kgdb: fix NMI hangs

This patch fixes the hang regression with kgdb when the NMI interrupt
comes in while the master core is returning from an exception.

Adjust the NMI logic such that KGDB will not stop NMI exceptions from
occurring by in general returning NOTIFY_DONE.  It is not possible to
distinguish the debug NMI sync vs the normal NMI apic interrupt so
kgdb needs to catch the unknown NMI if it the debugger was previously
active on one of the cpus.

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

1 files changed, 2 insertions, 0 deletions

diff --git a/arch/x86/kernel/traps_32.c b/arch/x86/kernel/traps_32.c
index b22c01e05a18..c5421f30d678 100644
--- a/arch/x86/kernel/traps_32.c
+++ b/arch/x86/kernel/traps_32.c
@@ -708,6 +708,8 @@ io_check_error(unsigned char reason, struct pt_regs * regs)
 static __kprobes void
 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 {
+        if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
+                return;
 #ifdef CONFIG_MCA
         /* Might actually be able to figure out what the guilty party
         * is. */