crypto: crc32c - Optimize CRC32C calculation with PCLMULQDQ instruction
authorTim Chen <tim.c.chen@linux.intel.com>
Thu, 27 Sep 2012 22:44:22 +0000 (15:44 -0700)
committerHerbert Xu <herbert@gondor.apana.org.au>
Mon, 15 Oct 2012 14:18:24 +0000 (22:18 +0800)
This patch adds the crc_pcl function that calculates CRC32C checksum using the
PCLMULQDQ instruction on processors that support this feature. This will
provide speedup over using CRC32 instruction only.
The usage of PCLMULQDQ necessitate the invocation of kernel_fpu_begin and
kernel_fpu_end and incur some overhead.  So the new crc_pcl function is only
invoked for buffer size of 512 bytes or more.  Larger sized
buffers will expect to see greater speedup.  This feature is best used coupled
with eager_fpu which reduces the kernel_fpu_begin/end overhead.  For
buffer size of 1K the speedup is around 1.6x and for buffer size greater than
4K, the speedup is around 3x compared to original implementation in crc32c-intel
module. Test was performed on Sandy Bridge based platform with constant frequency
set for cpu.

A white paper detailing the algorithm can be found here:
http://download.intel.com/design/intarch/papers/323405.pdf

Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
arch/x86/crypto/Makefile
arch/x86/crypto/crc32c-intel_glue.c
arch/x86/crypto/crc32c-pcl-intel-asm_64.S [new file with mode: 0644]
crypto/Kconfig

index b95778311178ba9234375da617b400bc2cea86af..84d7dbaba26e37d88489a6994ba36a2374788785 100644 (file)
@@ -48,3 +48,4 @@ 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
 sha1-ssse3-y := sha1_ssse3_asm.o sha1_ssse3_glue.o
 crc32c-intel-y := crc32c-intel_glue.o
+crc32c-intel-$(CONFIG_CRYPTO_CRC32C_X86_64) += crc32c-pcl-intel-asm_64.o
index 493f959261f7274eaf69729a7add9a2664d475fd..6812ad98355c3d0e93be5e07c47e5e3956ebc8a4 100644 (file)
@@ -32,6 +32,8 @@
 
 #include <asm/cpufeature.h>
 #include <asm/cpu_device_id.h>
+#include <asm/i387.h>
+#include <asm/fpu-internal.h>
 
 #define CHKSUM_BLOCK_SIZE      1
 #define CHKSUM_DIGEST_SIZE     4
 #define REX_PRE
 #endif
 
+#ifdef CONFIG_X86_64
+/*
+ * use carryless multiply version of crc32c when buffer
+ * size is >= 512 (when eager fpu is enabled) or
+ * >= 1024 (when eager fpu is disabled) to account
+ * for fpu state save/restore overhead.
+ */
+#define CRC32C_PCL_BREAKEVEN_EAGERFPU  512
+#define CRC32C_PCL_BREAKEVEN_NOEAGERFPU        1024
+
+asmlinkage unsigned int crc_pcl(const u8 *buffer, int len,
+                               unsigned int crc_init);
+static int crc32c_pcl_breakeven = CRC32C_PCL_BREAKEVEN_EAGERFPU;
+#if defined(X86_FEATURE_EAGER_FPU)
+#define set_pcl_breakeven_point()                                      \
+do {                                                                   \
+       if (!use_eager_fpu())                                           \
+               crc32c_pcl_breakeven = CRC32C_PCL_BREAKEVEN_NOEAGERFPU; \
+} while (0)
+#else
+#define set_pcl_breakeven_point()                                      \
+       (crc32c_pcl_breakeven = CRC32C_PCL_BREAKEVEN_NOEAGERFPU)
+#endif
+#endif /* CONFIG_X86_64 */
+
 static u32 crc32c_intel_le_hw_byte(u32 crc, unsigned char const *data, size_t length)
 {
        while (length--) {
@@ -154,6 +181,52 @@ static int crc32c_intel_cra_init(struct crypto_tfm *tfm)
        return 0;
 }
 
+#ifdef CONFIG_X86_64
+static int crc32c_pcl_intel_update(struct shash_desc *desc, const u8 *data,
+                              unsigned int len)
+{
+       u32 *crcp = shash_desc_ctx(desc);
+
+       /*
+        * use faster PCL version if datasize is large enough to
+        * overcome kernel fpu state save/restore overhead
+        */
+       if (len >= crc32c_pcl_breakeven && irq_fpu_usable()) {
+               kernel_fpu_begin();
+               *crcp = crc_pcl(data, len, *crcp);
+               kernel_fpu_end();
+       } else
+               *crcp = crc32c_intel_le_hw(*crcp, data, len);
+       return 0;
+}
+
+static int __crc32c_pcl_intel_finup(u32 *crcp, const u8 *data, unsigned int len,
+                               u8 *out)
+{
+       if (len >= crc32c_pcl_breakeven && irq_fpu_usable()) {
+               kernel_fpu_begin();
+               *(__le32 *)out = ~cpu_to_le32(crc_pcl(data, len, *crcp));
+               kernel_fpu_end();
+       } else
+               *(__le32 *)out =
+                       ~cpu_to_le32(crc32c_intel_le_hw(*crcp, data, len));
+       return 0;
+}
+
+static int crc32c_pcl_intel_finup(struct shash_desc *desc, const u8 *data,
+                             unsigned int len, u8 *out)
+{
+       return __crc32c_pcl_intel_finup(shash_desc_ctx(desc), data, len, out);
+}
+
+static int crc32c_pcl_intel_digest(struct shash_desc *desc, const u8 *data,
+                              unsigned int len, u8 *out)
+{
+       return __crc32c_pcl_intel_finup(crypto_shash_ctx(desc->tfm), data, len,
+                                   out);
+}
+#endif /* CONFIG_X86_64 */
+
 static struct shash_alg alg = {
        .setkey                 =       crc32c_intel_setkey,
        .init                   =       crc32c_intel_init,
@@ -184,6 +257,14 @@ static int __init crc32c_intel_mod_init(void)
 {
        if (!x86_match_cpu(crc32c_cpu_id))
                return -ENODEV;
+#ifdef CONFIG_X86_64
+       if (cpu_has_pclmulqdq) {
+               alg.update = crc32c_pcl_intel_update;
+               alg.finup = crc32c_pcl_intel_finup;
+               alg.digest = crc32c_pcl_intel_digest;
+               set_pcl_breakeven_point();
+       }
+#endif
        return crypto_register_shash(&alg);
 }
 
diff --git a/arch/x86/crypto/crc32c-pcl-intel-asm_64.S b/arch/x86/crypto/crc32c-pcl-intel-asm_64.S
new file mode 100644 (file)
index 0000000..93c6d39
--- /dev/null
@@ -0,0 +1,460 @@
+/*
+ * Implement fast CRC32C with PCLMULQDQ instructions. (x86_64)
+ *
+ * The white paper on CRC32C calculations with PCLMULQDQ instruction can be
+ * downloaded from:
+ * http://download.intel.com/design/intarch/papers/323405.pdf
+ *
+ * Copyright (C) 2012 Intel Corporation.
+ *
+ * Authors:
+ *     Wajdi Feghali <wajdi.k.feghali@intel.com>
+ *     James Guilford <james.guilford@intel.com>
+ *     David Cote <david.m.cote@intel.com>
+ *     Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+## ISCSI CRC 32 Implementation with crc32 and pclmulqdq Instruction
+
+.macro LABEL prefix n
+\prefix\n\():
+.endm
+
+.macro JMPTBL_ENTRY i
+.word crc_\i - crc_array
+.endm
+
+.macro JNC_LESS_THAN j
+       jnc less_than_\j
+.endm
+
+# Define threshold where buffers are considered "small" and routed to more
+# efficient "by-1" code. This "by-1" code only handles up to 255 bytes, so
+# SMALL_SIZE can be no larger than 255.
+
+#define SMALL_SIZE 200
+
+.if (SMALL_SIZE > 255)
+.error "SMALL_ SIZE must be < 256"
+.endif
+
+# unsigned int crc_pcl(u8 *buffer, int len, unsigned int crc_init);
+
+.global crc_pcl
+crc_pcl:
+#define    bufp                %rdi
+#define    bufp_dw     %edi
+#define    bufp_w      %di
+#define    bufp_b      %dil
+#define    bufptmp     %rcx
+#define    block_0     %rcx
+#define    block_1     %rdx
+#define    block_2     %r11
+#define    len         %rsi
+#define    len_dw      %esi
+#define    len_w       %si
+#define    len_b       %sil
+#define    crc_init_arg %rdx
+#define    tmp         %rbx
+#define    crc_init    %r8
+#define    crc_init_dw %r8d
+#define    crc1                %r9
+#define    crc2                %r10
+
+       pushq   %rbx
+       pushq   %rdi
+       pushq   %rsi
+
+       ## Move crc_init for Linux to a different
+       mov     crc_init_arg, crc_init
+
+       ################################################################
+       ## 1) ALIGN:
+       ################################################################
+
+       mov     bufp, bufptmp           # rdi = *buf
+       neg     bufp
+       and     $7, bufp                # calculate the unalignment amount of
+                                       # the address
+       je      proc_block              # Skip if aligned
+
+       ## If len is less than 8 and we're unaligned, we need to jump
+       ## to special code to avoid reading beyond the end of the buffer
+       cmp     $8, len
+       jae     do_align
+       # less_than_8 expects length in upper 3 bits of len_dw
+       # less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30]
+       shl     $32-3+1, len_dw
+       jmp     less_than_8_post_shl1
+
+do_align:
+       #### Calculate CRC of unaligned bytes of the buffer (if any)
+       movq    (bufptmp), tmp          # load a quadward from the buffer
+       add     bufp, bufptmp           # align buffer pointer for quadword
+                                       # processing
+       sub     bufp, len               # update buffer length
+align_loop:
+       crc32b  %bl, crc_init_dw        # compute crc32 of 1-byte
+       shr     $8, tmp                 # get next byte
+       dec     bufp
+       jne     align_loop
+
+proc_block:
+
+       ################################################################
+       ## 2) PROCESS  BLOCKS:
+       ################################################################
+
+       ## compute num of bytes to be processed
+       movq    len, tmp                # save num bytes in tmp
+
+       cmpq    $128*24, len
+       jae     full_block
+
+continue_block:
+       cmpq    $SMALL_SIZE, len
+       jb      small
+
+       ## len < 128*24
+       movq    $2731, %rax             # 2731 = ceil(2^16 / 24)
+       mul     len_dw
+       shrq    $16, %rax
+
+       ## eax contains floor(bytes / 24) = num 24-byte chunks to do
+
+       ## process rax 24-byte chunks (128 >= rax >= 0)
+
+       ## compute end address of each block
+       ## block 0 (base addr + RAX * 8)
+       ## block 1 (base addr + RAX * 16)
+       ## block 2 (base addr + RAX * 24)
+       lea     (bufptmp, %rax, 8), block_0
+       lea     (block_0, %rax, 8), block_1
+       lea     (block_1, %rax, 8), block_2
+
+       xor     crc1, crc1
+       xor     crc2, crc2
+
+       ## branch into array
+       lea     jump_table(%rip), bufp
+       movzxw  (bufp, %rax, 2), len
+       offset=crc_array-jump_table
+       lea     offset(bufp, len, 1), bufp
+       jmp     *bufp
+
+       ################################################################
+       ## 2a) PROCESS FULL BLOCKS:
+       ################################################################
+full_block:
+       movq    $128,%rax
+       lea     128*8*2(block_0), block_1
+       lea     128*8*3(block_0), block_2
+       add     $128*8*1, block_0
+
+       xor     crc1,crc1
+       xor     crc2,crc2
+
+       # Fall thruogh into top of crc array (crc_128)
+
+       ################################################################
+       ## 3) CRC Array:
+       ################################################################
+
+crc_array:
+       i=128
+.rept 128-1
+.altmacro
+LABEL crc_ %i
+.noaltmacro
+       crc32q   -i*8(block_0), crc_init
+       crc32q   -i*8(block_1), crc1
+       crc32q   -i*8(block_2), crc2
+       i=(i-1)
+.endr
+
+.altmacro
+LABEL crc_ %i
+.noaltmacro
+       crc32q   -i*8(block_0), crc_init
+       crc32q   -i*8(block_1), crc1
+# SKIP  crc32  -i*8(block_2), crc2 ; Don't do this one yet
+
+       mov     block_2, block_0
+
+       ################################################################
+       ## 4) Combine three results:
+       ################################################################
+
+       lea     (K_table-16)(%rip), bufp        # first entry is for idx 1
+       shlq    $3, %rax                        # rax *= 8
+       subq    %rax, tmp                       # tmp -= rax*8
+       shlq    $1, %rax
+       subq    %rax, tmp                       # tmp -= rax*16
+                                               # (total tmp -= rax*24)
+       addq    %rax, bufp
+
+       movdqa  (bufp), %xmm0                   # 2 consts: K1:K2
+
+       movq    crc_init, %xmm1                 # CRC for block 1
+       pclmulqdq $0x00,%xmm0,%xmm1             # Multiply by K2
+
+       movq    crc1, %xmm2                     # CRC for block 2
+       pclmulqdq $0x10, %xmm0, %xmm2           # Multiply by K1
+
+       pxor    %xmm2,%xmm1
+       movq    %xmm1, %rax
+       xor     -i*8(block_2), %rax
+       mov     crc2, crc_init
+       crc32   %rax, crc_init
+
+################################################################
+## 5) Check for end:
+################################################################
+
+LABEL crc_ 0
+       mov     tmp, len
+       cmp     $128*24, tmp
+       jae     full_block
+       cmp     $24, tmp
+       jae     continue_block
+
+less_than_24:
+       shl     $32-4, len_dw                   # less_than_16 expects length
+                                               # in upper 4 bits of len_dw
+       jnc     less_than_16
+       crc32q  (bufptmp), crc_init
+       crc32q  8(bufptmp), crc_init
+       jz      do_return
+       add     $16, bufptmp
+       # len is less than 8 if we got here
+       # less_than_8 expects length in upper 3 bits of len_dw
+       # less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30]
+       shl     $2, len_dw
+       jmp     less_than_8_post_shl1
+
+       #######################################################################
+       ## 6) LESS THAN 256-bytes REMAIN AT THIS POINT (8-bits of len are full)
+       #######################################################################
+small:
+       shl $32-8, len_dw               # Prepare len_dw for less_than_256
+       j=256
+.rept 5                                        # j = {256, 128, 64, 32, 16}
+.altmacro
+LABEL less_than_ %j                    # less_than_j: Length should be in
+                                       # upper lg(j) bits of len_dw
+       j=(j/2)
+       shl     $1, len_dw              # Get next MSB
+       JNC_LESS_THAN %j
+.noaltmacro
+       i=0
+.rept (j/8)
+       crc32q  i(bufptmp), crc_init    # Compute crc32 of 8-byte data
+       i=i+8
+.endr
+       jz      do_return               # Return if remaining length is zero
+       add     $j, bufptmp             # Advance buf
+.endr
+
+less_than_8:                           # Length should be stored in
+                                       # upper 3 bits of len_dw
+       shl     $1, len_dw
+less_than_8_post_shl1:
+       jnc     less_than_4
+       crc32l  (bufptmp), crc_init_dw  # CRC of 4 bytes
+       jz      do_return               # return if remaining data is zero
+       add     $4, bufptmp
+less_than_4:                           # Length should be stored in
+                                       # upper 2 bits of len_dw
+       shl     $1, len_dw
+       jnc     less_than_2
+       crc32w  (bufptmp), crc_init_dw  # CRC of 2 bytes
+       jz      do_return               # return if remaining data is zero
+       add     $2, bufptmp
+less_than_2:                           # Length should be stored in the MSB
+                                       # of len_dw
+       shl     $1, len_dw
+       jnc     less_than_1
+       crc32b  (bufptmp), crc_init_dw  # CRC of 1 byte
+less_than_1:                           # Length should be zero
+do_return:
+       movq    crc_init, %rax
+       popq    %rsi
+       popq    %rdi
+       popq    %rbx
+        ret
+
+        ################################################################
+        ## jump table        Table is 129 entries x 2 bytes each
+        ################################################################
+.align 4
+jump_table:
+       i=0
+.rept 129
+.altmacro
+JMPTBL_ENTRY %i
+.noaltmacro
+       i=i+1
+.endr
+       ################################################################
+       ## PCLMULQDQ tables
+       ## Table is 128 entries x 2 quad words each
+       ################################################################
+.data
+.align 64
+K_table:
+        .quad 0x14cd00bd6,0x105ec76f0
+        .quad 0x0ba4fc28e,0x14cd00bd6
+        .quad 0x1d82c63da,0x0f20c0dfe
+        .quad 0x09e4addf8,0x0ba4fc28e
+        .quad 0x039d3b296,0x1384aa63a
+        .quad 0x102f9b8a2,0x1d82c63da
+        .quad 0x14237f5e6,0x01c291d04
+        .quad 0x00d3b6092,0x09e4addf8
+        .quad 0x0c96cfdc0,0x0740eef02
+        .quad 0x18266e456,0x039d3b296
+        .quad 0x0daece73e,0x0083a6eec
+        .quad 0x0ab7aff2a,0x102f9b8a2
+        .quad 0x1248ea574,0x1c1733996
+        .quad 0x083348832,0x14237f5e6
+        .quad 0x12c743124,0x02ad91c30
+        .quad 0x0b9e02b86,0x00d3b6092
+        .quad 0x018b33a4e,0x06992cea2
+        .quad 0x1b331e26a,0x0c96cfdc0
+        .quad 0x17d35ba46,0x07e908048
+        .quad 0x1bf2e8b8a,0x18266e456
+        .quad 0x1a3e0968a,0x11ed1f9d8
+        .quad 0x0ce7f39f4,0x0daece73e
+        .quad 0x061d82e56,0x0f1d0f55e
+        .quad 0x0d270f1a2,0x0ab7aff2a
+        .quad 0x1c3f5f66c,0x0a87ab8a8
+        .quad 0x12ed0daac,0x1248ea574
+        .quad 0x065863b64,0x08462d800
+        .quad 0x11eef4f8e,0x083348832
+        .quad 0x1ee54f54c,0x071d111a8
+        .quad 0x0b3e32c28,0x12c743124
+        .quad 0x0064f7f26,0x0ffd852c6
+        .quad 0x0dd7e3b0c,0x0b9e02b86
+        .quad 0x0f285651c,0x0dcb17aa4
+        .quad 0x010746f3c,0x018b33a4e
+        .quad 0x1c24afea4,0x0f37c5aee
+        .quad 0x0271d9844,0x1b331e26a
+        .quad 0x08e766a0c,0x06051d5a2
+        .quad 0x093a5f730,0x17d35ba46
+        .quad 0x06cb08e5c,0x11d5ca20e
+        .quad 0x06b749fb2,0x1bf2e8b8a
+        .quad 0x1167f94f2,0x021f3d99c
+        .quad 0x0cec3662e,0x1a3e0968a
+        .quad 0x19329634a,0x08f158014
+        .quad 0x0e6fc4e6a,0x0ce7f39f4
+        .quad 0x08227bb8a,0x1a5e82106
+        .quad 0x0b0cd4768,0x061d82e56
+        .quad 0x13c2b89c4,0x188815ab2
+        .quad 0x0d7a4825c,0x0d270f1a2
+        .quad 0x10f5ff2ba,0x105405f3e
+        .quad 0x00167d312,0x1c3f5f66c
+        .quad 0x0f6076544,0x0e9adf796
+        .quad 0x026f6a60a,0x12ed0daac
+        .quad 0x1a2adb74e,0x096638b34
+        .quad 0x19d34af3a,0x065863b64
+        .quad 0x049c3cc9c,0x1e50585a0
+        .quad 0x068bce87a,0x11eef4f8e
+        .quad 0x1524fa6c6,0x19f1c69dc
+        .quad 0x16cba8aca,0x1ee54f54c
+        .quad 0x042d98888,0x12913343e
+        .quad 0x1329d9f7e,0x0b3e32c28
+        .quad 0x1b1c69528,0x088f25a3a
+        .quad 0x02178513a,0x0064f7f26
+        .quad 0x0e0ac139e,0x04e36f0b0
+        .quad 0x0170076fa,0x0dd7e3b0c
+        .quad 0x141a1a2e2,0x0bd6f81f8
+        .quad 0x16ad828b4,0x0f285651c
+        .quad 0x041d17b64,0x19425cbba
+        .quad 0x1fae1cc66,0x010746f3c
+        .quad 0x1a75b4b00,0x18db37e8a
+        .quad 0x0f872e54c,0x1c24afea4
+        .quad 0x01e41e9fc,0x04c144932
+        .quad 0x086d8e4d2,0x0271d9844
+        .quad 0x160f7af7a,0x052148f02
+        .quad 0x05bb8f1bc,0x08e766a0c
+        .quad 0x0a90fd27a,0x0a3c6f37a
+        .quad 0x0b3af077a,0x093a5f730
+        .quad 0x04984d782,0x1d22c238e
+        .quad 0x0ca6ef3ac,0x06cb08e5c
+        .quad 0x0234e0b26,0x063ded06a
+        .quad 0x1d88abd4a,0x06b749fb2
+        .quad 0x04597456a,0x04d56973c
+        .quad 0x0e9e28eb4,0x1167f94f2
+        .quad 0x07b3ff57a,0x19385bf2e
+        .quad 0x0c9c8b782,0x0cec3662e
+        .quad 0x13a9cba9e,0x0e417f38a
+        .quad 0x093e106a4,0x19329634a
+        .quad 0x167001a9c,0x14e727980
+        .quad 0x1ddffc5d4,0x0e6fc4e6a
+        .quad 0x00df04680,0x0d104b8fc
+        .quad 0x02342001e,0x08227bb8a
+        .quad 0x00a2a8d7e,0x05b397730
+        .quad 0x168763fa6,0x0b0cd4768
+        .quad 0x1ed5a407a,0x0e78eb416
+        .quad 0x0d2c3ed1a,0x13c2b89c4
+        .quad 0x0995a5724,0x1641378f0
+        .quad 0x19b1afbc4,0x0d7a4825c
+        .quad 0x109ffedc0,0x08d96551c
+        .quad 0x0f2271e60,0x10f5ff2ba
+        .quad 0x00b0bf8ca,0x00bf80dd2
+        .quad 0x123888b7a,0x00167d312
+        .quad 0x1e888f7dc,0x18dcddd1c
+        .quad 0x002ee03b2,0x0f6076544
+        .quad 0x183e8d8fe,0x06a45d2b2
+        .quad 0x133d7a042,0x026f6a60a
+        .quad 0x116b0f50c,0x1dd3e10e8
+        .quad 0x05fabe670,0x1a2adb74e
+        .quad 0x130004488,0x0de87806c
+        .quad 0x000bcf5f6,0x19d34af3a
+        .quad 0x18f0c7078,0x014338754
+        .quad 0x017f27698,0x049c3cc9c
+        .quad 0x058ca5f00,0x15e3e77ee
+        .quad 0x1af900c24,0x068bce87a
+        .quad 0x0b5cfca28,0x0dd07448e
+        .quad 0x0ded288f8,0x1524fa6c6
+        .quad 0x059f229bc,0x1d8048348
+        .quad 0x06d390dec,0x16cba8aca
+        .quad 0x037170390,0x0a3e3e02c
+        .quad 0x06353c1cc,0x042d98888
+        .quad 0x0c4584f5c,0x0d73c7bea
+        .quad 0x1f16a3418,0x1329d9f7e
+        .quad 0x0531377e2,0x185137662
+        .quad 0x1d8d9ca7c,0x1b1c69528
+        .quad 0x0b25b29f2,0x18a08b5bc
+        .quad 0x19fb2a8b0,0x02178513a
+        .quad 0x1a08fe6ac,0x1da758ae0
+        .quad 0x045cddf4e,0x0e0ac139e
+        .quad 0x1a91647f2,0x169cf9eb0
+        .quad 0x1a0f717c4,0x0170076fa
index 6563366bae809f8e5b052e003ad42c6c2f2adb33..b901b590635f1b5ca7f44b1270db837cfee746ca 100644 (file)
@@ -324,9 +324,19 @@ config CRYPTO_CRC32C
          by iSCSI for header and data digests and by others.
          See Castagnoli93.  Module will be crc32c.
 
+config CRYPTO_CRC32C_X86_64
+       bool
+       depends on X86 && 64BIT
+       select CRYPTO_HASH
+       help
+         In Intel processor with SSE4.2 supported, the processor will
+         support CRC32C calculation using hardware accelerated CRC32
+         instruction optimized with PCLMULQDQ instruction when available.
+
 config CRYPTO_CRC32C_INTEL
        tristate "CRC32c INTEL hardware acceleration"
        depends on X86
+       select CRYPTO_CRC32C_X86_64 if 64BIT
        select CRYPTO_HASH
        help
          In Intel processor with SSE4.2 supported, the processor will