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, 123 insertions, 0 deletions

diff --git a/net/sunrpc/timer.c b/net/sunrpc/timer.c
new file mode 100644
index 000000000..81ae35b37
--- /dev/null
+++ b/net/sunrpc/timer.c
@@ -0,0 +1,123 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * linux/net/sunrpc/timer.c
+ *
+ * Estimate RPC request round trip time.
+ *
+ * Based on packet round-trip and variance estimator algorithms described
+ * in appendix A of "Congestion Avoidance and Control" by Van Jacobson
+ * and Michael J. Karels (ACM Computer Communication Review; Proceedings
+ * of the Sigcomm '88 Symposium in Stanford, CA, August, 1988).
+ *
+ * This RTT estimator is used only for RPC over datagram protocols.
+ *
+ * Copyright (C) 2002 Trond Myklebust <trond.myklebust@fys.uio.no>
+ */
+
+#include <asm/param.h>
+
+#include <linux/types.h>
+#include <linux/unistd.h>
+#include <linux/module.h>
+
+#include <linux/sunrpc/clnt.h>
+
+#define RPC_RTO_MAX (60*HZ)
+#define RPC_RTO_INIT (HZ/5)
+#define RPC_RTO_MIN (HZ/10)
+
+/**
+ * rpc_init_rtt - Initialize an RPC RTT estimator context
+ * @rt: context to initialize
+ * @timeo: initial timeout value, in jiffies
+ *
+ */
+void rpc_init_rtt(struct rpc_rtt *rt, unsigned long timeo)
+{
+	unsigned long init = 0;
+	unsigned int i;
+
+	rt->timeo = timeo;
+
+	if (timeo > RPC_RTO_INIT)
+		init = (timeo - RPC_RTO_INIT) << 3;
+	for (i = 0; i < 5; i++) {
+		rt->srtt[i] = init;
+		rt->sdrtt[i] = RPC_RTO_INIT;
+		rt->ntimeouts[i] = 0;
+	}
+}
+EXPORT_SYMBOL_GPL(rpc_init_rtt);
+
+/**
+ * rpc_update_rtt - Update an RPC RTT estimator context
+ * @rt: context to update
+ * @timer: timer array index (request type)
+ * @m: recent actual RTT, in jiffies
+ *
+ * NB: When computing the smoothed RTT and standard deviation,
+ *     be careful not to produce negative intermediate results.
+ */
+void rpc_update_rtt(struct rpc_rtt *rt, unsigned int timer, long m)
+{
+	long *srtt, *sdrtt;
+
+	if (timer-- == 0)
+		return;
+
+	/* jiffies wrapped; ignore this one */
+	if (m < 0)
+		return;
+
+	if (m == 0)
+		m = 1L;
+
+	srtt = (long *)&rt->srtt[timer];
+	m -= *srtt >> 3;
+	*srtt += m;
+
+	if (m < 0)
+		m = -m;
+
+	sdrtt = (long *)&rt->sdrtt[timer];
+	m -= *sdrtt >> 2;
+	*sdrtt += m;
+
+	/* Set lower bound on the variance */
+	if (*sdrtt < RPC_RTO_MIN)
+		*sdrtt = RPC_RTO_MIN;
+}
+EXPORT_SYMBOL_GPL(rpc_update_rtt);
+
+/**
+ * rpc_calc_rto - Provide an estimated timeout value
+ * @rt: context to use for calculation
+ * @timer: timer array index (request type)
+ *
+ * Estimate RTO for an NFS RPC sent via an unreliable datagram.  Use
+ * the mean and mean deviation of RTT for the appropriate type of RPC
+ * for frequently issued RPCs, and a fixed default for the others.
+ *
+ * The justification for doing "other" this way is that these RPCs
+ * happen so infrequently that timer estimation would probably be
+ * stale.  Also, since many of these RPCs are non-idempotent, a
+ * conservative timeout is desired.
+ *
+ * getattr, lookup,
+ * read, write, commit     - A+4D
+ * other                   - timeo
+ */
+unsigned long rpc_calc_rto(struct rpc_rtt *rt, unsigned int timer)
+{
+	unsigned long res;
+
+	if (timer-- == 0)
+		return rt->timeo;
+
+	res = ((rt->srtt[timer] + 7) >> 3) + rt->sdrtt[timer];
+	if (res > RPC_RTO_MAX)
+		res = RPC_RTO_MAX;
+
+	return res;
+}
+EXPORT_SYMBOL_GPL(rpc_calc_rto);