[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <b8f675495a2c16cd742d1a95b3421ed15643aad2.camel@redhat.com>
Date: Wed, 31 Jan 2024 19:39:41 +0100
From: Paolo Abeni <pabeni@...hat.com>
To: Willem de Bruijn <willemdebruijn.kernel@...il.com>, Jakub Kicinski
<kuba@...nel.org>
Cc: netdev@...r.kernel.org, davem@...emloft.net, edumazet@...gle.com,
linux-kselftest@...r.kernel.org, Willem de Bruijn <willemb@...gle.com>
Subject: Re: [PATCH net-next] selftests/net: calibrate txtimestamp
On Wed, 2024-01-31 at 10:06 -0500, Willem de Bruijn wrote:
> Otherwise I'll start with the gro and so-txtime tests. They may not
> be so easily calibrated. As we cannot control the gro timeout, nor
> the FQ max horizon.
Note that we can control the GRO timeout to some extent, via
gro_flush_timeout, see commit 89abe628375301fedb68770644df845d49018d8b.
Unfortunately that is not enough for 'large' gro tests. I think the
root cause is that the process sending the packets can be de-scheduled
- even the qemu VM from the hypervisor CPU - causing an extremely large
gap between consecutive pkts.
I guess/hope that replacing multiple sendmsg() with a sendmmsg() could
improve a bit the scenario, but I fear it will not solve the issue
completely.
> In such cases we can use the environment variable to either skip the
> test entirely or --my preference-- run it to get code coverage, but
> suppress a failure if due to timing (only). Sounds good?
Sounds good to me! I was wondering about skipping the 'large' test
only, but suppressing the failure when KSFT_MACHINE_SLOW=yes only for
such test looks a better option.
Thanks!
Paolo
Powered by blists - more mailing lists