Loading...
1/*
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/gfp.h>
22#include <linux/raid/xor.h>
23#include <linux/jiffies.h>
24#include <linux/preempt.h>
25#include <asm/xor.h>
26
27/* The xor routines to use. */
28static struct xor_block_template *active_template;
29
30void
31xor_blocks(unsigned int src_count, unsigned int bytes, void *dest, void **srcs)
32{
33 unsigned long *p1, *p2, *p3, *p4;
34
35 p1 = (unsigned long *) srcs[0];
36 if (src_count == 1) {
37 active_template->do_2(bytes, dest, p1);
38 return;
39 }
40
41 p2 = (unsigned long *) srcs[1];
42 if (src_count == 2) {
43 active_template->do_3(bytes, dest, p1, p2);
44 return;
45 }
46
47 p3 = (unsigned long *) srcs[2];
48 if (src_count == 3) {
49 active_template->do_4(bytes, dest, p1, p2, p3);
50 return;
51 }
52
53 p4 = (unsigned long *) srcs[3];
54 active_template->do_5(bytes, dest, p1, p2, p3, p4);
55}
56EXPORT_SYMBOL(xor_blocks);
57
58/* Set of all registered templates. */
59static struct xor_block_template *__initdata template_list;
60
61#define BENCH_SIZE (PAGE_SIZE)
62
63static void __init
64do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
65{
66 int speed;
67 unsigned long now, j;
68 int i, count, max;
69
70 tmpl->next = template_list;
71 template_list = tmpl;
72
73 preempt_disable();
74
75 /*
76 * Count the number of XORs done during a whole jiffy, and use
77 * this to calculate the speed of checksumming. We use a 2-page
78 * allocation to have guaranteed color L1-cache layout.
79 */
80 max = 0;
81 for (i = 0; i < 5; i++) {
82 j = jiffies;
83 count = 0;
84 while ((now = jiffies) == j)
85 cpu_relax();
86 while (time_before(jiffies, now + 1)) {
87 mb(); /* prevent loop optimzation */
88 tmpl->do_2(BENCH_SIZE, b1, b2);
89 mb();
90 count++;
91 mb();
92 }
93 if (count > max)
94 max = count;
95 }
96
97 preempt_enable();
98
99 speed = max * (HZ * BENCH_SIZE / 1024);
100 tmpl->speed = speed;
101
102 printk(KERN_INFO " %-10s: %5d.%03d MB/sec\n", tmpl->name,
103 speed / 1000, speed % 1000);
104}
105
106static int __init
107calibrate_xor_blocks(void)
108{
109 void *b1, *b2;
110 struct xor_block_template *f, *fastest;
111
112 /*
113 * Note: Since the memory is not actually used for _anything_ but to
114 * test the XOR speed, we don't really want kmemcheck to warn about
115 * reading uninitialized bytes here.
116 */
117 b1 = (void *) __get_free_pages(GFP_KERNEL | __GFP_NOTRACK, 2);
118 if (!b1) {
119 printk(KERN_WARNING "xor: Yikes! No memory available.\n");
120 return -ENOMEM;
121 }
122 b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
123
124 /*
125 * If this arch/cpu has a short-circuited selection, don't loop through
126 * all the possible functions, just test the best one
127 */
128
129 fastest = NULL;
130
131#ifdef XOR_SELECT_TEMPLATE
132 fastest = XOR_SELECT_TEMPLATE(fastest);
133#endif
134
135#define xor_speed(templ) do_xor_speed((templ), b1, b2)
136
137 if (fastest) {
138 printk(KERN_INFO "xor: automatically using best "
139 "checksumming function:\n");
140 xor_speed(fastest);
141 goto out;
142 } else {
143 printk(KERN_INFO "xor: measuring software checksum speed\n");
144 XOR_TRY_TEMPLATES;
145 fastest = template_list;
146 for (f = fastest; f; f = f->next)
147 if (f->speed > fastest->speed)
148 fastest = f;
149 }
150
151 printk(KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n",
152 fastest->name, fastest->speed / 1000, fastest->speed % 1000);
153
154#undef xor_speed
155
156 out:
157 free_pages((unsigned long)b1, 2);
158
159 active_template = fastest;
160 return 0;
161}
162
163static __exit void xor_exit(void) { }
164
165MODULE_LICENSE("GPL");
166
167/* when built-in xor.o must initialize before drivers/md/md.o */
168core_initcall(calibrate_xor_blocks);
169module_exit(xor_exit);
1/*
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/gfp.h>
22#include <linux/raid/xor.h>
23#include <linux/jiffies.h>
24#include <linux/preempt.h>
25#include <asm/xor.h>
26
27#ifndef XOR_SELECT_TEMPLATE
28#define XOR_SELECT_TEMPLATE(x) (x)
29#endif
30
31/* The xor routines to use. */
32static struct xor_block_template *active_template;
33
34void
35xor_blocks(unsigned int src_count, unsigned int bytes, void *dest, void **srcs)
36{
37 unsigned long *p1, *p2, *p3, *p4;
38
39 p1 = (unsigned long *) srcs[0];
40 if (src_count == 1) {
41 active_template->do_2(bytes, dest, p1);
42 return;
43 }
44
45 p2 = (unsigned long *) srcs[1];
46 if (src_count == 2) {
47 active_template->do_3(bytes, dest, p1, p2);
48 return;
49 }
50
51 p3 = (unsigned long *) srcs[2];
52 if (src_count == 3) {
53 active_template->do_4(bytes, dest, p1, p2, p3);
54 return;
55 }
56
57 p4 = (unsigned long *) srcs[3];
58 active_template->do_5(bytes, dest, p1, p2, p3, p4);
59}
60EXPORT_SYMBOL(xor_blocks);
61
62/* Set of all registered templates. */
63static struct xor_block_template *__initdata template_list;
64
65#define BENCH_SIZE (PAGE_SIZE)
66
67static void __init
68do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
69{
70 int speed;
71 unsigned long now, j;
72 int i, count, max;
73
74 tmpl->next = template_list;
75 template_list = tmpl;
76
77 preempt_disable();
78
79 /*
80 * Count the number of XORs done during a whole jiffy, and use
81 * this to calculate the speed of checksumming. We use a 2-page
82 * allocation to have guaranteed color L1-cache layout.
83 */
84 max = 0;
85 for (i = 0; i < 5; i++) {
86 j = jiffies;
87 count = 0;
88 while ((now = jiffies) == j)
89 cpu_relax();
90 while (time_before(jiffies, now + 1)) {
91 mb(); /* prevent loop optimzation */
92 tmpl->do_2(BENCH_SIZE, b1, b2);
93 mb();
94 count++;
95 mb();
96 }
97 if (count > max)
98 max = count;
99 }
100
101 preempt_enable();
102
103 speed = max * (HZ * BENCH_SIZE / 1024);
104 tmpl->speed = speed;
105
106 printk(KERN_INFO " %-10s: %5d.%03d MB/sec\n", tmpl->name,
107 speed / 1000, speed % 1000);
108}
109
110static int __init
111calibrate_xor_blocks(void)
112{
113 void *b1, *b2;
114 struct xor_block_template *f, *fastest;
115
116 fastest = XOR_SELECT_TEMPLATE(NULL);
117
118 if (fastest) {
119 printk(KERN_INFO "xor: automatically using best "
120 "checksumming function %-10s\n",
121 fastest->name);
122 goto out;
123 }
124
125 b1 = (void *) __get_free_pages(GFP_KERNEL, 2);
126 if (!b1) {
127 printk(KERN_WARNING "xor: Yikes! No memory available.\n");
128 return -ENOMEM;
129 }
130 b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
131
132 /*
133 * If this arch/cpu has a short-circuited selection, don't loop through
134 * all the possible functions, just test the best one
135 */
136
137#define xor_speed(templ) do_xor_speed((templ), b1, b2)
138
139 printk(KERN_INFO "xor: measuring software checksum speed\n");
140 XOR_TRY_TEMPLATES;
141 fastest = template_list;
142 for (f = fastest; f; f = f->next)
143 if (f->speed > fastest->speed)
144 fastest = f;
145
146 printk(KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n",
147 fastest->name, fastest->speed / 1000, fastest->speed % 1000);
148
149#undef xor_speed
150
151 free_pages((unsigned long)b1, 2);
152out:
153 active_template = fastest;
154 return 0;
155}
156
157static __exit void xor_exit(void) { }
158
159MODULE_LICENSE("GPL");
160
161/* when built-in xor.o must initialize before drivers/md/md.o */
162core_initcall(calibrate_xor_blocks);
163module_exit(xor_exit);