diff --git a/lib/drmtest.c b/lib/drmtest.c
index 70483eac..81c5fca9 100644
--- a/lib/drmtest.c
+++ b/lib/drmtest.c
@@ -652,6 +652,12 @@ static enum {
 	CONT = 0, SKIP, FAIL
 } skip_subtests_henceforth = CONT;
 
+/* fork support state */
+pid_t *test_children;
+int num_test_children;
+int test_children_sz;
+bool test_child;
+
 bool __igt_fixture(void)
 {
 	assert(!in_fixture);
@@ -852,6 +858,8 @@ void igt_skip(const char *f, ...)
 	va_list args;
 	skipped_one = true;
 
+	assert(!test_child);
+
 	if (!igt_only_list_subtests()) {
 		va_start(args, f);
 		vprintf(f, args);
@@ -893,6 +901,10 @@ void igt_fail(int exitcode)
 
 	failed_one = true;
 
+	/* Silent exit, parent will do the yelling. */
+	if (test_child)
+		exit(exitcode);
+
 	if (in_subtest)
 		exit_subtest("FAIL");
 	else {
@@ -946,6 +958,59 @@ void igt_exit(void)
 		exit(77);
 }
 
+bool __igt_fork(void)
+{
+	assert(!test_with_subtests || in_subtest);
+
+	if (num_test_children >= test_children_sz) {
+		if (!test_children_sz)
+			test_children_sz = 4;
+		else
+			test_children_sz *= 2;
+
+		test_children = realloc(test_children,
+					sizeof(pid_t)*test_children_sz);
+		igt_assert(test_children);
+	}
+
+	switch (test_children[num_test_children++] = fork()) {
+	case -1:
+		igt_assert(0);
+	case 0:
+		test_child = true;
+		return true;
+	default:
+		return false;
+	}
+
+}
+
+/**
+ * igt_waitchildren - wait for all children forked with igt_fork
+ *
+ * The magic here is that exit codes from children will be correctly propagated
+ */
+void igt_waitchildren(void)
+{
+	assert(!test_child);
+
+	for (int nc = 0; nc < num_test_children; nc++) {
+		int status = -1;
+		while (waitpid(test_children[nc], &status, 0) == -1 &&
+		       errno == -EINTR)
+			;
+
+		if (status != 0) {
+			if (WIFEXITED(status))
+				igt_fail(WEXITSTATUS(status));
+			else
+				abort();
+		}
+	}
+
+	num_test_children = 0;
+}
+
 static bool env_set(const char *env_var, bool default_value)
 {
 	char *val;
diff --git a/lib/drmtest.h b/lib/drmtest.h
index ece478b1..27573bfd 100644
--- a/lib/drmtest.h
+++ b/lib/drmtest.h
@@ -204,6 +204,20 @@ void __igt_fixture_end(void) __attribute__((noreturn));
 			 igt_tokencat(__tmpint,__LINE__) ++, \
 			 __igt_fixture_complete())
 
+bool __igt_fork(void);
+/**
+ * igt_fork - fork parallel test threads with fork()
+ * @child: name of the int variable with the child number
+ * @num_children: number of children to fork
+ *
+ * Joining all test threads should be done with igt_waitchildren to ensure that
+ * the exit codes of all children are properly reflected in the test status.
+ */
+#define igt_fork(child, num_children) \
+	for (int child = 0; child < (num_children); child++) \
+		for (; __igt_fork(); exit(0))
+void igt_waitchildren(void);
+
 /* check functions which auto-skip tests by calling igt_skip() */
 void gem_require_caching(int fd);
 static inline void gem_require_ring(int fd, int ring_id)
diff --git a/tests/gem_concurrent_blit.c b/tests/gem_concurrent_blit.c
index 69477748..b6d316d1 100644
--- a/tests/gem_concurrent_blit.c
+++ b/tests/gem_concurrent_blit.c
@@ -267,47 +267,37 @@ static void run_forked(struct access_mode *mode,
 		       do_test do_test_func)
 {
 	const int old_num_buffers = num_buffers;
-	pid_t children[16];
 
-	num_buffers /= ARRAY_SIZE(children);
+	num_buffers /= 16;
 	num_buffers += 2;
 
 	igt_fork_signal_helper();
 
-	for (int nc = 0; nc < ARRAY_SIZE(children); nc++) {
-		switch ((children[nc] = fork())) {
-		case -1: igt_assert(0);
-		default: break;
-		case 0:
-			 /* recreate process local variables */
-			 bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
-			 drm_intel_bufmgr_gem_enable_reuse(bufmgr);
-			 batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
-			 for (int i = 0; i < num_buffers; i++) {
-				 src[i] = mode->create_bo(bufmgr, i, width, height);
-				 dst[i] = mode->create_bo(bufmgr, ~i, width, height);
-			 }
-			 dummy = mode->create_bo(bufmgr, 0, width, height);
-			 for (int loop = 0; loop < 10; loop++)
-				 do_test_func(mode, src, dst, dummy);
-			 /* as we borrow the fd, we need to reap our bo */
-			 for (int i = 0; i < num_buffers; i++) {
-				 drm_intel_bo_unreference(src[i]);
-				 drm_intel_bo_unreference(dst[i]);
-			 }
-			 drm_intel_bo_unreference(dummy);
-			 intel_batchbuffer_free(batch);
-			 drm_intel_bufmgr_destroy(bufmgr);
-			 exit(0);
+	igt_fork(child, 16) {
+		igt_fail(6);
+
+		/* recreate process local variables */
+		bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
+		drm_intel_bufmgr_gem_enable_reuse(bufmgr);
+		batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
+		for (int i = 0; i < num_buffers; i++) {
+			src[i] = mode->create_bo(bufmgr, i, width, height);
+			dst[i] = mode->create_bo(bufmgr, ~i, width, height);
 		}
+		dummy = mode->create_bo(bufmgr, 0, width, height);
+		for (int loop = 0; loop < 10; loop++)
+			do_test_func(mode, src, dst, dummy);
+		/* as we borrow the fd, we need to reap our bo */
+		for (int i = 0; i < num_buffers; i++) {
+			drm_intel_bo_unreference(src[i]);
+			drm_intel_bo_unreference(dst[i]);
+		}
+		drm_intel_bo_unreference(dummy);
+		intel_batchbuffer_free(batch);
+		drm_intel_bufmgr_destroy(bufmgr);
 	}
-	for (int nc = 0; nc < ARRAY_SIZE(children); nc++) {
-		int status = -1;
-		while (waitpid(children[nc], &status, 0) == -1 &&
-		       errno == -EINTR)
-			;
-		igt_assert(status == 0);
-	}
+
+	igt_waitchildren();
 
 	igt_stop_signal_helper();