Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote:
> @@ -1170,20 +1170,18 @@ nft_pipapo_avx2_lookup(const struct net *net, const struct nft_set *set,
> }
>
> m = rcu_dereference(priv->match);
> -
> + scratch = *raw_cpu_ptr(m->scratch);
> + if (unlikely(!scratch)) {
> + local_bh_enable();
> + return false;
The function has been changed upstream to return a pointer.
> + }
> + __local_lock_nested_bh(&scratch->bh_lock);
> /* Note that we don't need a valid MXCSR state for any of the
> * operations we use here, so pass 0 as mask and spare a LDMXCSR
> * instruction.
> */
> kernel_fpu_begin_mask(0);
Not sure this is correct. If RT allows to migrate in BH before
the local lock is taken, then the if (unlikely(!irq_fpu_usable()))
check needs to be done after the local lock was taken, no?
I will place a pending pipapo change in the nf-next:testing
branch shortly in case you need to resend.
If its fine as-is, I can also rebase the pending pipapo_avx2 patches.
Let me know.
