Re: Problem with find_first_bit function and kin

[Alexander van Heukelum]

On Sun, 17 Aug 2008 19:22:23 -0700 (PDT), "Alex Dubov" <oakad@xxxxxxxxx>
said:
> --- On Sun, 8/17/08, Alexander van Heukelum <heukelum@xxxxxxxxxxx> wrote:
> 
> > From: Alexander van Heukelum <heukelum@xxxxxxxxxxx>
> > Subject: Re: Problem with find_first_bit function and kin
> > To: oakad@xxxxxxxxx, "Linux kernel mailing list" <linux-kernel@xxxxxxxxxxxxxxx>
> > Date: Sunday, August 17, 2008, 4:58 AM
> > On Sat, 16 Aug 2008 12:26:54 -0700 (PDT), "Alex
> > Dubov" <oakad@xxxxxxxxx>
> > said:
> > > It's well may be that I'm just missing
> > something obvious.
> > > 
> > > It seems to me that find_first_bit/find_next_bit
> > functions return their
> > > offsets "base 1" - first set bit is
> > "1" and last is "bitmap size". This
> > > means that if only the last bit in the bitmap is set,
> > the returned value
> > > will be indistinguishable from no bits set situation.
> > Moreover, bit
> > > manipulation functions appear to use "base
> > 0" bit addresses, adding to
> > > the
> > > inconvenience.
> > > 
> > > Is this a desired behavior? And, if yes, how is one
> > supposed to deal with
> > > last bit of the bitmap?
> > 
> > Hi Alex,
> > 
> > If this is the behaviour you observe, it's a bug. How
> > did you
> > find out?
> > 
> > The intended behaviour is that the bits are enumerated in
> > "base 0"
> > style. If only the last bit in the bitmap is set it should
> > return
> > bitmapsize-1 and if no bit is set it should return
> > bitmapsize.
> > Some architecture-specific code gets the last detail wrong,
> > they
> > return a value that is slightly larger than the bitmap size
> > in
> > some cases.
> > 
> > Greetings,
> >     Alexander
> > -- 
> 
> I was compiling some stuff out-of-the tree and got "ffs" linked in
> instead of "__ffs", which caused all this weird behavior (on 64b
> platform). Sorry for the noise.
> 
> But then, why "ffs" behaves differently from "__ffs" and whats the
> reason they both exist?

The difference is that __ffs does base-0 indexing on a long type
and ffs does base-1 indexing on a regular int type. Also, __ffs(0)
is undefined, while ffs(0) returns 0. ffs is the odd one out in the
kernel, but its behaviour is set in stone, because gcc may replace
any call to ffs with its own (usually inlined) version, which
behaves as described.

Greetings,
    Alexander
-- 
  Alexander van Heukelum
  heukelum@xxxxxxxxxxx

-- 
http://www.fastmail.fm - Email service worth paying for. Try it for free

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/