On Sat, Dec 05, 2020 at 09:26:47PM -0500, Andrea Arcangeli wrote:
> Hi Mel,
>
> On Thu, Nov 26, 2020 at 10:47:20AM +0000, Mel Gorman wrote:
> > Agreed. This thread has a lot of different directions in it at this
> > point so what I'd hope for is first, a patch that initialises holes with
> > zone/node linkages within a 1<<(MAX_ORDER-1) alignment. If there is a
> > hole, it would be expected the pages are PageReserved. Second, a fix to
> > fast_isolate that forces PFNs returned to always be within the stated
> > zone boundaries.
>
> The last two patches should resolve the struct page
> initialization
> https://git.kernel.org/pub/scm/linux/kernel/git/andrea/aa.git/ and the
> VM_BUG_ON_PAGE never happened again as expected.
>
> So I looked back to see how the "distance" logic is accurate. I added
> those checks and I get negative hits:
>
> diff --git a/mm/compaction.c b/mm/compaction.c
> index cc1a7f600a86..844a90b0acdf 100644
> --- a/mm/compaction.c
> +++ b/mm/compaction.c
> @@ -1331,6 +1331,12 @@ fast_isolate_freepages(struct compact_control *cc)
> low_pfn = pageblock_start_pfn(cc->free_pfn - (distance >> 2));
> min_pfn = pageblock_start_pfn(cc->free_pfn - (distance >> 1));
>
> + WARN_ON_ONCE((long) low_pfn < 0);
> + WARN_ON_ONCE((long) min_pfn < 0);
> + if ((long) low_pfn < 0)
> + return cc->free_pfn;
> + if ((long) min_pfn < 0)
> + return cc->free_pfn;
> if (WARN_ON_ONCE(min_pfn > low_pfn))
> low_pfn = min_pfn;
>
> Both warn-on-once triggers, so it goes negative. While it appears not
> kernel crashing since pfn_valid won't succeed on negative entries and
> they'll always be higher than any pfn in the freelist, is this sign
> that there's further room for improvement here?
>
Possibly, checking the wrong pfns is simply risky. This is not tested
at all, just checking if it's in the right ballpark even. Intent is that
when the free/migrate PFNs are too close or already overlapping that no
attempt is made and it returns back to detect the scanners have met.
diff --git a/mm/compaction.c b/mm/compaction.c
index 13cb7a961b31..208cb5857446 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -1313,6 +1313,10 @@ fast_isolate_freepages(struct compact_control *cc)
if (cc->order <= 0)
return cc->free_pfn;
+ /* Ensure that migration and free scanner are not about to cross */
+ if (cc->migrate_pfn < cc->free_pfn)
+ return cc->free_pfn;
+
/*
* If starting the scan, use a deeper search and use the highest
* PFN found if a suitable one is not found.
@@ -1324,9 +1328,12 @@ fast_isolate_freepages(struct compact_control *cc)
/*
* Preferred point is in the top quarter of the scan space but take
- * a pfn from the top half if the search is problematic.
+ * a pfn from the top half if the search is problematic. Ensure
+ * there is enough distance to make the fast search worthwhile.
*/
distance = (cc->free_pfn - cc->migrate_pfn);
+ if (distance <= (pageblock_nr_pages << 2))
+ return cc->free_pfn;
low_pfn = pageblock_start_pfn(cc->free_pfn - (distance >> 2));
min_pfn = pageblock_start_pfn(cc->free_pfn - (distance >> 1));
--
Mel Gorman
SUSE Labs
