[PATCH net-next] net: stmmac: call stmmac_finalize_xdp_rx() on a condition

Leesoo Ahn lsahn at ooseel.net
Wed Mar 8 08:26:18 PST 2023


The current codebase calls the function no matter net device has XDP
programs or not. So the finalize function is being called everytime when RX
bottom-half in progress. It needs a few machine instructions for nothing
in the case that XDP programs are not attached at all.

Lets it call the function on a condition that if xdp_status variable has
not zero value. That means XDP programs are attached to the net device
and it should be finalized based on the variable.

The following instructions show that it's better than calling the function
unconditionally.

  0.31 │6b8:   ldr     w0, [sp, #196]
       │    ┌──cbz     w0, 6cc
       │    │  mov     x1, x0
       │    │  mov     x0, x27
       │    │→ bl     stmmac_finalize_xdp_rx
       │6cc:└─→ldr    x1, [sp, #176]

with 'if (xdp_status)' statement, jump to '6cc' label if xdp_status has
zero value.

Signed-off-by: Leesoo Ahn <lsahn at ooseel.net>
---
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index e4902a7bb61e..53c6e9b3a0c2 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -5145,7 +5145,8 @@ static int stmmac_rx_zc(struct stmmac_priv *priv, int limit, u32 queue)
 		rx_q->state.len = len;
 	}
 
-	stmmac_finalize_xdp_rx(priv, xdp_status);
+	if (xdp_status)
+		stmmac_finalize_xdp_rx(priv, xdp_status);
 
 	priv->xstats.rx_pkt_n += count;
 	priv->xstats.rxq_stats[queue].rx_pkt_n += count;
@@ -5425,7 +5426,8 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit, u32 queue)
 		rx_q->state.len = len;
 	}
 
-	stmmac_finalize_xdp_rx(priv, xdp_status);
+	if (xdp_status)
+		stmmac_finalize_xdp_rx(priv, xdp_status);
 
 	stmmac_rx_refill(priv, queue);
 
-- 
2.34.1




More information about the linux-arm-kernel mailing list