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 38struct rds_page_remainder { 39 struct page *r_page; 40 unsigned long r_offset; 41}; 42 43DEFINE_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 */ 53int 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 addr = kmap(page); 61 if (to_user) { 62 rds_stats_add(s_copy_to_user, bytes); 63 ret = copy_to_user(ptr, addr + offset, bytes); 64 } else { 65 rds_stats_add(s_copy_from_user, bytes); 66 ret = copy_from_user(addr + offset, ptr, bytes); 67 } 68 kunmap(page); 69 70 return ret ? -EFAULT : 0; 71} 72EXPORT_SYMBOL_GPL(rds_page_copy_user); 73 74/* 75 * Message allocation uses this to build up regions of a message. 76 * 77 * @bytes - the number of bytes needed. 78 * @gfp - the waiting behaviour of the allocation 79 * 80 * @gfp is always ored with __GFP_HIGHMEM. Callers must be prepared to 81 * kmap the pages, etc. 82 * 83 * If @bytes is at least a full page then this just returns a page from 84 * alloc_page(). 85 * 86 * If @bytes is a partial page then this stores the unused region of the 87 * page in a per-cpu structure. Future partial-page allocations may be 88 * satisfied from that cached region. This lets us waste less memory on 89 * small allocations with minimal complexity. It works because the transmit 90 * path passes read-only page regions down to devices. They hold a page 91 * reference until they are done with the region. 92 */ 93int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes, 94 gfp_t gfp) 95{ 96 struct rds_page_remainder *rem; 97 unsigned long flags; 98 struct page *page; 99 int ret; 100 101 gfp |= __GFP_HIGHMEM; 102 103 /* jump straight to allocation if we're trying for a huge page */ 104 if (bytes >= PAGE_SIZE) { 105 page = alloc_page(gfp); 106 if (page == NULL) { 107 ret = -ENOMEM; 108 } else { 109 sg_set_page(scat, page, PAGE_SIZE, 0); 110 ret = 0; 111 } 112 goto out; 113 } 114 115 rem = &per_cpu(rds_page_remainders, get_cpu()); 116 local_irq_save(flags); 117 118 while (1) { 119 /* avoid a tiny region getting stuck by tossing it */ 120 if (rem->r_page && bytes > (PAGE_SIZE - rem->r_offset)) { 121 rds_stats_inc(s_page_remainder_miss); 122 __free_page(rem->r_page); 123 rem->r_page = NULL; 124 } 125 126 /* hand out a fragment from the cached page */ 127 if (rem->r_page && bytes <= (PAGE_SIZE - rem->r_offset)) { 128 sg_set_page(scat, rem->r_page, bytes, rem->r_offset); 129 get_page(sg_page(scat)); 130 131 if (rem->r_offset != 0) 132 rds_stats_inc(s_page_remainder_hit); 133 134 rem->r_offset += bytes; 135 if (rem->r_offset == PAGE_SIZE) { 136 __free_page(rem->r_page); 137 rem->r_page = NULL; 138 } 139 ret = 0; 140 break; 141 } 142 143 /* alloc if there is nothing for us to use */ 144 local_irq_restore(flags); 145 put_cpu(); 146 147 page = alloc_page(gfp); 148 149 rem = &per_cpu(rds_page_remainders, get_cpu()); 150 local_irq_save(flags); 151 152 if (page == NULL) { 153 ret = -ENOMEM; 154 break; 155 } 156 157 /* did someone race to fill the remainder before us? */ 158 if (rem->r_page) { 159 __free_page(page); 160 continue; 161 } 162 163 /* otherwise install our page and loop around to alloc */ 164 rem->r_page = page; 165 rem->r_offset = 0; 166 } 167 168 local_irq_restore(flags); 169 put_cpu(); 170out: 171 rdsdebug("bytes %lu ret %d %p %u %u\n", bytes, ret, 172 ret ? NULL : sg_page(scat), ret ? 0 : scat->offset, 173 ret ? 0 : scat->length); 174 return ret; 175} 176 177static int rds_page_remainder_cpu_notify(struct notifier_block *self, 178 unsigned long action, void *hcpu) 179{ 180 struct rds_page_remainder *rem; 181 long cpu = (long)hcpu; 182 183 rem = &per_cpu(rds_page_remainders, cpu); 184 185 rdsdebug("cpu %ld action 0x%lx\n", cpu, action); 186 187 switch (action) { 188 case CPU_DEAD: 189 if (rem->r_page) 190 __free_page(rem->r_page); 191 rem->r_page = NULL; 192 break; 193 } 194 195 return 0; 196} 197 198static struct notifier_block rds_page_remainder_nb = { 199 .notifier_call = rds_page_remainder_cpu_notify, 200}; 201 202void rds_page_exit(void) 203{ 204 int i; 205 206 for_each_possible_cpu(i) 207 rds_page_remainder_cpu_notify(&rds_page_remainder_nb, 208 (unsigned long)CPU_DEAD, 209 (void *)(long)i); 210} 211