[PATCH net-next 4/4] net: rmnet: optimize rx handler by replacing skb_linearize with pskb_may_pull

[Koen Vandeputte]

The rmnet RX handler currently calls skb_linearize() unconditionally
on every incoming packet.

When receiving large, hardware-aggregated frames
(e.g., 16KB or 32KB from USB modems via qmi_wwan),
this forces the network stack to allocate a massive
contiguous buffer and memcpy the entire scattered payload
out of paged memory.

This severe performance bottleneck is completely unnecessary
because the RX handler only needs to inspect the initial
4-byte MAP header to determine the logical endpoint
and route the packet.

Replace skb_linearize() with pskb_may_pull().
This targeted check ensures the MAP header is safely accessible
in the linear data area while allowing the bulky payload to
remain untouched in paged memory fragments.
This preserves the zero-copy architecture for aggregated frames,
drastically reducing CPU overhead, cache thrashing,
and memory allocations on the RX hotpath.

On a slower embedded board, this improves throughput by ~10%.

Signed-off-by: Koen Vandeputte <koen.vandeputte@xxxxxxxxxxxx>
---
 drivers/net/ethernet/qualcomm/rmnet/rmnet_handlers.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/qualcomm/rmnet/rmnet_handlers.c b/drivers/net/ethernet/qualcomm/rmnet/rmnet_handlers.c
index 95c3e3934fd3..a2379df69a6d 100644
--- a/drivers/net/ethernet/qualcomm/rmnet/rmnet_handlers.c
+++ b/drivers/net/ethernet/qualcomm/rmnet/rmnet_handlers.c
@@ -217,7 +217,7 @@ rx_handler_result_t rmnet_rx_handler(struct sk_buff **pskb)
 	if (!skb)
 		goto done;
 
-	if (skb_linearize(skb)) {
+	if (!pskb_may_pull(skb, sizeof(struct rmnet_map_header))) {
 		kfree_skb(skb);
 		goto done;
 	}
-- 
2.43.0