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]; /* unused */
56 	struct radeon_bo	*cb_color_tile_bo[8]; /* unused */
57 	u32			cb_color_info[8];
58 	u32			cb_color_view[8];
59 	u32			cb_color_size_idx[8]; /* unused */
60 	u32			cb_target_mask;
61 	u32			cb_shader_mask;  /* unused */
62 	u32			cb_color_size[8];
63 	u32			vgt_strmout_en;
64 	u32			vgt_strmout_buffer_en;
65 	struct radeon_bo	*vgt_strmout_bo[4];
66 	u64			vgt_strmout_bo_mc[4]; /* unused */
67 	u32			vgt_strmout_bo_offset[4];
68 	u32			vgt_strmout_size[4];
69 	u32			db_depth_control;
70 	u32			db_depth_info;
71 	u32			db_depth_size_idx;
72 	u32			db_depth_view;
73 	u32			db_depth_size;
74 	u32			db_offset;
75 	struct radeon_bo	*db_bo;
76 	u64			db_bo_mc;
77 	bool			sx_misc_kill_all_prims;
78 	bool			cb_dirty;
79 	bool			db_dirty;
80 	bool			streamout_dirty;
81 	struct radeon_bo	*htile_bo;
82 	u64			htile_offset;
83 	u32			htile_surface;
84 };
85 
86 #define FMT_8_BIT(fmt, vc)   [fmt] = { 1, 1, 1, vc, CHIP_R600 }
87 #define FMT_16_BIT(fmt, vc)  [fmt] = { 1, 1, 2, vc, CHIP_R600 }
88 #define FMT_24_BIT(fmt)      [fmt] = { 1, 1, 4,  0, CHIP_R600 }
89 #define FMT_32_BIT(fmt, vc)  [fmt] = { 1, 1, 4, vc, CHIP_R600 }
90 #define FMT_48_BIT(fmt)      [fmt] = { 1, 1, 8,  0, CHIP_R600 }
91 #define FMT_64_BIT(fmt, vc)  [fmt] = { 1, 1, 8, vc, CHIP_R600 }
92 #define FMT_96_BIT(fmt)      [fmt] = { 1, 1, 12, 0, CHIP_R600 }
93 #define FMT_128_BIT(fmt, vc) [fmt] = { 1, 1, 16,vc, CHIP_R600 }
94 
95 struct gpu_formats {
96 	unsigned blockwidth;
97 	unsigned blockheight;
98 	unsigned blocksize;
99 	unsigned valid_color;
100 	enum radeon_family min_family;
101 };
102 
103 static const struct gpu_formats color_formats_table[] = {
104 	/* 8 bit */
105 	FMT_8_BIT(V_038004_COLOR_8, 1),
106 	FMT_8_BIT(V_038004_COLOR_4_4, 1),
107 	FMT_8_BIT(V_038004_COLOR_3_3_2, 1),
108 	FMT_8_BIT(V_038004_FMT_1, 0),
109 
110 	/* 16-bit */
111 	FMT_16_BIT(V_038004_COLOR_16, 1),
112 	FMT_16_BIT(V_038004_COLOR_16_FLOAT, 1),
113 	FMT_16_BIT(V_038004_COLOR_8_8, 1),
114 	FMT_16_BIT(V_038004_COLOR_5_6_5, 1),
115 	FMT_16_BIT(V_038004_COLOR_6_5_5, 1),
116 	FMT_16_BIT(V_038004_COLOR_1_5_5_5, 1),
117 	FMT_16_BIT(V_038004_COLOR_4_4_4_4, 1),
118 	FMT_16_BIT(V_038004_COLOR_5_5_5_1, 1),
119 
120 	/* 24-bit */
121 	FMT_24_BIT(V_038004_FMT_8_8_8),
122 
123 	/* 32-bit */
124 	FMT_32_BIT(V_038004_COLOR_32, 1),
125 	FMT_32_BIT(V_038004_COLOR_32_FLOAT, 1),
126 	FMT_32_BIT(V_038004_COLOR_16_16, 1),
127 	FMT_32_BIT(V_038004_COLOR_16_16_FLOAT, 1),
128 	FMT_32_BIT(V_038004_COLOR_8_24, 1),
129 	FMT_32_BIT(V_038004_COLOR_8_24_FLOAT, 1),
130 	FMT_32_BIT(V_038004_COLOR_24_8, 1),
131 	FMT_32_BIT(V_038004_COLOR_24_8_FLOAT, 1),
132 	FMT_32_BIT(V_038004_COLOR_10_11_11, 1),
133 	FMT_32_BIT(V_038004_COLOR_10_11_11_FLOAT, 1),
134 	FMT_32_BIT(V_038004_COLOR_11_11_10, 1),
135 	FMT_32_BIT(V_038004_COLOR_11_11_10_FLOAT, 1),
136 	FMT_32_BIT(V_038004_COLOR_2_10_10_10, 1),
137 	FMT_32_BIT(V_038004_COLOR_8_8_8_8, 1),
138 	FMT_32_BIT(V_038004_COLOR_10_10_10_2, 1),
139 	FMT_32_BIT(V_038004_FMT_5_9_9_9_SHAREDEXP, 0),
140 	FMT_32_BIT(V_038004_FMT_32_AS_8, 0),
141 	FMT_32_BIT(V_038004_FMT_32_AS_8_8, 0),
142 
143 	/* 48-bit */
144 	FMT_48_BIT(V_038004_FMT_16_16_16),
145 	FMT_48_BIT(V_038004_FMT_16_16_16_FLOAT),
146 
147 	/* 64-bit */
148 	FMT_64_BIT(V_038004_COLOR_X24_8_32_FLOAT, 1),
149 	FMT_64_BIT(V_038004_COLOR_32_32, 1),
150 	FMT_64_BIT(V_038004_COLOR_32_32_FLOAT, 1),
151 	FMT_64_BIT(V_038004_COLOR_16_16_16_16, 1),
152 	FMT_64_BIT(V_038004_COLOR_16_16_16_16_FLOAT, 1),
153 
154 	FMT_96_BIT(V_038004_FMT_32_32_32),
155 	FMT_96_BIT(V_038004_FMT_32_32_32_FLOAT),
156 
157 	/* 128-bit */
158 	FMT_128_BIT(V_038004_COLOR_32_32_32_32, 1),
159 	FMT_128_BIT(V_038004_COLOR_32_32_32_32_FLOAT, 1),
160 
161 	[V_038004_FMT_GB_GR] = { 2, 1, 4, 0 },
162 	[V_038004_FMT_BG_RG] = { 2, 1, 4, 0 },
163 
164 	/* block compressed formats */
165 	[V_038004_FMT_BC1] = { 4, 4, 8, 0 },
166 	[V_038004_FMT_BC2] = { 4, 4, 16, 0 },
167 	[V_038004_FMT_BC3] = { 4, 4, 16, 0 },
168 	[V_038004_FMT_BC4] = { 4, 4, 8, 0 },
169 	[V_038004_FMT_BC5] = { 4, 4, 16, 0},
170 	[V_038004_FMT_BC6] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
171 	[V_038004_FMT_BC7] = { 4, 4, 16, 0, CHIP_CEDAR}, /* Evergreen-only */
172 
173 	/* The other Evergreen formats */
174 	[V_038004_FMT_32_AS_32_32_32_32] = { 1, 1, 4, 0, CHIP_CEDAR},
175 };
176 
r600_fmt_is_valid_color(u32 format)177 bool r600_fmt_is_valid_color(u32 format)
178 {
179 	if (format >= ARRAY_SIZE(color_formats_table))
180 		return false;
181 
182 	if (color_formats_table[format].valid_color)
183 		return true;
184 
185 	return false;
186 }
187 
r600_fmt_is_valid_texture(u32 format,enum radeon_family family)188 bool r600_fmt_is_valid_texture(u32 format, enum radeon_family family)
189 {
190 	if (format >= ARRAY_SIZE(color_formats_table))
191 		return false;
192 
193 	if (family < color_formats_table[format].min_family)
194 		return false;
195 
196 	if (color_formats_table[format].blockwidth > 0)
197 		return true;
198 
199 	return false;
200 }
201 
r600_fmt_get_blocksize(u32 format)202 int r600_fmt_get_blocksize(u32 format)
203 {
204 	if (format >= ARRAY_SIZE(color_formats_table))
205 		return 0;
206 
207 	return color_formats_table[format].blocksize;
208 }
209 
r600_fmt_get_nblocksx(u32 format,u32 w)210 int r600_fmt_get_nblocksx(u32 format, u32 w)
211 {
212 	unsigned bw;
213 
214 	if (format >= ARRAY_SIZE(color_formats_table))
215 		return 0;
216 
217 	bw = color_formats_table[format].blockwidth;
218 	if (bw == 0)
219 		return 0;
220 
221 	return (w + bw - 1) / bw;
222 }
223 
r600_fmt_get_nblocksy(u32 format,u32 h)224 int r600_fmt_get_nblocksy(u32 format, u32 h)
225 {
226 	unsigned bh;
227 
228 	if (format >= ARRAY_SIZE(color_formats_table))
229 		return 0;
230 
231 	bh = color_formats_table[format].blockheight;
232 	if (bh == 0)
233 		return 0;
234 
235 	return (h + bh - 1) / bh;
236 }
237 
238 struct array_mode_checker {
239 	int array_mode;
240 	u32 group_size;
241 	u32 nbanks;
242 	u32 npipes;
243 	u32 nsamples;
244 	u32 blocksize;
245 };
246 
247 /* returns alignment in pixels for pitch/height/depth and bytes for base */
r600_get_array_mode_alignment(struct array_mode_checker * values,u32 * pitch_align,u32 * height_align,u32 * depth_align,u64 * base_align)248 static int r600_get_array_mode_alignment(struct array_mode_checker *values,
249 						u32 *pitch_align,
250 						u32 *height_align,
251 						u32 *depth_align,
252 						u64 *base_align)
253 {
254 	u32 tile_width = 8;
255 	u32 tile_height = 8;
256 	u32 macro_tile_width = values->nbanks;
257 	u32 macro_tile_height = values->npipes;
258 	u32 tile_bytes = tile_width * tile_height * values->blocksize * values->nsamples;
259 	u32 macro_tile_bytes = macro_tile_width * macro_tile_height * tile_bytes;
260 
261 	switch (values->array_mode) {
262 	case ARRAY_LINEAR_GENERAL:
263 		/* technically tile_width/_height for pitch/height */
264 		*pitch_align = 1; /* tile_width */
265 		*height_align = 1; /* tile_height */
266 		*depth_align = 1;
267 		*base_align = 1;
268 		break;
269 	case ARRAY_LINEAR_ALIGNED:
270 		*pitch_align = max((u32)64, (u32)(values->group_size / values->blocksize));
271 		*height_align = 1;
272 		*depth_align = 1;
273 		*base_align = values->group_size;
274 		break;
275 	case ARRAY_1D_TILED_THIN1:
276 		*pitch_align = max((u32)tile_width,
277 				   (u32)(values->group_size /
278 					 (tile_height * values->blocksize * values->nsamples)));
279 		*height_align = tile_height;
280 		*depth_align = 1;
281 		*base_align = values->group_size;
282 		break;
283 	case ARRAY_2D_TILED_THIN1:
284 		*pitch_align = max((u32)macro_tile_width * tile_width,
285 				(u32)((values->group_size * values->nbanks) /
286 				(values->blocksize * values->nsamples * tile_width)));
287 		*height_align = macro_tile_height * tile_height;
288 		*depth_align = 1;
289 		*base_align = max(macro_tile_bytes,
290 				  (*pitch_align) * values->blocksize * (*height_align) * values->nsamples);
291 		break;
292 	default:
293 		return -EINVAL;
294 	}
295 
296 	return 0;
297 }
298 
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_view[i] = 0xFFFFFFFF;
311 		track->cb_color_bo[i] = NULL;
312 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
313 		track->cb_color_bo_mc[i] = 0xFFFFFFFF;
314 	}
315 	track->cb_target_mask = 0xFFFFFFFF;
316 	track->cb_shader_mask = 0xFFFFFFFF;
317 	track->cb_dirty = true;
318 	track->db_bo = NULL;
319 	track->db_bo_mc = 0xFFFFFFFF;
320 	/* assume the biggest format and that htile is enabled */
321 	track->db_depth_info = 7 | (1 << 25);
322 	track->db_depth_view = 0xFFFFC000;
323 	track->db_depth_size = 0xFFFFFFFF;
324 	track->db_depth_size_idx = 0;
325 	track->db_depth_control = 0xFFFFFFFF;
326 	track->db_dirty = true;
327 	track->htile_bo = NULL;
328 	track->htile_offset = 0xFFFFFFFF;
329 	track->htile_surface = 0;
330 
331 	for (i = 0; i < 4; i++) {
332 		track->vgt_strmout_size[i] = 0;
333 		track->vgt_strmout_bo[i] = NULL;
334 		track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
335 		track->vgt_strmout_bo_mc[i] = 0xFFFFFFFF;
336 	}
337 	track->streamout_dirty = true;
338 	track->sx_misc_kill_all_prims = false;
339 }
340 
r600_cs_track_validate_cb(struct radeon_cs_parser * p,int i)341 static int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
342 {
343 	struct r600_cs_track *track = p->track;
344 	u32 slice_tile_max, size, tmp;
345 	u32 height, height_align, pitch, pitch_align, depth_align;
346 	u64 base_offset, base_align;
347 	struct array_mode_checker array_check;
348 	volatile u32 *ib = p->ib->ptr;
349 	unsigned array_mode;
350 	u32 format;
351 
352 	if (G_0280A0_TILE_MODE(track->cb_color_info[i])) {
353 		dev_warn(p->dev, "FMASK or CMASK buffer are not supported by this kernel\n");
354 		return -EINVAL;
355 	}
356 	size = radeon_bo_size(track->cb_color_bo[i]) - track->cb_color_bo_offset[i];
357 	format = G_0280A0_FORMAT(track->cb_color_info[i]);
358 	if (!r600_fmt_is_valid_color(format)) {
359 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08X)\n",
360 			 __func__, __LINE__, format,
361 			i, track->cb_color_info[i]);
362 		return -EINVAL;
363 	}
364 	/* pitch in pixels */
365 	pitch = (G_028060_PITCH_TILE_MAX(track->cb_color_size[i]) + 1) * 8;
366 	slice_tile_max = G_028060_SLICE_TILE_MAX(track->cb_color_size[i]) + 1;
367 	slice_tile_max *= 64;
368 	height = slice_tile_max / pitch;
369 	if (height > 8192)
370 		height = 8192;
371 	array_mode = G_0280A0_ARRAY_MODE(track->cb_color_info[i]);
372 
373 	base_offset = track->cb_color_bo_mc[i] + track->cb_color_bo_offset[i];
374 	array_check.array_mode = array_mode;
375 	array_check.group_size = track->group_size;
376 	array_check.nbanks = track->nbanks;
377 	array_check.npipes = track->npipes;
378 	array_check.nsamples = track->nsamples;
379 	array_check.blocksize = r600_fmt_get_blocksize(format);
380 	if (r600_get_array_mode_alignment(&array_check,
381 					  &pitch_align, &height_align, &depth_align, &base_align)) {
382 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
383 			 G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
384 			 track->cb_color_info[i]);
385 		return -EINVAL;
386 	}
387 	switch (array_mode) {
388 	case V_0280A0_ARRAY_LINEAR_GENERAL:
389 		break;
390 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
391 		break;
392 	case V_0280A0_ARRAY_1D_TILED_THIN1:
393 		/* avoid breaking userspace */
394 		if (height > 7)
395 			height &= ~0x7;
396 		break;
397 	case V_0280A0_ARRAY_2D_TILED_THIN1:
398 		break;
399 	default:
400 		dev_warn(p->dev, "%s invalid tiling %d for %d (0x%08X)\n", __func__,
401 			G_0280A0_ARRAY_MODE(track->cb_color_info[i]), i,
402 			track->cb_color_info[i]);
403 		return -EINVAL;
404 	}
405 
406 	if (!IS_ALIGNED(pitch, pitch_align)) {
407 		dev_warn(p->dev, "%s:%d cb pitch (%d, 0x%x, %d) invalid\n",
408 			 __func__, __LINE__, pitch, pitch_align, array_mode);
409 		return -EINVAL;
410 	}
411 	if (!IS_ALIGNED(height, height_align)) {
412 		dev_warn(p->dev, "%s:%d cb height (%d, 0x%x, %d) invalid\n",
413 			 __func__, __LINE__, height, height_align, array_mode);
414 		return -EINVAL;
415 	}
416 	if (!IS_ALIGNED(base_offset, base_align)) {
417 		dev_warn(p->dev, "%s offset[%d] 0x%llx 0x%llx, %d not aligned\n", __func__, i,
418 			 base_offset, base_align, array_mode);
419 		return -EINVAL;
420 	}
421 
422 	/* check offset */
423 	tmp = r600_fmt_get_nblocksy(format, height) * r600_fmt_get_nblocksx(format, pitch) * r600_fmt_get_blocksize(format);
424 	switch (array_mode) {
425 	default:
426 	case V_0280A0_ARRAY_LINEAR_GENERAL:
427 	case V_0280A0_ARRAY_LINEAR_ALIGNED:
428 		tmp += track->cb_color_view[i] & 0xFF;
429 		break;
430 	case V_0280A0_ARRAY_1D_TILED_THIN1:
431 	case V_0280A0_ARRAY_2D_TILED_THIN1:
432 		tmp += G_028080_SLICE_MAX(track->cb_color_view[i]) * tmp;
433 		break;
434 	}
435 	if ((tmp + track->cb_color_bo_offset[i]) > radeon_bo_size(track->cb_color_bo[i])) {
436 		if (array_mode == V_0280A0_ARRAY_LINEAR_GENERAL) {
437 			/* the initial DDX does bad things with the CB size occasionally */
438 			/* it rounds up height too far for slice tile max but the BO is smaller */
439 			/* r600c,g also seem to flush at bad times in some apps resulting in
440 			 * bogus values here. So for linear just allow anything to avoid breaking
441 			 * broken userspace.
442 			 */
443 		} else {
444 			dev_warn(p->dev, "%s offset[%d] %d %d %d %lu too big (%d %d) (%d %d %d)\n",
445 				 __func__, i, array_mode,
446 				 track->cb_color_bo_offset[i], tmp,
447 				 radeon_bo_size(track->cb_color_bo[i]),
448 				 pitch, height, r600_fmt_get_nblocksx(format, pitch),
449 				 r600_fmt_get_nblocksy(format, height),
450 				 r600_fmt_get_blocksize(format));
451 			return -EINVAL;
452 		}
453 	}
454 	/* limit max tile */
455 	tmp = (height * pitch) >> 6;
456 	if (tmp < slice_tile_max)
457 		slice_tile_max = tmp;
458 	tmp = S_028060_PITCH_TILE_MAX((pitch / 8) - 1) |
459 		S_028060_SLICE_TILE_MAX(slice_tile_max - 1);
460 	ib[track->cb_color_size_idx[i]] = tmp;
461 	return 0;
462 }
463 
r600_cs_track_validate_db(struct radeon_cs_parser * p)464 static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
465 {
466 	struct r600_cs_track *track = p->track;
467 	u32 nviews, bpe, ntiles, size, slice_tile_max, tmp;
468 	u32 height_align, pitch_align, depth_align;
469 	u32 pitch = 8192;
470 	u32 height = 8192;
471 	u64 base_offset, base_align;
472 	struct array_mode_checker array_check;
473 	int array_mode;
474 	volatile u32 *ib = p->ib->ptr;
475 
476 
477 	if (track->db_bo == NULL) {
478 		dev_warn(p->dev, "z/stencil with no depth buffer\n");
479 		return -EINVAL;
480 	}
481 	switch (G_028010_FORMAT(track->db_depth_info)) {
482 	case V_028010_DEPTH_16:
483 		bpe = 2;
484 		break;
485 	case V_028010_DEPTH_X8_24:
486 	case V_028010_DEPTH_8_24:
487 	case V_028010_DEPTH_X8_24_FLOAT:
488 	case V_028010_DEPTH_8_24_FLOAT:
489 	case V_028010_DEPTH_32_FLOAT:
490 		bpe = 4;
491 		break;
492 	case V_028010_DEPTH_X24_8_32_FLOAT:
493 		bpe = 8;
494 		break;
495 	default:
496 		dev_warn(p->dev, "z/stencil with invalid format %d\n", G_028010_FORMAT(track->db_depth_info));
497 		return -EINVAL;
498 	}
499 	if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
500 		if (!track->db_depth_size_idx) {
501 			dev_warn(p->dev, "z/stencil buffer size not set\n");
502 			return -EINVAL;
503 		}
504 		tmp = radeon_bo_size(track->db_bo) - track->db_offset;
505 		tmp = (tmp / bpe) >> 6;
506 		if (!tmp) {
507 			dev_warn(p->dev, "z/stencil buffer too small (0x%08X %d %d %ld)\n",
508 					track->db_depth_size, bpe, track->db_offset,
509 					radeon_bo_size(track->db_bo));
510 			return -EINVAL;
511 		}
512 		ib[track->db_depth_size_idx] = S_028000_SLICE_TILE_MAX(tmp - 1) | (track->db_depth_size & 0x3FF);
513 	} else {
514 		size = radeon_bo_size(track->db_bo);
515 		/* pitch in pixels */
516 		pitch = (G_028000_PITCH_TILE_MAX(track->db_depth_size) + 1) * 8;
517 		slice_tile_max = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
518 		slice_tile_max *= 64;
519 		height = slice_tile_max / pitch;
520 		if (height > 8192)
521 			height = 8192;
522 		base_offset = track->db_bo_mc + track->db_offset;
523 		array_mode = G_028010_ARRAY_MODE(track->db_depth_info);
524 		array_check.array_mode = array_mode;
525 		array_check.group_size = track->group_size;
526 		array_check.nbanks = track->nbanks;
527 		array_check.npipes = track->npipes;
528 		array_check.nsamples = track->nsamples;
529 		array_check.blocksize = bpe;
530 		if (r600_get_array_mode_alignment(&array_check,
531 					&pitch_align, &height_align, &depth_align, &base_align)) {
532 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
533 					G_028010_ARRAY_MODE(track->db_depth_info),
534 					track->db_depth_info);
535 			return -EINVAL;
536 		}
537 		switch (array_mode) {
538 		case V_028010_ARRAY_1D_TILED_THIN1:
539 			/* don't break userspace */
540 			height &= ~0x7;
541 			break;
542 		case V_028010_ARRAY_2D_TILED_THIN1:
543 			break;
544 		default:
545 			dev_warn(p->dev, "%s invalid tiling %d (0x%08X)\n", __func__,
546 					G_028010_ARRAY_MODE(track->db_depth_info),
547 					track->db_depth_info);
548 			return -EINVAL;
549 		}
550 
551 		if (!IS_ALIGNED(pitch, pitch_align)) {
552 			dev_warn(p->dev, "%s:%d db pitch (%d, 0x%x, %d) invalid\n",
553 					__func__, __LINE__, pitch, pitch_align, array_mode);
554 			return -EINVAL;
555 		}
556 		if (!IS_ALIGNED(height, height_align)) {
557 			dev_warn(p->dev, "%s:%d db height (%d, 0x%x, %d) invalid\n",
558 					__func__, __LINE__, height, height_align, array_mode);
559 			return -EINVAL;
560 		}
561 		if (!IS_ALIGNED(base_offset, base_align)) {
562 			dev_warn(p->dev, "%s offset 0x%llx, 0x%llx, %d not aligned\n", __func__,
563 					base_offset, base_align, array_mode);
564 			return -EINVAL;
565 		}
566 
567 		ntiles = G_028000_SLICE_TILE_MAX(track->db_depth_size) + 1;
568 		nviews = G_028004_SLICE_MAX(track->db_depth_view) + 1;
569 		tmp = ntiles * bpe * 64 * nviews;
570 		if ((tmp + track->db_offset) > radeon_bo_size(track->db_bo)) {
571 			dev_warn(p->dev, "z/stencil buffer (%d) too small (0x%08X %d %d %d -> %u have %lu)\n",
572 					array_mode,
573 					track->db_depth_size, ntiles, nviews, bpe, tmp + track->db_offset,
574 					radeon_bo_size(track->db_bo));
575 			return -EINVAL;
576 		}
577 	}
578 
579 	/* hyperz */
580 	if (G_028010_TILE_SURFACE_ENABLE(track->db_depth_info)) {
581 		unsigned long size;
582 		unsigned nbx, nby;
583 
584 		if (track->htile_bo == NULL) {
585 			dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
586 				 __func__, __LINE__, track->db_depth_info);
587 			return -EINVAL;
588 		}
589 		if ((track->db_depth_size & 0xFFFFFC00) == 0xFFFFFC00) {
590 			dev_warn(p->dev, "%s:%d htile can't be enabled with bogus db_depth_size 0x%08x\n",
591 				 __func__, __LINE__, track->db_depth_size);
592 			return -EINVAL;
593 		}
594 
595 		nbx = pitch;
596 		nby = height;
597 		if (G_028D24_LINEAR(track->htile_surface)) {
598 			/* nbx must be 16 htiles aligned == 16 * 8 pixel aligned */
599 			nbx = round_up(nbx, 16 * 8);
600 			/* nby is npipes htiles aligned == npipes * 8 pixel aligned */
601 			nby = round_up(nby, track->npipes * 8);
602 		} else {
603 			/* htile widht & nby (8 or 4) make 2 bits number */
604 			tmp = track->htile_surface & 3;
605 			/* align is htile align * 8, htile align vary according to
606 			 * number of pipe and tile width and nby
607 			 */
608 			switch (track->npipes) {
609 			case 8:
610 				switch (tmp) {
611 				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
612 					nbx = round_up(nbx, 64 * 8);
613 					nby = round_up(nby, 64 * 8);
614 					break;
615 				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
616 				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
617 					nbx = round_up(nbx, 64 * 8);
618 					nby = round_up(nby, 32 * 8);
619 					break;
620 				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
621 					nbx = round_up(nbx, 32 * 8);
622 					nby = round_up(nby, 32 * 8);
623 					break;
624 				default:
625 					return -EINVAL;
626 				}
627 				break;
628 			case 4:
629 				switch (tmp) {
630 				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
631 					nbx = round_up(nbx, 64 * 8);
632 					nby = round_up(nby, 32 * 8);
633 					break;
634 				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
635 				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
636 					nbx = round_up(nbx, 32 * 8);
637 					nby = round_up(nby, 32 * 8);
638 					break;
639 				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
640 					nbx = round_up(nbx, 32 * 8);
641 					nby = round_up(nby, 16 * 8);
642 					break;
643 				default:
644 					return -EINVAL;
645 				}
646 				break;
647 			case 2:
648 				switch (tmp) {
649 				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
650 					nbx = round_up(nbx, 32 * 8);
651 					nby = round_up(nby, 32 * 8);
652 					break;
653 				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
654 				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
655 					nbx = round_up(nbx, 32 * 8);
656 					nby = round_up(nby, 16 * 8);
657 					break;
658 				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
659 					nbx = round_up(nbx, 16 * 8);
660 					nby = round_up(nby, 16 * 8);
661 					break;
662 				default:
663 					return -EINVAL;
664 				}
665 				break;
666 			case 1:
667 				switch (tmp) {
668 				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
669 					nbx = round_up(nbx, 32 * 8);
670 					nby = round_up(nby, 16 * 8);
671 					break;
672 				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
673 				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
674 					nbx = round_up(nbx, 16 * 8);
675 					nby = round_up(nby, 16 * 8);
676 					break;
677 				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
678 					nbx = round_up(nbx, 16 * 8);
679 					nby = round_up(nby, 8 * 8);
680 					break;
681 				default:
682 					return -EINVAL;
683 				}
684 				break;
685 			default:
686 				dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
687 					 __func__, __LINE__, track->npipes);
688 				return -EINVAL;
689 			}
690 		}
691 		/* compute number of htile */
692 		nbx = G_028D24_HTILE_WIDTH(track->htile_surface) ? nbx / 8 : nbx / 4;
693 		nby = G_028D24_HTILE_HEIGHT(track->htile_surface) ? nby / 8 : nby / 4;
694 		size = nbx * nby * 4;
695 		size += track->htile_offset;
696 
697 		if (size > radeon_bo_size(track->htile_bo)) {
698 			dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
699 				 __func__, __LINE__, radeon_bo_size(track->htile_bo),
700 				 size, nbx, nby);
701 			return -EINVAL;
702 		}
703 	}
704 
705 	track->db_dirty = false;
706 	return 0;
707 }
708 
r600_cs_track_check(struct radeon_cs_parser * p)709 static int r600_cs_track_check(struct radeon_cs_parser *p)
710 {
711 	struct r600_cs_track *track = p->track;
712 	u32 tmp;
713 	int r, i;
714 
715 	/* on legacy kernel we don't perform advanced check */
716 	if (p->rdev == NULL)
717 		return 0;
718 
719 	/* check streamout */
720 	if (track->streamout_dirty && track->vgt_strmout_en) {
721 		for (i = 0; i < 4; i++) {
722 			if (track->vgt_strmout_buffer_en & (1 << i)) {
723 				if (track->vgt_strmout_bo[i]) {
724 					u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
725 						(u64)track->vgt_strmout_size[i];
726 					if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
727 						DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
728 							  i, offset,
729 							  radeon_bo_size(track->vgt_strmout_bo[i]));
730 						return -EINVAL;
731 					}
732 				} else {
733 					dev_warn(p->dev, "No buffer for streamout %d\n", i);
734 					return -EINVAL;
735 				}
736 			}
737 		}
738 		track->streamout_dirty = false;
739 	}
740 
741 	if (track->sx_misc_kill_all_prims)
742 		return 0;
743 
744 	/* check that we have a cb for each enabled target, we don't check
745 	 * shader_mask because it seems mesa isn't always setting it :(
746 	 */
747 	if (track->cb_dirty) {
748 		tmp = track->cb_target_mask;
749 		for (i = 0; i < 8; i++) {
750 			u32 format = G_0280A0_FORMAT(track->cb_color_info[i]);
751 
752 			if (format != V_0280A0_COLOR_INVALID &&
753 			    (tmp >> (i * 4)) & 0xF) {
754 				/* at least one component is enabled */
755 				if (track->cb_color_bo[i] == NULL) {
756 					dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
757 						__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
758 					return -EINVAL;
759 				}
760 				/* perform rewrite of CB_COLOR[0-7]_SIZE */
761 				r = r600_cs_track_validate_cb(p, i);
762 				if (r)
763 					return r;
764 			}
765 		}
766 		track->cb_dirty = false;
767 	}
768 
769 	/* Check depth buffer */
770 	if (track->db_dirty && (G_028800_STENCIL_ENABLE(track->db_depth_control) ||
771 		G_028800_Z_ENABLE(track->db_depth_control))) {
772 		r = r600_cs_track_validate_db(p);
773 		if (r)
774 			return r;
775 	}
776 
777 	return 0;
778 }
779 
780 /**
781  * r600_cs_packet_parse() - parse cp packet and point ib index to next packet
782  * @parser:	parser structure holding parsing context.
783  * @pkt:	where to store packet informations
784  *
785  * Assume that chunk_ib_index is properly set. Will return -EINVAL
786  * if packet is bigger than remaining ib size. or if packets is unknown.
787  **/
r600_cs_packet_parse(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,unsigned idx)788 int r600_cs_packet_parse(struct radeon_cs_parser *p,
789 			struct radeon_cs_packet *pkt,
790 			unsigned idx)
791 {
792 	struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
793 	uint32_t header;
794 
795 	if (idx >= ib_chunk->length_dw) {
796 		DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
797 			  idx, ib_chunk->length_dw);
798 		return -EINVAL;
799 	}
800 	header = radeon_get_ib_value(p, idx);
801 	pkt->idx = idx;
802 	pkt->type = CP_PACKET_GET_TYPE(header);
803 	pkt->count = CP_PACKET_GET_COUNT(header);
804 	pkt->one_reg_wr = 0;
805 	switch (pkt->type) {
806 	case PACKET_TYPE0:
807 		pkt->reg = CP_PACKET0_GET_REG(header);
808 		break;
809 	case PACKET_TYPE3:
810 		pkt->opcode = CP_PACKET3_GET_OPCODE(header);
811 		break;
812 	case PACKET_TYPE2:
813 		pkt->count = -1;
814 		break;
815 	default:
816 		DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
817 		return -EINVAL;
818 	}
819 	if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
820 		DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
821 			  pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
822 		return -EINVAL;
823 	}
824 	return 0;
825 }
826 
827 /**
828  * r600_cs_packet_next_reloc_mm() - parse next packet which should be reloc packet3
829  * @parser:		parser structure holding parsing context.
830  * @data:		pointer to relocation data
831  * @offset_start:	starting offset
832  * @offset_mask:	offset mask (to align start offset on)
833  * @reloc:		reloc informations
834  *
835  * Check next packet is relocation packet3, do bo validation and compute
836  * GPU offset using the provided start.
837  **/
r600_cs_packet_next_reloc_mm(struct radeon_cs_parser * p,struct radeon_cs_reloc ** cs_reloc)838 static int r600_cs_packet_next_reloc_mm(struct radeon_cs_parser *p,
839 					struct radeon_cs_reloc **cs_reloc)
840 {
841 	struct radeon_cs_chunk *relocs_chunk;
842 	struct radeon_cs_packet p3reloc;
843 	unsigned idx;
844 	int r;
845 
846 	if (p->chunk_relocs_idx == -1) {
847 		DRM_ERROR("No relocation chunk !\n");
848 		return -EINVAL;
849 	}
850 	*cs_reloc = NULL;
851 	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
852 	r = r600_cs_packet_parse(p, &p3reloc, p->idx);
853 	if (r) {
854 		return r;
855 	}
856 	p->idx += p3reloc.count + 2;
857 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
858 		DRM_ERROR("No packet3 for relocation for packet at %d.\n",
859 			  p3reloc.idx);
860 		return -EINVAL;
861 	}
862 	idx = radeon_get_ib_value(p, p3reloc.idx + 1);
863 	if (idx >= relocs_chunk->length_dw) {
864 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
865 			  idx, relocs_chunk->length_dw);
866 		return -EINVAL;
867 	}
868 	/* FIXME: we assume reloc size is 4 dwords */
869 	*cs_reloc = p->relocs_ptr[(idx / 4)];
870 	return 0;
871 }
872 
873 /**
874  * r600_cs_packet_next_reloc_nomm() - parse next packet which should be reloc packet3
875  * @parser:		parser structure holding parsing context.
876  * @data:		pointer to relocation data
877  * @offset_start:	starting offset
878  * @offset_mask:	offset mask (to align start offset on)
879  * @reloc:		reloc informations
880  *
881  * Check next packet is relocation packet3, do bo validation and compute
882  * GPU offset using the provided start.
883  **/
r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser * p,struct radeon_cs_reloc ** cs_reloc)884 static int r600_cs_packet_next_reloc_nomm(struct radeon_cs_parser *p,
885 					struct radeon_cs_reloc **cs_reloc)
886 {
887 	struct radeon_cs_chunk *relocs_chunk;
888 	struct radeon_cs_packet p3reloc;
889 	unsigned idx;
890 	int r;
891 
892 	if (p->chunk_relocs_idx == -1) {
893 		DRM_ERROR("No relocation chunk !\n");
894 		return -EINVAL;
895 	}
896 	*cs_reloc = NULL;
897 	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
898 	r = r600_cs_packet_parse(p, &p3reloc, p->idx);
899 	if (r) {
900 		return r;
901 	}
902 	p->idx += p3reloc.count + 2;
903 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
904 		DRM_ERROR("No packet3 for relocation for packet at %d.\n",
905 			  p3reloc.idx);
906 		return -EINVAL;
907 	}
908 	idx = radeon_get_ib_value(p, p3reloc.idx + 1);
909 	if (idx >= relocs_chunk->length_dw) {
910 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
911 			  idx, relocs_chunk->length_dw);
912 		return -EINVAL;
913 	}
914 	*cs_reloc = p->relocs;
915 	(*cs_reloc)->lobj.gpu_offset = (u64)relocs_chunk->kdata[idx + 3] << 32;
916 	(*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
917 	return 0;
918 }
919 
920 /**
921  * r600_cs_packet_next_is_pkt3_nop() - test if next packet is packet3 nop for reloc
922  * @parser:		parser structure holding parsing context.
923  *
924  * Check next packet is relocation packet3, do bo validation and compute
925  * GPU offset using the provided start.
926  **/
r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser * p)927 static int r600_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
928 {
929 	struct radeon_cs_packet p3reloc;
930 	int r;
931 
932 	r = r600_cs_packet_parse(p, &p3reloc, p->idx);
933 	if (r) {
934 		return 0;
935 	}
936 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
937 		return 0;
938 	}
939 	return 1;
940 }
941 
942 /**
943  * r600_cs_packet_next_vline() - parse userspace VLINE packet
944  * @parser:		parser structure holding parsing context.
945  *
946  * Userspace sends a special sequence for VLINE waits.
947  * PACKET0 - VLINE_START_END + value
948  * PACKET3 - WAIT_REG_MEM poll vline status reg
949  * RELOC (P3) - crtc_id in reloc.
950  *
951  * This function parses this and relocates the VLINE START END
952  * and WAIT_REG_MEM packets to the correct crtc.
953  * It also detects a switched off crtc and nulls out the
954  * wait in that case.
955  */
r600_cs_packet_parse_vline(struct radeon_cs_parser * p)956 static int r600_cs_packet_parse_vline(struct radeon_cs_parser *p)
957 {
958 	struct drm_mode_object *obj;
959 	struct drm_crtc *crtc;
960 	struct radeon_crtc *radeon_crtc;
961 	struct radeon_cs_packet p3reloc, wait_reg_mem;
962 	int crtc_id;
963 	int r;
964 	uint32_t header, h_idx, reg, wait_reg_mem_info;
965 	volatile uint32_t *ib;
966 
967 	ib = p->ib->ptr;
968 
969 	/* parse the WAIT_REG_MEM */
970 	r = r600_cs_packet_parse(p, &wait_reg_mem, p->idx);
971 	if (r)
972 		return r;
973 
974 	/* check its a WAIT_REG_MEM */
975 	if (wait_reg_mem.type != PACKET_TYPE3 ||
976 	    wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
977 		DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
978 		return -EINVAL;
979 	}
980 
981 	wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
982 	/* bit 4 is reg (0) or mem (1) */
983 	if (wait_reg_mem_info & 0x10) {
984 		DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
985 		return -EINVAL;
986 	}
987 	/* waiting for value to be equal */
988 	if ((wait_reg_mem_info & 0x7) != 0x3) {
989 		DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
990 		return -EINVAL;
991 	}
992 	if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != AVIVO_D1MODE_VLINE_STATUS) {
993 		DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
994 		return -EINVAL;
995 	}
996 
997 	if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != AVIVO_D1MODE_VLINE_STAT) {
998 		DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
999 		return -EINVAL;
1000 	}
1001 
1002 	/* jump over the NOP */
1003 	r = r600_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
1004 	if (r)
1005 		return r;
1006 
1007 	h_idx = p->idx - 2;
1008 	p->idx += wait_reg_mem.count + 2;
1009 	p->idx += p3reloc.count + 2;
1010 
1011 	header = radeon_get_ib_value(p, h_idx);
1012 	crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
1013 	reg = CP_PACKET0_GET_REG(header);
1014 
1015 	obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1016 	if (!obj) {
1017 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1018 		return -EINVAL;
1019 	}
1020 	crtc = obj_to_crtc(obj);
1021 	radeon_crtc = to_radeon_crtc(crtc);
1022 	crtc_id = radeon_crtc->crtc_id;
1023 
1024 	if (!crtc->enabled) {
1025 		/* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
1026 		ib[h_idx + 2] = PACKET2(0);
1027 		ib[h_idx + 3] = PACKET2(0);
1028 		ib[h_idx + 4] = PACKET2(0);
1029 		ib[h_idx + 5] = PACKET2(0);
1030 		ib[h_idx + 6] = PACKET2(0);
1031 		ib[h_idx + 7] = PACKET2(0);
1032 		ib[h_idx + 8] = PACKET2(0);
1033 	} else if (crtc_id == 1) {
1034 		switch (reg) {
1035 		case AVIVO_D1MODE_VLINE_START_END:
1036 			header &= ~R600_CP_PACKET0_REG_MASK;
1037 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1038 			break;
1039 		default:
1040 			DRM_ERROR("unknown crtc reloc\n");
1041 			return -EINVAL;
1042 		}
1043 		ib[h_idx] = header;
1044 		ib[h_idx + 4] = AVIVO_D2MODE_VLINE_STATUS >> 2;
1045 	}
1046 
1047 	return 0;
1048 }
1049 
r600_packet0_check(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,unsigned idx,unsigned reg)1050 static int r600_packet0_check(struct radeon_cs_parser *p,
1051 				struct radeon_cs_packet *pkt,
1052 				unsigned idx, unsigned reg)
1053 {
1054 	int r;
1055 
1056 	switch (reg) {
1057 	case AVIVO_D1MODE_VLINE_START_END:
1058 		r = r600_cs_packet_parse_vline(p);
1059 		if (r) {
1060 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1061 					idx, reg);
1062 			return r;
1063 		}
1064 		break;
1065 	default:
1066 		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1067 		       reg, idx);
1068 		return -EINVAL;
1069 	}
1070 	return 0;
1071 }
1072 
r600_cs_parse_packet0(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt)1073 static int r600_cs_parse_packet0(struct radeon_cs_parser *p,
1074 				struct radeon_cs_packet *pkt)
1075 {
1076 	unsigned reg, i;
1077 	unsigned idx;
1078 	int r;
1079 
1080 	idx = pkt->idx + 1;
1081 	reg = pkt->reg;
1082 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
1083 		r = r600_packet0_check(p, pkt, idx, reg);
1084 		if (r) {
1085 			return r;
1086 		}
1087 	}
1088 	return 0;
1089 }
1090 
1091 /**
1092  * r600_cs_check_reg() - check if register is authorized or not
1093  * @parser: parser structure holding parsing context
1094  * @reg: register we are testing
1095  * @idx: index into the cs buffer
1096  *
1097  * This function will test against r600_reg_safe_bm and return 0
1098  * if register is safe. If register is not flag as safe this function
1099  * will test it against a list of register needind special handling.
1100  */
r600_cs_check_reg(struct radeon_cs_parser * p,u32 reg,u32 idx)1101 static int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1102 {
1103 	struct r600_cs_track *track = (struct r600_cs_track *)p->track;
1104 	struct radeon_cs_reloc *reloc;
1105 	u32 m, i, tmp, *ib;
1106 	int r;
1107 
1108 	i = (reg >> 7);
1109 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
1110 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1111 		return -EINVAL;
1112 	}
1113 	m = 1 << ((reg >> 2) & 31);
1114 	if (!(r600_reg_safe_bm[i] & m))
1115 		return 0;
1116 	ib = p->ib->ptr;
1117 	switch (reg) {
1118 	/* force following reg to 0 in an attempt to disable out buffer
1119 	 * which will need us to better understand how it works to perform
1120 	 * security check on it (Jerome)
1121 	 */
1122 	case R_0288A8_SQ_ESGS_RING_ITEMSIZE:
1123 	case R_008C44_SQ_ESGS_RING_SIZE:
1124 	case R_0288B0_SQ_ESTMP_RING_ITEMSIZE:
1125 	case R_008C54_SQ_ESTMP_RING_SIZE:
1126 	case R_0288C0_SQ_FBUF_RING_ITEMSIZE:
1127 	case R_008C74_SQ_FBUF_RING_SIZE:
1128 	case R_0288B4_SQ_GSTMP_RING_ITEMSIZE:
1129 	case R_008C5C_SQ_GSTMP_RING_SIZE:
1130 	case R_0288AC_SQ_GSVS_RING_ITEMSIZE:
1131 	case R_008C4C_SQ_GSVS_RING_SIZE:
1132 	case R_0288BC_SQ_PSTMP_RING_ITEMSIZE:
1133 	case R_008C6C_SQ_PSTMP_RING_SIZE:
1134 	case R_0288C4_SQ_REDUC_RING_ITEMSIZE:
1135 	case R_008C7C_SQ_REDUC_RING_SIZE:
1136 	case R_0288B8_SQ_VSTMP_RING_ITEMSIZE:
1137 	case R_008C64_SQ_VSTMP_RING_SIZE:
1138 	case R_0288C8_SQ_GS_VERT_ITEMSIZE:
1139 		/* get value to populate the IB don't remove */
1140 		tmp =radeon_get_ib_value(p, idx);
1141 		ib[idx] = 0;
1142 		break;
1143 	case SQ_CONFIG:
1144 		track->sq_config = radeon_get_ib_value(p, idx);
1145 		break;
1146 	case R_028800_DB_DEPTH_CONTROL:
1147 		track->db_depth_control = radeon_get_ib_value(p, idx);
1148 		track->db_dirty = true;
1149 		break;
1150 	case R_028010_DB_DEPTH_INFO:
1151 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1152 		    r600_cs_packet_next_is_pkt3_nop(p)) {
1153 			r = r600_cs_packet_next_reloc(p, &reloc);
1154 			if (r) {
1155 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1156 					 "0x%04X\n", reg);
1157 				return -EINVAL;
1158 			}
1159 			track->db_depth_info = radeon_get_ib_value(p, idx);
1160 			ib[idx] &= C_028010_ARRAY_MODE;
1161 			track->db_depth_info &= C_028010_ARRAY_MODE;
1162 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1163 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1164 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_2D_TILED_THIN1);
1165 			} else {
1166 				ib[idx] |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1167 				track->db_depth_info |= S_028010_ARRAY_MODE(V_028010_ARRAY_1D_TILED_THIN1);
1168 			}
1169 		} else {
1170 			track->db_depth_info = radeon_get_ib_value(p, idx);
1171 		}
1172 		track->db_dirty = true;
1173 		break;
1174 	case R_028004_DB_DEPTH_VIEW:
1175 		track->db_depth_view = radeon_get_ib_value(p, idx);
1176 		track->db_dirty = true;
1177 		break;
1178 	case R_028000_DB_DEPTH_SIZE:
1179 		track->db_depth_size = radeon_get_ib_value(p, idx);
1180 		track->db_depth_size_idx = idx;
1181 		track->db_dirty = true;
1182 		break;
1183 	case R_028AB0_VGT_STRMOUT_EN:
1184 		track->vgt_strmout_en = radeon_get_ib_value(p, idx);
1185 		track->streamout_dirty = true;
1186 		break;
1187 	case R_028B20_VGT_STRMOUT_BUFFER_EN:
1188 		track->vgt_strmout_buffer_en = radeon_get_ib_value(p, idx);
1189 		track->streamout_dirty = true;
1190 		break;
1191 	case VGT_STRMOUT_BUFFER_BASE_0:
1192 	case VGT_STRMOUT_BUFFER_BASE_1:
1193 	case VGT_STRMOUT_BUFFER_BASE_2:
1194 	case VGT_STRMOUT_BUFFER_BASE_3:
1195 		r = r600_cs_packet_next_reloc(p, &reloc);
1196 		if (r) {
1197 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1198 					"0x%04X\n", reg);
1199 			return -EINVAL;
1200 		}
1201 		tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1202 		track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1203 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1204 		track->vgt_strmout_bo[tmp] = reloc->robj;
1205 		track->vgt_strmout_bo_mc[tmp] = reloc->lobj.gpu_offset;
1206 		track->streamout_dirty = true;
1207 		break;
1208 	case VGT_STRMOUT_BUFFER_SIZE_0:
1209 	case VGT_STRMOUT_BUFFER_SIZE_1:
1210 	case VGT_STRMOUT_BUFFER_SIZE_2:
1211 	case VGT_STRMOUT_BUFFER_SIZE_3:
1212 		tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1213 		/* size in register is DWs, convert to bytes */
1214 		track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1215 		track->streamout_dirty = true;
1216 		break;
1217 	case CP_COHER_BASE:
1218 		r = r600_cs_packet_next_reloc(p, &reloc);
1219 		if (r) {
1220 			dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1221 					"0x%04X\n", reg);
1222 			return -EINVAL;
1223 		}
1224 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1225 		break;
1226 	case R_028238_CB_TARGET_MASK:
1227 		track->cb_target_mask = radeon_get_ib_value(p, idx);
1228 		track->cb_dirty = true;
1229 		break;
1230 	case R_02823C_CB_SHADER_MASK:
1231 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
1232 		break;
1233 	case R_028C04_PA_SC_AA_CONFIG:
1234 		tmp = G_028C04_MSAA_NUM_SAMPLES(radeon_get_ib_value(p, idx));
1235 		track->nsamples = 1 << tmp;
1236 		track->cb_dirty = true;
1237 		break;
1238 	case R_0280A0_CB_COLOR0_INFO:
1239 	case R_0280A4_CB_COLOR1_INFO:
1240 	case R_0280A8_CB_COLOR2_INFO:
1241 	case R_0280AC_CB_COLOR3_INFO:
1242 	case R_0280B0_CB_COLOR4_INFO:
1243 	case R_0280B4_CB_COLOR5_INFO:
1244 	case R_0280B8_CB_COLOR6_INFO:
1245 	case R_0280BC_CB_COLOR7_INFO:
1246 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) &&
1247 		     r600_cs_packet_next_is_pkt3_nop(p)) {
1248 			r = r600_cs_packet_next_reloc(p, &reloc);
1249 			if (r) {
1250 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1251 				return -EINVAL;
1252 			}
1253 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1254 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1255 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
1256 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1257 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_2D_TILED_THIN1);
1258 			} else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
1259 				ib[idx] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1260 				track->cb_color_info[tmp] |= S_0280A0_ARRAY_MODE(V_0280A0_ARRAY_1D_TILED_THIN1);
1261 			}
1262 		} else {
1263 			tmp = (reg - R_0280A0_CB_COLOR0_INFO) / 4;
1264 			track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1265 		}
1266 		track->cb_dirty = true;
1267 		break;
1268 	case R_028080_CB_COLOR0_VIEW:
1269 	case R_028084_CB_COLOR1_VIEW:
1270 	case R_028088_CB_COLOR2_VIEW:
1271 	case R_02808C_CB_COLOR3_VIEW:
1272 	case R_028090_CB_COLOR4_VIEW:
1273 	case R_028094_CB_COLOR5_VIEW:
1274 	case R_028098_CB_COLOR6_VIEW:
1275 	case R_02809C_CB_COLOR7_VIEW:
1276 		tmp = (reg - R_028080_CB_COLOR0_VIEW) / 4;
1277 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1278 		track->cb_dirty = true;
1279 		break;
1280 	case R_028060_CB_COLOR0_SIZE:
1281 	case R_028064_CB_COLOR1_SIZE:
1282 	case R_028068_CB_COLOR2_SIZE:
1283 	case R_02806C_CB_COLOR3_SIZE:
1284 	case R_028070_CB_COLOR4_SIZE:
1285 	case R_028074_CB_COLOR5_SIZE:
1286 	case R_028078_CB_COLOR6_SIZE:
1287 	case R_02807C_CB_COLOR7_SIZE:
1288 		tmp = (reg - R_028060_CB_COLOR0_SIZE) / 4;
1289 		track->cb_color_size[tmp] = radeon_get_ib_value(p, idx);
1290 		track->cb_color_size_idx[tmp] = idx;
1291 		track->cb_dirty = true;
1292 		break;
1293 		/* This register were added late, there is userspace
1294 		 * which does provide relocation for those but set
1295 		 * 0 offset. In order to avoid breaking old userspace
1296 		 * we detect this and set address to point to last
1297 		 * CB_COLOR0_BASE, note that if userspace doesn't set
1298 		 * CB_COLOR0_BASE before this register we will report
1299 		 * error. Old userspace always set CB_COLOR0_BASE
1300 		 * before any of this.
1301 		 */
1302 	case R_0280E0_CB_COLOR0_FRAG:
1303 	case R_0280E4_CB_COLOR1_FRAG:
1304 	case R_0280E8_CB_COLOR2_FRAG:
1305 	case R_0280EC_CB_COLOR3_FRAG:
1306 	case R_0280F0_CB_COLOR4_FRAG:
1307 	case R_0280F4_CB_COLOR5_FRAG:
1308 	case R_0280F8_CB_COLOR6_FRAG:
1309 	case R_0280FC_CB_COLOR7_FRAG:
1310 		tmp = (reg - R_0280E0_CB_COLOR0_FRAG) / 4;
1311 		if (!r600_cs_packet_next_is_pkt3_nop(p)) {
1312 			if (!track->cb_color_base_last[tmp]) {
1313 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1314 				return -EINVAL;
1315 			}
1316 			ib[idx] = track->cb_color_base_last[tmp];
1317 			track->cb_color_frag_bo[tmp] = track->cb_color_bo[tmp];
1318 		} else {
1319 			r = r600_cs_packet_next_reloc(p, &reloc);
1320 			if (r) {
1321 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1322 				return -EINVAL;
1323 			}
1324 			ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1325 			track->cb_color_frag_bo[tmp] = reloc->robj;
1326 		}
1327 		break;
1328 	case R_0280C0_CB_COLOR0_TILE:
1329 	case R_0280C4_CB_COLOR1_TILE:
1330 	case R_0280C8_CB_COLOR2_TILE:
1331 	case R_0280CC_CB_COLOR3_TILE:
1332 	case R_0280D0_CB_COLOR4_TILE:
1333 	case R_0280D4_CB_COLOR5_TILE:
1334 	case R_0280D8_CB_COLOR6_TILE:
1335 	case R_0280DC_CB_COLOR7_TILE:
1336 		tmp = (reg - R_0280C0_CB_COLOR0_TILE) / 4;
1337 		if (!r600_cs_packet_next_is_pkt3_nop(p)) {
1338 			if (!track->cb_color_base_last[tmp]) {
1339 				dev_err(p->dev, "Broken old userspace ? no cb_color0_base supplied before trying to write 0x%08X\n", reg);
1340 				return -EINVAL;
1341 			}
1342 			ib[idx] = track->cb_color_base_last[tmp];
1343 			track->cb_color_tile_bo[tmp] = track->cb_color_bo[tmp];
1344 		} else {
1345 			r = r600_cs_packet_next_reloc(p, &reloc);
1346 			if (r) {
1347 				dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1348 				return -EINVAL;
1349 			}
1350 			ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1351 			track->cb_color_tile_bo[tmp] = reloc->robj;
1352 		}
1353 		break;
1354 	case CB_COLOR0_BASE:
1355 	case CB_COLOR1_BASE:
1356 	case CB_COLOR2_BASE:
1357 	case CB_COLOR3_BASE:
1358 	case CB_COLOR4_BASE:
1359 	case CB_COLOR5_BASE:
1360 	case CB_COLOR6_BASE:
1361 	case CB_COLOR7_BASE:
1362 		r = r600_cs_packet_next_reloc(p, &reloc);
1363 		if (r) {
1364 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1365 					"0x%04X\n", reg);
1366 			return -EINVAL;
1367 		}
1368 		tmp = (reg - CB_COLOR0_BASE) / 4;
1369 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1370 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1371 		track->cb_color_base_last[tmp] = ib[idx];
1372 		track->cb_color_bo[tmp] = reloc->robj;
1373 		track->cb_color_bo_mc[tmp] = reloc->lobj.gpu_offset;
1374 		track->cb_dirty = true;
1375 		break;
1376 	case DB_DEPTH_BASE:
1377 		r = r600_cs_packet_next_reloc(p, &reloc);
1378 		if (r) {
1379 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1380 					"0x%04X\n", reg);
1381 			return -EINVAL;
1382 		}
1383 		track->db_offset = radeon_get_ib_value(p, idx) << 8;
1384 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1385 		track->db_bo = reloc->robj;
1386 		track->db_bo_mc = reloc->lobj.gpu_offset;
1387 		track->db_dirty = true;
1388 		break;
1389 	case DB_HTILE_DATA_BASE:
1390 		r = r600_cs_packet_next_reloc(p, &reloc);
1391 		if (r) {
1392 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1393 					"0x%04X\n", reg);
1394 			return -EINVAL;
1395 		}
1396 		track->htile_offset = radeon_get_ib_value(p, idx) << 8;
1397 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1398 		track->htile_bo = reloc->robj;
1399 		track->db_dirty = true;
1400 		break;
1401 	case DB_HTILE_SURFACE:
1402 		track->htile_surface = radeon_get_ib_value(p, idx);
1403 		track->db_dirty = true;
1404 		break;
1405 	case SQ_PGM_START_FS:
1406 	case SQ_PGM_START_ES:
1407 	case SQ_PGM_START_VS:
1408 	case SQ_PGM_START_GS:
1409 	case SQ_PGM_START_PS:
1410 	case SQ_ALU_CONST_CACHE_GS_0:
1411 	case SQ_ALU_CONST_CACHE_GS_1:
1412 	case SQ_ALU_CONST_CACHE_GS_2:
1413 	case SQ_ALU_CONST_CACHE_GS_3:
1414 	case SQ_ALU_CONST_CACHE_GS_4:
1415 	case SQ_ALU_CONST_CACHE_GS_5:
1416 	case SQ_ALU_CONST_CACHE_GS_6:
1417 	case SQ_ALU_CONST_CACHE_GS_7:
1418 	case SQ_ALU_CONST_CACHE_GS_8:
1419 	case SQ_ALU_CONST_CACHE_GS_9:
1420 	case SQ_ALU_CONST_CACHE_GS_10:
1421 	case SQ_ALU_CONST_CACHE_GS_11:
1422 	case SQ_ALU_CONST_CACHE_GS_12:
1423 	case SQ_ALU_CONST_CACHE_GS_13:
1424 	case SQ_ALU_CONST_CACHE_GS_14:
1425 	case SQ_ALU_CONST_CACHE_GS_15:
1426 	case SQ_ALU_CONST_CACHE_PS_0:
1427 	case SQ_ALU_CONST_CACHE_PS_1:
1428 	case SQ_ALU_CONST_CACHE_PS_2:
1429 	case SQ_ALU_CONST_CACHE_PS_3:
1430 	case SQ_ALU_CONST_CACHE_PS_4:
1431 	case SQ_ALU_CONST_CACHE_PS_5:
1432 	case SQ_ALU_CONST_CACHE_PS_6:
1433 	case SQ_ALU_CONST_CACHE_PS_7:
1434 	case SQ_ALU_CONST_CACHE_PS_8:
1435 	case SQ_ALU_CONST_CACHE_PS_9:
1436 	case SQ_ALU_CONST_CACHE_PS_10:
1437 	case SQ_ALU_CONST_CACHE_PS_11:
1438 	case SQ_ALU_CONST_CACHE_PS_12:
1439 	case SQ_ALU_CONST_CACHE_PS_13:
1440 	case SQ_ALU_CONST_CACHE_PS_14:
1441 	case SQ_ALU_CONST_CACHE_PS_15:
1442 	case SQ_ALU_CONST_CACHE_VS_0:
1443 	case SQ_ALU_CONST_CACHE_VS_1:
1444 	case SQ_ALU_CONST_CACHE_VS_2:
1445 	case SQ_ALU_CONST_CACHE_VS_3:
1446 	case SQ_ALU_CONST_CACHE_VS_4:
1447 	case SQ_ALU_CONST_CACHE_VS_5:
1448 	case SQ_ALU_CONST_CACHE_VS_6:
1449 	case SQ_ALU_CONST_CACHE_VS_7:
1450 	case SQ_ALU_CONST_CACHE_VS_8:
1451 	case SQ_ALU_CONST_CACHE_VS_9:
1452 	case SQ_ALU_CONST_CACHE_VS_10:
1453 	case SQ_ALU_CONST_CACHE_VS_11:
1454 	case SQ_ALU_CONST_CACHE_VS_12:
1455 	case SQ_ALU_CONST_CACHE_VS_13:
1456 	case SQ_ALU_CONST_CACHE_VS_14:
1457 	case SQ_ALU_CONST_CACHE_VS_15:
1458 		r = r600_cs_packet_next_reloc(p, &reloc);
1459 		if (r) {
1460 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1461 					"0x%04X\n", reg);
1462 			return -EINVAL;
1463 		}
1464 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1465 		break;
1466 	case SX_MEMORY_EXPORT_BASE:
1467 		r = r600_cs_packet_next_reloc(p, &reloc);
1468 		if (r) {
1469 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1470 					"0x%04X\n", reg);
1471 			return -EINVAL;
1472 		}
1473 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1474 		break;
1475 	case SX_MISC:
1476 		track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1477 		break;
1478 	default:
1479 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1480 		return -EINVAL;
1481 	}
1482 	return 0;
1483 }
1484 
r600_mip_minify(unsigned size,unsigned level)1485 unsigned r600_mip_minify(unsigned size, unsigned level)
1486 {
1487 	unsigned val;
1488 
1489 	val = max(1U, size >> level);
1490 	if (level > 0)
1491 		val = roundup_pow_of_two(val);
1492 	return val;
1493 }
1494 
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)1495 static void r600_texture_size(unsigned nfaces, unsigned blevel, unsigned llevel,
1496 			      unsigned w0, unsigned h0, unsigned d0, unsigned format,
1497 			      unsigned block_align, unsigned height_align, unsigned base_align,
1498 			      unsigned *l0_size, unsigned *mipmap_size)
1499 {
1500 	unsigned offset, i, level;
1501 	unsigned width, height, depth, size;
1502 	unsigned blocksize;
1503 	unsigned nbx, nby;
1504 	unsigned nlevels = llevel - blevel + 1;
1505 
1506 	*l0_size = -1;
1507 	blocksize = r600_fmt_get_blocksize(format);
1508 
1509 	w0 = r600_mip_minify(w0, 0);
1510 	h0 = r600_mip_minify(h0, 0);
1511 	d0 = r600_mip_minify(d0, 0);
1512 	for(i = 0, offset = 0, level = blevel; i < nlevels; i++, level++) {
1513 		width = r600_mip_minify(w0, i);
1514 		nbx = r600_fmt_get_nblocksx(format, width);
1515 
1516 		nbx = round_up(nbx, block_align);
1517 
1518 		height = r600_mip_minify(h0, i);
1519 		nby = r600_fmt_get_nblocksy(format, height);
1520 		nby = round_up(nby, height_align);
1521 
1522 		depth = r600_mip_minify(d0, i);
1523 
1524 		size = nbx * nby * blocksize;
1525 		if (nfaces)
1526 			size *= nfaces;
1527 		else
1528 			size *= depth;
1529 
1530 		if (i == 0)
1531 			*l0_size = size;
1532 
1533 		if (i == 0 || i == 1)
1534 			offset = round_up(offset, base_align);
1535 
1536 		offset += size;
1537 	}
1538 	*mipmap_size = offset;
1539 	if (llevel == 0)
1540 		*mipmap_size = *l0_size;
1541 	if (!blevel)
1542 		*mipmap_size -= *l0_size;
1543 }
1544 
1545 /**
1546  * r600_check_texture_resource() - check if register is authorized or not
1547  * @p: parser structure holding parsing context
1548  * @idx: index into the cs buffer
1549  * @texture: texture's bo structure
1550  * @mipmap: mipmap's bo structure
1551  *
1552  * This function will check that the resource has valid field and that
1553  * the texture and mipmap bo object are big enough to cover this resource.
1554  */
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)1555 static int r600_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
1556 					      struct radeon_bo *texture,
1557 					      struct radeon_bo *mipmap,
1558 					      u64 base_offset,
1559 					      u64 mip_offset,
1560 					      u32 tiling_flags)
1561 {
1562 	struct r600_cs_track *track = p->track;
1563 	u32 nfaces, llevel, blevel, w0, h0, d0;
1564 	u32 word0, word1, l0_size, mipmap_size, word2, word3;
1565 	u32 height_align, pitch, pitch_align, depth_align;
1566 	u32 array, barray, larray;
1567 	u64 base_align;
1568 	struct array_mode_checker array_check;
1569 	u32 format;
1570 
1571 	/* on legacy kernel we don't perform advanced check */
1572 	if (p->rdev == NULL)
1573 		return 0;
1574 
1575 	/* convert to bytes */
1576 	base_offset <<= 8;
1577 	mip_offset <<= 8;
1578 
1579 	word0 = radeon_get_ib_value(p, idx + 0);
1580 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1581 		if (tiling_flags & RADEON_TILING_MACRO)
1582 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1583 		else if (tiling_flags & RADEON_TILING_MICRO)
1584 			word0 |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1585 	}
1586 	word1 = radeon_get_ib_value(p, idx + 1);
1587 	w0 = G_038000_TEX_WIDTH(word0) + 1;
1588 	h0 = G_038004_TEX_HEIGHT(word1) + 1;
1589 	d0 = G_038004_TEX_DEPTH(word1);
1590 	nfaces = 1;
1591 	array = 0;
1592 	switch (G_038000_DIM(word0)) {
1593 	case V_038000_SQ_TEX_DIM_1D:
1594 	case V_038000_SQ_TEX_DIM_2D:
1595 	case V_038000_SQ_TEX_DIM_3D:
1596 		break;
1597 	case V_038000_SQ_TEX_DIM_CUBEMAP:
1598 		if (p->family >= CHIP_RV770)
1599 			nfaces = 8;
1600 		else
1601 			nfaces = 6;
1602 		break;
1603 	case V_038000_SQ_TEX_DIM_1D_ARRAY:
1604 	case V_038000_SQ_TEX_DIM_2D_ARRAY:
1605 		array = 1;
1606 		break;
1607 	case V_038000_SQ_TEX_DIM_2D_MSAA:
1608 	case V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA:
1609 	default:
1610 		dev_warn(p->dev, "this kernel doesn't support %d texture dim\n", G_038000_DIM(word0));
1611 		return -EINVAL;
1612 	}
1613 	format = G_038004_DATA_FORMAT(word1);
1614 	if (!r600_fmt_is_valid_texture(format, p->family)) {
1615 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
1616 			 __func__, __LINE__, format);
1617 		return -EINVAL;
1618 	}
1619 
1620 	/* pitch in texels */
1621 	pitch = (G_038000_PITCH(word0) + 1) * 8;
1622 	array_check.array_mode = G_038000_TILE_MODE(word0);
1623 	array_check.group_size = track->group_size;
1624 	array_check.nbanks = track->nbanks;
1625 	array_check.npipes = track->npipes;
1626 	array_check.nsamples = 1;
1627 	array_check.blocksize = r600_fmt_get_blocksize(format);
1628 	if (r600_get_array_mode_alignment(&array_check,
1629 					  &pitch_align, &height_align, &depth_align, &base_align)) {
1630 		dev_warn(p->dev, "%s:%d tex array mode (%d) invalid\n",
1631 			 __func__, __LINE__, G_038000_TILE_MODE(word0));
1632 		return -EINVAL;
1633 	}
1634 
1635 	/* XXX check height as well... */
1636 
1637 	if (!IS_ALIGNED(pitch, pitch_align)) {
1638 		dev_warn(p->dev, "%s:%d tex pitch (%d, 0x%x, %d) invalid\n",
1639 			 __func__, __LINE__, pitch, pitch_align, G_038000_TILE_MODE(word0));
1640 		return -EINVAL;
1641 	}
1642 	if (!IS_ALIGNED(base_offset, base_align)) {
1643 		dev_warn(p->dev, "%s:%d tex base offset (0x%llx, 0x%llx, %d) invalid\n",
1644 			 __func__, __LINE__, base_offset, base_align, G_038000_TILE_MODE(word0));
1645 		return -EINVAL;
1646 	}
1647 	if (!IS_ALIGNED(mip_offset, base_align)) {
1648 		dev_warn(p->dev, "%s:%d tex mip offset (0x%llx, 0x%llx, %d) invalid\n",
1649 			 __func__, __LINE__, mip_offset, base_align, G_038000_TILE_MODE(word0));
1650 		return -EINVAL;
1651 	}
1652 
1653 	word2 = radeon_get_ib_value(p, idx + 2) << 8;
1654 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1655 
1656 	word0 = radeon_get_ib_value(p, idx + 4);
1657 	word1 = radeon_get_ib_value(p, idx + 5);
1658 	blevel = G_038010_BASE_LEVEL(word0);
1659 	llevel = G_038014_LAST_LEVEL(word1);
1660 	if (blevel > llevel) {
1661 		dev_warn(p->dev, "texture blevel %d > llevel %d\n",
1662 			 blevel, llevel);
1663 	}
1664 	if (array == 1) {
1665 		barray = G_038014_BASE_ARRAY(word1);
1666 		larray = G_038014_LAST_ARRAY(word1);
1667 
1668 		nfaces = larray - barray + 1;
1669 	}
1670 	r600_texture_size(nfaces, blevel, llevel, w0, h0, d0, format,
1671 			  pitch_align, height_align, base_align,
1672 			  &l0_size, &mipmap_size);
1673 	/* using get ib will give us the offset into the texture bo */
1674 	if ((l0_size + word2) > radeon_bo_size(texture)) {
1675 		dev_warn(p->dev, "texture bo too small ((%d %d) (%d %d) %d %d %d -> %d have %ld)\n",
1676 			 w0, h0, pitch_align, height_align,
1677 			 array_check.array_mode, format, word2,
1678 			 l0_size, radeon_bo_size(texture));
1679 		dev_warn(p->dev, "alignments %d %d %d %lld\n", pitch, pitch_align, height_align, base_align);
1680 		return -EINVAL;
1681 	}
1682 	/* using get ib will give us the offset into the mipmap bo */
1683 	word3 = radeon_get_ib_value(p, idx + 3) << 8;
1684 	if ((mipmap_size + word3) > radeon_bo_size(mipmap)) {
1685 		/*dev_warn(p->dev, "mipmap bo too small (%d %d %d %d %d %d -> %d have %ld)\n",
1686 		  w0, h0, format, blevel, nlevels, word3, mipmap_size, radeon_bo_size(texture));*/
1687 	}
1688 	return 0;
1689 }
1690 
r600_is_safe_reg(struct radeon_cs_parser * p,u32 reg,u32 idx)1691 static bool r600_is_safe_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1692 {
1693 	u32 m, i;
1694 
1695 	i = (reg >> 7);
1696 	if (i >= ARRAY_SIZE(r600_reg_safe_bm)) {
1697 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1698 		return false;
1699 	}
1700 	m = 1 << ((reg >> 2) & 31);
1701 	if (!(r600_reg_safe_bm[i] & m))
1702 		return true;
1703 	dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1704 	return false;
1705 }
1706 
r600_packet3_check(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt)1707 static int r600_packet3_check(struct radeon_cs_parser *p,
1708 				struct radeon_cs_packet *pkt)
1709 {
1710 	struct radeon_cs_reloc *reloc;
1711 	struct r600_cs_track *track;
1712 	volatile u32 *ib;
1713 	unsigned idx;
1714 	unsigned i;
1715 	unsigned start_reg, end_reg, reg;
1716 	int r;
1717 	u32 idx_value;
1718 
1719 	track = (struct r600_cs_track *)p->track;
1720 	ib = p->ib->ptr;
1721 	idx = pkt->idx + 1;
1722 	idx_value = radeon_get_ib_value(p, idx);
1723 
1724 	switch (pkt->opcode) {
1725 	case PACKET3_SET_PREDICATION:
1726 	{
1727 		int pred_op;
1728 		int tmp;
1729 		uint64_t offset;
1730 
1731 		if (pkt->count != 1) {
1732 			DRM_ERROR("bad SET PREDICATION\n");
1733 			return -EINVAL;
1734 		}
1735 
1736 		tmp = radeon_get_ib_value(p, idx + 1);
1737 		pred_op = (tmp >> 16) & 0x7;
1738 
1739 		/* for the clear predicate operation */
1740 		if (pred_op == 0)
1741 			return 0;
1742 
1743 		if (pred_op > 2) {
1744 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1745 			return -EINVAL;
1746 		}
1747 
1748 		r = r600_cs_packet_next_reloc(p, &reloc);
1749 		if (r) {
1750 			DRM_ERROR("bad SET PREDICATION\n");
1751 			return -EINVAL;
1752 		}
1753 
1754 		offset = reloc->lobj.gpu_offset +
1755 		         (idx_value & 0xfffffff0) +
1756 		         ((u64)(tmp & 0xff) << 32);
1757 
1758 		ib[idx + 0] = offset;
1759 		ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1760 	}
1761 	break;
1762 
1763 	case PACKET3_START_3D_CMDBUF:
1764 		if (p->family >= CHIP_RV770 || pkt->count) {
1765 			DRM_ERROR("bad START_3D\n");
1766 			return -EINVAL;
1767 		}
1768 		break;
1769 	case PACKET3_CONTEXT_CONTROL:
1770 		if (pkt->count != 1) {
1771 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1772 			return -EINVAL;
1773 		}
1774 		break;
1775 	case PACKET3_INDEX_TYPE:
1776 	case PACKET3_NUM_INSTANCES:
1777 		if (pkt->count) {
1778 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES\n");
1779 			return -EINVAL;
1780 		}
1781 		break;
1782 	case PACKET3_DRAW_INDEX:
1783 	{
1784 		uint64_t offset;
1785 		if (pkt->count != 3) {
1786 			DRM_ERROR("bad DRAW_INDEX\n");
1787 			return -EINVAL;
1788 		}
1789 		r = r600_cs_packet_next_reloc(p, &reloc);
1790 		if (r) {
1791 			DRM_ERROR("bad DRAW_INDEX\n");
1792 			return -EINVAL;
1793 		}
1794 
1795 		offset = reloc->lobj.gpu_offset +
1796 		         idx_value +
1797 		         ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1798 
1799 		ib[idx+0] = offset;
1800 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1801 
1802 		r = r600_cs_track_check(p);
1803 		if (r) {
1804 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1805 			return r;
1806 		}
1807 		break;
1808 	}
1809 	case PACKET3_DRAW_INDEX_AUTO:
1810 		if (pkt->count != 1) {
1811 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1812 			return -EINVAL;
1813 		}
1814 		r = r600_cs_track_check(p);
1815 		if (r) {
1816 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1817 			return r;
1818 		}
1819 		break;
1820 	case PACKET3_DRAW_INDEX_IMMD_BE:
1821 	case PACKET3_DRAW_INDEX_IMMD:
1822 		if (pkt->count < 2) {
1823 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1824 			return -EINVAL;
1825 		}
1826 		r = r600_cs_track_check(p);
1827 		if (r) {
1828 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1829 			return r;
1830 		}
1831 		break;
1832 	case PACKET3_WAIT_REG_MEM:
1833 		if (pkt->count != 5) {
1834 			DRM_ERROR("bad WAIT_REG_MEM\n");
1835 			return -EINVAL;
1836 		}
1837 		/* bit 4 is reg (0) or mem (1) */
1838 		if (idx_value & 0x10) {
1839 			uint64_t offset;
1840 
1841 			r = r600_cs_packet_next_reloc(p, &reloc);
1842 			if (r) {
1843 				DRM_ERROR("bad WAIT_REG_MEM\n");
1844 				return -EINVAL;
1845 			}
1846 
1847 			offset = reloc->lobj.gpu_offset +
1848 			         (radeon_get_ib_value(p, idx+1) & 0xfffffff0) +
1849 			         ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1850 
1851 			ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffff0);
1852 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1853 		}
1854 		break;
1855 	case PACKET3_SURFACE_SYNC:
1856 		if (pkt->count != 3) {
1857 			DRM_ERROR("bad SURFACE_SYNC\n");
1858 			return -EINVAL;
1859 		}
1860 		/* 0xffffffff/0x0 is flush all cache flag */
1861 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1862 		    radeon_get_ib_value(p, idx + 2) != 0) {
1863 			r = r600_cs_packet_next_reloc(p, &reloc);
1864 			if (r) {
1865 				DRM_ERROR("bad SURFACE_SYNC\n");
1866 				return -EINVAL;
1867 			}
1868 			ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1869 		}
1870 		break;
1871 	case PACKET3_EVENT_WRITE:
1872 		if (pkt->count != 2 && pkt->count != 0) {
1873 			DRM_ERROR("bad EVENT_WRITE\n");
1874 			return -EINVAL;
1875 		}
1876 		if (pkt->count) {
1877 			uint64_t offset;
1878 
1879 			r = r600_cs_packet_next_reloc(p, &reloc);
1880 			if (r) {
1881 				DRM_ERROR("bad EVENT_WRITE\n");
1882 				return -EINVAL;
1883 			}
1884 			offset = reloc->lobj.gpu_offset +
1885 			         (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
1886 			         ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1887 
1888 			ib[idx+1] = offset & 0xfffffff8;
1889 			ib[idx+2] = upper_32_bits(offset) & 0xff;
1890 		}
1891 		break;
1892 	case PACKET3_EVENT_WRITE_EOP:
1893 	{
1894 		uint64_t offset;
1895 
1896 		if (pkt->count != 4) {
1897 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1898 			return -EINVAL;
1899 		}
1900 		r = r600_cs_packet_next_reloc(p, &reloc);
1901 		if (r) {
1902 			DRM_ERROR("bad EVENT_WRITE\n");
1903 			return -EINVAL;
1904 		}
1905 
1906 		offset = reloc->lobj.gpu_offset +
1907 		         (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
1908 		         ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1909 
1910 		ib[idx+1] = offset & 0xfffffffc;
1911 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1912 		break;
1913 	}
1914 	case PACKET3_SET_CONFIG_REG:
1915 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_OFFSET;
1916 		end_reg = 4 * pkt->count + start_reg - 4;
1917 		if ((start_reg < PACKET3_SET_CONFIG_REG_OFFSET) ||
1918 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1919 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1920 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1921 			return -EINVAL;
1922 		}
1923 		for (i = 0; i < pkt->count; i++) {
1924 			reg = start_reg + (4 * i);
1925 			r = r600_cs_check_reg(p, reg, idx+1+i);
1926 			if (r)
1927 				return r;
1928 		}
1929 		break;
1930 	case PACKET3_SET_CONTEXT_REG:
1931 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_OFFSET;
1932 		end_reg = 4 * pkt->count + start_reg - 4;
1933 		if ((start_reg < PACKET3_SET_CONTEXT_REG_OFFSET) ||
1934 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1935 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1936 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1937 			return -EINVAL;
1938 		}
1939 		for (i = 0; i < pkt->count; i++) {
1940 			reg = start_reg + (4 * i);
1941 			r = r600_cs_check_reg(p, reg, idx+1+i);
1942 			if (r)
1943 				return r;
1944 		}
1945 		break;
1946 	case PACKET3_SET_RESOURCE:
1947 		if (pkt->count % 7) {
1948 			DRM_ERROR("bad SET_RESOURCE\n");
1949 			return -EINVAL;
1950 		}
1951 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_OFFSET;
1952 		end_reg = 4 * pkt->count + start_reg - 4;
1953 		if ((start_reg < PACKET3_SET_RESOURCE_OFFSET) ||
1954 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
1955 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
1956 			DRM_ERROR("bad SET_RESOURCE\n");
1957 			return -EINVAL;
1958 		}
1959 		for (i = 0; i < (pkt->count / 7); i++) {
1960 			struct radeon_bo *texture, *mipmap;
1961 			u32 size, offset, base_offset, mip_offset;
1962 
1963 			switch (G__SQ_VTX_CONSTANT_TYPE(radeon_get_ib_value(p, idx+(i*7)+6+1))) {
1964 			case SQ_TEX_VTX_VALID_TEXTURE:
1965 				/* tex base */
1966 				r = r600_cs_packet_next_reloc(p, &reloc);
1967 				if (r) {
1968 					DRM_ERROR("bad SET_RESOURCE\n");
1969 					return -EINVAL;
1970 				}
1971 				base_offset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1972 				if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1973 					if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1974 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_2D_TILED_THIN1);
1975 					else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1976 						ib[idx+1+(i*7)+0] |= S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
1977 				}
1978 				texture = reloc->robj;
1979 				/* tex mip base */
1980 				r = r600_cs_packet_next_reloc(p, &reloc);
1981 				if (r) {
1982 					DRM_ERROR("bad SET_RESOURCE\n");
1983 					return -EINVAL;
1984 				}
1985 				mip_offset = (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1986 				mipmap = reloc->robj;
1987 				r = r600_check_texture_resource(p,  idx+(i*7)+1,
1988 								texture, mipmap,
1989 								base_offset + radeon_get_ib_value(p, idx+1+(i*7)+2),
1990 								mip_offset + radeon_get_ib_value(p, idx+1+(i*7)+3),
1991 								reloc->lobj.tiling_flags);
1992 				if (r)
1993 					return r;
1994 				ib[idx+1+(i*7)+2] += base_offset;
1995 				ib[idx+1+(i*7)+3] += mip_offset;
1996 				break;
1997 			case SQ_TEX_VTX_VALID_BUFFER:
1998 			{
1999 				uint64_t offset64;
2000 				/* vtx base */
2001 				r = r600_cs_packet_next_reloc(p, &reloc);
2002 				if (r) {
2003 					DRM_ERROR("bad SET_RESOURCE\n");
2004 					return -EINVAL;
2005 				}
2006 				offset = radeon_get_ib_value(p, idx+1+(i*7)+0);
2007 				size = radeon_get_ib_value(p, idx+1+(i*7)+1) + 1;
2008 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2009 					/* force size to size of the buffer */
2010 					dev_warn(p->dev, "vbo resource seems too big (%d) for the bo (%ld)\n",
2011 						 size + offset, radeon_bo_size(reloc->robj));
2012 					ib[idx+1+(i*7)+1] = radeon_bo_size(reloc->robj) - offset;
2013 				}
2014 
2015 				offset64 = reloc->lobj.gpu_offset + offset;
2016 				ib[idx+1+(i*8)+0] = offset64;
2017 				ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2018 						    (upper_32_bits(offset64) & 0xff);
2019 				break;
2020 			}
2021 			case SQ_TEX_VTX_INVALID_TEXTURE:
2022 			case SQ_TEX_VTX_INVALID_BUFFER:
2023 			default:
2024 				DRM_ERROR("bad SET_RESOURCE\n");
2025 				return -EINVAL;
2026 			}
2027 		}
2028 		break;
2029 	case PACKET3_SET_ALU_CONST:
2030 		if (track->sq_config & DX9_CONSTS) {
2031 			start_reg = (idx_value << 2) + PACKET3_SET_ALU_CONST_OFFSET;
2032 			end_reg = 4 * pkt->count + start_reg - 4;
2033 			if ((start_reg < PACKET3_SET_ALU_CONST_OFFSET) ||
2034 			    (start_reg >= PACKET3_SET_ALU_CONST_END) ||
2035 			    (end_reg >= PACKET3_SET_ALU_CONST_END)) {
2036 				DRM_ERROR("bad SET_ALU_CONST\n");
2037 				return -EINVAL;
2038 			}
2039 		}
2040 		break;
2041 	case PACKET3_SET_BOOL_CONST:
2042 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_OFFSET;
2043 		end_reg = 4 * pkt->count + start_reg - 4;
2044 		if ((start_reg < PACKET3_SET_BOOL_CONST_OFFSET) ||
2045 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2046 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2047 			DRM_ERROR("bad SET_BOOL_CONST\n");
2048 			return -EINVAL;
2049 		}
2050 		break;
2051 	case PACKET3_SET_LOOP_CONST:
2052 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_OFFSET;
2053 		end_reg = 4 * pkt->count + start_reg - 4;
2054 		if ((start_reg < PACKET3_SET_LOOP_CONST_OFFSET) ||
2055 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2056 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2057 			DRM_ERROR("bad SET_LOOP_CONST\n");
2058 			return -EINVAL;
2059 		}
2060 		break;
2061 	case PACKET3_SET_CTL_CONST:
2062 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_OFFSET;
2063 		end_reg = 4 * pkt->count + start_reg - 4;
2064 		if ((start_reg < PACKET3_SET_CTL_CONST_OFFSET) ||
2065 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2066 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2067 			DRM_ERROR("bad SET_CTL_CONST\n");
2068 			return -EINVAL;
2069 		}
2070 		break;
2071 	case PACKET3_SET_SAMPLER:
2072 		if (pkt->count % 3) {
2073 			DRM_ERROR("bad SET_SAMPLER\n");
2074 			return -EINVAL;
2075 		}
2076 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_OFFSET;
2077 		end_reg = 4 * pkt->count + start_reg - 4;
2078 		if ((start_reg < PACKET3_SET_SAMPLER_OFFSET) ||
2079 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
2080 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
2081 			DRM_ERROR("bad SET_SAMPLER\n");
2082 			return -EINVAL;
2083 		}
2084 		break;
2085 	case PACKET3_SURFACE_BASE_UPDATE:
2086 		if (p->family >= CHIP_RV770 || p->family == CHIP_R600) {
2087 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2088 			return -EINVAL;
2089 		}
2090 		if (pkt->count) {
2091 			DRM_ERROR("bad SURFACE_BASE_UPDATE\n");
2092 			return -EINVAL;
2093 		}
2094 		break;
2095 	case PACKET3_STRMOUT_BUFFER_UPDATE:
2096 		if (pkt->count != 4) {
2097 			DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2098 			return -EINVAL;
2099 		}
2100 		/* Updating memory at DST_ADDRESS. */
2101 		if (idx_value & 0x1) {
2102 			u64 offset;
2103 			r = r600_cs_packet_next_reloc(p, &reloc);
2104 			if (r) {
2105 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2106 				return -EINVAL;
2107 			}
2108 			offset = radeon_get_ib_value(p, idx+1);
2109 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2110 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2111 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2112 					  offset + 4, radeon_bo_size(reloc->robj));
2113 				return -EINVAL;
2114 			}
2115 			offset += reloc->lobj.gpu_offset;
2116 			ib[idx+1] = offset;
2117 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2118 		}
2119 		/* Reading data from SRC_ADDRESS. */
2120 		if (((idx_value >> 1) & 0x3) == 2) {
2121 			u64 offset;
2122 			r = r600_cs_packet_next_reloc(p, &reloc);
2123 			if (r) {
2124 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2125 				return -EINVAL;
2126 			}
2127 			offset = radeon_get_ib_value(p, idx+3);
2128 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2129 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2130 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2131 					  offset + 4, radeon_bo_size(reloc->robj));
2132 				return -EINVAL;
2133 			}
2134 			offset += reloc->lobj.gpu_offset;
2135 			ib[idx+3] = offset;
2136 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2137 		}
2138 		break;
2139 	case PACKET3_COPY_DW:
2140 		if (pkt->count != 4) {
2141 			DRM_ERROR("bad COPY_DW (invalid count)\n");
2142 			return -EINVAL;
2143 		}
2144 		if (idx_value & 0x1) {
2145 			u64 offset;
2146 			/* SRC is memory. */
2147 			r = r600_cs_packet_next_reloc(p, &reloc);
2148 			if (r) {
2149 				DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2150 				return -EINVAL;
2151 			}
2152 			offset = radeon_get_ib_value(p, idx+1);
2153 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2154 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2155 				DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2156 					  offset + 4, radeon_bo_size(reloc->robj));
2157 				return -EINVAL;
2158 			}
2159 			offset += reloc->lobj.gpu_offset;
2160 			ib[idx+1] = offset;
2161 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2162 		} else {
2163 			/* SRC is a reg. */
2164 			reg = radeon_get_ib_value(p, idx+1) << 2;
2165 			if (!r600_is_safe_reg(p, reg, idx+1))
2166 				return -EINVAL;
2167 		}
2168 		if (idx_value & 0x2) {
2169 			u64 offset;
2170 			/* DST is memory. */
2171 			r = r600_cs_packet_next_reloc(p, &reloc);
2172 			if (r) {
2173 				DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2174 				return -EINVAL;
2175 			}
2176 			offset = radeon_get_ib_value(p, idx+3);
2177 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2178 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2179 				DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2180 					  offset + 4, radeon_bo_size(reloc->robj));
2181 				return -EINVAL;
2182 			}
2183 			offset += reloc->lobj.gpu_offset;
2184 			ib[idx+3] = offset;
2185 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2186 		} else {
2187 			/* DST is a reg. */
2188 			reg = radeon_get_ib_value(p, idx+3) << 2;
2189 			if (!r600_is_safe_reg(p, reg, idx+3))
2190 				return -EINVAL;
2191 		}
2192 		break;
2193 	case PACKET3_NOP:
2194 		break;
2195 	default:
2196 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2197 		return -EINVAL;
2198 	}
2199 	return 0;
2200 }
2201 
r600_cs_parse(struct radeon_cs_parser * p)2202 int r600_cs_parse(struct radeon_cs_parser *p)
2203 {
2204 	struct radeon_cs_packet pkt;
2205 	struct r600_cs_track *track;
2206 	int r;
2207 
2208 	if (p->track == NULL) {
2209 		/* initialize tracker, we are in kms */
2210 		track = kzalloc(sizeof(*track), GFP_KERNEL);
2211 		if (track == NULL)
2212 			return -ENOMEM;
2213 		r600_cs_track_init(track);
2214 		if (p->rdev->family < CHIP_RV770) {
2215 			track->npipes = p->rdev->config.r600.tiling_npipes;
2216 			track->nbanks = p->rdev->config.r600.tiling_nbanks;
2217 			track->group_size = p->rdev->config.r600.tiling_group_size;
2218 		} else if (p->rdev->family <= CHIP_RV740) {
2219 			track->npipes = p->rdev->config.rv770.tiling_npipes;
2220 			track->nbanks = p->rdev->config.rv770.tiling_nbanks;
2221 			track->group_size = p->rdev->config.rv770.tiling_group_size;
2222 		}
2223 		p->track = track;
2224 	}
2225 	do {
2226 		r = r600_cs_packet_parse(p, &pkt, p->idx);
2227 		if (r) {
2228 			kfree(p->track);
2229 			p->track = NULL;
2230 			return r;
2231 		}
2232 		p->idx += pkt.count + 2;
2233 		switch (pkt.type) {
2234 		case PACKET_TYPE0:
2235 			r = r600_cs_parse_packet0(p, &pkt);
2236 			break;
2237 		case PACKET_TYPE2:
2238 			break;
2239 		case PACKET_TYPE3:
2240 			r = r600_packet3_check(p, &pkt);
2241 			break;
2242 		default:
2243 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2244 			kfree(p->track);
2245 			p->track = NULL;
2246 			return -EINVAL;
2247 		}
2248 		if (r) {
2249 			kfree(p->track);
2250 			p->track = NULL;
2251 			return r;
2252 		}
2253 	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
2254 #if 0
2255 	for (r = 0; r < p->ib->length_dw; r++) {
2256 		printk(KERN_INFO "%05d  0x%08X\n", r, p->ib->ptr[r]);
2257 		mdelay(1);
2258 	}
2259 #endif
2260 	kfree(p->track);
2261 	p->track = NULL;
2262 	return 0;
2263 }
2264 
r600_cs_parser_relocs_legacy(struct radeon_cs_parser * p)2265 static int r600_cs_parser_relocs_legacy(struct radeon_cs_parser *p)
2266 {
2267 	if (p->chunk_relocs_idx == -1) {
2268 		return 0;
2269 	}
2270 	p->relocs = kzalloc(sizeof(struct radeon_cs_reloc), GFP_KERNEL);
2271 	if (p->relocs == NULL) {
2272 		return -ENOMEM;
2273 	}
2274 	return 0;
2275 }
2276 
2277 /**
2278  * cs_parser_fini() - clean parser states
2279  * @parser:	parser structure holding parsing context.
2280  * @error:	error number
2281  *
2282  * If error is set than unvalidate buffer, otherwise just free memory
2283  * used by parsing context.
2284  **/
r600_cs_parser_fini(struct radeon_cs_parser * parser,int error)2285 static void r600_cs_parser_fini(struct radeon_cs_parser *parser, int error)
2286 {
2287 	unsigned i;
2288 
2289 	kfree(parser->relocs);
2290 	for (i = 0; i < parser->nchunks; i++) {
2291 		kfree(parser->chunks[i].kdata);
2292 		kfree(parser->chunks[i].kpage[0]);
2293 		kfree(parser->chunks[i].kpage[1]);
2294 	}
2295 	kfree(parser->chunks);
2296 	kfree(parser->chunks_array);
2297 }
2298 
r600_cs_legacy(struct drm_device * dev,void * data,struct drm_file * filp,unsigned family,u32 * ib,int * l)2299 int r600_cs_legacy(struct drm_device *dev, void *data, struct drm_file *filp,
2300 			unsigned family, u32 *ib, int *l)
2301 {
2302 	struct radeon_cs_parser parser;
2303 	struct radeon_cs_chunk *ib_chunk;
2304 	struct radeon_ib fake_ib;
2305 	struct r600_cs_track *track;
2306 	int r;
2307 
2308 	/* initialize tracker */
2309 	track = kzalloc(sizeof(*track), GFP_KERNEL);
2310 	if (track == NULL)
2311 		return -ENOMEM;
2312 	r600_cs_track_init(track);
2313 	r600_cs_legacy_get_tiling_conf(dev, &track->npipes, &track->nbanks, &track->group_size);
2314 	/* initialize parser */
2315 	memset(&parser, 0, sizeof(struct radeon_cs_parser));
2316 	parser.filp = filp;
2317 	parser.dev = &dev->pdev->dev;
2318 	parser.rdev = NULL;
2319 	parser.family = family;
2320 	parser.ib = &fake_ib;
2321 	parser.track = track;
2322 	fake_ib.ptr = ib;
2323 	r = radeon_cs_parser_init(&parser, data);
2324 	if (r) {
2325 		DRM_ERROR("Failed to initialize parser !\n");
2326 		r600_cs_parser_fini(&parser, r);
2327 		return r;
2328 	}
2329 	r = r600_cs_parser_relocs_legacy(&parser);
2330 	if (r) {
2331 		DRM_ERROR("Failed to parse relocation !\n");
2332 		r600_cs_parser_fini(&parser, r);
2333 		return r;
2334 	}
2335 	/* Copy the packet into the IB, the parser will read from the
2336 	 * input memory (cached) and write to the IB (which can be
2337 	 * uncached). */
2338 	ib_chunk = &parser.chunks[parser.chunk_ib_idx];
2339 	parser.ib->length_dw = ib_chunk->length_dw;
2340 	*l = parser.ib->length_dw;
2341 	r = r600_cs_parse(&parser);
2342 	if (r) {
2343 		DRM_ERROR("Invalid command stream !\n");
2344 		r600_cs_parser_fini(&parser, r);
2345 		return r;
2346 	}
2347 	r = radeon_cs_finish_pages(&parser);
2348 	if (r) {
2349 		DRM_ERROR("Invalid command stream !\n");
2350 		r600_cs_parser_fini(&parser, r);
2351 		return r;
2352 	}
2353 	r600_cs_parser_fini(&parser, r);
2354 	return r;
2355 }
2356 
r600_cs_legacy_init(void)2357 void r600_cs_legacy_init(void)
2358 {
2359 	r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;
2360 }
2361