Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Add some new variants/rules that will run some tests using
additional cpu_model_flags values. Especially:

 * CPU models that require "+apic" to be run on RHEL-6 hosts on Intel
   CPUs
 * CPU models that require "-abm,-sse4a" on Intel CPUs
 * CPU models that require using "-svm" because we don't enable nesting
   by default
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

We don't have any dumps for nokvm mode, so there's no point in enabling
all those variants that won't run because of missing dump files.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

This will allow us to filter/document cases that require a specific host
CPU vendor. Enable only host_cpu_vendor=unknown by default.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

This way we can make filters that include/exclude all RHEL-6 versions.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Add extra CPU model flags to some CPU models when the full CPU model
can't be tested easily.

Note that the code loads the reference files using the CPU flags as part
of the file name too (so we can be sure we're reading the right dump
file)  This way we can be sure we are comparing the output of "-cpu
Foo,-flag" against a dump that was surely generated using "-cpu
Foo,-flag" as well (not just "-cpu Foo").
Signed-off-by: NEduardo Habkost <ehabkost@raisama.net>

Those bits were changed on RHEL-6.1 without any compat code, so we need
to ignore the differences we find.
Signed-off-by: NEduardo Habkost <ehabkost@raisama.net>

The CPUID dump test kernel doesn't need USB devices.

This allows some older QEMU versions to be tested.
Signed-off-by: NEduardo Habkost <ehabkost@raisama.net>

Simplify the config file a little bit.
Signed-off-by: NEduardo Habkost <ehabkost@raisama.net>

Signed-off-by: NEduardo Habkost <ehabkost@raisama.net>

I don't know why 486 was being excluded, as the CPUID instruction really
works, and QEMU has CPUID data being set for 486, too. The tests keep
working after removing the "no 486" line from the config file.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

The pclmulqdq bug is fixed on qemu.git on commit
56383703c060777fd01aaf8d63d5f46d660e9fb9, so we can stop ignoring the
bug and report as an actual failure.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

This adds a new test function to the cpuid.py test module, where the
CPUID data exposed by QEMU is compared with a full CPUID dump extracted
previously.

The new test code ensures that there are no ABI changes for a given
machine-type + CPU model combination, between QEMU versions.

The reference dumps were generated using the build-qemu-and-dump-cpus,
and copied using the copy-ref-dumps script.

There are some known QEMU bugs and past ABI changes that require
ignoring some differences. Those cases handled as rules that set
ignore_cpuid_leaves on qemu_cpu.cfg.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

There's no reason to create an additional test object and use confusing
tricks using __new__(), if we already have a test object as the "test"
parameter.

With this, we can convert all test classes to simple functions.

To make sure we kill the VM after running each test case, add
kill_vm=yes and kill_vm_gracefully=no to qemu_cpu.cfg.

Cc: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Keep only the "unknown QEMU version" variant enabled, but allow people
to enable other variants if desired.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Signed-off-by: NJiří Župka <jzupka@redhat.com>

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

Chapters 3.3/3.4 in Hypervisor Top-Level Functional
  Specification 2.0A: Windows Server 2008 R2

CPUID.0x40000000.EAX
Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

- leaf 0x40000000 idx 0x00 ebx,ecx,edx
     - kvm and hyperv variants
  - leaf 0x40000001 idx 0x00 eax
Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

Some tests are x86-specific.  Now that all guests call architectures the
same way in the tuples, filter them away.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This will help us check if we didn't break anything in KVM-specific or
TCG-specific code inside QEMU.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

This will allow us to more easily filter and combine CPU models with the
multiple test cases we have. We can easily introduce new tests that will
automatically be run once for each known CPU model.

Test cases that don't need to be re-run for each CPU model can use the
"cpu.unset" variant that doesn't set any specific cpu_model option.

This also introduces "kvm" and "nokvm" variants, to allow test cases to
be run with KVM disabled and enabled, if necessary.

This way the test output will now look like this:

    (1/94) qemu_cpu.check_models.kvm.cpu.intel.486.qemu13: PASS (0.42 s)
    (2/94) qemu_cpu.check_models.kvm.cpu.intel.core2duo.qemu13: PASS (0.44 s)
    [...]
    (49/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.intel.core2duo.qemu13: PASS (20.63 s)
    (50/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.intel.kvm32.qemu13: PASS (21.66 s)
    [...]
    (62/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.intel.SandyBridge.qemu13: PASS (20.58 s)
    (63/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.intel.Haswell.qemu13: PASS (20.52 s)
    (64/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.amd.qemu64.qemu13: PASS (21.51 s)
    (65/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.amd.phenom.qemu13: PASS (20.53 s)
    [...]
    (71/94) qemu_cpu.cpuid.default.vendor.nokvm.cpu.amd.Opteron_G5.qemu13: PASS (21.59 s)
    (72/94) qemu_cpu.cpuid.custom.vendor.normal.kvm.cpu.unset.qemu13: PASS (20.53 s)
    (73/94) qemu_cpu.cpuid.custom.vendor.tooshort.kvm.cpu.unset.qemu13: PASS (0.87 s)
    (74/94) qemu_cpu.cpuid.custom.vendor.empty.kvm.cpu.unset.qemu13: PASS (0.81 s)
    (75/94) qemu_cpu.cpuid.custom.level.level5.kvm.cpu.unset.qemu13: PASS (20.60 s)
    [...]
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

This shouldn't have any visible change, but it will make it easier to
review the changes made later when introducing the per-CPU-model
variants.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

I don't know yet what will be the best way to configure the tests, but
this changes the test code to allow both modes to be used: a single
model can be set using the "cpu_model" option, or multiple models may be
tested using the "cpu_models" option. Or, if cpu_models is "*", all
models reported by QEMU can be tested (that was the current behavior
when cpu_models was not set).
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Without this option, autotest will set a "cpu_model" config option by
itself.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Instead of just defining rhel64 and qemu13 and enabling both by default,
define rhel64, qemu13, and unknown_qemu.

unknown_qemu will be enabled by default, so we don't test for things
that may fail on non-RHEL-6 or non-qemu-1.3 hosts. This should make the
tests work by default on most hosts, until we implement some kind of
host-based autoconfiguration system.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

cpuid.custom will fail if qemu is built without kvm,
due to virt-test tries to start qemu in kvm mode.

To fix it run custom variants in TCG mode, since they do
not have anything kvm specific.
Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

If config has only 'disable-kvm=yes' but not 'enable-kvm=no'
qemu will be started in kvm mode and with kvm disabled at
compile time it fails to launch.

Currently disable/enable-kvm config options and qemu supported
options are intermingled in a bunch of ifs, separate apples
and oranges in different bins and make logic straightforward:

* alias disable-kvm to enable-kvm param
* get right option set to disable/enable kvm on qemu/qemu-kvm
* set appropriate option depending on value of param enable-kvm
Signed-off-by: NIgor Mammedov <imammedo@redhat.com>

The list of CPU models is the same on RHEL-6 and QEMU-1.3.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

The duplication of settings makes it very hard to spot mistakes. This is
a new version of the config file that uses nesting and cartesian
products of variants to avoid duplication.

To make it easier to test wheter the changes are working, this generates
exactly the same dictionaries as the original config, in exactly the
same sequence, including all the bugs. Look for the #FIXME comments for
places that will be cleaned up or fixed later.
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>
---
v2:
    * s/TestFail/TestNAError/
    * s/custom_model_id/custom.model_id/ in config

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>
---
v2:
    * s/TestFail/TestNAError/
    * s/custom_xlevel/custom.xlevel/ in config

Signed-off-by: NIgor Mammedov <imammedo@redhat.com>
---
v2:
   * s/TestFail/TestNAError/
   * s/custom_stepping/custom.stepping/ in config