On Thu, Oct 29, 2020 at 04:15:10PM +0800, Chuanjia Liu wrote: > Split the PCIe node and add pciecfg node to fix MSI issue. I still think if you are changing the binding this much, then further work should be done removing the slot nodes. > > Signed-off-by: Chuanjia Liu <chuanjia.liu@xxxxxxxxxxxx> > Acked-by: Ryder Lee <ryder.lee@xxxxxxxxxxxx> > --- > .../bindings/pci/mediatek-pcie-cfg.yaml | 39 ++++++ > .../devicetree/bindings/pci/mediatek-pcie.txt | 129 +++++++++++------- > 2 files changed, 118 insertions(+), 50 deletions(-) > create mode 100644 Documentation/devicetree/bindings/pci/mediatek-pcie-cfg.yaml > > diff --git a/Documentation/devicetree/bindings/pci/mediatek-pcie-cfg.yaml b/Documentation/devicetree/bindings/pci/mediatek-pcie-cfg.yaml > new file mode 100644 > index 000000000000..d3ecbcd032a2 > --- /dev/null > +++ b/Documentation/devicetree/bindings/pci/mediatek-pcie-cfg.yaml > @@ -0,0 +1,39 @@ > +# SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause > +%YAML 1.2 > +--- > +$id: http://devicetree.org/schemas/pci/mediatek-pcie-cfg.yaml# > +$schema: http://devicetree.org/meta-schemas/core.yaml# > + > +title: Mediatek PCIECFG controller > + > +maintainers: > + - Chuanjia Liu <chuanjia.liu@xxxxxxxxxxxx> > + - Jianjun Wang <jianjun.wang@xxxxxxxxxxxx> > + > +description: | > + The MediaTek PCIECFG controller controls some feature about > + LTSSM, ASPM and so on. > + > +properties: > + compatible: > + items: > + - enum: > + - mediatek,generic-pciecfg > + - const: syscon > + > + reg: > + maxItems: 1 > + > +required: > + - compatible > + - reg > + > +additionalProperties: false > + > +examples: > + - | > + pciecfg: pciecfg@1a140000 { > + compatible = "mediatek,generic-pciecfg", "syscon"; > + reg = <0x1a140000 0x1000>; > + }; > +... > diff --git a/Documentation/devicetree/bindings/pci/mediatek-pcie.txt b/Documentation/devicetree/bindings/pci/mediatek-pcie.txt > index 7468d666763a..c14a2745de37 100644 > --- a/Documentation/devicetree/bindings/pci/mediatek-pcie.txt > +++ b/Documentation/devicetree/bindings/pci/mediatek-pcie.txt > @@ -8,7 +8,7 @@ Required properties: > "mediatek,mt7623-pcie" > "mediatek,mt7629-pcie" > - device_type: Must be "pci" > -- reg: Base addresses and lengths of the PCIe subsys and root ports. > +- reg: Base addresses and lengths of the root ports. > - reg-names: Names of the above areas to use during resource lookup. > - #address-cells: Address representation for root ports (must be 3) > - #size-cells: Size representation for root ports (must be 2) > @@ -143,56 +143,71 @@ Examples for MT7623: > > Examples for MT2712: > > - pcie: pcie@11700000 { > + pcie1: pcie@112ff000 { > compatible = "mediatek,mt2712-pcie"; > device_type = "pci"; > - reg = <0 0x11700000 0 0x1000>, > - <0 0x112ff000 0 0x1000>; > - reg-names = "port0", "port1"; > + reg = <0 0x112ff000 0 0x1000>; > + reg-names = "port1"; > #address-cells = <3>; > #size-cells = <2>; > - interrupts = <GIC_SPI 115 IRQ_TYPE_LEVEL_HIGH>, > - <GIC_SPI 117 IRQ_TYPE_LEVEL_HIGH>; > - clocks = <&topckgen CLK_TOP_PE2_MAC_P0_SEL>, > - <&topckgen CLK_TOP_PE2_MAC_P1_SEL>, > - <&pericfg CLK_PERI_PCIE0>, > + interrupts = <GIC_SPI 117 IRQ_TYPE_LEVEL_HIGH>; > + interrupt-names = "pcie_irq"; > + clocks = <&topckgen CLK_TOP_PE2_MAC_P1_SEL>, > <&pericfg CLK_PERI_PCIE1>; > - clock-names = "sys_ck0", "sys_ck1", "ahb_ck0", "ahb_ck1"; > - phys = <&pcie0_phy PHY_TYPE_PCIE>, <&pcie1_phy PHY_TYPE_PCIE>; > - phy-names = "pcie-phy0", "pcie-phy1"; > + clock-names = "sys_ck1", "ahb_ck1"; > + phys = <&u3port1 PHY_TYPE_PCIE>; > + phy-names = "pcie-phy1"; > bus-range = <0x00 0xff>; > - ranges = <0x82000000 0 0x20000000 0x0 0x20000000 0 0x10000000>; > + ranges = <0x82000000 0 0x11400000 0x0 0x11400000 0 0x300000>; > > - pcie0: pcie@0,0 { > - reg = <0x0000 0 0 0 0>; > + slot1: pcie@1,0 { Does the driver still work if this is devfn 0 instead of 1 (or swap 0 and 1 slots)? I'll bet it does. The reason being is that AFAICT, the Mediatek PCI controller is Designware based. The registers at 0x70c and 0x73c are DWC 'port logic' registers. The DWC RC also has a quirk that it doesn't filter config accesses to only devfn 0 (see pci_dw_valid_device()), so your config accesses to the RP should work no matter what devfn you use. You'll have to get rid of the ports list and just get the mtk_pcie_port from bus->sysdata instead. Rob
