On Tue, Nov 21, 2017 at 12:46:23AM +0200, Jarkko Sakkinen wrote: > On Wed, Nov 15, 2017 at 12:54:12PM +0100, Peter Zijlstra wrote: > > On Tue, Nov 14, 2017 at 10:49:48PM +0200, Jarkko Sakkinen wrote: > > > In these cases IA32_FEATURE_CONTROL[17] would be zeroed before locking > > > the feature control, which would mean that the kernel could not write > > > new values with wrmsr for the root key hash. > > > > > The question is whether we want to allow this or not. If the answer is > > > no, a check can be added to the driver initialization code whether 17 is > > > set, and if not, it driver would fail to initialize. > > > > That has my vote; I would not trust a firmware/BIOS key. > > Please note that it does not have a key to look inside the enclave. The > enclave is protected by two means: > > 1. The CPU asserts the memory accesses to it. > 2. The CPU encrypts/decrypts in L1 in order to protect from physical > attacks and peripherals that have potential spy the bus. The encryption key is generated for every boot cycle and it is *never* leaked out of the CPU package. After thinking what you and Borislav said I position myself to the point of view that even if the MSRs would be read-only the kernel could allow running enclaves *with a condition*. I propose adding an additional check to the driver initialization: Try to start LE. If it doesn't start i.e. is signed with a different root key than the one inside MSRs, then fail the initialization. /Jarkko
