On 01/20/2014 02:13 AM, Peter Zijlstra wrote:
> On Mon, Jan 20, 2014 at 09:30:21AM +0100, Peter Zijlstra wrote:
>> Then make them so. The fact was that most of the mwait idle sites
>> were bloody broken. And the single mwait_idle_with_hints() function
>> presents a single nice function that does all the required magics.
>
> To stress this a bit more; have a look see at mwwait_idle_with_hints();
> it does a whole lot of subtle magic.
>
> - current_{set,clr}_polling*(), these are crucial in not missing and
> wrecking NEED_RESCHED state.
>
> - X86_FEATURE_CLFLUSH_MONTIOR quirk
>
> - Does the monitor(); if (!need_resched()) mwait() thing.
>
> All of those are required for a correct and functional idle loop. And
> I've seen sites where any or all of the above were missing/broken.
>
> Not unifying the lot into a simple usable function is just stupid --
> history has shown people simply cannot be trusted to get this right.
>
I don't think anyone is arguing that. The question is rather if the
implementation is correct, and if it is ready for the merge window.
-hpa
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
