Re: [PATCH iproute2] htb: report overhead attribute

[Jesper Dangaard Brouer <jbrouer@redhat.com>]

On Tue, 04 Jun 2013 06:58:31 -0700
Eric Dumazet <eric.dumazet@gmail.com> wrote:

> On Tue, 2013-06-04 at 13:11 +0200, Jesper Dangaard Brouer wrote:
> > On Mon, 03 Jun 2013 08:56:02 -0700
> > Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > 
> > > On Mon, 2013-06-03 at 08:45 -0700, Rick Jones wrote:
> > > 
> > > > Is it (still) possible to have a negative overhead?
> > > > 
> > > > http://www.linksysinfo.org/index.php?threads/speedmod-with-tc-atm-qos-patch-for-adsl.31541/
> > > 
> > > overhead always has been unsigned in the kernel.
> > > 
> > > What you describe is a userland hack in tc command.
> > > (or a bug)
> > 
> > Rick is referencing Russell Stuart's patches, where a negative
> > overhead was possible.
> >  http://ace-host.stuart.id.au/russell/files/tc/tc-atm/#history
> > 
> > But my patches got accepted into the kernel, where a negative
> > overhead was not possible. In retrospect, we should have supported a
> > negative overhead.
> > 
> > A negative overhead *is* a valid use-case, and we should work
> > towards supporting this. E.g. by changing the recent added "u16
> > overhead" in struct psched_ratecfg to be "s16" (ref [1]) ?
(My statement should have been corrected to "s16"->"s32", never mind)

> > 
> > [1] commit 01cb71d2d47 (net_sched: restore "overhead xxx" handling)
> >  https://git.kernel.org/cgit/linux/kernel/git/davem/net.git/commit/?id=01cb71d2d47b78354358e4bb938bb06323e17498
> > 
> 
> Again, you describe something that Vimal patch didn't broke, and
> should be addressed on net-next.

Yes, I know, I also said "work towards supporting this", meaning
"net-next".


> My concern was restoring the overhead attribute that Vimal broke, and
> this one was unsigned 16bits.
> 
> Allowing a negative offset is not free, it adds a conditional test,
> because (len + overhead) could be negative.

Yes, I do realize that. But Vimal patch actually also broke the
"mpu" (Minimum Packet Unit) feature.  And we could combine this, and
get negative offset fix by a max(mpu,len), where mpu would be default
init'ed. 

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Sr. Network Kernel Developer at Red Hat
  Author of http://www.iptv-analyzer.org
  LinkedIn: http://www.linkedin.com/in/brouer