| Message ID | 20200529130407.57176-1-vkuznets@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] selftests: kvm: introduce cpu_has_svm() check | expand |
On Fri, May 29, 2020 at 03:04:06PM +0200, Vitaly Kuznetsov wrote: > More tests may want to check if the CPU is Intel or AMD in > guest code, separate cpu_has_svm() and put it as static > inline to svm_util.h. > > Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com> > --- > tools/testing/selftests/kvm/include/x86_64/svm_util.h | 10 ++++++++++ > tools/testing/selftests/kvm/x86_64/state_test.c | 9 +-------- > 2 files changed, 11 insertions(+), 8 deletions(-) > > diff --git a/tools/testing/selftests/kvm/include/x86_64/svm_util.h b/tools/testing/selftests/kvm/include/x86_64/svm_util.h > index cd037917fece..b1057773206a 100644 > --- a/tools/testing/selftests/kvm/include/x86_64/svm_util.h > +++ b/tools/testing/selftests/kvm/include/x86_64/svm_util.h > @@ -35,4 +35,14 @@ void generic_svm_setup(struct svm_test_data *svm, void *guest_rip, void *guest_r > void run_guest(struct vmcb *vmcb, uint64_t vmcb_gpa); > void nested_svm_check_supported(void); > > +static inline bool cpu_has_svm(void) > +{ > + u32 eax = 0x80000001, ecx; > + > + asm volatile("cpuid" : > + "=a" (eax), "=c" (ecx) : "0" (eax) : "ebx", "edx"); u32 eax, ecx; asm("cpuid" : "=a" (eax), "=c" (ecx) : "a" (0x80000001) : "ebx", "edx"); The volatile shouldn't be needed, e.g. no one should be using this purely for its seralization properties, and I don't see any reason to put the leaf number into a variable. Alternatively, adding a proper cpuid framework to processor.h would likely be useful in the long run. > + > + return ecx & CPUID_SVM; > +} > +
Sean Christopherson <sean.j.christopherson@intel.com> writes: > On Fri, May 29, 2020 at 03:04:06PM +0200, Vitaly Kuznetsov wrote: >> More tests may want to check if the CPU is Intel or AMD in >> guest code, separate cpu_has_svm() and put it as static >> inline to svm_util.h. >> >> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com> >> --- >> tools/testing/selftests/kvm/include/x86_64/svm_util.h | 10 ++++++++++ >> tools/testing/selftests/kvm/x86_64/state_test.c | 9 +-------- >> 2 files changed, 11 insertions(+), 8 deletions(-) >> >> diff --git a/tools/testing/selftests/kvm/include/x86_64/svm_util.h b/tools/testing/selftests/kvm/include/x86_64/svm_util.h >> index cd037917fece..b1057773206a 100644 >> --- a/tools/testing/selftests/kvm/include/x86_64/svm_util.h >> +++ b/tools/testing/selftests/kvm/include/x86_64/svm_util.h >> @@ -35,4 +35,14 @@ void generic_svm_setup(struct svm_test_data *svm, void *guest_rip, void *guest_r >> void run_guest(struct vmcb *vmcb, uint64_t vmcb_gpa); >> void nested_svm_check_supported(void); >> >> +static inline bool cpu_has_svm(void) >> +{ >> + u32 eax = 0x80000001, ecx; >> + >> + asm volatile("cpuid" : >> + "=a" (eax), "=c" (ecx) : "0" (eax) : "ebx", "edx"); > > u32 eax, ecx; > > asm("cpuid" : "=a" (eax), "=c" (ecx) : "a" (0x80000001) : "ebx", "edx"); > > The volatile shouldn't be needed, e.g. no one should be using this purely > for its seralization properties, and I don't see any reason to put the leaf > number into a variable. > > Alternatively, adding a proper cpuid framework to processor.h would likely > be useful in the long run. > All true, even better would be to find a way to include the definition of native_cpuid*() from arch/x86/include/asm/processor.h but I didn't explore these options yet, was trying to address the immediate issue with Paolo's SVM series. It can probably be done when there is a second user of cpuid in tests which needs to check something different from SVM bit.
diff --git a/tools/testing/selftests/kvm/include/x86_64/svm_util.h b/tools/testing/selftests/kvm/include/x86_64/svm_util.h index cd037917fece..b1057773206a 100644 --- a/tools/testing/selftests/kvm/include/x86_64/svm_util.h +++ b/tools/testing/selftests/kvm/include/x86_64/svm_util.h @@ -35,4 +35,14 @@ void generic_svm_setup(struct svm_test_data *svm, void *guest_rip, void *guest_r void run_guest(struct vmcb *vmcb, uint64_t vmcb_gpa); void nested_svm_check_supported(void); +static inline bool cpu_has_svm(void) +{ + u32 eax = 0x80000001, ecx; + + asm volatile("cpuid" : + "=a" (eax), "=c" (ecx) : "0" (eax) : "ebx", "edx"); + + return ecx & CPUID_SVM; +} + #endif /* SELFTEST_KVM_SVM_UTILS_H */ diff --git a/tools/testing/selftests/kvm/x86_64/state_test.c b/tools/testing/selftests/kvm/x86_64/state_test.c index af8b6df6a13e..d43b6f99b66c 100644 --- a/tools/testing/selftests/kvm/x86_64/state_test.c +++ b/tools/testing/selftests/kvm/x86_64/state_test.c @@ -137,20 +137,13 @@ static void vmx_l1_guest_code(struct vmx_pages *vmx_pages) GUEST_ASSERT(vmresume()); } -static u32 cpuid_ecx(u32 eax) -{ - u32 ecx; - asm volatile("cpuid" : "=a" (eax), "=c" (ecx) : "0" (eax) : "ebx", "edx"); - return ecx; -} - static void __attribute__((__flatten__)) guest_code(void *arg) { GUEST_SYNC(1); GUEST_SYNC(2); if (arg) { - if (cpuid_ecx(0x80000001) & CPUID_SVM) + if (cpu_has_svm()) svm_l1_guest_code(arg); else vmx_l1_guest_code(arg);
More tests may want to check if the CPU is Intel or AMD in guest code, separate cpu_has_svm() and put it as static inline to svm_util.h. Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com> --- tools/testing/selftests/kvm/include/x86_64/svm_util.h | 10 ++++++++++ tools/testing/selftests/kvm/x86_64/state_test.c | 9 +-------- 2 files changed, 11 insertions(+), 8 deletions(-)