Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1471344 > unrolled thread

[PATCH 3/4] net/netfilter/nf_conntrack_core: update memory barriers.

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH 3/4] net/netfilter/nf_conntrack_core: update memory barriers. Manfred Spraul <manfred@colorfullife.com> - 2016-08-28 14:00 +0200
    Re: [PATCH 3/4] net/netfilter/nf_conntrack_core: update memory  barriers. Peter Zijlstra <peterz@infradead.org> - 2016-08-29 13:00 +0200

#1471344 — [PATCH 3/4] net/netfilter/nf_conntrack_core: update memory barriers.

Change the locking for nf_conntrack_lock and nf_conntract_lock_all()
to the approach used by ipc/sem.c:

- use smp_store_mb() instead of a raw smp_mb()
- use smp_mb__after_spin_lock().
- for nf_conntrack_lock, use spin_lock(&global_lock) instead of
  spin_unlock_wait(&global_lock) and loop backward.

The last change avoids that nf_conntrack_lock() could loop multiple times.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
---
 net/netfilter/nf_conntrack_core.c | 36 ++++++++++++++++++++++--------------
 1 file changed, 22 insertions(+), 14 deletions(-)

diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c
index dd2c43a..ded1adc 100644
--- a/net/netfilter/nf_conntrack_core.c
+++ b/net/netfilter/nf_conntrack_core.c
@@ -80,20 +80,29 @@ static __read_mostly bool nf_conntrack_locks_all;
 
 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
 {
+	/* Step 1: Acquire the lock */
 	spin_lock(lock);
-	while (unlikely(nf_conntrack_locks_all)) {
-		spin_unlock(lock);
 
-		/*
-		 * Order the 'nf_conntrack_locks_all' load vs. the
-		 * spin_unlock_wait() loads below, to ensure
-		 * that 'nf_conntrack_locks_all_lock' is indeed held:
-		 */
-		smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
-		spin_unlock_wait(&nf_conntrack_locks_all_lock);
-		spin_lock(lock);
-	}
+	/* Step 2: make it visible to all CPUs that we hold the lock */
+	smp_mb__after_spin_lock();
+
+	/* Step 3: read locks_all, with ACQUIRE semantics */
+	if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
+		return;
+
+	/* slow path: unlock */
+	spin_unlock(lock);
+
+	/* Slow path, step 1: get global lock */
+	spin_lock(&nf_conntrack_locks_all_lock);
+
+	/* Slow path, step 2: get the lock we want */
+	spin_lock(lock);
+
+	/* Slow path, step 3: release the global lock */
+	spin_unlock(&nf_conntrack_locks_all_lock);
 }
+
 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 
 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
@@ -133,15 +142,14 @@ static void nf_conntrack_all_lock(void)
 	int i;
 
 	spin_lock(&nf_conntrack_locks_all_lock);
-	nf_conntrack_locks_all = true;
 
 	/*
-	 * Order the above store of 'nf_conntrack_locks_all' against
+	 * Order the store of 'nf_conntrack_locks_all' against
 	 * the spin_unlock_wait() loads below, such that if
 	 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
 	 * we must observe nf_conntrack_locks[] held:
 	 */
-	smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
+	smp_store_mb(nf_conntrack_locks_all, true);
 
 	for (i = 0; i < CONNTRACK_LOCKS; i++) {
 		spin_unlock_wait(&nf_conntrack_locks[i]);
-- 
2.5.5

[toc] | [next] | [standalone]

#1471719 — Re: [PATCH 3/4] net/netfilter/nf_conntrack_core: update memory barriers.

On Sun, Aug 28, 2016 at 01:56:15PM +0200, Manfred Spraul wrote:
> +++ b/net/netfilter/nf_conntrack_core.c
> @@ -80,20 +80,29 @@ static __read_mostly bool nf_conntrack_locks_all;
>  
>  void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
>  {
> +	/* Step 1: Acquire the lock */
>  	spin_lock(lock);
> -	while (unlikely(nf_conntrack_locks_all)) {
> -		spin_unlock(lock);
>  
> -		/*
> -		 * Order the 'nf_conntrack_locks_all' load vs. the
> -		 * spin_unlock_wait() loads below, to ensure
> -		 * that 'nf_conntrack_locks_all_lock' is indeed held:
> -		 */
> -		smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
> -		spin_unlock_wait(&nf_conntrack_locks_all_lock);
> -		spin_lock(lock);
> -	}
> +	/* Step 2: make it visible to all CPUs that we hold the lock */
> +	smp_mb__after_spin_lock();

I hate this comment. A barrier does _not_ make visible anything.

A barrier _orders_ things.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web