drivers/gpu/drm/radeon/r600_cs.c

   1 /*
   2  * Copyright 2008 Advanced Micro Devices, Inc.
   3  * Copyright 2008 Red Hat Inc.
   4  * Copyright 2009 Jerome Glisse.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22  * OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors: Dave Airlie
  25  *          Alex Deucher
  26  *          Jerome Glisse
  27  */
  28 #include <linux/kernel.h>
  29 #include "drmP.h"
  30 #include "radeon.h"
  31 #include "r600d.h"
  32 #include "r600_reg_safe.h"
  33
  34 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
  35                                         struct radeon_cs_reloc **cs_reloc);
  36 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
  37                                         struct radeon_cs_reloc **cs_reloc);
  38 typedef int (*next_reloc_t)(struct radeon_cs_parser*, struct radeon_cs_reloc**);
  39 static next_reloc_t r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_mm;
  40 extern void r600_cs_legacy_get_tiling_conf(struct drm_device *dev, u32 *npipes, u32 *nbanks, u32 *group_size);
  41
  42
  43 struct r600_cs_track {
  44         /* configuration we miror so that we use same code btw kms/ums */
  45         u32                     group_size;
  46         u32                     nbanks;
  47         u32                     npipes;
  48         /* value we track */
  49         u32                     sq_config;
  50         u32                     nsamples;
  51         u32                     cb_color_base_last[8];
  52         struct radeon_bo        *cb_color_bo[8];
  53         u32                     cb_color_bo_offset[8];
  54         struct radeon_bo        *cb_color_frag_bo[8];
  55         struct radeon_bo        *cb_color_tile_bo[8];
  56         u32                     cb_color_info[8];
  57         u32                     cb_color_size_idx[8];
  58         u32                     cb_target_mask;
  59         u32                     cb_shader_mask;
  60         u32                     cb_color_size[8];
  61         u32                     vgt_strmout_en;
  62         u32                     vgt_strmout_buffer_en;
  63         u32                     db_depth_control;
  64         u32                     db_depth_info;
  65         u32                     db_depth_size_idx;
  66         u32                     db_depth_view;
  67         u32                     db_depth_size;
  68         u32                     db_offset;
  69         struct radeon_bo        *db_bo;
  70 };
  71
  72 static inline int r600_bpe_from_format(u32 *bpe, u32 format)
  73 {
  74         switch (format) {
  75         case V_038004_COLOR_8:
  76         case V_038004_COLOR_4_4:
  77         case V_038004_COLOR_3_3_2:
  78         case V_038004_FMT_1:
  79                 *bpe = 1;
  80                 break;
  81         case V_038004_COLOR_16:
  82         case V_038004_COLOR_16_FLOAT:
  83         case V_038004_COLOR_8_8:
  84         case V_038004_COLOR_5_6_5:
  85         case V_038004_COLOR_6_5_5:
  86         case V_038004_COLOR_1_5_5_5:
  87         case V_038004_COLOR_4_4_4_4:
  88         case V_038004_COLOR_5_5_5_1:
  89                 *bpe = 2;
  90                 break;
  91         case V_038004_FMT_8_8_8:
  92                 *bpe = 3;
  93                 break;
  94         case V_038004_COLOR_32:
  95         case V_038004_COLOR_32_FLOAT:
  96         case V_038004_COLOR_16_16:
  97         case V_038004_COLOR_16_16_FLOAT:
  98         case V_038004_COLOR_8_24:
  99         case V_038004_COLOR_8_24_FLOAT:
 100         case V_038004_COLOR_24_8:
 101         case V_038004_COLOR_24_8_FLOAT:
 102         case V_038004_COLOR_10_11_11:
 103         case V_038004_COLOR_10_11_11_FLOAT:
 104         case V_038004_COLOR_11_11_10:
 105         case V_038004_COLOR_11_11_10_FLOAT:
 106         case V_038004_COLOR_2_10_10_10:
 107         case V_038004_COLOR_8_8_8_8:
 108         case V_038004_COLOR_10_10_10_2:
 109         case V_038004_FMT_5_9_9_9_SHAREDEXP:
 110         case V_038004_FMT_32_AS_8:
 111         case V_038004_FMT_32_AS_8_8:
 112                 *bpe = 4;
 113                 break;
 114         case V_038004_COLOR_X24_8_32_FLOAT:
 115         case V_038004_COLOR_32_32:
 116         case V_038004_COLOR_32_32_FLOAT:
 117         case V_038004_COLOR_16_16_16_16:
 118         case V_038004_COLOR_16_16_16_16_FLOAT:
 119                 *bpe = 8;
 120                 break;
 121         case V_038004_FMT_16_16_16:
 122         case V_038004_FMT_16_16_16_FLOAT:
 123                 *bpe = 6;
 124                 break;
 125         case V_038004_FMT_32_32_32:
 126         case V_038004_FMT_32_32_32_FLOAT:
 127                 *bpe = 12;
 128                 break;
 129         case V_038004_COLOR_32_32_32_32:
 130         case V_038004_COLOR_32_32_32_32_FLOAT:
 131                 *bpe = 16;
 132                 break;
 133         case V_038004_FMT_GB_GR:
 134         case V_038004_FMT_BG_RG:
 135         case V_038004_COLOR_INVALID:
 136                 *bpe = 16;
 137                 return -EINVAL;
 138         }
 139         return 0;
 140 }
 141
 142 static void r600_cs_track_init(struct r600_cs_track *track)
 143 {
 144         int i;
 145
 146         /* assume DX9 mode */
 147         track->sq_config = DX9_CONSTS;
 148         for (i = 0; i < 8; i++) {
 149                 track->cb_color_base_last[i] = 0;
 150                 track->cb_color_size[i] = 0;
 151                 track->cb_color_size_idx[i] = 0;
 152                 track->cb_color_info[i] = 0;
 153                 track->cb_color_bo[i] = NULL;
 154                 track->cb_color_bo_offset[i] = 0xFFFFFFFF;
 155         }
 156         track->cb_target_mask = 0xFFFFFFFF;
 157         track->cb_shader_mask = 0xFFFFFFFF;
 158         track->db_bo = NULL;
 159         /* assume the biggest format and that htile is enabled */
 160         track->db_depth_info = 7 | (1 << 25);
 161         track->db_depth_view = 0xFFFFC000;
 162         track->db_depth_size = 0xFFFFFFFF;
 163         track->db_depth_size_idx = 0;
 164         track->db_depth_control = 0xFFFFFFFF;
 165 }
 166
 167 static inline int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
 168 {
 169         struct r600_cs_track *track = p->track;
 170         u32 bpe = 0, pitch, slice_tile_max, size, tmp, height, pitch_align;
 171         volatile u32 *ib = p->ib->ptr;
 172
 173         if (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
 174                 dev_warn(p->dev, "FMASK or CMASK buffer are not supported by this kernel\n");
 175                 return -EINVAL;
 176         }
 177         size = radeon_bo_size(track->cb_color_bo[i]);
 178         if (r600_bpe_from_format(&bpe, G_0280A0_FORMAT(track->cb_color_info[i]))) {
 179                 dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
 180                          __func__, __LINE__, G_0280A0_FORMAT(track->cb_color_info[i]),
 181                         i, track->cb_color_info[i]);
 182                 return -EINVAL;
 183         }
 184         /* pitch is the number of 8x8 tiles per row */
 185         pitch = G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1;
 186         slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
 187         height = size / (pitch * 8 * bpe);
 188         if (height > 8192)
 189                 height = 8192;
 190         if (height > 7)
 191                 height &= ~0x7;
 192         switch (G_0280A0_ARRAY_MODE(track->cb_color_info[i])) {
 193         case V_0280A0_ARRAY_LINEAR_GENERAL:
 194                 /* technically height & 0x7 */
 195                 break;
 196         case V_0280A0_ARRAY_LINEAR_ALIGNED:
 197                 pitch_align = max((u32)64, (u32)(track->group_size / bpe)) / 8;
 198                 if (!IS_ALIGNED(pitch, pitch_align)) {
 199                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 200                                  __func__, __LINE__, pitch);
 201                         return -EINVAL;
 202                 }
 203                 if (!IS_ALIGNED(height, 8)) {
 204                         dev_warn(p->dev, "%s:%d cb height (%d) invalid\n",
 205                                  __func__, __LINE__, height);
 206                         return -EINVAL;
 207                 }
 208                 break;
 209         case V_0280A0_ARRAY_1D_TILED_THIN1:
 210                 pitch_align = max((u32)8, (u32)(track->group_size / (8 * bpe * track->nsamples))) / 8;
 211                 if (!IS_ALIGNED(pitch, pitch_align)) {
 212                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 213                                  __func__, __LINE__, pitch);
 214                         return -EINVAL;
 215                 }
 216                 if (!IS_ALIGNED(height, 8)) {
 217                         dev_warn(p->dev, "%s:%d cb height (%d) invalid\n",
 218                                  __func__, __LINE__, height);
 219                         return -EINVAL;
 220                 }
 221                 break;
 222         case V_0280A0_ARRAY_2D_TILED_THIN1:
 223                 pitch_align = max((u32)track->nbanks,
 224                                   (u32)(((track->group_size / 8) / (bpe * track->nsamples)) * track->nbanks));
 225                 if (!IS_ALIGNED(pitch, pitch_align)) {
 226                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 227                                 __func__, __LINE__, pitch);
 228                         return -EINVAL;
 229                 }
 230                 if (!IS_ALIGNED((height / 8), track->nbanks)) {
 231                         dev_warn(p->dev, "%s:%d cb height (%d) invalid\n",
 232                                  __func__, __LINE__, height);
 233                         return -EINVAL;
 234                 }
 235                 break;
 236         default:
 237                 dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
 238                         G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
 239                         track->cb_color_info[i]);
 240                 return -EINVAL;
 241         }
 242         /* check offset */
 243         tmp = height * pitch * 8 * bpe;
 244         if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
 245                 dev_warn(p->dev, "%s offset[%d] %d too big\n", __func__, i, track->cb_color_bo_offset[i]);
 246                 return -EINVAL;
 247         }
 248         if (!IS_ALIGNED(track->cb_color_bo_offset[i], track->group_size)) {
 249                 dev_warn(p->dev, "%s offset[%d] %d not aligned\n", __func__, i, track->cb_color_bo_offset[i]);
 250                 return -EINVAL;
 251         }
 252         /* limit max tile */
 253         tmp = (height * pitch * 8) >> 6;
 254         if (tmp < slice_tile_max)
 255                 slice_tile_max = tmp;
 256         tmp = S_028060_PITCH_TILE_MAX(pitch - 1) |
 257                 S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
 258         ib[track->cb_color_size_idx[i]] = tmp;
 259         return 0;
 260 }
 261
 262 static int r600_cs_track_check(struct radeon_cs_parser *p)
 263 {
 264         struct r600_cs_track *track = p->track;
 265         u32 tmp;
 266         int r, i;
 267         volatile u32 *ib = p->ib->ptr;
 268
 269         /* on legacy kernel we don't perform advanced check */
 270         if (p->rdev == NULL)
 271                 return 0;
 272         /* we don't support out buffer yet */
 273         if (track->vgt_strmout_en || track->vgt_strmout_buffer_en) {
 274                 dev_warn(p->dev, "this kernel doesn't support SMX output buffer\n");
 275                 return -EINVAL;
 276         }
 277         /* check that we have a cb for each enabled target, we don't check
 278          * shader_mask because it seems mesa isn't always setting it :(
 279          */
 280         tmp = track->cb_target_mask;
 281         for (i = 0; i < 8; i++) {
 282                 if ((tmp >> (i * 4)) & 0xF) {
 283                         /* at least one component is enabled */
 284                         if (track->cb_color_bo[i] == NULL) {
 285                                 dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
 286                                         __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
 287                                 return -EINVAL;
 288                         }
 289                         /* perform rewrite of CB_COLOR[0-7]_SIZE */
 290                         r = r600_cs_track_validate_cb(p, i);
 291                         if (r)
 292                                 return r;
 293                 }
 294         }
 295         /* Check depth buffer */
 296         if (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
 297                 G_028800_Z_ENABLE(track->db_depth_control)) {
 298                 u32 nviews, bpe, ntiles, pitch, pitch_align, height, size;
 299                 if (track->db_bo == NULL) {
 300                         dev_warn(p->dev, "z/stencil with no depth buffer\n");
 301                         return -EINVAL;
 302                 }
 303                 if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
 304                         dev_warn(p->dev, "this kernel doesn't support z/stencil htile\n");
 305                         return -EINVAL;
 306                 }
 307                 switch (G_028010_FORMAT(track->db_depth_info)) {
 308                 case V_028010_DEPTH_16:
 309                         bpe = 2;
 310                         break;
 311                 case V_028010_DEPTH_X8_24:
 312                 case V_028010_DEPTH_8_24:
 313                 case V_028010_DEPTH_X8_24_FLOAT:
 314                 case V_028010_DEPTH_8_24_FLOAT:
 315                 case V_028010_DEPTH_32_FLOAT:
 316                         bpe = 4;
 317                         break;
 318                 case V_028010_DEPTH_X24_8_32_FLOAT:
 319                         bpe = 8;
 320                         break;
 321                 default:
 322                         dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
 323                         return -EINVAL;
 324                 }
 325                 if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
 326                         if (!track->db_depth_size_idx) {
 327                                 dev_warn(p->dev, "z/stencil buffer size not set\n");
 328                                 return -EINVAL;
 329                         }
 330                         printk_once(KERN_WARNING "You have old & broken userspace please consider updating mesa\n");
 331                         tmp = radeon_bo_size(track->db_bo) - track->db_offset;
 332                         tmp = (tmp / bpe) >> 6;
 333                         if (!tmp) {
 334                                 dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
 335                                                 track->db_depth_size, bpe, track->db_offset,
 336                                                 radeon_bo_size(track->db_bo));
 337                                 return -EINVAL;
 338                         }
 339                         ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
 340                 } else {
 341                         size = radeon_bo_size(track->db_bo);
 342                         pitch = G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1;
 343                         height = size / (pitch * 8 * bpe);
 344                         height &= ~0x7;
 345                         if (!height)
 346                                 height = 8;
 347
 348                         switch (G_028010_ARRAY_MODE(track->db_depth_info)) {
 349                         case V_028010_ARRAY_1D_TILED_THIN1:
 350                                 pitch_align = (max((u32)8, (u32)(track->group_size / (8 * bpe))) / 8);
 351                                 if (!IS_ALIGNED(pitch, pitch_align)) {
 352                                         dev_warn(p->dev, "%s:%d db pitch (%d) invalid\n",
 353                                                  __func__, __LINE__, pitch);
 354                                         return -EINVAL;
 355                                 }
 356                                 if (!IS_ALIGNED(height, 8)) {
 357                                         dev_warn(p->dev, "%s:%d db height (%d) invalid\n",
 358                                                  __func__, __LINE__, height);
 359                                         return -EINVAL;
 360                                 }
 361                                 break;
 362                         case V_028010_ARRAY_2D_TILED_THIN1:
 363                                 pitch_align = max((u32)track->nbanks,
 364                                                   (u32)(((track->group_size / 8) / bpe) * track->nbanks));
 365                                 if (!IS_ALIGNED(pitch, pitch_align)) {
 366                                         dev_warn(p->dev, "%s:%d db pitch (%d) invalid\n",
 367                                                  __func__, __LINE__, pitch);
 368                                         return -EINVAL;
 369                                 }
 370                                 if ((height / 8) & (track->nbanks - 1)) {
 371                                         dev_warn(p->dev, "%s:%d db height (%d) invalid\n",
 372                                                  __func__, __LINE__, height);
 373                                         return -EINVAL;
 374                                 }
 375                                 break;
 376                         default:
 377                                 dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
 378                                          G_028010_ARRAY_MODE(track->db_depth_info),
 379                                          track->db_depth_info);
 380                                 return -EINVAL;
 381                         }
 382                         if (!IS_ALIGNED(track->db_offset, track->group_size)) {
 383                                 dev_warn(p->dev, "%s offset[%d] %d not aligned\n", __func__, i, track->db_offset);
 384                                 return -EINVAL;
 385                         }
 386                         ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
 387                         nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
 388                         tmp = ntiles * bpe * 64 * nviews;
 389                         if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
 390                                 dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %d -> %d have %ld)\n",
 391                                                 track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
 392                                                 radeon_bo_size(track->db_bo));
 393                                 return -EINVAL;
 394                         }
 395                 }
 396         }
 397         return 0;
 398 }
 399
 400 /**
 401  * r600_cs_packet_parse() - parse cp packet and point ib index to next packet
 402  * @parser:     parser structure holding parsing context.
 403  * @pkt:        where to store packet informations
 404  *
 405  * Assume that chunk_ib_index is properly set. Will return -EINVAL
 406  * if packet is bigger than remaining ib size. or if packets is unknown.
 407  **/
 408 int r600_cs_packet_parse(struct radeon_cs_parser *p,
 409                         struct radeon_cs_packet *pkt,
 410                         unsigned idx)
 411 {
 412         struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
 413         uint32_t header;
 414
 415         if (idx >= ib_chunk->length_dw) {
 416                 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
 417                           idx, ib_chunk->length_dw);
 418                 return -EINVAL;
 419         }
 420         header = radeon_get_ib_value(p, idx);
 421         pkt->idx = idx;
 422         pkt->type = CP_PACKET_GET_TYPE(header);
 423         pkt->count = CP_PACKET_GET_COUNT(header);
 424         pkt->one_reg_wr = 0;
 425         switch (pkt->type) {
 426         case PACKET_TYPE0:
 427                 pkt->reg = CP_PACKET0_GET_REG(header);
 428                 break;
 429         case PACKET_TYPE3:
 430                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 431                 break;
 432         case PACKET_TYPE2:
 433                 pkt->count = -1;
 434                 break;
 435         default:
 436                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 437                 return -EINVAL;
 438         }
 439         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 440                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 441                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 442                 return -EINVAL;
 443         }
 444         return 0;
 445 }
 446
 447 /**
 448  * r600_cs_packet_next_reloc_mm() - parse next packet which should be reloc packet3
 449  * @parser:             parser structure holding parsing context.
 450  * @data:               pointer to relocation data
 451  * @offset_start:       starting offset
 452  * @offset_mask:        offset mask (to align start offset on)
 453  * @reloc:              reloc informations
 454  *
 455  * Check next packet is relocation packet3, do bo validation and compute
 456  * GPU offset using the provided start.
 457  **/
 458 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
 459                                         struct radeon_cs_reloc **cs_reloc)
 460 {
 461         struct radeon_cs_chunk *relocs_chunk;
 462         struct radeon_cs_packet p3reloc;
 463         unsigned idx;
 464         int r;
 465
 466         if (p->chunk_relocs_idx == -1) {
 467                 DRM_ERROR("No relocation chunk !\n");
 468                 return -EINVAL;
 469         }
 470         *cs_reloc = NULL;
 471         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 472         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 473         if (r) {
 474                 return r;
 475         }
 476         p->idx += p3reloc.count + 2;
 477         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 478                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 479                           p3reloc.idx);
 480                 return -EINVAL;
 481         }
 482         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
 483         if (idx >= relocs_chunk->length_dw) {
 484                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 485                           idx, relocs_chunk->length_dw);
 486                 return -EINVAL;
 487         }
 488         /* FIXME: we assume reloc size is 4 dwords */
 489         *cs_reloc = p->relocs_ptr[(idx / 4)];
 490         return 0;
 491 }
 492
 493 /**
 494  * r600_cs_packet_next_reloc_nomm() - parse next packet which should be reloc packet3
 495  * @parser:             parser structure holding parsing context.
 496  * @data:               pointer to relocation data
 497  * @offset_start:       starting offset
 498  * @offset_mask:        offset mask (to align start offset on)
 499  * @reloc:              reloc informations
 500  *
 501  * Check next packet is relocation packet3, do bo validation and compute
 502  * GPU offset using the provided start.
 503  **/
 504 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
 505                                         struct radeon_cs_reloc **cs_reloc)
 506 {
 507         struct radeon_cs_chunk *relocs_chunk;
 508         struct radeon_cs_packet p3reloc;
 509         unsigned idx;
 510         int r;
 511
 512         if (p->chunk_relocs_idx == -1) {
 513                 DRM_ERROR("No relocation chunk !\n");
 514                 return -EINVAL;
 515         }
 516         *cs_reloc = NULL;
 517         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 518         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 519         if (r) {
 520                 return r;
 521         }
 522         p->idx += p3reloc.count + 2;
 523         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 524                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 525                           p3reloc.idx);
 526                 return -EINVAL;
 527         }
 528         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
 529         if (idx >= relocs_chunk->length_dw) {
 530                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 531                           idx, relocs_chunk->length_dw);
 532                 return -EINVAL;
 533         }
 534         *cs_reloc = p->relocs;
 535         (*cs_reloc)->lobj.gpu_offset = (u64)relocs_chunk->kdata[idx + 3] << 32;
 536         (*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
 537         return 0;
 538 }
 539
 540 /**
 541  * r600_cs_packet_next_is_pkt3_nop() - test if next packet is packet3 nop for reloc
 542  * @parser:             parser structure holding parsing context.
 543  *
 544  * Check next packet is relocation packet3, do bo validation and compute
 545  * GPU offset using the provided start.
 546  **/
 547 static inline int r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
 548 {
 549         struct radeon_cs_packet p3reloc;
 550         int r;
 551
 552         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 553         if (r) {
 554                 return 0;
 555         }
 556         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 557                 return 0;
 558         }
 559         return 1;
 560 }
 561
 562 /**
 563  * r600_cs_packet_next_vline() - parse userspace VLINE packet
 564  * @parser:             parser structure holding parsing context.
 565  *
 566  * Userspace sends a special sequence for VLINE waits.
 567  * PACKET0 - VLINE_START_END + value
 568  * PACKET3 - WAIT_REG_MEM poll vline status reg
 569  * RELOC (P3) - crtc_id in reloc.
 570  *
 571  * This function parses this and relocates the VLINE START END
 572  * and WAIT_REG_MEM packets to the correct crtc.
 573  * It also detects a switched off crtc and nulls out the
 574  * wait in that case.
 575  */
 576 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
 577 {
 578         struct drm_mode_object *obj;
 579         struct drm_crtc *crtc;
 580         struct radeon_crtc *radeon_crtc;
 581         struct radeon_cs_packet p3reloc, wait_reg_mem;
 582         int crtc_id;
 583         int r;
 584         uint32_t header, h_idx, reg, wait_reg_mem_info;
 585         volatile uint32_t *ib;
 586
 587         ib = p->ib->ptr;
 588
 589         /* parse the WAIT_REG_MEM */
 590         r = r600_cs_packet_parse(p, &wait_reg_mem, p->idx);
 591         if (r)
 592                 return r;
 593
 594         /* check its a WAIT_REG_MEM */
 595         if (wait_reg_mem.type != PACKET_TYPE3 ||
 596             wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
 597                 DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
 598                 r = -EINVAL;
 599                 return r;
 600         }
 601
 602         wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
 603         /* bit 4 is reg (0) or mem (1) */
 604         if (wait_reg_mem_info & 0x10) {
 605                 DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
 606                 r = -EINVAL;
 607                 return r;
 608         }
 609         /* waiting for value to be equal */
 610         if ((wait_reg_mem_info & 0x7) != 0x3) {
 611                 DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
 612                 r = -EINVAL;
 613                 return r;
 614         }
 615         if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != AVIVO_D1MODE_VLINE_STATUS) {
 616                 DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
 617                 r = -EINVAL;
 618                 return r;
 619         }
 620
 621         if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != AVIVO_D1MODE_VLINE_STAT) {
 622                 DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
 623                 r = -EINVAL;
 624                 return r;
 625         }
 626
 627         /* jump over the NOP */
 628         r = r600_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
 629         if (r)
 630                 return r;
 631
 632         h_idx = p->idx - 2;
 633         p->idx += wait_reg_mem.count + 2;
 634         p->idx += p3reloc.count + 2;
 635
 636         header = radeon_get_ib_value(p, h_idx);
 637         crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
 638         reg = CP_PACKET0_GET_REG(header);
 639         mutex_lock(&p->rdev->ddev->mode_config.mutex);
 640         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 641         if (!obj) {
 642                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 643                 r = -EINVAL;
 644                 goto out;
 645         }
 646         crtc = obj_to_crtc(obj);
 647         radeon_crtc = to_radeon_crtc(crtc);
 648         crtc_id = radeon_crtc->crtc_id;
 649
 650         if (!crtc->enabled) {
 651                 /* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
 652                 ib[h_idx + 2] = PACKET2(0);
 653                 ib[h_idx + 3] = PACKET2(0);
 654                 ib[h_idx + 4] = PACKET2(0);
 655                 ib[h_idx + 5] = PACKET2(0);
 656                 ib[h_idx + 6] = PACKET2(0);
 657                 ib[h_idx + 7] = PACKET2(0);
 658                 ib[h_idx + 8] = PACKET2(0);
 659         } else if (crtc_id == 1) {
 660                 switch (reg) {
 661                 case AVIVO_D1MODE_VLINE_START_END:
 662                         header &= ~R600_CP_PACKET0_REG_MASK;
 663                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
 664                         break;
 665                 default:
 666                         DRM_ERROR("unknown crtc reloc\n");
 667                         r = -EINVAL;
 668                         goto out;
 669                 }
 670                 ib[h_idx] = header;
 671                 ib[h_idx + 4] = AVIVO_D2MODE_VLINE_STATUS >> 2;
 672         }
 673 out:
 674         mutex_unlock(&p->rdev->ddev->mode_config.mutex);
 675         return r;
 676 }
 677
 678 static int r600_packet0_check(struct radeon_cs_parser *p,
 679                                 struct radeon_cs_packet *pkt,
 680                                 unsigned idx, unsigned reg)
 681 {
 682         int r;
 683
 684         switch (reg) {
 685         case AVIVO_D1MODE_VLINE_START_END:
 686                 r = r600_cs_packet_parse_vline(p);
 687                 if (r) {
 688                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 689                                         idx, reg);
 690                         return r;
 691                 }
 692                 break;
 693         default:
 694                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
 695                        reg, idx);
 696                 return -EINVAL;
 697         }
 698         return 0;
 699 }
 700
 701 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
 702                                 struct radeon_cs_packet *pkt)
 703 {
 704         unsigned reg, i;
 705         unsigned idx;
 706         int r;
 707
 708         idx = pkt->idx + 1;
 709         reg = pkt->reg;
 710         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 711                 r = r600_packet0_check(p, pkt, idx, reg);
 712                 if (r) {
 713                         return r;
 714                 }
 715         }
 716         return 0;
 717 }
 718
 719 /**
 720  * r600_cs_check_reg() - check if register is authorized or not
 721  * @parser: parser structure holding parsing context
 722  * @reg: register we are testing
 723  * @idx: index into the cs buffer
 724  *
 725  * This function will test against r600_reg_safe_bm and return 0
 726  * if register is safe. If register is not flag as safe this function
 727  * will test it against a list of register needind special handling.
 728  */
 729 static inline int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
 730 {
 731         struct r600_cs_track *track = (struct r600_cs_track *)p->track;
 732         struct radeon_cs_reloc *reloc;
 733         u32 last_reg = ARRAY_SIZE(r600_reg_safe_bm);
 734         u32 m, i, tmp, *ib;
 735         int r;
 736
 737         i = (reg >> 7);
 738         if (i > last_reg) {
 739                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
 740                 return -EINVAL;
 741         }
 742         m = 1 << ((reg >> 2) & 31);
 743         if (!(r600_reg_safe_bm[i] & m))
 744                 return 0;
 745         ib = p->ib->ptr;
 746         switch (reg) {
 747         /* force following reg to 0 in an attemp to disable out buffer
 748          * which will need us to better understand how it works to perform
 749          * security check on it (Jerome)
 750          */
 751         case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
 752         case R_008C44_SQ_ESGS_RING_SIZE:
 753         case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
 754         case R_008C54_SQ_ESTMP_RING_SIZE:
 755         case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
 756         case R_008C74_SQ_FBUF_RING_SIZE:
 757         case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
 758         case R_008C5C_SQ_GSTMP_RING_SIZE:
 759         case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
 760         case R_008C4C_SQ_GSVS_RING_SIZE:
 761         case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
 762         case R_008C6C_SQ_PSTMP_RING_SIZE:
 763         case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
 764         case R_008C7C_SQ_REDUC_RING_SIZE:
 765         case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
 766         case R_008C64_SQ_VSTMP_RING_SIZE:
 767         case R_0288C8_SQ_GS_VERT_ITEMSIZE:
 768                 /* get value to populate the IB don't remove */
 769                 tmp =radeon_get_ib_value(p, idx);
 770                 ib[idx] = 0;
 771                 break;
 772         case SQ_CONFIG:
 773                 track->sq_config = radeon_get_ib_value(p, idx);
 774                 break;
 775         case R_028800_DB_DEPTH_CONTROL:
 776                 track->db_depth_control = radeon_get_ib_value(p, idx);
 777                 break;
 778         case R_028010_DB_DEPTH_INFO:
 779                 if (r600_cs_packet_next_is_pkt3_nop(p)) {
 780                         r = r600_cs_packet_next_reloc(p, &reloc);
 781                         if (r) {
 782                                 dev_warn(p->dev, "bad SET_CONTEXT_REG "
 783                                          "0x%04X\n", reg);
 784                                 return -EINVAL;
 785                         }
 786                         track->db_depth_info = radeon_get_ib_value(p, idx);
 787                         ib[idx] &= C_028010_ARRAY_MODE;
 788                         track->db_depth_info &= C_028010_ARRAY_MODE;
 789                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
 790                                 ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
 791                                 track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
 792                         } else {
 793                                 ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
 794                                 track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
 795                         }
 796                 } else
 797                         track->db_depth_info = radeon_get_ib_value(p, idx);
 798                 break;
 799         case R_028004_DB_DEPTH_VIEW:
 800                 track->db_depth_view = radeon_get_ib_value(p, idx);
 801                 break;
 802         case R_028000_DB_DEPTH_SIZE:
 803                 track->db_depth_size = radeon_get_ib_value(p, idx);
 804                 track->db_depth_size_idx = idx;
 805                 break;
 806         case R_028AB0_VGT_STRMOUT_EN:
 807                 track->vgt_strmout_en = radeon_get_ib_value(p, idx);
 808                 break;
 809         case R_028B20_VGT_STRMOUT_BUFFER_EN:
 810                 track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
 811                 break;
 812         case R_028238_CB_TARGET_MASK:
 813                 track->cb_target_mask = radeon_get_ib_value(p, idx);
 814                 break;
 815         case R_02823C_CB_SHADER_MASK:
 816                 track->cb_shader_mask = radeon_get_ib_value(p, idx);
 817                 break;
 818         case R_028C04_PA_SC_AA_CONFIG:
 819                 tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
 820                 track->nsamples = 1 << tmp;
 821                 break;
 822         case R_0280A0_CB_COLOR0_INFO:
 823         case R_0280A4_CB_COLOR1_INFO:
 824         case R_0280A8_CB_COLOR2_INFO:
 825         case R_0280AC_CB_COLOR3_INFO:
 826         case R_0280B0_CB_COLOR4_INFO:
 827         case R_0280B4_CB_COLOR5_INFO:
 828         case R_0280B8_CB_COLOR6_INFO:
 829         case R_0280BC_CB_COLOR7_INFO:
 830                 if (r600_cs_packet_next_is_pkt3_nop(p)) {
 831                         r = r600_cs_packet_next_reloc(p, &reloc);
 832                         if (r) {
 833                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
 834                                 return -EINVAL;
 835                         }
 836                         tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
 837                         track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
 838                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
 839                                 ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
 840                                 track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
 841                         } else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
 842                                 ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
 843                                 track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
 844                         }
 845                 } else {
 846                         tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
 847                         track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
 848                 }
 849                 break;
 850         case R_028060_CB_COLOR0_SIZE:
 851         case R_028064_CB_COLOR1_SIZE:
 852         case R_028068_CB_COLOR2_SIZE:
 853         case R_02806C_CB_COLOR3_SIZE:
 854         case R_028070_CB_COLOR4_SIZE:
 855         case R_028074_CB_COLOR5_SIZE:
 856         case R_028078_CB_COLOR6_SIZE:
 857         case R_02807C_CB_COLOR7_SIZE:
 858                 tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
 859                 track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
 860                 track->cb_color_size_idx[tmp] = idx;
 861                 break;
 862                 /* This register were added late, there is userspace
 863                  * which does provide relocation for those but set
 864                  * 0 offset. In order to avoid breaking old userspace
 865                  * we detect this and set address to point to last
 866                  * CB_COLOR0_BASE, note that if userspace doesn't set
 867                  * CB_COLOR0_BASE before this register we will report
 868                  * error. Old userspace always set CB_COLOR0_BASE
 869                  * before any of this.
 870                  */
 871         case R_0280E0_CB_COLOR0_FRAG:
 872         case R_0280E4_CB_COLOR1_FRAG:
 873         case R_0280E8_CB_COLOR2_FRAG:
 874         case R_0280EC_CB_COLOR3_FRAG:
 875         case R_0280F0_CB_COLOR4_FRAG:
 876         case R_0280F4_CB_COLOR5_FRAG:
 877         case R_0280F8_CB_COLOR6_FRAG:
 878         case R_0280FC_CB_COLOR7_FRAG:
 879                 tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
 880                 if (!r600_cs_packet_next_is_pkt3_nop(p)) {
 881                         if (!track->cb_color_base_last[tmp]) {
 882                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
 883                                 return -EINVAL;
 884                         }
 885                         ib[idx] = track->cb_color_base_last[tmp];
 886                         printk_once(KERN_WARNING "You have old & broken userspace "
 887                                         "please consider updating mesa & xf86-video-ati\n");
 888                         track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
 889                 } else {
 890                         r = r600_cs_packet_next_reloc(p, &reloc);
 891                         if (r) {
 892                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
 893                                 return -EINVAL;
 894                         }
 895                         ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 896                         track->cb_color_frag_bo[tmp] = reloc->robj;
 897                 }
 898                 break;
 899         case R_0280C0_CB_COLOR0_TILE:
 900         case R_0280C4_CB_COLOR1_TILE:
 901         case R_0280C8_CB_COLOR2_TILE:
 902         case R_0280CC_CB_COLOR3_TILE:
 903         case R_0280D0_CB_COLOR4_TILE:
 904         case R_0280D4_CB_COLOR5_TILE:
 905         case R_0280D8_CB_COLOR6_TILE:
 906         case R_0280DC_CB_COLOR7_TILE:
 907                 tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
 908                 if (!r600_cs_packet_next_is_pkt3_nop(p)) {
 909                         if (!track->cb_color_base_last[tmp]) {
 910                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
 911                                 return -EINVAL;
 912                         }
 913                         ib[idx] = track->cb_color_base_last[tmp];
 914                         printk_once(KERN_WARNING "You have old & broken userspace "
 915                                         "please consider updating mesa & xf86-video-ati\n");
 916                         track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
 917                 } else {
 918                         r = r600_cs_packet_next_reloc(p, &reloc);
 919                         if (r) {
 920                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
 921                                 return -EINVAL;
 922                         }
 923                         ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 924                         track->cb_color_tile_bo[tmp] = reloc->robj;
 925                 }
 926                 break;
 927         case CB_COLOR0_BASE:
 928         case CB_COLOR1_BASE:
 929         case CB_COLOR2_BASE:
 930         case CB_COLOR3_BASE:
 931         case CB_COLOR4_BASE:
 932         case CB_COLOR5_BASE:
 933         case CB_COLOR6_BASE:
 934         case CB_COLOR7_BASE:
 935                 r = r600_cs_packet_next_reloc(p, &reloc);
 936                 if (r) {
 937                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
 938                                         "0x%04X\n", reg);
 939                         return -EINVAL;
 940                 }
 941                 tmp = (reg - CB_COLOR0_BASE) / 4;
 942                 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
 943                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 944                 track->cb_color_base_last[tmp] = ib[idx];
 945                 track->cb_color_bo[tmp] = reloc->robj;
 946                 break;
 947         case DB_DEPTH_BASE:
 948                 r = r600_cs_packet_next_reloc(p, &reloc);
 949                 if (r) {
 950                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
 951                                         "0x%04X\n", reg);
 952                         return -EINVAL;
 953                 }
 954                 track->db_offset = radeon_get_ib_value(p, idx);
 955                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 956                 track->db_bo = reloc->robj;
 957                 break;
 958         case DB_HTILE_DATA_BASE:
 959         case SQ_PGM_START_FS:
 960         case SQ_PGM_START_ES:
 961         case SQ_PGM_START_VS:
 962         case SQ_PGM_START_GS:
 963         case SQ_PGM_START_PS:
 964         case SQ_ALU_CONST_CACHE_GS_0:
 965         case SQ_ALU_CONST_CACHE_GS_1:
 966         case SQ_ALU_CONST_CACHE_GS_2:
 967         case SQ_ALU_CONST_CACHE_GS_3:
 968         case SQ_ALU_CONST_CACHE_GS_4:
 969         case SQ_ALU_CONST_CACHE_GS_5:
 970         case SQ_ALU_CONST_CACHE_GS_6:
 971         case SQ_ALU_CONST_CACHE_GS_7:
 972         case SQ_ALU_CONST_CACHE_GS_8:
 973         case SQ_ALU_CONST_CACHE_GS_9:
 974         case SQ_ALU_CONST_CACHE_GS_10:
 975         case SQ_ALU_CONST_CACHE_GS_11:
 976         case SQ_ALU_CONST_CACHE_GS_12:
 977         case SQ_ALU_CONST_CACHE_GS_13:
 978         case SQ_ALU_CONST_CACHE_GS_14:
 979         case SQ_ALU_CONST_CACHE_GS_15:
 980         case SQ_ALU_CONST_CACHE_PS_0:
 981         case SQ_ALU_CONST_CACHE_PS_1:
 982         case SQ_ALU_CONST_CACHE_PS_2:
 983         case SQ_ALU_CONST_CACHE_PS_3:
 984         case SQ_ALU_CONST_CACHE_PS_4:
 985         case SQ_ALU_CONST_CACHE_PS_5:
 986         case SQ_ALU_CONST_CACHE_PS_6:
 987         case SQ_ALU_CONST_CACHE_PS_7:
 988         case SQ_ALU_CONST_CACHE_PS_8:
 989         case SQ_ALU_CONST_CACHE_PS_9:
 990         case SQ_ALU_CONST_CACHE_PS_10:
 991         case SQ_ALU_CONST_CACHE_PS_11:
 992         case SQ_ALU_CONST_CACHE_PS_12:
 993         case SQ_ALU_CONST_CACHE_PS_13:
 994         case SQ_ALU_CONST_CACHE_PS_14:
 995         case SQ_ALU_CONST_CACHE_PS_15:
 996         case SQ_ALU_CONST_CACHE_VS_0:
 997         case SQ_ALU_CONST_CACHE_VS_1:
 998         case SQ_ALU_CONST_CACHE_VS_2:
 999         case SQ_ALU_CONST_CACHE_VS_3:
1000         case SQ_ALU_CONST_CACHE_VS_4:
1001         case SQ_ALU_CONST_CACHE_VS_5:
1002         case SQ_ALU_CONST_CACHE_VS_6:
1003         case SQ_ALU_CONST_CACHE_VS_7:
1004         case SQ_ALU_CONST_CACHE_VS_8:
1005         case SQ_ALU_CONST_CACHE_VS_9:
1006         case SQ_ALU_CONST_CACHE_VS_10:
1007         case SQ_ALU_CONST_CACHE_VS_11:
1008         case SQ_ALU_CONST_CACHE_VS_12:
1009         case SQ_ALU_CONST_CACHE_VS_13:
1010         case SQ_ALU_CONST_CACHE_VS_14:
1011         case SQ_ALU_CONST_CACHE_VS_15:
1012                 r = r600_cs_packet_next_reloc(p, &reloc);
1013                 if (r) {
1014                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1015                                         "0x%04X\n", reg);
1016                         return -EINVAL;
1017                 }
1018                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1019                 break;
1020         default:
1021                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1022                 return -EINVAL;
1023         }
1024         return 0;
1025 }
1026
1027 static inline unsigned minify(unsigned size, unsigned levels)
1028 {
1029         size = size >> levels;
1030         if (size < 1)
1031                 size = 1;
1032         return size;
1033 }
1034
1035 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned nlevels,
1036                               unsigned w0, unsigned h0, unsigned d0, unsigned bpe,
1037                               unsigned pitch_align,
1038                               unsigned *l0_size, unsigned *mipmap_size)
1039 {
1040         unsigned offset, i, level, face;
1041         unsigned width, height, depth, rowstride, size;
1042
1043         w0 = minify(w0, 0);
1044         h0 = minify(h0, 0);
1045         d0 = minify(d0, 0);
1046         for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1047                 width = minify(w0, i);
1048                 height = minify(h0, i);
1049                 depth = minify(d0, i);
1050                 for(face = 0; face < nfaces; face++) {
1051                         rowstride = ALIGN((width * bpe), pitch_align);
1052                         size = height * rowstride * depth;
1053                         offset += size;
1054                         offset = (offset + 0x1f) & ~0x1f;
1055                 }
1056         }
1057         *l0_size = ALIGN((w0 * bpe), pitch_align) * h0 * d0;
1058         *mipmap_size = offset;
1059         if (!blevel)
1060                 *mipmap_size -= *l0_size;
1061         if (!nlevels)
1062                 *mipmap_size = *l0_size;
1063 }
1064
1065 /**
1066  * r600_check_texture_resource() - check if register is authorized or not
1067  * @p: parser structure holding parsing context
1068  * @idx: index into the cs buffer
1069  * @texture: texture's bo structure
1070  * @mipmap: mipmap's bo structure
1071  *
1072  * This function will check that the resource has valid field and that
1073  * the texture and mipmap bo object are big enough to cover this resource.
1074  */
1075 static inline int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1076                                               struct radeon_bo *texture,
1077                                               struct radeon_bo *mipmap,
1078                                               u32 tiling_flags)
1079 {
1080         struct r600_cs_track *track = p->track;
1081         u32 nfaces, nlevels, blevel, w0, h0, d0, bpe = 0;
1082         u32 word0, word1, l0_size, mipmap_size, pitch, pitch_align;
1083
1084         /* on legacy kernel we don't perform advanced check */
1085         if (p->rdev == NULL)
1086                 return 0;
1087
1088         word0 = radeon_get_ib_value(p, idx + 0);
1089         if (tiling_flags & RADEON_TILING_MACRO)
1090                 word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1091         else if (tiling_flags & RADEON_TILING_MICRO)
1092                 word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1093         word1 = radeon_get_ib_value(p, idx + 1);
1094         w0 = G_038000_TEX_WIDTH(word0) + 1;
1095         h0 = G_038004_TEX_HEIGHT(word1) + 1;
1096         d0 = G_038004_TEX_DEPTH(word1);
1097         nfaces = 1;
1098         switch (G_038000_DIM(word0)) {
1099         case V_038000_SQ_TEX_DIM_1D:
1100         case V_038000_SQ_TEX_DIM_2D:
1101         case V_038000_SQ_TEX_DIM_3D:
1102                 break;
1103         case V_038000_SQ_TEX_DIM_CUBEMAP:
1104                 nfaces = 6;
1105                 break;
1106         case V_038000_SQ_TEX_DIM_1D_ARRAY:
1107         case V_038000_SQ_TEX_DIM_2D_ARRAY:
1108         case V_038000_SQ_TEX_DIM_2D_MSAA:
1109         case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1110         default:
1111                 dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1112                 return -EINVAL;
1113         }
1114         if (r600_bpe_from_format(&bpe,  G_038004_DATA_FORMAT(word1))) {
1115                 dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1116                          __func__, __LINE__, G_038004_DATA_FORMAT(word1));
1117                 return -EINVAL;
1118         }
1119
1120         pitch = G_038000_PITCH(word0) + 1;
1121         switch (G_038000_TILE_MODE(word0)) {
1122         case V_038000_ARRAY_LINEAR_GENERAL:
1123                 pitch_align = 1;
1124                 /* XXX check height align */
1125                 break;
1126         case V_038000_ARRAY_LINEAR_ALIGNED:
1127                 pitch_align = max((u32)64, (u32)(track->group_size / bpe)) / 8;
1128                 if (!IS_ALIGNED(pitch, pitch_align)) {
1129                         dev_warn(p->dev, "%s:%d tex pitch (%d) invalid\n",
1130                                  __func__, __LINE__, pitch);
1131                         return -EINVAL;
1132                 }
1133                 /* XXX check height align */
1134                 break;
1135         case V_038000_ARRAY_1D_TILED_THIN1:
1136                 pitch_align = max((u32)8, (u32)(track->group_size / (8 * bpe))) / 8;
1137                 if (!IS_ALIGNED(pitch, pitch_align)) {
1138                         dev_warn(p->dev, "%s:%d tex pitch (%d) invalid\n",
1139                                  __func__, __LINE__, pitch);
1140                         return -EINVAL;
1141                 }
1142                 /* XXX check height align */
1143                 break;
1144         case V_038000_ARRAY_2D_TILED_THIN1:
1145                 pitch_align = max((u32)track->nbanks,
1146                                   (u32)(((track->group_size / 8) / bpe) * track->nbanks));
1147                 if (!IS_ALIGNED(pitch, pitch_align)) {
1148                         dev_warn(p->dev, "%s:%d tex pitch (%d) invalid\n",
1149                                 __func__, __LINE__, pitch);
1150                         return -EINVAL;
1151                 }
1152                 /* XXX check height align */
1153                 break;
1154         default:
1155                 dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
1156                          G_038000_TILE_MODE(word0), word0);
1157                 return -EINVAL;
1158         }
1159         /* XXX check offset align */
1160
1161         word0 = radeon_get_ib_value(p, idx + 4);
1162         word1 = radeon_get_ib_value(p, idx + 5);
1163         blevel = G_038010_BASE_LEVEL(word0);
1164         nlevels = G_038014_LAST_LEVEL(word1);
1165         r600_texture_size(nfaces, blevel, nlevels, w0, h0, d0, bpe,
1166                           (pitch_align * bpe),
1167                           &l0_size, &mipmap_size);
1168         /* using get ib will give us the offset into the texture bo */
1169         word0 = radeon_get_ib_value(p, idx + 2);
1170         if ((l0_size + word0) > radeon_bo_size(texture)) {
1171                 dev_warn(p->dev, "texture bo too small (%d %d %d %d -> %d have %ld)\n",
1172                         w0, h0, bpe, word0, l0_size, radeon_bo_size(texture));
1173                 return -EINVAL;
1174         }
1175         /* using get ib will give us the offset into the mipmap bo */
1176         word0 = radeon_get_ib_value(p, idx + 3);
1177         if ((mipmap_size + word0) > radeon_bo_size(mipmap)) {
1178                 dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1179                         w0, h0, bpe, blevel, nlevels, word0, mipmap_size, radeon_bo_size(texture));
1180                 return -EINVAL;
1181         }
1182         return 0;
1183 }
1184
1185 static int r600_packet3_check(struct radeon_cs_parser *p,
1186                                 struct radeon_cs_packet *pkt)
1187 {
1188         struct radeon_cs_reloc *reloc;
1189         struct r600_cs_track *track;
1190         volatile u32 *ib;
1191         unsigned idx;
1192         unsigned i;
1193         unsigned start_reg, end_reg, reg;
1194         int r;
1195         u32 idx_value;
1196
1197         track = (struct r600_cs_track *)p->track;
1198         ib = p->ib->ptr;
1199         idx = pkt->idx + 1;
1200         idx_value = radeon_get_ib_value(p, idx);
1201
1202         switch (pkt->opcode) {
1203         case PACKET3_START_3D_CMDBUF:
1204                 if (p->family >= CHIP_RV770 || pkt->count) {
1205                         DRM_ERROR("bad START_3D\n");
1206                         return -EINVAL;
1207                 }
1208                 break;
1209         case PACKET3_CONTEXT_CONTROL:
1210                 if (pkt->count != 1) {
1211                         DRM_ERROR("bad CONTEXT_CONTROL\n");
1212                         return -EINVAL;
1213                 }
1214                 break;
1215         case PACKET3_INDEX_TYPE:
1216         case PACKET3_NUM_INSTANCES:
1217                 if (pkt->count) {
1218                         DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1219                         return -EINVAL;
1220                 }
1221                 break;
1222         case PACKET3_DRAW_INDEX:
1223                 if (pkt->count != 3) {
1224                         DRM_ERROR("bad DRAW_INDEX\n");
1225                         return -EINVAL;
1226                 }
1227                 r = r600_cs_packet_next_reloc(p, &reloc);
1228                 if (r) {
1229                         DRM_ERROR("bad DRAW_INDEX\n");
1230                         return -EINVAL;
1231                 }
1232                 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1233                 ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1234                 r = r600_cs_track_check(p);
1235                 if (r) {
1236                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1237                         return r;
1238                 }
1239                 break;
1240         case PACKET3_DRAW_INDEX_AUTO:
1241                 if (pkt->count != 1) {
1242                         DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1243                         return -EINVAL;
1244                 }
1245                 r = r600_cs_track_check(p);
1246                 if (r) {
1247                         dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1248                         return r;
1249                 }
1250                 break;
1251         case PACKET3_DRAW_INDEX_IMMD_BE:
1252         case PACKET3_DRAW_INDEX_IMMD:
1253                 if (pkt->count < 2) {
1254                         DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1255                         return -EINVAL;
1256                 }
1257                 r = r600_cs_track_check(p);
1258                 if (r) {
1259                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1260                         return r;
1261                 }
1262                 break;
1263         case PACKET3_WAIT_REG_MEM:
1264                 if (pkt->count != 5) {
1265                         DRM_ERROR("bad WAIT_REG_MEM\n");
1266                         return -EINVAL;
1267                 }
1268                 /* bit 4 is reg (0) or mem (1) */
1269                 if (idx_value & 0x10) {
1270                         r = r600_cs_packet_next_reloc(p, &reloc);
1271                         if (r) {
1272                                 DRM_ERROR("bad WAIT_REG_MEM\n");
1273                                 return -EINVAL;
1274                         }
1275                         ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1276                         ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1277                 }
1278                 break;
1279         case PACKET3_SURFACE_SYNC:
1280                 if (pkt->count != 3) {
1281                         DRM_ERROR("bad SURFACE_SYNC\n");
1282                         return -EINVAL;
1283                 }
1284                 /* 0xffffffff/0x0 is flush all cache flag */
1285                 if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1286                     radeon_get_ib_value(p, idx + 2) != 0) {
1287                         r = r600_cs_packet_next_reloc(p, &reloc);
1288                         if (r) {
1289                                 DRM_ERROR("bad SURFACE_SYNC\n");
1290                                 return -EINVAL;
1291                         }
1292                         ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1293                 }
1294                 break;
1295         case PACKET3_EVENT_WRITE:
1296                 if (pkt->count != 2 && pkt->count != 0) {
1297                         DRM_ERROR("bad EVENT_WRITE\n");
1298                         return -EINVAL;
1299                 }
1300                 if (pkt->count) {
1301                         r = r600_cs_packet_next_reloc(p, &reloc);
1302                         if (r) {
1303                                 DRM_ERROR("bad EVENT_WRITE\n");
1304                                 return -EINVAL;
1305                         }
1306                         ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1307                         ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1308                 }
1309                 break;
1310         case PACKET3_EVENT_WRITE_EOP:
1311                 if (pkt->count != 4) {
1312                         DRM_ERROR("bad EVENT_WRITE_EOP\n");
1313                         return -EINVAL;
1314                 }
1315                 r = r600_cs_packet_next_reloc(p, &reloc);
1316                 if (r) {
1317                         DRM_ERROR("bad EVENT_WRITE\n");
1318                         return -EINVAL;
1319                 }
1320                 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1321                 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1322                 break;
1323         case PACKET3_SET_CONFIG_REG:
1324                 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1325                 end_reg = 4 * pkt->count + start_reg - 4;
1326                 if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1327                     (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1328                     (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1329                         DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1330                         return -EINVAL;
1331                 }
1332                 for (i = 0; i < pkt->count; i++) {
1333                         reg = start_reg + (4 * i);
1334                         r = r600_cs_check_reg(p, reg, idx+1+i);
1335                         if (r)
1336                                 return r;
1337                 }
1338                 break;
1339         case PACKET3_SET_CONTEXT_REG:
1340                 start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1341                 end_reg = 4 * pkt->count + start_reg - 4;
1342                 if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1343                     (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1344                     (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1345                         DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1346                         return -EINVAL;
1347                 }
1348                 for (i = 0; i < pkt->count; i++) {
1349                         reg = start_reg + (4 * i);
1350                         r = r600_cs_check_reg(p, reg, idx+1+i);
1351                         if (r)
1352                                 return r;
1353                 }
1354                 break;
1355         case PACKET3_SET_RESOURCE:
1356                 if (pkt->count % 7) {
1357                         DRM_ERROR("bad SET_RESOURCE\n");
1358                         return -EINVAL;
1359                 }
1360                 start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1361                 end_reg = 4 * pkt->count + start_reg - 4;
1362                 if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1363                     (start_reg >= PACKET3_SET_RESOURCE_END) ||
1364                     (end_reg >= PACKET3_SET_RESOURCE_END)) {
1365                         DRM_ERROR("bad SET_RESOURCE\n");
1366                         return -EINVAL;
1367                 }
1368                 for (i = 0; i < (pkt->count / 7); i++) {
1369                         struct radeon_bo *texture, *mipmap;
1370                         u32 size, offset;
1371
1372                         switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1373                         case SQ_TEX_VTX_VALID_TEXTURE:
1374                                 /* tex base */
1375                                 r = r600_cs_packet_next_reloc(p, &reloc);
1376                                 if (r) {
1377                                         DRM_ERROR("bad SET_RESOURCE\n");
1378                                         return -EINVAL;
1379                                 }
1380                                 ib[idx+1+(i*7)+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1381                                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1382                                         ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1383                                 else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1384                                         ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1385                                 texture = reloc->robj;
1386                                 /* tex mip base */
1387                                 r = r600_cs_packet_next_reloc(p, &reloc);
1388                                 if (r) {
1389                                         DRM_ERROR("bad SET_RESOURCE\n");
1390                                         return -EINVAL;
1391                                 }
1392                                 ib[idx+1+(i*7)+3] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1393                                 mipmap = reloc->robj;
1394                                 r = r600_check_texture_resource(p,  idx+(i*7)+1,
1395                                                                 texture, mipmap, reloc->lobj.tiling_flags);
1396                                 if (r)
1397                                         return r;
1398                                 break;
1399                         case SQ_TEX_VTX_VALID_BUFFER:
1400                                 /* vtx base */
1401                                 r = r600_cs_packet_next_reloc(p, &reloc);
1402                                 if (r) {
1403                                         DRM_ERROR("bad SET_RESOURCE\n");
1404                                         return -EINVAL;
1405                                 }
1406                                 offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
1407                                 size = radeon_get_ib_value(p, idx+1+(i*7)+1);
1408                                 if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
1409                                         /* force size to size of the buffer */
1410                                         dev_warn(p->dev, "vbo resource seems too big for the bo\n");
1411                                         ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj);
1412                                 }
1413                                 ib[idx+1+(i*7)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
1414                                 ib[idx+1+(i*7)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1415                                 break;
1416                         case SQ_TEX_VTX_INVALID_TEXTURE:
1417                         case SQ_TEX_VTX_INVALID_BUFFER:
1418                         default:
1419                                 DRM_ERROR("bad SET_RESOURCE\n");
1420                                 return -EINVAL;
1421                         }
1422                 }
1423                 break;
1424         case PACKET3_SET_ALU_CONST:
1425                 if (track->sq_config & DX9_CONSTS) {
1426                         start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
1427                         end_reg = 4 * pkt->count + start_reg - 4;
1428                         if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
1429                             (start_reg >= PACKET3_SET_ALU_CONST_END) ||
1430                             (end_reg >= PACKET3_SET_ALU_CONST_END)) {
1431                                 DRM_ERROR("bad SET_ALU_CONST\n");
1432                                 return -EINVAL;
1433                         }
1434                 }
1435                 break;
1436         case PACKET3_SET_BOOL_CONST:
1437                 start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
1438                 end_reg = 4 * pkt->count + start_reg - 4;
1439                 if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
1440                     (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
1441                     (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
1442                         DRM_ERROR("bad SET_BOOL_CONST\n");
1443                         return -EINVAL;
1444                 }
1445                 break;
1446         case PACKET3_SET_LOOP_CONST:
1447                 start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
1448                 end_reg = 4 * pkt->count + start_reg - 4;
1449                 if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
1450                     (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
1451                     (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
1452                         DRM_ERROR("bad SET_LOOP_CONST\n");
1453                         return -EINVAL;
1454                 }
1455                 break;
1456         case PACKET3_SET_CTL_CONST:
1457                 start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
1458                 end_reg = 4 * pkt->count + start_reg - 4;
1459                 if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
1460                     (start_reg >= PACKET3_SET_CTL_CONST_END) ||
1461                     (end_reg >= PACKET3_SET_CTL_CONST_END)) {
1462                         DRM_ERROR("bad SET_CTL_CONST\n");
1463                         return -EINVAL;
1464                 }
1465                 break;
1466         case PACKET3_SET_SAMPLER:
1467                 if (pkt->count % 3) {
1468                         DRM_ERROR("bad SET_SAMPLER\n");
1469                         return -EINVAL;
1470                 }
1471                 start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
1472                 end_reg = 4 * pkt->count + start_reg - 4;
1473                 if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
1474                     (start_reg >= PACKET3_SET_SAMPLER_END) ||
1475                     (end_reg >= PACKET3_SET_SAMPLER_END)) {
1476                         DRM_ERROR("bad SET_SAMPLER\n");
1477                         return -EINVAL;
1478                 }
1479                 break;
1480         case PACKET3_SURFACE_BASE_UPDATE:
1481                 if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
1482                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1483                         return -EINVAL;
1484                 }
1485                 if (pkt->count) {
1486                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1487                         return -EINVAL;
1488                 }
1489                 break;
1490         case PACKET3_NOP:
1491                 break;
1492         default:
1493                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1494                 return -EINVAL;
1495         }
1496         return 0;
1497 }
1498
1499 int r600_cs_parse(struct radeon_cs_parser *p)
1500 {
1501         struct radeon_cs_packet pkt;
1502         struct r600_cs_track *track;
1503         int r;
1504
1505         if (p->track == NULL) {
1506                 /* initialize tracker, we are in kms */
1507                 track = kzalloc(sizeof(*track), GFP_KERNEL);
1508                 if (track == NULL)
1509                         return -ENOMEM;
1510                 r600_cs_track_init(track);
1511                 if (p->rdev->family < CHIP_RV770) {
1512                         track->npipes = p->rdev->config.r600.tiling_npipes;
1513                         track->nbanks = p->rdev->config.r600.tiling_nbanks;
1514                         track->group_size = p->rdev->config.r600.tiling_group_size;
1515                 } else if (p->rdev->family <= CHIP_RV740) {
1516                         track->npipes = p->rdev->config.rv770.tiling_npipes;
1517                         track->nbanks = p->rdev->config.rv770.tiling_nbanks;
1518                         track->group_size = p->rdev->config.rv770.tiling_group_size;
1519                 }
1520                 p->track = track;
1521         }
1522         do {
1523                 r = r600_cs_packet_parse(p, &pkt, p->idx);
1524                 if (r) {
1525                         kfree(p->track);
1526                         p->track = NULL;
1527                         return r;
1528                 }
1529                 p->idx += pkt.count + 2;
1530                 switch (pkt.type) {
1531                 case PACKET_TYPE0:
1532                         r = r600_cs_parse_packet0(p, &pkt);
1533                         break;
1534                 case PACKET_TYPE2:
1535                         break;
1536                 case PACKET_TYPE3:
1537                         r = r600_packet3_check(p, &pkt);
1538                         break;
1539                 default:
1540                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1541                         kfree(p->track);
1542                         p->track = NULL;
1543                         return -EINVAL;
1544                 }
1545                 if (r) {
1546                         kfree(p->track);
1547                         p->track = NULL;
1548                         return r;
1549                 }
1550         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1551 #if 0
1552         for (r = 0; r < p->ib->length_dw; r++) {
1553                 printk(KERN_INFO "%05d  0x%08X\n", r, p->ib->ptr[r]);
1554                 mdelay(1);
1555         }
1556 #endif
1557         kfree(p->track);
1558         p->track = NULL;
1559         return 0;
1560 }
1561
1562 static int r600_cs_parser_relocs_legacy(struct radeon_cs_parser *p)
1563 {
1564         if (p->chunk_relocs_idx == -1) {
1565                 return 0;
1566         }
1567         p->relocs = kzalloc(sizeof(struct radeon_cs_reloc), GFP_KERNEL);
1568         if (p->relocs == NULL) {
1569                 return -ENOMEM;
1570         }
1571         return 0;
1572 }
1573
1574 /**
1575  * cs_parser_fini() - clean parser states
1576  * @parser:     parser structure holding parsing context.
1577  * @error:      error number
1578  *
1579  * If error is set than unvalidate buffer, otherwise just free memory
1580  * used by parsing context.
1581  **/
1582 static void r600_cs_parser_fini(struct radeon_cs_parser *parser, int error)
1583 {
1584         unsigned i;
1585
1586         kfree(parser->relocs);
1587         for (i = 0; i < parser->nchunks; i++) {
1588                 kfree(parser->chunks[i].kdata);
1589                 kfree(parser->chunks[i].kpage[0]);
1590                 kfree(parser->chunks[i].kpage[1]);
1591         }
1592         kfree(parser->chunks);
1593         kfree(parser->chunks_array);
1594 }
1595
1596 int r600_cs_legacy(struct drm_device *dev, void *data, struct drm_file *filp,
1597                         unsigned family, u32 *ib, int *l)
1598 {
1599         struct radeon_cs_parser parser;
1600         struct radeon_cs_chunk *ib_chunk;
1601         struct radeon_ib fake_ib;
1602         struct r600_cs_track *track;
1603         int r;
1604
1605         /* initialize tracker */
1606         track = kzalloc(sizeof(*track), GFP_KERNEL);
1607         if (track == NULL)
1608                 return -ENOMEM;
1609         r600_cs_track_init(track);
1610         r600_cs_legacy_get_tiling_conf(dev, &track->npipes, &track->nbanks, &track->group_size);
1611         /* initialize parser */
1612         memset(&parser, 0, sizeof(struct radeon_cs_parser));
1613         parser.filp = filp;
1614         parser.dev = &dev->pdev->dev;
1615         parser.rdev = NULL;
1616         parser.family = family;
1617         parser.ib = &fake_ib;
1618         parser.track = track;
1619         fake_ib.ptr = ib;
1620         r = radeon_cs_parser_init(&parser, data);
1621         if (r) {
1622                 DRM_ERROR("Failed to initialize parser !\n");
1623                 r600_cs_parser_fini(&parser, r);
1624                 return r;
1625         }
1626         r = r600_cs_parser_relocs_legacy(&parser);
1627         if (r) {
1628                 DRM_ERROR("Failed to parse relocation !\n");
1629                 r600_cs_parser_fini(&parser, r);
1630                 return r;
1631         }
1632         /* Copy the packet into the IB, the parser will read from the
1633          * input memory (cached) and write to the IB (which can be
1634          * uncached). */
1635         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
1636         parser.ib->length_dw = ib_chunk->length_dw;
1637         *l = parser.ib->length_dw;
1638         r = r600_cs_parse(&parser);
1639         if (r) {
1640                 DRM_ERROR("Invalid command stream !\n");
1641                 r600_cs_parser_fini(&parser, r);
1642                 return r;
1643         }
1644         r = radeon_cs_finish_pages(&parser);
1645         if (r) {
1646                 DRM_ERROR("Invalid command stream !\n");
1647                 r600_cs_parser_fini(&parser, r);
1648                 return r;
1649         }
1650         r600_cs_parser_fini(&parser, r);
1651         return r;
1652 }
1653
1654 void r600_cs_legacy_init(void)
1655 {
1656         r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;
1657 }