Received-SPF: pass (google.com: domain of linux-kernel+bounces-153420-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240422082238.5784-1-xuewen.yan@unisoc.com> <CAB8ipk-LhSuMsp0DdzjEJN-4XEBT1_ri6BPH_nvxSgFZONMT2Q@mail.gmail.com>
 <20240422094157.GA34453@noisy.programming.kicks-ass.net> <CAB8ipk9M1wLXV7MWMBYGMhKbY71QqwkvchiENeWTg7JERe6kZw@mail.gmail.com>
 <20240422111744.GO30852@noisy.programming.kicks-ass.net>
In-Reply-To: <20240422111744.GO30852@noisy.programming.kicks-ass.net>
From: Xuewen Yan <xuewen.yan94@gmail.com>
Date: Mon, 22 Apr 2024 21:12:12 +0800
Message-ID: <CAB8ipk8ZBSnJfw9Ow9My-VXG1TU6DVY7mOL_i0Ejdd7GUZCLWA@mail.gmail.com>
Subject: Re: [PATCH v2] sched/eevdf: Prevent vlag from going out of bounds
 when reweight_eevdf
To: Peter Zijlstra <peterz@infradead.org>
Cc: Xuewen Yan <xuewen.yan@unisoc.com>, mingo@redhat.com, juri.lelli@redhat.com, 
	vincent.guittot@linaro.org, dietmar.eggemann@arm.com, rostedt@goodmis.org, 
	bsegall@google.com, mgorman@suse.de, bristot@redhat.com, vschneid@redhat.com, 
	yu.c.chen@intel.com, ke.wang@unisoc.com, di.shen@unisoc.com, 
	linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hi peter,

On Mon, Apr 22, 2024 at 7:17=E2=80=AFPM Peter Zijlstra <peterz@infradead.or=
g> wrote:
>
> On Mon, Apr 22, 2024 at 07:07:25PM +0800, Xuewen Yan wrote:
> > On Mon, Apr 22, 2024 at 5:42=E2=80=AFPM Peter Zijlstra <peterz@infradea=
d.org> wrote:
> > >
> > > On Mon, Apr 22, 2024 at 04:33:37PM +0800, Xuewen Yan wrote:
> > >
> > > > On the Android system, the nice value of a task will change very
> > > > frequently. The limit can also be exceeded.
> > > > Maybe the !on_rq case is still necessary.
> > > > So I'm planning to propose another patch for !on_rq case later afte=
r
> > > > careful testing locally.
> > >
> > > So the scaling is: vlag =3D vlag * old_Weight / weight
> > >
> > > But given that integer devision is truncating, you could expect repea=
ted
> > > application of such scaling would eventually decrease the vlag instea=
d
> > > of grow it.
> > >
> > > Is there perhaps an invocation of reweight_task() missing? Looking at
> >
> > Is it necessary to add reweight_task in the prio_changed_fair()?
>
> I think that's the wrong place. Note how __setscheduler_params() already
> has set_load_weight(). And all other callers of ->prio_changed() already
> seem to do set_load_weight() as well.
>
> But that idle policy thing there still looks wrong, that sets the weight
> very low but doesn't re-adjust anything.

By adding a log to observe weight changes in reweight_entity, I found
that calc_group_shares() often causes new_weight to become very small:

Hardware name: Unisoc UMS-base Board (DT)
Call trace:
dump_backtrace+0xec/0x138
show_stack+0x18/0x24
dump_stack_lvl+0x60/0x84
dump_stack+0x18/0x24
reweight_entity+0x3e8/0x5f4
dequeue_task_fair+0x448/0x948
dequeue_task+0xc4/0x398
deactivate_task+0x1c/0x28
pull_tasks+0x200/0x334
newidle_balance+0x3cc/0x438
pick_next_task_fair+0x58/0x670
__schedule+0x204/0x9a0
schedule+0x128/0x1a8
schedule_timeout+0x44/0x1c8
__skb_wait_for_more_packets+0xd0/0x17c
__unix_dgram_recvmsg+0xdc/0x3a8
unix_seqpacket_recvmsg+0x64/0x74
__sys_recvfrom+0x14c/0x1e4
__arm64_sys_recvfrom+0x24/0x38
invoke_syscall+0x58/0x114
el0_svc_common+0xac/0xe0
do_el0_svc+0x1c/0x28
el0_svc+0x3c/0x70
el0t_64_sync_handler+0x68/0xbc
el0t_64_sync+0x1a8/0x1ac
reweight_entity: the lag=3D-831088603030 vruntime=3D2086205903
limit=3D3071999998 old_weight=3D237238 new_weight=3D2