From: Arjan van de Ven The skb_queue_head_init() function is used both in drivers for private use and in the core networking code. The usage models are vastly set of functions that is only softirq safe; while the driver usage tends to be more limited to a few hardirq safe accessor functions. Rather than annotating all 133+ driver usages, for now just split this lock into a per queue class. This change is obviously safe and probably should make 2.6.18. Signed-off-by: Arjan van de Ven Signed-off-by: Ingo Molnar Signed-off-by: Andrew Morton --- include/linux/skbuff.h | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff -puN include/linux/skbuff.h~lockdep-split-the-skb_queue_head_init-lock-class include/linux/skbuff.h --- a/include/linux/skbuff.h~lockdep-split-the-skb_queue_head_init-lock-class +++ a/include/linux/skbuff.h @@ -606,10 +606,17 @@ static inline __u32 skb_queue_len(const extern struct lock_class_key skb_queue_lock_key; +/* + * This function creates a split out lock class for each invocation; + * this is needed for now since a whole lot of users of the skb-queue + * infrastructure in drivers have different locking usage (in hardirq) + * than the networking core (in softirq only). In the long run either the + * network layer or drivers should need annotation to consolidate the + * main types of usage into 3 classes. + */ static inline void skb_queue_head_init(struct sk_buff_head *list) { spin_lock_init(&list->lock); - lockdep_set_class(&list->lock, &skb_queue_lock_key); list->prev = list->next = (struct sk_buff *)list; list->qlen = 0; } _