lkml.org 
[lkml]   [2018]   [Jan]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH 2/2] softirq: Per vector thread deferment
On Fri, Jan 12, 2018 at 10:07:25AM +0100, Paolo Abeni wrote:
> On Fri, 2018-01-12 at 06:35 +0100, Frederic Weisbecker wrote:
> > Some softirq vectors can be more CPU hungry than others. Especially
> > networking may sometimes deal with packet storm and need more CPU than
> > IRQ tail can offer without inducing scheduler latencies. In this case
> > the current code defers to ksoftirqd that behaves nicer. Now this nice
> > behaviour can be bad for other IRQ vectors that usually need quick
> > processing.
> >
> > To solve this we only defer to threading the vectors that outreached the
> > time limit on IRQ tail processing and leave the others inline on real
> > Soft-IRQs service. This is achieved using workqueues with
> > per-CPU/per-vector worklets.
> >
> > Note ksoftirqd is not removed as it is still needed for threaded IRQs
> > mode.
> >
> > Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
> > Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
> > Cc: Dmitry Safonov <dima@arista.com>
> > Cc: Eric Dumazet <edumazet@google.com>
> > Cc: Linus Torvalds <torvalds@linux-foundation.org>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Cc: David Miller <davem@davemloft.net>
> > Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
> > Cc: Ingo Molnar <mingo@kernel.org>
> > Cc: Levin Alexander <alexander.levin@verizon.com>
> > Cc: Paolo Abeni <pabeni@redhat.com>
> > Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > Cc: Radu Rendec <rrendec@arista.com>
> > Cc: Rik van Riel <riel@redhat.com>
> > Cc: Stanislaw Gruszka <sgruszka@redhat.com>
> > Cc: Thomas Gleixner <tglx@linutronix.de>
> > Cc: Wanpeng Li <wanpeng.li@hotmail.com>
> > ---
> > kernel/softirq.c | 90 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--
> > 1 file changed, 87 insertions(+), 3 deletions(-)
> >
> > diff --git a/kernel/softirq.c b/kernel/softirq.c
> > index fa267f7..0c817ec6 100644
> > --- a/kernel/softirq.c
> > +++ b/kernel/softirq.c
> > @@ -74,6 +74,13 @@ struct softirq_stat {
> >
> > static DEFINE_PER_CPU(struct softirq_stat, softirq_stat_cpu);
> >
> > +struct vector_work {
> > + int vec;
> > + struct work_struct work;
> > +};
> > +
> > +static DEFINE_PER_CPU(struct vector_work[NR_SOFTIRQS], vector_work_cpu);
> > +
> > /*
> > * we cannot loop indefinitely here to avoid userspace starvation,
> > * but we also don't want to introduce a worst case 1/HZ latency
> > @@ -251,6 +258,70 @@ static inline bool lockdep_softirq_start(void) { return false; }
> > static inline void lockdep_softirq_end(bool in_hardirq) { }
> > #endif
> >
> > +static void vector_work_func(struct work_struct *work)
> > +{
> > + struct vector_work *vector_work;
> > + u32 pending;
> > + int vec;
> > +
> > + vector_work = container_of(work, struct vector_work, work);
> > + vec = vector_work->vec;
> > +
> > + local_irq_disable();
> > + pending = local_softirq_pending();
> > + account_irq_enter_time(current);
> > + __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
> > + lockdep_softirq_enter();
> > + set_softirq_pending(pending & ~(1 << vec));
> > + local_irq_enable();
> > +
> > + if (pending & (1 << vec)) {
> > + struct softirq_action *sa = &softirq_vec[vec];
> > +
> > + kstat_incr_softirqs_this_cpu(vec);
> > + trace_softirq_entry(vec);
> > + sa->action(sa);
> > + trace_softirq_exit(vec);
> > + }
> > +
> > + local_irq_disable();
> > +
> > + pending = local_softirq_pending();
> > + if (pending & (1 << vec))
> > + schedule_work_on(smp_processor_id(), work);
>
> If we check for the overrun condition here, as done in the
> __do_softirq() main loop, we could avoid ksoftirqd completely and
> probably have less code duplication.

Yes that could be possible indeed. I guess having workqueues serializing
vector works is not much different that what ksoftirqd does.

I can try that.

>
> > +
> > + lockdep_softirq_exit();
> > + account_irq_exit_time(current);
> > + __local_bh_enable(SOFTIRQ_OFFSET);
> > + local_irq_enable();
> > +}
> > +
> > +static int do_softirq_overrun(u32 overrun, u32 pending)
> > +{
> > + struct softirq_action *h = softirq_vec;
> > + int softirq_bit;
> > +
> > + if (!overrun)
> > + return pending;
> > +
> > + overrun &= pending;
> > + pending &= ~overrun;
> > +
> > + while ((softirq_bit = ffs(overrun))) {
> > + struct vector_work *work;
> > + unsigned int vec_nr;
> > +
> > + h += softirq_bit - 1;
> > + vec_nr = h - softirq_vec;
> > + work = this_cpu_ptr(&vector_work_cpu[vec_nr]);
> > + schedule_work_on(smp_processor_id(), &work->work);
> > + h++;
> > + overrun >>= softirq_bit;
> > + }
> > +
> > + return pending;
> > +}
> > +
> > asmlinkage __visible void __softirq_entry __do_softirq(void)
> > {
> > struct softirq_stat *sstat = this_cpu_ptr(&softirq_stat_cpu);
> > @@ -321,10 +392,13 @@ asmlinkage __visible void __softirq_entry __do_softirq(void)
> >
> > pending = local_softirq_pending();
> > if (pending) {
> > - if (overrun || need_resched())
> > + if (need_resched()) {
> > wakeup_softirqd();
> > - else
> > - goto restart;
> > + } else {
> > + pending = do_softirq_overrun(overrun, pending);
> > + if (pending)
> > + goto restart;
> > + }
> > }
> >
> > lockdep_softirq_end(in_hardirq);
>
> This way the 'overrun' branch is not triggered if we (also) need
> resched, should we test for overrun first ?

Yes they could have similar treatment. If need_resched() we schedule
everything that is still pending: do_softirq_overrun(pending, pending),
otherwise we take the other branch and still do a goto restart.

In fact it can even be simplified this way:


if (need_resched())
overrun = pending;
pending = do_softirq_overrun(overrun, pending);
if (pending)
goto restart;

Thanks.

\
 
 \ /
  Last update: 2018-01-14 23:25    [W:0.268 / U:0.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site