xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_validate.c (revision a6ca5ac746d104019e76c29e69c2a1fc6dd2b29f)
1 /*
2  * Copyright © 2014 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: Command list validator for VC4.
26  *
27  * Since the VC4 has no IOMMU between it and system memory, a user
28  * with access to execute command lists could escalate privilege by
29  * overwriting system memory (drawing to it as a framebuffer) or
30  * reading system memory it shouldn't (reading it as a vertex buffer
31  * or index buffer)
32  *
33  * We validate binner command lists to ensure that all accesses are
34  * within the bounds of the GEM objects referenced by the submitted
35  * job.  It explicitly whitelists packets, and looks at the offsets in
36  * any address fields to make sure they're contained within the BOs
37  * they reference.
38  *
39  * Note that because CL validation is already reading the
40  * user-submitted CL and writing the validated copy out to the memory
41  * that the GPU will actually read, this is also where GEM relocation
42  * processing (turning BO references into actual addresses for the GPU
43  * to use) happens.
44  */
45 
46 #include "uapi/drm/vc4_drm.h"
47 #include "vc4_drv.h"
48 #include "vc4_packet.h"
49 
50 #define VALIDATE_ARGS \
51 	struct vc4_exec_info *exec,			\
52 	void *validated,				\
53 	void *untrusted
54 
55 /** Return the width in pixels of a 64-byte microtile. */
56 static uint32_t
57 utile_width(int cpp)
58 {
59 	switch (cpp) {
60 	case 1:
61 	case 2:
62 		return 8;
63 	case 4:
64 		return 4;
65 	case 8:
66 		return 2;
67 	default:
68 		DRM_ERROR("unknown cpp: %d\n", cpp);
69 		return 1;
70 	}
71 }
72 
73 /** Return the height in pixels of a 64-byte microtile. */
74 static uint32_t
75 utile_height(int cpp)
76 {
77 	switch (cpp) {
78 	case 1:
79 		return 8;
80 	case 2:
81 	case 4:
82 	case 8:
83 		return 4;
84 	default:
85 		DRM_ERROR("unknown cpp: %d\n", cpp);
86 		return 1;
87 	}
88 }
89 
90 /**
91  * size_is_lt() - Returns whether a miplevel of the given size will
92  * use the lineartile (LT) tiling layout rather than the normal T
93  * tiling layout.
94  * @width: Width in pixels of the miplevel
95  * @height: Height in pixels of the miplevel
96  * @cpp: Bytes per pixel of the pixel format
97  */
98 static bool
99 size_is_lt(uint32_t width, uint32_t height, int cpp)
100 {
101 	return (width <= 4 * utile_width(cpp) ||
102 		height <= 4 * utile_height(cpp));
103 }
104 
105 struct drm_gem_cma_object *
106 vc4_use_bo(struct vc4_exec_info *exec, uint32_t hindex)
107 {
108 	struct drm_gem_cma_object *obj;
109 	struct vc4_bo *bo;
110 
111 	if (hindex >= exec->bo_count) {
112 		DRM_ERROR("BO index %d greater than BO count %d\n",
113 			  hindex, exec->bo_count);
114 		return NULL;
115 	}
116 	obj = exec->bo[hindex];
117 	bo = to_vc4_bo(&obj->base);
118 
119 	if (bo->validated_shader) {
120 		DRM_ERROR("Trying to use shader BO as something other than "
121 			  "a shader\n");
122 		return NULL;
123 	}
124 
125 	return obj;
126 }
127 
128 static struct drm_gem_cma_object *
129 vc4_use_handle(struct vc4_exec_info *exec, uint32_t gem_handles_packet_index)
130 {
131 	return vc4_use_bo(exec, exec->bo_index[gem_handles_packet_index]);
132 }
133 
134 static bool
135 validate_bin_pos(struct vc4_exec_info *exec, void *untrusted, uint32_t pos)
136 {
137 	/* Note that the untrusted pointer passed to these functions is
138 	 * incremented past the packet byte.
139 	 */
140 	return (untrusted - 1 == exec->bin_u + pos);
141 }
142 
143 static uint32_t
144 gl_shader_rec_size(uint32_t pointer_bits)
145 {
146 	uint32_t attribute_count = pointer_bits & 7;
147 	bool extended = pointer_bits & 8;
148 
149 	if (attribute_count == 0)
150 		attribute_count = 8;
151 
152 	if (extended)
153 		return 100 + attribute_count * 4;
154 	else
155 		return 36 + attribute_count * 8;
156 }
157 
158 bool
159 vc4_check_tex_size(struct vc4_exec_info *exec, struct drm_gem_cma_object *fbo,
160 		   uint32_t offset, uint8_t tiling_format,
161 		   uint32_t width, uint32_t height, uint8_t cpp)
162 {
163 	uint32_t aligned_width, aligned_height, stride, size;
164 	uint32_t utile_w = utile_width(cpp);
165 	uint32_t utile_h = utile_height(cpp);
166 
167 	/* The shaded vertex format stores signed 12.4 fixed point
168 	 * (-2048,2047) offsets from the viewport center, so we should
169 	 * never have a render target larger than 4096.  The texture
170 	 * unit can only sample from 2048x2048, so it's even more
171 	 * restricted.  This lets us avoid worrying about overflow in
172 	 * our math.
173 	 */
174 	if (width > 4096 || height > 4096) {
175 		DRM_ERROR("Surface dimesions (%d,%d) too large", width, height);
176 		return false;
177 	}
178 
179 	switch (tiling_format) {
180 	case VC4_TILING_FORMAT_LINEAR:
181 		aligned_width = round_up(width, utile_w);
182 		aligned_height = height;
183 		break;
184 	case VC4_TILING_FORMAT_T:
185 		aligned_width = round_up(width, utile_w * 8);
186 		aligned_height = round_up(height, utile_h * 8);
187 		break;
188 	case VC4_TILING_FORMAT_LT:
189 		aligned_width = round_up(width, utile_w);
190 		aligned_height = round_up(height, utile_h);
191 		break;
192 	default:
193 		DRM_ERROR("buffer tiling %d unsupported\n", tiling_format);
194 		return false;
195 	}
196 
197 	stride = aligned_width * cpp;
198 	size = stride * aligned_height;
199 
200 	if (size + offset < size ||
201 	    size + offset > fbo->base.size) {
202 		DRM_ERROR("Overflow in %dx%d (%dx%d) fbo size (%d + %d > %zd)\n",
203 			  width, height,
204 			  aligned_width, aligned_height,
205 			  size, offset, fbo->base.size);
206 		return false;
207 	}
208 
209 	return true;
210 }
211 
212 static int
213 validate_flush(VALIDATE_ARGS)
214 {
215 	if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 1)) {
216 		DRM_ERROR("Bin CL must end with VC4_PACKET_FLUSH\n");
217 		return -EINVAL;
218 	}
219 	exec->found_flush = true;
220 
221 	return 0;
222 }
223 
224 static int
225 validate_start_tile_binning(VALIDATE_ARGS)
226 {
227 	if (exec->found_start_tile_binning_packet) {
228 		DRM_ERROR("Duplicate VC4_PACKET_START_TILE_BINNING\n");
229 		return -EINVAL;
230 	}
231 	exec->found_start_tile_binning_packet = true;
232 
233 	if (!exec->found_tile_binning_mode_config_packet) {
234 		DRM_ERROR("missing VC4_PACKET_TILE_BINNING_MODE_CONFIG\n");
235 		return -EINVAL;
236 	}
237 
238 	return 0;
239 }
240 
241 static int
242 validate_increment_semaphore(VALIDATE_ARGS)
243 {
244 	if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 2)) {
245 		DRM_ERROR("Bin CL must end with "
246 			  "VC4_PACKET_INCREMENT_SEMAPHORE\n");
247 		return -EINVAL;
248 	}
249 	exec->found_increment_semaphore_packet = true;
250 
251 	return 0;
252 }
253 
254 static int
255 validate_indexed_prim_list(VALIDATE_ARGS)
256 {
257 	struct drm_gem_cma_object *ib;
258 	uint32_t length = *(uint32_t *)(untrusted + 1);
259 	uint32_t offset = *(uint32_t *)(untrusted + 5);
260 	uint32_t max_index = *(uint32_t *)(untrusted + 9);
261 	uint32_t index_size = (*(uint8_t *)(untrusted + 0) >> 4) ? 2 : 1;
262 	struct vc4_shader_state *shader_state;
263 
264 	/* Check overflow condition */
265 	if (exec->shader_state_count == 0) {
266 		DRM_ERROR("shader state must precede primitives\n");
267 		return -EINVAL;
268 	}
269 	shader_state = &exec->shader_state[exec->shader_state_count - 1];
270 
271 	if (max_index > shader_state->max_index)
272 		shader_state->max_index = max_index;
273 
274 	ib = vc4_use_handle(exec, 0);
275 	if (!ib)
276 		return -EINVAL;
277 
278 	exec->bin_dep_seqno = max(exec->bin_dep_seqno,
279 				  to_vc4_bo(&ib->base)->write_seqno);
280 
281 	if (offset > ib->base.size ||
282 	    (ib->base.size - offset) / index_size < length) {
283 		DRM_ERROR("IB access overflow (%d + %d*%d > %zd)\n",
284 			  offset, length, index_size, ib->base.size);
285 		return -EINVAL;
286 	}
287 
288 	*(uint32_t *)(validated + 5) = ib->paddr + offset;
289 
290 	return 0;
291 }
292 
293 static int
294 validate_gl_array_primitive(VALIDATE_ARGS)
295 {
296 	uint32_t length = *(uint32_t *)(untrusted + 1);
297 	uint32_t base_index = *(uint32_t *)(untrusted + 5);
298 	uint32_t max_index;
299 	struct vc4_shader_state *shader_state;
300 
301 	/* Check overflow condition */
302 	if (exec->shader_state_count == 0) {
303 		DRM_ERROR("shader state must precede primitives\n");
304 		return -EINVAL;
305 	}
306 	shader_state = &exec->shader_state[exec->shader_state_count - 1];
307 
308 	if (length + base_index < length) {
309 		DRM_ERROR("primitive vertex count overflow\n");
310 		return -EINVAL;
311 	}
312 	max_index = length + base_index - 1;
313 
314 	if (max_index > shader_state->max_index)
315 		shader_state->max_index = max_index;
316 
317 	return 0;
318 }
319 
320 static int
321 validate_gl_shader_state(VALIDATE_ARGS)
322 {
323 	uint32_t i = exec->shader_state_count++;
324 
325 	if (i >= exec->shader_state_size) {
326 		DRM_ERROR("More requests for shader states than declared\n");
327 		return -EINVAL;
328 	}
329 
330 	exec->shader_state[i].addr = *(uint32_t *)untrusted;
331 	exec->shader_state[i].max_index = 0;
332 
333 	if (exec->shader_state[i].addr & ~0xf) {
334 		DRM_ERROR("high bits set in GL shader rec reference\n");
335 		return -EINVAL;
336 	}
337 
338 	*(uint32_t *)validated = (exec->shader_rec_p +
339 				  exec->shader_state[i].addr);
340 
341 	exec->shader_rec_p +=
342 		roundup(gl_shader_rec_size(exec->shader_state[i].addr), 16);
343 
344 	return 0;
345 }
346 
347 static int
348 validate_tile_binning_config(VALIDATE_ARGS)
349 {
350 	struct drm_device *dev = exec->exec_bo->base.dev;
351 	struct vc4_dev *vc4 = to_vc4_dev(dev);
352 	uint8_t flags;
353 	uint32_t tile_state_size;
354 	uint32_t tile_count, bin_addr;
355 	int bin_slot;
356 
357 	if (exec->found_tile_binning_mode_config_packet) {
358 		DRM_ERROR("Duplicate VC4_PACKET_TILE_BINNING_MODE_CONFIG\n");
359 		return -EINVAL;
360 	}
361 	exec->found_tile_binning_mode_config_packet = true;
362 
363 	exec->bin_tiles_x = *(uint8_t *)(untrusted + 12);
364 	exec->bin_tiles_y = *(uint8_t *)(untrusted + 13);
365 	tile_count = exec->bin_tiles_x * exec->bin_tiles_y;
366 	flags = *(uint8_t *)(untrusted + 14);
367 
368 	if (exec->bin_tiles_x == 0 ||
369 	    exec->bin_tiles_y == 0) {
370 		DRM_ERROR("Tile binning config of %dx%d too small\n",
371 			  exec->bin_tiles_x, exec->bin_tiles_y);
372 		return -EINVAL;
373 	}
374 
375 	if (flags & (VC4_BIN_CONFIG_DB_NON_MS |
376 		     VC4_BIN_CONFIG_TILE_BUFFER_64BIT)) {
377 		DRM_ERROR("unsupported binning config flags 0x%02x\n", flags);
378 		return -EINVAL;
379 	}
380 
381 	bin_slot = vc4_v3d_get_bin_slot(vc4);
382 	if (bin_slot < 0) {
383 		if (bin_slot != -EINTR && bin_slot != -ERESTARTSYS) {
384 			DRM_ERROR("Failed to allocate binner memory: %d\n",
385 				  bin_slot);
386 		}
387 		return bin_slot;
388 	}
389 
390 	/* The slot we allocated will only be used by this job, and is
391 	 * free when the job completes rendering.
392 	 */
393 	exec->bin_slots |= BIT(bin_slot);
394 	bin_addr = vc4->bin_bo->base.paddr + bin_slot * vc4->bin_alloc_size;
395 
396 	/* The tile state data array is 48 bytes per tile, and we put it at
397 	 * the start of a BO containing both it and the tile alloc.
398 	 */
399 	tile_state_size = 48 * tile_count;
400 
401 	/* Since the tile alloc array will follow us, align. */
402 	exec->tile_alloc_offset = bin_addr + roundup(tile_state_size, 4096);
403 
404 	*(uint8_t *)(validated + 14) =
405 		((flags & ~(VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_MASK |
406 			    VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_MASK)) |
407 		 VC4_BIN_CONFIG_AUTO_INIT_TSDA |
408 		 VC4_SET_FIELD(VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_32,
409 			       VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE) |
410 		 VC4_SET_FIELD(VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_128,
411 			       VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE));
412 
413 	/* tile alloc address. */
414 	*(uint32_t *)(validated + 0) = exec->tile_alloc_offset;
415 	/* tile alloc size. */
416 	*(uint32_t *)(validated + 4) = (bin_addr + vc4->bin_alloc_size -
417 					exec->tile_alloc_offset);
418 	/* tile state address. */
419 	*(uint32_t *)(validated + 8) = bin_addr;
420 
421 	return 0;
422 }
423 
424 static int
425 validate_gem_handles(VALIDATE_ARGS)
426 {
427 	memcpy(exec->bo_index, untrusted, sizeof(exec->bo_index));
428 	return 0;
429 }
430 
431 #define VC4_DEFINE_PACKET(packet, func) \
432 	[packet] = { packet ## _SIZE, #packet, func }
433 
434 static const struct cmd_info {
435 	uint16_t len;
436 	const char *name;
437 	int (*func)(struct vc4_exec_info *exec, void *validated,
438 		    void *untrusted);
439 } cmd_info[] = {
440 	VC4_DEFINE_PACKET(VC4_PACKET_HALT, NULL),
441 	VC4_DEFINE_PACKET(VC4_PACKET_NOP, NULL),
442 	VC4_DEFINE_PACKET(VC4_PACKET_FLUSH, validate_flush),
443 	VC4_DEFINE_PACKET(VC4_PACKET_FLUSH_ALL, NULL),
444 	VC4_DEFINE_PACKET(VC4_PACKET_START_TILE_BINNING,
445 			  validate_start_tile_binning),
446 	VC4_DEFINE_PACKET(VC4_PACKET_INCREMENT_SEMAPHORE,
447 			  validate_increment_semaphore),
448 
449 	VC4_DEFINE_PACKET(VC4_PACKET_GL_INDEXED_PRIMITIVE,
450 			  validate_indexed_prim_list),
451 	VC4_DEFINE_PACKET(VC4_PACKET_GL_ARRAY_PRIMITIVE,
452 			  validate_gl_array_primitive),
453 
454 	VC4_DEFINE_PACKET(VC4_PACKET_PRIMITIVE_LIST_FORMAT, NULL),
455 
456 	VC4_DEFINE_PACKET(VC4_PACKET_GL_SHADER_STATE, validate_gl_shader_state),
457 
458 	VC4_DEFINE_PACKET(VC4_PACKET_CONFIGURATION_BITS, NULL),
459 	VC4_DEFINE_PACKET(VC4_PACKET_FLAT_SHADE_FLAGS, NULL),
460 	VC4_DEFINE_PACKET(VC4_PACKET_POINT_SIZE, NULL),
461 	VC4_DEFINE_PACKET(VC4_PACKET_LINE_WIDTH, NULL),
462 	VC4_DEFINE_PACKET(VC4_PACKET_RHT_X_BOUNDARY, NULL),
463 	VC4_DEFINE_PACKET(VC4_PACKET_DEPTH_OFFSET, NULL),
464 	VC4_DEFINE_PACKET(VC4_PACKET_CLIP_WINDOW, NULL),
465 	VC4_DEFINE_PACKET(VC4_PACKET_VIEWPORT_OFFSET, NULL),
466 	VC4_DEFINE_PACKET(VC4_PACKET_CLIPPER_XY_SCALING, NULL),
467 	/* Note: The docs say this was also 105, but it was 106 in the
468 	 * initial userland code drop.
469 	 */
470 	VC4_DEFINE_PACKET(VC4_PACKET_CLIPPER_Z_SCALING, NULL),
471 
472 	VC4_DEFINE_PACKET(VC4_PACKET_TILE_BINNING_MODE_CONFIG,
473 			  validate_tile_binning_config),
474 
475 	VC4_DEFINE_PACKET(VC4_PACKET_GEM_HANDLES, validate_gem_handles),
476 };
477 
478 int
479 vc4_validate_bin_cl(struct drm_device *dev,
480 		    void *validated,
481 		    void *unvalidated,
482 		    struct vc4_exec_info *exec)
483 {
484 	uint32_t len = exec->args->bin_cl_size;
485 	uint32_t dst_offset = 0;
486 	uint32_t src_offset = 0;
487 
488 	while (src_offset < len) {
489 		void *dst_pkt = validated + dst_offset;
490 		void *src_pkt = unvalidated + src_offset;
491 		u8 cmd = *(uint8_t *)src_pkt;
492 		const struct cmd_info *info;
493 
494 		if (cmd >= ARRAY_SIZE(cmd_info)) {
495 			DRM_ERROR("0x%08x: packet %d out of bounds\n",
496 				  src_offset, cmd);
497 			return -EINVAL;
498 		}
499 
500 		info = &cmd_info[cmd];
501 		if (!info->name) {
502 			DRM_ERROR("0x%08x: packet %d invalid\n",
503 				  src_offset, cmd);
504 			return -EINVAL;
505 		}
506 
507 		if (src_offset + info->len > len) {
508 			DRM_ERROR("0x%08x: packet %d (%s) length 0x%08x "
509 				  "exceeds bounds (0x%08x)\n",
510 				  src_offset, cmd, info->name, info->len,
511 				  src_offset + len);
512 			return -EINVAL;
513 		}
514 
515 		if (cmd != VC4_PACKET_GEM_HANDLES)
516 			memcpy(dst_pkt, src_pkt, info->len);
517 
518 		if (info->func && info->func(exec,
519 					     dst_pkt + 1,
520 					     src_pkt + 1)) {
521 			DRM_ERROR("0x%08x: packet %d (%s) failed to validate\n",
522 				  src_offset, cmd, info->name);
523 			return -EINVAL;
524 		}
525 
526 		src_offset += info->len;
527 		/* GEM handle loading doesn't produce HW packets. */
528 		if (cmd != VC4_PACKET_GEM_HANDLES)
529 			dst_offset += info->len;
530 
531 		/* When the CL hits halt, it'll stop reading anything else. */
532 		if (cmd == VC4_PACKET_HALT)
533 			break;
534 	}
535 
536 	exec->ct0ea = exec->ct0ca + dst_offset;
537 
538 	if (!exec->found_start_tile_binning_packet) {
539 		DRM_ERROR("Bin CL missing VC4_PACKET_START_TILE_BINNING\n");
540 		return -EINVAL;
541 	}
542 
543 	/* The bin CL must be ended with INCREMENT_SEMAPHORE and FLUSH.  The
544 	 * semaphore is used to trigger the render CL to start up, and the
545 	 * FLUSH is what caps the bin lists with
546 	 * VC4_PACKET_RETURN_FROM_SUB_LIST (so they jump back to the main
547 	 * render CL when they get called to) and actually triggers the queued
548 	 * semaphore increment.
549 	 */
550 	if (!exec->found_increment_semaphore_packet || !exec->found_flush) {
551 		DRM_ERROR("Bin CL missing VC4_PACKET_INCREMENT_SEMAPHORE + "
552 			  "VC4_PACKET_FLUSH\n");
553 		return -EINVAL;
554 	}
555 
556 	return 0;
557 }
558 
559 static bool
560 reloc_tex(struct vc4_exec_info *exec,
561 	  void *uniform_data_u,
562 	  struct vc4_texture_sample_info *sample,
563 	  uint32_t texture_handle_index, bool is_cs)
564 {
565 	struct drm_gem_cma_object *tex;
566 	uint32_t p0 = *(uint32_t *)(uniform_data_u + sample->p_offset[0]);
567 	uint32_t p1 = *(uint32_t *)(uniform_data_u + sample->p_offset[1]);
568 	uint32_t p2 = (sample->p_offset[2] != ~0 ?
569 		       *(uint32_t *)(uniform_data_u + sample->p_offset[2]) : 0);
570 	uint32_t p3 = (sample->p_offset[3] != ~0 ?
571 		       *(uint32_t *)(uniform_data_u + sample->p_offset[3]) : 0);
572 	uint32_t *validated_p0 = exec->uniforms_v + sample->p_offset[0];
573 	uint32_t offset = p0 & VC4_TEX_P0_OFFSET_MASK;
574 	uint32_t miplevels = VC4_GET_FIELD(p0, VC4_TEX_P0_MIPLVLS);
575 	uint32_t width = VC4_GET_FIELD(p1, VC4_TEX_P1_WIDTH);
576 	uint32_t height = VC4_GET_FIELD(p1, VC4_TEX_P1_HEIGHT);
577 	uint32_t cpp, tiling_format, utile_w, utile_h;
578 	uint32_t i;
579 	uint32_t cube_map_stride = 0;
580 	enum vc4_texture_data_type type;
581 
582 	tex = vc4_use_bo(exec, texture_handle_index);
583 	if (!tex)
584 		return false;
585 
586 	if (sample->is_direct) {
587 		uint32_t remaining_size = tex->base.size - p0;
588 
589 		if (p0 > tex->base.size - 4) {
590 			DRM_ERROR("UBO offset greater than UBO size\n");
591 			goto fail;
592 		}
593 		if (p1 > remaining_size - 4) {
594 			DRM_ERROR("UBO clamp would allow reads "
595 				  "outside of UBO\n");
596 			goto fail;
597 		}
598 		*validated_p0 = tex->paddr + p0;
599 		return true;
600 	}
601 
602 	if (width == 0)
603 		width = 2048;
604 	if (height == 0)
605 		height = 2048;
606 
607 	if (p0 & VC4_TEX_P0_CMMODE_MASK) {
608 		if (VC4_GET_FIELD(p2, VC4_TEX_P2_PTYPE) ==
609 		    VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE)
610 			cube_map_stride = p2 & VC4_TEX_P2_CMST_MASK;
611 		if (VC4_GET_FIELD(p3, VC4_TEX_P2_PTYPE) ==
612 		    VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE) {
613 			if (cube_map_stride) {
614 				DRM_ERROR("Cube map stride set twice\n");
615 				goto fail;
616 			}
617 
618 			cube_map_stride = p3 & VC4_TEX_P2_CMST_MASK;
619 		}
620 		if (!cube_map_stride) {
621 			DRM_ERROR("Cube map stride not set\n");
622 			goto fail;
623 		}
624 	}
625 
626 	type = (VC4_GET_FIELD(p0, VC4_TEX_P0_TYPE) |
627 		(VC4_GET_FIELD(p1, VC4_TEX_P1_TYPE4) << 4));
628 
629 	switch (type) {
630 	case VC4_TEXTURE_TYPE_RGBA8888:
631 	case VC4_TEXTURE_TYPE_RGBX8888:
632 	case VC4_TEXTURE_TYPE_RGBA32R:
633 		cpp = 4;
634 		break;
635 	case VC4_TEXTURE_TYPE_RGBA4444:
636 	case VC4_TEXTURE_TYPE_RGBA5551:
637 	case VC4_TEXTURE_TYPE_RGB565:
638 	case VC4_TEXTURE_TYPE_LUMALPHA:
639 	case VC4_TEXTURE_TYPE_S16F:
640 	case VC4_TEXTURE_TYPE_S16:
641 		cpp = 2;
642 		break;
643 	case VC4_TEXTURE_TYPE_LUMINANCE:
644 	case VC4_TEXTURE_TYPE_ALPHA:
645 	case VC4_TEXTURE_TYPE_S8:
646 		cpp = 1;
647 		break;
648 	case VC4_TEXTURE_TYPE_ETC1:
649 		/* ETC1 is arranged as 64-bit blocks, where each block is 4x4
650 		 * pixels.
651 		 */
652 		cpp = 8;
653 		width = (width + 3) >> 2;
654 		height = (height + 3) >> 2;
655 		break;
656 	case VC4_TEXTURE_TYPE_BW1:
657 	case VC4_TEXTURE_TYPE_A4:
658 	case VC4_TEXTURE_TYPE_A1:
659 	case VC4_TEXTURE_TYPE_RGBA64:
660 	case VC4_TEXTURE_TYPE_YUV422R:
661 	default:
662 		DRM_ERROR("Texture format %d unsupported\n", type);
663 		goto fail;
664 	}
665 	utile_w = utile_width(cpp);
666 	utile_h = utile_height(cpp);
667 
668 	if (type == VC4_TEXTURE_TYPE_RGBA32R) {
669 		tiling_format = VC4_TILING_FORMAT_LINEAR;
670 	} else {
671 		if (size_is_lt(width, height, cpp))
672 			tiling_format = VC4_TILING_FORMAT_LT;
673 		else
674 			tiling_format = VC4_TILING_FORMAT_T;
675 	}
676 
677 	if (!vc4_check_tex_size(exec, tex, offset + cube_map_stride * 5,
678 				tiling_format, width, height, cpp)) {
679 		goto fail;
680 	}
681 
682 	/* The mipmap levels are stored before the base of the texture.  Make
683 	 * sure there is actually space in the BO.
684 	 */
685 	for (i = 1; i <= miplevels; i++) {
686 		uint32_t level_width = max(width >> i, 1u);
687 		uint32_t level_height = max(height >> i, 1u);
688 		uint32_t aligned_width, aligned_height;
689 		uint32_t level_size;
690 
691 		/* Once the levels get small enough, they drop from T to LT. */
692 		if (tiling_format == VC4_TILING_FORMAT_T &&
693 		    size_is_lt(level_width, level_height, cpp)) {
694 			tiling_format = VC4_TILING_FORMAT_LT;
695 		}
696 
697 		switch (tiling_format) {
698 		case VC4_TILING_FORMAT_T:
699 			aligned_width = round_up(level_width, utile_w * 8);
700 			aligned_height = round_up(level_height, utile_h * 8);
701 			break;
702 		case VC4_TILING_FORMAT_LT:
703 			aligned_width = round_up(level_width, utile_w);
704 			aligned_height = round_up(level_height, utile_h);
705 			break;
706 		default:
707 			aligned_width = round_up(level_width, utile_w);
708 			aligned_height = level_height;
709 			break;
710 		}
711 
712 		level_size = aligned_width * cpp * aligned_height;
713 
714 		if (offset < level_size) {
715 			DRM_ERROR("Level %d (%dx%d -> %dx%d) size %db "
716 				  "overflowed buffer bounds (offset %d)\n",
717 				  i, level_width, level_height,
718 				  aligned_width, aligned_height,
719 				  level_size, offset);
720 			goto fail;
721 		}
722 
723 		offset -= level_size;
724 	}
725 
726 	*validated_p0 = tex->paddr + p0;
727 
728 	if (is_cs) {
729 		exec->bin_dep_seqno = max(exec->bin_dep_seqno,
730 					  to_vc4_bo(&tex->base)->write_seqno);
731 	}
732 
733 	return true;
734  fail:
735 	DRM_INFO("Texture p0 at %d: 0x%08x\n", sample->p_offset[0], p0);
736 	DRM_INFO("Texture p1 at %d: 0x%08x\n", sample->p_offset[1], p1);
737 	DRM_INFO("Texture p2 at %d: 0x%08x\n", sample->p_offset[2], p2);
738 	DRM_INFO("Texture p3 at %d: 0x%08x\n", sample->p_offset[3], p3);
739 	return false;
740 }
741 
742 static int
743 validate_gl_shader_rec(struct drm_device *dev,
744 		       struct vc4_exec_info *exec,
745 		       struct vc4_shader_state *state)
746 {
747 	uint32_t *src_handles;
748 	void *pkt_u, *pkt_v;
749 	static const uint32_t shader_reloc_offsets[] = {
750 		4, /* fs */
751 		16, /* vs */
752 		28, /* cs */
753 	};
754 	uint32_t shader_reloc_count = ARRAY_SIZE(shader_reloc_offsets);
755 	struct drm_gem_cma_object *bo[shader_reloc_count + 8];
756 	uint32_t nr_attributes, nr_relocs, packet_size;
757 	int i;
758 
759 	nr_attributes = state->addr & 0x7;
760 	if (nr_attributes == 0)
761 		nr_attributes = 8;
762 	packet_size = gl_shader_rec_size(state->addr);
763 
764 	nr_relocs = ARRAY_SIZE(shader_reloc_offsets) + nr_attributes;
765 	if (nr_relocs * 4 > exec->shader_rec_size) {
766 		DRM_ERROR("overflowed shader recs reading %d handles "
767 			  "from %d bytes left\n",
768 			  nr_relocs, exec->shader_rec_size);
769 		return -EINVAL;
770 	}
771 	src_handles = exec->shader_rec_u;
772 	exec->shader_rec_u += nr_relocs * 4;
773 	exec->shader_rec_size -= nr_relocs * 4;
774 
775 	if (packet_size > exec->shader_rec_size) {
776 		DRM_ERROR("overflowed shader recs copying %db packet "
777 			  "from %d bytes left\n",
778 			  packet_size, exec->shader_rec_size);
779 		return -EINVAL;
780 	}
781 	pkt_u = exec->shader_rec_u;
782 	pkt_v = exec->shader_rec_v;
783 	memcpy(pkt_v, pkt_u, packet_size);
784 	exec->shader_rec_u += packet_size;
785 	/* Shader recs have to be aligned to 16 bytes (due to the attribute
786 	 * flags being in the low bytes), so round the next validated shader
787 	 * rec address up.  This should be safe, since we've got so many
788 	 * relocations in a shader rec packet.
789 	 */
790 	BUG_ON(roundup(packet_size, 16) - packet_size > nr_relocs * 4);
791 	exec->shader_rec_v += roundup(packet_size, 16);
792 	exec->shader_rec_size -= packet_size;
793 
794 	for (i = 0; i < shader_reloc_count; i++) {
795 		if (src_handles[i] > exec->bo_count) {
796 			DRM_ERROR("Shader handle %d too big\n", src_handles[i]);
797 			return -EINVAL;
798 		}
799 
800 		bo[i] = exec->bo[src_handles[i]];
801 		if (!bo[i])
802 			return -EINVAL;
803 	}
804 	for (i = shader_reloc_count; i < nr_relocs; i++) {
805 		bo[i] = vc4_use_bo(exec, src_handles[i]);
806 		if (!bo[i])
807 			return -EINVAL;
808 	}
809 
810 	if (((*(uint16_t *)pkt_u & VC4_SHADER_FLAG_FS_SINGLE_THREAD) == 0) !=
811 	    to_vc4_bo(&bo[0]->base)->validated_shader->is_threaded) {
812 		DRM_ERROR("Thread mode of CL and FS do not match\n");
813 		return -EINVAL;
814 	}
815 
816 	if (to_vc4_bo(&bo[1]->base)->validated_shader->is_threaded ||
817 	    to_vc4_bo(&bo[2]->base)->validated_shader->is_threaded) {
818 		DRM_ERROR("cs and vs cannot be threaded\n");
819 		return -EINVAL;
820 	}
821 
822 	for (i = 0; i < shader_reloc_count; i++) {
823 		struct vc4_validated_shader_info *validated_shader;
824 		uint32_t o = shader_reloc_offsets[i];
825 		uint32_t src_offset = *(uint32_t *)(pkt_u + o);
826 		uint32_t *texture_handles_u;
827 		void *uniform_data_u;
828 		uint32_t tex, uni;
829 
830 		*(uint32_t *)(pkt_v + o) = bo[i]->paddr + src_offset;
831 
832 		if (src_offset != 0) {
833 			DRM_ERROR("Shaders must be at offset 0 of "
834 				  "the BO.\n");
835 			return -EINVAL;
836 		}
837 
838 		validated_shader = to_vc4_bo(&bo[i]->base)->validated_shader;
839 		if (!validated_shader)
840 			return -EINVAL;
841 
842 		if (validated_shader->uniforms_src_size >
843 		    exec->uniforms_size) {
844 			DRM_ERROR("Uniforms src buffer overflow\n");
845 			return -EINVAL;
846 		}
847 
848 		texture_handles_u = exec->uniforms_u;
849 		uniform_data_u = (texture_handles_u +
850 				  validated_shader->num_texture_samples);
851 
852 		memcpy(exec->uniforms_v, uniform_data_u,
853 		       validated_shader->uniforms_size);
854 
855 		for (tex = 0;
856 		     tex < validated_shader->num_texture_samples;
857 		     tex++) {
858 			if (!reloc_tex(exec,
859 				       uniform_data_u,
860 				       &validated_shader->texture_samples[tex],
861 				       texture_handles_u[tex],
862 				       i == 2)) {
863 				return -EINVAL;
864 			}
865 		}
866 
867 		/* Fill in the uniform slots that need this shader's
868 		 * start-of-uniforms address (used for resetting the uniform
869 		 * stream in the presence of control flow).
870 		 */
871 		for (uni = 0;
872 		     uni < validated_shader->num_uniform_addr_offsets;
873 		     uni++) {
874 			uint32_t o = validated_shader->uniform_addr_offsets[uni];
875 			((uint32_t *)exec->uniforms_v)[o] = exec->uniforms_p;
876 		}
877 
878 		*(uint32_t *)(pkt_v + o + 4) = exec->uniforms_p;
879 
880 		exec->uniforms_u += validated_shader->uniforms_src_size;
881 		exec->uniforms_v += validated_shader->uniforms_size;
882 		exec->uniforms_p += validated_shader->uniforms_size;
883 	}
884 
885 	for (i = 0; i < nr_attributes; i++) {
886 		struct drm_gem_cma_object *vbo =
887 			bo[ARRAY_SIZE(shader_reloc_offsets) + i];
888 		uint32_t o = 36 + i * 8;
889 		uint32_t offset = *(uint32_t *)(pkt_u + o + 0);
890 		uint32_t attr_size = *(uint8_t *)(pkt_u + o + 4) + 1;
891 		uint32_t stride = *(uint8_t *)(pkt_u + o + 5);
892 		uint32_t max_index;
893 
894 		exec->bin_dep_seqno = max(exec->bin_dep_seqno,
895 					  to_vc4_bo(&vbo->base)->write_seqno);
896 
897 		if (state->addr & 0x8)
898 			stride |= (*(uint32_t *)(pkt_u + 100 + i * 4)) & ~0xff;
899 
900 		if (vbo->base.size < offset ||
901 		    vbo->base.size - offset < attr_size) {
902 			DRM_ERROR("BO offset overflow (%d + %d > %zu)\n",
903 				  offset, attr_size, vbo->base.size);
904 			return -EINVAL;
905 		}
906 
907 		if (stride != 0) {
908 			max_index = ((vbo->base.size - offset - attr_size) /
909 				     stride);
910 			if (state->max_index > max_index) {
911 				DRM_ERROR("primitives use index %d out of "
912 					  "supplied %d\n",
913 					  state->max_index, max_index);
914 				return -EINVAL;
915 			}
916 		}
917 
918 		*(uint32_t *)(pkt_v + o) = vbo->paddr + offset;
919 	}
920 
921 	return 0;
922 }
923 
924 int
925 vc4_validate_shader_recs(struct drm_device *dev,
926 			 struct vc4_exec_info *exec)
927 {
928 	uint32_t i;
929 	int ret = 0;
930 
931 	for (i = 0; i < exec->shader_state_count; i++) {
932 		ret = validate_gl_shader_rec(dev, exec, &exec->shader_state[i]);
933 		if (ret)
934 			return ret;
935 	}
936 
937 	return ret;
938 }
939