The patch can ensure that v7_exit_coherency_flush() in rockchip_cpu_die() executed in time. The mdelay(1) has enough time to fix the problem of CPU offlining. That's a workaround way in rockchip hotplug code, At least, we haven't a better way to solve it. Who know, that maybe fixed by chip (hardware) in the future. Signed-off-by: Caesar Wang <wxt at rock-chips.com> --- Changes in v6: - Fix the delay 1ms describing. Series-changes: 5 - Fix the patch decription. - Add the changelog. Series-changes: 2 - As Kever points out, Fix the subject typo WFI/WFE. arch/arm/mach-rockchip/platsmp.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/arch/arm/mach-rockchip/platsmp.c b/arch/arm/mach-rockchip/platsmp.c index b379cc8..d629206 100644 --- a/arch/arm/mach-rockchip/platsmp.c +++ b/arch/arm/mach-rockchip/platsmp.c @@ -322,6 +322,13 @@ static void __init rockchip_smp_prepare_cpus(unsigned int max_cpus) #ifdef CONFIG_HOTPLUG_CPU static int rockchip_cpu_kill(unsigned int cpu) { + /* + * We need a delay here to ensure that the dying CPU can finish + * executing v7_coherency_exit() and reach the WFI/WFE state + * prior to having the power domain disabled. + */ + mdelay(1); + pmu_set_power_domain(0 + cpu, false); return 1; } -- 1.9.1