[Xenomai-core] Xenomai-Solo: bug in vxWorks taskSpawn with twice the same name

[Xenomai-core] Xenomai-Solo: bug in vxWorks taskSpawn with twice the same name

[Niklaus Giger]

[-- Attachment #1: Type: text/plain, Size: 614 bytes --]

Hi

Thanks a lot to Philippe for fixing the previous bug so quickly! Never got
fixed a bug as fast by WindRIver despite having paid quite a lot of money
for it!

Now my testsuite passes almost all tests but crashes, when it spawns
twice a task with the same name. E.g. I get the following output:
> testTask done t1 0x901a0f4 t2 0x901a2b4
> Xenomai/SOLO: duplicate task name: Test
> assert passed at task1 line 24
> assert passed at task1 line 24
> assert passed at task1 line 26
> assert passed at task1 line 26
>
> Command terminated by signal 11

Attached my simple test case.

Best regards

-- 
NIklaus Giger


[-- Attachment #2: tst_main.cpp --]
[-- Type: text/x-c++src, Size: 2339 bytes --]

#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>


#define assert(cond) \
  if (!cond) printf("assert failed at %s line %d\n", __FUNCTION__, __LINE__); \
  else  printf("assert passed at %s line %d\n", __FUNCTION__, __LINE__);


#if defined(__SOLO__) || defined(_WRS_VXWORKS_MAJOR)
// #include <vxWorks.h>
#include <semLib.h>
#include <taskLib.h>
#include <tickLib.h>
#include <kernelLib.h>

void testTask(long a1, long a2, long a3, long a4, long a5,
	long a6, long a7, long a8, long a9, long a10);
bool finished;

  extern "C" void task1(long int, long int, long int, long int, long int, long int, long int, long int, long int, long int)
  {
	assert(1);
	taskDelay(30);	
	assert(2);
	finished = true;
  }

#endif

/**
 *  Main functions
 */
#if defined(PLAT_VXWORKS) || defined(_WRS_VXWORKS_MAJOR)
  extern "C" int tSysMain(const char* arg)
  {
    const char* argv[2];
    argv[0]  = "undefined";
    argv[1]  = arg;
    int argc = sizeof(argv) / sizeof(char*);
#else

  int main(int argc, char *const argv[])
  {
#endif

#ifdef __SOLO__

    kernelInit(testTask, argc, argv);
    while(!finished) { sleep(1); }
    exit(0);
}

void testTask(long a1, long a2, long a3, long a4, long a5,
	long a6, long a7, long a8, long a9, long a10)
{
   int argc=0; char *const argv[]={0};
#endif
   int t1 =taskSpawn("Test", // name of new task (stored at pStackBase)
                     90, // priority of new task
                     0, // task option word
                     20*1024, // size (bytes) of stack needed plus name
                     task1,  // entry point of new task
                     0, // first argument
                     0,0,0,0,0,0,0,0,0);   // unused parameters

   int t2 =taskSpawn("Test", // name of new task (stored at pStackBase)
                     90, // priority of new task
                     0, // task option word
                     20*1024, // size (bytes) of stack needed plus name
                     task1,  // entry point of new task
                     0, // first argument
                     0,0,0,0,0,0,0,0,0);   // unused parameters

  printf("%s done t1 0x%x t2 0x%x\n", __FUNCTION__, t1, t2);
  while (!finished) sleep(1);
#ifdef __SOLO__
  finished = true;
  return;
#else
  return 0;
#endif
}


//  ___________________________________________________________________________

Re: [Xenomai-core] Xenomai-Solo: bug in vxWorks taskSpawn with twice the same name

[Philippe Gerum]

Niklaus Giger wrote:
> Hi
> 
> Thanks a lot to Philippe for fixing the previous bug so quickly! Never got
> fixed a bug as fast by WindRIver despite having paid quite a lot of money
> for it!
>

inertia = success / competition

> Now my testsuite passes almost all tests but crashes, when it spawns
> twice a task with the same name. E.g. I get the following output:
>> testTask done t1 0x901a0f4 t2 0x901a2b4
>> Xenomai/SOLO: duplicate task name: Test
>> assert passed at task1 line 24
>> assert passed at task1 line 24
>> assert passed at task1 line 26
>> assert passed at task1 line 26
>>
>> Command terminated by signal 11
> 
> Attached my simple test case.
>

I eventually found out that this problem was triggered by the registry support,
in case duplicate names where found. This patch should fix the issue:

diff --git a/base/registry.c b/base/registry.c
index 6ab7461..d0ada62 100644
--- a/base/registry.c
+++ b/base/registry.c
@@ -116,11 +116,28 @@ done:
 	return ret;
 }

-int registry_add_file(struct fsobj *fsobj, struct registry_operations *ops,
-		      int mode, const char *fmt, ...)
+int registry_init_file(struct fsobj *fsobj,  struct registry_operations *ops)
 {
-	char path[PATH_MAX], *basename, *dir;
 	pthread_mutexattr_t mattr;
+
+	if (__no_registry_arg)
+		return 0;
+
+	fsobj->path = NULL;
+	fsobj->ops = ops;
+	holder_init(&fsobj->link);
+
+	pthread_mutexattr_init(&mattr);
+	pthread_mutexattr_setprotocol(&mattr, PTHREAD_PRIO_INHERIT);
+	pthread_mutex_init(&fsobj->lock, &mattr);
+	pthread_mutexattr_destroy(&mattr);
+
+	return 0;
+}
+
+int registry_add_file(struct fsobj *fsobj, int mode, const char *fmt, ...)
+{
+	char path[PATH_MAX], *basename, *dir;
 	struct hashobj *hobj;
 	struct regfs_dir *d;
 	va_list ap;
@@ -139,16 +156,9 @@ int registry_add_file(struct fsobj *fsobj, struct
registry_operations *ops,

 	fsobj->path = xnstrdup(path);
 	fsobj->basename = fsobj->path + (basename - path) + 1;
-	fsobj->ops = ops;
 	fsobj->mode = mode & O_ACCMODE;
 	clock_gettime(CLOCK_REALTIME, &fsobj->ctime);
 	fsobj->mtime = fsobj->ctime;
-	holder_init(&fsobj->link);
-
-	pthread_mutexattr_init(&mattr);
-	pthread_mutexattr_setprotocol(&mattr, PTHREAD_PRIO_INHERIT);
-	pthread_mutex_init(&fsobj->lock, &mattr);
-	pthread_mutexattr_destroy(&mattr);

 	pthread_mutex_lock(&regfs_lock);

@@ -161,7 +171,9 @@ int registry_add_file(struct fsobj *fsobj, struct
registry_operations *ops,
 	hobj = hash_search(&regfs_dirtable, dir);
 	if (hobj == NULL) {
 	fail:
+		hash_remove(&regfs_objtable, &fsobj->hobj);
 		xnfree(path);
+		fsobj->path = NULL;
 		ret = -ENOENT;
 		goto done;
 	}
@@ -184,6 +196,10 @@ void registry_remove_file(struct fsobj *fsobj)
 		return;

 	pthread_mutex_lock(&regfs_lock);
+
+	if (fsobj->path == NULL)
+		goto out;	/* Not registered. */
+
 	hash_remove(&regfs_objtable, &fsobj->hobj);

 	pthread_mutex_lock(&fsobj->lock);
@@ -194,7 +210,7 @@ void registry_remove_file(struct fsobj *fsobj)
 	xnfree(fsobj->path);
 	pthread_mutex_unlock(&fsobj->lock);
 	pthread_mutex_destroy(&fsobj->lock);
-
+out:
 	pthread_mutex_unlock(&regfs_lock);
 }

diff --git a/include/xenomai/registry.h b/include/xenomai/registry.h
index 15b98a3..e7f6247 100644
--- a/include/xenomai/registry.h
+++ b/include/xenomai/registry.h
@@ -57,8 +57,10 @@ extern "C" {

 int registry_add_dir(const char *fmt, ...);

+int registry_init_file(struct fsobj *fsobj,
+		       struct registry_operations *ops);
+
 int registry_add_file(struct fsobj *fsobj,
-		      struct registry_operations *ops,
 		      int mode,
 		      const char *fmt, ...);

@@ -89,8 +91,14 @@ int registry_add_dir(const char *fmt, ...)
 }

 static inline
+int registry_init_file(struct fsobj *fsobj,
+		       struct registry_operations *ops)
+{
+	return 0;
+}
+
+static inline
 int registry_add_file(struct fsobj *fsobj,
-		      struct registry_operations *ops,
 		      int mode,
 		      const char *fmt, ...)
 {
diff --git a/vxworks/taskLib.c b/vxworks/taskLib.c
index 8c1e612..f6253b2 100644
--- a/vxworks/taskLib.c
+++ b/vxworks/taskLib.c
@@ -234,17 +234,18 @@ static void *task_trampoline(void *arg)
 		goto done;
 	}

+	ret = registry_init_file(&task->fsobj, &registry_ops);
+
 	if (hash_enter(&wind_task_table, task->name, &task->obj)) {
 		warning("duplicate task name: %s", task->name);
 		/* Make sure we won't un-hash the previous one. */
 		strcpy(task->name, "(dup)");
-	} else {
-		ret = registry_add_file(&task->fsobj, &registry_ops, O_RDONLY,
+	} else if (ret == 0)
+		ret = registry_add_file(&task->fsobj, O_RDONLY,
 					"/vxworks/tasks/%s", task->name);
-		if (ret)
-			warning("failed to export task %s to registry",
-				task->name);
-	}
+	if (ret)
+		warning("failed to export task %s to registry",
+			task->name);

 	pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL);
 	pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, NULL);

-- 
Philippe.