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, 95 insertions, 0 deletions

diff --git a/Documentation/networking/netif-msg.rst b/Documentation/networking/netif-msg.rst
new file mode 100644
index 000000000..b20d265a7
--- /dev/null
+++ b/Documentation/networking/netif-msg.rst
@@ -0,0 +1,95 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+===============
+NETIF Msg Level
+===============
+
+The design of the network interface message level setting.
+
+History
+-------
+
+ The design of the debugging message interface was guided and
+ constrained by backwards compatibility previous practice.  It is useful
+ to understand the history and evolution in order to understand current
+ practice and relate it to older driver source code.
+
+ From the beginning of Linux, each network device driver has had a local
+ integer variable that controls the debug message level.  The message
+ level ranged from 0 to 7, and monotonically increased in verbosity.
+
+ The message level was not precisely defined past level 3, but were
+ always implemented within +-1 of the specified level.  Drivers tended
+ to shed the more verbose level messages as they matured.
+
+   - 0  Minimal messages, only essential information on fatal errors.
+   - 1  Standard messages, initialization status.  No run-time messages
+   - 2  Special media selection messages, generally timer-driver.
+   - 3  Interface starts and stops, including normal status messages
+   - 4  Tx and Rx frame error messages, and abnormal driver operation
+   - 5  Tx packet queue information, interrupt events.
+   - 6  Status on each completed Tx packet and received Rx packets
+   - 7  Initial contents of Tx and Rx packets
+
+ Initially this message level variable was uniquely named in each driver
+ e.g. "lance_debug", so that a kernel symbolic debugger could locate and
+ modify the setting.  When kernel modules became common, the variables
+ were consistently renamed to "debug" and allowed to be set as a module
+ parameter.
+
+ This approach worked well.  However there is always a demand for
+ additional features.  Over the years the following emerged as
+ reasonable and easily implemented enhancements
+
+   - Using an ioctl() call to modify the level.
+   - Per-interface rather than per-driver message level setting.
+   - More selective control over the type of messages emitted.
+
+ The netif_msg recommendation adds these features with only a minor
+ complexity and code size increase.
+
+ The recommendation is the following points
+
+  - Retaining the per-driver integer variable "debug" as a module
+    parameter with a default level of '1'.
+
+  - Adding a per-interface private variable named "msg_enable".  The
+    variable is a bit map rather than a level, and is initialized as::
+
+       1 << debug
+
+    Or more precisely::
+
+	debug < 0 ? 0 : 1 << min(sizeof(int)-1, debug)
+
+    Messages should changes from::
+
+      if (debug > 1)
+	   printk(MSG_DEBUG "%s: ...
+
+    to::
+
+      if (np->msg_enable & NETIF_MSG_LINK)
+	   printk(MSG_DEBUG "%s: ...
+
+
+The set of message levels is named
+
+
+  =========   ===================	============
+  Old level   Name			Bit position
+  =========   ===================	============
+    0         NETIF_MSG_DRV		0x0001
+    1         NETIF_MSG_PROBE		0x0002
+    2         NETIF_MSG_LINK		0x0004
+    2         NETIF_MSG_TIMER		0x0004
+    3         NETIF_MSG_IFDOWN		0x0008
+    3         NETIF_MSG_IFUP		0x0008
+    4         NETIF_MSG_RX_ERR		0x0010
+    4         NETIF_MSG_TX_ERR		0x0010
+    5         NETIF_MSG_TX_QUEUED	0x0020
+    5         NETIF_MSG_INTR		0x0020
+    6         NETIF_MSG_TX_DONE		0x0040
+    6         NETIF_MSG_RX_STATUS	0x0040
+    7         NETIF_MSG_PKTDATA		0x0080
+  =========   ===================	============