[OpenWrt-Devel] [RFC] ag71xx: Enable flow control for builtin switch port

[Conn O'Griofa]

On Mon, Apr 18, 2016 at 1:15 PM, Sven Eckelmann <
sven.eckelmann at open-mesh.com> wrote:

> From: Sven Eckelmann <sven.eckelmann at open-mesh.com>
>
> The port connected to the internal switch tends to drop a lot of packets
> when a lot of data is transferred over it. This is especially visible when
> IP fragmentation happens for large UDP and ICMP packets. An easy test for
> this is
>
>     # works
>     ping 192.168.1.23
>     # doesn't work
>     ping -s 65507 192.168.1.23
>
> But enabling flow control on ports without the builtin switch seems to
> break the connection completely in some situations. This was for example
> detected when a QCA955x with AR98533 on GMAC1 was started with an active
> link on this port. So only enable it for SoC ethernet devices with
> switch attached.
>
> This closes #19498
>
> Signed-off-by: Sven Eckelmann <sven.eckelmann at open-mesh.com>
> ---
> This is only an RFC because Felix reverted a similar change in r27034.
> Unfortunately, the revert commit doesn't contain enough information to
> find out if this problem would also happen when only activating flow
> control on the switch port port or when only enabling TX and not RX
> flow control (or the other way around).
>
>  .../ar71xx/files/drivers/net/ethernet/atheros/ag71xx/ag71xx_main.c  | 6
> +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git
> a/target/linux/ar71xx/files/drivers/net/ethernet/atheros/ag71xx/ag71xx_main.c
> b/target/linux/ar71xx/files/drivers/net/ethernet/atheros/ag71xx/ag71xx_main.c
> index 0832059..208b1d6 100644
> ---
> a/target/linux/ar71xx/files/drivers/net/ethernet/atheros/ag71xx/ag71xx_main.c
> +++
> b/target/linux/ar71xx/files/drivers/net/ethernet/atheros/ag71xx/ag71xx_main.c
> @@ -455,7 +455,11 @@ static void ag71xx_hw_setup(struct ag71xx *ag)
>         struct ag71xx_platform_data *pdata = ag71xx_get_pdata(ag);
>
>         /* setup MAC configuration registers */
> -       ag71xx_wr(ag, AG71XX_REG_MAC_CFG1, MAC_CFG1_INIT);
> +       if (pdata->is_ar724x && pdata->switch_data)
> +               ag71xx_wr(ag, AG71XX_REG_MAC_CFG1,
> +                         MAC_CFG1_INIT | MAC_CFG1_TFC | MAC_CFG1_RFC);
> +       else
> +               ag71xx_wr(ag, AG71XX_REG_MAC_CFG1, MAC_CFG1_INIT);
>
>         ag71xx_sb(ag, AG71XX_REG_MAC_CFG2,
>                   MAC_CFG2_PAD_CRC_EN | MAC_CFG2_LEN_CHECK);
> --
> 2.8.0.rc3
> _______________________________________________
> openwrt-devel mailing list
> openwrt-devel at lists.openwrt.org
> https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
>

It looks like you may have solved the original issue; this patch is working
fine on WR-842ND v1. More details:

* When I tried reverting r27034, the WAN port stopped functioning
correctly, showing a tx timeout  error - but the LAN ports still worked.
With your patch, all ports work correctly.
* Your testcase (ping -s 65507 from/to a wired client between LAN ports)
usually results in 20%-40% packet loss. With your patch (as with the
original), it's now typically 0% packet loss between LAN ports.

Conn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.infradead.org/pipermail/openwrt-devel/attachments/20160419/dbe210da/attachment.htm>
-------------- next part --------------
_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel