Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

[-- Attachment #1: Type: text/plain, Size: 137 bytes --]

Hi,

The _OSI("Darwin") support seems to break CPU power management for Attila.

Is there anything we can possibly do about that?

Rafael

[-- Attachment #2: bugzilla-daemon@bugzilla.kernel.org: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19 --]
[-- Type: message/rfc822, Size: 2449 bytes --]

From: bugzilla-daemon@bugzilla.kernel.org
To: rjw@rjwysocki.net
Subject: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19
Date: Tue, 10 Feb 2015 23:50:19 +0000
Message-ID: <bug-92111-76231-fcRlm4ifRg@https.bugzilla.kernel.org/>

https://bugzilla.kernel.org/show_bug.cgi?id=92111

--- Comment #32 from Attila <attila.jakosa@gmail.com> ---
Strange result, but git bisect shows this commit to be the first bad:

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=7bc5a2bad0b8d9d1ac9f7b8b33150e4ddf197334

I guess this acpi change results that CPU is not allowed to go deeper states
than pc3. I will do a build of 3.18 reverting this commit to make sure this was
the problem.

I will enclose a bisect log soon.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 01:29:36AM +0100, Rafael J. Wysocki wrote:
> Hi,
> 
> The _OSI("Darwin") support seems to break CPU power management for Attila.
> 
> Is there anything we can possibly do about that?

Hm. It's not clear to me what the bug is describing. If we're not waking 
up more frequently, why aren't we entering PC6?
-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 12:26:02 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 01:29:36AM +0100, Rafael J. Wysocki wrote:
> > Hi,
> > 
> > The _OSI("Darwin") support seems to break CPU power management for Attila.
> > 
> > Is there anything we can possibly do about that?
> 
> Hm. It's not clear to me what the bug is describing. If we're not waking 
> up more frequently, why aren't we entering PC6?

Good question. :-)

The bug is saying, pretty much, that after a positive _OSI response to "Darwin"
we cannot enter PC6 any more on that machine all other things being equal.

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 01:53:26AM +0100, Rafael J. Wysocki wrote:
> On Wednesday, February 11, 2015 12:26:02 AM Matthew Garrett wrote:
> > Hm. It's not clear to me what the bug is describing. If we're not waking 
> > up more frequently, why aren't we entering PC6?
> 
> Good question. :-)
> 
> The bug is saying, pretty much, that after a positive _OSI response to "Darwin"
> we cannot enter PC6 any more on that machine all other things being equal.

We're using intel_idle, so we should be making idle state determination 
regardless of what ACPI tells us. So is Darwin mode influencing the 
state of some MSR that controls whether PC6 can be entered? Is the GPU 
blocking it because of some configuration difference? I don't know 
enough about how the chip decides to enter PC6 to be able to tell.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 12:33:18 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 01:53:26AM +0100, Rafael J. Wysocki wrote:
> > On Wednesday, February 11, 2015 12:26:02 AM Matthew Garrett wrote:
> > > Hm. It's not clear to me what the bug is describing. If we're not waking 
> > > up more frequently, why aren't we entering PC6?
> > 
> > Good question. :-)
> > 
> > The bug is saying, pretty much, that after a positive _OSI response to "Darwin"
> > we cannot enter PC6 any more on that machine all other things being equal.
> 
> We're using intel_idle, so we should be making idle state determination 
> regardless of what ACPI tells us. So is Darwin mode influencing the 
> state of some MSR that controls whether PC6 can be entered?

I need to confirm that with Len, but I don't recall there being an MSR
controlling the PC6 availability on Haswell.  There is one for PC7, but
not for PC6 IIRC.

> Is the GPU blocking it because of some configuration difference?

That is possible.

> I don't know enough about how the chip decides to enter PC6 to be able to
> tell.

Well, so honestly how do you expect me to respond here?


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 02:02:06AM +0100, Rafael J. Wysocki wrote:
> On Wednesday, February 11, 2015 12:33:18 AM Matthew Garrett wrote:
> > We're using intel_idle, so we should be making idle state determination 
> > regardless of what ACPI tells us. So is Darwin mode influencing the 
> > state of some MSR that controls whether PC6 can be entered?
> 
> I need to confirm that with Len, but I don't recall there being an MSR
> controlling the PC6 availability on Haswell.  There is one for PC7, but
> not for PC6 IIRC.

Ok.

> > Is the GPU blocking it because of some configuration difference?
> 
> That is possible.

I know that it has some say in things here.

> > I don't know enough about how the chip decides to enter PC6 to be able to
> > tell.
> 
> Well, so honestly how do you expect me to respond here?

We really need feedback from people who know how the CPU makes its PC6 
determination - the firmware has clearly done something differently, but 
without knowing what factors feed into that determination we're pretty 
much helpless.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 12:42:12 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 02:02:06AM +0100, Rafael J. Wysocki wrote:
> > On Wednesday, February 11, 2015 12:33:18 AM Matthew Garrett wrote:
> > > We're using intel_idle, so we should be making idle state determination 
> > > regardless of what ACPI tells us. So is Darwin mode influencing the 
> > > state of some MSR that controls whether PC6 can be entered?
> > 
> > I need to confirm that with Len, but I don't recall there being an MSR
> > controlling the PC6 availability on Haswell.  There is one for PC7, but
> > not for PC6 IIRC.
> 
> Ok.
> 
> > > Is the GPU blocking it because of some configuration difference?
> > 
> > That is possible.
> 
> I know that it has some say in things here.
> 
> > > I don't know enough about how the chip decides to enter PC6 to be able to
> > > tell.
> > 
> > Well, so honestly how do you expect me to respond here?
> 
> We really need feedback from people who know how the CPU makes its PC6 
> determination - the firmware has clearly done something differently, but 
> without knowing what factors feed into that determination we're pretty 
> much helpless.

We may be able to get that feedback, but what do you suggest we do in the
meantime?  Nothing, I suppose?

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 02:09:17AM +0100, Rafael J. Wysocki wrote:

> We may be able to get that feedback, but what do you suggest we do in the
> meantime?  Nothing, I suppose?

We have the choice between increased power consumption or broken 
Thunderbolt - either choice is going to upset somebody. My gut feeling 
is that leaving us at increased power consumption might encourage 
someone at Intel to try to fix things, whereas nobody at Intel is going 
to care about Thunderbolt hotplug.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 12:48:04 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 02:09:17AM +0100, Rafael J. Wysocki wrote:
> 
> > We may be able to get that feedback, but what do you suggest we do in the
> > meantime?  Nothing, I suppose?
> 
> We have the choice between increased power consumption or broken 
> Thunderbolt - either choice is going to upset somebody. My gut feeling 
> is that leaving us at increased power consumption might encourage 
> someone at Intel to try to fix things,

Fix things how exactly?

> whereas nobody at Intel is going to care about Thunderbolt hotplug.

Apple is the only party that can fix anything in Apple products as
far as I can say.

Anyway, thanks for your input!

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 02:14:29AM +0100, Rafael J. Wysocki wrote:
> On Wednesday, February 11, 2015 12:48:04 AM Matthew Garrett wrote:
> > On Wed, Feb 11, 2015 at 02:09:17AM +0100, Rafael J. Wysocki wrote:
> > 
> > > We may be able to get that feedback, but what do you suggest we do in the
> > > meantime?  Nothing, I suppose?
> > 
> > We have the choice between increased power consumption or broken 
> > Thunderbolt - either choice is going to upset somebody. My gut feeling 
> > is that leaving us at increased power consumption might encourage 
> > someone at Intel to try to fix things,
> 
> Fix things how exactly?

Identify why the Intel CPU is refusing to enter PC6 despite all its 
cores being in C6 or deeper. Once we've done that we can figure out 
what's actually causing that change in behaviour and the right place to 
fix it.

> > whereas nobody at Intel is going to care about Thunderbolt hotplug.
> 
> Apple is the only party that can fix anything in Apple products as
> far as I can say.

There's an Intel part number on everything involved here...

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 12:53:37 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 02:14:29AM +0100, Rafael J. Wysocki wrote:
> > On Wednesday, February 11, 2015 12:48:04 AM Matthew Garrett wrote:
> > > On Wed, Feb 11, 2015 at 02:09:17AM +0100, Rafael J. Wysocki wrote:
> > > 
> > > > We may be able to get that feedback, but what do you suggest we do in the
> > > > meantime?  Nothing, I suppose?
> > > 
> > > We have the choice between increased power consumption or broken 
> > > Thunderbolt - either choice is going to upset somebody. My gut feeling 
> > > is that leaving us at increased power consumption might encourage 
> > > someone at Intel to try to fix things,
> > 
> > Fix things how exactly?
> 
> Identify why the Intel CPU is refusing to enter PC6 despite all its 
> cores being in C6 or deeper. Once we've done that we can figure out 
> what's actually causing that change in behaviour and the right place to 
> fix it.
> 
> > > whereas nobody at Intel is going to care about Thunderbolt hotplug.
> > 
> > Apple is the only party that can fix anything in Apple products as
> > far as I can say.
> 
> There's an Intel part number on everything involved here...

Not really on everything and the whole machine is still an Apple product.

You don't ask part suppliers to fix your car, do you?

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 02:20:26AM +0100, Rafael J. Wysocki wrote:

> Not really on everything and the whole machine is still an Apple product.
> 
> You don't ask part suppliers to fix your car, do you?

We don't ask system vendors to fix most bugs we find, because most of 
them won't talk to us. But we still try to do what we can with the 
contacts we have available. As far as we can tell there's no additional 
system wakeups here, so any failure to enter PC6 is because the CPU is 
in a different configuration. We have a good working relationship with 
Intel, so asking them if they can help us figure out what that different 
configuration is seems like a reasonable thing to do? I'm happy to take 
responsibility for working on it from that point, but I don't believe 
any of the documentation I'd need to find the starting point is public.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 01:01:50 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 02:20:26AM +0100, Rafael J. Wysocki wrote:
> 
> > Not really on everything and the whole machine is still an Apple product.
> > 
> > You don't ask part suppliers to fix your car, do you?
> 
> We don't ask system vendors to fix most bugs we find, because most of 
> them won't talk to us. But we still try to do what we can with the 
> contacts we have available. As far as we can tell there's no additional 
> system wakeups here, so any failure to enter PC6 is because the CPU is 
> in a different configuration. We have a good working relationship with 
> Intel, so asking them if they can help us figure out what that different 
> configuration is seems like a reasonable thing to do? I'm happy to take 
> responsibility for working on it from that point, but I don't believe 
> any of the documentation I'd need to find the starting point is public.

Well, it looks like we circled back to http://marc.info/?l=linux-acpi&m=142361557824868&w=4 :-)

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 02:57:48AM +0100, Rafael J. Wysocki wrote:

> Well, it looks like we circled back to http://marc.info/?l=linux-acpi&m=142361557824868&w=4 :-)

Right. And, like I said, if there's a "We're seeing increased power 
consumption" bug then there's an incentive for people from Intel to 
actually help identify the root source, whereas if there's a "We can't 
enable Thunderbolt hotplug on this platform Intel pretend has nothing to 
do with them because otherwise we'll see increased power consumption", 
it's going to be so far down people's list of priorities that it's 
probably not going to happen.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Rafael J. Wysocki]

On Wednesday, February 11, 2015 01:43:06 AM Matthew Garrett wrote:
> On Wed, Feb 11, 2015 at 02:57:48AM +0100, Rafael J. Wysocki wrote:
> 
> > Well, it looks like we circled back to http://marc.info/?l=linux-acpi&m=142361557824868&w=4 :-)
> 
> Right. And, like I said, if there's a "We're seeing increased power 
> consumption" bug then there's an incentive for people from Intel to 
> actually help identify the root source, whereas if there's a "We can't 
> enable Thunderbolt hotplug on this platform Intel pretend has nothing to 
> do with them because otherwise we'll see increased power consumption", 
> it's going to be so far down people's list of priorities that it's 
> probably not going to happen.

And you said that already too, so I guess we can circle like that forever.

Only one more comment:

In order to "identify the root cause" it is pretty much necessary to know
what _OSI("Darwin") causes the firmware to actually do and the only party
knowing that is Apple.  The only question I can ask hardware people is what
it *theoretically* may be doing to trigger the observed behavior and
honestly I doubt I'll get a useful answer to that.

Re: Fwd: [Bug 92111] Power regression from 3.17 to 3.18 and 3.19

[Matthew Garrett]

On Wed, Feb 11, 2015 at 03:26:35AM +0100, Rafael J. Wysocki wrote:

> In order to "identify the root cause" it is pretty much necessary to know
> what _OSI("Darwin") causes the firmware to actually do and the only party
> knowing that is Apple.  The only question I can ask hardware people is what
> it *theoretically* may be doing to trigger the observed behavior and
> honestly I doubt I'll get a useful answer to that.

The hardware has a fixed set of inputs that influence whether it enters 
PC6. If we know what those are then we can identify which is actually at 
play here, which gives us a high probability of identifying the root 
cause.

-- 
Matthew Garrett | mjg59@srcf.ucam.org