Re: [PATCH 2.5] report unknown NMI reasons only once

Re: [PATCH 2.5] report unknown NMI reasons only once

[Pascal Schmidt]

On Sun, 20 Apr 2003 01:00:19 +0200, you wrote in linux.kernel:

>> Those NMIs happen only rarely when the machine is lightly loaded, but
>> under load, I get several of them per second. This quickly makes
>> /var/log/messages grow.
> 
> I guess they are overheat traps then

At 41 degrees C reported CPU temperature, I don't think so. Even if this
is off by, say, 10 degrees, it's still not a problem. AMD allows 90 degrees
for my CPU type (Athlon XP 1700+, Thoroughbred core).

Maybe I should add that "under load" in this case means running OpenGL
games using DRI on a Voodoo4 card.

>> I don't think reporting any of those NMIs more than once provides
>> valuable information, so I've cooked up a patch which only reports each
>> unknown NMI reason once.
> 
> Its sitting there saying "Something is wrong" "Something is still
> wrong".

Beats me as to what could be wrong. It's not a memory problem and the
CPU does not overheat.

I'll go patch the kernel for my personal use then, but I'm not the only
one seeing those messages without any system problems.

-- 
Ciao,
Pascal

Re: [PATCH 2.5] report unknown NMI reasons only once

[Zwane Mwaikambo]

On Sun, 20 Apr 2003, Pascal Schmidt wrote:

> Beats me as to what could be wrong. It's not a memory problem and the
> CPU does not overheat.
> 
> I'll go patch the kernel for my personal use then, but I'm not the only
> one seeing those messages without any system problems.

It's all fun and games until NMIs turn into MCEs...

 \|/ ____ \|/
"@'/ ,. \`@"
/_| \__/ |_\
   \__U_/

Re: [PATCH 2.5] report unknown NMI reasons only once

[Pascal Schmidt]

On Sun, 20 Apr 2003 07:20:07 +0200, you wrote:

>> Beats me as to what could be wrong. It's not a memory problem and the
>> CPU does not overheat.
>> 
>> I'll go patch the kernel for my personal use then, but I'm not the only
>> one seeing those messages without any system problems.
> 
> It's all fun and games until NMIs turn into MCEs...

I have the MCE polling stuff enabled and will keep an eye on it. So far
I suspect flaky motherboard design (Athlon XPs didn't even exist when
this piece of hardware was designed).

It's definitely CPU-related since it never happened with the Duron
processor that I used before.

-- 
Ciao,
Pascal

Re: [PATCH 2.5] report unknown NMI reasons only once

[John Bradford]

> >> Beats me as to what could be wrong. It's not a memory problem and the
> >> CPU does not overheat.
> >> 
> >> I'll go patch the kernel for my personal use then, but I'm not the only
> >> one seeing those messages without any system problems.
> > 
> > It's all fun and games until NMIs turn into MCEs...
> 
> I have the MCE polling stuff enabled and will keep an eye on it. So far
> I suspect flaky motherboard design (Athlon XPs didn't even exist when
> this piece of hardware was designed).
> 
> It's definitely CPU-related since it never happened with the Duron
> processor that I used before.

Are you sure that the CPU voltage is correct?

John.

Re: [PATCH 2.5] report unknown NMI reasons only once

[Pascal Schmidt]

On Sun, 20 Apr 2003, John Bradford wrote:

> > It's definitely CPU-related since it never happened with the Duron
> > processor that I used before.
> 
> Are you sure that the CPU voltage is correct?

The motherboard is in autodetect mode...

According to AMD's docs, my CPU wants 1.50V.
According to the BIOS, what it gets is 1.48V.

AMD does not specify a tolerance for Vcore, but I can try to manually
set the motherboard for 1.50V and see if the NMIs disappear.

-- 
Ciao,
Pascal

[PATCH 2.5] report unknown NMI reasons only once

[Pascal Schmidt]

Hi people!

On my machine (Athlon XP on an old ALi Magik1 motherboard), I get tons of
messages like:

Uhhuh. NMI received for unknown reason 2d on CPU 0.
Dazed and confused, but trying to continue
Do you have a strange power saving mode enabled?

There's nothing I can do about them (ACPI or APIC support in the kernel or
not does not make a difference, nor do any of the BIOS power management
settings). The machine runs fine, no problems in memtest86, and I've not
noticed a problem with a system in the three or four months that I'm
running this configuration.

Those NMIs happen only rarely when the machine is lightly loaded, but
under load, I get several of them per second. This quickly makes
/var/log/messages grow.

I don't think reporting any of those NMIs more than once provides
valuable information, so I've cooked up a patch which only reports each
unknown NMI reason once.

I don't know who the maintainer of arch/i386/kernel/traps.c is supposed
to be, so I'm sending this to the list and CC to Alan since he is not
unlikely to take small patches like this. ;)

Patch against 2.5.67-bk10. Compiles, boots and does what it's supposed to
do on my machine.

Comments, anyone?

--- linux-2.5.67-bk10/arch/i386/kernel/traps.c	Sat Apr 19 20:00:42 2003
+++ work/arch/i386/kernel/traps.c	Sat Apr 19 20:17:40 2003
@@ -5,6 +5,8 @@
  *
  *  Pentium III FXSR, SSE support
  *	Gareth Hughes <gareth@valinux.com>, May 2000
+ *  Limit unknown NMI reporting
+ *     Pascal Schmidt <der.eremit@email.de>, April 2003
  */
 
 /*
@@ -48,6 +50,8 @@
 #include <asm/pgalloc.h>
 #include <asm/arch_hooks.h>
 
+#include <asm/bitops.h>
+
 #include <linux/irq.h>
 #include <linux/module.h>
 
@@ -417,6 +421,9 @@ static void io_check_error(unsigned char
 	outb(reason, 0x61);
 }
 
+/* bit field for already reported unknown NMI reasons */
+static int unknown_nmi_reported[8] = { 0, 0, 0, 0, 0, 0, 0, 0 };
+
 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 {
 #ifdef CONFIG_MCA
@@ -427,10 +434,13 @@ static void unknown_nmi_error(unsigned c
 		return;
 	}
 #endif
-	printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
-		reason, smp_processor_id());
-	printk("Dazed and confused, but trying to continue\n");
-	printk("Do you have a strange power saving mode enabled?\n");
+	if ( !test_bit(reason, (void *)&unknown_nmi_reported) ) {
+		printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
+			reason, smp_processor_id());
+		printk("Dazed and confused, but trying to continue\n");
+		printk("Do you have a strange power saving mode enabled?\n");
+		__set_bit(reason, (void *)&unknown_nmi_reported);
+	}
 }
 
 static void default_do_nmi(struct pt_regs * regs)

-- 
Ciao,
Pascal

Re: [PATCH 2.5] report unknown NMI reasons only once

[Alan Cox]

On Sad, 2003-04-19 at 19:21, Pascal Schmidt wrote:
> Those NMIs happen only rarely when the machine is lightly loaded, but
> under load, I get several of them per second. This quickly makes
> /var/log/messages grow.

I guess they are overheat traps then

> I don't think reporting any of those NMIs more than once provides
> valuable information, so I've cooked up a patch which only reports each
> unknown NMI reason once.

Its sitting there saying "Something is wrong" "Something is still
wrong". By all means kill it on your box, but this is not good for
general consumption.