Skip to contentSkip to conversation

A few gup refactorings and documentation updates

PATCH Superseded John Hubbard 3 participants 11 messages Aug 11, 2021 – Aug 12, 2021

Subsystems (6)
Memory management VFS MM core s390 +2 more subsystems

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

John Hubbard

author reviewer
#0 lore raw

v2 cover A few gup refactorings and documentation updates

OK, here is v2 with changes as recommended by Christoph Hellwig and Matthew Wilcox (thanks!):

* Changed refs to @refs, and added some more documentation as well.

* Completely removed try_get_page(). (I'm adding more people and lists to Cc, because of those call site changes.)

* Reversed the logic in try_grab_page() to make it a touch more readable.

Also, this has been rebased to today's linux-next (next-20210810), and re-tested on that.

Here is the v1 cover letter, edited slightly to keep up with the latest story.

While reviewing some of the other things going on around gup.c, I noticed that the documentation was wrong for a few of the routines that I wrote. And then I noticed that there was some significant code duplication too. So this fixes those issues.

This is not entirely risk-free, but after looking closely at this, I think it's actually a useful improvement, getting rid of the code duplication here.

However, it is possible I've overlooked something. I did some local LTP and other testing on an x86 test machine but failed to find any problems yet.

And the original v1 is here:
https://lore.kernel.org/r/[email protected]

John Hubbard (3):
mm/gup: documentation corrections for gup/pup
mm/gup: small refactoring: simplify try_grab_page() mm/gup: Remove try_get_page(), call try_get_compound_head() directly

 arch/s390/mm/fault.c |  2 +-
 fs/pipe.c            |  2 +-
 include/linux/mm.h   | 10 +-----
 mm/gup.c             | 79 ++++++++++++++++++++++----------------------
 4 files changed, 43 insertions(+), 50 deletions(-)

--
2.32.0

author reviewer
#1 lore raw

PATCH v2 1/3 mm/gup: documentation corrections for gup/pup

The documentation for try_grab_compound_head() and try_grab_page() has fallen a little out of date. Update and clarify a few points.

Also make it kerneldoc-correct, by adding @args documentation.

Cc Christoph Hellwig <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 mm/gup.c | 27 ++++++++++++++++++++-------
 1 file changed, 20 insertions(+), 7 deletions(-)
mm/gup.c +20 -7
index 2630ed1bb4f4..52f08e3177e9 100644
@@ -92,10 +92,17 @@ static inline struct page *try_get_compound_head(struct page *page, int refs)
 	return head;
 }
 
-/*
+/**
  * try_grab_compound_head() - attempt to elevate a page's refcount, by a
  * flags-dependent amount.
  *
+ * Even though the name includes "compound_head", this function is still
+ * appropriate for callers that have a non-compound @page to get.
+ *
+ * @page:  pointer to page to be grabbed
+ * @refs:  the value to (effectively) add to the page's refcount
+ * @flags: gup flags: these are the FOLL_* flag values.
+ *
  * "grab" names in this file mean, "look at flags to decide whether to use
  * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount.
  *
@@ -103,8 +110,14 @@ static inline struct page *try_get_compound_head(struct page *page, int refs)
  * same time. (That's true throughout the get_user_pages*() and
  * pin_user_pages*() APIs.) Cases:
  *
- *    FOLL_GET: page's refcount will be incremented by 1.
- *    FOLL_PIN: page's refcount will be incremented by GUP_PIN_COUNTING_BIAS.
+ *    FOLL_GET: page's refcount will be incremented by @refs.
+ *
+ *    FOLL_PIN on compound pages that are > two pages long: page's refcount will
+ *    be incremented by @refs, and page[2].hpage_pinned_refcount will be
+ *    incremented by @refs * GUP_PIN_COUNTING_BIAS.
+ *
+ *    FOLL_PIN on normal pages, or compound pages that are two pages long:
+ *    page's refcount will be incremented by @refs * GUP_PIN_COUNTING_BIAS.
  *
  * Return: head page (with refcount appropriately incremented) for success, or
  * NULL upon failure. If neither FOLL_GET nor FOLL_PIN was set, that's
@@ -141,6 +154,8 @@ __maybe_unused struct page *try_grab_compound_head(struct page *page,
 		 *
 		 * However, be sure to *also* increment the normal page refcount
 		 * field at least once, so that the page really is pinned.
+		 * That's why the refcount from the earlier
+		 * try_get_compound_head() is left intact.
 		 */
 		if (hpage_pincount_available(page))
 			hpage_pincount_add(page, refs);
@@ -184,10 +199,8 @@ static void put_compound_head(struct page *page, int refs, unsigned int flags)
  * @flags:   gup flags: these are the FOLL_* flag values.
  *
  * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same
- * time. Cases:
- *
- *    FOLL_GET: page's refcount will be incremented by 1.
- *    FOLL_PIN: page's refcount will be incremented by GUP_PIN_COUNTING_BIAS.
+ * time. Cases: please see the try_grab_compound_head() documentation, with
+ * "refs=1".
  *
  * Return: true for success, or if no action was required (if neither FOLL_PIN
  * nor FOLL_GET was set, nothing is done). False for failure: FOLL_GET or
Signature · 1 line
2.32.0
author reviewer
#2 lore raw

PATCH v2 2/3 mm/gup: small refactoring: simplify try_grab_page()

In reply to John Hubbard

try_grab_page() does the same thing as try_grab_compound_head(..., refs=1, ...), just with a different API. So there is a lot of code duplication there.

Change try_grab_page() to call try_grab_compound_head(), while keeping the API contract identical for callers.

Cc Christoph Hellwig <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 mm/gup.c | 31 +++----------------------------
 1 file changed, 3 insertions(+), 28 deletions(-)
mm/gup.c +3 -28
index 52f08e3177e9..64798d6b5043 100644
@@ -208,35 +208,10 @@ static void put_compound_head(struct page *page, int refs, unsigned int flags)
  */
 bool __must_check try_grab_page(struct page *page, unsigned int flags)
 {
-	WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN));
+	if (!(flags & (FOLL_GET | FOLL_PIN)))
+		return true;
 
-	if (flags & FOLL_GET)
-		return try_get_page(page);
-	else if (flags & FOLL_PIN) {
-		int refs = 1;
-
-		page = compound_head(page);
-
-		if (WARN_ON_ONCE(page_ref_count(page) <= 0))
-			return false;
-
-		if (hpage_pincount_available(page))
-			hpage_pincount_add(page, 1);
-		else
-			refs = GUP_PIN_COUNTING_BIAS;
-
-		/*
-		 * Similar to try_grab_compound_head(): even if using the
-		 * hpage_pincount_add/_sub() routines, be sure to
-		 * *also* increment the normal page refcount field at least
-		 * once, so that the page really is pinned.
-		 */
-		page_ref_add(page, refs);
-
-		mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1);
-	}
-
-	return true;
+	return try_grab_compound_head(page, 1, flags) != NULL;
 }
 
 /**
Signature · 1 line
2.32.0
author reviewer
#4 lore raw

Re: [PATCH v2 2/3] mm/gup: small refactoring: simplify try_grab_page()

Quoted 6 lines · Christoph Hellwig

Looks good,

Note: the __maybe_unused on try_grab_compound_head should be dropped now that there is always a user.

Good point, fixed in v3.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
author reviewer
#5 lore raw

PATCH v2 3/3 mm/gup: Remove try_get_page(), call try_get_compound_head() directly

In reply to John Hubbard

try_get_page() is very similar to try_get_compound_head(), and in fact try_get_page() has fallen a little behind in terms of maintenance:
try_get_compound_head() handles speculative page references more thoroughly.

There are only two try_get_page() callsites, so just call try_get_compound_head() directly from those, and remove try_get_page() entirely.

Also, seeing as how this changes try_get_compound_head() into a non-static function, provide some kerneldoc documentation for it.

Cc Christoph Hellwig <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 arch/s390/mm/fault.c |  2 +-
 fs/pipe.c            |  2 +-
 include/linux/mm.h   | 10 +---------
 mm/gup.c             | 21 +++++++++++++++++----
 4 files changed, 20 insertions(+), 15 deletions(-)
arch/s390/mm/fault.c +1 -1
index 212632d57db9..fe1d2c1dbe3b 100644
@@ -822,7 +822,7 @@ void do_secure_storage_access(struct pt_regs *regs)
 		break;
 	case KERNEL_FAULT:
 		page = phys_to_page(addr);
-		if (unlikely(!try_get_page(page)))
+		if (unlikely(try_get_compound_head(page, 1) == NULL))
 			break;
 		rc = arch_make_page_accessible(page);
 		put_page(page);
fs/pipe.c +1 -1
index 8e6ef62aeb1c..06ba9df37410 100644
@@ -191,7 +191,7 @@ EXPORT_SYMBOL(generic_pipe_buf_try_steal);
  */
 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
 {
-	return try_get_page(buf->page);
+	return try_get_compound_head(buf->page, 1) != NULL;
 }
 EXPORT_SYMBOL(generic_pipe_buf_get);
 
include/linux/mm.h +1 -9
index ce8fc0fd6d6e..cd00d1222235 100644
@@ -1207,15 +1207,7 @@ bool __must_check try_grab_page(struct page *page, unsigned int flags);
 __maybe_unused struct page *try_grab_compound_head(struct page *page, int refs,
 						   unsigned int flags);
 
-
-static inline __must_check bool try_get_page(struct page *page)
-{
-	page = compound_head(page);
-	if (WARN_ON_ONCE(page_ref_count(page) <= 0))
-		return false;
-	page_ref_inc(page);
-	return true;
-}
+struct page *try_get_compound_head(struct page *page, int refs);
 
 /**
  * folio_put - Decrement the reference count on a folio.
mm/gup.c +17 -4
index 64798d6b5043..c2d19d370c99 100644
@@ -62,11 +62,24 @@ static void put_page_refs(struct page *page, int refs)
 	put_page(page);
 }
 
-/*
- * Return the compound head page with ref appropriately incremented,
- * or NULL if that failed.
+/**
+ * try_get_compound_head() - return the compound head page with refcount
+ * appropriately incremented, or NULL if that failed.
+ *
+ * This handles potential refcount overflow correctly. It also works correctly
+ * for various lockless get_user_pages()-related callers, due to the use of
+ * page_cache_add_speculative().
+ *
+ * Even though the name includes "compound_head", this function is still
+ * appropriate for callers that have a non-compound @page to get.
+ *
+ * @page:  pointer to page to be gotten
+ * @refs:  the value to add to the page's refcount
+ *
+ * Return: head page (with refcount appropriately incremented) for success, or
+ * NULL upon failure.
  */
-static inline struct page *try_get_compound_head(struct page *page, int refs)
+struct page *try_get_compound_head(struct page *page, int refs)
 {
 	struct page *head = compound_head(page);
 
Signature · 1 line
2.32.0
#6 lore raw

Re: [PATCH v2 3/3] mm/gup: Remove try_get_page(), call try_get_compound_head() directly

I agree that try_get_page() should probably be removed entirely; is there a reason you didn't in v2 of the patch?

I'm also curious why you changed try_get_compound_head() into a routine from an inline.

If you want to retain try_get_page() it should be an inline as well, especially in its current implementation.

William Kucharski

Quoted 108 lines · John Hubbard

try_get_page() is very similar to try_get_compound_head(), and in fact try_get_page() has fallen a little behind in terms of maintenance:
try_get_compound_head() handles speculative page references more thoroughly.

There are only two try_get_page() callsites, so just call try_get_compound_head() directly from those, and remove try_get_page() entirely.

Also, seeing as how this changes try_get_compound_head() into a non-static function, provide some kerneldoc documentation for it.

Cc Christoph Hellwig <[email protected]>
Cc Matthew Wilcox <[email protected]>

---

arch/s390/mm/fault.c |  2 +-
fs/pipe.c            |  2 +-
include/linux/mm.h   | 10 +---------
mm/gup.c             | 21 +++++++++++++++++----
4 files changed, 20 insertions(+), 15 deletions(-)
arch/s390/mm/fault.c +1 -1
index 212632d57db9..fe1d2c1dbe3b 100644
@@ -822,7 +822,7 @@ void do_secure_storage_access(struct pt_regs *regs)
 		break;
 	case KERNEL_FAULT:
 		page = phys_to_page(addr);
-		if (unlikely(!try_get_page(page)))
+		if (unlikely(try_get_compound_head(page, 1) == NULL))
 			break;
 		rc = arch_make_page_accessible(page);
 		put_page(page);
fs/pipe.c +1 -1
index 8e6ef62aeb1c..06ba9df37410 100644
@@ -191,7 +191,7 @@ EXPORT_SYMBOL(generic_pipe_buf_try_steal);
 */
 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
 {
-	return try_get_page(buf->page);
+	return try_get_compound_head(buf->page, 1) != NULL;
 }
 EXPORT_SYMBOL(generic_pipe_buf_get);
 
include/linux/mm.h +1 -9
index ce8fc0fd6d6e..cd00d1222235 100644
@@ -1207,15 +1207,7 @@ bool __must_check try_grab_page(struct page *page, unsigned int flags);
 __maybe_unused struct page *try_grab_compound_head(struct page *page, int refs,
 						   unsigned int flags);
 
-
-static inline __must_check bool try_get_page(struct page *page)
-{
-	page = compound_head(page);
-	if (WARN_ON_ONCE(page_ref_count(page) <= 0))
-		return false;
-	page_ref_inc(page);
-	return true;
-}
+struct page *try_get_compound_head(struct page *page, int refs);
 
 /**
 * folio_put - Decrement the reference count on a folio.
mm/gup.c +17 -4
index 64798d6b5043..c2d19d370c99 100644
@@ -62,11 +62,24 @@ static void put_page_refs(struct page *page, int refs)
 	put_page(page);
 }
 
-/*
- * Return the compound head page with ref appropriately incremented,
- * or NULL if that failed.
+/**
+ * try_get_compound_head() - return the compound head page with refcount
+ * appropriately incremented, or NULL if that failed.
+ *
+ * This handles potential refcount overflow correctly. It also works correctly
+ * for various lockless get_user_pages()-related callers, due to the use of
+ * page_cache_add_speculative().
+ *
+ * Even though the name includes "compound_head", this function is still
+ * appropriate for callers that have a non-compound @page to get.
+ *
+ * @page:  pointer to page to be gotten
+ * @refs:  the value to add to the page's refcount
+ *
+ * Return: head page (with refcount appropriately incremented) for success, or
+ * NULL upon failure.
 */
-static inline struct page *try_get_compound_head(struct page *page, int refs)
+struct page *try_get_compound_head(struct page *page, int refs)
 {
 	struct page *head = compound_head(page);
 
Signature · 1 line
2.32.0
author reviewer
#7 lore raw

Re: [PATCH v2 3/3] mm/gup: Remove try_get_page(), call try_get_compound_head() directly

Quoted 2 lines · William Kucharski

I agree that try_get_page() should probably be removed entirely; is there a reason you didn't in v2 of the patch?

Hi William,

This patch *does* remove try_get_page() entirely! Look below. I'll reply inline, below, to show where that happens.

Quoted 3 lines

I'm also curious why you changed try_get_compound_head() into a routine from an inline.

It was part of the change to make it available to callers outside of gup.c. try_get_compound_head() is slightly messy and doesn't like to live in mm.h, because it calls page_cache_add_speculative(), which lives in linux-pagemap.h, which, in turn, has its own set of different headers that it pulls in.

So, leaving it in gup.c, and exposing it to the other callers as a non-static function, seemed appropriate here.

Quoted 64 lines

If you want to retain try_get_page() it should be an inline as well, especially in its current implementation.

William Kucharski

Quoted 58 lines · John Hubbard

try_get_page() is very similar to try_get_compound_head(), and in fact try_get_page() has fallen a little behind in terms of maintenance:
try_get_compound_head() handles speculative page references more thoroughly.

There are only two try_get_page() callsites, so just call try_get_compound_head() directly from those, and remove try_get_page() entirely.

Also, seeing as how this changes try_get_compound_head() into a non-static function, provide some kerneldoc documentation for it.

Cc Christoph Hellwig <[email protected]>
Cc Matthew Wilcox <[email protected]>

---

arch/s390/mm/fault.c |  2 +-
fs/pipe.c            |  2 +-
include/linux/mm.h   | 10 +---------
mm/gup.c             | 21 +++++++++++++++++----
4 files changed, 20 insertions(+), 15 deletions(-)
arch/s390/mm/fault.c +1 -1
index 212632d57db9..fe1d2c1dbe3b 100644
@@ -822,7 +822,7 @@ void do_secure_storage_access(struct pt_regs *regs)
 		break;
 	case KERNEL_FAULT:
 		page = phys_to_page(addr);
-		if (unlikely(!try_get_page(page)))
+		if (unlikely(try_get_compound_head(page, 1) == NULL))
 			break;
 		rc = arch_make_page_accessible(page);
 		put_page(page);
fs/pipe.c +1 -1
index 8e6ef62aeb1c..06ba9df37410 100644
@@ -191,7 +191,7 @@ EXPORT_SYMBOL(generic_pipe_buf_try_steal);
  */
 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
 {
-	return try_get_page(buf->page);
+	return try_get_compound_head(buf->page, 1) != NULL;
 }
 EXPORT_SYMBOL(generic_pipe_buf_get);
 
include/linux/mm.h +0 -0
index ce8fc0fd6d6e..cd00d1222235 100644
@@ -1207,15 +1207,7 @@ bool __must_check try_grab_page(struct page *page, unsigned int flags);
 __maybe_unused struct page *try_grab_compound_head(struct page *page, int refs,
 						   unsigned int flags);
 
Quoted 9 lines

include/linux/mm.h bool __must_check try_grab_page(struct page *page, unsigned int flags); · line 1218

Quoted 9 lines
include/linux/mm.h +0 -9
-
-static inline __must_check bool try_get_page(struct page *page)
-{
-	page = compound_head(page);
-	if (WARN_ON_ONCE(page_ref_count(page) <= 0))
-		return false;
-	page_ref_inc(page);
-	return true;
-}

This is where try_get_page() is removed entirely.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
Quoted 43 lines
Quoted 41 lines
include/linux/mm.h +1 -0
+struct page *try_get_compound_head(struct page *page, int refs);
 
 /**
  * folio_put - Decrement the reference count on a folio.
mm/gup.c +17 -4
index 64798d6b5043..c2d19d370c99 100644
@@ -62,11 +62,24 @@ static void put_page_refs(struct page *page, int refs)
 	put_page(page);
 }
 
-/*
- * Return the compound head page with ref appropriately incremented,
- * or NULL if that failed.
+/**
+ * try_get_compound_head() - return the compound head page with refcount
+ * appropriately incremented, or NULL if that failed.
+ *
+ * This handles potential refcount overflow correctly. It also works correctly
+ * for various lockless get_user_pages()-related callers, due to the use of
+ * page_cache_add_speculative().
+ *
+ * Even though the name includes "compound_head", this function is still
+ * appropriate for callers that have a non-compound @page to get.
+ *
+ * @page:  pointer to page to be gotten
+ * @refs:  the value to add to the page's refcount
+ *
+ * Return: head page (with refcount appropriately incremented) for success, or
+ * NULL upon failure.
  */
-static inline struct page *try_get_compound_head(struct page *page, int refs)
+struct page *try_get_compound_head(struct page *page, int refs)
 {
 	struct page *head = compound_head(page);
 
Signature · 1 line
2.32.0
Reviewed-by
#8 lore raw

Re: [PATCH v2 3/3] mm/gup: Remove try_get_page(), call try_get_compound_head() directly

Quoted 12 lines · John Hubbard
Quoted 2 lines · William Kucharski

I agree that try_get_page() should probably be removed entirely; is there a reason you didn't in v2 of the patch?

Hi William,

This patch *does* remove try_get_page() entirely! Look below. I'll reply inline, below, to show where that happens.

Ah, my bad. I was conflating it with try_grab_page() in patch 2/3, which also seems like it should be an inline, but given your explanation re:
try_get_compound_head() it makes perfect sense.

For the series:

Reviewed-by on v2 3/3
#9 lore raw

Re: [PATCH v2 3/3] mm/gup: Remove try_get_page(), call try_get_compound_head() directly

In reply to John Hubbard

arch/s390/mm/fault.c void do_secure_storage_access(struct pt_regs *regs) · line 825

Quoted 2 lines · John Hubbard
arch/s390/mm/fault.c +1 -1
-		if (unlikely(!try_get_page(page)))
+		if (unlikely(try_get_compound_head(page, 1) == NULL))

Why not a simple ! instead of the == NULL?

Otherwise looks good:

author reviewer
#10 lore raw

Re: [PATCH v2 3/3] mm/gup: Remove try_get_page(), call try_get_compound_head() directly

Quoted 5 lines · Christoph Hellwig

arch/s390/mm/fault.c void do_secure_storage_access(struct pt_regs *regs) · line 825

Quoted 2 lines · John Hubbard
arch/s390/mm/fault.c +1 -1
-		if (unlikely(!try_get_page(page)))
+		if (unlikely(try_get_compound_head(page, 1) == NULL))

Why not a simple ! instead of the == NULL?

I'll fix those up and post a v4.

Quoted 5 lines

Otherwise looks good:

Thanks for the reviews!

thanks,

Signature · 2 lines
John Hubbard
NVIDIA