[PATCH v3 1/2] cputime: fix invalid gtime in proc

[PATCH v3 1/2] cputime: fix invalid gtime in proc

[Hiroshi Shimamoto]

From: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>

/proc/stats shows invalid gtime when the thread is running in guest.
When vtime accounting is not enabled, we cannot get a valid delta.
The delta is calculated now - tsk->vtime_snap, but tsk->vtime_snap
is only updated when vtime accounting is enabled.

This patch makes task_gtime() just return gtime when vtime accounting
is not enabled.

Use context_tracking_is_enabled() to check if the vtime accounting on
current cpu. In future we should check the state of cpu which is running
target thread.

The kernel config contains CONFIG_VIRT_CPU_ACCOUNTING_GEN=y and
CONFIG_NO_HZ_FULL_ALL=n and boot without nohz_full.

I ran and stop a busy loop in VM and see the gtime in host.
Dump the 43rd field which shows the gtime in every second.
 # while :; do awk '{print $3" "$43}' /proc/3955/task/4014/stat; sleep 1; done
S 4348
R 7064566
R 7064766
R 7064967
R 7065168
S 4759
S 4759

During running busy loop, it returns large value.

After applying this patch, we can see right gtime.

 # while :; do awk '{print $3" "$43}' /proc/10913/task/10956/stat; sleep 1; done
S 5338
R 5365
R 5465
R 5566
R 5666
S 5726
S 5726

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Cc: stable@vger.kernel.org
---
v2: Update ChangeLog to put the script and show the point.
v3: Use context_tracking_is_enabled() instead of vtime_accounting_enabled().

 kernel/sched/cputime.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index 8cbc3db..63904e7 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -786,6 +786,9 @@ cputime_t task_gtime(struct task_struct *t)
 	unsigned int seq;
 	cputime_t gtime;
 
+	if (!context_tracking_is_enabled())
+		return t->gtime;
+
 	do {
 		seq = read_seqbegin(&t->vtime_seqlock);
 
-- 
1.8.3.1

Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc

[Peter Zijlstra]

On Fri, Oct 30, 2015 at 12:46:39AM +0000, Hiroshi Shimamoto wrote:
> +++ b/kernel/sched/cputime.c
> @@ -786,6 +786,9 @@ cputime_t task_gtime(struct task_struct *t)
>  	unsigned int seq;
>  	cputime_t gtime;
>  
> +	if (!context_tracking_is_enabled())
> +		return t->gtime;
> +

Yeah, not happy about that.. why do we have to touch context tracking
muck to find vtime state etc.

>  	do {
>  		seq = read_seqbegin(&t->vtime_seqlock);
>  
> -- 
> 1.8.3.1
> 

Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc

[Frederic Weisbecker]

On Mon, Nov 02, 2015 at 05:13:51PM +0100, Peter Zijlstra wrote:
> On Fri, Oct 30, 2015 at 12:46:39AM +0000, Hiroshi Shimamoto wrote:
> > +++ b/kernel/sched/cputime.c
> > @@ -786,6 +786,9 @@ cputime_t task_gtime(struct task_struct *t)
> >  	unsigned int seq;
> >  	cputime_t gtime;
> >  
> > +	if (!context_tracking_is_enabled())
> > +		return t->gtime;
> > +
> 
> Yeah, not happy about that.. why do we have to touch context tracking
> muck to find vtime state etc.

That's right, this is because it is deemed to be a quick and non invasive fix
to be backported.

Then will come the more invasive but proper fix consisting in having
vtime_accounting_enabled() telling if vtime is running on any CPU and
vtime_accounting_cpu_enabled(). The first will be used for remote readers
(as in this patch) and the second for writers.

Since we are dealing with a regression, it's better to minimize the changes.
AFAICT, the regression got introduced in 2012:

  6a61671bb2f3a1bd12cd17b8fca811a624782632
  ("cputime: Safely read cputime of full dynticks CPUs")

> 
> >  	do {
> >  		seq = read_seqbegin(&t->vtime_seqlock);
> >  
> > -- 
> > 1.8.3.1
> > 

RE: [PATCH v3 1/2] cputime: fix invalid gtime in proc

[Hiroshi Shimamoto]

> Subject: Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc
> 
> On Mon, Nov 02, 2015 at 05:13:51PM +0100, Peter Zijlstra wrote:
> > On Fri, Oct 30, 2015 at 12:46:39AM +0000, Hiroshi Shimamoto wrote:
> > > +++ b/kernel/sched/cputime.c
> > > @@ -786,6 +786,9 @@ cputime_t task_gtime(struct task_struct *t)
> > >  	unsigned int seq;
> > >  	cputime_t gtime;
> > >
> > > +	if (!context_tracking_is_enabled())
> > > +		return t->gtime;
> > > +
> >
> > Yeah, not happy about that.. why do we have to touch context tracking
> > muck to find vtime state etc.
> 
> That's right, this is because it is deemed to be a quick and non invasive fix
> to be backported.
> 
> Then will come the more invasive but proper fix consisting in having
> vtime_accounting_enabled() telling if vtime is running on any CPU and
> vtime_accounting_cpu_enabled(). The first will be used for remote readers
> (as in this patch) and the second for writers.
> 
> Since we are dealing with a regression, it's better to minimize the changes.
> AFAICT, the regression got introduced in 2012:
> 
>   6a61671bb2f3a1bd12cd17b8fca811a624782632
>   ("cputime: Safely read cputime of full dynticks CPUs")

Is this patch going to apply to fix the regression?

thanks,
Hiroshi

> 
> >
> > >  	do {
> > >  		seq = read_seqbegin(&t->vtime_seqlock);
> > >
> > > --
> > > 1.8.3.1
> > >

Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc

[Frederic Weisbecker]

On Tue, Nov 10, 2015 at 08:43:17AM +0000, Hiroshi Shimamoto wrote:
> > Subject: Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc
> > 
> > On Mon, Nov 02, 2015 at 05:13:51PM +0100, Peter Zijlstra wrote:
> > > On Fri, Oct 30, 2015 at 12:46:39AM +0000, Hiroshi Shimamoto wrote:
> > > > +++ b/kernel/sched/cputime.c
> > > > @@ -786,6 +786,9 @@ cputime_t task_gtime(struct task_struct *t)
> > > >  	unsigned int seq;
> > > >  	cputime_t gtime;
> > > >
> > > > +	if (!context_tracking_is_enabled())
> > > > +		return t->gtime;
> > > > +
> > >
> > > Yeah, not happy about that.. why do we have to touch context tracking
> > > muck to find vtime state etc.
> > 
> > That's right, this is because it is deemed to be a quick and non invasive fix
> > to be backported.
> > 
> > Then will come the more invasive but proper fix consisting in having
> > vtime_accounting_enabled() telling if vtime is running on any CPU and
> > vtime_accounting_cpu_enabled(). The first will be used for remote readers
> > (as in this patch) and the second for writers.
> > 
> > Since we are dealing with a regression, it's better to minimize the changes.
> > AFAICT, the regression got introduced in 2012:
> > 
> >   6a61671bb2f3a1bd12cd17b8fca811a624782632
> >   ("cputime: Safely read cputime of full dynticks CPUs")
> 
> Is this patch going to apply to fix the regression?

Peter, if you have doubts, I can integrate this change in a larger
series that does a proper cleanup. But this very patch will still need
to be tagged as stable for backport.

Thanks.

Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc

[Peter Zijlstra]

On Fri, Nov 13, 2015 at 04:33:01PM +0100, Frederic Weisbecker wrote:
> Peter, if you have doubts, I can integrate this change in a larger
> series that does a proper cleanup. But this very patch will still need
> to be tagged as stable for backport.

Yeah, I want to see it done proper. I really do not care about backports
much.

Thanks!

Re: [PATCH v3 1/2] cputime: fix invalid gtime in proc

[Frederic Weisbecker]

On Fri, Nov 13, 2015 at 05:10:28PM +0100, Peter Zijlstra wrote:
> On Fri, Nov 13, 2015 at 04:33:01PM +0100, Frederic Weisbecker wrote:
> > Peter, if you have doubts, I can integrate this change in a larger
> > series that does a proper cleanup. But this very patch will still need
> > to be tagged as stable for backport.
> 
> Yeah, I want to see it done proper. I really do not care about backports
> much.

Ok, I'll post something soon.

Thanks.