On Wed, Oct 03, 2018 at 11:38:54AM +0200, Lorenzo Bianconi wrote: > > +static DEFINE_PER_CPU(struct page_frag_cache, mt76_frag_cache); > > + > > +void *mt76_alloc_frag(unsigned int fragsz) > > +{ > > + struct page_frag_cache *fc; > > + unsigned long flags; > > + void *data; > > + > > + local_irq_save(flags); > > I like this approach since we will avoid a cache miss for the spinlock :) > Do we still need to disable local_irq here since (not considering fw upload) > I guess there is no contention for mt76_frag_cache I think is needed if we have more than one device, but I think we can change to local_irq_disable() / local_irq_enable() . Thanks Stanislaw