[PATCH] Quiet corrected errors (CMC/CPE)

[PATCH] Quiet corrected errors (CMC/CPE)

[Hidetoshi Seto]

Hi,

I think we don't always need this message to printk.

Is it better to have a choice between a daemon-A which quickly alerts
a user to the error or a daemon-B which silently takes the error statistic,
or a quiet with no daemon?

...BTW, could anyone tell me a link to the most recent salinfo_decode?


Thanks,
H.Seto

-----

diff -Nur linux-2.6.7/arch/ia64/kernel/mca.c linux-2.6.7-noprintk/arch/ia64/kernel/mca.c
--- linux-2.6.7/arch/ia64/kernel/mca.c  2004-06-24 14:04:32.000000000 +0900
+++ linux-2.6.7-noprintk/arch/ia64/kernel/mca.c 2004-06-24 14:05:14.000000000 +0900
@@ -253,7 +253,7 @@
        salinfo_log_wakeup(sal_info_type, buffer, size, irq_safe);

        if (irq_safe)
-               printk(KERN_INFO "CPU %d: SAL log contains %s error record\n",
+               IA64_MCA_DEBUG("CPU %d: SAL log contains %s error record\n",
                        smp_processor_id(),
                        sal_info_type < ARRAY_SIZE(rec_name) ? rec_name[sal_info_type] : "UNKNOWN");

Re: [PATCH] Quiet corrected errors (CMC/CPE)

[Grant Grundler]

On Thu, Jun 24, 2004 at 03:05:55PM +0900, Hidetoshi Seto wrote:
> ...BTW, could anyone tell me a link to the most recent salinfo_decode?

http://packages.debian.org/unstable/admin/salinfo

grant

Re: [PATCH] Quiet corrected errors (CMC/CPE)

[David Mosberger]

>>>>> On Thu, 24 Jun 2004 15:05:55 +0900, Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com> said:

  Hidetoshi> Hi, I think we don't always need this message to printk.

  Hidetoshi> Is it better to have a choice between a daemon-A which
  Hidetoshi> quickly alerts a user to the error or a daemon-B which
  Hidetoshi> silently takes the error statistic, or a quiet with no
  Hidetoshi> daemon?

OK, with me, but needs blessing from Tony (plus "Signed-off-by" trailers).

	--david

RE: [PATCH] Quiet corrected errors (CMC/CPE)

[Luck, Tony]

Hidetoshi> Hi, I think we don't always need this message to printk.

Hidetoshi> Is it better to have a choice between a daemon-A which
Hidetoshi> quickly alerts a user to the error or a daemon-B which
Hidetoshi> silently takes the error statistic, or a quiet with no
Hidetoshi> daemon?

David>OK, with me, but needs blessing from Tony (plus 
David>"Signed-off-by" trailers).

We definitely need some change here ... a single stuck bit in
a heavily used page of memory that generates a slew of
corrected errors can flood the console log with these messages.

I'm torn between the simplicity of just dropping the printk
completely and trusting the daemon, or adding yet another
rate limiter to make sure that some messages make it, but
we avoid the flood.

If Seto-san wants to go for extra credit and submit a patch
with a rate limiter ... then I'll bless that.  But if he's
busy doing more interesting error recovery and just wants to
stick with the current patch, then I'll provide my blessing
for that too.

-Tony

Re: [PATCH] Quiet corrected errors (CMC/CPE)

[Hidetoshi Seto]

> If Seto-san wants to go for extra credit and submit a patch
> with a rate limiter ... then I'll bless that.  But if he's
> busy doing more interesting error recovery and just wants to
> stick with the current patch, then I'll provide my blessing
> for that too.

The rate limiter is a good idea, but OTOH there is some user who
don't want to see such messages from kernel at all.

After all, current patch is a good quick-fix, I think.


H.Seto

Re: [PATCH] Quiet corrected errors (CMC/CPE)

[Hidetoshi Seto]

[-- Attachment #1: Type: text/plain, Size: 104 bytes --]

Trivial patch to quiet messages about corrected errors.

Signed-off-by: Tony Luck <tony.luck@intel.com>

[-- Attachment #2: mca_noprintk.patch --]
[-- Type: application/octet-stream, Size: 584 bytes --]

diff -Nur linux-2.6.7/arch/ia64/kernel/mca.c linux-2.6.7-noprintk/arch/ia64/kernel/mca.c
--- linux-2.6.7/arch/ia64/kernel/mca.c	2004-06-24 14:04:32.000000000 +0900
+++ linux-2.6.7-noprintk/arch/ia64/kernel/mca.c	2004-06-24 14:05:14.000000000 +0900
@@ -253,7 +253,7 @@
 	salinfo_log_wakeup(sal_info_type, buffer, size, irq_safe);
 
 	if (irq_safe)
-		printk(KERN_INFO "CPU %d: SAL log contains %s error record\n",
+		IA64_MCA_DEBUG("CPU %d: SAL log contains %s error record\n",
 			smp_processor_id(),
 			sal_info_type < ARRAY_SIZE(rec_name) ? rec_name[sal_info_type] : "UNKNOWN");