Messages in this thread | | | Date | Thu, 24 Feb 2011 19:02:47 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | Re: [RFC PATCH] page_cgroup: Reduce allocation overhead for page_cgroup array for CONFIG_SPARSEMEM |
| |
On Thu, 24 Feb 2011 10:35:19 +0100 Michal Hocko <mhocko@suse.cz> wrote:
> On Thu 24-02-11 08:52:27, KAMEZAWA Hiroyuki wrote: > > On Wed, 23 Feb 2011 10:19:22 -0800 > > Dave Hansen <dave@linux.vnet.ibm.com> wrote: > > > > > On Wed, 2011-02-23 at 16:10 +0100, Michal Hocko wrote: > > > > We can reduce this internal fragmentation by splitting the single > > > > page_cgroup array into more arrays where each one is well kmalloc > > > > aligned. This patch implements this idea. > > > > > > How about using alloc_pages_exact()? These things aren't allocated > > > often enough to really get most of the benefits of being in a slab. > > > That'll at least get you down to a maximum of about PAGE_SIZE wasted. > > > > > > > yes, alloc_pages_exact() is much better. > > > > packing page_cgroups for multiple sections causes breakage in memory hotplug logic. > > I am not sure I understand this. What do you mean by packing > page_cgroups for multiple sections? The patch I have posted doesn't do > any packing. Or do you mean that using a double array can break hotplog? > Not that this would change anything, alloc_pages_exact is really a > better solution, I am just curious ;) >
Sorry, it seems I failed to read code correctly. You just implemented 2 level table..
Thanks, -Kame
| |