Explicitly set the exception vector to #UD when potentially injecting an
exception in sync_regs_test's subtests that try to detect TOCTOU bugs
in KVM's handling of exceptions injected by userspace. A side effect of
the original KVM bug was that KVM would clear the vector, but relying on
KVM to clear the vector (i.e. make it #DE) makes it less likely that the
test would ever find *new* KVM bugs, e.g. because only the first iteration
would run with a legal vector to start.
Explicitly inject #UD for race_events_inj_pen() as well, e.g. so that it
doesn't inherit the illegal 255 vector from race_events_exc(), which
currently runs first.
Link: https://lore.kernel.org/r/20230817233430.1416463-3-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reload known good vCPU state if the vCPU triple faults in any of the
race_sync_regs() subtests, e.g. if KVM successfully injects an exception
(the vCPU isn't configured to handle exceptions). On Intel, the VMCS
is preserved even after shutdown, but AMD's APM states that the VMCB is
undefined after a shutdown and so KVM synthesizes an INIT to sanitize
vCPU/VMCB state, e.g. to guard against running with a garbage VMCB.
The synthetic INIT results in the vCPU never exiting to userspace, as it
gets put into Real Mode at the reset vector, which is full of zeros (as is
GPA 0 and beyond), and so executes ADD for a very, very long time.
Fixes: 60c4063b47 ("KVM: selftests: Extend x86's sync_regs_test to check for event vector races")
Cc: Michal Luczaj <mhal@rbox.co>
Link: https://lore.kernel.org/r/20230817233430.1416463-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
There is already an ASSERT_EQ macro in the file
tools/testing/selftests/kselftest_harness.h, so currently KVM selftests
can't include test_util.h from the KVM selftests together with that file.
Rename the macro in the KVM selftests to TEST_ASSERT_EQ to avoid the
problem - it is also more similar to the other macros in test_util.h that
way.
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Link: https://lore.kernel.org/r/20230712075910.22480-2-thuth@redhat.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Attempt to set the to-be-queued exception to be both pending and injected
_after_ KVM_CAP_SYNC_REGS's kvm_vcpu_ioctl_x86_set_vcpu_events() squashes
the pending exception (if there's also an injected exception). Buggy KVM
versions will eventually yell loudly about having impossible state when
processing queued excpetions, e.g.
WARNING: CPU: 0 PID: 1115 at arch/x86/kvm/x86.c:10095 kvm_check_and_inject_events+0x220/0x500 [kvm]
arch/x86/kvm/x86.c:kvm_check_and_inject_events():
WARN_ON_ONCE(vcpu->arch.exception.injected &&
vcpu->arch.exception.pending);
Signed-off-by: Michal Luczaj <mhal@rbox.co>
Link: https://lore.kernel.org/r/20230728001606.2275586-3-mhal@rbox.co
[sean: split to separate patch, massage changelog and comment]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Attempt to modify the to-be-injected exception vector to an illegal value
_after_ the sanity checks performed by KVM_CAP_SYNC_REGS's
arch/x86/kvm/x86.c:kvm_vcpu_ioctl_x86_set_vcpu_events(). Buggy KVM
versions will eventually yells loudly about attempting to inject a bogus
vector, e.g.
WARNING: CPU: 0 PID: 1107 at arch/x86/kvm/x86.c:547 kvm_check_and_inject_events+0x4a0/0x500 [kvm]
arch/x86/kvm/x86.c:exception_type():
WARN_ON(vector > 31 || vector == NMI_VECTOR)
Signed-off-by: Michal Luczaj <mhal@rbox.co>
Link: https://lore.kernel.org/r/20230728001606.2275586-3-mhal@rbox.co
[sean: split to separate patch]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Attempt to modify vcpu->run->s.regs _after_ the sanity checks performed by
KVM_CAP_SYNC_REGS's arch/x86/kvm/x86.c:sync_regs(). This can lead to some
nonsensical vCPU states accompanied by kernel splats, e.g. disabling PAE
while long mode is enabled makes KVM all kinds of confused:
WARNING: CPU: 0 PID: 1142 at arch/x86/kvm/mmu/paging_tmpl.h:358 paging32_walk_addr_generic+0x431/0x8f0 [kvm]
arch/x86/kvm/mmu/paging_tmpl.h:
KVM_BUG_ON(is_long_mode(vcpu) && !is_pae(vcpu), vcpu->kvm)
Signed-off-by: Michal Luczaj <mhal@rbox.co>
Link: https://lore.kernel.org/r/20230728001606.2275586-3-mhal@rbox.co
[sean: see link]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Add TEST_REQUIRE() and __TEST_REQUIRE() to replace the myriad open coded
instances of selftests exiting with KSFT_SKIP after printing an
informational message. In addition to reducing the amount of boilerplate
code in selftests, the UPPERCASE macro names make it easier to visually
identify a test's requirements.
Convert usage that erroneously uses something other than print_skip()
and/or "exits" with '0' or some other non-KSFT_SKIP value.
Intentionally drop a kvm_vm_free() in aarch64/debug-exceptions.c as part
of the conversion. All memory and file descriptors are freed on process
exit, so the explicit free is superfluous.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Take a vCPU directly instead of a VM+vcpu pair in all vCPU-scoped helpers
and ioctls.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Convert sync_regs_test to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID. Note, this
is a "functional" change in the sense that the test now creates a vCPU
with vcpu_id==0 instead of vcpu_id==5. The non-zero VCPU_ID was 100%
arbitrary and added little to no validation coverage. If testing
non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The guest in sync_regs_test does raw ucalls by directly accessing the
ucall IO port. It makes these ucalls without setting %rdi to a `struct
ucall`, which is what a ucall uses to pass messages. The issue is that
if the host did a get_ucall (the receiver side), it would try to access
the `struct ucall` at %rdi=0 which would lead to an error ("No mapping
for vm virtual address, gva: 0x0").
This issue is currently benign as there is no get_ucall in
sync_regs_test; however, that will change in the next commit as it
changes the unhandled exception reporting mechanism to use ucalls. In
that case, every vcpu_run is followed by a get_ucall to check if the
guest is trying to report an unhandled exception.
Fix this in advance by setting %rdi to a UCALL_NONE struct ucall for the
sync_regs_test guest.
Tested with gcc-[8,9,10], and clang-[9,11].
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20210611011020.3420067-3-ricarkol@google.com
Commit 204c91eff7 ("KVM: selftests: do not blindly clobber registers in
guest asm") was intended to make test more gcc-proof, however, the result
is exactly the opposite: on newer gccs (e.g. 8.2.1) the test breaks with
==== Test Assertion Failure ====
x86_64/sync_regs_test.c:168: run->s.regs.regs.rbx == 0xBAD1DEA + 1
pid=14170 tid=14170 - Invalid argument
1 0x00000000004015b3: main at sync_regs_test.c:166 (discriminator 6)
2 0x00007f413fb66412: ?? ??:0
3 0x000000000040191d: _start at ??:?
rbx sync regs value incorrect 0x1.
Apparently, compile is still free to play games with registers even
when they have variables attached.
Re-write guest code with 'asm volatile' by embedding ucall there and
making sure rbx is preserved.
Fixes: 204c91eff7 ("KVM: selftests: do not blindly clobber registers in guest asm")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The guest_code of sync_regs_test is assuming that the compiler will not
touch %r11 outside the asm that increments it, which is a bit brittle.
Instead, we can increment a variable and use a dummy asm to ensure the
increment is not optimized away. However, we also need to use a
callee-save register or the compiler will insert a save/restore around
the vmexit, breaking the whole idea behind the test.
(Yes, "if it ain't broken...", but I would like the test to be clean
before it is copied into the upcoming s390 selftests).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
