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

3 files changed, 150 insertions, 0 deletions

diff --git a/drivers/soc/litex/Kconfig b/drivers/soc/litex/Kconfig
new file mode 100644
index 000000000..f3f869639
--- /dev/null
+++ b/drivers/soc/litex/Kconfig
@@ -0,0 +1,20 @@
+# SPDX-License_Identifier: GPL-2.0
+
+menu "Enable LiteX SoC Builder specific drivers"
+
+config LITEX
+	bool
+
+config LITEX_SOC_CONTROLLER
+	tristate "Enable LiteX SoC Controller driver"
+	depends on OF
+	depends on HAS_IOMEM
+	select LITEX
+	help
+	  This option enables the SoC Controller Driver which verifies
+	  LiteX CSR access and provides common litex_[read|write]*
+	  accessors.
+	  All drivers that use functions from litex.h must depend on
+	  LITEX.
+
+endmenu
diff --git a/drivers/soc/litex/Makefile b/drivers/soc/litex/Makefile
new file mode 100644
index 000000000..98ff7325b
--- /dev/null
+++ b/drivers/soc/litex/Makefile
@@ -0,0 +1,3 @@
+# SPDX-License_Identifier: GPL-2.0
+
+obj-$(CONFIG_LITEX_SOC_CONTROLLER)	+= litex_soc_ctrl.o
diff --git a/drivers/soc/litex/litex_soc_ctrl.c b/drivers/soc/litex/litex_soc_ctrl.c
new file mode 100644
index 000000000..104a5f9bf
--- /dev/null
+++ b/drivers/soc/litex/litex_soc_ctrl.c
@@ -0,0 +1,127 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * LiteX SoC Controller Driver
+ *
+ * Copyright (C) 2020 Antmicro <www.antmicro.com>
+ *
+ */
+
+#include <linux/litex.h>
+#include <linux/device.h>
+#include <linux/errno.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/printk.h>
+#include <linux/module.h>
+#include <linux/io.h>
+#include <linux/reboot.h>
+
+/* reset register located at the base address */
+#define RESET_REG_OFF           0x00
+#define RESET_REG_VALUE         0x00000001
+
+#define SCRATCH_REG_OFF         0x04
+#define SCRATCH_REG_VALUE       0x12345678
+#define SCRATCH_TEST_VALUE      0xdeadbeef
+
+/*
+ * Check LiteX CSR read/write access
+ *
+ * This function reads and writes a scratch register in order to verify if CSR
+ * access works.
+ *
+ * In case any problems are detected, the driver should panic.
+ *
+ * Access to the LiteX CSR is, by design, done in CPU native endianness.
+ * The driver should not dynamically configure access functions when
+ * the endianness mismatch is detected. Such situation indicates problems in
+ * the soft SoC design and should be solved at the LiteX generator level,
+ * not in the software.
+ */
+static int litex_check_csr_access(void __iomem *reg_addr)
+{
+	unsigned long reg;
+
+	reg = litex_read32(reg_addr + SCRATCH_REG_OFF);
+
+	if (reg != SCRATCH_REG_VALUE) {
+		panic("Scratch register read error - the system is probably broken! Expected: 0x%x but got: 0x%lx",
+			SCRATCH_REG_VALUE, reg);
+		return -EINVAL;
+	}
+
+	litex_write32(reg_addr + SCRATCH_REG_OFF, SCRATCH_TEST_VALUE);
+	reg = litex_read32(reg_addr + SCRATCH_REG_OFF);
+
+	if (reg != SCRATCH_TEST_VALUE) {
+		panic("Scratch register write error - the system is probably broken! Expected: 0x%x but got: 0x%lx",
+			SCRATCH_TEST_VALUE, reg);
+		return -EINVAL;
+	}
+
+	/* restore original value of the SCRATCH register */
+	litex_write32(reg_addr + SCRATCH_REG_OFF, SCRATCH_REG_VALUE);
+
+	pr_info("LiteX SoC Controller driver initialized");
+
+	return 0;
+}
+
+struct litex_soc_ctrl_device {
+	void __iomem *base;
+};
+
+static int litex_reset_handler(struct sys_off_data *data)
+{
+	struct litex_soc_ctrl_device *soc_ctrl_dev = data->cb_data;
+
+	litex_write32(soc_ctrl_dev->base + RESET_REG_OFF, RESET_REG_VALUE);
+	return NOTIFY_DONE;
+}
+
+static const struct of_device_id litex_soc_ctrl_of_match[] = {
+	{.compatible = "litex,soc-controller"},
+	{},
+};
+MODULE_DEVICE_TABLE(of, litex_soc_ctrl_of_match);
+
+static int litex_soc_ctrl_probe(struct platform_device *pdev)
+{
+	struct litex_soc_ctrl_device *soc_ctrl_dev;
+	int error;
+
+	soc_ctrl_dev = devm_kzalloc(&pdev->dev, sizeof(*soc_ctrl_dev), GFP_KERNEL);
+	if (!soc_ctrl_dev)
+		return -ENOMEM;
+
+	soc_ctrl_dev->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(soc_ctrl_dev->base))
+		return PTR_ERR(soc_ctrl_dev->base);
+
+	error = litex_check_csr_access(soc_ctrl_dev->base);
+	if (error)
+		return error;
+
+	error = devm_register_restart_handler(&pdev->dev,
+					      litex_reset_handler,
+					      soc_ctrl_dev);
+	if (error) {
+		dev_warn(&pdev->dev, "cannot register restart handler: %d\n",
+			 error);
+	}
+
+	return 0;
+}
+
+static struct platform_driver litex_soc_ctrl_driver = {
+	.driver = {
+		.name = "litex-soc-controller",
+		.of_match_table = litex_soc_ctrl_of_match,
+	},
+	.probe = litex_soc_ctrl_probe,
+};
+
+module_platform_driver(litex_soc_ctrl_driver);
+MODULE_DESCRIPTION("LiteX SoC Controller driver");
+MODULE_AUTHOR("Antmicro <www.antmicro.com>");
+MODULE_LICENSE("GPL v2");