Re: [Intel-gfx] [PATCH 1/2] drm/i915/display: use fetch_and_zero if applicable

[Jani Nikula <jani.nikula@linux.intel.com>]

On Thu, 08 Dec 2022, "Vivi, Rodrigo" <rodrigo.vivi@intel.com> wrote:
> On Thu, 2022-12-08 at 14:32 +0200, Jani Nikula wrote:
>> On Thu, 08 Dec 2022, Andrzej Hajda <andrzej.hajda@intel.com> wrote:
>> > Simplify the code.
>> 
>> Personally, I absolutely hate fetch_and_zero().
>> 
>> I understand the point, but there are two main traps:
>> 
>> First, the name implies atomicity, which there is none at all.
>> 
>> Second, the name implies it's part of a kernel core header, which it
>> isn't, and this just amplifies the first point.
>> 
>> It's surprising and misleading, and those are not things I like about
>> interfaces in the kernel.
>> 
>> I would not like to see this proliferate. If fetch_and_zero() was
>> atomic
>> *and* part of a core kernel header, it would be a different matter.
>> But
>> I don't think that's going to happen, exactly because it won't be
>> atomic
>> and the name implies it is.
>
> +1 here.
>
> Please let's go the other way around and try to kill macros like this.
>
> we either kill or we ensure this gets accepted in the core kernel
> libraries.

Agreed. I'd be fine with either:

1) Get something like this accepted in core kernel headers:

#define fetch_and_zero(ptr) xchg(ptr, 0)

2) Do this in i915:

@@
expression E;
@@

- fetch_and_zero(E)
+ xchg(E, 0)


BR,
Jani.


-- 
Jani Nikula, Intel Open Source Graphics Center