Visible to Intel only — GUID: sfo1410068355201
Ixiasoft
Visible to Intel only — GUID: sfo1410068355201
Ixiasoft
10.4.1.2. Single Event Upset Protection
The L2 cache has the option of using ECCs to protect against Single Event Upset (SEU) errors in the cache RAM.
The L2 cache has two types of protection:
- The L2 cache tag RAMs are protected by parity.
- The L2 cache data RAMs are protected using single error correction, double error detection (SECDED) ECC Hamming code.
Enabling ECCs does not affect the performance of the L2 cache. The ECC bits are calculated only for writes to the data RAM that are 64 bits wide (8 bytes, or one‑quarter of the cache line length). The ECC logic does not perform a read‑modify‑write when calculating the ECC bits.
The ECC protection bits are not valid in the following cases:
- Data is written that is not 64-bit aligned in memory
- Data is written that is less than 64 bits in width
In these cases the Byte Write Error interrupt is asserted. Cache data is still written when such an error occurs. However, the ECC error detection and correction continues to function. Therefore, the cache data is likely to be incorrect on subsequent reads.
To use ECCs, the software and system must meet the following requirements:
- L1 and L2 cache must be configured as write‑back and write-allocate for any cacheable memory region
- Level 3 interconnect masters using the ACP must only perform the following types of data writes:
- 64-bit aligned in memory
- 64-bit wide accesses
If a correctable ECC error occurs, the ECC corrects the read data in parallel to asserting a correctable error signal on the AXI bus. If an uncorrectable error occurs in the L2 cache, the uncorrected data remains in the L2 cache and an AXI slave error (SLVERR) is sent to the L1 memory system. In addition, interrupts can be enabled for correctable and uncorrectable ECC errors through the GIC.