Re: [RFC PATCH 17/23] watchdog/hardlockup/hpet: Convert the timer's interrupt to NMI

[Ricardo Neri]

On Wed, Jun 13, 2018 at 11:07:20AM +0200, Peter Zijlstra wrote:
> On Tue, Jun 12, 2018 at 05:57:37PM -0700, Ricardo Neri wrote:
> 
> +static bool is_hpet_wdt_interrupt(struct hpet_hld_data *hdata)
> +{
> +       unsigned long this_isr;
> +       unsigned int lvl_trig;
> +
> +       this_isr = hpet_readl(HPET_STATUS) & BIT(hdata->num);
> +
> +       lvl_trig = hpet_readl(HPET_Tn_CFG(hdata->num)) & HPET_TN_LEVEL;
> +
> +       if (lvl_trig && this_isr)
> +               return true;
> +
> +       return false;
> +}
> 
> > +static int hardlockup_detector_nmi_handler(unsigned int val,
> > +					   struct pt_regs *regs)
> > +{
> > +	struct hpet_hld_data *hdata = hld_data;
> > +	unsigned int use_fsb;
> > +
> > +	/*
> > +	 * If FSB delivery mode is used, the timer interrupt is programmed as
> > +	 * edge-triggered and there is no need to check the ISR register.
> > +	 */
> > +	use_fsb = hdata->flags & HPET_DEV_FSB_CAP;
> 
> Please do explain.. That FSB thing basically means MSI. But there's only
> a single NMI vector. How do we know this NMI came from the HPET?

Indeed, I see now that this is wrong. There is no way to know. The only way
is to use an IO APIC interrupt and read the HPET status register.

> 
> > +
> > +	if (!use_fsb && !is_hpet_wdt_interrupt(hdata))
> 
> So you add _2_ HPET reads for every single NMI that gets triggered...
> and IIRC HPET reads are _sllooooowwwwww_.


Since the trigger mode of the HPET timer is not expected to change, 
perhaps is_hpet_wdt_interrupt() can only need the interrupt status
register. This would reduce the reads to one. Furthermore, the hardlockup
detector can skip an X number of NMIs and reduce further the frequency
of reads. Does this make sense?

> 
> > +		return NMI_DONE;
> > +
> > +	inspect_for_hardlockups(regs);
> > +
> > +	if (!(hdata->flags & HPET_DEV_PERI_CAP))
> > +		kick_timer(hdata);
> > +
> > +	/* Acknowledge interrupt if in level-triggered mode */
> > +	if (!use_fsb)
> > +		hpet_writel(BIT(hdata->num), HPET_STATUS);
> > +
> > +	return NMI_HANDLED;
> 
> So if I read this right, when in FSB/MSI mode, we'll basically _always_
> claim every single NMI as handled?
> 
> That's broken.

Yes, this is not correct. I will drop the functionality to use
FSB/MSI mode.

Thanks and BR,
Ricardo