[PATCH 8/8] watchdog: core: Update last_keepalive in watchdog_start()

[Guenter Roeck <linux@roeck-us.net>]

When starting a watchdog whose hardware is already running,
watchdog_start() calls __watchdog_ping() before marking the watchdog
active, and fails to update wd_data->last_keepalive. As a result:

1. wd_data->last_keepalive remains 0, causing watchdog_get_timeleft() to
   calculate the elapsed time since the epoch and return 0 time left.
2. watchdog_update_worker() inside __watchdog_ping() evaluates
   watchdog_need_worker() and watchdog_next_keepalive() while WDOG_ACTIVE
   is still clear, using wd_data->open_deadline instead of
   wd_data->last_keepalive. If open_deadline has expired, the background
   keepalive timer is not armed; conversely, if the active watchdog does
   not need a worker, the boot keepalive timer is not canceled.

Set WDOG_ACTIVE and wd_data->last_keepalive to started_at before calling
__watchdog_ping(), and revert WDOG_ACTIVE and update the worker if
__watchdog_ping() fails.

Fixes: fbbe35dfcf94 ("watchdog: use __watchdog_ping in startup")
Assisted-by: LLM
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
---
 drivers/watchdog/watchdog_dev.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/watchdog/watchdog_dev.c b/drivers/watchdog/watchdog_dev.c
index 24488c20539a..b9979f9957b8 100644
--- a/drivers/watchdog/watchdog_dev.c
+++ b/drivers/watchdog/watchdog_dev.c
@@ -264,10 +264,14 @@ static int watchdog_start(struct watchdog_device *wdd)
 
 	started_at = ktime_get();
 	if (watchdog_hw_running(wdd) && wdd->ops->ping) {
+		set_bit(WDOG_ACTIVE, &wdd->status);
+		wd_data->last_keepalive = started_at;
 		err = __watchdog_ping(wdd);
 		if (err == 0) {
-			set_bit(WDOG_ACTIVE, &wdd->status);
 			watchdog_hrtimer_pretimeout_start(wdd);
+		} else {
+			clear_bit(WDOG_ACTIVE, &wdd->status);
+			watchdog_update_worker(wdd);
 		}
 	} else {
 		err = wdd->ops->start(wdd);
-- 
2.45.2