svn commit: r189117 - head/sys/dev/bce

[Dmitry Sivachenko]

On Wed, Aug 12, 2009 at 03:57:26PM -0700, Pyun YongHyeon wrote:
> On Wed, Aug 12, 2009 at 02:44:29PM -0700, Xin LI wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> > 
> > Hi, Stanislav,
> > 
> > Stanislav Sedov wrote:
> > > On Thu, 06 Aug 2009 11:07:32 -0500
> > > Tom Judge <tom at tomjudge.com> mentioned:
> > > 
> > >> I can confirm that we are also seeing this problem on a number of Dell 
> > >> PE2950 systems running the bce driver from 7.2, the bug is more 
> > >> visible/producible on systems that are using Jumbo Frames (in our case 8Kb).
> > >>
> > >> The symptoms of this are an increase in the requests for jumbo clusters 
> > >> denied counter at 9k, and then random network stalls which sometimes 
> > >> recover on there own and others require a reboot.
> > >>
> > >> We have backed this change and so far it seems to have fixed the issue.
> > >>
> > > 
> > > David,
> > > 
> > > are you OK with reverting this patch?  I submitted the patch to re@ for
> > > review, and I can commit it if you don't mind doing so. :-)
> > 
> > I think there was some motivation supporting the SPLIT_HEADER disable in
> > the past (IIRC it was for reliability or data corruption), but I think I
> > have lost the mail archive (at least I couldn't find it at this moment
> > :-/  I'm not very sure if that was fixed by my busdma_sync fix, though,
> > can anyone confirm that?
> > 
> 
> Since I don't have bce(4) hardwares I can't confirm it. I think the
> busdma_sync fix would be prerequisite condition to fix data
> corruption problem but I think it would not necessarily fix
> previous data corruption issue.


I wish you reach some consensus there because bce(4) is indeed
broken and stas's patch seems to make it better.