On Mon, Apr 29, 2024 at 10:48:23AM -0500, Rob Herring wrote: > On Tue, Apr 02, 2024 at 10:33:46AM -0400, Frank Li wrote: > > From: Richard Zhu <hongxing.zhu@xxxxxxx> > > > > Add i.MX8Q PCIe "fsl,imx8q-pcie" compatible strings. > > > > Add "fsl,local-address" property for i.MX8Q platforms. fsl,local-address > > is address of PCIe module in high speed io (HSIO)subsystem bus fabric. HSIO > > bus fabric convert the incoming address base to this local-address. Two > > instances of PCI have difference local address. > > This is just some intermediate bus address? We really should be able to > describe this with standard ranges properties. Yes, Maybe dwc's implement have some problem. After read below doc again https://elinux.org/Device_Tree_Usage#PCI_Address_Translation ┌──────┐ ┌──────────┐ ┌────┐0x18001000 │ │ │ │ │CPU ├───────────►│ ├──┤ Others │ └────┘ │ │ │ │ │ │ └──────────┘ │ │ │ │ ┌─────────┐ │ │ │ │ ┌───────────┐ │ ├──►│ HSIO │ 0xB8001000 ├───────────┤ │ │ │ Fabric ├───────────►│Bar0 │ TLP mem 0xB8001000 │ │ │ │ │0xB8000000 ├───────► └──────┘ └─────────┘ │ │ Main Fabric ├───────────┤ │ │ │ │ │ │ │ │ │ │ │ │ │ DWC │ │ PCIe │ │ Controller│ │ │ │ │ └───────────┘ dts should be ranges = <0x82000000 0 0xB8000000 0x18000000 0 0x07f00000> ^^^^ u64 imx_pcie_cpu_addr_fixup(struct dw_pcie *pcie, u64 cpu_addr) offset = entry->res->start; { ... return (cpu_addr - entry->offset); } NVME can work. let me do more test. Frank > > > > > Signed-off-by: Richard Zhu <hongxing.zhu@xxxxxxx> > > Signed-off-by: Frank Li <Frank.Li@xxxxxxx> > > --- > > .../devicetree/bindings/pci/fsl,imx6q-pcie-common.yaml | 5 +++++ > > .../devicetree/bindings/pci/fsl,imx6q-pcie.yaml | 18 ++++++++++++++++++ > > 2 files changed, 23 insertions(+)