KVM: x86: Add a capability for GUEST_MAXPHYADDR < HOST_MAXPHYADDR sup…

…port

This patch adds a new capability KVM_CAP_SMALLER_MAXPHYADDR which
allows userspace to query if the underlying architecture would
support GUEST_MAXPHYADDR < HOST_MAXPHYADDR and hence act accordingly
(e.g. qemu can decide if it should warn for -cpu ..,phys-bits=X)

The complications in this patch are due to unexpected (but documented)
behaviour we see with NPF vmexit handling in AMD processor.  If
SVM is modified to add guest physical address checks in the NPF
and guest #PF paths, we see the followning error multiple times in
the 'access' test in kvm-unit-tests:

            test pte.p pte.36 pde.p: FAIL: pte 2000021 expected 2000001
            Dump mapping: address: 0x123400000000
            ------L4: 24c3027
            ------L3: 24c4027
            ------L2: 24c5021
            ------L1: 1002000021

This is because the PTE's accessed bit is set by the CPU hardware before
the NPF vmexit. This is handled completely by hardware and cannot be fixed
in software.

Therefore, availability of the new capability depends on a boolean variable
allow_smaller_maxphyaddr which is set individually by VMX and SVM init
routines. On VMX it's always set to true, on SVM it's only set to true
when NPT is not enabled.

CC: Tom Lendacky <thomas.lendacky@amd.com>
CC: Babu Moger <babu.moger@amd.com>
Signed-off-by: Mohammed Gamal <mgamal@redhat.com>
Message-Id: <20200710154811.418214-10-mgamal@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

arch/x86/include/asm/kvm_host.h

@@ -1263,7 +1263,7 @@ struct kvm_arch_async_pf {
 };
 
 extern u64 __read_mostly host_efer;
-
+extern bool __read_mostly allow_smaller_maxphyaddr;
 extern struct kvm_x86_ops kvm_x86_ops;
 
 #define __KVM_HAVE_ARCH_VM_ALLOC

arch/x86/kvm/svm/svm.c

@@ -924,6 +924,21 @@ static __init int svm_hardware_setup(void)
 
 	svm_set_cpu_caps();
 
+	/*
+	 * It seems that on AMD processors PTE's accessed bit is
+	 * being set by the CPU hardware before the NPF vmexit.
+	 * This is not expected behaviour and our tests fail because
+	 * of it.
+	 * A workaround here is to disable support for
+	 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
+	 * In this case userspace can know if there is support using
+	 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
+	 * it
+	 * If future AMD CPU models change the behaviour described above,
+	 * this variable can be changed accordingly
+	 */
+	allow_smaller_maxphyaddr = !npt_enabled;
+
 	return 0;
 
 err:

arch/x86/kvm/vmx/vmx.c

@@ -8309,6 +8309,13 @@ static int __init vmx_init(void)
 #endif
 	vmx_check_vmcs12_offsets();
 
+	/*
+	 * Intel processors don't have problems with
+	 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable
+	 * it for VMX by default
+	 */
+	allow_smaller_maxphyaddr = true;
+
 	return 0;
 }
 module_init(vmx_init);

arch/x86/kvm/x86.c

@@ -187,6 +187,9 @@ static struct kvm_shared_msrs __percpu *shared_msrs;
 u64 __read_mostly host_efer;
 EXPORT_SYMBOL_GPL(host_efer);
 
+bool __read_mostly allow_smaller_maxphyaddr;
+EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
+
 static u64 __read_mostly host_xss;
 u64 __read_mostly supported_xss;
 EXPORT_SYMBOL_GPL(supported_xss);
@@ -3574,6 +3577,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
 		break;
+	case KVM_CAP_SMALLER_MAXPHYADDR:
+		r = (int) allow_smaller_maxphyaddr;
+		break;
 	default:
 		break;
 	}

include/uapi/linux/kvm.h

@@ -1033,6 +1033,8 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_HALT_POLL 182
 #define KVM_CAP_ASYNC_PF_INT 183
 #define KVM_CAP_LAST_CPU 184
+#define KVM_CAP_SMALLER_MAXPHYADDR 185
+
 
 #ifdef KVM_CAP_IRQ_ROUTING