On Wed, Oct 16, 2019 at 12:24:09PM +0800, Lai Jiangshan wrote:
>
>
> On 2019/10/16 11:38 上午, Paul E. McKenney wrote:
> > On Tue, Oct 15, 2019 at 10:23:57AM +0000, Lai Jiangshan wrote:
> > > "rcu_wait" is incorrct here, use "rcu_run" instead.
> > >
> > > Signed-off-by: Lai Jiangshan <jiangshan...@gmail.com>
> > > Signed-off-by: Lai Jiangshan <la...@linux.alibaba.com>
> > > ---
> > > kernel/rcu/tree.c | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > > index 278798e58698..c351fc280945 100644
> > > --- a/kernel/rcu/tree.c
> > > +++ b/kernel/rcu/tree.c
> > > @@ -2485,7 +2485,7 @@ static void rcu_cpu_kthread(unsigned int cpu)
> > > int spincnt;
> > > for (spincnt = 0; spincnt < 10; spincnt++) {
> > > - trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait"));
> > > + trace_rcu_utilization(TPS("Start CPU kthread@rcu_run"));
> > > local_bh_disable();
> > > *statusp = RCU_KTHREAD_RUNNING;
> > > local_irq_disable();
> > > @@ -2496,7 +2496,7 @@ static void rcu_cpu_kthread(unsigned int cpu)
> > > rcu_core();
> > > local_bh_enable();
> > > if (*workp == 0) {
> > > - trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
> > > + trace_rcu_utilization(TPS("End CPU kthread@rcu_run"));
> >
> > This one needs to stay as it was because this is where we wait when out
> > of work.
>
> I don't fully understand those TPS marks.
>
> If it is all about "where we wait when out of work", it ought to
> be "Start ... wait", rather than "End ... wait". The later one
> ("End ... wait") should be put before
> "for (spincnt = 0; spincnt < 10; spincnt++)" and remove
> the whole "rcu_run" as this patch suggested. To be honest,
> "rcu_run" is redundant since we already has TPS("Start RCU core").
>
> Any ways, patch2&3 lose their relevance and should be dropped.
> Looking forward to your improved version.
Given that most of RCU's overhead is now in kthreads and in RCU_SOFTIRQ,
perhaps trace_rcu_utilization() has outlived its usefulness, especially
given the prospect of an RCU_SOFTIRQ-specific kthread.
Thanx, Paul
