FTBFS with linux-4.20 about linux-sgx-driver

[sergeay]

For which kernel are you compiling?

[mythi]

$ uname -r
4.20.3-685.native

[sergeay]

Yes, there is a change in kernel 4.20 that causes this build error and we are going to fix it.
Stay tuned or try an older kernel.

[mythi]

No worries, I'm not blocked because I can unpack the installer and

$ sed -i -e 's/vm_insert_pfn/vmf_insert_pfn/g' /opt/intel/sgxdriver/package/sgx_*.c

[mythi]

@sergeay there's also something else with 4.20: aesm_service fails to load AEs' with 4.20. The debug output is:

[enclave_initialize sgx_enclave_common.cpp:467]
SGX_IOC_ENCLAVE_INIT failed error = 4

the same aesm_service binary starts OK with an older kernel (that unfortunately has a different .config too). Any ideas where to look?

[sergeay]

This is unexpected...
Can you do the following for me:
$ sudo su
$ echo module isgx +p > /sys/kernel/debug/dynamic_debug/control
$ dmesg -wT | grep sgx

then run your scenario and drop the above dmesg output here.
Thanks

[mythi]

[Mon Jan 28 13:59:43 2019] calling  init_sgx_module+0x0/0x60 [isgx] @ 266
[Mon Jan 28 13:59:43 2019] intel_sgx: Intel SGX Driver v0.11
[Mon Jan 28 13:59:43 2019] intel_sgx INT0E0C:00: EPC bank 0x70200000-0x76000000
[Mon Jan 28 13:59:43 2019] intel_sgx: second initialization call skipped
[Mon Jan 28 13:59:43 2019] probe of intel_sgx.0 returned 1 after 17 usecs
[Mon Jan 28 13:59:43 2019] initcall init_sgx_module+0x0/0x60 [isgx] returned 0 a                                  fter 22278 usecs
[Mon Jan 28 14:00:45 2019] intel_sgx: [786:0x00000000432c5878] EINIT returned 4
[Mon Jan 28 14:01:00 2019] intel_sgx: [814:0x000000002e384b53] EINIT returned 4
[Mon Jan 28 14:01:22 2019] intel_sgx: [838:0x000000000313851d] EINIT returned 4
[Mon Jan 28 14:02:06 2019] intel_sgx: [853:0x00000000966f5758] EINIT returned 4

The last line repeats for all /opt/intel/sgxpsw/aesm/aesm_service attempts.

[sergeay]

Which version of psw are you using or do you build the open source package?

[mythi]

I'm pulling 'b2' branch from this repo, and psw (from linux-sgx) I've tried with master, 2.4, and 2.3.1.

[sergeay]

I can see that. It is reproducible with 2.1 as well.
Possibly some system call in the kernel has changed its behavior or exposed a bug.
We will check it.

[sergeay]

I suppose you are not blocked in the mean time and can use 4.19 or an older kernel.

[mythi]

@sergeay thanks for your help! I'm not blocked now that I know where the problem is (would prefer to stay with my distro's default kernel though, which is 4.20)

Does the AESM failure only happen with kernel 4.2? Can you reproduce the same issue (EINIT returned 4) with 4.19?

[mythi]

@ZongminGu I'm not able to check with 4.19 today. I'll get back to that question later in the week. It worked with 4.18 for sure.

[mythi]

The latest master helps to get aesm.service running. The latest SDK/PSW causes SIGILL with the SampleEnclave but that's probably something else.