[PATCH v2 1/2] PCI: keystone: Set mode as RootComplex for "ti,keystone-pcie" compatible

[Bjorn Helgaas]

On Wed, Nov 06, 2024 at 11:36:38AM +0530, Siddharth Vadapalli wrote:
> On Tue, Nov 05, 2024 at 06:57:58PM -0600, Bjorn Helgaas wrote:
> > On Fri, May 24, 2024 at 04:27:13PM +0530, Siddharth Vadapalli wrote:
> > > From: Kishon Vijay Abraham I <kishon at ti.com>
> > > 
> > > commit 23284ad677a9 ("PCI: keystone: Add support for PCIe EP in AM654x
> > > Platforms") introduced configuring "enum dw_pcie_device_mode" as part of
> > > device data ("struct ks_pcie_of_data"). However it failed to set mode
> > > for "ti,keystone-pcie" compatible. Set mode as RootComplex for
> > > "ti,keystone-pcie" compatible here.
> > 
> > 23284ad677a9 appeared in v5.10.  
> > 
> > But I guess RC support has not been broken since v5.10 because we
> > never used ks_pcie_rc_of_data.mode anyway?
> > 
> > It looks like the only use is here:
> > 
> >   #define DW_PCIE_VER_365A                0x3336352a
> >   #define DW_PCIE_VER_480A                0x3438302a
> > 
> >   ks_pcie_probe
> >   {
> >     ...
> >     mode = data->mode;
> >     ...
> >     if (dw_pcie_ver_is_ge(pci, 480A))
> >       ret = ks_pcie_am654_set_mode(dev, mode);
> >     else
> >       ret = ks_pcie_set_mode(dev);
> 
> "mode" is used later on during probe at:
> 
> ....
> 	switch (mode) {
> 	case DW_PCIE_RC_TYPE:
> 	...
> 	case DW_PCIE_EP_TYPE:
> 	...
> 	default:
> 		dev_err(dev, "INVALID device type %d\n", mode);
> 	}
> ....

How did I miss that? :)  It is literally two lines down.

> > so we don't even look at .mode unless the version is v4.80a or later,
> > and this is v3.65a?
> > 
> > So this is basically a cosmetic fix (but still worth doing for
> > readability!) and doesn't need a stable backport, right?
> 
> I suppose that "data->mode" will default to zero for v3.65a prior to
> this commit, corresponding to "DW_PCIE_UNKNOWN_TYPE" rather than the
> correct value of "DW_PCIE_RC_TYPE". Since I don't have an SoC with the
> v3.65a version of the controller, I cannot test it out, but I presume
> that the "INVALID device type 0" error will be displayed. Though the probe
> will not fail since the "default" case doesn't return an error code, the
> controller probably will not be functional as the configuration associated
> with the "DW_PCIE_RC_TYPE" case has been skipped. Hence, I believe that
> this fix should be backported.

I guess nobody really cares too much since it's been broken for almost
four years.

But indeed, sounds like it should have a stable tag and maybe a commit
log hint about what the failure looks like.

Thanks!

Bjorn