Re: 2.6.10-rc1-mm3

From: Greg KH
Date: Tue Nov 09 2004 - 03:12:20 EST


On Mon, Nov 08, 2004 at 11:14:55PM -0800, Greg KH wrote:
> On Mon, Nov 08, 2004 at 09:27:47PM -0800, Andrew Morton wrote:
> > Greg KH <greg@xxxxxxxxx> wrote:
> > >
> > > So I don't see how that could be failing here. And why I don't see this
> > > on my boxes...
> >
> > OK, progress. The oops is due to CONFIG_LEGACY_PTY_COUNT=512. I assume
> > anything greater than 256 will trigger it.
> >
> > - tty_register_driver() calls tty_register_device() for 512 devices.
> >
> > - tty_register_device() calls pty_line_name() for the 512 devices, but
> > pty_line_name() only understands 256 devices. After that, it starts
> > returning duplicated names.
> >
> > - class_simple_device_add() gets an -EEXIST return from
> > class_device_register() and then tries to kfree local variable s_dev, but
> > it's already free. Presumably all that icky refcounting under
> > class_device_register() did this for us already. Can you fix this one
> > Greg? Just enable slab debugging, set CONFIG_LEGACY_PTY_COUNT=512 and
> > watch the fun.
>
> Ick, yeah, I just tested that. I don't know why that's happening, I'll
> go fix it up now.

Ah, found it. Was caused by a patch from Tejun Heo <tj@xxxxxxxxxxx>
that went into the tree in my last round of driver core changes.

Tejun, the call to unlink() in the error path in kobject_add() does a
kobject_put(). Your patch added an extra kobject_put() which caused bad
things to happen when we failed.

Andrew, does the patch below fix the issue for you? It fixed my test
case.

thanks,

greg k-h

Subject: Remove extra kobject_put() in kobject_add() error path.

Also document the thing so no one tries to "fix" it again...

Signed-off-by: Greg Kroah-Hartman <greg@xxxxxxxxx>


--- a/lib/kobject.c 2004-11-05 10:06:33 -08:00
+++ b/lib/kobject.c 2004-11-08 23:58:02 -08:00
@@ -181,10 +181,10 @@ int kobject_add(struct kobject * kobj)

error = create_dir(kobj);
if (error) {
+ /* Does the kobject_put() for us */
unlink(kobj);
if (parent)
kobject_put(parent);
- kobject_put(kobj);
} else {
kobject_hotplug(kobj, KOBJ_ADD);
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/