crc32: deal with NULL buffer case required by Ceph

[changchengx]

Ceph need to deal with the case when the input buffer is NULL:
https://github.com/ceph/ceph/blob/master/src/test/common/test_crc32c.cc#L347

Signed-off-by: Changcheng Liu [email protected]

[changchengx]

@tchaikov Please help review.
We're going to use isa-l's crc32 to replace with src/common/crc32c_intel_fast assembler implementation

[tchaikov]

instead of forking intel/isa-l, i am inclined to bump up ceph/isa-l's master to track the upstream if it is not updated, i am afraid it's not the case. i think this source file is almost identical to https://github.com/intel/isa-l/blob/8074e3fe1b9398a9d3b717267790050fc5041594/crc/crc32_iscsi_00.asm . can we just use it instead?

[changchengx]

@tchaikov
I'm raising a similar question at: intel#159

[tchaikov]

@aclamk do you happen to know if we have any real-world use case of https://github.com/ceph/ceph/blob/master/src/test/common/test_crc32c.cc#L358 ? or it's created this way just because it was more convenient than creating a buffer filled with zero?

[aclamk]

@tchaikov

aclamk do you happen to know if we have any real-world use case of https://github.com/ceph/ceph/blob/master/src/test/common/test_crc32c.cc#L358 ? or it's created this way just because it was more convenient than creating a buffer filled with zero?

This code is testing performace of special crc function that calculates crc from zeros.
We are using this shortcut for caching crc values: https://github.com/ceph/ceph/blob/master/src/common/buffer.cc#L1975

[tchaikov]

thanks @aclamk . now i understand it better. seems we do need to have a dedicate function for calculating the crc of zero-filled buffers.

[tchaikov]

@changchengx i feel that it could be function which can also benefit other projects interested in high-performance CRC32 calculating using the technique of cached/precalculated CRC32. if this makes sense to you, probably we can upstream this change? before that, i'd suggest create a dedicated function and keep it in Ceph.

what do you think?

[changchengx]

@tchaikov

benefit other projects interested in high-performance CRC32 calculating using the technique of cached/precalculated CRC32.

I haven't understood this well.
I'll feedback to you later after checking https://github.com/ceph/ceph/blob/master/src/common/buffer.cc#L1975