[PATCH 0/3 v3] hrtimer: Fix timers queued locally from offline CPUs

[PATCH 0/3 v3] hrtimer: Fix timers queued locally from offline CPUs

[Frederic Weisbecker]

5c0930ccaad5 ("hrtimers: Push pending hrtimers away from outgoing CPU earlier")
was introduced to fix stalls with scheduler bandwidth timers getting
migrated while some kthreads handling CPU hotplug rely on bandwidth.

However this has introduced several other issues which used to be
confined to RCU. But not anymore as it is spreading to hotplug code
itself (https://lore.kernel.org/all/20241213203739.1519801-1-usamaarif642@gmail.com/)

Instead of introducing yet another new hackery, fix the problem in
hrtimers for everyone.

Changes since v2:

* Fix and add some tags
* Fix extra newlines

Frederic Weisbecker (3):
  hrtimers: Force migrate away hrtimers queued after
    CPUHP_AP_HRTIMERS_DYING
  rcu: Remove swake_up_one_online() bandaid
  Revert "rcu/nocb: Fix rcuog wake-up from offline softirq"

 include/linux/hrtimer_defs.h |  1 +
 kernel/rcu/tree.c            | 34 +----------------------
 kernel/rcu/tree_exp.h        |  2 +-
 kernel/rcu/tree_nocb.h       | 10 ++-----
 kernel/time/hrtimer.c        | 53 +++++++++++++++++++++++++++++++++---
 5 files changed, 54 insertions(+), 46 deletions(-)

-- 
2.46.0

[PATCH 1/3 v3] hrtimers: Force migrate away hrtimers queued after CPUHP_AP_HRTIMERS_DYING

[Frederic Weisbecker]

hrtimers are migrated away from the dying CPU to any online target at
the CPUHP_AP_HRTIMERS_DYING stage in order not to delay bandwidth timers
handling tasks involved in the CPU hotplug forward progress.

However wake ups can still be performed by the outgoing CPU after
CPUHP_AP_HRTIMERS_DYING. Those can result again in bandwidth timers
being armed. Depending on several considerations (crystal ball
power management based election, earliest timer already enqueued, timer
migration enabled or not), the target may eventually be the current
CPU even if offline. If that happens, the timer is eventually ignored.

The most notable example is RCU which had to deal with each an every of
those wake-ups by deferring them to an online CPU, along with related
workarounds:

_ e787644caf76 (rcu: Defer RCU kthreads wakeup when CPU is dying)
_ 9139f93209d1 (rcu/nocb: Fix RT throttling hrtimer armed from offline CPU)
_ f7345ccc62a4 (rcu/nocb: Fix rcuog wake-up from offline softirq)

The problem isn't confined to RCU though as the stop machine kthread
(which runs CPUHP_AP_HRTIMERS_DYING) reports its completion at the end
and performs a wake up that eventually arms the deadline server timer:

           WARNING: CPU: 94 PID: 588 at kernel/time/hrtimer.c:1086 hrtimer_start_range_ns+0x289/0x2d0
           Modules linked in:
           CPU: 94 UID: 0 PID: 588 Comm: migration/94 Not tainted
           Stopper: multi_cpu_stop+0x0/0x120 <- stop_machine_cpuslocked+0x66/0xc0
           RIP: 0010:hrtimer_start_range_ns+0x289/0x2d0
           Code: 41 5c 41 5d 41 5e 41 5f 5d e9 63 94 ea 00 0f 0b 48 83 c4 10 5b 41 5c 41 5d 41 5e 41 5f 5d e9 39 fc 15 01 0f 0b e9 c1 fd ff ff <0f> 0b 48 8b
+45 00 e9 59 ff ff ff f3 0f 1e fa 65 8b 05 1d ec e8 7e
           RSP: 0018:ffffc900019cbcc8 EFLAGS: 00010046
           RAX: ffff88bf449a4c40 RBX: 0000000000000082 RCX: 0000000000000001
           RDX: 0000000000000001 RSI: ffff88bf43224c80 RDI: ffff88bf449a4c40
           RBP: ffff88bf449a4c80 R08: ffff888280970090 R09: 0000000000000000
           R10: ffff88bf432252e0 R11: ffffffff811abf70 R12: ffff88bf449a4c40
           R13: ffff88bf43234b28 R14: ffff88bf43224c80 R15: 0000000000000000
           FS:  0000000000000000(0000) GS:ffff88bf44980000(0000) knlGS:0000000000000000
           CR2: 0000000000000000 CR3: 000000404b230001 CR4: 0000000000770ef0
           PKRU: 55555554
           Call Trace:
            <TASK>
            ? __warn+0xcf/0x1b0
            ? hrtimer_start_range_ns+0x289/0x2d0
            ? report_bug+0x120/0x1a0
            ? handle_bug+0x60/0x90
            ? exc_invalid_op+0x1a/0x50
            ? asm_exc_invalid_op+0x1a/0x20
            ? register_refined_jiffies+0xb0/0xb0
            ? hrtimer_start_range_ns+0x289/0x2d0
            ? hrtimer_start_range_ns+0x186/0x2d0
            start_dl_timer+0xfc/0x150
            enqueue_dl_entity+0x367/0x640
            dl_server_start+0x53/0xa0
            enqueue_task_fair+0x363/0x460
            enqueue_task+0x3c/0x200
            ttwu_do_activate+0x94/0x240
            try_to_wake_up+0x315/0x600
            complete+0x4b/0x80

Instead of providing yet another bandaid to work around the situation,
fix it from hrtimers infrastructure instead: always migrate away a
timer to an online target whenever it is enqueued from an offline CPU.

This will also allow to revert all the above RCU disgraceful hacks.

Reported-by: Vlad Poenaru <vlad.wing@gmail.com>
Reported-by: Usama Arif <usamaarif642@gmail.com>
Fixes: 5c0930ccaad5 ("hrtimers: Push pending hrtimers away from outgoing CPU earlier")
Closes: 20241213203739.1519801-1-usamaarif642@gmail.com
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
---
 include/linux/hrtimer_defs.h |  1 +
 kernel/time/hrtimer.c        | 53 +++++++++++++++++++++++++++++++++---
 2 files changed, 50 insertions(+), 4 deletions(-)

diff --git a/include/linux/hrtimer_defs.h b/include/linux/hrtimer_defs.h
index c3b4b7ed7c16..84a5045f80f3 100644
--- a/include/linux/hrtimer_defs.h
+++ b/include/linux/hrtimer_defs.h
@@ -125,6 +125,7 @@ struct hrtimer_cpu_base {
 	ktime_t				softirq_expires_next;
 	struct hrtimer			*softirq_next_timer;
 	struct hrtimer_clock_base	clock_base[HRTIMER_MAX_CLOCK_BASES];
+	call_single_data_t		csd;
 } ____cacheline_aligned;
 
 
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 80fe3749d2db..1ea9ef57d710 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -58,6 +58,8 @@
 #define HRTIMER_ACTIVE_SOFT	(HRTIMER_ACTIVE_HARD << MASK_SHIFT)
 #define HRTIMER_ACTIVE_ALL	(HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
 
+static void retrigger_next_event(void *arg);
+
 /*
  * The timer bases:
  *
@@ -111,7 +113,8 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 			.clockid = CLOCK_TAI,
 			.get_time = &ktime_get_clocktai,
 		},
-	}
+	},
+	.csd = CSD_INIT(retrigger_next_event, NULL)
 };
 
 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
@@ -716,8 +719,6 @@ static inline int hrtimer_is_hres_enabled(void)
 	return hrtimer_hres_enabled;
 }
 
-static void retrigger_next_event(void *arg);
-
 /*
  * Switch to high resolution mode
  */
@@ -1202,10 +1203,49 @@ hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
 	hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
 }
 
+/*
+ * If the current CPU is offline and timers have been already
+ * migrated away, make sure not to enqueue locally and perform
+ * a remote retrigger as a last resort.
+ */
+static void enqueue_hrtimer_offline(struct hrtimer *timer,
+				    struct hrtimer_clock_base *base,
+				    const enum hrtimer_mode mode)
+{
+#ifdef CONFIG_HOTPLUG_CPU
+	struct hrtimer_cpu_base *new_cpu_base, *old_cpu_base, *this_cpu_base;
+	struct hrtimer_clock_base *new_base;
+	int cpu;
+
+	old_cpu_base = base->cpu_base;
+	this_cpu_base = this_cpu_ptr(&hrtimer_bases);
+
+	if (old_cpu_base == this_cpu_base || !old_cpu_base->online) {
+		WARN_ON_ONCE(hrtimer_callback_running(timer));
+		cpu = cpumask_any_and(cpu_online_mask,
+				      housekeeping_cpumask(HK_TYPE_TIMER));
+		new_cpu_base = &per_cpu(hrtimer_bases, cpu);
+		new_base = &new_cpu_base->clock_base[base->index];
+		WRITE_ONCE(timer->base, &migration_base);
+		raw_spin_unlock(&old_cpu_base->lock);
+		raw_spin_lock(&new_cpu_base->lock);
+		WRITE_ONCE(timer->base, new_base);
+	} else {
+		new_base = base;
+		new_cpu_base = new_base->cpu_base;
+		cpu = new_cpu_base->cpu;
+	}
+
+	if (enqueue_hrtimer(timer, new_base, mode))
+		smp_call_function_single_async(cpu, &new_cpu_base->csd);
+#endif
+}
+
 static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 				    u64 delta_ns, const enum hrtimer_mode mode,
 				    struct hrtimer_clock_base *base)
 {
+	struct hrtimer_cpu_base *this_cpu_base = this_cpu_ptr(&hrtimer_bases);
 	struct hrtimer_clock_base *new_base;
 	bool force_local, first;
 
@@ -1217,7 +1257,7 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 	 * and enforce reprogramming after it is queued no matter whether
 	 * it is the new first expiring timer again or not.
 	 */
-	force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
+	force_local = base->cpu_base == this_cpu_base;
 	force_local &= base->cpu_base->next_timer == timer;
 
 	/*
@@ -1240,6 +1280,11 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 
 	hrtimer_set_expires_range_ns(timer, tim, delta_ns);
 
+	if (unlikely(!this_cpu_base->online)) {
+		enqueue_hrtimer_offline(timer, base, mode);
+		return 0;
+	}
+
 	/* Switch the timer base, if necessary: */
 	if (!force_local) {
 		new_base = switch_hrtimer_base(timer, base,
-- 
2.46.0

[PATCH 2/3 v3] rcu: Remove swake_up_one_online() bandaid

[Frederic Weisbecker]

It's now ok to perform a wake-up from an offline CPU because the
resulting armed scheduler bandwidth hrtimers are now correctly targeted
by hrtimer infrastructure.

Remove the obsolete hackerry.

Reviewed-by: Usama Arif <usamaarif642@gmail.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/rcu/tree.c      | 34 +---------------------------------
 kernel/rcu/tree_exp.h  |  2 +-
 kernel/rcu/tree_nocb.h |  2 +-
 3 files changed, 3 insertions(+), 35 deletions(-)

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index ff98233d4aa5..7c433c6b35ea 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1061,38 +1061,6 @@ static bool rcu_future_gp_cleanup(struct rcu_node *rnp)
 	return needmore;
 }
 
-static void swake_up_one_online_ipi(void *arg)
-{
-	struct swait_queue_head *wqh = arg;
-
-	swake_up_one(wqh);
-}
-
-static void swake_up_one_online(struct swait_queue_head *wqh)
-{
-	int cpu = get_cpu();
-
-	/*
-	 * If called from rcutree_report_cpu_starting(), wake up
-	 * is dangerous that late in the CPU-down hotplug process. The
-	 * scheduler might queue an ignored hrtimer. Defer the wake up
-	 * to an online CPU instead.
-	 */
-	if (unlikely(cpu_is_offline(cpu))) {
-		int target;
-
-		target = cpumask_any_and(housekeeping_cpumask(HK_TYPE_RCU),
-					 cpu_online_mask);
-
-		smp_call_function_single(target, swake_up_one_online_ipi,
-					 wqh, 0);
-		put_cpu();
-	} else {
-		put_cpu();
-		swake_up_one(wqh);
-	}
-}
-
 /*
  * Awaken the grace-period kthread.  Don't do a self-awaken (unless in an
  * interrupt or softirq handler, in which case we just might immediately
@@ -1117,7 +1085,7 @@ static void rcu_gp_kthread_wake(void)
 		return;
 	WRITE_ONCE(rcu_state.gp_wake_time, jiffies);
 	WRITE_ONCE(rcu_state.gp_wake_seq, READ_ONCE(rcu_state.gp_seq));
-	swake_up_one_online(&rcu_state.gp_wq);
+	swake_up_one(&rcu_state.gp_wq);
 }
 
 /*
diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h
index fb664d3a01c9..17b1a5e3590c 100644
--- a/kernel/rcu/tree_exp.h
+++ b/kernel/rcu/tree_exp.h
@@ -200,7 +200,7 @@ static void __rcu_report_exp_rnp(struct rcu_node *rnp,
 		if (rnp->parent == NULL) {
 			raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 			if (wake)
-				swake_up_one_online(&rcu_state.expedited_wq);
+				swake_up_one(&rcu_state.expedited_wq);
 
 			break;
 		}
diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
index 2605dd234a13..a43141a1b3a5 100644
--- a/kernel/rcu/tree_nocb.h
+++ b/kernel/rcu/tree_nocb.h
@@ -216,7 +216,7 @@ static bool __wake_nocb_gp(struct rcu_data *rdp_gp,
 	raw_spin_unlock_irqrestore(&rdp_gp->nocb_gp_lock, flags);
 	if (needwake) {
 		trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DoWake"));
-		swake_up_one_online(&rdp_gp->nocb_gp_wq);
+		swake_up_one(&rdp_gp->nocb_gp_wq);
 	}
 
 	return needwake;
-- 
2.46.0

[PATCH 3/3 v3] Revert "rcu/nocb: Fix rcuog wake-up from offline softirq"

[Frederic Weisbecker]

This reverts commit f7345ccc62a4b880cf76458db5f320725f28e400.

swake_up_one_online() has been removed because hrtimers can now assign
a proper online target to hrtimers queued from offline CPUs. Therefore
remove the related hackery.

Reviewed-by: Usama Arif <usamaarif642@gmail.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/rcu/tree_nocb.h | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
index a43141a1b3a5..a03fc19abde7 100644
--- a/kernel/rcu/tree_nocb.h
+++ b/kernel/rcu/tree_nocb.h
@@ -554,19 +554,13 @@ static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_alldone,
 			rcu_nocb_unlock(rdp);
 			wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE_LAZY,
 					   TPS("WakeLazy"));
-		} else if (!irqs_disabled_flags(flags) && cpu_online(rdp->cpu)) {
+		} else if (!irqs_disabled_flags(flags)) {
 			/* ... if queue was empty ... */
 			rcu_nocb_unlock(rdp);
 			wake_nocb_gp(rdp, false);
 			trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
 					    TPS("WakeEmpty"));
 		} else {
-			/*
-			 * Don't do the wake-up upfront on fragile paths.
-			 * Also offline CPUs can't call swake_up_one_online() from
-			 * (soft-)IRQs. Rely on the final deferred wake-up from
-			 * rcutree_report_cpu_dead()
-			 */
 			rcu_nocb_unlock(rdp);
 			wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE,
 					   TPS("WakeEmptyIsDeferred"));
-- 
2.46.0

Re: [PATCH 1/3 v3] hrtimers: Force migrate away hrtimers queued after CPUHP_AP_HRTIMERS_DYING

[Thomas Gleixner]

On Tue, Dec 31 2024 at 18:07, Frederic Weisbecker wrote:
> hrtimers are migrated away from the dying CPU to any online target at
> the CPUHP_AP_HRTIMERS_DYING stage in order not to delay bandwidth timers
> handling tasks involved in the CPU hotplug forward progress.
>
> However wake ups can still be performed by the outgoing CPU after
> CPUHP_AP_HRTIMERS_DYING. Those can result again in bandwidth timers
> being armed. Depending on several considerations (crystal ball
> power management based election, earliest timer already enqueued, timer
> migration enabled or not), the target may eventually be the current
> CPU even if offline. If that happens, the timer is eventually ignored.
>
> The most notable example is RCU which had to deal with each an every of
> those wake-ups by deferring them to an online CPU, along with related
> workarounds:
>
> _ e787644caf76 (rcu: Defer RCU kthreads wakeup when CPU is dying)
> _ 9139f93209d1 (rcu/nocb: Fix RT throttling hrtimer armed from offline CPU)
> _ f7345ccc62a4 (rcu/nocb: Fix rcuog wake-up from offline softirq)
>
> The problem isn't confined to RCU though as the stop machine kthread
> (which runs CPUHP_AP_HRTIMERS_DYING) reports its completion at the end
> and performs a wake up that eventually arms the deadline server timer:

Where does it report the completion?

>            WARNING: CPU: 94 PID: 588 at kernel/time/hrtimer.c:1086 hrtimer_start_range_ns+0x289/0x2d0
>            CPU: 94 UID: 0 PID: 588 Comm: migration/94 Not tainted
>            Stopper: multi_cpu_stop+0x0/0x120 <- stop_machine_cpuslocked+0x66/0xc0
>            RIP: 0010:hrtimer_start_range_ns+0x289/0x2d0
>            Call Trace:
>             ? __warn+0xcf/0x1b0
>             ? hrtimer_start_range_ns+0x289/0x2d0
>             ? hrtimer_start_range_ns+0x186/0x2d0
>             start_dl_timer+0xfc/0x150
>             enqueue_dl_entity+0x367/0x640
>             dl_server_start+0x53/0xa0
>             enqueue_task_fair+0x363/0x460
>             enqueue_task+0x3c/0x200
>             ttwu_do_activate+0x94/0x240
>             try_to_wake_up+0x315/0x600
>             complete+0x4b/0x80

You trimmed the backtrace at the wrong end. You left all the useless
register gunk around, but removed the call chain leading up to the
completion. :)

I assume it's the complete in stomp_machine() ....

> Instead of providing yet another bandaid to work around the situation,
> fix it from hrtimers infrastructure instead: always migrate away a
> timer to an online target whenever it is enqueued from an offline CPU.

> +/*
> + * If the current CPU is offline and timers have been already
> + * migrated away, make sure not to enqueue locally and perform
> + * a remote retrigger as a last resort.
> + */
> +static void enqueue_hrtimer_offline(struct hrtimer *timer,
> +				    struct hrtimer_clock_base *base,
> +				    const enum hrtimer_mode mode)
> +{
> +#ifdef CONFIG_HOTPLUG_CPU
> +	struct hrtimer_cpu_base *new_cpu_base, *old_cpu_base, *this_cpu_base;
> +	struct hrtimer_clock_base *new_base;
> +	int cpu;
> +
> +	old_cpu_base = base->cpu_base;
> +	this_cpu_base = this_cpu_ptr(&hrtimer_bases);
> +
> +	if (old_cpu_base == this_cpu_base || !old_cpu_base->online) {
> +		WARN_ON_ONCE(hrtimer_callback_running(timer));
> +		cpu = cpumask_any_and(cpu_online_mask,
> +				      housekeeping_cpumask(HK_TYPE_TIMER));
> +		new_cpu_base = &per_cpu(hrtimer_bases, cpu);
> +		new_base = &new_cpu_base->clock_base[base->index];
> +		WRITE_ONCE(timer->base, &migration_base);
> +		raw_spin_unlock(&old_cpu_base->lock);
> +		raw_spin_lock(&new_cpu_base->lock);
> +		WRITE_ONCE(timer->base, new_base);
> +	} else {
> +		new_base = base;
> +		new_cpu_base = new_base->cpu_base;
> +		cpu = new_cpu_base->cpu;
> +	}
> +
> +	if (enqueue_hrtimer(timer, new_base, mode))
> +		smp_call_function_single_async(cpu, &new_cpu_base->csd);

Duh. This reimplementation of switch_hrtimer_base() is really aweful. We
can be smarter than that. Untested patch below.

Thanks,

        tglx
---
--- a/include/linux/hrtimer_defs.h
+++ b/include/linux/hrtimer_defs.h
@@ -125,6 +125,7 @@ struct hrtimer_cpu_base {
 	ktime_t				softirq_expires_next;
 	struct hrtimer			*softirq_next_timer;
 	struct hrtimer_clock_base	clock_base[HRTIMER_MAX_CLOCK_BASES];
+	call_single_data_t		csd;
 } ____cacheline_aligned;
 
 
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -58,6 +58,8 @@
 #define HRTIMER_ACTIVE_SOFT	(HRTIMER_ACTIVE_HARD << MASK_SHIFT)
 #define HRTIMER_ACTIVE_ALL	(HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
 
+static void retrigger_next_event(void *arg);
+
 /*
  * The timer bases:
  *
@@ -111,7 +113,8 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base,
 			.clockid = CLOCK_TAI,
 			.get_time = &ktime_get_clocktai,
 		},
-	}
+	},
+	.csd = CSD_INIT(retrigger_next_event, NULL)
 };
 
 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
@@ -208,6 +211,13 @@ struct hrtimer_cpu_base *get_target_base
 	if (static_branch_likely(&timers_migration_enabled) && !pinned)
 		return &per_cpu(hrtimer_bases, get_nohz_timer_target());
 #endif
+#ifdef CONFIG_HOTPLUG_CPU
+	if (unlikely(!base->online)) {
+		int cpu = cpumask_any_and(cpu_online_mask, housekeeping_cpumask(HK_TYPE_TIMER));
+
+		return &per_cpu(hrtimer_bases, cpu);
+	}
+#endif
 	return base;
 }
 
@@ -254,7 +264,7 @@ switch_hrtimer_base(struct hrtimer *time
 		raw_spin_unlock(&base->cpu_base->lock);
 		raw_spin_lock(&new_base->cpu_base->lock);
 
-		if (new_cpu_base != this_cpu_base &&
+		if (new_cpu_base != this_cpu_base && this_cpu_base->online &&
 		    hrtimer_check_target(timer, new_base)) {
 			raw_spin_unlock(&new_base->cpu_base->lock);
 			raw_spin_lock(&base->cpu_base->lock);
@@ -716,8 +726,6 @@ static inline int hrtimer_is_hres_enable
 	return hrtimer_hres_enabled;
 }
 
-static void retrigger_next_event(void *arg);
-
 /*
  * Switch to high resolution mode
  */
@@ -1206,6 +1214,7 @@ static int __hrtimer_start_range_ns(stru
 				    u64 delta_ns, const enum hrtimer_mode mode,
 				    struct hrtimer_clock_base *base)
 {
+	struct hrtimer_cpu_base *this_cpu_base = this_cpu_ptr(&hrtimer_bases);
 	struct hrtimer_clock_base *new_base;
 	bool force_local, first;
 
@@ -1217,10 +1226,16 @@ static int __hrtimer_start_range_ns(stru
 	 * and enforce reprogramming after it is queued no matter whether
 	 * it is the new first expiring timer again or not.
 	 */
-	force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
+	force_local = base->cpu_base == this_cpu_base;
 	force_local &= base->cpu_base->next_timer == timer;
 
 	/*
+	 * Don't force local queuing if this enqueue happens on a unplugged
+	 * CPU after hrtimer_cpu_dying() has been invoked.
+	 */
+	force_local &= this_cpu_base->online;
+
+	/*
 	 * Remove an active timer from the queue. In case it is not queued
 	 * on the current CPU, make sure that remove_hrtimer() updates the
 	 * remote data correctly.
@@ -1249,8 +1264,17 @@ static int __hrtimer_start_range_ns(stru
 	}
 
 	first = enqueue_hrtimer(timer, new_base, mode);
-	if (!force_local)
-		return first;
+	if (!force_local) {
+		if (likely(this_cpu_base->online))
+			return first;
+
+		if (first) {
+			struct hrtimer_cpu_base *new_cpu_base = new_base->cpu_base;
+
+			smp_call_function_single_async(new_cpu_base->cpu, &new_cpu_base->csd);
+		}
+		return 0;
+	}
 
 	/*
 	 * Timer was forced to stay on the current CPU to avoid

Re: [PATCH 1/3 v3] hrtimers: Force migrate away hrtimers queued after CPUHP_AP_HRTIMERS_DYING

[Frederic Weisbecker]

Le Thu, Jan 16, 2025 at 11:59:48AM +0100, Thomas Gleixner a écrit :
> On Tue, Dec 31 2024 at 18:07, Frederic Weisbecker wrote:
> > hrtimers are migrated away from the dying CPU to any online target at
> > the CPUHP_AP_HRTIMERS_DYING stage in order not to delay bandwidth timers
> > handling tasks involved in the CPU hotplug forward progress.
> >
> > However wake ups can still be performed by the outgoing CPU after
> > CPUHP_AP_HRTIMERS_DYING. Those can result again in bandwidth timers
> > being armed. Depending on several considerations (crystal ball
> > power management based election, earliest timer already enqueued, timer
> > migration enabled or not), the target may eventually be the current
> > CPU even if offline. If that happens, the timer is eventually ignored.
> >
> > The most notable example is RCU which had to deal with each an every of
> > those wake-ups by deferring them to an online CPU, along with related
> > workarounds:
> >
> > _ e787644caf76 (rcu: Defer RCU kthreads wakeup when CPU is dying)
> > _ 9139f93209d1 (rcu/nocb: Fix RT throttling hrtimer armed from offline CPU)
> > _ f7345ccc62a4 (rcu/nocb: Fix rcuog wake-up from offline softirq)
> >
> > The problem isn't confined to RCU though as the stop machine kthread
> > (which runs CPUHP_AP_HRTIMERS_DYING) reports its completion at the end
> > and performs a wake up that eventually arms the deadline server timer:
> 
> Where does it report the completion?

Right, that's cpu_stop_signal_done()

> 
> >            WARNING: CPU: 94 PID: 588 at kernel/time/hrtimer.c:1086 hrtimer_start_range_ns+0x289/0x2d0
> >            CPU: 94 UID: 0 PID: 588 Comm: migration/94 Not tainted
> >            Stopper: multi_cpu_stop+0x0/0x120 <- stop_machine_cpuslocked+0x66/0xc0
> >            RIP: 0010:hrtimer_start_range_ns+0x289/0x2d0
> >            Call Trace:
> >             ? __warn+0xcf/0x1b0
> >             ? hrtimer_start_range_ns+0x289/0x2d0
> >             ? hrtimer_start_range_ns+0x186/0x2d0
> >             start_dl_timer+0xfc/0x150
> >             enqueue_dl_entity+0x367/0x640
> >             dl_server_start+0x53/0xa0
> >             enqueue_task_fair+0x363/0x460
> >             enqueue_task+0x3c/0x200
> >             ttwu_do_activate+0x94/0x240
> >             try_to_wake_up+0x315/0x600
> >             complete+0x4b/0x80
> 
> You trimmed the backtrace at the wrong end. You left all the useless
> register gunk around, but removed the call chain leading up to the
> completion. :)
> 
> I assume it's the complete in stomp_machine() ....

Bah! Indeed, lemme fix that...

> > +/*
> > + * If the current CPU is offline and timers have been already
> > + * migrated away, make sure not to enqueue locally and perform
> > + * a remote retrigger as a last resort.
> > + */
> > +static void enqueue_hrtimer_offline(struct hrtimer *timer,
> > +				    struct hrtimer_clock_base *base,
> > +				    const enum hrtimer_mode mode)
> > +{
> > +#ifdef CONFIG_HOTPLUG_CPU
> > +	struct hrtimer_cpu_base *new_cpu_base, *old_cpu_base, *this_cpu_base;
> > +	struct hrtimer_clock_base *new_base;
> > +	int cpu;
> > +
> > +	old_cpu_base = base->cpu_base;
> > +	this_cpu_base = this_cpu_ptr(&hrtimer_bases);
> > +
> > +	if (old_cpu_base == this_cpu_base || !old_cpu_base->online) {
> > +		WARN_ON_ONCE(hrtimer_callback_running(timer));
> > +		cpu = cpumask_any_and(cpu_online_mask,
> > +				      housekeeping_cpumask(HK_TYPE_TIMER));
> > +		new_cpu_base = &per_cpu(hrtimer_bases, cpu);
> > +		new_base = &new_cpu_base->clock_base[base->index];
> > +		WRITE_ONCE(timer->base, &migration_base);
> > +		raw_spin_unlock(&old_cpu_base->lock);
> > +		raw_spin_lock(&new_cpu_base->lock);
> > +		WRITE_ONCE(timer->base, new_base);
> > +	} else {
> > +		new_base = base;
> > +		new_cpu_base = new_base->cpu_base;
> > +		cpu = new_cpu_base->cpu;
> > +	}
> > +
> > +	if (enqueue_hrtimer(timer, new_base, mode))
> > +		smp_call_function_single_async(cpu, &new_cpu_base->csd);
> 
> Duh. This reimplementation of switch_hrtimer_base() is really aweful. We
> can be smarter than that. Untested patch below.

Indeed, and I tried to "fix" switch_hrtimer_base() but didn't managed to do
it properly. Looks like you did :-)

But I have a few comments:

> @@ -208,6 +211,13 @@ struct hrtimer_cpu_base *get_target_base
>  	if (static_branch_likely(&timers_migration_enabled) && !pinned)
>  		return &per_cpu(hrtimer_bases, get_nohz_timer_target());
>  #endif
> +#ifdef CONFIG_HOTPLUG_CPU
> +	if (unlikely(!base->online)) {
> +		int cpu = cpumask_any_and(cpu_online_mask, housekeeping_cpumask(HK_TYPE_TIMER));
> +
> +		return &per_cpu(hrtimer_bases, cpu);
> +	}
> +#endif

This should be at the beginning of get_target_base(), otherwise the target may
end up being the local one.


>  	return base;
>  }
>  
> @@ -254,7 +264,7 @@ switch_hrtimer_base(struct hrtimer *time
>  		raw_spin_unlock(&base->cpu_base->lock);
>  		raw_spin_lock(&new_base->cpu_base->lock);
>  
> -		if (new_cpu_base != this_cpu_base &&
> +		if (new_cpu_base != this_cpu_base && this_cpu_base->online &&
>  		    hrtimer_check_target(timer, new_base)) {
>  			raw_spin_unlock(&new_base->cpu_base->lock);
>  			raw_spin_lock(&base->cpu_base->lock);

This forget the case where the elected remote target is the same as the old one,
but its next event is after the timer. In that case this default to local, even
if it is offline.

How about this? (untested yet)

diff --git a/include/linux/hrtimer_defs.h b/include/linux/hrtimer_defs.h
index c3b4b7ed7c16..84a5045f80f3 100644
--- a/include/linux/hrtimer_defs.h
+++ b/include/linux/hrtimer_defs.h
@@ -125,6 +125,7 @@ struct hrtimer_cpu_base {
 	ktime_t				softirq_expires_next;
 	struct hrtimer			*softirq_next_timer;
 	struct hrtimer_clock_base	clock_base[HRTIMER_MAX_CLOCK_BASES];
+	call_single_data_t		csd;
 } ____cacheline_aligned;
 
 
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 80fe3749d2db..248330e02da7 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -58,6 +58,8 @@
 #define HRTIMER_ACTIVE_SOFT	(HRTIMER_ACTIVE_HARD << MASK_SHIFT)
 #define HRTIMER_ACTIVE_ALL	(HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
 
+static void retrigger_next_event(void *arg);
+
 /*
  * The timer bases:
  *
@@ -111,7 +113,8 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 			.clockid = CLOCK_TAI,
 			.get_time = &ktime_get_clocktai,
 		},
-	}
+	},
+	.csd = CSD_INIT(retrigger_next_event, NULL)
 };
 
 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
@@ -124,6 +127,14 @@ static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
 	[CLOCK_TAI]		= HRTIMER_BASE_TAI,
 };
 
+static inline bool hrtimer_base_is_online(struct hrtimer_cpu_base *base)
+{
+	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
+		return true;
+	else
+		return likely(base->online);
+}
+
 /*
  * Functions and macros which are different for UP/SMP systems are kept in a
  * single place
@@ -183,27 +194,55 @@ struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
 }
 
 /*
- * We do not migrate the timer when it is expiring before the next
- * event on the target cpu. When high resolution is enabled, we cannot
- * reprogram the target cpu hardware and we would cause it to fire
- * late. To keep it simple, we handle the high resolution enabled and
- * disabled case similar.
+ * Check if the elected target is suitable considering its next
+ * event and the hotplug state of the current CPU.
+ *
+ * If the elected target is remote and its next event is after the timer
+ * to queue, then a remote reprogram is necessary. However there is no
+ * guarantee the IPI handling the operation would arrive in time to meet
+ * the high resolution deadline. In this case the local CPU becomes a
+ * preferred target, unless it is offline.
+ *
+ * High and low resolution modes are handled the same way for simplicity.
  *
  * Called with cpu_base->lock of target cpu held.
  */
-static int
-hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
+static bool
+hrtimer_suitable_target(struct hrtimer *timer,
+			struct hrtimer_clock_base *new_base,
+			struct hrtimer_cpu_base *new_cpu_base,
+			struct hrtimer_cpu_base *this_cpu_base)
 {
 	ktime_t expires;
 
+	/* The local CPU clockevent can be reprogrammed */
+	if (new_cpu_base == this_cpu_base)
+		return true;
+
+	/*
+	 * The offline local CPU can't be the default target if the
+	 * next remote target event is after this timer. Keep the
+	 * elected new base. An IPI will we issued to reprogram
+	 * it as a last resort.
+	 */
+	if (!hrtimer_base_is_online(this_cpu_base))
+		return true;
+
 	expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
-	return expires < new_base->cpu_base->expires_next;
+
+	return expires >= new_base->cpu_base->expires_next;
 }
 
 static inline
 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
 					 int pinned)
 {
+	if (!hrtimer_base_is_online(base)) {
+		int cpu = cpumask_any_and(cpu_online_mask, housekeeping_cpumask(HK_TYPE_TIMER));
+
+		return &per_cpu(hrtimer_bases, cpu);
+	}
+
 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
 	if (static_branch_likely(&timers_migration_enabled) && !pinned)
 		return &per_cpu(hrtimer_bases, get_nohz_timer_target());
@@ -211,6 +250,8 @@ struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
 	return base;
 }
 
+
+
 /*
  * We switch the timer base to a power-optimized selected CPU target,
  * if:
@@ -254,8 +295,8 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 		raw_spin_unlock(&base->cpu_base->lock);
 		raw_spin_lock(&new_base->cpu_base->lock);
 
-		if (new_cpu_base != this_cpu_base &&
-		    hrtimer_check_target(timer, new_base)) {
+		if (!hrtimer_suitable_target(timer, new_base, new_cpu_base,
+					     this_cpu_base)) {
 			raw_spin_unlock(&new_base->cpu_base->lock);
 			raw_spin_lock(&base->cpu_base->lock);
 			new_cpu_base = this_cpu_base;
@@ -264,8 +305,8 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 		}
 		WRITE_ONCE(timer->base, new_base);
 	} else {
-		if (new_cpu_base != this_cpu_base &&
-		    hrtimer_check_target(timer, new_base)) {
+		if (!hrtimer_suitable_target(timer, new_base,  new_cpu_base,
+					     this_cpu_base)) {
 			new_cpu_base = this_cpu_base;
 			goto again;
 		}
@@ -716,8 +757,6 @@ static inline int hrtimer_is_hres_enabled(void)
 	return hrtimer_hres_enabled;
 }
 
-static void retrigger_next_event(void *arg);
-
 /*
  * Switch to high resolution mode
  */
@@ -1206,6 +1245,7 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 				    u64 delta_ns, const enum hrtimer_mode mode,
 				    struct hrtimer_clock_base *base)
 {
+	struct hrtimer_cpu_base *this_cpu_base = this_cpu_ptr(&hrtimer_bases);
 	struct hrtimer_clock_base *new_base;
 	bool force_local, first;
 
@@ -1217,9 +1257,15 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 	 * and enforce reprogramming after it is queued no matter whether
 	 * it is the new first expiring timer again or not.
 	 */
-	force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
+	force_local = base->cpu_base == this_cpu_base;
 	force_local &= base->cpu_base->next_timer == timer;
 
+	/*
+	 * Don't force local queuing if this enqueue happens on a unplugged
+	 * CPU after hrtimer_cpu_dying() has been invoked.
+	 */
+	force_local &= this_cpu_base->online;
+
 	/*
 	 * Remove an active timer from the queue. In case it is not queued
 	 * on the current CPU, make sure that remove_hrtimer() updates the
@@ -1249,8 +1295,17 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 	}
 
 	first = enqueue_hrtimer(timer, new_base, mode);
-	if (!force_local)
-		return first;
+	if (!force_local) {
+		if (hrtimer_base_is_online(this_cpu_base))
+			return first;
+
+		if (first) {
+			struct hrtimer_cpu_base *new_cpu_base = new_base->cpu_base;
+
+			smp_call_function_single_async(new_cpu_base->cpu, &new_cpu_base->csd);
+		}
+		return 0;
+	}
 
 	/*
 	 * Timer was forced to stay on the current CPU to avoid

Re: [PATCH 1/3 v3] hrtimers: Force migrate away hrtimers queued after CPUHP_AP_HRTIMERS_DYING

[Thomas Gleixner]

On Fri, Jan 17 2025 at 15:11, Frederic Weisbecker wrote:
> Le Thu, Jan 16, 2025 at 11:59:48AM +0100, Thomas Gleixner a écrit :
>> > +	if (enqueue_hrtimer(timer, new_base, mode))
>> > +		smp_call_function_single_async(cpu, &new_cpu_base->csd);
>> 
>> Duh. This reimplementation of switch_hrtimer_base() is really aweful. We
>> can be smarter than that. Untested patch below.
>
> Indeed, and I tried to "fix" switch_hrtimer_base() but didn't managed to do
> it properly. Looks like you did :-)
>
> But I have a few comments:
>
>> @@ -208,6 +211,13 @@ struct hrtimer_cpu_base *get_target_base
>>  	if (static_branch_likely(&timers_migration_enabled) && !pinned)
>>  		return &per_cpu(hrtimer_bases, get_nohz_timer_target());
>>  #endif
>> +#ifdef CONFIG_HOTPLUG_CPU
>> +	if (unlikely(!base->online)) {
>> +		int cpu = cpumask_any_and(cpu_online_mask, housekeeping_cpumask(HK_TYPE_TIMER));
>> +
>> +		return &per_cpu(hrtimer_bases, cpu);
>> +	}
>> +#endif
>
> This should be at the beginning of get_target_base(), otherwise the target may
> end up being the local one.

Indeed. As I said, it's untested.

>> -		if (new_cpu_base != this_cpu_base &&
>> +		if (new_cpu_base != this_cpu_base && this_cpu_base->online &&
>>  		    hrtimer_check_target(timer, new_base)) {
>>  			raw_spin_unlock(&new_base->cpu_base->lock);
>>  			raw_spin_lock(&base->cpu_base->lock);
>
> This forget the case where the elected remote target is the same as the old one,
> but its next event is after the timer. In that case this default to local, even
> if it is offline.

*blink*

> How about this? (untested yet)
> -static int
> -hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
> +static bool
> +hrtimer_suitable_target(struct hrtimer *timer,
> +			struct hrtimer_clock_base *new_base,
> +			struct hrtimer_cpu_base *new_cpu_base,
> +			struct hrtimer_cpu_base *this_cpu_base)
>  {
>  	ktime_t expires;
>  
> +	/* The local CPU clockevent can be reprogrammed */
> +	if (new_cpu_base == this_cpu_base)
> +		return true;

That needs a comment explaining that @new_cpu_base is guaranteed not to
be @this_cpu_base if @this_cpu_base->online == false due to the magic in
get_target_base(). That's far from obvious.

I had to stare at it 5 times to convince myself that this is correct.

Other than that, this looks about right.

Thanks,

        tglx