batman-adv: Deinline batadv_orig_hash_find, save 9024 bytes
Commit Message
This function compiles to 473 bytes of machine code.
21 callsites.
text data bss dec hex filename
95903266 20860288 35991552 152755106 91adba2 vmlinux_before
95894242 20860288 35991552 152746082 91ab862 vmlinux
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Marek Lindner <mareklindner@neomailbox.ch>
CC: Simon Wunderlich <sw@simonwunderlich.de>
CC: Antonio Quartulli <a@unstable.cc>
CC: Sven Eckelmann <sven@narfation.org>
CC: b.a.t.m.a.n@lists.open-mesh.org
CC: linux-kernel@vger.kernel.org
---
net/batman-adv/originator.c | 29 +++++++++++++++++++++++++++++
net/batman-adv/originator.h | 31 ++-----------------------------
2 files changed, 31 insertions(+), 29 deletions(-)
Comments
On Mon, Apr 25, 2016 at 03:25:22PM +0200, Denys Vlasenko wrote:
> This function compiles to 473 bytes of machine code.
> 21 callsites.
>
> text data bss dec hex filename
> 95903266 20860288 35991552 152755106 91adba2 vmlinux_before
> 95894242 20860288 35991552 152746082 91ab862 vmlinux
Hi Danys,
thanks for your patch. This function is used in a several performance critical
code paths (i.e. packet forwarding).
Are we sure we are not losing in performance here?
Cheers,
On 04/25/2016 03:39 PM, Antonio Quartulli wrote:
> On Mon, Apr 25, 2016 at 03:25:22PM +0200, Denys Vlasenko wrote:
>> This function compiles to 473 bytes of machine code.
>> 21 callsites.
>>
>> text data bss dec hex filename
>> 95903266 20860288 35991552 152755106 91adba2 vmlinux_before
>> 95894242 20860288 35991552 152746082 91ab862 vmlinux
>
> Hi Danys,
>
> thanks for your patch. This function is used in a several performance critical
> code paths (i.e. packet forwarding).
>
> Are we sure we are not losing in performance here?
Is this a common case?
if (!hash)
return NULL;
If yes, then we can inline this part only.
On Mon, Apr 25, 2016 at 03:45:20PM +0200, Denys Vlasenko wrote:
> On 04/25/2016 03:39 PM, Antonio Quartulli wrote:
> > On Mon, Apr 25, 2016 at 03:25:22PM +0200, Denys Vlasenko wrote:
> >> This function compiles to 473 bytes of machine code.
> >> 21 callsites.
> >>
> >> text data bss dec hex filename
> >> 95903266 20860288 35991552 152755106 91adba2 vmlinux_before
> >> 95894242 20860288 35991552 152746082 91ab862 vmlinux
> >
> > Hi Danys,
> >
> > thanks for your patch. This function is used in a several performance critical
> > code paths (i.e. packet forwarding).
> >
> > Are we sure we are not losing in performance here?
>
> Is this a common case?
>
> if (!hash)
> return NULL;
>
> If yes, then we can inline this part only.
Unfortunately not: this case is rather rare at runtime.
These hash tables are initialized when the batman virtual interface is created
and should be freed only upon interface shutdown.
(actually I believe this might be a good candidate for an unlikely())
Cheers,
On Monday 25 April 2016 15:25:22 Denys Vlasenko wrote:
> This function compiles to 473 bytes of machine code.
> 21 callsites.
>
> text data bss dec hex filename
> 95903266 20860288 35991552 152755106 91adba2 vmlinux_before
> 95894242 20860288 35991552 152746082 91ab862 vmlinux
>
> Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
> CC: Marek Lindner <mareklindner@neomailbox.ch>
> CC: Simon Wunderlich <sw@simonwunderlich.de>
> CC: Antonio Quartulli <a@unstable.cc>
> CC: Sven Eckelmann <sven@narfation.org>
> CC: b.a.t.m.a.n@lists.open-mesh.org
> CC: linux-kernel@vger.kernel.org
> ---
> net/batman-adv/originator.c | 29 +++++++++++++++++++++++++++++
> net/batman-adv/originator.h | 31 ++-----------------------------
> 2 files changed, 31 insertions(+), 29 deletions(-)
>
This patch should also remove following includes from originator.h:
-#include <linux/kref.h>
-#include <linux/rculist.h>
-#include <linux/rcupdate.h>
-#include <linux/stddef.h>
-#include "hash.h"
and add following includes to originator.c (please keep them in alphabetical
order):
+#include <linux/rcupdate.h>
+#include <linux/stddef.h>
Kind regards,
Sven
On Montag, 25. April 2016 21:39:50 CEST Antonio Quartulli wrote:
> On Mon, Apr 25, 2016 at 03:25:22PM +0200, Denys Vlasenko wrote:
> > This function compiles to 473 bytes of machine code.
> > 21 callsites.
> >
> > text data bss dec hex filename
> > 95903266 20860288 35991552 152755106 91adba2 vmlinux_before
> > 95894242 20860288 35991552 152746082 91ab862 vmlinux
>
> Hi Danys,
>
> thanks for your patch. This function is used in a several performance critical
> code paths (i.e. packet forwarding).
>
> Are we sure we are not losing in performance here?
Tested it with 2x OM5P-ACv2 (LEDE 17.01) which were connected via ethernet
cable on port eth1. iperf was started with reduced MSS to increase the packet
count on a PC which was connected to eth0:
$ iperf -c 192.168.10.1 -t 30 -i 1 -P8 -M 536 -y C > test01.csv
Another PC (running on the eth0 of the other device) was running the iperf-server.
The network configuration was modified to automatically enable bat0 when link was detected on eth1:
config interface 'loopback'
option ifname 'lo'
option proto 'static'
option ipaddr '127.0.0.1'
option netmask '255.0.0.0'
config globals 'globals'
option ula_prefix 'fdcb:4e2a:b274::/48'
config interface 'lan'
option type 'bridge'
option ifname 'eth0 bat0'
option proto 'static'
option ipaddr '192.168.1.2'
option netmask '255.255.255.0'
option ip6assign '60'
config interface 'batnet'
option mtu '1532'
option proto 'batadv'
option ifname 'eth1'
option mesh 'bat0'
test run | without patch (MiBit/s) | with patch (MiBit/s) | without batman-adv (MiBit/s)
---------+-------------------------+----------------------+-----------------------------
1 | 289 | 271 | 501
2 | 260 | 271 | 500
3 | 262 | 272 | 501
4 | 262 | 270 | 500
5 | 267 | 270 | 499
6 | 270 | 269 | 500
7 | 271 | 268 | 501
8 | 271 | 268 | 501
9 | 271 | 269 | 500
10 | 271 | 268 | 497
11 | 261 | 269 | 500
12 | 271 | 264 | 499
13 | 274 | 268 | 502
=========+=========================+======================+=============================
avg | 269 | 269 | 500
Kind regards,
Sven
@@ -47,6 +47,36 @@
/* hash class keys */
static struct lock_class_key batadv_orig_hash_lock_class_key;
+struct batadv_orig_node *
+batadv_orig_hash_find(struct batadv_priv *bat_priv, const void *data)
+{
+ struct batadv_hashtable *hash = bat_priv->orig_hash;
+ struct hlist_head *head;
+ struct batadv_orig_node *orig_node, *orig_node_tmp = NULL;
+ int index;
+
+ if (!hash)
+ return NULL;
+
+ index = batadv_choose_orig(data, hash->size);
+ head = &hash->table[index];
+
+ rcu_read_lock();
+ hlist_for_each_entry_rcu(orig_node, head, hash_entry) {
+ if (!batadv_compare_eth(orig_node, data))
+ continue;
+
+ if (!kref_get_unless_zero(&orig_node->refcount))
+ continue;
+
+ orig_node_tmp = orig_node;
+ break;
+ }
+ rcu_read_unlock();
+
+ return orig_node_tmp;
+}
+
static void batadv_purge_orig(struct work_struct *work);
/**
@@ -96,34 +96,7 @@ static inline u32 batadv_choose_orig(const void *data, u32 size)
return hash % size;
}
-static inline struct batadv_orig_node *
-batadv_orig_hash_find(struct batadv_priv *bat_priv, const void *data)
-{
- struct batadv_hashtable *hash = bat_priv->orig_hash;
- struct hlist_head *head;
- struct batadv_orig_node *orig_node, *orig_node_tmp = NULL;
- int index;
-
- if (!hash)
- return NULL;
-
- index = batadv_choose_orig(data, hash->size);
- head = &hash->table[index];
-
- rcu_read_lock();
- hlist_for_each_entry_rcu(orig_node, head, hash_entry) {
- if (!batadv_compare_eth(orig_node, data))
- continue;
-
- if (!kref_get_unless_zero(&orig_node->refcount))
- continue;
-
- orig_node_tmp = orig_node;
- break;
- }
- rcu_read_unlock();
-
- return orig_node_tmp;
-}
+struct batadv_orig_node *
+batadv_orig_hash_find(struct batadv_priv *bat_priv, const void *data);
#endif /* _NET_BATMAN_ADV_ORIGINATOR_H_ */