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

3 files changed, 72 insertions, 0 deletions

diff --git a/Documentation/admin-guide/namespaces/compatibility-list.rst b/Documentation/admin-guide/namespaces/compatibility-list.rst
new file mode 100644
index 000000000..318800b2a
--- /dev/null
+++ b/Documentation/admin-guide/namespaces/compatibility-list.rst
@@ -0,0 +1,43 @@
+=============================
+Namespaces compatibility list
+=============================
+
+This document contains the information about the problems user
+may have when creating tasks living in different namespaces.
+
+Here's the summary. This matrix shows the known problems, that
+occur when tasks share some namespace (the columns) while living
+in different other namespaces (the rows):
+
+====	===	===	===	===	====	===
+-	UTS	IPC	VFS	PID	User	Net
+====	===	===	===	===	====	===
+UTS	 X
+IPC		 X	 1
+VFS			 X
+PID		 1	 1	 X
+User		 2	 2		 X
+Net						 X
+====	===	===	===	===	====	===
+
+1. Both the IPC and the PID namespaces provide IDs to address
+   object inside the kernel. E.g. semaphore with IPCID or
+   process group with pid.
+
+   In both cases, tasks shouldn't try exposing this ID to some
+   other task living in a different namespace via a shared filesystem
+   or IPC shmem/message. The fact is that this ID is only valid
+   within the namespace it was obtained in and may refer to some
+   other object in another namespace.
+
+2. Intentionally, two equal user IDs in different user namespaces
+   should not be equal from the VFS point of view. In other
+   words, user 10 in one user namespace shouldn't have the same
+   access permissions to files, belonging to user 10 in another
+   namespace.
+
+   The same is true for the IPC namespaces being shared - two users
+   from different user namespaces should not access the same IPC objects
+   even having equal UIDs.
+
+   But currently this is not so.
diff --git a/Documentation/admin-guide/namespaces/index.rst b/Documentation/admin-guide/namespaces/index.rst
new file mode 100644
index 000000000..384f2e0f3
--- /dev/null
+++ b/Documentation/admin-guide/namespaces/index.rst
@@ -0,0 +1,11 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+==========
+Namespaces
+==========
+
+.. toctree::
+   :maxdepth: 1
+
+   compatibility-list
+   resource-control
diff --git a/Documentation/admin-guide/namespaces/resource-control.rst b/Documentation/admin-guide/namespaces/resource-control.rst
new file mode 100644
index 000000000..553a44803
--- /dev/null
+++ b/Documentation/admin-guide/namespaces/resource-control.rst
@@ -0,0 +1,18 @@
+====================================
+User namespaces and resource control
+====================================
+
+The kernel contains many kinds of objects that either don't have
+individual limits or that have limits which are ineffective when
+a set of processes is allowed to switch their UID. On a system
+where the admins don't trust their users or their users' programs,
+user namespaces expose the system to potential misuse of resources.
+
+In order to mitigate this, we recommend that admins enable memory
+control groups on any system that enables user namespaces.
+Furthermore, we recommend that admins configure the memory control
+groups to limit the maximum memory usable by any untrusted user.
+
+Memory control groups can be configured by installing the libcgroup
+package present on most distros editing /etc/cgrules.conf,
+/etc/cgconfig.conf and setting up libpam-cgroup.