lguest: fix on Intel when KVM loaded (unhandled trap 13)
Rusty Russell [Tue, 26 May 2009 11:24:41 +0000 (20:24 +0930)]
When KVM is loaded, and hence VT set up, the vmcall instruction in an
lguest guest causes a #GP, not #UD.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/lguest/x86/core.c

index 1a83910..eaf722f 100644 (file)
@@ -358,6 +358,16 @@ void lguest_arch_handle_trap(struct lg_cpu *cpu)
                        if (emulate_insn(cpu))
                                return;
                }
+               /* If KVM is active, the vmcall instruction triggers a
+                * General Protection Fault.  Normally it triggers an
+                * invalid opcode fault (6): */
+       case 6:
+               /* We need to check if ring == GUEST_PL and
+                * faulting instruction == vmcall. */
+               if (is_hypercall(cpu)) {
+                       rewrite_hypercall(cpu);
+                       return;
+               }
                break;
        case 14: /* We've intercepted a Page Fault. */
                /* The Guest accessed a virtual address that wasn't mapped.
@@ -403,15 +413,6 @@ void lguest_arch_handle_trap(struct lg_cpu *cpu)
                 * up the pointer now to indicate a hypercall is pending. */
                cpu->hcall = (struct hcall_args *)cpu->regs;
                return;
-       case 6:
-               /* kvm hypercalls trigger an invalid opcode fault (6).
-                * We need to check if ring == GUEST_PL and
-                * faulting instruction == vmcall. */
-               if (is_hypercall(cpu)) {
-                       rewrite_hypercall(cpu);
-                       return;
-               }
-               break;
        }
 
        /* We didn't handle the trap, so it needs to go to the Guest. */