xref: /openbmc/linux/drivers/gpu/drm/radeon/r600_cs.c (revision 2cf1c348)
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 extern void r600_cs_legacy_get_tiling_conf(struct drm_device *dev, u32 *npipes, u32 *nbanks, u32 *group_size);
38 
39 
40 struct r600_cs_track {
41 	/* configuration we mirror so that we use same code btw kms/ums */
42 	u32			group_size;
43 	u32			nbanks;
44 	u32			npipes;
45 	/* value we track */
46 	u32			sq_config;
47 	u32			log_nsamples;
48 	u32			nsamples;
49 	u32			cb_color_base_last[8];
50 	struct radeon_bo	*cb_color_bo[8];
51 	u64			cb_color_bo_mc[8];
52 	u64			cb_color_bo_offset[8];
53 	struct radeon_bo	*cb_color_frag_bo[8];
54 	u64			cb_color_frag_offset[8];
55 	struct radeon_bo	*cb_color_tile_bo[8];
56 	u64			cb_color_tile_offset[8];
57 	u32			cb_color_mask[8];
58 	u32			cb_color_info[8];
59 	u32			cb_color_view[8];
60 	u32			cb_color_size_idx[8]; /* unused */
61 	u32			cb_target_mask;
62 	u32			cb_shader_mask;  /* unused */
63 	bool			is_resolve;
64 	u32			cb_color_size[8];
65 	u32			vgt_strmout_en;
66 	u32			vgt_strmout_buffer_en;
67 	struct radeon_bo	*vgt_strmout_bo[4];
68 	u64			vgt_strmout_bo_mc[4]; /* unused */
69 	u32			vgt_strmout_bo_offset[4];
70 	u32			vgt_strmout_size[4];
71 	u32			db_depth_control;
72 	u32			db_depth_info;
73 	u32			db_depth_size_idx;
74 	u32			db_depth_view;
75 	u32			db_depth_size;
76 	u32			db_offset;
77 	struct radeon_bo	*db_bo;
78 	u64			db_bo_mc;
79 	bool			sx_misc_kill_all_prims;
80 	bool			cb_dirty;
81 	bool			db_dirty;
82 	bool			streamout_dirty;
83 	struct radeon_bo	*htile_bo;
84 	u64			htile_offset;
85 	u32			htile_surface;
86 };
87 
88 #define FMT_8_BIT(fmt, vc)   [fmt] = { 1, 1, 1, vc, CHIP_R600 }
89 #define FMT_16_BIT(fmt, vc)  [fmt] = { 1, 1, 2, vc, CHIP_R600 }
90 #define FMT_24_BIT(fmt)      [fmt] = { 1, 1, 4,  0, CHIP_R600 }
91 #define FMT_32_BIT(fmt, vc)  [fmt] = { 1, 1, 4, vc, CHIP_R600 }
92 #define FMT_48_BIT(fmt)      [fmt] = { 1, 1, 8,  0, CHIP_R600 }
93 #define FMT_64_BIT(fmt, vc)  [fmt] = { 1, 1, 8, vc, CHIP_R600 }
94 #define FMT_96_BIT(fmt)      [fmt] = { 1, 1, 12, 0, CHIP_R600 }
95 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16,vc, CHIP_R600 }
96 
97 struct gpu_formats {
98 	unsigned blockwidth;
99 	unsigned blockheight;
100 	unsigned blocksize;
101 	unsigned valid_color;
102 	enum radeon_family min_family;
103 };
104 
105 static const struct gpu_formats color_formats_table[] = {
106 	/* 8 bit */
107 	FMT_8_BIT(V_038004_COLOR_8, 1),
108 	FMT_8_BIT(V_038004_COLOR_4_4, 1),
109 	FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
110 	FMT_8_BIT(V_038004_FMT_1, 0),
111 
112 	/* 16-bit */
113 	FMT_16_BIT(V_038004_COLOR_16, 1),
114 	FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
115 	FMT_16_BIT(V_038004_COLOR_8_8, 1),
116 	FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
117 	FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
118 	FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
119 	FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
120 	FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
121 
122 	/* 24-bit */
123 	FMT_24_BIT(V_038004_FMT_8_8_8),
124 
125 	/* 32-bit */
126 	FMT_32_BIT(V_038004_COLOR_32, 1),
127 	FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
128 	FMT_32_BIT(V_038004_COLOR_16_16, 1),
129 	FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
130 	FMT_32_BIT(V_038004_COLOR_8_24, 1),
131 	FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
132 	FMT_32_BIT(V_038004_COLOR_24_8, 1),
133 	FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
134 	FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
135 	FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
136 	FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
137 	FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
138 	FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
139 	FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
140 	FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
141 	FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
142 	FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
143 	FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
144 
145 	/* 48-bit */
146 	FMT_48_BIT(V_038004_FMT_16_16_16),
147 	FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
148 
149 	/* 64-bit */
150 	FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
151 	FMT_64_BIT(V_038004_COLOR_32_32, 1),
152 	FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
153 	FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
154 	FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
155 
156 	FMT_96_BIT(V_038004_FMT_32_32_32),
157 	FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
158 
159 	/* 128-bit */
160 	FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
161 	FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
162 
163 	[V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
164 	[V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
165 
166 	/* block compressed formats */
167 	[V_038004_FMT_BC1] = { 4, 4, 8, 0 },
168 	[V_038004_FMT_BC2] = { 4, 4, 16, 0 },
169 	[V_038004_FMT_BC3] = { 4, 4, 16, 0 },
170 	[V_038004_FMT_BC4] = { 4, 4, 8, 0 },
171 	[V_038004_FMT_BC5] = { 4, 4, 16, 0},
172 	[V_038004_FMT_BC6] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
173 	[V_038004_FMT_BC7] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
174 
175 	/* The other Evergreen formats */
176 	[V_038004_FMT_32_AS_32_32_32_32] = { 1, 1, 4, 0, CHIP_CEDAR},
177 };
178 
179 bool r600_fmt_is_valid_color(u32 format)
180 {
181 	if (format >= ARRAY_SIZE(color_formats_table))
182 		return false;
183 
184 	if (color_formats_table[format].valid_color)
185 		return true;
186 
187 	return false;
188 }
189 
190 bool r600_fmt_is_valid_texture(u32 format, enum radeon_family family)
191 {
192 	if (format >= ARRAY_SIZE(color_formats_table))
193 		return false;
194 
195 	if (family < color_formats_table[format].min_family)
196 		return false;
197 
198 	if (color_formats_table[format].blockwidth > 0)
199 		return true;
200 
201 	return false;
202 }
203 
204 int r600_fmt_get_blocksize(u32 format)
205 {
206 	if (format >= ARRAY_SIZE(color_formats_table))
207 		return 0;
208 
209 	return color_formats_table[format].blocksize;
210 }
211 
212 int r600_fmt_get_nblocksx(u32 format, u32 w)
213 {
214 	unsigned bw;
215 
216 	if (format >= ARRAY_SIZE(color_formats_table))
217 		return 0;
218 
219 	bw = color_formats_table[format].blockwidth;
220 	if (bw == 0)
221 		return 0;
222 
223 	return DIV_ROUND_UP(w, bw);
224 }
225 
226 int r600_fmt_get_nblocksy(u32 format, u32 h)
227 {
228 	unsigned bh;
229 
230 	if (format >= ARRAY_SIZE(color_formats_table))
231 		return 0;
232 
233 	bh = color_formats_table[format].blockheight;
234 	if (bh == 0)
235 		return 0;
236 
237 	return DIV_ROUND_UP(h, bh);
238 }
239 
240 struct array_mode_checker {
241 	int array_mode;
242 	u32 group_size;
243 	u32 nbanks;
244 	u32 npipes;
245 	u32 nsamples;
246 	u32 blocksize;
247 };
248 
249 /* returns alignment in pixels for pitch/height/depth and bytes for base */
250 static int r600_get_array_mode_alignment(struct array_mode_checker *values,
251 						u32 *pitch_align,
252 						u32 *height_align,
253 						u32 *depth_align,
254 						u64 *base_align)
255 {
256 	u32 tile_width = 8;
257 	u32 tile_height = 8;
258 	u32 macro_tile_width = values->nbanks;
259 	u32 macro_tile_height = values->npipes;
260 	u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
261 	u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
262 
263 	switch (values->array_mode) {
264 	case ARRAY_LINEAR_GENERAL:
265 		/* technically tile_width/_height for pitch/height */
266 		*pitch_align = 1; /* tile_width */
267 		*height_align = 1; /* tile_height */
268 		*depth_align = 1;
269 		*base_align = 1;
270 		break;
271 	case ARRAY_LINEAR_ALIGNED:
272 		*pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
273 		*height_align = 1;
274 		*depth_align = 1;
275 		*base_align = values->group_size;
276 		break;
277 	case ARRAY_1D_TILED_THIN1:
278 		*pitch_align = max((u32)tile_width,
279 				   (u32)(values->group_size /
280 					 (tile_height * values->blocksize * values->nsamples)));
281 		*height_align = tile_height;
282 		*depth_align = 1;
283 		*base_align = values->group_size;
284 		break;
285 	case ARRAY_2D_TILED_THIN1:
286 		*pitch_align = max((u32)macro_tile_width * tile_width,
287 				(u32)((values->group_size * values->nbanks) /
288 				(values->blocksize * values->nsamples * tile_width)));
289 		*height_align = macro_tile_height * tile_height;
290 		*depth_align = 1;
291 		*base_align = max(macro_tile_bytes,
292 				  (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
293 		break;
294 	default:
295 		return -EINVAL;
296 	}
297 
298 	return 0;
299 }
300 
301 static void r600_cs_track_init(struct r600_cs_track *track)
302 {
303 	int i;
304 
305 	/* assume DX9 mode */
306 	track->sq_config = DX9_CONSTS;
307 	for (i = 0; i < 8; i++) {
308 		track->cb_color_base_last[i] = 0;
309 		track->cb_color_size[i] = 0;
310 		track->cb_color_size_idx[i] = 0;
311 		track->cb_color_info[i] = 0;
312 		track->cb_color_view[i] = 0xFFFFFFFF;
313 		track->cb_color_bo[i] = NULL;
314 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
315 		track->cb_color_bo_mc[i] = 0xFFFFFFFF;
316 		track->cb_color_frag_bo[i] = NULL;
317 		track->cb_color_frag_offset[i] = 0xFFFFFFFF;
318 		track->cb_color_tile_bo[i] = NULL;
319 		track->cb_color_tile_offset[i] = 0xFFFFFFFF;
320 		track->cb_color_mask[i] = 0xFFFFFFFF;
321 	}
322 	track->is_resolve = false;
323 	track->nsamples = 16;
324 	track->log_nsamples = 4;
325 	track->cb_target_mask = 0xFFFFFFFF;
326 	track->cb_shader_mask = 0xFFFFFFFF;
327 	track->cb_dirty = true;
328 	track->db_bo = NULL;
329 	track->db_bo_mc = 0xFFFFFFFF;
330 	/* assume the biggest format and that htile is enabled */
331 	track->db_depth_info = 7 | (1 << 25);
332 	track->db_depth_view = 0xFFFFC000;
333 	track->db_depth_size = 0xFFFFFFFF;
334 	track->db_depth_size_idx = 0;
335 	track->db_depth_control = 0xFFFFFFFF;
336 	track->db_dirty = true;
337 	track->htile_bo = NULL;
338 	track->htile_offset = 0xFFFFFFFF;
339 	track->htile_surface = 0;
340 
341 	for (i = 0; i < 4; i++) {
342 		track->vgt_strmout_size[i] = 0;
343 		track->vgt_strmout_bo[i] = NULL;
344 		track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
345 		track->vgt_strmout_bo_mc[i] = 0xFFFFFFFF;
346 	}
347 	track->streamout_dirty = true;
348 	track->sx_misc_kill_all_prims = false;
349 }
350 
351 static int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
352 {
353 	struct r600_cs_track *track = p->track;
354 	u32 slice_tile_max, tmp;
355 	u32 height, height_align, pitch, pitch_align, depth_align;
356 	u64 base_offset, base_align;
357 	struct array_mode_checker array_check;
358 	volatile u32 *ib = p->ib.ptr;
359 	unsigned array_mode;
360 	u32 format;
361 	/* When resolve is used, the second colorbuffer has always 1 sample. */
362 	unsigned nsamples = track->is_resolve && i == 1 ? 1 : track->nsamples;
363 
364 	format = G_0280A0_FORMAT(track->cb_color_info[i]);
365 	if (!r600_fmt_is_valid_color(format)) {
366 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
367 			 __func__, __LINE__, format,
368 			i, track->cb_color_info[i]);
369 		return -EINVAL;
370 	}
371 	/* pitch in pixels */
372 	pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
373 	slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
374 	slice_tile_max *= 64;
375 	height = slice_tile_max / pitch;
376 	if (height > 8192)
377 		height = 8192;
378 	array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
379 
380 	base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
381 	array_check.array_mode = array_mode;
382 	array_check.group_size = track->group_size;
383 	array_check.nbanks = track->nbanks;
384 	array_check.npipes = track->npipes;
385 	array_check.nsamples = nsamples;
386 	array_check.blocksize = r600_fmt_get_blocksize(format);
387 	if (r600_get_array_mode_alignment(&array_check,
388 					  &pitch_align, &height_align, &depth_align, &base_align)) {
389 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
390 			 G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
391 			 track->cb_color_info[i]);
392 		return -EINVAL;
393 	}
394 	switch (array_mode) {
395 	case V_0280A0_ARRAY_LINEAR_GENERAL:
396 		break;
397 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
398 		break;
399 	case V_0280A0_ARRAY_1D_TILED_THIN1:
400 		/* avoid breaking userspace */
401 		if (height > 7)
402 			height &= ~0x7;
403 		break;
404 	case V_0280A0_ARRAY_2D_TILED_THIN1:
405 		break;
406 	default:
407 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
408 			G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
409 			track->cb_color_info[i]);
410 		return -EINVAL;
411 	}
412 
413 	if (!IS_ALIGNED(pitch, pitch_align)) {
414 		dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
415 			 __func__, __LINE__, pitch, pitch_align, array_mode);
416 		return -EINVAL;
417 	}
418 	if (!IS_ALIGNED(height, height_align)) {
419 		dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
420 			 __func__, __LINE__, height, height_align, array_mode);
421 		return -EINVAL;
422 	}
423 	if (!IS_ALIGNED(base_offset, base_align)) {
424 		dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
425 			 base_offset, base_align, array_mode);
426 		return -EINVAL;
427 	}
428 
429 	/* check offset */
430 	tmp = r600_fmt_get_nblocksy(format, height) * r600_fmt_get_nblocksx(format, pitch) *
431 	      r600_fmt_get_blocksize(format) * nsamples;
432 	switch (array_mode) {
433 	default:
434 	case V_0280A0_ARRAY_LINEAR_GENERAL:
435 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
436 		tmp += track->cb_color_view[i] & 0xFF;
437 		break;
438 	case V_0280A0_ARRAY_1D_TILED_THIN1:
439 	case V_0280A0_ARRAY_2D_TILED_THIN1:
440 		tmp += G_028080_SLICE_MAX(track->cb_color_view[i]) * tmp;
441 		break;
442 	}
443 	if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
444 		if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
445 			/* the initial DDX does bad things with the CB size occasionally */
446 			/* it rounds up height too far for slice tile max but the BO is smaller */
447 			/* r600c,g also seem to flush at bad times in some apps resulting in
448 			 * bogus values here. So for linear just allow anything to avoid breaking
449 			 * broken userspace.
450 			 */
451 		} else {
452 			dev_warn(p->dev, "%s offset[%d] %d %llu %d %lu too big (%d %d) (%d %d %d)\n",
453 				 __func__, i, array_mode,
454 				 track->cb_color_bo_offset[i], tmp,
455 				 radeon_bo_size(track->cb_color_bo[i]),
456 				 pitch, height, r600_fmt_get_nblocksx(format, pitch),
457 				 r600_fmt_get_nblocksy(format, height),
458 				 r600_fmt_get_blocksize(format));
459 			return -EINVAL;
460 		}
461 	}
462 	/* limit max tile */
463 	tmp = (height * pitch) >> 6;
464 	if (tmp < slice_tile_max)
465 		slice_tile_max = tmp;
466 	tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
467 		S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
468 	ib[track->cb_color_size_idx[i]] = tmp;
469 
470 	/* FMASK/CMASK */
471 	switch (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
472 	case V_0280A0_TILE_DISABLE:
473 		break;
474 	case V_0280A0_FRAG_ENABLE:
475 		if (track->nsamples > 1) {
476 			uint32_t tile_max = G_028100_FMASK_TILE_MAX(track->cb_color_mask[i]);
477 			/* the tile size is 8x8, but the size is in units of bits.
478 			 * for bytes, do just * 8. */
479 			uint32_t bytes = track->nsamples * track->log_nsamples * 8 * (tile_max + 1);
480 
481 			if (bytes + track->cb_color_frag_offset[i] >
482 			    radeon_bo_size(track->cb_color_frag_bo[i])) {
483 				dev_warn(p->dev, "%s FMASK_TILE_MAX too large "
484 					 "(tile_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
485 					 __func__, tile_max, bytes,
486 					 track->cb_color_frag_offset[i],
487 					 radeon_bo_size(track->cb_color_frag_bo[i]));
488 				return -EINVAL;
489 			}
490 		}
491 		fallthrough;
492 	case V_0280A0_CLEAR_ENABLE:
493 	{
494 		uint32_t block_max = G_028100_CMASK_BLOCK_MAX(track->cb_color_mask[i]);
495 		/* One block = 128x128 pixels, one 8x8 tile has 4 bits..
496 		 * (128*128) / (8*8) / 2 = 128 bytes per block. */
497 		uint32_t bytes = (block_max + 1) * 128;
498 
499 		if (bytes + track->cb_color_tile_offset[i] >
500 		    radeon_bo_size(track->cb_color_tile_bo[i])) {
501 			dev_warn(p->dev, "%s CMASK_BLOCK_MAX too large "
502 				 "(block_max=%u, bytes=%u, offset=%llu, bo_size=%lu)\n",
503 				 __func__, block_max, bytes,
504 				 track->cb_color_tile_offset[i],
505 				 radeon_bo_size(track->cb_color_tile_bo[i]));
506 			return -EINVAL;
507 		}
508 		break;
509 	}
510 	default:
511 		dev_warn(p->dev, "%s invalid tile mode\n", __func__);
512 		return -EINVAL;
513 	}
514 	return 0;
515 }
516 
517 static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
518 {
519 	struct r600_cs_track *track = p->track;
520 	u32 nviews, bpe, ntiles, slice_tile_max, tmp;
521 	u32 height_align, pitch_align, depth_align;
522 	u32 pitch = 8192;
523 	u32 height = 8192;
524 	u64 base_offset, base_align;
525 	struct array_mode_checker array_check;
526 	int array_mode;
527 	volatile u32 *ib = p->ib.ptr;
528 
529 
530 	if (track->db_bo == NULL) {
531 		dev_warn(p->dev, "z/stencil with no depth buffer\n");
532 		return -EINVAL;
533 	}
534 	switch (G_028010_FORMAT(track->db_depth_info)) {
535 	case V_028010_DEPTH_16:
536 		bpe = 2;
537 		break;
538 	case V_028010_DEPTH_X8_24:
539 	case V_028010_DEPTH_8_24:
540 	case V_028010_DEPTH_X8_24_FLOAT:
541 	case V_028010_DEPTH_8_24_FLOAT:
542 	case V_028010_DEPTH_32_FLOAT:
543 		bpe = 4;
544 		break;
545 	case V_028010_DEPTH_X24_8_32_FLOAT:
546 		bpe = 8;
547 		break;
548 	default:
549 		dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
550 		return -EINVAL;
551 	}
552 	if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
553 		if (!track->db_depth_size_idx) {
554 			dev_warn(p->dev, "z/stencil buffer size not set\n");
555 			return -EINVAL;
556 		}
557 		tmp = radeon_bo_size(track->db_bo) - track->db_offset;
558 		tmp = (tmp / bpe) >> 6;
559 		if (!tmp) {
560 			dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
561 					track->db_depth_size, bpe, track->db_offset,
562 					radeon_bo_size(track->db_bo));
563 			return -EINVAL;
564 		}
565 		ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
566 	} else {
567 		/* pitch in pixels */
568 		pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
569 		slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
570 		slice_tile_max *= 64;
571 		height = slice_tile_max / pitch;
572 		if (height > 8192)
573 			height = 8192;
574 		base_offset = track->db_bo_mc + track->db_offset;
575 		array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
576 		array_check.array_mode = array_mode;
577 		array_check.group_size = track->group_size;
578 		array_check.nbanks = track->nbanks;
579 		array_check.npipes = track->npipes;
580 		array_check.nsamples = track->nsamples;
581 		array_check.blocksize = bpe;
582 		if (r600_get_array_mode_alignment(&array_check,
583 					&pitch_align, &height_align, &depth_align, &base_align)) {
584 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
585 					G_028010_ARRAY_MODE(track->db_depth_info),
586 					track->db_depth_info);
587 			return -EINVAL;
588 		}
589 		switch (array_mode) {
590 		case V_028010_ARRAY_1D_TILED_THIN1:
591 			/* don't break userspace */
592 			height &= ~0x7;
593 			break;
594 		case V_028010_ARRAY_2D_TILED_THIN1:
595 			break;
596 		default:
597 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
598 					G_028010_ARRAY_MODE(track->db_depth_info),
599 					track->db_depth_info);
600 			return -EINVAL;
601 		}
602 
603 		if (!IS_ALIGNED(pitch, pitch_align)) {
604 			dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
605 					__func__, __LINE__, pitch, pitch_align, array_mode);
606 			return -EINVAL;
607 		}
608 		if (!IS_ALIGNED(height, height_align)) {
609 			dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
610 					__func__, __LINE__, height, height_align, array_mode);
611 			return -EINVAL;
612 		}
613 		if (!IS_ALIGNED(base_offset, base_align)) {
614 			dev_warn(p->dev, "%s offset 0x%llx, 0x%llx, %d not aligned\n", __func__,
615 					base_offset, base_align, array_mode);
616 			return -EINVAL;
617 		}
618 
619 		ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
620 		nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
621 		tmp = ntiles * bpe * 64 * nviews * track->nsamples;
622 		if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
623 			dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
624 					array_mode,
625 					track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
626 					radeon_bo_size(track->db_bo));
627 			return -EINVAL;
628 		}
629 	}
630 
631 	/* hyperz */
632 	if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
633 		unsigned long size;
634 		unsigned nbx, nby;
635 
636 		if (track->htile_bo == NULL) {
637 			dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
638 				 __func__, __LINE__, track->db_depth_info);
639 			return -EINVAL;
640 		}
641 		if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
642 			dev_warn(p->dev, "%s:%d htile can't be enabled with bogus db_depth_size 0x%08x\n",
643 				 __func__, __LINE__, track->db_depth_size);
644 			return -EINVAL;
645 		}
646 
647 		nbx = pitch;
648 		nby = height;
649 		if (G_028D24_LINEAR(track->htile_surface)) {
650 			/* nbx must be 16 htiles aligned == 16 * 8 pixel aligned */
651 			nbx = round_up(nbx, 16 * 8);
652 			/* nby is npipes htiles aligned == npipes * 8 pixel aligned */
653 			nby = round_up(nby, track->npipes * 8);
654 		} else {
655 			/* always assume 8x8 htile */
656 			/* align is htile align * 8, htile align vary according to
657 			 * number of pipe and tile width and nby
658 			 */
659 			switch (track->npipes) {
660 			case 8:
661 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
662 				nbx = round_up(nbx, 64 * 8);
663 				nby = round_up(nby, 64 * 8);
664 				break;
665 			case 4:
666 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
667 				nbx = round_up(nbx, 64 * 8);
668 				nby = round_up(nby, 32 * 8);
669 				break;
670 			case 2:
671 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
672 				nbx = round_up(nbx, 32 * 8);
673 				nby = round_up(nby, 32 * 8);
674 				break;
675 			case 1:
676 				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
677 				nbx = round_up(nbx, 32 * 8);
678 				nby = round_up(nby, 16 * 8);
679 				break;
680 			default:
681 				dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
682 					 __func__, __LINE__, track->npipes);
683 				return -EINVAL;
684 			}
685 		}
686 		/* compute number of htile */
687 		nbx = nbx >> 3;
688 		nby = nby >> 3;
689 		/* size must be aligned on npipes * 2K boundary */
690 		size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
691 		size += track->htile_offset;
692 
693 		if (size > radeon_bo_size(track->htile_bo)) {
694 			dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
695 				 __func__, __LINE__, radeon_bo_size(track->htile_bo),
696 				 size, nbx, nby);
697 			return -EINVAL;
698 		}
699 	}
700 
701 	track->db_dirty = false;
702 	return 0;
703 }
704 
705 static int r600_cs_track_check(struct radeon_cs_parser *p)
706 {
707 	struct r600_cs_track *track = p->track;
708 	u32 tmp;
709 	int r, i;
710 
711 	/* on legacy kernel we don't perform advanced check */
712 	if (p->rdev == NULL)
713 		return 0;
714 
715 	/* check streamout */
716 	if (track->streamout_dirty && track->vgt_strmout_en) {
717 		for (i = 0; i < 4; i++) {
718 			if (track->vgt_strmout_buffer_en & (1 << i)) {
719 				if (track->vgt_strmout_bo[i]) {
720 					u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
721 						(u64)track->vgt_strmout_size[i];
722 					if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
723 						DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
724 							  i, offset,
725 							  radeon_bo_size(track->vgt_strmout_bo[i]));
726 						return -EINVAL;
727 					}
728 				} else {
729 					dev_warn(p->dev, "No buffer for streamout %d\n", i);
730 					return -EINVAL;
731 				}
732 			}
733 		}
734 		track->streamout_dirty = false;
735 	}
736 
737 	if (track->sx_misc_kill_all_prims)
738 		return 0;
739 
740 	/* check that we have a cb for each enabled target, we don't check
741 	 * shader_mask because it seems mesa isn't always setting it :(
742 	 */
743 	if (track->cb_dirty) {
744 		tmp = track->cb_target_mask;
745 
746 		/* We must check both colorbuffers for RESOLVE. */
747 		if (track->is_resolve) {
748 			tmp |= 0xff;
749 		}
750 
751 		for (i = 0; i < 8; i++) {
752 			u32 format = G_0280A0_FORMAT(track->cb_color_info[i]);
753 
754 			if (format != V_0280A0_COLOR_INVALID &&
755 			    (tmp >> (i * 4)) & 0xF) {
756 				/* at least one component is enabled */
757 				if (track->cb_color_bo[i] == NULL) {
758 					dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
759 						__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
760 					return -EINVAL;
761 				}
762 				/* perform rewrite of CB_COLOR[0-7]_SIZE */
763 				r = r600_cs_track_validate_cb(p, i);
764 				if (r)
765 					return r;
766 			}
767 		}
768 		track->cb_dirty = false;
769 	}
770 
771 	/* Check depth buffer */
772 	if (track->db_dirty &&
773 	    G_028010_FORMAT(track->db_depth_info) != V_028010_DEPTH_INVALID &&
774 	    (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
775 	     G_028800_Z_ENABLE(track->db_depth_control))) {
776 		r = r600_cs_track_validate_db(p);
777 		if (r)
778 			return r;
779 	}
780 
781 	return 0;
782 }
783 
784 /**
785  * r600_cs_packet_parse_vline() - parse userspace VLINE packet
786  * @p:		parser structure holding parsing context.
787  *
788  * This is an R600-specific function for parsing VLINE packets.
789  * Real work is done by r600_cs_common_vline_parse function.
790  * Here we just set up ASIC-specific register table and call
791  * the common implementation function.
792  */
793 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
794 {
795 	static uint32_t vline_start_end[2] = {AVIVO_D1MODE_VLINE_START_END,
796 					      AVIVO_D2MODE_VLINE_START_END};
797 	static uint32_t vline_status[2] = {AVIVO_D1MODE_VLINE_STATUS,
798 					   AVIVO_D2MODE_VLINE_STATUS};
799 
800 	return r600_cs_common_vline_parse(p, vline_start_end, vline_status);
801 }
802 
803 /**
804  * r600_cs_common_vline_parse() - common vline parser
805  * @p:			parser structure holding parsing context.
806  * @vline_start_end:    table of vline_start_end registers
807  * @vline_status:       table of vline_status registers
808  *
809  * Userspace sends a special sequence for VLINE waits.
810  * PACKET0 - VLINE_START_END + value
811  * PACKET3 - WAIT_REG_MEM poll vline status reg
812  * RELOC (P3) - crtc_id in reloc.
813  *
814  * This function parses this and relocates the VLINE START END
815  * and WAIT_REG_MEM packets to the correct crtc.
816  * It also detects a switched off crtc and nulls out the
817  * wait in that case. This function is common for all ASICs that
818  * are R600 and newer. The parsing algorithm is the same, and only
819  * differs in which registers are used.
820  *
821  * Caller is the ASIC-specific function which passes the parser
822  * context and ASIC-specific register table
823  */
824 int r600_cs_common_vline_parse(struct radeon_cs_parser *p,
825 			       uint32_t *vline_start_end,
826 			       uint32_t *vline_status)
827 {
828 	struct drm_crtc *crtc;
829 	struct radeon_crtc *radeon_crtc;
830 	struct radeon_cs_packet p3reloc, wait_reg_mem;
831 	int crtc_id;
832 	int r;
833 	uint32_t header, h_idx, reg, wait_reg_mem_info;
834 	volatile uint32_t *ib;
835 
836 	ib = p->ib.ptr;
837 
838 	/* parse the WAIT_REG_MEM */
839 	r = radeon_cs_packet_parse(p, &wait_reg_mem, p->idx);
840 	if (r)
841 		return r;
842 
843 	/* check its a WAIT_REG_MEM */
844 	if (wait_reg_mem.type != RADEON_PACKET_TYPE3 ||
845 	    wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
846 		DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
847 		return -EINVAL;
848 	}
849 
850 	wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
851 	/* bit 4 is reg (0) or mem (1) */
852 	if (wait_reg_mem_info & 0x10) {
853 		DRM_ERROR("vline WAIT_REG_MEM waiting on MEM instead of REG\n");
854 		return -EINVAL;
855 	}
856 	/* bit 8 is me (0) or pfp (1) */
857 	if (wait_reg_mem_info & 0x100) {
858 		DRM_ERROR("vline WAIT_REG_MEM waiting on PFP instead of ME\n");
859 		return -EINVAL;
860 	}
861 	/* waiting for value to be equal */
862 	if ((wait_reg_mem_info & 0x7) != 0x3) {
863 		DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
864 		return -EINVAL;
865 	}
866 	if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != vline_status[0]) {
867 		DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
868 		return -EINVAL;
869 	}
870 
871 	if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != RADEON_VLINE_STAT) {
872 		DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
873 		return -EINVAL;
874 	}
875 
876 	/* jump over the NOP */
877 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
878 	if (r)
879 		return r;
880 
881 	h_idx = p->idx - 2;
882 	p->idx += wait_reg_mem.count + 2;
883 	p->idx += p3reloc.count + 2;
884 
885 	header = radeon_get_ib_value(p, h_idx);
886 	crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
887 	reg = R600_CP_PACKET0_GET_REG(header);
888 
889 	crtc = drm_crtc_find(p->rdev->ddev, p->filp, crtc_id);
890 	if (!crtc) {
891 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
892 		return -ENOENT;
893 	}
894 	radeon_crtc = to_radeon_crtc(crtc);
895 	crtc_id = radeon_crtc->crtc_id;
896 
897 	if (!crtc->enabled) {
898 		/* CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
899 		ib[h_idx + 2] = PACKET2(0);
900 		ib[h_idx + 3] = PACKET2(0);
901 		ib[h_idx + 4] = PACKET2(0);
902 		ib[h_idx + 5] = PACKET2(0);
903 		ib[h_idx + 6] = PACKET2(0);
904 		ib[h_idx + 7] = PACKET2(0);
905 		ib[h_idx + 8] = PACKET2(0);
906 	} else if (reg == vline_start_end[0]) {
907 		header &= ~R600_CP_PACKET0_REG_MASK;
908 		header |= vline_start_end[crtc_id] >> 2;
909 		ib[h_idx] = header;
910 		ib[h_idx + 4] = vline_status[crtc_id] >> 2;
911 	} else {
912 		DRM_ERROR("unknown crtc reloc\n");
913 		return -EINVAL;
914 	}
915 	return 0;
916 }
917 
918 static int r600_packet0_check(struct radeon_cs_parser *p,
919 				struct radeon_cs_packet *pkt,
920 				unsigned idx, unsigned reg)
921 {
922 	int r;
923 
924 	switch (reg) {
925 	case AVIVO_D1MODE_VLINE_START_END:
926 		r = r600_cs_packet_parse_vline(p);
927 		if (r) {
928 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
929 					idx, reg);
930 			return r;
931 		}
932 		break;
933 	default:
934 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
935 		return -EINVAL;
936 	}
937 	return 0;
938 }
939 
940 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
941 				struct radeon_cs_packet *pkt)
942 {
943 	unsigned reg, i;
944 	unsigned idx;
945 	int r;
946 
947 	idx = pkt->idx + 1;
948 	reg = pkt->reg;
949 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
950 		r = r600_packet0_check(p, pkt, idx, reg);
951 		if (r) {
952 			return r;
953 		}
954 	}
955 	return 0;
956 }
957 
958 /**
959  * r600_cs_check_reg() - check if register is authorized or not
960  * @p: parser structure holding parsing context
961  * @reg: register we are testing
962  * @idx: index into the cs buffer
963  *
964  * This function will test against r600_reg_safe_bm and return 0
965  * if register is safe. If register is not flag as safe this function
966  * will test it against a list of register needing special handling.
967  */
968 static int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
969 {
970 	struct r600_cs_track *track = (struct r600_cs_track *)p->track;
971 	struct radeon_bo_list *reloc;
972 	u32 m, i, tmp, *ib;
973 	int r;
974 
975 	i = (reg >> 7);
976 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
977 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
978 		return -EINVAL;
979 	}
980 	m = 1 << ((reg >> 2) & 31);
981 	if (!(r600_reg_safe_bm[i] & m))
982 		return 0;
983 	ib = p->ib.ptr;
984 	switch (reg) {
985 	/* force following reg to 0 in an attempt to disable out buffer
986 	 * which will need us to better understand how it works to perform
987 	 * security check on it (Jerome)
988 	 */
989 	case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
990 	case R_008C44_SQ_ESGS_RING_SIZE:
991 	case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
992 	case R_008C54_SQ_ESTMP_RING_SIZE:
993 	case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
994 	case R_008C74_SQ_FBUF_RING_SIZE:
995 	case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
996 	case R_008C5C_SQ_GSTMP_RING_SIZE:
997 	case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
998 	case R_008C4C_SQ_GSVS_RING_SIZE:
999 	case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
1000 	case R_008C6C_SQ_PSTMP_RING_SIZE:
1001 	case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
1002 	case R_008C7C_SQ_REDUC_RING_SIZE:
1003 	case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
1004 	case R_008C64_SQ_VSTMP_RING_SIZE:
1005 	case R_0288C8_SQ_GS_VERT_ITEMSIZE:
1006 		/* get value to populate the IB don't remove */
1007 		/*tmp =radeon_get_ib_value(p, idx);
1008 		  ib[idx] = 0;*/
1009 		break;
1010 	case SQ_ESGS_RING_BASE:
1011 	case SQ_GSVS_RING_BASE:
1012 	case SQ_ESTMP_RING_BASE:
1013 	case SQ_GSTMP_RING_BASE:
1014 	case SQ_PSTMP_RING_BASE:
1015 	case SQ_VSTMP_RING_BASE:
1016 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1017 		if (r) {
1018 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1019 					"0x%04X\n", reg);
1020 			return -EINVAL;
1021 		}
1022 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1023 		break;
1024 	case SQ_CONFIG:
1025 		track->sq_config = radeon_get_ib_value(p, idx);
1026 		break;
1027 	case R_028800_DB_DEPTH_CONTROL:
1028 		track->db_depth_control = radeon_get_ib_value(p, idx);
1029 		track->db_dirty = true;
1030 		break;
1031 	case R_028010_DB_DEPTH_INFO:
1032 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1033 		    radeon_cs_packet_next_is_pkt3_nop(p)) {
1034 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1035 			if (r) {
1036 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1037 					 "0x%04X\n", reg);
1038 				return -EINVAL;
1039 			}
1040 			track->db_depth_info = radeon_get_ib_value(p, idx);
1041 			ib[idx] &= C_028010_ARRAY_MODE;
1042 			track->db_depth_info &= C_028010_ARRAY_MODE;
1043 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1044 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1045 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1046 			} else {
1047 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1048 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1049 			}
1050 		} else {
1051 			track->db_depth_info = radeon_get_ib_value(p, idx);
1052 		}
1053 		track->db_dirty = true;
1054 		break;
1055 	case R_028004_DB_DEPTH_VIEW:
1056 		track->db_depth_view = radeon_get_ib_value(p, idx);
1057 		track->db_dirty = true;
1058 		break;
1059 	case R_028000_DB_DEPTH_SIZE:
1060 		track->db_depth_size = radeon_get_ib_value(p, idx);
1061 		track->db_depth_size_idx = idx;
1062 		track->db_dirty = true;
1063 		break;
1064 	case R_028AB0_VGT_STRMOUT_EN:
1065 		track->vgt_strmout_en = radeon_get_ib_value(p, idx);
1066 		track->streamout_dirty = true;
1067 		break;
1068 	case R_028B20_VGT_STRMOUT_BUFFER_EN:
1069 		track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
1070 		track->streamout_dirty = true;
1071 		break;
1072 	case VGT_STRMOUT_BUFFER_BASE_0:
1073 	case VGT_STRMOUT_BUFFER_BASE_1:
1074 	case VGT_STRMOUT_BUFFER_BASE_2:
1075 	case VGT_STRMOUT_BUFFER_BASE_3:
1076 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1077 		if (r) {
1078 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1079 					"0x%04X\n", reg);
1080 			return -EINVAL;
1081 		}
1082 		tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1083 		track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1084 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1085 		track->vgt_strmout_bo[tmp] = reloc->robj;
1086 		track->vgt_strmout_bo_mc[tmp] = reloc->gpu_offset;
1087 		track->streamout_dirty = true;
1088 		break;
1089 	case VGT_STRMOUT_BUFFER_SIZE_0:
1090 	case VGT_STRMOUT_BUFFER_SIZE_1:
1091 	case VGT_STRMOUT_BUFFER_SIZE_2:
1092 	case VGT_STRMOUT_BUFFER_SIZE_3:
1093 		tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1094 		/* size in register is DWs, convert to bytes */
1095 		track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1096 		track->streamout_dirty = true;
1097 		break;
1098 	case CP_COHER_BASE:
1099 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1100 		if (r) {
1101 			dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1102 					"0x%04X\n", reg);
1103 			return -EINVAL;
1104 		}
1105 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1106 		break;
1107 	case R_028238_CB_TARGET_MASK:
1108 		track->cb_target_mask = radeon_get_ib_value(p, idx);
1109 		track->cb_dirty = true;
1110 		break;
1111 	case R_02823C_CB_SHADER_MASK:
1112 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
1113 		break;
1114 	case R_028C04_PA_SC_AA_CONFIG:
1115 		tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
1116 		track->log_nsamples = tmp;
1117 		track->nsamples = 1 << tmp;
1118 		track->cb_dirty = true;
1119 		break;
1120 	case R_028808_CB_COLOR_CONTROL:
1121 		tmp = G_028808_SPECIAL_OP(radeon_get_ib_value(p, idx));
1122 		track->is_resolve = tmp == V_028808_SPECIAL_RESOLVE_BOX;
1123 		track->cb_dirty = true;
1124 		break;
1125 	case R_0280A0_CB_COLOR0_INFO:
1126 	case R_0280A4_CB_COLOR1_INFO:
1127 	case R_0280A8_CB_COLOR2_INFO:
1128 	case R_0280AC_CB_COLOR3_INFO:
1129 	case R_0280B0_CB_COLOR4_INFO:
1130 	case R_0280B4_CB_COLOR5_INFO:
1131 	case R_0280B8_CB_COLOR6_INFO:
1132 	case R_0280BC_CB_COLOR7_INFO:
1133 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1134 		     radeon_cs_packet_next_is_pkt3_nop(p)) {
1135 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1136 			if (r) {
1137 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1138 				return -EINVAL;
1139 			}
1140 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1141 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1142 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1143 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1144 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1145 			} else if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1146 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1147 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1148 			}
1149 		} else {
1150 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1151 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1152 		}
1153 		track->cb_dirty = true;
1154 		break;
1155 	case R_028080_CB_COLOR0_VIEW:
1156 	case R_028084_CB_COLOR1_VIEW:
1157 	case R_028088_CB_COLOR2_VIEW:
1158 	case R_02808C_CB_COLOR3_VIEW:
1159 	case R_028090_CB_COLOR4_VIEW:
1160 	case R_028094_CB_COLOR5_VIEW:
1161 	case R_028098_CB_COLOR6_VIEW:
1162 	case R_02809C_CB_COLOR7_VIEW:
1163 		tmp = (reg - R_028080_CB_COLOR0_VIEW) / 4;
1164 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1165 		track->cb_dirty = true;
1166 		break;
1167 	case R_028060_CB_COLOR0_SIZE:
1168 	case R_028064_CB_COLOR1_SIZE:
1169 	case R_028068_CB_COLOR2_SIZE:
1170 	case R_02806C_CB_COLOR3_SIZE:
1171 	case R_028070_CB_COLOR4_SIZE:
1172 	case R_028074_CB_COLOR5_SIZE:
1173 	case R_028078_CB_COLOR6_SIZE:
1174 	case R_02807C_CB_COLOR7_SIZE:
1175 		tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1176 		track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1177 		track->cb_color_size_idx[tmp] = idx;
1178 		track->cb_dirty = true;
1179 		break;
1180 		/* This register were added late, there is userspace
1181 		 * which does provide relocation for those but set
1182 		 * 0 offset. In order to avoid breaking old userspace
1183 		 * we detect this and set address to point to last
1184 		 * CB_COLOR0_BASE, note that if userspace doesn't set
1185 		 * CB_COLOR0_BASE before this register we will report
1186 		 * error. Old userspace always set CB_COLOR0_BASE
1187 		 * before any of this.
1188 		 */
1189 	case R_0280E0_CB_COLOR0_FRAG:
1190 	case R_0280E4_CB_COLOR1_FRAG:
1191 	case R_0280E8_CB_COLOR2_FRAG:
1192 	case R_0280EC_CB_COLOR3_FRAG:
1193 	case R_0280F0_CB_COLOR4_FRAG:
1194 	case R_0280F4_CB_COLOR5_FRAG:
1195 	case R_0280F8_CB_COLOR6_FRAG:
1196 	case R_0280FC_CB_COLOR7_FRAG:
1197 		tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1198 		if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1199 			if (!track->cb_color_base_last[tmp]) {
1200 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1201 				return -EINVAL;
1202 			}
1203 			track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1204 			track->cb_color_frag_offset[tmp] = track->cb_color_bo_offset[tmp];
1205 			ib[idx] = track->cb_color_base_last[tmp];
1206 		} else {
1207 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1208 			if (r) {
1209 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1210 				return -EINVAL;
1211 			}
1212 			track->cb_color_frag_bo[tmp] = reloc->robj;
1213 			track->cb_color_frag_offset[tmp] = (u64)ib[idx] << 8;
1214 			ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1215 		}
1216 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1217 			track->cb_dirty = true;
1218 		}
1219 		break;
1220 	case R_0280C0_CB_COLOR0_TILE:
1221 	case R_0280C4_CB_COLOR1_TILE:
1222 	case R_0280C8_CB_COLOR2_TILE:
1223 	case R_0280CC_CB_COLOR3_TILE:
1224 	case R_0280D0_CB_COLOR4_TILE:
1225 	case R_0280D4_CB_COLOR5_TILE:
1226 	case R_0280D8_CB_COLOR6_TILE:
1227 	case R_0280DC_CB_COLOR7_TILE:
1228 		tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1229 		if (!radeon_cs_packet_next_is_pkt3_nop(p)) {
1230 			if (!track->cb_color_base_last[tmp]) {
1231 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1232 				return -EINVAL;
1233 			}
1234 			track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1235 			track->cb_color_tile_offset[tmp] = track->cb_color_bo_offset[tmp];
1236 			ib[idx] = track->cb_color_base_last[tmp];
1237 		} else {
1238 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1239 			if (r) {
1240 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1241 				return -EINVAL;
1242 			}
1243 			track->cb_color_tile_bo[tmp] = reloc->robj;
1244 			track->cb_color_tile_offset[tmp] = (u64)ib[idx] << 8;
1245 			ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1246 		}
1247 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1248 			track->cb_dirty = true;
1249 		}
1250 		break;
1251 	case R_028100_CB_COLOR0_MASK:
1252 	case R_028104_CB_COLOR1_MASK:
1253 	case R_028108_CB_COLOR2_MASK:
1254 	case R_02810C_CB_COLOR3_MASK:
1255 	case R_028110_CB_COLOR4_MASK:
1256 	case R_028114_CB_COLOR5_MASK:
1257 	case R_028118_CB_COLOR6_MASK:
1258 	case R_02811C_CB_COLOR7_MASK:
1259 		tmp = (reg - R_028100_CB_COLOR0_MASK) / 4;
1260 		track->cb_color_mask[tmp] = radeon_get_ib_value(p, idx);
1261 		if (G_0280A0_TILE_MODE(track->cb_color_info[tmp])) {
1262 			track->cb_dirty = true;
1263 		}
1264 		break;
1265 	case CB_COLOR0_BASE:
1266 	case CB_COLOR1_BASE:
1267 	case CB_COLOR2_BASE:
1268 	case CB_COLOR3_BASE:
1269 	case CB_COLOR4_BASE:
1270 	case CB_COLOR5_BASE:
1271 	case CB_COLOR6_BASE:
1272 	case CB_COLOR7_BASE:
1273 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1274 		if (r) {
1275 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1276 					"0x%04X\n", reg);
1277 			return -EINVAL;
1278 		}
1279 		tmp = (reg - CB_COLOR0_BASE) / 4;
1280 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1281 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1282 		track->cb_color_base_last[tmp] = ib[idx];
1283 		track->cb_color_bo[tmp] = reloc->robj;
1284 		track->cb_color_bo_mc[tmp] = reloc->gpu_offset;
1285 		track->cb_dirty = true;
1286 		break;
1287 	case DB_DEPTH_BASE:
1288 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1289 		if (r) {
1290 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1291 					"0x%04X\n", reg);
1292 			return -EINVAL;
1293 		}
1294 		track->db_offset = radeon_get_ib_value(p, idx) << 8;
1295 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1296 		track->db_bo = reloc->robj;
1297 		track->db_bo_mc = reloc->gpu_offset;
1298 		track->db_dirty = true;
1299 		break;
1300 	case DB_HTILE_DATA_BASE:
1301 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1302 		if (r) {
1303 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1304 					"0x%04X\n", reg);
1305 			return -EINVAL;
1306 		}
1307 		track->htile_offset = radeon_get_ib_value(p, idx) << 8;
1308 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1309 		track->htile_bo = reloc->robj;
1310 		track->db_dirty = true;
1311 		break;
1312 	case DB_HTILE_SURFACE:
1313 		track->htile_surface = radeon_get_ib_value(p, idx);
1314 		/* force 8x8 htile width and height */
1315 		ib[idx] |= 3;
1316 		track->db_dirty = true;
1317 		break;
1318 	case SQ_PGM_START_FS:
1319 	case SQ_PGM_START_ES:
1320 	case SQ_PGM_START_VS:
1321 	case SQ_PGM_START_GS:
1322 	case SQ_PGM_START_PS:
1323 	case SQ_ALU_CONST_CACHE_GS_0:
1324 	case SQ_ALU_CONST_CACHE_GS_1:
1325 	case SQ_ALU_CONST_CACHE_GS_2:
1326 	case SQ_ALU_CONST_CACHE_GS_3:
1327 	case SQ_ALU_CONST_CACHE_GS_4:
1328 	case SQ_ALU_CONST_CACHE_GS_5:
1329 	case SQ_ALU_CONST_CACHE_GS_6:
1330 	case SQ_ALU_CONST_CACHE_GS_7:
1331 	case SQ_ALU_CONST_CACHE_GS_8:
1332 	case SQ_ALU_CONST_CACHE_GS_9:
1333 	case SQ_ALU_CONST_CACHE_GS_10:
1334 	case SQ_ALU_CONST_CACHE_GS_11:
1335 	case SQ_ALU_CONST_CACHE_GS_12:
1336 	case SQ_ALU_CONST_CACHE_GS_13:
1337 	case SQ_ALU_CONST_CACHE_GS_14:
1338 	case SQ_ALU_CONST_CACHE_GS_15:
1339 	case SQ_ALU_CONST_CACHE_PS_0:
1340 	case SQ_ALU_CONST_CACHE_PS_1:
1341 	case SQ_ALU_CONST_CACHE_PS_2:
1342 	case SQ_ALU_CONST_CACHE_PS_3:
1343 	case SQ_ALU_CONST_CACHE_PS_4:
1344 	case SQ_ALU_CONST_CACHE_PS_5:
1345 	case SQ_ALU_CONST_CACHE_PS_6:
1346 	case SQ_ALU_CONST_CACHE_PS_7:
1347 	case SQ_ALU_CONST_CACHE_PS_8:
1348 	case SQ_ALU_CONST_CACHE_PS_9:
1349 	case SQ_ALU_CONST_CACHE_PS_10:
1350 	case SQ_ALU_CONST_CACHE_PS_11:
1351 	case SQ_ALU_CONST_CACHE_PS_12:
1352 	case SQ_ALU_CONST_CACHE_PS_13:
1353 	case SQ_ALU_CONST_CACHE_PS_14:
1354 	case SQ_ALU_CONST_CACHE_PS_15:
1355 	case SQ_ALU_CONST_CACHE_VS_0:
1356 	case SQ_ALU_CONST_CACHE_VS_1:
1357 	case SQ_ALU_CONST_CACHE_VS_2:
1358 	case SQ_ALU_CONST_CACHE_VS_3:
1359 	case SQ_ALU_CONST_CACHE_VS_4:
1360 	case SQ_ALU_CONST_CACHE_VS_5:
1361 	case SQ_ALU_CONST_CACHE_VS_6:
1362 	case SQ_ALU_CONST_CACHE_VS_7:
1363 	case SQ_ALU_CONST_CACHE_VS_8:
1364 	case SQ_ALU_CONST_CACHE_VS_9:
1365 	case SQ_ALU_CONST_CACHE_VS_10:
1366 	case SQ_ALU_CONST_CACHE_VS_11:
1367 	case SQ_ALU_CONST_CACHE_VS_12:
1368 	case SQ_ALU_CONST_CACHE_VS_13:
1369 	case SQ_ALU_CONST_CACHE_VS_14:
1370 	case SQ_ALU_CONST_CACHE_VS_15:
1371 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1372 		if (r) {
1373 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1374 					"0x%04X\n", reg);
1375 			return -EINVAL;
1376 		}
1377 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1378 		break;
1379 	case SX_MEMORY_EXPORT_BASE:
1380 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1381 		if (r) {
1382 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1383 					"0x%04X\n", reg);
1384 			return -EINVAL;
1385 		}
1386 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1387 		break;
1388 	case SX_MISC:
1389 		track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1390 		break;
1391 	default:
1392 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1393 		return -EINVAL;
1394 	}
1395 	return 0;
1396 }
1397 
1398 unsigned r600_mip_minify(unsigned size, unsigned level)
1399 {
1400 	unsigned val;
1401 
1402 	val = max(1U, size >> level);
1403 	if (level > 0)
1404 		val = roundup_pow_of_two(val);
1405 	return val;
1406 }
1407 
1408 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1409 			      unsigned w0, unsigned h0, unsigned d0, unsigned nsamples, unsigned format,
1410 			      unsigned block_align, unsigned height_align, unsigned base_align,
1411 			      unsigned *l0_size, unsigned *mipmap_size)
1412 {
1413 	unsigned offset, i, level;
1414 	unsigned width, height, depth, size;
1415 	unsigned blocksize;
1416 	unsigned nbx, nby;
1417 	unsigned nlevels = llevel - blevel + 1;
1418 
1419 	*l0_size = -1;
1420 	blocksize = r600_fmt_get_blocksize(format);
1421 
1422 	w0 = r600_mip_minify(w0, 0);
1423 	h0 = r600_mip_minify(h0, 0);
1424 	d0 = r600_mip_minify(d0, 0);
1425 	for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1426 		width = r600_mip_minify(w0, i);
1427 		nbx = r600_fmt_get_nblocksx(format, width);
1428 
1429 		nbx = round_up(nbx, block_align);
1430 
1431 		height = r600_mip_minify(h0, i);
1432 		nby = r600_fmt_get_nblocksy(format, height);
1433 		nby = round_up(nby, height_align);
1434 
1435 		depth = r600_mip_minify(d0, i);
1436 
1437 		size = nbx * nby * blocksize * nsamples;
1438 		if (nfaces)
1439 			size *= nfaces;
1440 		else
1441 			size *= depth;
1442 
1443 		if (i == 0)
1444 			*l0_size = size;
1445 
1446 		if (i == 0 || i == 1)
1447 			offset = round_up(offset, base_align);
1448 
1449 		offset += size;
1450 	}
1451 	*mipmap_size = offset;
1452 	if (llevel == 0)
1453 		*mipmap_size = *l0_size;
1454 	if (!blevel)
1455 		*mipmap_size -= *l0_size;
1456 }
1457 
1458 /**
1459  * r600_check_texture_resource() - check if register is authorized or not
1460  * @p: parser structure holding parsing context
1461  * @idx: index into the cs buffer
1462  * @texture: texture's bo structure
1463  * @mipmap: mipmap's bo structure
1464  * @base_offset: base offset (used for error checking)
1465  * @mip_offset: mip offset (used for error checking)
1466  * @tiling_flags: tiling flags
1467  *
1468  * This function will check that the resource has valid field and that
1469  * the texture and mipmap bo object are big enough to cover this resource.
1470  */
1471 static int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1472 					      struct radeon_bo *texture,
1473 					      struct radeon_bo *mipmap,
1474 					      u64 base_offset,
1475 					      u64 mip_offset,
1476 					      u32 tiling_flags)
1477 {
1478 	struct r600_cs_track *track = p->track;
1479 	u32 dim, nfaces, llevel, blevel, w0, h0, d0;
1480 	u32 word0, word1, l0_size, mipmap_size, word2, word3, word4, word5;
1481 	u32 height_align, pitch, pitch_align, depth_align;
1482 	u32 barray, larray;
1483 	u64 base_align;
1484 	struct array_mode_checker array_check;
1485 	u32 format;
1486 	bool is_array;
1487 
1488 	/* on legacy kernel we don't perform advanced check */
1489 	if (p->rdev == NULL)
1490 		return 0;
1491 
1492 	/* convert to bytes */
1493 	base_offset <<= 8;
1494 	mip_offset <<= 8;
1495 
1496 	word0 = radeon_get_ib_value(p, idx + 0);
1497 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1498 		if (tiling_flags & RADEON_TILING_MACRO)
1499 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1500 		else if (tiling_flags & RADEON_TILING_MICRO)
1501 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1502 	}
1503 	word1 = radeon_get_ib_value(p, idx + 1);
1504 	word2 = radeon_get_ib_value(p, idx + 2) << 8;
1505 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1506 	word4 = radeon_get_ib_value(p, idx + 4);
1507 	word5 = radeon_get_ib_value(p, idx + 5);
1508 	dim = G_038000_DIM(word0);
1509 	w0 = G_038000_TEX_WIDTH(word0) + 1;
1510 	pitch = (G_038000_PITCH(word0) + 1) * 8;
1511 	h0 = G_038004_TEX_HEIGHT(word1) + 1;
1512 	d0 = G_038004_TEX_DEPTH(word1);
1513 	format = G_038004_DATA_FORMAT(word1);
1514 	blevel = G_038010_BASE_LEVEL(word4);
1515 	llevel = G_038014_LAST_LEVEL(word5);
1516 	/* pitch in texels */
1517 	array_check.array_mode = G_038000_TILE_MODE(word0);
1518 	array_check.group_size = track->group_size;
1519 	array_check.nbanks = track->nbanks;
1520 	array_check.npipes = track->npipes;
1521 	array_check.nsamples = 1;
1522 	array_check.blocksize = r600_fmt_get_blocksize(format);
1523 	nfaces = 1;
1524 	is_array = false;
1525 	switch (dim) {
1526 	case V_038000_SQ_TEX_DIM_1D:
1527 	case V_038000_SQ_TEX_DIM_2D:
1528 	case V_038000_SQ_TEX_DIM_3D:
1529 		break;
1530 	case V_038000_SQ_TEX_DIM_CUBEMAP:
1531 		if (p->family >= CHIP_RV770)
1532 			nfaces = 8;
1533 		else
1534 			nfaces = 6;
1535 		break;
1536 	case V_038000_SQ_TEX_DIM_1D_ARRAY:
1537 	case V_038000_SQ_TEX_DIM_2D_ARRAY:
1538 		is_array = true;
1539 		break;
1540 	case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1541 		is_array = true;
1542 		fallthrough;
1543 	case V_038000_SQ_TEX_DIM_2D_MSAA:
1544 		array_check.nsamples = 1 << llevel;
1545 		llevel = 0;
1546 		break;
1547 	default:
1548 		dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1549 		return -EINVAL;
1550 	}
1551 	if (!r600_fmt_is_valid_texture(format, p->family)) {
1552 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1553 			 __func__, __LINE__, format);
1554 		return -EINVAL;
1555 	}
1556 
1557 	if (r600_get_array_mode_alignment(&array_check,
1558 					  &pitch_align, &height_align, &depth_align, &base_align)) {
1559 		dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1560 			 __func__, __LINE__, G_038000_TILE_MODE(word0));
1561 		return -EINVAL;
1562 	}
1563 
1564 	/* XXX check height as well... */
1565 
1566 	if (!IS_ALIGNED(pitch, pitch_align)) {
1567 		dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1568 			 __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1569 		return -EINVAL;
1570 	}
1571 	if (!IS_ALIGNED(base_offset, base_align)) {
1572 		dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1573 			 __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1574 		return -EINVAL;
1575 	}
1576 	if (!IS_ALIGNED(mip_offset, base_align)) {
1577 		dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1578 			 __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1579 		return -EINVAL;
1580 	}
1581 
1582 	if (blevel > llevel) {
1583 		dev_warn(p->dev, "texture blevel %d > llevel %d\n",
1584 			 blevel, llevel);
1585 	}
1586 	if (is_array) {
1587 		barray = G_038014_BASE_ARRAY(word5);
1588 		larray = G_038014_LAST_ARRAY(word5);
1589 
1590 		nfaces = larray - barray + 1;
1591 	}
1592 	r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, array_check.nsamples, format,
1593 			  pitch_align, height_align, base_align,
1594 			  &l0_size, &mipmap_size);
1595 	/* using get ib will give us the offset into the texture bo */
1596 	if ((l0_size + word2) > radeon_bo_size(texture)) {
1597 		dev_warn(p->dev, "texture bo too small ((%d %d) (%d %d) %d %d %d -> %d have %ld)\n",
1598 			 w0, h0, pitch_align, height_align,
1599 			 array_check.array_mode, format, word2,
1600 			 l0_size, radeon_bo_size(texture));
1601 		dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1602 		return -EINVAL;
1603 	}
1604 	/* using get ib will give us the offset into the mipmap bo */
1605 	if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1606 		/*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1607 		  w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1608 	}
1609 	return 0;
1610 }
1611 
1612 static bool r600_is_safe_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1613 {
1614 	u32 m, i;
1615 
1616 	i = (reg >> 7);
1617 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
1618 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1619 		return false;
1620 	}
1621 	m = 1 << ((reg >> 2) & 31);
1622 	if (!(r600_reg_safe_bm[i] & m))
1623 		return true;
1624 	dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1625 	return false;
1626 }
1627 
1628 static int r600_packet3_check(struct radeon_cs_parser *p,
1629 				struct radeon_cs_packet *pkt)
1630 {
1631 	struct radeon_bo_list *reloc;
1632 	struct r600_cs_track *track;
1633 	volatile u32 *ib;
1634 	unsigned idx;
1635 	unsigned i;
1636 	unsigned start_reg, end_reg, reg;
1637 	int r;
1638 	u32 idx_value;
1639 
1640 	track = (struct r600_cs_track *)p->track;
1641 	ib = p->ib.ptr;
1642 	idx = pkt->idx + 1;
1643 	idx_value = radeon_get_ib_value(p, idx);
1644 
1645 	switch (pkt->opcode) {
1646 	case PACKET3_SET_PREDICATION:
1647 	{
1648 		int pred_op;
1649 		int tmp;
1650 		uint64_t offset;
1651 
1652 		if (pkt->count != 1) {
1653 			DRM_ERROR("bad SET PREDICATION\n");
1654 			return -EINVAL;
1655 		}
1656 
1657 		tmp = radeon_get_ib_value(p, idx + 1);
1658 		pred_op = (tmp >> 16) & 0x7;
1659 
1660 		/* for the clear predicate operation */
1661 		if (pred_op == 0)
1662 			return 0;
1663 
1664 		if (pred_op > 2) {
1665 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1666 			return -EINVAL;
1667 		}
1668 
1669 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1670 		if (r) {
1671 			DRM_ERROR("bad SET PREDICATION\n");
1672 			return -EINVAL;
1673 		}
1674 
1675 		offset = reloc->gpu_offset +
1676 			 (idx_value & 0xfffffff0) +
1677 			 ((u64)(tmp & 0xff) << 32);
1678 
1679 		ib[idx + 0] = offset;
1680 		ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1681 	}
1682 	break;
1683 
1684 	case PACKET3_START_3D_CMDBUF:
1685 		if (p->family >= CHIP_RV770 || pkt->count) {
1686 			DRM_ERROR("bad START_3D\n");
1687 			return -EINVAL;
1688 		}
1689 		break;
1690 	case PACKET3_CONTEXT_CONTROL:
1691 		if (pkt->count != 1) {
1692 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1693 			return -EINVAL;
1694 		}
1695 		break;
1696 	case PACKET3_INDEX_TYPE:
1697 	case PACKET3_NUM_INSTANCES:
1698 		if (pkt->count) {
1699 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1700 			return -EINVAL;
1701 		}
1702 		break;
1703 	case PACKET3_DRAW_INDEX:
1704 	{
1705 		uint64_t offset;
1706 		if (pkt->count != 3) {
1707 			DRM_ERROR("bad DRAW_INDEX\n");
1708 			return -EINVAL;
1709 		}
1710 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1711 		if (r) {
1712 			DRM_ERROR("bad DRAW_INDEX\n");
1713 			return -EINVAL;
1714 		}
1715 
1716 		offset = reloc->gpu_offset +
1717 			 idx_value +
1718 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1719 
1720 		ib[idx+0] = offset;
1721 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1722 
1723 		r = r600_cs_track_check(p);
1724 		if (r) {
1725 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1726 			return r;
1727 		}
1728 		break;
1729 	}
1730 	case PACKET3_DRAW_INDEX_AUTO:
1731 		if (pkt->count != 1) {
1732 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1733 			return -EINVAL;
1734 		}
1735 		r = r600_cs_track_check(p);
1736 		if (r) {
1737 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1738 			return r;
1739 		}
1740 		break;
1741 	case PACKET3_DRAW_INDEX_IMMD_BE:
1742 	case PACKET3_DRAW_INDEX_IMMD:
1743 		if (pkt->count < 2) {
1744 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1745 			return -EINVAL;
1746 		}
1747 		r = r600_cs_track_check(p);
1748 		if (r) {
1749 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1750 			return r;
1751 		}
1752 		break;
1753 	case PACKET3_WAIT_REG_MEM:
1754 		if (pkt->count != 5) {
1755 			DRM_ERROR("bad WAIT_REG_MEM\n");
1756 			return -EINVAL;
1757 		}
1758 		/* bit 4 is reg (0) or mem (1) */
1759 		if (idx_value & 0x10) {
1760 			uint64_t offset;
1761 
1762 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1763 			if (r) {
1764 				DRM_ERROR("bad WAIT_REG_MEM\n");
1765 				return -EINVAL;
1766 			}
1767 
1768 			offset = reloc->gpu_offset +
1769 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff0) +
1770 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1771 
1772 			ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffff0);
1773 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1774 		} else if (idx_value & 0x100) {
1775 			DRM_ERROR("cannot use PFP on REG wait\n");
1776 			return -EINVAL;
1777 		}
1778 		break;
1779 	case PACKET3_CP_DMA:
1780 	{
1781 		u32 command, size;
1782 		u64 offset, tmp;
1783 		if (pkt->count != 4) {
1784 			DRM_ERROR("bad CP DMA\n");
1785 			return -EINVAL;
1786 		}
1787 		command = radeon_get_ib_value(p, idx+4);
1788 		size = command & 0x1fffff;
1789 		if (command & PACKET3_CP_DMA_CMD_SAS) {
1790 			/* src address space is register */
1791 			DRM_ERROR("CP DMA SAS not supported\n");
1792 			return -EINVAL;
1793 		} else {
1794 			if (command & PACKET3_CP_DMA_CMD_SAIC) {
1795 				DRM_ERROR("CP DMA SAIC only supported for registers\n");
1796 				return -EINVAL;
1797 			}
1798 			/* src address space is memory */
1799 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1800 			if (r) {
1801 				DRM_ERROR("bad CP DMA SRC\n");
1802 				return -EINVAL;
1803 			}
1804 
1805 			tmp = radeon_get_ib_value(p, idx) +
1806 				((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1807 
1808 			offset = reloc->gpu_offset + tmp;
1809 
1810 			if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1811 				dev_warn(p->dev, "CP DMA src buffer too small (%llu %lu)\n",
1812 					 tmp + size, radeon_bo_size(reloc->robj));
1813 				return -EINVAL;
1814 			}
1815 
1816 			ib[idx] = offset;
1817 			ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1818 		}
1819 		if (command & PACKET3_CP_DMA_CMD_DAS) {
1820 			/* dst address space is register */
1821 			DRM_ERROR("CP DMA DAS not supported\n");
1822 			return -EINVAL;
1823 		} else {
1824 			/* dst address space is memory */
1825 			if (command & PACKET3_CP_DMA_CMD_DAIC) {
1826 				DRM_ERROR("CP DMA DAIC only supported for registers\n");
1827 				return -EINVAL;
1828 			}
1829 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1830 			if (r) {
1831 				DRM_ERROR("bad CP DMA DST\n");
1832 				return -EINVAL;
1833 			}
1834 
1835 			tmp = radeon_get_ib_value(p, idx+2) +
1836 				((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
1837 
1838 			offset = reloc->gpu_offset + tmp;
1839 
1840 			if ((tmp + size) > radeon_bo_size(reloc->robj)) {
1841 				dev_warn(p->dev, "CP DMA dst buffer too small (%llu %lu)\n",
1842 					 tmp + size, radeon_bo_size(reloc->robj));
1843 				return -EINVAL;
1844 			}
1845 
1846 			ib[idx+2] = offset;
1847 			ib[idx+3] = upper_32_bits(offset) & 0xff;
1848 		}
1849 		break;
1850 	}
1851 	case PACKET3_SURFACE_SYNC:
1852 		if (pkt->count != 3) {
1853 			DRM_ERROR("bad SURFACE_SYNC\n");
1854 			return -EINVAL;
1855 		}
1856 		/* 0xffffffff/0x0 is flush all cache flag */
1857 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1858 		    radeon_get_ib_value(p, idx + 2) != 0) {
1859 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1860 			if (r) {
1861 				DRM_ERROR("bad SURFACE_SYNC\n");
1862 				return -EINVAL;
1863 			}
1864 			ib[idx+2] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1865 		}
1866 		break;
1867 	case PACKET3_EVENT_WRITE:
1868 		if (pkt->count != 2 && pkt->count != 0) {
1869 			DRM_ERROR("bad EVENT_WRITE\n");
1870 			return -EINVAL;
1871 		}
1872 		if (pkt->count) {
1873 			uint64_t offset;
1874 
1875 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1876 			if (r) {
1877 				DRM_ERROR("bad EVENT_WRITE\n");
1878 				return -EINVAL;
1879 			}
1880 			offset = reloc->gpu_offset +
1881 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
1882 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1883 
1884 			ib[idx+1] = offset & 0xfffffff8;
1885 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1886 		}
1887 		break;
1888 	case PACKET3_EVENT_WRITE_EOP:
1889 	{
1890 		uint64_t offset;
1891 
1892 		if (pkt->count != 4) {
1893 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1894 			return -EINVAL;
1895 		}
1896 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1897 		if (r) {
1898 			DRM_ERROR("bad EVENT_WRITE\n");
1899 			return -EINVAL;
1900 		}
1901 
1902 		offset = reloc->gpu_offset +
1903 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
1904 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1905 
1906 		ib[idx+1] = offset & 0xfffffffc;
1907 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1908 		break;
1909 	}
1910 	case PACKET3_SET_CONFIG_REG:
1911 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1912 		end_reg = 4 * pkt->count + start_reg - 4;
1913 		if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1914 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1915 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1916 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1917 			return -EINVAL;
1918 		}
1919 		for (i = 0; i < pkt->count; i++) {
1920 			reg = start_reg + (4 * i);
1921 			r = r600_cs_check_reg(p, reg, idx+1+i);
1922 			if (r)
1923 				return r;
1924 		}
1925 		break;
1926 	case PACKET3_SET_CONTEXT_REG:
1927 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1928 		end_reg = 4 * pkt->count + start_reg - 4;
1929 		if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1930 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1931 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1932 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1933 			return -EINVAL;
1934 		}
1935 		for (i = 0; i < pkt->count; i++) {
1936 			reg = start_reg + (4 * i);
1937 			r = r600_cs_check_reg(p, reg, idx+1+i);
1938 			if (r)
1939 				return r;
1940 		}
1941 		break;
1942 	case PACKET3_SET_RESOURCE:
1943 		if (pkt->count % 7) {
1944 			DRM_ERROR("bad SET_RESOURCE\n");
1945 			return -EINVAL;
1946 		}
1947 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1948 		end_reg = 4 * pkt->count + start_reg - 4;
1949 		if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1950 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
1951 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
1952 			DRM_ERROR("bad SET_RESOURCE\n");
1953 			return -EINVAL;
1954 		}
1955 		for (i = 0; i < (pkt->count / 7); i++) {
1956 			struct radeon_bo *texture, *mipmap;
1957 			u32 size, offset, base_offset, mip_offset;
1958 
1959 			switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1960 			case SQ_TEX_VTX_VALID_TEXTURE:
1961 				/* tex base */
1962 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1963 				if (r) {
1964 					DRM_ERROR("bad SET_RESOURCE\n");
1965 					return -EINVAL;
1966 				}
1967 				base_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1968 				if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1969 					if (reloc->tiling_flags & RADEON_TILING_MACRO)
1970 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1971 					else if (reloc->tiling_flags & RADEON_TILING_MICRO)
1972 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1973 				}
1974 				texture = reloc->robj;
1975 				/* tex mip base */
1976 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1977 				if (r) {
1978 					DRM_ERROR("bad SET_RESOURCE\n");
1979 					return -EINVAL;
1980 				}
1981 				mip_offset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1982 				mipmap = reloc->robj;
1983 				r = r600_check_texture_resource(p,  idx+(i*7)+1,
1984 								texture, mipmap,
1985 								base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1986 								mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1987 								reloc->tiling_flags);
1988 				if (r)
1989 					return r;
1990 				ib[idx+1+(i*7)+2] += base_offset;
1991 				ib[idx+1+(i*7)+3] += mip_offset;
1992 				break;
1993 			case SQ_TEX_VTX_VALID_BUFFER:
1994 			{
1995 				uint64_t offset64;
1996 				/* vtx base */
1997 				r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
1998 				if (r) {
1999 					DRM_ERROR("bad SET_RESOURCE\n");
2000 					return -EINVAL;
2001 				}
2002 				offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
2003 				size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
2004 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2005 					/* force size to size of the buffer */
2006 					dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
2007 						 size + offset, radeon_bo_size(reloc->robj));
2008 					ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj) - offset;
2009 				}
2010 
2011 				offset64 = reloc->gpu_offset + offset;
2012 				ib[idx+1+(i*8)+0] = offset64;
2013 				ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2014 						    (upper_32_bits(offset64) & 0xff);
2015 				break;
2016 			}
2017 			case SQ_TEX_VTX_INVALID_TEXTURE:
2018 			case SQ_TEX_VTX_INVALID_BUFFER:
2019 			default:
2020 				DRM_ERROR("bad SET_RESOURCE\n");
2021 				return -EINVAL;
2022 			}
2023 		}
2024 		break;
2025 	case PACKET3_SET_ALU_CONST:
2026 		if (track->sq_config & DX9_CONSTS) {
2027 			start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
2028 			end_reg = 4 * pkt->count + start_reg - 4;
2029 			if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
2030 			    (start_reg >= PACKET3_SET_ALU_CONST_END) ||
2031 			    (end_reg >= PACKET3_SET_ALU_CONST_END)) {
2032 				DRM_ERROR("bad SET_ALU_CONST\n");
2033 				return -EINVAL;
2034 			}
2035 		}
2036 		break;
2037 	case PACKET3_SET_BOOL_CONST:
2038 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
2039 		end_reg = 4 * pkt->count + start_reg - 4;
2040 		if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
2041 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2042 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2043 			DRM_ERROR("bad SET_BOOL_CONST\n");
2044 			return -EINVAL;
2045 		}
2046 		break;
2047 	case PACKET3_SET_LOOP_CONST:
2048 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
2049 		end_reg = 4 * pkt->count + start_reg - 4;
2050 		if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
2051 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2052 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2053 			DRM_ERROR("bad SET_LOOP_CONST\n");
2054 			return -EINVAL;
2055 		}
2056 		break;
2057 	case PACKET3_SET_CTL_CONST:
2058 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
2059 		end_reg = 4 * pkt->count + start_reg - 4;
2060 		if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
2061 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2062 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2063 			DRM_ERROR("bad SET_CTL_CONST\n");
2064 			return -EINVAL;
2065 		}
2066 		break;
2067 	case PACKET3_SET_SAMPLER:
2068 		if (pkt->count % 3) {
2069 			DRM_ERROR("bad SET_SAMPLER\n");
2070 			return -EINVAL;
2071 		}
2072 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
2073 		end_reg = 4 * pkt->count + start_reg - 4;
2074 		if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
2075 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
2076 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
2077 			DRM_ERROR("bad SET_SAMPLER\n");
2078 			return -EINVAL;
2079 		}
2080 		break;
2081 	case PACKET3_STRMOUT_BASE_UPDATE:
2082 		/* RS780 and RS880 also need this */
2083 		if (p->family < CHIP_RS780) {
2084 			DRM_ERROR("STRMOUT_BASE_UPDATE only supported on 7xx\n");
2085 			return -EINVAL;
2086 		}
2087 		if (pkt->count != 1) {
2088 			DRM_ERROR("bad STRMOUT_BASE_UPDATE packet count\n");
2089 			return -EINVAL;
2090 		}
2091 		if (idx_value > 3) {
2092 			DRM_ERROR("bad STRMOUT_BASE_UPDATE index\n");
2093 			return -EINVAL;
2094 		}
2095 		{
2096 			u64 offset;
2097 
2098 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2099 			if (r) {
2100 				DRM_ERROR("bad STRMOUT_BASE_UPDATE reloc\n");
2101 				return -EINVAL;
2102 			}
2103 
2104 			if (reloc->robj != track->vgt_strmout_bo[idx_value]) {
2105 				DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo does not match\n");
2106 				return -EINVAL;
2107 			}
2108 
2109 			offset = radeon_get_ib_value(p, idx+1) << 8;
2110 			if (offset != track->vgt_strmout_bo_offset[idx_value]) {
2111 				DRM_ERROR("bad STRMOUT_BASE_UPDATE, bo offset does not match: 0x%llx, 0x%x\n",
2112 					  offset, track->vgt_strmout_bo_offset[idx_value]);
2113 				return -EINVAL;
2114 			}
2115 
2116 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2117 				DRM_ERROR("bad STRMOUT_BASE_UPDATE bo too small: 0x%llx, 0x%lx\n",
2118 					  offset + 4, radeon_bo_size(reloc->robj));
2119 				return -EINVAL;
2120 			}
2121 			ib[idx+1] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2122 		}
2123 		break;
2124 	case PACKET3_SURFACE_BASE_UPDATE:
2125 		if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
2126 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2127 			return -EINVAL;
2128 		}
2129 		if (pkt->count) {
2130 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2131 			return -EINVAL;
2132 		}
2133 		break;
2134 	case PACKET3_STRMOUT_BUFFER_UPDATE:
2135 		if (pkt->count != 4) {
2136 			DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2137 			return -EINVAL;
2138 		}
2139 		/* Updating memory at DST_ADDRESS. */
2140 		if (idx_value & 0x1) {
2141 			u64 offset;
2142 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2143 			if (r) {
2144 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2145 				return -EINVAL;
2146 			}
2147 			offset = radeon_get_ib_value(p, idx+1);
2148 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2149 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2150 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2151 					  offset + 4, radeon_bo_size(reloc->robj));
2152 				return -EINVAL;
2153 			}
2154 			offset += reloc->gpu_offset;
2155 			ib[idx+1] = offset;
2156 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2157 		}
2158 		/* Reading data from SRC_ADDRESS. */
2159 		if (((idx_value >> 1) & 0x3) == 2) {
2160 			u64 offset;
2161 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2162 			if (r) {
2163 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2164 				return -EINVAL;
2165 			}
2166 			offset = radeon_get_ib_value(p, idx+3);
2167 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2168 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2169 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2170 					  offset + 4, radeon_bo_size(reloc->robj));
2171 				return -EINVAL;
2172 			}
2173 			offset += reloc->gpu_offset;
2174 			ib[idx+3] = offset;
2175 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2176 		}
2177 		break;
2178 	case PACKET3_MEM_WRITE:
2179 	{
2180 		u64 offset;
2181 
2182 		if (pkt->count != 3) {
2183 			DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2184 			return -EINVAL;
2185 		}
2186 		r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2187 		if (r) {
2188 			DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2189 			return -EINVAL;
2190 		}
2191 		offset = radeon_get_ib_value(p, idx+0);
2192 		offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2193 		if (offset & 0x7) {
2194 			DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2195 			return -EINVAL;
2196 		}
2197 		if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2198 			DRM_ERROR("bad MEM_WRITE bo too small: 0x%llx, 0x%lx\n",
2199 				  offset + 8, radeon_bo_size(reloc->robj));
2200 			return -EINVAL;
2201 		}
2202 		offset += reloc->gpu_offset;
2203 		ib[idx+0] = offset;
2204 		ib[idx+1] = upper_32_bits(offset) & 0xff;
2205 		break;
2206 	}
2207 	case PACKET3_COPY_DW:
2208 		if (pkt->count != 4) {
2209 			DRM_ERROR("bad COPY_DW (invalid count)\n");
2210 			return -EINVAL;
2211 		}
2212 		if (idx_value & 0x1) {
2213 			u64 offset;
2214 			/* SRC is memory. */
2215 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2216 			if (r) {
2217 				DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2218 				return -EINVAL;
2219 			}
2220 			offset = radeon_get_ib_value(p, idx+1);
2221 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2222 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2223 				DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2224 					  offset + 4, radeon_bo_size(reloc->robj));
2225 				return -EINVAL;
2226 			}
2227 			offset += reloc->gpu_offset;
2228 			ib[idx+1] = offset;
2229 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2230 		} else {
2231 			/* SRC is a reg. */
2232 			reg = radeon_get_ib_value(p, idx+1) << 2;
2233 			if (!r600_is_safe_reg(p, reg, idx+1))
2234 				return -EINVAL;
2235 		}
2236 		if (idx_value & 0x2) {
2237 			u64 offset;
2238 			/* DST is memory. */
2239 			r = radeon_cs_packet_next_reloc(p, &reloc, r600_nomm);
2240 			if (r) {
2241 				DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2242 				return -EINVAL;
2243 			}
2244 			offset = radeon_get_ib_value(p, idx+3);
2245 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2246 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2247 				DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2248 					  offset + 4, radeon_bo_size(reloc->robj));
2249 				return -EINVAL;
2250 			}
2251 			offset += reloc->gpu_offset;
2252 			ib[idx+3] = offset;
2253 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2254 		} else {
2255 			/* DST is a reg. */
2256 			reg = radeon_get_ib_value(p, idx+3) << 2;
2257 			if (!r600_is_safe_reg(p, reg, idx+3))
2258 				return -EINVAL;
2259 		}
2260 		break;
2261 	case PACKET3_NOP:
2262 		break;
2263 	default:
2264 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2265 		return -EINVAL;
2266 	}
2267 	return 0;
2268 }
2269 
2270 int r600_cs_parse(struct radeon_cs_parser *p)
2271 {
2272 	struct radeon_cs_packet pkt;
2273 	struct r600_cs_track *track;
2274 	int r;
2275 
2276 	if (p->track == NULL) {
2277 		/* initialize tracker, we are in kms */
2278 		track = kzalloc(sizeof(*track), GFP_KERNEL);
2279 		if (track == NULL)
2280 			return -ENOMEM;
2281 		r600_cs_track_init(track);
2282 		if (p->rdev->family < CHIP_RV770) {
2283 			track->npipes = p->rdev->config.r600.tiling_npipes;
2284 			track->nbanks = p->rdev->config.r600.tiling_nbanks;
2285 			track->group_size = p->rdev->config.r600.tiling_group_size;
2286 		} else if (p->rdev->family <= CHIP_RV740) {
2287 			track->npipes = p->rdev->config.rv770.tiling_npipes;
2288 			track->nbanks = p->rdev->config.rv770.tiling_nbanks;
2289 			track->group_size = p->rdev->config.rv770.tiling_group_size;
2290 		}
2291 		p->track = track;
2292 	}
2293 	do {
2294 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2295 		if (r) {
2296 			kfree(p->track);
2297 			p->track = NULL;
2298 			return r;
2299 		}
2300 		p->idx += pkt.count + 2;
2301 		switch (pkt.type) {
2302 		case RADEON_PACKET_TYPE0:
2303 			r = r600_cs_parse_packet0(p, &pkt);
2304 			break;
2305 		case RADEON_PACKET_TYPE2:
2306 			break;
2307 		case RADEON_PACKET_TYPE3:
2308 			r = r600_packet3_check(p, &pkt);
2309 			break;
2310 		default:
2311 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2312 			kfree(p->track);
2313 			p->track = NULL;
2314 			return -EINVAL;
2315 		}
2316 		if (r) {
2317 			kfree(p->track);
2318 			p->track = NULL;
2319 			return r;
2320 		}
2321 	} while (p->idx < p->chunk_ib->length_dw);
2322 #if 0
2323 	for (r = 0; r < p->ib.length_dw; r++) {
2324 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2325 		mdelay(1);
2326 	}
2327 #endif
2328 	kfree(p->track);
2329 	p->track = NULL;
2330 	return 0;
2331 }
2332 
2333 /*
2334  *  DMA
2335  */
2336 /**
2337  * r600_dma_cs_next_reloc() - parse next reloc
2338  * @p:		parser structure holding parsing context.
2339  * @cs_reloc:		reloc information
2340  *
2341  * Return the next reloc, do bo validation and compute
2342  * GPU offset using the provided start.
2343  **/
2344 int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
2345 			   struct radeon_bo_list **cs_reloc)
2346 {
2347 	unsigned idx;
2348 
2349 	*cs_reloc = NULL;
2350 	if (p->chunk_relocs == NULL) {
2351 		DRM_ERROR("No relocation chunk !\n");
2352 		return -EINVAL;
2353 	}
2354 	idx = p->dma_reloc_idx;
2355 	if (idx >= p->nrelocs) {
2356 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
2357 			  idx, p->nrelocs);
2358 		return -EINVAL;
2359 	}
2360 	*cs_reloc = &p->relocs[idx];
2361 	p->dma_reloc_idx++;
2362 	return 0;
2363 }
2364 
2365 #define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
2366 #define GET_DMA_COUNT(h) ((h) & 0x0000ffff)
2367 #define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
2368 
2369 /**
2370  * r600_dma_cs_parse() - parse the DMA IB
2371  * @p:		parser structure holding parsing context.
2372  *
2373  * Parses the DMA IB from the CS ioctl and updates
2374  * the GPU addresses based on the reloc information and
2375  * checks for errors. (R6xx-R7xx)
2376  * Returns 0 for success and an error on failure.
2377  **/
2378 int r600_dma_cs_parse(struct radeon_cs_parser *p)
2379 {
2380 	struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
2381 	struct radeon_bo_list *src_reloc, *dst_reloc;
2382 	u32 header, cmd, count, tiled;
2383 	volatile u32 *ib = p->ib.ptr;
2384 	u32 idx, idx_value;
2385 	u64 src_offset, dst_offset;
2386 	int r;
2387 
2388 	do {
2389 		if (p->idx >= ib_chunk->length_dw) {
2390 			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2391 				  p->idx, ib_chunk->length_dw);
2392 			return -EINVAL;
2393 		}
2394 		idx = p->idx;
2395 		header = radeon_get_ib_value(p, idx);
2396 		cmd = GET_DMA_CMD(header);
2397 		count = GET_DMA_COUNT(header);
2398 		tiled = GET_DMA_T(header);
2399 
2400 		switch (cmd) {
2401 		case DMA_PACKET_WRITE:
2402 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2403 			if (r) {
2404 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2405 				return -EINVAL;
2406 			}
2407 			if (tiled) {
2408 				dst_offset = radeon_get_ib_value(p, idx+1);
2409 				dst_offset <<= 8;
2410 
2411 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2412 				p->idx += count + 5;
2413 			} else {
2414 				dst_offset = radeon_get_ib_value(p, idx+1);
2415 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2416 
2417 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2418 				ib[idx+2] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2419 				p->idx += count + 3;
2420 			}
2421 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2422 				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
2423 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2424 				return -EINVAL;
2425 			}
2426 			break;
2427 		case DMA_PACKET_COPY:
2428 			r = r600_dma_cs_next_reloc(p, &src_reloc);
2429 			if (r) {
2430 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2431 				return -EINVAL;
2432 			}
2433 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2434 			if (r) {
2435 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2436 				return -EINVAL;
2437 			}
2438 			if (tiled) {
2439 				idx_value = radeon_get_ib_value(p, idx + 2);
2440 				/* detile bit */
2441 				if (idx_value & (1 << 31)) {
2442 					/* tiled src, linear dst */
2443 					src_offset = radeon_get_ib_value(p, idx+1);
2444 					src_offset <<= 8;
2445 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
2446 
2447 					dst_offset = radeon_get_ib_value(p, idx+5);
2448 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2449 					ib[idx+5] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2450 					ib[idx+6] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2451 				} else {
2452 					/* linear src, tiled dst */
2453 					src_offset = radeon_get_ib_value(p, idx+5);
2454 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2455 					ib[idx+5] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2456 					ib[idx+6] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2457 
2458 					dst_offset = radeon_get_ib_value(p, idx+1);
2459 					dst_offset <<= 8;
2460 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2461 				}
2462 				p->idx += 7;
2463 			} else {
2464 				if (p->family >= CHIP_RV770) {
2465 					src_offset = radeon_get_ib_value(p, idx+2);
2466 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2467 					dst_offset = radeon_get_ib_value(p, idx+1);
2468 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2469 
2470 					ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2471 					ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2472 					ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2473 					ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2474 					p->idx += 5;
2475 				} else {
2476 					src_offset = radeon_get_ib_value(p, idx+2);
2477 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2478 					dst_offset = radeon_get_ib_value(p, idx+1);
2479 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff0000)) << 16;
2480 
2481 					ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2482 					ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2483 					ib[idx+3] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2484 					ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) & 0xff) << 16;
2485 					p->idx += 4;
2486 				}
2487 			}
2488 			if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2489 				dev_warn(p->dev, "DMA copy src buffer too small (%llu %lu)\n",
2490 					 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2491 				return -EINVAL;
2492 			}
2493 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2494 				dev_warn(p->dev, "DMA write dst buffer too small (%llu %lu)\n",
2495 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2496 				return -EINVAL;
2497 			}
2498 			break;
2499 		case DMA_PACKET_CONSTANT_FILL:
2500 			if (p->family < CHIP_RV770) {
2501 				DRM_ERROR("Constant Fill is 7xx only !\n");
2502 				return -EINVAL;
2503 			}
2504 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2505 			if (r) {
2506 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2507 				return -EINVAL;
2508 			}
2509 			dst_offset = radeon_get_ib_value(p, idx+1);
2510 			dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
2511 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2512 				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
2513 					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2514 				return -EINVAL;
2515 			}
2516 			ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2517 			ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) << 16) & 0x00ff0000;
2518 			p->idx += 4;
2519 			break;
2520 		case DMA_PACKET_NOP:
2521 			p->idx += 1;
2522 			break;
2523 		default:
2524 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
2525 			return -EINVAL;
2526 		}
2527 	} while (p->idx < p->chunk_ib->length_dw);
2528 #if 0
2529 	for (r = 0; r < p->ib->length_dw; r++) {
2530 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2531 		mdelay(1);
2532 	}
2533 #endif
2534 	return 0;
2535 }
2536