Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
iovlock.c
Go to the documentation of this file.
1 /*
2  * Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
3  * Portions based on net/core/datagram.c and copyrighted by their authors.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License as published by the Free
7  * Software Foundation; either version 2 of the License, or (at your option)
8  * any later version.
9  *
10  * This program is distributed in the hope that it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program; if not, write to the Free Software Foundation, Inc., 59
17  * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18  *
19  * The full GNU General Public License is included in this distribution in the
20  * file called COPYING.
21  */
22 
23 /*
24  * This code allows the net stack to make use of a DMA engine for
25  * skb to iovec copies.
26  */
27 
28 #include <linux/dmaengine.h>
29 #include <linux/pagemap.h>
30 #include <linux/slab.h>
31 #include <net/tcp.h> /* for memcpy_toiovec */
32 #include <asm/io.h>
33 #include <asm/uaccess.h>
34 
35 static int num_pages_spanned(struct iovec *iov)
36 {
37  return
38  ((PAGE_ALIGN((unsigned long)iov->iov_base + iov->iov_len) -
39  ((unsigned long)iov->iov_base & PAGE_MASK)) >> PAGE_SHIFT);
40 }
41 
42 /*
43  * Pin down all the iovec pages needed for len bytes.
44  * Return a struct dma_pinned_list to keep track of pages pinned down.
45  *
46  * We are allocating a single chunk of memory, and then carving it up into
47  * 3 sections, the latter 2 whose size depends on the number of iovecs and the
48  * total number of pages, respectively.
49  */
50 struct dma_pinned_list *dma_pin_iovec_pages(struct iovec *iov, size_t len)
51 {
52  struct dma_pinned_list *local_list;
53  struct page **pages;
54  int i;
55  int ret;
56  int nr_iovecs = 0;
57  int iovec_len_used = 0;
58  int iovec_pages_used = 0;
59 
60  /* don't pin down non-user-based iovecs */
61  if (segment_eq(get_fs(), KERNEL_DS))
62  return NULL;
63 
64  /* determine how many iovecs/pages there are, up front */
65  do {
66  iovec_len_used += iov[nr_iovecs].iov_len;
67  iovec_pages_used += num_pages_spanned(&iov[nr_iovecs]);
68  nr_iovecs++;
69  } while (iovec_len_used < len);
70 
71  /* single kmalloc for pinned list, page_list[], and the page arrays */
72  local_list = kmalloc(sizeof(*local_list)
73  + (nr_iovecs * sizeof (struct dma_page_list))
74  + (iovec_pages_used * sizeof (struct page*)), GFP_KERNEL);
75  if (!local_list)
76  goto out;
77 
78  /* list of pages starts right after the page list array */
79  pages = (struct page **) &local_list->page_list[nr_iovecs];
80 
81  local_list->nr_iovecs = 0;
82 
83  for (i = 0; i < nr_iovecs; i++) {
84  struct dma_page_list *page_list = &local_list->page_list[i];
85 
86  len -= iov[i].iov_len;
87 
88  if (!access_ok(VERIFY_WRITE, iov[i].iov_base, iov[i].iov_len))
89  goto unpin;
90 
91  page_list->nr_pages = num_pages_spanned(&iov[i]);
92  page_list->base_address = iov[i].iov_base;
93 
94  page_list->pages = pages;
95  pages += page_list->nr_pages;
96 
97  /* pin pages down */
98  down_read(&current->mm->mmap_sem);
99  ret = get_user_pages(
100  current,
101  current->mm,
102  (unsigned long) iov[i].iov_base,
103  page_list->nr_pages,
104  1, /* write */
105  0, /* force */
106  page_list->pages,
107  NULL);
108  up_read(&current->mm->mmap_sem);
109 
110  if (ret != page_list->nr_pages)
111  goto unpin;
112 
113  local_list->nr_iovecs = i + 1;
114  }
115 
116  return local_list;
117 
118 unpin:
119  dma_unpin_iovec_pages(local_list);
120 out:
121  return NULL;
122 }
123 
124 void dma_unpin_iovec_pages(struct dma_pinned_list *pinned_list)
125 {
126  int i, j;
127 
128  if (!pinned_list)
129  return;
130 
131  for (i = 0; i < pinned_list->nr_iovecs; i++) {
132  struct dma_page_list *page_list = &pinned_list->page_list[i];
133  for (j = 0; j < page_list->nr_pages; j++) {
134  set_page_dirty_lock(page_list->pages[j]);
135  page_cache_release(page_list->pages[j]);
136  }
137  }
138 
139  kfree(pinned_list);
140 }
141 
142 
143 /*
144  * We have already pinned down the pages we will be using in the iovecs.
145  * Each entry in iov array has corresponding entry in pinned_list->page_list.
146  * Using array indexing to keep iov[] and page_list[] in sync.
147  * Initial elements in iov array's iov->iov_len will be 0 if already copied into
148  * by another call.
149  * iov array length remaining guaranteed to be bigger than len.
150  */
152  struct dma_pinned_list *pinned_list, unsigned char *kdata, size_t len)
153 {
154  int iov_byte_offset;
155  int copy;
156  dma_cookie_t dma_cookie = 0;
157  int iovec_idx;
158  int page_idx;
159 
160  if (!chan)
161  return memcpy_toiovec(iov, kdata, len);
162 
163  iovec_idx = 0;
164  while (iovec_idx < pinned_list->nr_iovecs) {
165  struct dma_page_list *page_list;
166 
167  /* skip already used-up iovecs */
168  while (!iov[iovec_idx].iov_len)
169  iovec_idx++;
170 
171  page_list = &pinned_list->page_list[iovec_idx];
172 
173  iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
174  page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
175  - ((unsigned long)page_list->base_address & PAGE_MASK)) >> PAGE_SHIFT;
176 
177  /* break up copies to not cross page boundary */
178  while (iov[iovec_idx].iov_len) {
179  copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
180  copy = min_t(int, copy, iov[iovec_idx].iov_len);
181 
182  dma_cookie = dma_async_memcpy_buf_to_pg(chan,
183  page_list->pages[page_idx],
184  iov_byte_offset,
185  kdata,
186  copy);
187  /* poll for a descriptor slot */
188  if (unlikely(dma_cookie < 0)) {
189  dma_async_issue_pending(chan);
190  continue;
191  }
192 
193  len -= copy;
194  iov[iovec_idx].iov_len -= copy;
195  iov[iovec_idx].iov_base += copy;
196 
197  if (!len)
198  return dma_cookie;
199 
200  kdata += copy;
201  iov_byte_offset = 0;
202  page_idx++;
203  }
204  iovec_idx++;
205  }
206 
207  /* really bad if we ever run out of iovecs */
208  BUG();
209  return -EFAULT;
210 }
211 
213  struct dma_pinned_list *pinned_list, struct page *page,
214  unsigned int offset, size_t len)
215 {
216  int iov_byte_offset;
217  int copy;
218  dma_cookie_t dma_cookie = 0;
219  int iovec_idx;
220  int page_idx;
221  int err;
222 
223  /* this needs as-yet-unimplemented buf-to-buff, so punt. */
224  /* TODO: use dma for this */
225  if (!chan || !pinned_list) {
226  u8 *vaddr = kmap(page);
227  err = memcpy_toiovec(iov, vaddr + offset, len);
228  kunmap(page);
229  return err;
230  }
231 
232  iovec_idx = 0;
233  while (iovec_idx < pinned_list->nr_iovecs) {
234  struct dma_page_list *page_list;
235 
236  /* skip already used-up iovecs */
237  while (!iov[iovec_idx].iov_len)
238  iovec_idx++;
239 
240  page_list = &pinned_list->page_list[iovec_idx];
241 
242  iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
243  page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
244  - ((unsigned long)page_list->base_address & PAGE_MASK)) >> PAGE_SHIFT;
245 
246  /* break up copies to not cross page boundary */
247  while (iov[iovec_idx].iov_len) {
248  copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
249  copy = min_t(int, copy, iov[iovec_idx].iov_len);
250 
251  dma_cookie = dma_async_memcpy_pg_to_pg(chan,
252  page_list->pages[page_idx],
253  iov_byte_offset,
254  page,
255  offset,
256  copy);
257  /* poll for a descriptor slot */
258  if (unlikely(dma_cookie < 0)) {
259  dma_async_issue_pending(chan);
260  continue;
261  }
262 
263  len -= copy;
264  iov[iovec_idx].iov_len -= copy;
265  iov[iovec_idx].iov_base += copy;
266 
267  if (!len)
268  return dma_cookie;
269 
270  offset += copy;
271  iov_byte_offset = 0;
272  page_idx++;
273  }
274  iovec_idx++;
275  }
276 
277  /* really bad if we ever run out of iovecs */
278  BUG();
279  return -EFAULT;
280 }