diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
index 1f2ccccaf009751a0ed830c6a91bfa4ea60278e3..4231034b8128de0265caf61b5f55967f7241f791 100644
--- a/arch/arm/kernel/smp.c
+++ b/arch/arm/kernel/smp.c
@@ -211,6 +211,13 @@ void __cpuinit __cpu_die(unsigned int cpu)
 	}
 	printk(KERN_NOTICE "CPU%u: shutdown\n", cpu);
 
+	/*
+	 * platform_cpu_kill() is generally expected to do the powering off
+	 * and/or cutting of clocks to the dying CPU.  Optionally, this may
+	 * be done by the CPU which is dying in preference to supporting
+	 * this call, but that means there is _no_ synchronisation between
+	 * the requesting CPU and the dying CPU actually losing power.
+	 */
 	if (!platform_cpu_kill(cpu))
 		printk("CPU%u: unable to kill\n", cpu);
 }
@@ -230,14 +237,41 @@ void __ref cpu_die(void)
 	idle_task_exit();
 
 	local_irq_disable();
-	mb();
 
-	/* Tell __cpu_die() that this CPU is now safe to dispose of */
+	/*
+	 * Flush the data out of the L1 cache for this CPU.  This must be
+	 * before the completion to ensure that data is safely written out
+	 * before platform_cpu_kill() gets called - which may disable
+	 * *this* CPU and power down its cache.
+	 */
+	flush_cache_louis();
+
+	/*
+	 * Tell __cpu_die() that this CPU is now safe to dispose of.  Once
+	 * this returns, power and/or clocks can be removed at any point
+	 * from this CPU and its cache by platform_cpu_kill().
+	 */
 	RCU_NONIDLE(complete(&cpu_died));
 
 	/*
-	 * actual CPU shutdown procedure is at least platform (if not
-	 * CPU) specific.
+	 * Ensure that the cache lines associated with that completion are
+	 * written out.  This covers the case where _this_ CPU is doing the
+	 * powering down, to ensure that the completion is visible to the
+	 * CPU waiting for this one.
+	 */
+	flush_cache_louis();
+
+	/*
+	 * The actual CPU shutdown procedure is at least platform (if not
+	 * CPU) specific.  This may remove power, or it may simply spin.
+	 *
+	 * Platforms are generally expected *NOT* to return from this call,
+	 * although there are some which do because they have no way to
+	 * power down the CPU.  These platforms are the _only_ reason we
+	 * have a return path which uses the fragment of assembly below.
+	 *
+	 * The return path should not be used for platforms which can
+	 * power off the CPU.
 	 */
 	if (smp_ops.cpu_die)
 		smp_ops.cpu_die(cpu);