Hi, About 3 years ago Liu Bo has try add xxhash support to btrfs[1]. For now (Kernel 4.14) kernel have a native library for xxhash.
Main reason to not add that support is what: - CRC32C - has hardware support and amazing fast - CRC32C - Can fix 1 bit error (Yes that possible, but that not used in btrfs and i hard to find example of correction in the kernel) So that a not a valuable reason. xxhash32/64 did not have a HW support but must outperform CRC32c in SW mode. I'm not sure that ARM boxes, as example, some NAS have a HW crc32c. ([2] show that CRC32 can have HW on some types of ARMv8). ([3] for ARM v8.1-A CRC hw must be) May be that make a sense to add support for both xxhash32/64? And do some type of testing on module load like raid6 have, and use fastest for newly written data? I test speed on xxhash32/64 and crc32c, and use fastest function for new written data? Thanks! P.S. May be some one has ARMv8 boxes? Can you check which module used by btrfs: ~ dmesg| grep crc [ 0.666469] Btrfs loaded, crc32c=crc32c-intel May be you want/can test xxhash32/64 on arm, i backport kernel version in [4]. (But [4] doesn't have HW CRC32 support, i hard to find example userspace implementation to stole on github and didn't have ARM box) Links: 1. http://linux-btrfs.vger.kernel.narkive.com/pO9lOshJ/patch-2-3-crypto-xxhash-add-tests 2. https://github.com/torvalds/linux/blob/5518b69b76680a4f2df96b1deca260059db0c2de/arch/arm/crypto/crc32-ce-glue.c 3. https://en.wikipedia.org/wiki/ARM_architecture#ARMv8.1-A 4. https://github.com/Nefelim4ag/CRC32C_AND_8Byte_parity -- Have a nice day, Timofey. -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
