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, 121 insertions, 0 deletions

diff --git a/Documentation/admin-guide/ldm.rst b/Documentation/admin-guide/ldm.rst
new file mode 100644
index 000000000..12c571368
--- /dev/null
+++ b/Documentation/admin-guide/ldm.rst
@@ -0,0 +1,121 @@
+==========================================
+LDM - Logical Disk Manager (Dynamic Disks)
+==========================================
+
+:Author: Originally Written by FlatCap - Richard Russon <ldm@flatcap.org>.
+:Last Updated: Anton Altaparmakov on 30 March 2007 for Windows Vista.
+
+Overview
+--------
+
+Windows 2000, XP, and Vista use a new partitioning scheme.  It is a complete
+replacement for the MSDOS style partitions.  It stores its information in a
+1MiB journalled database at the end of the physical disk.  The size of
+partitions is limited only by disk space.  The maximum number of partitions is
+nearly 2000.
+
+Any partitions created under the LDM are called "Dynamic Disks".  There are no
+longer any primary or extended partitions.  Normal MSDOS style partitions are
+now known as Basic Disks.
+
+If you wish to use Spanned, Striped, Mirrored or RAID 5 Volumes, you must use
+Dynamic Disks.  The journalling allows Windows to make changes to these
+partitions and filesystems without the need to reboot.
+
+Once the LDM driver has divided up the disk, you can use the MD driver to
+assemble any multi-partition volumes, e.g.  Stripes, RAID5.
+
+To prevent legacy applications from repartitioning the disk, the LDM creates a
+dummy MSDOS partition containing one disk-sized partition.  This is what is
+supported with the Linux LDM driver.
+
+A newer approach that has been implemented with Vista is to put LDM on top of a
+GPT label disk.  This is not supported by the Linux LDM driver yet.
+
+
+Example
+-------
+
+Below we have a 50MiB disk, divided into seven partitions.
+
+.. note::
+
+   The missing 1MiB at the end of the disk is where the LDM database is
+   stored.
+
++-------++--------------+---------+-----++--------------+---------+----+
+|Device || Offset Bytes | Sectors | MiB || Size   Bytes | Sectors | MiB|
++=======++==============+=========+=====++==============+=========+====+
+|hda    ||            0 |       0 |   0 ||     52428800 |  102400 |  50|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda1   ||     51380224 |  100352 |  49 ||      1048576 |    2048 |   1|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda2   ||        16384 |      32 |   0 ||      6979584 |   13632 |   6|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda3   ||      6995968 |   13664 |   6 ||     10485760 |   20480 |  10|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda4   ||     17481728 |   34144 |  16 ||      4194304 |    8192 |   4|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda5   ||     21676032 |   42336 |  20 ||      5242880 |   10240 |   5|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda6   ||     26918912 |   52576 |  25 ||     10485760 |   20480 |  10|
++-------++--------------+---------+-----++--------------+---------+----+
+|hda7   ||     37404672 |   73056 |  35 ||     13959168 |   27264 |  13|
++-------++--------------+---------+-----++--------------+---------+----+
+
+The LDM Database may not store the partitions in the order that they appear on
+disk, but the driver will sort them.
+
+When Linux boots, you will see something like::
+
+  hda: 102400 sectors w/32KiB Cache, CHS=50/64/32
+  hda: [LDM] hda1 hda2 hda3 hda4 hda5 hda6 hda7
+
+
+Compiling LDM Support
+---------------------
+
+To enable LDM, choose the following two options: 
+
+  - "Advanced partition selection" CONFIG_PARTITION_ADVANCED
+  - "Windows Logical Disk Manager (Dynamic Disk) support" CONFIG_LDM_PARTITION
+
+If you believe the driver isn't working as it should, you can enable the extra
+debugging code.  This will produce a LOT of output.  The option is:
+
+  - "Windows LDM extra logging" CONFIG_LDM_DEBUG
+
+N.B. The partition code cannot be compiled as a module.
+
+As with all the partition code, if the driver doesn't see signs of its type of
+partition, it will pass control to another driver, so there is no harm in
+enabling it.
+
+If you have Dynamic Disks but don't enable the driver, then all you will see
+is a dummy MSDOS partition filling the whole disk.  You won't be able to mount
+any of the volumes on the disk.
+
+
+Booting
+-------
+
+If you enable LDM support, then lilo is capable of booting from any of the
+discovered partitions.  However, grub does not understand the LDM partitioning
+and cannot boot from a Dynamic Disk.
+
+
+More Documentation
+------------------
+
+There is an Overview of the LDM together with complete Technical Documentation.
+It is available for download.
+
+  http://www.linux-ntfs.org/
+
+If you have any LDM questions that aren't answered in the documentation, email
+me.
+
+Cheers,
+    FlatCap - Richard Russon
+    ldm@flatcap.org
+