* [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently.
@ 2020-04-16 14:18 Muchun Song
2020-04-16 15:35 ` Steven Rostedt
0 siblings, 1 reply; 5+ messages in thread
From: Muchun Song @ 2020-04-16 14:18 UTC (permalink / raw)
To: mingo, peterz, juri.lelli, vincent.guittot, dietmar.eggemann,
rostedt, bsegall, mgorman, mingo
Cc: linux-kernel, Muchun Song
The user_mode(task_pt_regs(tsk)) always return true for
user thread, and false for kernel thread. So it means that
the cpuacct.usage_sys is the time that kernel thread uses
not the time that thread uses in the kernel mode. We can
use get_irq_regs() instead of task_pt_regs() to fix it.
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
---
kernel/sched/cpuacct.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
index 6448b0438ffb2..edfc62554648e 100644
--- a/kernel/sched/cpuacct.c
+++ b/kernel/sched/cpuacct.c
@@ -5,6 +5,7 @@
* Based on the work by Paul Menage (menage@google.com) and Balbir Singh
* (balbir@in.ibm.com).
*/
+#include <asm/irq_regs.h>
#include "sched.h"
/* Time spent by the tasks of the CPU accounting group executing in ... */
@@ -339,7 +340,7 @@ void cpuacct_charge(struct task_struct *tsk, u64 cputime)
{
struct cpuacct *ca;
int index = CPUACCT_STAT_SYSTEM;
- struct pt_regs *regs = task_pt_regs(tsk);
+ struct pt_regs *regs = get_irq_regs();
if (regs && user_mode(regs))
index = CPUACCT_STAT_USER;
--
2.11.0
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently.
2020-04-16 14:18 [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently Muchun Song
@ 2020-04-16 15:35 ` Steven Rostedt
2020-04-16 16:01 ` Steven Rostedt
2020-04-17 3:07 ` Muchun Song
0 siblings, 2 replies; 5+ messages in thread
From: Steven Rostedt @ 2020-04-16 15:35 UTC (permalink / raw)
To: Muchun Song
Cc: mingo, peterz, juri.lelli, vincent.guittot, dietmar.eggemann,
bsegall, mgorman, mingo, linux-kernel
On Thu, 16 Apr 2020 22:18:33 +0800
Muchun Song <songmuchun@bytedance.com> wrote:
> The user_mode(task_pt_regs(tsk)) always return true for
> user thread, and false for kernel thread. So it means that
> the cpuacct.usage_sys is the time that kernel thread uses
> not the time that thread uses in the kernel mode. We can
> use get_irq_regs() instead of task_pt_regs() to fix it.
>
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> ---
> kernel/sched/cpuacct.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
> index 6448b0438ffb2..edfc62554648e 100644
> --- a/kernel/sched/cpuacct.c
> +++ b/kernel/sched/cpuacct.c
> @@ -5,6 +5,7 @@
> * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
> * (balbir@in.ibm.com).
> */
> +#include <asm/irq_regs.h>
> #include "sched.h"
>
> /* Time spent by the tasks of the CPU accounting group executing in ... */
> @@ -339,7 +340,7 @@ void cpuacct_charge(struct task_struct *tsk, u64 cputime)
> {
> struct cpuacct *ca;
> int index = CPUACCT_STAT_SYSTEM;
> - struct pt_regs *regs = task_pt_regs(tsk);
> + struct pt_regs *regs = get_irq_regs();
But get_irq_regs() is only available from interrupt context. This will be
NULL most the time, whereas the original way will have regs existing for
the task.
>
> if (regs && user_mode(regs))
> index = CPUACCT_STAT_USER;
To show this, I applied your patch then did the following:
# echo 'p:cpuacct cpuacct_charge+0x36 regs=%ax' > /sys/kernel/tracing/kprobe_events
Where I found that the test of 'regs' is %rax at offset 0x36.
# trace-cmd start -p function -l cpuacct_charge -e kprobes
# trace-cmd show
# tracer: function
#
# entries-in-buffer/entries-written: 70664/70664 #P:8
#
# _-----=> irqs-off
# / _----=> need-resched
# | / _---=> hardirq/softirq
# || / _--=> preempt-depth
# ||| / delay
# TASK-PID CPU# |||| TIMESTAMP FUNCTION
# | | | |||| | |
<...>-1720 [002] d..2 306.430302: cpuacct_charge <-update_curr
<...>-1720 [002] d..3 306.430306: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
<...>-1720 [002] dN.2 306.430321: cpuacct_charge <-update_curr
<...>-1720 [002] dN.3 306.430322: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
<...>-1720 [002] d..2 306.430355: cpuacct_charge <-update_curr
<...>-1720 [002] d..3 306.430357: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
bash-1652 [006] d.h2 306.430799: cpuacct_charge <-update_curr
bash-1652 [006] d.h3 306.430802: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0xffffaf34012abdd8
<...>-199 [005] d.h2 306.430806: cpuacct_charge <-update_curr
<...>-199 [005] d.h3 306.430809: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0xffffaf3400347c38
<...>-16 [001] d..2 306.430873: cpuacct_charge <-update_curr
<...>-16 [001] d..3 306.430875: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
<...>-199 [005] d..2 306.430936: cpuacct_charge <-update_curr
<...>-199 [005] d..3 306.430937: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
bash-1652 [006] d..2 306.430944: cpuacct_charge <-update_curr
bash-1652 [006] d..3 306.430946: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
sshd-1649 [000] d..2 306.430990: cpuacct_charge <-update_curr
sshd-1649 [000] d..3 306.430992: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
rcu_preempt-10 [006] d..2 306.432844: cpuacct_charge <-update_curr
rcu_preempt-10 [006] d..3 306.432846: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
rcu_preempt-10 [006] d..2 306.436848: cpuacct_charge <-update_curr
rcu_preempt-10 [006] d..3 306.436850: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
rcu_preempt-10 [006] d..2 306.440868: cpuacct_charge <-update_curr
rcu_preempt-10 [006] d..3 306.440871: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
rcu_preempt-10 [006] d..2 306.444867: cpuacct_charge <-update_curr
rcu_preempt-10 [006] d..3 306.444870: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
kworker/2:1-127 [002] d..2 306.446925: cpuacct_charge <-update_curr
kworker/2:1-127 [002] d..3 306.446928: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
rcu_preempt-10 [006] d..2 306.448868: cpuacct_charge <-update_curr
rcu_preempt-10 [006] d..3 306.448870: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
rcu_preempt-10 [006] d..2 306.452869: cpuacct_charge <-update_curr
rcu_preempt-10 [006] d..3 306.452872: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
The only times regs has content is from the the interrupt handler (seen as
the 'h' in the status portion of the trace.
-- Steve
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently.
2020-04-16 15:35 ` Steven Rostedt
@ 2020-04-16 16:01 ` Steven Rostedt
2020-04-17 3:11 ` [External] " Muchun Song
2020-04-17 3:07 ` Muchun Song
1 sibling, 1 reply; 5+ messages in thread
From: Steven Rostedt @ 2020-04-16 16:01 UTC (permalink / raw)
To: Muchun Song
Cc: mingo, peterz, juri.lelli, vincent.guittot, dietmar.eggemann,
bsegall, mgorman, mingo, linux-kernel
On Thu, 16 Apr 2020 11:35:02 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:
> On Thu, 16 Apr 2020 22:18:33 +0800
> Muchun Song <songmuchun@bytedance.com> wrote:
>
> > The user_mode(task_pt_regs(tsk)) always return true for
> > user thread, and false for kernel thread. So it means that
> > the cpuacct.usage_sys is the time that kernel thread uses
> > not the time that thread uses in the kernel mode. We can
> > use get_irq_regs() instead of task_pt_regs() to fix it.
> >
> > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > ---
> > kernel/sched/cpuacct.c | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
> > index 6448b0438ffb2..edfc62554648e 100644
> > --- a/kernel/sched/cpuacct.c
> > +++ b/kernel/sched/cpuacct.c
> > @@ -5,6 +5,7 @@
> > * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
> > * (balbir@in.ibm.com).
> > */
> > +#include <asm/irq_regs.h>
> > #include "sched.h"
> >
> > /* Time spent by the tasks of the CPU accounting group executing in ... */
> > @@ -339,7 +340,7 @@ void cpuacct_charge(struct task_struct *tsk, u64 cputime)
> > {
> > struct cpuacct *ca;
> > int index = CPUACCT_STAT_SYSTEM;
> > - struct pt_regs *regs = task_pt_regs(tsk);
> > + struct pt_regs *regs = get_irq_regs();
>
> But get_irq_regs() is only available from interrupt context. This will be
> NULL most the time, whereas the original way will have regs existing for
> the task.
Perhaps you want:
regs = get_irqs_regs();
if (!regs)
regs = task_pt_regs(tsk);
?
-- Steve
>
> >
> > if (regs && user_mode(regs))
> > index = CPUACCT_STAT_USER;
>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [External] Re: [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently.
2020-04-16 15:35 ` Steven Rostedt
2020-04-16 16:01 ` Steven Rostedt
@ 2020-04-17 3:07 ` Muchun Song
1 sibling, 0 replies; 5+ messages in thread
From: Muchun Song @ 2020-04-17 3:07 UTC (permalink / raw)
To: Steven Rostedt
Cc: mingo, Peter Zijlstra, juri.lelli, Vincent Guittot,
dietmar.eggemann, Benjamin Segall, mgorman, mingo, linux-kernel
On Thu, Apr 16, 2020 at 11:35 PM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Thu, 16 Apr 2020 22:18:33 +0800
> Muchun Song <songmuchun@bytedance.com> wrote:
>
> > The user_mode(task_pt_regs(tsk)) always return true for
> > user thread, and false for kernel thread. So it means that
> > the cpuacct.usage_sys is the time that kernel thread uses
> > not the time that thread uses in the kernel mode. We can
> > use get_irq_regs() instead of task_pt_regs() to fix it.
> >
> > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > ---
> > kernel/sched/cpuacct.c | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
> > index 6448b0438ffb2..edfc62554648e 100644
> > --- a/kernel/sched/cpuacct.c
> > +++ b/kernel/sched/cpuacct.c
> > @@ -5,6 +5,7 @@
> > * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
> > * (balbir@in.ibm.com).
> > */
> > +#include <asm/irq_regs.h>
> > #include "sched.h"
> >
> > /* Time spent by the tasks of the CPU accounting group executing in ... */
> > @@ -339,7 +340,7 @@ void cpuacct_charge(struct task_struct *tsk, u64 cputime)
> > {
> > struct cpuacct *ca;
> > int index = CPUACCT_STAT_SYSTEM;
> > - struct pt_regs *regs = task_pt_regs(tsk);
> > + struct pt_regs *regs = get_irq_regs();
>
> But get_irq_regs() is only available from interrupt context. This will be
> NULL most the time, whereas the original way will have regs existing for
> the task.
>
> >
> > if (regs && user_mode(regs))
> > index = CPUACCT_STAT_USER;
>
> To show this, I applied your patch then did the following:
>
> # echo 'p:cpuacct cpuacct_charge+0x36 regs=%ax' > /sys/kernel/tracing/kprobe_events
>
> Where I found that the test of 'regs' is %rax at offset 0x36.
>
> # trace-cmd start -p function -l cpuacct_charge -e kprobes
> # trace-cmd show
> # tracer: function
> #
> # entries-in-buffer/entries-written: 70664/70664 #P:8
> #
> # _-----=> irqs-off
> # / _----=> need-resched
> # | / _---=> hardirq/softirq
> # || / _--=> preempt-depth
> # ||| / delay
> # TASK-PID CPU# |||| TIMESTAMP FUNCTION
> # | | | |||| | |
> <...>-1720 [002] d..2 306.430302: cpuacct_charge <-update_curr
> <...>-1720 [002] d..3 306.430306: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> <...>-1720 [002] dN.2 306.430321: cpuacct_charge <-update_curr
> <...>-1720 [002] dN.3 306.430322: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> <...>-1720 [002] d..2 306.430355: cpuacct_charge <-update_curr
> <...>-1720 [002] d..3 306.430357: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> bash-1652 [006] d.h2 306.430799: cpuacct_charge <-update_curr
> bash-1652 [006] d.h3 306.430802: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0xffffaf34012abdd8
> <...>-199 [005] d.h2 306.430806: cpuacct_charge <-update_curr
> <...>-199 [005] d.h3 306.430809: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0xffffaf3400347c38
> <...>-16 [001] d..2 306.430873: cpuacct_charge <-update_curr
> <...>-16 [001] d..3 306.430875: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> <...>-199 [005] d..2 306.430936: cpuacct_charge <-update_curr
> <...>-199 [005] d..3 306.430937: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> bash-1652 [006] d..2 306.430944: cpuacct_charge <-update_curr
> bash-1652 [006] d..3 306.430946: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> sshd-1649 [000] d..2 306.430990: cpuacct_charge <-update_curr
> sshd-1649 [000] d..3 306.430992: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> rcu_preempt-10 [006] d..2 306.432844: cpuacct_charge <-update_curr
> rcu_preempt-10 [006] d..3 306.432846: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> rcu_preempt-10 [006] d..2 306.436848: cpuacct_charge <-update_curr
> rcu_preempt-10 [006] d..3 306.436850: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> rcu_preempt-10 [006] d..2 306.440868: cpuacct_charge <-update_curr
> rcu_preempt-10 [006] d..3 306.440871: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> rcu_preempt-10 [006] d..2 306.444867: cpuacct_charge <-update_curr
> rcu_preempt-10 [006] d..3 306.444870: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> kworker/2:1-127 [002] d..2 306.446925: cpuacct_charge <-update_curr
> kworker/2:1-127 [002] d..3 306.446928: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> rcu_preempt-10 [006] d..2 306.448868: cpuacct_charge <-update_curr
> rcu_preempt-10 [006] d..3 306.448870: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
> rcu_preempt-10 [006] d..2 306.452869: cpuacct_charge <-update_curr
> rcu_preempt-10 [006] d..3 306.452872: cpuacct: (cpuacct_charge+0x36/0x1f0) regs=0x0
>
> The only times regs has content is from the the interrupt handler (seen as
> the 'h' in the status portion of the trace.
>
> -- Steve
Thanks for your test. You are right.
--
Yours,
Muchun
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [External] Re: [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently.
2020-04-16 16:01 ` Steven Rostedt
@ 2020-04-17 3:11 ` Muchun Song
0 siblings, 0 replies; 5+ messages in thread
From: Muchun Song @ 2020-04-17 3:11 UTC (permalink / raw)
To: Steven Rostedt
Cc: mingo, Peter Zijlstra, juri.lelli, Vincent Guittot,
dietmar.eggemann, Benjamin Segall, mgorman, mingo, linux-kernel
On Fri, Apr 17, 2020 at 12:01 AM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Thu, 16 Apr 2020 11:35:02 -0400
> Steven Rostedt <rostedt@goodmis.org> wrote:
>
> > On Thu, 16 Apr 2020 22:18:33 +0800
> > Muchun Song <songmuchun@bytedance.com> wrote:
> >
> > > The user_mode(task_pt_regs(tsk)) always return true for
> > > user thread, and false for kernel thread. So it means that
> > > the cpuacct.usage_sys is the time that kernel thread uses
> > > not the time that thread uses in the kernel mode. We can
> > > use get_irq_regs() instead of task_pt_regs() to fix it.
> > >
> > > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > > ---
> > > kernel/sched/cpuacct.c | 3 ++-
> > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
> > > index 6448b0438ffb2..edfc62554648e 100644
> > > --- a/kernel/sched/cpuacct.c
> > > +++ b/kernel/sched/cpuacct.c
> > > @@ -5,6 +5,7 @@
> > > * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
> > > * (balbir@in.ibm.com).
> > > */
> > > +#include <asm/irq_regs.h>
> > > #include "sched.h"
> > >
> > > /* Time spent by the tasks of the CPU accounting group executing in ... */
> > > @@ -339,7 +340,7 @@ void cpuacct_charge(struct task_struct *tsk, u64 cputime)
> > > {
> > > struct cpuacct *ca;
> > > int index = CPUACCT_STAT_SYSTEM;
> > > - struct pt_regs *regs = task_pt_regs(tsk);
> > > + struct pt_regs *regs = get_irq_regs();
> >
> > But get_irq_regs() is only available from interrupt context. This will be
> > NULL most the time, whereas the original way will have regs existing for
> > the task.
>
> Perhaps you want:
>
> regs = get_irqs_regs();
> if (!regs)
> regs = task_pt_regs(tsk);
>
> ?
Yeah, If regs is NULL, we can get it fall back to task_pt_regs.
Does Anyone else have suggestions?
>
> -- Steve
>
> >
> > >
> > > if (regs && user_mode(regs))
> > > index = CPUACCT_STAT_USER;
> >
--
Yours,
Muchun
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2020-04-17 3:12 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-04-16 14:18 [PATCH] sched/cpuacct: Fix charge cpuacct.usage_sys incorrently Muchun Song
2020-04-16 15:35 ` Steven Rostedt
2020-04-16 16:01 ` Steven Rostedt
2020-04-17 3:11 ` [External] " Muchun Song
2020-04-17 3:07 ` Muchun Song
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).