Re: [PATCH] [1/2] random: SMP locking

[Andrew Morton <akpm@osdl.org>]

Matt Mackall <mpm@selenic.com> wrote:
>
> > Are you really sure that all the decisions about where to use spin_lock()
>  > vs spin_lock_irq() vs spin_lock_irqsave() are correct?  They are
>  > non-obvious.
> 
>  Aside from the put_user stuff below, yes.

Well I see in Linus's current tree:

	ndev->regen_timer.function = ipv6_regen_rndid;

And ipv6_regen_rndid() ends up calling get_random_bytes() which calls
extract_entropy() which now does a bare spin_lock().

So I think if the timer is run while some process-context code on the same
CPU is running get_random_bytes() we deadlock don't we?

Probably, we should make get_random_bytes() callable from any context.