[v5,0/7] KVM: selftests: Implement ucall "pool" (for SEV)

Message

Sean Christopherson Aug. 25, 2022, 11:25 p.m. UTC

Non-KVM folks, y'all got pulled in because of the atomic_test_and_set_bit()
patch.

Rework the ucall infrastructure to use a pool of ucall structs to pass
memory instead of using the guest's stack.  For confidential VMs with
encrypted memory, e.g. SEV, the guest's stack "needs" to be private memory
and so can't be used to communicate with the host.

Convert all implementations to the pool as all of the complexity is hidden
in common code, and supporting multiple interfaces adds its own kind of
complexity. 

Tested on x86 and ARM, compile tested on s390 and RISC-V.

v5:
 - Use less convoluted method of writing per-VM "globals". [Oliver]
 - Add patch to drop ucall_uninit().

v4: https://lore.kernel.org/all/[e-mail]

Peter Gonda (2):
  tools: Add atomic_test_and_set_bit()
  KVM: selftests: Add ucall pool based implementation

Sean Christopherson (5):
  KVM: selftests: Consolidate common code for populating ucall struct
  KVM: selftests: Consolidate boilerplate code in get_ucall()
  KVM: selftests: Automatically do init_ucall() for non-barebones VMs
  KVM: selftests: Make arm64's MMIO ucall multi-VM friendly
  KVM: selftest: Drop now-unnecessary ucall_uninit()

 tools/arch/x86/include/asm/atomic.h           |   7 ++
 tools/include/asm-generic/atomic-gcc.h        |  12 +++
 tools/testing/selftests/kvm/Makefile          |   1 +
 .../selftests/kvm/aarch64/arch_timer.c        |   1 -
 .../selftests/kvm/aarch64/debug-exceptions.c  |   1 -
 .../selftests/kvm/aarch64/hypercalls.c        |   1 -
 .../testing/selftests/kvm/aarch64/psci_test.c |   1 -
 .../testing/selftests/kvm/aarch64/vgic_init.c |   2 -
 .../testing/selftests/kvm/aarch64/vgic_irq.c  |   1 -
 tools/testing/selftests/kvm/dirty_log_test.c  |   3 -
 .../selftests/kvm/include/kvm_util_base.h     |  15 +++
 .../selftests/kvm/include/ucall_common.h      |  10 +-
 .../selftests/kvm/kvm_page_table_test.c       |   2 -
 .../testing/selftests/kvm/lib/aarch64/ucall.c | 102 +++---------------
 tools/testing/selftests/kvm/lib/kvm_util.c    |  11 ++
 .../selftests/kvm/lib/perf_test_util.c        |   3 -
 tools/testing/selftests/kvm/lib/riscv/ucall.c |  42 ++------
 tools/testing/selftests/kvm/lib/s390x/ucall.c |  39 ++-----
 .../testing/selftests/kvm/lib/ucall_common.c  | 102 ++++++++++++++++++
 .../testing/selftests/kvm/lib/x86_64/ucall.c  |  39 ++-----
 .../testing/selftests/kvm/memslot_perf_test.c |   1 -
 tools/testing/selftests/kvm/rseq_test.c       |   1 -
 tools/testing/selftests/kvm/steal_time.c      |   1 -
 .../kvm/system_counter_offset_test.c          |   1 -
 24 files changed, 189 insertions(+), 210 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/lib/ucall_common.c


base-commit: 372d07084593dc7a399bf9bee815711b1fb1bcf2

Comments

Peter Gonda Aug. 26, 2022, 9:32 p.m. UTC | #1

On Thu, Aug 25, 2022 at 5:25 PM Sean Christopherson <[e-mail]> wrote:
>
> Non-KVM folks, y'all got pulled in because of the atomic_test_and_set_bit()
> patch.
>
> Rework the ucall infrastructure to use a pool of ucall structs to pass
> memory instead of using the guest's stack.  For confidential VMs with
> encrypted memory, e.g. SEV, the guest's stack "needs" to be private memory
> and so can't be used to communicate with the host.
>
> Convert all implementations to the pool as all of the complexity is hidden
> in common code, and supporting multiple interfaces adds its own kind of
> complexity.
>
> Tested on x86 and ARM, compile tested on s390 and RISC-V.
>

Thanks for the help on the ucall-pool Sean!

I rebased the SEV selftests on these and everything works as before

TESTED-BY: Peter Gonda <[e-mail]>