Hard Disk consumes lots of power in s2idle

Hard Disk consumes lots of power in s2idle

[Kai-Heng Feng]

Hi Srinivas,

Your previous work to support DEVSLP works well on SATA SSDs, so I am asking you the issue I am facing:
Once a laptop has a HDD installed, the power consumption during S2Idle increases ~0.4W, which is quite a lot.
However, HDDs don't seem to support DEVSLP, so I wonder if you know to do proper power management for HDDs?

Kai-Heng

Re: Hard Disk consumes lots of power in s2idle

[Srinivas Pandruvada]

Hi Kai,

On Wed, 2020-02-19 at 22:22 +0800, Kai-Heng Feng wrote:
> Hi Srinivas,
> 
> Your previous work to support DEVSLP works well on SATA SSDs, so I am
> asking you the issue I am facing:
> Once a laptop has a HDD installed, the power consumption during
> S2Idle increases ~0.4W, which is quite a lot.
> However, HDDs don't seem to support DEVSLP, so I wonder if you know
> to do proper power management for HDDs?
What is the default here
cat /sys/power/mem_sleep
s2idle or deep?

Please follow debug steps here:
https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-linux

We need to check whether you get any PC10 residency or not.

Thanks,
Srinivas


> 
> Kai-Heng

Re: Hard Disk consumes lots of power in s2idle

[Kai-Heng Feng]

Hi Srinivas,

> On Feb 20, 2020, at 02:36, Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> wrote:
> 
> Hi Kai,
> 
> On Wed, 2020-02-19 at 22:22 +0800, Kai-Heng Feng wrote:
>> Hi Srinivas,
>> 
>> Your previous work to support DEVSLP works well on SATA SSDs, so I am
>> asking you the issue I am facing:
>> Once a laptop has a HDD installed, the power consumption during
>> S2Idle increases ~0.4W, which is quite a lot.
>> However, HDDs don't seem to support DEVSLP, so I wonder if you know
>> to do proper power management for HDDs?
> What is the default here
> cat /sys/power/mem_sleep
> s2idle or deep?

It defaults to s2idle.

> 
> Please follow debug steps here:
> https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-linux
> 
> We need to check whether you get any PC10 residency or not.

Yes it reaches PC10. It doesn't reach SLP_S0 though.
The real number on S2Idle power consumption:
No HDD: ~1.4W
One HDD: ~1.8W

If the SoC doesn't hit PC10 the number should be significantly higher.
That's why I think the issue is the power management on HDD itself.

Kai-Heng

> 
> Thanks,
> Srinivas
> 
> 
>> 
>> Kai-Heng
> 

Re: Hard Disk consumes lots of power in s2idle

[Rafael J. Wysocki]

On Thu, Feb 20, 2020 at 9:08 AM Kai-Heng Feng
<kai.heng.feng@canonical.com> wrote:
>
> Hi Srinivas,
>
> > On Feb 20, 2020, at 02:36, Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> wrote:
> >
> > Hi Kai,
> >
> > On Wed, 2020-02-19 at 22:22 +0800, Kai-Heng Feng wrote:
> >> Hi Srinivas,
> >>
> >> Your previous work to support DEVSLP works well on SATA SSDs, so I am
> >> asking you the issue I am facing:
> >> Once a laptop has a HDD installed, the power consumption during
> >> S2Idle increases ~0.4W, which is quite a lot.
> >> However, HDDs don't seem to support DEVSLP, so I wonder if you know
> >> to do proper power management for HDDs?
> > What is the default here
> > cat /sys/power/mem_sleep
> > s2idle or deep?
>
> It defaults to s2idle.
>
> >
> > Please follow debug steps here:
> > https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-linux
> >
> > We need to check whether you get any PC10 residency or not.
>
> Yes it reaches PC10. It doesn't reach SLP_S0 though.
> The real number on S2Idle power consumption:
> No HDD: ~1.4W
> One HDD: ~1.8W
>
> If the SoC doesn't hit PC10 the number should be significantly higher.
> That's why I think the issue is the power management on HDD itself.

I'm assuming that you mean a non-SSD device here.

That would be handled via ata_port_suspend() I gather and whatever
that does should do the right thing.

Do you think that the disk doesn't spin down or it spins down, but the
logic stays on?

Re: Hard Disk consumes lots of power in s2idle

[Kai-Heng Feng]

> On Feb 20, 2020, at 18:12, Rafael J. Wysocki <rafael@kernel.org> wrote:
> 
> On Thu, Feb 20, 2020 at 9:08 AM Kai-Heng Feng
> <kai.heng.feng@canonical.com> wrote:
>> 
>> Hi Srinivas,
>> 
>>> On Feb 20, 2020, at 02:36, Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> wrote:
>>> 
>>> Hi Kai,
>>> 
>>> On Wed, 2020-02-19 at 22:22 +0800, Kai-Heng Feng wrote:
>>>> Hi Srinivas,
>>>> 
>>>> Your previous work to support DEVSLP works well on SATA SSDs, so I am
>>>> asking you the issue I am facing:
>>>> Once a laptop has a HDD installed, the power consumption during
>>>> S2Idle increases ~0.4W, which is quite a lot.
>>>> However, HDDs don't seem to support DEVSLP, so I wonder if you know
>>>> to do proper power management for HDDs?
>>> What is the default here
>>> cat /sys/power/mem_sleep
>>> s2idle or deep?
>> 
>> It defaults to s2idle.
>> 
>>> 
>>> Please follow debug steps here:
>>> https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-linux
>>> 
>>> We need to check whether you get any PC10 residency or not.
>> 
>> Yes it reaches PC10. It doesn't reach SLP_S0 though.
>> The real number on S2Idle power consumption:
>> No HDD: ~1.4W
>> One HDD: ~1.8W
>> 
>> If the SoC doesn't hit PC10 the number should be significantly higher.
>> That's why I think the issue is the power management on HDD itself.
> 
> I'm assuming that you mean a non-SSD device here.

Yes, it's spinning rust here.

> 
> That would be handled via ata_port_suspend() I gather and whatever
> that does should do the right thing.
> 
> Do you think that the disk doesn't spin down or it spins down, but the
> logic stays on?

The spin sound is audible, so I am certain the HDD spins down during S2Idle.

How do I know if the logic is on or off?

Kai-Heng

Re: Hard Disk consumes lots of power in s2idle

[Rafael J. Wysocki]

On Thu, Feb 20, 2020 at 11:25 AM Kai-Heng Feng
<kai.heng.feng@canonical.com> wrote:
>
>
>
> > On Feb 20, 2020, at 18:12, Rafael J. Wysocki <rafael@kernel.org> wrote:
> >
> > On Thu, Feb 20, 2020 at 9:08 AM Kai-Heng Feng
> > <kai.heng.feng@canonical.com> wrote:
> >>
> >> Hi Srinivas,
> >>
> >>> On Feb 20, 2020, at 02:36, Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> wrote:
> >>>
> >>> Hi Kai,
> >>>
> >>> On Wed, 2020-02-19 at 22:22 +0800, Kai-Heng Feng wrote:
> >>>> Hi Srinivas,
> >>>>
> >>>> Your previous work to support DEVSLP works well on SATA SSDs, so I am
> >>>> asking you the issue I am facing:
> >>>> Once a laptop has a HDD installed, the power consumption during
> >>>> S2Idle increases ~0.4W, which is quite a lot.
> >>>> However, HDDs don't seem to support DEVSLP, so I wonder if you know
> >>>> to do proper power management for HDDs?
> >>> What is the default here
> >>> cat /sys/power/mem_sleep
> >>> s2idle or deep?
> >>
> >> It defaults to s2idle.
> >>
> >>>
> >>> Please follow debug steps here:
> >>> https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-linux
> >>>
> >>> We need to check whether you get any PC10 residency or not.
> >>
> >> Yes it reaches PC10. It doesn't reach SLP_S0 though.
> >> The real number on S2Idle power consumption:
> >> No HDD: ~1.4W
> >> One HDD: ~1.8W
> >>
> >> If the SoC doesn't hit PC10 the number should be significantly higher.
> >> That's why I think the issue is the power management on HDD itself.
> >
> > I'm assuming that you mean a non-SSD device here.
>
> Yes, it's spinning rust here.
>
> >
> > That would be handled via ata_port_suspend() I gather and whatever
> > that does should do the right thing.
> >
> > Do you think that the disk doesn't spin down or it spins down, but the
> > logic stays on?
>
> The spin sound is audible, so I am certain the HDD spins down during S2Idle.

OK

> How do I know if the logic is on or off?

Well, if it were off, it would not draw power. :-)

So IMO it is reasonable to assume that the logic on the drive stays
on.  I'm not aware of anything that can be done to turn it off,
however.

Re: Hard Disk consumes lots of power in s2idle

[Kai-Heng Feng]

> On Feb 20, 2020, at 18:34, Rafael J. Wysocki <rafael@kernel.org> wrote:
> 
> On Thu, Feb 20, 2020 at 11:25 AM Kai-Heng Feng
> <kai.heng.feng@canonical.com> wrote:
>> 
>> 
>> 
>>> On Feb 20, 2020, at 18:12, Rafael J. Wysocki <rafael@kernel.org> wrote:
>>> 
>>> On Thu, Feb 20, 2020 at 9:08 AM Kai-Heng Feng
>>> <kai.heng.feng@canonical.com> wrote:
>>>> 
>>>> Hi Srinivas,
>>>> 
>>>>> On Feb 20, 2020, at 02:36, Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> wrote:
>>>>> 
>>>>> Hi Kai,
>>>>> 
>>>>> On Wed, 2020-02-19 at 22:22 +0800, Kai-Heng Feng wrote:
>>>>>> Hi Srinivas,
>>>>>> 
>>>>>> Your previous work to support DEVSLP works well on SATA SSDs, so I am
>>>>>> asking you the issue I am facing:
>>>>>> Once a laptop has a HDD installed, the power consumption during
>>>>>> S2Idle increases ~0.4W, which is quite a lot.
>>>>>> However, HDDs don't seem to support DEVSLP, so I wonder if you know
>>>>>> to do proper power management for HDDs?
>>>>> What is the default here
>>>>> cat /sys/power/mem_sleep
>>>>> s2idle or deep?
>>>> 
>>>> It defaults to s2idle.
>>>> 
>>>>> 
>>>>> Please follow debug steps here:
>>>>> https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-linux
>>>>> 
>>>>> We need to check whether you get any PC10 residency or not.
>>>> 
>>>> Yes it reaches PC10. It doesn't reach SLP_S0 though.
>>>> The real number on S2Idle power consumption:
>>>> No HDD: ~1.4W
>>>> One HDD: ~1.8W
>>>> 
>>>> If the SoC doesn't hit PC10 the number should be significantly higher.
>>>> That's why I think the issue is the power management on HDD itself.
>>> 
>>> I'm assuming that you mean a non-SSD device here.
>> 
>> Yes, it's spinning rust here.
>> 
>>> 
>>> That would be handled via ata_port_suspend() I gather and whatever
>>> that does should do the right thing.
>>> 
>>> Do you think that the disk doesn't spin down or it spins down, but the
>>> logic stays on?
>> 
>> The spin sound is audible, so I am certain the HDD spins down during S2Idle.
> 
> OK
> 
>> How do I know if the logic is on or off?
> 
> Well, if it were off, it would not draw power. :-)
> 
> So IMO it is reasonable to assume that the logic on the drive stays
> on.  I'm not aware of anything that can be done to turn it off,
> however.

Okay, thanks for the info.
I'll ask vendors the possibility to handle it at system firmware level.

Kai-Heng