xsk: Bring back busy polling support

Commit 86e25f4 ("net: napi: Add napi_config") moved napi->napi_id
assignment to a later point in time (napi_hash_add_with_id). This breaks
__xdp_rxq_info_reg which copies napi_id at an earlier time and now
stores 0 napi_id. It also makes sk_mark_napi_id_once_xdp and
__sk_mark_napi_id_once useless because they now work against 0 napi_id.
Since sk_busy_loop requires valid napi_id to busy-poll on, there is no way
to busy-poll AF_XDP sockets anymore.

Bring back the ability to busy-poll on XSK by resolving socket's napi_id
at bind time. This relies on relatively recent netif_queue_set_napi,
but (assume) at this point most popular drivers should have been converted.
This also removes per-tx/rx cycles which used to check and/or set
the napi_id value.

Confirmed by running a busy-polling AF_XDP socket
(github.com/fomichev/xskrtt) on mlx5 and looking at BusyPollRxPackets
from /proc/net/netstat.

Fixes: 86e25f4 ("net: napi: Add napi_config")
Signed-off-by: Stanislav Fomichev <[email protected]>
Acked-by: Magnus Karlsson <[email protected]>
Reviewed-by: Jakub Kicinski <[email protected]>
Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: Stanislav Fomichev

include/net/busy_poll.h

@@ -174,12 +174,4 @@ static inline void sk_mark_napi_id_once(struct sock *sk,
 #endif
 }
 
-static inline void sk_mark_napi_id_once_xdp(struct sock *sk,
-					    const struct xdp_buff *xdp)
-{
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	__sk_mark_napi_id_once(sk, xdp->rxq->napi_id);
-#endif
-}
-
 #endif /* _LINUX_NET_BUSY_POLL_H */

include/net/xdp.h

@@ -62,7 +62,6 @@ struct xdp_rxq_info {
 	u32 queue_index;
 	u32 reg_state;
 	struct xdp_mem_info mem;
-	unsigned int napi_id;
 	u32 frag_size;
 } ____cacheline_aligned; /* perf critical, avoid false-sharing */
 

include/net/xdp_sock_drv.h

@@ -59,15 +59,6 @@ static inline void xsk_pool_fill_cb(struct xsk_buff_pool *pool,
 	xp_fill_cb(pool, desc);
 }
 
-static inline unsigned int xsk_pool_get_napi_id(struct xsk_buff_pool *pool)
-{
-#ifdef CONFIG_NET_RX_BUSY_POLL
-	return pool->heads[0].xdp.rxq->napi_id;
-#else
-	return 0;
-#endif
-}
-
 static inline void xsk_pool_dma_unmap(struct xsk_buff_pool *pool,
 				      unsigned long attrs)
 {
@@ -306,11 +297,6 @@ static inline void xsk_pool_fill_cb(struct xsk_buff_pool *pool,
 {
 }
 
-static inline unsigned int xsk_pool_get_napi_id(struct xsk_buff_pool *pool)
-{
-	return 0;
-}
-
 static inline void xsk_pool_dma_unmap(struct xsk_buff_pool *pool,
 				      unsigned long attrs)
 {

net/core/xdp.c

@@ -186,7 +186,6 @@ int __xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
 	xdp_rxq_info_init(xdp_rxq);
 	xdp_rxq->dev = dev;
 	xdp_rxq->queue_index = queue_index;
-	xdp_rxq->napi_id = napi_id;
 	xdp_rxq->frag_size = frag_size;
 
 	xdp_rxq->reg_state = REG_STATE_REGISTERED;

net/xdp/xsk.c

@@ -322,7 +322,6 @@ static int xsk_rcv_check(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
 		return -ENOSPC;
 	}
 
-	sk_mark_napi_id_once_xdp(&xs->sk, xdp);
 	return 0;
 }
 
@@ -908,11 +907,8 @@ static int __xsk_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len
 	if (unlikely(!xs->tx))
 		return -ENOBUFS;
 
-	if (sk_can_busy_loop(sk)) {
-		if (xs->zc)
-			__sk_mark_napi_id_once(sk, xsk_pool_get_napi_id(xs->pool));
+	if (sk_can_busy_loop(sk))
 		sk_busy_loop(sk, 1); /* only support non-blocking sockets */
-	}
 
 	if (xs->zc && xsk_no_wakeup(sk))
 		return 0;
@@ -1298,6 +1294,14 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 	xs->queue_id = qid;
 	xp_add_xsk(xs->pool, xs);
 
+	if (xs->zc && qid < dev->real_num_rx_queues) {
+		struct netdev_rx_queue *rxq;
+
+		rxq = __netif_get_rx_queue(dev, qid);
+		if (rxq->napi)
+			__sk_mark_napi_id_once(sk, rxq->napi->napi_id);
+	}
+
 out_unlock:
 	if (err) {
 		dev_put(dev);