[PATCH] Don't loop endlessly in get_pbus

[Carl-Daniel Hailfinger]

If get_pbus() is called for a device which has no parent/ancestor bus
with nonzero PCI bus operations, get_pbus() will get stuck in a silent
endless loop.
Detect the endless loop and break out with an error message.

Such a situation can happen if the device tree is not yet
initialized/walked completely.

This fixes the unexplainable hang if pci_{read,write}_config{8,16,32}was
used in early mainboard code for the AMD DBM690T. Instead, the code will
now die() with a meaningful error message.

Thanks to Ward Vandewege for testing my patches to track down that bug.

Signed-off-by: Carl-Daniel Hailfinger

Index: LinuxBIOSv2-asus_m2a-vm/src/devices/pci_ops.c
===================================================================
--- LinuxBIOSv2-asus_m2a-vm/src/devices/pci_ops.c	(Revision 3967)
+++ LinuxBIOSv2-asus_m2a-vm/src/devices/pci_ops.c	(Arbeitskopie)
@@ -28,6 +28,12 @@
 {
    struct bus *pbus = dev->bus;
    while(pbus && pbus->dev && !ops_pci_bus(pbus)) {
+		if (pbus == pbus->dev->bus) {
+			printk_alert("%s in endless loop looking for a parent "
+				"bus with ops_pci_bus for %s, breaking out\n",
+				 __func__, dev_path(dev));
+			break;
+		}
    	pbus = pbus->dev->bus;
    }
    if (!pbus || !pbus->dev || !pbus->dev->ops || !pbus->dev->ops->ops_pci_bus) {

[Marc Jones]

On Tue, Mar 3, 2009 at 4:48 PM, Carl-Daniel Hailfinger
 wrote:
> If get_pbus() is called for a device which has no parent/ancestor bus
> with nonzero PCI bus operations, get_pbus() will get stuck in a silent
> endless loop.
> Detect the endless loop and break out with an error message.
>
> Such a situation can happen if the device tree is not yet
> initialized/walked completely.
>
> This fixes the unexplainable hang if pci_{read,write}_config{8,16,32}was
> used in early mainboard code for the AMD DBM690T. Instead, the code will
> now die() with a meaningful error message.
>
> Thanks to Ward Vandewege for testing my patches to track down that bug.
>
> Signed-off-by: Carl-Daniel Hailfinger
> Index: LinuxBIOSv2-asus_m2a-vm/src/devices/pci_ops.c
>
> --- LinuxBIOSv2-asus_m2a-vm/src/devices/pci_ops.c       (Revision 3967)
> +++ LinuxBIOSv2-asus_m2a-vm/src/devices/pci_ops.c       (Arbeitskopie)
> @@ -28,6 +28,12 @@
>  {
>        struct bus *pbus = dev->bus;
>        while(pbus && pbus->dev && !ops_pci_bus(pbus)) {
> if (pbus == pbus->dev->bus) {
> printk_alert("%s in endless loop looking for a parent "
> "bus with ops_pci_bus for %s, breaking out\n",
> __func__, dev_path(dev));
> break;
> }
>
> pbus = pbus->dev->bus;
>        }
>        if (!pbus || !pbus->dev || !pbus->dev->ops || !pbus->dev->ops->ops_pci_bus) {

Acked-by: Marc Jones

[Carl-Daniel Hailfinger]

On 05.03.2009 16:31, Marc Jones wrote:
> On Tue, Mar 3, 2009 at 4:48 PM, Carl-Daniel Hailfinger
>  wrote:
> > If get_pbus() is called for a device which has no parent/ancestor bus
> > with nonzero PCI bus operations, get_pbus() will get stuck in a silent
> > endless loop.
> > Detect the endless loop and break out with an error message.
> >
> > Such a situation can happen if the device tree is not yet
> > initialized/walked completely.
> >
> > This fixes the unexplainable hang if pci_{read,write}_config{8,16,32}was
> > used in early mainboard code for the AMD DBM690T. Instead, the code will
> > now die() with a meaningful error message.
> >
> > Thanks to Ward Vandewege for testing my patches to track down that bug.
> >
> > Signed-off-by: Carl-Daniel Hailfinger
>
> Acked-by: Marc Jones

Thanks, r3972.

Regards,
Carl-Daniel