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, 205 insertions, 0 deletions

diff --git a/drivers/nvmem/layouts.c b/drivers/nvmem/layouts.c
new file mode 100644
index 000000000..07a34be96
--- /dev/null
+++ b/drivers/nvmem/layouts.c
@@ -0,0 +1,205 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * NVMEM layout bus handling
+ *
+ * Copyright (C) 2023 Bootlin
+ * Author: Miquel Raynal <miquel.raynal@bootlin.com
+ */
+
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <linux/nvmem-consumer.h>
+#include <linux/nvmem-provider.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/of_irq.h>
+
+#include "internals.h"
+
+#define to_nvmem_layout_driver(drv) \
+	(container_of_const((drv), struct nvmem_layout_driver, driver))
+#define to_nvmem_layout_device(_dev) \
+	container_of((_dev), struct nvmem_layout, dev)
+
+static int nvmem_layout_bus_match(struct device *dev, const struct device_driver *drv)
+{
+	return of_driver_match_device(dev, drv);
+}
+
+static int nvmem_layout_bus_probe(struct device *dev)
+{
+	struct nvmem_layout_driver *drv = to_nvmem_layout_driver(dev->driver);
+	struct nvmem_layout *layout = to_nvmem_layout_device(dev);
+
+	if (!drv->probe || !drv->remove)
+		return -EINVAL;
+
+	return drv->probe(layout);
+}
+
+static void nvmem_layout_bus_remove(struct device *dev)
+{
+	struct nvmem_layout_driver *drv = to_nvmem_layout_driver(dev->driver);
+	struct nvmem_layout *layout = to_nvmem_layout_device(dev);
+
+	return drv->remove(layout);
+}
+
+static int nvmem_layout_bus_uevent(const struct device *dev,
+				   struct kobj_uevent_env *env)
+{
+	int ret;
+
+	ret = of_device_uevent_modalias(dev, env);
+	if (ret != -ENODEV)
+		return ret;
+
+	return 0;
+}
+
+static const struct bus_type nvmem_layout_bus_type = {
+	.name		= "nvmem-layout",
+	.match		= nvmem_layout_bus_match,
+	.probe		= nvmem_layout_bus_probe,
+	.remove		= nvmem_layout_bus_remove,
+	.uevent		= nvmem_layout_bus_uevent,
+};
+
+int __nvmem_layout_driver_register(struct nvmem_layout_driver *drv,
+				   struct module *owner)
+{
+	drv->driver.bus = &nvmem_layout_bus_type;
+	drv->driver.owner = owner;
+
+	return driver_register(&drv->driver);
+}
+EXPORT_SYMBOL_GPL(__nvmem_layout_driver_register);
+
+void nvmem_layout_driver_unregister(struct nvmem_layout_driver *drv)
+{
+	driver_unregister(&drv->driver);
+}
+EXPORT_SYMBOL_GPL(nvmem_layout_driver_unregister);
+
+static void nvmem_layout_release_device(struct device *dev)
+{
+	struct nvmem_layout *layout = to_nvmem_layout_device(dev);
+
+	of_node_put(layout->dev.of_node);
+	kfree(layout);
+}
+
+static int nvmem_layout_create_device(struct nvmem_device *nvmem,
+				      struct device_node *np)
+{
+	struct nvmem_layout *layout;
+	struct device *dev;
+	int ret;
+
+	layout = kzalloc_obj(*layout);
+	if (!layout)
+		return -ENOMEM;
+
+	/* Create a bidirectional link */
+	layout->nvmem = nvmem;
+	nvmem->layout = layout;
+
+	/* Device model registration */
+	dev = &layout->dev;
+	device_initialize(dev);
+	dev->parent = &nvmem->dev;
+	dev->bus = &nvmem_layout_bus_type;
+	dev->release = nvmem_layout_release_device;
+	dev->coherent_dma_mask = DMA_BIT_MASK(32);
+	dev->dma_mask = &dev->coherent_dma_mask;
+	device_set_node(dev, of_fwnode_handle(of_node_get(np)));
+	of_device_make_bus_id(dev);
+	of_msi_configure(dev, dev->of_node);
+
+	ret = device_add(dev);
+	if (ret) {
+		put_device(dev);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int nvmem_layout_bus_populate(struct nvmem_device *nvmem,
+				     struct device_node *layout_dn)
+{
+	int ret;
+
+	/* Make sure it has a compatible property */
+	if (!of_property_present(layout_dn, "compatible")) {
+		pr_debug("%s() - skipping %pOF, no compatible prop\n",
+			 __func__, layout_dn);
+		return 0;
+	}
+
+	if (of_node_check_flag(layout_dn, OF_POPULATED_BUS)) {
+		pr_debug("%s() - skipping %pOF, already populated\n",
+			 __func__, layout_dn);
+
+		return 0;
+	}
+
+	/* NVMEM layout buses expect only a single device representing the layout */
+	ret = nvmem_layout_create_device(nvmem, layout_dn);
+	if (ret)
+		return ret;
+
+	of_node_set_flag(layout_dn, OF_POPULATED_BUS);
+
+	return 0;
+}
+
+struct device_node *of_nvmem_layout_get_container(struct nvmem_device *nvmem)
+{
+	return of_get_child_by_name(nvmem->dev.of_node, "nvmem-layout");
+}
+EXPORT_SYMBOL_GPL(of_nvmem_layout_get_container);
+
+/*
+ * Returns the number of devices populated, 0 if the operation was not relevant
+ * for this nvmem device, an error code otherwise.
+ */
+int nvmem_populate_layout(struct nvmem_device *nvmem)
+{
+	struct device_node *layout_dn;
+	int ret;
+
+	layout_dn = of_nvmem_layout_get_container(nvmem);
+	if (!layout_dn)
+		return 0;
+
+	/* Populate the layout device */
+	device_links_supplier_sync_state_pause();
+	ret = nvmem_layout_bus_populate(nvmem, layout_dn);
+	device_links_supplier_sync_state_resume();
+
+	of_node_put(layout_dn);
+	return ret;
+}
+
+void nvmem_destroy_layout(struct nvmem_device *nvmem)
+{
+	struct device *dev;
+
+	if (!nvmem->layout)
+		return;
+
+	dev = &nvmem->layout->dev;
+	of_node_clear_flag(dev->of_node, OF_POPULATED_BUS);
+	device_unregister(dev);
+}
+
+int nvmem_layout_bus_register(void)
+{
+	return bus_register(&nvmem_layout_bus_type);
+}
+
+void nvmem_layout_bus_unregister(void)
+{
+	bus_unregister(&nvmem_layout_bus_type);
+}