commit 5c14a4d05f68415af9e41a4e667d1748d41d1baf upstream.

When we did the original tests for the optimal value of sk_pacing_shift, we came up with 6 ms of buffering as the default. Sadly, 6 is not a power of two, so when picking the shift value I erred on the size of less buffering and picked 4 ms instead of 8. This was probably wrong; those 2 ms of extra buffering makes a larger difference than I thought.

So, change the default pacing shift to 7, which corresponds to 8 ms of buffering. The point of diminishing returns really kicks in after 8 ms, and so having this as a default should cut down on the need for extensive per-device testing and overrides needed in the drivers.

Stable backport moves the change to where the value was set before the addition of per-device override support in 4.20.

Signed-off-by: Toke Høiland-Jørgensen toke@redhat.com --- net/mac80211/tx.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c index c7ccd7b71b15..743cde66aaf6 100644 --- a/net/mac80211/tx.c +++ b/net/mac80211/tx.c @@ -3614,10 +3614,10 @@ void __ieee80211_subif_start_xmit(struct sk_buff *skb, /* We need a bit of data queued to build aggregates properly, so * instruct the TCP stack to allow more than a single ms of data * to be queued in the stack. The value is a bit-shift of 1 - * second, so 8 is ~4ms of queued data. Only affects local TCP + * second, so 7 is ~8ms of queued data. Only affects local TCP * sockets. */ - sk_pacing_shift_update(skb->sk, 8); + sk_pacing_shift_update(skb->sk, 7);

fast_tx = rcu_dereference(sta->fast_tx);