Multiple components of switchover time were analyzed to assess optimization opportunities and the update of the interrupt vector map entries was found to be one of the prime factors impacting switchover time. The number of vectors to be updated varies from a few to the entire table (192 vectors). The update for a single entry can take about 5 cycles and hence the update of the vector table itself can take up to 960 cycles (4.8 us at 200MHz).

In order to reduce the switchover time, a shadow vector memory and capability to swap it with active vector memory is implemented. The switchover code can update the shadow vector memory when the application execution is in progress. Once the vector memory is updated, the swap be completed in one clock cycle. Both memories can be used in ping-pong fashion for successive software upgrades.

A representative implementation of Interrupt vector swap is shown in Figure 4-2. Figure 4-2(a) is the configuration before swap and Figure 4-2(b) after the swap. The overall vector memory is divided into two blocks – Block A that spans from address 0x0000_0D00 to 0x0000_0EFF and Block B that spans from address 0x0100_0900 to 0x0100_0AFF. Block A holds the active vector table and Block B holds the shadow vector table. During LFU, shadow memory entries get updated before switchover, and the swap is executed during switchover. This reduces the switchover time from a maximum of 960 cycles to a single cycle.