linux-kernel - Re: [PATCH] cpufreq: conservative: Do not use transition notifications

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: <20160613110851.GT27439@vireshk-i7>
Date:	Mon, 13 Jun 2016 16:38:51 +0530
From:	Viresh Kumar <viresh.kumar@...aro.org>
To:	"Rafael J. Wysocki" <rjw@...ysocki.net>
Cc:	Linux PM list <linux-pm@...r.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>
Subject: Re: [PATCH] cpufreq: conservative: Do not use transition
 notifications

On 10-06-16, 03:00, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
> 
> The conservative governor registers a transition notifier so it
> can update its internal requested_freq value if it falls out of the
> policy->min...policy->max range, but that's not the most
> straightforward way to achieve that.
> 
> To do it in a more straightforward way, first make sure that
> cs_dbs_timer() will only set frequencies between min and max.
> 
> With that, note that requested_freq will always fall between min
> and max unless either policy->min or policy->max changes and the
> governor's ->limits() callback will be invoked then.
> 
> Using this observation, add a ->limits callback pointer to
> struct dbs_governor, make cpufreq_dbs_governor_limits() invoke
> that callback if present, implement that callback in the conservative
> governor to update requested_freq if needed and drop the transition
> notifier from it, which also makes it possible to drop the
> struct cs_governor definition from there and simplify the code
> accordingly.

This code looks to me over-complicated and I am not sure if I
understand why we wanted the notifiers anyway? Why can't we replace
'dbs_info->requested_freq' with 'policy->cur' and kill the notifier
thing completely?

With requested_freq, we are trying to set the next freq to
requested_freq +- Delta, which I am not sure is the best approach
here.

What would go wrong if we will do, policy->cur +- delta instead?

The notifiers were added long back, to solve a problem which I don't
think will exist if we use policy->cur everywhere instead:

commit a8d7c3bc2396 ("[CPUFREQ] Make cpufreq_conservative handle
out-of-sync events properly")

Am I missing something?

-- 
viresh