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
xor_block(unsigned int count,struct buffer_head ** bh_ptr)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 #define BENCH_SIZE (PAGE_SIZE)
61
62 static void
do_xor_speed(struct xor_block_template * tmpl,void * b1,void * b2)63 do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
64 {
65 int speed;
66 unsigned long now;
67 int i, count, max;
68
69 tmpl->next = template_list;
70 template_list = tmpl;
71
72 /*
73 * Count the number of XORs done during a whole jiffy, and use
74 * this to calculate the speed of checksumming. We use a 2-page
75 * allocation to have guaranteed color L1-cache layout.
76 */
77 max = 0;
78 for (i = 0; i < 5; i++) {
79 now = jiffies;
80 count = 0;
81 while (jiffies == now) {
82 mb();
83 tmpl->do_2(BENCH_SIZE, b1, b2);
84 mb();
85 count++;
86 mb();
87 }
88 if (count > max)
89 max = count;
90 }
91
92 speed = max * (HZ * BENCH_SIZE / 1024);
93 tmpl->speed = speed;
94
95 printk(" %-10s: %5d.%03d MB/sec\n", tmpl->name,
96 speed / 1000, speed % 1000);
97 }
98
99 static int
calibrate_xor_block(void)100 calibrate_xor_block(void)
101 {
102 void *b1, *b2;
103 struct xor_block_template *f, *fastest;
104
105 b1 = (void *) md__get_free_pages(GFP_KERNEL, 2);
106 if (! b1) {
107 printk("raid5: Yikes! No memory available.\n");
108 return -ENOMEM;
109 }
110 b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
111
112 printk(KERN_INFO "raid5: measuring checksumming speed\n");
113 sti();
114
115 #define xor_speed(templ) do_xor_speed((templ), b1, b2)
116
117 XOR_TRY_TEMPLATES;
118
119 #undef xor_speed
120
121 free_pages((unsigned long)b1, 2);
122
123 fastest = template_list;
124 for (f = fastest; f; f = f->next)
125 if (f->speed > fastest->speed)
126 fastest = f;
127
128 #ifdef XOR_SELECT_TEMPLATE
129 fastest = XOR_SELECT_TEMPLATE(fastest);
130 #endif
131
132 active_template = fastest;
133 printk("raid5: using function: %s (%d.%03d MB/sec)\n",
134 fastest->name, fastest->speed / 1000, fastest->speed % 1000);
135
136 return 0;
137 }
138
139 MD_EXPORT_SYMBOL(xor_block);
140 MODULE_LICENSE("GPL");
141
142 module_init(calibrate_xor_block);
143