Skylake: intel_pstate: fix PCT_TO_HWP macro

Zhang, Xiong Y xiong.y.zhang at intel.com
Tue Sep 29 09:02:33 UTC 2015


I tried the latest 15.10 kernel with "fix PCT_TO_HWP macro" patch, it indeed have system hang on SKL-S machine.
Since upstream is still debugging this issue and seems the fix couldn't catch up 15.10 kernel freeze, I suggest Canonical to disable HWP on SKL-S with the attachment patch. 
The attachment patch is really small and won't generate regression, it only disable HWP on SKL-S and let legacy intel p state driver take effect on SKL-S. HWP is still enabled on SKL-Y/U

thanks
> -----Original Message-----
> From: Jason Gerard DeRose [mailto:jason at system76.com]
> Sent: Saturday, September 26, 2015 12:10 AM
> To: Wang, Xiaolong; Zhang, Xiong Y
> Cc: Ubuntu Kernel Team
> Subject: Re: Skylake: intel_pstate: fix PCT_TO_HWP macro
> 
> Wang,
> 
> On 09/24/2015 09:01 PM, Wang, Xiaolong wrote:
> > Hi Jason,
> > We tested
> >    a SKL-S Beta machine
> >      (Processor Stepping in BIOS is Q0, F/M/S is 6/94/1), and
> >    a SKL-S Qual machine
> >      (Processor Stepping in BIOS is R0/S0, F/M/S is 6/94/3)
> > On the Beta machine, Processor Frequency are found tied at the highest
> > possible value -- HWP does not work;
> > On the Qual machine, it hangs just after HWP enabled on each core.
> 
> Thanks for sharing these insights.
> 
> > Upstream have reproduced the hang on a Beta machine (weird),
> > but they say removing "quiet" from boot parameters can boot to OS.
> > I tried removing "quiet" on my Qual machine, it didn't work.
> 
> Yeah, I'd be kinda doubtful that "quiet" would have any effect either
> way, but I guess weirder things have happened :)
> 
> > As I know, upstream is still checking it but had not get the root cause
> > of that hang, and there is not a clear schedule to get that solved.
> 
> Thank for the info!
> -Jason
> 
> > -Wang, Xiaolong
> >
> > On 2015/9/25 9:25, Zhang, Xiong Y wrote:
> >> Hi, Xiaolong:
> >>     Could you help me answer this ?
> >>
> >> thanks
> >>> -----Original Message-----
> >>> From: Jason Gerard DeRose [mailto:jason at system76.com]
> >>> Sent: Thursday, September 24, 2015 10:44 PM
> >>> To: Zhang, Xiong Y; Ubuntu Kernel Team
> >>> Subject: Re: Skylake: intel_pstate: fix PCT_TO_HWP macro
> >>>
> >>> Thus far I've been testing this on SKL-S systems (one laptop, one
> >>> desktop) and haven't experienced any system hangs.
> >>>
> >>> Can you share any insight into the conditions under which these hangs
> >>> occur, so I can try to reproduce it?
> >>>
> >>> Also, when you say that hangs occur on "some SKL-S machines", are there
> >>> any common denominators among the problematic machines? Or put
> another
> >>> way, are there some SKL-S machines that are completely immune to these
> >>> hangs, and if so, what's the difference?
> >>>
> >>> Finally, an idea on the timeline for the complete fix?
> >>>
> >>> Thanks!
> >>>
> >>> On 09/23/2015 06:27 PM, Zhang, Xiong Y wrote:
> >>>> Yes, this patch will resolve p states on SKL-Y and SKL-U, but it also
> generate
> >>> system hang on some SKL-S machines. We are still waiting extra patches to
> fix
> >>> it.
> >>>>
> >>>> thanks
> >>>>> -----Original Message-----
> >>>>> From: kernel-team-bounces at lists.ubuntu.com
> >>>>> [mailto:kernel-team-bounces at lists.ubuntu.com] On Behalf Of Jason
> Gerard
> >>>>> DeRose
> >>>>> Sent: Thursday, September 24, 2015 2:52 AM
> >>>>> To: Ubuntu Kernel Team
> >>>>> Subject: Skylake: intel_pstate: fix PCT_TO_HWP macro
> >>>>>
> >>>>> I think commit 74da56ce5c6715630aed3ccc0fcb86a9210c1a56 is very
> >>> worthy
> >>>>> of cherry-picking before the kernel freeze.
> >>>>>
> >>>>>    From the bug I just filed:
> >>>>> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1499040
> >>>>>
> >>>>> The intel pstate driver is currently quite broken on on Skylake hardware.
> >>>>>
> >>>>> The symptom is that the pstate driver can scale the frequency up, but
> >>>>> wont then scale it back down. So as soon as you hit a decent amount of
> >>>>> CPU load, you get stuck fluctuating in the turbo frequency ranges, never
> >>>>> even dropping as low as the base clock frequency. Obviously this has
> >>>>> very negative consequences for battery life.
> >>>>>
> >>>>> Commit 74da56ce5c6715630aed3ccc0fcb86a9210c1a56 fixes this
> >>> upstream,
> >>>>> which is included in 4.3rc2:
> >>>>>
> >>>>>
> >>>
> https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=74d
> >>>>> a56ce5c6715630aed3ccc0fcb86a9210c1a56
> >>>>>
> >>>>> I built a custom Ubuntu kernel with this commit and tested it on a
> >>>>> laptop with a Skylake i7-6700 CPU. As measured by powerstat, this
> commit
> >>>>> dropped the idle power consumption by 11 watts (from 34 to 23 watts,
> >>>>> screen backlight was off during the test).
> >>>>>
> >>>>> I also tested this custom kernel on Haswell and Ivy Bridge hardware to
> >>>>> make sure it didn't introduce any behaviour changes in the pstate driver
> >>>>> on older hardware. As far as I can tell, on this older hardware pstate
> >>>>> is behaving exactly the same with and without this commit.
> >>>>>
> >>>>> Thanks!
> >>>>>
> >>>>> --
> >>>>> kernel-team mailing list
> >>>>> kernel-team at lists.ubuntu.com
> >>>>> https://lists.ubuntu.com/mailman/listinfo/kernel-team
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Disable-HWP-on-Skylake-S-processors.patch
Type: application/octet-stream
Size: 790 bytes
Desc: 0001-Disable-HWP-on-Skylake-S-processors.patch
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20150929/3c8dc580/attachment.obj>


More information about the kernel-team mailing list