net: ethernet: mtk_eth_soc: fix remaining throughput regression

Based on further tests, it seems that the QDMA shaper is not able to
perform shaping close to the MAC link rate without throughput loss.
This cannot be compensated by increasing the shaping rate, so it seems
to be an internal limit.

Fix the remaining throughput regression by detecting that condition and
limiting shaping to ports with lower link speed.

This patch intentionally ignores link speed gain from TRGMII, because
even on such links, shaping to 1000 Mbit/s incurs some throughput
degradation.

Fixes: f63959c7eec3 ("net: ethernet: mtk_eth_soc: implement multi-queue support for per-port queues")
Tested-By: Frank Wunderlich <frank-w@public-files.de>
Reported-by: Frank Wunderlich <frank-w@public-files.de>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
index 282f9435d5ff1f07d2edd9a0ad5cf1a4832d8d34..e14050e178624752c8c15f1f8e3a0347f40d4092 100644 (file)
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -753,6 +753,7 @@ static void mtk_mac_link_up(struct phylink_config *config,
                 MAC_MCR_FORCE_RX_FC);
 
        /* Configure speed */
+       mac->speed = speed;
        switch (speed) {
        case SPEED_2500:
        case SPEED_1000:
@@ -3235,6 +3236,9 @@ found:
        if (dp->index >= MTK_QDMA_NUM_QUEUES)
                return NOTIFY_DONE;
 
+       if (mac->speed > 0 && mac->speed <= s.base.speed)
+               s.base.speed = 0;
+
        mtk_set_queue_speed(eth, dp->index + 3, s.base.speed);
 
        return NOTIFY_DONE;