[IA64] kdump on INIT needs multi-nodes sync-up (v.2)

The current implementation of kdump on INIT events would enter
kdump processing on DIE_INIT_MONARCH_ENTER and DIE_INIT_SLAVE_ENTER
events. Thus, the monarch cpu would go ahead and boot up the kdump

On SN shub2 systems, this out-of-sync situation causes some slave
cpus on different nodes to enter POD.

This patch moves kdump entry points to DIE_INIT_MONARCH_LEAVE and
DIE_INIT_SLAVE_LEAVE. It also sets kdump_in_progress variable in
the DIE_INIT_MONARCH_PROCESS event to not dump all active stack
traces to the console in the case of kdump.

I have tested this patch on an SN machine and a HP RX2600.

Signed-off-by: Jay Lan <jlan@sgi.com>
Acked-by: Zou Nan hai <nanhai.zou@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

1 files changed, 4 insertions, 0 deletions

diff --git a/arch/ia64/kernel/mca.c b/arch/ia64/kernel/mca.c
index 26814de6c29a..93e82e4d99d6 100644
--- a/arch/ia64/kernel/mca.c
+++ b/arch/ia64/kernel/mca.c
@@ -1477,6 +1477,10 @@ default_monarch_init_process(struct notifier_block *self, unsigned long val, voi
         struct task_struct *g, *t;
         if (val != DIE_INIT_MONARCH_PROCESS)
                 return NOTIFY_DONE;
+#ifdef CONFIG_KEXEC
+        if (atomic_read(&kdump_in_progress))
+                return NOTIFY_DONE;
+#endif
 
         /*
          * FIXME: mlogbuf will brim over with INIT stack dumps.