Re: [PATCH v3 2/5] MFD: intel_pmt: Support non-PMT capabilities

[Greg KH <gregkh@linuxfoundation.org>]

On Mon, Sep 27, 2021 at 11:40:37AM -0700, David E. Box wrote:
> On Mon, 2021-09-27 at 19:36 +0200, Greg KH wrote:
> > On Wed, Sep 22, 2021 at 02:30:04PM -0700, David E. Box wrote:
> > > Intel Platform Monitoring Technology (PMT) support is indicated by presence
> > > of an Intel defined PCIe DVSEC structure with a PMT ID. However DVSEC
> > > structures may also be used by Intel to indicate support for other
> > > capabilities unrelated to PMT.  OOBMSM is a device that can have both PMT
> > > and non-PMT capabilities. In order to support these capabilities it is
> > > necessary to modify the intel_pmt driver to handle the creation of platform
> > > devices more generically.
> > 
> > I said this on your other driver submission, but why are you turning a
> > PCIe device into a set of platform devices and craming it into the MFD
> > subsystem?
> > 
> > PCIe devices are NOT platform devices.
> 
> But they *are* used to create platform devices when the PCIe device is multi-functional, which is
> what intel_pmt is.

That is an abuse of platform devices, as that is not what they are for.

> > Why not use the auxiliary bus for this thing if you have individual
> > drivers that need to "bind" to the different attributes that this single
> > PCIe device is exporting.
> 
> It wasn't clear in the beginning how this would evolve. MFD made sense for the PMT (platform
> monitoring technology) driver. PMT has 3 related but individually enumerable devices on the same IP,
> like lpss. But the same IP is now being used for other features too like SDSi. We could work on
> converting this to the auxiliary bus and then covert the cell drivers.

Please do so.

> > Or why not just fix the hardware to report individual PCIe devices, like
> > a sane system would do?
> 
> We have some systems with 1000+ PCIe devices. Each PCIe device adds cost to HW. So increasingly
> VSEC/DVSEC is used to expose features which are handled by the same micro-controller in the HW.

A PCIe device is a virtual thing, what HW cost do they have?

Anyway, a platform device should NOT ever be a child of a PCI device,
that is not ok and should be fixed here please.

A platform device is just that, something that the platform provides on
a non-discoverable bus.  A PCIe device is NOT that type of device at
all and never has been.

thanks,

greg k-h