Re: [patch] preempt-trace.patch (mono preempt-trace)

[Ingo Molnar]

* Dominik Karall <dominik.karall@xxxxxxx> wrote:

> With FRAME_POINTERS enabled:
> 
> BUG: mono[3193] exited with nonzero preempt_count 1!
> ---------------------------
> | preempt count: 00000001 ]
> | 1 level deep critical section nesting:
> ----------------------------------------
> .. [<ffffffff80400a46>] .... _spin_lock+0x16/0x80
> .....[<ffffffff801ed30c>] ..   ( <= sys_semtimedop+0x28c/0x7c0)

thanks. It seems semundo->lock somehow leaked. One possibility would be 
of semundo->refcount going from 2 to 1 while another thread has it 
locked. I dont see what prevents this scenario from happening. To test 
this theory, could you apply the patch below, which will do semundo 
locking not conditional on the refcount - does it fix the bug?

	Ingo

 ipc/sem.c |   10 +++-------
 1 files changed, 3 insertions(+), 7 deletions(-)

Index: linux-preempt-trace/ipc/sem.c
===================================================================
--- linux-preempt-trace.orig/ipc/sem.c
+++ linux-preempt-trace/ipc/sem.c
@@ -895,7 +895,7 @@ static inline void lock_semundo(void)
 	struct sem_undo_list *undo_list;
 
 	undo_list = current->sysvsem.undo_list;
-	if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
+	if (undo_list)
 		spin_lock(&undo_list->lock);
 }
 
@@ -915,7 +915,7 @@ static inline void unlock_semundo(void)
 	struct sem_undo_list *undo_list;
 
 	undo_list = current->sysvsem.undo_list;
-	if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
+	if (undo_list)
 		spin_unlock(&undo_list->lock);
 }
 
@@ -943,9 +943,7 @@ static inline int get_undo_list(struct s
 		if (undo_list == NULL)
 			return -ENOMEM;
 		memset(undo_list, 0, size);
-		/* don't initialize unodhd->lock here.  It's done
-		 * in copy_semundo() instead.
-		 */
+		spin_lock_init(&undo_list->lock);
 		atomic_set(&undo_list->refcnt, 1);
 		current->sysvsem.undo_list = undo_list;
 	}
@@ -1231,8 +1229,6 @@ int copy_semundo(unsigned long clone_fla
 		error = get_undo_list(&undo_list);
 		if (error)
 			return error;
-		if (atomic_read(&undo_list->refcnt) == 1)
-			spin_lock_init(&undo_list->lock);
 		atomic_inc(&undo_list->refcnt);
 		tsk->sysvsem.undo_list = undo_list;
 	} else 

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/