2021-10-26 21:22:34

by Jens Axboe

Subject: [PATCH] mm: move more expensive part of XA setup out of mapping check

The fast path here is not needing any writeback, yet we spend time setting
up the xarray lookup data upfront. Move the part that actually needs to
iterate the address space mapping into a separate helper, saving ~30% of
the time here.

Signed-off-by: Jens Axboe <[email protected]>

---

diff --git a/mm/filemap.c b/mm/filemap.c
index 850920276846..0e4021edac0b 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -638,6 +638,30 @@ static bool mapping_needs_writeback(struct address_space *mapping)
return mapping->nrpages;
}

+static bool filemap_range_has_writeback(struct address_space *mapping,
+ loff_t start_byte, loff_t end_byte)
+{
+ XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
+ pgoff_t max = end_byte >> PAGE_SHIFT;
+ struct page *page;
+
+ if (end_byte < start_byte)
+ return false;
+
+ rcu_read_lock();
+ xas_for_each(&xas, page, max) {
+ if (xas_retry(&xas, page))
+ continue;
+ if (xa_is_value(page))
+ continue;
+ if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
+ break;
+ }
+ rcu_read_unlock();
+ return page != NULL;
+
+}
+
/**
* filemap_range_needs_writeback - check if range potentially needs writeback
* @mapping: address space within which to check
@@ -655,29 +679,12 @@ static bool mapping_needs_writeback(struct address_space *mapping)
bool filemap_range_needs_writeback(struct address_space *mapping,
loff_t start_byte, loff_t end_byte)
{
- XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
- pgoff_t max = end_byte >> PAGE_SHIFT;
- struct page *page;
-
if (!mapping_needs_writeback(mapping))
return false;
if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
return false;
- if (end_byte < start_byte)
- return false;
-
- rcu_read_lock();
- xas_for_each(&xas, page, max) {
- if (xas_retry(&xas, page))
- continue;
- if (xa_is_value(page))
- continue;
- if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
- break;
- }
- rcu_read_unlock();
- return page != NULL;
+ return filemap_range_has_writeback(mapping, start_byte, end_byte);
}
EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);

--
Jens Axboe

2021-10-27 05:14:35

by Matthew Wilcox

Subject: Re: [PATCH] mm: move more expensive part of XA setup out of mapping check

On Tue, Oct 26, 2021 at 10:01:20AM -0600, Jens Axboe wrote:
> The fast path here is not needing any writeback, yet we spend time setting
> up the xarray lookup data upfront. Move the part that actually needs to
> iterate the address space mapping into a separate helper, saving ~30% of
> the time here.

No objection to this patch, but it did remind me that I never saw an answer to
https://lore.kernel.org/all/CAHk-=wg_-EwefQ_3Osz4iJxTrTk3tfrV53Z7-jaGg=tm9i5TXg@mail.gmail.com/
(and nor did lore)

2021-10-27 05:26:29

by Jens Axboe

Subject: Re: [PATCH] mm: move more expensive part of XA setup out of mapping check

On 10/26/21 12:08 PM, Matthew Wilcox wrote:
> On Tue, Oct 26, 2021 at 10:01:20AM -0600, Jens Axboe wrote:
>> The fast path here is not needing any writeback, yet we spend time setting
>> up the xarray lookup data upfront. Move the part that actually needs to
>> iterate the address space mapping into a separate helper, saving ~30% of
>> the time here.
>
> No objection to this patch, but it did remind me that I never saw an answer to
> https://lore.kernel.org/all/CAHk-=wg_-EwefQ_3Osz4iJxTrTk3tfrV53Z7-jaGg=tm9i5TXg@mail.gmail.com/
> (and nor did lore)

Totally missed that, and yes there could be cases where that lookup is
going to be way too slow. I'll take a look.

--
Jens Axboe