On Tue, Dec 03, 2024 at 12:29:16AM +0100, Heiko Stuebner wrote: > > On Wed, 20 Nov 2024 18:10:49 +0100, Niklas Cassel wrote: > > Add msi-parent for the pcie3x4_ep PCI endpoint node. > > > > The pcie3x4_ep node should use the same msi-parent as the pcie3x4 node > > (which represents the PCIe controller running in Root Complex mode). > > > > The GIC ITS can be used to trigger an IRQ on the endpoint when any of > > the endpoint's PCI BARs are written to by the host[1]. > > > > [...] > > Applied, thanks! > > [1/1] arm64: dts: rockchip: rk3588: add msi-parent for pcie3x4_ep > commit: b6f09f497b07008aa65c31341138cecafa78222c > Hello Heiko, When I sent this patch, I had tested it against Frank's v8 series which adds support for a PCIe endpoint triggering an interrupt (using GIC ITS) when the host side writes to the PCI BAR that the endpoint has configured as doorbell: https://lore.kernel.org/linux-pci/20241116-ep-msi-v8-0-6f1f68ffd1bb@xxxxxxx/ However, it seems that in v10 of his series, he has changed it so that the PCIe endpoint node now requires 'msi-map' instead of 'msi-parent' (just like how it is done in the PCIe root complex node): See: https://lore.kernel.org/linux-pci/20241204-ep-msi-v10-0-87c378dbcd6d@xxxxxxx/ """ Changes in v10: [...] - Use "msi-map" in pci ep controler node, instead of of msi-parent. first argument is (func_no << 8 | vfunc_no) """ I didn't realize that DT property required for this feature could change so drastically, since DT is supposed to describe hardware, and msi-parent made perfect sense, since this property is usually the one used for devices connected to a bus. (msi-map is usually only used on the host bus adapter / root complex.) Knowing what I now know, I should have waited until this feature had landed before submitting this patch. I apologize for this. Could you please drop this patch from your v6.14-armsoc/dts64 branch? Or should I send a revert? Once Frank's series has landed, I can resubmit a patch that adds whichever DT property he finally ends up using. Kind regards, Niklas
