]>
Commit | Line | Data |
---|---|---|
fb1d9738 JB |
1 | /************************************************************************** |
2 | * | |
3 | * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA | |
4 | * All Rights Reserved. | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a | |
7 | * copy of this software and associated documentation files (the | |
8 | * "Software"), to deal in the Software without restriction, including | |
9 | * without limitation the rights to use, copy, modify, merge, publish, | |
10 | * distribute, sub license, and/or sell copies of the Software, and to | |
11 | * permit persons to whom the Software is furnished to do so, subject to | |
12 | * the following conditions: | |
13 | * | |
14 | * The above copyright notice and this permission notice (including the | |
15 | * next paragraph) shall be included in all copies or substantial portions | |
16 | * of the Software. | |
17 | * | |
18 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
19 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
20 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL | |
21 | * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, | |
22 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR | |
23 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE | |
24 | * USE OR OTHER DEALINGS IN THE SOFTWARE. | |
25 | * | |
26 | **************************************************************************/ | |
27 | ||
28 | #include "vmwgfx_drv.h" | |
29 | #include "vmwgfx_reg.h" | |
30 | #include "ttm/ttm_bo_api.h" | |
31 | #include "ttm/ttm_placement.h" | |
32 | ||
33 | static int vmw_cmd_invalid(struct vmw_private *dev_priv, | |
34 | struct vmw_sw_context *sw_context, | |
35 | SVGA3dCmdHeader *header) | |
36 | { | |
37 | return capable(CAP_SYS_ADMIN) ? : -EINVAL; | |
38 | } | |
39 | ||
40 | static int vmw_cmd_ok(struct vmw_private *dev_priv, | |
41 | struct vmw_sw_context *sw_context, | |
42 | SVGA3dCmdHeader *header) | |
43 | { | |
44 | return 0; | |
45 | } | |
46 | ||
f18c8840 TH |
47 | static void vmw_resource_to_validate_list(struct vmw_sw_context *sw_context, |
48 | struct vmw_resource **p_res) | |
be38ab6e | 49 | { |
be38ab6e TH |
50 | struct vmw_resource *res = *p_res; |
51 | ||
f18c8840 TH |
52 | if (list_empty(&res->validate_head)) { |
53 | list_add_tail(&res->validate_head, &sw_context->resource_list); | |
54 | *p_res = NULL; | |
55 | } else | |
56 | vmw_resource_unreference(p_res); | |
be38ab6e TH |
57 | } |
58 | ||
e2fa3a76 TH |
59 | /** |
60 | * vmw_bo_to_validate_list - add a bo to a validate list | |
61 | * | |
62 | * @sw_context: The software context used for this command submission batch. | |
63 | * @bo: The buffer object to add. | |
64 | * @fence_flags: Fence flags to be or'ed with any other fence flags for | |
65 | * this buffer on this submission batch. | |
66 | * @p_val_node: If non-NULL Will be updated with the validate node number | |
67 | * on return. | |
68 | * | |
69 | * Returns -EINVAL if the limit of number of buffer objects per command | |
70 | * submission is reached. | |
71 | */ | |
72 | static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context, | |
73 | struct ttm_buffer_object *bo, | |
74 | uint32_t fence_flags, | |
75 | uint32_t *p_val_node) | |
76 | { | |
77 | uint32_t val_node; | |
78 | struct ttm_validate_buffer *val_buf; | |
79 | ||
80 | val_node = vmw_dmabuf_validate_node(bo, sw_context->cur_val_buf); | |
81 | ||
82 | if (unlikely(val_node >= VMWGFX_MAX_VALIDATIONS)) { | |
83 | DRM_ERROR("Max number of DMA buffers per submission" | |
84 | " exceeded.\n"); | |
85 | return -EINVAL; | |
86 | } | |
87 | ||
88 | val_buf = &sw_context->val_bufs[val_node]; | |
89 | if (unlikely(val_node == sw_context->cur_val_buf)) { | |
90 | val_buf->new_sync_obj_arg = NULL; | |
91 | val_buf->bo = ttm_bo_reference(bo); | |
92 | val_buf->usage = TTM_USAGE_READWRITE; | |
93 | list_add_tail(&val_buf->head, &sw_context->validate_nodes); | |
94 | ++sw_context->cur_val_buf; | |
95 | } | |
96 | ||
97 | val_buf->new_sync_obj_arg = (void *) | |
98 | ((unsigned long) val_buf->new_sync_obj_arg | fence_flags); | |
99 | sw_context->fence_flags |= fence_flags; | |
100 | ||
101 | if (p_val_node) | |
102 | *p_val_node = val_node; | |
103 | ||
104 | return 0; | |
105 | } | |
106 | ||
fb1d9738 JB |
107 | static int vmw_cmd_cid_check(struct vmw_private *dev_priv, |
108 | struct vmw_sw_context *sw_context, | |
109 | SVGA3dCmdHeader *header) | |
110 | { | |
be38ab6e TH |
111 | struct vmw_resource *ctx; |
112 | ||
fb1d9738 JB |
113 | struct vmw_cid_cmd { |
114 | SVGA3dCmdHeader header; | |
115 | __le32 cid; | |
116 | } *cmd; | |
117 | int ret; | |
118 | ||
119 | cmd = container_of(header, struct vmw_cid_cmd, header); | |
120 | if (likely(sw_context->cid_valid && cmd->cid == sw_context->last_cid)) | |
121 | return 0; | |
122 | ||
be38ab6e TH |
123 | ret = vmw_context_check(dev_priv, sw_context->tfile, cmd->cid, |
124 | &ctx); | |
fb1d9738 JB |
125 | if (unlikely(ret != 0)) { |
126 | DRM_ERROR("Could not find or use context %u\n", | |
127 | (unsigned) cmd->cid); | |
128 | return ret; | |
129 | } | |
130 | ||
131 | sw_context->last_cid = cmd->cid; | |
132 | sw_context->cid_valid = true; | |
e2fa3a76 | 133 | sw_context->cur_ctx = ctx; |
f18c8840 TH |
134 | vmw_resource_to_validate_list(sw_context, &ctx); |
135 | ||
136 | return 0; | |
fb1d9738 JB |
137 | } |
138 | ||
139 | static int vmw_cmd_sid_check(struct vmw_private *dev_priv, | |
140 | struct vmw_sw_context *sw_context, | |
7a73ba74 | 141 | uint32_t *sid) |
fb1d9738 | 142 | { |
be38ab6e TH |
143 | struct vmw_surface *srf; |
144 | int ret; | |
145 | struct vmw_resource *res; | |
146 | ||
7a73ba74 TH |
147 | if (*sid == SVGA3D_INVALID_ID) |
148 | return 0; | |
149 | ||
be38ab6e TH |
150 | if (likely((sw_context->sid_valid && |
151 | *sid == sw_context->last_sid))) { | |
7a73ba74 | 152 | *sid = sw_context->sid_translation; |
be38ab6e TH |
153 | return 0; |
154 | } | |
7a73ba74 | 155 | |
e2fa3a76 TH |
156 | ret = vmw_user_surface_lookup_handle(dev_priv, |
157 | sw_context->tfile, | |
be38ab6e TH |
158 | *sid, &srf); |
159 | if (unlikely(ret != 0)) { | |
160 | DRM_ERROR("Could ot find or use surface 0x%08x " | |
161 | "address 0x%08lx\n", | |
162 | (unsigned int) *sid, | |
163 | (unsigned long) sid); | |
164 | return ret; | |
165 | } | |
166 | ||
5bb39e81 TH |
167 | ret = vmw_surface_validate(dev_priv, srf); |
168 | if (unlikely(ret != 0)) { | |
169 | if (ret != -ERESTARTSYS) | |
170 | DRM_ERROR("Could not validate surface.\n"); | |
171 | vmw_surface_unreference(&srf); | |
172 | return ret; | |
173 | } | |
174 | ||
be38ab6e TH |
175 | sw_context->last_sid = *sid; |
176 | sw_context->sid_valid = true; | |
177 | sw_context->sid_translation = srf->res.id; | |
178 | *sid = sw_context->sid_translation; | |
179 | ||
180 | res = &srf->res; | |
f18c8840 TH |
181 | vmw_resource_to_validate_list(sw_context, &res); |
182 | ||
183 | return 0; | |
fb1d9738 JB |
184 | } |
185 | ||
186 | ||
187 | static int vmw_cmd_set_render_target_check(struct vmw_private *dev_priv, | |
188 | struct vmw_sw_context *sw_context, | |
189 | SVGA3dCmdHeader *header) | |
190 | { | |
191 | struct vmw_sid_cmd { | |
192 | SVGA3dCmdHeader header; | |
193 | SVGA3dCmdSetRenderTarget body; | |
194 | } *cmd; | |
195 | int ret; | |
196 | ||
197 | ret = vmw_cmd_cid_check(dev_priv, sw_context, header); | |
198 | if (unlikely(ret != 0)) | |
199 | return ret; | |
200 | ||
201 | cmd = container_of(header, struct vmw_sid_cmd, header); | |
7a73ba74 TH |
202 | ret = vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.target.sid); |
203 | return ret; | |
fb1d9738 JB |
204 | } |
205 | ||
206 | static int vmw_cmd_surface_copy_check(struct vmw_private *dev_priv, | |
207 | struct vmw_sw_context *sw_context, | |
208 | SVGA3dCmdHeader *header) | |
209 | { | |
210 | struct vmw_sid_cmd { | |
211 | SVGA3dCmdHeader header; | |
212 | SVGA3dCmdSurfaceCopy body; | |
213 | } *cmd; | |
214 | int ret; | |
215 | ||
216 | cmd = container_of(header, struct vmw_sid_cmd, header); | |
7a73ba74 | 217 | ret = vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.src.sid); |
fb1d9738 JB |
218 | if (unlikely(ret != 0)) |
219 | return ret; | |
7a73ba74 | 220 | return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.dest.sid); |
fb1d9738 JB |
221 | } |
222 | ||
223 | static int vmw_cmd_stretch_blt_check(struct vmw_private *dev_priv, | |
224 | struct vmw_sw_context *sw_context, | |
225 | SVGA3dCmdHeader *header) | |
226 | { | |
227 | struct vmw_sid_cmd { | |
228 | SVGA3dCmdHeader header; | |
229 | SVGA3dCmdSurfaceStretchBlt body; | |
230 | } *cmd; | |
231 | int ret; | |
232 | ||
233 | cmd = container_of(header, struct vmw_sid_cmd, header); | |
7a73ba74 | 234 | ret = vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.src.sid); |
fb1d9738 JB |
235 | if (unlikely(ret != 0)) |
236 | return ret; | |
7a73ba74 | 237 | return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.dest.sid); |
fb1d9738 JB |
238 | } |
239 | ||
240 | static int vmw_cmd_blt_surf_screen_check(struct vmw_private *dev_priv, | |
241 | struct vmw_sw_context *sw_context, | |
242 | SVGA3dCmdHeader *header) | |
243 | { | |
244 | struct vmw_sid_cmd { | |
245 | SVGA3dCmdHeader header; | |
246 | SVGA3dCmdBlitSurfaceToScreen body; | |
247 | } *cmd; | |
248 | ||
249 | cmd = container_of(header, struct vmw_sid_cmd, header); | |
0cff60c6 JB |
250 | |
251 | if (unlikely(!sw_context->kernel)) { | |
252 | DRM_ERROR("Kernel only SVGA3d command: %u.\n", cmd->header.id); | |
253 | return -EPERM; | |
254 | } | |
255 | ||
7a73ba74 | 256 | return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.srcImage.sid); |
fb1d9738 JB |
257 | } |
258 | ||
259 | static int vmw_cmd_present_check(struct vmw_private *dev_priv, | |
260 | struct vmw_sw_context *sw_context, | |
261 | SVGA3dCmdHeader *header) | |
262 | { | |
263 | struct vmw_sid_cmd { | |
264 | SVGA3dCmdHeader header; | |
265 | SVGA3dCmdPresent body; | |
266 | } *cmd; | |
267 | ||
5bb39e81 | 268 | |
fb1d9738 | 269 | cmd = container_of(header, struct vmw_sid_cmd, header); |
0cff60c6 JB |
270 | |
271 | if (unlikely(!sw_context->kernel)) { | |
272 | DRM_ERROR("Kernel only SVGA3d command: %u.\n", cmd->header.id); | |
273 | return -EPERM; | |
274 | } | |
275 | ||
7a73ba74 | 276 | return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.sid); |
fb1d9738 JB |
277 | } |
278 | ||
e2fa3a76 TH |
279 | /** |
280 | * vmw_query_bo_switch_prepare - Prepare to switch pinned buffer for queries. | |
281 | * | |
282 | * @dev_priv: The device private structure. | |
283 | * @cid: The hardware context for the next query. | |
284 | * @new_query_bo: The new buffer holding query results. | |
285 | * @sw_context: The software context used for this command submission. | |
286 | * | |
287 | * This function checks whether @new_query_bo is suitable for holding | |
288 | * query results, and if another buffer currently is pinned for query | |
289 | * results. If so, the function prepares the state of @sw_context for | |
290 | * switching pinned buffers after successful submission of the current | |
291 | * command batch. It also checks whether we're using a new query context. | |
292 | * In that case, it makes sure we emit a query barrier for the old | |
293 | * context before the current query buffer is fenced. | |
294 | */ | |
295 | static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv, | |
296 | uint32_t cid, | |
297 | struct ttm_buffer_object *new_query_bo, | |
298 | struct vmw_sw_context *sw_context) | |
299 | { | |
300 | int ret; | |
301 | bool add_cid = false; | |
302 | uint32_t cid_to_add; | |
303 | ||
304 | if (unlikely(new_query_bo != sw_context->cur_query_bo)) { | |
305 | ||
306 | if (unlikely(new_query_bo->num_pages > 4)) { | |
307 | DRM_ERROR("Query buffer too large.\n"); | |
308 | return -EINVAL; | |
309 | } | |
310 | ||
311 | if (unlikely(sw_context->cur_query_bo != NULL)) { | |
312 | BUG_ON(!sw_context->query_cid_valid); | |
313 | add_cid = true; | |
314 | cid_to_add = sw_context->cur_query_cid; | |
315 | ret = vmw_bo_to_validate_list(sw_context, | |
316 | sw_context->cur_query_bo, | |
317 | DRM_VMW_FENCE_FLAG_EXEC, | |
318 | NULL); | |
319 | if (unlikely(ret != 0)) | |
320 | return ret; | |
321 | } | |
322 | sw_context->cur_query_bo = new_query_bo; | |
323 | ||
324 | ret = vmw_bo_to_validate_list(sw_context, | |
325 | dev_priv->dummy_query_bo, | |
326 | DRM_VMW_FENCE_FLAG_EXEC, | |
327 | NULL); | |
328 | if (unlikely(ret != 0)) | |
329 | return ret; | |
330 | ||
331 | } | |
332 | ||
333 | if (unlikely(cid != sw_context->cur_query_cid && | |
334 | sw_context->query_cid_valid)) { | |
335 | add_cid = true; | |
336 | cid_to_add = sw_context->cur_query_cid; | |
337 | } | |
338 | ||
339 | sw_context->cur_query_cid = cid; | |
340 | sw_context->query_cid_valid = true; | |
341 | ||
342 | if (add_cid) { | |
343 | struct vmw_resource *ctx = sw_context->cur_ctx; | |
344 | ||
345 | if (list_empty(&ctx->query_head)) | |
346 | list_add_tail(&ctx->query_head, | |
347 | &sw_context->query_list); | |
348 | ret = vmw_bo_to_validate_list(sw_context, | |
349 | dev_priv->dummy_query_bo, | |
350 | DRM_VMW_FENCE_FLAG_EXEC, | |
351 | NULL); | |
352 | if (unlikely(ret != 0)) | |
353 | return ret; | |
354 | } | |
355 | return 0; | |
356 | } | |
357 | ||
358 | ||
359 | /** | |
360 | * vmw_query_bo_switch_commit - Finalize switching pinned query buffer | |
361 | * | |
362 | * @dev_priv: The device private structure. | |
363 | * @sw_context: The software context used for this command submission batch. | |
364 | * | |
365 | * This function will check if we're switching query buffers, and will then, | |
366 | * if no other query waits are issued this command submission batch, | |
367 | * issue a dummy occlusion query wait used as a query barrier. When the fence | |
368 | * object following that query wait has signaled, we are sure that all | |
369 | * preseding queries have finished, and the old query buffer can be unpinned. | |
370 | * However, since both the new query buffer and the old one are fenced with | |
371 | * that fence, we can do an asynchronus unpin now, and be sure that the | |
372 | * old query buffer won't be moved until the fence has signaled. | |
373 | * | |
374 | * As mentioned above, both the new - and old query buffers need to be fenced | |
375 | * using a sequence emitted *after* calling this function. | |
376 | */ | |
377 | static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv, | |
378 | struct vmw_sw_context *sw_context) | |
379 | { | |
380 | ||
381 | struct vmw_resource *ctx, *next_ctx; | |
382 | int ret; | |
383 | ||
384 | /* | |
385 | * The validate list should still hold references to all | |
386 | * contexts here. | |
387 | */ | |
388 | ||
389 | list_for_each_entry_safe(ctx, next_ctx, &sw_context->query_list, | |
390 | query_head) { | |
391 | list_del_init(&ctx->query_head); | |
392 | ||
f18c8840 | 393 | BUG_ON(list_empty(&ctx->validate_head)); |
e2fa3a76 TH |
394 | |
395 | ret = vmw_fifo_emit_dummy_query(dev_priv, ctx->id); | |
396 | ||
397 | if (unlikely(ret != 0)) | |
398 | DRM_ERROR("Out of fifo space for dummy query.\n"); | |
399 | } | |
400 | ||
401 | if (dev_priv->pinned_bo != sw_context->cur_query_bo) { | |
402 | if (dev_priv->pinned_bo) { | |
403 | vmw_bo_pin(dev_priv->pinned_bo, false); | |
404 | ttm_bo_unref(&dev_priv->pinned_bo); | |
405 | } | |
406 | ||
407 | vmw_bo_pin(sw_context->cur_query_bo, true); | |
408 | ||
409 | /* | |
410 | * We pin also the dummy_query_bo buffer so that we | |
411 | * don't need to validate it when emitting | |
412 | * dummy queries in context destroy paths. | |
413 | */ | |
414 | ||
415 | vmw_bo_pin(dev_priv->dummy_query_bo, true); | |
416 | dev_priv->dummy_query_bo_pinned = true; | |
417 | ||
418 | dev_priv->query_cid = sw_context->cur_query_cid; | |
419 | dev_priv->pinned_bo = | |
420 | ttm_bo_reference(sw_context->cur_query_bo); | |
421 | } | |
422 | } | |
423 | ||
424 | /** | |
425 | * vmw_query_switch_backoff - clear query barrier list | |
426 | * @sw_context: The sw context used for this submission batch. | |
427 | * | |
428 | * This function is used as part of an error path, where a previously | |
429 | * set up list of query barriers needs to be cleared. | |
430 | * | |
431 | */ | |
432 | static void vmw_query_switch_backoff(struct vmw_sw_context *sw_context) | |
433 | { | |
434 | struct list_head *list, *next; | |
435 | ||
436 | list_for_each_safe(list, next, &sw_context->query_list) { | |
437 | list_del_init(list); | |
438 | } | |
439 | } | |
440 | ||
4e4ddd47 TH |
441 | static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, |
442 | struct vmw_sw_context *sw_context, | |
443 | SVGAGuestPtr *ptr, | |
444 | struct vmw_dma_buffer **vmw_bo_p) | |
fb1d9738 | 445 | { |
fb1d9738 JB |
446 | struct vmw_dma_buffer *vmw_bo = NULL; |
447 | struct ttm_buffer_object *bo; | |
4e4ddd47 | 448 | uint32_t handle = ptr->gmrId; |
fb1d9738 | 449 | struct vmw_relocation *reloc; |
4e4ddd47 | 450 | int ret; |
fb1d9738 | 451 | |
fb1d9738 JB |
452 | ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo); |
453 | if (unlikely(ret != 0)) { | |
454 | DRM_ERROR("Could not find or use GMR region.\n"); | |
455 | return -EINVAL; | |
456 | } | |
457 | bo = &vmw_bo->base; | |
458 | ||
459 | if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) { | |
4e4ddd47 | 460 | DRM_ERROR("Max number relocations per submission" |
fb1d9738 JB |
461 | " exceeded\n"); |
462 | ret = -EINVAL; | |
463 | goto out_no_reloc; | |
464 | } | |
465 | ||
466 | reloc = &sw_context->relocs[sw_context->cur_reloc++]; | |
4e4ddd47 | 467 | reloc->location = ptr; |
fb1d9738 | 468 | |
e2fa3a76 TH |
469 | ret = vmw_bo_to_validate_list(sw_context, bo, DRM_VMW_FENCE_FLAG_EXEC, |
470 | &reloc->index); | |
471 | if (unlikely(ret != 0)) | |
fb1d9738 | 472 | goto out_no_reloc; |
fb1d9738 | 473 | |
4e4ddd47 TH |
474 | *vmw_bo_p = vmw_bo; |
475 | return 0; | |
476 | ||
477 | out_no_reloc: | |
478 | vmw_dmabuf_unreference(&vmw_bo); | |
479 | vmw_bo_p = NULL; | |
480 | return ret; | |
481 | } | |
482 | ||
483 | static int vmw_cmd_end_query(struct vmw_private *dev_priv, | |
484 | struct vmw_sw_context *sw_context, | |
485 | SVGA3dCmdHeader *header) | |
486 | { | |
487 | struct vmw_dma_buffer *vmw_bo; | |
488 | struct vmw_query_cmd { | |
489 | SVGA3dCmdHeader header; | |
490 | SVGA3dCmdEndQuery q; | |
491 | } *cmd; | |
492 | int ret; | |
493 | ||
494 | cmd = container_of(header, struct vmw_query_cmd, header); | |
495 | ret = vmw_cmd_cid_check(dev_priv, sw_context, header); | |
496 | if (unlikely(ret != 0)) | |
497 | return ret; | |
498 | ||
499 | ret = vmw_translate_guest_ptr(dev_priv, sw_context, | |
500 | &cmd->q.guestResult, | |
501 | &vmw_bo); | |
502 | if (unlikely(ret != 0)) | |
503 | return ret; | |
504 | ||
e2fa3a76 TH |
505 | ret = vmw_query_bo_switch_prepare(dev_priv, cmd->q.cid, |
506 | &vmw_bo->base, sw_context); | |
507 | ||
4e4ddd47 | 508 | vmw_dmabuf_unreference(&vmw_bo); |
e2fa3a76 | 509 | return ret; |
4e4ddd47 | 510 | } |
fb1d9738 | 511 | |
4e4ddd47 TH |
512 | static int vmw_cmd_wait_query(struct vmw_private *dev_priv, |
513 | struct vmw_sw_context *sw_context, | |
514 | SVGA3dCmdHeader *header) | |
515 | { | |
516 | struct vmw_dma_buffer *vmw_bo; | |
517 | struct vmw_query_cmd { | |
518 | SVGA3dCmdHeader header; | |
519 | SVGA3dCmdWaitForQuery q; | |
520 | } *cmd; | |
521 | int ret; | |
e2fa3a76 | 522 | struct vmw_resource *ctx; |
4e4ddd47 TH |
523 | |
524 | cmd = container_of(header, struct vmw_query_cmd, header); | |
525 | ret = vmw_cmd_cid_check(dev_priv, sw_context, header); | |
526 | if (unlikely(ret != 0)) | |
527 | return ret; | |
528 | ||
529 | ret = vmw_translate_guest_ptr(dev_priv, sw_context, | |
530 | &cmd->q.guestResult, | |
531 | &vmw_bo); | |
532 | if (unlikely(ret != 0)) | |
533 | return ret; | |
534 | ||
535 | vmw_dmabuf_unreference(&vmw_bo); | |
e2fa3a76 TH |
536 | |
537 | /* | |
538 | * This wait will act as a barrier for previous waits for this | |
539 | * context. | |
540 | */ | |
541 | ||
542 | ctx = sw_context->cur_ctx; | |
543 | if (!list_empty(&ctx->query_head)) | |
544 | list_del_init(&ctx->query_head); | |
545 | ||
4e4ddd47 TH |
546 | return 0; |
547 | } | |
548 | ||
4e4ddd47 TH |
549 | static int vmw_cmd_dma(struct vmw_private *dev_priv, |
550 | struct vmw_sw_context *sw_context, | |
551 | SVGA3dCmdHeader *header) | |
552 | { | |
553 | struct vmw_dma_buffer *vmw_bo = NULL; | |
554 | struct ttm_buffer_object *bo; | |
555 | struct vmw_surface *srf = NULL; | |
556 | struct vmw_dma_cmd { | |
557 | SVGA3dCmdHeader header; | |
558 | SVGA3dCmdSurfaceDMA dma; | |
559 | } *cmd; | |
560 | int ret; | |
be38ab6e | 561 | struct vmw_resource *res; |
4e4ddd47 TH |
562 | |
563 | cmd = container_of(header, struct vmw_dma_cmd, header); | |
564 | ret = vmw_translate_guest_ptr(dev_priv, sw_context, | |
565 | &cmd->dma.guest.ptr, | |
566 | &vmw_bo); | |
567 | if (unlikely(ret != 0)) | |
568 | return ret; | |
569 | ||
570 | bo = &vmw_bo->base; | |
7a73ba74 TH |
571 | ret = vmw_user_surface_lookup_handle(dev_priv, sw_context->tfile, |
572 | cmd->dma.host.sid, &srf); | |
fb1d9738 JB |
573 | if (ret) { |
574 | DRM_ERROR("could not find surface\n"); | |
575 | goto out_no_reloc; | |
576 | } | |
577 | ||
5bb39e81 TH |
578 | ret = vmw_surface_validate(dev_priv, srf); |
579 | if (unlikely(ret != 0)) { | |
580 | if (ret != -ERESTARTSYS) | |
581 | DRM_ERROR("Culd not validate surface.\n"); | |
582 | goto out_no_validate; | |
583 | } | |
584 | ||
be38ab6e | 585 | /* |
7a73ba74 TH |
586 | * Patch command stream with device SID. |
587 | */ | |
7a73ba74 | 588 | cmd->dma.host.sid = srf->res.id; |
fb1d9738 | 589 | vmw_kms_cursor_snoop(srf, sw_context->tfile, bo, header); |
be38ab6e TH |
590 | |
591 | vmw_dmabuf_unreference(&vmw_bo); | |
592 | ||
593 | res = &srf->res; | |
f18c8840 TH |
594 | vmw_resource_to_validate_list(sw_context, &res); |
595 | ||
596 | return 0; | |
fb1d9738 | 597 | |
5bb39e81 TH |
598 | out_no_validate: |
599 | vmw_surface_unreference(&srf); | |
fb1d9738 JB |
600 | out_no_reloc: |
601 | vmw_dmabuf_unreference(&vmw_bo); | |
602 | return ret; | |
603 | } | |
604 | ||
7a73ba74 TH |
605 | static int vmw_cmd_draw(struct vmw_private *dev_priv, |
606 | struct vmw_sw_context *sw_context, | |
607 | SVGA3dCmdHeader *header) | |
608 | { | |
609 | struct vmw_draw_cmd { | |
610 | SVGA3dCmdHeader header; | |
611 | SVGA3dCmdDrawPrimitives body; | |
612 | } *cmd; | |
613 | SVGA3dVertexDecl *decl = (SVGA3dVertexDecl *)( | |
614 | (unsigned long)header + sizeof(*cmd)); | |
615 | SVGA3dPrimitiveRange *range; | |
616 | uint32_t i; | |
617 | uint32_t maxnum; | |
618 | int ret; | |
619 | ||
620 | ret = vmw_cmd_cid_check(dev_priv, sw_context, header); | |
621 | if (unlikely(ret != 0)) | |
622 | return ret; | |
623 | ||
624 | cmd = container_of(header, struct vmw_draw_cmd, header); | |
625 | maxnum = (header->size - sizeof(cmd->body)) / sizeof(*decl); | |
626 | ||
627 | if (unlikely(cmd->body.numVertexDecls > maxnum)) { | |
628 | DRM_ERROR("Illegal number of vertex declarations.\n"); | |
629 | return -EINVAL; | |
630 | } | |
631 | ||
632 | for (i = 0; i < cmd->body.numVertexDecls; ++i, ++decl) { | |
633 | ret = vmw_cmd_sid_check(dev_priv, sw_context, | |
634 | &decl->array.surfaceId); | |
635 | if (unlikely(ret != 0)) | |
636 | return ret; | |
637 | } | |
638 | ||
639 | maxnum = (header->size - sizeof(cmd->body) - | |
640 | cmd->body.numVertexDecls * sizeof(*decl)) / sizeof(*range); | |
641 | if (unlikely(cmd->body.numRanges > maxnum)) { | |
642 | DRM_ERROR("Illegal number of index ranges.\n"); | |
643 | return -EINVAL; | |
644 | } | |
645 | ||
646 | range = (SVGA3dPrimitiveRange *) decl; | |
647 | for (i = 0; i < cmd->body.numRanges; ++i, ++range) { | |
648 | ret = vmw_cmd_sid_check(dev_priv, sw_context, | |
649 | &range->indexArray.surfaceId); | |
650 | if (unlikely(ret != 0)) | |
651 | return ret; | |
652 | } | |
653 | return 0; | |
654 | } | |
655 | ||
656 | ||
657 | static int vmw_cmd_tex_state(struct vmw_private *dev_priv, | |
658 | struct vmw_sw_context *sw_context, | |
659 | SVGA3dCmdHeader *header) | |
660 | { | |
661 | struct vmw_tex_state_cmd { | |
662 | SVGA3dCmdHeader header; | |
663 | SVGA3dCmdSetTextureState state; | |
664 | }; | |
665 | ||
666 | SVGA3dTextureState *last_state = (SVGA3dTextureState *) | |
667 | ((unsigned long) header + header->size + sizeof(header)); | |
668 | SVGA3dTextureState *cur_state = (SVGA3dTextureState *) | |
669 | ((unsigned long) header + sizeof(struct vmw_tex_state_cmd)); | |
670 | int ret; | |
671 | ||
672 | ret = vmw_cmd_cid_check(dev_priv, sw_context, header); | |
673 | if (unlikely(ret != 0)) | |
674 | return ret; | |
675 | ||
676 | for (; cur_state < last_state; ++cur_state) { | |
677 | if (likely(cur_state->name != SVGA3D_TS_BIND_TEXTURE)) | |
678 | continue; | |
679 | ||
680 | ret = vmw_cmd_sid_check(dev_priv, sw_context, | |
681 | &cur_state->value); | |
682 | if (unlikely(ret != 0)) | |
683 | return ret; | |
684 | } | |
685 | ||
686 | return 0; | |
687 | } | |
688 | ||
4084fb89 JB |
689 | static int vmw_cmd_check_define_gmrfb(struct vmw_private *dev_priv, |
690 | struct vmw_sw_context *sw_context, | |
691 | void *buf) | |
692 | { | |
693 | struct vmw_dma_buffer *vmw_bo; | |
694 | int ret; | |
695 | ||
696 | struct { | |
697 | uint32_t header; | |
698 | SVGAFifoCmdDefineGMRFB body; | |
699 | } *cmd = buf; | |
700 | ||
701 | ret = vmw_translate_guest_ptr(dev_priv, sw_context, | |
702 | &cmd->body.ptr, | |
703 | &vmw_bo); | |
704 | if (unlikely(ret != 0)) | |
705 | return ret; | |
706 | ||
707 | vmw_dmabuf_unreference(&vmw_bo); | |
708 | ||
709 | return ret; | |
710 | } | |
711 | ||
712 | static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv, | |
713 | struct vmw_sw_context *sw_context, | |
714 | void *buf, uint32_t *size) | |
715 | { | |
716 | uint32_t size_remaining = *size; | |
4084fb89 JB |
717 | uint32_t cmd_id; |
718 | ||
719 | cmd_id = le32_to_cpu(((uint32_t *)buf)[0]); | |
720 | switch (cmd_id) { | |
721 | case SVGA_CMD_UPDATE: | |
722 | *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdUpdate); | |
4084fb89 JB |
723 | break; |
724 | case SVGA_CMD_DEFINE_GMRFB: | |
725 | *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdDefineGMRFB); | |
726 | break; | |
727 | case SVGA_CMD_BLIT_GMRFB_TO_SCREEN: | |
728 | *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdBlitGMRFBToScreen); | |
729 | break; | |
730 | case SVGA_CMD_BLIT_SCREEN_TO_GMRFB: | |
731 | *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdBlitGMRFBToScreen); | |
732 | break; | |
733 | default: | |
734 | DRM_ERROR("Unsupported SVGA command: %u.\n", cmd_id); | |
735 | return -EINVAL; | |
736 | } | |
737 | ||
738 | if (*size > size_remaining) { | |
739 | DRM_ERROR("Invalid SVGA command (size mismatch):" | |
740 | " %u.\n", cmd_id); | |
741 | return -EINVAL; | |
742 | } | |
743 | ||
0cff60c6 | 744 | if (unlikely(!sw_context->kernel)) { |
4084fb89 JB |
745 | DRM_ERROR("Kernel only SVGA command: %u.\n", cmd_id); |
746 | return -EPERM; | |
747 | } | |
748 | ||
749 | if (cmd_id == SVGA_CMD_DEFINE_GMRFB) | |
750 | return vmw_cmd_check_define_gmrfb(dev_priv, sw_context, buf); | |
751 | ||
752 | return 0; | |
753 | } | |
fb1d9738 JB |
754 | |
755 | typedef int (*vmw_cmd_func) (struct vmw_private *, | |
756 | struct vmw_sw_context *, | |
757 | SVGA3dCmdHeader *); | |
758 | ||
759 | #define VMW_CMD_DEF(cmd, func) \ | |
760 | [cmd - SVGA_3D_CMD_BASE] = func | |
761 | ||
762 | static vmw_cmd_func vmw_cmd_funcs[SVGA_3D_CMD_MAX] = { | |
763 | VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE, &vmw_cmd_invalid), | |
764 | VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DESTROY, &vmw_cmd_invalid), | |
765 | VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_COPY, &vmw_cmd_surface_copy_check), | |
766 | VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_STRETCHBLT, &vmw_cmd_stretch_blt_check), | |
767 | VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DMA, &vmw_cmd_dma), | |
768 | VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DEFINE, &vmw_cmd_invalid), | |
769 | VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DESTROY, &vmw_cmd_invalid), | |
770 | VMW_CMD_DEF(SVGA_3D_CMD_SETTRANSFORM, &vmw_cmd_cid_check), | |
771 | VMW_CMD_DEF(SVGA_3D_CMD_SETZRANGE, &vmw_cmd_cid_check), | |
772 | VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERSTATE, &vmw_cmd_cid_check), | |
773 | VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERTARGET, | |
774 | &vmw_cmd_set_render_target_check), | |
7a73ba74 | 775 | VMW_CMD_DEF(SVGA_3D_CMD_SETTEXTURESTATE, &vmw_cmd_tex_state), |
fb1d9738 JB |
776 | VMW_CMD_DEF(SVGA_3D_CMD_SETMATERIAL, &vmw_cmd_cid_check), |
777 | VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTDATA, &vmw_cmd_cid_check), | |
778 | VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTENABLED, &vmw_cmd_cid_check), | |
779 | VMW_CMD_DEF(SVGA_3D_CMD_SETVIEWPORT, &vmw_cmd_cid_check), | |
780 | VMW_CMD_DEF(SVGA_3D_CMD_SETCLIPPLANE, &vmw_cmd_cid_check), | |
781 | VMW_CMD_DEF(SVGA_3D_CMD_CLEAR, &vmw_cmd_cid_check), | |
782 | VMW_CMD_DEF(SVGA_3D_CMD_PRESENT, &vmw_cmd_present_check), | |
783 | VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DEFINE, &vmw_cmd_cid_check), | |
784 | VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_cid_check), | |
785 | VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER, &vmw_cmd_cid_check), | |
786 | VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER_CONST, &vmw_cmd_cid_check), | |
7a73ba74 | 787 | VMW_CMD_DEF(SVGA_3D_CMD_DRAW_PRIMITIVES, &vmw_cmd_draw), |
fb1d9738 JB |
788 | VMW_CMD_DEF(SVGA_3D_CMD_SETSCISSORRECT, &vmw_cmd_cid_check), |
789 | VMW_CMD_DEF(SVGA_3D_CMD_BEGIN_QUERY, &vmw_cmd_cid_check), | |
4e4ddd47 TH |
790 | VMW_CMD_DEF(SVGA_3D_CMD_END_QUERY, &vmw_cmd_end_query), |
791 | VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_QUERY, &vmw_cmd_wait_query), | |
fb1d9738 JB |
792 | VMW_CMD_DEF(SVGA_3D_CMD_PRESENT_READBACK, &vmw_cmd_ok), |
793 | VMW_CMD_DEF(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN, | |
794 | &vmw_cmd_blt_surf_screen_check) | |
795 | }; | |
796 | ||
797 | static int vmw_cmd_check(struct vmw_private *dev_priv, | |
798 | struct vmw_sw_context *sw_context, | |
799 | void *buf, uint32_t *size) | |
800 | { | |
801 | uint32_t cmd_id; | |
7a73ba74 | 802 | uint32_t size_remaining = *size; |
fb1d9738 JB |
803 | SVGA3dCmdHeader *header = (SVGA3dCmdHeader *) buf; |
804 | int ret; | |
805 | ||
4084fb89 JB |
806 | cmd_id = le32_to_cpu(((uint32_t *)buf)[0]); |
807 | /* Handle any none 3D commands */ | |
808 | if (unlikely(cmd_id < SVGA_CMD_MAX)) | |
809 | return vmw_cmd_check_not_3d(dev_priv, sw_context, buf, size); | |
810 | ||
fb1d9738 JB |
811 | |
812 | cmd_id = le32_to_cpu(header->id); | |
813 | *size = le32_to_cpu(header->size) + sizeof(SVGA3dCmdHeader); | |
814 | ||
815 | cmd_id -= SVGA_3D_CMD_BASE; | |
7a73ba74 TH |
816 | if (unlikely(*size > size_remaining)) |
817 | goto out_err; | |
818 | ||
fb1d9738 JB |
819 | if (unlikely(cmd_id >= SVGA_3D_CMD_MAX - SVGA_3D_CMD_BASE)) |
820 | goto out_err; | |
821 | ||
822 | ret = vmw_cmd_funcs[cmd_id](dev_priv, sw_context, header); | |
823 | if (unlikely(ret != 0)) | |
824 | goto out_err; | |
825 | ||
826 | return 0; | |
827 | out_err: | |
828 | DRM_ERROR("Illegal / Invalid SVGA3D command: %d\n", | |
829 | cmd_id + SVGA_3D_CMD_BASE); | |
830 | return -EINVAL; | |
831 | } | |
832 | ||
833 | static int vmw_cmd_check_all(struct vmw_private *dev_priv, | |
834 | struct vmw_sw_context *sw_context, | |
922ade0d | 835 | void *buf, |
be38ab6e | 836 | uint32_t size) |
fb1d9738 JB |
837 | { |
838 | int32_t cur_size = size; | |
839 | int ret; | |
840 | ||
841 | while (cur_size > 0) { | |
7a73ba74 | 842 | size = cur_size; |
fb1d9738 JB |
843 | ret = vmw_cmd_check(dev_priv, sw_context, buf, &size); |
844 | if (unlikely(ret != 0)) | |
845 | return ret; | |
846 | buf = (void *)((unsigned long) buf + size); | |
847 | cur_size -= size; | |
848 | } | |
849 | ||
850 | if (unlikely(cur_size != 0)) { | |
851 | DRM_ERROR("Command verifier out of sync.\n"); | |
852 | return -EINVAL; | |
853 | } | |
854 | ||
855 | return 0; | |
856 | } | |
857 | ||
858 | static void vmw_free_relocations(struct vmw_sw_context *sw_context) | |
859 | { | |
860 | sw_context->cur_reloc = 0; | |
861 | } | |
862 | ||
863 | static void vmw_apply_relocations(struct vmw_sw_context *sw_context) | |
864 | { | |
865 | uint32_t i; | |
866 | struct vmw_relocation *reloc; | |
867 | struct ttm_validate_buffer *validate; | |
868 | struct ttm_buffer_object *bo; | |
869 | ||
870 | for (i = 0; i < sw_context->cur_reloc; ++i) { | |
871 | reloc = &sw_context->relocs[i]; | |
872 | validate = &sw_context->val_bufs[reloc->index]; | |
873 | bo = validate->bo; | |
135cba0d TH |
874 | if (bo->mem.mem_type == TTM_PL_VRAM) { |
875 | reloc->location->offset += bo->offset; | |
876 | reloc->location->gmrId = SVGA_GMR_FRAMEBUFFER; | |
877 | } else | |
878 | reloc->location->gmrId = bo->mem.start; | |
fb1d9738 JB |
879 | } |
880 | vmw_free_relocations(sw_context); | |
881 | } | |
882 | ||
883 | static void vmw_clear_validations(struct vmw_sw_context *sw_context) | |
884 | { | |
885 | struct ttm_validate_buffer *entry, *next; | |
f18c8840 | 886 | struct vmw_resource *res, *res_next; |
fb1d9738 | 887 | |
be38ab6e TH |
888 | /* |
889 | * Drop references to DMA buffers held during command submission. | |
890 | */ | |
fb1d9738 JB |
891 | list_for_each_entry_safe(entry, next, &sw_context->validate_nodes, |
892 | head) { | |
893 | list_del(&entry->head); | |
894 | vmw_dmabuf_validate_clear(entry->bo); | |
895 | ttm_bo_unref(&entry->bo); | |
896 | sw_context->cur_val_buf--; | |
897 | } | |
898 | BUG_ON(sw_context->cur_val_buf != 0); | |
be38ab6e TH |
899 | |
900 | /* | |
901 | * Drop references to resources held during command submission. | |
902 | */ | |
5bb39e81 | 903 | vmw_resource_unreserve(&sw_context->resource_list); |
f18c8840 TH |
904 | list_for_each_entry_safe(res, res_next, &sw_context->resource_list, |
905 | validate_head) { | |
906 | list_del_init(&res->validate_head); | |
907 | vmw_resource_unreference(&res); | |
be38ab6e | 908 | } |
fb1d9738 JB |
909 | } |
910 | ||
911 | static int vmw_validate_single_buffer(struct vmw_private *dev_priv, | |
912 | struct ttm_buffer_object *bo) | |
913 | { | |
914 | int ret; | |
915 | ||
e2fa3a76 TH |
916 | |
917 | /* | |
918 | * Don't validate pinned buffers. | |
919 | */ | |
920 | ||
921 | if (bo == dev_priv->pinned_bo || | |
922 | (bo == dev_priv->dummy_query_bo && | |
923 | dev_priv->dummy_query_bo_pinned)) | |
924 | return 0; | |
925 | ||
8ba5152a | 926 | /** |
135cba0d TH |
927 | * Put BO in VRAM if there is space, otherwise as a GMR. |
928 | * If there is no space in VRAM and GMR ids are all used up, | |
929 | * start evicting GMRs to make room. If the DMA buffer can't be | |
930 | * used as a GMR, this will return -ENOMEM. | |
8ba5152a TH |
931 | */ |
932 | ||
135cba0d | 933 | ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, true, false, false); |
3d3a5b32 | 934 | if (likely(ret == 0 || ret == -ERESTARTSYS)) |
fb1d9738 JB |
935 | return ret; |
936 | ||
8ba5152a TH |
937 | /** |
938 | * If that failed, try VRAM again, this time evicting | |
939 | * previous contents. | |
940 | */ | |
fb1d9738 | 941 | |
135cba0d | 942 | DRM_INFO("Falling through to VRAM.\n"); |
9d87fa21 | 943 | ret = ttm_bo_validate(bo, &vmw_vram_placement, true, false, false); |
fb1d9738 JB |
944 | return ret; |
945 | } | |
946 | ||
947 | ||
948 | static int vmw_validate_buffers(struct vmw_private *dev_priv, | |
949 | struct vmw_sw_context *sw_context) | |
950 | { | |
951 | struct ttm_validate_buffer *entry; | |
952 | int ret; | |
953 | ||
954 | list_for_each_entry(entry, &sw_context->validate_nodes, head) { | |
955 | ret = vmw_validate_single_buffer(dev_priv, entry->bo); | |
956 | if (unlikely(ret != 0)) | |
957 | return ret; | |
958 | } | |
959 | return 0; | |
960 | } | |
961 | ||
be38ab6e TH |
962 | static int vmw_resize_cmd_bounce(struct vmw_sw_context *sw_context, |
963 | uint32_t size) | |
964 | { | |
965 | if (likely(sw_context->cmd_bounce_size >= size)) | |
966 | return 0; | |
967 | ||
968 | if (sw_context->cmd_bounce_size == 0) | |
969 | sw_context->cmd_bounce_size = VMWGFX_CMD_BOUNCE_INIT_SIZE; | |
970 | ||
971 | while (sw_context->cmd_bounce_size < size) { | |
972 | sw_context->cmd_bounce_size = | |
973 | PAGE_ALIGN(sw_context->cmd_bounce_size + | |
974 | (sw_context->cmd_bounce_size >> 1)); | |
975 | } | |
976 | ||
977 | if (sw_context->cmd_bounce != NULL) | |
978 | vfree(sw_context->cmd_bounce); | |
979 | ||
980 | sw_context->cmd_bounce = vmalloc(sw_context->cmd_bounce_size); | |
981 | ||
982 | if (sw_context->cmd_bounce == NULL) { | |
983 | DRM_ERROR("Failed to allocate command bounce buffer.\n"); | |
984 | sw_context->cmd_bounce_size = 0; | |
985 | return -ENOMEM; | |
986 | } | |
987 | ||
988 | return 0; | |
989 | } | |
990 | ||
ae2a1040 TH |
991 | /** |
992 | * vmw_execbuf_fence_commands - create and submit a command stream fence | |
993 | * | |
994 | * Creates a fence object and submits a command stream marker. | |
995 | * If this fails for some reason, We sync the fifo and return NULL. | |
996 | * It is then safe to fence buffers with a NULL pointer. | |
6070e9fa JB |
997 | * |
998 | * If @p_handle is not NULL @file_priv must also not be NULL. Creates | |
999 | * a userspace handle if @p_handle is not NULL, otherwise not. | |
ae2a1040 TH |
1000 | */ |
1001 | ||
1002 | int vmw_execbuf_fence_commands(struct drm_file *file_priv, | |
1003 | struct vmw_private *dev_priv, | |
1004 | struct vmw_fence_obj **p_fence, | |
1005 | uint32_t *p_handle) | |
1006 | { | |
1007 | uint32_t sequence; | |
1008 | int ret; | |
1009 | bool synced = false; | |
1010 | ||
6070e9fa JB |
1011 | /* p_handle implies file_priv. */ |
1012 | BUG_ON(p_handle != NULL && file_priv == NULL); | |
ae2a1040 TH |
1013 | |
1014 | ret = vmw_fifo_send_fence(dev_priv, &sequence); | |
1015 | if (unlikely(ret != 0)) { | |
1016 | DRM_ERROR("Fence submission error. Syncing.\n"); | |
1017 | synced = true; | |
1018 | } | |
1019 | ||
1020 | if (p_handle != NULL) | |
1021 | ret = vmw_user_fence_create(file_priv, dev_priv->fman, | |
1022 | sequence, | |
1023 | DRM_VMW_FENCE_FLAG_EXEC, | |
1024 | p_fence, p_handle); | |
1025 | else | |
1026 | ret = vmw_fence_create(dev_priv->fman, sequence, | |
1027 | DRM_VMW_FENCE_FLAG_EXEC, | |
1028 | p_fence); | |
1029 | ||
1030 | if (unlikely(ret != 0 && !synced)) { | |
1031 | (void) vmw_fallback_wait(dev_priv, false, false, | |
1032 | sequence, false, | |
1033 | VMW_FENCE_WAIT_TIMEOUT); | |
1034 | *p_fence = NULL; | |
1035 | } | |
1036 | ||
1037 | return 0; | |
1038 | } | |
1039 | ||
922ade0d TH |
1040 | int vmw_execbuf_process(struct drm_file *file_priv, |
1041 | struct vmw_private *dev_priv, | |
1042 | void __user *user_commands, | |
1043 | void *kernel_commands, | |
1044 | uint32_t command_size, | |
1045 | uint64_t throttle_us, | |
1046 | struct drm_vmw_fence_rep __user *user_fence_rep) | |
fb1d9738 | 1047 | { |
fb1d9738 | 1048 | struct vmw_sw_context *sw_context = &dev_priv->ctx; |
922ade0d | 1049 | struct drm_vmw_fence_rep fence_rep; |
ae2a1040 TH |
1050 | struct vmw_fence_obj *fence; |
1051 | uint32_t handle; | |
922ade0d TH |
1052 | void *cmd; |
1053 | int ret; | |
fb1d9738 | 1054 | |
922ade0d | 1055 | ret = mutex_lock_interruptible(&dev_priv->cmdbuf_mutex); |
fb1d9738 | 1056 | if (unlikely(ret != 0)) |
922ade0d | 1057 | return -ERESTARTSYS; |
fb1d9738 | 1058 | |
922ade0d TH |
1059 | if (kernel_commands == NULL) { |
1060 | sw_context->kernel = false; | |
fb1d9738 | 1061 | |
922ade0d TH |
1062 | ret = vmw_resize_cmd_bounce(sw_context, command_size); |
1063 | if (unlikely(ret != 0)) | |
1064 | goto out_unlock; | |
fb1d9738 | 1065 | |
fb1d9738 | 1066 | |
922ade0d TH |
1067 | ret = copy_from_user(sw_context->cmd_bounce, |
1068 | user_commands, command_size); | |
1069 | ||
1070 | if (unlikely(ret != 0)) { | |
1071 | ret = -EFAULT; | |
1072 | DRM_ERROR("Failed copying commands.\n"); | |
1073 | goto out_unlock; | |
1074 | } | |
1075 | kernel_commands = sw_context->cmd_bounce; | |
1076 | } else | |
1077 | sw_context->kernel = true; | |
fb1d9738 JB |
1078 | |
1079 | sw_context->tfile = vmw_fpriv(file_priv)->tfile; | |
1080 | sw_context->cid_valid = false; | |
1081 | sw_context->sid_valid = false; | |
1082 | sw_context->cur_reloc = 0; | |
1083 | sw_context->cur_val_buf = 0; | |
e2fa3a76 TH |
1084 | sw_context->fence_flags = 0; |
1085 | INIT_LIST_HEAD(&sw_context->query_list); | |
f18c8840 | 1086 | INIT_LIST_HEAD(&sw_context->resource_list); |
e2fa3a76 TH |
1087 | sw_context->cur_query_bo = dev_priv->pinned_bo; |
1088 | sw_context->cur_query_cid = dev_priv->query_cid; | |
1089 | sw_context->query_cid_valid = (dev_priv->pinned_bo != NULL); | |
fb1d9738 JB |
1090 | |
1091 | INIT_LIST_HEAD(&sw_context->validate_nodes); | |
1092 | ||
922ade0d TH |
1093 | ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands, |
1094 | command_size); | |
fb1d9738 JB |
1095 | if (unlikely(ret != 0)) |
1096 | goto out_err; | |
be38ab6e | 1097 | |
65705962 | 1098 | ret = ttm_eu_reserve_buffers(&sw_context->validate_nodes); |
fb1d9738 JB |
1099 | if (unlikely(ret != 0)) |
1100 | goto out_err; | |
1101 | ||
1102 | ret = vmw_validate_buffers(dev_priv, sw_context); | |
1103 | if (unlikely(ret != 0)) | |
1104 | goto out_err; | |
1105 | ||
1106 | vmw_apply_relocations(sw_context); | |
1925d456 | 1107 | |
922ade0d | 1108 | if (throttle_us) { |
6bcd8d3c | 1109 | ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue, |
922ade0d | 1110 | throttle_us); |
1925d456 TH |
1111 | |
1112 | if (unlikely(ret != 0)) | |
be38ab6e TH |
1113 | goto out_throttle; |
1114 | } | |
1115 | ||
922ade0d | 1116 | cmd = vmw_fifo_reserve(dev_priv, command_size); |
be38ab6e TH |
1117 | if (unlikely(cmd == NULL)) { |
1118 | DRM_ERROR("Failed reserving fifo space for commands.\n"); | |
1119 | ret = -ENOMEM; | |
922ade0d | 1120 | goto out_throttle; |
1925d456 TH |
1121 | } |
1122 | ||
922ade0d TH |
1123 | memcpy(cmd, kernel_commands, command_size); |
1124 | vmw_fifo_commit(dev_priv, command_size); | |
fb1d9738 | 1125 | |
e2fa3a76 | 1126 | vmw_query_bo_switch_commit(dev_priv, sw_context); |
ae2a1040 TH |
1127 | ret = vmw_execbuf_fence_commands(file_priv, dev_priv, |
1128 | &fence, | |
1129 | (user_fence_rep) ? &handle : NULL); | |
fb1d9738 JB |
1130 | /* |
1131 | * This error is harmless, because if fence submission fails, | |
ae2a1040 TH |
1132 | * vmw_fifo_send_fence will sync. The error will be propagated to |
1133 | * user-space in @fence_rep | |
fb1d9738 JB |
1134 | */ |
1135 | ||
1136 | if (ret != 0) | |
1137 | DRM_ERROR("Fence submission error. Syncing.\n"); | |
1138 | ||
ae2a1040 TH |
1139 | ttm_eu_fence_buffer_objects(&sw_context->validate_nodes, |
1140 | (void *) fence); | |
fb1d9738 | 1141 | |
ae2a1040 | 1142 | vmw_clear_validations(sw_context); |
fb1d9738 | 1143 | |
ae2a1040 TH |
1144 | if (user_fence_rep) { |
1145 | fence_rep.error = ret; | |
1146 | fence_rep.handle = handle; | |
1147 | fence_rep.seqno = fence->seqno; | |
1148 | vmw_update_seqno(dev_priv, &dev_priv->fifo); | |
1149 | fence_rep.passed_seqno = dev_priv->last_read_seqno; | |
1150 | ||
1151 | /* | |
1152 | * copy_to_user errors will be detected by user space not | |
1153 | * seeing fence_rep::error filled in. Typically | |
1154 | * user-space would have pre-set that member to -EFAULT. | |
1155 | */ | |
1156 | ret = copy_to_user(user_fence_rep, &fence_rep, | |
1157 | sizeof(fence_rep)); | |
1158 | ||
1159 | /* | |
1160 | * User-space lost the fence object. We need to sync | |
1161 | * and unreference the handle. | |
1162 | */ | |
1163 | if (unlikely(ret != 0) && (fence_rep.error == 0)) { | |
1164 | BUG_ON(fence == NULL); | |
1165 | ||
1166 | ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile, | |
1167 | handle, TTM_REF_USAGE); | |
1168 | DRM_ERROR("Fence copy error. Syncing.\n"); | |
1169 | (void) vmw_fence_obj_wait(fence, | |
1170 | fence->signal_mask, | |
1171 | false, false, | |
1172 | VMW_FENCE_WAIT_TIMEOUT); | |
1173 | } | |
1174 | } | |
fb1d9738 | 1175 | |
ae2a1040 TH |
1176 | if (likely(fence != NULL)) |
1177 | vmw_fence_obj_unreference(&fence); | |
fb1d9738 | 1178 | |
922ade0d | 1179 | mutex_unlock(&dev_priv->cmdbuf_mutex); |
fb1d9738 | 1180 | return 0; |
922ade0d | 1181 | |
fb1d9738 JB |
1182 | out_err: |
1183 | vmw_free_relocations(sw_context); | |
be38ab6e | 1184 | out_throttle: |
e2fa3a76 | 1185 | vmw_query_switch_backoff(sw_context); |
fb1d9738 JB |
1186 | ttm_eu_backoff_reservation(&sw_context->validate_nodes); |
1187 | vmw_clear_validations(sw_context); | |
fb1d9738 JB |
1188 | out_unlock: |
1189 | mutex_unlock(&dev_priv->cmdbuf_mutex); | |
922ade0d TH |
1190 | return ret; |
1191 | } | |
1192 | ||
e2fa3a76 TH |
1193 | /** |
1194 | * vmw_execbuf_unpin_panic - Idle the fifo and unpin the query buffer. | |
1195 | * | |
1196 | * @dev_priv: The device private structure. | |
1197 | * | |
1198 | * This function is called to idle the fifo and unpin the query buffer | |
1199 | * if the normal way to do this hits an error, which should typically be | |
1200 | * extremely rare. | |
1201 | */ | |
1202 | static void vmw_execbuf_unpin_panic(struct vmw_private *dev_priv) | |
1203 | { | |
1204 | DRM_ERROR("Can't unpin query buffer. Trying to recover.\n"); | |
1205 | ||
1206 | (void) vmw_fallback_wait(dev_priv, false, true, 0, false, 10*HZ); | |
1207 | vmw_bo_pin(dev_priv->pinned_bo, false); | |
1208 | vmw_bo_pin(dev_priv->dummy_query_bo, false); | |
1209 | dev_priv->dummy_query_bo_pinned = false; | |
1210 | } | |
1211 | ||
1212 | ||
1213 | /** | |
1214 | * vmw_execbuf_release_pinned_bo - Flush queries and unpin the pinned | |
1215 | * query bo. | |
1216 | * | |
1217 | * @dev_priv: The device private structure. | |
1218 | * @only_on_cid_match: Only flush and unpin if the current active query cid | |
1219 | * matches @cid. | |
1220 | * @cid: Optional context id to match. | |
1221 | * | |
1222 | * This function should be used to unpin the pinned query bo, or | |
1223 | * as a query barrier when we need to make sure that all queries have | |
1224 | * finished before the next fifo command. (For example on hardware | |
1225 | * context destructions where the hardware may otherwise leak unfinished | |
1226 | * queries). | |
1227 | * | |
1228 | * This function does not return any failure codes, but make attempts | |
1229 | * to do safe unpinning in case of errors. | |
1230 | * | |
1231 | * The function will synchronize on the previous query barrier, and will | |
1232 | * thus not finish until that barrier has executed. | |
1233 | */ | |
1234 | void vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv, | |
1235 | bool only_on_cid_match, uint32_t cid) | |
1236 | { | |
1237 | int ret = 0; | |
1238 | struct list_head validate_list; | |
1239 | struct ttm_validate_buffer pinned_val, query_val; | |
1240 | struct vmw_fence_obj *fence; | |
1241 | ||
1242 | mutex_lock(&dev_priv->cmdbuf_mutex); | |
1243 | ||
1244 | if (dev_priv->pinned_bo == NULL) | |
1245 | goto out_unlock; | |
1246 | ||
1247 | if (only_on_cid_match && cid != dev_priv->query_cid) | |
1248 | goto out_unlock; | |
1249 | ||
1250 | INIT_LIST_HEAD(&validate_list); | |
1251 | ||
1252 | pinned_val.new_sync_obj_arg = (void *)(unsigned long) | |
1253 | DRM_VMW_FENCE_FLAG_EXEC; | |
1254 | pinned_val.bo = ttm_bo_reference(dev_priv->pinned_bo); | |
1255 | list_add_tail(&pinned_val.head, &validate_list); | |
1256 | ||
1257 | query_val.new_sync_obj_arg = pinned_val.new_sync_obj_arg; | |
1258 | query_val.bo = ttm_bo_reference(dev_priv->dummy_query_bo); | |
1259 | list_add_tail(&query_val.head, &validate_list); | |
1260 | ||
1261 | do { | |
1262 | ret = ttm_eu_reserve_buffers(&validate_list); | |
1263 | } while (ret == -ERESTARTSYS); | |
1264 | ||
1265 | if (unlikely(ret != 0)) { | |
1266 | vmw_execbuf_unpin_panic(dev_priv); | |
1267 | goto out_no_reserve; | |
1268 | } | |
1269 | ||
1270 | ret = vmw_fifo_emit_dummy_query(dev_priv, dev_priv->query_cid); | |
1271 | if (unlikely(ret != 0)) { | |
1272 | vmw_execbuf_unpin_panic(dev_priv); | |
1273 | goto out_no_emit; | |
1274 | } | |
1275 | ||
1276 | vmw_bo_pin(dev_priv->pinned_bo, false); | |
1277 | vmw_bo_pin(dev_priv->dummy_query_bo, false); | |
1278 | dev_priv->dummy_query_bo_pinned = false; | |
1279 | ||
1280 | (void) vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL); | |
1281 | ttm_eu_fence_buffer_objects(&validate_list, (void *) fence); | |
1282 | ||
1283 | ttm_bo_unref(&query_val.bo); | |
1284 | ttm_bo_unref(&pinned_val.bo); | |
1285 | ttm_bo_unref(&dev_priv->pinned_bo); | |
1286 | ||
1287 | out_unlock: | |
1288 | mutex_unlock(&dev_priv->cmdbuf_mutex); | |
1289 | return; | |
1290 | ||
1291 | out_no_emit: | |
1292 | ttm_eu_backoff_reservation(&validate_list); | |
1293 | out_no_reserve: | |
1294 | ttm_bo_unref(&query_val.bo); | |
1295 | ttm_bo_unref(&pinned_val.bo); | |
1296 | ttm_bo_unref(&dev_priv->pinned_bo); | |
1297 | mutex_unlock(&dev_priv->cmdbuf_mutex); | |
1298 | } | |
1299 | ||
922ade0d TH |
1300 | |
1301 | int vmw_execbuf_ioctl(struct drm_device *dev, void *data, | |
1302 | struct drm_file *file_priv) | |
1303 | { | |
1304 | struct vmw_private *dev_priv = vmw_priv(dev); | |
1305 | struct drm_vmw_execbuf_arg *arg = (struct drm_vmw_execbuf_arg *)data; | |
1306 | struct vmw_master *vmaster = vmw_master(file_priv->master); | |
1307 | int ret; | |
1308 | ||
1309 | /* | |
1310 | * This will allow us to extend the ioctl argument while | |
1311 | * maintaining backwards compatibility: | |
1312 | * We take different code paths depending on the value of | |
1313 | * arg->version. | |
1314 | */ | |
1315 | ||
1316 | if (unlikely(arg->version != DRM_VMW_EXECBUF_VERSION)) { | |
1317 | DRM_ERROR("Incorrect execbuf version.\n"); | |
1318 | DRM_ERROR("You're running outdated experimental " | |
1319 | "vmwgfx user-space drivers."); | |
1320 | return -EINVAL; | |
1321 | } | |
1322 | ||
1323 | ret = ttm_read_lock(&vmaster->lock, true); | |
1324 | if (unlikely(ret != 0)) | |
1325 | return ret; | |
1326 | ||
1327 | ret = vmw_execbuf_process(file_priv, dev_priv, | |
1328 | (void __user *)(unsigned long)arg->commands, | |
1329 | NULL, arg->command_size, arg->throttle_us, | |
1330 | (void __user *)(unsigned long)arg->fence_rep); | |
1331 | ||
1332 | if (unlikely(ret != 0)) | |
1333 | goto out_unlock; | |
1334 | ||
1335 | vmw_kms_cursor_post_execbuf(dev_priv); | |
1336 | ||
1337 | out_unlock: | |
fb1d9738 JB |
1338 | ttm_read_unlock(&vmaster->lock); |
1339 | return ret; | |
1340 | } |