[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240423114416.64308-1-hupu@oppo.com>
Date: Tue, 23 Apr 2024 19:44:16 +0800
From: <hupu@...o.com>
To: <peterz@...radead.org>, <mingo@...hat.com>, <juri.lelli@...hat.com>,
<vincent.guittot@...aro.org>, <dietmar.eggemann@....com>,
<rostedt@...dmis.org>, <bsegall@...gle.com>, <mgorman@...e.de>,
<bristot@...hat.com>, <vschneid@...hat.com>, <linux-kernel@...r.kernel.org>
CC: hupu <hupu@...o.com>
Subject: [PATCH] sched/fair.c: Fix the calculation method of 'lag' to ensure that the vlag of the task after placement is the same as before.
From: hupu <hupu@...o.com>
I think the 'lag' calculation here is inaccurate.
Assume that delta needs to be subtracted from v_i to ensure that the
vlag of task i after placement is the same as before. At this time, the
vlag of task i after placement should be:
vl'_i = V' - (v_i - delta)
>From the above formula, we know that vl'_i should be:
vl'_i = (vl_i * W)/(W + w_i)
That is to say:
V' - (v_i - delta) = (vl_i * W)/(W + w_i)
For a newly added entity, generally set v_i to V', and the above formula
can be converted into:
V' - (V' - delta) = (vl_i * W)/(W + w_i)
Therefore the value of delta should be as follows, where delta is the
'lag' in the code.
delta = (vl_i * W)/(W + w_i)
Signed-off-by: hupu <hupu@...o.com>
---
kernel/sched/fair.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 03be0d1330a6..c5f74f753be8 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5239,9 +5239,12 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
if (curr && curr->on_rq)
load += scale_load_down(curr->load.weight);
- lag *= load + scale_load_down(se->load.weight);
+ lag *= load;
+
+ load += scale_load_down(se->load.weight);
if (WARN_ON_ONCE(!load))
load = 1;
+
lag = div_s64(lag, load);
}
--
2.17.1
Powered by blists - more mailing lists