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, 76 insertions, 0 deletions

diff --git a/include/net/dscp.h b/include/net/dscp.h
new file mode 100644
index 000000000..ba4054086
--- /dev/null
+++ b/include/net/dscp.h
@@ -0,0 +1,76 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (c) 2024 Pengutronix, Oleksij Rempel <kernel@pengutronix.de> */
+
+#ifndef __DSCP_H__
+#define __DSCP_H__
+
+/*
+ * DSCP Pools and Codepoint Space Division:
+ *
+ * The Differentiated Services (Diffserv) architecture defines a method for
+ * classifying and managing network traffic using the DS field in IPv4 and IPv6
+ * packet headers. This field can carry one of 64 distinct DSCP (Differentiated
+ * Services Code Point) values, which are divided into three pools based on
+ * their Least Significant Bits (LSB) patterns and intended usage. Each pool has
+ * a specific registration procedure for assigning DSCP values:
+ *
+ * Pool 1 (Standards Action Pool):
+ * - Codepoint Space: xxxxx0
+ *   This pool includes DSCP values ending in '0' (binary), allocated via
+ *   Standards Action. It is intended for globally recognized traffic classes,
+ *   ensuring interoperability across the internet. This pool encompasses
+ *   well-known DSCP values such as CS0-CS7, AFxx, EF, and VOICE-ADMIT.
+ *
+ * Pool 2 (Experimental/Local Use Pool):
+ * - Codepoint Space: xxxx11
+ *   Reserved for DSCP values ending in '11' (binary), this pool is designated
+ *   for Experimental or Local Use. It allows for private or temporary traffic
+ *   marking schemes not intended for standardized global use, facilitating
+ *   testing and network-specific configurations without impacting
+ *   interoperability.
+ *
+ * Pool 3 (Preferential Standardization Pool):
+ * - Codepoint Space: xxxx01
+ *   Initially reserved for experimental or local use, this pool now serves as
+ *   a secondary standardization resource should Pool 1 become exhausted. DSCP
+ *   values ending in '01' (binary) are assigned via Standards Action, with a
+ *   focus on adopting new, standardized traffic classes as the need arises.
+ *
+ * For pool updates see:
+ * https://www.iana.org/assignments/dscp-registry/dscp-registry.xhtml
+ */
+
+/* Pool 1: Standardized DSCP values as per [RFC8126] */
+#define DSCP_CS0 0		/* 000000, [RFC2474] */
+/* CS0 is some times called default (DF) */
+#define DSCP_DF 0		/* 000000, [RFC2474] */
+#define DSCP_CS1 8		/* 001000, [RFC2474] */
+#define DSCP_CS2 16		/* 010000, [RFC2474] */
+#define DSCP_CS3 24		/* 011000, [RFC2474] */
+#define DSCP_CS4 32		/* 100000, [RFC2474] */
+#define DSCP_CS5 40		/* 101000, [RFC2474] */
+#define DSCP_CS6 48		/* 110000, [RFC2474] */
+#define DSCP_CS7 56		/* 111000, [RFC2474] */
+#define DSCP_AF11 10		/* 001010, [RFC2597] */
+#define DSCP_AF12 12		/* 001100, [RFC2597] */
+#define DSCP_AF13 14		/* 001110, [RFC2597] */
+#define DSCP_AF21 18		/* 010010, [RFC2597] */
+#define DSCP_AF22 20		/* 010100, [RFC2597] */
+#define DSCP_AF23 22		/* 010110, [RFC2597] */
+#define DSCP_AF31 26		/* 011010, [RFC2597] */
+#define DSCP_AF32 28		/* 011100, [RFC2597] */
+#define DSCP_AF33 30		/* 011110, [RFC2597] */
+#define DSCP_AF41 34		/* 100010, [RFC2597] */
+#define DSCP_AF42 36		/* 100100, [RFC2597] */
+#define DSCP_AF43 38		/* 100110, [RFC2597] */
+#define DSCP_EF 46		/* 101110, [RFC3246] */
+#define DSCP_VOICE_ADMIT 44	/* 101100, [RFC5865] */
+
+/* Pool 3: Standardized assignments, previously available for experimental/local
+ * use
+ */
+#define DSCP_LE 1		/* 000001, [RFC8622] */
+
+#define DSCP_MAX 64
+
+#endif /* __DSCP_H__ */