2015-12-08 15:42:52

by Arnd Bergmann

[permalink] [raw]
Subject: [PATCH] dm verity: use sector_div for division

The dm verity sec implementation uses do_div for dividing a sector_t,
which is slower than necessary when sector_t is a 32-bit type, and
we now get a warning for this case:

include/asm-generic/div64.h:224:22: warning: passing argument 1 of '__div64_32' from incompatible pointer type [-Wincompatible-pointer-types]
drivers/md/dm-verity-fec.c:725:6: note: in expansion of macro 'do_div'
if (do_div(f->rounds, f->rsn))

This changes the code to use sector_div instead, which does the
right thing and avoids the warning.

Signed-off-by: Arnd Bergmann <[email protected]>
---
drivers/md/dm-verity-fec.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

I believe we had a fix for this in the series that introduced the optimized
do_div, but the file got moved around and the fix got lost.

diff --git a/drivers/md/dm-verity-fec.c b/drivers/md/dm-verity-fec.c
index e722ce57520f..1bddaca37e98 100644
--- a/drivers/md/dm-verity-fec.c
+++ b/drivers/md/dm-verity-fec.c
@@ -722,7 +722,7 @@ int verity_fec_ctr(struct dm_verity *v)
}

f->rounds = f->blocks;
- if (do_div(f->rounds, f->rsn))
+ if (sector_div(f->rounds, f->rsn))
f->rounds++;

/*
--
2.1.0.rc2


2015-12-08 17:10:16

by Mike Snitzer

[permalink] [raw]
Subject: Re: dm verity: use sector_div for division

On Tue, Dec 08 2015 at 10:42am -0500,
Arnd Bergmann <[email protected]> wrote:

> The dm verity sec implementation uses do_div for dividing a sector_t,
> which is slower than necessary when sector_t is a 32-bit type, and
> we now get a warning for this case:
>
> include/asm-generic/div64.h:224:22: warning: passing argument 1 of '__div64_32' from incompatible pointer type [-Wincompatible-pointer-types]
> drivers/md/dm-verity-fec.c:725:6: note: in expansion of macro 'do_div'
> if (do_div(f->rounds, f->rsn))
>
> This changes the code to use sector_div instead, which does the
> right thing and avoids the warning.

Thanks, I'll fold this in.

(Btw, I did review this use of do_div(): but only to answer whether there
was potential for a 64bit divide. I concluded do_div() was fine but didn't
look to see the data type for f->rounds was actually sector_t.)