linux-kernel - Re: ACPI _CST introduced performance regresions on Haswll

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, 16 Oct 2020 15:41:12 +0200
From:   "Rafael J. Wysocki" <rafael@...nel.org>
To:     Mel Gorman <mgorman@...hsingularity.net>
Cc:     "Rafael J. Wysocki" <rafael.j.wysocki@...el.com>,
        Takashi Iwai <tiwai@...e.de>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Len Brown <lenb@...nel.org>,
        Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        Linux PM <linux-pm@...r.kernel.org>
Subject: Re: ACPI _CST introduced performance regresions on Haswll

On Thu, Oct 15, 2020 at 8:34 PM Mel Gorman <mgorman@...hsingularity.net> wrote:
>
> > Yes, it's well hidden but it's there. If the profile is made custom, then
> > the p-states can be selected and "custom" default enables C6 but not C3
> > (there is a note saying that it's not recommended for that CPU). If I
> > then switch it back to the normal profile, the c-states are not restored
> > so this is a one-way trip even if you disable the c-state in custom,
> > reboot, switch back, reboot. Same if the machine is reset to "optimal
> > default settings". Yey for BIOS developers.
> >
> > This means I have a limited number of attempts to do something about
> > this. 2 machines can no longer reproduce the problem reliably.
> >
>
> Turns out I didn't even have that. On another machine (same model,
> same cpu, different BIOS that cannot be updated), enabling the C6 state
> still did not enable it on boot and dmesg complained about CST not being
> usable. This is weird because one would expect that if CST was unusable
> that it would be the same as use_acpi == false.
>
> This could potentially be if the ACPI tables are unsuitable due to bad
> bad FFH information for a lower c-state. If _CST is not found or usable,
> should acpi_state_table.count be reset to go back to the old behaviour?

Yes, it should, although I would reset it in intel_idle_cst_usable()
right away before returning 'false'.

I can send a patch to do the above or please submit the one below as
it works too.

> diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
> index 13600c403035..3b84f8631b40 100644
> --- a/drivers/idle/intel_idle.c
> +++ b/drivers/idle/intel_idle.c
> @@ -1261,6 +1261,7 @@ static bool intel_idle_acpi_cst_extract(void)
>                 return true;
>         }
>
> +       acpi_state_table.count = 0;
>         pr_debug("ACPI _CST not found or not usable\n");
>         return false;
>  }
>
> --