arm64: Fix deadlock scenario with smp_send_stop()
authorArun KS <arunks.linux@gmail.com>
Wed, 7 May 2014 01:41:23 +0000 (02:41 +0100)
committerHuang, Tao <huangtao@rock-chips.com>
Fri, 27 Mar 2015 06:28:08 +0000 (14:28 +0800)
If one process calls sys_reboot and that process then stops other
CPUs while those CPUs are within a spin_lock() region we can
potentially encounter a deadlock scenario like below.

CPU 0                   CPU 1
-----                   -----
                        spin_lock(my_lock)
smp_send_stop()
 <send IPI>             handle_IPI()
                         disable_preemption/irqs
                          while(1);
 <PREEMPT>
spin_lock(my_lock) <--- Waits forever

We shouldn't attempt to run any other tasks after we send a stop
IPI to a CPU so disable preemption so that this task runs to
completion. We use local_irq_disable() here for cross-arch
consistency with x86.

Based-on-work-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Arun KS <getarunks@gmail.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit b9acc49ee9464f8f8232a790045d057eb158e869)

Conflicts:
arch/arm64/kernel/process.c

arch/arm64/kernel/process.c

index 94e13bc7dd9f1aad28432d5840c8f9f4d690aa45..11094c8d1227a7b5867541e1f29b7af7eae515a1 100644 (file)
@@ -123,6 +123,7 @@ void machine_shutdown(void)
  */
 void machine_halt(void)
 {
+       local_irq_disable();
        smp_send_stop();
        while (1);
 }
@@ -135,6 +136,7 @@ void machine_halt(void)
  */
 void machine_power_off(void)
 {
+       local_irq_disable();
        smp_send_stop();
        if (pm_power_off)
                pm_power_off();
@@ -153,11 +155,10 @@ void machine_power_off(void)
  */
 void machine_restart(char *cmd)
 {
-       smp_send_stop();
-
        /* Disable interrupts first */
        local_irq_disable();
        local_fiq_disable();
+       smp_send_stop();
 
        /* Now call the architecture specific reboot code. */
        if (arm_pm_restart)