Skip to contentSkip to conversation

x86/mm/pat: (un)track_pfn_copy() fix + improvements

PATCH Superseded David Hildenbrand 5 participants 10 messages Apr 4, 2025 – Apr 8, 2025

Subsystems (4)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/mm/pat: (un)track_pfn_copy() fix + improvements

We got a late smatch warning and some additional review feedback.

smatch warnings:
mm/memory.c:1428 copy_page_range() error: uninitialized symbol 'pfn'.

We actually use the pfn only when it is properly initialized; however, we may pass an uninitialized value to a function -- although it will not use it that likely still is UB in C.

Fix it by always initializing pfn when track_pfn_copy() returns 0 -- just as we document ("On success, stores the pfn to be passed to untrack_pfn_copy()"). In addition, to avoid further wrong smatch warnings, just initialize pfn = 0 in the caller as well.

While at it, clarify the doc of untrack_pfn_copy(), that internal checks make sure if we actually have to untrack anything.

Fixes dc84bc2aba85 ("x86/mm/pat: Fix VM_PAT handling when fork() fails in copy_page_range()")
Cc Andrew Morton <[email protected]>
Cc Lorenzo Stoakes <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc Rik van Riel <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Linus Torvalds <[email protected]>
---
 arch/x86/mm/pat/memtype.c | 4 +++-
 include/linux/pgtable.h   | 5 ++++-
 mm/memory.c               | 2 +-
 3 files changed, 8 insertions(+), 3 deletions(-)
arch/x86/mm/pat/memtype.c +3 -1
index 72d8cbc611583..9ad3e5b055d8a 100644
@@ -992,8 +992,10 @@ int track_pfn_copy(struct vm_area_struct *dst_vma,
 	pgprot_t pgprot;
 	int rc;
 
-	if (!(src_vma->vm_flags & VM_PAT))
+	if (!(src_vma->vm_flags & VM_PAT)) {
+		*pfn = 0;
 		return 0;
+	}
 
 	/*
 	 * Duplicate the PAT information for the dst VMA based on the src
include/linux/pgtable.h +4 -1
index e2b705c149454..9457064292141 100644
@@ -1517,12 +1517,15 @@ static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
 		struct vm_area_struct *src_vma, unsigned long *pfn)
 {
+	*pfn = 0;
 	return 0;
 }
 
 /*
  * untrack_pfn_copy is called when a VM_PFNMAP VMA failed to copy during
- * copy_page_range(), but after track_pfn_copy() was already called.
+ * copy_page_range(), but after track_pfn_copy() was already called. Can
+ * be called even if track_pfn_copy() did not actually track anything:
+ * handled internally.
  */
 static inline void untrack_pfn_copy(struct vm_area_struct *dst_vma,
 		unsigned long pfn)
mm/memory.c +1 -1
index 2d8c265fc7d60..1a35165622e1c 100644
@@ -1361,7 +1361,7 @@ copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
 	struct mm_struct *dst_mm = dst_vma->vm_mm;
 	struct mm_struct *src_mm = src_vma->vm_mm;
 	struct mmu_notifier_range range;
-	unsigned long next, pfn;
+	unsigned long next, pfn = 0;
 	bool is_cow;
 	int ret;
 
Signature · 1 line
2.48.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 17 lines · David Hildenbrand

We got a late smatch warning and some additional review feedback.

smatch warnings:
mm/memory.c:1428 copy_page_range() error: uninitialized symbol 'pfn'.

We actually use the pfn only when it is properly initialized; however, we may pass an uninitialized value to a function -- although it will not use it that likely still is UB in C.

Fix it by always initializing pfn when track_pfn_copy() returns 0 -- just as we document ("On success, stores the pfn to be passed to untrack_pfn_copy()"). In addition, to avoid further wrong smatch warnings, just initialize pfn = 0 in the caller as well.

While at it, clarify the doc of untrack_pfn_copy(), that internal checks make sure if we actually have to untrack anything.

Quoted 16 lines
Fixes dc84bc2aba85 ("x86/mm/pat: Fix VM_PAT handling when fork() fails in copy_page_range()")
Cc Andrew Morton <[email protected]>
Cc Lorenzo Stoakes <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc Rik van Riel <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Linus Torvalds <[email protected]>

LGTM,

Quoted 59 lines
---
 arch/x86/mm/pat/memtype.c | 4 +++-
 include/linux/pgtable.h   | 5 ++++-
 mm/memory.c               | 2 +-
 3 files changed, 8 insertions(+), 3 deletions(-)
arch/x86/mm/pat/memtype.c +3 -1
index 72d8cbc611583..9ad3e5b055d8a 100644
@@ -992,8 +992,10 @@ int track_pfn_copy(struct vm_area_struct *dst_vma,
 	pgprot_t pgprot;
 	int rc;
 
-	if (!(src_vma->vm_flags & VM_PAT))
+	if (!(src_vma->vm_flags & VM_PAT)) {
+		*pfn = 0;
 		return 0;
+	}
 
 	/*
 	 * Duplicate the PAT information for the dst VMA based on the src
include/linux/pgtable.h +4 -1
index e2b705c149454..9457064292141 100644
@@ -1517,12 +1517,15 @@ static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
 		struct vm_area_struct *src_vma, unsigned long *pfn)
 {
+	*pfn = 0;
 	return 0;
 }
 
 /*
  * untrack_pfn_copy is called when a VM_PFNMAP VMA failed to copy during
- * copy_page_range(), but after track_pfn_copy() was already called.
+ * copy_page_range(), but after track_pfn_copy() was already called. Can
+ * be called even if track_pfn_copy() did not actually track anything:
+ * handled internally.
  */
 static inline void untrack_pfn_copy(struct vm_area_struct *dst_vma,
 		unsigned long pfn)
mm/memory.c +1 -2
index 2d8c265fc7d60..1a35165622e1c 100644
@@ -1361,7 +1361,7 @@ copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
 	struct mm_struct *dst_mm = dst_vma->vm_mm;
 	struct mm_struct *src_mm = src_vma->vm_mm;
 	struct mmu_notifier_range range;
-	unsigned long next, pfn;
+	unsigned long next, pfn = 0;
 	bool is_cow;
 	int ret;
 
--

2.48.1

#2 lore raw
In reply to David Hildenbrand
Quoted 7 lines · David Hildenbrand

We got a late smatch warning and some additional review feedback.

smatch warnings:
mm/memory.c:1428 copy_page_range() error: uninitialized symbol 'pfn'.

arch/x86/mm/pat/memtype.c +1 -1
-	if (!(src_vma->vm_flags & VM_PAT))
+	if (!(src_vma->vm_flags & VM_PAT)) {

arch/x86/mm/pat/memtype.c int track_pfn_copy(struct vm_area_struct *dst_vma,

Quoted 10 lines
arch/x86/mm/pat/memtype.c +3 -0
+		*pfn = 0;
 		return 0;
+	}
 
 static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
 		struct vm_area_struct *src_vma, unsigned long *pfn)
 {
+	*pfn = 0;
 	return 0;
 }

That's way too ugly. There's nothing wrong with not touching 'pfn' in the error path: in fact it's pretty standard API where output pointers may not get set on errors.

If Smatch has a problem with it, Smatch should be fixed, or the false positive warning should be worked around by initializing 'pfn' in the callers.

Thanks,

Ingo

#3 lore raw
Quoted 14 lines · Ingo Molnar
Quoted 10 lines · David Hildenbrand

We got a late smatch warning and some additional review feedback.

smatch warnings:
mm/memory.c:1428 copy_page_range() error: uninitialized symbol 'pfn'.

(quoted diff) +3 -1
-	if (!(src_vma->vm_flags & VM_PAT))
+	if (!(src_vma->vm_flags & VM_PAT)) {
+		*pfn = 0;
  		return 0;
+	}
Quoted 10 lines

Review

Quoted 6 lines

static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
struct vm_area_struct *src_vma, unsigned long *pfn)
{

(quoted diff) +1 -0
+	*pfn = 0;
  	return 0;
  }

That's way too ugly. There's nothing wrong with not touching 'pfn' in the error path: in fact it's pretty standard API where output pointers may not get set on errors.

We're not concerned about the error path, though.

Quoted 4 lines

If Smatch has a problem with it, Smatch should be fixed, or the false positive warning should be worked around by initializing 'pfn' in the callers.

We could adjust the documentation of track_pfn_copy, to end up with the following:

include/linux/pgtable.h +6 -3
index e2b705c149454..b50447ef1c921 100644
@@ -1511,8 +1511,9 @@ static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 
 /*
  * track_pfn_copy is called when a VM_PFNMAP VMA is about to get the page
- * tables copied during copy_page_range(). On success, stores the pfn to be
- * passed to untrack_pfn_copy().
+ * tables copied during copy_page_range(). Will store the pfn to be
+ * passed to untrack_pfn_copy() only if there is something to be untracked.
+ * Callers should initialize the pfn to 0.
  */
 static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
                struct vm_area_struct *src_vma, unsigned long *pfn)
@@ -1522,7 +1523,9 @@ static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
 
 /*
  * untrack_pfn_copy is called when a VM_PFNMAP VMA failed to copy during
- * copy_page_range(), but after track_pfn_copy() was already called.
+ * copy_page_range(), but after track_pfn_copy() was already called. Can
+ * be called even if track_pfn_copy() did not actually track anything:
+ * handled internally.
  */
 static inline void untrack_pfn_copy(struct vm_area_struct *dst_vma,
                unsigned long pfn)
mm/memory.c +1 -1
index 2d8c265fc7d60..1a35165622e1c 100644
@@ -1361,7 +1361,7 @@ copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
        struct mm_struct *dst_mm = dst_vma->vm_mm;
        struct mm_struct *src_mm = src_vma->vm_mm;
        struct mmu_notifier_range range;
-       unsigned long next, pfn;
+       unsigned long next, pfn = 0;
        bool is_cow;
        int ret;
 
Signature · 3 lines
Cheers,

David / dhildenb
#4 lore raw
Quoted 26 lines · David Hildenbrand
Quoted 14 lines · Ingo Molnar
Quoted 10 lines · David Hildenbrand

We got a late smatch warning and some additional review feedback.

smatch warnings:
mm/memory.c:1428 copy_page_range() error: uninitialized symbol 'pfn'.

(quoted diff) +3 -1
-	if (!(src_vma->vm_flags & VM_PAT))
+	if (!(src_vma->vm_flags & VM_PAT)) {
+		*pfn = 0;
  		return 0;
+	}
Quoted 10 lines

Review

Quoted 6 lines

static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
struct vm_area_struct *src_vma, unsigned long *pfn)
{

(quoted diff) +1 -0
+	*pfn = 0;
  	return 0;
  }

That's way too ugly. There's nothing wrong with not touching 'pfn' in the error path: in fact it's pretty standard API where output pointers may not get set on errors.

Quoted 1 line

We're not concerned about the error path, though.

Sorry, indeed, not an error path, but the !VM_PAT path above - but still a similar argument applies IMHO.

Quoted 37 lines
Quoted 3 lines

If Smatch has a problem with it, Smatch should be fixed, or the false positive warning should be worked around by initializing 'pfn' in the callers.

We could adjust the documentation of track_pfn_copy, to end up with the following:

include/linux/pgtable.h +6 -3
index e2b705c149454..b50447ef1c921 100644
@@ -1511,8 +1511,9 @@ static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 /*
  * track_pfn_copy is called when a VM_PFNMAP VMA is about to get the page
- * tables copied during copy_page_range(). On success, stores the pfn to be
- * passed to untrack_pfn_copy().
+ * tables copied during copy_page_range(). Will store the pfn to be
+ * passed to untrack_pfn_copy() only if there is something to be untracked.
+ * Callers should initialize the pfn to 0.
  */
 static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
                struct vm_area_struct *src_vma, unsigned long *pfn)
@@ -1522,7 +1523,9 @@ static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
 /*
  * untrack_pfn_copy is called when a VM_PFNMAP VMA failed to copy during
- * copy_page_range(), but after track_pfn_copy() was already called.
+ * copy_page_range(), but after track_pfn_copy() was already called. Can
+ * be called even if track_pfn_copy() did not actually track anything:
+ * handled internally.
  */
 static inline void untrack_pfn_copy(struct vm_area_struct *dst_vma,
                unsigned long pfn)
mm/memory.c +0 -0
index 2d8c265fc7d60..1a35165622e1c 100644

mm/memory.c copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) · line 1364

Quoted 5 lines
mm/memory.c +1 -1
@@ -1361,7 +1361,7 @@ copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
        struct mm_struct *dst_mm = dst_vma->vm_mm;
        struct mm_struct *src_mm = src_vma->vm_mm;
        struct mmu_notifier_range range;
-       unsigned long next, pfn;
+       unsigned long next, pfn = 0;

Ack.

I hate it how uninitialized variables are even a thing in C, and why there's no compiler switch to turn it off for the kernel. (At least for non-struct variables. Even for structs I would zero-initialize and *maybe* allow a non-initialized opt-in for cases where it matters. It matters in very few cases in praxis. And don't get me started about the stupidity that is to not initialize holes in struct members ...)

Over the decades we've lived through numerous nasty bugs for very little tangible code generation benefits.

Thanks,

Ingo

#5 lore raw
Quoted 86 lines · Ingo Molnar
Quoted 26 lines · David Hildenbrand
Quoted 14 lines · Ingo Molnar
Quoted 10 lines · David Hildenbrand

We got a late smatch warning and some additional review feedback.

smatch warnings:
mm/memory.c:1428 copy_page_range() error: uninitialized symbol 'pfn'.

(quoted diff) +3 -1
-	if (!(src_vma->vm_flags & VM_PAT))
+	if (!(src_vma->vm_flags & VM_PAT)) {
+		*pfn = 0;
   		return 0;
+	}
Quoted 10 lines

Review

Quoted 6 lines

static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
struct vm_area_struct *src_vma, unsigned long *pfn)
{

(quoted diff) +1 -0
+	*pfn = 0;
   	return 0;
   }

That's way too ugly. There's nothing wrong with not touching 'pfn' in the error path: in fact it's pretty standard API where output pointers may not get set on errors.

Quoted 1 line

We're not concerned about the error path, though.

Sorry, indeed, not an error path, but the !VM_PAT path above - but still a similar argument applies IMHO.

Quoted 37 lines
Quoted 3 lines

If Smatch has a problem with it, Smatch should be fixed, or the false positive warning should be worked around by initializing 'pfn' in the callers.

We could adjust the documentation of track_pfn_copy, to end up with the following:

include/linux/pgtable.h +6 -3
index e2b705c149454..b50447ef1c921 100644
@@ -1511,8 +1511,9 @@ static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
  /*
   * track_pfn_copy is called when a VM_PFNMAP VMA is about to get the page
- * tables copied during copy_page_range(). On success, stores the pfn to be
- * passed to untrack_pfn_copy().
+ * tables copied during copy_page_range(). Will store the pfn to be
+ * passed to untrack_pfn_copy() only if there is something to be untracked.
+ * Callers should initialize the pfn to 0.
   */
  static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
                 struct vm_area_struct *src_vma, unsigned long *pfn)
@@ -1522,7 +1523,9 @@ static inline int track_pfn_copy(struct vm_area_struct *dst_vma,
  /*
   * untrack_pfn_copy is called when a VM_PFNMAP VMA failed to copy during
- * copy_page_range(), but after track_pfn_copy() was already called.
+ * copy_page_range(), but after track_pfn_copy() was already called. Can
+ * be called even if track_pfn_copy() did not actually track anything:
+ * handled internally.
   */
  static inline void untrack_pfn_copy(struct vm_area_struct *dst_vma,
                 unsigned long pfn)
mm/memory.c +0 -0
index 2d8c265fc7d60..1a35165622e1c 100644

mm/memory.c copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) · line 1364

Quoted 5 lines
mm/memory.c +1 -1
@@ -1361,7 +1361,7 @@ copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
         struct mm_struct *dst_mm = dst_vma->vm_mm;
         struct mm_struct *src_mm = src_vma->vm_mm;
         struct mmu_notifier_range range;
-       unsigned long next, pfn;
+       unsigned long next, pfn = 0;

Ack.

I hate it how uninitialized variables are even a thing in C, and why there's no compiler switch to turn it off for the kernel. (At least for non-struct variables. Even for structs I would zero-initialize and *maybe* allow a non-initialized opt-in for cases where it matters. It matters in very few cases in praxis. And don't get me started about the stupidity that is to not initialize holes in struct members ...)

Quoted 2 lines

Over the decades we've lived through numerous nasty bugs for very little tangible code generation benefits.

Ok, let me resend with that. (I'll still tag it as a fix due do the weird UB scenario when passing uninitialized values to a non-inline function ...)

Signature · 3 lines
Cheers,

David / dhildenb
#6 lore raw
In reply to Ingo Molnar
Quoted 13 lines · Ingo Molnar

mm/memory.c copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) · line 1364

Quoted 10 lines
mm/memory.c +1 -1
index 2d8c265fc7d60..1a35165622e1c 100644
@@ -1361,7 +1361,7 @@ copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma)
        struct mm_struct *dst_mm = dst_vma->vm_mm;
        struct mm_struct *src_mm = src_vma->vm_mm;
        struct mmu_notifier_range range;
-       unsigned long next, pfn;
+       unsigned long next, pfn = 0;

Ack.

Quoted 6 lines

I hate it how uninitialized variables are even a thing in C, and why there's no compiler switch to turn it off for the kernel. (At least for non-struct variables. Even for structs I would zero-initialize and *maybe* allow a non-initialized opt-in for cases where it matters. It matters in very few cases in praxis. And don't get me started about the stupidity that is to not initialize holes in struct members ...)

Everyone sane uses CONFIG_INIT_STACK_ALL_ZERO these days.

regards,
dan carpenter

#7 lore raw
Quoted 7 lines · Dan Carpenter
Quoted 5 lines

least for non-struct variables. Even for structs I would zero-initialize and *maybe* allow a non-initialized opt-in for cases where it matters. It matters in very few cases in praxis. And don't get me started about the stupidity that is to not initialize holes in struct members ...)

Everyone sane uses CONFIG_INIT_STACK_ALL_ZERO these days.

Good, although why is this compiler option named so weirdly in Clang:

CC_AUTO_VAR_INIT_ZERO_ENABLER := -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang

Hopefully it is named thusly because Clang has adopted GCC's -ftrivial-auto-var-init=zero?

Thanks,

Ingo

#8 lore raw
Quoted 6 lines · Ingo Molnar

Good, although why is this compiler option named so weirdly in Clang:

CC_AUTO_VAR_INIT_ZERO_ENABLER := -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang

Hopefully it is named thusly because Clang has adopted GCC's -ftrivial-auto-var-init=zero?

Clang did -ftrivial-auto-var-init first, where the original author added both pattern and zero but intended to remove zero once pattern has been optimized enough compared to zero (if I remember and understand correctly), so the "enabler" flag was added to try and make that clear. Eventually, Kees leveraged both Linus's stated desire for initializing stack variables [1] and GCC 12 landing -ftrivial-auto-var-init=zero without a separate enable option to deprecate the "enabler" flag in clang 16 [2] and remove it altogether in clang 18 [3].

[1]: https://lore.kernel.org/CAHk-=wgTM+cN7zyUZacGQDv3DuuoA4LORNPWgb1Y_Z1p4iedNQ@mail.gmail.com/
[2]: https://github.com/llvm/llvm-project/commit/aef03c9b3bed5cef5a1940774b80128aefcb4095
[3]: https://github.com/llvm/llvm-project/commit/00e54d04ae2802d498741097d4b83e898bc99c5b

Cheers,
Nathan

#9 lore raw
Quoted 17 lines · Nathan Chancellor
Quoted 6 lines · Ingo Molnar

Good, although why is this compiler option named so weirdly in Clang:

CC_AUTO_VAR_INIT_ZERO_ENABLER := -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang

Hopefully it is named thusly because Clang has adopted GCC's -ftrivial-auto-var-init=zero?

Clang did -ftrivial-auto-var-init first, where the original author added both pattern and zero but intended to remove zero once pattern has been optimized enough compared to zero (if I remember and understand correctly), so the "enabler" flag was added to try and make that clear. Eventually, Kees leveraged both Linus's stated desire for initializing stack variables [1] and GCC 12 landing -ftrivial-auto-var-init=zero without a separate enable option to deprecate the "enabler" flag in clang 16 [2] and remove it altogether in clang 18 [3].

Quoted 3 lines

[1]: https://lore.kernel.org/CAHk-=wgTM+cN7zyUZacGQDv3DuuoA4LORNPWgb1Y_Z1p4iedNQ@mail.gmail.com/
[2]: https://github.com/llvm/llvm-project/commit/aef03c9b3bed5cef5a1940774b80128aefcb4095
[3]: https://github.com/llvm/llvm-project/commit/00e54d04ae2802d498741097d4b83e898bc99c5b

Cool, thanks for the explanation!

Thanks,

Ingo