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, 65 insertions, 0 deletions

diff --git a/Documentation/filesystems/locks.rst b/Documentation/filesystems/locks.rst
new file mode 100644
index 000000000..26429317d
--- /dev/null
+++ b/Documentation/filesystems/locks.rst
@@ -0,0 +1,65 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+==========================
+File Locking Release Notes
+==========================
+
+		Andy Walker <andy@lysaker.kvaerner.no>
+
+			    12 May 1997
+
+
+1. What's New?
+==============
+
+1.1 Broken Flock Emulation
+--------------------------
+
+The old flock(2) emulation in the kernel was swapped for proper BSD
+compatible flock(2) support in the 1.3.x series of kernels. With the
+release of the 2.1.x kernel series, support for the old emulation has
+been totally removed, so that we don't need to carry this baggage
+forever.
+
+This should not cause problems for anybody, since everybody using a
+2.1.x kernel should have updated their C library to a suitable version
+anyway (see the file "Documentation/process/changes.rst".)
+
+1.2 Allow Mixed Locks Again
+---------------------------
+
+1.2.1 Typical Problems - Sendmail
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+Because sendmail was unable to use the old flock() emulation, many sendmail
+installations use fcntl() instead of flock(). This is true of Slackware 3.0
+for example. This gave rise to some other subtle problems if sendmail was
+configured to rebuild the alias file. Sendmail tried to lock the aliases.dir
+file with fcntl() at the same time as the GDBM routines tried to lock this
+file with flock(). With pre 1.3.96 kernels this could result in deadlocks that,
+over time, or under a very heavy mail load, would eventually cause the kernel
+to lock solid with deadlocked processes.
+
+
+1.2.2 The Solution
+^^^^^^^^^^^^^^^^^^
+The solution I have chosen, after much experimentation and discussion,
+is to make flock() and fcntl() locks oblivious to each other. Both can
+exists, and neither will have any effect on the other.
+
+I wanted the two lock styles to be cooperative, but there were so many
+race and deadlock conditions that the current solution was the only
+practical one. It puts us in the same position as, for example, SunOS
+4.1.x and several other commercial Unices. The only OS's that support
+cooperative flock()/fcntl() are those that emulate flock() using
+fcntl(), with all the problems that implies.
+
+
+1.3 Mandatory Locking As A Mount Option
+---------------------------------------
+
+Mandatory locking was prior to this release a general configuration option
+that was valid for all mounted filesystems.  This had a number of inherent
+dangers, not the least of which was the ability to freeze an NFS server by
+asking it to read a file for which a mandatory lock existed.
+
+Such option was dropped in Kernel v5.14.