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, 105 insertions, 0 deletions

diff --git a/Documentation/driver-api/serial/driver.rst b/Documentation/driver-api/serial/driver.rst
new file mode 100644
index 000000000..c1db6a1a6
--- /dev/null
+++ b/Documentation/driver-api/serial/driver.rst
@@ -0,0 +1,105 @@
+====================
+Low Level Serial API
+====================
+
+
+This document is meant as a brief overview of some aspects of the new serial
+driver.  It is not complete, any questions you have should be directed to
+<rmk@arm.linux.org.uk>
+
+The reference implementation is contained within amba-pl011.c.
+
+
+
+Low Level Serial Hardware Driver
+--------------------------------
+
+The low level serial hardware driver is responsible for supplying port
+information (defined by uart_port) and a set of control methods (defined
+by uart_ops) to the core serial driver.  The low level driver is also
+responsible for handling interrupts for the port, and providing any
+console support.
+
+
+Console Support
+---------------
+
+The serial core provides a few helper functions.  This includes
+decoding command line arguments (uart_parse_options()).
+
+There is also a helper function (uart_console_write()) which performs a
+character by character write, translating newlines to CRLF sequences.
+Driver writers are recommended to use this function rather than implementing
+their own version.
+
+
+Locking
+-------
+
+It is the responsibility of the low level hardware driver to perform the
+necessary locking using port->lock.  There are some exceptions (which
+are described in the struct uart_ops listing below.)
+
+There are two locks.  A per-port spinlock, and an overall semaphore.
+
+From the core driver perspective, the port->lock locks the following
+data::
+
+	port->mctrl
+	port->icount
+	port->state->xmit.head (circ_buf->head)
+	port->state->xmit.tail (circ_buf->tail)
+
+The low level driver is free to use this lock to provide any additional
+locking.
+
+The port_sem semaphore is used to protect against ports being added/
+removed or reconfigured at inappropriate times. Since v2.6.27, this
+semaphore has been the 'mutex' member of the tty_port struct, and
+commonly referred to as the port mutex.
+
+
+uart_ops
+--------
+
+.. kernel-doc:: include/linux/serial_core.h
+   :identifiers: uart_ops
+
+Other functions
+---------------
+
+.. kernel-doc:: drivers/tty/serial/serial_core.c
+   :identifiers: uart_update_timeout uart_get_baud_rate uart_get_divisor
+           uart_match_port uart_write_wakeup uart_register_driver
+           uart_unregister_driver uart_suspend_port uart_resume_port
+           uart_add_one_port uart_remove_one_port uart_console_write
+           uart_parse_earlycon uart_parse_options uart_set_options
+           uart_get_lsr_info uart_handle_dcd_change uart_handle_cts_change
+           uart_try_toggle_sysrq
+
+.. kernel-doc:: include/linux/serial_core.h
+   :identifiers: uart_port_tx_limited uart_port_tx
+
+Other notes
+-----------
+
+It is intended some day to drop the 'unused' entries from uart_port, and
+allow low level drivers to register their own individual uart_port's with
+the core.  This will allow drivers to use uart_port as a pointer to a
+structure containing both the uart_port entry with their own extensions,
+thus::
+
+	struct my_port {
+		struct uart_port	port;
+		int			my_stuff;
+	};
+
+Modem control lines via GPIO
+----------------------------
+
+Some helpers are provided in order to set/get modem control lines via GPIO.
+
+.. kernel-doc:: drivers/tty/serial/serial_mctrl_gpio.c
+   :identifiers: mctrl_gpio_init mctrl_gpio_to_gpiod
+           mctrl_gpio_set mctrl_gpio_get mctrl_gpio_enable_ms
+           mctrl_gpio_disable_ms_sync mctrl_gpio_disable_ms_no_sync