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 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 */
56int 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}
75EXPORT_SYMBOL_GPL(rds_page_copy_user);
76
77/**
78 * rds_page_remainder_alloc - build up regions of a message.
79 *
80 * @scat: Scatter list for message
81 * @bytes: the number of bytes needed.
82 * @gfp: the waiting behaviour of the allocation
83 *
84 * @gfp is always ored with __GFP_HIGHMEM.  Callers must be prepared to
85 * kmap the pages, etc.
86 *
87 * If @bytes is at least a full page then this just returns a page from
88 * alloc_page().
89 *
90 * If @bytes is a partial page then this stores the unused region of the
91 * page in a per-cpu structure.  Future partial-page allocations may be
92 * satisfied from that cached region.  This lets us waste less memory on
93 * small allocations with minimal complexity.  It works because the transmit
94 * path passes read-only page regions down to devices.  They hold a page
95 * reference until they are done with the region.
96 */
97int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes,
98			     gfp_t gfp)
99{
100	struct rds_page_remainder *rem;
101	unsigned long flags;
102	struct page *page;
103	int ret;
104
105	gfp |= __GFP_HIGHMEM;
106
107	/* jump straight to allocation if we're trying for a huge page */
108	if (bytes >= PAGE_SIZE) {
109		page = alloc_page(gfp);
110		if (!page) {
111			ret = -ENOMEM;
112		} else {
113			sg_set_page(scat, page, PAGE_SIZE, 0);
114			ret = 0;
115		}
116		goto out;
117	}
118
119	rem = &per_cpu(rds_page_remainders, get_cpu());
120	local_irq_save(flags);
121
122	while (1) {
123		/* avoid a tiny region getting stuck by tossing it */
124		if (rem->r_page && bytes > (PAGE_SIZE - rem->r_offset)) {
125			rds_stats_inc(s_page_remainder_miss);
126			__free_page(rem->r_page);
127			rem->r_page = NULL;
128		}
129
130		/* hand out a fragment from the cached page */
131		if (rem->r_page && bytes <= (PAGE_SIZE - rem->r_offset)) {
132			sg_set_page(scat, rem->r_page, bytes, rem->r_offset);
133			get_page(sg_page(scat));
134
135			if (rem->r_offset != 0)
136				rds_stats_inc(s_page_remainder_hit);
137
138			rem->r_offset += bytes;
139			if (rem->r_offset == PAGE_SIZE) {
140				__free_page(rem->r_page);
141				rem->r_page = NULL;
142			}
143			ret = 0;
144			break;
145		}
146
147		/* alloc if there is nothing for us to use */
148		local_irq_restore(flags);
149		put_cpu();
150
151		page = alloc_page(gfp);
152
153		rem = &per_cpu(rds_page_remainders, get_cpu());
154		local_irq_save(flags);
155
156		if (!page) {
157			ret = -ENOMEM;
158			break;
159		}
160
161		/* did someone race to fill the remainder before us? */
162		if (rem->r_page) {
163			__free_page(page);
164			continue;
165		}
166
167		/* otherwise install our page and loop around to alloc */
168		rem->r_page = page;
169		rem->r_offset = 0;
170	}
171
172	local_irq_restore(flags);
173	put_cpu();
174out:
175	rdsdebug("bytes %lu ret %d %p %u %u\n", bytes, ret,
176		 ret ? NULL : sg_page(scat), ret ? 0 : scat->offset,
177		 ret ? 0 : scat->length);
178	return ret;
179}
180EXPORT_SYMBOL_GPL(rds_page_remainder_alloc);
181
182static int rds_page_remainder_cpu_notify(struct notifier_block *self,
183					 unsigned long action, void *hcpu)
184{
185	struct rds_page_remainder *rem;
186	long cpu = (long)hcpu;
187
188	rem = &per_cpu(rds_page_remainders, cpu);
189
190	rdsdebug("cpu %ld action 0x%lx\n", cpu, action);
191
192	switch (action) {
193	case CPU_DEAD:
194		if (rem->r_page)
195			__free_page(rem->r_page);
196		rem->r_page = NULL;
197		break;
198	}
199
200	return 0;
201}
202
203static struct notifier_block rds_page_remainder_nb = {
204	.notifier_call = rds_page_remainder_cpu_notify,
205};
206
207void rds_page_exit(void)
208{
209	int i;
210
211	for_each_possible_cpu(i)
212		rds_page_remainder_cpu_notify(&rds_page_remainder_nb,
213					      (unsigned long)CPU_DEAD,
214					      (void *)(long)i);
215}
216