Real-time systems are becoming increasingly complex. A modern car, for example, requires a multitude of control tasks, such as braking, active suspension, and collision avoidance. These tasks not only exhibit different degrees of safety criticality but also change their criticalities as the driving mode changes. For instance, the suspension task is a critical part of the stability of the car at high speed, but it is only a comfort feature at low speed. Therefore, it is crucial to ensure timing guarantees for the system with respect to the tasks’ criticalities, not only within each mode but also during mode changes. This paper presents a partitioned multi-processor scheduling scheme for multi-modal mixed-criticality real-time systems. Our scheme consists of a packing algorithm and a scheduling algorithm for each processor that take into account both mode changes and criticalities. The packing algorithm maximizes the schedulable utilization across modes using the sustained criticality of each task, which captures the overall criticality of the task across modes. The scheduling algorithm combines Rate-Monotonic scheduling with a mode transition enforcement mechanism that relies on the transitional zero-slack instants of tasks to control low-criticality tasks during mode changes, so as to preserve the schedulability of high-criticality tasks. We also present an implementation of our scheduler in the Linux operating system, as well as an experimental evaluation to illustrate its practicality. Our evaluation shows that our scheme can provide close to twice as much tolerance to overloads (ductility) compared to a mode-agnostic scheme.