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, 179 insertions, 0 deletions

diff --git a/drivers/mtd/mtdsuper.c b/drivers/mtd/mtdsuper.c
new file mode 100644
index 000000000..c709ff7aa
--- /dev/null
+++ b/drivers/mtd/mtdsuper.c
@@ -0,0 +1,179 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* MTD-based superblock management
+ *
+ * Copyright © 2001-2007 Red Hat, Inc. All Rights Reserved.
+ * Copyright © 2001-2010 David Woodhouse <dwmw2@infradead.org>
+ *
+ * Written by:  David Howells <dhowells@redhat.com>
+ *              David Woodhouse <dwmw2@infradead.org>
+ */
+
+#include <linux/mtd/super.h>
+#include <linux/namei.h>
+#include <linux/export.h>
+#include <linux/ctype.h>
+#include <linux/slab.h>
+#include <linux/major.h>
+#include <linux/backing-dev.h>
+#include <linux/blkdev.h>
+#include <linux/fs_context.h>
+#include "mtdcore.h"
+
+/*
+ * get a superblock on an MTD-backed filesystem
+ */
+static int mtd_get_sb(struct fs_context *fc,
+		      struct mtd_info *mtd,
+		      int (*fill_super)(struct super_block *,
+					struct fs_context *))
+{
+	struct super_block *sb;
+	int ret;
+
+	sb = sget_dev(fc, MKDEV(MTD_BLOCK_MAJOR, mtd->index));
+	if (IS_ERR(sb))
+		return PTR_ERR(sb);
+
+	if (sb->s_root) {
+		/* new mountpoint for an already mounted superblock */
+		pr_debug("MTDSB: Device %d (\"%s\") is already mounted\n",
+			 mtd->index, mtd->name);
+		put_mtd_device(mtd);
+	} else {
+		/* fresh new superblock */
+		pr_debug("MTDSB: New superblock for device %d (\"%s\")\n",
+			 mtd->index, mtd->name);
+
+		/*
+		 * Would usually have been set with @sb_lock held but in
+		 * contrast to sb->s_bdev that's checked with only
+		 * @sb_lock held, nothing checks sb->s_mtd without also
+		 * holding sb->s_umount and we're holding sb->s_umount
+		 * here.
+		 */
+		sb->s_mtd = mtd;
+		sb->s_bdi = bdi_get(mtd_bdi);
+
+		ret = fill_super(sb, fc);
+		if (ret < 0)
+			goto error_sb;
+
+		sb->s_flags |= SB_ACTIVE;
+	}
+
+	BUG_ON(fc->root);
+	fc->root = dget(sb->s_root);
+	return 0;
+
+error_sb:
+	deactivate_locked_super(sb);
+	return ret;
+}
+
+/*
+ * get a superblock on an MTD-backed filesystem by MTD device number
+ */
+static int mtd_get_sb_by_nr(struct fs_context *fc, int mtdnr,
+			    int (*fill_super)(struct super_block *,
+					      struct fs_context *))
+{
+	struct mtd_info *mtd;
+
+	mtd = get_mtd_device(NULL, mtdnr);
+	if (IS_ERR(mtd)) {
+		errorf(fc, "MTDSB: Device #%u doesn't appear to exist\n", mtdnr);
+		return PTR_ERR(mtd);
+	}
+
+	return mtd_get_sb(fc, mtd, fill_super);
+}
+
+/**
+ * get_tree_mtd - Get a superblock based on a single MTD device
+ * @fc: The filesystem context holding the parameters
+ * @fill_super: Helper to initialise a new superblock
+ */
+int get_tree_mtd(struct fs_context *fc,
+	      int (*fill_super)(struct super_block *sb,
+				struct fs_context *fc))
+{
+#ifdef CONFIG_BLOCK
+	dev_t dev;
+	int ret;
+#endif
+	int mtdnr;
+
+	if (!fc->source)
+		return invalf(fc, "No source specified");
+
+	pr_debug("MTDSB: dev_name \"%s\"\n", fc->source);
+
+	/* the preferred way of mounting in future; especially when
+	 * CONFIG_BLOCK=n - we specify the underlying MTD device by number or
+	 * by name, so that we don't require block device support to be present
+	 * in the kernel.
+	 */
+	if (fc->source[0] == 'm' &&
+	    fc->source[1] == 't' &&
+	    fc->source[2] == 'd') {
+		if (fc->source[3] == ':') {
+			struct mtd_info *mtd;
+
+			/* mount by MTD device name */
+			pr_debug("MTDSB: mtd:%%s, name \"%s\"\n",
+				 fc->source + 4);
+
+			mtd = get_mtd_device_nm(fc->source + 4);
+			if (!IS_ERR(mtd))
+				return mtd_get_sb(fc, mtd, fill_super);
+
+			errorf(fc, "MTD: MTD device with name \"%s\" not found",
+			       fc->source + 4);
+
+		} else if (isdigit(fc->source[3])) {
+			/* mount by MTD device number name */
+			unsigned int mtdnr_val;
+
+			if (kstrtouint(fc->source + 3, 0, &mtdnr_val) == 0) {
+				mtdnr = mtdnr_val;
+				/* It was a valid number */
+				pr_debug("MTDSB: mtd%%d, mtdnr %u\n", mtdnr_val);
+				return mtd_get_sb_by_nr(fc, mtdnr, fill_super);
+			}
+		}
+	}
+
+#ifdef CONFIG_BLOCK
+	/* try the old way - the hack where we allowed users to mount
+	 * /dev/mtdblock$(n) but didn't actually _use_ the blockdev
+	 */
+	ret = lookup_bdev(fc->source, &dev);
+	if (ret) {
+		errorf(fc, "MTD: Couldn't look up '%s': %d", fc->source, ret);
+		return ret;
+	}
+	pr_debug("MTDSB: lookup_bdev() returned 0\n");
+
+	if (MAJOR(dev) == MTD_BLOCK_MAJOR)
+		return mtd_get_sb_by_nr(fc, MINOR(dev), fill_super);
+
+#endif /* CONFIG_BLOCK */
+
+	if (!(fc->sb_flags & SB_SILENT))
+		errorf(fc, "MTD: Attempt to mount non-MTD device \"%s\"",
+		       fc->source);
+	return -EINVAL;
+}
+EXPORT_SYMBOL_GPL(get_tree_mtd);
+
+/*
+ * destroy an MTD-based superblock
+ */
+void kill_mtd_super(struct super_block *sb)
+{
+	generic_shutdown_super(sb);
+	put_mtd_device(sb->s_mtd);
+	sb->s_mtd = NULL;
+}
+
+EXPORT_SYMBOL_GPL(kill_mtd_super);