Re: [PATCH v1 2/2] PCI: imx6: Use domain number replace the hardcodes

[Bjorn Helgaas <helgaas@kernel.org>]

On Wed, Mar 12, 2025 at 09:28:02AM +0100, Lucas Stach wrote:
> Am Mittwoch, dem 12.03.2025 um 04:05 +0000 schrieb Hongxing Zhu:
> > > -----Original Message-----
> > > From: Bjorn Helgaas <helgaas@kernel.org>
> > > Sent: 2025年3月11日 23:55
> > > On Tue, Mar 11, 2025 at 01:11:04AM +0000, Hongxing Zhu wrote:
> > > > > -----Original Message-----
> > > > > From: Bjorn Helgaas <helgaas@kernel.org>
> > > > > Sent: 2025年3月10日 23:11
> > > > > On Wed, Feb 26, 2025 at 10:42:56AM +0800, Richard Zhu wrote:
> > > > > > Use the domain number replace the hardcodes to uniquely identify
> > > > > > different controller on i.MX8MQ platforms. No function changes.
> > > > > > 
> > > > > > Signed-off-by: Richard Zhu <hongxing.zhu@nxp.com>
> > > > > > ---
> > > > > >  drivers/pci/controller/dwc/pci-imx6.c | 14 ++++++--------
> > > > > >  1 file changed, 6 insertions(+), 8 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/pci/controller/dwc/pci-imx6.c
> > > > > > b/drivers/pci/controller/dwc/pci-imx6.c
> > > > > > index 90ace941090f..ab9ebb783593 100644
> > > > > > --- a/drivers/pci/controller/dwc/pci-imx6.c
> > > > > > +++ b/drivers/pci/controller/dwc/pci-imx6.c
> > > > > > @@ -41,7 +41,6 @@
> > > > > >  #define IMX8MQ_GPR_PCIE_CLK_REQ_OVERRIDE	BIT(11)
> > > > > >  #define IMX8MQ_GPR_PCIE_VREG_BYPASS		BIT(12)
> > > > > >  #define IMX8MQ_GPR12_PCIE2_CTRL_DEVICE_TYPE	GENMASK(11,
> > > 8)
> > > > > > -#define IMX8MQ_PCIE2_BASE_ADDR			0x33c00000
> > > > > > 
> > > > > >  #define IMX95_PCIE_PHY_GEN_CTRL			0x0
> > > > > >  #define IMX95_PCIE_REF_USE_PAD			BIT(17)
> > > > > > @@ -1474,7 +1473,6 @@ static int imx_pcie_probe(struct
> > > > > > platform_device
> > > > > *pdev)
> > > > > >  	struct dw_pcie *pci;
> > > > > >  	struct imx_pcie *imx_pcie;
> > > > > >  	struct device_node *np;
> > > > > > -	struct resource *dbi_base;
> > > > > >  	struct device_node *node = dev->of_node;
> > > > > >  	int i, ret, req_cnt;
> > > > > >  	u16 val;
> > > > > > @@ -1515,10 +1513,6 @@ static int imx_pcie_probe(struct
> > > > > platform_device *pdev)
> > > > > >  			return PTR_ERR(imx_pcie->phy_base);
> > > > > >  	}
> > > > > > 
> > > > > > -	pci->dbi_base = devm_platform_get_and_ioremap_resource(pdev,
> > > 0,
> > > > > &dbi_base);
> > > > > > -	if (IS_ERR(pci->dbi_base))
> > > > > > -		return PTR_ERR(pci->dbi_base);

> > Use the domain number replace the hardcodes to uniquely identify
> > different controller on i.MX8MQ platforms. No function changes.
> > Please make sure the " linux,pci-domain" is set for i.MX8MQ correctly, since
> >  the controller id is relied on it totally.
> > 
> This breaks running a new kernel on an old DT without the
> linux,pci-domain property, which I'm absolutely no fan of. We tried
> really hard to keep this way around working in the i.MX world.
> 
> I'm fine with using the property if present and even mandating it for
> new platforms, but getting rid of the existing code for the i.MX8MQ
> platform is only a marginal cleanup of the driver code with the obvious
> downside of the above breakage.

I don't know the history of these DTs, but if there are any old DTs
for platforms that use controller 1 but lack 'linux,pci-domain', I
agree that we should not break them.

If we need to retain the dbi_base check so that old DTs continue to
work, I think it should look something like this:

  domain = of_get_pci_domain_nr(node);
  if (domain >= 0) {
      if (domain > 1)
          return dev_err_probe(..., "invalid domain %d\n", domain);
      imx_pcie->controller_id = domain;
  } else {
      dev_warn(..., "DT lacks linux,pci-domain, falling back to DBI addr\n");
      dbi_res = platform_get_resource(pdev, IORESOURCE_MEM, index);
      if (dbi_res->start == IMX8MQ_PCIE2_BASE_ADDR)
          imx_pcie->controller_id = 1;
  }

The previous code used devm_platform_get_and_ioremap_resource() and
set pci->dbi_base, but (1) there's no need to set pci->dbi_base since
the DWC core does that anyway, and (2) I think using ioremap() means
we depend on CPU virt == CPU phys, and I don't think we need to depend
on that.

Bjorn