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/include/trace/events/vmalloc.h b/include/trace/events/vmalloc.h
new file mode 100644
index 000000000..ad4e02191
--- /dev/null
+++ b/include/trace/events/vmalloc.h
@@ -0,0 +1,123 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM vmalloc
+
+#if !defined(_TRACE_VMALLOC_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_VMALLOC_H
+
+#include <linux/tracepoint.h>
+
+/**
+ * alloc_vmap_area - called when a new vmap allocation occurs
+ * @addr:	an allocated address
+ * @size:	a requested size
+ * @align:	a requested alignment
+ * @vstart:	a requested start range
+ * @vend:	a requested end range
+ * @failed:	an allocation failed or not
+ *
+ * This event is used for a debug purpose, it can give an extra
+ * information for a developer about how often it occurs and which
+ * parameters are passed for further validation.
+ */
+TRACE_EVENT(alloc_vmap_area,
+
+	TP_PROTO(unsigned long addr, unsigned long size, unsigned long align,
+		unsigned long vstart, unsigned long vend, int failed),
+
+	TP_ARGS(addr, size, align, vstart, vend, failed),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, addr)
+		__field(unsigned long, size)
+		__field(unsigned long, align)
+		__field(unsigned long, vstart)
+		__field(unsigned long, vend)
+		__field(int, failed)
+	),
+
+	TP_fast_assign(
+		__entry->addr = addr;
+		__entry->size = size;
+		__entry->align = align;
+		__entry->vstart = vstart;
+		__entry->vend = vend;
+		__entry->failed = failed;
+	),
+
+	TP_printk("va_start: %lu size=%lu align=%lu vstart=0x%lx vend=0x%lx failed=%d",
+		__entry->addr, __entry->size, __entry->align,
+		__entry->vstart, __entry->vend, __entry->failed)
+);
+
+/**
+ * purge_vmap_area_lazy - called when vmap areas were lazily freed
+ * @start:		purging start address
+ * @end:		purging end address
+ * @npurged:	numbed of purged vmap areas
+ *
+ * This event is used for a debug purpose. It gives some
+ * indication about start:end range and how many objects
+ * are released.
+ */
+TRACE_EVENT(purge_vmap_area_lazy,
+
+	TP_PROTO(unsigned long start, unsigned long end,
+		unsigned int npurged),
+
+	TP_ARGS(start, end, npurged),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, start)
+		__field(unsigned long, end)
+		__field(unsigned int, npurged)
+	),
+
+	TP_fast_assign(
+		__entry->start = start;
+		__entry->end = end;
+		__entry->npurged = npurged;
+	),
+
+	TP_printk("start=0x%lx end=0x%lx num_purged=%u",
+		__entry->start, __entry->end, __entry->npurged)
+);
+
+/**
+ * free_vmap_area_noflush - called when a vmap area is freed
+ * @va_start:		a start address of VA
+ * @nr_lazy:		number of current lazy pages
+ * @nr_lazy_max:	number of maximum lazy pages
+ *
+ * This event is used for a debug purpose. It gives some
+ * indication about a VA that is released, number of current
+ * outstanding areas and a maximum allowed threshold before
+ * dropping all of them.
+ */
+TRACE_EVENT(free_vmap_area_noflush,
+
+	TP_PROTO(unsigned long va_start, unsigned long nr_lazy,
+		unsigned long nr_lazy_max),
+
+	TP_ARGS(va_start, nr_lazy, nr_lazy_max),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, va_start)
+		__field(unsigned long, nr_lazy)
+		__field(unsigned long, nr_lazy_max)
+	),
+
+	TP_fast_assign(
+		__entry->va_start = va_start;
+		__entry->nr_lazy = nr_lazy;
+		__entry->nr_lazy_max = nr_lazy_max;
+	),
+
+	TP_printk("va_start=0x%lx nr_lazy=%lu nr_lazy_max=%lu",
+		__entry->va_start, __entry->nr_lazy, __entry->nr_lazy_max)
+);
+
+#endif /*  _TRACE_VMALLOC_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>