Re: [PATCH], issue EOI to APIC prior to calling crash_kexec in die_nmi path

[Ingo Molnar <mingo@elte.hu>]

* Vivek Goyal <vgoyal@redhat.com> wrote:

> On Wed, Feb 06, 2008 at 11:00:01PM +0100, Ingo Molnar wrote:
> > 
> > * Neil Horman <nhorman@tuxdriver.com> wrote:
> > 
> > >  	if (!user_mode_vm(regs)) {
> > > +		nmi_exit();
> > > +		local_irq_enable();
> > >  		current->thread.trap_no = 2;
> > >  		crash_kexec(regs);
> > 
> > looks good to me, but please move the local_irq_enable() to within 
> > crash_kexec() instead - probably inside the "got the kexec lock" 
> > section. That makes crash_kexec() use generally safer too i guess: right 
> > it seems that die() too can call crash_kexec() with irqs disabled - and 
> > can thus hang in smp_send_stop() [or wherever it hung before].
> > 
> 
> In general, I think we should not be servicing interrupts once the 
> system has crashed and crash_kexec() has been invoked.
> 
> In fact, right now machine_crash_shutdown() explicity disables 
> interrupt before sending NMIs to other cpus to stop these cpus and 
> which makes sense to me.
> 
> I am wondering if interrupts are disabled on crashing cpu or if 
> crashing cpu is inside die_nmi(), how would it stop/prevent delivery 
> of NMI IPI to other cpus.
> 
> Am I missing something obivious?

i wondered about that too. kexec should be as atomic as it can be - 
enabling interrupts only opens up a window for another crash (more 
memory corruption, etc. etc) to happen.

	Ingo