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
 40struct rds_page_remainder {
 41	struct page	*r_page;
 42	unsigned long	r_offset;
 43	local_lock_t	bh_lock;
 44};
 45
 46static DEFINE_PER_CPU_SHARED_ALIGNED(struct rds_page_remainder, rds_page_remainders) = {
 47	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
 48};
 49
 50/**
 51 * rds_page_remainder_alloc - build up regions of a message.
 52 *
 53 * @scat: Scatter list for message
 54 * @bytes: the number of bytes needed.
 55 * @gfp: the waiting behaviour of the allocation
 56 *
 57 * @gfp is always ored with __GFP_HIGHMEM.  Callers must be prepared to
 58 * kmap the pages, etc.
 59 *
 60 * If @bytes is at least a full page then this just returns a page from
 61 * alloc_page().
 62 *
 63 * If @bytes is a partial page then this stores the unused region of the
 64 * page in a per-cpu structure.  Future partial-page allocations may be
 65 * satisfied from that cached region.  This lets us waste less memory on
 66 * small allocations with minimal complexity.  It works because the transmit
 67 * path passes read-only page regions down to devices.  They hold a page
 68 * reference until they are done with the region.
 69 */
 70int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes,
 71			     gfp_t gfp)
 72{
 73	struct rds_page_remainder *rem;
 74	struct page *page;
 75	int ret;
 76
 77	gfp |= __GFP_HIGHMEM;
 78
 79	/* jump straight to allocation if we're trying for a huge page */
 80	if (bytes >= PAGE_SIZE) {
 81		page = alloc_page(gfp);
 82		if (!page) {
 83			ret = -ENOMEM;
 84		} else {
 85			sg_set_page(scat, page, PAGE_SIZE, 0);
 86			ret = 0;
 87		}
 88		goto out;
 89	}
 90
 91	local_bh_disable();
 92	local_lock_nested_bh(&rds_page_remainders.bh_lock);
 93	rem = this_cpu_ptr(&rds_page_remainders);
 94
 95	while (1) {
 96		/* avoid a tiny region getting stuck by tossing it */
 97		if (rem->r_page && bytes > (PAGE_SIZE - rem->r_offset)) {
 98			rds_stats_inc(s_page_remainder_miss);
 99			__free_page(rem->r_page);
100			rem->r_page = NULL;
101		}
102
103		/* hand out a fragment from the cached page */
104		if (rem->r_page && bytes <= (PAGE_SIZE - rem->r_offset)) {
105			sg_set_page(scat, rem->r_page, bytes, rem->r_offset);
106			get_page(sg_page(scat));
107
108			if (rem->r_offset != 0)
109				rds_stats_inc(s_page_remainder_hit);
110
111			rem->r_offset += ALIGN(bytes, 8);
112			if (rem->r_offset >= PAGE_SIZE) {
113				__free_page(rem->r_page);
114				rem->r_page = NULL;
115			}
116			ret = 0;
117			break;
118		}
119
120		/* alloc if there is nothing for us to use */
121		local_unlock_nested_bh(&rds_page_remainders.bh_lock);
122		local_bh_enable();
123
124		page = alloc_page(gfp);
125
126		local_bh_disable();
127		local_lock_nested_bh(&rds_page_remainders.bh_lock);
128		rem = this_cpu_ptr(&rds_page_remainders);
129
130		if (!page) {
131			ret = -ENOMEM;
132			break;
133		}
134
135		/* did someone race to fill the remainder before us? */
136		if (rem->r_page) {
137			__free_page(page);
138			continue;
139		}
140
141		/* otherwise install our page and loop around to alloc */
142		rem->r_page = page;
143		rem->r_offset = 0;
144	}
145
146	local_unlock_nested_bh(&rds_page_remainders.bh_lock);
147	local_bh_enable();
148out:
149	rdsdebug("bytes %lu ret %d %p %u %u\n", bytes, ret,
150		 ret ? NULL : sg_page(scat), ret ? 0 : scat->offset,
151		 ret ? 0 : scat->length);
152	return ret;
153}
154EXPORT_SYMBOL_GPL(rds_page_remainder_alloc);
155
156void rds_page_exit(void)
157{
158	unsigned int cpu;
159
160	for_each_possible_cpu(cpu) {
161		struct rds_page_remainder *rem;
162
163		rem = &per_cpu(rds_page_remainders, cpu);
164		rdsdebug("cpu %u\n", cpu);
165
166		if (rem->r_page)
167			__free_page(rem->r_page);
168		rem->r_page = NULL;
169	}
170}