On Thu, Aug 16, 2012 at 08:48:25PM -0700, Greg KH wrote:
> On Fri, Aug 17, 2012 at 10:00:46AM +0800, Fengguang Wu wrote:
> > On Sun, Aug 05, 2012 at 09:58:26AM -0700, Greg KH wrote:
> > > On Sun, Aug 05, 2012 at 10:59:38AM +0800, Fengguang Wu wrote:
> > > > Hi all,
> > > > 
> > > > This line triggers an oops in kvm boot test:
> > > > 
> > > > usb_match_id():
> > > > ==>      748         for (; id->idVendor || id->idProduct || id->bDeviceClass ||
> > > >          749                id->bInterfaceClass || id->driver_info; id++) {
> > > >          750                 if (usb_match_one_id(interface, id))
> > > >          751                         return id;
> > > >          752         }
> > > > 
> > > > It's an old bug and happens also in linux 3.0. It's very reproducible
> > > > for the attached config. I can send the initrd (yocto-minimal-i386.cgz)
> > > > on your request in private email.
> > > 
> > > Odds are a driver without a terminating NULL for the device id list is
> > > causing this to fail.
> > > 
> > > What devices are in the system and what drivers are trying to be bound?
> > 
> > The last match is for: drivers/usb/misc/emi62.c
> > 
> > Located down by Tianyu's debug patch:
> > 
> > [    2.206708] usb_device_match: device 1-1:1.0, driver cytherm
> > [    2.207627] usb_device_match: device 1-1:1.0, driver emi62 - firmware loader
> > [    2.208769] BUG: unable to handle kernel paging request at c1f7478e
> > [    2.209726] IP: [<c14ac1c0>] usb_match_id+0x5b/0xcd
> > 
> > > --- a/drivers/usb/core/driver.c
> > > +++ b/drivers/usb/core/driver.c
> > > @@ -778,7 +778,8 @@ static int usb_device_match(struct device *dev, struct device_driver *drv)
> > >
> > >                 intf = to_usb_interface(dev);
> > >                 usb_drv = to_usb_driver(drv);
> > > -
> > > +
> > > +               pr_info("%s: device %s, driver %s \n", dev_name(dev), drv->name);
> > >                 id = usb_match_id(intf, usb_drv->id_table);
> > >                 if (id)
> > >                         return 1;
> 
> Odd that it takes so long after you call that function for it to fail.
> 
> And that driver has a proper termination, so we aren't walking off the
> end of the list, so it must be in the probe function itself.
> 
> Care to add some more debugging for that driver?
> 
> Also, I don't see the dev_info() message from the driver itself, in the
> probe function, so it must not be getting called properly.
> 
> Something weird is happening...

Very true. I find it related to timing: after adding these debug
printks, it becomes very hard to reproduce. And I confirmed that
emi62_probe() is never called.

Anyway, I managed to get one oops. For comparison, one non-failing
dmesg for the same kernel is also attached.

--- tip.orig/drivers/usb/core/driver.c	2012-08-16 09:27:20.063315545 +0800
+++ tip/drivers/usb/core/driver.c	2012-08-17 12:27:32.441761637 +0800
@@ -747,6 +747,7 @@ const struct usb_device_id *usb_match_id
 	   device and interface. */
 	for (; id->idVendor || id->idProduct || id->bDeviceClass ||
 	       id->bInterfaceClass || id->driver_info; id++) {
+		printk(KERN_ERR "usb_match_id: id=%p idVendor=%#x idProduct=%#x\n", id, id->idVendor, id->idProduct);
 		if (usb_match_one_id(interface, id))
 			return id;
 	}
@@ -779,6 +780,7 @@ static int usb_device_match(struct devic
 		intf = to_usb_interface(dev);
 		usb_drv = to_usb_driver(drv);
 
+               pr_info("%s: device %s, driver %s \n", __func__, dev_name(dev), drv->name);
 		id = usb_match_id(intf, usb_drv->id_table);
 		if (id)
 			return 1;
--- tip.orig/drivers/usb/misc/emi62.c	2012-07-29 18:22:35.256996861 +0800
+++ tip/drivers/usb/misc/emi62.c	2012-08-17 12:46:49.761743138 +0800
@@ -242,9 +242,9 @@ MODULE_DEVICE_TABLE (usb, id_table);
 static int emi62_probe(struct usb_interface *intf, const struct usb_device_id *id)
 {
 	struct usb_device *dev = interface_to_usbdev(intf);
-	dev_dbg(&intf->dev, "emi62_probe\n");
+	dev_err(&intf->dev, "emi62_probe\n");
 
-	dev_info(&intf->dev, "%s start\n", __func__);
+	printk(KERN_ERR "%s start\n", __func__);
 
 	emi62_load_firmware(dev);
 
Thanks,
Fengguang