Skip to contentSkip to conversation

mm/gup: fix GUP-fast fallback for NULL-mapping order-0 folios

PATCH Superseded John Hubbard 6 participants 17 messages Apr 9 – Jun 11

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

John Hubbard

author reviewer
#0 lore raw

PATCH 1/1 mm/gup: fix GUP-fast fallback for NULL-mapping order-0 folios

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Fixes f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()")
---
 mm/gup.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
mm/gup.c +9 -4
index 8e7dc2c6ee73..3ea661e67eea 100644
@@ -2784,12 +2784,17 @@ static bool gup_fast_folio_allowed(struct folio *folio, unsigned int flags)
 	mapping = READ_ONCE(folio->mapping);
 
 	/*
-	 * The mapping may have been truncated, in any case we cannot determine
-	 * if this mapping is safe - fall back to slow path to determine how to
-	 * proceed.
+	 * If the mapping is NULL (truncated, or never set), we cannot
+	 * determine whether the folio is file-backed, so a long-term writable
+	 * pin must fall back to the slow path.
+	 *
+	 * Otherwise, a NULL mapping proves this is not a secretmem folio
+	 * (secretmem folios always have a valid mapping to the secretmem
+	 * inode's address_space), so in that case, we can continue with the
+	 * fast path.
 	 */
 	if (!mapping)
-		return false;
+		return !reject_file_backed;
 
 	/* Anonymous folios pose no problem. */
 	mapping_flags = (unsigned long)mapping & FOLIO_MAPPING_FLAGS;

base-commit: 7f87a5ea75f011d2c9bc8ac0167e5e2d1adb1594

Signature · 1 line
2.53.0
maintainer
#1 lore raw
Quoted 7 lines · On Wed, 8 Apr 2026 18:46:47 -0700 John Hubbard <[email protected]> wrote:

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

How significant is the slowdown?

Quoted 9 lines

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

author reviewer
#2 lore raw
Quoted 11 lines · Andrew Morton
Quoted 7 lines · On Wed, 8 Apr 2026 18:46:47 -0700 John Hubbard <[email protected]> wrote:

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

How significant is the slowdown?

About 2x!

Sourab has a bunch of peer to peer IO cases that use 4K pages, that he ran.

thanks,

Signature · 1 line
John Hubbard
#3 lore raw
In reply to John Hubbard
Quoted 14 lines · John Hubbard

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

Quoted 3 lines

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

Quoted 36 lines
Fixes f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()")
---
 mm/gup.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
mm/gup.c +9 -4
index 8e7dc2c6ee73..3ea661e67eea 100644
@@ -2784,12 +2784,17 @@ static bool gup_fast_folio_allowed(struct folio *folio, unsigned int flags)
 	mapping = READ_ONCE(folio->mapping);
 
 	/*
-	 * The mapping may have been truncated, in any case we cannot determine
-	 * if this mapping is safe - fall back to slow path to determine how to
-	 * proceed.
+	 * If the mapping is NULL (truncated, or never set), we cannot
+	 * determine whether the folio is file-backed, so a long-term writable
+	 * pin must fall back to the slow path.
+	 *
+	 * Otherwise, a NULL mapping proves this is not a secretmem folio
+	 * (secretmem folios always have a valid mapping to the secretmem
+	 * inode's address_space), so in that case, we can continue with the
+	 * fast path.
 	 */
 	if (!mapping)
-		return false;
+		return !reject_file_backed;
 
 	/* Anonymous folios pose no problem. */
 	mapping_flags = (unsigned long)mapping & FOLIO_MAPPING_FLAGS;

base-commit: 7f87a5ea75f011d2c9bc8ac0167e5e2d1adb1594

Signature · 3 lines
Cheers,

David
#4 lore raw
Quoted 19 lines · David Hildenbrand (Arm)
Quoted 14 lines · John Hubbard

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

Quoted 3 lines

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Quoted 6 lines

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

The idea is to give every allocated page a NotRmappable page_type and
1) let page_rmappable_folio() remove the page_type by setting mapcount to 0, since page_type overlaps with mapcount;
2) allow callers of alloc_page*() to change page_type from NotRmappable to other types.

After these, rmappable folios with any order is a folio without any page_type. HugeTLB is an exception, but rmap code has special handling for them, so it should be fine.

I have a draft[1], but have not had time to refresh it yet. I am planning to work on this after my two other jobs (make READ_ONLY_THP_FOR_FS enabled for all FS with large folio support and ensure ->private is 0 for tail pages).

[1] https://github.com/x-y-z/linux-dev/tree/remove_mapcount_for_non_folio-2026-02-16-23-27

Quoted 4 lines

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

Based on my understanding of some callers of alloc_page()/get_free_pages(), vm_insert_page(), they use GFP_COMP to get a compound page and vm_insert_page() individual pages of the compound page. These pages might be pinned. During past THP cabal meeting, we discussed about it if
1. we should disallow such uses and only allow order-0 allocations, or
2. we need a new type like new_type_folio for them and the type does not have folio fields except ->pincount.

Willy (cc’d) said he coded some experiment on it some time ago. I do not know if he is able to recover the code.

Quoted 33 lines
Quoted 33 lines
Fixes f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()")
---
 mm/gup.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
mm/gup.c +9 -4
index 8e7dc2c6ee73..3ea661e67eea 100644
@@ -2784,12 +2784,17 @@ static bool gup_fast_folio_allowed(struct folio *folio, unsigned int flags)
 	mapping = READ_ONCE(folio->mapping);
 
 	/*
-	 * The mapping may have been truncated, in any case we cannot determine
-	 * if this mapping is safe - fall back to slow path to determine how to
-	 * proceed.
+	 * If the mapping is NULL (truncated, or never set), we cannot
+	 * determine whether the folio is file-backed, so a long-term writable
+	 * pin must fall back to the slow path.
+	 *
+	 * Otherwise, a NULL mapping proves this is not a secretmem folio
+	 * (secretmem folios always have a valid mapping to the secretmem
+	 * inode's address_space), so in that case, we can continue with the
+	 * fast path.
 	 */
 	if (!mapping)
-		return false;
+		return !reject_file_backed;
 
Quoted 10 lines
Quoted 4 lines
mm/gup.c +0 -0
 	/* Anonymous folios pose no problem. */
 	mapping_flags = (unsigned long)mapping & FOLIO_MAPPING_FLAGS;

base-commit: 7f87a5ea75f011d2c9bc8ac0167e5e2d1adb1594

Signature · 3 lines
Cheers,

David

Best Regards,
Yan, Zi

#5 lore raw
Quoted 25 lines · Zi Yan
Quoted 23 lines · David Hildenbrand (Arm)
Quoted 14 lines · John Hubbard

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

Quoted 3 lines

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Quoted 8 lines
Quoted 2 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

The idea is to give every allocated page a NotRmappable page_type and
1) let page_rmappable_folio() remove the page_type by setting mapcount to 0, since page_type overlaps with mapcount;
2) allow callers of alloc_page*() to change page_type from NotRmappable to other types.

Could we instead just set the special type in vm_insert_page(), after verifying that it does not have any other type?

Quoted 16 lines

After these, rmappable folios with any order is a folio without any page_type. HugeTLB is an exception, but rmap code has special handling for them, so it should be fine.

I have a draft[1], but have not had time to refresh it yet. I am planning to work on this after my two other jobs (make READ_ONLY_THP_FOR_FS enabled for all FS with large folio support and ensure ->private is 0 for tail pages).

[1] https://github.com/x-y-z/linux-dev/tree/remove_mapcount_for_non_folio-2026-02-16-23-27

Quoted 4 lines

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

Quoted 5 lines

Based on my understanding of some callers of alloc_page()/get_free_pages(), vm_insert_page(), they use GFP_COMP to get a compound page and vm_insert_page() individual pages of the compound page. These pages might be pinned. During past THP cabal meeting, we discussed about it if
1. we should disallow such uses and only allow order-0 allocations, or

Right.

Quoted 2 lines

2. we need a new type like new_type_folio for them and the type does not have folio fields except ->pincount.

Right.

Signature · 3 lines
Cheers,

David
#6 lore raw
Quoted 29 lines · David Hildenbrand (Arm)
Quoted 28 lines · Zi Yan
Quoted 25 lines · David Hildenbrand (Arm)
Quoted 14 lines · John Hubbard

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

Quoted 3 lines

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Quoted 8 lines
Quoted 5 lines

The idea is to give every allocated page a NotRmappable page_type and
1) let page_rmappable_folio() remove the page_type by setting mapcount to 0, since page_type overlaps with mapcount;
2) allow callers of alloc_page*() to change page_type from NotRmappable to other types.

Could we instead just set the special type in vm_insert_page(), after verifying that it does not have any other type?

You mean in validate_page_before_insert() we set the page to NotRmappable if it can be inserted, then in the rmap code we reject NotRmappable page/folios? And NotRmappable will remain until the page/folio is freed. That should work. So the purpose is to avoid setting the page_type if a page/folio is never going through vm_insert_page()?

Quoted 23 lines
Quoted 17 lines

After these, rmappable folios with any order is a folio without any page_type. HugeTLB is an exception, but rmap code has special handling for them, so it should be fine.

I have a draft[1], but have not had time to refresh it yet. I am planning to work on this after my two other jobs (make READ_ONLY_THP_FOR_FS enabled for all FS with large folio support and ensure ->private is 0 for tail pages).

[1] https://github.com/x-y-z/linux-dev/tree/remove_mapcount_for_non_folio-2026-02-16-23-27

Quoted 4 lines

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

Quoted 5 lines

Based on my understanding of some callers of alloc_page()/get_free_pages(), vm_insert_page(), they use GFP_COMP to get a compound page and vm_insert_page() individual pages of the compound page. These pages might be pinned. During past THP cabal meeting, we discussed about it if
1. we should disallow such uses and only allow order-0 allocations, or

Quoted 11 lines

Right.

Quoted 2 lines

2. we need a new type like new_type_folio for them and the type does not have folio fields except ->pincount.

Right.

Signature · 3 lines
Cheers,

David

Best Regards,
Yan, Zi

#7 lore raw
Quoted 11 lines · Zi Yan
Quoted 9 lines · David Hildenbrand (Arm)
Quoted 7 lines · Zi Yan

The idea is to give every allocated page a NotRmappable page_type and
1) let page_rmappable_folio() remove the page_type by setting mapcount to 0, since page_type overlaps with mapcount;
2) allow callers of alloc_page*() to change page_type from NotRmappable to other types.

Quoted 8 lines
Quoted 2 lines

Could we instead just set the special type in vm_insert_page(), after verifying that it does not have any other type?

You mean in validate_page_before_insert() we set the page to NotRmappable if it can be inserted, then in the rmap code we reject NotRmappable page/folios? And NotRmappable will remain until the page/folio is freed. That should work. So the purpose is to avoid setting the page_type if a page/folio is never going through vm_insert_page()?

Yes, only these non-folio pages will show up process page tables.

We would add a safety check that the mapcount really is 0 (-1) or already the proper type (NotRmappable).

Rmap code would then simply skip any such thing.

Signature · 3 lines
Cheers,

David
#8 lore raw
Quoted 21 lines · David Hildenbrand (Arm)
Quoted 11 lines · Zi Yan
Quoted 9 lines · David Hildenbrand (Arm)
Quoted 7 lines · Zi Yan

The idea is to give every allocated page a NotRmappable page_type and
1) let page_rmappable_folio() remove the page_type by setting mapcount to 0, since page_type overlaps with mapcount;
2) allow callers of alloc_page*() to change page_type from NotRmappable to other types.

Quoted 8 lines
Quoted 2 lines

Could we instead just set the special type in vm_insert_page(), after verifying that it does not have any other type?

You mean in validate_page_before_insert() we set the page to NotRmappable if it can be inserted, then in the rmap code we reject NotRmappable page/folios? And NotRmappable will remain until the page/folio is freed. That should work. So the purpose is to avoid setting the page_type if a page/folio is never going through vm_insert_page()?

Quoted 6 lines

Yes, only these non-folio pages will show up process page tables.

We would add a safety check that the mapcount really is 0 (-1) or already the proper type (NotRmappable).

Rmap code would then simply skip any such thing.

OK, I can give this a try in my new version.

Best Regards,
Yan, Zi

#9 lore raw
In reply to David Hildenbrand (Arm)
Quoted 15 lines · David Hildenbrand (Arm)
Quoted 14 lines · John Hubbard

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

Quoted 8 lines
Quoted 3 lines

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

Quoted 7 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

I still think this is a short term fix worth having until we get Zi's fixes

Quoted 38 lines
Quoted 36 lines
Fixes f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()")
---
 mm/gup.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
mm/gup.c +9 -4
index 8e7dc2c6ee73..3ea661e67eea 100644
@@ -2784,12 +2784,17 @@ static bool gup_fast_folio_allowed(struct folio *folio, unsigned int flags)
 	mapping = READ_ONCE(folio->mapping);
 
 	/*
-	 * The mapping may have been truncated, in any case we cannot determine
-	 * if this mapping is safe - fall back to slow path to determine how to
-	 * proceed.
+	 * If the mapping is NULL (truncated, or never set), we cannot
+	 * determine whether the folio is file-backed, so a long-term writable
+	 * pin must fall back to the slow path.
+	 *
+	 * Otherwise, a NULL mapping proves this is not a secretmem folio
+	 * (secretmem folios always have a valid mapping to the secretmem
+	 * inode's address_space), so in that case, we can continue with the
+	 * fast path.
 	 */
 	if (!mapping)
-		return false;
+		return !reject_file_backed;
 
 	/* Anonymous folios pose no problem. */
 	mapping_flags = (unsigned long)mapping & FOLIO_MAPPING_FLAGS;

base-commit: 7f87a5ea75f011d2c9bc8ac0167e5e2d1adb1594

#10 lore raw
Quoted 20 lines · Balbir Singh
Quoted 19 lines · David Hildenbrand (Arm)
Quoted 17 lines · John Hubbard

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Quoted 8 lines
Quoted 4 lines

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

I thought FALLOC_FL_PUNCH_HOLE would work. But indeed, that does not seem to be wired up.

Signature · 3 lines
Cheers,

David
#11 lore raw
In reply to Balbir Singh

On 2026-05-15 at 10:09 +1000, Balbir Singh <[email protected]> wrote...

Quoted 32 lines · David Hildenbrand (Arm)
Quoted 15 lines · John Hubbard
Quoted 14 lines

Since commit f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()"), gup_fast_folio_allowed() falls back to the slow path for any order-0 folio with a NULL mapping when CONFIG_SECRETMEM=y. This causes a performance regression for drivers that allocate pages with alloc_page() and insert them into VMAs via vm_insert_page(). These pages legitimately have a NULL folio->mapping, but they cannot be secretmem pages.

Secretmem pages are always added to the secretmem inode's page cache via filemap_add_folio(), which sets folio->mapping to the inode's i_mapping. A folio with a NULL mapping can never be a secretmem folio. The NULL-mapping check was intended to handle truncated file-backed pages (a reject_file_backed concern), not secretmem detection.

Quoted 8 lines
Quoted 3 lines

When only check_secretmem is true (and reject_file_backed is false), a NULL mapping is sufficient to prove the folio is not secretmem, so the fast path can proceed.

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

Quoted 3 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Quoted 7 lines
Quoted 4 lines

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

I still think this is a short term fix worth having until we get Zi's fixes

Agreed - this clearly results in a performance regression and we have customers reporting it as such. I think Zi might be out of the office atm but when I last spoke to him he agreed this fix should go in for now and his approach will come later.

- Alistair

Quoted 40 lines
Quoted 38 lines
Quoted 36 lines
Fixes f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()")
---
 mm/gup.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
mm/gup.c +9 -4
index 8e7dc2c6ee73..3ea661e67eea 100644
@@ -2784,12 +2784,17 @@ static bool gup_fast_folio_allowed(struct folio *folio, unsigned int flags)
 	mapping = READ_ONCE(folio->mapping);
 
 	/*
-	 * The mapping may have been truncated, in any case we cannot determine
-	 * if this mapping is safe - fall back to slow path to determine how to
-	 * proceed.
+	 * If the mapping is NULL (truncated, or never set), we cannot
+	 * determine whether the folio is file-backed, so a long-term writable
+	 * pin must fall back to the slow path.
+	 *
+	 * Otherwise, a NULL mapping proves this is not a secretmem folio
+	 * (secretmem folios always have a valid mapping to the secretmem
+	 * inode's address_space), so in that case, we can continue with the
+	 * fast path.
 	 */
 	if (!mapping)
-		return false;
+		return !reject_file_backed;
 
 	/* Anonymous folios pose no problem. */
 	mapping_flags = (unsigned long)mapping & FOLIO_MAPPING_FLAGS;

base-commit: 7f87a5ea75f011d2c9bc8ac0167e5e2d1adb1594

#12 lore raw
Quoted 20 lines · Alistair Popple

On 2026-05-15 at 10:09 +1000, Balbir Singh <[email protected]> wrote...

Quoted 19 lines · David Hildenbrand (Arm)
Quoted 5 lines

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

Quoted 7 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

Quoted 7 lines
Quoted 2 lines

I still think this is a short term fix worth having until we get Zi's fixes

Agreed - this clearly results in a performance regression and we have customers reporting it as such. I think Zi might be out of the office atm but when I last spoke to him he agreed this fix should go in for now and his approach will come later.

He should be back now. But yeah, let's get this fix in.

Signature · 3 lines
Cheers,

David
#13 lore raw
Quoted 24 lines · David Hildenbrand (Arm)
Quoted 23 lines · Alistair Popple

On 2026-05-15 at 10:09 +1000, Balbir Singh <[email protected]> wrote...

Quoted 21 lines · David Hildenbrand (Arm)
Quoted 5 lines

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

Quoted 7 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

I still think this is a short term fix worth having until we get Zi's fixes

Quoted 6 lines
Quoted 4 lines

Agreed - this clearly results in a performance regression and we have customers reporting it as such. I think Zi might be out of the office atm but when I last spoke to him he agreed this fix should go in for now and his approach will come later.

He should be back now. But yeah, let's get this fix in.

Right, I will work on my solution soon. Meanwhile, this change is simpler and can be easily back ported.

BTW, should it be back ported to older kernels?

Best Regards,
Yan, Zi

#14 lore raw

On 2026-06-06 at 02:20 +1000, Zi Yan <[email protected]> wrote...

Quoted 31 lines · David Hildenbrand (Arm)
Quoted 30 lines · Alistair Popple
Quoted 27 lines

On 2026-05-15 at 10:09 +1000, Balbir Singh <[email protected]> wrote...

Quoted 14 lines · David Hildenbrand (Arm)
Quoted 5 lines

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

Quoted 3 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Quoted 7 lines
Quoted 4 lines

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

I still think this is a short term fix worth having until we get Zi's fixes

Agreed - this clearly results in a performance regression and we have customers reporting it as such. I think Zi might be out of the office atm but when I last spoke to him he agreed this fix should go in for now and his approach will come later.

Quoted 4 lines
Quoted 1 line

He should be back now. But yeah, let's get this fix in.

Right, I will work on my solution soon. Meanwhile, this change is simpler and can be easily back ported.

Thanks.

Quoted 1 line

BTW, should it be back ported to older kernels?

Yes to any with f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()") - we have got bug reports as a result of the lowered perf that commit causes.

- Alistair

Quoted 2 lines

Best Regards,
Yan, Zi

#15 lore raw
Quoted 39 lines · Alistair Popple

On 2026-06-06 at 02:20 +1000, Zi Yan <[email protected]> wrote...

Quoted 31 lines · David Hildenbrand (Arm)
Quoted 30 lines · Alistair Popple
Quoted 27 lines

On 2026-05-15 at 10:09 +1000, Balbir Singh <[email protected]> wrote...

Quoted 14 lines · David Hildenbrand (Arm)
Quoted 5 lines

Hm, what if secretmem folio just got truncated? I hate to rely on some handling in the caller to detect truncation differently during GUP-fast, but this function returning "true".

Can secretmem folios be truncated? I assume you are referring to ftruncate(), I am looking at the setattr implementation of secretmem and it does not seem like it can be truncated.

Quoted 3 lines

Zi is working on a way to distinguish folios from non-folio things: that we can identify whatever was added through vm_insert_page().

Quoted 7 lines
Quoted 4 lines

Because that's really the key problem here: vm_insert_page() pages are not actually folios, they just look like a folio today, but looking at fields like ->mapping does not make any sense.

I still think this is a short term fix worth having until we get Zi's fixes

Agreed - this clearly results in a performance regression and we have customers reporting it as such. I think Zi might be out of the office atm but when I last spoke to him he agreed this fix should go in for now and his approach will come later.

Quoted 4 lines
Quoted 1 line

He should be back now. But yeah, let's get this fix in.

Right, I will work on my solution soon. Meanwhile, this change is simpler and can be easily back ported.

Thanks.

Quoted 8 lines
Quoted 1 line

BTW, should it be back ported to older kernels?

Yes to any with f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()") - we have got bug reports as a result of the lowered perf that commit causes.

- Alistair

Ping @akpm to pick this up, if there is no objection @david

Balbir

maintainer Acked-by
#16 lore raw
Quoted 11 lines · Balbir Singh
Quoted 10 lines · Alistair Popple

On 2026-06-06 at 02:20 +1000, Zi Yan <[email protected]> wrote...

Quoted 4 lines

Right, I will work on my solution soon. Meanwhile, this change is simpler and can be easily back ported.

Thanks.

Quoted 1 line

BTW, should it be back ported to older kernels?

Quoted 8 lines
Quoted 6 lines

Yes to any with f002882ca369 ("mm: merge folio_is_secretmem() and folio_fast_pin_allowed() into gup_fast_folio_allowed()") - we have got bug reports as a result of the lowered perf that commit causes.

- Alistair

Ping @akpm to pick this up, if there is no objection @david

Fine with me

Signature · 3 lines
Cheers,

David