device driver memory leak in 5.1-20030726?

[Scott Long]

John-Mark Gurney wrote:

> Lukas Ertl wrote this message on Sun, Jul 27, 2003 at 16:43 +0200:
> 
>>On Sun, 27 Jul 2003, Mark Blackman wrote:
>>
>>
>>>Perhaps it's a USB bug. There seems to be some correspondence between
>>>the use of the USB Speedtouch ADSL modem and the out-of-control
>>>devbuf allocations.
>>
>>I'm too seeing these annoying kmem_malloc panics on recent -current
>>kernels. The laptop I'm using is way off of being overloaded at all, the
>>only thing I do is going online using a Bluetooth USB dongle. As soon as I
>>generate some network traffic, devbuf allocations go up, until at some
>>point the machine panics randomly in kmem_malloc.
> 
> 
> I must note that the USB changes only allocates memory in the M_USB
> area which is described by:
> usb.c:MALLOC_DEFINE(M_USB, "USB", "USB");
> 
> So, that means it wouldn't be in the devbuf area.  (This is the one of
> the points of malloc areas is to help track down stray allocations and
> memory leaks).
> 
> 
>>I have different core dumps and backtraces available, but they don't seem
>>to be of much use in this case. I really suspect the USB stuff to be
>>leaking.
> 
> 
> It may be leaking, but it won't be leaking devbuf memory.  The only
> thing that is in usb (in dev/usb) that uses M_DEVBUF is ukbd.
> 

bus_dma_tag_create() allocates out of M_DEVBUF.  Could it be that tags
are being created and never destroyed?

Scott