summaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
Diffstat (limited to 'arch')
-rw-r--r--arch/x86/crypto/Makefile8
-rw-r--r--arch/x86/crypto/sha1_ssse3_asm.S558
-rw-r--r--arch/x86/crypto/sha1_ssse3_glue.c240
-rw-r--r--arch/x86/include/asm/cpufeature.h3
4 files changed, 809 insertions, 0 deletions
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index c04f1b7a9139..57c7f7b4436d 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -13,6 +13,7 @@ obj-$(CONFIG_CRYPTO_AES_NI_INTEL) += aesni-intel.o
obj-$(CONFIG_CRYPTO_GHASH_CLMUL_NI_INTEL) += ghash-clmulni-intel.o
obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
+obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
aes-i586-y := aes-i586-asm_32.o aes_glue.o
twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
@@ -25,3 +26,10 @@ salsa20-x86_64-y := salsa20-x86_64-asm_64.o salsa20_glue.o
aesni-intel-y := aesni-intel_asm.o aesni-intel_glue.o fpu.o
ghash-clmulni-intel-y := ghash-clmulni-intel_asm.o ghash-clmulni-intel_glue.o
+
+# enable AVX support only when $(AS) can actually assemble the instructions
+ifeq ($(call as-instr,vpxor %xmm0$(comma)%xmm1$(comma)%xmm2,yes,no),yes)
+AFLAGS_sha1_ssse3_asm.o += -DSHA1_ENABLE_AVX_SUPPORT
+CFLAGS_sha1_ssse3_glue.o += -DSHA1_ENABLE_AVX_SUPPORT
+endif
+sha1-ssse3-y := sha1_ssse3_asm.o sha1_ssse3_glue.o
diff --git a/arch/x86/crypto/sha1_ssse3_asm.S b/arch/x86/crypto/sha1_ssse3_asm.S
new file mode 100644
index 000000000000..b2c2f57d70e8
--- /dev/null
+++ b/arch/x86/crypto/sha1_ssse3_asm.S
@@ -0,0 +1,558 @@
+/*
+ * This is a SIMD SHA-1 implementation. It requires the Intel(R) Supplemental
+ * SSE3 instruction set extensions introduced in Intel Core Microarchitecture
+ * processors. CPUs supporting Intel(R) AVX extensions will get an additional
+ * boost.
+ *
+ * This work was inspired by the vectorized implementation of Dean Gaudet.
+ * Additional information on it can be found at:
+ * http://www.arctic.org/~dean/crypto/sha1.html
+ *
+ * It was improved upon with more efficient vectorization of the message
+ * scheduling. This implementation has also been optimized for all current and
+ * several future generations of Intel CPUs.
+ *
+ * See this article for more information about the implementation details:
+ * http://software.intel.com/en-us/articles/improving-the-performance-of-the-secure-hash-algorithm-1/
+ *
+ * Copyright (C) 2010, Intel Corp.
+ * Authors: Maxim Locktyukhin <maxim.locktyukhin@intel.com>
+ * Ronen Zohar <ronen.zohar@intel.com>
+ *
+ * Converted to AT&T syntax and adapted for inclusion in the Linux kernel:
+ * Author: Mathias Krause <minipli@googlemail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+
+#define CTX %rdi // arg1
+#define BUF %rsi // arg2
+#define CNT %rdx // arg3
+
+#define REG_A %ecx
+#define REG_B %esi
+#define REG_C %edi
+#define REG_D %ebp
+#define REG_E %edx
+
+#define REG_T1 %eax
+#define REG_T2 %ebx
+
+#define K_BASE %r8
+#define HASH_PTR %r9
+#define BUFFER_PTR %r10
+#define BUFFER_END %r11
+
+#define W_TMP1 %xmm0
+#define W_TMP2 %xmm9
+
+#define W0 %xmm1
+#define W4 %xmm2
+#define W8 %xmm3
+#define W12 %xmm4
+#define W16 %xmm5
+#define W20 %xmm6
+#define W24 %xmm7
+#define W28 %xmm8
+
+#define XMM_SHUFB_BSWAP %xmm10
+
+/* we keep window of 64 w[i]+K pre-calculated values in a circular buffer */
+#define WK(t) (((t) & 15) * 4)(%rsp)
+#define W_PRECALC_AHEAD 16
+
+/*
+ * This macro implements the SHA-1 function's body for single 64-byte block
+ * param: function's name
+ */
+.macro SHA1_VECTOR_ASM name
+ .global \name
+ .type \name, @function
+ .align 32
+\name:
+ push %rbx
+ push %rbp
+ push %r12
+
+ mov %rsp, %r12
+ sub $64, %rsp # allocate workspace
+ and $~15, %rsp # align stack
+
+ mov CTX, HASH_PTR
+ mov BUF, BUFFER_PTR
+
+ shl $6, CNT # multiply by 64
+ add BUF, CNT
+ mov CNT, BUFFER_END
+
+ lea K_XMM_AR(%rip), K_BASE
+ xmm_mov BSWAP_SHUFB_CTL(%rip), XMM_SHUFB_BSWAP
+
+ SHA1_PIPELINED_MAIN_BODY
+
+ # cleanup workspace
+ mov $8, %ecx
+ mov %rsp, %rdi
+ xor %rax, %rax
+ rep stosq
+
+ mov %r12, %rsp # deallocate workspace
+
+ pop %r12
+ pop %rbp
+ pop %rbx
+ ret
+
+ .size \name, .-\name
+.endm
+
+/*
+ * This macro implements 80 rounds of SHA-1 for one 64-byte block
+ */
+.macro SHA1_PIPELINED_MAIN_BODY
+ INIT_REGALLOC
+
+ mov (HASH_PTR), A
+ mov 4(HASH_PTR), B
+ mov 8(HASH_PTR), C
+ mov 12(HASH_PTR), D
+ mov 16(HASH_PTR), E
+
+ .set i, 0
+ .rept W_PRECALC_AHEAD
+ W_PRECALC i
+ .set i, (i+1)
+ .endr
+
+.align 4
+1:
+ RR F1,A,B,C,D,E,0
+ RR F1,D,E,A,B,C,2
+ RR F1,B,C,D,E,A,4
+ RR F1,E,A,B,C,D,6
+ RR F1,C,D,E,A,B,8
+
+ RR F1,A,B,C,D,E,10
+ RR F1,D,E,A,B,C,12
+ RR F1,B,C,D,E,A,14
+ RR F1,E,A,B,C,D,16
+ RR F1,C,D,E,A,B,18
+
+ RR F2,A,B,C,D,E,20
+ RR F2,D,E,A,B,C,22
+ RR F2,B,C,D,E,A,24
+ RR F2,E,A,B,C,D,26
+ RR F2,C,D,E,A,B,28
+
+ RR F2,A,B,C,D,E,30
+ RR F2,D,E,A,B,C,32
+ RR F2,B,C,D,E,A,34
+ RR F2,E,A,B,C,D,36
+ RR F2,C,D,E,A,B,38
+
+ RR F3,A,B,C,D,E,40
+ RR F3,D,E,A,B,C,42
+ RR F3,B,C,D,E,A,44
+ RR F3,E,A,B,C,D,46
+ RR F3,C,D,E,A,B,48
+
+ RR F3,A,B,C,D,E,50
+ RR F3,D,E,A,B,C,52
+ RR F3,B,C,D,E,A,54
+ RR F3,E,A,B,C,D,56
+ RR F3,C,D,E,A,B,58
+
+ add $64, BUFFER_PTR # move to the next 64-byte block
+ cmp BUFFER_END, BUFFER_PTR # if the current is the last one use
+ cmovae K_BASE, BUFFER_PTR # dummy source to avoid buffer overrun
+
+ RR F4,A,B,C,D,E,60
+ RR F4,D,E,A,B,C,62
+ RR F4,B,C,D,E,A,64
+ RR F4,E,A,B,C,D,66
+ RR F4,C,D,E,A,B,68
+
+ RR F4,A,B,C,D,E,70
+ RR F4,D,E,A,B,C,72
+ RR F4,B,C,D,E,A,74
+ RR F4,E,A,B,C,D,76
+ RR F4,C,D,E,A,B,78
+
+ UPDATE_HASH (HASH_PTR), A
+ UPDATE_HASH 4(HASH_PTR), B
+ UPDATE_HASH 8(HASH_PTR), C
+ UPDATE_HASH 12(HASH_PTR), D
+ UPDATE_HASH 16(HASH_PTR), E
+
+ RESTORE_RENAMED_REGS
+ cmp K_BASE, BUFFER_PTR # K_BASE means, we reached the end
+ jne 1b
+.endm
+
+.macro INIT_REGALLOC
+ .set A, REG_A
+ .set B, REG_B
+ .set C, REG_C
+ .set D, REG_D
+ .set E, REG_E
+ .set T1, REG_T1
+ .set T2, REG_T2
+.endm
+
+.macro RESTORE_RENAMED_REGS
+ # order is important (REG_C is where it should be)
+ mov B, REG_B
+ mov D, REG_D
+ mov A, REG_A
+ mov E, REG_E
+.endm
+
+.macro SWAP_REG_NAMES a, b
+ .set _T, \a
+ .set \a, \b
+ .set \b, _T
+.endm
+
+.macro F1 b, c, d
+ mov \c, T1
+ SWAP_REG_NAMES \c, T1
+ xor \d, T1
+ and \b, T1
+ xor \d, T1
+.endm
+
+.macro F2 b, c, d
+ mov \d, T1
+ SWAP_REG_NAMES \d, T1
+ xor \c, T1
+ xor \b, T1
+.endm
+
+.macro F3 b, c ,d
+ mov \c, T1
+ SWAP_REG_NAMES \c, T1
+ mov \b, T2
+ or \b, T1
+ and \c, T2
+ and \d, T1
+ or T2, T1
+.endm
+
+.macro F4 b, c, d
+ F2 \b, \c, \d
+.endm
+
+.macro UPDATE_HASH hash, val
+ add \hash, \val
+ mov \val, \hash
+.endm
+
+/*
+ * RR does two rounds of SHA-1 back to back with W[] pre-calc
+ * t1 = F(b, c, d); e += w(i)
+ * e += t1; b <<= 30; d += w(i+1);
+ * t1 = F(a, b, c);
+ * d += t1; a <<= 5;
+ * e += a;
+ * t1 = e; a >>= 7;
+ * t1 <<= 5;
+ * d += t1;
+ */
+.macro RR F, a, b, c, d, e, round
+ add WK(\round), \e
+ \F \b, \c, \d # t1 = F(b, c, d);
+ W_PRECALC (\round + W_PRECALC_AHEAD)
+ rol $30, \b
+ add T1, \e
+ add WK(\round + 1), \d
+
+ \F \a, \b, \c
+ W_PRECALC (\round + W_PRECALC_AHEAD + 1)
+ rol $5, \a
+ add \a, \e
+ add T1, \d
+ ror $7, \a # (a <<r 5) >>r 7) => a <<r 30)
+
+ mov \e, T1
+ SWAP_REG_NAMES \e, T1
+
+ rol $5, T1
+ add T1, \d
+
+ # write: \a, \b
+ # rotate: \a<=\d, \b<=\e, \c<=\a, \d<=\b, \e<=\c
+.endm
+
+.macro W_PRECALC r
+ .set i, \r
+
+ .if (i < 20)
+ .set K_XMM, 0
+ .elseif (i < 40)
+ .set K_XMM, 16
+ .elseif (i < 60)
+ .set K_XMM, 32
+ .elseif (i < 80)
+ .set K_XMM, 48
+ .endif
+
+ .if ((i < 16) || ((i >= 80) && (i < (80 + W_PRECALC_AHEAD))))
+ .set i, ((\r) % 80) # pre-compute for the next iteration
+ .if (i == 0)
+ W_PRECALC_RESET
+ .endif
+ W_PRECALC_00_15
+ .elseif (i<32)
+ W_PRECALC_16_31
+ .elseif (i < 80) // rounds 32-79
+ W_PRECALC_32_79
+ .endif
+.endm
+
+.macro W_PRECALC_RESET
+ .set W, W0
+ .set W_minus_04, W4
+ .set W_minus_08, W8
+ .set W_minus_12, W12
+ .set W_minus_16, W16
+ .set W_minus_20, W20
+ .set W_minus_24, W24
+ .set W_minus_28, W28
+ .set W_minus_32, W
+.endm
+
+.macro W_PRECALC_ROTATE
+ .set W_minus_32, W_minus_28
+ .set W_minus_28, W_minus_24
+ .set W_minus_24, W_minus_20
+ .set W_minus_20, W_minus_16
+ .set W_minus_16, W_minus_12
+ .set W_minus_12, W_minus_08
+ .set W_minus_08, W_minus_04
+ .set W_minus_04, W
+ .set W, W_minus_32
+.endm
+
+.macro W_PRECALC_SSSE3
+
+.macro W_PRECALC_00_15
+ W_PRECALC_00_15_SSSE3
+.endm
+.macro W_PRECALC_16_31
+ W_PRECALC_16_31_SSSE3
+.endm
+.macro W_PRECALC_32_79
+ W_PRECALC_32_79_SSSE3
+.endm
+
+/* message scheduling pre-compute for rounds 0-15 */
+.macro W_PRECALC_00_15_SSSE3
+ .if ((i & 3) == 0)
+ movdqu (i*4)(BUFFER_PTR), W_TMP1
+ .elseif ((i & 3) == 1)
+ pshufb XMM_SHUFB_BSWAP, W_TMP1
+ movdqa W_TMP1, W
+ .elseif ((i & 3) == 2)
+ paddd (K_BASE), W_TMP1
+ .elseif ((i & 3) == 3)
+ movdqa W_TMP1, WK(i&~3)
+ W_PRECALC_ROTATE
+ .endif
+.endm
+
+/* message scheduling pre-compute for rounds 16-31
+ *
+ * - calculating last 32 w[i] values in 8 XMM registers
+ * - pre-calculate K+w[i] values and store to mem, for later load by ALU add
+ * instruction
+ *
+ * some "heavy-lifting" vectorization for rounds 16-31 due to w[i]->w[i-3]
+ * dependency, but improves for 32-79
+ */
+.macro W_PRECALC_16_31_SSSE3
+ # blended scheduling of vector and scalar instruction streams, one 4-wide
+ # vector iteration / 4 scalar rounds
+ .if ((i & 3) == 0)
+ movdqa W_minus_12, W
+ palignr $8, W_minus_16, W # w[i-14]
+ movdqa W_minus_04, W_TMP1
+ psrldq $4, W_TMP1 # w[i-3]
+ pxor W_minus_08, W
+ .elseif ((i & 3) == 1)
+ pxor W_minus_16, W_TMP1
+ pxor W_TMP1, W
+ movdqa W, W_TMP2
+ movdqa W, W_TMP1
+ pslldq $12, W_TMP2
+ .elseif ((i & 3) == 2)
+ psrld $31, W
+ pslld $1, W_TMP1
+ por W, W_TMP1
+ movdqa W_TMP2, W
+ psrld $30, W_TMP2
+ pslld $2, W
+ .elseif ((i & 3) == 3)
+ pxor W, W_TMP1
+ pxor W_TMP2, W_TMP1
+ movdqa W_TMP1, W
+ paddd K_XMM(K_BASE), W_TMP1
+ movdqa W_TMP1, WK(i&~3)
+ W_PRECALC_ROTATE
+ .endif
+.endm
+
+/* message scheduling pre-compute for rounds 32-79
+ *
+ * in SHA-1 specification: w[i] = (w[i-3] ^ w[i-8] ^ w[i-14] ^ w[i-16]) rol 1
+ * instead we do equal: w[i] = (w[i-6] ^ w[i-16] ^ w[i-28] ^ w[i-32]) rol 2
+ * allows more efficient vectorization since w[i]=>w[i-3] dependency is broken
+ */
+.macro W_PRECALC_32_79_SSSE3
+ .if ((i & 3) == 0)
+ movdqa W_minus_04, W_TMP1
+ pxor W_minus_28, W # W is W_minus_32 before xor
+ palignr $8, W_minus_08, W_TMP1
+ .elseif ((i & 3) == 1)
+ pxor W_minus_16, W
+ pxor W_TMP1, W
+ movdqa W, W_TMP1
+ .elseif ((i & 3) == 2)
+ psrld $30, W
+ pslld $2, W_TMP1
+ por W, W_TMP1
+ .elseif ((i & 3) == 3)
+ movdqa W_TMP1, W
+ paddd K_XMM(K_BASE), W_TMP1
+ movdqa W_TMP1, WK(i&~3)
+ W_PRECALC_ROTATE
+ .endif
+.endm
+
+.endm // W_PRECALC_SSSE3
+
+
+#define K1 0x5a827999
+#define K2 0x6ed9eba1
+#define K3 0x8f1bbcdc
+#define K4 0xca62c1d6
+
+.section .rodata
+.align 16
+
+K_XMM_AR:
+ .long K1, K1, K1, K1
+ .long K2, K2, K2, K2
+ .long K3, K3, K3, K3
+ .long K4, K4, K4, K4
+
+BSWAP_SHUFB_CTL:
+ .long 0x00010203
+ .long 0x04050607
+ .long 0x08090a0b
+ .long 0x0c0d0e0f
+
+
+.section .text
+
+W_PRECALC_SSSE3
+.macro xmm_mov a, b
+ movdqu \a,\b
+.endm
+
+/* SSSE3 optimized implementation:
+ * extern "C" void sha1_transform_ssse3(u32 *digest, const char *data, u32 *ws,
+ * unsigned int rounds);
+ */
+SHA1_VECTOR_ASM sha1_transform_ssse3
+
+#ifdef SHA1_ENABLE_AVX_SUPPORT
+
+.macro W_PRECALC_AVX
+
+.purgem W_PRECALC_00_15
+.macro W_PRECALC_00_15
+ W_PRECALC_00_15_AVX
+.endm
+.purgem W_PRECALC_16_31
+.macro W_PRECALC_16_31
+ W_PRECALC_16_31_AVX
+.endm
+.purgem W_PRECALC_32_79
+.macro W_PRECALC_32_79
+ W_PRECALC_32_79_AVX
+.endm
+
+.macro W_PRECALC_00_15_AVX
+ .if ((i & 3) == 0)
+ vmovdqu (i*4)(BUFFER_PTR), W_TMP1
+ .elseif ((i & 3) == 1)
+ vpshufb XMM_SHUFB_BSWAP, W_TMP1, W
+ .elseif ((i & 3) == 2)
+ vpaddd (K_BASE), W, W_TMP1
+ .elseif ((i & 3) == 3)
+ vmovdqa W_TMP1, WK(i&~3)
+ W_PRECALC_ROTATE
+ .endif
+.endm
+
+.macro W_PRECALC_16_31_AVX
+ .if ((i & 3) == 0)
+ vpalignr $8, W_minus_16, W_minus_12, W # w[i-14]
+ vpsrldq $4, W_minus_04, W_TMP1 # w[i-3]
+ vpxor W_minus_08, W, W
+ vpxor W_minus_16, W_TMP1, W_TMP1
+ .elseif ((i & 3) == 1)
+ vpxor W_TMP1, W, W
+ vpslldq $12, W, W_TMP2
+ vpslld $1, W, W_TMP1
+ .elseif ((i & 3) == 2)
+ vpsrld $31, W, W
+ vpor W, W_TMP1, W_TMP1
+ vpslld $2, W_TMP2, W
+ vpsrld $30, W_TMP2, W_TMP2
+ .elseif ((i & 3) == 3)
+ vpxor W, W_TMP1, W_TMP1
+ vpxor W_TMP2, W_TMP1, W
+ vpaddd K_XMM(K_BASE), W, W_TMP1
+ vmovdqu W_TMP1, WK(i&~3)
+ W_PRECALC_ROTATE
+ .endif
+.endm
+
+.macro W_PRECALC_32_79_AVX
+ .if ((i & 3) == 0)
+ vpalignr $8, W_minus_08, W_minus_04, W_TMP1
+ vpxor W_minus_28, W, W # W is W_minus_32 before xor
+ .elseif ((i & 3) == 1)
+ vpxor W_minus_16, W_TMP1, W_TMP1
+ vpxor W_TMP1, W, W
+ .elseif ((i & 3) == 2)
+ vpslld $2, W, W_TMP1
+ vpsrld $30, W, W
+ vpor W, W_TMP1, W
+ .elseif ((i & 3) == 3)
+ vpaddd K_XMM(K_BASE), W, W_TMP1
+ vmovdqu W_TMP1, WK(i&~3)
+ W_PRECALC_ROTATE
+ .endif
+.endm
+
+.endm // W_PRECALC_AVX
+
+W_PRECALC_AVX
+.purgem xmm_mov
+.macro xmm_mov a, b
+ vmovdqu \a,\b
+.endm
+
+
+/* AVX optimized implementation:
+ * extern "C" void sha1_transform_avx(u32 *digest, const char *data, u32 *ws,
+ * unsigned int rounds);
+ */
+SHA1_VECTOR_ASM sha1_transform_avx
+
+#endif
diff --git a/arch/x86/crypto/sha1_ssse3_glue.c b/arch/x86/crypto/sha1_ssse3_glue.c
new file mode 100644
index 000000000000..f916499d0abe
--- /dev/null
+++ b/arch/x86/crypto/sha1_ssse3_glue.c
@@ -0,0 +1,240 @@
+/*
+ * Cryptographic API.
+ *
+ * Glue code for the SHA1 Secure Hash Algorithm assembler implementation using
+ * Supplemental SSE3 instructions.
+ *
+ * This file is based on sha1_generic.c
+ *
+ * Copyright (c) Alan Smithee.
+ * Copyright (c) Andrew McDonald <andrew@mcdonald.org.uk>
+ * Copyright (c) Jean-Francois Dive <jef@linuxbe.org>
+ * Copyright (c) Mathias Krause <minipli@googlemail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/cryptohash.h>
+#include <linux/types.h>
+#include <crypto/sha.h>
+#include <asm/byteorder.h>
+#include <asm/i387.h>
+#include <asm/xcr.h>
+#include <asm/xsave.h>
+
+
+asmlinkage void sha1_transform_ssse3(u32 *digest, const char *data,
+ unsigned int rounds);
+#ifdef SHA1_ENABLE_AVX_SUPPORT
+asmlinkage void sha1_transform_avx(u32 *digest, const char *data,
+ unsigned int rounds);
+#endif
+
+static asmlinkage void (*sha1_transform_asm)(u32 *, const char *, unsigned int);
+
+
+static int sha1_ssse3_init(struct shash_desc *desc)
+{
+ struct sha1_state *sctx = shash_desc_ctx(desc);
+
+ *sctx = (struct sha1_state){
+ .state = { SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 },
+ };
+
+ return 0;
+}
+
+static int __sha1_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len, unsigned int partial)
+{
+ struct sha1_state *sctx = shash_desc_ctx(desc);
+ unsigned int done = 0;
+
+ sctx->count += len;
+
+ if (partial) {
+ done = SHA1_BLOCK_SIZE - partial;
+ memcpy(sctx->buffer + partial, data, done);
+ sha1_transform_asm(sctx->state, sctx->buffer, 1);
+ }
+
+ if (len - done >= SHA1_BLOCK_SIZE) {
+ const unsigned int rounds = (len - done) / SHA1_BLOCK_SIZE;
+
+ sha1_transform_asm(sctx->state, data + done, rounds);
+ done += rounds * SHA1_BLOCK_SIZE;
+ }
+
+ memcpy(sctx->buffer, data + done, len - done);
+
+ return 0;
+}
+
+static int sha1_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
+{
+ struct sha1_state *sctx = shash_desc_ctx(desc);
+ unsigned int partial = sctx->count % SHA1_BLOCK_SIZE;
+ int res;
+
+ /* Handle the fast case right here */
+ if (partial + len < SHA1_BLOCK_SIZE) {
+ sctx->count += len;
+ memcpy(sctx->buffer + partial, data, len);
+
+ return 0;
+ }
+
+ if (!irq_fpu_usable()) {
+ res = crypto_sha1_update(desc, data, len);
+ } else {
+ kernel_fpu_begin();
+ res = __sha1_ssse3_update(desc, data, len, partial);
+ kernel_fpu_end();
+ }
+
+ return res;
+}
+
+
+/* Add padding and return the message digest. */
+static int sha1_ssse3_final(struct shash_desc *desc, u8 *out)
+{
+ struct sha1_state *sctx = shash_desc_ctx(desc);
+ unsigned int i, index, padlen;
+ __be32 *dst = (__be32 *)out;
+ __be64 bits;
+ static const u8 padding[SHA1_BLOCK_SIZE] = { 0x80, };
+
+ bits = cpu_to_be64(sctx->count << 3);
+
+ /* Pad out to 56 mod 64 and append length */
+ index = sctx->count % SHA1_BLOCK_SIZE;
+ padlen = (index < 56) ? (56 - index) : ((SHA1_BLOCK_SIZE+56) - index);
+ if (!irq_fpu_usable()) {
+ crypto_sha1_update(desc, padding, padlen);
+ crypto_sha1_update(desc, (const u8 *)&bits, sizeof(bits));
+ } else {
+ kernel_fpu_begin();
+ /* We need to fill a whole block for __sha1_ssse3_update() */
+ if (padlen <= 56) {
+ sctx->count += padlen;
+ memcpy(sctx->buffer + index, padding, padlen);
+ } else {
+ __sha1_ssse3_update(desc, padding, padlen, index);
+ }
+ __sha1_ssse3_update(desc, (const u8 *)&bits, sizeof(bits), 56);
+ kernel_fpu_end();
+ }
+
+ /* Store state in digest */
+ for (i = 0; i < 5; i++)
+ dst[i] = cpu_to_be32(sctx->state[i]);
+
+ /* Wipe context */
+ memset(sctx, 0, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha1_ssse3_export(struct shash_desc *desc, void *out)
+{
+ struct sha1_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(out, sctx, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha1_ssse3_import(struct shash_desc *desc, const void *in)
+{
+ struct sha1_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(sctx, in, sizeof(*sctx));
+
+ return 0;
+}
+
+static struct shash_alg alg = {
+ .digestsize = SHA1_DIGEST_SIZE,
+ .init = sha1_ssse3_init,
+ .update = sha1_ssse3_update,
+ .final = sha1_ssse3_final,
+ .export = sha1_ssse3_export,
+ .import = sha1_ssse3_import,
+ .descsize = sizeof(struct sha1_state),
+ .statesize = sizeof(struct sha1_state),
+ .base = {
+ .cra_name = "sha1",
+ .cra_driver_name= "sha1-ssse3",
+ .cra_priority = 150,
+ .cra_flags = CRYPTO_ALG_TYPE_SHASH,
+ .cra_blocksize = SHA1_BLOCK_SIZE,
+ .cra_module = THIS_MODULE,
+ }
+};
+
+#ifdef SHA1_ENABLE_AVX_SUPPORT
+static bool __init avx_usable(void)
+{
+ u64 xcr0;
+
+ if (!cpu_has_avx || !cpu_has_osxsave)
+ return false;
+
+ xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
+ if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
+ pr_info("AVX detected but unusable.\n");
+
+ return false;
+ }
+
+ return true;
+}
+#endif
+
+static int __init sha1_ssse3_mod_init(void)
+{
+ /* test for SSSE3 first */
+ if (cpu_has_ssse3)
+ sha1_transform_asm = sha1_transform_ssse3;
+
+#ifdef SHA1_ENABLE_AVX_SUPPORT
+ /* allow AVX to override SSSE3, it's a little faster */
+ if (avx_usable())
+ sha1_transform_asm = sha1_transform_avx;
+#endif
+
+ if (sha1_transform_asm) {
+ pr_info("Using %s optimized SHA-1 implementation\n",
+ sha1_transform_asm == sha1_transform_ssse3 ? "SSSE3"
+ : "AVX");
+ return crypto_register_shash(&alg);
+ }
+ pr_info("Neither AVX nor SSSE3 is available/usable.\n");
+
+ return -ENODEV;
+}
+
+static void __exit sha1_ssse3_mod_fini(void)
+{
+ crypto_unregister_shash(&alg);
+}
+
+module_init(sha1_ssse3_mod_init);
+module_exit(sha1_ssse3_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("SHA1 Secure Hash Algorithm, Supplemental SSE3 accelerated");
+
+MODULE_ALIAS("sha1");
diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h
index 4258aac99a6e..48a93ef5c84b 100644
--- a/arch/x86/include/asm/cpufeature.h
+++ b/arch/x86/include/asm/cpufeature.h
@@ -257,7 +257,9 @@ extern const char * const x86_power_flags[32];
#define cpu_has_xmm boot_cpu_has(X86_FEATURE_XMM)
#define cpu_has_xmm2 boot_cpu_has(X86_FEATURE_XMM2)
#define cpu_has_xmm3 boot_cpu_has(X86_FEATURE_XMM3)
+#define cpu_has_ssse3 boot_cpu_has(X86_FEATURE_SSSE3)
#define cpu_has_aes boot_cpu_has(X86_FEATURE_AES)
+#define cpu_has_avx boot_cpu_has(X86_FEATURE_AVX)
#define cpu_has_ht boot_cpu_has(X86_FEATURE_HT)
#define cpu_has_mp boot_cpu_has(X86_FEATURE_MP)
#define cpu_has_nx boot_cpu_has(X86_FEATURE_NX)
@@ -285,6 +287,7 @@ extern const char * const x86_power_flags[32];
#define cpu_has_xmm4_2 boot_cpu_has(X86_FEATURE_XMM4_2)
#define cpu_has_x2apic boot_cpu_has(X86_FEATURE_X2APIC)
#define cpu_has_xsave boot_cpu_has(X86_FEATURE_XSAVE)
+#define cpu_has_osxsave boot_cpu_has(X86_FEATURE_OSXSAVE)
#define cpu_has_hypervisor boot_cpu_has(X86_FEATURE_HYPERVISOR)
#define cpu_has_pclmulqdq boot_cpu_has(X86_FEATURE_PCLMULQDQ)
#define cpu_has_perfctr_core boot_cpu_has(X86_FEATURE_PERFCTR_CORE)
OpenPOWER on IntegriCloud