[PATCH v6 1/5] powerpc/85xx: implement hardware timebase sync

Zhao Chenhui-B35336 B35336 at freescale.com
Fri Jun 29 20:33:45 EST 2012


> -----Original Message-----
> From: Linuxppc-dev [mailto:linuxppc-dev-bounces+chenhui.zhao=freescale.com at lists.ozlabs.org] On Behalf
> Of Kumar Gala
> Sent: Friday, June 29, 2012 2:30 AM
> To: Zhao Chenhui-B35336
> Cc: Wood Scott-B07421; linuxppc-dev at lists.ozlabs.org list; linux-kernel at vger.kernel.org list
> Subject: Re: [PATCH v6 1/5] powerpc/85xx: implement hardware timebase sync
> 
> 
> On Jun 28, 2012, at 5:50 AM, Benjamin Herrenschmidt wrote:
> 
> > On Thu, 2012-06-28 at 11:38 +0800, Zhao Chenhui wrote:
> >>
> >>
> >> The bootloader have done a timebase sync. If we do not need KEXEC or
> >> HOTPLUG_CPU feature, it is unnecessary to do it again at boot time of
> >> kernel. I only compile the timebase sync routines
> >> when users enable KEXEC or HOTPLUG_CPU.
> >
> > Still, how much are you really saving ? Is it worth the added mess and
> > loss of test coverage ?
> >
> > We have too many conditional stuff like that already.
> >
> > Cheers,
> > Ben.
> >
> 
> I'd also be interested to know how long it actually takes to do time base sync this way.  Since you
> are freezing the timers for some period how long does it really take between the freeze/unfreeze in
> mpc85xx_give_timebase()
> 
> +	mpc85xx_timebase_freeze(1);
> ...
> +	mpc85xx_timebase_freeze(0);
> 
> You can use ATBL/U as a way to see # of cycles taken.
> 
> - k

I measured it using ATBL on MPC8572DS with 1.5GHz core frequency and 600MHz CCB frequency.
The average of 10 times is 1019 clock. It seems that most of the time spent by isync and msync.

-Chenhui



More information about the Linuxppc-dev mailing list