Can we replace the msleep(250) in dev.c netdev_wait_allrefs?

I think there was a proposal to change it to schedule_timeout_uninterruptible(1); some time ago, but that change never made it upstream (at least not in 3.7.5)?

Thanks,
Ben

Quoted 6 lines · Ben Greear

I think there was a proposal to change it to schedule_timeout_uninterruptible(1); some time ago, but that change never made it upstream (at least not in 3.7.5)?

Thanks,
Ben

Honestly we should fix the bug at this point, not hide it.

We could use msleep(2500) to make sure people will complain ;)

Quoted 12 lines · Ben Greear

Quoted 8 lines

I think there was a proposal to change it to schedule_timeout_uninterruptible(1);
some time ago, but that change never made it upstream (at least not in 3.7.5)?

Thanks,
Ben

Honestly we should fix the bug at this point, not hide it.

Quoted 1 line

We could use msleep(2500) to make sure people will complain ;)

Well, few will notice that either, or at least are unlikely to know what to blame.

If we should just never reach that code, maybe a WARN_ON_ONCE() when we do?

If you only notice some time in the future that something has hung in this state, the original message could be lost. Therefore, I definitely prefer that this thing keeps barking into the logs periodically.

Quoted 4 lines

If you only notice some time in the future that something has hung in this state, the original message could be lost. Therefore, I definitely prefer that this thing keeps barking into the logs periodically.

Not sure we need the full stack, as it probably contains no useful context / information at this point.