blob: 263af9fb45ea281c94c41d5698d4d264d489bba9
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. */ |
32 | static struct xor_block_template *active_template; |
33 | |
34 | void |
35 | xor_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 | } |
60 | EXPORT_SYMBOL(xor_blocks); |
61 | |
62 | /* Set of all registered templates. */ |
63 | static struct xor_block_template *__initdata template_list; |
64 | |
65 | #define BENCH_SIZE (PAGE_SIZE) |
66 | |
67 | static void __init |
68 | do_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 | |
110 | static int __init |
111 | calibrate_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 | /* |
126 | * Note: Since the memory is not actually used for _anything_ but to |
127 | * test the XOR speed, we don't really want kmemcheck to warn about |
128 | * reading uninitialized bytes here. |
129 | */ |
130 | b1 = (void *) __get_free_pages(GFP_KERNEL | __GFP_NOTRACK, 2); |
131 | if (!b1) { |
132 | printk(KERN_WARNING "xor: Yikes! No memory available.\n"); |
133 | return -ENOMEM; |
134 | } |
135 | b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE; |
136 | |
137 | /* |
138 | * If this arch/cpu has a short-circuited selection, don't loop through |
139 | * all the possible functions, just test the best one |
140 | */ |
141 | |
142 | #define xor_speed(templ) do_xor_speed((templ), b1, b2) |
143 | |
144 | printk(KERN_INFO "xor: measuring software checksum speed\n"); |
145 | XOR_TRY_TEMPLATES; |
146 | fastest = template_list; |
147 | for (f = fastest; f; f = f->next) |
148 | if (f->speed > fastest->speed) |
149 | fastest = f; |
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 | free_pages((unsigned long)b1, 2); |
157 | out: |
158 | active_template = fastest; |
159 | return 0; |
160 | } |
161 | |
162 | static __exit void xor_exit(void) { } |
163 | |
164 | MODULE_LICENSE("GPL"); |
165 | |
166 | /* when built-in xor.o must initialize before drivers/md/md.o */ |
167 | core_initcall(calibrate_xor_blocks); |
168 | module_exit(xor_exit); |
169 |