We have a product that uses OCF to encrypt disk blocks. Since applying the FreeBSD-SA-18:07.lazyfpu patch we are seeing occasional corruption. The amount of corrupted data is always less than the request size and is always a multiple of the AES block size. Retrying the operation always succeeds leading to the conclusion that this seems to be a locking issue. The only thing so far that we can come up with is that this seems to result from either the OCF layer itself or the AES-NI driver below it. Looking for suggestions about where to proceed next.