Index: libexec/rtld-elf/Makefile
===================================================================
--- libexec/rtld-elf/Makefile
+++ libexec/rtld-elf/Makefile
@@ -47,7 +47,7 @@
 .if ${MACHINE_CPUARCH} == "amd64" || ${MACHINE_CPUARCH} == "i386"
 CFLAGS+=	-fvisibility=hidden
 .endif
-LDFLAGS+=	-shared -Wl,-Bsymbolic
+LDFLAGS+=	-shared -Wl,-Bsymbolic -Wl,-z,defs
 LIBADD=		c_pic
 .if ${MK_TOOLCHAIN} == "no"
 LDFLAGS+=	-L${LIBCDIR}
Index: libexec/rtld-elf/amd64/reloc.c
===================================================================
--- libexec/rtld-elf/amd64/reloc.c
+++ libexec/rtld-elf/amd64/reloc.c
@@ -34,6 +34,7 @@
 #include <sys/param.h>
 #include <sys/mman.h>
 #include <machine/sysarch.h>
+#include <machine/cpufunc.h>
 
 #include <dlfcn.h>
 #include <err.h>
@@ -406,7 +407,7 @@
 	  ptr = (Elf_Addr *)(obj->relocbase + rela->r_addend);
 	  where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 	  lock_release(rtld_bind_lock, lockstate);
-	  target = ((Elf_Addr (*)(void))ptr)();
+	  target = call_ifunc_resolver(ptr);
 	  wlock_acquire(rtld_bind_lock, lockstate);
 	  *where = target;
 	  break;
@@ -450,6 +451,25 @@
     return (0);
 }
 
+uint32_t cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2;
+
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT])
+{
+	u_int p[4], cpu_high;
+
+	do_cpuid(1, p);
+	cpu_feature = p[3];
+	cpu_feature2 = p[2];
+	do_cpuid(0, p);
+	cpu_high = p[0];
+	if (cpu_high >= 7) {
+		cpuid_count(7, 0, p);
+		cpu_stdext_feature = p[1];
+		cpu_stdext_feature2 = p[2];
+	}
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {
Index: libexec/rtld-elf/amd64/rtld_machdep.h
===================================================================
--- libexec/rtld-elf/amd64/rtld_machdep.h
+++ libexec/rtld-elf/amd64/rtld_machdep.h
@@ -61,6 +61,14 @@
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+extern uint32_t cpu_feature;
+extern uint32_t cpu_feature2;
+extern uint32_t cpu_stdext_feature;
+extern uint32_t cpu_stdext_feature2;
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(uint32_t, uint32_t, uint32_t, uint32_t))ptr)( \
+	    cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2))
+
 #define round(size, align) \
 	(((size) + (align) - 1) & ~((align) - 1))
 #define calculate_first_tls_offset(size, align) \
Index: libexec/rtld-elf/i386/reloc.c
===================================================================
--- libexec/rtld-elf/i386/reloc.c
+++ libexec/rtld-elf/i386/reloc.c
@@ -33,6 +33,7 @@
 
 #include <sys/param.h>
 #include <sys/mman.h>
+#include <machine/cpufunc.h>
 #include <machine/segments.h>
 #include <machine/sysarch.h>
 
@@ -359,7 +360,7 @@
 	case R_386_IRELATIVE:
 	  where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
 	  lock_release(rtld_bind_lock, lockstate);
-	  target = ((Elf_Addr (*)(void))(obj->relocbase + *where))();
+	  target = call_ifunc_resolver(obj->relocbase + *where);
 	  wlock_acquire(rtld_bind_lock, lockstate);
 	  *where = target;
 	  break;
@@ -404,6 +405,45 @@
     return (0);
 }
 
+uint32_t cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2;
+
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT])
+{
+	u_int p[4], cpu_high;
+	int cpuid_supported;
+
+	__asm __volatile(
+	    "	pushfl\n"
+	    "	popl	%%eax\n"
+	    "	movl    %%eax,%%ecx\n"
+	    "	xorl    $0x200000,%%eax\n"
+	    "	pushl	%%eax\n"
+	    "	popfl\n"
+	    "	pushfl\n"
+	    "	popl    %%eax\n"
+	    "	xorl    %%eax,%%ecx\n"
+	    "	je	1f\n"
+	    "	movl	$1,%0\n"
+	    "	jmp	2f\n"
+	    "1:	movl	$0,%0\n"
+	    "2:\n"
+	    : "=r" (cpuid_supported) : : "eax", "ecx");
+	if (!cpuid_supported)
+		return;
+
+	do_cpuid(1, p);
+	cpu_feature = p[3];
+	cpu_feature2 = p[2];
+	do_cpuid(0, p);
+	cpu_high = p[0];
+	if (cpu_high >= 7) {
+		cpuid_count(7, 0, p);
+		cpu_stdext_feature = p[1];
+		cpu_stdext_feature2 = p[2];
+	}
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {
Index: libexec/rtld-elf/i386/rtld_machdep.h
===================================================================
--- libexec/rtld-elf/i386/rtld_machdep.h
+++ libexec/rtld-elf/i386/rtld_machdep.h
@@ -61,6 +61,14 @@
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+extern uint32_t cpu_feature;
+extern uint32_t cpu_feature2;
+extern uint32_t cpu_stdext_feature;
+extern uint32_t cpu_stdext_feature2;
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(uint32_t, uint32_t, uint32_t, uint32_t))ptr)( \
+	    cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2))
+
 #define round(size, align) \
 	(((size) + (align) - 1) & ~((align) - 1))
 #define calculate_first_tls_offset(size, align) \
Index: libexec/rtld-elf/rtld.h
===================================================================
--- libexec/rtld-elf/rtld.h
+++ libexec/rtld-elf/rtld.h
@@ -367,6 +367,7 @@
 unsigned long elf_hash(const char *);
 const Elf_Sym *find_symdef(unsigned long, const Obj_Entry *,
   const Obj_Entry **, int, SymCache *, struct Struct_RtldLockState *);
+void ifunc_init(Elf_Auxinfo[static AT_COUNT]);
 void init_pltgot(Obj_Entry *);
 void lockdflt_init(void);
 void digest_notes(Obj_Entry *, Elf_Addr, Elf_Addr);
Index: libexec/rtld-elf/rtld.c
===================================================================
--- libexec/rtld-elf/rtld.c
+++ libexec/rtld-elf/rtld.c
@@ -642,6 +642,7 @@
     r_debug_state(NULL, &obj_main->linkmap); /* say hello to gdb! */
 
     map_stacks_exec(NULL);
+    ifunc_init(aux);
 
     dbg("resolving ifuncs");
     if (resolve_objects_ifunc(obj_main,
@@ -690,7 +691,7 @@
 	Elf_Addr target;
 
 	ptr = (void *)make_function_pointer(def, obj);
-	target = ((Elf_Addr (*)(void))ptr)();
+	target = call_ifunc_resolver(ptr);
 	return ((void *)target);
 }