Re: [PATCH v1 2/4] platform/x86: mlx-platform: Add differed bus functionality

[Darren Hart]

On Thu, Mar 22, 2018 at 10:04:51AM +0100, Greg Kroah-Hartman wrote:
> On Wed, Mar 21, 2018 at 03:58:18PM -0700, Darren Hart wrote:
> > On Tue, Feb 13, 2018 at 10:09:34PM +0000, Vadim Pasternak wrote:
> > > Add deferred bus functionality in order to enforce probing flow execution
> > > order. Driver mlx-platform activates platform driver i2c-mux-reg, which
> > > creates busses infrastructure, after that it activates mlxreg-hotplug
> > > driver, which uses these busses, for connecting devices. The possible
> > > miss-ordering can happened, for example in case the probing routine of
> > > mlxreg-hotplug is already started execution, while i2c-mux-reg probing
> > > routine is not completed yet. In such situation the first one could
> > > attempt to connect device to adapter number, which is not created yet.
> > > And as a result this connection will fail. In order to ensure the order of
> > > probing execution on mlxreg-hotplug probe routine will be deferred until
> > > all the busses is not created by  probe routine of i2c-mux-reg.
> > > In order to ensure the flow order, mlx-platform driver passes the highest
> > > bus number to the mlxreg-hotplug platform data, which in their turn could
> > > wait in the deferred state, until all the necessary buses topology is not
> > > exist.
> > 
> > Vadim,
> > 
> > I've got part way through a review of this series several times, and keep
> > running out of time trying to determine if the solution is appropriate
> > for the problem.
> > 
> > The deferred probing waiting for the i2c bus to be available seems like
> > something we might be able to handle more elegantly with:
> > 
> > 1) request_module() in the platform driver
> > 2) careful use of init levels (subsys, device, late)
> > 
> > This isn't something I'm well versed in - so I'll reach out here for
> > some advice from Greg and the i2c maintainer, Wolfram, and list (added
> > to Cc).
> 
> deferred device probe?
> 

This patch series adds the following, which I believe is what you're referring
to Greg? :

+       /* Defer probing if the necessary adapter is not configured yet. */
+       deferred_adap = i2c_get_adapter(pdata->deferred_nr);
+       if (!deferred_adap)
+               return -EPROBE_DEFER;

This is the preferred approach? Have the platform driver define the expected
adapter number, and then defer until that becomes available?

-- 
Darren Hart
VMware Open Source Technology Center