[PATCH net-next 2/4] net: rmnet: optimize rx deaggregation memory allocation

[Koen Vandeputte]

Currently, rmnet_map_deaggregate() uses alloc_skb() with GFP_ATOMIC to
allocate new socket buffers for individual packets extracted from an
aggregated frame. Under high-speed receive workloads, such as processing
large 16KB frames from a USB modem, repeatedly hitting the generic
kernel slab allocator introduces unnecessary lock contention and high
CPU overhead.

Replace alloc_skb() with netdev_alloc_skb() on the deaggregation path.
netdev_alloc_skb() is specifically designed for the network receive hot
path, utilizing a fast, lockless per-CPU NAPI cache. Furthermore, it
automatically provisions standard network padding (NET_SKB_PAD), ensuring
that the newly extracted IP headers are optimally cache-aligned for
efficient processing by the upper networking stack.

This change significantly reduces memory allocation overhead, prevents
slab allocator contention, and improves overall throughput when receiving
large bursts of aggregated traffic.

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

diff --git a/drivers/net/ethernet/qualcomm/rmnet/rmnet_map_data.c b/drivers/net/ethernet/qualcomm/rmnet/rmnet_map_data.c
index 2eafb1d969c1..ccdfde3711db 100644
--- a/drivers/net/ethernet/qualcomm/rmnet/rmnet_map_data.c
+++ b/drivers/net/ethernet/qualcomm/rmnet/rmnet_map_data.c
@@ -390,7 +390,7 @@ struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb,
 	if (!packet_len)
 		return NULL;
 
-	skbn = alloc_skb(packet_len + RMNET_MAP_DEAGGR_SPACING, GFP_ATOMIC);
+	skbn = netdev_alloc_skb(skb->dev, packet_len + RMNET_MAP_DEAGGR_SPACING);
 	if (!skbn)
 		return NULL;
 
-- 
2.43.0