Re: configurable discard parameters

["Martin K. Petersen" <martin.petersen@oracle.com>]

>>>>> "Tom" == Tom Yan <tom.ty89@gmail.com> writes:

Tom> First of all it wouldn't really solve problem for all devices which
Tom> have different limits.

Only by virtue of us generally aligning on large power of two
boundaries.

The only reason I am entertaining this in the first place is that I have
one drive that behaves in a way similar to yours. And if we can make
things slightly better (but not perfect) for several drives without
causing any regressions then that's worth exploring.

Tom> Secondly I doubt it's related to Deterministic Zero AT ALL. (For
Tom> one the SanDisk drive I have shows Deterministic Zero, still it
Tom> behaves similarly as the Intel drive.)

The drive reporting deterministic zero is not enough. It needs to be
explicitly whitelisted before we report discard_zeroes_data=1.

Tom> The only patching which would really mean something is to allow
Tom> user to configure blocks per range and ranges per command, so that
Tom> for users can tune kernel TRIM per device if they really want to,
Tom> while leaving the "safe default" intact.

I'll think about it.

Tom> However I am really curious how the drives "blow up" on less blocks
Tom> per range. Isn't that even more of a firmware bug than the problem
Tom> I have?

I have several older drives that expect a single contiguous LBA
range. They don't handle multiple discontiguous ranges at all.

-- 
Martin K. Petersen	Oracle Linux Engineering