Maintenance of virtualized datacenters is often needed for the purposes of introducing new features, fixing bugs or mitigating security problems. However, current maintenance methods are either highly disruptive to the operation of VMs, utilize large amounts of computing resources or require high development efforts. We build HyperTP, a generic framework which combines in a unified way two approaches: in-place server micro-reboot-based hypervisor transplant (noted InPlaceTP) and live VM migration-based hypervisor transplant (noted Migra-tionTP). HyperTP hinges on a VM state hierarchy for organizing different types of hypervisor memory states in terms of their relation to VM execution, and an Unified Intermediate State Representation that abstracts VM-relevant memory states between multiple different hypervisors. We describe our implementations of both approaches, including technical details of our UISR design and the transplant process. Our evaluation results show that HyperTP delivers satisfactory performance: (1) MigrationTP changes a VM's underlying hypervisor while taking the same time and impacting virtual machines (VMs) with the same performance degradation as normal live migration; and (2) InPlaceTP imposes minimal VM downtime, even under increasing number of VMs and memory sizes. Finally, we discuss how the combination of InPlaceTP and MigrationTP can be used to address the challenges of upgrading a hypervisor cluster, and to mitigate known unpatched hypervisor vulnerabilities.