diff --git a/libc/test/integration/src/pthread/CMakeLists.txt b/libc/test/integration/src/pthread/CMakeLists.txt
--- a/libc/test/integration/src/pthread/CMakeLists.txt
+++ b/libc/test/integration/src/pthread/CMakeLists.txt
@@ -129,6 +129,13 @@
     libc.include.stdio
     libc.src.pthread.pthread_create
     libc.src.pthread.pthread_join
+    libc.src.pthread.pthread_attr_setdetachstate
+    libc.src.pthread.pthread_attr_init
+    libc.src.pthread.pthread_attr_destroy
+    libc.src.stdlib.srand
+    libc.src.stdlib.rand
+    libc.src.__support.CPP.atomic
+    libc.src.__support.CPP.new
 )
 
 add_integration_test(
diff --git a/libc/test/integration/src/pthread/pthread_join_test.cpp b/libc/test/integration/src/pthread/pthread_join_test.cpp
--- a/libc/test/integration/src/pthread/pthread_join_test.cpp
+++ b/libc/test/integration/src/pthread/pthread_join_test.cpp
@@ -1,4 +1,5 @@
-//===-- Tests for pthread_join-- ------------------------------------------===//
+//===-- Tests for pthread_join
+//--------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -6,25 +7,125 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "src/pthread/pthread_attr_destroy.h"
+#include "src/pthread/pthread_attr_init.h"
+#include "src/pthread/pthread_attr_setdetachstate.h"
 #include "src/pthread/pthread_create.h"
 #include "src/pthread/pthread_join.h"
 
+#include "src/stdlib/rand.h"
+#include "src/stdlib/srand.h"
+
+#include "src/__support/CPP/atomic.h"
+#include "src/__support/CPP/new.h"
+
+#include "src/__support/threads/thread.h"
 #include "src/errno/libc_errno.h"
 
 #include "test/IntegrationTest/test.h"
+
+#include <linux/param.h> // For EXEC_PAGESIZE.
 #include <pthread.h>
 
-static void *simpleFunc(void *) { return nullptr; }
-static void nullJoinTest() {
-  pthread_t Tid;
-  ASSERT_EQ(__llvm_libc::pthread_create(&Tid, nullptr, simpleFunc, nullptr), 0);
+struct TestThreadArgs {
+  pthread_t tid;
+  int detached;
+  volatile int done;
+  size_t spin_cnt;
+  void *ret;
+};
+
+static __llvm_libc::AllocChecker global_ac;
+
+static void *func(void *arg) {
+  TestThreadArgs *th_arg = reinterpret_cast<TestThreadArgs *>(arg);
+  volatile size_t spin_cnt = th_arg->spin_cnt;
+
+  while (spin_cnt) {
+    --spin_cnt;
+  }
+
+  th_arg->done = 1;
+  return th_arg->ret;
+}
+
+static void *simple_func(void *) { return nullptr; }
+
+static TestThreadArgs *initialize_test_thread_args(size_t num_threads,
+                                                   unsigned detached_percent) {
+  TestThreadArgs *th_args = new (global_ac) TestThreadArgs[num_threads];
+  for (size_t i = 0; i < num_threads; ++i) {
+    unsigned rand_val = __llvm_libc::rand();
+    uintptr_t rand_ret = __llvm_libc::rand();
+    rand_ret *= __llvm_libc::rand();
+    th_args[i].detached = detached_percent > (rand_val % 100);
+    th_args[i].done = 0;
+    th_args[i].spin_cnt = rand_val % (1000 * 1000);
+    th_args[i].ret = reinterpret_cast<void *>(rand_ret);
+  }
+
+  pthread_attr_t attr;
+  ASSERT_EQ(__llvm_libc::pthread_attr_init(&attr), 0);
+  for (size_t i = 0; i < num_threads; ++i) {
+    ASSERT_EQ(__llvm_libc::pthread_attr_setdetachstate(
+                  &attr, th_args[i].detached ? PTHREAD_CREATE_DETACHED
+                                             : PTHREAD_CREATE_JOINABLE),
+              0);
+
+    ASSERT_EQ(
+        __llvm_libc::pthread_create(&th_args[i].tid, &attr, func,
+                                    reinterpret_cast<void *>(th_args + i)),
+        0);
+  }
+  return th_args;
+}
+
+static void run_join_tests(size_t num_threads, bool join_as_completed) {
+  TestThreadArgs *th_args = initialize_test_thread_args(num_threads, 0);
+  for (size_t join_cnt = 0; join_cnt < num_threads;) {
+    for (size_t i = 0; i < num_threads; ++i) {
+      if (th_args[i].detached) {
+        if (th_args[i].done == 1) {
+          th_args[i].done = 2;
+          ++join_cnt;
+        }
+      } else {
+        if (!join_as_completed || th_args[i].done == 1) {
+          void *th_ret;
+          ASSERT_EQ(__llvm_libc::pthread_join(th_args[i].tid, &th_ret), 0);
+          ASSERT_EQ(libc_errno, 0);
+
+          ASSERT_EQ(th_args[i].done, 1);
+          ASSERT_EQ(th_ret, th_args[i].ret);
+
+          th_args[i].done = 2;
+          ++join_cnt;
+        }
+      }
+    }
+  }
+  delete th_args;
+}
+
+static void null_join_test() {
+  pthread_t tid;
+  ASSERT_EQ(__llvm_libc::pthread_create(&tid, nullptr, simple_func, nullptr),
+            0);
   ASSERT_EQ(libc_errno, 0);
-  ASSERT_EQ(__llvm_libc::pthread_join(Tid, nullptr), 0);
+  ASSERT_EQ(__llvm_libc::pthread_join(tid, nullptr), 0);
   ASSERT_EQ(libc_errno, 0);
 }
 
 TEST_MAIN() {
   libc_errno = 0;
-  nullJoinTest();
+  __llvm_libc::srand(123);
+
+  null_join_test();
+
+  for (size_t num_threads = 1; num_threads < 8; ++num_threads) {
+    run_join_tests(num_threads, /*join_as_completed*/ true);
+    run_join_tests(num_threads, /*join_as_completed*/ false);
+  }
+
   return 0;
 }