Current computer systems offer main memory capacities that were not even imaginable decades ago.
However, memory management and abstractions are still mostly based on tiny 4-KiB fragments, allowing for fine-grained provisioning but at the expense of high management overhead for large objects.
With Morsels, we introduced a novel memory-management paradigm that shifts from the management of individual pages to larger virtual-memory objects, technically represented as subtrees of the page-table hierarchy.
This reduces management overhead and enables very fast transfer between address spaces.

Morsels are based on hardware-specific page-table formats, which limit them to a specific architecture, such as the AMD64 MMU format.
Although IO-page-table formats differ, we have previously demonstrated that it is possible to share Morsels with the IOMMU, enabling direct access from devices like NVMe SSDs, when both formats are similar.
However, this approach has several drawbacks. It restricts the use of advanced features (e.g., memory protection keys) and reduces the number of bits available for software use within each page table entry.
Furthermore, this method is infeasible if the formats are too different and important fields deviate.

To address these limitations, we propose an alternative approach, where multiple representations exist, one per MMU format.
One representation, likely the existing AMD64 implementation, will serve as the authoritative truth (primary), while others will be translation views (secondary) of the same memory object.
The authoritative truth ensures synchronization between threads and crash consistency using the existing log-free algorithms.
In contrast, translation views directly follow the object layout and therefore do not need to be persisted before system reboot.
Instead, they can be restored on first use and, for supported targets, populated lazily and partially on first access to a page within the Morsel.

As a first application for secondary views, we focus on virtual machines.
The guest physical address space of a virtual machine that runs on an AMD64 host is configured with page tables called Extended Page Tables (EPTs).
These page tables are incompatible with AMD64 MMU format.
Therefore, secondary views are required to map a morsel on both the host and the VM.
We show that morsels can be shared with virtual machines with only minor modifications to qemu and KVM.
Evaluation shows that it is not only feasible but also improves flexibility and the memory bandwidth compared to ivshmem, which is the current mechanism for sharing memory with a VM.