Re: [PATCH][BUG] Badness in smp_call_function at arch/i386/kernel/smp.c:552

[Andrew Morton <akpm@osdl.org>]

Zwane Mwaikambo <zwane@holomorphy.com> wrote:
>
> > Well, sort-of.
>  > 
>  > If __handle_sysrq was really a normal IRQ handler then the correct thing to
>  > do here is to replace spin_lock_irqsave() with spin_lock().  But
>  > __handle_sysrq() can also be called via /proc/sysrq-trigger and via the
>  > handlers of multiple interrupt sources.  So we're stuck with using
>  > spin_lock_irqsave().
>  > 
>  > However enabling interrupts as you've done menas that theoretically we
>  > could deadlock on sysrq_key_table_lock if another sysrq happens at the
>  > wrong time.
>  > 
>  > Which deadlock opportunity would you prefer? ;)
> 
>  Agreed, there is actually a higher chance of the smp_call_function 
>  deadlock occuring since the __handle_sysrq one relies on another sysrq 
>  event occuring via a different IRQ line interrupt handler, so 
>  we would have to do sysrq via serial and then sysrq via keyboard to cause 
>  the deadlock. Perhaps just make it a spin_trylock?

Well yeah, but it's so much fuss for such a silly problem.

How about a local_irq_enable() in sysrq_handle_reboot()?