I think this is the tldr; When operations use the (Intel) CPU cache they complete detectably faster than when using system RAM. When OpenSSL performs an encryption/decryption (five squaring operations followed by one multiplication) it divides up which cache banks it uses by multiplier number. If an attacker on the same system has filled the CPU cache banks they can detect which section of the cache bank has been written to and how long it takes. That information and knowledge of the OpenSSL cache use obfuscation scheme give the attacker the multiplier numbers used for encryption.