mm: replace vm_lock and detached flag with a reference count

rw_semaphore is a sizable structure of 40 bytes and consumes considerable
space for each vm_area_struct.  However vma_lock has two important
specifics which can be used to replace rw_semaphore with a simpler
structure:

1. Readers never wait.  They try to take the vma_lock and fall back to
   mmap_lock if that fails.

2. Only one writer at a time will ever try to write-lock a vma_lock
   because writers first take mmap_lock in write mode.  Because of these
   requirements, full rw_semaphore functionality is not needed and we can
   replace rw_semaphore and the vma->detached flag with a refcount
   (vm_refcnt).

When vma is in detached state, vm_refcnt is 0 and only a call to
vma_mark_attached() can take it out of this state.  Note that unlike
before, now we enforce both vma_mark_attached() and vma_mark_detached() to
be done only after vma has been write-locked.  vma_mark_attached() changes
vm_refcnt to 1 to indicate that it has been attached to the vma tree. 
When a reader takes read lock, it increments vm_refcnt, unless the top
usable bit of vm_refcnt (0x40000000) is set, indicating presence of a
writer.  When writer takes write lock, it sets the top usable bit to
indicate its presence.  If there are readers, writer will wait using newly
introduced mm->vma_writer_wait.  Since all writers take mmap_lock in write
mode first, there can be only one writer at a time.  The last reader to
release the lock will signal the writer to wake up.  refcount might
overflow if there are many competing readers, in which case read-locking
will fail.  Readers are expected to handle such failures.

In summary:
1. all readers increment the vm_refcnt;
2. writer sets top usable (writer) bit of vm_refcnt;
3. readers cannot increment the vm_refcnt if the writer bit is set;
4. in the presence of readers, writer must wait for the vm_refcnt to drop
to 1 (plus the VMA_LOCK_OFFSET writer bit), indicating an attached vma
with no readers;
5. vm_refcnt overflow is handled by the readers.

While this vm_lock replacement does not yet result in a smaller
vm_area_struct (it stays at 256 bytes due to cacheline alignment), it
allows for further size optimization by structure member regrouping to
bring the size of vm_area_struct below 192 bytes.

[surenb@google.com: fix a crash due to vma_end_read() that should have been removed]
  Link: https://lkml.kernel.org/r/20250220200208.323769-1-surenb@google.com
Link: https://lkml.kernel.org/r/20250213224655.1680278-13-surenb@google.com
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Matthew Wilcox <willy@infradead.org>
Tested-by: Shivank Garg <shivankg@amd.com>
  Link: https://lkml.kernel.org/r/5e19ec93-8307-47c2-bb13-3ddf7150624e@amd.com
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christian Brauner <brauner@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Klara Modin <klarasmodin@gmail.com>
Cc: Liam R. Howlett <Liam.Howlett@Oracle.com>
Cc: Lokesh Gidra <lokeshgidra@google.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mateusz Guzik <mjguzik@gmail.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Sourav Panda <souravpanda@google.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

2 files changed, 95 insertions, 55 deletions

diff --git a/include/linux/mm.h b/include/linux/mm.h
index c24c521e38a2..06f179c844c3 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -32,6 +32,7 @@
 #include <linux/memremap.h>
 #include <linux/slab.h>
 #include <linux/cacheinfo.h>
+#include <linux/rcuwait.h>
 
 struct mempolicy;
 struct anon_vma;
@@ -697,19 +698,54 @@ 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)
+static inline void vma_lock_init(struct vm_area_struct *vma, bool reset_refcnt)
 {
-	init_rwsem(&vma->vm_lock.lock);
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	static struct lock_class_key lockdep_key;
+
+	lockdep_init_map(&vma->vmlock_dep_map, "vm_lock", &lockdep_key, 0);
+#endif
+	if (reset_refcnt)
+		refcount_set(&vma->vm_refcnt, 0);
 	vma->vm_lock_seq = UINT_MAX;
 }
 
+static inline bool is_vma_writer_only(int refcnt)
+{
+	/*
+	 * With a writer and no readers, refcnt is VMA_LOCK_OFFSET if the vma
+	 * is detached and (VMA_LOCK_OFFSET + 1) if it is attached. Waiting on
+	 * a detached vma happens only in vma_mark_detached() and is a rare
+	 * case, therefore most of the time there will be no unnecessary wakeup.
+	 */
+	return refcnt & VMA_LOCK_OFFSET && refcnt <= VMA_LOCK_OFFSET + 1;
+}
+
+static inline void vma_refcount_put(struct vm_area_struct *vma)
+{
+	/* Use a copy of vm_mm in case vma is freed after we drop vm_refcnt */
+	struct mm_struct *mm = vma->vm_mm;
+	int oldcnt;
+
+	rwsem_release(&vma->vmlock_dep_map, _RET_IP_);
+	if (!__refcount_dec_and_test(&vma->vm_refcnt, &oldcnt)) {
+
+		if (is_vma_writer_only(oldcnt - 1))
+			rcuwait_wake_up(&mm->vma_writer_wait);
+	}
+}
+
 /*
  * 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.
+ * Returns the vma on success, NULL on failure to lock and EAGAIN if vma got
+ * detached.
  */
-static inline bool vma_start_read(struct vm_area_struct *vma)
+static inline struct vm_area_struct *vma_start_read(struct vm_area_struct *vma)
 {
+	int oldcnt;
+
 	/*
 	 * Check before locking. A race might cause false locked result.
 	 * We can use READ_ONCE() for the mm_lock_seq here, and don't need
@@ -718,15 +754,25 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
 	 * need ordering is below.
 	 */
 	if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq.sequence))
-		return false;
+		return NULL;
 
-	if (unlikely(down_read_trylock(&vma->vm_lock.lock) == 0))
-		return false;
+	/*
+	 * If VMA_LOCK_OFFSET is set, __refcount_inc_not_zero_limited_acquire()
+	 * will fail because VMA_REF_LIMIT is less than VMA_LOCK_OFFSET.
+	 * Acquire fence is required here to avoid reordering against later
+	 * vm_lock_seq check and checks inside lock_vma_under_rcu().
+	 */
+	if (unlikely(!__refcount_inc_not_zero_limited_acquire(&vma->vm_refcnt, &oldcnt,
+							      VMA_REF_LIMIT))) {
+		/* return EAGAIN if vma got detached from under us */
+		return oldcnt ? NULL : ERR_PTR(-EAGAIN);
+	}
 
+	rwsem_acquire_read(&vma->vmlock_dep_map, 0, 1, _RET_IP_);
 	/*
-	 * Overflow might produce false locked result.
+	 * Overflow of vm_lock_seq/mm_lock_seq might produce false locked result.
 	 * False unlocked result is impossible because we modify and check
-	 * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq
+	 * vma->vm_lock_seq under vma->vm_refcnt protection and mm->mm_lock_seq
 	 * modification invalidates all existing locks.
 	 *
 	 * We must use ACQUIRE semantics for the mm_lock_seq so that if we are
@@ -735,10 +781,11 @@ 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);
-		return false;
+		vma_refcount_put(vma);
+		return NULL;
 	}
-	return true;
+
+	return vma;
 }
 
 /*
@@ -749,8 +796,14 @@ static inline bool vma_start_read(struct vm_area_struct *vma)
  */
 static inline bool vma_start_read_locked_nested(struct vm_area_struct *vma, int subclass)
 {
+	int oldcnt;
+
 	mmap_assert_locked(vma->vm_mm);
-	down_read_nested(&vma->vm_lock.lock, subclass);
+	if (unlikely(!__refcount_inc_not_zero_limited_acquire(&vma->vm_refcnt, &oldcnt,
+							      VMA_REF_LIMIT)))
+		return false;
+
+	rwsem_acquire_read(&vma->vmlock_dep_map, 0, 1, _RET_IP_);
 	return true;
 }
 
@@ -762,16 +815,12 @@ static inline bool vma_start_read_locked_nested(struct vm_area_struct *vma, int
  */
 static inline bool vma_start_read_locked(struct vm_area_struct *vma)
 {
-	mmap_assert_locked(vma->vm_mm);
-	down_read(&vma->vm_lock.lock);
-	return true;
+	return vma_start_read_locked_nested(vma, 0);
 }
 
 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);
-	rcu_read_unlock();
+	vma_refcount_put(vma);
 }
 
 /* WARNING! Can only be used if mmap_lock is expected to be write-locked */
@@ -813,38 +862,35 @@ 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))
-		vma_assert_write_locked(vma);
+	unsigned int mm_lock_seq;
+
+	VM_BUG_ON_VMA(refcount_read(&vma->vm_refcnt) <= 1 &&
+		      !__is_vma_write_locked(vma, &mm_lock_seq), vma);
 }
 
+/*
+ * WARNING: to avoid racing with vma_mark_attached()/vma_mark_detached(), these
+ * assertions should be made either under mmap_write_lock or when the object
+ * has been isolated under mmap_write_lock, ensuring no competing writers.
+ */
 static inline void vma_assert_attached(struct vm_area_struct *vma)
 {
-	WARN_ON_ONCE(vma->detached);
+	WARN_ON_ONCE(!refcount_read(&vma->vm_refcnt));
 }
 
 static inline void vma_assert_detached(struct vm_area_struct *vma)
 {
-	WARN_ON_ONCE(!vma->detached);
+	WARN_ON_ONCE(refcount_read(&vma->vm_refcnt));
 }
 
 static inline void vma_mark_attached(struct vm_area_struct *vma)
 {
-	vma_assert_detached(vma);
-	vma->detached = false;
-}
-
-static inline void vma_mark_detached(struct vm_area_struct *vma)
-{
-	/* When detaching vma should be write-locked */
 	vma_assert_write_locked(vma);
-	vma_assert_attached(vma);
-	vma->detached = true;
+	vma_assert_detached(vma);
+	refcount_set(&vma->vm_refcnt, 1);
 }
 
-static inline bool is_vma_detached(struct vm_area_struct *vma)
-{
-	return vma->detached;
-}
+void vma_mark_detached(struct vm_area_struct *vma);
 
 static inline void release_fault_lock(struct vm_fault *vmf)
 {
@@ -867,9 +913,9 @@ 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_lock_init(struct vm_area_struct *vma, bool reset_refcnt) {}
+static inline struct vm_area_struct *vma_start_read(struct vm_area_struct *vma)
+		{ return NULL; }
 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)
@@ -910,12 +956,8 @@ 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);
-#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);
+	vma_lock_init(vma, false);
 }
 
 /* Use when VMA is not part of the VMA tree and needs no locking */
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 36dea20cd101..9de0a6cb3c2d 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -19,6 +19,7 @@
 #include <linux/workqueue.h>
 #include <linux/seqlock.h>
 #include <linux/percpu_counter.h>
+#include <linux/types.h>
 
 #include <asm/mmu.h>
 
@@ -629,9 +630,8 @@ static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
 }
 #endif
 
-struct vma_lock {
-	struct rw_semaphore lock;
-};
+#define VMA_LOCK_OFFSET	0x40000000
+#define VMA_REF_LIMIT	(VMA_LOCK_OFFSET - 1)
 
 struct vma_numab_state {
 	/*
@@ -710,18 +710,12 @@ struct vm_area_struct {
 
 #ifdef CONFIG_PER_VMA_LOCK
 	/*
-	 * Flag to indicate areas detached from the mm->mm_mt tree.
-	 * Unstable RCU readers are allowed to read this.
-	 */
-	bool detached;
-
-	/*
 	 * Can only be written (using WRITE_ONCE()) while holding both:
 	 *  - mmap_lock (in write mode)
-	 *  - vm_lock->lock (in write mode)
+	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set
 	 * Can be read reliably while holding one of:
 	 *  - mmap_lock (in read or write mode)
-	 *  - vm_lock->lock (in read or write mode)
+	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set or vm_refcnt > 1
 	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
 	 * while holding nothing (except RCU to keep the VMA struct allocated).
 	 *
@@ -784,7 +778,10 @@ struct vm_area_struct {
 	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;
+	refcount_t vm_refcnt ____cacheline_aligned_in_smp;
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	struct lockdep_map vmlock_dep_map;
+#endif
 #endif
 } __randomize_layout;
 
@@ -920,6 +917,7 @@ struct mm_struct {
 					  * by mmlist_lock
 					  */
 #ifdef CONFIG_PER_VMA_LOCK
+		struct rcuwait vma_writer_wait;
 		/*
 		 * This field has lock-like semantics, meaning it is sometimes
 		 * accessed with ACQUIRE/RELEASE semantics.