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 "drmP.h"
  29 #include "radeon.h"
  30 #include "r600d.h"
  31 #include "r600_reg_safe.h"
  32
  33 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
  34                                         struct radeon_cs_reloc **cs_reloc);
  35 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
  36                                         struct radeon_cs_reloc **cs_reloc);
  37 typedef int (*next_reloc_t)(struct radeon_cs_parser*, struct radeon_cs_reloc**);
  38 static next_reloc_t r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_mm;
  39 extern void r600_cs_legacy_get_tiling_conf(struct drm_device *dev, u32 *npipes, u32 *nbanks, u32 *group_size);
  40
  41
  42 struct r600_cs_track {
  43         /* configuration we miror so that we use same code btw kms/ums */
  44         u32                     group_size;
  45         u32                     nbanks;
  46         u32                     npipes;
  47         /* value we track */
  48         u32                     sq_config;
  49         u32                     nsamples;
  50         u32                     cb_color_base_last[8];
  51         struct radeon_bo        *cb_color_bo[8];
  52         u32                     cb_color_bo_offset[8];
  53         struct radeon_bo        *cb_color_frag_bo[8];
  54         struct radeon_bo        *cb_color_tile_bo[8];
  55         u32                     cb_color_info[8];
  56         u32                     cb_color_size_idx[8];
  57         u32                     cb_target_mask;
  58         u32                     cb_shader_mask;
  59         u32                     cb_color_size[8];
  60         u32                     vgt_strmout_en;
  61         u32                     vgt_strmout_buffer_en;
  62         u32                     db_depth_control;
  63         u32                     db_depth_info;
  64         u32                     db_depth_size_idx;
  65         u32                     db_depth_view;
  66         u32                     db_depth_size;
  67         u32                     db_offset;
  68         struct radeon_bo        *db_bo;
  69 };
  70
  71 static inline int r600_bpe_from_format(u32 *bpe, u32 format)
  72 {
  73         switch (format) {
  74         case V_038004_COLOR_8:
  75         case V_038004_COLOR_4_4:
  76         case V_038004_COLOR_3_3_2:
  77         case V_038004_FMT_1:
  78                 *bpe = 1;
  79                 break;
  80         case V_038004_COLOR_16:
  81         case V_038004_COLOR_16_FLOAT:
  82         case V_038004_COLOR_8_8:
  83         case V_038004_COLOR_5_6_5:
  84         case V_038004_COLOR_6_5_5:
  85         case V_038004_COLOR_1_5_5_5:
  86         case V_038004_COLOR_4_4_4_4:
  87         case V_038004_COLOR_5_5_5_1:
  88                 *bpe = 2;
  89                 break;
  90         case V_038004_FMT_8_8_8:
  91                 *bpe = 3;
  92                 break;
  93         case V_038004_COLOR_32:
  94         case V_038004_COLOR_32_FLOAT:
  95         case V_038004_COLOR_16_16:
  96         case V_038004_COLOR_16_16_FLOAT:
  97         case V_038004_COLOR_8_24:
  98         case V_038004_COLOR_8_24_FLOAT:
  99         case V_038004_COLOR_24_8:
 100         case V_038004_COLOR_24_8_FLOAT:
 101         case V_038004_COLOR_10_11_11:
 102         case V_038004_COLOR_10_11_11_FLOAT:
 103         case V_038004_COLOR_11_11_10:
 104         case V_038004_COLOR_11_11_10_FLOAT:
 105         case V_038004_COLOR_2_10_10_10:
 106         case V_038004_COLOR_8_8_8_8:
 107         case V_038004_COLOR_10_10_10_2:
 108         case V_038004_FMT_5_9_9_9_SHAREDEXP:
 109         case V_038004_FMT_32_AS_8:
 110         case V_038004_FMT_32_AS_8_8:
 111                 *bpe = 4;
 112                 break;
 113         case V_038004_COLOR_X24_8_32_FLOAT:
 114         case V_038004_COLOR_32_32:
 115         case V_038004_COLOR_32_32_FLOAT:
 116         case V_038004_COLOR_16_16_16_16:
 117         case V_038004_COLOR_16_16_16_16_FLOAT:
 118                 *bpe = 8;
 119                 break;
 120         case V_038004_FMT_16_16_16:
 121         case V_038004_FMT_16_16_16_FLOAT:
 122                 *bpe = 6;
 123                 break;
 124         case V_038004_FMT_32_32_32:
 125         case V_038004_FMT_32_32_32_FLOAT:
 126                 *bpe = 12;
 127                 break;
 128         case V_038004_COLOR_32_32_32_32:
 129         case V_038004_COLOR_32_32_32_32_FLOAT:
 130                 *bpe = 16;
 131                 break;
 132         case V_038004_FMT_GB_GR:
 133         case V_038004_FMT_BG_RG:
 134         case V_038004_COLOR_INVALID:
 135                 *bpe = 16;
 136                 return -EINVAL;
 137         }
 138         return 0;
 139 }
 140
 141 static void r600_cs_track_init(struct r600_cs_track *track)
 142 {
 143         int i;
 144
 145         /* assume DX9 mode */
 146         track->sq_config = DX9_CONSTS;
 147         for (i = 0; i < 8; i++) {
 148                 track->cb_color_base_last[i] = 0;
 149                 track->cb_color_size[i] = 0;
 150                 track->cb_color_size_idx[i] = 0;
 151                 track->cb_color_info[i] = 0;
 152                 track->cb_color_bo[i] = NULL;
 153                 track->cb_color_bo_offset[i] = 0xFFFFFFFF;
 154         }
 155         track->cb_target_mask = 0xFFFFFFFF;
 156         track->cb_shader_mask = 0xFFFFFFFF;
 157         track->db_bo = NULL;
 158         /* assume the biggest format and that htile is enabled */
 159         track->db_depth_info = 7 | (1 << 25);
 160         track->db_depth_view = 0xFFFFC000;
 161         track->db_depth_size = 0xFFFFFFFF;
 162         track->db_depth_size_idx = 0;
 163         track->db_depth_control = 0xFFFFFFFF;
 164 }
 165
 166 static inline int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
 167 {
 168         struct r600_cs_track *track = p->track;
 169         u32 bpe = 0, pitch, slice_tile_max, size, tmp, height;
 170         volatile u32 *ib = p->ib->ptr;
 171
 172         if (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
 173                 dev_warn(p->dev, "FMASK or CMASK buffer are not supported by this kernel\n");
 174                 return -EINVAL;
 175         }
 176         size = radeon_bo_size(track->cb_color_bo[i]);
 177         if (r600_bpe_from_format(&bpe, G_0280A0_FORMAT(track->cb_color_info[i]))) {
 178                 dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
 179                          __func__, __LINE__, G_0280A0_FORMAT(track->cb_color_info[i]),
 180                         i, track->cb_color_info[i]);
 181                 return -EINVAL;
 182         }
 183         pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) << 3;
 184         slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
 185         if (!pitch) {
 186                 dev_warn(p->dev, "%s:%d cb pitch (%d) for %d invalid (0x%08X)\n",
 187                         __func__, __LINE__, pitch, i, track->cb_color_size[i]);
 188                 return -EINVAL;
 189         }
 190         height = size / (pitch * bpe);
 191         if (height > 8192)
 192                 height = 8192;
 193         switch (G_0280A0_ARRAY_MODE(track->cb_color_info[i])) {
 194         case V_0280A0_ARRAY_LINEAR_GENERAL:
 195         case V_0280A0_ARRAY_LINEAR_ALIGNED:
 196                 if (pitch & 0x3f) {
 197                         dev_warn(p->dev, "%s:%d cb pitch (%d x %d = %d) invalid\n",
 198                                 __func__, __LINE__, pitch, bpe, pitch * bpe);
 199                         return -EINVAL;
 200                 }
 201                 if ((pitch * bpe) & (track->group_size - 1)) {
 202                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 203                                 __func__, __LINE__, pitch);
 204                         return -EINVAL;
 205                 }
 206                 break;
 207         case V_0280A0_ARRAY_1D_TILED_THIN1:
 208                 if ((pitch * 8 * bpe * track->nsamples) & (track->group_size - 1)) {
 209                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 210                                 __func__, __LINE__, pitch);
 211                         return -EINVAL;
 212                 }
 213                 height &= ~0x7;
 214                 if (!height)
 215                         height = 8;
 216                 break;
 217         case V_0280A0_ARRAY_2D_TILED_THIN1:
 218                 if (pitch & ((8 * track->nbanks) - 1)) {
 219                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 220                                 __func__, __LINE__, pitch);
 221                         return -EINVAL;
 222                 }
 223                 tmp = pitch * 8 * bpe * track->nsamples;
 224                 tmp = tmp / track->nbanks;
 225                 if (tmp & (track->group_size - 1)) {
 226                         dev_warn(p->dev, "%s:%d cb pitch (%d) invalid\n",
 227                                 __func__, __LINE__, pitch);
 228                         return -EINVAL;
 229                 }
 230                 height &= ~((16 * track->npipes) - 1);
 231                 if (!height)
 232                         height = 16 * track->npipes;
 233                 break;
 234         default:
 235                 dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
 236                         G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
 237                         track->cb_color_info[i]);
 238                 return -EINVAL;
 239         }
 240         /* check offset */
 241         tmp = height * pitch;
 242         if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
 243                 dev_warn(p->dev, "%s offset[%d] %d to big\n", __func__, i, track->cb_color_bo_offset[i]);
 244                 return -EINVAL;
 245         }
 246         /* limit max tile */
 247         tmp = (height * pitch) >> 6;
 248         if (tmp < slice_tile_max)
 249                 slice_tile_max = tmp;
 250         tmp = S_028060_PITCH_TILE_MAX((pitch >> 3) - 1) |
 251                 S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
 252         ib[track->cb_color_size_idx[i]] = tmp;
 253         return 0;
 254 }
 255
 256 static int r600_cs_track_check(struct radeon_cs_parser *p)
 257 {
 258         struct r600_cs_track *track = p->track;
 259         u32 tmp;
 260         int r, i;
 261         volatile u32 *ib = p->ib->ptr;
 262
 263         /* on legacy kernel we don't perform advanced check */
 264         if (p->rdev == NULL)
 265                 return 0;
 266         /* we don't support out buffer yet */
 267         if (track->vgt_strmout_en || track->vgt_strmout_buffer_en) {
 268                 dev_warn(p->dev, "this kernel doesn't support SMX output buffer\n");
 269                 return -EINVAL;
 270         }
 271         /* check that we have a cb for each enabled target, we don't check
 272          * shader_mask because it seems mesa isn't always setting it :(
 273          */
 274         tmp = track->cb_target_mask;
 275         for (i = 0; i < 8; i++) {
 276                 if ((tmp >> (i * 4)) & 0xF) {
 277                         /* at least one component is enabled */
 278                         if (track->cb_color_bo[i] == NULL) {
 279                                 dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
 280                                         __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
 281                                 return -EINVAL;
 282                         }
 283                         /* perform rewrite of CB_COLOR[0-7]_SIZE */
 284                         r = r600_cs_track_validate_cb(p, i);
 285                         if (r)
 286                                 return r;
 287                 }
 288         }
 289         /* Check depth buffer */
 290         if (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
 291                 G_028800_Z_ENABLE(track->db_depth_control)) {
 292                 u32 nviews, bpe, ntiles;
 293                 if (track->db_bo == NULL) {
 294                         dev_warn(p->dev, "z/stencil with no depth buffer\n");
 295                         return -EINVAL;
 296                 }
 297                 if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
 298                         dev_warn(p->dev, "this kernel doesn't support z/stencil htile\n");
 299                         return -EINVAL;
 300                 }
 301                 switch (G_028010_FORMAT(track->db_depth_info)) {
 302                 case V_028010_DEPTH_16:
 303                         bpe = 2;
 304                         break;
 305                 case V_028010_DEPTH_X8_24:
 306                 case V_028010_DEPTH_8_24:
 307                 case V_028010_DEPTH_X8_24_FLOAT:
 308                 case V_028010_DEPTH_8_24_FLOAT:
 309                 case V_028010_DEPTH_32_FLOAT:
 310                         bpe = 4;
 311                         break;
 312                 case V_028010_DEPTH_X24_8_32_FLOAT:
 313                         bpe = 8;
 314                         break;
 315                 default:
 316                         dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
 317                         return -EINVAL;
 318                 }
 319                 if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
 320                         if (!track->db_depth_size_idx) {
 321                                 dev_warn(p->dev, "z/stencil buffer size not set\n");
 322                                 return -EINVAL;
 323                         }
 324                         printk_once(KERN_WARNING "You have old & broken userspace please consider updating mesa\n");
 325                         tmp = radeon_bo_size(track->db_bo) - track->db_offset;
 326                         tmp = (tmp / bpe) >> 6;
 327                         if (!tmp) {
 328                                 dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
 329                                                 track->db_depth_size, bpe, track->db_offset,
 330                                                 radeon_bo_size(track->db_bo));
 331                                 return -EINVAL;
 332                         }
 333                         ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
 334                 } else {
 335                         ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
 336                         nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
 337                         tmp = ntiles * bpe * 64 * nviews;
 338                         if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
 339                                 dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %d -> %d have %ld)\n",
 340                                                 track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
 341                                                 radeon_bo_size(track->db_bo));
 342                                 return -EINVAL;
 343                         }
 344                 }
 345         }
 346         return 0;
 347 }
 348
 349 /**
 350  * r600_cs_packet_parse() - parse cp packet and point ib index to next packet
 351  * @parser:     parser structure holding parsing context.
 352  * @pkt:        where to store packet informations
 353  *
 354  * Assume that chunk_ib_index is properly set. Will return -EINVAL
 355  * if packet is bigger than remaining ib size. or if packets is unknown.
 356  **/
 357 int r600_cs_packet_parse(struct radeon_cs_parser *p,
 358                         struct radeon_cs_packet *pkt,
 359                         unsigned idx)
 360 {
 361         struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
 362         uint32_t header;
 363
 364         if (idx >= ib_chunk->length_dw) {
 365                 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
 366                           idx, ib_chunk->length_dw);
 367                 return -EINVAL;
 368         }
 369         header = radeon_get_ib_value(p, idx);
 370         pkt->idx = idx;
 371         pkt->type = CP_PACKET_GET_TYPE(header);
 372         pkt->count = CP_PACKET_GET_COUNT(header);
 373         pkt->one_reg_wr = 0;
 374         switch (pkt->type) {
 375         case PACKET_TYPE0:
 376                 pkt->reg = CP_PACKET0_GET_REG(header);
 377                 break;
 378         case PACKET_TYPE3:
 379                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 380                 break;
 381         case PACKET_TYPE2:
 382                 pkt->count = -1;
 383                 break;
 384         default:
 385                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 386                 return -EINVAL;
 387         }
 388         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 389                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 390                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 391                 return -EINVAL;
 392         }
 393         return 0;
 394 }
 395
 396 /**
 397  * r600_cs_packet_next_reloc_mm() - parse next packet which should be reloc packet3
 398  * @parser:             parser structure holding parsing context.
 399  * @data:               pointer to relocation data
 400  * @offset_start:       starting offset
 401  * @offset_mask:        offset mask (to align start offset on)
 402  * @reloc:              reloc informations
 403  *
 404  * Check next packet is relocation packet3, do bo validation and compute
 405  * GPU offset using the provided start.
 406  **/
 407 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
 408                                         struct radeon_cs_reloc **cs_reloc)
 409 {
 410         struct radeon_cs_chunk *relocs_chunk;
 411         struct radeon_cs_packet p3reloc;
 412         unsigned idx;
 413         int r;
 414
 415         if (p->chunk_relocs_idx == -1) {
 416                 DRM_ERROR("No relocation chunk !\n");
 417                 return -EINVAL;
 418         }
 419         *cs_reloc = NULL;
 420         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 421         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 422         if (r) {
 423                 return r;
 424         }
 425         p->idx += p3reloc.count + 2;
 426         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 427                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 428                           p3reloc.idx);
 429                 return -EINVAL;
 430         }
 431         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
 432         if (idx >= relocs_chunk->length_dw) {
 433                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 434                           idx, relocs_chunk->length_dw);
 435                 return -EINVAL;
 436         }
 437         /* FIXME: we assume reloc size is 4 dwords */
 438         *cs_reloc = p->relocs_ptr[(idx / 4)];
 439         return 0;
 440 }
 441
 442 /**
 443  * r600_cs_packet_next_reloc_nomm() - parse next packet which should be reloc packet3
 444  * @parser:             parser structure holding parsing context.
 445  * @data:               pointer to relocation data
 446  * @offset_start:       starting offset
 447  * @offset_mask:        offset mask (to align start offset on)
 448  * @reloc:              reloc informations
 449  *
 450  * Check next packet is relocation packet3, do bo validation and compute
 451  * GPU offset using the provided start.
 452  **/
 453 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
 454                                         struct radeon_cs_reloc **cs_reloc)
 455 {
 456         struct radeon_cs_chunk *relocs_chunk;
 457         struct radeon_cs_packet p3reloc;
 458         unsigned idx;
 459         int r;
 460
 461         if (p->chunk_relocs_idx == -1) {
 462                 DRM_ERROR("No relocation chunk !\n");
 463                 return -EINVAL;
 464         }
 465         *cs_reloc = NULL;
 466         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 467         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 468         if (r) {
 469                 return r;
 470         }
 471         p->idx += p3reloc.count + 2;
 472         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 473                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 474                           p3reloc.idx);
 475                 return -EINVAL;
 476         }
 477         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
 478         if (idx >= relocs_chunk->length_dw) {
 479                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 480                           idx, relocs_chunk->length_dw);
 481                 return -EINVAL;
 482         }
 483         *cs_reloc = p->relocs;
 484         (*cs_reloc)->lobj.gpu_offset = (u64)relocs_chunk->kdata[idx + 3] << 32;
 485         (*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
 486         return 0;
 487 }
 488
 489 /**
 490  * r600_cs_packet_next_is_pkt3_nop() - test if next packet is packet3 nop for reloc
 491  * @parser:             parser structure holding parsing context.
 492  *
 493  * Check next packet is relocation packet3, do bo validation and compute
 494  * GPU offset using the provided start.
 495  **/
 496 static inline int r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
 497 {
 498         struct radeon_cs_packet p3reloc;
 499         int r;
 500
 501         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 502         if (r) {
 503                 return 0;
 504         }
 505         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 506                 return 0;
 507         }
 508         return 1;
 509 }
 510
 511 /**
 512  * r600_cs_packet_next_vline() - parse userspace VLINE packet
 513  * @parser:             parser structure holding parsing context.
 514  *
 515  * Userspace sends a special sequence for VLINE waits.
 516  * PACKET0 - VLINE_START_END + value
 517  * PACKET3 - WAIT_REG_MEM poll vline status reg
 518  * RELOC (P3) - crtc_id in reloc.
 519  *
 520  * This function parses this and relocates the VLINE START END
 521  * and WAIT_REG_MEM packets to the correct crtc.
 522  * It also detects a switched off crtc and nulls out the
 523  * wait in that case.
 524  */
 525 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
 526 {
 527         struct drm_mode_object *obj;
 528         struct drm_crtc *crtc;
 529         struct radeon_crtc *radeon_crtc;
 530         struct radeon_cs_packet p3reloc, wait_reg_mem;
 531         int crtc_id;
 532         int r;
 533         uint32_t header, h_idx, reg, wait_reg_mem_info;
 534         volatile uint32_t *ib;
 535
 536         ib = p->ib->ptr;
 537
 538         /* parse the WAIT_REG_MEM */
 539         r = r600_cs_packet_parse(p, &wait_reg_mem, p->idx);
 540         if (r)
 541                 return r;
 542
 543         /* check its a WAIT_REG_MEM */
 544         if (wait_reg_mem.type != PACKET_TYPE3 ||
 545             wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
 546                 DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
 547                 r = -EINVAL;
 548                 return r;
 549         }
 550
 551         wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
 552         /* bit 4 is reg (0) or mem (1) */
 553         if (wait_reg_mem_info & 0x10) {
 554                 DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
 555                 r = -EINVAL;
 556                 return r;
 557         }
 558         /* waiting for value to be equal */
 559         if ((wait_reg_mem_info & 0x7) != 0x3) {
 560                 DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
 561                 r = -EINVAL;
 562                 return r;
 563         }
 564         if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != AVIVO_D1MODE_VLINE_STATUS) {
 565                 DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
 566                 r = -EINVAL;
 567                 return r;
 568         }
 569
 570         if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != AVIVO_D1MODE_VLINE_STAT) {
 571                 DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
 572                 r = -EINVAL;
 573                 return r;
 574         }
 575
 576         /* jump over the NOP */
 577         r = r600_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
 578         if (r)
 579                 return r;
 580
 581         h_idx = p->idx - 2;
 582         p->idx += wait_reg_mem.count + 2;
 583         p->idx += p3reloc.count + 2;
 584
 585         header = radeon_get_ib_value(p, h_idx);
 586         crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
 587         reg = CP_PACKET0_GET_REG(header);
 588         mutex_lock(&p->rdev->ddev->mode_config.mutex);
 589         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 590         if (!obj) {
 591                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 592                 r = -EINVAL;
 593                 goto out;
 594         }
 595         crtc = obj_to_crtc(obj);
 596         radeon_crtc = to_radeon_crtc(crtc);
 597         crtc_id = radeon_crtc->crtc_id;
 598
 599         if (!crtc->enabled) {
 600                 /* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
 601                 ib[h_idx + 2] = PACKET2(0);
 602                 ib[h_idx + 3] = PACKET2(0);
 603                 ib[h_idx + 4] = PACKET2(0);
 604                 ib[h_idx + 5] = PACKET2(0);
 605                 ib[h_idx + 6] = PACKET2(0);
 606                 ib[h_idx + 7] = PACKET2(0);
 607                 ib[h_idx + 8] = PACKET2(0);
 608         } else if (crtc_id == 1) {
 609                 switch (reg) {
 610                 case AVIVO_D1MODE_VLINE_START_END:
 611                         header &= ~R600_CP_PACKET0_REG_MASK;
 612                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
 613                         break;
 614                 default:
 615                         DRM_ERROR("unknown crtc reloc\n");
 616                         r = -EINVAL;
 617                         goto out;
 618                 }
 619                 ib[h_idx] = header;
 620                 ib[h_idx + 4] = AVIVO_D2MODE_VLINE_STATUS >> 2;
 621         }
 622 out:
 623         mutex_unlock(&p->rdev->ddev->mode_config.mutex);
 624         return r;
 625 }
 626
 627 static int r600_packet0_check(struct radeon_cs_parser *p,
 628                                 struct radeon_cs_packet *pkt,
 629                                 unsigned idx, unsigned reg)
 630 {
 631         int r;
 632
 633         switch (reg) {
 634         case AVIVO_D1MODE_VLINE_START_END:
 635                 r = r600_cs_packet_parse_vline(p);
 636                 if (r) {
 637                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 638                                         idx, reg);
 639                         return r;
 640                 }
 641                 break;
 642         default:
 643                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
 644                        reg, idx);
 645                 return -EINVAL;
 646         }
 647         return 0;
 648 }
 649
 650 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
 651                                 struct radeon_cs_packet *pkt)
 652 {
 653         unsigned reg, i;
 654         unsigned idx;
 655         int r;
 656
 657         idx = pkt->idx + 1;
 658         reg = pkt->reg;
 659         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 660                 r = r600_packet0_check(p, pkt, idx, reg);
 661                 if (r) {
 662                         return r;
 663                 }
 664         }
 665         return 0;
 666 }
 667
 668 /**
 669  * r600_cs_check_reg() - check if register is authorized or not
 670  * @parser: parser structure holding parsing context
 671  * @reg: register we are testing
 672  * @idx: index into the cs buffer
 673  *
 674  * This function will test against r600_reg_safe_bm and return 0
 675  * if register is safe. If register is not flag as safe this function
 676  * will test it against a list of register needind special handling.
 677  */
 678 static inline int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
 679 {
 680         struct r600_cs_track *track = (struct r600_cs_track *)p->track;
 681         struct radeon_cs_reloc *reloc;
 682         u32 last_reg = ARRAY_SIZE(r600_reg_safe_bm);
 683         u32 m, i, tmp, *ib;
 684         int r;
 685
 686         i = (reg >> 7);
 687         if (i > last_reg) {
 688                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
 689                 return -EINVAL;
 690         }
 691         m = 1 << ((reg >> 2) & 31);
 692         if (!(r600_reg_safe_bm[i] & m))
 693                 return 0;
 694         ib = p->ib->ptr;
 695         switch (reg) {
 696         /* force following reg to 0 in an attemp to disable out buffer
 697          * which will need us to better understand how it works to perform
 698          * security check on it (Jerome)
 699          */
 700         case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
 701         case R_008C44_SQ_ESGS_RING_SIZE:
 702         case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
 703         case R_008C54_SQ_ESTMP_RING_SIZE:
 704         case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
 705         case R_008C74_SQ_FBUF_RING_SIZE:
 706         case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
 707         case R_008C5C_SQ_GSTMP_RING_SIZE:
 708         case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
 709         case R_008C4C_SQ_GSVS_RING_SIZE:
 710         case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
 711         case R_008C6C_SQ_PSTMP_RING_SIZE:
 712         case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
 713         case R_008C7C_SQ_REDUC_RING_SIZE:
 714         case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
 715         case R_008C64_SQ_VSTMP_RING_SIZE:
 716         case R_0288C8_SQ_GS_VERT_ITEMSIZE:
 717                 /* get value to populate the IB don't remove */
 718                 tmp =radeon_get_ib_value(p, idx);
 719                 ib[idx] = 0;
 720                 break;
 721         case SQ_CONFIG:
 722                 track->sq_config = radeon_get_ib_value(p, idx);
 723                 break;
 724         case R_028800_DB_DEPTH_CONTROL:
 725                 track->db_depth_control = radeon_get_ib_value(p, idx);
 726                 break;
 727         case R_028010_DB_DEPTH_INFO:
 728                 track->db_depth_info = radeon_get_ib_value(p, idx);
 729                 break;
 730         case R_028004_DB_DEPTH_VIEW:
 731                 track->db_depth_view = radeon_get_ib_value(p, idx);
 732                 break;
 733         case R_028000_DB_DEPTH_SIZE:
 734                 track->db_depth_size = radeon_get_ib_value(p, idx);
 735                 track->db_depth_size_idx = idx;
 736                 break;
 737         case R_028AB0_VGT_STRMOUT_EN:
 738                 track->vgt_strmout_en = radeon_get_ib_value(p, idx);
 739                 break;
 740         case R_028B20_VGT_STRMOUT_BUFFER_EN:
 741                 track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
 742                 break;
 743         case R_028238_CB_TARGET_MASK:
 744                 track->cb_target_mask = radeon_get_ib_value(p, idx);
 745                 break;
 746         case R_02823C_CB_SHADER_MASK:
 747                 track->cb_shader_mask = radeon_get_ib_value(p, idx);
 748                 break;
 749         case R_028C04_PA_SC_AA_CONFIG:
 750                 tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
 751                 track->nsamples = 1 << tmp;
 752                 break;
 753         case R_0280A0_CB_COLOR0_INFO:
 754         case R_0280A4_CB_COLOR1_INFO:
 755         case R_0280A8_CB_COLOR2_INFO:
 756         case R_0280AC_CB_COLOR3_INFO:
 757         case R_0280B0_CB_COLOR4_INFO:
 758         case R_0280B4_CB_COLOR5_INFO:
 759         case R_0280B8_CB_COLOR6_INFO:
 760         case R_0280BC_CB_COLOR7_INFO:
 761                 tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
 762                 track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
 763                 break;
 764         case R_028060_CB_COLOR0_SIZE:
 765         case R_028064_CB_COLOR1_SIZE:
 766         case R_028068_CB_COLOR2_SIZE:
 767         case R_02806C_CB_COLOR3_SIZE:
 768         case R_028070_CB_COLOR4_SIZE:
 769         case R_028074_CB_COLOR5_SIZE:
 770         case R_028078_CB_COLOR6_SIZE:
 771         case R_02807C_CB_COLOR7_SIZE:
 772                 tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
 773                 track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
 774                 track->cb_color_size_idx[tmp] = idx;
 775                 break;
 776                 /* This register were added late, there is userspace
 777                  * which does provide relocation for those but set
 778                  * 0 offset. In order to avoid breaking old userspace
 779                  * we detect this and set address to point to last
 780                  * CB_COLOR0_BASE, note that if userspace doesn't set
 781                  * CB_COLOR0_BASE before this register we will report
 782                  * error. Old userspace always set CB_COLOR0_BASE
 783                  * before any of this.
 784                  */
 785         case R_0280E0_CB_COLOR0_FRAG:
 786         case R_0280E4_CB_COLOR1_FRAG:
 787         case R_0280E8_CB_COLOR2_FRAG:
 788         case R_0280EC_CB_COLOR3_FRAG:
 789         case R_0280F0_CB_COLOR4_FRAG:
 790         case R_0280F4_CB_COLOR5_FRAG:
 791         case R_0280F8_CB_COLOR6_FRAG:
 792         case R_0280FC_CB_COLOR7_FRAG:
 793                 tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
 794                 if (!r600_cs_packet_next_is_pkt3_nop(p)) {
 795                         if (!track->cb_color_base_last[tmp]) {
 796                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
 797                                 return -EINVAL;
 798                         }
 799                         ib[idx] = track->cb_color_base_last[tmp];
 800                         printk_once(KERN_WARNING "You have old & broken userspace "
 801                                         "please consider updating mesa & xf86-video-ati\n");
 802                         track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
 803                 } else {
 804                         r = r600_cs_packet_next_reloc(p, &reloc);
 805                         if (r) {
 806                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
 807                                 return -EINVAL;
 808                         }
 809                         ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 810                         track->cb_color_frag_bo[tmp] = reloc->robj;
 811                 }
 812                 break;
 813         case R_0280C0_CB_COLOR0_TILE:
 814         case R_0280C4_CB_COLOR1_TILE:
 815         case R_0280C8_CB_COLOR2_TILE:
 816         case R_0280CC_CB_COLOR3_TILE:
 817         case R_0280D0_CB_COLOR4_TILE:
 818         case R_0280D4_CB_COLOR5_TILE:
 819         case R_0280D8_CB_COLOR6_TILE:
 820         case R_0280DC_CB_COLOR7_TILE:
 821                 tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
 822                 if (!r600_cs_packet_next_is_pkt3_nop(p)) {
 823                         if (!track->cb_color_base_last[tmp]) {
 824                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
 825                                 return -EINVAL;
 826                         }
 827                         ib[idx] = track->cb_color_base_last[tmp];
 828                         printk_once(KERN_WARNING "You have old & broken userspace "
 829                                         "please consider updating mesa & xf86-video-ati\n");
 830                         track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
 831                 } else {
 832                         r = r600_cs_packet_next_reloc(p, &reloc);
 833                         if (r) {
 834                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
 835                                 return -EINVAL;
 836                         }
 837                         ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 838                         track->cb_color_tile_bo[tmp] = reloc->robj;
 839                 }
 840                 break;
 841         case CB_COLOR0_BASE:
 842         case CB_COLOR1_BASE:
 843         case CB_COLOR2_BASE:
 844         case CB_COLOR3_BASE:
 845         case CB_COLOR4_BASE:
 846         case CB_COLOR5_BASE:
 847         case CB_COLOR6_BASE:
 848         case CB_COLOR7_BASE:
 849                 r = r600_cs_packet_next_reloc(p, &reloc);
 850                 if (r) {
 851                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
 852                                         "0x%04X\n", reg);
 853                         return -EINVAL;
 854                 }
 855                 tmp = (reg - CB_COLOR0_BASE) / 4;
 856                 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
 857                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 858                 track->cb_color_base_last[tmp] = ib[idx];
 859                 track->cb_color_bo[tmp] = reloc->robj;
 860                 break;
 861         case DB_DEPTH_BASE:
 862                 r = r600_cs_packet_next_reloc(p, &reloc);
 863                 if (r) {
 864                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
 865                                         "0x%04X\n", reg);
 866                         return -EINVAL;
 867                 }
 868                 track->db_offset = radeon_get_ib_value(p, idx);
 869                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 870                 track->db_bo = reloc->robj;
 871                 break;
 872         case DB_HTILE_DATA_BASE:
 873         case SQ_PGM_START_FS:
 874         case SQ_PGM_START_ES:
 875         case SQ_PGM_START_VS:
 876         case SQ_PGM_START_GS:
 877         case SQ_PGM_START_PS:
 878         case SQ_ALU_CONST_CACHE_GS_0:
 879         case SQ_ALU_CONST_CACHE_GS_1:
 880         case SQ_ALU_CONST_CACHE_GS_2:
 881         case SQ_ALU_CONST_CACHE_GS_3:
 882         case SQ_ALU_CONST_CACHE_GS_4:
 883         case SQ_ALU_CONST_CACHE_GS_5:
 884         case SQ_ALU_CONST_CACHE_GS_6:
 885         case SQ_ALU_CONST_CACHE_GS_7:
 886         case SQ_ALU_CONST_CACHE_GS_8:
 887         case SQ_ALU_CONST_CACHE_GS_9:
 888         case SQ_ALU_CONST_CACHE_GS_10:
 889         case SQ_ALU_CONST_CACHE_GS_11:
 890         case SQ_ALU_CONST_CACHE_GS_12:
 891         case SQ_ALU_CONST_CACHE_GS_13:
 892         case SQ_ALU_CONST_CACHE_GS_14:
 893         case SQ_ALU_CONST_CACHE_GS_15:
 894         case SQ_ALU_CONST_CACHE_PS_0:
 895         case SQ_ALU_CONST_CACHE_PS_1:
 896         case SQ_ALU_CONST_CACHE_PS_2:
 897         case SQ_ALU_CONST_CACHE_PS_3:
 898         case SQ_ALU_CONST_CACHE_PS_4:
 899         case SQ_ALU_CONST_CACHE_PS_5:
 900         case SQ_ALU_CONST_CACHE_PS_6:
 901         case SQ_ALU_CONST_CACHE_PS_7:
 902         case SQ_ALU_CONST_CACHE_PS_8:
 903         case SQ_ALU_CONST_CACHE_PS_9:
 904         case SQ_ALU_CONST_CACHE_PS_10:
 905         case SQ_ALU_CONST_CACHE_PS_11:
 906         case SQ_ALU_CONST_CACHE_PS_12:
 907         case SQ_ALU_CONST_CACHE_PS_13:
 908         case SQ_ALU_CONST_CACHE_PS_14:
 909         case SQ_ALU_CONST_CACHE_PS_15:
 910         case SQ_ALU_CONST_CACHE_VS_0:
 911         case SQ_ALU_CONST_CACHE_VS_1:
 912         case SQ_ALU_CONST_CACHE_VS_2:
 913         case SQ_ALU_CONST_CACHE_VS_3:
 914         case SQ_ALU_CONST_CACHE_VS_4:
 915         case SQ_ALU_CONST_CACHE_VS_5:
 916         case SQ_ALU_CONST_CACHE_VS_6:
 917         case SQ_ALU_CONST_CACHE_VS_7:
 918         case SQ_ALU_CONST_CACHE_VS_8:
 919         case SQ_ALU_CONST_CACHE_VS_9:
 920         case SQ_ALU_CONST_CACHE_VS_10:
 921         case SQ_ALU_CONST_CACHE_VS_11:
 922         case SQ_ALU_CONST_CACHE_VS_12:
 923         case SQ_ALU_CONST_CACHE_VS_13:
 924         case SQ_ALU_CONST_CACHE_VS_14:
 925         case SQ_ALU_CONST_CACHE_VS_15:
 926                 r = r600_cs_packet_next_reloc(p, &reloc);
 927                 if (r) {
 928                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
 929                                         "0x%04X\n", reg);
 930                         return -EINVAL;
 931                 }
 932                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
 933                 break;
 934         default:
 935                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
 936                 return -EINVAL;
 937         }
 938         return 0;
 939 }
 940
 941 static inline unsigned minify(unsigned size, unsigned levels)
 942 {
 943         size = size >> levels;
 944         if (size < 1)
 945                 size = 1;
 946         return size;
 947 }
 948
 949 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned nlevels,
 950                                 unsigned w0, unsigned h0, unsigned d0, unsigned bpe,
 951                                 unsigned *l0_size, unsigned *mipmap_size)
 952 {
 953         unsigned offset, i, level, face;
 954         unsigned width, height, depth, rowstride, size;
 955
 956         w0 = minify(w0, 0);
 957         h0 = minify(h0, 0);
 958         d0 = minify(d0, 0);
 959         for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
 960                 width = minify(w0, i);
 961                 height = minify(h0, i);
 962                 depth = minify(d0, i);
 963                 for(face = 0; face < nfaces; face++) {
 964                         rowstride = ((width * bpe) + 255) & ~255;
 965                         size = height * rowstride * depth;
 966                         offset += size;
 967                         offset = (offset + 0x1f) & ~0x1f;
 968                 }
 969         }
 970         *l0_size = (((w0 * bpe) + 255) & ~255) * h0 * d0;
 971         *mipmap_size = offset;
 972         if (!blevel)
 973                 *mipmap_size -= *l0_size;
 974         if (!nlevels)
 975                 *mipmap_size = *l0_size;
 976 }
 977
 978 /**
 979  * r600_check_texture_resource() - check if register is authorized or not
 980  * @p: parser structure holding parsing context
 981  * @idx: index into the cs buffer
 982  * @texture: texture's bo structure
 983  * @mipmap: mipmap's bo structure
 984  *
 985  * This function will check that the resource has valid field and that
 986  * the texture and mipmap bo object are big enough to cover this resource.
 987  */
 988 static inline int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
 989                                                 struct radeon_bo *texture,
 990                                                 struct radeon_bo *mipmap)
 991 {
 992         u32 nfaces, nlevels, blevel, w0, h0, d0, bpe = 0;
 993         u32 word0, word1, l0_size, mipmap_size;
 994
 995         /* on legacy kernel we don't perform advanced check */
 996         if (p->rdev == NULL)
 997                 return 0;
 998         word0 = radeon_get_ib_value(p, idx + 0);
 999         word1 = radeon_get_ib_value(p, idx + 1);
1000         w0 = G_038000_TEX_WIDTH(word0) + 1;
1001         h0 = G_038004_TEX_HEIGHT(word1) + 1;
1002         d0 = G_038004_TEX_DEPTH(word1);
1003         nfaces = 1;
1004         switch (G_038000_DIM(word0)) {
1005         case V_038000_SQ_TEX_DIM_1D:
1006         case V_038000_SQ_TEX_DIM_2D:
1007         case V_038000_SQ_TEX_DIM_3D:
1008                 break;
1009         case V_038000_SQ_TEX_DIM_CUBEMAP:
1010                 nfaces = 6;
1011                 break;
1012         case V_038000_SQ_TEX_DIM_1D_ARRAY:
1013         case V_038000_SQ_TEX_DIM_2D_ARRAY:
1014         case V_038000_SQ_TEX_DIM_2D_MSAA:
1015         case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1016         default:
1017                 dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1018                 return -EINVAL;
1019         }
1020         if (r600_bpe_from_format(&bpe,  G_038004_DATA_FORMAT(word1))) {
1021                 dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1022                          __func__, __LINE__, G_038004_DATA_FORMAT(word1));
1023                 return -EINVAL;
1024         }
1025         word0 = radeon_get_ib_value(p, idx + 4);
1026         word1 = radeon_get_ib_value(p, idx + 5);
1027         blevel = G_038010_BASE_LEVEL(word0);
1028         nlevels = G_038014_LAST_LEVEL(word1);
1029         r600_texture_size(nfaces, blevel, nlevels, w0, h0, d0, bpe, &l0_size, &mipmap_size);
1030         /* using get ib will give us the offset into the texture bo */
1031         word0 = radeon_get_ib_value(p, idx + 2);
1032         if ((l0_size + word0) > radeon_bo_size(texture)) {
1033                 dev_warn(p->dev, "texture bo too small (%d %d %d %d -> %d have %ld)\n",
1034                         w0, h0, bpe, word0, l0_size, radeon_bo_size(texture));
1035                 return -EINVAL;
1036         }
1037         /* using get ib will give us the offset into the mipmap bo */
1038         word0 = radeon_get_ib_value(p, idx + 3);
1039         if ((mipmap_size + word0) > radeon_bo_size(mipmap)) {
1040                 dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1041                         w0, h0, bpe, blevel, nlevels, word0, mipmap_size, radeon_bo_size(texture));
1042                 return -EINVAL;
1043         }
1044         return 0;
1045 }
1046
1047 static int r600_packet3_check(struct radeon_cs_parser *p,
1048                                 struct radeon_cs_packet *pkt)
1049 {
1050         struct radeon_cs_reloc *reloc;
1051         struct r600_cs_track *track;
1052         volatile u32 *ib;
1053         unsigned idx;
1054         unsigned i;
1055         unsigned start_reg, end_reg, reg;
1056         int r;
1057         u32 idx_value;
1058
1059         track = (struct r600_cs_track *)p->track;
1060         ib = p->ib->ptr;
1061         idx = pkt->idx + 1;
1062         idx_value = radeon_get_ib_value(p, idx);
1063
1064         switch (pkt->opcode) {
1065         case PACKET3_START_3D_CMDBUF:
1066                 if (p->family >= CHIP_RV770 || pkt->count) {
1067                         DRM_ERROR("bad START_3D\n");
1068                         return -EINVAL;
1069                 }
1070                 break;
1071         case PACKET3_CONTEXT_CONTROL:
1072                 if (pkt->count != 1) {
1073                         DRM_ERROR("bad CONTEXT_CONTROL\n");
1074                         return -EINVAL;
1075                 }
1076                 break;
1077         case PACKET3_INDEX_TYPE:
1078         case PACKET3_NUM_INSTANCES:
1079                 if (pkt->count) {
1080                         DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1081                         return -EINVAL;
1082                 }
1083                 break;
1084         case PACKET3_DRAW_INDEX:
1085                 if (pkt->count != 3) {
1086                         DRM_ERROR("bad DRAW_INDEX\n");
1087                         return -EINVAL;
1088                 }
1089                 r = r600_cs_packet_next_reloc(p, &reloc);
1090                 if (r) {
1091                         DRM_ERROR("bad DRAW_INDEX\n");
1092                         return -EINVAL;
1093                 }
1094                 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1095                 ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1096                 r = r600_cs_track_check(p);
1097                 if (r) {
1098                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1099                         return r;
1100                 }
1101                 break;
1102         case PACKET3_DRAW_INDEX_AUTO:
1103                 if (pkt->count != 1) {
1104                         DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1105                         return -EINVAL;
1106                 }
1107                 r = r600_cs_track_check(p);
1108                 if (r) {
1109                         dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1110                         return r;
1111                 }
1112                 break;
1113         case PACKET3_DRAW_INDEX_IMMD_BE:
1114         case PACKET3_DRAW_INDEX_IMMD:
1115                 if (pkt->count < 2) {
1116                         DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1117                         return -EINVAL;
1118                 }
1119                 r = r600_cs_track_check(p);
1120                 if (r) {
1121                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1122                         return r;
1123                 }
1124                 break;
1125         case PACKET3_WAIT_REG_MEM:
1126                 if (pkt->count != 5) {
1127                         DRM_ERROR("bad WAIT_REG_MEM\n");
1128                         return -EINVAL;
1129                 }
1130                 /* bit 4 is reg (0) or mem (1) */
1131                 if (idx_value & 0x10) {
1132                         r = r600_cs_packet_next_reloc(p, &reloc);
1133                         if (r) {
1134                                 DRM_ERROR("bad WAIT_REG_MEM\n");
1135                                 return -EINVAL;
1136                         }
1137                         ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1138                         ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1139                 }
1140                 break;
1141         case PACKET3_SURFACE_SYNC:
1142                 if (pkt->count != 3) {
1143                         DRM_ERROR("bad SURFACE_SYNC\n");
1144                         return -EINVAL;
1145                 }
1146                 /* 0xffffffff/0x0 is flush all cache flag */
1147                 if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1148                     radeon_get_ib_value(p, idx + 2) != 0) {
1149                         r = r600_cs_packet_next_reloc(p, &reloc);
1150                         if (r) {
1151                                 DRM_ERROR("bad SURFACE_SYNC\n");
1152                                 return -EINVAL;
1153                         }
1154                         ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1155                 }
1156                 break;
1157         case PACKET3_EVENT_WRITE:
1158                 if (pkt->count != 2 && pkt->count != 0) {
1159                         DRM_ERROR("bad EVENT_WRITE\n");
1160                         return -EINVAL;
1161                 }
1162                 if (pkt->count) {
1163                         r = r600_cs_packet_next_reloc(p, &reloc);
1164                         if (r) {
1165                                 DRM_ERROR("bad EVENT_WRITE\n");
1166                                 return -EINVAL;
1167                         }
1168                         ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1169                         ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1170                 }
1171                 break;
1172         case PACKET3_EVENT_WRITE_EOP:
1173                 if (pkt->count != 4) {
1174                         DRM_ERROR("bad EVENT_WRITE_EOP\n");
1175                         return -EINVAL;
1176                 }
1177                 r = r600_cs_packet_next_reloc(p, &reloc);
1178                 if (r) {
1179                         DRM_ERROR("bad EVENT_WRITE\n");
1180                         return -EINVAL;
1181                 }
1182                 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1183                 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1184                 break;
1185         case PACKET3_SET_CONFIG_REG:
1186                 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1187                 end_reg = 4 * pkt->count + start_reg - 4;
1188                 if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1189                     (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1190                     (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1191                         DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1192                         return -EINVAL;
1193                 }
1194                 for (i = 0; i < pkt->count; i++) {
1195                         reg = start_reg + (4 * i);
1196                         r = r600_cs_check_reg(p, reg, idx+1+i);
1197                         if (r)
1198                                 return r;
1199                 }
1200                 break;
1201         case PACKET3_SET_CONTEXT_REG:
1202                 start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1203                 end_reg = 4 * pkt->count + start_reg - 4;
1204                 if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1205                     (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1206                     (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1207                         DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1208                         return -EINVAL;
1209                 }
1210                 for (i = 0; i < pkt->count; i++) {
1211                         reg = start_reg + (4 * i);
1212                         r = r600_cs_check_reg(p, reg, idx+1+i);
1213                         if (r)
1214                                 return r;
1215                 }
1216                 break;
1217         case PACKET3_SET_RESOURCE:
1218                 if (pkt->count % 7) {
1219                         DRM_ERROR("bad SET_RESOURCE\n");
1220                         return -EINVAL;
1221                 }
1222                 start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1223                 end_reg = 4 * pkt->count + start_reg - 4;
1224                 if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1225                     (start_reg >= PACKET3_SET_RESOURCE_END) ||
1226                     (end_reg >= PACKET3_SET_RESOURCE_END)) {
1227                         DRM_ERROR("bad SET_RESOURCE\n");
1228                         return -EINVAL;
1229                 }
1230                 for (i = 0; i < (pkt->count / 7); i++) {
1231                         struct radeon_bo *texture, *mipmap;
1232                         u32 size, offset;
1233
1234                         switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1235                         case SQ_TEX_VTX_VALID_TEXTURE:
1236                                 /* tex base */
1237                                 r = r600_cs_packet_next_reloc(p, &reloc);
1238                                 if (r) {
1239                                         DRM_ERROR("bad SET_RESOURCE\n");
1240                                         return -EINVAL;
1241                                 }
1242                                 ib[idx+1+(i*7)+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1243                                 texture = reloc->robj;
1244                                 /* tex mip base */
1245                                 r = r600_cs_packet_next_reloc(p, &reloc);
1246                                 if (r) {
1247                                         DRM_ERROR("bad SET_RESOURCE\n");
1248                                         return -EINVAL;
1249                                 }
1250                                 ib[idx+1+(i*7)+3] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1251                                 mipmap = reloc->robj;
1252                                 r = r600_check_texture_resource(p,  idx+(i*7)+1,
1253                                                 texture, mipmap);
1254                                 if (r)
1255                                         return r;
1256                                 break;
1257                         case SQ_TEX_VTX_VALID_BUFFER:
1258                                 /* vtx base */
1259                                 r = r600_cs_packet_next_reloc(p, &reloc);
1260                                 if (r) {
1261                                         DRM_ERROR("bad SET_RESOURCE\n");
1262                                         return -EINVAL;
1263                                 }
1264                                 offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
1265                                 size = radeon_get_ib_value(p, idx+1+(i*7)+1);
1266                                 if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
1267                                         /* force size to size of the buffer */
1268                                         dev_warn(p->dev, "vbo resource seems too big for the bo\n");
1269                                         ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj);
1270                                 }
1271                                 ib[idx+1+(i*7)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
1272                                 ib[idx+1+(i*7)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1273                                 break;
1274                         case SQ_TEX_VTX_INVALID_TEXTURE:
1275                         case SQ_TEX_VTX_INVALID_BUFFER:
1276                         default:
1277                                 DRM_ERROR("bad SET_RESOURCE\n");
1278                                 return -EINVAL;
1279                         }
1280                 }
1281                 break;
1282         case PACKET3_SET_ALU_CONST:
1283                 if (track->sq_config & DX9_CONSTS) {
1284                         start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
1285                         end_reg = 4 * pkt->count + start_reg - 4;
1286                         if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
1287                             (start_reg >= PACKET3_SET_ALU_CONST_END) ||
1288                             (end_reg >= PACKET3_SET_ALU_CONST_END)) {
1289                                 DRM_ERROR("bad SET_ALU_CONST\n");
1290                                 return -EINVAL;
1291                         }
1292                 }
1293                 break;
1294         case PACKET3_SET_BOOL_CONST:
1295                 start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
1296                 end_reg = 4 * pkt->count + start_reg - 4;
1297                 if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
1298                     (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
1299                     (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
1300                         DRM_ERROR("bad SET_BOOL_CONST\n");
1301                         return -EINVAL;
1302                 }
1303                 break;
1304         case PACKET3_SET_LOOP_CONST:
1305                 start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
1306                 end_reg = 4 * pkt->count + start_reg - 4;
1307                 if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
1308                     (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
1309                     (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
1310                         DRM_ERROR("bad SET_LOOP_CONST\n");
1311                         return -EINVAL;
1312                 }
1313                 break;
1314         case PACKET3_SET_CTL_CONST:
1315                 start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
1316                 end_reg = 4 * pkt->count + start_reg - 4;
1317                 if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
1318                     (start_reg >= PACKET3_SET_CTL_CONST_END) ||
1319                     (end_reg >= PACKET3_SET_CTL_CONST_END)) {
1320                         DRM_ERROR("bad SET_CTL_CONST\n");
1321                         return -EINVAL;
1322                 }
1323                 break;
1324         case PACKET3_SET_SAMPLER:
1325                 if (pkt->count % 3) {
1326                         DRM_ERROR("bad SET_SAMPLER\n");
1327                         return -EINVAL;
1328                 }
1329                 start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
1330                 end_reg = 4 * pkt->count + start_reg - 4;
1331                 if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
1332                     (start_reg >= PACKET3_SET_SAMPLER_END) ||
1333                     (end_reg >= PACKET3_SET_SAMPLER_END)) {
1334                         DRM_ERROR("bad SET_SAMPLER\n");
1335                         return -EINVAL;
1336                 }
1337                 break;
1338         case PACKET3_SURFACE_BASE_UPDATE:
1339                 if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
1340                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1341                         return -EINVAL;
1342                 }
1343                 if (pkt->count) {
1344                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1345                         return -EINVAL;
1346                 }
1347                 break;
1348         case PACKET3_NOP:
1349                 break;
1350         default:
1351                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1352                 return -EINVAL;
1353         }
1354         return 0;
1355 }
1356
1357 int r600_cs_parse(struct radeon_cs_parser *p)
1358 {
1359         struct radeon_cs_packet pkt;
1360         struct r600_cs_track *track;
1361         int r;
1362
1363         if (p->track == NULL) {
1364                 /* initialize tracker, we are in kms */
1365                 track = kzalloc(sizeof(*track), GFP_KERNEL);
1366                 if (track == NULL)
1367                         return -ENOMEM;
1368                 r600_cs_track_init(track);
1369                 if (p->rdev->family < CHIP_RV770) {
1370                         track->npipes = p->rdev->config.r600.tiling_npipes;
1371                         track->nbanks = p->rdev->config.r600.tiling_nbanks;
1372                         track->group_size = p->rdev->config.r600.tiling_group_size;
1373                 } else if (p->rdev->family <= CHIP_RV740) {
1374                         track->npipes = p->rdev->config.rv770.tiling_npipes;
1375                         track->nbanks = p->rdev->config.rv770.tiling_nbanks;
1376                         track->group_size = p->rdev->config.rv770.tiling_group_size;
1377                 }
1378                 p->track = track;
1379         }
1380         do {
1381                 r = r600_cs_packet_parse(p, &pkt, p->idx);
1382                 if (r) {
1383                         kfree(p->track);
1384                         p->track = NULL;
1385                         return r;
1386                 }
1387                 p->idx += pkt.count + 2;
1388                 switch (pkt.type) {
1389                 case PACKET_TYPE0:
1390                         r = r600_cs_parse_packet0(p, &pkt);
1391                         break;
1392                 case PACKET_TYPE2:
1393                         break;
1394                 case PACKET_TYPE3:
1395                         r = r600_packet3_check(p, &pkt);
1396                         break;
1397                 default:
1398                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1399                         kfree(p->track);
1400                         p->track = NULL;
1401                         return -EINVAL;
1402                 }
1403                 if (r) {
1404                         kfree(p->track);
1405                         p->track = NULL;
1406                         return r;
1407                 }
1408         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1409 #if 0
1410         for (r = 0; r < p->ib->length_dw; r++) {
1411                 printk(KERN_INFO "%05d  0x%08X\n", r, p->ib->ptr[r]);
1412                 mdelay(1);
1413         }
1414 #endif
1415         kfree(p->track);
1416         p->track = NULL;
1417         return 0;
1418 }
1419
1420 static int r600_cs_parser_relocs_legacy(struct radeon_cs_parser *p)
1421 {
1422         if (p->chunk_relocs_idx == -1) {
1423                 return 0;
1424         }
1425         p->relocs = kzalloc(sizeof(struct radeon_cs_reloc), GFP_KERNEL);
1426         if (p->relocs == NULL) {
1427                 return -ENOMEM;
1428         }
1429         return 0;
1430 }
1431
1432 /**
1433  * cs_parser_fini() - clean parser states
1434  * @parser:     parser structure holding parsing context.
1435  * @error:      error number
1436  *
1437  * If error is set than unvalidate buffer, otherwise just free memory
1438  * used by parsing context.
1439  **/
1440 static void r600_cs_parser_fini(struct radeon_cs_parser *parser, int error)
1441 {
1442         unsigned i;
1443
1444         kfree(parser->relocs);
1445         for (i = 0; i < parser->nchunks; i++) {
1446                 kfree(parser->chunks[i].kdata);
1447                 kfree(parser->chunks[i].kpage[0]);
1448                 kfree(parser->chunks[i].kpage[1]);
1449         }
1450         kfree(parser->chunks);
1451         kfree(parser->chunks_array);
1452 }
1453
1454 int r600_cs_legacy(struct drm_device *dev, void *data, struct drm_file *filp,
1455                         unsigned family, u32 *ib, int *l)
1456 {
1457         struct radeon_cs_parser parser;
1458         struct radeon_cs_chunk *ib_chunk;
1459         struct radeon_ib fake_ib;
1460         struct r600_cs_track *track;
1461         int r;
1462
1463         /* initialize tracker */
1464         track = kzalloc(sizeof(*track), GFP_KERNEL);
1465         if (track == NULL)
1466                 return -ENOMEM;
1467         r600_cs_track_init(track);
1468         r600_cs_legacy_get_tiling_conf(dev, &track->npipes, &track->nbanks, &track->group_size);
1469         /* initialize parser */
1470         memset(&parser, 0, sizeof(struct radeon_cs_parser));
1471         parser.filp = filp;
1472         parser.dev = &dev->pdev->dev;
1473         parser.rdev = NULL;
1474         parser.family = family;
1475         parser.ib = &fake_ib;
1476         parser.track = track;
1477         fake_ib.ptr = ib;
1478         r = radeon_cs_parser_init(&parser, data);
1479         if (r) {
1480                 DRM_ERROR("Failed to initialize parser !\n");
1481                 r600_cs_parser_fini(&parser, r);
1482                 return r;
1483         }
1484         r = r600_cs_parser_relocs_legacy(&parser);
1485         if (r) {
1486                 DRM_ERROR("Failed to parse relocation !\n");
1487                 r600_cs_parser_fini(&parser, r);
1488                 return r;
1489         }
1490         /* Copy the packet into the IB, the parser will read from the
1491          * input memory (cached) and write to the IB (which can be
1492          * uncached). */
1493         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
1494         parser.ib->length_dw = ib_chunk->length_dw;
1495         *l = parser.ib->length_dw;
1496         r = r600_cs_parse(&parser);
1497         if (r) {
1498                 DRM_ERROR("Invalid command stream !\n");
1499                 r600_cs_parser_fini(&parser, r);
1500                 return r;
1501         }
1502         r = radeon_cs_finish_pages(&parser);
1503         if (r) {
1504                 DRM_ERROR("Invalid command stream !\n");
1505                 r600_cs_parser_fini(&parser, r);
1506                 return r;
1507         }
1508         r600_cs_parser_fini(&parser, r);
1509         return r;
1510 }
1511
1512 void r600_cs_legacy_init(void)
1513 {
1514         r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;
1515 }