| 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 |
| 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 | #ifndef MODULE |
| 58 | static void __init do_xor_register(struct xor_block_template *tmpl) |
| 59 | { |
| 60 | tmpl->next = template_list; |
| 61 | template_list = tmpl; |
| 62 | } |
| 63 | |
| 64 | static int __init register_xor_blocks(void) |
| 65 | { |
| 66 | active_template = XOR_SELECT_TEMPLATE(NULL); |
| 67 | |
| 68 | if (!active_template) { |
| 69 | #define xor_speed do_xor_register |
| 70 | // register all the templates and pick the first as the default |
| 71 | XOR_TRY_TEMPLATES; |
| 72 | #undef xor_speed |
| 73 | active_template = template_list; |
| 74 | } |
| 75 | return 0; |
| 76 | } |
| 77 | #endif |
| 78 | |
| 79 | #define BENCH_SIZE 4096 |
| 80 | #define REPS 800U |
| 81 | |
| 82 | static void __init |
| 83 | do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2) |
| 84 | { |
| 85 | int speed; |
| 86 | int i, j; |
| 87 | ktime_t min, start, diff; |
| 88 | |
| 89 | tmpl->next = template_list; |
| 90 | template_list = tmpl; |
| 91 | |
| 92 | preempt_disable(); |
| 93 | |
| 94 | min = (ktime_t)S64_MAX; |
| 95 | for (i = 0; i < 3; i++) { |
| 96 | start = ktime_get(); |
| 97 | for (j = 0; j < REPS; j++) { |
| 98 | mb(); /* prevent loop optimization */ |
| 99 | tmpl->do_2(BENCH_SIZE, b1, b2); |
| 100 | mb(); |
| 101 | } |
| 102 | diff = ktime_sub(ktime_get(), start); |
| 103 | if (diff < min) |
| 104 | min = diff; |
| 105 | } |
| 106 | |
| 107 | preempt_enable(); |
| 108 | |
| 109 | // bytes/ns == GB/s, multiply by 1000 to get MB/s [not MiB/s] |
| 110 | if (!min) |
| 111 | min = 1; |
| 112 | speed = (1000 * REPS * BENCH_SIZE) / (unsigned int)ktime_to_ns(min); |
| 113 | tmpl->speed = speed; |
| 114 | |
| 115 | pr_info(" %-16s: %5d MB/sec\n", tmpl->name, speed); |
| 116 | } |
| 117 | |
| 118 | static int __init |
| 119 | calibrate_xor_blocks(void) |
| 120 | { |
| 121 | void *b1, *b2; |
| 122 | struct xor_block_template *f, *fastest; |
| 123 | |
| 124 | fastest = XOR_SELECT_TEMPLATE(NULL); |
| 125 | |
| 126 | if (fastest) { |
| 127 | printk(KERN_INFO "xor: automatically using best " |
| 128 | "checksumming function %-10s\n", |
| 129 | fastest->name); |
| 130 | goto out; |
| 131 | } |
| 132 | |
| 133 | b1 = (void *) __get_free_pages(GFP_KERNEL, 2); |
| 134 | if (!b1) { |
| 135 | printk(KERN_WARNING "xor: Yikes! No memory available.\n"); |
| 136 | return -ENOMEM; |
| 137 | } |
| 138 | b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE; |
| 139 | |
| 140 | /* |
| 141 | * If this arch/cpu has a short-circuited selection, don't loop through |
| 142 | * all the possible functions, just test the best one |
| 143 | */ |
| 144 | |
| 145 | #define xor_speed(templ) do_xor_speed((templ), b1, b2) |
| 146 | |
| 147 | printk(KERN_INFO "xor: measuring software checksum speed\n"); |
| 148 | template_list = NULL; |
| 149 | XOR_TRY_TEMPLATES; |
| 150 | fastest = template_list; |
| 151 | for (f = fastest; f; f = f->next) |
| 152 | if (f->speed > fastest->speed) |
| 153 | fastest = f; |
| 154 | |
| 155 | pr_info("xor: using function: %s (%d MB/sec)\n", |
| 156 | fastest->name, fastest->speed); |
| 157 | |
| 158 | #undef xor_speed |
| 159 | |
| 160 | free_pages((unsigned long)b1, 2); |
| 161 | out: |
| 162 | active_template = fastest; |
| 163 | return 0; |
| 164 | } |
| 165 | |
| 166 | static __exit void xor_exit(void) { } |
| 167 | |
| 168 | MODULE_LICENSE("GPL"); |
| 169 | |
| 170 | #ifndef MODULE |
| 171 | /* when built-in xor.o must initialize before drivers/md/md.o */ |
| 172 | core_initcall(register_xor_blocks); |
| 173 | #endif |
| 174 | |
| 175 | module_init(calibrate_xor_blocks); |
| 176 | module_exit(xor_exit); |