On Wed, Aug 09, 2023 at 05:45:49PM +0300, Vladimir Oltean wrote:
> Hi Bjorn,
>
> On Tue, Aug 08, 2023 at 05:21:07PM -0500, Bjorn Helgaas wrote:
> > I think this makes good sense, but let me make sure I understand how
> > this works.
> >
> > I *think* what's happening is that this Function 0 responds to config
> > reads, so PCI enumeration starts by discovering it normally. But
> > after 6fffbc7ae137 ("PCI: Honor firmware's device disabled status"),
> > we abort in pci_setup_device() if the DT or ACPI status is "disabled,"
> > which means there's no struct pci_dev for it, no quirks can run on it,
> > and no driver can bind to it. And, since PCI multi-function devices
> > must have a Function 0, we don't enumerate the other functions of this
> > device.
> >
> > That's a problem because (1) you need to do some initialization on
> > Function 0 even though you don't want a driver to claim it, and
> > (2) this is a multi-function device and you need to enumerate the
> > other functions.
>
> Correction: on functions 0, 1, 2 and 6 (all have
> PCI_VENDOR_ID_FREESCALE, ENETC_DEV_ID_PF), and not just on function
> 0. The particular nature of a hardware IP bug/afterthought makes
> this necessary.
Thanks, so (2) is only relevant to function 0, but (1) applies to
other functions as well.
> You may have given the netdev maintainers some mixed signals with the
> rewording suggestion plus the ack for my wording, and now we have commit
> 1a8c251cff20 ("PCI: move OF status = "disabled" detection to
> dev->match_driver") in the net.git tree.
>
> I think we are mostly on the same page about what is changing, it's just
> that we are focusing on different aspects of it in the description.
>
> I hope you're ok if we close the topic the way things are now? :)
Yep, my fault for forgetting how netdev works. Thanks for your
patience.
Bjorn
