On 8/20/2024 1:24 AM, Aithal, Srikanth wrote:
> On 8/17/2024 12:53 AM, Pratik R. Sampat wrote:
>> From: Michael Roth <michael.roth@xxxxxxx>
>>
>> SEV, SEV-ES, and SNP have a few corner cases where there is potential
>> for KVM_PRE_FAULT_MEMORY to behave differently depending on when it is
>> issued during initial guest setup. Exercising these various paths
>> requires a bit more fine-grained control over when the
>> KVM_PRE_FAULT_MEMORY requests are issued while setting up the guests.
>>
>> Since these CoCo-specific events are likely to be architecture-specific
>> KST helpers, take the existing generic test in pre_fault_memory_test.c
>> as a starting template, and then introduce an x86-specific version of
>> it with expanded coverage for SEV, SEV-ES, and SNP.
>>
>> Since there's a reasonable chance that TDX could extend this for similar
>> testing of TDX, give it a "coco-" prefix rather than an SEV-specific
>> one.
>>
>> Signed-off-by: Michael Roth <michael.roth@xxxxxxx>
>> Co-developed-by: Pratik R. Sampat <pratikrajesh.sampat@xxxxxxx>
>> Signed-off-by: Pratik R. Sampat <pratikrajesh.sampat@xxxxxxx>
>> ---
>> tools/testing/selftests/kvm/Makefile | 1 +
>> .../kvm/x86_64/coco_pre_fault_memory_test.c | 314 ++++++++++++++++++
>> 2 files changed, 315 insertions(+)
>> create mode 100644 tools/testing/selftests/kvm/x86_64/
>> coco_pre_fault_memory_test.c
>>
>> diff --git a/tools/testing/selftests/kvm/Makefile b/tools/testing/
>> selftests/kvm/Makefile
>> index 48d32c5aa3eb..65d19b277b06 100644
>> --- a/tools/testing/selftests/kvm/Makefile
>> +++ b/tools/testing/selftests/kvm/Makefile
>> @@ -129,6 +129,7 @@ TEST_GEN_PROGS_x86_64 += x86_64/amx_test
>> TEST_GEN_PROGS_x86_64 += x86_64/max_vcpuid_cap_test
>> TEST_GEN_PROGS_x86_64 += x86_64/triple_fault_event_test
>> TEST_GEN_PROGS_x86_64 += x86_64/recalc_apic_map_test
>> +TEST_GEN_PROGS_x86_64 += x86_64/coco_pre_fault_memory_test
>> TEST_GEN_PROGS_x86_64 += access_tracking_perf_test
>> TEST_GEN_PROGS_x86_64 += demand_paging_test
>> TEST_GEN_PROGS_x86_64 += dirty_log_test
>> diff --git a/tools/testing/selftests/kvm/x86_64/
>> coco_pre_fault_memory_test.c b/tools/testing/selftests/kvm/x86_64/
>> coco_pre_fault_memory_test.c
>> new file mode 100644
>> index 000000000000..e16fe185fb5a
>> --- /dev/null
>> +++ b/tools/testing/selftests/kvm/x86_64/coco_pre_fault_memory_test.c
>> @@ -0,0 +1,314 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +#include <linux/sizes.h>
>> +
>> +#include <test_util.h>
>> +#include <kvm_util.h>
>> +#include <processor.h>
>> +#include "sev.h"
>> +
>> +/* Arbitrarily chosen values */
>> +#define TEST_SIZE (SZ_2M + PAGE_SIZE)
>> +#define TEST_NPAGES (TEST_SIZE / PAGE_SIZE)
>> +#define TEST_SLOT 10
>> +#define TEST_GPA 0x100000000ul
>> +#define TEST_GVA 0x100000000ul
>> +
>> +enum prefault_snp_test_type {
>> + /* Skip pre-faulting tests. */
>> + NO_PREFAULT_TYPE = 0,
>> + /*
>> + * Issue KVM_PRE_FAULT_MEMORY for GFNs mapping non-private memory
>> + * before finalizing the initial guest contents (e.g. via
>> + * KVM_SEV_SNP_LAUNCH_FINISH for SNP guests).
>> + *
>> + * This should result in failure since KVM explicitly disallows
>> + * KVM_PRE_FAULT_MEMORY from being issued prior to finalizing the
>> + * initial guest contents.
>> + */
>> + PREFAULT_SHARED_BEFORE_FINALIZING = 0,
>> + /*
>> + * Issue KVM_PRE_FAULT_MEMORY for GFNs mapping private memory
>> + * before finalizing the initial guest contents (e.g. via
>> + * KVM_SEV_SNP_LAUNCH_FINISH for SNP guests).
>> + *
>> + * This should result in failure since KVM explicitly disallows
>> + * KVM_PRE_FAULT_MEMORY from being issued prior to finalizing the
>> + * initial guest contents.
>> + */
>> + PREFAULT_PRIVATE_BEFORE_FINALIZING,
>> + /*
>> + * Issue KVM_PRE_FAULT_MEMORY for GFNs mapping shared/private
>> + * memory after finalizing the initial guest contents
>> + * (e.g. via * KVM_SEV_SNP_LAUNCH_FINISH for SNP guests).
>> + *
>> + * This should succeed since pre-faulting is supported for both
>> + * non-private/private memory once the guest contents are finalized.
>> + */
>> + PREFAULT_PRIVATE_SHARED_AFTER_FINALIZING
>> +};
>> +
>> +static void guest_code_sev(void)
>> +{
>> + int i;
>> +
>> + GUEST_ASSERT(rdmsr(MSR_AMD64_SEV) & MSR_AMD64_SEV_ENABLED);
>> +
>> + for (i = 0; i < TEST_NPAGES; i++) {
>> + uint64_t *src = (uint64_t *)(TEST_GVA + i * PAGE_SIZE);
>> + uint64_t val = *src;
>> +
>> + /* Validate the data stored in the pages */
>> + if ((i < TEST_NPAGES / 2 && val != i + 1) ||
>> + (i >= TEST_NPAGES / 2 && val != 0)) {
>> + GUEST_FAIL("Inconsistent view of memory values in guest");
>> + }
>> + }
>> +
>> + if (rdmsr(MSR_AMD64_SEV) & MSR_AMD64_SEV_ES_ENABLED) {
>> + wrmsr(MSR_AMD64_SEV_ES_GHCB, GHCB_MSR_TERM_REQ);
>> + __asm__ __volatile__("rep; vmmcall");
>> + GUEST_FAIL("This should be unreachable.");
>> + }
>> +
>> + GUEST_DONE();
>> +}
>> +
>> +static void __pre_fault_memory(struct kvm_vcpu *vcpu, u64 gpa, u64 size,
>> + u64 left, bool expect_fail)
>> +{
>> + struct kvm_pre_fault_memory range = {
>> + .gpa = gpa,
>> + .size = size,
>> + .flags = 0,
>> + };
>> + int ret, save_errno;
>> + u64 prev;
>> +
>> + do {
>> + prev = range.size;
>> + ret = __vcpu_ioctl(vcpu, KVM_PRE_FAULT_MEMORY, &range);
>> + save_errno = errno;
>> + TEST_ASSERT((range.size < prev) ^ (ret < 0),
>> + "%sexpecting range.size to change on %s",
>> + ret < 0 ? "not " : "",
>> + ret < 0 ? "failure" : "success");
>> + } while (ret >= 0 ? range.size : save_errno == EINTR);
>> +
>> + TEST_ASSERT(expect_fail ? !(range.size == left) : (range.size ==
>> left),
>> + "[EXPECT %s] completed with %lld bytes left, expected %"
>> PRId64,
>> + expect_fail ? "FAIL" : "PASS",
>> + range.size, left);
>> +
>> + if (left == 0) {
>> + TEST_ASSERT(expect_fail ? ret : !ret,
>> + "[EXPECT %s] KVM_PRE_FAULT_MEMORY",
>> + expect_fail ? "FAIL" : "PASS");
>> + } else {
>> + /*
>> + * For shared memory, no memory slot causes RET_PF_EMULATE. It
>> + * results in -ENOENT.
>> + *
>> + * For private memory, no memory slot is an error case returning
>> + * -EFAULT, but it also possible the only the GPA ranges backed
>> + * by a slot are marked as private, in which case the noslot
>> + * range will also result in -ENOENT.
>> + *
>> + * So allow both errors for now, but in the future it would be
>> + * good to distinguish between these cases to tighten up the
>> + * error-checking.
>> + */
>> + TEST_ASSERT(expect_fail ? !ret :
>> + (ret && (save_errno == EFAULT || save_errno == ENOENT)),
>> + "[EXPECT %s] KVM_PRE_FAULT_MEMORY",
>> + expect_fail ? "FAIL" : "PASS");
>> + }
>> +}
>> +
>> +static void pre_fault_memory(struct kvm_vcpu *vcpu, u64 gpa,
>> + u64 size, u64 left)
>> +{
>> + __pre_fault_memory(vcpu, gpa, size, left, false);
>> +}
>> +
>> +static void pre_fault_memory_negative(struct kvm_vcpu *vcpu, u64 gpa,
>> + u64 size, u64 left)
>> +{
>> + __pre_fault_memory(vcpu, gpa, size, left, true);
>> +}
>> +
>> +static void pre_fault_memory_snp(struct kvm_vcpu *vcpu, struct kvm_vm
>> *vm,
>> + bool private, enum prefault_snp_test_type p_type)
>> +{
>> + if (p_type == PREFAULT_SHARED_BEFORE_FINALIZING)
>> + pre_fault_memory_negative(vcpu, TEST_GPA, SZ_2M, 0);
>> +
>> + snp_vm_launch_start(vm, SNP_POLICY);
>> +
>> + if (p_type == PREFAULT_SHARED_BEFORE_FINALIZING)
>> + pre_fault_memory_negative(vcpu, TEST_GPA, SZ_2M, 0);
>> +
>> + if (private) {
>> + /*
>> + * Make sure when pages are pre-faulted later after
>> + * finalization they are treated the same as a private
>> + * access by the guest so that the expected gmem
>> + * backing pages are used.
>> + */
>> + vm_mem_set_private(vm, TEST_GPA, TEST_SIZE);
>> + if (p_type == PREFAULT_PRIVATE_BEFORE_FINALIZING)
>> + pre_fault_memory_negative(vcpu, TEST_GPA, SZ_2M, 0);
>> + } else {
>> + if (p_type == PREFAULT_SHARED_BEFORE_FINALIZING)
>> + pre_fault_memory_negative(vcpu, TEST_GPA, SZ_2M, 0);
>> + }
>> +
>> + snp_vm_launch_update(vm);
>> +
>> + if (p_type == PREFAULT_SHARED_BEFORE_FINALIZING)
>> + pre_fault_memory_negative(vcpu, TEST_GPA, SZ_2M, 0);
>> +
>> + snp_vm_launch_finish(vm);
>> +
>> + /*
>> + * After finalization, pre-faulting either private or shared
>> + * ranges should work regardless of whether the pages were
>> + * encrypted as part of setting up initial guest state.
>> + */
>> + if (p_type == PREFAULT_PRIVATE_SHARED_AFTER_FINALIZING) {
>> + pre_fault_memory(vcpu, TEST_GPA, SZ_2M, 0);
>> + pre_fault_memory(vcpu, TEST_GPA + SZ_2M, PAGE_SIZE * 2,
>> PAGE_SIZE);
>> + pre_fault_memory(vcpu, TEST_GPA + TEST_SIZE, PAGE_SIZE,
>> PAGE_SIZE);
>> + }
>> +}
>> +
>> +static void pre_fault_memory_sev(unsigned long vm_type, struct
>> kvm_vcpu *vcpu,
>> + struct kvm_vm *vm)
>> +{
>> + uint32_t policy = (vm_type == KVM_X86_SEV_ES_VM) ?
>> SEV_POLICY_ES : 0;
>> +
>> + pre_fault_memory(vcpu, TEST_GPA, SZ_2M, 0);
>> + pre_fault_memory(vcpu, TEST_GPA + SZ_2M, PAGE_SIZE * 2, PAGE_SIZE);
>> + pre_fault_memory(vcpu, TEST_GPA + TEST_SIZE, PAGE_SIZE, PAGE_SIZE);
>> +
>> + sev_vm_launch(vm, policy);
>> +
>> + pre_fault_memory(vcpu, TEST_GPA, SZ_2M, 0);
>> + pre_fault_memory(vcpu, TEST_GPA + SZ_2M, PAGE_SIZE * 2, PAGE_SIZE);
>> + pre_fault_memory(vcpu, TEST_GPA + TEST_SIZE, PAGE_SIZE, PAGE_SIZE);
>> +
>> + sev_vm_launch_measure(vm, alloca(256));
>> +
>> + pre_fault_memory(vcpu, TEST_GPA, SZ_2M, 0);
>> + pre_fault_memory(vcpu, TEST_GPA + SZ_2M, PAGE_SIZE * 2, PAGE_SIZE);
>> + pre_fault_memory(vcpu, TEST_GPA + TEST_SIZE, PAGE_SIZE, PAGE_SIZE);
>> +
>> + sev_vm_launch_finish(vm);
>> +
>> + pre_fault_memory(vcpu, TEST_GPA, SZ_2M, 0);
>> + pre_fault_memory(vcpu, TEST_GPA + SZ_2M, PAGE_SIZE * 2, PAGE_SIZE);
>> + pre_fault_memory(vcpu, TEST_GPA + TEST_SIZE, PAGE_SIZE, PAGE_SIZE);
>> +}
>> +
>> +static void test_pre_fault_memory_sev(unsigned long vm_type, bool
>> private,
>> + enum prefault_snp_test_type p_type)
>> +{
>> + struct kvm_vcpu *vcpu;
>> + struct kvm_vm *vm;
>> + struct ucall uc;
>> + int i;
>> +
>> + vm = vm_sev_create_with_one_vcpu(vm_type, guest_code_sev, &vcpu);
>> +
>> + vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
>> + TEST_GPA, TEST_SLOT, TEST_NPAGES,
>> + (vm_type == KVM_X86_SNP_VM) ?
>> KVM_MEM_GUEST_MEMFD : 0);
>> +
>> + /*
>> + * Make sure guest page table is in agreement with what pages
>> will be
>> + * initially encrypted by the ASP.
>> + */
>> + if (private)
>> + vm_mem_set_protected(vm, TEST_SLOT, TEST_GPA, TEST_NPAGES);
>> +
>> + virt_map(vm, TEST_GVA, TEST_GPA, TEST_NPAGES);
>> +
>> + /*
>> + * Populate the pages to compare data read from the guest
>> + * Populate the first half with data and second half as all zeros.
>> + */
>> + for (i = 0; i < TEST_NPAGES; i++) {
>> + uint64_t *hva = addr_gva2hva(vm, TEST_GVA + i * PAGE_SIZE);
>> +
>> + if (i < TEST_NPAGES / 2)
>> + *hva = i + 1;
>> + else
>> + *hva = 0;
>> + }
>> +
>> + if (vm_type == KVM_X86_SNP_VM)
>> + pre_fault_memory_snp(vcpu, vm, private, p_type);
>> + else
>> + pre_fault_memory_sev(vm_type, vcpu, vm);
>> +
>> + vcpu_run(vcpu);
>> +
>> + if (vm->type == KVM_X86_SEV_ES_VM || vm->type == KVM_X86_SNP_VM) {
>> + TEST_ASSERT(vcpu->run->exit_reason == KVM_EXIT_SYSTEM_EVENT,
>> + "Wanted SYSTEM_EVENT, got %s",
>> + exit_reason_str(vcpu->run->exit_reason));
>> + TEST_ASSERT_EQ(vcpu->run->system_event.type,
>> KVM_SYSTEM_EVENT_SEV_TERM);
>> + TEST_ASSERT_EQ(vcpu->run->system_event.ndata, 1);
>> + TEST_ASSERT_EQ(vcpu->run->system_event.data[0],
>> GHCB_MSR_TERM_REQ);
>> + goto out;
>> + }
>> +
>> + switch (get_ucall(vcpu, &uc)) {
>> + case UCALL_DONE:
>> + break;
>> + case UCALL_ABORT:
>> + REPORT_GUEST_ASSERT(uc);
>> + default:
>> + TEST_FAIL("Unexpected exit: %s",
>> + exit_reason_str(vcpu->run->exit_reason));
>> + }
>> +
>> +out:
>> + kvm_vm_free(vm);
>> +}
>> +
>> +static void test_pre_fault_memory(unsigned long vm_type, bool private)
>> +{
>> + int pt;
>> +
>> + if (vm_type && !(kvm_check_cap(KVM_CAP_VM_TYPES) & BIT(vm_type))) {
>> + pr_info("Skipping tests for vm_type 0x%lx\n", vm_type);
>> + return;
>> + }
>> +
>> + switch (vm_type) {
>> + case KVM_X86_SEV_VM:
>> + case KVM_X86_SEV_ES_VM:
>> + test_pre_fault_memory_sev(vm_type, private, NO_PREFAULT_TYPE);
>> + break;
>> + case KVM_X86_SNP_VM:
>> + for (pt = 0; pt <= PREFAULT_PRIVATE_SHARED_AFTER_FINALIZING;
>> pt++)
>> + test_pre_fault_memory_sev(vm_type, private, pt);
>> + break;
>> + default:
>> + abort();
>> + }
>> +}
>> +
>> +int main(int argc, char *argv[])
>> +{
>> + TEST_REQUIRE(kvm_check_cap(KVM_CAP_PRE_FAULT_MEMORY));
>> +
>> + test_pre_fault_memory(KVM_X86_SEV_VM, false);
>> + test_pre_fault_memory(KVM_X86_SEV_VM, true);
>> + test_pre_fault_memory(KVM_X86_SEV_ES_VM, false);
>> + test_pre_fault_memory(KVM_X86_SEV_ES_VM, true);
>> + test_pre_fault_memory(KVM_X86_SNP_VM, false);
>> + test_pre_fault_memory(KVM_X86_SNP_VM, true);
>> +
>> + return 0;
>> +}
>
> Hello Pratik,
> I see below failure while running this test [kvm-x86/next + mentioned
> patches]:
>
> # selftests: kvm: coco_pre_fault_memory_test
> # Random seed: 0x6b8b4567
> # ==== Test Assertion Failure ====
> # x86_64/coco_pre_fault_memory_test.c:145: expect_fail ? !(range.size
> == left) : (range.size == left)
> # pid=202665 tid=202665 errno=9 - Bad file descriptor
> # 1 0x0000000000402870: __pre_fault_memory at
> coco_pre_fault_memory_test.c:145
> # 2 0x00000000004031c9: pre_fault_memory_negative at
> coco_pre_fault_memory_test.c:184
> # 3 (inlined by) pre_fault_memory_snp at
> coco_pre_fault_memory_test.c:202
> # 4 (inlined by) test_pre_fault_memory_sev at
> coco_pre_fault_memory_test.c:344
> # 5 0x00000000004033c0: test_pre_fault_memory at
> coco_pre_fault_memory_test.c:401 (discriminator 3)
> # 6 0x00000000004024d7: main at
> coco_pre_fault_memory_test.c:417 (discriminator 2)
> # 7 0x00007f9474829d8f: ?? ??:0
> # 8 0x00007f9474829e3f: ?? ??:0
> # 9 0x0000000000402574: _start at ??:?
> # [EXPECT FAIL] completed with 0 bytes left, expected 0
> not ok 66 selftests: kvm: coco_pre_fault_memory_test # exit=254
>
Hi Srikanth,
Thanks for testing these patches.
I believe that you may have to test these patches either over the
kvm-x86/fixes branch or over kvm/[queue/next] since there are a few
fixes (eg. KVM: x86: disallow pre-fault for SNP VMs before
initialization, etc.) which are not present in kvm-x86/next.
Do let me know if that works for you instead for the tests?
Thanks!
Pratik
>
>
