On Thu, Aug 11, 2016 at 12:51:31PM -0700, Linus Torvalds wrote:
> Ok. It does seem to also reset the active file page counts back, so
> that part did seem to be related, but yeah, from a performance
> standpoint that was clearly not a major issue.
> 
> Let's hope Dave can figure out something based on his numbers, because
> I'm out of ideas. Or maybe it's the pagefault-atomic thing that
> Christoph was looking at.

I can't really think of any reason why the pagefaul_disable() would
sіgnificantly change performance.  Anyway, the patch for the is below
(on top of the previous mark_page_accessed() one), so feel free to
re-run the test with it.  It would also be nice to see the profiles
with the two patches applied.

commit 43106eea246074acc4bb7d12fdb91f58002f52ed
Author: Christoph Hellwig <hch@lst.de>
Date:   Thu Aug 11 10:41:40 2016 -0700

    fs: remove superflous pagefault_disable from iomap_write_actor
    
    No idea where this really came from, generic_perform_write only briefly
    did a pagefaul_disable when trying a different prefault scheme.
    
    Signed-off-by: Christoph Hellwig <hch@lst.de>

diff --git a/fs/iomap.c b/fs/iomap.c
index f39c318..74712e2 100644
--- a/fs/iomap.c
+++ b/fs/iomap.c
@@ -194,9 +194,7 @@ again:
 		if (mapping_writably_mapped(inode->i_mapping))
 			flush_dcache_page(page);
 
-		pagefault_disable();
 		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
-		pagefault_enable();
 
 		flush_dcache_page(page);