[PATCH 08/10] netfs: Use folio_next_pos()

[PATCH 08/10] netfs: Use folio_next_pos()

[Matthew Wilcox (Oracle)]

This is one instruction more efficient than open-coding folio_pos() +
folio_size().  It's the equivalent of (x + y) << z rather than
x << z + y << z.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Paulo Alcantara <pc@manguebit.org>
Cc: netfs@lists.linux.dev
---
 fs/netfs/buffered_write.c | 2 +-
 fs/netfs/misc.c           | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/netfs/buffered_write.c b/fs/netfs/buffered_write.c
index 09394ac2c180..f9d62abef2ac 100644
--- a/fs/netfs/buffered_write.c
+++ b/fs/netfs/buffered_write.c
@@ -535,7 +535,7 @@ vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_gr
 		folio_unlock(folio);
 		err = filemap_fdatawrite_range(mapping,
 					       folio_pos(folio),
-					       folio_pos(folio) + folio_size(folio));
+					       folio_next_pos(folio));
 		switch (err) {
 		case 0:
 			ret = VM_FAULT_RETRY;
diff --git a/fs/netfs/misc.c b/fs/netfs/misc.c
index 486166460e17..82342c6d22cb 100644
--- a/fs/netfs/misc.c
+++ b/fs/netfs/misc.c
@@ -298,7 +298,7 @@ bool netfs_release_folio(struct folio *folio, gfp_t gfp)
 	if (folio_test_dirty(folio))
 		return false;
 
-	end = umin(folio_pos(folio) + folio_size(folio), i_size_read(&ctx->inode));
+	end = umin(folio_next_pos(folio), i_size_read(&ctx->inode));
 	if (end > ctx->zero_point)
 		ctx->zero_point = end;
 
-- 
2.47.2

Re: [PATCH 08/10] netfs: Use folio_next_pos()

[Paulo Alcantara]

"Matthew Wilcox (Oracle)" <willy@infradead.org> writes:

> This is one instruction more efficient than open-coding folio_pos() +
> folio_size().  It's the equivalent of (x + y) << z rather than
> x << z + y << z.
>
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> Cc: David Howells <dhowells@redhat.com>
> Cc: Paulo Alcantara <pc@manguebit.org>
> Cc: netfs@lists.linux.dev

Reviewed-by: Paulo Alcantara (Red Hat) <pc@manguebit.org>

Re: [PATCH 08/10] netfs: Use folio_next_pos()

[David Howells]

Matthew Wilcox (Oracle) <willy@infradead.org> wrote:

> This is one instruction more efficient than open-coding folio_pos() +
> folio_size().  It's the equivalent of (x + y) << z rather than
> x << z + y << z.

Should that be noted to the gcc bugzilla as a missed optimisation?

> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> Cc: David Howells <dhowells@redhat.com>
> Cc: Paulo Alcantara <pc@manguebit.org>
> Cc: netfs@lists.linux.dev

Acked-by: David Howells <dhowells@redhat.com>

Re: [PATCH 08/10] netfs: Use folio_next_pos()

[Matthew Wilcox]

On Mon, Oct 27, 2025 at 02:16:45PM +0000, David Howells wrote:
> Matthew Wilcox (Oracle) <willy@infradead.org> wrote:
> 
> > This is one instruction more efficient than open-coding folio_pos() +
> > folio_size().  It's the equivalent of (x + y) << z rather than
> > x << z + y << z.
> 
> Should that be noted to the gcc bugzilla as a missed optimisation?

I don't know?  Maybe there's a Law Of C Arithmetic that prevents it
from doing this optimisation that we can ignore because we know that
the calculation will never overflow.

I tried with both gcc and LLVM and both produce better code for g()
than for f():
https://godbolt.org/z/YTc883f6G

Interestingly, LLVM does better for f() than gcc does, but they produce
identical code for g() (minor difference like using sal instead of shl
but that feels unimportant).

If you want to file optimisation bug(s) against gcc, feel free to use
this example code.