Re: [Xenomai-core] xenomai-solo vxWorks: know whether a task is safe or not
Niklaus Giger wrote:
> Am Montag 22 September 2008 18.17:10 schrieb Philippe Gerum:
>> Niklaus Giger wrote:
>>> Hi
>>>
>>> For various reasons we were forced to examine under vxWorks 5.5 the
>>> private WIND_TCB safeCnt field to determine whether a task was "safe" or
>>> not.
>>>
>>> With the attached patch I try to add a taskIsSafe procedure to
>>> xenomai-solo. It is probably not free from race condition, as I do not
>>> know how to access a pthread_mutex.
>>>
>>> Would such (may be improved) patch find its way into the "official"
>>> trunk?
>> Assuming that VxWorks tracks the locking depth in safeCnt, that
>> implementation would not properly account for nested locking (safelock is a
>> recursive mutex). Additionally, since WIND_TCB is something of a
>> semi-public structure in 5.x and earlier versions, we could just track the
>> safe counter there as well, instead of adding a non-standard taskIsSafe()
>> call; we already do that for the status field anyway. Here is a possible
>> implementation:
>
> This implementation is okay for me, too. It seems to work, as far I could
> test it.
>
> I however still have some issues because we used sometimes (for historical
> reasons in order to emulate a pre-vxWorks home grown cooperative
> mulit-tasking
> operation system) a taskInit/taskActivate/taskRestart combination.
>
> Would it be a lot of work for you to add taskRestart to xenomai-solo?
>
I must admit that I always considered the task restart interface as a pure
manifestation of the Evil, so its absence from SOLO might not be totally
fortuitous. Trashing your system with that stuff is way to easy.
Anyway, here is a possible implementation based on - cough, cough... -
sigsetjmp(); I don't like it since it adds code to various hot paths just to
provide a rarely used feature, so I may make this optional using a configuration
knob at build time if/when I actually merge that, mmmh, thing.
Please let me know how that works for you.
--
Philippe.
diff --git a/include/vxworks/taskLib.h b/include/vxworks/taskLib.h
index 6334396..2f09de6 100644
--- a/include/vxworks/taskLib.h
+++ b/include/vxworks/taskLib.h
@@ -39,6 +39,7 @@
#define WIND_DELAY 0x4
#define WIND_DEAD 0x8
#define WIND_STOP 0x10 /* Never reported. */
+#define WIND_RESTART 0x20 /* Internal. */
typedef intptr_t TASK_ID;
@@ -78,6 +79,8 @@ STATUS taskInit(WIND_TCB *pTcb,
STATUS taskActivate(TASK_ID tid);
+STATUS taskRestart(TASK_ID tid);
+
STATUS taskDelete(TASK_ID tid);
STATUS taskDeleteForce(TASK_ID tid);
diff --git a/vxworks/msgQLib.c b/vxworks/msgQLib.c
index d591a7b..e3d071d 100644
--- a/vxworks/msgQLib.c
+++ b/vxworks/msgQLib.c
@@ -149,6 +149,8 @@ int msgQReceive(MSG_Q_ID msgQId, char *buffer, UINT maxNBytes, int timeout)
return ERROR;
}
+ check_task_restart(wind_task_current());
+
mq = get_mq_from_id(msgQId);
if (mq == NULL) {
objid_error:
@@ -217,6 +219,8 @@ STATUS msgQSend(MSG_Q_ID msgQId, const char *buffer, UINT bytes,
UINT maxbytes;
int ret;
+ check_task_restart(wind_task_current());
+
mq = get_mq_from_id(msgQId);
if (mq == NULL) {
objid_error:
diff --git a/vxworks/semLib.c b/vxworks/semLib.c
index 6467354..dfc37cc 100644
--- a/vxworks/semLib.c
+++ b/vxworks/semLib.c
@@ -61,7 +61,9 @@ static STATUS xsem_take(struct wind_sem *sem, int timeout)
if (threadobj_async_p())
return S_intLib_NOT_ISR_CALLABLE;
-
+
+ check_task_restart(wind_task_current());
+
if (syncobj_lock(&sem->u.xsem.sobj, &syns))
return S_objLib_OBJ_ID_ERROR;
@@ -196,8 +198,11 @@ static STATUS msem_take(struct wind_sem *sem, int timeout)
return S_intLib_NOT_ISR_CALLABLE;
current = wind_task_current();
- if (current != NULL && (sem->options & SEM_DELETE_SAFE))
- pthread_mutex_lock(¤t->safelock);
+ if (current != NULL) {
+ check_task_restart(current);
+ if (sem->options & SEM_DELETE_SAFE)
+ pthread_mutex_lock(¤t->safelock);
+ }
if (timeout == NO_WAIT) {
ret = pthread_mutex_trylock(&sem->u.msem.lock);
diff --git a/vxworks/taskLib.c b/vxworks/taskLib.c
index a9e12e7..7e74cf0 100644
--- a/vxworks/taskLib.c
+++ b/vxworks/taskLib.c
@@ -97,6 +97,8 @@ struct wind_task *get_wind_task_or_self(TASK_ID tid)
if (current == NULL)
return NULL;
+ check_task_restart(current);
+
/* This one might block but can't fail, it is ours. */
threadobj_lock(¤t->thobj);
@@ -114,6 +116,12 @@ void put_wind_task(struct wind_task *task)
threadobj_unlock(&task->thobj);
}
+void check_task_restart(struct wind_task *current)
+{
+ if (current && (current->tcb->status & WIND_RESTART) != 0)
+ siglongjmp(current->ienv, 1);
+}
+
static void task_finalizer(struct threadobj *thobj)
{
struct wind_task *task = container_of(thobj, struct wind_task, thobj);
@@ -253,6 +261,11 @@ static void *task_trampoline(void *arg)
/* Wait for someone to run taskActivate() upon us. */
sem_wait(&task->barrier);
+ threadobj_lock(&task->thobj);
+ sigsetjmp(task->ienv, 1); /* For taskRestart(). */
+ task->tcb->status &= ~WIND_REST
Re: [Xenomai-core] xenomai-solo vxWorks: know whether a task is safe or not
Niklaus Giger wrote:
> Am Montag 22 September 2008 18.17:10 schrieb Philippe Gerum:
>> Niklaus Giger wrote:
>>> Hi
>>>
>>> For various reasons we were forced to examine under vxWorks 5.5 the
>>> private WIND_TCB safeCnt field to determine whether a task was "safe" or
>>> not.
>>>
>>> With the attached patch I try to add a taskIsSafe procedure to
>>> xenomai-solo. It is probably not free from race condition, as I do not
>>> know how to access a pthread_mutex.
>>>
>>> Would such (may be improved) patch find its way into the "official"
>>> trunk?
>> Assuming that VxWorks tracks the locking depth in safeCnt, that
>> implementation would not properly account for nested locking (safelock is a
>> recursive mutex). Additionally, since WIND_TCB is something of a
>> semi-public structure in 5.x and earlier versions, we could just track the
>> safe counter there as well, instead of adding a non-standard taskIsSafe()
>> call; we already do that for the status field anyway. Here is a possible
>> implementation:
>
> This implementation is okay for me, too. It seems to work, as far I could
> test it.
>
> I however still have some issues because we used sometimes (for historical
> reasons in order to emulate a pre-vxWorks home grown cooperative
> mulit-tasking
> operation system) a taskInit/taskActivate/taskRestart combination.
>
> Would it be a lot of work for you to add taskRestart to xenomai-solo?
>
I must admit that I always considered the task restart interface as a pure
manifestation of the Evil, so it absence in SOLO might not be totally
fortuitous. Trashing your system with that stuff is way to easy.
Anyway, here is a possible implementation based on - cough, cough... -
sigsetjmp(); I don't like it since it adds code to various hot paths just to
provide a rarely used feature, so I may make this optional using a configuration
knob at build time if/when I actually merge that, mmmh, thing.
Please let me know how that works for you.
--
Philippe.
diff --git a/include/vxworks/taskLib.h b/include/vxworks/taskLib.h
index 6334396..2f09de6 100644
--- a/include/vxworks/taskLib.h
+++ b/include/vxworks/taskLib.h
@@ -39,6 +39,7 @@
#define WIND_DELAY 0x4
#define WIND_DEAD 0x8
#define WIND_STOP 0x10 /* Never reported. */
+#define WIND_RESTART 0x20 /* Internal. */
typedef intptr_t TASK_ID;
@@ -78,6 +79,8 @@ STATUS taskInit(WIND_TCB *pTcb,
STATUS taskActivate(TASK_ID tid);
+STATUS taskRestart(TASK_ID tid);
+
STATUS taskDelete(TASK_ID tid);
STATUS taskDeleteForce(TASK_ID tid);
diff --git a/vxworks/msgQLib.c b/vxworks/msgQLib.c
index d591a7b..e3d071d 100644
--- a/vxworks/msgQLib.c
+++ b/vxworks/msgQLib.c
@@ -149,6 +149,8 @@ int msgQReceive(MSG_Q_ID msgQId, char *buffer, UINT maxNBytes, int timeout)
return ERROR;
}
+ check_task_restart(wind_task_current());
+
mq = get_mq_from_id(msgQId);
if (mq == NULL) {
objid_error:
@@ -217,6 +219,8 @@ STATUS msgQSend(MSG_Q_ID msgQId, const char *buffer, UINT bytes,
UINT maxbytes;
int ret;
+ check_task_restart(wind_task_current());
+
mq = get_mq_from_id(msgQId);
if (mq == NULL) {
objid_error:
diff --git a/vxworks/semLib.c b/vxworks/semLib.c
index 6467354..dfc37cc 100644
--- a/vxworks/semLib.c
+++ b/vxworks/semLib.c
@@ -61,7 +61,9 @@ static STATUS xsem_take(struct wind_sem *sem, int timeout)
if (threadobj_async_p())
return S_intLib_NOT_ISR_CALLABLE;
-
+
+ check_task_restart(wind_task_current());
+
if (syncobj_lock(&sem->u.xsem.sobj, &syns))
return S_objLib_OBJ_ID_ERROR;
@@ -196,8 +198,11 @@ static STATUS msem_take(struct wind_sem *sem, int timeout)
return S_intLib_NOT_ISR_CALLABLE;
current = wind_task_current();
- if (current != NULL && (sem->options & SEM_DELETE_SAFE))
- pthread_mutex_lock(¤t->safelock);
+ if (current != NULL) {
+ check_task_restart(current);
+ if (sem->options & SEM_DELETE_SAFE)
+ pthread_mutex_lock(¤t->safelock);
+ }
if (timeout == NO_WAIT) {
ret = pthread_mutex_trylock(&sem->u.msem.lock);
diff --git a/vxworks/taskLib.c b/vxworks/taskLib.c
index a9e12e7..7e74cf0 100644
--- a/vxworks/taskLib.c
+++ b/vxworks/taskLib.c
@@ -97,6 +97,8 @@ struct wind_task *get_wind_task_or_self(TASK_ID tid)
if (current == NULL)
return NULL;
+ check_task_restart(current);
+
/* This one might block but can't fail, it is ours. */
threadobj_lock(¤t->thobj);
@@ -114,6 +116,12 @@ void put_wind_task(struct wind_task *task)
threadobj_unlock(&task->thobj);
}
+void check_task_restart(struct wind_task *current)
+{
+ if (current && (current->tcb->status & WIND_RESTART) != 0)
+ siglongjmp(current->ienv, 1);
+}
+
static void task_finalizer(struct threadobj *thobj)
{
struct wind_task *task = container_of(thobj, struct wind_task, thobj);
@@ -253,6 +261,11 @@ static void *task_trampoline(void *arg)
/* Wait for someone to run taskActivate() upon us. */
sem_wait(&task->barrier);
+ threadobj_lock(&task->thobj);
+ sigsetjmp(task->ienv, 1); /* For taskRestart(). */
+ task->tcb->status &= ~WIND_RESTART;
Re: [Xenomai-core] xenomai-solo vxWorks: know whether a task is safe or not
Niklaus Giger wrote:
> Hi
>
> For various reasons we were forced to examine under vxWorks 5.5 the private
> WIND_TCB safeCnt field to determine whether a task was "safe" or not.
>
> With the attached patch I try to add a taskIsSafe procedure to xenomai-solo.
> It is probably not free from race condition, as I do not know how to access
> a pthread_mutex.
>
> Would such (may be improved) patch find its way into the "official" trunk?
>
Assuming that VxWorks tracks the locking depth in safeCnt, that implementation
would not properly account for nested locking (safelock is a recursive mutex).
Additionally, since WIND_TCB is something of a semi-public structure in 5.x and
earlier versions, we could just track the safe counter there as well, instead of
adding a non-standard taskIsSafe() call; we already do that for the status field
anyway. Here is a possible implementation:
diff --git a/include/vxworks/taskLib.h b/include/vxworks/taskLib.h
index 510dabb..6334396 100644
--- a/include/vxworks/taskLib.h
+++ b/include/vxworks/taskLib.h
@@ -49,6 +49,7 @@ typedef struct WIND_TCB {
void *opaque;
int magic;
int status;
+ int safeCnt;
int flags;
FUNCPTR entry;
} WIND_TCB;
diff --git a/vxworks/taskLib.c b/vxworks/taskLib.c
index f6253b2..a9e12e7 100644
--- a/vxworks/taskLib.c
+++ b/vxworks/taskLib.c
@@ -283,7 +283,6 @@ static STATUS __taskInit(struct wind_task *task,
int ret;
ret = check_task_priority(prio);
-
if (ret) {
errno = ret;
return ERROR;
@@ -307,6 +306,7 @@ static STATUS __taskInit(struct wind_task *task,
tcb->magic = task_magic;
tcb->opaque = task;
tcb->status = WIND_SUSPEND;
+ tcb->safeCnt = 0;
tcb->flags = flags;
tcb->entry = entry;
@@ -588,13 +588,13 @@ STATUS taskSafe(void)
}
current = wind_task_current();
-
if (current == NULL) {
errno = S_objLib_OBJ_NO_METHOD;
return ERROR;
}
pthread_mutex_lock(¤t->safelock);
+ current->tcb->safeCnt++;
return OK;
}
@@ -602,6 +602,7 @@ STATUS taskSafe(void)
STATUS taskUnsafe(void)
{
struct wind_task *current;
+ int ret;
if (threadobj_async_p()) {
errno = S_intLib_NOT_ISR_CALLABLE;
@@ -609,13 +610,14 @@ STATUS taskUnsafe(void)
}
current = wind_task_current();
-
if (current == NULL) {
errno = S_objLib_OBJ_NO_METHOD;
return ERROR;
}
- pthread_mutex_unlock(¤t->safelock);
+ ret = pthread_mutex_unlock(¤t->safelock);
+ if (ret == 0)
+ current->tcb->safeCnt--;
return OK;
}
--
Philippe.
___
Xenomai-core mailing list
Xenomai-core@gna.org
https://mail.gna.org/listinfo/xenomai-core
