Re: [patch 05/50] sched: Provide schedule point for RT locks

[Peter Zijlstra]

On Wed, Jul 14, 2021 at 11:49:47AM +0200, Thomas Gleixner wrote:
> On Wed, Jul 14 2021 at 10:28, Peter Zijlstra wrote:
> > On Tue, Jul 13, 2021 at 05:10:59PM +0200, Thomas Gleixner wrote:
> >
> >> --- a/kernel/sched/core.c
> >> +++ b/kernel/sched/core.c
> >> @@ -5832,8 +5832,14 @@ pick_next_task(struct rq *rq, struct tas
> >>   */
> >>  #define	SM_NONE			0x0
> >>  #define	SM_PREEMPT		0x1
> >> -#define SM_MASK_PREEMPT		UINT_MAX
> >> -#define SM_MASK_STATE		SM_MASK_PREEMPT
> >> +#ifndef CONFIG_PREEMPT_RT
> >> +# define SM_MASK_PREEMPT	UINT_MAX
> >> +# define SM_MASK_STATE		SM_MASK_PREEMPT
> >> +#else
> >> +# define SM_RTLOCK_WAIT		0x2
> >> +# define SM_MASK_PREEMPT	SM_PREEMPT
> >> +# define SM_MASK_STATE		(SM_PREEMPT | SM_RTLOCK_WAIT)
> >> +#endif
> >
> > Wouldn't something like this:
> >
> >
> > #ifndef CONFIG_PREEMPT_RT
> > # define SM_MASK_PREEMPT        (~0U)
> > #else
> > # define SM_RTLOCK_WAIT         0x2
> > # define SM_MASK_PREEMPT        SM_PREEMPT
> > #endif
> >
> > #define SM_MASK_STATE          (~0U)
> >
> > Be even better?
> 
> SM_MASK_STATE is overengineered. See combo patch 4+5 below

Yep, that should result in similar code as my proposal, thanks!

nit: you like UINT_MAX better than (~0U) ?