All of lore.kernel.org
 help / color / mirror / Atom feed
From: Andi Kleen <ak@linux.intel.com>
To: Fengguang Wu <fengguang.wu@intel.com>
Cc: Linux-NFS <linux-nfs@vger.kernel.org>, Trond.Myklebust@netapp.com
Subject: Re: rpcauth_lookup_credcache() lock contentions
Date: Wed, 27 Jun 2012 11:03:53 -0700	[thread overview]
Message-ID: <20120627180353.GO4152@tassilo.jf.intel.com> (raw)
In-Reply-To: <20120625024211.GA11057@localhost>

> > Can you try this patch?
> 
> Thank you! This patch brings %sys down to 24%, a pretty large improvement!

I redid the patch now and fixed the race Trond pointed out.

Fengguang, can you retest please? Trond, does it look good now?

---

From: Andi Kleen <ak@linux.intel.com>
Date: Sun, 24 Jun 2012 14:31:06 -0700
Subject: [PATCH] sunrpc: Improve spinlocks in credential lookup path v2

Fengguang noticed that rpcauth_lookup_credcache has high lock contention
on the nfs server when doing kernel builds on nfsroot.

The only reason the spinlock is needed in the lockup path is to
synchronize with the garbage collector. First the object itself
is stable because it's RCU'ed.

So now we use an atomic_add_return and check for the 0 reference
count case.  When the count was 0 assume the garbage collector
is active on this entry and take the lock and recheck.

Otherwise the path is lockless.

v2: Add GC race check based on Trond's feedback
Cc: Trond.Myklebust@netapp.com
Reported-by: Fengguang Wu
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 net/sunrpc/auth.c |   19 +++++++++++++++----
 1 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/net/sunrpc/auth.c b/net/sunrpc/auth.c
index 727e506..645769e 100644
--- a/net/sunrpc/auth.c
+++ b/net/sunrpc/auth.c
@@ -364,13 +364,24 @@ rpcauth_lookup_credcache(struct rpc_auth *auth, struct auth_cred * acred,
 	hlist_for_each_entry_rcu(entry, pos, &cache->hashtable[nr], cr_hash) {
 		if (!entry->cr_ops->crmatch(acred, entry, flags))
 			continue;
-		spin_lock(&cache->lock);
 		if (test_bit(RPCAUTH_CRED_HASHED, &entry->cr_flags) == 0) {
-			spin_unlock(&cache->lock);
 			continue;
 		}
-		cred = get_rpccred(entry);
-		spin_unlock(&cache->lock);
+		cred = entry;
+		if (atomic_add_return(1, &cred->cr_count) == 1) {
+			/*
+			 * When the count was 0 we could race with the GC.
+			 * Take the lock and recheck.
+			 */
+			spin_lock(&cache->lock);
+			if (!test_bit(RPCAUTH_CRED_HASHED, &entry->cr_flags)) {
+				/* Lost the race. */
+				atomic_dec(&cred->cr_count);
+				spin_unlock(&cache->lock);
+				continue;
+			}
+			spin_unlock(&cache->lock);
+		}
 		break;
 	}
 	rcu_read_unlock();
-- 
1.7.7


  reply	other threads:[~2012-06-27 18:03 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20120623122604.GA10887@localhost>
2012-06-23 15:07 ` rpcauth_lookup_credcache() lock contentions Fengguang Wu
2012-06-24 21:34 ` Andi Kleen
2012-06-25  1:21   ` Myklebust, Trond
2012-06-25  2:45     ` Andi Kleen
2012-06-25  2:42   ` Fengguang Wu
2012-06-27 18:03     ` Andi Kleen [this message]
2012-06-27 18:36       ` Myklebust, Trond
2012-07-05 13:11       ` Fengguang Wu
2012-07-05 15:05         ` Malahal Naineni
2012-07-05 16:29           ` Andi Kleen
2012-07-05 16:27         ` Andi Kleen

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=20120627180353.GO4152@tassilo.jf.intel.com \
    --to=ak@linux.intel.com \
    --cc=Trond.Myklebust@netapp.com \
    --cc=fengguang.wu@intel.com \
    --cc=linux-nfs@vger.kernel.org \
    /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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.