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

2 files changed, 157 insertions, 0 deletions

diff --git a/samples/fprobe/Makefile b/samples/fprobe/Makefile
new file mode 100644
index 000000000..ecccbfa6e
--- /dev/null
+++ b/samples/fprobe/Makefile
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+
+obj-$(CONFIG_SAMPLE_FPROBE) += fprobe_example.o
diff --git a/samples/fprobe/fprobe_example.c b/samples/fprobe/fprobe_example.c
new file mode 100644
index 000000000..bfe98ce82
--- /dev/null
+++ b/samples/fprobe/fprobe_example.c
@@ -0,0 +1,154 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Here's a sample kernel module showing the use of fprobe to dump a
+ * stack trace and selected registers when kernel_clone() is called.
+ *
+ * For more information on theory of operation of kprobes, see
+ * Documentation/trace/kprobes.rst
+ *
+ * You will see the trace data in /var/log/messages and on the console
+ * whenever kernel_clone() is invoked to create a new process.
+ */
+
+#define pr_fmt(fmt) "%s: " fmt, __func__
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/fprobe.h>
+#include <linux/sched/debug.h>
+#include <linux/slab.h>
+
+#define BACKTRACE_DEPTH 16
+#define MAX_SYMBOL_LEN 4096
+static struct fprobe sample_probe;
+static unsigned long nhit;
+
+static char symbol[MAX_SYMBOL_LEN] = "kernel_clone";
+module_param_string(symbol, symbol, sizeof(symbol), 0644);
+MODULE_PARM_DESC(symbol, "Probed symbol(s), given by comma separated symbols or a wildcard pattern.");
+
+static char nosymbol[MAX_SYMBOL_LEN] = "";
+module_param_string(nosymbol, nosymbol, sizeof(nosymbol), 0644);
+MODULE_PARM_DESC(nosymbol, "Not-probed symbols, given by a wildcard pattern.");
+
+static bool stackdump = true;
+module_param(stackdump, bool, 0644);
+MODULE_PARM_DESC(stackdump, "Enable stackdump.");
+
+static bool use_trace = false;
+module_param(use_trace, bool, 0644);
+MODULE_PARM_DESC(use_trace, "Use trace_printk instead of printk. This is only for debugging.");
+
+static void show_backtrace(void)
+{
+	unsigned long stacks[BACKTRACE_DEPTH];
+	unsigned int len;
+
+	len = stack_trace_save(stacks, BACKTRACE_DEPTH, 2);
+	stack_trace_print(stacks, len, 24);
+}
+
+static int sample_entry_handler(struct fprobe *fp, unsigned long ip,
+				unsigned long ret_ip,
+				struct ftrace_regs *fregs, void *data)
+{
+	if (use_trace)
+		/*
+		 * This is just an example, no kernel code should call
+		 * trace_printk() except when actively debugging.
+		 */
+		trace_printk("Enter <%pS> ip = 0x%p\n", (void *)ip, (void *)ip);
+	else
+		pr_info("Enter <%pS> ip = 0x%p\n", (void *)ip, (void *)ip);
+	nhit++;
+	if (stackdump)
+		show_backtrace();
+	return 0;
+}
+
+static void sample_exit_handler(struct fprobe *fp, unsigned long ip,
+				unsigned long ret_ip, struct ftrace_regs *regs,
+				void *data)
+{
+	unsigned long rip = ret_ip;
+
+	if (use_trace)
+		/*
+		 * This is just an example, no kernel code should call
+		 * trace_printk() except when actively debugging.
+		 */
+		trace_printk("Return from <%pS> ip = 0x%p to rip = 0x%p (%pS)\n",
+			(void *)ip, (void *)ip, (void *)rip, (void *)rip);
+	else
+		pr_info("Return from <%pS> ip = 0x%p to rip = 0x%p (%pS)\n",
+			(void *)ip, (void *)ip, (void *)rip, (void *)rip);
+	nhit++;
+	if (stackdump)
+		show_backtrace();
+}
+
+static int __init fprobe_init(void)
+{
+	char *p, *symbuf = NULL;
+	const char **syms;
+	int ret, count, i;
+
+	sample_probe.entry_handler = sample_entry_handler;
+	sample_probe.exit_handler = sample_exit_handler;
+
+	if (strchr(symbol, '*')) {
+		/* filter based fprobe */
+		ret = register_fprobe(&sample_probe, symbol,
+				      nosymbol[0] == '\0' ? NULL : nosymbol);
+		goto out;
+	} else if (!strchr(symbol, ',')) {
+		symbuf = symbol;
+		ret = register_fprobe_syms(&sample_probe, (const char **)&symbuf, 1);
+		goto out;
+	}
+
+	/* Comma separated symbols */
+	symbuf = kstrdup(symbol, GFP_KERNEL);
+	if (!symbuf)
+		return -ENOMEM;
+	p = symbuf;
+	count = 1;
+	while ((p = strchr(++p, ',')) != NULL)
+		count++;
+
+	pr_info("%d symbols found\n", count);
+
+	syms = kcalloc(count, sizeof(char *), GFP_KERNEL);
+	if (!syms) {
+		kfree(symbuf);
+		return -ENOMEM;
+	}
+
+	p = symbuf;
+	for (i = 0; i < count; i++)
+		syms[i] = strsep(&p, ",");
+
+	ret = register_fprobe_syms(&sample_probe, syms, count);
+	kfree(syms);
+	kfree(symbuf);
+out:
+	if (ret < 0)
+		pr_err("register_fprobe failed, returned %d\n", ret);
+	else
+		pr_info("Planted fprobe at %s\n", symbol);
+
+	return ret;
+}
+
+static void __exit fprobe_exit(void)
+{
+	unregister_fprobe(&sample_probe);
+
+	pr_info("fprobe at %s unregistered. %ld times hit, %ld times missed\n",
+		symbol, nhit, sample_probe.nmissed);
+}
+
+module_init(fprobe_init)
+module_exit(fprobe_exit)
+MODULE_DESCRIPTION("sample kernel module showing the use of fprobe");
+MODULE_LICENSE("GPL");