score: Add self-contained futex implementation

6 files changed, 283 insertions, 0 deletions

diff --git a/cpukit/libmisc/monitor/mon-prmisc.c b/cpukit/libmisc/monitor/mon-prmisc.c
index 4656ed297a..14ba173256 100644
--- a/cpukit/libmisc/monitor/mon-prmisc.c
+++ b/cpukit/libmisc/monitor/mon-prmisc.c
@@ -134,6 +134,7 @@ static const rtems_assoc_t rtems_monitor_state_assoc[] = {
     { "Wseg",   STATES_WAITING_FOR_SEGMENT, 0 },
     { "Wsem",   STATES_WAITING_FOR_SEMAPHORE, 0 },
     { "Wsig",   STATES_WAITING_FOR_SIGNAL, 0 },
+    { "Wslftx", STATES_WAITING_FOR_SYS_LOCK_FUTEX, 0 },
     { "Wslmtx", STATES_WAITING_FOR_SYS_LOCK_MUTEX, 0 },
     { "Wslsem", STATES_WAITING_FOR_SYS_LOCK_SEMAPHORE, 0 },
     { "Wsysev", STATES_WAITING_FOR_SYSTEM_EVENT, 0 },
diff --git a/cpukit/score/Makefile.am b/cpukit/score/Makefile.am
index 4b4c1f5fce..460076c6a9 100644
--- a/cpukit/score/Makefile.am
+++ b/cpukit/score/Makefile.am
@@ -345,6 +345,7 @@ libscore_a_SOURCES += src/apiext.c src/chain.c src/chainappend.c \
     src/interr.c src/isr.c src/wkspace.c src/wkstringduplicate.c
 libscore_a_SOURCES += src/isrisinprogress.c
 libscore_a_SOURCES += src/debugisownerofallocator.c
+libscore_a_SOURCES += src/futex.c
 libscore_a_SOURCES += src/profilingisrentryexit.c
 libscore_a_SOURCES += src/mutex.c
 libscore_a_SOURCES += src/once.c
diff --git a/cpukit/score/include/rtems/score/statesimpl.h b/cpukit/score/include/rtems/score/statesimpl.h
index 7805ca4c9b..82d222c490 100644
--- a/cpukit/score/include/rtems/score/statesimpl.h
+++ b/cpukit/score/include/rtems/score/statesimpl.h
@@ -90,6 +90,8 @@ extern "C" {
 #define STATES_WAITING_FOR_SYS_LOCK_MUTEX      0x2000000
 /** This macro corresponds to a task waiting for a <sys/lock.h> semaphore. */
 #define STATES_WAITING_FOR_SYS_LOCK_SEMAPHORE  0x4000000
+/** This macro corresponds to a task waiting for a <sys/lock.h> futex. */
+#define STATES_WAITING_FOR_SYS_LOCK_FUTEX      0x8000000
 
 /** This macro corresponds to a task which is in an interruptible
  *  blocking state.
@@ -109,6 +111,7 @@ extern "C" {
                                  STATES_WAITING_FOR_BSD_WAKEUP         | \
                                  STATES_WAITING_FOR_SYS_LOCK_MUTEX     | \
                                  STATES_WAITING_FOR_SYS_LOCK_SEMAPHORE | \
+                                 STATES_WAITING_FOR_SYS_LOCK_FUTEX     | \
                                  STATES_WAITING_FOR_RWLOCK             )
 
 /** This macro corresponds to a task waiting which is blocked. */
diff --git a/cpukit/score/src/futex.c b/cpukit/score/src/futex.c
new file mode 100644
index 0000000000..2ec38a3747
--- /dev/null
+++ b/cpukit/score/src/futex.c
@@ -0,0 +1,199 @@
+/*
+ * Copyright (c) 2015 embedded brains GmbH.  All rights reserved.
+ *
+ *  embedded brains GmbH
+ *  Dornierstr. 4
+ *  82178 Puchheim
+ *  Germany
+ *  <rtems@embedded-brains.de>
+ *
+ * The license and distribution terms for this file may be
+ * found in the file LICENSE in this distribution or at
+ * http://www.rtems.org/license/LICENSE.
+ */
+
+#if HAVE_CONFIG_H
+  #include "config.h"
+#endif
+
+#if HAVE_STRUCT__THREAD_QUEUE_QUEUE
+
+#include <sys/lock.h>
+#include <errno.h>
+
+#include <rtems/score/atomic.h>
+#include <rtems/score/chainimpl.h>
+#include <rtems/score/threadimpl.h>
+#include <rtems/score/threadqimpl.h>
+
+#define FUTEX_TQ_OPERATIONS &_Thread_queue_Operations_FIFO
+
+typedef struct {
+  Thread_queue_Syslock_queue Queue;
+} Futex_Control;
+
+RTEMS_STATIC_ASSERT(
+  offsetof( Futex_Control, Queue )
+    == offsetof( struct _Futex_Control, _Queue ),
+  FUTEX_CONTROL_QUEUE
+);
+
+RTEMS_STATIC_ASSERT(
+  sizeof( Futex_Control ) == sizeof( struct _Futex_Control ),
+  FUTEX_CONTROL_SIZE
+);
+
+static Futex_Control *_Futex_Get( struct _Futex_Control *_futex )
+{
+  return (Futex_Control *) _futex;
+}
+
+static Thread_Control *_Futex_Queue_acquire(
+  Futex_Control    *futex,
+  ISR_lock_Context *lock_context
+)
+{
+  Thread_Control *executing;
+
+  _ISR_lock_ISR_disable( lock_context );
+  executing = _Thread_Executing;
+  _Thread_queue_Queue_acquire_critical(
+    &futex->Queue.Queue,
+    &executing->Potpourri_stats,
+    lock_context
+  );
+
+  return executing;
+}
+
+static void _Futex_Queue_release(
+  Futex_Control    *futex,
+  ISR_lock_Context *lock_context
+)
+{
+  _Thread_queue_Queue_release( &futex->Queue.Queue, lock_context );
+}
+
+int _Futex_Wait( struct _Futex_Control *_futex, int *uaddr, int val )
+{
+  Futex_Control    *futex;
+  ISR_lock_Context  lock_context;
+  Thread_Control   *executing;
+  int               eno;
+
+  futex = _Futex_Get( _futex );
+  executing = _Futex_Queue_acquire( futex, &lock_context );
+
+  if ( *uaddr == val ) {
+    _Thread_queue_Enqueue_critical(
+      &futex->Queue.Queue,
+      FUTEX_TQ_OPERATIONS,
+      executing,
+      STATES_WAITING_FOR_SYS_LOCK_FUTEX,
+      0,
+      0,
+      &lock_context
+    );
+    eno = 0;
+  } else {
+    _Futex_Queue_release( futex, &lock_context );
+    eno = EWOULDBLOCK;
+  }
+
+  return eno;
+}
+
+/*
+ * Use a noinline function to force the compiler to set up and tear down the
+ * large stack frame only in the slow case.
+ */
+static __attribute__((noinline)) int _Futex_Wake_slow(
+  Futex_Control      *futex,
+  int                 count,
+  Thread_queue_Heads *heads,
+  ISR_lock_Context   *lock_context
+)
+{
+  Chain_Control  unblock;
+  Chain_Node    *node;
+  Chain_Node    *tail;
+  int            woken;
+
+  woken = 0;
+  _Chain_Initialize_empty( &unblock );
+
+  while ( count > 0 && heads != NULL ) {
+    const Thread_queue_Operations *operations;
+    Thread_Control                *first;
+    bool                           do_unblock;
+
+    operations = FUTEX_TQ_OPERATIONS;
+    first = ( *operations->first )( heads );
+
+    do_unblock = _Thread_queue_Extract_locked(
+      &futex->Queue.Queue,
+      operations,
+      first
+    );
+    if (do_unblock) {
+      _Chain_Append_unprotected( &unblock, &first->Wait.Node.Chain );
+    }
+
+    ++woken;
+    --count;
+    heads = futex->Queue.Queue.heads;
+  }
+
+  node = _Chain_First( &unblock );
+  tail = _Chain_Tail( &unblock );
+
+  if ( node != tail ) {
+    Per_CPU_Control *cpu_self;
+
+    cpu_self = _Thread_Dispatch_disable_critical( lock_context );
+    _Futex_Queue_release( futex, lock_context );
+
+    do {
+      Thread_Control *thread;
+      Chain_Node     *next;
+
+      next = _Chain_Next( node );
+      thread = THREAD_CHAIN_NODE_TO_THREAD( node );
+      _Thread_Unblock( thread );
+
+      node = next;
+    } while ( node != tail );
+
+    _Thread_Dispatch_enable( cpu_self );
+  } else {
+    _Futex_Queue_release( futex, lock_context );
+  }
+
+  return woken;
+}
+
+int _Futex_Wake( struct _Futex_Control *_futex, int count )
+{
+  Futex_Control      *futex;
+  ISR_lock_Context    lock_context;
+  Thread_queue_Heads *heads;
+
+  futex = _Futex_Get( _futex );
+  _Futex_Queue_acquire( futex, &lock_context );
+
+  /*
+   * For some synchronization objects like barriers the _Futex_Wake() must be
+   * called in the fast path.  Normally there are no threads on the queue, so
+   * check this condition early.
+   */
+  heads = futex->Queue.Queue.heads;
+  if ( __predict_true( heads == NULL ) ) {
+    _Futex_Queue_release( futex, &lock_context );
+
+    return 0;
+  }
+
+  return _Futex_Wake_slow( futex, count, heads, &lock_context );
+}
+
+#endif /* HAVE_STRUCT__THREAD_QUEUE_QUEUE */
diff --git a/testsuites/sptests/spsyslock01/init.c b/testsuites/sptests/spsyslock01/init.c
index ece5943492..72e2ad28e7 100644
--- a/testsuites/sptests/spsyslock01/init.c
+++ b/testsuites/sptests/spsyslock01/init.c
@@ -19,6 +19,8 @@
 #include "tmacros.h"
 
 #include <sys/lock.h>
+#include <errno.h>
+#include <limits.h>
 #include <string.h>
 #include <time.h>
 
@@ -42,6 +44,8 @@ const char rtems_test_name[] = "SPSYSLOCK 1";
 
 #define EVENT_SEM_WAIT RTEMS_EVENT_7
 
+#define EVENT_FUTEX_WAIT RTEMS_EVENT_8
+
 typedef struct {
   rtems_id high[2];
   rtems_id mid;
@@ -49,6 +53,9 @@ typedef struct {
   struct _Mutex_Control mtx;
   struct _Mutex_recursive_Control rec_mtx;
   struct _Semaphore_Control sem;
+  struct _Futex_Control futex;
+  int val;
+  int eno[2];
   int generation[2];
   int current_generation[2];
 } test_context;
@@ -87,18 +94,22 @@ static void test_initialization(test_context *ctx)
   struct _Mutex_Control mtx = _MUTEX_INITIALIZER;
   struct _Mutex_recursive_Control rec_mtx = _MUTEX_RECURSIVE_INITIALIZER;
   struct _Semaphore_Control sem = _SEMAPHORE_INITIALIZER(1);
+  struct _Futex_Control futex = _FUTEX_INITIALIZER;
 
   _Mutex_Initialize(&ctx->mtx);
   _Mutex_recursive_Initialize(&ctx->rec_mtx);
   _Semaphore_Initialize(&ctx->sem, 1);
+  _Futex_Initialize(&ctx->futex);
 
   rtems_test_assert(memcmp(&mtx, &ctx->mtx, sizeof(mtx)) == 0);
   rtems_test_assert(memcmp(&rec_mtx, &ctx->rec_mtx, sizeof(rec_mtx)) == 0);
   rtems_test_assert(memcmp(&sem, &ctx->sem, sizeof(sem)) == 0);
+  rtems_test_assert(memcmp(&futex, &ctx->futex, sizeof(futex)) == 0);
 
   _Mutex_Destroy(&mtx);
   _Mutex_recursive_Destroy(&rec_mtx);
   _Semaphore_Destroy(&sem);
+  _Futex_Destroy(&futex);
 }
 
 static void test_recursive_acquire_normal(test_context *ctx)
@@ -321,6 +332,63 @@ static void test_sem_prio_wait_order(test_context *ctx)
   rtems_test_assert(ctx->generation[b] == gen_b + 1);
 }
 
+static void test_futex(test_context *ctx)
+{
+  struct _Futex_Control *futex = &ctx->futex;
+  size_t a = 0;
+  size_t b = 1;
+  int eno;
+  int woken;
+
+  eno = _Futex_Wait(futex, &ctx->val, 1);
+  rtems_test_assert(eno == EWOULDBLOCK);
+
+  woken = _Futex_Wake(futex, 0);
+  rtems_test_assert(woken == 0);
+
+  woken = _Futex_Wake(futex, 1);
+  rtems_test_assert(woken == 0);
+
+  ctx->val = 1;
+
+  ctx->eno[a] = -1;
+  send_event(ctx, a, EVENT_FUTEX_WAIT);
+  rtems_test_assert(ctx->eno[a] == -1);
+
+  woken = _Futex_Wake(futex, INT_MAX);
+  rtems_test_assert(woken == 1);
+  rtems_test_assert(ctx->eno[a] == 0);
+
+  ctx->eno[a] = -1;
+  ctx->eno[b] = -1;
+  send_event(ctx, a, EVENT_FUTEX_WAIT);
+  send_event(ctx, b, EVENT_FUTEX_WAIT);
+  rtems_test_assert(ctx->eno[a] == -1);
+  rtems_test_assert(ctx->eno[b] == -1);
+
+  woken = _Futex_Wake(futex, 1);
+  rtems_test_assert(woken == 1);
+  rtems_test_assert(ctx->eno[a] == 0);
+  rtems_test_assert(ctx->eno[b] == -1);
+
+  woken = _Futex_Wake(futex, 1);
+  rtems_test_assert(woken == 1);
+  rtems_test_assert(ctx->eno[a] == 0);
+  rtems_test_assert(ctx->eno[b] == 0);
+
+  ctx->eno[a] = -1;
+  ctx->eno[b] = -1;
+  send_event(ctx, a, EVENT_FUTEX_WAIT);
+  send_event(ctx, b, EVENT_FUTEX_WAIT);
+  rtems_test_assert(ctx->eno[a] == -1);
+  rtems_test_assert(ctx->eno[b] == -1);
+
+  woken = _Futex_Wake(futex, 2);
+  rtems_test_assert(woken == 2);
+  rtems_test_assert(ctx->eno[a] == 0);
+  rtems_test_assert(ctx->eno[b] == 0);
+}
+
 static void mid_task(rtems_task_argument arg)
 {
   rtems_test_assert(0);
@@ -402,6 +470,10 @@ static void high_task(rtems_task_argument idx)
       _Semaphore_Wait(&ctx->sem);
       ctx->generation[idx] = generation(ctx, idx);
     }
+
+    if ((events & EVENT_FUTEX_WAIT) != 0) {
+      ctx->eno[idx] = _Futex_Wait(&ctx->futex, &ctx->val, 1);
+    }
   }
 }
 
@@ -459,12 +531,14 @@ static void test(void)
   test_mtx_timeout_recursive(ctx);
   test_sem(ctx);
   test_sem_prio_wait_order(ctx);
+  test_futex(ctx);
 
   send_event(ctx, 0, EVENT_MTX_DEADLOCK);
 
   _Mutex_Destroy(&ctx->mtx);
   _Mutex_recursive_Destroy(&ctx->rec_mtx);
   _Semaphore_Destroy(&ctx->sem);
+  _Futex_Destroy(&ctx->futex);
 }
 
 static void Init(rtems_task_argument arg)
diff --git a/testsuites/sptests/spsyslock01/spsyslock01.doc b/testsuites/sptests/spsyslock01/spsyslock01.doc
index 1a7384ce7e..bf59cebeff 100644
--- a/testsuites/sptests/spsyslock01/spsyslock01.doc
+++ b/testsuites/sptests/spsyslock01/spsyslock01.doc
@@ -18,8 +18,13 @@ directives:
   - _Semaphore_Wait()
   - _Semaphore_Post()
   - _Semaphore_Destroy()
+  - _Futex_Initialize()
+  - _Futex_Wait()
+  - _Futex_Wake()
+  - _Futex_Destroy()
 
 concepts:
 
   - Ensure that self-contained mutexes and recursive mutexes work.
   - Ensure that self-contained semaphores work.
+  - Ensure that self-contained futexes work.