XenServer and viridian

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On XenServer/XCP-ng VMs can have 'Viridian enlightenments' enabled. This 
is used by Windows VMs to get some features that would otherwise require 
Hyper-V, for instance, VM generation counter as used by Active Directory 
to spot it if is has been rolled back.

These shouldn't affect any non-Windows VMs, but NetBSD mistakenly thinks 
it is running on Hyper-V rather than Xen, so the PVH code is not enabled.

I tried the following naive patch:

--- sys/arch/x86/x86/identcpu.c 2 Feb 2025 14:51:59 -0000       1.123.4.2
+++ sys/arch/x86/x86/identcpu.c 26 Mar 2025 17:20:44 -0000
@@ -1132,8 +1132,22 @@
                                vm_guest = VM_GUEST_VMWARE;
                        else if (memcmp(hv_vendor, "Microsoft Hv", 12) == 0) {
                                vm_guest = VM_GUEST_HV;
+                               p = pmf_get_platform("system-vendor");
+                               if (p != NULL) {
+                                       if (strncmp(p, "Xen", 3) == 0) {
+                                               /*
+                                                * XenServer VMs may have
+                                                * Viridian enlightenments
+                                                * enabled which may make
+                                                * Hyper-V be falsely matched
+                                                */
+                                               vm_guest = VM_GUEST_XENHVM;
+                                       }
+                               }
 #if NHYPERV > 0
-                               hyperv_early_init();
+                               if (vm_guest == VM_GUEST_HV) {
+                                       hyperv_early_init();
+                               }
 #endif
                        } else if (memcmp(hv_vendor, "KVMKVMKVM\0\0\0", 12) == 0)
                                vm_guest = VM_GUEST_KVM;


It correctly "Identified Guest XEN in HVM mode" instead of Hyper-C, but 
immediately panicked in xen_hvm_init().

Is there a less naive patch I could try? :-)

--
Stephen