diff --git a/src/_vmprof.c b/src/_vmprof.c
index f96feb20..e51dc9e3 100644
--- a/src/_vmprof.c
+++ b/src/_vmprof.c
@@ -393,7 +393,7 @@ insert_real_time_thread(PyObject *module, PyObject * noargs) {
     }
 
     vmprof_aquire_lock();
-    thread_count = insert_thread(pthread_self(), -1);
+    thread_count = insert_thread(GET_THREAD_ID(), -1);
     vmprof_release_lock();
 
     return PyLong_FromSsize_t(thread_count);
@@ -414,7 +414,7 @@ remove_real_time_thread(PyObject *module, PyObject * noargs) {
     }
 
     vmprof_aquire_lock();
-    thread_count = remove_thread(pthread_self(), -1);
+    thread_count = remove_thread(GET_THREAD_ID(), -1);
     vmprof_release_lock();
 
     return PyLong_FromSsize_t(thread_count);
diff --git a/src/vmprof_common.c b/src/vmprof_common.c
index d81b9119..eb64d51a 100644
--- a/src/vmprof_common.c
+++ b/src/vmprof_common.c
@@ -28,7 +28,7 @@ static long profile_interval_usec = 0;
 #ifdef VMPROF_UNIX
 static int signal_type = SIGPROF;
 static int itimer_type = ITIMER_PROF;
-static pthread_t *threads = NULL;
+static THREAD_ID *threads = NULL;
 static size_t threads_size = 0;
 static size_t thread_count = 0;
 static size_t threads_size_step = 8;
@@ -221,19 +221,19 @@ intptr_t vmprof_get_traceback(void *stack, void *ucontext,
 
 #ifdef VMPROF_UNIX
 
-ssize_t search_thread(pthread_t tid, ssize_t i)
+ssize_t search_thread(THREAD_ID tid, ssize_t i)
 {
     if (i < 0)
         i = 0;
     while ((size_t)i < thread_count) {
-        if (pthread_equal(threads[i], tid))
+        if (THREAD_EQUALS(threads[i], tid))
             return i;
         i++;
     }
     return -1;
 }
 
-ssize_t insert_thread(pthread_t tid, ssize_t i)
+ssize_t insert_thread(THREAD_ID tid, ssize_t i)
 {
     assert(signal_type == SIGALRM);
     i = search_thread(tid, i);
@@ -241,15 +241,15 @@ ssize_t insert_thread(pthread_t tid, ssize_t i)
         return -1;
     if (thread_count == threads_size) {
         threads_size += threads_size_step;
-        threads = realloc(threads, sizeof(pid_t) * threads_size);
+        threads = realloc(threads, sizeof(THREAD_ID) * threads_size);
         assert(threads != NULL);
-        memset(threads + thread_count, 0, sizeof(pid_t) * threads_size_step);
+        memset(threads + thread_count, 0, sizeof(THREAD_ID) * threads_size_step);
     }
     threads[thread_count++] = tid;
     return thread_count;
 }
 
-ssize_t remove_thread(pthread_t tid, ssize_t i)
+ssize_t remove_thread(THREAD_ID tid, ssize_t i)
 {
     assert(signal_type == SIGALRM);
     if (thread_count == 0)
@@ -280,13 +280,14 @@ int broadcast_signal_for_threads(void)
 {
     int done = 1;
     size_t i = 0;
-    pthread_t self = pthread_self();
-    pthread_t tid;
+    THREAD_ID self = GET_THREAD_ID();
+    THREAD_ID tid;
+    pid_t pid = getpid();
     while (i < thread_count) {
         tid = threads[i];
-        if (pthread_equal(tid, self)) {
+        if (THREAD_EQUALS(tid, self)) {
             done = 0;
-        } else if (pthread_kill(tid, SIGALRM)) {
+        } else if (THREAD_SIGNAL(pid, tid, SIGALRM)) {
             remove_thread(tid, i);
         }
         i++;
diff --git a/src/vmprof_common.h b/src/vmprof_common.h
index 9f55f796..5d496539 100644
--- a/src/vmprof_common.h
+++ b/src/vmprof_common.h
@@ -21,15 +21,31 @@
 
 #ifdef VMPROF_LINUX
 #include <syscall.h>
+#define THREAD_ID pid_t
+#define GET_THREAD_ID() ((pid_t) syscall(SYS_gettid))
+#define THREAD_EQUALS(thread_one, thread_two) (thread_one == thread_two)
+#define THREAD_SIGNAL(pid, tid, signal) syscall(SYS_tgkill, pid, tid, signal)
+#else
+#ifdef VMPROF_UNIX
+#define THREAD_ID pthread_t
+#define GET_THREAD_ID() ((pthread_t) pthread_self())
+#define THREAD_EQUALS(thread_one, thread_two) pthread_equal(thread_one, thread_two)
+#define THREAD_SIGNAL(pid, tid, signal) pthread_kill(tid, signal)
+#endif
 #endif
 
 #define MAX_FUNC_NAME 1024
 
+#ifdef VMPROF_LINUX
+#else
+#define THREAD_ID_TYPE
+#endif
+
 #ifdef VMPROF_UNIX
 
-ssize_t search_thread(pthread_t tid, ssize_t i);
-ssize_t insert_thread(pthread_t tid, ssize_t i);
-ssize_t remove_thread(pthread_t tid, ssize_t i);
+ssize_t search_thread(THREAD_ID tid, ssize_t i);
+ssize_t insert_thread(THREAD_ID tid, ssize_t i);
+ssize_t remove_thread(THREAD_ID tid, ssize_t i);
 ssize_t remove_threads(void);
 
 #endif
diff --git a/src/vmprof_unix.c b/src/vmprof_unix.c
index 1a85a10c..766e630d 100644
--- a/src/vmprof_unix.c
+++ b/src/vmprof_unix.c
@@ -359,7 +359,7 @@ int vmprof_enable(int memory, int native, int real_time)
     if (memory && setup_rss() == -1)
         goto error;
 #if VMPROF_UNIX
-    if (real_time && insert_thread(pthread_self(), -1) == -1)
+    if (real_time && insert_thread(GET_THREAD_ID(), -1) == -1)
         goto error;
 #endif
     if (install_pthread_atfork_hooks() == -1)
diff --git a/vmprof/stats.py b/vmprof/stats.py
index ee81f18e..2eb16a67 100644
--- a/vmprof/stats.py
+++ b/vmprof/stats.py
@@ -76,6 +76,18 @@ def generate_top(self):
                     self.functions[addr] = self.functions.get(addr, 0) + 1
                     current_iter[addr] = None
 
+    def all_profile(self):
+        all_functions = {}
+        for profile in self.profiles:
+            for i, addr in enumerate(profile[0]):
+                if self.profile_lines and i % 2 == 1:
+                    # this entry in the profile is a negative number indicating a line
+                    assert addr <= 0
+                    continue
+                else:
+                    all_functions[addr] = all_functions.get(addr, 0) + 1
+        return [(self._get_name(k), v) for (k, v) in six.iteritems(all_functions)]
+
     def top_profile(self):
         return [(self._get_name(k), v) for (k, v) in six.iteritems(self.functions)]
 
diff --git a/vmprof/test/test_run.py b/vmprof/test/test_run.py
index 29455978..69ff93e2 100644
--- a/vmprof/test/test_run.py
+++ b/vmprof/test/test_run.py
@@ -79,15 +79,33 @@ def function_bar():
 
 
 def functime_foo(t=0.05, insert=False):
+    if PY3K:
+        from threading import get_ident
+    else:
+        from thread import get_ident
+    thread_id = get_ident()
     if (insert):
-        vmprof.insert_real_time_thread()
-    return time.sleep(t)
+        thread_count = vmprof.insert_real_time_thread()
+    else:
+        thread_count = -2
+    sys.stderr.writelines("foo:%i:%i\n" % (thread_id, thread_count))
+    time.sleep(t)
+    return
 
 
 def functime_bar(t=0.05, remove=False):
+    if PY3K:
+        from threading import get_ident
+    else:
+        from thread import get_ident
+    thread_id = get_ident()
     if (remove):
-        vmprof.remove_real_time_thread()
-    return time.sleep(t)
+        thread_count = vmprof.remove_real_time_thread()
+    else:
+        thread_count = -2
+    sys.stderr.writelines("bar:%i:%i\n" % (thread_id, thread_count))
+    time.sleep(t)
+    return
 
 
 foo_full_name = "py:function_foo:%d:%s" % (function_foo.__code__.co_firstlineno,
@@ -255,7 +273,6 @@ def test_vmprof_real_time():
     assert d[foo_time_name] > 0
 
 
-@py.test.mark.xfail()
 @py.test.mark.skipif("'__pypy__' in sys.builtin_module_names")
 @py.test.mark.skipif("sys.platform == 'win32'")
 @py.test.mark.parametrize("insert_foo,remove_bar", [
@@ -266,16 +283,20 @@ def test_vmprof_real_time():
 ])
 def test_vmprof_real_time_threaded(insert_foo, remove_bar):
     import threading
+    import pprint
     prof = vmprof.Profiler()
     wait = 0.5
     thread = threading.Thread(target=functime_foo, args=[wait, insert_foo])
-    with prof.measure(period=0.25, real_time=True):
+    with prof.measure(period=wait/5, real_time=True):
         thread.start()
         functime_bar(wait, remove_bar)
         thread.join()
     stats = prof.get_stats()
     tprof = stats.top_profile()
+    aprof = stats.all_profile()
     d = dict(tprof)
+    f = pprint.pformat(d) + '\n' + pprint.pformat(dict(aprof)) + '\n'
+    sys.stderr.writelines(f)
     assert insert_foo == (foo_time_name in d)
     assert remove_bar != (bar_time_name in d)