Re: Xen Domu kernel crash at start of boot

[Jaromír Doleček <jaromir.dolecek%gmail.com@localhost>]

2018-06-22 2:54 GMT+02:00 Chuck Zmudzinski <frchuckz%gmail.com@localhost>:
> I am getting a kernel crash almost immediately after booting the current
> kernel. I am running NetBSD/xen amd64 on a Debian Linux 8.10 DOM0 which uses
> Xen-4.4. Last week's kernel was good. I built a kernel from a cvs update a
> couple of days ago and tried it. It crashed. I tried the most recent daily
> snapshot available from NetBSD daily builds. It crashed too. Here is the
> information from the console about the daily snapshot kernel that crashed
> (it was built earlier today):
> [   1.0000000] vcpu0: Intel(R) Core(TM) i5-4590S CPU @ 3.00GHz, id 0x306c3
> [   1.0000000] vcpu0: package 0, core 3, smt 0
> [   1.0000000] vcpu1 at hypervisor0
> [   1.0000000] vcpu1: Intel(R) Core(TM) i5-4590S CPU @ 3.00GHz, id 0x306c3
> [   1.0000000] vcpu1: package 0, core 3, smt 0
> [   1.0000000] xenbus0 at hypervisor0: Xen Virtual Bus Interface
> [   1.0000000] xencons0 at hypervisor0: Xen Virtual Console Driver
> [   1.0000030] fatal protection fault in supervisor mode
> [   1.0000030] trap type 4 code 0 rip 0xffffffff80205968 cs 0x10000e030
> rflags 0x10046 cr2 0 ilevel 0 rsp 0xffffa000a570fbf0
> [   1.0000030] curlwp 0xffffa0000642d4a0 pid 0.15 lowest kstack
> 0xffffa000a570b2c0
> kernel: protection fault trap, code=0
> Stopped in pid 0.15 (system) at ffffffff80205968:       fxsavel

That would be my fault.

Can you send me moral equivalent of "cpuctl identify 0" from the DOM0?
I want to know what CPUID is saying about supported features on the
CPU.

Can you also check whether you use no-xsave flag for your DOM0 by
chance? It should not be needed on Intel CPUs.

Meanwhile this change in sys/arch/xen/x86/cpu.c can be done to avoid this:

@@ -551,7 +551,7 @@ cpu_init(struct cpu_info *ci)
  * does, here we only set CR4_OSXSAVE if the feature is already
  * enabled according to CPUID.
  */
- if (cpu_feature[1] & CPUID2_OSXSAVE)
+ if (0 && cpu_feature[1] & CPUID2_OSXSAVE)
  cr4 |= CR4_OSXSAVE;
  else {
  x86_xsave_features = 0;

The change was tested on Xen 4.2 and Xen 4.8, I wonder if Xen 4.4 has
yet another quirk. Any chance you could try your DOM0 updated to newer
Xen?

Jaromir