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, 120 insertions, 0 deletions

diff --git a/Documentation/driver-api/driver-model/device.rst b/Documentation/driver-api/driver-model/device.rst
new file mode 100644
index 000000000..0833be568
--- /dev/null
+++ b/Documentation/driver-api/driver-model/device.rst
@@ -0,0 +1,120 @@
+==========================
+The Basic Device Structure
+==========================
+
+See the kerneldoc for the struct device.
+
+
+Programming Interface
+~~~~~~~~~~~~~~~~~~~~~
+The bus driver that discovers the device uses this to register the
+device with the core::
+
+  int device_register(struct device * dev);
+
+The bus should initialize the following fields:
+
+    - parent
+    - name
+    - bus_id
+    - bus
+
+A device is removed from the core when its reference count goes to
+0. The reference count can be adjusted using::
+
+  struct device * get_device(struct device * dev);
+  void put_device(struct device * dev);
+
+get_device() will return a pointer to the struct device passed to it
+if the reference is not already 0 (if it's in the process of being
+removed already).
+
+A driver can access the lock in the device structure using::
+
+  void lock_device(struct device * dev);
+  void unlock_device(struct device * dev);
+
+
+Attributes
+~~~~~~~~~~
+
+::
+
+  struct device_attribute {
+	struct attribute	attr;
+	ssize_t (*show)(struct device *dev, struct device_attribute *attr,
+			char *buf);
+	ssize_t (*store)(struct device *dev, struct device_attribute *attr,
+			 const char *buf, size_t count);
+  };
+
+Attributes of devices can be exported by a device driver through sysfs.
+
+Please see Documentation/filesystems/sysfs.rst for more information
+on how sysfs works.
+
+As explained in Documentation/core-api/kobject.rst, device attributes must be
+created before the KOBJ_ADD uevent is generated. The only way to realize
+that is by defining an attribute group.
+
+Attributes are declared using a macro called DEVICE_ATTR::
+
+  #define DEVICE_ATTR(name,mode,show,store)
+
+Example:::
+
+  static DEVICE_ATTR(type, 0444, type_show, NULL);
+  static DEVICE_ATTR(power, 0644, power_show, power_store);
+
+Helper macros are available for common values of mode, so the above examples
+can be simplified to:::
+
+  static DEVICE_ATTR_RO(type);
+  static DEVICE_ATTR_RW(power);
+
+This declares two structures of type struct device_attribute with respective
+names 'dev_attr_type' and 'dev_attr_power'. These two attributes can be
+organized as follows into a group::
+
+  static struct attribute *dev_attrs[] = {
+	&dev_attr_type.attr,
+	&dev_attr_power.attr,
+	NULL,
+  };
+
+  static struct attribute_group dev_group = {
+	.attrs = dev_attrs,
+  };
+
+  static const struct attribute_group *dev_groups[] = {
+	&dev_group,
+	NULL,
+  };
+
+A helper macro is available for the common case of a single group, so the
+above two structures can be declared using:::
+
+  ATTRIBUTE_GROUPS(dev);
+
+This array of groups can then be associated with a device by setting the
+group pointer in struct device before device_register() is invoked::
+
+        dev->groups = dev_groups;
+        device_register(dev);
+
+The device_register() function will use the 'groups' pointer to create the
+device attributes and the device_unregister() function will use this pointer
+to remove the device attributes.
+
+Word of warning:  While the kernel allows device_create_file() and
+device_remove_file() to be called on a device at any time, userspace has
+strict expectations on when attributes get created.  When a new device is
+registered in the kernel, a uevent is generated to notify userspace (like
+udev) that a new device is available.  If attributes are added after the
+device is registered, then userspace won't get notified and userspace will
+not know about the new attributes.
+
+This is important for device driver that need to publish additional
+attributes for a device at driver probe time.  If the device driver simply
+calls device_create_file() on the device structure passed to it, then
+userspace will never be notified of the new attributes.