If dual-port memory is used as system memory, the RPU can even have its own
connection to the memory and create a recovery point concurrently to processing
without performance impact.
On the other hand, the processor should also have accessed to this unit to
perform tests and instruct the unit to generate a recovery point. Direct access for the
RPU to the processor could also be allowed in order to store the full processor state
as well. To simplify the implementation, the software can also store the current
processor state (registers, program counter, etc.) to the memory and instruct the
RPU to include this memory area in the RP generation.
The RPU has two operation modes, one that generates the recovery point and the
checksum and one that only generates the checksum. The second mode is required
for diagnostic purposes and serves well for accelerating the search of a correct RP
(see Sect. 9.2.1 for details).
Of course, the checksum should be generated during the actual recovery point
creation. If required, the checksums could also be stored distinct from the recovery
point to prevent a fault from affecting the checksum and the recovery point at the
same time.
Fig. 8.3 Hardware design with RPU
124
8 Recovery Preparation