gve: process XSK TX descriptors as part of RX NAPI

commit ba0925c upstream.

When busy polling is enabled, xsk_sendmsg for AF_XDP zero copy marks
the NAPI ID corresponding to the memory pool allocated for the socket.
In GVE, this NAPI ID will never correspond to a NAPI ID of one of the
dedicated XDP TX queues registered with the umem because XDP TX is not
set up to share a NAPI with a corresponding RX queue.

This patch moves XSK TX descriptor processing from the TX NAPI to the RX
NAPI, and the gve_xsk_wakeup callback is updated to use the RX NAPI
instead of the TX NAPI, accordingly. The branch on if the wakeup is for
TX is removed, as the NAPI poll should be invoked whether the wakeup is
for TX or for RX.

Fixes: fd8e403 ("gve: Add AF_XDP zero-copy support for GQI-QPL format")
Cc: [email protected]
Signed-off-by: Praveen Kaligineedi <[email protected]>
Signed-off-by: Joshua Washington <[email protected]>
Reviewed-by: Willem de Bruijn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: josh8551021

drivers/net/ethernet/google/gve/gve.h

@@ -1134,6 +1134,7 @@ int gve_xdp_xmit_one(struct gve_priv *priv, struct gve_tx_ring *tx,
 void gve_xdp_tx_flush(struct gve_priv *priv, u32 xdp_qid);
 bool gve_tx_poll(struct gve_notify_block *block, int budget);
 bool gve_xdp_poll(struct gve_notify_block *block, int budget);
+int gve_xsk_tx_poll(struct gve_notify_block *block, int budget);
 int gve_tx_alloc_rings_gqi(struct gve_priv *priv,
 			   struct gve_tx_alloc_rings_cfg *cfg);
 void gve_tx_free_rings_gqi(struct gve_priv *priv,

drivers/net/ethernet/google/gve/gve_main.c

@@ -333,6 +333,14 @@ int gve_napi_poll(struct napi_struct *napi, int budget)
 
 	if (block->rx) {
 		work_done = gve_rx_poll(block, budget);
+
+		/* Poll XSK TX as part of RX NAPI. Setup re-poll based on max of
+		 * TX and RX work done.
+		 */
+		if (priv->xdp_prog)
+			work_done = max_t(int, work_done,
+					  gve_xsk_tx_poll(block, budget));
+
 		reschedule |= work_done == budget;
 	}
 

drivers/net/ethernet/google/gve/gve_tx.c

@@ -975,33 +975,41 @@ static int gve_xsk_tx(struct gve_priv *priv, struct gve_tx_ring *tx,
 	return sent;
 }
 
+int gve_xsk_tx_poll(struct gve_notify_block *rx_block, int budget)
+{
+	struct gve_rx_ring *rx = rx_block->rx;
+	struct gve_priv *priv = rx->gve;
+	struct gve_tx_ring *tx;
+	int sent = 0;
+
+	tx = &priv->tx[gve_xdp_tx_queue_id(priv, rx->q_num)];
+	if (tx->xsk_pool) {
+		sent = gve_xsk_tx(priv, tx, budget);
+
+		u64_stats_update_begin(&tx->statss);
+		tx->xdp_xsk_sent += sent;
+		u64_stats_update_end(&tx->statss);
+		if (xsk_uses_need_wakeup(tx->xsk_pool))
+			xsk_set_tx_need_wakeup(tx->xsk_pool);
+	}
+
+	return sent;
+}
+
 bool gve_xdp_poll(struct gve_notify_block *block, int budget)
 {
 	struct gve_priv *priv = block->priv;
 	struct gve_tx_ring *tx = block->tx;
 	u32 nic_done;
-	bool repoll;
 	u32 to_do;
 
 	/* Find out how much work there is to be done */
 	nic_done = gve_tx_load_event_counter(priv, tx);
 	to_do = min_t(u32, (nic_done - tx->done), budget);
 	gve_clean_xdp_done(priv, tx, to_do);
-	repoll = nic_done != tx->done;
-
-	if (tx->xsk_pool) {
-		int sent = gve_xsk_tx(priv, tx, budget);
-
-		u64_stats_update_begin(&tx->statss);
-		tx->xdp_xsk_sent += sent;
-		u64_stats_update_end(&tx->statss);
-		repoll |= (sent == budget);
-		if (xsk_uses_need_wakeup(tx->xsk_pool))
-			xsk_set_tx_need_wakeup(tx->xsk_pool);
-	}
 
 	/* If we still have work we want to repoll */
-	return repoll;
+	return nic_done != tx->done;
 }
 
 bool gve_tx_poll(struct gve_notify_block *block, int budget)