Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Byungchul Park]

On Thu, Jun 08, 2017 at 03:02:43PM +0100, Juri Lelli wrote:
> On 07/06/17 09:14, Byungchul Park wrote:
> > On Wed, Jun 07, 2017 at 08:42:24AM +0900, Byungchul Park wrote:
> > > On Tue, Jun 06, 2017 at 04:12:25PM +0100, Juri Lelli wrote:
> > > > Hi,
> > > > 
> > > > On 02/06/17 16:31, Byungchul Park wrote:
> 
> [...]
> 
> > > > >  
> > > > >  static inline int cpudl_maximum_cpu(struct cpudl *cp)
> > > > >  {
> > > > > -	return cp->elements[0].cpu;
> > > > > +	int cpu = cp->elements[0].cpu;
> > > > > +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
> > > > 
> > > > Mmm, don't we get a WARN from cpumask_check() if we return -1 here?
> > > 
> > > The function does not return -1 without my patch.
> > > 
> > > Right?
> > 
> 
> That's actually my point: with the change you are proposing we will
> start returning -1 and it looks to me that the WARN will start to fire.

Hi,

I see what you talk about. You are talking about WARN in cpumask_check().
Sorry for missing your words.

> What about the below instead (properly splitted in 2 patches I guess,
> and I'm not sure at all the macro thing is pretty at all) ?
> 
> --->8---
>  kernel/sched/cpudeadline.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index fba235c7d026..32e3dcef2b81 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -108,11 +108,17 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
>  		cpudl_heapify_down(cp, idx);
>  }
>  
> -static inline int cpudl_maximum(struct cpudl *cp)
> -{
> -	return cp->elements[0].cpu;
> +#define cpudl_maximum(field)				\
> +static inline int cpudl_maximum_##field			\
> +(struct cpudl *cp)					\
> +{							\
> +	return cp->elements[0].field;			\
>  }
>  
> +cpudl_maximum(cpu);
> +cpudl_maximum(dl);
> +cpudl_maximum(idx);
> +
>  /*
>   * cpudl_find - find the best (later-dl) CPU in the system
>   * @cp: the cpudl max-heap context
> @@ -131,9 +137,10 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
>  	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
>  		best_cpu = cpumask_any(later_mask);
>  		goto out;
> -	} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> -			dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
> -		best_cpu = cpudl_maximum(cp);
> +	} else if (cpudl_maximum_idx(cp) != IDX_INVALID &&
> +		   cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
> +		   dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
> +		best_cpu = cpudl_maximum_cpu(cp);

This would also work and avoid unnecessary warning. I missed the check
to avoid it. https://lkml.org/lkml/2017/3/23/175 was an original patch
doing it.

By the way, frankly speaking, I don't like accessing the cpudl instant
several times without protection. I rather prefer the following..

But whatever. I like both.

Thnaks,
Byungchul

----->8-----
diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index 9b314a9..1d369cf 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -137,11 +137,17 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
 	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
 		best_cpu = cpumask_any(later_mask);
 		goto out;
-	} else if (cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
-			dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
-		best_cpu = cpudl_maximum_cpu(cp);
-		if (later_mask)
-			cpumask_set_cpu(best_cpu, later_mask);
+	} else {
+		int max_cpu = cpudl_maximum_cpu(cp);
+		u64 max_dl = cpudl_maximum_dl(cp);
+
+		if (max_cpu != -1 &&
+		    cpumask_test_cpu(max_cpu, &p->cpus_allowed) &&
+		    dl_time_before(dl_se->deadline, max_dl)) {
+			best_cpu = max_cpu;
+			if (later_mask)
+				cpumask_set_cpu(best_cpu, later_mask);
+		}
 	}
 
 out: