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 static DEFINE_PER_CPU_SHARED_ALIGNED(struct rds_page_remainder,
  44                                      rds_page_remainders);
  45
  46 /*
  47  * returns 0 on success or -errno on failure.
  48  *
  49  * We don't have to worry about flush_dcache_page() as this only works
  50  * with private pages.  If, say, we were to do directed receive to pinned
  51  * user pages we'd have to worry more about cache coherence.  (Though
  52  * the flush_dcache_page() in get_user_pages() would probably be enough).
  53  */
  54 int rds_page_copy_user(struct page *page, unsigned long offset,
  55                        void __user *ptr, unsigned long bytes,
  56                        int to_user)
  57 {
  58         unsigned long ret;
  59         void *addr;
  60
  61         addr = kmap(page);
  62         if (to_user) {
  63                 rds_stats_add(s_copy_to_user, bytes);
  64                 ret = copy_to_user(ptr, addr + offset, bytes);
  65         } else {
  66                 rds_stats_add(s_copy_from_user, bytes);
  67                 ret = copy_from_user(addr + offset, ptr, bytes);
  68         }
  69         kunmap(page);
  70
  71         return ret ? -EFAULT : 0;
  72 }
  73 EXPORT_SYMBOL_GPL(rds_page_copy_user);
  74
  75 /*
  76  * Message allocation uses this to build up regions of a message.
  77  *
  78  * @bytes - the number of bytes needed.
  79  * @gfp - the waiting behaviour of the allocation
  80  *
  81  * @gfp is always ored with __GFP_HIGHMEM.  Callers must be prepared to
  82  * kmap the pages, etc.
  83  *
  84  * If @bytes is at least a full page then this just returns a page from
  85  * alloc_page().
  86  *
  87  * If @bytes is a partial page then this stores the unused region of the
  88  * page in a per-cpu structure.  Future partial-page allocations may be
  89  * satisfied from that cached region.  This lets us waste less memory on
  90  * small allocations with minimal complexity.  It works because the transmit
  91  * path passes read-only page regions down to devices.  They hold a page
  92  * reference until they are done with the region.
  93  */
  94 int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes,
  95                              gfp_t gfp)
  96 {
  97         struct rds_page_remainder *rem;
  98         unsigned long flags;
  99         struct page *page;
 100         int ret;
 101
 102         gfp |= __GFP_HIGHMEM;
 103
 104         /* jump straight to allocation if we're trying for a huge page */
 105         if (bytes >= PAGE_SIZE) {
 106                 page = alloc_page(gfp);
 107                 if (!page) {
 108                         ret = -ENOMEM;
 109                 } else {
 110                         sg_set_page(scat, page, PAGE_SIZE, 0);
 111                         ret = 0;
 112                 }
 113                 goto out;
 114         }
 115
 116         rem = &per_cpu(rds_page_remainders, get_cpu());
 117         local_irq_save(flags);
 118
 119         while (1) {
 120                 /* avoid a tiny region getting stuck by tossing it */
 121                 if (rem->r_page && bytes > (PAGE_SIZE - rem->r_offset)) {
 122                         rds_stats_inc(s_page_remainder_miss);
 123                         __free_page(rem->r_page);
 124                         rem->r_page = NULL;
 125                 }
 126
 127                 /* hand out a fragment from the cached page */
 128                 if (rem->r_page && bytes <= (PAGE_SIZE - rem->r_offset)) {
 129                         sg_set_page(scat, rem->r_page, bytes, rem->r_offset);
 130                         get_page(sg_page(scat));
 131
 132                         if (rem->r_offset != 0)
 133                                 rds_stats_inc(s_page_remainder_hit);
 134
 135                         rem->r_offset += bytes;
 136                         if (rem->r_offset == PAGE_SIZE) {
 137                                 __free_page(rem->r_page);
 138                                 rem->r_page = NULL;
 139                         }
 140                         ret = 0;
 141                         break;
 142                 }
 143
 144                 /* alloc if there is nothing for us to use */
 145                 local_irq_restore(flags);
 146                 put_cpu();
 147
 148                 page = alloc_page(gfp);
 149
 150                 rem = &per_cpu(rds_page_remainders, get_cpu());
 151                 local_irq_save(flags);
 152
 153                 if (!page) {
 154                         ret = -ENOMEM;
 155                         break;
 156                 }
 157
 158                 /* did someone race to fill the remainder before us? */
 159                 if (rem->r_page) {
 160                         __free_page(page);
 161                         continue;
 162                 }
 163
 164                 /* otherwise install our page and loop around to alloc */
 165                 rem->r_page = page;
 166                 rem->r_offset = 0;
 167         }
 168
 169         local_irq_restore(flags);
 170         put_cpu();
 171 out:
 172         rdsdebug("bytes %lu ret %d %p %u %u\n", bytes, ret,
 173                  ret ? NULL : sg_page(scat), ret ? 0 : scat->offset,
 174                  ret ? 0 : scat->length);
 175         return ret;
 176 }
 177 EXPORT_SYMBOL_GPL(rds_page_remainder_alloc);
 178
 179 static int rds_page_remainder_cpu_notify(struct notifier_block *self,
 180                                          unsigned long action, void *hcpu)
 181 {
 182         struct rds_page_remainder *rem;
 183         long cpu = (long)hcpu;
 184
 185         rem = &per_cpu(rds_page_remainders, cpu);
 186
 187         rdsdebug("cpu %ld action 0x%lx\n", cpu, action);
 188
 189         switch (action) {
 190         case CPU_DEAD:
 191                 if (rem->r_page)
 192                         __free_page(rem->r_page);
 193                 rem->r_page = NULL;
 194                 break;
 195         }
 196
 197         return 0;
 198 }
 199
 200 static struct notifier_block rds_page_remainder_nb = {
 201         .notifier_call = rds_page_remainder_cpu_notify,
 202 };
 203
 204 void rds_page_exit(void)
 205 {
 206         int i;
 207
 208         for_each_possible_cpu(i)
 209                 rds_page_remainder_cpu_notify(&rds_page_remainder_nb,
 210                                               (unsigned long)CPU_DEAD,
 211                                               (void *)(long)i);
 212 }