[05/12] KVM: x86: Use KVM-governed feature framework to track "XSAVES enabled"
Commit Message
Use the governed feature framework to track if XSAVES is "enabled", i.e.
if XSAVES can be used by the guest. Add a comment in the SVM code to
explain the very unintuitive logic of deliberately NOT checking if XSAVES
is enumerated in the guest CPUID model.
No functional change intended.
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
arch/x86/kvm/governed_features.h | 1 +
arch/x86/kvm/svm/svm.c | 17 ++++++++++++---
arch/x86/kvm/vmx/vmx.c | 36 ++++++++++++++++----------------
arch/x86/kvm/x86.c | 4 ++--
4 files changed, 35 insertions(+), 23 deletions(-)
Comments
On Fri, Feb 17, 2023 at 03:10:15PM -0800, Sean Christopherson wrote:
> Use the governed feature framework to track if XSAVES is "enabled", i.e.
> if XSAVES can be used by the guest. Add a comment in the SVM code to
> explain the very unintuitive logic of deliberately NOT checking if XSAVES
> is enumerated in the guest CPUID model.
>
> No functional change intended.
xsaves_enabled in struct kvm_vcpu_arch is no longer used. But instead of
just deleting it, maybe we could move 'bool load_eoi_exitmap_pending' to
its place, so 7 bytes can be saved for each struct kvm_vcpu_arch:
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index cd660de02f7b..0eef5469c165 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -740,7 +740,6 @@ struct kvm_vcpu_arch {
u64 efer;
u64 apic_base;
struct kvm_lapic *apic; /* kernel irqchip context */
- bool load_eoi_exitmap_pending;
DECLARE_BITMAP(ioapic_handled_vectors, 256);
unsigned long apic_attention;
int32_t apic_arb_prio;
@@ -750,7 +749,7 @@ struct kvm_vcpu_arch {
u64 smi_count;
bool at_instruction_boundary;
bool tpr_access_reporting;
- bool xsaves_enabled;
+ bool load_eoi_exitmap_pending;
bool xfd_no_write_intercept;
u64 ia32_xss;
u64 microcode_version;
B.R.
Yu
On Tue, Feb 21, 2023, Yu Zhang wrote:
> On Fri, Feb 17, 2023 at 03:10:15PM -0800, Sean Christopherson wrote:
> > Use the governed feature framework to track if XSAVES is "enabled", i.e.
> > if XSAVES can be used by the guest. Add a comment in the SVM code to
> > explain the very unintuitive logic of deliberately NOT checking if XSAVES
> > is enumerated in the guest CPUID model.
> >
> > No functional change intended.
>
> xsaves_enabled in struct kvm_vcpu_arch is no longer used. But instead of
> just deleting it, maybe we could move 'bool load_eoi_exitmap_pending' to
> its place, so 7 bytes can be saved for each struct kvm_vcpu_arch:
I prefer leaving load_eoi_exitmap_pending where it is so that it's co-located with
ioapic_handled_vectors. I agree wasting 7 bytes is unfortunate, but I don't want
to take an ad hoc approach to shrinking per-vCPU structs. See the link below for
more discussion.
https://lore.kernel.org/all/20230213163351.30704-1-minipli@grsecurity.net
>
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index cd660de02f7b..0eef5469c165 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -740,7 +740,6 @@ struct kvm_vcpu_arch {
> u64 efer;
> u64 apic_base;
> struct kvm_lapic *apic; /* kernel irqchip context */
> - bool load_eoi_exitmap_pending;
> DECLARE_BITMAP(ioapic_handled_vectors, 256);
> unsigned long apic_attention;
> int32_t apic_arb_prio;
> @@ -750,7 +749,7 @@ struct kvm_vcpu_arch {
> u64 smi_count;
> bool at_instruction_boundary;
> bool tpr_access_reporting;
> - bool xsaves_enabled;
> + bool load_eoi_exitmap_pending;
> bool xfd_no_write_intercept;
> u64 ia32_xss;
> u64 microcode_version;
>
> B.R.
> Yu
>
On Wed, Feb 22, 2023 at 10:56:04AM -0800, Sean Christopherson wrote:
> On Tue, Feb 21, 2023, Yu Zhang wrote:
> > On Fri, Feb 17, 2023 at 03:10:15PM -0800, Sean Christopherson wrote:
> > > Use the governed feature framework to track if XSAVES is "enabled", i.e.
> > > if XSAVES can be used by the guest. Add a comment in the SVM code to
> > > explain the very unintuitive logic of deliberately NOT checking if XSAVES
> > > is enumerated in the guest CPUID model.
> > >
> > > No functional change intended.
> >
> > xsaves_enabled in struct kvm_vcpu_arch is no longer used. But instead of
> > just deleting it, maybe we could move 'bool load_eoi_exitmap_pending' to
> > its place, so 7 bytes can be saved for each struct kvm_vcpu_arch:
>
> I prefer leaving load_eoi_exitmap_pending where it is so that it's co-located with
> ioapic_handled_vectors. I agree wasting 7 bytes is unfortunate, but I don't want
> to take an ad hoc approach to shrinking per-vCPU structs. See the link below for
> more discussion.
>
> https://lore.kernel.org/all/20230213163351.30704-1-minipli@grsecurity.net
Fair enough. :)
Thanks
Yu
@@ -6,6 +6,7 @@ BUILD_BUG()
#define KVM_GOVERNED_X86_FEATURE(x) KVM_GOVERNED_FEATURE(X86_FEATURE_##x)
KVM_GOVERNED_X86_FEATURE(GBPAGES)
+KVM_GOVERNED_X86_FEATURE(XSAVES)
#undef KVM_GOVERNED_X86_FEATURE
#undef KVM_GOVERNED_FEATURE
@@ -4128,9 +4128,20 @@ static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
struct vcpu_svm *svm = to_svm(vcpu);
struct kvm_cpuid_entry2 *best;
- vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
- boot_cpu_has(X86_FEATURE_XSAVE) &&
- boot_cpu_has(X86_FEATURE_XSAVES);
+ /*
+ * SVM doesn't provide a way to disable just XSAVES in the guest, KVM
+ * can only disable all variants of by disallowing CR4.OSXSAVE from
+ * being set. As a result, if the host has XSAVE and XSAVES, and the
+ * guest has XSAVE enabled, the guest can execute XSAVES without
+ * faulting. Treat XSAVES as enabled in this case regardless of
+ * whether it's advertised to the guest so that KVM context switches
+ * XSS on VM-Enter/VM-Exit. Failure to do so would effectively give
+ * the guest read/write access to the host's XSS.
+ */
+ if (boot_cpu_has(X86_FEATURE_XSAVE) &&
+ boot_cpu_has(X86_FEATURE_XSAVES) &&
+ guest_cpuid_has(vcpu, X86_FEATURE_XSAVE))
+ kvm_governed_feature_set(vcpu, X86_FEATURE_XSAVES);
/* Update nrips enabled cache */
svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
@@ -4551,16 +4551,19 @@ vmx_adjust_secondary_exec_control(struct vcpu_vmx *vmx, u32 *exec_control,
* based on a single guest CPUID bit, with a dedicated feature bit. This also
* verifies that the control is actually supported by KVM and hardware.
*/
-#define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting) \
-({ \
- bool __enabled; \
- \
- if (cpu_has_vmx_##name()) { \
- __enabled = guest_cpuid_has(&(vmx)->vcpu, \
- X86_FEATURE_##feat_name); \
- vmx_adjust_secondary_exec_control(vmx, exec_control, \
- SECONDARY_EXEC_##ctrl_name, __enabled, exiting); \
- } \
+#define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting) \
+({ \
+ struct kvm_vcpu *__vcpu = &(vmx)->vcpu; \
+ bool __enabled; \
+ \
+ if (cpu_has_vmx_##name()) { \
+ if (kvm_is_governed_feature(X86_FEATURE_##feat_name)) \
+ __enabled = guest_can_use(__vcpu, X86_FEATURE_##feat_name); \
+ else \
+ __enabled = guest_cpuid_has(__vcpu, X86_FEATURE_##feat_name); \
+ vmx_adjust_secondary_exec_control(vmx, exec_control, SECONDARY_EXEC_##ctrl_name,\
+ __enabled, exiting); \
+ } \
})
/* More macro magic for ENABLE_/opt-in versus _EXITING/opt-out controls. */
@@ -4620,10 +4623,7 @@ static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
if (!enable_pml || !atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
- if (cpu_has_vmx_xsaves())
- vmx_adjust_secondary_exec_control(vmx, &exec_control,
- SECONDARY_EXEC_ENABLE_XSAVES,
- vcpu->arch.xsaves_enabled, false);
+ vmx_adjust_sec_exec_feature(vmx, &exec_control, xsaves, XSAVES);
/*
* RDPID is also gated by ENABLE_RDTSCP, turn on the control if either
@@ -4642,6 +4642,7 @@ static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
SECONDARY_EXEC_ENABLE_RDTSCP,
rdpid_or_rdtscp_enabled, false);
}
+
vmx_adjust_sec_exec_feature(vmx, &exec_control, invpcid, INVPCID);
vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdrand, RDRAND);
@@ -7705,10 +7706,9 @@ static void vmx_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
* to the guest. XSAVES depends on CR4.OSXSAVE, and CR4.OSXSAVE can be
* set if and only if XSAVE is supported.
*/
- vcpu->arch.xsaves_enabled = cpu_has_vmx_xsaves() &&
- boot_cpu_has(X86_FEATURE_XSAVE) &&
- guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
- guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
+ if (cpu_has_vmx_xsaves() && boot_cpu_has(X86_FEATURE_XSAVE) &&
+ guest_cpuid_has(vcpu, X86_FEATURE_XSAVE))
+ kvm_governed_feature_check_and_set(vcpu, X86_FEATURE_XSAVES);
vmx_setup_uret_msrs(vmx);
@@ -988,7 +988,7 @@ void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
if (vcpu->arch.xcr0 != host_xcr0)
xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
- if (vcpu->arch.xsaves_enabled &&
+ if (guest_can_use(vcpu, X86_FEATURE_XSAVES) &&
vcpu->arch.ia32_xss != host_xss)
wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
}
@@ -1023,7 +1023,7 @@ void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
if (vcpu->arch.xcr0 != host_xcr0)
xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
- if (vcpu->arch.xsaves_enabled &&
+ if (guest_can_use(vcpu, X86_FEATURE_XSAVES) &&
vcpu->arch.ia32_xss != host_xss)
wrmsrl(MSR_IA32_XSS, host_xss);
}