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, 164 insertions, 0 deletions

diff --git a/drivers/iommu/intel/perf.c b/drivers/iommu/intel/perf.c
new file mode 100644
index 000000000..bec98dcbb
--- /dev/null
+++ b/drivers/iommu/intel/perf.c
@@ -0,0 +1,164 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * perf.c - performance monitor
+ *
+ * Copyright (C) 2021 Intel Corporation
+ *
+ * Author: Lu Baolu <baolu.lu@linux.intel.com>
+ *         Fenghua Yu <fenghua.yu@intel.com>
+ */
+
+#include <linux/spinlock.h>
+
+#include "iommu.h"
+#include "perf.h"
+
+static DEFINE_SPINLOCK(latency_lock);
+
+bool dmar_latency_enabled(struct intel_iommu *iommu, enum latency_type type)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+
+	return lstat && lstat[type].enabled;
+}
+
+int dmar_latency_enable(struct intel_iommu *iommu, enum latency_type type)
+{
+	struct latency_statistic *lstat;
+	unsigned long flags;
+	int ret = -EBUSY;
+
+	if (dmar_latency_enabled(iommu, type))
+		return 0;
+
+	spin_lock_irqsave(&latency_lock, flags);
+	if (!iommu->perf_statistic) {
+		iommu->perf_statistic = kzalloc_objs(*lstat, DMAR_LATENCY_NUM,
+						     GFP_ATOMIC);
+		if (!iommu->perf_statistic) {
+			ret = -ENOMEM;
+			goto unlock_out;
+		}
+	}
+
+	lstat = iommu->perf_statistic;
+
+	if (!lstat[type].enabled) {
+		lstat[type].enabled = true;
+		lstat[type].counter[COUNTS_MIN] = UINT_MAX;
+		ret = 0;
+	}
+unlock_out:
+	spin_unlock_irqrestore(&latency_lock, flags);
+
+	return ret;
+}
+
+void dmar_latency_disable(struct intel_iommu *iommu, enum latency_type type)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+	unsigned long flags;
+
+	if (!dmar_latency_enabled(iommu, type))
+		return;
+
+	spin_lock_irqsave(&latency_lock, flags);
+	memset(&lstat[type], 0, sizeof(*lstat));
+	spin_unlock_irqrestore(&latency_lock, flags);
+}
+
+void dmar_latency_update(struct intel_iommu *iommu, enum latency_type type, u64 latency)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+	unsigned long flags;
+	u64 min, max;
+
+	if (!dmar_latency_enabled(iommu, type))
+		return;
+
+	spin_lock_irqsave(&latency_lock, flags);
+	if (latency < 100)
+		lstat[type].counter[COUNTS_10e2]++;
+	else if (latency < 1000)
+		lstat[type].counter[COUNTS_10e3]++;
+	else if (latency < 10000)
+		lstat[type].counter[COUNTS_10e4]++;
+	else if (latency < 100000)
+		lstat[type].counter[COUNTS_10e5]++;
+	else if (latency < 1000000)
+		lstat[type].counter[COUNTS_10e6]++;
+	else if (latency < 10000000)
+		lstat[type].counter[COUNTS_10e7]++;
+	else
+		lstat[type].counter[COUNTS_10e8_plus]++;
+
+	min = lstat[type].counter[COUNTS_MIN];
+	max = lstat[type].counter[COUNTS_MAX];
+	lstat[type].counter[COUNTS_MIN] = min_t(u64, min, latency);
+	lstat[type].counter[COUNTS_MAX] = max_t(u64, max, latency);
+	lstat[type].counter[COUNTS_SUM] += latency;
+	lstat[type].samples++;
+	spin_unlock_irqrestore(&latency_lock, flags);
+}
+
+static char *latency_counter_names[] = {
+	"                  <0.1us",
+	"   0.1us-1us", "    1us-10us", "  10us-100us",
+	"   100us-1ms", "    1ms-10ms", "      >=10ms",
+	"     min(us)", "     max(us)", " average(us)"
+};
+
+static char *latency_type_names[] = {
+	"   inv_iotlb", "  inv_devtlb", "     inv_iec",
+	"     svm_prq"
+};
+
+void dmar_latency_snapshot(struct intel_iommu *iommu, char *str, size_t size)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+	unsigned long flags;
+	int bytes = 0, i, j;
+
+	memset(str, 0, size);
+
+	for (i = 0; i < COUNTS_NUM; i++)
+		bytes += scnprintf(str + bytes, size - bytes,
+				  "%s", latency_counter_names[i]);
+
+	spin_lock_irqsave(&latency_lock, flags);
+	for (i = 0; i < DMAR_LATENCY_NUM; i++) {
+		if (!dmar_latency_enabled(iommu, i))
+			continue;
+
+		bytes += scnprintf(str + bytes, size - bytes,
+				  "\n%s", latency_type_names[i]);
+
+		for (j = 0; j < COUNTS_NUM; j++) {
+			u64 val = lstat[i].counter[j];
+
+			switch (j) {
+			case COUNTS_MIN:
+				if (val == UINT_MAX)
+					val = 0;
+				else
+					val = div_u64(val, 1000);
+				break;
+			case COUNTS_MAX:
+				val = div_u64(val, 1000);
+				break;
+			case COUNTS_SUM:
+				if (lstat[i].samples)
+					val = div_u64(val, (lstat[i].samples * 1000));
+				else
+					val = 0;
+				break;
+			default:
+				break;
+			}
+
+			bytes += scnprintf(str + bytes, size - bytes,
+					  "%12lld", val);
+		}
+	}
+	spin_unlock_irqrestore(&latency_lock, flags);
+}