Skip to contentSkip to conversation

move per-vma lock into vm_area_struct

PATCH Superseded Suren Baghdasaryan 5 participants 27 messages Nov 20, 2024 – Nov 22, 2024

Subsystems (9)

v4 of 10 Superseded Latest is v10 →

lore
Share & tools
lore

Suren Baghdasaryan

author maintainer
#0 lore raw

v4 cover move per-vma lock into vm_area_struct

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].
Splitting single logical structure into multiple ones leads to more complicated management, extra pointer dereferences and overall less maintainable code. When that split-away part is a lock, it complicates things even further. With no performance benefits, there are no reasons for this split. Merging the vm_lock back into vm_area_struct also allows vm_area_struct to use SLAB_TYPESAFE_BY_RCU later in this patchset. This patchset:
1. moves vm_lock back into vm_area_struct, aligning it at the cacheline boundary and changing the cache to be cacheline-aligned to minimize cacheline sharing;
2. changes vm_area_struct initialization to mark new vma as detached until it is inserted into vma tree;
3. changes vm_area_struct cache to SLAB_TYPESAFE_BY_RCU to allow for their reuse and to minimize call_rcu() calls.
Pagefault microbenchmarks do not show noticeable performance change.

Changes since v3 [4]
- Added SOBs, per Lorenzo Stoakes and Davidlohr Bueso;
- Replaced vma write-locking in vma_mark_attached() with memory barriers to order accesses to vma->detached vs vm_mm/vm_start/vm_end

Patch applies over mm-unstable

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/ZsQyI%2F087V34JoIt@xsang-OptiPlex-9020/
[3] https://lore.kernel.org/all/CAJuCfpEisU8Lfe96AYJDZ+OM4NoPmnw9bP53cT_kbfP_pR+-2g@mail.gmail.com/
[4] https://lore.kernel.org/all/[email protected]/

Suren Baghdasaryan (5):
mm: introduce vma_start_read_locked{_nested} helpers
mm: move per-vma lock into vm_area_struct
mm: mark vma as detached until it's added into vma tree
mm: make vma cache SLAB_TYPESAFE_BY_RCU docs/mm: document latest changes to vm_lock

 Documentation/mm/process_addrs.rst |  10 ++-
 include/linux/mm.h                 | 129 ++++++++++++++++++++++++-----
 include/linux/mm_types.h           |  19 ++---
 kernel/fork.c                      |  88 +++++++-------------
 mm/memory.c                        |  17 ++--
 mm/userfaultfd.c                   |  22 ++---
 mm/vma.c                           |   8 +-
 mm/vma.h                           |   2 +
 tools/testing/vma/vma_internal.h   |  55 +++++-------
 9 files changed, 198 insertions(+), 152 deletions(-)

base-commit: 5a7056135bb69da2ce0a42eb8c07968c1331777b

Signature · 1 line
2.47.0.338.g60cca15819-goog
author maintainer
#1 lore raw

PATCH v4 1/5 mm: introduce vma_start_read_locked{_nested} helpers

Introduce helper functions which can be used to read-lock a VMA when holding mmap_lock for read. Replace direct accesses to vma->vm_lock with these new helpers.

---
 include/linux/mm.h | 24 ++++++++++++++++++++++++
 mm/userfaultfd.c   | 22 +++++-----------------
 2 files changed, 29 insertions(+), 17 deletions(-)
include/linux/mm.h +24 -0
index fecd47239fa9..1ba2e480ae63 100644
@@ -722,6 +722,30 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 	return true;
 }
 
+/*
+ * Use only while holding mmap read lock which guarantees that locking will not
+ * fail (nobody can concurrently write-lock the vma). vma_start_read() should
+ * not be used in such cases because it might fail due to mm_lock_seq overflow.
+ * This functionality is used to obtain vma read lock and drop the mmap read lock.
+ */
+static inline void vma_start_read_locked_nested(struct vm_area_struct *vma, int subclass)
+{
+	mmap_assert_locked(vma->vm_mm);
+	down_read_nested(&vma->vm_lock->lock, subclass);
+}
+
+/*
+ * Use only while holding mmap read lock which guarantees that locking will not
+ * fail (nobody can concurrently write-lock the vma). vma_start_read() should
+ * not be used in such cases because it might fail due to mm_lock_seq overflow.
+ * This functionality is used to obtain vma read lock and drop the mmap read lock.
+ */
+static inline void vma_start_read_locked(struct vm_area_struct *vma)
+{
+	mmap_assert_locked(vma->vm_mm);
+	down_read(&vma->vm_lock->lock);
+}
+
 static inline void vma_end_read(struct vm_area_struct *vma)
 {
 	rcu_read_lock(); /* keeps vma alive till the end of up_read */
mm/userfaultfd.c +5 -17
index 60a0be33766f..87db4b32b82a 100644
@@ -84,16 +84,8 @@ static struct vm_area_struct *uffd_lock_vma(struct mm_struct *mm,
 
 	mmap_read_lock(mm);
 	vma = find_vma_and_prepare_anon(mm, address);
-	if (!IS_ERR(vma)) {
-		/*
-		 * We cannot use vma_start_read() as it may fail due to
-		 * false locked (see comment in vma_start_read()). We
-		 * can avoid that by directly locking vm_lock under
-		 * mmap_lock, which guarantees that nobody can lock the
-		 * vma for write (vma_start_write()) under us.
-		 */
-		down_read(&vma->vm_lock->lock);
-	}
+	if (!IS_ERR(vma))
+		vma_start_read_locked(vma);
 
 	mmap_read_unlock(mm);
 	return vma;
@@ -1476,14 +1468,10 @@ static int uffd_move_lock(struct mm_struct *mm,
 	mmap_read_lock(mm);
 	err = find_vmas_mm_locked(mm, dst_start, src_start, dst_vmap, src_vmap);
 	if (!err) {
-		/*
-		 * See comment in uffd_lock_vma() as to why not using
-		 * vma_start_read() here.
-		 */
-		down_read(&(*dst_vmap)->vm_lock->lock);
+		vma_start_read_locked(*dst_vmap);
 		if (*dst_vmap != *src_vmap)
-			down_read_nested(&(*src_vmap)->vm_lock->lock,
-					 SINGLE_DEPTH_NESTING);
+			vma_start_read_locked_nested(*src_vmap,
+						SINGLE_DEPTH_NESTING);
 	}
 	mmap_read_unlock(mm);
 	return err;
Signature · 1 line
2.47.0.338.g60cca15819-goog
reviewer Reviewed-by on v4 1/5
#2 lore raw

Re: [PATCH v4 1/5] mm: introduce vma_start_read_locked{_nested} helpers

Quoted 7 lines · Suren Baghdasaryan

Introduce helper functions which can be used to read-lock a VMA when holding mmap_lock for read. Replace direct accesses to vma->vm_lock with these new helpers.

author maintainer
#3 lore raw

PATCH v4 2/5 mm: move per-vma lock into vm_area_struct

In reply to Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].
Splitting single logical structure into multiple ones leads to more complicated management, extra pointer dereferences and overall less maintainable code. When that split-away part is a lock, it complicates things even further. With no performance benefits, there are no reasons for this split. Merging the vm_lock back into vm_area_struct also allows vm_area_struct to use SLAB_TYPESAFE_BY_RCU later in this patchset. Move vm_lock back into vm_area_struct, aligning it at the cacheline boundary and changing the cache to be cacheline-aligned as well. With kernel compiled using defconfig, this causes VMA memory consumption to grow from 160 (vm_area_struct) + 40 (vm_lock) bytes to 256 bytes:

slabinfo before:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vma_lock ... 40 102 1 : ...
vm_area_struct ... 160 51 2 : ...

slabinfo after moving vm_lock:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vm_area_struct ... 256 32 2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 50 to 64 pages, which is 5.5MB per 100000 VMAs. Note that the size of this structure is dependent on the kernel configuration and typically the original size is higher than 160 bytes. Therefore these calculations are close to the worst case scenario. A more realistic vm_area_struct usage before this change is:

     <name>           ... <objsize> <objperslab> <pagesperslab> : ...
     vma_lock         ...     40  102    1 : ...
     vm_area_struct   ...    176   46    2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 54 to 64 pages, which is 3.9MB per 100000 VMAs.
This memory consumption growth can be addressed later by optimizing the vm_lock.

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/ZsQyI%2F087V34JoIt@xsang-OptiPlex-9020/
[3] https://lore.kernel.org/all/CAJuCfpEisU8Lfe96AYJDZ+OM4NoPmnw9bP53cT_kbfP_pR+-2g@mail.gmail.com/

---
 include/linux/mm.h               | 28 ++++++++++--------
 include/linux/mm_types.h         |  6 ++--
 kernel/fork.c                    | 49 ++++----------------------------
 tools/testing/vma/vma_internal.h | 33 +++++----------------
 4 files changed, 32 insertions(+), 84 deletions(-)
include/linux/mm.h +16 -12
index 1ba2e480ae63..737c003b0a1e 100644
@@ -684,6 +684,12 @@ static inline void vma_numab_state_free(struct vm_area_struct *vma) {}
 #endif /* CONFIG_NUMA_BALANCING */
 
 #ifdef CONFIG_PER_VMA_LOCK
+static inline void vma_lock_init(struct vm_area_struct *vma)
+{
+	init_rwsem(&vma->vm_lock.lock);
+	vma->vm_lock_seq = UINT_MAX;
+}
+
 /*
  * Try to read-lock a vma. The function is allowed to occasionally yield false
  * locked result to avoid performance overhead, in which case we fall back to
@@ -701,7 +707,7 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 	if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq.sequence))
 		return false;
 
-	if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0))
+	if (unlikely(down_read_trylock(&vma->vm_lock.lock) == 0))
 		return false;
 
 	/*
@@ -716,7 +722,7 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 	 * This pairs with RELEASE semantics in vma_end_write_all().
 	 */
 	if (unlikely(vma->vm_lock_seq == raw_read_seqcount(&vma->vm_mm->mm_lock_seq))) {
-		up_read(&vma->vm_lock->lock);
+		up_read(&vma->vm_lock.lock);
 		return false;
 	}
 	return true;
@@ -731,7 +737,7 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 static inline void vma_start_read_locked_nested(struct vm_area_struct *vma, int subclass)
 {
 	mmap_assert_locked(vma->vm_mm);
-	down_read_nested(&vma->vm_lock->lock, subclass);
+	down_read_nested(&vma->vm_lock.lock, subclass);
 }
 
 /*
@@ -743,13 +749,13 @@ static inline void vma_start_read_locked_nested(struct vm_area_struct *vma, int
 static inline void vma_start_read_locked(struct vm_area_struct *vma)
 {
 	mmap_assert_locked(vma->vm_mm);
-	down_read(&vma->vm_lock->lock);
+	down_read(&vma->vm_lock.lock);
 }
 
 static inline void vma_end_read(struct vm_area_struct *vma)
 {
 	rcu_read_lock(); /* keeps vma alive till the end of up_read */
-	up_read(&vma->vm_lock->lock);
+	up_read(&vma->vm_lock.lock);
 	rcu_read_unlock();
 }
 
@@ -778,7 +784,7 @@ static inline void vma_start_write(struct vm_area_struct *vma)
 	if (__is_vma_write_locked(vma, &mm_lock_seq))
 		return;
 
-	down_write(&vma->vm_lock->lock);
+	down_write(&vma->vm_lock.lock);
 	/*
 	 * We should use WRITE_ONCE() here because we can have concurrent reads
 	 * from the early lockless pessimistic check in vma_start_read().
@@ -786,7 +792,7 @@ static inline void vma_start_write(struct vm_area_struct *vma)
 	 * we should use WRITE_ONCE() for cleanliness and to keep KCSAN happy.
 	 */
 	WRITE_ONCE(vma->vm_lock_seq, mm_lock_seq);
-	up_write(&vma->vm_lock->lock);
+	up_write(&vma->vm_lock.lock);
 }
 
 static inline void vma_assert_write_locked(struct vm_area_struct *vma)
@@ -798,7 +804,7 @@ static inline void vma_assert_write_locked(struct vm_area_struct *vma)
 
 static inline void vma_assert_locked(struct vm_area_struct *vma)
 {
-	if (!rwsem_is_locked(&vma->vm_lock->lock))
+	if (!rwsem_is_locked(&vma->vm_lock.lock))
 		vma_assert_write_locked(vma);
 }
 
@@ -831,6 +837,7 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 
 #else /* CONFIG_PER_VMA_LOCK */
 
+static inline void vma_lock_init(struct vm_area_struct *vma) {}
 static inline bool vma_start_read(struct vm_area_struct *vma)
 		{ return false; }
 static inline void vma_end_read(struct vm_area_struct *vma) {}
@@ -865,10 +872,6 @@ static inline void assert_fault_locked(struct vm_fault *vmf)
 
 extern const struct vm_operations_struct vma_dummy_vm_ops;
 
-/*
- * WARNING: vma_init does not initialize vma->vm_lock.
- * Use vm_area_alloc()/vm_area_free() if vma needs locking.
- */
 static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 {
 	memset(vma, 0, sizeof(*vma));
@@ -877,6 +880,7 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
 	vma_mark_detached(vma, false);
 	vma_numab_state_init(vma);
+	vma_lock_init(vma);
 }
 
 /* Use when VMA is not part of the VMA tree and needs no locking */
include/linux/mm_types.h +4 -2
index 80fef38d9d64..5c4bfdcfac72 100644
@@ -716,8 +716,6 @@ struct vm_area_struct {
 	 * slowpath.
 	 */
 	unsigned int vm_lock_seq;
-	/* Unstable RCU readers are allowed to read this. */
-	struct vma_lock *vm_lock;
 #endif
 
 	/*
@@ -770,6 +768,10 @@ struct vm_area_struct {
 	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
 #endif
 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
+#ifdef CONFIG_PER_VMA_LOCK
+	/* Unstable RCU readers are allowed to read this. */
+	struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+#endif
 } __randomize_layout;
 
 #ifdef CONFIG_NUMA
kernel/fork.c +5 -44

85 lines not on this page · raw message

tools/testing/vma/vma_internal.h +7 -26
index 1d9fc97b8e80..11c2c38ca4e8 100644
@@ -230,10 +230,10 @@ struct vm_area_struct {
 	/*
 	 * Can only be written (using WRITE_ONCE()) while holding both:
 	 *  - mmap_lock (in write mode)
-	 *  - vm_lock->lock (in write mode)
+	 *  - vm_lock.lock (in write mode)
 	 * Can be read reliably while holding one of:
 	 *  - mmap_lock (in read or write mode)
-	 *  - vm_lock->lock (in read or write mode)
+	 *  - vm_lock.lock (in read or write mode)
 	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
 	 * while holding nothing (except RCU to keep the VMA struct allocated).
 	 *
@@ -242,7 +242,7 @@ struct vm_area_struct {
 	 * slowpath.
 	 */
 	unsigned int vm_lock_seq;
-	struct vma_lock *vm_lock;
+	struct vma_lock vm_lock;
 #endif
 
 	/*
@@ -408,17 +408,10 @@ static inline struct vm_area_struct *vma_next(struct vma_iterator *vmi)
 	return mas_find(&vmi->mas, ULONG_MAX);
 }
 
-static inline bool vma_lock_alloc(struct vm_area_struct *vma)
+static inline void vma_lock_init(struct vm_area_struct *vma)
 {
-	vma->vm_lock = calloc(1, sizeof(struct vma_lock));
-
-	if (!vma->vm_lock)
-		return false;
-
-	init_rwsem(&vma->vm_lock->lock);
+	init_rwsem(&vma->vm_lock.lock);
 	vma->vm_lock_seq = UINT_MAX;
-
-	return true;
 }
 
 static inline void vma_assert_write_locked(struct vm_area_struct *);
@@ -439,6 +432,7 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	vma->vm_ops = &vma_dummy_vm_ops;
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
 	vma_mark_detached(vma, false);
+	vma_lock_init(vma);
 }
 
 static inline struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
@@ -449,10 +443,6 @@ static inline struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 		return NULL;
 
 	vma_init(vma, mm);
-	if (!vma_lock_alloc(vma)) {
-		free(vma);
-		return NULL;
-	}
 
 	return vma;
 }
@@ -465,10 +455,7 @@ static inline struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 		return NULL;
 
 	memcpy(new, orig, sizeof(*new));
-	if (!vma_lock_alloc(new)) {
-		free(new);
-		return NULL;
-	}
+	vma_lock_init(new);
 	INIT_LIST_HEAD(&new->anon_vma_chain);
 
 	return new;
@@ -638,14 +625,8 @@ static inline void mpol_put(struct mempolicy *)
 {
 }
 
-static inline void vma_lock_free(struct vm_area_struct *vma)
-{
-	free(vma->vm_lock);
-}
-
 static inline void __vm_area_free(struct vm_area_struct *vma)
 {
-	vma_lock_free(vma);
 	free(vma);
 }
 
Signature · 1 line
2.47.0.338.g60cca15819-goog
reviewer Reviewed-by on v4 2/5
#4 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

Quoted 42 lines · Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].
Splitting single logical structure into multiple ones leads to more complicated management, extra pointer dereferences and overall less maintainable code. When that split-away part is a lock, it complicates things even further. With no performance benefits, there are no reasons for this split. Merging the vm_lock back into vm_area_struct also allows vm_area_struct to use SLAB_TYPESAFE_BY_RCU later in this patchset. Move vm_lock back into vm_area_struct, aligning it at the cacheline boundary and changing the cache to be cacheline-aligned as well. With kernel compiled using defconfig, this causes VMA memory consumption to grow from 160 (vm_area_struct) + 40 (vm_lock) bytes to 256 bytes:

slabinfo before:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vma_lock ... 40 102 1 : ...
vm_area_struct ... 160 51 2 : ...

slabinfo after moving vm_lock:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vm_area_struct ... 256 32 2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 50 to 64 pages, which is 5.5MB per 100000 VMAs. Note that the size of this structure is dependent on the kernel configuration and typically the original size is higher than 160 bytes. Therefore these calculations are close to the worst case scenario. A more realistic vm_area_struct usage before this change is:

     <name>           ... <objsize> <objperslab> <pagesperslab> : ...
     vma_lock         ...     40  102    1 : ...
     vm_area_struct   ...    176   46    2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 54 to 64 pages, which is 3.9MB per 100000 VMAs.
This memory consumption growth can be addressed later by optimizing the vm_lock.

Quoted 6 lines

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/ZsQyI%2F087V34JoIt@xsang-OptiPlex-9020/
[3] https://lore.kernel.org/all/CAJuCfpEisU8Lfe96AYJDZ+OM4NoPmnw9bP53cT_kbfP_pR+-2g@mail.gmail.com/

One question below.

Quoted 12 lines
include/linux/mm_types.h +0 -2
@@ -716,8 +716,6 @@ struct vm_area_struct {
 	 * slowpath.
 	 */
 	unsigned int vm_lock_seq;
-	/* Unstable RCU readers are allowed to read this. */
-	struct vma_lock *vm_lock;
 #endif
 
 	/*

include/linux/mm_types.h struct vm_area_struct { · line 775

Quoted 8 lines
include/linux/mm_types.h +4 -0
@@ -770,6 +768,10 @@ struct vm_area_struct {
 	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
 #endif
 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
+#ifdef CONFIG_PER_VMA_LOCK
+	/* Unstable RCU readers are allowed to read this. */
+	struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+#endif
 } __randomize_layout;

Do we just want 'struct vm_area_struct' to be cacheline aligned or do we want 'struct vma_lock vm_lock' to be on a separate cacheline as well?

author maintainer
#5 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

Quoted 44 lines · Shakeel Butt
Quoted 42 lines · Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].
Splitting single logical structure into multiple ones leads to more complicated management, extra pointer dereferences and overall less maintainable code. When that split-away part is a lock, it complicates things even further. With no performance benefits, there are no reasons for this split. Merging the vm_lock back into vm_area_struct also allows vm_area_struct to use SLAB_TYPESAFE_BY_RCU later in this patchset. Move vm_lock back into vm_area_struct, aligning it at the cacheline boundary and changing the cache to be cacheline-aligned as well. With kernel compiled using defconfig, this causes VMA memory consumption to grow from 160 (vm_area_struct) + 40 (vm_lock) bytes to 256 bytes:

slabinfo before:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vma_lock ... 40 102 1 : ...
vm_area_struct ... 160 51 2 : ...

slabinfo after moving vm_lock:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vm_area_struct ... 256 32 2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 50 to 64 pages, which is 5.5MB per 100000 VMAs. Note that the size of this structure is dependent on the kernel configuration and typically the original size is higher than 160 bytes. Therefore these calculations are close to the worst case scenario. A more realistic vm_area_struct usage before this change is:

     <name>           ... <objsize> <objperslab> <pagesperslab> : ...
     vma_lock         ...     40  102    1 : ...
     vm_area_struct   ...    176   46    2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 54 to 64 pages, which is 3.9MB per 100000 VMAs.
This memory consumption growth can be addressed later by optimizing the vm_lock.

Quoted 8 lines
Quoted 6 lines

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/ZsQyI%2F087V34JoIt@xsang-OptiPlex-9020/
[3] https://lore.kernel.org/all/CAJuCfpEisU8Lfe96AYJDZ+OM4NoPmnw9bP53cT_kbfP_pR+-2g@mail.gmail.com/

Thanks!

Quoted 14 lines

One question below.

Quoted 10 lines
include/linux/mm_types.h +0 -2
@@ -716,8 +716,6 @@ struct vm_area_struct {
       * slowpath.
       */
      unsigned int vm_lock_seq;
-     /* Unstable RCU readers are allowed to read this. */
-     struct vma_lock *vm_lock;
 #endif
 
Quoted 13 lines

include/linux/mm_types.h struct vm_area_struct { · line 775

Quoted 10 lines
include/linux/mm_types.h +4 -0
      /*
@@ -770,6 +768,10 @@ struct vm_area_struct {
      struct vma_numab_state *numab_state;    /* NUMA Balancing state */
 #endif
      struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
+#ifdef CONFIG_PER_VMA_LOCK
+     /* Unstable RCU readers are allowed to read this. */
+     struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+#endif
 } __randomize_layout;

Do we just want 'struct vm_area_struct' to be cacheline aligned or do we want 'struct vma_lock vm_lock' to be on a separate cacheline as well?

We want both to minimize cacheline sharing.

Quoted 1 line
reviewer
#6 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

Quoted 84 lines · Suren Baghdasaryan
Quoted 44 lines · Shakeel Butt
Quoted 42 lines · Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].
Splitting single logical structure into multiple ones leads to more complicated management, extra pointer dereferences and overall less maintainable code. When that split-away part is a lock, it complicates things even further. With no performance benefits, there are no reasons for this split. Merging the vm_lock back into vm_area_struct also allows vm_area_struct to use SLAB_TYPESAFE_BY_RCU later in this patchset. Move vm_lock back into vm_area_struct, aligning it at the cacheline boundary and changing the cache to be cacheline-aligned as well. With kernel compiled using defconfig, this causes VMA memory consumption to grow from 160 (vm_area_struct) + 40 (vm_lock) bytes to 256 bytes:

slabinfo before:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vma_lock ... 40 102 1 : ...
vm_area_struct ... 160 51 2 : ...

slabinfo after moving vm_lock:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vm_area_struct ... 256 32 2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 50 to 64 pages, which is 5.5MB per 100000 VMAs. Note that the size of this structure is dependent on the kernel configuration and typically the original size is higher than 160 bytes. Therefore these calculations are close to the worst case scenario. A more realistic vm_area_struct usage before this change is:

     <name>           ... <objsize> <objperslab> <pagesperslab> : ...
     vma_lock         ...     40  102    1 : ...
     vm_area_struct   ...    176   46    2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 54 to 64 pages, which is 3.9MB per 100000 VMAs.
This memory consumption growth can be addressed later by optimizing the vm_lock.

Quoted 8 lines
Quoted 6 lines

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/ZsQyI%2F087V34JoIt@xsang-OptiPlex-9020/
[3] https://lore.kernel.org/all/CAJuCfpEisU8Lfe96AYJDZ+OM4NoPmnw9bP53cT_kbfP_pR+-2g@mail.gmail.com/

Thanks!

Quoted 14 lines

One question below.

Quoted 10 lines
include/linux/mm_types.h +0 -2
@@ -716,8 +716,6 @@ struct vm_area_struct {
       * slowpath.
       */
      unsigned int vm_lock_seq;
-     /* Unstable RCU readers are allowed to read this. */
-     struct vma_lock *vm_lock;
 #endif
 
Quoted 13 lines

include/linux/mm_types.h struct vm_area_struct { · line 775

Quoted 10 lines
include/linux/mm_types.h +4 -0
      /*
@@ -770,6 +768,10 @@ struct vm_area_struct {
      struct vma_numab_state *numab_state;    /* NUMA Balancing state */
 #endif
      struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
+#ifdef CONFIG_PER_VMA_LOCK
+     /* Unstable RCU readers are allowed to read this. */
+     struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+#endif
 } __randomize_layout;

Do we just want 'struct vm_area_struct' to be cacheline aligned or do we want 'struct vma_lock vm_lock' to be on a separate cacheline as well?

Quoted 2 lines

We want both to minimize cacheline sharing.

For later, you will need to add a pad after vm_lock as well, so any future addition will not share the cacheline with vm_lock. I would do something like below. This is a nit and can be done later.

include/linux/mm_types.h +4 -2
index 7654c766cbe2..5cc4fff163a0 100644
@@ -751,10 +751,12 @@ struct vm_area_struct {
 #endif
 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 #ifdef CONFIG_PER_VMA_LOCK
+	CACHELINE_PADDING(__pad1__);
 	/* Unstable RCU readers are allowed to read this. */
-	struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+	struct vma_lock vm_lock;
+	CACHELINE_PADDING(__pad2__);
 #endif
-} __randomize_layout;
+} __randomize_layout ____cacheline_aligned_in_smp;
 
 #ifdef CONFIG_NUMA
 #define vma_policy(vma) ((vma)->vm_policy)
author maintainer
#7 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

Quoted 99 lines · Shakeel Butt
Quoted 84 lines · Suren Baghdasaryan
Quoted 44 lines · Shakeel Butt
Quoted 42 lines · Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].
Splitting single logical structure into multiple ones leads to more complicated management, extra pointer dereferences and overall less maintainable code. When that split-away part is a lock, it complicates things even further. With no performance benefits, there are no reasons for this split. Merging the vm_lock back into vm_area_struct also allows vm_area_struct to use SLAB_TYPESAFE_BY_RCU later in this patchset. Move vm_lock back into vm_area_struct, aligning it at the cacheline boundary and changing the cache to be cacheline-aligned as well. With kernel compiled using defconfig, this causes VMA memory consumption to grow from 160 (vm_area_struct) + 40 (vm_lock) bytes to 256 bytes:

slabinfo before:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vma_lock ... 40 102 1 : ...
vm_area_struct ... 160 51 2 : ...

slabinfo after moving vm_lock:
<name> ... <objsize> <objperslab> <pagesperslab> : ...
vm_area_struct ... 256 32 2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 50 to 64 pages, which is 5.5MB per 100000 VMAs. Note that the size of this structure is dependent on the kernel configuration and typically the original size is higher than 160 bytes. Therefore these calculations are close to the worst case scenario. A more realistic vm_area_struct usage before this change is:

     <name>           ... <objsize> <objperslab> <pagesperslab> : ...
     vma_lock         ...     40  102    1 : ...
     vm_area_struct   ...    176   46    2 : ...

Aggregate VMA memory consumption per 1000 VMAs grows from 54 to 64 pages, which is 3.9MB per 100000 VMAs.
This memory consumption growth can be addressed later by optimizing the vm_lock.

Quoted 8 lines
Quoted 6 lines

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/ZsQyI%2F087V34JoIt@xsang-OptiPlex-9020/
[3] https://lore.kernel.org/all/CAJuCfpEisU8Lfe96AYJDZ+OM4NoPmnw9bP53cT_kbfP_pR+-2g@mail.gmail.com/

Thanks!

Quoted 14 lines

One question below.

Quoted 10 lines
include/linux/mm_types.h +0 -2
@@ -716,8 +716,6 @@ struct vm_area_struct {
       * slowpath.
       */
      unsigned int vm_lock_seq;
-     /* Unstable RCU readers are allowed to read this. */
-     struct vma_lock *vm_lock;
 #endif
 
Quoted 13 lines

include/linux/mm_types.h struct vm_area_struct { · line 775

Quoted 10 lines
include/linux/mm_types.h +4 -0
      /*
@@ -770,6 +768,10 @@ struct vm_area_struct {
      struct vma_numab_state *numab_state;    /* NUMA Balancing state */
 #endif
      struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
+#ifdef CONFIG_PER_VMA_LOCK
+     /* Unstable RCU readers are allowed to read this. */
+     struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+#endif
 } __randomize_layout;

Do we just want 'struct vm_area_struct' to be cacheline aligned or do we want 'struct vma_lock vm_lock' to be on a separate cacheline as well?

Quoted 2 lines

We want both to minimize cacheline sharing.

For later, you will need to add a pad after vm_lock as well, so any future addition will not share the cacheline with vm_lock. I would do something like below. This is a nit and can be done later.

include/linux/mm_types.h +0 -0
index 7654c766cbe2..5cc4fff163a0 100644
@@ -751,10 +751,12 @@ struct vm_area_struct {
 #endif

include/linux/mm_types.h struct vm_area_struct { · line 759

Quoted 10 lines
include/linux/mm_types.h +4 -2
        struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 #ifdef CONFIG_PER_VMA_LOCK
+       CACHELINE_PADDING(__pad1__);
        /* Unstable RCU readers are allowed to read this. */
-       struct vma_lock vm_lock ____cacheline_aligned_in_smp;
+       struct vma_lock vm_lock;
+       CACHELINE_PADDING(__pad2__);
 #endif
-} __randomize_layout;
+} __randomize_layout ____cacheline_aligned_in_smp;

I thought SLAB_HWCACHE_ALIGN for vm_area_cachep added in this patch would have the same result, no?

Quoted 3 lines
include/linux/mm_types.h +0 -0
 #ifdef CONFIG_NUMA
 #define vma_policy(vma) ((vma)->vm_policy)
reviewer
#8 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

On Wed, Nov 20, 2024 at 04:33:37PM -0800, Suren Baghdasaryan wrote:
[...]

Quoted 22 lines
Quoted 22 lines
Quoted 6 lines
Quoted 3 lines

Do we just want 'struct vm_area_struct' to be cacheline aligned or do we want 'struct vma_lock vm_lock' to be on a separate cacheline as well?

We want both to minimize cacheline sharing.

For later, you will need to add a pad after vm_lock as well, so any future addition will not share the cacheline with vm_lock. I would do something like below. This is a nit and can be done later.

include/linux/mm_types.h +1 -1
index 7654c766cbe2..5cc4fff163a0 100644
@@ -751,10 +751,12 @@ struct vm_area_struct {
 #endif
        struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 #ifdef CONFIG_PER_VMA_LOCK
+       CACHELINE_PADDING(__pad1__);
        /* Unstable RCU readers are allowed to read this. */
-       struct vma_lock vm_lock ____cacheline_aligned_in_smp;
Quoted 9 lines

include/linux/mm_types.h struct vm_area_struct { · line 759

Quoted 5 lines
include/linux/mm_types.h +3 -1
+       struct vma_lock vm_lock;
+       CACHELINE_PADDING(__pad2__);
 #endif
-} __randomize_layout;
+} __randomize_layout ____cacheline_aligned_in_smp;

I thought SLAB_HWCACHE_ALIGN for vm_area_cachep added in this patch would have the same result, no?

SLAB_HWCACHE_ALIGN is more about the slub allocator allocating cache aligned memory. It does not say anything about the internals of the struct for which the kmem_cache is being created. The ____cacheline_aligned_in_smp tag in your patch made sure that the field vm_lock will be put in a new cacheline and there can be a hole between vm_lock and the previous field if the previous field is not ending at the cacheline boundary. Please note that if you add a new field after vm_lock (without cacheline alignment tag), it will be on the same cacheline as vm_lock. So, your code is achieving the vm_lock on its own cacheline goal but vm_lock being the only field on that cacheline is not being achieved.

author maintainer
#9 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

Quoted 35 lines · Shakeel Butt

On Wed, Nov 20, 2024 at 04:33:37PM -0800, Suren Baghdasaryan wrote:
[...]

Quoted 22 lines
Quoted 22 lines
Quoted 6 lines
Quoted 3 lines

Do we just want 'struct vm_area_struct' to be cacheline aligned or do we want 'struct vma_lock vm_lock' to be on a separate cacheline as well?

We want both to minimize cacheline sharing.

For later, you will need to add a pad after vm_lock as well, so any future addition will not share the cacheline with vm_lock. I would do something like below. This is a nit and can be done later.

include/linux/mm_types.h +1 -1
index 7654c766cbe2..5cc4fff163a0 100644
@@ -751,10 +751,12 @@ struct vm_area_struct {
 #endif
        struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 #ifdef CONFIG_PER_VMA_LOCK
+       CACHELINE_PADDING(__pad1__);
        /* Unstable RCU readers are allowed to read this. */
-       struct vma_lock vm_lock ____cacheline_aligned_in_smp;
Quoted 9 lines

include/linux/mm_types.h struct vm_area_struct { · line 759

Quoted 5 lines
include/linux/mm_types.h +3 -1
+       struct vma_lock vm_lock;
+       CACHELINE_PADDING(__pad2__);
 #endif
-} __randomize_layout;
+} __randomize_layout ____cacheline_aligned_in_smp;

I thought SLAB_HWCACHE_ALIGN for vm_area_cachep added in this patch would have the same result, no?

Quoted 11 lines

SLAB_HWCACHE_ALIGN is more about the slub allocator allocating cache aligned memory. It does not say anything about the internals of the struct for which the kmem_cache is being created. The ____cacheline_aligned_in_smp tag in your patch made sure that the field vm_lock will be put in a new cacheline and there can be a hole between vm_lock and the previous field if the previous field is not ending at the cacheline boundary. Please note that if you add a new field after vm_lock (without cacheline alignment tag), it will be on the same cacheline as vm_lock. So, your code is achieving the vm_lock on its own cacheline goal but vm_lock being the only field on that cacheline is not being achieved.

Sorry, I should have been more clear. It's ok if some fields which are rarely accessed in the pagefault path are placed in the same cacheling with vm_lock. In fact I've done that to pack them better in the previous version of this patchset here:
https://lore.kernel.org/all/[email protected]/
(removed for now based on the feedback). So, vm_lock being the only field on the cacheline is not my goal. After this patchset I'm planning to try packing the members better and save some memory.

Quoted 1 line
reviewer
#10 lore raw

Re: [PATCH v4 2/5] mm: move per-vma lock into vm_area_struct

On Thu, Nov 21, 2024 at 09:05:21AM -0800, Suren Baghdasaryan wrote:
[...]

Quoted 10 lines

Sorry, I should have been more clear. It's ok if some fields which are rarely accessed in the pagefault path are placed in the same cacheling with vm_lock. In fact I've done that to pack them better in the previous version of this patchset here:
https://lore.kernel.org/all/[email protected]/
(removed for now based on the feedback). So, vm_lock being the only field on the cacheline is not my goal. After this patchset I'm planning to try packing the members better and save some memory.

Nah, my bad, somehow I thought you want vm_lock to be on a cacheline alone.

author maintainer
#11 lore raw

PATCH v4 3/5 mm: mark vma as detached until it's added into vma tree

In reply to Suren Baghdasaryan

Current implementation does not set detached flag when a VMA is first allocated. This does not represent the real state of the VMA, which is detached until it is added into mm's VMA tree. Fix this by marking new VMAs as detached and resetting detached flag only after VMA is added into a tree.
Introduce vma_mark_attached() to make the API more readable and to simplify possible future cleanup when vma->vm_mm might be used to indicate detached vma and vma_mark_attached() will need an additional mm parameter.

---
 include/linux/mm.h               | 27 ++++++++++++++++++++-------
 kernel/fork.c                    |  4 ++++
 mm/memory.c                      |  2 +-
 mm/vma.c                         |  6 +++---
 mm/vma.h                         |  2 ++
 tools/testing/vma/vma_internal.h | 17 ++++++++++++-----
 6 files changed, 42 insertions(+), 16 deletions(-)
include/linux/mm.h +20 -7
index 737c003b0a1e..dd1b6190df28 100644
@@ -808,12 +808,21 @@ static inline void vma_assert_locked(struct vm_area_struct *vma)
 		vma_assert_write_locked(vma);
 }
 
-static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
+static inline void vma_mark_attached(struct vm_area_struct *vma)
+{
+	vma->detached = false;
+}
+
+static inline void vma_mark_detached(struct vm_area_struct *vma)
 {
 	/* When detaching vma should be write-locked */
-	if (detached)
-		vma_assert_write_locked(vma);
-	vma->detached = detached;
+	vma_assert_write_locked(vma);
+	vma->detached = true;
+}
+
+static inline bool is_vma_detached(struct vm_area_struct *vma)
+{
+	return vma->detached;
 }
 
 static inline void release_fault_lock(struct vm_fault *vmf)
@@ -844,8 +853,8 @@ static inline void vma_end_read(struct vm_area_struct *vma) {}
 static inline void vma_start_write(struct vm_area_struct *vma) {}
 static inline void vma_assert_write_locked(struct vm_area_struct *vma)
 		{ mmap_assert_write_locked(vma->vm_mm); }
-static inline void vma_mark_detached(struct vm_area_struct *vma,
-				     bool detached) {}
+static inline void vma_mark_attached(struct vm_area_struct *vma) {}
+static inline void vma_mark_detached(struct vm_area_struct *vma) {}
 
 static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 		unsigned long address)
@@ -878,7 +887,10 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	vma->vm_mm = mm;
 	vma->vm_ops = &vma_dummy_vm_ops;
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
-	vma_mark_detached(vma, false);
+#ifdef CONFIG_PER_VMA_LOCK
+	/* vma is not locked, can't use vma_mark_detached() */
+	vma->detached = true;
+#endif
 	vma_numab_state_init(vma);
 	vma_lock_init(vma);
 }
@@ -1073,6 +1085,7 @@ static inline int vma_iter_bulk_store(struct vma_iterator *vmi,
 	if (unlikely(mas_is_err(&vmi->mas)))
 		return -ENOMEM;
 
+	vma_mark_attached(vma);
 	return 0;
 }
 
kernel/fork.c +4 -0
index 7823797e31d2..f0cec673583c 100644
@@ -465,6 +465,10 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 	data_race(memcpy(new, orig, sizeof(*new)));
 	vma_lock_init(new);
 	INIT_LIST_HEAD(&new->anon_vma_chain);
+#ifdef CONFIG_PER_VMA_LOCK
+	/* vma is not locked, can't use vma_mark_detached() */
+	new->detached = true;
+#endif
 	vma_numab_state_init(new);
 	dup_anon_vma_name(orig, new);
 
mm/memory.c +1 -1
index 209885a4134f..d0197a0c0996 100644
@@ -6279,7 +6279,7 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 		goto inval;
 
 	/* Check if the VMA got isolated after we found it */
-	if (vma->detached) {
+	if (is_vma_detached(vma)) {
 		vma_end_read(vma);
 		count_vm_vma_lock_event(VMA_LOCK_MISS);
 		/* The area was replaced with another one */
mm/vma.c +3 -3
index 8a454a7bbc80..73104d434567 100644
@@ -295,7 +295,7 @@ static void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi,
 
 	if (vp->remove) {
 again:
-		vma_mark_detached(vp->remove, true);
+		vma_mark_detached(vp->remove);
 		if (vp->file) {
 			uprobe_munmap(vp->remove, vp->remove->vm_start,
 				      vp->remove->vm_end);
@@ -1220,7 +1220,7 @@ static void reattach_vmas(struct ma_state *mas_detach)
 
 	mas_set(mas_detach, 0);
 	mas_for_each(mas_detach, vma, ULONG_MAX)
-		vma_mark_detached(vma, false);
+		vma_mark_attached(vma);
 
 	__mt_destroy(mas_detach->tree);
 }
@@ -1295,7 +1295,7 @@ static int vms_gather_munmap_vmas(struct vma_munmap_struct *vms,
 		if (error)
 			goto munmap_gather_failed;
 
-		vma_mark_detached(next, true);
+		vma_mark_detached(next);
 		nrpages = vma_pages(next);
 
 		vms->nr_pages += nrpages;
mm/vma.h +2 -0
index 388d34748674..2e680f357ace 100644
@@ -162,6 +162,7 @@ static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
 	if (unlikely(mas_is_err(&vmi->mas)))
 		return -ENOMEM;
 
+	vma_mark_attached(vma);
 	return 0;
 }
 
@@ -385,6 +386,7 @@ static inline void vma_iter_store(struct vma_iterator *vmi,
 
 	__mas_set_range(&vmi->mas, vma->vm_start, vma->vm_end - 1);
 	mas_store_prealloc(&vmi->mas, vma);
+	vma_mark_attached(vma);
 }
 
 static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
tools/testing/vma/vma_internal.h +12 -5
index 11c2c38ca4e8..2fed366d20ef 100644
@@ -414,13 +414,17 @@ static inline void vma_lock_init(struct vm_area_struct *vma)
 	vma->vm_lock_seq = UINT_MAX;
 }
 
+static inline void vma_mark_attached(struct vm_area_struct *vma)
+{
+	vma->detached = false;
+}
+
 static inline void vma_assert_write_locked(struct vm_area_struct *);
-static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
+static inline void vma_mark_detached(struct vm_area_struct *vma)
 {
 	/* When detaching vma should be write-locked */
-	if (detached)
-		vma_assert_write_locked(vma);
-	vma->detached = detached;
+	vma_assert_write_locked(vma);
+	vma->detached = true;
 }
 
 extern const struct vm_operations_struct vma_dummy_vm_ops;
@@ -431,7 +435,8 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	vma->vm_mm = mm;
 	vma->vm_ops = &vma_dummy_vm_ops;
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
-	vma_mark_detached(vma, false);
+	/* vma is not locked, can't use vma_mark_detached() */
+	vma->detached = true;
 	vma_lock_init(vma);
 }
 
@@ -457,6 +462,8 @@ static inline struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 	memcpy(new, orig, sizeof(*new));
 	vma_lock_init(new);
 	INIT_LIST_HEAD(&new->anon_vma_chain);
+	/* vma is not locked, can't use vma_mark_detached() */
+	new->detached = true;
 
 	return new;
 }
Signature · 1 line
2.47.0.338.g60cca15819-goog
reviewer Reviewed-by on v4 3/5
#12 lore raw

Re: [PATCH v4 3/5] mm: mark vma as detached until it's added into vma tree

Quoted 11 lines · Suren Baghdasaryan

Current implementation does not set detached flag when a VMA is first allocated. This does not represent the real state of the VMA, which is detached until it is added into mm's VMA tree. Fix this by marking new VMAs as detached and resetting detached flag only after VMA is added into a tree.
Introduce vma_mark_attached() to make the API more readable and to simplify possible future cleanup when vma->vm_mm might be used to indicate detached vma and vma_mark_attached() will need an additional mm parameter.

Reviewed-by on v4 3/5
#13 lore raw

Re: [PATCH v4 3/5] mm: mark vma as detached until it's added into vma tree

In reply to Suren Baghdasaryan
Quoted 11 lines · Suren Baghdasaryan

Current implementation does not set detached flag when a VMA is first allocated. This does not represent the real state of the VMA, which is detached until it is added into mm's VMA tree. Fix this by marking new VMAs as detached and resetting detached flag only after VMA is added into a tree.
Introduce vma_mark_attached() to make the API more readable and to simplify possible future cleanup when vma->vm_mm might be used to indicate detached vma and vma_mark_attached() will need an additional mm parameter.

I tested this (whole series) locally and on real hardware and did a kernel compile on real hardware just to be sure :)) and all looks good.

The code looks sensible, so:

Quoted 154 lines
---
 include/linux/mm.h               | 27 ++++++++++++++++++++-------
 kernel/fork.c                    |  4 ++++
 mm/memory.c                      |  2 +-
 mm/vma.c                         |  6 +++---
 mm/vma.h                         |  2 ++
 tools/testing/vma/vma_internal.h | 17 ++++++++++++-----
 6 files changed, 42 insertions(+), 16 deletions(-)
include/linux/mm.h +20 -7
index 737c003b0a1e..dd1b6190df28 100644
@@ -808,12 +808,21 @@ static inline void vma_assert_locked(struct vm_area_struct *vma)
 		vma_assert_write_locked(vma);
 }
 
-static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
+static inline void vma_mark_attached(struct vm_area_struct *vma)
+{
+	vma->detached = false;
+}
+
+static inline void vma_mark_detached(struct vm_area_struct *vma)
 {
 	/* When detaching vma should be write-locked */
-	if (detached)
-		vma_assert_write_locked(vma);
-	vma->detached = detached;
+	vma_assert_write_locked(vma);
+	vma->detached = true;
+}
+
+static inline bool is_vma_detached(struct vm_area_struct *vma)
+{
+	return vma->detached;
 }
 
 static inline void release_fault_lock(struct vm_fault *vmf)
@@ -844,8 +853,8 @@ static inline void vma_end_read(struct vm_area_struct *vma) {}
 static inline void vma_start_write(struct vm_area_struct *vma) {}
 static inline void vma_assert_write_locked(struct vm_area_struct *vma)
 		{ mmap_assert_write_locked(vma->vm_mm); }
-static inline void vma_mark_detached(struct vm_area_struct *vma,
-				     bool detached) {}
+static inline void vma_mark_attached(struct vm_area_struct *vma) {}
+static inline void vma_mark_detached(struct vm_area_struct *vma) {}
 
 static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 		unsigned long address)
@@ -878,7 +887,10 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	vma->vm_mm = mm;
 	vma->vm_ops = &vma_dummy_vm_ops;
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
-	vma_mark_detached(vma, false);
+#ifdef CONFIG_PER_VMA_LOCK
+	/* vma is not locked, can't use vma_mark_detached() */
+	vma->detached = true;
+#endif
 	vma_numab_state_init(vma);
 	vma_lock_init(vma);
 }
@@ -1073,6 +1085,7 @@ static inline int vma_iter_bulk_store(struct vma_iterator *vmi,
 	if (unlikely(mas_is_err(&vmi->mas)))
 		return -ENOMEM;
 
+	vma_mark_attached(vma);
 	return 0;
 }
 
kernel/fork.c +4 -0
index 7823797e31d2..f0cec673583c 100644
@@ -465,6 +465,10 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 	data_race(memcpy(new, orig, sizeof(*new)));
 	vma_lock_init(new);
 	INIT_LIST_HEAD(&new->anon_vma_chain);
+#ifdef CONFIG_PER_VMA_LOCK
+	/* vma is not locked, can't use vma_mark_detached() */
+	new->detached = true;
+#endif
 	vma_numab_state_init(new);
 	dup_anon_vma_name(orig, new);
 
mm/memory.c +1 -1
index 209885a4134f..d0197a0c0996 100644
@@ -6279,7 +6279,7 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 		goto inval;
 
 	/* Check if the VMA got isolated after we found it */
-	if (vma->detached) {
+	if (is_vma_detached(vma)) {
 		vma_end_read(vma);
 		count_vm_vma_lock_event(VMA_LOCK_MISS);
 		/* The area was replaced with another one */
mm/vma.c +3 -3
index 8a454a7bbc80..73104d434567 100644
@@ -295,7 +295,7 @@ static void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi,
 
 	if (vp->remove) {
 again:
-		vma_mark_detached(vp->remove, true);
+		vma_mark_detached(vp->remove);
 		if (vp->file) {
 			uprobe_munmap(vp->remove, vp->remove->vm_start,
 				      vp->remove->vm_end);
@@ -1220,7 +1220,7 @@ static void reattach_vmas(struct ma_state *mas_detach)
 
 	mas_set(mas_detach, 0);
 	mas_for_each(mas_detach, vma, ULONG_MAX)
-		vma_mark_detached(vma, false);
+		vma_mark_attached(vma);
 
 	__mt_destroy(mas_detach->tree);
 }
@@ -1295,7 +1295,7 @@ static int vms_gather_munmap_vmas(struct vma_munmap_struct *vms,
 		if (error)
 			goto munmap_gather_failed;
 
-		vma_mark_detached(next, true);
+		vma_mark_detached(next);
 		nrpages = vma_pages(next);
 
 		vms->nr_pages += nrpages;
mm/vma.h +2 -0
index 388d34748674..2e680f357ace 100644
@@ -162,6 +162,7 @@ static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
 	if (unlikely(mas_is_err(&vmi->mas)))
 		return -ENOMEM;
 
+	vma_mark_attached(vma);
 	return 0;
 }
 
@@ -385,6 +386,7 @@ static inline void vma_iter_store(struct vma_iterator *vmi,
 
 	__mas_set_range(&vmi->mas, vma->vm_start, vma->vm_end - 1);
 	mas_store_prealloc(&vmi->mas, vma);
+	vma_mark_attached(vma);
 }
 
 static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
tools/testing/vma/vma_internal.h +0 -0
index 11c2c38ca4e8..2fed366d20ef 100644

tools/testing/vma/vma_internal.h static inline void vma_lock_init(struct vm_area_struct *vma) · line 420

Quoted 7 lines
tools/testing/vma/vma_internal.h +4 -0
@@ -414,13 +414,17 @@ static inline void vma_lock_init(struct vm_area_struct *vma)
 	vma->vm_lock_seq = UINT_MAX;
 }
 
+static inline void vma_mark_attached(struct vm_area_struct *vma)
+{
+	vma->detached = false;
+}

Yeah I think sensible to just accept that sometimes we are already attached when we mark attached.

Quoted 36 lines
tools/testing/vma/vma_internal.h +8 -6
+
 static inline void vma_assert_write_locked(struct vm_area_struct *);
-static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
+static inline void vma_mark_detached(struct vm_area_struct *vma)
 {
 	/* When detaching vma should be write-locked */
-	if (detached)
-		vma_assert_write_locked(vma);
-	vma->detached = detached;
+	vma_assert_write_locked(vma);
+	vma->detached = true;
 }
 
 extern const struct vm_operations_struct vma_dummy_vm_ops;
@@ -431,7 +435,8 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	vma->vm_mm = mm;
 	vma->vm_ops = &vma_dummy_vm_ops;
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
-	vma_mark_detached(vma, false);
+	/* vma is not locked, can't use vma_mark_detached() */
+	vma->detached = true;
 	vma_lock_init(vma);
 }
 
@@ -457,6 +462,8 @@ static inline struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 	memcpy(new, orig, sizeof(*new));
 	vma_lock_init(new);
 	INIT_LIST_HEAD(&new->anon_vma_chain);
+	/* vma is not locked, can't use vma_mark_detached() */
+	new->detached = true;
 
 	return new;
 }
--

2.47.0.338.g60cca15819-goog

author maintainer
#14 lore raw

Re: [PATCH v4 3/5] mm: mark vma as detached until it's added into vma tree

Quoted 12 lines · Lorenzo Stoakes
Quoted 10 lines · Suren Baghdasaryan

Current implementation does not set detached flag when a VMA is first allocated. This does not represent the real state of the VMA, which is detached until it is added into mm's VMA tree. Fix this by marking new VMAs as detached and resetting detached flag only after VMA is added into a tree.
Introduce vma_mark_attached() to make the API more readable and to simplify possible future cleanup when vma->vm_mm might be used to indicate detached vma and vma_mark_attached() will need an additional mm parameter.

Quoted 8 lines
Quoted 1 line

I tested this (whole series) locally and on real hardware and did a kernel compile on real hardware just to be sure :)) and all looks good.

The code looks sensible, so:

Thanks Lorenzo! I'll post a new version today, addressing Matthew's and other's comments but overall functionality should stay the same.

Quoted 202 lines
Quoted 155 lines
---
 include/linux/mm.h               | 27 ++++++++++++++++++++-------
 kernel/fork.c                    |  4 ++++
 mm/memory.c                      |  2 +-
 mm/vma.c                         |  6 +++---
 mm/vma.h                         |  2 ++
 tools/testing/vma/vma_internal.h | 17 ++++++++++++-----
 6 files changed, 42 insertions(+), 16 deletions(-)
include/linux/mm.h +20 -7
index 737c003b0a1e..dd1b6190df28 100644
@@ -808,12 +808,21 @@ static inline void vma_assert_locked(struct vm_area_struct *vma)
              vma_assert_write_locked(vma);
 }
 
-static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
+static inline void vma_mark_attached(struct vm_area_struct *vma)
+{
+     vma->detached = false;
+}
+
+static inline void vma_mark_detached(struct vm_area_struct *vma)
 {
      /* When detaching vma should be write-locked */
-     if (detached)
-             vma_assert_write_locked(vma);
-     vma->detached = detached;
+     vma_assert_write_locked(vma);
+     vma->detached = true;
+}
+
+static inline bool is_vma_detached(struct vm_area_struct *vma)
+{
+     return vma->detached;
 }
 
 static inline void release_fault_lock(struct vm_fault *vmf)
@@ -844,8 +853,8 @@ static inline void vma_end_read(struct vm_area_struct *vma) {}
 static inline void vma_start_write(struct vm_area_struct *vma) {}
 static inline void vma_assert_write_locked(struct vm_area_struct *vma)
              { mmap_assert_write_locked(vma->vm_mm); }
-static inline void vma_mark_detached(struct vm_area_struct *vma,
-                                  bool detached) {}
+static inline void vma_mark_attached(struct vm_area_struct *vma) {}
+static inline void vma_mark_detached(struct vm_area_struct *vma) {}
 
 static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
              unsigned long address)
@@ -878,7 +887,10 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
      vma->vm_mm = mm;
      vma->vm_ops = &vma_dummy_vm_ops;
      INIT_LIST_HEAD(&vma->anon_vma_chain);
-     vma_mark_detached(vma, false);
+#ifdef CONFIG_PER_VMA_LOCK
+     /* vma is not locked, can't use vma_mark_detached() */
+     vma->detached = true;
+#endif
      vma_numab_state_init(vma);
      vma_lock_init(vma);
 }
@@ -1073,6 +1085,7 @@ static inline int vma_iter_bulk_store(struct vma_iterator *vmi,
      if (unlikely(mas_is_err(&vmi->mas)))
              return -ENOMEM;
 
+     vma_mark_attached(vma);
      return 0;
 }
 
kernel/fork.c +4 -0
index 7823797e31d2..f0cec673583c 100644
@@ -465,6 +465,10 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
      data_race(memcpy(new, orig, sizeof(*new)));
      vma_lock_init(new);
      INIT_LIST_HEAD(&new->anon_vma_chain);
+#ifdef CONFIG_PER_VMA_LOCK
+     /* vma is not locked, can't use vma_mark_detached() */
+     new->detached = true;
+#endif
      vma_numab_state_init(new);
      dup_anon_vma_name(orig, new);
 
mm/memory.c +1 -1
index 209885a4134f..d0197a0c0996 100644
@@ -6279,7 +6279,7 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
              goto inval;
 
      /* Check if the VMA got isolated after we found it */
-     if (vma->detached) {
+     if (is_vma_detached(vma)) {
              vma_end_read(vma);
              count_vm_vma_lock_event(VMA_LOCK_MISS);
              /* The area was replaced with another one */
mm/vma.c +3 -3
index 8a454a7bbc80..73104d434567 100644
@@ -295,7 +295,7 @@ static void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi,
 
      if (vp->remove) {
 again:
-             vma_mark_detached(vp->remove, true);
+             vma_mark_detached(vp->remove);
              if (vp->file) {
                      uprobe_munmap(vp->remove, vp->remove->vm_start,
                                    vp->remove->vm_end);
@@ -1220,7 +1220,7 @@ static void reattach_vmas(struct ma_state *mas_detach)
 
      mas_set(mas_detach, 0);
      mas_for_each(mas_detach, vma, ULONG_MAX)
-             vma_mark_detached(vma, false);
+             vma_mark_attached(vma);
 
      __mt_destroy(mas_detach->tree);
 }
@@ -1295,7 +1295,7 @@ static int vms_gather_munmap_vmas(struct vma_munmap_struct *vms,
              if (error)
                      goto munmap_gather_failed;
 
-             vma_mark_detached(next, true);
+             vma_mark_detached(next);
              nrpages = vma_pages(next);
 
              vms->nr_pages += nrpages;
mm/vma.h +2 -0
index 388d34748674..2e680f357ace 100644
@@ -162,6 +162,7 @@ static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
      if (unlikely(mas_is_err(&vmi->mas)))
              return -ENOMEM;
 
+     vma_mark_attached(vma);
      return 0;
 }
 
@@ -385,6 +386,7 @@ static inline void vma_iter_store(struct vma_iterator *vmi,
 
      __mas_set_range(&vmi->mas, vma->vm_start, vma->vm_end - 1);
      mas_store_prealloc(&vmi->mas, vma);
+     vma_mark_attached(vma);
 }
 
 static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
tools/testing/vma/vma_internal.h +0 -0
index 11c2c38ca4e8..2fed366d20ef 100644

tools/testing/vma/vma_internal.h static inline void vma_lock_init(struct vm_area_struct *vma) · line 420

Quoted 7 lines
tools/testing/vma/vma_internal.h +4 -0
@@ -414,13 +414,17 @@ static inline void vma_lock_init(struct vm_area_struct *vma)
      vma->vm_lock_seq = UINT_MAX;
 }
 
+static inline void vma_mark_attached(struct vm_area_struct *vma)
+{
+     vma->detached = false;
+}

Yeah I think sensible to just accept that sometimes we are already attached when we mark attached.

Quoted 36 lines
tools/testing/vma/vma_internal.h +8 -6
+
 static inline void vma_assert_write_locked(struct vm_area_struct *);
-static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
+static inline void vma_mark_detached(struct vm_area_struct *vma)
 {
      /* When detaching vma should be write-locked */
-     if (detached)
-             vma_assert_write_locked(vma);
-     vma->detached = detached;
+     vma_assert_write_locked(vma);
+     vma->detached = true;
 }
 
 extern const struct vm_operations_struct vma_dummy_vm_ops;
@@ -431,7 +435,8 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
      vma->vm_mm = mm;
      vma->vm_ops = &vma_dummy_vm_ops;
      INIT_LIST_HEAD(&vma->anon_vma_chain);
-     vma_mark_detached(vma, false);
+     /* vma is not locked, can't use vma_mark_detached() */
+     vma->detached = true;
      vma_lock_init(vma);
 }
 
@@ -457,6 +462,8 @@ static inline struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
      memcpy(new, orig, sizeof(*new));
      vma_lock_init(new);
      INIT_LIST_HEAD(&new->anon_vma_chain);
+     /* vma is not locked, can't use vma_mark_detached() */
+     new->detached = true;
 
      return new;
 }
--

2.47.0.338.g60cca15819-goog

author maintainer
#15 lore raw

PATCH v4 4/5 mm: make vma cache SLAB_TYPESAFE_BY_RCU

In reply to Suren Baghdasaryan

To enable SLAB_TYPESAFE_BY_RCU for vma cache we need to ensure that object reuse before RCU grace period is over will be detected inside lock_vma_under_rcu().
lock_vma_under_rcu() enters RCU read section, finds the vma at the given address, locks the vma and checks if it got detached or remapped to cover a different address range. These last checks are there to ensure that the vma was not modified after we found it but before locking it.
vma reuse introduces several new possibilities:
1. vma can be reused after it was found but before it is locked;
2. vma can be reused and reinitialized (including changing its vm_mm) while being locked in vma_start_read();
3. vma can be reused and reinitialized after it was found but before it is locked, then attached at a new address or to a new mm while being read-locked;
For case #1 current checks will help detecting cases when:
- vma was reused but not yet added into the tree (detached check)
- vma was reused at a different address range (address check); We are missing the check for vm_mm to ensure the reused vma was not attached to a different mm. This patch adds the missing check. For case #2, we pass mm to vma_start_read() to prevent access to unstable vma->vm_mm.
For case #3, we ensure the order in which vma->detached flag and vm_start/vm_end/vm_mm are set and checked. vma gets attached after vm_start/vm_end/vm_mm were set and lock_vma_under_rcu() should check vma->detached before checking vm_start/vm_end/vm_mm. This is required because attaching vma happens without vma write-lock, as opposed to vma detaching, which requires vma write-lock. This patch adds memory barriers inside is_vma_detached() and vma_mark_attached() needed to order reads and writes to vma->detached vs vm_start/vm_end/vm_mm. After these provisions, SLAB_TYPESAFE_BY_RCU is added to vm_area_cachep. This will facilitate vm_area_struct reuse and will minimize the number of call_rcu() calls.
Adding a freeptr_t into vm_area_struct (unioned with vm_start/vm_end) could be used to avoids bloating the structure, however currently custom free pointers are not supported in combination with a ctor (see the comment for kmem_cache_args.freeptr_offset).

---
 include/linux/mm.h               | 60 +++++++++++++++++++++++++++-----
 include/linux/mm_types.h         | 13 +++----
 kernel/fork.c                    | 53 +++++++++++++++++-----------
 mm/memory.c                      | 15 +++++---
 mm/vma.c                         |  2 +-
 tools/testing/vma/vma_internal.h |  7 ++--
 6 files changed, 103 insertions(+), 47 deletions(-)
include/linux/mm.h +51 -9
index dd1b6190df28..2a4794b7a513 100644
@@ -257,7 +257,7 @@ struct vm_area_struct *vm_area_alloc(struct mm_struct *);
 struct vm_area_struct *vm_area_dup(struct vm_area_struct *);
 void vm_area_free(struct vm_area_struct *);
 /* Use only if VMA has no other users */
-void __vm_area_free(struct vm_area_struct *vma);
+void vm_area_free_unreachable(struct vm_area_struct *vma);
 
 #ifndef CONFIG_MMU
 extern struct rb_root nommu_region_tree;
@@ -690,12 +690,32 @@ static inline void vma_lock_init(struct vm_area_struct *vma)
 	vma->vm_lock_seq = UINT_MAX;
 }
 
+#define VMA_BEFORE_LOCK		offsetof(struct vm_area_struct, vm_lock)
+#define VMA_LOCK_END(vma)	\
+	(((void *)(vma)) + offsetofend(struct vm_area_struct, vm_lock))
+#define VMA_AFTER_LOCK		\
+	(sizeof(struct vm_area_struct) - offsetofend(struct vm_area_struct, vm_lock))
+
+static inline void vma_clear(struct vm_area_struct *vma)
+{
+	/* Preserve vma->vm_lock */
+	memset(vma, 0, VMA_BEFORE_LOCK);
+	memset(VMA_LOCK_END(vma), 0, VMA_AFTER_LOCK);
+}
+
+static inline void vma_copy(struct vm_area_struct *new, struct vm_area_struct *orig)
+{
+	/* Preserve vma->vm_lock */
+	data_race(memcpy(new, orig, VMA_BEFORE_LOCK));
+	data_race(memcpy(VMA_LOCK_END(new), VMA_LOCK_END(orig), VMA_AFTER_LOCK));
+}
+
 /*
  * Try to read-lock a vma. The function is allowed to occasionally yield false
  * locked result to avoid performance overhead, in which case we fall back to
  * using mmap_lock. The function should never yield false unlocked result.
  */
-static inline bool vma_start_read(struct vm_area_struct *vma)
+static inline bool vma_start_read(struct mm_struct *mm, struct vm_area_struct *vma)
 {
 	/*
 	 * Check before locking. A race might cause false locked result.
@@ -704,7 +724,7 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 	 * we don't rely on for anything - the mm_lock_seq read against which we
 	 * need ordering is below.
 	 */
-	if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq.sequence))
+	if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(mm->mm_lock_seq.sequence))
 		return false;
 
 	if (unlikely(down_read_trylock(&vma->vm_lock.lock) == 0))
@@ -721,7 +741,7 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 	 * after it has been unlocked.
 	 * This pairs with RELEASE semantics in vma_end_write_all().
 	 */
-	if (unlikely(vma->vm_lock_seq == raw_read_seqcount(&vma->vm_mm->mm_lock_seq))) {
+	if (unlikely(vma->vm_lock_seq == raw_read_seqcount(&mm->mm_lock_seq))) {
 		up_read(&vma->vm_lock.lock);
 		return false;
 	}
@@ -810,7 +830,15 @@ static inline void vma_assert_locked(struct vm_area_struct *vma)
 
 static inline void vma_mark_attached(struct vm_area_struct *vma)
 {
-	vma->detached = false;
+	/*
+	 * This pairs with smp_rmb() inside is_vma_detached().
+	 * vma is marked attached after all vma modifications are done and it
+	 * got added into the vma tree. All prior vma modifications should be
+	 * made visible before marking the vma attached.
+	 */
+	smp_wmb();
+	/* This pairs with READ_ONCE() in is_vma_detached(). */
+	WRITE_ONCE(vma->detached, false);
 }
 
 static inline void vma_mark_detached(struct vm_area_struct *vma)
@@ -822,7 +850,18 @@ static inline void vma_mark_detached(struct vm_area_struct *vma)
 
 static inline bool is_vma_detached(struct vm_area_struct *vma)
 {
-	return vma->detached;
+	bool detached;
+
+	/* This pairs with WRITE_ONCE() in vma_mark_attached(). */
+	detached = READ_ONCE(vma->detached);
+	/*
+	 * This pairs with smp_wmb() inside vma_mark_attached() to ensure
+	 * vma->detached is read before vma attributes read later inside
+	 * lock_vma_under_rcu().
+	 */
+	smp_rmb();
+
+	return detached;
 }
 
 static inline void release_fault_lock(struct vm_fault *vmf)
@@ -847,7 +886,11 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 #else /* CONFIG_PER_VMA_LOCK */
 
 static inline void vma_lock_init(struct vm_area_struct *vma) {}
-static inline bool vma_start_read(struct vm_area_struct *vma)
+static inline void vma_clear(struct vm_area_struct *vma)
+		{ memset(vma, 0, sizeof(*vma)); }
+static inline void vma_copy(struct vm_area_struct *new, struct vm_area_struct *orig)
+		{ data_race(memcpy(new, orig, sizeof(*new))); }
+static inline bool vma_start_read(struct mm_struct *mm, struct vm_area_struct *vma)
 		{ return false; }
 static inline void vma_end_read(struct vm_area_struct *vma) {}
 static inline void vma_start_write(struct vm_area_struct *vma) {}
@@ -883,7 +926,7 @@ extern const struct vm_operations_struct vma_dummy_vm_ops;
 
 static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 {
-	memset(vma, 0, sizeof(*vma));
+	vma_clear(vma);
 	vma->vm_mm = mm;
 	vma->vm_ops = &vma_dummy_vm_ops;
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
@@ -892,7 +935,6 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
 	vma->detached = true;
 #endif
 	vma_numab_state_init(vma);
-	vma_lock_init(vma);
 }
 
 /* Use when VMA is not part of the VMA tree and needs no locking */
include/linux/mm_types.h +4 -9
index 5c4bfdcfac72..8f6b0c935c2b 100644
@@ -667,15 +667,10 @@ struct vma_numab_state {
 struct vm_area_struct {
 	/* The first cache line has the info for VMA tree walking. */
 
-	union {
-		struct {
-			/* VMA covers [vm_start; vm_end) addresses within mm */
-			unsigned long vm_start;
-			unsigned long vm_end;
-		};
-#ifdef CONFIG_PER_VMA_LOCK
-		struct rcu_head vm_rcu;	/* Used for deferred freeing. */
-#endif
+	struct {
+		/* VMA covers [vm_start; vm_end) addresses within mm */
+		unsigned long vm_start;
+		unsigned long vm_end;
 	};
 
 	/*
kernel/fork.c +32 -21

87 lines not on this page · raw message

mm/memory.c +11 -4
index d0197a0c0996..b5fbc71b46bd 100644
@@ -6275,10 +6275,16 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 	if (!vma)
 		goto inval;
 
-	if (!vma_start_read(vma))
+	if (!vma_start_read(mm, vma))
 		goto inval;
 
-	/* Check if the VMA got isolated after we found it */
+	/*
+	 * Check if the VMA got isolated after we found it.
+	 * Note: vma we found could have been recycled and is being reattached.
+	 * It's possible to attach a vma while it is read-locked, however a
+	 * read-locked vma can't be detached (detaching requires write-locking).
+	 * Therefore if this check passes, we have an attached and stable vma.
+	 */
 	if (is_vma_detached(vma)) {
 		vma_end_read(vma);
 		count_vm_vma_lock_event(VMA_LOCK_MISS);
@@ -6292,8 +6298,9 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 	 * fields are accessible for RCU readers.
 	 */
 
-	/* Check since vm_start/vm_end might change before we lock the VMA */
-	if (unlikely(address < vma->vm_start || address >= vma->vm_end))
+	/* Check if the vma we locked is the right one. */
+	if (unlikely(vma->vm_mm != mm ||
+		     address < vma->vm_start || address >= vma->vm_end))
 		goto inval_end_read;
 
 	rcu_read_unlock();
mm/vma.c +1 -1
index 73104d434567..050b83df3df2 100644
@@ -382,7 +382,7 @@ void remove_vma(struct vm_area_struct *vma, bool unreachable)
 		fput(vma->vm_file);
 	mpol_put(vma_policy(vma));
 	if (unreachable)
-		__vm_area_free(vma);
+		vm_area_free_unreachable(vma);
 	else
 		vm_area_free(vma);
 }
tools/testing/vma/vma_internal.h +4 -3
index 2fed366d20ef..fd668d6cafc0 100644
@@ -632,14 +632,15 @@ static inline void mpol_put(struct mempolicy *)
 {
 }
 
-static inline void __vm_area_free(struct vm_area_struct *vma)
+static inline void vm_area_free(struct vm_area_struct *vma)
 {
 	free(vma);
 }
 
-static inline void vm_area_free(struct vm_area_struct *vma)
+static inline void vm_area_free_unreachable(struct vm_area_struct *vma)
 {
-	__vm_area_free(vma);
+	vma->detached = true;
+	vm_area_free(vma);
 }
 
 static inline void lru_add_drain(void)
Signature · 1 line
2.47.0.338.g60cca15819-goog
#16 lore raw

Re: [PATCH v4 4/5] mm: make vma cache SLAB_TYPESAFE_BY_RCU

include/linux/mm.h static inline void vma_lock_init(struct vm_area_struct *vma) · line 704

Quoted 6 lines · Suren Baghdasaryan
include/linux/mm.h +6 -0
+static inline void vma_clear(struct vm_area_struct *vma)
+{
+	/* Preserve vma->vm_lock */
+	memset(vma, 0, VMA_BEFORE_LOCK);
+	memset(VMA_LOCK_END(vma), 0, VMA_AFTER_LOCK);
+}

This isn't how you're supposed to handle constructors. You've fixed the immediate problem rather than writing the code in the intended style.

kernel/fork.c static struct kmem_cache *vm_area_cachep; · line 442

Quoted 4 lines
kernel/fork.c +4 -0
+static void vm_area_ctor(void *data)
+{
+	vma_lock_init(data);
+}

After the ctor has run, the object should be in the same state as it is after it's freed. If you want to memset the entire thing then you can do it in the ctor. But there should be no need to do it in vma_init().

And there's lots of things you can move from vma_init() to the ctor. For example, at free time, anon_vma_chain should be an empty list. So if you init it in the ctor, you can avoid doing it in vma_init(). I'd suggest that vma_numab_state_free() should be the place which sets vma->numab_state to NULL and we can delete vma_numab_state_init() entirely.

author maintainer
#17 lore raw

Re: [PATCH v4 4/5] mm: make vma cache SLAB_TYPESAFE_BY_RCU

Quoted 11 lines · Matthew Wilcox

include/linux/mm.h static inline void vma_lock_init(struct vm_area_struct *vma) · line 704

Quoted 6 lines · Suren Baghdasaryan
include/linux/mm.h +6 -0
+static inline void vma_clear(struct vm_area_struct *vma)
+{
+     /* Preserve vma->vm_lock */
+     memset(vma, 0, VMA_BEFORE_LOCK);
+     memset(VMA_LOCK_END(vma), 0, VMA_AFTER_LOCK);
+}

This isn't how you're supposed to handle constructors. You've fixed the immediate problem rather than writing the code in the intended style.

Yeah, I don't like this myself but the only alternative I can think of is to set the struct members individually.

Quoted 10 lines

kernel/fork.c static struct kmem_cache *vm_area_cachep; · line 442

Quoted 5 lines
kernel/fork.c +4 -0
+static void vm_area_ctor(void *data)
+{
+     vma_lock_init(data);
+}

After the ctor has run, the object should be in the same state as it is after it's freed. If you want to memset the entire thing then you can do it in the ctor. But there should be no need to do it in vma_init().

IIUC, your suggestion is to memset() the vma and initialize vm_lock inside the ctor. Then when it's time to free the vma, we reset all members except vm_lock before freeing the vma. As you mention later, members like anon_vma_chain, which are already clear, also won't need to be reset at this point. Am I understanding your proposal correctly?

BTW, if so, then vma_copy() will have to also copy vma members individually.

Quoted 4 lines

And there's lots of things you can move from vma_init() to the ctor. For example, at free time, anon_vma_chain should be an empty list. So if you init it in the ctor, you can avoid doing it in vma_init().

True.

Quoted 3 lines

I'd suggest that vma_numab_state_free() should be the place which sets vma->numab_state to NULL and we can delete vma_numab_state_init() entirely.

Sounds good to me.

Please confirm if I correctly got your idea and I'll update this patch. Thanks for the feedback!

Quoted 1 line
author maintainer
#18 lore raw

Re: [PATCH v4 4/5] mm: make vma cache SLAB_TYPESAFE_BY_RCU

Quoted 41 lines · Suren Baghdasaryan
Quoted 11 lines · Matthew Wilcox

include/linux/mm.h static inline void vma_lock_init(struct vm_area_struct *vma) · line 704

Quoted 6 lines · Suren Baghdasaryan
include/linux/mm.h +6 -0
+static inline void vma_clear(struct vm_area_struct *vma)
+{
+     /* Preserve vma->vm_lock */
+     memset(vma, 0, VMA_BEFORE_LOCK);
+     memset(VMA_LOCK_END(vma), 0, VMA_AFTER_LOCK);
+}

This isn't how you're supposed to handle constructors. You've fixed the immediate problem rather than writing the code in the intended style.

Yeah, I don't like this myself but the only alternative I can think of is to set the struct members individually.

Quoted 10 lines

kernel/fork.c static struct kmem_cache *vm_area_cachep; · line 442

Quoted 5 lines
kernel/fork.c +4 -0
+static void vm_area_ctor(void *data)
+{
+     vma_lock_init(data);
+}

After the ctor has run, the object should be in the same state as it is after it's freed. If you want to memset the entire thing then you can do it in the ctor. But there should be no need to do it in vma_init().

IIUC, your suggestion is to memset() the vma and initialize vm_lock inside the ctor. Then when it's time to free the vma, we reset all members except vm_lock before freeing the vma. As you mention later, members like anon_vma_chain, which are already clear, also won't need to be reset at this point. Am I understanding your proposal correctly?

BTW, if so, then vma_copy() will have to also copy vma members individually.

Quoted 4 lines

And there's lots of things you can move from vma_init() to the ctor. For example, at free time, anon_vma_chain should be an empty list. So if you init it in the ctor, you can avoid doing it in vma_init().

Quoted 7 lines

True.

Quoted 3 lines

I'd suggest that vma_numab_state_free() should be the place which sets vma->numab_state to NULL and we can delete vma_numab_state_init() entirely.

Sounds good to me.

I took a stab at it and the result does not look pretty... Couple notes:
- vma_init() is used in other places to initialize VMAs allocated on the stack, so I left it alone for such cases. VMAs like that are not allocated from vm_area_cachep, can't be reused anyway, therefore we can override their vm_lock.
- Since vma_init() has to stay, we can't retire vma_numab_state_init() because it's used in vma_init().
- I think resetting members before freeing might not be such a good idea because after resetting the object might not be reused at all.

Now, the main point:
I moved initializations of several members into ctor but even with that the code looks roughly like this:

static void vm_area_ctor(void *data)

{
    struct vm_area_struct *vma = (struct vm_area_struct *)data;

    vma->detached = true;
    INIT_LIST_HEAD(&vma->anon_vma_chain);
    vma_lock_init(vma);
}

struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)

{
    struct vm_area_struct *vma;

    vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
    if (!vma)
        return NULL;

    vma->vm_mm = mm;
    vma->vm_ops = &vma_dummy_vm_ops;
    vma->vm_start = 0;
    vma->vm_end = 0;
    memset(&vma->vm_page_prot, 0, sizeof(vma->vm_page_prot));
    vm_flags_init(vma, 0);
    vma_numab_state_init(vma);
    memset(&vma->shared, 0, sizeof(vma->shared));
    vma->anon_vma = NULL;
    vma->vm_pgoff = 0;
    vma->vm_file = NULL;
    vma->vm_private_data = NULL;
    memset(&vma->vm_userfaultfd_ctx, 0, sizeof(vma->vm_userfaultfd_ctx));
#ifdef CONFIG_ANON_VMA_NAME
    vma->anon_name = NULL;
#endif
#ifdef CONFIG_SWAP
    atomic_long_set(&vma->swap_readahead_info, 0);
#endif
#ifndef CONFIG_MMU
    vma->vm_region = NULL;
#endif
#ifdef CONFIG_NUMA
    vma->vm_policy = NULL;
#endif
#ifdef CONFIG_NUMA_BALANCING
    vma->numab_state = NULL;
#endif
    return vma;
}

I can of course add helper functions and get rid of the #ifdef's but still...

Matthew, want to double check if this looks like the solution you were proposing or am I completely off the target?

Quoted 5 lines

Please confirm if I correctly got your idea and I'll update this patch. Thanks for the feedback!

Quoted 1 line
#19 lore raw

Re: [PATCH v4 4/5] mm: make vma cache SLAB_TYPESAFE_BY_RCU

In reply to Suren Baghdasaryan
Quoted 14 lines · Suren Baghdasaryan

To enable SLAB_TYPESAFE_BY_RCU for vma cache we need to ensure that object reuse before RCU grace period is over will be detected inside lock_vma_under_rcu().
lock_vma_under_rcu() enters RCU read section, finds the vma at the given address, locks the vma and checks if it got detached or remapped to cover a different address range. These last checks are there to ensure that the vma was not modified after we found it but before locking it.
vma reuse introduces several new possibilities:
1. vma can be reused after it was found but before it is locked;
2. vma can be reused and reinitialized (including changing its vm_mm) while being locked in vma_start_read();
3. vma can be reused and reinitialized after it was found but before it is locked, then attached at a new address or to a new mm while being

Quoted 8 lines

read-locked;
For case #1 current checks will help detecting cases when:
- vma was reused but not yet added into the tree (detached check)
- vma was reused at a different address range (address check); We are missing the check for vm_mm to ensure the reused vma was not attached to a different mm. This patch adds the missing check. For case #2, we pass mm to vma_start_read() to prevent access to unstable vma->vm_mm.

So we may now be looking at different mm's mm_lock_seq.sequence and return a false unlocked result, right? I guess the mm validation in lock_vma_under_rcu() handles that, but maybe the comment of vma_start_read() needs updating.

Quoted 7 lines

For case #3, we ensure the order in which vma->detached flag and vm_start/vm_end/vm_mm are set and checked. vma gets attached after vm_start/vm_end/vm_mm were set and lock_vma_under_rcu() should check vma->detached before checking vm_start/vm_end/vm_mm. This is required because attaching vma happens without vma write-lock, as opposed to vma detaching, which requires vma write-lock. This patch adds memory barriers inside is_vma_detached() and vma_mark_attached() needed to

Quoted 8 lines

order reads and writes to vma->detached vs vm_start/vm_end/vm_mm. After these provisions, SLAB_TYPESAFE_BY_RCU is added to vm_area_cachep. This will facilitate vm_area_struct reuse and will minimize the number of call_rcu() calls.
Adding a freeptr_t into vm_area_struct (unioned with vm_start/vm_end) could be used to avoids bloating the structure, however currently custom free pointers are not supported in combination with a ctor (see the comment for kmem_cache_args.freeptr_offset).

I think there's nothing fundamental preventing to support that, there was just no user of it. We can do it later.

Quoted 16 lines
kernel/fork.c +5 -0
@@ -436,6 +436,11 @@ static struct kmem_cache *vm_area_cachep;
 /* SLAB cache for mm_struct structures (tsk->mm) */
 static struct kmem_cache *mm_cachep;
 
+static void vm_area_ctor(void *data)
+{
+	vma_lock_init(data);
+}
+
 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 {
 	struct vm_area_struct *vma;

kernel/fork.c struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) · line 473

Quoted 9 lines
kernel/fork.c +1 -2
@@ -462,8 +467,7 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 	 * orig->shared.rb may be modified concurrently, but the clone
 	 * will be reinitialized.
 	 */
-	data_race(memcpy(new, orig, sizeof(*new)));
-	vma_lock_init(new);
+	vma_copy(new, orig);
 	INIT_LIST_HEAD(&new->anon_vma_chain);
 #ifdef CONFIG_PER_VMA_LOCK
 	/* vma is not locked, can't use vma_mark_detached() */

Here we mark it detached but we might have already copied it as attached and confused a reader?

I think this will be covered by what you said in reply to willy:
"vma_copy() will have to also copy vma members individually."

Quoted 4 lines
kernel/fork.c +0 -0
@@ -475,32 +479,37 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 	return new;
 }
 
author maintainer
#20 lore raw

Re: [PATCH v4 4/5] mm: make vma cache SLAB_TYPESAFE_BY_RCU

Quoted 25 lines · Vlastimil Babka
Quoted 14 lines · Suren Baghdasaryan

To enable SLAB_TYPESAFE_BY_RCU for vma cache we need to ensure that object reuse before RCU grace period is over will be detected inside lock_vma_under_rcu().
lock_vma_under_rcu() enters RCU read section, finds the vma at the given address, locks the vma and checks if it got detached or remapped to cover a different address range. These last checks are there to ensure that the vma was not modified after we found it but before locking it.
vma reuse introduces several new possibilities:
1. vma can be reused after it was found but before it is locked;
2. vma can be reused and reinitialized (including changing its vm_mm) while being locked in vma_start_read();
3. vma can be reused and reinitialized after it was found but before it is locked, then attached at a new address or to a new mm while being

Quoted 8 lines

read-locked;
For case #1 current checks will help detecting cases when:
- vma was reused but not yet added into the tree (detached check)
- vma was reused at a different address range (address check); We are missing the check for vm_mm to ensure the reused vma was not attached to a different mm. This patch adds the missing check. For case #2, we pass mm to vma_start_read() to prevent access to unstable vma->vm_mm.

Quoted 4 lines

So we may now be looking at different mm's mm_lock_seq.sequence and return a false unlocked result, right? I guess the mm validation in lock_vma_under_rcu() handles that, but maybe the comment of vma_start_read() needs updating.

Correct. I'll add a comment about this.

Quoted 17 lines
Quoted 16 lines

For case #3, we ensure the order in which vma->detached flag and vm_start/vm_end/vm_mm are set and checked. vma gets attached after vm_start/vm_end/vm_mm were set and lock_vma_under_rcu() should check vma->detached before checking vm_start/vm_end/vm_mm. This is required because attaching vma happens without vma write-lock, as opposed to vma detaching, which requires vma write-lock. This patch adds memory barriers inside is_vma_detached() and vma_mark_attached() needed to order reads and writes to vma->detached vs vm_start/vm_end/vm_mm. After these provisions, SLAB_TYPESAFE_BY_RCU is added to vm_area_cachep. This will facilitate vm_area_struct reuse and will minimize the number of call_rcu() calls.
Adding a freeptr_t into vm_area_struct (unioned with vm_start/vm_end) could be used to avoids bloating the structure, however currently custom free pointers are not supported in combination with a ctor (see the comment for kmem_cache_args.freeptr_offset).

Quoted 2 lines

I think there's nothing fundamental preventing to support that, there was just no user of it. We can do it later.

Oh, ok. I can add it back so that we have one user and then when the mechanism is implemented it can be used for testing. Adding freeptr_t has no negative effects and will reduce later churn.

Quoted 21 lines
Quoted 21 lines
kernel/fork.c +5 -1
@@ -436,6 +436,11 @@ static struct kmem_cache *vm_area_cachep;
 /* SLAB cache for mm_struct structures (tsk->mm) */
 static struct kmem_cache *mm_cachep;
 
+static void vm_area_ctor(void *data)
+{
+     vma_lock_init(data);
+}
+
 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 {
      struct vm_area_struct *vma;
@@ -462,8 +467,7 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
       * orig->shared.rb may be modified concurrently, but the clone
       * will be reinitialized.
       */
-     data_race(memcpy(new, orig, sizeof(*new)));
Quoted 8 lines

kernel/fork.c struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) · line 473

Quoted 5 lines
kernel/fork.c +1 -1
-     vma_lock_init(new);
+     vma_copy(new, orig);
      INIT_LIST_HEAD(&new->anon_vma_chain);
 #ifdef CONFIG_PER_VMA_LOCK
      /* vma is not locked, can't use vma_mark_detached() */

Here we mark it detached but we might have already copied it as attached and confused a reader?

Very true. Thanks for catching this one!

Quoted 3 lines

I think this will be covered by what you said in reply to willy:
"vma_copy() will have to also copy vma members individually."

Yes, I think so. vma_copy() will need to copy most but not all members. vma->detached will be among those not copied. Thanks!

Quoted 5 lines
Quoted 5 lines
kernel/fork.c +0 -0
@@ -475,32 +479,37 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
      return new;
 }
 
author maintainer
#21 lore raw

PATCH v4 5/5 docs/mm: document latest changes to vm_lock

In reply to Suren Baghdasaryan

Change the documentation to reflect that vm_lock is integrated into vma. Document newly introduced vma_start_read_locked{_nested} functions.

---
 Documentation/mm/process_addrs.rst | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
Documentation/mm/process_addrs.rst +7 -3
index 1bf7ad010fc0..a18450b6496d 100644
@@ -686,7 +686,11 @@ calls :c:func:`!rcu_read_lock` to ensure that the VMA is looked up in an RCU
 critical section, then attempts to VMA lock it via :c:func:`!vma_start_read`,
 before releasing the RCU lock via :c:func:`!rcu_read_unlock`.
 
-VMA read locks hold the read lock on the :c:member:`!vma->vm_lock` semaphore for
+In cases when the user already holds mmap read lock, :c:func:`!vma_start_read_locked`
+and :c:func:`!vma_start_read_locked_nested` can be used. These functions always
+succeed in acquiring VMA read lock.
+
+VMA read locks hold the read lock on the :c:member:`!vma.vm_lock` semaphore for
 their duration and the caller of :c:func:`!lock_vma_under_rcu` must release it
 via :c:func:`!vma_end_read`.
 
@@ -750,7 +754,7 @@ keep VMAs locked across entirely separate write operations. It also maintains
 correct lock ordering.
 
 Each time a VMA read lock is acquired, we acquire a read lock on the
-:c:member:`!vma->vm_lock` read/write semaphore and hold it, while checking that
+:c:member:`!vma.vm_lock` read/write semaphore and hold it, while checking that
 the sequence count of the VMA does not match that of the mm.
 
 If it does, the read lock fails. If it does not, we hold the lock, excluding
@@ -760,7 +764,7 @@ Importantly, maple tree operations performed in :c:func:`!lock_vma_under_rcu`
 are also RCU safe, so the whole read lock operation is guaranteed to function
 correctly.
 
-On the write side, we acquire a write lock on the :c:member:`!vma->vm_lock`
+On the write side, we acquire a write lock on the :c:member:`!vma.vm_lock`
 read/write semaphore, before setting the VMA's sequence number under this lock,
 also simultaneously holding the mmap write lock.
 
Signature · 1 line
2.47.0.338.g60cca15819-goog
reviewer
#22 lore raw
In reply to Suren Baghdasaryan
Quoted 6 lines · Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].

If 'struct vm_area_struct' is prone to performance issues due to cacheline misalignments then we should do something about the __randomize_layout tag for it. I imagine we can identify the fields which might be performance critical to be on same cacheline or different cacheline due to false sharing then we can divide the fields into different cacheline groups and fields can be __randomize_layout within the group. WDYT?

author maintainer
#23 lore raw
Quoted 9 lines · Shakeel Butt
Quoted 6 lines · Suren Baghdasaryan

Back when per-vma locks were introduces, vm_lock was moved out of vm_area_struct in [1] because of the performance regression caused by false cacheline sharing. Recent investigation [2] revealed that the regressions is limited to a rather old Broadwell microarchitecture and even there it can be mitigated by disabling adjacent cacheline prefetching, see [3].

Quoted 7 lines

If 'struct vm_area_struct' is prone to performance issues due to cacheline misalignments then we should do something about the __randomize_layout tag for it. I imagine we can identify the fields which might be performance critical to be on same cacheline or different cacheline due to false sharing then we can divide the fields into different cacheline groups and fields can be __randomize_layout within the group. WDYT?

I think that's a good idea since shuffling these fields around does affect performance. I can look into it once these changes get finalized and the layout gets more stable.

Quoted 1 line
#24 lore raw
In reply to Shakeel Butt
Quoted 7 lines · Shakeel Butt

If 'struct vm_area_struct' is prone to performance issues due to cacheline misalignments then we should do something about the __randomize_layout tag for it. I imagine we can identify the fields which might be performance critical to be on same cacheline or different cacheline due to false sharing then we can divide the fields into different cacheline groups and fields can be __randomize_layout within the group. WDYT?

Pretty sure the people who think security is more important than performance are the only ones who randomize structs.

#25 lore raw
Quoted 11 lines · Matthew Wilcox
Quoted 7 lines · Shakeel Butt

If 'struct vm_area_struct' is prone to performance issues due to cacheline misalignments then we should do something about the __randomize_layout tag for it. I imagine we can identify the fields which might be performance critical to be on same cacheline or different cacheline due to false sharing then we can divide the fields into different cacheline groups and fields can be __randomize_layout within the group. WDYT?

Pretty sure the people who think security is more important than performance are the only ones who randomize structs.

I agree that I don't think we need concern ourselves with users of this setting for precisely this reason.

I wouldn't want supporting this to cause difficulty for users who do not enable this when those who do aren't really concerned about the perf issues as Matthew says.

author maintainer
#26 lore raw
Quoted 14 lines · Lorenzo Stoakes
Quoted 11 lines · Matthew Wilcox
Quoted 7 lines · Shakeel Butt

If 'struct vm_area_struct' is prone to performance issues due to cacheline misalignments then we should do something about the __randomize_layout tag for it. I imagine we can identify the fields which might be performance critical to be on same cacheline or different cacheline due to false sharing then we can divide the fields into different cacheline groups and fields can be __randomize_layout within the group. WDYT?

Pretty sure the people who think security is more important than performance are the only ones who randomize structs.

Quoted 6 lines

I agree that I don't think we need concern ourselves with users of this setting for precisely this reason.

I wouldn't want supporting this to cause difficulty for users who do not enable this when those who do aren't really concerned about the perf issues as Matthew says.

Ack. Will keep it as is. Thanks!