[PATCH] taskstats: fix sub-threads accounting

[PATCH] taskstats: fix sub-threads accounting

[Oleg Nesterov]

If there are no listeners, taskstats_exit_send() just returns because
taskstats_exit_alloc() didn't allocate *tidstats. This is wrong, each
sub-thread should do fill_tgid_exit() on exit, otherwise its ->delays
is not recorded in ->signal->stats and lost.

Q: We don't send TASKSTATS_TYPE_AGGR_TGID when single-threaded process
exits. Is it good? How can the listener figure out that it was actually
a process exit, not sub-thread?

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>

--- STATS/kernel/taskstats.c~2_send	2006-10-30 23:47:46.000000000 +0300
+++ STATS/kernel/taskstats.c	2006-10-31 00:14:42.000000000 +0300
@@ -446,10 +446,9 @@ void taskstats_exit_send(struct task_str
 	int is_thread_group;
 	struct nlattr *na;
 
-	if (!family_registered || !tidstats)
+	if (!family_registered)
 		return;
 
-	rc = 0;
 	/*
 	 * Size includes space for nested attributes
 	 */
@@ -457,8 +456,15 @@ void taskstats_exit_send(struct task_str
 		nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 
 	is_thread_group = (tsk->signal->stats != NULL);
-	if (is_thread_group)
-		size = 2 * size;	/* PID + STATS + TGID + STATS */
+	if (is_thread_group) {
+		/* PID + STATS + TGID + STATS */
+		size = 2 * size;
+		/* fill the tsk->signal->stats structure */
+		fill_tgid_exit(tsk);
+	}
+
+	if (!tidstats)
+		return;
 
 	rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 	if (rc < 0)
@@ -478,11 +484,8 @@ void taskstats_exit_send(struct task_str
 		goto send;
 
 	/*
-	 * tsk has/had a thread group so fill the tsk->signal->stats structure
 	 * Doesn't matter if tsk is the leader or the last group member leaving
 	 */
-
-	fill_tgid_exit(tsk);
 	if (!group_dead)
 		goto send;
 

Re: [PATCH] taskstats: fix sub-threads accounting

[Shailabh Nagar]

Oleg Nesterov wrote:
> If there are no listeners, taskstats_exit_send() just returns because
> taskstats_exit_alloc() didn't allocate *tidstats. This is wrong, each
> sub-thread should do fill_tgid_exit() on exit, otherwise its ->delays
> is not recorded in ->signal->stats and lost.

Good catch. Thanks for the detailed look at the delay accounting code.



> 
> Q: We don't send TASKSTATS_TYPE_AGGR_TGID when single-threaded process
> exits. Is it good? How can the listener figure out that it was actually
> a process exit, not sub-thread?

We had a detailed discussion on this on lkml earlier. The overhead of
sending essentially the same data twice (once as AGGR_TGID and once as
PID) was deemed too heavy esp. as the taskstats structure size grew.
Also, single threaded exit is a common case.

Using process events, its possible for user space to distinguish single
threaded process exits.

> 
> Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>

Acked-by: Shailabh Nagar <nagar@watson.ibm.com>

> 
> --- STATS/kernel/taskstats.c~2_send	2006-10-30 23:47:46.000000000 +0300
> +++ STATS/kernel/taskstats.c	2006-10-31 00:14:42.000000000 +0300
> @@ -446,10 +446,9 @@ void taskstats_exit_send(struct task_str
>  	int is_thread_group;
>  	struct nlattr *na;
>  
> -	if (!family_registered || !tidstats)
> +	if (!family_registered)
>  		return;
>  
> -	rc = 0;
>  	/*
>  	 * Size includes space for nested attributes
>  	 */
> @@ -457,8 +456,15 @@ void taskstats_exit_send(struct task_str
>  		nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
>  
>  	is_thread_group = (tsk->signal->stats != NULL);
> -	if (is_thread_group)
> -		size = 2 * size;	/* PID + STATS + TGID + STATS */
> +	if (is_thread_group) {
> +		/* PID + STATS + TGID + STATS */
> +		size = 2 * size;
> +		/* fill the tsk->signal->stats structure */
> +		fill_tgid_exit(tsk);
> +	}
> +
> +	if (!tidstats)
> +		return;
>  
>  	rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
>  	if (rc < 0)
> @@ -478,11 +484,8 @@ void taskstats_exit_send(struct task_str
>  		goto send;
>  
>  	/*
> -	 * tsk has/had a thread group so fill the tsk->signal->stats structure
>  	 * Doesn't matter if tsk is the leader or the last group member leaving
>  	 */
> -
> -	fill_tgid_exit(tsk);
>  	if (!group_dead)
>  		goto send;
>  
> 

Re: [PATCH] taskstats: fix sub-threads accounting

[Oleg Nesterov]

On 10/30, Shailabh Nagar wrote:
> 
> Oleg Nesterov wrote:
> >
> >Q: We don't send TASKSTATS_TYPE_AGGR_TGID when single-threaded process
> >exits. Is it good? How can the listener figure out that it was actually
> >a process exit, not sub-thread?
> 
> We had a detailed discussion on this on lkml earlier. The overhead of
> sending essentially the same data twice (once as AGGR_TGID and once as
> PID) was deemed too heavy esp. as the taskstats structure size grew.
> Also, single threaded exit is a common case.
> 
> Using process events, its possible for user space to distinguish single
> threaded process exits.

Ok, I see.

The taskstats's code is very clean and understandable, the only thing
I can't get is: why these listeners are per-cpu? It is very easy to add
'int exited_on_this_cpu' to struct taskstats.

Probaly this was done to filter out unneeded events (cpusets) ? In that
case it seems better to add cpumask_t to 'struct listener' but have a
single listener_array list.

Thanks!

Oleg.