Comment 645 for bug 1690085

I assume there is several (at least two) similar problems that cause spontaneous system hangs.

One of them is mwait bug listed in AMD errata. Looks like idle=halt is partial workaround for this. But, as said in AMD community forum, guest OS in virtual machine may execute mait instruction and provoke this bug. Luckily, mwait is not common for user-space applications (still not sure about that).

Another is power supply problem. This may be caused by unsupported PSU (no 0A 12V) or unsuitable power subsystem on motherboard. This may be partially (again!) solved by BIOS "typical current idle" option or disabling C6 states by .py script.
IMO this bios option may be implemented not properly in some mobos firmware. Looks like it only "says to OS" do not use C-states, but does not prevent deep sleep on hw level. Here we got [Firmware Bug]: ACPI MWAIT C-state 0x0 not supported by HW (0x0) in dmesg and problem persists.

Have read some forums with assures of completely stable Ryzen systems work under Linux without any tweks. So not really too many systems affected by mwait and PSU lockups. I think some combination of factors may provoke this behavior. Mobo+PSU, memory latency (why not?) or some vendor-provided bios config.

Also in some cases these two problems may appear together.
Obviously, segfaults or amdgpu crashes is not related to this bug.

PS:Sorry for my bad English.