[PATCH] RM64: dts: ls208xa: Add iommu-map property for pci

[Mark Rutland]

On Thu, Aug 31, 2017 at 11:49:37AM +0100, Marc Zyngier wrote:
> [Fixing Mark's address...]

Thanks!

> On 31/08/17 11:41, Bharat Bhushan wrote:
> > 
> >> -----Original Message-----
> >> From: Marc Zyngier [mailto:marc.zyngier at arm.com]
> >> Sent: Thursday, August 31, 2017 3:02 PM
> >> To: Bharat Bhushan <bharat.bhushan at nxp.com>; robh+dt at kernel.org;
> >> ark.rutland at arm.com; will.deacon at arm.com; oss at buserror.net; Gang Liu
> >> <gang.liu at nxp.com>; devicetree at vger.kernel.org; linux-arm-
> >> kernel at lists.infradead.org; linux-kernel at vger.kernel.org;
> >> catalin.marinas at arm.com
> >> Subject: Re: [PATCH] RM64: dts: ls208xa: Add iommu-map property for pci
> >>
> >> On 31/08/17 10:23, Bharat Bhushan wrote:
> >>> This patch adds iommu-map property for PCIe, which enables SMMU for
> >>> these devices on LS208xA devices.
> >>>
> >>> Signed-off-by: Bharat Bhushan <Bharat.Bhushan at nxp.com>
> >>> ---
> >>>  arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi | 4 ++++
> >>>  1 file changed, 4 insertions(+)
> >>>
> >>> diff --git a/arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi
> >>> b/arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi
> >>> index 94cdd30..67cf605 100644
> >>> --- a/arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi
> >>> +++ b/arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi
> >>> @@ -606,6 +606,7 @@
> >>>  			num-lanes = <4>;
> >>>  			bus-range = <0x0 0xff>;
> >>>  			msi-parent = <&its>;
> >>> +			iommu-map = <0 &smmu 0 1>;	/* This is fixed-up by
> >> u-boot */
> >>
> >> What does this do when your version of u-boot doesn't fill this in for you?
> > 
> > Good question, frankly I have not thought of this case before.
> > But if we pass length = 0 in above property then no fixup happen with
> > happen with older u-boot. In this case of_iommu_configure() will
> > return NULL iommu-ops and it switch to swio-tlb. Will that work?
> I really don't like this. You rely on having invalid data in the DT, and
> that seems just wrong.

Indeed.

Either the property should be valid (and correctly representing the HW),
or it shouldn't be present. Relying on kernel implementation details is
*not* OK.

Thanks,
Mark.