xref: /openbsd/sys/dev/pci/drm/radeon/r600_cs.c (revision ad8b1aaf)
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 "r600d.h"
33 #include "r600_reg_safe.h"
34 
35 static int r600_nomm;
36 extern void r600_cs_legacy_get_tiling_conf(struct drm_device *dev, u32 *npipes, u32 *nbanks, u32 *group_size);
37 
38 
39 struct r600_cs_track {
40 	/* configuration we miror so that we use same code btw kms/ums */
41 	u32			group_size;
42 	u32			nbanks;
43 	u32			npipes;
44 	/* value we track */
45 	u32			sq_config;
46 	u32			log_nsamples;
47 	u32			nsamples;
48 	u32			cb_color_base_last[8];
49 	struct radeon_bo	*cb_color_bo[8];
50 	u64			cb_color_bo_mc[8];
51 	u64			cb_color_bo_offset[8];
52 	struct radeon_bo	*cb_color_frag_bo[8];
53 	u64			cb_color_frag_offset[8];
54 	struct radeon_bo	*cb_color_tile_bo[8];
55 	u64			cb_color_tile_offset[8];
56 	u32			cb_color_mask[8];
57 	u32			cb_color_info[8];
58 	u32			cb_color_view[8];
59 	u32			cb_color_size_idx[8]; /* unused */
60 	u32			cb_target_mask;
61 	u32			cb_shader_mask;  /* unused */
62 	bool			is_resolve;
63 	u32			cb_color_size[8];
64 	u32			vgt_strmout_en;
65 	u32			vgt_strmout_buffer_en;
66 	struct radeon_bo	*vgt_strmout_bo[4];
67 	u64			vgt_strmout_bo_mc[4]; /* unused */
68 	u32			vgt_strmout_bo_offset[4];
69 	u32			vgt_strmout_size[4];
70 	u32			db_depth_control;
71 	u32			db_depth_info;
72 	u32			db_depth_size_idx;
73 	u32			db_depth_view;
74 	u32			db_depth_size;
75 	u32			db_offset;
76 	struct radeon_bo	*db_bo;
77 	u64			db_bo_mc;
78 	bool			sx_misc_kill_all_prims;
79 	bool			cb_dirty;
80 	bool			db_dirty;
81 	bool			streamout_dirty;
82 	struct radeon_bo	*htile_bo;
83 	u64			htile_offset;
84 	u32			htile_surface;
85 };
86 
87 #define FMT_8_BIT(fmt, vc)   [fmt] = { 1, 1, 1, vc, CHIP_R600 }
88 #define FMT_16_BIT(fmt, vc)  [fmt] = { 1, 1, 2, vc, CHIP_R600 }
89 #define FMT_24_BIT(fmt)      [fmt] = { 1, 1, 4,  0, CHIP_R600 }
90 #define FMT_32_BIT(fmt, vc)  [fmt] = { 1, 1, 4, vc, CHIP_R600 }
91 #define FMT_48_BIT(fmt)      [fmt] = { 1, 1, 8,  0, CHIP_R600 }
92 #define FMT_64_BIT(fmt, vc)  [fmt] = { 1, 1, 8, vc, CHIP_R600 }
93 #define FMT_96_BIT(fmt)      [fmt] = { 1, 1, 12, 0, CHIP_R600 }
94 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16,vc, CHIP_R600 }
95 
96 struct gpu_formats {
97 	unsigned blockwidth;
98 	unsigned blockheight;
99 	unsigned blocksize;
100 	unsigned valid_color;
101 	enum radeon_family min_family;
102 };
103 
104 static const struct gpu_formats color_formats_table[] = {
105 	/* 8 bit */
106 	FMT_8_BIT(V_038004_COLOR_8, 1),
107 	FMT_8_BIT(V_038004_COLOR_4_4, 1),
108 	FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
109 	FMT_8_BIT(V_038004_FMT_1, 0),
110 
111 	/* 16-bit */
112 	FMT_16_BIT(V_038004_COLOR_16, 1),
113 	FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
114 	FMT_16_BIT(V_038004_COLOR_8_8, 1),
115 	FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
116 	FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
117 	FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
118 	FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
119 	FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
120 
121 	/* 24-bit */
122 	FMT_24_BIT(V_038004_FMT_8_8_8),
123 
124 	/* 32-bit */
125 	FMT_32_BIT(V_038004_COLOR_32, 1),
126 	FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
127 	FMT_32_BIT(V_038004_COLOR_16_16, 1),
128 	FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
129 	FMT_32_BIT(V_038004_COLOR_8_24, 1),
130 	FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
131 	FMT_32_BIT(V_038004_COLOR_24_8, 1),
132 	FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
133 	FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
134 	FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
135 	FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
136 	FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
137 	FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
138 	FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
139 	FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
140 	FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
141 	FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
142 	FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
143 
144 	/* 48-bit */
145 	FMT_48_BIT(V_038004_FMT_16_16_16),
146 	FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
147 
148 	/* 64-bit */
149 	FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
150 	FMT_64_BIT(V_038004_COLOR_32_32, 1),
151 	FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
152 	FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
153 	FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
154 
155 	FMT_96_BIT(V_038004_FMT_32_32_32),
156 	FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
157 
158 	/* 128-bit */
159 	FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
160 	FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
161 
162 	[V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
163 	[V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
164 
165 	/* block compressed formats */
166 	[V_038004_FMT_BC1] = { 4, 4, 8, 0 },
167 	[V_038004_FMT_BC2] = { 4, 4, 16, 0 },
168 	[V_038004_FMT_BC3] = { 4, 4, 16, 0 },
169 	[V_038004_FMT_BC4] = { 4, 4, 8, 0 },
170 	[V_038004_FMT_BC5] = { 4, 4, 16, 0},
171 	[V_038004_FMT_BC6] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
172 	[V_038004_FMT_BC7] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
173 
174 	/* The other Evergreen formats */
175 	[V_038004_FMT_32_AS_32_32_32_32] = { 1, 1, 4, 0, CHIP_CEDAR},
176 };
177 
178 bool r600_fmt_is_valid_color(u32 format)
179 {
180 	if (format >= ARRAY_SIZE(color_formats_table))
181 		return false;
182 
183 	if (color_formats_table[format].valid_color)
184 		return true;
185 
186 	return false;
187 }
188 
189 bool r600_fmt_is_valid_texture(u32 format, enum radeon_family family)
190 {
191 	if (format >= ARRAY_SIZE(color_formats_table))
192 		return false;
193 
194 	if (family < color_formats_table[format].min_family)
195 		return false;
196 
197 	if (color_formats_table[format].blockwidth > 0)
198 		return true;
199 
200 	return false;
201 }
202 
203 int r600_fmt_get_blocksize(u32 format)
204 {
205 	if (format >= ARRAY_SIZE(color_formats_table))
206 		return 0;
207 
208 	return color_formats_table[format].blocksize;
209 }
210 
211 int r600_fmt_get_nblocksx(u32 format, u32 w)
212 {
213 	unsigned bw;
214 
215 	if (format >= ARRAY_SIZE(color_formats_table))
216 		return 0;
217 
218 	bw = color_formats_table[format].blockwidth;
219 	if (bw == 0)
220 		return 0;
221 
222 	return (w + bw - 1) / bw;
223 }
224 
225 int r600_fmt_get_nblocksy(u32 format, u32 h)
226 {
227 	unsigned bh;
228 
229 	if (format >= ARRAY_SIZE(color_formats_table))
230 		return 0;
231 
232 	bh = color_formats_table[format].blockheight;
233 	if (bh == 0)
234 		return 0;
235 
236 	return (h + bh - 1) / bh;
237 }
238 
239 struct array_mode_checker {
240 	int array_mode;
241 	u32 group_size;
242 	u32 nbanks;
243 	u32 npipes;
244 	u32 nsamples;
245 	u32 blocksize;
246 };
247 
248 /* returns alignment in pixels for pitch/height/depth and bytes for base */
249 static int r600_get_array_mode_alignment(struct array_mode_checker *values,
250 						u32 *pitch_align,
251 						u32 *height_align,
252 						u32 *depth_align,
253 						u64 *base_align)
254 {
255 	u32 tile_width = 8;
256 	u32 tile_height = 8;
257 	u32 macro_tile_width = values->nbanks;
258 	u32 macro_tile_height = values->npipes;
259 	u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
260 	u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
261 
262 	switch (values->array_mode) {
263 	case ARRAY_LINEAR_GENERAL:
264 		/* technically tile_width/_height for pitch/height */
265 		*pitch_align = 1; /* tile_width */
266 		*height_align = 1; /* tile_height */
267 		*depth_align = 1;
268 		*base_align = 1;
269 		break;
270 	case ARRAY_LINEAR_ALIGNED:
271 		*pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
272 		*height_align = 1;
273 		*depth_align = 1;
274 		*base_align = values->group_size;
275 		break;
276 	case ARRAY_1D_TILED_THIN1:
277 		*pitch_align = max((u32)tile_width,
278 				   (u32)(values->group_size /
279 					 (tile_height * values->blocksize * values->nsamples)));
280 		*height_align = tile_height;
281 		*depth_align = 1;
282 		*base_align = values->group_size;
283 		break;
284 	case ARRAY_2D_TILED_THIN1:
285 		*pitch_align = max((u32)macro_tile_width * tile_width,
286 				(u32)((values->group_size * values->nbanks) /
287 				(values->blocksize * values->nsamples * tile_width)));
288 		*height_align = macro_tile_height * tile_height;
289 		*depth_align = 1;
290 		*base_align = max(macro_tile_bytes,
291 				  (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
292 		break;
293 	default:
294 		return -EINVAL;
295 	}
296 
297 	return 0;
298 }
299 
300 static void r600_cs_track_init(struct r600_cs_track *track)
301 {
302 	int i;
303 
304 	/* assume DX9 mode */
305 	track->sq_config = DX9_CONSTS;
306 	for (i = 0; i < 8; i++) {
307 		track->cb_color_base_last[i] = 0;
308 		track->cb_color_size[i] = 0;
309 		track->cb_color_size_idx[i] = 0;
310 		track->cb_color_info[i] = 0;
311 		track->cb_color_view[i] = 0xFFFFFFFF;
312 		track->cb_color_bo[i] = NULL;
313 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
314 		track->cb_color_bo_mc[i] = 0xFFFFFFFF;
315 		track->cb_color_frag_bo[i] = NULL;
316 		track->cb_color_frag_offset[i] = 0xFFFFFFFF;
317 		track->cb_color_tile_bo[i] = NULL;
318 		track->cb_color_tile_offset[i] = 0xFFFFFFFF;
319 		track->cb_color_mask[i] = 0xFFFFFFFF;
320 	}
321 	track->is_resolve = false;
322 	track->nsamples = 16;
323 	track->log_nsamples = 4;
324 	track->cb_target_mask = 0xFFFFFFFF;
325 	track->cb_shader_mask = 0xFFFFFFFF;
326 	track->cb_dirty = true;
327 	track->db_bo = NULL;
328 	track->db_bo_mc = 0xFFFFFFFF;
329 	/* assume the biggest format and that htile is enabled */
330 	track->db_depth_info = 7 | (1 << 25);
331 	track->db_depth_view = 0xFFFFC000;
332 	track->db_depth_size = 0xFFFFFFFF;
333 	track->db_depth_size_idx = 0;
334 	track->db_depth_control = 0xFFFFFFFF;
335 	track->db_dirty = true;
336 	track->htile_bo = NULL;
337 	track->htile_offset = 0xFFFFFFFF;
338 	track->htile_surface = 0;
339 
340 	for (i = 0; i < 4; i++) {
341 		track->vgt_strmout_size[i] = 0;
342 		track->vgt_strmout_bo[i] = NULL;
343 		track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
344 		track->vgt_strmout_bo_mc[i] = 0xFFFFFFFF;
345 	}
346 	track->streamout_dirty = true;
347 	track->sx_misc_kill_all_prims = false;
348 }
349 
350 static int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
351 {
352 	struct r600_cs_track *track = p->track;
353 	u32 slice_tile_max, tmp;
354 	u32 height, height_align, pitch, pitch_align, depth_align;
355 	u64 base_offset, base_align;
356 	struct array_mode_checker array_check;
357 	volatile u32 *ib = p->ib.ptr;
358 	unsigned array_mode;
359 	u32 format;
360 	/* When resolve is used, the second colorbuffer has always 1 sample. */
361 	unsigned nsamples = track->is_resolve && i == 1 ? 1 : track->nsamples;
362 
363 	format = G_0280A0_FORMAT(track->cb_color_info[i]);
364 	if (!r600_fmt_is_valid_color(format)) {
365 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
366 			 __func__, __LINE__, format,
367 			i, track->cb_color_info[i]);
368 		return -EINVAL;
369 	}
370 	/* pitch in pixels */
371 	pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
372 	slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
373 	slice_tile_max *= 64;
374 	height = slice_tile_max / pitch;
375 	if (height > 8192)
376 		height = 8192;
377 	array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
378 
379 	base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
380 	array_check.array_mode = array_mode;
381 	array_check.group_size = track->group_size;
382 	array_check.nbanks = track->nbanks;
383 	array_check.npipes = track->npipes;
384 	array_check.nsamples = nsamples;
385 	array_check.blocksize = r600_fmt_get_blocksize(format);
386 	if (r600_get_array_mode_alignment(&array_check,
387 					  &pitch_align, &height_align, &depth_align, &base_align)) {
388 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
389 			 G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
390 			 track->cb_color_info[i]);
391 		return -EINVAL;
392 	}
393 	switch (array_mode) {
394 	case V_0280A0_ARRAY_LINEAR_GENERAL:
395 		break;
396 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
397 		break;
398 	case V_0280A0_ARRAY_1D_TILED_THIN1:
399 		/* avoid breaking userspace */
400 		if (height > 7)
401 			height &= ~0x7;
402 		break;
403 	case V_0280A0_ARRAY_2D_TILED_THIN1:
404 		break;
405 	default:
406 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
407 			G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
408 			track->cb_color_info[i]);
409 		return -EINVAL;
410 	}
411 
412 	if (!IS_ALIGNED(pitch, pitch_align)) {
413 		dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
414 			 __func__, __LINE__, pitch, pitch_align, array_mode);
415 		return -EINVAL;
416 	}
417 	if (!IS_ALIGNED(height, height_align)) {
418 		dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
419 			 __func__, __LINE__, height, height_align, array_mode);
420 		return -EINVAL;
421 	}
422 	if (!IS_ALIGNED(base_offset, base_align)) {
423 		dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
424 			 base_offset, base_align, array_mode);
425 		return -EINVAL;
426 	}
427 
428 	/* check offset */
429 	tmp = r600_fmt_get_nblocksy(format, height) * r600_fmt_get_nblocksx(format, pitch) *
430 	      r600_fmt_get_blocksize(format) * nsamples;
431 	switch (array_mode) {
432 	default:
433 	case V_0280A0_ARRAY_LINEAR_GENERAL:
434 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
435 		tmp += track->cb_color_view[i] & 0xFF;
436 		break;
437 	case V_0280A0_ARRAY_1D_TILED_THIN1:
438 	case V_0280A0_ARRAY_2D_TILED_THIN1:
439 		tmp += G_028080_SLICE_MAX(track->cb_color_view[i]) * tmp;
440 		break;
441 	}
442 	if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
443 		if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
444 			/* the initial DDX does bad things with the CB size occasionally */
445 			/* it rounds up height too far for slice tile max but the BO is smaller */
446 			/* r600c,g also seem to flush at bad times in some apps resulting in
447 			 * bogus values here. So for linear just allow anything to avoid breaking
448 			 * broken userspace.
449 			 */
450 		} else {
451 			dev_warn(p->dev, "%s offset[%d] %d %llu %d %lu too big (%d %d) (%d %d %d)\n",
452 				 __func__, i, array_mode,
453 				 track->cb_color_bo_offset[i], tmp,
454 				 radeon_bo_size(track->cb_color_bo[i]),
455 				 pitch, height, r600_fmt_get_nblocksx(format, pitch),
456 				 r600_fmt_get_nblocksy(format, height),
457 				 r600_fmt_get_blocksize(format));
458 			return -EINVAL;
459 		}
460 	}
461 	/* limit max tile */
462 	tmp = (height * pitch) >> 6;
463 	if (tmp < slice_tile_max)
464 		slice_tile_max = tmp;
465 	tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
466 		S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
467 	ib[track->cb_color_size_idx[i]] = tmp;
468 
469 	/* FMASK/CMASK */
470 	switch (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
471 	case V_0280A0_TILE_DISABLE:
472 		break;
473 	case V_0280A0_FRAG_ENABLE:
474 		if (track->nsamples > 1) {
475 			uint32_t tile_max = G_028100_FMASK_TILE_MAX(track->cb_color_mask[i]);
476 			/* the tile size is 8x8, but the size is in units of bits.
477 			 * for bytes, do just * 8. */
478 			uint32_t bytes = track->nsamples * track->log_nsamples * 8 * (tile_max + 1);
479 
480 			if (bytes + track->cb_color_frag_offset[i] >
481 			    radeon_bo_size(track->cb_color_frag_bo[i])) {
482 				dev_warn(p->dev, "%s FMASK_TILE_MAX too large "
483 					 "(tile_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
484 					 __func__, tile_max, bytes,
485 					 track->cb_color_frag_offset[i],
486 					 radeon_bo_size(track->cb_color_frag_bo[i]));
487 				return -EINVAL;
488 			}
489 		}
490 		fallthrough;
491 	case V_0280A0_CLEAR_ENABLE:
492 	{
493 		uint32_t block_max = G_028100_CMASK_BLOCK_MAX(track->cb_color_mask[i]);
494 		/* One block = 128x128 pixels, one 8x8 tile has 4 bits..
495 		 * (128*128) / (8*8) / 2 = 128 bytes per block. */
496 		uint32_t bytes = (block_max + 1) * 128;
497 
498 		if (bytes + track->cb_color_tile_offset[i] >
499 		    radeon_bo_size(track->cb_color_tile_bo[i])) {
500 			dev_warn(p->dev, "%s CMASK_BLOCK_MAX too large "
501 				 "(block_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
502 				 __func__, block_max, bytes,
503 				 track->cb_color_tile_offset[i],
504 				 radeon_bo_size(track->cb_color_tile_bo[i]));
505 			return -EINVAL;
506 		}
507 		break;
508 	}
509 	default:
510 		dev_warn(p->dev, "%s invalid tile mode\n", __func__);
511 		return -EINVAL;
512 	}
513 	return 0;
514 }
515 
516 static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
517 {
518 	struct r600_cs_track *track = p->track;
519 	u32 nviews, bpe, ntiles, slice_tile_max, tmp;
520 	u32 height_align, pitch_align, depth_align;
521 	u32 pitch = 8192;
522 	u32 height = 8192;
523 	u64 base_offset, base_align;
524 	struct array_mode_checker array_check;
525 	int array_mode;
526 	volatile u32 *ib = p->ib.ptr;
527 
528 
529 	if (track->db_bo == NULL) {
530 		dev_warn(p->dev, "z/stencil with no depth buffer\n");
531 		return -EINVAL;
532 	}
533 	switch (G_028010_FORMAT(track->db_depth_info)) {
534 	case V_028010_DEPTH_16:
535 		bpe = 2;
536 		break;
537 	case V_028010_DEPTH_X8_24:
538 	case V_028010_DEPTH_8_24:
539 	case V_028010_DEPTH_X8_24_FLOAT:
540 	case V_028010_DEPTH_8_24_FLOAT:
541 	case V_028010_DEPTH_32_FLOAT:
542 		bpe = 4;
543 		break;
544 	case V_028010_DEPTH_X24_8_32_FLOAT:
545 		bpe = 8;
546 		break;
547 	default:
548 		dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
549 		return -EINVAL;
550 	}
551 	if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
552 		if (!track->db_depth_size_idx) {
553 			dev_warn(p->dev, "z/stencil buffer size not set\n");
554 			return -EINVAL;
555 		}
556 		tmp = radeon_bo_size(track->db_bo) - track->db_offset;
557 		tmp = (tmp / bpe) >> 6;
558 		if (!tmp) {
559 			dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
560 					track->db_depth_size, bpe, track->db_offset,
561 					radeon_bo_size(track->db_bo));
562 			return -EINVAL;
563 		}
564 		ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
565 	} else {
566 		/* pitch in pixels */
567 		pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
568 		slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
569 		slice_tile_max *= 64;
570 		height = slice_tile_max / pitch;
571 		if (height > 8192)
572 			height = 8192;
573 		base_offset = track->db_bo_mc + track->db_offset;
574 		array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
575 		array_check.array_mode = array_mode;
576 		array_check.group_size = track->group_size;
577 		array_check.nbanks = track->nbanks;
578 		array_check.npipes = track->npipes;
579 		array_check.nsamples = track->nsamples;
580 		array_check.blocksize = bpe;
581 		if (r600_get_array_mode_alignment(&array_check,
582 					&pitch_align, &height_align, &depth_align, &base_align)) {
583 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
584 					G_028010_ARRAY_MODE(track->db_depth_info),
585 					track->db_depth_info);
586 			return -EINVAL;
587 		}
588 		switch (array_mode) {
589 		case V_028010_ARRAY_1D_TILED_THIN1:
590 			/* don't break userspace */
591 			height &= ~0x7;
592 			break;
593 		case V_028010_ARRAY_2D_TILED_THIN1:
594 			break;
595 		default:
596 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
597 					G_028010_ARRAY_MODE(track->db_depth_info),
598 					track->db_depth_info);
599 			return -EINVAL;
600 		}
601 
602 		if (!IS_ALIGNED(pitch, pitch_align)) {
603 			dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
604 					__func__, __LINE__, pitch, pitch_align, array_mode);
605 			return -EINVAL;
606 		}
607 		if (!IS_ALIGNED(height, height_align)) {
608 			dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
609 					__func__, __LINE__, height, height_align, array_mode);
610 			return -EINVAL;
611 		}
612 		if (!IS_ALIGNED(base_offset, base_align)) {
613 			dev_warn(p->dev, "%s offset 0x%llx, 0x%llx, %d not aligned\n", __func__,
614 					base_offset, base_align, array_mode);
615 			return -EINVAL;
616 		}
617 
618 		ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
619 		nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
620 		tmp = ntiles * bpe * 64 * nviews * track->nsamples;
621 		if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
622 			dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
623 					array_mode,
624 					track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
625 					radeon_bo_size(track->db_bo));
626 			return -EINVAL;
627 		}
628 	}
629 
630 	/* hyperz */
631 	if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
632 		unsigned long size;
633 		unsigned nbx, nby;
634 
635 		if (track->htile_bo == NULL) {
636 			dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
637 				 __func__, __LINE__, track->db_depth_info);
638 			return -EINVAL;
639 		}
640 		if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
641 			dev_warn(p->dev, "%s:%d htile can't be enabled with bogus db_depth_size 0x%08x\n",
642 				 __func__, __LINE__, track->db_depth_size);
643 			return -EINVAL;
644 		}
645 
646 		nbx = pitch;
647 		nby = height;
648 		if (G_028D24_LINEAR(track->htile_surface)) {
649 			/* nbx must be 16 htiles aligned == 16 * 8 pixel aligned */
650 			nbx = round_up(nbx, 16 * 8);
651 			/* nby is npipes htiles aligned == npipes * 8 pixel aligned */
652 			nby = round_up(nby, track->npipes * 8);
653 		} else {
654 			/* always assume 8x8 htile */
655 			/* align is htile align * 8, htile align vary according to
656 			 * number of pipe and tile width and nby
657 			 */
658 			switch (track->npipes) {
659 			case 8:
660 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
661 				nbx = round_up(nbx, 64 * 8);
662 				nby = round_up(nby, 64 * 8);
663 				break;
664 			case 4:
665 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
666 				nbx = round_up(nbx, 64 * 8);
667 				nby = round_up(nby, 32 * 8);
668 				break;
669 			case 2:
670 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
671 				nbx = round_up(nbx, 32 * 8);
672 				nby = round_up(nby, 32 * 8);
673 				break;
674 			case 1:
675 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
676 				nbx = round_up(nbx, 32 * 8);
677 				nby = round_up(nby, 16 * 8);
678 				break;
679 			default:
680 				dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
681 					 __func__, __LINE__, track->npipes);
682 				return -EINVAL;
683 			}
684 		}
685 		/* compute number of htile */
686 		nbx = nbx >> 3;
687 		nby = nby >> 3;
688 		/* size must be aligned on npipes * 2K boundary */
689 		size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
690 		size += track->htile_offset;
691 
692 		if (size > radeon_bo_size(track->htile_bo)) {
693 			dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
694 				 __func__, __LINE__, radeon_bo_size(track->htile_bo),
695 				 size, nbx, nby);
696 			return -EINVAL;
697 		}
698 	}
699 
700 	track->db_dirty = false;
701 	return 0;
702 }
703 
704 static int r600_cs_track_check(struct radeon_cs_parser *p)
705 {
706 	struct r600_cs_track *track = p->track;
707 	u32 tmp;
708 	int r, i;
709 
710 	/* on legacy kernel we don't perform advanced check */
711 	if (p->rdev == NULL)
712 		return 0;
713 
714 	/* check streamout */
715 	if (track->streamout_dirty && track->vgt_strmout_en) {
716 		for (i = 0; i < 4; i++) {
717 			if (track->vgt_strmout_buffer_en & (1 << i)) {
718 				if (track->vgt_strmout_bo[i]) {
719 					u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
720 						(u64)track->vgt_strmout_size[i];
721 					if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
722 						DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
723 							  i, offset,
724 							  radeon_bo_size(track->vgt_strmout_bo[i]));
725 						return -EINVAL;
726 					}
727 				} else {
728 					dev_warn(p->dev, "No buffer for streamout %d\n", i);
729 					return -EINVAL;
730 				}
731 			}
732 		}
733 		track->streamout_dirty = false;
734 	}
735 
736 	if (track->sx_misc_kill_all_prims)
737 		return 0;
738 
739 	/* check that we have a cb for each enabled target, we don't check
740 	 * shader_mask because it seems mesa isn't always setting it :(
741 	 */
742 	if (track->cb_dirty) {
743 		tmp = track->cb_target_mask;
744 
745 		/* We must check both colorbuffers for RESOLVE. */
746 		if (track->is_resolve) {
747 			tmp |= 0xff;
748 		}
749 
750 		for (i = 0; i < 8; i++) {
751 			u32 format = G_0280A0_FORMAT(track->cb_color_info[i]);
752 
753 			if (format != V_0280A0_COLOR_INVALID &&
754 			    (tmp >> (i * 4)) & 0xF) {
755 				/* at least one component is enabled */
756 				if (track->cb_color_bo[i] == NULL) {
757 					dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
758 						__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
759 					return -EINVAL;
760 				}
761 				/* perform rewrite of CB_COLOR[0-7]_SIZE */
762 				r = r600_cs_track_validate_cb(p, i);
763 				if (r)
764 					return r;
765 			}
766 		}
767 		track->cb_dirty = false;
768 	}
769 
770 	/* Check depth buffer */
771 	if (track->db_dirty &&
772 	    G_028010_FORMAT(track->db_depth_info) != V_028010_DEPTH_INVALID &&
773 	    (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
774 	     G_028800_Z_ENABLE(track->db_depth_control))) {
775 		r = r600_cs_track_validate_db(p);
776 		if (r)
777 			return r;
778 	}
779 
780 	return 0;
781 }
782 
783 /**
784  * r600_cs_packet_parse_vline() - parse userspace VLINE packet
785  * @parser:		parser structure holding parsing context.
786  *
787  * This is an R600-specific function for parsing VLINE packets.
788  * Real work is done by r600_cs_common_vline_parse function.
789  * Here we just set up ASIC-specific register table and call
790  * the common implementation function.
791  */
792 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
793 {
794 	static uint32_t vline_start_end[2] = {AVIVO_D1MODE_VLINE_START_END,
795 					      AVIVO_D2MODE_VLINE_START_END};
796 	static uint32_t vline_status[2] = {AVIVO_D1MODE_VLINE_STATUS,
797 					   AVIVO_D2MODE_VLINE_STATUS};
798 
799 	return r600_cs_common_vline_parse(p, vline_start_end, vline_status);
800 }
801 
802 /**
803  * r600_cs_common_vline_parse() - common vline parser
804  * @parser:		parser structure holding parsing context.
805  * @vline_start_end:    table of vline_start_end registers
806  * @vline_status:       table of vline_status registers
807  *
808  * Userspace sends a special sequence for VLINE waits.
809  * PACKET0 - VLINE_START_END + value
810  * PACKET3 - WAIT_REG_MEM poll vline status reg
811  * RELOC (P3) - crtc_id in reloc.
812  *
813  * This function parses this and relocates the VLINE START END
814  * and WAIT_REG_MEM packets to the correct crtc.
815  * It also detects a switched off crtc and nulls out the
816  * wait in that case. This function is common for all ASICs that
817  * are R600 and newer. The parsing algorithm is the same, and only
818  * differs in which registers are used.
819  *
820  * Caller is the ASIC-specific function which passes the parser
821  * context and ASIC-specific register table
822  */
823 int r600_cs_common_vline_parse(struct radeon_cs_parser *p,
824 			       uint32_t *vline_start_end,
825 			       uint32_t *vline_status)
826 {
827 	struct drm_crtc *crtc;
828 	struct radeon_crtc *radeon_crtc;
829 	struct radeon_cs_packet p3reloc, wait_reg_mem;
830 	int crtc_id;
831 	int r;
832 	uint32_t header, h_idx, reg, wait_reg_mem_info;
833 	volatile uint32_t *ib;
834 
835 	ib = p->ib.ptr;
836 
837 	/* parse the WAIT_REG_MEM */
838 	r = radeon_cs_packet_parse(p, &wait_reg_mem, p->idx);
839 	if (r)
840 		return r;
841 
842 	/* check its a WAIT_REG_MEM */
843 	if (wait_reg_mem.type != RADEON_PACKET_TYPE3 ||
844 	    wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
845 		DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
846 		return -EINVAL;
847 	}
848 
849 	wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
850 	/* bit 4 is reg (0) or mem (1) */
851 	if (wait_reg_mem_info & 0x10) {
852 		DRM_ERROR("vline WAIT_REG_MEM waiting on MEM instead of REG\n");
853 		return -EINVAL;
854 	}
855 	/* bit 8 is me (0) or pfp (1) */
856 	if (wait_reg_mem_info & 0x100) {
857 		DRM_ERROR("vline WAIT_REG_MEM waiting on PFP instead of ME\n");
858 		return -EINVAL;
859 	}
860 	/* waiting for value to be equal */
861 	if ((wait_reg_mem_info & 0x7) != 0x3) {
862 		DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
863 		return -EINVAL;
864 	}
865 	if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != vline_status[0]) {
866 		DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
867 		return -EINVAL;
868 	}
869 
870 	if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != RADEON_VLINE_STAT) {
871 		DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
872 		return -EINVAL;
873 	}
874 
875 	/* jump over the NOP */
876 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
877 	if (r)
878 		return r;
879 
880 	h_idx = p->idx - 2;
881 	p->idx += wait_reg_mem.count + 2;
882 	p->idx += p3reloc.count + 2;
883 
884 	header = radeon_get_ib_value(p, h_idx);
885 	crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
886 	reg = R600_CP_PACKET0_GET_REG(header);
887 
888 	crtc = drm_crtc_find(p->rdev->ddev, p->filp, crtc_id);
889 	if (!crtc) {
890 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
891 		return -ENOENT;
892 	}
893 	radeon_crtc = to_radeon_crtc(crtc);
894 	crtc_id = radeon_crtc->crtc_id;
895 
896 	if (!crtc->enabled) {
897 		/* CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
898 		ib[h_idx + 2] = PACKET2(0);
899 		ib[h_idx + 3] = PACKET2(0);
900 		ib[h_idx + 4] = PACKET2(0);
901 		ib[h_idx + 5] = PACKET2(0);
902 		ib[h_idx + 6] = PACKET2(0);
903 		ib[h_idx + 7] = PACKET2(0);
904 		ib[h_idx + 8] = PACKET2(0);
905 	} else if (reg == vline_start_end[0]) {
906 		header &= ~R600_CP_PACKET0_REG_MASK;
907 		header |= vline_start_end[crtc_id] >> 2;
908 		ib[h_idx] = header;
909 		ib[h_idx + 4] = vline_status[crtc_id] >> 2;
910 	} else {
911 		DRM_ERROR("unknown crtc reloc\n");
912 		return -EINVAL;
913 	}
914 	return 0;
915 }
916 
917 static int r600_packet0_check(struct radeon_cs_parser *p,
918 				struct radeon_cs_packet *pkt,
919 				unsigned idx, unsigned reg)
920 {
921 	int r;
922 
923 	switch (reg) {
924 	case AVIVO_D1MODE_VLINE_START_END:
925 		r = r600_cs_packet_parse_vline(p);
926 		if (r) {
927 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
928 					idx, reg);
929 			return r;
930 		}
931 		break;
932 	default:
933 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
934 		return -EINVAL;
935 	}
936 	return 0;
937 }
938 
939 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
940 				struct radeon_cs_packet *pkt)
941 {
942 	unsigned reg, i;
943 	unsigned idx;
944 	int r;
945 
946 	idx = pkt->idx + 1;
947 	reg = pkt->reg;
948 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
949 		r = r600_packet0_check(p, pkt, idx, reg);
950 		if (r) {
951 			return r;
952 		}
953 	}
954 	return 0;
955 }
956 
957 /**
958  * r600_cs_check_reg() - check if register is authorized or not
959  * @parser: parser structure holding parsing context
960  * @reg: register we are testing
961  * @idx: index into the cs buffer
962  *
963  * This function will test against r600_reg_safe_bm and return 0
964  * if register is safe. If register is not flag as safe this function
965  * will test it against a list of register needind special handling.
966  */
967 static int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
968 {
969 	struct r600_cs_track *track = (struct r600_cs_track *)p->track;
970 	struct radeon_bo_list *reloc;
971 	u32 m, i, tmp, *ib;
972 	int r;
973 
974 	i = (reg >> 7);
975 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
976 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
977 		return -EINVAL;
978 	}
979 	m = 1 << ((reg >> 2) & 31);
980 	if (!(r600_reg_safe_bm[i] & m))
981 		return 0;
982 	ib = p->ib.ptr;
983 	switch (reg) {
984 	/* force following reg to 0 in an attempt to disable out buffer
985 	 * which will need us to better understand how it works to perform
986 	 * security check on it (Jerome)
987 	 */
988 	case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
989 	case R_008C44_SQ_ESGS_RING_SIZE:
990 	case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
991 	case R_008C54_SQ_ESTMP_RING_SIZE:
992 	case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
993 	case R_008C74_SQ_FBUF_RING_SIZE:
994 	case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
995 	case R_008C5C_SQ_GSTMP_RING_SIZE:
996 	case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
997 	case R_008C4C_SQ_GSVS_RING_SIZE:
998 	case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
999 	case R_008C6C_SQ_PSTMP_RING_SIZE:
1000 	case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
1001 	case R_008C7C_SQ_REDUC_RING_SIZE:
1002 	case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
1003 	case R_008C64_SQ_VSTMP_RING_SIZE:
1004 	case R_0288C8_SQ_GS_VERT_ITEMSIZE:
1005 		/* get value to populate the IB don't remove */
1006 		/*tmp =radeon_get_ib_value(p, idx);
1007 		  ib[idx] = 0;*/
1008 		break;
1009 	case SQ_ESGS_RING_BASE:
1010 	case SQ_GSVS_RING_BASE:
1011 	case SQ_ESTMP_RING_BASE:
1012 	case SQ_GSTMP_RING_BASE:
1013 	case SQ_PSTMP_RING_BASE:
1014 	case SQ_VSTMP_RING_BASE:
1015 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1016 		if (r) {
1017 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1018 					"0x%04X\n", reg);
1019 			return -EINVAL;
1020 		}
1021 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1022 		break;
1023 	case SQ_CONFIG:
1024 		track->sq_config = radeon_get_ib_value(p, idx);
1025 		break;
1026 	case R_028800_DB_DEPTH_CONTROL:
1027 		track->db_depth_control = radeon_get_ib_value(p, idx);
1028 		track->db_dirty = true;
1029 		break;
1030 	case R_028010_DB_DEPTH_INFO:
1031 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1032 		    radeon_cs_packet_next_is_pkt3_nop(p)) {
1033 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1034 			if (r) {
1035 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1036 					 "0x%04X\n", reg);
1037 				return -EINVAL;
1038 			}
1039 			track->db_depth_info = radeon_get_ib_value(p, idx);
1040 			ib[idx] &= C_028010_ARRAY_MODE;
1041 			track->db_depth_info &= C_028010_ARRAY_MODE;
1042 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1043 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1044 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1045 			} else {
1046 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1047 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1048 			}
1049 		} else {
1050 			track->db_depth_info = radeon_get_ib_value(p, idx);
1051 		}
1052 		track->db_dirty = true;
1053 		break;
1054 	case R_028004_DB_DEPTH_VIEW:
1055 		track->db_depth_view = radeon_get_ib_value(p, idx);
1056 		track->db_dirty = true;
1057 		break;
1058 	case R_028000_DB_DEPTH_SIZE:
1059 		track->db_depth_size = radeon_get_ib_value(p, idx);
1060 		track->db_depth_size_idx = idx;
1061 		track->db_dirty = true;
1062 		break;
1063 	case R_028AB0_VGT_STRMOUT_EN:
1064 		track->vgt_strmout_en = radeon_get_ib_value(p, idx);
1065 		track->streamout_dirty = true;
1066 		break;
1067 	case R_028B20_VGT_STRMOUT_BUFFER_EN:
1068 		track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
1069 		track->streamout_dirty = true;
1070 		break;
1071 	case VGT_STRMOUT_BUFFER_BASE_0:
1072 	case VGT_STRMOUT_BUFFER_BASE_1:
1073 	case VGT_STRMOUT_BUFFER_BASE_2:
1074 	case VGT_STRMOUT_BUFFER_BASE_3:
1075 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1076 		if (r) {
1077 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1078 					"0x%04X\n", reg);
1079 			return -EINVAL;
1080 		}
1081 		tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1082 		track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1083 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1084 		track->vgt_strmout_bo[tmp] = reloc->robj;
1085 		track->vgt_strmout_bo_mc[tmp] = reloc->gpu_offset;
1086 		track->streamout_dirty = true;
1087 		break;
1088 	case VGT_STRMOUT_BUFFER_SIZE_0:
1089 	case VGT_STRMOUT_BUFFER_SIZE_1:
1090 	case VGT_STRMOUT_BUFFER_SIZE_2:
1091 	case VGT_STRMOUT_BUFFER_SIZE_3:
1092 		tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1093 		/* size in register is DWs, convert to bytes */
1094 		track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1095 		track->streamout_dirty = true;
1096 		break;
1097 	case CP_COHER_BASE:
1098 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1099 		if (r) {
1100 			dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1101 					"0x%04X\n", reg);
1102 			return -EINVAL;
1103 		}
1104 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1105 		break;
1106 	case R_028238_CB_TARGET_MASK:
1107 		track->cb_target_mask = radeon_get_ib_value(p, idx);
1108 		track->cb_dirty = true;
1109 		break;
1110 	case R_02823C_CB_SHADER_MASK:
1111 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
1112 		break;
1113 	case R_028C04_PA_SC_AA_CONFIG:
1114 		tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
1115 		track->log_nsamples = tmp;
1116 		track->nsamples = 1 << tmp;
1117 		track->cb_dirty = true;
1118 		break;
1119 	case R_028808_CB_COLOR_CONTROL:
1120 		tmp = G_028808_SPECIAL_OP(radeon_get_ib_value(p, idx));
1121 		track->is_resolve = tmp == V_028808_SPECIAL_RESOLVE_BOX;
1122 		track->cb_dirty = true;
1123 		break;
1124 	case R_0280A0_CB_COLOR0_INFO:
1125 	case R_0280A4_CB_COLOR1_INFO:
1126 	case R_0280A8_CB_COLOR2_INFO:
1127 	case R_0280AC_CB_COLOR3_INFO:
1128 	case R_0280B0_CB_COLOR4_INFO:
1129 	case R_0280B4_CB_COLOR5_INFO:
1130 	case R_0280B8_CB_COLOR6_INFO:
1131 	case R_0280BC_CB_COLOR7_INFO:
1132 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1133 		     radeon_cs_packet_next_is_pkt3_nop(p)) {
1134 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1135 			if (r) {
1136 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1137 				return -EINVAL;
1138 			}
1139 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1140 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1141 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1142 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1143 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1144 			} else if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1145 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1146 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1147 			}
1148 		} else {
1149 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1150 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1151 		}
1152 		track->cb_dirty = true;
1153 		break;
1154 	case R_028080_CB_COLOR0_VIEW:
1155 	case R_028084_CB_COLOR1_VIEW:
1156 	case R_028088_CB_COLOR2_VIEW:
1157 	case R_02808C_CB_COLOR3_VIEW:
1158 	case R_028090_CB_COLOR4_VIEW:
1159 	case R_028094_CB_COLOR5_VIEW:
1160 	case R_028098_CB_COLOR6_VIEW:
1161 	case R_02809C_CB_COLOR7_VIEW:
1162 		tmp = (reg - R_028080_CB_COLOR0_VIEW) / 4;
1163 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1164 		track->cb_dirty = true;
1165 		break;
1166 	case R_028060_CB_COLOR0_SIZE:
1167 	case R_028064_CB_COLOR1_SIZE:
1168 	case R_028068_CB_COLOR2_SIZE:
1169 	case R_02806C_CB_COLOR3_SIZE:
1170 	case R_028070_CB_COLOR4_SIZE:
1171 	case R_028074_CB_COLOR5_SIZE:
1172 	case R_028078_CB_COLOR6_SIZE:
1173 	case R_02807C_CB_COLOR7_SIZE:
1174 		tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1175 		track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1176 		track->cb_color_size_idx[tmp] = idx;
1177 		track->cb_dirty = true;
1178 		break;
1179 		/* This register were added late, there is userspace
1180 		 * which does provide relocation for those but set
1181 		 * 0 offset. In order to avoid breaking old userspace
1182 		 * we detect this and set address to point to last
1183 		 * CB_COLOR0_BASE, note that if userspace doesn't set
1184 		 * CB_COLOR0_BASE before this register we will report
1185 		 * error. Old userspace always set CB_COLOR0_BASE
1186 		 * before any of this.
1187 		 */
1188 	case R_0280E0_CB_COLOR0_FRAG:
1189 	case R_0280E4_CB_COLOR1_FRAG:
1190 	case R_0280E8_CB_COLOR2_FRAG:
1191 	case R_0280EC_CB_COLOR3_FRAG:
1192 	case R_0280F0_CB_COLOR4_FRAG:
1193 	case R_0280F4_CB_COLOR5_FRAG:
1194 	case R_0280F8_CB_COLOR6_FRAG:
1195 	case R_0280FC_CB_COLOR7_FRAG:
1196 		tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1197 		if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1198 			if (!track->cb_color_base_last[tmp]) {
1199 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1200 				return -EINVAL;
1201 			}
1202 			track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1203 			track->cb_color_frag_offset[tmp] = track->cb_color_bo_offset[tmp];
1204 			ib[idx] = track->cb_color_base_last[tmp];
1205 		} else {
1206 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1207 			if (r) {
1208 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1209 				return -EINVAL;
1210 			}
1211 			track->cb_color_frag_bo[tmp] = reloc->robj;
1212 			track->cb_color_frag_offset[tmp] = (u64)ib[idx] << 8;
1213 			ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1214 		}
1215 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1216 			track->cb_dirty = true;
1217 		}
1218 		break;
1219 	case R_0280C0_CB_COLOR0_TILE:
1220 	case R_0280C4_CB_COLOR1_TILE:
1221 	case R_0280C8_CB_COLOR2_TILE:
1222 	case R_0280CC_CB_COLOR3_TILE:
1223 	case R_0280D0_CB_COLOR4_TILE:
1224 	case R_0280D4_CB_COLOR5_TILE:
1225 	case R_0280D8_CB_COLOR6_TILE:
1226 	case R_0280DC_CB_COLOR7_TILE:
1227 		tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1228 		if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1229 			if (!track->cb_color_base_last[tmp]) {
1230 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1231 				return -EINVAL;
1232 			}
1233 			track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1234 			track->cb_color_tile_offset[tmp] = track->cb_color_bo_offset[tmp];
1235 			ib[idx] = track->cb_color_base_last[tmp];
1236 		} else {
1237 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1238 			if (r) {
1239 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1240 				return -EINVAL;
1241 			}
1242 			track->cb_color_tile_bo[tmp] = reloc->robj;
1243 			track->cb_color_tile_offset[tmp] = (u64)ib[idx] << 8;
1244 			ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1245 		}
1246 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1247 			track->cb_dirty = true;
1248 		}
1249 		break;
1250 	case R_028100_CB_COLOR0_MASK:
1251 	case R_028104_CB_COLOR1_MASK:
1252 	case R_028108_CB_COLOR2_MASK:
1253 	case R_02810C_CB_COLOR3_MASK:
1254 	case R_028110_CB_COLOR4_MASK:
1255 	case R_028114_CB_COLOR5_MASK:
1256 	case R_028118_CB_COLOR6_MASK:
1257 	case R_02811C_CB_COLOR7_MASK:
1258 		tmp = (reg - R_028100_CB_COLOR0_MASK) / 4;
1259 		track->cb_color_mask[tmp] = radeon_get_ib_value(p, idx);
1260 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1261 			track->cb_dirty = true;
1262 		}
1263 		break;
1264 	case CB_COLOR0_BASE:
1265 	case CB_COLOR1_BASE:
1266 	case CB_COLOR2_BASE:
1267 	case CB_COLOR3_BASE:
1268 	case CB_COLOR4_BASE:
1269 	case CB_COLOR5_BASE:
1270 	case CB_COLOR6_BASE:
1271 	case CB_COLOR7_BASE:
1272 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1273 		if (r) {
1274 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1275 					"0x%04X\n", reg);
1276 			return -EINVAL;
1277 		}
1278 		tmp = (reg - CB_COLOR0_BASE) / 4;
1279 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1280 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1281 		track->cb_color_base_last[tmp] = ib[idx];
1282 		track->cb_color_bo[tmp] = reloc->robj;
1283 		track->cb_color_bo_mc[tmp] = reloc->gpu_offset;
1284 		track->cb_dirty = true;
1285 		break;
1286 	case DB_DEPTH_BASE:
1287 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1288 		if (r) {
1289 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1290 					"0x%04X\n", reg);
1291 			return -EINVAL;
1292 		}
1293 		track->db_offset = radeon_get_ib_value(p, idx) << 8;
1294 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1295 		track->db_bo = reloc->robj;
1296 		track->db_bo_mc = reloc->gpu_offset;
1297 		track->db_dirty = true;
1298 		break;
1299 	case DB_HTILE_DATA_BASE:
1300 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1301 		if (r) {
1302 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1303 					"0x%04X\n", reg);
1304 			return -EINVAL;
1305 		}
1306 		track->htile_offset = radeon_get_ib_value(p, idx) << 8;
1307 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1308 		track->htile_bo = reloc->robj;
1309 		track->db_dirty = true;
1310 		break;
1311 	case DB_HTILE_SURFACE:
1312 		track->htile_surface = radeon_get_ib_value(p, idx);
1313 		/* force 8x8 htile width and height */
1314 		ib[idx] |= 3;
1315 		track->db_dirty = true;
1316 		break;
1317 	case SQ_PGM_START_FS:
1318 	case SQ_PGM_START_ES:
1319 	case SQ_PGM_START_VS:
1320 	case SQ_PGM_START_GS:
1321 	case SQ_PGM_START_PS:
1322 	case SQ_ALU_CONST_CACHE_GS_0:
1323 	case SQ_ALU_CONST_CACHE_GS_1:
1324 	case SQ_ALU_CONST_CACHE_GS_2:
1325 	case SQ_ALU_CONST_CACHE_GS_3:
1326 	case SQ_ALU_CONST_CACHE_GS_4:
1327 	case SQ_ALU_CONST_CACHE_GS_5:
1328 	case SQ_ALU_CONST_CACHE_GS_6:
1329 	case SQ_ALU_CONST_CACHE_GS_7:
1330 	case SQ_ALU_CONST_CACHE_GS_8:
1331 	case SQ_ALU_CONST_CACHE_GS_9:
1332 	case SQ_ALU_CONST_CACHE_GS_10:
1333 	case SQ_ALU_CONST_CACHE_GS_11:
1334 	case SQ_ALU_CONST_CACHE_GS_12:
1335 	case SQ_ALU_CONST_CACHE_GS_13:
1336 	case SQ_ALU_CONST_CACHE_GS_14:
1337 	case SQ_ALU_CONST_CACHE_GS_15:
1338 	case SQ_ALU_CONST_CACHE_PS_0:
1339 	case SQ_ALU_CONST_CACHE_PS_1:
1340 	case SQ_ALU_CONST_CACHE_PS_2:
1341 	case SQ_ALU_CONST_CACHE_PS_3:
1342 	case SQ_ALU_CONST_CACHE_PS_4:
1343 	case SQ_ALU_CONST_CACHE_PS_5:
1344 	case SQ_ALU_CONST_CACHE_PS_6:
1345 	case SQ_ALU_CONST_CACHE_PS_7:
1346 	case SQ_ALU_CONST_CACHE_PS_8:
1347 	case SQ_ALU_CONST_CACHE_PS_9:
1348 	case SQ_ALU_CONST_CACHE_PS_10:
1349 	case SQ_ALU_CONST_CACHE_PS_11:
1350 	case SQ_ALU_CONST_CACHE_PS_12:
1351 	case SQ_ALU_CONST_CACHE_PS_13:
1352 	case SQ_ALU_CONST_CACHE_PS_14:
1353 	case SQ_ALU_CONST_CACHE_PS_15:
1354 	case SQ_ALU_CONST_CACHE_VS_0:
1355 	case SQ_ALU_CONST_CACHE_VS_1:
1356 	case SQ_ALU_CONST_CACHE_VS_2:
1357 	case SQ_ALU_CONST_CACHE_VS_3:
1358 	case SQ_ALU_CONST_CACHE_VS_4:
1359 	case SQ_ALU_CONST_CACHE_VS_5:
1360 	case SQ_ALU_CONST_CACHE_VS_6:
1361 	case SQ_ALU_CONST_CACHE_VS_7:
1362 	case SQ_ALU_CONST_CACHE_VS_8:
1363 	case SQ_ALU_CONST_CACHE_VS_9:
1364 	case SQ_ALU_CONST_CACHE_VS_10:
1365 	case SQ_ALU_CONST_CACHE_VS_11:
1366 	case SQ_ALU_CONST_CACHE_VS_12:
1367 	case SQ_ALU_CONST_CACHE_VS_13:
1368 	case SQ_ALU_CONST_CACHE_VS_14:
1369 	case SQ_ALU_CONST_CACHE_VS_15:
1370 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1371 		if (r) {
1372 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1373 					"0x%04X\n", reg);
1374 			return -EINVAL;
1375 		}
1376 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1377 		break;
1378 	case SX_MEMORY_EXPORT_BASE:
1379 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1380 		if (r) {
1381 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1382 					"0x%04X\n", reg);
1383 			return -EINVAL;
1384 		}
1385 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1386 		break;
1387 	case SX_MISC:
1388 		track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1389 		break;
1390 	default:
1391 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1392 		return -EINVAL;
1393 	}
1394 	return 0;
1395 }
1396 
1397 unsigned r600_mip_minify(unsigned size, unsigned level)
1398 {
1399 	unsigned val;
1400 
1401 	val = max(1U, size >> level);
1402 	if (level > 0)
1403 		val = roundup_pow_of_two(val);
1404 	return val;
1405 }
1406 
1407 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1408 			      unsigned w0, unsigned h0, unsigned d0, unsigned nsamples, unsigned format,
1409 			      unsigned block_align, unsigned height_align, unsigned base_align,
1410 			      unsigned *l0_size, unsigned *mipmap_size)
1411 {
1412 	unsigned offset, i, level;
1413 	unsigned width, height, depth, size;
1414 	unsigned blocksize;
1415 	unsigned nbx, nby;
1416 	unsigned nlevels = llevel - blevel + 1;
1417 
1418 	*l0_size = -1;
1419 	blocksize = r600_fmt_get_blocksize(format);
1420 
1421 	w0 = r600_mip_minify(w0, 0);
1422 	h0 = r600_mip_minify(h0, 0);
1423 	d0 = r600_mip_minify(d0, 0);
1424 	for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1425 		width = r600_mip_minify(w0, i);
1426 		nbx = r600_fmt_get_nblocksx(format, width);
1427 
1428 		nbx = round_up(nbx, block_align);
1429 
1430 		height = r600_mip_minify(h0, i);
1431 		nby = r600_fmt_get_nblocksy(format, height);
1432 		nby = round_up(nby, height_align);
1433 
1434 		depth = r600_mip_minify(d0, i);
1435 
1436 		size = nbx * nby * blocksize * nsamples;
1437 		if (nfaces)
1438 			size *= nfaces;
1439 		else
1440 			size *= depth;
1441 
1442 		if (i == 0)
1443 			*l0_size = size;
1444 
1445 		if (i == 0 || i == 1)
1446 			offset = round_up(offset, base_align);
1447 
1448 		offset += size;
1449 	}
1450 	*mipmap_size = offset;
1451 	if (llevel == 0)
1452 		*mipmap_size = *l0_size;
1453 	if (!blevel)
1454 		*mipmap_size -= *l0_size;
1455 }
1456 
1457 /**
1458  * r600_check_texture_resource() - check if register is authorized or not
1459  * @p: parser structure holding parsing context
1460  * @idx: index into the cs buffer
1461  * @texture: texture's bo structure
1462  * @mipmap: mipmap's bo structure
1463  *
1464  * This function will check that the resource has valid field and that
1465  * the texture and mipmap bo object are big enough to cover this resource.
1466  */
1467 static int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1468 					      struct radeon_bo *texture,
1469 					      struct radeon_bo *mipmap,
1470 					      u64 base_offset,
1471 					      u64 mip_offset,
1472 					      u32 tiling_flags)
1473 {
1474 	struct r600_cs_track *track = p->track;
1475 	u32 dim, nfaces, llevel, blevel, w0, h0, d0;
1476 	u32 word0, word1, l0_size, mipmap_size, word2, word3, word4, word5;
1477 	u32 height_align, pitch, pitch_align, depth_align;
1478 	u32 barray, larray;
1479 	u64 base_align;
1480 	struct array_mode_checker array_check;
1481 	u32 format;
1482 	bool is_array;
1483 
1484 	/* on legacy kernel we don't perform advanced check */
1485 	if (p->rdev == NULL)
1486 		return 0;
1487 
1488 	/* convert to bytes */
1489 	base_offset <<= 8;
1490 	mip_offset <<= 8;
1491 
1492 	word0 = radeon_get_ib_value(p, idx + 0);
1493 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1494 		if (tiling_flags & RADEON_TILING_MACRO)
1495 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1496 		else if (tiling_flags & RADEON_TILING_MICRO)
1497 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1498 	}
1499 	word1 = radeon_get_ib_value(p, idx + 1);
1500 	word2 = radeon_get_ib_value(p, idx + 2) << 8;
1501 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1502 	word4 = radeon_get_ib_value(p, idx + 4);
1503 	word5 = radeon_get_ib_value(p, idx + 5);
1504 	dim = G_038000_DIM(word0);
1505 	w0 = G_038000_TEX_WIDTH(word0) + 1;
1506 	pitch = (G_038000_PITCH(word0) + 1) * 8;
1507 	h0 = G_038004_TEX_HEIGHT(word1) + 1;
1508 	d0 = G_038004_TEX_DEPTH(word1);
1509 	format = G_038004_DATA_FORMAT(word1);
1510 	blevel = G_038010_BASE_LEVEL(word4);
1511 	llevel = G_038014_LAST_LEVEL(word5);
1512 	/* pitch in texels */
1513 	array_check.array_mode = G_038000_TILE_MODE(word0);
1514 	array_check.group_size = track->group_size;
1515 	array_check.nbanks = track->nbanks;
1516 	array_check.npipes = track->npipes;
1517 	array_check.nsamples = 1;
1518 	array_check.blocksize = r600_fmt_get_blocksize(format);
1519 	nfaces = 1;
1520 	is_array = false;
1521 	switch (dim) {
1522 	case V_038000_SQ_TEX_DIM_1D:
1523 	case V_038000_SQ_TEX_DIM_2D:
1524 	case V_038000_SQ_TEX_DIM_3D:
1525 		break;
1526 	case V_038000_SQ_TEX_DIM_CUBEMAP:
1527 		if (p->family >= CHIP_RV770)
1528 			nfaces = 8;
1529 		else
1530 			nfaces = 6;
1531 		break;
1532 	case V_038000_SQ_TEX_DIM_1D_ARRAY:
1533 	case V_038000_SQ_TEX_DIM_2D_ARRAY:
1534 		is_array = true;
1535 		break;
1536 	case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1537 		is_array = true;
1538 		fallthrough;
1539 	case V_038000_SQ_TEX_DIM_2D_MSAA:
1540 		array_check.nsamples = 1 << llevel;
1541 		llevel = 0;
1542 		break;
1543 	default:
1544 		dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1545 		return -EINVAL;
1546 	}
1547 	if (!r600_fmt_is_valid_texture(format, p->family)) {
1548 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1549 			 __func__, __LINE__, format);
1550 		return -EINVAL;
1551 	}
1552 
1553 	if (r600_get_array_mode_alignment(&array_check,
1554 					  &pitch_align, &height_align, &depth_align, &base_align)) {
1555 		dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1556 			 __func__, __LINE__, G_038000_TILE_MODE(word0));
1557 		return -EINVAL;
1558 	}
1559 
1560 	/* XXX check height as well... */
1561 
1562 	if (!IS_ALIGNED(pitch, pitch_align)) {
1563 		dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1564 			 __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1565 		return -EINVAL;
1566 	}
1567 	if (!IS_ALIGNED(base_offset, base_align)) {
1568 		dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1569 			 __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1570 		return -EINVAL;
1571 	}
1572 	if (!IS_ALIGNED(mip_offset, base_align)) {
1573 		dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1574 			 __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1575 		return -EINVAL;
1576 	}
1577 
1578 	if (blevel > llevel) {
1579 		dev_warn(p->dev, "texture blevel %d > llevel %d\n",
1580 			 blevel, llevel);
1581 	}
1582 	if (is_array) {
1583 		barray = G_038014_BASE_ARRAY(word5);
1584 		larray = G_038014_LAST_ARRAY(word5);
1585 
1586 		nfaces = larray - barray + 1;
1587 	}
1588 	r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, array_check.nsamples, format,
1589 			  pitch_align, height_align, base_align,
1590 			  &l0_size, &mipmap_size);
1591 	/* using get ib will give us the offset into the texture bo */
1592 	if ((l0_size + word2) > radeon_bo_size(texture)) {
1593 		dev_warn(p->dev, "texture bo too small ((%d %d) (%d %d) %d %d %d -> %d have %ld)\n",
1594 			 w0, h0, pitch_align, height_align,
1595 			 array_check.array_mode, format, word2,
1596 			 l0_size, radeon_bo_size(texture));
1597 		dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1598 		return -EINVAL;
1599 	}
1600 	/* using get ib will give us the offset into the mipmap bo */
1601 	if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1602 		/*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1603 		  w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1604 	}
1605 	return 0;
1606 }
1607 
1608 static bool r600_is_safe_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1609 {
1610 	u32 m, i;
1611 
1612 	i = (reg >> 7);
1613 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
1614 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1615 		return false;
1616 	}
1617 	m = 1 << ((reg >> 2) & 31);
1618 	if (!(r600_reg_safe_bm[i] & m))
1619 		return true;
1620 	dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1621 	return false;
1622 }
1623 
1624 static int r600_packet3_check(struct radeon_cs_parser *p,
1625 				struct radeon_cs_packet *pkt)
1626 {
1627 	struct radeon_bo_list *reloc;
1628 	struct r600_cs_track *track;
1629 	volatile u32 *ib;
1630 	unsigned idx;
1631 	unsigned i;
1632 	unsigned start_reg, end_reg, reg;
1633 	int r;
1634 	u32 idx_value;
1635 
1636 	track = (struct r600_cs_track *)p->track;
1637 	ib = p->ib.ptr;
1638 	idx = pkt->idx + 1;
1639 	idx_value = radeon_get_ib_value(p, idx);
1640 
1641 	switch (pkt->opcode) {
1642 	case PACKET3_SET_PREDICATION:
1643 	{
1644 		int pred_op;
1645 		int tmp;
1646 		uint64_t offset;
1647 
1648 		if (pkt->count != 1) {
1649 			DRM_ERROR("bad SET PREDICATION\n");
1650 			return -EINVAL;
1651 		}
1652 
1653 		tmp = radeon_get_ib_value(p, idx + 1);
1654 		pred_op = (tmp >> 16) & 0x7;
1655 
1656 		/* for the clear predicate operation */
1657 		if (pred_op == 0)
1658 			return 0;
1659 
1660 		if (pred_op > 2) {
1661 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1662 			return -EINVAL;
1663 		}
1664 
1665 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1666 		if (r) {
1667 			DRM_ERROR("bad SET PREDICATION\n");
1668 			return -EINVAL;
1669 		}
1670 
1671 		offset = reloc->gpu_offset +
1672 			 (idx_value & 0xfffffff0) +
1673 			 ((u64)(tmp & 0xff) << 32);
1674 
1675 		ib[idx + 0] = offset;
1676 		ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1677 	}
1678 	break;
1679 
1680 	case PACKET3_START_3D_CMDBUF:
1681 		if (p->family >= CHIP_RV770 || pkt->count) {
1682 			DRM_ERROR("bad START_3D\n");
1683 			return -EINVAL;
1684 		}
1685 		break;
1686 	case PACKET3_CONTEXT_CONTROL:
1687 		if (pkt->count != 1) {
1688 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1689 			return -EINVAL;
1690 		}
1691 		break;
1692 	case PACKET3_INDEX_TYPE:
1693 	case PACKET3_NUM_INSTANCES:
1694 		if (pkt->count) {
1695 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1696 			return -EINVAL;
1697 		}
1698 		break;
1699 	case PACKET3_DRAW_INDEX:
1700 	{
1701 		uint64_t offset;
1702 		if (pkt->count != 3) {
1703 			DRM_ERROR("bad DRAW_INDEX\n");
1704 			return -EINVAL;
1705 		}
1706 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1707 		if (r) {
1708 			DRM_ERROR("bad DRAW_INDEX\n");
1709 			return -EINVAL;
1710 		}
1711 
1712 		offset = reloc->gpu_offset +
1713 			 idx_value +
1714 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1715 
1716 		ib[idx+0] = offset;
1717 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1718 
1719 		r = r600_cs_track_check(p);
1720 		if (r) {
1721 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1722 			return r;
1723 		}
1724 		break;
1725 	}
1726 	case PACKET3_DRAW_INDEX_AUTO:
1727 		if (pkt->count != 1) {
1728 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1729 			return -EINVAL;
1730 		}
1731 		r = r600_cs_track_check(p);
1732 		if (r) {
1733 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1734 			return r;
1735 		}
1736 		break;
1737 	case PACKET3_DRAW_INDEX_IMMD_BE:
1738 	case PACKET3_DRAW_INDEX_IMMD:
1739 		if (pkt->count < 2) {
1740 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1741 			return -EINVAL;
1742 		}
1743 		r = r600_cs_track_check(p);
1744 		if (r) {
1745 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1746 			return r;
1747 		}
1748 		break;
1749 	case PACKET3_WAIT_REG_MEM:
1750 		if (pkt->count != 5) {
1751 			DRM_ERROR("bad WAIT_REG_MEM\n");
1752 			return -EINVAL;
1753 		}
1754 		/* bit 4 is reg (0) or mem (1) */
1755 		if (idx_value & 0x10) {
1756 			uint64_t offset;
1757 
1758 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1759 			if (r) {
1760 				DRM_ERROR("bad WAIT_REG_MEM\n");
1761 				return -EINVAL;
1762 			}
1763 
1764 			offset = reloc->gpu_offset +
1765 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff0) +
1766 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1767 
1768 			ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffff0);
1769 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1770 		} else if (idx_value & 0x100) {
1771 			DRM_ERROR("cannot use PFP on REG wait\n");
1772 			return -EINVAL;
1773 		}
1774 		break;
1775 	case PACKET3_CP_DMA:
1776 	{
1777 		u32 command, size;
1778 		u64 offset, tmp;
1779 		if (pkt->count != 4) {
1780 			DRM_ERROR("bad CP DMA\n");
1781 			return -EINVAL;
1782 		}
1783 		command = radeon_get_ib_value(p, idx+4);
1784 		size = command & 0x1fffff;
1785 		if (command & PACKET3_CP_DMA_CMD_SAS) {
1786 			/* src address space is register */
1787 			DRM_ERROR("CP DMA SAS not supported\n");
1788 			return -EINVAL;
1789 		} else {
1790 			if (command & PACKET3_CP_DMA_CMD_SAIC) {
1791 				DRM_ERROR("CP DMA SAIC only supported for registers\n");
1792 				return -EINVAL;
1793 			}
1794 			/* src address space is memory */
1795 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1796 			if (r) {
1797 				DRM_ERROR("bad CP DMA SRC\n");
1798 				return -EINVAL;
1799 			}
1800 
1801 			tmp = radeon_get_ib_value(p, idx) +
1802 				((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1803 
1804 			offset = reloc->gpu_offset + tmp;
1805 
1806 			if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1807 				dev_warn(p->dev, "CP DMA src buffer too small (%llu %lu)\n",
1808 					 tmp + size, radeon_bo_size(reloc->robj));
1809 				return -EINVAL;
1810 			}
1811 
1812 			ib[idx] = offset;
1813 			ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1814 		}
1815 		if (command & PACKET3_CP_DMA_CMD_DAS) {
1816 			/* dst address space is register */
1817 			DRM_ERROR("CP DMA DAS not supported\n");
1818 			return -EINVAL;
1819 		} else {
1820 			/* dst address space is memory */
1821 			if (command & PACKET3_CP_DMA_CMD_DAIC) {
1822 				DRM_ERROR("CP DMA DAIC only supported for registers\n");
1823 				return -EINVAL;
1824 			}
1825 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1826 			if (r) {
1827 				DRM_ERROR("bad CP DMA DST\n");
1828 				return -EINVAL;
1829 			}
1830 
1831 			tmp = radeon_get_ib_value(p, idx+2) +
1832 				((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
1833 
1834 			offset = reloc->gpu_offset + tmp;
1835 
1836 			if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1837 				dev_warn(p->dev, "CP DMA dst buffer too small (%llu %lu)\n",
1838 					 tmp + size, radeon_bo_size(reloc->robj));
1839 				return -EINVAL;
1840 			}
1841 
1842 			ib[idx+2] = offset;
1843 			ib[idx+3] = upper_32_bits(offset) & 0xff;
1844 		}
1845 		break;
1846 	}
1847 	case PACKET3_SURFACE_SYNC:
1848 		if (pkt->count != 3) {
1849 			DRM_ERROR("bad SURFACE_SYNC\n");
1850 			return -EINVAL;
1851 		}
1852 		/* 0xffffffff/0x0 is flush all cache flag */
1853 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1854 		    radeon_get_ib_value(p, idx + 2) != 0) {
1855 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1856 			if (r) {
1857 				DRM_ERROR("bad SURFACE_SYNC\n");
1858 				return -EINVAL;
1859 			}
1860 			ib[idx+2] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1861 		}
1862 		break;
1863 	case PACKET3_EVENT_WRITE:
1864 		if (pkt->count != 2 && pkt->count != 0) {
1865 			DRM_ERROR("bad EVENT_WRITE\n");
1866 			return -EINVAL;
1867 		}
1868 		if (pkt->count) {
1869 			uint64_t offset;
1870 
1871 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1872 			if (r) {
1873 				DRM_ERROR("bad EVENT_WRITE\n");
1874 				return -EINVAL;
1875 			}
1876 			offset = reloc->gpu_offset +
1877 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
1878 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1879 
1880 			ib[idx+1] = offset & 0xfffffff8;
1881 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1882 		}
1883 		break;
1884 	case PACKET3_EVENT_WRITE_EOP:
1885 	{
1886 		uint64_t offset;
1887 
1888 		if (pkt->count != 4) {
1889 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1890 			return -EINVAL;
1891 		}
1892 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1893 		if (r) {
1894 			DRM_ERROR("bad EVENT_WRITE\n");
1895 			return -EINVAL;
1896 		}
1897 
1898 		offset = reloc->gpu_offset +
1899 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
1900 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1901 
1902 		ib[idx+1] = offset & 0xfffffffc;
1903 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1904 		break;
1905 	}
1906 	case PACKET3_SET_CONFIG_REG:
1907 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1908 		end_reg = 4 * pkt->count + start_reg - 4;
1909 		if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1910 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1911 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1912 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1913 			return -EINVAL;
1914 		}
1915 		for (i = 0; i < pkt->count; i++) {
1916 			reg = start_reg + (4 * i);
1917 			r = r600_cs_check_reg(p, reg, idx+1+i);
1918 			if (r)
1919 				return r;
1920 		}
1921 		break;
1922 	case PACKET3_SET_CONTEXT_REG:
1923 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1924 		end_reg = 4 * pkt->count + start_reg - 4;
1925 		if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1926 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1927 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1928 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1929 			return -EINVAL;
1930 		}
1931 		for (i = 0; i < pkt->count; i++) {
1932 			reg = start_reg + (4 * i);
1933 			r = r600_cs_check_reg(p, reg, idx+1+i);
1934 			if (r)
1935 				return r;
1936 		}
1937 		break;
1938 	case PACKET3_SET_RESOURCE:
1939 		if (pkt->count % 7) {
1940 			DRM_ERROR("bad SET_RESOURCE\n");
1941 			return -EINVAL;
1942 		}
1943 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1944 		end_reg = 4 * pkt->count + start_reg - 4;
1945 		if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1946 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
1947 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
1948 			DRM_ERROR("bad SET_RESOURCE\n");
1949 			return -EINVAL;
1950 		}
1951 		for (i = 0; i < (pkt->count / 7); i++) {
1952 			struct radeon_bo *texture, *mipmap;
1953 			u32 size, offset, base_offset, mip_offset;
1954 
1955 			switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1956 			case SQ_TEX_VTX_VALID_TEXTURE:
1957 				/* tex base */
1958 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1959 				if (r) {
1960 					DRM_ERROR("bad SET_RESOURCE\n");
1961 					return -EINVAL;
1962 				}
1963 				base_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1964 				if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1965 					if (reloc->tiling_flags & RADEON_TILING_MACRO)
1966 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1967 					else if (reloc->tiling_flags & RADEON_TILING_MICRO)
1968 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1969 				}
1970 				texture = reloc->robj;
1971 				/* tex mip base */
1972 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1973 				if (r) {
1974 					DRM_ERROR("bad SET_RESOURCE\n");
1975 					return -EINVAL;
1976 				}
1977 				mip_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1978 				mipmap = reloc->robj;
1979 				r = r600_check_texture_resource(p,  idx+(i*7)+1,
1980 								texture, mipmap,
1981 								base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1982 								mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1983 								reloc->tiling_flags);
1984 				if (r)
1985 					return r;
1986 				ib[idx+1+(i*7)+2] += base_offset;
1987 				ib[idx+1+(i*7)+3] += mip_offset;
1988 				break;
1989 			case SQ_TEX_VTX_VALID_BUFFER:
1990 			{
1991 				uint64_t offset64;
1992 				/* vtx base */
1993 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1994 				if (r) {
1995 					DRM_ERROR("bad SET_RESOURCE\n");
1996 					return -EINVAL;
1997 				}
1998 				offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
1999 				size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
2000 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2001 					/* force size to size of the buffer */
2002 					dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
2003 						 size + offset, radeon_bo_size(reloc->robj));
2004 					ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj) - offset;
2005 				}
2006 
2007 				offset64 = reloc->gpu_offset + offset;
2008 				ib[idx+1+(i*8)+0] = offset64;
2009 				ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2010 						    (upper_32_bits(offset64) & 0xff);
2011 				break;
2012 			}
2013 			case SQ_TEX_VTX_INVALID_TEXTURE:
2014 			case SQ_TEX_VTX_INVALID_BUFFER:
2015 			default:
2016 				DRM_ERROR("bad SET_RESOURCE\n");
2017 				return -EINVAL;
2018 			}
2019 		}
2020 		break;
2021 	case PACKET3_SET_ALU_CONST:
2022 		if (track->sq_config & DX9_CONSTS) {
2023 			start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
2024 			end_reg = 4 * pkt->count + start_reg - 4;
2025 			if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
2026 			    (start_reg >= PACKET3_SET_ALU_CONST_END) ||
2027 			    (end_reg >= PACKET3_SET_ALU_CONST_END)) {
2028 				DRM_ERROR("bad SET_ALU_CONST\n");
2029 				return -EINVAL;
2030 			}
2031 		}
2032 		break;
2033 	case PACKET3_SET_BOOL_CONST:
2034 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
2035 		end_reg = 4 * pkt->count + start_reg - 4;
2036 		if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
2037 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2038 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2039 			DRM_ERROR("bad SET_BOOL_CONST\n");
2040 			return -EINVAL;
2041 		}
2042 		break;
2043 	case PACKET3_SET_LOOP_CONST:
2044 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
2045 		end_reg = 4 * pkt->count + start_reg - 4;
2046 		if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
2047 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2048 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2049 			DRM_ERROR("bad SET_LOOP_CONST\n");
2050 			return -EINVAL;
2051 		}
2052 		break;
2053 	case PACKET3_SET_CTL_CONST:
2054 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
2055 		end_reg = 4 * pkt->count + start_reg - 4;
2056 		if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
2057 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2058 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2059 			DRM_ERROR("bad SET_CTL_CONST\n");
2060 			return -EINVAL;
2061 		}
2062 		break;
2063 	case PACKET3_SET_SAMPLER:
2064 		if (pkt->count % 3) {
2065 			DRM_ERROR("bad SET_SAMPLER\n");
2066 			return -EINVAL;
2067 		}
2068 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
2069 		end_reg = 4 * pkt->count + start_reg - 4;
2070 		if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
2071 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
2072 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
2073 			DRM_ERROR("bad SET_SAMPLER\n");
2074 			return -EINVAL;
2075 		}
2076 		break;
2077 	case PACKET3_STRMOUT_BASE_UPDATE:
2078 		/* RS780 and RS880 also need this */
2079 		if (p->family < CHIP_RS780) {
2080 			DRM_ERROR("STRMOUT_BASE_UPDATE only supported on 7xx\n");
2081 			return -EINVAL;
2082 		}
2083 		if (pkt->count != 1) {
2084 			DRM_ERROR("bad STRMOUT_BASE_UPDATE packet count\n");
2085 			return -EINVAL;
2086 		}
2087 		if (idx_value > 3) {
2088 			DRM_ERROR("bad STRMOUT_BASE_UPDATE index\n");
2089 			return -EINVAL;
2090 		}
2091 		{
2092 			u64 offset;
2093 
2094 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2095 			if (r) {
2096 				DRM_ERROR("bad STRMOUT_BASE_UPDATE reloc\n");
2097 				return -EINVAL;
2098 			}
2099 
2100 			if (reloc->robj != track->vgt_strmout_bo[idx_value]) {
2101 				DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo does not match\n");
2102 				return -EINVAL;
2103 			}
2104 
2105 			offset = radeon_get_ib_value(p, idx+1) << 8;
2106 			if (offset != track->vgt_strmout_bo_offset[idx_value]) {
2107 				DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo offset does not match: 0x%llx, 0x%x\n",
2108 					  offset, track->vgt_strmout_bo_offset[idx_value]);
2109 				return -EINVAL;
2110 			}
2111 
2112 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2113 				DRM_ERROR("bad STRMOUT_BASE_UPDATE bo too small: 0x%llx, 0x%lx\n",
2114 					  offset + 4, radeon_bo_size(reloc->robj));
2115 				return -EINVAL;
2116 			}
2117 			ib[idx+1] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2118 		}
2119 		break;
2120 	case PACKET3_SURFACE_BASE_UPDATE:
2121 		if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
2122 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2123 			return -EINVAL;
2124 		}
2125 		if (pkt->count) {
2126 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2127 			return -EINVAL;
2128 		}
2129 		break;
2130 	case PACKET3_STRMOUT_BUFFER_UPDATE:
2131 		if (pkt->count != 4) {
2132 			DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2133 			return -EINVAL;
2134 		}
2135 		/* Updating memory at DST_ADDRESS. */
2136 		if (idx_value & 0x1) {
2137 			u64 offset;
2138 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2139 			if (r) {
2140 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2141 				return -EINVAL;
2142 			}
2143 			offset = radeon_get_ib_value(p, idx+1);
2144 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2145 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2146 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2147 					  offset + 4, radeon_bo_size(reloc->robj));
2148 				return -EINVAL;
2149 			}
2150 			offset += reloc->gpu_offset;
2151 			ib[idx+1] = offset;
2152 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2153 		}
2154 		/* Reading data from SRC_ADDRESS. */
2155 		if (((idx_value >> 1) & 0x3) == 2) {
2156 			u64 offset;
2157 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2158 			if (r) {
2159 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2160 				return -EINVAL;
2161 			}
2162 			offset = radeon_get_ib_value(p, idx+3);
2163 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2164 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2165 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2166 					  offset + 4, radeon_bo_size(reloc->robj));
2167 				return -EINVAL;
2168 			}
2169 			offset += reloc->gpu_offset;
2170 			ib[idx+3] = offset;
2171 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2172 		}
2173 		break;
2174 	case PACKET3_MEM_WRITE:
2175 	{
2176 		u64 offset;
2177 
2178 		if (pkt->count != 3) {
2179 			DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2180 			return -EINVAL;
2181 		}
2182 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2183 		if (r) {
2184 			DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2185 			return -EINVAL;
2186 		}
2187 		offset = radeon_get_ib_value(p, idx+0);
2188 		offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2189 		if (offset & 0x7) {
2190 			DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2191 			return -EINVAL;
2192 		}
2193 		if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2194 			DRM_ERROR("bad MEM_WRITE bo too small: 0x%llx, 0x%lx\n",
2195 				  offset + 8, radeon_bo_size(reloc->robj));
2196 			return -EINVAL;
2197 		}
2198 		offset += reloc->gpu_offset;
2199 		ib[idx+0] = offset;
2200 		ib[idx+1] = upper_32_bits(offset) & 0xff;
2201 		break;
2202 	}
2203 	case PACKET3_COPY_DW:
2204 		if (pkt->count != 4) {
2205 			DRM_ERROR("bad COPY_DW (invalid count)\n");
2206 			return -EINVAL;
2207 		}
2208 		if (idx_value & 0x1) {
2209 			u64 offset;
2210 			/* SRC is memory. */
2211 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2212 			if (r) {
2213 				DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2214 				return -EINVAL;
2215 			}
2216 			offset = radeon_get_ib_value(p, idx+1);
2217 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2218 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2219 				DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2220 					  offset + 4, radeon_bo_size(reloc->robj));
2221 				return -EINVAL;
2222 			}
2223 			offset += reloc->gpu_offset;
2224 			ib[idx+1] = offset;
2225 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2226 		} else {
2227 			/* SRC is a reg. */
2228 			reg = radeon_get_ib_value(p, idx+1) << 2;
2229 			if (!r600_is_safe_reg(p, reg, idx+1))
2230 				return -EINVAL;
2231 		}
2232 		if (idx_value & 0x2) {
2233 			u64 offset;
2234 			/* DST is memory. */
2235 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2236 			if (r) {
2237 				DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2238 				return -EINVAL;
2239 			}
2240 			offset = radeon_get_ib_value(p, idx+3);
2241 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2242 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2243 				DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2244 					  offset + 4, radeon_bo_size(reloc->robj));
2245 				return -EINVAL;
2246 			}
2247 			offset += reloc->gpu_offset;
2248 			ib[idx+3] = offset;
2249 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2250 		} else {
2251 			/* DST is a reg. */
2252 			reg = radeon_get_ib_value(p, idx+3) << 2;
2253 			if (!r600_is_safe_reg(p, reg, idx+3))
2254 				return -EINVAL;
2255 		}
2256 		break;
2257 	case PACKET3_NOP:
2258 		break;
2259 	default:
2260 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2261 		return -EINVAL;
2262 	}
2263 	return 0;
2264 }
2265 
2266 int r600_cs_parse(struct radeon_cs_parser *p)
2267 {
2268 	struct radeon_cs_packet pkt;
2269 	struct r600_cs_track *track;
2270 	int r;
2271 
2272 	if (p->track == NULL) {
2273 		/* initialize tracker, we are in kms */
2274 		track = kzalloc(sizeof(*track), GFP_KERNEL);
2275 		if (track == NULL)
2276 			return -ENOMEM;
2277 		r600_cs_track_init(track);
2278 		if (p->rdev->family < CHIP_RV770) {
2279 			track->npipes = p->rdev->config.r600.tiling_npipes;
2280 			track->nbanks = p->rdev->config.r600.tiling_nbanks;
2281 			track->group_size = p->rdev->config.r600.tiling_group_size;
2282 		} else if (p->rdev->family <= CHIP_RV740) {
2283 			track->npipes = p->rdev->config.rv770.tiling_npipes;
2284 			track->nbanks = p->rdev->config.rv770.tiling_nbanks;
2285 			track->group_size = p->rdev->config.rv770.tiling_group_size;
2286 		}
2287 		p->track = track;
2288 	}
2289 	do {
2290 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2291 		if (r) {
2292 			kfree(p->track);
2293 			p->track = NULL;
2294 			return r;
2295 		}
2296 		p->idx += pkt.count + 2;
2297 		switch (pkt.type) {
2298 		case RADEON_PACKET_TYPE0:
2299 			r = r600_cs_parse_packet0(p, &pkt);
2300 			break;
2301 		case RADEON_PACKET_TYPE2:
2302 			break;
2303 		case RADEON_PACKET_TYPE3:
2304 			r = r600_packet3_check(p, &pkt);
2305 			break;
2306 		default:
2307 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2308 			kfree(p->track);
2309 			p->track = NULL;
2310 			return -EINVAL;
2311 		}
2312 		if (r) {
2313 			kfree(p->track);
2314 			p->track = NULL;
2315 			return r;
2316 		}
2317 	} while (p->idx < p->chunk_ib->length_dw);
2318 #if 0
2319 	for (r = 0; r < p->ib.length_dw; r++) {
2320 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2321 		mdelay(1);
2322 	}
2323 #endif
2324 	kfree(p->track);
2325 	p->track = NULL;
2326 	return 0;
2327 }
2328 
2329 /*
2330  *  DMA
2331  */
2332 /**
2333  * r600_dma_cs_next_reloc() - parse next reloc
2334  * @p:		parser structure holding parsing context.
2335  * @cs_reloc:		reloc informations
2336  *
2337  * Return the next reloc, do bo validation and compute
2338  * GPU offset using the provided start.
2339  **/
2340 int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
2341 			   struct radeon_bo_list **cs_reloc)
2342 {
2343 	unsigned idx;
2344 
2345 	*cs_reloc = NULL;
2346 	if (p->chunk_relocs == NULL) {
2347 		DRM_ERROR("No relocation chunk !\n");
2348 		return -EINVAL;
2349 	}
2350 	idx = p->dma_reloc_idx;
2351 	if (idx >= p->nrelocs) {
2352 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
2353 			  idx, p->nrelocs);
2354 		return -EINVAL;
2355 	}
2356 	*cs_reloc = &p->relocs[idx];
2357 	p->dma_reloc_idx++;
2358 	return 0;
2359 }
2360 
2361 #define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
2362 #define GET_DMA_COUNT(h) ((h) & 0x0000ffff)
2363 #define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
2364 
2365 /**
2366  * r600_dma_cs_parse() - parse the DMA IB
2367  * @p:		parser structure holding parsing context.
2368  *
2369  * Parses the DMA IB from the CS ioctl and updates
2370  * the GPU addresses based on the reloc information and
2371  * checks for errors. (R6xx-R7xx)
2372  * Returns 0 for success and an error on failure.
2373  **/
2374 int r600_dma_cs_parse(struct radeon_cs_parser *p)
2375 {
2376 	struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
2377 	struct radeon_bo_list *src_reloc, *dst_reloc;
2378 	u32 header, cmd, count, tiled;
2379 	volatile u32 *ib = p->ib.ptr;
2380 	u32 idx, idx_value;
2381 	u64 src_offset, dst_offset;
2382 	int r;
2383 
2384 	do {
2385 		if (p->idx >= ib_chunk->length_dw) {
2386 			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2387 				  p->idx, ib_chunk->length_dw);
2388 			return -EINVAL;
2389 		}
2390 		idx = p->idx;
2391 		header = radeon_get_ib_value(p, idx);
2392 		cmd = GET_DMA_CMD(header);
2393 		count = GET_DMA_COUNT(header);
2394 		tiled = GET_DMA_T(header);
2395 
2396 		switch (cmd) {
2397 		case DMA_PACKET_WRITE:
2398 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2399 			if (r) {
2400 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2401 				return -EINVAL;
2402 			}
2403 			if (tiled) {
2404 				dst_offset = radeon_get_ib_value(p, idx+1);
2405 				dst_offset <<= 8;
2406 
2407 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2408 				p->idx += count + 5;
2409 			} else {
2410 				dst_offset = radeon_get_ib_value(p, idx+1);
2411 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2412 
2413 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2414 				ib[idx+2] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2415 				p->idx += count + 3;
2416 			}
2417 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2418 				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
2419 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2420 				return -EINVAL;
2421 			}
2422 			break;
2423 		case DMA_PACKET_COPY:
2424 			r = r600_dma_cs_next_reloc(p, &src_reloc);
2425 			if (r) {
2426 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2427 				return -EINVAL;
2428 			}
2429 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2430 			if (r) {
2431 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2432 				return -EINVAL;
2433 			}
2434 			if (tiled) {
2435 				idx_value = radeon_get_ib_value(p, idx + 2);
2436 				/* detile bit */
2437 				if (idx_value & (1 << 31)) {
2438 					/* tiled src, linear dst */
2439 					src_offset = radeon_get_ib_value(p, idx+1);
2440 					src_offset <<= 8;
2441 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
2442 
2443 					dst_offset = radeon_get_ib_value(p, idx+5);
2444 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2445 					ib[idx+5] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2446 					ib[idx+6] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2447 				} else {
2448 					/* linear src, tiled dst */
2449 					src_offset = radeon_get_ib_value(p, idx+5);
2450 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2451 					ib[idx+5] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2452 					ib[idx+6] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2453 
2454 					dst_offset = radeon_get_ib_value(p, idx+1);
2455 					dst_offset <<= 8;
2456 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2457 				}
2458 				p->idx += 7;
2459 			} else {
2460 				if (p->family >= CHIP_RV770) {
2461 					src_offset = radeon_get_ib_value(p, idx+2);
2462 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2463 					dst_offset = radeon_get_ib_value(p, idx+1);
2464 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2465 
2466 					ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2467 					ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2468 					ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2469 					ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2470 					p->idx += 5;
2471 				} else {
2472 					src_offset = radeon_get_ib_value(p, idx+2);
2473 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2474 					dst_offset = radeon_get_ib_value(p, idx+1);
2475 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff0000)) << 16;
2476 
2477 					ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2478 					ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2479 					ib[idx+3] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2480 					ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) & 0xff) << 16;
2481 					p->idx += 4;
2482 				}
2483 			}
2484 			if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2485 				dev_warn(p->dev, "DMA copy src buffer too small (%llu %lu)\n",
2486 					 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2487 				return -EINVAL;
2488 			}
2489 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2490 				dev_warn(p->dev, "DMA write dst buffer too small (%llu %lu)\n",
2491 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2492 				return -EINVAL;
2493 			}
2494 			break;
2495 		case DMA_PACKET_CONSTANT_FILL:
2496 			if (p->family < CHIP_RV770) {
2497 				DRM_ERROR("Constant Fill is 7xx only !\n");
2498 				return -EINVAL;
2499 			}
2500 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2501 			if (r) {
2502 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2503 				return -EINVAL;
2504 			}
2505 			dst_offset = radeon_get_ib_value(p, idx+1);
2506 			dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
2507 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2508 				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
2509 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2510 				return -EINVAL;
2511 			}
2512 			ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2513 			ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) << 16) & 0x00ff0000;
2514 			p->idx += 4;
2515 			break;
2516 		case DMA_PACKET_NOP:
2517 			p->idx += 1;
2518 			break;
2519 		default:
2520 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
2521 			return -EINVAL;
2522 		}
2523 	} while (p->idx < p->chunk_ib->length_dw);
2524 #if 0
2525 	for (r = 0; r < p->ib->length_dw; r++) {
2526 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2527 		mdelay(1);
2528 	}
2529 #endif
2530 	return 0;
2531 }
2532