Re: [Xen-devel] why we need call xennet_alloc_rx_buffers() in xennet_poll() in Netfront module?

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Wed, Jun 19, 2013 at 11:05:40AM -0700, Deep Debroy wrote:
> On Mon, Jan 7, 2013 at 8:31 AM, Konrad Rzeszutek Wilk
> <konrad.wilk@xxxxxxxxxx> wrote:
> >
> > On Tue, Nov 27, 2012 at 11:42:11PM -0500, David Xu wrote:
> > > Hi all,
> > >
> > > Why we call xennet_alloc_rx_buffers function in xennet_poll of netfront?
> >
> > That is an easy answer - we need to allocate receive buffers :-)
> > >
> > > When I run the iperf benchmark to measure the TCP throughput between a
> > > physical machine and a VM, the TCP server which is a Xen VM crashed.
> > >
> > > I traced the source code and found that the error is from  xennet_poll
> > > =>  xennet_alloc_rx_buffers => __skb_dequeue => __skb_unlink
> >
> >
> > Hm, that is rather strange that you would hit this. Is this problem
> > still present with the latest released kernel?
> 
> Hi, I am seeing a crash in xennet_poll => xennet_alloc_rx_buffers in
> an old 2.6.27 kernel under somewhat similar circumstances (iperf on
> domu). Wondering if there was any specific root-cause determined for
> the crash reported below or if there are any specific patches that is
> expected to fix this in latest kernels that I can look at for
> back-porting.
> 

I skimmed the log and didn't see interesting commit logs.

2.6.27 is pretty old and I think you're pretty much on your own to look
through the git tree I'm afraid.

The rune I used:
  git log v2.6.27-rc9..HEAD -- drivers/net/xen-netfront.c


Wei.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel