diff --git a/tools/tests/x86_emulator/test_x86_emulator.c b/tools/tests/x86_emulator/test_x86_emulator.c
index 906fc6f398..5864a41b7a 100644
--- a/tools/tests/x86_emulator/test_x86_emulator.c
+++ b/tools/tests/x86_emulator/test_x86_emulator.c
@@ -127,6 +127,22 @@ static inline uint64_t xgetbv(uint32_t xcr)
     (ebx & (1U << 5)) != 0; \
 })
 
+static int read_cr(
+    unsigned int reg,
+    unsigned long *val,
+    struct x86_emulate_ctxt *ctxt)
+{
+    /* Fake just enough state for the emulator's _get_fpu() to be happy. */
+    switch ( reg )
+    {
+    case 0:
+        *val = 0x00000001; /* PE */
+        return X86EMUL_OKAY;
+    }
+
+    return X86EMUL_UNHANDLEABLE;
+}
+
 int get_fpu(
     void (*exception_callback)(void *, struct cpu_user_regs *),
     void *exception_callback_arg,
@@ -158,6 +174,7 @@ static struct x86_emulate_ops emulops = {
     .write      = write,
     .cmpxchg    = cmpxchg,
     .cpuid      = cpuid,
+    .read_cr    = read_cr,
     .get_fpu    = get_fpu,
 };
 
diff --git a/xen/arch/x86/hvm/emulate.c b/xen/arch/x86/hvm/emulate.c
index 868aa1df51..fab8ae324f 100644
--- a/xen/arch/x86/hvm/emulate.c
+++ b/xen/arch/x86/hvm/emulate.c
@@ -1070,6 +1070,7 @@ static int hvmemul_get_fpu(
     switch ( type )
     {
     case X86EMUL_FPU_fpu:
+    case X86EMUL_FPU_wait:
         break;
     case X86EMUL_FPU_mmx:
         if ( !cpu_has_mmx )
@@ -1077,7 +1078,6 @@ static int hvmemul_get_fpu(
         break;
     case X86EMUL_FPU_xmm:
         if ( !cpu_has_xmm ||
-             (curr->arch.hvm_vcpu.guest_cr[0] & X86_CR0_EM) ||
              !(curr->arch.hvm_vcpu.guest_cr[4] & X86_CR4_OSFXSR) )
             return X86EMUL_UNHANDLEABLE;
         break;
diff --git a/xen/arch/x86/x86_emulate/x86_emulate.c b/xen/arch/x86/x86_emulate/x86_emulate.c
index 4c90101f9c..81e268a848 100644
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -373,6 +373,9 @@ typedef union {
 
 /* Control register flags. */
 #define CR0_PE    (1<<0)
+#define CR0_MP    (1<<1)
+#define CR0_EM    (1<<2)
+#define CR0_TS    (1<<3)
 #define CR4_TSD   (1<<2)
 
 /* EFLAGS bit definitions. */
@@ -400,6 +403,7 @@ typedef union {
 #define EXC_OF  4
 #define EXC_BR  5
 #define EXC_UD  6
+#define EXC_NM  7
 #define EXC_TS 10
 #define EXC_NP 11
 #define EXC_SS 12
@@ -668,10 +672,45 @@ static void fpu_handle_exception(void *_fic, struct cpu_user_regs *regs)
     regs->eip += fic->insn_bytes;
 }
 
+static int _get_fpu(
+    enum x86_emulate_fpu_type type,
+    struct fpu_insn_ctxt *fic,
+    struct x86_emulate_ctxt *ctxt,
+    const struct x86_emulate_ops *ops)
+{
+    int rc;
+
+    fic->exn_raised = 0;
+
+    fail_if(!ops->get_fpu);
+    rc = ops->get_fpu(fpu_handle_exception, fic, type, ctxt);
+
+    if ( rc == X86EMUL_OKAY )
+    {
+        unsigned long cr0;
+
+        fail_if(!ops->read_cr);
+        rc = ops->read_cr(0, &cr0, ctxt);
+        if ( rc != X86EMUL_OKAY )
+            return rc;
+        if ( cr0 & CR0_EM )
+        {
+            generate_exception_if(type == X86EMUL_FPU_fpu, EXC_NM, -1);
+            generate_exception_if(type == X86EMUL_FPU_mmx, EXC_UD, -1);
+            generate_exception_if(type == X86EMUL_FPU_xmm, EXC_UD, -1);
+        }
+        generate_exception_if((cr0 & CR0_TS) &&
+                              (type != X86EMUL_FPU_wait || (cr0 & CR0_MP)),
+                              EXC_NM, -1);
+    }
+
+ done:
+    return rc;
+}
+
 #define get_fpu(_type, _fic)                                    \
-do{ (_fic)->exn_raised = 0;                                     \
-    fail_if(ops->get_fpu == NULL);                              \
-    rc = ops->get_fpu(fpu_handle_exception, _fic, _type, ctxt); \
+do {                                                            \
+    rc = _get_fpu(_type, _fic, ctxt, ops);                      \
     if ( rc ) goto done;                                        \
 } while (0)
 #define put_fpu(_fic)                                           \
@@ -2349,8 +2388,14 @@ x86_emulate(
     }
 
     case 0x9b:  /* wait/fwait */
-        emulate_fpu_insn("fwait");
+    {
+        struct fpu_insn_ctxt fic = { .insn_bytes = 1 };
+
+        get_fpu(X86EMUL_FPU_wait, &fic);
+        asm volatile ( "fwait" ::: "memory" );
+        put_fpu(&fic);
         break;
+    }
 
     case 0x9c: /* pushf */
         src.val = _regs.eflags;
diff --git a/xen/arch/x86/x86_emulate/x86_emulate.h b/xen/arch/x86/x86_emulate/x86_emulate.h
index 85bc4bcf76..73074236b6 100644
--- a/xen/arch/x86/x86_emulate/x86_emulate.h
+++ b/xen/arch/x86/x86_emulate/x86_emulate.h
@@ -99,6 +99,7 @@ struct segment_register {
 /* FPU sub-types which may be requested via ->get_fpu(). */
 enum x86_emulate_fpu_type {
     X86EMUL_FPU_fpu, /* Standard FPU coprocessor instruction set */
+    X86EMUL_FPU_wait, /* WAIT/FWAIT instruction */
     X86EMUL_FPU_mmx, /* MMX instruction set (%mm0-%mm7) */
     X86EMUL_FPU_xmm, /* SSE instruction set (%xmm0-%xmm7/15) */
     X86EMUL_FPU_ymm  /* AVX/XOP instruction set (%ymm0-%ymm7/15) */