Skip to contentSkip to conversation

x86/kvm/nVMX: optimize MMU switch between L1 and L2

RFC PATCH New Vitaly Kuznetsov 1 participant 8 messages Jul 20, 2018

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover x86/kvm/nVMX: optimize MMU switch between L1 and L2

Currently, when we switch from L1 to L2 we do the following:
- Re-initialize L1 MMU as shadow EPT MMU (nested_ept_init_mmu_context())
- Re-initialize 'nested' MMU (nested_vmx_load_cr3() -> init_kvm_nested_mmu())
- Reload MMU root upon guest entry.

When we switch back we do:
- Re-initialize L1 MMU (nested_vmx_load_cr3() -> init_kvm_tdp_mmu())
- Reload MMU root upon guest entry.

This seems to be sub-optimal. Initializing MMU is expensive (thanks to update_permission_bitmask(), update_pkru_bitmask(),..) and reloading MMU root doesn't come for free.

Try to approach the issue by splitting L1-normal and L1-nested MMUs and checking if MMU reset is really needed. This spares us about 1000 cpu cycles on nested vmexit.

RFC part:
- Does this look like a plausible solution?
- SVM nested can probably be optimized in the same way.
- Doesn mmu_update_needed() cover everything?

Vitaly Kuznetsov (7):
x86/kvm/mmu: make vcpu->mmu a pointer to the current MMU x86/kvm/mmu.c: set get_pdptr hook in kvm_init_shadow_ept_mmu() x86/kvm/mmu.c: add kvm_mmu parameter to kvm_mmu_free_roots() x86/kvm/mmu: introduce guest_mmu
x86/kvm/mmu: get rid of redundant kvm_mmu_setup() x86/kvm/nVMX: introduce scache for kvm_init_shadow_ept_mmu x86/kvm/nVMX: optimize MMU switch from nested_vmx_load_cr3()

 arch/x86/include/asm/kvm_host.h |  36 ++++-
 arch/x86/kvm/cpuid.c            |   2 +-
 arch/x86/kvm/mmu.c              | 282 ++++++++++++++++++++++++++--------------
 arch/x86/kvm/mmu.h              |   2 +-
 arch/x86/kvm/mmu_audit.c        |  12 +-
 arch/x86/kvm/paging_tmpl.h      |  17 +--
 arch/x86/kvm/svm.c              |  20 +--
 arch/x86/kvm/vmx.c              |  52 +++++---
 arch/x86/kvm/x86.c              |  34 ++---
 9 files changed, 292 insertions(+), 165 deletions(-)
Signature · 1 line
2.14.4
#1 lore raw

RFC PATCH 1/7 x86/kvm/mmu: make vcpu->mmu a pointer to the current MMU

As a preparation to full MMU split between L1 and L2 make vcpu->arch.mmu a pointer to the currently used mmu. For now, this is always vcpu->arch.root_mmu. No functional change.

---
 arch/x86/include/asm/kvm_host.h |   5 +-
 arch/x86/kvm/mmu.c              | 150 ++++++++++++++++++++--------------------
 arch/x86/kvm/mmu.h              |   2 +-
 arch/x86/kvm/mmu_audit.c        |  12 ++--
 arch/x86/kvm/paging_tmpl.h      |  17 ++---
 arch/x86/kvm/svm.c              |  14 ++--
 arch/x86/kvm/vmx.c              |  13 ++--
 arch/x86/kvm/x86.c              |  20 +++---
 8 files changed, 120 insertions(+), 113 deletions(-)
arch/x86/include/asm/kvm_host.h +4 -1
index c13cd28d9d1b..2aa3c43ca5e5 100644
@@ -521,7 +521,10 @@ struct kvm_vcpu_arch {
 	 * the paging mode of the l1 guest. This context is always used to
 	 * handle faults.
 	 */
-	struct kvm_mmu mmu;
+	struct kvm_mmu *mmu;
+
+	/* Non-nested MMU for L1 */
+	struct kvm_mmu root_mmu;
 
 	/*
 	 * Paging state of an L2 guest (used for nested npt)
arch/x86/kvm/mmu.c +76 -74

450 lines not on this page · raw message

arch/x86/kvm/mmu.h +1 -1
index 5b408c0ad612..8207c1146f4d 100644
@@ -79,7 +79,7 @@ static inline unsigned int kvm_mmu_available_pages(struct kvm *kvm)
 
 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
 {
-	if (likely(vcpu->arch.mmu.root_hpa != INVALID_PAGE))
+	if (likely(vcpu->arch.mmu->root_hpa != INVALID_PAGE))
 		return 0;
 
 	return kvm_mmu_load(vcpu);
arch/x86/kvm/mmu_audit.c +6 -6
index 1272861e77b9..abac7e208853 100644
@@ -59,19 +59,19 @@ static void mmu_spte_walk(struct kvm_vcpu *vcpu, inspect_spte_fn fn)
 	int i;
 	struct kvm_mmu_page *sp;
 
-	if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
+	if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
 		return;
 
-	if (vcpu->arch.mmu.root_level >= PT64_ROOT_4LEVEL) {
-		hpa_t root = vcpu->arch.mmu.root_hpa;
+	if (vcpu->arch.mmu->root_level >= PT64_ROOT_4LEVEL) {
+		hpa_t root = vcpu->arch.mmu->root_hpa;
 
 		sp = page_header(root);
-		__mmu_spte_walk(vcpu, sp, fn, vcpu->arch.mmu.root_level);
+		__mmu_spte_walk(vcpu, sp, fn, vcpu->arch.mmu->root_level);
 		return;
 	}
 
 	for (i = 0; i < 4; ++i) {
-		hpa_t root = vcpu->arch.mmu.pae_root[i];
+		hpa_t root = vcpu->arch.mmu->pae_root[i];
 
 		if (root && VALID_PAGE(root)) {
 			root &= PT64_BASE_ADDR_MASK;
@@ -122,7 +122,7 @@ static void audit_mappings(struct kvm_vcpu *vcpu, u64 *sptep, int level)
 	hpa =  pfn << PAGE_SHIFT;
 	if ((*sptep & PT64_BASE_ADDR_MASK) != hpa)
 		audit_printk(vcpu->kvm, "levels %d pfn %llx hpa %llx "
-			     "ent %llxn", vcpu->arch.mmu.root_level, pfn,
+			     "ent %llxn", vcpu->arch.mmu->root_level, pfn,
 			     hpa, *sptep);
 }
 
arch/x86/kvm/paging_tmpl.h +9 -8
index 6288e9d7068e..e86dc5b9d6cd 100644
@@ -158,14 +158,15 @@ static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu,
 				  struct kvm_mmu_page *sp, u64 *spte,
 				  u64 gpte)
 {
-	if (is_rsvd_bits_set(&vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL))
+	if (is_rsvd_bits_set(vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL))
 		goto no_present;
 
 	if (!FNAME(is_present_gpte)(gpte))
 		goto no_present;
 
 	/* if accessed bit is not supported prefetch non accessed gpte */
-	if (PT_HAVE_ACCESSED_DIRTY(&vcpu->arch.mmu) && !(gpte & PT_GUEST_ACCESSED_MASK))
+	if (PT_HAVE_ACCESSED_DIRTY(vcpu->arch.mmu) &&
+	    !(gpte & PT_GUEST_ACCESSED_MASK))
 		goto no_present;
 
 	return false;
@@ -480,7 +481,7 @@ static int FNAME(walk_addr_generic)(struct guest_walker *walker,
 static int FNAME(walk_addr)(struct guest_walker *walker,
 			    struct kvm_vcpu *vcpu, gva_t addr, u32 access)
 {
-	return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.mmu, addr,
+	return FNAME(walk_addr_generic)(walker, vcpu, vcpu->arch.mmu, addr,
 					access);
 }
 
@@ -509,7 +510,7 @@ FNAME(prefetch_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
 
 	gfn = gpte_to_gfn(gpte);
 	pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
-	FNAME(protect_clean_gpte)(&vcpu->arch.mmu, &pte_access, gpte);
+	FNAME(protect_clean_gpte)(vcpu->arch.mmu, &pte_access, gpte);
 	pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn,
 			no_dirty_log && (pte_access & ACC_WRITE_MASK));
 	if (is_error_pfn(pfn))
@@ -604,7 +605,7 @@ static int FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 
 	direct_access = gw->pte_access;
 
-	top_level = vcpu->arch.mmu.root_level;
+	top_level = vcpu->arch.mmu->root_level;
 	if (top_level == PT32E_ROOT_LEVEL)
 		top_level = PT32_ROOT_LEVEL;
 	/*
@@ -616,7 +617,7 @@ static int FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 	if (FNAME(gpte_changed)(vcpu, gw, top_level))
 		goto out_gpte_changed;
 
-	if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
+	if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
 		goto out_gpte_changed;
 
 	for (shadow_walk_init(&it, vcpu, addr);
@@ -871,7 +872,7 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 	 */
 	mmu_topup_memory_caches(vcpu);
 
-	if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) {
+	if (!VALID_PAGE(vcpu->arch.mmu->root_hpa)) {
 		WARN_ON(1);
 		return;
 	}
@@ -1003,7 +1004,7 @@ static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
 		gfn = gpte_to_gfn(gpte);
 		pte_access = sp->role.access;
 		pte_access &= FNAME(gpte_access)(vcpu, gpte);
-		FNAME(protect_clean_gpte)(&vcpu->arch.mmu, &pte_access, gpte);
+		FNAME(protect_clean_gpte)(vcpu->arch.mmu, &pte_access, gpte);
 
 		if (sync_mmio_spte(vcpu, &sp->spt[i], gfn, pte_access,
 		      &nr_present))
arch/x86/kvm/svm.c +7 -7
index f059a73f0fd0..3b3b9839c2b5 100644
@@ -2920,18 +2920,18 @@ static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
 {
 	WARN_ON(mmu_is_nested(vcpu));
 	kvm_init_shadow_mmu(vcpu);
-	vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
-	vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
-	vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
-	vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
-	vcpu->arch.mmu.shadow_root_level = get_npt_level(vcpu);
-	reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
+	vcpu->arch.mmu->set_cr3           = nested_svm_set_tdp_cr3;
+	vcpu->arch.mmu->get_cr3           = nested_svm_get_tdp_cr3;
+	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
+	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
+	vcpu->arch.mmu->shadow_root_level = get_npt_level(vcpu);
+	reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 }
 
 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 {
-	vcpu->arch.walk_mmu = &vcpu->arch.mmu;
+	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 }
 
 static int nested_svm_check_permissions(struct vcpu_svm *svm)
arch/x86/kvm/vmx.c +7 -6
index 559a12b6184d..418277fd1b7c 100644
@@ -4755,9 +4755,10 @@ static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
 				bool invalidate_gpa)
 {
 	if (enable_ept && (invalidate_gpa || !enable_vpid)) {
-		if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
+		if (!VALID_PAGE(vcpu->arch.mmu->root_hpa))
 			return;
-		ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
+		ept_sync_context(construct_eptp(vcpu,
+						vcpu->arch.mmu->root_hpa));
 	} else {
 		vpid_sync_context(vpid);
 	}
@@ -10577,9 +10578,9 @@ static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
 			to_vmx(vcpu)->nested.msrs.ept_caps &
 			VMX_EPT_EXECUTE_ONLY_BIT,
 			nested_ept_ad_enabled(vcpu));
-	vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
-	vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
-	vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
+	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
+	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
+	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
 
 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 	return 0;
@@ -10587,7 +10588,7 @@ static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
 
 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
 {
-	vcpu->arch.walk_mmu = &vcpu->arch.mmu;
+	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 }
 
 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
arch/x86/kvm/x86.c +10 -10
index 0046aa70205a..a98fafd6ac24 100644
@@ -502,7 +502,7 @@ static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fau
 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 	else
-		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
+		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
 
 	return fault->nested_page_fault;
 }
@@ -601,7 +601,7 @@ int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 		if ((pdpte[i] & PT_PRESENT_MASK) &&
 		    (pdpte[i] &
-		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
+		     vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
 			ret = 0;
 			goto out;
 		}
@@ -4700,7 +4700,7 @@ gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
 
 	/* NPT walks are always user-walks */
 	access |= PFERR_USER_MASK;
-	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
+	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
 
 	return t_gpa;
 }
@@ -5780,7 +5780,7 @@ static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
 		return false;
 
-	if (!vcpu->arch.mmu.direct_map) {
+	if (!vcpu->arch.mmu->direct_map) {
 		/*
 		 * Write permission should be allowed since only
 		 * write access need to be emulated.
@@ -5813,7 +5813,7 @@ static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
 	kvm_release_pfn_clean(pfn);
 
 	/* The instructions are well-emulated on direct mmu. */
-	if (vcpu->arch.mmu.direct_map) {
+	if (vcpu->arch.mmu->direct_map) {
 		unsigned int indirect_shadow_pages;
 
 		spin_lock(&vcpu->kvm->mmu_lock);
@@ -5877,7 +5877,7 @@ static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
 	vcpu->arch.last_retry_eip = ctxt->eip;
 	vcpu->arch.last_retry_addr = cr2;
 
-	if (!vcpu->arch.mmu.direct_map)
+	if (!vcpu->arch.mmu->direct_map)
 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
 
 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
@@ -9171,7 +9171,7 @@ void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
 {
 	int r;
 
-	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
+	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
 	      work->wakeup_all)
 		return;
 
@@ -9179,11 +9179,11 @@ void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
 	if (unlikely(r))
 		return;
 
-	if (!vcpu->arch.mmu.direct_map &&
-	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
+	if (!vcpu->arch.mmu->direct_map &&
+	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
 		return;
 
-	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
+	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
 }
 
 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
Signature · 1 line
2.14.4
#2 lore raw

RFC PATCH 2/7 x86/kvm/mmu.c: set get_pdptr hook in kvm_init_shadow_ept_mmu()

In reply to Vitaly Kuznetsov

kvm_init_shadow_ept_mmu() doesn't set get_pdptr() hook and is this not a problem just because MMU context is already initialized and this hook points to kvm_pdptr_read(). As we're intended to use a dedicated MMU for shadow EPT MMU set this hook explicitly.

---
 arch/x86/kvm/mmu.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/kvm/mmu.c +2 -0
index 914e4eea487a..10598dd396ed 100644
@@ -4570,6 +4570,8 @@ void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
 	context->direct_map = false;
 	context->base_role.ad_disabled = !accessed_dirty;
 	context->base_role.guest_mode = 1;
+	context->get_pdptr = kvm_pdptr_read;
+
 	update_permission_bitmask(vcpu, context, true);
 	update_pkru_bitmask(vcpu, context, true);
 	update_last_nonleaf_level(vcpu, context);
Signature · 1 line
2.14.4
#3 lore raw

RFC PATCH 3/7 x86/kvm/mmu.c: add kvm_mmu parameter to kvm_mmu_free_roots()

In reply to Vitaly Kuznetsov

Add an option to specify which MMU root we want to free. This will be used when nested and non-nested MMUs for L1 are split.

---
 arch/x86/include/asm/kvm_host.h | 2 +-
 arch/x86/kvm/mmu.c              | 7 +++----
 2 files changed, 4 insertions(+), 5 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -1
index 2aa3c43ca5e5..d819947d0a97 100644
@@ -1282,7 +1282,7 @@ void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu);
 int kvm_mmu_load(struct kvm_vcpu *vcpu);
 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu);
-void kvm_mmu_free_roots(struct kvm_vcpu *vcpu);
+void kvm_mmu_free_roots(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu);
 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
 			   struct x86_exception *exception);
 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
arch/x86/kvm/mmu.c +3 -4
index 10598dd396ed..8c5e5664d830 100644
@@ -3358,11 +3358,10 @@ static void mmu_free_root_page(struct kvm *kvm, hpa_t *root_hpa,
 	*root_hpa = INVALID_PAGE;
 }
 
-void kvm_mmu_free_roots(struct kvm_vcpu *vcpu)
+void kvm_mmu_free_roots(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu)
 {
 	int i;
 	LIST_HEAD(invalid_list);
-	struct kvm_mmu *mmu = vcpu->arch.mmu;
 
 	if (!VALID_PAGE(mmu->root_hpa))
 		return;
@@ -3954,7 +3953,7 @@ static void nonpaging_init_context(struct kvm_vcpu *vcpu,
 
 void kvm_mmu_new_cr3(struct kvm_vcpu *vcpu)
 {
-	kvm_mmu_free_roots(vcpu);
+	kvm_mmu_free_roots(vcpu, vcpu->arch.mmu);
 }
 
 static unsigned long get_cr3(struct kvm_vcpu *vcpu)
@@ -4671,7 +4670,7 @@ EXPORT_SYMBOL_GPL(kvm_mmu_load);
 
 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
 {
-	kvm_mmu_free_roots(vcpu);
+	kvm_mmu_free_roots(vcpu, vcpu->arch.mmu);
 	WARN_ON(VALID_PAGE(vcpu->arch.mmu->root_hpa));
 }
 EXPORT_SYMBOL_GPL(kvm_mmu_unload);
Signature · 1 line
2.14.4
#4 lore raw

RFC PATCH 4/7 x86/kvm/mmu: introduce guest_mmu

In reply to Vitaly Kuznetsov

When EPT is used for nested guest we need to re-init MMU as shadow EPT MMU (nested_ept_init_mmu_context() does that). When we return back from L2 to L1 kvm_mmu_reset_context() in nested_vmx_load_cr3() resets MMU back to normal TDP mode. Add a special 'guest_mmu' so we can avoid structure re-initialization and MMU root reload (in future, this is a preparatory patch).

---
 arch/x86/include/asm/kvm_host.h |  3 +++
 arch/x86/kvm/mmu.c              | 12 ++++++++----
 arch/x86/kvm/vmx.c              | 29 ++++++++++++++++++++---------
 3 files changed, 31 insertions(+), 13 deletions(-)
arch/x86/include/asm/kvm_host.h +3 -0
index d819947d0a97..3d0d26142619 100644
@@ -526,6 +526,9 @@ struct kvm_vcpu_arch {
 	/* Non-nested MMU for L1 */
 	struct kvm_mmu root_mmu;
 
+	/* L1 MMU when running nested */
+	struct kvm_mmu guest_mmu;
+
 	/*
 	 * Paging state of an L2 guest (used for nested npt)
 	 *
arch/x86/kvm/mmu.c +8 -4
index 8c5e5664d830..76849eda1a5b 100644
@@ -4670,8 +4670,10 @@ EXPORT_SYMBOL_GPL(kvm_mmu_load);
 
 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
 {
-	kvm_mmu_free_roots(vcpu, vcpu->arch.mmu);
-	WARN_ON(VALID_PAGE(vcpu->arch.mmu->root_hpa));
+	kvm_mmu_free_roots(vcpu, &vcpu->arch.root_mmu);
+	WARN_ON(VALID_PAGE(vcpu->arch.root_mmu.root_hpa));
+	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu);
+	WARN_ON(VALID_PAGE(vcpu->arch.guest_mmu.root_hpa));
 }
 EXPORT_SYMBOL_GPL(kvm_mmu_unload);
 
@@ -5051,8 +5053,10 @@ int kvm_mmu_create(struct kvm_vcpu *vcpu)
 {
 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
-	vcpu->arch.mmu->root_hpa = INVALID_PAGE;
-	vcpu->arch.mmu->translate_gpa = translate_gpa;
+	vcpu->arch.root_mmu.root_hpa = INVALID_PAGE;
+	vcpu->arch.root_mmu.translate_gpa = translate_gpa;
+	vcpu->arch.guest_mmu.root_hpa = INVALID_PAGE;
+	vcpu->arch.guest_mmu.translate_gpa = translate_gpa;
 	vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
 
 	return alloc_mmu_pages(vcpu);
arch/x86/kvm/vmx.c +20 -9
index 418277fd1b7c..5feb52991065 100644
@@ -8012,8 +8012,10 @@ static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
  * just stops using VMX.
  */
-static void free_nested(struct vcpu_vmx *vmx)
+static void free_nested(struct kvm_vcpu *vcpu)
 {
+	struct vcpu_vmx *vmx = to_vmx(vcpu);
+
 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
 		return;
 
@@ -8045,6 +8047,8 @@ static void free_nested(struct vcpu_vmx *vmx)
 		vmx->nested.pi_desc = NULL;
 	}
 
+	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu);
+
 	free_loaded_vmcs(&vmx->nested.vmcs02);
 }
 
@@ -8053,7 +8057,7 @@ static int handle_vmoff(struct kvm_vcpu *vcpu)
 {
 	if (!nested_vmx_check_permission(vcpu))
 		return 1;
-	free_nested(to_vmx(vcpu));
+	free_nested(vcpu);
 	nested_vmx_succeed(vcpu);
 	return kvm_skip_emulated_instruction(vcpu);
 }
@@ -8084,6 +8088,8 @@ static int handle_vmclear(struct kvm_vcpu *vcpu)
 	if (vmptr == vmx->nested.current_vmptr)
 		nested_release_vmcs12(vmx);
 
+	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu);
+
 	kvm_vcpu_write_guest(vcpu,
 			vmptr + offsetof(struct vmcs12, launch_state),
 			&zero, sizeof(zero));
@@ -8428,6 +8434,8 @@ static int handle_vmptrld(struct kvm_vcpu *vcpu)
 		}
 
 		nested_release_vmcs12(vmx);
+
+		kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu);
 		/*
 		 * Load VMCS12 from guest memory since it is not already
 		 * cached.
@@ -10244,12 +10252,12 @@ static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
  */
 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
 {
-       struct vcpu_vmx *vmx = to_vmx(vcpu);
+	struct vcpu_vmx *vmx = to_vmx(vcpu);
 
-       vcpu_load(vcpu);
-       vmx_switch_vmcs(vcpu, &vmx->vmcs01);
-       free_nested(vmx);
-       vcpu_put(vcpu);
+	vcpu_load(vcpu);
+	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
+	free_nested(vcpu);
+	vcpu_put(vcpu);
 }
 
 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
@@ -10573,7 +10581,9 @@ static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
 	if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
 		return 1;
 
-	kvm_mmu_unload(vcpu);
+	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
+	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu);
+
 	kvm_init_shadow_ept_mmu(vcpu,
 			to_vmx(vcpu)->nested.msrs.ept_caps &
 			VMX_EPT_EXECUTE_ONLY_BIT,
@@ -10588,6 +10598,7 @@ static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
 
 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
 {
+	vcpu->arch.mmu = &vcpu->arch.root_mmu;
 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 }
 
@@ -12444,7 +12455,7 @@ static void vmx_leave_nested(struct kvm_vcpu *vcpu)
 		to_vmx(vcpu)->nested.nested_run_pending = 0;
 		nested_vmx_vmexit(vcpu, -1, 0, 0);
 	}
-	free_nested(to_vmx(vcpu));
+	free_nested(vcpu);
 }
 
 /*
Signature · 1 line
2.14.4
#5 lore raw

RFC PATCH 5/7 x86/kvm/mmu: get rid of redundant kvm_mmu_setup()

In reply to Vitaly Kuznetsov

kvm_mmu_setup() is used only once in kvm_arch_vcpu_setup() and kvm_mmu_reset_context() is good enough: we have no root to free so the end result is not any different. init_kvm_mmu() can now be dropped too.

This is a preparatory change to optimizing kvm_mmu_reset_context().

---
 arch/x86/include/asm/kvm_host.h |  1 -
 arch/x86/kvm/mmu.c              | 17 +++--------------
 arch/x86/kvm/x86.c              |  2 +-
 3 files changed, 4 insertions(+), 16 deletions(-)
arch/x86/include/asm/kvm_host.h +0 -1
index 3d0d26142619..2c0b493e09f7 100644
@@ -1128,7 +1128,6 @@ void kvm_mmu_module_exit(void);
 
 void kvm_mmu_destroy(struct kvm_vcpu *vcpu);
 int kvm_mmu_create(struct kvm_vcpu *vcpu);
-void kvm_mmu_setup(struct kvm_vcpu *vcpu);
 void kvm_mmu_init_vm(struct kvm *kvm);
 void kvm_mmu_uninit_vm(struct kvm *kvm);
 void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
arch/x86/kvm/mmu.c +3 -14
index 76849eda1a5b..fb6652643b15 100644
@@ -4633,8 +4633,10 @@ static void init_kvm_nested_mmu(struct kvm_vcpu *vcpu)
 	update_last_nonleaf_level(vcpu, g_context);
 }
 
-static void init_kvm_mmu(struct kvm_vcpu *vcpu)
+void kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
 {
+	kvm_mmu_unload(vcpu);
+
 	if (mmu_is_nested(vcpu))
 		init_kvm_nested_mmu(vcpu);
 	else if (tdp_enabled)
@@ -4642,12 +4644,6 @@ static void init_kvm_mmu(struct kvm_vcpu *vcpu)
 	else
 		init_kvm_softmmu(vcpu);
 }
-
-void kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
-{
-	kvm_mmu_unload(vcpu);
-	init_kvm_mmu(vcpu);
-}
 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
 
 int kvm_mmu_load(struct kvm_vcpu *vcpu)
@@ -5062,13 +5058,6 @@ int kvm_mmu_create(struct kvm_vcpu *vcpu)
 	return alloc_mmu_pages(vcpu);
 }
 
-void kvm_mmu_setup(struct kvm_vcpu *vcpu)
-{
-	MMU_WARN_ON(VALID_PAGE(vcpu->arch.mmu->root_hpa));
-
-	init_kvm_mmu(vcpu);
-}
-
 static void kvm_mmu_invalidate_zap_pages_in_memslot(struct kvm *kvm,
 			struct kvm_memory_slot *slot,
 			struct kvm_page_track_notifier_node *node)
arch/x86/kvm/x86.c +1 -1
index a98fafd6ac24..5510a7f50195 100644
@@ -8333,7 +8333,7 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 	kvm_vcpu_mtrr_init(vcpu);
 	vcpu_load(vcpu);
 	kvm_vcpu_reset(vcpu, false);
-	kvm_mmu_setup(vcpu);
+	kvm_mmu_reset_context(vcpu);
 	vcpu_put(vcpu);
 	return 0;
 }
Signature · 1 line
2.14.4
#6 lore raw

RFC PATCH 6/7 x86/kvm/nVMX: introduce scache for kvm_init_shadow_ept_mmu

In reply to Vitaly Kuznetsov

MMU re-initialization is expensive, in particular, update_permission_bitmask() and update_pkru_bitmask() are.

Cache the data used to setup shadow EPT MMU and avoid full re-init when it is unchanged.

kvm_mmu_free_roots() can be dropped from nested_ept_init_mmu_context() as we always do kvm_mmu_reset_context() in nested_vmx_load_cr3().

---
 arch/x86/include/asm/kvm_host.h | 15 +++++++++++++
 arch/x86/kvm/mmu.c              | 50 ++++++++++++++++++++++++++++++++++++++++-
 arch/x86/kvm/vmx.c              |  5 +++--
 3 files changed, 67 insertions(+), 3 deletions(-)
arch/x86/include/asm/kvm_host.h +15 -0
index 2c0b493e09f7..fa73cf13c4d0 100644
@@ -325,6 +325,19 @@ struct rsvd_bits_validate {
 	u64 bad_mt_xwr;
 };
 
+/* Source data used to setup MMU */
+struct kvm_mmu_sdata_cache {
+	unsigned int valid:1;
+	unsigned int ept_ad:1;
+	unsigned int execonly:1;
+	unsigned int cr0_wp:1;
+	unsigned int cr4_pae:1;
+	unsigned int cr4_pse:1;
+	unsigned int cr4_pke:1;
+	unsigned int cr4_smap:1;
+	unsigned int cr4_smep:1;
+};
+
 /*
  * x86 supports 4 paging modes (5-level 64-bit, 4-level 64-bit, 3-level 32-bit,
  * and 2-level 32-bit).  The kvm_mmu structure abstracts the details of the
@@ -387,6 +400,8 @@ struct kvm_mmu {
 	bool nx;
 
 	u64 pdptrs[4]; /* pae */
+
+	struct kvm_mmu_sdata_cache scache;
 };
 
 enum pmc_type {
arch/x86/kvm/mmu.c +49 -1

61 lines not on this page · raw message

arch/x86/kvm/vmx.c +3 -2
index 5feb52991065..3467665a75d5 100644
@@ -10577,12 +10577,13 @@ static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
 
 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
 {
+	unsigned long cr3 = nested_ept_get_cr3(vcpu);
+
 	WARN_ON(mmu_is_nested(vcpu));
-	if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
+	if (!valid_ept_address(vcpu, cr3))
 		return 1;
 
 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
-	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu);
 
 	kvm_init_shadow_ept_mmu(vcpu,
 			to_vmx(vcpu)->nested.msrs.ept_caps &
Signature · 1 line
2.14.4
#7 lore raw

RFC PATCH 7/7 x86/kvm/nVMX: optimize MMU switch from nested_vmx_load_cr3()

In reply to Vitaly Kuznetsov

Now we have everything in place to stop doing MMU reload when we switch from L1 to L2 and back. Generalize shadow_ept_mmu_update_needed() making it suitable for kvm_mmu_reset_context().

---
 arch/x86/include/asm/kvm_host.h | 10 +++++-
 arch/x86/kvm/cpuid.c            |  2 +-
 arch/x86/kvm/mmu.c              | 74 +++++++++++++++++++++++++++++++----------
 arch/x86/kvm/svm.c              |  6 ++--
 arch/x86/kvm/vmx.c              |  7 ++--
 arch/x86/kvm/x86.c              | 14 ++++----
 6 files changed, 81 insertions(+), 32 deletions(-)
arch/x86/include/asm/kvm_host.h +9 -1
index fa73cf13c4d0..63ad28c40c1d 100644
@@ -327,15 +327,22 @@ struct rsvd_bits_validate {
 
 /* Source data used to setup MMU */
 struct kvm_mmu_sdata_cache {
+	unsigned long cr3;
+
 	unsigned int valid:1;
+	unsigned int smm:1;
 	unsigned int ept_ad:1;
 	unsigned int execonly:1;
+	unsigned int cr0_pg:1;
 	unsigned int cr0_wp:1;
 	unsigned int cr4_pae:1;
 	unsigned int cr4_pse:1;
 	unsigned int cr4_pke:1;
 	unsigned int cr4_smap:1;
 	unsigned int cr4_smep:1;
+	unsigned int cr4_la57:1;
+	unsigned int efer_lma:1;
+	unsigned int efer_nx:1;
 };
 
 /*
@@ -1149,7 +1156,8 @@ void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
 		u64 dirty_mask, u64 nx_mask, u64 x_mask, u64 p_mask,
 		u64 acc_track_mask, u64 me_mask);
 
-void kvm_mmu_reset_context(struct kvm_vcpu *vcpu);
+void kvm_mmu_reset_context(struct kvm_vcpu *vcpu, bool check_if_unchanged);
+
 void kvm_mmu_slot_remove_write_access(struct kvm *kvm,
 				      struct kvm_memory_slot *memslot);
 void kvm_mmu_zap_collapsible_sptes(struct kvm *kvm,
arch/x86/kvm/cpuid.c +1 -1
index 7e042e3d47fd..b0efd08075d8 100644
@@ -142,7 +142,7 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu)
 
 	/* Update physical-address width */
 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
-	kvm_mmu_reset_context(vcpu);
+	kvm_mmu_reset_context(vcpu, false);
 
 	kvm_pmu_refresh(vcpu);
 	return 0;
arch/x86/kvm/mmu.c +56 -18

140 lines not on this page · raw message

arch/x86/kvm/svm.c +3 -3
index 3b3b9839c2b5..6c1db96971c0 100644
@@ -1574,7 +1574,7 @@ static void init_vmcb(struct vcpu_svm *svm)
 	 * It also updates the guest-visible cr0 value.
 	 */
 	svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
-	kvm_mmu_reset_context(&svm->vcpu);
+	kvm_mmu_reset_context(&svm->vcpu, false);
 
 	save->cr4 = X86_CR4_PAE;
 	/* rdx = ?? */
@@ -3380,7 +3380,7 @@ static int nested_svm_vmexit(struct vcpu_svm *svm)
 	nested_svm_unmap(page);
 
 	nested_svm_uninit_mmu_context(&svm->vcpu);
-	kvm_mmu_reset_context(&svm->vcpu);
+	kvm_mmu_reset_context(&svm->vcpu, false);
 	kvm_mmu_load(&svm->vcpu);
 
 	return 0;
@@ -3466,7 +3466,7 @@ static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 		(void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
 
 	/* Guest paging mode is active - reset mmu */
-	kvm_mmu_reset_context(&svm->vcpu);
+	kvm_mmu_reset_context(&svm->vcpu, false);
 
 	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
 	kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
arch/x86/kvm/vmx.c +5 -2
index 3467665a75d5..a85ed004a4ba 100644
@@ -4696,7 +4696,7 @@ static void enter_rmode(struct kvm_vcpu *vcpu)
 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
 
-	kvm_mmu_reset_context(vcpu);
+	kvm_mmu_reset_context(vcpu, false);
 }
 
 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
@@ -11123,6 +11123,8 @@ static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
 			       u32 *entry_failure_code)
 {
+	bool mmu_reset_force = false;
+
 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
 		if (!nested_cr3_valid(vcpu, cr3)) {
 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
@@ -11135,6 +11137,7 @@ static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool ne
 		 */
 		if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
 		    !nested_ept) {
+			mmu_reset_force = true;
 			if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
 				*entry_failure_code = ENTRY_FAIL_PDPTE;
 				return 1;
@@ -11145,7 +11148,7 @@ static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool ne
 		__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 	}
 
-	kvm_mmu_reset_context(vcpu);
+	kvm_mmu_reset_context(vcpu, !mmu_reset_force);
 	return 0;
 }
 
arch/x86/kvm/x86.c +7 -7
index 5510a7f50195..3288a7e303ec 100644
@@ -695,7 +695,7 @@ int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 	}
 
 	if ((cr0 ^ old_cr0) & update_bits)
-		kvm_mmu_reset_context(vcpu);
+		kvm_mmu_reset_context(vcpu, false);
 
 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
@@ -836,7 +836,7 @@ int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 
 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
-		kvm_mmu_reset_context(vcpu);
+		kvm_mmu_reset_context(vcpu, false);
 
 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
 		kvm_update_cpuid(vcpu);
@@ -1162,7 +1162,7 @@ static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 
 	/* Update reserved bits */
 	if ((efer ^ old_efer) & EFER_NX)
-		kvm_mmu_reset_context(vcpu);
+		kvm_mmu_reset_context(vcpu, false);
 
 	return 0;
 }
@@ -5898,7 +5898,7 @@ static void kvm_smm_changed(struct kvm_vcpu *vcpu)
 		kvm_make_request(KVM_REQ_EVENT, vcpu);
 	}
 
-	kvm_mmu_reset_context(vcpu);
+	kvm_mmu_reset_context(vcpu, false);
 }
 
 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
@@ -7156,7 +7156,7 @@ static void enter_smm(struct kvm_vcpu *vcpu)
 		kvm_x86_ops->set_efer(vcpu, 0);
 
 	kvm_update_cpuid(vcpu);
-	kvm_mmu_reset_context(vcpu);
+	kvm_mmu_reset_context(vcpu, false);
 }
 
 static void process_smi(struct kvm_vcpu *vcpu)
@@ -8058,7 +8058,7 @@ static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
 
 	if (mmu_reset_needed)
-		kvm_mmu_reset_context(vcpu);
+		kvm_mmu_reset_context(vcpu, false);
 
 	max_bits = KVM_NR_INTERRUPTS;
 	pending_vec = find_first_bit(
@@ -8333,7 +8333,7 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 	kvm_vcpu_mtrr_init(vcpu);
 	vcpu_load(vcpu);
 	kvm_vcpu_reset(vcpu, false);
-	kvm_mmu_reset_context(vcpu);
+	kvm_mmu_reset_context(vcpu, false);
 	vcpu_put(vcpu);
 	return 0;
 }
Signature · 1 line
2.14.4