Re: [Bugme-new] [Bug 38102] New: BUG kmalloc-2048: Poison overwritten

[Eric Dumazet <eric.dumazet@gmail.com>]

Le lundi 04 juillet 2011 à 13:05 +0000, Michael Büsch a écrit :
> On Mon, 4 Jul 2011 15:48:31 +0400
> Alexey Zaytsev <alexey.zaytsev@gmail.com> wrote:
> > The skb is reinserted into the ring before its data is copied, it
> > seems. But this can't be the cause of my problem, as it would lead to
> > data corruption at most, not a write-after-free.
> 
> Recycling the skb does not imply that the device can reuse it immediately. The device is told at the very end of the RX function (after the loop) that it's now safe to put stuff into the recyceled/new buffers.
> 
> > And an other question. Why so we have the logic to work-around the 1Gb
> > DMA limit instead of just setting the dma mask?
> 
> Because the DMA mask does not work correctly on all arches for masks smaller than 4G.
> 
> And btw, I dont understand what that wmb() patch is supposed to fix. There may be a wmb() missing, but rather after the ctrl _and_ the address assignment to the descriptor.
> But I don't think this can cause this use-after-free anyway.
> 

I dont have the b44 specs, but :

For sure, addr should be set before ctl, just in case ctl allows chip to
start a dma transfert (to previous packet), because a OWN bit is unset
for example...

A second wmb() is not necessary.
It will be done eventually at next packet (we have a ring of 200
packets)