target/linux/brcm2708/patches-4.4/0120-drm-vc4-Synchronize-validation-code-for-v2-submissio.patch

   1 From f8453aacbe60712c31c57580a126017b798bd339 Mon Sep 17 00:00:00 2001
   2 From: Eric Anholt <eric@anholt.net>
   3 Date: Mon, 7 Dec 2015 12:35:01 -0800
   4 Subject: [PATCH 120/170] drm/vc4: Synchronize validation code for v2
   5  submission upstream.
   6
   7 Signed-off-by: Eric Anholt <eric@anholt.net>
   8 ---
   9  drivers/gpu/drm/vc4/vc4_drv.h       |  24 +--
  10  drivers/gpu/drm/vc4/vc4_gem.c       |  14 +-
  11  drivers/gpu/drm/vc4/vc4_render_cl.c |   6 +-
  12  drivers/gpu/drm/vc4/vc4_validate.c  | 287 +++++++++++++++---------------------
  13  4 files changed, 135 insertions(+), 196 deletions(-)
  14
  15 --- a/drivers/gpu/drm/vc4/vc4_drv.h
  16 +++ b/drivers/gpu/drm/vc4/vc4_drv.h
  17 @@ -189,17 +189,6 @@ to_vc4_encoder(struct drm_encoder *encod
  18  #define HVS_READ(offset) readl(vc4->hvs->regs + offset)
  19  #define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
  20
  21 -enum vc4_bo_mode {
  22 -       VC4_MODE_UNDECIDED,
  23 -       VC4_MODE_RENDER,
  24 -       VC4_MODE_SHADER,
  25 -};
  26 -
  27 -struct vc4_bo_exec_state {
  28 -       struct drm_gem_cma_object *bo;
  29 -       enum vc4_bo_mode mode;
  30 -};
  31 -
  32  struct vc4_exec_info {
  33         /* Sequence number for this bin/render job. */
  34         uint64_t seqno;
  35 @@ -210,7 +199,7 @@ struct vc4_exec_info {
  36         /* This is the array of BOs that were looked up at the start of exec.
  37          * Command validation will use indices into this array.
  38          */
  39 -       struct vc4_bo_exec_state *bo;
  40 +       struct drm_gem_cma_object **bo;
  41         uint32_t bo_count;
  42
  43         /* Pointers for our position in vc4->job_list */
  44 @@ -238,7 +227,6 @@ struct vc4_exec_info {
  45          * command lists.
  46          */
  47         struct vc4_shader_state {
  48 -               uint8_t packet;
  49                 uint32_t addr;
  50                 /* Maximum vertex index referenced by any primitive using this
  51                  * shader state.
  52 @@ -254,6 +242,7 @@ struct vc4_exec_info {
  53         bool found_tile_binning_mode_config_packet;
  54         bool found_start_tile_binning_packet;
  55         bool found_increment_semaphore_packet;
  56 +       bool found_flush;
  57         uint8_t bin_tiles_x, bin_tiles_y;
  58         struct drm_gem_cma_object *tile_bo;
  59         uint32_t tile_alloc_offset;
  60 @@ -265,6 +254,9 @@ struct vc4_exec_info {
  61         uint32_t ct0ca, ct0ea;
  62         uint32_t ct1ca, ct1ea;
  63
  64 +       /* Pointer to the unvalidated bin CL (if present). */
  65 +       void *bin_u;
  66 +
  67         /* Pointers to the shader recs.  These paddr gets incremented as CL
  68          * packets are relocated in validate_gl_shader_state, and the vaddrs
  69          * (u and v) get incremented and size decremented as the shader recs
  70 @@ -455,10 +447,8 @@ vc4_validate_bin_cl(struct drm_device *d
  71  int
  72  vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
  73
  74 -bool vc4_use_bo(struct vc4_exec_info *exec,
  75 -               uint32_t hindex,
  76 -               enum vc4_bo_mode mode,
  77 -               struct drm_gem_cma_object **obj);
  78 +struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
  79 +                                     uint32_t hindex);
  80
  81  int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
  82
  83 --- a/drivers/gpu/drm/vc4/vc4_gem.c
  84 +++ b/drivers/gpu/drm/vc4/vc4_gem.c
  85 @@ -169,8 +169,8 @@ vc4_save_hang_state(struct drm_device *d
  86         }
  87
  88         for (i = 0; i < exec->bo_count; i++) {
  89 -               drm_gem_object_reference(&exec->bo[i].bo->base);
  90 -               kernel_state->bo[i] = &exec->bo[i].bo->base;
  91 +               drm_gem_object_reference(&exec->bo[i]->base);
  92 +               kernel_state->bo[i] = &exec->bo[i]->base;
  93         }
  94
  95         list_for_each_entry(bo, &exec->unref_list, unref_head) {
  96 @@ -397,7 +397,7 @@ vc4_update_bo_seqnos(struct vc4_exec_inf
  97         unsigned i;
  98
  99         for (i = 0; i < exec->bo_count; i++) {
 100 -               bo = to_vc4_bo(&exec->bo[i].bo->base);
 101 +               bo = to_vc4_bo(&exec->bo[i]->base);
 102                 bo->seqno = seqno;
 103         }
 104
 105 @@ -467,7 +467,7 @@ vc4_cl_lookup_bos(struct drm_device *dev
 106                 return -EINVAL;
 107         }
 108
 109 -       exec->bo = kcalloc(exec->bo_count, sizeof(struct vc4_bo_exec_state),
 110 +       exec->bo = kcalloc(exec->bo_count, sizeof(struct drm_gem_cma_object *),
 111                            GFP_KERNEL);
 112         if (!exec->bo) {
 113                 DRM_ERROR("Failed to allocate validated BO pointers\n");
 114 @@ -500,7 +500,7 @@ vc4_cl_lookup_bos(struct drm_device *dev
 115                         goto fail;
 116                 }
 117                 drm_gem_object_reference(bo);
 118 -               exec->bo[i].bo = (struct drm_gem_cma_object *)bo;
 119 +               exec->bo[i] = (struct drm_gem_cma_object *)bo;
 120         }
 121         spin_unlock(&file_priv->table_lock);
 122
 123 @@ -591,6 +591,8 @@ vc4_get_bcl(struct drm_device *dev, stru
 124
 125         exec->ct0ca = exec->exec_bo->paddr + bin_offset;
 126
 127 +       exec->bin_u = bin;
 128 +
 129         exec->shader_rec_v = exec->exec_bo->vaddr + shader_rec_offset;
 130         exec->shader_rec_p = exec->exec_bo->paddr + shader_rec_offset;
 131         exec->shader_rec_size = args->shader_rec_size;
 132 @@ -622,7 +624,7 @@ vc4_complete_exec(struct drm_device *dev
 133         mutex_lock(&dev->struct_mutex);
 134         if (exec->bo) {
 135                 for (i = 0; i < exec->bo_count; i++)
 136 -                       drm_gem_object_unreference(&exec->bo[i].bo->base);
 137 +                       drm_gem_object_unreference(&exec->bo[i]->base);
 138                 kfree(exec->bo);
 139         }
 140
 141 --- a/drivers/gpu/drm/vc4/vc4_render_cl.c
 142 +++ b/drivers/gpu/drm/vc4/vc4_render_cl.c
 143 @@ -436,7 +436,8 @@ static int vc4_rcl_surface_setup(struct
 144         if (surf->hindex == ~0)
 145                 return 0;
 146
 147 -       if (!vc4_use_bo(exec, surf->hindex, VC4_MODE_RENDER, obj))
 148 +       *obj = vc4_use_bo(exec, surf->hindex);
 149 +       if (!*obj)
 150                 return -EINVAL;
 151
 152         if (surf->flags & VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
 153 @@ -537,7 +538,8 @@ vc4_rcl_render_config_surface_setup(stru
 154         if (surf->hindex == ~0)
 155                 return 0;
 156
 157 -       if (!vc4_use_bo(exec, surf->hindex, VC4_MODE_RENDER, obj))
 158 +       *obj = vc4_use_bo(exec, surf->hindex);
 159 +       if (!*obj)
 160                 return -EINVAL;
 161
 162         if (tiling > VC4_TILING_FORMAT_LT) {
 163 --- a/drivers/gpu/drm/vc4/vc4_validate.c
 164 +++ b/drivers/gpu/drm/vc4/vc4_validate.c
 165 @@ -94,42 +94,42 @@ size_is_lt(uint32_t width, uint32_t heig
 166                 height <= 4 * utile_height(cpp));
 167  }
 168
 169 -bool
 170 -vc4_use_bo(struct vc4_exec_info *exec,
 171 -          uint32_t hindex,
 172 -          enum vc4_bo_mode mode,
 173 -          struct drm_gem_cma_object **obj)
 174 +struct drm_gem_cma_object *
 175 +vc4_use_bo(struct vc4_exec_info *exec, uint32_t hindex)
 176  {
 177 -       *obj = NULL;
 178 +       struct drm_gem_cma_object *obj;
 179 +       struct vc4_bo *bo;
 180
 181         if (hindex >= exec->bo_count) {
 182                 DRM_ERROR("BO index %d greater than BO count %d\n",
 183                           hindex, exec->bo_count);
 184 -               return false;
 185 +               return NULL;
 186         }
 187 +       obj = exec->bo[hindex];
 188 +       bo = to_vc4_bo(&obj->base);
 189
 190 -       if (exec->bo[hindex].mode != mode) {
 191 -               if (exec->bo[hindex].mode == VC4_MODE_UNDECIDED) {
 192 -                       exec->bo[hindex].mode = mode;
 193 -               } else {
 194 -                       DRM_ERROR("BO index %d reused with mode %d vs %d\n",
 195 -                                 hindex, exec->bo[hindex].mode, mode);
 196 -                       return false;
 197 -               }
 198 +       if (bo->validated_shader) {
 199 +               DRM_ERROR("Trying to use shader BO as something other than "
 200 +                         "a shader\n");
 201 +               return NULL;
 202         }
 203
 204 -       *obj = exec->bo[hindex].bo;
 205 -       return true;
 206 +       return obj;
 207 +}
 208 +
 209 +static struct drm_gem_cma_object *
 210 +vc4_use_handle(struct vc4_exec_info *exec, uint32_t gem_handles_packet_index)
 211 +{
 212 +       return vc4_use_bo(exec, exec->bo_index[gem_handles_packet_index]);
 213  }
 214
 215  static bool
 216 -vc4_use_handle(struct vc4_exec_info *exec,
 217 -              uint32_t gem_handles_packet_index,
 218 -              enum vc4_bo_mode mode,
 219 -              struct drm_gem_cma_object **obj)
 220 +validate_bin_pos(struct vc4_exec_info *exec, void *untrusted, uint32_t pos)
 221  {
 222 -       return vc4_use_bo(exec, exec->bo_index[gem_handles_packet_index],
 223 -                         mode, obj);
 224 +       /* Note that the untrusted pointer passed to these functions is
 225 +        * incremented past the packet byte.
 226 +        */
 227 +       return (untrusted - 1 == exec->bin_u + pos);
 228  }
 229
 230  static uint32_t
 231 @@ -202,13 +202,13 @@ vc4_check_tex_size(struct vc4_exec_info
 232  }
 233
 234  static int
 235 -validate_flush_all(VALIDATE_ARGS)
 236 +validate_flush(VALIDATE_ARGS)
 237  {
 238 -       if (exec->found_increment_semaphore_packet) {
 239 -               DRM_ERROR("VC4_PACKET_FLUSH_ALL after "
 240 -                         "VC4_PACKET_INCREMENT_SEMAPHORE\n");
 241 +       if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 1)) {
 242 +               DRM_ERROR("Bin CL must end with VC4_PACKET_FLUSH\n");
 243                 return -EINVAL;
 244         }
 245 +       exec->found_flush = true;
 246
 247         return 0;
 248  }
 249 @@ -233,17 +233,13 @@ validate_start_tile_binning(VALIDATE_ARG
 250  static int
 251  validate_increment_semaphore(VALIDATE_ARGS)
 252  {
 253 -       if (exec->found_increment_semaphore_packet) {
 254 -               DRM_ERROR("Duplicate VC4_PACKET_INCREMENT_SEMAPHORE\n");
 255 +       if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 2)) {
 256 +               DRM_ERROR("Bin CL must end with "
 257 +                         "VC4_PACKET_INCREMENT_SEMAPHORE\n");
 258                 return -EINVAL;
 259         }
 260         exec->found_increment_semaphore_packet = true;
 261
 262 -       /* Once we've found the semaphore increment, there should be one FLUSH
 263 -        * then the end of the command list.  The FLUSH actually triggers the
 264 -        * increment, so we only need to make sure there
 265 -        */
 266 -
 267         return 0;
 268  }
 269
 270 @@ -257,11 +253,6 @@ validate_indexed_prim_list(VALIDATE_ARGS
 271         uint32_t index_size = (*(uint8_t *)(untrusted + 0) >> 4) ? 2 : 1;
 272         struct vc4_shader_state *shader_state;
 273
 274 -       if (exec->found_increment_semaphore_packet) {
 275 -               DRM_ERROR("Drawing after VC4_PACKET_INCREMENT_SEMAPHORE\n");
 276 -               return -EINVAL;
 277 -       }
 278 -
 279         /* Check overflow condition */
 280         if (exec->shader_state_count == 0) {
 281                 DRM_ERROR("shader state must precede primitives\n");
 282 @@ -272,7 +263,8 @@ validate_indexed_prim_list(VALIDATE_ARGS
 283         if (max_index > shader_state->max_index)
 284                 shader_state->max_index = max_index;
 285
 286 -       if (!vc4_use_handle(exec, 0, VC4_MODE_RENDER, &ib))
 287 +       ib = vc4_use_handle(exec, 0);
 288 +       if (!ib)
 289                 return -EINVAL;
 290
 291         if (offset > ib->base.size ||
 292 @@ -295,11 +287,6 @@ validate_gl_array_primitive(VALIDATE_ARG
 293         uint32_t max_index;
 294         struct vc4_shader_state *shader_state;
 295
 296 -       if (exec->found_increment_semaphore_packet) {
 297 -               DRM_ERROR("Drawing after VC4_PACKET_INCREMENT_SEMAPHORE\n");
 298 -               return -EINVAL;
 299 -       }
 300 -
 301         /* Check overflow condition */
 302         if (exec->shader_state_count == 0) {
 303                 DRM_ERROR("shader state must precede primitives\n");
 304 @@ -329,7 +316,6 @@ validate_gl_shader_state(VALIDATE_ARGS)
 305                 return -EINVAL;
 306         }
 307
 308 -       exec->shader_state[i].packet = VC4_PACKET_GL_SHADER_STATE;
 309         exec->shader_state[i].addr = *(uint32_t *)untrusted;
 310         exec->shader_state[i].max_index = 0;
 311
 312 @@ -348,31 +334,6 @@ validate_gl_shader_state(VALIDATE_ARGS)
 313  }
 314
 315  static int
 316 -validate_nv_shader_state(VALIDATE_ARGS)
 317 -{
 318 -       uint32_t i = exec->shader_state_count++;
 319 -
 320 -       if (i >= exec->shader_state_size) {
 321 -               DRM_ERROR("More requests for shader states than declared\n");
 322 -               return -EINVAL;
 323 -       }
 324 -
 325 -       exec->shader_state[i].packet = VC4_PACKET_NV_SHADER_STATE;
 326 -       exec->shader_state[i].addr = *(uint32_t *)untrusted;
 327 -
 328 -       if (exec->shader_state[i].addr & 15) {
 329 -               DRM_ERROR("NV shader state address 0x%08x misaligned\n",
 330 -                         exec->shader_state[i].addr);
 331 -               return -EINVAL;
 332 -       }
 333 -
 334 -       *(uint32_t *)validated = (exec->shader_state[i].addr +
 335 -                                 exec->shader_rec_p);
 336 -
 337 -       return 0;
 338 -}
 339 -
 340 -static int
 341  validate_tile_binning_config(VALIDATE_ARGS)
 342  {
 343         struct drm_device *dev = exec->exec_bo->base.dev;
 344 @@ -473,8 +434,8 @@ static const struct cmd_info {
 345  } cmd_info[] = {
 346         VC4_DEFINE_PACKET(VC4_PACKET_HALT, NULL),
 347         VC4_DEFINE_PACKET(VC4_PACKET_NOP, NULL),
 348 -       VC4_DEFINE_PACKET(VC4_PACKET_FLUSH, NULL),
 349 -       VC4_DEFINE_PACKET(VC4_PACKET_FLUSH_ALL, validate_flush_all),
 350 +       VC4_DEFINE_PACKET(VC4_PACKET_FLUSH, validate_flush),
 351 +       VC4_DEFINE_PACKET(VC4_PACKET_FLUSH_ALL, NULL),
 352         VC4_DEFINE_PACKET(VC4_PACKET_START_TILE_BINNING,
 353                           validate_start_tile_binning),
 354         VC4_DEFINE_PACKET(VC4_PACKET_INCREMENT_SEMAPHORE,
 355 @@ -488,7 +449,6 @@ static const struct cmd_info {
 356         VC4_DEFINE_PACKET(VC4_PACKET_PRIMITIVE_LIST_FORMAT, NULL),
 357
 358         VC4_DEFINE_PACKET(VC4_PACKET_GL_SHADER_STATE, validate_gl_shader_state),
 359 -       VC4_DEFINE_PACKET(VC4_PACKET_NV_SHADER_STATE, validate_nv_shader_state),
 360
 361         VC4_DEFINE_PACKET(VC4_PACKET_CONFIGURATION_BITS, NULL),
 362         VC4_DEFINE_PACKET(VC4_PACKET_FLAT_SHADE_FLAGS, NULL),
 363 @@ -575,8 +535,16 @@ vc4_validate_bin_cl(struct drm_device *d
 364                 return -EINVAL;
 365         }
 366
 367 -       if (!exec->found_increment_semaphore_packet) {
 368 -               DRM_ERROR("Bin CL missing VC4_PACKET_INCREMENT_SEMAPHORE\n");
 369 +       /* The bin CL must be ended with INCREMENT_SEMAPHORE and FLUSH.  The
 370 +        * semaphore is used to trigger the render CL to start up, and the
 371 +        * FLUSH is what caps the bin lists with
 372 +        * VC4_PACKET_RETURN_FROM_SUB_LIST (so they jump back to the main
 373 +        * render CL when they get called to) and actually triggers the queued
 374 +        * semaphore increment.
 375 +        */
 376 +       if (!exec->found_increment_semaphore_packet || !exec->found_flush) {
 377 +               DRM_ERROR("Bin CL missing VC4_PACKET_INCREMENT_SEMAPHORE + "
 378 +                         "VC4_PACKET_FLUSH\n");
 379                 return -EINVAL;
 380         }
 381
 382 @@ -607,7 +575,8 @@ reloc_tex(struct vc4_exec_info *exec,
 383         uint32_t cube_map_stride = 0;
 384         enum vc4_texture_data_type type;
 385
 386 -       if (!vc4_use_bo(exec, texture_handle_index, VC4_MODE_RENDER, &tex))
 387 +       tex = vc4_use_bo(exec, texture_handle_index);
 388 +       if (!tex)
 389                 return false;
 390
 391         if (sample->is_direct) {
 392 @@ -755,51 +724,28 @@ reloc_tex(struct vc4_exec_info *exec,
 393  }
 394
 395  static int
 396 -validate_shader_rec(struct drm_device *dev,
 397 -                   struct vc4_exec_info *exec,
 398 -                   struct vc4_shader_state *state)
 399 +validate_gl_shader_rec(struct drm_device *dev,
 400 +                      struct vc4_exec_info *exec,
 401 +                      struct vc4_shader_state *state)
 402  {
 403         uint32_t *src_handles;
 404         void *pkt_u, *pkt_v;
 405 -       enum shader_rec_reloc_type {
 406 -               RELOC_CODE,
 407 -               RELOC_VBO,
 408 -       };
 409 -       struct shader_rec_reloc {
 410 -               enum shader_rec_reloc_type type;
 411 -               uint32_t offset;
 412 -       };
 413 -       static const struct shader_rec_reloc gl_relocs[] = {
 414 -               { RELOC_CODE, 4 },  /* fs */
 415 -               { RELOC_CODE, 16 }, /* vs */
 416 -               { RELOC_CODE, 28 }, /* cs */
 417 -       };
 418 -       static const struct shader_rec_reloc nv_relocs[] = {
 419 -               { RELOC_CODE, 4 }, /* fs */
 420 -               { RELOC_VBO, 12 }
 421 +       static const uint32_t shader_reloc_offsets[] = {
 422 +               4, /* fs */
 423 +               16, /* vs */
 424 +               28, /* cs */
 425         };
 426 -       const struct shader_rec_reloc *relocs;
 427 -       struct drm_gem_cma_object *bo[ARRAY_SIZE(gl_relocs) + 8];
 428 -       uint32_t nr_attributes = 0, nr_fixed_relocs, nr_relocs, packet_size;
 429 +       uint32_t shader_reloc_count = ARRAY_SIZE(shader_reloc_offsets);
 430 +       struct drm_gem_cma_object *bo[shader_reloc_count + 8];
 431 +       uint32_t nr_attributes, nr_relocs, packet_size;
 432         int i;
 433 -       struct vc4_validated_shader_info *shader;
 434
 435 -       if (state->packet == VC4_PACKET_NV_SHADER_STATE) {
 436 -               relocs = nv_relocs;
 437 -               nr_fixed_relocs = ARRAY_SIZE(nv_relocs);
 438 -
 439 -               packet_size = 16;
 440 -       } else {
 441 -               relocs = gl_relocs;
 442 -               nr_fixed_relocs = ARRAY_SIZE(gl_relocs);
 443 -
 444 -               nr_attributes = state->addr & 0x7;
 445 -               if (nr_attributes == 0)
 446 -                       nr_attributes = 8;
 447 -               packet_size = gl_shader_rec_size(state->addr);
 448 -       }
 449 -       nr_relocs = nr_fixed_relocs + nr_attributes;
 450 +       nr_attributes = state->addr & 0x7;
 451 +       if (nr_attributes == 0)
 452 +               nr_attributes = 8;
 453 +       packet_size = gl_shader_rec_size(state->addr);
 454
 455 +       nr_relocs = ARRAY_SIZE(shader_reloc_offsets) + nr_attributes;
 456         if (nr_relocs * 4 > exec->shader_rec_size) {
 457                 DRM_ERROR("overflowed shader recs reading %d handles "
 458                           "from %d bytes left\n",
 459 @@ -829,21 +775,30 @@ validate_shader_rec(struct drm_device *d
 460         exec->shader_rec_v += roundup(packet_size, 16);
 461         exec->shader_rec_size -= packet_size;
 462
 463 -       for (i = 0; i < nr_relocs; i++) {
 464 -               enum vc4_bo_mode mode;
 465 +       if (!(*(uint16_t *)pkt_u & VC4_SHADER_FLAG_FS_SINGLE_THREAD)) {
 466 +               DRM_ERROR("Multi-threaded fragment shaders not supported.\n");
 467 +               return -EINVAL;
 468 +       }
 469
 470 -               if (i < nr_fixed_relocs && relocs[i].type == RELOC_CODE)
 471 -                       mode = VC4_MODE_SHADER;
 472 -               else
 473 -                       mode = VC4_MODE_RENDER;
 474 +       for (i = 0; i < shader_reloc_count; i++) {
 475 +               if (src_handles[i] > exec->bo_count) {
 476 +                       DRM_ERROR("Shader handle %d too big\n", src_handles[i]);
 477 +                       return -EINVAL;
 478 +               }
 479
 480 -               if (!vc4_use_bo(exec, src_handles[i], mode, &bo[i]))
 481 -                       return false;
 482 +               bo[i] = exec->bo[src_handles[i]];
 483 +               if (!bo[i])
 484 +                       return -EINVAL;
 485 +       }
 486 +       for (i = shader_reloc_count; i < nr_relocs; i++) {
 487 +               bo[i] = vc4_use_bo(exec, src_handles[i]);
 488 +               if (!bo[i])
 489 +                       return -EINVAL;
 490         }
 491
 492 -       for (i = 0; i < nr_fixed_relocs; i++) {
 493 -               struct vc4_bo *vc4_bo;
 494 -               uint32_t o = relocs[i].offset;
 495 +       for (i = 0; i < shader_reloc_count; i++) {
 496 +               struct vc4_validated_shader_info *validated_shader;
 497 +               uint32_t o = shader_reloc_offsets[i];
 498                 uint32_t src_offset = *(uint32_t *)(pkt_u + o);
 499                 uint32_t *texture_handles_u;
 500                 void *uniform_data_u;
 501 @@ -851,57 +806,50 @@ validate_shader_rec(struct drm_device *d
 502
 503                 *(uint32_t *)(pkt_v + o) = bo[i]->paddr + src_offset;
 504
 505 -               switch (relocs[i].type) {
 506 -               case RELOC_CODE:
 507 -                       if (src_offset != 0) {
 508 -                               DRM_ERROR("Shaders must be at offset 0 "
 509 -                                         "of the BO.\n");
 510 -                               goto fail;
 511 -                       }
 512 +               if (src_offset != 0) {
 513 +                       DRM_ERROR("Shaders must be at offset 0 of "
 514 +                                 "the BO.\n");
 515 +                       return -EINVAL;
 516 +               }
 517
 518 -                       vc4_bo = to_vc4_bo(&bo[i]->base);
 519 -                       shader = vc4_bo->validated_shader;
 520 -                       if (!shader)
 521 -                               goto fail;
 522 +               validated_shader = to_vc4_bo(&bo[i]->base)->validated_shader;
 523 +               if (!validated_shader)
 524 +                       return -EINVAL;
 525
 526 -                       if (shader->uniforms_src_size > exec->uniforms_size) {
 527 -                               DRM_ERROR("Uniforms src buffer overflow\n");
 528 -                               goto fail;
 529 -                       }
 530 +               if (validated_shader->uniforms_src_size >
 531 +                   exec->uniforms_size) {
 532 +                       DRM_ERROR("Uniforms src buffer overflow\n");
 533 +                       return -EINVAL;
 534 +               }
 535
 536 -                       texture_handles_u = exec->uniforms_u;
 537 -                       uniform_data_u = (texture_handles_u +
 538 -                                         shader->num_texture_samples);
 539 -
 540 -                       memcpy(exec->uniforms_v, uniform_data_u,
 541 -                              shader->uniforms_size);
 542 -
 543 -                       for (tex = 0;
 544 -                            tex < shader->num_texture_samples;
 545 -                            tex++) {
 546 -                               if (!reloc_tex(exec,
 547 -                                              uniform_data_u,
 548 -                                              &shader->texture_samples[tex],
 549 -                                              texture_handles_u[tex])) {
 550 -                                       goto fail;
 551 -                               }
 552 +               texture_handles_u = exec->uniforms_u;
 553 +               uniform_data_u = (texture_handles_u +
 554 +                                 validated_shader->num_texture_samples);
 555 +
 556 +               memcpy(exec->uniforms_v, uniform_data_u,
 557 +                      validated_shader->uniforms_size);
 558 +
 559 +               for (tex = 0;
 560 +                    tex < validated_shader->num_texture_samples;
 561 +                    tex++) {
 562 +                       if (!reloc_tex(exec,
 563 +                                      uniform_data_u,
 564 +                                      &validated_shader->texture_samples[tex],
 565 +                                      texture_handles_u[tex])) {
 566 +                               return -EINVAL;
 567                         }
 568 +               }
 569
 570 -                       *(uint32_t *)(pkt_v + o + 4) = exec->uniforms_p;
 571 -
 572 -                       exec->uniforms_u += shader->uniforms_src_size;
 573 -                       exec->uniforms_v += shader->uniforms_size;
 574 -                       exec->uniforms_p += shader->uniforms_size;
 575 -
 576 -                       break;
 577 +               *(uint32_t *)(pkt_v + o + 4) = exec->uniforms_p;
 578
 579 -               case RELOC_VBO:
 580 -                       break;
 581 -               }
 582 +               exec->uniforms_u += validated_shader->uniforms_src_size;
 583 +               exec->uniforms_v += validated_shader->uniforms_size;
 584 +               exec->uniforms_p += validated_shader->uniforms_size;
 585         }
 586
 587         for (i = 0; i < nr_attributes; i++) {
 588 -               struct drm_gem_cma_object *vbo = bo[nr_fixed_relocs + i];
 589 +               struct drm_gem_cma_object *vbo =
 590 +                       bo[ARRAY_SIZE(shader_reloc_offsets) + i];
 591                 uint32_t o = 36 + i * 8;
 592                 uint32_t offset = *(uint32_t *)(pkt_u + o + 0);
 593                 uint32_t attr_size = *(uint8_t *)(pkt_u + o + 4) + 1;
 594 @@ -933,9 +881,6 @@ validate_shader_rec(struct drm_device *d
 595         }
 596
 597         return 0;
 598 -
 599 -fail:
 600 -       return -EINVAL;
 601  }
 602
 603  int
 604 @@ -946,7 +891,7 @@ vc4_validate_shader_recs(struct drm_devi
 605         int ret = 0;
 606
 607         for (i = 0; i < exec->shader_state_count; i++) {
 608 -               ret = validate_shader_rec(dev, exec, &exec->shader_state[i]);
 609 +               ret = validate_gl_shader_rec(dev, exec, &exec->shader_state[i]);
 610                 if (ret)
 611                         return ret;
 612         }