542c852f8eb592b989b82b5a396bff41e7be07cb
[pandora-kernel.git] / drivers / gpu / drm / vmwgfx / vmwgfx_execbuf.c
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
47
48 static int vmw_resource_to_validate_list(struct vmw_sw_context *sw_context,
49                                          struct vmw_resource **p_res)
50 {
51         int ret = 0;
52         struct vmw_resource *res = *p_res;
53
54         if (!res->on_validate_list) {
55                 if (sw_context->num_ref_resources >= VMWGFX_MAX_VALIDATIONS) {
56                         DRM_ERROR("Too many resources referenced in "
57                                   "command stream.\n");
58                         ret = -ENOMEM;
59                         goto out;
60                 }
61                 sw_context->resources[sw_context->num_ref_resources++] = res;
62                 res->on_validate_list = true;
63                 return 0;
64         }
65
66 out:
67         vmw_resource_unreference(p_res);
68         return ret;
69 }
70
71 static int vmw_cmd_cid_check(struct vmw_private *dev_priv,
72                              struct vmw_sw_context *sw_context,
73                              SVGA3dCmdHeader *header)
74 {
75         struct vmw_resource *ctx;
76
77         struct vmw_cid_cmd {
78                 SVGA3dCmdHeader header;
79                 __le32 cid;
80         } *cmd;
81         int ret;
82
83         cmd = container_of(header, struct vmw_cid_cmd, header);
84         if (likely(sw_context->cid_valid && cmd->cid == sw_context->last_cid))
85                 return 0;
86
87         ret = vmw_context_check(dev_priv, sw_context->tfile, cmd->cid,
88                                 &ctx);
89         if (unlikely(ret != 0)) {
90                 DRM_ERROR("Could not find or use context %u\n",
91                           (unsigned) cmd->cid);
92                 return ret;
93         }
94
95         sw_context->last_cid = cmd->cid;
96         sw_context->cid_valid = true;
97         return vmw_resource_to_validate_list(sw_context, &ctx);
98 }
99
100 static int vmw_cmd_sid_check(struct vmw_private *dev_priv,
101                              struct vmw_sw_context *sw_context,
102                              uint32_t *sid)
103 {
104         struct vmw_surface *srf;
105         int ret;
106         struct vmw_resource *res;
107
108         if (*sid == SVGA3D_INVALID_ID)
109                 return 0;
110
111         if (likely((sw_context->sid_valid  &&
112                       *sid == sw_context->last_sid))) {
113                 *sid = sw_context->sid_translation;
114                 return 0;
115         }
116
117         ret = vmw_user_surface_lookup_handle(dev_priv, sw_context->tfile,
118                                              *sid, &srf);
119         if (unlikely(ret != 0)) {
120                 DRM_ERROR("Could ot find or use surface 0x%08x "
121                           "address 0x%08lx\n",
122                           (unsigned int) *sid,
123                           (unsigned long) sid);
124                 return ret;
125         }
126
127         sw_context->last_sid = *sid;
128         sw_context->sid_valid = true;
129         sw_context->sid_translation = srf->res.id;
130         *sid = sw_context->sid_translation;
131
132         res = &srf->res;
133         return vmw_resource_to_validate_list(sw_context, &res);
134 }
135
136
137 static int vmw_cmd_set_render_target_check(struct vmw_private *dev_priv,
138                                            struct vmw_sw_context *sw_context,
139                                            SVGA3dCmdHeader *header)
140 {
141         struct vmw_sid_cmd {
142                 SVGA3dCmdHeader header;
143                 SVGA3dCmdSetRenderTarget body;
144         } *cmd;
145         int ret;
146
147         ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
148         if (unlikely(ret != 0))
149                 return ret;
150
151         cmd = container_of(header, struct vmw_sid_cmd, header);
152         ret = vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.target.sid);
153         return ret;
154 }
155
156 static int vmw_cmd_surface_copy_check(struct vmw_private *dev_priv,
157                                       struct vmw_sw_context *sw_context,
158                                       SVGA3dCmdHeader *header)
159 {
160         struct vmw_sid_cmd {
161                 SVGA3dCmdHeader header;
162                 SVGA3dCmdSurfaceCopy body;
163         } *cmd;
164         int ret;
165
166         cmd = container_of(header, struct vmw_sid_cmd, header);
167         ret = vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.src.sid);
168         if (unlikely(ret != 0))
169                 return ret;
170         return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.dest.sid);
171 }
172
173 static int vmw_cmd_stretch_blt_check(struct vmw_private *dev_priv,
174                                      struct vmw_sw_context *sw_context,
175                                      SVGA3dCmdHeader *header)
176 {
177         struct vmw_sid_cmd {
178                 SVGA3dCmdHeader header;
179                 SVGA3dCmdSurfaceStretchBlt body;
180         } *cmd;
181         int ret;
182
183         cmd = container_of(header, struct vmw_sid_cmd, header);
184         ret = vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.src.sid);
185         if (unlikely(ret != 0))
186                 return ret;
187         return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.dest.sid);
188 }
189
190 static int vmw_cmd_blt_surf_screen_check(struct vmw_private *dev_priv,
191                                          struct vmw_sw_context *sw_context,
192                                          SVGA3dCmdHeader *header)
193 {
194         struct vmw_sid_cmd {
195                 SVGA3dCmdHeader header;
196                 SVGA3dCmdBlitSurfaceToScreen body;
197         } *cmd;
198
199         cmd = container_of(header, struct vmw_sid_cmd, header);
200         return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.srcImage.sid);
201 }
202
203 static int vmw_cmd_present_check(struct vmw_private *dev_priv,
204                                  struct vmw_sw_context *sw_context,
205                                  SVGA3dCmdHeader *header)
206 {
207         struct vmw_sid_cmd {
208                 SVGA3dCmdHeader header;
209                 SVGA3dCmdPresent body;
210         } *cmd;
211
212         cmd = container_of(header, struct vmw_sid_cmd, header);
213         return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.sid);
214 }
215
216 static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
217                                    struct vmw_sw_context *sw_context,
218                                    SVGAGuestPtr *ptr,
219                                    struct vmw_dma_buffer **vmw_bo_p)
220 {
221         struct vmw_dma_buffer *vmw_bo = NULL;
222         struct ttm_buffer_object *bo;
223         uint32_t handle = ptr->gmrId;
224         struct vmw_relocation *reloc;
225         uint32_t cur_validate_node;
226         struct ttm_validate_buffer *val_buf;
227         int ret;
228
229         ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo);
230         if (unlikely(ret != 0)) {
231                 DRM_ERROR("Could not find or use GMR region.\n");
232                 return -EINVAL;
233         }
234         bo = &vmw_bo->base;
235
236         if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
237                 DRM_ERROR("Max number relocations per submission"
238                           " exceeded\n");
239                 ret = -EINVAL;
240                 goto out_no_reloc;
241         }
242
243         reloc = &sw_context->relocs[sw_context->cur_reloc++];
244         reloc->location = ptr;
245
246         cur_validate_node = vmw_dmabuf_validate_node(bo, sw_context->cur_val_buf);
247         if (unlikely(cur_validate_node >= VMWGFX_MAX_VALIDATIONS)) {
248                 DRM_ERROR("Max number of DMA buffers per submission"
249                           " exceeded.\n");
250                 ret = -EINVAL;
251                 goto out_no_reloc;
252         }
253
254         reloc->index = cur_validate_node;
255         if (unlikely(cur_validate_node == sw_context->cur_val_buf)) {
256                 val_buf = &sw_context->val_bufs[cur_validate_node];
257                 val_buf->bo = ttm_bo_reference(bo);
258                 val_buf->usage = TTM_USAGE_READWRITE;
259                 val_buf->new_sync_obj_arg = (void *) DRM_VMW_FENCE_FLAG_EXEC;
260                 list_add_tail(&val_buf->head, &sw_context->validate_nodes);
261                 ++sw_context->cur_val_buf;
262         }
263         *vmw_bo_p = vmw_bo;
264         return 0;
265
266 out_no_reloc:
267         vmw_dmabuf_unreference(&vmw_bo);
268         vmw_bo_p = NULL;
269         return ret;
270 }
271
272 static int vmw_cmd_end_query(struct vmw_private *dev_priv,
273                              struct vmw_sw_context *sw_context,
274                              SVGA3dCmdHeader *header)
275 {
276         struct vmw_dma_buffer *vmw_bo;
277         struct vmw_query_cmd {
278                 SVGA3dCmdHeader header;
279                 SVGA3dCmdEndQuery q;
280         } *cmd;
281         int ret;
282
283         cmd = container_of(header, struct vmw_query_cmd, header);
284         ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
285         if (unlikely(ret != 0))
286                 return ret;
287
288         ret = vmw_translate_guest_ptr(dev_priv, sw_context,
289                                       &cmd->q.guestResult,
290                                       &vmw_bo);
291         if (unlikely(ret != 0))
292                 return ret;
293
294         vmw_dmabuf_unreference(&vmw_bo);
295         return 0;
296 }
297
298 static int vmw_cmd_wait_query(struct vmw_private *dev_priv,
299                               struct vmw_sw_context *sw_context,
300                               SVGA3dCmdHeader *header)
301 {
302         struct vmw_dma_buffer *vmw_bo;
303         struct vmw_query_cmd {
304                 SVGA3dCmdHeader header;
305                 SVGA3dCmdWaitForQuery q;
306         } *cmd;
307         int ret;
308
309         cmd = container_of(header, struct vmw_query_cmd, header);
310         ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
311         if (unlikely(ret != 0))
312                 return ret;
313
314         ret = vmw_translate_guest_ptr(dev_priv, sw_context,
315                                       &cmd->q.guestResult,
316                                       &vmw_bo);
317         if (unlikely(ret != 0))
318                 return ret;
319
320         vmw_dmabuf_unreference(&vmw_bo);
321         return 0;
322 }
323
324 static int vmw_cmd_dma(struct vmw_private *dev_priv,
325                        struct vmw_sw_context *sw_context,
326                        SVGA3dCmdHeader *header)
327 {
328         struct vmw_dma_buffer *vmw_bo = NULL;
329         struct ttm_buffer_object *bo;
330         struct vmw_surface *srf = NULL;
331         struct vmw_dma_cmd {
332                 SVGA3dCmdHeader header;
333                 SVGA3dCmdSurfaceDMA dma;
334         } *cmd;
335         int ret;
336         struct vmw_resource *res;
337
338         cmd = container_of(header, struct vmw_dma_cmd, header);
339         ret = vmw_translate_guest_ptr(dev_priv, sw_context,
340                                       &cmd->dma.guest.ptr,
341                                       &vmw_bo);
342         if (unlikely(ret != 0))
343                 return ret;
344
345         bo = &vmw_bo->base;
346         ret = vmw_user_surface_lookup_handle(dev_priv, sw_context->tfile,
347                                              cmd->dma.host.sid, &srf);
348         if (ret) {
349                 DRM_ERROR("could not find surface\n");
350                 goto out_no_reloc;
351         }
352
353         /*
354          * Patch command stream with device SID.
355          */
356         cmd->dma.host.sid = srf->res.id;
357         vmw_kms_cursor_snoop(srf, sw_context->tfile, bo, header);
358
359         vmw_dmabuf_unreference(&vmw_bo);
360
361         res = &srf->res;
362         return vmw_resource_to_validate_list(sw_context, &res);
363
364 out_no_reloc:
365         vmw_dmabuf_unreference(&vmw_bo);
366         return ret;
367 }
368
369 static int vmw_cmd_draw(struct vmw_private *dev_priv,
370                         struct vmw_sw_context *sw_context,
371                         SVGA3dCmdHeader *header)
372 {
373         struct vmw_draw_cmd {
374                 SVGA3dCmdHeader header;
375                 SVGA3dCmdDrawPrimitives body;
376         } *cmd;
377         SVGA3dVertexDecl *decl = (SVGA3dVertexDecl *)(
378                 (unsigned long)header + sizeof(*cmd));
379         SVGA3dPrimitiveRange *range;
380         uint32_t i;
381         uint32_t maxnum;
382         int ret;
383
384         ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
385         if (unlikely(ret != 0))
386                 return ret;
387
388         cmd = container_of(header, struct vmw_draw_cmd, header);
389         maxnum = (header->size - sizeof(cmd->body)) / sizeof(*decl);
390
391         if (unlikely(cmd->body.numVertexDecls > maxnum)) {
392                 DRM_ERROR("Illegal number of vertex declarations.\n");
393                 return -EINVAL;
394         }
395
396         for (i = 0; i < cmd->body.numVertexDecls; ++i, ++decl) {
397                 ret = vmw_cmd_sid_check(dev_priv, sw_context,
398                                         &decl->array.surfaceId);
399                 if (unlikely(ret != 0))
400                         return ret;
401         }
402
403         maxnum = (header->size - sizeof(cmd->body) -
404                   cmd->body.numVertexDecls * sizeof(*decl)) / sizeof(*range);
405         if (unlikely(cmd->body.numRanges > maxnum)) {
406                 DRM_ERROR("Illegal number of index ranges.\n");
407                 return -EINVAL;
408         }
409
410         range = (SVGA3dPrimitiveRange *) decl;
411         for (i = 0; i < cmd->body.numRanges; ++i, ++range) {
412                 ret = vmw_cmd_sid_check(dev_priv, sw_context,
413                                         &range->indexArray.surfaceId);
414                 if (unlikely(ret != 0))
415                         return ret;
416         }
417         return 0;
418 }
419
420
421 static int vmw_cmd_tex_state(struct vmw_private *dev_priv,
422                              struct vmw_sw_context *sw_context,
423                              SVGA3dCmdHeader *header)
424 {
425         struct vmw_tex_state_cmd {
426                 SVGA3dCmdHeader header;
427                 SVGA3dCmdSetTextureState state;
428         };
429
430         SVGA3dTextureState *last_state = (SVGA3dTextureState *)
431           ((unsigned long) header + header->size + sizeof(header));
432         SVGA3dTextureState *cur_state = (SVGA3dTextureState *)
433                 ((unsigned long) header + sizeof(struct vmw_tex_state_cmd));
434         int ret;
435
436         ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
437         if (unlikely(ret != 0))
438                 return ret;
439
440         for (; cur_state < last_state; ++cur_state) {
441                 if (likely(cur_state->name != SVGA3D_TS_BIND_TEXTURE))
442                         continue;
443
444                 ret = vmw_cmd_sid_check(dev_priv, sw_context,
445                                         &cur_state->value);
446                 if (unlikely(ret != 0))
447                         return ret;
448         }
449
450         return 0;
451 }
452
453
454 typedef int (*vmw_cmd_func) (struct vmw_private *,
455                              struct vmw_sw_context *,
456                              SVGA3dCmdHeader *);
457
458 #define VMW_CMD_DEF(cmd, func) \
459         [cmd - SVGA_3D_CMD_BASE] = func
460
461 static vmw_cmd_func vmw_cmd_funcs[SVGA_3D_CMD_MAX] = {
462         VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE, &vmw_cmd_invalid),
463         VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DESTROY, &vmw_cmd_invalid),
464         VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_COPY, &vmw_cmd_surface_copy_check),
465         VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_STRETCHBLT, &vmw_cmd_stretch_blt_check),
466         VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DMA, &vmw_cmd_dma),
467         VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DEFINE, &vmw_cmd_invalid),
468         VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DESTROY, &vmw_cmd_invalid),
469         VMW_CMD_DEF(SVGA_3D_CMD_SETTRANSFORM, &vmw_cmd_cid_check),
470         VMW_CMD_DEF(SVGA_3D_CMD_SETZRANGE, &vmw_cmd_cid_check),
471         VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERSTATE, &vmw_cmd_cid_check),
472         VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERTARGET,
473                     &vmw_cmd_set_render_target_check),
474         VMW_CMD_DEF(SVGA_3D_CMD_SETTEXTURESTATE, &vmw_cmd_tex_state),
475         VMW_CMD_DEF(SVGA_3D_CMD_SETMATERIAL, &vmw_cmd_cid_check),
476         VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTDATA, &vmw_cmd_cid_check),
477         VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTENABLED, &vmw_cmd_cid_check),
478         VMW_CMD_DEF(SVGA_3D_CMD_SETVIEWPORT, &vmw_cmd_cid_check),
479         VMW_CMD_DEF(SVGA_3D_CMD_SETCLIPPLANE, &vmw_cmd_cid_check),
480         VMW_CMD_DEF(SVGA_3D_CMD_CLEAR, &vmw_cmd_cid_check),
481         VMW_CMD_DEF(SVGA_3D_CMD_PRESENT, &vmw_cmd_present_check),
482         VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DEFINE, &vmw_cmd_cid_check),
483         VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_cid_check),
484         VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER, &vmw_cmd_cid_check),
485         VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER_CONST, &vmw_cmd_cid_check),
486         VMW_CMD_DEF(SVGA_3D_CMD_DRAW_PRIMITIVES, &vmw_cmd_draw),
487         VMW_CMD_DEF(SVGA_3D_CMD_SETSCISSORRECT, &vmw_cmd_cid_check),
488         VMW_CMD_DEF(SVGA_3D_CMD_BEGIN_QUERY, &vmw_cmd_cid_check),
489         VMW_CMD_DEF(SVGA_3D_CMD_END_QUERY, &vmw_cmd_end_query),
490         VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_QUERY, &vmw_cmd_wait_query),
491         VMW_CMD_DEF(SVGA_3D_CMD_PRESENT_READBACK, &vmw_cmd_ok),
492         VMW_CMD_DEF(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN,
493                     &vmw_cmd_blt_surf_screen_check)
494 };
495
496 static int vmw_cmd_check(struct vmw_private *dev_priv,
497                          struct vmw_sw_context *sw_context,
498                          void *buf, uint32_t *size)
499 {
500         uint32_t cmd_id;
501         uint32_t size_remaining = *size;
502         SVGA3dCmdHeader *header = (SVGA3dCmdHeader *) buf;
503         int ret;
504
505         cmd_id = ((uint32_t *)buf)[0];
506         if (cmd_id == SVGA_CMD_UPDATE) {
507                 *size = 5 << 2;
508                 return 0;
509         }
510
511         cmd_id = le32_to_cpu(header->id);
512         *size = le32_to_cpu(header->size) + sizeof(SVGA3dCmdHeader);
513
514         cmd_id -= SVGA_3D_CMD_BASE;
515         if (unlikely(*size > size_remaining))
516                 goto out_err;
517
518         if (unlikely(cmd_id >= SVGA_3D_CMD_MAX - SVGA_3D_CMD_BASE))
519                 goto out_err;
520
521         ret = vmw_cmd_funcs[cmd_id](dev_priv, sw_context, header);
522         if (unlikely(ret != 0))
523                 goto out_err;
524
525         return 0;
526 out_err:
527         DRM_ERROR("Illegal / Invalid SVGA3D command: %d\n",
528                   cmd_id + SVGA_3D_CMD_BASE);
529         return -EINVAL;
530 }
531
532 static int vmw_cmd_check_all(struct vmw_private *dev_priv,
533                              struct vmw_sw_context *sw_context,
534                              void *buf,
535                              uint32_t size)
536 {
537         int32_t cur_size = size;
538         int ret;
539
540         while (cur_size > 0) {
541                 size = cur_size;
542                 ret = vmw_cmd_check(dev_priv, sw_context, buf, &size);
543                 if (unlikely(ret != 0))
544                         return ret;
545                 buf = (void *)((unsigned long) buf + size);
546                 cur_size -= size;
547         }
548
549         if (unlikely(cur_size != 0)) {
550                 DRM_ERROR("Command verifier out of sync.\n");
551                 return -EINVAL;
552         }
553
554         return 0;
555 }
556
557 static void vmw_free_relocations(struct vmw_sw_context *sw_context)
558 {
559         sw_context->cur_reloc = 0;
560 }
561
562 static void vmw_apply_relocations(struct vmw_sw_context *sw_context)
563 {
564         uint32_t i;
565         struct vmw_relocation *reloc;
566         struct ttm_validate_buffer *validate;
567         struct ttm_buffer_object *bo;
568
569         for (i = 0; i < sw_context->cur_reloc; ++i) {
570                 reloc = &sw_context->relocs[i];
571                 validate = &sw_context->val_bufs[reloc->index];
572                 bo = validate->bo;
573                 if (bo->mem.mem_type == TTM_PL_VRAM) {
574                         reloc->location->offset += bo->offset;
575                         reloc->location->gmrId = SVGA_GMR_FRAMEBUFFER;
576                 } else
577                         reloc->location->gmrId = bo->mem.start;
578         }
579         vmw_free_relocations(sw_context);
580 }
581
582 static void vmw_clear_validations(struct vmw_sw_context *sw_context)
583 {
584         struct ttm_validate_buffer *entry, *next;
585         uint32_t i = sw_context->num_ref_resources;
586
587         /*
588          * Drop references to DMA buffers held during command submission.
589          */
590         list_for_each_entry_safe(entry, next, &sw_context->validate_nodes,
591                                  head) {
592                 list_del(&entry->head);
593                 vmw_dmabuf_validate_clear(entry->bo);
594                 ttm_bo_unref(&entry->bo);
595                 sw_context->cur_val_buf--;
596         }
597         BUG_ON(sw_context->cur_val_buf != 0);
598
599         /*
600          * Drop references to resources held during command submission.
601          */
602         while (i-- > 0) {
603                 sw_context->resources[i]->on_validate_list = false;
604                 vmw_resource_unreference(&sw_context->resources[i]);
605         }
606 }
607
608 static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
609                                       struct ttm_buffer_object *bo)
610 {
611         int ret;
612
613         /**
614          * Put BO in VRAM if there is space, otherwise as a GMR.
615          * If there is no space in VRAM and GMR ids are all used up,
616          * start evicting GMRs to make room. If the DMA buffer can't be
617          * used as a GMR, this will return -ENOMEM.
618          */
619
620         ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, true, false, false);
621         if (likely(ret == 0 || ret == -ERESTARTSYS))
622                 return ret;
623
624         /**
625          * If that failed, try VRAM again, this time evicting
626          * previous contents.
627          */
628
629         DRM_INFO("Falling through to VRAM.\n");
630         ret = ttm_bo_validate(bo, &vmw_vram_placement, true, false, false);
631         return ret;
632 }
633
634
635 static int vmw_validate_buffers(struct vmw_private *dev_priv,
636                                 struct vmw_sw_context *sw_context)
637 {
638         struct ttm_validate_buffer *entry;
639         int ret;
640
641         list_for_each_entry(entry, &sw_context->validate_nodes, head) {
642                 ret = vmw_validate_single_buffer(dev_priv, entry->bo);
643                 if (unlikely(ret != 0))
644                         return ret;
645         }
646         return 0;
647 }
648
649 static int vmw_resize_cmd_bounce(struct vmw_sw_context *sw_context,
650                                  uint32_t size)
651 {
652         if (likely(sw_context->cmd_bounce_size >= size))
653                 return 0;
654
655         if (sw_context->cmd_bounce_size == 0)
656                 sw_context->cmd_bounce_size = VMWGFX_CMD_BOUNCE_INIT_SIZE;
657
658         while (sw_context->cmd_bounce_size < size) {
659                 sw_context->cmd_bounce_size =
660                         PAGE_ALIGN(sw_context->cmd_bounce_size +
661                                    (sw_context->cmd_bounce_size >> 1));
662         }
663
664         if (sw_context->cmd_bounce != NULL)
665                 vfree(sw_context->cmd_bounce);
666
667         sw_context->cmd_bounce = vmalloc(sw_context->cmd_bounce_size);
668
669         if (sw_context->cmd_bounce == NULL) {
670                 DRM_ERROR("Failed to allocate command bounce buffer.\n");
671                 sw_context->cmd_bounce_size = 0;
672                 return -ENOMEM;
673         }
674
675         return 0;
676 }
677
678 /**
679  * vmw_execbuf_fence_commands - create and submit a command stream fence
680  *
681  * Creates a fence object and submits a command stream marker.
682  * If this fails for some reason, We sync the fifo and return NULL.
683  * It is then safe to fence buffers with a NULL pointer.
684  *
685  * If @p_handle is not NULL @file_priv must also not be NULL. Creates
686  * a userspace handle if @p_handle is not NULL, otherwise not.
687  */
688
689 int vmw_execbuf_fence_commands(struct drm_file *file_priv,
690                                struct vmw_private *dev_priv,
691                                struct vmw_fence_obj **p_fence,
692                                uint32_t *p_handle)
693 {
694         uint32_t sequence;
695         int ret;
696         bool synced = false;
697
698         /* p_handle implies file_priv. */
699         BUG_ON(p_handle != NULL && file_priv == NULL);
700
701         ret = vmw_fifo_send_fence(dev_priv, &sequence);
702         if (unlikely(ret != 0)) {
703                 DRM_ERROR("Fence submission error. Syncing.\n");
704                 synced = true;
705         }
706
707         if (p_handle != NULL)
708                 ret = vmw_user_fence_create(file_priv, dev_priv->fman,
709                                             sequence,
710                                             DRM_VMW_FENCE_FLAG_EXEC,
711                                             p_fence, p_handle);
712         else
713                 ret = vmw_fence_create(dev_priv->fman, sequence,
714                                        DRM_VMW_FENCE_FLAG_EXEC,
715                                        p_fence);
716
717         if (unlikely(ret != 0 && !synced)) {
718                 (void) vmw_fallback_wait(dev_priv, false, false,
719                                          sequence, false,
720                                          VMW_FENCE_WAIT_TIMEOUT);
721                 *p_fence = NULL;
722         }
723
724         return 0;
725 }
726
727 int vmw_execbuf_process(struct drm_file *file_priv,
728                         struct vmw_private *dev_priv,
729                         void __user *user_commands,
730                         void *kernel_commands,
731                         uint32_t command_size,
732                         uint64_t throttle_us,
733                         struct drm_vmw_fence_rep __user *user_fence_rep)
734 {
735         struct vmw_sw_context *sw_context = &dev_priv->ctx;
736         struct drm_vmw_fence_rep fence_rep;
737         struct vmw_fence_obj *fence;
738         uint32_t handle;
739         void *cmd;
740         int ret;
741
742         ret = mutex_lock_interruptible(&dev_priv->cmdbuf_mutex);
743         if (unlikely(ret != 0))
744                 return -ERESTARTSYS;
745
746         if (kernel_commands == NULL) {
747                 sw_context->kernel = false;
748
749                 ret = vmw_resize_cmd_bounce(sw_context, command_size);
750                 if (unlikely(ret != 0))
751                         goto out_unlock;
752
753
754                 ret = copy_from_user(sw_context->cmd_bounce,
755                                      user_commands, command_size);
756
757                 if (unlikely(ret != 0)) {
758                         ret = -EFAULT;
759                         DRM_ERROR("Failed copying commands.\n");
760                         goto out_unlock;
761                 }
762                 kernel_commands = sw_context->cmd_bounce;
763         } else
764                 sw_context->kernel = true;
765
766         sw_context->tfile = vmw_fpriv(file_priv)->tfile;
767         sw_context->cid_valid = false;
768         sw_context->sid_valid = false;
769         sw_context->cur_reloc = 0;
770         sw_context->cur_val_buf = 0;
771         sw_context->num_ref_resources = 0;
772
773         INIT_LIST_HEAD(&sw_context->validate_nodes);
774
775         ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands,
776                                 command_size);
777         if (unlikely(ret != 0))
778                 goto out_err;
779
780         ret = ttm_eu_reserve_buffers(&sw_context->validate_nodes);
781         if (unlikely(ret != 0))
782                 goto out_err;
783
784         ret = vmw_validate_buffers(dev_priv, sw_context);
785         if (unlikely(ret != 0))
786                 goto out_err;
787
788         vmw_apply_relocations(sw_context);
789
790         if (throttle_us) {
791                 ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue,
792                                    throttle_us);
793
794                 if (unlikely(ret != 0))
795                         goto out_throttle;
796         }
797
798         cmd = vmw_fifo_reserve(dev_priv, command_size);
799         if (unlikely(cmd == NULL)) {
800                 DRM_ERROR("Failed reserving fifo space for commands.\n");
801                 ret = -ENOMEM;
802                 goto out_throttle;
803         }
804
805         memcpy(cmd, kernel_commands, command_size);
806         vmw_fifo_commit(dev_priv, command_size);
807
808         ret = vmw_execbuf_fence_commands(file_priv, dev_priv,
809                                          &fence,
810                                          (user_fence_rep) ? &handle : NULL);
811         /*
812          * This error is harmless, because if fence submission fails,
813          * vmw_fifo_send_fence will sync. The error will be propagated to
814          * user-space in @fence_rep
815          */
816
817         if (ret != 0)
818                 DRM_ERROR("Fence submission error. Syncing.\n");
819
820         ttm_eu_fence_buffer_objects(&sw_context->validate_nodes,
821                                     (void *) fence);
822
823         vmw_clear_validations(sw_context);
824
825         if (user_fence_rep) {
826                 fence_rep.error = ret;
827                 fence_rep.handle = handle;
828                 fence_rep.seqno = fence->seqno;
829                 vmw_update_seqno(dev_priv, &dev_priv->fifo);
830                 fence_rep.passed_seqno = dev_priv->last_read_seqno;
831
832                 /*
833                  * copy_to_user errors will be detected by user space not
834                  * seeing fence_rep::error filled in. Typically
835                  * user-space would have pre-set that member to -EFAULT.
836                  */
837                 ret = copy_to_user(user_fence_rep, &fence_rep,
838                                    sizeof(fence_rep));
839
840                 /*
841                  * User-space lost the fence object. We need to sync
842                  * and unreference the handle.
843                  */
844                 if (unlikely(ret != 0) && (fence_rep.error == 0)) {
845                         BUG_ON(fence == NULL);
846
847                         ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile,
848                                                   handle, TTM_REF_USAGE);
849                         DRM_ERROR("Fence copy error. Syncing.\n");
850                         (void) vmw_fence_obj_wait(fence,
851                                                   fence->signal_mask,
852                                                   false, false,
853                                                   VMW_FENCE_WAIT_TIMEOUT);
854                 }
855         }
856
857         if (likely(fence != NULL))
858                 vmw_fence_obj_unreference(&fence);
859
860         mutex_unlock(&dev_priv->cmdbuf_mutex);
861         return 0;
862
863 out_err:
864         vmw_free_relocations(sw_context);
865 out_throttle:
866         ttm_eu_backoff_reservation(&sw_context->validate_nodes);
867         vmw_clear_validations(sw_context);
868 out_unlock:
869         mutex_unlock(&dev_priv->cmdbuf_mutex);
870         return ret;
871 }
872
873
874 int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
875                       struct drm_file *file_priv)
876 {
877         struct vmw_private *dev_priv = vmw_priv(dev);
878         struct drm_vmw_execbuf_arg *arg = (struct drm_vmw_execbuf_arg *)data;
879         struct vmw_master *vmaster = vmw_master(file_priv->master);
880         int ret;
881
882         /*
883          * This will allow us to extend the ioctl argument while
884          * maintaining backwards compatibility:
885          * We take different code paths depending on the value of
886          * arg->version.
887          */
888
889         if (unlikely(arg->version != DRM_VMW_EXECBUF_VERSION)) {
890                 DRM_ERROR("Incorrect execbuf version.\n");
891                 DRM_ERROR("You're running outdated experimental "
892                           "vmwgfx user-space drivers.");
893                 return -EINVAL;
894         }
895
896         ret = ttm_read_lock(&vmaster->lock, true);
897         if (unlikely(ret != 0))
898                 return ret;
899
900         ret = vmw_execbuf_process(file_priv, dev_priv,
901                                   (void __user *)(unsigned long)arg->commands,
902                                   NULL, arg->command_size, arg->throttle_us,
903                                   (void __user *)(unsigned long)arg->fence_rep);
904
905         if (unlikely(ret != 0))
906                 goto out_unlock;
907
908         vmw_kms_cursor_post_execbuf(dev_priv);
909
910 out_unlock:
911         ttm_read_unlock(&vmaster->lock);
912         return ret;
913 }