Fernando,
Thanks very much for your help. I read the entire 5105 kernel bug
discussion and didn't find the valuable information you presented,
i.e. how to change the clocksource at runtime so easily. I was going
to try the "notsc" kernel option. For what it's worth I changed the
clocksource as you mentioned and I haven't had a glitch yet (10
minutes since the change).. and this is unusual.
Aaron
On 11/21/05, Fernando Lopez-Lezcano <nando@email-addr-hidden> wrote:
> On Mon, 2005-11-21 at 15:25 -0500, Aaron Phillips wrote:
> > > Which cpu do you have?
> > I have AMD X2 also, the AMD X2 3800, interesting...
> >
> > > I have seen this coupled also to keyboard repeats
> > > being too fast because of kernel problems in keeping time (in my case
> > > related to using AMD X2 dual core processors).
> > Really? well should I turn down repeats in the BIOS or in Linux?
> >
> > > The X blackouts were actually the screensaver kicking in. It is supposed to be fixed...
> > In which version is it supposed to be fixed. I am using the latest
> > and greatest 2.6.14 plus the latest RT patch (rt13). The only thing I
> > don't quite understand is what to do with the post-release version
> > increments such as 2.6.14.2.. how do you patch that?
> >
> > Which kernel are you running and do you have the same problem still?
>
> I still have the problem and it is not fixed yet (just earlier today I
> think I finally understood what is actually happening).
>
> At this point this only happens on dual core systems (for example the
> Athlon X2 processor family) running an SMP i386 kernel, it does not
> happen on x86_64 systems. A patch has just been posted to lkml (by John
> Stultz) that should take care of this, but it will take a while for the
> stuff to get to the -rtxx patches.
>
> There is a workaround (thanks to John Stultz) which is to change the
> source clock for the timing to something other than the TSC[*]. You can
> see which source is being used by typing:
>
> cat /sys/devices/system/clocksource/clocksource0/clocksource
>
> The default should show something like this:
>
> acpi_pm jiffies *tsc pit
>
> note the "*" next to tsc, means it is the one being used.
> If you have acpi_pm change the source to it like this:
>
> echo "acpi_pm">/sys/devices/system/clocksource/clocksource0/clocksource
>
> That should fix the problem you are having.
>
> You can also add "clocksource=xyz" as a boot time kernel parameter,
> where xyz is the source you want to use.
>
> -- Fernando
>
> [*] "Time Stamp Clock": on Athlon X2 systems each cpu has its own and
> they can drift from each other, software that does not take this into
> account gets into big trouble, for example Jack.
>
>
>
Received on Tue Nov 22 12:15:06 2005
This archive was generated by hypermail 2.1.8 : Tue Nov 22 2005 - 12:15:06 EET