Re: [PATCH] NMI watchdog: setup before enabling NMI watchdog

[Aristeu Rozanski <aris@redhat.com>]

> Hi Aristeu,
> 
> thanks for the patch! I may be _absolutely_ wrong but could you
> explain me how we reach this site in traps
> 
> ---
> 	if (!(reason & 0xc0)) {
> 		if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason,
> 				2, SIGINT) == NOTIFY_STOP)
> 			return;
> 		/*
> 		 * Ok, so this is none of the documented NMI sources,
> 		 * so it must be the NMI watchdog.
> 			 ^^^^
> 		 */
> 		if (nmi_watchdog_tick(regs, reason))
> 			return;
> 		if (!do_nmi_callback(regs, cpu))
> 			unknown_nmi_error(reason, regs);
> 
> 		return;
> 	}
> ---
> 
> not having masked APIC registers as NMI entry yet (which is done during
> perfctl initialization)?
actually the comment is a bit misleading. we can get other "undocumented"
NMIs from different sources. Notice that if the nmi_watchdog_tick() doesn't
identifies it as a performance counter generated NMI (if LAPIC based, IOAPIC
always assume that the NMI is for the NMI watchdog), a default NMI callback
will be tried and if it fails, unknown_nmi_error() will be called. The first
case comes to my head is those NMI buttons present on development machines.

-- 
Aristeu