sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

alb_break_llc() decides whether to break LLC preference during active
load balance. It does so by testing that every runnable fair task on the
source rq prefers its LLC:

	env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable

But the two counters cover different sets. nr_pref_llc_running is updated
in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued,
so it follows queued tasks. h_nr_runnable is updated in set_delayed()/
clear_delayed() and drops delay-dequeued tasks.

So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted
in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks,
alb_break_llc() returns false, and active balance is free to pull a task
off its preferred LLC. Active balance only moves runnable tasks, and this
is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB
skips the per-task test in can_migrate_task(). The runnable set is the one
we want.

Fix it on the counter side. A task should be counted in
nr_pref_llc_running exactly while it is both queued on its preferred LLC
(pref_llc_queued) and runnable (!sched_delayed). Define that membership
once in task_pref_llc_runnable(), and adjust the counter only through
pref_llc_running_inc()/pref_llc_running_dec() from the four sites that
change either input: account_llc_enqueue(), account_llc_dequeue(),
set_delayed() and clear_delayed(). Gating every update on the same
predicate keeps the delay, wake and dequeue paths from double-counting
or underflowing; see the comments at those sites for the ordering.

nr_llc_running and sd->llc_counts are not touched and stay on queued
semantics.

Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC")
Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/
Reported-by: Zhan Xusheng <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

1 files changed, 171 insertions, 0 deletions

diff --git a/drivers/acpi/wakeup.c b/drivers/acpi/wakeup.c
new file mode 100644
index 000000000..0cbee0f06
--- /dev/null
+++ b/drivers/acpi/wakeup.c
@@ -0,0 +1,171 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * wakeup.c - support wakeup devices
+ * Copyright (C) 2004 Li Shaohua <shaohua.li@intel.com>
+ */
+
+#include <linux/init.h>
+#include <linux/acpi.h>
+#include <linux/kernel.h>
+#include <linux/types.h>
+
+#include "internal.h"
+#include "sleep.h"
+
+struct acpi_wakeup_handler {
+	struct list_head list_node;
+	bool (*wakeup)(void *context);
+	void *context;
+};
+
+static LIST_HEAD(acpi_wakeup_handler_head);
+static DEFINE_MUTEX(acpi_wakeup_handler_mutex);
+
+/*
+ * We didn't lock acpi_device_lock in the file, because it invokes oops in
+ * suspend/resume and isn't really required as this is called in S-state. At
+ * that time, there is no device hotplug
+ **/
+
+/**
+ * acpi_enable_wakeup_devices - Enable wake-up device GPEs.
+ * @sleep_state: ACPI system sleep state.
+ *
+ * Enable wakeup device power of devices with the state.enable flag set and set
+ * the wakeup enable mask bits in the GPE registers that correspond to wakeup
+ * devices.
+ */
+void acpi_enable_wakeup_devices(u8 sleep_state)
+{
+	struct acpi_device *dev, *tmp;
+
+	list_for_each_entry_safe(dev, tmp, &acpi_wakeup_device_list,
+				 wakeup_list) {
+		if (!dev->wakeup.flags.valid
+		    || sleep_state > dev->wakeup.sleep_state
+		    || !(device_may_wakeup(&dev->dev)
+			 || dev->wakeup.prepare_count))
+			continue;
+
+		if (device_may_wakeup(&dev->dev))
+			acpi_enable_wakeup_device_power(dev, sleep_state);
+
+		/* The wake-up power should have been enabled already. */
+		acpi_set_gpe_wake_mask(dev->wakeup.gpe_device, dev->wakeup.gpe_number,
+				ACPI_GPE_ENABLE);
+	}
+}
+
+/**
+ * acpi_disable_wakeup_devices - Disable devices' wakeup capability.
+ * @sleep_state: ACPI system sleep state.
+ */
+void acpi_disable_wakeup_devices(u8 sleep_state)
+{
+	struct acpi_device *dev, *tmp;
+
+	list_for_each_entry_safe(dev, tmp, &acpi_wakeup_device_list,
+				 wakeup_list) {
+		if (!dev->wakeup.flags.valid
+		    || sleep_state > dev->wakeup.sleep_state
+		    || !(device_may_wakeup(&dev->dev)
+			 || dev->wakeup.prepare_count))
+			continue;
+
+		acpi_set_gpe_wake_mask(dev->wakeup.gpe_device, dev->wakeup.gpe_number,
+				ACPI_GPE_DISABLE);
+
+		if (device_may_wakeup(&dev->dev))
+			acpi_disable_wakeup_device_power(dev);
+	}
+}
+
+int __init acpi_wakeup_device_init(void)
+{
+	struct acpi_device *dev, *tmp;
+
+	mutex_lock(&acpi_device_lock);
+	list_for_each_entry_safe(dev, tmp, &acpi_wakeup_device_list,
+				 wakeup_list) {
+		if (device_can_wakeup(&dev->dev)) {
+			/* Button GPEs are supposed to be always enabled. */
+			acpi_enable_gpe(dev->wakeup.gpe_device,
+					dev->wakeup.gpe_number);
+			device_set_wakeup_enable(&dev->dev, true);
+		}
+	}
+	mutex_unlock(&acpi_device_lock);
+	return 0;
+}
+
+/**
+ * acpi_register_wakeup_handler - Register wakeup handler
+ * @wake_irq: The IRQ through which the device may receive wakeups
+ * @wakeup:   Wakeup-handler to call when the SCI has triggered a wakeup
+ * @context:  Context to pass to the handler when calling it
+ *
+ * Drivers which may share an IRQ with the SCI can use this to register
+ * a handler which returns true when the device they are managing wants
+ * to trigger a wakeup.
+ */
+int acpi_register_wakeup_handler(int wake_irq, bool (*wakeup)(void *context),
+				 void *context)
+{
+	struct acpi_wakeup_handler *handler;
+
+	/*
+	 * If the device is not sharing its IRQ with the SCI, there is no
+	 * need to register the handler.
+	 */
+	if (!acpi_sci_irq_valid() || wake_irq != acpi_sci_irq)
+		return 0;
+
+	handler = kmalloc_obj(*handler);
+	if (!handler)
+		return -ENOMEM;
+
+	handler->wakeup = wakeup;
+	handler->context = context;
+
+	mutex_lock(&acpi_wakeup_handler_mutex);
+	list_add(&handler->list_node, &acpi_wakeup_handler_head);
+	mutex_unlock(&acpi_wakeup_handler_mutex);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(acpi_register_wakeup_handler);
+
+/**
+ * acpi_unregister_wakeup_handler - Unregister wakeup handler
+ * @wakeup:   Wakeup-handler passed to acpi_register_wakeup_handler()
+ * @context:  Context passed to acpi_register_wakeup_handler()
+ */
+void acpi_unregister_wakeup_handler(bool (*wakeup)(void *context),
+				    void *context)
+{
+	struct acpi_wakeup_handler *handler;
+
+	mutex_lock(&acpi_wakeup_handler_mutex);
+	list_for_each_entry(handler, &acpi_wakeup_handler_head, list_node) {
+		if (handler->wakeup == wakeup && handler->context == context) {
+			list_del(&handler->list_node);
+			kfree(handler);
+			break;
+		}
+	}
+	mutex_unlock(&acpi_wakeup_handler_mutex);
+}
+EXPORT_SYMBOL_GPL(acpi_unregister_wakeup_handler);
+
+bool acpi_check_wakeup_handlers(void)
+{
+	struct acpi_wakeup_handler *handler;
+
+	/* No need to lock, nothing else is running when we're called. */
+	list_for_each_entry(handler, &acpi_wakeup_handler_head, list_node) {
+		if (handler->wakeup(handler->context))
+			return true;
+	}
+
+	return false;
+}