KVM: PPC: Book3S HV: Recursively unmap all page table entries when unmapping

When partition scope mappings are unmapped with kvm_unmap_radix, the
pte is cleared, but the page table structure is left in place. If the
next page fault requests a different page table geometry (e.g., due to
THP promotion or split), kvmppc_create_pte is responsible for changing
the page tables.

When a page table entry is to be converted to a large pte, the page
table entry is cleared, the PWC flushed, then the page table it points
to freed. This will cause pte page tables to leak when a 1GB page is
to replace a pud entry points to a pmd table with pte tables under it:
The pmd table will be freed, but its pte tables will be missed.

Fix this by replacing the simple clear and free code with one that
walks down the page tables and frees children. Care must be taken to
clear the root entry being unmapped then flushing the PWC before
freeing any page tables, as explained in comments.

This requires PWC flush to logically become a flush-all-PWC (which it
already is in hardware, but the KVM API needs to be changed to avoid
confusion).

This code also checks that no unexpected pte entries exist in any page
table being freed, and unmaps those and emits a WARN. This is an
expensive operation for the pte page level, but partition scope
changes are rare, so it's unconditional for now to iron out bugs. It
can be put under a CONFIG option or removed after some time.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

1 files changed, 138 insertions, 54 deletions

diff --git a/arch/powerpc/kvm/book3s_64_mmu_radix.c b/arch/powerpc/kvm/book3s_64_mmu_radix.c
index 2c49b31..e514370 100644
--- a/arch/powerpc/kvm/book3s_64_mmu_radix.c
+++ b/arch/powerpc/kvm/book3s_64_mmu_radix.c
@@ -165,7 +165,7 @@ static void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
 	asm volatile("eieio ; tlbsync ; ptesync": : :"memory");
 }
 
-static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned long addr)
+static void kvmppc_radix_flush_pwc(struct kvm *kvm)
 {
 	unsigned long rb = 0x2 << PPC_BITLSHIFT(53); /* IS = 2 */
 
@@ -247,6 +247,139 @@ static void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte,
 	}
 }
 
+/*
+ * kvmppc_free_p?d are used to free existing page tables, and recursively
+ * descend and clear and free children.
+ * Callers are responsible for flushing the PWC.
+ *
+ * When page tables are being unmapped/freed as part of page fault path
+ * (full == false), ptes are not expected. There is code to unmap them
+ * and emit a warning if encountered, but there may already be data
+ * corruption due to the unexpected mappings.
+ */
+static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full)
+{
+	if (full) {
+		memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE);
+	} else {
+		pte_t *p = pte;
+		unsigned long it;
+
+		for (it = 0; it < PTRS_PER_PTE; ++it, ++p) {
+			if (pte_val(*p) == 0)
+				continue;
+			WARN_ON_ONCE(1);
+			kvmppc_unmap_pte(kvm, p,
+					 pte_pfn(*p) << PAGE_SHIFT,
+					 PAGE_SHIFT);
+		}
+	}
+
+	kvmppc_pte_free(pte);
+}
+
+static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full)
+{
+	unsigned long im;
+	pmd_t *p = pmd;
+
+	for (im = 0; im < PTRS_PER_PMD; ++im, ++p) {
+		if (!pmd_present(*p))
+			continue;
+		if (pmd_is_leaf(*p)) {
+			if (full) {
+				pmd_clear(p);
+			} else {
+				WARN_ON_ONCE(1);
+				kvmppc_unmap_pte(kvm, (pte_t *)p,
+					 pte_pfn(*(pte_t *)p) << PAGE_SHIFT,
+					 PMD_SHIFT);
+			}
+		} else {
+			pte_t *pte;
+
+			pte = pte_offset_map(p, 0);
+			kvmppc_unmap_free_pte(kvm, pte, full);
+			pmd_clear(p);
+		}
+	}
+	kvmppc_pmd_free(pmd);
+}
+
+static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud)
+{
+	unsigned long iu;
+	pud_t *p = pud;
+
+	for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) {
+		if (!pud_present(*p))
+			continue;
+		if (pud_huge(*p)) {
+			pud_clear(p);
+		} else {
+			pmd_t *pmd;
+
+			pmd = pmd_offset(p, 0);
+			kvmppc_unmap_free_pmd(kvm, pmd, true);
+			pud_clear(p);
+		}
+	}
+	pud_free(kvm->mm, pud);
+}
+
+void kvmppc_free_radix(struct kvm *kvm)
+{
+	unsigned long ig;
+	pgd_t *pgd;
+
+	if (!kvm->arch.pgtable)
+		return;
+	pgd = kvm->arch.pgtable;
+	for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
+		pud_t *pud;
+
+		if (!pgd_present(*pgd))
+			continue;
+		pud = pud_offset(pgd, 0);
+		kvmppc_unmap_free_pud(kvm, pud);
+		pgd_clear(pgd);
+	}
+	pgd_free(kvm->mm, kvm->arch.pgtable);
+	kvm->arch.pgtable = NULL;
+}
+
+static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd,
+					      unsigned long gpa)
+{
+	pte_t *pte = pte_offset_kernel(pmd, 0);
+
+	/*
+	 * Clearing the pmd entry then flushing the PWC ensures that the pte
+	 * page no longer be cached by the MMU, so can be freed without
+	 * flushing the PWC again.
+	 */
+	pmd_clear(pmd);
+	kvmppc_radix_flush_pwc(kvm);
+
+	kvmppc_unmap_free_pte(kvm, pte, false);
+}
+
+static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud,
+					unsigned long gpa)
+{
+	pmd_t *pmd = pmd_offset(pud, 0);
+
+	/*
+	 * Clearing the pud entry then flushing the PWC ensures that the pmd
+	 * page and any children pte pages will no longer be cached by the MMU,
+	 * so can be freed without flushing the PWC again.
+	 */
+	pud_clear(pud);
+	kvmppc_radix_flush_pwc(kvm);
+
+	kvmppc_unmap_free_pmd(kvm, pmd, false);
+}
+
 static int kvmppc_create_pte(struct kvm *kvm, pte_t pte, unsigned long gpa,
 			     unsigned int level, unsigned long mmu_seq)
 {
@@ -312,11 +445,9 @@ static int kvmppc_create_pte(struct kvm *kvm, pte_t pte, unsigned long gpa,
 			/*
 			 * There's a page table page here, but we wanted to
 			 * install a large page, so remove and free the page
-			 * table page.  new_pmd will be NULL since level == 2.
+			 * table page.
 			 */
-			new_pmd = pmd_offset(pud, 0);
-			pud_clear(pud);
-			kvmppc_radix_flush_pwc(kvm, gpa);
+			kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa);
 		}
 		kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte);
 		ret = 0;
@@ -353,11 +484,9 @@ static int kvmppc_create_pte(struct kvm *kvm, pte_t pte, unsigned long gpa,
 			/*
 			 * There's a page table page here, but we wanted to
 			 * install a large page, so remove and free the page
-			 * table page.  new_ptep will be NULL since level == 1.
+			 * table page.
 			 */
-			new_ptep = pte_offset_kernel(pmd, 0);
-			pmd_clear(pmd);
-			kvmppc_radix_flush_pwc(kvm, gpa);
+			kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa);
 		}
 		kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte);
 		ret = 0;
@@ -734,51 +863,6 @@ int kvmppc_init_vm_radix(struct kvm *kvm)
 	return 0;
 }
 
-void kvmppc_free_radix(struct kvm *kvm)
-{
-	unsigned long ig, iu, im;
-	pte_t *pte;
-	pmd_t *pmd;
-	pud_t *pud;
-	pgd_t *pgd;
-
-	if (!kvm->arch.pgtable)
-		return;
-	pgd = kvm->arch.pgtable;
-	for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
-		if (!pgd_present(*pgd))
-			continue;
-		pud = pud_offset(pgd, 0);
-		for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++pud) {
-			if (!pud_present(*pud))
-				continue;
-			if (pud_huge(*pud)) {
-				pud_clear(pud);
-				continue;
-			}
-			pmd = pmd_offset(pud, 0);
-			for (im = 0; im < PTRS_PER_PMD; ++im, ++pmd) {
-				if (pmd_is_leaf(*pmd)) {
-					pmd_clear(pmd);
-					continue;
-				}
-				if (!pmd_present(*pmd))
-					continue;
-				pte = pte_offset_map(pmd, 0);
-				memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE);
-				kvmppc_pte_free(pte);
-				pmd_clear(pmd);
-			}
-			kvmppc_pmd_free(pmd_offset(pud, 0));
-			pud_clear(pud);
-		}
-		pud_free(kvm->mm, pud_offset(pgd, 0));
-		pgd_clear(pgd);
-	}
-	pgd_free(kvm->mm, kvm->arch.pgtable);
-	kvm->arch.pgtable = NULL;
-}
-
 static void pte_ctor(void *addr)
 {
 	memset(addr, 0, RADIX_PTE_TABLE_SIZE);