SCHED_ENQUEUE: tighten checks on locking

Factor out the two different implementations and ensure
modifications happen with inteerrupts disabled.  This uncovered
a latent bug in our semaphore sleep code.

Also explicitly document our locking requirements for ProcReadyList.
The comment previously claimed that proc_forbid() would offer enough
protection, which is, of course, bullshit.

git-svn-id: https://src.develer.com/svnoss/bertos/trunk@1773 38d2e660-2303-0410-9eaa-f027e97ec537

diff --git a/bertos/kern/proc_p.h b/bertos/kern/proc_p.h
index 9d61b53123b369010532a3ad1765d6d5f239d103..d99d5fc187d6090eea283304d43c3a49e881d1e5 100644 (file)
--- a/bertos/kern/proc_p.h
+++ b/bertos/kern/proc_p.h
@@ -104,34 +104,31 @@ extern REGISTER Process   *CurrentProcess;
 /**
  * Track ready processes.
  *
- * Access to this list must be protected with a proc_forbid() / proc_premit()
- * pair, or with SCHED_ATOMIC()
+ * Access to this list must be performed with interrupts disabled
  */
 extern REGISTER List     ProcReadyList;
 
 #if CONFIG_KERN_PRI
-       /**
-        * Enqueue a task in the ready list.
-        *
-        * Always use this macro to instert a process in the ready list, as its
-        * might vary to implement a different scheduling algorithms.
-        *
-        * \note This macro is *NOT* protected against the scheduler.  Access to
-        *       this list must be performed with interrupts disabled.
-        */
-       #define SCHED_ENQUEUE(proc)  do { \
-                       LIST_ASSERT_VALID(&ProcReadyList); \
-                       LIST_ENQUEUE(&ProcReadyList, &(proc)->link); \
-               } while (0)
-
-#else // !CONFIG_KERN_PRI
-
-       #define SCHED_ENQUEUE(proc)  do { \
-                       LIST_ASSERT_VALID(&ProcReadyList); \
-                       ADDTAIL(&ProcReadyList, &(proc)->link); \
-               } while (0)
-
-#endif // !CONFIG_KERN_PRI
+       #define SCHED_ENQUEUE_INTERNAL(proc) LIST_ENQUEUE(&ProcReadyList, &(proc)->link)
+#else
+       #define SCHED_ENQUEUE_INTERNAL(proc) ADDTAIL(&ProcReadyList, &(proc)->link)
+#endif
+
+/**
+ * Enqueue a process in the ready list.
+ *
+ * Always use this macro to instert a process in the ready list, as its
+ * might vary to implement a different scheduling algorithms.
+ *
+ * \note Access to the scheduler ready list must be performed with
+ *       interrupts disabled.
+ */
+#define SCHED_ENQUEUE(proc)  do { \
+               IRQ_ASSERT_DISABLED(); \
+               LIST_ASSERT_VALID(&ProcReadyList); \
+               SCHED_ENQUEUE_INTERNAL(proc); \
+       } while (0)
+
 
 /// Schedule another process *without* adding the current one to the ready list.
 void proc_switch(void);

diff --git a/bertos/kern/sem.c b/bertos/kern/sem.c
index cc9234aa64a020531a6c1e354d93bd595fd80f6d..7d07f4d6f61585b928e1361186857c0071cca5fd 100644 (file)
--- a/bertos/kern/sem.c
+++ b/bertos/kern/sem.c
@@ -28,17 +28,16 @@
  *
  * Copyright 2001, 2004 Develer S.r.l. (http://www.develer.com/)
  * Copyright 1999, 2000, 2001 Bernie Innocenti <bernie@codewiz.org>
- *
  * -->
  *
  * \brief Semaphore based synchronization services.
  *
  * \version $Id$
- *
  * \author Bernie Innocenti <bernie@codewiz.org>
  */
 
 #include "sem.h"
+#include <cpu/irq.h> // ASSERT_IRQ_DISABLED()
 #include <kern/proc.h>
 #include <kern/proc_p.h>
 #include <kern/signal.h>
@@ -178,7 +177,7 @@ void sem_release(struct Semaphore *s)
                {
                        s->nest_count = 1;
                        s->owner = proc;
-                       SCHED_ENQUEUE(proc);
+                       ATOMIC(SCHED_ENQUEUE(proc));
                }
        }