]> www.infradead.org Git - users/jedix/linux-maple.git/commit
Merge branch 'net-mlx5e-shampo-enable-hw-gro-once-more'
authorJakub Kicinski <kuba@kernel.org>
Thu, 6 Jun 2024 03:13:10 +0000 (20:13 -0700)
committerJakub Kicinski <kuba@kernel.org>
Thu, 6 Jun 2024 03:20:47 +0000 (20:20 -0700)
commit7da375e2c7e023957b71fce44a72107559cfa6d0
treeac187f11c0650989716a44f6f752477936b4b6df
parented20142ed68c2b8819120508bc029e84d13cfe63
parent14ae2fd12be8c5089e43fee8a21cd8631699b97a
Merge branch 'net-mlx5e-shampo-enable-hw-gro-once-more'

Tariq Toukan says:

====================
net/mlx5e: SHAMPO, Enable HW GRO once more

This series enables hardware GRO for ConnectX-7 and newer NICs.
SHAMPO stands for Split Header And Merge Payload Offload.

The first part of the series contains important fixes and improvements.

The second part reworks the HW GRO counters.

Lastly, HW GRO is perf optimized and enabled.

Here are the bandwidth numbers for a simple iperf3 test over a single rq
where the application and irq are pinned to the same CPU:

+---------+--------+--------+-----------+-------------+
| streams | SW GRO | HW GRO | Unit      | Improvement |
+---------+--------+--------+-----------+-------------+
| 1       | 36     | 57     | Gbits/sec |    1.6 x    |
| 4       | 34     | 50     | Gbits/sec |    1.5 x    |
| 8       | 31     | 43     | Gbits/sec |    1.4 x    |
+---------+--------+--------+-----------+-------------+

Benchmark details:
VM based setup
CPU: Intel(R) Xeon(R) Platinum 8380 CPU, 24 cores
NIC: ConnectX-7 100GbE
iperf3 and irq running on same CPU over a single receive queue
====================

Link: https://lore.kernel.org/r/20240603212219.1037656-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>