On 19.06.24 17:48, Linus Torvalds wrote:
On Wed, 19 Jun 2024 at 07:31, Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
Actually, it's 11. We can't split an order-12 folio because we'd have
to allocate two levels of radix tree, and I decided that was too much
work. Also, I didn't know that ARM used order-13 PMD size at the time.
I think this is the best fix (modulo s/12/11/).
Can we use some more descriptive thing than the magic constant 11 that
is clearly very subtle.
Is it "XA_CHUNK_SHIFT * 2 - 1"
That's my best guess as well :)
IOW, something like
#define MAX_XAS_ORDER (XA_CHUNK_SHIFT * 2 - 1)
#define MAX_PAGECACHE_ORDER min(HPAGE_PMD_ORDER,12)
except for the non-TRANSPARENT_HUGEPAGE case where it currently does
#define MAX_PAGECACHE_ORDER 8
and I assume that "8" is just "random round value, smaller than 11"?
Yes, that matches my understanding.
Maybe to be safe for !THP as well, something ike:
+++ b/include/linux/pagemap.h
@@ -354,11 +354,18 @@ static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
* a good order (that's 1MB if you're using 4kB pages)
*/
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
-#define MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER
+#define WANTED_MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER
#else
-#define MAX_PAGECACHE_ORDER 8
+#define WANTED_MAX_PAGECACHE_ORDER 8
#endif
+/*
+ * xas_split_alloc() does not support arbitrary orders yet. This implies no
+ * 512MB THP on arm64 with 64k.
+ */
+#define MAX_XAS_ORDER (XA_CHUNK_SHIFT * 2 - 1)
+#define MAX_PAGECACHE_ORDER min(MAX_XAS_ORDER, WANTED_MAX_PAGECACHE_ORDER)
+
/**
* mapping_set_large_folios() - Indicate the file supports large folios.
* @mapping: The file.
--
2.45.2
@Gavin, do you have capacity to test+prepare an official patch? Also,
please double-check whether shmem must be fenced as well (very likely).
This implies no PMD-sized THPs in the pagecache/shmem on arm64 with 64k.
Could be worse, because as Willy said, they are rather rare and extremely
unpredictable.
--
Cheers,
David / dhildenb