kvm: mmu: ITLB_MULTIHIT mitigation

With some Intel processors, putting the same virtual address in the TLB
as both a 4 KiB and 2 MiB page can confuse the instruction fetch unit
and cause the processor to issue a machine check resulting in a CPU lockup.

Unfortunately when EPT page tables use huge pages, it is possible for a
malicious guest to cause this situation.

Add a knob to mark huge pages as non-executable. When the nx_huge_pages
parameter is enabled (and we are using EPT), all huge pages are marked as
NX. If the guest attempts to execute in one of those pages, the page is
broken down into 4K pages, which are then marked executable.

This is not an issue for shadow paging (except nested EPT), because then
the host is in control of TLB flushes and the problematic situation cannot
happen.  With nested EPT, again the nested guest can cause problems shadow
and direct EPT is treated in the same way.

[ tglx: Fixup default to auto and massage wording a bit ]

Originally-by: Junaid Shahid <junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Documentation/admin-guide/kernel-parameters.txt

@@ -2055,6 +2055,19 @@
 			KVM MMU at runtime.
 			Default is 0 (off)
 
+	kvm.nx_huge_pages=
+			[KVM] Controls the software workaround for the
+			X86_BUG_ITLB_MULTIHIT bug.
+			force	: Always deploy workaround.
+			off	: Never deploy workaround.
+			auto    : Deploy workaround based on the presence of
+				  X86_BUG_ITLB_MULTIHIT.
+
+			Default is 'auto'.
+
+			If the software workaround is enabled for the host,
+			guests do need not to enable it for nested guests.
+
 	kvm-amd.nested=	[KVM,AMD] Allow nested virtualization in KVM/SVM.
 			Default is 1 (enabled)
 
@@ -2637,6 +2650,12 @@
 					       l1tf=off [X86]
 					       mds=off [X86]
 					       tsx_async_abort=off [X86]
+					       kvm.nx_huge_pages=off [X86]
+
+				Exceptions:
+					       This does not have any effect on
+					       kvm.nx_huge_pages when
+					       kvm.nx_huge_pages=force.
 
 			auto (default)
 				Mitigate all CPU vulnerabilities, but leave SMT

arch/x86/include/asm/kvm_host.h

@@ -315,6 +315,7 @@ struct kvm_mmu_page {
 	bool unsync;
 	u8 mmu_valid_gen;
 	bool mmio_cached;
+	bool lpage_disallowed; /* Can't be replaced by an equiv large page */
 
 	/*
 	 * The following two entries are used to key the shadow page in the
@@ -946,6 +947,7 @@ struct kvm_vm_stat {
 	ulong mmu_unsync;
 	ulong remote_tlb_flush;
 	ulong lpages;
+	ulong nx_lpage_splits;
 	ulong max_mmu_page_hash_collisions;
 };
 

arch/x86/kernel/cpu/bugs.c

@@ -1257,6 +1257,9 @@ void x86_spec_ctrl_setup_ap(void)
 		x86_amd_ssb_disable();
 }
 
+bool itlb_multihit_kvm_mitigation;
+EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
+
 #undef pr_fmt
 #define pr_fmt(fmt)	"L1TF: " fmt
 
@@ -1412,17 +1415,25 @@ static ssize_t l1tf_show_state(char *buf)
 		       l1tf_vmx_states[l1tf_vmx_mitigation],
 		       sched_smt_active() ? "vulnerable" : "disabled");
 }
+
+static ssize_t itlb_multihit_show_state(char *buf)
+{
+	if (itlb_multihit_kvm_mitigation)
+		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
+	else
+		return sprintf(buf, "KVM: Vulnerable\n");
+}
 #else
 static ssize_t l1tf_show_state(char *buf)
 {
 	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
 }
-#endif
 
 static ssize_t itlb_multihit_show_state(char *buf)
 {
 	return sprintf(buf, "Processor vulnerable\n");
 }
+#endif
 
 static ssize_t mds_show_state(char *buf)
 {

arch/x86/kvm/mmu.c

@@ -47,6 +47,20 @@
 #include <asm/kvm_page_track.h>
 #include "trace.h"
 
+extern bool itlb_multihit_kvm_mitigation;
+
+static int __read_mostly nx_huge_pages = -1;
+
+static int set_nx_huge_pages(const char *val, const struct kernel_param *kp);
+
+static struct kernel_param_ops nx_huge_pages_ops = {
+	.set = set_nx_huge_pages,
+	.get = param_get_bool,
+};
+
+module_param_cb(nx_huge_pages, &nx_huge_pages_ops, &nx_huge_pages, 0644);
+__MODULE_PARM_TYPE(nx_huge_pages, "bool");
+
 /*
  * When setting this variable to true it enables Two-Dimensional-Paging
  * where the hardware walks 2 page tables:
@@ -352,6 +366,11 @@ static inline bool spte_ad_need_write_protect(u64 spte)
 	return (spte & SPTE_SPECIAL_MASK) != SPTE_AD_ENABLED_MASK;
 }
 
+static bool is_nx_huge_page_enabled(void)
+{
+	return READ_ONCE(nx_huge_pages);
+}
+
 static inline u64 spte_shadow_accessed_mask(u64 spte)
 {
 	MMU_WARN_ON(is_mmio_spte(spte));
@@ -1190,6 +1209,15 @@ static void account_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp)
 	kvm_mmu_gfn_disallow_lpage(slot, gfn);
 }
 
+static void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp)
+{
+	if (sp->lpage_disallowed)
+		return;
+
+	++kvm->stat.nx_lpage_splits;
+	sp->lpage_disallowed = true;
+}
+
 static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp)
 {
 	struct kvm_memslots *slots;
@@ -1207,6 +1235,12 @@ static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp)
 	kvm_mmu_gfn_allow_lpage(slot, gfn);
 }
 
+static void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp)
+{
+	--kvm->stat.nx_lpage_splits;
+	sp->lpage_disallowed = false;
+}
+
 static bool __mmu_gfn_lpage_is_disallowed(gfn_t gfn, int level,
 					  struct kvm_memory_slot *slot)
 {
@@ -2792,6 +2826,9 @@ static bool __kvm_mmu_prepare_zap_page(struct kvm *kvm,
 			kvm_reload_remote_mmus(kvm);
 	}
 
+	if (sp->lpage_disallowed)
+		unaccount_huge_nx_page(kvm, sp);
+
 	sp->role.invalid = 1;
 	return list_unstable;
 }
@@ -3013,6 +3050,11 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 	if (!speculative)
 		spte |= spte_shadow_accessed_mask(spte);
 
+	if (level > PT_PAGE_TABLE_LEVEL && (pte_access & ACC_EXEC_MASK) &&
+	    is_nx_huge_page_enabled()) {
+		pte_access &= ~ACC_EXEC_MASK;
+	}
+
 	if (pte_access & ACC_EXEC_MASK)
 		spte |= shadow_x_mask;
 	else
@@ -3233,9 +3275,32 @@ static void direct_pte_prefetch(struct kvm_vcpu *vcpu, u64 *sptep)
 	__direct_pte_prefetch(vcpu, sp, sptep);
 }
 
+static void disallowed_hugepage_adjust(struct kvm_shadow_walk_iterator it,
+				       gfn_t gfn, kvm_pfn_t *pfnp, int *levelp)
+{
+	int level = *levelp;
+	u64 spte = *it.sptep;
+
+	if (it.level == level && level > PT_PAGE_TABLE_LEVEL &&
+	    is_nx_huge_page_enabled() &&
+	    is_shadow_present_pte(spte) &&
+	    !is_large_pte(spte)) {
+		/*
+		 * A small SPTE exists for this pfn, but FNAME(fetch)
+		 * and __direct_map would like to create a large PTE
+		 * instead: just force them to go down another level,
+		 * patching back for them into pfn the next 9 bits of
+		 * the address.
+		 */
+		u64 page_mask = KVM_PAGES_PER_HPAGE(level) - KVM_PAGES_PER_HPAGE(level - 1);
+		*pfnp |= gfn & page_mask;
+		(*levelp)--;
+	}
+}
+
 static int __direct_map(struct kvm_vcpu *vcpu, gpa_t gpa, int write,
 			int map_writable, int level, kvm_pfn_t pfn,
-			bool prefault)
+			bool prefault, bool lpage_disallowed)
 {
 	struct kvm_shadow_walk_iterator it;
 	struct kvm_mmu_page *sp;
@@ -3248,6 +3313,12 @@ static int __direct_map(struct kvm_vcpu *vcpu, gpa_t gpa, int write,
 
 	trace_kvm_mmu_spte_requested(gpa, level, pfn);
 	for_each_shadow_entry(vcpu, gpa, it) {
+		/*
+		 * We cannot overwrite existing page tables with an NX
+		 * large page, as the leaf could be executable.
+		 */
+		disallowed_hugepage_adjust(it, gfn, &pfn, &level);
+
 		base_gfn = gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
 		if (it.level == level)
 			break;
@@ -3258,6 +3329,8 @@ static int __direct_map(struct kvm_vcpu *vcpu, gpa_t gpa, int write,
 					      it.level - 1, true, ACC_ALL);
 
 			link_shadow_page(vcpu, it.sptep, sp);
+			if (lpage_disallowed)
+				account_huge_nx_page(vcpu->kvm, sp);
 		}
 	}
 
@@ -3550,11 +3623,14 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, u32 error_code,
 {
 	int r;
 	int level;
-	bool force_pt_level = false;
+	bool force_pt_level;
 	kvm_pfn_t pfn;
 	unsigned long mmu_seq;
 	bool map_writable, write = error_code & PFERR_WRITE_MASK;
+	bool lpage_disallowed = (error_code & PFERR_FETCH_MASK) &&
+				is_nx_huge_page_enabled();
 
+	force_pt_level = lpage_disallowed;
 	level = mapping_level(vcpu, gfn, &force_pt_level);
 	if (likely(!force_pt_level)) {
 		/*
@@ -3588,7 +3664,8 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, u32 error_code,
 		goto out_unlock;
 	if (likely(!force_pt_level))
 		transparent_hugepage_adjust(vcpu, gfn, &pfn, &level);
-	r = __direct_map(vcpu, v, write, map_writable, level, pfn, prefault);
+	r = __direct_map(vcpu, v, write, map_writable, level, pfn,
+			 prefault, false);
 out_unlock:
 	spin_unlock(&vcpu->kvm->mmu_lock);
 	kvm_release_pfn_clean(pfn);
@@ -4174,6 +4251,8 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, u32 error_code,
 	unsigned long mmu_seq;
 	int write = error_code & PFERR_WRITE_MASK;
 	bool map_writable;
+	bool lpage_disallowed = (error_code & PFERR_FETCH_MASK) &&
+				is_nx_huge_page_enabled();
 
 	MMU_WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root_hpa));
 
@@ -4184,8 +4263,9 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, u32 error_code,
 	if (r)
 		return r;
 
-	force_pt_level = !check_hugepage_cache_consistency(vcpu, gfn,
+	force_pt_level =
-							   PT_DIRECTORY_LEVEL);
+		lpage_disallowed ||
+		!check_hugepage_cache_consistency(vcpu, gfn, PT_DIRECTORY_LEVEL);
 	level = mapping_level(vcpu, gfn, &force_pt_level);
 	if (likely(!force_pt_level)) {
 		if (level > PT_DIRECTORY_LEVEL &&
@@ -4214,7 +4294,8 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, u32 error_code,
 		goto out_unlock;
 	if (likely(!force_pt_level))
 		transparent_hugepage_adjust(vcpu, gfn, &pfn, &level);
-	r = __direct_map(vcpu, gpa, write, map_writable, level, pfn, prefault);
+	r = __direct_map(vcpu, gpa, write, map_writable, level, pfn,
+			 prefault, lpage_disallowed);
 out_unlock:
 	spin_unlock(&vcpu->kvm->mmu_lock);
 	kvm_release_pfn_clean(pfn);
@@ -6155,10 +6236,58 @@ static void kvm_set_mmio_spte_mask(void)
 	kvm_mmu_set_mmio_spte_mask(mask, mask, ACC_WRITE_MASK | ACC_USER_MASK);
 }
 
+static bool get_nx_auto_mode(void)
+{
+	/* Return true when CPU has the bug, and mitigations are ON */
+	return boot_cpu_has_bug(X86_BUG_ITLB_MULTIHIT) && !cpu_mitigations_off();
+}
+
+static void __set_nx_huge_pages(bool val)
+{
+	nx_huge_pages = itlb_multihit_kvm_mitigation = val;
+}
+
+static int set_nx_huge_pages(const char *val, const struct kernel_param *kp)
+{
+	bool old_val = nx_huge_pages;
+	bool new_val;
+
+	/* In "auto" mode deploy workaround only if CPU has the bug. */
+	if (sysfs_streq(val, "off"))
+		new_val = 0;
+	else if (sysfs_streq(val, "force"))
+		new_val = 1;
+	else if (sysfs_streq(val, "auto"))
+		new_val = get_nx_auto_mode();
+	else if (strtobool(val, &new_val) < 0)
+		return -EINVAL;
+
+	__set_nx_huge_pages(new_val);
+
+	if (new_val != old_val) {
+		struct kvm *kvm;
+		int idx;
+
+		mutex_lock(&kvm_lock);
+
+		list_for_each_entry(kvm, &vm_list, vm_list) {
+			idx = srcu_read_lock(&kvm->srcu);
+			kvm_mmu_zap_all_fast(kvm);
+			srcu_read_unlock(&kvm->srcu, idx);
+		}
+		mutex_unlock(&kvm_lock);
+	}
+
+	return 0;
+}
+
 int kvm_mmu_module_init(void)
 {
 	int ret = -ENOMEM;
 
+	if (nx_huge_pages == -1)
+		__set_nx_huge_pages(get_nx_auto_mode());
+
 	/*
 	 * MMU roles use union aliasing which is, generally speaking, an
 	 * undefined behavior. However, we supposedly know how compilers behave

arch/x86/kvm/paging_tmpl.h

@@ -614,13 +614,14 @@ static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
 static int FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 			 struct guest_walker *gw,
 			 int write_fault, int hlevel,
-			 kvm_pfn_t pfn, bool map_writable, bool prefault)
+			 kvm_pfn_t pfn, bool map_writable, bool prefault,
+			 bool lpage_disallowed)
 {
 	struct kvm_mmu_page *sp = NULL;
 	struct kvm_shadow_walk_iterator it;
 	unsigned direct_access, access = gw->pt_access;
 	int top_level, ret;
-	gfn_t base_gfn;
+	gfn_t gfn, base_gfn;
 
 	direct_access = gw->pte_access;
 
@@ -665,13 +666,25 @@ static int FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 			link_shadow_page(vcpu, it.sptep, sp);
 	}
 
-	base_gfn = gw->gfn;
+	/*
+	 * FNAME(page_fault) might have clobbered the bottom bits of
+	 * gw->gfn, restore them from the virtual address.
+	 */
+	gfn = gw->gfn | ((addr & PT_LVL_OFFSET_MASK(gw->level)) >> PAGE_SHIFT);
+	base_gfn = gfn;
 
 	trace_kvm_mmu_spte_requested(addr, gw->level, pfn);
 
 	for (; shadow_walk_okay(&it); shadow_walk_next(&it)) {
 		clear_sp_write_flooding_count(it.sptep);
-		base_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
+
+		/*
+		 * We cannot overwrite existing page tables with an NX
+		 * large page, as the leaf could be executable.
+		 */
+		disallowed_hugepage_adjust(it, gfn, &pfn, &hlevel);
+
+		base_gfn = gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
 		if (it.level == hlevel)
 			break;
 
@@ -683,6 +696,8 @@ static int FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 			sp = kvm_mmu_get_page(vcpu, base_gfn, addr,
 					      it.level - 1, true, direct_access);
 			link_shadow_page(vcpu, it.sptep, sp);
+			if (lpage_disallowed)
+				account_huge_nx_page(vcpu->kvm, sp);
 		}
 	}
 
@@ -759,9 +774,11 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
 	int r;
 	kvm_pfn_t pfn;
 	int level = PT_PAGE_TABLE_LEVEL;
-	bool force_pt_level = false;
 	unsigned long mmu_seq;
 	bool map_writable, is_self_change_mapping;
+	bool lpage_disallowed = (error_code & PFERR_FETCH_MASK) &&
+				is_nx_huge_page_enabled();
+	bool force_pt_level = lpage_disallowed;
 
 	pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
 
@@ -851,7 +868,7 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
 	if (!force_pt_level)
 		transparent_hugepage_adjust(vcpu, walker.gfn, &pfn, &level);
 	r = FNAME(fetch)(vcpu, addr, &walker, write_fault,
-			 level, pfn, map_writable, prefault);
+			 level, pfn, map_writable, prefault, lpage_disallowed);
 	kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
 
 out_unlock:

arch/x86/kvm/x86.c

@@ -213,6 +213,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
+	{ "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
 	{ "max_mmu_page_hash_collisions",
 		VM_STAT(max_mmu_page_hash_collisions) },
 	{ NULL }
@@ -1279,6 +1280,14 @@ static u64 kvm_get_arch_capabilities(void)
 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
 
+	/*
+	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
+	 * the nested hypervisor runs with NX huge pages.  If it is not,
+	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
+	 * L1 guests, so it need not worry about its own (L2) guests.
+	 */
+	data |= ARCH_CAP_PSCHANGE_MC_NO;
+
 	/*
 	 * If we're doing cache flushes (either "always" or "cond")
 	 * we will do one whenever the guest does a vmlaunch/vmresume.