xref: /openbmc/linux/drivers/gpu/drm/radeon/evergreen_cs.c (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
1 /*
2  * Copyright 2010 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include "drmP.h"
29 #include "radeon.h"
30 #include "evergreend.h"
31 #include "evergreen_reg_safe.h"
32 
33 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
34 					  struct radeon_cs_reloc **cs_reloc);
35 
36 struct evergreen_cs_track {
37 	u32			group_size;
38 	u32			nbanks;
39 	u32			npipes;
40 	/* value we track */
41 	u32			nsamples;
42 	u32			cb_color_base_last[12];
43 	struct radeon_bo	*cb_color_bo[12];
44 	u32			cb_color_bo_offset[12];
45 	struct radeon_bo	*cb_color_fmask_bo[8];
46 	struct radeon_bo	*cb_color_cmask_bo[8];
47 	u32			cb_color_info[12];
48 	u32			cb_color_view[12];
49 	u32			cb_color_pitch_idx[12];
50 	u32			cb_color_slice_idx[12];
51 	u32			cb_color_dim_idx[12];
52 	u32			cb_color_dim[12];
53 	u32			cb_color_pitch[12];
54 	u32			cb_color_slice[12];
55 	u32			cb_color_cmask_slice[8];
56 	u32			cb_color_fmask_slice[8];
57 	u32			cb_target_mask;
58 	u32			cb_shader_mask;
59 	u32			vgt_strmout_config;
60 	u32			vgt_strmout_buffer_config;
61 	u32			db_depth_control;
62 	u32			db_depth_view;
63 	u32			db_depth_size;
64 	u32			db_depth_size_idx;
65 	u32			db_z_info;
66 	u32			db_z_idx;
67 	u32			db_z_read_offset;
68 	u32			db_z_write_offset;
69 	struct radeon_bo	*db_z_read_bo;
70 	struct radeon_bo	*db_z_write_bo;
71 	u32			db_s_info;
72 	u32			db_s_idx;
73 	u32			db_s_read_offset;
74 	u32			db_s_write_offset;
75 	struct radeon_bo	*db_s_read_bo;
76 	struct radeon_bo	*db_s_write_bo;
77 };
78 
79 static void evergreen_cs_track_init(struct evergreen_cs_track *track)
80 {
81 	int i;
82 
83 	for (i = 0; i < 8; i++) {
84 		track->cb_color_fmask_bo[i] = NULL;
85 		track->cb_color_cmask_bo[i] = NULL;
86 		track->cb_color_cmask_slice[i] = 0;
87 		track->cb_color_fmask_slice[i] = 0;
88 	}
89 
90 	for (i = 0; i < 12; i++) {
91 		track->cb_color_base_last[i] = 0;
92 		track->cb_color_bo[i] = NULL;
93 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
94 		track->cb_color_info[i] = 0;
95 		track->cb_color_view[i] = 0;
96 		track->cb_color_pitch_idx[i] = 0;
97 		track->cb_color_slice_idx[i] = 0;
98 		track->cb_color_dim[i] = 0;
99 		track->cb_color_pitch[i] = 0;
100 		track->cb_color_slice[i] = 0;
101 		track->cb_color_dim[i] = 0;
102 	}
103 	track->cb_target_mask = 0xFFFFFFFF;
104 	track->cb_shader_mask = 0xFFFFFFFF;
105 
106 	track->db_depth_view = 0xFFFFC000;
107 	track->db_depth_size = 0xFFFFFFFF;
108 	track->db_depth_size_idx = 0;
109 	track->db_depth_control = 0xFFFFFFFF;
110 	track->db_z_info = 0xFFFFFFFF;
111 	track->db_z_idx = 0xFFFFFFFF;
112 	track->db_z_read_offset = 0xFFFFFFFF;
113 	track->db_z_write_offset = 0xFFFFFFFF;
114 	track->db_z_read_bo = NULL;
115 	track->db_z_write_bo = NULL;
116 	track->db_s_info = 0xFFFFFFFF;
117 	track->db_s_idx = 0xFFFFFFFF;
118 	track->db_s_read_offset = 0xFFFFFFFF;
119 	track->db_s_write_offset = 0xFFFFFFFF;
120 	track->db_s_read_bo = NULL;
121 	track->db_s_write_bo = NULL;
122 }
123 
124 static inline int evergreen_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
125 {
126 	/* XXX fill in */
127 	return 0;
128 }
129 
130 static int evergreen_cs_track_check(struct radeon_cs_parser *p)
131 {
132 	struct evergreen_cs_track *track = p->track;
133 
134 	/* we don't support stream out buffer yet */
135 	if (track->vgt_strmout_config || track->vgt_strmout_buffer_config) {
136 		dev_warn(p->dev, "this kernel doesn't support SMX output buffer\n");
137 		return -EINVAL;
138 	}
139 
140 	/* XXX fill in */
141 	return 0;
142 }
143 
144 /**
145  * evergreen_cs_packet_parse() - parse cp packet and point ib index to next packet
146  * @parser:	parser structure holding parsing context.
147  * @pkt:	where to store packet informations
148  *
149  * Assume that chunk_ib_index is properly set. Will return -EINVAL
150  * if packet is bigger than remaining ib size. or if packets is unknown.
151  **/
152 int evergreen_cs_packet_parse(struct radeon_cs_parser *p,
153 			      struct radeon_cs_packet *pkt,
154 			      unsigned idx)
155 {
156 	struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
157 	uint32_t header;
158 
159 	if (idx >= ib_chunk->length_dw) {
160 		DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
161 			  idx, ib_chunk->length_dw);
162 		return -EINVAL;
163 	}
164 	header = radeon_get_ib_value(p, idx);
165 	pkt->idx = idx;
166 	pkt->type = CP_PACKET_GET_TYPE(header);
167 	pkt->count = CP_PACKET_GET_COUNT(header);
168 	pkt->one_reg_wr = 0;
169 	switch (pkt->type) {
170 	case PACKET_TYPE0:
171 		pkt->reg = CP_PACKET0_GET_REG(header);
172 		break;
173 	case PACKET_TYPE3:
174 		pkt->opcode = CP_PACKET3_GET_OPCODE(header);
175 		break;
176 	case PACKET_TYPE2:
177 		pkt->count = -1;
178 		break;
179 	default:
180 		DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
181 		return -EINVAL;
182 	}
183 	if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
184 		DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
185 			  pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
186 		return -EINVAL;
187 	}
188 	return 0;
189 }
190 
191 /**
192  * evergreen_cs_packet_next_reloc() - parse next packet which should be reloc packet3
193  * @parser:		parser structure holding parsing context.
194  * @data:		pointer to relocation data
195  * @offset_start:	starting offset
196  * @offset_mask:	offset mask (to align start offset on)
197  * @reloc:		reloc informations
198  *
199  * Check next packet is relocation packet3, do bo validation and compute
200  * GPU offset using the provided start.
201  **/
202 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
203 					  struct radeon_cs_reloc **cs_reloc)
204 {
205 	struct radeon_cs_chunk *relocs_chunk;
206 	struct radeon_cs_packet p3reloc;
207 	unsigned idx;
208 	int r;
209 
210 	if (p->chunk_relocs_idx == -1) {
211 		DRM_ERROR("No relocation chunk !\n");
212 		return -EINVAL;
213 	}
214 	*cs_reloc = NULL;
215 	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
216 	r = evergreen_cs_packet_parse(p, &p3reloc, p->idx);
217 	if (r) {
218 		return r;
219 	}
220 	p->idx += p3reloc.count + 2;
221 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
222 		DRM_ERROR("No packet3 for relocation for packet at %d.\n",
223 			  p3reloc.idx);
224 		return -EINVAL;
225 	}
226 	idx = radeon_get_ib_value(p, p3reloc.idx + 1);
227 	if (idx >= relocs_chunk->length_dw) {
228 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
229 			  idx, relocs_chunk->length_dw);
230 		return -EINVAL;
231 	}
232 	/* FIXME: we assume reloc size is 4 dwords */
233 	*cs_reloc = p->relocs_ptr[(idx / 4)];
234 	return 0;
235 }
236 
237 /**
238  * evergreen_cs_packet_next_is_pkt3_nop() - test if next packet is packet3 nop for reloc
239  * @parser:		parser structure holding parsing context.
240  *
241  * Check next packet is relocation packet3, do bo validation and compute
242  * GPU offset using the provided start.
243  **/
244 static inline int evergreen_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
245 {
246 	struct radeon_cs_packet p3reloc;
247 	int r;
248 
249 	r = evergreen_cs_packet_parse(p, &p3reloc, p->idx);
250 	if (r) {
251 		return 0;
252 	}
253 	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
254 		return 0;
255 	}
256 	return 1;
257 }
258 
259 /**
260  * evergreen_cs_packet_next_vline() - parse userspace VLINE packet
261  * @parser:		parser structure holding parsing context.
262  *
263  * Userspace sends a special sequence for VLINE waits.
264  * PACKET0 - VLINE_START_END + value
265  * PACKET3 - WAIT_REG_MEM poll vline status reg
266  * RELOC (P3) - crtc_id in reloc.
267  *
268  * This function parses this and relocates the VLINE START END
269  * and WAIT_REG_MEM packets to the correct crtc.
270  * It also detects a switched off crtc and nulls out the
271  * wait in that case.
272  */
273 static int evergreen_cs_packet_parse_vline(struct radeon_cs_parser *p)
274 {
275 	struct drm_mode_object *obj;
276 	struct drm_crtc *crtc;
277 	struct radeon_crtc *radeon_crtc;
278 	struct radeon_cs_packet p3reloc, wait_reg_mem;
279 	int crtc_id;
280 	int r;
281 	uint32_t header, h_idx, reg, wait_reg_mem_info;
282 	volatile uint32_t *ib;
283 
284 	ib = p->ib->ptr;
285 
286 	/* parse the WAIT_REG_MEM */
287 	r = evergreen_cs_packet_parse(p, &wait_reg_mem, p->idx);
288 	if (r)
289 		return r;
290 
291 	/* check its a WAIT_REG_MEM */
292 	if (wait_reg_mem.type != PACKET_TYPE3 ||
293 	    wait_reg_mem.opcode != PACKET3_WAIT_REG_MEM) {
294 		DRM_ERROR("vline wait missing WAIT_REG_MEM segment\n");
295 		r = -EINVAL;
296 		return r;
297 	}
298 
299 	wait_reg_mem_info = radeon_get_ib_value(p, wait_reg_mem.idx + 1);
300 	/* bit 4 is reg (0) or mem (1) */
301 	if (wait_reg_mem_info & 0x10) {
302 		DRM_ERROR("vline WAIT_REG_MEM waiting on MEM rather than REG\n");
303 		r = -EINVAL;
304 		return r;
305 	}
306 	/* waiting for value to be equal */
307 	if ((wait_reg_mem_info & 0x7) != 0x3) {
308 		DRM_ERROR("vline WAIT_REG_MEM function not equal\n");
309 		r = -EINVAL;
310 		return r;
311 	}
312 	if ((radeon_get_ib_value(p, wait_reg_mem.idx + 2) << 2) != EVERGREEN_VLINE_STATUS) {
313 		DRM_ERROR("vline WAIT_REG_MEM bad reg\n");
314 		r = -EINVAL;
315 		return r;
316 	}
317 
318 	if (radeon_get_ib_value(p, wait_reg_mem.idx + 5) != EVERGREEN_VLINE_STAT) {
319 		DRM_ERROR("vline WAIT_REG_MEM bad bit mask\n");
320 		r = -EINVAL;
321 		return r;
322 	}
323 
324 	/* jump over the NOP */
325 	r = evergreen_cs_packet_parse(p, &p3reloc, p->idx + wait_reg_mem.count + 2);
326 	if (r)
327 		return r;
328 
329 	h_idx = p->idx - 2;
330 	p->idx += wait_reg_mem.count + 2;
331 	p->idx += p3reloc.count + 2;
332 
333 	header = radeon_get_ib_value(p, h_idx);
334 	crtc_id = radeon_get_ib_value(p, h_idx + 2 + 7 + 1);
335 	reg = CP_PACKET0_GET_REG(header);
336 	obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
337 	if (!obj) {
338 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
339 		r = -EINVAL;
340 		goto out;
341 	}
342 	crtc = obj_to_crtc(obj);
343 	radeon_crtc = to_radeon_crtc(crtc);
344 	crtc_id = radeon_crtc->crtc_id;
345 
346 	if (!crtc->enabled) {
347 		/* if the CRTC isn't enabled - we need to nop out the WAIT_REG_MEM */
348 		ib[h_idx + 2] = PACKET2(0);
349 		ib[h_idx + 3] = PACKET2(0);
350 		ib[h_idx + 4] = PACKET2(0);
351 		ib[h_idx + 5] = PACKET2(0);
352 		ib[h_idx + 6] = PACKET2(0);
353 		ib[h_idx + 7] = PACKET2(0);
354 		ib[h_idx + 8] = PACKET2(0);
355 	} else {
356 		switch (reg) {
357 		case EVERGREEN_VLINE_START_END:
358 			header &= ~R600_CP_PACKET0_REG_MASK;
359 			header |= (EVERGREEN_VLINE_START_END + radeon_crtc->crtc_offset) >> 2;
360 			ib[h_idx] = header;
361 			ib[h_idx + 4] = (EVERGREEN_VLINE_STATUS + radeon_crtc->crtc_offset) >> 2;
362 			break;
363 		default:
364 			DRM_ERROR("unknown crtc reloc\n");
365 			r = -EINVAL;
366 			goto out;
367 		}
368 	}
369 out:
370 	return r;
371 }
372 
373 static int evergreen_packet0_check(struct radeon_cs_parser *p,
374 				   struct radeon_cs_packet *pkt,
375 				   unsigned idx, unsigned reg)
376 {
377 	int r;
378 
379 	switch (reg) {
380 	case EVERGREEN_VLINE_START_END:
381 		r = evergreen_cs_packet_parse_vline(p);
382 		if (r) {
383 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
384 					idx, reg);
385 			return r;
386 		}
387 		break;
388 	default:
389 		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
390 		       reg, idx);
391 		return -EINVAL;
392 	}
393 	return 0;
394 }
395 
396 static int evergreen_cs_parse_packet0(struct radeon_cs_parser *p,
397 				      struct radeon_cs_packet *pkt)
398 {
399 	unsigned reg, i;
400 	unsigned idx;
401 	int r;
402 
403 	idx = pkt->idx + 1;
404 	reg = pkt->reg;
405 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
406 		r = evergreen_packet0_check(p, pkt, idx, reg);
407 		if (r) {
408 			return r;
409 		}
410 	}
411 	return 0;
412 }
413 
414 /**
415  * evergreen_cs_check_reg() - check if register is authorized or not
416  * @parser: parser structure holding parsing context
417  * @reg: register we are testing
418  * @idx: index into the cs buffer
419  *
420  * This function will test against evergreen_reg_safe_bm and return 0
421  * if register is safe. If register is not flag as safe this function
422  * will test it against a list of register needind special handling.
423  */
424 static inline int evergreen_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
425 {
426 	struct evergreen_cs_track *track = (struct evergreen_cs_track *)p->track;
427 	struct radeon_cs_reloc *reloc;
428 	u32 last_reg = ARRAY_SIZE(evergreen_reg_safe_bm);
429 	u32 m, i, tmp, *ib;
430 	int r;
431 
432 	i = (reg >> 7);
433 	if (i > last_reg) {
434 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
435 		return -EINVAL;
436 	}
437 	m = 1 << ((reg >> 2) & 31);
438 	if (!(evergreen_reg_safe_bm[i] & m))
439 		return 0;
440 	ib = p->ib->ptr;
441 	switch (reg) {
442 	/* force following reg to 0 in an attemp to disable out buffer
443 	 * which will need us to better understand how it works to perform
444 	 * security check on it (Jerome)
445 	 */
446 	case SQ_ESGS_RING_SIZE:
447 	case SQ_GSVS_RING_SIZE:
448 	case SQ_ESTMP_RING_SIZE:
449 	case SQ_GSTMP_RING_SIZE:
450 	case SQ_HSTMP_RING_SIZE:
451 	case SQ_LSTMP_RING_SIZE:
452 	case SQ_PSTMP_RING_SIZE:
453 	case SQ_VSTMP_RING_SIZE:
454 	case SQ_ESGS_RING_ITEMSIZE:
455 	case SQ_ESTMP_RING_ITEMSIZE:
456 	case SQ_GSTMP_RING_ITEMSIZE:
457 	case SQ_GSVS_RING_ITEMSIZE:
458 	case SQ_GS_VERT_ITEMSIZE:
459 	case SQ_GS_VERT_ITEMSIZE_1:
460 	case SQ_GS_VERT_ITEMSIZE_2:
461 	case SQ_GS_VERT_ITEMSIZE_3:
462 	case SQ_GSVS_RING_OFFSET_1:
463 	case SQ_GSVS_RING_OFFSET_2:
464 	case SQ_GSVS_RING_OFFSET_3:
465 	case SQ_HSTMP_RING_ITEMSIZE:
466 	case SQ_LSTMP_RING_ITEMSIZE:
467 	case SQ_PSTMP_RING_ITEMSIZE:
468 	case SQ_VSTMP_RING_ITEMSIZE:
469 	case VGT_TF_RING_SIZE:
470 		/* get value to populate the IB don't remove */
471 		tmp =radeon_get_ib_value(p, idx);
472 		ib[idx] = 0;
473 		break;
474 	case DB_DEPTH_CONTROL:
475 		track->db_depth_control = radeon_get_ib_value(p, idx);
476 		break;
477 	case DB_Z_INFO:
478 		r = evergreen_cs_packet_next_reloc(p, &reloc);
479 		if (r) {
480 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
481 					"0x%04X\n", reg);
482 			return -EINVAL;
483 		}
484 		track->db_z_info = radeon_get_ib_value(p, idx);
485 		ib[idx] &= ~Z_ARRAY_MODE(0xf);
486 		track->db_z_info &= ~Z_ARRAY_MODE(0xf);
487 		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
488 			ib[idx] |= Z_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
489 			track->db_z_info |= Z_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
490 		} else {
491 			ib[idx] |= Z_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
492 			track->db_z_info |= Z_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
493 		}
494 		break;
495 	case DB_STENCIL_INFO:
496 		track->db_s_info = radeon_get_ib_value(p, idx);
497 		break;
498 	case DB_DEPTH_VIEW:
499 		track->db_depth_view = radeon_get_ib_value(p, idx);
500 		break;
501 	case DB_DEPTH_SIZE:
502 		track->db_depth_size = radeon_get_ib_value(p, idx);
503 		track->db_depth_size_idx = idx;
504 		break;
505 	case DB_Z_READ_BASE:
506 		r = evergreen_cs_packet_next_reloc(p, &reloc);
507 		if (r) {
508 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
509 					"0x%04X\n", reg);
510 			return -EINVAL;
511 		}
512 		track->db_z_read_offset = radeon_get_ib_value(p, idx);
513 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
514 		track->db_z_read_bo = reloc->robj;
515 		break;
516 	case DB_Z_WRITE_BASE:
517 		r = evergreen_cs_packet_next_reloc(p, &reloc);
518 		if (r) {
519 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
520 					"0x%04X\n", reg);
521 			return -EINVAL;
522 		}
523 		track->db_z_write_offset = radeon_get_ib_value(p, idx);
524 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
525 		track->db_z_write_bo = reloc->robj;
526 		break;
527 	case DB_STENCIL_READ_BASE:
528 		r = evergreen_cs_packet_next_reloc(p, &reloc);
529 		if (r) {
530 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
531 					"0x%04X\n", reg);
532 			return -EINVAL;
533 		}
534 		track->db_s_read_offset = radeon_get_ib_value(p, idx);
535 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
536 		track->db_s_read_bo = reloc->robj;
537 		break;
538 	case DB_STENCIL_WRITE_BASE:
539 		r = evergreen_cs_packet_next_reloc(p, &reloc);
540 		if (r) {
541 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
542 					"0x%04X\n", reg);
543 			return -EINVAL;
544 		}
545 		track->db_s_write_offset = radeon_get_ib_value(p, idx);
546 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
547 		track->db_s_write_bo = reloc->robj;
548 		break;
549 	case VGT_STRMOUT_CONFIG:
550 		track->vgt_strmout_config = radeon_get_ib_value(p, idx);
551 		break;
552 	case VGT_STRMOUT_BUFFER_CONFIG:
553 		track->vgt_strmout_buffer_config = radeon_get_ib_value(p, idx);
554 		break;
555 	case CB_TARGET_MASK:
556 		track->cb_target_mask = radeon_get_ib_value(p, idx);
557 		break;
558 	case CB_SHADER_MASK:
559 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
560 		break;
561 	case PA_SC_AA_CONFIG:
562 		tmp = radeon_get_ib_value(p, idx) & MSAA_NUM_SAMPLES_MASK;
563 		track->nsamples = 1 << tmp;
564 		break;
565 	case CB_COLOR0_VIEW:
566 	case CB_COLOR1_VIEW:
567 	case CB_COLOR2_VIEW:
568 	case CB_COLOR3_VIEW:
569 	case CB_COLOR4_VIEW:
570 	case CB_COLOR5_VIEW:
571 	case CB_COLOR6_VIEW:
572 	case CB_COLOR7_VIEW:
573 		tmp = (reg - CB_COLOR0_VIEW) / 0x3c;
574 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
575 		break;
576 	case CB_COLOR8_VIEW:
577 	case CB_COLOR9_VIEW:
578 	case CB_COLOR10_VIEW:
579 	case CB_COLOR11_VIEW:
580 		tmp = ((reg - CB_COLOR8_VIEW) / 0x1c) + 8;
581 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
582 		break;
583 	case CB_COLOR0_INFO:
584 	case CB_COLOR1_INFO:
585 	case CB_COLOR2_INFO:
586 	case CB_COLOR3_INFO:
587 	case CB_COLOR4_INFO:
588 	case CB_COLOR5_INFO:
589 	case CB_COLOR6_INFO:
590 	case CB_COLOR7_INFO:
591 		r = evergreen_cs_packet_next_reloc(p, &reloc);
592 		if (r) {
593 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
594 					"0x%04X\n", reg);
595 			return -EINVAL;
596 		}
597 		tmp = (reg - CB_COLOR0_INFO) / 0x3c;
598 		track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
599 		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
600 			ib[idx] |= CB_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
601 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
602 		} else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
603 			ib[idx] |= CB_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
604 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
605 		}
606 		break;
607 	case CB_COLOR8_INFO:
608 	case CB_COLOR9_INFO:
609 	case CB_COLOR10_INFO:
610 	case CB_COLOR11_INFO:
611 		r = evergreen_cs_packet_next_reloc(p, &reloc);
612 		if (r) {
613 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
614 					"0x%04X\n", reg);
615 			return -EINVAL;
616 		}
617 		tmp = ((reg - CB_COLOR8_INFO) / 0x1c) + 8;
618 		track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
619 		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO) {
620 			ib[idx] |= CB_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
621 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
622 		} else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
623 			ib[idx] |= CB_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
624 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
625 		}
626 		break;
627 	case CB_COLOR0_PITCH:
628 	case CB_COLOR1_PITCH:
629 	case CB_COLOR2_PITCH:
630 	case CB_COLOR3_PITCH:
631 	case CB_COLOR4_PITCH:
632 	case CB_COLOR5_PITCH:
633 	case CB_COLOR6_PITCH:
634 	case CB_COLOR7_PITCH:
635 		tmp = (reg - CB_COLOR0_PITCH) / 0x3c;
636 		track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
637 		track->cb_color_pitch_idx[tmp] = idx;
638 		break;
639 	case CB_COLOR8_PITCH:
640 	case CB_COLOR9_PITCH:
641 	case CB_COLOR10_PITCH:
642 	case CB_COLOR11_PITCH:
643 		tmp = ((reg - CB_COLOR8_PITCH) / 0x1c) + 8;
644 		track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
645 		track->cb_color_pitch_idx[tmp] = idx;
646 		break;
647 	case CB_COLOR0_SLICE:
648 	case CB_COLOR1_SLICE:
649 	case CB_COLOR2_SLICE:
650 	case CB_COLOR3_SLICE:
651 	case CB_COLOR4_SLICE:
652 	case CB_COLOR5_SLICE:
653 	case CB_COLOR6_SLICE:
654 	case CB_COLOR7_SLICE:
655 		tmp = (reg - CB_COLOR0_SLICE) / 0x3c;
656 		track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
657 		track->cb_color_slice_idx[tmp] = idx;
658 		break;
659 	case CB_COLOR8_SLICE:
660 	case CB_COLOR9_SLICE:
661 	case CB_COLOR10_SLICE:
662 	case CB_COLOR11_SLICE:
663 		tmp = ((reg - CB_COLOR8_SLICE) / 0x1c) + 8;
664 		track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
665 		track->cb_color_slice_idx[tmp] = idx;
666 		break;
667 	case CB_COLOR0_ATTRIB:
668 	case CB_COLOR1_ATTRIB:
669 	case CB_COLOR2_ATTRIB:
670 	case CB_COLOR3_ATTRIB:
671 	case CB_COLOR4_ATTRIB:
672 	case CB_COLOR5_ATTRIB:
673 	case CB_COLOR6_ATTRIB:
674 	case CB_COLOR7_ATTRIB:
675 	case CB_COLOR8_ATTRIB:
676 	case CB_COLOR9_ATTRIB:
677 	case CB_COLOR10_ATTRIB:
678 	case CB_COLOR11_ATTRIB:
679 		break;
680 	case CB_COLOR0_DIM:
681 	case CB_COLOR1_DIM:
682 	case CB_COLOR2_DIM:
683 	case CB_COLOR3_DIM:
684 	case CB_COLOR4_DIM:
685 	case CB_COLOR5_DIM:
686 	case CB_COLOR6_DIM:
687 	case CB_COLOR7_DIM:
688 		tmp = (reg - CB_COLOR0_DIM) / 0x3c;
689 		track->cb_color_dim[tmp] = radeon_get_ib_value(p, idx);
690 		track->cb_color_dim_idx[tmp] = idx;
691 		break;
692 	case CB_COLOR8_DIM:
693 	case CB_COLOR9_DIM:
694 	case CB_COLOR10_DIM:
695 	case CB_COLOR11_DIM:
696 		tmp = ((reg - CB_COLOR8_DIM) / 0x1c) + 8;
697 		track->cb_color_dim[tmp] = radeon_get_ib_value(p, idx);
698 		track->cb_color_dim_idx[tmp] = idx;
699 		break;
700 	case CB_COLOR0_FMASK:
701 	case CB_COLOR1_FMASK:
702 	case CB_COLOR2_FMASK:
703 	case CB_COLOR3_FMASK:
704 	case CB_COLOR4_FMASK:
705 	case CB_COLOR5_FMASK:
706 	case CB_COLOR6_FMASK:
707 	case CB_COLOR7_FMASK:
708 		tmp = (reg - CB_COLOR0_FMASK) / 0x3c;
709 		r = evergreen_cs_packet_next_reloc(p, &reloc);
710 		if (r) {
711 			dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
712 			return -EINVAL;
713 		}
714 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
715 		track->cb_color_fmask_bo[tmp] = reloc->robj;
716 		break;
717 	case CB_COLOR0_CMASK:
718 	case CB_COLOR1_CMASK:
719 	case CB_COLOR2_CMASK:
720 	case CB_COLOR3_CMASK:
721 	case CB_COLOR4_CMASK:
722 	case CB_COLOR5_CMASK:
723 	case CB_COLOR6_CMASK:
724 	case CB_COLOR7_CMASK:
725 		tmp = (reg - CB_COLOR0_CMASK) / 0x3c;
726 		r = evergreen_cs_packet_next_reloc(p, &reloc);
727 		if (r) {
728 			dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
729 			return -EINVAL;
730 		}
731 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
732 		track->cb_color_cmask_bo[tmp] = reloc->robj;
733 		break;
734 	case CB_COLOR0_FMASK_SLICE:
735 	case CB_COLOR1_FMASK_SLICE:
736 	case CB_COLOR2_FMASK_SLICE:
737 	case CB_COLOR3_FMASK_SLICE:
738 	case CB_COLOR4_FMASK_SLICE:
739 	case CB_COLOR5_FMASK_SLICE:
740 	case CB_COLOR6_FMASK_SLICE:
741 	case CB_COLOR7_FMASK_SLICE:
742 		tmp = (reg - CB_COLOR0_FMASK_SLICE) / 0x3c;
743 		track->cb_color_fmask_slice[tmp] = radeon_get_ib_value(p, idx);
744 		break;
745 	case CB_COLOR0_CMASK_SLICE:
746 	case CB_COLOR1_CMASK_SLICE:
747 	case CB_COLOR2_CMASK_SLICE:
748 	case CB_COLOR3_CMASK_SLICE:
749 	case CB_COLOR4_CMASK_SLICE:
750 	case CB_COLOR5_CMASK_SLICE:
751 	case CB_COLOR6_CMASK_SLICE:
752 	case CB_COLOR7_CMASK_SLICE:
753 		tmp = (reg - CB_COLOR0_CMASK_SLICE) / 0x3c;
754 		track->cb_color_cmask_slice[tmp] = radeon_get_ib_value(p, idx);
755 		break;
756 	case CB_COLOR0_BASE:
757 	case CB_COLOR1_BASE:
758 	case CB_COLOR2_BASE:
759 	case CB_COLOR3_BASE:
760 	case CB_COLOR4_BASE:
761 	case CB_COLOR5_BASE:
762 	case CB_COLOR6_BASE:
763 	case CB_COLOR7_BASE:
764 		r = evergreen_cs_packet_next_reloc(p, &reloc);
765 		if (r) {
766 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
767 					"0x%04X\n", reg);
768 			return -EINVAL;
769 		}
770 		tmp = (reg - CB_COLOR0_BASE) / 0x3c;
771 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
772 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
773 		track->cb_color_base_last[tmp] = ib[idx];
774 		track->cb_color_bo[tmp] = reloc->robj;
775 		break;
776 	case CB_COLOR8_BASE:
777 	case CB_COLOR9_BASE:
778 	case CB_COLOR10_BASE:
779 	case CB_COLOR11_BASE:
780 		r = evergreen_cs_packet_next_reloc(p, &reloc);
781 		if (r) {
782 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
783 					"0x%04X\n", reg);
784 			return -EINVAL;
785 		}
786 		tmp = ((reg - CB_COLOR8_BASE) / 0x1c) + 8;
787 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
788 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
789 		track->cb_color_base_last[tmp] = ib[idx];
790 		track->cb_color_bo[tmp] = reloc->robj;
791 		break;
792 	case CB_IMMED0_BASE:
793 	case CB_IMMED1_BASE:
794 	case CB_IMMED2_BASE:
795 	case CB_IMMED3_BASE:
796 	case CB_IMMED4_BASE:
797 	case CB_IMMED5_BASE:
798 	case CB_IMMED6_BASE:
799 	case CB_IMMED7_BASE:
800 	case CB_IMMED8_BASE:
801 	case CB_IMMED9_BASE:
802 	case CB_IMMED10_BASE:
803 	case CB_IMMED11_BASE:
804 	case DB_HTILE_DATA_BASE:
805 	case SQ_PGM_START_FS:
806 	case SQ_PGM_START_ES:
807 	case SQ_PGM_START_VS:
808 	case SQ_PGM_START_GS:
809 	case SQ_PGM_START_PS:
810 	case SQ_PGM_START_HS:
811 	case SQ_PGM_START_LS:
812 	case GDS_ADDR_BASE:
813 	case SQ_CONST_MEM_BASE:
814 	case SQ_ALU_CONST_CACHE_GS_0:
815 	case SQ_ALU_CONST_CACHE_GS_1:
816 	case SQ_ALU_CONST_CACHE_GS_2:
817 	case SQ_ALU_CONST_CACHE_GS_3:
818 	case SQ_ALU_CONST_CACHE_GS_4:
819 	case SQ_ALU_CONST_CACHE_GS_5:
820 	case SQ_ALU_CONST_CACHE_GS_6:
821 	case SQ_ALU_CONST_CACHE_GS_7:
822 	case SQ_ALU_CONST_CACHE_GS_8:
823 	case SQ_ALU_CONST_CACHE_GS_9:
824 	case SQ_ALU_CONST_CACHE_GS_10:
825 	case SQ_ALU_CONST_CACHE_GS_11:
826 	case SQ_ALU_CONST_CACHE_GS_12:
827 	case SQ_ALU_CONST_CACHE_GS_13:
828 	case SQ_ALU_CONST_CACHE_GS_14:
829 	case SQ_ALU_CONST_CACHE_GS_15:
830 	case SQ_ALU_CONST_CACHE_PS_0:
831 	case SQ_ALU_CONST_CACHE_PS_1:
832 	case SQ_ALU_CONST_CACHE_PS_2:
833 	case SQ_ALU_CONST_CACHE_PS_3:
834 	case SQ_ALU_CONST_CACHE_PS_4:
835 	case SQ_ALU_CONST_CACHE_PS_5:
836 	case SQ_ALU_CONST_CACHE_PS_6:
837 	case SQ_ALU_CONST_CACHE_PS_7:
838 	case SQ_ALU_CONST_CACHE_PS_8:
839 	case SQ_ALU_CONST_CACHE_PS_9:
840 	case SQ_ALU_CONST_CACHE_PS_10:
841 	case SQ_ALU_CONST_CACHE_PS_11:
842 	case SQ_ALU_CONST_CACHE_PS_12:
843 	case SQ_ALU_CONST_CACHE_PS_13:
844 	case SQ_ALU_CONST_CACHE_PS_14:
845 	case SQ_ALU_CONST_CACHE_PS_15:
846 	case SQ_ALU_CONST_CACHE_VS_0:
847 	case SQ_ALU_CONST_CACHE_VS_1:
848 	case SQ_ALU_CONST_CACHE_VS_2:
849 	case SQ_ALU_CONST_CACHE_VS_3:
850 	case SQ_ALU_CONST_CACHE_VS_4:
851 	case SQ_ALU_CONST_CACHE_VS_5:
852 	case SQ_ALU_CONST_CACHE_VS_6:
853 	case SQ_ALU_CONST_CACHE_VS_7:
854 	case SQ_ALU_CONST_CACHE_VS_8:
855 	case SQ_ALU_CONST_CACHE_VS_9:
856 	case SQ_ALU_CONST_CACHE_VS_10:
857 	case SQ_ALU_CONST_CACHE_VS_11:
858 	case SQ_ALU_CONST_CACHE_VS_12:
859 	case SQ_ALU_CONST_CACHE_VS_13:
860 	case SQ_ALU_CONST_CACHE_VS_14:
861 	case SQ_ALU_CONST_CACHE_VS_15:
862 	case SQ_ALU_CONST_CACHE_HS_0:
863 	case SQ_ALU_CONST_CACHE_HS_1:
864 	case SQ_ALU_CONST_CACHE_HS_2:
865 	case SQ_ALU_CONST_CACHE_HS_3:
866 	case SQ_ALU_CONST_CACHE_HS_4:
867 	case SQ_ALU_CONST_CACHE_HS_5:
868 	case SQ_ALU_CONST_CACHE_HS_6:
869 	case SQ_ALU_CONST_CACHE_HS_7:
870 	case SQ_ALU_CONST_CACHE_HS_8:
871 	case SQ_ALU_CONST_CACHE_HS_9:
872 	case SQ_ALU_CONST_CACHE_HS_10:
873 	case SQ_ALU_CONST_CACHE_HS_11:
874 	case SQ_ALU_CONST_CACHE_HS_12:
875 	case SQ_ALU_CONST_CACHE_HS_13:
876 	case SQ_ALU_CONST_CACHE_HS_14:
877 	case SQ_ALU_CONST_CACHE_HS_15:
878 	case SQ_ALU_CONST_CACHE_LS_0:
879 	case SQ_ALU_CONST_CACHE_LS_1:
880 	case SQ_ALU_CONST_CACHE_LS_2:
881 	case SQ_ALU_CONST_CACHE_LS_3:
882 	case SQ_ALU_CONST_CACHE_LS_4:
883 	case SQ_ALU_CONST_CACHE_LS_5:
884 	case SQ_ALU_CONST_CACHE_LS_6:
885 	case SQ_ALU_CONST_CACHE_LS_7:
886 	case SQ_ALU_CONST_CACHE_LS_8:
887 	case SQ_ALU_CONST_CACHE_LS_9:
888 	case SQ_ALU_CONST_CACHE_LS_10:
889 	case SQ_ALU_CONST_CACHE_LS_11:
890 	case SQ_ALU_CONST_CACHE_LS_12:
891 	case SQ_ALU_CONST_CACHE_LS_13:
892 	case SQ_ALU_CONST_CACHE_LS_14:
893 	case SQ_ALU_CONST_CACHE_LS_15:
894 		r = evergreen_cs_packet_next_reloc(p, &reloc);
895 		if (r) {
896 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
897 					"0x%04X\n", reg);
898 			return -EINVAL;
899 		}
900 		ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
901 		break;
902 	default:
903 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
904 		return -EINVAL;
905 	}
906 	return 0;
907 }
908 
909 /**
910  * evergreen_check_texture_resource() - check if register is authorized or not
911  * @p: parser structure holding parsing context
912  * @idx: index into the cs buffer
913  * @texture: texture's bo structure
914  * @mipmap: mipmap's bo structure
915  *
916  * This function will check that the resource has valid field and that
917  * the texture and mipmap bo object are big enough to cover this resource.
918  */
919 static inline int evergreen_check_texture_resource(struct radeon_cs_parser *p,  u32 idx,
920 						   struct radeon_bo *texture,
921 						   struct radeon_bo *mipmap)
922 {
923 	/* XXX fill in */
924 	return 0;
925 }
926 
927 static int evergreen_packet3_check(struct radeon_cs_parser *p,
928 				   struct radeon_cs_packet *pkt)
929 {
930 	struct radeon_cs_reloc *reloc;
931 	struct evergreen_cs_track *track;
932 	volatile u32 *ib;
933 	unsigned idx;
934 	unsigned i;
935 	unsigned start_reg, end_reg, reg;
936 	int r;
937 	u32 idx_value;
938 
939 	track = (struct evergreen_cs_track *)p->track;
940 	ib = p->ib->ptr;
941 	idx = pkt->idx + 1;
942 	idx_value = radeon_get_ib_value(p, idx);
943 
944 	switch (pkt->opcode) {
945 	case PACKET3_CONTEXT_CONTROL:
946 		if (pkt->count != 1) {
947 			DRM_ERROR("bad CONTEXT_CONTROL\n");
948 			return -EINVAL;
949 		}
950 		break;
951 	case PACKET3_INDEX_TYPE:
952 	case PACKET3_NUM_INSTANCES:
953 	case PACKET3_CLEAR_STATE:
954 		if (pkt->count) {
955 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
956 			return -EINVAL;
957 		}
958 		break;
959 	case PACKET3_INDEX_BASE:
960 		if (pkt->count != 1) {
961 			DRM_ERROR("bad INDEX_BASE\n");
962 			return -EINVAL;
963 		}
964 		r = evergreen_cs_packet_next_reloc(p, &reloc);
965 		if (r) {
966 			DRM_ERROR("bad INDEX_BASE\n");
967 			return -EINVAL;
968 		}
969 		ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
970 		ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
971 		r = evergreen_cs_track_check(p);
972 		if (r) {
973 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
974 			return r;
975 		}
976 		break;
977 	case PACKET3_DRAW_INDEX:
978 		if (pkt->count != 3) {
979 			DRM_ERROR("bad DRAW_INDEX\n");
980 			return -EINVAL;
981 		}
982 		r = evergreen_cs_packet_next_reloc(p, &reloc);
983 		if (r) {
984 			DRM_ERROR("bad DRAW_INDEX\n");
985 			return -EINVAL;
986 		}
987 		ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
988 		ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
989 		r = evergreen_cs_track_check(p);
990 		if (r) {
991 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
992 			return r;
993 		}
994 		break;
995 	case PACKET3_DRAW_INDEX_2:
996 		if (pkt->count != 4) {
997 			DRM_ERROR("bad DRAW_INDEX_2\n");
998 			return -EINVAL;
999 		}
1000 		r = evergreen_cs_packet_next_reloc(p, &reloc);
1001 		if (r) {
1002 			DRM_ERROR("bad DRAW_INDEX_2\n");
1003 			return -EINVAL;
1004 		}
1005 		ib[idx+1] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1006 		ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1007 		r = evergreen_cs_track_check(p);
1008 		if (r) {
1009 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1010 			return r;
1011 		}
1012 		break;
1013 	case PACKET3_DRAW_INDEX_AUTO:
1014 		if (pkt->count != 1) {
1015 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1016 			return -EINVAL;
1017 		}
1018 		r = evergreen_cs_track_check(p);
1019 		if (r) {
1020 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1021 			return r;
1022 		}
1023 		break;
1024 	case PACKET3_DRAW_INDEX_MULTI_AUTO:
1025 		if (pkt->count != 2) {
1026 			DRM_ERROR("bad DRAW_INDEX_MULTI_AUTO\n");
1027 			return -EINVAL;
1028 		}
1029 		r = evergreen_cs_track_check(p);
1030 		if (r) {
1031 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1032 			return r;
1033 		}
1034 		break;
1035 	case PACKET3_DRAW_INDEX_IMMD:
1036 		if (pkt->count < 2) {
1037 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1038 			return -EINVAL;
1039 		}
1040 		r = evergreen_cs_track_check(p);
1041 		if (r) {
1042 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1043 			return r;
1044 		}
1045 		break;
1046 	case PACKET3_DRAW_INDEX_OFFSET:
1047 		if (pkt->count != 2) {
1048 			DRM_ERROR("bad DRAW_INDEX_OFFSET\n");
1049 			return -EINVAL;
1050 		}
1051 		r = evergreen_cs_track_check(p);
1052 		if (r) {
1053 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1054 			return r;
1055 		}
1056 		break;
1057 	case PACKET3_DRAW_INDEX_OFFSET_2:
1058 		if (pkt->count != 3) {
1059 			DRM_ERROR("bad DRAW_INDEX_OFFSET_2\n");
1060 			return -EINVAL;
1061 		}
1062 		r = evergreen_cs_track_check(p);
1063 		if (r) {
1064 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1065 			return r;
1066 		}
1067 		break;
1068 	case PACKET3_WAIT_REG_MEM:
1069 		if (pkt->count != 5) {
1070 			DRM_ERROR("bad WAIT_REG_MEM\n");
1071 			return -EINVAL;
1072 		}
1073 		/* bit 4 is reg (0) or mem (1) */
1074 		if (idx_value & 0x10) {
1075 			r = evergreen_cs_packet_next_reloc(p, &reloc);
1076 			if (r) {
1077 				DRM_ERROR("bad WAIT_REG_MEM\n");
1078 				return -EINVAL;
1079 			}
1080 			ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1081 			ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1082 		}
1083 		break;
1084 	case PACKET3_SURFACE_SYNC:
1085 		if (pkt->count != 3) {
1086 			DRM_ERROR("bad SURFACE_SYNC\n");
1087 			return -EINVAL;
1088 		}
1089 		/* 0xffffffff/0x0 is flush all cache flag */
1090 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
1091 		    radeon_get_ib_value(p, idx + 2) != 0) {
1092 			r = evergreen_cs_packet_next_reloc(p, &reloc);
1093 			if (r) {
1094 				DRM_ERROR("bad SURFACE_SYNC\n");
1095 				return -EINVAL;
1096 			}
1097 			ib[idx+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1098 		}
1099 		break;
1100 	case PACKET3_EVENT_WRITE:
1101 		if (pkt->count != 2 && pkt->count != 0) {
1102 			DRM_ERROR("bad EVENT_WRITE\n");
1103 			return -EINVAL;
1104 		}
1105 		if (pkt->count) {
1106 			r = evergreen_cs_packet_next_reloc(p, &reloc);
1107 			if (r) {
1108 				DRM_ERROR("bad EVENT_WRITE\n");
1109 				return -EINVAL;
1110 			}
1111 			ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1112 			ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1113 		}
1114 		break;
1115 	case PACKET3_EVENT_WRITE_EOP:
1116 		if (pkt->count != 4) {
1117 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1118 			return -EINVAL;
1119 		}
1120 		r = evergreen_cs_packet_next_reloc(p, &reloc);
1121 		if (r) {
1122 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
1123 			return -EINVAL;
1124 		}
1125 		ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1126 		ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1127 		break;
1128 	case PACKET3_EVENT_WRITE_EOS:
1129 		if (pkt->count != 3) {
1130 			DRM_ERROR("bad EVENT_WRITE_EOS\n");
1131 			return -EINVAL;
1132 		}
1133 		r = evergreen_cs_packet_next_reloc(p, &reloc);
1134 		if (r) {
1135 			DRM_ERROR("bad EVENT_WRITE_EOS\n");
1136 			return -EINVAL;
1137 		}
1138 		ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
1139 		ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1140 		break;
1141 	case PACKET3_SET_CONFIG_REG:
1142 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
1143 		end_reg = 4 * pkt->count + start_reg - 4;
1144 		if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
1145 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
1146 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
1147 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
1148 			return -EINVAL;
1149 		}
1150 		for (i = 0; i < pkt->count; i++) {
1151 			reg = start_reg + (4 * i);
1152 			r = evergreen_cs_check_reg(p, reg, idx+1+i);
1153 			if (r)
1154 				return r;
1155 		}
1156 		break;
1157 	case PACKET3_SET_CONTEXT_REG:
1158 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_START;
1159 		end_reg = 4 * pkt->count + start_reg - 4;
1160 		if ((start_reg < PACKET3_SET_CONTEXT_REG_START) ||
1161 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
1162 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
1163 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
1164 			return -EINVAL;
1165 		}
1166 		for (i = 0; i < pkt->count; i++) {
1167 			reg = start_reg + (4 * i);
1168 			r = evergreen_cs_check_reg(p, reg, idx+1+i);
1169 			if (r)
1170 				return r;
1171 		}
1172 		break;
1173 	case PACKET3_SET_RESOURCE:
1174 		if (pkt->count % 8) {
1175 			DRM_ERROR("bad SET_RESOURCE\n");
1176 			return -EINVAL;
1177 		}
1178 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_START;
1179 		end_reg = 4 * pkt->count + start_reg - 4;
1180 		if ((start_reg < PACKET3_SET_RESOURCE_START) ||
1181 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
1182 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
1183 			DRM_ERROR("bad SET_RESOURCE\n");
1184 			return -EINVAL;
1185 		}
1186 		for (i = 0; i < (pkt->count / 8); i++) {
1187 			struct radeon_bo *texture, *mipmap;
1188 			u32 size, offset;
1189 
1190 			switch (G__SQ_CONSTANT_TYPE(radeon_get_ib_value(p, idx+1+(i*8)+7))) {
1191 			case SQ_TEX_VTX_VALID_TEXTURE:
1192 				/* tex base */
1193 				r = evergreen_cs_packet_next_reloc(p, &reloc);
1194 				if (r) {
1195 					DRM_ERROR("bad SET_RESOURCE (tex)\n");
1196 					return -EINVAL;
1197 				}
1198 				ib[idx+1+(i*8)+2] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1199 				if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1200 					ib[idx+1+(i*8)+1] |= TEX_ARRAY_MODE(ARRAY_2D_TILED_THIN1);
1201 				else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1202 					ib[idx+1+(i*8)+1] |= TEX_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
1203 				texture = reloc->robj;
1204 				/* tex mip base */
1205 				r = evergreen_cs_packet_next_reloc(p, &reloc);
1206 				if (r) {
1207 					DRM_ERROR("bad SET_RESOURCE (tex)\n");
1208 					return -EINVAL;
1209 				}
1210 				ib[idx+1+(i*8)+3] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
1211 				mipmap = reloc->robj;
1212 				r = evergreen_check_texture_resource(p,  idx+1+(i*8),
1213 						texture, mipmap);
1214 				if (r)
1215 					return r;
1216 				break;
1217 			case SQ_TEX_VTX_VALID_BUFFER:
1218 				/* vtx base */
1219 				r = evergreen_cs_packet_next_reloc(p, &reloc);
1220 				if (r) {
1221 					DRM_ERROR("bad SET_RESOURCE (vtx)\n");
1222 					return -EINVAL;
1223 				}
1224 				offset = radeon_get_ib_value(p, idx+1+(i*8)+0);
1225 				size = radeon_get_ib_value(p, idx+1+(i*8)+1);
1226 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
1227 					/* force size to size of the buffer */
1228 					dev_warn(p->dev, "vbo resource seems too big for the bo\n");
1229 					ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj);
1230 				}
1231 				ib[idx+1+(i*8)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
1232 				ib[idx+1+(i*8)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
1233 				break;
1234 			case SQ_TEX_VTX_INVALID_TEXTURE:
1235 			case SQ_TEX_VTX_INVALID_BUFFER:
1236 			default:
1237 				DRM_ERROR("bad SET_RESOURCE\n");
1238 				return -EINVAL;
1239 			}
1240 		}
1241 		break;
1242 	case PACKET3_SET_ALU_CONST:
1243 		/* XXX fix me ALU const buffers only */
1244 		break;
1245 	case PACKET3_SET_BOOL_CONST:
1246 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_START;
1247 		end_reg = 4 * pkt->count + start_reg - 4;
1248 		if ((start_reg < PACKET3_SET_BOOL_CONST_START) ||
1249 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
1250 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
1251 			DRM_ERROR("bad SET_BOOL_CONST\n");
1252 			return -EINVAL;
1253 		}
1254 		break;
1255 	case PACKET3_SET_LOOP_CONST:
1256 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_START;
1257 		end_reg = 4 * pkt->count + start_reg - 4;
1258 		if ((start_reg < PACKET3_SET_LOOP_CONST_START) ||
1259 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
1260 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
1261 			DRM_ERROR("bad SET_LOOP_CONST\n");
1262 			return -EINVAL;
1263 		}
1264 		break;
1265 	case PACKET3_SET_CTL_CONST:
1266 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_START;
1267 		end_reg = 4 * pkt->count + start_reg - 4;
1268 		if ((start_reg < PACKET3_SET_CTL_CONST_START) ||
1269 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
1270 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
1271 			DRM_ERROR("bad SET_CTL_CONST\n");
1272 			return -EINVAL;
1273 		}
1274 		break;
1275 	case PACKET3_SET_SAMPLER:
1276 		if (pkt->count % 3) {
1277 			DRM_ERROR("bad SET_SAMPLER\n");
1278 			return -EINVAL;
1279 		}
1280 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_START;
1281 		end_reg = 4 * pkt->count + start_reg - 4;
1282 		if ((start_reg < PACKET3_SET_SAMPLER_START) ||
1283 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
1284 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
1285 			DRM_ERROR("bad SET_SAMPLER\n");
1286 			return -EINVAL;
1287 		}
1288 		break;
1289 	case PACKET3_NOP:
1290 		break;
1291 	default:
1292 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1293 		return -EINVAL;
1294 	}
1295 	return 0;
1296 }
1297 
1298 int evergreen_cs_parse(struct radeon_cs_parser *p)
1299 {
1300 	struct radeon_cs_packet pkt;
1301 	struct evergreen_cs_track *track;
1302 	int r;
1303 
1304 	if (p->track == NULL) {
1305 		/* initialize tracker, we are in kms */
1306 		track = kzalloc(sizeof(*track), GFP_KERNEL);
1307 		if (track == NULL)
1308 			return -ENOMEM;
1309 		evergreen_cs_track_init(track);
1310 		track->npipes = p->rdev->config.evergreen.tiling_npipes;
1311 		track->nbanks = p->rdev->config.evergreen.tiling_nbanks;
1312 		track->group_size = p->rdev->config.evergreen.tiling_group_size;
1313 		p->track = track;
1314 	}
1315 	do {
1316 		r = evergreen_cs_packet_parse(p, &pkt, p->idx);
1317 		if (r) {
1318 			kfree(p->track);
1319 			p->track = NULL;
1320 			return r;
1321 		}
1322 		p->idx += pkt.count + 2;
1323 		switch (pkt.type) {
1324 		case PACKET_TYPE0:
1325 			r = evergreen_cs_parse_packet0(p, &pkt);
1326 			break;
1327 		case PACKET_TYPE2:
1328 			break;
1329 		case PACKET_TYPE3:
1330 			r = evergreen_packet3_check(p, &pkt);
1331 			break;
1332 		default:
1333 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1334 			kfree(p->track);
1335 			p->track = NULL;
1336 			return -EINVAL;
1337 		}
1338 		if (r) {
1339 			kfree(p->track);
1340 			p->track = NULL;
1341 			return r;
1342 		}
1343 	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1344 #if 0
1345 	for (r = 0; r < p->ib->length_dw; r++) {
1346 		printk(KERN_INFO "%05d  0x%08X\n", r, p->ib->ptr[r]);
1347 		mdelay(1);
1348 	}
1349 #endif
1350 	kfree(p->track);
1351 	p->track = NULL;
1352 	return 0;
1353 }
1354 
1355