Re: [PATCH 0/4] usb: Use notifier for linking Type C ports.

[Heikki Krogerus]

On Tue, Nov 30, 2021 at 11:27:12AM -0800, Prashant Malani wrote:
> Hi Heikki,
> 
> Thanks for taking a look at the series.
> 
> On Tue, Nov 30, 2021 at 3:03 AM Heikki Krogerus
> <heikki.krogerus@xxxxxxxxxxxxxxx> wrote:
> >
> > Hi Prashant,
> >
> > On Fri, Nov 26, 2021 at 11:40:49AM +0200, Heikki Krogerus wrote:
> > > On Wed, Nov 24, 2021 at 03:10:06PM -0800, Prashant Malani wrote:
> > > > This series resolves the cyclic dependency error which was introduced by
> > > > commit 63cd78617350 ("usb: Link the ports to the connectors they are
> > > > attached to") which lead to it being reverted. The approach here is to
> > > > use a notifier to link a new Type C port to pre-existing USB ports
> > > > instead of calling an iterator of usb ports from the Type C connector
> > > > class. This allows commit 63cd78617350 ("usb: Link the ports to the
> > > > connectors they are attached to") to then be submitted without any
> > > > depmod cyclic dependency error.
> > > >
> > > > The final patch removes the usb port iterator since it is no longer
> > > > needed.
> > >
> > > This is not enough. Build the Type-C Class as a module and the USB bus
> > > statically, and the links will not get created.
> > >
> 
> I see. I suppose it is academic now (given your follow up email about converting
> port-mapper to component framework), but would reversing where the
> notifier block is i.e,
> have usbcore expose the notifier registration API instead of
> typec-class, resolve
> the issue? That would mean the dependency is the same as what it is right now
> in the code, right (typec -> usbcore)

Well, then you would have the same issue if you build the Type-C class
statically and USB as a module, no?

I'm sure that if we though about this hard enough, we would find a way
to make the notifiers work, most likely by handling every possible
scenario separately, but it would still not remove the core problem.
There is the dependency between these components/drivers. The proper
solution does not create that dependency.

Although I'm not sure that the component framework is the best (it is
in the end just a workaround as well, but at least it's there
available for everybody), by taking advantage of the _PLD we can
create a solution where both components can live completely
independently - the order in which they are registered becomes
irrelevant as well as are they build as modules or not.

> > > I'm not sure you actually achieve much with this series, and I'm not
> > > sure this approach will ever fully solve the problem. As long as we
> > > have to declare API, we will have the circular dependency issue on our
> > > hands. But there are ways to avoid that.
> > >
> > > There is for example the component framework (drivers/base/component.c)
> > > that I've been thinking about using here. In this case it would work
> > > so that you declare the USB Type-C part as your aggregate driver, and
> > > everything that is connected to it (so USB ports, DisplayPorts, TBT,
> > > etc.) would then just declare themselves as general components. Could
> > > you take a look at that?
> >
> > I'm preparing a patch where I store all _PLDs in the ACPI tables, and
> > create list of devices that share it. I can convert port-mapper.c to
> > it and the component framework while at it.
> 
> Great, thanks. We can help with testing once you have a patch series
> to share.

OK, cool.

thanks,

-- 
heikki