Re: REGRESSION: Performance regressions from switching anon_vma->lock to mutex

[Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>]

On Thu, Jun 16, 2011 at 2:05 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> This patch is UNTESTED!

It was also UNATTACHED!

Now it's attached.

                    Linus

 mm/rmap.c |   51 +++++++++++++++++++++++++++++++++++----------------
 1 files changed, 35 insertions(+), 16 deletions(-)

diff --git a/mm/rmap.c b/mm/rmap.c
index 0eb463ea88dd..d1d26900c082 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -208,13 +208,11 @@ static void anon_vma_chain_link(struct vm_area_struct *vma,
 	avc->anon_vma = anon_vma;
 	list_add(&avc->same_vma, &vma->anon_vma_chain);
 
-	anon_vma_lock(anon_vma);
 	/*
 	 * It's critical to add new vmas to the tail of the anon_vma,
 	 * see comment in huge_memory.c:__split_huge_page().
 	 */
 	list_add_tail(&avc->same_anon_vma, &anon_vma->head);
-	anon_vma_unlock(anon_vma);
 }
 
 /*
@@ -224,16 +222,30 @@ static void anon_vma_chain_link(struct vm_area_struct *vma,
 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
 {
 	struct anon_vma_chain *avc, *pavc;
+	struct anon_vma *root = NULL;
 
 	list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
+		struct anon_vma *anon_vma = pavc->anon_vma, *new_root = anon_vma->root;
+
+		if (new_root != root) {
+			if (WARN_ON_ONCE(root))
+				mutex_unlock(&root->mutex);
+			root = new_root;
+			mutex_lock(&root->mutex);
+		}
+
 		avc = anon_vma_chain_alloc();
 		if (!avc)
 			goto enomem_failure;
 		anon_vma_chain_link(dst, avc, pavc->anon_vma);
 	}
+	if (root)
+		mutex_unlock(&root->mutex);
 	return 0;
 
  enomem_failure:
+	if (root)
+		mutex_unlock(&root->mutex);
 	unlink_anon_vmas(dst);
 	return -ENOMEM;
 }
@@ -280,7 +292,9 @@ int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
 	get_anon_vma(anon_vma->root);
 	/* Mark this anon_vma as the one where our new (COWed) pages go. */
 	vma->anon_vma = anon_vma;
+	anon_vma_lock(anon_vma);
 	anon_vma_chain_link(vma, avc, anon_vma);
+	anon_vma_unlock(anon_vma);
 
 	return 0;
 
@@ -291,39 +305,44 @@ int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
 	return -ENOMEM;
 }
 
-static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
+static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain, struct anon_vma *anon_vma)
 {
-	struct anon_vma *anon_vma = anon_vma_chain->anon_vma;
-	int empty;
-
-	/* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
-	if (!anon_vma)
-		return;
-
-	anon_vma_lock(anon_vma);
 	list_del(&anon_vma_chain->same_anon_vma);
 
 	/* We must garbage collect the anon_vma if it's empty */
-	empty = list_empty(&anon_vma->head);
-	anon_vma_unlock(anon_vma);
-
-	if (empty)
+	if (list_empty(&anon_vma->head))
 		put_anon_vma(anon_vma);
 }
 
 void unlink_anon_vmas(struct vm_area_struct *vma)
 {
 	struct anon_vma_chain *avc, *next;
+	struct anon_vma *root = NULL;
 
 	/*
 	 * Unlink each anon_vma chained to the VMA.  This list is ordered
 	 * from newest to oldest, ensuring the root anon_vma gets freed last.
 	 */
 	list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
-		anon_vma_unlink(avc);
+		struct anon_vma *anon_vma = avc->anon_vma;
+
+		/* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
+		if (anon_vma) {
+			struct anon_vma *new_root = anon_vma->root;
+
+			if (new_root != root) {
+				if (WARN_ON_ONCE(root))
+					mutex_unlock(&root->mutex);
+				root = new_root;
+				mutex_lock(&root->mutex);
+			}
+			anon_vma_unlink(avc, anon_vma);
+		}
 		list_del(&avc->same_vma);
 		anon_vma_chain_free(avc);
 	}
+	if (root)
+		mutex_unlock(&root->mutex);
 }
 
 static void anon_vma_ctor(void *data)