[PATCH 4/5] soundwire: sysfs: remove sdw_slave_sysfs_init()

Vinod Koul vkoul at kernel.org
Wed Aug 3 13:33:41 CEST 2022


On 29-07-22, 17:13, Greg Kroah-Hartman wrote:
> On Fri, Jul 29, 2022 at 10:00:42AM -0500, Pierre-Louis Bossart wrote:
> > 
> > 
> > > diff --git a/drivers/soundwire/sysfs_slave_dpn.c b/drivers/soundwire/sysfs_slave_dpn.c
> > > index c4b6543c09fd..a3fb380ee519 100644
> > > --- a/drivers/soundwire/sysfs_slave_dpn.c
> > > +++ b/drivers/soundwire/sysfs_slave_dpn.c
> > > @@ -283,6 +283,9 @@ int sdw_slave_sysfs_dpn_init(struct sdw_slave *slave)
> > >  	int ret;
> > >  	int i;
> > >  
> > > +	if (!slave->prop.source_ports && !slave->prop.sink_ports)
> > > +		return 0;
> > > +
> > >  	mask = slave->prop.source_ports;
> > >  	for_each_set_bit(i, &mask, 32) {
> > >  		ret = add_all_attributes(&slave->dev, i, 1);
> > 
> > I am struggling with this one since the driver is still adding
> > attributes manually. You mentioned in the other thread that
> > 
> > "
> > That's what the is_visible() callback is for in the groups structure,
> > you determine if the attribute is visable or not at runtime, you don't
> > rely on the driver itself to add/remove attributes, that does not scale
> > and again, is racy.
> > "
> > 
> > I interpret that as "there's still a race here", no?
> 
> Yes, there is, BUT as you are creating all of these attributes "on the
> fly" for now, I don't see a simple conversion to fix that up.  Let me do
> these, the easy ones first.  Your dynamic attribute allocations are the
> harder things to do, let me think about those after I've fixed the rest
> of the tree up with the trivial ones :)

Sounds good to me.. Yes the dynamic ones are the one that need
attention. How do you propose to handle these?

-- 
~Vinod


More information about the Alsa-devel mailing list