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, 91 insertions, 0 deletions

diff --git a/Documentation/security/sak.rst b/Documentation/security/sak.rst
new file mode 100644
index 000000000..260e1d368
--- /dev/null
+++ b/Documentation/security/sak.rst
@@ -0,0 +1,91 @@
+=========================================
+Linux Secure Attention Key (SAK) handling
+=========================================
+
+:Date: 18 March 2001
+:Author: Andrew Morton
+
+An operating system's Secure Attention Key is a security tool which is
+provided as protection against trojan password capturing programs.  It
+is an undefeatable way of killing all programs which could be
+masquerading as login applications.  Users need to be taught to enter
+this key sequence before they log in to the system.
+
+From the PC keyboard, Linux has two similar but different ways of
+providing SAK.  One is the ALT-SYSRQ-K sequence.  You shouldn't use
+this sequence.  It is only available if the kernel was compiled with
+sysrq support.
+
+The proper way of generating a SAK is to define the key sequence using
+``loadkeys``.  This will work whether or not sysrq support is compiled
+into the kernel.
+
+SAK works correctly when the keyboard is in raw mode.  This means that
+once defined, SAK will kill a running X server.  If the system is in
+run level 5, the X server will restart.  This is what you want to
+happen.
+
+What key sequence should you use? Well, CTRL-ALT-DEL is used to reboot
+the machine.  CTRL-ALT-BACKSPACE is magical to the X server.  We'll
+choose CTRL-ALT-PAUSE.
+
+In your rc.sysinit (or rc.local) file, add the command::
+
+	echo "control alt keycode 101 = SAK" | /bin/loadkeys
+
+And that's it!  Only the superuser may reprogram the SAK key.
+
+
+.. note::
+
+  1. Linux SAK is said to be not a "true SAK" as is required by
+     systems which implement C2 level security.  This author does not
+     know why.
+
+
+  2. On the PC keyboard, SAK kills all applications which have
+     /dev/console opened.
+
+     Unfortunately this includes a number of things which you don't
+     actually want killed.  This is because these applications are
+     incorrectly holding /dev/console open.  Be sure to complain to your
+     Linux distributor about this!
+
+     You can identify processes which will be killed by SAK with the
+     command::
+
+	# ls -l /proc/[0-9]*/fd/* | grep console
+	l-wx------    1 root     root           64 Mar 18 00:46 /proc/579/fd/0 -> /dev/console
+
+     Then::
+
+	# ps aux|grep 579
+	root       579  0.0  0.1  1088  436 ?        S    00:43   0:00 gpm -t ps/2
+
+     So ``gpm`` will be killed by SAK.  This is a bug in gpm.  It should
+     be closing standard input.  You can work around this by finding the
+     initscript which launches gpm and changing it thusly:
+
+     Old::
+
+	daemon gpm
+
+     New::
+
+	daemon gpm < /dev/null
+
+     Vixie cron also seems to have this problem, and needs the same treatment.
+
+     Also, one prominent Linux distribution has the following three
+     lines in its rc.sysinit and rc scripts::
+
+	exec 3<&0
+	exec 4>&1
+	exec 5>&2
+
+     These commands cause **all** daemons which are launched by the
+     initscripts to have file descriptors 3, 4 and 5 attached to
+     /dev/console.  So SAK kills them all.  A workaround is to simply
+     delete these lines, but this may cause system management
+     applications to malfunction - test everything well.
+