Re: CI: another regression on linux-next with s390 sha384/sha512

[Eric Biggers <ebiggers@kernel.org>]

On Tue, Jun 24, 2025 at 01:38:54PM +0200, Ingo Franzki wrote:
> First of all sha512_s390.ko is no longer there. I guess that intended.

Yes.

> However, /proc/crypto does show
> 
>    name         : sha256
>    driver       : sha256-s390
>    module       : kernel
> 
> and
> 
>    name         : sha224
>    driver       : sha224-s390
>    module       : kernel
> 
> but no -s390 driver for sha512 and sha384. It only shows 
> 
>    name         : sha512
>    driver       : sha512-lib
>    module       : kernel
> 
> and 
> 
>    name         : sha384
>    driver       : sha384-lib
>    module       : kernel
> 
> The -lib variants are also shown for sha224 and sha256, but those also have the s390 variants.
> 
> So it looks like the s390 optimized sha384 and sha512 are now missing ? 
> 
> Similar, the -generic variants are only available for sha256 and sha224, but not for sha384 and sha512:
> 
>    name         : sha256
>    driver       : sha256-generic
>    module       : kernel
> 
> Can this please be fixed? We really want to keep the s390 optimized versions of all digests! 

sha384-lib and sha512-lib are s390 optimized (when the kernel is built for s390
and the CPU supports the CPACF_KIMD_SHA_512 instruction).  Please see
lib/crypto/s390/sha512.h.

So, SHA-384 and SHA-512 are still fully optimized for s390.

Yes, I'll be making the same change to SHA-224 and SHA-256 so that it's
consistent.  Sorry about the temporary inconsistency.  Note that sha224-lib and
sha256-lib are already s390-optimized as well; the difference is just that the
unnecessary sha224-generic, sha256-generic, sha224-$(ARCH), and sha256-$(ARCH)
algorithms are still being registered.  The *-lib ones just do the right thing,
and they are all that's needed.

- Eric