| Message ID | 1642716150-100589-1-git-send-email-jdamato@fastly.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net-next] sock: add sndbuff full perf trace event | expand |
On Thu, Jan 20, 2022 at 02:02:30PM -0800, Joe Damato wrote: > Calls to sock_alloc_send_pskb can fail or trigger a wait if there is no > sndbuf space available. > > Add a perf trace event to help users monitor when and how this occurs so > appropriate application level changes can be made, if necessary. > > Signed-off-by: Joe Damato <jdamato@fastly.com> > --- > include/trace/events/sock.h | 21 +++++++++++++++++++++ > net/core/sock.c | 1 + > 2 files changed, 22 insertions(+) There are more places where we set the SOCKWQ_ASYNC_NOSPACE bit, so it looks like your patch is incomplete. Thanks.
On Sun, Jan 23, 2022 at 12:17 PM Cong Wang <xiyou.wangcong@gmail.com> wrote: > > On Thu, Jan 20, 2022 at 02:02:30PM -0800, Joe Damato wrote: > > Calls to sock_alloc_send_pskb can fail or trigger a wait if there is no > > sndbuf space available. > > > > Add a perf trace event to help users monitor when and how this occurs so > > appropriate application level changes can be made, if necessary. > > > > Signed-off-by: Joe Damato <jdamato@fastly.com> > > --- > > include/trace/events/sock.h | 21 +++++++++++++++++++++ > > net/core/sock.c | 1 + > > 2 files changed, 22 insertions(+) > > There are more places where we set the SOCKWQ_ASYNC_NOSPACE bit, so > it looks like your patch is incomplete. Thanks for taking a look. I had originally taken your comment to suggest that the generic stream code should trigger the same tracepoint; I agree and am happy to generate a v2 with that case covered. Are you suggesting that all places where SOCKWQ_ASYNC_NOSPACE is set (e.g. the poll functions for various protocols) should trigger this tracepoint, as well? Some of these may not have a user configurable sndbuf, but I suppose knowing that this event is happening could be helpful. Thanks.
diff --git a/include/trace/events/sock.h b/include/trace/events/sock.h index 12c3157..de6f6a7 100644 --- a/include/trace/events/sock.h +++ b/include/trace/events/sock.h @@ -68,6 +68,27 @@ skmem_kind_names #define show_skmem_kind_names(val) \ __print_symbolic(val, skmem_kind_names) +TRACE_EVENT(sock_sndbuf_full, + TP_PROTO(struct sock *sk, long timeo), + + TP_ARGS(sk, timeo), + + TP_STRUCT__entry( + __field(int, wmem_alloc) + __field(int, sk_sndbuf) + __field(long, timeo) + ), + + TP_fast_assign( + __entry->wmem_alloc = refcount_read(&sk->sk_wmem_alloc); + __entry->sk_sndbuf = READ_ONCE(sk->sk_sndbuf); + __entry->timeo = timeo; + ), + + TP_printk("wmem_alloc=%d sk_sndbf=%d timeo=%ld", + __entry->wmem_alloc, __entry->sk_sndbuf, __entry->timeo) +); + TRACE_EVENT(sock_rcvqueue_full, TP_PROTO(struct sock *sk, struct sk_buff *skb), diff --git a/net/core/sock.c b/net/core/sock.c index e21485a..674627c 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -2571,6 +2571,7 @@ struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); + trace_sock_sndbuf_full(sk, timeo); err = -EAGAIN; if (!timeo) goto failure;
Calls to sock_alloc_send_pskb can fail or trigger a wait if there is no sndbuf space available. Add a perf trace event to help users monitor when and how this occurs so appropriate application level changes can be made, if necessary. Signed-off-by: Joe Damato <jdamato@fastly.com> --- include/trace/events/sock.h | 21 +++++++++++++++++++++ net/core/sock.c | 1 + 2 files changed, 22 insertions(+)