On 5/13/20 1:21 AM, Erik Skultety wrote:
2) check if /dev/sev device exist (aka firmware is detected) This seems reasonable. Shouldn't it have been documented in docs/kbase/launch_security_sev.rst? Sure, we can add a mention about this. Although, doesn't 1 imply 2? IOW can you have the kernel module parameter set to 1 and yet kernel doesn't expose the /dev/sev node?
Currently, 1 does not imply 2, KVM driver does not initialize the firmware during the feature probe (i.e does not access the /dev/sev). The firmware initialization is delayed until the first guest launch. So only sane way to know whether firmware is been detected is check the existence of the /dev/sev or issue a query-sev command . The query-sev command will send the platform_status request to the firmware, if the firmware is not ready then this command will fail. I see. Can query-sev fail or return that it's disabled, aka {"enabled": false,...} in the SevInfo QMP response, but at the same time succeed in returning the platform capabilities via query-sev-capabilities? I'm asking, because libvirt only issues the latter to fill in the QEMU capabilities structure.
Just looked at qemu code, If /dev/sev does not exist then query-sev-capabilities should fail, and if SEV is not enabled in the guest then query-sev should returns false. So, basically what libvirt is doing correct, it should be using query-sev-capabilities to populate QEMU capabilities. It was my bad, I should have mentioned the query-sev-capabilities and not the query-sev check. thanks