Re: [RFC PATCH v5 2/4] driver core: Add dev_is_drv_state_synced()

[Abel Vesa]

On 23-06-21 17:22:54, Greg Kroah-Hartman wrote:
> On Wed, Jun 21, 2023 at 05:40:17PM +0300, Abel Vesa wrote:
> > This can be used by drivers to figure out if a different device
> > driver has state synced or not for a specific device.
> > 
> > Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxx>
> > ---
> >  include/linux/device.h | 14 ++++++++++++++
> >  1 file changed, 14 insertions(+)
> > 
> > diff --git a/include/linux/device.h b/include/linux/device.h
> > index bae11928ef7e..8f042f04b5d9 100644
> > --- a/include/linux/device.h
> > +++ b/include/linux/device.h
> > @@ -1007,6 +1007,20 @@ static inline int dev_set_drv_sync_state(struct device *dev,
> >  	return 0;
> >  }
> >  
> > +static inline bool dev_is_drv_state_synced(struct device *dev)
> > +{
> > +	bool ret = false;
> > +
> > +	if (!dev)
> > +		return ret;
> > +
> > +	device_lock(dev);
> > +	ret = dev->state_synced;
> > +	device_unlock(dev);
> 
> This lock is "protecting" nothing, given that the value can instantly
> change after it is read.

Hmm, for some reason I thought it needs to be synchronized with the
sync state callback being called already. But I just noticed that call
to the sync state callback is independently locked after state_synced is
set. So I guess the lock can go away here.

> 
> Because it can change, how will this function actually show anything
> relevant?

The only usecase I can think of for this new API is for some driver
to delay an action until ultimately the driver for a specific device
gets state synced. So even if the value can change after it has
been checked, such consumer driver will most likely retry later on.

Hope that makes sense.

> 
> thanks,
> 
> greg k-h