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, 158 insertions, 0 deletions

diff --git a/fs/nfs/cache_lib.c b/fs/nfs/cache_lib.c
new file mode 100644
index 000000000..9738a1ae9
--- /dev/null
+++ b/fs/nfs/cache_lib.c
@@ -0,0 +1,158 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * linux/fs/nfs/cache_lib.c
+ *
+ * Helper routines for the NFS client caches
+ *
+ * Copyright (c) 2009 Trond Myklebust <Trond.Myklebust@netapp.com>
+ */
+#include <linux/kmod.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/mount.h>
+#include <linux/namei.h>
+#include <linux/slab.h>
+#include <linux/sunrpc/cache.h>
+#include <linux/sunrpc/rpc_pipe_fs.h>
+#include <net/net_namespace.h>
+
+#include "cache_lib.h"
+
+#define NFS_CACHE_UPCALL_PATHLEN 256
+#define NFS_CACHE_UPCALL_TIMEOUT 15
+
+static char nfs_cache_getent_prog[NFS_CACHE_UPCALL_PATHLEN] =
+				"/sbin/nfs_cache_getent";
+static unsigned long nfs_cache_getent_timeout = NFS_CACHE_UPCALL_TIMEOUT;
+
+module_param_string(cache_getent, nfs_cache_getent_prog,
+		sizeof(nfs_cache_getent_prog), 0600);
+MODULE_PARM_DESC(cache_getent, "Path to the client cache upcall program");
+module_param_named(cache_getent_timeout, nfs_cache_getent_timeout, ulong, 0600);
+MODULE_PARM_DESC(cache_getent_timeout, "Timeout (in seconds) after which "
+		"the cache upcall is assumed to have failed");
+
+int nfs_cache_upcall(struct cache_detail *cd, char *entry_name)
+{
+	static char *envp[] = { "HOME=/",
+		"TERM=linux",
+		"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
+		NULL
+	};
+	char *argv[] = {
+		nfs_cache_getent_prog,
+		cd->name,
+		entry_name,
+		NULL
+	};
+	int ret = -EACCES;
+
+	if (nfs_cache_getent_prog[0] == '\0')
+		goto out;
+	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
+	/*
+	 * Disable the upcall mechanism if we're getting an ENOENT or
+	 * EACCES error. The admin can re-enable it on the fly by using
+	 * sysfs to set the 'cache_getent' parameter once the problem
+	 * has been fixed.
+	 */
+	if (ret == -ENOENT || ret == -EACCES)
+		nfs_cache_getent_prog[0] = '\0';
+out:
+	return ret > 0 ? 0 : ret;
+}
+
+/*
+ * Deferred request handling
+ */
+void nfs_cache_defer_req_put(struct nfs_cache_defer_req *dreq)
+{
+	if (refcount_dec_and_test(&dreq->count))
+		kfree(dreq);
+}
+
+static void nfs_dns_cache_revisit(struct cache_deferred_req *d, int toomany)
+{
+	struct nfs_cache_defer_req *dreq;
+
+	dreq = container_of(d, struct nfs_cache_defer_req, deferred_req);
+
+	complete(&dreq->completion);
+	nfs_cache_defer_req_put(dreq);
+}
+
+static struct cache_deferred_req *nfs_dns_cache_defer(struct cache_req *req)
+{
+	struct nfs_cache_defer_req *dreq;
+
+	dreq = container_of(req, struct nfs_cache_defer_req, req);
+	dreq->deferred_req.revisit = nfs_dns_cache_revisit;
+	refcount_inc(&dreq->count);
+
+	return &dreq->deferred_req;
+}
+
+struct nfs_cache_defer_req *nfs_cache_defer_req_alloc(void)
+{
+	struct nfs_cache_defer_req *dreq;
+
+	dreq = kzalloc_obj(*dreq);
+	if (dreq) {
+		init_completion(&dreq->completion);
+		refcount_set(&dreq->count, 1);
+		dreq->req.defer = nfs_dns_cache_defer;
+	}
+	return dreq;
+}
+
+int nfs_cache_wait_for_upcall(struct nfs_cache_defer_req *dreq)
+{
+	if (wait_for_completion_timeout(&dreq->completion,
+			nfs_cache_getent_timeout * HZ) == 0)
+		return -ETIMEDOUT;
+	return 0;
+}
+
+int nfs_cache_register_sb(struct super_block *sb, struct cache_detail *cd)
+{
+	int ret;
+	struct dentry *dir;
+
+	dir = rpc_d_lookup_sb(sb, "cache");
+	ret = sunrpc_cache_register_pipefs(dir, cd->name, 0600, cd);
+	dput(dir);
+	return ret;
+}
+
+int nfs_cache_register_net(struct net *net, struct cache_detail *cd)
+{
+	struct super_block *pipefs_sb;
+	int ret = 0;
+
+	sunrpc_init_cache_detail(cd);
+	pipefs_sb = rpc_get_sb_net(net);
+	if (pipefs_sb) {
+		ret = nfs_cache_register_sb(pipefs_sb, cd);
+		rpc_put_sb_net(net);
+		if (ret)
+			sunrpc_destroy_cache_detail(cd);
+	}
+	return ret;
+}
+
+void nfs_cache_unregister_sb(struct super_block *sb, struct cache_detail *cd)
+{
+	sunrpc_cache_unregister_pipefs(cd);
+}
+
+void nfs_cache_unregister_net(struct net *net, struct cache_detail *cd)
+{
+	struct super_block *pipefs_sb;
+
+	pipefs_sb = rpc_get_sb_net(net);
+	if (pipefs_sb) {
+		nfs_cache_unregister_sb(pipefs_sb, cd);
+		rpc_put_sb_net(net);
+	}
+	sunrpc_destroy_cache_detail(cd);
+}