[ovs-dev] [PATCH v2 5/5] dpif-lookup: add avx512 gather implementation

[William Tu]

On Thu, May 21, 2020 at 6:04 AM Van Haaren, Harry
<harry.van.haaren at intel.com> wrote:
>
> > -----Original Message-----
> > From: William Tu <u9012063 at gmail.com>
> > Sent: Wednesday, May 20, 2020 4:15 PM
> > To: Van Haaren, Harry <harry.van.haaren at intel.com>
> > Cc: ovs-dev at openvswitch.org; i.maximets at ovn.org
> > Subject: Re: [ovs-dev] [PATCH v2 5/5] dpif-lookup: add avx512 gather
> > implementation
>
> <snip for required context only>
>
> > > 2020-05-20T14:15:20.184Z|00378|dpif_netdev(pmd-c00/id:9)|INFO|reprobing
> > > sub func, 4 1
> > > 2020-05-20T14:15:21.219Z|00379|dpif_netdev(pmd-c00/id:9)|INFO|reprobing
> > > sub func, 4 1
> > >
> > btw, looking at
> > ovs-appctl coverage/show, this counter is very high when enabling the avx512
> >   handler_duplicate_upcall 459645.4/sec 434475.500/sec
> > 17300.5372/sec   total: 64120526
>
> This counter seems to post some garbage to me if I run it before any traffic?
> Tested using OVS Master @ 48b1c7642 (not including any AVX512 patches):
>
> ./utilities/ovs-appctl coverage/show | grep duplicate_upcall
> 21:handler_duplicate_upcall   0.0/sec     0.000/sec        0.0000/sec   total: 10272710751479363764
>
> # re-runs show different numbers - indicates a garbage-initialized counter perhaps?
> 21:handler_duplicate_upcall   0.0/sec     0.000/sec        0.0000/sec   total: 1049338714623956653
> 21:handler_duplicate_upcall   0.0/sec     0.000/sec        0.0000/sec   total: 18343161283719775679
>

using the same pcap traffic (p0.pcap) on current master, I did not see
the above issue:
datapath_drop_upcall_error  57.4/sec     4.783/sec        0.0797/sec
total: 287
drop_action_of_pipeline  5909696.2/sec 492474.683/sec
8207.9114/sec   total: 52399553

William