Re: [PATCH net-next] net: bcmgenet: enable EEE TX LPI by default

From: Nicolai Buchwitz

Date: Tue Mar 17 2026 - 03:55:34 EST


On 16.3.2026 22:21, Florian Fainelli wrote:
On 3/16/26 13:59, Jakub Kicinski wrote:
On Mon, 16 Mar 2026 12:54:46 -0700 Florian Fainelli wrote:
On 3/3/26 08:02, Nicolai Buchwitz wrote:
GENET never initializes priv->eee.eee_enabled or
priv->eee.tx_lpi_enabled to true, so the link-change callback in
bcmgenet_mac_config() always calls bcmgenet_eee_enable_set() with
enable=false. The result is that EEE is negotiated at the PHY level
but the MAC never enters Low Power Idle, wasting the power savings
that EEE is designed to provide.

Users can work around this with 'ethtool --set-eee eth0 eee on tx-lpi
on tx-timer 250', but the MAC should enable TX LPI out of the box when
the hardware supports it, consistent with how phylink-managed MACs
behave (phylink calls phy_support_eee() which sets eee_enabled=true and
tx_lpi_enabled=true by default).

Initialize EEE software state during bcmgenet_open() on GENET v2+ so
that the link-change path activates EEE in hardware once the PHY
negotiates it. Use an LPI timer of 250 us, matching the default used
by other MAC drivers (mvneta, mvpp2).

Signed-off-by: Nicolai Buchwitz <nb@xxxxxxxxxxx>

Reviewed-by: Florian Fainelli <florian.fainelli@xxxxxxxxxxxx>

FWIW this patch is set to Changes Requested, I _think_ based on this
reply from Andrew:

https://lore.kernel.org/all/d352039f-4cbb-41e6-9aeb-0b4f3941b54c@xxxxxxx/

LMK if the patch is good as is.

It is good after 908c344d5cfac4160f49715da9efacdf5b6a28bd ("net: bcmgenet: fix broken EEE by converting to phylib-managed state") reaches net/next which is what Andrew was commenting on.

908c344d5cfac4160f49715da9efacdf5b6a28bd ("net: bcmgenet: fix broken EEE by converting to phylib-managed state") should cover it - phylib handles EEE activation automatically and sets the LPI tx timer from the phy default. I noted in the new patch that both earlier submissions are superseded, but forgot to mention it in the patches themselves. Apologies, still getting used to the list-based workflow!

Nicolai