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
  35 #include "rds.h"
  36
  37 struct rds_page_remainder {
  38         struct page     *r_page;
  39         unsigned long   r_offset;
  40 };
  41
  42 DEFINE_PER_CPU(struct rds_page_remainder, rds_page_remainders) ____cacheline_aligned;
  43
  44 /*
  45  * returns 0 on success or -errno on failure.
  46  *
  47  * We don't have to worry about flush_dcache_page() as this only works
  48  * with private pages.  If, say, we were to do directed receive to pinned
  49  * user pages we'd have to worry more about cache coherence.  (Though
  50  * the flush_dcache_page() in get_user_pages() would probably be enough).
  51  */
  52 int rds_page_copy_user(struct page *page, unsigned long offset,
  53                        void __user *ptr, unsigned long bytes,
  54                        int to_user)
  55 {
  56         unsigned long ret;
  57         void *addr;
  58
  59         if (to_user)
  60                 rds_stats_add(s_copy_to_user, bytes);
  61         else
  62                 rds_stats_add(s_copy_from_user, bytes);
  63
  64         addr = kmap_atomic(page, KM_USER0);
  65         if (to_user)
  66                 ret = __copy_to_user_inatomic(ptr, addr + offset, bytes);
  67         else
  68                 ret = __copy_from_user_inatomic(addr + offset, ptr, bytes);
  69         kunmap_atomic(addr, KM_USER0);
  70
  71         if (ret) {
  72                 addr = kmap(page);
  73                 if (to_user)
  74                         ret = copy_to_user(ptr, addr + offset, bytes);
  75                 else
  76                         ret = copy_from_user(addr + offset, ptr, bytes);
  77                 kunmap(page);
  78                 if (ret)
  79                         return -EFAULT;
  80         }
  81
  82         return 0;
  83 }
  84
  85 /*
  86  * Message allocation uses this to build up regions of a message.
  87  *
  88  * @bytes - the number of bytes needed.
  89  * @gfp - the waiting behaviour of the allocation
  90  *
  91  * @gfp is always ored with __GFP_HIGHMEM.  Callers must be prepared to
  92  * kmap the pages, etc.
  93  *
  94  * If @bytes is at least a full page then this just returns a page from
  95  * alloc_page().
  96  *
  97  * If @bytes is a partial page then this stores the unused region of the
  98  * page in a per-cpu structure.  Future partial-page allocations may be
  99  * satisfied from that cached region.  This lets us waste less memory on
 100  * small allocations with minimal complexity.  It works because the transmit
 101  * path passes read-only page regions down to devices.  They hold a page
 102  * reference until they are done with the region.
 103  */
 104 int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes,
 105                              gfp_t gfp)
 106 {
 107         struct rds_page_remainder *rem;
 108         unsigned long flags;
 109         struct page *page;
 110         int ret;
 111
 112         gfp |= __GFP_HIGHMEM;
 113
 114         /* jump straight to allocation if we're trying for a huge page */
 115         if (bytes >= PAGE_SIZE) {
 116                 page = alloc_page(gfp);
 117                 if (page == NULL) {
 118                         ret = -ENOMEM;
 119                 } else {
 120                         sg_set_page(scat, page, PAGE_SIZE, 0);
 121                         ret = 0;
 122                 }
 123                 goto out;
 124         }
 125
 126         rem = &per_cpu(rds_page_remainders, get_cpu());
 127         local_irq_save(flags);
 128
 129         while (1) {
 130                 /* avoid a tiny region getting stuck by tossing it */
 131                 if (rem->r_page && bytes > (PAGE_SIZE - rem->r_offset)) {
 132                         rds_stats_inc(s_page_remainder_miss);
 133                         __free_page(rem->r_page);
 134                         rem->r_page = NULL;
 135                 }
 136
 137                 /* hand out a fragment from the cached page */
 138                 if (rem->r_page && bytes <= (PAGE_SIZE - rem->r_offset)) {
 139                         sg_set_page(scat, rem->r_page, bytes, rem->r_offset);
 140                         get_page(sg_page(scat));
 141
 142                         if (rem->r_offset != 0)
 143                                 rds_stats_inc(s_page_remainder_hit);
 144
 145                         rem->r_offset += bytes;
 146                         if (rem->r_offset == PAGE_SIZE) {
 147                                 __free_page(rem->r_page);
 148                                 rem->r_page = NULL;
 149                         }
 150                         ret = 0;
 151                         break;
 152                 }
 153
 154                 /* alloc if there is nothing for us to use */
 155                 local_irq_restore(flags);
 156                 put_cpu();
 157
 158                 page = alloc_page(gfp);
 159
 160                 rem = &per_cpu(rds_page_remainders, get_cpu());
 161                 local_irq_save(flags);
 162
 163                 if (page == NULL) {
 164                         ret = -ENOMEM;
 165                         break;
 166                 }
 167
 168                 /* did someone race to fill the remainder before us? */
 169                 if (rem->r_page) {
 170                         __free_page(page);
 171                         continue;
 172                 }
 173
 174                 /* otherwise install our page and loop around to alloc */
 175                 rem->r_page = page;
 176                 rem->r_offset = 0;
 177         }
 178
 179         local_irq_restore(flags);
 180         put_cpu();
 181 out:
 182         rdsdebug("bytes %lu ret %d %p %u %u\n", bytes, ret,
 183                  ret ? NULL : sg_page(scat), ret ? 0 : scat->offset,
 184                  ret ? 0 : scat->length);
 185         return ret;
 186 }
 187
 188 static int rds_page_remainder_cpu_notify(struct notifier_block *self,
 189                                          unsigned long action, void *hcpu)
 190 {
 191         struct rds_page_remainder *rem;
 192         long cpu = (long)hcpu;
 193
 194         rem = &per_cpu(rds_page_remainders, cpu);
 195
 196         rdsdebug("cpu %ld action 0x%lx\n", cpu, action);
 197
 198         switch (action) {
 199         case CPU_DEAD:
 200                 if (rem->r_page)
 201                         __free_page(rem->r_page);
 202                 rem->r_page = NULL;
 203                 break;
 204         }
 205
 206         return 0;
 207 }
 208
 209 static struct notifier_block rds_page_remainder_nb = {
 210         .notifier_call = rds_page_remainder_cpu_notify,
 211 };
 212
 213 void rds_page_exit(void)
 214 {
 215         int i;
 216
 217         for_each_possible_cpu(i)
 218                 rds_page_remainder_cpu_notify(&rds_page_remainder_nb,
 219                                               (unsigned long)CPU_DEAD,
 220                                               (void *)(long)i);
 221 }