Re: [PATCH] lib/crypto: sparc: Drop optimized MD5 code

[Ard Biesheuvel]

On Fri, 27 Mar 2026, at 21:14, Eric Biggers wrote:
> On Thu, Mar 26, 2026 at 04:02:32PM -0700, Eric Biggers wrote:
>> On Thu, Mar 26, 2026 at 10:51:01PM +0100, John Paul Adrian Glaubitz wrote:
>> > On Thu, 2026-03-26 at 13:33 -0700, Eric Biggers wrote:
>> > > MD5 is obsolete.  Continuing to maintain architecture-optimized
>> > > implementations of MD5 is unnecessary and risky.  It diverts resources
>> > > from the modern algorithms that are actually important.
>> > 
>> > Why is it risky? That makes no sense.
>> 
>> Because there can be issues in architecture-optimized algorithm
>> implementations that don't exist in the generic implementations.  That's
>> a very common class of issue that has repeated over time.
>> 
>> > I also don't see how it diverts resources as no one is forced to work
>> > on the code.
>> > 
>> > SPARC is an architecture used by hobbyists and in space these days (in
>> > the form of Leon). I don't think any other kernel developer will have
>> > to take a look at it.
>> 
>> Huh?  We've been refactoring how the various crypto and CRC algorithms
>> are integrated, for all architectures.
>> 
>> So people outside the SPARC community, especially myself, been having to
>> spend quite a bit of time updating the SPARC code so that it can still
>> be used.
>> 
>> And this isn't new.  I've had to patch arch/sparc/crypto/ many times
>> over the years as things change in the crypto subsystem.  Many other
>> people, again outside the SPARC community, have as well.
>> 
>> The fact that you're denying that we've had to do this is really
>> frustrating.  There is a significant maintenance cost to keeping this
>> code working, which is being paid by people outside the SPARC community.
>> 
>> It seems best to at least focus that effort on modern algorithms like
>> AES and SHA-256, and not obsolete ones like MD5 and DES.  Note that
>> dropping those eliminates the need to add them to QEMU, as well.
>> 
>> I think that makes things easier for everyone.
>
> Let me know if you're aware of a real user that needs the obsolete MD5
> algorithm to be accelerated for SPARC in the kernel.
>
> Otherwise, I suggest we proceed with this patch, as this objection seems
> to based only on principles and misunderstandings.
>

Agreed. I do sympathise with the hobbyists, (and with the astronauts, for that matter), but the mainline kernel is not a museum. If someone has a desire to run obsolete code on an obsolete architecture, they are more than welcome to do so. But demanding that the rest of the world spends time on this is unreasonable, especially given that few people have access to the hardware.

The upshot of that is that the burden falls on the maintainers to chase bot reports and other reported regressions on obsolete architectures, and doing so for code such as MD5 which is entirely obsolete itself is just busywork.

Acked-by: Ard Biesheuvel <ardb@xxxxxxxxxx>