Re: AMD Bobcat cpufreq

Re: AMD Bobcat cpufreq

[Joonas Saarinen]

Borislav Petkov [bp@alien8.de] wrote:
> Once you've done that successfully, you need to run as root:
> 
> $ cd tools/power/cpupower/
> $ LD_LIBRARY_PATH=. ./cpupower frequency-info

This is what I get:

analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0
  CPUs which need to have their frequency coordinated by software: 0
  hardware limits: 800 MHz - 1000 MHz
  available frequency steps: 1000 MHz, 800 MHz
  available cpufreq governors: conservative, ondemand, userspace, powersave, performance
  current policy: frequency should be within 800 MHz and 1000 MHz.
                  The governor "ondemand" may decide which speed to use
                  within this range.
  current CPU frequency is 1000 MHz (asserted by call to hardware).
  cpufreq stats: 1000 MHz:-nan%, 800 MHz:-nan%  (646)
  boost state support:
    Supported: yes
    Active: yes

It's promising that the boost state support is "supported and active" 
but the output is still kind of sparse. There's no mention about
specific turbo states, nor do I get /proc/cpuinfo or that cpupower tool
ever to show an evidence that the CPU is going to 1333MHz. On Windows
CPU-Z showed it constantly visiting that state.

That's all for now.

Joonas

Re: AMD Bobcat cpufreq

[Borislav Petkov]

On Fri, Oct 18, 2013 at 04:19:49PM +0300, Joonas Saarinen wrote:
> It's promising that the boost state support is "supported and
> active" but the output is still kind of sparse. There's no mention
> about specific turbo states,

could be bug in the tool...

> nor do I get /proc/cpuinfo or that cpupower tool ever to show an
> evidence that the CPU is going to 1333MHz. On Windows CPU-Z showed it
> constantly visiting that state.

You can also try turbostat in tools/power/x86/turbostat/

# ./turbostat -i 1

to give you core freq. readout every second and then pin a workload on
one core in another shell, say kernel build:

$ taskset 1 make

You should be able to see core 0 boosting like in my case:

cor CPU   GHz  TSC
        3.90 4.01
  0   0 4.17 4.01
  1   1 4.17 4.01
  2   2 2.23 4.01
  3   3 2.74 4.01
  4   4 1.83 4.01
  5   5 3.23 4.01
  6   6 1.69 4.01
  7   7 1.43 4.01

cores 0 and 1 go over 4GHz which is their boosted state.

HTH.

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

Re: AMD Bobcat cpufreq

[Borislav Petkov]

On Fri, Oct 18, 2013 at 06:23:01PM +0300, Joonas Saarinen wrote:
> Borislav Petkov [bp@alien8.de] wrote:
> >You can also try turbostat in tools/power/x86/turbostat/
> >
> ># ./turbostat -i 1
> 
> We might be on to something. With work pinned on one core:
> 
> cor CPU   GHz  TSC
>        1.16 1.00
>  0   0 1.17 1.00
>  1   1 0.80 1.00
> 
> When both cores are occupied:
> 
> cor CPU   GHz  TSC
>        1.00 1.00
>  0   0 1.00 1.00
>  1   1 1.00 1.00
> 
> In this case both cores are limited to 1GHz, but it might be intentional
> to avoid overheating.

That's the boosting algorithm - the more parts of the silicon consume
power, the more it gets spread out amongst them so that each one of them
gets a smaller amount of the remaining power up to the TDP which is
available for consumption.

Also, AFAICT, on those machines the boosting algorithm involves the GPU
so that it boosts too. This might be the reason why you don't see 1.33
GHz with one of the cores.

> So there is some turbo functionality indeed. The frequency shown by
> turbostat is still a bit off the target, not sure if there is any
> actual problem though.

Yeah, see above.

I think all is fine with your box :)

HTH.

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

Re: AMD Bobcat cpufreq

[Joonas Saarinen]

Borislav Petkov [bp@alien8.de] wrote:
> You can also try turbostat in tools/power/x86/turbostat/
> 
> # ./turbostat -i 1

We might be on to something. With work pinned on one core:

cor CPU   GHz  TSC
        1.16 1.00
  0   0 1.17 1.00
  1   1 0.80 1.00

When both cores are occupied:

cor CPU   GHz  TSC
        1.00 1.00
  0   0 1.00 1.00
  1   1 1.00 1.00

In this case both cores are limited to 1GHz, but it might be intentional
to avoid overheating.

So there is some turbo functionality indeed. The frequency shown by 
turbostat is still a bit off the target, not sure if there is any actual
problem though.

Joonas

Re: AMD Bobcat cpufreq

[Borislav Petkov]

On Thu, Oct 17, 2013 at 11:54:28PM +0300, Joonas Saarinen wrote:
> All right. I'm finally back to re-investigate the issue. The
> requested data can be found at:
> 
> http://users.metropolia.fi/~joonasms/bobcat/
> 
> The CPU underclocks as expected but the turbo multiplier is never
> activated.

Right, first of all, when you reply to people on lkml, please make sure
to use "reply-to-all" in your mail client so that the people who you
reply to can get your message directly instead of possibly missing it in
the lkml flood.

Now, I'd guess this 1.3 GHz power state you're talking about is the
boosted state. Here's how you dump that: build and install the cpupower
tool in the kernel repository's tools/power/cpupower/ directory.

Then, do as root:

$ modprobe msr

because you need this module.

Once you've done that successfully, you need to run as root:

$ cd tools/power/cpupower/
$ LD_LIBRARY_PATH=. ./cpupower frequency-info

 [ Btw, the cpupower tool might already be available in your distro... ]

Anyways, once you run it successfully, you'll get a similar dump as
below which will show you all your power states. Pb0 should be 1333 in
your case and if it says "boost state support: ... Active" then all is
fine and your machine is switching into it but in a burstly manner - it
cannot be running in it all the time because it would explode the TDP
limits of the box.

HTH.

analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 4.0 us.
  hardware limits: 1.40 GHz - 4.00 GHz
  available frequency steps: 4.00 GHz, 3.40 GHz, 2.80 GHz, 2.10 GHz, 1.40 GHz
  available cpufreq governors: powersave, userspace, conservative, ondemand, performance
  current policy: frequency should be within 1.40 GHz and 4.00 GHz.
                  The governor "ondemand" may decide which speed to use
                  within this range.
  current CPU frequency is 2.80 GHz (asserted by call to hardware).
  cpufreq stats: 4.00 GHz:4.85%, 3.40 GHz:0.51%, 2.80 GHz:1.01%, 2.10 GHz:2.01%, 1.40 GHz:91.62%  (1281567)
  boost state support:
    Supported: yes
    Active: yes
    Boost States: 2
    Total States: 7
    Pstate-Pb0: 4200MHz (boost state)
    Pstate-Pb1: 4100MHz (boost state)
    Pstate-P0:  4000MHz
    Pstate-P1:  3400MHz
    Pstate-P2:  2800MHz
    Pstate-P3:  2100MHz
    Pstate-P4:  1400MHz

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

Re: AMD Bobcat cpufreq

[Joonas Saarinen]

On Thu, 22 Aug 2013 13:17:53 +0200 Borislav Petkov wrote:
> On Tue, Aug 20, 2013 at 10:58:07PM +0300, Joonas Saarinen wrote:
> > Hi guys,
> > 
> > The turbo core feature on AMD Bobcat platform seems not to be
> > working. I have a netbook with C-60 CPU which never cranks the
> > frequency over 1000MHz (it could reach 1333MHz).
>
> Yeah, how about more info like which kernel, /proc/cpuinfo, full
> dmesg, dmidecode... sounds like enough for starters.
>
> Thanks.

All right. I'm finally back to re-investigate the issue. The
requested data can be found at:

http://users.metropolia.fi/~joonasms/bobcat/

The CPU underclocks as expected but the turbo multiplier is never
activated.

Joonas

Re: AMD Bobcat cpufreq

[Borislav Petkov]

On Tue, Aug 20, 2013 at 10:58:07PM +0300, Joonas Saarinen wrote:
> Hi guys,
> 
> The turbo core feature on AMD Bobcat platform seems not to be
> working. I have a netbook with C-60 CPU which never cranks the
> frequency over 1000MHz (it could reach 1333MHz).

Yeah, how about more info like which kernel, /proc/cpuinfo, full dmesg,
dmidecode... sounds like enough for starters.

Thanks.

AMD Bobcat cpufreq

[Joonas Saarinen]

Hi guys,

The turbo core feature on AMD Bobcat platform seems not to be working. I 
have a netbook with C-60 CPU which never cranks the frequency over 
1000MHz (it could reach 1333MHz).

Can this be expected to work some day?

Thanks for any information,
Joonas