1 /*
2  * Copyright 2008 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include <linux/kernel.h>
29 #include "drmP.h"
30 #include "radeon.h"
31 #include "r600d.h"
32 #include "r600_reg_safe.h"
33 
34 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
35 					struct radeon_cs_reloc **cs_reloc);
36 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
37 					struct radeon_cs_reloc **cs_reloc);
38 typedef int (*next_reloc_t)(struct radeon_cs_parser*, struct radeon_cs_reloc**);
39 static next_reloc_t r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_mm;
40 extern void r600_cs_legacy_get_tiling_conf(struct drm_device *dev, u32 *npipes, u32 *nbanks, u32 *group_size);
41 
42 
43 struct r600_cs_track {
44 	/* configuration we miror so that we use same code btw kms/ums */
45 	u32			group_size;
46 	u32			nbanks;
47 	u32			npipes;
48 	/* value we track */
49 	u32			sq_config;
50 	u32			nsamples;
51 	u32			cb_color_base_last[8];
52 	struct radeon_bo	*cb_color_bo[8];
53 	u64			cb_color_bo_mc[8];
54 	u32			cb_color_bo_offset[8];
55 	struct radeon_bo	*cb_color_frag_bo[8];
56 	struct radeon_bo	*cb_color_tile_bo[8];
57 	u32			cb_color_info[8];
58 	u32			cb_color_size_idx[8];
59 	u32			cb_target_mask;
60 	u32			cb_shader_mask;
61 	u32			cb_color_size[8];
62 	u32			vgt_strmout_en;
63 	u32			vgt_strmout_buffer_en;
64 	u32			db_depth_control;
65 	u32			db_depth_info;
66 	u32			db_depth_size_idx;
67 	u32			db_depth_view;
68 	u32			db_depth_size;
69 	u32			db_offset;
70 	struct radeon_bo	*db_bo;
71 	u64			db_bo_mc;
72 };
73 
74 #define FMT_8_BIT(fmt, vc) [fmt] = { 1, 1, 1, vc }
75 #define FMT_16_BIT(fmt, vc) [fmt] = { 1, 1, 2, vc }
76 #define FMT_24_BIT(fmt) [fmt] = { 1, 1, 3, 0 }
77 #define FMT_32_BIT(fmt, vc) [fmt] = { 1, 1, 4, vc }
78 #define FMT_48_BIT(fmt) [fmt] = { 1, 1, 6, 0 }
79 #define FMT_64_BIT(fmt, vc) [fmt] = { 1, 1, 8, vc }
80 #define FMT_96_BIT(fmt) [fmt] = { 1, 1, 12, 0 }
81 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16, vc }
82 
83 struct gpu_formats {
84 	unsigned blockwidth;
85 	unsigned blockheight;
86 	unsigned blocksize;
87 	unsigned valid_color;
88 };
89 
90 static const struct gpu_formats color_formats_table[] = {
91 	/* 8 bit */
92 	FMT_8_BIT(V_038004_COLOR_8, 1),
93 	FMT_8_BIT(V_038004_COLOR_4_4, 1),
94 	FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
95 	FMT_8_BIT(V_038004_FMT_1, 0),
96 
97 	/* 16-bit */
98 	FMT_16_BIT(V_038004_COLOR_16, 1),
99 	FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
100 	FMT_16_BIT(V_038004_COLOR_8_8, 1),
101 	FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
102 	FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
103 	FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
104 	FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
105 	FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
106 
107 	/* 24-bit */
108 	FMT_24_BIT(V_038004_FMT_8_8_8),
109 
110 	/* 32-bit */
111 	FMT_32_BIT(V_038004_COLOR_32, 1),
112 	FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
113 	FMT_32_BIT(V_038004_COLOR_16_16, 1),
114 	FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
115 	FMT_32_BIT(V_038004_COLOR_8_24, 1),
116 	FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
117 	FMT_32_BIT(V_038004_COLOR_24_8, 1),
118 	FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
119 	FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
120 	FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
121 	FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
122 	FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
123 	FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
124 	FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
125 	FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
126 	FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
127 	FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
128 	FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
129 
130 	/* 48-bit */
131 	FMT_48_BIT(V_038004_FMT_16_16_16),
132 	FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
133 
134 	/* 64-bit */
135 	FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
136 	FMT_64_BIT(V_038004_COLOR_32_32, 1),
137 	FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
138 	FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
139 	FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
140 
141 	FMT_96_BIT(V_038004_FMT_32_32_32),
142 	FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
143 
144 	/* 128-bit */
145 	FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
146 	FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
147 
148 	[V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
149 	[V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
150 
151 	/* block compressed formats */
152 	[V_038004_FMT_BC1] = { 4, 4, 8, 0 },
153 	[V_038004_FMT_BC2] = { 4, 4, 16, 0 },
154 	[V_038004_FMT_BC3] = { 4, 4, 16, 0 },
155 	[V_038004_FMT_BC4] = { 4, 4, 8, 0 },
156 	[V_038004_FMT_BC5] = { 4, 4, 16, 0},
157 
158 };
159 
fmt_is_valid_color(u32 format)160 static inline bool fmt_is_valid_color(u32 format)
161 {
162 	if (format >= ARRAY_SIZE(color_formats_table))
163 		return false;
164 
165 	if (color_formats_table[format].valid_color)
166 		return true;
167 
168 	return false;
169 }
170 
fmt_is_valid_texture(u32 format)171 static inline bool fmt_is_valid_texture(u32 format)
172 {
173 	if (format >= ARRAY_SIZE(color_formats_table))
174 		return false;
175 
176 	if (color_formats_table[format].blockwidth > 0)
177 		return true;
178 
179 	return false;
180 }
181 
fmt_get_blocksize(u32 format)182 static inline int fmt_get_blocksize(u32 format)
183 {
184 	if (format >= ARRAY_SIZE(color_formats_table))
185 		return 0;
186 
187 	return color_formats_table[format].blocksize;
188 }
189 
fmt_get_nblocksx(u32 format,u32 w)190 static inline int fmt_get_nblocksx(u32 format, u32 w)
191 {
192 	unsigned bw;
193 
194 	if (format >= ARRAY_SIZE(color_formats_table))
195 		return 0;
196 
197 	bw = color_formats_table[format].blockwidth;
198 	if (bw == 0)
199 		return 0;
200 
201 	return (w + bw - 1) / bw;
202 }
203 
fmt_get_nblocksy(u32 format,u32 h)204 static inline int fmt_get_nblocksy(u32 format, u32 h)
205 {
206 	unsigned bh;
207 
208 	if (format >= ARRAY_SIZE(color_formats_table))
209 		return 0;
210 
211 	bh = color_formats_table[format].blockheight;
212 	if (bh == 0)
213 		return 0;
214 
215 	return (h + bh - 1) / bh;
216 }
217 
r600_bpe_from_format(u32 * bpe,u32 format)218 static inline int r600_bpe_from_format(u32 *bpe, u32 format)
219 {
220  	unsigned res;
221 
222 	if (format >= ARRAY_SIZE(color_formats_table))
223 		goto fail;
224 
225 	res = color_formats_table[format].blocksize;
226 	if (res == 0)
227 		goto fail;
228 
229 	*bpe = res;
230 	return 0;
231 
232 fail:
233 	*bpe = 16;
234 	return -EINVAL;
235 }
236 
237 struct array_mode_checker {
238 	int array_mode;
239 	u32 group_size;
240 	u32 nbanks;
241 	u32 npipes;
242 	u32 nsamples;
243 	u32 blocksize;
244 };
245 
246 /* returns alignment in pixels for pitch/height/depth and bytes for base */
r600_get_array_mode_alignment(struct array_mode_checker * values,u32 * pitch_align,u32 * height_align,u32 * depth_align,u64 * base_align)247 static inline int r600_get_array_mode_alignment(struct array_mode_checker *values,
248 						u32 *pitch_align,
249 						u32 *height_align,
250 						u32 *depth_align,
251 						u64 *base_align)
252 {
253 	u32 tile_width = 8;
254 	u32 tile_height = 8;
255 	u32 macro_tile_width = values->nbanks;
256 	u32 macro_tile_height = values->npipes;
257 	u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
258 	u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
259 
260 	switch (values->array_mode) {
261 	case ARRAY_LINEAR_GENERAL:
262 		/* technically tile_width/_height for pitch/height */
263 		*pitch_align = 1; /* tile_width */
264 		*height_align = 1; /* tile_height */
265 		*depth_align = 1;
266 		*base_align = 1;
267 		break;
268 	case ARRAY_LINEAR_ALIGNED:
269 		*pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
270 		*height_align = tile_height;
271 		*depth_align = 1;
272 		*base_align = values->group_size;
273 		break;
274 	case ARRAY_1D_TILED_THIN1:
275 		*pitch_align = max((u32)tile_width,
276 				   (u32)(values->group_size /
277 					 (tile_height * values->blocksize * values->nsamples)));
278 		*height_align = tile_height;
279 		*depth_align = 1;
280 		*base_align = values->group_size;
281 		break;
282 	case ARRAY_2D_TILED_THIN1:
283 		*pitch_align = max((u32)macro_tile_width,
284 				  (u32)(((values->group_size / tile_height) /
285 					 (values->blocksize * values->nsamples)) *
286 					values->nbanks)) * tile_width;
287 		*height_align = macro_tile_height * tile_height;
288 		*depth_align = 1;
289 		*base_align = max(macro_tile_bytes,
290 				  (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
291 		break;
292 	default:
293 		return -EINVAL;
294 	}
295 
296 	return 0;
297 }
298 
r600_cs_track_init(struct r600_cs_track * track)299 static void r600_cs_track_init(struct r600_cs_track *track)
300 {
301 	int i;
302 
303 	/* assume DX9 mode */
304 	track->sq_config = DX9_CONSTS;
305 	for (i = 0; i < 8; i++) {
306 		track->cb_color_base_last[i] = 0;
307 		track->cb_color_size[i] = 0;
308 		track->cb_color_size_idx[i] = 0;
309 		track->cb_color_info[i] = 0;
310 		track->cb_color_bo[i] = NULL;
311 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
312 		track->cb_color_bo_mc[i] = 0xFFFFFFFF;
313 	}
314 	track->cb_target_mask = 0xFFFFFFFF;
315 	track->cb_shader_mask = 0xFFFFFFFF;
316 	track->db_bo = NULL;
317 	track->db_bo_mc = 0xFFFFFFFF;
318 	/* assume the biggest format and that htile is enabled */
319 	track->db_depth_info = 7 | (1 << 25);
320 	track->db_depth_view = 0xFFFFC000;
321 	track->db_depth_size = 0xFFFFFFFF;
322 	track->db_depth_size_idx = 0;
323 	track->db_depth_control = 0xFFFFFFFF;
324 }
325 
r600_cs_track_validate_cb(struct radeon_cs_parser * p,int i)326 static inline int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
327 {
328 	struct r600_cs_track *track = p->track;
329 	u32 slice_tile_max, size, tmp;
330 	u32 height, height_align, pitch, pitch_align, depth_align;
331 	u64 base_offset, base_align;
332 	struct array_mode_checker array_check;
333 	volatile u32 *ib = p->ib->ptr;
334 	unsigned array_mode;
335 	u32 format;
336 	if (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
337 		dev_warn(p->dev, "FMASK or CMASK buffer are not supported by this kernel\n");
338 		return -EINVAL;
339 	}
340 	size = radeon_bo_size(track->cb_color_bo[i]) - track->cb_color_bo_offset[i];
341 	format = G_0280A0_FORMAT(track->cb_color_info[i]);
342 	if (!fmt_is_valid_color(format)) {
343 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
344 			 __func__, __LINE__, format,
345 			i, track->cb_color_info[i]);
346 		return -EINVAL;
347 	}
348 	/* pitch in pixels */
349 	pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
350 	slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
351 	slice_tile_max *= 64;
352 	height = slice_tile_max / pitch;
353 	if (height > 8192)
354 		height = 8192;
355 	array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
356 
357 	base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
358 	array_check.array_mode = array_mode;
359 	array_check.group_size = track->group_size;
360 	array_check.nbanks = track->nbanks;
361 	array_check.npipes = track->npipes;
362 	array_check.nsamples = track->nsamples;
363 	array_check.blocksize = fmt_get_blocksize(format);
364 	if (r600_get_array_mode_alignment(&array_check,
365 					  &pitch_align, &height_align, &depth_align, &base_align)) {
366 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
367 			 G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
368 			 track->cb_color_info[i]);
369 		return -EINVAL;
370 	}
371 	switch (array_mode) {
372 	case V_0280A0_ARRAY_LINEAR_GENERAL:
373 		break;
374 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
375 		break;
376 	case V_0280A0_ARRAY_1D_TILED_THIN1:
377 		/* avoid breaking userspace */
378 		if (height > 7)
379 			height &= ~0x7;
380 		break;
381 	case V_0280A0_ARRAY_2D_TILED_THIN1:
382 		break;
383 	default:
384 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
385 			G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
386 			track->cb_color_info[i]);
387 		return -EINVAL;
388 	}
389 
390 	if (!IS_ALIGNED(pitch, pitch_align)) {
391 		dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
392 			 __func__, __LINE__, pitch, pitch_align, array_mode);
393 		return -EINVAL;
394 	}
395 	if (!IS_ALIGNED(height, height_align)) {
396 		dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
397 			 __func__, __LINE__, height, height_align, array_mode);
398 		return -EINVAL;
399 	}
400 	if (!IS_ALIGNED(base_offset, base_align)) {
401 		dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
402 			 base_offset, base_align, array_mode);
403 		return -EINVAL;
404 	}
405 
406 	/* check offset */
407 	tmp = fmt_get_nblocksy(format, height) * fmt_get_nblocksx(format, pitch) * fmt_get_blocksize(format);
408 	if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
409 		if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
410 			/* the initial DDX does bad things with the CB size occasionally */
411 			/* it rounds up height too far for slice tile max but the BO is smaller */
412 			/* r600c,g also seem to flush at bad times in some apps resulting in
413 			 * bogus values here. So for linear just allow anything to avoid breaking
414 			 * broken userspace.
415 			 */
416 		} else {
417 			dev_warn(p->dev, "%s offset[%d] %d %d %d %lu too big\n", __func__, i,
418 				 array_mode,
419 				 track->cb_color_bo_offset[i], tmp,
420 				 radeon_bo_size(track->cb_color_bo[i]));
421 			return -EINVAL;
422 		}
423 	}
424 	/* limit max tile */
425 	tmp = (height * pitch) >> 6;
426 	if (tmp < slice_tile_max)
427 		slice_tile_max = tmp;
428 	tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
429 		S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
430 	ib[track->cb_color_size_idx[i]] = tmp;
431 	return 0;
432 }
433 
r600_cs_track_check(struct radeon_cs_parser * p)434 static int r600_cs_track_check(struct radeon_cs_parser *p)
435 {
436 	struct r600_cs_track *track = p->track;
437 	u32 tmp;
438 	int r, i;
439 	volatile u32 *ib = p->ib->ptr;
440 
441 	/* on legacy kernel we don't perform advanced check */
442 	if (p->rdev == NULL)
443 		return 0;
444 	/* we don't support out buffer yet */
445 	if (track->vgt_strmout_en || track->vgt_strmout_buffer_en) {
446 		dev_warn(p->dev, "this kernel doesn't support SMX output buffer\n");
447 		return -EINVAL;
448 	}
449 	/* check that we have a cb for each enabled target, we don't check
450 	 * shader_mask because it seems mesa isn't always setting it :(
451 	 */
452 	tmp = track->cb_target_mask;
453 	for (i = 0; i < 8; i++) {
454 		if ((tmp >> (i * 4)) & 0xF) {
455 			/* at least one component is enabled */
456 			if (track->cb_color_bo[i] == NULL) {
457 				dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
458 					__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
459 				return -EINVAL;
460 			}
461 			/* perform rewrite of CB_COLOR[0-7]_SIZE */
462 			r = r600_cs_track_validate_cb(p, i);
463 			if (r)
464 				return r;
465 		}
466 	}
467 	/* Check depth buffer */
468 	if (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
469 		G_028800_Z_ENABLE(track->db_depth_control)) {
470 		u32 nviews, bpe, ntiles, size, slice_tile_max;
471 		u32 height, height_align, pitch, pitch_align, depth_align;
472 		u64 base_offset, base_align;
473 		struct array_mode_checker array_check;
474 		int array_mode;
475 
476 		if (track->db_bo == NULL) {
477 			dev_warn(p->dev, "z/stencil with no depth buffer\n");
478 			return -EINVAL;
479 		}
480 		if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
481 			dev_warn(p->dev, "this kernel doesn't support z/stencil htile\n");
482 			return -EINVAL;
483 		}
484 		switch (G_028010_FORMAT(track->db_depth_info)) {
485 		case V_028010_DEPTH_16:
486 			bpe = 2;
487 			break;
488 		case V_028010_DEPTH_X8_24:
489 		case V_028010_DEPTH_8_24:
490 		case V_028010_DEPTH_X8_24_FLOAT:
491 		case V_028010_DEPTH_8_24_FLOAT:
492 		case V_028010_DEPTH_32_FLOAT:
493 			bpe = 4;
494 			break;
495 		case V_028010_DEPTH_X24_8_32_FLOAT:
496 			bpe = 8;
497 			break;
498 		default:
499 			dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
500 			return -EINVAL;
501 		}
502 		if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
503 			if (!track->db_depth_size_idx) {
504 				dev_warn(p->dev, "z/stencil buffer size not set\n");
505 				return -EINVAL;
506 			}
507 			tmp = radeon_bo_size(track->db_bo) - track->db_offset;
508 			tmp = (tmp / bpe) >> 6;
509 			if (!tmp) {
510 				dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
511 						track->db_depth_size, bpe, track->db_offset,
512 						radeon_bo_size(track->db_bo));
513 				return -EINVAL;
514 			}
515 			ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
516 		} else {
517 			size = radeon_bo_size(track->db_bo);
518 			/* pitch in pixels */
519 			pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
520 			slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
521 			slice_tile_max *= 64;
522 			height = slice_tile_max / pitch;
523 			if (height > 8192)
524 				height = 8192;
525 			base_offset = track->db_bo_mc + track->db_offset;
526 			array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
527 			array_check.array_mode = array_mode;
528 			array_check.group_size = track->group_size;
529 			array_check.nbanks = track->nbanks;
530 			array_check.npipes = track->npipes;
531 			array_check.nsamples = track->nsamples;
532 			array_check.blocksize = bpe;
533 			if (r600_get_array_mode_alignment(&array_check,
534 							  &pitch_align, &height_align, &depth_align, &base_align)) {
535 				dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
536 					 G_028010_ARRAY_MODE(track->db_depth_info),
537 					 track->db_depth_info);
538 				return -EINVAL;
539 			}
540 			switch (array_mode) {
541 			case V_028010_ARRAY_1D_TILED_THIN1:
542 				/* don't break userspace */
543 				height &= ~0x7;
544 				break;
545 			case V_028010_ARRAY_2D_TILED_THIN1:
546 				break;
547 			default:
548 				dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
549 					 G_028010_ARRAY_MODE(track->db_depth_info),
550 					 track->db_depth_info);
551 				return -EINVAL;
552 			}
553 
554 			if (!IS_ALIGNED(pitch, pitch_align)) {
555 				dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
556 					 __func__, __LINE__, pitch, pitch_align, array_mode);
557 				return -EINVAL;
558 			}
559 			if (!IS_ALIGNED(height, height_align)) {
560 				dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
561 					 __func__, __LINE__, height, height_align, array_mode);
562 				return -EINVAL;
563 			}
564 			if (!IS_ALIGNED(base_offset, base_align)) {
565 				dev_warn(p->dev, "%s offset[%d] 0x%llx, 0x%llx, %d not aligned\n", __func__, i,
566 					 base_offset, base_align, array_mode);
567 				return -EINVAL;
568 			}
569 
570 			ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
571 			nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
572 			tmp = ntiles * bpe * 64 * nviews;
573 			if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
574 				dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
575 					 array_mode,
576 					 track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
577 					 radeon_bo_size(track->db_bo));
578 				return -EINVAL;
579 			}
580 		}
581 	}
582 	return 0;
583 }
584 
585 /**
586  * r600_cs_packet_parse() - parse cp packet and point ib index to next packet
587  * @parser:	parser structure holding parsing context.
588  * @pkt:	where to store packet informations
589  *
590  * Assume that chunk_ib_index is properly set. Will return -EINVAL
591  * if packet is bigger than remaining ib size. or if packets is unknown.
592  **/
r600_cs_packet_parse(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,unsigned idx)593 int r600_cs_packet_parse(struct radeon_cs_parser *p,
594 			struct radeon_cs_packet *pkt,
595 			unsigned idx)
596 {
597 	struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
598 	uint32_t header;
599 
600 	if (idx >= ib_chunk->length_dw) {
601 		DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
602 			  idx, ib_chunk->length_dw);
603 		return -EINVAL;
604 	}
605 	header = radeon_get_ib_value(p, idx);
606 	pkt->idx = idx;
607 	pkt->type = CP_PACKET_GET_TYPE(header);
608 	pkt->count = CP_PACKET_GET_COUNT(header);
609 	pkt->one_reg_wr = 0;
610 	switch (pkt->type) {
611 	case PACKET_TYPE0:
612 		pkt->reg = CP_PACKET0_GET_REG(header);
613 		break;
614 	case PACKET_TYPE3:
615 		pkt->opcode = CP_PACKET3_GET_OPCODE(header);
616 		break;
617 	case PACKET_TYPE2:
618 		pkt->count = -1;
619 		break;
620 	default:
621 		DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
622 		return -EINVAL;
623 	}
624 	if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
625 		DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
626 			  pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
627 		return -EINVAL;
628 	}
629 	return 0;
630 }
631 
632 /**
633  * r600_cs_packet_next_reloc_mm() - parse next packet which should be reloc packet3
634  * @parser:		parser structure holding parsing context.
635  * @data:		pointer to relocation data
636  * @offset_start:	starting offset
637  * @offset_mask:	offset mask (to align start offset on)
638  * @reloc:		reloc informations
639  *
640  * Check next packet is relocation packet3, do bo validation and compute
641  * GPU offset using the provided start.
642  **/
r600_cs_packet_next_reloc_mm(struct radeon_cs_parser * p,struct radeon_cs_reloc ** cs_reloc)643 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
644 					struct radeon_cs_reloc **cs_reloc)
645 {
646 	struct radeon_cs_chunk *relocs_chunk;
647 	struct radeon_cs_packet p3reloc;
648 	unsigned idx;
649 	int r;
650 
651 	if (p->chunk_relocs_idx == -1) {
652 		DRM_ERROR("No relocation chunk !\n");
653 		return -EINVAL;
654 	}
655 	*cs_reloc = NULL;
656 	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
657 	r = r600_cs_packet_parse(p, &p3reloc, p->idx);
658 	if (r) {
659 		return r;
660 	}
661 	p->idx += p3reloc.count + 2;
662 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
663 		DRM_ERROR("No packet3 for relocation for packet at %d.\n",
664 			  p3reloc.idx);
665 		return -EINVAL;
666 	}
667 	idx = radeon_get_ib_value(p, p3reloc.idx + 1);
668 	if (idx >= relocs_chunk->length_dw) {
669 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
670 			  idx, relocs_chunk->length_dw);
671 		return -EINVAL;
672 	}
673 	/* FIXME: we assume reloc size is 4 dwords */
674 	*cs_reloc = p->relocs_ptr[(idx / 4)];
675 	return 0;
676 }
677 
678 /**
679  * r600_cs_packet_next_reloc_nomm() - parse next packet which should be reloc packet3
680  * @parser:		parser structure holding parsing context.
681  * @data:		pointer to relocation data
682  * @offset_start:	starting offset
683  * @offset_mask:	offset mask (to align start offset on)
684  * @reloc:		reloc informations
685  *
686  * Check next packet is relocation packet3, do bo validation and compute
687  * GPU offset using the provided start.
688  **/
r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser * p,struct radeon_cs_reloc ** cs_reloc)689 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
690 					struct radeon_cs_reloc **cs_reloc)
691 {
692 	struct radeon_cs_chunk *relocs_chunk;
693 	struct radeon_cs_packet p3reloc;
694 	unsigned idx;
695 	int r;
696 
697 	if (p->chunk_relocs_idx == -1) {
698 		DRM_ERROR("No relocation chunk !\n");
699 		return -EINVAL;
700 	}
701 	*cs_reloc = NULL;
702 	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
703 	r = r600_cs_packet_parse(p, &p3reloc, p->idx);
704 	if (r) {
705 		return r;
706 	}
707 	p->idx += p3reloc.count + 2;
708 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
709 		DRM_ERROR("No packet3 for relocation for packet at %d.\n",
710 			  p3reloc.idx);
711 		return -EINVAL;
712 	}
713 	idx = radeon_get_ib_value(p, p3reloc.idx + 1);
714 	if (idx >= relocs_chunk->length_dw) {
715 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
716 			  idx, relocs_chunk->length_dw);
717 		return -EINVAL;
718 	}
719 	*cs_reloc = p->relocs;
720 	(*cs_reloc)->lobj.gpu_offset = (u64)relocs_chunk->kdata[idx + 3] << 32;
721 	(*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
722 	return 0;
723 }
724 
725 /**
726  * r600_cs_packet_next_is_pkt3_nop() - test if next packet is packet3 nop for reloc
727  * @parser:		parser structure holding parsing context.
728  *
729  * Check next packet is relocation packet3, do bo validation and compute
730  * GPU offset using the provided start.
731  **/
r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser * p)732 static inline int r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
733 {
734 	struct radeon_cs_packet p3reloc;
735 	int r;
736 
737 	r = r600_cs_packet_parse(p, &p3reloc, p->idx);
738 	if (r) {
739 		return 0;
740 	}
741 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
742 		return 0;
743 	}
744 	return 1;
745 }
746 
747 /**
748  * r600_cs_packet_next_vline() - parse userspace VLINE packet
749  * @parser:		parser structure holding parsing context.
750  *
751  * Userspace sends a special sequence for VLINE waits.
752  * PACKET0 - VLINE_START_END + value
753  * PACKET3 - WAIT_REG_MEM poll vline status reg
754  * RELOC (P3) - crtc_id in reloc.
755  *
756  * This function parses this and relocates the VLINE START END
757  * and WAIT_REG_MEM packets to the correct crtc.
758  * It also detects a switched off crtc and nulls out the
759  * wait in that case.
760  */
r600_cs_packet_parse_vline(struct radeon_cs_parser * p)761 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
762 {
763 	struct drm_mode_object *obj;
764 	struct drm_crtc *crtc;
765 	struct radeon_crtc *radeon_crtc;
766 	struct radeon_cs_packet p3reloc, wait_reg_mem;
767 	int crtc_id;
768 	int r;
769 	uint32_t header, h_idx, reg, wait_reg_mem_info;
770 	volatile uint32_t *ib;
771 
772 	ib = p->ib->ptr;
773 
774 	/* parse the WAIT_REG_MEM */
775 	r = r600_cs_packet_parse(p, &wait_reg_mem, p->idx);
776 	if (r)
777 		return r;
778 
779 	/* check its a WAIT_REG_MEM */
780 	if (wait_reg_mem.type != PACKET_TYPE3 ||
781 	    wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
782 		DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
783 		return -EINVAL;
784 	}
785 
786 	wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
787 	/* bit 4 is reg (0) or mem (1) */
788 	if (wait_reg_mem_info & 0x10) {
789 		DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
790 		return -EINVAL;
791 	}
792 	/* waiting for value to be equal */
793 	if ((wait_reg_mem_info & 0x7) != 0x3) {
794 		DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
795 		return -EINVAL;
796 	}
797 	if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != AVIVO_D1MODE_VLINE_STATUS) {
798 		DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
799 		return -EINVAL;
800 	}
801 
802 	if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != AVIVO_D1MODE_VLINE_STAT) {
803 		DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
804 		return -EINVAL;
805 	}
806 
807 	/* jump over the NOP */
808 	r = r600_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
809 	if (r)
810 		return r;
811 
812 	h_idx = p->idx - 2;
813 	p->idx += wait_reg_mem.count + 2;
814 	p->idx += p3reloc.count + 2;
815 
816 	header = radeon_get_ib_value(p, h_idx);
817 	crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
818 	reg = CP_PACKET0_GET_REG(header);
819 
820 	obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
821 	if (!obj) {
822 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
823 		return -EINVAL;
824 	}
825 	crtc = obj_to_crtc(obj);
826 	radeon_crtc = to_radeon_crtc(crtc);
827 	crtc_id = radeon_crtc->crtc_id;
828 
829 	if (!crtc->enabled) {
830 		/* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
831 		ib[h_idx + 2] = PACKET2(0);
832 		ib[h_idx + 3] = PACKET2(0);
833 		ib[h_idx + 4] = PACKET2(0);
834 		ib[h_idx + 5] = PACKET2(0);
835 		ib[h_idx + 6] = PACKET2(0);
836 		ib[h_idx + 7] = PACKET2(0);
837 		ib[h_idx + 8] = PACKET2(0);
838 	} else if (crtc_id == 1) {
839 		switch (reg) {
840 		case AVIVO_D1MODE_VLINE_START_END:
841 			header &= ~R600_CP_PACKET0_REG_MASK;
842 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
843 			break;
844 		default:
845 			DRM_ERROR("unknown crtc reloc\n");
846 			return -EINVAL;
847 		}
848 		ib[h_idx] = header;
849 		ib[h_idx + 4] = AVIVO_D2MODE_VLINE_STATUS >> 2;
850 	}
851 
852 	return 0;
853 }
854 
r600_packet0_check(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,unsigned idx,unsigned reg)855 static int r600_packet0_check(struct radeon_cs_parser *p,
856 				struct radeon_cs_packet *pkt,
857 				unsigned idx, unsigned reg)
858 {
859 	int r;
860 
861 	switch (reg) {
862 	case AVIVO_D1MODE_VLINE_START_END:
863 		r = r600_cs_packet_parse_vline(p);
864 		if (r) {
865 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
866 					idx, reg);
867 			return r;
868 		}
869 		break;
870 	default:
871 		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
872 		       reg, idx);
873 		return -EINVAL;
874 	}
875 	return 0;
876 }
877 
r600_cs_parse_packet0(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt)878 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
879 				struct radeon_cs_packet *pkt)
880 {
881 	unsigned reg, i;
882 	unsigned idx;
883 	int r;
884 
885 	idx = pkt->idx + 1;
886 	reg = pkt->reg;
887 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
888 		r = r600_packet0_check(p, pkt, idx, reg);
889 		if (r) {
890 			return r;
891 		}
892 	}
893 	return 0;
894 }
895 
896 /**
897  * r600_cs_check_reg() - check if register is authorized or not
898  * @parser: parser structure holding parsing context
899  * @reg: register we are testing
900  * @idx: index into the cs buffer
901  *
902  * This function will test against r600_reg_safe_bm and return 0
903  * if register is safe. If register is not flag as safe this function
904  * will test it against a list of register needind special handling.
905  */
r600_cs_check_reg(struct radeon_cs_parser * p,u32 reg,u32 idx)906 static inline int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
907 {
908 	struct r600_cs_track *track = (struct r600_cs_track *)p->track;
909 	struct radeon_cs_reloc *reloc;
910 	u32 last_reg = ARRAY_SIZE(r600_reg_safe_bm);
911 	u32 m, i, tmp, *ib;
912 	int r;
913 
914 	i = (reg >> 7);
915 	if (i > last_reg) {
916 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
917 		return -EINVAL;
918 	}
919 	m = 1 << ((reg >> 2) & 31);
920 	if (!(r600_reg_safe_bm[i] & m))
921 		return 0;
922 	ib = p->ib->ptr;
923 	switch (reg) {
924 	/* force following reg to 0 in an attempt to disable out buffer
925 	 * which will need us to better understand how it works to perform
926 	 * security check on it (Jerome)
927 	 */
928 	case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
929 	case R_008C44_SQ_ESGS_RING_SIZE:
930 	case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
931 	case R_008C54_SQ_ESTMP_RING_SIZE:
932 	case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
933 	case R_008C74_SQ_FBUF_RING_SIZE:
934 	case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
935 	case R_008C5C_SQ_GSTMP_RING_SIZE:
936 	case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
937 	case R_008C4C_SQ_GSVS_RING_SIZE:
938 	case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
939 	case R_008C6C_SQ_PSTMP_RING_SIZE:
940 	case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
941 	case R_008C7C_SQ_REDUC_RING_SIZE:
942 	case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
943 	case R_008C64_SQ_VSTMP_RING_SIZE:
944 	case R_0288C8_SQ_GS_VERT_ITEMSIZE:
945 		/* get value to populate the IB don't remove */
946 		tmp =radeon_get_ib_value(p, idx);
947 		ib[idx] = 0;
948 		break;
949 	case SQ_CONFIG:
950 		track->sq_config = radeon_get_ib_value(p, idx);
951 		break;
952 	case R_028800_DB_DEPTH_CONTROL:
953 		track->db_depth_control = radeon_get_ib_value(p, idx);
954 		break;
955 	case R_028010_DB_DEPTH_INFO:
956 		if (r600_cs_packet_next_is_pkt3_nop(p)) {
957 			r = r600_cs_packet_next_reloc(p, &reloc);
958 			if (r) {
959 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
960 					 "0x%04X\n", reg);
961 				return -EINVAL;
962 			}
963 			track->db_depth_info = radeon_get_ib_value(p, idx);
964 			ib[idx] &= C_028010_ARRAY_MODE;
965 			track->db_depth_info &= C_028010_ARRAY_MODE;
966 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
967 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
968 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
969 			} else {
970 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
971 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
972 			}
973 		} else
974 			track->db_depth_info = radeon_get_ib_value(p, idx);
975 		break;
976 	case R_028004_DB_DEPTH_VIEW:
977 		track->db_depth_view = radeon_get_ib_value(p, idx);
978 		break;
979 	case R_028000_DB_DEPTH_SIZE:
980 		track->db_depth_size = radeon_get_ib_value(p, idx);
981 		track->db_depth_size_idx = idx;
982 		break;
983 	case R_028AB0_VGT_STRMOUT_EN:
984 		track->vgt_strmout_en = radeon_get_ib_value(p, idx);
985 		break;
986 	case R_028B20_VGT_STRMOUT_BUFFER_EN:
987 		track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
988 		break;
989 	case R_028238_CB_TARGET_MASK:
990 		track->cb_target_mask = radeon_get_ib_value(p, idx);
991 		break;
992 	case R_02823C_CB_SHADER_MASK:
993 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
994 		break;
995 	case R_028C04_PA_SC_AA_CONFIG:
996 		tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
997 		track->nsamples = 1 << tmp;
998 		break;
999 	case R_0280A0_CB_COLOR0_INFO:
1000 	case R_0280A4_CB_COLOR1_INFO:
1001 	case R_0280A8_CB_COLOR2_INFO:
1002 	case R_0280AC_CB_COLOR3_INFO:
1003 	case R_0280B0_CB_COLOR4_INFO:
1004 	case R_0280B4_CB_COLOR5_INFO:
1005 	case R_0280B8_CB_COLOR6_INFO:
1006 	case R_0280BC_CB_COLOR7_INFO:
1007 		if (r600_cs_packet_next_is_pkt3_nop(p)) {
1008 			r = r600_cs_packet_next_reloc(p, &reloc);
1009 			if (r) {
1010 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1011 				return -EINVAL;
1012 			}
1013 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1014 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1015 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1016 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1017 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1018 			} else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
1019 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1020 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1021 			}
1022 		} else {
1023 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1024 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1025 		}
1026 		break;
1027 	case R_028060_CB_COLOR0_SIZE:
1028 	case R_028064_CB_COLOR1_SIZE:
1029 	case R_028068_CB_COLOR2_SIZE:
1030 	case R_02806C_CB_COLOR3_SIZE:
1031 	case R_028070_CB_COLOR4_SIZE:
1032 	case R_028074_CB_COLOR5_SIZE:
1033 	case R_028078_CB_COLOR6_SIZE:
1034 	case R_02807C_CB_COLOR7_SIZE:
1035 		tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1036 		track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1037 		track->cb_color_size_idx[tmp] = idx;
1038 		break;
1039 		/* This register were added late, there is userspace
1040 		 * which does provide relocation for those but set
1041 		 * 0 offset. In order to avoid breaking old userspace
1042 		 * we detect this and set address to point to last
1043 		 * CB_COLOR0_BASE, note that if userspace doesn't set
1044 		 * CB_COLOR0_BASE before this register we will report
1045 		 * error. Old userspace always set CB_COLOR0_BASE
1046 		 * before any of this.
1047 		 */
1048 	case R_0280E0_CB_COLOR0_FRAG:
1049 	case R_0280E4_CB_COLOR1_FRAG:
1050 	case R_0280E8_CB_COLOR2_FRAG:
1051 	case R_0280EC_CB_COLOR3_FRAG:
1052 	case R_0280F0_CB_COLOR4_FRAG:
1053 	case R_0280F4_CB_COLOR5_FRAG:
1054 	case R_0280F8_CB_COLOR6_FRAG:
1055 	case R_0280FC_CB_COLOR7_FRAG:
1056 		tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1057 		if (!r600_cs_packet_next_is_pkt3_nop(p)) {
1058 			if (!track->cb_color_base_last[tmp]) {
1059 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1060 				return -EINVAL;
1061 			}
1062 			ib[idx] = track->cb_color_base_last[tmp];
1063 			track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1064 		} else {
1065 			r = r600_cs_packet_next_reloc(p, &reloc);
1066 			if (r) {
1067 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1068 				return -EINVAL;
1069 			}
1070 			ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1071 			track->cb_color_frag_bo[tmp] = reloc->robj;
1072 		}
1073 		break;
1074 	case R_0280C0_CB_COLOR0_TILE:
1075 	case R_0280C4_CB_COLOR1_TILE:
1076 	case R_0280C8_CB_COLOR2_TILE:
1077 	case R_0280CC_CB_COLOR3_TILE:
1078 	case R_0280D0_CB_COLOR4_TILE:
1079 	case R_0280D4_CB_COLOR5_TILE:
1080 	case R_0280D8_CB_COLOR6_TILE:
1081 	case R_0280DC_CB_COLOR7_TILE:
1082 		tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1083 		if (!r600_cs_packet_next_is_pkt3_nop(p)) {
1084 			if (!track->cb_color_base_last[tmp]) {
1085 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1086 				return -EINVAL;
1087 			}
1088 			ib[idx] = track->cb_color_base_last[tmp];
1089 			track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1090 		} else {
1091 			r = r600_cs_packet_next_reloc(p, &reloc);
1092 			if (r) {
1093 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1094 				return -EINVAL;
1095 			}
1096 			ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1097 			track->cb_color_tile_bo[tmp] = reloc->robj;
1098 		}
1099 		break;
1100 	case CB_COLOR0_BASE:
1101 	case CB_COLOR1_BASE:
1102 	case CB_COLOR2_BASE:
1103 	case CB_COLOR3_BASE:
1104 	case CB_COLOR4_BASE:
1105 	case CB_COLOR5_BASE:
1106 	case CB_COLOR6_BASE:
1107 	case CB_COLOR7_BASE:
1108 		r = r600_cs_packet_next_reloc(p, &reloc);
1109 		if (r) {
1110 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1111 					"0x%04X\n", reg);
1112 			return -EINVAL;
1113 		}
1114 		tmp = (reg - CB_COLOR0_BASE) / 4;
1115 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1116 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1117 		track->cb_color_base_last[tmp] = ib[idx];
1118 		track->cb_color_bo[tmp] = reloc->robj;
1119 		track->cb_color_bo_mc[tmp] = reloc->lobj.gpu_offset;
1120 		break;
1121 	case DB_DEPTH_BASE:
1122 		r = r600_cs_packet_next_reloc(p, &reloc);
1123 		if (r) {
1124 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1125 					"0x%04X\n", reg);
1126 			return -EINVAL;
1127 		}
1128 		track->db_offset = radeon_get_ib_value(p, idx) << 8;
1129 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1130 		track->db_bo = reloc->robj;
1131 		track->db_bo_mc = reloc->lobj.gpu_offset;
1132 		break;
1133 	case DB_HTILE_DATA_BASE:
1134 	case SQ_PGM_START_FS:
1135 	case SQ_PGM_START_ES:
1136 	case SQ_PGM_START_VS:
1137 	case SQ_PGM_START_GS:
1138 	case SQ_PGM_START_PS:
1139 	case SQ_ALU_CONST_CACHE_GS_0:
1140 	case SQ_ALU_CONST_CACHE_GS_1:
1141 	case SQ_ALU_CONST_CACHE_GS_2:
1142 	case SQ_ALU_CONST_CACHE_GS_3:
1143 	case SQ_ALU_CONST_CACHE_GS_4:
1144 	case SQ_ALU_CONST_CACHE_GS_5:
1145 	case SQ_ALU_CONST_CACHE_GS_6:
1146 	case SQ_ALU_CONST_CACHE_GS_7:
1147 	case SQ_ALU_CONST_CACHE_GS_8:
1148 	case SQ_ALU_CONST_CACHE_GS_9:
1149 	case SQ_ALU_CONST_CACHE_GS_10:
1150 	case SQ_ALU_CONST_CACHE_GS_11:
1151 	case SQ_ALU_CONST_CACHE_GS_12:
1152 	case SQ_ALU_CONST_CACHE_GS_13:
1153 	case SQ_ALU_CONST_CACHE_GS_14:
1154 	case SQ_ALU_CONST_CACHE_GS_15:
1155 	case SQ_ALU_CONST_CACHE_PS_0:
1156 	case SQ_ALU_CONST_CACHE_PS_1:
1157 	case SQ_ALU_CONST_CACHE_PS_2:
1158 	case SQ_ALU_CONST_CACHE_PS_3:
1159 	case SQ_ALU_CONST_CACHE_PS_4:
1160 	case SQ_ALU_CONST_CACHE_PS_5:
1161 	case SQ_ALU_CONST_CACHE_PS_6:
1162 	case SQ_ALU_CONST_CACHE_PS_7:
1163 	case SQ_ALU_CONST_CACHE_PS_8:
1164 	case SQ_ALU_CONST_CACHE_PS_9:
1165 	case SQ_ALU_CONST_CACHE_PS_10:
1166 	case SQ_ALU_CONST_CACHE_PS_11:
1167 	case SQ_ALU_CONST_CACHE_PS_12:
1168 	case SQ_ALU_CONST_CACHE_PS_13:
1169 	case SQ_ALU_CONST_CACHE_PS_14:
1170 	case SQ_ALU_CONST_CACHE_PS_15:
1171 	case SQ_ALU_CONST_CACHE_VS_0:
1172 	case SQ_ALU_CONST_CACHE_VS_1:
1173 	case SQ_ALU_CONST_CACHE_VS_2:
1174 	case SQ_ALU_CONST_CACHE_VS_3:
1175 	case SQ_ALU_CONST_CACHE_VS_4:
1176 	case SQ_ALU_CONST_CACHE_VS_5:
1177 	case SQ_ALU_CONST_CACHE_VS_6:
1178 	case SQ_ALU_CONST_CACHE_VS_7:
1179 	case SQ_ALU_CONST_CACHE_VS_8:
1180 	case SQ_ALU_CONST_CACHE_VS_9:
1181 	case SQ_ALU_CONST_CACHE_VS_10:
1182 	case SQ_ALU_CONST_CACHE_VS_11:
1183 	case SQ_ALU_CONST_CACHE_VS_12:
1184 	case SQ_ALU_CONST_CACHE_VS_13:
1185 	case SQ_ALU_CONST_CACHE_VS_14:
1186 	case SQ_ALU_CONST_CACHE_VS_15:
1187 		r = r600_cs_packet_next_reloc(p, &reloc);
1188 		if (r) {
1189 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1190 					"0x%04X\n", reg);
1191 			return -EINVAL;
1192 		}
1193 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1194 		break;
1195 	default:
1196 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1197 		return -EINVAL;
1198 	}
1199 	return 0;
1200 }
1201 
mip_minify(unsigned size,unsigned level)1202 static inline unsigned mip_minify(unsigned size, unsigned level)
1203 {
1204 	unsigned val;
1205 
1206 	val = max(1U, size >> level);
1207 	if (level > 0)
1208 		val = roundup_pow_of_two(val);
1209 	return val;
1210 }
1211 
r600_texture_size(unsigned nfaces,unsigned blevel,unsigned llevel,unsigned w0,unsigned h0,unsigned d0,unsigned format,unsigned block_align,unsigned height_align,unsigned base_align,unsigned * l0_size,unsigned * mipmap_size)1212 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1213 			      unsigned w0, unsigned h0, unsigned d0, unsigned format,
1214 			      unsigned block_align, unsigned height_align, unsigned base_align,
1215 			      unsigned *l0_size, unsigned *mipmap_size)
1216 {
1217 	unsigned offset, i, level;
1218 	unsigned width, height, depth, size;
1219 	unsigned blocksize;
1220 	unsigned nbx, nby;
1221 	unsigned nlevels = llevel - blevel + 1;
1222 
1223 	*l0_size = -1;
1224 	blocksize = fmt_get_blocksize(format);
1225 
1226 	w0 = mip_minify(w0, 0);
1227 	h0 = mip_minify(h0, 0);
1228 	d0 = mip_minify(d0, 0);
1229 	for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1230 		width = mip_minify(w0, i);
1231 		nbx = fmt_get_nblocksx(format, width);
1232 
1233 		nbx = round_up(nbx, block_align);
1234 
1235 		height = mip_minify(h0, i);
1236 		nby = fmt_get_nblocksy(format, height);
1237 		nby = round_up(nby, height_align);
1238 
1239 		depth = mip_minify(d0, i);
1240 
1241 		size = nbx * nby * blocksize;
1242 		if (nfaces)
1243 			size *= nfaces;
1244 		else
1245 			size *= depth;
1246 
1247 		if (i == 0)
1248 			*l0_size = size;
1249 
1250 		if (i == 0 || i == 1)
1251 			offset = round_up(offset, base_align);
1252 
1253 		offset += size;
1254 	}
1255 	*mipmap_size = offset;
1256 	if (llevel == 0)
1257 		*mipmap_size = *l0_size;
1258 	if (!blevel)
1259 		*mipmap_size -= *l0_size;
1260 }
1261 
1262 /**
1263  * r600_check_texture_resource() - check if register is authorized or not
1264  * @p: parser structure holding parsing context
1265  * @idx: index into the cs buffer
1266  * @texture: texture's bo structure
1267  * @mipmap: mipmap's bo structure
1268  *
1269  * This function will check that the resource has valid field and that
1270  * the texture and mipmap bo object are big enough to cover this resource.
1271  */
r600_check_texture_resource(struct radeon_cs_parser * p,u32 idx,struct radeon_bo * texture,struct radeon_bo * mipmap,u64 base_offset,u64 mip_offset,u32 tiling_flags)1272 static inline int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1273 					      struct radeon_bo *texture,
1274 					      struct radeon_bo *mipmap,
1275 					      u64 base_offset,
1276 					      u64 mip_offset,
1277 					      u32 tiling_flags)
1278 {
1279 	struct r600_cs_track *track = p->track;
1280 	u32 nfaces, llevel, blevel, w0, h0, d0;
1281 	u32 word0, word1, l0_size, mipmap_size, word2, word3;
1282 	u32 height_align, pitch, pitch_align, depth_align;
1283 	u32 array, barray, larray;
1284 	u64 base_align;
1285 	struct array_mode_checker array_check;
1286 	u32 format;
1287 
1288 	/* on legacy kernel we don't perform advanced check */
1289 	if (p->rdev == NULL)
1290 		return 0;
1291 
1292 	/* convert to bytes */
1293 	base_offset <<= 8;
1294 	mip_offset <<= 8;
1295 
1296 	word0 = radeon_get_ib_value(p, idx + 0);
1297 	if (tiling_flags & RADEON_TILING_MACRO)
1298 		word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1299 	else if (tiling_flags & RADEON_TILING_MICRO)
1300 		word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1301 	word1 = radeon_get_ib_value(p, idx + 1);
1302 	w0 = G_038000_TEX_WIDTH(word0) + 1;
1303 	h0 = G_038004_TEX_HEIGHT(word1) + 1;
1304 	d0 = G_038004_TEX_DEPTH(word1);
1305 	nfaces = 1;
1306 	switch (G_038000_DIM(word0)) {
1307 	case V_038000_SQ_TEX_DIM_1D:
1308 	case V_038000_SQ_TEX_DIM_2D:
1309 	case V_038000_SQ_TEX_DIM_3D:
1310 		break;
1311 	case V_038000_SQ_TEX_DIM_CUBEMAP:
1312 		if (p->family >= CHIP_RV770)
1313 			nfaces = 8;
1314 		else
1315 			nfaces = 6;
1316 		break;
1317 	case V_038000_SQ_TEX_DIM_1D_ARRAY:
1318 	case V_038000_SQ_TEX_DIM_2D_ARRAY:
1319 		array = 1;
1320 		break;
1321 	case V_038000_SQ_TEX_DIM_2D_MSAA:
1322 	case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1323 	default:
1324 		dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1325 		return -EINVAL;
1326 	}
1327 	format = G_038004_DATA_FORMAT(word1);
1328 	if (!fmt_is_valid_texture(format)) {
1329 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1330 			 __func__, __LINE__, format);
1331 		return -EINVAL;
1332 	}
1333 
1334 	/* pitch in texels */
1335 	pitch = (G_038000_PITCH(word0) + 1) * 8;
1336 	array_check.array_mode = G_038000_TILE_MODE(word0);
1337 	array_check.group_size = track->group_size;
1338 	array_check.nbanks = track->nbanks;
1339 	array_check.npipes = track->npipes;
1340 	array_check.nsamples = 1;
1341 	array_check.blocksize = fmt_get_blocksize(format);
1342 	if (r600_get_array_mode_alignment(&array_check,
1343 					  &pitch_align, &height_align, &depth_align, &base_align)) {
1344 		dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1345 			 __func__, __LINE__, G_038000_TILE_MODE(word0));
1346 		return -EINVAL;
1347 	}
1348 
1349 	/* XXX check height as well... */
1350 
1351 	if (!IS_ALIGNED(pitch, pitch_align)) {
1352 		dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1353 			 __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1354 		return -EINVAL;
1355 	}
1356 	if (!IS_ALIGNED(base_offset, base_align)) {
1357 		dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1358 			 __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1359 		return -EINVAL;
1360 	}
1361 	if (!IS_ALIGNED(mip_offset, base_align)) {
1362 		dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1363 			 __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1364 		return -EINVAL;
1365 	}
1366 
1367 	word2 = radeon_get_ib_value(p, idx + 2) << 8;
1368 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1369 
1370 	word0 = radeon_get_ib_value(p, idx + 4);
1371 	word1 = radeon_get_ib_value(p, idx + 5);
1372 	blevel = G_038010_BASE_LEVEL(word0);
1373 	llevel = G_038014_LAST_LEVEL(word1);
1374 	if (array == 1) {
1375 		barray = G_038014_BASE_ARRAY(word1);
1376 		larray = G_038014_LAST_ARRAY(word1);
1377 
1378 		nfaces = larray - barray + 1;
1379 	}
1380 	r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, format,
1381 			  pitch_align, height_align, base_align,
1382 			  &l0_size, &mipmap_size);
1383 	/* using get ib will give us the offset into the texture bo */
1384 	if ((l0_size + word2) > radeon_bo_size(texture)) {
1385 		dev_warn(p->dev, "texture bo too small (%d %d %d %d -> %d have %ld)\n",
1386 			w0, h0, format, word2, l0_size, radeon_bo_size(texture));
1387 		dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1388 		return -EINVAL;
1389 	}
1390 	/* using get ib will give us the offset into the mipmap bo */
1391 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1392 	if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1393 		/*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1394 		  w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1395 	}
1396 	return 0;
1397 }
1398 
r600_packet3_check(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt)1399 static int r600_packet3_check(struct radeon_cs_parser *p,
1400 				struct radeon_cs_packet *pkt)
1401 {
1402 	struct radeon_cs_reloc *reloc;
1403 	struct r600_cs_track *track;
1404 	volatile u32 *ib;
1405 	unsigned idx;
1406 	unsigned i;
1407 	unsigned start_reg, end_reg, reg;
1408 	int r;
1409 	u32 idx_value;
1410 
1411 	track = (struct r600_cs_track *)p->track;
1412 	ib = p->ib->ptr;
1413 	idx = pkt->idx + 1;
1414 	idx_value = radeon_get_ib_value(p, idx);
1415 
1416 	switch (pkt->opcode) {
1417 	case PACKET3_SET_PREDICATION:
1418 	{
1419 		int pred_op;
1420 		int tmp;
1421 		if (pkt->count != 1) {
1422 			DRM_ERROR("bad SET PREDICATION\n");
1423 			return -EINVAL;
1424 		}
1425 
1426 		tmp = radeon_get_ib_value(p, idx + 1);
1427 		pred_op = (tmp >> 16) & 0x7;
1428 
1429 		/* for the clear predicate operation */
1430 		if (pred_op == 0)
1431 			return 0;
1432 
1433 		if (pred_op > 2) {
1434 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1435 			return -EINVAL;
1436 		}
1437 
1438 		r = r600_cs_packet_next_reloc(p, &reloc);
1439 		if (r) {
1440 			DRM_ERROR("bad SET PREDICATION\n");
1441 			return -EINVAL;
1442 		}
1443 
1444 		ib[idx + 0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1445 		ib[idx + 1] = tmp + (upper_32_bits(reloc->lobj.gpu_offset) & 0xff);
1446 	}
1447 	break;
1448 
1449 	case PACKET3_START_3D_CMDBUF:
1450 		if (p->family >= CHIP_RV770 || pkt->count) {
1451 			DRM_ERROR("bad START_3D\n");
1452 			return -EINVAL;
1453 		}
1454 		break;
1455 	case PACKET3_CONTEXT_CONTROL:
1456 		if (pkt->count != 1) {
1457 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1458 			return -EINVAL;
1459 		}
1460 		break;
1461 	case PACKET3_INDEX_TYPE:
1462 	case PACKET3_NUM_INSTANCES:
1463 		if (pkt->count) {
1464 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1465 			return -EINVAL;
1466 		}
1467 		break;
1468 	case PACKET3_DRAW_INDEX:
1469 		if (pkt->count != 3) {
1470 			DRM_ERROR("bad DRAW_INDEX\n");
1471 			return -EINVAL;
1472 		}
1473 		r = r600_cs_packet_next_reloc(p, &reloc);
1474 		if (r) {
1475 			DRM_ERROR("bad DRAW_INDEX\n");
1476 			return -EINVAL;
1477 		}
1478 		ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1479 		ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1480 		r = r600_cs_track_check(p);
1481 		if (r) {
1482 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1483 			return r;
1484 		}
1485 		break;
1486 	case PACKET3_DRAW_INDEX_AUTO:
1487 		if (pkt->count != 1) {
1488 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1489 			return -EINVAL;
1490 		}
1491 		r = r600_cs_track_check(p);
1492 		if (r) {
1493 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1494 			return r;
1495 		}
1496 		break;
1497 	case PACKET3_DRAW_INDEX_IMMD_BE:
1498 	case PACKET3_DRAW_INDEX_IMMD:
1499 		if (pkt->count < 2) {
1500 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1501 			return -EINVAL;
1502 		}
1503 		r = r600_cs_track_check(p);
1504 		if (r) {
1505 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1506 			return r;
1507 		}
1508 		break;
1509 	case PACKET3_WAIT_REG_MEM:
1510 		if (pkt->count != 5) {
1511 			DRM_ERROR("bad WAIT_REG_MEM\n");
1512 			return -EINVAL;
1513 		}
1514 		/* bit 4 is reg (0) or mem (1) */
1515 		if (idx_value & 0x10) {
1516 			r = r600_cs_packet_next_reloc(p, &reloc);
1517 			if (r) {
1518 				DRM_ERROR("bad WAIT_REG_MEM\n");
1519 				return -EINVAL;
1520 			}
1521 			ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1522 			ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1523 		}
1524 		break;
1525 	case PACKET3_SURFACE_SYNC:
1526 		if (pkt->count != 3) {
1527 			DRM_ERROR("bad SURFACE_SYNC\n");
1528 			return -EINVAL;
1529 		}
1530 		/* 0xffffffff/0x0 is flush all cache flag */
1531 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1532 		    radeon_get_ib_value(p, idx + 2) != 0) {
1533 			r = r600_cs_packet_next_reloc(p, &reloc);
1534 			if (r) {
1535 				DRM_ERROR("bad SURFACE_SYNC\n");
1536 				return -EINVAL;
1537 			}
1538 			ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1539 		}
1540 		break;
1541 	case PACKET3_EVENT_WRITE:
1542 		if (pkt->count != 2 && pkt->count != 0) {
1543 			DRM_ERROR("bad EVENT_WRITE\n");
1544 			return -EINVAL;
1545 		}
1546 		if (pkt->count) {
1547 			r = r600_cs_packet_next_reloc(p, &reloc);
1548 			if (r) {
1549 				DRM_ERROR("bad EVENT_WRITE\n");
1550 				return -EINVAL;
1551 			}
1552 			ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1553 			ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1554 		}
1555 		break;
1556 	case PACKET3_EVENT_WRITE_EOP:
1557 		if (pkt->count != 4) {
1558 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1559 			return -EINVAL;
1560 		}
1561 		r = r600_cs_packet_next_reloc(p, &reloc);
1562 		if (r) {
1563 			DRM_ERROR("bad EVENT_WRITE\n");
1564 			return -EINVAL;
1565 		}
1566 		ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1567 		ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1568 		break;
1569 	case PACKET3_SET_CONFIG_REG:
1570 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1571 		end_reg = 4 * pkt->count + start_reg - 4;
1572 		if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1573 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1574 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1575 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1576 			return -EINVAL;
1577 		}
1578 		for (i = 0; i < pkt->count; i++) {
1579 			reg = start_reg + (4 * i);
1580 			r = r600_cs_check_reg(p, reg, idx+1+i);
1581 			if (r)
1582 				return r;
1583 		}
1584 		break;
1585 	case PACKET3_SET_CONTEXT_REG:
1586 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1587 		end_reg = 4 * pkt->count + start_reg - 4;
1588 		if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1589 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1590 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1591 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1592 			return -EINVAL;
1593 		}
1594 		for (i = 0; i < pkt->count; i++) {
1595 			reg = start_reg + (4 * i);
1596 			r = r600_cs_check_reg(p, reg, idx+1+i);
1597 			if (r)
1598 				return r;
1599 		}
1600 		break;
1601 	case PACKET3_SET_RESOURCE:
1602 		if (pkt->count % 7) {
1603 			DRM_ERROR("bad SET_RESOURCE\n");
1604 			return -EINVAL;
1605 		}
1606 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1607 		end_reg = 4 * pkt->count + start_reg - 4;
1608 		if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1609 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
1610 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
1611 			DRM_ERROR("bad SET_RESOURCE\n");
1612 			return -EINVAL;
1613 		}
1614 		for (i = 0; i < (pkt->count / 7); i++) {
1615 			struct radeon_bo *texture, *mipmap;
1616 			u32 size, offset, base_offset, mip_offset;
1617 
1618 			switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1619 			case SQ_TEX_VTX_VALID_TEXTURE:
1620 				/* tex base */
1621 				r = r600_cs_packet_next_reloc(p, &reloc);
1622 				if (r) {
1623 					DRM_ERROR("bad SET_RESOURCE\n");
1624 					return -EINVAL;
1625 				}
1626 				base_offset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1627 				if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1628 					ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1629 				else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1630 					ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1631 				texture = reloc->robj;
1632 				/* tex mip base */
1633 				r = r600_cs_packet_next_reloc(p, &reloc);
1634 				if (r) {
1635 					DRM_ERROR("bad SET_RESOURCE\n");
1636 					return -EINVAL;
1637 				}
1638 				mip_offset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1639 				mipmap = reloc->robj;
1640 				r = r600_check_texture_resource(p,  idx+(i*7)+1,
1641 								texture, mipmap,
1642 								base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1643 								mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1644 								reloc->lobj.tiling_flags);
1645 				if (r)
1646 					return r;
1647 				ib[idx+1+(i*7)+2] += base_offset;
1648 				ib[idx+1+(i*7)+3] += mip_offset;
1649 				break;
1650 			case SQ_TEX_VTX_VALID_BUFFER:
1651 				/* vtx base */
1652 				r = r600_cs_packet_next_reloc(p, &reloc);
1653 				if (r) {
1654 					DRM_ERROR("bad SET_RESOURCE\n");
1655 					return -EINVAL;
1656 				}
1657 				offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
1658 				size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
1659 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
1660 					/* force size to size of the buffer */
1661 					dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
1662 						 size + offset, radeon_bo_size(reloc->robj));
1663 					ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj);
1664 				}
1665 				ib[idx+1+(i*7)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
1666 				ib[idx+1+(i*7)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1667 				break;
1668 			case SQ_TEX_VTX_INVALID_TEXTURE:
1669 			case SQ_TEX_VTX_INVALID_BUFFER:
1670 			default:
1671 				DRM_ERROR("bad SET_RESOURCE\n");
1672 				return -EINVAL;
1673 			}
1674 		}
1675 		break;
1676 	case PACKET3_SET_ALU_CONST:
1677 		if (track->sq_config & DX9_CONSTS) {
1678 			start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
1679 			end_reg = 4 * pkt->count + start_reg - 4;
1680 			if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
1681 			    (start_reg >= PACKET3_SET_ALU_CONST_END) ||
1682 			    (end_reg >= PACKET3_SET_ALU_CONST_END)) {
1683 				DRM_ERROR("bad SET_ALU_CONST\n");
1684 				return -EINVAL;
1685 			}
1686 		}
1687 		break;
1688 	case PACKET3_SET_BOOL_CONST:
1689 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
1690 		end_reg = 4 * pkt->count + start_reg - 4;
1691 		if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
1692 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
1693 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
1694 			DRM_ERROR("bad SET_BOOL_CONST\n");
1695 			return -EINVAL;
1696 		}
1697 		break;
1698 	case PACKET3_SET_LOOP_CONST:
1699 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
1700 		end_reg = 4 * pkt->count + start_reg - 4;
1701 		if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
1702 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
1703 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
1704 			DRM_ERROR("bad SET_LOOP_CONST\n");
1705 			return -EINVAL;
1706 		}
1707 		break;
1708 	case PACKET3_SET_CTL_CONST:
1709 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
1710 		end_reg = 4 * pkt->count + start_reg - 4;
1711 		if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
1712 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
1713 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
1714 			DRM_ERROR("bad SET_CTL_CONST\n");
1715 			return -EINVAL;
1716 		}
1717 		break;
1718 	case PACKET3_SET_SAMPLER:
1719 		if (pkt->count % 3) {
1720 			DRM_ERROR("bad SET_SAMPLER\n");
1721 			return -EINVAL;
1722 		}
1723 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
1724 		end_reg = 4 * pkt->count + start_reg - 4;
1725 		if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
1726 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
1727 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
1728 			DRM_ERROR("bad SET_SAMPLER\n");
1729 			return -EINVAL;
1730 		}
1731 		break;
1732 	case PACKET3_SURFACE_BASE_UPDATE:
1733 		if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
1734 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1735 			return -EINVAL;
1736 		}
1737 		if (pkt->count) {
1738 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
1739 			return -EINVAL;
1740 		}
1741 		break;
1742 	case PACKET3_NOP:
1743 		break;
1744 	default:
1745 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1746 		return -EINVAL;
1747 	}
1748 	return 0;
1749 }
1750 
r600_cs_parse(struct radeon_cs_parser * p)1751 int r600_cs_parse(struct radeon_cs_parser *p)
1752 {
1753 	struct radeon_cs_packet pkt;
1754 	struct r600_cs_track *track;
1755 	int r;
1756 
1757 	if (p->track == NULL) {
1758 		/* initialize tracker, we are in kms */
1759 		track = kzalloc(sizeof(*track), GFP_KERNEL);
1760 		if (track == NULL)
1761 			return -ENOMEM;
1762 		r600_cs_track_init(track);
1763 		if (p->rdev->family < CHIP_RV770) {
1764 			track->npipes = p->rdev->config.r600.tiling_npipes;
1765 			track->nbanks = p->rdev->config.r600.tiling_nbanks;
1766 			track->group_size = p->rdev->config.r600.tiling_group_size;
1767 		} else if (p->rdev->family <= CHIP_RV740) {
1768 			track->npipes = p->rdev->config.rv770.tiling_npipes;
1769 			track->nbanks = p->rdev->config.rv770.tiling_nbanks;
1770 			track->group_size = p->rdev->config.rv770.tiling_group_size;
1771 		}
1772 		p->track = track;
1773 	}
1774 	do {
1775 		r = r600_cs_packet_parse(p, &pkt, p->idx);
1776 		if (r) {
1777 			kfree(p->track);
1778 			p->track = NULL;
1779 			return r;
1780 		}
1781 		p->idx += pkt.count + 2;
1782 		switch (pkt.type) {
1783 		case PACKET_TYPE0:
1784 			r = r600_cs_parse_packet0(p, &pkt);
1785 			break;
1786 		case PACKET_TYPE2:
1787 			break;
1788 		case PACKET_TYPE3:
1789 			r = r600_packet3_check(p, &pkt);
1790 			break;
1791 		default:
1792 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1793 			kfree(p->track);
1794 			p->track = NULL;
1795 			return -EINVAL;
1796 		}
1797 		if (r) {
1798 			kfree(p->track);
1799 			p->track = NULL;
1800 			return r;
1801 		}
1802 	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1803 #if 0
1804 	for (r = 0; r < p->ib->length_dw; r++) {
1805 		printk(KERN_INFO "%05d  0x%08X\n", r, p->ib->ptr[r]);
1806 		mdelay(1);
1807 	}
1808 #endif
1809 	kfree(p->track);
1810 	p->track = NULL;
1811 	return 0;
1812 }
1813 
r600_cs_parser_relocs_legacy(struct radeon_cs_parser * p)1814 static int r600_cs_parser_relocs_legacy(struct radeon_cs_parser *p)
1815 {
1816 	if (p->chunk_relocs_idx == -1) {
1817 		return 0;
1818 	}
1819 	p->relocs = kzalloc(sizeof(struct radeon_cs_reloc), GFP_KERNEL);
1820 	if (p->relocs == NULL) {
1821 		return -ENOMEM;
1822 	}
1823 	return 0;
1824 }
1825 
1826 /**
1827  * cs_parser_fini() - clean parser states
1828  * @parser:	parser structure holding parsing context.
1829  * @error:	error number
1830  *
1831  * If error is set than unvalidate buffer, otherwise just free memory
1832  * used by parsing context.
1833  **/
r600_cs_parser_fini(struct radeon_cs_parser * parser,int error)1834 static void r600_cs_parser_fini(struct radeon_cs_parser *parser, int error)
1835 {
1836 	unsigned i;
1837 
1838 	kfree(parser->relocs);
1839 	for (i = 0; i < parser->nchunks; i++) {
1840 		kfree(parser->chunks[i].kdata);
1841 		kfree(parser->chunks[i].kpage[0]);
1842 		kfree(parser->chunks[i].kpage[1]);
1843 	}
1844 	kfree(parser->chunks);
1845 	kfree(parser->chunks_array);
1846 }
1847 
r600_cs_legacy(struct drm_device * dev,void * data,struct drm_file * filp,unsigned family,u32 * ib,int * l)1848 int r600_cs_legacy(struct drm_device *dev, void *data, struct drm_file *filp,
1849 			unsigned family, u32 *ib, int *l)
1850 {
1851 	struct radeon_cs_parser parser;
1852 	struct radeon_cs_chunk *ib_chunk;
1853 	struct radeon_ib fake_ib;
1854 	struct r600_cs_track *track;
1855 	int r;
1856 
1857 	/* initialize tracker */
1858 	track = kzalloc(sizeof(*track), GFP_KERNEL);
1859 	if (track == NULL)
1860 		return -ENOMEM;
1861 	r600_cs_track_init(track);
1862 	r600_cs_legacy_get_tiling_conf(dev, &track->npipes, &track->nbanks, &track->group_size);
1863 	/* initialize parser */
1864 	memset(&parser, 0, sizeof(struct radeon_cs_parser));
1865 	parser.filp = filp;
1866 	parser.dev = &dev->pdev->dev;
1867 	parser.rdev = NULL;
1868 	parser.family = family;
1869 	parser.ib = &fake_ib;
1870 	parser.track = track;
1871 	fake_ib.ptr = ib;
1872 	r = radeon_cs_parser_init(&parser, data);
1873 	if (r) {
1874 		DRM_ERROR("Failed to initialize parser !\n");
1875 		r600_cs_parser_fini(&parser, r);
1876 		return r;
1877 	}
1878 	r = r600_cs_parser_relocs_legacy(&parser);
1879 	if (r) {
1880 		DRM_ERROR("Failed to parse relocation !\n");
1881 		r600_cs_parser_fini(&parser, r);
1882 		return r;
1883 	}
1884 	/* Copy the packet into the IB, the parser will read from the
1885 	 * input memory (cached) and write to the IB (which can be
1886 	 * uncached). */
1887 	ib_chunk = &parser.chunks[parser.chunk_ib_idx];
1888 	parser.ib->length_dw = ib_chunk->length_dw;
1889 	*l = parser.ib->length_dw;
1890 	r = r600_cs_parse(&parser);
1891 	if (r) {
1892 		DRM_ERROR("Invalid command stream !\n");
1893 		r600_cs_parser_fini(&parser, r);
1894 		return r;
1895 	}
1896 	r = radeon_cs_finish_pages(&parser);
1897 	if (r) {
1898 		DRM_ERROR("Invalid command stream !\n");
1899 		r600_cs_parser_fini(&parser, r);
1900 		return r;
1901 	}
1902 	r600_cs_parser_fini(&parser, r);
1903 	return r;
1904 }
1905 
r600_cs_legacy_init(void)1906 void r600_cs_legacy_init(void)
1907 {
1908 	r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;
1909 }
1910