[v5,0/2] selftests: kvm: Introduce the mem_slot_test test

Message

Wainer dos Santos Moschetta April 9, 2020, 10:09 p.m. UTC

This series introduces a new KVM selftest (mem_slot_test) that goal
is to verify memory slots can be added up to the maximum allowed. An
extra slot is attempted which should occur on error.

The patch 01 is needed so that the VM fd can be accessed from the
test code (for the ioctl call attempting to add an extra slot).

I ran the test successfully on x86_64, aarch64, and s390x.  This
is why it is enabled to build on those arches.

- Changelog -

v4 -> v5:
 - Initialize the guest_addr and mem_reg_size variables on definition
   [krish.sadhukhan]

v3 -> v4:
 - Discarded mem_reg_flags variable. Simply using 0 instead [drjones]
 - Discarded kvm_region pointer. Instead passing a compound literal in
   the ioctl [drjones]
 - All variables are declared on the declaration block [drjones]

v2 -> v3:
 - Keep alphabetical order of .gitignore and Makefile [drjones]
 - Use memory region flags equals to zero [drjones]
 - Changed mmap() assert from 'mem != NULL' to 'mem != MAP_FAILED' [drjones]
 - kvm_region is declared along side other variables and malloc()'ed
   later [drjones]
 - Combined two asserts into a single 'ret == -1 && errno == EINVAL'
   [drjones]

v1 -> v2:
 - Rebased to queue
 - vm_get_fd() returns int instead of unsigned int (patch 01) [drjones]
 - Removed MEM_REG_FLAGS and GUEST_VM_MODE defines [drjones]
 - Replaced DEBUG() with pr_info() [drjones]
 - Calculate number of guest pages with vm_calc_num_guest_pages()
   [drjones]
 - Using memory region of 1 MB sized (matches mininum needed
   for s390x)
 - Removed the increment of guest_addr after the loop [drjones]
 - Added assert for the errno when adding a slot beyond-the-limit [drjones]
 - Prefer KVM_MEM_READONLY flag but on s390x it switch to KVM_MEM_LOG_DIRTY_PAGES,
   so ensure the coverage of both flags. Also somewhat tests the KVM_CAP_READONLY_MEM capability check [drjones]
 - Moved the test logic to test_add_max_slots(), this allows to more easily add new cases in the "suite".

v1: https://lore.kernel.org/kvm/[e-mail]

Wainer dos Santos Moschetta (2):
  selftests: kvm: Add vm_get_fd() in kvm_util
  selftests: kvm: Add mem_slot_test test

 tools/testing/selftests/kvm/.gitignore        |  1 +
 tools/testing/selftests/kvm/Makefile          |  3 +
 .../testing/selftests/kvm/include/kvm_util.h  |  1 +
 tools/testing/selftests/kvm/lib/kvm_util.c    |  5 ++
 tools/testing/selftests/kvm/mem_slot_test.c   | 69 +++++++++++++++++++
 5 files changed, 79 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/mem_slot_test.c

Comments

Sean Christopherson April 10, 2020, 8:45 p.m. UTC | #1

On Thu, Apr 09, 2020 at 07:09:03PM -0300, Wainer dos Santos Moschetta wrote:
> This series introduces a new KVM selftest (mem_slot_test) that goal
> is to verify memory slots can be added up to the maximum allowed. An
> extra slot is attempted which should occur on error.
> 
> The patch 01 is needed so that the VM fd can be accessed from the
> test code (for the ioctl call attempting to add an extra slot).
> 
> I ran the test successfully on x86_64, aarch64, and s390x.  This
> is why it is enabled to build on those arches.

Any objection to folding these patches into a series I have to clean up
set_memory_region_test (which was mentioned in a prior version) and add
this as a testcase to set_memory_region_test instead of creating a whole
new test?

A large chunk of set_memory_region_test will still be x86_64 only, but
having the test reside in common code will hopefully make it easier to
extend to other architectures.

Andrew Jones April 14, 2020, 6:16 a.m. UTC | #2

On Fri, Apr 10, 2020 at 01:45:09PM -0700, Sean Christopherson wrote:
> On Thu, Apr 09, 2020 at 07:09:03PM -0300, Wainer dos Santos Moschetta wrote:
> > This series introduces a new KVM selftest (mem_slot_test) that goal
> > is to verify memory slots can be added up to the maximum allowed. An
> > extra slot is attempted which should occur on error.
> > 
> > The patch 01 is needed so that the VM fd can be accessed from the
> > test code (for the ioctl call attempting to add an extra slot).
> > 
> > I ran the test successfully on x86_64, aarch64, and s390x.  This
> > is why it is enabled to build on those arches.
> 
> Any objection to folding these patches into a series I have to clean up
> set_memory_region_test (which was mentioned in a prior version) and add
> this as a testcase to set_memory_region_test instead of creating a whole
> new test?
> 
> A large chunk of set_memory_region_test will still be x86_64 only, but
> having the test reside in common code will hopefully make it easier to
> extend to other architectures.
>

Yes, that would be my preference as well. Eventually I decided it could be
done later, but I still prefer it being done from the beginning.

Thanks,
drew