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, 191 insertions, 0 deletions

diff --git a/samples/seccomp/bpf-direct.c b/samples/seccomp/bpf-direct.c
new file mode 100644
index 000000000..c09e4a17a
--- /dev/null
+++ b/samples/seccomp/bpf-direct.c
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Seccomp filter example for x86 (32-bit and 64-bit) with BPF macros
+ *
+ * Copyright (c) 2012 The Chromium OS Authors <chromium-os-dev@chromium.org>
+ * Author: Will Drewry <wad@chromium.org>
+ *
+ * The code may be used by anyone for any purpose,
+ * and can serve as a starting point for developing
+ * applications using prctl(PR_SET_SECCOMP, 2, ...).
+ */
+#if defined(__i386__) || defined(__x86_64__)
+#define SUPPORTED_ARCH 1
+#endif
+
+#if defined(SUPPORTED_ARCH)
+#define __USE_GNU 1
+#define _GNU_SOURCE 1
+
+#include <linux/types.h>
+#include <linux/filter.h>
+#include <linux/seccomp.h>
+#include <linux/unistd.h>
+#include <signal.h>
+#include <stdio.h>
+#include <stddef.h>
+#include <string.h>
+#include <sys/prctl.h>
+#include <unistd.h>
+
+#define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
+#define syscall_nr (offsetof(struct seccomp_data, nr))
+
+#if defined(__i386__)
+#define REG_RESULT	REG_EAX
+#define REG_SYSCALL	REG_EAX
+#define REG_ARG0	REG_EBX
+#define REG_ARG1	REG_ECX
+#define REG_ARG2	REG_EDX
+#define REG_ARG3	REG_ESI
+#define REG_ARG4	REG_EDI
+#define REG_ARG5	REG_EBP
+#elif defined(__x86_64__)
+#define REG_RESULT	REG_RAX
+#define REG_SYSCALL	REG_RAX
+#define REG_ARG0	REG_RDI
+#define REG_ARG1	REG_RSI
+#define REG_ARG2	REG_RDX
+#define REG_ARG3	REG_R10
+#define REG_ARG4	REG_R8
+#define REG_ARG5	REG_R9
+#endif
+
+#ifndef PR_SET_NO_NEW_PRIVS
+#define PR_SET_NO_NEW_PRIVS 38
+#endif
+
+#ifndef SYS_SECCOMP
+#define SYS_SECCOMP 1
+#endif
+
+static void emulator(int nr, siginfo_t *info, void *void_context)
+{
+	ucontext_t *ctx = (ucontext_t *)(void_context);
+	int syscall;
+	char *buf;
+	ssize_t bytes;
+	size_t len;
+	if (info->si_code != SYS_SECCOMP)
+		return;
+	if (!ctx)
+		return;
+	syscall = ctx->uc_mcontext.gregs[REG_SYSCALL];
+	buf = (char *) ctx->uc_mcontext.gregs[REG_ARG1];
+	len = (size_t) ctx->uc_mcontext.gregs[REG_ARG2];
+
+	if (syscall != __NR_write)
+		return;
+	if (ctx->uc_mcontext.gregs[REG_ARG0] != STDERR_FILENO)
+		return;
+	/* Redirect stderr messages to stdout. Doesn't handle EINTR, etc */
+	ctx->uc_mcontext.gregs[REG_RESULT] = -1;
+	if (write(STDOUT_FILENO, "[ERR] ", 6) > 0) {
+		bytes = write(STDOUT_FILENO, buf, len);
+		ctx->uc_mcontext.gregs[REG_RESULT] = bytes;
+	}
+	return;
+}
+
+static int install_emulator(void)
+{
+	struct sigaction act;
+	sigset_t mask;
+	memset(&act, 0, sizeof(act));
+	sigemptyset(&mask);
+	sigaddset(&mask, SIGSYS);
+
+	act.sa_sigaction = &emulator;
+	act.sa_flags = SA_SIGINFO;
+	if (sigaction(SIGSYS, &act, NULL) < 0) {
+		perror("sigaction");
+		return -1;
+	}
+	if (sigprocmask(SIG_UNBLOCK, &mask, NULL)) {
+		perror("sigprocmask");
+		return -1;
+	}
+	return 0;
+}
+
+static int install_filter(void)
+{
+	struct sock_filter filter[] = {
+		/* Grab the system call number */
+		BPF_STMT(BPF_LD+BPF_W+BPF_ABS, syscall_nr),
+		/* Jump table for the allowed syscalls */
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_rt_sigreturn, 0, 1),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
+#ifdef __NR_sigreturn
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_sigreturn, 0, 1),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
+#endif
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_exit_group, 0, 1),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_exit, 0, 1),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_read, 1, 0),
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_write, 3, 2),
+
+		/* Check that read is only using stdin. */
+		BPF_STMT(BPF_LD+BPF_W+BPF_ABS, syscall_arg(0)),
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, STDIN_FILENO, 4, 0),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_KILL),
+
+		/* Check that write is only using stdout */
+		BPF_STMT(BPF_LD+BPF_W+BPF_ABS, syscall_arg(0)),
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, STDOUT_FILENO, 1, 0),
+		/* Trap attempts to write to stderr */
+		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, STDERR_FILENO, 1, 2),
+
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_TRAP),
+		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_KILL),
+	};
+	struct sock_fprog prog = {
+		.len = (unsigned short)(sizeof(filter)/sizeof(filter[0])),
+		.filter = filter,
+	};
+
+	if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
+		perror("prctl(NO_NEW_PRIVS)");
+		return 1;
+	}
+
+
+	if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog)) {
+		perror("prctl");
+		return 1;
+	}
+	return 0;
+}
+
+#define payload(_c) (_c), sizeof((_c))
+int main(int argc, char **argv)
+{
+	char buf[4096];
+	ssize_t bytes = 0;
+	if (install_emulator())
+		return 1;
+	if (install_filter())
+		return 1;
+	syscall(__NR_write, STDOUT_FILENO,
+		payload("OHAI! WHAT IS YOUR NAME? "));
+	bytes = syscall(__NR_read, STDIN_FILENO, buf, sizeof(buf));
+	syscall(__NR_write, STDOUT_FILENO, payload("HELLO, "));
+	syscall(__NR_write, STDOUT_FILENO, buf, bytes);
+	syscall(__NR_write, STDERR_FILENO,
+		payload("Error message going to STDERR\n"));
+	return 0;
+}
+#else	/* SUPPORTED_ARCH */
+/*
+ * This sample is x86-only.  Since kernel samples are compiled with the
+ * host toolchain, a non-x86 host will result in using only the main()
+ * below.
+ */
+int main(void)
+{
+	return 1;
+}
+#endif	/* SUPPORTED_ARCH */