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);
 
/** Initialize address space subsystem. */
void as_init(void)
{
    as_arch_init();
   
    as_slab = slab_cache_create("as_slab", sizeof(as_t), 0, NULL, NULL, 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);
    link_initialize(&as->inactive_as_with_asid_link);
    mutex_initialize(&as->lock);
    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+tee 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 an attempt
 *     to share non-anonymous address space area is detected.
 */
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);
        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);
}
 
/** @}
 */
 

Rev 1889	Rev 1890
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	/**
-		88	* Slab for as_t objects.
-		89	*/
-		90	static slab_cache_t *as_slab;
-		91
87	/** 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. */
88	SPINLOCK_INITIALIZE(inactive_as_with_asid_lock);	93	SPINLOCK_INITIALIZE(inactive_as_with_asid_lock);
89		94
90	/**	95	/**
91	* This list contains address spaces that are not active on any	96	* This list contains address spaces that are not active on any
92	* processor and that have valid ASID.	97	* processor and that have valid ASID.
93	*/	98	*/
94	LIST_INITIALIZE(inactive_as_with_asid_head);	99	LIST_INITIALIZE(inactive_as_with_asid_head);
95		100
96	/** Kernel address space. */	101	/** Kernel address space. */
97	as_t *AS_KERNEL = NULL;	102	as_t *AS_KERNEL = NULL;
98		103
99	static int area_flags_to_page_flags(int aflags);	104	static int area_flags_to_page_flags(int aflags);
100	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);
101	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);
102	static void sh_info_remove_reference(share_info_t *sh_info);	107	static void sh_info_remove_reference(share_info_t *sh_info);
103		108
104	/** Initialize address space subsystem. */	109	/** Initialize address space subsystem. */
105	void as_init(void)	110	void as_init(void)
106	{	111	{
107	as_arch_init();	112	as_arch_init();
-		113
-		114	as_slab = slab_cache_create("as_slab", sizeof(as_t), 0, NULL, NULL, SLAB_CACHE_MAGDEFERRED);
-		115
108	AS_KERNEL = as_create(FLAG_AS_KERNEL);	116	AS_KERNEL = as_create(FLAG_AS_KERNEL);
109	if (!AS_KERNEL)	117	if (!AS_KERNEL)
110	panic("can't create kernel address space\n");	118	panic("can't create kernel address space\n");
111		119
112	}	120	}
113		121
114	/** Create address space.	122	/** Create address space.
115	*	123	*
116	* @param flags Flags that influence way in wich the address space is created.	124	* @param flags Flags that influence way in wich the address space is created.
117	*/	125	*/
118	as_t *as_create(int flags)	126	as_t *as_create(int flags)
119	{	127	{
120	as_t *as;	128	as_t *as;
121		129
122	as = (as_t *) malloc(sizeof(as_t), 0);	130	as = (as_t *) slab_alloc(as_slab, 0);
123	link_initialize(&as->inactive_as_with_asid_link);	131	link_initialize(&as->inactive_as_with_asid_link);
124	mutex_initialize(&as->lock);	132	mutex_initialize(&as->lock);
125	btree_create(&as->as_area_btree);	133	btree_create(&as->as_area_btree);
126		134
127	if (flags & FLAG_AS_KERNEL)	135	if (flags & FLAG_AS_KERNEL)
128	as->asid = ASID_KERNEL;	136	as->asid = ASID_KERNEL;
129	else	137	else
130	as->asid = ASID_INVALID;	138	as->asid = ASID_INVALID;
131		139
132	as->refcount = 0;	140	as->refcount = 0;
133	as->cpu_refcount = 0;	141	as->cpu_refcount = 0;
134	as->page_table = page_table_create(flags);	142	as->page_table = page_table_create(flags);
135		143
136	return as;	144	return as;
137	}	145	}
138		146
139	/** Destroy adress space.	147	/** Destroy adress space.
140	*	148	*
141	* When there are no tasks referencing this address space (i.e. its refcount is zero),	149	* When there are no tasks referencing this address space (i.e. its refcount is zero),
142	* the address space can be destroyed.	150	* the address space can be destroyed.
143	*/	151	*/
144	void as_destroy(as_t *as)	152	void as_destroy(as_t *as)
145	{	153	{
146	ipl_t ipl;	154	ipl_t ipl;
147	bool cond;	155	bool cond;
148		156
149	ASSERT(as->refcount == 0);	157	ASSERT(as->refcount == 0);
150		158
151	/*	159	/*
152	* Since there is no reference to this area,	160	* Since there is no reference to this area,
153	* it is safe not to lock its mutex.	161	* it is safe not to lock its mutex.
154	*/	162	*/
155	ipl = interrupts_disable();	163	ipl = interrupts_disable();
156	spinlock_lock(&inactive_as_with_asid_lock);	164	spinlock_lock(&inactive_as_with_asid_lock);
157	if (as->asid != ASID_INVALID && as != AS_KERNEL) {	165	if (as->asid != ASID_INVALID && as != AS_KERNEL) {
158	if (as != AS && as->cpu_refcount == 0)	166	if (as != AS && as->cpu_refcount == 0)
159	list_remove(&as->inactive_as_with_asid_link);	167	list_remove(&as->inactive_as_with_asid_link);
160	asid_put(as->asid);	168	asid_put(as->asid);
161	}	169	}
162	spinlock_unlock(&inactive_as_with_asid_lock);	170	spinlock_unlock(&inactive_as_with_asid_lock);
163		171
164	/*	172	/*
165	* Destroy address space areas of the address space.	173	* Destroy address space areas of the address space.
166	* The B+tee must be walked carefully because it is	174	* The B+tee must be walked carefully because it is
167	* also being destroyed.	175	* also being destroyed.
168	*/	176	*/
169	for (cond = true; cond; ) {	177	for (cond = true; cond; ) {
170	btree_node_t *node;	178	btree_node_t *node;
171		179
172	ASSERT(!list_empty(&as->as_area_btree.leaf_head));	180	ASSERT(!list_empty(&as->as_area_btree.leaf_head));
173	node = list_get_instance(as->as_area_btree.leaf_head.next, btree_node_t, leaf_link);	181	node = list_get_instance(as->as_area_btree.leaf_head.next, btree_node_t, leaf_link);
174		182
175	if ((cond = node->keys)) {	183	if ((cond = node->keys)) {
176	as_area_destroy(as, node->key[0]);	184	as_area_destroy(as, node->key[0]);
177	}	185	}
178	}	186	}
179		187
180	btree_destroy(&as->as_area_btree);	188	btree_destroy(&as->as_area_btree);
181	page_table_destroy(as->page_table);	189	page_table_destroy(as->page_table);
182		190
183	interrupts_restore(ipl);	191	interrupts_restore(ipl);
184		192
185	free(as);	193	slab_free(as_slab, as);
186	}	194	}
187		195
188	/** Create address space area of common attributes.	196	/** Create address space area of common attributes.
189	*	197	*
190	* The created address space area is added to the target address space.	198	* The created address space area is added to the target address space.
191	*	199	*
192	* @param as Target address space.	200	* @param as Target address space.
193	* @param flags Flags of the area memory.	201	* @param flags Flags of the area memory.
194	* @param size Size of area.	202	* @param size Size of area.
195	* @param base Base address of area.	203	* @param base Base address of area.
196	* @param attrs Attributes of the area.	204	* @param attrs Attributes of the area.
197	* @param backend Address space area backend. NULL if no backend is used.	205	* @param backend Address space area backend. NULL if no backend is used.
198	* @param backend_data NULL or a pointer to an array holding two void *.	206	* @param backend_data NULL or a pointer to an array holding two void *.
199	*	207	*
200	* @return Address space area on success or NULL on failure.	208	* @return Address space area on success or NULL on failure.
201	*/	209	*/
202	as_area_t as_area_create(as_t as, int flags, size_t size, uintptr_t base, int attrs,	210	as_area_t as_area_create(as_t as, int flags, size_t size, uintptr_t base, int attrs,
203	mem_backend_t backend, mem_backend_data_t backend_data)	211	mem_backend_t backend, mem_backend_data_t backend_data)
204	{	212	{
205	ipl_t ipl;	213	ipl_t ipl;
206	as_area_t *a;	214	as_area_t *a;
207		215
208	if (base % PAGE_SIZE)	216	if (base % PAGE_SIZE)
209	return NULL;	217	return NULL;
210		218
211	if (!size)	219	if (!size)
212	return NULL;	220	return NULL;
213		221
214	/* Writeable executable areas are not supported. */	222	/* Writeable executable areas are not supported. */
215	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))	223	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))
216	return NULL;	224	return NULL;
217		225
218	ipl = interrupts_disable();	226	ipl = interrupts_disable();
219	mutex_lock(&as->lock);	227	mutex_lock(&as->lock);
220		228
221	if (!check_area_conflicts(as, base, size, NULL)) {	229	if (!check_area_conflicts(as, base, size, NULL)) {
222	mutex_unlock(&as->lock);	230	mutex_unlock(&as->lock);
223	interrupts_restore(ipl);	231	interrupts_restore(ipl);
224	return NULL;	232	return NULL;
225	}	233	}
226		234
227	a = (as_area_t *) malloc(sizeof(as_area_t), 0);	235	a = (as_area_t *) malloc(sizeof(as_area_t), 0);
228		236
229	mutex_initialize(&a->lock);	237	mutex_initialize(&a->lock);
230		238
231	a->as = as;	239	a->as = as;
232	a->flags = flags;	240	a->flags = flags;
233	a->attributes = attrs;	241	a->attributes = attrs;
234	a->pages = SIZE2FRAMES(size);	242	a->pages = SIZE2FRAMES(size);
235	a->base = base;	243	a->base = base;
236	a->sh_info = NULL;	244	a->sh_info = NULL;
237	a->backend = backend;	245	a->backend = backend;
238	if (backend_data)	246	if (backend_data)
239	a->backend_data = *backend_data;	247	a->backend_data = *backend_data;
240	else	248	else
241	memsetb((uintptr_t) &a->backend_data, sizeof(a->backend_data), 0);	249	memsetb((uintptr_t) &a->backend_data, sizeof(a->backend_data), 0);
242		250
243	btree_create(&a->used_space);	251	btree_create(&a->used_space);
244		252
245	btree_insert(&as->as_area_btree, base, (void *) a, NULL);	253	btree_insert(&as->as_area_btree, base, (void *) a, NULL);
246		254
247	mutex_unlock(&as->lock);	255	mutex_unlock(&as->lock);
248	interrupts_restore(ipl);	256	interrupts_restore(ipl);
249		257
250	return a;	258	return a;
251	}	259	}
252		260
253	/** Find address space area and change it.	261	/** Find address space area and change it.
254	*	262	*
255	* @param as Address space.	263	* @param as Address space.
256	* @param address Virtual address belonging to the area to be changed. Must be page-aligned.	264	* @param address Virtual address belonging to the area to be changed. Must be page-aligned.
257	* @param size New size of the virtual memory block starting at address.	265	* @param size New size of the virtual memory block starting at address.
258	* @param flags Flags influencing the remap operation. Currently unused.	266	* @param flags Flags influencing the remap operation. Currently unused.
259	*	267	*
260	* @return Zero on success or a value from @ref errno.h otherwise.	268	* @return Zero on success or a value from @ref errno.h otherwise.
261	*/	269	*/
262	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)	270	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)
263	{	271	{
264	as_area_t *area;	272	as_area_t *area;
265	ipl_t ipl;	273	ipl_t ipl;
266	size_t pages;	274	size_t pages;
267		275
268	ipl = interrupts_disable();	276	ipl = interrupts_disable();
269	mutex_lock(&as->lock);	277	mutex_lock(&as->lock);
270		278
271	/*	279	/*
272	* Locate the area.	280	* Locate the area.
273	*/	281	*/
274	area = find_area_and_lock(as, address);	282	area = find_area_and_lock(as, address);
275	if (!area) {	283	if (!area) {
276	mutex_unlock(&as->lock);	284	mutex_unlock(&as->lock);
277	interrupts_restore(ipl);	285	interrupts_restore(ipl);
278	return ENOENT;	286	return ENOENT;
279	}	287	}
280		288
281	if (area->backend == &phys_backend) {	289	if (area->backend == &phys_backend) {
282	/*	290	/*
283	* Remapping of address space areas associated	291	* Remapping of address space areas associated
284	* with memory mapped devices is not supported.	292	* with memory mapped devices is not supported.
285	*/	293	*/
286	mutex_unlock(&area->lock);	294	mutex_unlock(&area->lock);
287	mutex_unlock(&as->lock);	295	mutex_unlock(&as->lock);
288	interrupts_restore(ipl);	296	interrupts_restore(ipl);
289	return ENOTSUP;	297	return ENOTSUP;
290	}	298	}
291	if (area->sh_info) {	299	if (area->sh_info) {
292	/*	300	/*
293	* Remapping of shared address space areas	301	* Remapping of shared address space areas
294	* is not supported.	302	* is not supported.
295	*/	303	*/
296	mutex_unlock(&area->lock);	304	mutex_unlock(&area->lock);
297	mutex_unlock(&as->lock);	305	mutex_unlock(&as->lock);
298	interrupts_restore(ipl);	306	interrupts_restore(ipl);
299	return ENOTSUP;	307	return ENOTSUP;
300	}	308	}
301		309
302	pages = SIZE2FRAMES((address - area->base) + size);	310	pages = SIZE2FRAMES((address - area->base) + size);
303	if (!pages) {	311	if (!pages) {
304	/*	312	/*
305	* Zero size address space areas are not allowed.	313	* Zero size address space areas are not allowed.
306	*/	314	*/
307	mutex_unlock(&area->lock);	315	mutex_unlock(&area->lock);
308	mutex_unlock(&as->lock);	316	mutex_unlock(&as->lock);
309	interrupts_restore(ipl);	317	interrupts_restore(ipl);
310	return EPERM;	318	return EPERM;
311	}	319	}
312		320
313	if (pages < area->pages) {	321	if (pages < area->pages) {
314	bool cond;	322	bool cond;
315	uintptr_t start_free = area->base + pages*PAGE_SIZE;	323	uintptr_t start_free = area->base + pages*PAGE_SIZE;
316		324
317	/*	325	/*
318	* Shrinking the area.	326	* Shrinking the area.
319	* No need to check for overlaps.	327	* No need to check for overlaps.
320	*/	328	*/
321		329
322	/*	330	/*
323	* Start TLB shootdown sequence.	331	* Start TLB shootdown sequence.
324	*/	332	*/
325	tlb_shootdown_start(TLB_INVL_PAGES, AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);	333	tlb_shootdown_start(TLB_INVL_PAGES, AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);
326		334
327	/*	335	/*
328	* Remove frames belonging to used space starting from	336	* Remove frames belonging to used space starting from
329	* the highest addresses downwards until an overlap with	337	* the highest addresses downwards until an overlap with
330	* the resized address space area is found. Note that this	338	* the resized address space area is found. Note that this
331	* is also the right way to remove part of the used_space	339	* is also the right way to remove part of the used_space
332	* B+tree leaf list.	340	* B+tree leaf list.
333	*/	341	*/
334	for (cond = true; cond;) {	342	for (cond = true; cond;) {
335	btree_node_t *node;	343	btree_node_t *node;
336		344
337	ASSERT(!list_empty(&area->used_space.leaf_head));	345	ASSERT(!list_empty(&area->used_space.leaf_head));
338	node = list_get_instance(area->used_space.leaf_head.prev, btree_node_t, leaf_link);	346	node = list_get_instance(area->used_space.leaf_head.prev, btree_node_t, leaf_link);
339	if ((cond = (bool) node->keys)) {	347	if ((cond = (bool) node->keys)) {
340	uintptr_t b = node->key[node->keys - 1];	348	uintptr_t b = node->key[node->keys - 1];
341	count_t c = (count_t) node->value[node->keys - 1];	349	count_t c = (count_t) node->value[node->keys - 1];
342	int i = 0;	350	int i = 0;
343		351
344	if (overlaps(b, cPAGE_SIZE, area->base, pagesPAGE_SIZE)) {	352	if (overlaps(b, cPAGE_SIZE, area->base, pagesPAGE_SIZE)) {
345		353
346	if (b + c*PAGE_SIZE <= start_free) {	354	if (b + c*PAGE_SIZE <= start_free) {
347	/*	355	/*
348	* The whole interval fits completely	356	* The whole interval fits completely
349	* in the resized address space area.	357	* in the resized address space area.
350	*/	358	*/
351	break;	359	break;
352	}	360	}
353		361
354	/*	362	/*
355	* Part of the interval corresponding to b and c	363	* Part of the interval corresponding to b and c
356	* overlaps with the resized address space area.	364	* overlaps with the resized address space area.
357	*/	365	*/
358		366
359	cond = false; /* we are almost done */	367	cond = false; /* we are almost done */
360	i = (start_free - b) >> PAGE_WIDTH;	368	i = (start_free - b) >> PAGE_WIDTH;
361	if (!used_space_remove(area, start_free, c - i))	369	if (!used_space_remove(area, start_free, c - i))
362	panic("Could not remove used space.\n");	370	panic("Could not remove used space.\n");
363	} else {	371	} else {
364	/*	372	/*
365	* The interval of used space can be completely removed.	373	* The interval of used space can be completely removed.
366	*/	374	*/
367	if (!used_space_remove(area, b, c))	375	if (!used_space_remove(area, b, c))
368	panic("Could not remove used space.\n");	376	panic("Could not remove used space.\n");
369	}	377	}
370		378
371	for (; i < c; i++) {	379	for (; i < c; i++) {
372	pte_t *pte;	380	pte_t *pte;
373		381
374	page_table_lock(as, false);	382	page_table_lock(as, false);
375	pte = page_mapping_find(as, b + i*PAGE_SIZE);	383	pte = page_mapping_find(as, b + i*PAGE_SIZE);
376	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));	384	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));
377	if (area->backend && area->backend->frame_free) {	385	if (area->backend && area->backend->frame_free) {
378	area->backend->frame_free(area,	386	area->backend->frame_free(area,
379	b + i*PAGE_SIZE, PTE_GET_FRAME(pte));	387	b + i*PAGE_SIZE, PTE_GET_FRAME(pte));
380	}	388	}
381	page_mapping_remove(as, b + i*PAGE_SIZE);	389	page_mapping_remove(as, b + i*PAGE_SIZE);
382	page_table_unlock(as, false);	390	page_table_unlock(as, false);
383	}	391	}
384	}	392	}
385	}	393	}
386		394
387	/*	395	/*
388	* Finish TLB shootdown sequence.	396	* Finish TLB shootdown sequence.
389	*/	397	*/
390	tlb_invalidate_pages(AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);	398	tlb_invalidate_pages(AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);
391	tlb_shootdown_finalize();	399	tlb_shootdown_finalize();
392		400
393	/*	401	/*
394	* Invalidate software translation caches (e.g. TSB on sparc64).	402	* Invalidate software translation caches (e.g. TSB on sparc64).
395	*/	403	*/
396	as_invalidate_translation_cache(as, area->base + pages*PAGE_SIZE, area->pages - pages);	404	as_invalidate_translation_cache(as, area->base + pages*PAGE_SIZE, area->pages - pages);
397	} else {	405	} else {
398	/*	406	/*
399	* Growing the area.	407	* Growing the area.
400	* Check for overlaps with other address space areas.	408	* Check for overlaps with other address space areas.
401	*/	409	*/
402	if (!check_area_conflicts(as, address, pages * PAGE_SIZE, area)) {	410	if (!check_area_conflicts(as, address, pages * PAGE_SIZE, area)) {
403	mutex_unlock(&area->lock);	411	mutex_unlock(&area->lock);
404	mutex_unlock(&as->lock);	412	mutex_unlock(&as->lock);
405	interrupts_restore(ipl);	413	interrupts_restore(ipl);
406	return EADDRNOTAVAIL;	414	return EADDRNOTAVAIL;
407	}	415	}
408	}	416	}
409		417
410	area->pages = pages;	418	area->pages = pages;
411		419
412	mutex_unlock(&area->lock);	420	mutex_unlock(&area->lock);
413	mutex_unlock(&as->lock);	421	mutex_unlock(&as->lock);
414	interrupts_restore(ipl);	422	interrupts_restore(ipl);
415		423
416	return 0;	424	return 0;
417	}	425	}
418		426
419	/** Destroy address space area.	427	/** Destroy address space area.
420	*	428	*
421	* @param as Address space.	429	* @param as Address space.
422	* @param address Address withing the area to be deleted.	430	* @param address Address withing the area to be deleted.
423	*	431	*
424	* @return Zero on success or a value from @ref errno.h on failure.	432	* @return Zero on success or a value from @ref errno.h on failure.
425	*/	433	*/
426	int as_area_destroy(as_t *as, uintptr_t address)	434	int as_area_destroy(as_t *as, uintptr_t address)
427	{	435	{
428	as_area_t *area;	436	as_area_t *area;
429	uintptr_t base;	437	uintptr_t base;
430	link_t *cur;	438	link_t *cur;
431	ipl_t ipl;	439	ipl_t ipl;
432		440
433	ipl = interrupts_disable();	441	ipl = interrupts_disable();
434	mutex_lock(&as->lock);	442	mutex_lock(&as->lock);
435		443
436	area = find_area_and_lock(as, address);	444	area = find_area_and_lock(as, address);
437	if (!area) {	445	if (!area) {
438	mutex_unlock(&as->lock);	446	mutex_unlock(&as->lock);
439	interrupts_restore(ipl);	447	interrupts_restore(ipl);
440	return ENOENT;	448	return ENOENT;
441	}	449	}
442		450
443	base = area->base;	451	base = area->base;
444		452
445	/*	453	/*
446	* Start TLB shootdown sequence.	454	* Start TLB shootdown sequence.
447	*/	455	*/
448	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);	456	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
449		457
450	/*	458	/*
451	* Visit only the pages mapped by used_space B+tree.	459	* Visit only the pages mapped by used_space B+tree.
452	*/	460	*/
453	for (cur = area->used_space.leaf_head.next; cur != &area->used_space.leaf_head; cur = cur->next) {	461	for (cur = area->used_space.leaf_head.next; cur != &area->used_space.leaf_head; cur = cur->next) {
454	btree_node_t *node;	462	btree_node_t *node;
455	int i;	463	int i;
456		464
457	node = list_get_instance(cur, btree_node_t, leaf_link);	465	node = list_get_instance(cur, btree_node_t, leaf_link);
458	for (i = 0; i < node->keys; i++) {	466	for (i = 0; i < node->keys; i++) {
459	uintptr_t b = node->key[i];	467	uintptr_t b = node->key[i];
460	count_t j;	468	count_t j;
461	pte_t *pte;	469	pte_t *pte;
462		470
463	for (j = 0; j < (count_t) node->value[i]; j++) {	471	for (j = 0; j < (count_t) node->value[i]; j++) {
464	page_table_lock(as, false);	472	page_table_lock(as, false);
465	pte = page_mapping_find(as, b + j*PAGE_SIZE);	473	pte = page_mapping_find(as, b + j*PAGE_SIZE);
466	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));	474	ASSERT(pte && PTE_VALID(pte) && PTE_PRESENT(pte));
467	if (area->backend && area->backend->frame_free) {	475	if (area->backend && area->backend->frame_free) {
468	area->backend->frame_free(area,	476	area->backend->frame_free(area,
469	b + j*PAGE_SIZE, PTE_GET_FRAME(pte));	477	b + j*PAGE_SIZE, PTE_GET_FRAME(pte));
470	}	478	}
471	page_mapping_remove(as, b + j*PAGE_SIZE);	479	page_mapping_remove(as, b + j*PAGE_SIZE);
472	page_table_unlock(as, false);	480	page_table_unlock(as, false);
473	}	481	}
474	}	482	}
475	}	483	}
476		484
477	/*	485	/*
478	* Finish TLB shootdown sequence.	486	* Finish TLB shootdown sequence.
479	*/	487	*/
480	tlb_invalidate_pages(as->asid, area->base, area->pages);	488	tlb_invalidate_pages(as->asid, area->base, area->pages);
481	tlb_shootdown_finalize();	489	tlb_shootdown_finalize();
482		490
483	/*	491	/*
484	* Invalidate potential software translation caches (e.g. TSB on sparc64).	492	* Invalidate potential software translation caches (e.g. TSB on sparc64).
485	*/	493	*/
486	as_invalidate_translation_cache(as, area->base, area->pages);	494	as_invalidate_translation_cache(as, area->base, area->pages);
487		495
488	btree_destroy(&area->used_space);	496	btree_destroy(&area->used_space);
489		497
490	area->attributes \|= AS_AREA_ATTR_PARTIAL;	498	area->attributes \|= AS_AREA_ATTR_PARTIAL;
491		499
492	if (area->sh_info)	500	if (area->sh_info)

Subversion Repositories HelenOS

(root)/trunk/kernel/generic/src/mm/as.c @ 1891 – Rev 1889 → 1890