[PATCH v10 8/8] KVM: arm64: Add support for creating PUD hugepages at stage 2

[Suzuki K Poulose]

From: Punit Agrawal <punit.agrawal@xxxxxxx>

KVM only supports PMD hugepages at stage 2. Now that the various page
handling routines are updated, extend the stage 2 fault handling to
map in PUD hugepages.

Addition of PUD hugepage support enables additional page sizes (e.g.,
1G with 4K granule) which can be useful on cores that support mapping
larger block sizes in the TLB entries.

Signed-off-by: Punit Agrawal <punit.agrawal@xxxxxxx>
Cc: Christoffer Dall <christoffer.dall@xxxxxxx>
Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
Cc: Russell King <linux@xxxxxxxxxxxxxxx>
Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
Cc: Will Deacon <will.deacon@xxxxxxx>
[ Replace BUG() => WARN_ON(1) for arm32 PUD helpers ]
Signed-off-by: Suzuki Poulose <suzuki.poulose@xxxxxxx>
---
 arch/arm/include/asm/kvm_mmu.h         |  20 +++++++
 arch/arm/include/asm/stage2_pgtable.h  |   5 ++
 arch/arm64/include/asm/kvm_mmu.h       |  16 +++++
 arch/arm64/include/asm/pgtable-hwdef.h |   2 +
 arch/arm64/include/asm/pgtable.h       |   2 +
 virt/kvm/arm/mmu.c                     | 104 +++++++++++++++++++++++++++++++--
 6 files changed, 143 insertions(+), 6 deletions(-)

diff --git a/arch/arm/include/asm/kvm_mmu.h b/arch/arm/include/asm/kvm_mmu.h
index 4b4b214..3a875fc 100644
--- a/arch/arm/include/asm/kvm_mmu.h
+++ b/arch/arm/include/asm/kvm_mmu.h
@@ -84,11 +84,14 @@ void kvm_clear_hyp_idmap(void);
 
 #define kvm_pfn_pte(pfn, prot)	pfn_pte(pfn, prot)
 #define kvm_pfn_pmd(pfn, prot)	pfn_pmd(pfn, prot)
+#define kvm_pfn_pud(pfn, prot)	(__pud(0))
 
 #define kvm_pud_pfn(pud)	({ WARN_ON(1); 0; })
 
 
 #define kvm_pmd_mkhuge(pmd)	pmd_mkhuge(pmd)
+/* No support for pud hugepages */
+#define kvm_pud_mkhuge(pud)	( {WARN_ON(1); pud; })
 
 /*
  * The following kvm_*pud*() functions are provided strictly to allow
@@ -105,6 +108,23 @@ static inline bool kvm_s2pud_readonly(pud_t *pud)
 	return false;
 }
 
+static inline void kvm_set_pud(pud_t *pud, pud_t new_pud)
+{
+	WARN_ON(1);
+}
+
+static inline pud_t kvm_s2pud_mkwrite(pud_t pud)
+{
+	WARN_ON(1);
+	return pud;
+}
+
+static inline pud_t kvm_s2pud_mkexec(pud_t pud)
+{
+	WARN_ON(1);
+	return pud;
+}
+
 static inline bool kvm_s2pud_exec(pud_t *pud)
 {
 	WARN_ON(1);
diff --git a/arch/arm/include/asm/stage2_pgtable.h b/arch/arm/include/asm/stage2_pgtable.h
index f6a7ea8..f901716 100644
--- a/arch/arm/include/asm/stage2_pgtable.h
+++ b/arch/arm/include/asm/stage2_pgtable.h
@@ -68,4 +68,9 @@ stage2_pmd_addr_end(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
 #define stage2_pmd_table_empty(kvm, pmdp)	kvm_page_empty(pmdp)
 #define stage2_pud_table_empty(kvm, pudp)	false
 
+static inline bool kvm_stage2_has_pud(struct kvm *kvm)
+{
+	return false;
+}
+
 #endif	/* __ARM_S2_PGTABLE_H_ */
diff --git a/arch/arm64/include/asm/kvm_mmu.h b/arch/arm64/include/asm/kvm_mmu.h
index 9f941f7..8af4b1b 100644
--- a/arch/arm64/include/asm/kvm_mmu.h
+++ b/arch/arm64/include/asm/kvm_mmu.h
@@ -184,12 +184,16 @@ void kvm_clear_hyp_idmap(void);
 #define kvm_mk_pgd(pudp)					\
 	__pgd(__phys_to_pgd_val(__pa(pudp)) | PUD_TYPE_TABLE)
 
+#define kvm_set_pud(pudp, pud)		set_pud(pudp, pud)
+
 #define kvm_pfn_pte(pfn, prot)		pfn_pte(pfn, prot)
 #define kvm_pfn_pmd(pfn, prot)		pfn_pmd(pfn, prot)
+#define kvm_pfn_pud(pfn, prot)		pfn_pud(pfn, prot)
 
 #define kvm_pud_pfn(pud)		pud_pfn(pud)
 
 #define kvm_pmd_mkhuge(pmd)		pmd_mkhuge(pmd)
+#define kvm_pud_mkhuge(pud)		pud_mkhuge(pud)
 
 static inline pte_t kvm_s2pte_mkwrite(pte_t pte)
 {
@@ -203,6 +207,12 @@ static inline pmd_t kvm_s2pmd_mkwrite(pmd_t pmd)
 	return pmd;
 }
 
+static inline pud_t kvm_s2pud_mkwrite(pud_t pud)
+{
+	pud_val(pud) |= PUD_S2_RDWR;
+	return pud;
+}
+
 static inline pte_t kvm_s2pte_mkexec(pte_t pte)
 {
 	pte_val(pte) &= ~PTE_S2_XN;
@@ -215,6 +225,12 @@ static inline pmd_t kvm_s2pmd_mkexec(pmd_t pmd)
 	return pmd;
 }
 
+static inline pud_t kvm_s2pud_mkexec(pud_t pud)
+{
+	pud_val(pud) &= ~PUD_S2_XN;
+	return pud;
+}
+
 static inline void kvm_set_s2pte_readonly(pte_t *ptep)
 {
 	pteval_t old_pteval, pteval;
diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
index 336e24c..6f1c187 100644
--- a/arch/arm64/include/asm/pgtable-hwdef.h
+++ b/arch/arm64/include/asm/pgtable-hwdef.h
@@ -193,6 +193,8 @@
 #define PMD_S2_RDWR		(_AT(pmdval_t, 3) << 6)   /* HAP[2:1] */
 #define PMD_S2_XN		(_AT(pmdval_t, 2) << 53)  /* XN[1:0] */
 
+#define PUD_S2_RDONLY		(_AT(pudval_t, 1) << 6)   /* HAP[2:1] */
+#define PUD_S2_RDWR		(_AT(pudval_t, 3) << 6)   /* HAP[2:1] */
 #define PUD_S2_XN		(_AT(pudval_t, 2) << 53)  /* XN[1:0] */
 
 /*
diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
index bb0f3f1..5761286 100644
--- a/arch/arm64/include/asm/pgtable.h
+++ b/arch/arm64/include/asm/pgtable.h
@@ -390,6 +390,8 @@ static inline int pmd_protnone(pmd_t pmd)
 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
 #define pud_write(pud)		pte_write(pud_pte(pud))
 
+#define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
+
 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
index 3893ea6..2dcff38 100644
--- a/virt/kvm/arm/mmu.c
+++ b/virt/kvm/arm/mmu.c
@@ -115,6 +115,25 @@ static void stage2_dissolve_pmd(struct kvm *kvm, phys_addr_t addr, pmd_t *pmd)
 	put_page(virt_to_page(pmd));
 }
 
+/**
+ * stage2_dissolve_pud() - clear and flush huge PUD entry
+ * @kvm:	pointer to kvm structure.
+ * @addr:	IPA
+ * @pud:	pud pointer for IPA
+ *
+ * Function clears a PUD entry, flushes addr 1st and 2nd stage TLBs. Marks all
+ * pages in the range dirty.
+ */
+static void stage2_dissolve_pud(struct kvm *kvm, phys_addr_t addr, pud_t *pudp)
+{
+	if (!stage2_pud_huge(kvm, *pudp))
+		return;
+
+	stage2_pud_clear(kvm, pudp);
+	kvm_tlb_flush_vmid_ipa(kvm, addr);
+	put_page(virt_to_page(pudp));
+}
+
 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
 				  int min, int max)
 {
@@ -1022,7 +1041,7 @@ static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache
 	pmd_t *pmd;
 
 	pud = stage2_get_pud(kvm, cache, addr);
-	if (!pud)
+	if (!pud || stage2_pud_huge(kvm, *pud))
 		return NULL;
 
 	if (stage2_pud_none(kvm, *pud)) {
@@ -1083,6 +1102,36 @@ static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache
 	return 0;
 }
 
+static int stage2_set_pud_huge(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
+			       phys_addr_t addr, const pud_t *new_pudp)
+{
+	pud_t *pudp, old_pud;
+
+	pudp = stage2_get_pud(kvm, cache, addr);
+	VM_BUG_ON(!pudp);
+
+	old_pud = *pudp;
+
+	/*
+	 * A large number of vcpus faulting on the same stage 2 entry,
+	 * can lead to a refault due to the
+	 * stage2_pud_clear()/tlb_flush(). Skip updating the page
+	 * tables if there is no change.
+	 */
+	if (pud_val(old_pud) == pud_val(*new_pudp))
+		return 0;
+
+	if (stage2_pud_present(kvm, old_pud)) {
+		stage2_pud_clear(kvm, pudp);
+		kvm_tlb_flush_vmid_ipa(kvm, addr);
+	} else {
+		get_page(virt_to_page(pudp));
+	}
+
+	kvm_set_pud(pudp, *new_pudp);
+	return 0;
+}
+
 /*
  * stage2_get_leaf_entry - walk the stage2 VM page tables and return
  * true if a valid and present leaf-entry is found. A pointer to the
@@ -1149,6 +1198,7 @@ static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
 			  phys_addr_t addr, const pte_t *new_pte,
 			  unsigned long flags)
 {
+	pud_t *pud;
 	pmd_t *pmd;
 	pte_t *pte, old_pte;
 	bool iomap = flags & KVM_S2PTE_FLAG_IS_IOMAP;
@@ -1157,7 +1207,31 @@ static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
 	VM_BUG_ON(logging_active && !cache);
 
 	/* Create stage-2 page table mapping - Levels 0 and 1 */
-	pmd = stage2_get_pmd(kvm, cache, addr);
+	pud = stage2_get_pud(kvm, cache, addr);
+	if (!pud) {
+		/*
+		 * Ignore calls from kvm_set_spte_hva for unallocated
+		 * address ranges.
+		 */
+		return 0;
+	}
+
+	/*
+	 * While dirty page logging - dissolve huge PUD, then continue
+	 * on to allocate page.
+	 */
+	if (logging_active)
+		stage2_dissolve_pud(kvm, addr, pud);
+
+	if (stage2_pud_none(kvm, *pud)) {
+		if (!cache)
+			return 0; /* ignore calls from kvm_set_spte_hva */
+		pmd = mmu_memory_cache_alloc(cache);
+		stage2_pud_populate(kvm, pud, pmd);
+		get_page(virt_to_page(pud));
+	}
+
+	pmd = stage2_pmd_offset(kvm, pud, addr);
 	if (!pmd) {
 		/*
 		 * Ignore calls from kvm_set_spte_hva for unallocated
@@ -1557,12 +1631,19 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 	}
 
 	vma_pagesize = vma_kernel_pagesize(vma);
-	if (vma_pagesize == PMD_SIZE && !logging_active) {
-		gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
+	/*
+	 * PUD level may not exist for a VM but PMD is guaranteed to
+	 * exist.
+	 */
+	if ((vma_pagesize == PMD_SIZE ||
+	     (vma_pagesize == PUD_SIZE && kvm_stage2_has_pud(kvm))) &&
+	    !logging_active) {
+		gfn = (fault_ipa & huge_page_mask(hstate_vma(vma))) >> PAGE_SHIFT;
 	} else {
 		/*
 		 * Fallback to PTE if it's not one of the Stage 2
-		 * supported hugepage sizes
+		 * supported hugepage sizes or the corresponding level
+		 * doesn't exist
 		 */
 		vma_pagesize = PAGE_SIZE;
 
@@ -1661,7 +1742,18 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 	needs_exec = exec_fault ||
 		(fault_status == FSC_PERM && stage2_is_exec(kvm, fault_ipa));
 
-	if (vma_pagesize == PMD_SIZE) {
+	if (vma_pagesize == PUD_SIZE) {
+		pud_t new_pud = kvm_pfn_pud(pfn, mem_type);
+
+		new_pud = kvm_pud_mkhuge(new_pud);
+		if (writable)
+			new_pud = kvm_s2pud_mkwrite(new_pud);
+
+		if (needs_exec)
+			new_pud = kvm_s2pud_mkexec(new_pud);
+
+		ret = stage2_set_pud_huge(kvm, memcache, fault_ipa, &new_pud);
+	} else if (vma_pagesize == PMD_SIZE) {
 		pmd_t new_pmd = kvm_pfn_pmd(pfn, mem_type);
 
 		new_pmd = kvm_pmd_mkhuge(new_pmd);
-- 
2.7.4