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, 85 insertions, 0 deletions

diff --git a/include/xen/interface/callback.h b/include/xen/interface/callback.h
new file mode 100644
index 000000000..c67822a25
--- /dev/null
+++ b/include/xen/interface/callback.h
@@ -0,0 +1,85 @@
+/* SPDX-License-Identifier: MIT */
+/******************************************************************************
+ * callback.h
+ *
+ * Register guest OS callbacks with Xen.
+ *
+ * Copyright (c) 2006, Ian Campbell
+ */
+
+#ifndef __XEN_PUBLIC_CALLBACK_H__
+#define __XEN_PUBLIC_CALLBACK_H__
+
+#include <xen/interface/xen.h>
+
+/*
+ * Prototype for this hypercall is:
+ *   long callback_op(int cmd, void *extra_args)
+ * @cmd        == CALLBACKOP_??? (callback operation).
+ * @extra_args == Operation-specific extra arguments (NULL if none).
+ */
+
+/* x86: Callback for event delivery. */
+#define CALLBACKTYPE_event                 0
+
+/* x86: Failsafe callback when guest state cannot be restored by Xen. */
+#define CALLBACKTYPE_failsafe              1
+
+/* x86/64 hypervisor: Syscall by 64-bit guest app ('64-on-64-on-64'). */
+#define CALLBACKTYPE_syscall               2
+
+/*
+ * x86/32 hypervisor: Only available on x86/32 when supervisor_mode_kernel
+ *     feature is enabled. Do not use this callback type in new code.
+ */
+#define CALLBACKTYPE_sysenter_deprecated   3
+
+/* x86: Callback for NMI delivery. */
+#define CALLBACKTYPE_nmi                   4
+
+/*
+ * x86: sysenter is only available as follows:
+ * - 32-bit hypervisor: with the supervisor_mode_kernel feature enabled
+ * - 64-bit hypervisor: 32-bit guest applications on Intel CPUs
+ *                      ('32-on-32-on-64', '32-on-64-on-64')
+ *                      [nb. also 64-bit guest applications on Intel CPUs
+ *                           ('64-on-64-on-64'), but syscall is preferred]
+ */
+#define CALLBACKTYPE_sysenter              5
+
+/*
+ * x86/64 hypervisor: Syscall by 32-bit guest app on AMD CPUs
+ *                    ('32-on-32-on-64', '32-on-64-on-64')
+ */
+#define CALLBACKTYPE_syscall32             7
+
+/*
+ * Disable event deliver during callback? This flag is ignored for event and
+ * NMI callbacks: event delivery is unconditionally disabled.
+ */
+#define _CALLBACKF_mask_events             0
+#define CALLBACKF_mask_events              (1U << _CALLBACKF_mask_events)
+
+/*
+ * Register a callback.
+ */
+#define CALLBACKOP_register                0
+struct callback_register {
+	uint16_t type;
+	uint16_t flags;
+	xen_callback_t address;
+};
+
+/*
+ * Unregister a callback.
+ *
+ * Not all callbacks can be unregistered. -EINVAL will be returned if
+ * you attempt to unregister such a callback.
+ */
+#define CALLBACKOP_unregister              1
+struct callback_unregister {
+    uint16_t type;
+    uint16_t _unused;
+};
+
+#endif /* __XEN_PUBLIC_CALLBACK_H__ */