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, 141 insertions, 0 deletions

diff --git a/tools/testing/vma/shared.h b/tools/testing/vma/shared.h
new file mode 100644
index 000000000..97cd7a679
--- /dev/null
+++ b/tools/testing/vma/shared.h
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "generated/bit-length.h"
+#include "maple-shared.h"
+#include "vma_internal.h"
+#include "../../../mm/vma.h"
+
+/* Simple test runner. Assumes local num_[fail, tests] counters. */
+#define TEST(name)							\
+	do {								\
+		(*num_tests)++;						\
+		if (!test_##name()) {					\
+			(*num_fail)++;					\
+			fprintf(stderr, "Test " #name " FAILED\n");	\
+		}							\
+	} while (0)
+
+#define __ASSERT_TRUE(_expr, _fmt, ...)					   \
+	do {								   \
+		if (!(_expr)) {						   \
+			fprintf(stderr,					   \
+				"Assert FAILED at %s:%d:%s(): %s is FALSE" \
+				_fmt ".\n",				   \
+				__FILE__, __LINE__, __FUNCTION__, #_expr   \
+				__VA_OPT__(,) __VA_ARGS__);		   \
+			return false;					   \
+		}							   \
+	} while (0)
+
+#define __TO_SCALAR(x)	((unsigned long long)(uintptr_t)(x))
+
+#define ASSERT_TRUE(_expr) __ASSERT_TRUE(_expr, "")
+#define ASSERT_FALSE(_expr) ASSERT_TRUE(!(_expr))
+#define ASSERT_EQ(_val1, _val2) do {				 \
+	__typeof__(_val1) __val1 = (_val1);			 \
+	__typeof__(_val2) __val2 = (_val2);			 \
+	__ASSERT_TRUE(__val1 == __val2, " (0x%llx != 0x%llx)",	 \
+		      __TO_SCALAR(__val1), __TO_SCALAR(__val2)); \
+	} while (0)
+
+#define ASSERT_NE(_val1, _val2) do {				 \
+	__typeof__(_val1) __val1 = (_val1);			 \
+	__typeof__(_val2) __val2 = (_val2);			 \
+	__ASSERT_TRUE(__val1 != __val2, " (0x%llx == 0x%llx)",	 \
+		      __TO_SCALAR(__val1), __TO_SCALAR(__val2)); \
+	} while (0)
+
+#define ASSERT_FLAGS_SAME_MASK(_flags, _flags_other) \
+	ASSERT_TRUE(vma_flags_same_mask((_flags), (_flags_other)))
+
+#define ASSERT_FLAGS_NOT_SAME_MASK(_flags, _flags_other) \
+	ASSERT_FALSE(vma_flags_same_mask((_flags), (_flags_other)))
+
+#define ASSERT_FLAGS_SAME(_flags, ...) \
+	ASSERT_TRUE(vma_flags_same(_flags, __VA_ARGS__))
+
+#define ASSERT_FLAGS_NOT_SAME(_flags, ...) \
+	ASSERT_FALSE(vma_flags_same(_flags, __VA_ARGS__))
+
+#define ASSERT_FLAGS_EMPTY(_flags) \
+	ASSERT_TRUE(vma_flags_empty(_flags))
+
+#define ASSERT_FLAGS_NONEMPTY(_flags) \
+	ASSERT_FALSE(vma_flags_empty(_flags))
+
+extern bool fail_prealloc;
+
+/* Override vma_iter_prealloc() so we can choose to fail it. */
+#define vma_iter_prealloc(vmi, vma)					\
+	(fail_prealloc ? -ENOMEM : mas_preallocate(&(vmi)->mas, (vma), GFP_KERNEL))
+
+#define CONFIG_DEFAULT_MMAP_MIN_ADDR 65536
+
+extern unsigned long mmap_min_addr;
+extern unsigned long dac_mmap_min_addr;
+extern unsigned long stack_guard_gap;
+
+extern const struct vm_operations_struct vma_dummy_vm_ops;
+extern struct anon_vma dummy_anon_vma;
+extern struct task_struct __current;
+
+/*
+ * Helper function which provides a wrapper around a merge existing VMA
+ * operation.
+ *
+ * Declared in main.c as uses static VMA function.
+ */
+struct vm_area_struct *merge_existing(struct vma_merge_struct *vmg);
+
+/*
+ * Helper function to allocate a VMA and link it to the tree.
+ *
+ * Declared in main.c as uses static VMA function.
+ */
+int attach_vma(struct mm_struct *mm, struct vm_area_struct *vma);
+
+/* Helper function providing a dummy vm_ops->close() method.*/
+static inline void dummy_close(struct vm_area_struct *)
+{
+}
+
+/* Helper function to simply allocate a VMA. */
+struct vm_area_struct *alloc_vma(struct mm_struct *mm,
+		unsigned long start, unsigned long end,
+		pgoff_t pgoff, vma_flags_t vma_flags);
+
+/* Helper function to detach and free a VMA. */
+void detach_free_vma(struct vm_area_struct *vma);
+
+/* Helper function to allocate a VMA and link it to the tree. */
+struct vm_area_struct *alloc_and_link_vma(struct mm_struct *mm,
+		unsigned long start, unsigned long end,
+		pgoff_t pgoff, vma_flags_t vma_flags);
+
+/*
+ * Helper function to reset the dummy anon_vma to indicate it has not been
+ * duplicated.
+ */
+void reset_dummy_anon_vma(void);
+
+/*
+ * Helper function to remove all VMAs and destroy the maple tree associated with
+ * a virtual address space. Returns a count of VMAs in the tree.
+ */
+int cleanup_mm(struct mm_struct *mm, struct vma_iterator *vmi);
+
+/* Helper function to determine if VMA has had vma_start_write() performed. */
+bool vma_write_started(struct vm_area_struct *vma);
+
+void __vma_set_dummy_anon_vma(struct vm_area_struct *vma,
+		struct anon_vma_chain *avc, struct anon_vma *anon_vma);
+
+/* Provide a simple dummy VMA/anon_vma dummy setup for testing. */
+void vma_set_dummy_anon_vma(struct vm_area_struct *vma,
+			    struct anon_vma_chain *avc);