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, 207 insertions, 0 deletions

diff --git a/drivers/reset/reset-mpfs.c b/drivers/reset/reset-mpfs.c
new file mode 100644
index 000000000..8ffcc54ee
--- /dev/null
+++ b/drivers/reset/reset-mpfs.c
@@ -0,0 +1,207 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * PolarFire SoC (MPFS) Peripheral Clock Reset Controller
+ *
+ * Author: Conor Dooley <conor.dooley@microchip.com>
+ * Copyright (c) 2022 Microchip Technology Inc. and its subsidiaries.
+ *
+ */
+#include <linux/auxiliary_bus.h>
+#include <linux/delay.h>
+#include <linux/io.h>
+#include <linux/mfd/syscon.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/regmap.h>
+#include <linux/reset-controller.h>
+#include <linux/slab.h>
+#include <dt-bindings/clock/microchip,mpfs-clock.h>
+#include <soc/microchip/mpfs.h>
+
+/*
+ * The ENVM reset is the lowest bit in the register & I am using the CLK_FOO
+ * defines in the dt to make things easier to configure - so this is accounting
+ * for the offset of 3 there.
+ */
+#define MPFS_PERIPH_OFFSET	CLK_ENVM
+#define MPFS_NUM_RESETS		30u
+#define MPFS_SLEEP_MIN_US	100
+#define MPFS_SLEEP_MAX_US	200
+
+#define REG_SUBBLK_RESET_CR	0x88u
+
+struct mpfs_reset {
+	struct regmap *regmap;
+	struct reset_controller_dev rcdev;
+};
+
+static inline struct mpfs_reset *to_mpfs_reset(struct reset_controller_dev *rcdev)
+{
+	return container_of(rcdev, struct mpfs_reset, rcdev);
+}
+
+/*
+ * Peripheral clock resets
+ */
+static int mpfs_assert(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	struct mpfs_reset *rst = to_mpfs_reset(rcdev);
+
+	return regmap_set_bits(rst->regmap, REG_SUBBLK_RESET_CR, BIT(id));
+
+}
+
+static int mpfs_deassert(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	struct mpfs_reset *rst = to_mpfs_reset(rcdev);
+
+	return regmap_clear_bits(rst->regmap, REG_SUBBLK_RESET_CR, BIT(id));
+
+}
+
+static int mpfs_status(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	struct mpfs_reset *rst = to_mpfs_reset(rcdev);
+	u32 reg;
+
+	regmap_read(rst->regmap, REG_SUBBLK_RESET_CR, &reg);
+
+	/*
+	 * It is safe to return here as MPFS_NUM_RESETS makes sure the sign bit
+	 * is never hit.
+	 */
+	return (reg & BIT(id));
+}
+
+static int mpfs_reset(struct reset_controller_dev *rcdev, unsigned long id)
+{
+	mpfs_assert(rcdev, id);
+
+	usleep_range(MPFS_SLEEP_MIN_US, MPFS_SLEEP_MAX_US);
+
+	mpfs_deassert(rcdev, id);
+
+	return 0;
+}
+
+static const struct reset_control_ops mpfs_reset_ops = {
+	.reset = mpfs_reset,
+	.assert = mpfs_assert,
+	.deassert = mpfs_deassert,
+	.status = mpfs_status,
+};
+
+static int mpfs_reset_xlate(struct reset_controller_dev *rcdev,
+			    const struct of_phandle_args *reset_spec)
+{
+	unsigned int index = reset_spec->args[0];
+
+	/*
+	 * CLK_RESERVED does not map to a clock, but it does map to a reset,
+	 * so it has to be accounted for here. It is the reset for the fabric,
+	 * so if this reset gets called - do not reset it.
+	 */
+	if (index == CLK_RESERVED) {
+		dev_err(rcdev->dev, "Resetting the fabric is not supported\n");
+		return -EINVAL;
+	}
+
+	if (index < MPFS_PERIPH_OFFSET || index >= (MPFS_PERIPH_OFFSET + rcdev->nr_resets)) {
+		dev_err(rcdev->dev, "Invalid reset index %u\n", index);
+		return -EINVAL;
+	}
+
+	return index - MPFS_PERIPH_OFFSET;
+}
+
+static int mpfs_reset_mfd_probe(struct platform_device *pdev)
+{
+	struct reset_controller_dev *rcdev;
+	struct device *dev = &pdev->dev;
+	struct mpfs_reset *rst;
+
+	rst = devm_kzalloc(dev, sizeof(*rst), GFP_KERNEL);
+	if (!rst)
+		return -ENOMEM;
+
+	rcdev = &rst->rcdev;
+	rcdev->dev = dev;
+	rcdev->ops = &mpfs_reset_ops;
+
+	rcdev->of_node = pdev->dev.parent->of_node;
+	rcdev->of_reset_n_cells = 1;
+	rcdev->of_xlate = mpfs_reset_xlate;
+	rcdev->nr_resets = MPFS_NUM_RESETS;
+
+	rst->regmap = device_node_to_regmap(pdev->dev.parent->of_node);
+	if (IS_ERR(rst->regmap))
+		return dev_err_probe(dev, PTR_ERR(rst->regmap),
+				     "Failed to find syscon regmap\n");
+
+	return devm_reset_controller_register(dev, rcdev);
+}
+
+static struct platform_driver mpfs_reset_mfd_driver = {
+	.probe = mpfs_reset_mfd_probe,
+	.driver = {
+		.name = "mpfs-reset",
+	},
+};
+module_platform_driver(mpfs_reset_mfd_driver);
+
+static int mpfs_reset_adev_probe(struct auxiliary_device *adev,
+				 const struct auxiliary_device_id *id)
+{
+	struct reset_controller_dev *rcdev;
+	struct device *dev = &adev->dev;
+	struct mpfs_reset *rst;
+
+	rst = devm_kzalloc(dev, sizeof(*rst), GFP_KERNEL);
+	if (!rst)
+		return -ENOMEM;
+
+	rst->regmap = (struct regmap *)adev->dev.platform_data;
+
+	rcdev = &rst->rcdev;
+	rcdev->dev = dev;
+	rcdev->ops = &mpfs_reset_ops;
+
+	rcdev->of_node = dev->parent->of_node;
+	rcdev->of_reset_n_cells = 1;
+	rcdev->of_xlate = mpfs_reset_xlate;
+	rcdev->nr_resets = MPFS_NUM_RESETS;
+
+	return devm_reset_controller_register(dev, rcdev);
+}
+
+int mpfs_reset_controller_register(struct device *clk_dev, struct regmap *map)
+{
+	struct auxiliary_device *adev;
+
+	adev = devm_auxiliary_device_create(clk_dev, "reset-mpfs", (void *)map);
+	if (!adev)
+		return -ENODEV;
+
+	return 0;
+}
+EXPORT_SYMBOL_NS_GPL(mpfs_reset_controller_register, "MCHP_CLK_MPFS");
+
+static const struct auxiliary_device_id mpfs_reset_ids[] = {
+	{
+		.name = "reset_mpfs.reset-mpfs",
+	},
+	{ }
+};
+MODULE_DEVICE_TABLE(auxiliary, mpfs_reset_ids);
+
+static struct auxiliary_driver mpfs_reset_aux_driver = {
+	.probe		= mpfs_reset_adev_probe,
+	.id_table	= mpfs_reset_ids,
+};
+
+module_auxiliary_driver(mpfs_reset_aux_driver);
+
+MODULE_DESCRIPTION("Microchip PolarFire SoC Reset Driver");
+MODULE_AUTHOR("Conor Dooley <conor.dooley@microchip.com>");
+MODULE_IMPORT_NS("MCHP_CLK_MPFS");