2004-03-13 21:43:04

by Joerg Sommrey

[permalink] [raw]
Subject: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

Hello,

in my box (Tyan Tiger MPX / 2x AMD Athlon) the NMI watchdog never worked
on any kernel that I tried (2.4.x, 2.6.x). I always found:
| activating NMI Watchdog ... done.
| testing NMI watchdog ... CPU#0: NMI appears to be stuck!

But there is one exception: 2.6.3-mm4 shows:
| activating NMI Watchdog ... done.
| testing NMI watchdog ... OK.

[2.6.3-mm4 was the only -mmX kernel I tried so far.]

With 2.6.4-mm1 the NMI watchdog is again not functional in my box. Any
ideas?

-jo

--
-rw-r--r-- 1 jo users 80 2004-03-13 21:46 /home/jo/.signature


2004-03-14 00:51:11

by Andrew Morton

[permalink] [raw]
Subject: Re: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

Joerg Sommrey <[email protected]> wrote:
>
> n my box (Tyan Tiger MPX / 2x AMD Athlon) the NMI watchdog never worked
> on any kernel that I tried (2.4.x, 2.6.x). I always found:
> | activating NMI Watchdog ... done.
> | testing NMI watchdog ... CPU#0: NMI appears to be stuck!
>
> But there is one exception: 2.6.3-mm4 shows:
> | activating NMI Watchdog ... done.
> | testing NMI watchdog ... OK.
>
> [2.6.3-mm4 was the only -mmX kernel I tried so far.]
>
> With 2.6.4-mm1 the NMI watchdog is again not functional in my box. Any
> ideas?

Please try the next -mm. If that still has problems I'll drop both the
remaining NMI patches and would request that Phillippe ask you to test any
future NMI patches on that machine.

2004-03-14 12:12:12

by Mikael Pettersson

[permalink] [raw]
Subject: Re: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

On Sat, 13 Mar 2004 22:42:55 +0100, Joerg Sommrey wrote:
>in my box (Tyan Tiger MPX / 2x AMD Athlon) the NMI watchdog never worked
>on any kernel that I tried (2.4.x, 2.6.x). I always found:
>| activating NMI Watchdog ... done.
>| testing NMI watchdog ... CPU#0: NMI appears to be stuck!
>
>But there is one exception: 2.6.3-mm4 shows:
>| activating NMI Watchdog ... done.
>| testing NMI watchdog ... OK.
>
>[2.6.3-mm4 was the only -mmX kernel I tried so far.]
>
>With 2.6.4-mm1 the NMI watchdog is again not functional in my box. Any
>ideas?

Insufficient data. Please try a standard 2.6.4 or 2.4.25 kernel
and provide the complete dmesg boot log and the .config used.

nmi_watchdog=1 may be broken on some chipsets, but nmi_watchdog=2
should work, at least in a standard kernel with oprofile disabled.

/Mikael

2004-03-14 16:13:59

by Joerg Sommrey

[permalink] [raw]
Subject: Re: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

On Sun, Mar 14, 2004 at 01:12:05PM +0100, Mikael Pettersson wrote:
> On Sat, 13 Mar 2004 22:42:55 +0100, Joerg Sommrey wrote:
> >in my box (Tyan Tiger MPX / 2x AMD Athlon) the NMI watchdog never worked
> >on any kernel that I tried (2.4.x, 2.6.x). I always found:
> >| activating NMI Watchdog ... done.
> >| testing NMI watchdog ... CPU#0: NMI appears to be stuck!
> >
> >But there is one exception: 2.6.3-mm4 shows:
> >| activating NMI Watchdog ... done.
> >| testing NMI watchdog ... OK.
> >
> >[2.6.3-mm4 was the only -mmX kernel I tried so far.]
> >
> >With 2.6.4-mm1 the NMI watchdog is again not functional in my box. Any
> >ideas?
>
> Insufficient data. Please try a standard 2.6.4 or 2.4.25 kernel
> and provide the complete dmesg boot log and the .config used.
In the past I tried several plain vanilla kernels: 2.4.18/19/22/23,
2.6.1/2/3/4, some -ac kernels and at last the two -mm kernels mentioned.
>
> nmi_watchdog=1 may be broken on some chipsets, but nmi_watchdog=2
> should work, at least in a standard kernel with oprofile disabled.

nmi_watchdog=2 has never worked for me. Is this really supposed to work
on a SMP machine? In that case there isn't even a message
about activating the watchdog, but I get a nmi-count in /proc/interrupts.

I almost gave up, until I found the watchdog working in 2.6.3-mm4.
See attached files for further information.

-jo
--
-rw-r--r-- 1 jo users 80 2004-03-14 13:12 /home/jo/.signature


Attachments:
(No filename) (1.42 kB)
dmesg-2.6.3-mm4 (9.69 kB)
dmesg from 2.6.3-mm4
dmesg-2.6.4_nmi=1 (8.98 kB)
dmesg from 2.6.4 nmi_watchdog=1
dmesg-2.6.4_nmi=2 (8.89 kB)
dmesg from 2.6.4 nmi_watchdog=2
config-2.6.4 (25.41 kB)
.config for 2.6.4
Download all attachments

2004-03-14 17:10:24

by Philippe Elie

[permalink] [raw]
Subject: Re: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

On Sun, 14 Mar 2004 at 17:12 +0000, Joerg Sommrey wrote:

> nmi_watchdog=2 has never worked for me. Is this really supposed to work
> on a SMP machine? In that case there isn't even a message
> about activating the watchdog, but I get a nmi-count in /proc/interrupts.

nmi_watchdog=2 is not tested on SMP so it works w/o any notice.

regards,
Phil

2004-03-14 20:52:11

by Zwane Mwaikambo

[permalink] [raw]
Subject: Re: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

On Sun, 14 Mar 2004, Joerg Sommrey wrote:

> > >With 2.6.4-mm1 the NMI watchdog is again not functional in my box. Any
> > >ideas?

I think all we did was drop the NMI frequency in -mm1, but that would be
for nmi_watchdog=2

> nmi_watchdog=2 has never worked for me. Is this really supposed to work
> on a SMP machine? In that case there isn't even a message
> about activating the watchdog, but I get a nmi-count in /proc/interrupts.

It is activated, it's just not verbal about it, nmi_watchdog=1 is more
verbose because there are a number of interrupt delivery modes we try
before giving up.

> I almost gave up, until I found the watchdog working in 2.6.3-mm4.
> See attached files for further information.

I think the pending problem then is nmi_watchdog=1 working erratically.
Does 2.6.3 work for you?

2004-03-14 21:01:14

by Joerg Sommrey

[permalink] [raw]
Subject: Re: NMI watchdog in 2.6.3-mm4/2.6.4-mm1

On Sun, Mar 14, 2004 at 03:52:09PM -0500, Zwane Mwaikambo wrote:

>
> I think the pending problem then is nmi_watchdog=1 working erratically.
> Does 2.6.3 work for you?

Plain 2.6.3 with nmi_watchdog=1 does not work, 2.6.3-mm1 is the only
working kernel I have seen.

-jo

--
-rw-r--r-- 1 jo users 80 2004-03-14 21:18 /home/jo/.signature