diff mbox series

[v2,1/4] cpufreq: mediatek: CCI support SoC , the transition_delay set to 10 ms

Message ID 20241108063942.19744-2-chun-jen.tseng@mediatek.com (mailing list archive)
State New
Headers show
Series fixed mediatek-cpufreq has multi policy concurrency issue | expand

Commit Message

Mark Tseng Nov. 8, 2024, 6:39 a.m. UTC
SoC with CCI architecture should set transition_delay to 10 ms because
cpufreq need to call devfreq notifier in async mode. if delay less than
10 ms, it may get wrong OPP-level in devfreq passive governor.

Signed-off-by: Mark Tseng <chun-jen.tseng@mediatek.com>
---
 drivers/cpufreq/mediatek-cpufreq.c | 3 +++
 1 file changed, 3 insertions(+)

Comments

AngeloGioacchino Del Regno Nov. 14, 2024, 10:22 a.m. UTC | #1
Il 08/11/24 07:39, Mark Tseng ha scritto:
> SoC with CCI architecture should set transition_delay to 10 ms because
> cpufreq need to call devfreq notifier in async mode. if delay less than
> 10 ms, it may get wrong OPP-level in devfreq passive governor.
> 

This means that MediaTek SoCs can change their CPU frequency once every
10 milliseconds?!?!?!

I don't think that's really the case.

Besides, are you aware that this will have a *huge* impact on either power
consumption or performance?
We're going from a bunch of microseconds to *multiple* milliseconds here.

Regards,
Angelo
Mark Tseng Feb. 14, 2025, 7:41 a.m. UTC | #2
On Thu, 2024-11-14 at 11:22 +0100, AngeloGioacchino Del Regno wrote:

hi Angelo,

Thanks your review and recommendation.

I will fix this issue on next patch.

BRs,

Mark Tseng

> 
> External email : Please do not click links or open attachments until
> you have verified the sender or the content.
> 
> 
> Il 08/11/24 07:39, Mark Tseng ha scritto:
> > SoC with CCI architecture should set transition_delay to 10 ms
> > because
> > cpufreq need to call devfreq notifier in async mode. if delay less
> > than
> > 10 ms, it may get wrong OPP-level in devfreq passive governor.
> > 
> 
> This means that MediaTek SoCs can change their CPU frequency once
> every
> 10 milliseconds?!?!?!
> 
> I don't think that's really the case.
> 
> Besides, are you aware that this will have a *huge* impact on either
> power
> consumption or performance?
> We're going from a bunch of microseconds to *multiple* milliseconds
> here.
> 
> Regards,
> Angelo
>
diff mbox series

Patch

diff --git a/drivers/cpufreq/mediatek-cpufreq.c b/drivers/cpufreq/mediatek-cpufreq.c
index 663f61565cf7..f63183154e9a 100644
--- a/drivers/cpufreq/mediatek-cpufreq.c
+++ b/drivers/cpufreq/mediatek-cpufreq.c
@@ -597,6 +597,9 @@  static int mtk_cpufreq_init(struct cpufreq_policy *policy)
 	policy->driver_data = info;
 	policy->clk = info->cpu_clk;
 
+	if (info->soc_data->ccifreq_supported)
+		policy->transition_delay_us = 10000;
+
 	return 0;
 }