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, 87 insertions, 0 deletions

diff --git a/include/uapi/xen/gntalloc.h b/include/uapi/xen/gntalloc.h
new file mode 100644
index 000000000..310928267
--- /dev/null
+++ b/include/uapi/xen/gntalloc.h
@@ -0,0 +1,87 @@
+/******************************************************************************
+ * gntalloc.h
+ *
+ * Interface to /dev/xen/gntalloc.
+ *
+ * Author: Daniel De Graaf <dgdegra@tycho.nsa.gov>
+ *
+ * This file is in the public domain.
+ */
+
+#ifndef __LINUX_PUBLIC_GNTALLOC_H__
+#define __LINUX_PUBLIC_GNTALLOC_H__
+
+#include <linux/types.h>
+
+/*
+ * Allocates a new page and creates a new grant reference.
+ */
+#define IOCTL_GNTALLOC_ALLOC_GREF \
+_IOC(_IOC_NONE, 'G', 5, sizeof(struct ioctl_gntalloc_alloc_gref))
+struct ioctl_gntalloc_alloc_gref {
+	/* IN parameters */
+	/* The ID of the domain to be given access to the grants. */
+	__u16 domid;
+	/* Flags for this mapping */
+	__u16 flags;
+	/* Number of pages to map */
+	__u32 count;
+	/* OUT parameters */
+	/* The offset to be used on a subsequent call to mmap(). */
+	__u64 index;
+	/* The grant references of the newly created grant, one per page */
+	/* Variable size, depending on count */
+	union {
+		__u32 gref_ids[1];
+		__DECLARE_FLEX_ARRAY(__u32, gref_ids_flex);
+	};
+};
+
+#define GNTALLOC_FLAG_WRITABLE 1
+
+/*
+ * Deallocates the grant reference, allowing the associated page to be freed if
+ * no other domains are using it.
+ */
+#define IOCTL_GNTALLOC_DEALLOC_GREF \
+_IOC(_IOC_NONE, 'G', 6, sizeof(struct ioctl_gntalloc_dealloc_gref))
+struct ioctl_gntalloc_dealloc_gref {
+	/* IN parameters */
+	/* The offset returned in the map operation */
+	__u64 index;
+	/* Number of references to unmap */
+	__u32 count;
+};
+
+/*
+ * Sets up an unmap notification within the page, so that the other side can do
+ * cleanup if this side crashes. Required to implement cross-domain robust
+ * mutexes or close notification on communication channels.
+ *
+ * Each mapped page only supports one notification; multiple calls referring to
+ * the same page overwrite the previous notification. You must clear the
+ * notification prior to the IOCTL_GNTALLOC_DEALLOC_GREF if you do not want it
+ * to occur.
+ */
+#define IOCTL_GNTALLOC_SET_UNMAP_NOTIFY \
+_IOC(_IOC_NONE, 'G', 7, sizeof(struct ioctl_gntalloc_unmap_notify))
+struct ioctl_gntalloc_unmap_notify {
+	/* IN parameters */
+	/* Offset in the file descriptor for a byte within the page (same as
+	 * used in mmap). If using UNMAP_NOTIFY_CLEAR_BYTE, this is the byte to
+	 * be cleared. Otherwise, it can be any byte in the page whose
+	 * notification we are adjusting.
+	 */
+	__u64 index;
+	/* Action(s) to take on unmap */
+	__u32 action;
+	/* Event channel to notify */
+	__u32 event_channel_port;
+};
+
+/* Clear (set to zero) the byte specified by index */
+#define UNMAP_NOTIFY_CLEAR_BYTE 0x1
+/* Send an interrupt on the indicated event channel */
+#define UNMAP_NOTIFY_SEND_EVENT 0x2
+
+#endif /* __LINUX_PUBLIC_GNTALLOC_H__ */