Commit d839a731 authored by Sebastian Andrzej Siewior's avatar Sebastian Andrzej Siewior Committed by Paolo Abeni
Browse files

net: Optimize xdp_do_flush() with bpf_net_context infos. 



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: default avatarJakub Kicinski <kuba@kernel.org>
Signed-off-by: default avatarSebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: default avatarJakub Kicinski <kuba@kernel.org>
Signed-off-by: default avatarPaolo Abeni <pabeni@redhat.com>
parent 2896624b

include/linux/bpf.h

+4 −6
Original line number Diff line number Diff line
@@ -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)

{

}

include/linux/filter.h

+27 −0
Original line number Diff line number Diff line
@@ -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 (!)

include/net/xdp_sock.h

+2 −12
Original line number Diff line number Diff line
@@ -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 */

kernel/bpf/cpumap.c

+1 −12
Original line number Diff line number Diff line
@@ -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

kernel/bpf/devmap.c

+1 −12
Original line number Diff line number Diff line
@@ -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.