[PATCH -tip] x86, mce: CE in last bank prevents panic by unknownMCE

From: Hidetoshi Seto
Date: Wed Aug 26 2009 - 03:21:35 EST


[based on tip/x86/mce]

If MCE handler is called but none of mces_seen have machine check event
which might signal the MCE (i.e. event higher than MCE_KEEP_SEVERITY),
panic with "Machine check from unknown source" will be taken since the
MCE is assumed to be signaled from external agent or so.

Usually mces_seen never point MCE_KEEP_SEVERITY event such as CE.
But it can happen because initial value of mces_seen is accidentally
modified by mce_no_way_out() - in case if mce_no_way_out() run through
all banks and the last bank has the CE, mces_seen points the CE and
the "panic by unknown" will not be taken.

This patch fix this undesired behavior, and clarify the logic.

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@xxxxxxxxxxxxxx>
Reported-by: Jin Dongming <jin.dongming@xxxxxxxxxxxxxxxxxx>

---
arch/x86/kernel/cpu/mcheck/mce.c | 6 +++---
1 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index 54bd1b2..7b485e9 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -665,7 +665,7 @@ static void mce_reign(void)
* No machine check event found. Must be some external
* source or one CPU is hung. Panic.
*/
- if (!m && tolerant < 3)
+ if (global_worst <= MCE_KEEP_SEVERITY && tolerant < 3)
mce_panic("Machine check from unknown source", NULL, NULL);

/*
@@ -889,11 +889,11 @@ void do_machine_check(struct pt_regs *regs, long error_code)
mce_setup(&m);

m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS);
- no_way_out = mce_no_way_out(&m, &msg);
-
final = &__get_cpu_var(mces_seen);
*final = m;

+ no_way_out = mce_no_way_out(&m, &msg);
+
barrier();

/*
--
1.6.4.1


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/