Re: linux-next: manual merge of the gpio tree with the tip tree

[Peter Zijlstra]

On Thu, Nov 02, 2017 at 09:06:27AM +0100, Peter Zijlstra wrote:
> On Thu, Nov 02, 2017 at 03:59:52PM +1100, Stephen Rothwell wrote:
> > diff --cc include/linux/bitops.h
> > index 15a5bcfcd0a2,9a874deee6e2..000000000000
> > --- a/include/linux/bitops.h
> > +++ b/include/linux/bitops.h
> > @@@ -227,32 -227,30 +227,56 @@@ static inline unsigned long __ffs64(u6
> >   	return __ffs((unsigned long)word);
> >   }
> >   
> >  +/*
> >  + * clear_bit32 - Clear a bit in memory for u32 array
> >  + * @nr: Bit to clear
> >  + * @addr: u32 * address of bitmap
> >  + *
> >  + * Same as clear_bit, but avoids needing casts for u32 arrays.
> >  + */
> >  +
> >  +static __always_inline void clear_bit32(long nr, volatile u32 *addr)
> >  +{
> >  +	clear_bit(nr, (volatile unsigned long *)addr);
> >  +}
> >  +
> >  +/*
> >  + * set_bit32 - Set a bit in memory for u32 array
> >  + * @nr: Bit to clear
> >  + * @addr: u32 * address of bitmap
> >  + *
> >  + * Same as set_bit, but avoids needing casts for u32 arrays.
> >  + */
> >  +
> >  +static __always_inline void set_bit32(long nr, volatile u32 *addr)
> >  +{
> >  +	set_bit(nr, (volatile unsigned long *)addr);
> >  +}
> 
> How is that not fundamentally broken for big-endian?

I think its broken in another way as well, {set,clear}_bit() are atomic
ops that expect a naturally aligned 'unsigned long', and that cast can
break the alignment as well.

And I'm fairly sure we have architectures where the atomic ops
misbehave or maybe even trap when you don't feed them properly aligned
data.