Re: [PATCH] of/irq: Add msi-parent check to of_msi_xlate()

[Lorenzo Pieralisi <lpieralisi@kernel.org>]

On Thu, Sep 18, 2025 at 02:44:23PM -0500, Rob Herring wrote:
> On Thu, Sep 18, 2025 at 10:21 AM Lorenzo Pieralisi
> <lpieralisi@kernel.org> wrote:
> >
> > On Thu, Sep 18, 2025 at 08:55:55AM -0500, Rob Herring wrote:
> > > On Tue, Sep 16, 2025 at 11:18:58AM +0200, Lorenzo Pieralisi wrote:
> > > > In some legacy platforms the MSI controller for a PCI host
> > > > bridge is identified by an msi-parent property whose phandle
> > > > points at an MSI controller node with no #msi-cells property,
> > > > that implicitly means #msi-cells == 0.
> > > >
> > > > For such platforms, mapping a device ID and retrieving the
> > > > MSI controller node becomes simply a matter of checking
> > > > whether in the device hierarchy there is an msi-parent property
> > > > pointing at an MSI controller node with such characteristics.
> > > >
> > > > Add a helper function to of_msi_xlate() to check the msi-parent
> > > > property in addition to msi-map and retrieve the MSI controller
> > > > node (with a 1:1 ID deviceID-IN<->deviceID-OUT mapping) to
> > > > provide support for deviceID mapping and MSI controller node
> > > > retrieval for such platforms.
> > >
> > > Your line wrapping is a bit short.
> > >
> > > I had a look at who is parsing "msi-parent" themselves as that's
> > > typically a recipe for doing it incorrectly ('interrupt-map' anyone).
> > > Can we make iproc_pcie_msi_enable() use this? It's quite ugly reaching
> > > into the GICv3 node...
> >
> > I am not sure I get what you mean here. Possibly iproc_pcie_msi_enable()
> > can reuse this patch's code if I extend it and make it a global function,
> > yes and somehow use that function to carry out the check for an
> > msi-parent property with no #msi-cells property or an #msi-cells == 0.
> 
> I meant using of_msi_xlate() (or even of_msi_get_domain()).

https://lore.kernel.org/all/20240818161816.GA173148-robh@kernel.org/

Isn't of_msi_get_domain() leaking a reference when it finds an IRQ domain and
return ?

Possibly I can generalize it and use it in this patch too.

Thanks,
Lorenzo

> > Don't get what GICv3 node has to do with that though, sorry.
> 
> Just trace what the code there does after it gets the MSI parent. I
> didn't study it too closely, but why is a iProc PCIe parsing GICv3 MSI
> stuff itself? There's either some missing feature in the
> irqchip/domain APIs or it's being dumb.
> 
> Rob