Re: [BUG 2.6.17-git] kmem_cache_create: duplicate cache scsi_cmd_cache

[Greg KH <gregkh@suse.de>]

On Fri, May 12, 2006 at 03:37:59PM -0700, Linus Torvalds wrote:
> 
> 
> On Fri, 12 May 2006, Linus Torvalds wrote:
> > 
> > On Fri, 12 May 2006, Russell King wrote:
> > > 
> > > From: Todd Blumer <todd@sdgsystems.com>
> > > On a PXA27x handheld (iPAQ hx4700), when we eject a mounted SD memory
> > > card, we get a kernel panic (kernel trying to clean up non-existent
> > > device). One hack patch to avoid the panic is:
> > > 
> > > --- fs/partitions/check.c       10 Apr 2006 22:57:27 -0000      1.15
> > > +++ fs/partitions/check.c       4 May 2006 20:30:15 -0000
> > > @@ -491,6 +491,7 @@
> > >                         kfree(disk_name);
> > >                 }
> > >                 put_device(disk->driverfs_dev);
> > > +               disk->driverfs_dev = 0; /* HACK - what's the right solution? */
> > >         }
> > >         kobject_uevent(&disk->kobj, KOBJ_REMOVE);
> > >         kobject_del(&disk->kobj);
> > 
> > Btw, on the face it of, I really think that this patch is correct 
> > regardless of any other issues.
> 
> .. and I suspect it also shows what the "other issues" are.
> 
> Shouldn't that KOBJ_REMOVE uevent happen _before_ we do all the freeing of 
> the backing dev object? That KOBJ_REMOVE thing actually seems to want to 
> report the pathname for the disk it removes. Preferably before the thing 
> is gone and can't be reported on..
> 
> Ie shouldn't the diff be something like this?

It looks sane to me.  Russell, does it solve your oops too?

thanks,

greg k-h