[PATCH] v3: Fix one NULL pointer dereference, add code to debug another one.

[Carl-Daniel Hailfinger]

Fix one NULL pointer dereference, add code to debug another one.

Don't commit yet, I'd rather apply a fix for the NULL pointer
than just debugging it.

Signed-off-by: Carl-Daniel Hailfinger

Index: device/device.c
===================================================================
--- device/device.c	(Revision 630)
+++ device/device.c	(Arbeitskopie)
@@ -277,8 +277,9 @@
 {
    struct device *curdev;

-	printk(BIOS_SPEW, "%s: %s(%s) read_resources bus %d link: %d\n",
-	       __func__, bus->dev->dtsname, dev_path(bus->dev),
+	printk(BIOS_SPEW, "%s: bus->dev %p, bus->dev->dtsname %p, %s(%s) read_resources bus %d link: %d\n",
+	       __func__, bus->dev, bus->dev->dtsname,
+	       bus->dev->dtsname, dev_path(bus->dev),
           bus->secondary, bus->link);

/* Walk through all devices and find which resources they need. */
Index: device/pci_device.c
===================================================================
--- device/pci_device.c	(Revision 630)
+++ device/pci_device.c	(Arbeitskopie)
@@ -1101,7 +1101,7 @@
    	dev = pci_probe_dev(dev, bus, devfn);
    	printk(BIOS_SPEW,
    	       "PCI: pci_scan_bus pci_probe_dev returns dev %p(%s)\n",
-		       dev, dev->dtsname);
+		       dev, dev ? dev->dtsname : "None (not found)");

/* If this is not a multi function device, or the device is
    	 * not present don't waste time probing another function.

[PATCH] v3: Fix two NULL pointer dereferences in device code

[Carl-Daniel Hailfinger]

Ron? You can probably tell if bus->dev==NULL in read_resources(bus)
is an error or if we need to handle it cleanly.

Fix two NULL pointer dereferences in device code.
Add a nasty warning if one of the cases triggers because that should
not happen. We should fix the cases where the warning triggers.

Signed-off-by: Carl-Daniel Hailfinger

Index: LinuxBIOSv3-db800-forexport/device/device.c
===================================================================
--- LinuxBIOSv3-db800-forexport/device/device.c	(Revision 630)
+++ LinuxBIOSv3-db800-forexport/device/device.c	(Arbeitskopie)
@@ -278,8 +278,13 @@
    struct device *curdev;

printk(BIOS_SPEW, "%s: %s(%s) read_resources bus %d link: %d\n",
-	       __func__, bus->dev->dtsname, dev_path(bus->dev),
+	       __func__,
+	       (bus->dev ? bus->dev->dtsname : "No dtsname for NULL device"),
+	       (bus->dev ? dev_path(bus->dev) : "No path for NULL device"),
           bus->secondary, bus->link);
+	if (!bus->dev)
+		printk(BIOS_WARNING, "%s: ERROR: bus->dev is NULL!\n",
+		       __func__);

/* Walk through all devices and find which resources they need. */
    for (curdev = bus->children; curdev; curdev = curdev->sibling) {
Index: LinuxBIOSv3-db800-forexport/device/pci_device.c
===================================================================
--- LinuxBIOSv3-db800-forexport/device/pci_device.c	(Revision 630)
+++ LinuxBIOSv3-db800-forexport/device/pci_device.c	(Arbeitskopie)
@@ -1101,7 +1101,7 @@
    	dev = pci_probe_dev(dev, bus, devfn);
    	printk(BIOS_SPEW,
    	       "PCI: pci_scan_bus pci_probe_dev returns dev %p(%s)\n",
-		       dev, dev->dtsname);
+		       dev, dev ? dev->dtsname : "None (not found)");

/* If this is not a multi function device, or the device is
    	 * not present don't waste time probing another function.

[PATCH] v3: Fix two NULL pointer dereferences in device code

[ron minnich]

Acked-by: Ronald G. Minnich

This was on db800 right? I need to get a failure case.

[PATCH] v3: Fix two NULL pointer dereferences in device code

[Carl-Daniel Hailfinger]

On 05.03.2008 03:54, ron minnich wrote:
> Acked-by: Ronald G. Minnich

Thanks, committed in r631.
> This was on db800 right? I need to get a failure case.

db800, norwich, probably any AMD board triggered the NULL dereference in 
device/device.c. There is still one NULL dereference left in 
device/device.c and I need you to answer my question to fix it 
completely. Then again, you probably asked for a way to reproduce to see 
this. With the patch applied, look for "ERROR: bus->dev is NULL" in the 
logs and notice if that appears, there is garbage in the log later on.
> > Ron? You can probably tell if bus->dev==NULL in read_resources(bus)
> > is an error or if we need to handle it cleanly.

All boards (including qemu) triggered the NULL pointer in 
device/pci_device.c. Look for "None (not found)" in the logs with the 
patch applied. You'll see it 33 times in the qemu boot log.

Regards,
Carl-Daniel