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, 111 insertions, 0 deletions

diff --git a/include/clocksource/hyperv_timer.h b/include/clocksource/hyperv_timer.h
new file mode 100644
index 000000000..8d3befb7e
--- /dev/null
+++ b/include/clocksource/hyperv_timer.h
@@ -0,0 +1,111 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+/*
+ * Definitions for the clocksource provided by the Hyper-V
+ * hypervisor to guest VMs, as described in the Hyper-V Top
+ * Level Functional Spec (TLFS).
+ *
+ * Copyright (C) 2019, Microsoft, Inc.
+ *
+ * Author:  Michael Kelley <mikelley@microsoft.com>
+ */
+
+#ifndef __CLKSOURCE_HYPERV_TIMER_H
+#define __CLKSOURCE_HYPERV_TIMER_H
+
+#include <linux/clocksource.h>
+#include <linux/math64.h>
+#include <hyperv/hvhdk.h>
+
+#define HV_MAX_MAX_DELTA_TICKS 0xffffffff
+#define HV_MIN_DELTA_TICKS 1
+
+#ifdef CONFIG_HYPERV_TIMER
+
+#include <asm/hyperv_timer.h>
+
+/* Routines called by the VMbus driver */
+extern int hv_stimer_alloc(bool have_percpu_irqs);
+extern int hv_stimer_cleanup(unsigned int cpu);
+extern void hv_stimer_global_cleanup(void);
+
+extern void hv_init_clocksource(void);
+extern void hv_remap_tsc_clocksource(void);
+
+extern unsigned long hv_get_tsc_pfn(void);
+extern struct ms_hyperv_tsc_page *hv_get_tsc_page(void);
+
+extern void hv_adj_sched_clock_offset(u64 offset);
+
+static __always_inline bool
+hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
+		     u64 *cur_tsc, u64 *time)
+{
+	u64 scale, offset;
+	u32 sequence;
+
+	/*
+	 * The protocol for reading Hyper-V TSC page is specified in Hypervisor
+	 * Top-Level Functional Specification ver. 3.0 and above. To get the
+	 * reference time we must do the following:
+	 * - READ ReferenceTscSequence
+	 *   A special '0' value indicates the time source is unreliable and we
+	 *   need to use something else. The currently published specification
+	 *   versions (up to 4.0b) contain a mistake and wrongly claim '-1'
+	 *   instead of '0' as the special value, see commit c35b82ef0294.
+	 * - ReferenceTime =
+	 *        ((RDTSC() * ReferenceTscScale) >> 64) + ReferenceTscOffset
+	 * - READ ReferenceTscSequence again. In case its value has changed
+	 *   since our first reading we need to discard ReferenceTime and repeat
+	 *   the whole sequence as the hypervisor was updating the page in
+	 *   between.
+	 */
+	do {
+		sequence = READ_ONCE(tsc_pg->tsc_sequence);
+		if (!sequence)
+			return false;
+		/*
+		 * Make sure we read sequence before we read other values from
+		 * TSC page.
+		 */
+		smp_rmb();
+
+		scale = READ_ONCE(tsc_pg->tsc_scale);
+		offset = READ_ONCE(tsc_pg->tsc_offset);
+		*cur_tsc = hv_get_raw_timer();
+
+		/*
+		 * Make sure we read sequence after we read all other values
+		 * from TSC page.
+		 */
+		smp_rmb();
+
+	} while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
+
+	*time = mul_u64_u64_shr(*cur_tsc, scale, 64) + offset;
+	return true;
+}
+
+#else /* CONFIG_HYPERV_TIMER */
+static inline unsigned long hv_get_tsc_pfn(void)
+{
+	return 0;
+}
+
+static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
+{
+	return NULL;
+}
+
+static __always_inline bool
+hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg, u64 *cur_tsc, u64 *time)
+{
+	return false;
+}
+
+static inline int hv_stimer_cleanup(unsigned int cpu) { return 0; }
+static inline void hv_stimer_global_cleanup(void) {}
+
+#endif /* CONFIG_HYPERV_TIMER */
+
+#endif