Oops in make_class_name in 2.6.22.1 on Fedora

Oops in make_class_name in 2.6.22.1 on Fedora

[Pete Zaitcev]

Hi, Greg (and others):

I do not seem to be able to find an answer, sorry.
Do you happen to remember if this was fixed after 2.6.22.1:

localhost kernel: EIP is at make_class_name+0x27/0x7a
localhost kernel:  [<c05586f1>] class_device_del+0x97/0x119
localhost kernel:  [<c055877b>] class_device_unregister+0x8/0x10
localhost kernel:  [<e08735f4>] __scsi_remove_device+0x1d/0x60 [scsi_mod]
localhost kernel:  [<e08711ae>] scsi_forget_host+0x2d/0x4a [scsi_mod]
localhost kernel:  [<e086c49c>] scsi_remove_host+0x65/0xd5 [scsi_mod]
localhost kernel:  [<e0e76775>] storage_disconnect+0xe/0x16 [usb_storage]
localhost kernel:  [<c056f8b8>] usb_unbind_interface+0x44/0x85
localhost kernel:  [<c0557df7>] __device_release_driver+0x6e/0x8b

Obviously a known bug but all I see is users reporting it. In my case
it's this:
 https://bugzilla.redhat.com/show_bug.cgi?id=253424
I saw Alan giving it a try here:
 http://lkml.org/lkml/2007/7/12/259

Yours,
-- Pete

Re: Oops in make_class_name in 2.6.22.1 on Fedora

[Greg KH]

On Mon, Sep 10, 2007 at 08:26:47PM -0700, Pete Zaitcev wrote:
> Hi, Greg (and others):
> 
> I do not seem to be able to find an answer, sorry.
> Do you happen to remember if this was fixed after 2.6.22.1:
> 
> localhost kernel: EIP is at make_class_name+0x27/0x7a
> localhost kernel:  [<c05586f1>] class_device_del+0x97/0x119
> localhost kernel:  [<c055877b>] class_device_unregister+0x8/0x10
> localhost kernel:  [<e08735f4>] __scsi_remove_device+0x1d/0x60 [scsi_mod]
> localhost kernel:  [<e08711ae>] scsi_forget_host+0x2d/0x4a [scsi_mod]
> localhost kernel:  [<e086c49c>] scsi_remove_host+0x65/0xd5 [scsi_mod]
> localhost kernel:  [<e0e76775>] storage_disconnect+0xe/0x16 [usb_storage]
> localhost kernel:  [<c056f8b8>] usb_unbind_interface+0x44/0x85
> localhost kernel:  [<c0557df7>] __device_release_driver+0x6e/0x8b
> 
> Obviously a known bug but all I see is users reporting it. In my case
> it's this:
>  https://bugzilla.redhat.com/show_bug.cgi?id=253424
> I saw Alan giving it a try here:
>  http://lkml.org/lkml/2007/7/12/259

I think this was a scsi problem that has been fixed, but don't really
remember the exact commit.

Alan, any ideas?

It would probably be good to get this into the next -stable if we can
figure out which patch did fix it.

thanks,

greg k-h

Re: Oops in make_class_name in 2.6.22.1 on Fedora

[Cornelia Huck]

On Mon, 10 Sep 2007 23:12:26 -0700,
Greg KH <greg@kroah.com> wrote:

> On Mon, Sep 10, 2007 at 08:26:47PM -0700, Pete Zaitcev wrote:
> > Hi, Greg (and others):
> > 
> > I do not seem to be able to find an answer, sorry.
> > Do you happen to remember if this was fixed after 2.6.22.1:
> > 
> > localhost kernel: EIP is at make_class_name+0x27/0x7a
> > localhost kernel:  [<c05586f1>] class_device_del+0x97/0x119
> > localhost kernel:  [<c055877b>] class_device_unregister+0x8/0x10
> > localhost kernel:  [<e08735f4>] __scsi_remove_device+0x1d/0x60 [scsi_mod]
> > localhost kernel:  [<e08711ae>] scsi_forget_host+0x2d/0x4a [scsi_mod]
> > localhost kernel:  [<e086c49c>] scsi_remove_host+0x65/0xd5 [scsi_mod]
> > localhost kernel:  [<e0e76775>] storage_disconnect+0xe/0x16 [usb_storage]
> > localhost kernel:  [<c056f8b8>] usb_unbind_interface+0x44/0x85
> > localhost kernel:  [<c0557df7>] __device_release_driver+0x6e/0x8b
> > 
> > Obviously a known bug but all I see is users reporting it. In my case
> > it's this:
> >  https://bugzilla.redhat.com/show_bug.cgi?id=253424
> > I saw Alan giving it a try here:
> >  http://lkml.org/lkml/2007/7/12/259
> 
> I think this was a scsi problem that has been fixed, but don't really
> remember the exact commit.

It may be something like "removing a device before it has been
completely initialized" (class_device_unregister() racing against
class_device_register()). (Perhaps you can see if that is the case by
turning on verbose driver core and kobject debugging messages.)

Re: Oops in make_class_name in 2.6.22.1 on Fedora

[Alan Stern]

On Mon, 10 Sep 2007, Greg KH wrote:

> On Mon, Sep 10, 2007 at 08:26:47PM -0700, Pete Zaitcev wrote:
> > Hi, Greg (and others):
> > 
> > I do not seem to be able to find an answer, sorry.
> > Do you happen to remember if this was fixed after 2.6.22.1:
> > 
> > localhost kernel: EIP is at make_class_name+0x27/0x7a
> > localhost kernel:  [<c05586f1>] class_device_del+0x97/0x119
> > localhost kernel:  [<c055877b>] class_device_unregister+0x8/0x10
> > localhost kernel:  [<e08735f4>] __scsi_remove_device+0x1d/0x60 [scsi_mod]
> > localhost kernel:  [<e08711ae>] scsi_forget_host+0x2d/0x4a [scsi_mod]
> > localhost kernel:  [<e086c49c>] scsi_remove_host+0x65/0xd5 [scsi_mod]
> > localhost kernel:  [<e0e76775>] storage_disconnect+0xe/0x16 [usb_storage]
> > localhost kernel:  [<c056f8b8>] usb_unbind_interface+0x44/0x85
> > localhost kernel:  [<c0557df7>] __device_release_driver+0x6e/0x8b
> > 
> > Obviously a known bug but all I see is users reporting it. In my case
> > it's this:
> >  https://bugzilla.redhat.com/show_bug.cgi?id=253424
> > I saw Alan giving it a try here:
> >  http://lkml.org/lkml/2007/7/12/259
> 
> I think this was a scsi problem that has been fixed, but don't really
> remember the exact commit.

Yes.  There are some bugs in the SCSI async-scanning code, and there is
a patch to fix them:

http://git.kernel.org/?p=linux/kernel/git/jejb/scsi-misc-2.6.git;a=commit;h=a93a091df8232fad60867d41fbc3be855a0b78f2

It is scheduled for 2.6.24.  For now, people can work around the bug by 
disabling CONFIG_SCSI_SCAN_ASYNC.

> Alan, any ideas?
> 
> It would probably be good to get this into the next -stable if we can
> figure out which patch did fix it.

I have urged both James Bottomley and Andrew Morton to apply that patch 
as soon as possible.  Neither of them paid any attention.

Alan Stern

Re: Oops in make_class_name in 2.6.22.1 on Fedora

[Chuck Ebbert]

On 09/11/2007 11:11 AM, Alan Stern wrote:
>>>
>>> I do not seem to be able to find an answer, sorry.
>>> Do you happen to remember if this was fixed after 2.6.22.1:
>>>
>>> localhost kernel: EIP is at make_class_name+0x27/0x7a
>>> localhost kernel:  [<c05586f1>] class_device_del+0x97/0x119
>>> localhost kernel:  [<c055877b>] class_device_unregister+0x8/0x10
>>> localhost kernel:  [<e08735f4>] __scsi_remove_device+0x1d/0x60 [scsi_mod]
>>> localhost kernel:  [<e08711ae>] scsi_forget_host+0x2d/0x4a [scsi_mod]
>>> localhost kernel:  [<e086c49c>] scsi_remove_host+0x65/0xd5 [scsi_mod]
>>> localhost kernel:  [<e0e76775>] storage_disconnect+0xe/0x16 [usb_storage]
>>> localhost kernel:  [<c056f8b8>] usb_unbind_interface+0x44/0x85
>>> localhost kernel:  [<c0557df7>] __device_release_driver+0x6e/0x8b
>>>
>>> Obviously a known bug but all I see is users reporting it. In my case
>>> it's this:
>>>  https://bugzilla.redhat.com/show_bug.cgi?id=253424
>>> I saw Alan giving it a try here:
>>>  http://lkml.org/lkml/2007/7/12/259
>> I think this was a scsi problem that has been fixed, but don't really
>> remember the exact commit.
> 
> Yes.  There are some bugs in the SCSI async-scanning code, and there is
> a patch to fix them:
> 
> http://git.kernel.org/?p=linux/kernel/git/jejb/scsi-misc-2.6.git;a=commit;h=a93a091df8232fad60867d41fbc3be855a0b78f2
> 
> It is scheduled for 2.6.24.  For now, people can work around the bug by 
> disabling CONFIG_SCSI_SCAN_ASYNC.
> 

The bug still happens with that patch applied and CONFIG_SCSI_SCAN_ASYNC
enabled. In the marked line, class_dev->class->name is NULL and this causes
the oops in make_class_name().


static void remove_deprecated_class_device_links(struct class_device *class_dev)
{
        char *class_name;

        if (!class_dev->dev)
                return;

===>>   class_name = make_class_name(class_dev->class->name, &class_dev->kobj);
        if (class_name)
                sysfs_remove_link(&class_dev->dev->kobj, class_name);
        kfree(class_name);
}

Re: Oops in make_class_name in 2.6.22.1 on Fedora

[Alan Stern]

On Mon, 22 Oct 2007, Chuck Ebbert wrote:

> On 09/11/2007 11:11 AM, Alan Stern wrote:
> >>>
> >>> I do not seem to be able to find an answer, sorry.
> >>> Do you happen to remember if this was fixed after 2.6.22.1:
> >>>
> >>> localhost kernel: EIP is at make_class_name+0x27/0x7a
> >>> localhost kernel:  [<c05586f1>] class_device_del+0x97/0x119
> >>> localhost kernel:  [<c055877b>] class_device_unregister+0x8/0x10
> >>> localhost kernel:  [<e08735f4>] __scsi_remove_device+0x1d/0x60 [scsi_mod]
> >>> localhost kernel:  [<e08711ae>] scsi_forget_host+0x2d/0x4a [scsi_mod]
> >>> localhost kernel:  [<e086c49c>] scsi_remove_host+0x65/0xd5 [scsi_mod]
> >>> localhost kernel:  [<e0e76775>] storage_disconnect+0xe/0x16 [usb_storage]
> >>> localhost kernel:  [<c056f8b8>] usb_unbind_interface+0x44/0x85
> >>> localhost kernel:  [<c0557df7>] __device_release_driver+0x6e/0x8b
> >>>
> >>> Obviously a known bug but all I see is users reporting it. In my case
> >>> it's this:
> >>>  https://bugzilla.redhat.com/show_bug.cgi?id=253424
> >>> I saw Alan giving it a try here:
> >>>  http://lkml.org/lkml/2007/7/12/259
> >> I think this was a scsi problem that has been fixed, but don't really
> >> remember the exact commit.
> > 
> > Yes.  There are some bugs in the SCSI async-scanning code, and there is
> > a patch to fix them:
> > 
> > http://git.kernel.org/?p=linux/kernel/git/jejb/scsi-misc-2.6.git;a=commit;h=a93a091df8232fad60867d41fbc3be855a0b78f2
> > 
> > It is scheduled for 2.6.24.  For now, people can work around the bug by 
> > disabling CONFIG_SCSI_SCAN_ASYNC.
> > 
> 
> The bug still happens with that patch applied and CONFIG_SCSI_SCAN_ASYNC
> enabled. In the marked line, class_dev->class->name is NULL and this causes
> the oops in make_class_name().

Can anybody provide a dmesg log showing the error on a system with 
CONFIG_USB_DEBUG and CONFIG_USB_STORAGE_DEBUG enabled?

Does it still occur with 2.6.24-rc1?

Alan Stern