sparc64: Multi-page size support

Add support for using multiple hugepage sizes simultaneously
on mainline. Currently, support for 256M has been added which
can be used along with 8M pages.

Page tables are set like this (e.g. for 256M page):
    VA + (8M * x) -> PA + (8M * x) (sz bit = 256M) where x in [0, 31]

and TSB is set similarly:
    VA + (4M * x) -> PA + (4M * x) (sz bit = 256M) where x in [0, 63]

- Testing

Tested on Sonoma (which supports 256M pages) by running stream
benchmark instances in parallel: one instance uses 8M pages and
another uses 256M pages, consuming 48G each.

Boot params used:

default_hugepagesz=256M hugepagesz=256M hugepages=300 hugepagesz=8M
hugepages=10000

Signed-off-by: Nitin Gupta <nitin.m.gupta@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Author: Nitin Gupta

arch/sparc/include/asm/page_64.h

@@ -17,7 +17,7 @@
 
 #define HPAGE_SHIFT		23
 #define REAL_HPAGE_SHIFT	22
-
+#define HPAGE_256MB_SHIFT	28
 #define REAL_HPAGE_SIZE		(_AC(1,UL) << REAL_HPAGE_SHIFT)
 
 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
@@ -26,6 +26,7 @@
 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
 #define HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 #define REAL_HPAGE_PER_HPAGE	(_AC(1,UL) << (HPAGE_SHIFT - REAL_HPAGE_SHIFT))
+#define HUGE_MAX_HSTATE		2
 #endif
 
 #ifndef __ASSEMBLY__

arch/sparc/include/asm/pgtable_64.h

@@ -375,7 +375,10 @@ static inline pgprot_t pgprot_noncached(pgprot_t prot)
 #define pgprot_noncached pgprot_noncached
 
 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
-static inline unsigned long __pte_huge_mask(void)
+extern pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
+				struct page *page, int writable);
+#define arch_make_huge_pte arch_make_huge_pte
+static inline unsigned long __pte_default_huge_mask(void)
 {
 	unsigned long mask;
 
@@ -395,12 +398,14 @@ static inline unsigned long __pte_huge_mask(void)
 
 static inline pte_t pte_mkhuge(pte_t pte)
 {
-	return __pte(pte_val(pte) | _PAGE_PMD_HUGE | __pte_huge_mask());
+	return __pte(pte_val(pte) | __pte_default_huge_mask());
 }
 
-static inline bool is_hugetlb_pte(pte_t pte)
+static inline bool is_default_hugetlb_pte(pte_t pte)
 {
-	return !!(pte_val(pte) & __pte_huge_mask());
+	unsigned long mask = __pte_default_huge_mask();
+
+	return (pte_val(pte) & mask) == mask;
 }
 
 static inline bool is_hugetlb_pmd(pmd_t pmd)
@@ -875,10 +880,12 @@ static inline unsigned long pud_pfn(pud_t pud)
 
 /* Actual page table PTE updates.  */
 void tlb_batch_add(struct mm_struct *mm, unsigned long vaddr,
-		   pte_t *ptep, pte_t orig, int fullmm);
+		   pte_t *ptep, pte_t orig, int fullmm,
+		   unsigned int hugepage_shift);
 
 static void maybe_tlb_batch_add(struct mm_struct *mm, unsigned long vaddr,
-				pte_t *ptep, pte_t orig, int fullmm)
+				pte_t *ptep, pte_t orig, int fullmm,
+				unsigned int hugepage_shift)
 {
 	/* It is more efficient to let flush_tlb_kernel_range()
 	 * handle init_mm tlb flushes.
@@ -887,7 +894,7 @@ static void maybe_tlb_batch_add(struct mm_struct *mm, unsigned long vaddr,
 	 *             and SUN4V pte layout, so this inline test is fine.
 	 */
 	if (likely(mm != &init_mm) && pte_accessible(mm, orig))
-		tlb_batch_add(mm, vaddr, ptep, orig, fullmm);
+		tlb_batch_add(mm, vaddr, ptep, orig, fullmm, hugepage_shift);
 }
 
 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
@@ -906,7 +913,7 @@ static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
 	pte_t orig = *ptep;
 
 	*ptep = pte;
-	maybe_tlb_batch_add(mm, addr, ptep, orig, fullmm);
+	maybe_tlb_batch_add(mm, addr, ptep, orig, fullmm, PAGE_SHIFT);
 }
 
 #define set_pte_at(mm,addr,ptep,pte)	\

arch/sparc/include/asm/tlbflush_64.h

@@ -8,7 +8,7 @@
 #define TLB_BATCH_NR	192
 
 struct tlb_batch {
-	bool huge;
+	unsigned int hugepage_shift;
 	struct mm_struct *mm;
 	unsigned long tlb_nr;
 	unsigned long active;
@@ -17,7 +17,8 @@ struct tlb_batch {
 
 void flush_tsb_kernel_range(unsigned long start, unsigned long end);
 void flush_tsb_user(struct tlb_batch *tb);
-void flush_tsb_user_page(struct mm_struct *mm, unsigned long vaddr, bool huge);
+void flush_tsb_user_page(struct mm_struct *mm, unsigned long vaddr,
+			 unsigned int hugepage_shift);
 
 /* TLB flush operations. */
 

arch/sparc/kernel/tsb.S

@@ -117,26 +117,11 @@ tsb_miss_page_table_walk_sun4v_fastpath:
 	/* Valid PTE is now in %g5.  */
 
 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
-661:	sethi		%uhi(_PAGE_SZALL_4U), %g7
+	sethi		%uhi(_PAGE_PMD_HUGE), %g7
 	sllx		%g7, 32, %g7
-	.section	.sun4v_2insn_patch, "ax"
-	.word		661b
-	mov		_PAGE_SZALL_4V, %g7
-	nop
-	.previous
-
-	and		%g5, %g7, %g2
-
-661:	sethi		%uhi(_PAGE_SZHUGE_4U), %g7
-	sllx		%g7, 32, %g7
-	.section	.sun4v_2insn_patch, "ax"
-	.word		661b
-	mov		_PAGE_SZHUGE_4V, %g7
-	nop
-	.previous
 
-	cmp		%g2, %g7
-	bne,pt		%xcc, 60f
+	andcc		%g5, %g7, %g0
+	be,pt		%xcc, 60f
 	 nop
 
 	/* It is a huge page, use huge page TSB entry address we

arch/sparc/mm/hugetlbpage.c

@@ -28,6 +28,7 @@ static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *filp,
 							unsigned long pgoff,
 							unsigned long flags)
 {
+	struct hstate *h = hstate_file(filp);
 	unsigned long task_size = TASK_SIZE;
 	struct vm_unmapped_area_info info;
 
@@ -38,7 +39,7 @@ static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *filp,
 	info.length = len;
 	info.low_limit = TASK_UNMAPPED_BASE;
 	info.high_limit = min(task_size, VA_EXCLUDE_START);
-	info.align_mask = PAGE_MASK & ~HPAGE_MASK;
+	info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 	info.align_offset = 0;
 	addr = vm_unmapped_area(&info);
 
@@ -58,6 +59,7 @@ hugetlb_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 				  const unsigned long pgoff,
 				  const unsigned long flags)
 {
+	struct hstate *h = hstate_file(filp);
 	struct mm_struct *mm = current->mm;
 	unsigned long addr = addr0;
 	struct vm_unmapped_area_info info;
@@ -69,7 +71,7 @@ hugetlb_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 	info.length = len;
 	info.low_limit = PAGE_SIZE;
 	info.high_limit = mm->mmap_base;
-	info.align_mask = PAGE_MASK & ~HPAGE_MASK;
+	info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 	info.align_offset = 0;
 	addr = vm_unmapped_area(&info);
 
@@ -94,14 +96,15 @@ unsigned long
 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 		unsigned long len, unsigned long pgoff, unsigned long flags)
 {
+	struct hstate *h = hstate_file(file);
 	struct mm_struct *mm = current->mm;
 	struct vm_area_struct *vma;
 	unsigned long task_size = TASK_SIZE;
 
 	if (test_thread_flag(TIF_32BIT))
 		task_size = STACK_TOP32;
 
-	if (len & ~HPAGE_MASK)
+	if (len & ~huge_page_mask(h))
 		return -EINVAL;
 	if (len > task_size)
 		return -ENOMEM;
@@ -113,7 +116,7 @@ hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 	}
 
 	if (addr) {
-		addr = ALIGN(addr, HPAGE_SIZE);
+		addr = ALIGN(addr, huge_page_size(h));
 		vma = find_vma(mm, addr);
 		if (task_size - len >= addr &&
 		    (!vma || addr + len <= vma->vm_start))
@@ -127,6 +130,112 @@ hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 				pgoff, flags);
 }
 
+static pte_t sun4u_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
+{
+	return entry;
+}
+
+static pte_t sun4v_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
+{
+	unsigned long hugepage_size = _PAGE_SZ4MB_4V;
+
+	pte_val(entry) = pte_val(entry) & ~_PAGE_SZALL_4V;
+
+	switch (shift) {
+	case HPAGE_256MB_SHIFT:
+		hugepage_size = _PAGE_SZ256MB_4V;
+		pte_val(entry) |= _PAGE_PMD_HUGE;
+		break;
+	case HPAGE_SHIFT:
+		pte_val(entry) |= _PAGE_PMD_HUGE;
+		break;
+	default:
+		WARN_ONCE(1, "unsupported hugepage shift=%u\n", shift);
+	}
+
+	pte_val(entry) = pte_val(entry) | hugepage_size;
+	return entry;
+}
+
+static pte_t hugepage_shift_to_tte(pte_t entry, unsigned int shift)
+{
+	if (tlb_type == hypervisor)
+		return sun4v_hugepage_shift_to_tte(entry, shift);
+	else
+		return sun4u_hugepage_shift_to_tte(entry, shift);
+}
+
+pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
+			 struct page *page, int writeable)
+{
+	unsigned int shift = huge_page_shift(hstate_vma(vma));
+
+	return hugepage_shift_to_tte(entry, shift);
+}
+
+static unsigned int sun4v_huge_tte_to_shift(pte_t entry)
+{
+	unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4V;
+	unsigned int shift;
+
+	switch (tte_szbits) {
+	case _PAGE_SZ256MB_4V:
+		shift = HPAGE_256MB_SHIFT;
+		break;
+	case _PAGE_SZ4MB_4V:
+		shift = REAL_HPAGE_SHIFT;
+		break;
+	default:
+		shift = PAGE_SHIFT;
+		break;
+	}
+	return shift;
+}
+
+static unsigned int sun4u_huge_tte_to_shift(pte_t entry)
+{
+	unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4U;
+	unsigned int shift;
+
+	switch (tte_szbits) {
+	case _PAGE_SZ256MB_4U:
+		shift = HPAGE_256MB_SHIFT;
+		break;
+	case _PAGE_SZ4MB_4U:
+		shift = REAL_HPAGE_SHIFT;
+		break;
+	default:
+		shift = PAGE_SHIFT;
+		break;
+	}
+	return shift;
+}
+
+static unsigned int huge_tte_to_shift(pte_t entry)
+{
+	unsigned long shift;
+
+	if (tlb_type == hypervisor)
+		shift = sun4v_huge_tte_to_shift(entry);
+	else
+		shift = sun4u_huge_tte_to_shift(entry);
+
+	if (shift == PAGE_SHIFT)
+		WARN_ONCE(1, "tto_to_shift: invalid hugepage tte=0x%lx\n",
+			  pte_val(entry));
+
+	return shift;
+}
+
+static unsigned long huge_tte_to_size(pte_t pte)
+{
+	unsigned long size = 1UL << huge_tte_to_shift(pte);
+
+	if (size == REAL_HPAGE_SIZE)
+		size = HPAGE_SIZE;
+	return size;
+}
+
 pte_t *huge_pte_alloc(struct mm_struct *mm,
 			unsigned long addr, unsigned long sz)
 {
@@ -160,35 +269,54 @@ pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
 		     pte_t *ptep, pte_t entry)
 {
+	unsigned int i, nptes, hugepage_shift;
+	unsigned long size;
 	pte_t orig;
 
+	size = huge_tte_to_size(entry);
+	nptes = size >> PMD_SHIFT;
+
 	if (!pte_present(*ptep) && pte_present(entry))
-		mm->context.hugetlb_pte_count++;
+		mm->context.hugetlb_pte_count += nptes;
 
-	addr &= HPAGE_MASK;
+	addr &= ~(size - 1);
 	orig = *ptep;
-	*ptep = entry;
+	hugepage_shift = pte_none(orig) ? PAGE_SIZE : huge_tte_to_shift(orig);
 
-	/* Issue TLB flush at REAL_HPAGE_SIZE boundaries */
-	maybe_tlb_batch_add(mm, addr, ptep, orig, 0);
-	maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, orig, 0);
+	for (i = 0; i < nptes; i++)
+		ptep[i] = __pte(pte_val(entry) + (i << PMD_SHIFT));
+
+	maybe_tlb_batch_add(mm, addr, ptep, orig, 0, hugepage_shift);
+	/* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
+	if (size == HPAGE_SIZE)
+		maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, orig, 0,
+				    hugepage_shift);
 }
 
 pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
 			      pte_t *ptep)
 {
+	unsigned int i, nptes, hugepage_shift;
+	unsigned long size;
 	pte_t entry;
 
 	entry = *ptep;
+	size = huge_tte_to_size(entry);
+	nptes = size >> PMD_SHIFT;
+	hugepage_shift = pte_none(entry) ? PAGE_SIZE : huge_tte_to_shift(entry);
+
 	if (pte_present(entry))
-		mm->context.hugetlb_pte_count--;
+		mm->context.hugetlb_pte_count -= nptes;
 
-	addr &= HPAGE_MASK;
-	*ptep = __pte(0UL);
+	addr &= ~(size - 1);
+	for (i = 0; i < nptes; i++)
+		ptep[i] = __pte(0UL);
 
-	/* Issue TLB flush at REAL_HPAGE_SIZE boundaries */
-	maybe_tlb_batch_add(mm, addr, ptep, entry, 0);
-	maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, entry, 0);
+	maybe_tlb_batch_add(mm, addr, ptep, entry, 0, hugepage_shift);
+	/* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
+	if (size == HPAGE_SIZE)
+		maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, entry, 0,
+				    hugepage_shift);
 
 	return entry;
 }