linux-kernel - Re: [PATCH v3] cpufreq: fix race on cpufreq online

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: <20220513042705.nbnd6vccuiu6lb7a@vireshk-i7>
Date:   Fri, 13 May 2022 09:57:05 +0530
From:   Viresh Kumar <viresh.kumar@...aro.org>
To:     "Rafael J. Wysocki" <rafael@...nel.org>
Cc:     Schspa Shi <schspa@...il.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Linux PM <linux-pm@...r.kernel.org>
Subject: Re: [PATCH v3] cpufreq: fix race on cpufreq online

On 12-05-22, 12:49, Rafael J. Wysocki wrote:
> Well, would there be a problem with moving the
> cpufreq_policy_put_kobj() call to the front of cpufreq_policy_free()?

Emptying cpufreq_cpu_data first is required, else someone else will
end up doing kobject_get() again.

> If we did that, we'd know that everything could be torn down safely,
> because nobody would be holding references to the policy any more.

With the way we are progressing now, we will always have policy->cpus
empty while we reach cpufreq_policy_free(). With that I think we will
be safe with the current code here. I would also add a BUG_ON() here
for non empty policy->cpus to be safe.

> > > TBH, I'm not sure why show() doesn't check policy_is_inactive() under the rwsem.
> >
> > I agree, both show/store should have it.
> >
> > > Moreover, I'm not sure why the locking dance in store() is necessary.
> >
> > commit fdd320da84c6 ("cpufreq: Lock CPU online/offline in cpufreq_register_driver()")
> 
> I get that, but I'm wondering if locking CPU hotplug from store() is
> needed at all.  I mean, if we are in store(), we are holding an active
> reference to the policy kobject, so the policy cannot go away until we
> are done anyway.  Thus it should be sufficient to use the policy rwsem
> for synchronization.

I think after the current patchset is applied and we have the inactive
policy check in store(), we won't required the dance after all.

-- 
viresh