Subversion Repositories HelenOS

/*

 * Copyright (C) 2001-2006 Jakub Jermar

 * All rights reserved.

 *

 * Redistribution and use in source and binary forms, with or without

 * modification, are permitted provided that the following conditions

 * are met:

 *

 * - Redistributions of source code must retain the above copyright

 *   notice, this list of conditions and the following disclaimer.

 * - Redistributions in binary form must reproduce the above copyright

 *   notice, this list of conditions and the following disclaimer in the

 *   documentation and/or other materials provided with the distribution.

 * - The name of the author may not be used to endorse or promote products

 *   derived from this software without specific prior written permission.

 *

 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR

 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES

 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.

 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,

 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT

 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,

 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY

 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT

 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF

 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 */

/** @addtogroup genericmm

 * @{

 */

/**

 * @file

 * @brief   Address space related functions.

 *

 * This file contains address space manipulation functions.

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

 * Virtual Address Translation (VAT) subsystem.

 *

 * Functionality provided by this file allows one to

 * create address spaces and create, resize and share

 * address space areas.

 *

 * @see page.c

 *

 */

#include <mm/as.h>

#include <arch/mm/as.h>

#include <mm/page.h>

#include <mm/frame.h>

#include <mm/slab.h>

#include <mm/tlb.h>

#include <arch/mm/page.h>

#include <genarch/mm/page_pt.h>

#include <genarch/mm/page_ht.h>

#include <mm/asid.h>

#include <arch/mm/asid.h>

#include <synch/spinlock.h>

#include <synch/mutex.h>

#include <adt/list.h>

#include <adt/btree.h>

#include <proc/task.h>

#include <proc/thread.h>

#include <arch/asm.h>

#include <panic.h>

#include <debug.h>

#include <print.h>

#include <memstr.h>

#include <macros.h>

#include <arch.h>

#include <errno.h>

#include <config.h>

#include <align.h>

#include <arch/types.h>

#include <typedefs.h>

#include <syscall/copy.h>

#include <arch/interrupt.h>

/**

 * Each architecture decides what functions will be used to carry out

 * address space operations such as creating or locking page tables.

 */

as_operations_t *as_operations = NULL;

/**

 * Slab for as_t objects.

 */

static slab_cache_t *as_slab;

/** This lock protects inactive_as_with_asid_head list. It must be acquired before as_t mutex. */

SPINLOCK_INITIALIZE(inactive_as_with_asid_lock);

/**

 * This list contains address spaces that are not active on any

 * processor and that have valid ASID.

 */

LIST_INITIALIZE(inactive_as_with_asid_head);

/** Kernel address space. */

as_t *AS_KERNEL = NULL;

static int area_flags_to_page_flags(int aflags);

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

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

static void sh_info_remove_reference(share_info_t *sh_info);

static int as_constructor(void *obj, int flags)

{

    as_t *as = (as_t *) obj;

    int rc;

    link_initialize(&as->inactive_as_with_asid_link);

    mutex_initialize(&as->lock);   

    rc = as_constructor_arch(as, flags);

    return rc;

}

static int as_destructor(void *obj)

{

    as_t *as = (as_t *) obj;

    return as_destructor_arch(as);

}

/** Initialize address space subsystem. */

void as_init(void)

{

    as_arch_init();

    as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,

        as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);

    AS_KERNEL = as_create(FLAG_AS_KERNEL);

    if (!AS_KERNEL)

        panic("can't create kernel address space\n");

}

/** Create address space.

 *

 * @param flags Flags that influence way in wich the address space is created.

 */

as_t *as_create(int flags)

{

    as_t *as;

    as = (as_t *) slab_alloc(as_slab, 0);

    (void) as_create_arch(as, 0);

    btree_create(&as->as_area_btree);

    if (flags & FLAG_AS_KERNEL)

        as->asid = ASID_KERNEL;

    else

        as->asid = ASID_INVALID;

    as->refcount = 0;

    as->cpu_refcount = 0;

    as->page_table = page_table_create(flags);

    return as;

}

/** Destroy adress space.

 *

 * When there are no tasks referencing this address space (i.e. its refcount is zero),

 * the address space can be destroyed.

 */

void as_destroy(as_t *as)

{

    ipl_t ipl;

    bool cond;

    ASSERT(as->refcount == 0);

    /*

     * Since there is no reference to this area,

     * it is safe not to lock its mutex.

     */

    ipl = interrupts_disable();

    spinlock_lock(&inactive_as_with_asid_lock);

    if (as->asid != ASID_INVALID && as != AS_KERNEL) {

        if (as != AS && as->cpu_refcount == 0)

            list_remove(&as->inactive_as_with_asid_link);

        asid_put(as->asid);

    }

    spinlock_unlock(&inactive_as_with_asid_lock);

    /*

     * Destroy address space areas of the address space.

     * The B+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.