ACK: [ACT][PATCH 1/1] UBUNTU: SAUCE: ubuntu_kernel_selftests: Bump optmem_max for net bench tests
Tim Gardner
tim.gardner at canonical.com
Fri Feb 3 18:37:18 UTC 2023
On 2/3/23 11:14 AM, Andrei Gherzan wrote:
> BugLink: https://bugs.launchpad.net/ubuntu-kernel-tests/+bug/1960907
>
> One of the reasons why a zerocopy send operation could fail with ENOBUF
> is when the optmem limit is reached[1]. The default value is currently
> set at 20480 which is insufficient for edge cases especially around
> benchmarking tests.
>
> This change proposes an optmem_max increase to 2048000. It fixes for
> example "sendmesg: No buffer space available" sporadic errors in
> udpgso_bench_tx/udpgso_bench.sh
>
> [1] https://www.kernel.org/doc/html/latest/networking/msg_zerocopy.html#transmission
>
> Signed-off-by: Andrei Gherzan <andrei.gherzan at canonical.com>
> ---
> ubuntu_kernel_selftests/ubuntu_kernel_selftests.py | 12 ++++++++++--
> 1 file changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/ubuntu_kernel_selftests/ubuntu_kernel_selftests.py b/ubuntu_kernel_selftests/ubuntu_kernel_selftests.py
> index 5d87f82e..6335abac 100644
> --- a/ubuntu_kernel_selftests/ubuntu_kernel_selftests.py
> +++ b/ubuntu_kernel_selftests/ubuntu_kernel_selftests.py
> @@ -216,8 +216,16 @@ class ubuntu_kernel_selftests(test.test):
> if test_name.endswith('-build'):
> os.chdir(self.srcdir)
> if "net" in test_name:
> - cmd = "sh -c 'echo 1 > /proc/sys/net/ipv4/conf/all/accept_local'"
> - utils.system(cmd)
> + cmds = []
> + cmds.append("sh -c 'echo 1 > /proc/sys/net/ipv4/conf/all/accept_local'")
> + # The net benchmarching tests (e.g. udpgso) can fail when
> + # optmem limit is reached.
> + # https://www.kernel.org/doc/html/latest/networking/msg_zerocopy.html#transmission
> + # LP #1960907
> + cmds.append("sh -c 'echo 2048000 > /proc/sys/net/core/optmem_max'")
> + for cmd in cmds:
> + utils.system(cmd)
> +
> if self.kv >= 415:
> # net selftests use a module built by bpf selftests, bpf is available since bionic kernel
> if self.kv == 506:
Acked-by: Tim Gardner <tim.gardner at canonical.com>
Seems reasonable
--
-----------
Tim Gardner
Canonical, Inc
More information about the kernel-team
mailing list