Re: [PATCH 3/3] mem-hotplug: fix potential race while buildingzonelist for new populated zone

[Wu Fengguang]

On Tue, May 18, 2010 at 12:09:31AM +0800, Christoph Lameter wrote:
> On Mon, 17 May 2010, Haicheng Li wrote:
> 
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 72c1211..0729a82 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -2783,6 +2783,20 @@ static __init_refok int __build_all_zonelists(void
> > *data)
> >  {
> >  	int nid;
> >  	int cpu;
> > +#ifdef CONFIG_MEMORY_HOTPLUG
> > +	struct zone_online_info *new = (struct zone_online_info *)data;
> > +
> > +	/*
> > +	 * Populate the new zone before build zonelists, which could
> > +	 * happen only when onlining a new node after system is booted.
> > +	 */
> > +	if (new) {
> > +		/* We are expecting a new memory block here. */
> > +		WARN_ON(!new->onlined_pages);
> > +		new->zone->present_pages += new->onlined_pages;
> > +		new->zone->zone_pgdat->node_present_pages +=
> > new->onlined_pages;
> > +	}
> > +#endif
> 
> 
> Building a zonelist now has the potential side effect of changes to the
> size of the zone?

Yeah, this sounds a bit hacky.

> Can we have a global mutex that protects against size modification of
> zonelists instead? And it could also serialize the pageset setup?

Good suggestion. We could make zone_pageset_mutex a global mutex and
take it in all the functions that call build_all_zonelists() --
currently only online_pages() and numa_zonelist_order_handler().

This can equally fix the possible race:

    CPU0                                    CPU1                            CPU2
(1) zone->present_pages += online_pages;
(2)                                         build_all_zonelists();
(3)                                                                 alloc_page();
(4)                                                                 free_page();
(5) build_all_zonelists();
(6)   __build_all_zonelists();
(7)     zone->pageset = alloc_percpu();

In step (3,4), zone->pageset still points to boot_pageset, so bad
things may happen if 2+ nodes are in this state. Even if only 1 node
is accessing the boot_pageset, (3) may still consume too much memory
to fail the memory allocations in step (7).

Thanks,
Fengguang
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/