Re: [PATCH 1/6] rcu/nocb: Remove rdp from nocb list when de-offloaded

From: Paul E. McKenney
Date: Wed Nov 17 2021 - 13:53:46 EST


On Wed, Nov 17, 2021 at 04:56:32PM +0100, Frederic Weisbecker wrote:
> nocb_gp_wait() iterates all CPUs within the rcuog's group even if they
> are have been de-offloaded. This is suboptimal if only few CPUs are
> offloaded within the group. And this will become even more a problem
> when a nocb kthread will be created for all possible CPUs in the future.
>
> Therefore use a standard double linked list to link all the offloaded
> rdps and safely add/del their nodes as we (de-)offloaded them.
>
> Signed-off-by: Frederic Weisbecker <frederic@xxxxxxxxxx>
> Cc: Neeraj Upadhyay <quic_neeraju@xxxxxxxxxxx>
> Cc: Boqun Feng <boqun.feng@xxxxxxxxx>
> Cc: Uladzislau Rezki <urezki@xxxxxxxxx>
> Cc: Josh Triplett <josh@xxxxxxxxxxxxxxxx>
> Cc: Joel Fernandes <joel@xxxxxxxxxxxxxxxxx>
> ---
> kernel/rcu/tree.h | 7 +++++--
> kernel/rcu/tree_nocb.h | 29 ++++++++++++++++++++++-------
> 2 files changed, 27 insertions(+), 9 deletions(-)
>
> diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
> index deeaf2fee714..486fc901bd08 100644
> --- a/kernel/rcu/tree.h
> +++ b/kernel/rcu/tree.h
> @@ -221,8 +221,11 @@ struct rcu_data {
> struct swait_queue_head nocb_gp_wq; /* For nocb kthreads to sleep on. */
> bool nocb_cb_sleep; /* Is the nocb CB thread asleep? */
> struct task_struct *nocb_cb_kthread;
> - struct rcu_data *nocb_next_cb_rdp;
> - /* Next rcu_data in wakeup chain. */
> + struct list_head nocb_head_rdp; /*
> + * Head of rcu_data list in wakeup chain,
> + * if rdp_gp.
> + */
> + struct list_head nocb_entry_rdp; /* rcu_data node in wakeup chain. */
>
> /* The following fields are used by CB kthread, hence new cacheline. */
> struct rcu_data *nocb_gp_rdp ____cacheline_internodealigned_in_smp;
> diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
> index 2461fe8d0c23..e728128be02a 100644
> --- a/kernel/rcu/tree_nocb.h
> +++ b/kernel/rcu/tree_nocb.h
> @@ -625,7 +625,7 @@ static void nocb_gp_wait(struct rcu_data *my_rdp)
> * and the global grace-period kthread are awakened if needed.
> */
> WARN_ON_ONCE(my_rdp->nocb_gp_rdp != my_rdp);
> - for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_cb_rdp) {
> + list_for_each_entry_rcu(rdp, &my_rdp->nocb_head_rdp, nocb_entry_rdp, 1) {
> bool needwake_state = false;
>
> if (!nocb_gp_enabled_cb(rdp))
> @@ -1003,6 +1003,8 @@ static long rcu_nocb_rdp_deoffload(void *arg)
> swait_event_exclusive(rdp->nocb_state_wq,
> !rcu_segcblist_test_flags(cblist, SEGCBLIST_KTHREAD_CB |
> SEGCBLIST_KTHREAD_GP));
> + /* Don't bother iterate this one anymore on nocb_gp_wait() */
> + list_del_rcu(&rdp->nocb_entry_rdp);
> /*
> * Lock one last time to acquire latest callback updates from kthreads
> * so we can later handle callbacks locally without locking.
> @@ -1066,6 +1068,15 @@ static long rcu_nocb_rdp_offload(void *arg)
> return -EINVAL;
>
> pr_info("Offloading %d\n", rdp->cpu);
> +
> + /*
> + * Iterate this CPU on nocb_gp_wait(). We do it before locking nocb_gp_lock,
> + * resetting nocb_gp_sleep and waking up the related "rcuog". Since nocb_gp_wait()
> + * in turn locks nocb_gp_lock before setting nocb_gp_sleep again, we are guaranteed
> + * to iterate this new rdp before "rcuog" goes to sleep again.

Just to make sure that I understand...

The ->nocb_entry_rdp list is RCU-protected, with an odd flavor of RCU.
The list_add_tail_rcu() handles list insertion. For list deletion,
on the one hand, the rcu_data structures are never freed, so their
lifetime never ends. But one could be removed during an de-offloading
operation, then re-added by a later offloading operation. It would be
bad if a reader came along for that ride because that reader would end
up skipping all the rcu_data structures that were in the list after the
initial position of the one that was removed and added back.

The trick seems to be that the de-offloading process cannot complete
until the relevant rcuog kthread has acknowledged the de-offloading,
which it cannot do until it has completed the list_for_each_entry_rcu()
loop. And the rcuog kthread is the only thing that traverses the list,
except for the show_rcu_nocb_state() function, more on which later.

Therefore, it is not possible for the rcuog kthread to come along for
that ride.

On to show_rcu_nocb_state()...

This does not actually traverse the list, but instead looks at the ->cpu
field of the next structure. Because the ->next pointer is never NULLed,
the worst that can happen is that a confusing ->cpu field is printed,
for example, the one that was in effect prior to the de-offloading
operation. But that number would have been printed anyway had the
show_rcu_nocb_state() function not been delayed across the de-offloading
and offloading.

So no harm done.

Did I get it right? If so, the comment might need help. If not,
what am I missing?

Thanx, Paul

> + */
> + list_add_tail_rcu(&rdp->nocb_entry_rdp, &rdp->nocb_gp_rdp->nocb_head_rdp);
> +
> /*
> * Can't use rcu_nocb_lock_irqsave() before SEGCBLIST_LOCKING
> * is set.
> @@ -1268,7 +1279,6 @@ static void __init rcu_organize_nocb_kthreads(void)
> int nl = 0; /* Next GP kthread. */
> struct rcu_data *rdp;
> struct rcu_data *rdp_gp = NULL; /* Suppress misguided gcc warn. */
> - struct rcu_data *rdp_prev = NULL;
>
> if (!cpumask_available(rcu_nocb_mask))
> return;
> @@ -1288,8 +1298,8 @@ static void __init rcu_organize_nocb_kthreads(void)
> /* New GP kthread, set up for CBs & next GP. */
> gotnocbs = true;
> nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls;
> - rdp->nocb_gp_rdp = rdp;
> rdp_gp = rdp;
> + INIT_LIST_HEAD(&rdp->nocb_head_rdp);
> if (dump_tree) {
> if (!firsttime)
> pr_cont("%s\n", gotnocbscbs
> @@ -1302,12 +1312,11 @@ static void __init rcu_organize_nocb_kthreads(void)
> } else {
> /* Another CB kthread, link to previous GP kthread. */
> gotnocbscbs = true;
> - rdp->nocb_gp_rdp = rdp_gp;
> - rdp_prev->nocb_next_cb_rdp = rdp;
> if (dump_tree)
> pr_cont(" %d", cpu);
> }
> - rdp_prev = rdp;
> + rdp->nocb_gp_rdp = rdp_gp;
> + list_add_tail(&rdp->nocb_entry_rdp, &rdp_gp->nocb_head_rdp);
> }
> if (gotnocbs && dump_tree)
> pr_cont("%s\n", gotnocbscbs ? "" : " (self only)");
> @@ -1369,6 +1378,7 @@ static void show_rcu_nocb_state(struct rcu_data *rdp)
> {
> char bufw[20];
> char bufr[20];
> + struct rcu_data *nocb_next_rdp = NULL;
> struct rcu_segcblist *rsclp = &rdp->cblist;
> bool waslocked;
> bool wassleep;
> @@ -1376,11 +1386,16 @@ static void show_rcu_nocb_state(struct rcu_data *rdp)
> if (rdp->nocb_gp_rdp == rdp)
> show_rcu_nocb_gp_state(rdp);
>
> + nocb_next_rdp = list_next_or_null_rcu(&rdp->nocb_gp_rdp->nocb_head_rdp,
> + &rdp->nocb_entry_rdp,
> + typeof(*rdp),
> + nocb_entry_rdp);
> +
> sprintf(bufw, "%ld", rsclp->gp_seq[RCU_WAIT_TAIL]);
> sprintf(bufr, "%ld", rsclp->gp_seq[RCU_NEXT_READY_TAIL]);
> pr_info(" CB %d^%d->%d %c%c%c%c%c%c F%ld L%ld C%d %c%c%s%c%s%c%c q%ld %c CPU %d%s\n",
> rdp->cpu, rdp->nocb_gp_rdp->cpu,
> - rdp->nocb_next_cb_rdp ? rdp->nocb_next_cb_rdp->cpu : -1,
> + nocb_next_rdp ? nocb_next_rdp->cpu : -1,
> "kK"[!!rdp->nocb_cb_kthread],
> "bB"[raw_spin_is_locked(&rdp->nocb_bypass_lock)],
> "cC"[!!atomic_read(&rdp->nocb_lock_contended)],
> --
> 2.25.1
>