On 2019-03-24 12:31, Jesper Dangaard Brouer wrote:
> On Sun, 24 Mar 2019 07:58:34 +0100
> Felix Fietkau <n...@nbd.name> wrote:
>
>> Since we're freeing multiple skbs, we might as well use bulk free to save a
>> few cycles. Use the same conditions for bulk free as in napi_consume_skb.
>>
>> Signed-off-by: Felix Fietkau <n...@nbd.name>
>
> Thanks for working on this, it's been on my todo list for a very long
> time. I just discussed this with Florian at NetDevconf.
No problem. It was showing up on my perf traces while improving
mac80211/mt76 performance, so I decided to deal with it now :)
>> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
>> index 2415d9cb9b89..ec030ab7f1e7 100644
>> --- a/net/core/skbuff.c
>> +++ b/net/core/skbuff.c
>> @@ -666,12 +666,39 @@ EXPORT_SYMBOL(kfree_skb);
>>
>> void kfree_skb_list(struct sk_buff *segs)
>> {
>> - while (segs) {
>> - struct sk_buff *next = segs->next;
>> + struct sk_buff *next = segs;
>> + void *skbs[16];
>> + int n_skbs = 0;
>>
>> - kfree_skb(segs);
>> - segs = next;
>> + while ((segs = next) != NULL) {
>> + next = segs->next;
>> +
>> + if (!skb_unref(segs))
>> + continue;
>> +
>> + if (segs->fclone != SKB_FCLONE_UNAVAILABLE ||
>> + n_skbs >= ARRAY_SIZE(skbs)) {
>
> You could call kmem_cache_free_bulk() here and reset n_skbs=0.
Sure, good idea. I'll send v2 shortly.
- Felix
