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, 134 insertions, 0 deletions

diff --git a/tools/perf/util/sample.c b/tools/perf/util/sample.c
new file mode 100644
index 000000000..bccc19e2a
--- /dev/null
+++ b/tools/perf/util/sample.c
@@ -0,0 +1,134 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#include "sample.h"
+
+#include <stdlib.h>
+#include <string.h>
+
+#include <elf.h>
+#include <linux/zalloc.h>
+
+#include "../../arch/x86/include/asm/insn.h"
+#include "debug.h"
+#include "evsel.h"
+#include "thread.h"
+
+#ifndef EM_CSKY
+#define EM_CSKY		252
+#endif
+#ifndef EM_LOONGARCH
+#define EM_LOONGARCH	258
+#endif
+
+void perf_sample__init(struct perf_sample *sample, bool all)
+{
+	if (all) {
+		memset(sample, 0, sizeof(*sample));
+	} else {
+		sample->evsel = NULL;
+		sample->user_regs = NULL;
+		sample->intr_regs = NULL;
+		sample->merged_callchain = false;
+		sample->callchain = NULL;
+	}
+}
+
+void perf_sample__exit(struct perf_sample *sample)
+{
+	evsel__put(sample->evsel);
+	sample->evsel = NULL;
+	zfree(&sample->user_regs);
+	zfree(&sample->intr_regs);
+	if (sample->merged_callchain) {
+		zfree(&sample->callchain);
+		sample->merged_callchain = false;
+	}
+}
+
+struct regs_dump *perf_sample__user_regs(struct perf_sample *sample)
+{
+	if (!sample->user_regs) {
+		sample->user_regs = zalloc(sizeof(*sample->user_regs));
+		if (!sample->user_regs)
+			pr_err("Failure to allocate sample user_regs");
+	}
+	return sample->user_regs;
+}
+
+
+struct regs_dump *perf_sample__intr_regs(struct perf_sample *sample)
+{
+	if (!sample->intr_regs) {
+		sample->intr_regs = zalloc(sizeof(*sample->intr_regs));
+		if (!sample->intr_regs)
+			pr_err("Failure to allocate sample intr_regs");
+	}
+	return sample->intr_regs;
+}
+
+static int elf_machine_max_instruction_length(uint16_t e_machine)
+{
+	switch (e_machine) {
+	/* Fixed 4-byte (32-bit) architectures */
+	case EM_AARCH64:
+	case EM_PPC:
+	case EM_PPC64:
+	case EM_MIPS:
+	case EM_SPARC:
+	case EM_SPARCV9:
+	case EM_ALPHA:
+	case EM_LOONGARCH:
+	case EM_PARISC:
+	case EM_SH:
+		return 4;
+
+	/* Variable length or mixed-mode architectures */
+	case EM_ARM:    /* Variable due to Thumb/Thumb-2 */
+	case EM_RISCV:  /* Variable due to Compressed (C) extension */
+	case EM_CSKY:   /* Variable (16 or 32 bit) */
+	case EM_ARC:    /* Variable (ARCompact) */
+		return 4;
+	case EM_S390:   /* Variable (2, 4, or 6 bytes) */
+		return 6;
+	case EM_68K:
+		return 10;
+	case EM_386:
+	case EM_X86_64:
+		return 15;
+	case EM_XTENSA: /* Variable (FLIX) */
+		return 16;
+	default:
+		return MAX_INSN;
+	}
+}
+
+void perf_sample__fetch_insn(struct perf_sample *sample,
+			     struct thread *thread,
+			     struct machine *machine)
+{
+	int ret, len;
+	bool is64bit = false;
+	uint16_t e_machine;
+
+	if (!sample->ip || sample->insn_len != 0)
+		return;
+
+	e_machine = thread__e_machine(thread, machine, /*e_flags=*/NULL);
+	len = elf_machine_max_instruction_length(e_machine);
+	len = thread__memcpy(thread, machine, sample->insn,
+			     sample->ip, len,
+			     &is64bit);
+	if (len <= 0)
+		return;
+
+	sample->insn_len = len;
+
+	if (e_machine == EM_386 || e_machine == EM_X86_64) {
+		/* Refine the x86 instruction length with the decoder. */
+		struct insn insn;
+
+		ret = insn_decode(&insn, sample->insn, len,
+				  is64bit ? INSN_MODE_64 : INSN_MODE_32);
+		if (ret >= 0 && insn.length <= len)
+			sample->insn_len = insn.length;
+	}
+}