Netdev List
 help / color / mirror / Atom feed
From: Eric Dumazet <eric.dumazet@gmail.com>
To: David Miller <davem@davemloft.net>
Cc: netdev <netdev@vger.kernel.org>,
	Tom Herbert <therbert@google.com>,
	Mahesh Bandewar <maheshb@google.com>
Subject: [PATCH] net: remove delay at device dismantle
Date: Fri, 10 Aug 2012 16:46:06 +0200	[thread overview]
Message-ID: <1344609966.31104.2705.camel@edumazet-glaptop> (raw)
In-Reply-To: <1344596809.31104.2358.camel@edumazet-glaptop>

From: Eric Dumazet <edumazet@google.com>

I noticed extra one second delay in device dismantle, tracked down to
a call to dst_dev_event() while some call_rcu() are still in RCU queues.

These call_rcu() were posted by rt_free(struct rtable *rt) calls.

We then wait a little (but one second) in netdev_wait_allrefs() before
kicking again NETDEV_UNREGISTER.

As the call_rcu() are now completed, dst_dev_event() can do the needed
device swap on busy dst.

To solve this problem, add a new NETDEV_UNREGISTER_FINAL, called
in rollback_registered_many() after a rcu_barrier().

Change dst_dev_event() handler to react to NETDEV_UNREGISTER_FINAL

Also remove NETDEV_UNREGISTER_BATCH, as its not used anymore after
IP cache removal.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
---
 include/linux/netdevice.h |    2 +-
 net/core/dev.c            |   23 +++++++----------------
 net/core/dst.c            |    2 +-
 net/core/rtnetlink.c      |    2 +-
 net/ipv4/fib_frontend.c   |    2 --
 5 files changed, 10 insertions(+), 21 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index a9db4f3..e3a43fd 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1550,7 +1550,7 @@ struct packet_type {
 #define NETDEV_PRE_TYPE_CHANGE	0x000E
 #define NETDEV_POST_TYPE_CHANGE	0x000F
 #define NETDEV_POST_INIT	0x0010
-#define NETDEV_UNREGISTER_BATCH 0x0011
+#define NETDEV_UNREGISTER_FINAL 0x0011
 #define NETDEV_RELEASE		0x0012
 #define NETDEV_NOTIFY_PEERS	0x0013
 #define NETDEV_JOIN		0x0014
diff --git a/net/core/dev.c b/net/core/dev.c
index a39354e..be3a5af 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1394,7 +1394,6 @@ rollback:
 				nb->notifier_call(nb, NETDEV_DOWN, dev);
 			}
 			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
-			nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
 		}
 	}
 
@@ -1436,7 +1435,6 @@ int unregister_netdevice_notifier(struct notifier_block *nb)
 				nb->notifier_call(nb, NETDEV_DOWN, dev);
 			}
 			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
-			nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
 		}
 	}
 unlock:
@@ -5307,14 +5305,14 @@ static void rollback_registered_many(struct list_head *head)
 		netdev_unregister_kobject(dev);
 	}
 
-	/* Process any work delayed until the end of the batch */
-	dev = list_first_entry(head, struct net_device, unreg_list);
-	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
-
 	synchronize_net();
+	/* Wait for rcu callbacks to finish before next phase */
+	rcu_barrier();
 
-	list_for_each_entry(dev, head, unreg_list)
+	list_for_each_entry(dev, head, unreg_list) {
+		call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
 		dev_put(dev);
+	}
 }
 
 static void rollback_registered(struct net_device *dev)
@@ -5757,9 +5755,8 @@ static void netdev_wait_allrefs(struct net_device *dev)
 
 			/* Rebroadcast unregister notification */
 			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
-			/* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
-			 * should have already handle it the first time */
-
+			rcu_barrier();
+			call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
 			if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
 				     &dev->state)) {
 				/* We must not have linkwatch events
@@ -5821,11 +5818,6 @@ void netdev_run_todo(void)
 
 	__rtnl_unlock();
 
-	/* Wait for rcu callbacks to finish before attempting to drain
-	 * the device list.  This usually avoids a 250ms wait.
-	 */
-	if (!list_empty(&list))
-		rcu_barrier();
 
 	while (!list_empty(&list)) {
 		struct net_device *dev
@@ -6226,7 +6218,6 @@ int dev_change_net_namespace(struct net_device *dev, struct net *net, const char
 	   the device is just moving and can keep their slaves up.
 	*/
 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
-	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
 	rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
 
 	/*
diff --git a/net/core/dst.c b/net/core/dst.c
index 56d6361..f6593d2 100644
--- a/net/core/dst.c
+++ b/net/core/dst.c
@@ -374,7 +374,7 @@ static int dst_dev_event(struct notifier_block *this, unsigned long event,
 	struct dst_entry *dst, *last = NULL;
 
 	switch (event) {
-	case NETDEV_UNREGISTER:
+	case NETDEV_UNREGISTER_FINAL:
 	case NETDEV_DOWN:
 		mutex_lock(&dst_gc_mutex);
 		for (dst = dst_busy_list; dst; dst = dst->next) {
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index 2c5a0a0..30f5a06 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -2356,7 +2356,7 @@ static int rtnetlink_event(struct notifier_block *this, unsigned long event, voi
 	case NETDEV_PRE_TYPE_CHANGE:
 	case NETDEV_GOING_DOWN:
 	case NETDEV_UNREGISTER:
-	case NETDEV_UNREGISTER_BATCH:
+	case NETDEV_UNREGISTER_FINAL:
 	case NETDEV_RELEASE:
 	case NETDEV_JOIN:
 		break;
diff --git a/net/ipv4/fib_frontend.c b/net/ipv4/fib_frontend.c
index c43ae3f..4fe50184 100644
--- a/net/ipv4/fib_frontend.c
+++ b/net/ipv4/fib_frontend.c
@@ -1071,8 +1071,6 @@ static int fib_netdev_event(struct notifier_block *this, unsigned long event, vo
 	case NETDEV_CHANGE:
 		rt_cache_flush(dev_net(dev), 0);
 		break;
-	case NETDEV_UNREGISTER_BATCH:
-		break;
 	}
 	return NOTIFY_DONE;
 }

  parent reply	other threads:[~2012-08-10 14:46 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-08-10  9:27 [RFC] NETDEV_UNREGISTER_BATCH seems unused nowaday ? Eric Dumazet
2012-08-10 10:42 ` David Miller
2012-08-10 11:06   ` Eric Dumazet
2012-08-10 14:45     ` Eric W. Biederman
2012-08-10 15:01       ` Eric Dumazet
2012-08-10 17:55       ` Paul E. McKenney
2012-08-10 14:46     ` Eric Dumazet [this message]
2012-08-10 15:42       ` [PATCH net-next v2] net: remove delay at device dismantle Eric Dumazet
2012-08-11  5:54         ` Eric Dumazet
2012-08-11  5:57           ` David Miller
2012-08-23  2:18           ` David Miller
2012-08-23  2:25           ` Gao feng
2012-08-23  2:51             ` David Miller
2012-08-23  2:58               ` Eric Dumazet

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1344609966.31104.2705.camel@edumazet-glaptop \
    --to=eric.dumazet@gmail.com \
    --cc=davem@davemloft.net \
    --cc=maheshb@google.com \
    --cc=netdev@vger.kernel.org \
    --cc=therbert@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox