Re: [BUG] get_rtc_time() triggers NMI watchdog in hpet_rtc_interrupt()

[Alan Jenkins <alan-jenkins@tuffmail.co.uk>]

Ingo Molnar wrote:
> * Alan Jenkins <alan-jenkins@tuffmail.co.uk> wrote:
>
>   
>> I beat you :-).  I noticed the problem myself.  It does trigger if I 
>> run hwclock in a loop.
>>
>> while true; do hwclock; done
>>
>> However it is fine in latest git.  Looks like Ingo's patch has been 
>> merged, and has fixed it.
>>     
>
> ah, good - thanks for testing it. I havent seen a lockup on the T60 
> either - but it was very sporadic to begin with. (and now we understand 
> why: hwclock is running only once per reboot.)
>
> Plus, the real bug has to be found as well: who in the rtc ioctl path 
> calls this code with irqs disabled - it's not nice.
>   

Isn't the problem that it gets called from a timer IRQ?  So irqs are
enabled globally, but the timer IRQ is masked.

 [<c0139cc0>] hrtimer_run_pending+0x20/0x90
 [<c01478a8>] handle_IRQ_event+0x28/0x50
 [<c0148f71>] handle_edge_irq+0xa1/0x120
 [<c010615b>] do_IRQ+0x3b/0x70
 [<c0113225>] smp_apic_timer_interrupt+0x55/0x80
 [<c0103c4f>] common_interrupt+0x23/0x28

Which in turn is probably because the userspace "RTC interrupt"
interface is emulated using RTC reads within the main timer interrupt
handler.

Alan