[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+FuTSd9K+s1rXUFpb_RWEC-uAgwU1Vz44zaUPaZK0cfsX4kwA@mail.gmail.com>
Date: Wed, 5 Aug 2020 10:06:41 +0200
From: Willem de Bruijn <willemdebruijn.kernel@...il.com>
To: Eric Dumazet <eric.dumazet@...il.com>
Cc: Colin King <colin.king@...onical.com>,
"David S . Miller" <davem@...emloft.net>,
Shuah Khan <shuah@...nel.org>,
Network Development <netdev@...r.kernel.org>,
"open list:KERNEL SELFTEST FRAMEWORK"
<linux-kselftest@...r.kernel.org>, kernel-janitors@...r.kernel.org,
linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] selftests/net: skip msg_zerocopy test if we have less
than 4 CPUs
On Wed, Aug 5, 2020 at 2:54 AM Eric Dumazet <eric.dumazet@...il.com> wrote:
>
>
>
> On 8/4/20 5:30 AM, Colin King wrote:
> > From: Colin Ian King <colin.king@...onical.com>
> >
> > The current test will exit with a failure if it cannot set affinity on
> > specific CPUs which is problematic when running this on single CPU
> > systems. Add a check for the number of CPUs and skip the test if
> > the CPU requirement is not met.
> >
> > Signed-off-by: Colin Ian King <colin.king@...onical.com>
> > ---
> > tools/testing/selftests/net/msg_zerocopy.sh | 5 +++++
> > 1 file changed, 5 insertions(+)
> >
> > diff --git a/tools/testing/selftests/net/msg_zerocopy.sh b/tools/testing/selftests/net/msg_zerocopy.sh
> > index 825ffec85cea..97bc527e1297 100755
> > --- a/tools/testing/selftests/net/msg_zerocopy.sh
> > +++ b/tools/testing/selftests/net/msg_zerocopy.sh
> > @@ -21,6 +21,11 @@ readonly DADDR6='fd::2'
> >
> > readonly path_sysctl_mem="net.core.optmem_max"
> >
> > +if [[ $(nproc) -lt 4 ]]; then
> > + echo "SKIP: test requires at least 4 CPUs"
> > + exit 4
> > +fi
> > +
> > # No arguments: automated test
> > if [[ "$#" -eq "0" ]]; then
> > $0 4 tcp -t 1
> >
>
> Test explicitly uses CPU 2 and 3, right ?
>
> nproc could be 500, yet cpu 2 or 3 could be offline
>
> # cat /sys/devices/system/cpu/cpu3/online
> 0
> # echo $(nproc)
> 71
The cpu affinity is only set to bring some stability across runs.
The test does not actually verify that a run with zerocopy is some
factor faster than without, as that factor is hard to choose across
all platforms. As a result the automated run mainly gives code coverage.
It's preferable to always run. And on sched_setaffinity failure log a
message about possible jitter and continue. I can send that patch, if
the approach sounds good.
Powered by blists - more mailing lists