xref: /linux/drivers/gpu/drm/radeon/evergreen_cs.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /*
2  * Copyright 2010 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 
29 #include "radeon.h"
30 #include "radeon_asic.h"
31 #include "r600.h"
32 #include "evergreend.h"
33 #include "evergreen_reg_safe.h"
34 #include "cayman_reg_safe.h"
35 
36 #ifndef MIN
37 #define MAX(a, b)                   (((a) > (b)) ? (a) : (b))
38 #define MIN(a, b)                   (((a) < (b)) ? (a) : (b))
39 #endif
40 
41 #define REG_SAFE_BM_SIZE ARRAY_SIZE(evergreen_reg_safe_bm)
42 
43 struct evergreen_cs_track {
44 	u32			group_size;
45 	u32			nbanks;
46 	u32			npipes;
47 	u32			row_size;
48 	/* value we track */
49 	u32			nsamples;		/* unused */
50 	struct radeon_bo	*cb_color_bo[12];
51 	u32			cb_color_bo_offset[12];
52 	struct radeon_bo	*cb_color_fmask_bo[8];	/* unused */
53 	struct radeon_bo	*cb_color_cmask_bo[8];	/* unused */
54 	u32			cb_color_info[12];
55 	u32			cb_color_view[12];
56 	u32			cb_color_pitch[12];
57 	u32			cb_color_slice[12];
58 	u32			cb_color_slice_idx[12];
59 	u32			cb_color_attrib[12];
60 	u32			cb_color_cmask_slice[8];/* unused */
61 	u32			cb_color_fmask_slice[8];/* unused */
62 	u32			cb_target_mask;
63 	u32			cb_shader_mask; /* unused */
64 	u32			vgt_strmout_config;
65 	u32			vgt_strmout_buffer_config;
66 	struct radeon_bo	*vgt_strmout_bo[4];
67 	u32			vgt_strmout_bo_offset[4];
68 	u32			vgt_strmout_size[4];
69 	u32			db_depth_control;
70 	u32			db_depth_view;
71 	u32			db_depth_slice;
72 	u32			db_depth_size;
73 	u32			db_z_info;
74 	u32			db_z_read_offset;
75 	u32			db_z_write_offset;
76 	struct radeon_bo	*db_z_read_bo;
77 	struct radeon_bo	*db_z_write_bo;
78 	u32			db_s_info;
79 	u32			db_s_read_offset;
80 	u32			db_s_write_offset;
81 	struct radeon_bo	*db_s_read_bo;
82 	struct radeon_bo	*db_s_write_bo;
83 	bool			sx_misc_kill_all_prims;
84 	bool			cb_dirty;
85 	bool			db_dirty;
86 	bool			streamout_dirty;
87 	u32			htile_offset;
88 	u32			htile_surface;
89 	struct radeon_bo	*htile_bo;
90 	unsigned long		indirect_draw_buffer_size;
91 	const unsigned		*reg_safe_bm;
92 };
93 
94 static u32 evergreen_cs_get_aray_mode(u32 tiling_flags)
95 {
96 	if (tiling_flags & RADEON_TILING_MACRO)
97 		return ARRAY_2D_TILED_THIN1;
98 	else if (tiling_flags & RADEON_TILING_MICRO)
99 		return ARRAY_1D_TILED_THIN1;
100 	else
101 		return ARRAY_LINEAR_GENERAL;
102 }
103 
104 static u32 evergreen_cs_get_num_banks(u32 nbanks)
105 {
106 	switch (nbanks) {
107 	case 2:
108 		return ADDR_SURF_2_BANK;
109 	case 4:
110 		return ADDR_SURF_4_BANK;
111 	case 8:
112 	default:
113 		return ADDR_SURF_8_BANK;
114 	case 16:
115 		return ADDR_SURF_16_BANK;
116 	}
117 }
118 
119 static void evergreen_cs_track_init(struct evergreen_cs_track *track)
120 {
121 	int i;
122 
123 	for (i = 0; i < 8; i++) {
124 		track->cb_color_fmask_bo[i] = NULL;
125 		track->cb_color_cmask_bo[i] = NULL;
126 		track->cb_color_cmask_slice[i] = 0;
127 		track->cb_color_fmask_slice[i] = 0;
128 	}
129 
130 	for (i = 0; i < 12; i++) {
131 		track->cb_color_bo[i] = NULL;
132 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
133 		track->cb_color_info[i] = 0;
134 		track->cb_color_view[i] = 0xFFFFFFFF;
135 		track->cb_color_pitch[i] = 0;
136 		track->cb_color_slice[i] = 0xfffffff;
137 		track->cb_color_slice_idx[i] = 0;
138 	}
139 	track->cb_target_mask = 0xFFFFFFFF;
140 	track->cb_shader_mask = 0xFFFFFFFF;
141 	track->cb_dirty = true;
142 
143 	track->db_depth_slice = 0xffffffff;
144 	track->db_depth_view = 0xFFFFC000;
145 	track->db_depth_size = 0xFFFFFFFF;
146 	track->db_depth_control = 0xFFFFFFFF;
147 	track->db_z_info = 0xFFFFFFFF;
148 	track->db_z_read_offset = 0xFFFFFFFF;
149 	track->db_z_write_offset = 0xFFFFFFFF;
150 	track->db_z_read_bo = NULL;
151 	track->db_z_write_bo = NULL;
152 	track->db_s_info = 0xFFFFFFFF;
153 	track->db_s_read_offset = 0xFFFFFFFF;
154 	track->db_s_write_offset = 0xFFFFFFFF;
155 	track->db_s_read_bo = NULL;
156 	track->db_s_write_bo = NULL;
157 	track->db_dirty = true;
158 	track->htile_bo = NULL;
159 	track->htile_offset = 0xFFFFFFFF;
160 	track->htile_surface = 0;
161 
162 	for (i = 0; i < 4; i++) {
163 		track->vgt_strmout_size[i] = 0;
164 		track->vgt_strmout_bo[i] = NULL;
165 		track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
166 	}
167 	track->streamout_dirty = true;
168 	track->sx_misc_kill_all_prims = false;
169 }
170 
171 struct eg_surface {
172 	/* value gathered from cs */
173 	unsigned	nbx;
174 	unsigned	nby;
175 	unsigned	format;
176 	unsigned	mode;
177 	unsigned	nbanks;
178 	unsigned	bankw;
179 	unsigned	bankh;
180 	unsigned	tsplit;
181 	unsigned	mtilea;
182 	unsigned	nsamples;
183 	/* output value */
184 	unsigned	bpe;
185 	unsigned	layer_size;
186 	unsigned	palign;
187 	unsigned	halign;
188 	unsigned long	base_align;
189 };
190 
191 static int evergreen_surface_check_linear(struct radeon_cs_parser *p,
192 					  struct eg_surface *surf,
193 					  const char *prefix)
194 {
195 	surf->layer_size = surf->nbx * surf->nby * surf->bpe * surf->nsamples;
196 	surf->base_align = surf->bpe;
197 	surf->palign = 1;
198 	surf->halign = 1;
199 	return 0;
200 }
201 
202 static int evergreen_surface_check_linear_aligned(struct radeon_cs_parser *p,
203 						  struct eg_surface *surf,
204 						  const char *prefix)
205 {
206 	struct evergreen_cs_track *track = p->track;
207 	unsigned palign;
208 
209 	palign = MAX(64, track->group_size / surf->bpe);
210 	surf->layer_size = surf->nbx * surf->nby * surf->bpe * surf->nsamples;
211 	surf->base_align = track->group_size;
212 	surf->palign = palign;
213 	surf->halign = 1;
214 	if (surf->nbx & (palign - 1)) {
215 		if (prefix) {
216 			dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d\n",
217 				 __func__, __LINE__, prefix, surf->nbx, palign);
218 		}
219 		return -EINVAL;
220 	}
221 	return 0;
222 }
223 
224 static int evergreen_surface_check_1d(struct radeon_cs_parser *p,
225 				      struct eg_surface *surf,
226 				      const char *prefix)
227 {
228 	struct evergreen_cs_track *track = p->track;
229 	unsigned palign;
230 
231 	palign = track->group_size / (8 * surf->bpe * surf->nsamples);
232 	palign = MAX(8, palign);
233 	surf->layer_size = surf->nbx * surf->nby * surf->bpe;
234 	surf->base_align = track->group_size;
235 	surf->palign = palign;
236 	surf->halign = 8;
237 	if ((surf->nbx & (palign - 1))) {
238 		if (prefix) {
239 			dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d (%d %d %d)\n",
240 				 __func__, __LINE__, prefix, surf->nbx, palign,
241 				 track->group_size, surf->bpe, surf->nsamples);
242 		}
243 		return -EINVAL;
244 	}
245 	if ((surf->nby & (8 - 1))) {
246 		if (prefix) {
247 			dev_warn(p->dev, "%s:%d %s height %d invalid must be aligned with 8\n",
248 				 __func__, __LINE__, prefix, surf->nby);
249 		}
250 		return -EINVAL;
251 	}
252 	return 0;
253 }
254 
255 static int evergreen_surface_check_2d(struct radeon_cs_parser *p,
256 				      struct eg_surface *surf,
257 				      const char *prefix)
258 {
259 	struct evergreen_cs_track *track = p->track;
260 	unsigned palign, halign, tileb, slice_pt;
261 	unsigned mtile_pr, mtile_ps, mtileb;
262 
263 	tileb = 64 * surf->bpe * surf->nsamples;
264 	slice_pt = 1;
265 	if (tileb > surf->tsplit) {
266 		slice_pt = tileb / surf->tsplit;
267 	}
268 	tileb = tileb / slice_pt;
269 	/* macro tile width & height */
270 	palign = (8 * surf->bankw * track->npipes) * surf->mtilea;
271 	halign = (8 * surf->bankh * surf->nbanks) / surf->mtilea;
272 	mtileb = (palign / 8) * (halign / 8) * tileb;
273 	mtile_pr = surf->nbx / palign;
274 	mtile_ps = (mtile_pr * surf->nby) / halign;
275 	surf->layer_size = mtile_ps * mtileb * slice_pt;
276 	surf->base_align = (palign / 8) * (halign / 8) * tileb;
277 	surf->palign = palign;
278 	surf->halign = halign;
279 
280 	if ((surf->nbx & (palign - 1))) {
281 		if (prefix) {
282 			dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d\n",
283 				 __func__, __LINE__, prefix, surf->nbx, palign);
284 		}
285 		return -EINVAL;
286 	}
287 	if ((surf->nby & (halign - 1))) {
288 		if (prefix) {
289 			dev_warn(p->dev, "%s:%d %s height %d invalid must be aligned with %d\n",
290 				 __func__, __LINE__, prefix, surf->nby, halign);
291 		}
292 		return -EINVAL;
293 	}
294 
295 	return 0;
296 }
297 
298 static int evergreen_surface_check(struct radeon_cs_parser *p,
299 				   struct eg_surface *surf,
300 				   const char *prefix)
301 {
302 	/* some common value computed here */
303 	surf->bpe = r600_fmt_get_blocksize(surf->format);
304 
305 	switch (surf->mode) {
306 	case ARRAY_LINEAR_GENERAL:
307 		return evergreen_surface_check_linear(p, surf, prefix);
308 	case ARRAY_LINEAR_ALIGNED:
309 		return evergreen_surface_check_linear_aligned(p, surf, prefix);
310 	case ARRAY_1D_TILED_THIN1:
311 		return evergreen_surface_check_1d(p, surf, prefix);
312 	case ARRAY_2D_TILED_THIN1:
313 		return evergreen_surface_check_2d(p, surf, prefix);
314 	default:
315 		dev_warn(p->dev, "%s:%d %s invalid array mode %d\n",
316 				__func__, __LINE__, prefix, surf->mode);
317 		return -EINVAL;
318 	}
319 	return -EINVAL;
320 }
321 
322 static int evergreen_surface_value_conv_check(struct radeon_cs_parser *p,
323 					      struct eg_surface *surf,
324 					      const char *prefix)
325 {
326 	switch (surf->mode) {
327 	case ARRAY_2D_TILED_THIN1:
328 		break;
329 	case ARRAY_LINEAR_GENERAL:
330 	case ARRAY_LINEAR_ALIGNED:
331 	case ARRAY_1D_TILED_THIN1:
332 		return 0;
333 	default:
334 		dev_warn(p->dev, "%s:%d %s invalid array mode %d\n",
335 				__func__, __LINE__, prefix, surf->mode);
336 		return -EINVAL;
337 	}
338 
339 	switch (surf->nbanks) {
340 	case 0: surf->nbanks = 2; break;
341 	case 1: surf->nbanks = 4; break;
342 	case 2: surf->nbanks = 8; break;
343 	case 3: surf->nbanks = 16; break;
344 	default:
345 		dev_warn(p->dev, "%s:%d %s invalid number of banks %d\n",
346 			 __func__, __LINE__, prefix, surf->nbanks);
347 		return -EINVAL;
348 	}
349 	switch (surf->bankw) {
350 	case 0: surf->bankw = 1; break;
351 	case 1: surf->bankw = 2; break;
352 	case 2: surf->bankw = 4; break;
353 	case 3: surf->bankw = 8; break;
354 	default:
355 		dev_warn(p->dev, "%s:%d %s invalid bankw %d\n",
356 			 __func__, __LINE__, prefix, surf->bankw);
357 		return -EINVAL;
358 	}
359 	switch (surf->bankh) {
360 	case 0: surf->bankh = 1; break;
361 	case 1: surf->bankh = 2; break;
362 	case 2: surf->bankh = 4; break;
363 	case 3: surf->bankh = 8; break;
364 	default:
365 		dev_warn(p->dev, "%s:%d %s invalid bankh %d\n",
366 			 __func__, __LINE__, prefix, surf->bankh);
367 		return -EINVAL;
368 	}
369 	switch (surf->mtilea) {
370 	case 0: surf->mtilea = 1; break;
371 	case 1: surf->mtilea = 2; break;
372 	case 2: surf->mtilea = 4; break;
373 	case 3: surf->mtilea = 8; break;
374 	default:
375 		dev_warn(p->dev, "%s:%d %s invalid macro tile aspect %d\n",
376 			 __func__, __LINE__, prefix, surf->mtilea);
377 		return -EINVAL;
378 	}
379 	switch (surf->tsplit) {
380 	case 0: surf->tsplit = 64; break;
381 	case 1: surf->tsplit = 128; break;
382 	case 2: surf->tsplit = 256; break;
383 	case 3: surf->tsplit = 512; break;
384 	case 4: surf->tsplit = 1024; break;
385 	case 5: surf->tsplit = 2048; break;
386 	case 6: surf->tsplit = 4096; break;
387 	default:
388 		dev_warn(p->dev, "%s:%d %s invalid tile split %d\n",
389 			 __func__, __LINE__, prefix, surf->tsplit);
390 		return -EINVAL;
391 	}
392 	return 0;
393 }
394 
395 static int evergreen_cs_track_validate_cb(struct radeon_cs_parser *p, unsigned id)
396 {
397 	struct evergreen_cs_track *track = p->track;
398 	struct eg_surface surf;
399 	unsigned pitch, slice, mslice;
400 	u64 offset;
401 	int r;
402 
403 	mslice = G_028C6C_SLICE_MAX(track->cb_color_view[id]) + 1;
404 	pitch = track->cb_color_pitch[id];
405 	slice = track->cb_color_slice[id];
406 	surf.nbx = (pitch + 1) * 8;
407 	surf.nby = ((slice + 1) * 64) / surf.nbx;
408 	surf.mode = G_028C70_ARRAY_MODE(track->cb_color_info[id]);
409 	surf.format = G_028C70_FORMAT(track->cb_color_info[id]);
410 	surf.tsplit = G_028C74_TILE_SPLIT(track->cb_color_attrib[id]);
411 	surf.nbanks = G_028C74_NUM_BANKS(track->cb_color_attrib[id]);
412 	surf.bankw = G_028C74_BANK_WIDTH(track->cb_color_attrib[id]);
413 	surf.bankh = G_028C74_BANK_HEIGHT(track->cb_color_attrib[id]);
414 	surf.mtilea = G_028C74_MACRO_TILE_ASPECT(track->cb_color_attrib[id]);
415 	surf.nsamples = 1;
416 
417 	if (!r600_fmt_is_valid_color(surf.format)) {
418 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08x)\n",
419 			 __func__, __LINE__, surf.format,
420 			id, track->cb_color_info[id]);
421 		return -EINVAL;
422 	}
423 
424 	r = evergreen_surface_value_conv_check(p, &surf, "cb");
425 	if (r) {
426 		return r;
427 	}
428 
429 	r = evergreen_surface_check(p, &surf, "cb");
430 	if (r) {
431 		dev_warn(p->dev, "%s:%d cb[%d] invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
432 			 __func__, __LINE__, id, track->cb_color_pitch[id],
433 			 track->cb_color_slice[id], track->cb_color_attrib[id],
434 			 track->cb_color_info[id]);
435 		return r;
436 	}
437 
438 	offset = (u64)track->cb_color_bo_offset[id] << 8;
439 	if (offset & (surf.base_align - 1)) {
440 		dev_warn(p->dev, "%s:%d cb[%d] bo base %llu not aligned with %ld\n",
441 			 __func__, __LINE__, id, offset, surf.base_align);
442 		return -EINVAL;
443 	}
444 
445 	offset += (u64)surf.layer_size * mslice;
446 	if (offset > radeon_bo_size(track->cb_color_bo[id])) {
447 		/* old ddx are broken they allocate bo with w*h*bpp but
448 		 * program slice with ALIGN(h, 8), catch this and patch
449 		 * command stream.
450 		 */
451 		if (!surf.mode) {
452 			uint32_t *ib = p->ib.ptr;
453 			u64 tmp, nby, bsize, size, min = 0;
454 
455 			/* find the height the ddx wants */
456 			if (surf.nby > 8) {
457 				min = surf.nby - 8;
458 			}
459 			bsize = radeon_bo_size(track->cb_color_bo[id]);
460 			tmp = (u64)track->cb_color_bo_offset[id] << 8;
461 			for (nby = surf.nby; nby > min; nby--) {
462 				size = nby * surf.nbx * surf.bpe * surf.nsamples;
463 				if ((tmp + size * mslice) <= bsize) {
464 					break;
465 				}
466 			}
467 			if (nby > min) {
468 				surf.nby = nby;
469 				slice = ((nby * surf.nbx) / 64) - 1;
470 				if (!evergreen_surface_check(p, &surf, "cb")) {
471 					/* check if this one works */
472 					tmp += (u64)surf.layer_size * mslice;
473 					if (tmp <= bsize) {
474 						ib[track->cb_color_slice_idx[id]] = slice;
475 						goto old_ddx_ok;
476 					}
477 				}
478 			}
479 		}
480 		dev_warn(p->dev, "%s:%d cb[%d] bo too small (layer size %d, "
481 			 "offset %llu, max layer %d, bo size %ld, slice %d)\n",
482 			 __func__, __LINE__, id, surf.layer_size,
483 			(u64)track->cb_color_bo_offset[id] << 8, mslice,
484 			radeon_bo_size(track->cb_color_bo[id]), slice);
485 		dev_warn(p->dev, "%s:%d problematic surf: (%d %d) (%d %d %d %d %d %d %d)\n",
486 			 __func__, __LINE__, surf.nbx, surf.nby,
487 			surf.mode, surf.bpe, surf.nsamples,
488 			surf.bankw, surf.bankh,
489 			surf.tsplit, surf.mtilea);
490 		return -EINVAL;
491 	}
492 old_ddx_ok:
493 
494 	return 0;
495 }
496 
497 static int evergreen_cs_track_validate_htile(struct radeon_cs_parser *p,
498 						unsigned nbx, unsigned nby)
499 {
500 	struct evergreen_cs_track *track = p->track;
501 	unsigned long size;
502 
503 	if (track->htile_bo == NULL) {
504 		dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
505 				__func__, __LINE__, track->db_z_info);
506 		return -EINVAL;
507 	}
508 
509 	if (G_028ABC_LINEAR(track->htile_surface)) {
510 		/* pitch must be 16 htiles aligned == 16 * 8 pixel aligned */
511 		nbx = round_up(nbx, 16 * 8);
512 		/* height is npipes htiles aligned == npipes * 8 pixel aligned */
513 		nby = round_up(nby, track->npipes * 8);
514 	} else {
515 		/* always assume 8x8 htile */
516 		/* align is htile align * 8, htile align vary according to
517 		 * number of pipe and tile width and nby
518 		 */
519 		switch (track->npipes) {
520 		case 8:
521 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
522 			nbx = round_up(nbx, 64 * 8);
523 			nby = round_up(nby, 64 * 8);
524 			break;
525 		case 4:
526 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
527 			nbx = round_up(nbx, 64 * 8);
528 			nby = round_up(nby, 32 * 8);
529 			break;
530 		case 2:
531 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
532 			nbx = round_up(nbx, 32 * 8);
533 			nby = round_up(nby, 32 * 8);
534 			break;
535 		case 1:
536 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
537 			nbx = round_up(nbx, 32 * 8);
538 			nby = round_up(nby, 16 * 8);
539 			break;
540 		default:
541 			dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
542 					__func__, __LINE__, track->npipes);
543 			return -EINVAL;
544 		}
545 	}
546 	/* compute number of htile */
547 	nbx = nbx >> 3;
548 	nby = nby >> 3;
549 	/* size must be aligned on npipes * 2K boundary */
550 	size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
551 	size += track->htile_offset;
552 
553 	if (size > radeon_bo_size(track->htile_bo)) {
554 		dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
555 				__func__, __LINE__, radeon_bo_size(track->htile_bo),
556 				size, nbx, nby);
557 		return -EINVAL;
558 	}
559 	return 0;
560 }
561 
562 static int evergreen_cs_track_validate_stencil(struct radeon_cs_parser *p)
563 {
564 	struct evergreen_cs_track *track = p->track;
565 	struct eg_surface surf;
566 	unsigned pitch, slice, mslice;
567 	u64 offset;
568 	int r;
569 
570 	mslice = G_028008_SLICE_MAX(track->db_depth_view) + 1;
571 	pitch = G_028058_PITCH_TILE_MAX(track->db_depth_size);
572 	slice = track->db_depth_slice;
573 	surf.nbx = (pitch + 1) * 8;
574 	surf.nby = ((slice + 1) * 64) / surf.nbx;
575 	surf.mode = G_028040_ARRAY_MODE(track->db_z_info);
576 	surf.format = G_028044_FORMAT(track->db_s_info);
577 	surf.tsplit = G_028044_TILE_SPLIT(track->db_s_info);
578 	surf.nbanks = G_028040_NUM_BANKS(track->db_z_info);
579 	surf.bankw = G_028040_BANK_WIDTH(track->db_z_info);
580 	surf.bankh = G_028040_BANK_HEIGHT(track->db_z_info);
581 	surf.mtilea = G_028040_MACRO_TILE_ASPECT(track->db_z_info);
582 	surf.nsamples = 1;
583 
584 	if (surf.format != 1) {
585 		dev_warn(p->dev, "%s:%d stencil invalid format %d\n",
586 			 __func__, __LINE__, surf.format);
587 		return -EINVAL;
588 	}
589 	/* replace by color format so we can use same code */
590 	surf.format = V_028C70_COLOR_8;
591 
592 	r = evergreen_surface_value_conv_check(p, &surf, "stencil");
593 	if (r) {
594 		return r;
595 	}
596 
597 	r = evergreen_surface_check(p, &surf, NULL);
598 	if (r) {
599 		/* old userspace doesn't compute proper depth/stencil alignment
600 		 * check that alignment against a bigger byte per elements and
601 		 * only report if that alignment is wrong too.
602 		 */
603 		surf.format = V_028C70_COLOR_8_8_8_8;
604 		r = evergreen_surface_check(p, &surf, "stencil");
605 		if (r) {
606 			dev_warn(p->dev, "%s:%d stencil invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
607 				 __func__, __LINE__, track->db_depth_size,
608 				 track->db_depth_slice, track->db_s_info, track->db_z_info);
609 		}
610 		return r;
611 	}
612 
613 	offset = (u64)track->db_s_read_offset << 8;
614 	if (offset & (surf.base_align - 1)) {
615 		dev_warn(p->dev, "%s:%d stencil read bo base %llu not aligned with %ld\n",
616 			 __func__, __LINE__, offset, surf.base_align);
617 		return -EINVAL;
618 	}
619 	offset += (u64)surf.layer_size * mslice;
620 	if (offset > radeon_bo_size(track->db_s_read_bo)) {
621 		dev_warn(p->dev, "%s:%d stencil read bo too small (layer size %d, "
622 			 "offset %llu, max layer %d, bo size %ld)\n",
623 			 __func__, __LINE__, surf.layer_size,
624 			(u64)track->db_s_read_offset << 8, mslice,
625 			radeon_bo_size(track->db_s_read_bo));
626 		dev_warn(p->dev, "%s:%d stencil invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
627 			 __func__, __LINE__, track->db_depth_size,
628 			 track->db_depth_slice, track->db_s_info, track->db_z_info);
629 		return -EINVAL;
630 	}
631 
632 	offset = (u64)track->db_s_write_offset << 8;
633 	if (offset & (surf.base_align - 1)) {
634 		dev_warn(p->dev, "%s:%d stencil write bo base %llu not aligned with %ld\n",
635 			 __func__, __LINE__, offset, surf.base_align);
636 		return -EINVAL;
637 	}
638 	offset += (u64)surf.layer_size * mslice;
639 	if (offset > radeon_bo_size(track->db_s_write_bo)) {
640 		dev_warn(p->dev, "%s:%d stencil write bo too small (layer size %d, "
641 			 "offset %llu, max layer %d, bo size %ld)\n",
642 			 __func__, __LINE__, surf.layer_size,
643 			(u64)track->db_s_write_offset << 8, mslice,
644 			radeon_bo_size(track->db_s_write_bo));
645 		return -EINVAL;
646 	}
647 
648 	/* hyperz */
649 	if (G_028040_TILE_SURFACE_ENABLE(track->db_z_info)) {
650 		r = evergreen_cs_track_validate_htile(p, surf.nbx, surf.nby);
651 		if (r) {
652 			return r;
653 		}
654 	}
655 
656 	return 0;
657 }
658 
659 static int evergreen_cs_track_validate_depth(struct radeon_cs_parser *p)
660 {
661 	struct evergreen_cs_track *track = p->track;
662 	struct eg_surface surf;
663 	unsigned pitch, slice, mslice;
664 	u64 offset;
665 	int r;
666 
667 	mslice = G_028008_SLICE_MAX(track->db_depth_view) + 1;
668 	pitch = G_028058_PITCH_TILE_MAX(track->db_depth_size);
669 	slice = track->db_depth_slice;
670 	surf.nbx = (pitch + 1) * 8;
671 	surf.nby = ((slice + 1) * 64) / surf.nbx;
672 	surf.mode = G_028040_ARRAY_MODE(track->db_z_info);
673 	surf.format = G_028040_FORMAT(track->db_z_info);
674 	surf.tsplit = G_028040_TILE_SPLIT(track->db_z_info);
675 	surf.nbanks = G_028040_NUM_BANKS(track->db_z_info);
676 	surf.bankw = G_028040_BANK_WIDTH(track->db_z_info);
677 	surf.bankh = G_028040_BANK_HEIGHT(track->db_z_info);
678 	surf.mtilea = G_028040_MACRO_TILE_ASPECT(track->db_z_info);
679 	surf.nsamples = 1;
680 
681 	switch (surf.format) {
682 	case V_028040_Z_16:
683 		surf.format = V_028C70_COLOR_16;
684 		break;
685 	case V_028040_Z_24:
686 	case V_028040_Z_32_FLOAT:
687 		surf.format = V_028C70_COLOR_8_8_8_8;
688 		break;
689 	default:
690 		dev_warn(p->dev, "%s:%d depth invalid format %d\n",
691 			 __func__, __LINE__, surf.format);
692 		return -EINVAL;
693 	}
694 
695 	r = evergreen_surface_value_conv_check(p, &surf, "depth");
696 	if (r) {
697 		dev_warn(p->dev, "%s:%d depth invalid (0x%08x 0x%08x 0x%08x)\n",
698 			 __func__, __LINE__, track->db_depth_size,
699 			 track->db_depth_slice, track->db_z_info);
700 		return r;
701 	}
702 
703 	r = evergreen_surface_check(p, &surf, "depth");
704 	if (r) {
705 		dev_warn(p->dev, "%s:%d depth invalid (0x%08x 0x%08x 0x%08x)\n",
706 			 __func__, __LINE__, track->db_depth_size,
707 			 track->db_depth_slice, track->db_z_info);
708 		return r;
709 	}
710 
711 	offset = (u64)track->db_z_read_offset << 8;
712 	if (offset & (surf.base_align - 1)) {
713 		dev_warn(p->dev, "%s:%d stencil read bo base %llu not aligned with %ld\n",
714 			 __func__, __LINE__, offset, surf.base_align);
715 		return -EINVAL;
716 	}
717 	offset += (u64)surf.layer_size * mslice;
718 	if (offset > radeon_bo_size(track->db_z_read_bo)) {
719 		dev_warn(p->dev, "%s:%d depth read bo too small (layer size %d, "
720 			 "offset %llu, max layer %d, bo size %ld)\n",
721 			 __func__, __LINE__, surf.layer_size,
722 			(u64)track->db_z_read_offset << 8, mslice,
723 			radeon_bo_size(track->db_z_read_bo));
724 		return -EINVAL;
725 	}
726 
727 	offset = (u64)track->db_z_write_offset << 8;
728 	if (offset & (surf.base_align - 1)) {
729 		dev_warn(p->dev, "%s:%d stencil write bo base %llu not aligned with %ld\n",
730 			 __func__, __LINE__, offset, surf.base_align);
731 		return -EINVAL;
732 	}
733 	offset += (u64)surf.layer_size * mslice;
734 	if (offset > radeon_bo_size(track->db_z_write_bo)) {
735 		dev_warn(p->dev, "%s:%d depth write bo too small (layer size %d, "
736 			 "offset %llu, max layer %d, bo size %ld)\n",
737 			 __func__, __LINE__, surf.layer_size,
738 			(u64)track->db_z_write_offset << 8, mslice,
739 			radeon_bo_size(track->db_z_write_bo));
740 		return -EINVAL;
741 	}
742 
743 	/* hyperz */
744 	if (G_028040_TILE_SURFACE_ENABLE(track->db_z_info)) {
745 		r = evergreen_cs_track_validate_htile(p, surf.nbx, surf.nby);
746 		if (r) {
747 			return r;
748 		}
749 	}
750 
751 	return 0;
752 }
753 
754 static int evergreen_cs_track_validate_texture(struct radeon_cs_parser *p,
755 					       struct radeon_bo *texture,
756 					       struct radeon_bo *mipmap,
757 					       unsigned idx)
758 {
759 	struct eg_surface surf;
760 	unsigned long toffset, moffset;
761 	unsigned dim, llevel, mslice, width, height, depth, i;
762 	u32 texdw[8];
763 	int r;
764 
765 	texdw[0] = radeon_get_ib_value(p, idx + 0);
766 	texdw[1] = radeon_get_ib_value(p, idx + 1);
767 	texdw[2] = radeon_get_ib_value(p, idx + 2);
768 	texdw[3] = radeon_get_ib_value(p, idx + 3);
769 	texdw[4] = radeon_get_ib_value(p, idx + 4);
770 	texdw[5] = radeon_get_ib_value(p, idx + 5);
771 	texdw[6] = radeon_get_ib_value(p, idx + 6);
772 	texdw[7] = radeon_get_ib_value(p, idx + 7);
773 	dim = G_030000_DIM(texdw[0]);
774 	llevel = G_030014_LAST_LEVEL(texdw[5]);
775 	mslice = G_030014_LAST_ARRAY(texdw[5]) + 1;
776 	width = G_030000_TEX_WIDTH(texdw[0]) + 1;
777 	height =  G_030004_TEX_HEIGHT(texdw[1]) + 1;
778 	depth = G_030004_TEX_DEPTH(texdw[1]) + 1;
779 	surf.format = G_03001C_DATA_FORMAT(texdw[7]);
780 	surf.nbx = (G_030000_PITCH(texdw[0]) + 1) * 8;
781 	surf.nbx = r600_fmt_get_nblocksx(surf.format, surf.nbx);
782 	surf.nby = r600_fmt_get_nblocksy(surf.format, height);
783 	surf.mode = G_030004_ARRAY_MODE(texdw[1]);
784 	surf.tsplit = G_030018_TILE_SPLIT(texdw[6]);
785 	surf.nbanks = G_03001C_NUM_BANKS(texdw[7]);
786 	surf.bankw = G_03001C_BANK_WIDTH(texdw[7]);
787 	surf.bankh = G_03001C_BANK_HEIGHT(texdw[7]);
788 	surf.mtilea = G_03001C_MACRO_TILE_ASPECT(texdw[7]);
789 	surf.nsamples = 1;
790 	toffset = texdw[2] << 8;
791 	moffset = texdw[3] << 8;
792 
793 	if (!r600_fmt_is_valid_texture(surf.format, p->family)) {
794 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
795 			 __func__, __LINE__, surf.format);
796 		return -EINVAL;
797 	}
798 	switch (dim) {
799 	case V_030000_SQ_TEX_DIM_1D:
800 	case V_030000_SQ_TEX_DIM_2D:
801 	case V_030000_SQ_TEX_DIM_CUBEMAP:
802 	case V_030000_SQ_TEX_DIM_1D_ARRAY:
803 	case V_030000_SQ_TEX_DIM_2D_ARRAY:
804 		depth = 1;
805 		break;
806 	case V_030000_SQ_TEX_DIM_2D_MSAA:
807 	case V_030000_SQ_TEX_DIM_2D_ARRAY_MSAA:
808 		surf.nsamples = 1 << llevel;
809 		llevel = 0;
810 		depth = 1;
811 		break;
812 	case V_030000_SQ_TEX_DIM_3D:
813 		break;
814 	default:
815 		dev_warn(p->dev, "%s:%d texture invalid dimension %d\n",
816 			 __func__, __LINE__, dim);
817 		return -EINVAL;
818 	}
819 
820 	r = evergreen_surface_value_conv_check(p, &surf, "texture");
821 	if (r) {
822 		return r;
823 	}
824 
825 	/* align height */
826 	evergreen_surface_check(p, &surf, NULL);
827 	surf.nby = ALIGN(surf.nby, surf.halign);
828 
829 	r = evergreen_surface_check(p, &surf, "texture");
830 	if (r) {
831 		dev_warn(p->dev, "%s:%d texture invalid 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x\n",
832 			 __func__, __LINE__, texdw[0], texdw[1], texdw[4],
833 			 texdw[5], texdw[6], texdw[7]);
834 		return r;
835 	}
836 
837 	/* check texture size */
838 	if (toffset & (surf.base_align - 1)) {
839 		dev_warn(p->dev, "%s:%d texture bo base %ld not aligned with %ld\n",
840 			 __func__, __LINE__, toffset, surf.base_align);
841 		return -EINVAL;
842 	}
843 	if (surf.nsamples <= 1 && moffset & (surf.base_align - 1)) {
844 		dev_warn(p->dev, "%s:%d mipmap bo base %ld not aligned with %ld\n",
845 			 __func__, __LINE__, moffset, surf.base_align);
846 		return -EINVAL;
847 	}
848 	if (dim == SQ_TEX_DIM_3D) {
849 		toffset += surf.layer_size * depth;
850 	} else {
851 		toffset += surf.layer_size * mslice;
852 	}
853 	if (toffset > radeon_bo_size(texture)) {
854 		dev_warn(p->dev, "%s:%d texture bo too small (layer size %d, "
855 			 "offset %ld, max layer %d, depth %d, bo size %ld) (%d %d)\n",
856 			 __func__, __LINE__, surf.layer_size,
857 			(unsigned long)texdw[2] << 8, mslice,
858 			depth, radeon_bo_size(texture),
859 			surf.nbx, surf.nby);
860 		return -EINVAL;
861 	}
862 
863 	if (!mipmap) {
864 		if (llevel) {
865 			dev_warn(p->dev, "%s:%i got NULL MIP_ADDRESS relocation\n",
866 				 __func__, __LINE__);
867 			return -EINVAL;
868 		} else {
869 			return 0; /* everything's ok */
870 		}
871 	}
872 
873 	/* check mipmap size */
874 	for (i = 1; i <= llevel; i++) {
875 		unsigned w, h, d;
876 
877 		w = r600_mip_minify(width, i);
878 		h = r600_mip_minify(height, i);
879 		d = r600_mip_minify(depth, i);
880 		surf.nbx = r600_fmt_get_nblocksx(surf.format, w);
881 		surf.nby = r600_fmt_get_nblocksy(surf.format, h);
882 
883 		switch (surf.mode) {
884 		case ARRAY_2D_TILED_THIN1:
885 			if (surf.nbx < surf.palign || surf.nby < surf.halign) {
886 				surf.mode = ARRAY_1D_TILED_THIN1;
887 			}
888 			/* recompute alignment */
889 			evergreen_surface_check(p, &surf, NULL);
890 			break;
891 		case ARRAY_LINEAR_GENERAL:
892 		case ARRAY_LINEAR_ALIGNED:
893 		case ARRAY_1D_TILED_THIN1:
894 			break;
895 		default:
896 			dev_warn(p->dev, "%s:%d invalid array mode %d\n",
897 				 __func__, __LINE__, surf.mode);
898 			return -EINVAL;
899 		}
900 		surf.nbx = ALIGN(surf.nbx, surf.palign);
901 		surf.nby = ALIGN(surf.nby, surf.halign);
902 
903 		r = evergreen_surface_check(p, &surf, "mipmap");
904 		if (r) {
905 			return r;
906 		}
907 
908 		if (dim == SQ_TEX_DIM_3D) {
909 			moffset += surf.layer_size * d;
910 		} else {
911 			moffset += surf.layer_size * mslice;
912 		}
913 		if (moffset > radeon_bo_size(mipmap)) {
914 			dev_warn(p->dev, "%s:%d mipmap [%d] bo too small (layer size %d, "
915 					"offset %ld, coffset %ld, max layer %d, depth %d, "
916 					"bo size %ld) level0 (%d %d %d)\n",
917 					__func__, __LINE__, i, surf.layer_size,
918 					(unsigned long)texdw[3] << 8, moffset, mslice,
919 					d, radeon_bo_size(mipmap),
920 					width, height, depth);
921 			dev_warn(p->dev, "%s:%d problematic surf: (%d %d) (%d %d %d %d %d %d %d)\n",
922 				 __func__, __LINE__, surf.nbx, surf.nby,
923 				surf.mode, surf.bpe, surf.nsamples,
924 				surf.bankw, surf.bankh,
925 				surf.tsplit, surf.mtilea);
926 			return -EINVAL;
927 		}
928 	}
929 
930 	return 0;
931 }
932 
933 static int evergreen_cs_track_check(struct radeon_cs_parser *p)
934 {
935 	struct evergreen_cs_track *track = p->track;
936 	unsigned tmp, i;
937 	int r;
938 	unsigned buffer_mask = 0;
939 
940 	/* check streamout */
941 	if (track->streamout_dirty && track->vgt_strmout_config) {
942 		for (i = 0; i < 4; i++) {
943 			if (track->vgt_strmout_config & (1 << i)) {
944 				buffer_mask |= (track->vgt_strmout_buffer_config >> (i * 4)) & 0xf;
945 			}
946 		}
947 
948 		for (i = 0; i < 4; i++) {
949 			if (buffer_mask & (1 << i)) {
950 				if (track->vgt_strmout_bo[i]) {
951 					u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
952 							(u64)track->vgt_strmout_size[i];
953 					if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
954 						DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
955 							  i, offset,
956 							  radeon_bo_size(track->vgt_strmout_bo[i]));
957 						return -EINVAL;
958 					}
959 				} else {
960 					dev_warn(p->dev, "No buffer for streamout %d\n", i);
961 					return -EINVAL;
962 				}
963 			}
964 		}
965 		track->streamout_dirty = false;
966 	}
967 
968 	if (track->sx_misc_kill_all_prims)
969 		return 0;
970 
971 	/* check that we have a cb for each enabled target
972 	 */
973 	if (track->cb_dirty) {
974 		tmp = track->cb_target_mask;
975 		for (i = 0; i < 8; i++) {
976 			u32 format = G_028C70_FORMAT(track->cb_color_info[i]);
977 
978 			if (format != V_028C70_COLOR_INVALID &&
979 			    (tmp >> (i * 4)) & 0xF) {
980 				/* at least one component is enabled */
981 				if (track->cb_color_bo[i] == NULL) {
982 					dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
983 						__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
984 					return -EINVAL;
985 				}
986 				/* check cb */
987 				r = evergreen_cs_track_validate_cb(p, i);
988 				if (r) {
989 					return r;
990 				}
991 			}
992 		}
993 		track->cb_dirty = false;
994 	}
995 
996 	if (track->db_dirty) {
997 		/* Check stencil buffer */
998 		if (G_028044_FORMAT(track->db_s_info) != V_028044_STENCIL_INVALID &&
999 		    G_028800_STENCIL_ENABLE(track->db_depth_control)) {
1000 			r = evergreen_cs_track_validate_stencil(p);
1001 			if (r)
1002 				return r;
1003 		}
1004 		/* Check depth buffer */
1005 		if (G_028040_FORMAT(track->db_z_info) != V_028040_Z_INVALID &&
1006 		    G_028800_Z_ENABLE(track->db_depth_control)) {
1007 			r = evergreen_cs_track_validate_depth(p);
1008 			if (r)
1009 				return r;
1010 		}
1011 		track->db_dirty = false;
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 /**
1018  * evergreen_cs_packet_parse_vline() - parse userspace VLINE packet
1019  * @p:		parser structure holding parsing context.
1020  *
1021  * This is an Evergreen(+)-specific function for parsing VLINE packets.
1022  * Real work is done by r600_cs_common_vline_parse function.
1023  * Here we just set up ASIC-specific register table and call
1024  * the common implementation function.
1025  */
1026 static int evergreen_cs_packet_parse_vline(struct radeon_cs_parser *p)
1027 {
1028 
1029 	static uint32_t vline_start_end[6] = {
1030 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC0_REGISTER_OFFSET,
1031 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC1_REGISTER_OFFSET,
1032 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC2_REGISTER_OFFSET,
1033 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC3_REGISTER_OFFSET,
1034 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC4_REGISTER_OFFSET,
1035 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC5_REGISTER_OFFSET
1036 	};
1037 	static uint32_t vline_status[6] = {
1038 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET,
1039 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET,
1040 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET,
1041 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET,
1042 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET,
1043 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET
1044 	};
1045 
1046 	return r600_cs_common_vline_parse(p, vline_start_end, vline_status);
1047 }
1048 
1049 static int evergreen_packet0_check(struct radeon_cs_parser *p,
1050 				   struct radeon_cs_packet *pkt,
1051 				   unsigned idx, unsigned reg)
1052 {
1053 	int r;
1054 
1055 	switch (reg) {
1056 	case EVERGREEN_VLINE_START_END:
1057 		r = evergreen_cs_packet_parse_vline(p);
1058 		if (r) {
1059 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1060 					idx, reg);
1061 			return r;
1062 		}
1063 		break;
1064 	default:
1065 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1066 		return -EINVAL;
1067 	}
1068 	return 0;
1069 }
1070 
1071 static int evergreen_cs_parse_packet0(struct radeon_cs_parser *p,
1072 				      struct radeon_cs_packet *pkt)
1073 {
1074 	unsigned reg, i;
1075 	unsigned idx;
1076 	int r;
1077 
1078 	idx = pkt->idx + 1;
1079 	reg = pkt->reg;
1080 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
1081 		r = evergreen_packet0_check(p, pkt, idx, reg);
1082 		if (r) {
1083 			return r;
1084 		}
1085 	}
1086 	return 0;
1087 }
1088 
1089 /**
1090  * evergreen_cs_handle_reg() - process registers that need special handling.
1091  * @p: parser structure holding parsing context
1092  * @reg: register we are testing
1093  * @idx: index into the cs buffer
1094  */
1095 static int evergreen_cs_handle_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1096 {
1097 	struct evergreen_cs_track *track = (struct evergreen_cs_track *)p->track;
1098 	struct radeon_bo_list *reloc;
1099 	u32 tmp, *ib;
1100 	int r;
1101 
1102 	ib = p->ib.ptr;
1103 	switch (reg) {
1104 	/* force following reg to 0 in an attempt to disable out buffer
1105 	 * which will need us to better understand how it works to perform
1106 	 * security check on it (Jerome)
1107 	 */
1108 	case SQ_ESGS_RING_SIZE:
1109 	case SQ_GSVS_RING_SIZE:
1110 	case SQ_ESTMP_RING_SIZE:
1111 	case SQ_GSTMP_RING_SIZE:
1112 	case SQ_HSTMP_RING_SIZE:
1113 	case SQ_LSTMP_RING_SIZE:
1114 	case SQ_PSTMP_RING_SIZE:
1115 	case SQ_VSTMP_RING_SIZE:
1116 	case SQ_ESGS_RING_ITEMSIZE:
1117 	case SQ_ESTMP_RING_ITEMSIZE:
1118 	case SQ_GSTMP_RING_ITEMSIZE:
1119 	case SQ_GSVS_RING_ITEMSIZE:
1120 	case SQ_GS_VERT_ITEMSIZE:
1121 	case SQ_GS_VERT_ITEMSIZE_1:
1122 	case SQ_GS_VERT_ITEMSIZE_2:
1123 	case SQ_GS_VERT_ITEMSIZE_3:
1124 	case SQ_GSVS_RING_OFFSET_1:
1125 	case SQ_GSVS_RING_OFFSET_2:
1126 	case SQ_GSVS_RING_OFFSET_3:
1127 	case SQ_HSTMP_RING_ITEMSIZE:
1128 	case SQ_LSTMP_RING_ITEMSIZE:
1129 	case SQ_PSTMP_RING_ITEMSIZE:
1130 	case SQ_VSTMP_RING_ITEMSIZE:
1131 	case VGT_TF_RING_SIZE:
1132 		/* get value to populate the IB don't remove */
1133 		/*tmp =radeon_get_ib_value(p, idx);
1134 		  ib[idx] = 0;*/
1135 		break;
1136 	case SQ_ESGS_RING_BASE:
1137 	case SQ_GSVS_RING_BASE:
1138 	case SQ_ESTMP_RING_BASE:
1139 	case SQ_GSTMP_RING_BASE:
1140 	case SQ_HSTMP_RING_BASE:
1141 	case SQ_LSTMP_RING_BASE:
1142 	case SQ_PSTMP_RING_BASE:
1143 	case SQ_VSTMP_RING_BASE:
1144 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1145 		if (r) {
1146 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1147 					"0x%04X\n", reg);
1148 			return -EINVAL;
1149 		}
1150 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1151 		break;
1152 	case DB_DEPTH_CONTROL:
1153 		track->db_depth_control = radeon_get_ib_value(p, idx);
1154 		track->db_dirty = true;
1155 		break;
1156 	case CAYMAN_DB_EQAA:
1157 		if (p->rdev->family < CHIP_CAYMAN) {
1158 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1159 				 "0x%04X\n", reg);
1160 			return -EINVAL;
1161 		}
1162 		break;
1163 	case CAYMAN_DB_DEPTH_INFO:
1164 		if (p->rdev->family < CHIP_CAYMAN) {
1165 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1166 				 "0x%04X\n", reg);
1167 			return -EINVAL;
1168 		}
1169 		break;
1170 	case DB_Z_INFO:
1171 		track->db_z_info = radeon_get_ib_value(p, idx);
1172 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1173 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1174 			if (r) {
1175 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1176 						"0x%04X\n", reg);
1177 				return -EINVAL;
1178 			}
1179 			ib[idx] &= ~Z_ARRAY_MODE(0xf);
1180 			track->db_z_info &= ~Z_ARRAY_MODE(0xf);
1181 			ib[idx] |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1182 			track->db_z_info |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1183 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1184 				unsigned bankw, bankh, mtaspect, tile_split;
1185 
1186 				evergreen_tiling_fields(reloc->tiling_flags,
1187 							&bankw, &bankh, &mtaspect,
1188 							&tile_split);
1189 				ib[idx] |= DB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1190 				ib[idx] |= DB_TILE_SPLIT(tile_split) |
1191 						DB_BANK_WIDTH(bankw) |
1192 						DB_BANK_HEIGHT(bankh) |
1193 						DB_MACRO_TILE_ASPECT(mtaspect);
1194 			}
1195 		}
1196 		track->db_dirty = true;
1197 		break;
1198 	case DB_STENCIL_INFO:
1199 		track->db_s_info = radeon_get_ib_value(p, idx);
1200 		track->db_dirty = true;
1201 		break;
1202 	case DB_DEPTH_VIEW:
1203 		track->db_depth_view = radeon_get_ib_value(p, idx);
1204 		track->db_dirty = true;
1205 		break;
1206 	case DB_DEPTH_SIZE:
1207 		track->db_depth_size = radeon_get_ib_value(p, idx);
1208 		track->db_dirty = true;
1209 		break;
1210 	case R_02805C_DB_DEPTH_SLICE:
1211 		track->db_depth_slice = radeon_get_ib_value(p, idx);
1212 		track->db_dirty = true;
1213 		break;
1214 	case DB_Z_READ_BASE:
1215 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1216 		if (r) {
1217 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1218 					"0x%04X\n", reg);
1219 			return -EINVAL;
1220 		}
1221 		track->db_z_read_offset = radeon_get_ib_value(p, idx);
1222 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1223 		track->db_z_read_bo = reloc->robj;
1224 		track->db_dirty = true;
1225 		break;
1226 	case DB_Z_WRITE_BASE:
1227 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1228 		if (r) {
1229 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1230 					"0x%04X\n", reg);
1231 			return -EINVAL;
1232 		}
1233 		track->db_z_write_offset = radeon_get_ib_value(p, idx);
1234 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1235 		track->db_z_write_bo = reloc->robj;
1236 		track->db_dirty = true;
1237 		break;
1238 	case DB_STENCIL_READ_BASE:
1239 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1240 		if (r) {
1241 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1242 					"0x%04X\n", reg);
1243 			return -EINVAL;
1244 		}
1245 		track->db_s_read_offset = radeon_get_ib_value(p, idx);
1246 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1247 		track->db_s_read_bo = reloc->robj;
1248 		track->db_dirty = true;
1249 		break;
1250 	case DB_STENCIL_WRITE_BASE:
1251 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1252 		if (r) {
1253 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1254 					"0x%04X\n", reg);
1255 			return -EINVAL;
1256 		}
1257 		track->db_s_write_offset = radeon_get_ib_value(p, idx);
1258 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1259 		track->db_s_write_bo = reloc->robj;
1260 		track->db_dirty = true;
1261 		break;
1262 	case VGT_STRMOUT_CONFIG:
1263 		track->vgt_strmout_config = radeon_get_ib_value(p, idx);
1264 		track->streamout_dirty = true;
1265 		break;
1266 	case VGT_STRMOUT_BUFFER_CONFIG:
1267 		track->vgt_strmout_buffer_config = radeon_get_ib_value(p, idx);
1268 		track->streamout_dirty = true;
1269 		break;
1270 	case VGT_STRMOUT_BUFFER_BASE_0:
1271 	case VGT_STRMOUT_BUFFER_BASE_1:
1272 	case VGT_STRMOUT_BUFFER_BASE_2:
1273 	case VGT_STRMOUT_BUFFER_BASE_3:
1274 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1275 		if (r) {
1276 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1277 					"0x%04X\n", reg);
1278 			return -EINVAL;
1279 		}
1280 		tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1281 		track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1282 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1283 		track->vgt_strmout_bo[tmp] = reloc->robj;
1284 		track->streamout_dirty = true;
1285 		break;
1286 	case VGT_STRMOUT_BUFFER_SIZE_0:
1287 	case VGT_STRMOUT_BUFFER_SIZE_1:
1288 	case VGT_STRMOUT_BUFFER_SIZE_2:
1289 	case VGT_STRMOUT_BUFFER_SIZE_3:
1290 		tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1291 		/* size in register is DWs, convert to bytes */
1292 		track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1293 		track->streamout_dirty = true;
1294 		break;
1295 	case CP_COHER_BASE:
1296 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1297 		if (r) {
1298 			dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1299 					"0x%04X\n", reg);
1300 			return -EINVAL;
1301 		}
1302 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1303 		break;
1304 	case CB_TARGET_MASK:
1305 		track->cb_target_mask = radeon_get_ib_value(p, idx);
1306 		track->cb_dirty = true;
1307 		break;
1308 	case CB_SHADER_MASK:
1309 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
1310 		track->cb_dirty = true;
1311 		break;
1312 	case PA_SC_AA_CONFIG:
1313 		if (p->rdev->family >= CHIP_CAYMAN) {
1314 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1315 				 "0x%04X\n", reg);
1316 			return -EINVAL;
1317 		}
1318 		tmp = radeon_get_ib_value(p, idx) & MSAA_NUM_SAMPLES_MASK;
1319 		track->nsamples = 1 << tmp;
1320 		break;
1321 	case CAYMAN_PA_SC_AA_CONFIG:
1322 		if (p->rdev->family < CHIP_CAYMAN) {
1323 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1324 				 "0x%04X\n", reg);
1325 			return -EINVAL;
1326 		}
1327 		tmp = radeon_get_ib_value(p, idx) & CAYMAN_MSAA_NUM_SAMPLES_MASK;
1328 		track->nsamples = 1 << tmp;
1329 		break;
1330 	case CB_COLOR0_VIEW:
1331 	case CB_COLOR1_VIEW:
1332 	case CB_COLOR2_VIEW:
1333 	case CB_COLOR3_VIEW:
1334 	case CB_COLOR4_VIEW:
1335 	case CB_COLOR5_VIEW:
1336 	case CB_COLOR6_VIEW:
1337 	case CB_COLOR7_VIEW:
1338 		tmp = (reg - CB_COLOR0_VIEW) / 0x3c;
1339 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1340 		track->cb_dirty = true;
1341 		break;
1342 	case CB_COLOR8_VIEW:
1343 	case CB_COLOR9_VIEW:
1344 	case CB_COLOR10_VIEW:
1345 	case CB_COLOR11_VIEW:
1346 		tmp = ((reg - CB_COLOR8_VIEW) / 0x1c) + 8;
1347 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1348 		track->cb_dirty = true;
1349 		break;
1350 	case CB_COLOR0_INFO:
1351 	case CB_COLOR1_INFO:
1352 	case CB_COLOR2_INFO:
1353 	case CB_COLOR3_INFO:
1354 	case CB_COLOR4_INFO:
1355 	case CB_COLOR5_INFO:
1356 	case CB_COLOR6_INFO:
1357 	case CB_COLOR7_INFO:
1358 		tmp = (reg - CB_COLOR0_INFO) / 0x3c;
1359 		track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1360 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1361 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1362 			if (r) {
1363 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1364 						"0x%04X\n", reg);
1365 				return -EINVAL;
1366 			}
1367 			ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1368 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1369 		}
1370 		track->cb_dirty = true;
1371 		break;
1372 	case CB_COLOR8_INFO:
1373 	case CB_COLOR9_INFO:
1374 	case CB_COLOR10_INFO:
1375 	case CB_COLOR11_INFO:
1376 		tmp = ((reg - CB_COLOR8_INFO) / 0x1c) + 8;
1377 		track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1378 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1379 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1380 			if (r) {
1381 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1382 						"0x%04X\n", reg);
1383 				return -EINVAL;
1384 			}
1385 			ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1386 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1387 		}
1388 		track->cb_dirty = true;
1389 		break;
1390 	case CB_COLOR0_PITCH:
1391 	case CB_COLOR1_PITCH:
1392 	case CB_COLOR2_PITCH:
1393 	case CB_COLOR3_PITCH:
1394 	case CB_COLOR4_PITCH:
1395 	case CB_COLOR5_PITCH:
1396 	case CB_COLOR6_PITCH:
1397 	case CB_COLOR7_PITCH:
1398 		tmp = (reg - CB_COLOR0_PITCH) / 0x3c;
1399 		track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
1400 		track->cb_dirty = true;
1401 		break;
1402 	case CB_COLOR8_PITCH:
1403 	case CB_COLOR9_PITCH:
1404 	case CB_COLOR10_PITCH:
1405 	case CB_COLOR11_PITCH:
1406 		tmp = ((reg - CB_COLOR8_PITCH) / 0x1c) + 8;
1407 		track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
1408 		track->cb_dirty = true;
1409 		break;
1410 	case CB_COLOR0_SLICE:
1411 	case CB_COLOR1_SLICE:
1412 	case CB_COLOR2_SLICE:
1413 	case CB_COLOR3_SLICE:
1414 	case CB_COLOR4_SLICE:
1415 	case CB_COLOR5_SLICE:
1416 	case CB_COLOR6_SLICE:
1417 	case CB_COLOR7_SLICE:
1418 		tmp = (reg - CB_COLOR0_SLICE) / 0x3c;
1419 		track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
1420 		track->cb_color_slice_idx[tmp] = idx;
1421 		track->cb_dirty = true;
1422 		break;
1423 	case CB_COLOR8_SLICE:
1424 	case CB_COLOR9_SLICE:
1425 	case CB_COLOR10_SLICE:
1426 	case CB_COLOR11_SLICE:
1427 		tmp = ((reg - CB_COLOR8_SLICE) / 0x1c) + 8;
1428 		track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
1429 		track->cb_color_slice_idx[tmp] = idx;
1430 		track->cb_dirty = true;
1431 		break;
1432 	case CB_COLOR0_ATTRIB:
1433 	case CB_COLOR1_ATTRIB:
1434 	case CB_COLOR2_ATTRIB:
1435 	case CB_COLOR3_ATTRIB:
1436 	case CB_COLOR4_ATTRIB:
1437 	case CB_COLOR5_ATTRIB:
1438 	case CB_COLOR6_ATTRIB:
1439 	case CB_COLOR7_ATTRIB:
1440 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1441 		if (r) {
1442 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1443 					"0x%04X\n", reg);
1444 			return -EINVAL;
1445 		}
1446 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1447 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1448 				unsigned bankw, bankh, mtaspect, tile_split;
1449 
1450 				evergreen_tiling_fields(reloc->tiling_flags,
1451 							&bankw, &bankh, &mtaspect,
1452 							&tile_split);
1453 				ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1454 				ib[idx] |= CB_TILE_SPLIT(tile_split) |
1455 					   CB_BANK_WIDTH(bankw) |
1456 					   CB_BANK_HEIGHT(bankh) |
1457 					   CB_MACRO_TILE_ASPECT(mtaspect);
1458 			}
1459 		}
1460 		tmp = ((reg - CB_COLOR0_ATTRIB) / 0x3c);
1461 		track->cb_color_attrib[tmp] = ib[idx];
1462 		track->cb_dirty = true;
1463 		break;
1464 	case CB_COLOR8_ATTRIB:
1465 	case CB_COLOR9_ATTRIB:
1466 	case CB_COLOR10_ATTRIB:
1467 	case CB_COLOR11_ATTRIB:
1468 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1469 		if (r) {
1470 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1471 					"0x%04X\n", reg);
1472 			return -EINVAL;
1473 		}
1474 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1475 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1476 				unsigned bankw, bankh, mtaspect, tile_split;
1477 
1478 				evergreen_tiling_fields(reloc->tiling_flags,
1479 							&bankw, &bankh, &mtaspect,
1480 							&tile_split);
1481 				ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1482 				ib[idx] |= CB_TILE_SPLIT(tile_split) |
1483 					   CB_BANK_WIDTH(bankw) |
1484 					   CB_BANK_HEIGHT(bankh) |
1485 					   CB_MACRO_TILE_ASPECT(mtaspect);
1486 			}
1487 		}
1488 		tmp = ((reg - CB_COLOR8_ATTRIB) / 0x1c) + 8;
1489 		track->cb_color_attrib[tmp] = ib[idx];
1490 		track->cb_dirty = true;
1491 		break;
1492 	case CB_COLOR0_FMASK:
1493 	case CB_COLOR1_FMASK:
1494 	case CB_COLOR2_FMASK:
1495 	case CB_COLOR3_FMASK:
1496 	case CB_COLOR4_FMASK:
1497 	case CB_COLOR5_FMASK:
1498 	case CB_COLOR6_FMASK:
1499 	case CB_COLOR7_FMASK:
1500 		tmp = (reg - CB_COLOR0_FMASK) / 0x3c;
1501 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1502 		if (r) {
1503 			dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1504 			return -EINVAL;
1505 		}
1506 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1507 		track->cb_color_fmask_bo[tmp] = reloc->robj;
1508 		break;
1509 	case CB_COLOR0_CMASK:
1510 	case CB_COLOR1_CMASK:
1511 	case CB_COLOR2_CMASK:
1512 	case CB_COLOR3_CMASK:
1513 	case CB_COLOR4_CMASK:
1514 	case CB_COLOR5_CMASK:
1515 	case CB_COLOR6_CMASK:
1516 	case CB_COLOR7_CMASK:
1517 		tmp = (reg - CB_COLOR0_CMASK) / 0x3c;
1518 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1519 		if (r) {
1520 			dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1521 			return -EINVAL;
1522 		}
1523 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1524 		track->cb_color_cmask_bo[tmp] = reloc->robj;
1525 		break;
1526 	case CB_COLOR0_FMASK_SLICE:
1527 	case CB_COLOR1_FMASK_SLICE:
1528 	case CB_COLOR2_FMASK_SLICE:
1529 	case CB_COLOR3_FMASK_SLICE:
1530 	case CB_COLOR4_FMASK_SLICE:
1531 	case CB_COLOR5_FMASK_SLICE:
1532 	case CB_COLOR6_FMASK_SLICE:
1533 	case CB_COLOR7_FMASK_SLICE:
1534 		tmp = (reg - CB_COLOR0_FMASK_SLICE) / 0x3c;
1535 		track->cb_color_fmask_slice[tmp] = radeon_get_ib_value(p, idx);
1536 		break;
1537 	case CB_COLOR0_CMASK_SLICE:
1538 	case CB_COLOR1_CMASK_SLICE:
1539 	case CB_COLOR2_CMASK_SLICE:
1540 	case CB_COLOR3_CMASK_SLICE:
1541 	case CB_COLOR4_CMASK_SLICE:
1542 	case CB_COLOR5_CMASK_SLICE:
1543 	case CB_COLOR6_CMASK_SLICE:
1544 	case CB_COLOR7_CMASK_SLICE:
1545 		tmp = (reg - CB_COLOR0_CMASK_SLICE) / 0x3c;
1546 		track->cb_color_cmask_slice[tmp] = radeon_get_ib_value(p, idx);
1547 		break;
1548 	case CB_COLOR0_BASE:
1549 	case CB_COLOR1_BASE:
1550 	case CB_COLOR2_BASE:
1551 	case CB_COLOR3_BASE:
1552 	case CB_COLOR4_BASE:
1553 	case CB_COLOR5_BASE:
1554 	case CB_COLOR6_BASE:
1555 	case CB_COLOR7_BASE:
1556 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1557 		if (r) {
1558 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1559 					"0x%04X\n", reg);
1560 			return -EINVAL;
1561 		}
1562 		tmp = (reg - CB_COLOR0_BASE) / 0x3c;
1563 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
1564 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1565 		track->cb_color_bo[tmp] = reloc->robj;
1566 		track->cb_dirty = true;
1567 		break;
1568 	case CB_COLOR8_BASE:
1569 	case CB_COLOR9_BASE:
1570 	case CB_COLOR10_BASE:
1571 	case CB_COLOR11_BASE:
1572 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1573 		if (r) {
1574 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1575 					"0x%04X\n", reg);
1576 			return -EINVAL;
1577 		}
1578 		tmp = ((reg - CB_COLOR8_BASE) / 0x1c) + 8;
1579 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
1580 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1581 		track->cb_color_bo[tmp] = reloc->robj;
1582 		track->cb_dirty = true;
1583 		break;
1584 	case DB_HTILE_DATA_BASE:
1585 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1586 		if (r) {
1587 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1588 					"0x%04X\n", reg);
1589 			return -EINVAL;
1590 		}
1591 		track->htile_offset = radeon_get_ib_value(p, idx);
1592 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1593 		track->htile_bo = reloc->robj;
1594 		track->db_dirty = true;
1595 		break;
1596 	case DB_HTILE_SURFACE:
1597 		/* 8x8 only */
1598 		track->htile_surface = radeon_get_ib_value(p, idx);
1599 		/* force 8x8 htile width and height */
1600 		ib[idx] |= 3;
1601 		track->db_dirty = true;
1602 		break;
1603 	case CB_IMMED0_BASE:
1604 	case CB_IMMED1_BASE:
1605 	case CB_IMMED2_BASE:
1606 	case CB_IMMED3_BASE:
1607 	case CB_IMMED4_BASE:
1608 	case CB_IMMED5_BASE:
1609 	case CB_IMMED6_BASE:
1610 	case CB_IMMED7_BASE:
1611 	case CB_IMMED8_BASE:
1612 	case CB_IMMED9_BASE:
1613 	case CB_IMMED10_BASE:
1614 	case CB_IMMED11_BASE:
1615 	case SQ_PGM_START_FS:
1616 	case SQ_PGM_START_ES:
1617 	case SQ_PGM_START_VS:
1618 	case SQ_PGM_START_GS:
1619 	case SQ_PGM_START_PS:
1620 	case SQ_PGM_START_HS:
1621 	case SQ_PGM_START_LS:
1622 	case SQ_CONST_MEM_BASE:
1623 	case SQ_ALU_CONST_CACHE_GS_0:
1624 	case SQ_ALU_CONST_CACHE_GS_1:
1625 	case SQ_ALU_CONST_CACHE_GS_2:
1626 	case SQ_ALU_CONST_CACHE_GS_3:
1627 	case SQ_ALU_CONST_CACHE_GS_4:
1628 	case SQ_ALU_CONST_CACHE_GS_5:
1629 	case SQ_ALU_CONST_CACHE_GS_6:
1630 	case SQ_ALU_CONST_CACHE_GS_7:
1631 	case SQ_ALU_CONST_CACHE_GS_8:
1632 	case SQ_ALU_CONST_CACHE_GS_9:
1633 	case SQ_ALU_CONST_CACHE_GS_10:
1634 	case SQ_ALU_CONST_CACHE_GS_11:
1635 	case SQ_ALU_CONST_CACHE_GS_12:
1636 	case SQ_ALU_CONST_CACHE_GS_13:
1637 	case SQ_ALU_CONST_CACHE_GS_14:
1638 	case SQ_ALU_CONST_CACHE_GS_15:
1639 	case SQ_ALU_CONST_CACHE_PS_0:
1640 	case SQ_ALU_CONST_CACHE_PS_1:
1641 	case SQ_ALU_CONST_CACHE_PS_2:
1642 	case SQ_ALU_CONST_CACHE_PS_3:
1643 	case SQ_ALU_CONST_CACHE_PS_4:
1644 	case SQ_ALU_CONST_CACHE_PS_5:
1645 	case SQ_ALU_CONST_CACHE_PS_6:
1646 	case SQ_ALU_CONST_CACHE_PS_7:
1647 	case SQ_ALU_CONST_CACHE_PS_8:
1648 	case SQ_ALU_CONST_CACHE_PS_9:
1649 	case SQ_ALU_CONST_CACHE_PS_10:
1650 	case SQ_ALU_CONST_CACHE_PS_11:
1651 	case SQ_ALU_CONST_CACHE_PS_12:
1652 	case SQ_ALU_CONST_CACHE_PS_13:
1653 	case SQ_ALU_CONST_CACHE_PS_14:
1654 	case SQ_ALU_CONST_CACHE_PS_15:
1655 	case SQ_ALU_CONST_CACHE_VS_0:
1656 	case SQ_ALU_CONST_CACHE_VS_1:
1657 	case SQ_ALU_CONST_CACHE_VS_2:
1658 	case SQ_ALU_CONST_CACHE_VS_3:
1659 	case SQ_ALU_CONST_CACHE_VS_4:
1660 	case SQ_ALU_CONST_CACHE_VS_5:
1661 	case SQ_ALU_CONST_CACHE_VS_6:
1662 	case SQ_ALU_CONST_CACHE_VS_7:
1663 	case SQ_ALU_CONST_CACHE_VS_8:
1664 	case SQ_ALU_CONST_CACHE_VS_9:
1665 	case SQ_ALU_CONST_CACHE_VS_10:
1666 	case SQ_ALU_CONST_CACHE_VS_11:
1667 	case SQ_ALU_CONST_CACHE_VS_12:
1668 	case SQ_ALU_CONST_CACHE_VS_13:
1669 	case SQ_ALU_CONST_CACHE_VS_14:
1670 	case SQ_ALU_CONST_CACHE_VS_15:
1671 	case SQ_ALU_CONST_CACHE_HS_0:
1672 	case SQ_ALU_CONST_CACHE_HS_1:
1673 	case SQ_ALU_CONST_CACHE_HS_2:
1674 	case SQ_ALU_CONST_CACHE_HS_3:
1675 	case SQ_ALU_CONST_CACHE_HS_4:
1676 	case SQ_ALU_CONST_CACHE_HS_5:
1677 	case SQ_ALU_CONST_CACHE_HS_6:
1678 	case SQ_ALU_CONST_CACHE_HS_7:
1679 	case SQ_ALU_CONST_CACHE_HS_8:
1680 	case SQ_ALU_CONST_CACHE_HS_9:
1681 	case SQ_ALU_CONST_CACHE_HS_10:
1682 	case SQ_ALU_CONST_CACHE_HS_11:
1683 	case SQ_ALU_CONST_CACHE_HS_12:
1684 	case SQ_ALU_CONST_CACHE_HS_13:
1685 	case SQ_ALU_CONST_CACHE_HS_14:
1686 	case SQ_ALU_CONST_CACHE_HS_15:
1687 	case SQ_ALU_CONST_CACHE_LS_0:
1688 	case SQ_ALU_CONST_CACHE_LS_1:
1689 	case SQ_ALU_CONST_CACHE_LS_2:
1690 	case SQ_ALU_CONST_CACHE_LS_3:
1691 	case SQ_ALU_CONST_CACHE_LS_4:
1692 	case SQ_ALU_CONST_CACHE_LS_5:
1693 	case SQ_ALU_CONST_CACHE_LS_6:
1694 	case SQ_ALU_CONST_CACHE_LS_7:
1695 	case SQ_ALU_CONST_CACHE_LS_8:
1696 	case SQ_ALU_CONST_CACHE_LS_9:
1697 	case SQ_ALU_CONST_CACHE_LS_10:
1698 	case SQ_ALU_CONST_CACHE_LS_11:
1699 	case SQ_ALU_CONST_CACHE_LS_12:
1700 	case SQ_ALU_CONST_CACHE_LS_13:
1701 	case SQ_ALU_CONST_CACHE_LS_14:
1702 	case SQ_ALU_CONST_CACHE_LS_15:
1703 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1704 		if (r) {
1705 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1706 					"0x%04X\n", reg);
1707 			return -EINVAL;
1708 		}
1709 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1710 		break;
1711 	case SX_MEMORY_EXPORT_BASE:
1712 		if (p->rdev->family >= CHIP_CAYMAN) {
1713 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1714 				 "0x%04X\n", reg);
1715 			return -EINVAL;
1716 		}
1717 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1718 		if (r) {
1719 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1720 					"0x%04X\n", reg);
1721 			return -EINVAL;
1722 		}
1723 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1724 		break;
1725 	case CAYMAN_SX_SCATTER_EXPORT_BASE:
1726 		if (p->rdev->family < CHIP_CAYMAN) {
1727 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1728 				 "0x%04X\n", reg);
1729 			return -EINVAL;
1730 		}
1731 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1732 		if (r) {
1733 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1734 					"0x%04X\n", reg);
1735 			return -EINVAL;
1736 		}
1737 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1738 		break;
1739 	case SX_MISC:
1740 		track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1741 		break;
1742 	default:
1743 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1744 		return -EINVAL;
1745 	}
1746 	return 0;
1747 }
1748 
1749 /**
1750  * evergreen_is_safe_reg() - check if register is authorized or not
1751  * @p: parser structure holding parsing context
1752  * @reg: register we are testing
1753  *
1754  * This function will test against reg_safe_bm and return true
1755  * if register is safe or false otherwise.
1756  */
1757 static inline bool evergreen_is_safe_reg(struct radeon_cs_parser *p, u32 reg)
1758 {
1759 	struct evergreen_cs_track *track = p->track;
1760 	u32 m, i;
1761 
1762 	i = (reg >> 7);
1763 	if (unlikely(i >= REG_SAFE_BM_SIZE)) {
1764 		return false;
1765 	}
1766 	m = 1 << ((reg >> 2) & 31);
1767 	if (!(track->reg_safe_bm[i] & m))
1768 		return true;
1769 
1770 	return false;
1771 }
1772 
1773 static int evergreen_packet3_check(struct radeon_cs_parser *p,
1774 				   struct radeon_cs_packet *pkt)
1775 {
1776 	struct radeon_bo_list *reloc;
1777 	struct evergreen_cs_track *track;
1778 	uint32_t *ib;
1779 	unsigned idx;
1780 	unsigned i;
1781 	unsigned start_reg, end_reg, reg;
1782 	int r;
1783 	u32 idx_value;
1784 
1785 	track = (struct evergreen_cs_track *)p->track;
1786 	ib = p->ib.ptr;
1787 	idx = pkt->idx + 1;
1788 	idx_value = radeon_get_ib_value(p, idx);
1789 
1790 	switch (pkt->opcode) {
1791 	case PACKET3_SET_PREDICATION:
1792 	{
1793 		int pred_op;
1794 		int tmp;
1795 		uint64_t offset;
1796 
1797 		if (pkt->count != 1) {
1798 			DRM_ERROR("bad SET PREDICATION\n");
1799 			return -EINVAL;
1800 		}
1801 
1802 		tmp = radeon_get_ib_value(p, idx + 1);
1803 		pred_op = (tmp >> 16) & 0x7;
1804 
1805 		/* for the clear predicate operation */
1806 		if (pred_op == 0)
1807 			return 0;
1808 
1809 		if (pred_op > 2) {
1810 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1811 			return -EINVAL;
1812 		}
1813 
1814 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1815 		if (r) {
1816 			DRM_ERROR("bad SET PREDICATION\n");
1817 			return -EINVAL;
1818 		}
1819 
1820 		offset = reloc->gpu_offset +
1821 			 (idx_value & 0xfffffff0) +
1822 			 ((u64)(tmp & 0xff) << 32);
1823 
1824 		ib[idx + 0] = offset;
1825 		ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1826 	}
1827 	break;
1828 	case PACKET3_CONTEXT_CONTROL:
1829 		if (pkt->count != 1) {
1830 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1831 			return -EINVAL;
1832 		}
1833 		break;
1834 	case PACKET3_INDEX_TYPE:
1835 	case PACKET3_NUM_INSTANCES:
1836 	case PACKET3_CLEAR_STATE:
1837 		if (pkt->count) {
1838 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
1839 			return -EINVAL;
1840 		}
1841 		break;
1842 	case CAYMAN_PACKET3_DEALLOC_STATE:
1843 		if (p->rdev->family < CHIP_CAYMAN) {
1844 			DRM_ERROR("bad PACKET3_DEALLOC_STATE\n");
1845 			return -EINVAL;
1846 		}
1847 		if (pkt->count) {
1848 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
1849 			return -EINVAL;
1850 		}
1851 		break;
1852 	case PACKET3_INDEX_BASE:
1853 	{
1854 		uint64_t offset;
1855 
1856 		if (pkt->count != 1) {
1857 			DRM_ERROR("bad INDEX_BASE\n");
1858 			return -EINVAL;
1859 		}
1860 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1861 		if (r) {
1862 			DRM_ERROR("bad INDEX_BASE\n");
1863 			return -EINVAL;
1864 		}
1865 
1866 		offset = reloc->gpu_offset +
1867 			 idx_value +
1868 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1869 
1870 		ib[idx+0] = offset;
1871 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1872 
1873 		r = evergreen_cs_track_check(p);
1874 		if (r) {
1875 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1876 			return r;
1877 		}
1878 		break;
1879 	}
1880 	case PACKET3_INDEX_BUFFER_SIZE:
1881 	{
1882 		if (pkt->count != 0) {
1883 			DRM_ERROR("bad INDEX_BUFFER_SIZE\n");
1884 			return -EINVAL;
1885 		}
1886 		break;
1887 	}
1888 	case PACKET3_DRAW_INDEX:
1889 	{
1890 		uint64_t offset;
1891 		if (pkt->count != 3) {
1892 			DRM_ERROR("bad DRAW_INDEX\n");
1893 			return -EINVAL;
1894 		}
1895 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1896 		if (r) {
1897 			DRM_ERROR("bad DRAW_INDEX\n");
1898 			return -EINVAL;
1899 		}
1900 
1901 		offset = reloc->gpu_offset +
1902 			 idx_value +
1903 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1904 
1905 		ib[idx+0] = offset;
1906 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1907 
1908 		r = evergreen_cs_track_check(p);
1909 		if (r) {
1910 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1911 			return r;
1912 		}
1913 		break;
1914 	}
1915 	case PACKET3_DRAW_INDEX_2:
1916 	{
1917 		uint64_t offset;
1918 
1919 		if (pkt->count != 4) {
1920 			DRM_ERROR("bad DRAW_INDEX_2\n");
1921 			return -EINVAL;
1922 		}
1923 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1924 		if (r) {
1925 			DRM_ERROR("bad DRAW_INDEX_2\n");
1926 			return -EINVAL;
1927 		}
1928 
1929 		offset = reloc->gpu_offset +
1930 			 radeon_get_ib_value(p, idx+1) +
1931 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1932 
1933 		ib[idx+1] = offset;
1934 		ib[idx+2] = upper_32_bits(offset) & 0xff;
1935 
1936 		r = evergreen_cs_track_check(p);
1937 		if (r) {
1938 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1939 			return r;
1940 		}
1941 		break;
1942 	}
1943 	case PACKET3_DRAW_INDEX_AUTO:
1944 		if (pkt->count != 1) {
1945 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1946 			return -EINVAL;
1947 		}
1948 		r = evergreen_cs_track_check(p);
1949 		if (r) {
1950 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1951 			return r;
1952 		}
1953 		break;
1954 	case PACKET3_DRAW_INDEX_MULTI_AUTO:
1955 		if (pkt->count != 2) {
1956 			DRM_ERROR("bad DRAW_INDEX_MULTI_AUTO\n");
1957 			return -EINVAL;
1958 		}
1959 		r = evergreen_cs_track_check(p);
1960 		if (r) {
1961 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1962 			return r;
1963 		}
1964 		break;
1965 	case PACKET3_DRAW_INDEX_IMMD:
1966 		if (pkt->count < 2) {
1967 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1968 			return -EINVAL;
1969 		}
1970 		r = evergreen_cs_track_check(p);
1971 		if (r) {
1972 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1973 			return r;
1974 		}
1975 		break;
1976 	case PACKET3_DRAW_INDEX_OFFSET:
1977 		if (pkt->count != 2) {
1978 			DRM_ERROR("bad DRAW_INDEX_OFFSET\n");
1979 			return -EINVAL;
1980 		}
1981 		r = evergreen_cs_track_check(p);
1982 		if (r) {
1983 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1984 			return r;
1985 		}
1986 		break;
1987 	case PACKET3_DRAW_INDEX_OFFSET_2:
1988 		if (pkt->count != 3) {
1989 			DRM_ERROR("bad DRAW_INDEX_OFFSET_2\n");
1990 			return -EINVAL;
1991 		}
1992 		r = evergreen_cs_track_check(p);
1993 		if (r) {
1994 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1995 			return r;
1996 		}
1997 		break;
1998 	case PACKET3_SET_BASE:
1999 	{
2000 		/*
2001 		DW 1 HEADER Header of the packet. Shader_Type in bit 1 of the Header will correspond to the shader type of the Load, see Type-3 Packet.
2002 		   2 BASE_INDEX Bits [3:0] BASE_INDEX - Base Index specifies which base address is specified in the last two DWs.
2003 		     0001: DX11 Draw_Index_Indirect Patch Table Base: Base address for Draw_Index_Indirect data.
2004 		   3 ADDRESS_LO Bits [31:3] - Lower bits of QWORD-Aligned Address. Bits [2:0] - Reserved
2005 		   4 ADDRESS_HI Bits [31:8] - Reserved. Bits [7:0] - Upper bits of Address [47:32]
2006 		*/
2007 		if (pkt->count != 2) {
2008 			DRM_ERROR("bad SET_BASE\n");
2009 			return -EINVAL;
2010 		}
2011 
2012 		/* currently only supporting setting indirect draw buffer base address */
2013 		if (idx_value != 1) {
2014 			DRM_ERROR("bad SET_BASE\n");
2015 			return -EINVAL;
2016 		}
2017 
2018 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2019 		if (r) {
2020 			DRM_ERROR("bad SET_BASE\n");
2021 			return -EINVAL;
2022 		}
2023 
2024 		track->indirect_draw_buffer_size = radeon_bo_size(reloc->robj);
2025 
2026 		ib[idx+1] = reloc->gpu_offset;
2027 		ib[idx+2] = upper_32_bits(reloc->gpu_offset) & 0xff;
2028 
2029 		break;
2030 	}
2031 	case PACKET3_DRAW_INDIRECT:
2032 	case PACKET3_DRAW_INDEX_INDIRECT:
2033 	{
2034 		u64 size = pkt->opcode == PACKET3_DRAW_INDIRECT ? 16 : 20;
2035 
2036 		/*
2037 		DW 1 HEADER
2038 		   2 DATA_OFFSET Bits [31:0] + byte aligned offset where the required data structure starts. Bits 1:0 are zero
2039 		   3 DRAW_INITIATOR Draw Initiator Register. Written to the VGT_DRAW_INITIATOR register for the assigned context
2040 		*/
2041 		if (pkt->count != 1) {
2042 			DRM_ERROR("bad DRAW_INDIRECT\n");
2043 			return -EINVAL;
2044 		}
2045 
2046 		if (idx_value + size > track->indirect_draw_buffer_size) {
2047 			dev_warn(p->dev, "DRAW_INDIRECT buffer too small %u + %llu > %lu\n",
2048 				idx_value, size, track->indirect_draw_buffer_size);
2049 			return -EINVAL;
2050 		}
2051 
2052 		r = evergreen_cs_track_check(p);
2053 		if (r) {
2054 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2055 			return r;
2056 		}
2057 		break;
2058 	}
2059 	case PACKET3_DISPATCH_DIRECT:
2060 		if (pkt->count != 3) {
2061 			DRM_ERROR("bad DISPATCH_DIRECT\n");
2062 			return -EINVAL;
2063 		}
2064 		r = evergreen_cs_track_check(p);
2065 		if (r) {
2066 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
2067 			return r;
2068 		}
2069 		break;
2070 	case PACKET3_DISPATCH_INDIRECT:
2071 		if (pkt->count != 1) {
2072 			DRM_ERROR("bad DISPATCH_INDIRECT\n");
2073 			return -EINVAL;
2074 		}
2075 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2076 		if (r) {
2077 			DRM_ERROR("bad DISPATCH_INDIRECT\n");
2078 			return -EINVAL;
2079 		}
2080 		ib[idx+0] = idx_value + (u32)(reloc->gpu_offset & 0xffffffff);
2081 		r = evergreen_cs_track_check(p);
2082 		if (r) {
2083 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2084 			return r;
2085 		}
2086 		break;
2087 	case PACKET3_WAIT_REG_MEM:
2088 		if (pkt->count != 5) {
2089 			DRM_ERROR("bad WAIT_REG_MEM\n");
2090 			return -EINVAL;
2091 		}
2092 		/* bit 4 is reg (0) or mem (1) */
2093 		if (idx_value & 0x10) {
2094 			uint64_t offset;
2095 
2096 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2097 			if (r) {
2098 				DRM_ERROR("bad WAIT_REG_MEM\n");
2099 				return -EINVAL;
2100 			}
2101 
2102 			offset = reloc->gpu_offset +
2103 				 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2104 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2105 
2106 			ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffffc);
2107 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2108 		} else if (idx_value & 0x100) {
2109 			DRM_ERROR("cannot use PFP on REG wait\n");
2110 			return -EINVAL;
2111 		}
2112 		break;
2113 	case PACKET3_CP_DMA:
2114 	{
2115 		u32 command, size, info;
2116 		u64 offset, tmp;
2117 		if (pkt->count != 4) {
2118 			DRM_ERROR("bad CP DMA\n");
2119 			return -EINVAL;
2120 		}
2121 		command = radeon_get_ib_value(p, idx+4);
2122 		size = command & 0x1fffff;
2123 		info = radeon_get_ib_value(p, idx+1);
2124 		if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
2125 		    (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
2126 		    ((((info & 0x00300000) >> 20) == 0) &&
2127 		     (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
2128 		    ((((info & 0x60000000) >> 29) == 0) &&
2129 		     (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
2130 			/* non mem to mem copies requires dw aligned count */
2131 			if (size % 4) {
2132 				DRM_ERROR("CP DMA command requires dw count alignment\n");
2133 				return -EINVAL;
2134 			}
2135 		}
2136 		if (command & PACKET3_CP_DMA_CMD_SAS) {
2137 			/* src address space is register */
2138 			/* GDS is ok */
2139 			if (((info & 0x60000000) >> 29) != 1) {
2140 				DRM_ERROR("CP DMA SAS not supported\n");
2141 				return -EINVAL;
2142 			}
2143 		} else {
2144 			if (command & PACKET3_CP_DMA_CMD_SAIC) {
2145 				DRM_ERROR("CP DMA SAIC only supported for registers\n");
2146 				return -EINVAL;
2147 			}
2148 			/* src address space is memory */
2149 			if (((info & 0x60000000) >> 29) == 0) {
2150 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2151 				if (r) {
2152 					DRM_ERROR("bad CP DMA SRC\n");
2153 					return -EINVAL;
2154 				}
2155 
2156 				tmp = radeon_get_ib_value(p, idx) +
2157 					((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
2158 
2159 				offset = reloc->gpu_offset + tmp;
2160 
2161 				if ((tmp + size) > radeon_bo_size(reloc->robj)) {
2162 					dev_warn(p->dev, "CP DMA src buffer too small (%llu %lu)\n",
2163 						 tmp + size, radeon_bo_size(reloc->robj));
2164 					return -EINVAL;
2165 				}
2166 
2167 				ib[idx] = offset;
2168 				ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2169 			} else if (((info & 0x60000000) >> 29) != 2) {
2170 				DRM_ERROR("bad CP DMA SRC_SEL\n");
2171 				return -EINVAL;
2172 			}
2173 		}
2174 		if (command & PACKET3_CP_DMA_CMD_DAS) {
2175 			/* dst address space is register */
2176 			/* GDS is ok */
2177 			if (((info & 0x00300000) >> 20) != 1) {
2178 				DRM_ERROR("CP DMA DAS not supported\n");
2179 				return -EINVAL;
2180 			}
2181 		} else {
2182 			/* dst address space is memory */
2183 			if (command & PACKET3_CP_DMA_CMD_DAIC) {
2184 				DRM_ERROR("CP DMA DAIC only supported for registers\n");
2185 				return -EINVAL;
2186 			}
2187 			if (((info & 0x00300000) >> 20) == 0) {
2188 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2189 				if (r) {
2190 					DRM_ERROR("bad CP DMA DST\n");
2191 					return -EINVAL;
2192 				}
2193 
2194 				tmp = radeon_get_ib_value(p, idx+2) +
2195 					((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
2196 
2197 				offset = reloc->gpu_offset + tmp;
2198 
2199 				if ((tmp + size) > radeon_bo_size(reloc->robj)) {
2200 					dev_warn(p->dev, "CP DMA dst buffer too small (%llu %lu)\n",
2201 						 tmp + size, radeon_bo_size(reloc->robj));
2202 					return -EINVAL;
2203 				}
2204 
2205 				ib[idx+2] = offset;
2206 				ib[idx+3] = upper_32_bits(offset) & 0xff;
2207 			} else {
2208 				DRM_ERROR("bad CP DMA DST_SEL\n");
2209 				return -EINVAL;
2210 			}
2211 		}
2212 		break;
2213 	}
2214 	case PACKET3_PFP_SYNC_ME:
2215 		if (pkt->count) {
2216 			DRM_ERROR("bad PFP_SYNC_ME\n");
2217 			return -EINVAL;
2218 		}
2219 		break;
2220 	case PACKET3_SURFACE_SYNC:
2221 		if (pkt->count != 3) {
2222 			DRM_ERROR("bad SURFACE_SYNC\n");
2223 			return -EINVAL;
2224 		}
2225 		/* 0xffffffff/0x0 is flush all cache flag */
2226 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
2227 		    radeon_get_ib_value(p, idx + 2) != 0) {
2228 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2229 			if (r) {
2230 				DRM_ERROR("bad SURFACE_SYNC\n");
2231 				return -EINVAL;
2232 			}
2233 			ib[idx+2] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2234 		}
2235 		break;
2236 	case PACKET3_EVENT_WRITE:
2237 		if (pkt->count != 2 && pkt->count != 0) {
2238 			DRM_ERROR("bad EVENT_WRITE\n");
2239 			return -EINVAL;
2240 		}
2241 		if (pkt->count) {
2242 			uint64_t offset;
2243 
2244 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2245 			if (r) {
2246 				DRM_ERROR("bad EVENT_WRITE\n");
2247 				return -EINVAL;
2248 			}
2249 			offset = reloc->gpu_offset +
2250 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
2251 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2252 
2253 			ib[idx+1] = offset & 0xfffffff8;
2254 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2255 		}
2256 		break;
2257 	case PACKET3_EVENT_WRITE_EOP:
2258 	{
2259 		uint64_t offset;
2260 
2261 		if (pkt->count != 4) {
2262 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
2263 			return -EINVAL;
2264 		}
2265 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2266 		if (r) {
2267 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
2268 			return -EINVAL;
2269 		}
2270 
2271 		offset = reloc->gpu_offset +
2272 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2273 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2274 
2275 		ib[idx+1] = offset & 0xfffffffc;
2276 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2277 		break;
2278 	}
2279 	case PACKET3_EVENT_WRITE_EOS:
2280 	{
2281 		uint64_t offset;
2282 
2283 		if (pkt->count != 3) {
2284 			DRM_ERROR("bad EVENT_WRITE_EOS\n");
2285 			return -EINVAL;
2286 		}
2287 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2288 		if (r) {
2289 			DRM_ERROR("bad EVENT_WRITE_EOS\n");
2290 			return -EINVAL;
2291 		}
2292 
2293 		offset = reloc->gpu_offset +
2294 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2295 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2296 
2297 		ib[idx+1] = offset & 0xfffffffc;
2298 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2299 		break;
2300 	}
2301 	case PACKET3_SET_CONFIG_REG:
2302 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2303 		end_reg = 4 * pkt->count + start_reg - 4;
2304 		if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2305 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2306 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2307 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2308 			return -EINVAL;
2309 		}
2310 		for (reg = start_reg, idx++; reg <= end_reg; reg += 4, idx++) {
2311 			if (evergreen_is_safe_reg(p, reg))
2312 				continue;
2313 			r = evergreen_cs_handle_reg(p, reg, idx);
2314 			if (r)
2315 				return r;
2316 		}
2317 		break;
2318 	case PACKET3_SET_CONTEXT_REG:
2319 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_START;
2320 		end_reg = 4 * pkt->count + start_reg - 4;
2321 		if ((start_reg < PACKET3_SET_CONTEXT_REG_START) ||
2322 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
2323 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
2324 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
2325 			return -EINVAL;
2326 		}
2327 		for (reg = start_reg, idx++; reg <= end_reg; reg += 4, idx++) {
2328 			if (evergreen_is_safe_reg(p, reg))
2329 				continue;
2330 			r = evergreen_cs_handle_reg(p, reg, idx);
2331 			if (r)
2332 				return r;
2333 		}
2334 		break;
2335 	case PACKET3_SET_RESOURCE:
2336 		if (pkt->count % 8) {
2337 			DRM_ERROR("bad SET_RESOURCE\n");
2338 			return -EINVAL;
2339 		}
2340 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_START;
2341 		end_reg = 4 * pkt->count + start_reg - 4;
2342 		if ((start_reg < PACKET3_SET_RESOURCE_START) ||
2343 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
2344 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
2345 			DRM_ERROR("bad SET_RESOURCE\n");
2346 			return -EINVAL;
2347 		}
2348 		for (i = 0; i < (pkt->count / 8); i++) {
2349 			struct radeon_bo *texture, *mipmap;
2350 			u32 toffset, moffset;
2351 			u32 size, offset, mip_address, tex_dim;
2352 
2353 			switch (G__SQ_CONSTANT_TYPE(radeon_get_ib_value(p, idx+1+(i*8)+7))) {
2354 			case SQ_TEX_VTX_VALID_TEXTURE:
2355 				/* tex base */
2356 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2357 				if (r) {
2358 					DRM_ERROR("bad SET_RESOURCE (tex)\n");
2359 					return -EINVAL;
2360 				}
2361 				if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
2362 					ib[idx+1+(i*8)+1] |=
2363 						TEX_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
2364 					if (reloc->tiling_flags & RADEON_TILING_MACRO) {
2365 						unsigned bankw, bankh, mtaspect, tile_split;
2366 
2367 						evergreen_tiling_fields(reloc->tiling_flags,
2368 									&bankw, &bankh, &mtaspect,
2369 									&tile_split);
2370 						ib[idx+1+(i*8)+6] |= TEX_TILE_SPLIT(tile_split);
2371 						ib[idx+1+(i*8)+7] |=
2372 							TEX_BANK_WIDTH(bankw) |
2373 							TEX_BANK_HEIGHT(bankh) |
2374 							MACRO_TILE_ASPECT(mtaspect) |
2375 							TEX_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
2376 					}
2377 				}
2378 				texture = reloc->robj;
2379 				toffset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2380 
2381 				/* tex mip base */
2382 				tex_dim = ib[idx+1+(i*8)+0] & 0x7;
2383 				mip_address = ib[idx+1+(i*8)+3];
2384 
2385 				if ((tex_dim == SQ_TEX_DIM_2D_MSAA || tex_dim == SQ_TEX_DIM_2D_ARRAY_MSAA) &&
2386 				    !mip_address &&
2387 				    !radeon_cs_packet_next_is_pkt3_nop(p)) {
2388 					/* MIP_ADDRESS should point to FMASK for an MSAA texture.
2389 					 * It should be 0 if FMASK is disabled. */
2390 					moffset = 0;
2391 					mipmap = NULL;
2392 				} else {
2393 					r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2394 					if (r) {
2395 						DRM_ERROR("bad SET_RESOURCE (tex)\n");
2396 						return -EINVAL;
2397 					}
2398 					moffset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2399 					mipmap = reloc->robj;
2400 				}
2401 
2402 				r = evergreen_cs_track_validate_texture(p, texture, mipmap, idx+1+(i*8));
2403 				if (r)
2404 					return r;
2405 				ib[idx+1+(i*8)+2] += toffset;
2406 				ib[idx+1+(i*8)+3] += moffset;
2407 				break;
2408 			case SQ_TEX_VTX_VALID_BUFFER:
2409 			{
2410 				uint64_t offset64;
2411 				/* vtx base */
2412 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2413 				if (r) {
2414 					DRM_ERROR("bad SET_RESOURCE (vtx)\n");
2415 					return -EINVAL;
2416 				}
2417 				offset = radeon_get_ib_value(p, idx+1+(i*8)+0);
2418 				size = radeon_get_ib_value(p, idx+1+(i*8)+1);
2419 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2420 					/* force size to size of the buffer */
2421 					dev_warn_ratelimited(p->dev, "vbo resource seems too big for the bo\n");
2422 					ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj) - offset;
2423 				}
2424 
2425 				offset64 = reloc->gpu_offset + offset;
2426 				ib[idx+1+(i*8)+0] = offset64;
2427 				ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2428 						    (upper_32_bits(offset64) & 0xff);
2429 				break;
2430 			}
2431 			case SQ_TEX_VTX_INVALID_TEXTURE:
2432 			case SQ_TEX_VTX_INVALID_BUFFER:
2433 			default:
2434 				DRM_ERROR("bad SET_RESOURCE\n");
2435 				return -EINVAL;
2436 			}
2437 		}
2438 		break;
2439 	case PACKET3_SET_ALU_CONST:
2440 		/* XXX fix me ALU const buffers only */
2441 		break;
2442 	case PACKET3_SET_BOOL_CONST:
2443 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_START;
2444 		end_reg = 4 * pkt->count + start_reg - 4;
2445 		if ((start_reg < PACKET3_SET_BOOL_CONST_START) ||
2446 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2447 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2448 			DRM_ERROR("bad SET_BOOL_CONST\n");
2449 			return -EINVAL;
2450 		}
2451 		break;
2452 	case PACKET3_SET_LOOP_CONST:
2453 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_START;
2454 		end_reg = 4 * pkt->count + start_reg - 4;
2455 		if ((start_reg < PACKET3_SET_LOOP_CONST_START) ||
2456 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2457 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2458 			DRM_ERROR("bad SET_LOOP_CONST\n");
2459 			return -EINVAL;
2460 		}
2461 		break;
2462 	case PACKET3_SET_CTL_CONST:
2463 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_START;
2464 		end_reg = 4 * pkt->count + start_reg - 4;
2465 		if ((start_reg < PACKET3_SET_CTL_CONST_START) ||
2466 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2467 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2468 			DRM_ERROR("bad SET_CTL_CONST\n");
2469 			return -EINVAL;
2470 		}
2471 		break;
2472 	case PACKET3_SET_SAMPLER:
2473 		if (pkt->count % 3) {
2474 			DRM_ERROR("bad SET_SAMPLER\n");
2475 			return -EINVAL;
2476 		}
2477 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_START;
2478 		end_reg = 4 * pkt->count + start_reg - 4;
2479 		if ((start_reg < PACKET3_SET_SAMPLER_START) ||
2480 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
2481 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
2482 			DRM_ERROR("bad SET_SAMPLER\n");
2483 			return -EINVAL;
2484 		}
2485 		break;
2486 	case PACKET3_STRMOUT_BUFFER_UPDATE:
2487 		if (pkt->count != 4) {
2488 			DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2489 			return -EINVAL;
2490 		}
2491 		/* Updating memory at DST_ADDRESS. */
2492 		if (idx_value & 0x1) {
2493 			u64 offset;
2494 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2495 			if (r) {
2496 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2497 				return -EINVAL;
2498 			}
2499 			offset = radeon_get_ib_value(p, idx+1);
2500 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2501 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2502 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2503 					  offset + 4, radeon_bo_size(reloc->robj));
2504 				return -EINVAL;
2505 			}
2506 			offset += reloc->gpu_offset;
2507 			ib[idx+1] = offset;
2508 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2509 		}
2510 		/* Reading data from SRC_ADDRESS. */
2511 		if (((idx_value >> 1) & 0x3) == 2) {
2512 			u64 offset;
2513 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2514 			if (r) {
2515 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2516 				return -EINVAL;
2517 			}
2518 			offset = radeon_get_ib_value(p, idx+3);
2519 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2520 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2521 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2522 					  offset + 4, radeon_bo_size(reloc->robj));
2523 				return -EINVAL;
2524 			}
2525 			offset += reloc->gpu_offset;
2526 			ib[idx+3] = offset;
2527 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2528 		}
2529 		break;
2530 	case PACKET3_MEM_WRITE:
2531 	{
2532 		u64 offset;
2533 
2534 		if (pkt->count != 3) {
2535 			DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2536 			return -EINVAL;
2537 		}
2538 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2539 		if (r) {
2540 			DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2541 			return -EINVAL;
2542 		}
2543 		offset = radeon_get_ib_value(p, idx+0);
2544 		offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2545 		if (offset & 0x7) {
2546 			DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2547 			return -EINVAL;
2548 		}
2549 		if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2550 			DRM_ERROR("bad MEM_WRITE bo too small: 0x%llx, 0x%lx\n",
2551 				  offset + 8, radeon_bo_size(reloc->robj));
2552 			return -EINVAL;
2553 		}
2554 		offset += reloc->gpu_offset;
2555 		ib[idx+0] = offset;
2556 		ib[idx+1] = upper_32_bits(offset) & 0xff;
2557 		break;
2558 	}
2559 	case PACKET3_COPY_DW:
2560 		if (pkt->count != 4) {
2561 			DRM_ERROR("bad COPY_DW (invalid count)\n");
2562 			return -EINVAL;
2563 		}
2564 		if (idx_value & 0x1) {
2565 			u64 offset;
2566 			/* SRC is memory. */
2567 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2568 			if (r) {
2569 				DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2570 				return -EINVAL;
2571 			}
2572 			offset = radeon_get_ib_value(p, idx+1);
2573 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2574 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2575 				DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2576 					  offset + 4, radeon_bo_size(reloc->robj));
2577 				return -EINVAL;
2578 			}
2579 			offset += reloc->gpu_offset;
2580 			ib[idx+1] = offset;
2581 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2582 		} else {
2583 			/* SRC is a reg. */
2584 			reg = radeon_get_ib_value(p, idx+1) << 2;
2585 			if (!evergreen_is_safe_reg(p, reg)) {
2586 				dev_warn(p->dev, "forbidden register 0x%08x at %d\n",
2587 					 reg, idx + 1);
2588 				return -EINVAL;
2589 			}
2590 		}
2591 		if (idx_value & 0x2) {
2592 			u64 offset;
2593 			/* DST is memory. */
2594 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2595 			if (r) {
2596 				DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2597 				return -EINVAL;
2598 			}
2599 			offset = radeon_get_ib_value(p, idx+3);
2600 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2601 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2602 				DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2603 					  offset + 4, radeon_bo_size(reloc->robj));
2604 				return -EINVAL;
2605 			}
2606 			offset += reloc->gpu_offset;
2607 			ib[idx+3] = offset;
2608 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2609 		} else {
2610 			/* DST is a reg. */
2611 			reg = radeon_get_ib_value(p, idx+3) << 2;
2612 			if (!evergreen_is_safe_reg(p, reg)) {
2613 				dev_warn(p->dev, "forbidden register 0x%08x at %d\n",
2614 					 reg, idx + 3);
2615 				return -EINVAL;
2616 			}
2617 		}
2618 		break;
2619 	case PACKET3_SET_APPEND_CNT:
2620 	{
2621 		uint32_t areg;
2622 		uint32_t allowed_reg_base;
2623 		uint32_t source_sel;
2624 		if (pkt->count != 2) {
2625 			DRM_ERROR("bad SET_APPEND_CNT (invalid count)\n");
2626 			return -EINVAL;
2627 		}
2628 
2629 		allowed_reg_base = GDS_APPEND_COUNT_0;
2630 		allowed_reg_base -= PACKET3_SET_CONTEXT_REG_START;
2631 		allowed_reg_base >>= 2;
2632 
2633 		areg = idx_value >> 16;
2634 		if (areg < allowed_reg_base || areg > (allowed_reg_base + 11)) {
2635 			dev_warn(p->dev, "forbidden register for append cnt 0x%08x at %d\n",
2636 				 areg, idx);
2637 			return -EINVAL;
2638 		}
2639 
2640 		source_sel = G_PACKET3_SET_APPEND_CNT_SRC_SELECT(idx_value);
2641 		if (source_sel == PACKET3_SAC_SRC_SEL_MEM) {
2642 			uint64_t offset;
2643 			uint32_t swap;
2644 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2645 			if (r) {
2646 				DRM_ERROR("bad SET_APPEND_CNT (missing reloc)\n");
2647 				return -EINVAL;
2648 			}
2649 			offset = radeon_get_ib_value(p, idx + 1);
2650 			swap = offset & 0x3;
2651 			offset &= ~0x3;
2652 
2653 			offset += ((u64)(radeon_get_ib_value(p, idx + 2) & 0xff)) << 32;
2654 
2655 			offset += reloc->gpu_offset;
2656 			ib[idx+1] = (offset & 0xfffffffc) | swap;
2657 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2658 		} else {
2659 			DRM_ERROR("bad SET_APPEND_CNT (unsupported operation)\n");
2660 			return -EINVAL;
2661 		}
2662 		break;
2663 	}
2664 	case PACKET3_NOP:
2665 		break;
2666 	default:
2667 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2668 		return -EINVAL;
2669 	}
2670 	return 0;
2671 }
2672 
2673 int evergreen_cs_parse(struct radeon_cs_parser *p)
2674 {
2675 	struct radeon_cs_packet pkt;
2676 	struct evergreen_cs_track *track;
2677 	u32 tmp;
2678 	int r;
2679 
2680 	if (p->track == NULL) {
2681 		/* initialize tracker, we are in kms */
2682 		track = kzalloc(sizeof(*track), GFP_KERNEL);
2683 		if (track == NULL)
2684 			return -ENOMEM;
2685 		evergreen_cs_track_init(track);
2686 		if (p->rdev->family >= CHIP_CAYMAN) {
2687 			tmp = p->rdev->config.cayman.tile_config;
2688 			track->reg_safe_bm = cayman_reg_safe_bm;
2689 		} else {
2690 			tmp = p->rdev->config.evergreen.tile_config;
2691 			track->reg_safe_bm = evergreen_reg_safe_bm;
2692 		}
2693 		BUILD_BUG_ON(ARRAY_SIZE(cayman_reg_safe_bm) != REG_SAFE_BM_SIZE);
2694 		BUILD_BUG_ON(ARRAY_SIZE(evergreen_reg_safe_bm) != REG_SAFE_BM_SIZE);
2695 		switch (tmp & 0xf) {
2696 		case 0:
2697 			track->npipes = 1;
2698 			break;
2699 		case 1:
2700 		default:
2701 			track->npipes = 2;
2702 			break;
2703 		case 2:
2704 			track->npipes = 4;
2705 			break;
2706 		case 3:
2707 			track->npipes = 8;
2708 			break;
2709 		}
2710 
2711 		switch ((tmp & 0xf0) >> 4) {
2712 		case 0:
2713 			track->nbanks = 4;
2714 			break;
2715 		case 1:
2716 		default:
2717 			track->nbanks = 8;
2718 			break;
2719 		case 2:
2720 			track->nbanks = 16;
2721 			break;
2722 		}
2723 
2724 		switch ((tmp & 0xf00) >> 8) {
2725 		case 0:
2726 			track->group_size = 256;
2727 			break;
2728 		case 1:
2729 		default:
2730 			track->group_size = 512;
2731 			break;
2732 		}
2733 
2734 		switch ((tmp & 0xf000) >> 12) {
2735 		case 0:
2736 			track->row_size = 1;
2737 			break;
2738 		case 1:
2739 		default:
2740 			track->row_size = 2;
2741 			break;
2742 		case 2:
2743 			track->row_size = 4;
2744 			break;
2745 		}
2746 
2747 		p->track = track;
2748 	}
2749 	do {
2750 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2751 		if (r) {
2752 			kfree(p->track);
2753 			p->track = NULL;
2754 			return r;
2755 		}
2756 		p->idx += pkt.count + 2;
2757 		switch (pkt.type) {
2758 		case RADEON_PACKET_TYPE0:
2759 			r = evergreen_cs_parse_packet0(p, &pkt);
2760 			break;
2761 		case RADEON_PACKET_TYPE2:
2762 			break;
2763 		case RADEON_PACKET_TYPE3:
2764 			r = evergreen_packet3_check(p, &pkt);
2765 			break;
2766 		default:
2767 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2768 			kfree(p->track);
2769 			p->track = NULL;
2770 			return -EINVAL;
2771 		}
2772 		if (r) {
2773 			kfree(p->track);
2774 			p->track = NULL;
2775 			return r;
2776 		}
2777 	} while (p->idx < p->chunk_ib->length_dw);
2778 #if 0
2779 	for (r = 0; r < p->ib.length_dw; r++) {
2780 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2781 		mdelay(1);
2782 	}
2783 #endif
2784 	kfree(p->track);
2785 	p->track = NULL;
2786 	return 0;
2787 }
2788 
2789 /**
2790  * evergreen_dma_cs_parse() - parse the DMA IB
2791  * @p:		parser structure holding parsing context.
2792  *
2793  * Parses the DMA IB from the CS ioctl and updates
2794  * the GPU addresses based on the reloc information and
2795  * checks for errors. (Evergreen-Cayman)
2796  * Returns 0 for success and an error on failure.
2797  **/
2798 int evergreen_dma_cs_parse(struct radeon_cs_parser *p)
2799 {
2800 	struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
2801 	struct radeon_bo_list *src_reloc, *dst_reloc, *dst2_reloc;
2802 	u32 header, cmd, count, sub_cmd;
2803 	uint32_t *ib = p->ib.ptr;
2804 	u32 idx;
2805 	u64 src_offset, dst_offset, dst2_offset;
2806 	int r;
2807 
2808 	do {
2809 		if (p->idx >= ib_chunk->length_dw) {
2810 			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2811 				  p->idx, ib_chunk->length_dw);
2812 			return -EINVAL;
2813 		}
2814 		idx = p->idx;
2815 		header = radeon_get_ib_value(p, idx);
2816 		cmd = GET_DMA_CMD(header);
2817 		count = GET_DMA_COUNT(header);
2818 		sub_cmd = GET_DMA_SUB_CMD(header);
2819 
2820 		switch (cmd) {
2821 		case DMA_PACKET_WRITE:
2822 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2823 			if (r) {
2824 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2825 				return -EINVAL;
2826 			}
2827 			switch (sub_cmd) {
2828 			/* tiled */
2829 			case 8:
2830 				dst_offset = radeon_get_ib_value(p, idx+1);
2831 				dst_offset <<= 8;
2832 
2833 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2834 				p->idx += count + 7;
2835 				break;
2836 			/* linear */
2837 			case 0:
2838 				dst_offset = radeon_get_ib_value(p, idx+1);
2839 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2840 
2841 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2842 				ib[idx+2] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2843 				p->idx += count + 3;
2844 				break;
2845 			default:
2846 				DRM_ERROR("bad DMA_PACKET_WRITE [%6d] 0x%08x sub cmd is not 0 or 8\n", idx, header);
2847 				return -EINVAL;
2848 			}
2849 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2850 				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
2851 					 dst_offset, radeon_bo_size(dst_reloc->robj));
2852 				return -EINVAL;
2853 			}
2854 			break;
2855 		case DMA_PACKET_COPY:
2856 			r = r600_dma_cs_next_reloc(p, &src_reloc);
2857 			if (r) {
2858 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2859 				return -EINVAL;
2860 			}
2861 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2862 			if (r) {
2863 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2864 				return -EINVAL;
2865 			}
2866 			switch (sub_cmd) {
2867 			/* Copy L2L, DW aligned */
2868 			case 0x00:
2869 				/* L2L, dw */
2870 				src_offset = radeon_get_ib_value(p, idx+2);
2871 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2872 				dst_offset = radeon_get_ib_value(p, idx+1);
2873 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2874 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2875 					dev_warn(p->dev, "DMA L2L, dw src buffer too small (%llu %lu)\n",
2876 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2877 					return -EINVAL;
2878 				}
2879 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2880 					dev_warn(p->dev, "DMA L2L, dw dst buffer too small (%llu %lu)\n",
2881 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2882 					return -EINVAL;
2883 				}
2884 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2885 				ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2886 				ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2887 				ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2888 				p->idx += 5;
2889 				break;
2890 			/* Copy L2T/T2L */
2891 			case 0x08:
2892 				/* detile bit */
2893 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
2894 					/* tiled src, linear dst */
2895 					src_offset = radeon_get_ib_value(p, idx+1);
2896 					src_offset <<= 8;
2897 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
2898 
2899 					dst_offset = radeon_get_ib_value(p, idx + 7);
2900 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
2901 					ib[idx+7] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2902 					ib[idx+8] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2903 				} else {
2904 					/* linear src, tiled dst */
2905 					src_offset = radeon_get_ib_value(p, idx+7);
2906 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
2907 					ib[idx+7] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2908 					ib[idx+8] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2909 
2910 					dst_offset = radeon_get_ib_value(p, idx+1);
2911 					dst_offset <<= 8;
2912 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2913 				}
2914 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2915 					dev_warn(p->dev, "DMA L2T, src buffer too small (%llu %lu)\n",
2916 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2917 					return -EINVAL;
2918 				}
2919 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2920 					dev_warn(p->dev, "DMA L2T, dst buffer too small (%llu %lu)\n",
2921 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2922 					return -EINVAL;
2923 				}
2924 				p->idx += 9;
2925 				break;
2926 			/* Copy L2L, byte aligned */
2927 			case 0x40:
2928 				/* L2L, byte */
2929 				src_offset = radeon_get_ib_value(p, idx+2);
2930 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2931 				dst_offset = radeon_get_ib_value(p, idx+1);
2932 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2933 				if ((src_offset + count) > radeon_bo_size(src_reloc->robj)) {
2934 					dev_warn(p->dev, "DMA L2L, byte src buffer too small (%llu %lu)\n",
2935 							src_offset + count, radeon_bo_size(src_reloc->robj));
2936 					return -EINVAL;
2937 				}
2938 				if ((dst_offset + count) > radeon_bo_size(dst_reloc->robj)) {
2939 					dev_warn(p->dev, "DMA L2L, byte dst buffer too small (%llu %lu)\n",
2940 							dst_offset + count, radeon_bo_size(dst_reloc->robj));
2941 					return -EINVAL;
2942 				}
2943 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xffffffff);
2944 				ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xffffffff);
2945 				ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2946 				ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2947 				p->idx += 5;
2948 				break;
2949 			/* Copy L2L, partial */
2950 			case 0x41:
2951 				/* L2L, partial */
2952 				if (p->family < CHIP_CAYMAN) {
2953 					DRM_ERROR("L2L Partial is cayman only !\n");
2954 					return -EINVAL;
2955 				}
2956 				ib[idx+1] += (u32)(src_reloc->gpu_offset & 0xffffffff);
2957 				ib[idx+2] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2958 				ib[idx+4] += (u32)(dst_reloc->gpu_offset & 0xffffffff);
2959 				ib[idx+5] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2960 
2961 				p->idx += 9;
2962 				break;
2963 			/* Copy L2L, DW aligned, broadcast */
2964 			case 0x44:
2965 				/* L2L, dw, broadcast */
2966 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
2967 				if (r) {
2968 					DRM_ERROR("bad L2L, dw, broadcast DMA_PACKET_COPY\n");
2969 					return -EINVAL;
2970 				}
2971 				dst_offset = radeon_get_ib_value(p, idx+1);
2972 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2973 				dst2_offset = radeon_get_ib_value(p, idx+2);
2974 				dst2_offset |= ((u64)(radeon_get_ib_value(p, idx+5) & 0xff)) << 32;
2975 				src_offset = radeon_get_ib_value(p, idx+3);
2976 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2977 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2978 					dev_warn(p->dev, "DMA L2L, dw, broadcast src buffer too small (%llu %lu)\n",
2979 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2980 					return -EINVAL;
2981 				}
2982 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2983 					dev_warn(p->dev, "DMA L2L, dw, broadcast dst buffer too small (%llu %lu)\n",
2984 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2985 					return -EINVAL;
2986 				}
2987 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
2988 					dev_warn(p->dev, "DMA L2L, dw, broadcast dst2 buffer too small (%llu %lu)\n",
2989 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
2990 					return -EINVAL;
2991 				}
2992 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2993 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset & 0xfffffffc);
2994 				ib[idx+3] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2995 				ib[idx+4] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2996 				ib[idx+5] += upper_32_bits(dst2_reloc->gpu_offset) & 0xff;
2997 				ib[idx+6] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2998 				p->idx += 7;
2999 				break;
3000 			/* Copy L2T Frame to Field */
3001 			case 0x48:
3002 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3003 					DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
3004 					return -EINVAL;
3005 				}
3006 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3007 				if (r) {
3008 					DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
3009 					return -EINVAL;
3010 				}
3011 				dst_offset = radeon_get_ib_value(p, idx+1);
3012 				dst_offset <<= 8;
3013 				dst2_offset = radeon_get_ib_value(p, idx+2);
3014 				dst2_offset <<= 8;
3015 				src_offset = radeon_get_ib_value(p, idx+8);
3016 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3017 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3018 					dev_warn(p->dev, "DMA L2T, frame to fields src buffer too small (%llu %lu)\n",
3019 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3020 					return -EINVAL;
3021 				}
3022 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3023 					dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%llu %lu)\n",
3024 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3025 					return -EINVAL;
3026 				}
3027 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3028 					dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%llu %lu)\n",
3029 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3030 					return -EINVAL;
3031 				}
3032 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3033 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset >> 8);
3034 				ib[idx+8] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3035 				ib[idx+9] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3036 				p->idx += 10;
3037 				break;
3038 			/* Copy L2T/T2L, partial */
3039 			case 0x49:
3040 				/* L2T, T2L partial */
3041 				if (p->family < CHIP_CAYMAN) {
3042 					DRM_ERROR("L2T, T2L Partial is cayman only !\n");
3043 					return -EINVAL;
3044 				}
3045 				/* detile bit */
3046 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3047 					/* tiled src, linear dst */
3048 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
3049 
3050 					ib[idx+7] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
3051 					ib[idx+8] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
3052 				} else {
3053 					/* linear src, tiled dst */
3054 					ib[idx+7] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3055 					ib[idx+8] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3056 
3057 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3058 				}
3059 				p->idx += 12;
3060 				break;
3061 			/* Copy L2T broadcast */
3062 			case 0x4b:
3063 				/* L2T, broadcast */
3064 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3065 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3066 					return -EINVAL;
3067 				}
3068 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3069 				if (r) {
3070 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3071 					return -EINVAL;
3072 				}
3073 				dst_offset = radeon_get_ib_value(p, idx+1);
3074 				dst_offset <<= 8;
3075 				dst2_offset = radeon_get_ib_value(p, idx+2);
3076 				dst2_offset <<= 8;
3077 				src_offset = radeon_get_ib_value(p, idx+8);
3078 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3079 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3080 					dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
3081 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3082 					return -EINVAL;
3083 				}
3084 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3085 					dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
3086 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3087 					return -EINVAL;
3088 				}
3089 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3090 					dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%llu %lu)\n",
3091 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3092 					return -EINVAL;
3093 				}
3094 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3095 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset >> 8);
3096 				ib[idx+8] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3097 				ib[idx+9] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3098 				p->idx += 10;
3099 				break;
3100 			/* Copy L2T/T2L (tile units) */
3101 			case 0x4c:
3102 				/* L2T, T2L */
3103 				/* detile bit */
3104 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3105 					/* tiled src, linear dst */
3106 					src_offset = radeon_get_ib_value(p, idx+1);
3107 					src_offset <<= 8;
3108 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
3109 
3110 					dst_offset = radeon_get_ib_value(p, idx+7);
3111 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3112 					ib[idx+7] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
3113 					ib[idx+8] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
3114 				} else {
3115 					/* linear src, tiled dst */
3116 					src_offset = radeon_get_ib_value(p, idx+7);
3117 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3118 					ib[idx+7] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3119 					ib[idx+8] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3120 
3121 					dst_offset = radeon_get_ib_value(p, idx+1);
3122 					dst_offset <<= 8;
3123 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3124 				}
3125 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3126 					dev_warn(p->dev, "DMA L2T, T2L src buffer too small (%llu %lu)\n",
3127 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3128 					return -EINVAL;
3129 				}
3130 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3131 					dev_warn(p->dev, "DMA L2T, T2L dst buffer too small (%llu %lu)\n",
3132 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3133 					return -EINVAL;
3134 				}
3135 				p->idx += 9;
3136 				break;
3137 			/* Copy T2T, partial (tile units) */
3138 			case 0x4d:
3139 				/* T2T partial */
3140 				if (p->family < CHIP_CAYMAN) {
3141 					DRM_ERROR("L2T, T2L Partial is cayman only !\n");
3142 					return -EINVAL;
3143 				}
3144 				ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
3145 				ib[idx+4] += (u32)(dst_reloc->gpu_offset >> 8);
3146 				p->idx += 13;
3147 				break;
3148 			/* Copy L2T broadcast (tile units) */
3149 			case 0x4f:
3150 				/* L2T, broadcast */
3151 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3152 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3153 					return -EINVAL;
3154 				}
3155 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3156 				if (r) {
3157 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3158 					return -EINVAL;
3159 				}
3160 				dst_offset = radeon_get_ib_value(p, idx+1);
3161 				dst_offset <<= 8;
3162 				dst2_offset = radeon_get_ib_value(p, idx+2);
3163 				dst2_offset <<= 8;
3164 				src_offset = radeon_get_ib_value(p, idx+8);
3165 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3166 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3167 					dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
3168 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3169 					return -EINVAL;
3170 				}
3171 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3172 					dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
3173 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3174 					return -EINVAL;
3175 				}
3176 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3177 					dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%llu %lu)\n",
3178 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3179 					return -EINVAL;
3180 				}
3181 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3182 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset >> 8);
3183 				ib[idx+8] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3184 				ib[idx+9] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3185 				p->idx += 10;
3186 				break;
3187 			default:
3188 				DRM_ERROR("bad DMA_PACKET_COPY [%6d] 0x%08x invalid sub cmd\n", idx, header);
3189 				return -EINVAL;
3190 			}
3191 			break;
3192 		case DMA_PACKET_CONSTANT_FILL:
3193 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
3194 			if (r) {
3195 				DRM_ERROR("bad DMA_PACKET_CONSTANT_FILL\n");
3196 				return -EINVAL;
3197 			}
3198 			dst_offset = radeon_get_ib_value(p, idx+1);
3199 			dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
3200 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3201 				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
3202 					 dst_offset, radeon_bo_size(dst_reloc->robj));
3203 				return -EINVAL;
3204 			}
3205 			ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
3206 			ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) << 16) & 0x00ff0000;
3207 			p->idx += 4;
3208 			break;
3209 		case DMA_PACKET_NOP:
3210 			p->idx += 1;
3211 			break;
3212 		default:
3213 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
3214 			return -EINVAL;
3215 		}
3216 	} while (p->idx < p->chunk_ib->length_dw);
3217 #if 0
3218 	for (r = 0; r < p->ib->length_dw; r++) {
3219 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
3220 		mdelay(1);
3221 	}
3222 #endif
3223 	return 0;
3224 }
3225 
3226 /* vm parser */
3227 static bool evergreen_vm_reg_valid(u32 reg)
3228 {
3229 	/* context regs are fine */
3230 	if (reg >= 0x28000)
3231 		return true;
3232 
3233 	/* check config regs */
3234 	switch (reg) {
3235 	case WAIT_UNTIL:
3236 	case GRBM_GFX_INDEX:
3237 	case CP_STRMOUT_CNTL:
3238 	case CP_COHER_CNTL:
3239 	case CP_COHER_SIZE:
3240 	case VGT_VTX_VECT_EJECT_REG:
3241 	case VGT_CACHE_INVALIDATION:
3242 	case VGT_GS_VERTEX_REUSE:
3243 	case VGT_PRIMITIVE_TYPE:
3244 	case VGT_INDEX_TYPE:
3245 	case VGT_NUM_INDICES:
3246 	case VGT_NUM_INSTANCES:
3247 	case VGT_COMPUTE_DIM_X:
3248 	case VGT_COMPUTE_DIM_Y:
3249 	case VGT_COMPUTE_DIM_Z:
3250 	case VGT_COMPUTE_START_X:
3251 	case VGT_COMPUTE_START_Y:
3252 	case VGT_COMPUTE_START_Z:
3253 	case VGT_COMPUTE_INDEX:
3254 	case VGT_COMPUTE_THREAD_GROUP_SIZE:
3255 	case VGT_HS_OFFCHIP_PARAM:
3256 	case PA_CL_ENHANCE:
3257 	case PA_SU_LINE_STIPPLE_VALUE:
3258 	case PA_SC_LINE_STIPPLE_STATE:
3259 	case PA_SC_ENHANCE:
3260 	case SQ_DYN_GPR_CNTL_PS_FLUSH_REQ:
3261 	case SQ_DYN_GPR_SIMD_LOCK_EN:
3262 	case SQ_CONFIG:
3263 	case SQ_GPR_RESOURCE_MGMT_1:
3264 	case SQ_GLOBAL_GPR_RESOURCE_MGMT_1:
3265 	case SQ_GLOBAL_GPR_RESOURCE_MGMT_2:
3266 	case SQ_CONST_MEM_BASE:
3267 	case SQ_STATIC_THREAD_MGMT_1:
3268 	case SQ_STATIC_THREAD_MGMT_2:
3269 	case SQ_STATIC_THREAD_MGMT_3:
3270 	case SPI_CONFIG_CNTL:
3271 	case SPI_CONFIG_CNTL_1:
3272 	case TA_CNTL_AUX:
3273 	case DB_DEBUG:
3274 	case DB_DEBUG2:
3275 	case DB_DEBUG3:
3276 	case DB_DEBUG4:
3277 	case DB_WATERMARKS:
3278 	case TD_PS_BORDER_COLOR_INDEX:
3279 	case TD_PS_BORDER_COLOR_RED:
3280 	case TD_PS_BORDER_COLOR_GREEN:
3281 	case TD_PS_BORDER_COLOR_BLUE:
3282 	case TD_PS_BORDER_COLOR_ALPHA:
3283 	case TD_VS_BORDER_COLOR_INDEX:
3284 	case TD_VS_BORDER_COLOR_RED:
3285 	case TD_VS_BORDER_COLOR_GREEN:
3286 	case TD_VS_BORDER_COLOR_BLUE:
3287 	case TD_VS_BORDER_COLOR_ALPHA:
3288 	case TD_GS_BORDER_COLOR_INDEX:
3289 	case TD_GS_BORDER_COLOR_RED:
3290 	case TD_GS_BORDER_COLOR_GREEN:
3291 	case TD_GS_BORDER_COLOR_BLUE:
3292 	case TD_GS_BORDER_COLOR_ALPHA:
3293 	case TD_HS_BORDER_COLOR_INDEX:
3294 	case TD_HS_BORDER_COLOR_RED:
3295 	case TD_HS_BORDER_COLOR_GREEN:
3296 	case TD_HS_BORDER_COLOR_BLUE:
3297 	case TD_HS_BORDER_COLOR_ALPHA:
3298 	case TD_LS_BORDER_COLOR_INDEX:
3299 	case TD_LS_BORDER_COLOR_RED:
3300 	case TD_LS_BORDER_COLOR_GREEN:
3301 	case TD_LS_BORDER_COLOR_BLUE:
3302 	case TD_LS_BORDER_COLOR_ALPHA:
3303 	case TD_CS_BORDER_COLOR_INDEX:
3304 	case TD_CS_BORDER_COLOR_RED:
3305 	case TD_CS_BORDER_COLOR_GREEN:
3306 	case TD_CS_BORDER_COLOR_BLUE:
3307 	case TD_CS_BORDER_COLOR_ALPHA:
3308 	case SQ_ESGS_RING_SIZE:
3309 	case SQ_GSVS_RING_SIZE:
3310 	case SQ_ESTMP_RING_SIZE:
3311 	case SQ_GSTMP_RING_SIZE:
3312 	case SQ_HSTMP_RING_SIZE:
3313 	case SQ_LSTMP_RING_SIZE:
3314 	case SQ_PSTMP_RING_SIZE:
3315 	case SQ_VSTMP_RING_SIZE:
3316 	case SQ_ESGS_RING_ITEMSIZE:
3317 	case SQ_ESTMP_RING_ITEMSIZE:
3318 	case SQ_GSTMP_RING_ITEMSIZE:
3319 	case SQ_GSVS_RING_ITEMSIZE:
3320 	case SQ_GS_VERT_ITEMSIZE:
3321 	case SQ_GS_VERT_ITEMSIZE_1:
3322 	case SQ_GS_VERT_ITEMSIZE_2:
3323 	case SQ_GS_VERT_ITEMSIZE_3:
3324 	case SQ_GSVS_RING_OFFSET_1:
3325 	case SQ_GSVS_RING_OFFSET_2:
3326 	case SQ_GSVS_RING_OFFSET_3:
3327 	case SQ_HSTMP_RING_ITEMSIZE:
3328 	case SQ_LSTMP_RING_ITEMSIZE:
3329 	case SQ_PSTMP_RING_ITEMSIZE:
3330 	case SQ_VSTMP_RING_ITEMSIZE:
3331 	case VGT_TF_RING_SIZE:
3332 	case SQ_ESGS_RING_BASE:
3333 	case SQ_GSVS_RING_BASE:
3334 	case SQ_ESTMP_RING_BASE:
3335 	case SQ_GSTMP_RING_BASE:
3336 	case SQ_HSTMP_RING_BASE:
3337 	case SQ_LSTMP_RING_BASE:
3338 	case SQ_PSTMP_RING_BASE:
3339 	case SQ_VSTMP_RING_BASE:
3340 	case CAYMAN_VGT_OFFCHIP_LDS_BASE:
3341 	case CAYMAN_SQ_EX_ALLOC_TABLE_SLOTS:
3342 		return true;
3343 	default:
3344 		DRM_ERROR("Invalid register 0x%x in CS\n", reg);
3345 		return false;
3346 	}
3347 }
3348 
3349 static int evergreen_vm_packet3_check(struct radeon_device *rdev,
3350 				      u32 *ib, struct radeon_cs_packet *pkt)
3351 {
3352 	u32 idx = pkt->idx + 1;
3353 	u32 idx_value = ib[idx];
3354 	u32 start_reg, end_reg, reg, i;
3355 	u32 command, info;
3356 
3357 	switch (pkt->opcode) {
3358 	case PACKET3_NOP:
3359 		break;
3360 	case PACKET3_SET_BASE:
3361 		if (idx_value != 1) {
3362 			DRM_ERROR("bad SET_BASE");
3363 			return -EINVAL;
3364 		}
3365 		break;
3366 	case PACKET3_CLEAR_STATE:
3367 	case PACKET3_INDEX_BUFFER_SIZE:
3368 	case PACKET3_DISPATCH_DIRECT:
3369 	case PACKET3_DISPATCH_INDIRECT:
3370 	case PACKET3_MODE_CONTROL:
3371 	case PACKET3_SET_PREDICATION:
3372 	case PACKET3_COND_EXEC:
3373 	case PACKET3_PRED_EXEC:
3374 	case PACKET3_DRAW_INDIRECT:
3375 	case PACKET3_DRAW_INDEX_INDIRECT:
3376 	case PACKET3_INDEX_BASE:
3377 	case PACKET3_DRAW_INDEX_2:
3378 	case PACKET3_CONTEXT_CONTROL:
3379 	case PACKET3_DRAW_INDEX_OFFSET:
3380 	case PACKET3_INDEX_TYPE:
3381 	case PACKET3_DRAW_INDEX:
3382 	case PACKET3_DRAW_INDEX_AUTO:
3383 	case PACKET3_DRAW_INDEX_IMMD:
3384 	case PACKET3_NUM_INSTANCES:
3385 	case PACKET3_DRAW_INDEX_MULTI_AUTO:
3386 	case PACKET3_STRMOUT_BUFFER_UPDATE:
3387 	case PACKET3_DRAW_INDEX_OFFSET_2:
3388 	case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
3389 	case PACKET3_MPEG_INDEX:
3390 	case PACKET3_WAIT_REG_MEM:
3391 	case PACKET3_MEM_WRITE:
3392 	case PACKET3_PFP_SYNC_ME:
3393 	case PACKET3_SURFACE_SYNC:
3394 	case PACKET3_EVENT_WRITE:
3395 	case PACKET3_EVENT_WRITE_EOP:
3396 	case PACKET3_EVENT_WRITE_EOS:
3397 	case PACKET3_SET_CONTEXT_REG:
3398 	case PACKET3_SET_BOOL_CONST:
3399 	case PACKET3_SET_LOOP_CONST:
3400 	case PACKET3_SET_RESOURCE:
3401 	case PACKET3_SET_SAMPLER:
3402 	case PACKET3_SET_CTL_CONST:
3403 	case PACKET3_SET_RESOURCE_OFFSET:
3404 	case PACKET3_SET_CONTEXT_REG_INDIRECT:
3405 	case PACKET3_SET_RESOURCE_INDIRECT:
3406 	case CAYMAN_PACKET3_DEALLOC_STATE:
3407 		break;
3408 	case PACKET3_COND_WRITE:
3409 		if (idx_value & 0x100) {
3410 			reg = ib[idx + 5] * 4;
3411 			if (!evergreen_vm_reg_valid(reg))
3412 				return -EINVAL;
3413 		}
3414 		break;
3415 	case PACKET3_COPY_DW:
3416 		if (idx_value & 0x2) {
3417 			reg = ib[idx + 3] * 4;
3418 			if (!evergreen_vm_reg_valid(reg))
3419 				return -EINVAL;
3420 		}
3421 		break;
3422 	case PACKET3_SET_CONFIG_REG:
3423 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
3424 		end_reg = 4 * pkt->count + start_reg - 4;
3425 		if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
3426 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
3427 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
3428 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
3429 			return -EINVAL;
3430 		}
3431 		for (i = 0; i < pkt->count; i++) {
3432 			reg = start_reg + (4 * i);
3433 			if (!evergreen_vm_reg_valid(reg))
3434 				return -EINVAL;
3435 		}
3436 		break;
3437 	case PACKET3_CP_DMA:
3438 		command = ib[idx + 4];
3439 		info = ib[idx + 1];
3440 		if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
3441 		    (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
3442 		    ((((info & 0x00300000) >> 20) == 0) &&
3443 		     (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
3444 		    ((((info & 0x60000000) >> 29) == 0) &&
3445 		     (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
3446 			/* non mem to mem copies requires dw aligned count */
3447 			if ((command & 0x1fffff) % 4) {
3448 				DRM_ERROR("CP DMA command requires dw count alignment\n");
3449 				return -EINVAL;
3450 			}
3451 		}
3452 		if (command & PACKET3_CP_DMA_CMD_SAS) {
3453 			/* src address space is register */
3454 			if (((info & 0x60000000) >> 29) == 0) {
3455 				start_reg = idx_value << 2;
3456 				if (command & PACKET3_CP_DMA_CMD_SAIC) {
3457 					reg = start_reg;
3458 					if (!evergreen_vm_reg_valid(reg)) {
3459 						DRM_ERROR("CP DMA Bad SRC register\n");
3460 						return -EINVAL;
3461 					}
3462 				} else {
3463 					for (i = 0; i < (command & 0x1fffff); i++) {
3464 						reg = start_reg + (4 * i);
3465 						if (!evergreen_vm_reg_valid(reg)) {
3466 							DRM_ERROR("CP DMA Bad SRC register\n");
3467 							return -EINVAL;
3468 						}
3469 					}
3470 				}
3471 			}
3472 		}
3473 		if (command & PACKET3_CP_DMA_CMD_DAS) {
3474 			/* dst address space is register */
3475 			if (((info & 0x00300000) >> 20) == 0) {
3476 				start_reg = ib[idx + 2];
3477 				if (command & PACKET3_CP_DMA_CMD_DAIC) {
3478 					reg = start_reg;
3479 					if (!evergreen_vm_reg_valid(reg)) {
3480 						DRM_ERROR("CP DMA Bad DST register\n");
3481 						return -EINVAL;
3482 					}
3483 				} else {
3484 					for (i = 0; i < (command & 0x1fffff); i++) {
3485 						reg = start_reg + (4 * i);
3486 						if (!evergreen_vm_reg_valid(reg)) {
3487 							DRM_ERROR("CP DMA Bad DST register\n");
3488 							return -EINVAL;
3489 						}
3490 					}
3491 				}
3492 			}
3493 		}
3494 		break;
3495 	case PACKET3_SET_APPEND_CNT: {
3496 		uint32_t areg;
3497 		uint32_t allowed_reg_base;
3498 
3499 		if (pkt->count != 2) {
3500 			DRM_ERROR("bad SET_APPEND_CNT (invalid count)\n");
3501 			return -EINVAL;
3502 		}
3503 
3504 		allowed_reg_base = GDS_APPEND_COUNT_0;
3505 		allowed_reg_base -= PACKET3_SET_CONTEXT_REG_START;
3506 		allowed_reg_base >>= 2;
3507 
3508 		areg = idx_value >> 16;
3509 		if (areg < allowed_reg_base || areg > (allowed_reg_base + 11)) {
3510 			DRM_ERROR("forbidden register for append cnt 0x%08x at %d\n",
3511 				  areg, idx);
3512 			return -EINVAL;
3513 		}
3514 		break;
3515 	}
3516 	default:
3517 		return -EINVAL;
3518 	}
3519 	return 0;
3520 }
3521 
3522 int evergreen_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3523 {
3524 	int ret = 0;
3525 	u32 idx = 0;
3526 	struct radeon_cs_packet pkt;
3527 
3528 	do {
3529 		pkt.idx = idx;
3530 		pkt.type = RADEON_CP_PACKET_GET_TYPE(ib->ptr[idx]);
3531 		pkt.count = RADEON_CP_PACKET_GET_COUNT(ib->ptr[idx]);
3532 		pkt.one_reg_wr = 0;
3533 		switch (pkt.type) {
3534 		case RADEON_PACKET_TYPE0:
3535 			dev_err(rdev->dev, "Packet0 not allowed!\n");
3536 			ret = -EINVAL;
3537 			break;
3538 		case RADEON_PACKET_TYPE2:
3539 			idx += 1;
3540 			break;
3541 		case RADEON_PACKET_TYPE3:
3542 			pkt.opcode = RADEON_CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
3543 			ret = evergreen_vm_packet3_check(rdev, ib->ptr, &pkt);
3544 			idx += pkt.count + 2;
3545 			break;
3546 		default:
3547 			dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
3548 			ret = -EINVAL;
3549 			break;
3550 		}
3551 		if (ret)
3552 			break;
3553 	} while (idx < ib->length_dw);
3554 
3555 	return ret;
3556 }
3557 
3558 /**
3559  * evergreen_dma_ib_parse() - parse the DMA IB for VM
3560  * @rdev: radeon_device pointer
3561  * @ib:	radeon_ib pointer
3562  *
3563  * Parses the DMA IB from the VM CS ioctl
3564  * checks for errors. (Cayman-SI)
3565  * Returns 0 for success and an error on failure.
3566  **/
3567 int evergreen_dma_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3568 {
3569 	u32 idx = 0;
3570 	u32 header, cmd, count, sub_cmd;
3571 
3572 	do {
3573 		header = ib->ptr[idx];
3574 		cmd = GET_DMA_CMD(header);
3575 		count = GET_DMA_COUNT(header);
3576 		sub_cmd = GET_DMA_SUB_CMD(header);
3577 
3578 		switch (cmd) {
3579 		case DMA_PACKET_WRITE:
3580 			switch (sub_cmd) {
3581 			/* tiled */
3582 			case 8:
3583 				idx += count + 7;
3584 				break;
3585 			/* linear */
3586 			case 0:
3587 				idx += count + 3;
3588 				break;
3589 			default:
3590 				DRM_ERROR("bad DMA_PACKET_WRITE [%6d] 0x%08x sub cmd is not 0 or 8\n", idx, ib->ptr[idx]);
3591 				return -EINVAL;
3592 			}
3593 			break;
3594 		case DMA_PACKET_COPY:
3595 			switch (sub_cmd) {
3596 			/* Copy L2L, DW aligned */
3597 			case 0x00:
3598 				idx += 5;
3599 				break;
3600 			/* Copy L2T/T2L */
3601 			case 0x08:
3602 				idx += 9;
3603 				break;
3604 			/* Copy L2L, byte aligned */
3605 			case 0x40:
3606 				idx += 5;
3607 				break;
3608 			/* Copy L2L, partial */
3609 			case 0x41:
3610 				idx += 9;
3611 				break;
3612 			/* Copy L2L, DW aligned, broadcast */
3613 			case 0x44:
3614 				idx += 7;
3615 				break;
3616 			/* Copy L2T Frame to Field */
3617 			case 0x48:
3618 				idx += 10;
3619 				break;
3620 			/* Copy L2T/T2L, partial */
3621 			case 0x49:
3622 				idx += 12;
3623 				break;
3624 			/* Copy L2T broadcast */
3625 			case 0x4b:
3626 				idx += 10;
3627 				break;
3628 			/* Copy L2T/T2L (tile units) */
3629 			case 0x4c:
3630 				idx += 9;
3631 				break;
3632 			/* Copy T2T, partial (tile units) */
3633 			case 0x4d:
3634 				idx += 13;
3635 				break;
3636 			/* Copy L2T broadcast (tile units) */
3637 			case 0x4f:
3638 				idx += 10;
3639 				break;
3640 			default:
3641 				DRM_ERROR("bad DMA_PACKET_COPY [%6d] 0x%08x invalid sub cmd\n", idx, ib->ptr[idx]);
3642 				return -EINVAL;
3643 			}
3644 			break;
3645 		case DMA_PACKET_CONSTANT_FILL:
3646 			idx += 4;
3647 			break;
3648 		case DMA_PACKET_NOP:
3649 			idx += 1;
3650 			break;
3651 		default:
3652 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
3653 			return -EINVAL;
3654 		}
3655 	} while (idx < ib->length_dw);
3656 
3657 	return 0;
3658 }
3659