powerpc: Hugetlb for BookE

Enable hugepages on Freescale BookE processors.  This allows the kernel to
use huge TLB entries to map pages, which can greatly reduce the number of
TLB misses and the amount of TLB thrashing experienced by applications with
large memory footprints.  Care should be taken when using this on FSL
processors, as the number of large TLB entries supported by the core is low
(16-64) on current processors.

The supported set of hugepage sizes include 4m, 16m, 64m, 256m, and 1g.
Page sizes larger than the max zone size are called "gigantic" pages and
must be allocated on the command line (and cannot be deallocated).

This is currently only fully implemented for Freescale 32-bit BookE
processors, but there is some infrastructure in the code for
64-bit BooKE.

Signed-off-by: Becky Bruce <beckyb@kernel.crashing.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

1 files changed, 337 insertions, 42 deletions

diff --git a/arch/powerpc/mm/hugetlbpage.c b/arch/powerpc/mm/hugetlbpage.c
index 0b9a5c1..3a5f59d 100644
--- a/arch/powerpc/mm/hugetlbpage.c
+++ b/arch/powerpc/mm/hugetlbpage.c
@@ -1,7 +1,8 @@
 /*
- * PPC64 (POWER4) Huge TLB Page Support for Kernel.
+ * PPC Huge TLB Page Support for Kernel.
  *
  * Copyright (C) 2003 David Gibson, IBM Corporation.
+ * Copyright (C) 2011 Becky Bruce, Freescale Semiconductor
  *
  * Based on the IA-32 version:
  * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com>
@@ -11,24 +12,39 @@
 #include <linux/io.h>
 #include <linux/slab.h>
 #include <linux/hugetlb.h>
+#include <linux/of_fdt.h>
+#include <linux/memblock.h>
+#include <linux/bootmem.h>
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
 #include <asm/tlb.h>
+#include <asm/setup.h>
 
 #define PAGE_SHIFT_64K	16
 #define PAGE_SHIFT_16M	24
 #define PAGE_SHIFT_16G	34
 
-#define MAX_NUMBER_GPAGES	1024
+unsigned int HPAGE_SHIFT;
 
-/* Tracks the 16G pages after the device tree is scanned and before the
- * huge_boot_pages list is ready.  */
-static unsigned long gpage_freearray[MAX_NUMBER_GPAGES];
+/*
+ * Tracks gpages after the device tree is scanned and before the
+ * huge_boot_pages list is ready.  On 64-bit implementations, this is
+ * just used to track 16G pages and so is a single array.  32-bit
+ * implementations may have more than one gpage size due to limitations
+ * of the memory allocators, so we need multiple arrays
+ */
+#ifdef CONFIG_PPC64
+#define MAX_NUMBER_GPAGES	1024
+static u64 gpage_freearray[MAX_NUMBER_GPAGES];
 static unsigned nr_gpages;
-
-/* Flag to mark huge PD pointers.  This means pmd_bad() and pud_bad()
- * will choke on pointers to hugepte tables, which is handy for
- * catching screwups early. */
+#else
+#define MAX_NUMBER_GPAGES	128
+struct psize_gpages {
+	u64 gpage_list[MAX_NUMBER_GPAGES];
+	unsigned int nr_gpages;
+};
+static struct psize_gpages gpage_freearray[MMU_PAGE_COUNT];
+#endif
 
 static inline int shift_to_mmu_psize(unsigned int shift)
 {
@@ -49,25 +65,6 @@ static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
 
 #define hugepd_none(hpd)	((hpd).pd == 0)
 
-static inline pte_t *hugepd_page(hugepd_t hpd)
-{
-	BUG_ON(!hugepd_ok(hpd));
-	return (pte_t *)((hpd.pd & ~HUGEPD_SHIFT_MASK) | 0xc000000000000000);
-}
-
-static inline unsigned int hugepd_shift(hugepd_t hpd)
-{
-	return hpd.pd & HUGEPD_SHIFT_MASK;
-}
-
-static inline pte_t *hugepte_offset(hugepd_t *hpdp, unsigned long addr, unsigned pdshift)
-{
-	unsigned long idx = (addr & ((1UL << pdshift) - 1)) >> hugepd_shift(*hpdp);
-	pte_t *dir = hugepd_page(*hpdp);
-
-	return dir + idx;
-}
-
 pte_t *find_linux_pte_or_hugepte(pgd_t *pgdir, unsigned long ea, unsigned *shift)
 {
 	pgd_t *pg;
@@ -93,7 +90,7 @@ pte_t *find_linux_pte_or_hugepte(pgd_t *pgdir, unsigned long ea, unsigned *shift
 			if (is_hugepd(pm))
 				hpdp = (hugepd_t *)pm;
 			else if (!pmd_none(*pm)) {
-				return pte_offset_map(pm, ea);
+				return pte_offset_kernel(pm, ea);
 			}
 		}
 	}
@@ -114,8 +111,18 @@ pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
 static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
 			   unsigned long address, unsigned pdshift, unsigned pshift)
 {
-	pte_t *new = kmem_cache_zalloc(PGT_CACHE(pdshift - pshift),
-				       GFP_KERNEL|__GFP_REPEAT);
+	struct kmem_cache *cachep;
+	pte_t *new;
+
+#ifdef CONFIG_PPC64
+	cachep = PGT_CACHE(pdshift - pshift);
+#else
+	int i;
+	int num_hugepd = 1 << (pshift - pdshift);
+	cachep = hugepte_cache;
+#endif
+
+	new = kmem_cache_zalloc(cachep, GFP_KERNEL|__GFP_REPEAT);
 
 	BUG_ON(pshift > HUGEPD_SHIFT_MASK);
 	BUG_ON((unsigned long)new & HUGEPD_SHIFT_MASK);
@@ -124,10 +131,31 @@ static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
 		return -ENOMEM;
 
 	spin_lock(&mm->page_table_lock);
+#ifdef CONFIG_PPC64
 	if (!hugepd_none(*hpdp))
-		kmem_cache_free(PGT_CACHE(pdshift - pshift), new);
+		kmem_cache_free(cachep, new);
 	else
-		hpdp->pd = ((unsigned long)new & ~0x8000000000000000) | pshift;
+		hpdp->pd = ((unsigned long)new & ~PD_HUGE) | pshift;
+#else
+	/*
+	 * We have multiple higher-level entries that point to the same
+	 * actual pte location.  Fill in each as we go and backtrack on error.
+	 * We need all of these so the DTLB pgtable walk code can find the
+	 * right higher-level entry without knowing if it's a hugepage or not.
+	 */
+	for (i = 0; i < num_hugepd; i++, hpdp++) {
+		if (unlikely(!hugepd_none(*hpdp)))
+			break;
+		else
+			hpdp->pd = ((unsigned long)new & ~PD_HUGE) | pshift;
+	}
+	/* If we bailed from the for loop early, an error occurred, clean up */
+	if (i < num_hugepd) {
+		for (i = i - 1 ; i >= 0; i--, hpdp--)
+			hpdp->pd = 0;
+		kmem_cache_free(cachep, new);
+	}
+#endif
 	spin_unlock(&mm->page_table_lock);
 	return 0;
 }
@@ -169,11 +197,132 @@ pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr, unsigned long sz
 	return hugepte_offset(hpdp, addr, pdshift);
 }
 
+#ifdef CONFIG_PPC32
 /* Build list of addresses of gigantic pages.  This function is used in early
  * boot before the buddy or bootmem allocator is setup.
  */
-void add_gpage(unsigned long addr, unsigned long page_size,
-	unsigned long number_of_pages)
+void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages)
+{
+	unsigned int idx = shift_to_mmu_psize(__ffs(page_size));
+	int i;
+
+	if (addr == 0)
+		return;
+
+	gpage_freearray[idx].nr_gpages = number_of_pages;
+
+	for (i = 0; i < number_of_pages; i++) {
+		gpage_freearray[idx].gpage_list[i] = addr;
+		addr += page_size;
+	}
+}
+
+/*
+ * Moves the gigantic page addresses from the temporary list to the
+ * huge_boot_pages list.
+ */
+int alloc_bootmem_huge_page(struct hstate *hstate)
+{
+	struct huge_bootmem_page *m;
+	int idx = shift_to_mmu_psize(hstate->order + PAGE_SHIFT);
+	int nr_gpages = gpage_freearray[idx].nr_gpages;
+
+	if (nr_gpages == 0)
+		return 0;
+
+#ifdef CONFIG_HIGHMEM
+	/*
+	 * If gpages can be in highmem we can't use the trick of storing the
+	 * data structure in the page; allocate space for this
+	 */
+	m = alloc_bootmem(sizeof(struct huge_bootmem_page));
+	m->phys = gpage_freearray[idx].gpage_list[--nr_gpages];
+#else
+	m = phys_to_virt(gpage_freearray[idx].gpage_list[--nr_gpages]);
+#endif
+
+	list_add(&m->list, &huge_boot_pages);
+	gpage_freearray[idx].nr_gpages = nr_gpages;
+	gpage_freearray[idx].gpage_list[nr_gpages] = 0;
+	m->hstate = hstate;
+
+	return 1;
+}
+/*
+ * Scan the command line hugepagesz= options for gigantic pages; store those in
+ * a list that we use to allocate the memory once all options are parsed.
+ */
+
+unsigned long gpage_npages[MMU_PAGE_COUNT];
+
+static int __init do_gpage_early_setup(char *param, char *val)
+{
+	static phys_addr_t size;
+	unsigned long npages;
+
+	/*
+	 * The hugepagesz and hugepages cmdline options are interleaved.  We
+	 * use the size variable to keep track of whether or not this was done
+	 * properly and skip over instances where it is incorrect.  Other
+	 * command-line parsing code will issue warnings, so we don't need to.
+	 *
+	 */
+	if ((strcmp(param, "default_hugepagesz") == 0) ||
+	    (strcmp(param, "hugepagesz") == 0)) {
+		size = memparse(val, NULL);
+	} else if (strcmp(param, "hugepages") == 0) {
+		if (size != 0) {
+			if (sscanf(val, "%lu", &npages) <= 0)
+				npages = 0;
+			gpage_npages[shift_to_mmu_psize(__ffs(size))] = npages;
+			size = 0;
+		}
+	}
+	return 0;
+}
+
+
+/*
+ * This function allocates physical space for pages that are larger than the
+ * buddy allocator can handle.  We want to allocate these in highmem because
+ * the amount of lowmem is limited.  This means that this function MUST be
+ * called before lowmem_end_addr is set up in MMU_init() in order for the lmb
+ * allocate to grab highmem.
+ */
+void __init reserve_hugetlb_gpages(void)
+{
+	static __initdata char cmdline[COMMAND_LINE_SIZE];
+	phys_addr_t size, base;
+	int i;
+
+	strlcpy(cmdline, boot_command_line, COMMAND_LINE_SIZE);
+	parse_args("hugetlb gpages", cmdline, NULL, 0, &do_gpage_early_setup);
+
+	/*
+	 * Walk gpage list in reverse, allocating larger page sizes first.
+	 * Skip over unsupported sizes, or sizes that have 0 gpages allocated.
+	 * When we reach the point in the list where pages are no longer
+	 * considered gpages, we're done.
+	 */
+	for (i = MMU_PAGE_COUNT-1; i >= 0; i--) {
+		if (mmu_psize_defs[i].shift == 0 || gpage_npages[i] == 0)
+			continue;
+		else if (mmu_psize_to_shift(i) < (MAX_ORDER + PAGE_SHIFT))
+			break;
+
+		size = (phys_addr_t)(1ULL << mmu_psize_to_shift(i));
+		base = memblock_alloc_base(size * gpage_npages[i], size,
+					   MEMBLOCK_ALLOC_ANYWHERE);
+		add_gpage(base, size, gpage_npages[i]);
+	}
+}
+
+#else /* PPC64 */
+
+/* Build list of addresses of gigantic pages.  This function is used in early
+ * boot before the buddy or bootmem allocator is setup.
+ */
+void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages)
 {
 	if (!addr)
 		return;
@@ -199,19 +348,79 @@ int alloc_bootmem_huge_page(struct hstate *hstate)
 	m->hstate = hstate;
 	return 1;
 }
+#endif
 
 int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
 {
 	return 0;
 }
 
+#ifdef CONFIG_PPC32
+#define HUGEPD_FREELIST_SIZE \
+	((PAGE_SIZE - sizeof(struct hugepd_freelist)) / sizeof(pte_t))
+
+struct hugepd_freelist {
+	struct rcu_head	rcu;
+	unsigned int index;
+	void *ptes[0];
+};
+
+static DEFINE_PER_CPU(struct hugepd_freelist *, hugepd_freelist_cur);
+
+static void hugepd_free_rcu_callback(struct rcu_head *head)
+{
+	struct hugepd_freelist *batch =
+		container_of(head, struct hugepd_freelist, rcu);
+	unsigned int i;
+
+	for (i = 0; i < batch->index; i++)
+		kmem_cache_free(hugepte_cache, batch->ptes[i]);
+
+	free_page((unsigned long)batch);
+}
+
+static void hugepd_free(struct mmu_gather *tlb, void *hugepte)
+{
+	struct hugepd_freelist **batchp;
+
+	batchp = &__get_cpu_var(hugepd_freelist_cur);
+
+	if (atomic_read(&tlb->mm->mm_users) < 2 ||
+	    cpumask_equal(mm_cpumask(tlb->mm),
+			  cpumask_of(smp_processor_id()))) {
+		kmem_cache_free(hugepte_cache, hugepte);
+		return;
+	}
+
+	if (*batchp == NULL) {
+		*batchp = (struct hugepd_freelist *)__get_free_page(GFP_ATOMIC);
+		(*batchp)->index = 0;
+	}
+
+	(*batchp)->ptes[(*batchp)->index++] = hugepte;
+	if ((*batchp)->index == HUGEPD_FREELIST_SIZE) {
+		call_rcu_sched(&(*batchp)->rcu, hugepd_free_rcu_callback);
+		*batchp = NULL;
+	}
+}
+#endif
+
 static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshift,
 			      unsigned long start, unsigned long end,
 			      unsigned long floor, unsigned long ceiling)
 {
 	pte_t *hugepte = hugepd_page(*hpdp);
-	unsigned shift = hugepd_shift(*hpdp);
+	int i;
+
 	unsigned long pdmask = ~((1UL << pdshift) - 1);
+	unsigned int num_hugepd = 1;
+
+#ifdef CONFIG_PPC64
+	unsigned int shift = hugepd_shift(*hpdp);
+#else
+	/* Note: On 32-bit the hpdp may be the first of several */
+	num_hugepd = (1 << (hugepd_shift(*hpdp) - pdshift));
+#endif
 
 	start &= pdmask;
 	if (start < floor)
@@ -224,9 +433,15 @@ static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshif
 	if (end - 1 > ceiling - 1)
 		return;
 
-	hpdp->pd = 0;
+	for (i = 0; i < num_hugepd; i++, hpdp++)
+		hpdp->pd = 0;
+
 	tlb->need_flush = 1;
+#ifdef CONFIG_PPC64
 	pgtable_free_tlb(tlb, hugepte, pdshift - shift);
+#else
+	hugepd_free(tlb, hugepte);
+#endif
 }
 
 static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
@@ -331,18 +546,27 @@ void hugetlb_free_pgd_range(struct mmu_gather *tlb,
 	 * too.
 	 */
 
-	pgd = pgd_offset(tlb->mm, addr);
 	do {
 		next = pgd_addr_end(addr, end);
+		pgd = pgd_offset(tlb->mm, addr);
 		if (!is_hugepd(pgd)) {
 			if (pgd_none_or_clear_bad(pgd))
 				continue;
 			hugetlb_free_pud_range(tlb, pgd, addr, next, floor, ceiling);
 		} else {
+#ifdef CONFIG_PPC32
+			/*
+			 * Increment next by the size of the huge mapping since
+			 * on 32-bit there may be more than one entry at the pgd
+			 * level for a single hugepage, but all of them point to
+			 * the same kmem cache that holds the hugepte.
+			 */
+			next = addr + (1 << hugepd_shift(*(hugepd_t *)pgd));
+#endif
 			free_hugepd_range(tlb, (hugepd_t *)pgd, PGDIR_SHIFT,
 					  addr, next, floor, ceiling);
 		}
-	} while (pgd++, addr = next, addr != end);
+	} while (addr = next, addr != end);
 }
 
 struct page *
@@ -466,17 +690,35 @@ unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 					unsigned long len, unsigned long pgoff,
 					unsigned long flags)
 {
+#ifdef CONFIG_MM_SLICES
 	struct hstate *hstate = hstate_file(file);
 	int mmu_psize = shift_to_mmu_psize(huge_page_shift(hstate));
 
 	return slice_get_unmapped_area(addr, len, flags, mmu_psize, 1, 0);
+#else
+	return get_unmapped_area(file, addr, len, pgoff, flags);
+#endif
 }
 
 unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
 {
+#ifdef CONFIG_MM_SLICES
 	unsigned int psize = get_slice_psize(vma->vm_mm, vma->vm_start);
 
 	return 1UL << mmu_psize_to_shift(psize);
+#else
+	if (!is_vm_hugetlb_page(vma))
+		return PAGE_SIZE;
+
+	return huge_page_size(hstate_vma(vma));
+#endif
+}
+
+static inline bool is_power_of_4(unsigned long x)
+{
+	if (is_power_of_2(x))
+		return (__ilog2(x) % 2) ? false : true;
+	return false;
 }
 
 static int __init add_huge_page_size(unsigned long long size)
@@ -486,9 +728,14 @@ static int __init add_huge_page_size(unsigned long long size)
 
 	/* Check that it is a page size supported by the hardware and
 	 * that it fits within pagetable and slice limits. */
+#ifdef CONFIG_PPC_FSL_BOOK3E
+	if ((size < PAGE_SIZE) || !is_power_of_4(size))
+		return -EINVAL;
+#else
 	if (!is_power_of_2(size)
 	    || (shift > SLICE_HIGH_SHIFT) || (shift <= PAGE_SHIFT))
 		return -EINVAL;
+#endif
 
 	if ((mmu_psize = shift_to_mmu_psize(shift)) < 0)
 		return -EINVAL;
@@ -525,6 +772,46 @@ static int __init hugepage_setup_sz(char *str)
 }
 __setup("hugepagesz=", hugepage_setup_sz);
 
+#ifdef CONFIG_FSL_BOOKE
+struct kmem_cache *hugepte_cache;
+static int __init hugetlbpage_init(void)
+{
+	int psize;
+
+	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
+		unsigned shift;
+
+		if (!mmu_psize_defs[psize].shift)
+			continue;
+
+		shift = mmu_psize_to_shift(psize);
+
+		/* Don't treat normal page sizes as huge... */
+		if (shift != PAGE_SHIFT)
+			if (add_huge_page_size(1ULL << shift) < 0)
+				continue;
+	}
+
+	/*
+	 * Create a kmem cache for hugeptes.  The bottom bits in the pte have
+	 * size information encoded in them, so align them to allow this
+	 */
+	hugepte_cache =  kmem_cache_create("hugepte-cache", sizeof(pte_t),
+					   HUGEPD_SHIFT_MASK + 1, 0, NULL);
+	if (hugepte_cache == NULL)
+		panic("%s: Unable to create kmem cache for hugeptes\n",
+		      __func__);
+
+	/* Default hpage size = 4M */
+	if (mmu_psize_defs[MMU_PAGE_4M].shift)
+		HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_4M].shift;
+	else
+		panic("%s: Unable to set default huge page size\n", __func__);
+
+
+	return 0;
+}
+#else
 static int __init hugetlbpage_init(void)
 {
 	int psize;
@@ -567,15 +854,23 @@ static int __init hugetlbpage_init(void)
 
 	return 0;
 }
-
+#endif
 module_init(hugetlbpage_init);
 
 void flush_dcache_icache_hugepage(struct page *page)
 {
 	int i;
+	void *start;
 
 	BUG_ON(!PageCompound(page));
 
-	for (i = 0; i < (1UL << compound_order(page)); i++)
-		__flush_dcache_icache(page_address(page+i));
+	for (i = 0; i < (1UL << compound_order(page)); i++) {
+		if (!PageHighMem(page)) {
+			__flush_dcache_icache(page_address(page+i));
+		} else {
+			start = kmap_atomic(page+i, KM_PPC_SYNC_ICACHE);
+			__flush_dcache_icache(start);
+			kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
+		}
+	}
 }