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, 125 insertions, 0 deletions

diff --git a/include/drm/drm_module.h b/include/drm/drm_module.h
new file mode 100644
index 000000000..4db1ae03d
--- /dev/null
+++ b/include/drm/drm_module.h
@@ -0,0 +1,125 @@
+/* SPDX-License-Identifier: MIT */
+
+#ifndef DRM_MODULE_H
+#define DRM_MODULE_H
+
+#include <linux/pci.h>
+#include <linux/platform_device.h>
+
+#include <drm/drm_drv.h>
+
+/**
+ * DOC: overview
+ *
+ * This library provides helpers registering DRM drivers during module
+ * initialization and shutdown. The provided helpers act like bus-specific
+ * module helpers, such as module_pci_driver(), but respect additional
+ * parameters that control DRM driver registration.
+ *
+ * Below is an example of initializing a DRM driver for a device on the
+ * PCI bus.
+ *
+ * .. code-block:: c
+ *
+ *	struct pci_driver my_pci_drv = {
+ *	};
+ *
+ *	drm_module_pci_driver(my_pci_drv);
+ *
+ * The generated code will test if DRM drivers are enabled and register
+ * the PCI driver my_pci_drv. For more complex module initialization, you
+ * can still use module_init() and module_exit() in your driver.
+ */
+
+/*
+ * PCI drivers
+ */
+
+static inline int __init drm_pci_register_driver(struct pci_driver *pci_drv)
+{
+	if (drm_firmware_drivers_only())
+		return -ENODEV;
+
+	return pci_register_driver(pci_drv);
+}
+
+/**
+ * drm_module_pci_driver - Register a DRM driver for PCI-based devices
+ * @__pci_drv: the PCI driver structure
+ *
+ * Registers a DRM driver for devices on the PCI bus. The helper
+ * macro behaves like module_pci_driver() but tests the state of
+ * drm_firmware_drivers_only(). For more complex module initialization,
+ * use module_init() and module_exit() directly.
+ *
+ * Each module may only use this macro once. Calling it replaces
+ * module_init() and module_exit().
+ */
+#define drm_module_pci_driver(__pci_drv) \
+	module_driver(__pci_drv, drm_pci_register_driver, pci_unregister_driver)
+
+static inline int __init
+drm_pci_register_driver_if_modeset(struct pci_driver *pci_drv, int modeset)
+{
+	if (drm_firmware_drivers_only() && modeset == -1)
+		return -ENODEV;
+	if (modeset == 0)
+		return -ENODEV;
+
+	return pci_register_driver(pci_drv);
+}
+
+static inline void __exit
+drm_pci_unregister_driver_if_modeset(struct pci_driver *pci_drv, int modeset)
+{
+	pci_unregister_driver(pci_drv);
+}
+
+/**
+ * drm_module_pci_driver_if_modeset - Register a DRM driver for PCI-based devices
+ * @__pci_drv: the PCI driver structure
+ * @__modeset: an additional parameter that disables the driver
+ *
+ * This macro is deprecated and only provided for existing drivers. For
+ * new drivers, use drm_module_pci_driver().
+ *
+ * Registers a DRM driver for devices on the PCI bus. The helper macro
+ * behaves like drm_module_pci_driver() with an additional driver-specific
+ * flag. If __modeset is 0, the driver has been disabled, if __modeset is
+ * -1 the driver state depends on the global DRM state. For all other
+ * values, the PCI driver has been enabled. The default should be -1.
+ */
+#define drm_module_pci_driver_if_modeset(__pci_drv, __modeset) \
+	module_driver(__pci_drv, drm_pci_register_driver_if_modeset, \
+		      drm_pci_unregister_driver_if_modeset, __modeset)
+
+/*
+ * Platform drivers
+ */
+
+static inline int __init
+drm_platform_driver_register(struct platform_driver *platform_drv)
+{
+	if (drm_firmware_drivers_only())
+		return -ENODEV;
+
+	return platform_driver_register(platform_drv);
+}
+
+/**
+ * drm_module_platform_driver - Register a DRM driver for platform devices
+ * @__platform_drv: the platform driver structure
+ *
+ * Registers a DRM driver for devices on the platform bus. The helper
+ * macro behaves like module_platform_driver() but tests the state of
+ * drm_firmware_drivers_only(). For more complex module initialization,
+ * use module_init() and module_exit() directly.
+ *
+ * Each module may only use this macro once. Calling it replaces
+ * module_init() and module_exit().
+ */
+#define drm_module_platform_driver(__platform_drv) \
+	module_driver(__platform_drv, drm_platform_driver_register, \
+		      platform_driver_unregister)
+
+#endif