TCM ECC error events are only signaled for TCM reads, although this includes the read-modify-write sequence performed for some stores. Most errors detected by the ECC logic are signaled twice:

The TCM-centric events consist of two events per TCM port, one for fatal, that is, 2-bit ECC errors and one for correctable, that is, 1-bit ECC errors. These events are generated three clock cycles after the data read cycle. Consequently, these events are sometimes signaled on speculative TCM reads, such as instructions that are prefetched but never executed because of a branch earlier in the instruction sequence.

The processor-centric TCM events are only signaled for errors in data that would have otherwise been used by the processor. Errors on speculative reads never generate these errors. They consist of fatal and correctable events for: