What you should know about the diag288 watchdog device driver

6.10 LPAR mode z/VM guest KVM guest

The watchdog function comprises two components: a watchdog application on the Linux® instance being controlled and a watchdog timer outside the Linux instance.

For Linux in LPAR mode, the timer runs in the IBM Z® firmware. For Linux on z/VM® the timer is provided by z/VM CP. For Linux on KVM, the timer runs on the KVM host.

While the Linux instance operates satisfactorily, the watchdog application reports a positive status to the watchdog timer at regular intervals. The watchdog application uses a device node to pass these status reports to the timer (Figure 1).

Figure 1. Watchdog application and timer for Linux in LPAR mode
This graphic shows a watchdog application in user space using the /dev/watchdog device node to connect to the watchdog timer.

The watchdog application typically derives its status by monitoring critical network connections, file systems, and processes on the Linux instance. If a specified time elapses without a positive report being received by the watchdog timer, the watchdog timer assumes that the Linux instance is in an error state. The watchdog timer then triggers a predefined action against the Linux instance. For example, Linux might be shut down or rebooted, or a system dump might be initiated, see Setting the timeout action.

For information about setting the default timer and performing other actions, see the external programming interfaces of the device driver.

Linux on z/VM only: Loading or saving a DCSS can take a long time during which the virtual machine does not respond, depending on the size of the DCSS. As a result, a watchdog might time out and restart the guest. You are advised not to use the watchdog in combination with loading or saving DCSSs.

See also the generic watchdog documentation in your Linux kernel source tree under Documentation/watchdog.