]>
Commit | Line | Data |
---|---|---|
771fe6b9 JG |
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]; | |
16557f1e | 61 | for (j = 0; j < i; j++) { |
771fe6b9 JG |
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); | |
bf79cb91 | 75 | return -ENOENT; |
771fe6b9 JG |
76 | } |
77 | p->relocs_ptr[i] = &p->relocs[i]; | |
7e4d15d9 | 78 | p->relocs[i].robj = gem_to_radeon_bo(p->relocs[i].gobj); |
4c788679 | 79 | p->relocs[i].lobj.bo = p->relocs[i].robj; |
771fe6b9 | 80 | p->relocs[i].lobj.wdomain = r->write_domain; |
147666fb TH |
81 | p->relocs[i].lobj.rdomain = r->read_domains; |
82 | p->relocs[i].lobj.tv.bo = &p->relocs[i].robj->tbo; | |
771fe6b9 JG |
83 | p->relocs[i].handle = r->handle; |
84 | p->relocs[i].flags = r->flags; | |
4c788679 | 85 | radeon_bo_list_add_object(&p->relocs[i].lobj, |
147666fb | 86 | &p->validated); |
93504fce | 87 | |
16557f1e CK |
88 | } else |
89 | p->relocs[i].handle = 0; | |
771fe6b9 | 90 | } |
94429bb6 | 91 | return radeon_bo_list_validate(&p->validated); |
771fe6b9 JG |
92 | } |
93 | ||
721604a1 JG |
94 | static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority) |
95 | { | |
96 | p->priority = priority; | |
97 | ||
98 | switch (ring) { | |
99 | default: | |
100 | DRM_ERROR("unknown ring id: %d\n", ring); | |
101 | return -EINVAL; | |
102 | case RADEON_CS_RING_GFX: | |
103 | p->ring = RADEON_RING_TYPE_GFX_INDEX; | |
104 | break; | |
105 | case RADEON_CS_RING_COMPUTE: | |
8d5ef7b1 AD |
106 | if (p->rdev->family >= CHIP_TAHITI) { |
107 | if (p->priority > 0) | |
108 | p->ring = CAYMAN_RING_TYPE_CP1_INDEX; | |
109 | else | |
110 | p->ring = CAYMAN_RING_TYPE_CP2_INDEX; | |
111 | } else | |
112 | p->ring = RADEON_RING_TYPE_GFX_INDEX; | |
721604a1 JG |
113 | break; |
114 | } | |
115 | return 0; | |
116 | } | |
117 | ||
220907d9 | 118 | static void radeon_cs_sync_rings(struct radeon_cs_parser *p) |
93504fce | 119 | { |
220907d9 | 120 | int i; |
93504fce | 121 | |
cdac5504 | 122 | for (i = 0; i < p->nrelocs; i++) { |
220907d9 | 123 | struct radeon_fence *a, *b; |
133f4cb3 | 124 | |
cdac5504 CK |
125 | if (!p->relocs[i].robj || !p->relocs[i].robj->tbo.sync_obj) |
126 | continue; | |
127 | ||
220907d9 CK |
128 | a = p->relocs[i].robj->tbo.sync_obj; |
129 | b = p->ib.sync_to[a->ring]; | |
130 | p->ib.sync_to[a->ring] = radeon_fence_later(a, b); | |
8f676c4c | 131 | } |
93504fce CK |
132 | } |
133 | ||
9b00147d | 134 | /* XXX: note that this is called from the legacy UMS CS ioctl as well */ |
771fe6b9 JG |
135 | int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data) |
136 | { | |
137 | struct drm_radeon_cs *cs = data; | |
138 | uint64_t *chunk_array_ptr; | |
721604a1 JG |
139 | unsigned size, i; |
140 | u32 ring = RADEON_CS_RING_GFX; | |
141 | s32 priority = 0; | |
771fe6b9 JG |
142 | |
143 | if (!cs->num_chunks) { | |
144 | return 0; | |
145 | } | |
146 | /* get chunks */ | |
147 | INIT_LIST_HEAD(&p->validated); | |
148 | p->idx = 0; | |
f2e39221 JG |
149 | p->ib.sa_bo = NULL; |
150 | p->ib.semaphore = NULL; | |
151 | p->const_ib.sa_bo = NULL; | |
152 | p->const_ib.semaphore = NULL; | |
771fe6b9 JG |
153 | p->chunk_ib_idx = -1; |
154 | p->chunk_relocs_idx = -1; | |
721604a1 | 155 | p->chunk_flags_idx = -1; |
dfcf5f36 | 156 | p->chunk_const_ib_idx = -1; |
771fe6b9 JG |
157 | p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL); |
158 | if (p->chunks_array == NULL) { | |
159 | return -ENOMEM; | |
160 | } | |
161 | chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks); | |
162 | if (DRM_COPY_FROM_USER(p->chunks_array, chunk_array_ptr, | |
163 | sizeof(uint64_t)*cs->num_chunks)) { | |
164 | return -EFAULT; | |
165 | } | |
721604a1 | 166 | p->cs_flags = 0; |
771fe6b9 JG |
167 | p->nchunks = cs->num_chunks; |
168 | p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL); | |
169 | if (p->chunks == NULL) { | |
170 | return -ENOMEM; | |
171 | } | |
172 | for (i = 0; i < p->nchunks; i++) { | |
173 | struct drm_radeon_cs_chunk __user **chunk_ptr = NULL; | |
174 | struct drm_radeon_cs_chunk user_chunk; | |
175 | uint32_t __user *cdata; | |
176 | ||
177 | chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i]; | |
178 | if (DRM_COPY_FROM_USER(&user_chunk, chunk_ptr, | |
179 | sizeof(struct drm_radeon_cs_chunk))) { | |
180 | return -EFAULT; | |
181 | } | |
5176fdc4 DA |
182 | p->chunks[i].length_dw = user_chunk.length_dw; |
183 | p->chunks[i].kdata = NULL; | |
771fe6b9 | 184 | p->chunks[i].chunk_id = user_chunk.chunk_id; |
5176fdc4 | 185 | |
771fe6b9 JG |
186 | if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) { |
187 | p->chunk_relocs_idx = i; | |
188 | } | |
189 | if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_IB) { | |
190 | p->chunk_ib_idx = i; | |
5176fdc4 DA |
191 | /* zero length IB isn't useful */ |
192 | if (p->chunks[i].length_dw == 0) | |
193 | return -EINVAL; | |
771fe6b9 | 194 | } |
dfcf5f36 AD |
195 | if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_CONST_IB) { |
196 | p->chunk_const_ib_idx = i; | |
197 | /* zero length CONST IB isn't useful */ | |
198 | if (p->chunks[i].length_dw == 0) | |
199 | return -EINVAL; | |
200 | } | |
721604a1 JG |
201 | if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS) { |
202 | p->chunk_flags_idx = i; | |
203 | /* zero length flags aren't useful */ | |
204 | if (p->chunks[i].length_dw == 0) | |
205 | return -EINVAL; | |
e70f224c | 206 | } |
5176fdc4 | 207 | |
771fe6b9 | 208 | p->chunks[i].length_dw = user_chunk.length_dw; |
513bcb46 | 209 | p->chunks[i].user_ptr = (void __user *)(unsigned long)user_chunk.chunk_data; |
771fe6b9 | 210 | |
513bcb46 | 211 | cdata = (uint32_t *)(unsigned long)user_chunk.chunk_data; |
721604a1 JG |
212 | if ((p->chunks[i].chunk_id == RADEON_CHUNK_ID_RELOCS) || |
213 | (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS)) { | |
513bcb46 DA |
214 | size = p->chunks[i].length_dw * sizeof(uint32_t); |
215 | p->chunks[i].kdata = kmalloc(size, GFP_KERNEL); | |
216 | if (p->chunks[i].kdata == NULL) { | |
217 | return -ENOMEM; | |
218 | } | |
219 | if (DRM_COPY_FROM_USER(p->chunks[i].kdata, | |
220 | p->chunks[i].user_ptr, size)) { | |
221 | return -EFAULT; | |
222 | } | |
e70f224c | 223 | if (p->chunks[i].chunk_id == RADEON_CHUNK_ID_FLAGS) { |
721604a1 JG |
224 | p->cs_flags = p->chunks[i].kdata[0]; |
225 | if (p->chunks[i].length_dw > 1) | |
226 | ring = p->chunks[i].kdata[1]; | |
227 | if (p->chunks[i].length_dw > 2) | |
228 | priority = (s32)p->chunks[i].kdata[2]; | |
e70f224c | 229 | } |
771fe6b9 JG |
230 | } |
231 | } | |
721604a1 | 232 | |
9b00147d AD |
233 | /* these are KMS only */ |
234 | if (p->rdev) { | |
235 | if ((p->cs_flags & RADEON_CS_USE_VM) && | |
236 | !p->rdev->vm_manager.enabled) { | |
237 | DRM_ERROR("VM not active on asic!\n"); | |
238 | return -EINVAL; | |
239 | } | |
1b5475db | 240 | |
9b00147d AD |
241 | /* we only support VM on SI+ */ |
242 | if ((p->rdev->family >= CHIP_TAHITI) && | |
243 | ((p->cs_flags & RADEON_CS_USE_VM) == 0)) { | |
244 | DRM_ERROR("VM required on SI+!\n"); | |
245 | return -EINVAL; | |
246 | } | |
721604a1 | 247 | |
9b00147d AD |
248 | if (radeon_cs_get_ring(p, ring, priority)) |
249 | return -EINVAL; | |
250 | } | |
721604a1 JG |
251 | |
252 | /* deal with non-vm */ | |
253 | if ((p->chunk_ib_idx != -1) && | |
254 | ((p->cs_flags & RADEON_CS_USE_VM) == 0) && | |
255 | (p->chunks[p->chunk_ib_idx].chunk_id == RADEON_CHUNK_ID_IB)) { | |
256 | if (p->chunks[p->chunk_ib_idx].length_dw > (16 * 1024)) { | |
257 | DRM_ERROR("cs IB too big: %d\n", | |
258 | p->chunks[p->chunk_ib_idx].length_dw); | |
259 | return -EINVAL; | |
260 | } | |
6a7068b4 DA |
261 | if ((p->rdev->flags & RADEON_IS_AGP)) { |
262 | p->chunks[p->chunk_ib_idx].kpage[0] = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
263 | p->chunks[p->chunk_ib_idx].kpage[1] = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
264 | if (p->chunks[p->chunk_ib_idx].kpage[0] == NULL || | |
265 | p->chunks[p->chunk_ib_idx].kpage[1] == NULL) { | |
266 | kfree(p->chunks[i].kpage[0]); | |
267 | kfree(p->chunks[i].kpage[1]); | |
268 | return -ENOMEM; | |
269 | } | |
270 | } | |
721604a1 JG |
271 | p->chunks[p->chunk_ib_idx].kpage_idx[0] = -1; |
272 | p->chunks[p->chunk_ib_idx].kpage_idx[1] = -1; | |
273 | p->chunks[p->chunk_ib_idx].last_copied_page = -1; | |
274 | p->chunks[p->chunk_ib_idx].last_page_index = | |
275 | ((p->chunks[p->chunk_ib_idx].length_dw * 4) - 1) / PAGE_SIZE; | |
276 | } | |
277 | ||
771fe6b9 JG |
278 | return 0; |
279 | } | |
280 | ||
281 | /** | |
282 | * cs_parser_fini() - clean parser states | |
283 | * @parser: parser structure holding parsing context. | |
284 | * @error: error number | |
285 | * | |
286 | * If error is set than unvalidate buffer, otherwise just free memory | |
287 | * used by parsing context. | |
288 | **/ | |
289 | static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error) | |
290 | { | |
291 | unsigned i; | |
292 | ||
f2e39221 | 293 | if (!error) |
147666fb | 294 | ttm_eu_fence_buffer_objects(&parser->validated, |
f2e39221 | 295 | parser->ib.fence); |
147666fb TH |
296 | else |
297 | ttm_eu_backoff_reservation(&parser->validated); | |
298 | ||
fcbc451b PN |
299 | if (parser->relocs != NULL) { |
300 | for (i = 0; i < parser->nrelocs; i++) { | |
301 | if (parser->relocs[i].gobj) | |
302 | drm_gem_object_unreference_unlocked(parser->relocs[i].gobj); | |
303 | } | |
771fe6b9 | 304 | } |
48e113e5 | 305 | kfree(parser->track); |
771fe6b9 JG |
306 | kfree(parser->relocs); |
307 | kfree(parser->relocs_ptr); | |
308 | for (i = 0; i < parser->nchunks; i++) { | |
309 | kfree(parser->chunks[i].kdata); | |
6a7068b4 DA |
310 | if ((parser->rdev->flags & RADEON_IS_AGP)) { |
311 | kfree(parser->chunks[i].kpage[0]); | |
312 | kfree(parser->chunks[i].kpage[1]); | |
313 | } | |
771fe6b9 JG |
314 | } |
315 | kfree(parser->chunks); | |
316 | kfree(parser->chunks_array); | |
317 | radeon_ib_free(parser->rdev, &parser->ib); | |
f2e39221 | 318 | radeon_ib_free(parser->rdev, &parser->const_ib); |
771fe6b9 JG |
319 | } |
320 | ||
721604a1 JG |
321 | static int radeon_cs_ib_chunk(struct radeon_device *rdev, |
322 | struct radeon_cs_parser *parser) | |
323 | { | |
324 | struct radeon_cs_chunk *ib_chunk; | |
325 | int r; | |
326 | ||
327 | if (parser->chunk_ib_idx == -1) | |
328 | return 0; | |
329 | ||
330 | if (parser->cs_flags & RADEON_CS_USE_VM) | |
331 | return 0; | |
332 | ||
333 | ib_chunk = &parser->chunks[parser->chunk_ib_idx]; | |
334 | /* Copy the packet into the IB, the parser will read from the | |
335 | * input memory (cached) and write to the IB (which can be | |
336 | * uncached). | |
337 | */ | |
338 | r = radeon_ib_get(rdev, parser->ring, &parser->ib, | |
339 | ib_chunk->length_dw * 4); | |
340 | if (r) { | |
341 | DRM_ERROR("Failed to get ib !\n"); | |
342 | return r; | |
343 | } | |
f2e39221 | 344 | parser->ib.length_dw = ib_chunk->length_dw; |
eb0c19c5 | 345 | r = radeon_cs_parse(rdev, parser->ring, parser); |
721604a1 JG |
346 | if (r || parser->parser_error) { |
347 | DRM_ERROR("Invalid command stream !\n"); | |
348 | return r; | |
349 | } | |
350 | r = radeon_cs_finish_pages(parser); | |
351 | if (r) { | |
352 | DRM_ERROR("Invalid command stream !\n"); | |
353 | return r; | |
354 | } | |
220907d9 | 355 | radeon_cs_sync_rings(parser); |
f2e39221 JG |
356 | parser->ib.vm_id = 0; |
357 | r = radeon_ib_schedule(rdev, &parser->ib); | |
721604a1 JG |
358 | if (r) { |
359 | DRM_ERROR("Failed to schedule IB !\n"); | |
360 | } | |
93bf888c | 361 | return r; |
721604a1 JG |
362 | } |
363 | ||
364 | static int radeon_bo_vm_update_pte(struct radeon_cs_parser *parser, | |
365 | struct radeon_vm *vm) | |
366 | { | |
367 | struct radeon_bo_list *lobj; | |
368 | struct radeon_bo *bo; | |
369 | int r; | |
370 | ||
371 | list_for_each_entry(lobj, &parser->validated, tv.head) { | |
372 | bo = lobj->bo; | |
373 | r = radeon_vm_bo_update_pte(parser->rdev, vm, bo, &bo->tbo.mem); | |
374 | if (r) { | |
375 | return r; | |
376 | } | |
377 | } | |
378 | return 0; | |
379 | } | |
380 | ||
381 | static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev, | |
382 | struct radeon_cs_parser *parser) | |
383 | { | |
384 | struct radeon_cs_chunk *ib_chunk; | |
385 | struct radeon_fpriv *fpriv = parser->filp->driver_priv; | |
386 | struct radeon_vm *vm = &fpriv->vm; | |
387 | int r; | |
388 | ||
389 | if (parser->chunk_ib_idx == -1) | |
390 | return 0; | |
391 | ||
392 | if ((parser->cs_flags & RADEON_CS_USE_VM) == 0) | |
393 | return 0; | |
394 | ||
dfcf5f36 AD |
395 | if ((rdev->family >= CHIP_TAHITI) && |
396 | (parser->chunk_const_ib_idx != -1)) { | |
397 | ib_chunk = &parser->chunks[parser->chunk_const_ib_idx]; | |
398 | if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) { | |
399 | DRM_ERROR("cs IB CONST too big: %d\n", ib_chunk->length_dw); | |
400 | return -EINVAL; | |
401 | } | |
402 | r = radeon_ib_get(rdev, parser->ring, &parser->const_ib, | |
403 | ib_chunk->length_dw * 4); | |
404 | if (r) { | |
405 | DRM_ERROR("Failed to get const ib !\n"); | |
406 | return r; | |
407 | } | |
f2e39221 JG |
408 | parser->const_ib.is_const_ib = true; |
409 | parser->const_ib.length_dw = ib_chunk->length_dw; | |
dfcf5f36 | 410 | /* Copy the packet into the IB */ |
f2e39221 | 411 | if (DRM_COPY_FROM_USER(parser->const_ib.ptr, ib_chunk->user_ptr, |
dfcf5f36 AD |
412 | ib_chunk->length_dw * 4)) { |
413 | return -EFAULT; | |
414 | } | |
f2e39221 | 415 | r = radeon_ring_ib_parse(rdev, parser->ring, &parser->const_ib); |
dfcf5f36 AD |
416 | if (r) { |
417 | return r; | |
418 | } | |
419 | } | |
420 | ||
721604a1 JG |
421 | ib_chunk = &parser->chunks[parser->chunk_ib_idx]; |
422 | if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) { | |
423 | DRM_ERROR("cs IB too big: %d\n", ib_chunk->length_dw); | |
424 | return -EINVAL; | |
425 | } | |
426 | r = radeon_ib_get(rdev, parser->ring, &parser->ib, | |
427 | ib_chunk->length_dw * 4); | |
428 | if (r) { | |
429 | DRM_ERROR("Failed to get ib !\n"); | |
430 | return r; | |
431 | } | |
f2e39221 | 432 | parser->ib.length_dw = ib_chunk->length_dw; |
721604a1 | 433 | /* Copy the packet into the IB */ |
f2e39221 | 434 | if (DRM_COPY_FROM_USER(parser->ib.ptr, ib_chunk->user_ptr, |
721604a1 JG |
435 | ib_chunk->length_dw * 4)) { |
436 | return -EFAULT; | |
437 | } | |
f2e39221 | 438 | r = radeon_ring_ib_parse(rdev, parser->ring, &parser->ib); |
721604a1 JG |
439 | if (r) { |
440 | return r; | |
441 | } | |
442 | ||
36ff39c4 | 443 | mutex_lock(&rdev->vm_manager.lock); |
721604a1 JG |
444 | mutex_lock(&vm->mutex); |
445 | r = radeon_vm_bind(rdev, vm); | |
446 | if (r) { | |
447 | goto out; | |
448 | } | |
449 | r = radeon_bo_vm_update_pte(parser, vm); | |
450 | if (r) { | |
451 | goto out; | |
452 | } | |
220907d9 | 453 | radeon_cs_sync_rings(parser); |
dfcf5f36 AD |
454 | |
455 | if ((rdev->family >= CHIP_TAHITI) && | |
456 | (parser->chunk_const_ib_idx != -1)) { | |
f2e39221 | 457 | parser->const_ib.vm_id = vm->id; |
dfcf5f36 AD |
458 | /* ib pool is bind at 0 in virtual address space to gpu_addr is the |
459 | * offset inside the pool bo | |
460 | */ | |
f2e39221 JG |
461 | parser->const_ib.gpu_addr = parser->const_ib.sa_bo->soffset; |
462 | r = radeon_ib_schedule(rdev, &parser->const_ib); | |
dfcf5f36 AD |
463 | if (r) |
464 | goto out; | |
465 | } | |
466 | ||
f2e39221 | 467 | parser->ib.vm_id = vm->id; |
721604a1 JG |
468 | /* ib pool is bind at 0 in virtual address space to gpu_addr is the |
469 | * offset inside the pool bo | |
470 | */ | |
f2e39221 JG |
471 | parser->ib.gpu_addr = parser->ib.sa_bo->soffset; |
472 | parser->ib.is_const_ib = false; | |
473 | r = radeon_ib_schedule(rdev, &parser->ib); | |
721604a1 JG |
474 | out: |
475 | if (!r) { | |
476 | if (vm->fence) { | |
477 | radeon_fence_unref(&vm->fence); | |
478 | } | |
f2e39221 | 479 | vm->fence = radeon_fence_ref(parser->ib.fence); |
721604a1 | 480 | } |
36ff39c4 CK |
481 | mutex_unlock(&vm->mutex); |
482 | mutex_unlock(&rdev->vm_manager.lock); | |
721604a1 JG |
483 | return r; |
484 | } | |
485 | ||
6c6f4783 CK |
486 | static int radeon_cs_handle_lockup(struct radeon_device *rdev, int r) |
487 | { | |
488 | if (r == -EDEADLK) { | |
489 | r = radeon_gpu_reset(rdev); | |
490 | if (!r) | |
491 | r = -EAGAIN; | |
492 | } | |
493 | return r; | |
494 | } | |
495 | ||
771fe6b9 JG |
496 | int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) |
497 | { | |
498 | struct radeon_device *rdev = dev->dev_private; | |
499 | struct radeon_cs_parser parser; | |
771fe6b9 JG |
500 | int r; |
501 | ||
dee53e7f | 502 | down_read(&rdev->exclusive_lock); |
6b7746e8 | 503 | if (!rdev->accel_working) { |
dee53e7f | 504 | up_read(&rdev->exclusive_lock); |
6b7746e8 JG |
505 | return -EBUSY; |
506 | } | |
771fe6b9 JG |
507 | /* initialize parser */ |
508 | memset(&parser, 0, sizeof(struct radeon_cs_parser)); | |
509 | parser.filp = filp; | |
510 | parser.rdev = rdev; | |
c8c15ff1 | 511 | parser.dev = rdev->dev; |
428c6e36 | 512 | parser.family = rdev->family; |
771fe6b9 JG |
513 | r = radeon_cs_parser_init(&parser, data); |
514 | if (r) { | |
515 | DRM_ERROR("Failed to initialize parser !\n"); | |
516 | radeon_cs_parser_fini(&parser, r); | |
dee53e7f | 517 | up_read(&rdev->exclusive_lock); |
6c6f4783 | 518 | r = radeon_cs_handle_lockup(rdev, r); |
771fe6b9 JG |
519 | return r; |
520 | } | |
771fe6b9 JG |
521 | r = radeon_cs_parser_relocs(&parser); |
522 | if (r) { | |
97f23b3d DA |
523 | if (r != -ERESTARTSYS) |
524 | DRM_ERROR("Failed to parse relocation %d!\n", r); | |
771fe6b9 | 525 | radeon_cs_parser_fini(&parser, r); |
dee53e7f | 526 | up_read(&rdev->exclusive_lock); |
6c6f4783 | 527 | r = radeon_cs_handle_lockup(rdev, r); |
771fe6b9 JG |
528 | return r; |
529 | } | |
721604a1 | 530 | r = radeon_cs_ib_chunk(rdev, &parser); |
771fe6b9 | 531 | if (r) { |
721604a1 | 532 | goto out; |
771fe6b9 | 533 | } |
721604a1 | 534 | r = radeon_cs_ib_vm_chunk(rdev, &parser); |
771fe6b9 | 535 | if (r) { |
721604a1 | 536 | goto out; |
771fe6b9 | 537 | } |
721604a1 | 538 | out: |
771fe6b9 | 539 | radeon_cs_parser_fini(&parser, r); |
dee53e7f | 540 | up_read(&rdev->exclusive_lock); |
6c6f4783 | 541 | r = radeon_cs_handle_lockup(rdev, r); |
771fe6b9 JG |
542 | return r; |
543 | } | |
513bcb46 DA |
544 | |
545 | int radeon_cs_finish_pages(struct radeon_cs_parser *p) | |
546 | { | |
547 | struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx]; | |
548 | int i; | |
549 | int size = PAGE_SIZE; | |
550 | ||
551 | for (i = ibc->last_copied_page + 1; i <= ibc->last_page_index; i++) { | |
552 | if (i == ibc->last_page_index) { | |
553 | size = (ibc->length_dw * 4) % PAGE_SIZE; | |
554 | if (size == 0) | |
555 | size = PAGE_SIZE; | |
556 | } | |
557 | ||
f2e39221 | 558 | if (DRM_COPY_FROM_USER(p->ib.ptr + (i * (PAGE_SIZE/4)), |
513bcb46 DA |
559 | ibc->user_ptr + (i * PAGE_SIZE), |
560 | size)) | |
561 | return -EFAULT; | |
562 | } | |
563 | return 0; | |
564 | } | |
565 | ||
c4c7f314 | 566 | static int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx) |
513bcb46 DA |
567 | { |
568 | int new_page; | |
513bcb46 DA |
569 | struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx]; |
570 | int i; | |
571 | int size = PAGE_SIZE; | |
6a7068b4 | 572 | bool copy1 = (p->rdev->flags & RADEON_IS_AGP) ? false : true; |
513bcb46 | 573 | |
c5e617e2 | 574 | for (i = ibc->last_copied_page + 1; i < pg_idx; i++) { |
f2e39221 | 575 | if (DRM_COPY_FROM_USER(p->ib.ptr + (i * (PAGE_SIZE/4)), |
513bcb46 DA |
576 | ibc->user_ptr + (i * PAGE_SIZE), |
577 | PAGE_SIZE)) { | |
578 | p->parser_error = -EFAULT; | |
579 | return 0; | |
580 | } | |
581 | } | |
582 | ||
513bcb46 DA |
583 | if (pg_idx == ibc->last_page_index) { |
584 | size = (ibc->length_dw * 4) % PAGE_SIZE; | |
6a7068b4 DA |
585 | if (size == 0) |
586 | size = PAGE_SIZE; | |
513bcb46 DA |
587 | } |
588 | ||
6a7068b4 DA |
589 | new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1; |
590 | if (copy1) | |
f2e39221 | 591 | ibc->kpage[new_page] = p->ib.ptr + (pg_idx * (PAGE_SIZE / 4)); |
6a7068b4 | 592 | |
513bcb46 DA |
593 | if (DRM_COPY_FROM_USER(ibc->kpage[new_page], |
594 | ibc->user_ptr + (pg_idx * PAGE_SIZE), | |
595 | size)) { | |
596 | p->parser_error = -EFAULT; | |
597 | return 0; | |
598 | } | |
599 | ||
6a7068b4 DA |
600 | /* copy to IB for non single case */ |
601 | if (!copy1) | |
f2e39221 | 602 | memcpy((void *)(p->ib.ptr+(pg_idx*(PAGE_SIZE/4))), ibc->kpage[new_page], size); |
513bcb46 DA |
603 | |
604 | ibc->last_copied_page = pg_idx; | |
605 | ibc->kpage_idx[new_page] = pg_idx; | |
606 | ||
607 | return new_page; | |
608 | } | |
c4c7f314 DA |
609 | |
610 | u32 radeon_get_ib_value(struct radeon_cs_parser *p, int idx) | |
611 | { | |
612 | struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx]; | |
613 | u32 pg_idx, pg_offset; | |
614 | u32 idx_value = 0; | |
615 | int new_page; | |
616 | ||
617 | pg_idx = (idx * 4) / PAGE_SIZE; | |
618 | pg_offset = (idx * 4) % PAGE_SIZE; | |
619 | ||
620 | if (ibc->kpage_idx[0] == pg_idx) | |
621 | return ibc->kpage[0][pg_offset/4]; | |
622 | if (ibc->kpage_idx[1] == pg_idx) | |
623 | return ibc->kpage[1][pg_offset/4]; | |
624 | ||
625 | new_page = radeon_cs_update_pages(p, pg_idx); | |
626 | if (new_page < 0) { | |
627 | p->parser_error = new_page; | |
628 | return 0; | |
629 | } | |
630 | ||
631 | idx_value = ibc->kpage[new_page][pg_offset/4]; | |
632 | return idx_value; | |
633 | } |