KVM: Synchronize guest physical memory map to host virtual memory map

Synchronize changes to host virtual addresses which are part of
a KVM memory slot to the KVM shadow mmu.  This allows pte operations
like swapping, page migration, and madvise() to transparently work
with KVM.
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>

arch/x86/kvm/mmu.c

@@ -653,6 +653,84 @@ static void rmap_write_protect(struct kvm *kvm, u64 gfn)
 	account_shadowed(kvm, gfn);
 }
 
+static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
+{
+	u64 *spte;
+	int need_tlb_flush = 0;
+
+	while ((spte = rmap_next(kvm, rmapp, NULL))) {
+		BUG_ON(!(*spte & PT_PRESENT_MASK));
+		rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte);
+		rmap_remove(kvm, spte);
+		set_shadow_pte(spte, shadow_trap_nonpresent_pte);
+		need_tlb_flush = 1;
+	}
+	return need_tlb_flush;
+}
+
+static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
+			  int (*handler)(struct kvm *kvm, unsigned long *rmapp))
+{
+	int i;
+	int retval = 0;
+
+	/*
+	 * If mmap_sem isn't taken, we can look the memslots with only
+	 * the mmu_lock by skipping over the slots with userspace_addr == 0.
+	 */
+	for (i = 0; i < kvm->nmemslots; i++) {
+		struct kvm_memory_slot *memslot = &kvm->memslots[i];
+		unsigned long start = memslot->userspace_addr;
+		unsigned long end;
+
+		/* mmu_lock protects userspace_addr */
+		if (!start)
+			continue;
+
+		end = start + (memslot->npages << PAGE_SHIFT);
+		if (hva >= start && hva < end) {
+			gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT;
+			retval |= handler(kvm, &memslot->rmap[gfn_offset]);
+			retval |= handler(kvm,
+					  &memslot->lpage_info[
+						  gfn_offset /
+						  KVM_PAGES_PER_HPAGE].rmap_pde);
+		}
+	}
+
+	return retval;
+}
+
+int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
+{
+	return kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
+}
+
+static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp)
+{
+	u64 *spte;
+	int young = 0;
+
+	spte = rmap_next(kvm, rmapp, NULL);
+	while (spte) {
+		int _young;
+		u64 _spte = *spte;
+		BUG_ON(!(_spte & PT_PRESENT_MASK));
+		_young = _spte & PT_ACCESSED_MASK;
+		if (_young) {
+			young = 1;
+			clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
+		}
+		spte = rmap_next(kvm, rmapp, spte);
+	}
+	return young;
+}
+
+int kvm_age_hva(struct kvm *kvm, unsigned long hva)
+{
+	return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
+}
+
 #ifdef MMU_DEBUG
 static int is_empty_shadow_page(u64 *spt)
 {
@@ -1203,6 +1281,7 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 	int r;
 	int largepage = 0;
 	pfn_t pfn;
+	unsigned long mmu_seq;
 
 	down_read(&current->mm->mmap_sem);
 	if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
@@ -1210,6 +1289,8 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 		largepage = 1;
 	}
 
+	mmu_seq = vcpu->kvm->mmu_notifier_seq;
+	/* implicit mb(), we'll read before PT lock is unlocked */
 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
 	up_read(&current->mm->mmap_sem);
 
@@ -1220,6 +1301,8 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 	}
 
 	spin_lock(&vcpu->kvm->mmu_lock);
+	if (mmu_notifier_retry(vcpu, mmu_seq))
+		goto out_unlock;
 	kvm_mmu_free_some_pages(vcpu);
 	r = __direct_map(vcpu, v, write, largepage, gfn, pfn,
 			 PT32E_ROOT_LEVEL);
@@ -1227,6 +1310,11 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 
 
 	return r;
+
+out_unlock:
+	spin_unlock(&vcpu->kvm->mmu_lock);
+	kvm_release_pfn_clean(pfn);
+	return 0;
 }
 
 
@@ -1345,6 +1433,7 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 	int r;
 	int largepage = 0;
 	gfn_t gfn = gpa >> PAGE_SHIFT;
+	unsigned long mmu_seq;
 
 	ASSERT(vcpu);
 	ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
@@ -1358,6 +1447,8 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 		gfn &= ~(KVM_PAGES_PER_HPAGE-1);
 		largepage = 1;
 	}
+	mmu_seq = vcpu->kvm->mmu_notifier_seq;
+	/* implicit mb(), we'll read before PT lock is unlocked */
 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
 	up_read(&current->mm->mmap_sem);
 	if (is_error_pfn(pfn)) {
@@ -1365,12 +1456,19 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 		return 1;
 	}
 	spin_lock(&vcpu->kvm->mmu_lock);
+	if (mmu_notifier_retry(vcpu, mmu_seq))
+		goto out_unlock;
 	kvm_mmu_free_some_pages(vcpu);
 	r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
 			 largepage, gfn, pfn, kvm_x86_ops->get_tdp_level());
 	spin_unlock(&vcpu->kvm->mmu_lock);
 
 	return r;
+
+out_unlock:
+	spin_unlock(&vcpu->kvm->mmu_lock);
+	kvm_release_pfn_clean(pfn);
+	return 0;
 }
 
 static void nonpaging_free(struct kvm_vcpu *vcpu)
@@ -1670,6 +1768,8 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 		gfn &= ~(KVM_PAGES_PER_HPAGE-1);
 		vcpu->arch.update_pte.largepage = 1;
 	}
+	vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq;
+	/* implicit mb(), we'll read before PT lock is unlocked */
 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
 	up_read(&current->mm->mmap_sem);
 

arch/x86/kvm/paging_tmpl.h

@@ -263,6 +263,8 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
 	pfn = vcpu->arch.update_pte.pfn;
 	if (is_error_pfn(pfn))
 		return;
+	if (mmu_notifier_retry(vcpu, vcpu->arch.update_pte.mmu_seq))
+		return;
 	kvm_get_pfn(pfn);
 	mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
 		     gpte & PT_DIRTY_MASK, NULL, largepage, gpte_to_gfn(gpte),
@@ -380,6 +382,7 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	int r;
 	pfn_t pfn;
 	int largepage = 0;
+	unsigned long mmu_seq;
 
 	pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
 	kvm_mmu_audit(vcpu, "pre page fault");
@@ -413,6 +416,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 			largepage = 1;
 		}
 	}
+	mmu_seq = vcpu->kvm->mmu_notifier_seq;
+	/* implicit mb(), we'll read before PT lock is unlocked */
 	pfn = gfn_to_pfn(vcpu->kvm, walker.gfn);
 	up_read(&current->mm->mmap_sem);
 
@@ -424,6 +429,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	}
 
 	spin_lock(&vcpu->kvm->mmu_lock);
+	if (mmu_notifier_retry(vcpu, mmu_seq))
+		goto out_unlock;
 	kvm_mmu_free_some_pages(vcpu);
 	shadow_pte = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
 				  largepage, &write_pt, pfn);
@@ -439,6 +446,11 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	spin_unlock(&vcpu->kvm->mmu_lock);
 
 	return write_pt;
+
+out_unlock:
+	spin_unlock(&vcpu->kvm->mmu_lock);
+	kvm_release_pfn_clean(pfn);
+	return 0;
 }
 
 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)

include/asm-x86/kvm_host.h

@@ -13,6 +13,7 @@
 
 #include <linux/types.h>
 #include <linux/mm.h>
+#include <linux/mmu_notifier.h>
 
 #include <linux/kvm.h>
 #include <linux/kvm_para.h>
@@ -251,6 +252,7 @@ struct kvm_vcpu_arch {
 		gfn_t gfn;	/* presumed gfn during guest pte update */
 		pfn_t pfn;	/* pfn corresponding to that gfn */
 		int largepage;
+		unsigned long mmu_seq;
 	} update_pte;
 
 	struct i387_fxsave_struct host_fx_image;
@@ -729,4 +731,8 @@ asmlinkage void kvm_handle_fault_on_reboot(void);
 	KVM_EX_ENTRY " 666b, 667b \n\t" \
 	".popsection"
 
+#define KVM_ARCH_WANT_MMU_NOTIFIER
+int kvm_unmap_hva(struct kvm *kvm, unsigned long hva);
+int kvm_age_hva(struct kvm *kvm, unsigned long hva);
+
 #endif

include/linux/kvm_host.h

@@ -121,6 +121,12 @@ struct kvm {
 	struct kvm_coalesced_mmio_dev *coalesced_mmio_dev;
 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
 #endif
+
+#ifdef KVM_ARCH_WANT_MMU_NOTIFIER
+	struct mmu_notifier mmu_notifier;
+	unsigned long mmu_notifier_seq;
+	long mmu_notifier_count;
+#endif
 };
 
 /* The guest did something we don't support. */
@@ -332,4 +338,22 @@ int kvm_trace_ioctl(unsigned int ioctl, unsigned long arg)
 #define kvm_trace_cleanup() ((void)0)
 #endif
 
+#ifdef KVM_ARCH_WANT_MMU_NOTIFIER
+static inline int mmu_notifier_retry(struct kvm_vcpu *vcpu, unsigned long mmu_seq)
+{
+	if (unlikely(vcpu->kvm->mmu_notifier_count))
+		return 1;
+	/*
+	 * Both reads happen under the mmu_lock and both values are
+	 * modified under mmu_lock, so there's no need of smb_rmb()
+	 * here in between, otherwise mmu_notifier_count should be
+	 * read before mmu_notifier_seq, see
+	 * mmu_notifier_invalidate_range_end write side.
+	 */
+	if (vcpu->kvm->mmu_notifier_seq != mmu_seq)
+		return 1;
+	return 0;
+}
+#endif
+
 #endif

virt/kvm/kvm_main.c

@@ -192,6 +192,123 @@ void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 }
 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 
+#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
+static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
+{
+	return container_of(mn, struct kvm, mmu_notifier);
+}
+
+static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
+					     struct mm_struct *mm,
+					     unsigned long address)
+{
+	struct kvm *kvm = mmu_notifier_to_kvm(mn);
+	int need_tlb_flush;
+
+	/*
+	 * When ->invalidate_page runs, the linux pte has been zapped
+	 * already but the page is still allocated until
+	 * ->invalidate_page returns. So if we increase the sequence
+	 * here the kvm page fault will notice if the spte can't be
+	 * established because the page is going to be freed. If
+	 * instead the kvm page fault establishes the spte before
+	 * ->invalidate_page runs, kvm_unmap_hva will release it
+	 * before returning.
+	 *
+	 * The sequence increase only need to be seen at spin_unlock
+	 * time, and not at spin_lock time.
+	 *
+	 * Increasing the sequence after the spin_unlock would be
+	 * unsafe because the kvm page fault could then establish the
+	 * pte after kvm_unmap_hva returned, without noticing the page
+	 * is going to be freed.
+	 */
+	spin_lock(&kvm->mmu_lock);
+	kvm->mmu_notifier_seq++;
+	need_tlb_flush = kvm_unmap_hva(kvm, address);
+	spin_unlock(&kvm->mmu_lock);
+
+	/* we've to flush the tlb before the pages can be freed */
+	if (need_tlb_flush)
+		kvm_flush_remote_tlbs(kvm);
+
+}
+
+static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
+						    struct mm_struct *mm,
+						    unsigned long start,
+						    unsigned long end)
+{
+	struct kvm *kvm = mmu_notifier_to_kvm(mn);
+	int need_tlb_flush = 0;
+
+	spin_lock(&kvm->mmu_lock);
+	/*
+	 * The count increase must become visible at unlock time as no
+	 * spte can be established without taking the mmu_lock and
+	 * count is also read inside the mmu_lock critical section.
+	 */
+	kvm->mmu_notifier_count++;
+	for (; start < end; start += PAGE_SIZE)
+		need_tlb_flush |= kvm_unmap_hva(kvm, start);
+	spin_unlock(&kvm->mmu_lock);
+
+	/* we've to flush the tlb before the pages can be freed */
+	if (need_tlb_flush)
+		kvm_flush_remote_tlbs(kvm);
+}
+
+static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
+						  struct mm_struct *mm,
+						  unsigned long start,
+						  unsigned long end)
+{
+	struct kvm *kvm = mmu_notifier_to_kvm(mn);
+
+	spin_lock(&kvm->mmu_lock);
+	/*
+	 * This sequence increase will notify the kvm page fault that
+	 * the page that is going to be mapped in the spte could have
+	 * been freed.
+	 */
+	kvm->mmu_notifier_seq++;
+	/*
+	 * The above sequence increase must be visible before the
+	 * below count decrease but both values are read by the kvm
+	 * page fault under mmu_lock spinlock so we don't need to add
+	 * a smb_wmb() here in between the two.
+	 */
+	kvm->mmu_notifier_count--;
+	spin_unlock(&kvm->mmu_lock);
+
+	BUG_ON(kvm->mmu_notifier_count < 0);
+}
+
+static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
+					      struct mm_struct *mm,
+					      unsigned long address)
+{
+	struct kvm *kvm = mmu_notifier_to_kvm(mn);
+	int young;
+
+	spin_lock(&kvm->mmu_lock);
+	young = kvm_age_hva(kvm, address);
+	spin_unlock(&kvm->mmu_lock);
+
+	if (young)
+		kvm_flush_remote_tlbs(kvm);
+
+	return young;
+}
+
+static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
+	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
+	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
+	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
+	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
+};
+#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
+
 static struct kvm *kvm_create_vm(void)
 {
 	struct kvm *kvm = kvm_arch_create_vm();
@@ -212,6 +329,21 @@ static struct kvm *kvm_create_vm(void)
 			(struct kvm_coalesced_mmio_ring *)page_address(page);
 #endif
 
+#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
+	{
+		int err;
+		kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
+		err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
+		if (err) {
+#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
+			put_page(page);
+#endif
+			kfree(kvm);
+			return ERR_PTR(err);
+		}
+	}
+#endif
+
 	kvm->mm = current->mm;
 	atomic_inc(&kvm->mm->mm_count);
 	spin_lock_init(&kvm->mmu_lock);
@@ -271,6 +403,9 @@ static void kvm_destroy_vm(struct kvm *kvm)
 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 	if (kvm->coalesced_mmio_ring != NULL)
 		free_page((unsigned long)kvm->coalesced_mmio_ring);
+#endif
+#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
+	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 #endif
 	kvm_arch_destroy_vm(kvm);
 	mmdrop(mm);