void *validated, \
void *untrusted
-
/** Return the width in pixels of a 64-byte microtile. */
static uint32_t
utile_width(int cpp)
height <= 4 * utile_height(cpp));
}
-static bool
-vc4_use_bo(struct vc4_exec_info *exec,
- uint32_t hindex,
- enum vc4_bo_mode mode,
- struct drm_gem_cma_object **obj)
+struct drm_gem_cma_object *
+vc4_use_bo(struct vc4_exec_info *exec, uint32_t hindex)
{
- *obj = NULL;
+ struct drm_gem_cma_object *obj;
+ struct drm_vc4_bo *bo;
if (hindex >= exec->bo_count) {
DRM_ERROR("BO index %d greater than BO count %d\n",
hindex, exec->bo_count);
- return false;
+ return NULL;
}
+ obj = exec->bo[hindex];
+ bo = to_vc4_bo(&obj->base);
- if (exec->bo[hindex].mode != mode) {
- if (exec->bo[hindex].mode == VC4_MODE_UNDECIDED) {
- exec->bo[hindex].mode = mode;
- } else {
- DRM_ERROR("BO index %d reused with mode %d vs %d\n",
- hindex, exec->bo[hindex].mode, mode);
- return false;
- }
+ if (bo->validated_shader) {
+ DRM_ERROR("Trying to use shader BO as something other than "
+ "a shader\n");
+ return NULL;
}
- *obj = exec->bo[hindex].bo;
- return true;
+ return obj;
+}
+
+static struct drm_gem_cma_object *
+vc4_use_handle(struct vc4_exec_info *exec, uint32_t gem_handles_packet_index)
+{
+ return vc4_use_bo(exec, exec->bo_index[gem_handles_packet_index]);
}
static bool
-vc4_use_handle(struct vc4_exec_info *exec,
- uint32_t gem_handles_packet_index,
- enum vc4_bo_mode mode,
- struct drm_gem_cma_object **obj)
+validate_bin_pos(struct vc4_exec_info *exec, void *untrusted, uint32_t pos)
{
- return vc4_use_bo(exec, exec->bo_index[gem_handles_packet_index],
- mode, obj);
+ /* Note that the untrusted pointer passed to these functions is
+ * incremented past the packet byte.
+ */
+ return (untrusted - 1 == exec->bin_u + pos);
}
static uint32_t
return 36 + attribute_count * 8;
}
-static bool
-check_tex_size(struct vc4_exec_info *exec, struct drm_gem_cma_object *fbo,
- uint32_t offset, uint8_t tiling_format,
- uint32_t width, uint32_t height, uint8_t cpp)
+bool
+vc4_check_tex_size(struct vc4_exec_info *exec, struct drm_gem_cma_object *fbo,
+ uint32_t offset, uint8_t tiling_format,
+ uint32_t width, uint32_t height, uint8_t cpp)
{
uint32_t aligned_width, aligned_height, stride, size;
uint32_t utile_w = utile_width(cpp);
if (size + offset < size ||
size + offset > fbo->base.size) {
- DRM_ERROR("Overflow in %dx%d (%dx%d) fbo size (%d + %d > %d)\n",
+ DRM_ERROR("Overflow in %dx%d (%dx%d) fbo size (%d + %d > %zd)\n",
width, height,
aligned_width, aligned_height,
size, offset, fbo->base.size);
}
static int
-validate_flush_all(VALIDATE_ARGS)
+validate_flush(VALIDATE_ARGS)
{
- if (exec->found_increment_semaphore_packet) {
- DRM_ERROR("VC4_PACKET_FLUSH_ALL after "
- "VC4_PACKET_INCREMENT_SEMAPHORE\n");
+ if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 1)) {
+ DRM_ERROR("Bin CL must end with VC4_PACKET_FLUSH\n");
return -EINVAL;
}
+ exec->found_flush = true;
return 0;
}
static int
validate_increment_semaphore(VALIDATE_ARGS)
{
- if (exec->found_increment_semaphore_packet) {
- DRM_ERROR("Duplicate VC4_PACKET_INCREMENT_SEMAPHORE\n");
- return -EINVAL;
- }
- exec->found_increment_semaphore_packet = true;
-
- /* Once we've found the semaphore increment, there should be one FLUSH
- * then the end of the command list. The FLUSH actually triggers the
- * increment, so we only need to make sure there
- */
-
- return 0;
-}
-
-static int
-validate_wait_on_semaphore(VALIDATE_ARGS)
-{
- if (exec->found_wait_on_semaphore_packet) {
- DRM_ERROR("Duplicate VC4_PACKET_WAIT_ON_SEMAPHORE\n");
- return -EINVAL;
- }
- exec->found_wait_on_semaphore_packet = true;
-
- if (!exec->found_increment_semaphore_packet) {
- DRM_ERROR("VC4_PACKET_WAIT_ON_SEMAPHORE without "
+ if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 2)) {
+ DRM_ERROR("Bin CL must end with "
"VC4_PACKET_INCREMENT_SEMAPHORE\n");
return -EINVAL;
}
-
- return 0;
-}
-
-static int
-validate_branch_to_sublist(VALIDATE_ARGS)
-{
- uint32_t offset;
-
- if (!exec->tile_alloc_bo) {
- DRM_ERROR("VC4_PACKET_BRANCH_TO_SUB_LIST seen before "
- "binner setup\n");
- return -EINVAL;
- }
-
- if (!exec->found_wait_on_semaphore_packet) {
- DRM_ERROR("Jumping to tile alloc before binning finished.\n");
- return -EINVAL;
- }
-
- offset = *(uint32_t *)(untrusted + 0);
- if (offset & exec->tile_alloc_init_block_mask ||
- offset > exec->tile_alloc_init_block_last) {
- DRM_ERROR("VC4_PACKET_BRANCH_TO_SUB_LIST must jump to initial "
- "tile allocation space.\n");
- return -EINVAL;
- }
-
- *(uint32_t *)(validated + 0) = exec->tile_alloc_bo->paddr + offset;
-
- return 0;
-}
-
-/**
- * validate_loadstore_tile_buffer_general() - Validation for
- * VC4_PACKET_LOAD_TILE_BUFFER_GENERAL and
- * VC4_PACKET_STORE_TILE_BUFFER_GENERAL.
- *
- * The two packets are nearly the same, except for the TLB-clearing management
- * bits not being present for loads. Additionally, while stores are executed
- * immediately (using the current tile coordinates), loads are queued to be
- * executed when the tile coordinates packet occurs.
- *
- * Note that coordinates packets are validated to be within the declared
- * bin_x/y, which themselves are verified to match the rendering-configuration
- * FB width and height (which the hardware uses to clip loads and stores).
- */
-static int
-validate_loadstore_tile_buffer_general(VALIDATE_ARGS)
-{
- uint32_t packet_b0 = *(uint8_t *)(untrusted + 0);
- uint32_t packet_b1 = *(uint8_t *)(untrusted + 1);
- struct drm_gem_cma_object *fbo;
- uint32_t buffer_type = packet_b0 & 0xf;
- uint32_t untrusted_address, offset, cpp;
-
- switch (buffer_type) {
- case VC4_LOADSTORE_TILE_BUFFER_NONE:
- return 0;
- case VC4_LOADSTORE_TILE_BUFFER_COLOR:
- if ((packet_b1 & VC4_LOADSTORE_TILE_BUFFER_MASK) ==
- VC4_LOADSTORE_TILE_BUFFER_RGBA8888) {
- cpp = 4;
- } else {
- cpp = 2;
- }
- break;
-
- case VC4_LOADSTORE_TILE_BUFFER_Z:
- case VC4_LOADSTORE_TILE_BUFFER_ZS:
- cpp = 4;
- break;
-
- default:
- DRM_ERROR("Load/store type %d unsupported\n", buffer_type);
- return -EINVAL;
- }
-
- if (!vc4_use_handle(exec, 0, VC4_MODE_RENDER, &fbo))
- return -EINVAL;
-
- untrusted_address = *(uint32_t *)(untrusted + 2);
- offset = untrusted_address & ~0xf;
-
- if (!check_tex_size(exec, fbo, offset,
- ((packet_b0 &
- VC4_LOADSTORE_TILE_BUFFER_FORMAT_MASK) >>
- VC4_LOADSTORE_TILE_BUFFER_FORMAT_SHIFT),
- exec->fb_width, exec->fb_height, cpp)) {
- return -EINVAL;
- }
-
- *(uint32_t *)(validated + 2) = (offset + fbo->paddr +
- (untrusted_address & 0xf));
+ exec->found_increment_semaphore_packet = true;
return 0;
}
uint32_t index_size = (*(uint8_t *)(untrusted + 0) >> 4) ? 2 : 1;
struct vc4_shader_state *shader_state;
- if (exec->found_increment_semaphore_packet) {
- DRM_ERROR("Drawing after VC4_PACKET_INCREMENT_SEMAPHORE\n");
- return -EINVAL;
- }
-
/* Check overflow condition */
if (exec->shader_state_count == 0) {
DRM_ERROR("shader state must precede primitives\n");
if (max_index > shader_state->max_index)
shader_state->max_index = max_index;
- if (!vc4_use_handle(exec, 0, VC4_MODE_RENDER, &ib))
+ ib = vc4_use_handle(exec, 0);
+ if (!ib)
return -EINVAL;
if (offset > ib->base.size ||
(ib->base.size - offset) / index_size < length) {
- DRM_ERROR("IB access overflow (%d + %d*%d > %d)\n",
+ DRM_ERROR("IB access overflow (%d + %d*%d > %zd)\n",
offset, length, index_size, ib->base.size);
return -EINVAL;
}
uint32_t max_index;
struct vc4_shader_state *shader_state;
- if (exec->found_increment_semaphore_packet) {
- DRM_ERROR("Drawing after VC4_PACKET_INCREMENT_SEMAPHORE\n");
- return -EINVAL;
- }
-
/* Check overflow condition */
if (exec->shader_state_count == 0) {
DRM_ERROR("shader state must precede primitives\n");
return -EINVAL;
}
- exec->shader_state[i].packet = VC4_PACKET_GL_SHADER_STATE;
exec->shader_state[i].addr = *(uint32_t *)untrusted;
exec->shader_state[i].max_index = 0;
return 0;
}
-static int
-validate_nv_shader_state(VALIDATE_ARGS)
-{
- uint32_t i = exec->shader_state_count++;
-
- if (i >= exec->shader_state_size) {
- DRM_ERROR("More requests for shader states than declared\n");
- return -EINVAL;
- }
-
- exec->shader_state[i].packet = VC4_PACKET_NV_SHADER_STATE;
- exec->shader_state[i].addr = *(uint32_t *)untrusted;
-
- if (exec->shader_state[i].addr & 15) {
- DRM_ERROR("NV shader state address 0x%08x misaligned\n",
- exec->shader_state[i].addr);
- return -EINVAL;
- }
-
- *(uint32_t *)validated = (exec->shader_state[i].addr +
- exec->shader_rec_p);
-
- return 0;
-}
-
static int
validate_tile_binning_config(VALIDATE_ARGS)
{
- struct drm_gem_cma_object *tile_allocation;
- struct drm_gem_cma_object *tile_state_data_array;
+ struct drm_device *dev = exec->exec_bo->base.dev;
uint8_t flags;
- uint32_t tile_allocation_size;
- uint32_t tile_alloc_init_block_size;
-
- if (!vc4_use_handle(exec, 0, VC4_MODE_TILE_ALLOC, &tile_allocation) ||
- !vc4_use_handle(exec, 1, VC4_MODE_TSDA, &tile_state_data_array))
- return -EINVAL;
+ uint32_t tile_state_size, tile_alloc_size;
+ uint32_t tile_count;
if (exec->found_tile_binning_mode_config_packet) {
DRM_ERROR("Duplicate VC4_PACKET_TILE_BINNING_MODE_CONFIG\n");
exec->bin_tiles_x = *(uint8_t *)(untrusted + 12);
exec->bin_tiles_y = *(uint8_t *)(untrusted + 13);
+ tile_count = exec->bin_tiles_x * exec->bin_tiles_y;
flags = *(uint8_t *)(untrusted + 14);
if (exec->bin_tiles_x == 0 ||
return -EINVAL;
}
- /* Our validation relies on the user not getting to set up their own
- * tile state/tile allocation BO contents.
- */
- if (!(flags & VC4_BIN_CONFIG_AUTO_INIT_TSDA)) {
- DRM_ERROR("binning config missing "
- "VC4_BIN_CONFIG_AUTO_INIT_TSDA\n");
- return -EINVAL;
- }
-
if (flags & (VC4_BIN_CONFIG_DB_NON_MS |
- VC4_BIN_CONFIG_TILE_BUFFER_64BIT |
- VC4_BIN_CONFIG_MS_MODE_4X)) {
- DRM_ERROR("unsupported bining config flags 0x%02x\n", flags);
- return -EINVAL;
- }
-
- if (*(uint32_t *)(untrusted + 0) != 0) {
- DRM_ERROR("tile allocation offset != 0 unsupported\n");
- return -EINVAL;
- }
- tile_allocation_size = *(uint32_t *)(untrusted + 4);
- if (tile_allocation_size > tile_allocation->base.size) {
- DRM_ERROR("tile allocation size %d > BO size %d\n",
- tile_allocation_size, tile_allocation->base.size);
- return -EINVAL;
- }
- *(uint32_t *)validated = tile_allocation->paddr;
- exec->tile_alloc_bo = tile_allocation;
-
- tile_alloc_init_block_size = 1 << (5 + ((flags >> 5) & 3));
- if (exec->bin_tiles_x * exec->bin_tiles_y *
- tile_alloc_init_block_size > tile_allocation_size) {
- DRM_ERROR("tile init exceeds tile alloc size (%d vs %d)\n",
- exec->bin_tiles_x * exec->bin_tiles_y *
- tile_alloc_init_block_size,
- tile_allocation_size);
- return -EINVAL;
- }
- exec->tile_alloc_init_block_mask = tile_alloc_init_block_size - 1;
- exec->tile_alloc_init_block_last = tile_alloc_init_block_size *
- (exec->bin_tiles_x * exec->bin_tiles_y - 1);
-
- if (*(uint32_t *)(untrusted + 8) != 0) {
- DRM_ERROR("TSDA offset != 0 unsupported\n");
- return -EINVAL;
- }
- if (exec->bin_tiles_x * exec->bin_tiles_y * 48 >
- tile_state_data_array->base.size) {
- DRM_ERROR("TSDA of %db too small for %dx%d bin config\n",
- tile_state_data_array->base.size,
- exec->bin_tiles_x, exec->bin_tiles_y);
- }
- *(uint32_t *)(validated + 8) = tile_state_data_array->paddr;
-
- return 0;
-}
-
-static int
-validate_tile_rendering_mode_config(VALIDATE_ARGS)
-{
- struct drm_gem_cma_object *fbo;
- uint32_t flags, offset, cpp;
-
- if (exec->found_tile_rendering_mode_config_packet) {
- DRM_ERROR("Duplicate VC4_PACKET_TILE_RENDERING_MODE_CONFIG\n");
+ VC4_BIN_CONFIG_TILE_BUFFER_64BIT)) {
+ DRM_ERROR("unsupported binning config flags 0x%02x\n", flags);
return -EINVAL;
}
- exec->found_tile_rendering_mode_config_packet = true;
-
- if (!vc4_use_handle(exec, 0, VC4_MODE_RENDER, &fbo))
- return -EINVAL;
-
- exec->fb_width = *(uint16_t *)(untrusted + 4);
- exec->fb_height = *(uint16_t *)(untrusted + 6);
-
- flags = *(uint16_t *)(untrusted + 8);
- if ((flags & VC4_RENDER_CONFIG_FORMAT_MASK) ==
- VC4_RENDER_CONFIG_FORMAT_RGBA8888) {
- cpp = 4;
- } else {
- cpp = 2;
- }
-
- offset = *(uint32_t *)untrusted;
- if (!check_tex_size(exec, fbo, offset,
- ((flags &
- VC4_RENDER_CONFIG_MEMORY_FORMAT_MASK) >>
- VC4_RENDER_CONFIG_MEMORY_FORMAT_SHIFT),
- exec->fb_width, exec->fb_height, cpp)) {
- return -EINVAL;
- }
-
- *(uint32_t *)validated = fbo->paddr + offset;
-
- return 0;
-}
-static int
-validate_tile_coordinates(VALIDATE_ARGS)
-{
- uint8_t tile_x = *(uint8_t *)(untrusted + 0);
- uint8_t tile_y = *(uint8_t *)(untrusted + 1);
+ /* The tile state data array is 48 bytes per tile, and we put it at
+ * the start of a BO containing both it and the tile alloc.
+ */
+ tile_state_size = 48 * tile_count;
+
+ /* Since the tile alloc array will follow us, align. */
+ exec->tile_alloc_offset = roundup(tile_state_size, 4096);
+
+ *(uint8_t *)(validated + 14) =
+ ((flags & ~(VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_MASK |
+ VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_MASK)) |
+ VC4_BIN_CONFIG_AUTO_INIT_TSDA |
+ VC4_SET_FIELD(VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_32,
+ VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE) |
+ VC4_SET_FIELD(VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_128,
+ VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE));
+
+ /* Initial block size. */
+ tile_alloc_size = 32 * tile_count;
+
+ /*
+ * The initial allocation gets rounded to the next 256 bytes before
+ * the hardware starts fulfilling further allocations.
+ */
+ tile_alloc_size = roundup(tile_alloc_size, 256);
- if (tile_x * 64 >= exec->fb_width || tile_y * 64 >= exec->fb_height) {
- DRM_ERROR("Tile coordinates %d,%d > render config %dx%d\n",
- tile_x, tile_y, exec->fb_width, exec->fb_height);
- return -EINVAL;
- }
+ /* Add space for the extra allocations. This is what gets used first,
+ * before overflow memory. It must have at least 4096 bytes, but we
+ * want to avoid overflow memory usage if possible.
+ */
+ tile_alloc_size += 1024 * 1024;
+
+ exec->tile_bo = drm_gem_cma_create(dev, exec->tile_alloc_offset +
+ tile_alloc_size);
+ if (!exec->tile_bo)
+ return -ENOMEM;
+ list_addtail(&to_vc4_bo(&exec->tile_bo->base)->unref_head,
+ &exec->unref_list);
+
+ /* tile alloc address. */
+ *(uint32_t *)(validated + 0) = (exec->tile_bo->paddr +
+ exec->tile_alloc_offset);
+ /* tile alloc size. */
+ *(uint32_t *)(validated + 4) = tile_alloc_size;
+ /* tile state address. */
+ *(uint32_t *)(validated + 8) = exec->tile_bo->paddr;
return 0;
}
return 0;
}
+#define VC4_DEFINE_PACKET(packet, func) \
+ [packet] = { packet ## _SIZE, #packet, func }
+
static const struct cmd_info {
- bool bin;
- bool render;
uint16_t len;
const char *name;
int (*func)(struct vc4_exec_info *exec, void *validated,
void *untrusted);
} cmd_info[] = {
- [VC4_PACKET_HALT] = { 1, 1, 1, "halt", NULL },
- [VC4_PACKET_NOP] = { 1, 1, 1, "nop", NULL },
- [VC4_PACKET_FLUSH] = { 1, 1, 1, "flush", NULL },
- [VC4_PACKET_FLUSH_ALL] = { 1, 0, 1, "flush all state", validate_flush_all },
- [VC4_PACKET_START_TILE_BINNING] = { 1, 0, 1, "start tile binning", validate_start_tile_binning },
- [VC4_PACKET_INCREMENT_SEMAPHORE] = { 1, 0, 1, "increment semaphore", validate_increment_semaphore },
- [VC4_PACKET_WAIT_ON_SEMAPHORE] = { 0, 1, 1, "wait on semaphore", validate_wait_on_semaphore },
- /* BRANCH_TO_SUB_LIST is actually supported in the binner as well, but
- * we only use it from the render CL in order to jump into the tile
- * allocation BO.
- */
- [VC4_PACKET_BRANCH_TO_SUB_LIST] = { 0, 1, 5, "branch to sublist", validate_branch_to_sublist },
- [VC4_PACKET_STORE_MS_TILE_BUFFER] = { 0, 1, 1, "store MS resolved tile color buffer", NULL },
- [VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF] = { 0, 1, 1, "store MS resolved tile color buffer and EOF", NULL },
-
- [VC4_PACKET_STORE_TILE_BUFFER_GENERAL] = { 0, 1, 7, "Store Tile Buffer General", validate_loadstore_tile_buffer_general },
- [VC4_PACKET_LOAD_TILE_BUFFER_GENERAL] = { 0, 1, 7, "Load Tile Buffer General", validate_loadstore_tile_buffer_general },
-
- [VC4_PACKET_GL_INDEXED_PRIMITIVE] = { 1, 1, 14, "Indexed Primitive List", validate_indexed_prim_list },
-
- [VC4_PACKET_GL_ARRAY_PRIMITIVE] = { 1, 1, 10, "Vertex Array Primitives", validate_gl_array_primitive },
-
- /* This is only used by clipped primitives (packets 48 and 49), which
- * we don't support parsing yet.
- */
- [VC4_PACKET_PRIMITIVE_LIST_FORMAT] = { 1, 1, 2, "primitive list format", NULL },
-
- [VC4_PACKET_GL_SHADER_STATE] = { 1, 1, 5, "GL Shader State", validate_gl_shader_state },
- [VC4_PACKET_NV_SHADER_STATE] = { 1, 1, 5, "NV Shader State", validate_nv_shader_state },
-
- [VC4_PACKET_CONFIGURATION_BITS] = { 1, 1, 4, "configuration bits", NULL },
- [VC4_PACKET_FLAT_SHADE_FLAGS] = { 1, 1, 5, "flat shade flags", NULL },
- [VC4_PACKET_POINT_SIZE] = { 1, 1, 5, "point size", NULL },
- [VC4_PACKET_LINE_WIDTH] = { 1, 1, 5, "line width", NULL },
- [VC4_PACKET_RHT_X_BOUNDARY] = { 1, 1, 3, "RHT X boundary", NULL },
- [VC4_PACKET_DEPTH_OFFSET] = { 1, 1, 5, "Depth Offset", NULL },
- [VC4_PACKET_CLIP_WINDOW] = { 1, 1, 9, "Clip Window", NULL },
- [VC4_PACKET_VIEWPORT_OFFSET] = { 1, 1, 5, "Viewport Offset", NULL },
- [VC4_PACKET_CLIPPER_XY_SCALING] = { 1, 1, 9, "Clipper XY Scaling", NULL },
+ VC4_DEFINE_PACKET(VC4_PACKET_HALT, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_NOP, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_FLUSH, validate_flush),
+ VC4_DEFINE_PACKET(VC4_PACKET_FLUSH_ALL, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_START_TILE_BINNING,
+ validate_start_tile_binning),
+ VC4_DEFINE_PACKET(VC4_PACKET_INCREMENT_SEMAPHORE,
+ validate_increment_semaphore),
+
+ VC4_DEFINE_PACKET(VC4_PACKET_GL_INDEXED_PRIMITIVE,
+ validate_indexed_prim_list),
+ VC4_DEFINE_PACKET(VC4_PACKET_GL_ARRAY_PRIMITIVE,
+ validate_gl_array_primitive),
+
+ VC4_DEFINE_PACKET(VC4_PACKET_PRIMITIVE_LIST_FORMAT, NULL),
+
+ VC4_DEFINE_PACKET(VC4_PACKET_GL_SHADER_STATE, validate_gl_shader_state),
+
+ VC4_DEFINE_PACKET(VC4_PACKET_CONFIGURATION_BITS, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_FLAT_SHADE_FLAGS, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_POINT_SIZE, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_LINE_WIDTH, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_RHT_X_BOUNDARY, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_DEPTH_OFFSET, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_CLIP_WINDOW, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_VIEWPORT_OFFSET, NULL),
+ VC4_DEFINE_PACKET(VC4_PACKET_CLIPPER_XY_SCALING, NULL),
/* Note: The docs say this was also 105, but it was 106 in the
* initial userland code drop.
*/
- [VC4_PACKET_CLIPPER_Z_SCALING] = { 1, 1, 9, "Clipper Z Scale and Offset", NULL },
+ VC4_DEFINE_PACKET(VC4_PACKET_CLIPPER_Z_SCALING, NULL),
- [VC4_PACKET_TILE_BINNING_MODE_CONFIG] = { 1, 0, 16, "tile binning configuration", validate_tile_binning_config },
+ VC4_DEFINE_PACKET(VC4_PACKET_TILE_BINNING_MODE_CONFIG,
+ validate_tile_binning_config),
- [VC4_PACKET_TILE_RENDERING_MODE_CONFIG] = { 0, 1, 11, "tile rendering mode configuration", validate_tile_rendering_mode_config},
-
- [VC4_PACKET_CLEAR_COLORS] = { 0, 1, 14, "Clear Colors", NULL },
-
- [VC4_PACKET_TILE_COORDINATES] = { 0, 1, 3, "Tile Coordinates", validate_tile_coordinates },
-
- [VC4_PACKET_GEM_HANDLES] = { 1, 1, 9, "GEM handles", validate_gem_handles },
+ VC4_DEFINE_PACKET(VC4_PACKET_GEM_HANDLES, validate_gem_handles),
};
int
-vc4_validate_cl(struct drm_device *dev,
- void *validated,
- void *unvalidated,
- uint32_t len,
- bool is_bin,
- bool has_bin,
- struct vc4_exec_info *exec)
+vc4_validate_bin_cl(struct drm_device *dev,
+ void *validated,
+ void *unvalidated,
+ struct vc4_exec_info *exec)
{
+ uint32_t len = exec->args->bin_cl_size;
uint32_t dst_offset = 0;
uint32_t src_offset = 0;
u8 cmd = *(uint8_t *)src_pkt;
const struct cmd_info *info;
- if (cmd > ARRAY_SIZE(cmd_info)) {
+ if (cmd >= ARRAY_SIZE(cmd_info)) {
DRM_ERROR("0x%08x: packet %d out of bounds\n",
src_offset, cmd);
return -EINVAL;
return -EINVAL;
}
-#if 0
- DRM_INFO("0x%08x: packet %d (%s) size %d processing...\n",
- src_offset, cmd, info->name, info->len);
-#endif
-
- if ((is_bin && !info->bin) ||
- (!is_bin && !info->render)) {
- DRM_ERROR("0x%08x: packet %d (%s) invalid for %s\n",
- src_offset, cmd, info->name,
- is_bin ? "binner" : "render");
- return -EINVAL;
- }
-
if (src_offset + info->len > len) {
DRM_ERROR("0x%08x: packet %d (%s) length 0x%08x "
"exceeds bounds (0x%08x)\n",
if (info->func && info->func(exec,
dst_pkt + 1,
src_pkt + 1)) {
- DRM_ERROR("0x%08x: packet %d (%s) failed to "
- "validate\n",
+ DRM_ERROR("0x%08x: packet %d (%s) failed to validate\n",
src_offset, cmd, info->name);
return -EINVAL;
}
break;
}
- if (is_bin) {
- exec->ct0ea = exec->ct0ca + dst_offset;
+ exec->ct0ea = exec->ct0ca + dst_offset;
- if (has_bin && !exec->found_start_tile_binning_packet) {
- DRM_ERROR("Bin CL missing VC4_PACKET_START_TILE_BINNING\n");
- return -EINVAL;
- }
- } else {
- if (!exec->found_tile_rendering_mode_config_packet) {
- DRM_ERROR("Render CL missing VC4_PACKET_TILE_RENDERING_MODE_CONFIG\n");
- return -EINVAL;
- }
+ if (!exec->found_start_tile_binning_packet) {
+ DRM_ERROR("Bin CL missing VC4_PACKET_START_TILE_BINNING\n");
+ return -EINVAL;
+ }
- /* Make sure that they actually consumed the semaphore
- * increment from the bin CL. Otherwise a later submit would
- * have render execute immediately.
- */
- if (exec->found_wait_on_semaphore_packet != has_bin) {
- DRM_ERROR("Render CL %s VC4_PACKET_WAIT_ON_SEMAPHORE\n",
- exec->found_wait_on_semaphore_packet ?
- "has" : "missing");
- return -EINVAL;
- }
- exec->ct1ea = exec->ct1ca + dst_offset;
+ /* The bin CL must be ended with INCREMENT_SEMAPHORE and FLUSH. The
+ * semaphore is used to trigger the render CL to start up, and the
+ * FLUSH is what caps the bin lists with
+ * VC4_PACKET_RETURN_FROM_SUB_LIST (so they jump back to the main
+ * render CL when they get called to) and actually triggers the queued
+ * semaphore increment.
+ */
+ if (!exec->found_increment_semaphore_packet || !exec->found_flush) {
+ DRM_ERROR("Bin CL missing VC4_PACKET_INCREMENT_SEMAPHORE + "
+ "VC4_PACKET_FLUSH\n");
+ return -EINVAL;
}
return 0;
uint32_t p3 = (sample->p_offset[3] != ~0 ?
*(uint32_t *)(uniform_data_u + sample->p_offset[3]) : 0);
uint32_t *validated_p0 = exec->uniforms_v + sample->p_offset[0];
- uint32_t offset = p0 & ~0xfff;
- uint32_t miplevels = (p0 & 15);
- uint32_t width = (p1 >> 8) & 2047;
- uint32_t height = (p1 >> 20) & 2047;
+ uint32_t offset = p0 & VC4_TEX_P0_OFFSET_MASK;
+ uint32_t miplevels = VC4_GET_FIELD(p0, VC4_TEX_P0_MIPLVLS);
+ uint32_t width = VC4_GET_FIELD(p1, VC4_TEX_P1_WIDTH);
+ uint32_t height = VC4_GET_FIELD(p1, VC4_TEX_P1_HEIGHT);
uint32_t cpp, tiling_format, utile_w, utile_h;
uint32_t i;
uint32_t cube_map_stride = 0;
enum vc4_texture_data_type type;
- if (!vc4_use_bo(exec, texture_handle_index, VC4_MODE_RENDER, &tex))
+ tex = vc4_use_bo(exec, texture_handle_index);
+ if (!tex)
return false;
if (sample->is_direct) {
uint32_t remaining_size = tex->base.size - p0;
+
if (p0 > tex->base.size - 4) {
DRM_ERROR("UBO offset greater than UBO size\n");
- return false;
+ goto fail;
}
if (p1 > remaining_size - 4) {
- DRM_ERROR("UBO clamp would allow reads outside of UBO\n");
- return false;
+ DRM_ERROR("UBO clamp would allow reads "
+ "outside of UBO\n");
+ goto fail;
}
*validated_p0 = tex->paddr + p0;
return true;
if (height == 0)
height = 2048;
- if (p0 & (1 << 9)) {
- if ((p2 & (3 << 30)) == (1 << 30))
- cube_map_stride = p2 & 0x3ffff000;
- if ((p3 & (3 << 30)) == (1 << 30)) {
+ if (p0 & VC4_TEX_P0_CMMODE_MASK) {
+ if (VC4_GET_FIELD(p2, VC4_TEX_P2_PTYPE) ==
+ VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE)
+ cube_map_stride = p2 & VC4_TEX_P2_CMST_MASK;
+ if (VC4_GET_FIELD(p3, VC4_TEX_P2_PTYPE) ==
+ VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE) {
if (cube_map_stride) {
DRM_ERROR("Cube map stride set twice\n");
- return false;
+ goto fail;
}
- cube_map_stride = p3 & 0x3ffff000;
+ cube_map_stride = p3 & VC4_TEX_P2_CMST_MASK;
}
if (!cube_map_stride) {
DRM_ERROR("Cube map stride not set\n");
- return false;
+ goto fail;
}
}
- type = ((p0 >> 4) & 15) | ((p1 >> 31) << 4);
+ type = (VC4_GET_FIELD(p0, VC4_TEX_P0_TYPE) |
+ (VC4_GET_FIELD(p1, VC4_TEX_P1_TYPE4) << 4));
switch (type) {
case VC4_TEXTURE_TYPE_RGBA8888:
cpp = 1;
break;
case VC4_TEXTURE_TYPE_ETC1:
+ /* ETC1 is arranged as 64-bit blocks, where each block is 4x4
+ * pixels.
+ */
+ cpp = 8;
+ width = (width + 3) >> 2;
+ height = (height + 3) >> 2;
+ break;
case VC4_TEXTURE_TYPE_BW1:
case VC4_TEXTURE_TYPE_A4:
case VC4_TEXTURE_TYPE_A1:
case VC4_TEXTURE_TYPE_YUV422R:
default:
DRM_ERROR("Texture format %d unsupported\n", type);
- return false;
+ goto fail;
}
utile_w = utile_width(cpp);
utile_h = utile_height(cpp);
tiling_format = VC4_TILING_FORMAT_T;
}
- if (!check_tex_size(exec, tex, offset + cube_map_stride * 5,
- tiling_format, width, height, cpp)) {
- return false;
+ if (!vc4_check_tex_size(exec, tex, offset + cube_map_stride * 5,
+ tiling_format, width, height, cpp)) {
+ goto fail;
}
/* The mipmap levels are stored before the base of the texture. Make
i, level_width, level_height,
aligned_width, aligned_height,
level_size, offset);
- return false;
+ goto fail;
}
offset -= level_size;
*validated_p0 = tex->paddr + p0;
return true;
+ fail:
+ DRM_INFO("Texture p0 at %d: 0x%08x\n", sample->p_offset[0], p0);
+ DRM_INFO("Texture p1 at %d: 0x%08x\n", sample->p_offset[1], p1);
+ DRM_INFO("Texture p2 at %d: 0x%08x\n", sample->p_offset[2], p2);
+ DRM_INFO("Texture p3 at %d: 0x%08x\n", sample->p_offset[3], p3);
+ return false;
}
static int
-validate_shader_rec(struct drm_device *dev,
- struct vc4_exec_info *exec,
- struct vc4_shader_state *state)
+validate_gl_shader_rec(struct drm_device *dev,
+ struct vc4_exec_info *exec,
+ struct vc4_shader_state *state)
{
uint32_t *src_handles;
void *pkt_u, *pkt_v;
- enum shader_rec_reloc_type {
- RELOC_CODE,
- RELOC_VBO,
- };
- struct shader_rec_reloc {
- enum shader_rec_reloc_type type;
- uint32_t offset;
- };
- static const struct shader_rec_reloc gl_relocs[] = {
- { RELOC_CODE, 4 }, /* fs */
- { RELOC_CODE, 16 }, /* vs */
- { RELOC_CODE, 28 }, /* cs */
+ static const uint32_t shader_reloc_offsets[] = {
+ 4, /* fs */
+ 16, /* vs */
+ 28, /* cs */
};
- static const struct shader_rec_reloc nv_relocs[] = {
- { RELOC_CODE, 4 }, /* fs */
- { RELOC_VBO, 12 }
- };
- const struct shader_rec_reloc *relocs;
- struct drm_gem_cma_object *bo[ARRAY_SIZE(gl_relocs) + 8];
- uint32_t nr_attributes = 0, nr_fixed_relocs, nr_relocs, packet_size;
+ uint32_t shader_reloc_count = ARRAY_SIZE(shader_reloc_offsets);
+ struct drm_gem_cma_object *bo[shader_reloc_count + 8];
+ uint32_t nr_attributes, nr_relocs, packet_size;
int i;
- struct vc4_validated_shader_info *validated_shader = NULL;
-
- if (state->packet == VC4_PACKET_NV_SHADER_STATE) {
- relocs = nv_relocs;
- nr_fixed_relocs = ARRAY_SIZE(nv_relocs);
- packet_size = 16;
- } else {
- relocs = gl_relocs;
- nr_fixed_relocs = ARRAY_SIZE(gl_relocs);
-
- nr_attributes = state->addr & 0x7;
- if (nr_attributes == 0)
- nr_attributes = 8;
- packet_size = gl_shader_rec_size(state->addr);
- }
- nr_relocs = nr_fixed_relocs + nr_attributes;
+ nr_attributes = state->addr & 0x7;
+ if (nr_attributes == 0)
+ nr_attributes = 8;
+ packet_size = gl_shader_rec_size(state->addr);
+ nr_relocs = ARRAY_SIZE(shader_reloc_offsets) + nr_attributes;
if (nr_relocs * 4 > exec->shader_rec_size) {
DRM_ERROR("overflowed shader recs reading %d handles "
"from %d bytes left\n",
exec->shader_rec_v += roundup(packet_size, 16);
exec->shader_rec_size -= packet_size;
- for (i = 0; i < nr_relocs; i++) {
- enum vc4_bo_mode mode;
+ for (i = 0; i < shader_reloc_count; i++) {
+ if (src_handles[i] > exec->bo_count) {
+ DRM_ERROR("Shader handle %d too big\n", src_handles[i]);
+ return -EINVAL;
+ }
- if (i < nr_fixed_relocs && relocs[i].type == RELOC_CODE)
- mode = VC4_MODE_SHADER;
- else
- mode = VC4_MODE_RENDER;
+ bo[i] = exec->bo[src_handles[i]];
+ if (!bo[i])
+ return -EINVAL;
+ }
+ for (i = shader_reloc_count; i < nr_relocs; i++) {
+ bo[i] = vc4_use_bo(exec, src_handles[i]);
+ if (!bo[i])
+ return -EINVAL;
+ }
- if (!vc4_use_bo(exec, src_handles[i], mode, &bo[i])) {
- return false;
- }
+ if (((*(uint16_t *)pkt_u & VC4_SHADER_FLAG_FS_SINGLE_THREAD) == 0) !=
+ to_vc4_bo(&bo[0]->base)->validated_shader->is_threaded) {
+ DRM_ERROR("Thread mode of CL and FS do not match\n");
+ return -EINVAL;
+ }
+
+ if (to_vc4_bo(&bo[1]->base)->validated_shader->is_threaded ||
+ to_vc4_bo(&bo[2]->base)->validated_shader->is_threaded) {
+ DRM_ERROR("cs and vs cannot be threaded\n");
+ return -EINVAL;
}
- for (i = 0; i < nr_fixed_relocs; i++) {
- uint32_t o = relocs[i].offset;
+ for (i = 0; i < shader_reloc_count; i++) {
+ struct vc4_validated_shader_info *validated_shader;
+ uint32_t o = shader_reloc_offsets[i];
uint32_t src_offset = *(uint32_t *)(pkt_u + o);
uint32_t *texture_handles_u;
void *uniform_data_u;
- uint32_t tex;
+ uint32_t tex, uni;
*(uint32_t *)(pkt_v + o) = bo[i]->paddr + src_offset;
- switch (relocs[i].type) {
- case RELOC_CODE:
- if (src_offset != 0) {
- DRM_ERROR("Shaders must be at offset 0 of "
- "the BO.\n");
- goto fail;
- }
-
- kfree(validated_shader);
- validated_shader = vc4_validate_shader(bo[i]);
- if (!validated_shader)
- goto fail;
+ if (src_offset != 0) {
+ DRM_ERROR("Shaders must be at offset 0 of "
+ "the BO.\n");
+ return -EINVAL;
+ }
- if (validated_shader->uniforms_src_size >
- exec->uniforms_size) {
- DRM_ERROR("Uniforms src buffer overflow\n");
- goto fail;
- }
+ validated_shader = to_vc4_bo(&bo[i]->base)->validated_shader;
+ if (!validated_shader)
+ return -EINVAL;
- texture_handles_u = exec->uniforms_u;
- uniform_data_u = (texture_handles_u +
- validated_shader->num_texture_samples);
-
- memcpy(exec->uniforms_v, uniform_data_u,
- validated_shader->uniforms_size);
-
- for (tex = 0;
- tex < validated_shader->num_texture_samples;
- tex++) {
- if (!reloc_tex(exec,
- uniform_data_u,
- &validated_shader->texture_samples[tex],
- texture_handles_u[tex])) {
- goto fail;
- }
- }
+ if (validated_shader->uniforms_src_size >
+ exec->uniforms_size) {
+ DRM_ERROR("Uniforms src buffer overflow\n");
+ return -EINVAL;
+ }
- *(uint32_t *)(pkt_v + o + 4) = exec->uniforms_p;
+ texture_handles_u = exec->uniforms_u;
+ uniform_data_u = (texture_handles_u +
+ validated_shader->num_texture_samples);
- exec->uniforms_u += validated_shader->uniforms_src_size;
- exec->uniforms_v += validated_shader->uniforms_size;
- exec->uniforms_p += validated_shader->uniforms_size;
+ memcpy(exec->uniforms_v, uniform_data_u,
+ validated_shader->uniforms_size);
- break;
+ for (tex = 0;
+ tex < validated_shader->num_texture_samples;
+ tex++) {
+ if (!reloc_tex(exec,
+ uniform_data_u,
+ &validated_shader->texture_samples[tex],
+ texture_handles_u[tex])) {
+ return -EINVAL;
+ }
+ }
- case RELOC_VBO:
- break;
+ /* Fill in the uniform slots that need this shader's
+ * start-of-uniforms address (used for resetting the uniform
+ * stream in the presence of control flow).
+ */
+ for (uni = 0;
+ uni < validated_shader->num_uniform_addr_offsets;
+ uni++) {
+ uint32_t o = validated_shader->uniform_addr_offsets[uni];
+ ((uint32_t *)exec->uniforms_v)[o] = exec->uniforms_p;
}
+
+ *(uint32_t *)(pkt_v + o + 4) = exec->uniforms_p;
+
+ exec->uniforms_u += validated_shader->uniforms_src_size;
+ exec->uniforms_v += validated_shader->uniforms_size;
+ exec->uniforms_p += validated_shader->uniforms_size;
}
for (i = 0; i < nr_attributes; i++) {
- struct drm_gem_cma_object *vbo = bo[nr_fixed_relocs + i];
+ struct drm_gem_cma_object *vbo =
+ bo[ARRAY_SIZE(shader_reloc_offsets) + i];
uint32_t o = 36 + i * 8;
uint32_t offset = *(uint32_t *)(pkt_u + o + 0);
uint32_t attr_size = *(uint8_t *)(pkt_u + o + 4) + 1;
if (vbo->base.size < offset ||
vbo->base.size - offset < attr_size) {
- DRM_ERROR("BO offset overflow (%d + %d > %d)\n",
+ DRM_ERROR("BO offset overflow (%d + %d > %zd)\n",
offset, attr_size, vbo->base.size);
return -EINVAL;
}
max_index = ((vbo->base.size - offset - attr_size) /
stride);
if (state->max_index > max_index) {
- DRM_ERROR("primitives use index %d out of supplied %d\n",
+ DRM_ERROR("primitives use index %d out of "
+ "supplied %d\n",
state->max_index, max_index);
return -EINVAL;
}
*(uint32_t *)(pkt_v + o) = vbo->paddr + offset;
}
- kfree(validated_shader);
-
return 0;
-
-fail:
- kfree(validated_shader);
- return -EINVAL;
}
int
int ret = 0;
for (i = 0; i < exec->shader_state_count; i++) {
- ret = validate_shader_rec(dev, exec, &exec->shader_state[i]);
+ ret = validate_gl_shader_rec(dev, exec, &exec->shader_state[i]);
if (ret)
return ret;
}