On Fri, 29 Jun 2018, Ax0n wrote:
> Updated and recompiled. Here's what it's doing now. No idea what the Lock
> Order Reversal noise is between the two VM starts, but I've seen a lot of
> it on this laptop lately, even without using vmm. Probably unrelated.
Yep, that's from the WITNESS work and not part of the xcr0 problem.
> vmm_fpurestore: guest attempted to set invalid bits in xcr0 (guest
> %xcr0=0x1, host mask=0xffffffffffffffff)
Oh, duh: this box doesn't have XSAVE at all but we init guests as if it
does. Try this diff on the host.
Philip Guenther
Index: amd64/vmm.c
===================================================================
RCS file: /data/src/openbsd/src/sys/arch/amd64/amd64/vmm.c,v
retrieving revision 1.202
diff -u -p -r1.202 vmm.c
--- amd64/vmm.c 22 Jun 2018 05:21:45 -0000 1.202
+++ amd64/vmm.c 29 Jun 2018 17:36:34 -0000
@@ -1971,7 +1971,7 @@ vcpu_reset_regs_svm(struct vcpu *vcpu, s
ret = vcpu_writeregs_svm(vcpu, VM_RWREGS_ALL, vrs);
/* xcr0 power on default sets bit 0 (x87 state) */
- vcpu->vc_gueststate.vg_xcr0 = XCR0_X87;
+ vcpu->vc_gueststate.vg_xcr0 = XCR0_X87 & xsave_mask;
exit:
return ret;
@@ -2764,7 +2764,7 @@ vcpu_reset_regs_vmx(struct vcpu *vcpu, s
/* XXX CR4 shadow */
/* xcr0 power on default sets bit 0 (x87 state) */
- vcpu->vc_gueststate.vg_xcr0 = XCR0_X87;
+ vcpu->vc_gueststate.vg_xcr0 = XCR0_X87 & xsave_mask;
/* Flush the VMCS */
if (vmclear(&vcpu->vc_control_pa)) {
