2019-10-31 07:47:28

by Christophe JAILLET

[permalink] [raw]
Subject: [PATCH] batman-adv: Simplify 'batadv_v_ogm_aggr_list_free()'

Use 'skb_queue_purge()' instead of re-implementing it.

Signed-off-by: Christophe JAILLET <[email protected]>
---
BTW, I don't really see the need of 'aggr_list_lock'. I think that the code
could be refactored to drop 'aggr_list_lock' and use the already existing
'aggr_list.lock'.
This would require to use the lock-free __skb_... variants when working on
'aggr_list'.

As far as I understand, the use of 'aggr_list' and 'aggr_list_lock' is
limited to bat_v_ogm.c'. So the impact would be limited.
This would avoid a useless locking that never fails, so the performance
gain should be really limited.

So, I'm not sure this would be more readable and/or future proof, so
I just note it here to open the discussion.

If interested, I have a (compiled tested only) patch that implements this
change.
---
net/batman-adv/bat_v_ogm.c | 6 +-----
1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/net/batman-adv/bat_v_ogm.c b/net/batman-adv/bat_v_ogm.c
index dc4f7430cb5a..b841c83d9c3b 100644
--- a/net/batman-adv/bat_v_ogm.c
+++ b/net/batman-adv/bat_v_ogm.c
@@ -177,13 +177,9 @@ static bool batadv_v_ogm_queue_left(struct sk_buff *skb,
*/
static void batadv_v_ogm_aggr_list_free(struct batadv_hard_iface *hard_iface)
{
- struct sk_buff *skb;
-
lockdep_assert_held(&hard_iface->bat_v.aggr_list_lock);

- while ((skb = skb_dequeue(&hard_iface->bat_v.aggr_list)))
- kfree_skb(skb);
-
+ skb_queue_purge(&hard_iface->bat_v.aggr_list);
hard_iface->bat_v.aggr_len = 0;
}

--
2.20.1


2019-10-31 08:01:39

by Sven Eckelmann

[permalink] [raw]
Subject: Re: [PATCH] batman-adv: Simplify 'batadv_v_ogm_aggr_list_free()'

On Thursday, 31 October 2019 08:42:55 CET Christophe JAILLET wrote:
> Use 'skb_queue_purge()' instead of re-implementing it.
>
> Signed-off-by: Christophe JAILLET <[email protected]>

Consider this patch applied. I just have to leave now and thus I will only
apply after my return.

> ---
> BTW, I don't really see the need of 'aggr_list_lock'. I think that the code
> could be refactored to drop 'aggr_list_lock' and use the already existing
> 'aggr_list.lock'.
> This would require to use the lock-free __skb_... variants when working on
> 'aggr_list'.
>
> As far as I understand, the use of 'aggr_list' and 'aggr_list_lock' is
> limited to bat_v_ogm.c'. So the impact would be limited.
> This would avoid a useless locking that never fails, so the performance
> gain should be really limited.
>
> So, I'm not sure this would be more readable and/or future proof, so
> I just note it here to open the discussion.
>
> If interested, I have a (compiled tested only) patch that implements this
> change.

Yes, please send it over.

Kind regards,
Sven


Attachments:
signature.asc (849.00 B)
This is a digitally signed message part.