Re: [PATCH 1/3] taskstats: separate PID/TGID stats producers to complete the TGID ones

[Balbir Singh <balbir@linux.vnet.ibm.com>]

Guillaume Chazarain wrote:
> TASKSTATS_CMD_ATTR_TGID used to return only the delay accounting stats, not
> the basic and extended accounting.  With this patch,
> TASKSTATS_CMD_ATTR_TGID also aggregates the accounting info for all threads
> of a thread group.
> 
> TASKSTATS_CMD_ATTR_PID output should be unchanged
> TASKSTATS_CMD_ATTR_TGID output should have all fields set, unlike before the
> patch where most of the fiels were set to 0.
> 
> To this aim, two functions were introduced: fill_threadgroup() and add_tsk().
> These functions are responsible for aggregating the subsystem specific
> accounting information. Taskstats requesters (fill_pid(), fill_tgid() and
> fill_tgid_exit()) should only call add_tsk() and fill_threadgroup() to get the
> stats.
> 
> Signed-off-by: Guillaume Chazarain <guichaz@yahoo.fr>
> Cc: Balbir Singh <balbir@in.ibm.com>
> Cc: Jay Lan <jlan@engr.sgi.com>
> Cc: Jonathan Lim <jlim@sgi.com>
> Cc: Oleg Nesterov <oleg@tv-sign.ru>
> ---
> 
>  Documentation/accounting/getdelays.c |    2 -
>  include/linux/tsacct_kern.h          |   12 ++-
>  kernel/taskstats.c                   |  131 ++++++++++++++++++++++------------
>  kernel/tsacct.c                      |  106 ++++++++++++++++------------
>  4 files changed, 155 insertions(+), 96 deletions(-)
> 
> diff --git a/Documentation/accounting/getdelays.c b/Documentation/accounting/getdelays.c
> index cbee3a2..78773c0 100644
> --- a/Documentation/accounting/getdelays.c
> +++ b/Documentation/accounting/getdelays.c
> @@ -76,7 +76,7 @@ static void usage(void)
>  	fprintf(stderr, "getdelays [-dilv] [-w logfile] [-r bufsize] "
>  			"[-m cpumask] [-t tgid] [-p pid]\n");
>  	fprintf(stderr, "  -d: print delayacct stats\n");
> -	fprintf(stderr, "  -i: print IO accounting (works only with -p)\n");
> +	fprintf(stderr, "  -i: print IO accounting\n");
>  	fprintf(stderr, "  -l: listen forever\n");
>  	fprintf(stderr, "  -v: debug on\n");
>  }
> diff --git a/include/linux/tsacct_kern.h b/include/linux/tsacct_kern.h
> index 7e50ac7..93dffc2 100644
> --- a/include/linux/tsacct_kern.h
> +++ b/include/linux/tsacct_kern.h
> @@ -10,17 +10,23 @@
>  #include <linux/taskstats.h>
> 
>  #ifdef CONFIG_TASKSTATS
> -extern void bacct_add_tsk(struct taskstats *stats, struct task_struct *tsk);
> +void bacct_fill_threadgroup(struct taskstats *stats, struct task_struct *task);
> +void bacct_add_tsk(struct taskstats *stats, struct task_struct *task);
>  #else
> -static inline void bacct_add_tsk(struct taskstats *stats, struct task_struct *tsk)
> +static inline void bacct_fill_threadgroup(struct taskstats *stats, struct task_struct *task)
> +{}
> +static inline void bacct_add_tsk(struct taskstats *stats, struct task_struct *task)
>  {}
>  #endif /* CONFIG_TASKSTATS */
> 
>  #ifdef CONFIG_TASK_XACCT
> -extern void xacct_add_tsk(struct taskstats *stats, struct task_struct *p);
> +void xacct_fill_threadgroup(struct taskstats *stats, struct task_struct *task);
> +void xacct_add_tsk(struct taskstats *stats, struct task_struct *p);
>  extern void acct_update_integrals(struct task_struct *tsk);
>  extern void acct_clear_integrals(struct task_struct *tsk);
>  #else
> +static inline void xacct_fill_threadgroup(struct taskstats *stats, struct task_struct *task)
> +{}
>  static inline void xacct_add_tsk(struct taskstats *stats, struct task_struct *p)
>  {}
>  static inline void acct_update_integrals(struct task_struct *tsk)
> diff --git a/kernel/taskstats.c b/kernel/taskstats.c
> index 059431e..ce43fae 100644
> --- a/kernel/taskstats.c
> +++ b/kernel/taskstats.c
> @@ -168,6 +168,68 @@ static void send_cpu_listeners(struct sk_buff *skb,
>  	up_write(&listeners->sem);
>  }
> 
> +/**
> + * fill_threadgroup - initialize some common stats for the thread group
> + * @stats: the taskstats to write into
> + * @task: the thread representing the whole group
> + *
> + * There are two types of taskstats fields when considering a thread group:
> + *	- those that can be aggregated from each thread in the group (like CPU
> + *	times),
> + *	- those that cannot be aggregated (like UID) or are identical (like
> + *	memory usage), so are taken from the group leader.
> + * XXX_threadgroup() methods deal with the first type while XXX_add_tsk() with
> + * the second.
> + */
> +static void fill_threadgroup(struct taskstats *stats, struct task_struct *task)
> +{

How about calling this one fill_threadgroup_stats()?

> +	/*
> +	 * Each accounting subsystem adds calls to its functions to initialize
> +	 * relevant parts of struct taskstsats for a single tgid as follows:
> +	 *
> +	 *	per-task-foo-fill_threadgroup(stats, task);
> +	 */
> +
> +	stats->version = TASKSTATS_VERSION;
> +
> +	/* fill in basic acct fields */
> +	bacct_fill_threadgroup(stats, task);
> +
> +	/* fill in extended acct fields */
> +	xacct_fill_threadgroup(stats, task);
> +}
> +
> +/**
> + * add_tsk - combine some thread specific stats in a taskstats
> + * @stats: the taskstats to write into
> + * @task: the thread to combine
> + *
> + * Stats specific to each thread in the thread group. Stats of @task should be
> + * combined with those already present in @stats. add_tsk() works in
> + * conjunction with fill_threadgroup(), taskstats fields should not be touched
> + * by both functions.
> + */
> +static void add_tsk(struct taskstats *stats, struct task_struct *task)
> +{

How about we call function add_tsk_stats()?

> +	/*
> +	 * Each accounting subsystem adds calls to its functions to combine
> +	 * relevant parts of struct taskstsats for a single pid as follows:
> +	 *
> +	 *	per-task-foo-add_tsk(stats, task);
> +	 */
> +	stats->nvcsw  += task->nvcsw;
> +	stats->nivcsw += task->nivcsw;
> +
> +	/* fill in delay acct fields */
> +	delayacct_add_tsk(stats, task);
> +
> +	/* fill in basic acct fields */
> +	bacct_add_tsk(stats, task);
> +
> +	/* fill in extended acct fields */
> +	xacct_add_tsk(stats, task);
> +}
> +
>  static int fill_pid(pid_t pid, struct task_struct *tsk,
>  		struct taskstats *stats)
>  {
> @@ -185,23 +247,8 @@ static int fill_pid(pid_t pid, struct task_struct *tsk,
>  		get_task_struct(tsk);
> 
>  	memset(stats, 0, sizeof(*stats));
> -	/*
> -	 * Each accounting subsystem adds calls to its functions to
> -	 * fill in relevant parts of struct taskstsats as follows
> -	 *
> -	 *	per-task-foo(stats, tsk);
> -	 */
> -
> -	delayacct_add_tsk(stats, tsk);
> -
> -	/* fill in basic acct fields */
> -	stats->version = TASKSTATS_VERSION;
> -	stats->nvcsw = tsk->nvcsw;
> -	stats->nivcsw = tsk->nivcsw;
> -	bacct_add_tsk(stats, tsk);
> -
> -	/* fill in extended acct fields */
> -	xacct_add_tsk(stats, tsk);
> +	add_tsk(stats, tsk);
> +	fill_threadgroup(stats, tsk);
> 

So we always call fill_threadgroup later, is there a reason for
that. Can the order of calls be arbitrary or is there a dependency?

>  	/* Define err: label here if needed */
>  	put_task_struct(tsk);
> @@ -233,31 +280,20 @@ static int fill_tgid(pid_t tgid, struct task_struct *first,
>  		memset(stats, 0, sizeof(*stats));
> 
>  	tsk = first;
> -	do {
> -		if (tsk->exit_state)
> -			continue;
> -		/*
> -		 * Accounting subsystem can call its functions here to
> -		 * fill in relevant parts of struct taskstsats as follows
> -		 *
> -		 *	per-task-foo(stats, tsk);
> -		 */
> -		delayacct_add_tsk(stats, tsk);
> -
> -		stats->nvcsw += tsk->nvcsw;
> -		stats->nivcsw += tsk->nivcsw;
> -	} while_each_thread(first, tsk);
> -
> +	do
> +		if (!tsk->exit_state)
> +			/*
> +			 * This check is racy as a thread could exit just right
> +			 * now and have its statistics accounted twice.
> +			 */
> +			add_tsk(stats, tsk);
> +	while_each_thread(first, tsk);
> +

I still prefer braces around do <--> while, I think the code is easier
to read with them.

> +	fill_threadgroup(stats, first->group_leader);
>  	unlock_task_sighand(first, &flags);
>  	rc = 0;
>  out:
>  	rcu_read_unlock();
> -
> -	stats->version = TASKSTATS_VERSION;
> -	/*
> -	 * Accounting subsytems can also add calls here to modify
> -	 * fields of taskstats.
> -	 */
>  	return rc;
>  }
> 
> @@ -267,19 +303,14 @@ static void fill_tgid_exit(struct task_struct *tsk)
>  	unsigned long flags;
> 
>  	spin_lock_irqsave(&tsk->sighand->siglock, flags);
> -	if (!tsk->signal->stats)
> -		goto ret;
> 
>  	/*
> -	 * Each accounting subsystem calls its functions here to
> -	 * accumalate its per-task stats for tsk, into the per-tgid structure
> -	 *
> -	 *	per-task-foo(tsk->signal->stats, tsk);
> +	 * The fill_threadgroup() part of the statistics will be added by the
> +	 * stats requester, i.e. fill_tgid() or taskstats_exit()
>  	 */
> -	delayacct_add_tsk(tsk->signal->stats, tsk);
> -ret:
> +	add_tsk(tsk->signal->stats, tsk);
> +
>  	spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
> -	return;
>  }
> 
>  static int add_del_listener(pid_t pid, cpumask_t *maskp, int isadd)
> @@ -508,7 +539,13 @@ void taskstats_exit(struct task_struct *tsk, int group_dead)
>  	if (!stats)
>  		goto err;
> 
> +	/*
> +	 * Race here: the last thread may have not finished its
> +	 * fill_tgid_exit(), leaving an incomplete tsk->signal->stats
> +	 */
> +
>  	memcpy(stats, tsk->signal->stats, sizeof(*stats));
> +	fill_threadgroup(stats, tsk->group_leader);
> 
>  send:
>  	send_cpu_listeners(rep_skb, listeners);
> diff --git a/kernel/tsacct.c b/kernel/tsacct.c
> index 4ab1b58..9541a1a 100644
> --- a/kernel/tsacct.c
> +++ b/kernel/tsacct.c
> @@ -22,54 +22,68 @@
>  #include <linux/acct.h>
>  #include <linux/jiffies.h>
> 
> -/*
> - * fill in basic accounting fields
> - */


Could we further split the tsacct and delayacct/taskstats patches.

> -void bacct_add_tsk(struct taskstats *stats, struct task_struct *tsk)
> +static void fill_wall_times(struct taskstats *stats, struct task_struct *tsk)
>  {
>  	struct timespec uptime, ts;
>  	s64 ac_etime;
> 
> -	BUILD_BUG_ON(TS_COMM_LEN < TASK_COMM_LEN);
> -
>  	/* calculate task elapsed time in timespec */
>  	do_posix_clock_monotonic_gettime(&uptime);
>  	ts = timespec_sub(uptime, tsk->start_time);
>  	/* rebase elapsed time to usec */
>  	ac_etime = timespec_to_ns(&ts);
>  	do_div(ac_etime, NSEC_PER_USEC);
> -	stats->ac_etime = ac_etime;
> -	stats->ac_btime = get_seconds() - ts.tv_sec;
> -	if (thread_group_leader(tsk)) {
> -		stats->ac_exitcode = tsk->exit_code;
> -		if (tsk->flags & PF_FORKNOEXEC)
> +	stats->ac_etime	= ac_etime;
> +	stats->ac_btime	= get_seconds() - ts.tv_sec;
> +}
> +
> +/*
> + * fill in basic accounting fields
> + */
> +


-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL