ARM: vexpress: fix CPU idle C1 state enablementtracking-tracking-armlt-tc2-pm-ll-20120727.0

This patch defines C1 as disabled by default on TC2. If the state is
later enabled but deep shutdown states are required to be enabled on
a cluster basis, a coupled barrier must be added to the coupled enter
function so that coupled CPUs cannot exit the polling C1 state while
other ones are still entering.

This patch fixes a race in C1 code where one CPU exits a coupled
C-state while another one, part of the coupled mask is still
selecting the coupled C-state value. This can cause some CPUs to
select the -1 state index which triggers an undefined instruction fault.

The added coupled C-state barrier fixes the issue by forcing coupled
C-states to be entered and exited at the same time by all CPUs even
if cluster shutdown is disabled.

Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>

1 files changed, 5 insertions, 1 deletions

diff --git a/arch/arm/mach-vexpress/cpuidle-tc2.c b/arch/arm/mach-vexpress/cpuidle-tc2.c
index 11030b5c0ec4..cb9c64695b3c 100644
--- a/arch/arm/mach-vexpress/cpuidle-tc2.c
+++ b/arch/arm/mach-vexpress/cpuidle-tc2.c
@@ -72,6 +72,7 @@ static struct cpuidle_state tc2_cpuidle_set[] __initdata = {
 							CPUIDLE_FLAG_COUPLED,
 		.name			= "C1",
 		.desc			= "ARM power down",
+		.disable		= 1,
 	},
 };
 
@@ -138,8 +139,11 @@ static int tc2_enter_coupled(struct cpuidle_device *dev,
 	/* Used to keep track of the total time in idle */
 	getnstimeofday(&ts_preidle);
 
-	if (!cpu_isset(cluster, cluster_mask))
+	if (!cpu_isset(cluster, cluster_mask)) {
+			cpuidle_coupled_parallel_barrier(dev,
+					&abort_barrier[cluster]);
 			goto shallow_out;
+	}
 
 	BUG_ON(!irqs_disabled());