Re: [PATCH 5/5] arm64: dts: qcom: qcs8550: add QCS8550 RB5Gen2 board support

[Manivannan Sadhasivam]

On Tue, Apr 07, 2026 at 12:39:25PM +0100, Joe Sandom wrote:

[...]

> > > +&pcie0 {
> > > +	wake-gpios = <&tlmm 96 GPIO_ACTIVE_HIGH>;
> > > +	perst-gpios = <&tlmm 94 GPIO_ACTIVE_LOW>;
> > > +
> > > +	pinctrl-0 = <&pcie0_default_state>;
> > > +	pinctrl-names = "default";
> > > +
> > > +	iommu-map = <0x0 &apps_smmu 0x1400 0x1>,
> > > +		    <0x100 &apps_smmu 0x1401 0x1>,
> > > +		    <0x208 &apps_smmu 0x1402 0x1>,
> > > +		    <0x210 &apps_smmu 0x1403 0x1>,
> > > +		    <0x218 &apps_smmu 0x1404 0x1>,
> > > +		    <0x300 &apps_smmu 0x1407 0x1>,
> > > +		    <0x400 &apps_smmu 0x1408 0x1>,
> > > +		    <0x500 &apps_smmu 0x140c 0x1>,
> > > +		    <0x501 &apps_smmu 0x140e 0x1>;
> > > +
> > > +	/delete-property/ msi-map;
> > 
> > Why?
> I tried extending the msi-map to cover the RIDs from the QPS615
> PCIe switch (matching the iommu-map entries), but this caused
> ITS MAPD command timeouts.

I'm not aware of any specific issue with ITS on this chipset. At what time did
you see the timeout? During probe?

> From what I could gather, deleting
> msi-map forces the PCIe controller to fall back to the internal
> iMSI-RX module, where this worked properly.
> 

That's true.

> For reference, I checked the RB3gen2 since it also uses a QPS615
> and there doesn't seem to be any msi-map defined (in kodiak.dtsi).
> 

But Kodiak has no MSI support (no LPIs). That's why the ITS node is disabled and
iommu-map is used.

> Any recommendations to resolve this properly?

I will also check internally in the meantime.

- Mani

-- 
மணிவண்ணன் சதாசிவம்