From: Don Zickus <dzickus@redhat.com>
Subject: [RHEL-5.1 PATCH] x86: allow edac to panic with memory corruption on non-kdump kernels
Date: Thu, 12 Jul 2007 12:15:44 -0400
Bugzilla: 237950
Message-Id: <20070712161544.GL22926@redhat.com>
Changelog: [edac] allow edac to panic with memory corruption on non-kdump kernels


https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=237950

This patch reverts part of a previous patch that disables a memory
corruption failure on a non-kdump kernel using the edac module.

The initial reason for disabling the panic was that kdump kernels were
panicing in the AGP subsystem, most likely due to in-flight dma requests
from the previous kernel.  This breaks the idea of kdump.  So a change to
a printk was made.

Upon thinking about it, this is really a bad idea.  Memory corruption is
something that kernel should _not_ continue to run with.  The hack below
allows the panic for non-kdump kernels, but continues to only offer
printks for the kdump kernel on x86 arches.

Hopefully, a more correct solution will present itself for 5.2 but that is
still being fleshed out upstream and will not be ready for 5.1.  Hence the
temporary solution proposed here.

Please ACK for 5.1.  Yeah this is about as late as it gets...

Thanks to Alan for noticing this flaw.

Thanks to Konrad for testing on a system that could reproduce the original
problem.

Cheers,
Don


--- linux-2.6.18.noarch/drivers/edac/k8_edac.c.orig	2007-06-22 11:05:00.000000000 -0400
+++ linux-2.6.18.noarch/drivers/edac/k8_edac.c	2007-06-22 11:05:06.000000000 -0400
@@ -1749,8 +1749,12 @@
 		edac_mc_handle_ue_no_info(mci, "UE bit is set\n");
 	}
 
-	if (regs->nbsh & BIT(25))
-		k8_mc_printk(mci, KERN_CRIT, "processor context corrupt\n");
+	if (regs->nbsh & BIT(25)) {
+		if (reset_devices == 0)
+			panic("MC%d: processor context corrupt", mci->mc_idx);
+		else
+			k8_mc_printk(mci, KERN_CRIT, "processor context corrupt\n");
+	}
 
 	return 1;
 }