Re: [PATCH] armpmu: broadcast overflow irq on multi-core system having one muxed SPI for PMU.

[Mark Rutland]

On Fri, May 11, 2018 at 08:20:49AM +0900, ïïÈïï wrote:
> Thank you for the reply.
> 
> > -----Original Message-----
> > From: Mark Rutland [mailto:mark.rutland@xxxxxxx]
> > Sent: Thursday, May 10, 2018 7:21 PM
> > To: Hoeun Ryu <hoeun.ryu@xxxxxxxxxxx>
> > Cc: Will Deacon <will.deacon@xxxxxxx>; Hoeun Ryu <hoeun.ryu@xxxxxxx>;
> > linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> > Subject: Re: [PATCH] armpmu: broadcast overflow irq on multi-core system
> > having one muxed SPI for PMU.

> > Muxing the PMU IRQs is a really broken system design, and there's no good
> > way of supporting it.

> > What we should do for such systems is:
> > 
> > * Add a flag to the DT to describe that the IRQs are muxed, as this
> >   cannot be probed.
> > 
> > * Add hrtimer code to periodically update the counters, to avoid
> >   overflow (e.g. as we do in the l2x0 PMU).
> > 
> > * Reject sampling for such systems, as this cannot be done reliably or
> >   efficiently.
> > 
> > NAK to broadcasting the IRQ -- there are a number of issues with the
> > general approach.
> 
> The second solution would be good if sampling is necessary even like those
> systems.

Please note that I mean *all* of the above. There would be no sampling
on systems with muxed PMU IRQs, since there's no correlation between
overflow events and the hrtimer interrupts -- the results of sampling
would be misleading.

> Actually I'm working on FIQ available ARM32 system and trying to enable the
> hard lockup detector by routing the PMU IRQ to FIQ.
> Because of that, I really need the interrupt event if it is a muxed SPI,
> beside I also need to make an dedicated IPI FIQ to broadcast the IRQ in
> this approach.
> What would you do if you were in the same situation ?

I don't think that this can work with a muxed IRQ, sorry.

It would be better to use some kind of timer.

[...]

> > Futher, If you ever encounter a case where you need to avoid preemption
> > across enabling IRQs, preemption must be disabled *before* enabling IRQs.
> 
> Ah, OK.
> Enabling IRQs can cause scheduling tasks in the end of exception or other
> scheduling points, right ?

Yes. If an IRQ was taken *between* enabling IRQs and disabling
preemption, preemption may occur as part of the exception return.

Thanks,
Mark.