Re: [patch v12 09/13] task isolation: add preempt notifier to sync per-CPU vmstat dirty info to thread info

[Marcelo Tosatti <mtosatti@redhat.com>]

On Wed, Apr 27, 2022 at 02:09:16PM +0200, Thomas Gleixner wrote:
> On Wed, Apr 27 2022 at 09:11, Thomas Gleixner wrote:
> > On Tue, Mar 15 2022 at 12:31, Marcelo Tosatti wrote:
> >> If a thread has task isolation activated, is preempted by thread B,
> >> which marks vmstat information dirty, and is preempted back in,
> >> one might return to userspace with vmstat dirty information on the 
> >> CPU in question.
> >>
> >> To address this problem, add a preempt notifier that transfers vmstat dirty
> >> information to TIF_TASK_ISOL thread flag.
> >
> > How does this compile with CONFIG_KVM=n?
> 
> Aside of that, the existance of this preempt notifier alone tells me
> that this is either a design fail or has no design in the first place.
> 
> The state of vmstat does not matter at all at the point where a task is
> scheduled in. It matters when an isolated task goes out to user space or
> enters a VM.

If the following happens, with two threads with names that mean whether
a thread has task isolation enabled or not:

Thread-no-task-isol, Thread-task-isol.

Events:

not-runnable  		Thread-task-isol
runnable      		Thread-task-no-isol
marks vmstat dirty	Thread-task-no-isol (writes to some per-CPU vmstat
counter)
not-runnable		Thread-task-no-isol
runnable		Thread-task-isol

Then we have to transfer the "vmstat dirty" information from per-CPU 
bool to per-thread TIF_TASK_ISOL bit (so that the
task_isolation_process_work thing executes on return to userspace).

> We already have something similar in the exit to user path:
> 
>    tick_nohz_user_enter_prepare()
> 
> So you can do something like the below and have:
> 
> static inline void task_isol_exit_to_user_prepare(void)
> {
>         if (unlikely(current_needs_isol_exit_to_user())
>         	__task_isol_exit_to_user_prepare();
> }
> 
> where current_needs_isol_exit_to_user() is a simple check of either an
> existing mechanism like
> 
>          task->syscall_work & SYSCALL_WORK_TASK_ISOL_EXIT
> 
> or of some new task isolation specific member of task_struct which is
> placed so it is cache hot at that point:
> 
>         task->isol_work & SYSCALL_TASK_ISOL_EXIT
> 
> which is going to be almost zero overhead for any non isolated task.

Sure, but who sets SYSCALL_TASK_ISOL_EXIT or SYSCALL_TASK_ISOL_EXIT ?

> It's trivial enough to encode the real stuff into task->isol_work and
> I'm pretty sure, that a 32bit member is sufficient for that. There is
> absolutely no need for a potential 64x64 bit feature matrix.

Well, OK, the meaning of TIF_TASK_ISOL thread flag is ambiguous:

1) We set it when quiescing vmstat feature of task isolation.
2) We set it when switching between tasks A and B, B has 
task isolation configured and activated, and per-CPU vmstat information 
was dirty.
3) We clear it on return to userspace:

	if (test_bit(TIF_TASK_ISOL, thread->flags)) {
		clear_bit(TIF_TASK_ISOL, thread->flags))
		process_task_isol_work();
	}

So you prefer to separate:

Use TIF_TASK_ISOL for "task isolation configured and activated,
quiesce vmstat work on return to userspace" only, and then have
the "is vmstat per-CPU data dirty?" information held on 
task->syscall_work or task->isol_work ? (that will be probably be two
cachelines).

You'd still need the preempt notifier, though (unless i am missing
something).

Happy with either case.

Thanks for the review!

> Thanks,
> 
>         tglx
> ---
> --- a/kernel/entry/common.c
> +++ b/kernel/entry/common.c
> @@ -142,6 +142,12 @@ void noinstr exit_to_user_mode(void)
>  /* Workaround to allow gradual conversion of architecture code */
>  void __weak arch_do_signal_or_restart(struct pt_regs *regs) { }
>  
> +static void exit_to_user_update_work(void)
> +{
> +	tick_nohz_user_enter_prepare();
> +	task_isol_exit_to_user_prepare();
> +}
> +
>  static unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
>  					    unsigned long ti_work)
>  {
> @@ -178,8 +184,7 @@ static unsigned long exit_to_user_mode_l
>  		 */
>  		local_irq_disable_exit_to_user();
>  
> -		/* Check if any of the above work has queued a deferred wakeup */
> -		tick_nohz_user_enter_prepare();
> +		exit_to_user_update_work();
>  
>  		ti_work = read_thread_flags();
>  	}
> @@ -194,8 +199,7 @@ static void exit_to_user_mode_prepare(st
>  
>  	lockdep_assert_irqs_disabled();
>  
> -	/* Flush pending rcuog wakeup before the last need_resched() check */
> -	tick_nohz_user_enter_prepare();
> +	exit_to_user_update_work();
>  
>  	if (unlikely(ti_work & EXIT_TO_USER_MODE_WORK))
>  		ti_work = exit_to_user_mode_loop(regs, ti_work);
> 
>