[PATCH] kfifo: DECLARE_KIFO_PTR(fifo, u64) does not work on arm 32 bit

[PATCH] kfifo: DECLARE_KIFO_PTR(fifo, u64) does not work on arm 32 bit

[Matthew Weber]

From: Sean Young <sean@mess.org>

If you try to store u64 in a kfifo (or a struct with u64 members),
then the buf member of __STRUCT_KFIFO_PTR will cause 4 bytes
padding due to alignment (note that struct __kfifo is 20 bytes
on 32 bit).

That in turn causes the __is_kfifo_ptr() to fail, which is caught
by kfifo_alloc(), which now returns EINVAL.

So, ensure that __is_kfifo_ptr() compares to the right structure.

Signed-off-by: Sean Young <sean@mess.org>
Acked-by: Stefani Seibold <stefani@seibold.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
Signed-off-by: Matthew Weber <matthew.weber@collins.com>
---

This patch originally made it into 4.16 as 
8a866fee3909c49738e1c4429a8d2b9bf27e015d but is a bug on at least 4.14
for any 32bit system(PPC/ARM/...) using kfifo with u64 datatypes.

Please add to linux-4.14.y for the next LTS tag.
(Below patch is a cherry-pick of the commit onto linux-4.14.y and was
 verified on 32bit hardware.)

---
 include/linux/kfifo.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/include/linux/kfifo.h b/include/linux/kfifo.h
index 41eb6fdf87a8..86b5fb08e96c 100644
--- a/include/linux/kfifo.h
+++ b/include/linux/kfifo.h
@@ -113,7 +113,8 @@ struct kfifo_rec_ptr_2 __STRUCT_KFIFO_PTR(unsigned char, 2, void);
  * array is a part of the structure and the fifo type where the array is
  * outside of the fifo structure.
  */
-#define	__is_kfifo_ptr(fifo)	(sizeof(*fifo) == sizeof(struct __kfifo))
+#define	__is_kfifo_ptr(fifo) \
+	(sizeof(*fifo) == sizeof(STRUCT_KFIFO_PTR(typeof(*(fifo)->type))))
 
 /**
  * DECLARE_KFIFO_PTR - macro to declare a fifo pointer object
-- 
2.17.1

Re: [PATCH] kfifo: DECLARE_KIFO_PTR(fifo, u64) does not work on arm 32 bit

[Greg KH]

On Thu, Jul 08, 2021 at 10:23:41AM -0500, Matthew Weber wrote:
> From: Sean Young <sean@mess.org>
> 
> If you try to store u64 in a kfifo (or a struct with u64 members),
> then the buf member of __STRUCT_KFIFO_PTR will cause 4 bytes
> padding due to alignment (note that struct __kfifo is 20 bytes
> on 32 bit).
> 
> That in turn causes the __is_kfifo_ptr() to fail, which is caught
> by kfifo_alloc(), which now returns EINVAL.
> 
> So, ensure that __is_kfifo_ptr() compares to the right structure.
> 
> Signed-off-by: Sean Young <sean@mess.org>
> Acked-by: Stefani Seibold <stefani@seibold.net>
> Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
> Signed-off-by: Matthew Weber <matthew.weber@collins.com>
> ---
> 
> This patch originally made it into 4.16 as 
> 8a866fee3909c49738e1c4429a8d2b9bf27e015d but is a bug on at least 4.14
> for any 32bit system(PPC/ARM/...) using kfifo with u64 datatypes.
> 
> Please add to linux-4.14.y for the next LTS tag.
> (Below patch is a cherry-pick of the commit onto linux-4.14.y and was
>  verified on 32bit hardware.)

Now queued up, thanks.

greg k-h