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, 73 insertions, 0 deletions

diff --git a/include/asm-generic/pgtable-nopmd.h b/include/asm-generic/pgtable-nopmd.h
new file mode 100644
index 000000000..36b6490ed
--- /dev/null
+++ b/include/asm-generic/pgtable-nopmd.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _PGTABLE_NOPMD_H
+#define _PGTABLE_NOPMD_H
+
+#ifndef __ASSEMBLER__
+
+#include <asm-generic/pgtable-nopud.h>
+
+struct mm_struct;
+
+#define __PAGETABLE_PMD_FOLDED 1
+
+/*
+ * Having the pmd type consist of a pud gets the size right, and allows
+ * us to conceptually access the pud entry that this pmd is folded into
+ * without casting.
+ */
+typedef struct { pud_t pud; } pmd_t;
+
+#define PMD_SHIFT	PUD_SHIFT
+#define PTRS_PER_PMD	1
+#define PMD_SIZE  	(1UL << PMD_SHIFT)
+#define PMD_MASK  	(~(PMD_SIZE-1))
+
+/*
+ * The "pud_xxx()" functions here are trivial for a folded two-level
+ * setup: the pmd is never bad, and a pmd always exists (as it's folded
+ * into the pud entry)
+ */
+static inline int pud_none(pud_t pud)		{ return 0; }
+static inline int pud_bad(pud_t pud)		{ return 0; }
+static inline int pud_present(pud_t pud)	{ return 1; }
+static inline int pud_user(pud_t pud)		{ return 0; }
+static inline int pud_leaf(pud_t pud)		{ return 0; }
+static inline void pud_clear(pud_t *pud)	{ }
+#define pmd_ERROR(pmd)				(pud_ERROR((pmd).pud))
+
+#define pud_populate(mm, pmd, pte)		do { } while (0)
+
+/*
+ * (pmds are folded into puds so this doesn't get actually called,
+ * but the define is needed for a generic inline function.)
+ */
+#define set_pud(pudptr, pudval)			set_pmd((pmd_t *)(pudptr), (pmd_t) { pudval })
+
+static inline pmd_t * pmd_offset(pud_t * pud, unsigned long address)
+{
+	return (pmd_t *)pud;
+}
+#define pmd_offset pmd_offset
+
+#define pmd_val(x)				(pud_val((x).pud))
+#define __pmd(x)				((pmd_t) { __pud(x) } )
+
+#define pud_page(pud)				(pmd_page((pmd_t){ pud }))
+#define pud_pgtable(pud)			((pmd_t *)(pmd_page_vaddr((pmd_t){ pud })))
+
+/*
+ * allocating and freeing a pmd is trivial: the 1-entry pmd is
+ * inside the pud, so has no extra memory associated with it.
+ */
+#define pmd_alloc_one(mm, address)		NULL
+static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
+{
+}
+#define pmd_free_tlb(tlb, x, a)		do { } while (0)
+
+#undef  pmd_addr_end
+#define pmd_addr_end(addr, end)			(end)
+
+#endif /* __ASSEMBLER__ */
+
+#endif /* _PGTABLE_NOPMD_H */