Skip to contentSkip to conversation

CFS related question

DISCUSSION Cyrill Gorcunov 3 participants 5 messages Oct 18, 2008 – Oct 20, 2008

lore
Share & tools
lore
#0 lore raw

Hi Ingo, Peter,

I just curious, look we have the following

static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)

{
	struct sched_entity *se = NULL;

	if (first_fair(cfs_rq)) {
		se = __pick_next_entity(cfs_rq);
		se = pick_next(cfs_rq, se);
		set_next_entity(cfs_rq, se);
	}

	return se;
}

which I presume may return NULL so the following piece could fail

static struct task_struct *pick_next_task_fair(struct rq *rq)

{
	struct task_struct *p;
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;

	if (unlikely(!cfs_rq->nr_running))
		return NULL;

	do {
-->		se = pick_next_entity(cfs_rq);
--> OOPs	cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

	p = task_of(se);
	hrtick_start_fair(rq, p);

	return p;
}

Did I miss something? Or it comepletely can NOT happen?

- Cyrill -

#1 lore raw
Quoted 40 lines · Cyrill Gorcunov

Hi Ingo, Peter,

I just curious, look we have the following

static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)

{
	struct sched_entity *se = NULL;

	if (first_fair(cfs_rq)) {
		se = __pick_next_entity(cfs_rq);
		se = pick_next(cfs_rq, se);
		set_next_entity(cfs_rq, se);
	}

	return se;
}

which I presume may return NULL so the following piece could fail

static struct task_struct *pick_next_task_fair(struct rq *rq)

{
	struct task_struct *p;
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;

	if (unlikely(!cfs_rq->nr_running))
		return NULL;

	do {
-->		se = pick_next_entity(cfs_rq);
--> OOPs	cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

	p = task_of(se);
	hrtick_start_fair(rq, p);

	return p;
}
Quoted 1 line

Did I miss something? Or it comepletely can NOT happen?

pick_next_entity() only returns NULL when !first_fair(), which is when ! nr_running.

So the initial !nr_running check in pick_next_task_fair() will catch that. Further nested RQs will never have !nr_running because then they get dequeued.

#2 lore raw

[Peter Zijlstra - Sat, Oct 18, 2008 at 10:28:18PM +0200]

| On Sun, 2008-10-19 at 00:03 +0400, Cyrill Gorcunov wrote:
| > Hi Ingo, Peter,
| > 
| > I just curious, look we have the following
| > 
| > static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
| > {
| > 	struct sched_entity *se = NULL;
| > 
| > 	if (first_fair(cfs_rq)) {
| > 		se = __pick_next_entity(cfs_rq);
| > 		se = pick_next(cfs_rq, se);
| > 		set_next_entity(cfs_rq, se);
| > 	}
| > 
| > 	return se;
| > }
| > 
| > which I presume may return NULL so the following piece
| > could fail
| > 
| > static struct task_struct *pick_next_task_fair(struct rq *rq)
| > {
| > 	struct task_struct *p;
| > 	struct cfs_rq *cfs_rq = &rq->cfs;
| > 	struct sched_entity *se;
| > 
| > 	if (unlikely(!cfs_rq->nr_running))
| > 		return NULL;
| > 
| > 	do {
| > -->		se = pick_next_entity(cfs_rq);
| > --> OOPs	cfs_rq = group_cfs_rq(se);
| > 	} while (cfs_rq);
| > 
| > 	p = task_of(se);
| > 	hrtick_start_fair(rq, p);
| > 
| > 	return p;
| > }
| > 
| > Did I miss something? Or it comepletely can NOT happen?
| 
| pick_next_entity() only returns NULL when !first_fair(), which is when !
| nr_running.
| 
| So the initial !nr_running check in pick_next_task_fair() will catch
| that. Further nested RQs will never have !nr_running because then they
| get dequeued.
| 
| 
| 

Thanks Peter!

- Cyrill -

#3 lore raw
In reply to Peter Zijlstra
Quoted 41 lines · Peter Zijlstra
Quoted 40 lines · Cyrill Gorcunov

Hi Ingo, Peter,

I just curious, look we have the following

static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)

{
	struct sched_entity *se = NULL;

	if (first_fair(cfs_rq)) {
		se = __pick_next_entity(cfs_rq);
		se = pick_next(cfs_rq, se);
		set_next_entity(cfs_rq, se);
	}

	return se;
}

which I presume may return NULL so the following piece could fail

static struct task_struct *pick_next_task_fair(struct rq *rq)

{
	struct task_struct *p;
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;

	if (unlikely(!cfs_rq->nr_running))
		return NULL;

	do {
-->		se = pick_next_entity(cfs_rq);
--> OOPs	cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

	p = task_of(se);
	hrtick_start_fair(rq, p);

	return p;
}
Quoted 8 lines
Quoted 1 line

Did I miss something? Or it comepletely can NOT happen?

pick_next_entity() only returns NULL when !first_fair(), which is when ! nr_running.

So the initial !nr_running check in pick_next_task_fair() will catch that. Further nested RQs will never have !nr_running because then they get dequeued.

Hi Peter,

pick_next_entity() is used in pick_next_task_fair() only. So, checking first_fair() never fail, and if fails it means bug. Right?

How about the below patch?
--------
From: Hiroshi Shimamoto <[email protected]>
Subject: [PATCH] sched: replace check with BUG_ON in pick_next_entity()

BUG_ON instead of returning NULL in pick_next_entity() when !first_fair(). Basically first_fair() is always true, and returning NULL will cause oops later.

---
 kernel/sched_fair.c |   12 ++++++------
 1 files changed, 6 insertions(+), 6 deletions(-)
kernel/sched_fair.c +6 -6
index 9573c33..3ce7c25 100644
@@ -758,13 +758,13 @@ pick_next(struct cfs_rq *cfs_rq, struct sched_entity *se)
 
 static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
 {
-	struct sched_entity *se = NULL;
+	struct sched_entity *se;
 
-	if (first_fair(cfs_rq)) {
-		se = __pick_next_entity(cfs_rq);
-		se = pick_next(cfs_rq, se);
-		set_next_entity(cfs_rq, se);
-	}
+	BUG_ON(!first_fair(cfs_rq));
+
+	se = __pick_next_entity(cfs_rq);
+	se = pick_next(cfs_rq, se);
+	set_next_entity(cfs_rq, se);
 
 	return se;
 }
Signature · 1 line
1.5.6
#4 lore raw

[Hiroshi Shimamoto - Mon, Oct 20, 2008 at 11:12:36AM -0700]

| Peter Zijlstra wrote:
| > On Sun, 2008-10-19 at 00:03 +0400, Cyrill Gorcunov wrote:
| >> Hi Ingo, Peter,
| >>
| >> I just curious, look we have the following
| >>
| >> static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
| >> {
| >> 	struct sched_entity *se = NULL;
| >>
| >> 	if (first_fair(cfs_rq)) {
| >> 		se = __pick_next_entity(cfs_rq);
| >> 		se = pick_next(cfs_rq, se);
| >> 		set_next_entity(cfs_rq, se);
| >> 	}
| >>
| >> 	return se;
| >> }
| >>
| >> which I presume may return NULL so the following piece
| >> could fail
| >>
| >> static struct task_struct *pick_next_task_fair(struct rq *rq)
| >> {
| >> 	struct task_struct *p;
| >> 	struct cfs_rq *cfs_rq = &rq->cfs;
| >> 	struct sched_entity *se;
| >>
| >> 	if (unlikely(!cfs_rq->nr_running))
| >> 		return NULL;
| >>
| >> 	do {
| >> -->		se = pick_next_entity(cfs_rq);
| >> --> OOPs	cfs_rq = group_cfs_rq(se);
| >> 	} while (cfs_rq);
| >>
| >> 	p = task_of(se);
| >> 	hrtick_start_fair(rq, p);
| >>
| >> 	return p;
| >> }
| >>
| >> Did I miss something? Or it comepletely can NOT happen?
| > 
| > pick_next_entity() only returns NULL when !first_fair(), which is when !
| > nr_running.
| > 
| > So the initial !nr_running check in pick_next_task_fair() will catch
| > that. Further nested RQs will never have !nr_running because then they
| > get dequeued.
| 
| Hi Peter,
| 
| pick_next_entity() is used in pick_next_task_fair() only.
| So, checking first_fair() never fail, and if fails it means bug. Right?
| 
| How about the below patch?
| --------
| From: Hiroshi Shimamoto <[email protected]>
| Subject: [PATCH] sched: replace check with BUG_ON in pick_next_entity()
| 
| BUG_ON instead of returning NULL in pick_next_entity() when !first_fair().
| Basically first_fair() is always true, and returning NULL will cause oops later.
| 
| Signed-off-by: Hiroshi Shimamoto <[email protected]>
| ---
|  kernel/sched_fair.c |   12 ++++++------
|  1 files changed, 6 insertions(+), 6 deletions(-)
| 
| diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
| index 9573c33..3ce7c25 100644
| --- a/kernel/sched_fair.c
| +++ b/kernel/sched_fair.c
| @@ -758,13 +758,13 @@ pick_next(struct cfs_rq *cfs_rq, struct sched_entity *se)
|  
|  static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
|  {
| -	struct sched_entity *se = NULL;
| +	struct sched_entity *se;
|  
| -	if (first_fair(cfs_rq)) {
| -		se = __pick_next_entity(cfs_rq);
| -		se = pick_next(cfs_rq, se);
| -		set_next_entity(cfs_rq, se);
| -	}
| +	BUG_ON(!first_fair(cfs_rq));
| +
| +	se = __pick_next_entity(cfs_rq);
| +	se = pick_next(cfs_rq, se);
| +	set_next_entity(cfs_rq, se);
|  
|  	return se;
|  }
| -- 
| 1.5.6
|

Technically it should not happen (as Peter explained). But it seems it happens sometime -- and most probably 'cause of error in another part of kernel (ie indirect error). So if Peter would not object against -- I would really like to better have BUG_ON check there :) Peter?

	http://lkml.org/lkml/2008/8/3/269
	http://lists.openwall.net/linux-kernel/2008/05/14/130
 
		- Cyrill -