Messages in this thread | | | Date | Wed, 11 Dec 2013 10:30:38 +0100 | From | Sebastian Andrzej Siewior <> | Subject | Re: [PATCH v2] debugobject: add support for kref |
| |
On 12/11/2013 07:45 AM, Greg KH wrote: > I need an ack from Thomas or some other debugobjects-knowledgable > developer before I can take this...
I have a v3 which I haven't posted yet. Basically the whole thing seems to work. "Seems" means that I get a bunch of backtraces which I haven't checked to see if it is a kref-accounting bug (something that should be fixed) or something that is correct but kref isn't used properly.
One example is serial port init via tty_port_init(): It initializes &port->kref. To use that kref you need to use tty_port_get()/put() and more importantly dynamically allocate tty_port _or_ provide port->ops->destruct callback. This seems not be done by everyone, for instance gs_port_alloc() [0]. The question is this a bug and should be fixed (and kref's debugobject is right here) or it is "legal" behavior.
I get another backtrace on "rmmod sg" which shows ODEBUG: free active (active state 0) object type: kref hint: kset_init+0xe/0x30 [<c112fec8>] kfree+0x128/0x280 [<c13784c2>] ? class_release+0x22/0x30 [<c13784c2>] class_release+0x22/0x30 [<c12da3bd>] kobject_cleanup+0x2d/0x70 [<c12da2c7>] kobject_put+0x67/0xa0 [<c12da34f>] kset_unregister+0xf/0x20 [<c13787b3>] class_destroy+0x23/0x30 [<f850ee62>] exit_sg+0x17/0x3c [sg]
so it is triggered by class_destory() and I haven't have the time check what is wrong here.
Based on this I would like to know what could be the best way forward here. Post v3 and work on this backtraces before it gets merged or after it gets merged.
[0] drivers/usb/gadget/u_serial.c
> thanks, > > greg k-h > Sebastian
| |