26.06.2021 03:15, Jim Mattson пишет:
If the squashed exception was a trap, it's now lost.
I am pretty sure this will do it:
---
--- x86.c.old 2021-03-20 12:51:14.000000000 +0300
+++ x86.c 2021-06-27 15:02:45.126161812 +0300
@@ -9093,7 +9093,11 @@
cancel_injection:
if (req_immediate_exit)
kvm_make_request(KVM_REQ_EVENT, vcpu);
- kvm_x86_ops.cancel_injection(vcpu);
+ if (vcpu->arch.exception.injected) {
+ kvm_x86_ops.cancel_injection(vcpu);
+ vcpu->arch.exception.injected = false;
+ vcpu->arch.exception.pending = true;
+ }
if (unlikely(vcpu->arch.apic_attention))
kvm_lapic_sync_from_vapic(vcpu);
out:
@@ -9464,6 +9468,7 @@
kvm_rip_write(vcpu, regs->rip);
kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
+ WARN_ON_ONCE(vcpu->arch.exception.injected);
vcpu->arch.exception.pending = false;
kvm_make_request(KVM_REQ_EVENT, vcpu);
---
In cancel_injection, the injected/pending
members were getting out of sync with
vmcs.
We need to move it back to pending,
and if user-space does SET_REGS, then
it is cleared (not sure why SET_SREGS
doesn't clear it also).
But if the .injected member is stuck,
then its not cleared by SET_REGS, and
I added WARN_ON_ONCE() for that case.
Does this make sense?
