Messages in this thread | | | Date | Tue, 3 Aug 2010 13:25:59 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | Re: [PATCH -mm 5/5] memcg: use spinlock in page_cgroup instead of bit_spinlock |
| |
On Tue, 3 Aug 2010 09:36:45 +0530 Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> * KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2010-08-02 19:20:06]: > > > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> > > > > This patch replaces bit_spinlock with spinlock. In general, > > spinlock has good functinality than bit_spin_lock and we should use > > it if we have a room for it. In 64bit arch, we have extra 4bytes. > > Let's use it. > > expected effects: > > - use better codes. > > - ticket lock on x86-64 > > - para-vitualization aware lock > > etc.. > > > > Chagelog: 20090729 > > - fixed page_cgroup_is_locked(). > > > > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> > > -- > > The additional space usage is a big concern, I think saving space > would be of highest priority. I understand the expected benefits, but > a spinlock_t per page_cgroup is quite expensive at the moment. If > anything I think it should be a config option under CONFIG_DEBUG or > something else to play with and see the side effects. >
Hmm. As I already wrote, packing id to flags is not easy. leave 4 bytes space _pad for a while and drop this patch ?
I don't like to add CONFIG_DEBUG in this core.
Thanks, -Kame
| |