Shutdown problem in SMP system happened on Tegra20

[Shawn Guo]

On Sat, Aug 25, 2012 at 08:52:21AM +0100, Russell King - ARM Linux wrote:
> On Sat, Aug 25, 2012 at 03:36:27PM +0800, Shawn Guo wrote:
> > On Fri, Aug 24, 2012 at 07:21:33PM +0100, Russell King - ARM Linux wrote:
> > > Why not just ensure that CONFIG_PM_SLEEP_SMP is enabled if your platform
> > > requires that the lowest CPU number be the CPU dealing with reboot?
> > 
> > I have CONFIG_PM_SLEEP_SMP enabled for imx6q, but still see the imx6q
> > restart hook running on cpu1 than cpu0.  It seems that
> > disable_nonboot_cpus is only called in kernel_power_off but never
> > kernel_restart.  We should probably patch kernel_restart or
> > machine_restart to have disable_nonboot_cpus called? 
> 
> Remember that this path gets called from IRQ context which makes calling
> functions which sleep very dodgy.

I'm so familiar with the code/context.  Are you saying disable_nonboot_cpus
sleeps and kernel_power_off never gets called from IRQ context?

I tested the code change below on kernel_restart with
CONFIG_DEBUG_ATOMIC_SLEEP enabled.  It fixes imx6q restart issue (with
no change on cpu_relax) while no sleep-inside-atomic-section warning
is seen. 

Regards,
Shawn

diff --git a/kernel/sys.c b/kernel/sys.c
index 241507f..0359328 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -373,6 +373,7 @@ void kernel_restart(char *cmd)
        else
                printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
        kmsg_dump(KMSG_DUMP_RESTART);
+       disable_nonboot_cpus();
        machine_restart(cmd);
 }
 EXPORT_SYMBOL_GPL(kernel_restart);