arch/tile/mm/pgtable.c

/*
 * Copyright 2010 Tilera Corporation. All Rights Reserved.
 *
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU General Public License
 *   as published by the Free Software Foundation, version 2.
 *
 *   This program is distributed in the hope that it will be useful, but
 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 */

#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/mm.h>
#include <linux/swap.h>
#include <linux/highmem.h>
#include <linux/slab.h>
#include <linux/pagemap.h>
#include <linux/spinlock.h>
#include <linux/cpumask.h>
#include <linux/module.h>
#include <linux/io.h>
#include <linux/vmalloc.h>
#include <linux/smp.h>

#include <asm/pgtable.h>
#include <asm/pgalloc.h>
#include <asm/fixmap.h>
#include <asm/tlb.h>
#include <asm/tlbflush.h>
#include <asm/homecache.h>

#define K(x) ((x) << (PAGE_SHIFT-10))

/*
 * The normal show_free_areas() is too verbose on Tile, with dozens
 * of processors and often four NUMA zones each with high and lowmem.
 */
void show_mem(unsigned int filter)
{
	struct zone *zone;

	pr_err("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu"
	       " free:%lu\n slab:%lu mapped:%lu pagetables:%lu bounce:%lu"
	       " pagecache:%lu swap:%lu\n",
	       (global_page_state(NR_ACTIVE_ANON) +
		global_page_state(NR_ACTIVE_FILE)),
	       (global_page_state(NR_INACTIVE_ANON) +
		global_page_state(NR_INACTIVE_FILE)),
	       global_page_state(NR_FILE_DIRTY),
	       global_page_state(NR_WRITEBACK),
	       global_page_state(NR_UNSTABLE_NFS),
	       global_page_state(NR_FREE_PAGES),
	       (global_page_state(NR_SLAB_RECLAIMABLE) +
		global_page_state(NR_SLAB_UNRECLAIMABLE)),
	       global_page_state(NR_FILE_MAPPED),
	       global_page_state(NR_PAGETABLE),
	       global_page_state(NR_BOUNCE),
	       global_page_state(NR_FILE_PAGES),
	       get_nr_swap_pages());

	for_each_zone(zone) {
		unsigned long flags, order, total = 0, largest_order = -1;

		if (!populated_zone(zone))
			continue;

		spin_lock_irqsave(&zone->lock, flags);
		for (order = 0; order < MAX_ORDER; order++) {
			int nr = zone->free_area[order].nr_free;
			total += nr << order;
			if (nr)
				largest_order = order;
		}
		spin_unlock_irqrestore(&zone->lock, flags);
		pr_err("Node %d %7s: %lukB (largest %luKb)\n",
		       zone_to_nid(zone), zone->name,
		       K(total), largest_order ? K(1UL) << largest_order : 0);
	}
}

/**
 * shatter_huge_page() - ensure a given address is mapped by a small page.
 *
 * This function converts a huge PTE mapping kernel LOWMEM into a bunch
 * of small PTEs with the same caching.  No cache flush required, but we
 * must do a global TLB flush.
 *
 * Any caller that wishes to modify a kernel mapping that might
 * have been made with a huge page should call this function,
 * since doing so properly avoids race conditions with installing the
 * newly-shattered page and then flushing all the TLB entries.
 *
 * @addr: Address at which to shatter any existing huge page.
 */
void shatter_huge_page(unsigned long addr)
{
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
	unsigned long flags = 0;  /* happy compiler */
#ifdef __PAGETABLE_PMD_FOLDED
	struct list_head *pos;
#endif

	/* Get a pointer to the pmd entry that we need to change. */
	addr &= HPAGE_MASK;
	BUG_ON(pgd_addr_invalid(addr));
	BUG_ON(addr < PAGE_OFFSET);  /* only for kernel LOWMEM */
	pgd = swapper_pg_dir + pgd_index(addr);
	pud = pud_offset(pgd, addr);
	BUG_ON(!pud_present(*pud));
	pmd = pmd_offset(pud, addr);
	BUG_ON(!pmd_present(*pmd));
	if (!pmd_huge_page(*pmd))
		return;

	spin_lock_irqsave(&init_mm.page_table_lock, flags);
	if (!pmd_huge_page(*pmd)) {
		/* Lost the race to convert the huge page. */
		spin_unlock_irqrestore(&init_mm.page_table_lock, flags);
		return;
	}

	/* Shatter the huge page into the preallocated L2 page table. */
	pmd_populate_kernel(&init_mm, pmd,
			    get_prealloc_pte(pte_pfn(*(pte_t *)pmd)));

#ifdef __PAGETABLE_PMD_FOLDED
	/* Walk every pgd on the system and update the pmd there. */
	spin_lock(&pgd_lock);
	list_for_each(pos, &pgd_list) {
		pmd_t *copy_pmd;
		pgd = list_to_pgd(pos) + pgd_index(addr);
		pud = pud_offset(pgd, addr);
		copy_pmd = pmd_offset(pud, addr);
		__set_pmd(copy_pmd, *pmd);
	}
	spin_unlock(&pgd_lock);
#endif

	/* Tell every cpu to notice the change. */
	flush_remote(0, 0, NULL, addr, HPAGE_SIZE, HPAGE_SIZE,
		     cpu_possible_mask, NULL, 0);

	/* Hold the lock until the TLB flush is finished to avoid races. */
	spin_unlock_irqrestore(&init_mm.page_table_lock, flags);
}

/*
 * List of all pgd's needed so it can invalidate entries in both cached
 * and uncached pgd's. This is essentially codepath-based locking
 * against pageattr.c; it is the unique case in which a valid change
 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 * vmalloc faults work because attached pagetables are never freed.
 *
 * The lock is always taken with interrupts disabled, unlike on x86
 * and other platforms, because we need to take the lock in
 * shatter_huge_page(), which may be called from an interrupt context.
 * We are not at risk from the tlbflush IPI deadlock that was seen on
 * x86, since we use the flush_remote() API to have the hypervisor do
 * the TLB flushes regardless of irq disabling.
 */
DEFINE_SPINLOCK(pgd_lock);
LIST_HEAD(pgd_list);

static inline void pgd_list_add(pgd_t *pgd)
{
	list_add(pgd_to_list(pgd), &pgd_list);
}

static inline void pgd_list_del(pgd_t *pgd)
{
	list_del(pgd_to_list(pgd));
}

#define KERNEL_PGD_INDEX_START pgd_index(PAGE_OFFSET)
#define KERNEL_PGD_PTRS (PTRS_PER_PGD - KERNEL_PGD_INDEX_START)

static void pgd_ctor(pgd_t *pgd)
{
	unsigned long flags;

	memset(pgd, 0, KERNEL_PGD_INDEX_START*sizeof(pgd_t));
	spin_lock_irqsave(&pgd_lock, flags);

#ifndef __tilegx__
	/*
	 * Check that the user interrupt vector has no L2.
	 * It never should for the swapper, and new page tables
	 * should always start with an empty user interrupt vector.
	 */
	BUG_ON(((u64 *)swapper_pg_dir)[pgd_index(MEM_USER_INTRPT)] != 0);
#endif

	memcpy(pgd + KERNEL_PGD_INDEX_START,
	       swapper_pg_dir + KERNEL_PGD_INDEX_START,
	       KERNEL_PGD_PTRS * sizeof(pgd_t));

	pgd_list_add(pgd);
	spin_unlock_irqrestore(&pgd_lock, flags);
}

static void pgd_dtor(pgd_t *pgd)
{
	unsigned long flags; /* can be called from interrupt context */

	spin_lock_irqsave(&pgd_lock, flags);
	pgd_list_del(pgd);
	spin_unlock_irqrestore(&pgd_lock, flags);
}

pgd_t *pgd_alloc(struct mm_struct *mm)
{
	pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
	if (pgd)
		pgd_ctor(pgd);
	return pgd;
}

void pgd_free(struct mm_struct *mm, pgd_t *pgd)
{
	pgd_dtor(pgd);
	kmem_cache_free(pgd_cache, pgd);
}


#define L2_USER_PGTABLE_PAGES (1 << L2_USER_PGTABLE_ORDER)

struct page *pgtable_alloc_one(struct mm_struct *mm, unsigned long address,
			       int order)
{
	gfp_t flags = GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO;
	struct page *p;
	int i;

	p = alloc_pages(flags, L2_USER_PGTABLE_ORDER);
	if (p == NULL)
		return NULL;

	/*
	 * Make every page have a page_count() of one, not just the first.
	 * We don't use __GFP_COMP since it doesn't look like it works
	 * correctly with tlb_remove_page().
	 */
	for (i = 1; i < order; ++i) {
		init_page_count(p+i);
		inc_zone_page_state(p+i, NR_PAGETABLE);
	}

	pgtable_page_ctor(p);
	return p;
}

/*
 * Free page immediately (used in __pte_alloc if we raced with another
 * process).  We have to correct whatever pte_alloc_one() did before
 * returning the pages to the allocator.
 */
void pgtable_free(struct mm_struct *mm, struct page *p, int order)
{
	int i;

	pgtable_page_dtor(p);
	__free_page(p);

	for (i = 1; i < order; ++i) {
		__free_page(p+i);
		dec_zone_page_state(p+i, NR_PAGETABLE);
	}
}

void __pgtable_free_tlb(struct mmu_gather *tlb, struct page *pte,
			unsigned long address, int order)
{
	int i;

	pgtable_page_dtor(pte);
	tlb_remove_page(tlb, pte);

	for (i = 1; i < order; ++i) {
		tlb_remove_page(tlb, pte + i);
		dec_zone_page_state(pte + i, NR_PAGETABLE);
	}
}

#ifndef __tilegx__

/*
 * FIXME: needs to be atomic vs hypervisor writes.  For now we make the
 * window of vulnerability a bit smaller by doing an unlocked 8-bit update.
 */
int ptep_test_and_clear_young(struct vm_area_struct *vma,
			      unsigned long addr, pte_t *ptep)
{
#if HV_PTE_INDEX_ACCESSED < 8 || HV_PTE_INDEX_ACCESSED >= 16
# error Code assumes HV_PTE "accessed" bit in second byte
#endif
	u8 *tmp = (u8 *)ptep;
	u8 second_byte = tmp[1];
	if (!(second_byte & (1 << (HV_PTE_INDEX_ACCESSED - 8))))
		return 0;
	tmp[1] = second_byte & ~(1 << (HV_PTE_INDEX_ACCESSED - 8));
	return 1;
}

/*
 * This implementation is atomic vs hypervisor writes, since the hypervisor
 * always writes the low word (where "accessed" and "dirty" are) and this
 * routine only writes the high word.
 */
void ptep_set_wrprotect(struct mm_struct *mm,
			unsigned long addr, pte_t *ptep)
{
#if HV_PTE_INDEX_WRITABLE < 32
# error Code assumes HV_PTE "writable" bit in high word
#endif
	u32 *tmp = (u32 *)ptep;
	tmp[1] = tmp[1] & ~(1 << (HV_PTE_INDEX_WRITABLE - 32));
}

#endif

pte_t *virt_to_pte(struct mm_struct* mm, unsigned long addr)
{
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;

	if (pgd_addr_invalid(addr))
		return NULL;

	pgd = mm ? pgd_offset(mm, addr) : swapper_pg_dir + pgd_index(addr);
	pud = pud_offset(pgd, addr);
	if (!pud_present(*pud))
		return NULL;
	pmd = pmd_offset(pud, addr);
	if (pmd_huge_page(*pmd))
		return (pte_t *)pmd;
	if (!pmd_present(*pmd))
		return NULL;
	return pte_offset_kernel(pmd, addr);
}

pgprot_t set_remote_cache_cpu(pgprot_t prot, int cpu)
{
	unsigned int width = smp_width;
	int x = cpu % width;
	int y = cpu / width;
	BUG_ON(y >= smp_height);
	BUG_ON(hv_pte_get_mode(prot) != HV_PTE_MODE_CACHE_TILE_L3);
	BUG_ON(cpu < 0 || cpu >= NR_CPUS);
	BUG_ON(!cpu_is_valid_lotar(cpu));
	return hv_pte_set_lotar(prot, HV_XY_TO_LOTAR(x, y));
}

int get_remote_cache_cpu(pgprot_t prot)
{
	HV_LOTAR lotar = hv_pte_get_lotar(prot);
	int x = HV_LOTAR_X(lotar);
	int y = HV_LOTAR_Y(lotar);
	BUG_ON(hv_pte_get_mode(prot) != HV_PTE_MODE_CACHE_TILE_L3);
	return x + y * smp_width;
}

/*
 * Convert a kernel VA to a PA and homing information.
 */
int va_to_cpa_and_pte(void *va, unsigned long long *cpa, pte_t *pte)
{
	struct page *page = virt_to_page(va);
	pte_t null_pte = { 0 };

	*cpa = __pa(va);

	/* Note that this is not writing a page table, just returning a pte. */
	*pte = pte_set_home(null_pte, page_home(page));

	return 0; /* return non-zero if not hfh? */
}
EXPORT_SYMBOL(va_to_cpa_and_pte);

void __set_pte(pte_t *ptep, pte_t pte)
{
#ifdef __tilegx__
	*ptep = pte;
#else
# if HV_PTE_INDEX_PRESENT >= 32 || HV_PTE_INDEX_MIGRATING >= 32
#  error Must write the present and migrating bits last
# endif
	if (pte_present(pte)) {
		((u32 *)ptep)[1] = (u32)(pte_val(pte) >> 32);
		barrier();
		((u32 *)ptep)[0] = (u32)(pte_val(pte));
	} else {
		((u32 *)ptep)[0] = (u32)(pte_val(pte));
		barrier();
		((u32 *)ptep)[1] = (u32)(pte_val(pte) >> 32);
	}
#endif /* __tilegx__ */
}

void set_pte(pte_t *ptep, pte_t pte)
{
	if (pte_present(pte) &&
	    (!CHIP_HAS_MMIO() || hv_pte_get_mode(pte) != HV_PTE_MODE_MMIO)) {
		/* The PTE actually references physical memory. */
		unsigned long pfn = pte_pfn(pte);
		if (pfn_valid(pfn)) {
			/* Update the home of the PTE from the struct page. */
			pte = pte_set_home(pte, page_home(pfn_to_page(pfn)));
		} else if (hv_pte_get_mode(pte) == 0) {
			/* remap_pfn_range(), etc, must supply PTE mode. */
			panic("set_pte(): out-of-range PFN and mode 0\n");
		}
	}

	__set_pte(ptep, pte);
}

/* Can this mm load a PTE with cached_priority set? */
static inline int mm_is_priority_cached(struct mm_struct *mm)
{
	return mm->context.priority_cached != 0;
}

/*
 * Add a priority mapping to an mm_context and
 * notify the hypervisor if this is the first one.
 */
void start_mm_caching(struct mm_struct *mm)
{
	if (!mm_is_priority_cached(mm)) {
		mm->context.priority_cached = -1UL;
		hv_set_caching(-1UL);
	}
}

/*
 * Validate and return the priority_cached flag.  We know if it's zero
 * that we don't need to scan, since we immediately set it non-zero
 * when we first consider a MAP_CACHE_PRIORITY mapping.
 *
 * We only _try_ to acquire the mmap_sem semaphore; if we can't acquire it,
 * since we're in an interrupt context (servicing switch_mm) we don't
 * worry about it and don't unset the "priority_cached" field.
 * Presumably we'll come back later and have more luck and clear
 * the value then; for now we'll just keep the cache marked for priority.
 */
static unsigned long update_priority_cached(struct mm_struct *mm)
{
	if (mm->context.priority_cached && down_write_trylock(&mm->mmap_sem)) {
		struct vm_area_struct *vm;
		for (vm = mm->mmap; vm; vm = vm->vm_next) {
			if (hv_pte_get_cached_priority(vm->vm_page_prot))
				break;
		}
		if (vm == NULL)
			mm->context.priority_cached = 0;
		up_write(&mm->mmap_sem);
	}
	return mm->context.priority_cached;
}

/* Set caching correctly for an mm that we are switching to. */
void check_mm_caching(struct mm_struct *prev, struct mm_struct *next)
{
	if (!mm_is_priority_cached(next)) {
		/*
		 * If the new mm doesn't use priority caching, just see if we
		 * need the hv_set_caching(), or can assume it's already zero.
		 */
		if (mm_is_priority_cached(prev))
			hv_set_caching(0);
	} else {
		hv_set_caching(update_priority_cached(next));
	}
}

#if CHIP_HAS_MMIO()

/* Map an arbitrary MMIO address, homed according to pgprot, into VA space. */
void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size,
			   pgprot_t home)
{
	void *addr;
	struct vm_struct *area;
	unsigned long offset, last_addr;
	pgprot_t pgprot;

	/* Don't allow wraparound or zero size */
	last_addr = phys_addr + size - 1;
	if (!size || last_addr < phys_addr)
		return NULL;

	/* Create a read/write, MMIO VA mapping homed at the requested shim. */
	pgprot = PAGE_KERNEL;
	pgprot = hv_pte_set_mode(pgprot, HV_PTE_MODE_MMIO);
	pgprot = hv_pte_set_lotar(pgprot, hv_pte_get_lotar(home));

	/*
	 * Mappings have to be page-aligned
	 */
	offset = phys_addr & ~PAGE_MASK;
	phys_addr &= PAGE_MASK;
	size = PAGE_ALIGN(last_addr+1) - phys_addr;

	/*
	 * Ok, go for it..
	 */
	area = get_vm_area(size, VM_IOREMAP /* | other flags? */);
	if (!area)
		return NULL;
	area->phys_addr = phys_addr;
	addr = area->addr;
	if (ioremap_page_range((unsigned long)addr, (unsigned long)addr + size,
			       phys_addr, pgprot)) {
		free_vm_area(area);
		return NULL;
	}
	return (__force void __iomem *) (offset + (char *)addr);
}
EXPORT_SYMBOL(ioremap_prot);

/* Unmap an MMIO VA mapping. */
void iounmap(volatile void __iomem *addr_in)
{
	volatile void __iomem *addr = (volatile void __iomem *)
		(PAGE_MASK & (unsigned long __force)addr_in);
#if 1
	vunmap((void * __force)addr);
#else
	/* x86 uses this complicated flow instead of vunmap().  Is
	 * there any particular reason we should do the same? */
	struct vm_struct *p, *o;

	/* Use the vm area unlocked, assuming the caller
	   ensures there isn't another iounmap for the same address
	   in parallel. Reuse of the virtual address is prevented by
	   leaving it in the global lists until we're done with it.
	   cpa takes care of the direct mappings. */
	p = find_vm_area((void *)addr);

	if (!p) {
		pr_err("iounmap: bad address %p\n", addr);
		dump_stack();
		return;
	}

	/* Finally remove it */
	o = remove_vm_area((void *)addr);
	BUG_ON(p != o || o == NULL);
	kfree(p);
#endif
}
EXPORT_SYMBOL(iounmap);

#endif /* CHIP_HAS_MMIO() */