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, 97 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-mce b/Documentation/ABI/testing/sysfs-mce
new file mode 100644
index 000000000..83172f50e
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-mce
@@ -0,0 +1,97 @@
+What:		/sys/devices/system/machinecheck/machinecheckX/
+Contact:	Andi Kleen <ak@linux.intel.com>
+Date:		Feb, 2007
+Description:
+		(X = CPU number)
+
+		Machine checks report internal hardware error conditions
+		detected by the CPU. Uncorrected errors typically cause a
+		machine check (often with panic), corrected ones cause a
+		machine check log entry.
+
+		For more details about the x86 machine check architecture
+		see the Intel and AMD architecture manuals from their
+		developer websites.
+
+		For more details about the architecture
+		see http://one.firstfloor.org/~andi/mce.pdf
+
+		Each CPU has its own directory.
+
+What:		/sys/devices/system/machinecheck/machinecheckX/bank<Y>
+Contact:	Andi Kleen <ak@linux.intel.com>
+Date:		Feb, 2007
+Description:
+		(Y bank number)
+
+		64bit Hex bitmask enabling/disabling specific subevents for
+		bank Y.
+
+		When a bit in the bitmask is zero then the respective
+		subevent will not be reported.
+
+		By default all events are enabled.
+
+		Note that BIOS maintain another mask to disable specific events
+		per bank.  This is not visible here
+
+What:		/sys/devices/system/machinecheck/machinecheckX/check_interval
+Contact:	Andi Kleen <ak@linux.intel.com>
+Date:		Feb, 2007
+Description:
+		The entries appear for each CPU, but they are truly shared
+		between all CPUs.
+
+		How often to poll for corrected machine check errors, in
+		seconds (Note output is hexadecimal). Default 5 minutes.
+		When the poller finds MCEs it triggers an exponential speedup
+		(poll more often) on the polling interval.  When the poller
+		stops finding MCEs, it triggers an exponential backoff
+		(poll less often) on the polling interval. The check_interval
+		variable is both the initial and maximum polling interval.
+		0 means no polling for corrected machine check errors
+		(but some corrected errors might be still reported
+		in other ways)
+
+What:		/sys/devices/system/machinecheck/machinecheckX/trigger
+Contact:	Andi Kleen <ak@linux.intel.com>
+Date:		Feb, 2007
+Description:
+		The entries appear for each CPU, but they are truly shared
+		between all CPUs.
+
+		Program to run when a machine check event is detected.
+		This is an alternative to running mcelog regularly from cron
+		and allows to detect events faster.
+
+What:		/sys/devices/system/machinecheck/machinecheckX/monarch_timeout
+Contact:	Andi Kleen <ak@linux.intel.com>
+Date:		Feb, 2007
+Description:
+		How long to wait for the other CPUs to machine check too on a
+		exception. 0 to disable waiting for other CPUs.
+
+		Unit: us
+
+What:		/sys/devices/system/machinecheck/machinecheckX/ignore_ce
+Contact:	Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
+Date:		Jun 2009
+Description:
+		Disables polling and CMCI for corrected errors.
+		All corrected events are not cleared and kept in bank MSRs.
+
+What:		/sys/devices/system/machinecheck/machinecheckX/dont_log_ce
+Contact:	Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
+Date:		Jun 2009
+Description:
+		Disables logging for corrected errors.
+		All reported corrected errors will be cleared silently.
+
+		This option will be useful if you never care about corrected
+		errors.
+
+What:		/sys/devices/system/machinecheck/machinecheckX/cmci_disabled
+Contact:	Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
+Date:		Jun 2009
+Description:
+		Disables the CMCI feature.