crypto/xor.c

   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 <asm/xor.h>
  25
  26 /* The xor routines to use.  */
  27 static struct xor_block_template *active_template;
  28
  29 void
  30 xor_blocks(unsigned int src_count, unsigned int bytes, void *dest, void **srcs)
  31 {
  32         unsigned long *p1, *p2, *p3, *p4;
  33
  34         p1 = (unsigned long *) srcs[0];
  35         if (src_count == 1) {
  36                 active_template->do_2(bytes, dest, p1);
  37                 return;
  38         }
  39
  40         p2 = (unsigned long *) srcs[1];
  41         if (src_count == 2) {
  42                 active_template->do_3(bytes, dest, p1, p2);
  43                 return;
  44         }
  45
  46         p3 = (unsigned long *) srcs[2];
  47         if (src_count == 3) {
  48                 active_template->do_4(bytes, dest, p1, p2, p3);
  49                 return;
  50         }
  51
  52         p4 = (unsigned long *) srcs[3];
  53         active_template->do_5(bytes, dest, p1, p2, p3, p4);
  54 }
  55 EXPORT_SYMBOL(xor_blocks);
  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
  63 do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
  64 {
  65         int speed;
  66         unsigned long now, j;
  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                 j = jiffies;
  80                 count = 0;
  81                 while ((now = jiffies) == j)
  82                         cpu_relax();
  83                 while (time_before(jiffies, now + 1)) {
  84                         mb(); /* prevent loop optimzation */
  85                         tmpl->do_2(BENCH_SIZE, b1, b2);
  86                         mb();
  87                         count++;
  88                         mb();
  89                 }
  90                 if (count > max)
  91                         max = count;
  92         }
  93
  94         speed = max * (HZ * BENCH_SIZE / 1024);
  95         tmpl->speed = speed;
  96
  97         printk(KERN_INFO "   %-10s: %5d.%03d MB/sec\n", tmpl->name,
  98                speed / 1000, speed % 1000);
  99 }
 100
 101 static int __init
 102 calibrate_xor_blocks(void)
 103 {
 104         void *b1, *b2;
 105         struct xor_block_template *f, *fastest;
 106
 107         /*
 108          * Note: Since the memory is not actually used for _anything_ but to
 109          * test the XOR speed, we don't really want kmemcheck to warn about
 110          * reading uninitialized bytes here.
 111          */
 112         b1 = (void *) __get_free_pages(GFP_KERNEL | __GFP_NOTRACK, 2);
 113         if (!b1) {
 114                 printk(KERN_WARNING "xor: Yikes!  No memory available.\n");
 115                 return -ENOMEM;
 116         }
 117         b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
 118
 119         /*
 120          * If this arch/cpu has a short-circuited selection, don't loop through
 121          * all the possible functions, just test the best one
 122          */
 123
 124         fastest = NULL;
 125
 126 #ifdef XOR_SELECT_TEMPLATE
 127                 fastest = XOR_SELECT_TEMPLATE(fastest);
 128 #endif
 129
 130 #define xor_speed(templ)        do_xor_speed((templ), b1, b2)
 131
 132         if (fastest) {
 133                 printk(KERN_INFO "xor: automatically using best "
 134                         "checksumming function: %s\n",
 135                         fastest->name);
 136                 xor_speed(fastest);
 137         } else {
 138                 printk(KERN_INFO "xor: measuring software checksum speed\n");
 139                 XOR_TRY_TEMPLATES;
 140                 fastest = template_list;
 141                 for (f = fastest; f; f = f->next)
 142                         if (f->speed > fastest->speed)
 143                                 fastest = f;
 144         }
 145
 146         printk(KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n",
 147                fastest->name, fastest->speed / 1000, fastest->speed % 1000);
 148
 149 #undef xor_speed
 150
 151         free_pages((unsigned long)b1, 2);
 152
 153         active_template = fastest;
 154         return 0;
 155 }
 156
 157 static __exit void xor_exit(void) { }
 158
 159 MODULE_LICENSE("GPL");
 160
 161 /* when built-in xor.o must initialize before drivers/md/md.o */
 162 core_initcall(calibrate_xor_blocks);
 163 module_exit(xor_exit);