Re: [PATCH 4/4] PM / devfreq: Handle monitor start/stop in the devfreq core

[Matthias Kaehlcke <mka@chromium.org>]

Hi Lukasz,

On Thu, Feb 14, 2019 at 07:01:36PM +0100, Lukasz Luba wrote:
> Hi Matthias,
> 
> I have compiled and run your changes on Odroid xu3 and v5.0-rc6.
> There are kernel warnings because of mutex not held in function
> devfreq_monitor_[start|stop]() in use cases:
>   1) a few times during registration of new devices devfreq_add_device()
>   2) poking the device from sysfs

Thanks testing!

I messed up carrying over changes from my v4.19 device kernel
to the tree used for upstreaming :(

Do you still see warnings with the below patch?

Thanks

Matthias

diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
index a42c37543c190..866fe711b43ca 100644
--- a/drivers/devfreq/devfreq.c
+++ b/drivers/devfreq/devfreq.c
@@ -440,7 +440,6 @@ static void devfreq_monitor_stop(struct devfreq *devfreq)

        mutex_lock(&devfreq->lock);
        devfreq->monitor_state = DEVFREQ_MONITOR_STOPPED;
-       mutex_unlock(&devfreq->lock);
 }

 /**
@@ -596,7 +595,9 @@ static int governor_start(struct devfreq *devfreq)
                return err;
        }

+       mutex_lock(&devfreq->lock);
        devfreq_monitor_start(devfreq);
+       mutex_unlock(&devfreq->lock);

        return 0;
 }
@@ -614,7 +615,9 @@ static int governor_stop(struct devfreq *devfreq)
                 "mutex must *not* be held by the caller\n"))
                return -EINVAL;

+       mutex_lock(&devfreq->lock);
        devfreq_monitor_stop(devfreq);
+       mutex_unlock(&devfreq->lock);

        err = devfreq->governor->event_handler(devfreq, DEVFREQ_GOV_STOP, NULL);
        if (err) {