linux-kernel - Re: [PATCH] cpufreq: Fix initialization with disabled boost

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: <e9fc6154-7199-4709-b428-3f848f1597e8@arm.com>
Date: Tue, 17 Jun 2025 09:20:52 +0100
From: Christian Loehle <christian.loehle@....com>
To: "zhenglifeng (A)" <zhenglifeng1@...wei.com>,
 Robin Murphy <robin.murphy@....com>,
 "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
 linux-pm <linux-pm@...r.kernel.org>, Viresh Kumar <viresh.kumar@...aro.org>,
 "Rafael J. Wysocki" <rafael@...nel.org>
Subject: Re: [PATCH] cpufreq: Fix initialization with disabled boost

On 6/17/25 03:14, zhenglifeng (A) wrote:
> On 2025/6/17 3:10, Robin Murphy wrote:
>> On 2025-06-16 6:25 pm, Christian Loehle wrote:
>>> The boost_enabled early return in policy_set_boost() caused
>>> the boost disabled at initialization to not actually set the
>>> initial policy->max, therefore effectively enabling boost while
>>> it should have been enabled.
>>>
>>> Fixes: 27241c8b63bd ("cpufreq: Introduce policy_set_boost()")
>>
>> I think it's a bit older than that - I noticed this with 6.15 stable, prior to that refactoring, and from a poke through the history the underlying logic appears to date back to dd016f379ebc ("cpufreq: Introduce a more generic way to set default per-policy boost flag"). Hopefully someone can figure out the appropriate stable backport.
>>
>> I can at least confirm that equivalently hacking out the "&& policy->boost_enabled != cpufreq_boost_enabled()" condition previously here does have the desired effect for me of initialising scaling_max_freq correctly at boot, but I'm not sure that's entirely correct on its own...
>>
>> Thanks,
>> Robin.
>>
>>> Reported-by: Robin Murphy <robin.murphy@....com>
>>> Signed-off-by: Christian Loehle <christian.loehle@....com>
>>> ---
>>>   drivers/cpufreq/cpufreq.c | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
>>> index d7426e1d8bdd..e85139bd0436 100644
>>> --- a/drivers/cpufreq/cpufreq.c
>>> +++ b/drivers/cpufreq/cpufreq.c
>>> @@ -1630,7 +1630,7 @@ static int cpufreq_online(unsigned int cpu)
>>>        */
>>>       if (cpufreq_driver->set_boost && policy->boost_supported &&
>>>           (new_policy || !cpufreq_boost_enabled())) {
>>> -        ret = policy_set_boost(policy, cpufreq_boost_enabled());
>>> +        ret = cpufreq_driver->set_boost(policy, cpufreq_boost_enabled());
>>>           if (ret) {
>>>               /* If the set_boost fails, the online operation is not affected */
>>>               pr_info("%s: CPU%d: Cannot %s BOOST\n", __func__, policy->cpu,
>>
>>
> 
> I don't quite understand what problem you've met. It semms like you guys
> propose that set_boost() should be called no matter what
> policy->boost_enabled is. Having more details would help to clarify things,
> such as which driver you use and what you expect but not be achieved.
> 

so calling policy_set_boost(policy, enable) is a noop here if
policy->boost_enabled == cpufreq_boost_enabled():

	if (policy->boost_enabled == enable)
		return 0;

We have policy->boost_enabled == false on boot, thus never actually
setting policy->max up ever, which leads to the following:

# cat /sys/devices/system/cpu/cpufreq/policy4/scaling_boost_frequencies
2016000 
# cat /sys/devices/system/cpu/cpufreq/policy4/scaling_max_freq
2016000 
# cat /sys/devices/system/cpu/cpufreq/boost
0
# echo 1 > /sys/devices/system/cpu/cpufreq/boost
# echo 0 > /sys/devices/system/cpu/cpufreq/boost
# cat /sys/devices/system/cpu/cpufreq/policy4/scaling_boost_frequencies
1800000

Anyway I'll bisect some more to find the actual first bad commit and
resend.