linux-kernel - Re: [intel-pstate driver regression] processor frequency very high even if in idle

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:	Thu, 31 Mar 2016 09:45:00 +0200
From:	Sedat Dilek <sedat.dilek@...il.com>
To:	"Rafael J. Wysocki" <rafael@...nel.org>
Cc:	"Pandruvada, Srinivas" <srinivas.pandruvada@...el.com>,
	"Wysocki, Rafael J" <rafael.j.wysocki@...el.com>,
	"dsmythies@...us.net" <dsmythies@...us.net>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"jrg.otte@...il.com" <jrg.otte@...il.com>,
	"linux-pm@...r.kernel.org" <linux-pm@...r.kernel.org>
Subject: Re: [intel-pstate driver regression] processor frequency very high
 even if in idle

On Thu, Mar 31, 2016 at 8:38 AM, Sedat Dilek <sedat.dilek@...il.com> wrote:
> On Thu, Mar 31, 2016 at 1:41 AM, Rafael J. Wysocki <rafael@...nel.org> wrote:
>> On Thu, Mar 31, 2016 at 1:32 AM, Sedat Dilek <sedat.dilek@...il.com> wrote:
>>> On Thu, Mar 31, 2016 at 1:28 AM, Rafael J. Wysocki <rafael@...nel.org> wrote:
>>>> On Thu, Mar 31, 2016 at 1:25 AM, Sedat Dilek <sedat.dilek@...il.com> wrote:
>>>>> On Thu, Mar 31, 2016 at 1:17 AM, Pandruvada, Srinivas
>>>>> <srinivas.pandruvada@...el.com> wrote:
>>>>>> On Wed, 2016-03-30 at 15:46 -0700, Srinivas Pandruvada wrote:
>>>>>>> On Thu, 2016-03-31 at 00:25 +0200, Rafael J. Wysocki wrote:
>>>>>>> >
>>>>>>> > On 3/31/2016 12:18 AM, Srinivas Pandruvada wrote:
>>>>>>> > >
>>>>>>> > >
>>>>>>> > > On Wed, 2016-03-30 at 23:41 +0200, Sedat Dilek wrote:
>>>>>>> > > >
>>>>>>> > > >
>>>>>>> > > > Hi,
>>>>>>> > > >
>>>>>>> > > > I am using Intel-PState-Driver here with v4.6-rc1 and Intel-
>>>>>>> > > > SandyBridge-CPU.
>>>>>>> > > >
>>>>>>> > > > Here are my turbostat results attached.
>>>>>>> > > >
>>>>>>> > > > $ cd $BUILD_DIR
>>>>>>> > > > $ LC_ALL=C make -C tools/ turbostat
>>>>>>> > > >
>>>>>>> > > > $ sudo ./turbostat -i 1 --msr=0x199 --debug --out
>>>>>>> > > > /tmp/turbostat-i-1-msr-0x199-debug.txt
>>>>>>> > > >
>>>>>>> > > > Will try <https://patchwork.kernel.org/patch/8702071/>.
>>>>>>> > > >
>>>>>>> > > > Please see attached files.
>>>>>>> > > >
>>>>>>> > > Thanks. Your logs make sense. You have config set to performance
>>>>>>> > > mode
>>>>>>> > > by default (Which I believe default in all kernel Ubuntu). So as
>>>>>>> > > expected Intel P state was asking for max. So there is no issue
>>>>>>> > > here.
>>>>>>> > But the behavior is different from what it used to be, isn't it?
>>>>>>> >
>>>>
>>>> [cut]
>>>>
>>>>>
>>>>> Too much questions.
>>>>
>>>> OK
>>>>
>>>>> I switched to CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE=y and tested with
>>>>> the simplified revert-patch of Rafael.
>>>>
>>>> Why do you need the revert patch in the first place?
>>>
>>> I simply wanted to test CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE=y and
>>> keep the rest.
>>
>> But it is not clear to us whether or not your system has regressed
>> with respect to 4.5, which is the whole point of this thread.  Do you
>> see a regression here?
>>
>> Jörg in particular is saying that he sees the problem with both
>> performance and powersave settings.  What about you?
>
> I have only tested with 4.6-rc1.
> My 4.5 kernel has no CONFIG_X86_MSR=m which is mandatory for turbostat.
>
> With CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE=y and without your
> simplified revert-patch.
> The values of Bzy_MHz look good for me.
>
> CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y is somehow showing higher
> values than TSC_MHz.
> Not sure if this really make sense if TSC_MHz is a max-value (which I
> don't know it is).
>

I have tested with Linux v4.5 and
CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y and CONFIG_X86_MSR=m.

Both turbostat versions (v4.8 from Linux-4.5 and v4.11 from
Linux-4.6-rc1) show higher values in Bzy_MHz than TSC_MHz.

Checking the statement, after 1 minute performance->powersave freq-gov
auto-change on Ubuntu...
I ran approx. 3 minutes...

$ sudo ./turbostat --interval=1 --msr=0x199 --debug <--- turbostat v4.8

...this is the last line...

    Core     CPU Avg_MHz   %Busy Bzy_MHz TSC_MHz   MSR 0x199     SMI
CPU%c1  CPU%c3  CPU%c6  CPU%c7 CoreTmp  PkgTmp Pkg%pc2 Pkg%pc3 Pkg%pc6
Pkg%pc7 PkgWatt CorWatt GFXWatt
       -       -       7    0.40    1754    1596  0x00000000       0
 1.31    0.00    0.00   98.29      55      55    0.71    0.00    1.25
 93.96    2.62    0.16    0.08
       0       0      10    0.63    1616    1596  0x00001700       0
 2.01    0.00    0.00   97.36      54      55    0.71    0.00    1.25
 93.96    2.62    0.16    0.08
       0       1       4    0.27    1566    1596  0x00001700       0    2.37
       1       2      11    0.55    2073    1596  0x00001700       0
 0.23    0.00    0.00   99.22      55
       1       3       2    0.16    1525    1596  0x00001700       0    0.62

Dunno why CPU%c7 and Pkg%pc7 have so high values and what they mean.

See also attached files.

Hope this helps with less confusion.

- Sedat -

View attachment "dmesg_4.5.0-4-iniza-small.txt" of type "text/plain" (54433 bytes)

Download attachment "config-4.5.0-4-iniza-small" of type "application/octet-stream" (133587 bytes)

View attachment "turbostat-i-1-msr-0x199-debug_4.5.0-4-iniza-small.txt" of type "text/plain" (88878 bytes)

View attachment "turbostat-interval-1-msr-0x199-debug_4.5.0-4-iniza-small_turbostat-4-8.txt" of type "text/plain" (25937 bytes)

View attachment "4.5.0-4-iniza-small.patch" of type "text/x-diff" (3567 bytes)