This device includes an Error Profiling Controller (EPC) module. The main goal of this module is to enable the system to tolerate a certain amount of ECC correctable errors on the same address repeatedly in the memory system with minimal runtime overhead. Main features implemented in this device are described below.
- Capture the address of correctable ECC faults from different sources (for example, CPU, L2RAM, Interconnect) into a 16-entry Content Addressable Memory (CAM).
- For correctable faults, the error handling depends on the below conditions:
- if the incoming address is already in the 16-entry CAM, discard the fail. No error generated to ESM
- if the address is not in the CAM list, and the CAM has empty entries, add the address into the CAM list. In addition, raise the error signal to the ESM group 1 if enabled.
- if the address is not in the CAM list, and the CAM has no empty entries, always raise a signal to the ESM group 1.
- A 4-entry FIFO to store correctable error events and addresses for each IP interface.
- For uncorrectable faults of non-CPU access, capture the address and raise a signal to the ESM group 2.
- The CAM is implemented in memory mapped registers. The CPU can read and write to any entry for diagnostic test as if a real CAM memory macro.