[PATCH 2.6.24-rc4] PCI: refuse to re-add a device to a bus upon pci_scan_child_bus()

From: Eran Liberty
Date: Sun Jul 20 2008 - 06:41:59 EST

Dear Penguins,

As a follow up on my own post ( http://lkml.org/lkml/2008/6/18/195 ), I have solved a small bug/misbehavior within the pci_scan_single_device().

The misbehavior manifest itself upon calling pci_scan_child_bus() with a bus which already contain some devices. After scanning is done devices that already existed will be present twice in the PCI bus devise list. Once with is_added indication and once without. Trying to add this bus will cause a resource conflict as the same device is already present and initialized.

This patch will simply prevent a device to be added to a bus list if it is already there.

Points to consider:
1. I am not a PCI Guru and might have over looked the bigger picture. Is it OK to prevent a device of being re-added?
2. I have decided that two devices are, in fact, the same instance if it has the same: vendor, device, and devfn. Is there a finer test for a device identity?

p.s. As per http://www.kernel.org/pub/linux/docs/lkml/#s1-10 it is preferred for Mozilla Firefox clients to attach the patch, hence my patch is attached.


Signed-off-by: Eran Liberty <liberty@xxxxxxxxxxxxx>
Index: drivers/pci/probe.c
--- drivers/pci/probe.c (revision 102)
+++ drivers/pci/probe.c (revision 103)
@@ -996,11 +996,21 @@
struct pci_dev *__ref pci_scan_single_device(struct pci_bus *bus, int devfn)
struct pci_dev *dev;
+ struct list_head *ln;

dev = pci_scan_device(bus, devfn);
if (!dev)
return NULL;

+ /* refuse to re-add a device */
+ list_for_each(ln, &bus->devices) {
+ struct pci_dev *d = pci_dev_b(ln);
+ if ((d->vendor == dev->vendor) &&
+ (d->device == dev->device) &&
+ (d->devfn == dev->devfn))
+ return NULL;
+ }
pci_device_add(dev, bus);

return dev;