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

[Tim Chen]

On Tue, 2012-08-21 at 09:21 -0400, Matthew Wilcox wrote:

> 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:

I like your proposal and will try to test with a new patch along your
suggestions.  Though I will be out the rest of the week and may be
delayed a bit getting the testing completed.

Tim

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/