linux-kernel - Re: [PATCH v3 1/3] cpufreq: mediatek: using global lock avoid race condition

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <06356a55cdb3c34dfc716349d1967f95655b0ab2.camel@mediatek.com>
Date: Mon, 14 Apr 2025 08:42:22 +0000
From: Chun-Jen Tseng (曾俊仁)
	<Chun-Jen.Tseng@...iatek.com>
To: "viresh.kumar@...aro.org" <viresh.kumar@...aro.org>
CC: "cw00.choi@...sung.com" <cw00.choi@...sung.com>, "rafael@...nel.org"
	<rafael@...nel.org>, Project_Global_Chrome_Upstream_Group
	<Project_Global_Chrome_Upstream_Group@...iatek.com>, "AngeloGioacchino Del
 Regno" <angelogioacchino.delregno@...labora.com>, "myungjoo.ham@...sung.com"
	<myungjoo.ham@...sung.com>, "linux-kernel@...r.kernel.org"
	<linux-kernel@...r.kernel.org>, "kyungmin.park@...sung.com"
	<kyungmin.park@...sung.com>, "linux-arm-kernel@...ts.infradead.org"
	<linux-arm-kernel@...ts.infradead.org>, "matthias.bgg@...il.com"
	<matthias.bgg@...il.com>, "linux-pm@...r.kernel.org"
	<linux-pm@...r.kernel.org>, "linux-mediatek@...ts.infradead.org"
	<linux-mediatek@...ts.infradead.org>
Subject: Re: [PATCH v3 1/3] cpufreq: mediatek: using global lock avoid race
 condition

Hi Viresh,

The CCI level choose by Max_Level(LCPU & BCPU frequency) in devfreq
driver.
without global lock, It may choose wrong CCI level and cause system
stall.

I hope this flow is serial setting like, BCPU / LCPU set frequency ->
set CCI level -> BCPU / LCPU set frequency -> set CCI level -> ......

without global lock, it could be LCPU / BCPU set frequency -> set CCI
level(during this time, it may change BCPU / LCPU frequency and cause
system stall.

I also can only do global lock on ccifreq_support SoC.

BRs,

Mark Tseng

On Mon, 2025-03-24 at 11:13 +0530, Viresh Kumar wrote:
> 
> External email : Please do not click links or open attachments until
> you have verified the sender or the content.
> 
> 
> Hi,
> 
> Thanks for sharing the details this time, it makes it much clearer
> now.
> 
> On 24-03-25, 03:21, Chun-Jen Tseng (曾俊仁) wrote:
> > I think the best configuration sequence is as follows:
> >   cpufreq policy -> set frequency -> CCI governor get
> > CPUFREQ_POSTCHANGE NB -> choose CCI frequency -> set CCI frequency
> > 
> > However, in drivers/devfreq/governor_passive.c#L77,
> > get_target_freq_with_cpufreq() retrieves the current frequency of
> > each
> > policy,
> > and it determines the CCI frequency based on the frequency of each
> > policy.
> > 
> > But if policy-0 and policy-6 enter simultaneously, the CCI governor
> > might get an incorrect frequency.
> 
> Yes it may fetch the current frequency (or last known one), but that
> shouldn't be a problem as the postchange notification for policy-6
> should get called right after and should fix the issue. Right ?
> 
> I don't think this is a race and if this requires fixing. clk_get()
> for any device, will always return the last configured value, while
> the clock might be changing at the same time.
> 
> What's important is that you don't get an incorrect frequency (as in
> based on intermediate values of registers, etc). Note that the last
> configured frequency isn't an incorrect frequency.
> 
> > cpufreq policy-0 -> set frequency -> CCI governor get
> > CPUFREQ_POSTCHANGE NB -> choose CCI frequency -> set CCI frequency
> > => during this time, the CCI governor gets policy-0 and policy-6,
> > BUT
> > policy-6 may change frequency by cpufreq driver at the same time.
> 
> Sure, and I don't see a problem with that. The issue is there only if
> we can reach a state where CCI is left configured in the wrong state.
> Which I don't think would happen here as the postchange notifier will
> get called again, forcing a switch of frequency again.
> 
> --
> viresh