On Thu, Feb 14, 2019 at 04:24:16PM +0100, Paolo Bonzini wrote: > On 14/02/19 14:30, Paolo Bonzini wrote: > > On 13/02/19 00:34, Sean Christopherson wrote: > >> Sean Christopherson (15): > >> KVM: Add a "-t" option to run a specific test > >> KVM: x86/config: Consolidate EPT access tests into a single test > >> KVM: nVMX: Exclude the EPT access tests from the VMX testcase > >> KVM: nVMX: Drop the bogus 2gb requirement from the "vmx" testcase > >> KVM: nVMX: Drop the bogus 2gb requirement from the shadow VMCS test > >> KVM: nVMX: Drop the bogus SMP requirement from VMCS shadowing test > >> KVM: nVMX: Rename VMCS shadowing test to "vmcs_shadow" > >> KVM: nVMX: Exclude the VMCS shadowing test from the "vmx" testcase > >> KVM: nVMX: Drop testcases that are redundant with the primary VMX > >> testcase > >> KVM: nVMX: Drop the SMP configuration of the ST APIC passthrough test > >> KVM: nVMX: Consolidate the SMP tests into a single testcase > >> KVM: nVMX: Rename SMP tests to being with vmx_smp_* > >> KVM: nVMX: Exclude the SMP tests from the primary "vmx" testcases > >> KVM: nVMX: Drop the bogus 2gb requirement from the SMP tests > >> KVM: nVMX: Drop the bogus 2gb requirement from the HLT+RVI testcase > >> > >> run_tests.sh | 6 +- > >> scripts/runtime.bash | 6 +- > >> x86/unittests.cfg | 397 +------------------------------------------ > >> x86/vmx_tests.c | 8 +- > >> 4 files changed, 20 insertions(+), 397 deletions(-) > >> > > > > > > Nice, thanks. "-t" is taken by TAP output so I dropped that patch, and > > I also dropped patch 7 because it's nice to have "vmx_" in the name of > > the tests. Otherwise, it's always good to clean up after our past selves... > > Hmm, spoke too soon. The two SMP tests hang when grouped in a single > execution (at least on a Haswell Xeon E5), I can reproduce it also with > upstream kvm-unit-tests and > > ./x86/run x86/vmx.flat -cpu host,+vmx -smp 2 \ > -append \ > "vmx_eoi_bitmap_ioapic_scan_test vmx_apic_passthrough_thread_test" None of my systems have full vAPIC support, i.e. I can't run vmx_eoi_bitmap_ioapic_scan_test. My guess is it doesn't clean up after itself. On the topic of self-cleanup, a lofty long-term goal would be to enhance (or rewrite) the framework to properly sandbox each test. The unit tests are riddled with hidden dependencies.
