From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paul Durrant Subject: Re: xennet: skb rides the rocket: 20 slots Date: Tue, 8 Jan 2013 10:16:29 +0000 Message-ID: <291EDFCB1E9E224A99088639C4762022012CCFEB80D7@LONPMAILBOX01.citrite.net> References: <72958707.20130104172854@eikelenboom.it> <1357556115.7989.13.camel@zakaz.uk.xensource.com> <50EB8091.90705@oracle.com> <1357639549.7989.147.camel@zakaz.uk.xensource.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1357639549.7989.147.camel@zakaz.uk.xensource.com> Content-Language: en-US List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Ian Campbell , ANNIE LI Cc: Sander Eikelenboom , xen-devel , Konrad Rzeszutek Wilk List-Id: xen-devel@lists.xenproject.org > -----Original Message----- > From: xen-devel-bounces@lists.xen.org [mailto:xen-devel- > bounces@lists.xen.org] On Behalf Of Ian Campbell > Sent: 08 January 2013 10:06 > To: ANNIE LI > Cc: Sander Eikelenboom; Konrad Rzeszutek Wilk; xen-devel > Subject: Re: [Xen-devel] xennet: skb rides the rocket: 20 slots > > On Tue, 2013-01-08 at 02:12 +0000, ANNIE LI wrote: > > > >> > > >> I have added some extra info, but i don't have enough knowledge if > > >> this could/should be prevented from happening ? > > > MAX_SKB_FRAGS has never, AFAIK, been as big as 19 or 20 so I'm not > > > sure this can have ever worked. > > > > > > These SKBs seem to be pretty big (not quite 64KB), seemingly most of > > > the data is contained in a smallish number of frags, which suggests > > > compound pages and therefore a reasonably modern kernel? > > > > > > This probably relates somewhat to the issues described in the > > > "netchannel vs MAX_SKB_FRAGS" thread last year, or at least the > > > solution to that would necessarily involve fixing this issue. > > > > If netback complains about "Too many frags", then it should be > > MAX_SKB_FRAGS limitation in netback results in dropping packets in > > netfront. It is possible that other netfronts(windows?) also hit this. > > It's very possible. I rather suspect that non-Linux frontends have > workarounds (e.g. manual resegmenting etc) for this case. > Citrix Windows PV drivers certainly do; packets will be copied if they have more than MAX_SKB_FRAGS fragments. We also choose a maximal TSO size carefully to try to avoid Windows' network stack sending down more than MAX_SKB_FRAGS fragments in the first place. Paul