Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1605381 > unrolled thread

[PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology

Started byByungchul Park <byungchul.park@lge.com>
First post2017-03-21 09:00 +0100
Last post2017-03-23 00:30 +0100
Articles 5 — 3 participants

Back to article view | Back to linux.kernel


Contents

  [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology Byungchul Park <byungchul.park@lge.com> - 2017-03-21 09:00 +0100
    Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer  cpu in topology Daniel Bristot de Oliveira <bristot@redhat.com> - 2017-03-21 14:30 +0100
      Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer  cpu in topology Byungchul Park <byungchul.park@lge.com> - 2017-03-22 03:50 +0100
    Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer  cpu in topology Peter Zijlstra <peterz@infradead.org> - 2017-03-22 13:40 +0100
      Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer  cpu in topology Byungchul Park <byungchul.park@lge.com> - 2017-03-23 00:30 +0100

#1605381 — [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology

FromByungchul Park <byungchul.park@lge.com>
Date2017-03-21 09:00 +0100
Subject[PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology
Message-ID<tnmRX-fw-1@gated-at.bofh.it>
When cpudl_find() returns any among free_cpus, the cpu might not be
closer than others, considering sched domain. For example:

   this_cpu: 15
   free_cpus: 0, 1,..., 14 (== later_mask)
   best_cpu: 0

   topology:

   0 --+
       +--+
   1 --+  |
          +-- ... --+
   2 --+  |         |
       +--+         |
   3 --+            |

   ...             ...

   12 --+           |
        +--+        |
   13 --+  |        |
           +-- ... -+
   14 --+  |
        +--+
   15 --+

In this case, it would be best to select 14 since it's a free cpu and
closest to 15(this_cpu). However, currently the code select 0(best_cpu)
even though that's just any among free_cpus. Fix it.

Signed-off-by: Byungchul Park <byungchul.park@lge.com>
---
 kernel/sched/deadline.c | 29 +++++++++++++++--------------
 1 file changed, 15 insertions(+), 14 deletions(-)

diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index a2ce590..49c93b9 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -1324,7 +1324,7 @@ static int find_later_rq(struct task_struct *task)
 	struct sched_domain *sd;
 	struct cpumask *later_mask = this_cpu_cpumask_var_ptr(local_cpu_mask_dl);
 	int this_cpu = smp_processor_id();
-	int best_cpu, cpu = task_cpu(task);
+	int cpu = task_cpu(task);
 
 	/* Make sure the mask is initialized first */
 	if (unlikely(!later_mask))
@@ -1337,17 +1337,14 @@ static int find_later_rq(struct task_struct *task)
 	 * We have to consider system topology and task affinity
 	 * first, then we can look for a suitable cpu.
 	 */
-	best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
-			task, later_mask);
-	if (best_cpu == -1)
+	if (cpudl_find(&task_rq(task)->rd->cpudl, task, later_mask) == -1)
 		return -1;
 
 	/*
-	 * If we are here, some target has been found,
-	 * the most suitable of which is cached in best_cpu.
-	 * This is, among the runqueues where the current tasks
-	 * have later deadlines than the task's one, the rq
-	 * with the latest possible one.
+	 * If we are here, some targets have been found, including
+	 * the most suitable which is, among the runqueues where the
+	 * current tasks have later deadlines than the task's one, the
+	 * rq with the latest possible one.
 	 *
 	 * Now we check how well this matches with task's
 	 * affinity and system topology.
@@ -1367,6 +1364,7 @@ static int find_later_rq(struct task_struct *task)
 	rcu_read_lock();
 	for_each_domain(cpu, sd) {
 		if (sd->flags & SD_WAKE_AFFINE) {
+			int closest_cpu;
 
 			/*
 			 * If possible, preempting this_cpu is
@@ -1378,14 +1376,17 @@ static int find_later_rq(struct task_struct *task)
 				return this_cpu;
 			}
 
+			closest_cpu = cpumask_first_and(later_mask,
+							sched_domain_span(sd));
 			/*
-			 * Last chance: if best_cpu is valid and is
-			 * in the mask, that becomes our choice.
+			 * Last chance: if a cpu being in both later_mask
+			 * and current sd span is valid, that becomes our
+			 * choice. Of course, the latest possible cpu is
+			 * already under consideration through later_mask.
 			 */
-			if (best_cpu < nr_cpu_ids &&
-			    cpumask_test_cpu(best_cpu, sched_domain_span(sd))) {
+			if (closest_cpu < nr_cpu_ids) {
 				rcu_read_unlock();
-				return best_cpu;
+				return closest_cpu;
 			}
 		}
 	}
-- 
1.9.1

[toc] | [next] | [standalone]


#1605588 — Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology

FromDaniel Bristot de Oliveira <bristot@redhat.com>
Date2017-03-21 14:30 +0100
SubjectRe: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology
Message-ID<tns1j-3SO-9@gated-at.bofh.it>
In reply to#1605381
On 03/21/2017 08:52 AM, Byungchul Park wrote:
> When cpudl_find() returns any among free_cpus, the cpu might not be
> closer than others, considering sched domain. For example:
> 
>    this_cpu: 15
>    free_cpus: 0, 1,..., 14 (== later_mask)
>    best_cpu: 0
> 
>    topology:
> 
>    0 --+
>        +--+
>    1 --+  |
>           +-- ... --+
>    2 --+  |         |
>        +--+         |
>    3 --+            |
> 
>    ...             ...
> 
>    12 --+           |
>         +--+        |
>    13 --+  |        |
>            +-- ... -+
>    14 --+  |
>         +--+
>    15 --+
> 
> In this case, it would be best to select 14 since it's a free cpu and
> closest to 15(this_cpu). However, currently the code select 0(best_cpu)
> even though that's just any among free_cpus. Fix it.

That is a nice patch! But I wonder what would be the behavior with your
patch in the following hw:

# numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 2 4 6 8 10 12 14
node 0 size: 16159 MB
node 0 free: 15308 MB
node 1 cpus: 1 3 5 7 9 11 13 15
node 1 size: 16384 MB
node 1 free: 15028 MB
node distances:
node   0   1
  0:  10  21
  1:  21  10

-- Daniel

[toc] | [prev] | [next] | [standalone]


#1606171 — Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology

FromByungchul Park <byungchul.park@lge.com>
Date2017-03-22 03:50 +0100
SubjectRe: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology
Message-ID<tnEvw-3Xg-25@gated-at.bofh.it>
In reply to#1605588
On Tue, Mar 21, 2017 at 02:28:50PM +0100, Daniel Bristot de Oliveira wrote:
> On 03/21/2017 08:52 AM, Byungchul Park wrote:
> > When cpudl_find() returns any among free_cpus, the cpu might not be
> > closer than others, considering sched domain. For example:
> > 
> >    this_cpu: 15
> >    free_cpus: 0, 1,..., 14 (== later_mask)
> >    best_cpu: 0
> > 
> >    topology:
> > 
> >    0 --+
> >        +--+
> >    1 --+  |
> >           +-- ... --+
> >    2 --+  |         |
> >        +--+         |
> >    3 --+            |
> > 
> >    ...             ...
> > 
> >    12 --+           |
> >         +--+        |
> >    13 --+  |        |
> >            +-- ... -+
> >    14 --+  |
> >         +--+
> >    15 --+
> > 
> > In this case, it would be best to select 14 since it's a free cpu and
> > closest to 15(this_cpu). However, currently the code select 0(best_cpu)
> > even though that's just any among free_cpus. Fix it.
> 
> That is a nice patch! But I wonder what would be the behavior with your
> patch in the following hw:
> 
> # numactl --hardware
> available: 2 nodes (0-1)
> node 0 cpus: 0 2 4 6 8 10 12 14
> node 0 size: 16159 MB
> node 0 free: 15308 MB
> node 1 cpus: 1 3 5 7 9 11 13 15
> node 1 size: 16384 MB
> node 1 free: 15028 MB
> node distances:
> node   0   1
>   0:  10  21
>   1:  21  10

Hi,

In this case, I guess the topology looks like:

    0 --+
        +--+
    2 --+  |
           +-- ... --+
    4 --+  |         |
        +--+         |
    6 --+            |

    ...             ...

     9 --+           |
         +--+        |
    11 --+  |        |
            +-- ... -+
    13 --+  |
         +--+
    15 --+

And sched_domain would also reflect that. So the dl's push works well.
Do I miss something?

In addition, IMHO, it's not an issue for dl's push but one for building
sched_domains. Wrong?

Thanks,
Byungchul

[toc] | [prev] | [next] | [standalone]


#1606425 — Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology

FromPeter Zijlstra <peterz@infradead.org>
Date2017-03-22 13:40 +0100
SubjectRe: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology
Message-ID<tnNIt-2p2-17@gated-at.bofh.it>
In reply to#1605381
On Tue, Mar 21, 2017 at 04:52:24PM +0900, Byungchul Park wrote:
> When cpudl_find() returns any among free_cpus, the cpu might not be
> closer than others, considering sched domain. For example:
> 
>    this_cpu: 15
>    free_cpus: 0, 1,..., 14 (== later_mask)
>    best_cpu: 0
> 
>    topology:
> 
>    0 --+
>        +--+
>    1 --+  |
>           +-- ... --+
>    2 --+  |         |
>        +--+         |
>    3 --+            |
> 
>    ...             ...
> 
>    12 --+           |
>         +--+        |
>    13 --+  |        |
>            +-- ... -+
>    14 --+  |
>         +--+
>    15 --+
> 
> In this case, it would be best to select 14 since it's a free cpu and
> closest to 15(this_cpu). However, currently the code select 0(best_cpu)
> even though that's just any among free_cpus. Fix it.

This would result in picking the HT sibling, if available. Which is
typically the worst possible pick.

If you add support for SD_PREFER_SIBLING, which denotes a preference for
any other sibling domain above this one, this might work.

[toc] | [prev] | [next] | [standalone]


#1607068 — Re: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology

FromByungchul Park <byungchul.park@lge.com>
Date2017-03-23 00:30 +0100
SubjectRe: [PATCH v2] sched/deadline: Make find_later_rq() choose a closer cpu in topology
Message-ID<tnXRv-1Qw-1@gated-at.bofh.it>
In reply to#1606425
On Wed, Mar 22, 2017 at 01:37:28PM +0100, Peter Zijlstra wrote:
> On Tue, Mar 21, 2017 at 04:52:24PM +0900, Byungchul Park wrote:
> > When cpudl_find() returns any among free_cpus, the cpu might not be
> > closer than others, considering sched domain. For example:
> > 
> >    this_cpu: 15
> >    free_cpus: 0, 1,..., 14 (== later_mask)
> >    best_cpu: 0
> > 
> >    topology:
> > 
> >    0 --+
> >        +--+
> >    1 --+  |
> >           +-- ... --+
> >    2 --+  |         |
> >        +--+         |
> >    3 --+            |
> > 
> >    ...             ...
> > 
> >    12 --+           |
> >         +--+        |
> >    13 --+  |        |
> >            +-- ... -+
> >    14 --+  |
> >         +--+
> >    15 --+
> > 
> > In this case, it would be best to select 14 since it's a free cpu and
> > closest to 15(this_cpu). However, currently the code select 0(best_cpu)
> > even though that's just any among free_cpus. Fix it.
> 
> This would result in picking the HT sibling, if available. Which is
> typically the worst possible pick.
> 
> If you add support for SD_PREFER_SIBLING, which denotes a preference for
> any other sibling domain above this one, this might work.

Sure. I will add that support as well. Thank you.

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web