Host tries to get ReturnValue after guest cancellation despite memory reset

[ludfjig]

It's possible that the host tries to read a returnvalue after cancelling the guest, even if the guest memory was reset. This surfaces as error=Error("Unable to pop data from buffer: Stack pointer is out of bounds. Stack pointer: 8, Buffer size: 16384") since there is nothing to pop since the memory is reset

[ludfjig]

The above is incorrect

During contention, when the main thread dispatches a guest function call, it sends a HypervisorHandlerAction::DispatchCallFromHost message to the hypervisor-handler thread and waits for a response. If the main thread times out while waiting, it calls terminate_hypervisor_handler_execution_and_reinitialise. To handle cases where the guest function completes execution just as the main thread gives up waiting, an atomic flag is checked to determine if the guest has finished executing.

The issue arises because, under contention, this atomic flag may never have been set to true in the first place—particularly if the hypervisor-handler thread was never scheduled. As a result, the main thread mistakenly assumes that the hypervisor thread has completed execution, since it interprets the atomic being false as an indication that it was initially true and later cleared upon completion. However, if the flag was never set to true, this assumption is incorrect. Consequently, the main thread attempts to read a ReturnValue from guest memory, even though the guest function never executed, leading to an error.

See code here. The info! call suggests that execution finished before we had time to cancel it. This is untrue. In fact, it's possible for execution to never had started.