[PATCH v2 1/3] powerpc: use NMI IPI for smp_send_stop

[Nicholas Piggin]

On Tue, 3 Apr 2018 10:13:26 +1000
Balbir Singh <bsingharora at gmail.com> wrote:

> On Sun,  1 Apr 2018 20:36:13 +1000
> Nicholas Piggin <npiggin at gmail.com> wrote:
> 
> > Use the NMI IPI rather than smp_call_function for smp_send_stop.
> > Have stopped CPUs hard disable interrupts rather than just soft
> > disable.
> > 
> > This function is used in crash/panic/shutdown paths to bring other
> > CPUs down as quickly and reliably as possible, and minimizing their
> > potential to cause trouble.
> > 
> > Avoiding the Linux smp_call_function infrastructure and (if supported)
> > using true NMI IPIs makes this more robust.
> > 
> > Signed-off-by: Nicholas Piggin <npiggin at gmail.com>
> > ---
> >  arch/powerpc/kernel/smp.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> > 
> > diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
> > index cfc08b099c49..db88660bf6bd 100644
> > --- a/arch/powerpc/kernel/smp.c
> > +++ b/arch/powerpc/kernel/smp.c
> > @@ -565,7 +565,11 @@ void crash_send_ipi(void (*crash_ipi_callback)(struct pt_regs *))
> >  }
> >  #endif
> >  
> > +#ifdef CONFIG_NMI_IPI
> > +static void stop_this_cpu(struct pt_regs *regs)
> > +#else
> >  static void stop_this_cpu(void *dummy)
> > +#endif
> >  {
> >  	/* Remove this CPU */
> >  	set_cpu_online(smp_processor_id(), false);
> > @@ -577,7 +581,11 @@ static void stop_this_cpu(void *dummy)
> >  
> >  void smp_send_stop(void)
> >  {
> > +#ifdef CONFIG_NMI_IPI
> > +	smp_send_nmi_ipi(NMI_IPI_ALL_OTHERS, stop_this_cpu, 1000000);  
> 
> I wonder if the delay_us should be a function of number of cpus and the
> callee should figure this out on its own? May be not in this series, but
> in the longer run.

It possibly should. The big per-cpu/core serialized delay is in firmware
(worst case we have to do a special wakeup on each core and bring it out
of deep stop), and that gets done before the delay starts. So we count
delay from after all threads are woken and given their sreset command, so
this is probably okay even for very big systems.

But it should at least be a #define constant if not something more
sophisticated like you suggest. We have a few delays like that including
in xmon and other crash paths that could use some more thought.

Thanks,
Nick


> 
> 
> Balbir Singh.