[net-next,2/3] net: Optimize xdp_do_flush() with bpf_net_context infos.

Commit Message

Sebastian Andrzej Siewior June 28, 2024, 10:18 a.m. UTC

Every NIC driver utilizing XDP should invoke xdp_do_flush() after
processing all packages. With the introduction of the bpf_net_context
logic the flush lists (for dev, CPU-map and xsk) are lazy initialized
only if used. However xdp_do_flush() tries to flush all three of them so
all three lists are always initialized and the likely empty lists are
"iterated".
Without the usage of XDP but with CONFIG_DEBUG_NET the lists are also
initialized due to xdp_do_check_flushed().

Jakub suggest to utilize the hints in bpf_net_context and avoid invoking
the flush function. This will also avoiding initializing the lists which
are otherwise unused.

Introduce bpf_net_ctx_get_all_used_flush_lists() to return the
individual list if not-empty. Use the logic in xdp_do_flush() and
xdp_do_check_flushed(). Remove the not needed .*_check_flush().

Suggested-by: Jakub Kicinski <[e-mail]>
Signed-off-by: Sebastian Andrzej Siewior <[e-mail]>
---
 include/linux/bpf.h    | 10 ++++------
 include/linux/filter.h | 27 +++++++++++++++++++++++++++
 include/net/xdp_sock.h | 14 ++------------
 kernel/bpf/cpumap.c    | 13 +------------
 kernel/bpf/devmap.c    | 13 +------------
 net/core/filter.c      | 33 +++++++++++++++++++++++++--------
 net/xdp/xsk.c          | 13 +------------
 7 files changed, 61 insertions(+), 62 deletions(-)

Patch

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index a834f4b761bc5..f5c6bc9093a6b 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2494,7 +2494,7 @@  struct sk_buff;
 struct bpf_dtab_netdev;
 struct bpf_cpu_map_entry;
 
-void __dev_flush(void);
+void __dev_flush(struct list_head *flush_list);
 int dev_xdp_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
 		    struct net_device *dev_rx);
 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_frame *xdpf,
@@ -2507,7 +2507,7 @@  int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb,
 			   struct bpf_prog *xdp_prog, struct bpf_map *map,
 			   bool exclude_ingress);
 
-void __cpu_map_flush(void);
+void __cpu_map_flush(struct list_head *flush_list);
 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf,
 		    struct net_device *dev_rx);
 int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu,
@@ -2644,8 +2644,6 @@  void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
 void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr);
 void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr);
 
-bool dev_check_flush(void);
-bool cpu_map_check_flush(void);
 #else /* !CONFIG_BPF_SYSCALL */
 static inline struct bpf_prog *bpf_prog_get(u32 ufd)
 {
@@ -2738,7 +2736,7 @@  static inline struct bpf_token *bpf_token_get_from_fd(u32 ufd)
 	return ERR_PTR(-EOPNOTSUPP);
 }
 
-static inline void __dev_flush(void)
+static inline void __dev_flush(struct list_head *flush_list)
 {
 }
 
@@ -2784,7 +2782,7 @@  int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb,
 	return 0;
 }
 
-static inline void __cpu_map_flush(void)
+static inline void __cpu_map_flush(struct list_head *flush_list)
 {
 }
 
diff --git a/include/linux/filter.h b/include/linux/filter.h
index c0349522de8fb..02ddcfdf94c46 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -829,6 +829,33 @@  static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
 	return &bpf_net_ctx->xskmap_map_flush_list;
 }
 
+static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map,
+							struct list_head **lh_dev,
+							struct list_head **lh_xsk)
+{
+	struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
+	u32 kern_flags = bpf_net_ctx->ri.kern_flags;
+	struct list_head *lh;
+
+	*lh_map = *lh_dev = *lh_xsk = NULL;
+
+	if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
+		return;
+
+	lh = &bpf_net_ctx->dev_map_flush_list;
+	if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh))
+		*lh_dev = lh;
+
+	lh = &bpf_net_ctx->cpu_map_flush_list;
+	if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh))
+		*lh_map = lh;
+
+	lh = &bpf_net_ctx->xskmap_map_flush_list;
+	if (IS_ENABLED(CONFIG_XDP_SOCKETS) &&
+	    kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh))
+		*lh_xsk = lh;
+}
+
 /* Compute the linear packet data range [data, data_end) which
  * will be accessed by various program types (cls_bpf, act_bpf,
  * lwt, ...). Subsystems allowing direct data access must (!)
diff --git a/include/net/xdp_sock.h b/include/net/xdp_sock.h
index 3d54de168a6d9..bfe625b55d55d 100644
--- a/include/net/xdp_sock.h
+++ b/include/net/xdp_sock.h
@@ -121,7 +121,7 @@  struct xsk_tx_metadata_ops {
 
 int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp);
 int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp);
-void __xsk_map_flush(void);
+void __xsk_map_flush(struct list_head *flush_list);
 
 /**
  *  xsk_tx_metadata_to_compl - Save enough relevant metadata information
@@ -206,7 +206,7 @@  static inline int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp)
 	return -EOPNOTSUPP;
 }
 
-static inline void __xsk_map_flush(void)
+static inline void __xsk_map_flush(struct list_head *flush_list)
 {
 }
 
@@ -228,14 +228,4 @@  static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl,
 }
 
 #endif /* CONFIG_XDP_SOCKETS */
-
-#if defined(CONFIG_XDP_SOCKETS) && defined(CONFIG_DEBUG_NET)
-bool xsk_map_check_flush(void);
-#else
-static inline bool xsk_map_check_flush(void)
-{
-	return false;
-}
-#endif
-
 #endif /* _LINUX_XDP_SOCK_H */
diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index 068e994ed781a..4acf90cd79eb4 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -757,9 +757,8 @@  int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu,
 	return ret;
 }
 
-void __cpu_map_flush(void)
+void __cpu_map_flush(struct list_head *flush_list)
 {
-	struct list_head *flush_list = bpf_net_ctx_get_cpu_map_flush_list();
 	struct xdp_bulk_queue *bq, *tmp;
 
 	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {
@@ -769,13 +768,3 @@  void __cpu_map_flush(void)
 		wake_up_process(bq->obj->kthread);
 	}
 }
-
-#ifdef CONFIG_DEBUG_NET
-bool cpu_map_check_flush(void)
-{
-	if (list_empty(bpf_net_ctx_get_cpu_map_flush_list()))
-		return false;
-	__cpu_map_flush();
-	return true;
-}
-#endif
diff --git a/kernel/bpf/devmap.c b/kernel/bpf/devmap.c
index 317ac2d66ebd1..9ca47eaacdd5e 100644
--- a/kernel/bpf/devmap.c
+++ b/kernel/bpf/devmap.c
@@ -412,9 +412,8 @@  static void bq_xmit_all(struct xdp_dev_bulk_queue *bq, u32 flags)
  * driver before returning from its napi->poll() routine. See the comment above
  * xdp_do_flush() in filter.c.
  */
-void __dev_flush(void)
+void __dev_flush(struct list_head *flush_list)
 {
-	struct list_head *flush_list = bpf_net_ctx_get_dev_flush_list();
 	struct xdp_dev_bulk_queue *bq, *tmp;
 
 	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {
@@ -425,16 +424,6 @@  void __dev_flush(void)
 	}
 }
 
-#ifdef CONFIG_DEBUG_NET
-bool dev_check_flush(void)
-{
-	if (list_empty(bpf_net_ctx_get_dev_flush_list()))
-		return false;
-	__dev_flush();
-	return true;
-}
-#endif
-
 /* Elements are kept alive by RCU; either by rcu_read_lock() (from syscall) or
  * by local_bh_disable() (from XDP calls inside NAPI). The
  * rcu_read_lock_bh_held() below makes lockdep accept both.
diff --git a/net/core/filter.c b/net/core/filter.c
index eb1c4425c06f3..403d23faf22e1 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -4277,22 +4277,39 @@  static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
  */
 void xdp_do_flush(void)
 {
-	__dev_flush();
-	__cpu_map_flush();
-	__xsk_map_flush();
+	struct list_head *lh_map, *lh_dev, *lh_xsk;
+
+	bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
+	if (lh_dev)
+		__dev_flush(lh_dev);
+	if (lh_map)
+		__cpu_map_flush(lh_map);
+	if (lh_xsk)
+		__xsk_map_flush(lh_xsk);
 }
 EXPORT_SYMBOL_GPL(xdp_do_flush);
 
 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
 void xdp_do_check_flushed(struct napi_struct *napi)
 {
-	bool ret;
+	struct list_head *lh_map, *lh_dev, *lh_xsk;
+	bool missed = false;
 
-	ret = dev_check_flush();
-	ret |= cpu_map_check_flush();
-	ret |= xsk_map_check_flush();
+	bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
+	if (lh_dev) {
+		__dev_flush(lh_dev);
+		missed = true;
+	}
+	if (lh_map) {
+		__cpu_map_flush(lh_map);
+		missed = true;
+	}
+	if (lh_xsk) {
+		__xsk_map_flush(lh_xsk);
+		missed = true;
+	}
 
-	WARN_ONCE(ret, "Missing xdp_do_flush() invocation after NAPI by %ps\n",
+	WARN_ONCE(missed, "Missing xdp_do_flush() invocation after NAPI by %ps\n",
 		  napi->poll);
 }
 #endif
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index ed062e0383896..de9c0322bc294 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -383,9 +383,8 @@  int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp)
 	return 0;
 }
 
-void __xsk_map_flush(void)
+void __xsk_map_flush(struct list_head *flush_list)
 {
-	struct list_head *flush_list = bpf_net_ctx_get_xskmap_flush_list();
 	struct xdp_sock *xs, *tmp;
 
 	list_for_each_entry_safe(xs, tmp, flush_list, flush_node) {
@@ -394,16 +393,6 @@  void __xsk_map_flush(void)
 	}
 }
 
-#ifdef CONFIG_DEBUG_NET
-bool xsk_map_check_flush(void)
-{
-	if (list_empty(bpf_net_ctx_get_xskmap_flush_list()))
-		return false;
-	__xsk_map_flush();
-	return true;
-}
-#endif
-
 void xsk_tx_completed(struct xsk_buff_pool *pool, u32 nb_entries)
 {
 	xskq_prod_submit_n(pool->cq, nb_entries);