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. | |
eb8ffbfe RD |
33 | */ |
34 | ||
35 | #include <linux/mm.h> | |
36 | #include <linux/dma-mapping.h> | |
e8edc6e0 | 37 | #include <linux/sched.h> |
b108d976 | 38 | #include <linux/export.h> |
c8d8beea | 39 | #include <linux/hugetlb.h> |
cb9fbc5c | 40 | #include <linux/dma-attrs.h> |
5a0e3ad6 | 41 | #include <linux/slab.h> |
eb8ffbfe RD |
42 | |
43 | #include "uverbs.h" | |
44 | ||
92ddc447 | 45 | |
eb8ffbfe RD |
46 | static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty) |
47 | { | |
eeb8461e YH |
48 | struct scatterlist *sg; |
49 | struct page *page; | |
eb8ffbfe RD |
50 | int i; |
51 | ||
eeb8461e YH |
52 | if (umem->nmap > 0) |
53 | ib_dma_unmap_sg(dev, umem->sg_head.sgl, | |
54 | umem->nmap, | |
55 | DMA_BIDIRECTIONAL); | |
45711f1a | 56 | |
eeb8461e | 57 | for_each_sg(umem->sg_head.sgl, sg, umem->npages, i) { |
eb8ffbfe | 58 | |
eeb8461e YH |
59 | page = sg_page(sg); |
60 | if (umem->writable && dirty) | |
61 | set_page_dirty_lock(page); | |
62 | put_page(page); | |
eb8ffbfe | 63 | } |
eeb8461e YH |
64 | |
65 | sg_free_table(&umem->sg_head); | |
66 | return; | |
67 | ||
eb8ffbfe RD |
68 | } |
69 | ||
f7c6a7b5 RD |
70 | /** |
71 | * ib_umem_get - Pin and DMA map userspace memory. | |
72 | * @context: userspace context to pin memory for | |
73 | * @addr: userspace virtual address to start at | |
74 | * @size: length of region to pin | |
75 | * @access: IB_ACCESS_xxx flags for memory being pinned | |
cb9fbc5c | 76 | * @dmasync: flush in-flight DMA when the memory region is written |
f7c6a7b5 RD |
77 | */ |
78 | struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, | |
cb9fbc5c | 79 | size_t size, int access, int dmasync) |
eb8ffbfe | 80 | { |
f7c6a7b5 | 81 | struct ib_umem *umem; |
eb8ffbfe | 82 | struct page **page_list; |
c8d8beea | 83 | struct vm_area_struct **vma_list; |
eb8ffbfe RD |
84 | unsigned long locked; |
85 | unsigned long lock_limit; | |
86 | unsigned long cur_base; | |
87 | unsigned long npages; | |
f7c6a7b5 | 88 | int ret; |
eb8ffbfe | 89 | int i; |
cb9fbc5c | 90 | DEFINE_DMA_ATTRS(attrs); |
eeb8461e YH |
91 | struct scatterlist *sg, *sg_list_start; |
92 | int need_release = 0; | |
cb9fbc5c AK |
93 | |
94 | if (dmasync) | |
95 | dma_set_attr(DMA_ATTR_WRITE_BARRIER, &attrs); | |
eb8ffbfe RD |
96 | |
97 | if (!can_do_mlock()) | |
f7c6a7b5 | 98 | return ERR_PTR(-EPERM); |
eb8ffbfe | 99 | |
eeb8461e | 100 | umem = kzalloc(sizeof *umem, GFP_KERNEL); |
f7c6a7b5 RD |
101 | if (!umem) |
102 | return ERR_PTR(-ENOMEM); | |
eb8ffbfe | 103 | |
f7c6a7b5 RD |
104 | umem->context = context; |
105 | umem->length = size; | |
106 | umem->offset = addr & ~PAGE_MASK; | |
107 | umem->page_size = PAGE_SIZE; | |
108 | /* | |
109 | * We ask for writable memory if any access flags other than | |
110 | * "remote read" are set. "Local write" and "remote write" | |
111 | * obviously require write access. "Remote atomic" can do | |
112 | * things like fetch and add, which will modify memory, and | |
113 | * "MW bind" can change permissions by binding a window. | |
114 | */ | |
115 | umem->writable = !!(access & ~IB_ACCESS_REMOTE_READ); | |
eb8ffbfe | 116 | |
c8d8beea JF |
117 | /* We assume the memory is from hugetlb until proved otherwise */ |
118 | umem->hugetlb = 1; | |
119 | ||
f7c6a7b5 RD |
120 | page_list = (struct page **) __get_free_page(GFP_KERNEL); |
121 | if (!page_list) { | |
122 | kfree(umem); | |
123 | return ERR_PTR(-ENOMEM); | |
124 | } | |
eb8ffbfe | 125 | |
c8d8beea JF |
126 | /* |
127 | * if we can't alloc the vma_list, it's not so bad; | |
128 | * just assume the memory is not hugetlb memory | |
129 | */ | |
130 | vma_list = (struct vm_area_struct **) __get_free_page(GFP_KERNEL); | |
131 | if (!vma_list) | |
132 | umem->hugetlb = 0; | |
133 | ||
f7c6a7b5 | 134 | npages = PAGE_ALIGN(size + umem->offset) >> PAGE_SHIFT; |
eb8ffbfe RD |
135 | |
136 | down_write(¤t->mm->mmap_sem); | |
137 | ||
bc3e53f6 | 138 | locked = npages + current->mm->pinned_vm; |
ccbe9f0b | 139 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; |
eb8ffbfe RD |
140 | |
141 | if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { | |
142 | ret = -ENOMEM; | |
143 | goto out; | |
144 | } | |
145 | ||
f7c6a7b5 | 146 | cur_base = addr & PAGE_MASK; |
eb8ffbfe | 147 | |
eeb8461e YH |
148 | if (npages == 0) { |
149 | ret = -EINVAL; | |
150 | goto out; | |
151 | } | |
152 | ||
153 | ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL); | |
154 | if (ret) | |
155 | goto out; | |
156 | ||
157 | need_release = 1; | |
158 | sg_list_start = umem->sg_head.sgl; | |
159 | ||
eb8ffbfe RD |
160 | while (npages) { |
161 | ret = get_user_pages(current, current->mm, cur_base, | |
8079ffa0 | 162 | min_t(unsigned long, npages, |
eb8ffbfe | 163 | PAGE_SIZE / sizeof (struct page *)), |
c8d8beea | 164 | 1, !umem->writable, page_list, vma_list); |
eb8ffbfe RD |
165 | |
166 | if (ret < 0) | |
167 | goto out; | |
168 | ||
eeb8461e | 169 | umem->npages += ret; |
eb8ffbfe RD |
170 | cur_base += ret * PAGE_SIZE; |
171 | npages -= ret; | |
172 | ||
eeb8461e YH |
173 | for_each_sg(sg_list_start, sg, ret, i) { |
174 | if (vma_list && !is_vm_hugetlb_page(vma_list[i])) | |
175 | umem->hugetlb = 0; | |
176 | ||
177 | sg_set_page(sg, page_list[i], PAGE_SIZE, 0); | |
eb8ffbfe RD |
178 | } |
179 | ||
eeb8461e YH |
180 | /* preparing for next loop */ |
181 | sg_list_start = sg; | |
eb8ffbfe RD |
182 | } |
183 | ||
eeb8461e YH |
184 | umem->nmap = ib_dma_map_sg_attrs(context->device, |
185 | umem->sg_head.sgl, | |
186 | umem->npages, | |
187 | DMA_BIDIRECTIONAL, | |
188 | &attrs); | |
189 | ||
190 | if (umem->nmap <= 0) { | |
191 | ret = -ENOMEM; | |
192 | goto out; | |
193 | } | |
194 | ||
195 | ret = 0; | |
196 | ||
eb8ffbfe | 197 | out: |
f7c6a7b5 | 198 | if (ret < 0) { |
eeb8461e YH |
199 | if (need_release) |
200 | __ib_umem_release(context->device, umem, 0); | |
f7c6a7b5 RD |
201 | kfree(umem); |
202 | } else | |
bc3e53f6 | 203 | current->mm->pinned_vm = locked; |
eb8ffbfe RD |
204 | |
205 | up_write(¤t->mm->mmap_sem); | |
c8d8beea JF |
206 | if (vma_list) |
207 | free_page((unsigned long) vma_list); | |
eb8ffbfe RD |
208 | free_page((unsigned long) page_list); |
209 | ||
f7c6a7b5 | 210 | return ret < 0 ? ERR_PTR(ret) : umem; |
eb8ffbfe | 211 | } |
f7c6a7b5 | 212 | EXPORT_SYMBOL(ib_umem_get); |
eb8ffbfe | 213 | |
1bf66a30 | 214 | static void ib_umem_account(struct work_struct *work) |
eb8ffbfe | 215 | { |
1bf66a30 RD |
216 | struct ib_umem *umem = container_of(work, struct ib_umem, work); |
217 | ||
218 | down_write(&umem->mm->mmap_sem); | |
bc3e53f6 | 219 | umem->mm->pinned_vm -= umem->diff; |
1bf66a30 RD |
220 | up_write(&umem->mm->mmap_sem); |
221 | mmput(umem->mm); | |
222 | kfree(umem); | |
eb8ffbfe RD |
223 | } |
224 | ||
f7c6a7b5 RD |
225 | /** |
226 | * ib_umem_release - release memory pinned with ib_umem_get | |
227 | * @umem: umem struct to release | |
228 | */ | |
229 | void ib_umem_release(struct ib_umem *umem) | |
eb8ffbfe | 230 | { |
f7c6a7b5 | 231 | struct ib_ucontext *context = umem->context; |
eb8ffbfe | 232 | struct mm_struct *mm; |
f7c6a7b5 | 233 | unsigned long diff; |
eb8ffbfe | 234 | |
f7c6a7b5 | 235 | __ib_umem_release(umem->context->device, umem, 1); |
eb8ffbfe RD |
236 | |
237 | mm = get_task_mm(current); | |
7b82cd8e EC |
238 | if (!mm) { |
239 | kfree(umem); | |
eb8ffbfe | 240 | return; |
7b82cd8e | 241 | } |
eb8ffbfe | 242 | |
f7c6a7b5 | 243 | diff = PAGE_ALIGN(umem->length + umem->offset) >> PAGE_SHIFT; |
f7c6a7b5 | 244 | |
eb8ffbfe RD |
245 | /* |
246 | * We may be called with the mm's mmap_sem already held. This | |
247 | * can happen when a userspace munmap() is the call that drops | |
248 | * the last reference to our file and calls our release | |
249 | * method. If there are memory regions to destroy, we'll end | |
f7c6a7b5 RD |
250 | * up here and not be able to take the mmap_sem. In that case |
251 | * we defer the vm_locked accounting to the system workqueue. | |
eb8ffbfe | 252 | */ |
24bce508 RD |
253 | if (context->closing) { |
254 | if (!down_write_trylock(&mm->mmap_sem)) { | |
255 | INIT_WORK(&umem->work, ib_umem_account); | |
256 | umem->mm = mm; | |
257 | umem->diff = diff; | |
258 | ||
f0626710 | 259 | queue_work(ib_wq, &umem->work); |
24bce508 RD |
260 | return; |
261 | } | |
f7c6a7b5 RD |
262 | } else |
263 | down_write(&mm->mmap_sem); | |
264 | ||
c4870eb8 | 265 | current->mm->pinned_vm -= diff; |
f7c6a7b5 RD |
266 | up_write(&mm->mmap_sem); |
267 | mmput(mm); | |
1bf66a30 | 268 | kfree(umem); |
f7c6a7b5 RD |
269 | } |
270 | EXPORT_SYMBOL(ib_umem_release); | |
271 | ||
272 | int ib_umem_page_count(struct ib_umem *umem) | |
273 | { | |
f7c6a7b5 RD |
274 | int shift; |
275 | int i; | |
276 | int n; | |
eeb8461e | 277 | struct scatterlist *sg; |
f7c6a7b5 RD |
278 | |
279 | shift = ilog2(umem->page_size); | |
eb8ffbfe | 280 | |
f7c6a7b5 | 281 | n = 0; |
eeb8461e YH |
282 | for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i) |
283 | n += sg_dma_len(sg) >> shift; | |
eb8ffbfe | 284 | |
f7c6a7b5 | 285 | return n; |
eb8ffbfe | 286 | } |
f7c6a7b5 | 287 | EXPORT_SYMBOL(ib_umem_page_count); |