Re: [PATCH v2] PCI: dw-rockchip: Enable async probe by default

[Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>]

On Fri, Jan 03, 2025 at 08:59:51PM +0530, Anand Moon wrote:
> Hi Niklas
> 
> On Fri, 3 Jan 2025 at 20:40, Niklas Cassel <cassel@kernel.org> wrote:
> >
> > On Fri, Jan 03, 2025 at 08:36:18PM +0530, Anand Moon wrote:
> > > > >
> > > > > We need to enable the GMAC PHY and reset it using the proper GPIO pin
> > > > > (PCIE_PERST_L).
> > > > > Please refer to the schematic for more details.
> > > >
> > > > The PERST# GPIO is already asserted + deasserted from the PCIe Root Complex
> > > > (host) driver:
> > > > https://github.com/torvalds/linux/blob/v6.13-rc5/drivers/pci/controller/dwc/pcie-dw-rockchip.c#L191-L206
> > > >
> > > > which will cause the endpoint device (a RTL8125 NIC in this case)
> > > > to be reset during bootup.
> > > >
> > > Thanks for letting me know. It seems like a workaround.
> > > I'll try to disable this and test it again.
> > >
> > > My point is that we haven't enabled the GMAC PHY (device nodes)
> > > and must properly reset the GMAC.
> > >
> > > We're relying on the code above hack to do that job.
> >
> > I do not think it is a hack.
> >
> > If you look in most PCIe controller drivers, they toggle PERST before
> > enumerating the bus:
> > $ git grep gpiod_set_value drivers/pci/controller/
> >
> 
> Ok, understood. However, we have multiple reset lines per controller,
> so the PCIe driver will reset these lines using gpiod_set_value.
> 
> PCIE30X4_PERSTn_M1_L
> PCIE30x1_0_PERSTn_M1_L
> PCIE_PERST_L

PERST# gpio is unique per controller instance and will be asserted/deasserted
by the PCIe controller driver itself. Endpoint drivers should not touch these.

And most of the PCIe endpoint devices do not need to be described in devicetree
as PCIe is a discoverable bus. But we do define some of them if they require any
special board configuration.

- Mani

-- 
மணிவண்ணன் சதாசிவம்