[PATCH v2] PCI: dwc: Fix resume failure if no EP is connected at some platforms

[Hongxing Zhu]

> -----Original Message-----
> From: Bjorn Helgaas <helgaas at kernel.org>
> Sent: 2024年11月7日 6:30
> To: Hongxing Zhu <hongxing.zhu at nxp.com>
> Cc: kwilczynski at kernel.org; bhelgaas at google.com;
> lorenzo.pieralisi at arm.com; Frank Li <frank.li at nxp.com>; mani at kernel.org;
> linux-pci at vger.kernel.org; linux-arm-kernel at lists.infradead.org;
> linux-kernel at vger.kernel.org; kernel at pengutronix.de; imx at lists.linux.dev
> Subject: Re: [PATCH v2] PCI: dwc: Fix resume failure if no EP is connected at
> some platforms
> 
> On Wed, Nov 06, 2024 at 01:59:41AM +0000, Hongxing Zhu wrote:
> > > -----Original Message-----
> > > From: Bjorn Helgaas <helgaas at kernel.org>
> > > Sent: 2024年11月6日 7:27
> > > To: Hongxing Zhu <hongxing.zhu at nxp.com>
> > > Cc: kwilczynski at kernel.org; bhelgaas at google.com;
> > > lorenzo.pieralisi at arm.com; Frank Li <frank.li at nxp.com>;
> > > mani at kernel.org; linux-pci at vger.kernel.org;
> > > linux-arm-kernel at lists.infradead.org;
> > > linux-kernel at vger.kernel.org; kernel at pengutronix.de;
> > > imx at lists.linux.dev
> > > Subject: Re: [PATCH v2] PCI: dwc: Fix resume failure if no EP is
> > > connected at some platforms
> > >
> > > On Mon, Jul 22, 2024 at 02:15:13PM +0800, Richard Zhu wrote:
> > > > The dw_pcie_suspend_noirq() function currently returns success
> > > > directly if no endpoint (EP) device is connected. However, on some
> > > > platforms, power loss occurs during suspend, causing dw_resume()
> > > > to do nothing in this case.
> > > > This results in a system halt because the DWC controller is not
> > > > initialized after power-on during resume.
> 
> > > > @@ -933,23 +933,23 @@ int dw_pcie_suspend_noirq(struct dw_pcie
> *pci)
> > > >  	if (dw_pcie_readw_dbi(pci, offset + PCI_EXP_LNKCTL) &
> > > PCI_EXP_LNKCTL_ASPM_L1)
> > > >  		return 0;
> > > >
> > > > -	if (dw_pcie_get_ltssm(pci) <= DW_PCIE_LTSSM_DETECT_ACT)
> > > > -		return 0;
> > > > -
> > > > -	if (pci->pp.ops->pme_turn_off)
> > > > -		pci->pp.ops->pme_turn_off(&pci->pp);
> > > > -	else
> > > > -		ret = dw_pcie_pme_turn_off(pci);
> > > > +	if (dw_pcie_get_ltssm(pci) > DW_PCIE_LTSSM_DETECT_ACT) {
> > > > +		/* Only send out PME_TURN_OFF when PCIE link is up */
> > > > +		if (pci->pp.ops->pme_turn_off)
> > > > +			pci->pp.ops->pme_turn_off(&pci->pp);
> > > > +		else
> > > > +			ret = dw_pcie_pme_turn_off(pci);
> > >
> > > This looks possibly racy since the link can go down at any point.
> >
> > When link is down and without this commit changes,
> > dw_pcie_suspend_noirq() return directly, and the PME_TURN_OFF wouldn't
> > be kicked off.
> 
> Right, that's the code change.
> 
> > I change the behavior to issue the PME_TURN_OFF when link is up here.
> 
> But I don't think you responded to the race question.  What happens here?
> 
>   if (dw_pcie_get_ltssm(pci) > DW_PCIE_LTSSM_DETECT_ACT) {
>     --> link goes down here <--
>     pci->pp.ops->pme_turn_off(&pci->pp);
> 
> You decide the LTSSM is active and the link is up.  Then the link goes down.
> Then you send PME_Turn_off.  Now what?
> 
> If it's safe to try to send PME_Turn_off regardless of whether the link is up or
> down, there would be no need to test the LTSSM state.
I made a local tests on i.MX95/i.MX8QM/i.MX8MP/i.MX8MM/i.MX8MQ, and
confirm that it's safe to send PME_TURN_OFF although the link is down.
You're right. It's no need to test LTSSM state here.
So do the L2 poll listed above after PME_TURN_OFF is sent.

Since the 6.13 merge window is almost closed.
How about I prepare another Fix patch to do the following items for incoming
6.14?
- Before sending PME_TURN_OFF, don't test the LTSSM stat.
- Remove the L2 stat poll, after sending PME_TURN_OFF.

Best Regards
Richard Zhu

> 
> Bjorn