From: Aristeu Rozanski <arozansk@redhat.com> Subject: [PATCH RHEL5.1] k8_edac: don't panic on PCC check Date: Mon, 4 Jun 2007 10:16:27 -0400 Bugzilla: 237950 Message-Id: <20070604141627.GB31131@redhat.com> Changelog: [edac] k8_edac: don't panic on PCC check Processor Context Corrupt check (PCC) is part of the generic machine check infrastructure and it defines that the check handler doesn't has enough information to resume the instruction stream. Currently, the EDAC layer checks for errors using polling and panics if PCC bit is set. The problem is that PCC bit doesn't has the same meaning if read outside a machine check handler. This patch makes k8_edac driver to print a warning instead of calling panic(). IBM is hitting this problem while using kdump in BZ#237950. This problem has been discussed on: http://sourceforge.net/mailarchive/message.php?msg_id=461B2D2C.40301%40in.ibm.com and http://sourceforge.net/mailarchive/forum.php?thread_name=20070424100935.GA3039%40in.ibm.com&forum_name=bluesmoke-devel And it was commited to bluesmoke CVS (k8_edac is not upstream yet): http://bluesmoke.cvs.sourceforge.net/bluesmoke/bluesmoke/edac/patches/driver-edac-new-k8-rev-f.patch?r1=1.1.2.7&r2=1.1.2.8&pathrev=edac Please ACK --- linux-2.6.18.noarch.orig/drivers/edac/k8_edac.c 2007-05-07 15:39:45.000000000 -0400 +++ linux-2.6.18.noarch/drivers/edac/k8_edac.c 2007-05-07 15:49:39.000000000 -0400 @@ -1750,7 +1750,7 @@ } if (regs->nbsh & BIT(25)) - panic("MC%d: processor context corrupt", mci->mc_idx); + k8_mc_printk(mci, KERN_CRIT, "processor context corrupt\n"); return 1; }