Re: [PATCH] ARM: tegra: disable nonboot CPUs when reboot

[Will Deacon <will.deacon-5wv7dgnIgG8@public.gmane.org>]

On Fri, Jun 07, 2013 at 11:55:12PM +0100, Russell King - ARM Linux wrote:
> On Fri, Jun 07, 2013 at 04:39:32PM -0600, Stephen Warren wrote:
> > On 06/07/2013 04:15 PM, Russell King - ARM Linux wrote:
> > > For reboot, the real solution there is not to use software-based
> > > reboot, but bring the other cores to a halt (which is what
> > > ipi_send_stop is doing) and then issue a hardware reset to the whole
> > > system, including the other CPUs.
> > 
> > Ignoring the issues with oops in reboot, I think there's a bug in that
> > when hotplug is enabled, smp_kill_cpus() calls platform_cpu_kill(), but
> > nothing causes the failing CPU to ever execute smp_ops.cpu_die(). Hence,
> > if the implementation of smp_ops.cpu_kill() relies on the target CPU
> > having run smp_ops.cpu_die(), then smp_ops.cpu_kill() may not operate
> > correctly.
> 
> Well, smp_kill_cpus() was added to get around the kexec problem -
> transitioning from one kernel to the next kernel without going through
> a hardware reset.  Maybe if we take a step back...
> 
> 1. remove smp_kill_cpus() from smp_send_stop().
> 2. remove machine_shutdown() from machine_halt(), machine_power_off()
>    and machine_restart().
> 3. call smp_send_stop() only from machine_halt(), machine_power_off() and
>    machine_restart()
> 4. require a hardware-based reboot method for all SMP implementations;
>    using soft_reboot() is not an option.
> 
> This should get us into the situation where we have a reliable method of
> halting and rebooting the kernel everywhere, leaving kexec as being the
> remaining problem case.
> 
> Currently, for that we effectively do smp_send_stop() followed by
> smp_kill_cpus().  The no-op change for kexec there is to allow
> smp_kill_cpus() to be called directly from machine_shutdown() - but
> I suspect there will still be stuff that's broken with that...
> 
> So the ongoing problem remains - how to deal with kexec in a SMP
> environment where it's difficult to reliably take a secondary CPU
> offline to a safe place and then be able to restart it into the
> next kernel...

For kexec, I think it's perfectly reasonable to mandate hardware-based
offlining for the secondary cores (hence the half-hearted dependency on
HOTPLUG_CPU). In that case, the only guy that has to go down the soft
reboot path is the primary CPU which shouldn't be too problematic, right?

Supporting sort-reboot of secondaries is a total PITA, even if you have some
`safe place' to put them. You still have to synchronise with non-coherent
cores so that you know when it's safe to clobber the old image, which
requires complex locking algorithms and a prevailing wind.

Will