Re: [PATCH] rcu: Reduce synchronize_rcu() delays when all wait heads are in use

[Frederic Weisbecker]

Le Wed, Mar 13, 2024 at 09:41:58PM +0530, Neeraj Upadhyay a écrit :
> Hi Frederic,
> 
> On 3/13/2024 8:48 PM, Frederic Weisbecker wrote:
> > Le Wed, Mar 13, 2024 at 02:02:28PM +0530, Neeraj Upadhyay a écrit :
> >> When all wait heads are in use, which can happen when
> >> rcu_sr_normal_gp_cleanup_work()'s callback processing
> >> is slow, any new synchronize_rcu() user's rcu_synchronize
> >> node's processing is deferred to future GP periods. This
> >> can result in long list of synchronize_rcu() invocations
> >> waiting for full grace period processing, which can delay
> >> freeing of memory. Mitigate this problem by using first
> >> node in the list as wait tail when all wait heads are in use.
> >> While methods to speed up callback processing would be needed
> >> to recover from this situation, allowing new nodes to complete
> >> their grace period can help prevent delays due to a fixed
> >> number of wait head nodes.
> >>
> >> Signed-off-by: Neeraj Upadhyay <Neeraj.Upadhyay@xxxxxxx>
> >> ---
> >>  kernel/rcu/tree.c | 27 +++++++++++++--------------
> >>  1 file changed, 13 insertions(+), 14 deletions(-)
> >>
> >> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> >> index 9fbb5ab57c84..bdccce1ed62f 100644
> >> --- a/kernel/rcu/tree.c
> >> +++ b/kernel/rcu/tree.c
> >> @@ -1470,14 +1470,11 @@ static void rcu_poll_gp_seq_end_unlocked(unsigned long *snap)
> >>   * for this new grace period. Given that there are a fixed
> >>   * number of wait nodes, if all wait nodes are in use
> >>   * (which can happen when kworker callback processing
> >> - * is delayed) and additional grace period is requested.
> >> - * This means, a system is slow in processing callbacks.
> >> - *
> >> - * TODO: If a slow processing is detected, a first node
> >> - * in the llist should be used as a wait-tail for this
> >> - * grace period, therefore users which should wait due
> >> - * to a slow process are handled by _this_ grace period
> >> - * and not next.
> >> + * is delayed), first node in the llist is used as wait
> >> + * tail for this grace period. This means, the first node
> >> + * has to go through additional grace periods before it is
> >> + * part of the wait callbacks. This should be ok, as
> >> + * the system is slow in processing callbacks anyway.
> >>   *
> >>   * Below is an illustration of how the done and wait
> >>   * tail pointers move from one set of rcu_synchronize nodes
> >> @@ -1725,15 +1722,17 @@ static bool rcu_sr_normal_gp_init(void)
> >>  		return start_new_poll;
> >>  
> >>  	wait_head = rcu_sr_get_wait_head();
> >> -	if (!wait_head) {
> >> -		// Kick another GP to retry.
> >> +	if (wait_head) {
> >> +		/* Inject a wait-dummy-node. */
> >> +		llist_add(wait_head, &rcu_state.srs_next);
> >> +	} else {
> >> +		// Kick another GP for first node.
> >>  		start_new_poll = true;
> >> -		return start_new_poll;
> >> +		if (first == rcu_state.srs_done_tail)
> >> +			return start_new_poll;
> >> +		wait_head = first;
> > 
> > This means you're setting a non-wait-head as srs_wait_tail, right?
> > Doesn't it trigger the following warning in rcu_sr_normal_gp_cleanup():
> > 
> > 	WARN_ON_ONCE(!rcu_sr_is_wait_head(wait_tail));
> > 
> 
> Oh I missed it. Will fix it, thanks!
> 
> > Also there is a risk that this non-wait-head gets later assigned as
> > rcu_state.srs_done_tail. And then this pending sr may not be completed
> > until the next grace period calling rcu_sr_normal_gp_cleanup()? (Because
> > the work doesn't take care of rcu_state.srs_done_tail itself). And then
> > the delay can be arbitrary.
> > 
> 
> That is correct. Only the first node suffers from deferred GP.
> If there are large number of callbacks which got added after
> last available wait head was queued, all those callbacks (except one)
> can still have a GP assigned to them.
> 
> > And the next grace period completing this sr (that non-wait-head set
> > as rcu_state.srs_done_tail) and thus allowing its caller to wipe it out
> > of its stack may race with the cleanup work dereferencing it?
> > 
> 
> This sr can only be completed when done tail moves to new node. Till
> then, it gets deferred continuously. So, we won't be entering into
> the situation where the sr processing is complete while done tail is pointing
> to it. Please correct me if I am missing something here.

Ok I'm confused as usual. Let's take a practical case. Is the following
sequence possible?

1) wait_tail = NULL
   done_tail = WH4->SR4->WH3->SR3->WH2->SR2->WH1->SR1...

Initial layout

2) wait_tail = SR5 -> WH4...
   done_tail = WH4->SR4->WH3->SR3->WH2->SR2->WH1->SR1...

New GP

3) wait_tail = NULL
   done_tail = SR5->WH4->SR4->WH3->SR3->WH2->SR2->WH1->SR1...

GP completes, normal cleanup

3) wait_tail = SR6->SR5...
   done_tail = SR5->WH4->SR4->WH3->SR2->WH2->SR1->WH1->SR1...

New GP

4) GP completes and SR5 is completed by rcu_sr_normal_gp_cleanup(). So
   the caller releases it from the stack. But before rcu_sr_normal_gp_cleanup()
   overwrites done_tail to SR6->WH4->SR4.... , the workqueue manages to run
   and concurrently dereferences SR5.

But I bet I'm missing something obvious in the middle, preventing that...