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, 203 insertions, 0 deletions

diff --git a/Documentation/driver-api/media/media-committers.rst b/Documentation/driver-api/media/media-committers.rst
new file mode 100644
index 000000000..a905856f6
--- /dev/null
+++ b/Documentation/driver-api/media/media-committers.rst
@@ -0,0 +1,203 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+.. _Media Committers:
+
+Media Committers
+================
+
+Who is a Media Committer?
+-------------------------
+
+A Media Committer is a Media Maintainer with patchwork access who has been
+granted commit access to push patches from other developers and their own
+patches to the
+`media-committers <https://gitlab.freedesktop.org/linux-media/media-committers>`_
+tree.
+
+These commit rights are granted with expectation of responsibility:
+committers are people who care about the Linux Kernel as a whole and
+about the Linux media subsystem and want to advance its development. It
+is also based on a trust relationship among other committers, maintainers
+and the Linux Media community.
+
+As Media Committer you have the following additional responsibilities:
+
+1. Patches you authored must have a ``Signed-off-by``, ``Reviewed-by``
+   or ``Acked-by`` from another Media Maintainer;
+2. If a patch introduces a regression, then that must be corrected as soon
+   as possible. Typically the patch is either reverted, or an additional
+   patch is committed to fix the regression;
+3. If patches are fixing bugs against already released Kernels, including
+   the reverts mentioned above, the Media Committer shall add the needed
+   tags. Please see :ref:`Media development workflow` for more details.
+4. All Media Committers are responsible for maintaining
+   `Patchwork <https://patchwork.linuxtv.org/project/linux-media/list/>`_,
+   updating the state of the patches they review or merge.
+
+
+Becoming a Media Committer
+--------------------------
+
+Existing Media Committers can nominate a Media Maintainer to be granted
+commit rights. The Media Maintainer must have patchwork access,
+have been reviewing patches from third parties for some time, and has
+demonstrated a good understanding of the maintainer's duties and processes.
+
+The ultimate responsibility for accepting a nominated committer is up to
+the Media Subsystem Maintainers. The nominated committer must have earned a
+trust relationship with all Media Subsystem Maintainers, as, by granting you
+commit rights, part of their responsibilities are handed over to you.
+
+Due to that, to become a Media Committer, a consensus between all Media
+Subsystem Maintainers is required.
+
+.. Note::
+
+   In order to preserve/protect the developers that could have their commit
+   rights granted, denied or removed as well as the subsystem maintainers who
+   have the task to accept or deny commit rights, all communication related to
+   changing commit rights should happen in private as much as possible.
+
+.. _media-committer-agreement:
+
+Media Committer's agreement
+---------------------------
+
+Once a nominated committer is accepted by all Media Subsystem Maintainers,
+they will ask if the developer is interested in the nomination and discuss
+what area(s) of the media subsystem the committer will be responsible for.
+Those areas will typically be the same as the areas that the nominated
+committer is already maintaining.
+
+When the developer accepts being a committer, the new committer shall
+explicitly accept the Kernel development policies described under its
+Documentation/, and in particular to the rules in this document, by writing
+an e-mail to media-committers@linuxtv.org, with a declaration of intent
+following the model below::
+
+   I, John Doe, would like to change my status to: Committer
+
+   As Media Maintainer I accept commit rights for the following areas of
+   the media subsystem:
+
+   ...
+
+   For the purpose of committing patches to the media-committers tree,
+   I'll be using my user https://gitlab.freedesktop.org/users/<username>.
+
+Followed by a formal declaration of agreement with the Kernel development
+rules::
+
+   I agree to follow the Kernel development rules described at:
+
+   https://www.kernel.org/doc/html/latest/driver-api/media/media-committers.rst
+
+   and to the Linux Kernel development process rules.
+
+   I agree to abide by the Code of Conduct as documented in:
+   https://www.kernel.org/doc/html/latest/process/code-of-conduct.rst
+
+   I am aware that I can, at any point of time, retire. In that case, I will
+   send an e-mail to notify the Media Subsystem Maintainers for them to revoke
+   my commit rights.
+
+   I am aware that the Kernel development rules change over time.
+   By doing a new push to media-committers tree, I understand that I agree
+   to follow the rules in effect at the time of the commit.
+
+That e-mail shall be signed via the Kernel Web of trust with a PGP key cross
+signed by other Kernel and media developers. As described at
+:ref:`media-developers-gpg`, the PGP signature, together with the gitlab user
+security are fundamental components that ensure the authenticity of the merge
+requests that will happen at the media-committers.git tree.
+
+In case the kernel development process changes, by merging new commits to the
+`media-committers tree <https://gitlab.freedesktop.org/linux-media/media-committers>`_,
+the Media Committer implicitly declares their agreement with the latest
+version of the documented process including the contents of this file.
+
+If a Media Committer decides to retire, it is the committer's duty to
+notify the Media Subsystem Maintainers about that decision.
+
+.. note::
+
+   1. Changes to the kernel media development process shall be announced in
+      the media-committers mailing list with a reasonable review period. All
+      committers are automatically subscribed to that mailing list;
+   2. Due to the distributed nature of the Kernel development, it is
+      possible that kernel development process changes may end being
+      reviewed/merged at the Linux Docs and/or at the Linux Kernel mailing
+      lists, especially for the contents under Documentation/process and for
+      trivial typo fixes.
+
+Media Core Committers
+---------------------
+
+A Media Core Committer is a Media Core Maintainer with commit rights.
+
+As described in Documentation/driver-api/media/maintainer-entry-profile.rst,
+a Media Core Maintainer maintains media core frameworks as well, besides
+just drivers, and so is allowed to change core files and the media subsystem's
+Kernel API. The extent of the core committer's grants will be detailed by the
+Media Subsystem Maintainers when they nominate a Media Core Committer.
+
+Existing Media Committers may become Media Core Committers and vice versa.
+Such decisions will be taken in consensus among the Media Subsystem
+Maintainers.
+
+Media committers rules
+----------------------
+
+Media committers shall do their best efforts to avoid merging patches that
+would break any existing drivers. If it breaks, fixup or revert patches
+shall be merged as soon as possible, aiming to be merged at the same Kernel
+cycle the bug is reported.
+
+Media committers shall behave accordingly to the rights granted by
+the Media Subsystem Maintainers, especially with regards of the scope of changes
+they may apply directly at the media-committers tree. That scope can
+change over time on a mutual agreement between Media Committers and
+Media Subsystem Maintainers.
+
+The Media Committer workflow is described at :ref:`Media development workflow`.
+
+.. _Maintain Media Status:
+
+Maintaining Media Maintainer or Committer status
+------------------------------------------------
+
+A community of maintainers working together to move the Linux Kernel
+forward is essential to creating successful projects that are rewarding
+to work on. If there are problems or disagreements within the community,
+they can usually be solved through healthy discussion and debate.
+
+In the unhappy event that a Media Maintainer or Committer continues to
+disregard good citizenship (or actively disrupts the project), we may need
+to revoke that person's status. In such cases, if someone suggests the
+revocation with a good reason, then after discussing this among the Media
+Maintainers, the final decision is taken by the Media Subsystem Maintainers.
+
+As the decision to become a Media Maintainer or Committer comes from a
+consensus between Media Subsystem Maintainers, a single Media Subsystem
+Maintainer not trusting the Media Maintainer or Committer anymore is enough
+to revoke their maintenance, Patchwork grants and/or commit rights.
+
+Having commit rights revoked doesn't prevent Media Maintainers to keep
+contributing to the subsystem either via the pull request or via email workflow
+as documented at the :ref:`Media development workflow`.
+
+If a maintainer is inactive for more than a couple of Kernel cycles,
+maintainers will try to reach you via e-mail. If not possible, they may
+revoke their maintainer/patchwork and committer rights and update MAINTAINERS
+file entries accordingly. If you wish to resume contributing as maintainer
+later on, then contact the Media Subsystem Maintainers to ask if your
+maintenance, Patchwork grants and commit rights can be restored.
+
+References
+----------
+
+Much of this was inspired by/copied from the committer policies of:
+
+- `Chromium <https://chromium.googlesource.com/chromium/src/+/main/docs/contributing.md>`_;
+- `WebKit <https://webkit.org/commit-and-review-policy/>`_;
+- `Mozilla <https://www.mozilla.org/hacking/committer/>`_.