Re: [PATCH v2 1/1] PCI: Fix BAR resizing when VF BARs are assigned

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 2661 bytes --]

+ Jakub
+ Alexander

On Mon, 17 Mar 2025, Alex Williamson wrote:
> On Mon, 17 Mar 2025 19:18:03 +0100
> Michał Winiarski <michal.winiarski@intel.com> wrote:
> > On Fri, Mar 14, 2025 at 08:56:49AM -0600, Alex Williamson wrote:
> > > On Fri,  7 Mar 2025 16:03:49 +0200
> > > Ilpo Järvinen <ilpo.jarvinen@linux.intel.com> wrote:
> > >   
> > > > __resource_resize_store() attempts to release all resources of the
> > > > device before attempting the resize. The loop, however, only covers
> > > > standard BARs (< PCI_STD_NUM_BARS). If a device has VF BARs that are
> > > > assigned, pci_reassign_bridge_resources() finds the bridge window still
> > > > has some assigned child resources and returns -NOENT which makes
> > > > pci_resize_resource() to detect an error and abort the resize.
> > > > 
> > > > Change the release loop to cover all resources up to VF BARs which
> > > > allows the resize operation to release the bridge windows and attempt
> > > > to assigned them again with the different size.
> > > > 
> > > > As __resource_resize_store() checks first that no driver is bound to
> > > > the PCI device before resizing is allowed, SR-IOV cannot be enabled
> > > > during resize so it is safe to release also the IOV resources.  
> > > 
> > > Is this true?  pci-pf-stub doesn't teardown SR-IOV on release, which I
> > > understand is done intentionally.  Thanks,  

Thanks for reviewing. I'm sorry I just took Michał's word on this for 
granted so I didn't check it myself.

I could amend __resource_resize_store() to return -EBUSY if SR-IOV is 
there despite no driver being present, but lets hear if Alexander or Jakub 
has some input on this.

> > Is that really intentional?
> > PCI warns when that scenario occurs:
> > https://elixir.bootlin.com/linux/v6.13.7/source/drivers/pci/iov.c#L936
> 
> Yep, it warns.  It doesn't prevent it from happening though.
> 
> > I thought that the usecase is binding pci-pf-stub, creating VFs, and
> > letting the driver be.
> > But unbinding after creating VFs? What's the goal of that?
> > Perhaps we're just missing .remove() in pci-pf-stub?
> 
> I guess I don't actually know that leaving SR-IOV enabled was
> intentional, maybe it was an oversight.  The original commit only
> mentions the case of a device that requires nothing but this shim as
> the PF driver.  A pci_warn() isn't much disincentive, the system might
> already have taints.  If it's something that we really want to show as
> broken, it'd probably need to be a WARN_ON. 

Added Alexander and Jakub, perhaps they remember something or if there are 
caveats going either way.

-- 
 i.