Mailing List Archive

Do RTC and HPET need to be included as periodic timers? They are never used
by Linux for ticks (it always uses PIT for jiffy updates) and I think
Windows is more robust against (or at least doesn't noticeably complain
about) time jumps.

-- Keir

On 6/12/06 5:15 am, "Yang, Xiaowei" <xiaowei.yang@intel.com> wrote:

> This patch enables more than one platform timer (PIT/RTC/HPET)
> programmed as periodic timer and adds them to abstract layer, which
> keeps track of pending_intr_nr to avoid time interrupt lost and sync'ed
> timer with TSC.
>
> It also makes some cleanup to the time related code.
>
> Signed-off-by: Xiaowei Yang <xiaowei.yang@intel.com>
> Signed-off-by: Yunhong Jiang yunhong.jiang@intel.com
>
> Thanks,
> Xiaowei
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>Do RTC and HPET need to be included as periodic timers? They are never
used
>by Linux for ticks (it always uses PIT for jiffy updates) and I think
>Windows is more robust against (or at least doesn't noticeably complain
>about) time jumps.
>

For RTC, yes! As we know some version of Windows uses RTC as main time
resource for timekeeping. We noticed if it's not added as periodic
timer, Windows time may go slow in SMP environment.
In case of HPET, the answer may be no, as OS can read the monotonically
increased counter to see time jumps.

And as periodic timer list is per vcpu, it's also helpful to sync one
vcpu's lapic time with its TSC, which is needed when guest calibrates
lapic at least.

Thanks,
Xiaowei

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> >Do RTC and HPET need to be included as periodic timers? They are
never
> used
> >by Linux for ticks (it always uses PIT for jiffy updates) and I think
> >Windows is more robust against (or at least doesn't noticeably
complain
> >about) time jumps.
> >
>
> For RTC, yes! As we know some version of Windows uses RTC as main time
> resource for timekeeping.

I suspect that it only does this because during boot it decides that it
doesn't like the look of our pm_timer (I don't blame it!)

It would be good to invest some effort understanding why it selects RTC
-- I'm sure that it's a fall back.

Ian


We noticed if it's not added as periodic
> timer, Windows time may go slow in SMP environment.
> In case of HPET, the answer may be no, as OS can read the
monotonically
> increased counter to see time jumps.
>
> And as periodic timer list is per vcpu, it's also helpful to sync one
> vcpu's lapic time with its TSC, which is needed when guest calibrates
> lapic at least.
>
> Thanks,
> Xiaowei
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>> >Do RTC and HPET need to be included as periodic timers? They are
>never
>> used
>> >by Linux for ticks (it always uses PIT for jiffy updates) and I
think
>> >Windows is more robust against (or at least doesn't noticeably
>complain
>> >about) time jumps.
>> >
>>
>> For RTC, yes! As we know some version of Windows uses RTC as main
time
>> resource for timekeeping.
>
>I suspect that it only does this because during boot it decides that it
>doesn't like the look of our pm_timer (I don't blame it!)
>
>It would be good to invest some effort understanding why it selects RTC
>-- I'm sure that it's a fall back.
>

I am not sure whether RTC is a fall back choice or not. The below
paragraph is from a paper of timekeeping in VMware. From the text, it
seems the answer is no.

------------
The timer device used and the number of interrupts generated per second
vary depending on which specific version of Microsoft Windows, and which
Windows HAL (hardware architecture layer), is installed. Most
uniprocessor Windows installations use the PIT as their main system
timer, but multiprocessor HALs and some ACPI uniprocessor HALs use the
CMOS periodic timer instead. For systems using the PIT, the base
interrupt rate is usually 100Hz, although Windows 98 uses 200Hz. For
systems that use the CMOS timer, the base interrupt rate is usually
64Hz.
------------

Thanks,
Xiaowei

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel