Merge pull request torvalds#225 from tavip/syscall-threads-autocleanup

Syscall threads autocleanup

arch/lkl/include/uapi/asm/host_ops.h

@@ -38,10 +38,12 @@ typedef unsigned long lkl_thread_t;
  * @thread_join - wait for the given thread to terminate. Returns 0
  * for success, -1 otherwise
  *
- * @tls_alloc - allocate a thread local storage key; returns 0 if succesful
+ * @tls_alloc - allocate a thread local storage key; returns 0 if successful; if
+ * destructor is not NULL it will be called when a thread terminates with its
+ * argument set to the current thread local storage value
  * @tls_free - frees a thread local storage key; returns 0 if succesful
  * @tls_set - associate data to the thread local storage key; returns 0 if
- * succesful
+ * successful
  * @tls_get - return data associated with the thread local storage key or NULL
  * on error
  *
@@ -85,7 +87,7 @@ struct lkl_host_operations {
 	void (*thread_exit)(void);
 	int (*thread_join)(lkl_thread_t tid);
 
-	int (*tls_alloc)(unsigned int *key);
+	int (*tls_alloc)(unsigned int *key, void (*destructor)(void *));
 	int (*tls_free)(unsigned int key);
 	int (*tls_set)(unsigned int key, void *data);
 	void *(*tls_get)(unsigned int key);

arch/lkl/kernel/syscalls.c

@@ -80,8 +80,8 @@ static irqreturn_t syscall_irq_handler(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
-static int __lkl_stop_syscall_thread(struct syscall_thread_data *data,
-				     bool host);
+static void cleanup_syscall_threads(void);
+
 int syscall_thread(void *_data)
 {
 	struct syscall_thread_data *data;
@@ -131,15 +131,8 @@ int syscall_thread(void *_data)
 		lkl_ops->sem_up(data->completion);
 	}
 
-	if (data == &default_syscall_thread_data) {
-		struct syscall_thread_data *i = NULL, *aux;
-
-		list_for_each_entry_safe(i, aux, &syscall_threads, list) {
-			if (i == &default_syscall_thread_data)
-				continue;
-			__lkl_stop_syscall_thread(i, false);
-		}
-	}
+	if (data == &default_syscall_thread_data)
+		cleanup_syscall_threads();
 
 	pr_info("lkl: exiting syscall thread %s\n", current->comm);
 
@@ -331,12 +324,82 @@ sys_create_syscall_thread(struct syscall_thread_data *data)
 	return 0;
 }
 
+
+/*
+ * A synchronization algorithm between cleanup_syscall_threads (which terminates
+ * all remaining syscall threads) and destructors functions (which frees a
+ * syscall thread as soon as the associated host thread terminates) is required
+ * since destructor functions run in host context and is not subject to kernel
+ * scheduling.
+ *
+ * An atomic counter is used to count the number of running destructor functions
+ * and allows the cleanup function to wait for the running destructor functions
+ * to complete.
+ *
+ * The cleanup functions adds MAX_SYSCALL_THREADS to this counter and this
+ * allows the destructor functions to check if the cleanup process has started
+ * and abort execution. This prevents "late" destructors from trying to free the
+ * syscall threads.
+ *
+ * This algorithm assumes that we never have more the MAX_SYSCALL_THREADS
+ * running.
+ */
+#define MAX_SYSCALL_THREADS 1000000
+static unsigned int destrs;
+
+/*
+ * This is called when the host thread terminates if auto_syscall_threads is
+ * enabled. We use it to remove the associated kernel syscall thread since it is
+ * not going to be used anymore.
+ *
+ * Note that this run in host context, not kernel context.
+ *
+ * To avoid races between the destructor and lkl_sys_halt we announce that a
+ * destructor is running and also check to see if lkl_sys_halt is running, in
+ * which case we bail out - the kernel thread is going to be / has been stopped
+ * by lkl_sys_halt.
+ */
+static void syscall_thread_destructor(void *_data)
+{
+	struct syscall_thread_data *data = _data;
+
+	if (!data)
+		return;
+
+	if (__sync_fetch_and_add(&destrs, 1) < MAX_SYSCALL_THREADS)
+		__lkl_stop_syscall_thread(data, true);
+	__sync_fetch_and_sub(&destrs, 1);
+}
+
+static void cleanup_syscall_threads(void)
+{
+	struct syscall_thread_data *i = NULL, *aux;
+
+	/* announce destructors that we are stopping */
+	__sync_fetch_and_add(&destrs, MAX_SYSCALL_THREADS);
+
+	/* wait for any pending destructors to complete */
+	while (__sync_fetch_and_add(&destrs, 0) > MAX_SYSCALL_THREADS)
+		schedule_timeout(1);
+
+	/* no more destructors, we can safely remove the remaining threads */
+	list_for_each_entry_safe(i, aux, &syscall_threads, list) {
+		if (i == &default_syscall_thread_data)
+			continue;
+		__lkl_stop_syscall_thread(i, false);
+	}
+}
+
 int initial_syscall_thread(void *sem)
 {
+	void (*destr)(void *) = NULL;
 	int ret = 0;
 
+	if (auto_syscall_threads)
+		destr = syscall_thread_destructor;
+
 	if (lkl_ops->tls_alloc)
-		ret = lkl_ops->tls_alloc(&syscall_thread_data_key);
+		ret = lkl_ops->tls_alloc(&syscall_thread_data_key, destr);
 	if (ret)
 		return ret;
 

tools/lkl/lib/Build

@@ -1,5 +1,6 @@
 CFLAGS_posix-host.o += -D_FILE_OFFSET_BITS=64
 CFLAGS_virtio_net_vde.o += $(shell pkg-config --cflags vdeplug)
+CFLAGS_nt-host.o += -D_WIN32_WINNT=0x0600
 
 lkl-y += fs.o
 lkl-y += iomem.o

tools/lkl/lib/nt-host.c

@@ -88,25 +88,28 @@ static int thread_join(lkl_thread_t tid)
 	return 0;
 }
 
-static int tls_alloc(unsigned int *key)
+static int tls_alloc(unsigned int *key, void (*destructor)(void *))
 {
-	*key = TlsAlloc();
+	*key = FlsAlloc((PFLS_CALLBACK_FUNCTION)destructor);
 	return *key == TLS_OUT_OF_INDEXES ? -1 : 0;
 }
 
 static int tls_free(unsigned int key)
 {
-	return TlsFree(key) ? 0 : -1;
+	/* setting to NULL first to prevent the callback from being called */
+	if (!FlsSetValue(key, NULL))
+		return -1;
+	return FlsFree(key) ? 0 : -1;
 }
 
 static int tls_set(unsigned int key, void *data)
 {
-	return TlsSetValue(key, data) ? 0 : -1;
+	return FlsSetValue(key, data) ? 0 : -1;
 }
 
 static void *tls_get(unsigned int key)
 {
-	return TlsGetValue(key);
+	return FlsGetValue(key);
 }
 
 

tools/lkl/lib/posix-host.c

@@ -201,9 +201,9 @@ static int thread_join(lkl_thread_t tid)
 		return 0;
 }
 
-static int tls_alloc(unsigned int *key)
+static int tls_alloc(unsigned int *key, void (*destructor)(void *))
 {
-	return pthread_key_create((pthread_key_t*)key, NULL);
+	return pthread_key_create((pthread_key_t *)key, destructor);
 }
 
 static int tls_free(unsigned int key)

tools/lkl/tests/boot.c

@@ -729,7 +729,6 @@ static void test_thread(void *data)
 		fprintf(stderr, "%s: %s\n", __func__, lkl_strerror(ret));
 	}
 
-	lkl_stop_syscall_thread();
 }
 
 static int test_syscall_thread(char *str, int len)
@@ -775,7 +774,36 @@ static int test_syscall_thread(char *str, int len)
 	return TEST_SUCCESS;
 }
 
-void thread_quit_immediately(void *unused)
+#ifndef __MINGW32__
+static void thread_get_pid(void *unused)
+{
+	lkl_sys_getpid();
+}
+
+static int test_many_syscall_threads(char *str, int len)
+{
+	lkl_thread_t tid;
+	int count = 65, ret;
+
+	while (--count > 0) {
+		tid = lkl_host_ops.thread_create(thread_get_pid, NULL);
+		if (!tid) {
+			snprintf(str, len, "failed to create thread");
+			return TEST_FAILURE;
+		}
+
+		ret = lkl_host_ops.thread_join(tid);
+		if (ret) {
+			snprintf(str, len, "failed to join thread");
+			return TEST_FAILURE;
+		}
+	}
+
+	return TEST_SUCCESS;
+}
+#endif
+
+static void thread_quit_immediately(void *unused)
 {
 }
 
@@ -911,6 +939,14 @@ int main(int argc, char **argv)
 	TEST(gettid);
 	TEST(syscall_thread);
 	TEST(join);
+	/*
+	 * Wine has an issue where the FlsCallback is not called when the thread
+	 * terminates which makes testing the automatic syscall threads cleanup
+	 * impossible under wine.
+	 */
+#ifndef __MINGW32__
+	TEST(many_syscall_threads);
+#endif
 
 	lkl_sys_halt();