Re: Changing the default sata link-power-management policy for mobile chipsets
Hi,
On 12/18/2017 07:27 PM, Laura Abbott wrote:
On 12/12/2017 12:04 AM, Hans de Goede wrote:
> Hi All,
>
> As part of: https://fedoraproject.org/wiki/Changes/ImprovedLaptopBatteryLife
> I've been working on some kernel patches to allow configuring the default
> sata link-power-management policy for mobile chipsets. Before this the
> kernel would always default to leaving things as setup by the BIOS.
>
> These patches have been accepted upstream and are queued for merging
> into 4.16:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/tj/libata.git/log/?h=for-...
>
> I would like to add these patches to the Fedora 4.15 kernels *and*
> set the new CONFIG_SATA_MOBILE_LPM_POLICY Kconfig option to "3"
> which translates to "med_power_with_dipm".
>
> The last bit, setting the default for mobile chipsets to
> "med_power_with_dipm" is the important bit here.
>
> On most laptops with a SATA disk / SSD this leads to power-savings of
> 1.0 - 1.5 Watts (when idle). Modern ultrabooks idle around 6W (at 50%
> screen brightness), 1.0 - 1.5W is a significant chunk of this.
>
> But this change is not entirely uncontroversial, enabling LPM by
> default is not entirely without risk of regressions. At least min_power
> is known to cause issues with some disks, including some reports of data
> corruption.
>
> med_power_with_dipm option which mirrors the Intel RST Windows drivers
> defaults and as such should be pretty safe, but the proof is in the
> pudding. I've done a blog-post a while ago to ask users to test this:
> https://hansdegoede.livejournal.com/18412.html
>
> Test results from this can be found here:
> https://fedoraproject.org/wiki/Changes/ImprovedLaptopBatteryLife#How_To_Test
>
> All testing so far has shown that the med_power_with_dipm option seems to
> be safe, even with SSDs which are known to corrupt data with the min_power
> setting!
>
> Note that the patches provide a ahci.mobile_lpm_policy kernel cmdline option,
> so for debugging disk issues users can simply pass ahci.mobile_lpm_policy=0
> on the kernel commandline.
>
> Summarizing: I would like the Fedora kernel's team permission to enable
> med_power_with_dipm SATA LPM policy by default for mobile chipsets, this is
> not entirely without risk but I believe it is safe and the power-savings to
> be had are simply to good to leave them on the table.
>
> I hereby promise that the Fedora kernel team can re-assing any SATA related
> bugs to me for the F28 cycle and I will take care of them ASAP.
>
> Regards,
>
> Hans
>
>
> p.s.
>
> 1) Note that these power-savings have been laying on the table since 2015:
> https://mjg59.dreamwidth.org/34868.html
> I really believe that we should not leave these on the table for another
> 2+ years.
>
> 2) This clearly is a rawhide / F28 only thing and should not be carried
> over to F27 / F26 once 4.15 gets into the stabilization branch.
I talked with Justin and you can go ahead and turn it on in rawhide.
Justin pointed out a) we should get more testing before it goes
upstream b) it is easy to turn off.
This is something that has the potential to cause corruption though
so I think this deserves a post a few more places (devel@ ? users@?)
for wider information.
Ok, I've just send a heads up mail about this to the Fedora devel and
users lists. I will push the changes to pkgs.fedoraproject.org/kernel
soon, so this will be in the next build you kick of.
Regards,
Hans