Re: [PATCH] PCI/PM: Skip the suspend_noirq config save if runtime-suspended

[Navon John Lukose]

On Wed, Sep 23, 2026 at 12:27:27PM +0200, Lukas Wunner wrote:
> Hm, pci_pm_suspend_noirq() should bail out earlier because of the
> dev_pm_skip_suspend() check.  That check resolves to:
>
>   dev_pm_smart_suspend(dev) && pm_runtime_status_suspended(dev)
>
> The code comment in device_prepare_smart_suspend() explains that
> smart_suspend is true for devices without PM callbacks.  But only
> if the parent has smart suspend enabled as well.  I guess that's
> the sticking point?  That the port above the card reader doesn't
> have smart suspend enabled?

It is that check, but it fails on the card reader itself, before the
parent is looked at.  no_pm_callbacks is only set when the bus has no
PM ops either, and pci_bus_type always has pci_dev_pm_ops, so no PCI
device ever counts as having no callbacks.  With no driver to set
DPM_FLAG_SMART_SUSPEND, the reader never gets smart_suspend.

pcieport does set the flag for the port, and on this machine the port
stays in D3hot through suspend.

Thanks,
Navon