blob: 5040efeae9030005af296bfa72c07414535144a4 [file] [log] [blame]
Jens Axboe38455912008-08-04 15:35:26 +02001#include <inttypes.h>
Jens Axboe5d7c5d32010-06-21 15:08:17 +02002#include <string.h>
3#include <unistd.h>
4#include <stdlib.h>
5#include <signal.h>
6#include <sys/types.h>
7#include <sys/wait.h>
Aaron Carroll419484b2008-09-12 10:57:13 +02008#include "crc32c.h"
Jens Axboe38455912008-08-04 15:35:26 +02009
10/*
11 * Based on a posting to lkml by Austin Zhang <austin.zhang@intel.com>
12 *
13 * Using hardware provided CRC32 instruction to accelerate the CRC32 disposal.
14 * CRC32C polynomial:0x1EDC6F41(BE)/0x82F63B78(LE)
15 * CRC32 is a new instruction in Intel SSE4.2, the reference can be found at:
16 * http://www.intel.com/products/processor/manuals/
17 * Intel(R) 64 and IA-32 Architectures Software Developer's Manual
18 * Volume 2A: Instruction Set Reference, A-M
19 */
20
Jens Axboe2f681242010-10-21 08:15:59 +020021#ifdef ARCH_HAVE_SSE4_2
Aaron Carroll419484b2008-09-12 10:57:13 +020022
Jens Axboe38455912008-08-04 15:35:26 +020023#if BITS_PER_LONG == 64
24#define REX_PRE "0x48, "
25#define SCALE_F 8
26#else
27#define REX_PRE
28#define SCALE_F 4
29#endif
30
Jens Axboecc62ea72012-02-09 21:17:06 +010031static uint32_t crc32c_intel_le_hw_byte(uint32_t crc, unsigned char const *data,
32 unsigned long length)
Jens Axboe38455912008-08-04 15:35:26 +020033{
34 while (length--) {
35 __asm__ __volatile__(
36 ".byte 0xf2, 0xf, 0x38, 0xf0, 0xf1"
37 :"=S"(crc)
38 :"0"(crc), "c"(*data)
39 );
40 data++;
41 }
42
43 return crc;
44}
45
46/*
47 * Steps through buffer one byte at at time, calculates reflected
48 * crc using table.
49 */
50uint32_t crc32c_intel(unsigned char const *data, unsigned long length)
51{
52 unsigned int iquotient = length / SCALE_F;
53 unsigned int iremainder = length % SCALE_F;
54#if BITS_PER_LONG == 64
55 uint64_t *ptmp = (uint64_t *) data;
56#else
57 uint32_t *ptmp = (uint32_t *) data;
58#endif
59 uint32_t crc = ~0;
60
61 while (iquotient--) {
62 __asm__ __volatile__(
63 ".byte 0xf2, " REX_PRE "0xf, 0x38, 0xf1, 0xf1;"
64 :"=S"(crc)
65 :"0"(crc), "c"(*ptmp)
66 );
67 ptmp++;
68 }
69
70 if (iremainder)
71 crc = crc32c_intel_le_hw_byte(crc, (unsigned char *)ptmp,
72 iremainder);
73
74 return crc;
75}
Aaron Carroll419484b2008-09-12 10:57:13 +020076
Jens Axboee0ab5f92010-06-29 10:07:13 +020077static void do_cpuid(unsigned int *eax, unsigned int *ebx, unsigned int *ecx,
78 unsigned int *edx)
Jens Axboe5d7c5d32010-06-21 15:08:17 +020079{
Jens Axboee0ab5f92010-06-29 10:07:13 +020080 int id = *eax;
Aaron Carroll419484b2008-09-12 10:57:13 +020081
Jens Axboee0ab5f92010-06-29 10:07:13 +020082 asm("movl %4, %%eax;"
83 "cpuid;"
84 "movl %%eax, %0;"
85 "movl %%ebx, %1;"
86 "movl %%ecx, %2;"
87 "movl %%edx, %3;"
88 : "=r" (*eax), "=r" (*ebx), "=r" (*ecx), "=r" (*edx)
89 : "r" (id)
90 : "eax", "ebx", "ecx", "edx");
Jens Axboe5d7c5d32010-06-21 15:08:17 +020091}
92
93int crc32c_intel_works(void)
94{
Jens Axboee0ab5f92010-06-29 10:07:13 +020095 unsigned int eax, ebx, ecx, edx;
Jens Axboe5d7c5d32010-06-21 15:08:17 +020096
Jens Axboee0ab5f92010-06-29 10:07:13 +020097 eax = 1;
98
99 do_cpuid(&eax, &ebx, &ecx, &edx);
100 return (ecx & (1 << 20)) != 0;
Jens Axboe5d7c5d32010-06-21 15:08:17 +0200101}
102
103#endif /* ARCH_HAVE_SSE */