1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * xor.c : Multiple Devices driver for Linux
4 *
5 * Copyright (C) 1996, 1997, 1998, 1999, 2000,
6 * Ingo Molnar, Matti Aarnio, Jakub Jelinek, Richard Henderson.
7 *
8 * Dispatch optimized RAID-5 checksumming functions.
9 */
10
11 #define BH_TRACE 0
12 #include <linux/module.h>
13 #include <linux/gfp.h>
14 #include <linux/raid/xor.h>
15 #include <linux/jiffies.h>
16 #include <linux/preempt.h>
17 #include <asm/xor.h>
18
19 #ifndef XOR_SELECT_TEMPLATE
20 #define XOR_SELECT_TEMPLATE(x) (x)
21 #endif
22
23 /* The xor routines to use. */
24 static struct xor_block_template *active_template;
25
26 void
xor_blocks(unsigned int src_count,unsigned int bytes,void * dest,void ** srcs)27 xor_blocks(unsigned int src_count, unsigned int bytes, void *dest, void **srcs)
28 {
29 unsigned long *p1, *p2, *p3, *p4;
30
31 p1 = (unsigned long *) srcs[0];
32 if (src_count == 1) {
33 active_template->do_2(bytes, dest, p1);
34 return;
35 }
36
37 p2 = (unsigned long *) srcs[1];
38 if (src_count == 2) {
39 active_template->do_3(bytes, dest, p1, p2);
40 return;
41 }
42
43 p3 = (unsigned long *) srcs[2];
44 if (src_count == 3) {
45 active_template->do_4(bytes, dest, p1, p2, p3);
46 return;
47 }
48
49 p4 = (unsigned long *) srcs[3];
50 active_template->do_5(bytes, dest, p1, p2, p3, p4);
51 }
52 EXPORT_SYMBOL(xor_blocks);
53
54 /* Set of all registered templates. */
55 static struct xor_block_template *__initdata template_list;
56
57 #define BENCH_SIZE (PAGE_SIZE)
58
59 static void __init
do_xor_speed(struct xor_block_template * tmpl,void * b1,void * b2)60 do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
61 {
62 int speed;
63 unsigned long now, j;
64 int i, count, max;
65
66 tmpl->next = template_list;
67 template_list = tmpl;
68
69 preempt_disable();
70
71 /*
72 * Count the number of XORs done during a whole jiffy, and use
73 * this to calculate the speed of checksumming. We use a 2-page
74 * allocation to have guaranteed color L1-cache layout.
75 */
76 max = 0;
77 for (i = 0; i < 5; i++) {
78 j = jiffies;
79 count = 0;
80 while ((now = jiffies) == j)
81 cpu_relax();
82 while (time_before(jiffies, now + 1)) {
83 mb(); /* prevent loop optimzation */
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 preempt_enable();
94
95 speed = max * (HZ * BENCH_SIZE / 1024);
96 tmpl->speed = speed;
97
98 printk(KERN_INFO " %-10s: %5d.%03d MB/sec\n", tmpl->name,
99 speed / 1000, speed % 1000);
100 }
101
102 static int __init
calibrate_xor_blocks(void)103 calibrate_xor_blocks(void)
104 {
105 void *b1, *b2;
106 struct xor_block_template *f, *fastest;
107
108 fastest = XOR_SELECT_TEMPLATE(NULL);
109
110 if (fastest) {
111 printk(KERN_INFO "xor: automatically using best "
112 "checksumming function %-10s\n",
113 fastest->name);
114 goto out;
115 }
116
117 b1 = (void *) __get_free_pages(GFP_KERNEL, 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 #define xor_speed(templ) do_xor_speed((templ), b1, b2)
130
131 printk(KERN_INFO "xor: measuring software checksum speed\n");
132 XOR_TRY_TEMPLATES;
133 fastest = template_list;
134 for (f = fastest; f; f = f->next)
135 if (f->speed > fastest->speed)
136 fastest = f;
137
138 printk(KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n",
139 fastest->name, fastest->speed / 1000, fastest->speed % 1000);
140
141 #undef xor_speed
142
143 free_pages((unsigned long)b1, 2);
144 out:
145 active_template = fastest;
146 return 0;
147 }
148
xor_exit(void)149 static __exit void xor_exit(void) { }
150
151 MODULE_LICENSE("GPL");
152
153 /* when built-in xor.o must initialize before drivers/md/md.o */
154 core_initcall(calibrate_xor_blocks);
155 module_exit(xor_exit);
156
