Re: [PATCH] clk: qcom: gcc-eliza: Use retention for PCIe power domains

[Abel Vesa]

On 26-10-28 04:29:31, Manivannan Sadhasivam wrote:
> On Tue, Oct 20, 2026 at 03:41:54PM +0300, Abel Vesa wrote:
> > On 26-10-14 14:09:04, Loic Poulain wrote:
> > > On Wed, Oct 14, 2026 at 1:59 PM Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On 26-10-12 14:58:43, Jagadeesh Kona wrote:
> > > > >
> > > > >
> > > > > On 10/12/2026 1:29 AM, Abel Vesa wrote:
> > > > > > When resuming from system suspend on Eliza CQS EVK, gcc_pcie_0_gdsc
> > > > > > gets stuck off when powering back on, preventing the system from resuming.
> > > > > >
> > > > > > Use retention for all four PCIe GDSCs, as on SM8550, to preserve their
> > > > > > state across suspend and allow the system to resume.
> > > > > >
> > > > > > Fixes: 3d356ab4a1ec ("clk: qcom: Add support for Global clock controller on Eliza")
> > > > > > Assisted-by: LLM
> > > > > > Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx>
> > > > > > ---
> > > > > >  drivers/clk/qcom/gcc-eliza.c | 8 ++++----
> > > > > >  1 file changed, 4 insertions(+), 4 deletions(-)
> > > > > >
> > > > >
> > > > >
> > > > > Hi Abel,
> > > > >
> > > > > Using PWRSTS_RET_ON keeps the GDSC forever ON after first enable and the GDSC never
> > > > > collapses.
> > > > >
> > > > > For similar issue on Lemans (pcie_0_gdsc gets stuck off when powering back on), picking
> > > > > the below changes from PCIE side helped to power ON the GDSC properly without any issue.
> > > > >
> > > > > https://lore.kernel.org/all/20260728-hotplug_pci-v1-1-d1f38cfb9122@xxxxxxxxxxxxxxxx/
> > > > > https://lore.kernel.org/all/20260729071514.859778-1-manivannan.sadhasivam@xxxxxxxxxxxxxxxx/
> > > > >
> > > > > Can you please try if above changes are helping on Eliza also?
> > > >
> > > > Thanks for the hint. Mani's patch (v2) is already upstream, so just enabling
> > > > CONFIG_HOTPLUG_PCI_PCIE did the trick.
> > > >
> > > > Suspend/resume works fine now on Eliza, so this patch can be ignored then.
> > > 
> > > So should the retention configuration above be conditional on
> > > CONFIG_HOTPLUG_PCI_PCIE, or should we simply depend on
> > > CONFIG_HOTPLUG_PCI_PCIE?
> > 
> > None. I think QCOM_PCIE should select it.
> 
> Nope. It should be selected by defconfig [1].
> 
> Just to clarify on how the hotplug support is fixing this issue. Once we enable
> the hotplug driver, the Root Port will be allowed to enter D3Hot during suspend,
> which is a pre-requisite for powering off the PCIe devices and the controller.
> So once all the conditions are satisfied, the controller and devices will be
> powered off. So even if the PCIe GDSC gets powered off at the end of suspend, it
> won't affect PCIe devices as they already lost their power. So we don't need
> GDSC retention.
> 
> But that being said, if any of the PCIe devices support wakeup like WOL from
> D3Hot, then we do not power off the controller and devices. In that case, GDSC
> retention would be required to keep the controller/device state intact. You can
> reproduce this issue by enabling wakeup for the PCIe device from sysfs and
> attempt suspend.

Thanks for explaining it.

> 
> So to cleanly fix this issue, we need the PCIe controller driver to signal
> retention state to the GDSC driver or the GDSC driver has to figure this on its

So until we have a solution for controling the GDSC retention from PCIe
all the way through provider, what do you think should be the stop gap
solution?