 surf.mtilea = G_028040_MACRO_TILE_ASPECT(track->db_z_info);
681 switch (surf.format) {
683 surf.format = V_028C70_COLOR_16;
686 case V_028040_Z_32_FLOAT:
687 surf.format = V_028C70_COLOR_8_8_8_8;
690 dev_warn(p->dev, "%s:%d depth invalid format %d\n",
691 __func__, __LINE__, surf.format);
695 r = evergreen_surface_value_conv_check(p, &surf, "depth");
697 dev_warn(p->dev, "%s:%d depth invalid (0x%08x 0x%08x 0x%08x)\n",
698 __func__, __LINE__, track->db_depth_size,
699 track->db_depth_slice, track->db_z_info);
703 r = evergreen_surface_check(p, &surf, "depth");
705 dev_warn(p->dev, "%s:%d depth invalid (0x%08x 0x%08x 0x%08x)\n",
706 __func__, __LINE__, track->db_depth_size,
707 track->db_depth_slice, track->db_z_info);
711 offset = track->db_z_read_offset << 8;
712 if (offset & (surf.base_align - 1)) {
713 dev_warn(p->dev, "%s:%d stencil read bo base %ld not aligned with %ld\n",
714 __func__, __LINE__, offset, surf.base_align);
717 offset += surf.layer_size * mslice;
718 if (offset > radeon_bo_size(track->db_z_read_bo)) {
719 dev_warn(p->dev, "%s:%d depth read bo too small (layer size %d, "
720 "offset %ld, max layer %d, bo size %ld)\n",
721 __func__, __LINE__, surf.layer_size,
722 (unsigned long)track->db_z_read_offset << 8, mslice,
723 radeon_bo_size(track->db_z_read_bo));
727 offset = track->db_z_write_offset << 8;
728 if (offset & (surf.base_align - 1)) {
729 dev_warn(p->dev, "%s:%d stencil write bo base %ld not aligned with %ld\n",
730 __func__, __LINE__, offset, surf.base_align);
733 offset += surf.layer_size * mslice;
734 if (offset > radeon_bo_size(track->db_z_write_bo)) {
735 dev_warn(p->dev, "%s:%d depth write bo too small (layer size %d, "
736 "offset %ld, max layer %d, bo size %ld)\n",
737 __func__, __LINE__, surf.layer_size,
738 (unsigned long)track->db_z_write_offset << 8, mslice,
739 radeon_bo_size(track->db_z_write_bo));
744 if (G_028040_TILE_SURFACE_ENABLE(track->db_z_info)) {
745 r = evergreen_cs_track_validate_htile(p, surf.nbx, surf.nby);
754 static int evergreen_cs_track_validate_texture(struct radeon_cs_parser *p,
755 struct radeon_bo *texture,
756 struct radeon_bo *mipmap,
759 struct eg_surface surf;
760 unsigned long toffset, moffset;
761 unsigned dim, llevel, mslice, width, height, depth, i;
765 texdw[0] = radeon_get_ib_value(p, idx + 0);
766 texdw[1] = radeon_get_ib_value(p, idx + 1);
767 texdw[2] = radeon_get_ib_value(p, idx + 2);
768 texdw[3] = radeon_get_ib_value(p, idx + 3);
769 texdw[4] = radeon_get_ib_value(p, idx + 4);
770 texdw[5] = radeon_get_ib_value(p, idx + 5);
771 texdw[6] = radeon_get_ib_value(p, idx + 6);
772 texdw[7] = radeon_get_ib_value(p, idx + 7);
773 dim = G_030000_DIM(texdw[0]);
774 llevel = G_030014_LAST_LEVEL(texdw[5]);
775 mslice = G_030014_LAST_ARRAY(texdw[5]) + 1;
776 width = G_030000_TEX_WIDTH(texdw[0]) + 1;
777 height = G_030004_TEX_HEIGHT(texdw[1]) + 1;
778 depth = G_030004_TEX_DEPTH(texdw[1]) + 1;
779 surf.format = G_03001C_DATA_FORMAT(texdw[7]);
780 surf.nbx = (G_030000_PITCH(texdw[0]) + 1) * 8;
781 surf.nbx = r600_fmt_get_nblocksx(surf.format, surf.nbx);
782 surf.nby = r600_fmt_get_nblocksy(surf.format, height);
783 surf.mode = G_030004_ARRAY_MODE(texdw[1]);
784 surf.tsplit = G_030018_TILE_SPLIT(texdw[6]);
785 surf.nbanks = G_03001C_NUM_BANKS(texdw[7]);
786 surf.bankw = G_03001C_BANK_WIDTH(texdw[7]);
787 surf.bankh = G_03001C_BANK_HEIGHT(texdw[7]);
788 surf.mtilea = G_03001C_MACRO_TILE_ASPECT(texdw[7]);
790 toffset = texdw[2] << 8;
791 moffset = texdw[3] << 8;
793 if (!r600_fmt_is_valid_texture(surf.format, p->family)) {
794 dev_warn(p->dev, "%s:%d texture invalid format %d\n",
795 __func__, __LINE__, surf.format);
799 case V_030000_SQ_TEX_DIM_1D:
800 case V_030000_SQ_TEX_DIM_2D:
801 case V_030000_SQ_TEX_DIM_CUBEMAP:
802 case V_030000_SQ_TEX_DIM_1D_ARRAY:
803 case V_030000_SQ_TEX_DIM_2D_ARRAY:
806 case V_030000_SQ_TEX_DIM_2D_MSAA:
807 case V_030000_SQ_TEX_DIM_2D_ARRAY_MSAA:
808 surf.nsamples = 1 << llevel;
812 case V_030000_SQ_TEX_DIM_3D:
815 dev_warn(p->dev, "%s:%d texture invalid dimension %d\n",
816 __func__, __LINE__, dim);
820 r = evergreen_surface_value_conv_check(p, &surf, "texture");
826 evergreen_surface_check(p, &surf, NULL);
827 surf.nby = roundup(surf.nby, surf.halign);
829 r = evergreen_surface_check(p, &surf, "texture");
831 dev_warn(p->dev, "%s:%d texture invalid 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x\n",
832 __func__, __LINE__, texdw[0], texdw[1], texdw[4],
833 texdw[5], texdw[6], texdw[7]);
837 /* check texture size */
838 if (toffset & (surf.base_align - 1)) {
839 dev_warn(p->dev, "%s:%d texture bo base %ld not aligned with %ld\n",
840 __func__, __LINE__, toffset, surf.base_align);
843 if (moffset & (surf.base_align - 1)) {
844 dev_warn(p->dev, "%s:%d mipmap bo base %ld not aligned with %ld\n",
845 __func__, __LINE__, moffset, surf.base_align);
848 if (dim == SQ_TEX_DIM_3D) {
849 toffset += surf.layer_size * depth;
851 toffset += surf.layer_size * mslice;
853 if (toffset > radeon_bo_size(texture)) {
854 dev_warn(p->dev, "%s:%d texture bo too small (layer size %d, "
855 "offset %ld, max layer %d, depth %d, bo size %ld) (%d %d)\n",
856 __func__, __LINE__, surf.layer_size,
857 (unsigned long)texdw[2] << 8, mslice,
858 depth, radeon_bo_size(texture),
865 dev_warn(p->dev, "%s:%i got NULL MIP_ADDRESS relocation\n",
869 return 0; /* everything's ok */
873 /* check mipmap size */
874 for (i = 1; i <= llevel; i++) {
877 w = r600_mip_minify(width, i);
878 h = r600_mip_minify(height, i);
879 d = r600_mip_minify(depth, i);
880 surf.nbx = r600_fmt_get_nblocksx(surf.format, w);
881 surf.nby = r600_fmt_get_nblocksy(surf.format, h);
884 case ARRAY_2D_TILED_THIN1:
885 if (surf.nbx < surf.palign || surf.nby < surf.halign) {
886 surf.mode = ARRAY_1D_TILED_THIN1;
888 /* recompute alignment */
889 evergreen_surface_check(p, &surf, NULL);
891 case ARRAY_LINEAR_GENERAL:
892 case ARRAY_LINEAR_ALIGNED:
893 case ARRAY_1D_TILED_THIN1:
896 dev_warn(p->dev, "%s:%d invalid array mode %d\n",
897 __func__, __LINE__, surf.mode);
900 surf.nbx = roundup(surf.nbx, surf.palign);
901 surf.nby = roundup(surf.nby, surf.halign);
903 r = evergreen_surface_check(p, &surf, "mipmap");
908 if (dim == SQ_TEX_DIM_3D) {
909 moffset += surf.layer_size * d;
911 moffset += surf.layer_size * mslice;
913 if (moffset > radeon_bo_size(mipmap)) {
914 dev_warn(p->dev, "%s:%d mipmap [%d] bo too small (layer size %d, "
915 "offset %ld, coffset %ld, max layer %d, depth %d, "
916 "bo size %ld) level0 (%d %d %d)\n",
917 __func__, __LINE__, i, surf.layer_size,
918 (unsigned long)texdw[3] << 8, moffset, mslice,
919 d, radeon_bo_size(mipmap),
920 width, height, depth);
921 dev_warn(p->dev, "%s:%d problematic surf: (%d %d) (%d %d %d %d %d %d %d)\n",
922 __func__, __LINE__, surf.nbx, surf.nby,
923 surf.mode, surf.bpe, surf.nsamples,
924 surf.bankw, surf.bankh,
925 surf.tsplit, surf.mtilea);
933 static int evergreen_cs_track_check(struct radeon_cs_parser *p)
935 struct evergreen_cs_track *track = p->track;
938 unsigned buffer_mask = 0;
940 /* check streamout */
941 if (track->streamout_dirty && track->vgt_strmout_config) {
942 for (i = 0; i < 4; i++) {
943 if (track->vgt_strmout_config & (1 << i)) {
944 buffer_mask |= (track->vgt_strmout_buffer_config >> (i * 4)) & 0xf;
948 for (i = 0; i < 4; i++) {
949 if (buffer_mask & (1 << i)) {
950 if (track->vgt_strmout_bo[i]) {
951 u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
952 (u64)track->vgt_strmout_size[i];
953 if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
954 DRM_ERROR("streamout %d bo too small: 0x%jx, 0x%lx\n",
955 i, (uintmax_t)offset,
956 radeon_bo_size(track->vgt_strmout_bo[i]));
960 dev_warn(p->dev, "No buffer for streamout %d\n", i);
965 track->streamout_dirty = false;
968 if (track->sx_misc_kill_all_prims)
971 /* check that we have a cb for each enabled target
973 if (track->cb_dirty) {
974 tmp = track->cb_target_mask;
975 for (i = 0; i < 8; i++) {
976 if ((tmp >> (i * 4)) & 0xF) {
977 /* at least one component is enabled */
978 if (track->cb_color_bo[i] == NULL) {
979 dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
980 __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
984 r = evergreen_cs_track_validate_cb(p, i);
990 track->cb_dirty = false;
993 if (track->db_dirty) {
994 /* Check stencil buffer */
995 if (G_028044_FORMAT(track->db_s_info) != V_028044_STENCIL_INVALID &&
996 G_028800_STENCIL_ENABLE(track->db_depth_control)) {
997 r = evergreen_cs_track_validate_stencil(p);
1001 /* Check depth buffer */
1002 if (G_028040_FORMAT(track->db_z_info) != V_028040_Z_INVALID &&
1003 G_028800_Z_ENABLE(track->db_depth_control)) {
1004 r = evergreen_cs_track_validate_depth(p);
1008 track->db_dirty = false;
1015 * evergreen_cs_packet_parse() - parse cp packet and point ib index to next packet
1016 * @parser: parser structure holding parsing context.
1017 * @pkt: where to store packet informations
1019 * Assume that chunk_ib_index is properly set. Will return -EINVAL
1020 * if packet is bigger than remaining ib size. or if packets is unknown.
1022 static int evergreen_cs_packet_parse(struct radeon_cs_parser *p,
1023 struct radeon_cs_packet *pkt,
1026 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
1029 if (idx >= ib_chunk->length_dw) {
1030 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
1031 idx, ib_chunk->length_dw);
1034 header = radeon_get_ib_value(p, idx);
1036 pkt->type = CP_PACKET_GET_TYPE(header);
1037 pkt->count = CP_PACKET_GET_COUNT(header);
1038 pkt->one_reg_wr = 0;
1039 switch (pkt->type) {
1041 pkt->reg = CP_PACKET0_GET_REG(header);
1044 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
1050 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
1053 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
1054 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
1055 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
1062 * evergreen_cs_packet_next_reloc() - parse next packet which should be reloc packet3
1063 * @parser: parser structure holding parsing context.
1064 * @data: pointer to relocation data
1065 * @offset_start: starting offset
1066 * @offset_mask: offset mask (to align start offset on)
1067 * @reloc: reloc informations
1069 * Check next packet is relocation packet3, do bo validation and compute
1070 * GPU offset using the provided start.
1072 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
1073 struct radeon_cs_reloc **cs_reloc)
1075 struct radeon_cs_chunk *relocs_chunk;
1076 struct radeon_cs_packet p3reloc;
1080 if (p->chunk_relocs_idx == -1) {
1081 DRM_ERROR("No relocation chunk !\n");
1085 relocs_chunk = &p->chunks[p->chunk_relocs_idx];
1086 r = evergreen_cs_packet_parse(p, &p3reloc, p->idx);
1090 p->idx += p3reloc.count + 2;
1091 if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
1092 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
1096 idx = radeon_get_ib_value(p, p3reloc.idx + 1);
1097 if (idx >= relocs_chunk->length_dw) {
1098 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
1099 idx, relocs_chunk->length_dw);
1102 /* FIXME: we assume reloc size is 4 dwords */
1103 *cs_reloc = p->relocs_ptr[(idx / 4)];
1108 * evergreen_cs_packet_next_is_pkt3_nop() - test if the next packet is NOP
1109 * @p: structure holding the parser context.
1111 * Check if the next packet is a relocation packet3.
1113 static bool evergreen_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
1115 struct radeon_cs_packet p3reloc;
1118 r = evergreen_cs_packet_parse(p, &p3reloc, p->idx);
1122 if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
1129 * evergreen_cs_packet_next_vline() - parse userspace VLINE packet
1130 * @parser: parser structure holding parsing context.
1132 * Userspace sends a special sequence for VLINE waits.
1133 * PACKET0 - VLINE_START_END + value
1134 * PACKET3 - WAIT_REG_MEM poll vline status reg
1135 * RELOC (P3) - crtc_id in reloc.
1137 * This function parses this and relocates the VLINE START END
1138 * and WAIT_REG_MEM packets to the correct crtc.
1139 * It also detects a switched off crtc and nulls out the
1140 * wait in that case.
1142 static int evergreen_cs_packet_parse_vline(struct radeon_cs_parser *p)
1144 struct drm_mode_object *obj;
1145 struct drm_crtc *crtc;
1146 struct radeon_crtc *radeon_crtc;
1147 struct radeon_cs_packet p3reloc, wait_reg_mem;
1150 uint32_t header, h_idx, reg, wait_reg_mem_info;
1151 volatile uint32_t *ib;
1155 /* parse the WAIT_REG_MEM */
1156 r = evergreen_cs_packet_parse(p, &wait_reg_mem, p->idx);
1160 /* check its a WAIT_REG_MEM */
1161 if (wait_reg_mem.type != PACKET_TYPE3 ||
1162 wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
1163 DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
1167 wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
1168 /* bit 4 is reg (0) or mem (1) */
1169 if (wait_reg_mem_info & 0x10) {
1170 DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
1173 /* waiting for value to be equal */
1174 if ((wait_reg_mem_info & 0x7) != 0x3) {
1175 DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
1178 if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != EVERGREEN_VLINE_STATUS) {
1179 DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
1183 if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != EVERGREEN_VLINE_STAT) {
1184 DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
1188 /* jump over the NOP */
1189 r = evergreen_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
1194 p->idx += wait_reg_mem.count + 2;
1195 p->idx += p3reloc.count + 2;
1197 header = radeon_get_ib_value(p, h_idx);
1198 crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
1199 reg = CP_PACKET0_GET_REG(header);
1200 obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1202 DRM_ERROR("cannot find crtc %d\n", crtc_id);
1205 crtc = obj_to_crtc(obj);
1206 radeon_crtc = to_radeon_crtc(crtc);
1207 crtc_id = radeon_crtc->crtc_id;
1209 if (!crtc->enabled) {
1210 /* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
1211 ib[h_idx + 2] = PACKET2(0);
1212 ib[h_idx + 3] = PACKET2(0);
1213 ib[h_idx + 4] = PACKET2(0);
1214 ib[h_idx + 5] = PACKET2(0);
1215 ib[h_idx + 6] = PACKET2(0);
1216 ib[h_idx + 7] = PACKET2(0);
1217 ib[h_idx + 8] = PACKET2(0);
1220 case EVERGREEN_VLINE_START_END:
1221 header &= ~R600_CP_PACKET0_REG_MASK;
1222 header |= (EVERGREEN_VLINE_START_END + radeon_crtc->crtc_offset) >> 2;
1224 ib[h_idx + 4] = (EVERGREEN_VLINE_STATUS + radeon_crtc->crtc_offset) >> 2;
1227 DRM_ERROR("unknown crtc reloc\n");
1234 static int evergreen_packet0_check(struct radeon_cs_parser *p,
1235 struct radeon_cs_packet *pkt,
1236 unsigned idx, unsigned reg)
1241 case EVERGREEN_VLINE_START_END:
1242 r = evergreen_cs_packet_parse_vline(p);
1244 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1250 DRM_ERROR("Forbidden register 0x%04X in cs at %d\n",
1257 static int evergreen_cs_parse_packet0(struct radeon_cs_parser *p,
1258 struct radeon_cs_packet *pkt)
1266 for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
1267 r = evergreen_packet0_check(p, pkt, idx, reg);
1276 * evergreen_cs_check_reg() - check if register is authorized or not
1277 * @parser: parser structure holding parsing context
1278 * @reg: register we are testing
1279 * @idx: index into the cs buffer
1281 * This function will test against evergreen_reg_safe_bm and return 0
1282 * if register is safe. If register is not flag as safe this function
1283 * will test it against a list of register needind special handling.
1285 static int evergreen_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1287 struct evergreen_cs_track *track = (struct evergreen_cs_track *)p->track;
1288 struct radeon_cs_reloc *reloc;
1293 if (p->rdev->family >= CHIP_CAYMAN)
1294 last_reg = DRM_ARRAY_SIZE(cayman_reg_safe_bm);
1296 last_reg = DRM_ARRAY_SIZE(evergreen_reg_safe_bm);
1299 if (i >= last_reg) {
1300 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1303 m = 1 << ((reg >> 2) & 31);
1304 if (p->rdev->family >= CHIP_CAYMAN) {
1305 if (!(cayman_reg_safe_bm[i] & m))
1308 if (!(evergreen_reg_safe_bm[i] & m))
1313 /* force following reg to 0 in an attempt to disable out buffer
1314 * which will need us to better understand how it works to perform
1315 * security check on it (Jerome)
1317 case SQ_ESGS_RING_SIZE:
1318 case SQ_GSVS_RING_SIZE:
1319 case SQ_ESTMP_RING_SIZE:
1320 case SQ_GSTMP_RING_SIZE:
1321 case SQ_HSTMP_RING_SIZE:
1322 case SQ_LSTMP_RING_SIZE:
1323 case SQ_PSTMP_RING_SIZE:
1324 case SQ_VSTMP_RING_SIZE:
1325 case SQ_ESGS_RING_ITEMSIZE:
1326 case SQ_ESTMP_RING_ITEMSIZE:
1327 case SQ_GSTMP_RING_ITEMSIZE:
1328 case SQ_GSVS_RING_ITEMSIZE:
1329 case SQ_GS_VERT_ITEMSIZE:
1330 case SQ_GS_VERT_ITEMSIZE_1:
1331 case SQ_GS_VERT_ITEMSIZE_2:
1332 case SQ_GS_VERT_ITEMSIZE_3:
1333 case SQ_GSVS_RING_OFFSET_1:
1334 case SQ_GSVS_RING_OFFSET_2:
1335 case SQ_GSVS_RING_OFFSET_3:
1336 case SQ_HSTMP_RING_ITEMSIZE:
1337 case SQ_LSTMP_RING_ITEMSIZE:
1338 case SQ_PSTMP_RING_ITEMSIZE:
1339 case SQ_VSTMP_RING_ITEMSIZE:
1340 case VGT_TF_RING_SIZE:
1341 /* get value to populate the IB don't remove */
1342 /*tmp =radeon_get_ib_value(p, idx);
1345 case SQ_ESGS_RING_BASE:
1346 case SQ_GSVS_RING_BASE:
1347 case SQ_ESTMP_RING_BASE:
1348 case SQ_GSTMP_RING_BASE:
1349 case SQ_HSTMP_RING_BASE:
1350 case SQ_LSTMP_RING_BASE:
1351 case SQ_PSTMP_RING_BASE:
1352 case SQ_VSTMP_RING_BASE:
1353 r = evergreen_cs_packet_next_reloc(p, &reloc);
1355 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1359 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1361 case DB_DEPTH_CONTROL:
1362 track->db_depth_control = radeon_get_ib_value(p, idx);
1363 track->db_dirty = true;
1365 case CAYMAN_DB_EQAA:
1366 if (p->rdev->family < CHIP_CAYMAN) {
1367 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1372 case CAYMAN_DB_DEPTH_INFO:
1373 if (p->rdev->family < CHIP_CAYMAN) {
1374 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1380 track->db_z_info = radeon_get_ib_value(p, idx);
1381 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1382 r = evergreen_cs_packet_next_reloc(p, &reloc);
1384 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1388 ib[idx] &= ~Z_ARRAY_MODE(0xf);
1389 track->db_z_info &= ~Z_ARRAY_MODE(0xf);
1390 ib[idx] |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1391 track->db_z_info |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1392 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1393 unsigned bankw, bankh, mtaspect, tile_split;
1395 evergreen_tiling_fields(reloc->lobj.tiling_flags,
1396 &bankw, &bankh, &mtaspect,
1398 ib[idx] |= DB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1399 ib[idx] |= DB_TILE_SPLIT(tile_split) |
1400 DB_BANK_WIDTH(bankw) |
1401 DB_BANK_HEIGHT(bankh) |
1402 DB_MACRO_TILE_ASPECT(mtaspect);
1405 track->db_dirty = true;
1407 case DB_STENCIL_INFO:
1408 track->db_s_info = radeon_get_ib_value(p, idx);
1409 track->db_dirty = true;
1412 track->db_depth_view = radeon_get_ib_value(p, idx);
1413 track->db_dirty = true;
1416 track->db_depth_size = radeon_get_ib_value(p, idx);
1417 track->db_dirty = true;
1419 case R_02805C_DB_DEPTH_SLICE:
1420 track->db_depth_slice = radeon_get_ib_value(p, idx);
1421 track->db_dirty = true;
1423 case DB_Z_READ_BASE:
1424 r = evergreen_cs_packet_next_reloc(p, &reloc);
1426 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1430 track->db_z_read_offset = radeon_get_ib_value(p, idx);
1431 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1432 track->db_z_read_bo = reloc->robj;
1433 track->db_dirty = true;
1435 case DB_Z_WRITE_BASE:
1436 r = evergreen_cs_packet_next_reloc(p, &reloc);
1438 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1442 track->db_z_write_offset = radeon_get_ib_value(p, idx);
1443 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1444 track->db_z_write_bo = reloc->robj;
1445 track->db_dirty = true;
1447 case DB_STENCIL_READ_BASE:
1448 r = evergreen_cs_packet_next_reloc(p, &reloc);
1450 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1454 track->db_s_read_offset = radeon_get_ib_value(p, idx);
1455 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1456 track->db_s_read_bo = reloc->robj;
1457 track->db_dirty = true;
1459 case DB_STENCIL_WRITE_BASE:
1460 r = evergreen_cs_packet_next_reloc(p, &reloc);
1462 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1466 track->db_s_write_offset = radeon_get_ib_value(p, idx);
1467 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1468 track->db_s_write_bo = reloc->robj;
1469 track->db_dirty = true;
1471 case VGT_STRMOUT_CONFIG:
1472 track->vgt_strmout_config = radeon_get_ib_value(p, idx);
1473 track->streamout_dirty = true;
1475 case VGT_STRMOUT_BUFFER_CONFIG:
1476 track->vgt_strmout_buffer_config = radeon_get_ib_value(p, idx);
1477 track->streamout_dirty = true;
1479 case VGT_STRMOUT_BUFFER_BASE_0:
1480 case VGT_STRMOUT_BUFFER_BASE_1:
1481 case VGT_STRMOUT_BUFFER_BASE_2:
1482 case VGT_STRMOUT_BUFFER_BASE_3:
1483 r = evergreen_cs_packet_next_reloc(p, &reloc);
1485 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1489 tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1490 track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1491 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1492 track->vgt_strmout_bo[tmp] = reloc->robj;
1493 track->streamout_dirty = true;
1495 case VGT_STRMOUT_BUFFER_SIZE_0:
1496 case VGT_STRMOUT_BUFFER_SIZE_1:
1497 case VGT_STRMOUT_BUFFER_SIZE_2:
1498 case VGT_STRMOUT_BUFFER_SIZE_3:
1499 tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1500 /* size in register is DWs, convert to bytes */
1501 track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1502 track->streamout_dirty = true;
1505 r = evergreen_cs_packet_next_reloc(p, &reloc);
1507 dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1511 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1512 case CB_TARGET_MASK:
1513 track->cb_target_mask = radeon_get_ib_value(p, idx);
1514 track->cb_dirty = true;
1516 case CB_SHADER_MASK:
1517 track->cb_shader_mask = radeon_get_ib_value(p, idx);
1518 track->cb_dirty = true;
1520 case PA_SC_AA_CONFIG:
1521 if (p->rdev->family >= CHIP_CAYMAN) {
1522 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1526 tmp = radeon_get_ib_value(p, idx) & MSAA_NUM_SAMPLES_MASK;
1527 track->nsamples = 1 << tmp;
1529 case CAYMAN_PA_SC_AA_CONFIG:
1530 if (p->rdev->family < CHIP_CAYMAN) {
1531 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1535 tmp = radeon_get_ib_value(p, idx) & CAYMAN_MSAA_NUM_SAMPLES_MASK;
1536 track->nsamples = 1 << tmp;
1538 case CB_COLOR0_VIEW:
1539 case CB_COLOR1_VIEW:
1540 case CB_COLOR2_VIEW:
1541 case CB_COLOR3_VIEW:
1542 case CB_COLOR4_VIEW:
1543 case CB_COLOR5_VIEW:
1544 case CB_COLOR6_VIEW:
1545 case CB_COLOR7_VIEW:
1546 tmp = (reg - CB_COLOR0_VIEW) / 0x3c;
1547 track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1548 track->cb_dirty = true;
1550 case CB_COLOR8_VIEW:
1551 case CB_COLOR9_VIEW:
1552 case CB_COLOR10_VIEW:
1553 case CB_COLOR11_VIEW:
1554 tmp = ((reg - CB_COLOR8_VIEW) / 0x1c) + 8;
1555 track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1556 track->cb_dirty = true;
1558 case CB_COLOR0_INFO:
1559 case CB_COLOR1_INFO:
1560 case CB_COLOR2_INFO:
1561 case CB_COLOR3_INFO:
1562 case CB_COLOR4_INFO:
1563 case CB_COLOR5_INFO:
1564 case CB_COLOR6_INFO:
1565 case CB_COLOR7_INFO:
1566 tmp = (reg - CB_COLOR0_INFO) / 0x3c;
1567 track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1568 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1569 r = evergreen_cs_packet_next_reloc(p, &reloc);
1571 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1575 ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1576 track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1578 track->cb_dirty = true;
1580 case CB_COLOR8_INFO:
1581 case CB_COLOR9_INFO:
1582 case CB_COLOR10_INFO:
1583 case CB_COLOR11_INFO:
1584 tmp = ((reg - CB_COLOR8_INFO) / 0x1c) + 8;
1585 track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1586 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1587 r = evergreen_cs_packet_next_reloc(p, &reloc);
1589 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1593 ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1594 track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
1596 track->cb_dirty = true;
1598 case CB_COLOR0_PITCH:
1599 case CB_COLOR1_PITCH:
1600 case CB_COLOR2_PITCH:
1601 case CB_COLOR3_PITCH:
1602 case CB_COLOR4_PITCH:
1603 case CB_COLOR5_PITCH:
1604 case CB_COLOR6_PITCH:
1605 case CB_COLOR7_PITCH:
1606 tmp = (reg - CB_COLOR0_PITCH) / 0x3c;
1607 track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
1608 track->cb_dirty = true;
1610 case CB_COLOR8_PITCH:
1611 case CB_COLOR9_PITCH:
1612 case CB_COLOR10_PITCH:
1613 case CB_COLOR11_PITCH:
1614 tmp = ((reg - CB_COLOR8_PITCH) / 0x1c) + 8;
1615 track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
1616 track->cb_dirty = true;
1618 case CB_COLOR0_SLICE:
1619 case CB_COLOR1_SLICE:
1620 case CB_COLOR2_SLICE:
1621 case CB_COLOR3_SLICE:
1622 case CB_COLOR4_SLICE:
1623 case CB_COLOR5_SLICE:
1624 case CB_COLOR6_SLICE:
1625 case CB_COLOR7_SLICE:
1626 tmp = (reg - CB_COLOR0_SLICE) / 0x3c;
1627 track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
1628 track->cb_color_slice_idx[tmp] = idx;
1629 track->cb_dirty = true;
1631 case CB_COLOR8_SLICE:
1632 case CB_COLOR9_SLICE:
1633 case CB_COLOR10_SLICE:
1634 case CB_COLOR11_SLICE:
1635 tmp = ((reg - CB_COLOR8_SLICE) / 0x1c) + 8;
1636 track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
1637 track->cb_color_slice_idx[tmp] = idx;
1638 track->cb_dirty = true;
1640 case CB_COLOR0_ATTRIB:
1641 case CB_COLOR1_ATTRIB:
1642 case CB_COLOR2_ATTRIB:
1643 case CB_COLOR3_ATTRIB:
1644 case CB_COLOR4_ATTRIB:
1645 case CB_COLOR5_ATTRIB:
1646 case CB_COLOR6_ATTRIB:
1647 case CB_COLOR7_ATTRIB:
1648 r = evergreen_cs_packet_next_reloc(p, &reloc);
1650 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1654 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1655 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1656 unsigned bankw, bankh, mtaspect, tile_split;
1658 evergreen_tiling_fields(reloc->lobj.tiling_flags,
1659 &bankw, &bankh, &mtaspect,
1661 ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1662 ib[idx] |= CB_TILE_SPLIT(tile_split) |
1663 CB_BANK_WIDTH(bankw) |
1664 CB_BANK_HEIGHT(bankh) |
1665 CB_MACRO_TILE_ASPECT(mtaspect);
1668 tmp = ((reg - CB_COLOR0_ATTRIB) / 0x3c);
1669 track->cb_color_attrib[tmp] = ib[idx];
1670 track->cb_dirty = true;
1672 case CB_COLOR8_ATTRIB:
1673 case CB_COLOR9_ATTRIB:
1674 case CB_COLOR10_ATTRIB:
1675 case CB_COLOR11_ATTRIB:
1676 r = evergreen_cs_packet_next_reloc(p, &reloc);
1678 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1682 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1683 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1684 unsigned bankw, bankh, mtaspect, tile_split;
1686 evergreen_tiling_fields(reloc->lobj.tiling_flags,
1687 &bankw, &bankh, &mtaspect,
1689 ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1690 ib[idx] |= CB_TILE_SPLIT(tile_split) |
1691 CB_BANK_WIDTH(bankw) |
1692 CB_BANK_HEIGHT(bankh) |
1693 CB_MACRO_TILE_ASPECT(mtaspect);
1696 tmp = ((reg - CB_COLOR8_ATTRIB) / 0x1c) + 8;
1697 track->cb_color_attrib[tmp] = ib[idx];
1698 track->cb_dirty = true;
1700 case CB_COLOR0_FMASK:
1701 case CB_COLOR1_FMASK:
1702 case CB_COLOR2_FMASK:
1703 case CB_COLOR3_FMASK:
1704 case CB_COLOR4_FMASK:
1705 case CB_COLOR5_FMASK:
1706 case CB_COLOR6_FMASK:
1707 case CB_COLOR7_FMASK:
1708 tmp = (reg - CB_COLOR0_FMASK) / 0x3c;
1709 r = evergreen_cs_packet_next_reloc(p, &reloc);
1711 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1714 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1715 track->cb_color_fmask_bo[tmp] = reloc->robj;
1717 case CB_COLOR0_CMASK:
1718 case CB_COLOR1_CMASK:
1719 case CB_COLOR2_CMASK:
1720 case CB_COLOR3_CMASK:
1721 case CB_COLOR4_CMASK:
1722 case CB_COLOR5_CMASK:
1723 case CB_COLOR6_CMASK:
1724 case CB_COLOR7_CMASK:
1725 tmp = (reg - CB_COLOR0_CMASK) / 0x3c;
1726 r = evergreen_cs_packet_next_reloc(p, &reloc);
1728 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1731 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1732 track->cb_color_cmask_bo[tmp] = reloc->robj;
1734 case CB_COLOR0_FMASK_SLICE:
1735 case CB_COLOR1_FMASK_SLICE:
1736 case CB_COLOR2_FMASK_SLICE:
1737 case CB_COLOR3_FMASK_SLICE:
1738 case CB_COLOR4_FMASK_SLICE:
1739 case CB_COLOR5_FMASK_SLICE:
1740 case CB_COLOR6_FMASK_SLICE:
1741 case CB_COLOR7_FMASK_SLICE:
1742 tmp = (reg - CB_COLOR0_FMASK_SLICE) / 0x3c;
1743 track->cb_color_fmask_slice[tmp] = radeon_get_ib_value(p, idx);
1745 case CB_COLOR0_CMASK_SLICE:
1746 case CB_COLOR1_CMASK_SLICE:
1747 case CB_COLOR2_CMASK_SLICE:
1748 case CB_COLOR3_CMASK_SLICE:
1749 case CB_COLOR4_CMASK_SLICE:
1750 case CB_COLOR5_CMASK_SLICE:
1751 case CB_COLOR6_CMASK_SLICE:
1752 case CB_COLOR7_CMASK_SLICE:
1753 tmp = (reg - CB_COLOR0_CMASK_SLICE) / 0x3c;
1754 track->cb_color_cmask_slice[tmp] = radeon_get_ib_value(p, idx);
1756 case CB_COLOR0_BASE:
1757 case CB_COLOR1_BASE:
1758 case CB_COLOR2_BASE:
1759 case CB_COLOR3_BASE:
1760 case CB_COLOR4_BASE:
1761 case CB_COLOR5_BASE:
1762 case CB_COLOR6_BASE:
1763 case CB_COLOR7_BASE:
1764 r = evergreen_cs_packet_next_reloc(p, &reloc);
1766 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1770 tmp = (reg - CB_COLOR0_BASE) / 0x3c;
1771 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
1772 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1773 track->cb_color_bo[tmp] = reloc->robj;
1774 track->cb_dirty = true;
1776 case CB_COLOR8_BASE:
1777 case CB_COLOR9_BASE:
1778 case CB_COLOR10_BASE:
1779 case CB_COLOR11_BASE:
1780 r = evergreen_cs_packet_next_reloc(p, &reloc);
1782 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1786 tmp = ((reg - CB_COLOR8_BASE) / 0x1c) + 8;
1787 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
1788 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1789 track->cb_color_bo[tmp] = reloc->robj;
1790 track->cb_dirty = true;
1792 case DB_HTILE_DATA_BASE:
1793 r = evergreen_cs_packet_next_reloc(p, &reloc);
1795 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1799 track->htile_offset = radeon_get_ib_value(p, idx);
1800 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1801 track->htile_bo = reloc->robj;
1802 track->db_dirty = true;
1804 case DB_HTILE_SURFACE:
1806 track->htile_surface = radeon_get_ib_value(p, idx);
1807 /* force 8x8 htile width and height */
1809 track->db_dirty = true;
1811 case CB_IMMED0_BASE:
1812 case CB_IMMED1_BASE:
1813 case CB_IMMED2_BASE:
1814 case CB_IMMED3_BASE:
1815 case CB_IMMED4_BASE:
1816 case CB_IMMED5_BASE:
1817 case CB_IMMED6_BASE:
1818 case CB_IMMED7_BASE:
1819 case CB_IMMED8_BASE:
1820 case CB_IMMED9_BASE:
1821 case CB_IMMED10_BASE:
1822 case CB_IMMED11_BASE:
1823 case SQ_PGM_START_FS:
1824 case SQ_PGM_START_ES:
1825 case SQ_PGM_START_VS:
1826 case SQ_PGM_START_GS:
1827 case SQ_PGM_START_PS:
1828 case SQ_PGM_START_HS:
1829 case SQ_PGM_START_LS:
1830 case SQ_CONST_MEM_BASE:
1831 case SQ_ALU_CONST_CACHE_GS_0:
1832 case SQ_ALU_CONST_CACHE_GS_1:
1833 case SQ_ALU_CONST_CACHE_GS_2:
1834 case SQ_ALU_CONST_CACHE_GS_3:
1835 case SQ_ALU_CONST_CACHE_GS_4:
1836 case SQ_ALU_CONST_CACHE_GS_5:
1837 case SQ_ALU_CONST_CACHE_GS_6:
1838 case SQ_ALU_CONST_CACHE_GS_7:
1839 case SQ_ALU_CONST_CACHE_GS_8:
1840 case SQ_ALU_CONST_CACHE_GS_9:
1841 case SQ_ALU_CONST_CACHE_GS_10:
1842 case SQ_ALU_CONST_CACHE_GS_11:
1843 case SQ_ALU_CONST_CACHE_GS_12:
1844 case SQ_ALU_CONST_CACHE_GS_13:
1845 case SQ_ALU_CONST_CACHE_GS_14:
1846 case SQ_ALU_CONST_CACHE_GS_15:
1847 case SQ_ALU_CONST_CACHE_PS_0:
1848 case SQ_ALU_CONST_CACHE_PS_1:
1849 case SQ_ALU_CONST_CACHE_PS_2:
1850 case SQ_ALU_CONST_CACHE_PS_3:
1851 case SQ_ALU_CONST_CACHE_PS_4:
1852 case SQ_ALU_CONST_CACHE_PS_5:
1853 case SQ_ALU_CONST_CACHE_PS_6:
1854 case SQ_ALU_CONST_CACHE_PS_7:
1855 case SQ_ALU_CONST_CACHE_PS_8:
1856 case SQ_ALU_CONST_CACHE_PS_9:
1857 case SQ_ALU_CONST_CACHE_PS_10:
1858 case SQ_ALU_CONST_CACHE_PS_11:
1859 case SQ_ALU_CONST_CACHE_PS_12:
1860 case SQ_ALU_CONST_CACHE_PS_13:
1861 case SQ_ALU_CONST_CACHE_PS_14:
1862 case SQ_ALU_CONST_CACHE_PS_15:
1863 case SQ_ALU_CONST_CACHE_VS_0:
1864 case SQ_ALU_CONST_CACHE_VS_1:
1865 case SQ_ALU_CONST_CACHE_VS_2:
1866 case SQ_ALU_CONST_CACHE_VS_3:
1867 case SQ_ALU_CONST_CACHE_VS_4:
1868 case SQ_ALU_CONST_CACHE_VS_5:
1869 case SQ_ALU_CONST_CACHE_VS_6:
1870 case SQ_ALU_CONST_CACHE_VS_7:
1871 case SQ_ALU_CONST_CACHE_VS_8:
1872 case SQ_ALU_CONST_CACHE_VS_9:
1873 case SQ_ALU_CONST_CACHE_VS_10:
1874 case SQ_ALU_CONST_CACHE_VS_11:
1875 case SQ_ALU_CONST_CACHE_VS_12:
1876 case SQ_ALU_CONST_CACHE_VS_13:
1877 case SQ_ALU_CONST_CACHE_VS_14:
1878 case SQ_ALU_CONST_CACHE_VS_15:
1879 case SQ_ALU_CONST_CACHE_HS_0:
1880 case SQ_ALU_CONST_CACHE_HS_1:
1881 case SQ_ALU_CONST_CACHE_HS_2:
1882 case SQ_ALU_CONST_CACHE_HS_3:
1883 case SQ_ALU_CONST_CACHE_HS_4:
1884 case SQ_ALU_CONST_CACHE_HS_5:
1885 case SQ_ALU_CONST_CACHE_HS_6:
1886 case SQ_ALU_CONST_CACHE_HS_7:
1887 case SQ_ALU_CONST_CACHE_HS_8:
1888 case SQ_ALU_CONST_CACHE_HS_9:
1889 case SQ_ALU_CONST_CACHE_HS_10:
1890 case SQ_ALU_CONST_CACHE_HS_11:
1891 case SQ_ALU_CONST_CACHE_HS_12:
1892 case SQ_ALU_CONST_CACHE_HS_13:
1893 case SQ_ALU_CONST_CACHE_HS_14:
1894 case SQ_ALU_CONST_CACHE_HS_15:
1895 case SQ_ALU_CONST_CACHE_LS_0:
1896 case SQ_ALU_CONST_CACHE_LS_1:
1897 case SQ_ALU_CONST_CACHE_LS_2:
1898 case SQ_ALU_CONST_CACHE_LS_3:
1899 case SQ_ALU_CONST_CACHE_LS_4:
1900 case SQ_ALU_CONST_CACHE_LS_5:
1901 case SQ_ALU_CONST_CACHE_LS_6:
1902 case SQ_ALU_CONST_CACHE_LS_7:
1903 case SQ_ALU_CONST_CACHE_LS_8:
1904 case SQ_ALU_CONST_CACHE_LS_9:
1905 case SQ_ALU_CONST_CACHE_LS_10:
1906 case SQ_ALU_CONST_CACHE_LS_11:
1907 case SQ_ALU_CONST_CACHE_LS_12:
1908 case SQ_ALU_CONST_CACHE_LS_13:
1909 case SQ_ALU_CONST_CACHE_LS_14:
1910 case SQ_ALU_CONST_CACHE_LS_15:
1911 r = evergreen_cs_packet_next_reloc(p, &reloc);
1913 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1917 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1919 case SX_MEMORY_EXPORT_BASE:
1920 if (p->rdev->family >= CHIP_CAYMAN) {
1921 dev_warn(p->dev, "bad SET_CONFIG_REG "
1925 r = evergreen_cs_packet_next_reloc(p, &reloc);
1927 dev_warn(p->dev, "bad SET_CONFIG_REG "
1931 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1933 case CAYMAN_SX_SCATTER_EXPORT_BASE:
1934 if (p->rdev->family < CHIP_CAYMAN) {
1935 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1939 r = evergreen_cs_packet_next_reloc(p, &reloc);
1941 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1945 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1948 track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1951 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1957 static bool evergreen_is_safe_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1961 if (p->rdev->family >= CHIP_CAYMAN)
1962 last_reg = DRM_ARRAY_SIZE(cayman_reg_safe_bm);
1964 last_reg = DRM_ARRAY_SIZE(evergreen_reg_safe_bm);
1967 if (i >= last_reg) {
1968 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1971 m = 1 << ((reg >> 2) & 31);
1972 if (p->rdev->family >= CHIP_CAYMAN) {
1973 if (!(cayman_reg_safe_bm[i] & m))
1976 if (!(evergreen_reg_safe_bm[i] & m))
1979 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1983 static int evergreen_packet3_check(struct radeon_cs_parser *p,
1984 struct radeon_cs_packet *pkt)
1986 struct radeon_cs_reloc *reloc;
1987 struct evergreen_cs_track *track;
1991 unsigned start_reg, end_reg, reg;
1995 track = (struct evergreen_cs_track *)p->track;
1998 idx_value = radeon_get_ib_value(p, idx);
2000 switch (pkt->opcode) {
2001 case PACKET3_SET_PREDICATION:
2007 if (pkt->count != 1) {
2008 DRM_ERROR("bad SET PREDICATION\n");
2012 tmp = radeon_get_ib_value(p, idx + 1);
2013 pred_op = (tmp >> 16) & 0x7;
2015 /* for the clear predicate operation */
2020 DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
2024 r = evergreen_cs_packet_next_reloc(p, &reloc);
2026 DRM_ERROR("bad SET PREDICATION\n");
2030 offset = reloc->lobj.gpu_offset +
2031 (idx_value & 0xfffffff0) +
2032 ((u64)(tmp & 0xff) << 32);
2034 ib[idx + 0] = offset;
2035 ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2038 case PACKET3_CONTEXT_CONTROL:
2039 if (pkt->count != 1) {
2040 DRM_ERROR("bad CONTEXT_CONTROL\n");
2044 case PACKET3_INDEX_TYPE:
2045 case PACKET3_NUM_INSTANCES:
2046 case PACKET3_CLEAR_STATE:
2048 DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
2052 case CAYMAN_PACKET3_DEALLOC_STATE:
2053 if (p->rdev->family < CHIP_CAYMAN) {
2054 DRM_ERROR("bad PACKET3_DEALLOC_STATE\n");
2058 DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
2062 case PACKET3_INDEX_BASE:
2066 if (pkt->count != 1) {
2067 DRM_ERROR("bad INDEX_BASE\n");
2070 r = evergreen_cs_packet_next_reloc(p, &reloc);
2072 DRM_ERROR("bad INDEX_BASE\n");
2076 offset = reloc->lobj.gpu_offset +
2078 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
2081 ib[idx+1] = upper_32_bits(offset) & 0xff;
2083 r = evergreen_cs_track_check(p);
2085 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2090 case PACKET3_DRAW_INDEX:
2093 if (pkt->count != 3) {
2094 DRM_ERROR("bad DRAW_INDEX\n");
2097 r = evergreen_cs_packet_next_reloc(p, &reloc);
2099 DRM_ERROR("bad DRAW_INDEX\n");
2103 offset = reloc->lobj.gpu_offset +
2105 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
2108 ib[idx+1] = upper_32_bits(offset) & 0xff;
2110 r = evergreen_cs_track_check(p);
2112 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2117 case PACKET3_DRAW_INDEX_2:
2121 if (pkt->count != 4) {
2122 DRM_ERROR("bad DRAW_INDEX_2\n");
2125 r = evergreen_cs_packet_next_reloc(p, &reloc);
2127 DRM_ERROR("bad DRAW_INDEX_2\n");
2131 offset = reloc->lobj.gpu_offset +
2132 radeon_get_ib_value(p, idx+1) +
2133 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2136 ib[idx+2] = upper_32_bits(offset) & 0xff;
2138 r = evergreen_cs_track_check(p);
2140 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2145 case PACKET3_DRAW_INDEX_AUTO:
2146 if (pkt->count != 1) {
2147 DRM_ERROR("bad DRAW_INDEX_AUTO\n");
2150 r = evergreen_cs_track_check(p);
2152 dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
2156 case PACKET3_DRAW_INDEX_MULTI_AUTO:
2157 if (pkt->count != 2) {
2158 DRM_ERROR("bad DRAW_INDEX_MULTI_AUTO\n");
2161 r = evergreen_cs_track_check(p);
2163 dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
2167 case PACKET3_DRAW_INDEX_IMMD:
2168 if (pkt->count < 2) {
2169 DRM_ERROR("bad DRAW_INDEX_IMMD\n");
2172 r = evergreen_cs_track_check(p);
2174 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2178 case PACKET3_DRAW_INDEX_OFFSET:
2179 if (pkt->count != 2) {
2180 DRM_ERROR("bad DRAW_INDEX_OFFSET\n");
2183 r = evergreen_cs_track_check(p);
2185 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2189 case PACKET3_DRAW_INDEX_OFFSET_2:
2190 if (pkt->count != 3) {
2191 DRM_ERROR("bad DRAW_INDEX_OFFSET_2\n");
2194 r = evergreen_cs_track_check(p);
2196 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2200 case PACKET3_DISPATCH_DIRECT:
2201 if (pkt->count != 3) {
2202 DRM_ERROR("bad DISPATCH_DIRECT\n");
2205 r = evergreen_cs_track_check(p);
2207 dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
2211 case PACKET3_DISPATCH_INDIRECT:
2212 if (pkt->count != 1) {
2213 DRM_ERROR("bad DISPATCH_INDIRECT\n");
2216 r = evergreen_cs_packet_next_reloc(p, &reloc);
2218 DRM_ERROR("bad DISPATCH_INDIRECT\n");
2221 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
2222 r = evergreen_cs_track_check(p);
2224 dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2228 case PACKET3_WAIT_REG_MEM:
2229 if (pkt->count != 5) {
2230 DRM_ERROR("bad WAIT_REG_MEM\n");
2233 /* bit 4 is reg (0) or mem (1) */
2234 if (idx_value & 0x10) {
2237 r = evergreen_cs_packet_next_reloc(p, &reloc);
2239 DRM_ERROR("bad WAIT_REG_MEM\n");
2243 offset = reloc->lobj.gpu_offset +
2244 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2245 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2247 ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffffc);
2248 ib[idx+2] = upper_32_bits(offset) & 0xff;
2251 case PACKET3_CP_DMA:
2253 u32 command, size, info;
2255 if (pkt->count != 4) {
2256 DRM_ERROR("bad CP DMA\n");
2259 command = radeon_get_ib_value(p, idx+4);
2260 size = command & 0x1fffff;
2261 info = radeon_get_ib_value(p, idx+1);
2262 if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
2263 (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
2264 ((((info & 0x00300000) >> 20) == 0) &&
2265 (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
2266 ((((info & 0x60000000) >> 29) == 0) &&
2267 (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
2268 /* non mem to mem copies requires dw aligned count */
2270 DRM_ERROR("CP DMA command requires dw count alignment\n");
2274 if (command & PACKET3_CP_DMA_CMD_SAS) {
2275 /* src address space is register */
2277 if (((info & 0x60000000) >> 29) != 1) {
2278 DRM_ERROR("CP DMA SAS not supported\n");
2282 if (command & PACKET3_CP_DMA_CMD_SAIC) {
2283 DRM_ERROR("CP DMA SAIC only supported for registers\n");
2286 /* src address space is memory */
2287 if (((info & 0x60000000) >> 29) == 0) {
2288 r = evergreen_cs_packet_next_reloc(p, &reloc);
2290 DRM_ERROR("bad CP DMA SRC\n");
2294 tmp = radeon_get_ib_value(p, idx) +
2295 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
2297 offset = reloc->lobj.gpu_offset + tmp;
2299 if ((tmp + size) > radeon_bo_size(reloc->robj)) {
2300 dev_warn(p->dev, "CP DMA src buffer too small (%ju %lu)\n",
2301 (uintmax_t)tmp + size, radeon_bo_size(reloc->robj));
2306 ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2307 } else if (((info & 0x60000000) >> 29) != 2) {
2308 DRM_ERROR("bad CP DMA SRC_SEL\n");
2312 if (command & PACKET3_CP_DMA_CMD_DAS) {
2313 /* dst address space is register */
2315 if (((info & 0x00300000) >> 20) != 1) {
2316 DRM_ERROR("CP DMA DAS not supported\n");
2320 /* dst address space is memory */
2321 if (command & PACKET3_CP_DMA_CMD_DAIC) {
2322 DRM_ERROR("CP DMA DAIC only supported for registers\n");
2325 if (((info & 0x00300000) >> 20) == 0) {
2326 r = evergreen_cs_packet_next_reloc(p, &reloc);
2328 DRM_ERROR("bad CP DMA DST\n");
2332 tmp = radeon_get_ib_value(p, idx+2) +
2333 ((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
2335 offset = reloc->lobj.gpu_offset + tmp;
2337 if ((tmp + size) > radeon_bo_size(reloc->robj)) {
2338 dev_warn(p->dev, "CP DMA dst buffer too small (%ju %lu)\n",
2339 (uintmax_t)tmp + size, radeon_bo_size(reloc->robj));
2344 ib[idx+3] = upper_32_bits(offset) & 0xff;
2346 DRM_ERROR("bad CP DMA DST_SEL\n");
2352 case PACKET3_SURFACE_SYNC:
2353 if (pkt->count != 3) {
2354 DRM_ERROR("bad SURFACE_SYNC\n");
2357 /* 0xffffffff/0x0 is flush all cache flag */
2358 if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
2359 radeon_get_ib_value(p, idx + 2) != 0) {
2360 r = evergreen_cs_packet_next_reloc(p, &reloc);
2362 DRM_ERROR("bad SURFACE_SYNC\n");
2365 ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
2368 case PACKET3_EVENT_WRITE:
2369 if (pkt->count != 2 && pkt->count != 0) {
2370 DRM_ERROR("bad EVENT_WRITE\n");
2376 r = evergreen_cs_packet_next_reloc(p, &reloc);
2378 DRM_ERROR("bad EVENT_WRITE\n");
2381 offset = reloc->lobj.gpu_offset +
2382 (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
2383 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2385 ib[idx+1] = offset & 0xfffffff8;
2386 ib[idx+2] = upper_32_bits(offset) & 0xff;
2389 case PACKET3_EVENT_WRITE_EOP:
2393 if (pkt->count != 4) {
2394 DRM_ERROR("bad EVENT_WRITE_EOP\n");
2397 r = evergreen_cs_packet_next_reloc(p, &reloc);
2399 DRM_ERROR("bad EVENT_WRITE_EOP\n");
2403 offset = reloc->lobj.gpu_offset +
2404 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2405 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2407 ib[idx+1] = offset & 0xfffffffc;
2408 ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2411 case PACKET3_EVENT_WRITE_EOS:
2415 if (pkt->count != 3) {
2416 DRM_ERROR("bad EVENT_WRITE_EOS\n");
2419 r = evergreen_cs_packet_next_reloc(p, &reloc);
2421 DRM_ERROR("bad EVENT_WRITE_EOS\n");
2425 offset = reloc->lobj.gpu_offset +
2426 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2427 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2429 ib[idx+1] = offset & 0xfffffffc;
2430 ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2433 case PACKET3_SET_CONFIG_REG:
2434 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2435 end_reg = 4 * pkt->count + start_reg - 4;
2436 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2437 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2438 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2439 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2442 for (i = 0; i < pkt->count; i++) {
2443 reg = start_reg + (4 * i);
2444 r = evergreen_cs_check_reg(p, reg, idx+1+i);
2449 case PACKET3_SET_CONTEXT_REG:
2450 start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_START;
2451 end_reg = 4 * pkt->count + start_reg - 4;
2452 if ((start_reg < PACKET3_SET_CONTEXT_REG_START) ||
2453 (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
2454 (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
2455 DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
2458 for (i = 0; i < pkt->count; i++) {
2459 reg = start_reg + (4 * i);
2460 r = evergreen_cs_check_reg(p, reg, idx+1+i);
2465 case PACKET3_SET_RESOURCE:
2466 if (pkt->count % 8) {
2467 DRM_ERROR("bad SET_RESOURCE\n");
2470 start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_START;
2471 end_reg = 4 * pkt->count + start_reg - 4;
2472 if ((start_reg < PACKET3_SET_RESOURCE_START) ||
2473 (start_reg >= PACKET3_SET_RESOURCE_END) ||
2474 (end_reg >= PACKET3_SET_RESOURCE_END)) {
2475 DRM_ERROR("bad SET_RESOURCE\n");
2478 for (i = 0; i < (pkt->count / 8); i++) {
2479 struct radeon_bo *texture, *mipmap;
2480 u32 toffset, moffset;
2481 u32 size, offset, mip_address, tex_dim;
2483 switch (G__SQ_CONSTANT_TYPE(radeon_get_ib_value(p, idx+1+(i*8)+7))) {
2484 case SQ_TEX_VTX_VALID_TEXTURE:
2486 r = evergreen_cs_packet_next_reloc(p, &reloc);
2488 DRM_ERROR("bad SET_RESOURCE (tex)\n");
2491 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
2492 ib[idx+1+(i*8)+1] |=
2493 TEX_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
2494 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
2495 unsigned bankw, bankh, mtaspect, tile_split;
2497 evergreen_tiling_fields(reloc->lobj.tiling_flags,
2498 &bankw, &bankh, &mtaspect,
2500 ib[idx+1+(i*8)+6] |= TEX_TILE_SPLIT(tile_split);
2501 ib[idx+1+(i*8)+7] |=
2502 TEX_BANK_WIDTH(bankw) |
2503 TEX_BANK_HEIGHT(bankh) |
2504 MACRO_TILE_ASPECT(mtaspect) |
2505 TEX_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
2508 texture = reloc->robj;
2509 toffset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
2512 tex_dim = ib[idx+1+(i*8)+0] & 0x7;
2513 mip_address = ib[idx+1+(i*8)+3];
2515 if ((tex_dim == SQ_TEX_DIM_2D_MSAA || tex_dim == SQ_TEX_DIM_2D_ARRAY_MSAA) &&
2517 !evergreen_cs_packet_next_is_pkt3_nop(p)) {
2518 /* MIP_ADDRESS should point to FMASK for an MSAA texture.
2519 * It should be 0 if FMASK is disabled. */
2523 r = evergreen_cs_packet_next_reloc(p, &reloc);
2525 DRM_ERROR("bad SET_RESOURCE (tex)\n");
2528 moffset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
2529 mipmap = reloc->robj;
2532 r = evergreen_cs_track_validate_texture(p, texture, mipmap, idx+1+(i*8));
2535 ib[idx+1+(i*8)+2] += toffset;
2536 ib[idx+1+(i*8)+3] += moffset;
2538 case SQ_TEX_VTX_VALID_BUFFER:
2542 r = evergreen_cs_packet_next_reloc(p, &reloc);
2544 DRM_ERROR("bad SET_RESOURCE (vtx)\n");
2547 offset = radeon_get_ib_value(p, idx+1+(i*8)+0);
2548 size = radeon_get_ib_value(p, idx+1+(i*8)+1);
2549 if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2550 /* force size to size of the buffer */
2551 dev_warn(p->dev, "vbo resource seems too big for the bo\n");
2552 ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj) - offset;
2555 offset64 = reloc->lobj.gpu_offset + offset;
2556 ib[idx+1+(i*8)+0] = offset64;
2557 ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2558 (upper_32_bits(offset64) & 0xff);
2561 case SQ_TEX_VTX_INVALID_TEXTURE:
2562 case SQ_TEX_VTX_INVALID_BUFFER:
2564 DRM_ERROR("bad SET_RESOURCE\n");
2569 case PACKET3_SET_ALU_CONST:
2570 /* XXX fix me ALU const buffers only */
2572 case PACKET3_SET_BOOL_CONST:
2573 start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_START;
2574 end_reg = 4 * pkt->count + start_reg - 4;
2575 if ((start_reg < PACKET3_SET_BOOL_CONST_START) ||
2576 (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2577 (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2578 DRM_ERROR("bad SET_BOOL_CONST\n");
2582 case PACKET3_SET_LOOP_CONST:
2583 start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_START;
2584 end_reg = 4 * pkt->count + start_reg - 4;
2585 if ((start_reg < PACKET3_SET_LOOP_CONST_START) ||
2586 (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2587 (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2588 DRM_ERROR("bad SET_LOOP_CONST\n");
2592 case PACKET3_SET_CTL_CONST:
2593 start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_START;
2594 end_reg = 4 * pkt->count + start_reg - 4;
2595 if ((start_reg < PACKET3_SET_CTL_CONST_START) ||
2596 (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2597 (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2598 DRM_ERROR("bad SET_CTL_CONST\n");
2602 case PACKET3_SET_SAMPLER:
2603 if (pkt->count % 3) {
2604 DRM_ERROR("bad SET_SAMPLER\n");
2607 start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_START;
2608 end_reg = 4 * pkt->count + start_reg - 4;
2609 if ((start_reg < PACKET3_SET_SAMPLER_START) ||
2610 (start_reg >= PACKET3_SET_SAMPLER_END) ||
2611 (end_reg >= PACKET3_SET_SAMPLER_END)) {
2612 DRM_ERROR("bad SET_SAMPLER\n");
2616 case PACKET3_STRMOUT_BUFFER_UPDATE:
2617 if (pkt->count != 4) {
2618 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2621 /* Updating memory at DST_ADDRESS. */
2622 if (idx_value & 0x1) {
2624 r = evergreen_cs_packet_next_reloc(p, &reloc);
2626 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2629 offset = radeon_get_ib_value(p, idx+1);
2630 offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2631 if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2632 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%jx, 0x%lx\n",
2633 (uintmax_t)offset + 4, radeon_bo_size(reloc->robj));
2636 offset += reloc->lobj.gpu_offset;
2638 ib[idx+2] = upper_32_bits(offset) & 0xff;
2640 /* Reading data from SRC_ADDRESS. */
2641 if (((idx_value >> 1) & 0x3) == 2) {
2643 r = evergreen_cs_packet_next_reloc(p, &reloc);
2645 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2648 offset = radeon_get_ib_value(p, idx+3);
2649 offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2650 if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2651 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%jx, 0x%lx\n",
2652 (uintmax_t)offset + 4, radeon_bo_size(reloc->robj));
2655 offset += reloc->lobj.gpu_offset;
2657 ib[idx+4] = upper_32_bits(offset) & 0xff;
2660 case PACKET3_MEM_WRITE:
2664 if (pkt->count != 3) {
2665 DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2668 r = evergreen_cs_packet_next_reloc(p, &reloc);
2670 DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2673 offset = radeon_get_ib_value(p, idx+0);
2674 offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2676 DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2679 if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2680 DRM_ERROR("bad MEM_WRITE bo too small: 0x%jx, 0x%lx\n",
2681 (uintmax_t)offset + 8, radeon_bo_size(reloc->robj));
2684 offset += reloc->lobj.gpu_offset;
2686 ib[idx+1] = upper_32_bits(offset) & 0xff;
2689 case PACKET3_COPY_DW:
2690 if (pkt->count != 4) {
2691 DRM_ERROR("bad COPY_DW (invalid count)\n");
2694 if (idx_value & 0x1) {
2696 /* SRC is memory. */
2697 r = evergreen_cs_packet_next_reloc(p, &reloc);
2699 DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2702 offset = radeon_get_ib_value(p, idx+1);
2703 offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2704 if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2705 DRM_ERROR("bad COPY_DW src bo too small: 0x%jx, 0x%lx\n",
2706 (uintmax_t)offset + 4, radeon_bo_size(reloc->robj));
2709 offset += reloc->lobj.gpu_offset;
2711 ib[idx+2] = upper_32_bits(offset) & 0xff;
2714 reg = radeon_get_ib_value(p, idx+1) << 2;
2715 if (!evergreen_is_safe_reg(p, reg, idx+1))
2718 if (idx_value & 0x2) {
2720 /* DST is memory. */
2721 r = evergreen_cs_packet_next_reloc(p, &reloc);
2723 DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2726 offset = radeon_get_ib_value(p, idx+3);
2727 offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2728 if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2729 DRM_ERROR("bad COPY_DW dst bo too small: 0x%jx, 0x%lx\n",
2730 (uintmax_t)offset + 4, radeon_bo_size(reloc->robj));
2733 offset += reloc->lobj.gpu_offset;
2735 ib[idx+4] = upper_32_bits(offset) & 0xff;
2738 reg = radeon_get_ib_value(p, idx+3) << 2;
2739 if (!evergreen_is_safe_reg(p, reg, idx+3))
2746 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2752 int evergreen_cs_parse(struct radeon_cs_parser *p)
2754 struct radeon_cs_packet pkt;
2755 struct evergreen_cs_track *track;
2759 if (p->track == NULL) {
2760 /* initialize tracker, we are in kms */
2761 track = kmalloc(sizeof(*track), M_DRM,
2765 evergreen_cs_track_init(track);
2766 if (p->rdev->family >= CHIP_CAYMAN)
2767 tmp = p->rdev->config.cayman.tile_config;
2769 tmp = p->rdev->config.evergreen.tile_config;
2771 switch (tmp & 0xf) {
2787 switch ((tmp & 0xf0) >> 4) {
2800 switch ((tmp & 0xf00) >> 8) {
2802 track->group_size = 256;
2806 track->group_size = 512;
2810 switch ((tmp & 0xf000) >> 12) {
2812 track->row_size = 1;
2816 track->row_size = 2;
2819 track->row_size = 4;
2826 r = evergreen_cs_packet_parse(p, &pkt, p->idx);
2828 drm_free(p->track, M_DRM);
2832 p->idx += pkt.count + 2;
2835 r = evergreen_cs_parse_packet0(p, &pkt);
2840 r = evergreen_packet3_check(p, &pkt);
2843 DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2844 drm_free(p->track, M_DRM);
2849 drm_free(p->track, M_DRM);
2853 } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
2855 for (r = 0; r < p->ib.length_dw; r++) {
2856 DRM_INFO("%05d 0x%08X\n", r, p->ib.ptr[r]);
2860 drm_free(p->track, M_DRM);
2869 #define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
2870 #define GET_DMA_COUNT(h) ((h) & 0x000fffff)
2871 #define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
2872 #define GET_DMA_NEW(h) (((h) & 0x04000000) >> 26)
2873 #define GET_DMA_MISC(h) (((h) & 0x0700000) >> 20)
2876 * evergreen_dma_cs_parse() - parse the DMA IB
2877 * @p: parser structure holding parsing context.
2879 * Parses the DMA IB from the CS ioctl and updates
2880 * the GPU addresses based on the reloc information and
2881 * checks for errors. (Evergreen-Cayman)
2882 * Returns 0 for success and an error on failure.
2884 int evergreen_dma_cs_parse(struct radeon_cs_parser *p)
2886 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
2887 struct radeon_cs_reloc *src_reloc, *dst_reloc, *dst2_reloc;
2888 u32 header, cmd, count, tiled, new_cmd, misc;
2889 volatile u32 *ib = p->ib.ptr;
2891 u64 src_offset, dst_offset, dst2_offset;
2895 if (p->idx >= ib_chunk->length_dw) {
2896 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2897 p->idx, ib_chunk->length_dw);
2901 header = radeon_get_ib_value(p, idx);
2902 cmd = GET_DMA_CMD(header);
2903 count = GET_DMA_COUNT(header);
2904 tiled = GET_DMA_T(header);
2905 new_cmd = GET_DMA_NEW(header);
2906 misc = GET_DMA_MISC(header);
2909 case DMA_PACKET_WRITE:
2910 r = r600_dma_cs_next_reloc(p, &dst_reloc);
2912 DRM_ERROR("bad DMA_PACKET_WRITE\n");
2916 dst_offset = radeon_get_ib_value(p, idx+1);
2919 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
2920 p->idx += count + 7;
2922 dst_offset = radeon_get_ib_value(p, idx+1);
2923 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2925 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
2926 ib[idx+2] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
2927 p->idx += count + 3;
2929 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2930 dev_warn(p->dev, "DMA write buffer too small (%ju %lu)\n",
2931 (uintmax_t)dst_offset, radeon_bo_size(dst_reloc->robj));
2935 case DMA_PACKET_COPY:
2936 r = r600_dma_cs_next_reloc(p, &src_reloc);
2938 DRM_ERROR("bad DMA_PACKET_COPY\n");
2941 r = r600_dma_cs_next_reloc(p, &dst_reloc);
2943 DRM_ERROR("bad DMA_PACKET_COPY\n");
2947 idx_value = radeon_get_ib_value(p, idx + 2);
2951 /* L2T, frame to fields */
2952 if (idx_value & (1 << 31)) {
2953 DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
2956 r = r600_dma_cs_next_reloc(p, &dst2_reloc);
2958 DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
2961 dst_offset = radeon_get_ib_value(p, idx+1);
2963 dst2_offset = radeon_get_ib_value(p, idx+2);
2965 src_offset = radeon_get_ib_value(p, idx+8);
2966 src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
2967 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2968 dev_warn(p->dev, "DMA L2T, frame to fields src buffer too small (%ju %lu)\n",
2969 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2972 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2973 dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%ju %lu)\n",
2974 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2977 if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
2978 dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%ju %lu)\n",
2979 (uintmax_t)dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
2982 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
2983 ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset >> 8);
2984 ib[idx+8] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
2985 ib[idx+9] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
2989 /* L2T, T2L partial */
2990 if (p->family < CHIP_CAYMAN) {
2991 DRM_ERROR("L2T, T2L Partial is cayman only !\n");
2995 if (idx_value & (1 << 31)) {
2996 /* tiled src, linear dst */
2997 ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
2999 ib[idx+7] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
3000 ib[idx+8] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3002 /* linear src, tiled dst */
3003 ib[idx+7] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3004 ib[idx+8] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3006 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
3011 /* L2T, broadcast */
3012 if (idx_value & (1 << 31)) {
3013 DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3016 r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3018 DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3021 dst_offset = radeon_get_ib_value(p, idx+1);
3023 dst2_offset = radeon_get_ib_value(p, idx+2);
3025 src_offset = radeon_get_ib_value(p, idx+8);
3026 src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3027 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3028 dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%ju %lu)\n",
3029 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3032 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3033 dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%ju %lu)\n",
3034 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3037 if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3038 dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%ju %lu)\n",
3039 (uintmax_t)dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3042 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
3043 ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset >> 8);
3044 ib[idx+8] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3045 ib[idx+9] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3051 if (idx_value & (1 << 31)) {
3052 /* tiled src, linear dst */
3053 src_offset = radeon_get_ib_value(p, idx+1);
3055 ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
3057 dst_offset = radeon_get_ib_value(p, idx+7);
3058 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3059 ib[idx+7] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
3060 ib[idx+8] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3062 /* linear src, tiled dst */
3063 src_offset = radeon_get_ib_value(p, idx+7);
3064 src_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3065 ib[idx+7] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3066 ib[idx+8] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3068 dst_offset = radeon_get_ib_value(p, idx+1);
3070 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
3072 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3073 dev_warn(p->dev, "DMA L2T, T2L src buffer too small (%ju %lu)\n",
3074 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3077 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3078 dev_warn(p->dev, "DMA L2T, T2L dst buffer too small (%ju %lu)\n",
3079 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3086 if (p->family < CHIP_CAYMAN) {
3087 DRM_ERROR("L2T, T2L Partial is cayman only !\n");
3090 ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
3091 ib[idx+4] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
3095 /* L2T, broadcast */
3096 if (idx_value & (1 << 31)) {
3097 DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3100 r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3102 DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3105 dst_offset = radeon_get_ib_value(p, idx+1);
3107 dst2_offset = radeon_get_ib_value(p, idx+2);
3109 src_offset = radeon_get_ib_value(p, idx+8);
3110 src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3111 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3112 dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%ju %lu)\n",
3113 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3116 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3117 dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%ju %lu)\n",
3118 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3121 if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3122 dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%ju %lu)\n",
3123 (uintmax_t)dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3126 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
3127 ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset >> 8);
3128 ib[idx+8] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3129 ib[idx+9] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3133 DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
3140 if (idx_value & (1 << 31)) {
3141 /* tiled src, linear dst */
3142 src_offset = radeon_get_ib_value(p, idx+1);
3144 ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
3146 dst_offset = radeon_get_ib_value(p, idx+7);
3147 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3148 ib[idx+7] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
3149 ib[idx+8] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3151 /* linear src, tiled dst */
3152 src_offset = radeon_get_ib_value(p, idx+7);
3153 src_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3154 ib[idx+7] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3155 ib[idx+8] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3157 dst_offset = radeon_get_ib_value(p, idx+1);
3159 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
3161 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3162 dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%ju %lu)\n",
3163 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3166 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3167 dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%ju %lu)\n",
3168 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3174 DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
3183 src_offset = radeon_get_ib_value(p, idx+2);
3184 src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
3185 dst_offset = radeon_get_ib_value(p, idx+1);
3186 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
3187 if ((src_offset + count) > radeon_bo_size(src_reloc->robj)) {
3188 dev_warn(p->dev, "DMA L2L, byte src buffer too small (%ju %lu)\n",
3189 (uintmax_t)src_offset + count, radeon_bo_size(src_reloc->robj));
3192 if ((dst_offset + count) > radeon_bo_size(dst_reloc->robj)) {
3193 dev_warn(p->dev, "DMA L2L, byte dst buffer too small (%ju %lu)\n",
3194 (uintmax_t)dst_offset + count, radeon_bo_size(dst_reloc->robj));
3197 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xffffffff);
3198 ib[idx+2] += (u32)(src_reloc->lobj.gpu_offset & 0xffffffff);
3199 ib[idx+3] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3200 ib[idx+4] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3205 if (p->family < CHIP_CAYMAN) {
3206 DRM_ERROR("L2L Partial is cayman only !\n");
3209 ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset & 0xffffffff);
3210 ib[idx+2] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3211 ib[idx+4] += (u32)(dst_reloc->lobj.gpu_offset & 0xffffffff);
3212 ib[idx+5] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3217 /* L2L, dw, broadcast */
3218 r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3220 DRM_ERROR("bad L2L, dw, broadcast DMA_PACKET_COPY\n");
3223 dst_offset = radeon_get_ib_value(p, idx+1);
3224 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
3225 dst2_offset = radeon_get_ib_value(p, idx+2);
3226 dst2_offset |= ((u64)(radeon_get_ib_value(p, idx+5) & 0xff)) << 32;
3227 src_offset = radeon_get_ib_value(p, idx+3);
3228 src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
3229 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3230 dev_warn(p->dev, "DMA L2L, dw, broadcast src buffer too small (%ju %lu)\n",
3231 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3234 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3235 dev_warn(p->dev, "DMA L2L, dw, broadcast dst buffer too small (%ju %lu)\n",
3236 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3239 if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3240 dev_warn(p->dev, "DMA L2L, dw, broadcast dst2 buffer too small (%ju %lu)\n",
3241 (uintmax_t)dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3244 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
3245 ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset & 0xfffffffc);
3246 ib[idx+3] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3247 ib[idx+4] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3248 ib[idx+5] += upper_32_bits(dst2_reloc->lobj.gpu_offset) & 0xff;
3249 ib[idx+6] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3253 DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
3258 src_offset = radeon_get_ib_value(p, idx+2);
3259 src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
3260 dst_offset = radeon_get_ib_value(p, idx+1);
3261 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
3262 if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3263 dev_warn(p->dev, "DMA L2L, dw src buffer too small (%ju %lu)\n",
3264 (uintmax_t)src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3267 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3268 dev_warn(p->dev, "DMA L2L, dw dst buffer too small (%ju %lu)\n",
3269 (uintmax_t)dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3272 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
3273 ib[idx+2] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
3274 ib[idx+3] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
3275 ib[idx+4] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
3280 case DMA_PACKET_CONSTANT_FILL:
3281 r = r600_dma_cs_next_reloc(p, &dst_reloc);
3283 DRM_ERROR("bad DMA_PACKET_CONSTANT_FILL\n");
3286 dst_offset = radeon_get_ib_value(p, idx+1);
3287 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
3288 if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3289 dev_warn(p->dev, "DMA constant fill buffer too small (%ju %lu)\n",
3290 (uintmax_t)dst_offset, radeon_bo_size(dst_reloc->robj));
3293 ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
3294 ib[idx+3] += (upper_32_bits(dst_reloc->lobj.gpu_offset) << 16) & 0x00ff0000;
3297 case DMA_PACKET_NOP:
3301 DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
3304 } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
3306 for (r = 0; r < p->ib->length_dw; r++) {
3307 DRM_INFO("%05d 0x%08X\n", r, p->ib.ptr[r]);
3315 static bool evergreen_vm_reg_valid(u32 reg)
3317 /* context regs are fine */
3321 /* check config regs */
3324 case GRBM_GFX_INDEX:
3325 case CP_STRMOUT_CNTL:
3328 case VGT_VTX_VECT_EJECT_REG:
3329 case VGT_CACHE_INVALIDATION:
3330 case VGT_GS_VERTEX_REUSE:
3331 case VGT_PRIMITIVE_TYPE:
3332 case VGT_INDEX_TYPE:
3333 case VGT_NUM_INDICES:
3334 case VGT_NUM_INSTANCES:
3335 case VGT_COMPUTE_DIM_X:
3336 case VGT_COMPUTE_DIM_Y:
3337 case VGT_COMPUTE_DIM_Z:
3338 case VGT_COMPUTE_START_X:
3339 case VGT_COMPUTE_START_Y:
3340 case VGT_COMPUTE_START_Z:
3341 case VGT_COMPUTE_INDEX:
3342 case VGT_COMPUTE_THREAD_GROUP_SIZE:
3343 case VGT_HS_OFFCHIP_PARAM:
3345 case PA_SU_LINE_STIPPLE_VALUE:
3346 case PA_SC_LINE_STIPPLE_STATE:
3348 case SQ_DYN_GPR_CNTL_PS_FLUSH_REQ:
3349 case SQ_DYN_GPR_SIMD_LOCK_EN:
3351 case SQ_GPR_RESOURCE_MGMT_1:
3352 case SQ_GLOBAL_GPR_RESOURCE_MGMT_1:
3353 case SQ_GLOBAL_GPR_RESOURCE_MGMT_2:
3354 case SQ_CONST_MEM_BASE:
3355 case SQ_STATIC_THREAD_MGMT_1:
3356 case SQ_STATIC_THREAD_MGMT_2:
3357 case SQ_STATIC_THREAD_MGMT_3:
3358 case SPI_CONFIG_CNTL:
3359 case SPI_CONFIG_CNTL_1:
3366 case TD_PS_BORDER_COLOR_INDEX:
3367 case TD_PS_BORDER_COLOR_RED:
3368 case TD_PS_BORDER_COLOR_GREEN:
3369 case TD_PS_BORDER_COLOR_BLUE:
3370 case TD_PS_BORDER_COLOR_ALPHA:
3371 case TD_VS_BORDER_COLOR_INDEX:
3372 case TD_VS_BORDER_COLOR_RED:
3373 case TD_VS_BORDER_COLOR_GREEN:
3374 case TD_VS_BORDER_COLOR_BLUE:
3375 case TD_VS_BORDER_COLOR_ALPHA:
3376 case TD_GS_BORDER_COLOR_INDEX:
3377 case TD_GS_BORDER_COLOR_RED:
3378 case TD_GS_BORDER_COLOR_GREEN:
3379 case TD_GS_BORDER_COLOR_BLUE:
3380 case TD_GS_BORDER_COLOR_ALPHA:
3381 case TD_HS_BORDER_COLOR_INDEX:
3382 case TD_HS_BORDER_COLOR_RED:
3383 case TD_HS_BORDER_COLOR_GREEN:
3384 case TD_HS_BORDER_COLOR_BLUE:
3385 case TD_HS_BORDER_COLOR_ALPHA:
3386 case TD_LS_BORDER_COLOR_INDEX:
3387 case TD_LS_BORDER_COLOR_RED:
3388 case TD_LS_BORDER_COLOR_GREEN:
3389 case TD_LS_BORDER_COLOR_BLUE:
3390 case TD_LS_BORDER_COLOR_ALPHA:
3391 case TD_CS_BORDER_COLOR_INDEX:
3392 case TD_CS_BORDER_COLOR_RED:
3393 case TD_CS_BORDER_COLOR_GREEN:
3394 case TD_CS_BORDER_COLOR_BLUE:
3395 case TD_CS_BORDER_COLOR_ALPHA:
3396 case SQ_ESGS_RING_SIZE:
3397 case SQ_GSVS_RING_SIZE:
3398 case SQ_ESTMP_RING_SIZE:
3399 case SQ_GSTMP_RING_SIZE:
3400 case SQ_HSTMP_RING_SIZE:
3401 case SQ_LSTMP_RING_SIZE:
3402 case SQ_PSTMP_RING_SIZE:
3403 case SQ_VSTMP_RING_SIZE:
3404 case SQ_ESGS_RING_ITEMSIZE:
3405 case SQ_ESTMP_RING_ITEMSIZE:
3406 case SQ_GSTMP_RING_ITEMSIZE:
3407 case SQ_GSVS_RING_ITEMSIZE:
3408 case SQ_GS_VERT_ITEMSIZE:
3409 case SQ_GS_VERT_ITEMSIZE_1:
3410 case SQ_GS_VERT_ITEMSIZE_2:
3411 case SQ_GS_VERT_ITEMSIZE_3:
3412 case SQ_GSVS_RING_OFFSET_1:
3413 case SQ_GSVS_RING_OFFSET_2:
3414 case SQ_GSVS_RING_OFFSET_3:
3415 case SQ_HSTMP_RING_ITEMSIZE:
3416 case SQ_LSTMP_RING_ITEMSIZE:
3417 case SQ_PSTMP_RING_ITEMSIZE:
3418 case SQ_VSTMP_RING_ITEMSIZE:
3419 case VGT_TF_RING_SIZE:
3420 case SQ_ESGS_RING_BASE:
3421 case SQ_GSVS_RING_BASE:
3422 case SQ_ESTMP_RING_BASE:
3423 case SQ_GSTMP_RING_BASE:
3424 case SQ_HSTMP_RING_BASE:
3425 case SQ_LSTMP_RING_BASE:
3426 case SQ_PSTMP_RING_BASE:
3427 case SQ_VSTMP_RING_BASE:
3428 case CAYMAN_VGT_OFFCHIP_LDS_BASE:
3429 case CAYMAN_SQ_EX_ALLOC_TABLE_SLOTS:
3432 DRM_ERROR("Invalid register 0x%x in CS\n", reg);
3437 static int evergreen_vm_packet3_check(struct radeon_device *rdev,
3438 u32 *ib, struct radeon_cs_packet *pkt)
3440 u32 idx = pkt->idx + 1;
3441 u32 idx_value = ib[idx];
3442 u32 start_reg, end_reg, reg, i;
3445 switch (pkt->opcode) {
3447 case PACKET3_SET_BASE:
3448 case PACKET3_CLEAR_STATE:
3449 case PACKET3_INDEX_BUFFER_SIZE:
3450 case PACKET3_DISPATCH_DIRECT:
3451 case PACKET3_DISPATCH_INDIRECT:
3452 case PACKET3_MODE_CONTROL:
3453 case PACKET3_SET_PREDICATION:
3454 case PACKET3_COND_EXEC:
3455 case PACKET3_PRED_EXEC:
3456 case PACKET3_DRAW_INDIRECT:
3457 case PACKET3_DRAW_INDEX_INDIRECT:
3458 case PACKET3_INDEX_BASE:
3459 case PACKET3_DRAW_INDEX_2:
3460 case PACKET3_CONTEXT_CONTROL:
3461 case PACKET3_DRAW_INDEX_OFFSET:
3462 case PACKET3_INDEX_TYPE:
3463 case PACKET3_DRAW_INDEX:
3464 case PACKET3_DRAW_INDEX_AUTO:
3465 case PACKET3_DRAW_INDEX_IMMD:
3466 case PACKET3_NUM_INSTANCES:
3467 case PACKET3_DRAW_INDEX_MULTI_AUTO:
3468 case PACKET3_STRMOUT_BUFFER_UPDATE:
3469 case PACKET3_DRAW_INDEX_OFFSET_2:
3470 case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
3471 case PACKET3_MPEG_INDEX:
3472 case PACKET3_WAIT_REG_MEM:
3473 case PACKET3_MEM_WRITE:
3474 case PACKET3_SURFACE_SYNC:
3475 case PACKET3_EVENT_WRITE:
3476 case PACKET3_EVENT_WRITE_EOP:
3477 case PACKET3_EVENT_WRITE_EOS:
3478 case PACKET3_SET_CONTEXT_REG:
3479 case PACKET3_SET_BOOL_CONST:
3480 case PACKET3_SET_LOOP_CONST:
3481 case PACKET3_SET_RESOURCE:
3482 case PACKET3_SET_SAMPLER:
3483 case PACKET3_SET_CTL_CONST:
3484 case PACKET3_SET_RESOURCE_OFFSET:
3485 case PACKET3_SET_CONTEXT_REG_INDIRECT:
3486 case PACKET3_SET_RESOURCE_INDIRECT:
3487 case CAYMAN_PACKET3_DEALLOC_STATE:
3489 case PACKET3_COND_WRITE:
3490 if (idx_value & 0x100) {
3491 reg = ib[idx + 5] * 4;
3492 if (!evergreen_vm_reg_valid(reg))
3496 case PACKET3_COPY_DW:
3497 if (idx_value & 0x2) {
3498 reg = ib[idx + 3] * 4;
3499 if (!evergreen_vm_reg_valid(reg))
3503 case PACKET3_SET_CONFIG_REG:
3504 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
3505 end_reg = 4 * pkt->count + start_reg - 4;
3506 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
3507 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
3508 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
3509 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
3512 for (i = 0; i < pkt->count; i++) {
3513 reg = start_reg + (4 * i);
3514 if (!evergreen_vm_reg_valid(reg))
3518 case PACKET3_CP_DMA:
3519 command = ib[idx + 4];
3521 if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
3522 (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
3523 ((((info & 0x00300000) >> 20) == 0) &&
3524 (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
3525 ((((info & 0x60000000) >> 29) == 0) &&
3526 (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
3527 /* non mem to mem copies requires dw aligned count */
3528 if ((command & 0x1fffff) % 4) {
3529 DRM_ERROR("CP DMA command requires dw count alignment\n");
3533 if (command & PACKET3_CP_DMA_CMD_SAS) {
3534 /* src address space is register */
3535 if (((info & 0x60000000) >> 29) == 0) {
3536 start_reg = idx_value << 2;
3537 if (command & PACKET3_CP_DMA_CMD_SAIC) {
3539 if (!evergreen_vm_reg_valid(reg)) {
3540 DRM_ERROR("CP DMA Bad SRC register\n");
3544 for (i = 0; i < (command & 0x1fffff); i++) {
3545 reg = start_reg + (4 * i);
3546 if (!evergreen_vm_reg_valid(reg)) {
3547 DRM_ERROR("CP DMA Bad SRC register\n");
3554 if (command & PACKET3_CP_DMA_CMD_DAS) {
3555 /* dst address space is register */
3556 if (((info & 0x00300000) >> 20) == 0) {
3557 start_reg = ib[idx + 2];
3558 if (command & PACKET3_CP_DMA_CMD_DAIC) {
3560 if (!evergreen_vm_reg_valid(reg)) {
3561 DRM_ERROR("CP DMA Bad DST register\n");
3565 for (i = 0; i < (command & 0x1fffff); i++) {
3566 reg = start_reg + (4 * i);
3567 if (!evergreen_vm_reg_valid(reg)) {
3568 DRM_ERROR("CP DMA Bad DST register\n");
3582 int evergreen_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3586 struct radeon_cs_packet pkt;
3590 pkt.type = CP_PACKET_GET_TYPE(ib->ptr[idx]);
3591 pkt.count = CP_PACKET_GET_COUNT(ib->ptr[idx]);
3595 dev_err(rdev->dev, "Packet0 not allowed!\n");
3602 pkt.opcode = CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
3603 ret = evergreen_vm_packet3_check(rdev, ib->ptr, &pkt);
3604 idx += pkt.count + 2;
3607 dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
3613 } while (idx < ib->length_dw);
3619 * evergreen_dma_ib_parse() - parse the DMA IB for VM
3620 * @rdev: radeon_device pointer
3621 * @ib: radeon_ib pointer
3623 * Parses the DMA IB from the VM CS ioctl
3624 * checks for errors. (Cayman-SI)
3625 * Returns 0 for success and an error on failure.
3627 int evergreen_dma_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3630 u32 header, cmd, count, tiled, new_cmd, misc;
3633 header = ib->ptr[idx];
3634 cmd = GET_DMA_CMD(header);
3635 count = GET_DMA_COUNT(header);
3636 tiled = GET_DMA_T(header);
3637 new_cmd = GET_DMA_NEW(header);
3638 misc = GET_DMA_MISC(header);
3641 case DMA_PACKET_WRITE:
3647 case DMA_PACKET_COPY:
3652 /* L2T, frame to fields */
3656 /* L2T, T2L partial */
3660 /* L2T, broadcast */
3672 /* L2T, broadcast */
3676 DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
3685 DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
3701 /* L2L, dw, broadcast */
3705 DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
3714 case DMA_PACKET_CONSTANT_FILL:
3717 case DMA_PACKET_NOP:
3721 DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
3724 } while (idx < ib->length_dw);