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, 146 insertions, 0 deletions

diff --git a/Documentation/driver-api/driver-model/bus.rst b/Documentation/driver-api/driver-model/bus.rst
new file mode 100644
index 000000000..9709ab62a
--- /dev/null
+++ b/Documentation/driver-api/driver-model/bus.rst
@@ -0,0 +1,146 @@
+=========
+Bus Types
+=========
+
+Definition
+~~~~~~~~~~
+See the kerneldoc for the struct bus_type.
+
+int bus_register(struct bus_type * bus);
+
+
+Declaration
+~~~~~~~~~~~
+
+Each bus type in the kernel (PCI, USB, etc) should declare one static
+object of this type. They must initialize the name field, and may
+optionally initialize the match callback::
+
+   struct bus_type pci_bus_type = {
+          .name	= "pci",
+          .match	= pci_bus_match,
+   };
+
+The structure should be exported to drivers in a header file:
+
+extern struct bus_type pci_bus_type;
+
+
+Registration
+~~~~~~~~~~~~
+
+When a bus driver is initialized, it calls bus_register. This
+initializes the rest of the fields in the bus object and inserts it
+into a global list of bus types. Once the bus object is registered,
+the fields in it are usable by the bus driver.
+
+
+Callbacks
+~~~~~~~~~
+
+match(): Attaching Drivers to Devices
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The format of device ID structures and the semantics for comparing
+them are inherently bus-specific. Drivers typically declare an array
+of device IDs of devices they support that reside in a bus-specific
+driver structure.
+
+The purpose of the match callback is to give the bus an opportunity to
+determine if a particular driver supports a particular device by
+comparing the device IDs the driver supports with the device ID of a
+particular device, without sacrificing bus-specific functionality or
+type-safety.
+
+When a driver is registered with the bus, the bus's list of devices is
+iterated over, and the match callback is called for each device that
+does not have a driver associated with it.
+
+
+
+Device and Driver Lists
+~~~~~~~~~~~~~~~~~~~~~~~
+
+The lists of devices and drivers are intended to replace the local
+lists that many buses keep. They are lists of struct devices and
+struct device_drivers, respectively. Bus drivers are free to use the
+lists as they please, but conversion to the bus-specific type may be
+necessary.
+
+The LDM core provides helper functions for iterating over each list::
+
+  int bus_for_each_dev(struct bus_type * bus, struct device * start,
+		       void * data,
+		       int (*fn)(struct device *, void *));
+
+  int bus_for_each_drv(struct bus_type * bus, struct device_driver * start,
+		       void * data, int (*fn)(struct device_driver *, void *));
+
+These helpers iterate over the respective list, and call the callback
+for each device or driver in the list. All list accesses are
+synchronized by taking the bus's lock (read currently). The reference
+count on each object in the list is incremented before the callback is
+called; it is decremented after the next object has been obtained. The
+lock is not held when calling the callback.
+
+
+sysfs
+~~~~~~~~
+There is a top-level directory named 'bus'.
+
+Each bus gets a directory in the bus directory, along with two default
+directories::
+
+	/sys/bus/pci/
+	|-- devices
+	`-- drivers
+
+Drivers registered with the bus get a directory in the bus's drivers
+directory::
+
+	/sys/bus/pci/
+	|-- devices
+	`-- drivers
+	    |-- Intel ICH
+	    |-- Intel ICH Joystick
+	    |-- agpgart
+	    `-- e100
+
+Each device that is discovered on a bus of that type gets a symlink in
+the bus's devices directory to the device's directory in the physical
+hierarchy::
+
+	/sys/bus/pci/
+	|-- devices
+	|   |-- 00:00.0 -> ../../../root/pci0/00:00.0
+	|   |-- 00:01.0 -> ../../../root/pci0/00:01.0
+	|   `-- 00:02.0 -> ../../../root/pci0/00:02.0
+	`-- drivers
+
+
+Exporting Attributes
+~~~~~~~~~~~~~~~~~~~~
+
+::
+
+  struct bus_attribute {
+	struct attribute	attr;
+	ssize_t (*show)(const struct bus_type *, char * buf);
+	ssize_t (*store)(const struct bus_type *, const char * buf, size_t count);
+  };
+
+Bus drivers can export attributes using the BUS_ATTR_RW macro that works
+similarly to the DEVICE_ATTR_RW macro for devices. For example, a
+definition like this::
+
+	static BUS_ATTR_RW(debug);
+
+is equivalent to declaring::
+
+	static bus_attribute bus_attr_debug;
+
+This can then be used to add and remove the attribute from the bus's
+sysfs directory using::
+
+	int bus_create_file(struct bus_type *, struct bus_attribute *);
+	void bus_remove_file(struct bus_type *, struct bus_attribute *);