On Thu, Oct 24, 2024 at 06:30:15PM +0200, Marek Vasut wrote:
>
> @@ -582,15 +585,12 @@ void hwrng_unregister(struct hwrng *rng)
> }
>
> new_rng = get_current_rng_nolock();
> - if (list_empty(&rng_list)) {
> - mutex_unlock(&rng_mutex);
> - if (hwrng_fill)
> - kthread_stop(hwrng_fill);
> - } else
> - mutex_unlock(&rng_mutex);
> + mutex_unlock(&rng_mutex);
>
> if (new_rng)
> put_rng(new_rng);
> + else
> + kthread_park(hwrng_fill);
The kthread_park should be moved back into the locked region
of rng_mute). The kthread_stop was moved out because it could
dead-lock waiting on the kthread that's also taking the same
lock. This is no longer an issue with kthread_park since it
simply sets a flag.
Having it outside of the locked region is potentially dangerous
since a pair of hwrng_unregister and hwrng_register could be
re-ordered resulting in the kthread being parked forever.
Thanks,
--
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
