Gregory,

On Fri, 03 Jul 2015 13:39:49 +0200, Gregory CLEMENT wrote:

Having 2 initcall does not work because, there is a dependency between these
2 calls. And actually the suspend_ops is registered before the board specific
hook. As soon as the suspend_ops is registered, mvebu_pm_valid() is called but
at this point mvebu_board_pm_enter is NULL so PM_SUSPEND_MEM is not available.

And? It will become available soon afterwards.

All the complexity of the original patch was to allow registering a handler
without needed to get the resource(gpio device) that are not available when using
arch_initcall(). However the device_initcall_sync comes latter enough to
get all the devices registered but it still happens before the late_initcall,
so I will use this one and I will add a comment around it.

I don't think we care about the order in which the initcalls are called.

If the SoC level init call registering the suspend_ops gets called
first, then at the beginning there is only support for standby. The
support for suspend to RAM will be enabled once the board-level init
call gets called.

If the board level init call is called first, then it will set
mvebu_board_pm_enter. It is not useful at this point. Until the SoC
level init call registers the suspend_ops.

I believe that the ->valid() method of suspend_ops gets called when the
user actually enters the suspend state by writing to /sys/power/state.
And by that time, both init calls will have been called.

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com