mm: change anon_vma linking to fix multi-process server scalability issue

The old anon_vma code can lead to scalability issues with heavily forking
workloads.  Specifically, each anon_vma will be shared between the parent
process and all its child processes.

In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes.  However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.

This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock.  This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands.  Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.

This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA.  At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated.  The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.

This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
 This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.

The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations.  This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures.  This in
turn means error handling needs to be added to the calling functions.

A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock.  To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag.  This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.

Some test results:

Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.

With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time.  The anon_vma lock contention appears to be resolved.

[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 130 insertions, 26 deletions

diff --git a/mm/rmap.c b/mm/rmap.c
index 5cb47111f79..be34094e459 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -62,6 +62,7 @@
 #include "internal.h"
 
 static struct kmem_cache *anon_vma_cachep;
+static struct kmem_cache *anon_vma_chain_cachep;
 
 static inline struct anon_vma *anon_vma_alloc(void)
 {
@@ -73,6 +74,16 @@ void anon_vma_free(struct anon_vma *anon_vma)
 	kmem_cache_free(anon_vma_cachep, anon_vma);
 }
 
+static inline struct anon_vma_chain *anon_vma_chain_alloc(void)
+{
+	return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL);
+}
+
+void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
+{
+	kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
+}
+
 /**
  * anon_vma_prepare - attach an anon_vma to a memory region
  * @vma: the memory region in question
@@ -103,18 +114,23 @@ void anon_vma_free(struct anon_vma *anon_vma)
 int anon_vma_prepare(struct vm_area_struct *vma)
 {
 	struct anon_vma *anon_vma = vma->anon_vma;
+	struct anon_vma_chain *avc;
 
 	might_sleep();
 	if (unlikely(!anon_vma)) {
 		struct mm_struct *mm = vma->vm_mm;
 		struct anon_vma *allocated;
 
+		avc = anon_vma_chain_alloc();
+		if (!avc)
+			goto out_enomem;
+
 		anon_vma = find_mergeable_anon_vma(vma);
 		allocated = NULL;
 		if (!anon_vma) {
 			anon_vma = anon_vma_alloc();
 			if (unlikely(!anon_vma))
-				return -ENOMEM;
+				goto out_enomem_free_avc;
 			allocated = anon_vma;
 		}
 		spin_lock(&anon_vma->lock);
@@ -123,53 +139,113 @@ int anon_vma_prepare(struct vm_area_struct *vma)
 		spin_lock(&mm->page_table_lock);
 		if (likely(!vma->anon_vma)) {
 			vma->anon_vma = anon_vma;
-			list_add_tail(&vma->anon_vma_node, &anon_vma->head);
+			avc->anon_vma = anon_vma;
+			avc->vma = vma;
+			list_add(&avc->same_vma, &vma->anon_vma_chain);
+			list_add(&avc->same_anon_vma, &anon_vma->head);
 			allocated = NULL;
 		}
 		spin_unlock(&mm->page_table_lock);
 
 		spin_unlock(&anon_vma->lock);
-		if (unlikely(allocated))
+		if (unlikely(allocated)) {
 			anon_vma_free(allocated);
+			anon_vma_chain_free(avc);
+		}
 	}
 	return 0;
+
+ out_enomem_free_avc:
+	anon_vma_chain_free(avc);
+ out_enomem:
+	return -ENOMEM;
 }
 
-void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
+static void anon_vma_chain_link(struct vm_area_struct *vma,
+				struct anon_vma_chain *avc,
+				struct anon_vma *anon_vma)
 {
-	BUG_ON(vma->anon_vma != next->anon_vma);
-	list_del(&next->anon_vma_node);
+	avc->vma = vma;
+	avc->anon_vma = anon_vma;
+	list_add(&avc->same_vma, &vma->anon_vma_chain);
+
+	spin_lock(&anon_vma->lock);
+	list_add_tail(&avc->same_anon_vma, &anon_vma->head);
+	spin_unlock(&anon_vma->lock);
 }
 
-void __anon_vma_link(struct vm_area_struct *vma)
+/*
+ * Attach the anon_vmas from src to dst.
+ * Returns 0 on success, -ENOMEM on failure.
+ */
+int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
 {
-	struct anon_vma *anon_vma = vma->anon_vma;
+	struct anon_vma_chain *avc, *pavc;
+
+	list_for_each_entry(pavc, &src->anon_vma_chain, same_vma) {
+		avc = anon_vma_chain_alloc();
+		if (!avc)
+			goto enomem_failure;
+		anon_vma_chain_link(dst, avc, pavc->anon_vma);
+	}
+	return 0;
 
-	if (anon_vma)
-		list_add_tail(&vma->anon_vma_node, &anon_vma->head);
+ enomem_failure:
+	unlink_anon_vmas(dst);
+	return -ENOMEM;
 }
 
-void anon_vma_link(struct vm_area_struct *vma)
+/*
+ * Attach vma to its own anon_vma, as well as to the anon_vmas that
+ * the corresponding VMA in the parent process is attached to.
+ * Returns 0 on success, non-zero on failure.
+ */
+int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
 {
-	struct anon_vma *anon_vma = vma->anon_vma;
+	struct anon_vma_chain *avc;
+	struct anon_vma *anon_vma;
 
-	if (anon_vma) {
-		spin_lock(&anon_vma->lock);
-		list_add_tail(&vma->anon_vma_node, &anon_vma->head);
-		spin_unlock(&anon_vma->lock);
-	}
+	/* Don't bother if the parent process has no anon_vma here. */
+	if (!pvma->anon_vma)
+		return 0;
+
+	/*
+	 * First, attach the new VMA to the parent VMA's anon_vmas,
+	 * so rmap can find non-COWed pages in child processes.
+	 */
+	if (anon_vma_clone(vma, pvma))
+		return -ENOMEM;
+
+	/* Then add our own anon_vma. */
+	anon_vma = anon_vma_alloc();
+	if (!anon_vma)
+		goto out_error;
+	avc = anon_vma_chain_alloc();
+	if (!avc)
+		goto out_error_free_anon_vma;
+	anon_vma_chain_link(vma, avc, anon_vma);
+	/* Mark this anon_vma as the one where our new (COWed) pages go. */
+	vma->anon_vma = anon_vma;
+
+	return 0;
+
+ out_error_free_anon_vma:
+	anon_vma_free(anon_vma);
+ out_error:
+	return -ENOMEM;
 }
 
-void anon_vma_unlink(struct vm_area_struct *vma)
+static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
 {
-	struct anon_vma *anon_vma = 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;
 
 	spin_lock(&anon_vma->lock);
-	list_del(&vma->anon_vma_node);
+	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) && !ksm_refcount(anon_vma);
@@ -179,6 +255,18 @@ void anon_vma_unlink(struct vm_area_struct *vma)
 		anon_vma_free(anon_vma);
 }
 
+void unlink_anon_vmas(struct vm_area_struct *vma)
+{
+	struct anon_vma_chain *avc, *next;
+
+	/* Unlink each anon_vma chained to the VMA. */
+	list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
+		anon_vma_unlink(avc);
+		list_del(&avc->same_vma);
+		anon_vma_chain_free(avc);
+	}
+}
+
 static void anon_vma_ctor(void *data)
 {
 	struct anon_vma *anon_vma = data;
@@ -192,6 +280,7 @@ void __init anon_vma_init(void)
 {
 	anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
 			0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
+	anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
 }
 
 /*
@@ -240,6 +329,18 @@ vma_address(struct page *page, struct vm_area_struct *vma)
 		/* page should be within @vma mapping range */
 		return -EFAULT;
 	}
+	if (unlikely(vma->vm_flags & VM_LOCK_RMAP)) {
+		/*
+		 * This VMA is being unlinked or is not yet linked into the
+		 * VMA tree.  Do not try to follow this rmap.  This race
+		 * condition can result in page_referenced() ignoring a
+		 * reference or in try_to_unmap() failing to unmap a page.
+		 * The VMA cannot be freed under us because we hold the
+		 * anon_vma->lock, which the munmap code takes while
+		 * unlinking the anon_vmas from the VMA.
+		 */
+		return -EFAULT;
+	}
 	return address;
 }
 
@@ -396,7 +497,7 @@ static int page_referenced_anon(struct page *page,
 {
 	unsigned int mapcount;
 	struct anon_vma *anon_vma;
-	struct vm_area_struct *vma;
+	struct anon_vma_chain *avc;
 	int referenced = 0;
 
 	anon_vma = page_lock_anon_vma(page);
@@ -404,7 +505,8 @@ static int page_referenced_anon(struct page *page,
 		return referenced;
 
 	mapcount = page_mapcount(page);
-	list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
+	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
+		struct vm_area_struct *vma = avc->vma;
 		unsigned long address = vma_address(page, vma);
 		if (address == -EFAULT)
 			continue;
@@ -1025,14 +1127,15 @@ static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
 static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
 {
 	struct anon_vma *anon_vma;
-	struct vm_area_struct *vma;
+	struct anon_vma_chain *avc;
 	int ret = SWAP_AGAIN;
 
 	anon_vma = page_lock_anon_vma(page);
 	if (!anon_vma)
 		return ret;
 
-	list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
+	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
+		struct vm_area_struct *vma = avc->vma;
 		unsigned long address = vma_address(page, vma);
 		if (address == -EFAULT)
 			continue;
@@ -1223,7 +1326,7 @@ static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
 		struct vm_area_struct *, unsigned long, void *), void *arg)
 {
 	struct anon_vma *anon_vma;
-	struct vm_area_struct *vma;
+	struct anon_vma_chain *avc;
 	int ret = SWAP_AGAIN;
 
 	/*
@@ -1238,7 +1341,8 @@ static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
 	if (!anon_vma)
 		return ret;
 	spin_lock(&anon_vma->lock);
-	list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
+	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
+		struct vm_area_struct *vma = avc->vma;
 		unsigned long address = vma_address(page, vma);
 		if (address == -EFAULT)
 			continue;