Re: [PATCH 03/10] soft-offline: use migrate_pages() instead of migrate_huge_page()

[Michal Hocko <mhocko@suse.cz>]

On Mon 01-04-13 10:43:14, Aneesh Kumar K.V wrote:
> Michal Hocko <mhocko@suse.cz> writes:
> 
> > On Tue 26-03-13 16:59:40, Aneesh Kumar K.V wrote:
> >> Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> writes:
> > [...]
> >> > diff --git v3.9-rc3.orig/mm/memory-failure.c v3.9-rc3/mm/memory-failure.c
> >> > index df0694c..4e01082 100644
> >> > --- v3.9-rc3.orig/mm/memory-failure.c
> >> > +++ v3.9-rc3/mm/memory-failure.c
> >> > @@ -1467,6 +1467,7 @@ static int soft_offline_huge_page(struct page *page, int flags)
> >> >  	int ret;
> >> >  	unsigned long pfn = page_to_pfn(page);
> >> >  	struct page *hpage = compound_head(page);
> >> > +	LIST_HEAD(pagelist);
> >> >
> >> >  	/*
> >> >  	 * This double-check of PageHWPoison is to avoid the race with
> >> > @@ -1482,12 +1483,20 @@ static int soft_offline_huge_page(struct page *page, int flags)
> >> >  	unlock_page(hpage);
> >> >
> >> >  	/* Keep page count to indicate a given hugepage is isolated. */
> >> > -	ret = migrate_huge_page(hpage, new_page, MPOL_MF_MOVE_ALL,
> >> > -				MIGRATE_SYNC);
> >> > -	put_page(hpage);
> >> > +	list_move(&hpage->lru, &pagelist);
> >> 
> >> we use hpage->lru to add the hpage to h->hugepage_activelist. This will
> >> break a hugetlb cgroup removal isn't it ?
> >
> > This particular part will not break removal because
> > hugetlb_cgroup_css_offline loops until hugetlb_cgroup_have_usage is 0.
> >
> 
> But we still need to hold hugetlb_lock around that right ?

Right. Racing hugetlb_cgroup_move_parent and hugetlb_cgroup_migrate could
lead to newpage pointing to NULL cgroup. That could be fixed by checking
old page cgroup for NULL inside hugetlb_lock and using
list_for_each_safe in hugetlb_cgroup_css_offline no?
-- 
Michal Hocko
SUSE Labs