On Fri, Aug 09, 2019 at 05:03:55PM -0700, Sean Christopherson wrote: > On Sat, Aug 10, 2019 at 02:40:32AM +0300, Jarkko Sakkinen wrote: > > On Wed, Aug 07, 2019 at 05:12:50PM -0700, Sean Christopherson wrote: > > > Move the EADD/EINIT checks on SGX_ENCL_CREATED to the very beginning of > > > the ioctl() flows. Deferring the check until the core code is fragile > > > as all code leading up to that point must be careful that it only uses > > > members of @encl that are initialized at allocation time. For example, > > > the flush_work() call in sgx_encl_init() will crash if the enclave has > > > not been created. > > > > > > Note, there is no need to take encl->lock to check SGX_ENCL_CREATED so > > > long as SGX_ENCL_CREATED is set only after the enclave is fully > > > initialized, it's not the kernel's responsibility to guard against > > > sgx_encl_create() racing with EADD/EINIT. Add a comment to highlight > > > the dependency. > > > > > > Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx> > > > > 07/11, 08/11 and 09/11 have been squashed and pushed. > > > > I'm now observing this kind of behavior with the self-test: > > > > jsakkine at jsakkine-lab2 in ~/devel/linux-sgx (master●) > > $ sudo tools/testing/selftests/x86/sgx/test_sgx > > Binary size 24576 (0x6000), SIGSTRUCT size 1808 > > Loading the enclave. > > ECREATE failed rc=-1, err=22. > > Doh, I forgot to update/test the selftest. I'll do so now. Oh, it's probably due to changing sgx_get_unmapped_area() to not align the address, i.e. the selftest is likely passing in an unaligned ELRANGE.
