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, 249 insertions, 0 deletions

diff --git a/block/ioprio.c b/block/ioprio.c
new file mode 100644
index 000000000..f0ee27985
--- /dev/null
+++ b/block/ioprio.c
@@ -0,0 +1,249 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * fs/ioprio.c
+ *
+ * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
+ *
+ * Helper functions for setting/querying io priorities of processes. The
+ * system calls closely mimmick getpriority/setpriority, see the man page for
+ * those. The prio argument is a composite of prio class and prio data, where
+ * the data argument has meaning within that class. The standard scheduling
+ * classes have 8 distinct prio levels, with 0 being the highest prio and 7
+ * being the lowest.
+ *
+ * IOW, setting BE scheduling class with prio 2 is done ala:
+ *
+ * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
+ *
+ * ioprio_set(PRIO_PROCESS, pid, prio);
+ *
+ * See also Documentation/block/ioprio.rst
+ *
+ */
+#include <linux/gfp.h>
+#include <linux/kernel.h>
+#include <linux/ioprio.h>
+#include <linux/cred.h>
+#include <linux/blkdev.h>
+#include <linux/capability.h>
+#include <linux/syscalls.h>
+#include <linux/security.h>
+#include <linux/pid_namespace.h>
+
+int ioprio_check_cap(int ioprio)
+{
+	int class = IOPRIO_PRIO_CLASS(ioprio);
+	int level = IOPRIO_PRIO_LEVEL(ioprio);
+
+	switch (class) {
+		case IOPRIO_CLASS_RT:
+			/*
+			 * Originally this only checked for CAP_SYS_ADMIN,
+			 * which was implicitly allowed for pid 0 by security
+			 * modules such as SELinux. Make sure we check
+			 * CAP_SYS_ADMIN first to avoid a denial/avc for
+			 * possibly missing CAP_SYS_NICE permission.
+			 */
+			if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
+				return -EPERM;
+			break;
+		case IOPRIO_CLASS_BE:
+		case IOPRIO_CLASS_IDLE:
+			break;
+		case IOPRIO_CLASS_NONE:
+			if (level)
+				return -EINVAL;
+			break;
+		case IOPRIO_CLASS_INVALID:
+		default:
+			return -EINVAL;
+	}
+
+	return 0;
+}
+
+SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
+{
+	struct task_struct *p, *g;
+	struct user_struct *user;
+	struct pid *pgrp;
+	kuid_t uid;
+	int ret;
+
+	ret = ioprio_check_cap(ioprio);
+	if (ret)
+		return ret;
+
+	ret = -ESRCH;
+	rcu_read_lock();
+	switch (which) {
+		case IOPRIO_WHO_PROCESS:
+			if (!who)
+				p = current;
+			else
+				p = find_task_by_vpid(who);
+			if (p)
+				ret = set_task_ioprio(p, ioprio);
+			break;
+		case IOPRIO_WHO_PGRP:
+			if (!who)
+				pgrp = task_pgrp(current);
+			else
+				pgrp = find_vpid(who);
+
+			read_lock(&tasklist_lock);
+			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
+				ret = set_task_ioprio(p, ioprio);
+				if (ret) {
+					read_unlock(&tasklist_lock);
+					goto out;
+				}
+			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
+			read_unlock(&tasklist_lock);
+
+			break;
+		case IOPRIO_WHO_USER:
+			uid = make_kuid(current_user_ns(), who);
+			if (!uid_valid(uid))
+				break;
+			if (!who)
+				user = current_user();
+			else
+				user = find_user(uid);
+
+			if (!user)
+				break;
+
+			for_each_process_thread(g, p) {
+				if (!uid_eq(task_uid(p), uid) ||
+				    !task_pid_vnr(p))
+					continue;
+				ret = set_task_ioprio(p, ioprio);
+				if (ret)
+					goto free_uid;
+			}
+free_uid:
+			if (who)
+				free_uid(user);
+			break;
+		default:
+			ret = -EINVAL;
+	}
+
+out:
+	rcu_read_unlock();
+	return ret;
+}
+
+static int get_task_ioprio(struct task_struct *p)
+{
+	int ret;
+
+	ret = security_task_getioprio(p);
+	if (ret)
+		goto out;
+	task_lock(p);
+	ret = __get_task_ioprio(p);
+	task_unlock(p);
+out:
+	return ret;
+}
+
+/*
+ * Return raw IO priority value as set by userspace. We use this for
+ * ioprio_get(pid, IOPRIO_WHO_PROCESS) so that we keep historical behavior and
+ * also so that userspace can distinguish unset IO priority (which just gets
+ * overriden based on task's nice value) from IO priority set to some value.
+ */
+static int get_task_raw_ioprio(struct task_struct *p)
+{
+	int ret;
+
+	ret = security_task_getioprio(p);
+	if (ret)
+		goto out;
+	task_lock(p);
+	if (p->io_context)
+		ret = p->io_context->ioprio;
+	else
+		ret = IOPRIO_DEFAULT;
+	task_unlock(p);
+out:
+	return ret;
+}
+
+static int ioprio_best(unsigned short aprio, unsigned short bprio)
+{
+	return min(aprio, bprio);
+}
+
+SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
+{
+	struct task_struct *g, *p;
+	struct user_struct *user;
+	struct pid *pgrp;
+	kuid_t uid;
+	int ret = -ESRCH;
+	int tmpio;
+
+	rcu_read_lock();
+	switch (which) {
+		case IOPRIO_WHO_PROCESS:
+			if (!who)
+				p = current;
+			else
+				p = find_task_by_vpid(who);
+			if (p)
+				ret = get_task_raw_ioprio(p);
+			break;
+		case IOPRIO_WHO_PGRP:
+			if (!who)
+				pgrp = task_pgrp(current);
+			else
+				pgrp = find_vpid(who);
+			read_lock(&tasklist_lock);
+			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
+				tmpio = get_task_ioprio(p);
+				if (tmpio < 0)
+					continue;
+				if (ret == -ESRCH)
+					ret = tmpio;
+				else
+					ret = ioprio_best(ret, tmpio);
+			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
+			read_unlock(&tasklist_lock);
+
+			break;
+		case IOPRIO_WHO_USER:
+			uid = make_kuid(current_user_ns(), who);
+			if (!who)
+				user = current_user();
+			else
+				user = find_user(uid);
+
+			if (!user)
+				break;
+
+			for_each_process_thread(g, p) {
+				if (!uid_eq(task_uid(p), user->uid) ||
+				    !task_pid_vnr(p))
+					continue;
+				tmpio = get_task_ioprio(p);
+				if (tmpio < 0)
+					continue;
+				if (ret == -ESRCH)
+					ret = tmpio;
+				else
+					ret = ioprio_best(ret, tmpio);
+			}
+
+			if (who)
+				free_uid(user);
+			break;
+		default:
+			ret = -EINVAL;
+	}
+
+	rcu_read_unlock();
+	return ret;
+}