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, 124 insertions, 0 deletions

diff --git a/Documentation/driver-api/driver-model/overview.rst b/Documentation/driver-api/driver-model/overview.rst
new file mode 100644
index 000000000..b3f447bf9
--- /dev/null
+++ b/Documentation/driver-api/driver-model/overview.rst
@@ -0,0 +1,124 @@
+=============================
+The Linux Kernel Device Model
+=============================
+
+Patrick Mochel	<mochel@digitalimplant.org>
+
+Drafted 26 August 2002
+Updated 31 January 2006
+
+
+Overview
+~~~~~~~~
+
+The Linux Kernel Driver Model is a unification of all the disparate driver
+models that were previously used in the kernel. It is intended to augment the
+bus-specific drivers for bridges and devices by consolidating a set of data
+and operations into globally accessible data structures.
+
+Traditional driver models implemented some sort of tree-like structure
+(sometimes just a list) for the devices they control. There wasn't any
+uniformity across the different bus types.
+
+The current driver model provides a common, uniform data model for describing
+a bus and the devices that can appear under the bus. The unified bus
+model includes a set of common attributes which all buses carry, and a set
+of common callbacks, such as device discovery during bus probing, bus
+shutdown, bus power management, etc.
+
+The common device and bridge interface reflects the goals of the modern
+computer: namely the ability to do seamless device "plug and play", power
+management, and hot plug. In particular, the model dictated by Intel and
+Microsoft (namely ACPI) ensures that almost every device on almost any bus
+on an x86-compatible system can work within this paradigm.  Of course,
+not every bus is able to support all such operations, although most
+buses support most of those operations.
+
+
+Downstream Access
+~~~~~~~~~~~~~~~~~
+
+Common data fields have been moved out of individual bus layers into a common
+data structure. These fields must still be accessed by the bus layers,
+and sometimes by the device-specific drivers.
+
+Other bus layers are encouraged to do what has been done for the PCI layer.
+struct pci_dev now looks like this::
+
+  struct pci_dev {
+	...
+
+	struct device dev;     /* Generic device interface */
+	...
+  };
+
+Note first that the struct device dev within the struct pci_dev is
+statically allocated. This means only one allocation on device discovery.
+
+Note also that that struct device dev is not necessarily defined at the
+front of the pci_dev structure.  This is to make people think about what
+they're doing when switching between the bus driver and the global driver,
+and to discourage meaningless and incorrect casts between the two.
+
+The PCI bus layer freely accesses the fields of struct device. It knows about
+the structure of struct pci_dev, and it should know the structure of struct
+device. Individual PCI device drivers that have been converted to the current
+driver model generally do not and should not touch the fields of struct device,
+unless there is a compelling reason to do so.
+
+The above abstraction prevents unnecessary pain during transitional phases.
+If it were not done this way, then when a field was renamed or removed, every
+downstream driver would break.  On the other hand, if only the bus layer
+(and not the device layer) accesses the struct device, it is only the bus
+layer that needs to change.
+
+
+User Interface
+~~~~~~~~~~~~~~
+
+By virtue of having a complete hierarchical view of all the devices in the
+system, exporting a complete hierarchical view to userspace becomes relatively
+easy. This has been accomplished by implementing a special purpose virtual
+file system named sysfs.
+
+Almost all mainstream Linux distros mount this filesystem automatically; you
+can see some variation of the following in the output of the "mount" command::
+
+  $ mount
+  ...
+  none on /sys type sysfs (rw,noexec,nosuid,nodev)
+  ...
+  $
+
+The auto-mounting of sysfs is typically accomplished by an entry similar to
+the following in the /etc/fstab file::
+
+  none     	/sys	sysfs    defaults	  	0 0
+
+or something similar in the /lib/init/fstab file on Debian-based systems::
+
+  none            /sys    sysfs    nodev,noexec,nosuid    0 0
+
+If sysfs is not automatically mounted, you can always do it manually with::
+
+	# mount -t sysfs sysfs /sys
+
+Whenever a device is inserted into the tree, a directory is created for it.
+This directory may be populated at each layer of discovery - the global layer,
+the bus layer, or the device layer.
+
+The global layer currently creates two files - 'name' and 'power'. The
+former only reports the name of the device. The latter reports the
+current power state of the device. It will also be used to set the current
+power state.
+
+The bus layer may also create files for the devices it finds while probing the
+bus. For example, the PCI layer currently creates 'irq' and 'resource' files
+for each PCI device.
+
+A device-specific driver may also export files in its directory to expose
+device-specific data or tunable interfaces.
+
+More information about the sysfs directory layout can be found in
+the other documents in this directory and in the file
+Documentation/filesystems/sysfs.rst.