On Tue, Feb 25, 2020 at 07:14:55PM -0800, Paul E. McKenney wrote:
> On Tue, Feb 25, 2020 at 06:24:36PM +0800, Boqun Feng wrote:
> > Hi Paul,
> >
> > On Fri, Feb 14, 2020 at 03:56:07PM -0800, paulmck@xxxxxxxxxx wrote:
> > > From: "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> > >
> > > Currently, rcu_barrier() ignores offline CPUs, However, it is possible
> > > for an offline no-CBs CPU to have callbacks queued, and rcu_barrier()
> > > must wait for those callbacks. This commit therefore makes rcu_barrier()
> > > directly invoke the rcu_barrier_func() with interrupts disabled for such
> > > CPUs. This requires passing the CPU number into this function so that
> > > it can entrain the rcu_barrier() callback onto the correct CPU's callback
> > > list, given that the code must instead execute on the current CPU.
> > >
> > > While in the area, this commit fixes a bug where the first CPU's callback
> > > might have been invoked before rcu_segcblist_entrain() returned, which
> > > would also result in an early wakeup.
> > >
> > > Fixes: 5d6742b37727 ("rcu/nocb: Use rcu_segcblist for no-CBs CPUs")
> > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > Cc: <stable@xxxxxxxxxxxxxxx> # 5.5.x
> > > ---
> > > include/trace/events/rcu.h | 1 +
> > > kernel/rcu/tree.c | 32 ++++++++++++++++++++------------
> > > 2 files changed, 21 insertions(+), 12 deletions(-)
> > >
> > > diff --git a/include/trace/events/rcu.h b/include/trace/events/rcu.h
> > > index 5e49b06..d56d54c 100644
> > > --- a/include/trace/events/rcu.h
> > > +++ b/include/trace/events/rcu.h
> > > @@ -712,6 +712,7 @@ TRACE_EVENT_RCU(rcu_torture_read,
> > > * "Begin": rcu_barrier() started.
> > > * "EarlyExit": rcu_barrier() piggybacked, thus early exit.
> > > * "Inc1": rcu_barrier() piggyback check counter incremented.
> > > + * "OfflineNoCBQ": rcu_barrier() found offline no-CBs CPU with callbacks.
> > > * "OnlineQ": rcu_barrier() found online CPU with callbacks.
> > > * "OnlineNQ": rcu_barrier() found online CPU, no callbacks.
> > > * "IRQ": An rcu_barrier_callback() callback posted on remote CPU.
> > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > > index d15041f..160643e 100644
> > > --- a/kernel/rcu/tree.c
> > > +++ b/kernel/rcu/tree.c
> > > @@ -3098,9 +3098,10 @@ static void rcu_barrier_callback(struct rcu_head *rhp)
> > > /*
> > > * Called with preemption disabled, and from cross-cpu IRQ context.
> > > */
> > > -static void rcu_barrier_func(void *unused)
> > > +static void rcu_barrier_func(void *cpu_in)
> > > {
> > > - struct rcu_data *rdp = raw_cpu_ptr(&rcu_data);
> > > + uintptr_t cpu = (uintptr_t)cpu_in;
> > > + struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
> > >
> > > rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence);
> > > rdp->barrier_head.func = rcu_barrier_callback;
> > > @@ -3127,7 +3128,7 @@ static void rcu_barrier_func(void *unused)
> > > */
> > > void rcu_barrier(void)
> > > {
> > > - int cpu;
> > > + uintptr_t cpu;
> > > struct rcu_data *rdp;
> > > unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence);
> > >
> > > @@ -3150,13 +3151,14 @@ void rcu_barrier(void)
> > > rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence);
> > >
> > > /*
> > > - * Initialize the count to one rather than to zero in order to
> > > - * avoid a too-soon return to zero in case of a short grace period
> > > - * (or preemption of this task). Exclude CPU-hotplug operations
> > > - * to ensure that no offline CPU has callbacks queued.
> > > + * Initialize the count to two rather than to zero in order
> > > + * to avoid a too-soon return to zero in case of an immediate
> > > + * invocation of the just-enqueued callback (or preemption of
> > > + * this task). Exclude CPU-hotplug operations to ensure that no
> > > + * offline non-offloaded CPU has callbacks queued.
> > > */
> > > init_completion(&rcu_state.barrier_completion);
> > > - atomic_set(&rcu_state.barrier_cpu_count, 1);
> > > + atomic_set(&rcu_state.barrier_cpu_count, 2);
> > > get_online_cpus();
> > >
> > > /*
> > > @@ -3166,13 +3168,19 @@ void rcu_barrier(void)
> > > */
> > > for_each_possible_cpu(cpu) {
> > > rdp = per_cpu_ptr(&rcu_data, cpu);
> > > - if (!cpu_online(cpu) &&
> > > + if (cpu_is_offline(cpu) &&
> > > !rcu_segcblist_is_offloaded(&rdp->cblist))
> > > continue;
> > > - if (rcu_segcblist_n_cbs(&rdp->cblist)) {
> > > + if (rcu_segcblist_n_cbs(&rdp->cblist) && cpu_online(cpu)) {
> > > rcu_barrier_trace(TPS("OnlineQ"), cpu,
> > > rcu_state.barrier_sequence);
> > > - smp_call_function_single(cpu, rcu_barrier_func, NULL, 1);
> > > + smp_call_function_single(cpu, rcu_barrier_func, (void *)cpu, 1);
> > > + } else if (cpu_is_offline(cpu)) {
> >
> > I wonder whether this should be:
> >
> > else if (rcu_segcblist_n_cbs(&rdp->cblist) && cpu_is_offline(cpu))
> >
> > ? Because I think we only want to queue the barrier call back if there
> > are callbacks for a particular CPU. Am I missing something subtle?
>
> I don't believe that you are missing anything at all!
>
> Thank you very much -- this bug would not have shown up in any validation
> setup that I am aware of. ;-)
And with additional adjustment to make tracing accurate.
Thanx, Paul
