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 <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 <typedefs.h>
#include <syscall/copy.h>
#include <arch/interrupt.h>
 
/**
 * 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 protects inactive_as_with_asid_head list. It must be acquired before as_t mutex. */
SPINLOCK_INITIALIZE(inactive_as_with_asid_lock);
 
/**
 * 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 aflags);
static as_area_t *find_area_and_lock(as_t *as, uintptr_t va);
static bool check_area_conflicts(as_t *as, uintptr_t va, size_t size, as_area_t *avoid_area);
static void sh_info_remove_reference(share_info_t *sh_info);
 
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);   
   
    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("can't create kernel address space\n");
   
}
 
/** Create address space.
 *
 * @param flags Flags that influence 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;
   
    as->refcount = 0;
    as->cpu_refcount = 0;
    as->page_table = page_table_create(flags);
 
    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.
 */
void as_destroy(as_t *as)
{
    ipl_t ipl;
    bool cond;
 
    ASSERT(as->refcount == 0);
   
    /*
     * Since there is no reference to this area,
     * it is safe not to lock its mutex.
     */
    ipl = interrupts_disable();
    spinlock_lock(&inactive_as_with_asid_lock);
    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(&inactive_as_with_asid_lock);
 
    /*
     * 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);
    page_table_destroy(as->page_table);
 
    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);
   
    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((uintptr_t) &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];
                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("Could not remove used space.\n");
                } else {
                    /*
                     * The interval of used space can be completely removed.
                     */
                    if (!used_space_remove(area, b, c))
                        panic("Could not remove used space.\n");
                }
           
                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);
        tlb_shootdown_finalize();
       
        /*
         * Invalidate software translation caches (e.g. TSB on sparc64).
         */
        as_invalidate_translation_cache(as, area->base + pages*PAGE_SIZE, area->pages - pages);
    } 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 withing 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;
        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);
    tlb_shootdown_finalize();
   
    /*
     * Invalidate potential software translation caches (e.g. TSB on sparc64).
     */
    as_invalidate_translation_cache(as, area->base, area->pages);
   
    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. It can be also returned if the architecture uses virtually indexed
 * caches and the source and destination areas start at pages with different
 * page colors.
 */
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 0   /* disable the check for now */
#ifdef CONFIG_VIRT_IDX_CACHE
    if (PAGE_COLOR(src_area->base) != PAGE_COLOR(dst_base)) {
        /*
         * Refuse to create illegal address alias.
         */
        mutex_unlock(&src_area->lock);
        mutex_unlock(&src_as->lock);
        interrupts_restore(ipl);
        return ENOTSUP;
    }
#endif /* CONFIG_VIRT_IDX_CACHE */
#endif
 
    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);
        sh_info->refcount = 2;
        btree_create(&sh_info->pagemap);
        src_area->sh_info = sh_info;
    } else {
        mutex_lock(&sh_info->lock);
        sh_info->refcount++;
        mutex_unlock(&sh_info->lock);
    }
 
    src_area->backend->share(src_area);
 
    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_area->lock);
    dst_area->attributes &= ~AS_AREA_ATTR_PARTIAL;
    dst_area->sh_info = sh_info;
    mutex_unlock(&dst_area->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;
}
 
/** 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 fault (i.e. read/write/exec).
 * @param istate Pointer to 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.
 *
 * @param old Old address space or NULL.
 * @param new New address space.
 */
void as_switch(as_t *old, as_t *new)
{
    ipl_t ipl;
    bool needs_asid = false;
   
    ipl = interrupts_disable();
    spinlock_lock(&inactive_as_with_asid_lock);
 
    /*
     * First, take care of the old address space.
     */
    if (old) {
        mutex_lock_active(&old->lock);
        ASSERT(old->cpu_refcount);
        if((--old->cpu_refcount == 0) && (old != 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->asid != ASID_INVALID);
             list_append(&old->inactive_as_with_asid_link, &inactive_as_with_asid_head);
        }
        mutex_unlock(&old->lock);
 
        /*
         * Perform architecture-specific tasks when the address space
         * is being removed from the CPU.
         */
        as_deinstall_arch(old);
    }
 
    /*
     * Second, prepare the new address space.
     */
    mutex_lock_active(&new->lock);
    if ((new->cpu_refcount++ == 0) && (new != AS_KERNEL)) {
        if (new->asid != ASID_INVALID)
            list_remove(&new->inactive_as_with_asid_link);
        else
            needs_asid = true;  /* defer call to asid_get() until new->lock is released */
    }
    SET_PTL0_ADDRESS(new->page_table);
    mutex_unlock(&new->lock);
 
    if (needs_asid) {
        /*
         * Allocation of new ASID was deferred
         * until now in order to avoid deadlock.
         */
        asid_t asid;
       
        asid = asid_get();
        mutex_lock_active(&new->lock);
        new->asid = asid;
        mutex_unlock(&new->lock);
    }
    spinlock_unlock(&inactive_as_with_asid_lock);
    interrupts_restore(ipl);
   
    /*
     * Perform architecture-specific steps.
     * (e.g. write ASID to hardware register etc.)
     */
    as_install_arch(new);
   
    AS = new;
}
 
/** 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 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;
    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.
     */
    if ((lnode = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf))) {
        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;
    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);
    }
    if ((node = btree_leaf_node_right_neighbour(&as->as_area_btree, leaf))) {
        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.  */
size_t as_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 0 on failure and 1 on success.
 */
int used_space_insert(as_area_t *a, uintptr_t page, count_t count)
{
    btree_node_t *leaf, *node;
    count_t pages;
    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], right_pg = leaf->key[0];
        count_t left_cnt = (count_t) node->value[node->keys - 1], 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], right_pg = node->key[0];
        count_t left_cnt = (count_t) leaf->value[leaf->keys - 1], 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], right_pg = leaf->key[i];
            count_t left_cnt = (count_t) leaf->value[i - 1], 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 %d pages of used space at %p.\n", 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 0 on failure and 1 on success.
 */
int used_space_remove(as_area_t *a, uintptr_t page, count_t count)
{
    btree_node_t *leaf, *node;
    count_t pages;
    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 %d pages of used space from %p.\n", 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;
            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_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 = list_get_instance(cur, btree_node_t, leaf_link);
       
        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=%d (%p - %p)\n",
                area, area->base, area->pages, area->base, area->base + area->pages*PAGE_SIZE);
            mutex_unlock(&area->lock);
        }
    }
   
    mutex_unlock(&as->lock);
    interrupts_restore(ipl);
}
 
/** @}
 */
 

Rev 1954	Rev 2007
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 <synch/spinlock.h>	60	#include <synch/spinlock.h>
61	#include <synch/mutex.h>	61	#include <synch/mutex.h>
62	#include <adt/list.h>	62	#include <adt/list.h>
63	#include <adt/btree.h>	63	#include <adt/btree.h>
64	#include <proc/task.h>	64	#include <proc/task.h>
65	#include <proc/thread.h>	65	#include <proc/thread.h>
66	#include <arch/asm.h>	66	#include <arch/asm.h>
67	#include <panic.h>	67	#include <panic.h>
68	#include <debug.h>	68	#include <debug.h>
69	#include <print.h>	69	#include <print.h>
70	#include <memstr.h>	70	#include <memstr.h>
71	#include <macros.h>	71	#include <macros.h>
72	#include <arch.h>	72	#include <arch.h>
73	#include <errno.h>	73	#include <errno.h>
74	#include <config.h>	74	#include <config.h>
75	#include <align.h>	75	#include <align.h>
76	#include <arch/types.h>	76	#include <arch/types.h>
77	#include <typedefs.h>	77	#include <typedefs.h>
78	#include <syscall/copy.h>	78	#include <syscall/copy.h>
79	#include <arch/interrupt.h>	79	#include <arch/interrupt.h>
80		80
81	/**	81	/**
82	* Each architecture decides what functions will be used to carry out	82	* Each architecture decides what functions will be used to carry out
83	* address space operations such as creating or locking page tables.	83	* address space operations such as creating or locking page tables.
84	*/	84	*/
85	as_operations_t *as_operations = NULL;	85	as_operations_t *as_operations = NULL;
86		86
87	/**	87	/**
88	* Slab for as_t objects.	88	* Slab for as_t objects.
89	*/	89	*/
90	static slab_cache_t *as_slab;	90	static slab_cache_t *as_slab;
91		91
92	/** This lock protects inactive_as_with_asid_head list. It must be acquired before as_t mutex. */	92	/** This lock protects inactive_as_with_asid_head list. It must be acquired before as_t mutex. */
93	SPINLOCK_INITIALIZE(inactive_as_with_asid_lock);	93	SPINLOCK_INITIALIZE(inactive_as_with_asid_lock);
94		94
95	/**	95	/**
96	* This list contains address spaces that are not active on any	96	* This list contains address spaces that are not active on any
97	* processor and that have valid ASID.	97	* processor and that have valid ASID.
98	*/	98	*/
99	LIST_INITIALIZE(inactive_as_with_asid_head);	99	LIST_INITIALIZE(inactive_as_with_asid_head);
100		100
101	/** Kernel address space. */	101	/** Kernel address space. */
102	as_t *AS_KERNEL = NULL;	102	as_t *AS_KERNEL = NULL;
103		103
104	static int area_flags_to_page_flags(int aflags);	104	static int area_flags_to_page_flags(int aflags);
105	static as_area_t find_area_and_lock(as_t as, uintptr_t va);	105	static as_area_t find_area_and_lock(as_t as, uintptr_t va);
106	static bool check_area_conflicts(as_t as, uintptr_t va, size_t size, as_area_t avoid_area);	106	static bool check_area_conflicts(as_t as, uintptr_t va, size_t size, as_area_t avoid_area);
107	static void sh_info_remove_reference(share_info_t *sh_info);	107	static void sh_info_remove_reference(share_info_t *sh_info);
108		108
109	static int as_constructor(void *obj, int flags)	109	static int as_constructor(void *obj, int flags)
110	{	110	{
111	as_t as = (as_t ) obj;	111	as_t as = (as_t ) obj;
112	int rc;	112	int rc;
113		113
114	link_initialize(&as->inactive_as_with_asid_link);	114	link_initialize(&as->inactive_as_with_asid_link);
115	mutex_initialize(&as->lock);	115	mutex_initialize(&as->lock);
116		116
117	rc = as_constructor_arch(as, flags);	117	rc = as_constructor_arch(as, flags);
118		118
119	return rc;	119	return rc;
120	}	120	}
121		121
122	static int as_destructor(void *obj)	122	static int as_destructor(void *obj)
123	{	123	{
124	as_t as = (as_t ) obj;	124	as_t as = (as_t ) obj;
125		125
126	return as_destructor_arch(as);	126	return as_destructor_arch(as);
127	}	127	}
128		128
129	/** Initialize address space subsystem. */	129	/** Initialize address space subsystem. */
130	void as_init(void)	130	void as_init(void)
131	{	131	{
132	as_arch_init();	132	as_arch_init();
133		133
134	as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,	134	as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,
135	as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);	135	as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);
136		136
137	AS_KERNEL = as_create(FLAG_AS_KERNEL);	137	AS_KERNEL = as_create(FLAG_AS_KERNEL);
138	if (!AS_KERNEL)	138	if (!AS_KERNEL)
139	panic("can't create kernel address space\n");	139	panic("can't create kernel address space\n");
140		140
141	}	141	}
142		142
143	/** Create address space.	143	/** Create address space.
144	*	144	*
145	* @param flags Flags that influence way in wich the address space is created.	145	* @param flags Flags that influence way in wich the address space is created.
146	*/	146	*/
147	as_t *as_create(int flags)	147	as_t *as_create(int flags)
148	{	148	{
149	as_t *as;	149	as_t *as;
150		150
151	as = (as_t *) slab_alloc(as_slab, 0);	151	as = (as_t *) slab_alloc(as_slab, 0);
152	(void) as_create_arch(as, 0);	152	(void) as_create_arch(as, 0);
153		153
154	btree_create(&as->as_area_btree);	154	btree_create(&as->as_area_btree);
155		155
156	if (flags & FLAG_AS_KERNEL)	156	if (flags & FLAG_AS_KERNEL)
157	as->asid = ASID_KERNEL;	157	as->asid = ASID_KERNEL;
158	else	158	else
159	as->asid = ASID_INVALID;	159	as->asid = ASID_INVALID;
160		160
161	as->refcount = 0;	161	as->refcount = 0;
162	as->cpu_refcount = 0;	162	as->cpu_refcount = 0;
163	as->page_table = page_table_create(flags);	163	as->page_table = page_table_create(flags);
164		164
165	return as;	165	return as;
166	}	166	}
167		167
168	/** Destroy adress space.	168	/** Destroy adress space.
169	*	169	*
170	* When there are no tasks referencing this address space (i.e. its refcount is zero),	170	* When there are no tasks referencing this address space (i.e. its refcount is zero),
171	* the address space can be destroyed.	171	* the address space can be destroyed.
172	*/	172	*/
173	void as_destroy(as_t *as)	173	void as_destroy(as_t *as)
174	{	174	{
175	ipl_t ipl;	175	ipl_t ipl;
176	bool cond;	176	bool cond;
177		177
178	ASSERT(as->refcount == 0);	178	ASSERT(as->refcount == 0);
179		179
180	/*	180	/*
181	* Since there is no reference to this area,	181	* Since there is no reference to this area,
182	* it is safe not to lock its mutex.	182	* it is safe not to lock its mutex.
183	*/	183	*/
184	ipl = interrupts_disable();	184	ipl = interrupts_disable();
185	spinlock_lock(&inactive_as_with_asid_lock);	185	spinlock_lock(&inactive_as_with_asid_lock);
186	if (as->asid != ASID_INVALID && as != AS_KERNEL) {	186	if (as->asid != ASID_INVALID && as != AS_KERNEL) {
187	if (as != AS && as->cpu_refcount == 0)	187	if (as != AS && as->cpu_refcount == 0)
188	list_remove(&as->inactive_as_with_asid_link);	188	list_remove(&as->inactive_as_with_asid_link);
189	asid_put(as->asid);	189	asid_put(as->asid);
190	}	190	}
191	spinlock_unlock(&inactive_as_with_asid_lock);	191	spinlock_unlock(&inactive_as_with_asid_lock);
192		192
193	/*	193	/*
194	* Destroy address space areas of the address space.	194	* Destroy address space areas of the address space.
195	* The B+tree must be walked carefully because it is	195	* The B+tree must be walked carefully because it is
196	* also being destroyed.	196	* also being destroyed.
197	*/	197	*/
198	for (cond = true; cond; ) {	198	for (cond = true; cond; ) {
199	btree_node_t *node;	199	btree_node_t *node;
200		200
201	ASSERT(!list_empty(&as->as_area_btree.leaf_head));	201	ASSERT(!list_empty(&as->as_area_btree.leaf_head));
202	node = list_get_instance(as->as_area_btree.leaf_head.next, btree_node_t, leaf_link);	202	node = list_get_instance(as->as_area_btree.leaf_head.next, btree_node_t, leaf_link);
203		203
204	if ((cond = node->keys)) {	204	if ((cond = node->keys)) {
205	as_area_destroy(as, node->key[0]);	205	as_area_destroy(as, node->key[0]);
206	}	206	}
207	}	207	}
208		208
209	btree_destroy(&as->as_area_btree);	209	btree_destroy(&as->as_area_btree);
210	page_table_destroy(as->page_table);	210	page_table_destroy(as->page_table);
211		211
212	interrupts_restore(ipl);	212	interrupts_restore(ipl);
213		213
214	slab_free(as_slab, as);	214	slab_free(as_slab, as);
215	}	215	}
216		216
217	/** Create address space area of common attributes.	217	/** Create address space area of common attributes.
218	*	218	*
219	* The created address space area is added to the target address space.	219	* The created address space area is added to the target address space.
220	*	220	*
221	* @param as Target address space.	221	* @param as Target address space.
222	* @param flags Flags of the area memory.	222	* @param flags Flags of the area memory.
223	* @param size Size of area.	223	* @param size Size of area.
224	* @param base Base address of area.	224	* @param base Base address of area.
225	* @param attrs Attributes of the area.	225	* @param attrs Attributes of the area.
226	* @param backend Address space area backend. NULL if no backend is used.	226	* @param backend Address space area backend. NULL if no backend is used.
227	* @param backend_data NULL or a pointer to an array holding two void *.	227	* @param backend_data NULL or a pointer to an array holding two void *.
228	*	228	*
229	* @return Address space area on success or NULL on failure.	229	* @return Address space area on success or NULL on failure.
230	*/	230	*/
231	as_area_t as_area_create(as_t as, int flags, size_t size, uintptr_t base, int attrs,	231	as_area_t as_area_create(as_t as, int flags, size_t size, uintptr_t base, int attrs,
232	mem_backend_t backend, mem_backend_data_t backend_data)	232	mem_backend_t backend, mem_backend_data_t backend_data)
233	{	233	{
234	ipl_t ipl;	234	ipl_t ipl;
235	as_area_t *a;	235	as_area_t *a;
236		236
237	if (base % PAGE_SIZE)	237	if (base % PAGE_SIZE)
238	return NULL;	238	return NULL;
239		239
240	if (!size)	240	if (!size)
241	return NULL;	241	return NULL;
242		242
243	/* Writeable executable areas are not supported. */	243	/* Writeable executable areas are not supported. */
244	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))	244	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))
245	return NULL;	245	return NULL;
246		246
247	ipl = interrupts_disable();	247	ipl = interrupts_disable();
248	mutex_lock(&as->lock);	248	mutex_lock(&as->lock);
249		249
250	if (!check_area_conflicts(as, base, size, NULL)) {	250	if (!check_area_conflicts(as, base, size, NULL)) {
251	mutex_unlock(&as->lock);	251	mutex_unlock(&as->lock);
252	interrupts_restore(ipl);	252	interrupts_restore(ipl);
253	return NULL;	253	return NULL;
254	}	254	}
255		255
256	a = (as_area_t *) malloc(sizeof(as_area_t), 0);	256	a = (as_area_t *) malloc(sizeof(as_area_t), 0);
257		257
258	mutex_initialize(&a->lock);	258	mutex_initialize(&a->lock);
259		259
260	a->as = as;	260	a->as = as;
261	a->flags = flags;	261	a->flags = flags;
262	a->attributes = attrs;	262	a->attributes = attrs;
263	a->pages = SIZE2FRAMES(size);	263	a->pages = SIZE2FRAMES(size);
264	a->base = base;	264	a->base = base;
265	a->sh_info = NULL;	265	a->sh_info = NULL;
266	a->backend = backend;	266	a->backend = backend;
267	if (backend_data)	267	if (backend_data)
268	a->backend_data = *backend_data;	268	a->backend_data = *backend_data;
269	else	269	else
270	memsetb((uintptr_t) &a->backend_data, sizeof(a->backend_data), 0);	270	memsetb((uintptr_t) &a->backend_data, sizeof(a->backend_data), 0);
271		271
272	btree_create(&a->used_space);	272	btree_create(&a->used_space);
273		273
274	btree_insert(&as->as_area_btree, base, (void *) a, NULL);	274	btree_insert(&as->as_area_btree, base, (void *) a, NULL);
275		275
276	mutex_unlock(&as->lock);	276	mutex_unlock(&as->lock);
277	interrupts_restore(ipl);	277	interrupts_restore(ipl);
278		278
279	return a;	279	return a;
280	}	280	}
281		281
282	/** Find address space area and change it.	282	/** Find address space area and change it.
283	*	283	*
284	* @param as Address space.	284	* @param as Address space.
285	* @param address Virtual address belonging to the area to be changed. Must be page-aligned.	285	* @param address Virtual address belonging to the area to be changed. Must be page-aligned.
286	* @param size New size of the virtual memory block starting at address.	286	* @param size New size of the virtual memory block starting at address.
287	* @param flags Flags influencing the remap operation. Currently unused.	287	* @param flags Flags influencing the remap operation. Currently unused.
288	*	288	*
289	* @return Zero on success or a value from @ref errno.h otherwise.	289	* @return Zero on success or a value from @ref errno.h otherwise.
290	*/	290	*/
291	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)	291	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)
292	{	292	{
293	as_area_t *area;	293	as_area_t *area;
294	ipl_t ipl;	294	ipl_t ipl;
295	size_t pages;	295	size_t pages;
296		296
297	ipl = interrupts_disable();	297	ipl = interrupts_disable();
298	mutex_lock(&as->lock);	298	mutex_lock(&as->lock);
299		299
300	/*	300	/*
301	* Locate the area.	301	* Locate the area.
302	*/	302	*/
303	area = find_area_and_lock(as, address);	303	area = find_area_and_lock(as, address);
304	if (!area) {	304	if (!area) {
305	mutex_unlock(&as->lock);	305	mutex_unlock(&as->lock);
306	interrupts_restore(ipl);	306	interrupts_restore(ipl);
307	return ENOENT;	307	return ENOENT;
308	}	308	}
309		309
310	if (area->backend == &phys_backend) {	310	if (area->backend == &phys_backend) {
311	/*	311	/*
312	* Remapping of address space areas associated	312	* Remapping of address space areas associated
313	* with memory mapped devices is not supported.	313	* with memory mapped devices is not supported.
314	*/	314	*/
315	mutex_unlock(&area->lock);	315	mutex_unlock(&area->lock);
316	mutex_unlock(&as->lock);	316	mutex_unlock(&as->lock);
317	interrupts_restore(ipl);	317	interrupts_restore(ipl);
318	return ENOTSUP;	318	return ENOTSUP;
319	}	319	}
320	if (area->sh_info) {	320	if (area->sh_info) {
321	/*	321	/*
322	* Remapping of shared address space areas	322	* Remapping of shared address space areas
323	* is not supported.	323	* is not supported.
324	*/	324	*/
325	mutex_unlock(&area->lock);	325	mutex_unlock(&area->lock);
326	mutex_unlock(&as->lock);	326	mutex_unlock(&as->lock);
327	interrupts_restore(ipl);	327	interrupts_restore(ipl);
328	return ENOTSUP;	328	return ENOTSUP;
329	}	329	}
330		330
331	pages = SIZE2FRAMES((address - area->base) + size);	331	pages = SIZE2FRAMES((address - area->base) + size);
332	if (!pages) {	332	if (!pages) {
333	/*	333	/*
334	* Zero size address space areas are not allowed.	334	* Zero size address space areas are not allowed.
335	*/	335	*/
336	mutex_unlock(&area->lock);	336	mutex_unlock(&area->lock);
337	mutex_unlock(&as->lock);	337	mutex_unlock(&as->lock);
338	interrupts_restore(ipl);	338	interrupts_restore(ipl);
339	return EPERM;	339	return EPERM;
340	}	340	}
341		341
342	if (pages < area->pages) {	342	if (pages < area->pages) {
343	bool cond;	343	bool cond;
344	uintptr_t start_free = area->base + pages*PAGE_SIZE;	344	uintptr_t start_free = area->base + pages*PAGE_SIZE;
345		345
346	/*	346	/*
347	* Shrinking the area.	347	* Shrinking the area.
348	* No need to check for overlaps.	348	* No need to check for overlaps.
349	*/	349	*/
350		350
351	/*	351	/*
352	* Start TLB shootdown sequence.	352	* Start TLB shootdown sequence.
353	*/	353	*/
354	tlb_shootdown_start(TLB_INVL_PAGES, AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);	354	tlb_shootdown_start(TLB_INVL_PAGES, AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);
355		355
356	/*	356	/*
357	* Remove frames belonging to used space starting from	357	* Remove frames belonging to used space starting from
358	* the highest addresses downwards until an overlap with	358	* the highest addresses downwards until an overlap with
359	* the resized address space area is found. Note that this	359	* the resized address space area is found. Note that this
360	* is also the right way to remove part of the used_space	360	* is also the right way to remove part of the used_space
361	* B+tree leaf list.	361	* B+tree leaf list.
362	*/	362	*/
363	for (cond = true; cond;) {	363	for (cond = true; cond;) {
364	btree_node_t *node;	364	btree_node_t *node;
365		365
366	ASSERT(!list_empty(&area->used_space.leaf_head));	366	ASSERT(!list_empty(&area->used_space.leaf_head));
367	node = list_get_instance(area->used_space.leaf_head.prev, btree_node_t, leaf_link);	367	node = list_get_instance(area->used_space.leaf_head.prev, btree_node_t, leaf_link);
368	if ((cond = (bool) node->keys)) {	368	if ((cond = (bool) node->keys)) {
369	uintptr_t b = node->key[node->keys - 1];	369	uintptr_t b = node->key[node->keys - 1];
370	count_t c = (count_t) node->value[node->keys - 1];	370	count_t c = (count_t) node->value[node->keys - 1];
371	int i = 0;	371	int i = 0;
372		372
373	if (overlaps(b, cPAGE_SIZE, area->base, pagesPAGE_SIZE)) {	373	if (overlaps(b, cPAGE_SIZE, area->base, pagesPAGE_SIZE)) {
374		374
375	if (b + c*PAGE_SIZE <= start_free) {	375	if (b + c*PAGE_SIZE <= start_free) {
376	/*	376	/*
377	* The whole interval fits completely	377	* The whole interval fits completely
378	* in the resized address space area.	378	* in the resized address space area.
379	*/	379	*/
380	break;	380	break;
381	}	381	}
382		382
383	/*	383	/*
384	* Part of the interval corresponding to b and c	384	* Part of the interval corresponding to b and c
385	* overlaps with the resized address space area.	385	* overlaps with the resized address space area.
386	*/	386	*/
387		387
388	cond = false; /* we are almost done */	388	cond = false; /* we are almost done */
389	i = (start_free - b) >> PAGE_WIDTH;	389	i = (start_free - b) >> PAGE_WIDTH;
390	if (!used_space_remove(area, start_free, c - i))	390	if (!used_space_remove(area, start_free, c - i))
391	panic("Could not remove used space.\n");	391	panic("Could not remove used space.\n");
392	} else {	392	} else {
393	/*	393	/*
394	* The interval of used space can be completely removed.	394	* The interval of used space can be completely removed.
395	*/	395	*/
396	if (!used_space_remove(area, b, c))	396	if (!used_space_remove(area, b, c))
397	panic("Could not remove used space.\n");	397	panic("Could not remove used space.\n");
398	}	398	}
399		399
400	for (; i < c; i++) {	400	for (; i < c; i++) {
401	pte_t *pte;	401	pte_t *pte;
402		402
403	page_table_lock(as, false);	403	page_table_lock(as, false);
404	pte = page_mapping_find(as, b + i*PAGE_SIZE);	404	pte = page_mapping_find(as, b + i*PAGE_SIZE);
405	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));	405	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));
406	if (area->backend && area->backend->frame_free) {	406	if (area->backend && area->backend->frame_free) {
407	area->backend->frame_free(area,	407	area->backend->frame_free(area,
408	b + i*PAGE_SIZE, PTE_GET_FRAME(pte));	408	b + i*PAGE_SIZE, PTE_GET_FRAME(pte));
409	}	409	}
410	page_mapping_remove(as, b + i*PAGE_SIZE);	410	page_mapping_remove(as, b + i*PAGE_SIZE);
411	page_table_unlock(as, false);	411	page_table_unlock(as, false);
412	}	412	}
413	}	413	}
414	}	414	}
415		415
416	/*	416	/*
417	* Finish TLB shootdown sequence.	417	* Finish TLB shootdown sequence.
418	*/	418	*/
419	tlb_invalidate_pages(as->asid, area->base + pages*PAGE_SIZE, area->pages - pages);	419	tlb_invalidate_pages(as->asid, area->base + pages*PAGE_SIZE, area->pages - pages);
420	tlb_shootdown_finalize();	420	tlb_shootdown_finalize();
421		421
422	/*	422	/*
423	* Invalidate software translation caches (e.g. TSB on sparc64).	423	* Invalidate software translation caches (e.g. TSB on sparc64).
424	*/	424	*/
425	as_invalidate_translation_cache(as, area->base + pages*PAGE_SIZE, area->pages - pages);	425	as_invalidate_translation_cache(as, area->base + pages*PAGE_SIZE, area->pages - pages);
426	} else {	426	} else {
427	/*	427	/*
428	* Growing the area.	428	* Growing the area.
429	* Check for overlaps with other address space areas.	429	* Check for overlaps with other address space areas.
430	*/	430	*/
431	if (!check_area_conflicts(as, address, pages * PAGE_SIZE, area)) {	431	if (!check_area_conflicts(as, address, pages * PAGE_SIZE, area)) {
432	mutex_unlock(&area->lock);	432	mutex_unlock(&area->lock);
433	mutex_unlock(&as->lock);	433	mutex_unlock(&as->lock);
434	interrupts_restore(ipl);	434	interrupts_restore(ipl);
435	return EADDRNOTAVAIL;	435	return EADDRNOTAVAIL;
436	}	436	}
437	}	437	}
438		438
439	area->pages = pages;	439	area->pages = pages;
440		440
441	mutex_unlock(&area->lock);	441	mutex_unlock(&area->lock);
442	mutex_unlock(&as->lock);	442	mutex_unlock(&as->lock);
443	interrupts_restore(ipl);	443	interrupts_restore(ipl);
444		444
445	return 0;	445	return 0;
446	}	446	}
447		447
448	/** Destroy address space area.	448	/** Destroy address space area.
449	*	449	*
450	* @param as Address space.	450	* @param as Address space.
451	* @param address Address withing the area to be deleted.	451	* @param address Address withing the area to be deleted.
452	*	452	*
453	* @return Zero on success or a value from @ref errno.h on failure.	453	* @return Zero on success or a value from @ref errno.h on failure.
454	*/	454	*/
455	int as_area_destroy(as_t *as, uintptr_t address)	455	int as_area_destroy(as_t *as, uintptr_t address)
456	{	456	{
457	as_area_t *area;	457	as_area_t *area;
458	uintptr_t base;	458	uintptr_t base;
459	link_t *cur;	459	link_t *cur;
460	ipl_t ipl;	460	ipl_t ipl;
461		461
462	ipl = interrupts_disable();	462	ipl = interrupts_disable();
463	mutex_lock(&as->lock);	463	mutex_lock(&as->lock);
464		464
465	area = find_area_and_lock(as, address);	465	area = find_area_and_lock(as, address);
466	if (!area) {	466	if (!area) {
467	mutex_unlock(&as->lock);	467	mutex_unlock(&as->lock);
468	interrupts_restore(ipl);	468	interrupts_restore(ipl);
469	return ENOENT;	469	return ENOENT;
470	}	470	}
471		471
472	base = area->base;	472	base = area->base;
473		473
474	/*	474	/*
475	* Start TLB shootdown sequence.	475	* Start TLB shootdown sequence.
476	*/	476	*/
477	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);	477	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
478		478
479	/*	479	/*
480	* Visit only the pages mapped by used_space B+tree.	480	* Visit only the pages mapped by used_space B+tree.
481	*/	481	*/
482	for (cur = area->used_space.leaf_head.next; cur != &area->used_space.leaf_head; cur = cur->next) {	482	for (cur = area->used_space.leaf_head.next; cur != &area->used_space.leaf_head; cur = cur->next) {
483	btree_node_t *node;	483	btree_node_t *node;
484	int i;	484	int i;
485		485
486	node = list_get_instance(cur, btree_node_t, leaf_link);	486	node = list_get_instance(cur, btree_node_t, leaf_link);
487	for (i = 0; i < node->keys; i++) {	487	for (i = 0; i < node->keys; i++) {
488	uintptr_t b = node->key[i];	488	uintptr_t b = node->key[i];
489	count_t j;	489	count_t j;
490	pte_t *pte;	490	pte_t *pte;
491		491
492	for (j = 0; j < (count_t) node->value[i]; j++) {	492	for (j = 0; j < (count_t) node->value[i]; j++) {
493	page_table_lock(as, false);	493	page_table_lock(as, false);
494	pte = page_mapping_find(as, b + j*PAGE_SIZE);	494	pte = page_mapping_find(as, b + j*PAGE_SIZE);
495	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));	495	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));
496	if (area->backend && area->backend->frame_free) {	496	if (area->backend && area->backend->frame_free) {
497	area->backend->frame_free(area,	497	area->backend->frame_free(area,
498	b + j*PAGE_SIZE, PTE_GET_FRAME(pte));	498	b + j*PAGE_SIZE, PTE_GET_FRAME(pte));
499	}	499	}
500	page_mapping_remove(as, b + j*PAGE_SIZE);	500	page_mapping_remove(as, b + j*PAGE_SIZE);

Subversion Repositories HelenOS

(root)/trunk/kernel/generic/src/mm/as.c – Rev 1954 → 2007