Re: [PATCH v3] cpufreq: fix race on cpufreq online

[Viresh Kumar]

On 11-05-22, 14:59, Rafael J. Wysocki wrote:
> There are two things.
> 
> One is the possible race with respect to the sysfs access occurring
> during failing initialization and the other is that ->offline() or
> ->exit() can be called with or without holding the policy rwsem
> depending on the code path.
> 
> Namely, cpufreq_offline() calls them under the policy rwsem, but
> cpufreq_remove_dev() calls ->exit() outside the rwsem.  Also they are
> called outside the rwsem in cpufreq_online().

Right.

> Moreover, ->offline() and ->exit() cannot expect policy->cpus to be
> populated, because they are called when it is empty from
> cpufreq_offline().

Correct.

> So the $subject patch is correct AFAICS even though it doesn't address
> all of the above.

-- 
viresh