Re: [PATCH 1/3] driver core: enable drivers to use deferred probe from init

["Luis R. Rodriguez" <mcgrof@suse.com>]

On Mon, Jul 28, 2014 at 06:52:48PM +0200, Luis R. Rodriguez wrote:
> On Mon, Jul 28, 2014 at 03:46:32PM +0000, Yuval Mintz wrote:
> > > Subject: Re: [PATCH 1/3] driver core: enable drivers to use deferred probe from
> > > init
> > > 
> > > On Mon, Jul 28, 2014 at 03:12:11PM +0000, Yuval Mintz wrote:
> > > > Perhaps this is a silly question, but what guarantees that the
> > > > deferred probe list will actually be triggered, e.g., in case the
> > > > delayed device is the last device in the system?
> > > 
> > > The dev->init_delayed_probe is used to ensure that we'd add the device to the
> > > deferred probe list once making this a per device thing if the driver has the field
> > > delay_probe set to true. This technically also allows this to be a per device thing
> > > so with some more work we could enable drivers to only enable this for specific
> > > devices but at this point this did not seem required.
> > 
> > Sorry for not being clear, but I didn't meant 'what guarantees that the device
> > will be added to the deferred probe', but rather what guarantees that the
> > deferred workqueue will be scheduled.
> > 
> > To the best of my knowledge the deferring mechanism works only if one device
> > is dependent upon another, e.g., for Multi-function devices where one device
> > probe is dependent upon the others - which are soon-to-be probed.
> 
> The workqueue will be kicked when driver_deferred_probe_trigger() gets
> poked, we do that in the late_initcall(deferred_probe_initcall), it
> also gets flushed there with a flush_workqueue(deferred_wq).

But come to think of it that will work well for devices already plugged in
so indeed I think that driver_deferred_probe_add() needs a check added
for for if (!driver_deferred_probe_enable) then we have to
driver_deferred_probe_trigger(). The driver_deferred_probe_enable is set
to false upon init, but later on late init it gets set to true so with
that check we'd  only generate the trigger after late init call.

I can fold that in the v2.

  Luis