Rev 4104 | Rev 4371 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 4104 | Rev 4246 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright © 2008,2010 Intel Corporation |
2 | * Copyright © 2008,2010 Intel Corporation |
3 | * |
3 | * |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * copy of this software and associated documentation files (the "Software"), |
5 | * copy of this software and associated documentation files (the "Software"), |
6 | * to deal in the Software without restriction, including without limitation |
6 | * to deal in the Software without restriction, including without limitation |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
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 |
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: |
9 | * Software is furnished to do so, subject to the following conditions: |
10 | * |
10 | * |
11 | * The above copyright notice and this permission notice (including the next |
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 |
12 | * paragraph) shall be included in all copies or substantial portions of the |
13 | * Software. |
13 | * Software. |
14 | * |
14 | * |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
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, |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
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 |
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 |
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 |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
21 | * IN THE SOFTWARE. |
21 | * IN THE SOFTWARE. |
22 | * |
22 | * |
23 | * Authors: |
23 | * Authors: |
24 | * Eric Anholt |
24 | * Eric Anholt |
25 | * Chris Wilson |
25 | * Chris Wilson |
26 | * |
26 | * |
27 | */ |
27 | */ |
28 | 28 | ||
29 | #include |
29 | #include |
30 | #include |
30 | #include |
31 | #include "i915_drv.h" |
31 | #include "i915_drv.h" |
32 | #include "i915_trace.h" |
32 | #include "i915_trace.h" |
33 | #include "intel_drv.h" |
33 | #include "intel_drv.h" |
34 | //#include |
34 | //#include |
35 | 35 | ||
36 | #define I915_EXEC_SECURE (1<<9) |
36 | #define I915_EXEC_SECURE (1<<9) |
37 | #define I915_EXEC_IS_PINNED (1<<10) |
37 | #define I915_EXEC_IS_PINNED (1<<10) |
38 | #define I915_EXEC_VEBOX (4<<0) |
38 | #define I915_EXEC_VEBOX (4<<0) |
39 | 39 | ||
40 | #define wmb() asm volatile ("sfence") |
40 | #define wmb() asm volatile ("sfence") |
41 | 41 | ||
42 | struct drm_i915_gem_object *get_fb_obj(); |
42 | struct drm_i915_gem_object *get_fb_obj(); |
43 | 43 | ||
44 | 44 | ||
45 | static unsigned long |
45 | static unsigned long |
46 | copy_to_user(void __user *to, const void *from, unsigned long n) |
46 | copy_to_user(void __user *to, const void *from, unsigned long n) |
47 | { |
47 | { |
48 | memcpy(to, from, n); |
48 | memcpy(to, from, n); |
49 | return 0; |
49 | return 0; |
50 | } |
50 | } |
51 | 51 | ||
52 | static unsigned long |
52 | static unsigned long |
53 | copy_from_user(void *to, const void __user *from, unsigned long n) |
53 | copy_from_user(void *to, const void __user *from, unsigned long n) |
54 | { |
54 | { |
55 | memcpy(to, from, n); |
55 | memcpy(to, from, n); |
56 | return 0; |
56 | return 0; |
57 | } |
57 | } |
58 | 58 | ||
59 | struct eb_objects { |
59 | struct eb_objects { |
60 | struct list_head objects; |
60 | struct list_head objects; |
61 | int and; |
61 | int and; |
62 | union { |
62 | union { |
63 | struct drm_i915_gem_object *lut[0]; |
63 | struct drm_i915_gem_object *lut[0]; |
64 | struct hlist_head buckets[0]; |
64 | struct hlist_head buckets[0]; |
65 | }; |
65 | }; |
66 | }; |
66 | }; |
67 | 67 | ||
68 | static struct eb_objects * |
68 | static struct eb_objects * |
69 | eb_create(struct drm_i915_gem_execbuffer2 *args) |
69 | eb_create(struct drm_i915_gem_execbuffer2 *args) |
70 | { |
70 | { |
71 | struct eb_objects *eb = NULL; |
71 | struct eb_objects *eb = NULL; |
72 | 72 | ||
73 | if (args->flags & I915_EXEC_HANDLE_LUT) { |
73 | if (args->flags & I915_EXEC_HANDLE_LUT) { |
74 | int size = args->buffer_count; |
74 | int size = args->buffer_count; |
75 | size *= sizeof(struct drm_i915_gem_object *); |
75 | size *= sizeof(struct drm_i915_gem_object *); |
76 | size += sizeof(struct eb_objects); |
76 | size += sizeof(struct eb_objects); |
77 | eb = kmalloc(size, GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY); |
77 | eb = kmalloc(size, GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY); |
78 | } |
78 | } |
79 | 79 | ||
80 | if (eb == NULL) { |
80 | if (eb == NULL) { |
81 | int size = args->buffer_count; |
81 | int size = args->buffer_count; |
82 | int count = PAGE_SIZE / sizeof(struct hlist_head) / 2; |
82 | int count = PAGE_SIZE / sizeof(struct hlist_head) / 2; |
83 | BUILD_BUG_ON_NOT_POWER_OF_2(PAGE_SIZE / sizeof(struct hlist_head)); |
83 | BUILD_BUG_ON_NOT_POWER_OF_2(PAGE_SIZE / sizeof(struct hlist_head)); |
84 | while (count > 2*size) |
84 | while (count > 2*size) |
85 | count >>= 1; |
85 | count >>= 1; |
86 | eb = kzalloc(count*sizeof(struct hlist_head) + |
86 | eb = kzalloc(count*sizeof(struct hlist_head) + |
87 | sizeof(struct eb_objects), |
87 | sizeof(struct eb_objects), |
88 | GFP_TEMPORARY); |
88 | GFP_TEMPORARY); |
89 | if (eb == NULL) |
89 | if (eb == NULL) |
90 | return eb; |
90 | return eb; |
91 | 91 | ||
92 | eb->and = count - 1; |
92 | eb->and = count - 1; |
93 | } else |
93 | } else |
94 | eb->and = -args->buffer_count; |
94 | eb->and = -args->buffer_count; |
95 | 95 | ||
96 | INIT_LIST_HEAD(&eb->objects); |
96 | INIT_LIST_HEAD(&eb->objects); |
97 | return eb; |
97 | return eb; |
98 | } |
98 | } |
99 | 99 | ||
100 | static void |
100 | static void |
101 | eb_reset(struct eb_objects *eb) |
101 | eb_reset(struct eb_objects *eb) |
102 | { |
102 | { |
103 | if (eb->and >= 0) |
103 | if (eb->and >= 0) |
104 | memset(eb->buckets, 0, (eb->and+1)*sizeof(struct hlist_head)); |
104 | memset(eb->buckets, 0, (eb->and+1)*sizeof(struct hlist_head)); |
105 | } |
105 | } |
106 | 106 | ||
107 | static int |
107 | static int |
108 | eb_lookup_objects(struct eb_objects *eb, |
108 | eb_lookup_objects(struct eb_objects *eb, |
109 | struct drm_i915_gem_exec_object2 *exec, |
109 | struct drm_i915_gem_exec_object2 *exec, |
110 | const struct drm_i915_gem_execbuffer2 *args, |
110 | const struct drm_i915_gem_execbuffer2 *args, |
111 | struct drm_file *file) |
111 | struct drm_file *file) |
112 | { |
112 | { |
113 | int i; |
113 | int i; |
114 | 114 | ||
115 | spin_lock(&file->table_lock); |
115 | spin_lock(&file->table_lock); |
116 | for (i = 0; i < args->buffer_count; i++) { |
116 | for (i = 0; i < args->buffer_count; i++) { |
117 | struct drm_i915_gem_object *obj; |
117 | struct drm_i915_gem_object *obj; |
118 | 118 | ||
119 | if(exec[i].handle == -2) |
119 | if(exec[i].handle == -2) |
120 | obj = get_fb_obj(); |
120 | obj = get_fb_obj(); |
121 | else |
121 | else |
122 | obj = to_intel_bo(idr_find(&file->object_idr, exec[i].handle)); |
122 | obj = to_intel_bo(idr_find(&file->object_idr, exec[i].handle)); |
123 | if (obj == NULL) { |
123 | if (obj == NULL) { |
124 | spin_unlock(&file->table_lock); |
124 | spin_unlock(&file->table_lock); |
125 | DRM_DEBUG("Invalid object handle %d at index %d\n", |
125 | DRM_DEBUG("Invalid object handle %d at index %d\n", |
126 | exec[i].handle, i); |
126 | exec[i].handle, i); |
127 | return -ENOENT; |
127 | return -ENOENT; |
128 | } |
128 | } |
129 | 129 | ||
130 | if (!list_empty(&obj->exec_list)) { |
130 | if (!list_empty(&obj->exec_list)) { |
131 | spin_unlock(&file->table_lock); |
131 | spin_unlock(&file->table_lock); |
132 | DRM_DEBUG("Object %p [handle %d, index %d] appears more than once in object list\n", |
132 | DRM_DEBUG("Object %p [handle %d, index %d] appears more than once in object list\n", |
133 | obj, exec[i].handle, i); |
133 | obj, exec[i].handle, i); |
134 | return -EINVAL; |
134 | return -EINVAL; |
135 | } |
135 | } |
136 | 136 | ||
137 | drm_gem_object_reference(&obj->base); |
137 | drm_gem_object_reference(&obj->base); |
138 | list_add_tail(&obj->exec_list, &eb->objects); |
138 | list_add_tail(&obj->exec_list, &eb->objects); |
139 | 139 | ||
140 | obj->exec_entry = &exec[i]; |
140 | obj->exec_entry = &exec[i]; |
- | 141 | ||
- | 142 | if(exec[i].handle == -2) |
|
- | 143 | continue; |
|
- | 144 | ||
141 | if (eb->and < 0) { |
145 | if (eb->and < 0) { |
142 | eb->lut[i] = obj; |
146 | eb->lut[i] = obj; |
143 | } else { |
147 | } else { |
144 | uint32_t handle = args->flags & I915_EXEC_HANDLE_LUT ? i : exec[i].handle; |
148 | uint32_t handle = args->flags & I915_EXEC_HANDLE_LUT ? i : exec[i].handle; |
145 | obj->exec_handle = handle; |
149 | obj->exec_handle = handle; |
146 | hlist_add_head(&obj->exec_node, |
150 | hlist_add_head(&obj->exec_node, |
147 | &eb->buckets[handle & eb->and]); |
151 | &eb->buckets[handle & eb->and]); |
148 | } |
152 | } |
149 | } |
153 | } |
150 | spin_unlock(&file->table_lock); |
154 | spin_unlock(&file->table_lock); |
151 | 155 | ||
152 | return 0; |
156 | return 0; |
153 | } |
157 | } |
154 | 158 | ||
155 | static struct drm_i915_gem_object * |
159 | static struct drm_i915_gem_object * |
156 | eb_get_object(struct eb_objects *eb, unsigned long handle) |
160 | eb_get_object(struct eb_objects *eb, unsigned long handle) |
157 | { |
161 | { |
- | 162 | ||
- | 163 | if(handle == -2) |
|
- | 164 | return get_fb_obj(); |
|
- | 165 | ||
158 | if (eb->and < 0) { |
166 | if (eb->and < 0) { |
159 | if (handle >= -eb->and) |
167 | if (handle >= -eb->and) |
160 | return NULL; |
168 | return NULL; |
161 | return eb->lut[handle]; |
169 | return eb->lut[handle]; |
162 | } else { |
170 | } else { |
163 | struct hlist_head *head; |
171 | struct hlist_head *head; |
164 | struct hlist_node *node; |
172 | struct hlist_node *node; |
165 | 173 | ||
166 | head = &eb->buckets[handle & eb->and]; |
174 | head = &eb->buckets[handle & eb->and]; |
167 | hlist_for_each(node, head) { |
175 | hlist_for_each(node, head) { |
168 | struct drm_i915_gem_object *obj; |
176 | struct drm_i915_gem_object *obj; |
169 | 177 | ||
170 | obj = hlist_entry(node, struct drm_i915_gem_object, exec_node); |
178 | obj = hlist_entry(node, struct drm_i915_gem_object, exec_node); |
171 | if (obj->exec_handle == handle) |
179 | if (obj->exec_handle == handle) |
172 | return obj; |
180 | return obj; |
173 | } |
181 | } |
174 | return NULL; |
182 | return NULL; |
175 | } |
183 | } |
176 | } |
184 | } |
177 | 185 | ||
178 | static void |
186 | static void |
179 | eb_destroy(struct eb_objects *eb) |
187 | eb_destroy(struct eb_objects *eb) |
180 | { |
188 | { |
181 | while (!list_empty(&eb->objects)) { |
189 | while (!list_empty(&eb->objects)) { |
182 | struct drm_i915_gem_object *obj; |
190 | struct drm_i915_gem_object *obj; |
183 | 191 | ||
184 | obj = list_first_entry(&eb->objects, |
192 | obj = list_first_entry(&eb->objects, |
185 | struct drm_i915_gem_object, |
193 | struct drm_i915_gem_object, |
186 | exec_list); |
194 | exec_list); |
187 | list_del_init(&obj->exec_list); |
195 | list_del_init(&obj->exec_list); |
188 | drm_gem_object_unreference(&obj->base); |
196 | drm_gem_object_unreference(&obj->base); |
189 | } |
197 | } |
190 | kfree(eb); |
198 | kfree(eb); |
191 | } |
199 | } |
192 | 200 | ||
193 | static inline int use_cpu_reloc(struct drm_i915_gem_object *obj) |
201 | static inline int use_cpu_reloc(struct drm_i915_gem_object *obj) |
194 | { |
202 | { |
195 | return (obj->base.write_domain == I915_GEM_DOMAIN_CPU || |
203 | return (obj->base.write_domain == I915_GEM_DOMAIN_CPU || |
196 | !obj->map_and_fenceable || |
204 | !obj->map_and_fenceable || |
197 | obj->cache_level != I915_CACHE_NONE); |
205 | obj->cache_level != I915_CACHE_NONE); |
198 | } |
206 | } |
199 | 207 | ||
200 | static int |
208 | static int |
201 | i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj, |
209 | i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj, |
202 | struct eb_objects *eb, |
210 | struct eb_objects *eb, |
203 | struct drm_i915_gem_relocation_entry *reloc, |
211 | struct drm_i915_gem_relocation_entry *reloc, |
204 | struct i915_address_space *vm) |
212 | struct i915_address_space *vm) |
205 | { |
213 | { |
206 | struct drm_device *dev = obj->base.dev; |
214 | struct drm_device *dev = obj->base.dev; |
207 | struct drm_gem_object *target_obj; |
215 | struct drm_gem_object *target_obj; |
208 | struct drm_i915_gem_object *target_i915_obj; |
216 | struct drm_i915_gem_object *target_i915_obj; |
209 | uint32_t target_offset; |
217 | uint32_t target_offset; |
210 | int ret = -EINVAL; |
218 | int ret = -EINVAL; |
211 | 219 | ||
212 | /* we've already hold a reference to all valid objects */ |
220 | /* we've already hold a reference to all valid objects */ |
213 | target_obj = &eb_get_object(eb, reloc->target_handle)->base; |
221 | target_obj = &eb_get_object(eb, reloc->target_handle)->base; |
214 | if (unlikely(target_obj == NULL)) |
222 | if (unlikely(target_obj == NULL)) |
215 | return -ENOENT; |
223 | return -ENOENT; |
216 | 224 | ||
217 | target_i915_obj = to_intel_bo(target_obj); |
225 | target_i915_obj = to_intel_bo(target_obj); |
218 | target_offset = i915_gem_obj_ggtt_offset(target_i915_obj); |
226 | target_offset = i915_gem_obj_ggtt_offset(target_i915_obj); |
219 | 227 | ||
220 | /* Sandybridge PPGTT errata: We need a global gtt mapping for MI and |
228 | /* Sandybridge PPGTT errata: We need a global gtt mapping for MI and |
221 | * pipe_control writes because the gpu doesn't properly redirect them |
229 | * pipe_control writes because the gpu doesn't properly redirect them |
222 | * through the ppgtt for non_secure batchbuffers. */ |
230 | * through the ppgtt for non_secure batchbuffers. */ |
223 | if (unlikely(IS_GEN6(dev) && |
231 | if (unlikely(IS_GEN6(dev) && |
224 | reloc->write_domain == I915_GEM_DOMAIN_INSTRUCTION && |
232 | reloc->write_domain == I915_GEM_DOMAIN_INSTRUCTION && |
225 | !target_i915_obj->has_global_gtt_mapping)) { |
233 | !target_i915_obj->has_global_gtt_mapping)) { |
226 | i915_gem_gtt_bind_object(target_i915_obj, |
234 | i915_gem_gtt_bind_object(target_i915_obj, |
227 | target_i915_obj->cache_level); |
235 | target_i915_obj->cache_level); |
228 | } |
236 | } |
229 | 237 | ||
230 | /* Validate that the target is in a valid r/w GPU domain */ |
238 | /* Validate that the target is in a valid r/w GPU domain */ |
231 | if (unlikely(reloc->write_domain & (reloc->write_domain - 1))) { |
239 | if (unlikely(reloc->write_domain & (reloc->write_domain - 1))) { |
232 | DRM_DEBUG("reloc with multiple write domains: " |
240 | DRM_DEBUG("reloc with multiple write domains: " |
233 | "obj %p target %d offset %d " |
241 | "obj %p target %d offset %d " |
234 | "read %08x write %08x", |
242 | "read %08x write %08x", |
235 | obj, reloc->target_handle, |
243 | obj, reloc->target_handle, |
236 | (int) reloc->offset, |
244 | (int) reloc->offset, |
237 | reloc->read_domains, |
245 | reloc->read_domains, |
238 | reloc->write_domain); |
246 | reloc->write_domain); |
239 | return ret; |
247 | return ret; |
240 | } |
248 | } |
241 | if (unlikely((reloc->write_domain | reloc->read_domains) |
249 | if (unlikely((reloc->write_domain | reloc->read_domains) |
242 | & ~I915_GEM_GPU_DOMAINS)) { |
250 | & ~I915_GEM_GPU_DOMAINS)) { |
243 | DRM_DEBUG("reloc with read/write non-GPU domains: " |
251 | DRM_DEBUG("reloc with read/write non-GPU domains: " |
244 | "obj %p target %d offset %d " |
252 | "obj %p target %d offset %d " |
245 | "read %08x write %08x", |
253 | "read %08x write %08x", |
246 | obj, reloc->target_handle, |
254 | obj, reloc->target_handle, |
247 | (int) reloc->offset, |
255 | (int) reloc->offset, |
248 | reloc->read_domains, |
256 | reloc->read_domains, |
249 | reloc->write_domain); |
257 | reloc->write_domain); |
250 | return ret; |
258 | return ret; |
251 | } |
259 | } |
252 | 260 | ||
253 | target_obj->pending_read_domains |= reloc->read_domains; |
261 | target_obj->pending_read_domains |= reloc->read_domains; |
254 | target_obj->pending_write_domain |= reloc->write_domain; |
262 | target_obj->pending_write_domain |= reloc->write_domain; |
255 | 263 | ||
256 | /* If the relocation already has the right value in it, no |
264 | /* If the relocation already has the right value in it, no |
257 | * more work needs to be done. |
265 | * more work needs to be done. |
258 | */ |
266 | */ |
259 | if (target_offset == reloc->presumed_offset) |
267 | if (target_offset == reloc->presumed_offset) |
260 | return 0; |
268 | return 0; |
261 | 269 | ||
262 | /* Check that the relocation address is valid... */ |
270 | /* Check that the relocation address is valid... */ |
263 | if (unlikely(reloc->offset > obj->base.size - 4)) { |
271 | if (unlikely(reloc->offset > obj->base.size - 4)) { |
264 | DRM_DEBUG("Relocation beyond object bounds: " |
272 | DRM_DEBUG("Relocation beyond object bounds: " |
265 | "obj %p target %d offset %d size %d.\n", |
273 | "obj %p target %d offset %d size %d.\n", |
266 | obj, reloc->target_handle, |
274 | obj, reloc->target_handle, |
267 | (int) reloc->offset, |
275 | (int) reloc->offset, |
268 | (int) obj->base.size); |
276 | (int) obj->base.size); |
269 | return ret; |
277 | return ret; |
270 | } |
278 | } |
271 | if (unlikely(reloc->offset & 3)) { |
279 | if (unlikely(reloc->offset & 3)) { |
272 | DRM_DEBUG("Relocation not 4-byte aligned: " |
280 | DRM_DEBUG("Relocation not 4-byte aligned: " |
273 | "obj %p target %d offset %d.\n", |
281 | "obj %p target %d offset %d.\n", |
274 | obj, reloc->target_handle, |
282 | obj, reloc->target_handle, |
275 | (int) reloc->offset); |
283 | (int) reloc->offset); |
276 | return ret; |
284 | return ret; |
277 | } |
285 | } |
278 | 286 | ||
279 | /* We can't wait for rendering with pagefaults disabled */ |
287 | /* We can't wait for rendering with pagefaults disabled */ |
280 | 288 | ||
281 | reloc->delta += target_offset; |
289 | reloc->delta += target_offset; |
282 | if (use_cpu_reloc(obj)) { |
290 | if (use_cpu_reloc(obj)) { |
283 | uint32_t page_offset = reloc->offset & ~PAGE_MASK; |
291 | uint32_t page_offset = reloc->offset & ~PAGE_MASK; |
284 | char *vaddr; |
292 | char *vaddr; |
285 | 293 | ||
286 | ret = i915_gem_object_set_to_cpu_domain(obj, 1); |
294 | ret = i915_gem_object_set_to_cpu_domain(obj, 1); |
287 | if (ret) |
295 | if (ret) |
288 | return ret; |
296 | return ret; |
289 | 297 | ||
290 | vaddr = (char *)MapIoMem((addr_t)i915_gem_object_get_page(obj, |
298 | vaddr = (char *)MapIoMem((addr_t)i915_gem_object_get_page(obj, |
291 | reloc->offset >> PAGE_SHIFT), 4096, 3); |
299 | reloc->offset >> PAGE_SHIFT), 4096, 3); |
292 | *(uint32_t *)(vaddr + page_offset) = reloc->delta; |
300 | *(uint32_t *)(vaddr + page_offset) = reloc->delta; |
293 | FreeKernelSpace(vaddr); |
301 | FreeKernelSpace(vaddr); |
294 | } else { |
302 | } else { |
295 | struct drm_i915_private *dev_priv = dev->dev_private; |
303 | struct drm_i915_private *dev_priv = dev->dev_private; |
296 | uint32_t __iomem *reloc_entry; |
304 | uint32_t __iomem *reloc_entry; |
297 | void __iomem *reloc_page; |
305 | void __iomem *reloc_page; |
298 | 306 | ||
299 | ret = i915_gem_object_set_to_gtt_domain(obj, true); |
307 | ret = i915_gem_object_set_to_gtt_domain(obj, true); |
300 | if (ret) |
308 | if (ret) |
301 | return ret; |
309 | return ret; |
302 | 310 | ||
303 | ret = i915_gem_object_put_fence(obj); |
311 | ret = i915_gem_object_put_fence(obj); |
304 | if (ret) |
312 | if (ret) |
305 | return ret; |
313 | return ret; |
306 | 314 | ||
307 | /* Map the page containing the relocation we're going to perform. */ |
315 | /* Map the page containing the relocation we're going to perform. */ |
308 | reloc->offset += i915_gem_obj_ggtt_offset(obj); |
316 | reloc->offset += i915_gem_obj_ggtt_offset(obj); |
309 | reloc_page = (void*)MapIoMem(reloc->offset & PAGE_MASK, 4096, 3); |
317 | reloc_page = (void*)MapIoMem(reloc->offset & PAGE_MASK, 4096, 3); |
310 | reloc_entry = (uint32_t __iomem *) |
318 | reloc_entry = (uint32_t __iomem *) |
311 | (reloc_page + (reloc->offset & ~PAGE_MASK)); |
319 | (reloc_page + (reloc->offset & ~PAGE_MASK)); |
312 | iowrite32(reloc->delta, reloc_entry); |
320 | iowrite32(reloc->delta, reloc_entry); |
313 | FreeKernelSpace(reloc_page); |
321 | FreeKernelSpace(reloc_page); |
314 | } |
322 | } |
315 | 323 | ||
316 | /* and update the user's relocation entry */ |
324 | /* and update the user's relocation entry */ |
317 | reloc->presumed_offset = target_offset; |
325 | reloc->presumed_offset = target_offset; |
318 | 326 | ||
319 | return 0; |
327 | return 0; |
320 | } |
328 | } |
321 | 329 | ||
322 | static int |
330 | static int |
323 | i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj, |
331 | i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj, |
324 | struct eb_objects *eb, |
332 | struct eb_objects *eb, |
325 | struct i915_address_space *vm) |
333 | struct i915_address_space *vm) |
326 | { |
334 | { |
327 | #define N_RELOC(x) ((x) / sizeof(struct drm_i915_gem_relocation_entry)) |
335 | #define N_RELOC(x) ((x) / sizeof(struct drm_i915_gem_relocation_entry)) |
328 | struct drm_i915_gem_relocation_entry stack_reloc[N_RELOC(64)]; |
336 | struct drm_i915_gem_relocation_entry stack_reloc[N_RELOC(64)]; |
329 | struct drm_i915_gem_relocation_entry __user *user_relocs; |
337 | struct drm_i915_gem_relocation_entry __user *user_relocs; |
330 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
338 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
331 | int remain, ret; |
339 | int remain, ret; |
332 | 340 | ||
333 | user_relocs = (void __user *)(uintptr_t)entry->relocs_ptr; |
341 | user_relocs = (void __user *)(uintptr_t)entry->relocs_ptr; |
334 | 342 | ||
335 | remain = entry->relocation_count; |
343 | remain = entry->relocation_count; |
336 | while (remain) { |
344 | while (remain) { |
337 | struct drm_i915_gem_relocation_entry *r = stack_reloc; |
345 | struct drm_i915_gem_relocation_entry *r = stack_reloc; |
338 | int count = remain; |
346 | int count = remain; |
339 | if (count > ARRAY_SIZE(stack_reloc)) |
347 | if (count > ARRAY_SIZE(stack_reloc)) |
340 | count = ARRAY_SIZE(stack_reloc); |
348 | count = ARRAY_SIZE(stack_reloc); |
341 | remain -= count; |
349 | remain -= count; |
342 | 350 | ||
343 | memcpy(r, user_relocs, count*sizeof(r[0])); |
351 | memcpy(r, user_relocs, count*sizeof(r[0])); |
344 | 352 | ||
345 | do { |
353 | do { |
346 | u64 offset = r->presumed_offset; |
354 | u64 offset = r->presumed_offset; |
347 | 355 | ||
348 | ret = i915_gem_execbuffer_relocate_entry(obj, eb, r, |
356 | ret = i915_gem_execbuffer_relocate_entry(obj, eb, r, |
349 | vm); |
357 | vm); |
350 | if (ret) |
358 | if (ret) |
351 | return ret; |
359 | return ret; |
352 | 360 | ||
353 | memcpy(&user_relocs->presumed_offset, |
361 | memcpy(&user_relocs->presumed_offset, |
354 | &r->presumed_offset, |
362 | &r->presumed_offset, |
355 | sizeof(r->presumed_offset)); |
363 | sizeof(r->presumed_offset)); |
356 | 364 | ||
357 | user_relocs++; |
365 | user_relocs++; |
358 | r++; |
366 | r++; |
359 | } while (--count); |
367 | } while (--count); |
360 | } |
368 | } |
361 | 369 | ||
362 | return 0; |
370 | return 0; |
363 | #undef N_RELOC |
371 | #undef N_RELOC |
364 | } |
372 | } |
365 | 373 | ||
366 | static int |
374 | static int |
367 | i915_gem_execbuffer_relocate_object_slow(struct drm_i915_gem_object *obj, |
375 | i915_gem_execbuffer_relocate_object_slow(struct drm_i915_gem_object *obj, |
368 | struct eb_objects *eb, |
376 | struct eb_objects *eb, |
369 | struct drm_i915_gem_relocation_entry *relocs, |
377 | struct drm_i915_gem_relocation_entry *relocs, |
370 | struct i915_address_space *vm) |
378 | struct i915_address_space *vm) |
371 | { |
379 | { |
372 | const struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
380 | const struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
373 | int i, ret; |
381 | int i, ret; |
374 | 382 | ||
375 | for (i = 0; i < entry->relocation_count; i++) { |
383 | for (i = 0; i < entry->relocation_count; i++) { |
376 | ret = i915_gem_execbuffer_relocate_entry(obj, eb, &relocs[i], |
384 | ret = i915_gem_execbuffer_relocate_entry(obj, eb, &relocs[i], |
377 | vm); |
385 | vm); |
378 | if (ret) |
386 | if (ret) |
379 | return ret; |
387 | return ret; |
380 | } |
388 | } |
381 | 389 | ||
382 | return 0; |
390 | return 0; |
383 | } |
391 | } |
384 | 392 | ||
385 | static int |
393 | static int |
386 | i915_gem_execbuffer_relocate(struct eb_objects *eb, |
394 | i915_gem_execbuffer_relocate(struct eb_objects *eb, |
387 | struct i915_address_space *vm) |
395 | struct i915_address_space *vm) |
388 | { |
396 | { |
389 | struct drm_i915_gem_object *obj; |
397 | struct drm_i915_gem_object *obj; |
390 | int ret = 0; |
398 | int ret = 0; |
391 | 399 | ||
392 | /* This is the fast path and we cannot handle a pagefault whilst |
400 | /* This is the fast path and we cannot handle a pagefault whilst |
393 | * holding the struct mutex lest the user pass in the relocations |
401 | * holding the struct mutex lest the user pass in the relocations |
394 | * contained within a mmaped bo. For in such a case we, the page |
402 | * contained within a mmaped bo. For in such a case we, the page |
395 | * fault handler would call i915_gem_fault() and we would try to |
403 | * fault handler would call i915_gem_fault() and we would try to |
396 | * acquire the struct mutex again. Obviously this is bad and so |
404 | * acquire the struct mutex again. Obviously this is bad and so |
397 | * lockdep complains vehemently. |
405 | * lockdep complains vehemently. |
398 | */ |
406 | */ |
399 | // pagefault_disable(); |
407 | // pagefault_disable(); |
400 | list_for_each_entry(obj, &eb->objects, exec_list) { |
408 | list_for_each_entry(obj, &eb->objects, exec_list) { |
401 | ret = i915_gem_execbuffer_relocate_object(obj, eb, vm); |
409 | ret = i915_gem_execbuffer_relocate_object(obj, eb, vm); |
402 | if (ret) |
410 | if (ret) |
403 | break; |
411 | break; |
404 | } |
412 | } |
405 | // pagefault_enable(); |
413 | // pagefault_enable(); |
406 | 414 | ||
407 | return ret; |
415 | return ret; |
408 | } |
416 | } |
409 | 417 | ||
410 | #define __EXEC_OBJECT_HAS_PIN (1<<31) |
418 | #define __EXEC_OBJECT_HAS_PIN (1<<31) |
411 | #define __EXEC_OBJECT_HAS_FENCE (1<<30) |
419 | #define __EXEC_OBJECT_HAS_FENCE (1<<30) |
412 | 420 | ||
413 | static int |
421 | static int |
414 | need_reloc_mappable(struct drm_i915_gem_object *obj) |
422 | need_reloc_mappable(struct drm_i915_gem_object *obj) |
415 | { |
423 | { |
416 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
424 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
417 | return entry->relocation_count && !use_cpu_reloc(obj); |
425 | return entry->relocation_count && !use_cpu_reloc(obj); |
418 | } |
426 | } |
419 | 427 | ||
420 | static int |
428 | static int |
421 | i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, |
429 | i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, |
422 | struct intel_ring_buffer *ring, |
430 | struct intel_ring_buffer *ring, |
423 | struct i915_address_space *vm, |
431 | struct i915_address_space *vm, |
424 | bool *need_reloc) |
432 | bool *need_reloc) |
425 | { |
433 | { |
426 | struct drm_i915_private *dev_priv = obj->base.dev->dev_private; |
434 | struct drm_i915_private *dev_priv = obj->base.dev->dev_private; |
427 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
435 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
428 | bool has_fenced_gpu_access = INTEL_INFO(ring->dev)->gen < 4; |
436 | bool has_fenced_gpu_access = INTEL_INFO(ring->dev)->gen < 4; |
429 | bool need_fence, need_mappable; |
437 | bool need_fence, need_mappable; |
430 | int ret; |
438 | int ret; |
431 | 439 | ||
432 | need_fence = |
440 | need_fence = |
433 | has_fenced_gpu_access && |
441 | has_fenced_gpu_access && |
434 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
442 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
435 | obj->tiling_mode != I915_TILING_NONE; |
443 | obj->tiling_mode != I915_TILING_NONE; |
436 | need_mappable = need_fence || need_reloc_mappable(obj); |
444 | need_mappable = need_fence || need_reloc_mappable(obj); |
437 | 445 | ||
438 | ret = i915_gem_object_pin(obj, vm, entry->alignment, need_mappable, |
446 | ret = i915_gem_object_pin(obj, vm, entry->alignment, need_mappable, |
439 | false); |
447 | false); |
440 | if (ret) |
448 | if (ret) |
441 | return ret; |
449 | return ret; |
442 | 450 | ||
443 | entry->flags |= __EXEC_OBJECT_HAS_PIN; |
451 | entry->flags |= __EXEC_OBJECT_HAS_PIN; |
444 | 452 | ||
445 | if (has_fenced_gpu_access) { |
453 | if (has_fenced_gpu_access) { |
446 | if (entry->flags & EXEC_OBJECT_NEEDS_FENCE) { |
454 | if (entry->flags & EXEC_OBJECT_NEEDS_FENCE) { |
447 | ret = i915_gem_object_get_fence(obj); |
455 | ret = i915_gem_object_get_fence(obj); |
448 | if (ret) |
456 | if (ret) |
449 | return ret; |
457 | return ret; |
450 | 458 | ||
451 | if (i915_gem_object_pin_fence(obj)) |
459 | if (i915_gem_object_pin_fence(obj)) |
452 | entry->flags |= __EXEC_OBJECT_HAS_FENCE; |
460 | entry->flags |= __EXEC_OBJECT_HAS_FENCE; |
453 | 461 | ||
454 | obj->pending_fenced_gpu_access = true; |
462 | obj->pending_fenced_gpu_access = true; |
455 | } |
463 | } |
456 | } |
464 | } |
457 | 465 | ||
458 | /* Ensure ppgtt mapping exists if needed */ |
466 | /* Ensure ppgtt mapping exists if needed */ |
459 | if (dev_priv->mm.aliasing_ppgtt && !obj->has_aliasing_ppgtt_mapping) { |
467 | if (dev_priv->mm.aliasing_ppgtt && !obj->has_aliasing_ppgtt_mapping) { |
460 | i915_ppgtt_bind_object(dev_priv->mm.aliasing_ppgtt, |
468 | i915_ppgtt_bind_object(dev_priv->mm.aliasing_ppgtt, |
461 | obj, obj->cache_level); |
469 | obj, obj->cache_level); |
462 | 470 | ||
463 | obj->has_aliasing_ppgtt_mapping = 1; |
471 | obj->has_aliasing_ppgtt_mapping = 1; |
464 | } |
472 | } |
465 | 473 | ||
466 | if (entry->offset != i915_gem_obj_offset(obj, vm)) { |
474 | if (entry->offset != i915_gem_obj_offset(obj, vm)) { |
467 | entry->offset = i915_gem_obj_offset(obj, vm); |
475 | entry->offset = i915_gem_obj_offset(obj, vm); |
468 | *need_reloc = true; |
476 | *need_reloc = true; |
469 | } |
477 | } |
470 | 478 | ||
471 | if (entry->flags & EXEC_OBJECT_WRITE) { |
479 | if (entry->flags & EXEC_OBJECT_WRITE) { |
472 | obj->base.pending_read_domains = I915_GEM_DOMAIN_RENDER; |
480 | obj->base.pending_read_domains = I915_GEM_DOMAIN_RENDER; |
473 | obj->base.pending_write_domain = I915_GEM_DOMAIN_RENDER; |
481 | obj->base.pending_write_domain = I915_GEM_DOMAIN_RENDER; |
474 | } |
482 | } |
475 | 483 | ||
476 | if (entry->flags & EXEC_OBJECT_NEEDS_GTT && |
484 | if (entry->flags & EXEC_OBJECT_NEEDS_GTT && |
477 | !obj->has_global_gtt_mapping) |
485 | !obj->has_global_gtt_mapping) |
478 | i915_gem_gtt_bind_object(obj, obj->cache_level); |
486 | i915_gem_gtt_bind_object(obj, obj->cache_level); |
479 | 487 | ||
480 | return 0; |
488 | return 0; |
481 | } |
489 | } |
482 | 490 | ||
483 | static void |
491 | static void |
484 | i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj) |
492 | i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj) |
485 | { |
493 | { |
486 | struct drm_i915_gem_exec_object2 *entry; |
494 | struct drm_i915_gem_exec_object2 *entry; |
487 | 495 | ||
488 | if (!i915_gem_obj_bound_any(obj)) |
496 | if (!i915_gem_obj_bound_any(obj)) |
489 | return; |
497 | return; |
490 | 498 | ||
491 | entry = obj->exec_entry; |
499 | entry = obj->exec_entry; |
492 | 500 | ||
493 | if (entry->flags & __EXEC_OBJECT_HAS_FENCE) |
501 | if (entry->flags & __EXEC_OBJECT_HAS_FENCE) |
494 | i915_gem_object_unpin_fence(obj); |
502 | i915_gem_object_unpin_fence(obj); |
495 | 503 | ||
496 | if (entry->flags & __EXEC_OBJECT_HAS_PIN) |
504 | if (entry->flags & __EXEC_OBJECT_HAS_PIN) |
497 | i915_gem_object_unpin(obj); |
505 | i915_gem_object_unpin(obj); |
498 | 506 | ||
499 | entry->flags &= ~(__EXEC_OBJECT_HAS_FENCE | __EXEC_OBJECT_HAS_PIN); |
507 | entry->flags &= ~(__EXEC_OBJECT_HAS_FENCE | __EXEC_OBJECT_HAS_PIN); |
500 | } |
508 | } |
501 | 509 | ||
502 | static int |
510 | static int |
503 | i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring, |
511 | i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring, |
504 | struct list_head *objects, |
512 | struct list_head *objects, |
505 | struct i915_address_space *vm, |
513 | struct i915_address_space *vm, |
506 | bool *need_relocs) |
514 | bool *need_relocs) |
507 | { |
515 | { |
508 | struct drm_i915_gem_object *obj; |
516 | struct drm_i915_gem_object *obj; |
509 | struct list_head ordered_objects; |
517 | struct list_head ordered_objects; |
510 | bool has_fenced_gpu_access = INTEL_INFO(ring->dev)->gen < 4; |
518 | bool has_fenced_gpu_access = INTEL_INFO(ring->dev)->gen < 4; |
511 | int retry; |
519 | int retry; |
512 | 520 | ||
513 | INIT_LIST_HEAD(&ordered_objects); |
521 | INIT_LIST_HEAD(&ordered_objects); |
514 | while (!list_empty(objects)) { |
522 | while (!list_empty(objects)) { |
515 | struct drm_i915_gem_exec_object2 *entry; |
523 | struct drm_i915_gem_exec_object2 *entry; |
516 | bool need_fence, need_mappable; |
524 | bool need_fence, need_mappable; |
517 | 525 | ||
518 | obj = list_first_entry(objects, |
526 | obj = list_first_entry(objects, |
519 | struct drm_i915_gem_object, |
527 | struct drm_i915_gem_object, |
520 | exec_list); |
528 | exec_list); |
521 | entry = obj->exec_entry; |
529 | entry = obj->exec_entry; |
522 | 530 | ||
523 | need_fence = |
531 | need_fence = |
524 | has_fenced_gpu_access && |
532 | has_fenced_gpu_access && |
525 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
533 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
526 | obj->tiling_mode != I915_TILING_NONE; |
534 | obj->tiling_mode != I915_TILING_NONE; |
527 | need_mappable = need_fence || need_reloc_mappable(obj); |
535 | need_mappable = need_fence || need_reloc_mappable(obj); |
528 | 536 | ||
529 | if (need_mappable) |
537 | if (need_mappable) |
530 | list_move(&obj->exec_list, &ordered_objects); |
538 | list_move(&obj->exec_list, &ordered_objects); |
531 | else |
539 | else |
532 | list_move_tail(&obj->exec_list, &ordered_objects); |
540 | list_move_tail(&obj->exec_list, &ordered_objects); |
533 | 541 | ||
534 | obj->base.pending_read_domains = I915_GEM_GPU_DOMAINS & ~I915_GEM_DOMAIN_COMMAND; |
542 | obj->base.pending_read_domains = I915_GEM_GPU_DOMAINS & ~I915_GEM_DOMAIN_COMMAND; |
535 | obj->base.pending_write_domain = 0; |
543 | obj->base.pending_write_domain = 0; |
536 | obj->pending_fenced_gpu_access = false; |
544 | obj->pending_fenced_gpu_access = false; |
537 | } |
545 | } |
538 | list_splice(&ordered_objects, objects); |
546 | list_splice(&ordered_objects, objects); |
539 | 547 | ||
540 | /* Attempt to pin all of the buffers into the GTT. |
548 | /* Attempt to pin all of the buffers into the GTT. |
541 | * This is done in 3 phases: |
549 | * This is done in 3 phases: |
542 | * |
550 | * |
543 | * 1a. Unbind all objects that do not match the GTT constraints for |
551 | * 1a. Unbind all objects that do not match the GTT constraints for |
544 | * the execbuffer (fenceable, mappable, alignment etc). |
552 | * the execbuffer (fenceable, mappable, alignment etc). |
545 | * 1b. Increment pin count for already bound objects. |
553 | * 1b. Increment pin count for already bound objects. |
546 | * 2. Bind new objects. |
554 | * 2. Bind new objects. |
547 | * 3. Decrement pin count. |
555 | * 3. Decrement pin count. |
548 | * |
556 | * |
549 | * This avoid unnecessary unbinding of later objects in order to make |
557 | * This avoid unnecessary unbinding of later objects in order to make |
550 | * room for the earlier objects *unless* we need to defragment. |
558 | * room for the earlier objects *unless* we need to defragment. |
551 | */ |
559 | */ |
552 | retry = 0; |
560 | retry = 0; |
553 | do { |
561 | do { |
554 | int ret = 0; |
562 | int ret = 0; |
555 | 563 | ||
556 | /* Unbind any ill-fitting objects or pin. */ |
564 | /* Unbind any ill-fitting objects or pin. */ |
557 | list_for_each_entry(obj, objects, exec_list) { |
565 | list_for_each_entry(obj, objects, exec_list) { |
558 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
566 | struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; |
559 | bool need_fence, need_mappable; |
567 | bool need_fence, need_mappable; |
560 | u32 obj_offset; |
568 | u32 obj_offset; |
561 | 569 | ||
562 | if (!i915_gem_obj_bound(obj, vm)) |
570 | if (!i915_gem_obj_bound(obj, vm)) |
563 | continue; |
571 | continue; |
564 | 572 | ||
565 | obj_offset = i915_gem_obj_offset(obj, vm); |
573 | obj_offset = i915_gem_obj_offset(obj, vm); |
566 | need_fence = |
574 | need_fence = |
567 | has_fenced_gpu_access && |
575 | has_fenced_gpu_access && |
568 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
576 | entry->flags & EXEC_OBJECT_NEEDS_FENCE && |
569 | obj->tiling_mode != I915_TILING_NONE; |
577 | obj->tiling_mode != I915_TILING_NONE; |
570 | need_mappable = need_fence || need_reloc_mappable(obj); |
578 | need_mappable = need_fence || need_reloc_mappable(obj); |
- | 579 | ||
- | 580 | WARN_ON((need_mappable || need_fence) && |
|
571 | 581 | !i915_is_ggtt(vm)); |
|
572 | 582 | ||
573 | if ((entry->alignment && |
583 | if ((entry->alignment && |
574 | obj_offset & (entry->alignment - 1)) || |
584 | obj_offset & (entry->alignment - 1)) || |
575 | (need_mappable && !obj->map_and_fenceable)) |
585 | (need_mappable && !obj->map_and_fenceable)) |
576 | ret = i915_vma_unbind(i915_gem_obj_to_vma(obj, vm)); |
586 | ret = i915_vma_unbind(i915_gem_obj_to_vma(obj, vm)); |
577 | else |
587 | else |
578 | ret = i915_gem_execbuffer_reserve_object(obj, ring, vm, need_relocs); |
588 | ret = i915_gem_execbuffer_reserve_object(obj, ring, vm, need_relocs); |
579 | if (ret) |
589 | if (ret) |
580 | goto err; |
590 | goto err; |
581 | } |
591 | } |
582 | 592 | ||
583 | /* Bind fresh objects */ |
593 | /* Bind fresh objects */ |
584 | list_for_each_entry(obj, objects, exec_list) { |
594 | list_for_each_entry(obj, objects, exec_list) { |
585 | if (i915_gem_obj_bound(obj, vm)) |
595 | if (i915_gem_obj_bound(obj, vm)) |
586 | continue; |
596 | continue; |
587 | 597 | ||
588 | ret = i915_gem_execbuffer_reserve_object(obj, ring, vm, need_relocs); |
598 | ret = i915_gem_execbuffer_reserve_object(obj, ring, vm, need_relocs); |
589 | if (ret) |
599 | if (ret) |
590 | goto err; |
600 | goto err; |
591 | } |
601 | } |
592 | 602 | ||
593 | err: /* Decrement pin count for bound objects */ |
603 | err: /* Decrement pin count for bound objects */ |
594 | list_for_each_entry(obj, objects, exec_list) |
604 | list_for_each_entry(obj, objects, exec_list) |
595 | i915_gem_execbuffer_unreserve_object(obj); |
605 | i915_gem_execbuffer_unreserve_object(obj); |
596 | 606 | ||
597 | if (ret != -ENOSPC || retry++) |
607 | if (ret != -ENOSPC || retry++) |
598 | return ret; |
608 | return ret; |
599 | 609 | ||
600 | // ret = i915_gem_evict_everything(ring->dev); |
610 | // ret = i915_gem_evict_everything(ring->dev); |
601 | if (ret) |
611 | if (ret) |
602 | return ret; |
612 | return ret; |
603 | } while (1); |
613 | } while (1); |
604 | } |
614 | } |
605 | 615 | ||
606 | static int |
616 | static int |
607 | i915_gem_execbuffer_relocate_slow(struct drm_device *dev, |
617 | i915_gem_execbuffer_relocate_slow(struct drm_device *dev, |
608 | struct drm_i915_gem_execbuffer2 *args, |
618 | struct drm_i915_gem_execbuffer2 *args, |
609 | struct drm_file *file, |
619 | struct drm_file *file, |
610 | struct intel_ring_buffer *ring, |
620 | struct intel_ring_buffer *ring, |
611 | struct eb_objects *eb, |
621 | struct eb_objects *eb, |
612 | struct drm_i915_gem_exec_object2 *exec, |
622 | struct drm_i915_gem_exec_object2 *exec, |
613 | struct i915_address_space *vm) |
623 | struct i915_address_space *vm) |
614 | { |
624 | { |
615 | struct drm_i915_gem_relocation_entry *reloc; |
625 | struct drm_i915_gem_relocation_entry *reloc; |
616 | struct drm_i915_gem_object *obj; |
626 | struct drm_i915_gem_object *obj; |
617 | bool need_relocs; |
627 | bool need_relocs; |
618 | int *reloc_offset; |
628 | int *reloc_offset; |
619 | int i, total, ret; |
629 | int i, total, ret; |
620 | int count = args->buffer_count; |
630 | int count = args->buffer_count; |
621 | 631 | ||
622 | /* We may process another execbuffer during the unlock... */ |
632 | /* We may process another execbuffer during the unlock... */ |
623 | while (!list_empty(&eb->objects)) { |
633 | while (!list_empty(&eb->objects)) { |
624 | obj = list_first_entry(&eb->objects, |
634 | obj = list_first_entry(&eb->objects, |
625 | struct drm_i915_gem_object, |
635 | struct drm_i915_gem_object, |
626 | exec_list); |
636 | exec_list); |
627 | list_del_init(&obj->exec_list); |
637 | list_del_init(&obj->exec_list); |
628 | drm_gem_object_unreference(&obj->base); |
638 | drm_gem_object_unreference(&obj->base); |
629 | } |
639 | } |
630 | 640 | ||
631 | mutex_unlock(&dev->struct_mutex); |
641 | mutex_unlock(&dev->struct_mutex); |
632 | 642 | ||
633 | total = 0; |
643 | total = 0; |
634 | for (i = 0; i < count; i++) |
644 | for (i = 0; i < count; i++) |
635 | total += exec[i].relocation_count; |
645 | total += exec[i].relocation_count; |
636 | 646 | ||
637 | reloc_offset = malloc(count * sizeof(*reloc_offset)); |
647 | reloc_offset = malloc(count * sizeof(*reloc_offset)); |
638 | reloc = malloc(total * sizeof(*reloc)); |
648 | reloc = malloc(total * sizeof(*reloc)); |
639 | if (reloc == NULL || reloc_offset == NULL) { |
649 | if (reloc == NULL || reloc_offset == NULL) { |
640 | kfree(reloc); |
650 | kfree(reloc); |
641 | kfree(reloc_offset); |
651 | kfree(reloc_offset); |
642 | mutex_lock(&dev->struct_mutex); |
652 | mutex_lock(&dev->struct_mutex); |
643 | return -ENOMEM; |
653 | return -ENOMEM; |
644 | } |
654 | } |
645 | 655 | ||
646 | total = 0; |
656 | total = 0; |
647 | for (i = 0; i < count; i++) { |
657 | for (i = 0; i < count; i++) { |
648 | struct drm_i915_gem_relocation_entry __user *user_relocs; |
658 | struct drm_i915_gem_relocation_entry __user *user_relocs; |
649 | u64 invalid_offset = (u64)-1; |
659 | u64 invalid_offset = (u64)-1; |
650 | int j; |
660 | int j; |
651 | 661 | ||
652 | user_relocs = (void __user *)(uintptr_t)exec[i].relocs_ptr; |
662 | user_relocs = (void __user *)(uintptr_t)exec[i].relocs_ptr; |
653 | 663 | ||
654 | if (copy_from_user(reloc+total, user_relocs, |
664 | if (copy_from_user(reloc+total, user_relocs, |
655 | exec[i].relocation_count * sizeof(*reloc))) { |
665 | exec[i].relocation_count * sizeof(*reloc))) { |
656 | ret = -EFAULT; |
666 | ret = -EFAULT; |
657 | mutex_lock(&dev->struct_mutex); |
667 | mutex_lock(&dev->struct_mutex); |
658 | goto err; |
668 | goto err; |
659 | } |
669 | } |
660 | 670 | ||
661 | /* As we do not update the known relocation offsets after |
671 | /* As we do not update the known relocation offsets after |
662 | * relocating (due to the complexities in lock handling), |
672 | * relocating (due to the complexities in lock handling), |
663 | * we need to mark them as invalid now so that we force the |
673 | * we need to mark them as invalid now so that we force the |
664 | * relocation processing next time. Just in case the target |
674 | * relocation processing next time. Just in case the target |
665 | * object is evicted and then rebound into its old |
675 | * object is evicted and then rebound into its old |
666 | * presumed_offset before the next execbuffer - if that |
676 | * presumed_offset before the next execbuffer - if that |
667 | * happened we would make the mistake of assuming that the |
677 | * happened we would make the mistake of assuming that the |
668 | * relocations were valid. |
678 | * relocations were valid. |
669 | */ |
679 | */ |
670 | for (j = 0; j < exec[i].relocation_count; j++) { |
680 | for (j = 0; j < exec[i].relocation_count; j++) { |
671 | if (copy_to_user(&user_relocs[j].presumed_offset, |
681 | if (copy_to_user(&user_relocs[j].presumed_offset, |
672 | &invalid_offset, |
682 | &invalid_offset, |
673 | sizeof(invalid_offset))) { |
683 | sizeof(invalid_offset))) { |
674 | ret = -EFAULT; |
684 | ret = -EFAULT; |
675 | mutex_lock(&dev->struct_mutex); |
685 | mutex_lock(&dev->struct_mutex); |
676 | goto err; |
686 | goto err; |
677 | } |
687 | } |
678 | } |
688 | } |
679 | 689 | ||
680 | reloc_offset[i] = total; |
690 | reloc_offset[i] = total; |
681 | total += exec[i].relocation_count; |
691 | total += exec[i].relocation_count; |
682 | } |
692 | } |
683 | 693 | ||
684 | ret = i915_mutex_lock_interruptible(dev); |
694 | ret = i915_mutex_lock_interruptible(dev); |
685 | if (ret) { |
695 | if (ret) { |
686 | mutex_lock(&dev->struct_mutex); |
696 | mutex_lock(&dev->struct_mutex); |
687 | goto err; |
697 | goto err; |
688 | } |
698 | } |
689 | 699 | ||
690 | /* reacquire the objects */ |
700 | /* reacquire the objects */ |
691 | eb_reset(eb); |
701 | eb_reset(eb); |
692 | ret = eb_lookup_objects(eb, exec, args, file); |
702 | ret = eb_lookup_objects(eb, exec, args, file); |
693 | if (ret) |
703 | if (ret) |
694 | goto err; |
704 | goto err; |
695 | 705 | ||
696 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
706 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
697 | ret = i915_gem_execbuffer_reserve(ring, &eb->objects, vm, &need_relocs); |
707 | ret = i915_gem_execbuffer_reserve(ring, &eb->objects, vm, &need_relocs); |
698 | if (ret) |
708 | if (ret) |
699 | goto err; |
709 | goto err; |
700 | 710 | ||
701 | list_for_each_entry(obj, &eb->objects, exec_list) { |
711 | list_for_each_entry(obj, &eb->objects, exec_list) { |
702 | int offset = obj->exec_entry - exec; |
712 | int offset = obj->exec_entry - exec; |
703 | ret = i915_gem_execbuffer_relocate_object_slow(obj, eb, |
713 | ret = i915_gem_execbuffer_relocate_object_slow(obj, eb, |
704 | reloc + reloc_offset[offset], |
714 | reloc + reloc_offset[offset], |
705 | vm); |
715 | vm); |
706 | if (ret) |
716 | if (ret) |
707 | goto err; |
717 | goto err; |
708 | } |
718 | } |
709 | 719 | ||
710 | /* Leave the user relocations as are, this is the painfully slow path, |
720 | /* Leave the user relocations as are, this is the painfully slow path, |
711 | * and we want to avoid the complication of dropping the lock whilst |
721 | * and we want to avoid the complication of dropping the lock whilst |
712 | * having buffers reserved in the aperture and so causing spurious |
722 | * having buffers reserved in the aperture and so causing spurious |
713 | * ENOSPC for random operations. |
723 | * ENOSPC for random operations. |
714 | */ |
724 | */ |
715 | 725 | ||
716 | err: |
726 | err: |
717 | kfree(reloc); |
727 | kfree(reloc); |
718 | kfree(reloc_offset); |
728 | kfree(reloc_offset); |
719 | return ret; |
729 | return ret; |
720 | } |
730 | } |
721 | 731 | ||
722 | static int |
732 | static int |
723 | i915_gem_execbuffer_move_to_gpu(struct intel_ring_buffer *ring, |
733 | i915_gem_execbuffer_move_to_gpu(struct intel_ring_buffer *ring, |
724 | struct list_head *objects) |
734 | struct list_head *objects) |
725 | { |
735 | { |
726 | struct drm_i915_gem_object *obj; |
736 | struct drm_i915_gem_object *obj; |
727 | uint32_t flush_domains = 0; |
737 | uint32_t flush_domains = 0; |
728 | bool flush_chipset = false; |
738 | bool flush_chipset = false; |
729 | int ret; |
739 | int ret; |
730 | 740 | ||
731 | list_for_each_entry(obj, objects, exec_list) { |
741 | list_for_each_entry(obj, objects, exec_list) { |
732 | ret = i915_gem_object_sync(obj, ring); |
742 | ret = i915_gem_object_sync(obj, ring); |
733 | if (ret) |
743 | if (ret) |
734 | return ret; |
744 | return ret; |
735 | 745 | ||
736 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
746 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
737 | flush_chipset |= i915_gem_clflush_object(obj, false); |
747 | flush_chipset |= i915_gem_clflush_object(obj, false); |
738 | 748 | ||
739 | flush_domains |= obj->base.write_domain; |
749 | flush_domains |= obj->base.write_domain; |
740 | } |
750 | } |
741 | 751 | ||
742 | if (flush_chipset) |
752 | if (flush_chipset) |
743 | i915_gem_chipset_flush(ring->dev); |
753 | i915_gem_chipset_flush(ring->dev); |
744 | 754 | ||
745 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
755 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
746 | wmb(); |
756 | wmb(); |
747 | 757 | ||
748 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
758 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
749 | * any residual writes from the previous batch. |
759 | * any residual writes from the previous batch. |
750 | */ |
760 | */ |
751 | return intel_ring_invalidate_all_caches(ring); |
761 | return intel_ring_invalidate_all_caches(ring); |
752 | } |
762 | } |
753 | 763 | ||
754 | static bool |
764 | static bool |
755 | i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec) |
765 | i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec) |
756 | { |
766 | { |
757 | if (exec->flags & __I915_EXEC_UNKNOWN_FLAGS) |
767 | if (exec->flags & __I915_EXEC_UNKNOWN_FLAGS) |
758 | return false; |
768 | return false; |
759 | 769 | ||
760 | return ((exec->batch_start_offset | exec->batch_len) & 0x7) == 0; |
770 | return ((exec->batch_start_offset | exec->batch_len) & 0x7) == 0; |
761 | } |
771 | } |
762 | 772 | ||
763 | static int |
773 | static int |
764 | validate_exec_list(struct drm_i915_gem_exec_object2 *exec, |
774 | validate_exec_list(struct drm_i915_gem_exec_object2 *exec, |
765 | int count) |
775 | int count) |
766 | { |
776 | { |
767 | int i; |
777 | int i; |
768 | int relocs_total = 0; |
778 | int relocs_total = 0; |
769 | int relocs_max = INT_MAX / sizeof(struct drm_i915_gem_relocation_entry); |
779 | int relocs_max = INT_MAX / sizeof(struct drm_i915_gem_relocation_entry); |
770 | 780 | ||
771 | for (i = 0; i < count; i++) { |
781 | for (i = 0; i < count; i++) { |
772 | char __user *ptr = to_user_ptr(exec[i].relocs_ptr); |
782 | char __user *ptr = to_user_ptr(exec[i].relocs_ptr); |
773 | int length; /* limited by fault_in_pages_readable() */ |
783 | int length; /* limited by fault_in_pages_readable() */ |
774 | 784 | ||
775 | if (exec[i].flags & __EXEC_OBJECT_UNKNOWN_FLAGS) |
785 | if (exec[i].flags & __EXEC_OBJECT_UNKNOWN_FLAGS) |
776 | return -EINVAL; |
786 | return -EINVAL; |
777 | 787 | ||
778 | /* First check for malicious input causing overflow in |
788 | /* First check for malicious input causing overflow in |
779 | * the worst case where we need to allocate the entire |
789 | * the worst case where we need to allocate the entire |
780 | * relocation tree as a single array. |
790 | * relocation tree as a single array. |
781 | */ |
791 | */ |
782 | if (exec[i].relocation_count > relocs_max - relocs_total) |
792 | if (exec[i].relocation_count > relocs_max - relocs_total) |
783 | return -EINVAL; |
793 | return -EINVAL; |
784 | relocs_total += exec[i].relocation_count; |
794 | relocs_total += exec[i].relocation_count; |
785 | 795 | ||
786 | length = exec[i].relocation_count * |
796 | length = exec[i].relocation_count * |
787 | sizeof(struct drm_i915_gem_relocation_entry); |
797 | sizeof(struct drm_i915_gem_relocation_entry); |
788 | /* |
798 | /* |
789 | * We must check that the entire relocation array is safe |
799 | * We must check that the entire relocation array is safe |
790 | * to read, but since we may need to update the presumed |
800 | * to read, but since we may need to update the presumed |
791 | * offsets during execution, check for full write access. |
801 | * offsets during execution, check for full write access. |
792 | */ |
802 | */ |
793 | // if (!access_ok(VERIFY_WRITE, ptr, length)) |
803 | // if (!access_ok(VERIFY_WRITE, ptr, length)) |
794 | // return -EFAULT; |
804 | // return -EFAULT; |
795 | 805 | ||
796 | // if (fault_in_multipages_readable(ptr, length)) |
806 | // if (fault_in_multipages_readable(ptr, length)) |
797 | // return -EFAULT; |
807 | // return -EFAULT; |
798 | } |
808 | } |
799 | 809 | ||
800 | return 0; |
810 | return 0; |
801 | } |
811 | } |
802 | 812 | ||
803 | static void |
813 | static void |
804 | i915_gem_execbuffer_move_to_active(struct list_head *objects, |
814 | i915_gem_execbuffer_move_to_active(struct list_head *objects, |
805 | struct i915_address_space *vm, |
815 | struct i915_address_space *vm, |
806 | struct intel_ring_buffer *ring) |
816 | struct intel_ring_buffer *ring) |
807 | { |
817 | { |
808 | struct drm_i915_gem_object *obj; |
818 | struct drm_i915_gem_object *obj; |
809 | 819 | ||
810 | list_for_each_entry(obj, objects, exec_list) { |
820 | list_for_each_entry(obj, objects, exec_list) { |
811 | u32 old_read = obj->base.read_domains; |
821 | u32 old_read = obj->base.read_domains; |
812 | u32 old_write = obj->base.write_domain; |
822 | u32 old_write = obj->base.write_domain; |
813 | 823 | ||
814 | obj->base.write_domain = obj->base.pending_write_domain; |
824 | obj->base.write_domain = obj->base.pending_write_domain; |
815 | if (obj->base.write_domain == 0) |
825 | if (obj->base.write_domain == 0) |
816 | obj->base.pending_read_domains |= obj->base.read_domains; |
826 | obj->base.pending_read_domains |= obj->base.read_domains; |
817 | obj->base.read_domains = obj->base.pending_read_domains; |
827 | obj->base.read_domains = obj->base.pending_read_domains; |
818 | obj->fenced_gpu_access = obj->pending_fenced_gpu_access; |
828 | obj->fenced_gpu_access = obj->pending_fenced_gpu_access; |
819 | 829 | ||
820 | /* FIXME: This lookup gets fixed later <-- danvet */ |
830 | /* FIXME: This lookup gets fixed later <-- danvet */ |
821 | list_move_tail(&i915_gem_obj_to_vma(obj, vm)->mm_list, &vm->active_list); |
831 | list_move_tail(&i915_gem_obj_to_vma(obj, vm)->mm_list, &vm->active_list); |
822 | i915_gem_object_move_to_active(obj, ring); |
832 | i915_gem_object_move_to_active(obj, ring); |
823 | if (obj->base.write_domain) { |
833 | if (obj->base.write_domain) { |
824 | obj->dirty = 1; |
834 | obj->dirty = 1; |
825 | obj->last_write_seqno = intel_ring_get_seqno(ring); |
835 | obj->last_write_seqno = intel_ring_get_seqno(ring); |
826 | if (obj->pin_count) /* check for potential scanout */ |
836 | if (obj->pin_count) /* check for potential scanout */ |
827 | intel_mark_fb_busy(obj, ring); |
837 | intel_mark_fb_busy(obj, ring); |
828 | } |
838 | } |
829 | 839 | ||
830 | trace_i915_gem_object_change_domain(obj, old_read, old_write); |
840 | trace_i915_gem_object_change_domain(obj, old_read, old_write); |
831 | } |
841 | } |
832 | } |
842 | } |
833 | 843 | ||
834 | static void |
844 | static void |
835 | i915_gem_execbuffer_retire_commands(struct drm_device *dev, |
845 | i915_gem_execbuffer_retire_commands(struct drm_device *dev, |
836 | struct drm_file *file, |
846 | struct drm_file *file, |
837 | struct intel_ring_buffer *ring, |
847 | struct intel_ring_buffer *ring, |
838 | struct drm_i915_gem_object *obj) |
848 | struct drm_i915_gem_object *obj) |
839 | { |
849 | { |
840 | /* Unconditionally force add_request to emit a full flush. */ |
850 | /* Unconditionally force add_request to emit a full flush. */ |
841 | ring->gpu_caches_dirty = true; |
851 | ring->gpu_caches_dirty = true; |
842 | 852 | ||
843 | /* Add a breadcrumb for the completion of the batch buffer */ |
853 | /* Add a breadcrumb for the completion of the batch buffer */ |
844 | (void)__i915_add_request(ring, file, obj, NULL); |
854 | (void)__i915_add_request(ring, file, obj, NULL); |
845 | } |
855 | } |
846 | 856 | ||
847 | static int |
857 | static int |
848 | i915_reset_gen7_sol_offsets(struct drm_device *dev, |
858 | i915_reset_gen7_sol_offsets(struct drm_device *dev, |
849 | struct intel_ring_buffer *ring) |
859 | struct intel_ring_buffer *ring) |
850 | { |
860 | { |
851 | drm_i915_private_t *dev_priv = dev->dev_private; |
861 | drm_i915_private_t *dev_priv = dev->dev_private; |
852 | int ret, i; |
862 | int ret, i; |
853 | 863 | ||
854 | if (!IS_GEN7(dev) || ring != &dev_priv->ring[RCS]) |
864 | if (!IS_GEN7(dev) || ring != &dev_priv->ring[RCS]) |
855 | return 0; |
865 | return 0; |
856 | 866 | ||
857 | ret = intel_ring_begin(ring, 4 * 3); |
867 | ret = intel_ring_begin(ring, 4 * 3); |
858 | if (ret) |
868 | if (ret) |
859 | return ret; |
869 | return ret; |
860 | 870 | ||
861 | for (i = 0; i < 4; i++) { |
871 | for (i = 0; i < 4; i++) { |
862 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); |
872 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); |
863 | intel_ring_emit(ring, GEN7_SO_WRITE_OFFSET(i)); |
873 | intel_ring_emit(ring, GEN7_SO_WRITE_OFFSET(i)); |
864 | intel_ring_emit(ring, 0); |
874 | intel_ring_emit(ring, 0); |
865 | } |
875 | } |
866 | 876 | ||
867 | intel_ring_advance(ring); |
877 | intel_ring_advance(ring); |
868 | 878 | ||
869 | return 0; |
879 | return 0; |
870 | } |
880 | } |
871 | 881 | ||
872 | static int |
882 | static int |
873 | i915_gem_do_execbuffer(struct drm_device *dev, void *data, |
883 | i915_gem_do_execbuffer(struct drm_device *dev, void *data, |
874 | struct drm_file *file, |
884 | struct drm_file *file, |
875 | struct drm_i915_gem_execbuffer2 *args, |
885 | struct drm_i915_gem_execbuffer2 *args, |
876 | struct drm_i915_gem_exec_object2 *exec, |
886 | struct drm_i915_gem_exec_object2 *exec, |
877 | struct i915_address_space *vm) |
887 | struct i915_address_space *vm) |
878 | { |
888 | { |
879 | drm_i915_private_t *dev_priv = dev->dev_private; |
889 | drm_i915_private_t *dev_priv = dev->dev_private; |
880 | struct eb_objects *eb; |
890 | struct eb_objects *eb; |
881 | struct drm_i915_gem_object *batch_obj; |
891 | struct drm_i915_gem_object *batch_obj; |
882 | struct drm_clip_rect *cliprects = NULL; |
892 | struct drm_clip_rect *cliprects = NULL; |
883 | struct intel_ring_buffer *ring; |
893 | struct intel_ring_buffer *ring; |
884 | u32 ctx_id = i915_execbuffer2_get_context_id(*args); |
894 | u32 ctx_id = i915_execbuffer2_get_context_id(*args); |
885 | u32 exec_start, exec_len; |
895 | u32 exec_start, exec_len; |
886 | u32 mask, flags; |
896 | u32 mask, flags; |
887 | int ret, mode, i; |
897 | int ret, mode, i; |
888 | bool need_relocs; |
898 | bool need_relocs; |
889 | 899 | ||
890 | if (!i915_gem_check_execbuffer(args)) |
900 | if (!i915_gem_check_execbuffer(args)) |
891 | return -EINVAL; |
901 | return -EINVAL; |
892 | 902 | ||
893 | ret = validate_exec_list(exec, args->buffer_count); |
903 | ret = validate_exec_list(exec, args->buffer_count); |
894 | if (ret) |
904 | if (ret) |
895 | return ret; |
905 | return ret; |
896 | 906 | ||
897 | flags = 0; |
907 | flags = 0; |
898 | if (args->flags & I915_EXEC_SECURE) { |
908 | if (args->flags & I915_EXEC_SECURE) { |
899 | 909 | ||
900 | flags |= I915_DISPATCH_SECURE; |
910 | flags |= I915_DISPATCH_SECURE; |
901 | } |
911 | } |
902 | if (args->flags & I915_EXEC_IS_PINNED) |
912 | if (args->flags & I915_EXEC_IS_PINNED) |
903 | flags |= I915_DISPATCH_PINNED; |
913 | flags |= I915_DISPATCH_PINNED; |
904 | 914 | ||
905 | switch (args->flags & I915_EXEC_RING_MASK) { |
915 | switch (args->flags & I915_EXEC_RING_MASK) { |
906 | case I915_EXEC_DEFAULT: |
916 | case I915_EXEC_DEFAULT: |
907 | case I915_EXEC_RENDER: |
917 | case I915_EXEC_RENDER: |
908 | ring = &dev_priv->ring[RCS]; |
918 | ring = &dev_priv->ring[RCS]; |
909 | break; |
919 | break; |
910 | case I915_EXEC_BSD: |
920 | case I915_EXEC_BSD: |
911 | ring = &dev_priv->ring[VCS]; |
921 | ring = &dev_priv->ring[VCS]; |
912 | if (ctx_id != DEFAULT_CONTEXT_ID) { |
922 | if (ctx_id != DEFAULT_CONTEXT_ID) { |
913 | DRM_DEBUG("Ring %s doesn't support contexts\n", |
923 | DRM_DEBUG("Ring %s doesn't support contexts\n", |
914 | ring->name); |
924 | ring->name); |
915 | return -EPERM; |
925 | return -EPERM; |
916 | } |
926 | } |
917 | break; |
927 | break; |
918 | case I915_EXEC_BLT: |
928 | case I915_EXEC_BLT: |
919 | ring = &dev_priv->ring[BCS]; |
929 | ring = &dev_priv->ring[BCS]; |
920 | if (ctx_id != DEFAULT_CONTEXT_ID) { |
930 | if (ctx_id != DEFAULT_CONTEXT_ID) { |
921 | DRM_DEBUG("Ring %s doesn't support contexts\n", |
931 | DRM_DEBUG("Ring %s doesn't support contexts\n", |
922 | ring->name); |
932 | ring->name); |
923 | return -EPERM; |
933 | return -EPERM; |
924 | } |
934 | } |
925 | break; |
935 | break; |
926 | case I915_EXEC_VEBOX: |
936 | case I915_EXEC_VEBOX: |
927 | ring = &dev_priv->ring[VECS]; |
937 | ring = &dev_priv->ring[VECS]; |
928 | if (ctx_id != DEFAULT_CONTEXT_ID) { |
938 | if (ctx_id != DEFAULT_CONTEXT_ID) { |
929 | DRM_DEBUG("Ring %s doesn't support contexts\n", |
939 | DRM_DEBUG("Ring %s doesn't support contexts\n", |
930 | ring->name); |
940 | ring->name); |
931 | return -EPERM; |
941 | return -EPERM; |
932 | } |
942 | } |
933 | break; |
943 | break; |
934 | 944 | ||
935 | default: |
945 | default: |
936 | DRM_DEBUG("execbuf with unknown ring: %d\n", |
946 | DRM_DEBUG("execbuf with unknown ring: %d\n", |
937 | (int)(args->flags & I915_EXEC_RING_MASK)); |
947 | (int)(args->flags & I915_EXEC_RING_MASK)); |
938 | return -EINVAL; |
948 | return -EINVAL; |
939 | } |
949 | } |
940 | if (!intel_ring_initialized(ring)) { |
950 | if (!intel_ring_initialized(ring)) { |
941 | DRM_DEBUG("execbuf with invalid ring: %d\n", |
951 | DRM_DEBUG("execbuf with invalid ring: %d\n", |
942 | (int)(args->flags & I915_EXEC_RING_MASK)); |
952 | (int)(args->flags & I915_EXEC_RING_MASK)); |
943 | return -EINVAL; |
953 | return -EINVAL; |
944 | } |
954 | } |
945 | 955 | ||
946 | mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
956 | mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
947 | mask = I915_EXEC_CONSTANTS_MASK; |
957 | mask = I915_EXEC_CONSTANTS_MASK; |
948 | switch (mode) { |
958 | switch (mode) { |
949 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
959 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
950 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
960 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
951 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
961 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
952 | if (ring == &dev_priv->ring[RCS] && |
962 | if (ring == &dev_priv->ring[RCS] && |
953 | mode != dev_priv->relative_constants_mode) { |
963 | mode != dev_priv->relative_constants_mode) { |
954 | if (INTEL_INFO(dev)->gen < 4) |
964 | if (INTEL_INFO(dev)->gen < 4) |
955 | return -EINVAL; |
965 | return -EINVAL; |
956 | 966 | ||
957 | if (INTEL_INFO(dev)->gen > 5 && |
967 | if (INTEL_INFO(dev)->gen > 5 && |
958 | mode == I915_EXEC_CONSTANTS_REL_SURFACE) |
968 | mode == I915_EXEC_CONSTANTS_REL_SURFACE) |
959 | return -EINVAL; |
969 | return -EINVAL; |
960 | 970 | ||
961 | /* The HW changed the meaning on this bit on gen6 */ |
971 | /* The HW changed the meaning on this bit on gen6 */ |
962 | if (INTEL_INFO(dev)->gen >= 6) |
972 | if (INTEL_INFO(dev)->gen >= 6) |
963 | mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
973 | mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
964 | } |
974 | } |
965 | break; |
975 | break; |
966 | default: |
976 | default: |
967 | DRM_DEBUG("execbuf with unknown constants: %d\n", mode); |
977 | DRM_DEBUG("execbuf with unknown constants: %d\n", mode); |
968 | return -EINVAL; |
978 | return -EINVAL; |
969 | } |
979 | } |
970 | 980 | ||
971 | if (args->buffer_count < 1) { |
981 | if (args->buffer_count < 1) { |
972 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
982 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
973 | return -EINVAL; |
983 | return -EINVAL; |
974 | } |
984 | } |
975 | 985 | ||
976 | if (args->num_cliprects != 0) { |
986 | if (args->num_cliprects != 0) { |
977 | if (ring != &dev_priv->ring[RCS]) { |
987 | if (ring != &dev_priv->ring[RCS]) { |
978 | DRM_DEBUG("clip rectangles are only valid with the render ring\n"); |
988 | DRM_DEBUG("clip rectangles are only valid with the render ring\n"); |
979 | return -EINVAL; |
989 | return -EINVAL; |
980 | } |
990 | } |
981 | 991 | ||
982 | if (INTEL_INFO(dev)->gen >= 5) { |
992 | if (INTEL_INFO(dev)->gen >= 5) { |
983 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); |
993 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); |
984 | return -EINVAL; |
994 | return -EINVAL; |
985 | } |
995 | } |
986 | 996 | ||
987 | if (args->num_cliprects > UINT_MAX / sizeof(*cliprects)) { |
997 | if (args->num_cliprects > UINT_MAX / sizeof(*cliprects)) { |
988 | DRM_DEBUG("execbuf with %u cliprects\n", |
998 | DRM_DEBUG("execbuf with %u cliprects\n", |
989 | args->num_cliprects); |
999 | args->num_cliprects); |
990 | return -EINVAL; |
1000 | return -EINVAL; |
991 | } |
1001 | } |
992 | 1002 | ||
993 | cliprects = kmalloc(args->num_cliprects * sizeof(*cliprects), |
1003 | cliprects = kmalloc(args->num_cliprects * sizeof(*cliprects), |
994 | GFP_KERNEL); |
1004 | GFP_KERNEL); |
995 | if (cliprects == NULL) { |
1005 | if (cliprects == NULL) { |
996 | ret = -ENOMEM; |
1006 | ret = -ENOMEM; |
997 | goto pre_mutex_err; |
1007 | goto pre_mutex_err; |
998 | } |
1008 | } |
999 | 1009 | ||
1000 | if (copy_from_user(cliprects, |
1010 | if (copy_from_user(cliprects, |
1001 | to_user_ptr(args->cliprects_ptr), |
1011 | to_user_ptr(args->cliprects_ptr), |
1002 | sizeof(*cliprects)*args->num_cliprects)) { |
1012 | sizeof(*cliprects)*args->num_cliprects)) { |
1003 | ret = -EFAULT; |
1013 | ret = -EFAULT; |
1004 | goto pre_mutex_err; |
1014 | goto pre_mutex_err; |
1005 | } |
1015 | } |
1006 | } |
1016 | } |
1007 | 1017 | ||
1008 | ret = i915_mutex_lock_interruptible(dev); |
1018 | ret = i915_mutex_lock_interruptible(dev); |
1009 | if (ret) |
1019 | if (ret) |
1010 | goto pre_mutex_err; |
1020 | goto pre_mutex_err; |
1011 | 1021 | ||
1012 | if (dev_priv->ums.mm_suspended) { |
1022 | if (dev_priv->ums.mm_suspended) { |
1013 | mutex_unlock(&dev->struct_mutex); |
1023 | mutex_unlock(&dev->struct_mutex); |
1014 | ret = -EBUSY; |
1024 | ret = -EBUSY; |
1015 | goto pre_mutex_err; |
1025 | goto pre_mutex_err; |
1016 | } |
1026 | } |
1017 | 1027 | ||
1018 | eb = eb_create(args); |
1028 | eb = eb_create(args); |
1019 | if (eb == NULL) { |
1029 | if (eb == NULL) { |
1020 | mutex_unlock(&dev->struct_mutex); |
1030 | mutex_unlock(&dev->struct_mutex); |
1021 | ret = -ENOMEM; |
1031 | ret = -ENOMEM; |
1022 | goto pre_mutex_err; |
1032 | goto pre_mutex_err; |
1023 | } |
1033 | } |
1024 | 1034 | ||
1025 | /* Look up object handles */ |
1035 | /* Look up object handles */ |
1026 | ret = eb_lookup_objects(eb, exec, args, file); |
1036 | ret = eb_lookup_objects(eb, exec, args, file); |
1027 | if (ret) |
1037 | if (ret) |
1028 | goto err; |
1038 | goto err; |
1029 | 1039 | ||
1030 | /* take note of the batch buffer before we might reorder the lists */ |
1040 | /* take note of the batch buffer before we might reorder the lists */ |
1031 | batch_obj = list_entry(eb->objects.prev, |
1041 | batch_obj = list_entry(eb->objects.prev, |
1032 | struct drm_i915_gem_object, |
1042 | struct drm_i915_gem_object, |
1033 | exec_list); |
1043 | exec_list); |
1034 | 1044 | ||
1035 | /* Move the objects en-masse into the GTT, evicting if necessary. */ |
1045 | /* Move the objects en-masse into the GTT, evicting if necessary. */ |
1036 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
1046 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
1037 | ret = i915_gem_execbuffer_reserve(ring, &eb->objects, vm, &need_relocs); |
1047 | ret = i915_gem_execbuffer_reserve(ring, &eb->objects, vm, &need_relocs); |
1038 | if (ret) |
1048 | if (ret) |
1039 | goto err; |
1049 | goto err; |
1040 | 1050 | ||
1041 | /* The objects are in their final locations, apply the relocations. */ |
1051 | /* The objects are in their final locations, apply the relocations. */ |
1042 | if (need_relocs) |
1052 | if (need_relocs) |
1043 | ret = i915_gem_execbuffer_relocate(eb, vm); |
1053 | ret = i915_gem_execbuffer_relocate(eb, vm); |
1044 | if (ret) { |
1054 | if (ret) { |
1045 | if (ret == -EFAULT) { |
1055 | if (ret == -EFAULT) { |
1046 | ret = i915_gem_execbuffer_relocate_slow(dev, args, file, ring, |
1056 | ret = i915_gem_execbuffer_relocate_slow(dev, args, file, ring, |
1047 | eb, exec, vm); |
1057 | eb, exec, vm); |
1048 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
1058 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
1049 | } |
1059 | } |
1050 | if (ret) |
1060 | if (ret) |
1051 | goto err; |
1061 | goto err; |
1052 | } |
1062 | } |
1053 | 1063 | ||
1054 | /* Set the pending read domains for the batch buffer to COMMAND */ |
1064 | /* Set the pending read domains for the batch buffer to COMMAND */ |
1055 | if (batch_obj->base.pending_write_domain) { |
1065 | if (batch_obj->base.pending_write_domain) { |
1056 | DRM_DEBUG("Attempting to use self-modifying batch buffer\n"); |
1066 | DRM_DEBUG("Attempting to use self-modifying batch buffer\n"); |
1057 | ret = -EINVAL; |
1067 | ret = -EINVAL; |
1058 | goto err; |
1068 | goto err; |
1059 | } |
1069 | } |
1060 | batch_obj->base.pending_read_domains |= I915_GEM_DOMAIN_COMMAND; |
1070 | batch_obj->base.pending_read_domains |= I915_GEM_DOMAIN_COMMAND; |
1061 | 1071 | ||
1062 | /* snb/ivb/vlv conflate the "batch in ppgtt" bit with the "non-secure |
1072 | /* snb/ivb/vlv conflate the "batch in ppgtt" bit with the "non-secure |
1063 | * batch" bit. Hence we need to pin secure batches into the global gtt. |
1073 | * batch" bit. Hence we need to pin secure batches into the global gtt. |
1064 | * hsw should have this fixed, but let's be paranoid and do it |
1074 | * hsw should have this fixed, but let's be paranoid and do it |
1065 | * unconditionally for now. */ |
1075 | * unconditionally for now. */ |
1066 | if (flags & I915_DISPATCH_SECURE && !batch_obj->has_global_gtt_mapping) |
1076 | if (flags & I915_DISPATCH_SECURE && !batch_obj->has_global_gtt_mapping) |
1067 | i915_gem_gtt_bind_object(batch_obj, batch_obj->cache_level); |
1077 | i915_gem_gtt_bind_object(batch_obj, batch_obj->cache_level); |
1068 | 1078 | ||
1069 | ret = i915_gem_execbuffer_move_to_gpu(ring, &eb->objects); |
1079 | ret = i915_gem_execbuffer_move_to_gpu(ring, &eb->objects); |
1070 | if (ret) |
1080 | if (ret) |
1071 | goto err; |
1081 | goto err; |
1072 | 1082 | ||
1073 | ret = i915_switch_context(ring, file, ctx_id); |
1083 | ret = i915_switch_context(ring, file, ctx_id); |
1074 | if (ret) |
1084 | if (ret) |
1075 | goto err; |
1085 | goto err; |
1076 | 1086 | ||
1077 | if (ring == &dev_priv->ring[RCS] && |
1087 | if (ring == &dev_priv->ring[RCS] && |
1078 | mode != dev_priv->relative_constants_mode) { |
1088 | mode != dev_priv->relative_constants_mode) { |
1079 | ret = intel_ring_begin(ring, 4); |
1089 | ret = intel_ring_begin(ring, 4); |
1080 | if (ret) |
1090 | if (ret) |
1081 | goto err; |
1091 | goto err; |
1082 | 1092 | ||
1083 | intel_ring_emit(ring, MI_NOOP); |
1093 | intel_ring_emit(ring, MI_NOOP); |
1084 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); |
1094 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); |
1085 | intel_ring_emit(ring, INSTPM); |
1095 | intel_ring_emit(ring, INSTPM); |
1086 | intel_ring_emit(ring, mask << 16 | mode); |
1096 | intel_ring_emit(ring, mask << 16 | mode); |
1087 | intel_ring_advance(ring); |
1097 | intel_ring_advance(ring); |
1088 | 1098 | ||
1089 | dev_priv->relative_constants_mode = mode; |
1099 | dev_priv->relative_constants_mode = mode; |
1090 | } |
1100 | } |
1091 | 1101 | ||
1092 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
1102 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
1093 | ret = i915_reset_gen7_sol_offsets(dev, ring); |
1103 | ret = i915_reset_gen7_sol_offsets(dev, ring); |
1094 | if (ret) |
1104 | if (ret) |
1095 | goto err; |
1105 | goto err; |
1096 | } |
1106 | } |
1097 | 1107 | ||
1098 | exec_start = i915_gem_obj_offset(batch_obj, vm) + |
1108 | exec_start = i915_gem_obj_offset(batch_obj, vm) + |
1099 | args->batch_start_offset; |
1109 | args->batch_start_offset; |
1100 | exec_len = args->batch_len; |
1110 | exec_len = args->batch_len; |
1101 | if (cliprects) { |
1111 | if (cliprects) { |
- | 1112 | for (i = 0; i < args->num_cliprects; i++) { |
|
- | 1113 | ret = i915_emit_box(dev, &cliprects[i], |
|
- | 1114 | args->DR1, args->DR4); |
|
- | 1115 | if (ret) |
|
- | 1116 | goto err; |
|
- | 1117 | ||
- | 1118 | ret = ring->dispatch_execbuffer(ring, |
|
- | 1119 | exec_start, exec_len, |
|
- | 1120 | flags); |
|
- | 1121 | if (ret) |
|
- | 1122 | goto err; |
|
1102 | 1123 | } |
|
1103 | } else { |
1124 | } else { |
1104 | ret = ring->dispatch_execbuffer(ring, |
1125 | ret = ring->dispatch_execbuffer(ring, |
1105 | exec_start, exec_len, |
1126 | exec_start, exec_len, |
1106 | flags); |
1127 | flags); |
1107 | if (ret) |
1128 | if (ret) |
1108 | goto err; |
1129 | goto err; |
1109 | } |
1130 | } |
1110 | 1131 | ||
1111 | trace_i915_gem_ring_dispatch(ring, intel_ring_get_seqno(ring), flags); |
1132 | trace_i915_gem_ring_dispatch(ring, intel_ring_get_seqno(ring), flags); |
1112 | 1133 | ||
1113 | i915_gem_execbuffer_move_to_active(&eb->objects, vm, ring); |
1134 | i915_gem_execbuffer_move_to_active(&eb->objects, vm, ring); |
1114 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); |
1135 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); |
1115 | 1136 | ||
1116 | err: |
1137 | err: |
1117 | eb_destroy(eb); |
1138 | eb_destroy(eb); |
1118 | 1139 | ||
1119 | mutex_unlock(&dev->struct_mutex); |
1140 | mutex_unlock(&dev->struct_mutex); |
1120 | 1141 | ||
1121 | pre_mutex_err: |
1142 | pre_mutex_err: |
1122 | kfree(cliprects); |
1143 | kfree(cliprects); |
1123 | return ret; |
1144 | return ret; |
1124 | } |
1145 | } |
- | 1146 | ||
- | 1147 | #if 0 |
|
- | 1148 | /* |
|
- | 1149 | * Legacy execbuffer just creates an exec2 list from the original exec object |
|
- | 1150 | * list array and passes it to the real function. |
|
- | 1151 | */ |
|
- | 1152 | int |
|
- | 1153 | i915_gem_execbuffer(struct drm_device *dev, void *data, |
|
- | 1154 | struct drm_file *file) |
|
- | 1155 | { |
|
- | 1156 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
- | 1157 | struct drm_i915_gem_execbuffer *args = data; |
|
- | 1158 | struct drm_i915_gem_execbuffer2 exec2; |
|
- | 1159 | struct drm_i915_gem_exec_object *exec_list = NULL; |
|
- | 1160 | struct drm_i915_gem_exec_object2 *exec2_list = NULL; |
|
- | 1161 | int ret, i; |
|
- | 1162 | ||
- | 1163 | if (args->buffer_count < 1) { |
|
- | 1164 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
|
- | 1165 | return -EINVAL; |
|
- | 1166 | } |
|
- | 1167 | ||
- | 1168 | /* Copy in the exec list from userland */ |
|
- | 1169 | exec_list = drm_malloc_ab(sizeof(*exec_list), args->buffer_count); |
|
- | 1170 | exec2_list = drm_malloc_ab(sizeof(*exec2_list), args->buffer_count); |
|
- | 1171 | if (exec_list == NULL || exec2_list == NULL) { |
|
- | 1172 | DRM_DEBUG("Failed to allocate exec list for %d buffers\n", |
|
- | 1173 | args->buffer_count); |
|
- | 1174 | drm_free_large(exec_list); |
|
- | 1175 | drm_free_large(exec2_list); |
|
- | 1176 | return -ENOMEM; |
|
- | 1177 | } |
|
- | 1178 | ret = copy_from_user(exec_list, |
|
- | 1179 | to_user_ptr(args->buffers_ptr), |
|
- | 1180 | sizeof(*exec_list) * args->buffer_count); |
|
- | 1181 | if (ret != 0) { |
|
- | 1182 | DRM_DEBUG("copy %d exec entries failed %d\n", |
|
- | 1183 | args->buffer_count, ret); |
|
- | 1184 | drm_free_large(exec_list); |
|
- | 1185 | drm_free_large(exec2_list); |
|
- | 1186 | return -EFAULT; |
|
- | 1187 | } |
|
- | 1188 | ||
- | 1189 | for (i = 0; i < args->buffer_count; i++) { |
|
- | 1190 | exec2_list[i].handle = exec_list[i].handle; |
|
- | 1191 | exec2_list[i].relocation_count = exec_list[i].relocation_count; |
|
- | 1192 | exec2_list[i].relocs_ptr = exec_list[i].relocs_ptr; |
|
- | 1193 | exec2_list[i].alignment = exec_list[i].alignment; |
|
- | 1194 | exec2_list[i].offset = exec_list[i].offset; |
|
- | 1195 | if (INTEL_INFO(dev)->gen < 4) |
|
- | 1196 | exec2_list[i].flags = EXEC_OBJECT_NEEDS_FENCE; |
|
- | 1197 | else |
|
- | 1198 | exec2_list[i].flags = 0; |
|
- | 1199 | } |
|
- | 1200 | ||
- | 1201 | exec2.buffers_ptr = args->buffers_ptr; |
|
- | 1202 | exec2.buffer_count = args->buffer_count; |
|
- | 1203 | exec2.batch_start_offset = args->batch_start_offset; |
|
- | 1204 | exec2.batch_len = args->batch_len; |
|
- | 1205 | exec2.DR1 = args->DR1; |
|
- | 1206 | exec2.DR4 = args->DR4; |
|
- | 1207 | exec2.num_cliprects = args->num_cliprects; |
|
- | 1208 | exec2.cliprects_ptr = args->cliprects_ptr; |
|
- | 1209 | exec2.flags = I915_EXEC_RENDER; |
|
- | 1210 | i915_execbuffer2_set_context_id(exec2, 0); |
|
- | 1211 | ||
- | 1212 | ret = i915_gem_do_execbuffer(dev, data, file, &exec2, exec2_list, |
|
- | 1213 | &dev_priv->gtt.base); |
|
- | 1214 | if (!ret) { |
|
- | 1215 | /* Copy the new buffer offsets back to the user's exec list. */ |
|
- | 1216 | for (i = 0; i < args->buffer_count; i++) |
|
- | 1217 | exec_list[i].offset = exec2_list[i].offset; |
|
- | 1218 | /* ... and back out to userspace */ |
|
- | 1219 | ret = copy_to_user(to_user_ptr(args->buffers_ptr), |
|
- | 1220 | exec_list, |
|
- | 1221 | sizeof(*exec_list) * args->buffer_count); |
|
- | 1222 | if (ret) { |
|
- | 1223 | ret = -EFAULT; |
|
- | 1224 | DRM_DEBUG("failed to copy %d exec entries " |
|
- | 1225 | "back to user (%d)\n", |
|
- | 1226 | args->buffer_count, ret); |
|
- | 1227 | } |
|
- | 1228 | } |
|
- | 1229 | ||
- | 1230 | drm_free_large(exec_list); |
|
- | 1231 | drm_free_large(exec2_list); |
|
- | 1232 | return ret; |
|
1125 | 1233 | } |
|
1126 | 1234 | #endif |
|
1127 | 1235 | ||
1128 | int |
1236 | int |
1129 | i915_gem_execbuffer2(struct drm_device *dev, void *data, |
1237 | i915_gem_execbuffer2(struct drm_device *dev, void *data, |
1130 | struct drm_file *file) |
1238 | struct drm_file *file) |
1131 | { |
1239 | { |
1132 | struct drm_i915_private *dev_priv = dev->dev_private; |
1240 | struct drm_i915_private *dev_priv = dev->dev_private; |
1133 | struct drm_i915_gem_execbuffer2 *args = data; |
1241 | struct drm_i915_gem_execbuffer2 *args = data; |
1134 | struct drm_i915_gem_exec_object2 *exec2_list = NULL; |
1242 | struct drm_i915_gem_exec_object2 *exec2_list = NULL; |
1135 | int ret; |
1243 | int ret; |
1136 | 1244 | ||
1137 | if (args->buffer_count < 1 || |
1245 | if (args->buffer_count < 1 || |
1138 | args->buffer_count > UINT_MAX / sizeof(*exec2_list)) { |
1246 | args->buffer_count > UINT_MAX / sizeof(*exec2_list)) { |
1139 | DRM_DEBUG("execbuf2 with %d buffers\n", args->buffer_count); |
1247 | DRM_DEBUG("execbuf2 with %d buffers\n", args->buffer_count); |
1140 | return -EINVAL; |
1248 | return -EINVAL; |
1141 | } |
1249 | } |
1142 | 1250 | ||
1143 | exec2_list = kmalloc(sizeof(*exec2_list)*args->buffer_count, |
1251 | exec2_list = kmalloc(sizeof(*exec2_list)*args->buffer_count, |
1144 | GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY); |
1252 | GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY); |
1145 | if (exec2_list == NULL) { |
1253 | if (exec2_list == NULL) { |
1146 | DRM_DEBUG("Failed to allocate exec list for %d buffers\n", |
1254 | DRM_DEBUG("Failed to allocate exec list for %d buffers\n", |
1147 | args->buffer_count); |
1255 | args->buffer_count); |
1148 | return -ENOMEM; |
1256 | return -ENOMEM; |
1149 | } |
1257 | } |
1150 | ret = copy_from_user(exec2_list, |
1258 | ret = copy_from_user(exec2_list, |
1151 | (struct drm_i915_relocation_entry __user *) |
1259 | (struct drm_i915_relocation_entry __user *) |
1152 | (uintptr_t) args->buffers_ptr, |
1260 | (uintptr_t) args->buffers_ptr, |
1153 | sizeof(*exec2_list) * args->buffer_count); |
1261 | sizeof(*exec2_list) * args->buffer_count); |
1154 | if (ret != 0) { |
1262 | if (ret != 0) { |
1155 | DRM_DEBUG("copy %d exec entries failed %d\n", |
1263 | DRM_DEBUG("copy %d exec entries failed %d\n", |
1156 | args->buffer_count, ret); |
1264 | args->buffer_count, ret); |
1157 | kfree(exec2_list); |
1265 | kfree(exec2_list); |
1158 | FAIL(); |
1266 | FAIL(); |
1159 | return -EFAULT; |
1267 | return -EFAULT; |
1160 | } |
1268 | } |
1161 | 1269 | ||
1162 | ret = i915_gem_do_execbuffer(dev, data, file, args, exec2_list, |
1270 | ret = i915_gem_do_execbuffer(dev, data, file, args, exec2_list, |
1163 | &dev_priv->gtt.base); |
1271 | &dev_priv->gtt.base); |
1164 | if (!ret) { |
1272 | if (!ret) { |
1165 | /* Copy the new buffer offsets back to the user's exec list. */ |
1273 | /* Copy the new buffer offsets back to the user's exec list. */ |
1166 | ret = copy_to_user((void __user *)(uintptr_t)args->buffers_ptr, |
1274 | ret = copy_to_user((void __user *)(uintptr_t)args->buffers_ptr, |
1167 | exec2_list, |
1275 | exec2_list, |
1168 | sizeof(*exec2_list) * args->buffer_count); |
1276 | sizeof(*exec2_list) * args->buffer_count); |
1169 | if (ret) { |
1277 | if (ret) { |
1170 | ret = -EFAULT; |
1278 | ret = -EFAULT; |
1171 | DRM_DEBUG("failed to copy %d exec entries " |
1279 | DRM_DEBUG("failed to copy %d exec entries " |
1172 | "back to user (%d)\n", |
1280 | "back to user (%d)\n", |
1173 | args->buffer_count, ret); |
1281 | args->buffer_count, ret); |
1174 | } |
1282 | } |
1175 | } |
1283 | } |
1176 | 1284 | ||
1177 | kfree(exec2_list); |
1285 | kfree(exec2_list); |
1178 | - | ||
1179 | // LEAVE(); |
- | |
1180 | - | ||
1181 | return ret; |
1286 | return ret; |
1182 | }>><>>>>-->>>>>>>30) |
1287 | }>>>>>>><>>>>-->>>>>>>30) |
1183 | 1288 | ||
1184 | static><30) |
1289 | static><30) |
1185 | 1290 | ||
1186 | static>31) |
1291 | static>31) |
1187 | #define><31) |
1292 | #define><31) |
1188 | #define>>>>>0) |
1293 | #define>>>>>0) |
1189 | 1294 | ||
1190 | #define><0) |
1295 | #define><0) |
1191 | 1296 | ||
1192 | #define>10) |
1297 | #define>10) |
1193 | #define><10) |
1298 | #define><10) |
1194 | #define>9) |
1299 | #define>9) |
1195 | #define><9) |
1300 | #define><9) |
1196 | #define> |
1301 | #define> |