gro_cells: mark napi struct as not busy poll candidates
authorEric Dumazet <edumazet@google.com>
Tue, 15 Nov 2016 00:28:42 +0000 (16:28 -0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 16 Nov 2016 03:27:27 +0000 (22:27 -0500)
commite88a2766143a27bfe6704b4493b214de4094cf29
treebb89a8330ce7936f26ef120d829e23b9a9de2527
parent73e2d5e34b6cdd1080038daf3d6d6d744a9eefe6
gro_cells: mark napi struct as not busy poll candidates

Rolf Neugebauer reported very long delays at netns dismantle.

Eric W. Biederman was kind enough to look at this problem
and noticed synchronize_net() occurring from netif_napi_del() that was
added in linux-4.5

Busy polling makes no sense for tunnels NAPI.
If busy poll is used for sessions over tunnels, the poller will need to
poll the physical device queue anyway.

netif_tx_napi_add() could be used here, but function name is misleading,
and renaming it is not stable material, so set NAPI_STATE_NO_BUSY_POLL
bit directly.

This will avoid inserting gro_cells napi structures in napi_hash[]
and avoid the problematic synchronize_net() (per possible cpu) that
Rolf reported.

Fixes: 93d05d4a320c ("net: provide generic busy polling to all NAPI drivers")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Rolf Neugebauer <rolf.neugebauer@docker.com>
Reported-by: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Tested-by: Rolf Neugebauer <rolf.neugebauer@docker.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/gro_cells.h