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, 324 insertions, 0 deletions

diff --git a/drivers/rtc/rtc-cpcap.c b/drivers/rtc/rtc-cpcap.c
new file mode 100644
index 000000000..a7db3173d
--- /dev/null
+++ b/drivers/rtc/rtc-cpcap.c
@@ -0,0 +1,324 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Motorola CPCAP PMIC RTC driver
+ *
+ * Based on cpcap-regulator.c from Motorola Linux kernel tree
+ * Copyright (C) 2009 Motorola, Inc.
+ *
+ * Rewritten for mainline kernel
+ *  - use DT
+ *  - use regmap
+ *  - use standard interrupt framework
+ *  - use managed device resources
+ *  - remove custom "secure clock daemon" helpers
+ *
+ * Copyright (C) 2017 Sebastian Reichel <sre@kernel.org>
+ */
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/device.h>
+#include <linux/platform_device.h>
+#include <linux/rtc.h>
+#include <linux/err.h>
+#include <linux/regmap.h>
+#include <linux/mfd/motorola-cpcap.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+
+#define SECS_PER_DAY 86400
+#define DAY_MASK  0x7FFF
+#define TOD1_MASK 0x00FF
+#define TOD2_MASK 0x01FF
+
+struct cpcap_time {
+	int day;
+	int tod1;
+	int tod2;
+};
+
+struct cpcap_rtc {
+	struct regmap *regmap;
+	struct rtc_device *rtc_dev;
+	u16 vendor;
+	int alarm_irq;
+	bool alarm_enabled;
+	int update_irq;
+	bool update_enabled;
+};
+
+static void cpcap2rtc_time(struct rtc_time *rtc, struct cpcap_time *cpcap)
+{
+	unsigned long int tod;
+	unsigned long int time;
+
+	tod = (cpcap->tod1 & TOD1_MASK) | ((cpcap->tod2 & TOD2_MASK) << 8);
+	time = tod + ((cpcap->day & DAY_MASK) * SECS_PER_DAY);
+
+	rtc_time64_to_tm(time, rtc);
+}
+
+static void rtc2cpcap_time(struct cpcap_time *cpcap, struct rtc_time *rtc)
+{
+	unsigned long time;
+
+	time = rtc_tm_to_time64(rtc);
+
+	cpcap->day = time / SECS_PER_DAY;
+	time %= SECS_PER_DAY;
+	cpcap->tod2 = (time >> 8) & TOD2_MASK;
+	cpcap->tod1 = time & TOD1_MASK;
+}
+
+static int cpcap_rtc_alarm_irq_enable(struct device *dev, unsigned int enabled)
+{
+	struct cpcap_rtc *rtc = dev_get_drvdata(dev);
+
+	if (rtc->alarm_enabled == enabled)
+		return 0;
+
+	if (enabled)
+		enable_irq(rtc->alarm_irq);
+	else
+		disable_irq(rtc->alarm_irq);
+
+	rtc->alarm_enabled = !!enabled;
+
+	return 0;
+}
+
+static int cpcap_rtc_read_time(struct device *dev, struct rtc_time *tm)
+{
+	struct cpcap_rtc *rtc;
+	struct cpcap_time cpcap_tm;
+	int temp_tod2;
+	int ret;
+
+	rtc = dev_get_drvdata(dev);
+
+	ret = regmap_read(rtc->regmap, CPCAP_REG_TOD2, &temp_tod2);
+	ret |= regmap_read(rtc->regmap, CPCAP_REG_DAY, &cpcap_tm.day);
+	ret |= regmap_read(rtc->regmap, CPCAP_REG_TOD1, &cpcap_tm.tod1);
+	ret |= regmap_read(rtc->regmap, CPCAP_REG_TOD2, &cpcap_tm.tod2);
+
+	if (temp_tod2 > cpcap_tm.tod2)
+		ret |= regmap_read(rtc->regmap, CPCAP_REG_DAY, &cpcap_tm.day);
+
+	if (ret) {
+		dev_err(dev, "Failed to read time\n");
+		return -EIO;
+	}
+
+	cpcap2rtc_time(tm, &cpcap_tm);
+
+	return 0;
+}
+
+static int cpcap_rtc_set_time(struct device *dev, struct rtc_time *tm)
+{
+	struct cpcap_rtc *rtc;
+	struct cpcap_time cpcap_tm;
+	int ret = 0;
+
+	rtc = dev_get_drvdata(dev);
+
+	rtc2cpcap_time(&cpcap_tm, tm);
+
+	if (rtc->alarm_enabled)
+		disable_irq(rtc->alarm_irq);
+	if (rtc->update_enabled)
+		disable_irq(rtc->update_irq);
+
+	if (rtc->vendor == CPCAP_VENDOR_ST) {
+		/* The TOD1 and TOD2 registers MUST be written in this order
+		 * for the change to properly set.
+		 */
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TOD1,
+					  TOD1_MASK, cpcap_tm.tod1);
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TOD2,
+					  TOD2_MASK, cpcap_tm.tod2);
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_DAY,
+					  DAY_MASK, cpcap_tm.day);
+	} else {
+		/* Clearing the upper lower 8 bits of the TOD guarantees that
+		 * the upper half of TOD (TOD2) will not increment for 0xFF RTC
+		 * ticks (255 seconds).  During this time we can safely write
+		 * to DAY, TOD2, then TOD1 (in that order) and expect RTC to be
+		 * synchronized to the exact time requested upon the final write
+		 * to TOD1.
+		 */
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TOD1,
+					  TOD1_MASK, 0);
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_DAY,
+					  DAY_MASK, cpcap_tm.day);
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TOD2,
+					  TOD2_MASK, cpcap_tm.tod2);
+		ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TOD1,
+					  TOD1_MASK, cpcap_tm.tod1);
+	}
+
+	if (rtc->update_enabled)
+		enable_irq(rtc->update_irq);
+	if (rtc->alarm_enabled)
+		enable_irq(rtc->alarm_irq);
+
+	return ret;
+}
+
+static int cpcap_rtc_read_alarm(struct device *dev, struct rtc_wkalrm *alrm)
+{
+	struct cpcap_rtc *rtc;
+	struct cpcap_time cpcap_tm;
+	int ret;
+
+	rtc = dev_get_drvdata(dev);
+
+	alrm->enabled = rtc->alarm_enabled;
+
+	ret = regmap_read(rtc->regmap, CPCAP_REG_DAYA, &cpcap_tm.day);
+	ret |= regmap_read(rtc->regmap, CPCAP_REG_TODA2, &cpcap_tm.tod2);
+	ret |= regmap_read(rtc->regmap, CPCAP_REG_TODA1, &cpcap_tm.tod1);
+
+	if (ret) {
+		dev_err(dev, "Failed to read time\n");
+		return -EIO;
+	}
+
+	cpcap2rtc_time(&alrm->time, &cpcap_tm);
+	return rtc_valid_tm(&alrm->time);
+}
+
+static int cpcap_rtc_set_alarm(struct device *dev, struct rtc_wkalrm *alrm)
+{
+	struct cpcap_rtc *rtc;
+	struct cpcap_time cpcap_tm;
+	int ret;
+
+	rtc = dev_get_drvdata(dev);
+
+	rtc2cpcap_time(&cpcap_tm, &alrm->time);
+
+	if (rtc->alarm_enabled)
+		disable_irq(rtc->alarm_irq);
+
+	ret = regmap_update_bits(rtc->regmap, CPCAP_REG_DAYA, DAY_MASK,
+				 cpcap_tm.day);
+	ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TODA2, TOD2_MASK,
+				  cpcap_tm.tod2);
+	ret |= regmap_update_bits(rtc->regmap, CPCAP_REG_TODA1, TOD1_MASK,
+				  cpcap_tm.tod1);
+
+	if (!ret) {
+		enable_irq(rtc->alarm_irq);
+		rtc->alarm_enabled = true;
+	}
+
+	return ret;
+}
+
+static const struct rtc_class_ops cpcap_rtc_ops = {
+	.read_time		= cpcap_rtc_read_time,
+	.set_time		= cpcap_rtc_set_time,
+	.read_alarm		= cpcap_rtc_read_alarm,
+	.set_alarm		= cpcap_rtc_set_alarm,
+	.alarm_irq_enable	= cpcap_rtc_alarm_irq_enable,
+};
+
+static irqreturn_t cpcap_rtc_alarm_irq(int irq, void *data)
+{
+	struct cpcap_rtc *rtc = data;
+
+	rtc_update_irq(rtc->rtc_dev, 1, RTC_AF | RTC_IRQF);
+	return IRQ_HANDLED;
+}
+
+static irqreturn_t cpcap_rtc_update_irq(int irq, void *data)
+{
+	struct cpcap_rtc *rtc = data;
+
+	rtc_update_irq(rtc->rtc_dev, 1, RTC_UF | RTC_IRQF);
+	return IRQ_HANDLED;
+}
+
+static int cpcap_rtc_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct cpcap_rtc *rtc;
+	int err;
+
+	rtc = devm_kzalloc(dev, sizeof(*rtc), GFP_KERNEL);
+	if (!rtc)
+		return -ENOMEM;
+
+	rtc->regmap = dev_get_regmap(dev->parent, NULL);
+	if (!rtc->regmap)
+		return -ENODEV;
+
+	platform_set_drvdata(pdev, rtc);
+	rtc->rtc_dev = devm_rtc_allocate_device(dev);
+	if (IS_ERR(rtc->rtc_dev))
+		return PTR_ERR(rtc->rtc_dev);
+
+	rtc->rtc_dev->ops = &cpcap_rtc_ops;
+	rtc->rtc_dev->range_max = (timeu64_t) (DAY_MASK + 1) * SECS_PER_DAY - 1;
+
+	err = cpcap_get_vendor(dev, rtc->regmap, &rtc->vendor);
+	if (err)
+		return err;
+
+	rtc->alarm_irq = platform_get_irq(pdev, 0);
+	err = devm_request_threaded_irq(dev, rtc->alarm_irq, NULL,
+					cpcap_rtc_alarm_irq,
+					IRQF_TRIGGER_NONE | IRQF_ONESHOT,
+					"rtc_alarm", rtc);
+	if (err) {
+		dev_err(dev, "Could not request alarm irq: %d\n", err);
+		return err;
+	}
+	disable_irq(rtc->alarm_irq);
+
+	/* Stock Android uses the 1 Hz interrupt for "secure clock daemon",
+	 * which is not supported by the mainline kernel. The mainline kernel
+	 * does not use the irq at the moment, but we explicitly request and
+	 * disable it, so that its masked and does not wake up the processor
+	 * every second.
+	 */
+	rtc->update_irq = platform_get_irq(pdev, 1);
+	err = devm_request_threaded_irq(dev, rtc->update_irq, NULL,
+					cpcap_rtc_update_irq,
+					IRQF_TRIGGER_NONE | IRQF_ONESHOT,
+					"rtc_1hz", rtc);
+	if (err) {
+		dev_err(dev, "Could not request update irq: %d\n", err);
+		return err;
+	}
+	disable_irq(rtc->update_irq);
+
+	err = device_init_wakeup(dev, true);
+	if (err) {
+		dev_err(dev, "wakeup initialization failed (%d)\n", err);
+		/* ignore error and continue without wakeup support */
+	}
+
+	return devm_rtc_register_device(rtc->rtc_dev);
+}
+
+static const struct of_device_id cpcap_rtc_of_match[] = {
+	{ .compatible = "motorola,cpcap-rtc", },
+	{},
+};
+MODULE_DEVICE_TABLE(of, cpcap_rtc_of_match);
+
+static struct platform_driver cpcap_rtc_driver = {
+	.probe		= cpcap_rtc_probe,
+	.driver		= {
+		.name	= "cpcap-rtc",
+		.of_match_table = cpcap_rtc_of_match,
+	},
+};
+
+module_platform_driver(cpcap_rtc_driver);
+
+MODULE_DESCRIPTION("CPCAP RTC driver");
+MODULE_AUTHOR("Sebastian Reichel <sre@kernel.org>");
+MODULE_LICENSE("GPL");