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, 173 insertions, 0 deletions

diff --git a/io_uring/eventfd.c b/io_uring/eventfd.c
new file mode 100644
index 000000000..63fe6e5d7
--- /dev/null
+++ b/io_uring/eventfd.c
@@ -0,0 +1,173 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/eventfd.h>
+#include <linux/eventpoll.h>
+#include <linux/io_uring.h>
+#include <linux/io_uring_types.h>
+
+#include "io-wq.h"
+#include "eventfd.h"
+
+struct io_ev_fd {
+	struct eventfd_ctx	*cq_ev_fd;
+	unsigned int		eventfd_async;
+	/* protected by ->completion_lock */
+	unsigned		last_cq_tail;
+	refcount_t		refs;
+	atomic_t		ops;
+	struct rcu_head		rcu;
+};
+
+enum {
+	IO_EVENTFD_OP_SIGNAL_BIT,
+};
+
+static void io_eventfd_free(struct rcu_head *rcu)
+{
+	struct io_ev_fd *ev_fd = container_of(rcu, struct io_ev_fd, rcu);
+
+	eventfd_ctx_put(ev_fd->cq_ev_fd);
+	kfree(ev_fd);
+}
+
+static void io_eventfd_put(struct io_ev_fd *ev_fd)
+{
+	if (refcount_dec_and_test(&ev_fd->refs))
+		call_rcu(&ev_fd->rcu, io_eventfd_free);
+}
+
+static void io_eventfd_do_signal(struct rcu_head *rcu)
+{
+	struct io_ev_fd *ev_fd = container_of(rcu, struct io_ev_fd, rcu);
+
+	atomic_andnot(BIT(IO_EVENTFD_OP_SIGNAL_BIT), &ev_fd->ops);
+	eventfd_signal_mask(ev_fd->cq_ev_fd, EPOLL_URING_WAKE);
+	io_eventfd_put(ev_fd);
+}
+
+/*
+ * Returns true if the caller should put the ev_fd reference, false if not.
+ */
+static bool __io_eventfd_signal(struct io_ev_fd *ev_fd, bool defer)
+{
+	if (!defer && eventfd_signal_allowed()) {
+		eventfd_signal_mask(ev_fd->cq_ev_fd, EPOLL_URING_WAKE);
+		return true;
+	}
+	if (!atomic_fetch_or(BIT(IO_EVENTFD_OP_SIGNAL_BIT), &ev_fd->ops)) {
+		call_rcu_hurry(&ev_fd->rcu, io_eventfd_do_signal);
+		return false;
+	}
+	return true;
+}
+
+/*
+ * Trigger if eventfd_async isn't set, or if it's set and the caller is
+ * an async worker.
+ */
+static bool io_eventfd_trigger(struct io_ev_fd *ev_fd)
+{
+	return !ev_fd->eventfd_async || io_wq_current_is_worker();
+}
+
+void io_eventfd_signal(struct io_ring_ctx *ctx, bool cqe_event, bool defer)
+{
+	bool skip = false;
+	struct io_ev_fd *ev_fd;
+	struct io_rings *rings;
+
+	guard(rcu)();
+
+	rings = rcu_dereference(ctx->rings_rcu);
+	if (!rings)
+		return;
+	if (READ_ONCE(rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
+		return;
+	ev_fd = rcu_dereference(ctx->io_ev_fd);
+	/*
+	 * Check again if ev_fd exists in case an io_eventfd_unregister call
+	 * completed between the NULL check of ctx->io_ev_fd at the start of
+	 * the function and rcu_read_lock.
+	 */
+	if (!ev_fd)
+		return;
+	if (!io_eventfd_trigger(ev_fd) || !refcount_inc_not_zero(&ev_fd->refs))
+		return;
+
+	if (cqe_event) {
+		/*
+		 * Eventfd should only get triggered when at least one event
+		 * has been posted. Some applications rely on the eventfd
+		 * notification count only changing IFF a new CQE has been
+		 * added to the CQ ring. There's no dependency on 1:1
+		 * relationship between how many times this function is called
+		 * (and hence the eventfd count) and number of CQEs posted to
+		 * the CQ ring.
+		 */
+		spin_lock(&ctx->completion_lock);
+		skip = ctx->cached_cq_tail == ev_fd->last_cq_tail;
+		ev_fd->last_cq_tail = ctx->cached_cq_tail;
+		spin_unlock(&ctx->completion_lock);
+	}
+
+	if (skip || __io_eventfd_signal(ev_fd, defer))
+		io_eventfd_put(ev_fd);
+}
+
+int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg,
+			unsigned int eventfd_async)
+{
+	struct io_ev_fd *ev_fd;
+	__s32 __user *fds = arg;
+	int fd;
+
+	ev_fd = rcu_dereference_protected(ctx->io_ev_fd,
+					lockdep_is_held(&ctx->uring_lock));
+	if (ev_fd)
+		return -EBUSY;
+
+	if (copy_from_user(&fd, fds, sizeof(*fds)))
+		return -EFAULT;
+
+	ev_fd = kmalloc_obj(*ev_fd);
+	if (!ev_fd)
+		return -ENOMEM;
+
+	ev_fd->cq_ev_fd = eventfd_ctx_fdget(fd);
+	if (IS_ERR(ev_fd->cq_ev_fd)) {
+		int ret = PTR_ERR(ev_fd->cq_ev_fd);
+
+		kfree(ev_fd);
+		return ret;
+	}
+
+	spin_lock(&ctx->completion_lock);
+	ev_fd->last_cq_tail = ctx->cached_cq_tail;
+	spin_unlock(&ctx->completion_lock);
+
+	ev_fd->eventfd_async = eventfd_async;
+	ctx->int_flags |= IO_RING_F_HAS_EVFD;
+	refcount_set(&ev_fd->refs, 1);
+	atomic_set(&ev_fd->ops, 0);
+	rcu_assign_pointer(ctx->io_ev_fd, ev_fd);
+	return 0;
+}
+
+int io_eventfd_unregister(struct io_ring_ctx *ctx)
+{
+	struct io_ev_fd *ev_fd;
+
+	ev_fd = rcu_dereference_protected(ctx->io_ev_fd,
+					lockdep_is_held(&ctx->uring_lock));
+	if (ev_fd) {
+		ctx->int_flags &= ~IO_RING_F_HAS_EVFD;
+		rcu_assign_pointer(ctx->io_ev_fd, NULL);
+		io_eventfd_put(ev_fd);
+		return 0;
+	}
+
+	return -ENXIO;
+}