On 7/5/2023 6:42 AM, Alex Bennée wrote:
zhanghao1 <zhangh...@kylinos.cn> writes:
Each vcpu creates a corresponding timer task. The watchdog
is driven by a timer according to a certain period. Each time
the timer expires, the counter is decremented. When the counter
is "0", the watchdog considers the vcpu to be stalling and resets
the VM. To avoid watchdog expiration, the guest kernel driver
needs to periodically send a pet event to update the counter.
Signed-off-by: zhanghao1 <zhangh...@kylinos.cn>
---
v2:
- change the function name and remove the redundant word 'stall'
- add trace-events to replace DPRINTF and qemu_log
- call 'watchdog_perform_action()' to reset vm
- update g_new0 to replace malloc
- update only use '.generic_name'
- update the bool variable 'is_initialized' to uint32_t
v1:
https://lore.kernel.org/qemu-devel/20230615061302.301754-1-zhangh...@kylinos.cn/
<snip>
+static void virtio_vcpu_watchdog_device_realize(DeviceState *dev, Error **errp)
+{
+ VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+ vwdt = VIRTIO_VCPU_WATCHDOG(dev);
+
+ virtio_init(vdev, VIRTIO_ID_WATCHDOG, 0);
This will never compile because VIRTIO_ID_WATCHDOG isn't defined
anywhere.
Next time you post you need to also include a link to the kernel side of
the driver and the virtio specification (or inflight patch for it).
Is this similar to vcpu stall based watchdog driver submitted and merged
in kernel.org (by Google) sometime back?
---Trilok Soni
