xref: /openbmc/linux/drivers/gpu/drm/radeon/r600_cs.c (revision 987abe82)
1 /*
2  * Copyright 2008 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include <linux/kernel.h>
29 
30 #include "radeon.h"
31 #include "radeon_asic.h"
32 #include "r600.h"
33 #include "r600d.h"
34 #include "r600_reg_safe.h"
35 
36 static int r600_nomm;
37 
38 struct r600_cs_track {
39 	/* configuration we mirror so that we use same code btw kms/ums */
40 	u32			group_size;
41 	u32			nbanks;
42 	u32			npipes;
43 	/* value we track */
44 	u32			sq_config;
45 	u32			log_nsamples;
46 	u32			nsamples;
47 	u32			cb_color_base_last[8];
48 	struct radeon_bo	*cb_color_bo[8];
49 	u64			cb_color_bo_mc[8];
50 	u64			cb_color_bo_offset[8];
51 	struct radeon_bo	*cb_color_frag_bo[8];
52 	u64			cb_color_frag_offset[8];
53 	struct radeon_bo	*cb_color_tile_bo[8];
54 	u64			cb_color_tile_offset[8];
55 	u32			cb_color_mask[8];
56 	u32			cb_color_info[8];
57 	u32			cb_color_view[8];
58 	u32			cb_color_size_idx[8]; /* unused */
59 	u32			cb_target_mask;
60 	u32			cb_shader_mask;  /* unused */
61 	bool			is_resolve;
62 	u32			cb_color_size[8];
63 	u32			vgt_strmout_en;
64 	u32			vgt_strmout_buffer_en;
65 	struct radeon_bo	*vgt_strmout_bo[4];
66 	u64			vgt_strmout_bo_mc[4]; /* unused */
67 	u32			vgt_strmout_bo_offset[4];
68 	u32			vgt_strmout_size[4];
69 	u32			db_depth_control;
70 	u32			db_depth_info;
71 	u32			db_depth_size_idx;
72 	u32			db_depth_view;
73 	u32			db_depth_size;
74 	u32			db_offset;
75 	struct radeon_bo	*db_bo;
76 	u64			db_bo_mc;
77 	bool			sx_misc_kill_all_prims;
78 	bool			cb_dirty;
79 	bool			db_dirty;
80 	bool			streamout_dirty;
81 	struct radeon_bo	*htile_bo;
82 	u64			htile_offset;
83 	u32			htile_surface;
84 };
85 
86 #define FMT_8_BIT(fmt, vc)   [fmt] = { 1, 1, 1, vc, CHIP_R600 }
87 #define FMT_16_BIT(fmt, vc)  [fmt] = { 1, 1, 2, vc, CHIP_R600 }
88 #define FMT_24_BIT(fmt)      [fmt] = { 1, 1, 4,  0, CHIP_R600 }
89 #define FMT_32_BIT(fmt, vc)  [fmt] = { 1, 1, 4, vc, CHIP_R600 }
90 #define FMT_48_BIT(fmt)      [fmt] = { 1, 1, 8,  0, CHIP_R600 }
91 #define FMT_64_BIT(fmt, vc)  [fmt] = { 1, 1, 8, vc, CHIP_R600 }
92 #define FMT_96_BIT(fmt)      [fmt] = { 1, 1, 12, 0, CHIP_R600 }
93 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16,vc, CHIP_R600 }
94 
95 struct gpu_formats {
96 	unsigned blockwidth;
97 	unsigned blockheight;
98 	unsigned blocksize;
99 	unsigned valid_color;
100 	enum radeon_family min_family;
101 };
102 
103 static const struct gpu_formats color_formats_table[] = {
104 	/* 8 bit */
105 	FMT_8_BIT(V_038004_COLOR_8, 1),
106 	FMT_8_BIT(V_038004_COLOR_4_4, 1),
107 	FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
108 	FMT_8_BIT(V_038004_FMT_1, 0),
109 
110 	/* 16-bit */
111 	FMT_16_BIT(V_038004_COLOR_16, 1),
112 	FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
113 	FMT_16_BIT(V_038004_COLOR_8_8, 1),
114 	FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
115 	FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
116 	FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
117 	FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
118 	FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
119 
120 	/* 24-bit */
121 	FMT_24_BIT(V_038004_FMT_8_8_8),
122 
123 	/* 32-bit */
124 	FMT_32_BIT(V_038004_COLOR_32, 1),
125 	FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
126 	FMT_32_BIT(V_038004_COLOR_16_16, 1),
127 	FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
128 	FMT_32_BIT(V_038004_COLOR_8_24, 1),
129 	FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
130 	FMT_32_BIT(V_038004_COLOR_24_8, 1),
131 	FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
132 	FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
133 	FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
134 	FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
135 	FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
136 	FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
137 	FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
138 	FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
139 	FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
140 	FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
141 	FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
142 
143 	/* 48-bit */
144 	FMT_48_BIT(V_038004_FMT_16_16_16),
145 	FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
146 
147 	/* 64-bit */
148 	FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
149 	FMT_64_BIT(V_038004_COLOR_32_32, 1),
150 	FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
151 	FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
152 	FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
153 
154 	FMT_96_BIT(V_038004_FMT_32_32_32),
155 	FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
156 
157 	/* 128-bit */
158 	FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
159 	FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
160 
161 	[V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
162 	[V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
163 
164 	/* block compressed formats */
165 	[V_038004_FMT_BC1] = { 4, 4, 8, 0 },
166 	[V_038004_FMT_BC2] = { 4, 4, 16, 0 },
167 	[V_038004_FMT_BC3] = { 4, 4, 16, 0 },
168 	[V_038004_FMT_BC4] = { 4, 4, 8, 0 },
169 	[V_038004_FMT_BC5] = { 4, 4, 16, 0},
170 	[V_038004_FMT_BC6] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
171 	[V_038004_FMT_BC7] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
172 
173 	/* The other Evergreen formats */
174 	[V_038004_FMT_32_AS_32_32_32_32] = { 1, 1, 4, 0, CHIP_CEDAR},
175 };
176 
177 bool r600_fmt_is_valid_color(u32 format)
178 {
179 	if (format >= ARRAY_SIZE(color_formats_table))
180 		return false;
181 
182 	if (color_formats_table[format].valid_color)
183 		return true;
184 
185 	return false;
186 }
187 
188 bool r600_fmt_is_valid_texture(u32 format, enum radeon_family family)
189 {
190 	if (format >= ARRAY_SIZE(color_formats_table))
191 		return false;
192 
193 	if (family < color_formats_table[format].min_family)
194 		return false;
195 
196 	if (color_formats_table[format].blockwidth > 0)
197 		return true;
198 
199 	return false;
200 }
201 
202 int r600_fmt_get_blocksize(u32 format)
203 {
204 	if (format >= ARRAY_SIZE(color_formats_table))
205 		return 0;
206 
207 	return color_formats_table[format].blocksize;
208 }
209 
210 int r600_fmt_get_nblocksx(u32 format, u32 w)
211 {
212 	unsigned bw;
213 
214 	if (format >= ARRAY_SIZE(color_formats_table))
215 		return 0;
216 
217 	bw = color_formats_table[format].blockwidth;
218 	if (bw == 0)
219 		return 0;
220 
221 	return DIV_ROUND_UP(w, bw);
222 }
223 
224 int r600_fmt_get_nblocksy(u32 format, u32 h)
225 {
226 	unsigned bh;
227 
228 	if (format >= ARRAY_SIZE(color_formats_table))
229 		return 0;
230 
231 	bh = color_formats_table[format].blockheight;
232 	if (bh == 0)
233 		return 0;
234 
235 	return DIV_ROUND_UP(h, bh);
236 }
237 
238 struct array_mode_checker {
239 	int array_mode;
240 	u32 group_size;
241 	u32 nbanks;
242 	u32 npipes;
243 	u32 nsamples;
244 	u32 blocksize;
245 };
246 
247 /* returns alignment in pixels for pitch/height/depth and bytes for base */
248 static int r600_get_array_mode_alignment(struct array_mode_checker *values,
249 						u32 *pitch_align,
250 						u32 *height_align,
251 						u32 *depth_align,
252 						u64 *base_align)
253 {
254 	u32 tile_width = 8;
255 	u32 tile_height = 8;
256 	u32 macro_tile_width = values->nbanks;
257 	u32 macro_tile_height = values->npipes;
258 	u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
259 	u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
260 
261 	switch (values->array_mode) {
262 	case ARRAY_LINEAR_GENERAL:
263 		/* technically tile_width/_height for pitch/height */
264 		*pitch_align = 1; /* tile_width */
265 		*height_align = 1; /* tile_height */
266 		*depth_align = 1;
267 		*base_align = 1;
268 		break;
269 	case ARRAY_LINEAR_ALIGNED:
270 		*pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
271 		*height_align = 1;
272 		*depth_align = 1;
273 		*base_align = values->group_size;
274 		break;
275 	case ARRAY_1D_TILED_THIN1:
276 		*pitch_align = max((u32)tile_width,
277 				   (u32)(values->group_size /
278 					 (tile_height * values->blocksize * values->nsamples)));
279 		*height_align = tile_height;
280 		*depth_align = 1;
281 		*base_align = values->group_size;
282 		break;
283 	case ARRAY_2D_TILED_THIN1:
284 		*pitch_align = max((u32)macro_tile_width * tile_width,
285 				(u32)((values->group_size * values->nbanks) /
286 				(values->blocksize * values->nsamples * tile_width)));
287 		*height_align = macro_tile_height * tile_height;
288 		*depth_align = 1;
289 		*base_align = max(macro_tile_bytes,
290 				  (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
291 		break;
292 	default:
293 		return -EINVAL;
294 	}
295 
296 	return 0;
297 }
298 
299 static void r600_cs_track_init(struct r600_cs_track *track)
300 {
301 	int i;
302 
303 	/* assume DX9 mode */
304 	track->sq_config = DX9_CONSTS;
305 	for (i = 0; i < 8; i++) {
306 		track->cb_color_base_last[i] = 0;
307 		track->cb_color_size[i] = 0;
308 		track->cb_color_size_idx[i] = 0;
309 		track->cb_color_info[i] = 0;
310 		track->cb_color_view[i] = 0xFFFFFFFF;
311 		track->cb_color_bo[i] = NULL;
312 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
313 		track->cb_color_bo_mc[i] = 0xFFFFFFFF;
314 		track->cb_color_frag_bo[i] = NULL;
315 		track->cb_color_frag_offset[i] = 0xFFFFFFFF;
316 		track->cb_color_tile_bo[i] = NULL;
317 		track->cb_color_tile_offset[i] = 0xFFFFFFFF;
318 		track->cb_color_mask[i] = 0xFFFFFFFF;
319 	}
320 	track->is_resolve = false;
321 	track->nsamples = 16;
322 	track->log_nsamples = 4;
323 	track->cb_target_mask = 0xFFFFFFFF;
324 	track->cb_shader_mask = 0xFFFFFFFF;
325 	track->cb_dirty = true;
326 	track->db_bo = NULL;
327 	track->db_bo_mc = 0xFFFFFFFF;
328 	/* assume the biggest format and that htile is enabled */
329 	track->db_depth_info = 7 | (1 << 25);
330 	track->db_depth_view = 0xFFFFC000;
331 	track->db_depth_size = 0xFFFFFFFF;
332 	track->db_depth_size_idx = 0;
333 	track->db_depth_control = 0xFFFFFFFF;
334 	track->db_dirty = true;
335 	track->htile_bo = NULL;
336 	track->htile_offset = 0xFFFFFFFF;
337 	track->htile_surface = 0;
338 
339 	for (i = 0; i < 4; i++) {
340 		track->vgt_strmout_size[i] = 0;
341 		track->vgt_strmout_bo[i] = NULL;
342 		track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
343 		track->vgt_strmout_bo_mc[i] = 0xFFFFFFFF;
344 	}
345 	track->streamout_dirty = true;
346 	track->sx_misc_kill_all_prims = false;
347 }
348 
349 static int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
350 {
351 	struct r600_cs_track *track = p->track;
352 	u32 slice_tile_max, tmp;
353 	u32 height, height_align, pitch, pitch_align, depth_align;
354 	u64 base_offset, base_align;
355 	struct array_mode_checker array_check;
356 	volatile u32 *ib = p->ib.ptr;
357 	unsigned array_mode;
358 	u32 format;
359 	/* When resolve is used, the second colorbuffer has always 1 sample. */
360 	unsigned nsamples = track->is_resolve && i == 1 ? 1 : track->nsamples;
361 
362 	format = G_0280A0_FORMAT(track->cb_color_info[i]);
363 	if (!r600_fmt_is_valid_color(format)) {
364 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
365 			 __func__, __LINE__, format,
366 			i, track->cb_color_info[i]);
367 		return -EINVAL;
368 	}
369 	/* pitch in pixels */
370 	pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
371 	slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
372 	slice_tile_max *= 64;
373 	height = slice_tile_max / pitch;
374 	if (height > 8192)
375 		height = 8192;
376 	array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
377 
378 	base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
379 	array_check.array_mode = array_mode;
380 	array_check.group_size = track->group_size;
381 	array_check.nbanks = track->nbanks;
382 	array_check.npipes = track->npipes;
383 	array_check.nsamples = nsamples;
384 	array_check.blocksize = r600_fmt_get_blocksize(format);
385 	if (r600_get_array_mode_alignment(&array_check,
386 					  &pitch_align, &height_align, &depth_align, &base_align)) {
387 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
388 			 G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
389 			 track->cb_color_info[i]);
390 		return -EINVAL;
391 	}
392 	switch (array_mode) {
393 	case V_0280A0_ARRAY_LINEAR_GENERAL:
394 		break;
395 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
396 		break;
397 	case V_0280A0_ARRAY_1D_TILED_THIN1:
398 		/* avoid breaking userspace */
399 		if (height > 7)
400 			height &= ~0x7;
401 		break;
402 	case V_0280A0_ARRAY_2D_TILED_THIN1:
403 		break;
404 	default:
405 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
406 			G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
407 			track->cb_color_info[i]);
408 		return -EINVAL;
409 	}
410 
411 	if (!IS_ALIGNED(pitch, pitch_align)) {
412 		dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
413 			 __func__, __LINE__, pitch, pitch_align, array_mode);
414 		return -EINVAL;
415 	}
416 	if (!IS_ALIGNED(height, height_align)) {
417 		dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
418 			 __func__, __LINE__, height, height_align, array_mode);
419 		return -EINVAL;
420 	}
421 	if (!IS_ALIGNED(base_offset, base_align)) {
422 		dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
423 			 base_offset, base_align, array_mode);
424 		return -EINVAL;
425 	}
426 
427 	/* check offset */
428 	tmp = r600_fmt_get_nblocksy(format, height) * r600_fmt_get_nblocksx(format, pitch) *
429 	      r600_fmt_get_blocksize(format) * nsamples;
430 	switch (array_mode) {
431 	default:
432 	case V_0280A0_ARRAY_LINEAR_GENERAL:
433 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
434 		tmp += track->cb_color_view[i] & 0xFF;
435 		break;
436 	case V_0280A0_ARRAY_1D_TILED_THIN1:
437 	case V_0280A0_ARRAY_2D_TILED_THIN1:
438 		tmp += G_028080_SLICE_MAX(track->cb_color_view[i]) * tmp;
439 		break;
440 	}
441 	if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
442 		if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
443 			/* the initial DDX does bad things with the CB size occasionally */
444 			/* it rounds up height too far for slice tile max but the BO is smaller */
445 			/* r600c,g also seem to flush at bad times in some apps resulting in
446 			 * bogus values here. So for linear just allow anything to avoid breaking
447 			 * broken userspace.
448 			 */
449 		} else {
450 			dev_warn(p->dev, "%s offset[%d] %d %llu %d %lu too big (%d %d) (%d %d %d)\n",
451 				 __func__, i, array_mode,
452 				 track->cb_color_bo_offset[i], tmp,
453 				 radeon_bo_size(track->cb_color_bo[i]),
454 				 pitch, height, r600_fmt_get_nblocksx(format, pitch),
455 				 r600_fmt_get_nblocksy(format, height),
456 				 r600_fmt_get_blocksize(format));
457 			return -EINVAL;
458 		}
459 	}
460 	/* limit max tile */
461 	tmp = (height * pitch) >> 6;
462 	if (tmp < slice_tile_max)
463 		slice_tile_max = tmp;
464 	tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
465 		S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
466 	ib[track->cb_color_size_idx[i]] = tmp;
467 
468 	/* FMASK/CMASK */
469 	switch (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
470 	case V_0280A0_TILE_DISABLE:
471 		break;
472 	case V_0280A0_FRAG_ENABLE:
473 		if (track->nsamples > 1) {
474 			uint32_t tile_max = G_028100_FMASK_TILE_MAX(track->cb_color_mask[i]);
475 			/* the tile size is 8x8, but the size is in units of bits.
476 			 * for bytes, do just * 8. */
477 			uint32_t bytes = track->nsamples * track->log_nsamples * 8 * (tile_max + 1);
478 
479 			if (bytes + track->cb_color_frag_offset[i] >
480 			    radeon_bo_size(track->cb_color_frag_bo[i])) {
481 				dev_warn(p->dev, "%s FMASK_TILE_MAX too large "
482 					 "(tile_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
483 					 __func__, tile_max, bytes,
484 					 track->cb_color_frag_offset[i],
485 					 radeon_bo_size(track->cb_color_frag_bo[i]));
486 				return -EINVAL;
487 			}
488 		}
489 		fallthrough;
490 	case V_0280A0_CLEAR_ENABLE:
491 	{
492 		uint32_t block_max = G_028100_CMASK_BLOCK_MAX(track->cb_color_mask[i]);
493 		/* One block = 128x128 pixels, one 8x8 tile has 4 bits..
494 		 * (128*128) / (8*8) / 2 = 128 bytes per block. */
495 		uint32_t bytes = (block_max + 1) * 128;
496 
497 		if (bytes + track->cb_color_tile_offset[i] >
498 		    radeon_bo_size(track->cb_color_tile_bo[i])) {
499 			dev_warn(p->dev, "%s CMASK_BLOCK_MAX too large "
500 				 "(block_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
501 				 __func__, block_max, bytes,
502 				 track->cb_color_tile_offset[i],
503 				 radeon_bo_size(track->cb_color_tile_bo[i]));
504 			return -EINVAL;
505 		}
506 		break;
507 	}
508 	default:
509 		dev_warn(p->dev, "%s invalid tile mode\n", __func__);
510 		return -EINVAL;
511 	}
512 	return 0;
513 }
514 
515 static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
516 {
517 	struct r600_cs_track *track = p->track;
518 	u32 nviews, bpe, ntiles, slice_tile_max, tmp;
519 	u32 height_align, pitch_align, depth_align;
520 	u32 pitch = 8192;
521 	u32 height = 8192;
522 	u64 base_offset, base_align;
523 	struct array_mode_checker array_check;
524 	int array_mode;
525 	volatile u32 *ib = p->ib.ptr;
526 
527 
528 	if (track->db_bo == NULL) {
529 		dev_warn(p->dev, "z/stencil with no depth buffer\n");
530 		return -EINVAL;
531 	}
532 	switch (G_028010_FORMAT(track->db_depth_info)) {
533 	case V_028010_DEPTH_16:
534 		bpe = 2;
535 		break;
536 	case V_028010_DEPTH_X8_24:
537 	case V_028010_DEPTH_8_24:
538 	case V_028010_DEPTH_X8_24_FLOAT:
539 	case V_028010_DEPTH_8_24_FLOAT:
540 	case V_028010_DEPTH_32_FLOAT:
541 		bpe = 4;
542 		break;
543 	case V_028010_DEPTH_X24_8_32_FLOAT:
544 		bpe = 8;
545 		break;
546 	default:
547 		dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
548 		return -EINVAL;
549 	}
550 	if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
551 		if (!track->db_depth_size_idx) {
552 			dev_warn(p->dev, "z/stencil buffer size not set\n");
553 			return -EINVAL;
554 		}
555 		tmp = radeon_bo_size(track->db_bo) - track->db_offset;
556 		tmp = (tmp / bpe) >> 6;
557 		if (!tmp) {
558 			dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
559 					track->db_depth_size, bpe, track->db_offset,
560 					radeon_bo_size(track->db_bo));
561 			return -EINVAL;
562 		}
563 		ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
564 	} else {
565 		/* pitch in pixels */
566 		pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
567 		slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
568 		slice_tile_max *= 64;
569 		height = slice_tile_max / pitch;
570 		if (height > 8192)
571 			height = 8192;
572 		base_offset = track->db_bo_mc + track->db_offset;
573 		array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
574 		array_check.array_mode = array_mode;
575 		array_check.group_size = track->group_size;
576 		array_check.nbanks = track->nbanks;
577 		array_check.npipes = track->npipes;
578 		array_check.nsamples = track->nsamples;
579 		array_check.blocksize = bpe;
580 		if (r600_get_array_mode_alignment(&array_check,
581 					&pitch_align, &height_align, &depth_align, &base_align)) {
582 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
583 					G_028010_ARRAY_MODE(track->db_depth_info),
584 					track->db_depth_info);
585 			return -EINVAL;
586 		}
587 		switch (array_mode) {
588 		case V_028010_ARRAY_1D_TILED_THIN1:
589 			/* don't break userspace */
590 			height &= ~0x7;
591 			break;
592 		case V_028010_ARRAY_2D_TILED_THIN1:
593 			break;
594 		default:
595 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
596 					G_028010_ARRAY_MODE(track->db_depth_info),
597 					track->db_depth_info);
598 			return -EINVAL;
599 		}
600 
601 		if (!IS_ALIGNED(pitch, pitch_align)) {
602 			dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
603 					__func__, __LINE__, pitch, pitch_align, array_mode);
604 			return -EINVAL;
605 		}
606 		if (!IS_ALIGNED(height, height_align)) {
607 			dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
608 					__func__, __LINE__, height, height_align, array_mode);
609 			return -EINVAL;
610 		}
611 		if (!IS_ALIGNED(base_offset, base_align)) {
612 			dev_warn(p->dev, "%s offset 0x%llx, 0x%llx, %d not aligned\n", __func__,
613 					base_offset, base_align, array_mode);
614 			return -EINVAL;
615 		}
616 
617 		ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
618 		nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
619 		tmp = ntiles * bpe * 64 * nviews * track->nsamples;
620 		if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
621 			dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
622 					array_mode,
623 					track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
624 					radeon_bo_size(track->db_bo));
625 			return -EINVAL;
626 		}
627 	}
628 
629 	/* hyperz */
630 	if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
631 		unsigned long size;
632 		unsigned nbx, nby;
633 
634 		if (track->htile_bo == NULL) {
635 			dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
636 				 __func__, __LINE__, track->db_depth_info);
637 			return -EINVAL;
638 		}
639 		if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
640 			dev_warn(p->dev, "%s:%d htile can't be enabled with bogus db_depth_size 0x%08x\n",
641 				 __func__, __LINE__, track->db_depth_size);
642 			return -EINVAL;
643 		}
644 
645 		nbx = pitch;
646 		nby = height;
647 		if (G_028D24_LINEAR(track->htile_surface)) {
648 			/* nbx must be 16 htiles aligned == 16 * 8 pixel aligned */
649 			nbx = round_up(nbx, 16 * 8);
650 			/* nby is npipes htiles aligned == npipes * 8 pixel aligned */
651 			nby = round_up(nby, track->npipes * 8);
652 		} else {
653 			/* always assume 8x8 htile */
654 			/* align is htile align * 8, htile align vary according to
655 			 * number of pipe and tile width and nby
656 			 */
657 			switch (track->npipes) {
658 			case 8:
659 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
660 				nbx = round_up(nbx, 64 * 8);
661 				nby = round_up(nby, 64 * 8);
662 				break;
663 			case 4:
664 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
665 				nbx = round_up(nbx, 64 * 8);
666 				nby = round_up(nby, 32 * 8);
667 				break;
668 			case 2:
669 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
670 				nbx = round_up(nbx, 32 * 8);
671 				nby = round_up(nby, 32 * 8);
672 				break;
673 			case 1:
674 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
675 				nbx = round_up(nbx, 32 * 8);
676 				nby = round_up(nby, 16 * 8);
677 				break;
678 			default:
679 				dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
680 					 __func__, __LINE__, track->npipes);
681 				return -EINVAL;
682 			}
683 		}
684 		/* compute number of htile */
685 		nbx = nbx >> 3;
686 		nby = nby >> 3;
687 		/* size must be aligned on npipes * 2K boundary */
688 		size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
689 		size += track->htile_offset;
690 
691 		if (size > radeon_bo_size(track->htile_bo)) {
692 			dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
693 				 __func__, __LINE__, radeon_bo_size(track->htile_bo),
694 				 size, nbx, nby);
695 			return -EINVAL;
696 		}
697 	}
698 
699 	track->db_dirty = false;
700 	return 0;
701 }
702 
703 static int r600_cs_track_check(struct radeon_cs_parser *p)
704 {
705 	struct r600_cs_track *track = p->track;
706 	u32 tmp;
707 	int r, i;
708 
709 	/* on legacy kernel we don't perform advanced check */
710 	if (p->rdev == NULL)
711 		return 0;
712 
713 	/* check streamout */
714 	if (track->streamout_dirty && track->vgt_strmout_en) {
715 		for (i = 0; i < 4; i++) {
716 			if (track->vgt_strmout_buffer_en & (1 << i)) {
717 				if (track->vgt_strmout_bo[i]) {
718 					u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
719 						(u64)track->vgt_strmout_size[i];
720 					if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
721 						DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
722 							  i, offset,
723 							  radeon_bo_size(track->vgt_strmout_bo[i]));
724 						return -EINVAL;
725 					}
726 				} else {
727 					dev_warn(p->dev, "No buffer for streamout %d\n", i);
728 					return -EINVAL;
729 				}
730 			}
731 		}
732 		track->streamout_dirty = false;
733 	}
734 
735 	if (track->sx_misc_kill_all_prims)
736 		return 0;
737 
738 	/* check that we have a cb for each enabled target, we don't check
739 	 * shader_mask because it seems mesa isn't always setting it :(
740 	 */
741 	if (track->cb_dirty) {
742 		tmp = track->cb_target_mask;
743 
744 		/* We must check both colorbuffers for RESOLVE. */
745 		if (track->is_resolve) {
746 			tmp |= 0xff;
747 		}
748 
749 		for (i = 0; i < 8; i++) {
750 			u32 format = G_0280A0_FORMAT(track->cb_color_info[i]);
751 
752 			if (format != V_0280A0_COLOR_INVALID &&
753 			    (tmp >> (i * 4)) & 0xF) {
754 				/* at least one component is enabled */
755 				if (track->cb_color_bo[i] == NULL) {
756 					dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
757 						__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
758 					return -EINVAL;
759 				}
760 				/* perform rewrite of CB_COLOR[0-7]_SIZE */
761 				r = r600_cs_track_validate_cb(p, i);
762 				if (r)
763 					return r;
764 			}
765 		}
766 		track->cb_dirty = false;
767 	}
768 
769 	/* Check depth buffer */
770 	if (track->db_dirty &&
771 	    G_028010_FORMAT(track->db_depth_info) != V_028010_DEPTH_INVALID &&
772 	    (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
773 	     G_028800_Z_ENABLE(track->db_depth_control))) {
774 		r = r600_cs_track_validate_db(p);
775 		if (r)
776 			return r;
777 	}
778 
779 	return 0;
780 }
781 
782 /**
783  * r600_cs_packet_parse_vline() - parse userspace VLINE packet
784  * @p:		parser structure holding parsing context.
785  *
786  * This is an R600-specific function for parsing VLINE packets.
787  * Real work is done by r600_cs_common_vline_parse function.
788  * Here we just set up ASIC-specific register table and call
789  * the common implementation function.
790  */
791 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
792 {
793 	static uint32_t vline_start_end[2] = {AVIVO_D1MODE_VLINE_START_END,
794 					      AVIVO_D2MODE_VLINE_START_END};
795 	static uint32_t vline_status[2] = {AVIVO_D1MODE_VLINE_STATUS,
796 					   AVIVO_D2MODE_VLINE_STATUS};
797 
798 	return r600_cs_common_vline_parse(p, vline_start_end, vline_status);
799 }
800 
801 /**
802  * r600_cs_common_vline_parse() - common vline parser
803  * @p:			parser structure holding parsing context.
804  * @vline_start_end:    table of vline_start_end registers
805  * @vline_status:       table of vline_status registers
806  *
807  * Userspace sends a special sequence for VLINE waits.
808  * PACKET0 - VLINE_START_END + value
809  * PACKET3 - WAIT_REG_MEM poll vline status reg
810  * RELOC (P3) - crtc_id in reloc.
811  *
812  * This function parses this and relocates the VLINE START END
813  * and WAIT_REG_MEM packets to the correct crtc.
814  * It also detects a switched off crtc and nulls out the
815  * wait in that case. This function is common for all ASICs that
816  * are R600 and newer. The parsing algorithm is the same, and only
817  * differs in which registers are used.
818  *
819  * Caller is the ASIC-specific function which passes the parser
820  * context and ASIC-specific register table
821  */
822 int r600_cs_common_vline_parse(struct radeon_cs_parser *p,
823 			       uint32_t *vline_start_end,
824 			       uint32_t *vline_status)
825 {
826 	struct drm_crtc *crtc;
827 	struct radeon_crtc *radeon_crtc;
828 	struct radeon_cs_packet p3reloc, wait_reg_mem;
829 	int crtc_id;
830 	int r;
831 	uint32_t header, h_idx, reg, wait_reg_mem_info;
832 	volatile uint32_t *ib;
833 
834 	ib = p->ib.ptr;
835 
836 	/* parse the WAIT_REG_MEM */
837 	r = radeon_cs_packet_parse(p, &wait_reg_mem, p->idx);
838 	if (r)
839 		return r;
840 
841 	/* check its a WAIT_REG_MEM */
842 	if (wait_reg_mem.type != RADEON_PACKET_TYPE3 ||
843 	    wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
844 		DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
845 		return -EINVAL;
846 	}
847 
848 	wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
849 	/* bit 4 is reg (0) or mem (1) */
850 	if (wait_reg_mem_info & 0x10) {
851 		DRM_ERROR("vline WAIT_REG_MEM waiting on MEM instead of REG\n");
852 		return -EINVAL;
853 	}
854 	/* bit 8 is me (0) or pfp (1) */
855 	if (wait_reg_mem_info & 0x100) {
856 		DRM_ERROR("vline WAIT_REG_MEM waiting on PFP instead of ME\n");
857 		return -EINVAL;
858 	}
859 	/* waiting for value to be equal */
860 	if ((wait_reg_mem_info & 0x7) != 0x3) {
861 		DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
862 		return -EINVAL;
863 	}
864 	if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != vline_status[0]) {
865 		DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
866 		return -EINVAL;
867 	}
868 
869 	if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != RADEON_VLINE_STAT) {
870 		DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
871 		return -EINVAL;
872 	}
873 
874 	/* jump over the NOP */
875 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
876 	if (r)
877 		return r;
878 
879 	h_idx = p->idx - 2;
880 	p->idx += wait_reg_mem.count + 2;
881 	p->idx += p3reloc.count + 2;
882 
883 	header = radeon_get_ib_value(p, h_idx);
884 	crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
885 	reg = R600_CP_PACKET0_GET_REG(header);
886 
887 	crtc = drm_crtc_find(rdev_to_drm(p->rdev), p->filp, crtc_id);
888 	if (!crtc) {
889 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
890 		return -ENOENT;
891 	}
892 	radeon_crtc = to_radeon_crtc(crtc);
893 	crtc_id = radeon_crtc->crtc_id;
894 
895 	if (!crtc->enabled) {
896 		/* CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
897 		ib[h_idx + 2] = PACKET2(0);
898 		ib[h_idx + 3] = PACKET2(0);
899 		ib[h_idx + 4] = PACKET2(0);
900 		ib[h_idx + 5] = PACKET2(0);
901 		ib[h_idx + 6] = PACKET2(0);
902 		ib[h_idx + 7] = PACKET2(0);
903 		ib[h_idx + 8] = PACKET2(0);
904 	} else if (reg == vline_start_end[0]) {
905 		header &= ~R600_CP_PACKET0_REG_MASK;
906 		header |= vline_start_end[crtc_id] >> 2;
907 		ib[h_idx] = header;
908 		ib[h_idx + 4] = vline_status[crtc_id] >> 2;
909 	} else {
910 		DRM_ERROR("unknown crtc reloc\n");
911 		return -EINVAL;
912 	}
913 	return 0;
914 }
915 
916 static int r600_packet0_check(struct radeon_cs_parser *p,
917 				struct radeon_cs_packet *pkt,
918 				unsigned idx, unsigned reg)
919 {
920 	int r;
921 
922 	switch (reg) {
923 	case AVIVO_D1MODE_VLINE_START_END:
924 		r = r600_cs_packet_parse_vline(p);
925 		if (r) {
926 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
927 					idx, reg);
928 			return r;
929 		}
930 		break;
931 	default:
932 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
933 		return -EINVAL;
934 	}
935 	return 0;
936 }
937 
938 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
939 				struct radeon_cs_packet *pkt)
940 {
941 	unsigned reg, i;
942 	unsigned idx;
943 	int r;
944 
945 	idx = pkt->idx + 1;
946 	reg = pkt->reg;
947 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
948 		r = r600_packet0_check(p, pkt, idx, reg);
949 		if (r) {
950 			return r;
951 		}
952 	}
953 	return 0;
954 }
955 
956 /**
957  * r600_cs_check_reg() - check if register is authorized or not
958  * @p: parser structure holding parsing context
959  * @reg: register we are testing
960  * @idx: index into the cs buffer
961  *
962  * This function will test against r600_reg_safe_bm and return 0
963  * if register is safe. If register is not flag as safe this function
964  * will test it against a list of register needing special handling.
965  */
966 static int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
967 {
968 	struct r600_cs_track *track = (struct r600_cs_track *)p->track;
969 	struct radeon_bo_list *reloc;
970 	u32 m, i, tmp, *ib;
971 	int r;
972 
973 	i = (reg >> 7);
974 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
975 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
976 		return -EINVAL;
977 	}
978 	m = 1 << ((reg >> 2) & 31);
979 	if (!(r600_reg_safe_bm[i] & m))
980 		return 0;
981 	ib = p->ib.ptr;
982 	switch (reg) {
983 	/* force following reg to 0 in an attempt to disable out buffer
984 	 * which will need us to better understand how it works to perform
985 	 * security check on it (Jerome)
986 	 */
987 	case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
988 	case R_008C44_SQ_ESGS_RING_SIZE:
989 	case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
990 	case R_008C54_SQ_ESTMP_RING_SIZE:
991 	case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
992 	case R_008C74_SQ_FBUF_RING_SIZE:
993 	case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
994 	case R_008C5C_SQ_GSTMP_RING_SIZE:
995 	case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
996 	case R_008C4C_SQ_GSVS_RING_SIZE:
997 	case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
998 	case R_008C6C_SQ_PSTMP_RING_SIZE:
999 	case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
1000 	case R_008C7C_SQ_REDUC_RING_SIZE:
1001 	case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
1002 	case R_008C64_SQ_VSTMP_RING_SIZE:
1003 	case R_0288C8_SQ_GS_VERT_ITEMSIZE:
1004 		/* get value to populate the IB don't remove */
1005 		/*tmp =radeon_get_ib_value(p, idx);
1006 		  ib[idx] = 0;*/
1007 		break;
1008 	case SQ_ESGS_RING_BASE:
1009 	case SQ_GSVS_RING_BASE:
1010 	case SQ_ESTMP_RING_BASE:
1011 	case SQ_GSTMP_RING_BASE:
1012 	case SQ_PSTMP_RING_BASE:
1013 	case SQ_VSTMP_RING_BASE:
1014 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1015 		if (r) {
1016 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1017 					"0x%04X\n", reg);
1018 			return -EINVAL;
1019 		}
1020 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1021 		break;
1022 	case SQ_CONFIG:
1023 		track->sq_config = radeon_get_ib_value(p, idx);
1024 		break;
1025 	case R_028800_DB_DEPTH_CONTROL:
1026 		track->db_depth_control = radeon_get_ib_value(p, idx);
1027 		track->db_dirty = true;
1028 		break;
1029 	case R_028010_DB_DEPTH_INFO:
1030 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1031 		    radeon_cs_packet_next_is_pkt3_nop(p)) {
1032 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1033 			if (r) {
1034 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1035 					 "0x%04X\n", reg);
1036 				return -EINVAL;
1037 			}
1038 			track->db_depth_info = radeon_get_ib_value(p, idx);
1039 			ib[idx] &= C_028010_ARRAY_MODE;
1040 			track->db_depth_info &= C_028010_ARRAY_MODE;
1041 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1042 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1043 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1044 			} else {
1045 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1046 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1047 			}
1048 		} else {
1049 			track->db_depth_info = radeon_get_ib_value(p, idx);
1050 		}
1051 		track->db_dirty = true;
1052 		break;
1053 	case R_028004_DB_DEPTH_VIEW:
1054 		track->db_depth_view = radeon_get_ib_value(p, idx);
1055 		track->db_dirty = true;
1056 		break;
1057 	case R_028000_DB_DEPTH_SIZE:
1058 		track->db_depth_size = radeon_get_ib_value(p, idx);
1059 		track->db_depth_size_idx = idx;
1060 		track->db_dirty = true;
1061 		break;
1062 	case R_028AB0_VGT_STRMOUT_EN:
1063 		track->vgt_strmout_en = radeon_get_ib_value(p, idx);
1064 		track->streamout_dirty = true;
1065 		break;
1066 	case R_028B20_VGT_STRMOUT_BUFFER_EN:
1067 		track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
1068 		track->streamout_dirty = true;
1069 		break;
1070 	case VGT_STRMOUT_BUFFER_BASE_0:
1071 	case VGT_STRMOUT_BUFFER_BASE_1:
1072 	case VGT_STRMOUT_BUFFER_BASE_2:
1073 	case VGT_STRMOUT_BUFFER_BASE_3:
1074 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1075 		if (r) {
1076 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1077 					"0x%04X\n", reg);
1078 			return -EINVAL;
1079 		}
1080 		tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1081 		track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1082 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1083 		track->vgt_strmout_bo[tmp] = reloc->robj;
1084 		track->vgt_strmout_bo_mc[tmp] = reloc->gpu_offset;
1085 		track->streamout_dirty = true;
1086 		break;
1087 	case VGT_STRMOUT_BUFFER_SIZE_0:
1088 	case VGT_STRMOUT_BUFFER_SIZE_1:
1089 	case VGT_STRMOUT_BUFFER_SIZE_2:
1090 	case VGT_STRMOUT_BUFFER_SIZE_3:
1091 		tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1092 		/* size in register is DWs, convert to bytes */
1093 		track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1094 		track->streamout_dirty = true;
1095 		break;
1096 	case CP_COHER_BASE:
1097 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1098 		if (r) {
1099 			dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1100 					"0x%04X\n", reg);
1101 			return -EINVAL;
1102 		}
1103 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1104 		break;
1105 	case R_028238_CB_TARGET_MASK:
1106 		track->cb_target_mask = radeon_get_ib_value(p, idx);
1107 		track->cb_dirty = true;
1108 		break;
1109 	case R_02823C_CB_SHADER_MASK:
1110 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
1111 		break;
1112 	case R_028C04_PA_SC_AA_CONFIG:
1113 		tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
1114 		track->log_nsamples = tmp;
1115 		track->nsamples = 1 << tmp;
1116 		track->cb_dirty = true;
1117 		break;
1118 	case R_028808_CB_COLOR_CONTROL:
1119 		tmp = G_028808_SPECIAL_OP(radeon_get_ib_value(p, idx));
1120 		track->is_resolve = tmp == V_028808_SPECIAL_RESOLVE_BOX;
1121 		track->cb_dirty = true;
1122 		break;
1123 	case R_0280A0_CB_COLOR0_INFO:
1124 	case R_0280A4_CB_COLOR1_INFO:
1125 	case R_0280A8_CB_COLOR2_INFO:
1126 	case R_0280AC_CB_COLOR3_INFO:
1127 	case R_0280B0_CB_COLOR4_INFO:
1128 	case R_0280B4_CB_COLOR5_INFO:
1129 	case R_0280B8_CB_COLOR6_INFO:
1130 	case R_0280BC_CB_COLOR7_INFO:
1131 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1132 		     radeon_cs_packet_next_is_pkt3_nop(p)) {
1133 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1134 			if (r) {
1135 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1136 				return -EINVAL;
1137 			}
1138 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1139 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1140 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1141 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1142 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1143 			} else if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1144 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1145 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1146 			}
1147 		} else {
1148 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1149 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1150 		}
1151 		track->cb_dirty = true;
1152 		break;
1153 	case R_028080_CB_COLOR0_VIEW:
1154 	case R_028084_CB_COLOR1_VIEW:
1155 	case R_028088_CB_COLOR2_VIEW:
1156 	case R_02808C_CB_COLOR3_VIEW:
1157 	case R_028090_CB_COLOR4_VIEW:
1158 	case R_028094_CB_COLOR5_VIEW:
1159 	case R_028098_CB_COLOR6_VIEW:
1160 	case R_02809C_CB_COLOR7_VIEW:
1161 		tmp = (reg - R_028080_CB_COLOR0_VIEW) / 4;
1162 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1163 		track->cb_dirty = true;
1164 		break;
1165 	case R_028060_CB_COLOR0_SIZE:
1166 	case R_028064_CB_COLOR1_SIZE:
1167 	case R_028068_CB_COLOR2_SIZE:
1168 	case R_02806C_CB_COLOR3_SIZE:
1169 	case R_028070_CB_COLOR4_SIZE:
1170 	case R_028074_CB_COLOR5_SIZE:
1171 	case R_028078_CB_COLOR6_SIZE:
1172 	case R_02807C_CB_COLOR7_SIZE:
1173 		tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1174 		track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1175 		track->cb_color_size_idx[tmp] = idx;
1176 		track->cb_dirty = true;
1177 		break;
1178 		/* This register were added late, there is userspace
1179 		 * which does provide relocation for those but set
1180 		 * 0 offset. In order to avoid breaking old userspace
1181 		 * we detect this and set address to point to last
1182 		 * CB_COLOR0_BASE, note that if userspace doesn't set
1183 		 * CB_COLOR0_BASE before this register we will report
1184 		 * error. Old userspace always set CB_COLOR0_BASE
1185 		 * before any of this.
1186 		 */
1187 	case R_0280E0_CB_COLOR0_FRAG:
1188 	case R_0280E4_CB_COLOR1_FRAG:
1189 	case R_0280E8_CB_COLOR2_FRAG:
1190 	case R_0280EC_CB_COLOR3_FRAG:
1191 	case R_0280F0_CB_COLOR4_FRAG:
1192 	case R_0280F4_CB_COLOR5_FRAG:
1193 	case R_0280F8_CB_COLOR6_FRAG:
1194 	case R_0280FC_CB_COLOR7_FRAG:
1195 		tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1196 		if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1197 			if (!track->cb_color_base_last[tmp]) {
1198 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1199 				return -EINVAL;
1200 			}
1201 			track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1202 			track->cb_color_frag_offset[tmp] = track->cb_color_bo_offset[tmp];
1203 			ib[idx] = track->cb_color_base_last[tmp];
1204 		} else {
1205 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1206 			if (r) {
1207 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1208 				return -EINVAL;
1209 			}
1210 			track->cb_color_frag_bo[tmp] = reloc->robj;
1211 			track->cb_color_frag_offset[tmp] = (u64)ib[idx] << 8;
1212 			ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1213 		}
1214 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1215 			track->cb_dirty = true;
1216 		}
1217 		break;
1218 	case R_0280C0_CB_COLOR0_TILE:
1219 	case R_0280C4_CB_COLOR1_TILE:
1220 	case R_0280C8_CB_COLOR2_TILE:
1221 	case R_0280CC_CB_COLOR3_TILE:
1222 	case R_0280D0_CB_COLOR4_TILE:
1223 	case R_0280D4_CB_COLOR5_TILE:
1224 	case R_0280D8_CB_COLOR6_TILE:
1225 	case R_0280DC_CB_COLOR7_TILE:
1226 		tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1227 		if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1228 			if (!track->cb_color_base_last[tmp]) {
1229 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1230 				return -EINVAL;
1231 			}
1232 			track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1233 			track->cb_color_tile_offset[tmp] = track->cb_color_bo_offset[tmp];
1234 			ib[idx] = track->cb_color_base_last[tmp];
1235 		} else {
1236 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1237 			if (r) {
1238 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1239 				return -EINVAL;
1240 			}
1241 			track->cb_color_tile_bo[tmp] = reloc->robj;
1242 			track->cb_color_tile_offset[tmp] = (u64)ib[idx] << 8;
1243 			ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1244 		}
1245 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1246 			track->cb_dirty = true;
1247 		}
1248 		break;
1249 	case R_028100_CB_COLOR0_MASK:
1250 	case R_028104_CB_COLOR1_MASK:
1251 	case R_028108_CB_COLOR2_MASK:
1252 	case R_02810C_CB_COLOR3_MASK:
1253 	case R_028110_CB_COLOR4_MASK:
1254 	case R_028114_CB_COLOR5_MASK:
1255 	case R_028118_CB_COLOR6_MASK:
1256 	case R_02811C_CB_COLOR7_MASK:
1257 		tmp = (reg - R_028100_CB_COLOR0_MASK) / 4;
1258 		track->cb_color_mask[tmp] = radeon_get_ib_value(p, idx);
1259 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1260 			track->cb_dirty = true;
1261 		}
1262 		break;
1263 	case CB_COLOR0_BASE:
1264 	case CB_COLOR1_BASE:
1265 	case CB_COLOR2_BASE:
1266 	case CB_COLOR3_BASE:
1267 	case CB_COLOR4_BASE:
1268 	case CB_COLOR5_BASE:
1269 	case CB_COLOR6_BASE:
1270 	case CB_COLOR7_BASE:
1271 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1272 		if (r) {
1273 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1274 					"0x%04X\n", reg);
1275 			return -EINVAL;
1276 		}
1277 		tmp = (reg - CB_COLOR0_BASE) / 4;
1278 		track->cb_color_bo_offset[tmp] = (u64)radeon_get_ib_value(p, idx) << 8;
1279 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1280 		track->cb_color_base_last[tmp] = ib[idx];
1281 		track->cb_color_bo[tmp] = reloc->robj;
1282 		track->cb_color_bo_mc[tmp] = reloc->gpu_offset;
1283 		track->cb_dirty = true;
1284 		break;
1285 	case DB_DEPTH_BASE:
1286 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1287 		if (r) {
1288 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1289 					"0x%04X\n", reg);
1290 			return -EINVAL;
1291 		}
1292 		track->db_offset = radeon_get_ib_value(p, idx) << 8;
1293 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1294 		track->db_bo = reloc->robj;
1295 		track->db_bo_mc = reloc->gpu_offset;
1296 		track->db_dirty = true;
1297 		break;
1298 	case DB_HTILE_DATA_BASE:
1299 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1300 		if (r) {
1301 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1302 					"0x%04X\n", reg);
1303 			return -EINVAL;
1304 		}
1305 		track->htile_offset = (u64)radeon_get_ib_value(p, idx) << 8;
1306 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1307 		track->htile_bo = reloc->robj;
1308 		track->db_dirty = true;
1309 		break;
1310 	case DB_HTILE_SURFACE:
1311 		track->htile_surface = radeon_get_ib_value(p, idx);
1312 		/* force 8x8 htile width and height */
1313 		ib[idx] |= 3;
1314 		track->db_dirty = true;
1315 		break;
1316 	case SQ_PGM_START_FS:
1317 	case SQ_PGM_START_ES:
1318 	case SQ_PGM_START_VS:
1319 	case SQ_PGM_START_GS:
1320 	case SQ_PGM_START_PS:
1321 	case SQ_ALU_CONST_CACHE_GS_0:
1322 	case SQ_ALU_CONST_CACHE_GS_1:
1323 	case SQ_ALU_CONST_CACHE_GS_2:
1324 	case SQ_ALU_CONST_CACHE_GS_3:
1325 	case SQ_ALU_CONST_CACHE_GS_4:
1326 	case SQ_ALU_CONST_CACHE_GS_5:
1327 	case SQ_ALU_CONST_CACHE_GS_6:
1328 	case SQ_ALU_CONST_CACHE_GS_7:
1329 	case SQ_ALU_CONST_CACHE_GS_8:
1330 	case SQ_ALU_CONST_CACHE_GS_9:
1331 	case SQ_ALU_CONST_CACHE_GS_10:
1332 	case SQ_ALU_CONST_CACHE_GS_11:
1333 	case SQ_ALU_CONST_CACHE_GS_12:
1334 	case SQ_ALU_CONST_CACHE_GS_13:
1335 	case SQ_ALU_CONST_CACHE_GS_14:
1336 	case SQ_ALU_CONST_CACHE_GS_15:
1337 	case SQ_ALU_CONST_CACHE_PS_0:
1338 	case SQ_ALU_CONST_CACHE_PS_1:
1339 	case SQ_ALU_CONST_CACHE_PS_2:
1340 	case SQ_ALU_CONST_CACHE_PS_3:
1341 	case SQ_ALU_CONST_CACHE_PS_4:
1342 	case SQ_ALU_CONST_CACHE_PS_5:
1343 	case SQ_ALU_CONST_CACHE_PS_6:
1344 	case SQ_ALU_CONST_CACHE_PS_7:
1345 	case SQ_ALU_CONST_CACHE_PS_8:
1346 	case SQ_ALU_CONST_CACHE_PS_9:
1347 	case SQ_ALU_CONST_CACHE_PS_10:
1348 	case SQ_ALU_CONST_CACHE_PS_11:
1349 	case SQ_ALU_CONST_CACHE_PS_12:
1350 	case SQ_ALU_CONST_CACHE_PS_13:
1351 	case SQ_ALU_CONST_CACHE_PS_14:
1352 	case SQ_ALU_CONST_CACHE_PS_15:
1353 	case SQ_ALU_CONST_CACHE_VS_0:
1354 	case SQ_ALU_CONST_CACHE_VS_1:
1355 	case SQ_ALU_CONST_CACHE_VS_2:
1356 	case SQ_ALU_CONST_CACHE_VS_3:
1357 	case SQ_ALU_CONST_CACHE_VS_4:
1358 	case SQ_ALU_CONST_CACHE_VS_5:
1359 	case SQ_ALU_CONST_CACHE_VS_6:
1360 	case SQ_ALU_CONST_CACHE_VS_7:
1361 	case SQ_ALU_CONST_CACHE_VS_8:
1362 	case SQ_ALU_CONST_CACHE_VS_9:
1363 	case SQ_ALU_CONST_CACHE_VS_10:
1364 	case SQ_ALU_CONST_CACHE_VS_11:
1365 	case SQ_ALU_CONST_CACHE_VS_12:
1366 	case SQ_ALU_CONST_CACHE_VS_13:
1367 	case SQ_ALU_CONST_CACHE_VS_14:
1368 	case SQ_ALU_CONST_CACHE_VS_15:
1369 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1370 		if (r) {
1371 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1372 					"0x%04X\n", reg);
1373 			return -EINVAL;
1374 		}
1375 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1376 		break;
1377 	case SX_MEMORY_EXPORT_BASE:
1378 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1379 		if (r) {
1380 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1381 					"0x%04X\n", reg);
1382 			return -EINVAL;
1383 		}
1384 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1385 		break;
1386 	case SX_MISC:
1387 		track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1388 		break;
1389 	default:
1390 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1391 		return -EINVAL;
1392 	}
1393 	return 0;
1394 }
1395 
1396 unsigned r600_mip_minify(unsigned size, unsigned level)
1397 {
1398 	unsigned val;
1399 
1400 	val = max(1U, size >> level);
1401 	if (level > 0)
1402 		val = roundup_pow_of_two(val);
1403 	return val;
1404 }
1405 
1406 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1407 			      unsigned w0, unsigned h0, unsigned d0, unsigned nsamples, unsigned format,
1408 			      unsigned block_align, unsigned height_align, unsigned base_align,
1409 			      unsigned *l0_size, unsigned *mipmap_size)
1410 {
1411 	unsigned offset, i, level;
1412 	unsigned width, height, depth, size;
1413 	unsigned blocksize;
1414 	unsigned nbx, nby;
1415 	unsigned nlevels = llevel - blevel + 1;
1416 
1417 	*l0_size = -1;
1418 	blocksize = r600_fmt_get_blocksize(format);
1419 
1420 	w0 = r600_mip_minify(w0, 0);
1421 	h0 = r600_mip_minify(h0, 0);
1422 	d0 = r600_mip_minify(d0, 0);
1423 	for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1424 		width = r600_mip_minify(w0, i);
1425 		nbx = r600_fmt_get_nblocksx(format, width);
1426 
1427 		nbx = round_up(nbx, block_align);
1428 
1429 		height = r600_mip_minify(h0, i);
1430 		nby = r600_fmt_get_nblocksy(format, height);
1431 		nby = round_up(nby, height_align);
1432 
1433 		depth = r600_mip_minify(d0, i);
1434 
1435 		size = nbx * nby * blocksize * nsamples;
1436 		if (nfaces)
1437 			size *= nfaces;
1438 		else
1439 			size *= depth;
1440 
1441 		if (i == 0)
1442 			*l0_size = size;
1443 
1444 		if (i == 0 || i == 1)
1445 			offset = round_up(offset, base_align);
1446 
1447 		offset += size;
1448 	}
1449 	*mipmap_size = offset;
1450 	if (llevel == 0)
1451 		*mipmap_size = *l0_size;
1452 	if (!blevel)
1453 		*mipmap_size -= *l0_size;
1454 }
1455 
1456 /**
1457  * r600_check_texture_resource() - check if register is authorized or not
1458  * @p: parser structure holding parsing context
1459  * @idx: index into the cs buffer
1460  * @texture: texture's bo structure
1461  * @mipmap: mipmap's bo structure
1462  * @base_offset: base offset (used for error checking)
1463  * @mip_offset: mip offset (used for error checking)
1464  * @tiling_flags: tiling flags
1465  *
1466  * This function will check that the resource has valid field and that
1467  * the texture and mipmap bo object are big enough to cover this resource.
1468  */
1469 static int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1470 					      struct radeon_bo *texture,
1471 					      struct radeon_bo *mipmap,
1472 					      u64 base_offset,
1473 					      u64 mip_offset,
1474 					      u32 tiling_flags)
1475 {
1476 	struct r600_cs_track *track = p->track;
1477 	u32 dim, nfaces, llevel, blevel, w0, h0, d0;
1478 	u32 word0, word1, l0_size, mipmap_size, word2, word3, word4, word5;
1479 	u32 height_align, pitch, pitch_align, depth_align;
1480 	u32 barray, larray;
1481 	u64 base_align;
1482 	struct array_mode_checker array_check;
1483 	u32 format;
1484 	bool is_array;
1485 
1486 	/* on legacy kernel we don't perform advanced check */
1487 	if (p->rdev == NULL)
1488 		return 0;
1489 
1490 	/* convert to bytes */
1491 	base_offset <<= 8;
1492 	mip_offset <<= 8;
1493 
1494 	word0 = radeon_get_ib_value(p, idx + 0);
1495 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1496 		if (tiling_flags & RADEON_TILING_MACRO)
1497 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1498 		else if (tiling_flags & RADEON_TILING_MICRO)
1499 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1500 	}
1501 	word1 = radeon_get_ib_value(p, idx + 1);
1502 	word2 = radeon_get_ib_value(p, idx + 2) << 8;
1503 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1504 	word4 = radeon_get_ib_value(p, idx + 4);
1505 	word5 = radeon_get_ib_value(p, idx + 5);
1506 	dim = G_038000_DIM(word0);
1507 	w0 = G_038000_TEX_WIDTH(word0) + 1;
1508 	pitch = (G_038000_PITCH(word0) + 1) * 8;
1509 	h0 = G_038004_TEX_HEIGHT(word1) + 1;
1510 	d0 = G_038004_TEX_DEPTH(word1);
1511 	format = G_038004_DATA_FORMAT(word1);
1512 	blevel = G_038010_BASE_LEVEL(word4);
1513 	llevel = G_038014_LAST_LEVEL(word5);
1514 	/* pitch in texels */
1515 	array_check.array_mode = G_038000_TILE_MODE(word0);
1516 	array_check.group_size = track->group_size;
1517 	array_check.nbanks = track->nbanks;
1518 	array_check.npipes = track->npipes;
1519 	array_check.nsamples = 1;
1520 	array_check.blocksize = r600_fmt_get_blocksize(format);
1521 	nfaces = 1;
1522 	is_array = false;
1523 	switch (dim) {
1524 	case V_038000_SQ_TEX_DIM_1D:
1525 	case V_038000_SQ_TEX_DIM_2D:
1526 	case V_038000_SQ_TEX_DIM_3D:
1527 		break;
1528 	case V_038000_SQ_TEX_DIM_CUBEMAP:
1529 		if (p->family >= CHIP_RV770)
1530 			nfaces = 8;
1531 		else
1532 			nfaces = 6;
1533 		break;
1534 	case V_038000_SQ_TEX_DIM_1D_ARRAY:
1535 	case V_038000_SQ_TEX_DIM_2D_ARRAY:
1536 		is_array = true;
1537 		break;
1538 	case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1539 		is_array = true;
1540 		fallthrough;
1541 	case V_038000_SQ_TEX_DIM_2D_MSAA:
1542 		array_check.nsamples = 1 << llevel;
1543 		llevel = 0;
1544 		break;
1545 	default:
1546 		dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1547 		return -EINVAL;
1548 	}
1549 	if (!r600_fmt_is_valid_texture(format, p->family)) {
1550 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1551 			 __func__, __LINE__, format);
1552 		return -EINVAL;
1553 	}
1554 
1555 	if (r600_get_array_mode_alignment(&array_check,
1556 					  &pitch_align, &height_align, &depth_align, &base_align)) {
1557 		dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1558 			 __func__, __LINE__, G_038000_TILE_MODE(word0));
1559 		return -EINVAL;
1560 	}
1561 
1562 	/* XXX check height as well... */
1563 
1564 	if (!IS_ALIGNED(pitch, pitch_align)) {
1565 		dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1566 			 __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1567 		return -EINVAL;
1568 	}
1569 	if (!IS_ALIGNED(base_offset, base_align)) {
1570 		dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1571 			 __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1572 		return -EINVAL;
1573 	}
1574 	if (!IS_ALIGNED(mip_offset, base_align)) {
1575 		dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1576 			 __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1577 		return -EINVAL;
1578 	}
1579 
1580 	if (blevel > llevel) {
1581 		dev_warn(p->dev, "texture blevel %d > llevel %d\n",
1582 			 blevel, llevel);
1583 	}
1584 	if (is_array) {
1585 		barray = G_038014_BASE_ARRAY(word5);
1586 		larray = G_038014_LAST_ARRAY(word5);
1587 
1588 		nfaces = larray - barray + 1;
1589 	}
1590 	r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, array_check.nsamples, format,
1591 			  pitch_align, height_align, base_align,
1592 			  &l0_size, &mipmap_size);
1593 	/* using get ib will give us the offset into the texture bo */
1594 	if ((l0_size + word2) > radeon_bo_size(texture)) {
1595 		dev_warn(p->dev, "texture bo too small ((%d %d) (%d %d) %d %d %d -> %d have %ld)\n",
1596 			 w0, h0, pitch_align, height_align,
1597 			 array_check.array_mode, format, word2,
1598 			 l0_size, radeon_bo_size(texture));
1599 		dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1600 		return -EINVAL;
1601 	}
1602 	/* using get ib will give us the offset into the mipmap bo */
1603 	if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1604 		/*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1605 		  w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1606 	}
1607 	return 0;
1608 }
1609 
1610 static bool r600_is_safe_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1611 {
1612 	u32 m, i;
1613 
1614 	i = (reg >> 7);
1615 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
1616 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1617 		return false;
1618 	}
1619 	m = 1 << ((reg >> 2) & 31);
1620 	if (!(r600_reg_safe_bm[i] & m))
1621 		return true;
1622 	dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1623 	return false;
1624 }
1625 
1626 static int r600_packet3_check(struct radeon_cs_parser *p,
1627 				struct radeon_cs_packet *pkt)
1628 {
1629 	struct radeon_bo_list *reloc;
1630 	struct r600_cs_track *track;
1631 	volatile u32 *ib;
1632 	unsigned idx;
1633 	unsigned i;
1634 	unsigned start_reg, end_reg, reg;
1635 	int r;
1636 	u32 idx_value;
1637 
1638 	track = (struct r600_cs_track *)p->track;
1639 	ib = p->ib.ptr;
1640 	idx = pkt->idx + 1;
1641 	idx_value = radeon_get_ib_value(p, idx);
1642 
1643 	switch (pkt->opcode) {
1644 	case PACKET3_SET_PREDICATION:
1645 	{
1646 		int pred_op;
1647 		int tmp;
1648 		uint64_t offset;
1649 
1650 		if (pkt->count != 1) {
1651 			DRM_ERROR("bad SET PREDICATION\n");
1652 			return -EINVAL;
1653 		}
1654 
1655 		tmp = radeon_get_ib_value(p, idx + 1);
1656 		pred_op = (tmp >> 16) & 0x7;
1657 
1658 		/* for the clear predicate operation */
1659 		if (pred_op == 0)
1660 			return 0;
1661 
1662 		if (pred_op > 2) {
1663 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1664 			return -EINVAL;
1665 		}
1666 
1667 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1668 		if (r) {
1669 			DRM_ERROR("bad SET PREDICATION\n");
1670 			return -EINVAL;
1671 		}
1672 
1673 		offset = reloc->gpu_offset +
1674 			 (idx_value & 0xfffffff0) +
1675 			 ((u64)(tmp & 0xff) << 32);
1676 
1677 		ib[idx + 0] = offset;
1678 		ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1679 	}
1680 	break;
1681 
1682 	case PACKET3_START_3D_CMDBUF:
1683 		if (p->family >= CHIP_RV770 || pkt->count) {
1684 			DRM_ERROR("bad START_3D\n");
1685 			return -EINVAL;
1686 		}
1687 		break;
1688 	case PACKET3_CONTEXT_CONTROL:
1689 		if (pkt->count != 1) {
1690 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1691 			return -EINVAL;
1692 		}
1693 		break;
1694 	case PACKET3_INDEX_TYPE:
1695 	case PACKET3_NUM_INSTANCES:
1696 		if (pkt->count) {
1697 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1698 			return -EINVAL;
1699 		}
1700 		break;
1701 	case PACKET3_DRAW_INDEX:
1702 	{
1703 		uint64_t offset;
1704 		if (pkt->count != 3) {
1705 			DRM_ERROR("bad DRAW_INDEX\n");
1706 			return -EINVAL;
1707 		}
1708 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1709 		if (r) {
1710 			DRM_ERROR("bad DRAW_INDEX\n");
1711 			return -EINVAL;
1712 		}
1713 
1714 		offset = reloc->gpu_offset +
1715 			 idx_value +
1716 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1717 
1718 		ib[idx+0] = offset;
1719 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1720 
1721 		r = r600_cs_track_check(p);
1722 		if (r) {
1723 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1724 			return r;
1725 		}
1726 		break;
1727 	}
1728 	case PACKET3_DRAW_INDEX_AUTO:
1729 		if (pkt->count != 1) {
1730 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1731 			return -EINVAL;
1732 		}
1733 		r = r600_cs_track_check(p);
1734 		if (r) {
1735 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1736 			return r;
1737 		}
1738 		break;
1739 	case PACKET3_DRAW_INDEX_IMMD_BE:
1740 	case PACKET3_DRAW_INDEX_IMMD:
1741 		if (pkt->count < 2) {
1742 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1743 			return -EINVAL;
1744 		}
1745 		r = r600_cs_track_check(p);
1746 		if (r) {
1747 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1748 			return r;
1749 		}
1750 		break;
1751 	case PACKET3_WAIT_REG_MEM:
1752 		if (pkt->count != 5) {
1753 			DRM_ERROR("bad WAIT_REG_MEM\n");
1754 			return -EINVAL;
1755 		}
1756 		/* bit 4 is reg (0) or mem (1) */
1757 		if (idx_value & 0x10) {
1758 			uint64_t offset;
1759 
1760 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1761 			if (r) {
1762 				DRM_ERROR("bad WAIT_REG_MEM\n");
1763 				return -EINVAL;
1764 			}
1765 
1766 			offset = reloc->gpu_offset +
1767 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff0) +
1768 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1769 
1770 			ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffff0);
1771 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1772 		} else if (idx_value & 0x100) {
1773 			DRM_ERROR("cannot use PFP on REG wait\n");
1774 			return -EINVAL;
1775 		}
1776 		break;
1777 	case PACKET3_CP_DMA:
1778 	{
1779 		u32 command, size;
1780 		u64 offset, tmp;
1781 		if (pkt->count != 4) {
1782 			DRM_ERROR("bad CP DMA\n");
1783 			return -EINVAL;
1784 		}
1785 		command = radeon_get_ib_value(p, idx+4);
1786 		size = command & 0x1fffff;
1787 		if (command & PACKET3_CP_DMA_CMD_SAS) {
1788 			/* src address space is register */
1789 			DRM_ERROR("CP DMA SAS not supported\n");
1790 			return -EINVAL;
1791 		} else {
1792 			if (command & PACKET3_CP_DMA_CMD_SAIC) {
1793 				DRM_ERROR("CP DMA SAIC only supported for registers\n");
1794 				return -EINVAL;
1795 			}
1796 			/* src address space is memory */
1797 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1798 			if (r) {
1799 				DRM_ERROR("bad CP DMA SRC\n");
1800 				return -EINVAL;
1801 			}
1802 
1803 			tmp = radeon_get_ib_value(p, idx) +
1804 				((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1805 
1806 			offset = reloc->gpu_offset + tmp;
1807 
1808 			if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1809 				dev_warn(p->dev, "CP DMA src buffer too small (%llu %lu)\n",
1810 					 tmp + size, radeon_bo_size(reloc->robj));
1811 				return -EINVAL;
1812 			}
1813 
1814 			ib[idx] = offset;
1815 			ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1816 		}
1817 		if (command & PACKET3_CP_DMA_CMD_DAS) {
1818 			/* dst address space is register */
1819 			DRM_ERROR("CP DMA DAS not supported\n");
1820 			return -EINVAL;
1821 		} else {
1822 			/* dst address space is memory */
1823 			if (command & PACKET3_CP_DMA_CMD_DAIC) {
1824 				DRM_ERROR("CP DMA DAIC only supported for registers\n");
1825 				return -EINVAL;
1826 			}
1827 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1828 			if (r) {
1829 				DRM_ERROR("bad CP DMA DST\n");
1830 				return -EINVAL;
1831 			}
1832 
1833 			tmp = radeon_get_ib_value(p, idx+2) +
1834 				((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
1835 
1836 			offset = reloc->gpu_offset + tmp;
1837 
1838 			if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1839 				dev_warn(p->dev, "CP DMA dst buffer too small (%llu %lu)\n",
1840 					 tmp + size, radeon_bo_size(reloc->robj));
1841 				return -EINVAL;
1842 			}
1843 
1844 			ib[idx+2] = offset;
1845 			ib[idx+3] = upper_32_bits(offset) & 0xff;
1846 		}
1847 		break;
1848 	}
1849 	case PACKET3_SURFACE_SYNC:
1850 		if (pkt->count != 3) {
1851 			DRM_ERROR("bad SURFACE_SYNC\n");
1852 			return -EINVAL;
1853 		}
1854 		/* 0xffffffff/0x0 is flush all cache flag */
1855 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1856 		    radeon_get_ib_value(p, idx + 2) != 0) {
1857 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1858 			if (r) {
1859 				DRM_ERROR("bad SURFACE_SYNC\n");
1860 				return -EINVAL;
1861 			}
1862 			ib[idx+2] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1863 		}
1864 		break;
1865 	case PACKET3_EVENT_WRITE:
1866 		if (pkt->count != 2 && pkt->count != 0) {
1867 			DRM_ERROR("bad EVENT_WRITE\n");
1868 			return -EINVAL;
1869 		}
1870 		if (pkt->count) {
1871 			uint64_t offset;
1872 
1873 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1874 			if (r) {
1875 				DRM_ERROR("bad EVENT_WRITE\n");
1876 				return -EINVAL;
1877 			}
1878 			offset = reloc->gpu_offset +
1879 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
1880 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1881 
1882 			ib[idx+1] = offset & 0xfffffff8;
1883 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1884 		}
1885 		break;
1886 	case PACKET3_EVENT_WRITE_EOP:
1887 	{
1888 		uint64_t offset;
1889 
1890 		if (pkt->count != 4) {
1891 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1892 			return -EINVAL;
1893 		}
1894 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1895 		if (r) {
1896 			DRM_ERROR("bad EVENT_WRITE\n");
1897 			return -EINVAL;
1898 		}
1899 
1900 		offset = reloc->gpu_offset +
1901 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
1902 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1903 
1904 		ib[idx+1] = offset & 0xfffffffc;
1905 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1906 		break;
1907 	}
1908 	case PACKET3_SET_CONFIG_REG:
1909 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1910 		end_reg = 4 * pkt->count + start_reg - 4;
1911 		if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1912 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1913 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1914 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1915 			return -EINVAL;
1916 		}
1917 		for (i = 0; i < pkt->count; i++) {
1918 			reg = start_reg + (4 * i);
1919 			r = r600_cs_check_reg(p, reg, idx+1+i);
1920 			if (r)
1921 				return r;
1922 		}
1923 		break;
1924 	case PACKET3_SET_CONTEXT_REG:
1925 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1926 		end_reg = 4 * pkt->count + start_reg - 4;
1927 		if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1928 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1929 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1930 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1931 			return -EINVAL;
1932 		}
1933 		for (i = 0; i < pkt->count; i++) {
1934 			reg = start_reg + (4 * i);
1935 			r = r600_cs_check_reg(p, reg, idx+1+i);
1936 			if (r)
1937 				return r;
1938 		}
1939 		break;
1940 	case PACKET3_SET_RESOURCE:
1941 		if (pkt->count % 7) {
1942 			DRM_ERROR("bad SET_RESOURCE\n");
1943 			return -EINVAL;
1944 		}
1945 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1946 		end_reg = 4 * pkt->count + start_reg - 4;
1947 		if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1948 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
1949 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
1950 			DRM_ERROR("bad SET_RESOURCE\n");
1951 			return -EINVAL;
1952 		}
1953 		for (i = 0; i < (pkt->count / 7); i++) {
1954 			struct radeon_bo *texture, *mipmap;
1955 			u32 size, offset, base_offset, mip_offset;
1956 
1957 			switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1958 			case SQ_TEX_VTX_VALID_TEXTURE:
1959 				/* tex base */
1960 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1961 				if (r) {
1962 					DRM_ERROR("bad SET_RESOURCE\n");
1963 					return -EINVAL;
1964 				}
1965 				base_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1966 				if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1967 					if (reloc->tiling_flags & RADEON_TILING_MACRO)
1968 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1969 					else if (reloc->tiling_flags & RADEON_TILING_MICRO)
1970 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1971 				}
1972 				texture = reloc->robj;
1973 				/* tex mip base */
1974 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1975 				if (r) {
1976 					DRM_ERROR("bad SET_RESOURCE\n");
1977 					return -EINVAL;
1978 				}
1979 				mip_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1980 				mipmap = reloc->robj;
1981 				r = r600_check_texture_resource(p,  idx+(i*7)+1,
1982 								texture, mipmap,
1983 								base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1984 								mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1985 								reloc->tiling_flags);
1986 				if (r)
1987 					return r;
1988 				ib[idx+1+(i*7)+2] += base_offset;
1989 				ib[idx+1+(i*7)+3] += mip_offset;
1990 				break;
1991 			case SQ_TEX_VTX_VALID_BUFFER:
1992 			{
1993 				uint64_t offset64;
1994 				/* vtx base */
1995 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1996 				if (r) {
1997 					DRM_ERROR("bad SET_RESOURCE\n");
1998 					return -EINVAL;
1999 				}
2000 				offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
2001 				size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
2002 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2003 					/* force size to size of the buffer */
2004 					dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
2005 						 size + offset, radeon_bo_size(reloc->robj));
2006 					ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj) - offset;
2007 				}
2008 
2009 				offset64 = reloc->gpu_offset + offset;
2010 				ib[idx+1+(i*8)+0] = offset64;
2011 				ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2012 						    (upper_32_bits(offset64) & 0xff);
2013 				break;
2014 			}
2015 			case SQ_TEX_VTX_INVALID_TEXTURE:
2016 			case SQ_TEX_VTX_INVALID_BUFFER:
2017 			default:
2018 				DRM_ERROR("bad SET_RESOURCE\n");
2019 				return -EINVAL;
2020 			}
2021 		}
2022 		break;
2023 	case PACKET3_SET_ALU_CONST:
2024 		if (track->sq_config & DX9_CONSTS) {
2025 			start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
2026 			end_reg = 4 * pkt->count + start_reg - 4;
2027 			if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
2028 			    (start_reg >= PACKET3_SET_ALU_CONST_END) ||
2029 			    (end_reg >= PACKET3_SET_ALU_CONST_END)) {
2030 				DRM_ERROR("bad SET_ALU_CONST\n");
2031 				return -EINVAL;
2032 			}
2033 		}
2034 		break;
2035 	case PACKET3_SET_BOOL_CONST:
2036 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
2037 		end_reg = 4 * pkt->count + start_reg - 4;
2038 		if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
2039 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2040 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2041 			DRM_ERROR("bad SET_BOOL_CONST\n");
2042 			return -EINVAL;
2043 		}
2044 		break;
2045 	case PACKET3_SET_LOOP_CONST:
2046 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
2047 		end_reg = 4 * pkt->count + start_reg - 4;
2048 		if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
2049 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2050 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2051 			DRM_ERROR("bad SET_LOOP_CONST\n");
2052 			return -EINVAL;
2053 		}
2054 		break;
2055 	case PACKET3_SET_CTL_CONST:
2056 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
2057 		end_reg = 4 * pkt->count + start_reg - 4;
2058 		if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
2059 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2060 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2061 			DRM_ERROR("bad SET_CTL_CONST\n");
2062 			return -EINVAL;
2063 		}
2064 		break;
2065 	case PACKET3_SET_SAMPLER:
2066 		if (pkt->count % 3) {
2067 			DRM_ERROR("bad SET_SAMPLER\n");
2068 			return -EINVAL;
2069 		}
2070 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
2071 		end_reg = 4 * pkt->count + start_reg - 4;
2072 		if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
2073 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
2074 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
2075 			DRM_ERROR("bad SET_SAMPLER\n");
2076 			return -EINVAL;
2077 		}
2078 		break;
2079 	case PACKET3_STRMOUT_BASE_UPDATE:
2080 		/* RS780 and RS880 also need this */
2081 		if (p->family < CHIP_RS780) {
2082 			DRM_ERROR("STRMOUT_BASE_UPDATE only supported on 7xx\n");
2083 			return -EINVAL;
2084 		}
2085 		if (pkt->count != 1) {
2086 			DRM_ERROR("bad STRMOUT_BASE_UPDATE packet count\n");
2087 			return -EINVAL;
2088 		}
2089 		if (idx_value > 3) {
2090 			DRM_ERROR("bad STRMOUT_BASE_UPDATE index\n");
2091 			return -EINVAL;
2092 		}
2093 		{
2094 			u64 offset;
2095 
2096 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2097 			if (r) {
2098 				DRM_ERROR("bad STRMOUT_BASE_UPDATE reloc\n");
2099 				return -EINVAL;
2100 			}
2101 
2102 			if (reloc->robj != track->vgt_strmout_bo[idx_value]) {
2103 				DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo does not match\n");
2104 				return -EINVAL;
2105 			}
2106 
2107 			offset = (u64)radeon_get_ib_value(p, idx+1) << 8;
2108 			if (offset != track->vgt_strmout_bo_offset[idx_value]) {
2109 				DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo offset does not match: 0x%llx, 0x%x\n",
2110 					  offset, track->vgt_strmout_bo_offset[idx_value]);
2111 				return -EINVAL;
2112 			}
2113 
2114 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2115 				DRM_ERROR("bad STRMOUT_BASE_UPDATE bo too small: 0x%llx, 0x%lx\n",
2116 					  offset + 4, radeon_bo_size(reloc->robj));
2117 				return -EINVAL;
2118 			}
2119 			ib[idx+1] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2120 		}
2121 		break;
2122 	case PACKET3_SURFACE_BASE_UPDATE:
2123 		if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
2124 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2125 			return -EINVAL;
2126 		}
2127 		if (pkt->count) {
2128 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2129 			return -EINVAL;
2130 		}
2131 		break;
2132 	case PACKET3_STRMOUT_BUFFER_UPDATE:
2133 		if (pkt->count != 4) {
2134 			DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2135 			return -EINVAL;
2136 		}
2137 		/* Updating memory at DST_ADDRESS. */
2138 		if (idx_value & 0x1) {
2139 			u64 offset;
2140 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2141 			if (r) {
2142 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2143 				return -EINVAL;
2144 			}
2145 			offset = radeon_get_ib_value(p, idx+1);
2146 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2147 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2148 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2149 					  offset + 4, radeon_bo_size(reloc->robj));
2150 				return -EINVAL;
2151 			}
2152 			offset += reloc->gpu_offset;
2153 			ib[idx+1] = offset;
2154 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2155 		}
2156 		/* Reading data from SRC_ADDRESS. */
2157 		if (((idx_value >> 1) & 0x3) == 2) {
2158 			u64 offset;
2159 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2160 			if (r) {
2161 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2162 				return -EINVAL;
2163 			}
2164 			offset = radeon_get_ib_value(p, idx+3);
2165 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2166 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2167 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2168 					  offset + 4, radeon_bo_size(reloc->robj));
2169 				return -EINVAL;
2170 			}
2171 			offset += reloc->gpu_offset;
2172 			ib[idx+3] = offset;
2173 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2174 		}
2175 		break;
2176 	case PACKET3_MEM_WRITE:
2177 	{
2178 		u64 offset;
2179 
2180 		if (pkt->count != 3) {
2181 			DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2182 			return -EINVAL;
2183 		}
2184 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2185 		if (r) {
2186 			DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2187 			return -EINVAL;
2188 		}
2189 		offset = radeon_get_ib_value(p, idx+0);
2190 		offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2191 		if (offset & 0x7) {
2192 			DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2193 			return -EINVAL;
2194 		}
2195 		if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2196 			DRM_ERROR("bad MEM_WRITE bo too small: 0x%llx, 0x%lx\n",
2197 				  offset + 8, radeon_bo_size(reloc->robj));
2198 			return -EINVAL;
2199 		}
2200 		offset += reloc->gpu_offset;
2201 		ib[idx+0] = offset;
2202 		ib[idx+1] = upper_32_bits(offset) & 0xff;
2203 		break;
2204 	}
2205 	case PACKET3_COPY_DW:
2206 		if (pkt->count != 4) {
2207 			DRM_ERROR("bad COPY_DW (invalid count)\n");
2208 			return -EINVAL;
2209 		}
2210 		if (idx_value & 0x1) {
2211 			u64 offset;
2212 			/* SRC is memory. */
2213 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2214 			if (r) {
2215 				DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2216 				return -EINVAL;
2217 			}
2218 			offset = radeon_get_ib_value(p, idx+1);
2219 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2220 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2221 				DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2222 					  offset + 4, radeon_bo_size(reloc->robj));
2223 				return -EINVAL;
2224 			}
2225 			offset += reloc->gpu_offset;
2226 			ib[idx+1] = offset;
2227 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2228 		} else {
2229 			/* SRC is a reg. */
2230 			reg = radeon_get_ib_value(p, idx+1) << 2;
2231 			if (!r600_is_safe_reg(p, reg, idx+1))
2232 				return -EINVAL;
2233 		}
2234 		if (idx_value & 0x2) {
2235 			u64 offset;
2236 			/* DST is memory. */
2237 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2238 			if (r) {
2239 				DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2240 				return -EINVAL;
2241 			}
2242 			offset = radeon_get_ib_value(p, idx+3);
2243 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2244 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2245 				DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2246 					  offset + 4, radeon_bo_size(reloc->robj));
2247 				return -EINVAL;
2248 			}
2249 			offset += reloc->gpu_offset;
2250 			ib[idx+3] = offset;
2251 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2252 		} else {
2253 			/* DST is a reg. */
2254 			reg = radeon_get_ib_value(p, idx+3) << 2;
2255 			if (!r600_is_safe_reg(p, reg, idx+3))
2256 				return -EINVAL;
2257 		}
2258 		break;
2259 	case PACKET3_NOP:
2260 		break;
2261 	default:
2262 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2263 		return -EINVAL;
2264 	}
2265 	return 0;
2266 }
2267 
2268 int r600_cs_parse(struct radeon_cs_parser *p)
2269 {
2270 	struct radeon_cs_packet pkt;
2271 	struct r600_cs_track *track;
2272 	int r;
2273 
2274 	if (p->track == NULL) {
2275 		/* initialize tracker, we are in kms */
2276 		track = kzalloc(sizeof(*track), GFP_KERNEL);
2277 		if (track == NULL)
2278 			return -ENOMEM;
2279 		r600_cs_track_init(track);
2280 		if (p->rdev->family < CHIP_RV770) {
2281 			track->npipes = p->rdev->config.r600.tiling_npipes;
2282 			track->nbanks = p->rdev->config.r600.tiling_nbanks;
2283 			track->group_size = p->rdev->config.r600.tiling_group_size;
2284 		} else if (p->rdev->family <= CHIP_RV740) {
2285 			track->npipes = p->rdev->config.rv770.tiling_npipes;
2286 			track->nbanks = p->rdev->config.rv770.tiling_nbanks;
2287 			track->group_size = p->rdev->config.rv770.tiling_group_size;
2288 		}
2289 		p->track = track;
2290 	}
2291 	do {
2292 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2293 		if (r) {
2294 			kfree(p->track);
2295 			p->track = NULL;
2296 			return r;
2297 		}
2298 		p->idx += pkt.count + 2;
2299 		switch (pkt.type) {
2300 		case RADEON_PACKET_TYPE0:
2301 			r = r600_cs_parse_packet0(p, &pkt);
2302 			break;
2303 		case RADEON_PACKET_TYPE2:
2304 			break;
2305 		case RADEON_PACKET_TYPE3:
2306 			r = r600_packet3_check(p, &pkt);
2307 			break;
2308 		default:
2309 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2310 			kfree(p->track);
2311 			p->track = NULL;
2312 			return -EINVAL;
2313 		}
2314 		if (r) {
2315 			kfree(p->track);
2316 			p->track = NULL;
2317 			return r;
2318 		}
2319 	} while (p->idx < p->chunk_ib->length_dw);
2320 #if 0
2321 	for (r = 0; r < p->ib.length_dw; r++) {
2322 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2323 		mdelay(1);
2324 	}
2325 #endif
2326 	kfree(p->track);
2327 	p->track = NULL;
2328 	return 0;
2329 }
2330 
2331 /*
2332  *  DMA
2333  */
2334 /**
2335  * r600_dma_cs_next_reloc() - parse next reloc
2336  * @p:		parser structure holding parsing context.
2337  * @cs_reloc:		reloc information
2338  *
2339  * Return the next reloc, do bo validation and compute
2340  * GPU offset using the provided start.
2341  **/
2342 int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
2343 			   struct radeon_bo_list **cs_reloc)
2344 {
2345 	unsigned idx;
2346 
2347 	*cs_reloc = NULL;
2348 	if (p->chunk_relocs == NULL) {
2349 		DRM_ERROR("No relocation chunk !\n");
2350 		return -EINVAL;
2351 	}
2352 	idx = p->dma_reloc_idx;
2353 	if (idx >= p->nrelocs) {
2354 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
2355 			  idx, p->nrelocs);
2356 		return -EINVAL;
2357 	}
2358 	*cs_reloc = &p->relocs[idx];
2359 	p->dma_reloc_idx++;
2360 	return 0;
2361 }
2362 
2363 #define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
2364 #define GET_DMA_COUNT(h) ((h) & 0x0000ffff)
2365 #define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
2366 
2367 /**
2368  * r600_dma_cs_parse() - parse the DMA IB
2369  * @p:		parser structure holding parsing context.
2370  *
2371  * Parses the DMA IB from the CS ioctl and updates
2372  * the GPU addresses based on the reloc information and
2373  * checks for errors. (R6xx-R7xx)
2374  * Returns 0 for success and an error on failure.
2375  **/
2376 int r600_dma_cs_parse(struct radeon_cs_parser *p)
2377 {
2378 	struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
2379 	struct radeon_bo_list *src_reloc, *dst_reloc;
2380 	u32 header, cmd, count, tiled;
2381 	volatile u32 *ib = p->ib.ptr;
2382 	u32 idx, idx_value;
2383 	u64 src_offset, dst_offset;
2384 	int r;
2385 
2386 	do {
2387 		if (p->idx >= ib_chunk->length_dw) {
2388 			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2389 				  p->idx, ib_chunk->length_dw);
2390 			return -EINVAL;
2391 		}
2392 		idx = p->idx;
2393 		header = radeon_get_ib_value(p, idx);
2394 		cmd = GET_DMA_CMD(header);
2395 		count = GET_DMA_COUNT(header);
2396 		tiled = GET_DMA_T(header);
2397 
2398 		switch (cmd) {
2399 		case DMA_PACKET_WRITE:
2400 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2401 			if (r) {
2402 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2403 				return -EINVAL;
2404 			}
2405 			if (tiled) {
2406 				dst_offset = radeon_get_ib_value(p, idx+1);
2407 				dst_offset <<= 8;
2408 
2409 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2410 				p->idx += count + 5;
2411 			} else {
2412 				dst_offset = radeon_get_ib_value(p, idx+1);
2413 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2414 
2415 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2416 				ib[idx+2] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2417 				p->idx += count + 3;
2418 			}
2419 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2420 				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
2421 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2422 				return -EINVAL;
2423 			}
2424 			break;
2425 		case DMA_PACKET_COPY:
2426 			r = r600_dma_cs_next_reloc(p, &src_reloc);
2427 			if (r) {
2428 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2429 				return -EINVAL;
2430 			}
2431 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2432 			if (r) {
2433 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2434 				return -EINVAL;
2435 			}
2436 			if (tiled) {
2437 				idx_value = radeon_get_ib_value(p, idx + 2);
2438 				/* detile bit */
2439 				if (idx_value & (1 << 31)) {
2440 					/* tiled src, linear dst */
2441 					src_offset = radeon_get_ib_value(p, idx+1);
2442 					src_offset <<= 8;
2443 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
2444 
2445 					dst_offset = radeon_get_ib_value(p, idx+5);
2446 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2447 					ib[idx+5] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2448 					ib[idx+6] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2449 				} else {
2450 					/* linear src, tiled dst */
2451 					src_offset = radeon_get_ib_value(p, idx+5);
2452 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2453 					ib[idx+5] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2454 					ib[idx+6] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2455 
2456 					dst_offset = radeon_get_ib_value(p, idx+1);
2457 					dst_offset <<= 8;
2458 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2459 				}
2460 				p->idx += 7;
2461 			} else {
2462 				if (p->family >= CHIP_RV770) {
2463 					src_offset = radeon_get_ib_value(p, idx+2);
2464 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2465 					dst_offset = radeon_get_ib_value(p, idx+1);
2466 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2467 
2468 					ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2469 					ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2470 					ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2471 					ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2472 					p->idx += 5;
2473 				} else {
2474 					src_offset = radeon_get_ib_value(p, idx+2);
2475 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2476 					dst_offset = radeon_get_ib_value(p, idx+1);
2477 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff0000)) << 16;
2478 
2479 					ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2480 					ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2481 					ib[idx+3] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2482 					ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) & 0xff) << 16;
2483 					p->idx += 4;
2484 				}
2485 			}
2486 			if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2487 				dev_warn(p->dev, "DMA copy src buffer too small (%llu %lu)\n",
2488 					 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2489 				return -EINVAL;
2490 			}
2491 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2492 				dev_warn(p->dev, "DMA write dst buffer too small (%llu %lu)\n",
2493 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2494 				return -EINVAL;
2495 			}
2496 			break;
2497 		case DMA_PACKET_CONSTANT_FILL:
2498 			if (p->family < CHIP_RV770) {
2499 				DRM_ERROR("Constant Fill is 7xx only !\n");
2500 				return -EINVAL;
2501 			}
2502 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2503 			if (r) {
2504 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2505 				return -EINVAL;
2506 			}
2507 			dst_offset = radeon_get_ib_value(p, idx+1);
2508 			dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
2509 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2510 				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
2511 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2512 				return -EINVAL;
2513 			}
2514 			ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2515 			ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) << 16) & 0x00ff0000;
2516 			p->idx += 4;
2517 			break;
2518 		case DMA_PACKET_NOP:
2519 			p->idx += 1;
2520 			break;
2521 		default:
2522 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
2523 			return -EINVAL;
2524 		}
2525 	} while (p->idx < p->chunk_ib->length_dw);
2526 #if 0
2527 	for (r = 0; r < p->ib->length_dw; r++) {
2528 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2529 		mdelay(1);
2530 	}
2531 #endif
2532 	return 0;
2533 }
2534