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, 121 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-bus-cdx b/Documentation/ABI/testing/sysfs-bus-cdx
new file mode 100644
index 000000000..e84277531
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-bus-cdx
@@ -0,0 +1,121 @@
+What:		/sys/bus/cdx/rescan
+Date:		March 2023
+Contact:	nipun.gupta@amd.com
+Description:
+		Writing y/1/on to this file will cause rescan of the bus
+		and devices on the CDX bus. Any new devices are scanned and
+		added to the list of Linux devices and any devices removed are
+		also deleted from Linux.
+
+		For example::
+
+		  # echo 1 > /sys/bus/cdx/rescan
+
+What:		/sys/bus/cdx/devices/.../vendor
+Date:		March 2023
+Contact:	nipun.gupta@amd.com
+Description:
+		Vendor ID for this CDX device, in hexadecimal. Vendor ID is
+		16 bit identifier which is specific to the device manufacturer.
+		Combination of Vendor ID and Device ID identifies a device.
+
+What:		/sys/bus/cdx/devices/.../device
+Date:		March 2023
+Contact:	nipun.gupta@amd.com
+Description:
+		Device ID for this CDX device, in hexadecimal. Device ID is
+		16 bit identifier to identify a device type within the range
+		of a device manufacturer.
+		Combination of Vendor ID and Device ID identifies a device.
+
+What:		/sys/bus/cdx/devices/.../subsystem_vendor
+Date:		July 2023
+Contact:	puneet.gupta@amd.com
+Description:
+		Subsystem Vendor ID for this CDX device, in hexadecimal.
+		Subsystem Vendor ID is 16 bit identifier specific to the
+		card manufacturer.
+
+What:		/sys/bus/cdx/devices/.../subsystem_device
+Date:		July 2023
+Contact:	puneet.gupta@amd.com
+Description:
+		Subsystem Device ID for this CDX device, in hexadecimal
+		Subsystem Device ID is 16 bit identifier specific to the
+		card manufacturer.
+
+What:		/sys/bus/cdx/devices/.../class
+Date:		July 2023
+Contact:	puneet.gupta@amd.com
+Description:
+		This file contains the class of the CDX device, in hexadecimal.
+		Class is 24 bit identifier specifies the functionality of the device.
+
+What:		/sys/bus/cdx/devices/.../revision
+Date:		July 2023
+Contact:	puneet.gupta@amd.com
+Description:
+		This file contains the revision field of the CDX device, in hexadecimal.
+		Revision is 8 bit revision identifier of the device.
+
+What:		/sys/bus/cdx/devices/.../enable
+Date:		October 2023
+Contact:	abhijit.gangurde@amd.com
+Description:
+		CDX bus should be disabled before updating the devices in FPGA.
+		Writing n/0/off will attempt to disable the CDX bus and.
+		writing y/1/on will attempt to enable the CDX bus. Reading this file
+		gives the current state of the bus, 1 for enabled and 0 for disabled.
+
+		For example::
+
+		  # echo 1 > /sys/bus/cdx/.../enable
+
+What:		/sys/bus/cdx/devices/.../reset
+Date:		March 2023
+Contact:	nipun.gupta@amd.com
+Description:
+		Writing y/1/on to this file resets the CDX device or all devices
+		on the bus. On resetting the device, the corresponding driver is
+		notified twice, once before the device is being reset, and again
+		after the reset has been complete.
+
+		For example::
+
+		  # echo 1 > /sys/bus/cdx/.../reset
+
+What:		/sys/bus/cdx/devices/.../remove
+Date:		March 2023
+Contact:	tarak.reddy@amd.com
+Description:
+		Writing y/1/on to this file removes the corresponding
+		device from the CDX bus. If the device is to be reconfigured
+		reconfigured in the Hardware, the device can be removed, so
+		that the device driver does not access the device while it is
+		being reconfigured.
+
+		For example::
+
+		  # echo 1 > /sys/bus/cdx/devices/.../remove
+
+What:		/sys/bus/cdx/devices/.../resource<N>
+Date:		July 2023
+Contact:	puneet.gupta@amd.com
+Description:
+		The resource binary file contains the content of the memory
+		regions. These files can be m'maped from userspace.
+
+What:		/sys/bus/cdx/devices/.../modalias
+Date:		July 2023
+Contact:	nipun.gupta@amd.com
+Description:
+		This attribute indicates the CDX ID of the device.
+		That is in the format:
+		cdx:vXXXXdXXXXsvXXXXsdXXXXcXXXXXX,
+		where:
+
+		    - vXXXX contains the vendor ID;
+		    - dXXXX contains the device ID;
+		    - svXXXX contains the subsystem vendor ID;
+		    - sdXXXX contains the subsystem device ID;
+		    - cXXXXXX contains the device class.