On Wed, 19 May 2021 at 13:29, Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote: > > On Wed, May 19, 2021 at 01:22:34PM +0200, Ard Biesheuvel wrote: > > > > crypto_stats_get(alg); > > - if (crypto_aead_get_flags(aead) & CRYPTO_TFM_NEED_KEY) > > + if (!(alg->cra_flags & CRYPTO_ALG_ASYNC) && > > + WARN_ONCE(!in_task() && !in_serving_softirq(), > > + "synchronous call from invalid context\n")) > > + ret = -EBUSY; > > I don't think we've ever supported crypto in hard IRQ contexts. > So this should be done regardless of ASYNC. > OK. > Then again, do we really need this since the assumption has > always been that the crypto API can only be invoked in user or > softirq context? > With this series applied, some of the arm64 accelerated s/w implementations will no longer work correctly when this rule is violated, and so it would be nice to have a sanity check somewhere. And policing rules like these is best done in generic code, right? So if we do need to check this, we should check it here. If we don't, then we can drop these patches.
