Re: [PATCH] proc/schedstat: Expose /proc/<pid>/schedstat if delay accounting is enabled

[Ingo Molnar <mingo@kernel.org>]

* Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> wrote:

> On 2015/05/29 10:04AM, Ingo Molnar wrote:
> > 
> > * Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> wrote:
> > 
> > > /proc/<pid>/schedstat is currently only available if CONFIG_SCHEDSTATS is
> > > enabled. But, all the fields that this exposes are available and valid
> > > if CONFIG_TASK_DELAY_ACCT is enabled as well.
> > > 
> > > Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
> > > ---
> > >  fs/proc/base.c | 11 +++++++----
> > >  1 file changed, 7 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/fs/proc/base.c b/fs/proc/base.c
> > > index 093ca14..3ece303 100644
> > > --- a/fs/proc/base.c
> > > +++ b/fs/proc/base.c
> > > @@ -304,14 +304,17 @@ static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
> > >  }
> > >  #endif
> > >  
> > > +#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
> > > +#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
> > > +#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
> > 
> 
> Hi Ingo,
> Thanks for the review.
> 
> > So such #ifdef parades are ugly and are usually a sign of some problem with the 
> > patch - as in this case.
> > 
> > But there's deeper problems as well:
> > 
> > /*
> >  * Provides /proc/PID/schedstat
> >  */
> > static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
> >                               struct pid *pid, struct task_struct *task)
> > {
> >         seq_printf(m, "%llu %llu %lu\n",
> >                    (unsigned long long)task->se.sum_exec_runtime,
> >                    (unsigned long long)task->sched_info.run_delay,
> >                    task->sched_info.pcount);
> > 
> >         return 0;
> > }
> > 
> >  - The sum_exec_runtime field is available unconditionally.
> > 
> >  - But the sched_info.run_delay field is only maintained if CONFIG_SCHEDSTATS is
> >    enabled.
> > 
> >  - Also, the sched_info.pcount field is again only maintained if CONFIG_SCHEDSTATS 
> >    is enabled.
> 
> I may be missing something, but from my reading of the code, the above 
> are maintained if any one of CONFIG_SCHEDSTATS or CONFIG_TASK_DELAY_ACCT 
> are enabled (from kernel/sched/stats.h).

Hm, indeed - I mis-read the rq-specific code - sorry.

So all this should really be cleaned up:

include/linux/sched.h:#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
include/linux/sched.h:#endif /* defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) */
include/linux/sched.h:#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
kernel/sched/core.c:#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
kernel/sched/stats.h:#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
kernel/sched/stats.h:#endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */

by introducing an intermediate Kconfig variable, named CONFIG_SCHED_INFO or so, 
and selected by both SCHEDSTATS and TASK_DELAY_ACCT.

Please make it two patches: the first one adds CONFIG_SCHED_INFO and cleans up the 
code to use it, the second one uses it for the procps change.

Thanks,

	Ingo