arch/tile: support multiple huge page sizes dynamically

This change adds support for a new "super" bit in the PTE, using the new
arch_make_huge_pte() method.  The Tilera hypervisor sees the bit set at a
given level of the page table and gangs together 4, 16, or 64 consecutive
pages from that level of the hierarchy to create a larger TLB entry.

One extra "super" page size can be specified at each of the three levels
of the page table hierarchy on tilegx, using the "hugepagesz" argument
on the boot command line.  A new hypervisor API is added to allow Linux
to tell the hypervisor how many PTEs to gang together at each level of
the page table.

To allow pre-allocating huge pages larger than the buddy allocator can
handle, this change modifies the Tilera bootmem support to put all of
memory on tilegx platforms into bootmem.

As part of this change I eliminate the vestigial CONFIG_HIGHPTE support,
which never worked anyway, and eliminate the hv_page_size() API in favor
of the standard vma_kernel_pagesize() API.

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>

5 files changed, 226 insertions, 79 deletions

diff --git a/arch/tile/mm/fault.c b/arch/tile/mm/fault.c
index 22e58f51ed2..54f18fc25ed 100644
--- a/arch/tile/mm/fault.c
+++ b/arch/tile/mm/fault.c
@@ -187,7 +187,7 @@ static pgd_t *get_current_pgd(void)
 	HV_Context ctx = hv_inquire_context();
 	unsigned long pgd_pfn = ctx.page_table >> PAGE_SHIFT;
 	struct page *pgd_page = pfn_to_page(pgd_pfn);
-	BUG_ON(PageHighMem(pgd_page));   /* oops, HIGHPTE? */
+	BUG_ON(PageHighMem(pgd_page));
 	return (pgd_t *) __va(ctx.page_table);
 }
 
diff --git a/arch/tile/mm/homecache.c b/arch/tile/mm/homecache.c
index 499f73770b0..dbcbdf7b8aa 100644
--- a/arch/tile/mm/homecache.c
+++ b/arch/tile/mm/homecache.c
@@ -30,6 +30,7 @@
 #include <linux/cache.h>
 #include <linux/smp.h>
 #include <linux/module.h>
+#include <linux/hugetlb.h>
 
 #include <asm/page.h>
 #include <asm/sections.h>
diff --git a/arch/tile/mm/hugetlbpage.c b/arch/tile/mm/hugetlbpage.c
index 42cfcba4e1e..812e2d03797 100644
--- a/arch/tile/mm/hugetlbpage.c
+++ b/arch/tile/mm/hugetlbpage.c
@@ -27,85 +27,161 @@
 #include <linux/mman.h>
 #include <asm/tlb.h>
 #include <asm/tlbflush.h>
+#include <asm/setup.h>
+
+#ifdef CONFIG_HUGETLB_SUPER_PAGES
+
+/*
+ * Provide an additional huge page size (in addition to the regular default
+ * huge page size) if no "hugepagesz" arguments are specified.
+ * Note that it must be smaller than the default huge page size so
+ * that it's possible to allocate them on demand from the buddy allocator.
+ * You can change this to 64K (on a 16K build), 256K, 1M, or 4M,
+ * or not define it at all.
+ */
+#define ADDITIONAL_HUGE_SIZE (1024 * 1024UL)
+
+/* "Extra" page-size multipliers, one per level of the page table. */
+int huge_shift[HUGE_SHIFT_ENTRIES] = {
+#ifdef ADDITIONAL_HUGE_SIZE
+#define ADDITIONAL_HUGE_SHIFT __builtin_ctzl(ADDITIONAL_HUGE_SIZE / PAGE_SIZE)
+	[HUGE_SHIFT_PAGE] = ADDITIONAL_HUGE_SHIFT
+#endif
+};
+
+/*
+ * This routine is a hybrid of pte_alloc_map() and pte_alloc_kernel().
+ * It assumes that L2 PTEs are never in HIGHMEM (we don't support that).
+ * It locks the user pagetable, and bumps up the mm->nr_ptes field,
+ * but otherwise allocate the page table using the kernel versions.
+ */
+static pte_t *pte_alloc_hugetlb(struct mm_struct *mm, pmd_t *pmd,
+				unsigned long address)
+{
+	pte_t *new;
+
+	if (pmd_none(*pmd)) {
+		new = pte_alloc_one_kernel(mm, address);
+		if (!new)
+			return NULL;
+
+		smp_wmb(); /* See comment in __pte_alloc */
+
+		spin_lock(&mm->page_table_lock);
+		if (likely(pmd_none(*pmd))) {  /* Has another populated it ? */
+			mm->nr_ptes++;
+			pmd_populate_kernel(mm, pmd, new);
+			new = NULL;
+		} else
+			VM_BUG_ON(pmd_trans_splitting(*pmd));
+		spin_unlock(&mm->page_table_lock);
+		if (new)
+			pte_free_kernel(mm, new);
+	}
+
+	return pte_offset_kernel(pmd, address);
+}
+#endif
 
 pte_t *huge_pte_alloc(struct mm_struct *mm,
 		      unsigned long addr, unsigned long sz)
 {
 	pgd_t *pgd;
 	pud_t *pud;
-	pte_t *pte = NULL;
 
-	/* We do not yet support multiple huge page sizes. */
-	BUG_ON(sz != PMD_SIZE);
+	addr &= -sz;   /* Mask off any low bits in the address. */
 
 	pgd = pgd_offset(mm, addr);
 	pud = pud_alloc(mm, pgd, addr);
-	if (pud)
-		pte = (pte_t *) pmd_alloc(mm, pud, addr);
-	BUG_ON(pte && !pte_none(*pte) && !pte_huge(*pte));
 
-	return pte;
+#ifdef CONFIG_HUGETLB_SUPER_PAGES
+	if (sz >= PGDIR_SIZE) {
+		BUG_ON(sz != PGDIR_SIZE &&
+		       sz != PGDIR_SIZE << huge_shift[HUGE_SHIFT_PGDIR]);
+		return (pte_t *)pud;
+	} else {
+		pmd_t *pmd = pmd_alloc(mm, pud, addr);
+		if (sz >= PMD_SIZE) {
+			BUG_ON(sz != PMD_SIZE &&
+			       sz != (PMD_SIZE << huge_shift[HUGE_SHIFT_PMD]));
+			return (pte_t *)pmd;
+		}
+		else {
+			if (sz != PAGE_SIZE << huge_shift[HUGE_SHIFT_PAGE])
+				panic("Unexpected page size %#lx\n", sz);
+			return pte_alloc_hugetlb(mm, pmd, addr);
+		}
+	}
+#else
+	BUG_ON(sz != PMD_SIZE);
+	return (pte_t *) pmd_alloc(mm, pud, addr);
+#endif
 }
 
-pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
+static pte_t *get_pte(pte_t *base, int index, int level)
 {
-	pgd_t *pgd;
-	pud_t *pud;
-	pmd_t *pmd = NULL;
-
-	pgd = pgd_offset(mm, addr);
-	if (pgd_present(*pgd)) {
-		pud = pud_offset(pgd, addr);
-		if (pud_present(*pud))
-			pmd = pmd_offset(pud, addr);
+	pte_t *ptep = base + index;
+#ifdef CONFIG_HUGETLB_SUPER_PAGES
+	if (!pte_present(*ptep) && huge_shift[level] != 0) {
+		unsigned long mask = -1UL << huge_shift[level];
+		pte_t *super_ptep = base + (index & mask);
+		pte_t pte = *super_ptep;
+		if (pte_present(pte) && pte_super(pte))
+			ptep = super_ptep;
 	}
-	return (pte_t *) pmd;
+#endif
+	return ptep;
 }
 
-#ifdef HUGETLB_TEST
-struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
-			      int write)
+pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
 {
-	unsigned long start = address;
-	int length = 1;
-	int nr;
-	struct page *page;
-	struct vm_area_struct *vma;
-
-	vma = find_vma(mm, addr);
-	if (!vma || !is_vm_hugetlb_page(vma))
-		return ERR_PTR(-EINVAL);
-
-	pte = huge_pte_offset(mm, address);
+	pgd_t *pgd;
+	pud_t *pud;
+	pmd_t *pmd;
+#ifdef CONFIG_HUGETLB_SUPER_PAGES
+	pte_t *pte;
+#endif
 
-	/* hugetlb should be locked, and hence, prefaulted */
-	WARN_ON(!pte || pte_none(*pte));
+	/* Get the top-level page table entry. */
+	pgd = (pgd_t *)get_pte((pte_t *)mm->pgd, pgd_index(addr), 0);
+	if (!pgd_present(*pgd))
+		return NULL;
 
-	page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
+	/* We don't have four levels. */
+	pud = pud_offset(pgd, addr);
+#ifndef __PAGETABLE_PUD_FOLDED
+# error support fourth page table level
+#endif
 
-	WARN_ON(!PageHead(page));
+	/* Check for an L0 huge PTE, if we have three levels. */
+#ifndef __PAGETABLE_PMD_FOLDED
+	if (pud_huge(*pud))
+		return (pte_t *)pud;
 
-	return page;
-}
-
-int pmd_huge(pmd_t pmd)
-{
-	return 0;
-}
+	pmd = (pmd_t *)get_pte((pte_t *)pud_page_vaddr(*pud),
+			       pmd_index(addr), 1);
+	if (!pmd_present(*pmd))
+		return NULL;
+#else
+	pmd = pmd_offset(pud, addr);
+#endif
 
-int pud_huge(pud_t pud)
-{
-	return 0;
-}
+	/* Check for an L1 huge PTE. */
+	if (pmd_huge(*pmd))
+		return (pte_t *)pmd;
+
+#ifdef CONFIG_HUGETLB_SUPER_PAGES
+	/* Check for an L2 huge PTE. */
+	pte = get_pte((pte_t *)pmd_page_vaddr(*pmd), pte_index(addr), 2);
+	if (!pte_present(*pte))
+		return NULL;
+	if (pte_super(*pte))
+		return pte;
+#endif
 
-struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
-			     pmd_t *pmd, int write)
-{
 	return NULL;
 }
 
-#else
-
 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
 			      int write)
 {
@@ -149,8 +225,6 @@ int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
 	return 0;
 }
 
-#endif
-
 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *file,
 		unsigned long addr, unsigned long len,
@@ -322,21 +396,102 @@ unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 		return hugetlb_get_unmapped_area_topdown(file, addr, len,
 				pgoff, flags);
 }
+#endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
 
-static __init int setup_hugepagesz(char *opt)
+#ifdef CONFIG_HUGETLB_SUPER_PAGES
+static __init int __setup_hugepagesz(unsigned long ps)
 {
-	unsigned long ps = memparse(opt, &opt);
-	if (ps == PMD_SIZE) {
-		hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
-	} else if (ps == PUD_SIZE) {
-		hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
+	int log_ps = __builtin_ctzl(ps);
+	int level, base_shift;
+
+	if ((1UL << log_ps) != ps || (log_ps & 1) != 0) {
+		pr_warn("Not enabling %ld byte huge pages;"
+			" must be a power of four.\n", ps);
+		return -EINVAL;
+	}
+
+	if (ps > 64*1024*1024*1024UL) {
+		pr_warn("Not enabling %ld MB huge pages;"
+			" largest legal value is 64 GB .\n", ps >> 20);
+		return -EINVAL;
+	} else if (ps >= PUD_SIZE) {
+		static long hv_jpage_size;
+		if (hv_jpage_size == 0)
+			hv_jpage_size = hv_sysconf(HV_SYSCONF_PAGE_SIZE_JUMBO);
+		if (hv_jpage_size != PUD_SIZE) {
+			pr_warn("Not enabling >= %ld MB huge pages:"
+				" hypervisor reports size %ld\n",
+				PUD_SIZE >> 20, hv_jpage_size);
+			return -EINVAL;
+		}
+		level = 0;
+		base_shift = PUD_SHIFT;
+	} else if (ps >= PMD_SIZE) {
+		level = 1;
+		base_shift = PMD_SHIFT;
+	} else if (ps > PAGE_SIZE) {
+		level = 2;
+		base_shift = PAGE_SHIFT;
 	} else {
-		pr_err("hugepagesz: Unsupported page size %lu M\n",
-			ps >> 20);
-		return 0;
+		pr_err("hugepagesz: huge page size %ld too small\n", ps);
+		return -EINVAL;
 	}
-	return 1;
+
+	if (log_ps != base_shift) {
+		int shift_val = log_ps - base_shift;
+		if (huge_shift[level] != 0) {
+			int old_shift = base_shift + huge_shift[level];
+			pr_warn("Not enabling %ld MB huge pages;"
+				" already have size %ld MB.\n",
+				ps >> 20, (1UL << old_shift) >> 20);
+			return -EINVAL;
+		}
+		if (hv_set_pte_super_shift(level, shift_val) != 0) {
+			pr_warn("Not enabling %ld MB huge pages;"
+				" no hypervisor support.\n", ps >> 20);
+			return -EINVAL;
+		}
+		printk(KERN_DEBUG "Enabled %ld MB huge pages\n", ps >> 20);
+		huge_shift[level] = shift_val;
+	}
+
+	hugetlb_add_hstate(log_ps - PAGE_SHIFT);
+
+	return 0;
+}
+
+static bool saw_hugepagesz;
+
+static __init int setup_hugepagesz(char *opt)
+{
+	if (!saw_hugepagesz) {
+		saw_hugepagesz = true;
+		memset(huge_shift, 0, sizeof(huge_shift));
+	}
+	return __setup_hugepagesz(memparse(opt, NULL));
 }
 __setup("hugepagesz=", setup_hugepagesz);
 
-#endif /*HAVE_ARCH_HUGETLB_UNMAPPED_AREA*/
+#ifdef ADDITIONAL_HUGE_SIZE
+/*
+ * Provide an additional huge page size if no "hugepagesz" args are given.
+ * In that case, all the cores have properly set up their hv super_shift
+ * already, but we need to notify the hugetlb code to enable the
+ * new huge page size from the Linux point of view.
+ */
+static __init int add_default_hugepagesz(void)
+{
+	if (!saw_hugepagesz) {
+		BUILD_BUG_ON(ADDITIONAL_HUGE_SIZE >= PMD_SIZE ||
+			     ADDITIONAL_HUGE_SIZE <= PAGE_SIZE);
+		BUILD_BUG_ON((PAGE_SIZE << ADDITIONAL_HUGE_SHIFT) !=
+			     ADDITIONAL_HUGE_SIZE);
+		BUILD_BUG_ON(ADDITIONAL_HUGE_SHIFT & 1);
+		hugetlb_add_hstate(ADDITIONAL_HUGE_SHIFT);
+	}
+	return 0;
+}
+arch_initcall(add_default_hugepagesz);
+#endif
+
+#endif /* CONFIG_HUGETLB_SUPER_PAGES */
diff --git a/arch/tile/mm/init.c b/arch/tile/mm/init.c
index c04fbfd93fc..630dd2ce2af 100644
--- a/arch/tile/mm/init.c
+++ b/arch/tile/mm/init.c
@@ -698,6 +698,7 @@ static void __init permanent_kmaps_init(pgd_t *pgd_base)
 #endif /* CONFIG_HIGHMEM */
 
 
+#ifndef CONFIG_64BIT
 static void __init init_free_pfn_range(unsigned long start, unsigned long end)
 {
 	unsigned long pfn;
@@ -770,6 +771,7 @@ static void __init set_non_bootmem_pages_init(void)
 		init_free_pfn_range(start, end);
 	}
 }
+#endif
 
 /*
  * paging_init() sets up the page tables - note that all of lowmem is
@@ -858,8 +860,10 @@ void __init mem_init(void)
 	/* this will put all bootmem onto the freelists */
 	totalram_pages += free_all_bootmem();
 
+#ifndef CONFIG_64BIT
 	/* count all remaining LOWMEM and give all HIGHMEM to page allocator */
 	set_non_bootmem_pages_init();
+#endif
 
 	codesize =  (unsigned long)&_etext - (unsigned long)&_text;
 	datasize =  (unsigned long)&_end - (unsigned long)&_sdata;
diff --git a/arch/tile/mm/pgtable.c b/arch/tile/mm/pgtable.c
index 3d7074347e6..345edfed9fc 100644
--- a/arch/tile/mm/pgtable.c
+++ b/arch/tile/mm/pgtable.c
@@ -132,15 +132,6 @@ void __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 	set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 }
 
-#if defined(CONFIG_HIGHPTE)
-pte_t *_pte_offset_map(pmd_t *dir, unsigned long address)
-{
-	pte_t *pte = kmap_atomic(pmd_page(*dir)) +
-		(pmd_ptfn(*dir) << HV_LOG2_PAGE_TABLE_ALIGN) & ~PAGE_MASK;
-	return &pte[pte_index(address)];
-}
-#endif
-
 /**
  * shatter_huge_page() - ensure a given address is mapped by a small page.
  *
@@ -296,10 +287,6 @@ struct page *pgtable_alloc_one(struct mm_struct *mm, unsigned long address,
 	struct page *p;
 	int i;
 
-#ifdef CONFIG_HIGHPTE
-	flags |= __GFP_HIGHMEM;
-#endif
-
 	p = alloc_pages(flags, L2_USER_PGTABLE_ORDER);
 	if (p == NULL)
 		return NULL;