Currently the CRC-T10DIF checksum is computed using a generic table lookup
algorithm. By switching the checksum to PCLMULQDQ based computation,
we can speedup the computation by 8x for checksumming 512 bytes and
even more for larger buffer size. This will improve performance of SCSI
drivers turning on the CRC-T10IDF checksum. In our SSD based experiments,
we have seen increase disk throughput by 3.5x with T10DIF for 512 byte
block size.
This patch set provides the x86_64 routine using PCLMULQDQ instruction
and switches the crc_t10dif library function to use the faster PCLMULQDQ
based routine when available.
Tim
v1->v2
1. Get rid of unnecessary xmm registers save and restore and fix ENDPROC
position in PCLMULQDQ version of crc t10dif computation.
2. Fix URL to paper reference of CRC computation with PCLMULQDQ.
3. Add one additional tcrypt test case to exercise more code paths through
crc t10dif computation.
4. Fix config dependencies of CRYPTO_CRCT10DIF.
Thanks to Matthew and Jussi who reviewed the patches and Keith
for testing version 1 of the patch set.
Tim Chen (4):
Wrap crc_t10dif function all to use crypto transform framework
Accelerated CRC T10 DIF computation with PCLMULQDQ instruction
Glue code to cast accelerated CRCT10DIF assembly as a crypto
transform
Simple correctness and speed test for CRCT10DIF hash
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/crct10dif-pcl-asm_64.S | 643 ++++++++++++++++++++++++++++++++
arch/x86/crypto/crct10dif-pclmul_glue.c | 153 ++++++++
crypto/Kconfig | 21 ++
crypto/tcrypt.c | 8 +
crypto/testmgr.c | 10 +
crypto/testmgr.h | 33 ++
include/linux/crc-t10dif.h | 10 +
lib/crc-t10dif.c | 96 +++++
9 files changed, 976 insertions(+)
create mode 100644 arch/x86/crypto/crct10dif-pcl-asm_64.S
create mode 100644 arch/x86/crypto/crct10dif-pclmul_glue.c
--
1.7.11.7
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
