Re: [PATCH 1/2] IB/hfi1: Fix a parameter of find_first_bit.

[Leon Romanovsky]

On Fri, Aug 26, 2016 at 03:34:48PM -0400, Doug Ledford wrote:
> On 8/26/2016 3:29 PM, Leon Romanovsky wrote:
> > On Fri, Aug 26, 2016 at 02:01:55PM -0400, Doug Ledford wrote:
> >> On 8/26/2016 9:35 AM, Doug Ledford wrote:
> >>> On 8/26/2016 12:49 AM, Christophe JAILLET wrote:
> >>>> The 2nd parameter of 'find_first_bit' is the number of bits to search.
> >>>> In this case, we are passing 'sizeof(unsigned long)' which is likely to
> >>>> be 4 or 8.
> >>>
> >>> If the size can be 4 or 8, then using 64 universally is not correct.
> >>> Why not use sizeof() * 8 (or << 3)?
> >>
> >> Better yet, why not put this patch in the kernel first:
> >>
> >> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> >> index d96a6118d26a..a8838c87668e 100644
> >> --- a/include/linux/kernel.h
> >> +++ b/include/linux/kernel.h
> >> @@ -52,6 +52,8 @@
> >>
> >>  #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]) +
> >> __must_be_array(arr))
> >>
> >> +#define bitsizeof(x)   (sizeof((x)) << 3)
> >> +
> >>  #define u64_to_user_ptr(x) (           \
> >>  {                                      \
> >>         typecheck(u64, x);              \
> >>
> >> then start going around replacing all these hard coded numbers with the
> >> use of bitsizeof().  It can be applied not just to the find_first*bit()
> >> routines, but to a bunch of other routines too.  Just look at
> >> include/linux/bitmap.h and any that have nbits as an argument are
> >> candidates.
> >
> > There is BITS_PER_LONG define for that. There is actual use of it in mlx5 for
> > the similar code pieces.
>
> BITS_PER_LONG only works if your bitfield is a single long.  It doesn't
> work for other bitfields.  What I posted above will work for anything.

Yes, the question to ask if it is really needed.

>
>
> --
> Doug Ledford <dledford@xxxxxxxxxx>
>     GPG Key ID: 0E572FDD
>

Attachment: signature.asc
Description: PGP signature