Re: Linux-next: crash in alloc_huge_page()
From: Qian Cai
Date: Wed Aug 11 2021 - 10:25:24 EST
On 8/11/2021 10:11 AM, Mike Rapoport wrote:
> On Wed, Aug 11, 2021 at 04:06:18AM +0100, Matthew Wilcox wrote:
>> On Tue, Aug 10, 2021 at 10:22:37PM -0400, Qian Cai wrote:
>>> and the page->lru has an address fffffffffffffffc for some reasons. Does it sound like some error code
>>> had not been handled properly and had been propagated here instead? I tried reverting a few recent
>>> commits for mm/hugetlb.c and mm/memblock.c without luck so far.
>>
>> Yes, ff..fc is going to be at offset 8 from the actual address, so
>> that's -12 and -12 is ...
>>
>> #define ENOMEM 12 /* Out of memory */
>>
>> so something's returning ERR_PTR(-ENOMEM) instead of NULL.
>
> page is not initialized in alloc_buddy_huge_page_with_mpol() and after
> commit 2cfa8b23744f ("mm-hugetlb-add-support-for-mempolicy-mpol_preferred_many-fix") we have
Good catch, Mike! Pretty sure I missed to test that commit thought that was an old commit along
with the rest of the mpol_preferred series.
It is a dream that one day mm tree could like other subsystem trees where "git tag --contains"
would work to indicate which linux-next tags contains a particular commit to tell the timeline
of it. Right now, we have those commits ID always changed and commit date is meaningless for
mm commits in linux-next.
>
> struct page *page;
>
> ...
>
> if (mpol_is_preferred_many(mpol)) {
> gfp_t gfp = gfp_mask | __GFP_NOWARN;
>
> gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL);
> page = alloc_surplus_huge_page(h, gfp, nid, nodemask, false);
>
> /* Fallback to all nodes if page==NULL */
> nodemask = NULL;
> }
>
> if (!page)
> page = alloc_surplus_huge_page(h, gfp_mask, nid, nodemask, false
>
> mpol_cond_put(mpol);
> return page;
>
> so for !mpol_is_preferred_many() we return an uninitialized variable.
>
> This should fix it:
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 008662083fec..6337697f7ee4 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -2152,7 +2152,7 @@ static
> struct page *alloc_buddy_huge_page_with_mpol(struct hstate *h,
> struct vm_area_struct *vma, unsigned long addr)
> {
> - struct page *page;
> + struct page *page = NULL;
> struct mempolicy *mpol;
> gfp_t gfp_mask = htlb_alloc_mask(h);
> int nid;
>
>