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, 123 insertions, 0 deletions

diff --git a/drivers/mfd/vx855.c b/drivers/mfd/vx855.c
new file mode 100644
index 000000000..985f81c17
--- /dev/null
+++ b/drivers/mfd/vx855.c
@@ -0,0 +1,123 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Linux multi-function-device driver (MFD) for the integrated peripherals
+ * of the VIA VX855 chipset
+ *
+ * Copyright (C) 2009 VIA Technologies, Inc.
+ * Copyright (C) 2010 One Laptop per Child
+ * Author: Harald Welte <HaraldWelte@viatech.com>
+ * All rights reserved.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/platform_device.h>
+#include <linux/pci.h>
+#include <linux/mfd/core.h>
+
+/* offset into pci config space indicating the 16bit register containing
+ * the power management IO space base */
+#define VX855_CFG_PMIO_OFFSET	0x88
+
+/* ACPI I/O Space registers */
+#define VX855_PMIO_ACPI		0x00
+#define VX855_PMIO_ACPI_LEN	0x0b
+
+/* Processor Power Management */
+#define VX855_PMIO_PPM		0x10
+#define VX855_PMIO_PPM_LEN	0x08
+
+/* General Purpose Power Management */
+#define VX855_PMIO_GPPM		0x20
+#define VX855_PMIO_R_GPI	0x48
+#define VX855_PMIO_R_GPO	0x4c
+#define VX855_PMIO_GPPM_LEN	0x33
+
+#define VSPIC_MMIO_SIZE	0x1000
+
+static struct resource vx855_gpio_resources[] = {
+	{
+		.flags = IORESOURCE_IO,
+	},
+	{
+		.flags = IORESOURCE_IO,
+	},
+};
+
+static const struct mfd_cell vx855_cells[] = {
+	{
+		.name = "vx855_gpio",
+		.num_resources = ARRAY_SIZE(vx855_gpio_resources),
+		.resources = vx855_gpio_resources,
+
+		/* we must ignore resource conflicts, for reasons outlined in
+		 * the vx855_gpio driver */
+		.ignore_resource_conflicts = true,
+	},
+};
+
+static int vx855_probe(struct pci_dev *pdev,
+				 const struct pci_device_id *id)
+{
+	int ret;
+	u16 gpio_io_offset;
+
+	ret = pci_enable_device(pdev);
+	if (ret)
+		return -ENODEV;
+
+	pci_read_config_word(pdev, VX855_CFG_PMIO_OFFSET, &gpio_io_offset);
+	if (!gpio_io_offset) {
+		dev_warn(&pdev->dev,
+			"BIOS did not assign PMIO base offset?!?\n");
+		ret = -ENODEV;
+		goto out;
+	}
+
+	/* mask out the lowest seven bits, as they are always zero, but
+	 * hardware returns them as 0x01 */
+	gpio_io_offset &= 0xff80;
+
+	/* As the region identified here includes many non-GPIO things, we
+	 * only work with the specific registers that concern us. */
+	vx855_gpio_resources[0].start = gpio_io_offset + VX855_PMIO_R_GPI;
+	vx855_gpio_resources[0].end = vx855_gpio_resources[0].start + 3;
+	vx855_gpio_resources[1].start = gpio_io_offset + VX855_PMIO_R_GPO;
+	vx855_gpio_resources[1].end = vx855_gpio_resources[1].start + 3;
+
+	ret = mfd_add_devices(&pdev->dev, -1, vx855_cells, ARRAY_SIZE(vx855_cells),
+			NULL, 0, NULL);
+
+	/* we always return -ENODEV here in order to enable other
+	 * drivers like old, not-yet-platform_device ported i2c-viapro */
+	return -ENODEV;
+out:
+	pci_disable_device(pdev);
+	return ret;
+}
+
+static void vx855_remove(struct pci_dev *pdev)
+{
+	mfd_remove_devices(&pdev->dev);
+	pci_disable_device(pdev);
+}
+
+static const struct pci_device_id vx855_pci_tbl[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VX855) },
+	{ 0, }
+};
+MODULE_DEVICE_TABLE(pci, vx855_pci_tbl);
+
+static struct pci_driver vx855_pci_driver = {
+	.name		= "vx855",
+	.id_table	= vx855_pci_tbl,
+	.probe		= vx855_probe,
+	.remove		= vx855_remove,
+};
+
+module_pci_driver(vx855_pci_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Harald Welte <HaraldWelte@viatech.com>");
+MODULE_DESCRIPTION("Driver for the VIA VX855 chipset");