[solved] Throttled CPU after Kernel upgrade 5.4 -> 5.10

[bheinze]

Hello everyone,

some time ago I upgraded the kernel gentoo-sources from 5.4.97 to 5.10.27. Afterwards I noticed that the CPU doesn't get so hot anymore while compiling ebuilds. Coincidently the ebuilds now take approximately 2.5 times longer than with the old kernel. So it seems like the CPU gets somehow throttled.

What I tried so far:

• Update kernel config with make olddefconfig (first I used make oldconfig)
  
• compare the output of lscpu
  
• compare the output of dmidecode
  
• compare the output of /proc/cpuinfo

The only difference I found is that with kernel 5.4 I have Virtualization VT-x active. However I'm not using it actively, and I didn't knowingly disable it for kernel 5.10.

I'm running a Thinkpad T430 with an Intel i5-3320M cpu. Has anyone a pointer where to look for the cause of the slow system?

Thanks for your help!

[Ralphred]

Check your

[Goverp]

AFAIR sometime in the recent past few kernels, the governors defaults started changing between ondemand and schedutil and possibly even powersave. Check your config for _CONFIG_CPU_FREQ entries in case you're not getting what you expect.
_________________
Greybeard

[Ralphred]

[bheinze]

Thanks all, the governor was a good starting point, however in fact the intel_pstate-driver is to blame:

With kernel 5.4.97:

[dmpogo]

I always considered intel_pstate to be superior governor (when it uses its one). It is interesting that newer kernels have intel-pstate that dropped support for old CPU's that where previously supported. If so, for me it would be an argument to stick to older kernel.

[ilightwas]

[Jaglover]

[eccerr0r]

I'd call this a bug, even if it still functions correctly...

Then again nobody calls computer slowdowns by software changes as bugs, they call it... "new features"

(Suspecting I'll run into this issue on my machine, sigh. At least it won't overheat in the current heat wave.)
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

[figueroa]

This calls for a do-over. Do you still have your kernel 5.4.97 configuration. Upgrade to current stable 5.4 (now 109) using make oldconfig. Then, be happy, or try to upgrade again to 5.10, but using make oldconfig. (Me, I'm using 5.4 for next 3-4 years.)
_________________
Andy Figueroa
hp pavilion hpe h8-1260t/2AB5; spinning rust x3
i7-2600 @ 3.40GHz; 16 gb; Radeon HD 7570
amd64/23.0/split-usr/desktop (stable), OpenRC, -systemd -pulseaudio -uefi

[eccerr0r]

Incidentally it looks like I upgraded my i5-3317U to 5.10.27 a few weeks ago, and apparently I had the config file set in a way such that transferring the 5.4.97 config seemed to not make that much of a difference in speed by default. Perhaps I had more than the absolute minimum selected. Normally I make menuconfig and do a quick check of the options when I upgrade, so it might be a different situation here...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

[figueroa]

[eccerr0r]

I'm not exactly sure what I had in my .config, but every time a kernel comes out, I copy the .config to the new kernel and run menuconfig from scratch, I never run oldconfig. Menuconfig tends to also correct changes but you're right, it usually does not tell me what are no longer available (sometimes when I exit it gives a few warnings). However I do scan over all the options to see if there are any that are blatantly incorrect, like making sure the network drivers are still there, etc.

I have not gone through the update on my big iron machine (the i7) yet, it's been having overheating problems. The i5 has been as well, but it seems to always boot properly - just throttles and sometimes gets real slow and later pops back to normal. Not so with my i7...it doesn't even seem to want to automatically throttle when it's at the 90°C mark, just overheats and beeps...

Gone are the days of the core2 machines behavior...always staying at the same speed. Then again turboboost does sound like the right solution if you're only running one thread. Alas when running Gentoo and running make -j ... turbo boost is not going to help that. When the weather is cooler the i7 isn't too much slower when running all cores versus one core (3.9GHz vs 3.5GHz), but the i5 it's a dog when it can't turboboost (2.6GHz vs 1.7GHz).

Despite that, oddly enough, the (newer) i5 with 2C4T and turboboost issues, it still gives my Core2 Quads (4C4T 2.6GHz 100%) a run for their money building rust...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

[figueroa]

Sounds like time to reseat the i5 & i7 CPUs and give the heatsink a good blow. Nothing quite like hauling the whole box outside and putting the shop-vac on blow.
_________________
Andy Figueroa
hp pavilion hpe h8-1260t/2AB5; spinning rust x3
i7-2600 @ 3.40GHz; 16 gb; Radeon HD 7570
amd64/23.0/split-usr/desktop (stable), OpenRC, -systemd -pulseaudio -uefi

[eccerr0r]

Nah, it's just hot out (or in) now. When ambient is 27°C there's just that much less thermal margin.

What's weird is that core2 quads don't seem to get as hot as the i5/i7 do, which is somewhat baffling to me. Maybe the sensors are misreporting CPU temperatures but they stay just a few degrees above ambient, maybe only 2°C or so when idling, unlike the i5/i7 which usually stay 8-10°C when idling. I think all are using their stock CPU HS/F's.

Granted the i5 is a laptop.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

[ilightwas]

[sl70]

I had the same problem. I read this thread but didn't know how to add the intel_pstate kernel option, and couldn't tell from the other messages in the thread whether it was even a good thing to do. However, I found this thread that suggested changing one line in the .config file: