OSDN Git Service

KVM: PPC: Book3S HV: Sanitise hv_regs on nested guest entry
authorSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Mon, 8 Oct 2018 05:31:14 +0000 (16:31 +1100)
committerMichael Ellerman <mpe@ellerman.id.au>
Tue, 9 Oct 2018 05:04:27 +0000 (16:04 +1100)
restore_hv_regs() is used to copy the hv_regs L1 wants to set to run the
nested (L2) guest into the vcpu structure. We need to sanitise these
values to ensure we don't let the L1 guest hypervisor do things we don't
want it to.

We don't let data address watchpoints or completed instruction address
breakpoints be set to match in hypervisor state.

We also don't let L1 enable features in the hypervisor facility status
and control register (HFSCR) for L2 which we have disabled for L1. That
is L2 will get the subset of features which the L0 hypervisor has
enabled for L1 and the features L1 wants to enable for L2. This could
mean we give L1 a hypervisor facility unavailable interrupt for a
facility it thinks it has enabled, however it shouldn't have enabled a
facility it itself doesn't have for the L2 guest.

We sanitise the registers when copying in the L2 hv_regs. We don't need
to sanitise when copying back the L1 hv_regs since these shouldn't be
able to contain invalid values as they're just what was copied out.

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/include/asm/reg.h
arch/powerpc/kvm/book3s_hv_nested.c

index 6fda746..c906989 100644 (file)
 #define   HFSCR_DSCR   __MASK(FSCR_DSCR_LG)
 #define   HFSCR_VECVSX __MASK(FSCR_VECVSX_LG)
 #define   HFSCR_FP     __MASK(FSCR_FP_LG)
+#define   HFSCR_INTR_CAUSE (ASM_CONST(0xFF) << 56)     /* interrupt cause */
 #define SPRN_TAR       0x32f   /* Target Address Register */
 #define SPRN_LPCR      0x13E   /* LPAR Control Register */
 #define   LPCR_VPM0            ASM_CONST(0x8000000000000000)
index a876dc3..e230596 100644 (file)
@@ -86,6 +86,22 @@ static void save_hv_return_state(struct kvm_vcpu *vcpu, int trap,
        }
 }
 
+static void sanitise_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
+{
+       /*
+        * Don't let L1 enable features for L2 which we've disabled for L1,
+        * but preserve the interrupt cause field.
+        */
+       hr->hfscr &= (HFSCR_INTR_CAUSE | vcpu->arch.hfscr);
+
+       /* Don't let data address watchpoint match in hypervisor state */
+       hr->dawrx0 &= ~DAWRX_HYP;
+
+       /* Don't let completed instruction address breakpt match in HV state */
+       if ((hr->ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
+               hr->ciabr &= ~CIABR_PRIV;
+}
+
 static void restore_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
 {
        struct kvmppc_vcore *vc = vcpu->arch.vcore;
@@ -198,6 +214,7 @@ long kvmhv_enter_nested_guest(struct kvm_vcpu *vcpu)
        mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD |
                LPCR_LPES | LPCR_MER;
        lpcr = (vc->lpcr & ~mask) | (l2_hv.lpcr & mask);
+       sanitise_hv_regs(vcpu, &l2_hv);
        restore_hv_regs(vcpu, &l2_hv);
 
        vcpu->arch.ret = RESUME_GUEST;