Re: [PATCH v6 08/11] PCI: dwc: Invoke post_init in dw_pcie_resume_noirq()

[Bjorn Helgaas]

On Fri, Jan 09, 2026 at 02:10:07AM +0000, Hongxing Zhu wrote:
> > -----Original Message-----
> > From: Bjorn Helgaas <helgaas@xxxxxxxxxx>
> > Sent: 2026年1月9日 5:50
> > To: Hongxing Zhu <hongxing.zhu@xxxxxxx>
> > Cc: Frank Li <frank.li@xxxxxxx>; l.stach@xxxxxxxxxxxxxx;
> > lpieralisi@xxxxxxxxxx; kwilczynski@xxxxxxxxxx; mani@xxxxxxxxxx;
> > robh@xxxxxxxxxx; krzk+dt@xxxxxxxxxx; conor+dt@xxxxxxxxxx;
> > bhelgaas@xxxxxxxxxx; shawnguo@xxxxxxxxxx; s.hauer@xxxxxxxxxxxxxx;
> > kernel@xxxxxxxxxxxxxx; festevam@xxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx;
> > linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx;
> > imx@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> > Subject: Re: [PATCH v6 08/11] PCI: dwc: Invoke post_init in
> > dw_pcie_resume_noirq()
> > 
> > On Wed, Oct 15, 2025 at 11:04:25AM +0800, Richard Zhu wrote:
> > > If the ops has post_init callback, invoke it in dw_pcie_resume_noirq().
> > 
> > I'm trying to write the merge commit log for this branch, and I don't quite
> > understand this.
> > 
> > The effect is to apply the GEN3_ZRXDC_NONCOMPL workaround for the
> > ERR051586 erratum, and Mani added the hint that this enables REFCLK
> > during resume.  But it seems weird that we apply a REFCLK workaround
> > after the link is already up.
> > 
> > During probe, .post_init() is run after pci_host_probe(), so we apply the
> > workaround after enumerating all the devices, which means REFCLK must
> > already be valid and the link is already up.
> > 
> > Is "enabling REFCLK" actually what imx_pcie_host_post_init() does?
>
> The codes are used to clean up the CLKREQ# override active low
> configurations after link is up and the CLKREQ# is drove to low by
> remote endpoint device at this point(support-clkreq is TRUE).
>
> It paves the way to support the CLKREQ# toggling mandatory required
> by L1SS.
>
> > Could the workaround be done in imx_pcie_host_init() before the
> > link is brought up?  If it could, it looks like we wouldn't need
> > imx_pcie_host_post_init() at all.
>
> Two actions are done in imx_pcie_post_init().
>
> One is to apply the workaround of ERR051586 by commit 744a1c20ce93
> ("PCI: imx6: Add workaround for errata ERR051586"). It should be
> applied after link is up.
>
> The other one is to clean up the CLKREQ# override active low
> configurations previous set in imx_pcie_host_init().

Right.  I'm not asking about the CLKREQ# part.  I'm asking about the
IMX_PCIE_FLAG_8GT_ECN_ERR051586 part.

The current commit log for this patch is this:

  Some SoCs like i.MX95 require enabling REFCLK after resuming from suspend
  in their post_init callback. So invoke the callback at the end of
  dw_pcie_resume_noirq() if available.

When .post_init() is called, the link is already up and PCIe devices
have already been enumerated.  I think REFCLK is required for the link
to come up, so it doesn't sound right to me that .post_init() would be
*enabling* REFCLK.

> > For now, I put this in the merge commit log:
> > 
> >   - Apply i.MX95 ERR051586 erratum workaround for REFCLK issue during
> >     resume (Richard Zhu)