Merge tag 'kvm-x86-svm-6.12' of https://github.com/kvm-x86/linux into HEAD

KVM SVM changes for 6.12:

 - Don't stuff the RSB after VM-Exit when RETPOLINE=y and AutoIBRS is enabled,
   i.e. when the CPU has already flushed the RSB.

 - Trace the per-CPU host save area as a VMCB pointer to improve readability
   and cleanup the retrieval of the SEV-ES host save area.

 - Remove unnecessary accounting of temporary nested VMCB related allocations.

Author: bonzini

arch/x86/include/asm/svm.h

@@ -516,6 +516,20 @@ struct ghcb {
 	u32 ghcb_usage;
 } __packed;
 
+struct vmcb {
+	struct vmcb_control_area control;
+	union {
+		struct vmcb_save_area save;
+
+		/*
+		 * For SEV-ES VMs, the save area in the VMCB is used only to
+		 * save/load host state.  Guest state resides in a separate
+		 * page, the aptly named VM Save Area (VMSA), that is encrypted
+		 * with the guest's private key.
+		 */
+		struct sev_es_save_area host_sev_es_save;
+	};
+} __packed;
 
 #define EXPECTED_VMCB_SAVE_AREA_SIZE		744
 #define EXPECTED_GHCB_SAVE_AREA_SIZE		1032
@@ -532,6 +546,7 @@ static inline void __unused_size_checks(void)
 	BUILD_BUG_ON(sizeof(struct ghcb_save_area)	!= EXPECTED_GHCB_SAVE_AREA_SIZE);
 	BUILD_BUG_ON(sizeof(struct sev_es_save_area)	!= EXPECTED_SEV_ES_SAVE_AREA_SIZE);
 	BUILD_BUG_ON(sizeof(struct vmcb_control_area)	!= EXPECTED_VMCB_CONTROL_AREA_SIZE);
+	BUILD_BUG_ON(offsetof(struct vmcb, save)	!= EXPECTED_VMCB_CONTROL_AREA_SIZE);
 	BUILD_BUG_ON(sizeof(struct ghcb)		!= EXPECTED_GHCB_SIZE);
 
 	/* Check offsets of reserved fields */
@@ -568,11 +583,6 @@ static inline void __unused_size_checks(void)
 	BUILD_BUG_RESERVED_OFFSET(ghcb, 0xff0);
 }
 
-struct vmcb {
-	struct vmcb_control_area control;
-	struct vmcb_save_area save;
-} __packed;
-
 #define SVM_CPUID_FUNC 0x8000000a
 
 #define SVM_SELECTOR_S_SHIFT 4

arch/x86/kvm/svm/nested.c

@@ -1693,8 +1693,8 @@ static int svm_set_nested_state(struct kvm_vcpu *vcpu,
 		return -EINVAL;
 
 	ret  = -ENOMEM;
-	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
-	save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
+	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
+	save = kzalloc(sizeof(*save), GFP_KERNEL);
 	if (!ctl || !save)
 		goto out_free;
 

arch/x86/kvm/svm/svm.c

@@ -573,7 +573,7 @@ static void __svm_write_tsc_multiplier(u64 multiplier)
 
 static __always_inline struct sev_es_save_area *sev_es_host_save_area(struct svm_cpu_data *sd)
 {
-	return page_address(sd->save_area) + 0x400;
+	return &sd->save_area->host_sev_es_save;
 }
 
 static inline void kvm_cpu_svm_disable(void)
@@ -696,31 +696,32 @@ static void svm_cpu_uninit(int cpu)
 		return;
 
 	kfree(sd->sev_vmcbs);
-	__free_page(sd->save_area);
+	__free_page(__sme_pa_to_page(sd->save_area_pa));
 	sd->save_area_pa = 0;
 	sd->save_area = NULL;
 }
 
 static int svm_cpu_init(int cpu)
 {
 	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, cpu);
+	struct page *save_area_page;
 	int ret = -ENOMEM;
 
 	memset(sd, 0, sizeof(struct svm_cpu_data));
-	sd->save_area = snp_safe_alloc_page_node(cpu_to_node(cpu), GFP_KERNEL);
-	if (!sd->save_area)
+	save_area_page = snp_safe_alloc_page_node(cpu_to_node(cpu), GFP_KERNEL);
+	if (!save_area_page)
 		return ret;
 
 	ret = sev_cpu_init(sd);
 	if (ret)
 		goto free_save_area;
 
-	sd->save_area_pa = __sme_page_pa(sd->save_area);
+	sd->save_area = page_address(save_area_page);
+	sd->save_area_pa = __sme_page_pa(save_area_page);
 	return 0;
 
 free_save_area:
-	__free_page(sd->save_area);
-	sd->save_area = NULL;
+	__free_page(save_area_page);
 	return ret;
 
 }
@@ -1124,8 +1125,7 @@ static void svm_hardware_unsetup(void)
 	for_each_possible_cpu(cpu)
 		svm_cpu_uninit(cpu);
 
-	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT),
-	get_order(IOPM_SIZE));
+	__free_pages(__sme_pa_to_page(iopm_base), get_order(IOPM_SIZE));
 	iopm_base = 0;
 }
 
@@ -1301,7 +1301,7 @@ static void init_vmcb(struct kvm_vcpu *vcpu)
 	if (!kvm_hlt_in_guest(vcpu->kvm))
 		svm_set_intercept(svm, INTERCEPT_HLT);
 
-	control->iopm_base_pa = __sme_set(iopm_base);
+	control->iopm_base_pa = iopm_base;
 	control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
 	control->int_ctl = V_INTR_MASKING_MASK;
 
@@ -1503,7 +1503,7 @@ static void svm_vcpu_free(struct kvm_vcpu *vcpu)
 
 	sev_free_vcpu(vcpu);
 
-	__free_page(pfn_to_page(__sme_clr(svm->vmcb01.pa) >> PAGE_SHIFT));
+	__free_page(__sme_pa_to_page(svm->vmcb01.pa));
 	__free_pages(virt_to_page(svm->msrpm), get_order(MSRPM_SIZE));
 }
 
@@ -5281,7 +5281,7 @@ static __init int svm_hardware_setup(void)
 
 	iopm_va = page_address(iopm_pages);
 	memset(iopm_va, 0xff, PAGE_SIZE * (1 << order));
-	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
+	iopm_base = __sme_page_pa(iopm_pages);
 
 	init_msrpm_offsets();
 

arch/x86/kvm/svm/svm.h

@@ -25,7 +25,21 @@
 #include "cpuid.h"
 #include "kvm_cache_regs.h"
 
-#define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
+/*
+ * Helpers to convert to/from physical addresses for pages whose address is
+ * consumed directly by hardware.  Even though it's a physical address, SVM
+ * often restricts the address to the natural width, hence 'unsigned long'
+ * instead of 'hpa_t'.
+ */
+static inline unsigned long __sme_page_pa(struct page *page)
+{
+	return __sme_set(page_to_pfn(page) << PAGE_SHIFT);
+}
+
+static inline struct page *__sme_pa_to_page(unsigned long pa)
+{
+	return pfn_to_page(__sme_clr(pa) >> PAGE_SHIFT);
+}
 
 #define	IOPM_SIZE PAGE_SIZE * 3
 #define	MSRPM_SIZE PAGE_SIZE * 2
@@ -321,7 +335,7 @@ struct svm_cpu_data {
 	u32 next_asid;
 	u32 min_asid;
 
-	struct page *save_area;
+	struct vmcb *save_area;
 	unsigned long save_area_pa;
 
 	struct vmcb *current_vmcb;

arch/x86/kvm/svm/vmenter.S

@@ -209,10 +209,8 @@ SYM_FUNC_START(__svm_vcpu_run)
 7:	vmload %_ASM_AX
 8:
 
-#ifdef CONFIG_MITIGATION_RETPOLINE
 	/* IMPORTANT: Stuff the RSB immediately after VM-Exit, before RET! */
-	FILL_RETURN_BUFFER %_ASM_AX, RSB_CLEAR_LOOPS, X86_FEATURE_RETPOLINE
-#endif
+	FILL_RETURN_BUFFER %_ASM_AX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT
 
 	/* Clobbers RAX, RCX, RDX.  */
 	RESTORE_HOST_SPEC_CTRL
@@ -348,10 +346,8 @@ SYM_FUNC_START(__svm_sev_es_vcpu_run)
 
 2:	cli
 
-#ifdef CONFIG_MITIGATION_RETPOLINE
 	/* IMPORTANT: Stuff the RSB immediately after VM-Exit, before RET! */
-	FILL_RETURN_BUFFER %rax, RSB_CLEAR_LOOPS, X86_FEATURE_RETPOLINE
-#endif
+	FILL_RETURN_BUFFER %rax, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT
 
 	/* Clobbers RAX, RCX, RDX, consumes RDI (@svm) and RSI (@spec_ctrl_intercepted). */
 	RESTORE_HOST_SPEC_CTRL