Re: [tip:x86/mce] x86/bitops: Move BIT_64() for a wider use

["H. Peter Anvin" <hpa@zytor.com>]

On 05/23/2012 09:43 AM, Linus Torvalds wrote:
> On Wed, May 23, 2012 at 9:31 AM, H. Peter Anvin <hpa@zytor.com> wrote:
>>>>
>>>> And it should return UL for shift values < 32 and ULL otherwise.
>>>
>>
>> Why do you want that behavior?  That seems bizarre...
> 
> We *have* to have that behavior.
> 
> A 64-bit value on a 32-bit architecture has fundamentally different
> semantics than a 32-bit one.
> 
> It expands arithmetic, but it has other semantic differences too.
> Think "printf()" etc. We don't want to force people to do 64-bit
> arithmetic on x86-32 when they are working with BIT(0), for chrissake!
> 
> So if people make BIT(0) be a 64-bit value on a 32-bit architecture,
> I'm going to run around naked with a chainsaw, and call people morons.
> That's just not acceptable.
> 

BIT(0), okay.  I thought we were talking about BIT_64() here...

Any reason we can't just tell people to use BIT() for a native "unsigned
long" type (32/64 bits) and BIT_64() if they really want a 64-bit result?

There are good reasons for the latter.  Consider, for example:

	u64 msr;
	...
	msr &= ~BIT_64(1);

This *better* not be an unsigned 32 bit value, or we just chopped off
the upper half.  In this case and similar ones the 64-bitness of the
result really matters.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.