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, 66 insertions, 0 deletions

diff --git a/Documentation/admin-guide/jfs.rst b/Documentation/admin-guide/jfs.rst
new file mode 100644
index 000000000..9e12d936b
--- /dev/null
+++ b/Documentation/admin-guide/jfs.rst
@@ -0,0 +1,66 @@
+===========================================
+IBM's Journaled File System (JFS) for Linux
+===========================================
+
+JFS Homepage:  http://jfs.sourceforge.net/
+
+The following mount options are supported:
+
+(*) == default
+
+iocharset=name
+                Character set to use for converting from Unicode to
+		ASCII.  The default is to do no conversion.  Use
+		iocharset=utf8 for UTF-8 translations.  This requires
+		CONFIG_NLS_UTF8 to be set in the kernel .config file.
+		iocharset=none specifies the default behavior explicitly.
+
+resize=value
+                Resize the volume to <value> blocks.  JFS only supports
+		growing a volume, not shrinking it.  This option is only
+		valid during a remount, when the volume is mounted
+		read-write.  The resize keyword with no value will grow
+		the volume to the full size of the partition.
+
+nointegrity
+                Do not write to the journal.  The primary use of this option
+		is to allow for higher performance when restoring a volume
+		from backup media.  The integrity of the volume is not
+		guaranteed if the system abnormally abends.
+
+integrity(*)
+                Commit metadata changes to the journal.  Use this option to
+		remount a volume where the nointegrity option was
+		previously specified in order to restore normal behavior.
+
+errors=continue
+                        Keep going on a filesystem error.
+errors=remount-ro(*)
+                        Remount the filesystem read-only on an error.
+errors=panic
+                        Panic and halt the machine if an error occurs.
+
+uid=value
+                Override on-disk uid with specified value
+gid=value
+                Override on-disk gid with specified value
+umask=value
+                Override on-disk umask with specified octal value. For
+                directories, the execute bit will be set if the corresponding
+		read bit is set.
+
+discard=minlen, discard/nodiscard(*)
+                This enables/disables the use of discard/TRIM commands.
+		The discard/TRIM commands are sent to the underlying
+                block device when blocks are freed. This is useful for SSD
+                devices and sparse/thinly-provisioned LUNs.  The FITRIM ioctl
+		command is also available together with the nodiscard option.
+		The value of minlen specifies the minimum blockcount, when
+		a TRIM command to the block device is considered useful.
+		When no value is given to the discard option, it defaults to
+		64 blocks, which means 256KiB in JFS.
+		The minlen value of discard overrides the minlen value given
+		on an FITRIM ioctl().
+
+The JFS mailing list can be subscribed to by using the link labeled
+"Mail list Subscribe" at our web page http://jfs.sourceforge.net/