| 1 |
/* |
| 2 |
* Copyright © 2014-2015 Broadcom |
| 3 |
* |
| 4 |
* Permission is hereby granted, free of charge, to any person obtaining a |
| 5 |
* copy of this software and associated documentation files (the "Software"), |
| 6 |
* to deal in the Software without restriction, including without limitation |
| 7 |
* the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 8 |
* and/or sell copies of the Software, and to permit persons to whom the |
| 9 |
* Software is furnished to do so, subject to the following conditions: |
| 10 |
* |
| 11 |
* The above copyright notice and this permission notice (including the next |
| 12 |
* paragraph) shall be included in all copies or substantial portions of the |
| 13 |
* Software. |
| 14 |
* |
| 15 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 16 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 17 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 18 |
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 19 |
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
| 20 |
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
| 21 |
* IN THE SOFTWARE. |
| 22 |
*/ |
| 23 |
|
| 24 |
/** |
| 25 |
* DOC: Render command list generation |
| 26 |
* |
| 27 |
* In the VC4 driver, render command list generation is performed by the |
| 28 |
* kernel instead of userspace. We do this because validating a |
| 29 |
* user-submitted command list is hard to get right and has high CPU overhead, |
| 30 |
* while the number of valid configurations for render command lists is |
| 31 |
* actually fairly low. |
| 32 |
*/ |
| 33 |
|
| 34 |
#include "vc4_drv.h" |
| 35 |
#include "vc4_packet.h" |
| 36 |
|
| 37 |
struct vc4_rcl_setup { |
| 38 |
struct drm_gem_cma_object *color_read; |
| 39 |
struct drm_gem_cma_object *color_write; |
| 40 |
struct drm_gem_cma_object *zs_read; |
| 41 |
struct drm_gem_cma_object *zs_write; |
| 42 |
struct drm_gem_cma_object *msaa_color_write; |
| 43 |
struct drm_gem_cma_object *msaa_zs_write; |
| 44 |
|
| 45 |
struct drm_gem_cma_object *rcl; |
| 46 |
u32 next_offset; |
| 47 |
}; |
| 48 |
|
| 49 |
static inline void rcl_u8(struct vc4_rcl_setup *setup, u8 val) |
| 50 |
{ |
| 51 |
*(u8 *)(setup->rcl->vaddr + setup->next_offset) = val; |
| 52 |
setup->next_offset += 1; |
| 53 |
} |
| 54 |
|
| 55 |
static inline void rcl_u16(struct vc4_rcl_setup *setup, u16 val) |
| 56 |
{ |
| 57 |
*(u16 *)(setup->rcl->vaddr + setup->next_offset) = val; |
| 58 |
setup->next_offset += 2; |
| 59 |
} |
| 60 |
|
| 61 |
static inline void rcl_u32(struct vc4_rcl_setup *setup, u32 val) |
| 62 |
{ |
| 63 |
*(u32 *)(setup->rcl->vaddr + setup->next_offset) = val; |
| 64 |
setup->next_offset += 4; |
| 65 |
} |
| 66 |
|
| 67 |
/* |
| 68 |
* Emits a no-op STORE_TILE_BUFFER_GENERAL. |
| 69 |
* |
| 70 |
* If we emit a PACKET_TILE_COORDINATES, it must be followed by a store of |
| 71 |
* some sort before another load is triggered. |
| 72 |
*/ |
| 73 |
static void vc4_store_before_load(struct vc4_rcl_setup *setup) |
| 74 |
{ |
| 75 |
rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL); |
| 76 |
rcl_u16(setup, |
| 77 |
VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_NONE, |
| 78 |
VC4_LOADSTORE_TILE_BUFFER_BUFFER) | |
| 79 |
VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR | |
| 80 |
VC4_STORE_TILE_BUFFER_DISABLE_ZS_CLEAR | |
| 81 |
VC4_STORE_TILE_BUFFER_DISABLE_VG_MASK_CLEAR); |
| 82 |
rcl_u32(setup, 0); /* no address, since we're in None mode */ |
| 83 |
} |
| 84 |
|
| 85 |
/* |
| 86 |
* Calculates the physical address of the start of a tile in a RCL surface. |
| 87 |
* |
| 88 |
* Unlike the other load/store packets, |
| 89 |
* VC4_PACKET_LOAD/STORE_FULL_RES_TILE_BUFFER don't look at the tile |
| 90 |
* coordinates packet, and instead just store to the address given. |
| 91 |
*/ |
| 92 |
static uint32_t vc4_full_res_offset(struct vc4_exec_info *exec, |
| 93 |
struct drm_gem_cma_object *bo, |
| 94 |
struct drm_vc4_submit_rcl_surface *surf, |
| 95 |
uint8_t x, uint8_t y) |
| 96 |
{ |
| 97 |
return bo->paddr + surf->offset + VC4_TILE_BUFFER_SIZE * |
| 98 |
(DIV_ROUND_UP(exec->args->width, exec->tile_width) * y + x); |
| 99 |
} |
| 100 |
|
| 101 |
/* |
| 102 |
* Emits a PACKET_TILE_COORDINATES if one isn't already pending. |
| 103 |
* |
| 104 |
* The tile coordinates packet triggers a pending load if there is one, are |
| 105 |
* used for clipping during rendering, and determine where loads/stores happen |
| 106 |
* relative to their base address. |
| 107 |
*/ |
| 108 |
static void vc4_tile_coordinates(struct vc4_rcl_setup *setup, |
| 109 |
uint32_t x, uint32_t y) |
| 110 |
{ |
| 111 |
rcl_u8(setup, VC4_PACKET_TILE_COORDINATES); |
| 112 |
rcl_u8(setup, x); |
| 113 |
rcl_u8(setup, y); |
| 114 |
} |
| 115 |
|
| 116 |
static void emit_tile(struct vc4_exec_info *exec, |
| 117 |
struct vc4_rcl_setup *setup, |
| 118 |
uint8_t x, uint8_t y, bool first, bool last) |
| 119 |
{ |
| 120 |
struct drm_vc4_submit_cl *args = exec->args; |
| 121 |
bool has_bin = args->bin_cl_size != 0; |
| 122 |
|
| 123 |
/* Note that the load doesn't actually occur until the |
| 124 |
* tile coords packet is processed, and only one load |
| 125 |
* may be outstanding at a time. |
| 126 |
*/ |
| 127 |
if (setup->color_read) { |
| 128 |
if (args->color_read.flags & |
| 129 |
VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) { |
| 130 |
rcl_u8(setup, VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER); |
| 131 |
rcl_u32(setup, |
| 132 |
vc4_full_res_offset(exec, setup->color_read, |
| 133 |
&args->color_read, x, y) | |
| 134 |
VC4_LOADSTORE_FULL_RES_DISABLE_ZS); |
| 135 |
} else { |
| 136 |
rcl_u8(setup, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL); |
| 137 |
rcl_u16(setup, args->color_read.bits); |
| 138 |
rcl_u32(setup, setup->color_read->paddr + |
| 139 |
args->color_read.offset); |
| 140 |
} |
| 141 |
} |
| 142 |
|
| 143 |
if (setup->zs_read) { |
| 144 |
if (setup->color_read) { |
| 145 |
/* Exec previous load. */ |
| 146 |
vc4_tile_coordinates(setup, x, y); |
| 147 |
vc4_store_before_load(setup); |
| 148 |
} |
| 149 |
|
| 150 |
if (args->zs_read.flags & |
| 151 |
VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) { |
| 152 |
rcl_u8(setup, VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER); |
| 153 |
rcl_u32(setup, |
| 154 |
vc4_full_res_offset(exec, setup->zs_read, |
| 155 |
&args->zs_read, x, y) | |
| 156 |
VC4_LOADSTORE_FULL_RES_DISABLE_COLOR); |
| 157 |
} else { |
| 158 |
rcl_u8(setup, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL); |
| 159 |
rcl_u16(setup, args->zs_read.bits); |
| 160 |
rcl_u32(setup, setup->zs_read->paddr + |
| 161 |
args->zs_read.offset); |
| 162 |
} |
| 163 |
} |
| 164 |
|
| 165 |
/* Clipping depends on tile coordinates having been |
| 166 |
* emitted, so we always need one here. |
| 167 |
*/ |
| 168 |
vc4_tile_coordinates(setup, x, y); |
| 169 |
|
| 170 |
/* Wait for the binner before jumping to the first |
| 171 |
* tile's lists. |
| 172 |
*/ |
| 173 |
if (first && has_bin) |
| 174 |
rcl_u8(setup, VC4_PACKET_WAIT_ON_SEMAPHORE); |
| 175 |
|
| 176 |
if (has_bin) { |
| 177 |
rcl_u8(setup, VC4_PACKET_BRANCH_TO_SUB_LIST); |
| 178 |
rcl_u32(setup, (exec->tile_bo->paddr + |
| 179 |
exec->tile_alloc_offset + |
| 180 |
(y * exec->bin_tiles_x + x) * 32)); |
| 181 |
} |
| 182 |
|
| 183 |
if (setup->msaa_color_write) { |
| 184 |
bool last_tile_write = (!setup->msaa_zs_write && |
| 185 |
!setup->zs_write && |
| 186 |
!setup->color_write); |
| 187 |
uint32_t bits = VC4_LOADSTORE_FULL_RES_DISABLE_ZS; |
| 188 |
|
| 189 |
if (!last_tile_write) |
| 190 |
bits |= VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL; |
| 191 |
else if (last) |
| 192 |
bits |= VC4_LOADSTORE_FULL_RES_EOF; |
| 193 |
rcl_u8(setup, VC4_PACKET_STORE_FULL_RES_TILE_BUFFER); |
| 194 |
rcl_u32(setup, |
| 195 |
vc4_full_res_offset(exec, setup->msaa_color_write, |
| 196 |
&args->msaa_color_write, x, y) | |
| 197 |
bits); |
| 198 |
} |
| 199 |
|
| 200 |
if (setup->msaa_zs_write) { |
| 201 |
bool last_tile_write = (!setup->zs_write && |
| 202 |
!setup->color_write); |
| 203 |
uint32_t bits = VC4_LOADSTORE_FULL_RES_DISABLE_COLOR; |
| 204 |
|
| 205 |
if (setup->msaa_color_write) |
| 206 |
vc4_tile_coordinates(setup, x, y); |
| 207 |
if (!last_tile_write) |
| 208 |
bits |= VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL; |
| 209 |
else if (last) |
| 210 |
bits |= VC4_LOADSTORE_FULL_RES_EOF; |
| 211 |
rcl_u8(setup, VC4_PACKET_STORE_FULL_RES_TILE_BUFFER); |
| 212 |
rcl_u32(setup, |
| 213 |
vc4_full_res_offset(exec, setup->msaa_zs_write, |
| 214 |
&args->msaa_zs_write, x, y) | |
| 215 |
bits); |
| 216 |
} |
| 217 |
|
| 218 |
if (setup->zs_write) { |
| 219 |
bool last_tile_write = !setup->color_write; |
| 220 |
|
| 221 |
if (setup->msaa_color_write || setup->msaa_zs_write) |
| 222 |
vc4_tile_coordinates(setup, x, y); |
| 223 |
|
| 224 |
rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL); |
| 225 |
rcl_u16(setup, args->zs_write.bits | |
| 226 |
(last_tile_write ? |
| 227 |
0 : VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR)); |
| 228 |
rcl_u32(setup, |
| 229 |
(setup->zs_write->paddr + args->zs_write.offset) | |
| 230 |
((last && last_tile_write) ? |
| 231 |
VC4_LOADSTORE_TILE_BUFFER_EOF : 0)); |
| 232 |
} |
| 233 |
|
| 234 |
if (setup->color_write) { |
| 235 |
if (setup->msaa_color_write || setup->msaa_zs_write || |
| 236 |
setup->zs_write) { |
| 237 |
vc4_tile_coordinates(setup, x, y); |
| 238 |
} |
| 239 |
|
| 240 |
if (last) |
| 241 |
rcl_u8(setup, VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF); |
| 242 |
else |
| 243 |
rcl_u8(setup, VC4_PACKET_STORE_MS_TILE_BUFFER); |
| 244 |
} |
| 245 |
} |
| 246 |
|
| 247 |
static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec, |
| 248 |
struct vc4_rcl_setup *setup) |
| 249 |
{ |
| 250 |
struct drm_vc4_submit_cl *args = exec->args; |
| 251 |
bool has_bin = args->bin_cl_size != 0; |
| 252 |
uint8_t min_x_tile = args->min_x_tile; |
| 253 |
uint8_t min_y_tile = args->min_y_tile; |
| 254 |
uint8_t max_x_tile = args->max_x_tile; |
| 255 |
uint8_t max_y_tile = args->max_y_tile; |
| 256 |
uint8_t xtiles = max_x_tile - min_x_tile + 1; |
| 257 |
uint8_t ytiles = max_y_tile - min_y_tile + 1; |
| 258 |
uint8_t xi, yi; |
| 259 |
uint32_t size, loop_body_size; |
| 260 |
bool positive_x = true; |
| 261 |
bool positive_y = true; |
| 262 |
|
| 263 |
if (args->flags & VC4_SUBMIT_CL_FIXED_RCL_ORDER) { |
| 264 |
if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X)) |
| 265 |
positive_x = false; |
| 266 |
if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y)) |
| 267 |
positive_y = false; |
| 268 |
} |
| 269 |
|
| 270 |
size = VC4_PACKET_TILE_RENDERING_MODE_CONFIG_SIZE; |
| 271 |
loop_body_size = VC4_PACKET_TILE_COORDINATES_SIZE; |
| 272 |
|
| 273 |
if (args->flags & VC4_SUBMIT_CL_USE_CLEAR_COLOR) { |
| 274 |
size += VC4_PACKET_CLEAR_COLORS_SIZE + |
| 275 |
VC4_PACKET_TILE_COORDINATES_SIZE + |
| 276 |
VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE; |
| 277 |
} |
| 278 |
|
| 279 |
if (setup->color_read) { |
| 280 |
if (args->color_read.flags & |
| 281 |
VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) { |
| 282 |
loop_body_size += VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE; |
| 283 |
} else { |
| 284 |
loop_body_size += VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE; |
| 285 |
} |
| 286 |
} |
| 287 |
if (setup->zs_read) { |
| 288 |
if (setup->color_read) { |
| 289 |
loop_body_size += VC4_PACKET_TILE_COORDINATES_SIZE; |
| 290 |
loop_body_size += VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE; |
| 291 |
} |
| 292 |
|
| 293 |
if (args->zs_read.flags & |
| 294 |
VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) { |
| 295 |
loop_body_size += VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE; |
| 296 |
} else { |
| 297 |
loop_body_size += VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE; |
| 298 |
} |
| 299 |
} |
| 300 |
|
| 301 |
if (has_bin) { |
| 302 |
size += VC4_PACKET_WAIT_ON_SEMAPHORE_SIZE; |
| 303 |
loop_body_size += VC4_PACKET_BRANCH_TO_SUB_LIST_SIZE; |
| 304 |
} |
| 305 |
|
| 306 |
if (setup->msaa_color_write) |
| 307 |
loop_body_size += VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE; |
| 308 |
if (setup->msaa_zs_write) |
| 309 |
loop_body_size += VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE; |
| 310 |
|
| 311 |
if (setup->zs_write) |
| 312 |
loop_body_size += VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE; |
| 313 |
if (setup->color_write) |
| 314 |
loop_body_size += VC4_PACKET_STORE_MS_TILE_BUFFER_SIZE; |
| 315 |
|
| 316 |
/* We need a VC4_PACKET_TILE_COORDINATES in between each store. */ |
| 317 |
loop_body_size += VC4_PACKET_TILE_COORDINATES_SIZE * |
| 318 |
((setup->msaa_color_write != NULL) + |
| 319 |
(setup->msaa_zs_write != NULL) + |
| 320 |
(setup->color_write != NULL) + |
| 321 |
(setup->zs_write != NULL) - 1); |
| 322 |
|
| 323 |
size += xtiles * ytiles * loop_body_size; |
| 324 |
|
| 325 |
setup->rcl = drm_gem_cma_create(dev, size); |
| 326 |
if (!setup->rcl) |
| 327 |
return -ENOMEM; |
| 328 |
list_addtail(&to_vc4_bo(&setup->rcl->base)->unref_head, |
| 329 |
&exec->unref_list); |
| 330 |
|
| 331 |
rcl_u8(setup, VC4_PACKET_TILE_RENDERING_MODE_CONFIG); |
| 332 |
rcl_u32(setup, |
| 333 |
(setup->color_write ? (setup->color_write->paddr + |
| 334 |
args->color_write.offset) : |
| 335 |
0)); |
| 336 |
rcl_u16(setup, args->width); |
| 337 |
rcl_u16(setup, args->height); |
| 338 |
rcl_u16(setup, args->color_write.bits); |
| 339 |
|
| 340 |
/* The tile buffer gets cleared when the previous tile is stored. If |
| 341 |
* the clear values changed between frames, then the tile buffer has |
| 342 |
* stale clear values in it, so we have to do a store in None mode (no |
| 343 |
* writes) so that we trigger the tile buffer clear. |
| 344 |
*/ |
| 345 |
if (args->flags & VC4_SUBMIT_CL_USE_CLEAR_COLOR) { |
| 346 |
rcl_u8(setup, VC4_PACKET_CLEAR_COLORS); |
| 347 |
rcl_u32(setup, args->clear_color[0]); |
| 348 |
rcl_u32(setup, args->clear_color[1]); |
| 349 |
rcl_u32(setup, args->clear_z); |
| 350 |
rcl_u8(setup, args->clear_s); |
| 351 |
|
| 352 |
vc4_tile_coordinates(setup, 0, 0); |
| 353 |
|
| 354 |
rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL); |
| 355 |
rcl_u16(setup, VC4_LOADSTORE_TILE_BUFFER_NONE); |
| 356 |
rcl_u32(setup, 0); /* no address, since we're in None mode */ |
| 357 |
} |
| 358 |
|
| 359 |
for (yi = 0; yi < ytiles; yi++) { |
| 360 |
int y = positive_y ? min_y_tile + yi : max_y_tile - yi; |
| 361 |
for (xi = 0; xi < xtiles; xi++) { |
| 362 |
int x = positive_x ? min_x_tile + xi : max_x_tile - xi; |
| 363 |
bool first = (xi == 0 && yi == 0); |
| 364 |
bool last = (xi == xtiles - 1 && yi == ytiles - 1); |
| 365 |
|
| 366 |
emit_tile(exec, setup, x, y, first, last); |
| 367 |
} |
| 368 |
} |
| 369 |
|
| 370 |
BUG_ON(setup->next_offset != size); |
| 371 |
exec->ct1ca = setup->rcl->paddr; |
| 372 |
exec->ct1ea = setup->rcl->paddr + setup->next_offset; |
| 373 |
|
| 374 |
return 0; |
| 375 |
} |
| 376 |
|
| 377 |
static int vc4_full_res_bounds_check(struct vc4_exec_info *exec, |
| 378 |
struct drm_gem_cma_object *obj, |
| 379 |
struct drm_vc4_submit_rcl_surface *surf) |
| 380 |
{ |
| 381 |
struct drm_vc4_submit_cl *args = exec->args; |
| 382 |
u32 render_tiles_stride = DIV_ROUND_UP(exec->args->width, |
| 383 |
exec->tile_width); |
| 384 |
|
| 385 |
if (surf->offset > obj->base.size) { |
| 386 |
DRM_ERROR("surface offset %d > BO size %zd\n", |
| 387 |
surf->offset, obj->base.size); |
| 388 |
return -EINVAL; |
| 389 |
} |
| 390 |
|
| 391 |
if ((obj->base.size - surf->offset) / VC4_TILE_BUFFER_SIZE < |
| 392 |
render_tiles_stride * args->max_y_tile + args->max_x_tile) { |
| 393 |
DRM_ERROR("MSAA tile %d, %d out of bounds " |
| 394 |
"(bo size %zd, offset %d).\n", |
| 395 |
args->max_x_tile, args->max_y_tile, |
| 396 |
obj->base.size, |
| 397 |
surf->offset); |
| 398 |
return -EINVAL; |
| 399 |
} |
| 400 |
|
| 401 |
return 0; |
| 402 |
} |
| 403 |
|
| 404 |
static int vc4_rcl_msaa_surface_setup(struct vc4_exec_info *exec, |
| 405 |
struct drm_gem_cma_object **obj, |
| 406 |
struct drm_vc4_submit_rcl_surface *surf) |
| 407 |
{ |
| 408 |
if (surf->flags != 0 || surf->bits != 0) { |
| 409 |
DRM_ERROR("MSAA surface had nonzero flags/bits\n"); |
| 410 |
return -EINVAL; |
| 411 |
} |
| 412 |
|
| 413 |
if (surf->hindex == ~0) |
| 414 |
return 0; |
| 415 |
|
| 416 |
*obj = vc4_use_bo(exec, surf->hindex); |
| 417 |
if (!*obj) |
| 418 |
return -EINVAL; |
| 419 |
|
| 420 |
if (surf->offset & 0xf) { |
| 421 |
DRM_ERROR("MSAA write must be 16b aligned.\n"); |
| 422 |
return -EINVAL; |
| 423 |
} |
| 424 |
|
| 425 |
return vc4_full_res_bounds_check(exec, *obj, surf); |
| 426 |
} |
| 427 |
|
| 428 |
static int vc4_rcl_surface_setup(struct vc4_exec_info *exec, |
| 429 |
struct drm_gem_cma_object **obj, |
| 430 |
struct drm_vc4_submit_rcl_surface *surf) |
| 431 |
{ |
| 432 |
uint8_t tiling = VC4_GET_FIELD(surf->bits, |
| 433 |
VC4_LOADSTORE_TILE_BUFFER_TILING); |
| 434 |
uint8_t buffer = VC4_GET_FIELD(surf->bits, |
| 435 |
VC4_LOADSTORE_TILE_BUFFER_BUFFER); |
| 436 |
uint8_t format = VC4_GET_FIELD(surf->bits, |
| 437 |
VC4_LOADSTORE_TILE_BUFFER_FORMAT); |
| 438 |
int cpp; |
| 439 |
int ret; |
| 440 |
|
| 441 |
if (surf->flags & ~VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) { |
| 442 |
DRM_ERROR("Extra flags set\n"); |
| 443 |
return -EINVAL; |
| 444 |
} |
| 445 |
|
| 446 |
if (surf->hindex == ~0) |
| 447 |
return 0; |
| 448 |
|
| 449 |
*obj = vc4_use_bo(exec, surf->hindex); |
| 450 |
if (!*obj) |
| 451 |
return -EINVAL; |
| 452 |
|
| 453 |
if (surf->flags & VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) { |
| 454 |
if (surf == &exec->args->zs_write) { |
| 455 |
DRM_ERROR("general zs write may not be a full-res.\n"); |
| 456 |
return -EINVAL; |
| 457 |
} |
| 458 |
|
| 459 |
if (surf->bits != 0) { |
| 460 |
DRM_ERROR("load/store general bits set with " |
| 461 |
"full res load/store.\n"); |
| 462 |
return -EINVAL; |
| 463 |
} |
| 464 |
|
| 465 |
ret = vc4_full_res_bounds_check(exec, *obj, surf); |
| 466 |
if (!ret) |
| 467 |
return ret; |
| 468 |
|
| 469 |
return 0; |
| 470 |
} |
| 471 |
|
| 472 |
if (surf->bits & ~(VC4_LOADSTORE_TILE_BUFFER_TILING_MASK | |
| 473 |
VC4_LOADSTORE_TILE_BUFFER_BUFFER_MASK | |
| 474 |
VC4_LOADSTORE_TILE_BUFFER_FORMAT_MASK)) { |
| 475 |
DRM_ERROR("Unknown bits in load/store: 0x%04x\n", |
| 476 |
surf->bits); |
| 477 |
return -EINVAL; |
| 478 |
} |
| 479 |
|
| 480 |
if (tiling > VC4_TILING_FORMAT_LT) { |
| 481 |
DRM_ERROR("Bad tiling format\n"); |
| 482 |
return -EINVAL; |
| 483 |
} |
| 484 |
|
| 485 |
if (buffer == VC4_LOADSTORE_TILE_BUFFER_ZS) { |
| 486 |
if (format != 0) { |
| 487 |
DRM_ERROR("No color format should be set for ZS\n"); |
| 488 |
return -EINVAL; |
| 489 |
} |
| 490 |
cpp = 4; |
| 491 |
} else if (buffer == VC4_LOADSTORE_TILE_BUFFER_COLOR) { |
| 492 |
switch (format) { |
| 493 |
case VC4_LOADSTORE_TILE_BUFFER_BGR565: |
| 494 |
case VC4_LOADSTORE_TILE_BUFFER_BGR565_DITHER: |
| 495 |
cpp = 2; |
| 496 |
break; |
| 497 |
case VC4_LOADSTORE_TILE_BUFFER_RGBA8888: |
| 498 |
cpp = 4; |
| 499 |
break; |
| 500 |
default: |
| 501 |
DRM_ERROR("Bad tile buffer format\n"); |
| 502 |
return -EINVAL; |
| 503 |
} |
| 504 |
} else { |
| 505 |
DRM_ERROR("Bad load/store buffer %d.\n", buffer); |
| 506 |
return -EINVAL; |
| 507 |
} |
| 508 |
|
| 509 |
if (surf->offset & 0xf) { |
| 510 |
DRM_ERROR("load/store buffer must be 16b aligned.\n"); |
| 511 |
return -EINVAL; |
| 512 |
} |
| 513 |
|
| 514 |
if (!vc4_check_tex_size(exec, *obj, surf->offset, tiling, |
| 515 |
exec->args->width, exec->args->height, cpp)) { |
| 516 |
return -EINVAL; |
| 517 |
} |
| 518 |
|
| 519 |
return 0; |
| 520 |
} |
| 521 |
|
| 522 |
static int |
| 523 |
vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec, |
| 524 |
struct vc4_rcl_setup *setup, |
| 525 |
struct drm_gem_cma_object **obj, |
| 526 |
struct drm_vc4_submit_rcl_surface *surf) |
| 527 |
{ |
| 528 |
uint8_t tiling = VC4_GET_FIELD(surf->bits, |
| 529 |
VC4_RENDER_CONFIG_MEMORY_FORMAT); |
| 530 |
uint8_t format = VC4_GET_FIELD(surf->bits, |
| 531 |
VC4_RENDER_CONFIG_FORMAT); |
| 532 |
int cpp; |
| 533 |
|
| 534 |
if (surf->flags != 0) { |
| 535 |
DRM_ERROR("No flags supported on render config.\n"); |
| 536 |
return -EINVAL; |
| 537 |
} |
| 538 |
|
| 539 |
if (surf->bits & ~(VC4_RENDER_CONFIG_MEMORY_FORMAT_MASK | |
| 540 |
VC4_RENDER_CONFIG_FORMAT_MASK | |
| 541 |
VC4_RENDER_CONFIG_MS_MODE_4X | |
| 542 |
VC4_RENDER_CONFIG_DECIMATE_MODE_4X)) { |
| 543 |
DRM_ERROR("Unknown bits in render config: 0x%04x\n", |
| 544 |
surf->bits); |
| 545 |
return -EINVAL; |
| 546 |
} |
| 547 |
|
| 548 |
if (surf->hindex == ~0) |
| 549 |
return 0; |
| 550 |
|
| 551 |
*obj = vc4_use_bo(exec, surf->hindex); |
| 552 |
if (!*obj) |
| 553 |
return -EINVAL; |
| 554 |
|
| 555 |
if (tiling > VC4_TILING_FORMAT_LT) { |
| 556 |
DRM_ERROR("Bad tiling format\n"); |
| 557 |
return -EINVAL; |
| 558 |
} |
| 559 |
|
| 560 |
switch (format) { |
| 561 |
case VC4_RENDER_CONFIG_FORMAT_BGR565_DITHERED: |
| 562 |
case VC4_RENDER_CONFIG_FORMAT_BGR565: |
| 563 |
cpp = 2; |
| 564 |
break; |
| 565 |
case VC4_RENDER_CONFIG_FORMAT_RGBA8888: |
| 566 |
cpp = 4; |
| 567 |
break; |
| 568 |
default: |
| 569 |
DRM_ERROR("Bad tile buffer format\n"); |
| 570 |
return -EINVAL; |
| 571 |
} |
| 572 |
|
| 573 |
if (!vc4_check_tex_size(exec, *obj, surf->offset, tiling, |
| 574 |
exec->args->width, exec->args->height, cpp)) { |
| 575 |
return -EINVAL; |
| 576 |
} |
| 577 |
|
| 578 |
return 0; |
| 579 |
} |
| 580 |
|
| 581 |
int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec) |
| 582 |
{ |
| 583 |
struct vc4_rcl_setup setup = {0}; |
| 584 |
struct drm_vc4_submit_cl *args = exec->args; |
| 585 |
bool has_bin = args->bin_cl_size != 0; |
| 586 |
int ret; |
| 587 |
|
| 588 |
if (args->min_x_tile > args->max_x_tile || |
| 589 |
args->min_y_tile > args->max_y_tile) { |
| 590 |
DRM_ERROR("Bad render tile set (%d,%d)-(%d,%d)\n", |
| 591 |
args->min_x_tile, args->min_y_tile, |
| 592 |
args->max_x_tile, args->max_y_tile); |
| 593 |
return -EINVAL; |
| 594 |
} |
| 595 |
|
| 596 |
if (has_bin && |
| 597 |
(args->max_x_tile > exec->bin_tiles_x || |
| 598 |
args->max_y_tile > exec->bin_tiles_y)) { |
| 599 |
DRM_ERROR("Render tiles (%d,%d) outside of bin config " |
| 600 |
"(%d,%d)\n", |
| 601 |
args->max_x_tile, args->max_y_tile, |
| 602 |
exec->bin_tiles_x, exec->bin_tiles_y); |
| 603 |
return -EINVAL; |
| 604 |
} |
| 605 |
|
| 606 |
ret = vc4_rcl_render_config_surface_setup(exec, &setup, |
| 607 |
&setup.color_write, |
| 608 |
&args->color_write); |
| 609 |
if (ret) |
| 610 |
return ret; |
| 611 |
|
| 612 |
ret = vc4_rcl_surface_setup(exec, &setup.color_read, &args->color_read); |
| 613 |
if (ret) |
| 614 |
return ret; |
| 615 |
|
| 616 |
ret = vc4_rcl_surface_setup(exec, &setup.zs_read, &args->zs_read); |
| 617 |
if (ret) |
| 618 |
return ret; |
| 619 |
|
| 620 |
ret = vc4_rcl_surface_setup(exec, &setup.zs_write, &args->zs_write); |
| 621 |
if (ret) |
| 622 |
return ret; |
| 623 |
|
| 624 |
ret = vc4_rcl_msaa_surface_setup(exec, &setup.msaa_color_write, |
| 625 |
&args->msaa_color_write); |
| 626 |
if (ret) |
| 627 |
return ret; |
| 628 |
|
| 629 |
ret = vc4_rcl_msaa_surface_setup(exec, &setup.msaa_zs_write, |
| 630 |
&args->msaa_zs_write); |
| 631 |
if (ret) |
| 632 |
return ret; |
| 633 |
|
| 634 |
/* We shouldn't even have the job submitted to us if there's no |
| 635 |
* surface to write out. |
| 636 |
*/ |
| 637 |
if (!setup.color_write && !setup.zs_write && |
| 638 |
!setup.msaa_color_write && !setup.msaa_zs_write) { |
| 639 |
DRM_ERROR("RCL requires color or Z/S write\n"); |
| 640 |
return -EINVAL; |
| 641 |
} |
| 642 |
|
| 643 |
return vc4_create_rcl_bo(dev, exec, &setup); |
| 644 |
} |