| Message ID | 20181109222131.60885-1-jmattson@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2,1/4] kvm: vmx: Set IA32_TSC_AUX for legacy mode guests | expand |
On Fri, Nov 9, 2018 at 2:21 PM, Jim Mattson <jmattson@google.com> wrote: > The SYSCALL instruction is only supported in 64-bit mode on Intel > CPUs. With VT-x, A legacy-mode hypervisor can't launch a long-mode > guest. Therefore, we can omit MSR_STAR support for i386 builds. > > Note that the elided comment has not been relevant since move_msr_up() > was introduced in commit a75beee6e4f5d ("KVM: VMX: Avoid saving and > restoring msrs on lightweight vmexit"). I realize now that eliminating the storage for MSR_STAR from i386 builds breaks syscall emulation on Intel hardware when the guest CPUID claims "AuthenticAMD" or "AMDisbetter!" (?) I'll send out a v3 that leaves the storage for MSR_STAR in i386 builds and clearly documents the reason for it.
> On 12 Nov 2018, at 19:22, Jim Mattson <jmattson@google.com> wrote: > > On Fri, Nov 9, 2018 at 2:21 PM, Jim Mattson <jmattson@google.com> wrote: >> The SYSCALL instruction is only supported in 64-bit mode on Intel >> CPUs. With VT-x, A legacy-mode hypervisor can't launch a long-mode >> guest. Therefore, we can omit MSR_STAR support for i386 builds. >> >> Note that the elided comment has not been relevant since move_msr_up() >> was introduced in commit a75beee6e4f5d ("KVM: VMX: Avoid saving and >> restoring msrs on lightweight vmexit"). > > I realize now that eliminating the storage for MSR_STAR from i386 > builds breaks syscall emulation on Intel hardware when the guest CPUID > claims "AuthenticAMD" or "AMDisbetter!" (?) Heh. “AMDisbetter!” seems to be a joke that was reported from early engineering samples of AMD K5 processor. Go figure ;) > > I'll send out a v3 that leaves the storage for MSR_STAR in i386 builds > and clearly documents the reason for it.
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c index 120fc97a63fc..da7f43457d49 100644 --- a/arch/x86/kvm/vmx.c +++ b/arch/x86/kvm/vmx.c @@ -1387,15 +1387,11 @@ static u64 host_efer; static void ept_save_pdptrs(struct kvm_vcpu *vcpu); -/* - * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it - * away by decrementing the array size. - */ static const u32 vmx_msr_index[] = { #ifdef CONFIG_X86_64 - MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, + MSR_SYSCALL_MASK, MSR_STAR, MSR_LSTAR, MSR_CSTAR, #endif - MSR_EFER, MSR_TSC_AUX, MSR_STAR, + MSR_EFER, MSR_TSC_AUX }; DEFINE_STATIC_KEY_FALSE(enable_evmcs);
The SYSCALL instruction is only supported in 64-bit mode on Intel CPUs. With VT-x, A legacy-mode hypervisor can't launch a long-mode guest. Therefore, we can omit MSR_STAR support for i386 builds. Note that the elided comment has not been relevant since move_msr_up() was introduced in commit a75beee6e4f5d ("KVM: VMX: Avoid saving and restoring msrs on lightweight vmexit"). Signed-off-by: Jim Mattson <jmattson@google.com> Reviewed-by: Peter Shier <pshier@google.com> Reviewed-by: Marc Orr <marcorr@google.com> --- arch/x86/kvm/vmx.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)