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, 178 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-mic b/Documentation/ABI/testing/sysfs-class-mic
new file mode 100644
index 000000000..5e5f36d10
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-mic
@@ -0,0 +1,178 @@
+What:		/sys/class/mic/
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		The mic class directory belongs to Intel MIC devices and
+		provides information per MIC device. An Intel MIC device is a
+		PCIe form factor add-in Coprocessor card based on the Intel Many
+		Integrated Core (MIC) architecture that runs a Linux OS.
+
+What:		/sys/class/mic/mic<X>
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		The directories /sys/class/mic/mic0, /sys/class/mic/mic1 etc.,
+		represent MIC devices (0,1,..etc). Each directory has
+		information specific to that MIC device.
+
+What:		/sys/class/mic/mic<X>/family
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		Provides information about the Coprocessor family for an Intel
+		MIC device. For example - "x100"
+
+What:		/sys/class/mic/mic<X>/stepping
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		Provides information about the silicon stepping for an Intel
+		MIC device. For example - "A0" or "B0"
+
+What:		/sys/class/mic/mic<X>/state
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		When read, this entry provides the current state of an Intel
+		MIC device in the context of the card OS. Possible values that
+		will be read are:
+
+
+		===============  ===============================================
+		"ready"		 The MIC device is ready to boot the card OS.
+				 On reading this entry after an OSPM resume,
+				 a "boot" has to be written to this entry if
+				 the card was previously shutdown during OSPM
+				 suspend.
+		"booting"	 The MIC device has initiated booting a card OS.
+		"online"	 The MIC device has completed boot and is online
+		"shutting_down"	 The card OS is shutting down.
+		"resetting"	 A reset has been initiated for the MIC device
+		"reset_failed"	 The MIC device has failed to reset.
+		===============  ===============================================
+
+		When written, this sysfs entry triggers different state change
+		operations depending upon the current state of the card OS.
+		Acceptable values are:
+
+
+		==========  ===================================================
+		"boot"      Boot the card OS image specified by the combination
+			    of firmware, ramdisk, cmdline and bootmode
+			    sysfs entries.
+		"reset"     Initiates device reset.
+		"shutdown"  Initiates card OS shutdown.
+		==========  ===================================================
+
+What:		/sys/class/mic/mic<X>/shutdown_status
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		An Intel MIC device runs a Linux OS during its operation. This
+		OS can shutdown because of various reasons. When read, this
+		entry provides the status on why the card OS was shutdown.
+		Possible values are:
+
+		==========  ===================================================
+		"nop"       shutdown status is not applicable, when the card OS
+			    is "online"
+		"crashed"   Shutdown because of a HW or SW crash.
+		"halted"    Shutdown because of a halt command.
+		"poweroff"  Shutdown because of a poweroff command.
+		"restart"   Shutdown because of a restart command.
+		==========  ===================================================
+
+What:		/sys/class/mic/mic<X>/cmdline
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		An Intel MIC device runs a Linux OS during its operation. Before
+		booting this card OS, it is possible to pass kernel command line
+		options to configure various features in it, similar to
+		self-bootable machines. When read, this entry provides
+		information about the current kernel command line options set to
+		boot the card OS. This entry can be written to change the
+		existing kernel command line options. Typically, the user would
+		want to read the current command line options, append new ones
+		or modify existing ones and then write the whole kernel command
+		line back to this entry.
+
+What:		/sys/class/mic/mic<X>/firmware
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		When read, this sysfs entry provides the path name under
+		/lib/firmware/ where the firmware image to be booted on the
+		card can be found. The entry can be written to change the
+		firmware image location under /lib/firmware/.
+
+What:		/sys/class/mic/mic<X>/ramdisk
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		When read, this sysfs entry provides the path name under
+		/lib/firmware/ where the ramdisk image to be used during card
+		OS boot can be found. The entry can be written to change
+		the ramdisk image location under /lib/firmware/.
+
+What:		/sys/class/mic/mic<X>/bootmode
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		When read, this sysfs entry provides the current bootmode for
+		the card. This sysfs entry can be written with the following
+		valid strings:
+		a) linux - Boot a Linux image.
+		b) flash - Boot an image for flash updates.
+
+What:		/sys/class/mic/mic<X>/log_buf_addr
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		An Intel MIC device runs a Linux OS during its operation. For
+		debugging purpose and early kernel boot messages, the user can
+		access the card OS log buffer via debugfs. When read, this entry
+		provides the kernel virtual address of the buffer where the card
+		OS log buffer can be read. This entry is written by the host
+		configuration daemon to set the log buffer address. The correct
+		log buffer address to be written can be found in the System.map
+		file of the card OS.
+
+What:		/sys/class/mic/mic<X>/log_buf_len
+Date:		October 2013
+KernelVersion:	3.13
+Contact:	Sudeep Dutt <sudeep.dutt@intel.com>
+Description:
+		An Intel MIC device runs a Linux OS during its operation. For
+		debugging purpose and early kernel boot messages, the user can
+		access the card OS log buffer via debugfs. When read, this entry
+		provides the kernel virtual address where the card OS log buffer
+		length can be read. This entry is written by host configuration
+		daemon to set the log buffer length address. The correct log
+		buffer length address to be written can be found in the
+		System.map file of the card OS.
+
+What:		/sys/class/mic/mic<X>/heartbeat_enable
+Date:		March 2015
+KernelVersion:	4.4
+Contact:	Ashutosh Dixit <ashutosh.dixit@intel.com>
+Description:
+		The MIC drivers detect and inform user space about card crashes
+		via a heartbeat mechanism (see the description of
+		shutdown_status above). User space can turn off this
+		notification by setting heartbeat_enable to 0 and enable it by
+		setting this entry to 1. If this notification is disabled it is
+		the responsibility of user space to detect card crashes via
+		alternative means such as a network ping. This setting is
+		enabled by default.