From: Eric Dumazet <eduma...@google.com>
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 <eduma...@google.com>
Reported-by: Rolf Neugebauer <rolf.neugeba...@docker.com>
Reported-by: Eric W. Biederman <ebied...@xmission.com>
---
include/net/gro_cells.h | 3 +++
1 file changed, 3 insertions(+)
diff --git a/include/net/gro_cells.h b/include/net/gro_cells.h
index
d15214d673b2e8e08fd6437b572278fb1359f10d..2a1abbf8da74368cd01adc40cef6c0644e059ef2
100644
--- a/include/net/gro_cells.h
+++ b/include/net/gro_cells.h
@@ -68,6 +68,9 @@ static inline int gro_cells_init(struct gro_cells *gcells,
struct net_device *de
struct gro_cell *cell = per_cpu_ptr(gcells->cells, i);
__skb_queue_head_init(&cell->napi_skbs);
+
+ set_bit(NAPI_STATE_NO_BUSY_POLL, &cell->napi.state);
+
netif_napi_add(dev, &cell->napi, gro_cell_poll, 64);
napi_enable(&cell->napi);
}
