watchdog_get_timeleft() calls the driver's ->get_timeleft() regardless of whether the watchdog is running. It is reachable through the sysfs timeleft attribute, which is world readable with CONFIG_WATCHDOG_SYSFS=y, and through WDIOC_GETTIMELEFT, also after the watchdog was stopped with WDIOS_DISABLECARD while the device is still open. Drivers may not be able to access the hardware while the watchdog is stopped. omap_wdt only holds a runtime PM reference while the watchdog is running, so on a BeagleBone Black (AM335x) reading the counter of a stopped watchdog triggers an external abort. Any unprivileged user can hit it through sysfs: $ cat /sys/class/watchdog/watchdog0/timeleft Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35028 Internal error: : 1028 [#1] SMP ARM PC is at omap_wdt_get_timeleft+0x4/0x14 LR is at watchdog_get_timeleft+0x24/0x130 Call trace: omap_wdt_get_timeleft from watchdog_get_timeleft+0x24/0x130 watchdog_get_timeleft from timeleft_show+0x3c/0x80 timeleft_show from dev_attr_show+0x18/0x4c dev_attr_show from sysfs_kf_seq_show+0xa0/0x138 ... and the same happens with WDIOS_DISABLECARD followed by WDIOC_GETTIMELEFT on the open device, once the module has idled: Call trace: omap_wdt_get_timeleft from watchdog_get_timeleft+0x24/0x130 watchdog_get_timeleft from watchdog_ioctl+0x494/0x668 watchdog_ioctl from sys_ioctl+0x128/0xbcc sys_ioctl from ret_fast_syscall+0x0/0x54 If the ioctl comes before the module has idled, the read succeeds but reports 59 seconds left for a watchdog that is stopped. A stopped watchdog will not reboot the system, so report 0 time left when neither WDOG_ACTIVE nor WDOG_HW_RUNNING is set, without calling into the driver. This matches the software fallback for drivers without ->get_timeleft(), which already decays to 0 once the watchdog is no longer pinged. Tested on a BeagleBone Black: with the watchdog stopped, both the sysfs attribute and WDIOC_GETTIMELEFT after WDIOS_DISABLECARD return 0; with the watchdog started through /dev/watchdog0, timeleft counts down as before (59, then 56 three seconds later). Fixes: 452fafed839e ("watchdog: omap_wdt: implement get_timeleft") Cc: stable@vger.kernel.org Suggested-by: Guenter Roeck Signed-off-by: Nhat Tran --- v1->v2: - Moved the check from omap_wdt into watchdog_get_timeleft(), as suggested by Guenter, so it covers all drivers - Return 0 for a stopped watchdog; this also covers WDIOC_GETTIMELEFT after WDIOS_DISABLECARD on an open device v1: https://lore.kernel.org/all/20261120124009.24264-1-nhat.tranminhk2@gmail.com/ --- drivers/watchdog/watchdog_dev.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/watchdog/watchdog_dev.c b/drivers/watchdog/watchdog_dev.c index e8c1fcb53c56..b67a68316bda 100644 --- a/drivers/watchdog/watchdog_dev.c +++ b/drivers/watchdog/watchdog_dev.c @@ -435,6 +435,9 @@ static void watchdog_get_timeleft(struct watchdog_device *wdd, { *timeleft = 0; + if (!watchdog_active(wdd) && !watchdog_hw_running(wdd)) + return; + if (wdd->ops->get_timeleft) { *timeleft = wdd->ops->get_timeleft(wdd); } else { -- 2.25.1