watchdog: i6300esb: do not stop an already running watchdog

[PATCH] watchdog: i6300esb: do not stop an already running watchdog

Posted by Jan Kiszka 2 weeks, 2 days ago

From: Jan Kiszka <jan.kiszka@siemens.com>

esb_initdevice() unconditionally disabled the WDT, stopping a
watchdog that firmware had left running instead of taking over its
care. Detect the enable bit in the lock register, set
WDOG_HW_RUNNING like the other watchdog drivers do, and simply
reset the timeout to the configured value.

Assisted-by: opencode:Qwen3.8-27B
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---

Background to patch this: The i6300esb is the only watchdog model that 
QEMU provides for non-x86 virt machines. It can serve there to test 
complex system integrations. Patches to add a U-Boot driver were sent as 
well:
https://lore.kernel.org/u-boot/cover.1788526864.git.jan.kiszka@siemens.com/

Background for this unusual as assistance: While crafting this with own 
hands or using some cloud-hosted frontier model would have been much 
quicker, I used this practical case to try out what could be done in the 
background on my ordinary notebook already. Reviewing, testing (it 
worked immediately) and a bit shortening was done by me, analysis and 
patch fell out of the opencode after some half-day of thinking.

 drivers/watchdog/i6300esb.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)

diff --git a/drivers/watchdog/i6300esb.c b/drivers/watchdog/i6300esb.c
index a30835f547b3..13daafe05c43 100644
--- a/drivers/watchdog/i6300esb.c
+++ b/drivers/watchdog/i6300esb.c
@@ -264,13 +264,21 @@ static void esb_initdevice(struct esb_dev *edev)
 	 */
 	pci_write_config_word(edev->pdev, ESB_CONFIG_REG, 0x0003);
 
-	/* Check that the WDT isn't already locked */
+	/* Check the current state of the WDT */
 	pci_read_config_byte(edev->pdev, ESB_LOCK_REG, &val1);
 	if (val1 & ESB_WDT_LOCK)
 		dev_warn(&edev->pdev->dev, "nowayout already set\n");
 
-	/* Set the timer to watchdog mode and disable it for now */
-	pci_write_config_byte(edev->pdev, ESB_LOCK_REG, 0x00);
+	if (val1 & ESB_WDT_ENABLE) {
+		/*
+		 * The watchdog is already running, e.g. enabled by
+		 * firmware. Do not stop it, just mark it as running.
+		 */
+		set_bit(WDOG_HW_RUNNING, &edev->wdd.status);
+	} else {
+		/* Set the timer to watchdog mode and disable it for now */
+		pci_write_config_byte(edev->pdev, ESB_LOCK_REG, 0x00);
+	}
 
 	/* Check if the watchdog was previously triggered */
 	esb_unlock_registers(edev);
@@ -301,7 +309,10 @@ static int esb_probe(struct pci_dev *pdev,
 	if (!esb_getdevice(edev))
 		return -ENODEV;
 
-	/* Initialize the watchdog and make sure it does not run */
+	/*
+	 * Initialize the watchdog, keeping it running if it was already
+	 * started, e.g. by firmware.
+	 */
 	edev->wdd.info = &esb_info;
 	edev->wdd.ops = &esb_ops;
 	edev->wdd.min_timeout = ESB_HEARTBEAT_MIN;
-- 
2.47.3

Re: [PATCH] watchdog: i6300esb: do not stop an already running watchdog

Posted by Guenter Roeck 2 weeks, 1 day ago

On Wed, Sep 09, 2026 at 01:15:33PM +0200, Jan Kiszka wrote:
> From: Jan Kiszka <jan.kiszka@siemens.com>
> 
> esb_initdevice() unconditionally disabled the WDT, stopping a
> watchdog that firmware had left running instead of taking over its
> care. Detect the enable bit in the lock register, set
> WDOG_HW_RUNNING like the other watchdog drivers do, and simply
> reset the timeout to the configured value.
> 
> Assisted-by: opencode:Qwen3.8-27B
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Applied.

Thanks,
Guenter