[v2] net: lan966x: Improve TX/RX of frames from/to CPU

[PATCH net-next v2 0/2] net: lan966x: Improve TX/RX of frames from/to CPU

Horatiu Vultur posted 2 patches 2 years, 11 months ago

Diff against v1
Download series mbox

.../net/ethernet/microchip/lan966x/Kconfig    |  1 -
.../ethernet/microchip/lan966x/lan966x_fdma.c |  2 +-
.../ethernet/microchip/lan966x/lan966x_main.c | 76 +++++++++++++------
.../ethernet/microchip/lan966x/lan966x_main.h |  5 +-
.../ethernet/microchip/lan966x/lan966x_ptp.c  | 20 ++---
5 files changed, 65 insertions(+), 39 deletions(-)

Expand all Fold all

[PATCH net-next v2 0/2] net: lan966x: Improve TX/RX of frames from/to CPU

Posted by Horatiu Vultur 2 years, 11 months ago

The first patch of this series improves the RX side. As it seems to be
an expensive operation to read the RX timestamp for every frame, then
read it only if it is required. This will give an improvement of ~70mbit
on the RX side.
The second patch stops using the packing library. This improves mostly
the TX side as this library is used to set diffent bits in the IFH. If
this library is replaced with a more simple/shorter implementation,
this gives an improvement of more than 100mbit on TX side.
All the measurements were done using iperf3.

v1->v2:
- update lan966x_ifh_set to set the bytes and not each bit individually

Horatiu Vultur (2):
  net: lan966x: Don't read RX timestamp if not needed
  net: lan966x: Stop using packing library

 .../net/ethernet/microchip/lan966x/Kconfig    |  1 -
 .../ethernet/microchip/lan966x/lan966x_fdma.c |  2 +-
 .../ethernet/microchip/lan966x/lan966x_main.c | 76 +++++++++++++------
 .../ethernet/microchip/lan966x/lan966x_main.h |  5 +-
 .../ethernet/microchip/lan966x/lan966x_ptp.c  | 20 ++---
 5 files changed, 65 insertions(+), 39 deletions(-)

-- 
2.38.0

Re: [PATCH net-next v2 0/2] net: lan966x: Improve TX/RX of frames from/to CPU

Posted by patchwork-bot+netdevbpf@kernel.org 2 years, 10 months ago

Hello:

This series was applied to netdev/net-next.git (main)
by David S. Miller <davem@davemloft.net>:

On Fri, 17 Mar 2023 16:27:11 +0100 you wrote:
> The first patch of this series improves the RX side. As it seems to be
> an expensive operation to read the RX timestamp for every frame, then
> read it only if it is required. This will give an improvement of ~70mbit
> on the RX side.
> The second patch stops using the packing library. This improves mostly
> the TX side as this library is used to set diffent bits in the IFH. If
> this library is replaced with a more simple/shorter implementation,
> this gives an improvement of more than 100mbit on TX side.
> All the measurements were done using iperf3.
> 
> [...]

Here is the summary with links:
  - [net-next,v2,1/2] net: lan966x: Don't read RX timestamp if not needed
    https://git.kernel.org/netdev/net-next/c/ff89ac704e2c
  - [net-next,v2,2/2] net: lan966x: Stop using packing library
    https://git.kernel.org/netdev/net-next/c/fd7627833ddf

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html