[RFC PATCH] arch/arm64: put crash_save_cpu() at last before die

[Ding Hui]

When we debug a vmcore by crash utility, sometimes we got
backtraces on non-panic core like this:

crash> bt -c 1
PID: 0      TASK: ffff80014107a000  CPU: 1   COMMAND: "swapper/1"
 #0 [1ffff0002f73c7dc] crash_save_cpu at ffff20002031cb68

The crash utility do backtrace by search LR from stack,
which pushed by callee of crash_save_cpu().

So the LR is easier to be destroyed by subsequent invoking other
functions, and lead to crash failed to backtrace on non-panic core

I tried to put crash_save_cpu() after sdei_mask_local_cpu(), and
it turns out bt is working to me.

But I'm not sure the order of crash_save_cpu() is whether strictly
required, so I send this RFC patch to get some suggestion.

Signed-off-by: Ding Hui <dinghui@xxxxxxxxxxxxxx>
---
 arch/arm64/kernel/smp.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
index ffc5d76cf695..5f1defe76f3d 100644
--- a/arch/arm64/kernel/smp.c
+++ b/arch/arm64/kernel/smp.c
@@ -842,13 +842,13 @@ static atomic_t waiting_for_crash_ipi = ATOMIC_INIT(0);
 static void ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs)
 {
 #ifdef CONFIG_KEXEC_CORE
-	crash_save_cpu(regs, cpu);
-
 	atomic_dec(&waiting_for_crash_ipi);
 
 	local_irq_disable();
 	sdei_mask_local_cpu();
 
+	crash_save_cpu(regs, cpu);
+
 	if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
 		__cpu_try_die(cpu);
 
-- 
2.17.1