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/uapi/asm-generic/stat.h b/include/uapi/asm-generic/stat.h
new file mode 100644
index 000000000..0d962ecd1
--- /dev/null
+++ b/include/uapi/asm-generic/stat.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+#ifndef __ASM_GENERIC_STAT_H
+#define __ASM_GENERIC_STAT_H
+
+/*
+ * Everybody gets this wrong and has to stick with it for all
+ * eternity. Hopefully, this version gets used by new architectures
+ * so they don't fall into the same traps.
+ *
+ * stat64 is copied from powerpc64, with explicit padding added.
+ * stat is the same structure layout on 64-bit, without the 'long long'
+ * types.
+ *
+ * By convention, 64 bit architectures use the stat interface, while
+ * 32 bit architectures use the stat64 interface. Note that we don't
+ * provide an __old_kernel_stat here, which new architecture should
+ * not have to start with.
+ */
+
+#include <asm/bitsperlong.h>
+
+#define STAT_HAVE_NSEC 1
+
+struct stat {
+	unsigned long	st_dev;		/* Device.  */
+	unsigned long	st_ino;		/* File serial number.  */
+	unsigned int	st_mode;	/* File mode.  */
+	unsigned int	st_nlink;	/* Link count.  */
+	unsigned int	st_uid;		/* User ID of the file's owner.  */
+	unsigned int	st_gid;		/* Group ID of the file's group. */
+	unsigned long	st_rdev;	/* Device number, if device.  */
+	unsigned long	__pad1;
+	long		st_size;	/* Size of file, in bytes.  */
+	int		st_blksize;	/* Optimal block size for I/O.  */
+	int		__pad2;
+	long		st_blocks;	/* Number 512-byte blocks allocated. */
+	long		st_atime;	/* Time of last access.  */
+	unsigned long	st_atime_nsec;
+	long		st_mtime;	/* Time of last modification.  */
+	unsigned long	st_mtime_nsec;
+	long		st_ctime;	/* Time of last status change.  */
+	unsigned long	st_ctime_nsec;
+	unsigned int	__unused4;
+	unsigned int	__unused5;
+};
+
+/* This matches struct stat64 in glibc2.1. Only used for 32 bit. */
+#if __BITS_PER_LONG != 64 || defined(__ARCH_WANT_STAT64)
+struct stat64 {
+	unsigned long long st_dev;	/* Device.  */
+	unsigned long long st_ino;	/* File serial number.  */
+	unsigned int	st_mode;	/* File mode.  */
+	unsigned int	st_nlink;	/* Link count.  */
+	unsigned int	st_uid;		/* User ID of the file's owner.  */
+	unsigned int	st_gid;		/* Group ID of the file's group. */
+	unsigned long long st_rdev;	/* Device number, if device.  */
+	unsigned long long __pad1;
+	long long	st_size;	/* Size of file, in bytes.  */
+	int		st_blksize;	/* Optimal block size for I/O.  */
+	int		__pad2;
+	long long	st_blocks;	/* Number 512-byte blocks allocated. */
+	int		st_atime;	/* Time of last access.  */
+	unsigned int	st_atime_nsec;
+	int		st_mtime;	/* Time of last modification.  */
+	unsigned int	st_mtime_nsec;
+	int		st_ctime;	/* Time of last status change.  */
+	unsigned int	st_ctime_nsec;
+	unsigned int	__unused4;
+	unsigned int	__unused5;
+};
+#endif
+
+#endif /* __ASM_GENERIC_STAT_H */