Re: [PATCH v2 2/6] pci/hotplug/pnv_php: Work around switches with broken

[Bjorn Helgaas]

On Wed, Jun 18, 2025 at 02:50:04PM -0500, Timothy Pearson wrote:
> ----- Original Message -----
> > From: "Bjorn Helgaas" <helgaas@xxxxxxxxxx>
> > To: "Timothy Pearson" <tpearson@xxxxxxxxxxxxxxxxxxxxx>
> > Cc: "linuxppc-dev" <linuxppc-dev@xxxxxxxxxxxxxxxx>, "linux-kernel" <linux-kernel@xxxxxxxxxxxxxxx>, "linux-pci"
> > <linux-pci@xxxxxxxxxxxxxxx>, "Madhavan Srinivasan" <maddy@xxxxxxxxxxxxx>, "Michael Ellerman" <mpe@xxxxxxxxxxxxxx>,
> > "christophe leroy" <christophe.leroy@xxxxxxxxxx>, "Naveen N Rao" <naveen@xxxxxxxxxx>, "Bjorn Helgaas"
> > <bhelgaas@xxxxxxxxxx>, "Shawn Anastasio" <sanastasio@xxxxxxxxxxxxxxxxxxxxx>, "Lukas Wunner" <lukas@xxxxxxxxx>
> > Sent: Wednesday, June 18, 2025 2:44:00 PM
> > Subject: Re: [PATCH v2 2/6] pci/hotplug/pnv_php: Work around switches with broken
> 
> > [+cc Lukas, pciehp expert]
> > 
> > On Wed, Jun 18, 2025 at 11:56:54AM -0500, Timothy Pearson wrote:
> >>  presence detection
> > 
> > (subject/commit wrapping seems to be on all of these patches)
> > 
> >> The Microsemi Switchtec PM8533 PFX 48xG3 [11f8:8533] PCIe switch system
> >> was observed to incorrectly assert the Presence Detect Set bit in its
> >> capabilities when tested on a Raptor Computing Systems Blackbird system,
> >> resulting in the hot insert path never attempting a rescan of the bus
> >> and any downstream devices not being re-detected.
> > 
> > Seems like this switch supports standard PCIe hotplug?  Quite a bit of
> > this driver looks similar to things in pciehp.  Is there some reason
> > we can't use pciehp directly?  Maybe pciehp could work if there were
> > hooks for the PPC-specific bits?
> 
> While that is a good long term goal that Raptor is willing to work
> toward, it is non-trivial and will require buy-in from other
> stakeholders (e.g. IBM).  If practical, I'd like to get this series
> merged first, to fix the broken hotplug on our hardware that is
> deployed worldwide, then in parallel see what can be done to merge
> PowerNV support into pciehp.  Would that work?

Yeah, it wouldn't make sense to switch horses at this stage.

I guess I was triggered by this patch, which seems to be a workaround
for a defect in a device that is probably also used on non-PPC
systems, and pciehp would need a similar workaround.  But I guess you
go on to say that pciehp already does something similar, so it guess
it's already covered.

Bjorn