Re: [PATCH] net: raise RCU qs after each threaded NAPI poll

[Yan Zhai <yan@cloudflare.com>]

On Wed, Feb 28, 2024 at 9:10 AM Paul E. McKenney <paulmck@kernel.org> wrote:
>
> On Wed, Feb 28, 2024 at 12:50:53PM +0100, Toke Høiland-Jørgensen wrote:
> > "Paul E. McKenney" <paulmck@kernel.org> writes:
> >
> > > On Tue, Feb 27, 2024 at 05:44:17PM +0100, Eric Dumazet wrote:
> > >> On Tue, Feb 27, 2024 at 4:44 PM Yan Zhai <yan@cloudflare.com> wrote:
> > >> >
> > >> > We noticed task RCUs being blocked when threaded NAPIs are very busy in
> > >> > production: detaching any BPF tracing programs, i.e. removing a ftrace
> > >> > trampoline, will simply block for very long in rcu_tasks_wait_gp. This
> > >> > ranges from hundreds of seconds to even an hour, severely harming any
> > >> > observability tools that rely on BPF tracing programs. It can be
> > >> > easily reproduced locally with following setup:
> > >> >
> > >> > ip netns add test1
> > >> > ip netns add test2
> > >> >
> > >> > ip -n test1 link add veth1 type veth peer name veth2 netns test2
> > >> >
> > >> > ip -n test1 link set veth1 up
> > >> > ip -n test1 link set lo up
> > >> > ip -n test2 link set veth2 up
> > >> > ip -n test2 link set lo up
> > >> >
> > >> > ip -n test1 addr add 192.168.1.2/31 dev veth1
> > >> > ip -n test1 addr add 1.1.1.1/32 dev lo
> > >> > ip -n test2 addr add 192.168.1.3/31 dev veth2
> > >> > ip -n test2 addr add 2.2.2.2/31 dev lo
> > >> >
> > >> > ip -n test1 route add default via 192.168.1.3
> > >> > ip -n test2 route add default via 192.168.1.2
> > >> >
> > >> > for i in `seq 10 210`; do
> > >> >  for j in `seq 10 210`; do
> > >> >     ip netns exec test2 iptables -I INPUT -s 3.3.$i.$j -p udp --dport 5201
> > >> >  done
> > >> > done
> > >> >
> > >> > ip netns exec test2 ethtool -K veth2 gro on
> > >> > ip netns exec test2 bash -c 'echo 1 > /sys/class/net/veth2/threaded'
> > >> > ip netns exec test1 ethtool -K veth1 tso off
> > >> >
> > >> > Then run an iperf3 client/server and a bpftrace script can trigger it:
> > >> >
> > >> > ip netns exec test2 iperf3 -s -B 2.2.2.2 >/dev/null&
> > >> > ip netns exec test1 iperf3 -c 2.2.2.2 -B 1.1.1.1 -u -l 1500 -b 3g -t 100 >/dev/null&
> > >> > bpftrace -e 'kfunc:__napi_poll{@=count();} interval:s:1{exit();}'
> > >> >
> > >> > Above reproduce for net-next kernel with following RCU and preempt
> > >> > configuraitons:
> > >> >
> > >> > # RCU Subsystem
> > >> > CONFIG_TREE_RCU=y
> > >> > CONFIG_PREEMPT_RCU=y
> > >> > # CONFIG_RCU_EXPERT is not set
> > >> > CONFIG_SRCU=y
> > >> > CONFIG_TREE_SRCU=y
> > >> > CONFIG_TASKS_RCU_GENERIC=y
> > >> > CONFIG_TASKS_RCU=y
> > >> > CONFIG_TASKS_RUDE_RCU=y
> > >> > CONFIG_TASKS_TRACE_RCU=y
> > >> > CONFIG_RCU_STALL_COMMON=y
> > >> > CONFIG_RCU_NEED_SEGCBLIST=y
> > >> > # end of RCU Subsystem
> > >> > # RCU Debugging
> > >> > # CONFIG_RCU_SCALE_TEST is not set
> > >> > # CONFIG_RCU_TORTURE_TEST is not set
> > >> > # CONFIG_RCU_REF_SCALE_TEST is not set
> > >> > CONFIG_RCU_CPU_STALL_TIMEOUT=21
> > >> > CONFIG_RCU_EXP_CPU_STALL_TIMEOUT=0
> > >> > # CONFIG_RCU_TRACE is not set
> > >> > # CONFIG_RCU_EQS_DEBUG is not set
> > >> > # end of RCU Debugging
> > >> >
> > >> > CONFIG_PREEMPT_BUILD=y
> > >> > # CONFIG_PREEMPT_NONE is not set
> > >> > CONFIG_PREEMPT_VOLUNTARY=y
> > >> > # CONFIG_PREEMPT is not set
> > >> > CONFIG_PREEMPT_COUNT=y
> > >> > CONFIG_PREEMPTION=y
> > >> > CONFIG_PREEMPT_DYNAMIC=y
> > >> > CONFIG_PREEMPT_RCU=y
> > >> > CONFIG_HAVE_PREEMPT_DYNAMIC=y
> > >> > CONFIG_HAVE_PREEMPT_DYNAMIC_CALL=y
> > >> > CONFIG_PREEMPT_NOTIFIERS=y
> > >> > # CONFIG_DEBUG_PREEMPT is not set
> > >> > # CONFIG_PREEMPT_TRACER is not set
> > >> > # CONFIG_PREEMPTIRQ_DELAY_TEST is not set
> > >> >
> > >> > An interesting observation is that, while tasks RCUs are blocked,
> > >> > related NAPI thread is still being scheduled (even across cores)
> > >> > regularly. Looking at the gp conditions, I am inclining to cond_resched
> > >> > after each __napi_poll being the problem: cond_resched enters the
> > >> > scheduler with PREEMPT bit, which does not account as a gp for tasks
> > >> > RCUs. Meanwhile, since the thread has been frequently resched, the
> > >> > normal scheduling point (no PREEMPT bit, accounted as a task RCU gp)
> > >> > seems to have very little chance to kick in. Given the nature of "busy
> > >> > polling" program, such NAPI thread won't have task->nvcsw or task->on_rq
> > >> > updated (other gp conditions), the result is that such NAPI thread is
> > >> > put on RCU holdouts list for indefinitely long time.
> > >> >
> > >> > This is simply fixed by mirroring the ksoftirqd behavior: after
> > >> > NAPI/softirq work, raise a RCU QS to help expedite the RCU period. No
> > >> > more blocking afterwards for the same setup.
> > >> >
> > >> > Fixes: 29863d41bb6e ("net: implement threaded-able napi poll loop support")
> > >> > Signed-off-by: Yan Zhai <yan@cloudflare.com>
> > >> > ---
> > >> >  net/core/dev.c | 4 ++++
> > >> >  1 file changed, 4 insertions(+)
> > >> >
> > >> > diff --git a/net/core/dev.c b/net/core/dev.c
> > >> > index 275fd5259a4a..6e41263ff5d3 100644
> > >> > --- a/net/core/dev.c
> > >> > +++ b/net/core/dev.c
> > >> > @@ -6773,6 +6773,10 @@ static int napi_threaded_poll(void *data)
> > >> >                                 net_rps_action_and_irq_enable(sd);
> > >> >                         }
> > >> >                         skb_defer_free_flush(sd);
> > >
> > > Please put a comment here stating that RCU readers cannot cross
> > > this point.
> > >
> > > I need to add lockdep to rcu_softirq_qs() to catch placing this in an
> > > RCU read-side critical section.  And a header comment noting that from
> > > an RCU perspective, it acts as a momentary enabling of preemption.
> >
> > OK, so one question here: for XDP, we're basically treating
> > local_bh_disable/enable() as the RCU critical section, cf the discussion
> > we had a few years ago that led to this being documented[0]. So why is
> > it OK to have the rcu_softirq_qs() inside the bh disable/enable pair,
> > but not inside an rcu_read_lock() section?
>
> In general, it is not OK.  And it is not OK in this case if this happens
> to be one of the local_bh_disable() regions that XDP is waiting on.
> Except that that region ends right after the rcu_softirq_qs(), so that
> should not be a problem.
>
> But you are quite right, that is an accident waiting to happen, so it
> would be better if the patch did something like this:
>
>         local_bh_enable();
>         if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
>                 preempt_disable();
>                 rcu_softirq_qs();
>                 preempt_enable();
>         }
>
Yeah we need preempt for this call. When I first attempt it after
local_bh_enable, I got the bug call:
[ 1166.384279] BUG: using __this_cpu_read() in preemptible [00000000]
code: napi/veth2-66/8439
[ 1166.385337] caller is rcu_softirq_qs+0x16/0x130
[ 1166.385900] CPU: 3 PID: 8439 Comm: napi/veth2-66 Not tainted
6.7.0-rc8-g3fbf61207c66-dirty #75
[ 1166.386950] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS 1.16.2-debian-1.16.2-1 04/01/2014
[ 1166.388110] Call Trace:
[ 1166.388417]  <TASK>
[ 1166.388684]  dump_stack_lvl+0x36/0x50
[ 1166.389147]  check_preemption_disabled+0xd1/0xe0
[ 1166.389725]  rcu_softirq_qs+0x16/0x130
[ 1166.390190]  napi_threaded_poll+0x21e/0x260
[ 1166.390702]  ? __pfx_napi_threaded_poll+0x10/0x10
[ 1166.391277]  kthread+0xf7/0x130
[ 1166.391643]  ? __pfx_kthread+0x10/0x10
[ 1166.392130]  ret_from_fork+0x34/0x50
[ 1166.392574]  ? __pfx_kthread+0x10/0x10
[ 1166.393048]  ret_from_fork_asm+0x1b/0x30
[ 1166.393530]  </TASK>

Since this patch is trying to mirror what __do_softirq has, should the
similar notes/changes apply to that side as well?


> Though maybe something like this would be better:
>
>         local_bh_enable();
>         if (!IS_ENABLED(CONFIG_PREEMPT_RT))
>                 rcu_softirq_qs_enable(local_bh_enable());
>         else
>                 local_bh_enable();
>
> A bit ugly, but it does allow exact checking of the rules and also
> avoids extra overhead.
>
> I could imagine pulling the CONFIG_PREEMPT_RT check into the body of
> rcu_softirq_qs_enable().
>
> But is there a better way?
>
> > Also, looking at the patch in question:
> >
> > >> > +                       if (!IS_ENABLED(CONFIG_PREEMPT_RT))
> > >> > +                               rcu_softirq_qs();
> > >> > +
> > >> >                         local_bh_enable();
> >
> > Why does that local_bh_enable() not accomplish the same thing as the qs?
>
> In this case, because it does not create the appearance of a voluntary
> context switch needed by RCU Tasks.  So the wait for trampoline evacuation
> could still take a very long time.
>
>                                                         Thanx, Paul
>
> > -Toke
> >
> > [0] https://lore.kernel.org/bpf/20210624160609.292325-6-toke@redhat.com/
> >