Re: [PATCH 04/25] vtime: Spare a seqcount lock/unlock cycle on context switch

[Peter Zijlstra]

On Wed, Nov 14, 2018 at 03:45:48AM +0100, Frederic Weisbecker wrote:

So I definitely like avoiding that superfluous atomic op, however:

> @@ -730,19 +728,25 @@ static void vtime_account_guest(struct task_struct *tsk,
>  	}
>  }
>  
> +static void __vtime_account_kernel(struct task_struct *tsk,
> +				   struct vtime *vtime)

Your last patch removed a __function, and now you're adding one back :/

>  {
>  	/* We might have scheduled out from guest path */
>  	if (tsk->flags & PF_VCPU)
>  		vtime_account_guest(tsk, vtime);
>  	else
>  		vtime_account_system(tsk, vtime);
> +}
> +
> +void vtime_account_kernel(struct task_struct *tsk)
> +{
> +	struct vtime *vtime = &tsk->vtime;
> +
> +	if (!vtime_delta(vtime))
> +		return;
> +

See here the fast path (is it worth it?)

> +	write_seqcount_begin(&vtime->seqcount);
> +	__vtime_account_kernel(tsk, vtime);
>  	write_seqcount_end(&vtime->seqcount);
>  }

> +void vtime_task_switch_generic(struct task_struct *prev)
>  {
>  	struct vtime *vtime = &prev->vtime;

And observe a distinct lack of that same fast path..

>  
>  	write_seqcount_begin(&vtime->seqcount);
> +	if (is_idle_task(prev))
> +		vtime_account_idle(prev);
> +	else
> +		__vtime_account_kernel(prev, vtime);
>  	vtime->state = VTIME_INACTIVE;
>  	write_seqcount_end(&vtime->seqcount);
>  
> -- 
> 2.7.4
>