net/mlx5e: Replace reciprocal_scale in TX select queue function
authorShay Agroskin <shayag@mellanox.com>
Sun, 28 Apr 2019 07:14:23 +0000 (10:14 +0300)
committerSaeed Mahameed <saeedm@mellanox.com>
Fri, 7 Jun 2019 21:40:37 +0000 (14:40 -0700)
commit57c70d8740f740498a52f9c0c0d7295829b944de
treecc309b729e2ee9d678adca239c208a015fb1c5c8
parentd3cbd4254df881777e2efb68ee10ede0d9dc0647
net/mlx5e: Replace reciprocal_scale in TX select queue function

The TX queue index returned by the fallback function ranges
between [0,NUM CHANNELS - 1] if QoS isn't set and
[0, (NUM CHANNELS)*(NUM TCs) -1] otherwise.

Our HW uses different TC mapping than the fallback function
(which is denoted as 'up', user priority) so we only need to extract
a channel number out of the returned value.

Since (NUM CHANNELS)*(NUM TCs) is a relatively small number, using
reciprocal scale almost always returns zero.
We instead access the 'txq2sq' table to extract the sq (and with it the
channel number) associated with the tx queue, thus getting
a more evenly distributed channel number.

Perf:

Rx/Tx side with Intel(R) Xeon(R) Silver 4108 CPU @ 1.80GHz and ConnectX-5.
Used 'iperf' UDP traffic, 10 threads, and priority 5.

Before: 0.566Mpps
After:  2.37Mpps

As expected, releasing the existing bottleneck of steering all traffic
to TX queue zero significantly improves transmission rates.

Fixes: 7ccdd0841b30 ("net/mlx5e: Fix select queue callback")
Signed-off-by: Shay Agroskin <shayag@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/en.h
drivers/net/ethernet/mellanox/mlx5/core/en_main.c
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c