Re: [PATCH] crypto: marvell/cesa - Avoid empty transfer descriptor

[Herbert Xu]

On Tue, Jun 17, 2025 at 07:32:28AM +0200, Klaus Kudielka wrote:
> Hello,
> 
> Plain v6.16-rc1 with marvell-cesa as module and CRYPTO_SELFTESTS enabled.
> Identical behaviour as reported previously with the cryptodev tree (see below).
> 
> According to /proc/crypto, all marvell-cesa selftests (including ahash) appear to pass reliably again.
> 
> If we can trust the information in /proc/crypto, I would suggest to revert commit
> e845d2399a00f866f287e0cefbd4fc7d8ef0d2f7 ("crypto: marvell/cesa - Disable hash algorithms")

Corentin is still able to reproduce failures reliably.  While
I have a patch that seems to reduce the failure rate for him it
is not yet a complete fix.  There is also no full explanation of
why my patch even works to the extent that it does (the DMA memory
is supposed to be cache-coherent but perhaps it isn't).

So I think it is still premature at this point.  There is also a
known problem with the hmac code where it fails to deal with zero-length
updates correctly.

Cheers,
-- 
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt