It's buggy to disable the cpu that is being hot-unplugged in .cpu_die
hook which runs on the cpu itself. Instead, it should be done in
.cpu_kill which runs on the thread (another cpu) that asks for shutting
down the cpu. Move imx_enable_cpu(cpu, false) call into .cpu_kill
hook, and leave the cpu to be hot-unplugged in WFI within .cpu_die,
so that we can get a more stable cpu hot-plug operation.
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Rename mach-imx/include/mach/common.h to mach-imx/common.h and update
all users to include common.h rather than mach/common.h.
It also removes an unneeded inclusion to common.h in
mach-imx/devices/devices.c.
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Acked-by: Sascha Hauer <s.hauer@pengutronix.de>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Convert the imx6q platform to use struct smp_operations to provide
its SMP and CPU hotplug operations.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Nicolas Pitre <nico@linaro.org>
Cc: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Though commit 602bf40 (ARM: imx6: exit coherency when shutting down
a cpu) improves the stability of imx6q cpu hotplug a lot, there are
still hangs seen with a more stressful hotplug testing.
It's expected that once imx_enable_cpu(cpu, false) is called, the cpu
will be taken down by hardware immediately, and the code after that
will not get any chance to execute. However, this is not always the
case from the testing. The cpu could possibly be alive for a few
cycles before hardware actually takes it down. So rather than letting
cpu execute some code that could cause a hang in these cycles, let's
make the cpu spin there and wait for hardware to take it down.
Cc: <stable@vger.kernel.org>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
There is a system hang issue on imx6q which can easily be seen with
running a cpu hotplug stress testing (hotplug secondary cores from
user space via sysfs interface for thousands iterations).
It turns out that the issue is caused by coherency of the cpu that
is being shut down. When shutting down a cpu, we need to have the
cpu exit coherency to prevent it from receiving cache, TLB, or BTB
maintenance operations broadcast by other CPUs in the cluster.
Copy cpu_enter_lowpower() and cpu_leave_lowpower() from mach-vexpress
to have coherency properly handled in platform_cpu_die(), thus fix
the issue.
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Cc: stable@kernel.org
Shawn Guo