https://www.theregister.co.uk/2018/06/22/intel_tlbleed_key_data_leak/
"If software is written to be perfectly side-channel resistant – control flow and data flow are constant and do not depend on the key being processed – [TLBleed] will not work, but these implementations are rare. Also, the fact that the TLB uses data accesses and not the code path is unusual enough that side-channel-resistant implementations have been found to be vulnerable to TLBleed, eg: RSA in libgcrypt."