Re: [PATCH v2] sched: Remove obscure comment from select_task_rq_fair

[Joel Fernandes]

On Mon, May 28, 2018 at 09:15:08AM +0200, Peter Zijlstra wrote:
> On Sun, May 27, 2018 at 06:24:47PM -0700, Joel Fernandes (Google) wrote:
> >  This patch removes the obscure comment which was
> > saying this path can be entered only for wake-balance.
> 
> The comment was meant to ask if we want to always do
> select_idle_siblings() and not only on wakeups.

Ok, thanks for the clarification. Makes sense, and sorry for the noise.

thanks,

 - Joel

Re: [PATCH v2] sched: Remove obscure comment from select_task_rq_fair

[Joel Fernandes]

On Mon, May 28, 2018 at 09:15:08AM +0200, Peter Zijlstra wrote:
> On Sun, May 27, 2018 at 06:24:47PM -0700, Joel Fernandes (Google) wrote:
> >  This patch removes the obscure comment which was
> > saying this path can be entered only for wake-balance.
> 
> The comment was meant to ask if we want to always do
> select_idle_siblings() and not only on wakeups.

Ok, thanks for the clarification. Makes sense, and sorry for the noise.

thanks,

 - Joel

Re: [PATCH v2] sched: Remove obscure comment from select_task_rq_fair

[Peter Zijlstra]

On Sun, May 27, 2018 at 06:24:47PM -0700, Joel Fernandes (Google) wrote:
>  This patch removes the obscure comment which was
> saying this path can be entered only for wake-balance.

The comment was meant to ask if we want to always do
select_idle_siblings() and not only on wakeups.

Re: [PATCH v2] sched: Remove obscure comment from select_task_rq_fair

[Peter Zijlstra]

On Sun, May 27, 2018 at 06:24:47PM -0700, Joel Fernandes (Google) wrote:
>  This patch removes the obscure comment which was
> saying this path can be entered only for wake-balance.

The comment was meant to ask if we want to always do
select_idle_siblings() and not only on wakeups.

[PATCH v2] sched: Remove obscure comment from select_task_rq_fair

[Joel Fernandes (Google)]

I was playing with cpusets and sched_load_balance flag and notice that
the fast-path (select_idle_sibling) can also be attempted for
exec-balance, not just wake-balance if the waker cpu's cpuset has
sched_load_balance = 0. This patch removes the obscure comment which was
saying this path can be entered only for wake-balance.

To trigger this, I just do:
mkdir /cpuset
mount -t cpuset none /cpuset
echo 0 > sched_load_balance

Then did some random activity and dumped the stack from 'if (!sd)' for
the non wake-balance cases.

Following is one of the stacks:
dump_stack+0x46/0x5b
select_task_rq_fair+0x101d/0x1030
sched_exec+0x4f/0xc0
do_execveat_common.isra.41+0x1e3/0x7c0
__x64_sys_execve+0x2d/0x40
do_syscall_64+0x43/0xf0
entry_SYSCALL_64_after_hwframe+0x44/0xa9

Turns out the same case occurs also during boot up when kthreadd tries
to create threads before domains are attached so lets fix the comment.

Cc: Dietmar Eggemann 
Cc: Morten Ramussen 
Cc: Ingo Molnar 
Cc: Peter Zijlstra 
Cc: Juri Lelli 
Cc: Vincent Guittot 
Cc: Patrick Bellasi 
Cc: Rohit Jain 
Cc: kernel-t...@android.com
Signed-off-by: Joel Fernandes (Google) 
---
v1->v2: Resending without "XXX" in subject since otherwise LKML thinks
its junk.

 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 54dc31e7ab9b..dd07794141d0 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -6665,7 +6665,7 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, 
int sd_flag, int wake_f
 
if (!sd) {
 pick_cpu:
-   if (sd_flag & SD_BALANCE_WAKE) { /* XXX always ? */
+   if (sd_flag & SD_BALANCE_WAKE) {
new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
 
if (want_affine)
-- 
2.17.0.921.gf22659ad46-goog

[PATCH v2] sched: Remove obscure comment from select_task_rq_fair

[Joel Fernandes (Google)]

I was playing with cpusets and sched_load_balance flag and notice that
the fast-path (select_idle_sibling) can also be attempted for
exec-balance, not just wake-balance if the waker cpu's cpuset has
sched_load_balance = 0. This patch removes the obscure comment which was
saying this path can be entered only for wake-balance.

To trigger this, I just do:
mkdir /cpuset
mount -t cpuset none /cpuset
echo 0 > sched_load_balance

Then did some random activity and dumped the stack from 'if (!sd)' for
the non wake-balance cases.

Following is one of the stacks:
dump_stack+0x46/0x5b
select_task_rq_fair+0x101d/0x1030
sched_exec+0x4f/0xc0
do_execveat_common.isra.41+0x1e3/0x7c0
__x64_sys_execve+0x2d/0x40
do_syscall_64+0x43/0xf0
entry_SYSCALL_64_after_hwframe+0x44/0xa9

Turns out the same case occurs also during boot up when kthreadd tries
to create threads before domains are attached so lets fix the comment.

Cc: Dietmar Eggemann 
Cc: Morten Ramussen 
Cc: Ingo Molnar 
Cc: Peter Zijlstra 
Cc: Juri Lelli 
Cc: Vincent Guittot 
Cc: Patrick Bellasi 
Cc: Rohit Jain 
Cc: kernel-t...@android.com
Signed-off-by: Joel Fernandes (Google) 
---
v1->v2: Resending without "XXX" in subject since otherwise LKML thinks
its junk.

 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 54dc31e7ab9b..dd07794141d0 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -6665,7 +6665,7 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, 
int sd_flag, int wake_f
 
if (!sd) {
 pick_cpu:
-   if (sd_flag & SD_BALANCE_WAKE) { /* XXX always ? */
+   if (sd_flag & SD_BALANCE_WAKE) {
new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
 
if (want_affine)
-- 
2.17.0.921.gf22659ad46-goog