]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 | 22 | #include <linux/raid/xor.h> |
bff61975 | 23 | #include <linux/jiffies.h> |
56a51991 | 24 | #include <linux/preempt.h> |
1da177e4 LT |
25 | #include <asm/xor.h> |
26 | ||
27 | /* The xor routines to use. */ | |
28 | static struct xor_block_template *active_template; | |
29 | ||
30 | void | |
9bc89cd8 | 31 | xor_blocks(unsigned int src_count, unsigned int bytes, void *dest, void **srcs) |
1da177e4 | 32 | { |
9bc89cd8 | 33 | unsigned long *p1, *p2, *p3, *p4; |
1da177e4 | 34 | |
9bc89cd8 DW |
35 | p1 = (unsigned long *) srcs[0]; |
36 | if (src_count == 1) { | |
37 | active_template->do_2(bytes, dest, p1); | |
1da177e4 LT |
38 | return; |
39 | } | |
40 | ||
9bc89cd8 DW |
41 | p2 = (unsigned long *) srcs[1]; |
42 | if (src_count == 2) { | |
43 | active_template->do_3(bytes, dest, p1, p2); | |
1da177e4 LT |
44 | return; |
45 | } | |
46 | ||
9bc89cd8 DW |
47 | p3 = (unsigned long *) srcs[2]; |
48 | if (src_count == 3) { | |
49 | active_template->do_4(bytes, dest, p1, p2, p3); | |
1da177e4 LT |
50 | return; |
51 | } | |
52 | ||
9bc89cd8 DW |
53 | p4 = (unsigned long *) srcs[3]; |
54 | active_template->do_5(bytes, dest, p1, p2, p3, p4); | |
1da177e4 | 55 | } |
685784aa | 56 | EXPORT_SYMBOL(xor_blocks); |
1da177e4 LT |
57 | |
58 | /* Set of all registered templates. */ | |
af7cf25d | 59 | static struct xor_block_template *__initdata template_list; |
1da177e4 LT |
60 | |
61 | #define BENCH_SIZE (PAGE_SIZE) | |
62 | ||
af7cf25d | 63 | static void __init |
1da177e4 LT |
64 | do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2) |
65 | { | |
66 | int speed; | |
6a328475 | 67 | unsigned long now, j; |
1da177e4 LT |
68 | int i, count, max; |
69 | ||
70 | tmpl->next = template_list; | |
71 | template_list = tmpl; | |
72 | ||
56a51991 JK |
73 | preempt_disable(); |
74 | ||
1da177e4 LT |
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++) { | |
6a328475 | 82 | j = jiffies; |
1da177e4 | 83 | count = 0; |
6a328475 JK |
84 | while ((now = jiffies) == j) |
85 | cpu_relax(); | |
86 | while (time_before(jiffies, now + 1)) { | |
685784aa | 87 | mb(); /* prevent loop optimzation */ |
1da177e4 LT |
88 | tmpl->do_2(BENCH_SIZE, b1, b2); |
89 | mb(); | |
90 | count++; | |
91 | mb(); | |
92 | } | |
93 | if (count > max) | |
94 | max = count; | |
95 | } | |
96 | ||
56a51991 JK |
97 | preempt_enable(); |
98 | ||
1da177e4 LT |
99 | speed = max * (HZ * BENCH_SIZE / 1024); |
100 | tmpl->speed = speed; | |
101 | ||
685784aa | 102 | printk(KERN_INFO " %-10s: %5d.%03d MB/sec\n", tmpl->name, |
1da177e4 LT |
103 | speed / 1000, speed % 1000); |
104 | } | |
105 | ||
685784aa DW |
106 | static int __init |
107 | calibrate_xor_blocks(void) | |
1da177e4 LT |
108 | { |
109 | void *b1, *b2; | |
110 | struct xor_block_template *f, *fastest; | |
111 | ||
33f65df7 VN |
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); | |
685784aa DW |
118 | if (!b1) { |
119 | printk(KERN_WARNING "xor: Yikes! No memory available.\n"); | |
1da177e4 LT |
120 | return -ENOMEM; |
121 | } | |
122 | b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE; | |
123 | ||
124 | /* | |
685784aa DW |
125 | * If this arch/cpu has a short-circuited selection, don't loop through |
126 | * all the possible functions, just test the best one | |
1da177e4 LT |
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) { | |
685784aa | 138 | printk(KERN_INFO "xor: automatically using best " |
d788fec8 | 139 | "checksumming function:\n"); |
1da177e4 | 140 | xor_speed(fastest); |
d788fec8 | 141 | goto out; |
1da177e4 | 142 | } else { |
9bc89cd8 | 143 | printk(KERN_INFO "xor: measuring software checksum speed\n"); |
1da177e4 LT |
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 | ||
685784aa | 151 | printk(KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n", |
1da177e4 LT |
152 | fastest->name, fastest->speed / 1000, fastest->speed % 1000); |
153 | ||
154 | #undef xor_speed | |
155 | ||
d788fec8 | 156 | out: |
1da177e4 LT |
157 | free_pages((unsigned long)b1, 2); |
158 | ||
159 | active_template = fastest; | |
160 | return 0; | |
161 | } | |
162 | ||
163 | static __exit void xor_exit(void) { } | |
164 | ||
1da177e4 LT |
165 | MODULE_LICENSE("GPL"); |
166 | ||
685784aa DW |
167 | /* when built-in xor.o must initialize before drivers/md/md.o */ |
168 | core_initcall(calibrate_xor_blocks); | |
1da177e4 | 169 | module_exit(xor_exit); |