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