[PATCH] [PATCH] PM: docs: Add comprehensive wakeup_count documentation

[chenheyun]

The current Documentation/power/wakeup-count.rst is empty and lacks
description of the race-free suspend mechanism, sysfs ABI semantics,
blocking behavior, and standard userspace usage.

Add complete documentation for /sys/power/wakeup_count, including
overview, interface semantics, usage example, and related interfaces.

Also update Documentation/power/index.rst to include the new document.

Signed-off-by: chenheyun <chen_heyun@xxxxxxx>
---
 Documentation/power/index.rst        |  1 +
 Documentation/power/wakeup-count.rst | 63 ++++++++++++++++++++++++++++
 2 files changed, 64 insertions(+)
 create mode 100644 Documentation/power/wakeup-count.rst

diff --git a/Documentation/power/index.rst b/Documentation/power/index.rst
index b4581e4ae785..901268049d7c 100644
--- a/Documentation/power/index.rst
+++ b/Documentation/power/index.rst
@@ -27,6 +27,7 @@ Power Management
     swsusp
     video
     tricks
+    wakeup-count
 
     userland-swsusp
 
diff --git a/Documentation/power/wakeup-count.rst b/Documentation/power/wakeup-count.rst
new file mode 100644
index 000000000000..5f3a1ca654ce
--- /dev/null
+++ b/Documentation/power/wakeup-count.rst
@@ -0,0 +1,63 @@
+.. SPDX-License-Identifier: GPL-2.0
+.. Copyright (C) 2025 The Linux Foundation
+
+The wakeup_count mechanism for race-free suspend
+================================================
+
+Overview
+--------
+
+The ``/sys/power/wakeup_count`` sysfs interface provides a stable userspace
+mechanism to perform race-free system suspend transitions. It eliminates the
+well-known race condition between suspend permission check and actual system
+suspend entry.
+
+Userspace may use it in a standard three-step sequence:
+
+1. Read the current global wakeup event counter. The read operation blocks
+   until all ongoing wakeup event processing is finished, returning a stable value.
+2. Perform necessary suspend preparation steps in userspace.
+3. Write the previously-read counter value back to the interface.
+   The write operation will only succeed if no new wakeup events have occurred
+   since the read.
+
+Only after a successful write may userspace safely trigger system suspend.
+
+Interface semantics
+-------------------
+
+``/sys/power/wakeup_count``
+
+**Read**
+    Returns the global monotonically-increasing wakeup event counter.
+    This call blocks until there are no wakeup events under active processing
+    inside the kernel. If interrupted by a signal, it returns -EINTR.
+
+**Write**
+    Accepts the counter value obtained from a prior read.
+    The write succeeds only if the kernel's current counter exactly matches
+    the written value. Mismatch indicates new wakeup events arrived during
+    userspace preparation, and suspend must be aborted.
+
+Standard userspace usage example
+--------------------------------
+
+.. code-block:: shell
+
+    count=$(cat /sys/power/wakeup_count)
+    do_suspend_preparation
+    echo "$count" > /sys/power/wakeup_count && echo mem > /sys/power/state
+
+Blocking behavior
+-----------------
+
+The blocking read ensures that userspace never observes an inconsistent state
+where wakeup events are still being handled within the kernel. This stability
+is the core guarantee of the interface.
+
+Related kernel interfaces
+-------------------------
+
+- ``/sys/power/state``: System suspend state control interface.
+- ``/sys/kernel/debug/wakeup_sources``: Per-device wakeup source statistics.
+- ``Documentation/power/wakeup-events.rst``: General wakeup event framework.
\ No newline at end of file
-- 
2.25.1