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, 161 insertions, 0 deletions

diff --git a/drivers/soc/tegra/common.c b/drivers/soc/tegra/common.c
new file mode 100644
index 000000000..0864eb018
--- /dev/null
+++ b/drivers/soc/tegra/common.c
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2014 NVIDIA CORPORATION.  All rights reserved.
+ */
+
+#define dev_fmt(fmt)	"tegra-soc: " fmt
+
+#include <linux/clk.h>
+#include <linux/device.h>
+#include <linux/export.h>
+#include <linux/of.h>
+#include <linux/pm_opp.h>
+#include <linux/pm_runtime.h>
+
+#include <soc/tegra/common.h>
+#include <soc/tegra/fuse.h>
+
+static const struct of_device_id tegra_machine_match[] = {
+	{ .compatible = "nvidia,tegra20", },
+	{ .compatible = "nvidia,tegra30", },
+	{ .compatible = "nvidia,tegra114", },
+	{ .compatible = "nvidia,tegra124", },
+	{ .compatible = "nvidia,tegra132", },
+	{ .compatible = "nvidia,tegra210", },
+	{ }
+};
+
+bool soc_is_tegra(void)
+{
+	return of_machine_device_match(tegra_machine_match);
+}
+
+static int tegra_core_dev_init_opp_state(struct device *dev)
+{
+	unsigned long rate;
+	struct clk *clk;
+	bool rpm_enabled;
+	int err;
+
+	clk = devm_clk_get(dev, NULL);
+	if (IS_ERR(clk)) {
+		dev_err(dev, "failed to get clk: %pe\n", clk);
+		return PTR_ERR(clk);
+	}
+
+	rate = clk_get_rate(clk);
+	if (!rate) {
+		dev_err(dev, "failed to get clk rate\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * Runtime PM of the device must be enabled in order to set up
+	 * GENPD's performance properly because GENPD core checks whether
+	 * device is suspended and this check doesn't work while RPM is
+	 * disabled. This makes sure the OPP vote below gets cached in
+	 * GENPD for the device. Instead, the vote is done the next time
+	 * the device gets runtime resumed.
+	 */
+	rpm_enabled = pm_runtime_enabled(dev);
+	if (!rpm_enabled)
+		pm_runtime_enable(dev);
+
+	/* should never happen in practice */
+	if (!pm_runtime_enabled(dev)) {
+		dev_WARN(dev, "failed to enable runtime PM\n");
+		pm_runtime_disable(dev);
+		return -EINVAL;
+	}
+
+	/* first dummy rate-setting initializes voltage vote */
+	err = dev_pm_opp_set_rate(dev, rate);
+
+	if (!rpm_enabled)
+		pm_runtime_disable(dev);
+
+	if (err) {
+		dev_err(dev, "failed to initialize OPP clock: %d\n", err);
+		return err;
+	}
+
+	return 0;
+}
+
+/**
+ * devm_tegra_core_dev_init_opp_table() - initialize OPP table
+ * @dev: device for which OPP table is initialized
+ * @params: pointer to the OPP table configuration
+ *
+ * This function will initialize OPP table and sync OPP state of a Tegra SoC
+ * core device.
+ *
+ * Return: 0 on success or errorno.
+ */
+int devm_tegra_core_dev_init_opp_table(struct device *dev,
+				       struct tegra_core_opp_params *params)
+{
+	u32 hw_version;
+	int err;
+	/*
+	 * The clk's connection id to set is NULL and this is a NULL terminated
+	 * array, hence two NULL entries.
+	 */
+	const char *clk_names[] = { NULL, NULL };
+	struct dev_pm_opp_config config = {
+		/*
+		 * For some devices we don't have any OPP table in the DT, and
+		 * in order to use the same code path for all the devices, we
+		 * create a dummy OPP table for them via this. The dummy OPP
+		 * table is only capable of doing clk_set_rate() on invocation
+		 * of dev_pm_opp_set_rate() and doesn't provide any other
+		 * functionality.
+		 */
+		.clk_names = clk_names,
+	};
+
+	if (of_machine_is_compatible("nvidia,tegra20")) {
+		hw_version = BIT(tegra_sku_info.soc_process_id);
+		config.supported_hw = &hw_version;
+		config.supported_hw_count = 1;
+	} else if (of_machine_is_compatible("nvidia,tegra30") ||
+		   of_machine_is_compatible("nvidia,tegra114")) {
+		hw_version = BIT(tegra_sku_info.soc_speedo_id);
+		config.supported_hw = &hw_version;
+		config.supported_hw_count = 1;
+	}
+
+	err = devm_pm_opp_set_config(dev, &config);
+	if (err) {
+		dev_err(dev, "failed to set OPP config: %d\n", err);
+		return err;
+	}
+
+	/*
+	 * Tegra124+ doesn't support OPP yet, return early for pre-Tegra124
+	 * case.
+	 */
+	if (!config.supported_hw)
+		return -ENODEV;
+
+	/*
+	 * Older device-trees have an empty OPP table, we will get
+	 * -ENODEV from devm_pm_opp_of_add_table() in this case.
+	 */
+	err = devm_pm_opp_of_add_table(dev);
+	if (err) {
+		if (err != -ENODEV)
+			dev_err(dev, "failed to add OPP table: %d\n", err);
+
+		return err;
+	}
+
+	if (params->init_state) {
+		err = tegra_core_dev_init_opp_state(dev);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(devm_tegra_core_dev_init_opp_table);