Re: 2.6.24-rc5 sysfs pci bridge duplicate symlink

[Russell King - ARM Linux]

On Fri, Dec 28, 2007 at 03:03:22PM -0800, Andrew Morton wrote:
> On Fri, 28 Dec 2007 13:11:37 +0000 "Bahadir Balban" <bahadir.balban@xxxxxxxxx> wrote:
> 
> > Hi,
> > 
> > On ARM with PCI, I get this error since -rc2 (didn't try rc1):
> > 
> > sysfs: duplicate filename 'bridge' can not be created
> > WARNING: at fs/sysfs/dir.c:424 sysfs_add_one()
> > [<c00287fc>] (dump_stack+0x0/0x14) from [<c00cbc64>] (sysfs_add_one+0x50/0xf4)
> > [<c00cbc14>] (sysfs_add_one+0x0/0xf4) from [<c00ccd30>]
> > (sysfs_create_link+0xec/0x184)
> >  r6:cfca8b48 r5:cfca96c8 r4:cfc2fec0
> > [<c00ccc44>] (sysfs_create_link+0x0/0x184) from [<c0160b44>]
> > (pci_bus_add_devices+0xf0/0x144)
> >  r7:cfc5a2d4 r6:cfc5a2d4 r5:cfc5a2c0 r4:cfcad800
> > [<c0160a54>] (pci_bus_add_devices+0x0/0x144) from [<c0160b30>]
> > (pci_bus_add_devices+0xdc/0x144)
> >  r7:cfc5a3d4 r6:cfc5a3d4 r5:cfc5a3c0 r4:cfcadc00
> > [<c0160a54>] (pci_bus_add_devices+0x0/0x144) from [<c0160b30>]
> > (pci_bus_add_devices+0xdc/0x144)
> >  r7:cfc5a4d4 r6:cfc5a4d4 r5:cfc5a4c0 r4:cfc5d400
> > [<c0160a54>] (pci_bus_add_devices+0x0/0x144) from [<c000a934>]
> > (pci_common_init+0x148/0x184)
> >  r7:c035fd58 r6:cfc3e760 r5:c001e030 r4:cfc5a4c0
> > 
> > 
> > Any idea why?
> > 
> 
> (suitable cc added)

Looks to me as if the code added to pci_bus_add_devices() is off it's
rocker.

In brief, pci_bus_add_devices() is supposed to be callable given *any*
state of the bus, and it will add any new devices found in the bus tree
from the bus that it's called for downwards.  (That's how I designed
that bit of PCI code - so it can cope with hotpluggable PCI bridges.)

It still does this.  However, some bright spark decided to add sysfs
links to it - and break it.  Now, whenever we descend into a subordinate
bus (and return to a higher level) we will try to create a sysfs symlink
even if the bus is one that we already registered.

So... that sysfs_create_link() call in there should not be unconditional.
It should _only_ ever do that call if list_empty(&dev->subordinate->node)
was true.  IOW, we only create symlinks for subordinate bus that we don't
already know about.

It looks like it was introduced by Greg's "PCI: fix __must_check warnings"
but why a patch advertised _solely_ as a fix is adding additional sysfs
attributes I've no idea.

Something like the following should fix it.  Bahadir - please test.

diff --git a/drivers/pci/bus.c b/drivers/pci/bus.c
index 9e5ea07..132a91f 100644
--- a/drivers/pci/bus.c
+++ b/drivers/pci/bus.c
@@ -131,18 +131,21 @@ void pci_bus_add_devices(struct pci_bus *bus)
 		 * it and then scan for unattached PCI devices.
 		 */
 		if (dev->subordinate) {
-		       if (list_empty(&dev->subordinate->node)) {
+			int new_bus = list_empty(&dev->subordinate->node);
+			if (new_bus) {
 			       down_write(&pci_bus_sem);
 			       list_add_tail(&dev->subordinate->node,
 					       &dev->bus->children);
 			       up_write(&pci_bus_sem);
 			}
 			pci_bus_add_devices(dev->subordinate);
-			retval = sysfs_create_link(&dev->subordinate->class_dev.kobj,
-						   &dev->dev.kobj, "bridge");
-			if (retval)
-				dev_err(&dev->dev, "Error creating sysfs "
-					"bridge symlink, continuing...\n");
+			if (new_bus) {
+				retval = sysfs_create_link(&dev->subordinate->class_dev.kobj,
+							   &dev->dev.kobj, "bridge");
+				if (retval)
+					dev_err(&dev->dev, "Error creating sysfs "
+						"bridge symlink, continuing...\n");
+			}
 		}
 	}
 }

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/