Re: [PATCH 1/2] exit: add a tracepoint for profiling a task that is starting to exit

[Mathieu Desnoyers]

On 2024-02-22 11:04, wenyang.linux@xxxxxxxxxxx wrote:
> From: Wen Yang <wenyang.linux@xxxxxxxxxxx>
>
> Currently coredump_task_exit() takes some time to wait for the generation
> of the dump file. But if the user-space wants to receive a notification
> as soon as possible it maybe inconvenient.
>
> Commit 2d4bcf886e42 ("exit: Remove profile_task_exit & profile_munmap")
> simplified the code, but also removed profile_task_exit(), which may
> prevent third-party kernel modules from detecting process exits timely.
>
> Add the new trace_sched_profile_task_exit() this way a user-space monitor
> could detect the exits and potentially make some preparations in advance.

I don't see any explanation justifying adding an extra tracepoint
rather than just moving trace_sched_process_exit() earlier in do_exit().

Why is moving trace_sched_process_exit() earlier in do_exit() an issue,
considering that any tracer interested in knowing the point where a task
is really reclaimed (from zombie state) is trace_sched_process_free()
called from delayed_put_task_struct() ?

Thanks,

Mathieu

> Signed-off-by: Wen Yang <wenyang.linux@xxxxxxxxxxx>
> Cc: Oleg Nesterov <oleg@xxxxxxxxxx>
> Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
> Cc: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> Cc: Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> Cc: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> ---
>   include/trace/events/sched.h | 28 ++++++++++++++++++++++++++++
>   kernel/exit.c                |  1 +
>   2 files changed, 29 insertions(+)
>
> diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
> index dbb01b4b7451..750b2f0bdf69 100644
> --- a/include/trace/events/sched.h
> +++ b/include/trace/events/sched.h
> @@ -341,6 +341,34 @@ DEFINE_EVENT(sched_process_template, sched_wait_task,
>   	TP_PROTO(struct task_struct *p),
>   	TP_ARGS(p));
>   
> +/*
> + * Tracepoint for profiling a task that is starting to exit:
> + */
> +TRACE_EVENT(sched_profile_task_exit,
> +
> +	TP_PROTO(struct task_struct *task, long code),
> +
> +	TP_ARGS(task, code),
> +
> +	TP_STRUCT__entry(
> +		__array(	char,	comm,	TASK_COMM_LEN	)
> +		__field(	pid_t,	pid			)
> +		__field(	int,	prio			)
> +		__field(	long,	code			)
> +	),
> +
> +	TP_fast_assign(
> +		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
> +		__entry->pid		= task->pid;
> +		__entry->prio		= task->prio;
> +		__entry->code		= code;
> +	),
> +
> +	TP_printk("comm=%s pid=%d prio=%d exit_code=0x%lx",
> +		  __entry->comm, __entry->pid, __entry->prio,
> +		  __entry->code)
> +);
> +
>   /*
>    * Tracepoint for a waiting task:
>    */
> diff --git a/kernel/exit.c b/kernel/exit.c
> index 493647fd7c07..f675f879a1b2 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -826,6 +826,7 @@ void __noreturn do_exit(long code)
>   
>   	WARN_ON(tsk->plug);
>   
> +	trace_sched_profile_task_exit(tsk, code);
>   	kcov_task_exit(tsk);
>   	kmsan_task_exit(tsk);
>   

--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com