Re: [RFC PATCH 2/2] mm: Batch page_check_references in shrink_page_list sharing the same i_mmap_mutex

[Matthew Wilcox <willy@linux.intel.com>]

On Mon, Aug 20, 2012 at 09:43:02AM -0700, Tim Chen wrote:
> > In shrink_page_list, call to page_referenced_file will causes the
> > acquisition/release of mapping->i_mmap_mutex for each page in the page
> > list.  However, it is very likely that successive pages in the list
> > share the same mapping and we can reduce the frequency of i_mmap_mutex
> > acquisition by holding the mutex in shrink_page_list. This improves the
> > performance when the system has a lot page reclamations for file mapped
> > pages if workloads are using a lot of memory for page cache.

Is there a (performant) way to avoid passing around the
'mmap_mutex_locked' state?

For example, does it hurt to have all the callers hold the i_mmap_mutex()
over the entire call, or do we rely on being able to execute large chunks
of this in parallel?

Here's what I'm thinking:

1. Rename the existing page_referenced implementation to __page_referenced().
2. Add:

int needs_page_mmap_mutex(struct page *page)
{
	return page->mapping && page_mapped(page) && page_rmapping(page) &&
		!PageKsm(page) && !PageAnon(page);
}

int page_referenced(struct page *page, int is_locked, struct mem_cgroup *memcg,
						unsigned long *vm_flags)
{
	int result, needs_lock;

	needs_lock = needs_page_mmap_mutex(page);
	if (needs_lock)
		mutex_lock(&page->mapping->i_mmap_mutex);
	result = __page_referenced(page, is_locked, memcg, vm_flags);
	if (needs_lock)
		mutex_unlock(&page->mapping->i_mmap_mutex);
	return result;
}

3. Rename the existing try_to_unmap() to __try_to_unmap()
4. Add:

int try_to_unmap(struct page *page, enum ttu_flags flags)
{
	int result, needs_lock;
	
	needs_lock = needs_page_mmap_mutex(page);
	if (needs_lock)
		mutex_lock(&page->mapping->i_mmap_mutex);
	result = __try_to_unmap(page, is_locked, memcg, vm_flags);
	if (needs_lock)
		mutex_unlock(&page->mapping->i_mmap_mutex);
	return result;
}

5. Change page_check_references to always call __page_referenced (since it
now always holds the mutex)
6. Replace the mutex_lock() calls in page_referenced_file() and
try_to_unmap_file() with
	BUG_ON(!mutex_is_locked(&mapping->i_mmap_mutex));
7. I think you can simplify this:

> -		references = page_check_references(page, mz, sc);
> +		if (page->mapping) {
> +			if (i_mmap_mutex == &page->mapping->i_mmap_mutex) {
> +				mmap_mutex_locked = 1;
> +			} else {
> +				if (page_mapped(page) && page_rmapping(page)
> +					&& !PageKsm(page) && !PageAnon(page)) {
> +					if (i_mmap_mutex)
> +						mutex_unlock(i_mmap_mutex);
> +					i_mmap_mutex = &page->mapping->i_mmap_mutex;
> +					mutex_lock(i_mmap_mutex);
> +					mmap_mutex_locked = 1;
> +				}
> +			}
> +		}
> +
> +		references = page_check_references(page, mz, sc,
> +						   mmap_mutex_locked);

to just:

+		if (needs_page_mmap_mutex(page) &&
+				i_mmap_mutex != &page->mapping->i_mmap_mutex) {
+			if (i_mmap_mutex)
+				mutex_unlock(i_mmap_mutex);
+			i_mmap_mutex = &page->mapping->i_mmap_mutex;
+			mutex_lock(i_mmap_mutex);
+		}
	 	references = page_check_references(page, mz, sc);

The only clunky bit would seem to be this bit:

>  		if (page_mapped(page) && mapping) {
> -			switch (try_to_unmap(page, TTU_UNMAP)) {
> +			switch (try_to_unmap(page, TTU_UNMAP,
> +						mmap_mutex_locked)) {

Which I think has to look like this:

		if (page_mapped(page) && mapping) {
-			switch (try_to_unmap(page, TTU_UNMAP)) {
+			int result;
+			if (i_mmap_mutex)
+				result = __try_to_unmap(page, TTU_UNMAP);
+			else
+				result = try_to_unmap(page, TTU_UNMAP);
+			switch (result) {

Either that, or use a function pointer.  But ... bleugh.

> diff --git a/include/linux/rmap.h b/include/linux/rmap.h
> index fd07c45..f0174ae 100644
> --- a/include/linux/rmap.h
> +++ b/include/linux/rmap.h
> @@ -156,8 +156,8 @@ static inline void page_dup_rmap(struct page *page)
>  /*
>   * Called from mm/vmscan.c to handle paging out
>   */
> -int page_referenced(struct page *, int is_locked,
> -			struct mem_cgroup *memcg, unsigned long *vm_flags);
> +int page_referenced(struct page *, int is_locked, struct mem_cgroup *memcg,
> +				unsigned long *vm_flags, int mmap_mutex_locked);
>  int page_referenced_one(struct page *, struct vm_area_struct *,
>  	unsigned long address, unsigned int *mapcount, unsigned long *vm_flags);
>  
> @@ -175,7 +175,7 @@ enum ttu_flags {
>  
>  bool is_vma_temporary_stack(struct vm_area_struct *vma);
>  
> -int try_to_unmap(struct page *, enum ttu_flags flags);
> +int try_to_unmap(struct page *, enum ttu_flags flags, int mmap_mutex_locked);
>  int try_to_unmap_one(struct page *, struct vm_area_struct *,
>  			unsigned long address, enum ttu_flags flags);
>  
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 97cc273..5f162c7 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -953,7 +953,7 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
>  	if (hpage != ppage)
>  		lock_page(ppage);
>  
> -	ret = try_to_unmap(ppage, ttu);
> +	ret = try_to_unmap(ppage, ttu, 0);
>  	if (ret != SWAP_SUCCESS)
>  		printk(KERN_ERR "MCE %#lx: failed to unmap page (mapcount=%d)\n",
>  				pfn, page_mapcount(ppage));
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 1107238..1aa37b1 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -802,7 +802,7 @@ static int __unmap_and_move(struct page *page, struct page *newpage,
>  	}
>  
>  	/* Establish migration ptes or remove ptes */
> -	try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
> +	try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS, 0);
>  
>  skip_unmap:
>  	if (!page_mapped(page))
> @@ -918,7 +918,8 @@ static int unmap_and_move_huge_page(new_page_t get_new_page,
>  	if (PageAnon(hpage))
>  		anon_vma = page_get_anon_vma(hpage);
>  
> -	try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
> +	try_to_unmap(hpage,
> +		     TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS, 0);
>  
>  	if (!page_mapped(hpage))
>  		rc = move_to_new_page(new_hpage, hpage, 1, mode);
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 5b5ad58..63fc84b 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -820,7 +820,7 @@ static int page_referenced_anon(struct page *page,
>   */
>  static int page_referenced_file(struct page *page,
>  				struct mem_cgroup *memcg,
> -				unsigned long *vm_flags)
> +				unsigned long *vm_flags, int mmap_mutex_locked)
>  {
>  	unsigned int mapcount;
>  	struct address_space *mapping = page->mapping;
> @@ -844,7 +844,8 @@ static int page_referenced_file(struct page *page,
>  	 */
>  	BUG_ON(!PageLocked(page));
>  
> -	mutex_lock(&mapping->i_mmap_mutex);
> +	if (!mmap_mutex_locked)
> +		mutex_lock(&mapping->i_mmap_mutex);
>  
>  	/*
>  	 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
> @@ -869,7 +870,8 @@ static int page_referenced_file(struct page *page,
>  			break;
>  	}
>  
> -	mutex_unlock(&mapping->i_mmap_mutex);
> +	if (!mmap_mutex_locked)
> +		mutex_unlock(&mapping->i_mmap_mutex);
>  	return referenced;
>  }
>  
> @@ -886,7 +888,7 @@ static int page_referenced_file(struct page *page,
>  int page_referenced(struct page *page,
>  		    int is_locked,
>  		    struct mem_cgroup *memcg,
> -		    unsigned long *vm_flags)
> +		    unsigned long *vm_flags, int mmap_mutex_locked)
>  {
>  	int referenced = 0;
>  	int we_locked = 0;
> @@ -908,7 +910,7 @@ int page_referenced(struct page *page,
>  								vm_flags);
>  		else if (page->mapping)
>  			referenced += page_referenced_file(page, memcg,
> -								vm_flags);
> +						 vm_flags, mmap_mutex_locked);
>  		if (we_locked)
>  			unlock_page(page);
>  
> @@ -1548,7 +1550,8 @@ static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
>   * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
>   * 'LOCKED.
>   */
> -static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
> +static int try_to_unmap_file(struct page *page, enum ttu_flags flags,
> +				int mmap_mutex_locked)
>  {
>  	struct address_space *mapping = page->mapping;
>  	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
> @@ -1560,7 +1563,8 @@ static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
>  	unsigned long max_nl_size = 0;
>  	unsigned int mapcount;
>  
> -	mutex_lock(&mapping->i_mmap_mutex);
> +	if (!mmap_mutex_locked)
> +		mutex_lock(&mapping->i_mmap_mutex);
>  	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
>  		unsigned long address = vma_address(page, vma);
>  		if (address == -EFAULT)
> @@ -1640,7 +1644,8 @@ static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
>  	list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
>  		vma->vm_private_data = NULL;
>  out:
> -	mutex_unlock(&mapping->i_mmap_mutex);
> +	if (!mmap_mutex_locked)
> +		mutex_unlock(&mapping->i_mmap_mutex);
>  	return ret;
>  }
>  
> @@ -1658,7 +1663,7 @@ out:
>   * SWAP_FAIL	- the page is unswappable
>   * SWAP_MLOCK	- page is mlocked.
>   */
> -int try_to_unmap(struct page *page, enum ttu_flags flags)
> +int try_to_unmap(struct page *page, enum ttu_flags flags, int mmap_mutex_locked)
>  {
>  	int ret;
>  
> @@ -1670,7 +1675,7 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
>  	else if (PageAnon(page))
>  		ret = try_to_unmap_anon(page, flags);
>  	else
> -		ret = try_to_unmap_file(page, flags);
> +		ret = try_to_unmap_file(page, flags, mmap_mutex_locked);
>  	if (ret != SWAP_MLOCK && !page_mapped(page))
>  		ret = SWAP_SUCCESS;
>  	return ret;
> @@ -1700,7 +1705,7 @@ int try_to_munlock(struct page *page)
>  	else if (PageAnon(page))
>  		return try_to_unmap_anon(page, TTU_MUNLOCK);
>  	else
> -		return try_to_unmap_file(page, TTU_MUNLOCK);
> +		return try_to_unmap_file(page, TTU_MUNLOCK, 0);
>  }
>  
>  void __put_anon_vma(struct anon_vma *anon_vma)
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index a538565..1af1f1c 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -27,6 +27,7 @@
>  					buffer_heads_over_limit */
>  #include <linux/mm_inline.h>
>  #include <linux/backing-dev.h>
> +#include <linux/ksm.h>
>  #include <linux/rmap.h>
>  #include <linux/topology.h>
>  #include <linux/cpu.h>
> @@ -781,12 +782,14 @@ enum page_references {
>  
>  static enum page_references page_check_references(struct page *page,
>  						  struct mem_cgroup_zone *mz,
> -						  struct scan_control *sc)
> +						  struct scan_control *sc,
> +						  int mmap_mutex_locked)
>  {
>  	int referenced_ptes, referenced_page;
>  	unsigned long vm_flags;
>  
> -	referenced_ptes = page_referenced(page, 1, mz->mem_cgroup, &vm_flags);
> +	referenced_ptes = page_referenced(page, 1, mz->mem_cgroup, &vm_flags,
> +					  mmap_mutex_locked);
>  	referenced_page = TestClearPageReferenced(page);
>  
>  	/* Lumpy reclaim - ignore references */
> @@ -856,6 +859,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>  	unsigned long nr_congested = 0;
>  	unsigned long nr_reclaimed = 0;
>  	unsigned long nr_writeback = 0;
> +	struct mutex *i_mmap_mutex = NULL;
>  
>  	cond_resched();
>  
> @@ -865,12 +869,14 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>  		struct address_space *mapping;
>  		struct page *page;
>  		int may_enter_fs;
> +		int mmap_mutex_locked;
>  
>  		cond_resched();
>  
>  		page = lru_to_page(page_list);
>  		list_del(&page->lru);
>  
> +		mmap_mutex_locked = 0;
>  		if (!trylock_page(page))
>  			goto keep;
>  
> @@ -909,7 +915,23 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>  			}
>  		}
>  
> -		references = page_check_references(page, mz, sc);
> +		if (page->mapping) {
> +			if (i_mmap_mutex == &page->mapping->i_mmap_mutex) {
> +				mmap_mutex_locked = 1;
> +			} else {
> +				if (page_mapped(page) && page_rmapping(page)
> +					&& !PageKsm(page) && !PageAnon(page)) {
> +					if (i_mmap_mutex)
> +						mutex_unlock(i_mmap_mutex);
> +					i_mmap_mutex = &page->mapping->i_mmap_mutex;
> +					mutex_lock(i_mmap_mutex);
> +					mmap_mutex_locked = 1;
> +				}
> +			}
> +		}
> +
> +		references = page_check_references(page, mz, sc,
> +						   mmap_mutex_locked);
>  		switch (references) {
>  		case PAGEREF_ACTIVATE:
>  			goto activate_locked;
> @@ -939,7 +961,8 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>  		 * processes. Try to unmap it here.
>  		 */
>  		if (page_mapped(page) && mapping) {
> -			switch (try_to_unmap(page, TTU_UNMAP)) {
> +			switch (try_to_unmap(page, TTU_UNMAP,
> +						mmap_mutex_locked)) {
>  			case SWAP_FAIL:
>  				goto activate_locked;
>  			case SWAP_AGAIN:
> @@ -1090,6 +1113,8 @@ keep_lumpy:
>  		VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
>  	}
>  
> +	if (i_mmap_mutex)
> +		mutex_unlock(i_mmap_mutex);
>  	nr_reclaimed += __remove_mapping_batch(&unmap_pages, &ret_pages,
>  					       &free_pages);
>  
> @@ -1818,7 +1843,7 @@ static void shrink_active_list(unsigned long nr_to_scan,
>  			}
>  		}
>  
> -		if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags)) {
> +		if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags, 0)) {
>  			nr_rotated += hpage_nr_pages(page);
>  			/*
>  			 * Identify referenced, file-backed active pages and
>