Comment 597 for bug 1690085

(In reply to Maxim Bakulin from comment #492)
> Created attachment 280669 [details]
> dmesg of freeze with 4.20.3 kernel and nomwait, rcu_nocbs, max_cstate applied
>
> some older info here:
> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1690085/comments/566
>
> I have three machines with new 2700x CPUs, and all three of them experience
> freezes in xubuntu 18.04 after some time of work. I use compiling QEMU with
> make -j16 in a loop to test for stability.
>
> I'm not sure it is the same bug, because I observe different behaviour: one
> machine, that was compiling QEMU, froze during the night, and the one left
> idle worked for 1 day. Another thing that SEEM to help is disabling SMT:

The Processor errata lists two bugs (SMT-related) for Ryzen 1 and Ryzen 2 (1095 and 1109) with status "no fix planned". If you disable MWAIT but enable SMT, you are left with bug 1095: "Potential Violation of Read Ordering In Lock Operation In SMT (Simultaneous Multithreading) Mode". This can cause crashes. Not necessarily the cause of your crashes :-D

# lsmsr -r 0xc0011020
warning: unknown MSR c0011020
unknown = 0x0006800000000010

On my Ryzen 1600X bit 57 (no idea what it does) is 0. (But I have nosmt=force.) Linux kernel doesn't seem to touch that bit.

Also, if you get "ACPI MWAIT C-state 0x0 not supported by HW (0x0)", mwait is not used by kernel.