Re: [PATCH v4] scsi: arm: Convert acornscsi-io.S to C

[Breno Rodrigues Alves]

Hi Jonathan,

You are right, and I should have been clearer about the motivation.

The reason this patch exists is that the original assembly has a latent
bug: for odd transfer lengths, it drops the final byte.  The C version
fixes that, and the odd-length case is now exercised by the userspace
unit test.  But that bug cannot be triggered by any current caller, so
even the bug-fix justification is weak - it is a fix for a path that
nothing currently exercises, in a driver for hardware almost nobody
still has.

You are also right about the broader point: "assembly to C" on its own
is not a justification.  The kernel's assembly is there for reasons, and
translating it for its own sake is churn that reviewers have to pay for.
I have been approaching this as a translation exercise, and that is not
the right frame.  I will not propose further conversions of that kind
without a concrete functional benefit - something that fixes a bug that
can actually be hit, or enables a change that cannot be done in assembly.

Bart's Reviewed-by: stands on the technical merits of the change as a
change (the C code is equivalent to the assembly for even lengths and
correct for odd ones).  Your concern is a different question - whether
the change should exist at all.  Both are valid, and the decision
belongs to the SCSI maintainers, not to me.

If the consensus is that this patch is not worth the churn, I am happy
to withdraw it.  I would rather withdraw it than have it sit in the tree
as an example of exactly the kind of change you are warning against.

Thanks for the direct feedback - it is a useful correction to how I was
approaching this, and more valuable than the patch itself.

Breno