s390/mm: add support for 2GB hugepages

This adds support for 2GB hugetlbfs pages on s390.
Reviewed-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/hugetlb.h

@@ -41,7 +41,10 @@ static inline int prepare_hugepage_range(struct file *file,
 static inline void huge_pte_clear(struct mm_struct *mm, unsigned long addr,
 				  pte_t *ptep)
 {
-	pte_val(*ptep) = _SEGMENT_ENTRY_EMPTY;
+	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
+		pte_val(*ptep) = _REGION3_ENTRY_EMPTY;
+	else
+		pte_val(*ptep) = _SEGMENT_ENTRY_EMPTY;
 }
 
 static inline void huge_ptep_clear_flush(struct vm_area_struct *vma,

arch/s390/include/asm/page.h

@@ -21,6 +21,7 @@
 #define HPAGE_SIZE	(1UL << HPAGE_SHIFT)
 #define HPAGE_MASK	(~(HPAGE_SIZE - 1))
 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
+#define HUGE_MAX_HSTATE		2
 
 #define ARCH_HAS_SETCLEAR_HUGE_PTE
 #define ARCH_HAS_HUGE_PTE_TYPE

arch/s390/include/asm/pgtable.h

@@ -306,6 +306,9 @@ static inline int is_module_addr(void *addr)
 #define _REGION3_ENTRY_SOFT_DIRTY 0x0000 /* SW region soft dirty bit */
 #endif
 
+#define _REGION_ENTRY_BITS	 0xfffffffffffff227UL
+#define _REGION_ENTRY_BITS_LARGE 0xffffffff8000fe27UL
+
 /* Bits in the segment table entry */
 #define _SEGMENT_ENTRY_BITS	0xfffffffffffffe33UL
 #define _SEGMENT_ENTRY_BITS_LARGE 0xfffffffffff0ff33UL
@@ -573,7 +576,7 @@ static inline int pud_none(pud_t pud)
 {
 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
 		return 0;
-	return (pud_val(pud) & _REGION_ENTRY_INVALID) != 0UL;
+	return pud_val(pud) == _REGION3_ENTRY_EMPTY;
 }
 
 static inline int pud_large(pud_t pud)
@@ -593,17 +596,25 @@ static inline unsigned long pud_pfn(pud_t pud)
 	return (pud_val(pud) & origin_mask) >> PAGE_SHIFT;
 }
 
+static inline int pmd_large(pmd_t pmd)
+{
+	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
+}
+
+static inline int pmd_bad(pmd_t pmd)
+{
+	if (pmd_large(pmd))
+		return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS_LARGE) != 0;
+	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
+}
+
 static inline int pud_bad(pud_t pud)
 {
-	/*
-	 * With dynamic page table levels the pud can be a region table
-	 * entry or a segment table entry. Check for the bit that are
-	 * invalid for either table entry.
-	 */
-	unsigned long mask =
-		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
-		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
-	return (pud_val(pud) & mask) != 0;
+	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
+		return pmd_bad(__pmd(pud_val(pud)));
+	if (pud_large(pud))
+		return (pud_val(pud) & ~_REGION_ENTRY_BITS_LARGE) != 0;
+	return (pud_val(pud) & ~_REGION_ENTRY_BITS) != 0;
 }
 
 static inline int pmd_present(pmd_t pmd)
@@ -616,11 +627,6 @@ static inline int pmd_none(pmd_t pmd)
 	return pmd_val(pmd) == _SEGMENT_ENTRY_INVALID;
 }
 
-static inline int pmd_large(pmd_t pmd)
-{
-	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
-}
-
 static inline unsigned long pmd_pfn(pmd_t pmd)
 {
 	unsigned long origin_mask;
@@ -631,13 +637,6 @@ static inline unsigned long pmd_pfn(pmd_t pmd)
 	return (pmd_val(pmd) & origin_mask) >> PAGE_SHIFT;
 }
 
-static inline int pmd_bad(pmd_t pmd)
-{
-	if (pmd_large(pmd))
-		return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS_LARGE) != 0;
-	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
-}
-
 #define __HAVE_ARCH_PMD_WRITE
 static inline int pmd_write(pmd_t pmd)
 {
@@ -1081,6 +1080,7 @@ static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
 #define pte_page(x) pfn_to_page(pte_pfn(x))
 
 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd))
+#define pud_page(pud) pfn_to_page(pud_pfn(pud))
 
 /* Find an entry in the lowest level page table.. */
 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
@@ -1238,6 +1238,19 @@ static inline void __pmdp_idte(unsigned long address, pmd_t *pmdp)
 		: "cc" );
 }
 
+static inline void __pudp_idte(unsigned long address, pud_t *pudp)
+{
+	unsigned long r3o;
+
+	r3o = (unsigned long) pudp - pud_index(address) * sizeof(pud_t);
+	r3o |= _ASCE_TYPE_REGION3;
+	asm volatile(
+		"	.insn	rrf,0xb98e0000,%2,%3,0,0"
+		: "=m" (*pudp)
+		: "m" (*pudp), "a" (r3o), "a" ((address & PUD_MASK))
+		: "cc");
+}
+
 static inline void __pmdp_idte_local(unsigned long address, pmd_t *pmdp)
 {
 	unsigned long sto;
@@ -1250,8 +1263,22 @@ static inline void __pmdp_idte_local(unsigned long address, pmd_t *pmdp)
 		: "cc" );
 }
 
+static inline void __pudp_idte_local(unsigned long address, pud_t *pudp)
+{
+	unsigned long r3o;
+
+	r3o = (unsigned long) pudp - pud_index(address) * sizeof(pud_t);
+	r3o |= _ASCE_TYPE_REGION3;
+	asm volatile(
+		"	.insn	rrf,0xb98e0000,%2,%3,0,1"
+		: "=m" (*pudp)
+		: "m" (*pudp), "a" (r3o), "a" ((address & PUD_MASK))
+		: "cc");
+}
+
 pmd_t pmdp_xchg_direct(struct mm_struct *, unsigned long, pmd_t *, pmd_t);
 pmd_t pmdp_xchg_lazy(struct mm_struct *, unsigned long, pmd_t *, pmd_t);
+pud_t pudp_xchg_direct(struct mm_struct *, unsigned long, pud_t *, pud_t);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 

arch/s390/mm/gmap.c

@@ -430,6 +430,9 @@ int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
 	VM_BUG_ON(pgd_none(*pgd));
 	pud = pud_offset(pgd, vmaddr);
 	VM_BUG_ON(pud_none(*pud));
+	/* large puds cannot yet be handled */
+	if (pud_large(*pud))
+		return -EFAULT;
 	pmd = pmd_offset(pud, vmaddr);
 	VM_BUG_ON(pmd_none(*pmd));
 	/* large pmds cannot yet be handled */

arch/s390/mm/gup.c

@@ -128,6 +128,44 @@ static inline int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr,
 	return 1;
 }
 
+static int gup_huge_pud(pud_t *pudp, pud_t pud, unsigned long addr,
+		unsigned long end, int write, struct page **pages, int *nr)
+{
+	struct page *head, *page;
+	unsigned long mask;
+	int refs;
+
+	mask = (write ? _REGION_ENTRY_PROTECT : 0) | _REGION_ENTRY_INVALID;
+	if ((pud_val(pud) & mask) != 0)
+		return 0;
+	VM_BUG_ON(!pfn_valid(pud_pfn(pud)));
+
+	refs = 0;
+	head = pud_page(pud);
+	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
+	do {
+		VM_BUG_ON_PAGE(compound_head(page) != head, page);
+		pages[*nr] = page;
+		(*nr)++;
+		page++;
+		refs++;
+	} while (addr += PAGE_SIZE, addr != end);
+
+	if (!page_cache_add_speculative(head, refs)) {
+		*nr -= refs;
+		return 0;
+	}
+
+	if (unlikely(pud_val(pud) != pud_val(*pudp))) {
+		*nr -= refs;
+		while (refs--)
+			put_page(head);
+		return 0;
+	}
+
+	return 1;
+}
+
 static inline int gup_pud_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr,
 		unsigned long end, int write, struct page **pages, int *nr)
 {
@@ -144,7 +182,12 @@ static inline int gup_pud_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr,
 		next = pud_addr_end(addr, end);
 		if (pud_none(pud))
 			return 0;
-		if (!gup_pmd_range(pudp, pud, addr, next, write, pages, nr))
+		if (unlikely(pud_large(pud))) {
+			if (!gup_huge_pud(pudp, pud, addr, next, write, pages,
+					  nr))
+				return 0;
+		} else if (!gup_pmd_range(pudp, pud, addr, next, write, pages,
+					  nr))
 			return 0;
 	} while (pudp++, addr = next, addr != end);
 

arch/s390/mm/hugetlbpage.c

 /*
  *  IBM System z Huge TLB Page Support for Kernel.
  *
- *    Copyright IBM Corp. 2007
+ *    Copyright IBM Corp. 2007,2016
  *    Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
  */
 
+#define KMSG_COMPONENT "hugetlb"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
 #include <linux/mm.h>
 #include <linux/hugetlb.h>
 
-static inline pmd_t __pte_to_pmd(pte_t pte)
+static inline unsigned long __pte_to_rste(pte_t pte)
 {
-	pmd_t pmd;
+	unsigned long rste;
 
 	/*
-	 * Convert encoding		  pte bits	   pmd bits
+	 * Convert encoding		  pte bits	pmd / pud bits
 	 *				lIR.uswrdy.p	dy..R...I...wr
 	 * empty			010.000000.0 -> 00..0...1...00
 	 * prot-none, clean, old	111.000000.1 -> 00..1...1...00
@@ -33,25 +36,31 @@ static inline pmd_t __pte_to_pmd(pte_t pte)
 	 *	    u unused, l large
 	 */
 	if (pte_present(pte)) {
-		pmd_val(pmd) = pte_val(pte) & PAGE_MASK;
-		pmd_val(pmd) |= (pte_val(pte) & _PAGE_READ) >> 4;
-		pmd_val(pmd) |= (pte_val(pte) & _PAGE_WRITE) >> 4;
-		pmd_val(pmd) |=	(pte_val(pte) & _PAGE_INVALID) >> 5;
-		pmd_val(pmd) |= (pte_val(pte) & _PAGE_PROTECT);
-		pmd_val(pmd) |= (pte_val(pte) & _PAGE_DIRTY) << 10;
-		pmd_val(pmd) |= (pte_val(pte) & _PAGE_YOUNG) << 10;
-		pmd_val(pmd) |= (pte_val(pte) & _PAGE_SOFT_DIRTY) << 13;
+		rste = pte_val(pte) & PAGE_MASK;
+		rste |= (pte_val(pte) & _PAGE_READ) >> 4;
+		rste |= (pte_val(pte) & _PAGE_WRITE) >> 4;
+		rste |= (pte_val(pte) & _PAGE_INVALID) >> 5;
+		rste |= (pte_val(pte) & _PAGE_PROTECT);
+		rste |= (pte_val(pte) & _PAGE_DIRTY) << 10;
+		rste |= (pte_val(pte) & _PAGE_YOUNG) << 10;
+		rste |= (pte_val(pte) & _PAGE_SOFT_DIRTY) << 13;
 	} else
-		pmd_val(pmd) = _SEGMENT_ENTRY_INVALID;
-	return pmd;
+		rste = _SEGMENT_ENTRY_INVALID;
+	return rste;
 }
 
-static inline pte_t __pmd_to_pte(pmd_t pmd)
+static inline pte_t __rste_to_pte(unsigned long rste)
 {
+	int present;
 	pte_t pte;
 
+	if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
+		present = pud_present(__pud(rste));
+	else
+		present = pmd_present(__pmd(rste));
+
 	/*
-	 * Convert encoding		   pmd bits	    pte bits
+	 * Convert encoding		pmd / pud bits	    pte bits
 	 *				dy..R...I...wr	  lIR.uswrdy.p
 	 * empty			00..0...1...00 -> 010.000000.0
 	 * prot-none, clean, old	00..1...1...00 -> 111.000000.1
@@ -70,16 +79,16 @@ static inline pte_t __pmd_to_pte(pmd_t pmd)
 	 * SW-bits: p present, y young, d dirty, r read, w write, s special,
 	 *	    u unused, l large
 	 */
-	if (pmd_present(pmd)) {
-		pte_val(pte) = pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN_LARGE;
+	if (present) {
+		pte_val(pte) = rste & _SEGMENT_ENTRY_ORIGIN_LARGE;
 		pte_val(pte) |= _PAGE_LARGE | _PAGE_PRESENT;
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_READ) << 4;
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) << 4;
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_INVALID) << 5;
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT);
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) >> 10;
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) >> 10;
-		pte_val(pte) |= (pmd_val(pmd) & _SEGMENT_ENTRY_SOFT_DIRTY) >> 13;
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_READ) << 4;
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_WRITE) << 4;
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_INVALID) << 5;
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_PROTECT);
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_DIRTY) >> 10;
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_YOUNG) >> 10;
+		pte_val(pte) |= (rste & _SEGMENT_ENTRY_SOFT_DIRTY) >> 13;
 	} else
 		pte_val(pte) = _PAGE_INVALID;
 	return pte;
@@ -88,27 +97,33 @@ static inline pte_t __pmd_to_pte(pmd_t pmd)
 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
 		     pte_t *ptep, pte_t pte)
 {
-	pmd_t pmd = __pte_to_pmd(pte);
-
-	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
-	*(pmd_t *) ptep = pmd;
+	unsigned long rste = __pte_to_rste(pte);
+
+	/* Set correct table type for 2G hugepages */
+	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
+		rste |= _REGION_ENTRY_TYPE_R3 | _REGION3_ENTRY_LARGE;
+	else
+		rste |= _SEGMENT_ENTRY_LARGE;
+	pte_val(*ptep) = rste;
 }
 
 pte_t huge_ptep_get(pte_t *ptep)
 {
-	pmd_t pmd = *(pmd_t *) ptep;
-
-	return __pmd_to_pte(pmd);
+	return __rste_to_pte(pte_val(*ptep));
 }
 
 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
 			      unsigned long addr, pte_t *ptep)
 {
+	pte_t pte = huge_ptep_get(ptep);
 	pmd_t *pmdp = (pmd_t *) ptep;
-	pmd_t old;
+	pud_t *pudp = (pud_t *) ptep;
 
-	old = pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
-	return __pmd_to_pte(old);
+	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
+		pudp_xchg_direct(mm, addr, pudp, __pud(_REGION3_ENTRY_EMPTY));
+	else
+		pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
+	return pte;
 }
 
 pte_t *huge_pte_alloc(struct mm_struct *mm,
@@ -120,8 +135,12 @@ pte_t *huge_pte_alloc(struct mm_struct *mm,
 
 	pgdp = pgd_offset(mm, addr);
 	pudp = pud_alloc(mm, pgdp, addr);
-	if (pudp)
-		pmdp = pmd_alloc(mm, pudp, addr);
+	if (pudp) {
+		if (sz == PUD_SIZE)
+			return (pte_t *) pudp;
+		else if (sz == PMD_SIZE)
+			pmdp = pmd_alloc(mm, pudp, addr);
+	}
 	return (pte_t *) pmdp;
 }
 
@@ -134,8 +153,11 @@ pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
 	pgdp = pgd_offset(mm, addr);
 	if (pgd_present(*pgdp)) {
 		pudp = pud_offset(pgdp, addr);
-		if (pud_present(*pudp))
+		if (pud_present(*pudp)) {
+			if (pud_large(*pudp))
+				return (pte_t *) pudp;
 			pmdp = pmd_offset(pudp, addr);
+		}
 	}
 	return (pte_t *) pmdp;
 }
@@ -147,5 +169,34 @@ int pmd_huge(pmd_t pmd)
 
 int pud_huge(pud_t pud)
 {
-	return 0;
+	return pud_large(pud);
+}
+
+struct page *
+follow_huge_pud(struct mm_struct *mm, unsigned long address,
+		pud_t *pud, int flags)
+{
+	if (flags & FOLL_GET)
+		return NULL;
+
+	return pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
+}
+
+static __init int setup_hugepagesz(char *opt)
+{
+	unsigned long size;
+	char *string = opt;
+
+	size = memparse(opt, &opt);
+	if (MACHINE_HAS_EDAT1 && size == PMD_SIZE) {
+		hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
+	} else if (MACHINE_HAS_EDAT2 && size == PUD_SIZE) {
+		hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
+	} else {
+		pr_err("hugepagesz= specifies an unsupported page size %s\n",
+			string);
+		return 0;
+	}
+	return 1;
 }
+__setup("hugepagesz=", setup_hugepagesz);

arch/s390/mm/pgtable.c

@@ -352,6 +352,45 @@ pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
 }
 EXPORT_SYMBOL(pmdp_xchg_lazy);
 
+static inline pud_t pudp_flush_direct(struct mm_struct *mm,
+				      unsigned long addr, pud_t *pudp)
+{
+	pud_t old;
+
+	old = *pudp;
+	if (pud_val(old) & _REGION_ENTRY_INVALID)
+		return old;
+	if (!MACHINE_HAS_IDTE) {
+		/*
+		 * Invalid bit position is the same for pmd and pud, so we can
+		 * re-use _pmd_csp() here
+		 */
+		__pmdp_csp((pmd_t *) pudp);
+		return old;
+	}
+	atomic_inc(&mm->context.flush_count);
+	if (MACHINE_HAS_TLB_LC &&
+	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
+		__pudp_idte_local(addr, pudp);
+	else
+		__pudp_idte(addr, pudp);
+	atomic_dec(&mm->context.flush_count);
+	return old;
+}
+
+pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
+		       pud_t *pudp, pud_t new)
+{
+	pud_t old;
+
+	preempt_disable();
+	old = pudp_flush_direct(mm, addr, pudp);
+	*pudp = new;
+	preempt_enable();
+	return old;
+}
+EXPORT_SYMBOL(pudp_xchg_direct);
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 				pgtable_t pgtable)

mm/hugetlb.c

@@ -1022,7 +1022,9 @@ static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed)
 		((node = hstate_next_node_to_free(hs, mask)) || 1);	\
 		nr_nodes--)
 
-#if defined(CONFIG_X86_64) && ((defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA))
+#if (defined(CONFIG_X86_64) || defined(CONFIG_S390)) && \
+	((defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || \
+	defined(CONFIG_CMA))
 static void destroy_compound_gigantic_page(struct page *page,
 					unsigned int order)
 {