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, 226 insertions, 0 deletions

diff --git a/drivers/hwmon/k8temp.c b/drivers/hwmon/k8temp.c
new file mode 100644
index 000000000..532411645
--- /dev/null
+++ b/drivers/hwmon/k8temp.c
@@ -0,0 +1,226 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * k8temp.c - Linux kernel module for hardware monitoring
+ *
+ * Copyright (C) 2006 Rudolf Marek <r.marek@assembler.cz>
+ *
+ * Inspired from the w83785 and amd756 drivers.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/pci.h>
+#include <linux/hwmon.h>
+#include <linux/err.h>
+#include <linux/mutex.h>
+#include <asm/processor.h>
+#include <asm/cpuid/api.h>
+
+#define TEMP_FROM_REG(val)	(((((val) >> 16) & 0xff) - 49) * 1000)
+#define REG_TEMP	0xe4
+#define SEL_PLACE	0x40
+#define SEL_CORE	0x04
+
+struct k8temp_data {
+	struct mutex update_lock;
+
+	/* registers values */
+	u8 sensorsp;		/* sensor presence bits - SEL_CORE, SEL_PLACE */
+	u8 swap_core_select;    /* meaning of SEL_CORE is inverted */
+	u32 temp_offset;
+};
+
+static const struct pci_device_id k8temp_ids[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB_MISC) },
+	{ 0 },
+};
+MODULE_DEVICE_TABLE(pci, k8temp_ids);
+
+static int is_rev_g_desktop(u8 model)
+{
+	u32 brandidx;
+
+	if (model < 0x69)
+		return 0;
+
+	if (model == 0xc1 || model == 0x6c || model == 0x7c)
+		return 0;
+
+	/*
+	 * Differentiate between AM2 and ASB1.
+	 * See "Constructing the processor Name String" in "Revision
+	 * Guide for AMD NPT Family 0Fh Processors" (33610).
+	 */
+	brandidx = cpuid_ebx(0x80000001);
+	brandidx = (brandidx >> 9) & 0x1f;
+
+	/* Single core */
+	if ((model == 0x6f || model == 0x7f) &&
+	    (brandidx == 0x7 || brandidx == 0x9 || brandidx == 0xc))
+		return 0;
+
+	/* Dual core */
+	if (model == 0x6b &&
+	    (brandidx == 0xb || brandidx == 0xc))
+		return 0;
+
+	return 1;
+}
+
+static umode_t
+k8temp_is_visible(const void *drvdata, enum hwmon_sensor_types type,
+		  u32 attr, int channel)
+{
+	const struct k8temp_data *data = drvdata;
+
+	if ((channel & 1) && !(data->sensorsp & SEL_PLACE))
+		return 0;
+
+	if ((channel & 2) && !(data->sensorsp & SEL_CORE))
+		return 0;
+
+	return 0444;
+}
+
+static int
+k8temp_read(struct device *dev, enum hwmon_sensor_types type,
+	    u32 attr, int channel, long *val)
+{
+	struct k8temp_data *data = dev_get_drvdata(dev);
+	struct pci_dev *pdev = to_pci_dev(dev->parent);
+	int core, place;
+	u32 temp;
+	u8 tmp;
+
+	core = (channel >> 1) & 1;
+	place = channel & 1;
+
+	core ^= data->swap_core_select;
+
+	mutex_lock(&data->update_lock);
+	pci_read_config_byte(pdev, REG_TEMP, &tmp);
+	tmp &= ~(SEL_PLACE | SEL_CORE);
+	if (core)
+		tmp |= SEL_CORE;
+	if (place)
+		tmp |= SEL_PLACE;
+	pci_write_config_byte(pdev, REG_TEMP, tmp);
+	pci_read_config_dword(pdev, REG_TEMP, &temp);
+	mutex_unlock(&data->update_lock);
+
+	*val = TEMP_FROM_REG(temp) + data->temp_offset;
+
+	return 0;
+}
+
+static const struct hwmon_ops k8temp_ops = {
+	.is_visible = k8temp_is_visible,
+	.read = k8temp_read,
+};
+
+static const struct hwmon_channel_info * const k8temp_info[] = {
+	HWMON_CHANNEL_INFO(temp,
+		HWMON_T_INPUT, HWMON_T_INPUT, HWMON_T_INPUT, HWMON_T_INPUT),
+	NULL
+};
+
+static const struct hwmon_chip_info k8temp_chip_info = {
+	.ops = &k8temp_ops,
+	.info = k8temp_info,
+};
+
+static int k8temp_probe(struct pci_dev *pdev,
+				  const struct pci_device_id *id)
+{
+	u8 scfg;
+	u32 temp;
+	u8 model, stepping;
+	struct k8temp_data *data;
+	struct device *hwmon_dev;
+
+	data = devm_kzalloc(&pdev->dev, sizeof(struct k8temp_data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	model = boot_cpu_data.x86_model;
+	stepping = boot_cpu_data.x86_stepping;
+
+	/* feature available since SH-C0, exclude older revisions */
+	if ((model == 4 && stepping == 0) ||
+	    (model == 5 && stepping <= 1))
+		return -ENODEV;
+
+	/*
+	 * AMD NPT family 0fh, i.e. RevF and RevG:
+	 * meaning of SEL_CORE bit is inverted
+	 */
+	if (model >= 0x40) {
+		data->swap_core_select = 1;
+		dev_warn(&pdev->dev,
+			 "Temperature readouts might be wrong - check erratum #141\n");
+	}
+
+	/*
+	 * RevG desktop CPUs (i.e. no socket S1G1 or ASB1 parts) need
+	 * additional offset, otherwise reported temperature is below
+	 * ambient temperature
+	 */
+	if (is_rev_g_desktop(model))
+		data->temp_offset = 21000;
+
+	pci_read_config_byte(pdev, REG_TEMP, &scfg);
+	scfg &= ~(SEL_PLACE | SEL_CORE);	/* Select sensor 0, core0 */
+	pci_write_config_byte(pdev, REG_TEMP, scfg);
+	pci_read_config_byte(pdev, REG_TEMP, &scfg);
+
+	if (scfg & (SEL_PLACE | SEL_CORE)) {
+		dev_err(&pdev->dev, "Configuration bit(s) stuck at 1!\n");
+		return -ENODEV;
+	}
+
+	scfg |= (SEL_PLACE | SEL_CORE);
+	pci_write_config_byte(pdev, REG_TEMP, scfg);
+
+	/* now we know if we can change core and/or sensor */
+	pci_read_config_byte(pdev, REG_TEMP, &data->sensorsp);
+
+	if (data->sensorsp & SEL_PLACE) {
+		scfg &= ~SEL_CORE;	/* Select sensor 1, core0 */
+		pci_write_config_byte(pdev, REG_TEMP, scfg);
+		pci_read_config_dword(pdev, REG_TEMP, &temp);
+		scfg |= SEL_CORE;	/* prepare for next selection */
+		if (!((temp >> 16) & 0xff)) /* if temp is 0 -49C is unlikely */
+			data->sensorsp &= ~SEL_PLACE;
+	}
+
+	if (data->sensorsp & SEL_CORE) {
+		scfg &= ~SEL_PLACE;	/* Select sensor 0, core1 */
+		pci_write_config_byte(pdev, REG_TEMP, scfg);
+		pci_read_config_dword(pdev, REG_TEMP, &temp);
+		if (!((temp >> 16) & 0xff)) /* if temp is 0 -49C is unlikely */
+			data->sensorsp &= ~SEL_CORE;
+	}
+
+	mutex_init(&data->update_lock);
+
+	hwmon_dev = devm_hwmon_device_register_with_info(&pdev->dev,
+							 "k8temp",
+							 data,
+							 &k8temp_chip_info,
+							 NULL);
+
+	return PTR_ERR_OR_ZERO(hwmon_dev);
+}
+
+static struct pci_driver k8temp_driver = {
+	.name = "k8temp",
+	.id_table = k8temp_ids,
+	.probe = k8temp_probe,
+};
+
+module_pci_driver(k8temp_driver);
+
+MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
+MODULE_DESCRIPTION("AMD K8 core temperature monitor");
+MODULE_LICENSE("GPL");