[ovs-dev] Patch adding additional QoS strategies to OVS

[Ben Pfaff]

On Tue, Mar 10, 2015 at 10:07:31PM +0100, Jonathan Vestin wrote:
> Ok, quite some things to do, I will fix the things you mentioned after 
> sorting out the tc_load problem.
> 
> Regarding tc_load:
> See the comment in: tc_parse_qdisc at 4414 to 4420:
> * To avoid the OOPS, we must not make a request that would attempt to dump
> * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
> * few others. There are a few ways that I can see to do this, but most of
> * them seem to be racy (and if you lose the race the kernel OOPSes). The
> * technique chosen here is to assume that any non-default qdisc that we
> * create will have a class with handle 1:0. The built-in qdiscs only have
> * a class with handle 0:0.
> 
> I presume this means that in order for tc_load to be called, the qdisc 
> *must* have one class, which classless qdiscs cannot have. Please correct me 
> if I'm wrong.

That's a good point, thanks for pointing it out.

I sent out a possible fix.  Would you mind reviewing it and making
sure that it actually fixes the problem for you?  The patch is at:
        http://openvswitch.org/pipermail/dev/2015-March/052268.html