| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20090714183308.GB3090@ami.dom.local>
Date: Tue, 14 Jul 2009 20:33:08 +0200
From: Jarek Poplawski <jarkao2@...il.com>
To: David Miller <davem@...emloft.net>
Cc: "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>,
Paweł Staszewski <pstaszewski@...are.pl>,
Linux Network Development list <netdev@...r.kernel.org>,
Robert Olsson <robert@...ur.slu.se>
Subject: [PATCH net-next] Re: rib_trie / Fix inflate_threshold_root. Now=15
size=11 bits
On Sun, Jul 05, 2009 at 02:32:32PM -0700, Paul E. McKenney wrote:
> On Sun, Jul 05, 2009 at 07:32:08PM +0200, Jarek Poplawski wrote:
> > On Sun, Jul 05, 2009 at 06:20:03PM +0200, Jarek Poplawski wrote:
> > > On Sun, Jul 05, 2009 at 02:30:03AM +0200, Paweł Staszewski wrote:
> > > > Oh
> > > >
> > > > I forgot - please Jarek give me patch with sync rcu and i will make test
> > > > on preempt kernel
> > >
> > > Probably non-preempt kernel might need something like this more, but
> > > comparing is always interesting. This patch is based on Paul's
> > > suggestion (I hope).
> >
> > Hold on ;-) Here is something even better... Syncing after 128 pages
> > might be still too slow, so here is a higher initial value, 1000, plus
> > you can change this while testing in:
> >
> > /sys/module/fib_trie/parameters/sync_pages
> >
> > It would be interesting to find the lowest acceptable value.
>
> Looks like a promising approach to me!
>
> Thanx, Paul
Below is a simpler version of this patch, without the sysfs parameter.
(I left the previous version quoted for comparison.) Thanks.
> > Jarek P.
> > ---> (synchronize take 8; apply on top of the 2.6.29.x with the last
> > all-in-one patch, or net-2.6)
> >
> > net/ipv4/fib_trie.c | 12 ++++++++++++
> > 1 files changed, 12 insertions(+), 0 deletions(-)
> >
> > diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
> > index 00a54b2..decc8d0 100644
> > --- a/net/ipv4/fib_trie.c
> > +++ b/net/ipv4/fib_trie.c
> > @@ -71,6 +71,7 @@
> > #include <linux/netlink.h>
> > #include <linux/init.h>
> > #include <linux/list.h>
> > +#include <linux/moduleparam.h>
> > #include <net/net_namespace.h>
> > #include <net/ip.h>
> > #include <net/protocol.h>
> > @@ -164,6 +165,10 @@ static struct tnode *inflate(struct trie *t, struct tnode *tn);
> > static struct tnode *halve(struct trie *t, struct tnode *tn);
> > /* tnodes to free after resize(); protected by RTNL */
> > static struct tnode *tnode_free_head;
> > +static size_t tnode_free_size;
> > +
> > +static int sync_pages __read_mostly = 1000;
> > +module_param(sync_pages, int, 0640);
> >
> > static struct kmem_cache *fn_alias_kmem __read_mostly;
> > static struct kmem_cache *trie_leaf_kmem __read_mostly;
> > @@ -393,6 +398,8 @@ static void tnode_free_safe(struct tnode *tn)
> > BUG_ON(IS_LEAF(tn));
> > tn->tnode_free = tnode_free_head;
> > tnode_free_head = tn;
> > + tnode_free_size += sizeof(struct tnode) +
> > + (sizeof(struct node *) << tn->bits);
> > }
> >
> > static void tnode_free_flush(void)
> > @@ -404,6 +411,11 @@ static void tnode_free_flush(void)
> > tn->tnode_free = NULL;
> > tnode_free(tn);
> > }
> > +
> > + if (tnode_free_size >= PAGE_SIZE * sync_pages) {
> > + tnode_free_size = 0;
> > + synchronize_rcu();
> > + }
> > }
> >
> > static struct leaf *leaf_new(void)
> > --
------------------------>
ipv4: Use synchronize_rcu() during trie_rebalance()
During trie_rebalance() we free memory after resizing with call_rcu(),
but large updates, especially with PREEMPT_NONE configs, can cause
memory stresses, so this patch calls synchronize_rcu() in
tnode_free_flush() after each sync_pages to guarantee such freeing
(especially before resizing the root node).
The value of sync_pages = 128 is based on Pawel Staszewski's tests as
the lowest which doesn't hinder updating times. (For testing purposes
there was a sysfs module parameter to change it on demand, but it's
removed until we're sure it could be really useful.)
The patch is based on suggestions by: Paul E. McKenney
<paulmck@...ux.vnet.ibm.com>
Reported-by: Pawel Staszewski <pstaszewski@...are.pl>
Tested-by: Pawel Staszewski <pstaszewski@...are.pl>
Signed-off-by: Jarek Poplawski <jarkao2@...il.com>
---
net/ipv4/fib_trie.c | 15 +++++++++++++++
1 files changed, 15 insertions(+), 0 deletions(-)
diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
index 63c2fa7..58ba9f4 100644
--- a/net/ipv4/fib_trie.c
+++ b/net/ipv4/fib_trie.c
@@ -164,6 +164,14 @@ static struct tnode *inflate(struct trie *t, struct tnode *tn);
static struct tnode *halve(struct trie *t, struct tnode *tn);
/* tnodes to free after resize(); protected by RTNL */
static struct tnode *tnode_free_head;
+static size_t tnode_free_size;
+
+/*
+ * synchronize_rcu after call_rcu for that many pages; it should be especially
+ * useful before resizing the root node with PREEMPT_NONE configs; the value was
+ * obtained experimentally, aiming to avoid visible slowdown.
+ */
+static const int sync_pages = 128;
static struct kmem_cache *fn_alias_kmem __read_mostly;
static struct kmem_cache *trie_leaf_kmem __read_mostly;
@@ -393,6 +401,8 @@ static void tnode_free_safe(struct tnode *tn)
BUG_ON(IS_LEAF(tn));
tn->tnode_free = tnode_free_head;
tnode_free_head = tn;
+ tnode_free_size += sizeof(struct tnode) +
+ (sizeof(struct node *) << tn->bits);
}
static void tnode_free_flush(void)
@@ -404,6 +414,11 @@ static void tnode_free_flush(void)
tn->tnode_free = NULL;
tnode_free(tn);
}
+
+ if (tnode_free_size >= PAGE_SIZE * sync_pages) {
+ tnode_free_size = 0;
+ synchronize_rcu();
+ }
}
static struct leaf *leaf_new(void)
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists