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