xref: /linux/drivers/gpu/drm/radeon/radeon_cs.c (revision 9e8ba5f3ec35cba4fd8a8bebda548c4db2651e40)
1 /*
2  * Copyright 2008 Jerome Glisse.
3  * All Rights Reserved.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * 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  * PRECISION INSIGHT AND/OR ITS SUPPLIERS 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 OTHER
22  * DEALINGS IN THE SOFTWARE.
23  *
24  * Authors:
25  *    Jerome Glisse <glisse@freedesktop.org>
26  */
27 #include "drmP.h"
28 #include "radeon_drm.h"
29 #include "radeon_reg.h"
30 #include "radeon.h"
31 
32 void r100_cs_dump_packet(struct radeon_cs_parser *p,
33 			 struct radeon_cs_packet *pkt);
34 
35 int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
36 {
37 	struct drm_device *ddev = p->rdev->ddev;
38 	struct radeon_cs_chunk *chunk;
39 	unsigned i, j;
40 	bool duplicate;
41 
42 	if (p->chunk_relocs_idx == -1) {
43 		return 0;
44 	}
45 	chunk = &p->chunks[p->chunk_relocs_idx];
46 	/* FIXME: we assume that each relocs use 4 dwords */
47 	p->nrelocs = chunk->length_dw / 4;
48 	p->relocs_ptr = kcalloc(p->nrelocs, sizeof(void *), GFP_KERNEL);
49 	if (p->relocs_ptr == NULL) {
50 		return -ENOMEM;
51 	}
52 	p->relocs = kcalloc(p->nrelocs, sizeof(struct radeon_cs_reloc), GFP_KERNEL);
53 	if (p->relocs == NULL) {
54 		return -ENOMEM;
55 	}
56 	for (i = 0; i < p->nrelocs; i++) {
57 		struct drm_radeon_cs_reloc *r;
58 
59 		duplicate = false;
60 		r = (struct drm_radeon_cs_reloc *)&chunk->kdata[i*4];
61 		for (j = 0; j < p->nrelocs; j++) {
62 			if (r->handle == p->relocs[j].handle) {
63 				p->relocs_ptr[i] = &p->relocs[j];
64 				duplicate = true;
65 				break;
66 			}
67 		}
68 		if (!duplicate) {
69 			p->relocs[i].gobj = drm_gem_object_lookup(ddev,
70 								  p->filp,
71 								  r->handle);
72 			if (p->relocs[i].gobj == NULL) {
73 				DRM_ERROR("gem object lookup failed 0x%x\n",
74 					  r->handle);
75 				return -ENOENT;
76 			}
77 			p->relocs_ptr[i] = &p->relocs[i];
78 			p->relocs[i].robj = gem_to_radeon_bo(p->relocs[i].gobj);
79 			p->relocs[i].lobj.bo = p->relocs[i].robj;
80 			p->relocs[i].lobj.wdomain = r->write_domain;
81 			p->relocs[i].lobj.rdomain = r->read_domains;
82 			p->relocs[i].lobj.tv.bo = &p->relocs[i].robj->tbo;
83 			p->relocs[i].handle = r->handle;
84 			p->relocs[i].flags = r->flags;
85 			radeon_bo_list_add_object(&p->relocs[i].lobj,
86 						  &p->validated);
87 		}
88 	}
89 	return radeon_bo_list_validate(&p->validated);
90 }
91 
92 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
93 {
94 	struct drm_radeon_cs *cs = data;
95 	uint64_t *chunk_array_ptr;
96 	unsigned size, i, flags = 0;
97 
98 	if (!cs->num_chunks) {
99 		return 0;
100 	}
101 	/* get chunks */
102 	INIT_LIST_HEAD(&p->validated);
103 	p->idx = 0;
104 	p->chunk_ib_idx = -1;
105 	p->chunk_relocs_idx = -1;
106 	p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
107 	if (p->chunks_array == NULL) {
108 		return -ENOMEM;
109 	}
110 	chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks);
111 	if (DRM_COPY_FROM_USER(p->chunks_array, chunk_array_ptr,
112 			       sizeof(uint64_t)*cs->num_chunks)) {
113 		return -EFAULT;
114 	}
115 	p->nchunks = cs->num_chunks;
116 	p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
117 	if (p->chunks == NULL) {
118 		return -ENOMEM;
119 	}
120 	for (i = 0; i < p->nchunks; i++) {
121 		struct drm_radeon_cs_chunk __user **chunk_ptr = NULL;
122 		struct drm_radeon_cs_chunk user_chunk;
123 		uint32_t __user *cdata;
124 
125 		chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i];
126 		if (DRM_COPY_FROM_USER(&user_chunk, chunk_ptr,
127 				       sizeof(struct drm_radeon_cs_chunk))) {
128 			return -EFAULT;
129 		}
130 		p->chunks[i].length_dw = user_chunk.length_dw;
131 		p->chunks[i].kdata = NULL;
132 		p->chunks[i].chunk_id = user_chunk.chunk_id;
133 
134 		if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) {
135 			p->chunk_relocs_idx = i;
136 		}
137 		if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_IB) {
138 			p->chunk_ib_idx = i;
139 			/* zero length IB isn't useful */
140 			if (p->chunks[i].length_dw == 0)
141 				return -EINVAL;
142 		}
143 		if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS &&
144 		    !p->chunks[i].length_dw) {
145 			return -EINVAL;
146 		}
147 
148 		p->chunks[i].length_dw = user_chunk.length_dw;
149 		p->chunks[i].user_ptr = (void __user *)(unsigned long)user_chunk.chunk_data;
150 
151 		cdata = (uint32_t *)(unsigned long)user_chunk.chunk_data;
152 		if (p->chunks[i].chunk_id != RADEON_CHUNK_ID_IB) {
153 			size = p->chunks[i].length_dw * sizeof(uint32_t);
154 			p->chunks[i].kdata = kmalloc(size, GFP_KERNEL);
155 			if (p->chunks[i].kdata == NULL) {
156 				return -ENOMEM;
157 			}
158 			if (DRM_COPY_FROM_USER(p->chunks[i].kdata,
159 					       p->chunks[i].user_ptr, size)) {
160 				return -EFAULT;
161 			}
162 			if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS) {
163 				flags = p->chunks[i].kdata[0];
164 			}
165 		} else {
166 			p->chunks[i].kpage[0] = kmalloc(PAGE_SIZE, GFP_KERNEL);
167 			p->chunks[i].kpage[1] = kmalloc(PAGE_SIZE, GFP_KERNEL);
168 			if (p->chunks[i].kpage[0] == NULL || p->chunks[i].kpage[1] == NULL) {
169 				kfree(p->chunks[i].kpage[0]);
170 				kfree(p->chunks[i].kpage[1]);
171 				return -ENOMEM;
172 			}
173 			p->chunks[i].kpage_idx[0] = -1;
174 			p->chunks[i].kpage_idx[1] = -1;
175 			p->chunks[i].last_copied_page = -1;
176 			p->chunks[i].last_page_index = ((p->chunks[i].length_dw * 4) - 1) / PAGE_SIZE;
177 		}
178 	}
179 	if (p->chunks[p->chunk_ib_idx].length_dw > (16 * 1024)) {
180 		DRM_ERROR("cs IB too big: %d\n",
181 			  p->chunks[p->chunk_ib_idx].length_dw);
182 		return -EINVAL;
183 	}
184 
185 	p->keep_tiling_flags = (flags & RADEON_CS_KEEP_TILING_FLAGS) != 0;
186 	return 0;
187 }
188 
189 /**
190  * cs_parser_fini() - clean parser states
191  * @parser:	parser structure holding parsing context.
192  * @error:	error number
193  *
194  * If error is set than unvalidate buffer, otherwise just free memory
195  * used by parsing context.
196  **/
197 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
198 {
199 	unsigned i;
200 
201 
202 	if (!error && parser->ib)
203 		ttm_eu_fence_buffer_objects(&parser->validated,
204 					    parser->ib->fence);
205 	else
206 		ttm_eu_backoff_reservation(&parser->validated);
207 
208 	if (parser->relocs != NULL) {
209 		for (i = 0; i < parser->nrelocs; i++) {
210 			if (parser->relocs[i].gobj)
211 				drm_gem_object_unreference_unlocked(parser->relocs[i].gobj);
212 		}
213 	}
214 	kfree(parser->track);
215 	kfree(parser->relocs);
216 	kfree(parser->relocs_ptr);
217 	for (i = 0; i < parser->nchunks; i++) {
218 		kfree(parser->chunks[i].kdata);
219 		kfree(parser->chunks[i].kpage[0]);
220 		kfree(parser->chunks[i].kpage[1]);
221 	}
222 	kfree(parser->chunks);
223 	kfree(parser->chunks_array);
224 	radeon_ib_free(parser->rdev, &parser->ib);
225 }
226 
227 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
228 {
229 	struct radeon_device *rdev = dev->dev_private;
230 	struct radeon_cs_parser parser;
231 	struct radeon_cs_chunk *ib_chunk;
232 	int r;
233 
234 	radeon_mutex_lock(&rdev->cs_mutex);
235 	/* initialize parser */
236 	memset(&parser, 0, sizeof(struct radeon_cs_parser));
237 	parser.filp = filp;
238 	parser.rdev = rdev;
239 	parser.dev = rdev->dev;
240 	parser.family = rdev->family;
241 	r = radeon_cs_parser_init(&parser, data);
242 	if (r) {
243 		DRM_ERROR("Failed to initialize parser !\n");
244 		radeon_cs_parser_fini(&parser, r);
245 		radeon_mutex_unlock(&rdev->cs_mutex);
246 		return r;
247 	}
248 	r =  radeon_ib_get(rdev, &parser.ib);
249 	if (r) {
250 		DRM_ERROR("Failed to get ib !\n");
251 		radeon_cs_parser_fini(&parser, r);
252 		radeon_mutex_unlock(&rdev->cs_mutex);
253 		return r;
254 	}
255 	r = radeon_cs_parser_relocs(&parser);
256 	if (r) {
257 		if (r != -ERESTARTSYS)
258 			DRM_ERROR("Failed to parse relocation %d!\n", r);
259 		radeon_cs_parser_fini(&parser, r);
260 		radeon_mutex_unlock(&rdev->cs_mutex);
261 		return r;
262 	}
263 	/* Copy the packet into the IB, the parser will read from the
264 	 * input memory (cached) and write to the IB (which can be
265 	 * uncached). */
266 	ib_chunk = &parser.chunks[parser.chunk_ib_idx];
267 	parser.ib->length_dw = ib_chunk->length_dw;
268 	r = radeon_cs_parse(&parser);
269 	if (r || parser.parser_error) {
270 		DRM_ERROR("Invalid command stream !\n");
271 		radeon_cs_parser_fini(&parser, r);
272 		radeon_mutex_unlock(&rdev->cs_mutex);
273 		return r;
274 	}
275 	r = radeon_cs_finish_pages(&parser);
276 	if (r) {
277 		DRM_ERROR("Invalid command stream !\n");
278 		radeon_cs_parser_fini(&parser, r);
279 		radeon_mutex_unlock(&rdev->cs_mutex);
280 		return r;
281 	}
282 	r = radeon_ib_schedule(rdev, parser.ib);
283 	if (r) {
284 		DRM_ERROR("Failed to schedule IB !\n");
285 	}
286 	radeon_cs_parser_fini(&parser, r);
287 	radeon_mutex_unlock(&rdev->cs_mutex);
288 	return r;
289 }
290 
291 int radeon_cs_finish_pages(struct radeon_cs_parser *p)
292 {
293 	struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
294 	int i;
295 	int size = PAGE_SIZE;
296 
297 	for (i = ibc->last_copied_page + 1; i <= ibc->last_page_index; i++) {
298 		if (i == ibc->last_page_index) {
299 			size = (ibc->length_dw * 4) % PAGE_SIZE;
300 			if (size == 0)
301 				size = PAGE_SIZE;
302 		}
303 
304 		if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
305 				       ibc->user_ptr + (i * PAGE_SIZE),
306 				       size))
307 			return -EFAULT;
308 	}
309 	return 0;
310 }
311 
312 int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
313 {
314 	int new_page;
315 	struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
316 	int i;
317 	int size = PAGE_SIZE;
318 
319 	for (i = ibc->last_copied_page + 1; i < pg_idx; i++) {
320 		if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
321 				       ibc->user_ptr + (i * PAGE_SIZE),
322 				       PAGE_SIZE)) {
323 			p->parser_error = -EFAULT;
324 			return 0;
325 		}
326 	}
327 
328 	new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1;
329 
330 	if (pg_idx == ibc->last_page_index) {
331 		size = (ibc->length_dw * 4) % PAGE_SIZE;
332 			if (size == 0)
333 				size = PAGE_SIZE;
334 	}
335 
336 	if (DRM_COPY_FROM_USER(ibc->kpage[new_page],
337 			       ibc->user_ptr + (pg_idx * PAGE_SIZE),
338 			       size)) {
339 		p->parser_error = -EFAULT;
340 		return 0;
341 	}
342 
343 	/* copy to IB here */
344 	memcpy((void *)(p->ib->ptr+(pg_idx*(PAGE_SIZE/4))), ibc->kpage[new_page], size);
345 
346 	ibc->last_copied_page = pg_idx;
347 	ibc->kpage_idx[new_page] = pg_idx;
348 
349 	return new_page;
350 }
351