Re: [PATCH] pci: Workaround Stratus broken PCIE hierarchy

[Matthew Wilcox <matthew@wil.cx>]

On Mon, Nov 14, 2011 at 09:44:50AM -0500, Prarit Bhargava wrote:
> On 11/11/2011 03:35 PM, Don Dutile wrote:
>  We should make it work automatically, either with
> >> some sort of machine-dependent quirk, or (preferably) with a change to
> >> the generic algorithm so it can handle these "broken" topologies along
> >> with all the correct ones.
> 
> The problem is when I'm examining "this" PCIE bridge I have no knowledge of the remaining hierarchy below it.  In order to find the broken topology we need to do a two pass enumeration (NO.) or minimally query all possible devices immediately connected to it.  If we do that then what is the purpose of the only_one_child() check?

We have enough information to do this quirk.

pci_scan_slot calls pci_scan_single_device() for the devfn 0.
We can look at the device we found and quirk the parent device.
Something like this ...

 int pci_scan_slot(struct pci_bus *bus, int devfn)
[...]
 	dev = pci_scan_single_device(bus, devfn);
	if (!dev)
		return 0;
	if (!dev->is_added)
		nr++;
+	/* Quirk to fix NEC/Stratus broken PCIe topologies */
+	if ((dev->pcie_type == PCI_EXP_TYPE_DOWNSTREAM) &&
+	    (bus->self->pcie_type == PCI_EXP_TYPE_DOWNSTREAM))
+		bus->self->pcie_type = PCI_EXP_TYPE_UPSTREAM;

	if (pci_ari_enabled(bus))
		next_fn = next_ari_fn;
[...]

As an aside, I'm quite shocked that this patch has been in the kernel
for almost two years and this is the first time anybody's tested it on
one of these systems.

-- 
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."