Message-ID: <CACS=qqKgiwdCR_5+z-vkZ0X8DfzOPD7_ooJ_imeBnx+X1zw2qg@mail.gmail.com>
Date:   Wed, 26 Aug 2020 10:38:43 +0800
From:   Kehuan Feng <kehuan.feng@...il.com>
To:     Hillf Danton <hdanton@...a.com>
Cc:     Jike Song <albcamus@...il.com>, Josh Hunt <johunt@...mai.com>,
        Paolo Abeni <pabeni@...hat.com>,
        Jonas Bonn <jonas.bonn@...rounds.com>,
        Cong Wang <xiyou.wangcong@...il.com>,
        Michael Zhivich <mzhivich@...mai.com>,
        David Miller <davem@...emloft.net>,
        John Fastabend <john.fastabend@...il.com>,
        LKML <linux-kernel@...r.kernel.org>,
        Netdev <netdev@...r.kernel.org>
Subject: Re: Packet gets stuck in NOLOCK pfifo_fast qdisc

Hi Hillf,

Thanks for the patch.
I just tried it and it looks better than previous one. The issue
appeared only once over ~30 mins stressing (without the patch , it
shows up within 1 mins in usual, so I feel like we are getting close
to the final fix)
(pasted the modifications on my tree in case of any missing)

--- ./include/net/sch_generic.h.orig 2020-08-21 15:13:51.787952710 +0800
+++ ./include/net/sch_generic.h 2020-08-26 09:41:04.647173869 +0800
@@ -79,6 +79,7 @@
 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
+ int                     pkt_seq;
  u32 limit;
  const struct Qdisc_ops *ops;
  struct qdisc_size_table __rcu *stab;
--- ./include/net/pkt_sched.h.orig 2020-08-21 15:13:51.787952710 +0800
+++ ./include/net/pkt_sched.h 2020-08-26 09:42:14.491377514 +0800
@@ -117,8 +117,15 @@
 static inline void qdisc_run(struct Qdisc *q)
 {
  if (qdisc_run_begin(q)) {
+ q->pkt_seq = 0;
+
  __qdisc_run(q);
  qdisc_run_end(q);
+
+ /* reschedule qdisc if there are packets enqueued */
+ if (q->pkt_seq != 0)
+ __netif_schedule(q);
+
  }
 }

--- ./net/core/dev.c.orig 2020-03-19 16:31:27.000000000 +0800
+++ ./net/core/dev.c 2020-08-26 09:47:57.783165885 +0800
@@ -2721,6 +2721,7 @@

  local_irq_save(flags);
  sd = this_cpu_ptr(&softnet_data);
+ q->pkt_seq = 0;
  q->next_sched = NULL;
  *sd->output_queue_tailp = q;
  sd->output_queue_tailp = &q->next_sched;
--- ./net/sched/sch_generic.c.orig 2020-08-24 22:02:04.589830751 +0800
+++ ./net/sched/sch_generic.c 2020-08-26 09:43:40.987852551 +0800
@@ -403,6 +403,9 @@
  */
  quota -= packets;
  if (quota <= 0 || need_resched()) {
+ /* info caller to reschedule qdisc outside q->seqlock */
+ q->pkt_seq = 1;
+
  __netif_schedule(q);
  break;
  }


Hillf Danton <hdanton@...a.com> 于2020年8月26日周三 上午12:26写道：
>
>
> Hi Feng,
>
> On Tue, 25 Aug 2020 15:14:12 +0800 Fengkehuan Feng wrote:
> > Hi Hillf,
> >
> > I just tried the updated version and the system can boot up now.
>
> Thanks again for your testing.
>
> > It does mitigate the issue a lot but still couldn't get rid of it
> > thoroughly. It seems to me like the effect of Cong's patch.
>
> Your echoes show we're still march in the dark and let's try another
> direction in which qdisc is rescheduled outside seqlock to make sure
> tx softirq is raised when there're more packets on the pfifo_fast to
> be transmitted.
>
>         CPU0                            CPU1
>         ----                            ----
>         seqlock
>         test __QDISC_STATE_SCHED
>                 raise tx softirq
>                                         clear __QDISC_STATE_SCHED
>                                         try seqlock
>                                         __qdisc_run(q);
>                                         sequnlock
>         sequnlock
>
>
> --- a/include/net/sch_generic.h
> +++ b/include/net/sch_generic.h
> @@ -79,6 +79,7 @@ struct Qdisc {
>  #define TCQ_F_INVISIBLE                0x80 /* invisible by default in dump */
>  #define TCQ_F_NOLOCK           0x100 /* qdisc does not require locking */
>  #define TCQ_F_OFFLOADED                0x200 /* qdisc is offloaded to HW */
> +       int                     pkt_seq;
>         u32                     limit;
>         const struct Qdisc_ops  *ops;
>         struct qdisc_size_table __rcu *stab;
> --- a/include/net/pkt_sched.h
> +++ b/include/net/pkt_sched.h
> @@ -118,6 +118,8 @@ void __qdisc_run(struct Qdisc *q);
>  static inline void qdisc_run(struct Qdisc *q)
>  {
>         if (qdisc_run_begin(q)) {
> +               q->pkt_seq = 0;
> +
>                 /* NOLOCK qdisc must check 'state' under the qdisc seqlock
>                  * to avoid racing with dev_qdisc_reset()
>                  */
> @@ -125,6 +127,10 @@ static inline void qdisc_run(struct Qdis
>                     likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
>                         __qdisc_run(q);
>                 qdisc_run_end(q);
> +
> +               /* reschedule qdisc if there are packets enqueued */
> +               if (q->pkt_seq != 0)
> +                       __netif_schedule(q);
>         }
>  }
>
> --- a/net/sched/sch_generic.c
> +++ b/net/sched/sch_generic.c
> @@ -384,6 +384,8 @@ void __qdisc_run(struct Qdisc *q)
>         while (qdisc_restart(q, &packets)) {
>                 quota -= packets;
>                 if (quota <= 0) {
> +                       /* info caller to reschedule qdisc outside q->seqlock */
> +                       q->pkt_seq = 1;
>                         __netif_schedule(q);
>                         break;
>                 }
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -3031,6 +3031,7 @@ static void __netif_reschedule(struct Qd
>
>         local_irq_save(flags);
>         sd = this_cpu_ptr(&softnet_data);
> +       q->pkt_seq = 0;
>         q->next_sched = NULL;
>         *sd->output_queue_tailp = q;
>         sd->output_queue_tailp = &q->next_sched;
> --
>

Powered by blists - more mailing lists