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

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Thu, 2011-06-16 at 20:58 -0700, Linus Torvalds wrote:
> Ok, I'm still thinking. I have an approach that I think will handle it
> fairly cleanly, but that involves walking the same_vma list twice:
> once to actually unlink the anon_vma's under the lock, and then a
> second pass that does the rest. It should work. 

Something like so? Compiles and runs the benchmark in question.

---
Index: linux-2.6/mm/rmap.c
===================================================================
--- linux-2.6.orig/mm/rmap.c
+++ linux-2.6/mm/rmap.c
@@ -324,36 +324,52 @@ int anon_vma_fork(struct vm_area_struct
 	return -ENOMEM;
 }
 
-static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
+static int 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 (list_empty(&anon_vma->head))
+		return 1;
 
-	if (empty)
-		put_anon_vma(anon_vma);
+	return 0;
 }
 
 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) {
+			root = lock_anon_vma_root(root, anon_vma);
+			/* Leave empty anon_vmas on the list. */
+			if (anon_vma_unlink(avc, anon_vma))
+				continue;
+		}
+		list_del(&avc->same_vma);
+		anon_vma_chain_free(avc);
+	}
+	unlock_anon_vma_root(root);
+
+	/*
+	 * Iterate the list once more, it now only contains empty and unlinked
+	 * anon_vmas, destroy them. Could not do before due to anon_vma_free()
+	 * needing to acquire the anon_vma->root->mutex.
+	 */
+	list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
+		struct anon_vma *anon_vma = avc->anon_vma;
+
+		if (anon_vma)
+			put_anon_vma(anon_vma);
+
 		list_del(&avc->same_vma);
 		anon_vma_chain_free(avc);
 	}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>