Commit | Line | Data |
---|---|---|
54cf91dc CW |
1 | /* |
2 | * Copyright © 2008,2010 Intel Corporation | |
3 | * | |
4 | * Permission is hereby granted, free of charge, to any person obtaining a | |
5 | * copy of this software and associated documentation files (the "Software"), | |
6 | * to deal in the Software without restriction, including without limitation | |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
8 | * and/or sell copies of the Software, and to permit persons to whom the | |
9 | * Software is furnished to do so, subject to the following conditions: | |
10 | * | |
11 | * The above copyright notice and this permission notice (including the next | |
12 | * paragraph) shall be included in all copies or substantial portions of the | |
13 | * Software. | |
14 | * | |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
21 | * IN THE SOFTWARE. | |
22 | * | |
23 | * Authors: | |
24 | * Eric Anholt <eric@anholt.net> | |
25 | * Chris Wilson <chris@chris-wilson.co.uk> | |
26 | * | |
27 | */ | |
28 | ||
ad778f89 CW |
29 | #include <linux/dma_remapping.h> |
30 | #include <linux/reservation.h> | |
31 | #include <linux/uaccess.h> | |
32 | ||
760285e7 DH |
33 | #include <drm/drmP.h> |
34 | #include <drm/i915_drm.h> | |
ad778f89 | 35 | |
54cf91dc | 36 | #include "i915_drv.h" |
ad778f89 | 37 | #include "i915_gem_dmabuf.h" |
54cf91dc CW |
38 | #include "i915_trace.h" |
39 | #include "intel_drv.h" | |
5d723d7a | 40 | #include "intel_frontbuffer.h" |
54cf91dc | 41 | |
9e2793f6 DG |
42 | #define __EXEC_OBJECT_HAS_PIN (1<<31) |
43 | #define __EXEC_OBJECT_HAS_FENCE (1<<30) | |
44 | #define __EXEC_OBJECT_NEEDS_MAP (1<<29) | |
45 | #define __EXEC_OBJECT_NEEDS_BIAS (1<<28) | |
46 | #define __EXEC_OBJECT_INTERNAL_FLAGS (0xf<<28) /* all of the above */ | |
d23db88c CW |
47 | |
48 | #define BATCH_OFFSET_BIAS (256*1024) | |
a415d355 | 49 | |
5b043f4e CW |
50 | struct i915_execbuffer_params { |
51 | struct drm_device *dev; | |
52 | struct drm_file *file; | |
59bfa124 CW |
53 | struct i915_vma *batch; |
54 | u32 dispatch_flags; | |
55 | u32 args_batch_start_offset; | |
5b043f4e | 56 | struct intel_engine_cs *engine; |
5b043f4e CW |
57 | struct i915_gem_context *ctx; |
58 | struct drm_i915_gem_request *request; | |
59 | }; | |
60 | ||
27173f1f BW |
61 | struct eb_vmas { |
62 | struct list_head vmas; | |
67731b87 | 63 | int and; |
eef90ccb | 64 | union { |
27173f1f | 65 | struct i915_vma *lut[0]; |
eef90ccb CW |
66 | struct hlist_head buckets[0]; |
67 | }; | |
67731b87 CW |
68 | }; |
69 | ||
27173f1f | 70 | static struct eb_vmas * |
17601cbc | 71 | eb_create(struct drm_i915_gem_execbuffer2 *args) |
67731b87 | 72 | { |
27173f1f | 73 | struct eb_vmas *eb = NULL; |
eef90ccb CW |
74 | |
75 | if (args->flags & I915_EXEC_HANDLE_LUT) { | |
b205ca57 | 76 | unsigned size = args->buffer_count; |
27173f1f BW |
77 | size *= sizeof(struct i915_vma *); |
78 | size += sizeof(struct eb_vmas); | |
eef90ccb CW |
79 | eb = kmalloc(size, GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY); |
80 | } | |
81 | ||
82 | if (eb == NULL) { | |
b205ca57 DV |
83 | unsigned size = args->buffer_count; |
84 | unsigned count = PAGE_SIZE / sizeof(struct hlist_head) / 2; | |
27b7c63a | 85 | BUILD_BUG_ON_NOT_POWER_OF_2(PAGE_SIZE / sizeof(struct hlist_head)); |
eef90ccb CW |
86 | while (count > 2*size) |
87 | count >>= 1; | |
88 | eb = kzalloc(count*sizeof(struct hlist_head) + | |
27173f1f | 89 | sizeof(struct eb_vmas), |
eef90ccb CW |
90 | GFP_TEMPORARY); |
91 | if (eb == NULL) | |
92 | return eb; | |
93 | ||
94 | eb->and = count - 1; | |
95 | } else | |
96 | eb->and = -args->buffer_count; | |
97 | ||
27173f1f | 98 | INIT_LIST_HEAD(&eb->vmas); |
67731b87 CW |
99 | return eb; |
100 | } | |
101 | ||
102 | static void | |
27173f1f | 103 | eb_reset(struct eb_vmas *eb) |
67731b87 | 104 | { |
eef90ccb CW |
105 | if (eb->and >= 0) |
106 | memset(eb->buckets, 0, (eb->and+1)*sizeof(struct hlist_head)); | |
67731b87 CW |
107 | } |
108 | ||
59bfa124 CW |
109 | static struct i915_vma * |
110 | eb_get_batch(struct eb_vmas *eb) | |
111 | { | |
112 | struct i915_vma *vma = list_entry(eb->vmas.prev, typeof(*vma), exec_list); | |
113 | ||
114 | /* | |
115 | * SNA is doing fancy tricks with compressing batch buffers, which leads | |
116 | * to negative relocation deltas. Usually that works out ok since the | |
117 | * relocate address is still positive, except when the batch is placed | |
118 | * very low in the GTT. Ensure this doesn't happen. | |
119 | * | |
120 | * Note that actual hangs have only been observed on gen7, but for | |
121 | * paranoia do it everywhere. | |
122 | */ | |
123 | if ((vma->exec_entry->flags & EXEC_OBJECT_PINNED) == 0) | |
124 | vma->exec_entry->flags |= __EXEC_OBJECT_NEEDS_BIAS; | |
125 | ||
126 | return vma; | |
127 | } | |
128 | ||
3b96eff4 | 129 | static int |
27173f1f BW |
130 | eb_lookup_vmas(struct eb_vmas *eb, |
131 | struct drm_i915_gem_exec_object2 *exec, | |
132 | const struct drm_i915_gem_execbuffer2 *args, | |
133 | struct i915_address_space *vm, | |
134 | struct drm_file *file) | |
3b96eff4 | 135 | { |
27173f1f BW |
136 | struct drm_i915_gem_object *obj; |
137 | struct list_head objects; | |
9ae9ab52 | 138 | int i, ret; |
3b96eff4 | 139 | |
27173f1f | 140 | INIT_LIST_HEAD(&objects); |
3b96eff4 | 141 | spin_lock(&file->table_lock); |
27173f1f BW |
142 | /* Grab a reference to the object and release the lock so we can lookup |
143 | * or create the VMA without using GFP_ATOMIC */ | |
eef90ccb | 144 | for (i = 0; i < args->buffer_count; i++) { |
3b96eff4 CW |
145 | obj = to_intel_bo(idr_find(&file->object_idr, exec[i].handle)); |
146 | if (obj == NULL) { | |
147 | spin_unlock(&file->table_lock); | |
148 | DRM_DEBUG("Invalid object handle %d at index %d\n", | |
149 | exec[i].handle, i); | |
27173f1f | 150 | ret = -ENOENT; |
9ae9ab52 | 151 | goto err; |
3b96eff4 CW |
152 | } |
153 | ||
27173f1f | 154 | if (!list_empty(&obj->obj_exec_link)) { |
3b96eff4 CW |
155 | spin_unlock(&file->table_lock); |
156 | DRM_DEBUG("Object %p [handle %d, index %d] appears more than once in object list\n", | |
157 | obj, exec[i].handle, i); | |
27173f1f | 158 | ret = -EINVAL; |
9ae9ab52 | 159 | goto err; |
3b96eff4 CW |
160 | } |
161 | ||
25dc556a | 162 | i915_gem_object_get(obj); |
27173f1f BW |
163 | list_add_tail(&obj->obj_exec_link, &objects); |
164 | } | |
165 | spin_unlock(&file->table_lock); | |
3b96eff4 | 166 | |
27173f1f | 167 | i = 0; |
9ae9ab52 | 168 | while (!list_empty(&objects)) { |
27173f1f | 169 | struct i915_vma *vma; |
6f65e29a | 170 | |
9ae9ab52 CW |
171 | obj = list_first_entry(&objects, |
172 | struct drm_i915_gem_object, | |
173 | obj_exec_link); | |
174 | ||
e656a6cb DV |
175 | /* |
176 | * NOTE: We can leak any vmas created here when something fails | |
177 | * later on. But that's no issue since vma_unbind can deal with | |
178 | * vmas which are not actually bound. And since only | |
179 | * lookup_or_create exists as an interface to get at the vma | |
180 | * from the (obj, vm) we don't run the risk of creating | |
181 | * duplicated vmas for the same vm. | |
182 | */ | |
da51a1e7 | 183 | vma = i915_gem_obj_lookup_or_create_vma(obj, vm); |
27173f1f | 184 | if (IS_ERR(vma)) { |
27173f1f BW |
185 | DRM_DEBUG("Failed to lookup VMA\n"); |
186 | ret = PTR_ERR(vma); | |
9ae9ab52 | 187 | goto err; |
27173f1f BW |
188 | } |
189 | ||
9ae9ab52 | 190 | /* Transfer ownership from the objects list to the vmas list. */ |
27173f1f | 191 | list_add_tail(&vma->exec_list, &eb->vmas); |
9ae9ab52 | 192 | list_del_init(&obj->obj_exec_link); |
27173f1f BW |
193 | |
194 | vma->exec_entry = &exec[i]; | |
eef90ccb | 195 | if (eb->and < 0) { |
27173f1f | 196 | eb->lut[i] = vma; |
eef90ccb CW |
197 | } else { |
198 | uint32_t handle = args->flags & I915_EXEC_HANDLE_LUT ? i : exec[i].handle; | |
27173f1f BW |
199 | vma->exec_handle = handle; |
200 | hlist_add_head(&vma->exec_node, | |
eef90ccb CW |
201 | &eb->buckets[handle & eb->and]); |
202 | } | |
27173f1f | 203 | ++i; |
3b96eff4 | 204 | } |
3b96eff4 | 205 | |
9ae9ab52 | 206 | return 0; |
27173f1f | 207 | |
27173f1f | 208 | |
9ae9ab52 | 209 | err: |
27173f1f BW |
210 | while (!list_empty(&objects)) { |
211 | obj = list_first_entry(&objects, | |
212 | struct drm_i915_gem_object, | |
213 | obj_exec_link); | |
214 | list_del_init(&obj->obj_exec_link); | |
f8c417cd | 215 | i915_gem_object_put(obj); |
27173f1f | 216 | } |
9ae9ab52 CW |
217 | /* |
218 | * Objects already transfered to the vmas list will be unreferenced by | |
219 | * eb_destroy. | |
220 | */ | |
221 | ||
27173f1f | 222 | return ret; |
3b96eff4 CW |
223 | } |
224 | ||
27173f1f | 225 | static struct i915_vma *eb_get_vma(struct eb_vmas *eb, unsigned long handle) |
67731b87 | 226 | { |
eef90ccb CW |
227 | if (eb->and < 0) { |
228 | if (handle >= -eb->and) | |
229 | return NULL; | |
230 | return eb->lut[handle]; | |
231 | } else { | |
232 | struct hlist_head *head; | |
aa45950b | 233 | struct i915_vma *vma; |
67731b87 | 234 | |
eef90ccb | 235 | head = &eb->buckets[handle & eb->and]; |
aa45950b | 236 | hlist_for_each_entry(vma, head, exec_node) { |
27173f1f BW |
237 | if (vma->exec_handle == handle) |
238 | return vma; | |
eef90ccb CW |
239 | } |
240 | return NULL; | |
241 | } | |
67731b87 CW |
242 | } |
243 | ||
a415d355 CW |
244 | static void |
245 | i915_gem_execbuffer_unreserve_vma(struct i915_vma *vma) | |
246 | { | |
247 | struct drm_i915_gem_exec_object2 *entry; | |
248 | struct drm_i915_gem_object *obj = vma->obj; | |
249 | ||
250 | if (!drm_mm_node_allocated(&vma->node)) | |
251 | return; | |
252 | ||
253 | entry = vma->exec_entry; | |
254 | ||
255 | if (entry->flags & __EXEC_OBJECT_HAS_FENCE) | |
256 | i915_gem_object_unpin_fence(obj); | |
257 | ||
258 | if (entry->flags & __EXEC_OBJECT_HAS_PIN) | |
20dfbde4 | 259 | __i915_vma_unpin(vma); |
a415d355 | 260 | |
de4e783a | 261 | entry->flags &= ~(__EXEC_OBJECT_HAS_FENCE | __EXEC_OBJECT_HAS_PIN); |
a415d355 CW |
262 | } |
263 | ||
264 | static void eb_destroy(struct eb_vmas *eb) | |
265 | { | |
27173f1f BW |
266 | while (!list_empty(&eb->vmas)) { |
267 | struct i915_vma *vma; | |
bcffc3fa | 268 | |
27173f1f BW |
269 | vma = list_first_entry(&eb->vmas, |
270 | struct i915_vma, | |
bcffc3fa | 271 | exec_list); |
27173f1f | 272 | list_del_init(&vma->exec_list); |
a415d355 | 273 | i915_gem_execbuffer_unreserve_vma(vma); |
624192cf | 274 | i915_vma_put(vma); |
bcffc3fa | 275 | } |
67731b87 CW |
276 | kfree(eb); |
277 | } | |
278 | ||
dabdfe02 CW |
279 | static inline int use_cpu_reloc(struct drm_i915_gem_object *obj) |
280 | { | |
2cc86b82 CW |
281 | return (HAS_LLC(obj->base.dev) || |
282 | obj->base.write_domain == I915_GEM_DOMAIN_CPU || | |
dabdfe02 CW |
283 | obj->cache_level != I915_CACHE_NONE); |
284 | } | |
285 | ||
934acce3 MW |
286 | /* Used to convert any address to canonical form. |
287 | * Starting from gen8, some commands (e.g. STATE_BASE_ADDRESS, | |
288 | * MI_LOAD_REGISTER_MEM and others, see Broadwell PRM Vol2a) require the | |
289 | * addresses to be in a canonical form: | |
290 | * "GraphicsAddress[63:48] are ignored by the HW and assumed to be in correct | |
291 | * canonical form [63:48] == [47]." | |
292 | */ | |
293 | #define GEN8_HIGH_ADDRESS_BIT 47 | |
294 | static inline uint64_t gen8_canonical_addr(uint64_t address) | |
295 | { | |
296 | return sign_extend64(address, GEN8_HIGH_ADDRESS_BIT); | |
297 | } | |
298 | ||
299 | static inline uint64_t gen8_noncanonical_addr(uint64_t address) | |
300 | { | |
301 | return address & ((1ULL << (GEN8_HIGH_ADDRESS_BIT + 1)) - 1); | |
302 | } | |
303 | ||
304 | static inline uint64_t | |
305 | relocation_target(struct drm_i915_gem_relocation_entry *reloc, | |
306 | uint64_t target_offset) | |
307 | { | |
308 | return gen8_canonical_addr((int)reloc->delta + target_offset); | |
309 | } | |
310 | ||
5032d871 RB |
311 | static int |
312 | relocate_entry_cpu(struct drm_i915_gem_object *obj, | |
d9ceb957 BW |
313 | struct drm_i915_gem_relocation_entry *reloc, |
314 | uint64_t target_offset) | |
5032d871 | 315 | { |
3c94ceee | 316 | struct drm_device *dev = obj->base.dev; |
5032d871 | 317 | uint32_t page_offset = offset_in_page(reloc->offset); |
934acce3 | 318 | uint64_t delta = relocation_target(reloc, target_offset); |
5032d871 | 319 | char *vaddr; |
8b78f0e5 | 320 | int ret; |
5032d871 | 321 | |
2cc86b82 | 322 | ret = i915_gem_object_set_to_cpu_domain(obj, true); |
5032d871 RB |
323 | if (ret) |
324 | return ret; | |
325 | ||
033908ae | 326 | vaddr = kmap_atomic(i915_gem_object_get_dirty_page(obj, |
5032d871 | 327 | reloc->offset >> PAGE_SHIFT)); |
d9ceb957 | 328 | *(uint32_t *)(vaddr + page_offset) = lower_32_bits(delta); |
3c94ceee BW |
329 | |
330 | if (INTEL_INFO(dev)->gen >= 8) { | |
331 | page_offset = offset_in_page(page_offset + sizeof(uint32_t)); | |
332 | ||
333 | if (page_offset == 0) { | |
334 | kunmap_atomic(vaddr); | |
033908ae | 335 | vaddr = kmap_atomic(i915_gem_object_get_dirty_page(obj, |
3c94ceee BW |
336 | (reloc->offset + sizeof(uint32_t)) >> PAGE_SHIFT)); |
337 | } | |
338 | ||
d9ceb957 | 339 | *(uint32_t *)(vaddr + page_offset) = upper_32_bits(delta); |
3c94ceee BW |
340 | } |
341 | ||
5032d871 RB |
342 | kunmap_atomic(vaddr); |
343 | ||
344 | return 0; | |
345 | } | |
346 | ||
347 | static int | |
348 | relocate_entry_gtt(struct drm_i915_gem_object *obj, | |
d9ceb957 BW |
349 | struct drm_i915_gem_relocation_entry *reloc, |
350 | uint64_t target_offset) | |
5032d871 RB |
351 | { |
352 | struct drm_device *dev = obj->base.dev; | |
72e96d64 JL |
353 | struct drm_i915_private *dev_priv = to_i915(dev); |
354 | struct i915_ggtt *ggtt = &dev_priv->ggtt; | |
934acce3 | 355 | uint64_t delta = relocation_target(reloc, target_offset); |
906843c3 | 356 | uint64_t offset; |
5032d871 | 357 | void __iomem *reloc_page; |
8b78f0e5 | 358 | int ret; |
5032d871 RB |
359 | |
360 | ret = i915_gem_object_set_to_gtt_domain(obj, true); | |
361 | if (ret) | |
362 | return ret; | |
363 | ||
364 | ret = i915_gem_object_put_fence(obj); | |
365 | if (ret) | |
366 | return ret; | |
367 | ||
368 | /* Map the page containing the relocation we're going to perform. */ | |
906843c3 CW |
369 | offset = i915_gem_obj_ggtt_offset(obj); |
370 | offset += reloc->offset; | |
72e96d64 | 371 | reloc_page = io_mapping_map_atomic_wc(ggtt->mappable, |
906843c3 CW |
372 | offset & PAGE_MASK); |
373 | iowrite32(lower_32_bits(delta), reloc_page + offset_in_page(offset)); | |
3c94ceee BW |
374 | |
375 | if (INTEL_INFO(dev)->gen >= 8) { | |
906843c3 | 376 | offset += sizeof(uint32_t); |
3c94ceee | 377 | |
906843c3 | 378 | if (offset_in_page(offset) == 0) { |
3c94ceee | 379 | io_mapping_unmap_atomic(reloc_page); |
906843c3 | 380 | reloc_page = |
72e96d64 | 381 | io_mapping_map_atomic_wc(ggtt->mappable, |
906843c3 | 382 | offset); |
3c94ceee BW |
383 | } |
384 | ||
906843c3 CW |
385 | iowrite32(upper_32_bits(delta), |
386 | reloc_page + offset_in_page(offset)); | |
3c94ceee BW |
387 | } |
388 | ||
5032d871 RB |
389 | io_mapping_unmap_atomic(reloc_page); |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
edf4427b CW |
394 | static void |
395 | clflush_write32(void *addr, uint32_t value) | |
396 | { | |
397 | /* This is not a fast path, so KISS. */ | |
398 | drm_clflush_virt_range(addr, sizeof(uint32_t)); | |
399 | *(uint32_t *)addr = value; | |
400 | drm_clflush_virt_range(addr, sizeof(uint32_t)); | |
401 | } | |
402 | ||
403 | static int | |
404 | relocate_entry_clflush(struct drm_i915_gem_object *obj, | |
405 | struct drm_i915_gem_relocation_entry *reloc, | |
406 | uint64_t target_offset) | |
407 | { | |
408 | struct drm_device *dev = obj->base.dev; | |
409 | uint32_t page_offset = offset_in_page(reloc->offset); | |
934acce3 | 410 | uint64_t delta = relocation_target(reloc, target_offset); |
edf4427b CW |
411 | char *vaddr; |
412 | int ret; | |
413 | ||
414 | ret = i915_gem_object_set_to_gtt_domain(obj, true); | |
415 | if (ret) | |
416 | return ret; | |
417 | ||
033908ae | 418 | vaddr = kmap_atomic(i915_gem_object_get_dirty_page(obj, |
edf4427b CW |
419 | reloc->offset >> PAGE_SHIFT)); |
420 | clflush_write32(vaddr + page_offset, lower_32_bits(delta)); | |
421 | ||
422 | if (INTEL_INFO(dev)->gen >= 8) { | |
423 | page_offset = offset_in_page(page_offset + sizeof(uint32_t)); | |
424 | ||
425 | if (page_offset == 0) { | |
426 | kunmap_atomic(vaddr); | |
033908ae | 427 | vaddr = kmap_atomic(i915_gem_object_get_dirty_page(obj, |
edf4427b CW |
428 | (reloc->offset + sizeof(uint32_t)) >> PAGE_SHIFT)); |
429 | } | |
430 | ||
431 | clflush_write32(vaddr + page_offset, upper_32_bits(delta)); | |
432 | } | |
433 | ||
434 | kunmap_atomic(vaddr); | |
435 | ||
436 | return 0; | |
437 | } | |
438 | ||
909d074c CW |
439 | static bool object_is_idle(struct drm_i915_gem_object *obj) |
440 | { | |
573adb39 | 441 | unsigned long active = i915_gem_object_get_active(obj); |
909d074c CW |
442 | int idx; |
443 | ||
444 | for_each_active(active, idx) { | |
445 | if (!i915_gem_active_is_idle(&obj->last_read[idx], | |
446 | &obj->base.dev->struct_mutex)) | |
447 | return false; | |
448 | } | |
449 | ||
450 | return true; | |
451 | } | |
452 | ||
54cf91dc CW |
453 | static int |
454 | i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj, | |
27173f1f | 455 | struct eb_vmas *eb, |
3e7a0322 | 456 | struct drm_i915_gem_relocation_entry *reloc) |
54cf91dc CW |
457 | { |
458 | struct drm_device *dev = obj->base.dev; | |
459 | struct drm_gem_object *target_obj; | |
149c8407 | 460 | struct drm_i915_gem_object *target_i915_obj; |
27173f1f | 461 | struct i915_vma *target_vma; |
d9ceb957 | 462 | uint64_t target_offset; |
8b78f0e5 | 463 | int ret; |
54cf91dc | 464 | |
67731b87 | 465 | /* we've already hold a reference to all valid objects */ |
27173f1f BW |
466 | target_vma = eb_get_vma(eb, reloc->target_handle); |
467 | if (unlikely(target_vma == NULL)) | |
54cf91dc | 468 | return -ENOENT; |
27173f1f BW |
469 | target_i915_obj = target_vma->obj; |
470 | target_obj = &target_vma->obj->base; | |
54cf91dc | 471 | |
934acce3 | 472 | target_offset = gen8_canonical_addr(target_vma->node.start); |
54cf91dc | 473 | |
e844b990 EA |
474 | /* Sandybridge PPGTT errata: We need a global gtt mapping for MI and |
475 | * pipe_control writes because the gpu doesn't properly redirect them | |
476 | * through the ppgtt for non_secure batchbuffers. */ | |
477 | if (unlikely(IS_GEN6(dev) && | |
0875546c | 478 | reloc->write_domain == I915_GEM_DOMAIN_INSTRUCTION)) { |
fe14d5f4 | 479 | ret = i915_vma_bind(target_vma, target_i915_obj->cache_level, |
0875546c | 480 | PIN_GLOBAL); |
fe14d5f4 TU |
481 | if (WARN_ONCE(ret, "Unexpected failure to bind target VMA!")) |
482 | return ret; | |
483 | } | |
e844b990 | 484 | |
54cf91dc | 485 | /* Validate that the target is in a valid r/w GPU domain */ |
b8f7ab17 | 486 | if (unlikely(reloc->write_domain & (reloc->write_domain - 1))) { |
ff240199 | 487 | DRM_DEBUG("reloc with multiple write domains: " |
54cf91dc CW |
488 | "obj %p target %d offset %d " |
489 | "read %08x write %08x", | |
490 | obj, reloc->target_handle, | |
491 | (int) reloc->offset, | |
492 | reloc->read_domains, | |
493 | reloc->write_domain); | |
8b78f0e5 | 494 | return -EINVAL; |
54cf91dc | 495 | } |
4ca4a250 DV |
496 | if (unlikely((reloc->write_domain | reloc->read_domains) |
497 | & ~I915_GEM_GPU_DOMAINS)) { | |
ff240199 | 498 | DRM_DEBUG("reloc with read/write non-GPU domains: " |
54cf91dc CW |
499 | "obj %p target %d offset %d " |
500 | "read %08x write %08x", | |
501 | obj, reloc->target_handle, | |
502 | (int) reloc->offset, | |
503 | reloc->read_domains, | |
504 | reloc->write_domain); | |
8b78f0e5 | 505 | return -EINVAL; |
54cf91dc | 506 | } |
54cf91dc CW |
507 | |
508 | target_obj->pending_read_domains |= reloc->read_domains; | |
509 | target_obj->pending_write_domain |= reloc->write_domain; | |
510 | ||
511 | /* If the relocation already has the right value in it, no | |
512 | * more work needs to be done. | |
513 | */ | |
514 | if (target_offset == reloc->presumed_offset) | |
67731b87 | 515 | return 0; |
54cf91dc CW |
516 | |
517 | /* Check that the relocation address is valid... */ | |
3c94ceee BW |
518 | if (unlikely(reloc->offset > |
519 | obj->base.size - (INTEL_INFO(dev)->gen >= 8 ? 8 : 4))) { | |
ff240199 | 520 | DRM_DEBUG("Relocation beyond object bounds: " |
54cf91dc CW |
521 | "obj %p target %d offset %d size %d.\n", |
522 | obj, reloc->target_handle, | |
523 | (int) reloc->offset, | |
524 | (int) obj->base.size); | |
8b78f0e5 | 525 | return -EINVAL; |
54cf91dc | 526 | } |
b8f7ab17 | 527 | if (unlikely(reloc->offset & 3)) { |
ff240199 | 528 | DRM_DEBUG("Relocation not 4-byte aligned: " |
54cf91dc CW |
529 | "obj %p target %d offset %d.\n", |
530 | obj, reloc->target_handle, | |
531 | (int) reloc->offset); | |
8b78f0e5 | 532 | return -EINVAL; |
54cf91dc CW |
533 | } |
534 | ||
dabdfe02 | 535 | /* We can't wait for rendering with pagefaults disabled */ |
909d074c | 536 | if (pagefault_disabled() && !object_is_idle(obj)) |
dabdfe02 CW |
537 | return -EFAULT; |
538 | ||
5032d871 | 539 | if (use_cpu_reloc(obj)) |
d9ceb957 | 540 | ret = relocate_entry_cpu(obj, reloc, target_offset); |
edf4427b | 541 | else if (obj->map_and_fenceable) |
d9ceb957 | 542 | ret = relocate_entry_gtt(obj, reloc, target_offset); |
906bf7fd | 543 | else if (static_cpu_has(X86_FEATURE_CLFLUSH)) |
edf4427b CW |
544 | ret = relocate_entry_clflush(obj, reloc, target_offset); |
545 | else { | |
546 | WARN_ONCE(1, "Impossible case in relocation handling\n"); | |
547 | ret = -ENODEV; | |
548 | } | |
54cf91dc | 549 | |
d4d36014 DV |
550 | if (ret) |
551 | return ret; | |
552 | ||
54cf91dc CW |
553 | /* and update the user's relocation entry */ |
554 | reloc->presumed_offset = target_offset; | |
555 | ||
67731b87 | 556 | return 0; |
54cf91dc CW |
557 | } |
558 | ||
559 | static int | |
27173f1f BW |
560 | i915_gem_execbuffer_relocate_vma(struct i915_vma *vma, |
561 | struct eb_vmas *eb) | |
54cf91dc | 562 | { |
1d83f442 CW |
563 | #define N_RELOC(x) ((x) / sizeof(struct drm_i915_gem_relocation_entry)) |
564 | struct drm_i915_gem_relocation_entry stack_reloc[N_RELOC(512)]; | |
54cf91dc | 565 | struct drm_i915_gem_relocation_entry __user *user_relocs; |
27173f1f | 566 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
1d83f442 | 567 | int remain, ret; |
54cf91dc | 568 | |
3ed605bc | 569 | user_relocs = u64_to_user_ptr(entry->relocs_ptr); |
54cf91dc | 570 | |
1d83f442 CW |
571 | remain = entry->relocation_count; |
572 | while (remain) { | |
573 | struct drm_i915_gem_relocation_entry *r = stack_reloc; | |
574 | int count = remain; | |
575 | if (count > ARRAY_SIZE(stack_reloc)) | |
576 | count = ARRAY_SIZE(stack_reloc); | |
577 | remain -= count; | |
578 | ||
579 | if (__copy_from_user_inatomic(r, user_relocs, count*sizeof(r[0]))) | |
54cf91dc CW |
580 | return -EFAULT; |
581 | ||
1d83f442 CW |
582 | do { |
583 | u64 offset = r->presumed_offset; | |
54cf91dc | 584 | |
3e7a0322 | 585 | ret = i915_gem_execbuffer_relocate_entry(vma->obj, eb, r); |
1d83f442 CW |
586 | if (ret) |
587 | return ret; | |
588 | ||
589 | if (r->presumed_offset != offset && | |
5b09c3ed | 590 | __put_user(r->presumed_offset, &user_relocs->presumed_offset)) { |
1d83f442 CW |
591 | return -EFAULT; |
592 | } | |
593 | ||
594 | user_relocs++; | |
595 | r++; | |
596 | } while (--count); | |
54cf91dc CW |
597 | } |
598 | ||
599 | return 0; | |
1d83f442 | 600 | #undef N_RELOC |
54cf91dc CW |
601 | } |
602 | ||
603 | static int | |
27173f1f BW |
604 | i915_gem_execbuffer_relocate_vma_slow(struct i915_vma *vma, |
605 | struct eb_vmas *eb, | |
606 | struct drm_i915_gem_relocation_entry *relocs) | |
54cf91dc | 607 | { |
27173f1f | 608 | const struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
54cf91dc CW |
609 | int i, ret; |
610 | ||
611 | for (i = 0; i < entry->relocation_count; i++) { | |
3e7a0322 | 612 | ret = i915_gem_execbuffer_relocate_entry(vma->obj, eb, &relocs[i]); |
54cf91dc CW |
613 | if (ret) |
614 | return ret; | |
615 | } | |
616 | ||
617 | return 0; | |
618 | } | |
619 | ||
620 | static int | |
17601cbc | 621 | i915_gem_execbuffer_relocate(struct eb_vmas *eb) |
54cf91dc | 622 | { |
27173f1f | 623 | struct i915_vma *vma; |
d4aeee77 CW |
624 | int ret = 0; |
625 | ||
626 | /* This is the fast path and we cannot handle a pagefault whilst | |
627 | * holding the struct mutex lest the user pass in the relocations | |
628 | * contained within a mmaped bo. For in such a case we, the page | |
629 | * fault handler would call i915_gem_fault() and we would try to | |
630 | * acquire the struct mutex again. Obviously this is bad and so | |
631 | * lockdep complains vehemently. | |
632 | */ | |
633 | pagefault_disable(); | |
27173f1f BW |
634 | list_for_each_entry(vma, &eb->vmas, exec_list) { |
635 | ret = i915_gem_execbuffer_relocate_vma(vma, eb); | |
54cf91dc | 636 | if (ret) |
d4aeee77 | 637 | break; |
54cf91dc | 638 | } |
d4aeee77 | 639 | pagefault_enable(); |
54cf91dc | 640 | |
d4aeee77 | 641 | return ret; |
54cf91dc CW |
642 | } |
643 | ||
edf4427b CW |
644 | static bool only_mappable_for_reloc(unsigned int flags) |
645 | { | |
646 | return (flags & (EXEC_OBJECT_NEEDS_FENCE | __EXEC_OBJECT_NEEDS_MAP)) == | |
647 | __EXEC_OBJECT_NEEDS_MAP; | |
648 | } | |
649 | ||
1690e1eb | 650 | static int |
27173f1f | 651 | i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, |
0bc40be8 | 652 | struct intel_engine_cs *engine, |
27173f1f | 653 | bool *need_reloc) |
1690e1eb | 654 | { |
6f65e29a | 655 | struct drm_i915_gem_object *obj = vma->obj; |
27173f1f | 656 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
d23db88c | 657 | uint64_t flags; |
1690e1eb CW |
658 | int ret; |
659 | ||
0875546c | 660 | flags = PIN_USER; |
0229da32 DV |
661 | if (entry->flags & EXEC_OBJECT_NEEDS_GTT) |
662 | flags |= PIN_GLOBAL; | |
663 | ||
edf4427b | 664 | if (!drm_mm_node_allocated(&vma->node)) { |
101b506a MT |
665 | /* Wa32bitGeneralStateOffset & Wa32bitInstructionBaseOffset, |
666 | * limit address to the first 4GBs for unflagged objects. | |
667 | */ | |
668 | if ((entry->flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS) == 0) | |
669 | flags |= PIN_ZONE_4G; | |
edf4427b CW |
670 | if (entry->flags & __EXEC_OBJECT_NEEDS_MAP) |
671 | flags |= PIN_GLOBAL | PIN_MAPPABLE; | |
edf4427b CW |
672 | if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS) |
673 | flags |= BATCH_OFFSET_BIAS | PIN_OFFSET_BIAS; | |
506a8e87 CW |
674 | if (entry->flags & EXEC_OBJECT_PINNED) |
675 | flags |= entry->offset | PIN_OFFSET_FIXED; | |
101b506a MT |
676 | if ((flags & PIN_MAPPABLE) == 0) |
677 | flags |= PIN_HIGH; | |
edf4427b | 678 | } |
1ec9e26d | 679 | |
59bfa124 CW |
680 | ret = i915_vma_pin(vma, |
681 | entry->pad_to_size, | |
682 | entry->alignment, | |
683 | flags); | |
684 | if ((ret == -ENOSPC || ret == -E2BIG) && | |
edf4427b | 685 | only_mappable_for_reloc(entry->flags)) |
59bfa124 CW |
686 | ret = i915_vma_pin(vma, |
687 | entry->pad_to_size, | |
688 | entry->alignment, | |
689 | flags & ~PIN_MAPPABLE); | |
1690e1eb CW |
690 | if (ret) |
691 | return ret; | |
692 | ||
7788a765 CW |
693 | entry->flags |= __EXEC_OBJECT_HAS_PIN; |
694 | ||
82b6b6d7 CW |
695 | if (entry->flags & EXEC_OBJECT_NEEDS_FENCE) { |
696 | ret = i915_gem_object_get_fence(obj); | |
697 | if (ret) | |
698 | return ret; | |
9a5a53b3 | 699 | |
82b6b6d7 CW |
700 | if (i915_gem_object_pin_fence(obj)) |
701 | entry->flags |= __EXEC_OBJECT_HAS_FENCE; | |
1690e1eb CW |
702 | } |
703 | ||
27173f1f BW |
704 | if (entry->offset != vma->node.start) { |
705 | entry->offset = vma->node.start; | |
ed5982e6 DV |
706 | *need_reloc = true; |
707 | } | |
708 | ||
709 | if (entry->flags & EXEC_OBJECT_WRITE) { | |
710 | obj->base.pending_read_domains = I915_GEM_DOMAIN_RENDER; | |
711 | obj->base.pending_write_domain = I915_GEM_DOMAIN_RENDER; | |
712 | } | |
713 | ||
1690e1eb | 714 | return 0; |
7788a765 | 715 | } |
1690e1eb | 716 | |
d23db88c | 717 | static bool |
e6a84468 | 718 | need_reloc_mappable(struct i915_vma *vma) |
d23db88c CW |
719 | { |
720 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; | |
d23db88c | 721 | |
e6a84468 CW |
722 | if (entry->relocation_count == 0) |
723 | return false; | |
724 | ||
3272db53 | 725 | if (!i915_vma_is_ggtt(vma)) |
e6a84468 CW |
726 | return false; |
727 | ||
728 | /* See also use_cpu_reloc() */ | |
729 | if (HAS_LLC(vma->obj->base.dev)) | |
730 | return false; | |
731 | ||
732 | if (vma->obj->base.write_domain == I915_GEM_DOMAIN_CPU) | |
733 | return false; | |
734 | ||
735 | return true; | |
736 | } | |
737 | ||
738 | static bool | |
739 | eb_vma_misplaced(struct i915_vma *vma) | |
740 | { | |
741 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; | |
742 | struct drm_i915_gem_object *obj = vma->obj; | |
d23db88c | 743 | |
3272db53 CW |
744 | WARN_ON(entry->flags & __EXEC_OBJECT_NEEDS_MAP && |
745 | !i915_vma_is_ggtt(vma)); | |
d23db88c CW |
746 | |
747 | if (entry->alignment && | |
748 | vma->node.start & (entry->alignment - 1)) | |
749 | return true; | |
750 | ||
91b2db6f CW |
751 | if (vma->node.size < entry->pad_to_size) |
752 | return true; | |
753 | ||
506a8e87 CW |
754 | if (entry->flags & EXEC_OBJECT_PINNED && |
755 | vma->node.start != entry->offset) | |
756 | return true; | |
757 | ||
d23db88c CW |
758 | if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS && |
759 | vma->node.start < BATCH_OFFSET_BIAS) | |
760 | return true; | |
761 | ||
edf4427b CW |
762 | /* avoid costly ping-pong once a batch bo ended up non-mappable */ |
763 | if (entry->flags & __EXEC_OBJECT_NEEDS_MAP && !obj->map_and_fenceable) | |
764 | return !only_mappable_for_reloc(entry->flags); | |
765 | ||
101b506a MT |
766 | if ((entry->flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS) == 0 && |
767 | (vma->node.start + vma->node.size - 1) >> 32) | |
768 | return true; | |
769 | ||
d23db88c CW |
770 | return false; |
771 | } | |
772 | ||
54cf91dc | 773 | static int |
0bc40be8 | 774 | i915_gem_execbuffer_reserve(struct intel_engine_cs *engine, |
27173f1f | 775 | struct list_head *vmas, |
e2efd130 | 776 | struct i915_gem_context *ctx, |
ed5982e6 | 777 | bool *need_relocs) |
54cf91dc | 778 | { |
432e58ed | 779 | struct drm_i915_gem_object *obj; |
27173f1f | 780 | struct i915_vma *vma; |
68c8c17f | 781 | struct i915_address_space *vm; |
27173f1f | 782 | struct list_head ordered_vmas; |
506a8e87 | 783 | struct list_head pinned_vmas; |
c033666a | 784 | bool has_fenced_gpu_access = INTEL_GEN(engine->i915) < 4; |
7788a765 | 785 | int retry; |
6fe4f140 | 786 | |
68c8c17f BW |
787 | vm = list_first_entry(vmas, struct i915_vma, exec_list)->vm; |
788 | ||
27173f1f | 789 | INIT_LIST_HEAD(&ordered_vmas); |
506a8e87 | 790 | INIT_LIST_HEAD(&pinned_vmas); |
27173f1f | 791 | while (!list_empty(vmas)) { |
6fe4f140 CW |
792 | struct drm_i915_gem_exec_object2 *entry; |
793 | bool need_fence, need_mappable; | |
794 | ||
27173f1f BW |
795 | vma = list_first_entry(vmas, struct i915_vma, exec_list); |
796 | obj = vma->obj; | |
797 | entry = vma->exec_entry; | |
6fe4f140 | 798 | |
b1b38278 DW |
799 | if (ctx->flags & CONTEXT_NO_ZEROMAP) |
800 | entry->flags |= __EXEC_OBJECT_NEEDS_BIAS; | |
801 | ||
82b6b6d7 CW |
802 | if (!has_fenced_gpu_access) |
803 | entry->flags &= ~EXEC_OBJECT_NEEDS_FENCE; | |
6fe4f140 | 804 | need_fence = |
6fe4f140 | 805 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
3e510a8e | 806 | i915_gem_object_is_tiled(obj); |
27173f1f | 807 | need_mappable = need_fence || need_reloc_mappable(vma); |
6fe4f140 | 808 | |
506a8e87 CW |
809 | if (entry->flags & EXEC_OBJECT_PINNED) |
810 | list_move_tail(&vma->exec_list, &pinned_vmas); | |
811 | else if (need_mappable) { | |
e6a84468 | 812 | entry->flags |= __EXEC_OBJECT_NEEDS_MAP; |
27173f1f | 813 | list_move(&vma->exec_list, &ordered_vmas); |
e6a84468 | 814 | } else |
27173f1f | 815 | list_move_tail(&vma->exec_list, &ordered_vmas); |
595dad76 | 816 | |
ed5982e6 | 817 | obj->base.pending_read_domains = I915_GEM_GPU_DOMAINS & ~I915_GEM_DOMAIN_COMMAND; |
595dad76 | 818 | obj->base.pending_write_domain = 0; |
6fe4f140 | 819 | } |
27173f1f | 820 | list_splice(&ordered_vmas, vmas); |
506a8e87 | 821 | list_splice(&pinned_vmas, vmas); |
54cf91dc CW |
822 | |
823 | /* Attempt to pin all of the buffers into the GTT. | |
824 | * This is done in 3 phases: | |
825 | * | |
826 | * 1a. Unbind all objects that do not match the GTT constraints for | |
827 | * the execbuffer (fenceable, mappable, alignment etc). | |
828 | * 1b. Increment pin count for already bound objects. | |
829 | * 2. Bind new objects. | |
830 | * 3. Decrement pin count. | |
831 | * | |
7788a765 | 832 | * This avoid unnecessary unbinding of later objects in order to make |
54cf91dc CW |
833 | * room for the earlier objects *unless* we need to defragment. |
834 | */ | |
835 | retry = 0; | |
836 | do { | |
7788a765 | 837 | int ret = 0; |
54cf91dc CW |
838 | |
839 | /* Unbind any ill-fitting objects or pin. */ | |
27173f1f | 840 | list_for_each_entry(vma, vmas, exec_list) { |
27173f1f | 841 | if (!drm_mm_node_allocated(&vma->node)) |
54cf91dc CW |
842 | continue; |
843 | ||
e6a84468 | 844 | if (eb_vma_misplaced(vma)) |
27173f1f | 845 | ret = i915_vma_unbind(vma); |
54cf91dc | 846 | else |
0bc40be8 TU |
847 | ret = i915_gem_execbuffer_reserve_vma(vma, |
848 | engine, | |
849 | need_relocs); | |
432e58ed | 850 | if (ret) |
54cf91dc | 851 | goto err; |
54cf91dc CW |
852 | } |
853 | ||
854 | /* Bind fresh objects */ | |
27173f1f BW |
855 | list_for_each_entry(vma, vmas, exec_list) { |
856 | if (drm_mm_node_allocated(&vma->node)) | |
1690e1eb | 857 | continue; |
54cf91dc | 858 | |
0bc40be8 TU |
859 | ret = i915_gem_execbuffer_reserve_vma(vma, engine, |
860 | need_relocs); | |
7788a765 CW |
861 | if (ret) |
862 | goto err; | |
54cf91dc CW |
863 | } |
864 | ||
a415d355 | 865 | err: |
6c085a72 | 866 | if (ret != -ENOSPC || retry++) |
54cf91dc CW |
867 | return ret; |
868 | ||
a415d355 CW |
869 | /* Decrement pin count for bound objects */ |
870 | list_for_each_entry(vma, vmas, exec_list) | |
871 | i915_gem_execbuffer_unreserve_vma(vma); | |
872 | ||
68c8c17f | 873 | ret = i915_gem_evict_vm(vm, true); |
54cf91dc CW |
874 | if (ret) |
875 | return ret; | |
54cf91dc CW |
876 | } while (1); |
877 | } | |
878 | ||
879 | static int | |
880 | i915_gem_execbuffer_relocate_slow(struct drm_device *dev, | |
ed5982e6 | 881 | struct drm_i915_gem_execbuffer2 *args, |
54cf91dc | 882 | struct drm_file *file, |
0bc40be8 | 883 | struct intel_engine_cs *engine, |
27173f1f | 884 | struct eb_vmas *eb, |
b1b38278 | 885 | struct drm_i915_gem_exec_object2 *exec, |
e2efd130 | 886 | struct i915_gem_context *ctx) |
54cf91dc CW |
887 | { |
888 | struct drm_i915_gem_relocation_entry *reloc; | |
27173f1f BW |
889 | struct i915_address_space *vm; |
890 | struct i915_vma *vma; | |
ed5982e6 | 891 | bool need_relocs; |
dd6864a4 | 892 | int *reloc_offset; |
54cf91dc | 893 | int i, total, ret; |
b205ca57 | 894 | unsigned count = args->buffer_count; |
54cf91dc | 895 | |
27173f1f BW |
896 | vm = list_first_entry(&eb->vmas, struct i915_vma, exec_list)->vm; |
897 | ||
67731b87 | 898 | /* We may process another execbuffer during the unlock... */ |
27173f1f BW |
899 | while (!list_empty(&eb->vmas)) { |
900 | vma = list_first_entry(&eb->vmas, struct i915_vma, exec_list); | |
901 | list_del_init(&vma->exec_list); | |
a415d355 | 902 | i915_gem_execbuffer_unreserve_vma(vma); |
624192cf | 903 | i915_vma_put(vma); |
67731b87 CW |
904 | } |
905 | ||
54cf91dc CW |
906 | mutex_unlock(&dev->struct_mutex); |
907 | ||
908 | total = 0; | |
909 | for (i = 0; i < count; i++) | |
432e58ed | 910 | total += exec[i].relocation_count; |
54cf91dc | 911 | |
dd6864a4 | 912 | reloc_offset = drm_malloc_ab(count, sizeof(*reloc_offset)); |
54cf91dc | 913 | reloc = drm_malloc_ab(total, sizeof(*reloc)); |
dd6864a4 CW |
914 | if (reloc == NULL || reloc_offset == NULL) { |
915 | drm_free_large(reloc); | |
916 | drm_free_large(reloc_offset); | |
54cf91dc CW |
917 | mutex_lock(&dev->struct_mutex); |
918 | return -ENOMEM; | |
919 | } | |
920 | ||
921 | total = 0; | |
922 | for (i = 0; i < count; i++) { | |
923 | struct drm_i915_gem_relocation_entry __user *user_relocs; | |
262b6d36 CW |
924 | u64 invalid_offset = (u64)-1; |
925 | int j; | |
54cf91dc | 926 | |
3ed605bc | 927 | user_relocs = u64_to_user_ptr(exec[i].relocs_ptr); |
54cf91dc CW |
928 | |
929 | if (copy_from_user(reloc+total, user_relocs, | |
432e58ed | 930 | exec[i].relocation_count * sizeof(*reloc))) { |
54cf91dc CW |
931 | ret = -EFAULT; |
932 | mutex_lock(&dev->struct_mutex); | |
933 | goto err; | |
934 | } | |
935 | ||
262b6d36 CW |
936 | /* As we do not update the known relocation offsets after |
937 | * relocating (due to the complexities in lock handling), | |
938 | * we need to mark them as invalid now so that we force the | |
939 | * relocation processing next time. Just in case the target | |
940 | * object is evicted and then rebound into its old | |
941 | * presumed_offset before the next execbuffer - if that | |
942 | * happened we would make the mistake of assuming that the | |
943 | * relocations were valid. | |
944 | */ | |
945 | for (j = 0; j < exec[i].relocation_count; j++) { | |
9aab8bff CW |
946 | if (__copy_to_user(&user_relocs[j].presumed_offset, |
947 | &invalid_offset, | |
948 | sizeof(invalid_offset))) { | |
262b6d36 CW |
949 | ret = -EFAULT; |
950 | mutex_lock(&dev->struct_mutex); | |
951 | goto err; | |
952 | } | |
953 | } | |
954 | ||
dd6864a4 | 955 | reloc_offset[i] = total; |
432e58ed | 956 | total += exec[i].relocation_count; |
54cf91dc CW |
957 | } |
958 | ||
959 | ret = i915_mutex_lock_interruptible(dev); | |
960 | if (ret) { | |
961 | mutex_lock(&dev->struct_mutex); | |
962 | goto err; | |
963 | } | |
964 | ||
67731b87 | 965 | /* reacquire the objects */ |
67731b87 | 966 | eb_reset(eb); |
27173f1f | 967 | ret = eb_lookup_vmas(eb, exec, args, vm, file); |
3b96eff4 CW |
968 | if (ret) |
969 | goto err; | |
67731b87 | 970 | |
ed5982e6 | 971 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
0bc40be8 TU |
972 | ret = i915_gem_execbuffer_reserve(engine, &eb->vmas, ctx, |
973 | &need_relocs); | |
54cf91dc CW |
974 | if (ret) |
975 | goto err; | |
976 | ||
27173f1f BW |
977 | list_for_each_entry(vma, &eb->vmas, exec_list) { |
978 | int offset = vma->exec_entry - exec; | |
979 | ret = i915_gem_execbuffer_relocate_vma_slow(vma, eb, | |
980 | reloc + reloc_offset[offset]); | |
54cf91dc CW |
981 | if (ret) |
982 | goto err; | |
54cf91dc CW |
983 | } |
984 | ||
985 | /* Leave the user relocations as are, this is the painfully slow path, | |
986 | * and we want to avoid the complication of dropping the lock whilst | |
987 | * having buffers reserved in the aperture and so causing spurious | |
988 | * ENOSPC for random operations. | |
989 | */ | |
990 | ||
991 | err: | |
992 | drm_free_large(reloc); | |
dd6864a4 | 993 | drm_free_large(reloc_offset); |
54cf91dc CW |
994 | return ret; |
995 | } | |
996 | ||
573adb39 CW |
997 | static unsigned int eb_other_engines(struct drm_i915_gem_request *req) |
998 | { | |
999 | unsigned int mask; | |
1000 | ||
1001 | mask = ~intel_engine_flag(req->engine) & I915_BO_ACTIVE_MASK; | |
1002 | mask <<= I915_BO_ACTIVE_SHIFT; | |
1003 | ||
1004 | return mask; | |
1005 | } | |
1006 | ||
54cf91dc | 1007 | static int |
535fbe82 | 1008 | i915_gem_execbuffer_move_to_gpu(struct drm_i915_gem_request *req, |
27173f1f | 1009 | struct list_head *vmas) |
54cf91dc | 1010 | { |
573adb39 | 1011 | const unsigned int other_rings = eb_other_engines(req); |
27173f1f | 1012 | struct i915_vma *vma; |
6ac42f41 | 1013 | uint32_t flush_domains = 0; |
000433b6 | 1014 | bool flush_chipset = false; |
432e58ed | 1015 | int ret; |
54cf91dc | 1016 | |
27173f1f BW |
1017 | list_for_each_entry(vma, vmas, exec_list) { |
1018 | struct drm_i915_gem_object *obj = vma->obj; | |
03ade511 | 1019 | |
573adb39 | 1020 | if (obj->flags & other_rings) { |
8e637178 | 1021 | ret = i915_gem_object_sync(obj, req); |
03ade511 CW |
1022 | if (ret) |
1023 | return ret; | |
1024 | } | |
6ac42f41 DV |
1025 | |
1026 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) | |
000433b6 | 1027 | flush_chipset |= i915_gem_clflush_object(obj, false); |
6ac42f41 | 1028 | |
6ac42f41 | 1029 | flush_domains |= obj->base.write_domain; |
c59a333f CW |
1030 | } |
1031 | ||
000433b6 | 1032 | if (flush_chipset) |
c033666a | 1033 | i915_gem_chipset_flush(req->engine->i915); |
6ac42f41 DV |
1034 | |
1035 | if (flush_domains & I915_GEM_DOMAIN_GTT) | |
1036 | wmb(); | |
1037 | ||
c7fe7d25 | 1038 | /* Unconditionally invalidate GPU caches and TLBs. */ |
7c9cf4e3 | 1039 | return req->engine->emit_flush(req, EMIT_INVALIDATE); |
54cf91dc CW |
1040 | } |
1041 | ||
432e58ed CW |
1042 | static bool |
1043 | i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec) | |
54cf91dc | 1044 | { |
ed5982e6 DV |
1045 | if (exec->flags & __I915_EXEC_UNKNOWN_FLAGS) |
1046 | return false; | |
1047 | ||
2f5945bc CW |
1048 | /* Kernel clipping was a DRI1 misfeature */ |
1049 | if (exec->num_cliprects || exec->cliprects_ptr) | |
1050 | return false; | |
1051 | ||
1052 | if (exec->DR4 == 0xffffffff) { | |
1053 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); | |
1054 | exec->DR4 = 0; | |
1055 | } | |
1056 | if (exec->DR1 || exec->DR4) | |
1057 | return false; | |
1058 | ||
1059 | if ((exec->batch_start_offset | exec->batch_len) & 0x7) | |
1060 | return false; | |
1061 | ||
1062 | return true; | |
54cf91dc CW |
1063 | } |
1064 | ||
1065 | static int | |
ad19f10b CW |
1066 | validate_exec_list(struct drm_device *dev, |
1067 | struct drm_i915_gem_exec_object2 *exec, | |
54cf91dc CW |
1068 | int count) |
1069 | { | |
b205ca57 DV |
1070 | unsigned relocs_total = 0; |
1071 | unsigned relocs_max = UINT_MAX / sizeof(struct drm_i915_gem_relocation_entry); | |
ad19f10b CW |
1072 | unsigned invalid_flags; |
1073 | int i; | |
1074 | ||
9e2793f6 DG |
1075 | /* INTERNAL flags must not overlap with external ones */ |
1076 | BUILD_BUG_ON(__EXEC_OBJECT_INTERNAL_FLAGS & ~__EXEC_OBJECT_UNKNOWN_FLAGS); | |
1077 | ||
ad19f10b CW |
1078 | invalid_flags = __EXEC_OBJECT_UNKNOWN_FLAGS; |
1079 | if (USES_FULL_PPGTT(dev)) | |
1080 | invalid_flags |= EXEC_OBJECT_NEEDS_GTT; | |
54cf91dc CW |
1081 | |
1082 | for (i = 0; i < count; i++) { | |
3ed605bc | 1083 | char __user *ptr = u64_to_user_ptr(exec[i].relocs_ptr); |
54cf91dc CW |
1084 | int length; /* limited by fault_in_pages_readable() */ |
1085 | ||
ad19f10b | 1086 | if (exec[i].flags & invalid_flags) |
ed5982e6 DV |
1087 | return -EINVAL; |
1088 | ||
934acce3 MW |
1089 | /* Offset can be used as input (EXEC_OBJECT_PINNED), reject |
1090 | * any non-page-aligned or non-canonical addresses. | |
1091 | */ | |
1092 | if (exec[i].flags & EXEC_OBJECT_PINNED) { | |
1093 | if (exec[i].offset != | |
1094 | gen8_canonical_addr(exec[i].offset & PAGE_MASK)) | |
1095 | return -EINVAL; | |
1096 | ||
1097 | /* From drm_mm perspective address space is continuous, | |
1098 | * so from this point we're always using non-canonical | |
1099 | * form internally. | |
1100 | */ | |
1101 | exec[i].offset = gen8_noncanonical_addr(exec[i].offset); | |
1102 | } | |
1103 | ||
55a9785d CW |
1104 | if (exec[i].alignment && !is_power_of_2(exec[i].alignment)) |
1105 | return -EINVAL; | |
1106 | ||
91b2db6f CW |
1107 | /* pad_to_size was once a reserved field, so sanitize it */ |
1108 | if (exec[i].flags & EXEC_OBJECT_PAD_TO_SIZE) { | |
1109 | if (offset_in_page(exec[i].pad_to_size)) | |
1110 | return -EINVAL; | |
1111 | } else { | |
1112 | exec[i].pad_to_size = 0; | |
1113 | } | |
1114 | ||
3118a4f6 KC |
1115 | /* First check for malicious input causing overflow in |
1116 | * the worst case where we need to allocate the entire | |
1117 | * relocation tree as a single array. | |
1118 | */ | |
1119 | if (exec[i].relocation_count > relocs_max - relocs_total) | |
54cf91dc | 1120 | return -EINVAL; |
3118a4f6 | 1121 | relocs_total += exec[i].relocation_count; |
54cf91dc CW |
1122 | |
1123 | length = exec[i].relocation_count * | |
1124 | sizeof(struct drm_i915_gem_relocation_entry); | |
30587535 KC |
1125 | /* |
1126 | * We must check that the entire relocation array is safe | |
1127 | * to read, but since we may need to update the presumed | |
1128 | * offsets during execution, check for full write access. | |
1129 | */ | |
54cf91dc CW |
1130 | if (!access_ok(VERIFY_WRITE, ptr, length)) |
1131 | return -EFAULT; | |
1132 | ||
d330a953 | 1133 | if (likely(!i915.prefault_disable)) { |
0b74b508 XZ |
1134 | if (fault_in_multipages_readable(ptr, length)) |
1135 | return -EFAULT; | |
1136 | } | |
54cf91dc CW |
1137 | } |
1138 | ||
1139 | return 0; | |
1140 | } | |
1141 | ||
e2efd130 | 1142 | static struct i915_gem_context * |
d299cce7 | 1143 | i915_gem_validate_context(struct drm_device *dev, struct drm_file *file, |
0bc40be8 | 1144 | struct intel_engine_cs *engine, const u32 ctx_id) |
d299cce7 | 1145 | { |
e2efd130 | 1146 | struct i915_gem_context *ctx = NULL; |
d299cce7 MK |
1147 | struct i915_ctx_hang_stats *hs; |
1148 | ||
0bc40be8 | 1149 | if (engine->id != RCS && ctx_id != DEFAULT_CONTEXT_HANDLE) |
7c9c4b8f DV |
1150 | return ERR_PTR(-EINVAL); |
1151 | ||
ca585b5d | 1152 | ctx = i915_gem_context_lookup(file->driver_priv, ctx_id); |
72ad5c45 | 1153 | if (IS_ERR(ctx)) |
41bde553 | 1154 | return ctx; |
d299cce7 | 1155 | |
41bde553 | 1156 | hs = &ctx->hang_stats; |
d299cce7 MK |
1157 | if (hs->banned) { |
1158 | DRM_DEBUG("Context %u tried to submit while banned\n", ctx_id); | |
41bde553 | 1159 | return ERR_PTR(-EIO); |
d299cce7 MK |
1160 | } |
1161 | ||
41bde553 | 1162 | return ctx; |
d299cce7 MK |
1163 | } |
1164 | ||
5cf3d280 CW |
1165 | void i915_vma_move_to_active(struct i915_vma *vma, |
1166 | struct drm_i915_gem_request *req, | |
1167 | unsigned int flags) | |
1168 | { | |
1169 | struct drm_i915_gem_object *obj = vma->obj; | |
1170 | const unsigned int idx = req->engine->id; | |
1171 | ||
1172 | GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); | |
1173 | ||
1174 | obj->dirty = 1; /* be paranoid */ | |
1175 | ||
b0decaf7 CW |
1176 | /* Add a reference if we're newly entering the active list. |
1177 | * The order in which we add operations to the retirement queue is | |
1178 | * vital here: mark_active adds to the start of the callback list, | |
1179 | * such that subsequent callbacks are called first. Therefore we | |
1180 | * add the active reference first and queue for it to be dropped | |
1181 | * *last*. | |
1182 | */ | |
573adb39 | 1183 | if (!i915_gem_object_is_active(obj)) |
5cf3d280 | 1184 | i915_gem_object_get(obj); |
573adb39 | 1185 | i915_gem_object_set_active(obj, idx); |
5cf3d280 CW |
1186 | i915_gem_active_set(&obj->last_read[idx], req); |
1187 | ||
1188 | if (flags & EXEC_OBJECT_WRITE) { | |
1189 | i915_gem_active_set(&obj->last_write, req); | |
1190 | ||
1191 | intel_fb_obj_invalidate(obj, ORIGIN_CS); | |
1192 | ||
1193 | /* update for the implicit flush after a batch */ | |
1194 | obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS; | |
1195 | } | |
1196 | ||
1197 | if (flags & EXEC_OBJECT_NEEDS_FENCE) { | |
1198 | i915_gem_active_set(&obj->last_fence, req); | |
1199 | if (flags & __EXEC_OBJECT_HAS_FENCE) { | |
1200 | struct drm_i915_private *dev_priv = req->i915; | |
1201 | ||
1202 | list_move_tail(&dev_priv->fence_regs[obj->fence_reg].lru_list, | |
1203 | &dev_priv->mm.fence_list); | |
1204 | } | |
1205 | } | |
1206 | ||
b0decaf7 CW |
1207 | i915_vma_set_active(vma, idx); |
1208 | i915_gem_active_set(&vma->last_read[idx], req); | |
5cf3d280 CW |
1209 | list_move_tail(&vma->vm_link, &vma->vm->active_list); |
1210 | } | |
1211 | ||
ad778f89 CW |
1212 | static void eb_export_fence(struct drm_i915_gem_object *obj, |
1213 | struct drm_i915_gem_request *req, | |
1214 | unsigned int flags) | |
1215 | { | |
1216 | struct reservation_object *resv; | |
1217 | ||
1218 | resv = i915_gem_object_get_dmabuf_resv(obj); | |
1219 | if (!resv) | |
1220 | return; | |
1221 | ||
1222 | /* Ignore errors from failing to allocate the new fence, we can't | |
1223 | * handle an error right now. Worst case should be missed | |
1224 | * synchronisation leading to rendering corruption. | |
1225 | */ | |
1226 | ww_mutex_lock(&resv->lock, NULL); | |
1227 | if (flags & EXEC_OBJECT_WRITE) | |
1228 | reservation_object_add_excl_fence(resv, &req->fence); | |
1229 | else if (reservation_object_reserve_shared(resv) == 0) | |
1230 | reservation_object_add_shared_fence(resv, &req->fence); | |
1231 | ww_mutex_unlock(&resv->lock); | |
1232 | } | |
1233 | ||
5b043f4e | 1234 | static void |
27173f1f | 1235 | i915_gem_execbuffer_move_to_active(struct list_head *vmas, |
8a8edb59 | 1236 | struct drm_i915_gem_request *req) |
432e58ed | 1237 | { |
27173f1f | 1238 | struct i915_vma *vma; |
432e58ed | 1239 | |
27173f1f BW |
1240 | list_for_each_entry(vma, vmas, exec_list) { |
1241 | struct drm_i915_gem_object *obj = vma->obj; | |
69c2fc89 CW |
1242 | u32 old_read = obj->base.read_domains; |
1243 | u32 old_write = obj->base.write_domain; | |
db53a302 | 1244 | |
432e58ed | 1245 | obj->base.write_domain = obj->base.pending_write_domain; |
5cf3d280 CW |
1246 | if (obj->base.write_domain) |
1247 | vma->exec_entry->flags |= EXEC_OBJECT_WRITE; | |
1248 | else | |
ed5982e6 DV |
1249 | obj->base.pending_read_domains |= obj->base.read_domains; |
1250 | obj->base.read_domains = obj->base.pending_read_domains; | |
432e58ed | 1251 | |
5cf3d280 | 1252 | i915_vma_move_to_active(vma, req, vma->exec_entry->flags); |
ad778f89 | 1253 | eb_export_fence(obj, req, vma->exec_entry->flags); |
db53a302 | 1254 | trace_i915_gem_object_change_domain(obj, old_read, old_write); |
432e58ed CW |
1255 | } |
1256 | } | |
1257 | ||
ae662d31 | 1258 | static int |
b5321f30 | 1259 | i915_reset_gen7_sol_offsets(struct drm_i915_gem_request *req) |
ae662d31 | 1260 | { |
7e37f889 | 1261 | struct intel_ring *ring = req->ring; |
ae662d31 EA |
1262 | int ret, i; |
1263 | ||
b5321f30 | 1264 | if (!IS_GEN7(req->i915) || req->engine->id != RCS) { |
9d662da8 DV |
1265 | DRM_DEBUG("sol reset is gen7/rcs only\n"); |
1266 | return -EINVAL; | |
1267 | } | |
ae662d31 | 1268 | |
5fb9de1a | 1269 | ret = intel_ring_begin(req, 4 * 3); |
ae662d31 EA |
1270 | if (ret) |
1271 | return ret; | |
1272 | ||
1273 | for (i = 0; i < 4; i++) { | |
b5321f30 CW |
1274 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); |
1275 | intel_ring_emit_reg(ring, GEN7_SO_WRITE_OFFSET(i)); | |
1276 | intel_ring_emit(ring, 0); | |
ae662d31 EA |
1277 | } |
1278 | ||
b5321f30 | 1279 | intel_ring_advance(ring); |
ae662d31 EA |
1280 | |
1281 | return 0; | |
1282 | } | |
1283 | ||
59bfa124 | 1284 | static struct i915_vma* |
0bc40be8 | 1285 | i915_gem_execbuffer_parse(struct intel_engine_cs *engine, |
71745376 | 1286 | struct drm_i915_gem_exec_object2 *shadow_exec_entry, |
71745376 | 1287 | struct drm_i915_gem_object *batch_obj, |
59bfa124 | 1288 | struct eb_vmas *eb, |
71745376 BV |
1289 | u32 batch_start_offset, |
1290 | u32 batch_len, | |
17cabf57 | 1291 | bool is_master) |
71745376 | 1292 | { |
71745376 | 1293 | struct drm_i915_gem_object *shadow_batch_obj; |
17cabf57 | 1294 | struct i915_vma *vma; |
71745376 BV |
1295 | int ret; |
1296 | ||
0bc40be8 | 1297 | shadow_batch_obj = i915_gem_batch_pool_get(&engine->batch_pool, |
17cabf57 | 1298 | PAGE_ALIGN(batch_len)); |
71745376 | 1299 | if (IS_ERR(shadow_batch_obj)) |
59bfa124 | 1300 | return ERR_CAST(shadow_batch_obj); |
71745376 | 1301 | |
33a051a5 CW |
1302 | ret = intel_engine_cmd_parser(engine, |
1303 | batch_obj, | |
1304 | shadow_batch_obj, | |
1305 | batch_start_offset, | |
1306 | batch_len, | |
1307 | is_master); | |
17cabf57 CW |
1308 | if (ret) |
1309 | goto err; | |
71745376 | 1310 | |
de895082 | 1311 | ret = i915_gem_object_ggtt_pin(shadow_batch_obj, NULL, 0, 0, 0); |
17cabf57 CW |
1312 | if (ret) |
1313 | goto err; | |
71745376 | 1314 | |
de4e783a CW |
1315 | i915_gem_object_unpin_pages(shadow_batch_obj); |
1316 | ||
17cabf57 | 1317 | memset(shadow_exec_entry, 0, sizeof(*shadow_exec_entry)); |
71745376 | 1318 | |
17cabf57 CW |
1319 | vma = i915_gem_obj_to_ggtt(shadow_batch_obj); |
1320 | vma->exec_entry = shadow_exec_entry; | |
de4e783a | 1321 | vma->exec_entry->flags = __EXEC_OBJECT_HAS_PIN; |
25dc556a | 1322 | i915_gem_object_get(shadow_batch_obj); |
17cabf57 | 1323 | list_add_tail(&vma->exec_list, &eb->vmas); |
71745376 | 1324 | |
59bfa124 | 1325 | return vma; |
71745376 | 1326 | |
17cabf57 | 1327 | err: |
de4e783a | 1328 | i915_gem_object_unpin_pages(shadow_batch_obj); |
17cabf57 | 1329 | if (ret == -EACCES) /* unhandled chained batch */ |
59bfa124 | 1330 | return NULL; |
17cabf57 CW |
1331 | else |
1332 | return ERR_PTR(ret); | |
71745376 | 1333 | } |
5c6c6003 | 1334 | |
5b043f4e CW |
1335 | static int |
1336 | execbuf_submit(struct i915_execbuffer_params *params, | |
1337 | struct drm_i915_gem_execbuffer2 *args, | |
1338 | struct list_head *vmas) | |
78382593 | 1339 | { |
b5321f30 | 1340 | struct drm_i915_private *dev_priv = params->request->i915; |
5f19e2bf | 1341 | u64 exec_start, exec_len; |
78382593 OM |
1342 | int instp_mode; |
1343 | u32 instp_mask; | |
2f5945bc | 1344 | int ret; |
78382593 | 1345 | |
535fbe82 | 1346 | ret = i915_gem_execbuffer_move_to_gpu(params->request, vmas); |
78382593 | 1347 | if (ret) |
2f5945bc | 1348 | return ret; |
78382593 | 1349 | |
ba01cc93 | 1350 | ret = i915_switch_context(params->request); |
78382593 | 1351 | if (ret) |
2f5945bc | 1352 | return ret; |
78382593 OM |
1353 | |
1354 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; | |
1355 | instp_mask = I915_EXEC_CONSTANTS_MASK; | |
1356 | switch (instp_mode) { | |
1357 | case I915_EXEC_CONSTANTS_REL_GENERAL: | |
1358 | case I915_EXEC_CONSTANTS_ABSOLUTE: | |
1359 | case I915_EXEC_CONSTANTS_REL_SURFACE: | |
b5321f30 | 1360 | if (instp_mode != 0 && params->engine->id != RCS) { |
78382593 | 1361 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); |
2f5945bc | 1362 | return -EINVAL; |
78382593 OM |
1363 | } |
1364 | ||
1365 | if (instp_mode != dev_priv->relative_constants_mode) { | |
b5321f30 | 1366 | if (INTEL_INFO(dev_priv)->gen < 4) { |
78382593 | 1367 | DRM_DEBUG("no rel constants on pre-gen4\n"); |
2f5945bc | 1368 | return -EINVAL; |
78382593 OM |
1369 | } |
1370 | ||
b5321f30 | 1371 | if (INTEL_INFO(dev_priv)->gen > 5 && |
78382593 OM |
1372 | instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { |
1373 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); | |
2f5945bc | 1374 | return -EINVAL; |
78382593 OM |
1375 | } |
1376 | ||
1377 | /* The HW changed the meaning on this bit on gen6 */ | |
b5321f30 | 1378 | if (INTEL_INFO(dev_priv)->gen >= 6) |
78382593 OM |
1379 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
1380 | } | |
1381 | break; | |
1382 | default: | |
1383 | DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); | |
2f5945bc | 1384 | return -EINVAL; |
78382593 OM |
1385 | } |
1386 | ||
b5321f30 | 1387 | if (params->engine->id == RCS && |
2f5945bc | 1388 | instp_mode != dev_priv->relative_constants_mode) { |
7e37f889 | 1389 | struct intel_ring *ring = params->request->ring; |
b5321f30 | 1390 | |
5fb9de1a | 1391 | ret = intel_ring_begin(params->request, 4); |
78382593 | 1392 | if (ret) |
2f5945bc | 1393 | return ret; |
78382593 | 1394 | |
b5321f30 CW |
1395 | intel_ring_emit(ring, MI_NOOP); |
1396 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); | |
1397 | intel_ring_emit_reg(ring, INSTPM); | |
1398 | intel_ring_emit(ring, instp_mask << 16 | instp_mode); | |
1399 | intel_ring_advance(ring); | |
78382593 OM |
1400 | |
1401 | dev_priv->relative_constants_mode = instp_mode; | |
1402 | } | |
1403 | ||
1404 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { | |
b5321f30 | 1405 | ret = i915_reset_gen7_sol_offsets(params->request); |
78382593 | 1406 | if (ret) |
2f5945bc | 1407 | return ret; |
78382593 OM |
1408 | } |
1409 | ||
5f19e2bf | 1410 | exec_len = args->batch_len; |
59bfa124 | 1411 | exec_start = params->batch->node.start + |
5f19e2bf JH |
1412 | params->args_batch_start_offset; |
1413 | ||
9d611c03 | 1414 | if (exec_len == 0) |
59bfa124 | 1415 | exec_len = params->batch->size; |
9d611c03 | 1416 | |
803688ba CW |
1417 | ret = params->engine->emit_bb_start(params->request, |
1418 | exec_start, exec_len, | |
1419 | params->dispatch_flags); | |
2f5945bc CW |
1420 | if (ret) |
1421 | return ret; | |
78382593 | 1422 | |
95c24161 | 1423 | trace_i915_gem_ring_dispatch(params->request, params->dispatch_flags); |
78382593 | 1424 | |
8a8edb59 | 1425 | i915_gem_execbuffer_move_to_active(vmas, params->request); |
78382593 | 1426 | |
2f5945bc | 1427 | return 0; |
78382593 OM |
1428 | } |
1429 | ||
a8ebba75 ZY |
1430 | /** |
1431 | * Find one BSD ring to dispatch the corresponding BSD command. | |
c80ff16e | 1432 | * The engine index is returned. |
a8ebba75 | 1433 | */ |
de1add36 | 1434 | static unsigned int |
c80ff16e CW |
1435 | gen8_dispatch_bsd_engine(struct drm_i915_private *dev_priv, |
1436 | struct drm_file *file) | |
a8ebba75 | 1437 | { |
a8ebba75 ZY |
1438 | struct drm_i915_file_private *file_priv = file->driver_priv; |
1439 | ||
de1add36 | 1440 | /* Check whether the file_priv has already selected one ring. */ |
c80ff16e | 1441 | if ((int)file_priv->bsd_engine < 0) { |
de1add36 | 1442 | /* If not, use the ping-pong mechanism to select one. */ |
91c8a326 | 1443 | mutex_lock(&dev_priv->drm.struct_mutex); |
c80ff16e CW |
1444 | file_priv->bsd_engine = dev_priv->mm.bsd_engine_dispatch_index; |
1445 | dev_priv->mm.bsd_engine_dispatch_index ^= 1; | |
91c8a326 | 1446 | mutex_unlock(&dev_priv->drm.struct_mutex); |
a8ebba75 | 1447 | } |
de1add36 | 1448 | |
c80ff16e | 1449 | return file_priv->bsd_engine; |
a8ebba75 ZY |
1450 | } |
1451 | ||
de1add36 TU |
1452 | #define I915_USER_RINGS (4) |
1453 | ||
117897f4 | 1454 | static const enum intel_engine_id user_ring_map[I915_USER_RINGS + 1] = { |
de1add36 TU |
1455 | [I915_EXEC_DEFAULT] = RCS, |
1456 | [I915_EXEC_RENDER] = RCS, | |
1457 | [I915_EXEC_BLT] = BCS, | |
1458 | [I915_EXEC_BSD] = VCS, | |
1459 | [I915_EXEC_VEBOX] = VECS | |
1460 | }; | |
1461 | ||
f8ca0c07 DG |
1462 | static struct intel_engine_cs * |
1463 | eb_select_engine(struct drm_i915_private *dev_priv, | |
1464 | struct drm_file *file, | |
1465 | struct drm_i915_gem_execbuffer2 *args) | |
de1add36 TU |
1466 | { |
1467 | unsigned int user_ring_id = args->flags & I915_EXEC_RING_MASK; | |
f8ca0c07 | 1468 | struct intel_engine_cs *engine; |
de1add36 TU |
1469 | |
1470 | if (user_ring_id > I915_USER_RINGS) { | |
1471 | DRM_DEBUG("execbuf with unknown ring: %u\n", user_ring_id); | |
f8ca0c07 | 1472 | return NULL; |
de1add36 TU |
1473 | } |
1474 | ||
1475 | if ((user_ring_id != I915_EXEC_BSD) && | |
1476 | ((args->flags & I915_EXEC_BSD_MASK) != 0)) { | |
1477 | DRM_DEBUG("execbuf with non bsd ring but with invalid " | |
1478 | "bsd dispatch flags: %d\n", (int)(args->flags)); | |
f8ca0c07 | 1479 | return NULL; |
de1add36 TU |
1480 | } |
1481 | ||
1482 | if (user_ring_id == I915_EXEC_BSD && HAS_BSD2(dev_priv)) { | |
1483 | unsigned int bsd_idx = args->flags & I915_EXEC_BSD_MASK; | |
1484 | ||
1485 | if (bsd_idx == I915_EXEC_BSD_DEFAULT) { | |
c80ff16e | 1486 | bsd_idx = gen8_dispatch_bsd_engine(dev_priv, file); |
de1add36 TU |
1487 | } else if (bsd_idx >= I915_EXEC_BSD_RING1 && |
1488 | bsd_idx <= I915_EXEC_BSD_RING2) { | |
d9da6aa0 | 1489 | bsd_idx >>= I915_EXEC_BSD_SHIFT; |
de1add36 TU |
1490 | bsd_idx--; |
1491 | } else { | |
1492 | DRM_DEBUG("execbuf with unknown bsd ring: %u\n", | |
1493 | bsd_idx); | |
f8ca0c07 | 1494 | return NULL; |
de1add36 TU |
1495 | } |
1496 | ||
f8ca0c07 | 1497 | engine = &dev_priv->engine[_VCS(bsd_idx)]; |
de1add36 | 1498 | } else { |
f8ca0c07 | 1499 | engine = &dev_priv->engine[user_ring_map[user_ring_id]]; |
de1add36 TU |
1500 | } |
1501 | ||
f8ca0c07 | 1502 | if (!intel_engine_initialized(engine)) { |
de1add36 | 1503 | DRM_DEBUG("execbuf with invalid ring: %u\n", user_ring_id); |
f8ca0c07 | 1504 | return NULL; |
de1add36 TU |
1505 | } |
1506 | ||
f8ca0c07 | 1507 | return engine; |
de1add36 TU |
1508 | } |
1509 | ||
54cf91dc CW |
1510 | static int |
1511 | i915_gem_do_execbuffer(struct drm_device *dev, void *data, | |
1512 | struct drm_file *file, | |
1513 | struct drm_i915_gem_execbuffer2 *args, | |
41bde553 | 1514 | struct drm_i915_gem_exec_object2 *exec) |
54cf91dc | 1515 | { |
72e96d64 JL |
1516 | struct drm_i915_private *dev_priv = to_i915(dev); |
1517 | struct i915_ggtt *ggtt = &dev_priv->ggtt; | |
27173f1f | 1518 | struct eb_vmas *eb; |
78a42377 | 1519 | struct drm_i915_gem_exec_object2 shadow_exec_entry; |
e2f80391 | 1520 | struct intel_engine_cs *engine; |
e2efd130 | 1521 | struct i915_gem_context *ctx; |
41bde553 | 1522 | struct i915_address_space *vm; |
5f19e2bf JH |
1523 | struct i915_execbuffer_params params_master; /* XXX: will be removed later */ |
1524 | struct i915_execbuffer_params *params = ¶ms_master; | |
d299cce7 | 1525 | const u32 ctx_id = i915_execbuffer2_get_context_id(*args); |
8e004efc | 1526 | u32 dispatch_flags; |
78382593 | 1527 | int ret; |
ed5982e6 | 1528 | bool need_relocs; |
54cf91dc | 1529 | |
ed5982e6 | 1530 | if (!i915_gem_check_execbuffer(args)) |
432e58ed | 1531 | return -EINVAL; |
432e58ed | 1532 | |
ad19f10b | 1533 | ret = validate_exec_list(dev, exec, args->buffer_count); |
54cf91dc CW |
1534 | if (ret) |
1535 | return ret; | |
1536 | ||
8e004efc | 1537 | dispatch_flags = 0; |
d7d4eedd | 1538 | if (args->flags & I915_EXEC_SECURE) { |
b3ac9f25 | 1539 | if (!drm_is_current_master(file) || !capable(CAP_SYS_ADMIN)) |
d7d4eedd CW |
1540 | return -EPERM; |
1541 | ||
8e004efc | 1542 | dispatch_flags |= I915_DISPATCH_SECURE; |
d7d4eedd | 1543 | } |
b45305fc | 1544 | if (args->flags & I915_EXEC_IS_PINNED) |
8e004efc | 1545 | dispatch_flags |= I915_DISPATCH_PINNED; |
d7d4eedd | 1546 | |
f8ca0c07 DG |
1547 | engine = eb_select_engine(dev_priv, file, args); |
1548 | if (!engine) | |
1549 | return -EINVAL; | |
54cf91dc CW |
1550 | |
1551 | if (args->buffer_count < 1) { | |
ff240199 | 1552 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
54cf91dc CW |
1553 | return -EINVAL; |
1554 | } | |
54cf91dc | 1555 | |
a9ed33ca AJ |
1556 | if (args->flags & I915_EXEC_RESOURCE_STREAMER) { |
1557 | if (!HAS_RESOURCE_STREAMER(dev)) { | |
1558 | DRM_DEBUG("RS is only allowed for Haswell, Gen8 and above\n"); | |
1559 | return -EINVAL; | |
1560 | } | |
e2f80391 | 1561 | if (engine->id != RCS) { |
a9ed33ca | 1562 | DRM_DEBUG("RS is not available on %s\n", |
e2f80391 | 1563 | engine->name); |
a9ed33ca AJ |
1564 | return -EINVAL; |
1565 | } | |
1566 | ||
1567 | dispatch_flags |= I915_DISPATCH_RS; | |
1568 | } | |
1569 | ||
67d97da3 CW |
1570 | /* Take a local wakeref for preparing to dispatch the execbuf as |
1571 | * we expect to access the hardware fairly frequently in the | |
1572 | * process. Upon first dispatch, we acquire another prolonged | |
1573 | * wakeref that we hold until the GPU has been idle for at least | |
1574 | * 100ms. | |
1575 | */ | |
f65c9168 PZ |
1576 | intel_runtime_pm_get(dev_priv); |
1577 | ||
54cf91dc CW |
1578 | ret = i915_mutex_lock_interruptible(dev); |
1579 | if (ret) | |
1580 | goto pre_mutex_err; | |
1581 | ||
e2f80391 | 1582 | ctx = i915_gem_validate_context(dev, file, engine, ctx_id); |
72ad5c45 | 1583 | if (IS_ERR(ctx)) { |
d299cce7 | 1584 | mutex_unlock(&dev->struct_mutex); |
41bde553 | 1585 | ret = PTR_ERR(ctx); |
d299cce7 | 1586 | goto pre_mutex_err; |
935f38d6 | 1587 | } |
41bde553 | 1588 | |
9a6feaf0 | 1589 | i915_gem_context_get(ctx); |
41bde553 | 1590 | |
ae6c4806 DV |
1591 | if (ctx->ppgtt) |
1592 | vm = &ctx->ppgtt->base; | |
1593 | else | |
72e96d64 | 1594 | vm = &ggtt->base; |
d299cce7 | 1595 | |
5f19e2bf JH |
1596 | memset(¶ms_master, 0x00, sizeof(params_master)); |
1597 | ||
17601cbc | 1598 | eb = eb_create(args); |
67731b87 | 1599 | if (eb == NULL) { |
9a6feaf0 | 1600 | i915_gem_context_put(ctx); |
67731b87 CW |
1601 | mutex_unlock(&dev->struct_mutex); |
1602 | ret = -ENOMEM; | |
1603 | goto pre_mutex_err; | |
1604 | } | |
1605 | ||
54cf91dc | 1606 | /* Look up object handles */ |
27173f1f | 1607 | ret = eb_lookup_vmas(eb, exec, args, vm, file); |
3b96eff4 CW |
1608 | if (ret) |
1609 | goto err; | |
54cf91dc | 1610 | |
6fe4f140 | 1611 | /* take note of the batch buffer before we might reorder the lists */ |
59bfa124 | 1612 | params->batch = eb_get_batch(eb); |
6fe4f140 | 1613 | |
54cf91dc | 1614 | /* Move the objects en-masse into the GTT, evicting if necessary. */ |
ed5982e6 | 1615 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
e2f80391 TU |
1616 | ret = i915_gem_execbuffer_reserve(engine, &eb->vmas, ctx, |
1617 | &need_relocs); | |
54cf91dc CW |
1618 | if (ret) |
1619 | goto err; | |
1620 | ||
1621 | /* The objects are in their final locations, apply the relocations. */ | |
ed5982e6 | 1622 | if (need_relocs) |
17601cbc | 1623 | ret = i915_gem_execbuffer_relocate(eb); |
54cf91dc CW |
1624 | if (ret) { |
1625 | if (ret == -EFAULT) { | |
e2f80391 TU |
1626 | ret = i915_gem_execbuffer_relocate_slow(dev, args, file, |
1627 | engine, | |
b1b38278 | 1628 | eb, exec, ctx); |
54cf91dc CW |
1629 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
1630 | } | |
1631 | if (ret) | |
1632 | goto err; | |
1633 | } | |
1634 | ||
1635 | /* Set the pending read domains for the batch buffer to COMMAND */ | |
59bfa124 | 1636 | if (params->batch->obj->base.pending_write_domain) { |
ff240199 | 1637 | DRM_DEBUG("Attempting to use self-modifying batch buffer\n"); |
54cf91dc CW |
1638 | ret = -EINVAL; |
1639 | goto err; | |
1640 | } | |
54cf91dc | 1641 | |
5f19e2bf | 1642 | params->args_batch_start_offset = args->batch_start_offset; |
33a051a5 | 1643 | if (intel_engine_needs_cmd_parser(engine) && args->batch_len) { |
59bfa124 CW |
1644 | struct i915_vma *vma; |
1645 | ||
1646 | vma = i915_gem_execbuffer_parse(engine, &shadow_exec_entry, | |
1647 | params->batch->obj, | |
1648 | eb, | |
1649 | args->batch_start_offset, | |
1650 | args->batch_len, | |
1651 | drm_is_current_master(file)); | |
1652 | if (IS_ERR(vma)) { | |
1653 | ret = PTR_ERR(vma); | |
78a42377 BV |
1654 | goto err; |
1655 | } | |
17cabf57 | 1656 | |
59bfa124 | 1657 | if (vma) { |
c7c7372e RP |
1658 | /* |
1659 | * Batch parsed and accepted: | |
1660 | * | |
1661 | * Set the DISPATCH_SECURE bit to remove the NON_SECURE | |
1662 | * bit from MI_BATCH_BUFFER_START commands issued in | |
1663 | * the dispatch_execbuffer implementations. We | |
1664 | * specifically don't want that set on batches the | |
1665 | * command parser has accepted. | |
1666 | */ | |
1667 | dispatch_flags |= I915_DISPATCH_SECURE; | |
5f19e2bf | 1668 | params->args_batch_start_offset = 0; |
59bfa124 | 1669 | params->batch = vma; |
c7c7372e | 1670 | } |
351e3db2 BV |
1671 | } |
1672 | ||
59bfa124 | 1673 | params->batch->obj->base.pending_read_domains |= I915_GEM_DOMAIN_COMMAND; |
78a42377 | 1674 | |
d7d4eedd CW |
1675 | /* snb/ivb/vlv conflate the "batch in ppgtt" bit with the "non-secure |
1676 | * batch" bit. Hence we need to pin secure batches into the global gtt. | |
28cf5415 | 1677 | * hsw should have this fixed, but bdw mucks it up again. */ |
8e004efc | 1678 | if (dispatch_flags & I915_DISPATCH_SECURE) { |
59bfa124 CW |
1679 | struct drm_i915_gem_object *obj = params->batch->obj; |
1680 | ||
da51a1e7 DV |
1681 | /* |
1682 | * So on first glance it looks freaky that we pin the batch here | |
1683 | * outside of the reservation loop. But: | |
1684 | * - The batch is already pinned into the relevant ppgtt, so we | |
1685 | * already have the backing storage fully allocated. | |
1686 | * - No other BO uses the global gtt (well contexts, but meh), | |
fd0753cf | 1687 | * so we don't really have issues with multiple objects not |
da51a1e7 DV |
1688 | * fitting due to fragmentation. |
1689 | * So this is actually safe. | |
1690 | */ | |
de895082 | 1691 | ret = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, 0); |
da51a1e7 DV |
1692 | if (ret) |
1693 | goto err; | |
d7d4eedd | 1694 | |
59bfa124 CW |
1695 | params->batch = i915_gem_obj_to_ggtt(obj); |
1696 | } | |
d7d4eedd | 1697 | |
0c8dac88 | 1698 | /* Allocate a request for this batch buffer nice and early. */ |
8e637178 CW |
1699 | params->request = i915_gem_request_alloc(engine, ctx); |
1700 | if (IS_ERR(params->request)) { | |
1701 | ret = PTR_ERR(params->request); | |
0c8dac88 | 1702 | goto err_batch_unpin; |
26827088 | 1703 | } |
0c8dac88 | 1704 | |
17f298cf CW |
1705 | /* Whilst this request exists, batch_obj will be on the |
1706 | * active_list, and so will hold the active reference. Only when this | |
1707 | * request is retired will the the batch_obj be moved onto the | |
1708 | * inactive_list and lose its active reference. Hence we do not need | |
1709 | * to explicitly hold another reference here. | |
1710 | */ | |
1711 | params->request->batch_obj = params->batch->obj; | |
1712 | ||
8e637178 | 1713 | ret = i915_gem_request_add_to_client(params->request, file); |
fcfa423c | 1714 | if (ret) |
aa9b7810 | 1715 | goto err_request; |
fcfa423c | 1716 | |
5f19e2bf JH |
1717 | /* |
1718 | * Save assorted stuff away to pass through to *_submission(). | |
1719 | * NB: This data should be 'persistent' and not local as it will | |
1720 | * kept around beyond the duration of the IOCTL once the GPU | |
1721 | * scheduler arrives. | |
1722 | */ | |
1723 | params->dev = dev; | |
1724 | params->file = file; | |
4a570db5 | 1725 | params->engine = engine; |
5f19e2bf | 1726 | params->dispatch_flags = dispatch_flags; |
5f19e2bf JH |
1727 | params->ctx = ctx; |
1728 | ||
5b043f4e | 1729 | ret = execbuf_submit(params, args, &eb->vmas); |
aa9b7810 | 1730 | err_request: |
17f298cf | 1731 | __i915_add_request(params->request, ret == 0); |
54cf91dc | 1732 | |
0c8dac88 | 1733 | err_batch_unpin: |
da51a1e7 DV |
1734 | /* |
1735 | * FIXME: We crucially rely upon the active tracking for the (ppgtt) | |
1736 | * batch vma for correctness. For less ugly and less fragility this | |
1737 | * needs to be adjusted to also track the ggtt batch vma properly as | |
1738 | * active. | |
1739 | */ | |
8e004efc | 1740 | if (dispatch_flags & I915_DISPATCH_SECURE) |
59bfa124 | 1741 | i915_vma_unpin(params->batch); |
54cf91dc | 1742 | err: |
41bde553 | 1743 | /* the request owns the ref now */ |
9a6feaf0 | 1744 | i915_gem_context_put(ctx); |
67731b87 | 1745 | eb_destroy(eb); |
54cf91dc CW |
1746 | |
1747 | mutex_unlock(&dev->struct_mutex); | |
1748 | ||
1749 | pre_mutex_err: | |
f65c9168 PZ |
1750 | /* intel_gpu_busy should also get a ref, so it will free when the device |
1751 | * is really idle. */ | |
1752 | intel_runtime_pm_put(dev_priv); | |
54cf91dc CW |
1753 | return ret; |
1754 | } | |
1755 | ||
1756 | /* | |
1757 | * Legacy execbuffer just creates an exec2 list from the original exec object | |
1758 | * list array and passes it to the real function. | |
1759 | */ | |
1760 | int | |
1761 | i915_gem_execbuffer(struct drm_device *dev, void *data, | |
1762 | struct drm_file *file) | |
1763 | { | |
1764 | struct drm_i915_gem_execbuffer *args = data; | |
1765 | struct drm_i915_gem_execbuffer2 exec2; | |
1766 | struct drm_i915_gem_exec_object *exec_list = NULL; | |
1767 | struct drm_i915_gem_exec_object2 *exec2_list = NULL; | |
1768 | int ret, i; | |
1769 | ||
54cf91dc | 1770 | if (args->buffer_count < 1) { |
ff240199 | 1771 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
54cf91dc CW |
1772 | return -EINVAL; |
1773 | } | |
1774 | ||
1775 | /* Copy in the exec list from userland */ | |
1776 | exec_list = drm_malloc_ab(sizeof(*exec_list), args->buffer_count); | |
1777 | exec2_list = drm_malloc_ab(sizeof(*exec2_list), args->buffer_count); | |
1778 | if (exec_list == NULL || exec2_list == NULL) { | |
ff240199 | 1779 | DRM_DEBUG("Failed to allocate exec list for %d buffers\n", |
54cf91dc CW |
1780 | args->buffer_count); |
1781 | drm_free_large(exec_list); | |
1782 | drm_free_large(exec2_list); | |
1783 | return -ENOMEM; | |
1784 | } | |
1785 | ret = copy_from_user(exec_list, | |
3ed605bc | 1786 | u64_to_user_ptr(args->buffers_ptr), |
54cf91dc CW |
1787 | sizeof(*exec_list) * args->buffer_count); |
1788 | if (ret != 0) { | |
ff240199 | 1789 | DRM_DEBUG("copy %d exec entries failed %d\n", |
54cf91dc CW |
1790 | args->buffer_count, ret); |
1791 | drm_free_large(exec_list); | |
1792 | drm_free_large(exec2_list); | |
1793 | return -EFAULT; | |
1794 | } | |
1795 | ||
1796 | for (i = 0; i < args->buffer_count; i++) { | |
1797 | exec2_list[i].handle = exec_list[i].handle; | |
1798 | exec2_list[i].relocation_count = exec_list[i].relocation_count; | |
1799 | exec2_list[i].relocs_ptr = exec_list[i].relocs_ptr; | |
1800 | exec2_list[i].alignment = exec_list[i].alignment; | |
1801 | exec2_list[i].offset = exec_list[i].offset; | |
1802 | if (INTEL_INFO(dev)->gen < 4) | |
1803 | exec2_list[i].flags = EXEC_OBJECT_NEEDS_FENCE; | |
1804 | else | |
1805 | exec2_list[i].flags = 0; | |
1806 | } | |
1807 | ||
1808 | exec2.buffers_ptr = args->buffers_ptr; | |
1809 | exec2.buffer_count = args->buffer_count; | |
1810 | exec2.batch_start_offset = args->batch_start_offset; | |
1811 | exec2.batch_len = args->batch_len; | |
1812 | exec2.DR1 = args->DR1; | |
1813 | exec2.DR4 = args->DR4; | |
1814 | exec2.num_cliprects = args->num_cliprects; | |
1815 | exec2.cliprects_ptr = args->cliprects_ptr; | |
1816 | exec2.flags = I915_EXEC_RENDER; | |
6e0a69db | 1817 | i915_execbuffer2_set_context_id(exec2, 0); |
54cf91dc | 1818 | |
41bde553 | 1819 | ret = i915_gem_do_execbuffer(dev, data, file, &exec2, exec2_list); |
54cf91dc | 1820 | if (!ret) { |
9aab8bff | 1821 | struct drm_i915_gem_exec_object __user *user_exec_list = |
3ed605bc | 1822 | u64_to_user_ptr(args->buffers_ptr); |
9aab8bff | 1823 | |
54cf91dc | 1824 | /* Copy the new buffer offsets back to the user's exec list. */ |
9aab8bff | 1825 | for (i = 0; i < args->buffer_count; i++) { |
934acce3 MW |
1826 | exec2_list[i].offset = |
1827 | gen8_canonical_addr(exec2_list[i].offset); | |
9aab8bff CW |
1828 | ret = __copy_to_user(&user_exec_list[i].offset, |
1829 | &exec2_list[i].offset, | |
1830 | sizeof(user_exec_list[i].offset)); | |
1831 | if (ret) { | |
1832 | ret = -EFAULT; | |
1833 | DRM_DEBUG("failed to copy %d exec entries " | |
1834 | "back to user (%d)\n", | |
1835 | args->buffer_count, ret); | |
1836 | break; | |
1837 | } | |
54cf91dc CW |
1838 | } |
1839 | } | |
1840 | ||
1841 | drm_free_large(exec_list); | |
1842 | drm_free_large(exec2_list); | |
1843 | return ret; | |
1844 | } | |
1845 | ||
1846 | int | |
1847 | i915_gem_execbuffer2(struct drm_device *dev, void *data, | |
1848 | struct drm_file *file) | |
1849 | { | |
1850 | struct drm_i915_gem_execbuffer2 *args = data; | |
1851 | struct drm_i915_gem_exec_object2 *exec2_list = NULL; | |
1852 | int ret; | |
1853 | ||
ed8cd3b2 XW |
1854 | if (args->buffer_count < 1 || |
1855 | args->buffer_count > UINT_MAX / sizeof(*exec2_list)) { | |
ff240199 | 1856 | DRM_DEBUG("execbuf2 with %d buffers\n", args->buffer_count); |
54cf91dc CW |
1857 | return -EINVAL; |
1858 | } | |
1859 | ||
9cb34664 DV |
1860 | if (args->rsvd2 != 0) { |
1861 | DRM_DEBUG("dirty rvsd2 field\n"); | |
1862 | return -EINVAL; | |
1863 | } | |
1864 | ||
f2a85e19 CW |
1865 | exec2_list = drm_malloc_gfp(args->buffer_count, |
1866 | sizeof(*exec2_list), | |
1867 | GFP_TEMPORARY); | |
54cf91dc | 1868 | if (exec2_list == NULL) { |
ff240199 | 1869 | DRM_DEBUG("Failed to allocate exec list for %d buffers\n", |
54cf91dc CW |
1870 | args->buffer_count); |
1871 | return -ENOMEM; | |
1872 | } | |
1873 | ret = copy_from_user(exec2_list, | |
3ed605bc | 1874 | u64_to_user_ptr(args->buffers_ptr), |
54cf91dc CW |
1875 | sizeof(*exec2_list) * args->buffer_count); |
1876 | if (ret != 0) { | |
ff240199 | 1877 | DRM_DEBUG("copy %d exec entries failed %d\n", |
54cf91dc CW |
1878 | args->buffer_count, ret); |
1879 | drm_free_large(exec2_list); | |
1880 | return -EFAULT; | |
1881 | } | |
1882 | ||
41bde553 | 1883 | ret = i915_gem_do_execbuffer(dev, data, file, args, exec2_list); |
54cf91dc CW |
1884 | if (!ret) { |
1885 | /* Copy the new buffer offsets back to the user's exec list. */ | |
d593d992 | 1886 | struct drm_i915_gem_exec_object2 __user *user_exec_list = |
3ed605bc | 1887 | u64_to_user_ptr(args->buffers_ptr); |
9aab8bff CW |
1888 | int i; |
1889 | ||
1890 | for (i = 0; i < args->buffer_count; i++) { | |
934acce3 MW |
1891 | exec2_list[i].offset = |
1892 | gen8_canonical_addr(exec2_list[i].offset); | |
9aab8bff CW |
1893 | ret = __copy_to_user(&user_exec_list[i].offset, |
1894 | &exec2_list[i].offset, | |
1895 | sizeof(user_exec_list[i].offset)); | |
1896 | if (ret) { | |
1897 | ret = -EFAULT; | |
1898 | DRM_DEBUG("failed to copy %d exec entries " | |
1899 | "back to user\n", | |
1900 | args->buffer_count); | |
1901 | break; | |
1902 | } | |
54cf91dc CW |
1903 | } |
1904 | } | |
1905 | ||
1906 | drm_free_large(exec2_list); | |
1907 | return ret; | |
1908 | } |