Commit | Line | Data |
---|---|---|
d38ceaf9 AD |
1 | /* |
2 | * Copyright 2015 Advanced Micro Devices, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
5 | * Permission is hereby granted, free of charge, to any person obtaining a | |
6 | * copy of this software and associated documentation files (the | |
7 | * "Software"), to deal in the Software without restriction, including | |
8 | * without limitation the rights to use, copy, modify, merge, publish, | |
9 | * distribute, sub license, and/or sell copies of the Software, and to | |
10 | * permit persons to whom the Software is furnished to do so, subject to | |
11 | * the following conditions: | |
12 | * | |
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL | |
16 | * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, | |
17 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR | |
18 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE | |
19 | * USE OR OTHER DEALINGS IN THE SOFTWARE. | |
20 | * | |
21 | * The above copyright notice and this permission notice (including the | |
22 | * next paragraph) shall be included in all copies or substantial portions | |
23 | * of the Software. | |
24 | * | |
25 | */ | |
26 | /* | |
27 | * Authors: | |
28 | * Christian König <deathsimple@vodafone.de> | |
29 | */ | |
30 | ||
31 | #include <drm/drmP.h> | |
32 | #include "amdgpu.h" | |
ec74407a | 33 | #include "amdgpu_trace.h" |
d38ceaf9 | 34 | |
636ce25c CK |
35 | #define AMDGPU_BO_LIST_MAX_PRIORITY 32u |
36 | #define AMDGPU_BO_LIST_NUM_BUCKETS (AMDGPU_BO_LIST_MAX_PRIORITY + 1) | |
37 | ||
d38ceaf9 AD |
38 | static int amdgpu_bo_list_create(struct amdgpu_fpriv *fpriv, |
39 | struct amdgpu_bo_list **result, | |
40 | int *id) | |
41 | { | |
42 | int r; | |
43 | ||
44 | *result = kzalloc(sizeof(struct amdgpu_bo_list), GFP_KERNEL); | |
45 | if (!*result) | |
46 | return -ENOMEM; | |
47 | ||
48 | mutex_lock(&fpriv->bo_list_lock); | |
49 | r = idr_alloc(&fpriv->bo_list_handles, *result, | |
decee87a | 50 | 1, 0, GFP_KERNEL); |
d38ceaf9 AD |
51 | if (r < 0) { |
52 | mutex_unlock(&fpriv->bo_list_lock); | |
53 | kfree(*result); | |
54 | return r; | |
55 | } | |
56 | *id = r; | |
57 | ||
58 | mutex_init(&(*result)->lock); | |
59 | (*result)->num_entries = 0; | |
60 | (*result)->array = NULL; | |
61 | ||
62 | mutex_lock(&(*result)->lock); | |
63 | mutex_unlock(&fpriv->bo_list_lock); | |
64 | ||
65 | return 0; | |
66 | } | |
67 | ||
68 | static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id) | |
69 | { | |
70 | struct amdgpu_bo_list *list; | |
71 | ||
72 | mutex_lock(&fpriv->bo_list_lock); | |
73 | list = idr_find(&fpriv->bo_list_handles, id); | |
74 | if (list) { | |
75 | mutex_lock(&list->lock); | |
76 | idr_remove(&fpriv->bo_list_handles, id); | |
77 | mutex_unlock(&list->lock); | |
78 | amdgpu_bo_list_free(list); | |
79 | } | |
80 | mutex_unlock(&fpriv->bo_list_lock); | |
81 | } | |
82 | ||
83 | static int amdgpu_bo_list_set(struct amdgpu_device *adev, | |
84 | struct drm_file *filp, | |
85 | struct amdgpu_bo_list *list, | |
86 | struct drm_amdgpu_bo_list_entry *info, | |
87 | unsigned num_entries) | |
88 | { | |
89 | struct amdgpu_bo_list_entry *array; | |
90 | struct amdgpu_bo *gds_obj = adev->gds.gds_gfx_bo; | |
91 | struct amdgpu_bo *gws_obj = adev->gds.gws_gfx_bo; | |
92 | struct amdgpu_bo *oa_obj = adev->gds.oa_gfx_bo; | |
93 | ||
94 | bool has_userptr = false; | |
95 | unsigned i; | |
cc325d19 | 96 | int r; |
d38ceaf9 AD |
97 | |
98 | array = drm_malloc_ab(num_entries, sizeof(struct amdgpu_bo_list_entry)); | |
99 | if (!array) | |
100 | return -ENOMEM; | |
101 | memset(array, 0, num_entries * sizeof(struct amdgpu_bo_list_entry)); | |
102 | ||
103 | for (i = 0; i < num_entries; ++i) { | |
104 | struct amdgpu_bo_list_entry *entry = &array[i]; | |
105 | struct drm_gem_object *gobj; | |
cc325d19 | 106 | struct mm_struct *usermm; |
d38ceaf9 AD |
107 | |
108 | gobj = drm_gem_object_lookup(adev->ddev, filp, info[i].bo_handle); | |
cc325d19 CK |
109 | if (!gobj) { |
110 | r = -ENOENT; | |
d38ceaf9 | 111 | goto error_free; |
cc325d19 | 112 | } |
d38ceaf9 AD |
113 | |
114 | entry->robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); | |
115 | drm_gem_object_unreference_unlocked(gobj); | |
636ce25c CK |
116 | entry->priority = min(info[i].bo_priority, |
117 | AMDGPU_BO_LIST_MAX_PRIORITY); | |
cc325d19 CK |
118 | usermm = amdgpu_ttm_tt_get_usermm(entry->robj->tbo.ttm); |
119 | if (usermm) { | |
120 | if (usermm != current->mm) { | |
121 | r = -EPERM; | |
122 | goto error_free; | |
123 | } | |
d38ceaf9 | 124 | has_userptr = true; |
cc325d19 | 125 | } |
d38ceaf9 AD |
126 | entry->tv.bo = &entry->robj->tbo; |
127 | entry->tv.shared = true; | |
128 | ||
1ea863fd | 129 | if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GDS) |
d38ceaf9 | 130 | gds_obj = entry->robj; |
1ea863fd | 131 | if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GWS) |
d38ceaf9 | 132 | gws_obj = entry->robj; |
1ea863fd | 133 | if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_OA) |
d38ceaf9 | 134 | oa_obj = entry->robj; |
ec74407a CK |
135 | |
136 | trace_amdgpu_bo_list_set(list, entry->robj); | |
d38ceaf9 AD |
137 | } |
138 | ||
139 | for (i = 0; i < list->num_entries; ++i) | |
140 | amdgpu_bo_unref(&list->array[i].robj); | |
141 | ||
142 | drm_free_large(list->array); | |
143 | ||
144 | list->gds_obj = gds_obj; | |
145 | list->gws_obj = gws_obj; | |
146 | list->oa_obj = oa_obj; | |
147 | list->has_userptr = has_userptr; | |
148 | list->array = array; | |
149 | list->num_entries = num_entries; | |
150 | ||
151 | return 0; | |
152 | ||
153 | error_free: | |
154 | drm_free_large(array); | |
cc325d19 | 155 | return r; |
d38ceaf9 AD |
156 | } |
157 | ||
158 | struct amdgpu_bo_list * | |
159 | amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id) | |
160 | { | |
161 | struct amdgpu_bo_list *result; | |
162 | ||
163 | mutex_lock(&fpriv->bo_list_lock); | |
164 | result = idr_find(&fpriv->bo_list_handles, id); | |
165 | if (result) | |
166 | mutex_lock(&result->lock); | |
167 | mutex_unlock(&fpriv->bo_list_lock); | |
168 | return result; | |
169 | } | |
170 | ||
636ce25c CK |
171 | void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list, |
172 | struct list_head *validated) | |
173 | { | |
174 | /* This is based on the bucket sort with O(n) time complexity. | |
175 | * An item with priority "i" is added to bucket[i]. The lists are then | |
176 | * concatenated in descending order. | |
177 | */ | |
178 | struct list_head bucket[AMDGPU_BO_LIST_NUM_BUCKETS]; | |
179 | unsigned i; | |
180 | ||
181 | for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++) | |
182 | INIT_LIST_HEAD(&bucket[i]); | |
183 | ||
184 | /* Since buffers which appear sooner in the relocation list are | |
185 | * likely to be used more often than buffers which appear later | |
186 | * in the list, the sort mustn't change the ordering of buffers | |
187 | * with the same priority, i.e. it must be stable. | |
188 | */ | |
189 | for (i = 0; i < list->num_entries; i++) { | |
190 | unsigned priority = list->array[i].priority; | |
191 | ||
192 | list_add_tail(&list->array[i].tv.head, | |
193 | &bucket[priority]); | |
194 | } | |
195 | ||
196 | /* Connect the sorted buckets in the output list. */ | |
197 | for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++) | |
198 | list_splice(&bucket[i], validated); | |
199 | } | |
200 | ||
d38ceaf9 AD |
201 | void amdgpu_bo_list_put(struct amdgpu_bo_list *list) |
202 | { | |
203 | mutex_unlock(&list->lock); | |
204 | } | |
205 | ||
206 | void amdgpu_bo_list_free(struct amdgpu_bo_list *list) | |
207 | { | |
208 | unsigned i; | |
209 | ||
210 | for (i = 0; i < list->num_entries; ++i) | |
211 | amdgpu_bo_unref(&list->array[i].robj); | |
212 | ||
213 | mutex_destroy(&list->lock); | |
214 | drm_free_large(list->array); | |
215 | kfree(list); | |
216 | } | |
217 | ||
218 | int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data, | |
219 | struct drm_file *filp) | |
220 | { | |
221 | const uint32_t info_size = sizeof(struct drm_amdgpu_bo_list_entry); | |
222 | ||
223 | struct amdgpu_device *adev = dev->dev_private; | |
224 | struct amdgpu_fpriv *fpriv = filp->driver_priv; | |
225 | union drm_amdgpu_bo_list *args = data; | |
226 | uint32_t handle = args->in.list_handle; | |
227 | const void __user *uptr = (const void*)(long)args->in.bo_info_ptr; | |
228 | ||
229 | struct drm_amdgpu_bo_list_entry *info; | |
230 | struct amdgpu_bo_list *list; | |
231 | ||
232 | int r; | |
233 | ||
234 | info = drm_malloc_ab(args->in.bo_number, | |
235 | sizeof(struct drm_amdgpu_bo_list_entry)); | |
236 | if (!info) | |
237 | return -ENOMEM; | |
238 | ||
239 | /* copy the handle array from userspace to a kernel buffer */ | |
240 | r = -EFAULT; | |
241 | if (likely(info_size == args->in.bo_info_size)) { | |
242 | unsigned long bytes = args->in.bo_number * | |
243 | args->in.bo_info_size; | |
244 | ||
245 | if (copy_from_user(info, uptr, bytes)) | |
246 | goto error_free; | |
247 | ||
248 | } else { | |
249 | unsigned long bytes = min(args->in.bo_info_size, info_size); | |
250 | unsigned i; | |
251 | ||
252 | memset(info, 0, args->in.bo_number * info_size); | |
253 | for (i = 0; i < args->in.bo_number; ++i) { | |
254 | if (copy_from_user(&info[i], uptr, bytes)) | |
255 | goto error_free; | |
256 | ||
257 | uptr += args->in.bo_info_size; | |
258 | } | |
259 | } | |
260 | ||
261 | switch (args->in.operation) { | |
262 | case AMDGPU_BO_LIST_OP_CREATE: | |
263 | r = amdgpu_bo_list_create(fpriv, &list, &handle); | |
264 | if (r) | |
265 | goto error_free; | |
266 | ||
267 | r = amdgpu_bo_list_set(adev, filp, list, info, | |
268 | args->in.bo_number); | |
269 | amdgpu_bo_list_put(list); | |
270 | if (r) | |
271 | goto error_free; | |
272 | ||
273 | break; | |
274 | ||
275 | case AMDGPU_BO_LIST_OP_DESTROY: | |
276 | amdgpu_bo_list_destroy(fpriv, handle); | |
277 | handle = 0; | |
278 | break; | |
279 | ||
280 | case AMDGPU_BO_LIST_OP_UPDATE: | |
281 | r = -ENOENT; | |
282 | list = amdgpu_bo_list_get(fpriv, handle); | |
283 | if (!list) | |
284 | goto error_free; | |
285 | ||
286 | r = amdgpu_bo_list_set(adev, filp, list, info, | |
287 | args->in.bo_number); | |
288 | amdgpu_bo_list_put(list); | |
289 | if (r) | |
290 | goto error_free; | |
291 | ||
292 | break; | |
293 | ||
294 | default: | |
295 | r = -EINVAL; | |
296 | goto error_free; | |
297 | } | |
298 | ||
299 | memset(args, 0, sizeof(*args)); | |
300 | args->out.list_handle = handle; | |
301 | drm_free_large(info); | |
302 | ||
303 | return 0; | |
304 | ||
305 | error_free: | |
306 | drm_free_large(info); | |
307 | return r; | |
308 | } |