Re: [PATCH] pmdomain: rockchip: mark PD_PIPE as active wakeup on RK3568

From: Ulf Hansson

Date: Fri Nov 13 2026 - 18:44:35 EST


On Thu, Oct 22, 2026 at 5:04 AM Coia Prant <coiaprant@xxxxxxxxx> wrote:
>
> Coia Prant <coiaprant@xxxxxxxxx> 于2026年10月6日周二 07:27写道:
> >
> > On September 25, 2026 12:17:51 PM GMT+08:00, Coia Prant <coiaprant@xxxxxxxxx> wrote:
> > >The PD_PIPE power domain on RK3568 hosts the XPCS, the PCIe
> > >controllers and the SATA controllers. All of them can act as
> > >wakeup sources for the system:
> > >
> > > - the XPCS for MAC WoL, through the SGMII link
> > > - the PCIe controllers for PME#
> > > - the SATA controllers for hotplug wake
> > >
> > >The domain is registered with active_wakeup = false, so genpd powers
> > >it down during system suspend even when a consumer is on the wakeup
> > >path. That kills the XPCS and the SerDes, which breaks MAC WoL on
> > >RK3568 SGMII boards, and it also powers down the PCIe and SATA
> > >controllers while their drivers still expect the link to be alive.
> > >
> > >Mark PD_PIPE as active_wakeup so that genpd honours the wakeup path
> > >of its consumers. Consumers that call device_set_wakeup_path() keep
> > >the domain powered; consumers that do not mark themselves as wakeup
> > >sources are unaffected and the domain is still powered down when
> > >nothing needs it.
> > >
> > >Several other Rockchip SoCs already mark their PCIe and USB power
> > >domains as active wakeup, for example RK3588's pcie and usb domains
> > >and RK3576's usb domain. PD_PIPE hosts the same kind of
> > >wakeup-capable controllers on RK3568, so this follows the same
> > >pattern.
> > >
> > >Signed-off-by: Coia Prant <coiaprant@xxxxxxxxx>
>
> Hi Ulf,
>
> Gentle ping on this patch. It was sent on Sep 25 and Patchwork shows it
> as "Handled Elsewhere", but I can't find it merged in linux-pm,
> linux-rockchip, linux-next or mainline, and I haven't received any
> review feedback.
>
> Since then, the XPCS driver and DTS that depend on PD_PIPE staying
> powered across suspend have been merged and sent upstream:
>
> - XPCS driver in net-next: https://git.kernel.org/netdev/net-next/c/5e91edfb0e45
>
> Could you let me know which tree this should go through, or whether it
> was already picked somewhere? If it needs to be routed differently, I'm
> happy to resend it.
>
> Thanks,
> Coia

Apologize for the huge delay. I was waiting for an ack from some of
the rockchip maintainers.

Anyway, this looks fine to me so, applied for next, thanks!

Kind regards
Uffe