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, 232 insertions, 0 deletions

diff --git a/drivers/reset/reset-ma35d1.c b/drivers/reset/reset-ma35d1.c
new file mode 100644
index 000000000..4ee901f00
--- /dev/null
+++ b/drivers/reset/reset-ma35d1.c
@@ -0,0 +1,232 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2023 Nuvoton Technology Corp.
+ * Author: Chi-Fang Li <cfli0@nuvoton.com>
+ */
+
+#include <linux/bits.h>
+#include <linux/container_of.h>
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/reboot.h>
+#include <linux/reset-controller.h>
+#include <linux/spinlock.h>
+#include <dt-bindings/reset/nuvoton,ma35d1-reset.h>
+
+struct ma35d1_reset_data {
+	struct reset_controller_dev rcdev;
+	void __iomem *base;
+	/* protect registers against concurrent read-modify-write */
+	spinlock_t lock;
+};
+
+static const struct {
+	u32 reg_ofs;
+	u32 bit;
+} ma35d1_reset_map[] = {
+	[MA35D1_RESET_CHIP] =    {0x20, 0},
+	[MA35D1_RESET_CA35CR0] = {0x20, 1},
+	[MA35D1_RESET_CA35CR1] = {0x20, 2},
+	[MA35D1_RESET_CM4] =     {0x20, 3},
+	[MA35D1_RESET_PDMA0] =   {0x20, 4},
+	[MA35D1_RESET_PDMA1] =   {0x20, 5},
+	[MA35D1_RESET_PDMA2] =   {0x20, 6},
+	[MA35D1_RESET_PDMA3] =   {0x20, 7},
+	[MA35D1_RESET_DISP] =    {0x20, 9},
+	[MA35D1_RESET_VCAP0] =   {0x20, 10},
+	[MA35D1_RESET_VCAP1] =   {0x20, 11},
+	[MA35D1_RESET_GFX] =     {0x20, 12},
+	[MA35D1_RESET_VDEC] =    {0x20, 13},
+	[MA35D1_RESET_WHC0] =    {0x20, 14},
+	[MA35D1_RESET_WHC1] =    {0x20, 15},
+	[MA35D1_RESET_GMAC0] =   {0x20, 16},
+	[MA35D1_RESET_GMAC1] =   {0x20, 17},
+	[MA35D1_RESET_HWSEM] =   {0x20, 18},
+	[MA35D1_RESET_EBI] =     {0x20, 19},
+	[MA35D1_RESET_HSUSBH0] = {0x20, 20},
+	[MA35D1_RESET_HSUSBH1] = {0x20, 21},
+	[MA35D1_RESET_HSUSBD] =  {0x20, 22},
+	[MA35D1_RESET_USBHL] =   {0x20, 23},
+	[MA35D1_RESET_SDH0] =    {0x20, 24},
+	[MA35D1_RESET_SDH1] =    {0x20, 25},
+	[MA35D1_RESET_NAND] =    {0x20, 26},
+	[MA35D1_RESET_GPIO] =    {0x20, 27},
+	[MA35D1_RESET_MCTLP] =   {0x20, 28},
+	[MA35D1_RESET_MCTLC] =   {0x20, 29},
+	[MA35D1_RESET_DDRPUB] =  {0x20, 30},
+	[MA35D1_RESET_TMR0] =    {0x24, 2},
+	[MA35D1_RESET_TMR1] =    {0x24, 3},
+	[MA35D1_RESET_TMR2] =    {0x24, 4},
+	[MA35D1_RESET_TMR3] =    {0x24, 5},
+	[MA35D1_RESET_I2C0] =    {0x24, 8},
+	[MA35D1_RESET_I2C1] =    {0x24, 9},
+	[MA35D1_RESET_I2C2] =    {0x24, 10},
+	[MA35D1_RESET_I2C3] =    {0x24, 11},
+	[MA35D1_RESET_QSPI0] =   {0x24, 12},
+	[MA35D1_RESET_SPI0] =    {0x24, 13},
+	[MA35D1_RESET_SPI1] =    {0x24, 14},
+	[MA35D1_RESET_SPI2] =    {0x24, 15},
+	[MA35D1_RESET_UART0] =   {0x24, 16},
+	[MA35D1_RESET_UART1] =   {0x24, 17},
+	[MA35D1_RESET_UART2] =   {0x24, 18},
+	[MA35D1_RESET_UART3] =   {0x24, 19},
+	[MA35D1_RESET_UART4] =   {0x24, 20},
+	[MA35D1_RESET_UART5] =   {0x24, 21},
+	[MA35D1_RESET_UART6] =   {0x24, 22},
+	[MA35D1_RESET_UART7] =   {0x24, 23},
+	[MA35D1_RESET_CANFD0] =  {0x24, 24},
+	[MA35D1_RESET_CANFD1] =  {0x24, 25},
+	[MA35D1_RESET_EADC0] =   {0x24, 28},
+	[MA35D1_RESET_I2S0] =    {0x24, 29},
+	[MA35D1_RESET_SC0] =     {0x28, 0},
+	[MA35D1_RESET_SC1] =     {0x28, 1},
+	[MA35D1_RESET_QSPI1] =   {0x28, 4},
+	[MA35D1_RESET_SPI3] =    {0x28, 6},
+	[MA35D1_RESET_EPWM0] =   {0x28, 16},
+	[MA35D1_RESET_EPWM1] =   {0x28, 17},
+	[MA35D1_RESET_QEI0] =    {0x28, 22},
+	[MA35D1_RESET_QEI1] =    {0x28, 23},
+	[MA35D1_RESET_ECAP0] =   {0x28, 26},
+	[MA35D1_RESET_ECAP1] =   {0x28, 27},
+	[MA35D1_RESET_CANFD2] =  {0x28, 28},
+	[MA35D1_RESET_ADC0] =    {0x28, 31},
+	[MA35D1_RESET_TMR4] =    {0x2C, 0},
+	[MA35D1_RESET_TMR5] =    {0x2C, 1},
+	[MA35D1_RESET_TMR6] =    {0x2C, 2},
+	[MA35D1_RESET_TMR7] =    {0x2C, 3},
+	[MA35D1_RESET_TMR8] =    {0x2C, 4},
+	[MA35D1_RESET_TMR9] =    {0x2C, 5},
+	[MA35D1_RESET_TMR10] =   {0x2C, 6},
+	[MA35D1_RESET_TMR11] =   {0x2C, 7},
+	[MA35D1_RESET_UART8] =   {0x2C, 8},
+	[MA35D1_RESET_UART9] =   {0x2C, 9},
+	[MA35D1_RESET_UART10] =  {0x2C, 10},
+	[MA35D1_RESET_UART11] =  {0x2C, 11},
+	[MA35D1_RESET_UART12] =  {0x2C, 12},
+	[MA35D1_RESET_UART13] =  {0x2C, 13},
+	[MA35D1_RESET_UART14] =  {0x2C, 14},
+	[MA35D1_RESET_UART15] =  {0x2C, 15},
+	[MA35D1_RESET_UART16] =  {0x2C, 16},
+	[MA35D1_RESET_I2S1] =    {0x2C, 17},
+	[MA35D1_RESET_I2C4] =    {0x2C, 18},
+	[MA35D1_RESET_I2C5] =    {0x2C, 19},
+	[MA35D1_RESET_EPWM2] =   {0x2C, 20},
+	[MA35D1_RESET_ECAP2] =   {0x2C, 21},
+	[MA35D1_RESET_QEI2] =    {0x2C, 22},
+	[MA35D1_RESET_CANFD3] =  {0x2C, 23},
+	[MA35D1_RESET_KPI] =     {0x2C, 24},
+	[MA35D1_RESET_GIC] =     {0x2C, 28},
+	[MA35D1_RESET_SSMCC] =   {0x2C, 30},
+	[MA35D1_RESET_SSPCC] =   {0x2C, 31}
+};
+
+static int ma35d1_restart_handler(struct sys_off_data *sys_off_data)
+{
+	struct ma35d1_reset_data *data = sys_off_data->cb_data;
+	u32 id = MA35D1_RESET_CHIP;
+
+	writel_relaxed(BIT(ma35d1_reset_map[id].bit),
+		       data->base + ma35d1_reset_map[id].reg_ofs);
+	return 0;
+}
+
+static int ma35d1_reset_update(struct reset_controller_dev *rcdev, unsigned long id, bool assert)
+{
+	struct ma35d1_reset_data *data = container_of(rcdev, struct ma35d1_reset_data, rcdev);
+	unsigned long flags;
+	u32 reg;
+
+	if (WARN_ON_ONCE(id >= ARRAY_SIZE(ma35d1_reset_map)))
+		return -EINVAL;
+
+	spin_lock_irqsave(&data->lock, flags);
+	reg = readl_relaxed(data->base + ma35d1_reset_map[id].reg_ofs);
+	if (assert)
+		reg |= BIT(ma35d1_reset_map[id].bit);
+	else
+		reg &= ~(BIT(ma35d1_reset_map[id].bit));
+	writel_relaxed(reg, data->base + ma35d1_reset_map[id].reg_ofs);
+	spin_unlock_irqrestore(&data->lock, flags);
+
+	return 0;
+}
+
+static int ma35d1_reset_assert(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	return ma35d1_reset_update(rcdev, id, true);
+}
+
+static int ma35d1_reset_deassert(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	return ma35d1_reset_update(rcdev, id, false);
+}
+
+static int ma35d1_reset_status(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	struct ma35d1_reset_data *data = container_of(rcdev, struct ma35d1_reset_data, rcdev);
+	u32 reg;
+
+	if (WARN_ON_ONCE(id >= ARRAY_SIZE(ma35d1_reset_map)))
+		return -EINVAL;
+
+	reg = readl_relaxed(data->base + ma35d1_reset_map[id].reg_ofs);
+	return !!(reg & BIT(ma35d1_reset_map[id].bit));
+}
+
+static const struct reset_control_ops ma35d1_reset_ops = {
+	.assert = ma35d1_reset_assert,
+	.deassert = ma35d1_reset_deassert,
+	.status = ma35d1_reset_status,
+};
+
+static const struct of_device_id ma35d1_reset_dt_ids[] = {
+	{ .compatible = "nuvoton,ma35d1-reset" },
+	{ },
+};
+
+static int ma35d1_reset_probe(struct platform_device *pdev)
+{
+	struct ma35d1_reset_data *reset_data;
+	struct device *dev = &pdev->dev;
+	int err;
+
+	if (!pdev->dev.of_node) {
+		dev_err(&pdev->dev, "Device tree node not found\n");
+		return -EINVAL;
+	}
+
+	reset_data = devm_kzalloc(dev, sizeof(*reset_data), GFP_KERNEL);
+	if (!reset_data)
+		return -ENOMEM;
+
+	reset_data->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(reset_data->base))
+		return PTR_ERR(reset_data->base);
+
+	reset_data->rcdev.owner = THIS_MODULE;
+	reset_data->rcdev.nr_resets = MA35D1_RESET_COUNT;
+	reset_data->rcdev.ops = &ma35d1_reset_ops;
+	reset_data->rcdev.of_node = dev->of_node;
+	spin_lock_init(&reset_data->lock);
+
+	err = devm_register_sys_off_handler(dev, SYS_OFF_MODE_RESTART, 192,
+					    ma35d1_restart_handler, reset_data);
+	if (err)
+		dev_warn(&pdev->dev, "failed to register restart handler\n");
+
+	return devm_reset_controller_register(dev, &reset_data->rcdev);
+}
+
+static struct platform_driver ma35d1_reset_driver = {
+	.probe = ma35d1_reset_probe,
+	.driver = {
+		.name = "ma35d1-reset",
+		.of_match_table	= ma35d1_reset_dt_ids,
+	},
+};
+
+builtin_platform_driver(ma35d1_reset_driver);