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