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; | |
406f9e5f | 106 | umem->address = addr; |
f7c6a7b5 | 107 | umem->page_size = PAGE_SIZE; |
87773dd5 | 108 | umem->pid = get_task_pid(current, PIDTYPE_PID); |
f7c6a7b5 | 109 | /* |
860f10a7 SG |
110 | * We ask for writable memory if any of the following |
111 | * access flags are set. "Local write" and "remote write" | |
f7c6a7b5 RD |
112 | * obviously require write access. "Remote atomic" can do |
113 | * things like fetch and add, which will modify memory, and | |
114 | * "MW bind" can change permissions by binding a window. | |
115 | */ | |
860f10a7 SG |
116 | umem->writable = !!(access & |
117 | (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | | |
118 | IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND)); | |
eb8ffbfe | 119 | |
c8d8beea JF |
120 | /* We assume the memory is from hugetlb until proved otherwise */ |
121 | umem->hugetlb = 1; | |
122 | ||
f7c6a7b5 RD |
123 | page_list = (struct page **) __get_free_page(GFP_KERNEL); |
124 | if (!page_list) { | |
125 | kfree(umem); | |
126 | return ERR_PTR(-ENOMEM); | |
127 | } | |
eb8ffbfe | 128 | |
c8d8beea JF |
129 | /* |
130 | * if we can't alloc the vma_list, it's not so bad; | |
131 | * just assume the memory is not hugetlb memory | |
132 | */ | |
133 | vma_list = (struct vm_area_struct **) __get_free_page(GFP_KERNEL); | |
134 | if (!vma_list) | |
135 | umem->hugetlb = 0; | |
136 | ||
406f9e5f | 137 | npages = ib_umem_num_pages(umem); |
eb8ffbfe RD |
138 | |
139 | down_write(¤t->mm->mmap_sem); | |
140 | ||
bc3e53f6 | 141 | locked = npages + current->mm->pinned_vm; |
ccbe9f0b | 142 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; |
eb8ffbfe RD |
143 | |
144 | if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { | |
145 | ret = -ENOMEM; | |
146 | goto out; | |
147 | } | |
148 | ||
f7c6a7b5 | 149 | cur_base = addr & PAGE_MASK; |
eb8ffbfe | 150 | |
eeb8461e YH |
151 | if (npages == 0) { |
152 | ret = -EINVAL; | |
153 | goto out; | |
154 | } | |
155 | ||
156 | ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL); | |
157 | if (ret) | |
158 | goto out; | |
159 | ||
160 | need_release = 1; | |
161 | sg_list_start = umem->sg_head.sgl; | |
162 | ||
eb8ffbfe RD |
163 | while (npages) { |
164 | ret = get_user_pages(current, current->mm, cur_base, | |
8079ffa0 | 165 | min_t(unsigned long, npages, |
eb8ffbfe | 166 | PAGE_SIZE / sizeof (struct page *)), |
c8d8beea | 167 | 1, !umem->writable, page_list, vma_list); |
eb8ffbfe RD |
168 | |
169 | if (ret < 0) | |
170 | goto out; | |
171 | ||
eeb8461e | 172 | umem->npages += ret; |
eb8ffbfe RD |
173 | cur_base += ret * PAGE_SIZE; |
174 | npages -= ret; | |
175 | ||
eeb8461e YH |
176 | for_each_sg(sg_list_start, sg, ret, i) { |
177 | if (vma_list && !is_vm_hugetlb_page(vma_list[i])) | |
178 | umem->hugetlb = 0; | |
179 | ||
180 | sg_set_page(sg, page_list[i], PAGE_SIZE, 0); | |
eb8ffbfe RD |
181 | } |
182 | ||
eeb8461e YH |
183 | /* preparing for next loop */ |
184 | sg_list_start = sg; | |
eb8ffbfe RD |
185 | } |
186 | ||
eeb8461e YH |
187 | umem->nmap = ib_dma_map_sg_attrs(context->device, |
188 | umem->sg_head.sgl, | |
189 | umem->npages, | |
190 | DMA_BIDIRECTIONAL, | |
191 | &attrs); | |
192 | ||
193 | if (umem->nmap <= 0) { | |
194 | ret = -ENOMEM; | |
195 | goto out; | |
196 | } | |
197 | ||
198 | ret = 0; | |
199 | ||
eb8ffbfe | 200 | out: |
f7c6a7b5 | 201 | if (ret < 0) { |
eeb8461e YH |
202 | if (need_release) |
203 | __ib_umem_release(context->device, umem, 0); | |
87773dd5 | 204 | put_pid(umem->pid); |
f7c6a7b5 RD |
205 | kfree(umem); |
206 | } else | |
bc3e53f6 | 207 | current->mm->pinned_vm = locked; |
eb8ffbfe RD |
208 | |
209 | up_write(¤t->mm->mmap_sem); | |
c8d8beea JF |
210 | if (vma_list) |
211 | free_page((unsigned long) vma_list); | |
eb8ffbfe RD |
212 | free_page((unsigned long) page_list); |
213 | ||
f7c6a7b5 | 214 | return ret < 0 ? ERR_PTR(ret) : umem; |
eb8ffbfe | 215 | } |
f7c6a7b5 | 216 | EXPORT_SYMBOL(ib_umem_get); |
eb8ffbfe | 217 | |
1bf66a30 | 218 | static void ib_umem_account(struct work_struct *work) |
eb8ffbfe | 219 | { |
1bf66a30 RD |
220 | struct ib_umem *umem = container_of(work, struct ib_umem, work); |
221 | ||
222 | down_write(&umem->mm->mmap_sem); | |
bc3e53f6 | 223 | umem->mm->pinned_vm -= umem->diff; |
1bf66a30 RD |
224 | up_write(&umem->mm->mmap_sem); |
225 | mmput(umem->mm); | |
226 | kfree(umem); | |
eb8ffbfe RD |
227 | } |
228 | ||
f7c6a7b5 RD |
229 | /** |
230 | * ib_umem_release - release memory pinned with ib_umem_get | |
231 | * @umem: umem struct to release | |
232 | */ | |
233 | void ib_umem_release(struct ib_umem *umem) | |
eb8ffbfe | 234 | { |
f7c6a7b5 | 235 | struct ib_ucontext *context = umem->context; |
eb8ffbfe | 236 | struct mm_struct *mm; |
87773dd5 | 237 | struct task_struct *task; |
f7c6a7b5 | 238 | unsigned long diff; |
eb8ffbfe | 239 | |
f7c6a7b5 | 240 | __ib_umem_release(umem->context->device, umem, 1); |
eb8ffbfe | 241 | |
87773dd5 SB |
242 | task = get_pid_task(umem->pid, PIDTYPE_PID); |
243 | put_pid(umem->pid); | |
244 | if (!task) | |
245 | goto out; | |
246 | mm = get_task_mm(task); | |
247 | put_task_struct(task); | |
248 | if (!mm) | |
249 | goto out; | |
eb8ffbfe | 250 | |
406f9e5f | 251 | diff = ib_umem_num_pages(umem); |
f7c6a7b5 | 252 | |
eb8ffbfe RD |
253 | /* |
254 | * We may be called with the mm's mmap_sem already held. This | |
255 | * can happen when a userspace munmap() is the call that drops | |
256 | * the last reference to our file and calls our release | |
257 | * method. If there are memory regions to destroy, we'll end | |
f7c6a7b5 RD |
258 | * up here and not be able to take the mmap_sem. In that case |
259 | * we defer the vm_locked accounting to the system workqueue. | |
eb8ffbfe | 260 | */ |
24bce508 RD |
261 | if (context->closing) { |
262 | if (!down_write_trylock(&mm->mmap_sem)) { | |
263 | INIT_WORK(&umem->work, ib_umem_account); | |
264 | umem->mm = mm; | |
265 | umem->diff = diff; | |
266 | ||
f0626710 | 267 | queue_work(ib_wq, &umem->work); |
24bce508 RD |
268 | return; |
269 | } | |
f7c6a7b5 RD |
270 | } else |
271 | down_write(&mm->mmap_sem); | |
272 | ||
87773dd5 | 273 | mm->pinned_vm -= diff; |
f7c6a7b5 RD |
274 | up_write(&mm->mmap_sem); |
275 | mmput(mm); | |
87773dd5 | 276 | out: |
1bf66a30 | 277 | kfree(umem); |
f7c6a7b5 RD |
278 | } |
279 | EXPORT_SYMBOL(ib_umem_release); | |
280 | ||
281 | int ib_umem_page_count(struct ib_umem *umem) | |
282 | { | |
f7c6a7b5 RD |
283 | int shift; |
284 | int i; | |
285 | int n; | |
eeb8461e | 286 | struct scatterlist *sg; |
f7c6a7b5 RD |
287 | |
288 | shift = ilog2(umem->page_size); | |
eb8ffbfe | 289 | |
f7c6a7b5 | 290 | n = 0; |
eeb8461e YH |
291 | for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i) |
292 | n += sg_dma_len(sg) >> shift; | |
eb8ffbfe | 293 | |
f7c6a7b5 | 294 | return n; |
eb8ffbfe | 295 | } |
f7c6a7b5 | 296 | EXPORT_SYMBOL(ib_umem_page_count); |
c5d76f13 HE |
297 | |
298 | /* | |
299 | * Copy from the given ib_umem's pages to the given buffer. | |
300 | * | |
301 | * umem - the umem to copy from | |
302 | * offset - offset to start copying from | |
303 | * dst - destination buffer | |
304 | * length - buffer length | |
305 | * | |
306 | * Returns 0 on success, or an error code. | |
307 | */ | |
308 | int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset, | |
309 | size_t length) | |
310 | { | |
311 | size_t end = offset + length; | |
312 | int ret; | |
313 | ||
314 | if (offset > umem->length || length > umem->length - offset) { | |
315 | pr_err("ib_umem_copy_from not in range. offset: %zd umem length: %zd end: %zd\n", | |
316 | offset, umem->length, end); | |
317 | return -EINVAL; | |
318 | } | |
319 | ||
320 | ret = sg_pcopy_to_buffer(umem->sg_head.sgl, umem->nmap, dst, length, | |
321 | offset + ib_umem_offset(umem)); | |
322 | ||
323 | if (ret < 0) | |
324 | return ret; | |
325 | else if (ret != length) | |
326 | return -EINVAL; | |
327 | else | |
328 | return 0; | |
329 | } | |
330 | EXPORT_SYMBOL(ib_umem_copy_from); |