Skip to contentSkip to conversation

idle_balance() does not call load_balance_newidle()

BUG Applied Vaidyanathan Srinivasan 3 participants 6 messages Dec 8, 2008 – Dec 9, 2008

lore
Share & tools
lore
#0 lore raw

Hi,

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

pulled_task is initialised to -1 and checked for non-zero which is always true if the lowest level sched_domain does not have SD_BALANCE_NEWIDLE flag set.

Trivial fix to initialise pulled_task to zero.
Patch against 2.6.28-rc7

Thanks,
Vaidy

---

 kernel/sched.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/sched.c +1 -1
index f79c7c4..4abdce1 100644
@@ -3801,7 +3801,7 @@ out_balanced:
 static void idle_balance(int this_cpu, struct rq *this_rq)
 {
 	struct sched_domain *sd;
-	int pulled_task = -1;
+	int pulled_task = 0;
 	unsigned long next_balance = jiffies + HZ;
 	cpumask_t tmpmask;
 
#1 lore raw
Quoted 9 lines · Vaidyanathan Srinivasan

Hi,

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

pulled_task is initialised to -1 and checked for non-zero which is always true if the lowest level sched_domain does not have SD_BALANCE_NEWIDLE flag set.

Quoted 7 lines

Trivial fix to initialise pulled_task to zero.
Patch against 2.6.28-rc7

Thanks,
Vaidy

Seems sane

Quoted 19 lines

---

 kernel/sched.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/sched.c +1 -1
index f79c7c4..4abdce1 100644
@@ -3801,7 +3801,7 @@ out_balanced:
 static void idle_balance(int this_cpu, struct rq *this_rq)
 {
 	struct sched_domain *sd;
-	int pulled_task = -1;
+	int pulled_task = 0;
 	unsigned long next_balance = jiffies + HZ;
 	cpumask_t tmpmask;
 
Applied
#2 lore raw
In reply to Vaidyanathan Srinivasan
Quoted 11 lines · Vaidyanathan Srinivasan

Hi,

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

pulled_task is initialised to -1 and checked for non-zero which is always true if the lowest level sched_domain does not have SD_BALANCE_NEWIDLE flag set.

Trivial fix to initialise pulled_task to zero.
Patch against 2.6.28-rc7

applied to tip/sched/core, thanks! (Not for v2.6.28 because this could affect performance.)

Ingo

#3 lore raw
Quoted 12 lines · Ingo Molnar
Quoted 9 lines · Vaidyanathan Srinivasan

Hi,

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

pulled_task is initialised to -1 and checked for non-zero which is always true if the lowest level sched_domain does not have SD_BALANCE_NEWIDLE flag set.

Quoted 5 lines
Quoted 2 lines

Trivial fix to initialise pulled_task to zero.
Patch against 2.6.28-rc7

applied to tip/sched/core, thanks! (Not for v2.6.28 because this could affect performance.)

Thanks Ingo. This patch does not change any functionality in v2.6.28 and hence will not affect performance. The SD flags are not touched. I found this bug while setting different SD flags at MC level and CPU level in my power saving balance patches.

--Vaidy

#4 lore raw
Quoted 17 lines · Vaidyanathan Srinivasan
Quoted 15 lines · Ingo Molnar
Quoted 11 lines · Vaidyanathan Srinivasan

Hi,

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

pulled_task is initialised to -1 and checked for non-zero which is always true if the lowest level sched_domain does not have SD_BALANCE_NEWIDLE flag set.

Trivial fix to initialise pulled_task to zero.
Patch against 2.6.28-rc7

Quoted 7 lines
Quoted 2 lines

applied to tip/sched/core, thanks! (Not for v2.6.28 because this could affect performance.)

Thanks Ingo. This patch does not change any functionality in v2.6.28 and hence will not affect performance. The SD flags are not touched. I found this bug while setting different SD flags at MC level and CPU level in my power saving balance patches.

if it does not change any functionality then we would not be doing the change, right?

It does change functionality, because:

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

even though it's a bug fix, it affects how the SD flags are interpreted and acted upon by the load balancer - i.e. the change can impact performance.

Ingo

#5 lore raw
Quoted 23 lines · Ingo Molnar
Quoted 20 lines · Vaidyanathan Srinivasan
Quoted 17 lines · Ingo Molnar
Quoted 11 lines · Vaidyanathan Srinivasan

Hi,

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

pulled_task is initialised to -1 and checked for non-zero which is always true if the lowest level sched_domain does not have SD_BALANCE_NEWIDLE flag set.

Trivial fix to initialise pulled_task to zero.
Patch against 2.6.28-rc7

applied to tip/sched/core, thanks! (Not for v2.6.28 because this could affect performance.)

Quoted 7 lines
Quoted 4 lines

Thanks Ingo. This patch does not change any functionality in v2.6.28 and hence will not affect performance. The SD flags are not touched. I found this bug while setting different SD flags at MC level and CPU level in my power saving balance patches.

if it does not change any functionality then we would not be doing the change, right?

:)

Quoted 9 lines

It does change functionality, because:

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is set at higher level domain (3-CPU) and not in low level domain (2-MC).

even though it's a bug fix, it affects how the SD flags are interpreted and acted upon by the load balancer - i.e. the change can impact performance.

Agreed. In my test setup with two levels of sched_domains, SD_BALANCE_NEWIDLE has not been set in both CPU level and MC level starting from 2.6.28-rc4. In a setup with more levels of sched_domains and if the higher levels have SD_BALANCE_NEWIDLE set, then we will have a functional impact.

You are right in queueing it for 2.6.29.

Thanks for the clarification.

--Vaidy