[PATCH 0/5] mm: i_mmap_mutex to rwsem

[PATCH 0/5] mm: i_mmap_mutex to rwsem

[Davidlohr Bueso]

This patchset extends the work started by Ingo Molnar in late 2012,
optimizing the anon-vma mutex lock, converting it from a exclusive mutex
to a rwsem, and sharing the lock for read-only paths when walking the
the vma-interval tree. More specifically commits 5a505085 and 4fc3f1d6.

The i_mmap mutex has similar responsibilities with the anon-vma, protecting
file backed pages. Therefore we can use similar locking techniques: covert
the mutex to a rwsem and share the lock when possible.

With these changes, and the rwsem optimizations discussed in
http://lkml.org/lkml/2013/6/16/38 we can see performance improvements.
For instance, on a 8 socket, 80 core DL980, when compared to a vanilla 3.10-rc5, 
aim7 benefits in throughput, with the following workloads (beyond 500 users):

- alltests (+14.5%)
- custom (+17%)
- disk (+11%)
- high_systime (+5%)
- shared (+15%)
- short (+4%)

For lower amounts of users, there are no significant differences as all numbers
are within the 0-2% noise range.

Davidlohr Bueso (5):
  mm,fs: introduce helpers around i_mmap_mutex
  mm: use new helper functions around the i_mmap_mutex
  mm: convert i_mmap_mutex to rwsem
  mm/rmap: share the i_mmap_rwsem
  mm: rename leftover i_mmap_mutex

 Documentation/lockstat.txt   |  2 +-
 Documentation/vm/locking     |  2 +-
 arch/x86/mm/hugetlbpage.c    |  6 +++---
 fs/hugetlbfs/inode.c         |  4 ++--
 fs/inode.c                   |  2 +-
 include/linux/fs.h           | 22 +++++++++++++++++++++-
 include/linux/mmu_notifier.h |  2 +-
 kernel/events/uprobes.c      |  6 +++---
 kernel/fork.c                |  4 ++--
 mm/filemap.c                 | 10 +++++-----
 mm/filemap_xip.c             |  4 ++--
 mm/fremap.c                  |  4 ++--
 mm/hugetlb.c                 | 16 ++++++++--------
 mm/memory-failure.c          |  7 +++----
 mm/memory.c                  |  8 ++++----
 mm/mmap.c                    | 22 +++++++++++-----------
 mm/mremap.c                  |  6 +++---
 mm/nommu.c                   | 14 +++++++-------
 mm/rmap.c                    | 24 ++++++++++++------------
 19 files changed, 92 insertions(+), 73 deletions(-)

-- 
1.7.11.7

[PATCH 1/5] mm,fs: introduce helpers around i_mmap_mutex

[Davidlohr Bueso]

Various parts of the kernel acquire and release this mutex,
so add i_mmap_lock_write() and immap_unlock_write() helper
functions that will encapsulate this logic. The next patch
will make use of these.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
---
 include/linux/fs.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 65c2be2..1ea6c68 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -475,6 +475,16 @@ struct block_device {
 
 int mapping_tagged(struct address_space *mapping, int tag);
 
+static inline void i_mmap_lock_write(struct address_space *mapping)
+{
+	mutex_lock(&mapping->i_mmap_mutex);
+}
+
+static inline void i_mmap_unlock_write(struct address_space *mapping)
+{
+	mutex_unlock(&mapping->i_mmap_mutex);
+}
+
 /*
  * Might pages of this file be mapped into userspace?
  */
-- 
1.7.11.7

[PATCH 2/5] mm: use new helper functions around the i_mmap_mutex

[Davidlohr Bueso]

Convert all open coded mutex_lock/unlock calls to the
i_mmap_[lock/unlock]_write() helpers.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
---
 arch/x86/mm/hugetlbpage.c |  4 ++--
 fs/hugetlbfs/inode.c      |  4 ++--
 kernel/events/uprobes.c   |  4 ++--
 kernel/fork.c             |  4 ++--
 mm/filemap_xip.c          |  4 ++--
 mm/fremap.c               |  4 ++--
 mm/hugetlb.c              |  8 ++++----
 mm/memory-failure.c       |  4 ++--
 mm/memory.c               |  8 ++++----
 mm/mmap.c                 | 14 +++++++-------
 mm/mremap.c               |  4 ++--
 mm/nommu.c                | 14 +++++++-------
 mm/rmap.c                 | 16 ++++++++--------
 13 files changed, 46 insertions(+), 46 deletions(-)

diff --git a/arch/x86/mm/hugetlbpage.c b/arch/x86/mm/hugetlbpage.c
index ae1aa71..9c61a1e 100644
--- a/arch/x86/mm/hugetlbpage.c
+++ b/arch/x86/mm/hugetlbpage.c
@@ -79,7 +79,7 @@ huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud)
 	if (!vma_shareable(vma, addr))
 		return (pte_t *)pmd_alloc(mm, pud, addr);
 
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(svma, &mapping->i_mmap, idx, idx) {
 		if (svma == vma)
 			continue;
@@ -105,7 +105,7 @@ huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud)
 	spin_unlock(&mm->page_table_lock);
 out:
 	pte = (pte_t *)pmd_alloc(mm, pud, addr);
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 	return pte;
 }
 
diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index a3f868a..35eebfc 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -404,10 +404,10 @@ static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 	pgoff = offset >> PAGE_SHIFT;
 
 	i_size_write(inode, offset);
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	if (!RB_EMPTY_ROOT(&mapping->i_mmap))
 		hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 	truncate_hugepages(inode, offset);
 	return 0;
 }
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index f356974..c7b9f45 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -693,7 +693,7 @@ build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
 	int more = 0;
 
  again:
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		if (!valid_vma(vma, is_register))
 			continue;
@@ -724,7 +724,7 @@ build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
 		info->mm = vma->vm_mm;
 		info->vaddr = offset_to_vaddr(vma, offset);
 	}
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 
 	if (!more)
 		goto out;
diff --git a/kernel/fork.c b/kernel/fork.c
index 987b28a..13226f1 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -420,7 +420,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 			get_file(file);
 			if (tmp->vm_flags & VM_DENYWRITE)
 				atomic_dec(&inode->i_writecount);
-			mutex_lock(&mapping->i_mmap_mutex);
+			i_mmap_lock_write(mapping);
 			if (tmp->vm_flags & VM_SHARED)
 				mapping->i_mmap_writable++;
 			flush_dcache_mmap_lock(mapping);
@@ -432,7 +432,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 				vma_interval_tree_insert_after(tmp, mpnt,
 							&mapping->i_mmap);
 			flush_dcache_mmap_unlock(mapping);
-			mutex_unlock(&mapping->i_mmap_mutex);
+			i_mmap_unlock_write(mapping);
 		}
 
 		/*
diff --git a/mm/filemap_xip.c b/mm/filemap_xip.c
index 28fe26b..c851586 100644
--- a/mm/filemap_xip.c
+++ b/mm/filemap_xip.c
@@ -182,7 +182,7 @@ __xip_unmap (struct address_space * mapping,
 		return;
 
 retry:
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		mm = vma->vm_mm;
 		address = vma->vm_start +
@@ -202,7 +202,7 @@ retry:
 			page_cache_release(page);
 		}
 	}
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 
 	if (locked) {
 		mutex_unlock(&xip_sparse_mutex);
diff --git a/mm/fremap.c b/mm/fremap.c
index 87da359..fa49f3d 100644
--- a/mm/fremap.c
+++ b/mm/fremap.c
@@ -215,13 +215,13 @@ get_write_lock:
 			}
 			goto out;
 		}
-		mutex_lock(&mapping->i_mmap_mutex);
+		i_mmap_lock_write(mapping);
 		flush_dcache_mmap_lock(mapping);
 		vma->vm_flags |= VM_NONLINEAR;
 		vma_interval_tree_remove(vma, &mapping->i_mmap);
 		vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 		flush_dcache_mmap_unlock(mapping);
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 	}
 
 	if (vma->vm_flags & VM_LOCKED) {
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index e2bfbf7..1becd0a 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2506,7 +2506,7 @@ static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma,
 	 * this mapping should be shared between all the VMAs,
 	 * __unmap_hugepage_range() is called as the lock is already held
 	 */
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(iter_vma, &mapping->i_mmap, pgoff, pgoff) {
 		/* Do not unmap the current VMA */
 		if (iter_vma == vma)
@@ -2523,7 +2523,7 @@ static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma,
 			unmap_hugepage_range(iter_vma, address,
 					     address + huge_page_size(h), page);
 	}
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 
 	return 1;
 }
@@ -3047,7 +3047,7 @@ unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 	BUG_ON(address >= end);
 	flush_cache_range(vma, address, end);
 
-	mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);
+	i_mmap_lock_write(vma->vm_file->f_mapping);
 	spin_lock(&mm->page_table_lock);
 	for (; address < end; address += huge_page_size(h)) {
 		ptep = huge_pte_offset(mm, address);
@@ -3073,7 +3073,7 @@ unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 	 * and that page table be reused and filled with junk.
 	 */
 	flush_tlb_range(vma, start, end);
-	mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);
+	i_mmap_unlock_write(vma->vm_file->f_mapping);
 
 	return pages << h->order;
 }
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index ceb0c7f..e7e0f90 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -436,7 +436,7 @@ static void collect_procs_file(struct page *page, struct list_head *to_kill,
 	struct task_struct *tsk;
 	struct address_space *mapping = page->mapping;
 
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	read_lock(&tasklist_lock);
 	for_each_process(tsk) {
 		pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
@@ -458,7 +458,7 @@ static void collect_procs_file(struct page *page, struct list_head *to_kill,
 		}
 	}
 	read_unlock(&tasklist_lock);
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 }
 
 /*
diff --git a/mm/memory.c b/mm/memory.c
index 61a262b..344b4cf 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1340,9 +1340,9 @@ static void unmap_single_vma(struct mmu_gather *tlb,
 			 * safe to do nothing in this case.
 			 */
 			if (vma->vm_file) {
-				mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);
+				i_mmap_lock_write(vma->vm_file->f_mapping);
 				__unmap_hugepage_range_final(tlb, vma, start, end, NULL);
-				mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);
+				i_mmap_unlock_write(vma->vm_file->f_mapping);
 			}
 		} else
 			unmap_page_range(tlb, vma, start, end, details);
@@ -2974,12 +2974,12 @@ void unmap_mapping_range(struct address_space *mapping,
 		details.last_index = ULONG_MAX;
 
 
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
 		unmap_mapping_range_tree(&mapping->i_mmap, &details);
 	if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
 		unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 }
 EXPORT_SYMBOL(unmap_mapping_range);
 
diff --git a/mm/mmap.c b/mm/mmap.c
index f681e18..01a9876 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -233,9 +233,9 @@ void unlink_file_vma(struct vm_area_struct *vma)
 
 	if (file) {
 		struct address_space *mapping = file->f_mapping;
-		mutex_lock(&mapping->i_mmap_mutex);
+		i_mmap_lock_write(mapping);
 		__remove_shared_vm_struct(vma, file, mapping);
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 	}
 }
 
@@ -644,13 +644,13 @@ static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 		mapping = vma->vm_file->f_mapping;
 
 	if (mapping)
-		mutex_lock(&mapping->i_mmap_mutex);
+		i_mmap_lock_write(mapping);
 
 	__vma_link(mm, vma, prev, rb_link, rb_parent);
 	__vma_link_file(vma);
 
 	if (mapping)
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 
 	mm->map_count++;
 	validate_mm(mm);
@@ -761,7 +761,7 @@ again:			remove_next = 1 + (end > next->vm_end);
 							next->vm_end);
 		}
 
-		mutex_lock(&mapping->i_mmap_mutex);
+		i_mmap_lock_write(mapping);
 		if (insert) {
 			/*
 			 * Put into interval tree now, so instantiated pages
@@ -848,7 +848,7 @@ again:			remove_next = 1 + (end > next->vm_end);
 		anon_vma_unlock_write(anon_vma);
 	}
 	if (mapping)
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 
 	if (root) {
 		uprobe_mmap(vma);
@@ -3112,7 +3112,7 @@ static void vm_unlock_mapping(struct address_space *mapping)
 		 * AS_MM_ALL_LOCKS can't change to 0 from under us
 		 * because we hold the mm_all_locks_mutex.
 		 */
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 		if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
 					&mapping->flags))
 			BUG();
diff --git a/mm/mremap.c b/mm/mremap.c
index 463a257..02fc5df 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -101,7 +101,7 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
 	if (need_rmap_locks) {
 		if (vma->vm_file) {
 			mapping = vma->vm_file->f_mapping;
-			mutex_lock(&mapping->i_mmap_mutex);
+			i_mmap_lock_write(mapping);
 		}
 		if (vma->anon_vma) {
 			anon_vma = vma->anon_vma;
@@ -137,7 +137,7 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
 	if (anon_vma)
 		anon_vma_unlock_write(anon_vma);
 	if (mapping)
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 }
 
 #define LATENCY_LIMIT	(64 * PAGE_SIZE)
diff --git a/mm/nommu.c b/mm/nommu.c
index 298884d..5a11e45 100644
--- a/mm/nommu.c
+++ b/mm/nommu.c
@@ -714,11 +714,11 @@ static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 	if (vma->vm_file) {
 		mapping = vma->vm_file->f_mapping;
 
-		mutex_lock(&mapping->i_mmap_mutex);
+		i_mmap_lock_write(mapping);
 		flush_dcache_mmap_lock(mapping);
 		vma_interval_tree_insert(vma, &mapping->i_mmap);
 		flush_dcache_mmap_unlock(mapping);
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 	}
 
 	/* add the VMA to the tree */
@@ -780,11 +780,11 @@ static void delete_vma_from_mm(struct vm_area_struct *vma)
 	if (vma->vm_file) {
 		mapping = vma->vm_file->f_mapping;
 
-		mutex_lock(&mapping->i_mmap_mutex);
+		i_mmap_lock_write(mapping);
 		flush_dcache_mmap_lock(mapping);
 		vma_interval_tree_remove(vma, &mapping->i_mmap);
 		flush_dcache_mmap_unlock(mapping);
-		mutex_unlock(&mapping->i_mmap_mutex);
+		i_mmap_unlock_write(mapping);
 	}
 
 	/* remove from the MM's tree and list */
@@ -2087,14 +2087,14 @@ int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
 	high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 
 	down_write(&nommu_region_sem);
-	mutex_lock(&inode->i_mapping->i_mmap_mutex);
+	i_mmap_lock_write(inode->i_mapping);
 
 	/* search for VMAs that fall within the dead zone */
 	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
 		/* found one - only interested if it's shared out of the page
 		 * cache */
 		if (vma->vm_flags & VM_SHARED) {
-			mutex_unlock(&inode->i_mapping->i_mmap_mutex);
+			i_mmap_unlock_write(inode->i_mapping);
 			up_write(&nommu_region_sem);
 			return -ETXTBSY; /* not quite true, but near enough */
 		}
@@ -2122,7 +2122,7 @@ int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
 		}
 	}
 
-	mutex_unlock(&inode->i_mapping->i_mmap_mutex);
+	i_mmap_unlock_write(inode->i_mapping);
 	up_write(&nommu_region_sem);
 	return 0;
 }
diff --git a/mm/rmap.c b/mm/rmap.c
index 6280da8..bc8eeb5 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -808,7 +808,7 @@ static int page_referenced_file(struct page *page,
 	 */
 	BUG_ON(!PageLocked(page));
 
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 
 	/*
 	 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
@@ -831,7 +831,7 @@ static int page_referenced_file(struct page *page,
 			break;
 	}
 
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 	return referenced;
 }
 
@@ -920,14 +920,14 @@ static int page_mkclean_file(struct address_space *mapping, struct page *page)
 
 	BUG_ON(PageAnon(page));
 
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		if (vma->vm_flags & VM_SHARED) {
 			unsigned long address = vma_address(page, vma);
 			ret += page_mkclean_one(page, vma, address);
 		}
 	}
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 	return ret;
 }
 
@@ -1516,7 +1516,7 @@ static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
 	if (PageHuge(page))
 		pgoff = page->index << compound_order(page);
 
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		unsigned long address = vma_address(page, vma);
 		ret = try_to_unmap_one(page, vma, address, flags);
@@ -1594,7 +1594,7 @@ static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
 	list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
 		vma->vm_private_data = NULL;
 out:
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 	return ret;
 }
 
@@ -1711,7 +1711,7 @@ static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
 
 	if (!mapping)
 		return ret;
-	mutex_lock(&mapping->i_mmap_mutex);
+	i_mmap_lock_write(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		unsigned long address = vma_address(page, vma);
 		ret = rmap_one(page, vma, address, arg);
@@ -1723,7 +1723,7 @@ static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
 	 * never contain migration ptes.  Decide what to do about this
 	 * limitation to linear when we need rmap_walk() on nonlinear.
 	 */
-	mutex_unlock(&mapping->i_mmap_mutex);
+	i_mmap_unlock_write(mapping);
 	return ret;
 }
 
-- 
1.7.11.7

[PATCH 3/5] mm: convert i_mmap_mutex to rwsem

[Davidlohr Bueso]

This conversion is straightforward. All users take the write
lock, so there is really not much difference with the previous
mutex lock.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
---
 fs/inode.c         | 2 +-
 include/linux/fs.h | 6 +++---
 mm/mmap.c          | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/fs/inode.c b/fs/inode.c
index 00d5fc3..af5f0ea 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -345,7 +345,7 @@ void address_space_init_once(struct address_space *mapping)
 	memset(mapping, 0, sizeof(*mapping));
 	INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
 	spin_lock_init(&mapping->tree_lock);
-	mutex_init(&mapping->i_mmap_mutex);
+	init_rwsem(&mapping->i_mmap_rwsem);
 	INIT_LIST_HEAD(&mapping->private_list);
 	spin_lock_init(&mapping->private_lock);
 	mapping->i_mmap = RB_ROOT;
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 1ea6c68..79b8548 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -410,7 +410,7 @@ struct address_space {
 	unsigned int		i_mmap_writable;/* count VM_SHARED mappings */
 	struct rb_root		i_mmap;		/* tree of private and shared mappings */
 	struct list_head	i_mmap_nonlinear;/*list VM_NONLINEAR mappings */
-	struct mutex		i_mmap_mutex;	/* protect tree, count, list */
+	struct rw_semaphore     i_mmap_rwsem;	/* protect tree, count, list */
 	/* Protected by tree_lock together with the radix tree */
 	unsigned long		nrpages;	/* number of total pages */
 	pgoff_t			writeback_index;/* writeback starts here */
@@ -477,12 +477,12 @@ int mapping_tagged(struct address_space *mapping, int tag);
 
 static inline void i_mmap_lock_write(struct address_space *mapping)
 {
-	mutex_lock(&mapping->i_mmap_mutex);
+	down_write(&mapping->i_mmap_rwsem);
 }
 
 static inline void i_mmap_unlock_write(struct address_space *mapping)
 {
-	mutex_unlock(&mapping->i_mmap_mutex);
+	up_write(&mapping->i_mmap_rwsem);
 }
 
 /*
diff --git a/mm/mmap.c b/mm/mmap.c
index 01a9876..b4e142a 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -3016,7 +3016,7 @@ static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
 		 */
 		if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
 			BUG();
-		mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
+		down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
 	}
 }
 
-- 
1.7.11.7

[PATCH 4/5] mm/rmap: share the i_mmap_rwsem

[Davidlohr Bueso]

Similar to commit 4fc3f1d6, which optimized the anon-vma rwsem, we can share
the i_mmap_rwsem among multiple readers for rmap_walk_file(),
try_to_unmap_file() and collect_procs_file().

With this change, and the rwsem optimizations discussed in
http://lkml.org/lkml/2013/6/16/38 we can see performance improvements.
On a 8 socket, 80 core DL980, when compared to a vanilla 3.10-rc5, aim7
benefits in throughput, with the following workloads (beyond 500 users):

- alltests (+14.5%)
- custom (+17%)
- disk (+11%)
- high_systime (+5%)
- shared (+15%)
- short (+4%)

For lower amounts of users, there are no significant differences as all numbers
are within the 0-2% noise range.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
---
 include/linux/fs.h  | 10 ++++++++++
 mm/memory-failure.c |  7 +++----
 mm/rmap.c           | 12 ++++++------
 3 files changed, 19 insertions(+), 10 deletions(-)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 79b8548..5646641 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -485,6 +485,16 @@ static inline void i_mmap_unlock_write(struct address_space *mapping)
 	up_write(&mapping->i_mmap_rwsem);
 }
 
+static inline void i_mmap_lock_read(struct address_space *mapping)
+{
+	down_read(&mapping->i_mmap_rwsem);
+}
+
+static inline void i_mmap_unlock_read(struct address_space *mapping)
+{
+	up_read(&mapping->i_mmap_rwsem);
+}
+
 /*
  * Might pages of this file be mapped into userspace?
  */
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index e7e0f90..6db44eb 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -436,7 +436,7 @@ static void collect_procs_file(struct page *page, struct list_head *to_kill,
 	struct task_struct *tsk;
 	struct address_space *mapping = page->mapping;
 
-	i_mmap_lock_write(mapping);
+	i_mmap_lock_read(mapping);
 	read_lock(&tasklist_lock);
 	for_each_process(tsk) {
 		pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
@@ -444,8 +444,7 @@ static void collect_procs_file(struct page *page, struct list_head *to_kill,
 		if (!task_early_kill(tsk))
 			continue;
 
-		vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff,
-				      pgoff) {
+		vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 			/*
 			 * Send early kill signal to tasks where a vma covers
 			 * the page but the corrupted page is not necessarily
@@ -458,7 +457,7 @@ static void collect_procs_file(struct page *page, struct list_head *to_kill,
 		}
 	}
 	read_unlock(&tasklist_lock);
-	i_mmap_unlock_write(mapping);
+	i_mmap_unlock_read(mapping);
 }
 
 /*
diff --git a/mm/rmap.c b/mm/rmap.c
index bc8eeb5..98b986d 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -808,7 +808,7 @@ static int page_referenced_file(struct page *page,
 	 */
 	BUG_ON(!PageLocked(page));
 
-	i_mmap_lock_write(mapping);
+	i_mmap_lock_read(mapping);
 
 	/*
 	 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
@@ -831,7 +831,7 @@ static int page_referenced_file(struct page *page,
 			break;
 	}
 
-	i_mmap_unlock_write(mapping);
+	i_mmap_unlock_read(mapping);
 	return referenced;
 }
 
@@ -1516,7 +1516,7 @@ static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
 	if (PageHuge(page))
 		pgoff = page->index << compound_order(page);
 
-	i_mmap_lock_write(mapping);
+	i_mmap_lock_read(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		unsigned long address = vma_address(page, vma);
 		ret = try_to_unmap_one(page, vma, address, flags);
@@ -1594,7 +1594,7 @@ static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
 	list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
 		vma->vm_private_data = NULL;
 out:
-	i_mmap_unlock_write(mapping);
+	i_mmap_unlock_read(mapping);
 	return ret;
 }
 
@@ -1711,7 +1711,7 @@ static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
 
 	if (!mapping)
 		return ret;
-	i_mmap_lock_write(mapping);
+	i_mmap_lock_read(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		unsigned long address = vma_address(page, vma);
 		ret = rmap_one(page, vma, address, arg);
@@ -1723,7 +1723,7 @@ static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
 	 * never contain migration ptes.  Decide what to do about this
 	 * limitation to linear when we need rmap_walk() on nonlinear.
 	 */
-	i_mmap_unlock_write(mapping);
+	i_mmap_unlock_read(mapping);
 	return ret;
 }
 
-- 
1.7.11.7

[PATCH 5/5] mm: rename leftover i_mmap_mutex

[Davidlohr Bueso]

Update the lock to i_mmap_rwsem throughout the kernel.
All changes are in comments and documentation.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
---
 Documentation/lockstat.txt   |  2 +-
 Documentation/vm/locking     |  2 +-
 arch/x86/mm/hugetlbpage.c    |  2 +-
 include/linux/mmu_notifier.h |  2 +-
 kernel/events/uprobes.c      |  2 +-
 mm/filemap.c                 | 10 +++++-----
 mm/hugetlb.c                 |  8 ++++----
 mm/mmap.c                    |  6 +++---
 mm/mremap.c                  |  2 +-
 mm/rmap.c                    |  8 ++++----
 10 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/Documentation/lockstat.txt b/Documentation/lockstat.txt
index dd2f7b2..96b8233 100644
--- a/Documentation/lockstat.txt
+++ b/Documentation/lockstat.txt
@@ -168,7 +168,7 @@ View the top contending locks:
                              dcache_lock:          1037           1161           0.38          45.32         774.51           6611         243371           0.15         306.48       77387.24
                          &inode->i_mutex:           161            286 18446744073709       62882.54     1244614.55           3653          20598 18446744073709       62318.60     1693822.74
                          &zone->lru_lock:            94             94           0.53           7.33          92.10           4366          32690           0.29          59.81       16350.06
-              &inode->i_data.i_mmap_mutex:            79             79           0.40           3.77          53.03          11779          87755           0.28         116.93       29898.44
+              &inode->i_data.i_mmap_rwsem:           79             79           0.40           3.77          53.03          11779          87755           0.28         116.93       29898.44
                         &q->__queue_lock:            48             50           0.52          31.62          86.31            774          13131           0.17         113.08       12277.52
                         &rq->rq_lock_key:            43             47           0.74          68.50         170.63           3706          33929           0.22         107.99       17460.62
                       &rq->rq_lock_key#2:            39             46           0.75           6.68          49.03           2979          32292           0.17         125.17       17137.63
diff --git a/Documentation/vm/locking b/Documentation/vm/locking
index f61228b..fb64028 100644
--- a/Documentation/vm/locking
+++ b/Documentation/vm/locking
@@ -66,7 +66,7 @@ in some cases it is not really needed. Eg, vm_start is modified by
 expand_stack(), it is hard to come up with a destructive scenario without 
 having the vmlist protection in this case.
 
-The page_table_lock nests with the inode i_mmap_mutex and the kmem cache
+The page_table_lock nests with the inode i_mmap_rwsem and the kmem cache
 c_spinlock spinlocks.  This is okay, since the kmem code asks for pages after
 dropping c_spinlock.  The page_table_lock also nests with pagecache_lock and
 pagemap_lru_lock spinlocks, and no code asks for memory with these locks
diff --git a/arch/x86/mm/hugetlbpage.c b/arch/x86/mm/hugetlbpage.c
index 9c61a1e..df68d13 100644
--- a/arch/x86/mm/hugetlbpage.c
+++ b/arch/x86/mm/hugetlbpage.c
@@ -60,7 +60,7 @@ static int vma_shareable(struct vm_area_struct *vma, unsigned long addr)
  * and returns the corresponding pte. While this is not necessary for the
  * !shared pmd case because we can allocate the pmd later as well, it makes the
  * code much cleaner. pmd allocation is essential for the shared case because
- * pud has to be populated inside the same i_mmap_mutex section - otherwise
+ * pud has to be populated inside the same i_mmap_rwsem section - otherwise
  * racing tasks could either miss the sharing (see huge_pte_offset) or select a
  * bad pmd for sharing.
  */
diff --git a/include/linux/mmu_notifier.h b/include/linux/mmu_notifier.h
index deca874..f9c11ab 100644
--- a/include/linux/mmu_notifier.h
+++ b/include/linux/mmu_notifier.h
@@ -151,7 +151,7 @@ struct mmu_notifier_ops {
  * Therefore notifier chains can only be traversed when either
  *
  * 1. mmap_sem is held.
- * 2. One of the reverse map locks is held (i_mmap_mutex or anon_vma->rwsem).
+ * 2. One of the reverse map locks is held (i_mmap_rwsem or anon_vma->rwsem).
  * 3. No other concurrent thread can access the list (release)
  */
 struct mmu_notifier {
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index c7b9f45..4ca146e 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -700,7 +700,7 @@ build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
 
 		if (!prev && !more) {
 			/*
-			 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
+			 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
 			 * reclaim. This is optimistic, no harm done if it fails.
 			 */
 			prev = kmalloc(sizeof(struct map_info),
diff --git a/mm/filemap.c b/mm/filemap.c
index 7905fe7..5d3ae93 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -60,16 +60,16 @@
 /*
  * Lock ordering:
  *
- *  ->i_mmap_mutex		(truncate_pagecache)
+ *  ->i_mmap_rwsem		(truncate_pagecache)
  *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  *      ->swap_lock		(exclusive_swap_page, others)
  *        ->mapping->tree_lock
  *
  *  ->i_mutex
- *    ->i_mmap_mutex		(truncate->unmap_mapping_range)
+ *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  *
  *  ->mmap_sem
- *    ->i_mmap_mutex
+ *    ->i_mmap_rwsem
  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  *        ->mapping->tree_lock	(arch-dependent flush_dcache_mmap_lock)
  *
@@ -83,7 +83,7 @@
  *    sb_lock			(fs/fs-writeback.c)
  *    ->mapping->tree_lock	(__sync_single_inode)
  *
- *  ->i_mmap_mutex
+ *  ->i_mmap_rwsem
  *    ->anon_vma.lock		(vma_adjust)
  *
  *  ->anon_vma.lock
@@ -103,7 +103,7 @@
  *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
  *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
  *
- * ->i_mmap_mutex
+ * ->i_mmap_rwsem
  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
  */
 
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 1becd0a..041f58f 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2458,9 +2458,9 @@ void __unmap_hugepage_range_final(struct mmu_gather *tlb,
 	 * on its way out.  We're lucky that the flag has such an appropriate
 	 * name, and can in fact be safely cleared here. We could clear it
 	 * before the __unmap_hugepage_range above, but all that's necessary
-	 * is to clear it before releasing the i_mmap_mutex. This works
+	 * is to clear it before releasing the i_mmap_rwsem. This works
 	 * because in the context this is called, the VMA is about to be
-	 * destroyed and the i_mmap_mutex is held.
+	 * destroyed and the i_mmap_rwsem is held.
 	 */
 	vma->vm_flags &= ~VM_MAYSHARE;
 }
@@ -3067,9 +3067,9 @@ unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 	}
 	spin_unlock(&mm->page_table_lock);
 	/*
-	 * Must flush TLB before releasing i_mmap_mutex: x86's huge_pmd_unshare
+	 * Must flush TLB before releasing i_mmap_rwsem: x86's huge_pmd_unshare
 	 * may have cleared our pud entry and done put_page on the page table:
-	 * once we release i_mmap_mutex, another task can do the final put_page
+	 * once we release i_mmap_rwsem, another task can do the final put_page
 	 * and that page table be reused and filled with junk.
 	 */
 	flush_tlb_range(vma, start, end);
diff --git a/mm/mmap.c b/mm/mmap.c
index b4e142a..b4ca52a 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -205,7 +205,7 @@ error:
 }
 
 /*
- * Requires inode->i_mapping->i_mmap_mutex
+ * Requires inode->i_mapping->i_mmap_rwsem
  */
 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 		struct file *file, struct address_space *mapping)
@@ -2759,7 +2759,7 @@ void exit_mmap(struct mm_struct *mm)
 
 /* Insert vm structure into process list sorted by address
  * and into the inode's i_mmap tree.  If vm_file is non-NULL
- * then i_mmap_mutex is taken here.
+ * then i_mmap_rwsem is taken here.
  */
 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
 {
@@ -3043,7 +3043,7 @@ static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
  * vma in this mm is backed by the same anon_vma or address_space.
  *
  * We can take all the locks in random order because the VM code
- * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
+ * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
  * takes more than one of them in a row. Secondly we're protected
  * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
  *
diff --git a/mm/mremap.c b/mm/mremap.c
index 02fc5df..742fe28 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -81,7 +81,7 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
 	spinlock_t *old_ptl, *new_ptl;
 
 	/*
-	 * When need_rmap_locks is true, we take the i_mmap_mutex and anon_vma
+	 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
 	 * locks to ensure that rmap will always observe either the old or the
 	 * new ptes. This is the easiest way to avoid races with
 	 * truncate_pagecache(), page migration, etc...
diff --git a/mm/rmap.c b/mm/rmap.c
index 98b986d..1bfde51 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -23,7 +23,7 @@
  * inode->i_mutex	(while writing or truncating, not reading or faulting)
  *   mm->mmap_sem
  *     page->flags PG_locked (lock_page)
- *       mapping->i_mmap_mutex
+ *       mapping->i_mmap_rwsem
  *         anon_vma->rwsem
  *           mm->page_table_lock or pte_lock
  *             zone->lru_lock (in mark_page_accessed, isolate_lru_page)
@@ -804,14 +804,14 @@ static int page_referenced_file(struct page *page,
 	 * The page lock not only makes sure that page->mapping cannot
 	 * suddenly be NULLified by truncation, it makes sure that the
 	 * structure at mapping cannot be freed and reused yet,
-	 * so we can safely take mapping->i_mmap_mutex.
+	 * so we can safely take mapping->i_mmap_rwsem.
 	 */
 	BUG_ON(!PageLocked(page));
 
 	i_mmap_lock_read(mapping);
 
 	/*
-	 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
+	 * i_mmap_rwsem does not stabilize mapcount at all, but mapcount
 	 * is more likely to be accurate if we note it after spinning.
 	 */
 	mapcount = page_mapcount(page);
@@ -1291,7 +1291,7 @@ out_mlock:
 	/*
 	 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
 	 * unstable result and race. Plus, We can't wait here because
-	 * we now hold anon_vma->rwsem or mapping->i_mmap_mutex.
+	 * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem.
 	 * if trylock failed, the page remain in evictable lru and later
 	 * vmscan could retry to move the page to unevictable lru if the
 	 * page is actually mlocked.
-- 
1.7.11.7

Re: [PATCH 1/5] mm,fs: introduce helpers around i_mmap_mutex

[Rik van Riel]

On 05/22/2014 11:33 PM, Davidlohr Bueso wrote:
> Various parts of the kernel acquire and release this mutex,
> so add i_mmap_lock_write() and immap_unlock_write() helper
> functions that will encapsulate this logic. The next patch
> will make use of these.
> 
> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>

Reviewed-by: Rik van Riel <riel@redhat.com>

-- 
All rights reversed

[PATCH 1/5] mm,fs: introduce helpers around i_mmap_mutex

[Davidlohr Bueso]

Various parts of the kernel acquire and release this mutex,
so add i_mmap_lock_write() and immap_unlock_write() helper
functions that will encapsulate this logic. The next patch
will make use of these.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
---
 include/linux/fs.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 1cab2f8..524d2c1 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -456,6 +456,16 @@ struct block_device {
 
 int mapping_tagged(struct address_space *mapping, int tag);
 
+static inline void i_mmap_lock_write(struct address_space *mapping)
+{
+	mutex_lock(&mapping->i_mmap_mutex);
+}
+
+static inline void i_mmap_unlock_write(struct address_space *mapping)
+{
+	mutex_unlock(&mapping->i_mmap_mutex);
+}
+
 /*
  * Might pages of this file be mapped into userspace?
  */
-- 
1.8.1.4