Re: [PATCH 07/15] x86: add xadd helper macro

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 08/22/2011 04:29 PM, H. Peter Anvin wrote:
> On 08/22/2011 04:15 PM, Jeremy Fitzhardinge wrote:
>> From: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
>>
>> Add a common xadd implementation.
>>
>> This has the side effect of generating a bad instruction if you try to
>> use it on a 64-bit value on a 32-bit system - but don't do that.
>>
> It would be better to barf at that point, so we get the error with a C
> line... also, there needs to be a default clause with
> __compiletime_error() in it.

OK, but the "standard of care" here is the old "calling undefined
function" link error; getting a bad asm instruction is a little more
helpful, in a sense.  But agreed on the default: case.  I'll see if I
can fix up xadd and cmpxchg to fail better as well.

> There are a few additional xadd users which should be converted unless
> I'm mistaken:
>
> rwsem_atomic_update() in asm/rwsem.h.
>
> atomic_add_return() in asm/atomic.h.
>
> atomic64_add_return() in asm/atomic64_64.h.
>
> atom_asr() in asm/uv/uv_bau.h (*VOMIT* - the UV people have created a
> whole different type in private code...)
>

Yeah, I hadn't done a full xadd audit.  I'll take a look.

    J