On Tue, Nov 21, 2023 at 04:38:34PM -0500, Steven Rostedt wrote:
> On Tue, 21 Nov 2023 13:14:16 -0800
> "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
>
> > On Tue, Nov 21, 2023 at 09:30:49PM +0100, Peter Zijlstra wrote:
> > > On Tue, Nov 21, 2023 at 11:25:18AM -0800, Paul E. McKenney wrote:
> > > > #define preempt_enable() \
> > > > do { \
> > > > barrier(); \
> > > > if (!IS_ENABLED(CONFIG_PREEMPT_RCU) && raw_cpu_read(rcu_data.rcu_urgent_qs) && \
> > > > (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK | HARDIRQ_MASK | NMI_MASK) == PREEMPT_OFFSET) &&
> > > > !irqs_disabled()) \
>
> Could we make the above an else case of the below if ?
Wouldn't that cause the above preempt_count() test to always fail?
Another approach is to bury the test in preempt_count_dec_and_test(),
but I suspect that this would not make Peter any more happy than my
earlier suggestion. ;-)
> > > > rcu_all_qs(); \
> > > > if (unlikely(preempt_count_dec_and_test())) { \
> > > > __preempt_schedule(); \
> > > > } \
> > > > } while (0)
> > >
> > > Aaaaahhh, please no. We spend so much time reducing preempt_enable() to
> > > the minimal thing it is today, this will make it blow up into something
> > > giant again.
>
> Note, the above is only true with "CONFIG_PREEMPT_RCU is not set", which
> keeps the preempt_count() for preemptable kernels with PREEMPT_RCU still minimal.
Agreed, and there is probably some workload that does not like this.
After all, current CONFIG_PREEMPT_DYNAMIC=y booted with preempt=none
would have those cond_resched() invocations. I was leary of checking
dynamic information, but maybe sched_feat() is faster than I am thinking?
(It should be with the static_branch, but not sure about the other two
access modes.)
Thanx, Paul
