blob: b2e6db075e4993eac4fcffb9004b82ee4a0bca71 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * xor.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1996, 1997, 1998, 1999, 2000,
5 * Ingo Molnar, Matti Aarnio, Jakub Jelinek, Richard Henderson.
6 *
7 * Dispatch optimized RAID-5 checksumming functions.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2, or (at your option)
12 * any later version.
13 *
14 * You should have received a copy of the GNU General Public License
15 * (for example /usr/src/linux/COPYING); if not, write to the Free
16 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
17 */
18
19#define BH_TRACE 0
20#include <linux/module.h>
21#include <linux/raid/md.h>
22#include <linux/raid/xor.h>
23#include <asm/xor.h>
24
25/* The xor routines to use. */
26static struct xor_block_template *active_template;
27
28void
Dan Williams9bc89cd2007-01-02 11:10:44 -070029xor_blocks(unsigned int src_count, unsigned int bytes, void *dest, void **srcs)
Linus Torvalds1da177e2005-04-16 15:20:36 -070030{
Dan Williams9bc89cd2007-01-02 11:10:44 -070031 unsigned long *p1, *p2, *p3, *p4;
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Dan Williams9bc89cd2007-01-02 11:10:44 -070033 p1 = (unsigned long *) srcs[0];
34 if (src_count == 1) {
35 active_template->do_2(bytes, dest, p1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 return;
37 }
38
Dan Williams9bc89cd2007-01-02 11:10:44 -070039 p2 = (unsigned long *) srcs[1];
40 if (src_count == 2) {
41 active_template->do_3(bytes, dest, p1, p2);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 return;
43 }
44
Dan Williams9bc89cd2007-01-02 11:10:44 -070045 p3 = (unsigned long *) srcs[2];
46 if (src_count == 3) {
47 active_template->do_4(bytes, dest, p1, p2, p3);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 return;
49 }
50
Dan Williams9bc89cd2007-01-02 11:10:44 -070051 p4 = (unsigned long *) srcs[3];
52 active_template->do_5(bytes, dest, p1, p2, p3, p4);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053}
Dan Williams685784a2007-07-09 11:56:42 -070054EXPORT_SYMBOL(xor_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/* Set of all registered templates. */
57static struct xor_block_template *template_list;
58
59#define BENCH_SIZE (PAGE_SIZE)
60
61static void
62do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
63{
64 int speed;
65 unsigned long now;
66 int i, count, max;
67
68 tmpl->next = template_list;
69 template_list = tmpl;
70
71 /*
72 * Count the number of XORs done during a whole jiffy, and use
73 * this to calculate the speed of checksumming. We use a 2-page
74 * allocation to have guaranteed color L1-cache layout.
75 */
76 max = 0;
77 for (i = 0; i < 5; i++) {
78 now = jiffies;
79 count = 0;
80 while (jiffies == now) {
Dan Williams685784a2007-07-09 11:56:42 -070081 mb(); /* prevent loop optimzation */
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 tmpl->do_2(BENCH_SIZE, b1, b2);
83 mb();
84 count++;
85 mb();
86 }
87 if (count > max)
88 max = count;
89 }
90
91 speed = max * (HZ * BENCH_SIZE / 1024);
92 tmpl->speed = speed;
93
Dan Williams685784a2007-07-09 11:56:42 -070094 printk(KERN_INFO " %-10s: %5d.%03d MB/sec\n", tmpl->name,
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 speed / 1000, speed % 1000);
96}
97
Dan Williams685784a2007-07-09 11:56:42 -070098static int __init
99calibrate_xor_blocks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
101 void *b1, *b2;
102 struct xor_block_template *f, *fastest;
103
104 b1 = (void *) __get_free_pages(GFP_KERNEL, 2);
Dan Williams685784a2007-07-09 11:56:42 -0700105 if (!b1) {
106 printk(KERN_WARNING "xor: Yikes! No memory available.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 return -ENOMEM;
108 }
109 b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
110
111 /*
Dan Williams685784a2007-07-09 11:56:42 -0700112 * If this arch/cpu has a short-circuited selection, don't loop through
113 * all the possible functions, just test the best one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115
116 fastest = NULL;
117
118#ifdef XOR_SELECT_TEMPLATE
119 fastest = XOR_SELECT_TEMPLATE(fastest);
120#endif
121
122#define xor_speed(templ) do_xor_speed((templ), b1, b2)
123
124 if (fastest) {
Dan Williams685784a2007-07-09 11:56:42 -0700125 printk(KERN_INFO "xor: automatically using best "
126 "checksumming function: %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 fastest->name);
128 xor_speed(fastest);
129 } else {
Dan Williams9bc89cd2007-01-02 11:10:44 -0700130 printk(KERN_INFO "xor: measuring software checksum speed\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 XOR_TRY_TEMPLATES;
132 fastest = template_list;
133 for (f = fastest; f; f = f->next)
134 if (f->speed > fastest->speed)
135 fastest = f;
136 }
137
Dan Williams685784a2007-07-09 11:56:42 -0700138 printk(KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 fastest->name, fastest->speed / 1000, fastest->speed % 1000);
140
141#undef xor_speed
142
143 free_pages((unsigned long)b1, 2);
144
145 active_template = fastest;
146 return 0;
147}
148
149static __exit void xor_exit(void) { }
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151MODULE_LICENSE("GPL");
152
Dan Williams685784a2007-07-09 11:56:42 -0700153/* when built-in xor.o must initialize before drivers/md/md.o */
154core_initcall(calibrate_xor_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155module_exit(xor_exit);