RE: [PATCH v2 19/19] crypto: x86/sha - register only the best function

From: Elliott, Robert (Servers)
Date: Thu Oct 13 2022 - 19:00:11 EST




> -----Original Message-----
> From: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
> Sent: Thursday, October 13, 2022 2:52 AM
> To: Eric Biggers <ebiggers@xxxxxxxxxx>
> Cc: Elliott, Robert (Servers) <elliott@xxxxxxx>; davem@xxxxxxxxxxxxx;
> tim.c.chen@xxxxxxxxxxxxxxx; ap420073@xxxxxxxxx; ardb@xxxxxxxxxx; linux-
> crypto@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH v2 19/19] crypto: x86/sha - register only the best
> function
>
> On Wed, Oct 12, 2022 at 11:07:43PM -0700, Eric Biggers wrote:
> >
> > I thought that it's done the way it is so that it's easy to run the self-
> tests
> > for all the different variants.
>
> Yes, we should keep it that way so that it's easy to test the
> different code paths for correctness and/or speed.

I have done some testing with extra patches that do that for
that very reason. Is there much overhead from having a module
loaded and registered in the crypto system, but not being
chosen for use?

The current sha modules register SSSE3, then register AVX and
unregister SSSE3, then register AVX2 and unregister AVX...
good testing for the unregister function, but not real helpful
for users.