Re: [PATCH v2] timekeeping: Allow runtime PM from change_clocksource()

[Wolfram Sang]

On Thu, Feb 11, 2021 at 02:43:18PM +0100, Niklas Söderlund wrote:
> The struct clocksource callbacks enable() and disable() are described as
> a way to allow clock sources to enter a power save mode [1]. But using
> runtime PM from these callbacks triggers a cyclic lockdep warning when
> switching clock source using change_clocksource().
> 
> This change allows the new clocksource to be enabled() and the old one
> to be disabled() without holding the timekeeper_lock. This solution is
> modeled on timekeeping_resume() and timekeeping_suspend() where the
> struct clocksource resume() and suspend() callbacks are called without
> holding the timekeeper_lock.
> 
> Triggering and log of the deadlock warning,
> 
>   # echo e60f0000.timer > /sys/devices/system/clocksource/clocksource0/current_clocksource

...

I had the same issue when running 'clocksource-switch' from the Kernel
selftests with a Renesas Falcon board using a R-Car V3U. This patch
fixed the deadlock warning, so:

Tested-by: Wolfram Sang <wsa+renesas@xxxxxxxxxxxxxxxxxxxx>

Attachment: signature.asc
Description: PGP signature