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, 120 insertions, 0 deletions

diff --git a/include/net/gue.h b/include/net/gue.h
new file mode 100644
index 000000000..caefd6da8
--- /dev/null
+++ b/include/net/gue.h
@@ -0,0 +1,120 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __NET_GUE_H
+#define __NET_GUE_H
+
+/* Definitions for the GUE header, standard and private flags, lengths
+ * of optional fields are below.
+ *
+ * Diagram of GUE header:
+ *
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |Ver|C|  Hlen   | Proto/ctype   |        Standard flags       |P|
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |                                                               |
+ * ~                      Fields (optional)                        ~
+ * |                                                               |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |            Private flags (optional, P bit is set)             |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |                                                               |
+ * ~                   Private fields (optional)                   ~
+ * |                                                               |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *
+ * C bit indicates control message when set, data message when unset.
+ * For a control message, proto/ctype is interpreted as a type of
+ * control message. For data messages, proto/ctype is the IP protocol
+ * of the next header.
+ *
+ * P bit indicates private flags field is present. The private flags
+ * may refer to options placed after this field.
+ */
+
+#include <asm/byteorder.h>
+#include <linux/types.h>
+
+struct guehdr {
+	union {
+		struct {
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+			__u8	hlen:5,
+				control:1,
+				version:2;
+#elif defined (__BIG_ENDIAN_BITFIELD)
+			__u8	version:2,
+				control:1,
+				hlen:5;
+#else
+#error  "Please fix <asm/byteorder.h>"
+#endif
+			__u8	proto_ctype;
+			__be16	flags;
+		};
+		__be32	word;
+	};
+};
+
+/* Standard flags in GUE header */
+
+#define GUE_FLAG_PRIV	htons(1<<0)	/* Private flags are in options */
+#define GUE_LEN_PRIV	4
+
+#define GUE_FLAGS_ALL	(GUE_FLAG_PRIV)
+
+/* Private flags in the private option extension */
+
+#define GUE_PFLAG_REMCSUM	htonl(1U << 31)
+#define GUE_PLEN_REMCSUM	4
+
+#define GUE_PFLAGS_ALL	(GUE_PFLAG_REMCSUM)
+
+/* Functions to compute options length corresponding to flags.
+ * If we ever have a lot of flags this can be potentially be
+ * converted to a more optimized algorithm (table lookup
+ * for instance).
+ */
+static inline size_t guehdr_flags_len(__be16 flags)
+{
+	return ((flags & GUE_FLAG_PRIV) ? GUE_LEN_PRIV : 0);
+}
+
+static inline size_t guehdr_priv_flags_len(__be32 flags)
+{
+	return (flags & GUE_PFLAG_REMCSUM) ? GUE_PLEN_REMCSUM : 0;
+}
+
+/* Validate standard and private flags. Returns non-zero (meaning invalid)
+ * if there is an unknown standard or private flags, or the options length for
+ * the flags exceeds the options length specific in hlen of the GUE header.
+ */
+static inline int validate_gue_flags(struct guehdr *guehdr, size_t optlen)
+{
+	__be16 flags = guehdr->flags;
+	size_t len;
+
+	if (flags & ~GUE_FLAGS_ALL)
+		return 1;
+
+	len = guehdr_flags_len(flags);
+	if (len > optlen)
+		return 1;
+
+	if (flags & GUE_FLAG_PRIV) {
+		/* Private flags are last four bytes accounted in
+		 * guehdr_flags_len
+		 */
+		__be32 pflags = *(__be32 *)((void *)&guehdr[1] +
+					    len - GUE_LEN_PRIV);
+
+		if (pflags & ~GUE_PFLAGS_ALL)
+			return 1;
+
+		len += guehdr_priv_flags_len(pflags);
+		if (len > optlen)
+			return 1;
+	}
+
+	return 0;
+}
+
+#endif