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         u64                     cb_color_bo_mc[8];
  54         u32                     cb_color_bo_offset[8];
  55         struct radeon_bo        *cb_color_frag_bo[8];
  56         struct radeon_bo        *cb_color_tile_bo[8];
  57         u32                     cb_color_info[8];
  58         u32                     cb_color_size_idx[8];
  59         u32                     cb_target_mask;
  60         u32                     cb_shader_mask;
  61         u32                     cb_color_size[8];
  62         u32                     vgt_strmout_en;
  63         u32                     vgt_strmout_buffer_en;
  64         u32                     db_depth_control;
  65         u32                     db_depth_info;
  66         u32                     db_depth_size_idx;
  67         u32                     db_depth_view;
  68         u32                     db_depth_size;
  69         u32                     db_offset;
  70         struct radeon_bo        *db_bo;
  71         u64                     db_bo_mc;
  72 };
  73
  74 #define FMT_8_BIT(fmt, vc)   [fmt] = { 1, 1, 1, vc, CHIP_R600 }
  75 #define FMT_16_BIT(fmt, vc)  [fmt] = { 1, 1, 2, vc, CHIP_R600 }
  76 #define FMT_24_BIT(fmt)      [fmt] = { 1, 1, 3,  0, CHIP_R600 }
  77 #define FMT_32_BIT(fmt, vc)  [fmt] = { 1, 1, 4, vc, CHIP_R600 }
  78 #define FMT_48_BIT(fmt)      [fmt] = { 1, 1, 6,  0, CHIP_R600 }
  79 #define FMT_64_BIT(fmt, vc)  [fmt] = { 1, 1, 8, vc, CHIP_R600 }
  80 #define FMT_96_BIT(fmt)      [fmt] = { 1, 1, 12, 0, CHIP_R600 }
  81 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16,vc, CHIP_R600 }
  82
  83 struct gpu_formats {
  84         unsigned blockwidth;
  85         unsigned blockheight;
  86         unsigned blocksize;
  87         unsigned valid_color;
  88         enum radeon_family min_family;
  89 };
  90
  91 static const struct gpu_formats color_formats_table[] = {
  92         /* 8 bit */
  93         FMT_8_BIT(V_038004_COLOR_8, 1),
  94         FMT_8_BIT(V_038004_COLOR_4_4, 1),
  95         FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
  96         FMT_8_BIT(V_038004_FMT_1, 0),
  97
  98         /* 16-bit */
  99         FMT_16_BIT(V_038004_COLOR_16, 1),
 100         FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
 101         FMT_16_BIT(V_038004_COLOR_8_8, 1),
 102         FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
 103         FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
 104         FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
 105         FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
 106         FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
 107
 108         /* 24-bit */
 109         FMT_24_BIT(V_038004_FMT_8_8_8),
 110
 111         /* 32-bit */
 112         FMT_32_BIT(V_038004_COLOR_32, 1),
 113         FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
 114         FMT_32_BIT(V_038004_COLOR_16_16, 1),
 115         FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
 116         FMT_32_BIT(V_038004_COLOR_8_24, 1),
 117         FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
 118         FMT_32_BIT(V_038004_COLOR_24_8, 1),
 119         FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
 120         FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
 121         FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
 122         FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
 123         FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
 124         FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
 125         FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
 126         FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
 127         FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
 128         FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
 129         FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
 130
 131         /* 48-bit */
 132         FMT_48_BIT(V_038004_FMT_16_16_16),
 133         FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
 134
 135         /* 64-bit */
 136         FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
 137         FMT_64_BIT(V_038004_COLOR_32_32, 1),
 138         FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
 139         FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
 140         FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
 141
 142         FMT_96_BIT(V_038004_FMT_32_32_32),
 143         FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
 144
 145         /* 128-bit */
 146         FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
 147         FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
 148
 149         [V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
 150         [V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
 151
 152         /* block compressed formats */
 153         [V_038004_FMT_BC1] = { 4, 4, 8, 0 },
 154         [V_038004_FMT_BC2] = { 4, 4, 16, 0 },
 155         [V_038004_FMT_BC3] = { 4, 4, 16, 0 },
 156         [V_038004_FMT_BC4] = { 4, 4, 8, 0 },
 157         [V_038004_FMT_BC5] = { 4, 4, 16, 0},
 158         [V_038004_FMT_BC6] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
 159         [V_038004_FMT_BC7] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
 160
 161         /* The other Evergreen formats */
 162         [V_038004_FMT_32_AS_32_32_32_32] = { 1, 1, 4, 0, CHIP_CEDAR},
 163 };
 164
 165 static inline bool fmt_is_valid_color(u32 format)
 166 {
 167         if (format >= ARRAY_SIZE(color_formats_table))
 168                 return false;
 169
 170         if (color_formats_table[format].valid_color)
 171                 return true;
 172
 173         return false;
 174 }
 175
 176 static inline bool fmt_is_valid_texture(u32 format, enum radeon_family family)
 177 {
 178         if (format >= ARRAY_SIZE(color_formats_table))
 179                 return false;
 180
 181         if (family < color_formats_table[format].min_family)
 182                 return false;
 183
 184         if (color_formats_table[format].blockwidth > 0)
 185                 return true;
 186
 187         return false;
 188 }
 189
 190 static inline int fmt_get_blocksize(u32 format)
 191 {
 192         if (format >= ARRAY_SIZE(color_formats_table))
 193                 return 0;
 194
 195         return color_formats_table[format].blocksize;
 196 }
 197
 198 static inline int fmt_get_nblocksx(u32 format, u32 w)
 199 {
 200         unsigned bw;
 201
 202         if (format >= ARRAY_SIZE(color_formats_table))
 203                 return 0;
 204
 205         bw = color_formats_table[format].blockwidth;
 206         if (bw == 0)
 207                 return 0;
 208
 209         return (w + bw - 1) / bw;
 210 }
 211
 212 static inline int fmt_get_nblocksy(u32 format, u32 h)
 213 {
 214         unsigned bh;
 215
 216         if (format >= ARRAY_SIZE(color_formats_table))
 217                 return 0;
 218
 219         bh = color_formats_table[format].blockheight;
 220         if (bh == 0)
 221                 return 0;
 222
 223         return (h + bh - 1) / bh;
 224 }
 225
 226 static inline int r600_bpe_from_format(u32 *bpe, u32 format)
 227 {
 228         unsigned res;
 229
 230         if (format >= ARRAY_SIZE(color_formats_table))
 231                 goto fail;
 232
 233         res = color_formats_table[format].blocksize;
 234         if (res == 0)
 235                 goto fail;
 236
 237         *bpe = res;
 238         return 0;
 239
 240 fail:
 241         *bpe = 16;
 242         return -EINVAL;
 243 }
 244
 245 struct array_mode_checker {
 246         int array_mode;
 247         u32 group_size;
 248         u32 nbanks;
 249         u32 npipes;
 250         u32 nsamples;
 251         u32 blocksize;
 252 };
 253
 254 /* returns alignment in pixels for pitch/height/depth and bytes for base */
 255 static inline int r600_get_array_mode_alignment(struct array_mode_checker *values,
 256                                                 u32 *pitch_align,
 257                                                 u32 *height_align,
 258                                                 u32 *depth_align,
 259                                                 u64 *base_align)
 260 {
 261         u32 tile_width = 8;
 262         u32 tile_height = 8;
 263         u32 macro_tile_width = values->nbanks;
 264         u32 macro_tile_height = values->npipes;
 265         u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
 266         u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
 267
 268         switch (values->array_mode) {
 269         case ARRAY_LINEAR_GENERAL:
 270                 /* technically tile_width/_height for pitch/height */
 271                 *pitch_align = 1; /* tile_width */
 272                 *height_align = 1; /* tile_height */
 273                 *depth_align = 1;
 274                 *base_align = 1;
 275                 break;
 276         case ARRAY_LINEAR_ALIGNED:
 277                 *pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
 278                 *height_align = tile_height;
 279                 *depth_align = 1;
 280                 *base_align = values->group_size;
 281                 break;
 282         case ARRAY_1D_TILED_THIN1:
 283                 *pitch_align = max((u32)tile_width,
 284                                    (u32)(values->group_size /
 285                                          (tile_height * values->blocksize * values->nsamples)));
 286                 *height_align = tile_height;
 287                 *depth_align = 1;
 288                 *base_align = values->group_size;
 289                 break;
 290         case ARRAY_2D_TILED_THIN1:
 291                 *pitch_align = max((u32)macro_tile_width,
 292                                   (u32)(((values->group_size / tile_height) /
 293                                          (values->blocksize * values->nsamples)) *
 294                                         values->nbanks)) * tile_width;
 295                 *height_align = macro_tile_height * tile_height;
 296                 *depth_align = 1;
 297                 *base_align = max(macro_tile_bytes,
 298                                   (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
 299                 break;
 300         default:
 301                 return -EINVAL;
 302         }
 303
 304         return 0;
 305 }
 306
 307 static void r600_cs_track_init(struct r600_cs_track *track)
 308 {
 309         int i;
 310
 311         /* assume DX9 mode */
 312         track->sq_config = DX9_CONSTS;
 313         for (i = 0; i < 8; i++) {
 314                 track->cb_color_base_last[i] = 0;
 315                 track->cb_color_size[i] = 0;
 316                 track->cb_color_size_idx[i] = 0;
 317                 track->cb_color_info[i] = 0;
 318                 track->cb_color_bo[i] = NULL;
 319                 track->cb_color_bo_offset[i] = 0xFFFFFFFF;
 320                 track->cb_color_bo_mc[i] = 0xFFFFFFFF;
 321         }
 322         track->cb_target_mask = 0xFFFFFFFF;
 323         track->cb_shader_mask = 0xFFFFFFFF;
 324         track->db_bo = NULL;
 325         track->db_bo_mc = 0xFFFFFFFF;
 326         /* assume the biggest format and that htile is enabled */
 327         track->db_depth_info = 7 | (1 << 25);
 328         track->db_depth_view = 0xFFFFC000;
 329         track->db_depth_size = 0xFFFFFFFF;
 330         track->db_depth_size_idx = 0;
 331         track->db_depth_control = 0xFFFFFFFF;
 332 }
 333
 334 static inline int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
 335 {
 336         struct r600_cs_track *track = p->track;
 337         u32 slice_tile_max, size, tmp;
 338         u32 height, height_align, pitch, pitch_align, depth_align;
 339         u64 base_offset, base_align;
 340         struct array_mode_checker array_check;
 341         volatile u32 *ib = p->ib->ptr;
 342         unsigned array_mode;
 343         u32 format;
 344         if (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
 345                 dev_warn(p->dev, "FMASK or CMASK buffer are not supported by this kernel\n");
 346                 return -EINVAL;
 347         }
 348         size = radeon_bo_size(track->cb_color_bo[i]) - track->cb_color_bo_offset[i];
 349         format = G_0280A0_FORMAT(track->cb_color_info[i]);
 350         if (!fmt_is_valid_color(format)) {
 351                 dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
 352                          __func__, __LINE__, format,
 353                         i, track->cb_color_info[i]);
 354                 return -EINVAL;
 355         }
 356         /* pitch in pixels */
 357         pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
 358         slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
 359         slice_tile_max *= 64;
 360         height = slice_tile_max / pitch;
 361         if (height > 8192)
 362                 height = 8192;
 363         array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
 364
 365         base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
 366         array_check.array_mode = array_mode;
 367         array_check.group_size = track->group_size;
 368         array_check.nbanks = track->nbanks;
 369         array_check.npipes = track->npipes;
 370         array_check.nsamples = track->nsamples;
 371         array_check.blocksize = fmt_get_blocksize(format);
 372         if (r600_get_array_mode_alignment(&array_check,
 373                                           &pitch_align, &height_align, &depth_align, &base_align)) {
 374                 dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
 375                          G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
 376                          track->cb_color_info[i]);
 377                 return -EINVAL;
 378         }
 379         switch (array_mode) {
 380         case V_0280A0_ARRAY_LINEAR_GENERAL:
 381                 break;
 382         case V_0280A0_ARRAY_LINEAR_ALIGNED:
 383                 break;
 384         case V_0280A0_ARRAY_1D_TILED_THIN1:
 385                 /* avoid breaking userspace */
 386                 if (height > 7)
 387                         height &= ~0x7;
 388                 break;
 389         case V_0280A0_ARRAY_2D_TILED_THIN1:
 390                 break;
 391         default:
 392                 dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
 393                         G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
 394                         track->cb_color_info[i]);
 395                 return -EINVAL;
 396         }
 397
 398         if (!IS_ALIGNED(pitch, pitch_align)) {
 399                 dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
 400                          __func__, __LINE__, pitch, pitch_align, array_mode);
 401                 return -EINVAL;
 402         }
 403         if (!IS_ALIGNED(height, height_align)) {
 404                 dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
 405                          __func__, __LINE__, height, height_align, array_mode);
 406                 return -EINVAL;
 407         }
 408         if (!IS_ALIGNED(base_offset, base_align)) {
 409                 dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
 410                          base_offset, base_align, array_mode);
 411                 return -EINVAL;
 412         }
 413
 414         /* check offset */
 415         tmp = fmt_get_nblocksy(format, height) * fmt_get_nblocksx(format, pitch) * fmt_get_blocksize(format);
 416         if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
 417                 if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
 418                         /* the initial DDX does bad things with the CB size occasionally */
 419                         /* it rounds up height too far for slice tile max but the BO is smaller */
 420                         /* r600c,g also seem to flush at bad times in some apps resulting in
 421                          * bogus values here. So for linear just allow anything to avoid breaking
 422                          * broken userspace.
 423                          */
 424                 } else {
 425                         dev_warn(p->dev, "%s offset[%d] %d %d %d %lu too big\n", __func__, i,
 426                                  array_mode,
 427                                  track->cb_color_bo_offset[i], tmp,
 428                                  radeon_bo_size(track->cb_color_bo[i]));
 429                         return -EINVAL;
 430                 }
 431         }
 432         /* limit max tile */
 433         tmp = (height * pitch) >> 6;
 434         if (tmp < slice_tile_max)
 435                 slice_tile_max = tmp;
 436         tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
 437                 S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
 438         ib[track->cb_color_size_idx[i]] = tmp;
 439         return 0;
 440 }
 441
 442 static int r600_cs_track_check(struct radeon_cs_parser *p)
 443 {
 444         struct r600_cs_track *track = p->track;
 445         u32 tmp;
 446         int r, i;
 447         volatile u32 *ib = p->ib->ptr;
 448
 449         /* on legacy kernel we don't perform advanced check */
 450         if (p->rdev == NULL)
 451                 return 0;
 452         /* we don't support out buffer yet */
 453         if (track->vgt_strmout_en || track->vgt_strmout_buffer_en) {
 454                 dev_warn(p->dev, "this kernel doesn't support SMX output buffer\n");
 455                 return -EINVAL;
 456         }
 457         /* check that we have a cb for each enabled target, we don't check
 458          * shader_mask because it seems mesa isn't always setting it :(
 459          */
 460         tmp = track->cb_target_mask;
 461         for (i = 0; i < 8; i++) {
 462                 if ((tmp >> (i * 4)) & 0xF) {
 463                         /* at least one component is enabled */
 464                         if (track->cb_color_bo[i] == NULL) {
 465                                 dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
 466                                         __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
 467                                 return -EINVAL;
 468                         }
 469                         /* perform rewrite of CB_COLOR[0-7]_SIZE */
 470                         r = r600_cs_track_validate_cb(p, i);
 471                         if (r)
 472                                 return r;
 473                 }
 474         }
 475         /* Check depth buffer */
 476         if (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
 477                 G_028800_Z_ENABLE(track->db_depth_control)) {
 478                 u32 nviews, bpe, ntiles, size, slice_tile_max;
 479                 u32 height, height_align, pitch, pitch_align, depth_align;
 480                 u64 base_offset, base_align;
 481                 struct array_mode_checker array_check;
 482                 int array_mode;
 483
 484                 if (track->db_bo == NULL) {
 485                         dev_warn(p->dev, "z/stencil with no depth buffer\n");
 486                         return -EINVAL;
 487                 }
 488                 if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
 489                         dev_warn(p->dev, "this kernel doesn't support z/stencil htile\n");
 490                         return -EINVAL;
 491                 }
 492                 switch (G_028010_FORMAT(track->db_depth_info)) {
 493                 case V_028010_DEPTH_16:
 494                         bpe = 2;
 495                         break;
 496                 case V_028010_DEPTH_X8_24:
 497                 case V_028010_DEPTH_8_24:
 498                 case V_028010_DEPTH_X8_24_FLOAT:
 499                 case V_028010_DEPTH_8_24_FLOAT:
 500                 case V_028010_DEPTH_32_FLOAT:
 501                         bpe = 4;
 502                         break;
 503                 case V_028010_DEPTH_X24_8_32_FLOAT:
 504                         bpe = 8;
 505                         break;
 506                 default:
 507                         dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
 508                         return -EINVAL;
 509                 }
 510                 if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
 511                         if (!track->db_depth_size_idx) {
 512                                 dev_warn(p->dev, "z/stencil buffer size not set\n");
 513                                 return -EINVAL;
 514                         }
 515                         tmp = radeon_bo_size(track->db_bo) - track->db_offset;
 516                         tmp = (tmp / bpe) >> 6;
 517                         if (!tmp) {
 518                                 dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
 519                                                 track->db_depth_size, bpe, track->db_offset,
 520                                                 radeon_bo_size(track->db_bo));
 521                                 return -EINVAL;
 522                         }
 523                         ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
 524                 } else {
 525                         size = radeon_bo_size(track->db_bo);
 526                         /* pitch in pixels */
 527                         pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
 528                         slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
 529                         slice_tile_max *= 64;
 530                         height = slice_tile_max / pitch;
 531                         if (height > 8192)
 532                                 height = 8192;
 533                         base_offset = track->db_bo_mc + track->db_offset;
 534                         array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
 535                         array_check.array_mode = array_mode;
 536                         array_check.group_size = track->group_size;
 537                         array_check.nbanks = track->nbanks;
 538                         array_check.npipes = track->npipes;
 539                         array_check.nsamples = track->nsamples;
 540                         array_check.blocksize = bpe;
 541                         if (r600_get_array_mode_alignment(&array_check,
 542                                                           &pitch_align, &height_align, &depth_align, &base_align)) {
 543                                 dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
 544                                          G_028010_ARRAY_MODE(track->db_depth_info),
 545                                          track->db_depth_info);
 546                                 return -EINVAL;
 547                         }
 548                         switch (array_mode) {
 549                         case V_028010_ARRAY_1D_TILED_THIN1:
 550                                 /* don't break userspace */
 551                                 height &= ~0x7;
 552                                 break;
 553                         case V_028010_ARRAY_2D_TILED_THIN1:
 554                                 break;
 555                         default:
 556                                 dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
 557                                          G_028010_ARRAY_MODE(track->db_depth_info),
 558                                          track->db_depth_info);
 559                                 return -EINVAL;
 560                         }
 561
 562                         if (!IS_ALIGNED(pitch, pitch_align)) {
 563                                 dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
 564                                          __func__, __LINE__, pitch, pitch_align, array_mode);
 565                                 return -EINVAL;
 566                         }
 567                         if (!IS_ALIGNED(height, height_align)) {
 568                                 dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
 569                                          __func__, __LINE__, height, height_align, array_mode);
 570                                 return -EINVAL;
 571                         }
 572                         if (!IS_ALIGNED(base_offset, base_align)) {
 573                                 dev_warn(p->dev, "%s offset[%d] 0x%llx, 0x%llx, %d not aligned\n", __func__, i,
 574                                          base_offset, base_align, array_mode);
 575                                 return -EINVAL;
 576                         }
 577
 578                         ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
 579                         nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
 580                         tmp = ntiles * bpe * 64 * nviews;
 581                         if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
 582                                 dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
 583                                          array_mode,
 584                                          track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
 585                                          radeon_bo_size(track->db_bo));
 586                                 return -EINVAL;
 587                         }
 588                 }
 589         }
 590         return 0;
 591 }
 592
 593 /**
 594  * r600_cs_packet_parse() - parse cp packet and point ib index to next packet
 595  * @parser:     parser structure holding parsing context.
 596  * @pkt:        where to store packet informations
 597  *
 598  * Assume that chunk_ib_index is properly set. Will return -EINVAL
 599  * if packet is bigger than remaining ib size. or if packets is unknown.
 600  **/
 601 int r600_cs_packet_parse(struct radeon_cs_parser *p,
 602                         struct radeon_cs_packet *pkt,
 603                         unsigned idx)
 604 {
 605         struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
 606         uint32_t header;
 607
 608         if (idx >= ib_chunk->length_dw) {
 609                 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
 610                           idx, ib_chunk->length_dw);
 611                 return -EINVAL;
 612         }
 613         header = radeon_get_ib_value(p, idx);
 614         pkt->idx = idx;
 615         pkt->type = CP_PACKET_GET_TYPE(header);
 616         pkt->count = CP_PACKET_GET_COUNT(header);
 617         pkt->one_reg_wr = 0;
 618         switch (pkt->type) {
 619         case PACKET_TYPE0:
 620                 pkt->reg = CP_PACKET0_GET_REG(header);
 621                 break;
 622         case PACKET_TYPE3:
 623                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 624                 break;
 625         case PACKET_TYPE2:
 626                 pkt->count = -1;
 627                 break;
 628         default:
 629                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 630                 return -EINVAL;
 631         }
 632         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 633                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 634                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 635                 return -EINVAL;
 636         }
 637         return 0;
 638 }
 639
 640 /**
 641  * r600_cs_packet_next_reloc_mm() - parse next packet which should be reloc packet3
 642  * @parser:             parser structure holding parsing context.
 643  * @data:               pointer to relocation data
 644  * @offset_start:       starting offset
 645  * @offset_mask:        offset mask (to align start offset on)
 646  * @reloc:              reloc informations
 647  *
 648  * Check next packet is relocation packet3, do bo validation and compute
 649  * GPU offset using the provided start.
 650  **/
 651 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
 652                                         struct radeon_cs_reloc **cs_reloc)
 653 {
 654         struct radeon_cs_chunk *relocs_chunk;
 655         struct radeon_cs_packet p3reloc;
 656         unsigned idx;
 657         int r;
 658
 659         if (p->chunk_relocs_idx == -1) {
 660                 DRM_ERROR("No relocation chunk !\n");
 661                 return -EINVAL;
 662         }
 663         *cs_reloc = NULL;
 664         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 665         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 666         if (r) {
 667                 return r;
 668         }
 669         p->idx += p3reloc.count + 2;
 670         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 671                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 672                           p3reloc.idx);
 673                 return -EINVAL;
 674         }
 675         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
 676         if (idx >= relocs_chunk->length_dw) {
 677                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 678                           idx, relocs_chunk->length_dw);
 679                 return -EINVAL;
 680         }
 681         /* FIXME: we assume reloc size is 4 dwords */
 682         *cs_reloc = p->relocs_ptr[(idx / 4)];
 683         return 0;
 684 }
 685
 686 /**
 687  * r600_cs_packet_next_reloc_nomm() - parse next packet which should be reloc packet3
 688  * @parser:             parser structure holding parsing context.
 689  * @data:               pointer to relocation data
 690  * @offset_start:       starting offset
 691  * @offset_mask:        offset mask (to align start offset on)
 692  * @reloc:              reloc informations
 693  *
 694  * Check next packet is relocation packet3, do bo validation and compute
 695  * GPU offset using the provided start.
 696  **/
 697 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
 698                                         struct radeon_cs_reloc **cs_reloc)
 699 {
 700         struct radeon_cs_chunk *relocs_chunk;
 701         struct radeon_cs_packet p3reloc;
 702         unsigned idx;
 703         int r;
 704
 705         if (p->chunk_relocs_idx == -1) {
 706                 DRM_ERROR("No relocation chunk !\n");
 707                 return -EINVAL;
 708         }
 709         *cs_reloc = NULL;
 710         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 711         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 712         if (r) {
 713                 return r;
 714         }
 715         p->idx += p3reloc.count + 2;
 716         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 717                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 718                           p3reloc.idx);
 719                 return -EINVAL;
 720         }
 721         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
 722         if (idx >= relocs_chunk->length_dw) {
 723                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 724                           idx, relocs_chunk->length_dw);
 725                 return -EINVAL;
 726         }
 727         *cs_reloc = p->relocs;
 728         (*cs_reloc)->lobj.gpu_offset = (u64)relocs_chunk->kdata[idx + 3] << 32;
 729         (*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
 730         return 0;
 731 }
 732
 733 /**
 734  * r600_cs_packet_next_is_pkt3_nop() - test if next packet is packet3 nop for reloc
 735  * @parser:             parser structure holding parsing context.
 736  *
 737  * Check next packet is relocation packet3, do bo validation and compute
 738  * GPU offset using the provided start.
 739  **/
 740 static inline int r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
 741 {
 742         struct radeon_cs_packet p3reloc;
 743         int r;
 744
 745         r = r600_cs_packet_parse(p, &p3reloc, p->idx);
 746         if (r) {
 747                 return 0;
 748         }
 749         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 750                 return 0;
 751         }
 752         return 1;
 753 }
 754
 755 /**
 756  * r600_cs_packet_next_vline() - parse userspace VLINE packet
 757  * @parser:             parser structure holding parsing context.
 758  *
 759  * Userspace sends a special sequence for VLINE waits.
 760  * PACKET0 - VLINE_START_END + value
 761  * PACKET3 - WAIT_REG_MEM poll vline status reg
 762  * RELOC (P3) - crtc_id in reloc.
 763  *
 764  * This function parses this and relocates the VLINE START END
 765  * and WAIT_REG_MEM packets to the correct crtc.
 766  * It also detects a switched off crtc and nulls out the
 767  * wait in that case.
 768  */
 769 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
 770 {
 771         struct drm_mode_object *obj;
 772         struct drm_crtc *crtc;
 773         struct radeon_crtc *radeon_crtc;
 774         struct radeon_cs_packet p3reloc, wait_reg_mem;
 775         int crtc_id;
 776         int r;
 777         uint32_t header, h_idx, reg, wait_reg_mem_info;
 778         volatile uint32_t *ib;
 779
 780         ib = p->ib->ptr;
 781
 782         /* parse the WAIT_REG_MEM */
 783         r = r600_cs_packet_parse(p, &wait_reg_mem, p->idx);
 784         if (r)
 785                 return r;
 786
 787         /* check its a WAIT_REG_MEM */
 788         if (wait_reg_mem.type != PACKET_TYPE3 ||
 789             wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
 790                 DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
 791                 return -EINVAL;
 792         }
 793
 794         wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
 795         /* bit 4 is reg (0) or mem (1) */
 796         if (wait_reg_mem_info & 0x10) {
 797                 DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
 798                 return -EINVAL;
 799         }
 800         /* waiting for value to be equal */
 801         if ((wait_reg_mem_info & 0x7) != 0x3) {
 802                 DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
 803                 return -EINVAL;
 804         }
 805         if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != AVIVO_D1MODE_VLINE_STATUS) {
 806                 DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
 807                 return -EINVAL;
 808         }
 809
 810         if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != AVIVO_D1MODE_VLINE_STAT) {
 811                 DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
 812                 return -EINVAL;
 813         }
 814
 815         /* jump over the NOP */
 816         r = r600_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
 817         if (r)
 818                 return r;
 819
 820         h_idx = p->idx - 2;
 821         p->idx += wait_reg_mem.count + 2;
 822         p->idx += p3reloc.count + 2;
 823
 824         header = radeon_get_ib_value(p, h_idx);
 825         crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
 826         reg = CP_PACKET0_GET_REG(header);
 827
 828         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 829         if (!obj) {
 830                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 831                 return -EINVAL;
 832         }
 833         crtc = obj_to_crtc(obj);
 834         radeon_crtc = to_radeon_crtc(crtc);
 835         crtc_id = radeon_crtc->crtc_id;
 836
 837         if (!crtc->enabled) {
 838                 /* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
 839                 ib[h_idx + 2] = PACKET2(0);
 840                 ib[h_idx + 3] = PACKET2(0);
 841                 ib[h_idx + 4] = PACKET2(0);
 842                 ib[h_idx + 5] = PACKET2(0);
 843                 ib[h_idx + 6] = PACKET2(0);
 844                 ib[h_idx + 7] = PACKET2(0);
 845                 ib[h_idx + 8] = PACKET2(0);
 846         } else if (crtc_id == 1) {
 847                 switch (reg) {
 848                 case AVIVO_D1MODE_VLINE_START_END:
 849                         header &= ~R600_CP_PACKET0_REG_MASK;
 850                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
 851                         break;
 852                 default:
 853                         DRM_ERROR("unknown crtc reloc\n");
 854                         return -EINVAL;
 855                 }
 856                 ib[h_idx] = header;
 857                 ib[h_idx + 4] = AVIVO_D2MODE_VLINE_STATUS >> 2;
 858         }
 859
 860         return 0;
 861 }
 862
 863 static int r600_packet0_check(struct radeon_cs_parser *p,
 864                                 struct radeon_cs_packet *pkt,
 865                                 unsigned idx, unsigned reg)
 866 {
 867         int r;
 868
 869         switch (reg) {
 870         case AVIVO_D1MODE_VLINE_START_END:
 871                 r = r600_cs_packet_parse_vline(p);
 872                 if (r) {
 873                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 874                                         idx, reg);
 875                         return r;
 876                 }
 877                 break;
 878         default:
 879                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
 880                        reg, idx);
 881                 return -EINVAL;
 882         }
 883         return 0;
 884 }
 885
 886 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
 887                                 struct radeon_cs_packet *pkt)
 888 {
 889         unsigned reg, i;
 890         unsigned idx;
 891         int r;
 892
 893         idx = pkt->idx + 1;
 894         reg = pkt->reg;
 895         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 896                 r = r600_packet0_check(p, pkt, idx, reg);
 897                 if (r) {
 898                         return r;
 899                 }
 900         }
 901         return 0;
 902 }
 903
 904 /**
 905  * r600_cs_check_reg() - check if register is authorized or not
 906  * @parser: parser structure holding parsing context
 907  * @reg: register we are testing
 908  * @idx: index into the cs buffer
 909  *
 910  * This function will test against r600_reg_safe_bm and return 0
 911  * if register is safe. If register is not flag as safe this function
 912  * will test it against a list of register needind special handling.
 913  */
 914 static inline int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
 915 {
 916         struct r600_cs_track *track = (struct r600_cs_track *)p->track;
 917         struct radeon_cs_reloc *reloc;
 918         u32 m, i, tmp, *ib;
 919         int r;
 920
 921         i = (reg >> 7);
 922         if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
 923                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
 924                 return -EINVAL;
 925         }
 926         m = 1 << ((reg >> 2) & 31);
 927         if (!(r600_reg_safe_bm[i] & m))
 928                 return 0;
 929         ib = p->ib->ptr;
 930         switch (reg) {
 931         /* force following reg to 0 in an attempt to disable out buffer
 932          * which will need us to better understand how it works to perform
 933          * security check on it (Jerome)
 934          */
 935         case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
 936         case R_008C44_SQ_ESGS_RING_SIZE:
 937         case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
 938         case R_008C54_SQ_ESTMP_RING_SIZE:
 939         case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
 940         case R_008C74_SQ_FBUF_RING_SIZE:
 941         case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
 942         case R_008C5C_SQ_GSTMP_RING_SIZE:
 943         case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
 944         case R_008C4C_SQ_GSVS_RING_SIZE:
 945         case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
 946         case R_008C6C_SQ_PSTMP_RING_SIZE:
 947         case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
 948         case R_008C7C_SQ_REDUC_RING_SIZE:
 949         case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
 950         case R_008C64_SQ_VSTMP_RING_SIZE:
 951         case R_0288C8_SQ_GS_VERT_ITEMSIZE:
 952                 /* get value to populate the IB don't remove */
 953                 tmp =radeon_get_ib_value(p, idx);
 954                 ib[idx] = 0;
 955                 break;
 956         case SQ_CONFIG:
 957                 track->sq_config = radeon_get_ib_value(p, idx);
 958                 break;
 959         case R_028800_DB_DEPTH_CONTROL:
 960                 track->db_depth_control = radeon_get_ib_value(p, idx);
 961                 break;
 962         case R_028010_DB_DEPTH_INFO:
 963                 if (r600_cs_packet_next_is_pkt3_nop(p)) {
 964                         r = r600_cs_packet_next_reloc(p, &reloc);
 965                         if (r) {
 966                                 dev_warn(p->dev, "bad SET_CONTEXT_REG "
 967                                          "0x%04X\n", reg);
 968                                 return -EINVAL;
 969                         }
 970                         track->db_depth_info = radeon_get_ib_value(p, idx);
 971                         ib[idx] &= C_028010_ARRAY_MODE;
 972                         track->db_depth_info &= C_028010_ARRAY_MODE;
 973                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
 974                                 ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
 975                                 track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
 976                         } else {
 977                                 ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
 978                                 track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
 979                         }
 980                 } else
 981                         track->db_depth_info = radeon_get_ib_value(p, idx);
 982                 break;
 983         case R_028004_DB_DEPTH_VIEW:
 984                 track->db_depth_view = radeon_get_ib_value(p, idx);
 985                 break;
 986         case R_028000_DB_DEPTH_SIZE:
 987                 track->db_depth_size = radeon_get_ib_value(p, idx);
 988                 track->db_depth_size_idx = idx;
 989                 break;
 990         case R_028AB0_VGT_STRMOUT_EN:
 991                 track->vgt_strmout_en = radeon_get_ib_value(p, idx);
 992                 break;
 993         case R_028B20_VGT_STRMOUT_BUFFER_EN:
 994                 track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
 995                 break;
 996         case R_028238_CB_TARGET_MASK:
 997                 track->cb_target_mask = radeon_get_ib_value(p, idx);
 998                 break;
 999         case R_02823C_CB_SHADER_MASK:
1000                 track->cb_shader_mask = radeon_get_ib_value(p, idx);
1001                 break;
1002         case R_028C04_PA_SC_AA_CONFIG:
1003                 tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
1004                 track->nsamples = 1 << tmp;
1005                 break;
1006         case R_0280A0_CB_COLOR0_INFO:
1007         case R_0280A4_CB_COLOR1_INFO:
1008         case R_0280A8_CB_COLOR2_INFO:
1009         case R_0280AC_CB_COLOR3_INFO:
1010         case R_0280B0_CB_COLOR4_INFO:
1011         case R_0280B4_CB_COLOR5_INFO:
1012         case R_0280B8_CB_COLOR6_INFO:
1013         case R_0280BC_CB_COLOR7_INFO:
1014                 if (r600_cs_packet_next_is_pkt3_nop(p)) {
1015                         r = r600_cs_packet_next_reloc(p, &reloc);
1016                         if (r) {
1017                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1018                                 return -EINVAL;
1019                         }
1020                         tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1021                         track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1022                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1023                                 ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1024                                 track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1025                         } else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
1026                                 ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1027                                 track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1028                         }
1029                 } else {
1030                         tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1031                         track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1032                 }
1033                 break;
1034         case R_028060_CB_COLOR0_SIZE:
1035         case R_028064_CB_COLOR1_SIZE:
1036         case R_028068_CB_COLOR2_SIZE:
1037         case R_02806C_CB_COLOR3_SIZE:
1038         case R_028070_CB_COLOR4_SIZE:
1039         case R_028074_CB_COLOR5_SIZE:
1040         case R_028078_CB_COLOR6_SIZE:
1041         case R_02807C_CB_COLOR7_SIZE:
1042                 tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1043                 track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1044                 track->cb_color_size_idx[tmp] = idx;
1045                 break;
1046                 /* This register were added late, there is userspace
1047                  * which does provide relocation for those but set
1048                  * 0 offset. In order to avoid breaking old userspace
1049                  * we detect this and set address to point to last
1050                  * CB_COLOR0_BASE, note that if userspace doesn't set
1051                  * CB_COLOR0_BASE before this register we will report
1052                  * error. Old userspace always set CB_COLOR0_BASE
1053                  * before any of this.
1054                  */
1055         case R_0280E0_CB_COLOR0_FRAG:
1056         case R_0280E4_CB_COLOR1_FRAG:
1057         case R_0280E8_CB_COLOR2_FRAG:
1058         case R_0280EC_CB_COLOR3_FRAG:
1059         case R_0280F0_CB_COLOR4_FRAG:
1060         case R_0280F4_CB_COLOR5_FRAG:
1061         case R_0280F8_CB_COLOR6_FRAG:
1062         case R_0280FC_CB_COLOR7_FRAG:
1063                 tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1064                 if (!r600_cs_packet_next_is_pkt3_nop(p)) {
1065                         if (!track->cb_color_base_last[tmp]) {
1066                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1067                                 return -EINVAL;
1068                         }
1069                         ib[idx] = track->cb_color_base_last[tmp];
1070                         track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1071                 } else {
1072                         r = r600_cs_packet_next_reloc(p, &reloc);
1073                         if (r) {
1074                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1075                                 return -EINVAL;
1076                         }
1077                         ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1078                         track->cb_color_frag_bo[tmp] = reloc->robj;
1079                 }
1080                 break;
1081         case R_0280C0_CB_COLOR0_TILE:
1082         case R_0280C4_CB_COLOR1_TILE:
1083         case R_0280C8_CB_COLOR2_TILE:
1084         case R_0280CC_CB_COLOR3_TILE:
1085         case R_0280D0_CB_COLOR4_TILE:
1086         case R_0280D4_CB_COLOR5_TILE:
1087         case R_0280D8_CB_COLOR6_TILE:
1088         case R_0280DC_CB_COLOR7_TILE:
1089                 tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1090                 if (!r600_cs_packet_next_is_pkt3_nop(p)) {
1091                         if (!track->cb_color_base_last[tmp]) {
1092                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1093                                 return -EINVAL;
1094                         }
1095                         ib[idx] = track->cb_color_base_last[tmp];
1096                         track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1097                 } else {
1098                         r = r600_cs_packet_next_reloc(p, &reloc);
1099                         if (r) {
1100                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1101                                 return -EINVAL;
1102                         }
1103                         ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1104                         track->cb_color_tile_bo[tmp] = reloc->robj;
1105                 }
1106                 break;
1107         case CB_COLOR0_BASE:
1108         case CB_COLOR1_BASE:
1109         case CB_COLOR2_BASE:
1110         case CB_COLOR3_BASE:
1111         case CB_COLOR4_BASE:
1112         case CB_COLOR5_BASE:
1113         case CB_COLOR6_BASE:
1114         case CB_COLOR7_BASE:
1115                 r = r600_cs_packet_next_reloc(p, &reloc);
1116                 if (r) {
1117                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1118                                         "0x%04X\n", reg);
1119                         return -EINVAL;
1120                 }
1121                 tmp = (reg - CB_COLOR0_BASE) / 4;
1122                 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1123                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1124                 track->cb_color_base_last[tmp] = ib[idx];
1125                 track->cb_color_bo[tmp] = reloc->robj;
1126                 track->cb_color_bo_mc[tmp] = reloc->lobj.gpu_offset;
1127                 break;
1128         case DB_DEPTH_BASE:
1129                 r = r600_cs_packet_next_reloc(p, &reloc);
1130                 if (r) {
1131                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1132                                         "0x%04X\n", reg);
1133                         return -EINVAL;
1134                 }
1135                 track->db_offset = radeon_get_ib_value(p, idx) << 8;
1136                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1137                 track->db_bo = reloc->robj;
1138                 track->db_bo_mc = reloc->lobj.gpu_offset;
1139                 break;
1140         case DB_HTILE_DATA_BASE:
1141         case SQ_PGM_START_FS:
1142         case SQ_PGM_START_ES:
1143         case SQ_PGM_START_VS:
1144         case SQ_PGM_START_GS:
1145         case SQ_PGM_START_PS:
1146         case SQ_ALU_CONST_CACHE_GS_0:
1147         case SQ_ALU_CONST_CACHE_GS_1:
1148         case SQ_ALU_CONST_CACHE_GS_2:
1149         case SQ_ALU_CONST_CACHE_GS_3:
1150         case SQ_ALU_CONST_CACHE_GS_4:
1151         case SQ_ALU_CONST_CACHE_GS_5:
1152         case SQ_ALU_CONST_CACHE_GS_6:
1153         case SQ_ALU_CONST_CACHE_GS_7:
1154         case SQ_ALU_CONST_CACHE_GS_8:
1155         case SQ_ALU_CONST_CACHE_GS_9:
1156         case SQ_ALU_CONST_CACHE_GS_10:
1157         case SQ_ALU_CONST_CACHE_GS_11:
1158         case SQ_ALU_CONST_CACHE_GS_12:
1159         case SQ_ALU_CONST_CACHE_GS_13:
1160         case SQ_ALU_CONST_CACHE_GS_14:
1161         case SQ_ALU_CONST_CACHE_GS_15:
1162         case SQ_ALU_CONST_CACHE_PS_0:
1163         case SQ_ALU_CONST_CACHE_PS_1:
1164         case SQ_ALU_CONST_CACHE_PS_2:
1165         case SQ_ALU_CONST_CACHE_PS_3:
1166         case SQ_ALU_CONST_CACHE_PS_4:
1167         case SQ_ALU_CONST_CACHE_PS_5:
1168         case SQ_ALU_CONST_CACHE_PS_6:
1169         case SQ_ALU_CONST_CACHE_PS_7:
1170         case SQ_ALU_CONST_CACHE_PS_8:
1171         case SQ_ALU_CONST_CACHE_PS_9:
1172         case SQ_ALU_CONST_CACHE_PS_10:
1173         case SQ_ALU_CONST_CACHE_PS_11:
1174         case SQ_ALU_CONST_CACHE_PS_12:
1175         case SQ_ALU_CONST_CACHE_PS_13:
1176         case SQ_ALU_CONST_CACHE_PS_14:
1177         case SQ_ALU_CONST_CACHE_PS_15:
1178         case SQ_ALU_CONST_CACHE_VS_0:
1179         case SQ_ALU_CONST_CACHE_VS_1:
1180         case SQ_ALU_CONST_CACHE_VS_2:
1181         case SQ_ALU_CONST_CACHE_VS_3:
1182         case SQ_ALU_CONST_CACHE_VS_4:
1183         case SQ_ALU_CONST_CACHE_VS_5:
1184         case SQ_ALU_CONST_CACHE_VS_6:
1185         case SQ_ALU_CONST_CACHE_VS_7:
1186         case SQ_ALU_CONST_CACHE_VS_8:
1187         case SQ_ALU_CONST_CACHE_VS_9:
1188         case SQ_ALU_CONST_CACHE_VS_10:
1189         case SQ_ALU_CONST_CACHE_VS_11:
1190         case SQ_ALU_CONST_CACHE_VS_12:
1191         case SQ_ALU_CONST_CACHE_VS_13:
1192         case SQ_ALU_CONST_CACHE_VS_14:
1193         case SQ_ALU_CONST_CACHE_VS_15:
1194                 r = r600_cs_packet_next_reloc(p, &reloc);
1195                 if (r) {
1196                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1197                                         "0x%04X\n", reg);
1198                         return -EINVAL;
1199                 }
1200                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1201                 break;
1202         case SX_MEMORY_EXPORT_BASE:
1203                 r = r600_cs_packet_next_reloc(p, &reloc);
1204                 if (r) {
1205                         dev_warn(p->dev, "bad SET_CONFIG_REG "
1206                                         "0x%04X\n", reg);
1207                         return -EINVAL;
1208                 }
1209                 ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1210                 break;
1211         default:
1212                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1213                 return -EINVAL;
1214         }
1215         return 0;
1216 }
1217
1218 static inline unsigned mip_minify(unsigned size, unsigned level)
1219 {
1220         unsigned val;
1221
1222         val = max(1U, size >> level);
1223         if (level > 0)
1224                 val = roundup_pow_of_two(val);
1225         return val;
1226 }
1227
1228 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1229                               unsigned w0, unsigned h0, unsigned d0, unsigned format,
1230                               unsigned block_align, unsigned height_align, unsigned base_align,
1231                               unsigned *l0_size, unsigned *mipmap_size)
1232 {
1233         unsigned offset, i, level;
1234         unsigned width, height, depth, size;
1235         unsigned blocksize;
1236         unsigned nbx, nby;
1237         unsigned nlevels = llevel - blevel + 1;
1238
1239         *l0_size = -1;
1240         blocksize = fmt_get_blocksize(format);
1241
1242         w0 = mip_minify(w0, 0);
1243         h0 = mip_minify(h0, 0);
1244         d0 = mip_minify(d0, 0);
1245         for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1246                 width = mip_minify(w0, i);
1247                 nbx = fmt_get_nblocksx(format, width);
1248
1249                 nbx = round_up(nbx, block_align);
1250
1251                 height = mip_minify(h0, i);
1252                 nby = fmt_get_nblocksy(format, height);
1253                 nby = round_up(nby, height_align);
1254
1255                 depth = mip_minify(d0, i);
1256
1257                 size = nbx * nby * blocksize;
1258                 if (nfaces)
1259                         size *= nfaces;
1260                 else
1261                         size *= depth;
1262
1263                 if (i == 0)
1264                         *l0_size = size;
1265
1266                 if (i == 0 || i == 1)
1267                         offset = round_up(offset, base_align);
1268
1269                 offset += size;
1270         }
1271         *mipmap_size = offset;
1272         if (llevel == 0)
1273                 *mipmap_size = *l0_size;
1274         if (!blevel)
1275                 *mipmap_size -= *l0_size;
1276 }
1277
1278 /**
1279  * r600_check_texture_resource() - check if register is authorized or not
1280  * @p: parser structure holding parsing context
1281  * @idx: index into the cs buffer
1282  * @texture: texture's bo structure
1283  * @mipmap: mipmap's bo structure
1284  *
1285  * This function will check that the resource has valid field and that
1286  * the texture and mipmap bo object are big enough to cover this resource.
1287  */
1288 static inline int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1289                                               struct radeon_bo *texture,
1290                                               struct radeon_bo *mipmap,
1291                                               u64 base_offset,
1292                                               u64 mip_offset,
1293                                               u32 tiling_flags)
1294 {
1295         struct r600_cs_track *track = p->track;
1296         u32 nfaces, llevel, blevel, w0, h0, d0;
1297         u32 word0, word1, l0_size, mipmap_size, word2, word3;
1298         u32 height_align, pitch, pitch_align, depth_align;
1299         u32 array, barray, larray;
1300         u64 base_align;
1301         struct array_mode_checker array_check;
1302         u32 format;
1303
1304         /* on legacy kernel we don't perform advanced check */
1305         if (p->rdev == NULL)
1306                 return 0;
1307
1308         /* convert to bytes */
1309         base_offset <<= 8;
1310         mip_offset <<= 8;
1311
1312         word0 = radeon_get_ib_value(p, idx + 0);
1313         if (tiling_flags & RADEON_TILING_MACRO)
1314                 word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1315         else if (tiling_flags & RADEON_TILING_MICRO)
1316                 word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1317         word1 = radeon_get_ib_value(p, idx + 1);
1318         w0 = G_038000_TEX_WIDTH(word0) + 1;
1319         h0 = G_038004_TEX_HEIGHT(word1) + 1;
1320         d0 = G_038004_TEX_DEPTH(word1);
1321         nfaces = 1;
1322         switch (G_038000_DIM(word0)) {
1323         case V_038000_SQ_TEX_DIM_1D:
1324         case V_038000_SQ_TEX_DIM_2D:
1325         case V_038000_SQ_TEX_DIM_3D:
1326                 break;
1327         case V_038000_SQ_TEX_DIM_CUBEMAP:
1328                 if (p->family >= CHIP_RV770)
1329                         nfaces = 8;
1330                 else
1331                         nfaces = 6;
1332                 break;
1333         case V_038000_SQ_TEX_DIM_1D_ARRAY:
1334         case V_038000_SQ_TEX_DIM_2D_ARRAY:
1335                 array = 1;
1336                 break;
1337         case V_038000_SQ_TEX_DIM_2D_MSAA:
1338         case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1339         default:
1340                 dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1341                 return -EINVAL;
1342         }
1343         format = G_038004_DATA_FORMAT(word1);
1344         if (!fmt_is_valid_texture(format, p->family)) {
1345                 dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1346                          __func__, __LINE__, format);
1347                 return -EINVAL;
1348         }
1349
1350         /* pitch in texels */
1351         pitch = (G_038000_PITCH(word0) + 1) * 8;
1352         array_check.array_mode = G_038000_TILE_MODE(word0);
1353         array_check.group_size = track->group_size;
1354         array_check.nbanks = track->nbanks;
1355         array_check.npipes = track->npipes;
1356         array_check.nsamples = 1;
1357         array_check.blocksize = fmt_get_blocksize(format);
1358         if (r600_get_array_mode_alignment(&array_check,
1359                                           &pitch_align, &height_align, &depth_align, &base_align)) {
1360                 dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1361                          __func__, __LINE__, G_038000_TILE_MODE(word0));
1362                 return -EINVAL;
1363         }
1364
1365         /* XXX check height as well... */
1366
1367         if (!IS_ALIGNED(pitch, pitch_align)) {
1368                 dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1369                          __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1370                 return -EINVAL;
1371         }
1372         if (!IS_ALIGNED(base_offset, base_align)) {
1373                 dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1374                          __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1375                 return -EINVAL;
1376         }
1377         if (!IS_ALIGNED(mip_offset, base_align)) {
1378                 dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1379                          __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1380                 return -EINVAL;
1381         }
1382
1383         word2 = radeon_get_ib_value(p, idx + 2) << 8;
1384         word3 = radeon_get_ib_value(p, idx + 3) << 8;
1385
1386         word0 = radeon_get_ib_value(p, idx + 4);
1387         word1 = radeon_get_ib_value(p, idx + 5);
1388         blevel = G_038010_BASE_LEVEL(word0);
1389         llevel = G_038014_LAST_LEVEL(word1);
1390         if (array == 1) {
1391                 barray = G_038014_BASE_ARRAY(word1);
1392                 larray = G_038014_LAST_ARRAY(word1);
1393
1394                 nfaces = larray - barray + 1;
1395         }
1396         r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, format,
1397                           pitch_align, height_align, base_align,
1398                           &l0_size, &mipmap_size);
1399         /* using get ib will give us the offset into the texture bo */
1400         if ((l0_size + word2) > radeon_bo_size(texture)) {
1401                 dev_warn(p->dev, "texture bo too small (%d %d %d %d -> %d have %ld)\n",
1402                         w0, h0, format, word2, l0_size, radeon_bo_size(texture));
1403                 dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1404                 return -EINVAL;
1405         }
1406         /* using get ib will give us the offset into the mipmap bo */
1407         word3 = radeon_get_ib_value(p, idx + 3) << 8;
1408         if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1409                 /*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1410                   w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1411         }
1412         return 0;
1413 }
1414
1415 static int r600_packet3_check(struct radeon_cs_parser *p,
1416                                 struct radeon_cs_packet *pkt)
1417 {
1418         struct radeon_cs_reloc *reloc;
1419         struct r600_cs_track *track;
1420         volatile u32 *ib;
1421         unsigned idx;
1422         unsigned i;
1423         unsigned start_reg, end_reg, reg;
1424         int r;
1425         u32 idx_value;
1426
1427         track = (struct r600_cs_track *)p->track;
1428         ib = p->ib->ptr;
1429         idx = pkt->idx + 1;
1430         idx_value = radeon_get_ib_value(p, idx);
1431
1432         switch (pkt->opcode) {
1433         case PACKET3_SET_PREDICATION:
1434         {
1435                 int pred_op;
1436                 int tmp;
1437                 if (pkt->count != 1) {
1438                         DRM_ERROR("bad SET PREDICATION\n");
1439                         return -EINVAL;
1440                 }
1441
1442                 tmp = radeon_get_ib_value(p, idx + 1);
1443                 pred_op = (tmp >> 16) & 0x7;
1444
1445                 /* for the clear predicate operation */
1446                 if (pred_op == 0)
1447                         return 0;
1448
1449                 if (pred_op > 2) {
1450                         DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1451                         return -EINVAL;
1452                 }
1453
1454                 r = r600_cs_packet_next_reloc(p, &reloc);
1455                 if (r) {
1456                         DRM_ERROR("bad SET PREDICATION\n");
1457                         return -EINVAL;
1458                 }
1459
1460                 ib[idx + 0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1461                 ib[idx + 1] = tmp + (upper_32_bits(reloc->lobj.gpu_offset) & 0xff);
1462         }
1463         break;
1464
1465         case PACKET3_START_3D_CMDBUF:
1466                 if (p->family >= CHIP_RV770 || pkt->count) {
1467                         DRM_ERROR("bad START_3D\n");
1468                         return -EINVAL;
1469                 }
1470                 break;
1471         case PACKET3_CONTEXT_CONTROL:
1472                 if (pkt->count != 1) {
1473                         DRM_ERROR("bad CONTEXT_CONTROL\n");
1474                         return -EINVAL;
1475                 }
1476                 break;
1477         case PACKET3_INDEX_TYPE:
1478         case PACKET3_NUM_INSTANCES:
1479                 if (pkt->count) {
1480                         DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1481                         return -EINVAL;
1482                 }
1483                 break;
1484         case PACKET3_DRAW_INDEX:
1485                 if (pkt->count != 3) {
1486                         DRM_ERROR("bad DRAW_INDEX\n");
1487                         return -EINVAL;
1488                 }
1489                 r = r600_cs_packet_next_reloc(p, &reloc);
1490                 if (r) {
1491                         DRM_ERROR("bad DRAW_INDEX\n");
1492                         return -EINVAL;
1493                 }
1494                 ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1495                 ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1496                 r = r600_cs_track_check(p);
1497                 if (r) {
1498                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1499                         return r;
1500                 }
1501                 break;
1502         case PACKET3_DRAW_INDEX_AUTO:
1503                 if (pkt->count != 1) {
1504                         DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1505                         return -EINVAL;
1506                 }
1507                 r = r600_cs_track_check(p);
1508                 if (r) {
1509                         dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1510                         return r;
1511                 }
1512                 break;
1513         case PACKET3_DRAW_INDEX_IMMD_BE:
1514         case PACKET3_DRAW_INDEX_IMMD:
1515                 if (pkt->count < 2) {
1516                         DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1517                         return -EINVAL;
1518                 }
1519                 r = r600_cs_track_check(p);
1520                 if (r) {
1521                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1522                         return r;
1523                 }
1524                 break;
1525         case PACKET3_WAIT_REG_MEM:
1526                 if (pkt->count != 5) {
1527                         DRM_ERROR("bad WAIT_REG_MEM\n");
1528                         return -EINVAL;
1529                 }
1530                 /* bit 4 is reg (0) or mem (1) */
1531                 if (idx_value & 0x10) {
1532                         r = r600_cs_packet_next_reloc(p, &reloc);
1533                         if (r) {
1534                                 DRM_ERROR("bad WAIT_REG_MEM\n");
1535                                 return -EINVAL;
1536                         }
1537                         ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1538                         ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1539                 }
1540                 break;
1541         case PACKET3_SURFACE_SYNC:
1542                 if (pkt->count != 3) {
1543                         DRM_ERROR("bad SURFACE_SYNC\n");
1544                         return -EINVAL;
1545                 }
1546                 /* 0xffffffff/0x0 is flush all cache flag */
1547                 if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1548                     radeon_get_ib_value(p, idx + 2) != 0) {
1549                         r = r600_cs_packet_next_reloc(p, &reloc);
1550                         if (r) {
1551                                 DRM_ERROR("bad SURFACE_SYNC\n");
1552                                 return -EINVAL;
1553                         }
1554                         ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1555                 }
1556                 break;
1557         case PACKET3_EVENT_WRITE:
1558                 if (pkt->count != 2 && pkt->count != 0) {
1559                         DRM_ERROR("bad EVENT_WRITE\n");
1560                         return -EINVAL;
1561                 }
1562                 if (pkt->count) {
1563                         r = r600_cs_packet_next_reloc(p, &reloc);
1564                         if (r) {
1565                                 DRM_ERROR("bad EVENT_WRITE\n");
1566                                 return -EINVAL;
1567                         }
1568                         ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1569                         ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1570                 }
1571                 break;
1572         case PACKET3_EVENT_WRITE_EOP:
1573                 if (pkt->count != 4) {
1574                         DRM_ERROR("bad EVENT_WRITE_EOP\n");
1575                         return -EINVAL;
1576                 }
1577                 r = r600_cs_packet_next_reloc(p, &reloc);
1578                 if (r) {
1579                         DRM_ERROR("bad EVENT_WRITE\n");
1580                         return -EINVAL;
1581                 }
1582                 ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1583                 ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1584                 break;
1585         case PACKET3_SET_CONFIG_REG:
1586                 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1587                 end_reg = 4 * pkt->count + start_reg - 4;
1588                 if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1589                     (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1590                     (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1591                         DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1592                         return -EINVAL;
1593                 }
1594                 for (i = 0; i < pkt->count; i++) {
1595                         reg = start_reg + (4 * i);
1596                         r = r600_cs_check_reg(p, reg, idx+1+i);
1597                         if (r)
1598                                 return r;
1599                 }
1600                 break;
1601         case PACKET3_SET_CONTEXT_REG:
1602                 start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1603                 end_reg = 4 * pkt->count + start_reg - 4;
1604                 if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1605                     (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1606                     (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1607                         DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1608                         return -EINVAL;
1609                 }
1610                 for (i = 0; i < pkt->count; i++) {
1611                         reg = start_reg + (4 * i);
1612                         r = r600_cs_check_reg(p, reg, idx+1+i);
1613                         if (r)
1614                                 return r;
1615                 }
1616                 break;
1617         case PACKET3_SET_RESOURCE:
1618                 if (pkt->count % 7) {
1619                         DRM_ERROR("bad SET_RESOURCE\n");
1620                         return -EINVAL;
1621                 }
1622                 start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1623                 end_reg = 4 * pkt->count + start_reg - 4;
1624                 if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1625                     (start_reg >= PACKET3_SET_RESOURCE_END) ||
1626                     (end_reg >= PACKET3_SET_RESOURCE_END)) {
1627                         DRM_ERROR("bad SET_RESOURCE\n");
1628                         return -EINVAL;
1629                 }
1630                 for (i = 0; i < (pkt->count / 7); i++) {
1631                         struct radeon_bo *texture, *mipmap;
1632                         u32 size, offset, base_offset, mip_offset;
1633
1634                         switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1635                         case SQ_TEX_VTX_VALID_TEXTURE:
1636                                 /* tex base */
1637                                 r = r600_cs_packet_next_reloc(p, &reloc);
1638                                 if (r) {
1639                                         DRM_ERROR("bad SET_RESOURCE\n");
1640                                         return -EINVAL;
1641                                 }
1642                                 base_offset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1643                                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1644                                         ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1645                                 else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1646                                         ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1647                                 texture = reloc->robj;
1648                                 /* tex mip base */
1649                                 r = r600_cs_packet_next_reloc(p, &reloc);
1650                                 if (r) {
1651                                         DRM_ERROR("bad SET_RESOURCE\n");
1652                                         return -EINVAL;
1653                                 }
1654                                 mip_offset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1655                                 mipmap = reloc->robj;
1656                                 r = r600_check_texture_resource(p,  idx+(i*7)+1,
1657                                                                 texture, mipmap,
1658                                                                 base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1659                                                                 mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1660                                                                 reloc->lobj.tiling_flags);
1661                                 if (r)
1662                                         return r;
1663                                 ib[idx+1+(i*7)+2] += base_offset;
1664                                 ib[idx+1+(i*7)+3] += mip_offset;
1665                                 break;
1666                         case SQ_TEX_VTX_VALID_BUFFER:
1667                                 /* vtx base */
1668                                 r = r600_cs_packet_next_reloc(p, &reloc);
1669                                 if (r) {
1670                                         DRM_ERROR("bad SET_RESOURCE\n");
1671                                         return -EINVAL;
1672                                 }
1673                                 offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
1674                                 size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
1675                                 if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
1676                                         /* force size to size of the buffer */
1677                                         dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
1678                                                  size + offset, radeon_bo_size(reloc->robj));
1679                                         ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj);
1680                                 }
1681                                 ib[idx+1+(i*7)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
1682                                 ib[idx+1+(i*7)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1683                                 break;
1684                         case SQ_TEX_VTX_INVALID_TEXTURE:
1685                         case SQ_TEX_VTX_INVALID_BUFFER:
1686                         default:
1687                                 DRM_ERROR("bad SET_RESOURCE\n");
1688                                 return -EINVAL;
1689                         }
1690                 }
1691                 break;
1692         case PACKET3_SET_ALU_CONST:
1693                 if (track->sq_config & DX9_CONSTS) {
1694                         start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
1695                         end_reg = 4 * pkt->count + start_reg - 4;
1696                         if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
1697                             (start_reg >= PACKET3_SET_ALU_CONST_END) ||
1698                             (end_reg >= PACKET3_SET_ALU_CONST_END)) {
1699                                 DRM_ERROR("bad SET_ALU_CONST\n");
1700                                 return -EINVAL;
1701                         }
1702                 }
1703                 break;
1704         case PACKET3_SET_BOOL_CONST:
1705                 start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
1706                 end_reg = 4 * pkt->count + start_reg - 4;
1707                 if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
1708                     (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
1709                     (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
1710                         DRM_ERROR("bad SET_BOOL_CONST\n");
1711                         return -EINVAL;
1712                 }
1713                 break;
1714         case PACKET3_SET_LOOP_CONST:
1715                 start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
1716                 end_reg = 4 * pkt->count + start_reg - 4;
1717                 if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
1718                     (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
1719                     (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
1720                         DRM_ERROR("bad SET_LOOP_CONST\n");
1721                         return -EINVAL;
1722                 }
1723                 break;
1724         case PACKET3_SET_CTL_CONST:
1725                 start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
1726                 end_reg = 4 * pkt->count + start_reg - 4;
1727                 if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
1728                     (start_reg >= PACKET3_SET_CTL_CONST_END) ||
1729                     (end_reg >= PACKET3_SET_CTL_CONST_END)) {
1730                         DRM_ERROR("bad SET_CTL_CONST\n");
1731                         return -EINVAL;
1732                 }
1733                 break;
1734         case PACKET3_SET_SAMPLER:
1735                 if (pkt->count % 3) {
1736                         DRM_ERROR("bad SET_SAMPLER\n");
1737                         return -EINVAL;
1738                 }
1739                 start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
1740                 end_reg = 4 * pkt->count + start_reg - 4;
1741                 if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
1742                     (start_reg >= PACKET3_SET_SAMPLER_END) ||
1743                     (end_reg >= PACKET3_SET_SAMPLER_END)) {
1744                         DRM_ERROR("bad SET_SAMPLER\n");
1745                         return -EINVAL;
1746                 }
1747                 break;
1748         case PACKET3_SURFACE_BASE_UPDATE:
1749                 if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
1750                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1751                         return -EINVAL;
1752                 }
1753                 if (pkt->count) {
1754                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1755                         return -EINVAL;
1756                 }
1757                 break;
1758         case PACKET3_NOP:
1759                 break;
1760         default:
1761                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1762                 return -EINVAL;
1763         }
1764         return 0;
1765 }
1766
1767 int r600_cs_parse(struct radeon_cs_parser *p)
1768 {
1769         struct radeon_cs_packet pkt;
1770         struct r600_cs_track *track;
1771         int r;
1772
1773         if (p->track == NULL) {
1774                 /* initialize tracker, we are in kms */
1775                 track = kzalloc(sizeof(*track), GFP_KERNEL);
1776                 if (track == NULL)
1777                         return -ENOMEM;
1778                 r600_cs_track_init(track);
1779                 if (p->rdev->family < CHIP_RV770) {
1780                         track->npipes = p->rdev->config.r600.tiling_npipes;
1781                         track->nbanks = p->rdev->config.r600.tiling_nbanks;
1782                         track->group_size = p->rdev->config.r600.tiling_group_size;
1783                 } else if (p->rdev->family <= CHIP_RV740) {
1784                         track->npipes = p->rdev->config.rv770.tiling_npipes;
1785                         track->nbanks = p->rdev->config.rv770.tiling_nbanks;
1786                         track->group_size = p->rdev->config.rv770.tiling_group_size;
1787                 }
1788                 p->track = track;
1789         }
1790         do {
1791                 r = r600_cs_packet_parse(p, &pkt, p->idx);
1792                 if (r) {
1793                         kfree(p->track);
1794                         p->track = NULL;
1795                         return r;
1796                 }
1797                 p->idx += pkt.count + 2;
1798                 switch (pkt.type) {
1799                 case PACKET_TYPE0:
1800                         r = r600_cs_parse_packet0(p, &pkt);
1801                         break;
1802                 case PACKET_TYPE2:
1803                         break;
1804                 case PACKET_TYPE3:
1805                         r = r600_packet3_check(p, &pkt);
1806                         break;
1807                 default:
1808                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1809                         kfree(p->track);
1810                         p->track = NULL;
1811                         return -EINVAL;
1812                 }
1813                 if (r) {
1814                         kfree(p->track);
1815                         p->track = NULL;
1816                         return r;
1817                 }
1818         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1819 #if 0
1820         for (r = 0; r < p->ib->length_dw; r++) {
1821                 printk(KERN_INFO "%05d  0x%08X\n", r, p->ib->ptr[r]);
1822                 mdelay(1);
1823         }
1824 #endif
1825         kfree(p->track);
1826         p->track = NULL;
1827         return 0;
1828 }
1829
1830 static int r600_cs_parser_relocs_legacy(struct radeon_cs_parser *p)
1831 {
1832         if (p->chunk_relocs_idx == -1) {
1833                 return 0;
1834         }
1835         p->relocs = kzalloc(sizeof(struct radeon_cs_reloc), GFP_KERNEL);
1836         if (p->relocs == NULL) {
1837                 return -ENOMEM;
1838         }
1839         return 0;
1840 }
1841
1842 /**
1843  * cs_parser_fini() - clean parser states
1844  * @parser:     parser structure holding parsing context.
1845  * @error:      error number
1846  *
1847  * If error is set than unvalidate buffer, otherwise just free memory
1848  * used by parsing context.
1849  **/
1850 static void r600_cs_parser_fini(struct radeon_cs_parser *parser, int error)
1851 {
1852         unsigned i;
1853
1854         kfree(parser->relocs);
1855         for (i = 0; i < parser->nchunks; i++) {
1856                 kfree(parser->chunks[i].kdata);
1857                 kfree(parser->chunks[i].kpage[0]);
1858                 kfree(parser->chunks[i].kpage[1]);
1859         }
1860         kfree(parser->chunks);
1861         kfree(parser->chunks_array);
1862 }
1863
1864 int r600_cs_legacy(struct drm_device *dev, void *data, struct drm_file *filp,
1865                         unsigned family, u32 *ib, int *l)
1866 {
1867         struct radeon_cs_parser parser;
1868         struct radeon_cs_chunk *ib_chunk;
1869         struct radeon_ib fake_ib;
1870         struct r600_cs_track *track;
1871         int r;
1872
1873         /* initialize tracker */
1874         track = kzalloc(sizeof(*track), GFP_KERNEL);
1875         if (track == NULL)
1876                 return -ENOMEM;
1877         r600_cs_track_init(track);
1878         r600_cs_legacy_get_tiling_conf(dev, &track->npipes, &track->nbanks, &track->group_size);
1879         /* initialize parser */
1880         memset(&parser, 0, sizeof(struct radeon_cs_parser));
1881         parser.filp = filp;
1882         parser.dev = &dev->pdev->dev;
1883         parser.rdev = NULL;
1884         parser.family = family;
1885         parser.ib = &fake_ib;
1886         parser.track = track;
1887         fake_ib.ptr = ib;
1888         r = radeon_cs_parser_init(&parser, data);
1889         if (r) {
1890                 DRM_ERROR("Failed to initialize parser !\n");
1891                 r600_cs_parser_fini(&parser, r);
1892                 return r;
1893         }
1894         r = r600_cs_parser_relocs_legacy(&parser);
1895         if (r) {
1896                 DRM_ERROR("Failed to parse relocation !\n");
1897                 r600_cs_parser_fini(&parser, r);
1898                 return r;
1899         }
1900         /* Copy the packet into the IB, the parser will read from the
1901          * input memory (cached) and write to the IB (which can be
1902          * uncached). */
1903         ib_chunk = &parser.chunks[parser.chunk_ib_idx];
1904         parser.ib->length_dw = ib_chunk->length_dw;
1905         *l = parser.ib->length_dw;
1906         r = r600_cs_parse(&parser);
1907         if (r) {
1908                 DRM_ERROR("Invalid command stream !\n");
1909                 r600_cs_parser_fini(&parser, r);
1910                 return r;
1911         }
1912         r = radeon_cs_finish_pages(&parser);
1913         if (r) {
1914                 DRM_ERROR("Invalid command stream !\n");
1915                 r600_cs_parser_fini(&parser, r);
1916                 return r;
1917         }
1918         r600_cs_parser_fini(&parser, r);
1919         return r;
1920 }
1921
1922 void r600_cs_legacy_init(void)
1923 {
1924         r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;
1925 }