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 <drm/drmP.h>
  30 #include "radeon.h"
  31 #include "radeon_asic.h"
  32 #include "r600d.h"
  33 #include "r600_reg_safe.h"
  34
  35 static int r600_nomm;
  36 extern void r600_cs_legacy_get_tiling_conf(struct drm_device *dev, u32 *npipes, u32 *nbanks, u32 *group_size);
  37
  38
  39 struct r600_cs_track {
  40         /* configuration we miror so that we use same code btw kms/ums */
  41         u32                     group_size;
  42         u32                     nbanks;
  43         u32                     npipes;
  44         /* value we track */
  45         u32                     sq_config;
  46         u32                     log_nsamples;
  47         u32                     nsamples;
  48         u32                     cb_color_base_last[8];
  49         struct radeon_bo        *cb_color_bo[8];
  50         u64                     cb_color_bo_mc[8];
  51         u64                     cb_color_bo_offset[8];
  52         struct radeon_bo        *cb_color_frag_bo[8];
  53         u64                     cb_color_frag_offset[8];
  54         struct radeon_bo        *cb_color_tile_bo[8];
  55         u64                     cb_color_tile_offset[8];
  56         u32                     cb_color_mask[8];
  57         u32                     cb_color_info[8];
  58         u32                     cb_color_view[8];
  59         u32                     cb_color_size_idx[8]; /* unused */
  60         u32                     cb_target_mask;
  61         u32                     cb_shader_mask;  /* unused */
  62         bool                    is_resolve;
  63         u32                     cb_color_size[8];
  64         u32                     vgt_strmout_en;
  65         u32                     vgt_strmout_buffer_en;
  66         struct radeon_bo        *vgt_strmout_bo[4];
  67         u64                     vgt_strmout_bo_mc[4]; /* unused */
  68         u32                     vgt_strmout_bo_offset[4];
  69         u32                     vgt_strmout_size[4];
  70         u32                     db_depth_control;
  71         u32                     db_depth_info;
  72         u32                     db_depth_size_idx;
  73         u32                     db_depth_view;
  74         u32                     db_depth_size;
  75         u32                     db_offset;
  76         struct radeon_bo        *db_bo;
  77         u64                     db_bo_mc;
  78         bool                    sx_misc_kill_all_prims;
  79         bool                    cb_dirty;
  80         bool                    db_dirty;
  81         bool                    streamout_dirty;
  82         struct radeon_bo        *htile_bo;
  83         u64                     htile_offset;
  84         u32                     htile_surface;
  85 };
  86
  87 #define FMT_8_BIT(fmt, vc)   [fmt] = { 1, 1, 1, vc, CHIP_R600 }
  88 #define FMT_16_BIT(fmt, vc)  [fmt] = { 1, 1, 2, vc, CHIP_R600 }
  89 #define FMT_24_BIT(fmt)      [fmt] = { 1, 1, 4,  0, CHIP_R600 }
  90 #define FMT_32_BIT(fmt, vc)  [fmt] = { 1, 1, 4, vc, CHIP_R600 }
  91 #define FMT_48_BIT(fmt)      [fmt] = { 1, 1, 8,  0, CHIP_R600 }
  92 #define FMT_64_BIT(fmt, vc)  [fmt] = { 1, 1, 8, vc, CHIP_R600 }
  93 #define FMT_96_BIT(fmt)      [fmt] = { 1, 1, 12, 0, CHIP_R600 }
  94 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16,vc, CHIP_R600 }
  95
  96 struct gpu_formats {
  97         unsigned blockwidth;
  98         unsigned blockheight;
  99         unsigned blocksize;
 100         unsigned valid_color;
 101         enum radeon_family min_family;
 102 };
 103
 104 static const struct gpu_formats color_formats_table[] = {
 105         /* 8 bit */
 106         FMT_8_BIT(V_038004_COLOR_8, 1),
 107         FMT_8_BIT(V_038004_COLOR_4_4, 1),
 108         FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
 109         FMT_8_BIT(V_038004_FMT_1, 0),
 110
 111         /* 16-bit */
 112         FMT_16_BIT(V_038004_COLOR_16, 1),
 113         FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
 114         FMT_16_BIT(V_038004_COLOR_8_8, 1),
 115         FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
 116         FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
 117         FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
 118         FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
 119         FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
 120
 121         /* 24-bit */
 122         FMT_24_BIT(V_038004_FMT_8_8_8),
 123
 124         /* 32-bit */
 125         FMT_32_BIT(V_038004_COLOR_32, 1),
 126         FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
 127         FMT_32_BIT(V_038004_COLOR_16_16, 1),
 128         FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
 129         FMT_32_BIT(V_038004_COLOR_8_24, 1),
 130         FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
 131         FMT_32_BIT(V_038004_COLOR_24_8, 1),
 132         FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
 133         FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
 134         FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
 135         FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
 136         FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
 137         FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
 138         FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
 139         FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
 140         FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
 141         FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
 142         FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
 143
 144         /* 48-bit */
 145         FMT_48_BIT(V_038004_FMT_16_16_16),
 146         FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
 147
 148         /* 64-bit */
 149         FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
 150         FMT_64_BIT(V_038004_COLOR_32_32, 1),
 151         FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
 152         FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
 153         FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
 154
 155         FMT_96_BIT(V_038004_FMT_32_32_32),
 156         FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
 157
 158         /* 128-bit */
 159         FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
 160         FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
 161
 162         [V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
 163         [V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
 164
 165         /* block compressed formats */
 166         [V_038004_FMT_BC1] = { 4, 4, 8, 0 },
 167         [V_038004_FMT_BC2] = { 4, 4, 16, 0 },
 168         [V_038004_FMT_BC3] = { 4, 4, 16, 0 },
 169         [V_038004_FMT_BC4] = { 4, 4, 8, 0 },
 170         [V_038004_FMT_BC5] = { 4, 4, 16, 0},
 171         [V_038004_FMT_BC6] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
 172         [V_038004_FMT_BC7] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
 173
 174         /* The other Evergreen formats */
 175         [V_038004_FMT_32_AS_32_32_32_32] = { 1, 1, 4, 0, CHIP_CEDAR},
 176 };
 177
 178 bool r600_fmt_is_valid_color(u32 format)
 179 {
 180         if (format >= ARRAY_SIZE(color_formats_table))
 181                 return false;
 182
 183         if (color_formats_table[format].valid_color)
 184                 return true;
 185
 186         return false;
 187 }
 188
 189 bool r600_fmt_is_valid_texture(u32 format, enum radeon_family family)
 190 {
 191         if (format >= ARRAY_SIZE(color_formats_table))
 192                 return false;
 193
 194         if (family < color_formats_table[format].min_family)
 195                 return false;
 196
 197         if (color_formats_table[format].blockwidth > 0)
 198                 return true;
 199
 200         return false;
 201 }
 202
 203 int r600_fmt_get_blocksize(u32 format)
 204 {
 205         if (format >= ARRAY_SIZE(color_formats_table))
 206                 return 0;
 207
 208         return color_formats_table[format].blocksize;
 209 }
 210
 211 int r600_fmt_get_nblocksx(u32 format, u32 w)
 212 {
 213         unsigned bw;
 214
 215         if (format >= ARRAY_SIZE(color_formats_table))
 216                 return 0;
 217
 218         bw = color_formats_table[format].blockwidth;
 219         if (bw == 0)
 220                 return 0;
 221
 222         return (w + bw - 1) / bw;
 223 }
 224
 225 int r600_fmt_get_nblocksy(u32 format, u32 h)
 226 {
 227         unsigned bh;
 228
 229         if (format >= ARRAY_SIZE(color_formats_table))
 230                 return 0;
 231
 232         bh = color_formats_table[format].blockheight;
 233         if (bh == 0)
 234                 return 0;
 235
 236         return (h + bh - 1) / bh;
 237 }
 238
 239 struct array_mode_checker {
 240         int array_mode;
 241         u32 group_size;
 242         u32 nbanks;
 243         u32 npipes;
 244         u32 nsamples;
 245         u32 blocksize;
 246 };
 247
 248 /* returns alignment in pixels for pitch/height/depth and bytes for base */
 249 static int r600_get_array_mode_alignment(struct array_mode_checker *values,
 250                                                 u32 *pitch_align,
 251                                                 u32 *height_align,
 252                                                 u32 *depth_align,
 253                                                 u64 *base_align)
 254 {
 255         u32 tile_width = 8;
 256         u32 tile_height = 8;
 257         u32 macro_tile_width = values->nbanks;
 258         u32 macro_tile_height = values->npipes;
 259         u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
 260         u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
 261
 262         switch (values->array_mode) {
 263         case ARRAY_LINEAR_GENERAL:
 264                 /* technically tile_width/_height for pitch/height */
 265                 *pitch_align = 1; /* tile_width */
 266                 *height_align = 1; /* tile_height */
 267                 *depth_align = 1;
 268                 *base_align = 1;
 269                 break;
 270         case ARRAY_LINEAR_ALIGNED:
 271                 *pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
 272                 *height_align = 1;
 273                 *depth_align = 1;
 274                 *base_align = values->group_size;
 275                 break;
 276         case ARRAY_1D_TILED_THIN1:
 277                 *pitch_align = max((u32)tile_width,
 278                                    (u32)(values->group_size /
 279                                          (tile_height * values->blocksize * values->nsamples)));
 280                 *height_align = tile_height;
 281                 *depth_align = 1;
 282                 *base_align = values->group_size;
 283                 break;
 284         case ARRAY_2D_TILED_THIN1:
 285                 *pitch_align = max((u32)macro_tile_width * tile_width,
 286                                 (u32)((values->group_size * values->nbanks) /
 287                                 (values->blocksize * values->nsamples * tile_width)));
 288                 *height_align = macro_tile_height * tile_height;
 289                 *depth_align = 1;
 290                 *base_align = max(macro_tile_bytes,
 291                                   (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
 292                 break;
 293         default:
 294                 return -EINVAL;
 295         }
 296
 297         return 0;
 298 }
 299
 300 static void r600_cs_track_init(struct r600_cs_track *track)
 301 {
 302         int i;
 303
 304         /* assume DX9 mode */
 305         track->sq_config = DX9_CONSTS;
 306         for (i = 0; i < 8; i++) {
 307                 track->cb_color_base_last[i] = 0;
 308                 track->cb_color_size[i] = 0;
 309                 track->cb_color_size_idx[i] = 0;
 310                 track->cb_color_info[i] = 0;
 311                 track->cb_color_view[i] = 0xFFFFFFFF;
 312                 track->cb_color_bo[i] = NULL;
 313                 track->cb_color_bo_offset[i] = 0xFFFFFFFF;
 314                 track->cb_color_bo_mc[i] = 0xFFFFFFFF;
 315                 track->cb_color_frag_bo[i] = NULL;
 316                 track->cb_color_frag_offset[i] = 0xFFFFFFFF;
 317                 track->cb_color_tile_bo[i] = NULL;
 318                 track->cb_color_tile_offset[i] = 0xFFFFFFFF;
 319                 track->cb_color_mask[i] = 0xFFFFFFFF;
 320         }
 321         track->is_resolve = false;
 322         track->nsamples = 16;
 323         track->log_nsamples = 4;
 324         track->cb_target_mask = 0xFFFFFFFF;
 325         track->cb_shader_mask = 0xFFFFFFFF;
 326         track->cb_dirty = true;
 327         track->db_bo = NULL;
 328         track->db_bo_mc = 0xFFFFFFFF;
 329         /* assume the biggest format and that htile is enabled */
 330         track->db_depth_info = 7 | (1 << 25);
 331         track->db_depth_view = 0xFFFFC000;
 332         track->db_depth_size = 0xFFFFFFFF;
 333         track->db_depth_size_idx = 0;
 334         track->db_depth_control = 0xFFFFFFFF;
 335         track->db_dirty = true;
 336         track->htile_bo = NULL;
 337         track->htile_offset = 0xFFFFFFFF;
 338         track->htile_surface = 0;
 339
 340         for (i = 0; i < 4; i++) {
 341                 track->vgt_strmout_size[i] = 0;
 342                 track->vgt_strmout_bo[i] = NULL;
 343                 track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
 344                 track->vgt_strmout_bo_mc[i] = 0xFFFFFFFF;
 345         }
 346         track->streamout_dirty = true;
 347         track->sx_misc_kill_all_prims = false;
 348 }
 349
 350 static int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
 351 {
 352         struct r600_cs_track *track = p->track;
 353         u32 slice_tile_max, size, tmp;
 354         u32 height, height_align, pitch, pitch_align, depth_align;
 355         u64 base_offset, base_align;
 356         struct array_mode_checker array_check;
 357         volatile u32 *ib = p->ib.ptr;
 358         unsigned array_mode;
 359         u32 format;
 360         /* When resolve is used, the second colorbuffer has always 1 sample. */
 361         unsigned nsamples = track->is_resolve && i == 1 ? 1 : track->nsamples;
 362
 363         size = radeon_bo_size(track->cb_color_bo[i]) - track->cb_color_bo_offset[i];
 364         format = G_0280A0_FORMAT(track->cb_color_info[i]);
 365         if (!r600_fmt_is_valid_color(format)) {
 366                 dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
 367                          __func__, __LINE__, format,
 368                         i, track->cb_color_info[i]);
 369                 return -EINVAL;
 370         }
 371         /* pitch in pixels */
 372         pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
 373         slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
 374         slice_tile_max *= 64;
 375         height = slice_tile_max / pitch;
 376         if (height > 8192)
 377                 height = 8192;
 378         array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
 379
 380         base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
 381         array_check.array_mode = array_mode;
 382         array_check.group_size = track->group_size;
 383         array_check.nbanks = track->nbanks;
 384         array_check.npipes = track->npipes;
 385         array_check.nsamples = nsamples;
 386         array_check.blocksize = r600_fmt_get_blocksize(format);
 387         if (r600_get_array_mode_alignment(&array_check,
 388                                           &pitch_align, &height_align, &depth_align, &base_align)) {
 389                 dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
 390                          G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
 391                          track->cb_color_info[i]);
 392                 return -EINVAL;
 393         }
 394         switch (array_mode) {
 395         case V_0280A0_ARRAY_LINEAR_GENERAL:
 396                 break;
 397         case V_0280A0_ARRAY_LINEAR_ALIGNED:
 398                 break;
 399         case V_0280A0_ARRAY_1D_TILED_THIN1:
 400                 /* avoid breaking userspace */
 401                 if (height > 7)
 402                         height &= ~0x7;
 403                 break;
 404         case V_0280A0_ARRAY_2D_TILED_THIN1:
 405                 break;
 406         default:
 407                 dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
 408                         G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
 409                         track->cb_color_info[i]);
 410                 return -EINVAL;
 411         }
 412
 413         if (!IS_ALIGNED(pitch, pitch_align)) {
 414                 dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
 415                          __func__, __LINE__, pitch, pitch_align, array_mode);
 416                 return -EINVAL;
 417         }
 418         if (!IS_ALIGNED(height, height_align)) {
 419                 dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
 420                          __func__, __LINE__, height, height_align, array_mode);
 421                 return -EINVAL;
 422         }
 423         if (!IS_ALIGNED(base_offset, base_align)) {
 424                 dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
 425                          base_offset, base_align, array_mode);
 426                 return -EINVAL;
 427         }
 428
 429         /* check offset */
 430         tmp = r600_fmt_get_nblocksy(format, height) * r600_fmt_get_nblocksx(format, pitch) *
 431               r600_fmt_get_blocksize(format) * nsamples;
 432         switch (array_mode) {
 433         default:
 434         case V_0280A0_ARRAY_LINEAR_GENERAL:
 435         case V_0280A0_ARRAY_LINEAR_ALIGNED:
 436                 tmp += track->cb_color_view[i] & 0xFF;
 437                 break;
 438         case V_0280A0_ARRAY_1D_TILED_THIN1:
 439         case V_0280A0_ARRAY_2D_TILED_THIN1:
 440                 tmp += G_028080_SLICE_MAX(track->cb_color_view[i]) * tmp;
 441                 break;
 442         }
 443         if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
 444                 if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
 445                         /* the initial DDX does bad things with the CB size occasionally */
 446                         /* it rounds up height too far for slice tile max but the BO is smaller */
 447                         /* r600c,g also seem to flush at bad times in some apps resulting in
 448                          * bogus values here. So for linear just allow anything to avoid breaking
 449                          * broken userspace.
 450                          */
 451                 } else {
 452                         dev_warn(p->dev, "%s offset[%d] %d %llu %d %lu too big (%d %d) (%d %d %d)\n",
 453                                  __func__, i, array_mode,
 454                                  track->cb_color_bo_offset[i], tmp,
 455                                  radeon_bo_size(track->cb_color_bo[i]),
 456                                  pitch, height, r600_fmt_get_nblocksx(format, pitch),
 457                                  r600_fmt_get_nblocksy(format, height),
 458                                  r600_fmt_get_blocksize(format));
 459                         return -EINVAL;
 460                 }
 461         }
 462         /* limit max tile */
 463         tmp = (height * pitch) >> 6;
 464         if (tmp < slice_tile_max)
 465                 slice_tile_max = tmp;
 466         tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
 467                 S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
 468         ib[track->cb_color_size_idx[i]] = tmp;
 469
 470         /* FMASK/CMASK */
 471         switch (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
 472         case V_0280A0_TILE_DISABLE:
 473                 break;
 474         case V_0280A0_FRAG_ENABLE:
 475                 if (track->nsamples > 1) {
 476                         uint32_t tile_max = G_028100_FMASK_TILE_MAX(track->cb_color_mask[i]);
 477                         /* the tile size is 8x8, but the size is in units of bits.
 478                          * for bytes, do just * 8. */
 479                         uint32_t bytes = track->nsamples * track->log_nsamples * 8 * (tile_max + 1);
 480
 481                         if (bytes + track->cb_color_frag_offset[i] >
 482                             radeon_bo_size(track->cb_color_frag_bo[i])) {
 483                                 dev_warn(p->dev, "%s FMASK_TILE_MAX too large "
 484                                          "(tile_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
 485                                          __func__, tile_max, bytes,
 486                                          track->cb_color_frag_offset[i],
 487                                          radeon_bo_size(track->cb_color_frag_bo[i]));
 488                                 return -EINVAL;
 489                         }
 490                 }
 491                 /* fall through */
 492         case V_0280A0_CLEAR_ENABLE:
 493         {
 494                 uint32_t block_max = G_028100_CMASK_BLOCK_MAX(track->cb_color_mask[i]);
 495                 /* One block = 128x128 pixels, one 8x8 tile has 4 bits..
 496                  * (128*128) / (8*8) / 2 = 128 bytes per block. */
 497                 uint32_t bytes = (block_max + 1) * 128;
 498
 499                 if (bytes + track->cb_color_tile_offset[i] >
 500                     radeon_bo_size(track->cb_color_tile_bo[i])) {
 501                         dev_warn(p->dev, "%s CMASK_BLOCK_MAX too large "
 502                                  "(block_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
 503                                  __func__, block_max, bytes,
 504                                  track->cb_color_tile_offset[i],
 505                                  radeon_bo_size(track->cb_color_tile_bo[i]));
 506                         return -EINVAL;
 507                 }
 508                 break;
 509         }
 510         default:
 511                 dev_warn(p->dev, "%s invalid tile mode\n", __func__);
 512                 return -EINVAL;
 513         }
 514         return 0;
 515 }
 516
 517 static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
 518 {
 519         struct r600_cs_track *track = p->track;
 520         u32 nviews, bpe, ntiles, size, slice_tile_max, tmp;
 521         u32 height_align, pitch_align, depth_align;
 522         u32 pitch = 8192;
 523         u32 height = 8192;
 524         u64 base_offset, base_align;
 525         struct array_mode_checker array_check;
 526         int array_mode;
 527         volatile u32 *ib = p->ib.ptr;
 528
 529
 530         if (track->db_bo == NULL) {
 531                 dev_warn(p->dev, "z/stencil with no depth buffer\n");
 532                 return -EINVAL;
 533         }
 534         switch (G_028010_FORMAT(track->db_depth_info)) {
 535         case V_028010_DEPTH_16:
 536                 bpe = 2;
 537                 break;
 538         case V_028010_DEPTH_X8_24:
 539         case V_028010_DEPTH_8_24:
 540         case V_028010_DEPTH_X8_24_FLOAT:
 541         case V_028010_DEPTH_8_24_FLOAT:
 542         case V_028010_DEPTH_32_FLOAT:
 543                 bpe = 4;
 544                 break;
 545         case V_028010_DEPTH_X24_8_32_FLOAT:
 546                 bpe = 8;
 547                 break;
 548         default:
 549                 dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
 550                 return -EINVAL;
 551         }
 552         if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
 553                 if (!track->db_depth_size_idx) {
 554                         dev_warn(p->dev, "z/stencil buffer size not set\n");
 555                         return -EINVAL;
 556                 }
 557                 tmp = radeon_bo_size(track->db_bo) - track->db_offset;
 558                 tmp = (tmp / bpe) >> 6;
 559                 if (!tmp) {
 560                         dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
 561                                         track->db_depth_size, bpe, track->db_offset,
 562                                         radeon_bo_size(track->db_bo));
 563                         return -EINVAL;
 564                 }
 565                 ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
 566         } else {
 567                 size = radeon_bo_size(track->db_bo);
 568                 /* pitch in pixels */
 569                 pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
 570                 slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
 571                 slice_tile_max *= 64;
 572                 height = slice_tile_max / pitch;
 573                 if (height > 8192)
 574                         height = 8192;
 575                 base_offset = track->db_bo_mc + track->db_offset;
 576                 array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
 577                 array_check.array_mode = array_mode;
 578                 array_check.group_size = track->group_size;
 579                 array_check.nbanks = track->nbanks;
 580                 array_check.npipes = track->npipes;
 581                 array_check.nsamples = track->nsamples;
 582                 array_check.blocksize = bpe;
 583                 if (r600_get_array_mode_alignment(&array_check,
 584                                         &pitch_align, &height_align, &depth_align, &base_align)) {
 585                         dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
 586                                         G_028010_ARRAY_MODE(track->db_depth_info),
 587                                         track->db_depth_info);
 588                         return -EINVAL;
 589                 }
 590                 switch (array_mode) {
 591                 case V_028010_ARRAY_1D_TILED_THIN1:
 592                         /* don't break userspace */
 593                         height &= ~0x7;
 594                         break;
 595                 case V_028010_ARRAY_2D_TILED_THIN1:
 596                         break;
 597                 default:
 598                         dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
 599                                         G_028010_ARRAY_MODE(track->db_depth_info),
 600                                         track->db_depth_info);
 601                         return -EINVAL;
 602                 }
 603
 604                 if (!IS_ALIGNED(pitch, pitch_align)) {
 605                         dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
 606                                         __func__, __LINE__, pitch, pitch_align, array_mode);
 607                         return -EINVAL;
 608                 }
 609                 if (!IS_ALIGNED(height, height_align)) {
 610                         dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
 611                                         __func__, __LINE__, height, height_align, array_mode);
 612                         return -EINVAL;
 613                 }
 614                 if (!IS_ALIGNED(base_offset, base_align)) {
 615                         dev_warn(p->dev, "%s offset 0x%llx, 0x%llx, %d not aligned\n", __func__,
 616                                         base_offset, base_align, array_mode);
 617                         return -EINVAL;
 618                 }
 619
 620                 ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
 621                 nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
 622                 tmp = ntiles * bpe * 64 * nviews * track->nsamples;
 623                 if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
 624                         dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
 625                                         array_mode,
 626                                         track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
 627                                         radeon_bo_size(track->db_bo));
 628                         return -EINVAL;
 629                 }
 630         }
 631
 632         /* hyperz */
 633         if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
 634                 unsigned long size;
 635                 unsigned nbx, nby;
 636
 637                 if (track->htile_bo == NULL) {
 638                         dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
 639                                  __func__, __LINE__, track->db_depth_info);
 640                         return -EINVAL;
 641                 }
 642                 if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
 643                         dev_warn(p->dev, "%s:%d htile can't be enabled with bogus db_depth_size 0x%08x\n",
 644                                  __func__, __LINE__, track->db_depth_size);
 645                         return -EINVAL;
 646                 }
 647
 648                 nbx = pitch;
 649                 nby = height;
 650                 if (G_028D24_LINEAR(track->htile_surface)) {
 651                         /* nbx must be 16 htiles aligned == 16 * 8 pixel aligned */
 652                         nbx = round_up(nbx, 16 * 8);
 653                         /* nby is npipes htiles aligned == npipes * 8 pixel aligned */
 654                         nby = round_up(nby, track->npipes * 8);
 655                 } else {
 656                         /* always assume 8x8 htile */
 657                         /* align is htile align * 8, htile align vary according to
 658                          * number of pipe and tile width and nby
 659                          */
 660                         switch (track->npipes) {
 661                         case 8:
 662                                 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 663                                 nbx = round_up(nbx, 64 * 8);
 664                                 nby = round_up(nby, 64 * 8);
 665                                 break;
 666                         case 4:
 667                                 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 668                                 nbx = round_up(nbx, 64 * 8);
 669                                 nby = round_up(nby, 32 * 8);
 670                                 break;
 671                         case 2:
 672                                 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 673                                 nbx = round_up(nbx, 32 * 8);
 674                                 nby = round_up(nby, 32 * 8);
 675                                 break;
 676                         case 1:
 677                                 /* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 678                                 nbx = round_up(nbx, 32 * 8);
 679                                 nby = round_up(nby, 16 * 8);
 680                                 break;
 681                         default:
 682                                 dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
 683                                          __func__, __LINE__, track->npipes);
 684                                 return -EINVAL;
 685                         }
 686                 }
 687                 /* compute number of htile */
 688                 nbx = nbx >> 3;
 689                 nby = nby >> 3;
 690                 /* size must be aligned on npipes * 2K boundary */
 691                 size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
 692                 size += track->htile_offset;
 693
 694                 if (size > radeon_bo_size(track->htile_bo)) {
 695                         dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
 696                                  __func__, __LINE__, radeon_bo_size(track->htile_bo),
 697                                  size, nbx, nby);
 698                         return -EINVAL;
 699                 }
 700         }
 701
 702         track->db_dirty = false;
 703         return 0;
 704 }
 705
 706 static int r600_cs_track_check(struct radeon_cs_parser *p)
 707 {
 708         struct r600_cs_track *track = p->track;
 709         u32 tmp;
 710         int r, i;
 711
 712         /* on legacy kernel we don't perform advanced check */
 713         if (p->rdev == NULL)
 714                 return 0;
 715
 716         /* check streamout */
 717         if (track->streamout_dirty && track->vgt_strmout_en) {
 718                 for (i = 0; i < 4; i++) {
 719                         if (track->vgt_strmout_buffer_en & (1 << i)) {
 720                                 if (track->vgt_strmout_bo[i]) {
 721                                         u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
 722                                                 (u64)track->vgt_strmout_size[i];
 723                                         if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
 724                                                 DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
 725                                                           i, offset,
 726                                                           radeon_bo_size(track->vgt_strmout_bo[i]));
 727                                                 return -EINVAL;
 728                                         }
 729                                 } else {
 730                                         dev_warn(p->dev, "No buffer for streamout %d\n", i);
 731                                         return -EINVAL;
 732                                 }
 733                         }
 734                 }
 735                 track->streamout_dirty = false;
 736         }
 737
 738         if (track->sx_misc_kill_all_prims)
 739                 return 0;
 740
 741         /* check that we have a cb for each enabled target, we don't check
 742          * shader_mask because it seems mesa isn't always setting it :(
 743          */
 744         if (track->cb_dirty) {
 745                 tmp = track->cb_target_mask;
 746
 747                 /* We must check both colorbuffers for RESOLVE. */
 748                 if (track->is_resolve) {
 749                         tmp |= 0xff;
 750                 }
 751
 752                 for (i = 0; i < 8; i++) {
 753                         u32 format = G_0280A0_FORMAT(track->cb_color_info[i]);
 754
 755                         if (format != V_0280A0_COLOR_INVALID &&
 756                             (tmp >> (i * 4)) & 0xF) {
 757                                 /* at least one component is enabled */
 758                                 if (track->cb_color_bo[i] == NULL) {
 759                                         dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
 760                                                 __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
 761                                         return -EINVAL;
 762                                 }
 763                                 /* perform rewrite of CB_COLOR[0-7]_SIZE */
 764                                 r = r600_cs_track_validate_cb(p, i);
 765                                 if (r)
 766                                         return r;
 767                         }
 768                 }
 769                 track->cb_dirty = false;
 770         }
 771
 772         /* Check depth buffer */
 773         if (track->db_dirty &&
 774             G_028010_FORMAT(track->db_depth_info) != V_028010_DEPTH_INVALID &&
 775             (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
 776              G_028800_Z_ENABLE(track->db_depth_control))) {
 777                 r = r600_cs_track_validate_db(p);
 778                 if (r)
 779                         return r;
 780         }
 781
 782         return 0;
 783 }
 784
 785 /**
 786  * r600_cs_packet_parse_vline() - parse userspace VLINE packet
 787  * @parser:             parser structure holding parsing context.
 788  *
 789  * This is an R600-specific function for parsing VLINE packets.
 790  * Real work is done by r600_cs_common_vline_parse function.
 791  * Here we just set up ASIC-specific register table and call
 792  * the common implementation function.
 793  */
 794 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
 795 {
 796         static uint32_t vline_start_end[2] = {AVIVO_D1MODE_VLINE_START_END,
 797                                               AVIVO_D2MODE_VLINE_START_END};
 798         static uint32_t vline_status[2] = {AVIVO_D1MODE_VLINE_STATUS,
 799                                            AVIVO_D2MODE_VLINE_STATUS};
 800
 801         return r600_cs_common_vline_parse(p, vline_start_end, vline_status);
 802 }
 803
 804 /**
 805  * r600_cs_common_vline_parse() - common vline parser
 806  * @parser:             parser structure holding parsing context.
 807  * @vline_start_end:    table of vline_start_end registers
 808  * @vline_status:       table of vline_status registers
 809  *
 810  * Userspace sends a special sequence for VLINE waits.
 811  * PACKET0 - VLINE_START_END + value
 812  * PACKET3 - WAIT_REG_MEM poll vline status reg
 813  * RELOC (P3) - crtc_id in reloc.
 814  *
 815  * This function parses this and relocates the VLINE START END
 816  * and WAIT_REG_MEM packets to the correct crtc.
 817  * It also detects a switched off crtc and nulls out the
 818  * wait in that case. This function is common for all ASICs that
 819  * are R600 and newer. The parsing algorithm is the same, and only
 820  * differs in which registers are used.
 821  *
 822  * Caller is the ASIC-specific function which passes the parser
 823  * context and ASIC-specific register table
 824  */
 825 int r600_cs_common_vline_parse(struct radeon_cs_parser *p,
 826                                uint32_t *vline_start_end,
 827                                uint32_t *vline_status)
 828 {
 829         struct drm_crtc *crtc;
 830         struct radeon_crtc *radeon_crtc;
 831         struct radeon_cs_packet p3reloc, wait_reg_mem;
 832         int crtc_id;
 833         int r;
 834         uint32_t header, h_idx, reg, wait_reg_mem_info;
 835         volatile uint32_t *ib;
 836
 837         ib = p->ib.ptr;
 838
 839         /* parse the WAIT_REG_MEM */
 840         r = radeon_cs_packet_parse(p, &wait_reg_mem, p->idx);
 841         if (r)
 842                 return r;
 843
 844         /* check its a WAIT_REG_MEM */
 845         if (wait_reg_mem.type != RADEON_PACKET_TYPE3 ||
 846             wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
 847                 DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
 848                 return -EINVAL;
 849         }
 850
 851         wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
 852         /* bit 4 is reg (0) or mem (1) */
 853         if (wait_reg_mem_info & 0x10) {
 854                 DRM_ERROR("vline WAIT_REG_MEM waiting on MEM instead of REG\n");
 855                 return -EINVAL;
 856         }
 857         /* bit 8 is me (0) or pfp (1) */
 858         if (wait_reg_mem_info & 0x100) {
 859                 DRM_ERROR("vline WAIT_REG_MEM waiting on PFP instead of ME\n");
 860                 return -EINVAL;
 861         }
 862         /* waiting for value to be equal */
 863         if ((wait_reg_mem_info & 0x7) != 0x3) {
 864                 DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
 865                 return -EINVAL;
 866         }
 867         if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != vline_status[0]) {
 868                 DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
 869                 return -EINVAL;
 870         }
 871
 872         if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != RADEON_VLINE_STAT) {
 873                 DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
 874                 return -EINVAL;
 875         }
 876
 877         /* jump over the NOP */
 878         r = radeon_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
 879         if (r)
 880                 return r;
 881
 882         h_idx = p->idx - 2;
 883         p->idx += wait_reg_mem.count + 2;
 884         p->idx += p3reloc.count + 2;
 885
 886         header = radeon_get_ib_value(p, h_idx);
 887         crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
 888         reg = R600_CP_PACKET0_GET_REG(header);
 889
 890         crtc = drm_crtc_find(p->rdev->ddev, crtc_id);
 891         if (!crtc) {
 892                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 893                 return -ENOENT;
 894         }
 895         radeon_crtc = to_radeon_crtc(crtc);
 896         crtc_id = radeon_crtc->crtc_id;
 897
 898         if (!crtc->enabled) {
 899                 /* CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
 900                 ib[h_idx + 2] = PACKET2(0);
 901                 ib[h_idx + 3] = PACKET2(0);
 902                 ib[h_idx + 4] = PACKET2(0);
 903                 ib[h_idx + 5] = PACKET2(0);
 904                 ib[h_idx + 6] = PACKET2(0);
 905                 ib[h_idx + 7] = PACKET2(0);
 906                 ib[h_idx + 8] = PACKET2(0);
 907         } else if (reg == vline_start_end[0]) {
 908                 header &= ~R600_CP_PACKET0_REG_MASK;
 909                 header |= vline_start_end[crtc_id] >> 2;
 910                 ib[h_idx] = header;
 911                 ib[h_idx + 4] = vline_status[crtc_id] >> 2;
 912         } else {
 913                 DRM_ERROR("unknown crtc reloc\n");
 914                 return -EINVAL;
 915         }
 916         return 0;
 917 }
 918
 919 static int r600_packet0_check(struct radeon_cs_parser *p,
 920                                 struct radeon_cs_packet *pkt,
 921                                 unsigned idx, unsigned reg)
 922 {
 923         int r;
 924
 925         switch (reg) {
 926         case AVIVO_D1MODE_VLINE_START_END:
 927                 r = r600_cs_packet_parse_vline(p);
 928                 if (r) {
 929                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 930                                         idx, reg);
 931                         return r;
 932                 }
 933                 break;
 934         default:
 935                 pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
 936                 return -EINVAL;
 937         }
 938         return 0;
 939 }
 940
 941 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
 942                                 struct radeon_cs_packet *pkt)
 943 {
 944         unsigned reg, i;
 945         unsigned idx;
 946         int r;
 947
 948         idx = pkt->idx + 1;
 949         reg = pkt->reg;
 950         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 951                 r = r600_packet0_check(p, pkt, idx, reg);
 952                 if (r) {
 953                         return r;
 954                 }
 955         }
 956         return 0;
 957 }
 958
 959 /**
 960  * r600_cs_check_reg() - check if register is authorized or not
 961  * @parser: parser structure holding parsing context
 962  * @reg: register we are testing
 963  * @idx: index into the cs buffer
 964  *
 965  * This function will test against r600_reg_safe_bm and return 0
 966  * if register is safe. If register is not flag as safe this function
 967  * will test it against a list of register needind special handling.
 968  */
 969 static int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
 970 {
 971         struct r600_cs_track *track = (struct r600_cs_track *)p->track;
 972         struct radeon_bo_list *reloc;
 973         u32 m, i, tmp, *ib;
 974         int r;
 975
 976         i = (reg >> 7);
 977         if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
 978                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
 979                 return -EINVAL;
 980         }
 981         m = 1 << ((reg >> 2) & 31);
 982         if (!(r600_reg_safe_bm[i] & m))
 983                 return 0;
 984         ib = p->ib.ptr;
 985         switch (reg) {
 986         /* force following reg to 0 in an attempt to disable out buffer
 987          * which will need us to better understand how it works to perform
 988          * security check on it (Jerome)
 989          */
 990         case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
 991         case R_008C44_SQ_ESGS_RING_SIZE:
 992         case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
 993         case R_008C54_SQ_ESTMP_RING_SIZE:
 994         case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
 995         case R_008C74_SQ_FBUF_RING_SIZE:
 996         case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
 997         case R_008C5C_SQ_GSTMP_RING_SIZE:
 998         case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
 999         case R_008C4C_SQ_GSVS_RING_SIZE:
1000         case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
1001         case R_008C6C_SQ_PSTMP_RING_SIZE:
1002         case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
1003         case R_008C7C_SQ_REDUC_RING_SIZE:
1004         case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
1005         case R_008C64_SQ_VSTMP_RING_SIZE:
1006         case R_0288C8_SQ_GS_VERT_ITEMSIZE:
1007                 /* get value to populate the IB don't remove */
1008                 /*tmp =radeon_get_ib_value(p, idx);
1009                   ib[idx] = 0;*/
1010                 break;
1011         case SQ_ESGS_RING_BASE:
1012         case SQ_GSVS_RING_BASE:
1013         case SQ_ESTMP_RING_BASE:
1014         case SQ_GSTMP_RING_BASE:
1015         case SQ_PSTMP_RING_BASE:
1016         case SQ_VSTMP_RING_BASE:
1017                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1018                 if (r) {
1019                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1020                                         "0x%04X\n", reg);
1021                         return -EINVAL;
1022                 }
1023                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1024                 break;
1025         case SQ_CONFIG:
1026                 track->sq_config = radeon_get_ib_value(p, idx);
1027                 break;
1028         case R_028800_DB_DEPTH_CONTROL:
1029                 track->db_depth_control = radeon_get_ib_value(p, idx);
1030                 track->db_dirty = true;
1031                 break;
1032         case R_028010_DB_DEPTH_INFO:
1033                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1034                     radeon_cs_packet_next_is_pkt3_nop(p)) {
1035                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1036                         if (r) {
1037                                 dev_warn(p->dev, "bad SET_CONTEXT_REG "
1038                                          "0x%04X\n", reg);
1039                                 return -EINVAL;
1040                         }
1041                         track->db_depth_info = radeon_get_ib_value(p, idx);
1042                         ib[idx] &= C_028010_ARRAY_MODE;
1043                         track->db_depth_info &= C_028010_ARRAY_MODE;
1044                         if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1045                                 ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1046                                 track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1047                         } else {
1048                                 ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1049                                 track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1050                         }
1051                 } else {
1052                         track->db_depth_info = radeon_get_ib_value(p, idx);
1053                 }
1054                 track->db_dirty = true;
1055                 break;
1056         case R_028004_DB_DEPTH_VIEW:
1057                 track->db_depth_view = radeon_get_ib_value(p, idx);
1058                 track->db_dirty = true;
1059                 break;
1060         case R_028000_DB_DEPTH_SIZE:
1061                 track->db_depth_size = radeon_get_ib_value(p, idx);
1062                 track->db_depth_size_idx = idx;
1063                 track->db_dirty = true;
1064                 break;
1065         case R_028AB0_VGT_STRMOUT_EN:
1066                 track->vgt_strmout_en = radeon_get_ib_value(p, idx);
1067                 track->streamout_dirty = true;
1068                 break;
1069         case R_028B20_VGT_STRMOUT_BUFFER_EN:
1070                 track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
1071                 track->streamout_dirty = true;
1072                 break;
1073         case VGT_STRMOUT_BUFFER_BASE_0:
1074         case VGT_STRMOUT_BUFFER_BASE_1:
1075         case VGT_STRMOUT_BUFFER_BASE_2:
1076         case VGT_STRMOUT_BUFFER_BASE_3:
1077                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1078                 if (r) {
1079                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1080                                         "0x%04X\n", reg);
1081                         return -EINVAL;
1082                 }
1083                 tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1084                 track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1085                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1086                 track->vgt_strmout_bo[tmp] = reloc->robj;
1087                 track->vgt_strmout_bo_mc[tmp] = reloc->gpu_offset;
1088                 track->streamout_dirty = true;
1089                 break;
1090         case VGT_STRMOUT_BUFFER_SIZE_0:
1091         case VGT_STRMOUT_BUFFER_SIZE_1:
1092         case VGT_STRMOUT_BUFFER_SIZE_2:
1093         case VGT_STRMOUT_BUFFER_SIZE_3:
1094                 tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1095                 /* size in register is DWs, convert to bytes */
1096                 track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1097                 track->streamout_dirty = true;
1098                 break;
1099         case CP_COHER_BASE:
1100                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1101                 if (r) {
1102                         dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1103                                         "0x%04X\n", reg);
1104                         return -EINVAL;
1105                 }
1106                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1107                 break;
1108         case R_028238_CB_TARGET_MASK:
1109                 track->cb_target_mask = radeon_get_ib_value(p, idx);
1110                 track->cb_dirty = true;
1111                 break;
1112         case R_02823C_CB_SHADER_MASK:
1113                 track->cb_shader_mask = radeon_get_ib_value(p, idx);
1114                 break;
1115         case R_028C04_PA_SC_AA_CONFIG:
1116                 tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
1117                 track->log_nsamples = tmp;
1118                 track->nsamples = 1 << tmp;
1119                 track->cb_dirty = true;
1120                 break;
1121         case R_028808_CB_COLOR_CONTROL:
1122                 tmp = G_028808_SPECIAL_OP(radeon_get_ib_value(p, idx));
1123                 track->is_resolve = tmp == V_028808_SPECIAL_RESOLVE_BOX;
1124                 track->cb_dirty = true;
1125                 break;
1126         case R_0280A0_CB_COLOR0_INFO:
1127         case R_0280A4_CB_COLOR1_INFO:
1128         case R_0280A8_CB_COLOR2_INFO:
1129         case R_0280AC_CB_COLOR3_INFO:
1130         case R_0280B0_CB_COLOR4_INFO:
1131         case R_0280B4_CB_COLOR5_INFO:
1132         case R_0280B8_CB_COLOR6_INFO:
1133         case R_0280BC_CB_COLOR7_INFO:
1134                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1135                      radeon_cs_packet_next_is_pkt3_nop(p)) {
1136                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1137                         if (r) {
1138                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1139                                 return -EINVAL;
1140                         }
1141                         tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1142                         track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1143                         if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1144                                 ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1145                                 track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1146                         } else if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1147                                 ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1148                                 track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1149                         }
1150                 } else {
1151                         tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1152                         track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1153                 }
1154                 track->cb_dirty = true;
1155                 break;
1156         case R_028080_CB_COLOR0_VIEW:
1157         case R_028084_CB_COLOR1_VIEW:
1158         case R_028088_CB_COLOR2_VIEW:
1159         case R_02808C_CB_COLOR3_VIEW:
1160         case R_028090_CB_COLOR4_VIEW:
1161         case R_028094_CB_COLOR5_VIEW:
1162         case R_028098_CB_COLOR6_VIEW:
1163         case R_02809C_CB_COLOR7_VIEW:
1164                 tmp = (reg - R_028080_CB_COLOR0_VIEW) / 4;
1165                 track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1166                 track->cb_dirty = true;
1167                 break;
1168         case R_028060_CB_COLOR0_SIZE:
1169         case R_028064_CB_COLOR1_SIZE:
1170         case R_028068_CB_COLOR2_SIZE:
1171         case R_02806C_CB_COLOR3_SIZE:
1172         case R_028070_CB_COLOR4_SIZE:
1173         case R_028074_CB_COLOR5_SIZE:
1174         case R_028078_CB_COLOR6_SIZE:
1175         case R_02807C_CB_COLOR7_SIZE:
1176                 tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1177                 track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1178                 track->cb_color_size_idx[tmp] = idx;
1179                 track->cb_dirty = true;
1180                 break;
1181                 /* This register were added late, there is userspace
1182                  * which does provide relocation for those but set
1183                  * 0 offset. In order to avoid breaking old userspace
1184                  * we detect this and set address to point to last
1185                  * CB_COLOR0_BASE, note that if userspace doesn't set
1186                  * CB_COLOR0_BASE before this register we will report
1187                  * error. Old userspace always set CB_COLOR0_BASE
1188                  * before any of this.
1189                  */
1190         case R_0280E0_CB_COLOR0_FRAG:
1191         case R_0280E4_CB_COLOR1_FRAG:
1192         case R_0280E8_CB_COLOR2_FRAG:
1193         case R_0280EC_CB_COLOR3_FRAG:
1194         case R_0280F0_CB_COLOR4_FRAG:
1195         case R_0280F4_CB_COLOR5_FRAG:
1196         case R_0280F8_CB_COLOR6_FRAG:
1197         case R_0280FC_CB_COLOR7_FRAG:
1198                 tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1199                 if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1200                         if (!track->cb_color_base_last[tmp]) {
1201                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1202                                 return -EINVAL;
1203                         }
1204                         track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1205                         track->cb_color_frag_offset[tmp] = track->cb_color_bo_offset[tmp];
1206                         ib[idx] = track->cb_color_base_last[tmp];
1207                 } else {
1208                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1209                         if (r) {
1210                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1211                                 return -EINVAL;
1212                         }
1213                         track->cb_color_frag_bo[tmp] = reloc->robj;
1214                         track->cb_color_frag_offset[tmp] = (u64)ib[idx] << 8;
1215                         ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1216                 }
1217                 if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1218                         track->cb_dirty = true;
1219                 }
1220                 break;
1221         case R_0280C0_CB_COLOR0_TILE:
1222         case R_0280C4_CB_COLOR1_TILE:
1223         case R_0280C8_CB_COLOR2_TILE:
1224         case R_0280CC_CB_COLOR3_TILE:
1225         case R_0280D0_CB_COLOR4_TILE:
1226         case R_0280D4_CB_COLOR5_TILE:
1227         case R_0280D8_CB_COLOR6_TILE:
1228         case R_0280DC_CB_COLOR7_TILE:
1229                 tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1230                 if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1231                         if (!track->cb_color_base_last[tmp]) {
1232                                 dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1233                                 return -EINVAL;
1234                         }
1235                         track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1236                         track->cb_color_tile_offset[tmp] = track->cb_color_bo_offset[tmp];
1237                         ib[idx] = track->cb_color_base_last[tmp];
1238                 } else {
1239                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1240                         if (r) {
1241                                 dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1242                                 return -EINVAL;
1243                         }
1244                         track->cb_color_tile_bo[tmp] = reloc->robj;
1245                         track->cb_color_tile_offset[tmp] = (u64)ib[idx] << 8;
1246                         ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1247                 }
1248                 if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1249                         track->cb_dirty = true;
1250                 }
1251                 break;
1252         case R_028100_CB_COLOR0_MASK:
1253         case R_028104_CB_COLOR1_MASK:
1254         case R_028108_CB_COLOR2_MASK:
1255         case R_02810C_CB_COLOR3_MASK:
1256         case R_028110_CB_COLOR4_MASK:
1257         case R_028114_CB_COLOR5_MASK:
1258         case R_028118_CB_COLOR6_MASK:
1259         case R_02811C_CB_COLOR7_MASK:
1260                 tmp = (reg - R_028100_CB_COLOR0_MASK) / 4;
1261                 track->cb_color_mask[tmp] = radeon_get_ib_value(p, idx);
1262                 if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1263                         track->cb_dirty = true;
1264                 }
1265                 break;
1266         case CB_COLOR0_BASE:
1267         case CB_COLOR1_BASE:
1268         case CB_COLOR2_BASE:
1269         case CB_COLOR3_BASE:
1270         case CB_COLOR4_BASE:
1271         case CB_COLOR5_BASE:
1272         case CB_COLOR6_BASE:
1273         case CB_COLOR7_BASE:
1274                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1275                 if (r) {
1276                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1277                                         "0x%04X\n", reg);
1278                         return -EINVAL;
1279                 }
1280                 tmp = (reg - CB_COLOR0_BASE) / 4;
1281                 track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1282                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1283                 track->cb_color_base_last[tmp] = ib[idx];
1284                 track->cb_color_bo[tmp] = reloc->robj;
1285                 track->cb_color_bo_mc[tmp] = reloc->gpu_offset;
1286                 track->cb_dirty = true;
1287                 break;
1288         case DB_DEPTH_BASE:
1289                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1290                 if (r) {
1291                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1292                                         "0x%04X\n", reg);
1293                         return -EINVAL;
1294                 }
1295                 track->db_offset = radeon_get_ib_value(p, idx) << 8;
1296                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1297                 track->db_bo = reloc->robj;
1298                 track->db_bo_mc = reloc->gpu_offset;
1299                 track->db_dirty = true;
1300                 break;
1301         case DB_HTILE_DATA_BASE:
1302                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1303                 if (r) {
1304                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1305                                         "0x%04X\n", reg);
1306                         return -EINVAL;
1307                 }
1308                 track->htile_offset = radeon_get_ib_value(p, idx) << 8;
1309                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1310                 track->htile_bo = reloc->robj;
1311                 track->db_dirty = true;
1312                 break;
1313         case DB_HTILE_SURFACE:
1314                 track->htile_surface = radeon_get_ib_value(p, idx);
1315                 /* force 8x8 htile width and height */
1316                 ib[idx] |= 3;
1317                 track->db_dirty = true;
1318                 break;
1319         case SQ_PGM_START_FS:
1320         case SQ_PGM_START_ES:
1321         case SQ_PGM_START_VS:
1322         case SQ_PGM_START_GS:
1323         case SQ_PGM_START_PS:
1324         case SQ_ALU_CONST_CACHE_GS_0:
1325         case SQ_ALU_CONST_CACHE_GS_1:
1326         case SQ_ALU_CONST_CACHE_GS_2:
1327         case SQ_ALU_CONST_CACHE_GS_3:
1328         case SQ_ALU_CONST_CACHE_GS_4:
1329         case SQ_ALU_CONST_CACHE_GS_5:
1330         case SQ_ALU_CONST_CACHE_GS_6:
1331         case SQ_ALU_CONST_CACHE_GS_7:
1332         case SQ_ALU_CONST_CACHE_GS_8:
1333         case SQ_ALU_CONST_CACHE_GS_9:
1334         case SQ_ALU_CONST_CACHE_GS_10:
1335         case SQ_ALU_CONST_CACHE_GS_11:
1336         case SQ_ALU_CONST_CACHE_GS_12:
1337         case SQ_ALU_CONST_CACHE_GS_13:
1338         case SQ_ALU_CONST_CACHE_GS_14:
1339         case SQ_ALU_CONST_CACHE_GS_15:
1340         case SQ_ALU_CONST_CACHE_PS_0:
1341         case SQ_ALU_CONST_CACHE_PS_1:
1342         case SQ_ALU_CONST_CACHE_PS_2:
1343         case SQ_ALU_CONST_CACHE_PS_3:
1344         case SQ_ALU_CONST_CACHE_PS_4:
1345         case SQ_ALU_CONST_CACHE_PS_5:
1346         case SQ_ALU_CONST_CACHE_PS_6:
1347         case SQ_ALU_CONST_CACHE_PS_7:
1348         case SQ_ALU_CONST_CACHE_PS_8:
1349         case SQ_ALU_CONST_CACHE_PS_9:
1350         case SQ_ALU_CONST_CACHE_PS_10:
1351         case SQ_ALU_CONST_CACHE_PS_11:
1352         case SQ_ALU_CONST_CACHE_PS_12:
1353         case SQ_ALU_CONST_CACHE_PS_13:
1354         case SQ_ALU_CONST_CACHE_PS_14:
1355         case SQ_ALU_CONST_CACHE_PS_15:
1356         case SQ_ALU_CONST_CACHE_VS_0:
1357         case SQ_ALU_CONST_CACHE_VS_1:
1358         case SQ_ALU_CONST_CACHE_VS_2:
1359         case SQ_ALU_CONST_CACHE_VS_3:
1360         case SQ_ALU_CONST_CACHE_VS_4:
1361         case SQ_ALU_CONST_CACHE_VS_5:
1362         case SQ_ALU_CONST_CACHE_VS_6:
1363         case SQ_ALU_CONST_CACHE_VS_7:
1364         case SQ_ALU_CONST_CACHE_VS_8:
1365         case SQ_ALU_CONST_CACHE_VS_9:
1366         case SQ_ALU_CONST_CACHE_VS_10:
1367         case SQ_ALU_CONST_CACHE_VS_11:
1368         case SQ_ALU_CONST_CACHE_VS_12:
1369         case SQ_ALU_CONST_CACHE_VS_13:
1370         case SQ_ALU_CONST_CACHE_VS_14:
1371         case SQ_ALU_CONST_CACHE_VS_15:
1372                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1373                 if (r) {
1374                         dev_warn(p->dev, "bad SET_CONTEXT_REG "
1375                                         "0x%04X\n", reg);
1376                         return -EINVAL;
1377                 }
1378                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1379                 break;
1380         case SX_MEMORY_EXPORT_BASE:
1381                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1382                 if (r) {
1383                         dev_warn(p->dev, "bad SET_CONFIG_REG "
1384                                         "0x%04X\n", reg);
1385                         return -EINVAL;
1386                 }
1387                 ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1388                 break;
1389         case SX_MISC:
1390                 track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1391                 break;
1392         default:
1393                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1394                 return -EINVAL;
1395         }
1396         return 0;
1397 }
1398
1399 unsigned r600_mip_minify(unsigned size, unsigned level)
1400 {
1401         unsigned val;
1402
1403         val = max(1U, size >> level);
1404         if (level > 0)
1405                 val = roundup_pow_of_two(val);
1406         return val;
1407 }
1408
1409 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1410                               unsigned w0, unsigned h0, unsigned d0, unsigned nsamples, unsigned format,
1411                               unsigned block_align, unsigned height_align, unsigned base_align,
1412                               unsigned *l0_size, unsigned *mipmap_size)
1413 {
1414         unsigned offset, i, level;
1415         unsigned width, height, depth, size;
1416         unsigned blocksize;
1417         unsigned nbx, nby;
1418         unsigned nlevels = llevel - blevel + 1;
1419
1420         *l0_size = -1;
1421         blocksize = r600_fmt_get_blocksize(format);
1422
1423         w0 = r600_mip_minify(w0, 0);
1424         h0 = r600_mip_minify(h0, 0);
1425         d0 = r600_mip_minify(d0, 0);
1426         for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1427                 width = r600_mip_minify(w0, i);
1428                 nbx = r600_fmt_get_nblocksx(format, width);
1429
1430                 nbx = round_up(nbx, block_align);
1431
1432                 height = r600_mip_minify(h0, i);
1433                 nby = r600_fmt_get_nblocksy(format, height);
1434                 nby = round_up(nby, height_align);
1435
1436                 depth = r600_mip_minify(d0, i);
1437
1438                 size = nbx * nby * blocksize * nsamples;
1439                 if (nfaces)
1440                         size *= nfaces;
1441                 else
1442                         size *= depth;
1443
1444                 if (i == 0)
1445                         *l0_size = size;
1446
1447                 if (i == 0 || i == 1)
1448                         offset = round_up(offset, base_align);
1449
1450                 offset += size;
1451         }
1452         *mipmap_size = offset;
1453         if (llevel == 0)
1454                 *mipmap_size = *l0_size;
1455         if (!blevel)
1456                 *mipmap_size -= *l0_size;
1457 }
1458
1459 /**
1460  * r600_check_texture_resource() - check if register is authorized or not
1461  * @p: parser structure holding parsing context
1462  * @idx: index into the cs buffer
1463  * @texture: texture's bo structure
1464  * @mipmap: mipmap's bo structure
1465  *
1466  * This function will check that the resource has valid field and that
1467  * the texture and mipmap bo object are big enough to cover this resource.
1468  */
1469 static int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1470                                               struct radeon_bo *texture,
1471                                               struct radeon_bo *mipmap,
1472                                               u64 base_offset,
1473                                               u64 mip_offset,
1474                                               u32 tiling_flags)
1475 {
1476         struct r600_cs_track *track = p->track;
1477         u32 dim, nfaces, llevel, blevel, w0, h0, d0;
1478         u32 word0, word1, l0_size, mipmap_size, word2, word3, word4, word5;
1479         u32 height_align, pitch, pitch_align, depth_align;
1480         u32 barray, larray;
1481         u64 base_align;
1482         struct array_mode_checker array_check;
1483         u32 format;
1484         bool is_array;
1485
1486         /* on legacy kernel we don't perform advanced check */
1487         if (p->rdev == NULL)
1488                 return 0;
1489
1490         /* convert to bytes */
1491         base_offset <<= 8;
1492         mip_offset <<= 8;
1493
1494         word0 = radeon_get_ib_value(p, idx + 0);
1495         if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1496                 if (tiling_flags & RADEON_TILING_MACRO)
1497                         word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1498                 else if (tiling_flags & RADEON_TILING_MICRO)
1499                         word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1500         }
1501         word1 = radeon_get_ib_value(p, idx + 1);
1502         word2 = radeon_get_ib_value(p, idx + 2) << 8;
1503         word3 = radeon_get_ib_value(p, idx + 3) << 8;
1504         word4 = radeon_get_ib_value(p, idx + 4);
1505         word5 = radeon_get_ib_value(p, idx + 5);
1506         dim = G_038000_DIM(word0);
1507         w0 = G_038000_TEX_WIDTH(word0) + 1;
1508         pitch = (G_038000_PITCH(word0) + 1) * 8;
1509         h0 = G_038004_TEX_HEIGHT(word1) + 1;
1510         d0 = G_038004_TEX_DEPTH(word1);
1511         format = G_038004_DATA_FORMAT(word1);
1512         blevel = G_038010_BASE_LEVEL(word4);
1513         llevel = G_038014_LAST_LEVEL(word5);
1514         /* pitch in texels */
1515         array_check.array_mode = G_038000_TILE_MODE(word0);
1516         array_check.group_size = track->group_size;
1517         array_check.nbanks = track->nbanks;
1518         array_check.npipes = track->npipes;
1519         array_check.nsamples = 1;
1520         array_check.blocksize = r600_fmt_get_blocksize(format);
1521         nfaces = 1;
1522         is_array = false;
1523         switch (dim) {
1524         case V_038000_SQ_TEX_DIM_1D:
1525         case V_038000_SQ_TEX_DIM_2D:
1526         case V_038000_SQ_TEX_DIM_3D:
1527                 break;
1528         case V_038000_SQ_TEX_DIM_CUBEMAP:
1529                 if (p->family >= CHIP_RV770)
1530                         nfaces = 8;
1531                 else
1532                         nfaces = 6;
1533                 break;
1534         case V_038000_SQ_TEX_DIM_1D_ARRAY:
1535         case V_038000_SQ_TEX_DIM_2D_ARRAY:
1536                 is_array = true;
1537                 break;
1538         case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1539                 is_array = true;
1540                 /* fall through */
1541         case V_038000_SQ_TEX_DIM_2D_MSAA:
1542                 array_check.nsamples = 1 << llevel;
1543                 llevel = 0;
1544                 break;
1545         default:
1546                 dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1547                 return -EINVAL;
1548         }
1549         if (!r600_fmt_is_valid_texture(format, p->family)) {
1550                 dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1551                          __func__, __LINE__, format);
1552                 return -EINVAL;
1553         }
1554
1555         if (r600_get_array_mode_alignment(&array_check,
1556                                           &pitch_align, &height_align, &depth_align, &base_align)) {
1557                 dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1558                          __func__, __LINE__, G_038000_TILE_MODE(word0));
1559                 return -EINVAL;
1560         }
1561
1562         /* XXX check height as well... */
1563
1564         if (!IS_ALIGNED(pitch, pitch_align)) {
1565                 dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1566                          __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1567                 return -EINVAL;
1568         }
1569         if (!IS_ALIGNED(base_offset, base_align)) {
1570                 dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1571                          __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1572                 return -EINVAL;
1573         }
1574         if (!IS_ALIGNED(mip_offset, base_align)) {
1575                 dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1576                          __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1577                 return -EINVAL;
1578         }
1579
1580         if (blevel > llevel) {
1581                 dev_warn(p->dev, "texture blevel %d > llevel %d\n",
1582                          blevel, llevel);
1583         }
1584         if (is_array) {
1585                 barray = G_038014_BASE_ARRAY(word5);
1586                 larray = G_038014_LAST_ARRAY(word5);
1587
1588                 nfaces = larray - barray + 1;
1589         }
1590         r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, array_check.nsamples, format,
1591                           pitch_align, height_align, base_align,
1592                           &l0_size, &mipmap_size);
1593         /* using get ib will give us the offset into the texture bo */
1594         if ((l0_size + word2) > radeon_bo_size(texture)) {
1595                 dev_warn(p->dev, "texture bo too small ((%d %d) (%d %d) %d %d %d -> %d have %ld)\n",
1596                          w0, h0, pitch_align, height_align,
1597                          array_check.array_mode, format, word2,
1598                          l0_size, radeon_bo_size(texture));
1599                 dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1600                 return -EINVAL;
1601         }
1602         /* using get ib will give us the offset into the mipmap bo */
1603         if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1604                 /*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1605                   w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1606         }
1607         return 0;
1608 }
1609
1610 static bool r600_is_safe_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1611 {
1612         u32 m, i;
1613
1614         i = (reg >> 7);
1615         if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
1616                 dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1617                 return false;
1618         }
1619         m = 1 << ((reg >> 2) & 31);
1620         if (!(r600_reg_safe_bm[i] & m))
1621                 return true;
1622         dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1623         return false;
1624 }
1625
1626 static int r600_packet3_check(struct radeon_cs_parser *p,
1627                                 struct radeon_cs_packet *pkt)
1628 {
1629         struct radeon_bo_list *reloc;
1630         struct r600_cs_track *track;
1631         volatile u32 *ib;
1632         unsigned idx;
1633         unsigned i;
1634         unsigned start_reg, end_reg, reg;
1635         int r;
1636         u32 idx_value;
1637
1638         track = (struct r600_cs_track *)p->track;
1639         ib = p->ib.ptr;
1640         idx = pkt->idx + 1;
1641         idx_value = radeon_get_ib_value(p, idx);
1642
1643         switch (pkt->opcode) {
1644         case PACKET3_SET_PREDICATION:
1645         {
1646                 int pred_op;
1647                 int tmp;
1648                 uint64_t offset;
1649
1650                 if (pkt->count != 1) {
1651                         DRM_ERROR("bad SET PREDICATION\n");
1652                         return -EINVAL;
1653                 }
1654
1655                 tmp = radeon_get_ib_value(p, idx + 1);
1656                 pred_op = (tmp >> 16) & 0x7;
1657
1658                 /* for the clear predicate operation */
1659                 if (pred_op == 0)
1660                         return 0;
1661
1662                 if (pred_op > 2) {
1663                         DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1664                         return -EINVAL;
1665                 }
1666
1667                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1668                 if (r) {
1669                         DRM_ERROR("bad SET PREDICATION\n");
1670                         return -EINVAL;
1671                 }
1672
1673                 offset = reloc->gpu_offset +
1674                          (idx_value & 0xfffffff0) +
1675                          ((u64)(tmp & 0xff) << 32);
1676
1677                 ib[idx + 0] = offset;
1678                 ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1679         }
1680         break;
1681
1682         case PACKET3_START_3D_CMDBUF:
1683                 if (p->family >= CHIP_RV770 || pkt->count) {
1684                         DRM_ERROR("bad START_3D\n");
1685                         return -EINVAL;
1686                 }
1687                 break;
1688         case PACKET3_CONTEXT_CONTROL:
1689                 if (pkt->count != 1) {
1690                         DRM_ERROR("bad CONTEXT_CONTROL\n");
1691                         return -EINVAL;
1692                 }
1693                 break;
1694         case PACKET3_INDEX_TYPE:
1695         case PACKET3_NUM_INSTANCES:
1696                 if (pkt->count) {
1697                         DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1698                         return -EINVAL;
1699                 }
1700                 break;
1701         case PACKET3_DRAW_INDEX:
1702         {
1703                 uint64_t offset;
1704                 if (pkt->count != 3) {
1705                         DRM_ERROR("bad DRAW_INDEX\n");
1706                         return -EINVAL;
1707                 }
1708                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1709                 if (r) {
1710                         DRM_ERROR("bad DRAW_INDEX\n");
1711                         return -EINVAL;
1712                 }
1713
1714                 offset = reloc->gpu_offset +
1715                          idx_value +
1716                          ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1717
1718                 ib[idx+0] = offset;
1719                 ib[idx+1] = upper_32_bits(offset) & 0xff;
1720
1721                 r = r600_cs_track_check(p);
1722                 if (r) {
1723                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1724                         return r;
1725                 }
1726                 break;
1727         }
1728         case PACKET3_DRAW_INDEX_AUTO:
1729                 if (pkt->count != 1) {
1730                         DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1731                         return -EINVAL;
1732                 }
1733                 r = r600_cs_track_check(p);
1734                 if (r) {
1735                         dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1736                         return r;
1737                 }
1738                 break;
1739         case PACKET3_DRAW_INDEX_IMMD_BE:
1740         case PACKET3_DRAW_INDEX_IMMD:
1741                 if (pkt->count < 2) {
1742                         DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1743                         return -EINVAL;
1744                 }
1745                 r = r600_cs_track_check(p);
1746                 if (r) {
1747                         dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1748                         return r;
1749                 }
1750                 break;
1751         case PACKET3_WAIT_REG_MEM:
1752                 if (pkt->count != 5) {
1753                         DRM_ERROR("bad WAIT_REG_MEM\n");
1754                         return -EINVAL;
1755                 }
1756                 /* bit 4 is reg (0) or mem (1) */
1757                 if (idx_value & 0x10) {
1758                         uint64_t offset;
1759
1760                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1761                         if (r) {
1762                                 DRM_ERROR("bad WAIT_REG_MEM\n");
1763                                 return -EINVAL;
1764                         }
1765
1766                         offset = reloc->gpu_offset +
1767                                  (radeon_get_ib_value(p, idx+1) & 0xfffffff0) +
1768                                  ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1769
1770                         ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffff0);
1771                         ib[idx+2] = upper_32_bits(offset) & 0xff;
1772                 } else if (idx_value & 0x100) {
1773                         DRM_ERROR("cannot use PFP on REG wait\n");
1774                         return -EINVAL;
1775                 }
1776                 break;
1777         case PACKET3_CP_DMA:
1778         {
1779                 u32 command, size;
1780                 u64 offset, tmp;
1781                 if (pkt->count != 4) {
1782                         DRM_ERROR("bad CP DMA\n");
1783                         return -EINVAL;
1784                 }
1785                 command = radeon_get_ib_value(p, idx+4);
1786                 size = command & 0x1fffff;
1787                 if (command & PACKET3_CP_DMA_CMD_SAS) {
1788                         /* src address space is register */
1789                         DRM_ERROR("CP DMA SAS not supported\n");
1790                         return -EINVAL;
1791                 } else {
1792                         if (command & PACKET3_CP_DMA_CMD_SAIC) {
1793                                 DRM_ERROR("CP DMA SAIC only supported for registers\n");
1794                                 return -EINVAL;
1795                         }
1796                         /* src address space is memory */
1797                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1798                         if (r) {
1799                                 DRM_ERROR("bad CP DMA SRC\n");
1800                                 return -EINVAL;
1801                         }
1802
1803                         tmp = radeon_get_ib_value(p, idx) +
1804                                 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1805
1806                         offset = reloc->gpu_offset + tmp;
1807
1808                         if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1809                                 dev_warn(p->dev, "CP DMA src buffer too small (%llu %lu)\n",
1810                                          tmp + size, radeon_bo_size(reloc->robj));
1811                                 return -EINVAL;
1812                         }
1813
1814                         ib[idx] = offset;
1815                         ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1816                 }
1817                 if (command & PACKET3_CP_DMA_CMD_DAS) {
1818                         /* dst address space is register */
1819                         DRM_ERROR("CP DMA DAS not supported\n");
1820                         return -EINVAL;
1821                 } else {
1822                         /* dst address space is memory */
1823                         if (command & PACKET3_CP_DMA_CMD_DAIC) {
1824                                 DRM_ERROR("CP DMA DAIC only supported for registers\n");
1825                                 return -EINVAL;
1826                         }
1827                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1828                         if (r) {
1829                                 DRM_ERROR("bad CP DMA DST\n");
1830                                 return -EINVAL;
1831                         }
1832
1833                         tmp = radeon_get_ib_value(p, idx+2) +
1834                                 ((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
1835
1836                         offset = reloc->gpu_offset + tmp;
1837
1838                         if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1839                                 dev_warn(p->dev, "CP DMA dst buffer too small (%llu %lu)\n",
1840                                          tmp + size, radeon_bo_size(reloc->robj));
1841                                 return -EINVAL;
1842                         }
1843
1844                         ib[idx+2] = offset;
1845                         ib[idx+3] = upper_32_bits(offset) & 0xff;
1846                 }
1847                 break;
1848         }
1849         case PACKET3_SURFACE_SYNC:
1850                 if (pkt->count != 3) {
1851                         DRM_ERROR("bad SURFACE_SYNC\n");
1852                         return -EINVAL;
1853                 }
1854                 /* 0xffffffff/0x0 is flush all cache flag */
1855                 if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1856                     radeon_get_ib_value(p, idx + 2) != 0) {
1857                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1858                         if (r) {
1859                                 DRM_ERROR("bad SURFACE_SYNC\n");
1860                                 return -EINVAL;
1861                         }
1862                         ib[idx+2] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1863                 }
1864                 break;
1865         case PACKET3_EVENT_WRITE:
1866                 if (pkt->count != 2 && pkt->count != 0) {
1867                         DRM_ERROR("bad EVENT_WRITE\n");
1868                         return -EINVAL;
1869                 }
1870                 if (pkt->count) {
1871                         uint64_t offset;
1872
1873                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1874                         if (r) {
1875                                 DRM_ERROR("bad EVENT_WRITE\n");
1876                                 return -EINVAL;
1877                         }
1878                         offset = reloc->gpu_offset +
1879                                  (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
1880                                  ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1881
1882                         ib[idx+1] = offset & 0xfffffff8;
1883                         ib[idx+2] = upper_32_bits(offset) & 0xff;
1884                 }
1885                 break;
1886         case PACKET3_EVENT_WRITE_EOP:
1887         {
1888                 uint64_t offset;
1889
1890                 if (pkt->count != 4) {
1891                         DRM_ERROR("bad EVENT_WRITE_EOP\n");
1892                         return -EINVAL;
1893                 }
1894                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1895                 if (r) {
1896                         DRM_ERROR("bad EVENT_WRITE\n");
1897                         return -EINVAL;
1898                 }
1899
1900                 offset = reloc->gpu_offset +
1901                          (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
1902                          ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1903
1904                 ib[idx+1] = offset & 0xfffffffc;
1905                 ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1906                 break;
1907         }
1908         case PACKET3_SET_CONFIG_REG:
1909                 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1910                 end_reg = 4 * pkt->count + start_reg - 4;
1911                 if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1912                     (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1913                     (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1914                         DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1915                         return -EINVAL;
1916                 }
1917                 for (i = 0; i < pkt->count; i++) {
1918                         reg = start_reg + (4 * i);
1919                         r = r600_cs_check_reg(p, reg, idx+1+i);
1920                         if (r)
1921                                 return r;
1922                 }
1923                 break;
1924         case PACKET3_SET_CONTEXT_REG:
1925                 start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1926                 end_reg = 4 * pkt->count + start_reg - 4;
1927                 if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1928                     (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1929                     (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1930                         DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1931                         return -EINVAL;
1932                 }
1933                 for (i = 0; i < pkt->count; i++) {
1934                         reg = start_reg + (4 * i);
1935                         r = r600_cs_check_reg(p, reg, idx+1+i);
1936                         if (r)
1937                                 return r;
1938                 }
1939                 break;
1940         case PACKET3_SET_RESOURCE:
1941                 if (pkt->count % 7) {
1942                         DRM_ERROR("bad SET_RESOURCE\n");
1943                         return -EINVAL;
1944                 }
1945                 start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1946                 end_reg = 4 * pkt->count + start_reg - 4;
1947                 if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1948                     (start_reg >= PACKET3_SET_RESOURCE_END) ||
1949                     (end_reg >= PACKET3_SET_RESOURCE_END)) {
1950                         DRM_ERROR("bad SET_RESOURCE\n");
1951                         return -EINVAL;
1952                 }
1953                 for (i = 0; i < (pkt->count / 7); i++) {
1954                         struct radeon_bo *texture, *mipmap;
1955                         u32 size, offset, base_offset, mip_offset;
1956
1957                         switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1958                         case SQ_TEX_VTX_VALID_TEXTURE:
1959                                 /* tex base */
1960                                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1961                                 if (r) {
1962                                         DRM_ERROR("bad SET_RESOURCE\n");
1963                                         return -EINVAL;
1964                                 }
1965                                 base_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1966                                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1967                                         if (reloc->tiling_flags & RADEON_TILING_MACRO)
1968                                                 ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1969                                         else if (reloc->tiling_flags & RADEON_TILING_MICRO)
1970                                                 ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1971                                 }
1972                                 texture = reloc->robj;
1973                                 /* tex mip base */
1974                                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1975                                 if (r) {
1976                                         DRM_ERROR("bad SET_RESOURCE\n");
1977                                         return -EINVAL;
1978                                 }
1979                                 mip_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1980                                 mipmap = reloc->robj;
1981                                 r = r600_check_texture_resource(p,  idx+(i*7)+1,
1982                                                                 texture, mipmap,
1983                                                                 base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1984                                                                 mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1985                                                                 reloc->tiling_flags);
1986                                 if (r)
1987                                         return r;
1988                                 ib[idx+1+(i*7)+2] += base_offset;
1989                                 ib[idx+1+(i*7)+3] += mip_offset;
1990                                 break;
1991                         case SQ_TEX_VTX_VALID_BUFFER:
1992                         {
1993                                 uint64_t offset64;
1994                                 /* vtx base */
1995                                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1996                                 if (r) {
1997                                         DRM_ERROR("bad SET_RESOURCE\n");
1998                                         return -EINVAL;
1999                                 }
2000                                 offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
2001                                 size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
2002                                 if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2003                                         /* force size to size of the buffer */
2004                                         dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
2005                                                  size + offset, radeon_bo_size(reloc->robj));
2006                                         ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj) - offset;
2007                                 }
2008
2009                                 offset64 = reloc->gpu_offset + offset;
2010                                 ib[idx+1+(i*8)+0] = offset64;
2011                                 ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2012                                                     (upper_32_bits(offset64) & 0xff);
2013                                 break;
2014                         }
2015                         case SQ_TEX_VTX_INVALID_TEXTURE:
2016                         case SQ_TEX_VTX_INVALID_BUFFER:
2017                         default:
2018                                 DRM_ERROR("bad SET_RESOURCE\n");
2019                                 return -EINVAL;
2020                         }
2021                 }
2022                 break;
2023         case PACKET3_SET_ALU_CONST:
2024                 if (track->sq_config & DX9_CONSTS) {
2025                         start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
2026                         end_reg = 4 * pkt->count + start_reg - 4;
2027                         if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
2028                             (start_reg >= PACKET3_SET_ALU_CONST_END) ||
2029                             (end_reg >= PACKET3_SET_ALU_CONST_END)) {
2030                                 DRM_ERROR("bad SET_ALU_CONST\n");
2031                                 return -EINVAL;
2032                         }
2033                 }
2034                 break;
2035         case PACKET3_SET_BOOL_CONST:
2036                 start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
2037                 end_reg = 4 * pkt->count + start_reg - 4;
2038                 if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
2039                     (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2040                     (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2041                         DRM_ERROR("bad SET_BOOL_CONST\n");
2042                         return -EINVAL;
2043                 }
2044                 break;
2045         case PACKET3_SET_LOOP_CONST:
2046                 start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
2047                 end_reg = 4 * pkt->count + start_reg - 4;
2048                 if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
2049                     (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2050                     (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2051                         DRM_ERROR("bad SET_LOOP_CONST\n");
2052                         return -EINVAL;
2053                 }
2054                 break;
2055         case PACKET3_SET_CTL_CONST:
2056                 start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
2057                 end_reg = 4 * pkt->count + start_reg - 4;
2058                 if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
2059                     (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2060                     (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2061                         DRM_ERROR("bad SET_CTL_CONST\n");
2062                         return -EINVAL;
2063                 }
2064                 break;
2065         case PACKET3_SET_SAMPLER:
2066                 if (pkt->count % 3) {
2067                         DRM_ERROR("bad SET_SAMPLER\n");
2068                         return -EINVAL;
2069                 }
2070                 start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
2071                 end_reg = 4 * pkt->count + start_reg - 4;
2072                 if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
2073                     (start_reg >= PACKET3_SET_SAMPLER_END) ||
2074                     (end_reg >= PACKET3_SET_SAMPLER_END)) {
2075                         DRM_ERROR("bad SET_SAMPLER\n");
2076                         return -EINVAL;
2077                 }
2078                 break;
2079         case PACKET3_STRMOUT_BASE_UPDATE:
2080                 /* RS780 and RS880 also need this */
2081                 if (p->family < CHIP_RS780) {
2082                         DRM_ERROR("STRMOUT_BASE_UPDATE only supported on 7xx\n");
2083                         return -EINVAL;
2084                 }
2085                 if (pkt->count != 1) {
2086                         DRM_ERROR("bad STRMOUT_BASE_UPDATE packet count\n");
2087                         return -EINVAL;
2088                 }
2089                 if (idx_value > 3) {
2090                         DRM_ERROR("bad STRMOUT_BASE_UPDATE index\n");
2091                         return -EINVAL;
2092                 }
2093                 {
2094                         u64 offset;
2095
2096                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2097                         if (r) {
2098                                 DRM_ERROR("bad STRMOUT_BASE_UPDATE reloc\n");
2099                                 return -EINVAL;
2100                         }
2101
2102                         if (reloc->robj != track->vgt_strmout_bo[idx_value]) {
2103                                 DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo does not match\n");
2104                                 return -EINVAL;
2105                         }
2106
2107                         offset = radeon_get_ib_value(p, idx+1) << 8;
2108                         if (offset != track->vgt_strmout_bo_offset[idx_value]) {
2109                                 DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo offset does not match: 0x%llx, 0x%x\n",
2110                                           offset, track->vgt_strmout_bo_offset[idx_value]);
2111                                 return -EINVAL;
2112                         }
2113
2114                         if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2115                                 DRM_ERROR("bad STRMOUT_BASE_UPDATE bo too small: 0x%llx, 0x%lx\n",
2116                                           offset + 4, radeon_bo_size(reloc->robj));
2117                                 return -EINVAL;
2118                         }
2119                         ib[idx+1] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2120                 }
2121                 break;
2122         case PACKET3_SURFACE_BASE_UPDATE:
2123                 if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
2124                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2125                         return -EINVAL;
2126                 }
2127                 if (pkt->count) {
2128                         DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2129                         return -EINVAL;
2130                 }
2131                 break;
2132         case PACKET3_STRMOUT_BUFFER_UPDATE:
2133                 if (pkt->count != 4) {
2134                         DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2135                         return -EINVAL;
2136                 }
2137                 /* Updating memory at DST_ADDRESS. */
2138                 if (idx_value & 0x1) {
2139                         u64 offset;
2140                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2141                         if (r) {
2142                                 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2143                                 return -EINVAL;
2144                         }
2145                         offset = radeon_get_ib_value(p, idx+1);
2146                         offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2147                         if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2148                                 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2149                                           offset + 4, radeon_bo_size(reloc->robj));
2150                                 return -EINVAL;
2151                         }
2152                         offset += reloc->gpu_offset;
2153                         ib[idx+1] = offset;
2154                         ib[idx+2] = upper_32_bits(offset) & 0xff;
2155                 }
2156                 /* Reading data from SRC_ADDRESS. */
2157                 if (((idx_value >> 1) & 0x3) == 2) {
2158                         u64 offset;
2159                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2160                         if (r) {
2161                                 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2162                                 return -EINVAL;
2163                         }
2164                         offset = radeon_get_ib_value(p, idx+3);
2165                         offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2166                         if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2167                                 DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2168                                           offset + 4, radeon_bo_size(reloc->robj));
2169                                 return -EINVAL;
2170                         }
2171                         offset += reloc->gpu_offset;
2172                         ib[idx+3] = offset;
2173                         ib[idx+4] = upper_32_bits(offset) & 0xff;
2174                 }
2175                 break;
2176         case PACKET3_MEM_WRITE:
2177         {
2178                 u64 offset;
2179
2180                 if (pkt->count != 3) {
2181                         DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2182                         return -EINVAL;
2183                 }
2184                 r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2185                 if (r) {
2186                         DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2187                         return -EINVAL;
2188                 }
2189                 offset = radeon_get_ib_value(p, idx+0);
2190                 offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2191                 if (offset & 0x7) {
2192                         DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2193                         return -EINVAL;
2194                 }
2195                 if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2196                         DRM_ERROR("bad MEM_WRITE bo too small: 0x%llx, 0x%lx\n",
2197                                   offset + 8, radeon_bo_size(reloc->robj));
2198                         return -EINVAL;
2199                 }
2200                 offset += reloc->gpu_offset;
2201                 ib[idx+0] = offset;
2202                 ib[idx+1] = upper_32_bits(offset) & 0xff;
2203                 break;
2204         }
2205         case PACKET3_COPY_DW:
2206                 if (pkt->count != 4) {
2207                         DRM_ERROR("bad COPY_DW (invalid count)\n");
2208                         return -EINVAL;
2209                 }
2210                 if (idx_value & 0x1) {
2211                         u64 offset;
2212                         /* SRC is memory. */
2213                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2214                         if (r) {
2215                                 DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2216                                 return -EINVAL;
2217                         }
2218                         offset = radeon_get_ib_value(p, idx+1);
2219                         offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2220                         if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2221                                 DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2222                                           offset + 4, radeon_bo_size(reloc->robj));
2223                                 return -EINVAL;
2224                         }
2225                         offset += reloc->gpu_offset;
2226                         ib[idx+1] = offset;
2227                         ib[idx+2] = upper_32_bits(offset) & 0xff;
2228                 } else {
2229                         /* SRC is a reg. */
2230                         reg = radeon_get_ib_value(p, idx+1) << 2;
2231                         if (!r600_is_safe_reg(p, reg, idx+1))
2232                                 return -EINVAL;
2233                 }
2234                 if (idx_value & 0x2) {
2235                         u64 offset;
2236                         /* DST is memory. */
2237                         r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2238                         if (r) {
2239                                 DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2240                                 return -EINVAL;
2241                         }
2242                         offset = radeon_get_ib_value(p, idx+3);
2243                         offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2244                         if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2245                                 DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2246                                           offset + 4, radeon_bo_size(reloc->robj));
2247                                 return -EINVAL;
2248                         }
2249                         offset += reloc->gpu_offset;
2250                         ib[idx+3] = offset;
2251                         ib[idx+4] = upper_32_bits(offset) & 0xff;
2252                 } else {
2253                         /* DST is a reg. */
2254                         reg = radeon_get_ib_value(p, idx+3) << 2;
2255                         if (!r600_is_safe_reg(p, reg, idx+3))
2256                                 return -EINVAL;
2257                 }
2258                 break;
2259         case PACKET3_NOP:
2260                 break;
2261         default:
2262                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2263                 return -EINVAL;
2264         }
2265         return 0;
2266 }
2267
2268 int r600_cs_parse(struct radeon_cs_parser *p)
2269 {
2270         struct radeon_cs_packet pkt;
2271         struct r600_cs_track *track;
2272         int r;
2273
2274         if (p->track == NULL) {
2275                 /* initialize tracker, we are in kms */
2276                 track = kzalloc(sizeof(*track), GFP_KERNEL);
2277                 if (track == NULL)
2278                         return -ENOMEM;
2279                 r600_cs_track_init(track);
2280                 if (p->rdev->family < CHIP_RV770) {
2281                         track->npipes = p->rdev->config.r600.tiling_npipes;
2282                         track->nbanks = p->rdev->config.r600.tiling_nbanks;
2283                         track->group_size = p->rdev->config.r600.tiling_group_size;
2284                 } else if (p->rdev->family <= CHIP_RV740) {
2285                         track->npipes = p->rdev->config.rv770.tiling_npipes;
2286                         track->nbanks = p->rdev->config.rv770.tiling_nbanks;
2287                         track->group_size = p->rdev->config.rv770.tiling_group_size;
2288                 }
2289                 p->track = track;
2290         }
2291         do {
2292                 r = radeon_cs_packet_parse(p, &pkt, p->idx);
2293                 if (r) {
2294                         kfree(p->track);
2295                         p->track = NULL;
2296                         return r;
2297                 }
2298                 p->idx += pkt.count + 2;
2299                 switch (pkt.type) {
2300                 case RADEON_PACKET_TYPE0:
2301                         r = r600_cs_parse_packet0(p, &pkt);
2302                         break;
2303                 case RADEON_PACKET_TYPE2:
2304                         break;
2305                 case RADEON_PACKET_TYPE3:
2306                         r = r600_packet3_check(p, &pkt);
2307                         break;
2308                 default:
2309                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2310                         kfree(p->track);
2311                         p->track = NULL;
2312                         return -EINVAL;
2313                 }
2314                 if (r) {
2315                         kfree(p->track);
2316                         p->track = NULL;
2317                         return r;
2318                 }
2319         } while (p->idx < p->chunk_ib->length_dw);
2320 #if 0
2321         for (r = 0; r < p->ib.length_dw; r++) {
2322                 pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2323                 mdelay(1);
2324         }
2325 #endif
2326         kfree(p->track);
2327         p->track = NULL;
2328         return 0;
2329 }
2330
2331 /*
2332  *  DMA
2333  */
2334 /**
2335  * r600_dma_cs_next_reloc() - parse next reloc
2336  * @p:          parser structure holding parsing context.
2337  * @cs_reloc:           reloc informations
2338  *
2339  * Return the next reloc, do bo validation and compute
2340  * GPU offset using the provided start.
2341  **/
2342 int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
2343                            struct radeon_bo_list **cs_reloc)
2344 {
2345         struct radeon_cs_chunk *relocs_chunk;
2346         unsigned idx;
2347
2348         *cs_reloc = NULL;
2349         if (p->chunk_relocs == NULL) {
2350                 DRM_ERROR("No relocation chunk !\n");
2351                 return -EINVAL;
2352         }
2353         relocs_chunk = p->chunk_relocs;
2354         idx = p->dma_reloc_idx;
2355         if (idx >= p->nrelocs) {
2356                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
2357                           idx, p->nrelocs);
2358                 return -EINVAL;
2359         }
2360         *cs_reloc = &p->relocs[idx];
2361         p->dma_reloc_idx++;
2362         return 0;
2363 }
2364
2365 #define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
2366 #define GET_DMA_COUNT(h) ((h) & 0x0000ffff)
2367 #define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
2368
2369 /**
2370  * r600_dma_cs_parse() - parse the DMA IB
2371  * @p:          parser structure holding parsing context.
2372  *
2373  * Parses the DMA IB from the CS ioctl and updates
2374  * the GPU addresses based on the reloc information and
2375  * checks for errors. (R6xx-R7xx)
2376  * Returns 0 for success and an error on failure.
2377  **/
2378 int r600_dma_cs_parse(struct radeon_cs_parser *p)
2379 {
2380         struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
2381         struct radeon_bo_list *src_reloc, *dst_reloc;
2382         u32 header, cmd, count, tiled;
2383         volatile u32 *ib = p->ib.ptr;
2384         u32 idx, idx_value;
2385         u64 src_offset, dst_offset;
2386         int r;
2387
2388         do {
2389                 if (p->idx >= ib_chunk->length_dw) {
2390                         DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2391                                   p->idx, ib_chunk->length_dw);
2392                         return -EINVAL;
2393                 }
2394                 idx = p->idx;
2395                 header = radeon_get_ib_value(p, idx);
2396                 cmd = GET_DMA_CMD(header);
2397                 count = GET_DMA_COUNT(header);
2398                 tiled = GET_DMA_T(header);
2399
2400                 switch (cmd) {
2401                 case DMA_PACKET_WRITE:
2402                         r = r600_dma_cs_next_reloc(p, &dst_reloc);
2403                         if (r) {
2404                                 DRM_ERROR("bad DMA_PACKET_WRITE\n");
2405                                 return -EINVAL;
2406                         }
2407                         if (tiled) {
2408                                 dst_offset = radeon_get_ib_value(p, idx+1);
2409                                 dst_offset <<= 8;
2410
2411                                 ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2412                                 p->idx += count + 5;
2413                         } else {
2414                                 dst_offset = radeon_get_ib_value(p, idx+1);
2415                                 dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2416
2417                                 ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2418                                 ib[idx+2] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2419                                 p->idx += count + 3;
2420                         }
2421                         if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2422                                 dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
2423                                          dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2424                                 return -EINVAL;
2425                         }
2426                         break;
2427                 case DMA_PACKET_COPY:
2428                         r = r600_dma_cs_next_reloc(p, &src_reloc);
2429                         if (r) {
2430                                 DRM_ERROR("bad DMA_PACKET_COPY\n");
2431                                 return -EINVAL;
2432                         }
2433                         r = r600_dma_cs_next_reloc(p, &dst_reloc);
2434                         if (r) {
2435                                 DRM_ERROR("bad DMA_PACKET_COPY\n");
2436                                 return -EINVAL;
2437                         }
2438                         if (tiled) {
2439                                 idx_value = radeon_get_ib_value(p, idx + 2);
2440                                 /* detile bit */
2441                                 if (idx_value & (1 << 31)) {
2442                                         /* tiled src, linear dst */
2443                                         src_offset = radeon_get_ib_value(p, idx+1);
2444                                         src_offset <<= 8;
2445                                         ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
2446
2447                                         dst_offset = radeon_get_ib_value(p, idx+5);
2448                                         dst_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2449                                         ib[idx+5] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2450                                         ib[idx+6] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2451                                 } else {
2452                                         /* linear src, tiled dst */
2453                                         src_offset = radeon_get_ib_value(p, idx+5);
2454                                         src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2455                                         ib[idx+5] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2456                                         ib[idx+6] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2457
2458                                         dst_offset = radeon_get_ib_value(p, idx+1);
2459                                         dst_offset <<= 8;
2460                                         ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2461                                 }
2462                                 p->idx += 7;
2463                         } else {
2464                                 if (p->family >= CHIP_RV770) {
2465                                         src_offset = radeon_get_ib_value(p, idx+2);
2466                                         src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2467                                         dst_offset = radeon_get_ib_value(p, idx+1);
2468                                         dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2469
2470                                         ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2471                                         ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2472                                         ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2473                                         ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2474                                         p->idx += 5;
2475                                 } else {
2476                                         src_offset = radeon_get_ib_value(p, idx+2);
2477                                         src_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2478                                         dst_offset = radeon_get_ib_value(p, idx+1);
2479                                         dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff0000)) << 16;
2480
2481                                         ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2482                                         ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2483                                         ib[idx+3] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2484                                         ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) & 0xff) << 16;
2485                                         p->idx += 4;
2486                                 }
2487                         }
2488                         if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2489                                 dev_warn(p->dev, "DMA copy src buffer too small (%llu %lu)\n",
2490                                          src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2491                                 return -EINVAL;
2492                         }
2493                         if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2494                                 dev_warn(p->dev, "DMA write dst buffer too small (%llu %lu)\n",
2495                                          dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2496                                 return -EINVAL;
2497                         }
2498                         break;
2499                 case DMA_PACKET_CONSTANT_FILL:
2500                         if (p->family < CHIP_RV770) {
2501                                 DRM_ERROR("Constant Fill is 7xx only !\n");
2502                                 return -EINVAL;
2503                         }
2504                         r = r600_dma_cs_next_reloc(p, &dst_reloc);
2505                         if (r) {
2506                                 DRM_ERROR("bad DMA_PACKET_WRITE\n");
2507                                 return -EINVAL;
2508                         }
2509                         dst_offset = radeon_get_ib_value(p, idx+1);
2510                         dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
2511                         if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2512                                 dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
2513                                          dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2514                                 return -EINVAL;
2515                         }
2516                         ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2517                         ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) << 16) & 0x00ff0000;
2518                         p->idx += 4;
2519                         break;
2520                 case DMA_PACKET_NOP:
2521                         p->idx += 1;
2522                         break;
2523                 default:
2524                         DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
2525                         return -EINVAL;
2526                 }
2527         } while (p->idx < p->chunk_ib->length_dw);
2528 #if 0
2529         for (r = 0; r < p->ib->length_dw; r++) {
2530                 pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2531                 mdelay(1);
2532         }
2533 #endif
2534         return 0;
2535 }