Re: [PATCH v2] PCI/portdrv: Allow probing even without child services

[Brian Norris <briannorris@chromium.org>]

Hi Lukas, Bjorn,

On Mon, Jul 20, 2026 at 05:25:56PM -0700, Brian Norris wrote:
> On Mon, Jul 20, 2026 at 03:44:19PM -0500, Bjorn Helgaas wrote:
> > On Fri, Jul 17, 2026 at 10:26:53AM +0200, Lukas Wunner wrote:
> > > On Thu, Jul 16, 2026 at 12:26:41PM -0500, Bjorn Helgaas wrote:
> > > > On Mon, Feb 09, 2026 at 05:15:35PM -0800, Brian Norris wrote:
> > > > > The PCIe port driver fails to probe if it finds no child services,
> > > > > presumably under the assumption that the driver is not useful in that
> > > > > case. However, the driver *can* still be useful for power management
> > > > > support -- namely, it still configures the port for runtime PM / D3,
> > > > > which may be important for allowing a bridge to enter low power modes.
> > > > > 
> > > > > Thus, allow probe to succeed even if no IRQs and no child services are
> > > > > available. This also mirrors existing behavior for ports that have no
> > > > > PCIe capabilities, where we'd also probe successfully.
> > > > > 
> > > > > This change is a bit more important after commit f5cd8a929c82 ("PCI:
> > > > > dwc: Remove MSI/MSIX capability for Root Port if iMSI-RX is used as MSI
> > > > > controller"), because it's common for some DWC-based systems to:
> > > > > 
> > > > > 1. have only have the "aer" and "pcie_pme" port services available and
> > > > > 2. not define legacy INTx interrupts properly in their device tree.
> > > > > 
> > > > > After commit f5cd8a929c82, such systems may fail
> > > > > pcie_init_service_irqs() and so exit with -ENODEV.
> > > > > 
> > > > > Link: https://lore.kernel.org/all/nyada24tqwlkzdceyoxbzitzygvp4elvj5oajnqdwb33xkcdwk@76vnrx45fsfd/
> > > > > Signed-off-by: Brian Norris <briannorris@chromium.org>
> > > > 
> > > > Applied to pci/portdrv for v7.3, thank you!
> > > 
> > > Some thoughts on the version now applied to pci/portdrv:
> > > 
> > > * After this patch, the return value of pcie_device_init() is no longer
> > >   evaluated, so that function should be changed to return void as part
> > >   of this patch.
> 
> Yes, seems reasonable.
> 
> But then, possibly negated below.

Not done in v4, since we now use the return code differently.

> > > * If no port services are initalized, e.g. because pcie_device_init()
> > >   failed for all of them, the allocated IRQ vectors are not freed.
> > >   They're hogged, which may be a problem on systems which only support
> > >   very few IRQ vectors such as Sophgo SG2042:
> > > 
> > >   https://lore.kernel.org/r/0f42afefd9322779af5463b696c55b08d2296ea8.camel@iscas.ac.cn
> > 
> > Seems like a good idea to free IRQ vectors we don't use.  It *looks*
> > like it should be safe to call pci_free_irq_vectors() here and then
> > again in pcie_port_device_remove() if a port is removed, but Brian
> > wasn't sure, and I didn't look any deeper than pci_disable_msix() and
> > pci_disable_msi().
> 
> I think everything in pci_free_irq_vectors() is written such that it can
> safely be called twice. I'll see if I can verify that better and submit
> an update. I was just unsure at the moment of review, especially
> considering you had other real double-cleanup bugs in your proposal.
> 
> > I guess if we did this we'd need to retain the pcie_device_init().
> 
> Ack.

Done in v4.

> > > * It might be clearer to call pci_clear_master() directly in the
> > >   if-block which is amended to "goto out".  Then the "goto out"
> > >   can be changed to a "return 0".  It seems "if (!capabilities)"
> > >   cannot happen through any other code path, so it's unclear why
> > >   the "out" label is necessary.
> > 
> > It does look nicer that way.
> 
> Hmm, yes. I may have rewritten it enough times that it was never clear
> how many error paths we were going to have. As soon as you get more than
> 1 error path, it gets tricky to do this.
> 
> I'll revisit this one along with the previous suggestions.

The additional "free IRQs" suggestion means we again have 2 paths that
need the same cleanup, so I chose to keep the 'goto'. There are other
creative ways to avoid the 'goto', but I felt like "creative" often
doesn't help readability and maintainability.

> > > * There's a duplicate "have" in the commit message ("have only have").
> > 
> > I fixed this on pci/portdrv.
> 
> Commit message looks good, thanks.
> 
> > What do you think, Brian?
> 
> I'll give the above a shot. I suppose that would work best as a v4
> (pretending your pci/portdrv is only a reference, and not a patch-basis
> target).

Submitted v4:

https://lore.kernel.org/all/20260722160942.v4.1.I5fd5d83f518681b3949d8ab2f16ba8244fd3e774@changeid/

Brian