Messages in this thread | | | Date | Mon, 26 Sep 2011 12:28:14 +0300 | Subject | Re: [PATCH 5/5] slub: Only IPI CPUs that have per cpu obj to flush | From | Pekka Enberg <> |
| |
On Mon, Sep 26, 2011 at 11:35 AM, Gilad Ben-Yossef <gilad@benyossef.com> wrote: >> Right, having to do that for_each_oneline_cpu() loop only to then IPI >> them can cause a massive cacheline bounce fest.. Ideally you'd want to >> keep a cpumask per kmem_cache, although I bet the memory overhead of >> that isn't attractive. >> >> Also, what Pekka says, having that alloc here isn't good either. > > Yes, the alloc in the flush_all path definitively needs to go. I > wonder if just to resolve that allocating the mask per cpu and not in > kmem_cache itself is not better - after all, all we need is a single > mask per cpu when we wish to do a flush_all and no per cache. The > memory overhead of that is slightly better. This doesn't cover the > cahce bounce issue.
I'm fine with whatever works for you as long as you don't add a kmalloc() call in flush_all().
Pekka
| |