Commit | Line | Data |
---|---|---|
eb8ffbfe RD |
1 | /* |
2 | * Copyright (c) 2005 Topspin Communications. All rights reserved. | |
3 | * Copyright (c) 2005 Cisco Systems. All rights reserved. | |
2a1d9b7f | 4 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. |
eb8ffbfe RD |
5 | * |
6 | * This software is available to you under a choice of one of two | |
7 | * licenses. You may choose to be licensed under the terms of the GNU | |
8 | * General Public License (GPL) Version 2, available from the file | |
9 | * COPYING in the main directory of this source tree, or the | |
10 | * OpenIB.org BSD license below: | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or | |
13 | * without modification, are permitted provided that the following | |
14 | * conditions are met: | |
15 | * | |
16 | * - Redistributions of source code must retain the above | |
17 | * copyright notice, this list of conditions and the following | |
18 | * disclaimer. | |
19 | * | |
20 | * - Redistributions in binary form must reproduce the above | |
21 | * copyright notice, this list of conditions and the following | |
22 | * disclaimer in the documentation and/or other materials | |
23 | * provided with the distribution. | |
24 | * | |
25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
26 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
27 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
28 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
29 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
30 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
31 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
32 | * SOFTWARE. | |
33 | * | |
34 | * $Id: uverbs_mem.c 2743 2005-06-28 22:27:59Z roland $ | |
35 | */ | |
36 | ||
37 | #include <linux/mm.h> | |
38 | #include <linux/dma-mapping.h> | |
39 | ||
40 | #include "uverbs.h" | |
41 | ||
eb8ffbfe RD |
42 | static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty) |
43 | { | |
44 | struct ib_umem_chunk *chunk, *tmp; | |
45 | int i; | |
46 | ||
47 | list_for_each_entry_safe(chunk, tmp, &umem->chunk_list, list) { | |
1527106f RC |
48 | ib_dma_unmap_sg(dev, chunk->page_list, |
49 | chunk->nents, DMA_BIDIRECTIONAL); | |
eb8ffbfe RD |
50 | for (i = 0; i < chunk->nents; ++i) { |
51 | if (umem->writable && dirty) | |
52 | set_page_dirty_lock(chunk->page_list[i].page); | |
53 | put_page(chunk->page_list[i].page); | |
54 | } | |
55 | ||
56 | kfree(chunk); | |
57 | } | |
58 | } | |
59 | ||
f7c6a7b5 RD |
60 | /** |
61 | * ib_umem_get - Pin and DMA map userspace memory. | |
62 | * @context: userspace context to pin memory for | |
63 | * @addr: userspace virtual address to start at | |
64 | * @size: length of region to pin | |
65 | * @access: IB_ACCESS_xxx flags for memory being pinned | |
66 | */ | |
67 | struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, | |
68 | size_t size, int access) | |
eb8ffbfe | 69 | { |
f7c6a7b5 | 70 | struct ib_umem *umem; |
eb8ffbfe RD |
71 | struct page **page_list; |
72 | struct ib_umem_chunk *chunk; | |
73 | unsigned long locked; | |
74 | unsigned long lock_limit; | |
75 | unsigned long cur_base; | |
76 | unsigned long npages; | |
f7c6a7b5 | 77 | int ret; |
eb8ffbfe RD |
78 | int off; |
79 | int i; | |
80 | ||
81 | if (!can_do_mlock()) | |
f7c6a7b5 | 82 | return ERR_PTR(-EPERM); |
eb8ffbfe | 83 | |
f7c6a7b5 RD |
84 | umem = kmalloc(sizeof *umem, GFP_KERNEL); |
85 | if (!umem) | |
86 | return ERR_PTR(-ENOMEM); | |
eb8ffbfe | 87 | |
f7c6a7b5 RD |
88 | umem->context = context; |
89 | umem->length = size; | |
90 | umem->offset = addr & ~PAGE_MASK; | |
91 | umem->page_size = PAGE_SIZE; | |
92 | /* | |
93 | * We ask for writable memory if any access flags other than | |
94 | * "remote read" are set. "Local write" and "remote write" | |
95 | * obviously require write access. "Remote atomic" can do | |
96 | * things like fetch and add, which will modify memory, and | |
97 | * "MW bind" can change permissions by binding a window. | |
98 | */ | |
99 | umem->writable = !!(access & ~IB_ACCESS_REMOTE_READ); | |
eb8ffbfe | 100 | |
f7c6a7b5 RD |
101 | INIT_LIST_HEAD(&umem->chunk_list); |
102 | ||
103 | page_list = (struct page **) __get_free_page(GFP_KERNEL); | |
104 | if (!page_list) { | |
105 | kfree(umem); | |
106 | return ERR_PTR(-ENOMEM); | |
107 | } | |
eb8ffbfe | 108 | |
f7c6a7b5 | 109 | npages = PAGE_ALIGN(size + umem->offset) >> PAGE_SHIFT; |
eb8ffbfe RD |
110 | |
111 | down_write(¤t->mm->mmap_sem); | |
112 | ||
113 | locked = npages + current->mm->locked_vm; | |
114 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; | |
115 | ||
116 | if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { | |
117 | ret = -ENOMEM; | |
118 | goto out; | |
119 | } | |
120 | ||
f7c6a7b5 | 121 | cur_base = addr & PAGE_MASK; |
eb8ffbfe RD |
122 | |
123 | while (npages) { | |
124 | ret = get_user_pages(current, current->mm, cur_base, | |
125 | min_t(int, npages, | |
126 | PAGE_SIZE / sizeof (struct page *)), | |
f7c6a7b5 | 127 | 1, !umem->writable, page_list, NULL); |
eb8ffbfe RD |
128 | |
129 | if (ret < 0) | |
130 | goto out; | |
131 | ||
132 | cur_base += ret * PAGE_SIZE; | |
133 | npages -= ret; | |
134 | ||
135 | off = 0; | |
136 | ||
137 | while (ret) { | |
138 | chunk = kmalloc(sizeof *chunk + sizeof (struct scatterlist) * | |
139 | min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK), | |
140 | GFP_KERNEL); | |
141 | if (!chunk) { | |
142 | ret = -ENOMEM; | |
143 | goto out; | |
144 | } | |
145 | ||
146 | chunk->nents = min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK); | |
147 | for (i = 0; i < chunk->nents; ++i) { | |
148 | chunk->page_list[i].page = page_list[i + off]; | |
149 | chunk->page_list[i].offset = 0; | |
150 | chunk->page_list[i].length = PAGE_SIZE; | |
151 | } | |
152 | ||
f7c6a7b5 | 153 | chunk->nmap = ib_dma_map_sg(context->device, |
1527106f RC |
154 | &chunk->page_list[0], |
155 | chunk->nents, | |
156 | DMA_BIDIRECTIONAL); | |
eb8ffbfe RD |
157 | if (chunk->nmap <= 0) { |
158 | for (i = 0; i < chunk->nents; ++i) | |
159 | put_page(chunk->page_list[i].page); | |
160 | kfree(chunk); | |
161 | ||
162 | ret = -ENOMEM; | |
163 | goto out; | |
164 | } | |
165 | ||
166 | ret -= chunk->nents; | |
167 | off += chunk->nents; | |
f7c6a7b5 | 168 | list_add_tail(&chunk->list, &umem->chunk_list); |
eb8ffbfe RD |
169 | } |
170 | ||
171 | ret = 0; | |
172 | } | |
173 | ||
174 | out: | |
f7c6a7b5 RD |
175 | if (ret < 0) { |
176 | __ib_umem_release(context->device, umem, 0); | |
177 | kfree(umem); | |
178 | } else | |
eb8ffbfe RD |
179 | current->mm->locked_vm = locked; |
180 | ||
181 | up_write(¤t->mm->mmap_sem); | |
182 | free_page((unsigned long) page_list); | |
183 | ||
f7c6a7b5 | 184 | return ret < 0 ? ERR_PTR(ret) : umem; |
eb8ffbfe | 185 | } |
f7c6a7b5 | 186 | EXPORT_SYMBOL(ib_umem_get); |
eb8ffbfe | 187 | |
1bf66a30 | 188 | static void ib_umem_account(struct work_struct *work) |
eb8ffbfe | 189 | { |
1bf66a30 RD |
190 | struct ib_umem *umem = container_of(work, struct ib_umem, work); |
191 | ||
192 | down_write(&umem->mm->mmap_sem); | |
193 | umem->mm->locked_vm -= umem->diff; | |
194 | up_write(&umem->mm->mmap_sem); | |
195 | mmput(umem->mm); | |
196 | kfree(umem); | |
eb8ffbfe RD |
197 | } |
198 | ||
f7c6a7b5 RD |
199 | /** |
200 | * ib_umem_release - release memory pinned with ib_umem_get | |
201 | * @umem: umem struct to release | |
202 | */ | |
203 | void ib_umem_release(struct ib_umem *umem) | |
eb8ffbfe | 204 | { |
f7c6a7b5 | 205 | struct ib_ucontext *context = umem->context; |
eb8ffbfe | 206 | struct mm_struct *mm; |
f7c6a7b5 | 207 | unsigned long diff; |
eb8ffbfe | 208 | |
f7c6a7b5 | 209 | __ib_umem_release(umem->context->device, umem, 1); |
eb8ffbfe RD |
210 | |
211 | mm = get_task_mm(current); | |
212 | if (!mm) | |
213 | return; | |
214 | ||
f7c6a7b5 | 215 | diff = PAGE_ALIGN(umem->length + umem->offset) >> PAGE_SHIFT; |
f7c6a7b5 | 216 | |
eb8ffbfe RD |
217 | /* |
218 | * We may be called with the mm's mmap_sem already held. This | |
219 | * can happen when a userspace munmap() is the call that drops | |
220 | * the last reference to our file and calls our release | |
221 | * method. If there are memory regions to destroy, we'll end | |
f7c6a7b5 RD |
222 | * up here and not be able to take the mmap_sem. In that case |
223 | * we defer the vm_locked accounting to the system workqueue. | |
eb8ffbfe | 224 | */ |
f7c6a7b5 | 225 | if (context->closing && !down_write_trylock(&mm->mmap_sem)) { |
1bf66a30 RD |
226 | INIT_WORK(&umem->work, ib_umem_account); |
227 | umem->mm = mm; | |
228 | umem->diff = diff; | |
eb8ffbfe | 229 | |
1bf66a30 | 230 | schedule_work(&umem->work); |
eb8ffbfe | 231 | return; |
f7c6a7b5 RD |
232 | } else |
233 | down_write(&mm->mmap_sem); | |
234 | ||
235 | current->mm->locked_vm -= diff; | |
236 | up_write(&mm->mmap_sem); | |
237 | mmput(mm); | |
1bf66a30 | 238 | kfree(umem); |
f7c6a7b5 RD |
239 | } |
240 | EXPORT_SYMBOL(ib_umem_release); | |
241 | ||
242 | int ib_umem_page_count(struct ib_umem *umem) | |
243 | { | |
244 | struct ib_umem_chunk *chunk; | |
245 | int shift; | |
246 | int i; | |
247 | int n; | |
248 | ||
249 | shift = ilog2(umem->page_size); | |
eb8ffbfe | 250 | |
f7c6a7b5 RD |
251 | n = 0; |
252 | list_for_each_entry(chunk, &umem->chunk_list, list) | |
253 | for (i = 0; i < chunk->nmap; ++i) | |
254 | n += sg_dma_len(&chunk->page_list[i]) >> shift; | |
eb8ffbfe | 255 | |
f7c6a7b5 | 256 | return n; |
eb8ffbfe | 257 | } |
f7c6a7b5 | 258 | EXPORT_SYMBOL(ib_umem_page_count); |