Re: [RFC PATCH v1 0/3] Userspace MFR Policy via memfd

[Jiaqi Yan <jiaqiyan@google.com>]

On Mon, Nov 3, 2025 at 12:53 AM Harry Yoo <harry.yoo@oracle.com> wrote:
>
> On Mon, Nov 03, 2025 at 05:16:33PM +0900, Harry Yoo wrote:
> > On Thu, Oct 30, 2025 at 10:28:48AM -0700, Jiaqi Yan wrote:
> > > On Thu, Oct 30, 2025 at 4:51 AM Miaohe Lin <linmiaohe@huawei.com> wrote:
> > > > On 2025/10/28 15:00, Harry Yoo wrote:
> > > > > On Mon, Oct 27, 2025 at 09:17:31PM -0700, Jiaqi Yan wrote:
> > > > >> On Wed, Oct 22, 2025 at 6:09 AM Harry Yoo <harry.yoo@oracle.com> wrote:
> > > > >>> On Mon, Oct 13, 2025 at 03:14:32PM -0700, Jiaqi Yan wrote:
> > > > >>>> On Fri, Sep 19, 2025 at 8:58 AM “William Roche <william.roche@oracle.com> wrote:
> > > > >>> But even after fixing that we need to fix the race condition.
> > > > >>
> > > > >> What exactly is the race condition you are referring to?
> > > > >
> > > > > When you free a high-order page, the buddy allocator doesn't not check
> > > > > PageHWPoison() on the page and its subpages. It checks PageHWPoison()
> > > > > only when you free a base (order-0) page, see free_pages_prepare().
> > > >
> > > > I think we might could check PageHWPoison() for subpages as what free_page_is_bad()
> > > > does. If any subpage has HWPoisoned flag set, simply drop the folio. Even we could
> > >
> > > Agree, I think as a starter I could try to, for example, let
> > > free_pages_prepare scan HWPoison-ed subpages if the base page is high
> > > order. In the optimal case, HugeTLB does move PageHWPoison flag from
> > > head page to the raw error pages.
> >
> > [+Cc page allocator folks]
> >
> > AFAICT enabling page sanity check in page alloc/free path would be against
> > past efforts to reduce sanity check overhead.
> >
> > [1] https://lore.kernel.org/linux-mm/1460711275-1130-15-git-send-email-mgorman@techsingularity.net/
> > [2] https://lore.kernel.org/linux-mm/1460711275-1130-16-git-send-email-mgorman@techsingularity.net/
> > [3] https://lore.kernel.org/all/20230216095131.17336-1-vbabka@suse.cz
> >
> > I'd recommend to check hwpoison flag before freeing it to the buddy
> > when we know a memory error has occurred (I guess that's also what Miaohe
> > suggested).
> >
> > > > do it better -- Split the folio and let healthy subpages join the buddy while reject
> > > > the hwpoisoned one.
> > > >
> > > > >
> > > > > AFAICT there is nothing that prevents the poisoned page to be
> > > > > allocated back to users because the buddy doesn't check PageHWPoison()
> > > > > on allocation as well (by default).
> > > > >
> > > > > So rather than freeing the high-order page as-is in
> > > > > dissolve_free_hugetlb_folio(), I think we have to split it to base pages
> > > > > and then free them one by one.
> > > >
> > > > It might not be worth to do that as this would significantly increase the overhead
> > > > of the function while memory failure event is really rare.
> > >
> > > IIUC, Harry's idea is to do the split in dissolve_free_hugetlb_folio
> > > only if folio is HWPoison-ed, similar to what Miaohe suggested
> > > earlier.
> >
> > Yes, and if we do the check before moving HWPoison flag to raw pages,
> > it'll be just a single folio_test_hwpoison() call.
> >
> > > BTW, I believe this race condition already exists today when
> > > memory_failure handles HWPoison-ed free hugetlb page; it is not
> > > something introduced via this patchset. I will fix or improve this in
> > > a separate patchset.
> >
> > That makes sense.
>
> Wait, without this patchset, do we even free the hugetlb folio when
> its subpage is hwpoisoned? I don't think we do, but I'm not expert at MFR...

Based on my reading of try_memory_failure_hugetlb, me_huge_page, and
__page_handle_poison, I think mainline kernel frees dissolved hugetlb
folio to buddy allocator in two cases:
1. it was a free hugetlb page at the moment of try_memory_failure_hugetlb
2. it was an anonomous hugetlb page

Let me know if my understanding is wrong.

>
> If we don't, the mainline kernel should not be affected by this yet?
>
> > Thanks for working on this!
> >
> > > > > That way, free_pages_prepare() will catch that it's poisoned and won't
> > > > > add it back to the freelist. Otherwise there will always be a window
> > > > > where the poisoned page can be allocated to users - before it's taken
> > > > > off from the buddy.
>
> --
> Cheers,
> Harry / Hyeonggon