Re: [PATCH 3/3] IO_URING: Statistics of the true utilization of sq threads.

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Sep 28, 2023 at 10:22:28AM +0800, Xiaobing Li wrote:
> Since the sq thread has a while(1) structure, during this process, there
> may be a lot of time that is not processing IO but does not exceed the
> timeout period, therefore, the sqpoll thread will keep running and will
> keep occupying the CPU. Obviously, the CPU is wasted at this time;Our
> goal is to count the part of the time that the sqpoll thread actually
> processes IO, so as to reflect the part of the CPU it uses to process
> IO, which can be used to help improve the actual utilization of the CPU
> in the future.
> 
> Signed-off-by: Xiaobing Li <xiaobing.li@samsung.com>
> ---
>  io_uring/sqpoll.c | 26 +++++++++++++++++++++++++-
>  1 file changed, 25 insertions(+), 1 deletion(-)
> 
> diff --git a/io_uring/sqpoll.c b/io_uring/sqpoll.c
> index bd6c2c7959a5..2c5fc4d95fa8 100644
> --- a/io_uring/sqpoll.c
> +++ b/io_uring/sqpoll.c
> @@ -11,6 +11,7 @@
>  #include <linux/audit.h>
>  #include <linux/security.h>
>  #include <linux/io_uring.h>
> +#include <linux/time.h>
>  
>  #include <uapi/linux/io_uring.h>
>  
> @@ -235,6 +236,10 @@ static int io_sq_thread(void *data)
>  		set_cpus_allowed_ptr(current, cpu_online_mask);
>  
>  	mutex_lock(&sqd->lock);
> +	bool first = true;
> +	struct timespec64 ts_start, ts_end;
> +	struct timespec64 ts_delta;
> +	struct sched_entity *se = &sqd->thread->se;
>  	while (1) {
>  		bool cap_entries, sqt_spin = false;
>  
> @@ -243,7 +248,16 @@ static int io_sq_thread(void *data)
>  				break;
>  			timeout = jiffies + sqd->sq_thread_idle;
>  		}
> -
> +		ktime_get_boottime_ts64(&ts_start);
> +		ts_delta = timespec64_sub(ts_start, ts_end);
> +		unsigned long long now = ts_delta.tv_sec * NSEC_PER_SEC + ts_delta.tv_nsec +
> +		se->sq_avg.last_update_time;
> +
> +		if (first) {
> +			now = 0;
> +			first = false;
> +		}
> +		__update_sq_avg_block(now, se);
>  		cap_entries = !list_is_singular(&sqd->ctx_list);
>  		list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
>  			int ret = __io_sq_thread(ctx, cap_entries);
> @@ -251,6 +265,16 @@ static int io_sq_thread(void *data)
>  			if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list)))
>  				sqt_spin = true;
>  		}
> +
> +		ktime_get_boottime_ts64(&ts_end);
> +		ts_delta = timespec64_sub(ts_end, ts_start);
> +		now = ts_delta.tv_sec * NSEC_PER_SEC + ts_delta.tv_nsec +
> +		se->sq_avg.last_update_time;
> +
> +		if (sqt_spin)
> +			__update_sq_avg(now, se);
> +		else
> +			__update_sq_avg_block(now, se);
>  		if (io_run_task_work())
>  			sqt_spin = true;
>  

All of this is quite insane, but the above is actually broken. You're
using wall-time to measure runtime of a preemptible thread.

On top of that, for extra insanity, you're using the frigging insane
timespec interface, because clearly the _ns() interfaces are too
complicated or something?

And that whole first thing is daft too, pull now out of the loop and
set it before, then all that goes away.

Now, I see what you're trying to do, but who actually uses this data?

Finally, please don't scream in the subject :/