On Wed, Feb 15, 2017 at 1:34 PM, Michael Tokarev <m...@tls.msk.ru> wrote:
> 06.02.2017 17:44, Igor Pavlikevich wrote:
>> TCO timer does not actually stop
>>
>> Signed-off-by: Igor Pavlikevich <ipavlikev...@gmail.com>
>> ---
>> hw/acpi/tco.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/hw/acpi/tco.c b/hw/acpi/tco.c
>> index 8ce7daf..b4adac8 100644
>> --- a/hw/acpi/tco.c
>> +++ b/hw/acpi/tco.c
>> @@ -49,6 +49,7 @@ static inline void tco_timer_reload(TCOIORegs *tr)
>> static inline void tco_timer_stop(TCOIORegs *tr)
>> {
>> tr->expire_time = -1;
>> + timer_del(tr->tco_timer);
>> }
>>
>> static void tco_timer_expired(void *opaque)
>>
>
> I'm not sure this is really trivial change.
> Can you elaborate a bit please?
When linux guest goes to first reboot, it calls
watchdog_open()->watchdog_start()->iTCO_wdt_start() and write 0 to
TCO1_CNT.
After that can_start_tco_timer() returns true, starts timer in
tco_timer_reload() and guest kernel starts flooding
[ 590.392055] Uhhuh. NMI received for unknown reason 30 on CPU 0.
[ 590.392055] Do you have a strange power saving mode enabled?
[ 590.392055] Dazed and confused, but trying to continue
[ 620.392035] Uhhuh. NMI received for unknown reason 20 on CPU 0.
[ 620.392035] Do you have a strange power saving mode enabled?
[ 620.392035] Dazed and confused, but trying to continue
Timer keeps reloading after ich9_generate_nmi() so there is only one
way to stop reloading by clearing ICH9_CC_GCS_NO_REBOOT and reset
guest via watchdog_perform_action(), becasue tco_timer_stop() does
not actually modify tr->tco_timer.
Tested on Debian wheezy/jessie guests with default kernels and qemu
2.5+ q35 machine.
>
> And please Cc qemu-devel@ too (doing this now).
>
> Thanks,
>
> /mjt
