WebSVN – HelenOS – Diff – /trunk/kernel/generic/src/mm/as.c


/*
 * Copyright (c) 2001-2006 Jakub Jermar
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 *   notice, this list of conditions and the following disclaimer.
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 * - The name of the author may not be used to endorse or promote products
 *   derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
/** @addtogroup genericmm
 * @{
 */
 
/**
 * @file
 * @brief   Address space related functions.
 *
 * This file contains address space manipulation functions.
 * Roughly speaking, this is a higher-level client of
 * Virtual Address Translation (VAT) subsystem.
 *
 * Functionality provided by this file allows one to
 * create address spaces and create, resize and share
 * address space areas.
 *
 * @see page.c
 *
 */
 
#include <mm/as.h>
#include <arch/mm/as.h>
#include <mm/page.h>
#include <mm/frame.h>
#include <mm/slab.h>
#include <mm/tlb.h>
#include <arch/mm/page.h>
#include <genarch/mm/page_pt.h>
#include <genarch/mm/page_ht.h>
#include <mm/asid.h>
#include <arch/mm/asid.h>
#include <preemption.h>
#include <synch/spinlock.h>
#include <synch/mutex.h>
#include <adt/list.h>
#include <adt/btree.h>
#include <proc/task.h>
#include <proc/thread.h>
#include <arch/asm.h>
#include <panic.h>
#include <debug.h>
#include <print.h>
#include <memstr.h>
#include <macros.h>
#include <arch.h>
#include <errno.h>
#include <config.h>
#include <align.h>
#include <arch/types.h>
#include <syscall/copy.h>
#include <arch/interrupt.h>
 
#ifdef CONFIG_VIRT_IDX_DCACHE
#include <arch/mm/cache.h>
#endif /* CONFIG_VIRT_IDX_DCACHE */
 
/**
 * Each architecture decides what functions will be used to carry out
 * address space operations such as creating or locking page tables.
 */
as_operations_t *as_operations = NULL;
 
/**
 * Slab for as_t objects.
 */
static slab_cache_t *as_slab;
 
/**
 * This lock serializes access to the ASID subsystem.
 * It protects:
 * - inactive_as_with_asid_head list
 * - as->asid for each as of the as_t type
 * - asids_allocated counter
 */
SPINLOCK_INITIALIZE(asidlock);
 
/**
 * This list contains address spaces that are not active on any
 * processor and that have valid ASID.
 */
LIST_INITIALIZE(inactive_as_with_asid_head);
 
/** Kernel address space. */
as_t *AS_KERNEL = NULL;
 
static int area_flags_to_page_flags(int);
static as_area_t *find_area_and_lock(as_t *, uintptr_t);
static bool check_area_conflicts(as_t *, uintptr_t, size_t, as_area_t *);
static void sh_info_remove_reference(share_info_t *);
 
static int as_constructor(void *obj, int flags)
{
    as_t *as = (as_t *) obj;
    int rc;
 
    link_initialize(&as->inactive_as_with_asid_link);
    mutex_initialize(&as->lock, MUTEX_PASSIVE);
   
    rc = as_constructor_arch(as, flags);
   
    return rc;
}
 
static int as_destructor(void *obj)
{
    as_t *as = (as_t *) obj;
 
    return as_destructor_arch(as);
}
 
/** Initialize address space subsystem. */
void as_init(void)
{
    as_arch_init();
 
    as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,
        as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);
   
    AS_KERNEL = as_create(FLAG_AS_KERNEL);
    if (!AS_KERNEL)
        panic("Cannot create kernel address space.");
   
    /* Make sure the kernel address space
     * reference count never drops to zero.
     */
    atomic_set(&AS_KERNEL->refcount, 1);
}
 
/** Create address space.
 *
 * @param flags     Flags that influence the way in wich the address space
 *          is created.
 */
as_t *as_create(int flags)
{
    as_t *as;
 
    as = (as_t *) slab_alloc(as_slab, 0);
    (void) as_create_arch(as, 0);
   
    btree_create(&as->as_area_btree);
   
    if (flags & FLAG_AS_KERNEL)
        as->asid = ASID_KERNEL;
    else
        as->asid = ASID_INVALID;
   
    atomic_set(&as->refcount, 0);
    as->cpu_refcount = 0;
#ifdef AS_PAGE_TABLE
    as->genarch.page_table = page_table_create(flags);
#else
    page_table_create(flags);
#endif
   
    return as;
}
 
/** Destroy adress space.
 *
 * When there are no tasks referencing this address space (i.e. its refcount is
 * zero), the address space can be destroyed.
 *
 * We know that we don't hold any spinlock.
 *
 * @param as        Address space to be destroyed.
 */
void as_destroy(as_t *as)
{
    ipl_t ipl;
    bool cond;
    DEADLOCK_PROBE_INIT(p_asidlock);
 
    ASSERT(atomic_get(&as->refcount) == 0);
   
    /*
     * Since there is no reference to this area,
     * it is safe not to lock its mutex.
     */
 
    /*
     * We need to avoid deadlock between TLB shootdown and asidlock.
     * We therefore try to take asid conditionally and if we don't succeed,
     * we enable interrupts and try again. This is done while preemption is
     * disabled to prevent nested context switches. We also depend on the
     * fact that so far no spinlocks are held.
     */
    preemption_disable();
    ipl = interrupts_read();
retry:
    interrupts_disable();
    if (!spinlock_trylock(&asidlock)) {
        interrupts_enable();
        DEADLOCK_PROBE(p_asidlock, DEADLOCK_THRESHOLD);
        goto retry;
    }
    preemption_enable();    /* Interrupts disabled, enable preemption */
    if (as->asid != ASID_INVALID && as != AS_KERNEL) {
        if (as != AS && as->cpu_refcount == 0)
            list_remove(&as->inactive_as_with_asid_link);
        asid_put(as->asid);
    }
    spinlock_unlock(&asidlock);
 
    /*
     * Destroy address space areas of the address space.
     * The B+tree must be walked carefully because it is
     * also being destroyed.
     */
    for (cond = true; cond; ) {
        btree_node_t *node;
 
        ASSERT(!list_empty(&as->as_area_btree.leaf_head));
        node = list_get_instance(as->as_area_btree.leaf_head.next,
            btree_node_t, leaf_link);
 
        if ((cond = node->keys)) {
            as_area_destroy(as, node->key[0]);
        }
    }
 
    btree_destroy(&as->as_area_btree);
#ifdef AS_PAGE_TABLE
    page_table_destroy(as->genarch.page_table);
#else
    page_table_destroy(NULL);
#endif
 
    interrupts_restore(ipl);
 
    slab_free(as_slab, as);
}
 
/** Create address space area of common attributes.
 *
 * The created address space area is added to the target address space.
 *
 * @param as        Target address space.
 * @param flags     Flags of the area memory.
 * @param size      Size of area.
 * @param base      Base address of area.
 * @param attrs     Attributes of the area.
 * @param backend   Address space area backend. NULL if no backend is used.
 * @param backend_data  NULL or a pointer to an array holding two void *.
 *
 * @return      Address space area on success or NULL on failure.
 */
as_area_t *
as_area_create(as_t *as, int flags, size_t size, uintptr_t base, int attrs,
    mem_backend_t *backend, mem_backend_data_t *backend_data)
{
    ipl_t ipl;
    as_area_t *a;
   
    if (base % PAGE_SIZE)
        return NULL;
 
    if (!size)
        return NULL;
 
    /* Writeable executable areas are not supported. */
    if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))
        return NULL;
   
    ipl = interrupts_disable();
    mutex_lock(&as->lock);
   
    if (!check_area_conflicts(as, base, size, NULL)) {
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return NULL;
    }
   
    a = (as_area_t *) malloc(sizeof(as_area_t), 0);
 
    mutex_initialize(&a->lock, MUTEX_PASSIVE);
   
    a->as = as;
    a->flags = flags;
    a->attributes = attrs;
    a->pages = SIZE2FRAMES(size);
    a->base = base;
    a->sh_info = NULL;
    a->backend = backend;
    if (backend_data)
        a->backend_data = *backend_data;
    else
        memsetb(&a->backend_data, sizeof(a->backend_data), 0);
 
    btree_create(&a->used_space);
   
    btree_insert(&as->as_area_btree, base, (void *) a, NULL);
 
    mutex_unlock(&as->lock);
    interrupts_restore(ipl);
 
    return a;
}
 
/** Find address space area and change it.
 *
 * @param as        Address space.
 * @param address   Virtual address belonging to the area to be changed.
 *          Must be page-aligned.
 * @param size      New size of the virtual memory block starting at
 *          address.
 * @param flags     Flags influencing the remap operation. Currently unused.
 *
 * @return      Zero on success or a value from @ref errno.h otherwise.
 */
int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)
{
    as_area_t *area;
    ipl_t ipl;
    size_t pages;
   
    ipl = interrupts_disable();
    mutex_lock(&as->lock);
   
    /*
     * Locate the area.
     */
    area = find_area_and_lock(as, address);
    if (!area) {
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return ENOENT;
    }
 
    if (area->backend == &phys_backend) {
        /*
         * Remapping of address space areas associated
         * with memory mapped devices is not supported.
         */
        mutex_unlock(&area->lock);
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return ENOTSUP;
    }
    if (area->sh_info) {
        /*
         * Remapping of shared address space areas
         * is not supported.
         */
        mutex_unlock(&area->lock);
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return ENOTSUP;
    }
 
    pages = SIZE2FRAMES((address - area->base) + size);
    if (!pages) {
        /*
         * Zero size address space areas are not allowed.
         */
        mutex_unlock(&area->lock);
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return EPERM;
    }
   
    if (pages < area->pages) {
        bool cond;
        uintptr_t start_free = area->base + pages * PAGE_SIZE;
 
        /*
         * Shrinking the area.
         * No need to check for overlaps.
         */
 
        /*
         * Start TLB shootdown sequence.
         */
        tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base +
            pages * PAGE_SIZE, area->pages - pages);
 
        /*
         * Remove frames belonging to used space starting from
         * the highest addresses downwards until an overlap with
         * the resized address space area is found. Note that this
         * is also the right way to remove part of the used_space
         * B+tree leaf list.
         */    
        for (cond = true; cond;) {
            btree_node_t *node;
       
            ASSERT(!list_empty(&area->used_space.leaf_head));
            node =
                list_get_instance(area->used_space.leaf_head.prev,
                btree_node_t, leaf_link);
            if ((cond = (bool) node->keys)) {
                uintptr_t b = node->key[node->keys - 1];
                count_t c =
                    (count_t) node->value[node->keys - 1];
                unsigned int i = 0;
           
                if (overlaps(b, c * PAGE_SIZE, area->base,
                    pages * PAGE_SIZE)) {
                   
                    if (b + c * PAGE_SIZE <= start_free) {
                        /*
                         * The whole interval fits
                         * completely in the resized
                         * address space area.
                         */
                        break;
                    }
       
                    /*
                     * Part of the interval corresponding
                     * to b and c overlaps with the resized
                     * address space area.
                     */
       
                    cond = false;   /* we are almost done */
                    i = (start_free - b) >> PAGE_WIDTH;
                    if (!used_space_remove(area, start_free,
                        c - i))
                        panic("Cannot remove used "
                            "space.");
                } else {
                    /*
                     * The interval of used space can be
                     * completely removed.
                     */
                    if (!used_space_remove(area, b, c))
                        panic("Cannot remove used "
                            "space.");
                }
           
                for (; i < c; i++) {
                    pte_t *pte;
           
                    page_table_lock(as, false);
                    pte = page_mapping_find(as, b +
                        i * PAGE_SIZE);
                    ASSERT(pte && PTE_VALID(pte) &&
                        PTE_PRESENT(pte));
                    if (area->backend &&
                        area->backend->frame_free) {
                        area->backend->frame_free(area,
                            b + i * PAGE_SIZE,
                            PTE_GET_FRAME(pte));
                    }
                    page_mapping_remove(as, b +
                        i * PAGE_SIZE);
                    page_table_unlock(as, false);
                }
            }
        }
 
        /*
         * Finish TLB shootdown sequence.
         */
 
        tlb_invalidate_pages(as->asid, area->base + pages * PAGE_SIZE,
            area->pages - pages);
        /*
         * Invalidate software translation caches (e.g. TSB on sparc64).
         */
        as_invalidate_translation_cache(as, area->base +
            pages * PAGE_SIZE, area->pages - pages);
        tlb_shootdown_finalize();
       
    } else {
        /*
         * Growing the area.
         * Check for overlaps with other address space areas.
         */
        if (!check_area_conflicts(as, address, pages * PAGE_SIZE,
            area)) {
            mutex_unlock(&area->lock);
            mutex_unlock(&as->lock);       
            interrupts_restore(ipl);
            return EADDRNOTAVAIL;
        }
    }
 
    area->pages = pages;
   
    mutex_unlock(&area->lock);
    mutex_unlock(&as->lock);
    interrupts_restore(ipl);
 
    return 0;
}
 
/** Destroy address space area.
 *
 * @param as        Address space.
 * @param address   Address within the area to be deleted.
 *
 * @return      Zero on success or a value from @ref errno.h on failure.
 */
int as_area_destroy(as_t *as, uintptr_t address)
{
    as_area_t *area;
    uintptr_t base;
    link_t *cur;
    ipl_t ipl;
 
    ipl = interrupts_disable();
    mutex_lock(&as->lock);
 
    area = find_area_and_lock(as, address);
    if (!area) {
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return ENOENT;
    }
 
    base = area->base;
 
    /*
     * Start TLB shootdown sequence.
     */
    tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
 
    /*
     * Visit only the pages mapped by used_space B+tree.
     */
    for (cur = area->used_space.leaf_head.next;
        cur != &area->used_space.leaf_head; cur = cur->next) {
        btree_node_t *node;
        unsigned int i;
       
        node = list_get_instance(cur, btree_node_t, leaf_link);
        for (i = 0; i < node->keys; i++) {
            uintptr_t b = node->key[i];
            count_t j;
            pte_t *pte;
           
            for (j = 0; j < (count_t) node->value[i]; j++) {
                page_table_lock(as, false);
                pte = page_mapping_find(as, b + j * PAGE_SIZE);
                ASSERT(pte && PTE_VALID(pte) &&
                    PTE_PRESENT(pte));
                if (area->backend &&
                    area->backend->frame_free) {
                    area->backend->frame_free(area, b +
                        j * PAGE_SIZE, PTE_GET_FRAME(pte));
                }
                page_mapping_remove(as, b + j * PAGE_SIZE);            
                page_table_unlock(as, false);
            }
        }
    }
 
    /*
     * Finish TLB shootdown sequence.
     */
 
    tlb_invalidate_pages(as->asid, area->base, area->pages);
    /*
     * Invalidate potential software translation caches (e.g. TSB on
     * sparc64).
     */
    as_invalidate_translation_cache(as, area->base, area->pages);
    tlb_shootdown_finalize();
   
    btree_destroy(&area->used_space);
 
    area->attributes |= AS_AREA_ATTR_PARTIAL;
   
    if (area->sh_info)
        sh_info_remove_reference(area->sh_info);
       
    mutex_unlock(&area->lock);
 
    /*
     * Remove the empty area from address space.
     */
    btree_remove(&as->as_area_btree, base, NULL);
   
    free(area);
   
    mutex_unlock(&as->lock);
    interrupts_restore(ipl);
    return 0;
}
 
/** Share address space area with another or the same address space.
 *
 * Address space area mapping is shared with a new address space area.
 * If the source address space area has not been shared so far,
 * a new sh_info is created. The new address space area simply gets the
 * sh_info of the source area. The process of duplicating the
 * mapping is done through the backend share function.
 *
 * @param src_as    Pointer to source address space.
 * @param src_base  Base address of the source address space area.
 * @param acc_size  Expected size of the source area.
 * @param dst_as    Pointer to destination address space.
 * @param dst_base  Target base address.
 * @param dst_flags_mask Destination address space area flags mask.
 *
 * @return      Zero on success or ENOENT if there is no such task or if
 *          there is no such address space area, EPERM if there was
 *          a problem in accepting the area or ENOMEM if there was a
 *          problem in allocating destination address space area.
 *          ENOTSUP is returned if the address space area backend
 *          does not support sharing.
 */
int as_area_share(as_t *src_as, uintptr_t src_base, size_t acc_size,
    as_t *dst_as, uintptr_t dst_base, int dst_flags_mask)
{
    ipl_t ipl;
    int src_flags;
    size_t src_size;
    as_area_t *src_area, *dst_area;
    share_info_t *sh_info;
    mem_backend_t *src_backend;
    mem_backend_data_t src_backend_data;
   
    ipl = interrupts_disable();
    mutex_lock(&src_as->lock);
    src_area = find_area_and_lock(src_as, src_base);
    if (!src_area) {
        /*
         * Could not find the source address space area.
         */
        mutex_unlock(&src_as->lock);
        interrupts_restore(ipl);
        return ENOENT;
    }
 
    if (!src_area->backend || !src_area->backend->share) {
        /*
         * There is no backend or the backend does not
         * know how to share the area.
         */
        mutex_unlock(&src_area->lock);
        mutex_unlock(&src_as->lock);
        interrupts_restore(ipl);
        return ENOTSUP;
    }
   
    src_size = src_area->pages * PAGE_SIZE;
    src_flags = src_area->flags;
    src_backend = src_area->backend;
    src_backend_data = src_area->backend_data;
 
    /* Share the cacheable flag from the original mapping */
    if (src_flags & AS_AREA_CACHEABLE)
        dst_flags_mask |= AS_AREA_CACHEABLE;
 
    if (src_size != acc_size ||
        (src_flags & dst_flags_mask) != dst_flags_mask) {
        mutex_unlock(&src_area->lock);
        mutex_unlock(&src_as->lock);
        interrupts_restore(ipl);
        return EPERM;
    }
 
    /*
     * Now we are committed to sharing the area.
     * First, prepare the area for sharing.
     * Then it will be safe to unlock it.
     */
    sh_info = src_area->sh_info;
    if (!sh_info) {
        sh_info = (share_info_t *) malloc(sizeof(share_info_t), 0);
        mutex_initialize(&sh_info->lock, MUTEX_PASSIVE);
        sh_info->refcount = 2;
        btree_create(&sh_info->pagemap);
        src_area->sh_info = sh_info;
        /*
         * Call the backend to setup sharing.
         */
        src_area->backend->share(src_area);
    } else {
        mutex_lock(&sh_info->lock);
        sh_info->refcount++;
        mutex_unlock(&sh_info->lock);
    }
 
    mutex_unlock(&src_area->lock);
    mutex_unlock(&src_as->lock);
 
    /*
     * Create copy of the source address space area.
     * The destination area is created with AS_AREA_ATTR_PARTIAL
     * attribute set which prevents race condition with
     * preliminary as_page_fault() calls.
     * The flags of the source area are masked against dst_flags_mask
     * to support sharing in less privileged mode.
     */
    dst_area = as_area_create(dst_as, dst_flags_mask, src_size, dst_base,
        AS_AREA_ATTR_PARTIAL, src_backend, &src_backend_data);
    if (!dst_area) {
        /*
         * Destination address space area could not be created.
         */
        sh_info_remove_reference(sh_info);
       
        interrupts_restore(ipl);
        return ENOMEM;
    }
 
    /*
     * Now the destination address space area has been
     * fully initialized. Clear the AS_AREA_ATTR_PARTIAL
     * attribute and set the sh_info.
     */
    mutex_lock(&dst_as->lock); 
    mutex_lock(&dst_area->lock);
    dst_area->attributes &= ~AS_AREA_ATTR_PARTIAL;
    dst_area->sh_info = sh_info;
    mutex_unlock(&dst_area->lock);
    mutex_unlock(&dst_as->lock);   
 
    interrupts_restore(ipl);
   
    return 0;
}
 
/** Check access mode for address space area.
 *
 * The address space area must be locked prior to this call.
 *
 * @param area      Address space area.
 * @param access    Access mode.
 *
 * @return      False if access violates area's permissions, true
 *          otherwise.
 */
bool as_area_check_access(as_area_t *area, pf_access_t access)
{
    int flagmap[] = {
        [PF_ACCESS_READ] = AS_AREA_READ,
        [PF_ACCESS_WRITE] = AS_AREA_WRITE,
        [PF_ACCESS_EXEC] = AS_AREA_EXEC
    };
 
    if (!(area->flags & flagmap[access]))
        return false;
   
    return true;
}
 
/** Change adress space area flags.
 *
 * The idea is to have the same data, but with a different access mode.
 * This is needed e.g. for writing code into memory and then executing it.
 * In order for this to work properly, this may copy the data
 * into private anonymous memory (unless it's already there).
 *
 * @param as      Address space.
 * @param flags   Flags of the area memory.
 * @param address Address within the area to be changed.
 *
 * @return Zero on success or a value from @ref errno.h on failure.
 *
 */
int as_area_change_flags(as_t *as, int flags, uintptr_t address)
{
    as_area_t *area;
    uintptr_t base;
    link_t *cur;
    ipl_t ipl;
    int page_flags;
    uintptr_t *old_frame;
    index_t frame_idx;
    count_t used_pages;
   
    /* Flags for the new memory mapping */
    page_flags = area_flags_to_page_flags(flags);
 
    ipl = interrupts_disable();
    mutex_lock(&as->lock);
 
    area = find_area_and_lock(as, address);
    if (!area) {
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return ENOENT;
    }
 
    if ((area->sh_info) || (area->backend != &anon_backend)) {
        /* Copying shared areas not supported yet */
        /* Copying non-anonymous memory not supported yet */
        mutex_unlock(&area->lock);
        mutex_unlock(&as->lock);
        interrupts_restore(ipl);
        return ENOTSUP;
    }
 
    base = area->base;
 
    /*
     * Compute total number of used pages in the used_space B+tree
     */
    used_pages = 0;
 
    for (cur = area->used_space.leaf_head.next;
        cur != &area->used_space.leaf_head; cur = cur->next) {
        btree_node_t *node;
        unsigned int i;
       
        node = list_get_instance(cur, btree_node_t, leaf_link);
        for (i = 0; i < node->keys; i++) {
            used_pages += (count_t) node->value[i];
        }
    }
 
    /* An array for storing frame numbers */
    old_frame = malloc(used_pages * sizeof(uintptr_t), 0);
 
    /*
     * Start TLB shootdown sequence.
     */
    tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
 
    /*
     * Remove used pages from page tables and remember their frame
     * numbers.
     */
    frame_idx = 0;
 
    for (cur = area->used_space.leaf_head.next;
        cur != &area->used_space.leaf_head; cur = cur->next) {
        btree_node_t *node;
        unsigned int i;
       
        node = list_get_instance(cur, btree_node_t, leaf_link);
        for (i = 0; i < node->keys; i++) {
            uintptr_t b = node->key[i];
            count_t j;
            pte_t *pte;
           
            for (j = 0; j < (count_t) node->value[i]; j++) {
                page_table_lock(as, false);
                pte = page_mapping_find(as, b + j * PAGE_SIZE);
                ASSERT(pte && PTE_VALID(pte) &&
                    PTE_PRESENT(pte));
                old_frame[frame_idx++] = PTE_GET_FRAME(pte);
 
                /* Remove old mapping */
                page_mapping_remove(as, b + j * PAGE_SIZE);
                page_table_unlock(as, false);
            }
        }
    }
 
    /*
     * Finish TLB shootdown sequence.
     */
 
    tlb_invalidate_pages(as->asid, area->base, area->pages);
   
    /*
     * Invalidate potential software translation caches (e.g. TSB on
     * sparc64).
     */
    as_invalidate_translation_cache(as, area->base, area->pages);
    tlb_shootdown_finalize();
 
    /*
     * Set the new flags.
     */
    area->flags = flags;
 
    /*
     * Map pages back in with new flags. This step is kept separate
     * so that the memory area could not be accesed with both the old and
     * the new flags at once.
     */
    frame_idx = 0;
 
    for (cur = area->used_space.leaf_head.next;
        cur != &area->used_space.leaf_head; cur = cur->next) {
        btree_node_t *node;
        unsigned int i;
       
        node = list_get_instance(cur, btree_node_t, leaf_link);
        for (i = 0; i < node->keys; i++) {
            uintptr_t b = node->key[i];
            count_t j;
           
            for (j = 0; j < (count_t) node->value[i]; j++) {
                page_table_lock(as, false);
 
                /* Insert the new mapping */
                page_mapping_insert(as, b + j * PAGE_SIZE,
                    old_frame[frame_idx++], page_flags);
 
                page_table_unlock(as, false);
            }
        }
    }
 
    free(old_frame);
 
    mutex_unlock(&area->lock);
    mutex_unlock(&as->lock);
    interrupts_restore(ipl);
 
    return 0;
}
 
 
/** Handle page fault within the current address space.
 *
 * This is the high-level page fault handler. It decides whether the page fault
 * can be resolved by any backend and if so, it invokes the backend to resolve
 * the page fault.
 *
 * Interrupts are assumed disabled.
 *
 * @param page      Faulting page.
 * @param access    Access mode that caused the page fault (i.e.
 *          read/write/exec).
 * @param istate    Pointer to the interrupted state.
 *
 * @return      AS_PF_FAULT on page fault, AS_PF_OK on success or
 *          AS_PF_DEFER if the fault was caused by copy_to_uspace()
 *          or copy_from_uspace().
 */
int as_page_fault(uintptr_t page, pf_access_t access, istate_t *istate)
{
    pte_t *pte;
    as_area_t *area;
   
    if (!THREAD)
        return AS_PF_FAULT;
       
    ASSERT(AS);
 
    mutex_lock(&AS->lock);
    area = find_area_and_lock(AS, page);   
    if (!area) {
        /*
         * No area contained mapping for 'page'.
         * Signal page fault to low-level handler.
         */
        mutex_unlock(&AS->lock);
        goto page_fault;
    }
 
    if (area->attributes & AS_AREA_ATTR_PARTIAL) {
        /*
         * The address space area is not fully initialized.
         * Avoid possible race by returning error.
         */
        mutex_unlock(&area->lock);
        mutex_unlock(&AS->lock);
        goto page_fault;       
    }
 
    if (!area->backend || !area->backend->page_fault) {
        /*
         * The address space area is not backed by any backend
         * or the backend cannot handle page faults.
         */
        mutex_unlock(&area->lock);
        mutex_unlock(&AS->lock);
        goto page_fault;       
    }
 
    page_table_lock(AS, false);
   
    /*
     * To avoid race condition between two page faults on the same address,
     * we need to make sure the mapping has not been already inserted.
     */
    if ((pte = page_mapping_find(AS, page))) {
        if (PTE_PRESENT(pte)) {
            if (((access == PF_ACCESS_READ) && PTE_READABLE(pte)) ||
                (access == PF_ACCESS_WRITE && PTE_WRITABLE(pte)) ||
                (access == PF_ACCESS_EXEC && PTE_EXECUTABLE(pte))) {
                page_table_unlock(AS, false);
                mutex_unlock(&area->lock);
                mutex_unlock(&AS->lock);
                return AS_PF_OK;
            }
        }
    }
   
    /*
     * Resort to the backend page fault handler.
     */
    if (area->backend->page_fault(area, page, access) != AS_PF_OK) {
        page_table_unlock(AS, false);
        mutex_unlock(&area->lock);
        mutex_unlock(&AS->lock);
        goto page_fault;
    }
   
    page_table_unlock(AS, false);
    mutex_unlock(&area->lock);
    mutex_unlock(&AS->lock);
    return AS_PF_OK;
 
page_fault:
    if (THREAD->in_copy_from_uspace) {
        THREAD->in_copy_from_uspace = false;
        istate_set_retaddr(istate,
            (uintptr_t) &memcpy_from_uspace_failover_address);
    } else if (THREAD->in_copy_to_uspace) {
        THREAD->in_copy_to_uspace = false;
        istate_set_retaddr(istate,
            (uintptr_t) &memcpy_to_uspace_failover_address);
    } else {
        return AS_PF_FAULT;
    }
 
    return AS_PF_DEFER;
}
 
/** Switch address spaces.
 *
 * Note that this function cannot sleep as it is essentially a part of
 * scheduling. Sleeping here would lead to deadlock on wakeup. Another
 * thing which is forbidden in this context is locking the address space.
 *
 * When this function is enetered, no spinlocks may be held.
 *
 * @param old       Old address space or NULL.
 * @param new       New address space.
 */
void as_switch(as_t *old_as, as_t *new_as)
{
    DEADLOCK_PROBE_INIT(p_asidlock);
    preemption_disable();
retry:
    (void) interrupts_disable();
    if (!spinlock_trylock(&asidlock)) {
        /*
         * Avoid deadlock with TLB shootdown.
         * We can enable interrupts here because
         * preemption is disabled. We should not be
         * holding any other lock.
         */
        (void) interrupts_enable();
        DEADLOCK_PROBE(p_asidlock, DEADLOCK_THRESHOLD);
        goto retry;
    }
    preemption_enable();
 
    /*
     * First, take care of the old address space.
     */
    if (old_as) {
        ASSERT(old_as->cpu_refcount);
        if((--old_as->cpu_refcount == 0) && (old_as != AS_KERNEL)) {
            /*
             * The old address space is no longer active on
             * any processor. It can be appended to the
             * list of inactive address spaces with assigned
             * ASID.
             */
            ASSERT(old_as->asid != ASID_INVALID);
            list_append(&old_as->inactive_as_with_asid_link,
                &inactive_as_with_asid_head);
        }
 
        /*
         * Perform architecture-specific tasks when the address space
         * is being removed from the CPU.
         */
        as_deinstall_arch(old_as);
    }
 
    /*
     * Second, prepare the new address space.
     */
    if ((new_as->cpu_refcount++ == 0) && (new_as != AS_KERNEL)) {
        if (new_as->asid != ASID_INVALID)
            list_remove(&new_as->inactive_as_with_asid_link);
        else
            new_as->asid = asid_get();
    }
#ifdef AS_PAGE_TABLE
    SET_PTL0_ADDRESS(new_as->genarch.page_table);
#endif
   
    /*
     * Perform architecture-specific steps.
     * (e.g. write ASID to hardware register etc.)
     */
    as_install_arch(new_as);
 
    spinlock_unlock(&asidlock);
   
    AS = new_as;
}
 
/** Convert address space area flags to page flags.
 *
 * @param aflags    Flags of some address space area.
 *
 * @return      Flags to be passed to page_mapping_insert().
 */
int area_flags_to_page_flags(int aflags)
{
    int flags;
 
    flags = PAGE_USER | PAGE_PRESENT;
   
    if (aflags & AS_AREA_READ)
        flags |= PAGE_READ;
       
    if (aflags & AS_AREA_WRITE)
        flags |= PAGE_WRITE;
   
    if (aflags & AS_AREA_EXEC)
        flags |= PAGE_EXEC;
   
    if (aflags & AS_AREA_CACHEABLE)
        flags |= PAGE_CACHEABLE;
       
    return flags;
}
 
/** Compute flags for virtual address translation subsytem.
 *
 * The address space area must be locked.
 * Interrupts must be disabled.
 *
 * @param a     Address space area.
 *
 * @return      Flags to be used in page_mapping_insert().
 */
int as_area_get_flags(as_area_t *a)
{
    return area_flags_to_page_flags(a->flags);
}
 
/** Create page table.
 *
 * Depending on architecture, create either address space private or global page
 * table.
 *
 * @param flags     Flags saying whether the page table is for the kernel
 *          address space.
 *
 * @return      First entry of the page table.
 */
pte_t *page_table_create(int flags)
{
    ASSERT(as_operations);
    ASSERT(as_operations->page_table_create);
   
    return as_operations->page_table_create(flags);
}
 
/** Destroy page table.
 *
 * Destroy page table in architecture specific way.
 *
 * @param page_table    Physical address of PTL0.
 */
void page_table_destroy(pte_t *page_table)
{
    ASSERT(as_operations);
    ASSERT(as_operations->page_table_destroy);
   
    as_operations->page_table_destroy(page_table);
}
 
/** Lock page table.
 *
 * This function should be called before any page_mapping_insert(),
 * page_mapping_remove() and page_mapping_find().
 *
 * Locking order is such that address space areas must be locked
 * prior to this call. Address space can be locked prior to this
 * call in which case the lock argument is false.
 *
 * @param as        Address space.
 * @param lock      If false, do not attempt to lock as->lock.
 */
void page_table_lock(as_t *as, bool lock)
{
    ASSERT(as_operations);
    ASSERT(as_operations->page_table_lock);
   
    as_operations->page_table_lock(as, lock);
}
 
/** Unlock page table.
 *
 * @param as        Address space.
 * @param unlock    If false, do not attempt to unlock as->lock.
 */
void page_table_unlock(as_t *as, bool unlock)
{
    ASSERT(as_operations);
    ASSERT(as_operations->page_table_unlock);
   
    as_operations->page_table_unlock(as, unlock);
}
 
 
/** Find address space area and lock it.
 *
 * The address space must be locked and interrupts must be disabled.
 *
 * @param as        Address space.
 * @param va        Virtual address.
 *
 * @return      Locked address space area containing va on success or
 *          NULL on failure.
 */
as_area_t *find_area_and_lock(as_t *as, uintptr_t va)
{
    as_area_t *a;
    btree_node_t *leaf, *lnode;
    unsigned int i;
   
    a = (as_area_t *) btree_search(&as->as_area_btree, va, &leaf);
    if (a) {
        /* va is the base address of an address space area */
        mutex_lock(&a->lock);
        return a;
    }
   
    /*
     * Search the leaf node and the righmost record of its left neighbour
     * to find out whether this is a miss or va belongs to an address
     * space area found there.
     */
   
    /* First, search the leaf node itself. */
    for (i = 0; i < leaf->keys; i++) {
        a = (as_area_t *) leaf->value[i];
        mutex_lock(&a->lock);
        if ((a->base <= va) && (va < a->base + a->pages * PAGE_SIZE)) {
            return a;
        }
        mutex_unlock(&a->lock);
    }
 
    /*
     * Second, locate the left neighbour and test its last record.
     * Because of its position in the B+tree, it must have base < va.
     */
    lnode = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf);
    if (lnode) {
        a = (as_area_t *) lnode->value[lnode->keys - 1];
        mutex_lock(&a->lock);
        if (va < a->base + a->pages * PAGE_SIZE) {
            return a;
        }
        mutex_unlock(&a->lock);
    }
 
    return NULL;
}
 
/** Check area conflicts with other areas.
 *
 * The address space must be locked and interrupts must be disabled.
 *
 * @param as        Address space.
 * @param va        Starting virtual address of the area being tested.
 * @param size      Size of the area being tested.
 * @param avoid_area    Do not touch this area.
 *
 * @return      True if there is no conflict, false otherwise.
 */
bool
check_area_conflicts(as_t *as, uintptr_t va, size_t size, as_area_t *avoid_area)
{
    as_area_t *a;
    btree_node_t *leaf, *node;
    unsigned int i;
   
    /*
     * We don't want any area to have conflicts with NULL page.
     */
    if (overlaps(va, size, NULL, PAGE_SIZE))
        return false;
   
    /*
     * The leaf node is found in O(log n), where n is proportional to
     * the number of address space areas belonging to as.
     * The check for conflicts is then attempted on the rightmost
     * record in the left neighbour, the leftmost record in the right
     * neighbour and all records in the leaf node itself.
     */
   
    if ((a = (as_area_t *) btree_search(&as->as_area_btree, va, &leaf))) {
        if (a != avoid_area)
            return false;
    }
   
    /* First, check the two border cases. */
    if ((node = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf))) {
        a = (as_area_t *) node->value[node->keys - 1];
        mutex_lock(&a->lock);
        if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
            mutex_unlock(&a->lock);
            return false;
        }
        mutex_unlock(&a->lock);
    }
    node = btree_leaf_node_right_neighbour(&as->as_area_btree, leaf);
    if (node) {
        a = (as_area_t *) node->value[0];
        mutex_lock(&a->lock);
        if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
            mutex_unlock(&a->lock);
            return false;
        }
        mutex_unlock(&a->lock);
    }
   
    /* Second, check the leaf node. */
    for (i = 0; i < leaf->keys; i++) {
        a = (as_area_t *) leaf->value[i];
   
        if (a == avoid_area)
            continue;
   
        mutex_lock(&a->lock);
        if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
            mutex_unlock(&a->lock);
            return false;
        }
        mutex_unlock(&a->lock);
    }
 
    /*
     * So far, the area does not conflict with other areas.
     * Check if it doesn't conflict with kernel address space.
     */  
    if (!KERNEL_ADDRESS_SPACE_SHADOWED) {
        return !overlaps(va, size,
            KERNEL_ADDRESS_SPACE_START,
            KERNEL_ADDRESS_SPACE_END - KERNEL_ADDRESS_SPACE_START);
    }
 
    return true;
}
 
/** Return size of the address space area with given base.
 *
 * @param base      Arbitrary address insede the address space area.
 *
 * @return      Size of the address space area in bytes or zero if it
 *          does not exist.
 */
size_t as_area_get_size(uintptr_t base)
{
    ipl_t ipl;
    as_area_t *src_area;
    size_t size;
 
    ipl = interrupts_disable();
    src_area = find_area_and_lock(AS, base);
    if (src_area) {
        size = src_area->pages * PAGE_SIZE;
        mutex_unlock(&src_area->lock);
    } else {
        size = 0;
    }
    interrupts_restore(ipl);
    return size;
}
 
/** Mark portion of address space area as used.
 *
 * The address space area must be already locked.
 *
 * @param a     Address space area.
 * @param page      First page to be marked.
 * @param count     Number of page to be marked.
 *
 * @return      Zero on failure and non-zero on success.
 */
int used_space_insert(as_area_t *a, uintptr_t page, count_t count)
{
    btree_node_t *leaf, *node;
    count_t pages;
    unsigned int i;
 
    ASSERT(page == ALIGN_DOWN(page, PAGE_SIZE));
    ASSERT(count);
 
    pages = (count_t) btree_search(&a->used_space, page, &leaf);
    if (pages) {
        /*
         * We hit the beginning of some used space.
         */
        return 0;
    }
 
    if (!leaf->keys) {
        btree_insert(&a->used_space, page, (void *) count, leaf);
        return 1;
    }
 
    node = btree_leaf_node_left_neighbour(&a->used_space, leaf);
    if (node) {
        uintptr_t left_pg = node->key[node->keys - 1];
        uintptr_t right_pg = leaf->key[0];
        count_t left_cnt = (count_t) node->value[node->keys - 1];
        count_t right_cnt = (count_t) leaf->value[0];
       
        /*
         * Examine the possibility that the interval fits
         * somewhere between the rightmost interval of
         * the left neigbour and the first interval of the leaf.
         */
         
        if (page >= right_pg) {
            /* Do nothing. */
        } else if (overlaps(page, count * PAGE_SIZE, left_pg,
            left_cnt * PAGE_SIZE)) {
            /* The interval intersects with the left interval. */
            return 0;
        } else if (overlaps(page, count * PAGE_SIZE, right_pg,
            right_cnt * PAGE_SIZE)) {
            /* The interval intersects with the right interval. */
            return 0;          
        } else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
            (page + count * PAGE_SIZE == right_pg)) {
            /*
             * The interval can be added by merging the two already
             * present intervals.
             */
            node->value[node->keys - 1] += count + right_cnt;
            btree_remove(&a->used_space, right_pg, leaf);
            return 1;
        } else if (page == left_pg + left_cnt * PAGE_SIZE) {
            /*
             * The interval can be added by simply growing the left
             * interval.
             */
            node->value[node->keys - 1] += count;
            return 1;
        } else if (page + count * PAGE_SIZE == right_pg) {
            /*
             * The interval can be addded by simply moving base of
             * the right interval down and increasing its size
             * accordingly.
             */
            leaf->value[0] += count;
            leaf->key[0] = page;
            return 1;
        } else {
            /*
             * The interval is between both neigbouring intervals,
             * but cannot be merged with any of them.
             */
            btree_insert(&a->used_space, page, (void *) count,
                leaf);
            return 1;
        }
    } else if (page < leaf->key[0]) {
        uintptr_t right_pg = leaf->key[0];
        count_t right_cnt = (count_t) leaf->value[0];
   
        /*
         * Investigate the border case in which the left neighbour does
         * not exist but the interval fits from the left.
         */
         
        if (overlaps(page, count * PAGE_SIZE, right_pg,
            right_cnt * PAGE_SIZE)) {
            /* The interval intersects with the right interval. */
            return 0;
        } else if (page + count * PAGE_SIZE == right_pg) {
            /*
             * The interval can be added by moving the base of the
             * right interval down and increasing its size
             * accordingly.
             */
            leaf->key[0] = page;
            leaf->value[0] += count;
            return 1;
        } else {
            /*
             * The interval doesn't adjoin with the right interval.
             * It must be added individually.
             */
            btree_insert(&a->used_space, page, (void *) count,
                leaf);
            return 1;
        }
    }
 
    node = btree_leaf_node_right_neighbour(&a->used_space, leaf);
    if (node) {
        uintptr_t left_pg = leaf->key[leaf->keys - 1];
        uintptr_t right_pg = node->key[0];
        count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
        count_t right_cnt = (count_t) node->value[0];
       
        /*
         * Examine the possibility that the interval fits
         * somewhere between the leftmost interval of
         * the right neigbour and the last interval of the leaf.
         */
 
        if (page < left_pg) {
            /* Do nothing. */
        } else if (overlaps(page, count * PAGE_SIZE, left_pg,
            left_cnt * PAGE_SIZE)) {
            /* The interval intersects with the left interval. */
            return 0;
        } else if (overlaps(page, count * PAGE_SIZE, right_pg,
            right_cnt * PAGE_SIZE)) {
            /* The interval intersects with the right interval. */
            return 0;          
        } else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
            (page + count * PAGE_SIZE == right_pg)) {
            /*
             * The interval can be added by merging the two already
             * present intervals.
             * */
            leaf->value[leaf->keys - 1] += count + right_cnt;
            btree_remove(&a->used_space, right_pg, node);
            return 1;
        } else if (page == left_pg + left_cnt * PAGE_SIZE) {
            /*
             * The interval can be added by simply growing the left
             * interval.
             * */
            leaf->value[leaf->keys - 1] +=  count;
            return 1;
        } else if (page + count * PAGE_SIZE == right_pg) {
            /*
             * The interval can be addded by simply moving base of
             * the right interval down and increasing its size
             * accordingly.
             */
            node->value[0] += count;
            node->key[0] = page;
            return 1;
        } else {
            /*
             * The interval is between both neigbouring intervals,
             * but cannot be merged with any of them.
             */
            btree_insert(&a->used_space, page, (void *) count,
                leaf);
            return 1;
        }
    } else if (page >= leaf->key[leaf->keys - 1]) {
        uintptr_t left_pg = leaf->key[leaf->keys - 1];
        count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
   
        /*
         * Investigate the border case in which the right neighbour
         * does not exist but the interval fits from the right.
         */
         
        if (overlaps(page, count * PAGE_SIZE, left_pg,
            left_cnt * PAGE_SIZE)) {
            /* The interval intersects with the left interval. */
            return 0;
        } else if (left_pg + left_cnt * PAGE_SIZE == page) {
            /*
             * The interval can be added by growing the left
             * interval.
             */
            leaf->value[leaf->keys - 1] += count;
            return 1;
        } else {
            /*
             * The interval doesn't adjoin with the left interval.
             * It must be added individually.
             */
            btree_insert(&a->used_space, page, (void *) count,
                leaf);
            return 1;
        }
    }
   
    /*
     * Note that if the algorithm made it thus far, the interval can fit
     * only between two other intervals of the leaf. The two border cases
     * were already resolved.
     */
    for (i = 1; i < leaf->keys; i++) {
        if (page < leaf->key[i]) {
            uintptr_t left_pg = leaf->key[i - 1];
            uintptr_t right_pg = leaf->key[i];
            count_t left_cnt = (count_t) leaf->value[i - 1];
            count_t right_cnt = (count_t) leaf->value[i];
 
            /*
             * The interval fits between left_pg and right_pg.
             */
 
            if (overlaps(page, count * PAGE_SIZE, left_pg,
                left_cnt * PAGE_SIZE)) {
                /*
                 * The interval intersects with the left
                 * interval.
                 */
                return 0;
            } else if (overlaps(page, count * PAGE_SIZE, right_pg,
                right_cnt * PAGE_SIZE)) {
                /*
                 * The interval intersects with the right
                 * interval.
                 */
                return 0;          
            } else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
                (page + count * PAGE_SIZE == right_pg)) {
                /*
                 * The interval can be added by merging the two
                 * already present intervals.
                 */
                leaf->value[i - 1] += count + right_cnt;
                btree_remove(&a->used_space, right_pg, leaf);
                return 1;
            } else if (page == left_pg + left_cnt * PAGE_SIZE) {
                /*
                 * The interval can be added by simply growing
                 * the left interval.
                 */
                leaf->value[i - 1] += count;
                return 1;
            } else if (page + count * PAGE_SIZE == right_pg) {
                /*
                     * The interval can be addded by simply moving
                 * base of the right interval down and
                 * increasing its size accordingly.
                 */
                leaf->value[i] += count;
                leaf->key[i] = page;
                return 1;
            } else {
                /*
                 * The interval is between both neigbouring
                 * intervals, but cannot be merged with any of
                 * them.
                 */
                btree_insert(&a->used_space, page,
                    (void *) count, leaf);
                return 1;
            }
        }
    }
 
    panic("Inconsistency detected while adding %" PRIc " pages of used "
        "space at %p.", count, page);
}
 
/** Mark portion of address space area as unused.
 *
 * The address space area must be already locked.
 *
 * @param a     Address space area.
 * @param page      First page to be marked.
 * @param count     Number of page to be marked.
 *
 * @return      Zero on failure and non-zero on success.
 */
int used_space_remove(as_area_t *a, uintptr_t page, count_t count)
{
    btree_node_t *leaf, *node;
    count_t pages;
    unsigned int i;
 
    ASSERT(page == ALIGN_DOWN(page, PAGE_SIZE));
    ASSERT(count);
 
    pages = (count_t) btree_search(&a->used_space, page, &leaf);
    if (pages) {
        /*
         * We are lucky, page is the beginning of some interval.
         */
        if (count > pages) {
            return 0;
        } else if (count == pages) {
            btree_remove(&a->used_space, page, leaf);
            return 1;
        } else {
            /*
             * Find the respective interval.
             * Decrease its size and relocate its start address.
             */
            for (i = 0; i < leaf->keys; i++) {
                if (leaf->key[i] == page) {
                    leaf->key[i] += count * PAGE_SIZE;
                    leaf->value[i] -= count;
                    return 1;
                }
            }
            goto error;
        }
    }
 
    node = btree_leaf_node_left_neighbour(&a->used_space, leaf);
    if (node && page < leaf->key[0]) {
        uintptr_t left_pg = node->key[node->keys - 1];
        count_t left_cnt = (count_t) node->value[node->keys - 1];
 
        if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
            count * PAGE_SIZE)) {
            if (page + count * PAGE_SIZE ==
                left_pg + left_cnt * PAGE_SIZE) {
                /*
                 * The interval is contained in the rightmost
                 * interval of the left neighbour and can be
                 * removed by updating the size of the bigger
                 * interval.
                 */
                node->value[node->keys - 1] -= count;
                return 1;
            } else if (page + count * PAGE_SIZE <
                left_pg + left_cnt*PAGE_SIZE) {
                count_t new_cnt;
               
                /*
                 * The interval is contained in the rightmost
                 * interval of the left neighbour but its
                 * removal requires both updating the size of
                 * the original interval and also inserting a
                 * new interval.
                 */
                new_cnt = ((left_pg + left_cnt * PAGE_SIZE) -
                    (page + count*PAGE_SIZE)) >> PAGE_WIDTH;
                node->value[node->keys - 1] -= count + new_cnt;
                btree_insert(&a->used_space, page +
                    count * PAGE_SIZE, (void *) new_cnt, leaf);
                return 1;
            }
        }
        return 0;
    } else if (page < leaf->key[0]) {
        return 0;
    }
   
    if (page > leaf->key[leaf->keys - 1]) {
        uintptr_t left_pg = leaf->key[leaf->keys - 1];
        count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
 
        if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
            count * PAGE_SIZE)) {
            if (page + count * PAGE_SIZE ==
                left_pg + left_cnt * PAGE_SIZE) {
                /*
                 * The interval is contained in the rightmost
                 * interval of the leaf and can be removed by
                 * updating the size of the bigger interval.
                 */
                leaf->value[leaf->keys - 1] -= count;
                return 1;
            } else if (page + count * PAGE_SIZE < left_pg +
                left_cnt * PAGE_SIZE) {
                count_t new_cnt;
               
                /*
                 * The interval is contained in the rightmost
                 * interval of the leaf but its removal
                 * requires both updating the size of the
                 * original interval and also inserting a new
                 * interval.
                 */
                new_cnt = ((left_pg + left_cnt * PAGE_SIZE) -
                    (page + count * PAGE_SIZE)) >> PAGE_WIDTH;
                leaf->value[leaf->keys - 1] -= count + new_cnt;
                btree_insert(&a->used_space, page +
                    count * PAGE_SIZE, (void *) new_cnt, leaf);
                return 1;
            }
        }
        return 0;
    }  
   
    /*
     * The border cases have been already resolved.
     * Now the interval can be only between intervals of the leaf.
     */
    for (i = 1; i < leaf->keys - 1; i++) {
        if (page < leaf->key[i]) {
            uintptr_t left_pg = leaf->key[i - 1];
            count_t left_cnt = (count_t) leaf->value[i - 1];
 
            /*
             * Now the interval is between intervals corresponding
             * to (i - 1) and i.
             */
            if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
                count * PAGE_SIZE)) {
                if (page + count * PAGE_SIZE ==
                    left_pg + left_cnt*PAGE_SIZE) {
                    /*
                     * The interval is contained in the
                     * interval (i - 1) of the leaf and can
                     * be removed by updating the size of
                     * the bigger interval.
                     */
                    leaf->value[i - 1] -= count;
                    return 1;
                } else if (page + count * PAGE_SIZE <
                    left_pg + left_cnt * PAGE_SIZE) {
                    count_t new_cnt;
               
                    /*
                     * The interval is contained in the
                     * interval (i - 1) of the leaf but its
                     * removal requires both updating the
                     * size of the original interval and
                     * also inserting a new interval.
                     */
                    new_cnt = ((left_pg +
                        left_cnt * PAGE_SIZE) -
                        (page + count * PAGE_SIZE)) >>
                        PAGE_WIDTH;
                    leaf->value[i - 1] -= count + new_cnt;
                    btree_insert(&a->used_space, page +
                        count * PAGE_SIZE, (void *) new_cnt,
                        leaf);
                    return 1;
                }
            }
            return 0;
        }
    }
 
error:
    panic("Inconsistency detected while removing %" PRIc " pages of used "
        "space from %p.", count, page);
}
 
/** Remove reference to address space area share info.
 *
 * If the reference count drops to 0, the sh_info is deallocated.
 *
 * @param sh_info   Pointer to address space area share info.
 */
void sh_info_remove_reference(share_info_t *sh_info)
{
    bool dealloc = false;
 
    mutex_lock(&sh_info->lock);
    ASSERT(sh_info->refcount);
    if (--sh_info->refcount == 0) {
        dealloc = true;
        link_t *cur;
       
        /*
         * Now walk carefully the pagemap B+tree and free/remove
         * reference from all frames found there.
         */
        for (cur = sh_info->pagemap.leaf_head.next;
            cur != &sh_info->pagemap.leaf_head; cur = cur->next) {
            btree_node_t *node;
            unsigned int i;
           
            node = list_get_instance(cur, btree_node_t, leaf_link);
            for (i = 0; i < node->keys; i++)
                frame_free((uintptr_t) node->value[i]);
        }
       
    }
    mutex_unlock(&sh_info->lock);
   
    if (dealloc) {
        btree_destroy(&sh_info->pagemap);
        free(sh_info);
    }
}
 
/*
 * Address space related syscalls.
 */
 
/** Wrapper for as_area_create(). */
unative_t sys_as_area_create(uintptr_t address, size_t size, int flags)
{
    if (as_area_create(AS, flags | AS_AREA_CACHEABLE, size, address,
        AS_AREA_ATTR_NONE, &anon_backend, NULL))
        return (unative_t) address;
    else
        return (unative_t) -1;
}
 
/** Wrapper for as_area_resize(). */
unative_t sys_as_area_resize(uintptr_t address, size_t size, int flags)
{
    return (unative_t) as_area_resize(AS, address, size, 0);
}
 
/** Wrapper for as_area_change_flags(). */
unative_t sys_as_area_change_flags(uintptr_t address, int flags)
{
    return (unative_t) as_area_change_flags(AS, flags, address);
}
 
/** Wrapper for as_area_destroy(). */
unative_t sys_as_area_destroy(uintptr_t address)
{
    return (unative_t) as_area_destroy(AS, address);
}
 
/** Print out information about address space.
 *
 * @param as        Address space.
 */
void as_print(as_t *as)
{
    ipl_t ipl;
   
    ipl = interrupts_disable();
    mutex_lock(&as->lock);
   
    /* print out info about address space areas */
    link_t *cur;
    for (cur = as->as_area_btree.leaf_head.next;
        cur != &as->as_area_btree.leaf_head; cur = cur->next) {
        btree_node_t *node;
       
        node = list_get_instance(cur, btree_node_t, leaf_link);
       
        unsigned int i;
        for (i = 0; i < node->keys; i++) {
            as_area_t *area = node->value[i];
       
            mutex_lock(&area->lock);
            printf("as_area: %p, base=%p, pages=%" PRIc
                " (%p - %p)\n", area, area->base, area->pages,
                area->base, area->base + FRAMES2SIZE(area->pages));
            mutex_unlock(&area->lock);
        }
    }
   
    mutex_unlock(&as->lock);
    interrupts_restore(ipl);
}
 
/** @}
 */
 

Rev 3707	Rev 3790
1	/*	1	/*
2	* Copyright (c) 2001-2006 Jakub Jermar	2	* Copyright (c) 2001-2006 Jakub Jermar
3	* All rights reserved.	3	* All rights reserved.
4	*	4	*
5	* Redistribution and use in source and binary forms, with or without	5	* Redistribution and use in source and binary forms, with or without
6	* modification, are permitted provided that the following conditions	6	* modification, are permitted provided that the following conditions
7	* are met:	7	* are met:
8	*	8	*
9	* - Redistributions of source code must retain the above copyright	9	* - Redistributions of source code must retain the above copyright
10	* notice, this list of conditions and the following disclaimer.	10	* notice, this list of conditions and the following disclaimer.
11	* - Redistributions in binary form must reproduce the above copyright	11	* - Redistributions in binary form must reproduce the above copyright
12	* notice, this list of conditions and the following disclaimer in the	12	* notice, this list of conditions and the following disclaimer in the
13	* documentation and/or other materials provided with the distribution.	13	* documentation and/or other materials provided with the distribution.
14	* - The name of the author may not be used to endorse or promote products	14	* - The name of the author may not be used to endorse or promote products
15	* derived from this software without specific prior written permission.	15	* derived from this software without specific prior written permission.
16	*	16	*
17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR	17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18	* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES	18	* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19	* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.	19	* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20	* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,	20	* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21	* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT	21	* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22	* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,	22	* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23	* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY	23	* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24	* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT	24	* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25	* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF	25	* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26	* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.	26	* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27	*/	27	*/
28		28
29	/** @addtogroup genericmm	29	/** @addtogroup genericmm
30	* @{	30	* @{
31	*/	31	*/
32		32
33	/**	33	/**
34	* @file	34	* @file
35	* @brief Address space related functions.	35	* @brief Address space related functions.
36	*	36	*
37	* This file contains address space manipulation functions.	37	* This file contains address space manipulation functions.
38	* Roughly speaking, this is a higher-level client of	38	* Roughly speaking, this is a higher-level client of
39	* Virtual Address Translation (VAT) subsystem.	39	* Virtual Address Translation (VAT) subsystem.
40	*	40	*
41	* Functionality provided by this file allows one to	41	* Functionality provided by this file allows one to
42	* create address spaces and create, resize and share	42	* create address spaces and create, resize and share
43	* address space areas.	43	* address space areas.
44	*	44	*
45	* @see page.c	45	* @see page.c
46	*	46	*
47	*/	47	*/
48		48
49	#include <mm/as.h>	49	#include <mm/as.h>
50	#include <arch/mm/as.h>	50	#include <arch/mm/as.h>
51	#include <mm/page.h>	51	#include <mm/page.h>
52	#include <mm/frame.h>	52	#include <mm/frame.h>
53	#include <mm/slab.h>	53	#include <mm/slab.h>
54	#include <mm/tlb.h>	54	#include <mm/tlb.h>
55	#include <arch/mm/page.h>	55	#include <arch/mm/page.h>
56	#include <genarch/mm/page_pt.h>	56	#include <genarch/mm/page_pt.h>
57	#include <genarch/mm/page_ht.h>	57	#include <genarch/mm/page_ht.h>
58	#include <mm/asid.h>	58	#include <mm/asid.h>
59	#include <arch/mm/asid.h>	59	#include <arch/mm/asid.h>
60	#include <preemption.h>	60	#include <preemption.h>
61	#include <synch/spinlock.h>	61	#include <synch/spinlock.h>
62	#include <synch/mutex.h>	62	#include <synch/mutex.h>
63	#include <adt/list.h>	63	#include <adt/list.h>
64	#include <adt/btree.h>	64	#include <adt/btree.h>
65	#include <proc/task.h>	65	#include <proc/task.h>
66	#include <proc/thread.h>	66	#include <proc/thread.h>
67	#include <arch/asm.h>	67	#include <arch/asm.h>
68	#include <panic.h>	68	#include <panic.h>
69	#include <debug.h>	69	#include <debug.h>
70	#include <print.h>	70	#include <print.h>
71	#include <memstr.h>	71	#include <memstr.h>
72	#include <macros.h>	72	#include <macros.h>
73	#include <arch.h>	73	#include <arch.h>
74	#include <errno.h>	74	#include <errno.h>
75	#include <config.h>	75	#include <config.h>
76	#include <align.h>	76	#include <align.h>
77	#include <arch/types.h>	77	#include <arch/types.h>
78	#include <syscall/copy.h>	78	#include <syscall/copy.h>
79	#include <arch/interrupt.h>	79	#include <arch/interrupt.h>
80		80
81	#ifdef CONFIG_VIRT_IDX_DCACHE	81	#ifdef CONFIG_VIRT_IDX_DCACHE
82	#include <arch/mm/cache.h>	82	#include <arch/mm/cache.h>
83	#endif /* CONFIG_VIRT_IDX_DCACHE */	83	#endif /* CONFIG_VIRT_IDX_DCACHE */
84		84
85	/**	85	/**
86	* Each architecture decides what functions will be used to carry out	86	* Each architecture decides what functions will be used to carry out
87	* address space operations such as creating or locking page tables.	87	* address space operations such as creating or locking page tables.
88	*/	88	*/
89	as_operations_t *as_operations = NULL;	89	as_operations_t *as_operations = NULL;
90		90
91	/**	91	/**
92	* Slab for as_t objects.	92	* Slab for as_t objects.
93	*/	93	*/
94	static slab_cache_t *as_slab;	94	static slab_cache_t *as_slab;
95		95
96	/**	96	/**
97	* This lock serializes access to the ASID subsystem.	97	* This lock serializes access to the ASID subsystem.
98	* It protects:	98	* It protects:
99	* - inactive_as_with_asid_head list	99	* - inactive_as_with_asid_head list
100	* - as->asid for each as of the as_t type	100	* - as->asid for each as of the as_t type
101	* - asids_allocated counter	101	* - asids_allocated counter
102	*/	102	*/
103	SPINLOCK_INITIALIZE(asidlock);	103	SPINLOCK_INITIALIZE(asidlock);
104		104
105	/**	105	/**
106	* This list contains address spaces that are not active on any	106	* This list contains address spaces that are not active on any
107	* processor and that have valid ASID.	107	* processor and that have valid ASID.
108	*/	108	*/
109	LIST_INITIALIZE(inactive_as_with_asid_head);	109	LIST_INITIALIZE(inactive_as_with_asid_head);
110		110
111	/** Kernel address space. */	111	/** Kernel address space. */
112	as_t *AS_KERNEL = NULL;	112	as_t *AS_KERNEL = NULL;
113		113
114	static int area_flags_to_page_flags(int);	114	static int area_flags_to_page_flags(int);
115	static as_area_t find_area_and_lock(as_t , uintptr_t);	115	static as_area_t find_area_and_lock(as_t , uintptr_t);
116	static bool check_area_conflicts(as_t , uintptr_t, size_t, as_area_t );	116	static bool check_area_conflicts(as_t , uintptr_t, size_t, as_area_t );
117	static void sh_info_remove_reference(share_info_t *);	117	static void sh_info_remove_reference(share_info_t *);
118		118
119	static int as_constructor(void *obj, int flags)	119	static int as_constructor(void *obj, int flags)
120	{	120	{
121	as_t as = (as_t ) obj;	121	as_t as = (as_t ) obj;
122	int rc;	122	int rc;
123		123
124	link_initialize(&as->inactive_as_with_asid_link);	124	link_initialize(&as->inactive_as_with_asid_link);
125	mutex_initialize(&as->lock, MUTEX_PASSIVE);	125	mutex_initialize(&as->lock, MUTEX_PASSIVE);
126		126
127	rc = as_constructor_arch(as, flags);	127	rc = as_constructor_arch(as, flags);
128		128
129	return rc;	129	return rc;
130	}	130	}
131		131
132	static int as_destructor(void *obj)	132	static int as_destructor(void *obj)
133	{	133	{
134	as_t as = (as_t ) obj;	134	as_t as = (as_t ) obj;
135		135
136	return as_destructor_arch(as);	136	return as_destructor_arch(as);
137	}	137	}
138		138
139	/** Initialize address space subsystem. */	139	/** Initialize address space subsystem. */
140	void as_init(void)	140	void as_init(void)
141	{	141	{
142	as_arch_init();	142	as_arch_init();
143		143
144	as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,	144	as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,
145	as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);	145	as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);
146		146
147	AS_KERNEL = as_create(FLAG_AS_KERNEL);	147	AS_KERNEL = as_create(FLAG_AS_KERNEL);
148	if (!AS_KERNEL)	148	if (!AS_KERNEL)
149	panic("Cannot create kernel address space\n");	149	panic("Cannot create kernel address space.");
150		150
151	/* Make sure the kernel address space	151	/* Make sure the kernel address space
152	* reference count never drops to zero.	152	* reference count never drops to zero.
153	*/	153	*/
154	atomic_set(&AS_KERNEL->refcount, 1);	154	atomic_set(&AS_KERNEL->refcount, 1);
155	}	155	}
156		156
157	/** Create address space.	157	/** Create address space.
158	*	158	*
159	* @param flags Flags that influence the way in wich the address space	159	* @param flags Flags that influence the way in wich the address space
160	* is created.	160	* is created.
161	*/	161	*/
162	as_t *as_create(int flags)	162	as_t *as_create(int flags)
163	{	163	{
164	as_t *as;	164	as_t *as;
165		165
166	as = (as_t *) slab_alloc(as_slab, 0);	166	as = (as_t *) slab_alloc(as_slab, 0);
167	(void) as_create_arch(as, 0);	167	(void) as_create_arch(as, 0);
168		168
169	btree_create(&as->as_area_btree);	169	btree_create(&as->as_area_btree);
170		170
171	if (flags & FLAG_AS_KERNEL)	171	if (flags & FLAG_AS_KERNEL)
172	as->asid = ASID_KERNEL;	172	as->asid = ASID_KERNEL;
173	else	173	else
174	as->asid = ASID_INVALID;	174	as->asid = ASID_INVALID;
175		175
176	atomic_set(&as->refcount, 0);	176	atomic_set(&as->refcount, 0);
177	as->cpu_refcount = 0;	177	as->cpu_refcount = 0;
178	#ifdef AS_PAGE_TABLE	178	#ifdef AS_PAGE_TABLE
179	as->genarch.page_table = page_table_create(flags);	179	as->genarch.page_table = page_table_create(flags);
180	#else	180	#else
181	page_table_create(flags);	181	page_table_create(flags);
182	#endif	182	#endif
183		183
184	return as;	184	return as;
185	}	185	}
186		186
187	/** Destroy adress space.	187	/** Destroy adress space.
188	*	188	*
189	* When there are no tasks referencing this address space (i.e. its refcount is	189	* When there are no tasks referencing this address space (i.e. its refcount is
190	* zero), the address space can be destroyed.	190	* zero), the address space can be destroyed.
191	*	191	*
192	* We know that we don't hold any spinlock.	192	* We know that we don't hold any spinlock.
193	*	193	*
194	* @param as Address space to be destroyed.	194	* @param as Address space to be destroyed.
195	*/	195	*/
196	void as_destroy(as_t *as)	196	void as_destroy(as_t *as)
197	{	197	{
198	ipl_t ipl;	198	ipl_t ipl;
199	bool cond;	199	bool cond;
200	DEADLOCK_PROBE_INIT(p_asidlock);	200	DEADLOCK_PROBE_INIT(p_asidlock);
201		201
202	ASSERT(atomic_get(&as->refcount) == 0);	202	ASSERT(atomic_get(&as->refcount) == 0);
203		203
204	/*	204	/*
205	* Since there is no reference to this area,	205	* Since there is no reference to this area,
206	* it is safe not to lock its mutex.	206	* it is safe not to lock its mutex.
207	*/	207	*/
208		208
209	/*	209	/*
210	* We need to avoid deadlock between TLB shootdown and asidlock.	210	* We need to avoid deadlock between TLB shootdown and asidlock.
211	* We therefore try to take asid conditionally and if we don't succeed,	211	* We therefore try to take asid conditionally and if we don't succeed,
212	* we enable interrupts and try again. This is done while preemption is	212	* we enable interrupts and try again. This is done while preemption is
213	* disabled to prevent nested context switches. We also depend on the	213	* disabled to prevent nested context switches. We also depend on the
214	* fact that so far no spinlocks are held.	214	* fact that so far no spinlocks are held.
215	*/	215	*/
216	preemption_disable();	216	preemption_disable();
217	ipl = interrupts_read();	217	ipl = interrupts_read();
218	retry:	218	retry:
219	interrupts_disable();	219	interrupts_disable();
220	if (!spinlock_trylock(&asidlock)) {	220	if (!spinlock_trylock(&asidlock)) {
221	interrupts_enable();	221	interrupts_enable();
222	DEADLOCK_PROBE(p_asidlock, DEADLOCK_THRESHOLD);	222	DEADLOCK_PROBE(p_asidlock, DEADLOCK_THRESHOLD);
223	goto retry;	223	goto retry;
224	}	224	}
225	preemption_enable(); /* Interrupts disabled, enable preemption */	225	preemption_enable(); /* Interrupts disabled, enable preemption */
226	if (as->asid != ASID_INVALID && as != AS_KERNEL) {	226	if (as->asid != ASID_INVALID && as != AS_KERNEL) {
227	if (as != AS && as->cpu_refcount == 0)	227	if (as != AS && as->cpu_refcount == 0)
228	list_remove(&as->inactive_as_with_asid_link);	228	list_remove(&as->inactive_as_with_asid_link);
229	asid_put(as->asid);	229	asid_put(as->asid);
230	}	230	}
231	spinlock_unlock(&asidlock);	231	spinlock_unlock(&asidlock);
232		232
233	/*	233	/*
234	* Destroy address space areas of the address space.	234	* Destroy address space areas of the address space.
235	* The B+tree must be walked carefully because it is	235	* The B+tree must be walked carefully because it is
236	* also being destroyed.	236	* also being destroyed.
237	*/	237	*/
238	for (cond = true; cond; ) {	238	for (cond = true; cond; ) {
239	btree_node_t *node;	239	btree_node_t *node;
240		240
241	ASSERT(!list_empty(&as->as_area_btree.leaf_head));	241	ASSERT(!list_empty(&as->as_area_btree.leaf_head));
242	node = list_get_instance(as->as_area_btree.leaf_head.next,	242	node = list_get_instance(as->as_area_btree.leaf_head.next,
243	btree_node_t, leaf_link);	243	btree_node_t, leaf_link);
244		244
245	if ((cond = node->keys)) {	245	if ((cond = node->keys)) {
246	as_area_destroy(as, node->key[0]);	246	as_area_destroy(as, node->key[0]);
247	}	247	}
248	}	248	}
249		249
250	btree_destroy(&as->as_area_btree);	250	btree_destroy(&as->as_area_btree);
251	#ifdef AS_PAGE_TABLE	251	#ifdef AS_PAGE_TABLE
252	page_table_destroy(as->genarch.page_table);	252	page_table_destroy(as->genarch.page_table);
253	#else	253	#else
254	page_table_destroy(NULL);	254	page_table_destroy(NULL);
255	#endif	255	#endif
256		256
257	interrupts_restore(ipl);	257	interrupts_restore(ipl);
258		258
259	slab_free(as_slab, as);	259	slab_free(as_slab, as);
260	}	260	}
261		261
262	/** Create address space area of common attributes.	262	/** Create address space area of common attributes.
263	*	263	*
264	* The created address space area is added to the target address space.	264	* The created address space area is added to the target address space.
265	*	265	*
266	* @param as Target address space.	266	* @param as Target address space.
267	* @param flags Flags of the area memory.	267	* @param flags Flags of the area memory.
268	* @param size Size of area.	268	* @param size Size of area.
269	* @param base Base address of area.	269	* @param base Base address of area.
270	* @param attrs Attributes of the area.	270	* @param attrs Attributes of the area.
271	* @param backend Address space area backend. NULL if no backend is used.	271	* @param backend Address space area backend. NULL if no backend is used.
272	* @param backend_data NULL or a pointer to an array holding two void *.	272	* @param backend_data NULL or a pointer to an array holding two void *.
273	*	273	*
274	* @return Address space area on success or NULL on failure.	274	* @return Address space area on success or NULL on failure.
275	*/	275	*/
276	as_area_t *	276	as_area_t *
277	as_area_create(as_t *as, int flags, size_t size, uintptr_t base, int attrs,	277	as_area_create(as_t *as, int flags, size_t size, uintptr_t base, int attrs,
278	mem_backend_t backend, mem_backend_data_t backend_data)	278	mem_backend_t backend, mem_backend_data_t backend_data)
279	{	279	{
280	ipl_t ipl;	280	ipl_t ipl;
281	as_area_t *a;	281	as_area_t *a;
282		282
283	if (base % PAGE_SIZE)	283	if (base % PAGE_SIZE)
284	return NULL;	284	return NULL;
285		285
286	if (!size)	286	if (!size)
287	return NULL;	287	return NULL;
288		288
289	/* Writeable executable areas are not supported. */	289	/* Writeable executable areas are not supported. */
290	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))	290	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))
291	return NULL;	291	return NULL;
292		292
293	ipl = interrupts_disable();	293	ipl = interrupts_disable();
294	mutex_lock(&as->lock);	294	mutex_lock(&as->lock);
295		295
296	if (!check_area_conflicts(as, base, size, NULL)) {	296	if (!check_area_conflicts(as, base, size, NULL)) {
297	mutex_unlock(&as->lock);	297	mutex_unlock(&as->lock);
298	interrupts_restore(ipl);	298	interrupts_restore(ipl);
299	return NULL;	299	return NULL;
300	}	300	}
301		301
302	a = (as_area_t *) malloc(sizeof(as_area_t), 0);	302	a = (as_area_t *) malloc(sizeof(as_area_t), 0);
303		303
304	mutex_initialize(&a->lock, MUTEX_PASSIVE);	304	mutex_initialize(&a->lock, MUTEX_PASSIVE);
305		305
306	a->as = as;	306	a->as = as;
307	a->flags = flags;	307	a->flags = flags;
308	a->attributes = attrs;	308	a->attributes = attrs;
309	a->pages = SIZE2FRAMES(size);	309	a->pages = SIZE2FRAMES(size);
310	a->base = base;	310	a->base = base;
311	a->sh_info = NULL;	311	a->sh_info = NULL;
312	a->backend = backend;	312	a->backend = backend;
313	if (backend_data)	313	if (backend_data)
314	a->backend_data = *backend_data;	314	a->backend_data = *backend_data;
315	else	315	else
316	memsetb(&a->backend_data, sizeof(a->backend_data), 0);	316	memsetb(&a->backend_data, sizeof(a->backend_data), 0);
317		317
318	btree_create(&a->used_space);	318	btree_create(&a->used_space);
319		319
320	btree_insert(&as->as_area_btree, base, (void *) a, NULL);	320	btree_insert(&as->as_area_btree, base, (void *) a, NULL);
321		321
322	mutex_unlock(&as->lock);	322	mutex_unlock(&as->lock);
323	interrupts_restore(ipl);	323	interrupts_restore(ipl);
324		324
325	return a;	325	return a;
326	}	326	}
327		327
328	/** Find address space area and change it.	328	/** Find address space area and change it.
329	*	329	*
330	* @param as Address space.	330	* @param as Address space.
331	* @param address Virtual address belonging to the area to be changed.	331	* @param address Virtual address belonging to the area to be changed.
332	* Must be page-aligned.	332	* Must be page-aligned.
333	* @param size New size of the virtual memory block starting at	333	* @param size New size of the virtual memory block starting at
334	* address.	334	* address.
335	* @param flags Flags influencing the remap operation. Currently unused.	335	* @param flags Flags influencing the remap operation. Currently unused.
336	*	336	*
337	* @return Zero on success or a value from @ref errno.h otherwise.	337	* @return Zero on success or a value from @ref errno.h otherwise.
338	*/	338	*/
339	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)	339	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)
340	{	340	{
341	as_area_t *area;	341	as_area_t *area;
342	ipl_t ipl;	342	ipl_t ipl;
343	size_t pages;	343	size_t pages;
344		344
345	ipl = interrupts_disable();	345	ipl = interrupts_disable();
346	mutex_lock(&as->lock);	346	mutex_lock(&as->lock);
347		347
348	/*	348	/*
349	* Locate the area.	349	* Locate the area.
350	*/	350	*/
351	area = find_area_and_lock(as, address);	351	area = find_area_and_lock(as, address);
352	if (!area) {	352	if (!area) {
353	mutex_unlock(&as->lock);	353	mutex_unlock(&as->lock);
354	interrupts_restore(ipl);	354	interrupts_restore(ipl);
355	return ENOENT;	355	return ENOENT;
356	}	356	}
357		357
358	if (area->backend == &phys_backend) {	358	if (area->backend == &phys_backend) {
359	/*	359	/*
360	* Remapping of address space areas associated	360	* Remapping of address space areas associated
361	* with memory mapped devices is not supported.	361	* with memory mapped devices is not supported.
362	*/	362	*/
363	mutex_unlock(&area->lock);	363	mutex_unlock(&area->lock);
364	mutex_unlock(&as->lock);	364	mutex_unlock(&as->lock);
365	interrupts_restore(ipl);	365	interrupts_restore(ipl);
366	return ENOTSUP;	366	return ENOTSUP;
367	}	367	}
368	if (area->sh_info) {	368	if (area->sh_info) {
369	/*	369	/*
370	* Remapping of shared address space areas	370	* Remapping of shared address space areas
371	* is not supported.	371	* is not supported.
372	*/	372	*/
373	mutex_unlock(&area->lock);	373	mutex_unlock(&area->lock);
374	mutex_unlock(&as->lock);	374	mutex_unlock(&as->lock);
375	interrupts_restore(ipl);	375	interrupts_restore(ipl);
376	return ENOTSUP;	376	return ENOTSUP;
377	}	377	}
378		378
379	pages = SIZE2FRAMES((address - area->base) + size);	379	pages = SIZE2FRAMES((address - area->base) + size);
380	if (!pages) {	380	if (!pages) {
381	/*	381	/*
382	* Zero size address space areas are not allowed.	382	* Zero size address space areas are not allowed.
383	*/	383	*/
384	mutex_unlock(&area->lock);	384	mutex_unlock(&area->lock);
385	mutex_unlock(&as->lock);	385	mutex_unlock(&as->lock);
386	interrupts_restore(ipl);	386	interrupts_restore(ipl);
387	return EPERM;	387	return EPERM;
388	}	388	}
389		389
390	if (pages < area->pages) {	390	if (pages < area->pages) {
391	bool cond;	391	bool cond;
392	uintptr_t start_free = area->base + pages * PAGE_SIZE;	392	uintptr_t start_free = area->base + pages * PAGE_SIZE;
393		393
394	/*	394	/*
395	* Shrinking the area.	395	* Shrinking the area.
396	* No need to check for overlaps.	396	* No need to check for overlaps.
397	*/	397	*/
398		398
399	/*	399	/*
400	* Start TLB shootdown sequence.	400	* Start TLB shootdown sequence.
401	*/	401	*/
402	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base +	402	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base +
403	pages * PAGE_SIZE, area->pages - pages);	403	pages * PAGE_SIZE, area->pages - pages);
404		404
405	/*	405	/*
406	* Remove frames belonging to used space starting from	406	* Remove frames belonging to used space starting from
407	* the highest addresses downwards until an overlap with	407	* the highest addresses downwards until an overlap with
408	* the resized address space area is found. Note that this	408	* the resized address space area is found. Note that this
409	* is also the right way to remove part of the used_space	409	* is also the right way to remove part of the used_space
410	* B+tree leaf list.	410	* B+tree leaf list.
411	*/	411	*/
412	for (cond = true; cond;) {	412	for (cond = true; cond;) {
413	btree_node_t *node;	413	btree_node_t *node;
414		414
415	ASSERT(!list_empty(&area->used_space.leaf_head));	415	ASSERT(!list_empty(&area->used_space.leaf_head));
416	node =	416	node =
417	list_get_instance(area->used_space.leaf_head.prev,	417	list_get_instance(area->used_space.leaf_head.prev,
418	btree_node_t, leaf_link);	418	btree_node_t, leaf_link);
419	if ((cond = (bool) node->keys)) {	419	if ((cond = (bool) node->keys)) {
420	uintptr_t b = node->key[node->keys - 1];	420	uintptr_t b = node->key[node->keys - 1];
421	count_t c =	421	count_t c =
422	(count_t) node->value[node->keys - 1];	422	(count_t) node->value[node->keys - 1];
423	unsigned int i = 0;	423	unsigned int i = 0;
424		424
425	if (overlaps(b, c * PAGE_SIZE, area->base,	425	if (overlaps(b, c * PAGE_SIZE, area->base,
426	pages * PAGE_SIZE)) {	426	pages * PAGE_SIZE)) {
427		427
428	if (b + c * PAGE_SIZE <= start_free) {	428	if (b + c * PAGE_SIZE <= start_free) {
429	/*	429	/*
430	* The whole interval fits	430	* The whole interval fits
431	* completely in the resized	431	* completely in the resized
432	* address space area.	432	* address space area.
433	*/	433	*/
434	break;	434	break;
435	}	435	}
436		436
437	/*	437	/*
438	* Part of the interval corresponding	438	* Part of the interval corresponding
439	* to b and c overlaps with the resized	439	* to b and c overlaps with the resized
440	* address space area.	440	* address space area.
441	*/	441	*/
442		442
443	cond = false; /* we are almost done */	443	cond = false; /* we are almost done */
444	i = (start_free - b) >> PAGE_WIDTH;	444	i = (start_free - b) >> PAGE_WIDTH;
445	if (!used_space_remove(area, start_free,	445	if (!used_space_remove(area, start_free,
446	c - i))	446	c - i))
447	panic("Could not remove used "	447	panic("Cannot remove used "
448	"space.\n");	448	"space.");
449	} else {	449	} else {
450	/*	450	/*
451	* The interval of used space can be	451	* The interval of used space can be
452	* completely removed.	452	* completely removed.
453	*/	453	*/
454	if (!used_space_remove(area, b, c))	454	if (!used_space_remove(area, b, c))
455	panic("Could not remove used "	455	panic("Cannot remove used "
456	"space.\n");	456	"space.");
457	}	457	}
458		458
459	for (; i < c; i++) {	459	for (; i < c; i++) {
460	pte_t *pte;	460	pte_t *pte;
461		461
462	page_table_lock(as, false);	462	page_table_lock(as, false);
463	pte = page_mapping_find(as, b +	463	pte = page_mapping_find(as, b +
464	i * PAGE_SIZE);	464	i * PAGE_SIZE);
465	ASSERT(pte && PTE_VALID(pte) &&	465	ASSERT(pte && PTE_VALID(pte) &&
466	PTE_PRESENT(pte));	466	PTE_PRESENT(pte));
467	if (area->backend &&	467	if (area->backend &&
468	area->backend->frame_free) {	468	area->backend->frame_free) {
469	area->backend->frame_free(area,	469	area->backend->frame_free(area,
470	b + i * PAGE_SIZE,	470	b + i * PAGE_SIZE,
471	PTE_GET_FRAME(pte));	471	PTE_GET_FRAME(pte));
472	}	472	}
473	page_mapping_remove(as, b +	473	page_mapping_remove(as, b +
474	i * PAGE_SIZE);	474	i * PAGE_SIZE);
475	page_table_unlock(as, false);	475	page_table_unlock(as, false);
476	}	476	}
477	}	477	}
478	}	478	}
479		479
480	/*	480	/*
481	* Finish TLB shootdown sequence.	481	* Finish TLB shootdown sequence.
482	*/	482	*/
483		483
484	tlb_invalidate_pages(as->asid, area->base + pages * PAGE_SIZE,	484	tlb_invalidate_pages(as->asid, area->base + pages * PAGE_SIZE,
485	area->pages - pages);	485	area->pages - pages);
486	/*	486	/*
487	* Invalidate software translation caches (e.g. TSB on sparc64).	487	* Invalidate software translation caches (e.g. TSB on sparc64).
488	*/	488	*/
489	as_invalidate_translation_cache(as, area->base +	489	as_invalidate_translation_cache(as, area->base +
490	pages * PAGE_SIZE, area->pages - pages);	490	pages * PAGE_SIZE, area->pages - pages);
491	tlb_shootdown_finalize();	491	tlb_shootdown_finalize();
492		492
493	} else {	493	} else {
494	/*	494	/*
495	* Growing the area.	495	* Growing the area.
496	* Check for overlaps with other address space areas.	496	* Check for overlaps with other address space areas.
497	*/	497	*/
498	if (!check_area_conflicts(as, address, pages * PAGE_SIZE,	498	if (!check_area_conflicts(as, address, pages * PAGE_SIZE,
499	area)) {	499	area)) {
500	mutex_unlock(&area->lock);	500	mutex_unlock(&area->lock);

Subversion Repositories HelenOS

(root)/trunk/kernel/generic/src/mm/as.c – Rev 3707 → 3790