Missed replying to some comments.. On Tue, May 1, 2018 at 7:24 AM Steven Rostedt <rost...@goodmis.org> wrote:
> On Mon, 30 Apr 2018 18:42:03 -0700 > Joel Fernandes <joe...@google.com> wrote: > > In recent tests with IRQ on/off tracepoints, a large performance > > overhead ~10% is noticed when running hackbench. This is root caused to > > calls to rcu_irq_enter_irqson and rcu_irq_exit_irqson from the > > tracepoint code. Following a long discussion on the list [1] about this, > > we concluded that srcu is a better alternative for use during rcu idle. > > Although it does involve extra barriers, its lighter than the sched-rcu > > version which has to do additional RCU calls to notify RCU idle about > > entry into RCU sections. > > > > In this patch, we change the underlying implementation of the > > trace_*_rcuidle API to use SRCU. This has shown to improve performance > > alot for the high frequency irq enable/disable tracepoints. > Can you post some numbers? Sure, I will post them in the next revision. > > Test: Tested idle and preempt/irq tracepoints. > > > > [1] https://patchwork.kernel.org/patch/10344297/ > > [...] > > include/linux/tracepoint.h | 46 +++++++++++++++++++++++++++++++------- > > kernel/tracepoint.c | 10 ++++++++- > > 2 files changed, 47 insertions(+), 9 deletions(-) > > > > diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h > > index c94f466d57ef..4135e08fb5f1 100644 > > --- a/include/linux/tracepoint.h > > +++ b/include/linux/tracepoint.h > > @@ -15,6 +15,7 @@ > > */ > > > > #include <linux/smp.h> > > +#include <linux/srcu.h> > > #include <linux/errno.h> > > #include <linux/types.h> > > #include <linux/cpumask.h> > > @@ -33,6 +34,8 @@ struct trace_eval_map { > > > > #define TRACEPOINT_DEFAULT_PRIO 10 > > > > +extern struct srcu_struct tracepoint_srcu; > > + > > extern int > > tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); > > extern int > > @@ -77,6 +80,9 @@ int unregister_tracepoint_module_notifier(struct notifier_block *nb) > > */ > > static inline void tracepoint_synchronize_unregister(void) > > { > > +#ifdef CONFIG_TRACEPOINTS > > + synchronize_srcu(&tracepoint_srcu); > > +#endif > Not related to your patch, but I find it interesting that we don't make > this function a nop if CONFIG_TRACEPOINTS is not set. Is it because > something might rely on our implementation that we call > synchronize_sched here? I think that's a too tight of a coupling for > others to rely on this, especially since it's not in the comments about > this function. If there's no CONFIG_TRACEPOINTS, then nothing should be replying on the implementation? Basically, if !TRACEPOINTS, then there shouldn't be any active rcu read sections calling probes. > Again, not related to this series, but something we should probably > consider in the future. It would require auditing users of this too. Yes, probably could be a noop in the future. > > synchronize_sched(); > > } > > > > @@ -129,18 +135,38 @@ extern void syscall_unregfunc(void); > > * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just > > * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto". > > */ > > -#define __DO_TRACE(tp, proto, args, cond, rcucheck) \ > > +#define __DO_TRACE(tp, proto, args, cond, rcuidle) \ > > do { \ > > struct tracepoint_func *it_func_ptr; \ > > void *it_func; \ > > void *__data; \ > > + int __maybe_unused idx = 0; \ > > \ > > if (!(cond)) \ > > return; \ > > - if (rcucheck) \ > > - rcu_irq_enter_irqson(); \ > > - rcu_read_lock_sched_notrace(); \ > > - it_func_ptr = rcu_dereference_sched((tp)->funcs); \ > > + \ > > + /* \ > > + * For rcuidle callers, use srcu since sched-rcu \ > > + * doesn't work from the idle path. \ > > + */ \ > > + if (rcuidle) { \ > > + if (in_nmi()) { \ > > + WARN_ON_ONCE(1); \ > > + return; /* no srcu from nmi */ \ > > + } \ > > + \ > > + /* To keep it consistent with !rcuidle path */ \ > > + preempt_disable_notrace(); \ > Why not disable preemption after taking the srcu lock? Sure. I don't have a strong preference for either way so I could disable it after. [...] > > #ifndef MODULE > > diff --git a/kernel/tracepoint.c b/kernel/tracepoint.c > > index 671b13457387..b3b1d65a2460 100644 > > --- a/kernel/tracepoint.c > > +++ b/kernel/tracepoint.c > > @@ -31,6 +31,9 @@ > > extern struct tracepoint * const __start___tracepoints_ptrs[]; > > extern struct tracepoint * const __stop___tracepoints_ptrs[]; > > > > +DEFINE_SRCU(tracepoint_srcu); > > +EXPORT_SYMBOL_GPL(tracepoint_srcu); > > + > > /* Set to 1 to enable tracepoint debug output */ > > static const int tracepoint_debug; > > > > @@ -67,11 +70,16 @@ static inline void *allocate_probes(int count) > > return p == NULL ? NULL : p->probes; > > } > > > > -static void rcu_free_old_probes(struct rcu_head *head) > > +static void srcu_free_old_probes(struct rcu_head *head) > > { > > kfree(container_of(head, struct tp_probes, rcu)); > > } > > > > +static void rcu_free_old_probes(struct rcu_head *head) > > +{ > > + call_srcu(&tracepoint_srcu, head, srcu_free_old_probes); > Hmm, is it OK to call call_srcu() from a call_rcu() callback? I guess > it would be. > I think we should add a comment to why we are doing this. Something > like: > /* > * Tracepoint probes are protected by both sched RCU and SRCU, by > * calling the SRCU callback in the sched RCU callback we cover > * both cases. > */ > Or something along those lines. Ok I'll add these. Thanks, - Joel
