Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
From:   Joel Fernandes <joel@joelfernandes.org>
Mime-Version: 1.0 (1.0)
Subject: Re: [PATCH -rcu] rcu: Disable lazy if call_rcu() called when GPs expedited
Date:   Mon, 9 Jan 2023 18:20:55 -0500
Message-Id: <C461EF52-B060-4871-8C20-30824983E787@joelfernandes.org>
References: <20230109231416.GA4028633@paulmck-ThinkPad-P17-Gen-1>
Cc:     linux-kernel@vger.kernel.org,
        Josh Triplett <josh@joshtriplett.org>,
        Lai Jiangshan <jiangshanlai@gmail.com>,
        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
        rcu@vger.kernel.org, Steven Rostedt <rostedt@goodmis.org>,
        foo@bar.com
In-Reply-To: <20230109231416.GA4028633@paulmck-ThinkPad-P17-Gen-1>
To:     paulmck@kernel.org
Precedence: bulk


> On Jan 9, 2023, at 6:14 PM, Paul E. McKenney <paulmck@kernel.org> wrote:
>=20
> =EF=BB=BFOn Mon, Jan 09, 2023 at 10:17:56PM +0000, Joel Fernandes (Google)=
 wrote:
>> During suspend, we see failures to suspend 1 in 300-500 suspends.
>> Looking closer, it appears that we are queuing lazy callbacks even
>> though rcu_gp_is_expedited(). These delays appear to not be very welcome
>> by the suspend/resume code as evidenced by these occasional suspend
>> failures.
>>=20
>> This commit therefore checks if rcu_gp_is_expedited() and ignores the
>> lazy hint if so.
>>=20
>> Ignoring the lazy hint if rcu_gp_is_expedited() makes the 3000
>> suspend/resume cycles pass reliably on a 12th gen 12-core Intel CPU.
>=20
> Yow!!!  ;-)

:-D

>> Fixes: 3cb278e73be5 ("rcu: Make call_rcu() lazy to save power")
>> Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
>> ---
>> Paul, could we take this for 6.2 -rc cycle? Thanks.
>>=20
>> kernel/rcu/tree.c | 5 +++--
>> 1 file changed, 3 insertions(+), 2 deletions(-)
>>=20
>> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
>> index 63545d79da51..93eb03f8ed99 100644
>> --- a/kernel/rcu/tree.c
>> +++ b/kernel/rcu/tree.c
>> @@ -2594,12 +2594,12 @@ static void check_cb_ovld(struct rcu_data *rdp)
>> }
>>=20
>> static void
>> -__call_rcu_common(struct rcu_head *head, rcu_callback_t func, bool lazy)=

>> +__call_rcu_common(struct rcu_head *head, rcu_callback_t func, bool lazy_=
in)
>> {
>>    static atomic_t doublefrees;
>>    unsigned long flags;
>>    struct rcu_data *rdp;
>> -    bool was_alldone;
>> +    bool was_alldone, lazy;
>=20
> Please put "lazy" in alpha order.  Except that...

Ah sure.

>=20
>>    /* Misaligned rcu_head! */
>>    WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));
>> @@ -2622,6 +2622,7 @@ __call_rcu_common(struct rcu_head *head, rcu_callba=
ck_t func, bool lazy)
>>    kasan_record_aux_stack_noalloc(head);
>>    local_irq_save(flags);
>>    rdp =3D this_cpu_ptr(&rcu_data);
>> +    lazy =3D lazy_in && !rcu_gp_is_expedited();
>=20
> Doesn't this completely disable laziness on Android?

Good point, I am not sure but it could be. Maybe it is safer that I add
a new suspend-indicator then, with corresponding
suspend entry/exit calls like we do for expedited.

That way anyone doing it this way will not disable
lazy fully.

Thoughts?

Thanks!

 - Joel=20


>=20
>                            Thanx, Paul
>=20
>>    /* Add the callback to our list. */
>>    if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist))) {
>> --=20
>> 2.39.0.314.g84b9a713c41-goog