Hi Hans, On Wed, Jul 19, 2017 at 01:24:54PM +0200, Hans Verkuil wrote: > On 18/07/17 21:03, Sakari Ailus wrote: > > The async notifier supports three callbacks to the notifier: bound, unbound > > and complete. The complete callback has been traditionally used for > > creating the sub-device nodes. > > > > This approach has an inherent weakness: if registration of a single > > sub-device fails for whatever reason, it renders the entire media device > > unusable even if only that piece of hardware is not working. This is a > > problem in particular in systems with multiple independent image pipelines > > on a single device. We have had such devices (e.g. omap3isp) supported for > > a number of years and the problem is growing more pressing as time passes > > so there is an incentive to resolve this. > > I don't think this is a good reason. If one of the subdevices fail, then your > hardware is messed up and there is no point in continuing. That's entirely untrue in general case. If you have e.g. a mobile phone with a single camera, yes, you're right. But most mobile phones have two cameras these days. Embedded systems may have many, think of automotive use cases: you could have five or ten cameras there. It is not feasible to prevent the entire system from working if a single component is at fault --- this is really any component such as a lens controller. -- Sakari Ailus e-mail: sakari.ailus@xxxxxx XMPP: sailus@xxxxxxxxxxxxxx
