On Wed, Jun 08, 2011 at 08:45:38AM +0900, KAMEZAWA Hiroyuki wrote:
> On Tue, 7 Jun 2011 12:26:11 +0200
> Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> > On Tue, Jun 07, 2011 at 04:55:37PM +0900, KAMEZAWA Hiroyuki wrote:
> > > On Tue, 7 Jun 2011 09:51:31 +0200
> > > Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> > > > @@ -283,23 +285,30 @@ static int __meminit page_cgroup_callback(struct notifier_block *self,
> > > >
> > > > void __init page_cgroup_init(void)
> > > > {
> > > > - unsigned long pfn;
> > > > - int fail = 0;
> > > > + pg_data_t *pgdat;
> > > >
> > > > if (mem_cgroup_disabled())
> > > > return;
> > > >
> > > > - for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) {
> > > > - if (!pfn_present(pfn))
> > > > - continue;
> > > > - fail = init_section_page_cgroup(pfn);
> > > > - }
> > > > - if (fail) {
> > > > - printk(KERN_CRIT "try 'cgroup_disable=memory' boot option\n");
> > > > - panic("Out of memory");
> > > > - } else {
> > > > - hotplug_memory_notifier(page_cgroup_callback, 0);
> > > > + for_each_online_pgdat(pgdat) {
> > > > + unsigned long start;
> > > > + unsigned long end;
> > > > + unsigned long pfn;
> > > > +
> > > > + start = pgdat->node_start_pfn & ~(PAGES_PER_SECTION - 1);
> > > > + end = ALIGN(pgdat->node_start_pfn + pgdat->node_spanned_pages,
> > > > + PAGES_PER_SECTION);
> > > > + for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
> > > > + if (!pfn_present(pfn))
> > > > + continue;
> > > > + if (!init_section_page_cgroup(pgdat->node_id, pfn))
> > > > + continue;
> > >
> > > AFAIK, nodes can overlap. So, this [start, end) scan doesn't work. sections
> > > may be initizalised mulitple times ...in wrong way. At here, what we can trust
> > > is nid in page->flags or early_node_map[]?.
> >
> > Sections are not be initialized multiple times. Once their
> > page_cgroup array is allocated they are skipped if considered again
> > later.
> >
> > Second, even if there are two nodes backing the memory of a single
> > section, there is still just a single page_cgroup array per section,
> > we have to take the memory from one node or the other.
> >
> > So if both node N1 and N2 fall into section SN, SN->page_cgroup will
> > be an array of page_cgroup structures, allocated on N1, to represent
> > the pages of SN.
> >
> > The first section considered when walking the PFNs of N2 will be SN,
> > which will be skipped because of !!SN->page_cgroup.
> >
> > I do not see the problem.
> >
>
> Assume a host with node layout
>
> N0 | N1 | N2 | N3 | N4 | N0
>
>
> pages for all node will be allocated onto node-0.
That makes sense, thanks for the explanation!
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>