Re: [PATCH v3 3/4] lib/find_bit: optimize find_next_bit() functions

[Yury Norov]

On Wed, Sep 07, 2022 at 05:27:08PM +0100, Valentin Schneider wrote:
> On 27/08/22 10:58, Yury Norov wrote:
> > +#define FIND_NEXT_BIT(FETCH, MUNGE, size, start)				\
> > +({										\
> > +	unsigned long mask, idx, tmp, sz = (size), __start = (start);		\
> > +										\
> > +	if (unlikely(__start >= sz))						\
> > +		goto out;							\
> > +										\
> > +	mask = MUNGE(BITMAP_FIRST_WORD_MASK(__start));				\
> > +	idx = __start / BITS_PER_LONG;						\
> > +										\
> > +	for (tmp = (FETCH) & mask; !tmp; tmp = (FETCH)) {			\
> > +		if (idx > sz / BITS_PER_LONG)					\
> 
> Does that want to be

Yes, I already fixed this.
 
>                 if (idx + 1 >= sz / BITS_PER_LONG)
> 
> ?
> 
> Consider this as used in _find_next_bit() for an all-zero 128-bit wide
> bitmap (two ULL's), providing the memory contiguous to the bitmap is also
> zero then this will only stop at idx=3, so that's fetching two ULLs too
> far.
> 
> > +			goto out;						\
> > +		idx++;								\
> > +	}									\
> > +										\
> > +	sz = min(idx * BITS_PER_LONG + __ffs(MUNGE(tmp)), sz);			\
> > +out:										\
> > +	sz;									\
> > +})