• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "uapi/drm/vc4_drm.h"
35 #include "vc4_drv.h"
36 #include "vc4_packet.h"
37 
38 struct vc4_rcl_setup {
39 	struct drm_gem_cma_object *color_read;
40 	struct drm_gem_cma_object *color_write;
41 	struct drm_gem_cma_object *zs_read;
42 	struct drm_gem_cma_object *zs_write;
43 	struct drm_gem_cma_object *msaa_color_write;
44 	struct drm_gem_cma_object *msaa_zs_write;
45 
46 	struct drm_gem_cma_object *rcl;
47 	u32 next_offset;
48 
49 	u32 next_write_bo_index;
50 };
51 
rcl_u8(struct vc4_rcl_setup * setup,u8 val)52 static inline void rcl_u8(struct vc4_rcl_setup *setup, u8 val)
53 {
54 	*(u8 *)(setup->rcl->vaddr + setup->next_offset) = val;
55 	setup->next_offset += 1;
56 }
57 
rcl_u16(struct vc4_rcl_setup * setup,u16 val)58 static inline void rcl_u16(struct vc4_rcl_setup *setup, u16 val)
59 {
60 	*(u16 *)(setup->rcl->vaddr + setup->next_offset) = val;
61 	setup->next_offset += 2;
62 }
63 
rcl_u32(struct vc4_rcl_setup * setup,u32 val)64 static inline void rcl_u32(struct vc4_rcl_setup *setup, u32 val)
65 {
66 	*(u32 *)(setup->rcl->vaddr + setup->next_offset) = val;
67 	setup->next_offset += 4;
68 }
69 
70 /*
71  * Emits a no-op STORE_TILE_BUFFER_GENERAL.
72  *
73  * If we emit a PACKET_TILE_COORDINATES, it must be followed by a store of
74  * some sort before another load is triggered.
75  */
vc4_store_before_load(struct vc4_rcl_setup * setup)76 static void vc4_store_before_load(struct vc4_rcl_setup *setup)
77 {
78 	rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
79 	rcl_u16(setup,
80 		VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_NONE,
81 			      VC4_LOADSTORE_TILE_BUFFER_BUFFER) |
82 		VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR |
83 		VC4_STORE_TILE_BUFFER_DISABLE_ZS_CLEAR |
84 		VC4_STORE_TILE_BUFFER_DISABLE_VG_MASK_CLEAR);
85 	rcl_u32(setup, 0); /* no address, since we're in None mode */
86 }
87 
88 /*
89  * Calculates the physical address of the start of a tile in a RCL surface.
90  *
91  * Unlike the other load/store packets,
92  * VC4_PACKET_LOAD/STORE_FULL_RES_TILE_BUFFER don't look at the tile
93  * coordinates packet, and instead just store to the address given.
94  */
vc4_full_res_offset(struct vc4_exec_info * exec,struct drm_gem_cma_object * bo,struct drm_vc4_submit_rcl_surface * surf,uint8_t x,uint8_t y)95 static uint32_t vc4_full_res_offset(struct vc4_exec_info *exec,
96 				    struct drm_gem_cma_object *bo,
97 				    struct drm_vc4_submit_rcl_surface *surf,
98 				    uint8_t x, uint8_t y)
99 {
100 	return bo->paddr + surf->offset + VC4_TILE_BUFFER_SIZE *
101 		(DIV_ROUND_UP(exec->args->width, 32) * y + x);
102 }
103 
104 /*
105  * Emits a PACKET_TILE_COORDINATES if one isn't already pending.
106  *
107  * The tile coordinates packet triggers a pending load if there is one, are
108  * used for clipping during rendering, and determine where loads/stores happen
109  * relative to their base address.
110  */
vc4_tile_coordinates(struct vc4_rcl_setup * setup,uint32_t x,uint32_t y)111 static void vc4_tile_coordinates(struct vc4_rcl_setup *setup,
112 				 uint32_t x, uint32_t y)
113 {
114 	rcl_u8(setup, VC4_PACKET_TILE_COORDINATES);
115 	rcl_u8(setup, x);
116 	rcl_u8(setup, y);
117 }
118 
emit_tile(struct vc4_exec_info * exec,struct vc4_rcl_setup * setup,uint8_t x,uint8_t y,bool first,bool last)119 static void emit_tile(struct vc4_exec_info *exec,
120 		      struct vc4_rcl_setup *setup,
121 		      uint8_t x, uint8_t y, bool first, bool last)
122 {
123 	struct drm_vc4_submit_cl *args = exec->args;
124 	bool has_bin = args->bin_cl_size != 0;
125 
126 	/* Note that the load doesn't actually occur until the
127 	 * tile coords packet is processed, and only one load
128 	 * may be outstanding at a time.
129 	 */
130 	if (setup->color_read) {
131 		if (args->color_read.flags &
132 		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
133 			rcl_u8(setup, VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER);
134 			rcl_u32(setup,
135 				vc4_full_res_offset(exec, setup->color_read,
136 						    &args->color_read, x, y) |
137 				VC4_LOADSTORE_FULL_RES_DISABLE_ZS);
138 		} else {
139 			rcl_u8(setup, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL);
140 			rcl_u16(setup, args->color_read.bits);
141 			rcl_u32(setup, setup->color_read->paddr +
142 				args->color_read.offset);
143 		}
144 	}
145 
146 	if (setup->zs_read) {
147 		if (args->zs_read.flags &
148 		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
149 			rcl_u8(setup, VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER);
150 			rcl_u32(setup,
151 				vc4_full_res_offset(exec, setup->zs_read,
152 						    &args->zs_read, x, y) |
153 				VC4_LOADSTORE_FULL_RES_DISABLE_COLOR);
154 		} else {
155 			if (setup->color_read) {
156 				/* Exec previous load. */
157 				vc4_tile_coordinates(setup, x, y);
158 				vc4_store_before_load(setup);
159 			}
160 
161 			rcl_u8(setup, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL);
162 			rcl_u16(setup, args->zs_read.bits);
163 			rcl_u32(setup, setup->zs_read->paddr +
164 				args->zs_read.offset);
165 		}
166 	}
167 
168 	/* Clipping depends on tile coordinates having been
169 	 * emitted, so we always need one here.
170 	 */
171 	vc4_tile_coordinates(setup, x, y);
172 
173 	/* Wait for the binner before jumping to the first
174 	 * tile's lists.
175 	 */
176 	if (first && has_bin)
177 		rcl_u8(setup, VC4_PACKET_WAIT_ON_SEMAPHORE);
178 
179 	if (has_bin) {
180 		rcl_u8(setup, VC4_PACKET_BRANCH_TO_SUB_LIST);
181 		rcl_u32(setup, (exec->tile_bo->paddr +
182 				exec->tile_alloc_offset +
183 				(y * exec->bin_tiles_x + x) * 32));
184 	}
185 
186 	if (setup->msaa_color_write) {
187 		bool last_tile_write = (!setup->msaa_zs_write &&
188 					!setup->zs_write &&
189 					!setup->color_write);
190 		uint32_t bits = VC4_LOADSTORE_FULL_RES_DISABLE_ZS;
191 
192 		if (!last_tile_write)
193 			bits |= VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL;
194 		else if (last)
195 			bits |= VC4_LOADSTORE_FULL_RES_EOF;
196 		rcl_u8(setup, VC4_PACKET_STORE_FULL_RES_TILE_BUFFER);
197 		rcl_u32(setup,
198 			vc4_full_res_offset(exec, setup->msaa_color_write,
199 					    &args->msaa_color_write, x, y) |
200 			bits);
201 	}
202 
203 	if (setup->msaa_zs_write) {
204 		bool last_tile_write = (!setup->zs_write &&
205 					!setup->color_write);
206 		uint32_t bits = VC4_LOADSTORE_FULL_RES_DISABLE_COLOR;
207 
208 		if (setup->msaa_color_write)
209 			vc4_tile_coordinates(setup, x, y);
210 		if (!last_tile_write)
211 			bits |= VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL;
212 		else if (last)
213 			bits |= VC4_LOADSTORE_FULL_RES_EOF;
214 		rcl_u8(setup, VC4_PACKET_STORE_FULL_RES_TILE_BUFFER);
215 		rcl_u32(setup,
216 			vc4_full_res_offset(exec, setup->msaa_zs_write,
217 					    &args->msaa_zs_write, x, y) |
218 			bits);
219 	}
220 
221 	if (setup->zs_write) {
222 		bool last_tile_write = !setup->color_write;
223 
224 		if (setup->msaa_color_write || setup->msaa_zs_write)
225 			vc4_tile_coordinates(setup, x, y);
226 
227 		rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
228 		rcl_u16(setup, args->zs_write.bits |
229 			(last_tile_write ?
230 			 0 : VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR));
231 		rcl_u32(setup,
232 			(setup->zs_write->paddr + args->zs_write.offset) |
233 			((last && last_tile_write) ?
234 			 VC4_LOADSTORE_TILE_BUFFER_EOF : 0));
235 	}
236 
237 	if (setup->color_write) {
238 		if (setup->msaa_color_write || setup->msaa_zs_write ||
239 		    setup->zs_write) {
240 			vc4_tile_coordinates(setup, x, y);
241 		}
242 
243 		if (last)
244 			rcl_u8(setup, VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF);
245 		else
246 			rcl_u8(setup, VC4_PACKET_STORE_MS_TILE_BUFFER);
247 	}
248 }
249 
vc4_create_rcl_bo(struct drm_device * dev,struct vc4_exec_info * exec,struct vc4_rcl_setup * setup)250 static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
251 			     struct vc4_rcl_setup *setup)
252 {
253 	struct drm_vc4_submit_cl *args = exec->args;
254 	bool has_bin = args->bin_cl_size != 0;
255 	uint8_t min_x_tile = args->min_x_tile;
256 	uint8_t min_y_tile = args->min_y_tile;
257 	uint8_t max_x_tile = args->max_x_tile;
258 	uint8_t max_y_tile = args->max_y_tile;
259 	uint8_t xtiles = max_x_tile - min_x_tile + 1;
260 	uint8_t ytiles = max_y_tile - min_y_tile + 1;
261 	uint8_t x, y;
262 	uint32_t size, loop_body_size;
263 
264 	size = VC4_PACKET_TILE_RENDERING_MODE_CONFIG_SIZE;
265 	loop_body_size = VC4_PACKET_TILE_COORDINATES_SIZE;
266 
267 	if (args->flags & VC4_SUBMIT_CL_USE_CLEAR_COLOR) {
268 		size += VC4_PACKET_CLEAR_COLORS_SIZE +
269 			VC4_PACKET_TILE_COORDINATES_SIZE +
270 			VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE;
271 	}
272 
273 	if (setup->color_read) {
274 		if (args->color_read.flags &
275 		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
276 			loop_body_size += VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE;
277 		} else {
278 			loop_body_size += VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE;
279 		}
280 	}
281 	if (setup->zs_read) {
282 		if (args->zs_read.flags &
283 		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
284 			loop_body_size += VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE;
285 		} else {
286 			if (setup->color_read &&
287 			    !(args->color_read.flags &
288 			      VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES)) {
289 				loop_body_size += VC4_PACKET_TILE_COORDINATES_SIZE;
290 				loop_body_size += VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE;
291 			}
292 			loop_body_size += VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE;
293 		}
294 	}
295 
296 	if (has_bin) {
297 		size += VC4_PACKET_WAIT_ON_SEMAPHORE_SIZE;
298 		loop_body_size += VC4_PACKET_BRANCH_TO_SUB_LIST_SIZE;
299 	}
300 
301 	if (setup->msaa_color_write)
302 		loop_body_size += VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE;
303 	if (setup->msaa_zs_write)
304 		loop_body_size += VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE;
305 
306 	if (setup->zs_write)
307 		loop_body_size += VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE;
308 	if (setup->color_write)
309 		loop_body_size += VC4_PACKET_STORE_MS_TILE_BUFFER_SIZE;
310 
311 	/* We need a VC4_PACKET_TILE_COORDINATES in between each store. */
312 	loop_body_size += VC4_PACKET_TILE_COORDINATES_SIZE *
313 		((setup->msaa_color_write != NULL) +
314 		 (setup->msaa_zs_write != NULL) +
315 		 (setup->color_write != NULL) +
316 		 (setup->zs_write != NULL) - 1);
317 
318 	size += xtiles * ytiles * loop_body_size;
319 
320 	setup->rcl = &vc4_bo_create(dev, size, true)->base;
321 	if (IS_ERR(setup->rcl))
322 		return PTR_ERR(setup->rcl);
323 	list_add_tail(&to_vc4_bo(&setup->rcl->base)->unref_head,
324 		      &exec->unref_list);
325 
326 	/* The tile buffer gets cleared when the previous tile is stored.  If
327 	 * the clear values changed between frames, then the tile buffer has
328 	 * stale clear values in it, so we have to do a store in None mode (no
329 	 * writes) so that we trigger the tile buffer clear.
330 	 */
331 	if (args->flags & VC4_SUBMIT_CL_USE_CLEAR_COLOR) {
332 		rcl_u8(setup, VC4_PACKET_CLEAR_COLORS);
333 		rcl_u32(setup, args->clear_color[0]);
334 		rcl_u32(setup, args->clear_color[1]);
335 		rcl_u32(setup, args->clear_z);
336 		rcl_u8(setup, args->clear_s);
337 
338 		vc4_tile_coordinates(setup, 0, 0);
339 
340 		rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
341 		rcl_u16(setup, VC4_LOADSTORE_TILE_BUFFER_NONE);
342 		rcl_u32(setup, 0); /* no address, since we're in None mode */
343 	}
344 
345 	rcl_u8(setup, VC4_PACKET_TILE_RENDERING_MODE_CONFIG);
346 	rcl_u32(setup,
347 		(setup->color_write ? (setup->color_write->paddr +
348 				       args->color_write.offset) :
349 		 0));
350 	rcl_u16(setup, args->width);
351 	rcl_u16(setup, args->height);
352 	rcl_u16(setup, args->color_write.bits);
353 
354 	for (y = min_y_tile; y <= max_y_tile; y++) {
355 		for (x = min_x_tile; x <= max_x_tile; x++) {
356 			bool first = (x == min_x_tile && y == min_y_tile);
357 			bool last = (x == max_x_tile && y == max_y_tile);
358 
359 			emit_tile(exec, setup, x, y, first, last);
360 		}
361 	}
362 
363 	BUG_ON(setup->next_offset != size);
364 	exec->ct1ca = setup->rcl->paddr;
365 	exec->ct1ea = setup->rcl->paddr + setup->next_offset;
366 
367 	return 0;
368 }
369 
vc4_full_res_bounds_check(struct vc4_exec_info * exec,struct drm_gem_cma_object * obj,struct drm_vc4_submit_rcl_surface * surf)370 static int vc4_full_res_bounds_check(struct vc4_exec_info *exec,
371 				     struct drm_gem_cma_object *obj,
372 				     struct drm_vc4_submit_rcl_surface *surf)
373 {
374 	struct drm_vc4_submit_cl *args = exec->args;
375 	u32 render_tiles_stride = DIV_ROUND_UP(exec->args->width, 32);
376 
377 	if (surf->offset > obj->base.size) {
378 		DRM_ERROR("surface offset %d > BO size %zd\n",
379 			  surf->offset, obj->base.size);
380 		return -EINVAL;
381 	}
382 
383 	if ((obj->base.size - surf->offset) / VC4_TILE_BUFFER_SIZE <
384 	    render_tiles_stride * args->max_y_tile + args->max_x_tile) {
385 		DRM_ERROR("MSAA tile %d, %d out of bounds "
386 			  "(bo size %zd, offset %d).\n",
387 			  args->max_x_tile, args->max_y_tile,
388 			  obj->base.size,
389 			  surf->offset);
390 		return -EINVAL;
391 	}
392 
393 	return 0;
394 }
395 
vc4_rcl_msaa_surface_setup(struct vc4_exec_info * exec,struct drm_gem_cma_object ** obj,struct drm_vc4_submit_rcl_surface * surf)396 static int vc4_rcl_msaa_surface_setup(struct vc4_exec_info *exec,
397 				      struct drm_gem_cma_object **obj,
398 				      struct drm_vc4_submit_rcl_surface *surf)
399 {
400 	if (surf->flags != 0 || surf->bits != 0) {
401 		DRM_ERROR("MSAA surface had nonzero flags/bits\n");
402 		return -EINVAL;
403 	}
404 
405 	if (surf->hindex == ~0)
406 		return 0;
407 
408 	*obj = vc4_use_bo(exec, surf->hindex);
409 	if (!*obj)
410 		return -EINVAL;
411 
412 	exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
413 
414 	if (surf->offset & 0xf) {
415 		DRM_ERROR("MSAA write must be 16b aligned.\n");
416 		return -EINVAL;
417 	}
418 
419 	return vc4_full_res_bounds_check(exec, *obj, surf);
420 }
421 
vc4_rcl_surface_setup(struct vc4_exec_info * exec,struct drm_gem_cma_object ** obj,struct drm_vc4_submit_rcl_surface * surf,bool is_write)422 static int vc4_rcl_surface_setup(struct vc4_exec_info *exec,
423 				 struct drm_gem_cma_object **obj,
424 				 struct drm_vc4_submit_rcl_surface *surf,
425 				 bool is_write)
426 {
427 	uint8_t tiling = VC4_GET_FIELD(surf->bits,
428 				       VC4_LOADSTORE_TILE_BUFFER_TILING);
429 	uint8_t buffer = VC4_GET_FIELD(surf->bits,
430 				       VC4_LOADSTORE_TILE_BUFFER_BUFFER);
431 	uint8_t format = VC4_GET_FIELD(surf->bits,
432 				       VC4_LOADSTORE_TILE_BUFFER_FORMAT);
433 	int cpp;
434 	int ret;
435 
436 	if (surf->flags & ~VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
437 		DRM_ERROR("Extra flags set\n");
438 		return -EINVAL;
439 	}
440 
441 	if (surf->hindex == ~0)
442 		return 0;
443 
444 	*obj = vc4_use_bo(exec, surf->hindex);
445 	if (!*obj)
446 		return -EINVAL;
447 
448 	if (is_write)
449 		exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
450 
451 	if (surf->flags & VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
452 		if (surf == &exec->args->zs_write) {
453 			DRM_ERROR("general zs write may not be a full-res.\n");
454 			return -EINVAL;
455 		}
456 
457 		if (surf->bits != 0) {
458 			DRM_ERROR("load/store general bits set with "
459 				  "full res load/store.\n");
460 			return -EINVAL;
461 		}
462 
463 		ret = vc4_full_res_bounds_check(exec, *obj, surf);
464 		if (ret)
465 			return ret;
466 
467 		return 0;
468 	}
469 
470 	if (surf->bits & ~(VC4_LOADSTORE_TILE_BUFFER_TILING_MASK |
471 			   VC4_LOADSTORE_TILE_BUFFER_BUFFER_MASK |
472 			   VC4_LOADSTORE_TILE_BUFFER_FORMAT_MASK)) {
473 		DRM_ERROR("Unknown bits in load/store: 0x%04x\n",
474 			  surf->bits);
475 		return -EINVAL;
476 	}
477 
478 	if (tiling > VC4_TILING_FORMAT_LT) {
479 		DRM_ERROR("Bad tiling format\n");
480 		return -EINVAL;
481 	}
482 
483 	if (buffer == VC4_LOADSTORE_TILE_BUFFER_ZS) {
484 		if (format != 0) {
485 			DRM_ERROR("No color format should be set for ZS\n");
486 			return -EINVAL;
487 		}
488 		cpp = 4;
489 	} else if (buffer == VC4_LOADSTORE_TILE_BUFFER_COLOR) {
490 		switch (format) {
491 		case VC4_LOADSTORE_TILE_BUFFER_BGR565:
492 		case VC4_LOADSTORE_TILE_BUFFER_BGR565_DITHER:
493 			cpp = 2;
494 			break;
495 		case VC4_LOADSTORE_TILE_BUFFER_RGBA8888:
496 			cpp = 4;
497 			break;
498 		default:
499 			DRM_ERROR("Bad tile buffer format\n");
500 			return -EINVAL;
501 		}
502 	} else {
503 		DRM_ERROR("Bad load/store buffer %d.\n", buffer);
504 		return -EINVAL;
505 	}
506 
507 	if (surf->offset & 0xf) {
508 		DRM_ERROR("load/store buffer must be 16b aligned.\n");
509 		return -EINVAL;
510 	}
511 
512 	if (!vc4_check_tex_size(exec, *obj, surf->offset, tiling,
513 				exec->args->width, exec->args->height, cpp)) {
514 		return -EINVAL;
515 	}
516 
517 	return 0;
518 }
519 
520 static int
vc4_rcl_render_config_surface_setup(struct vc4_exec_info * exec,struct vc4_rcl_setup * setup,struct drm_gem_cma_object ** obj,struct drm_vc4_submit_rcl_surface * surf)521 vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec,
522 				    struct vc4_rcl_setup *setup,
523 				    struct drm_gem_cma_object **obj,
524 				    struct drm_vc4_submit_rcl_surface *surf)
525 {
526 	uint8_t tiling = VC4_GET_FIELD(surf->bits,
527 				       VC4_RENDER_CONFIG_MEMORY_FORMAT);
528 	uint8_t format = VC4_GET_FIELD(surf->bits,
529 				       VC4_RENDER_CONFIG_FORMAT);
530 	int cpp;
531 
532 	if (surf->flags != 0) {
533 		DRM_ERROR("No flags supported on render config.\n");
534 		return -EINVAL;
535 	}
536 
537 	if (surf->bits & ~(VC4_RENDER_CONFIG_MEMORY_FORMAT_MASK |
538 			   VC4_RENDER_CONFIG_FORMAT_MASK |
539 			   VC4_RENDER_CONFIG_MS_MODE_4X |
540 			   VC4_RENDER_CONFIG_DECIMATE_MODE_4X)) {
541 		DRM_ERROR("Unknown bits in render config: 0x%04x\n",
542 			  surf->bits);
543 		return -EINVAL;
544 	}
545 
546 	if (surf->hindex == ~0)
547 		return 0;
548 
549 	*obj = vc4_use_bo(exec, surf->hindex);
550 	if (!*obj)
551 		return -EINVAL;
552 
553 	exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
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 
vc4_get_rcl(struct drm_device * dev,struct vc4_exec_info * exec)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 				    false);
614 	if (ret)
615 		return ret;
616 
617 	ret = vc4_rcl_surface_setup(exec, &setup.zs_read, &args->zs_read,
618 				    false);
619 	if (ret)
620 		return ret;
621 
622 	ret = vc4_rcl_surface_setup(exec, &setup.zs_write, &args->zs_write,
623 				    true);
624 	if (ret)
625 		return ret;
626 
627 	ret = vc4_rcl_msaa_surface_setup(exec, &setup.msaa_color_write,
628 					 &args->msaa_color_write);
629 	if (ret)
630 		return ret;
631 
632 	ret = vc4_rcl_msaa_surface_setup(exec, &setup.msaa_zs_write,
633 					 &args->msaa_zs_write);
634 	if (ret)
635 		return ret;
636 
637 	/* We shouldn't even have the job submitted to us if there's no
638 	 * surface to write out.
639 	 */
640 	if (!setup.color_write && !setup.zs_write &&
641 	    !setup.msaa_color_write && !setup.msaa_zs_write) {
642 		DRM_ERROR("RCL requires color or Z/S write\n");
643 		return -EINVAL;
644 	}
645 
646 	return vc4_create_rcl_bo(dev, exec, &setup);
647 }
648