On Wed, Jan 26, 2022 at 04:01:04PM +0100, Philipp Zabel wrote:
> On Wed, 2022-01-26 at 15:53 +0100, Philipp Zabel wrote:
> [...]
> > This can happen during cleanup if the error path is taken for a built-
> > in algorithm, before crypto_algapi_init() was called.
>
> I see this happen on ARM with CONFIG_CRYPTO_AES_ARM_BS=y since v5.16-rc1
> because the simd_skcipher_create_compat("ecb(aes)", "ecb-aes-neonbs",
> "__ecb-aes-neonbs") call in arch/arm/crypto/aes-neonbs-glue.c returns
> -ENOENT. I believe that is the same issue as reported in [1].
Good catch. We definitely need to fix the crash from the unregister
while the algorithm is still being referenced.
However, the fact that the simd create is failing is probably a bug
too. Could you take a look at that aspect and whether the failure
is related to the testmgr changes in question?
Thanks!
--
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
