Re: [TIP] BUG kmalloc-4096: Poison overwritten (ath5k_rx_skb_alloc)

[Bob Copeland <me@bobcopeland.com>]

On Sun, Jun 28, 2009 at 09:23:29PM +0100, Sitsofe Wheeler wrote:
> OK I'm still trying to follow this. I forwarded your patch to 2.6.31-rc1
> but this time I have kmemleak enabled. I left it streaming radio for the
> past few hours and the following appeared in the logs:
> 
> Jun 28 17:01:34 eeepc kernel: [  744.083787] kmemleak: unreferenced object 0xf5845770 (size 64):
> Jun 28 17:01:34 eeepc kernel: [  744.083795] kmemleak:   comm "swapper", pid 1, jiffies 4294673468
> Jun 28 17:01:34 eeepc kernel: [  744.083799] kmemleak:   backtrace:
> Jun 28 17:01:34 eeepc kernel: [  744.083811] kmemleak:     [<c01a749d>] kmemleak_alloc+0x11d/0x2a0
> Jun 28 17:01:34 eeepc kernel: [  744.083818] kmemleak:     [<c01a4376>] __kmalloc+0x136/0x210
> Jun 28 17:01:34 eeepc kernel: [  744.083828] kmemleak:     [<c043f923>] ieee80211_register_hw+0x83/0x4b0
[snip]


> There were quite a few leaks before this. I dunno if kmemleak is
> spurious or not. As always, any ideas?

I can't seem to boot a kmemleak kernel myself.  But there were a number
of fixes to kmemleak recently to track memory allocated in
ieee80211_register_hw, IIRC -- so it would be good to know if this
persists with all the latest fixes.

As always, no ideas :(  I'm guessing nothing in .31-rc has magically
fixed any problems?  I still say it's some kind of race condition
made worse by the eeepc performance, but unless I can find one or
replicate it with introduced load, I'm afraid I'm stumped.  

(Well, I guess we can always go off and rewrite the rx loop without
the circular terminator and see how well that works on your HW.)

-- 
Bob Copeland %% www.bobcopeland.com