Re: [PATCH] Fix x86_64 _spin_lock_irqsave()

[Suleiman Souhlal <ssouhlal@FreeBSD.org>]

Andi Kleen wrote:
> On Thursday 24 August 2006 13:04, Suleiman Souhlal wrote:
>
> > Andi Kleen wrote:
> >
> > > Edward Falk <efalk@google.com> writes:
> > >
> > >
> > >
> > > > Add spin_lock_string_flags and _raw_spin_lock_flags() to
> > > > asm-x86_64/spinlock.h so that _spin_lock_irqsave() has the same
> > > > semantics on x86_64 as it does on i386 and does *not* have interrupts
> > > > disabled while it is waiting for the lock.
> > >
> > >
> > > Did it fix anything for you?
> >
> > I think this was to work around the fact that some buggy drivers try to 
> > grab spinlocks without disabling interrupts when they should, which 
> > would cause deadlocks when trying to rendez-vous every cpu via IPIs.
>
>
> That doesn't help them at all because they could then deadlock later.

If the driver uses spin_lock() when it knows that the hardware won't 
generate the interrupt that would need to be masked, and 
spin_lock_irqsave() elsewhere, there shouldn't be any deadlocks unless 
IPIs are involved.

For example, say a driver uses spin_lock(&driver_lock) in its interrupt 
handler and spin_lock_irqsave(&driver_lock) elsewhere.
Imagine CPU1 is handling a such a interrupt while CPU2 is trying to send 
a packet (for example).

In a regular situation, CPU1 shouldn't be interrupted by anything 
needing driver_lock, and so it should be able to complete and let CPU2 
acquire the lock.

Now, if CPU3 is trying to do an IPI rendez-vous, it will interrupt CPU1 
and try to interrupt CPU2 too. However, since spin_lock_irqsave() spins 
with interrupts disabled, the system will deadlock.

With this patch, IPI rendez-vous shouldn't cause these problems, since 
it will let the rendez-vous will be able to complete. Or am I missing 
something?

-- Suleiman
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/