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, 102 insertions, 0 deletions

diff --git a/Documentation/driver-api/switchtec.rst b/Documentation/driver-api/switchtec.rst
new file mode 100644
index 000000000..7611fdc53
--- /dev/null
+++ b/Documentation/driver-api/switchtec.rst
@@ -0,0 +1,102 @@
+========================
+Linux Switchtec Support
+========================
+
+Microsemi's "Switchtec" line of PCI switch devices is already
+supported by the kernel with standard PCI switch drivers. However, the
+Switchtec device advertises a special management endpoint which
+enables some additional functionality. This includes:
+
+* Packet and Byte Counters
+* Firmware Upgrades
+* Event and Error logs
+* Querying port link status
+* Custom user firmware commands
+
+The switchtec kernel module implements this functionality.
+
+
+Interface
+=========
+
+The primary means of communicating with the Switchtec management firmware is
+through the Memory-mapped Remote Procedure Call (MRPC) interface.
+Commands are submitted to the interface with a 4-byte command
+identifier and up to 1KB of command specific data. The firmware will
+respond with a 4-byte return code and up to 1KB of command-specific
+data. The interface only processes a single command at a time.
+
+
+Userspace Interface
+===================
+
+The MRPC interface will be exposed to userspace through a simple char
+device: /dev/switchtec#, one for each management endpoint in the system.
+
+The char device has the following semantics:
+
+* A write must consist of at least 4 bytes and no more than 1028 bytes.
+  The first 4 bytes will be interpreted as the Command ID and the
+  remainder will be used as the input data. A write will send the
+  command to the firmware to begin processing.
+
+* Each write must be followed by exactly one read. Any double write will
+  produce an error and any read that doesn't follow a write will
+  produce an error.
+
+* A read will block until the firmware completes the command and return
+  the 4-byte Command Return Value plus up to 1024 bytes of output
+  data. (The length will be specified by the size parameter of the read
+  call -- reading less than 4 bytes will produce an error.)
+
+* The poll call will also be supported for userspace applications that
+  need to do other things while waiting for the command to complete.
+
+The following IOCTLs are also supported by the device:
+
+* SWITCHTEC_IOCTL_FLASH_INFO - Retrieve firmware length and number
+  of partitions in the device.
+
+* SWITCHTEC_IOCTL_FLASH_PART_INFO - Retrieve address and lengeth for
+  any specified partition in flash.
+
+* SWITCHTEC_IOCTL_EVENT_SUMMARY - Read a structure of bitmaps
+  indicating all uncleared events.
+
+* SWITCHTEC_IOCTL_EVENT_CTL - Get the current count, clear and set flags
+  for any event. This ioctl takes in a switchtec_ioctl_event_ctl struct
+  with the event_id, index and flags set (index being the partition or PFF
+  number for non-global events). It returns whether the event has
+  occurred, the number of times and any event specific data. The flags
+  can be used to clear the count or enable and disable actions to
+  happen when the event occurs.
+  By using the SWITCHTEC_IOCTL_EVENT_FLAG_EN_POLL flag,
+  you can set an event to trigger a poll command to return with
+  POLLPRI. In this way, userspace can wait for events to occur.
+
+* SWITCHTEC_IOCTL_PFF_TO_PORT and SWITCHTEC_IOCTL_PORT_TO_PFF convert
+  between PCI Function Framework number (used by the event system)
+  and Switchtec Logic Port ID and Partition number (which is more
+  user friendly).
+
+
+Non-Transparent Bridge (NTB) Driver
+===================================
+
+An NTB hardware driver is provided for the Switchtec hardware in
+ntb_hw_switchtec. Currently, it only supports switches configured with
+exactly 2 NT partitions and zero or more non-NT partitions. It also requires
+the following configuration settings:
+
+* Both NT partitions must be able to access each other's GAS spaces.
+  Thus, the bits in the GAS Access Vector under Management Settings
+  must be set to support this.
+* Kernel configuration MUST include support for NTB (CONFIG_NTB needs
+  to be set)
+
+NT EP BAR 2 will be dynamically configured as a Direct Window, and
+the configuration file does not need to configure it explicitly.
+
+Please refer to Documentation/driver-api/ntb.rst in Linux source tree for an overall
+understanding of the Linux NTB stack. ntb_hw_switchtec works as an NTB
+Hardware Driver in this stack.