[Bug 109051] intel_idle.max_cstate=1 required on baytrail to prevent crashes
Robert Small
robert at robertsmall.org
Thu Sep 22 00:07:33 CEST 2016
As requested.
# cstateInfo.sh
cpu0 State Name Disabled Latency Residency Time Usage
0 POLL 0 0 0 320349 23243
1 C1-BYT 0 1 1 35018930703 14858769
2 C6N-BYT 1 300 275 225138 319
3 C6S-BYT 1 500 560 1517455 999
4 C7-BYT 0 1200 4000 34965703628 6557188
5 C7S-BYT 0 10000 20000 6707737452 564406
cpu1 State Name Disabled Latency Residency Time Usage
0 POLL 0 0 0 144668 14927
1 C1-BYT 0 1 1 15728639117 4710264
2 C6N-BYT 1 300 275 114082 172
3 C6S-BYT 1 500 560 1410496 624
4 C7-BYT 0 1200 4000 50011254628 7988000
5 C7S-BYT 0 10000 20000 13239825298 605157
cpu2 State Name Disabled Latency Residency Time Usage
0 POLL 0 0 0 781399 31216
1 C1-BYT 0 1 1 14661348811 2820901
2 C6N-BYT 1 300 275 293099 423
3 C6S-BYT 1 500 560 1615360 1021
4 C7-BYT 0 1200 4000 12125826131 1639244
5 C7S-BYT 0 10000 20000 22839073312 728724
cpu3 State Name Disabled Latency Residency Time Usage
0 POLL 0 0 0 1046058 33476
1 C1-BYT 0 1 1 18025094795 3457563
2 C6N-BYT 1 300 275 321547 584
3 C6S-BYT 1 500 560 1689307 896
4 C7-BYT 0 1200 4000 12417037473 1725348
5 C7S-BYT 0 10000 20000 20938233703 922453
# uptime
23:04:17 up 1 day, 1 min, 4 users, load average: 1.15, 1.04, 1.01
Thanks,
*Robert Small*
e: Robert at RobertSmall.org
On 21 September 2016 at 09:23, Robert Small <robert at robertsmall.org> wrote:
> Sending from my phone outside school gates so I'm not able to get all the
> details. However, previously on 4.5.7 where I don't need the cstates
> parameter at all I've found it to be significantly cooler and by manually
> checking and trying to remember it seemed to be lower power usage
> (powertop).
>
> Incidentally, those commandline parameters were what was needed to drop
> the cstates parameter from the commandline with 4.5.7. Not sure how many
> are needed now.
>
> Another possibility is that some of these crashes are software not kernel.
> I went from stable to testing yesterday and everything kept crashing. I
> upgraded to unstable which has proven to be stable to the same level as
> stable. Also, I noticed one crash which recovered itself.
>
> Thanks,
> Robert.
>
> On Wed, 21 Sep 2016, 8:40 a.m. Nicolas Huillard, <nicolas at huillard.net>
> wrote:
>
>> Le mardi 20 septembre 2016 à 20:19 +0100, Robert Small a écrit :
>> > Not only can I confirm that this works it makes sound stable on
>> 4.7.3.10!
>>
>> Great info, thank you Robert !
>>
>> > cpu0 State Name Disabled Latency Residency Time Usage
>> > 0 POLL 0 0 0 139827 655
>> > 1 C1-BYT 0 1 1 278968871 443703
>> > 2 C6N-BYT 1 300 275 529818 620
>> > 3 C6S-BYT 1 500 560 1498249 1034
>> > 4 C7-BYT 0 1200 4000 42598904 11160
>> > 5 C7S-BYT 0 10000 20000 5278866 421
>> ...
>>
>> I suppose you can confirm that C6* time and usage doesn't change after
>> starting the script. C7* are not used that much, but that's probably
>> because you got this snapshot shortly after boot/tests.
>>
>> > cat /proc/cmdline
>> > BOOT_IMAGE=/boot/vmlinuz-4.7.3.10 root=/dev/mapper/vgDrudwyn-root64 ro
>> > i915.modeset=1 threadirqs intel_iommu=igfx_off clocksource=tsc
>> tsc=reliable
>> > sdhci.debug_quirks=0x10000 quiet splash
>>
>> Wow... That's a lot of parameters... I'll check mine when I can.
>>
>> What about battery life or device temperature when you leave it alone
>> (it's a bit warm when I leave it alone without C7*) ?
>>
>> --
>> NH
>>
>> _______________________________________________
>> click mailing list
>> click at tinc-vpn.org
>> https://www.tinc-vpn.org/cgi-bin/mailman/listinfo/click
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.tinc-vpn.org/pipermail/click/attachments/20160921/41e32bb1/attachment.html>
More information about the click
mailing list