Subversion Repositories HelenOS-historic

/*

 * 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.

 */

/*

 * This file contains address space manipulation functions.

 * Roughly speaking, this is a higher-level client of

 * Virtual Address Translation (VAT) subsystem.

 */

#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 <adt/list.h>

#include <adt/btree.h>

#include <proc/task.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 <arch/types.h>

#include <typedefs.h>

as_operations_t *as_operations = NULL;

/** Address space lock. It protects inactive_as_with_asid_head. */

SPINLOCK_INITIALIZE(as_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 int get_area_flags(as_area_t *a);

static as_area_t *find_area_and_lock(as_t *as, __address va);

static bool check_area_conflicts(as_t *as, __address va, size_t size, as_area_t *avoid_area);

/** Initialize address space subsystem. */

void as_init(void)

{

    as_arch_init();

    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 *) malloc(sizeof(as_t), 0);

    link_initialize(&as->inactive_as_with_asid_link);

    spinlock_initialize(&as->lock, "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->page_table = page_table_create(flags);

    return as;

}

/** Free Adress space */

void as_free(as_t *as)

{

    ASSERT(as->refcount == 0);

    /* TODO: free as_areas and other resources held by as */

    /* TODO: free page table */

    free(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.

 * @param size Size of area.

 * @param base Base address of area.

 *

 * @return Address space area on success or NULL on failure.

 */

as_area_t *as_area_create(as_t *as, int flags, size_t size, __address base)

{

    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();

    spinlock_lock(&as->lock);

    if (!check_area_conflicts(as, base, size, NULL)) {

        spinlock_unlock(&as->lock);

        interrupts_restore(ipl);

        return NULL;

    }

    a = (as_area_t *) malloc(sizeof(as_area_t), 0);

    spinlock_initialize(&a->lock, "as_area_lock");

    a->flags = flags;

    a->pages = SIZE2FRAMES(size);

    a->base = base;

    btree_insert(&as->as_area_btree, base, (void *) a, NULL);

    spinlock_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 address on success, (__address) -1 otherwise.

 */

__address as_area_resize(as_t *as, __address address, size_t size, int flags)

{

    as_area_t *area = NULL;

    ipl_t ipl;

    size_t pages;

    ipl = interrupts_disable();

    spinlock_lock(&as->lock);

    /*

     * Locate the area.

     */

    area = find_area_and_lock(as, address);

    if (!area) {

        spinlock_unlock(&as->lock);

        interrupts_restore(ipl);

        return (__address) -1;

    }

    if (area->flags & AS_AREA_DEVICE) {

        /*

         * Remapping of address space areas associated

         * with memory mapped devices is not supported.

         */

        spinlock_unlock(&area->lock);

        spinlock_unlock(&as->lock);

        interrupts_restore(ipl);

        return (__address) -1;

    }

    pages = SIZE2FRAMES((address - area->base) + size);

    if (!pages) {

        /*

         * Zero size address space areas are not allowed.

         */

        spinlock_unlock(&area->lock);

        spinlock_unlock(&as->lock);

        interrupts_restore(ipl);

        return (__address) -1;

    }

    if (pages < area->pages) {

        int i;

        /*

         * Shrinking the area.

         * No need to check for overlaps.

         */

        for (i = pages; i < area->pages; i++) {

            pte_t *pte;

            /*

             * Releasing physical memory.

             * This depends on the fact that the memory was allocated using frame_alloc().

             */

            page_table_lock(as, false);

            pte = page_mapping_find(as, area->base + i*PAGE_SIZE);

            if (pte && PTE_VALID(pte)) {

                __address frame;

                ASSERT(PTE_PRESENT(pte));

                frame = PTE_GET_FRAME(pte);

                page_mapping_remove(as, area->base + i*PAGE_SIZE);

                page_table_unlock(as, false);

                frame_free(ADDR2PFN(frame));

            } else {

                page_table_unlock(as, false);

            }

        }

        /*

         * Invalidate TLB's.

         */

        tlb_shootdown_start(TLB_INVL_PAGES, AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);

        tlb_invalidate_pages(AS->asid, area->base + pages*PAGE_SIZE, area->pages - pages);

        tlb_shootdown_finalize();

    } else {

        /*

         * Growing the area.

         * Check for overlaps with other address space areas.

         */

        if (!check_area_conflicts(as, address, pages * PAGE_SIZE, area)) {

            spinlock_unlock(&area->lock);

            spinlock_unlock(&as->lock);    

            interrupts_restore(ipl);

            return (__address) -1;

        }

    }

    area->pages = pages;

    spinlock_unlock(&area->lock);

    spinlock_unlock(&as->lock);

    interrupts_restore(ipl);

    return address;

}

/** Send address space area to another task.

 *

 * Address space area is sent to the specified task.

 * If the destination task is willing to accept the

 * area, a new area is created according to the

 * source area. Moreover, any existing mapping

 * is copied as well, providing thus a mechanism

 * for sharing group of pages. The source address

 * space area and any associated mapping is preserved.

 *

 * @param id Task ID of the accepting task.

 * @param base Base address of the source address space area.

 * @param size Size of the source address space area.

 * @param flags Flags of the source address space area.

 *

 * @return 0 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.

 */

int as_area_send(task_id_t id, __address base, size_t size, int flags)

{

    ipl_t ipl;

    task_t *t;

    count_t i;

    as_t *as;

    __address dst_base;

    ipl = interrupts_disable();

    spinlock_lock(&tasks_lock);

    t = task_find_by_id(id);

    if (!NULL) {

        spinlock_unlock(&tasks_lock);

        interrupts_restore(ipl);

        return ENOENT;

    }

    spinlock_lock(&t->lock);

    spinlock_unlock(&tasks_lock);

    as = t->as;

    dst_base = (__address) t->accept_arg.base;

    if (as == AS) {

        /*

         * The two tasks share the entire address space.

         * Return error since there is no point in continuing.

         */

        spinlock_unlock(&t->lock);

        interrupts_restore(ipl);

        return EPERM;

    }

    if ((t->accept_arg.task_id != TASK->taskid) || (t->accept_arg.size != size) ||

        (t->accept_arg.flags != flags)) {

        /*

         * Discrepancy in either task ID, size or flags.

         */

        spinlock_unlock(&t->lock);

        interrupts_restore(ipl);

        return EPERM;

    }

    /*

     * Create copy of the address space area.

     */

    if (!as_area_create(as, flags, size, dst_base)) {

        /*

         * Destination address space area could not be created.

         */

        spinlock_unlock(&t->lock);

        interrupts_restore(ipl);

        return ENOMEM;

    }

    /*

     * NOTE: we have just introduced a race condition.

     * The destination task can try to attempt the newly

     * created area before its mapping is copied from

     * the source address space area. In result, frames

     * can get lost.

     *

     * Currently, this race is not solved, but one of the

     * possible solutions would be to sleep in as_page_fault()

     * when this situation is detected.

     */

    memsetb((__address) &t->accept_arg, sizeof(as_area_acptsnd_arg_t), 0);

    spinlock_unlock(&t->lock);

    /*

     * Avoid deadlock by first locking the address space with lower address.

     */

    if (as < AS) {

        spinlock_lock(&as->lock);

        spinlock_lock(&AS->lock);

    } else {

        spinlock_lock(&AS->lock);

        spinlock_lock(&as->lock);

    }

    for (i = 0; i < SIZE2FRAMES(size); i++) {

        pte_t *pte;

        __address frame;

        page_table_lock(AS, false);

        pte = page_mapping_find(AS, base + i*PAGE_SIZE);

        if (pte && PTE_VALID(pte)) {

            ASSERT(PTE_PRESENT(pte));

            frame = PTE_GET_FRAME(pte);

            if (!(flags & AS_AREA_DEVICE)) {

                /* TODO: increment frame reference count */

            }

            page_table_unlock(AS, false);

        } else {

            page_table_unlock(AS, false);

            continue;

        }

        page_table_lock(as, false);

        page_mapping_insert(as, dst_base + i*PAGE_SIZE, frame, area_flags_to_page_flags(flags));

        page_table_unlock(as, false);

    }

    spinlock_unlock(&AS->lock);

    spinlock_unlock(&as->lock);

    interrupts_restore(ipl);

    return 0;

}

/** Initialize mapping for one page of address space.

 *

 * This functions maps 'page' to 'frame' according

 * to attributes of the address space area to

 * wich 'page' belongs.

 *

 * @param as Target address space.

 * @param page Virtual page within the area.

 * @param frame Physical frame to which page will be mapped.

 */

void as_set_mapping(as_t *as, __address page, __address frame)

{

    as_area_t *area;

    ipl_t ipl;

    ipl = interrupts_disable();

    page_table_lock(as, true);

    area = find_area_and_lock(as, page);

    if (!area) {

        panic("page not part of any as_area\n");

    }

    page_mapping_insert(as, page, frame, get_area_flags(area));

    spinlock_unlock(&area->lock);

    page_table_unlock(as, true);

    interrupts_restore(ipl);

}

/** Handle page fault within the current address space.

 *

 * This is the high-level page fault handler.

 * Interrupts are assumed disabled.

 *

 * @param page Faulting page.

 *

 * @return 0 on page fault, 1 on success.

 */

int as_page_fault(__address page)

{

    pte_t *pte;

    as_area_t *area;

    __address frame;

    ASSERT(AS);

    spinlock_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.

         */

        spinlock_unlock(&AS->lock);

        return 0;

    }

    ASSERT(!(area->flags & AS_AREA_DEVICE));

    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)) {

            page_table_unlock(AS, false);

            spinlock_unlock(&area->lock);

            spinlock_unlock(&AS->lock);

            return 1;

        }

    }

    /*

     * In general, there can be several reasons that

     * can have caused this fault.

     *

     * - non-existent mapping: the area is a scratch

     *   area (e.g. stack) and so far has not been

     *   allocated a frame for the faulting page

     *

     * - non-present mapping: another possibility,

     *   currently not implemented, would be frame

     *   reuse; when this becomes a possibility,

     *   do not forget to distinguish between

     *   the different causes

     */

    frame = PFN2ADDR(frame_alloc(ONE_FRAME, 0));

    memsetb(PA2KA(frame), FRAME_SIZE, 0);

    /*

     * Map 'page' to 'frame'.

     * Note that TLB shootdown is not attempted as only new information is being

     * inserted into page tables.

     */

    page_mapping_insert(AS, page, frame, get_area_flags(area));

    page_table_unlock(AS, false);

    spinlock_unlock(&area->lock);

    spinlock_unlock(&AS->lock);

    return 1;

}

/** Switch address spaces.

 *

 * @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(&as_lock);

    /*

     * First, take care of the old address space.

     */

    if (old) {

        spinlock_lock(&old->lock);

        ASSERT(old->refcount);

        if((--old->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);

        }

        spinlock_unlock(&old->lock);

    }

    /*

     * Second, prepare the new address space.

     */

    spinlock_lock(&new->lock);

    if ((new->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);

    spinlock_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();

        spinlock_lock(&new->lock);

        new->asid = asid;

        spinlock_unlock(&new->lock);

    }

    spinlock_unlock(&as_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_DEVICE))

        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 get_area_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);

}

/** 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 as_locked 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 as_locked 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, __address 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 */

        spinlock_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];

        spinlock_lock(&a->lock);

        if ((a->base <= va) && (va < a->base + a->pages * PAGE_SIZE)) {

            return a;

        }

        spinlock_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];

        spinlock_lock(&a->lock);

        if (va < a->base + a->pages * PAGE_SIZE) {

            return a;

        }

        spinlock_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, __address 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];

        spinlock_lock(&a->lock);

        if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {

            spinlock_unlock(&a->lock);

            return false;

        }

        spinlock_unlock(&a->lock);

    }

    if ((node = btree_leaf_node_right_neighbour(&as->as_area_btree, leaf))) {

        a = (as_area_t *) node->value[0];

        spinlock_lock(&a->lock);

        if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {

            spinlock_unlock(&a->lock);

            return false;

        }

        spinlock_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;

        spinlock_lock(&a->lock);

        if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {

            spinlock_unlock(&a->lock);

            return false;

        }

        spinlock_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;

}

/*

 * Address space related syscalls.

 */

/** Wrapper for as_area_create(). */