linux-kernel - Re: [PATCH] cpufreq,cppc: fix issue when hotplugging out policy->cpu

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]

Date:   Fri, 4 Sep 2020 10:43:03 +0100
From:   Ionela Voinescu <ionela.voinescu@....com>
To:     Viresh Kumar <viresh.kumar@...aro.org>
Cc:     rjw@...ysocki.net, sudeep.holla@....com, linux-pm@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH] cpufreq,cppc: fix issue when hotplugging out policy->cpu

Hi Viresh,

On Friday 04 Sep 2020 at 10:36:04 (+0530), Viresh Kumar wrote:
[..]
> >  /* Per CPU container for runtime CPPC management. */
> >  struct cppc_cpudata {
> > -	int cpu;
> >  	struct cppc_perf_caps perf_caps;
> >  	struct cppc_perf_ctrls perf_ctrls;
> >  	struct cppc_perf_fb_ctrs perf_fb_ctrs;
> 
> With the way things are designed, I believe this is one of the bugs
> out of many.
> 
> The structure cppc_cpudata must be shared across all CPUs of the same
> policy, so they all end up using the same set of values for different
> variables. i.e. it shouldn't be a per-cpu thing at all. Just allocate
> it from cpufreq_driver->init and store in policy->driver_data for use
> elsewhere.
> 
> That would be a proper fix IMO, we just avoided one of the bugs here
> otherwise.
> 

Do you know why it was designed this way in the first place?

I assumed it was designed like this (per-cpu cppc_cpudata structures) to
allow for the future addition of support for the HW_ALL CPPC coordination
type. In that case you can still have PSD (dependency) domains but the
desired performance controls would be per-cpu, with the coordination
done in hardware/firmware. So, in the HW_ALL case you'd end up having
different performance controls even for CPUs in the same policy.
Currently the CPPC driver only supports SW_ANY which is the traditional
cpufreq approach.

Thanks,
Ionela.

> -- 
> viresh