net/rds/page.c

   1 /*
   2  * Copyright (c) 2006 Oracle.  All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  *
  32  */
  33 #include <linux/highmem.h>
  34 #include <linux/gfp.h>
  35
  36 #include "rds.h"
  37
  38 struct rds_page_remainder {
  39         struct page     *r_page;
  40         unsigned long   r_offset;
  41 };
  42
  43 DEFINE_PER_CPU_SHARED_ALIGNED(struct rds_page_remainder, rds_page_remainders);
  44
  45 /*
  46  * returns 0 on success or -errno on failure.
  47  *
  48  * We don't have to worry about flush_dcache_page() as this only works
  49  * with private pages.  If, say, we were to do directed receive to pinned
  50  * user pages we'd have to worry more about cache coherence.  (Though
  51  * the flush_dcache_page() in get_user_pages() would probably be enough).
  52  */
  53 int rds_page_copy_user(struct page *page, unsigned long offset,
  54                        void __user *ptr, unsigned long bytes,
  55                        int to_user)
  56 {
  57         unsigned long ret;
  58         void *addr;
  59
  60         if (to_user)
  61                 rds_stats_add(s_copy_to_user, bytes);
  62         else
  63                 rds_stats_add(s_copy_from_user, bytes);
  64
  65         addr = kmap_atomic(page, KM_USER0);
  66         if (to_user)
  67                 ret = __copy_to_user_inatomic(ptr, addr + offset, bytes);
  68         else
  69                 ret = __copy_from_user_inatomic(addr + offset, ptr, bytes);
  70         kunmap_atomic(addr, KM_USER0);
  71
  72         if (ret) {
  73                 addr = kmap(page);
  74                 if (to_user)
  75                         ret = copy_to_user(ptr, addr + offset, bytes);
  76                 else
  77                         ret = copy_from_user(addr + offset, ptr, bytes);
  78                 kunmap(page);
  79                 if (ret)
  80                         return -EFAULT;
  81         }
  82
  83         return 0;
  84 }
  85 EXPORT_SYMBOL_GPL(rds_page_copy_user);
  86
  87 /*
  88  * Message allocation uses this to build up regions of a message.
  89  *
  90  * @bytes - the number of bytes needed.
  91  * @gfp - the waiting behaviour of the allocation
  92  *
  93  * @gfp is always ored with __GFP_HIGHMEM.  Callers must be prepared to
  94  * kmap the pages, etc.
  95  *
  96  * If @bytes is at least a full page then this just returns a page from
  97  * alloc_page().
  98  *
  99  * If @bytes is a partial page then this stores the unused region of the
 100  * page in a per-cpu structure.  Future partial-page allocations may be
 101  * satisfied from that cached region.  This lets us waste less memory on
 102  * small allocations with minimal complexity.  It works because the transmit
 103  * path passes read-only page regions down to devices.  They hold a page
 104  * reference until they are done with the region.
 105  */
 106 int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes,
 107                              gfp_t gfp)
 108 {
 109         struct rds_page_remainder *rem;
 110         unsigned long flags;
 111         struct page *page;
 112         int ret;
 113
 114         gfp |= __GFP_HIGHMEM;
 115
 116         /* jump straight to allocation if we're trying for a huge page */
 117         if (bytes >= PAGE_SIZE) {
 118                 page = alloc_page(gfp);
 119                 if (page == NULL) {
 120                         ret = -ENOMEM;
 121                 } else {
 122                         sg_set_page(scat, page, PAGE_SIZE, 0);
 123                         ret = 0;
 124                 }
 125                 goto out;
 126         }
 127
 128         rem = &per_cpu(rds_page_remainders, get_cpu());
 129         local_irq_save(flags);
 130
 131         while (1) {
 132                 /* avoid a tiny region getting stuck by tossing it */
 133                 if (rem->r_page && bytes > (PAGE_SIZE - rem->r_offset)) {
 134                         rds_stats_inc(s_page_remainder_miss);
 135                         __free_page(rem->r_page);
 136                         rem->r_page = NULL;
 137                 }
 138
 139                 /* hand out a fragment from the cached page */
 140                 if (rem->r_page && bytes <= (PAGE_SIZE - rem->r_offset)) {
 141                         sg_set_page(scat, rem->r_page, bytes, rem->r_offset);
 142                         get_page(sg_page(scat));
 143
 144                         if (rem->r_offset != 0)
 145                                 rds_stats_inc(s_page_remainder_hit);
 146
 147                         rem->r_offset += bytes;
 148                         if (rem->r_offset == PAGE_SIZE) {
 149                                 __free_page(rem->r_page);
 150                                 rem->r_page = NULL;
 151                         }
 152                         ret = 0;
 153                         break;
 154                 }
 155
 156                 /* alloc if there is nothing for us to use */
 157                 local_irq_restore(flags);
 158                 put_cpu();
 159
 160                 page = alloc_page(gfp);
 161
 162                 rem = &per_cpu(rds_page_remainders, get_cpu());
 163                 local_irq_save(flags);
 164
 165                 if (page == NULL) {
 166                         ret = -ENOMEM;
 167                         break;
 168                 }
 169
 170                 /* did someone race to fill the remainder before us? */
 171                 if (rem->r_page) {
 172                         __free_page(page);
 173                         continue;
 174                 }
 175
 176                 /* otherwise install our page and loop around to alloc */
 177                 rem->r_page = page;
 178                 rem->r_offset = 0;
 179         }
 180
 181         local_irq_restore(flags);
 182         put_cpu();
 183 out:
 184         rdsdebug("bytes %lu ret %d %p %u %u\n", bytes, ret,
 185                  ret ? NULL : sg_page(scat), ret ? 0 : scat->offset,
 186                  ret ? 0 : scat->length);
 187         return ret;
 188 }
 189
 190 static int rds_page_remainder_cpu_notify(struct notifier_block *self,
 191                                          unsigned long action, void *hcpu)
 192 {
 193         struct rds_page_remainder *rem;
 194         long cpu = (long)hcpu;
 195
 196         rem = &per_cpu(rds_page_remainders, cpu);
 197
 198         rdsdebug("cpu %ld action 0x%lx\n", cpu, action);
 199
 200         switch (action) {
 201         case CPU_DEAD:
 202                 if (rem->r_page)
 203                         __free_page(rem->r_page);
 204                 rem->r_page = NULL;
 205                 break;
 206         }
 207
 208         return 0;
 209 }
 210
 211 static struct notifier_block rds_page_remainder_nb = {
 212         .notifier_call = rds_page_remainder_cpu_notify,
 213 };
 214
 215 void rds_page_exit(void)
 216 {
 217         int i;
 218
 219         for_each_possible_cpu(i)
 220                 rds_page_remainder_cpu_notify(&rds_page_remainder_nb,
 221                                               (unsigned long)CPU_DEAD,
 222                                               (void *)(long)i);
 223 }