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/raid/md.h>
22 #include <linux/raid/xor.h>
23 #include <asm/xor.h>
24
25 /* The xor routines to use. */
26 static struct xor_block_template *active_template;
27
28 void
29 xor_block(unsigned int count, struct buffer_head **bh_ptr)
30 {
31 unsigned long *p0, *p1, *p2, *p3, *p4;
32 unsigned long bytes = bh_ptr[0]->b_size;
33
34 p0 = (unsigned long *) bh_ptr[0]->b_data;
35 p1 = (unsigned long *) bh_ptr[1]->b_data;
36 if (count == 2) {
37 active_template->do_2(bytes, p0, p1);
38 return;
39 }
40
41 p2 = (unsigned long *) bh_ptr[2]->b_data;
42 if (count == 3) {
43 active_template->do_3(bytes, p0, p1, p2);
44 return;
45 }
46
47 p3 = (unsigned long *) bh_ptr[3]->b_data;
48 if (count == 4) {
49 active_template->do_4(bytes, p0, p1, p2, p3);
50 return;
51 }
52
53 p4 = (unsigned long *) bh_ptr[4]->b_data;
54 active_template->do_5(bytes, p0, p1, p2, p3, p4);
55 }
56
57 /* Set of all registered templates. */
58 static struct xor_block_template *template_list;
59
60 /* The -6*32 shift factor colors the cache. */
61 #define BENCH_SIZE (PAGE_SIZE-6*32)
62
63 static void
64 do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
65 {
66 int speed;
67 unsigned long now;
68 int i, count, max;
69
70 tmpl->next = template_list;
71 template_list = tmpl;
72
73 /*
74 * Count the number of XORs done during a whole jiffy, and use
75 * this to calculate the speed of checksumming. We use a 2-page
76 * allocation to have guaranteed color L1-cache layout.
77 */
78 max = 0;
79 for (i = 0; i < 5; i++) {
80 now = jiffies;
81 count = 0;
82 while (jiffies == now) {
83 mb();
84 tmpl->do_2(BENCH_SIZE, b1, b2);
85 mb();
86 count++;
87 mb();
88 }
89 if (count > max)
90 max = count;
91 }
92
93 speed = max * (HZ * BENCH_SIZE / 1024);
94 tmpl->speed = speed;
95
96 printk(" %-10s: %5d.%03d MB/sec\n", tmpl->name,
97 speed / 1000, speed % 1000);
98 }
99
100 static int
101 calibrate_xor_block(void)
102 {
103 void *b1, *b2;
104 struct xor_block_template *f, *fastest;
105
106 b1 = (void *) md__get_free_pages(GFP_KERNEL, 2);
107 if (! b1) {
108 printk("raid5: Yikes! No memory available.\n");
109 return -ENOMEM;
110 }
111 b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
112
113 printk(KERN_INFO "raid5: measuring checksumming speed\n");
114 sti();
115
116 #define xor_speed(templ) do_xor_speed((templ), b1, b2)
117
118 XOR_TRY_TEMPLATES;
119
120 #undef xor_speed
121
122 free_pages((unsigned long)b1, 2);
123
124 fastest = template_list;
125 for (f = fastest; f; f = f->next)
126 if (f->speed > fastest->speed)
127 fastest = f;
128
129 #ifdef XOR_SELECT_TEMPLATE
130 fastest = XOR_SELECT_TEMPLATE(fastest);
131 #endif
132
133 active_template = fastest;
134 printk("raid5: using function: %s (%d.%03d MB/sec)\n",
135 fastest->name, fastest->speed / 1000, fastest->speed % 1000);
136
137 return 0;
138 }
139
140 MD_EXPORT_SYMBOL(xor_block);
141
142 module_init(calibrate_xor_block);
143
This page was automatically generated by the
LXR engine.
Visit the LXR main site for more
information.