[PATCH v6 5/5] KVM: SVM: Allow direct access to MSR_IA32_SPEC_CTRL

[KarimAllah Ahmed <karahmed@amazon.de>]

[ Based on a patch from Paolo Bonzini <pbonzini@redhat.com> ]

... basically doing exactly what we do for VMX:

- Passthrough SPEC_CTRL to guests (if enabled in guest CPUID)
- Save and restore SPEC_CTRL around VMExit and VMEntry only if the guest
  actually used it.

Cc: Asit Mallick <asit.k.mallick@intel.com>
Cc: Arjan Van De Ven <arjan.van.de.ven@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jun Nakajima <jun.nakajima@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: David Woodhouse <dwmw@amazon.co.uk>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: KarimAllah Ahmed <karahmed@amazon.de>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
---
v5:
- Add SPEC_CTRL to direct_access_msrs.
---
 arch/x86/kvm/svm.c | 59 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 59 insertions(+)

diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 254eefb..c6ab343 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -184,6 +184,9 @@ struct vcpu_svm {
 		u64 gs_base;
 	} host;
 
+	u64 spec_ctrl;
+	bool save_spec_ctrl_on_exit;
+
 	u32 *msrpm;
 
 	ulong nmi_iret_rip;
@@ -249,6 +252,7 @@ static const struct svm_direct_access_msrs {
 	{ .index = MSR_CSTAR,				.always = true  },
 	{ .index = MSR_SYSCALL_MASK,			.always = true  },
 #endif
+	{ .index = MSR_IA32_SPEC_CTRL,			.always = false },
 	{ .index = MSR_IA32_PRED_CMD,			.always = false },
 	{ .index = MSR_IA32_LASTBRANCHFROMIP,		.always = false },
 	{ .index = MSR_IA32_LASTBRANCHTOIP,		.always = false },
@@ -1584,6 +1588,8 @@ static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 	u32 dummy;
 	u32 eax = 1;
 
+	svm->spec_ctrl = 0;
+
 	if (!init_event) {
 		svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
 					   MSR_IA32_APICBASE_ENABLE;
@@ -3605,6 +3611,13 @@ static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	case MSR_VM_CR:
 		msr_info->data = svm->nested.vm_cr_msr;
 		break;
+	case MSR_IA32_SPEC_CTRL:
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_IBRS))
+			return 1;
+
+		msr_info->data = svm->spec_ctrl;
+		break;
 	case MSR_IA32_UCODE_REV:
 		msr_info->data = 0x01000065;
 		break;
@@ -3696,6 +3709,30 @@ static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
 	case MSR_IA32_TSC:
 		kvm_write_tsc(vcpu, msr);
 		break;
+	case MSR_IA32_SPEC_CTRL:
+		if (!msr->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_IBRS))
+			return 1;
+
+		/* The STIBP bit doesn't fault even if it's not advertised */
+		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP))
+			return 1;
+
+		svm->spec_ctrl = data;
+
+		/*
+		 * When it's written (to non-zero) for the first time, pass
+		 * it through. This means we don't have to take the perf
+		 * hit of saving it on vmexit for the common case of guests
+		 * that don't use it.
+		 */
+		if (data && !svm->save_spec_ctrl_on_exit) {
+			svm->save_spec_ctrl_on_exit = true;
+			if (is_guest_mode(vcpu))
+				break;
+			set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
+		}
+		break;
 	case MSR_IA32_PRED_CMD:
 		if (!msr->host_initiated &&
 		    !guest_cpuid_has(vcpu, X86_FEATURE_IBPB))
@@ -4964,6 +5001,15 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 
 	local_irq_enable();
 
+	/*
+	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
+	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
+	 * is no need to worry about the conditional branch over the wrmsr
+	 * being speculatively taken.
+	 */
+	if (svm->spec_ctrl)
+		wrmsrl(MSR_IA32_SPEC_CTRL, svm->spec_ctrl);
+
 	asm volatile (
 		"push %%" _ASM_BP "; \n\t"
 		"mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
@@ -5056,6 +5102,19 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 #endif
 		);
 
+	/*
+	 * We do not use IBRS in the kernel. If this vCPU has used the
+	 * SPEC_CTRL MSR it may have left it on; save the value and
+	 * turn it off. This is much more efficient than blindly adding
+	 * it to the atomic save/restore list. Especially as the former
+	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
+	 */
+	if (svm->save_spec_ctrl_on_exit)
+		rdmsrl(MSR_IA32_SPEC_CTRL, svm->spec_ctrl);
+
+	if (svm->spec_ctrl)
+		wrmsrl(MSR_IA32_SPEC_CTRL, 0);
+
 	/* Eliminate branch target predictions from guest mode */
 	vmexit_fill_RSB();
 
-- 
2.7.4