aboutsummaryrefslogtreecommitdiffstats
path: root/lib
diff options
context:
space:
mode:
authorMandeep Singh Baines <msb@chromium.org>2011-08-05 21:46:27 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2011-08-06 14:26:52 -0400
commit1eb19a12bd2214cdcad5273d472b062a4ba97fa1 (patch)
treea88b76f31a4ea5e3631581be249759fcf4b05a08 /lib
parentde96355c111679dd6e2c5c73e25e814c72510c58 (diff)
lib/sha1: use the git implementation of SHA-1
For ChromiumOS, we use SHA-1 to verify the integrity of the root filesystem. The speed of the kernel sha-1 implementation has a major impact on our boot performance. To improve boot performance, we investigated using the heavily optimized sha-1 implementation used in git. With the git sha-1 implementation, we see a 11.7% improvement in boot time. 10 reboots, remove slowest/fastest. Before: Mean: 6.58 seconds Stdev: 0.14 After (with git sha-1, this patch): Mean: 5.89 seconds Stdev: 0.07 The other cool thing about the git SHA-1 implementation is that it only needs 64 bytes of stack for the workspace while the original kernel implementation needed 320 bytes. Signed-off-by: Mandeep Singh Baines <msb@chromium.org> Cc: Ramsay Jones <ramsay@ramsay1.demon.co.uk> Cc: Nicolas Pitre <nico@cam.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: David S. Miller <davem@davemloft.net> Cc: linux-crypto@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'lib')
-rw-r--r--lib/sha1.c212
1 files changed, 158 insertions, 54 deletions
diff --git a/lib/sha1.c b/lib/sha1.c
index 4c45fd50e913..f33271dd00cb 100644
--- a/lib/sha1.c
+++ b/lib/sha1.c
@@ -1,31 +1,72 @@
1/* 1/*
2 * SHA transform algorithm, originally taken from code written by 2 * SHA1 routine optimized to do word accesses rather than byte accesses,
3 * Peter Gutmann, and placed in the public domain. 3 * and to avoid unnecessary copies into the context array.
4 *
5 * This was based on the git SHA1 implementation.
4 */ 6 */
5 7
6#include <linux/kernel.h> 8#include <linux/kernel.h>
7#include <linux/module.h> 9#include <linux/module.h>
8#include <linux/cryptohash.h> 10#include <linux/bitops.h>
11#include <asm/unaligned.h>
9 12
10/* The SHA f()-functions. */ 13/*
14 * If you have 32 registers or more, the compiler can (and should)
15 * try to change the array[] accesses into registers. However, on
16 * machines with less than ~25 registers, that won't really work,
17 * and at least gcc will make an unholy mess of it.
18 *
19 * So to avoid that mess which just slows things down, we force
20 * the stores to memory to actually happen (we might be better off
21 * with a 'W(t)=(val);asm("":"+m" (W(t))' there instead, as
22 * suggested by Artur Skawina - that will also make gcc unable to
23 * try to do the silly "optimize away loads" part because it won't
24 * see what the value will be).
25 *
26 * Ben Herrenschmidt reports that on PPC, the C version comes close
27 * to the optimized asm with this (ie on PPC you don't want that
28 * 'volatile', since there are lots of registers).
29 *
30 * On ARM we get the best code generation by forcing a full memory barrier
31 * between each SHA_ROUND, otherwise gcc happily get wild with spilling and
32 * the stack frame size simply explode and performance goes down the drain.
33 */
11 34
12#define f1(x,y,z) (z ^ (x & (y ^ z))) /* x ? y : z */ 35#ifdef CONFIG_X86
13#define f2(x,y,z) (x ^ y ^ z) /* XOR */ 36 #define setW(x, val) (*(volatile __u32 *)&W(x) = (val))
14#define f3(x,y,z) ((x & y) + (z & (x ^ y))) /* majority */ 37#elif defined(CONFIG_ARM)
38 #define setW(x, val) do { W(x) = (val); __asm__("":::"memory"); } while (0)
39#else
40 #define setW(x, val) (W(x) = (val))
41#endif
15 42
16/* The SHA Mysterious Constants */ 43/* This "rolls" over the 512-bit array */
44#define W(x) (array[(x)&15])
17 45
18#define K1 0x5A827999L /* Rounds 0-19: sqrt(2) * 2^30 */ 46/*
19#define K2 0x6ED9EBA1L /* Rounds 20-39: sqrt(3) * 2^30 */ 47 * Where do we get the source from? The first 16 iterations get it from
20#define K3 0x8F1BBCDCL /* Rounds 40-59: sqrt(5) * 2^30 */ 48 * the input data, the next mix it from the 512-bit array.
21#define K4 0xCA62C1D6L /* Rounds 60-79: sqrt(10) * 2^30 */ 49 */
50#define SHA_SRC(t) get_unaligned_be32((__u32 *)data + t)
51#define SHA_MIX(t) rol32(W(t+13) ^ W(t+8) ^ W(t+2) ^ W(t), 1)
52
53#define SHA_ROUND(t, input, fn, constant, A, B, C, D, E) do { \
54 __u32 TEMP = input(t); setW(t, TEMP); \
55 E += TEMP + rol32(A,5) + (fn) + (constant); \
56 B = ror32(B, 2); } while (0)
57
58#define T_0_15(t, A, B, C, D, E) SHA_ROUND(t, SHA_SRC, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
59#define T_16_19(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
60#define T_20_39(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0x6ed9eba1, A, B, C, D, E )
61#define T_40_59(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, ((B&C)+(D&(B^C))) , 0x8f1bbcdc, A, B, C, D, E )
62#define T_60_79(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0xca62c1d6, A, B, C, D, E )
22 63
23/** 64/**
24 * sha_transform - single block SHA1 transform 65 * sha_transform - single block SHA1 transform
25 * 66 *
26 * @digest: 160 bit digest to update 67 * @digest: 160 bit digest to update
27 * @data: 512 bits of data to hash 68 * @data: 512 bits of data to hash
28 * @W: 80 words of workspace (see note) 69 * @array: 16 words of workspace (see note)
29 * 70 *
30 * This function generates a SHA1 digest for a single 512-bit block. 71 * This function generates a SHA1 digest for a single 512-bit block.
31 * Be warned, it does not handle padding and message digest, do not 72 * Be warned, it does not handle padding and message digest, do not
@@ -36,47 +77,111 @@
36 * to clear the workspace. This is left to the caller to avoid 77 * to clear the workspace. This is left to the caller to avoid
37 * unnecessary clears between chained hashing operations. 78 * unnecessary clears between chained hashing operations.
38 */ 79 */
39void sha_transform(__u32 *digest, const char *in, __u32 *W) 80void sha_transform(__u32 *digest, const char *data, __u32 *array)
40{ 81{
41 __u32 a, b, c, d, e, t, i; 82 __u32 A, B, C, D, E;
42 83
43 for (i = 0; i < 16; i++) 84 A = digest[0];
44 W[i] = be32_to_cpu(((const __be32 *)in)[i]); 85 B = digest[1];
45 86 C = digest[2];
46 for (i = 0; i < 64; i++) 87 D = digest[3];
47 W[i+16] = rol32(W[i+13] ^ W[i+8] ^ W[i+2] ^ W[i], 1); 88 E = digest[4];
48 89
49 a = digest[0]; 90 /* Round 1 - iterations 0-16 take their input from 'data' */
50 b = digest[1]; 91 T_0_15( 0, A, B, C, D, E);
51 c = digest[2]; 92 T_0_15( 1, E, A, B, C, D);
52 d = digest[3]; 93 T_0_15( 2, D, E, A, B, C);
53 e = digest[4]; 94 T_0_15( 3, C, D, E, A, B);
54 95 T_0_15( 4, B, C, D, E, A);
55 for (i = 0; i < 20; i++) { 96 T_0_15( 5, A, B, C, D, E);
56 t = f1(b, c, d) + K1 + rol32(a, 5) + e + W[i]; 97 T_0_15( 6, E, A, B, C, D);
57 e = d; d = c; c = rol32(b, 30); b = a; a = t; 98 T_0_15( 7, D, E, A, B, C);
58 } 99 T_0_15( 8, C, D, E, A, B);
59 100 T_0_15( 9, B, C, D, E, A);
60 for (; i < 40; i ++) { 101 T_0_15(10, A, B, C, D, E);
61 t = f2(b, c, d) + K2 + rol32(a, 5) + e + W[i]; 102 T_0_15(11, E, A, B, C, D);
62 e = d; d = c; c = rol32(b, 30); b = a; a = t; 103 T_0_15(12, D, E, A, B, C);
63 } 104 T_0_15(13, C, D, E, A, B);
64 105 T_0_15(14, B, C, D, E, A);
65 for (; i < 60; i ++) { 106 T_0_15(15, A, B, C, D, E);
66 t = f3(b, c, d) + K3 + rol32(a, 5) + e + W[i]; 107
67 e = d; d = c; c = rol32(b, 30); b = a; a = t; 108 /* Round 1 - tail. Input from 512-bit mixing array */
68 } 109 T_16_19(16, E, A, B, C, D);
69 110 T_16_19(17, D, E, A, B, C);
70 for (; i < 80; i ++) { 111 T_16_19(18, C, D, E, A, B);
71 t = f2(b, c, d) + K4 + rol32(a, 5) + e + W[i]; 112 T_16_19(19, B, C, D, E, A);
72 e = d; d = c; c = rol32(b, 30); b = a; a = t; 113
73 } 114 /* Round 2 */
74 115 T_20_39(20, A, B, C, D, E);
75 digest[0] += a; 116 T_20_39(21, E, A, B, C, D);
76 digest[1] += b; 117 T_20_39(22, D, E, A, B, C);
77 digest[2] += c; 118 T_20_39(23, C, D, E, A, B);
78 digest[3] += d; 119 T_20_39(24, B, C, D, E, A);
79 digest[4] += e; 120 T_20_39(25, A, B, C, D, E);
121 T_20_39(26, E, A, B, C, D);
122 T_20_39(27, D, E, A, B, C);
123 T_20_39(28, C, D, E, A, B);
124 T_20_39(29, B, C, D, E, A);
125 T_20_39(30, A, B, C, D, E);
126 T_20_39(31, E, A, B, C, D);
127 T_20_39(32, D, E, A, B, C);
128 T_20_39(33, C, D, E, A, B);
129 T_20_39(34, B, C, D, E, A);
130 T_20_39(35, A, B, C, D, E);
131 T_20_39(36, E, A, B, C, D);
132 T_20_39(37, D, E, A, B, C);
133 T_20_39(38, C, D, E, A, B);
134 T_20_39(39, B, C, D, E, A);
135
136 /* Round 3 */
137 T_40_59(40, A, B, C, D, E);
138 T_40_59(41, E, A, B, C, D);
139 T_40_59(42, D, E, A, B, C);
140 T_40_59(43, C, D, E, A, B);
141 T_40_59(44, B, C, D, E, A);
142 T_40_59(45, A, B, C, D, E);
143 T_40_59(46, E, A, B, C, D);
144 T_40_59(47, D, E, A, B, C);
145 T_40_59(48, C, D, E, A, B);
146 T_40_59(49, B, C, D, E, A);
147 T_40_59(50, A, B, C, D, E);
148 T_40_59(51, E, A, B, C, D);
149 T_40_59(52, D, E, A, B, C);
150 T_40_59(53, C, D, E, A, B);
151 T_40_59(54, B, C, D, E, A);
152 T_40_59(55, A, B, C, D, E);
153 T_40_59(56, E, A, B, C, D);
154 T_40_59(57, D, E, A, B, C);
155 T_40_59(58, C, D, E, A, B);
156 T_40_59(59, B, C, D, E, A);
157
158 /* Round 4 */
159 T_60_79(60, A, B, C, D, E);
160 T_60_79(61, E, A, B, C, D);
161 T_60_79(62, D, E, A, B, C);
162 T_60_79(63, C, D, E, A, B);
163 T_60_79(64, B, C, D, E, A);
164 T_60_79(65, A, B, C, D, E);
165 T_60_79(66, E, A, B, C, D);
166 T_60_79(67, D, E, A, B, C);
167 T_60_79(68, C, D, E, A, B);
168 T_60_79(69, B, C, D, E, A);
169 T_60_79(70, A, B, C, D, E);
170 T_60_79(71, E, A, B, C, D);
171 T_60_79(72, D, E, A, B, C);
172 T_60_79(73, C, D, E, A, B);
173 T_60_79(74, B, C, D, E, A);
174 T_60_79(75, A, B, C, D, E);
175 T_60_79(76, E, A, B, C, D);
176 T_60_79(77, D, E, A, B, C);
177 T_60_79(78, C, D, E, A, B);
178 T_60_79(79, B, C, D, E, A);
179
180 digest[0] += A;
181 digest[1] += B;
182 digest[2] += C;
183 digest[3] += D;
184 digest[4] += E;
80} 185}
81EXPORT_SYMBOL(sha_transform); 186EXPORT_SYMBOL(sha_transform);
82 187
@@ -92,4 +197,3 @@ void sha_init(__u32 *buf)
92 buf[3] = 0x10325476; 197 buf[3] = 0x10325476;
93 buf[4] = 0xc3d2e1f0; 198 buf[4] = 0xc3d2e1f0;
94} 199}
95