Thanks for pointing that out. They do indeed originate from OneNAND. I should really have looked at the attached dmesg output before posting. My post came from previous experience having seen main memory ECC errors in hundreds of servers (before flash memory was commonplace) and it never occurred to me that ECC would now be available in flash memory devices.

The same ECC principles should still apply though regardless of whether the memory is volatile or non-volatile.

Interestingly enough, OneNAND is actually known as "fusion" memory which not only consists of flash memory but also includes a 5KB SRAM buffer (as well as controller logic and hardware ECC) on the same chip so it's possible (but not likely) that the errors come from the SRAM buffer.