mirror of
https://github.com/outbackdingo/mt76.git
synced 2026-10-24 13:26:48 +00:00
Filip BakreskiandFelix Fietkau
114c702810
The USB RX path delivers frames to the stack via mt76_rx_complete() with a NULL napi pointer, taking the netif_receive_skb_list() path, so it never benefits from GRO -- unlike the DMA-based mt76 drivers, which pass a real napi and use napi_gro_receive(). For bulk TCP traffic this is costly, as every segment traverses the stack individually. Service the MT_RXQ_MAIN queue from a threaded NAPI, reusing mt76_dev's existing napi_dev and napi[] rather than adding new fields. The URB completion handler schedules the napi; its poll drains the URBs, builds the skbs, resubmits and delivers them through napi_gro_receive(). The MCU queue stays on the existing RX worker. This enables GRO and moves RX processing into its own kernel thread, parallelising the datapath. On mt7921u at HE-MCS 11 (2x2, 80 MHz; fast.com, multiple streams) this averages ~588 Mbit/s, versus ~424 Mbit/s when the same napi is instead driven manually from the RX worker, and ~380 Mbit/s for the unmodified driver. Suggested-by: Lorenzo Bianconi <[email protected]> Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Filip Bakreski <[email protected]> Acked-by: Lorenzo Bianconi <[email protected]> Link: https://patch.msgid.link/[email protected]