equivalent of touch_nmi_watchdog

equivalent of touch_nmi_watchdog

[Rajarshi Das]

Hi,
I am running some code which needs interrupts to be disabled for a
reasonable amount of time. A "touch_nmi_watchdog" call is done before
this code gets called on i386 platform. How do I do it on ppc64 ? I am
using a power 3 machine (type 7044-170) (uniprocessor) and a 2.6.6
kernel.

Thanks in advance,
Rajarshi.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: equivalent of touch_nmi_watchdog

[Benjamin Herrenschmidt]

On Tue, 2004-06-29 at 06:49, Rajarshi Das wrote:
> Hi,
> I am running some code which needs interrupts to be disabled for a
> reasonable amount of time. A "touch_nmi_watchdog" call is done before
> this code gets called on i386 platform. How do I do it on ppc64 ? I am
> using a power 3 machine (type 7044-170) (uniprocessor) and a 2.6.6
> kernel.

Can you tell us more about your code ? It's never a good idea to disable
interrupts for more than a few microseconds ...

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: equivalent of touch_nmi_watchdog

[Rajarshi Das]

Its basically dumping code. After a panic has happened, interrupts are
inhibited, all CPUS (except the host cpu) are frozen, and for each page
of data, the dump module writes the dump to disk by calling the host
adapter's interrupt handler till I/O completes.

Rajarshi.

On Tue, 2004-06-29 at 18:36, Benjamin Herrenschmidt wrote:
> On Tue, 2004-06-29 at 06:49, Rajarshi Das wrote:
> > Hi,
> > I am running some code which needs interrupts to be disabled for a
> > reasonable amount of time. A "touch_nmi_watchdog" call is done before
> > this code gets called on i386 platform. How do I do it on ppc64 ? I am
> > using a power 3 machine (type 7044-170) (uniprocessor) and a 2.6.6
> > kernel.
>
> Can you tell us more about your code ? It's never a good idea to disable
> interrupts for more than a few microseconds ...
>
> Ben.
>
>
>


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: equivalent of touch_nmi_watchdog

[Benjamin Herrenschmidt]

On Tue, 2004-06-29 at 09:00, Rajarshi Das wrote:
> Its basically dumping code. After a panic has happened, interrupts are
> inhibited, all CPUS (except the host cpu) are frozen, and for each page
> of data, the dump module writes the dump to disk by calling the host
> adapter's interrupt handler till I/O completes.

Hrm... a bit scary but might actually work. I don't think we have
a clean interface for whatever watchdog service HV may provide
though, but I'll let others comment on that.

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: equivalent of touch_nmi_watchdog

[Anton Blanchard]

Hi,

> I am running some code which needs interrupts to be disabled for a
> reasonable amount of time. A "touch_nmi_watchdog" call is done before
> this code gets called on i386 platform. How do I do it on ppc64 ? I am
> using a power 3 machine (type 7044-170) (uniprocessor) and a 2.6.6
> kernel.

Check out arch/ppc64/xmon/xmon.c: disable_surveillance() in a recent 2.6
kernel. It might be worth moving that function into rtas.c if you want
to use it as well.

Anton

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/