intel-staging · cxdong · Oct 12, 2025 · Oct 12, 2025 · Nov 10, 2025 · Oct 12, 2025
diff --git a/arch/x86/include/asm/kvm_pkvm.h b/arch/x86/include/asm/kvm_pkvm.h
@@ -69,6 +69,9 @@ union pkvm_hc_data {
 	struct {
 		struct pkvm_memcache memcache;
 	} vcpu_free;
+	struct {
+		u64 data;
+	} get_msr;
 	struct {
 		u64 data[PKVM_HC_DATA_MAX_NUM];
 	} raw;
@@ -222,8 +225,12 @@ extern u64 pkvm_sym(sme_me_mask);
 extern struct cpumask pkvm_sym(__cpu_possible_mask);
 extern unsigned int pkvm_sym(nr_cpu_ids);
 DECLARE_STATIC_KEY_FALSE(pkvm_sym(switch_vcpu_ibpb));
+extern u64 pkvm_sym(x86_pred_cmd);
 extern struct fpu_state_config pkvm_sym(fpu_kernel_cfg);
 extern struct fpu_state_config pkvm_sym(fpu_user_cfg);
+#ifdef CONFIG_X86_64
+DECLARE_STATIC_KEY_FALSE(pkvm_sym(__fpu_state_size_dynamic));
+#endif
 
 u64 pkvm_total_reserve_pages(void);
 PKVM_DECLARE(void *, pkvm_early_alloc_page, (void));

diff --git a/arch/x86/include/asm/pkvm_hypercalls.h b/arch/x86/include/asm/pkvm_hypercalls.h
@@ -21,6 +21,12 @@ PKVM_HC(vm_init)
 PKVM_HC_OUT(vm_destroy)
 PKVM_HC(vcpu_create)
 PKVM_HC_OUT(vcpu_free)
+PKVM_HC(vcpu_load)
+PKVM_HC(vcpu_put)
+PKVM_HC(update_exception_bitmap)
+PKVM_HC(set_efer)
+PKVM_HC(set_msr)
+PKVM_HC_OUT(get_msr)
 
 #undef PKVM_HC
 #undef PKVM_HC_OUT
diff --git a/arch/x86/include/asm/pkvm_image_vars.h b/arch/x86/include/asm/pkvm_image_vars.h
@@ -19,6 +19,7 @@ PKVM_ALIAS(__trace_bprintk);
 PKVM_ALIAS(__dynamic_pr_debug);
 PKVM_ALIAS(mem_dump_obj);
 PKVM_ALIAS(vmalloc_base);
+PKVM_ALIAS(get_cpu_entry_area);
 #endif
 
 #endif /* _ASM_x86_PKVM_IMAGE_VARS_H */
diff --git a/arch/x86/kernel/fpu/core.c b/arch/x86/kernel/fpu/core.c
@@ -29,7 +29,7 @@
 #define CREATE_TRACE_POINTS
 #include <asm/trace/fpu.h>
 
-#if defined(CONFIG_X86_64) && !defined(__PKVM_HYP__)
+#ifdef CONFIG_X86_64
 DEFINE_STATIC_KEY_FALSE(__fpu_state_size_dynamic);
 DEFINE_PER_CPU(u64, xfd_state);
 #endif
@@ -217,8 +217,10 @@ void fpu_reset_from_exception_fixup(void)
 {
 	restore_fpregs_from_fpstate(&init_fpstate, XFEATURE_MASK_FPSTATE);
 }
+#endif /* !__PKVM_HYP__ */
 
 #if IS_ENABLED(CONFIG_KVM)
+#ifndef __PKVM_HYP__
 static void __fpstate_reset(struct fpstate *fpstate);
 
 static void fpu_lock_guest_permissions(void)
@@ -315,18 +317,24 @@ int fpu_enable_guest_xfd_features(struct fpu_guest *guest_fpu, u64 xfeatures)
 	return __xfd_enable_feature(xfeatures, guest_fpu);
 }
 EXPORT_SYMBOL_GPL(fpu_enable_guest_xfd_features);
+#endif /* !__PKVM_HYP__ */
 
 #ifdef CONFIG_X86_64
 void fpu_update_guest_xfd(struct fpu_guest *guest_fpu, u64 xfd)
 {
+#ifndef __PKVM_HYP__
 	fpregs_lock();
+#endif
 	guest_fpu->fpstate->xfd = xfd;
 	if (guest_fpu->fpstate->in_use)
 		xfd_update_state(guest_fpu->fpstate);
+#ifndef __PKVM_HYP__
 	fpregs_unlock();
+#endif
 }
 EXPORT_SYMBOL_GPL(fpu_update_guest_xfd);
 
+#ifndef __PKVM_HYP__
 /**
  * fpu_sync_guest_vmexit_xfd_state - Synchronize XFD MSR and software state
  *
@@ -350,8 +358,10 @@ void fpu_sync_guest_vmexit_xfd_state(void)
 	}
 }
 EXPORT_SYMBOL_GPL(fpu_sync_guest_vmexit_xfd_state);
+#endif /* !__PKVM_HYP__ */
 #endif /* CONFIG_X86_64 */
 
+#ifndef __PKVM_HYP__
 int fpu_swap_kvm_fpstate(struct fpu_guest *guest_fpu, bool enter_guest)
 {
 	struct fpstate *guest_fps = guest_fpu->fpstate;
@@ -441,8 +451,10 @@ int fpu_copy_uabi_to_guest_fpstate(struct fpu_guest *gfpu, const void *buf,
 	return copy_uabi_from_kernel_to_xstate(kstate, ustate, vpkru);
 }
 EXPORT_SYMBOL_GPL(fpu_copy_uabi_to_guest_fpstate);
+#endif /* !__PKVM_HYP__ */
 #endif /* CONFIG_KVM */
 
+#ifndef __PKVM_HYP__
 void kernel_fpu_begin_mask(unsigned int kfpu_mask)
 {
 	if (!irqs_disabled())

diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
@@ -81,6 +81,7 @@ u32 xstate_required_size(u64 xstate_bv, bool compacted)
 
 	return ret;
 }
+#endif /* !__PKVM_HYP__ */
 
 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry2(
 	struct kvm_cpuid_entry2 *entries, int nent, u32 function, u64 index)
@@ -134,6 +135,7 @@ struct kvm_cpuid_entry2 *kvm_find_cpuid_entry2(
 }
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_find_cpuid_entry2);
 
+#ifndef __PKVM_HYP__
 static int kvm_check_cpuid(struct kvm_vcpu *vcpu)
 {
 	struct kvm_cpuid_entry2 *best;
@@ -479,6 +481,7 @@ int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 not_found:
 	return 36;
 }
+#endif /* !__PKVM_HYP__ */
 
 int cpuid_query_maxguestphyaddr(struct kvm_vcpu *vcpu)
 {
@@ -504,6 +507,7 @@ u64 kvm_vcpu_reserved_gpa_bits_raw(struct kvm_vcpu *vcpu)
 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
 }
 
+#ifndef __PKVM_HYP__
 static int kvm_set_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *e2,
                         int nent)
 {

diff --git a/arch/x86/kvm/kvm_cache_regs.h b/arch/x86/kvm/kvm_cache_regs.h
@@ -44,6 +44,7 @@ BUILD_KVM_GPR_ACCESSORS(r14, R14)
 BUILD_KVM_GPR_ACCESSORS(r15, R15)
 #endif
 
+#ifndef __PKVM_HYP__
 /*
  * Using the register cache from interrupt context is generally not allowed, as
  * caching a register and marking it available/dirty can't be done atomically,
@@ -55,6 +56,9 @@ BUILD_KVM_GPR_ACCESSORS(r15, R15)
  */
 #define kvm_assert_register_caching_allowed(vcpu)		\
 	lockdep_assert_once(in_task() || kvm_arch_pmi_in_guest(vcpu))
+#else
+#define kvm_assert_register_caching_allowed(vcpu)
+#endif
 
 /*
  * avail  dirty

diff --git a/arch/x86/kvm/pkvm/Makefile b/arch/x86/kvm/pkvm/Makefile
@@ -37,7 +37,7 @@ pkvm-hyp-y			+= $(kernel-lib)/sort.o $(kernel-lib)/bsearch.o \
 				   $(kernel-lib)/ctype.o
 
 kvm				:= ..
-pkvm-hyp-y			+= $(kvm)/x86.o $(kvm)/cpuid.o
+pkvm-hyp-y			+= $(kvm)/x86.o $(kvm)/cpuid.o $(kvm)/mtrr.o
 
 pkvm-hyp-$(CONFIG_PKVM_INTEL)	+= vmx/host_vmentry.o vmx/host_vmx.o \
 				   $(kvm)/vmx/vmx.o vmx/idt.o vmx/ept.o \

diff --git a/arch/x86/kvm/pkvm/cpu.c b/arch/x86/kvm/pkvm/cpu.c
@@ -21,6 +21,7 @@ struct cpumask __cpu_possible_mask __ro_after_init;
 unsigned int nr_cpu_ids;
 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
 DEFINE_STATIC_KEY_FALSE(switch_vcpu_ibpb);
+u64 x86_pred_cmd = PRED_CMD_IBPB;
 
 /*
  * Used to switch the FPU state between the host VM and pVMs. The fpu struct is

diff --git a/arch/x86/kvm/pkvm/entry.S b/arch/x86/kvm/pkvm/entry.S
@@ -2,6 +2,7 @@
 #include <linux/linkage.h>
 #include <asm/asm.h>
 #include <asm/cache.h>
+#include <asm/nospec-branch.h>
 #include <asm/segment.h>
 #include <asm/trapnr.h>
 #include <asm/unwind_hints.h>
@@ -83,3 +84,16 @@ SYM_CODE_START_NOALIGN(x86_verw_sel)
 SYM_CODE_END(x86_verw_sel);
 
 .popsection
+
+/* Clobbers AX, CX, DX */
+SYM_FUNC_START(write_ibpb)
+	ANNOTATE_NOENDBR
+	movl	$MSR_IA32_PRED_CMD, %ecx
+	movl	_ASM_RIP(x86_pred_cmd), %eax
+	xorl	%edx, %edx
+	wrmsr
+
+	/* Make sure IBPB clears return stack preductions too. */
+	FILL_RETURN_BUFFER %rax, RSB_CLEAR_LOOPS, X86_BUG_IBPB_NO_RET
+	RET
+SYM_FUNC_END(write_ibpb)
diff --git a/arch/x86/kvm/pkvm/idt.c b/arch/x86/kvm/pkvm/idt.c
@@ -114,6 +114,8 @@ static bool pkvm_fixup_exception(struct pt_regs *regs)
 	reg  = FIELD_GET(EX_DATA_REG_MASK,  e->data);
 
 	switch (type) {
+	case EX_TYPE_DEFAULT:
+		return ex_handler_default(e, regs);
 	case EX_TYPE_WRMSR:
 		return ex_handler_msr(e, regs, true, false, reg);
 	case EX_TYPE_RDMSR: