Comment 508 for bug 1690085

(In reply to Michaël Colignon from comment #420)
> ...
> I want to add another thing, about fixing it directly from the kernel.
> As typical idle current seems (i am testing it again for see) not needed
> under the other OS, that's because the core parking under this other OS is
> partially disabled by DEFAULT on this OS as i said earlier in this thread.
>
> So it seems possible to fix it via the kernel, without make use of an uefi
> parameters not enabled by default...

I can perhaps confirm that. But I have to say first, I'm not a developer and I do not understand anything about Linux.

I use an AMD Ryzen 5 1600 Six-Core @ 12x 3.2GHz with Gentoo as the operating system. So it's very easy for me to build a kernel just for my machine. (Without the whole Intel, ARM64, Android and Big-Data stuff.)

I try the kernels with standard bios setup, without kernel parameters like "idle = nomwait", without OC. My bios only has Global C-states Auto (which is default) and enable/disable options. There is no "Idle Control to "Typical Current Idle"" in the BIOS. Only CONFIG_RCU_NOCB_CPU=y is set in the .config.

BOOT_IMAGE=/boot/kernel-x86_64-4.19.0-rc3 root=UUID=465f9c93-09d0-441b-9440-8ec7799b557c ro real_root=/dev/sdd4 resume=/dev/sdd3 init=/usr/lib/systemd/systemd
rootfstype=ext4

./zenstates.py -l
...
C6 State - Package - Enabled
C6 State - Core - Enabled
...
On this basis,
It is to be confirmed by me that the kernel 4.17.14 works really well. But the series 4.18.x is cruel. Also the currently as "stable" declarated 4.18.7. Bad. Freeze or reboot every 2 - 3 hours.

To my surprise, the 4.19.0-rcX work well. So far no soft lockups.

I do not know why. But these are the subjective results of my playing with the kernels.

I can also understand that the few kernel developers are working on priority issues. Please think of Specter and Meltdown.

But a few clarifying information would be very welcome..