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, 166 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-tty b/Documentation/ABI/testing/sysfs-tty
new file mode 100644
index 000000000..895c47f05
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-tty
@@ -0,0 +1,166 @@
+What:		/sys/class/tty/console/active
+Date:		Nov 2010
+Contact:	Kay Sievers <kay.sievers@vrfy.org>
+Description:
+		 Shows the list of currently configured
+		 console devices, like 'tty1 ttyS0'.
+		 The last entry in the file is the active
+		 device connected to /dev/console.
+		 The file supports poll() to detect virtual
+		 console switches.
+
+What:		/sys/class/tty/tty<x>/active
+Date:		Nov 2010
+Contact:	Kay Sievers <kay.sievers@vrfy.org>
+Description:
+		 Shows the currently active virtual console
+		 device, like 'tty1'.
+		 The file supports poll() to detect virtual
+		 console switches.
+
+What:		/sys/class/tty/ttyS<x>/uartclk
+Date:		Sep 2012
+Contact:	Tomas Hlavacek <tmshlvck@gmail.com>
+Description:
+		 Shows the current uartclk value associated with the
+		 UART port in serial_core, that is bound to TTY like ttyS0.
+		 uartclk = 16 * baud_base
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/type
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Shows the current tty type for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/line
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Shows the current tty line number for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/port
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Shows the current tty port I/O address for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/irq
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Shows the current primary interrupt for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/flags
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Show the tty port status flags for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/xmit_fifo_size
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Show the transmit FIFO size for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/close_delay
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		Show the closing delay time for this port in centiseconds.
+
+		These sysfs values expose the TIOCGSERIAL interface via
+		sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/closing_wait
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		Show the close wait time for this port in centiseconds.
+
+		Waiting forever is represented as 0. If waiting on close is
+		disabled then the value is 65535.
+
+		These sysfs values expose the TIOCGSERIAL interface via
+		sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/custom_divisor
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Show the custom divisor if any that is set on this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/io_type
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Show the I/O type that is to be used with the iomem base
+		 address.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/iomem_base
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 The I/O memory base for this port.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/iomem_reg_shift
+Date:		October 2012
+Contact:	Alan Cox <alan@linux.intel.com>
+Description:
+		 Show the register shift indicating the spacing to be used
+		 for accesses on this iomem address.
+
+		 These sysfs values expose the TIOCGSERIAL interface via
+		 sysfs rather than via ioctls.
+
+What:		/sys/class/tty/ttyS<x>/rx_trig_bytes
+Date:		May 2014
+Contact:	Yoshihiro YUNOMAE <yoshihiro.yunomae.ez@hitachi.com>
+Description:
+		 Shows current RX interrupt trigger bytes or sets the
+		 user specified value to change it for the FIFO buffer.
+		 Users can show or set this value regardless of opening the
+		 serial device file or not.
+
+		 The RX trigger can be set one of four kinds of values for UART
+		 serials. When users input a meaning less value to this I/F,
+		 the RX trigger is changed to the nearest lower value for the
+		 device specification. For example, when user sets 7bytes on
+		 16550A, which has 1/4/8/14 bytes trigger, the RX trigger is
+		 automatically changed to 4 bytes.
+
+What:		/sys/class/tty/ttyS<x>/console
+Date:		February 2020
+Contact:	Andy Shevchenko <andriy.shevchenko@linux.intel.com>
+Description:
+		 Allows user to detach or attach back the given device as
+		 kernel console. It shows and accepts a boolean variable.