Re: [PATCH v2 1/3] mm: khugepaged: fix swap entry value to folio_pfn()

On Mon, Aug 17, 2026 at 05:10:39PM +0100, Lorenzo Stoakes (ARM) wrote:
> On Sat, Aug 15, 2026 at 01:19:22PM +0800, Vernon Yang wrote:
> > From: Vernon Yang <[email protected]>
> >
> > When the swap entries found exceed max_ptes_swap, the loop is left via
> > break with folio still holding the xarray value that encodes the swap
> > entry, not valid folio pointer.
> >
> > That value is passed to trace_mm_khugepaged_scan_file(), which feeds it
> > to folio_pfn(). On FLATMEM and SPARSEMEM_VMEMMAP, the page_to_pfn() is
> > plain pointer arithmetic, so the trace event merely prints bogus
> > scan_pfn. On classic SPARSEMEM, the page_to_pfn() reads page->flags,
> > dereferencing the tiny encoded integer and oopsing khugepaged whenever
> > the trace event is enabled.
>
> Hmm I seem to recall that we were no longer spporting !VMEMMAP SPARSEMEM, but
> maybe I was imagining that :)

OK seems some museum piece architectures are causing us issues again, fun times
(parisc and some mips).

>
> But this needs fixing in general anyway.
>
> >
> > So when folio is the swap entry value, simply set pfn to -1, just like
> > exhausted scan naturally.
> >
> > And the folio_put() has maybe dropped the last reference of folio. The
> > trace_mm_khugepaged_scan_file() is left with a dangling folio pointer.
> > so using the folio_pfn() before dropping the reference, closing
> > use-after-free window.
> >
> > Fixes: d41fd2016ed0 ("mm/khugepaged: add tracepoint to hpage_collapse_scan_file()")
> > Cc: [email protected]
> > Signed-off-by: Vernon Yang <[email protected]>
> > ---
> >  include/trace/events/huge_memory.h |  6 +++---
> >  mm/khugepaged.c                    | 14 +++++++++-----
> >  2 files changed, 12 insertions(+), 8 deletions(-)
> >
> > diff --git a/include/trace/events/huge_memory.h b/include/trace/events/huge_memory.h
> > index 291fae364c62..d3572d4ef453 100644
> > --- a/include/trace/events/huge_memory.h
> > +++ b/include/trace/events/huge_memory.h
> > @@ -178,10 +178,10 @@ TRACE_EVENT(mm_collapse_huge_page_swapin,
> >
> >  TRACE_EVENT(mm_khugepaged_scan_file,
> >
> > -	TP_PROTO(struct mm_struct *mm, struct folio *folio, struct file *file,
> > +	TP_PROTO(struct mm_struct *mm, unsigned long pfn, struct file *file,
> >  		 int present, int swap, int result),
> >
> > -	TP_ARGS(mm, folio, file, present, swap, result),
> > +	TP_ARGS(mm, pfn, file, present, swap, result),
> >
> >  	TP_STRUCT__entry(
> >  		__field(struct mm_struct *, mm)
> > @@ -194,7 +194,7 @@ TRACE_EVENT(mm_khugepaged_scan_file,
> >
> >  	TP_fast_assign(
> >  		__entry->mm = mm;
> > -		__entry->pfn = folio ? folio_pfn(folio) : -1;
> > +		__entry->pfn = pfn;
>
> I wonder how easy it is for people to interpret that this is a PFN of something
> only in the case of an early exit.
>
> It's a bit of a mess that we're exposing internal implementation details like
> this, quite honestly.
>
> I hope there is no expectation of this being there indefinitely.
>
> >  		__assign_str(filename);
> >  		__entry->present = present;
> >  		__entry->swap = swap;
> > diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> > index 617bca76db49..e7830761d3a2 100644
> > --- a/mm/khugepaged.c
> > +++ b/mm/khugepaged.c
> > @@ -2683,6 +2683,7 @@ static enum scan_result collapse_scan_file(struct mm_struct *mm,
> >  	int present, swap;
> >  	int node = NUMA_NO_NODE;
> >  	enum scan_result result = SCAN_SUCCEED;
> > +	unsigned long pfn;
>
> See below I think we should drop it.

OK we can't because of museum piece architectures that will actually deref the
folio to get the PFN...

I'd startr by initialising this to -1 then since I guess in theory it's possible
xas_for_each() could just not do anything is it? (folio is being set to NULL so
that suggests so).

>
> >
> >  	present = 0;
> >  	swap = 0;
> > @@ -2720,27 +2721,23 @@ static enum scan_result collapse_scan_file(struct mm_struct *mm,
> >  			 * PMD-sized THP implies that we can only try
> >  			 * retracting the PTE table.
> >  			 */
> > -			folio_put(folio);

I still hate the idea of moving these outside of the loop, it's just asking for trouble.

So I guess instead:

	xas_for_each(&xas, folio, start + HPAGE_PMD_NR - 1) {
		pfn = -1;
		...
		if (xa_is_value(folio)) {
			...
		}
		pfn = folio_pfn(folio);
		...
	}
	...
	trace_mm_khugepaged_scan_file(mm, pfn, file, present, swap, result);

That way you avoid the horror or moving the folio put somewhere deeply
unintuitive and inconsistent, and maintain pfn lifetime for some arches nobody
uses.


> >  			break;
> >  		}
> >
> >  		node = folio_nid(folio);
> >  		if (collapse_scan_abort(node, cc)) {
> >  			result = SCAN_SCAN_ABORT;
> > -			folio_put(folio);
> >  			break;
> >  		}
> >  		cc->node_load[node]++;
> >
> >  		if (!folio_test_lru(folio)) {
> >  			result = SCAN_PAGE_LRU;
> > -			folio_put(folio);
> >  			break;
> >  		}
> >
> >  		if (folio_expected_ref_count(folio) + 1 != folio_ref_count(folio)) {
> >  			result = SCAN_PAGE_COUNT;
> > -			folio_put(folio);
> >  			break;
> >  		}
> >
> > @@ -2759,7 +2756,14 @@ static enum scan_result collapse_scan_file(struct mm_struct *mm,
> >  			cond_resched_rcu();
> >  		}
> >  	}
> > +	if (!folio || xa_is_value(folio)) {
> > +		pfn = -1;
> > +	} else {
> > +		pfn = folio_pfn(folio);
> > +		folio_put(folio);
> > +	}
>
> Hmm kinda hate this you're now mixing clean up logic with some trace logic.
>
> You have to go check out xas_for_each() and see that xas_next_entry() sets folio
> to NULL at the end for the normal case to know this is OK and then figure out
> that this is just for the break cases.
>
> That's not intuitive in already not intuitive code.
>
> And this is a patch you're wanting to backport to fix a bug too... :)
>
> I think you should keep the folio_put()'s where they are for now and limit this
> fix to the trace code please.
>
> I think you should drop the pfn var altogether and do this far more simply, see
> below.
>
> >  	rcu_read_unlock();
> > +
>
> Stray newline added, let's not do that for a backported fix please :)
>
> >  	if (result == SCAN_PTE_MAPPED_HUGEPAGE)
> >  		cc->progress++;
> >  	else
> > @@ -2774,7 +2778,7 @@ static enum scan_result collapse_scan_file(struct mm_struct *mm,
> >  		}
> >  	}
> >
> > -	trace_mm_khugepaged_scan_file(mm, folio, file, present, swap, result);
> > +	trace_mm_khugepaged_scan_file(mm, pfn, file, present, swap, result);
>
> I mean maybe better to just reduce this to something like:
>
> 	/* folio is NULL unless exited early. */
> 	trace_mm_khugepaged_scan_file(mm,
> 			(!folio || xa_is_value(folio)) ? -1 : folio_pfn(folio),
> 			file, present, swap, result);
>
> As the least invasive way of fixing the trace + limiting the contageon?

Yeah as above this isn't right.

>
> >  	return result;
> >  }
> >
> > --
> > 2.53.0
> >
>
> --
> Cheers, Lorenzo

--
Cheers, Lorenzo