On 21/11/2017 14:54, Liran Alon wrote:
>
> Hi all,
>
> It seems that when running latest kvm-unit-tests repo, all the following
> unit-tests suddenly fails:
>
> FAIL vmx (3643 tests, 20 unexpected failures, 10 expected failures)
> FAIL vmx_ept_access_test_not_present (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_read_only (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_write_only (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_read_write (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_execute_only (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_read_execute (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_write_execute (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_read_write_execute (3 tests, 1 unexpected
> failures)
> FAIL vmx_ept_access_test_reserved_bits (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_ignored_bits (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_paddr_not_present_ad_disabled (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_paddr_not_present_ad_enabled (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_paddr_read_only_ad_disabled (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_paddr_read_only_ad_enabled (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_paddr_read_write (3 tests, 1 unexpected failures)
> FAIL vmx_ept_access_test_paddr_read_write_execute (3 tests, 1 unexpected
> failures)
> FAIL vmx_ept_access_test_paddr_read_execute_ad_disabled (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_paddr_read_execute_ad_enabled (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_paddr_not_present_page_fault (3 tests, 1
> unexpected failures)
> FAIL vmx_ept_access_test_force_2m_page (3 tests, 1 unexpected failures)
>
> It seems that reverting the latest commit cb67196a3624 ("x86: setup:
> fixes to alloc_phys setup") solves the issue.
>
> Didn't investigate root-cause yet.
There are two causes at least. One is failure to find a 1G page, which
is fixed easily by giving the tests 2.5 G of memory (-m 2560). It then
fails with a page fault, haven't yet looked at it.
Paolo
