Re: [PATCH] fix altsysrq deadlock

[Andrew Morton <akpm@osdl.org>]

Jason Baron <jbaron@redhat.com> wrote:
>
> 
> This patch fixes a deadlock in the handle_sysrq function.
> ...
>> -	__sysrq_lock_table();
> +	if(!__sysrq_trylock_table()) {
> +		if(in_interrupt())
> +			return;
> +		else
> +			__sysrq_lock_table();
> +	}
> +

This is only a partial solution - __sysrq_trylock_table() is exported to
modules which do who know what with it and they don't know how to handle
locking failures - they'll just go ahead and do a spin_unlock() of an
unlocked lock and mayhem will ensue.

What we need to do is to move all those inlined functions out of sysrq.h,
into sysrq.c then withdraw all those exported-to-modules helper functions
then remove __sysrq_trylock_table() altogether and then use
spin_lock_irqsave() in the appropriate places.