xnu-1228.9.59.tar.gz

[apple/xnu.git] / osfmk / vm / vm_shared_region.c

/*
 * Copyright (c) 2007 Apple Inc. All rights reserved.
 *
 * @APPLE_LICENSE_HEADER_START@
 * 
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this
 * file.
 * 
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 * 
 * @APPLE_LICENSE_HEADER_END@
 */

/*
 * Shared region (... and comm page)
 *
 * This file handles the VM shared region and comm page.
 *
 */
/*
 * SHARED REGIONS
 * --------------
 *
 * A shared region is a submap that contains the most common system shared
 * libraries for a given environment.
 * An environment is defined by (cpu-type, 64-bitness, root directory).
 *
 * The point of a shared region is to reduce the setup overhead when exec'ing
 * a new process.
 * A shared region uses a shared VM submap that gets mapped automatically
 * at exec() time (see vm_map_exec()).  The first process of a given
 * environment sets up the shared region and all further processes in that
 * environment can re-use that shared region without having to re-create
 * the same mappings in their VM map.  All they need is contained in the shared
 * region.
 * It can also shared a pmap (mostly for read-only parts but also for the
 * initial version of some writable parts), which gets "nested" into the 
 * process's pmap.  This reduces the number of soft faults:  once one process
 * brings in a page in the shared region, all the other processes can access
 * it without having to enter it in their own pmap.
 *
 *
 * When a process is being exec'ed, vm_map_exec() calls vm_shared_region_enter()
 * to map the appropriate shared region in the process's address space.
 * We look up the appropriate shared region for the process's environment.
 * If we can't find one, we create a new (empty) one and add it to the list.
 * Otherwise, we just take an extra reference on the shared region we found.
 *
 * The "dyld" runtime (mapped into the process's address space at exec() time)
 * will then use the shared_region_check_np() and shared_region_map_np()
 * system call to validate and/or populate the shared region with the
 * appropriate dyld_shared_cache file.
 *
 * The shared region is inherited on fork() and the child simply takes an
 * extra reference on its parent's shared region.
 *
 * When the task terminates, we release a reference on its shared region.
 * When the last reference is released, we destroy the shared region.
 *
 * After a chroot(), the calling process keeps using its original shared region,
 * since that's what was mapped when it was started.  But its children
 * will use a different shared region, because they need to use the shared
 * cache that's relative to the new root directory.
 */
/*
 * COMM PAGE
 *
 * A "comm page" is an area of memory that is populated by the kernel with
 * the appropriate platform-specific version of some commonly used code.
 * There is one "comm page" per platform (cpu-type, 64-bitness) but only
 * for the native cpu-type.  No need to overly optimize translated code
 * for hardware that is not really there !
 *
 * The comm pages are created and populated at boot time.
 *
 * The appropriate comm page is mapped into a process's address space
 * at exec() time, in vm_map_exec().
 * It is then inherited on fork().
 *
 * The comm page is shared between the kernel and all applications of
 * a given platform.  Only the kernel can modify it.
 *
 * Applications just branch to fixed addresses in the comm page and find
 * the right version of the code for the platform.  There is also some
 * data provided and updated by the kernel for processes to retrieve easily
 * without having to do a system call.
 */

#include <debug.h>

#include <kern/ipc_tt.h>
#include <kern/kalloc.h>

#include <mach/mach_vm.h>

#include <vm/vm_map.h>
#include <vm/vm_shared_region.h>

#include <vm/vm_protos.h>

#include <machine/commpage.h>
#include <machine/cpu_capabilities.h>

/* "dyld" uses this to figure out what the kernel supports */
int shared_region_version = 3;

/* should local (non-chroot) shared regions persist when no task uses them ? */
int shared_region_persistence = 1;      /* yes by default */

/* trace level, output is sent to the system log file */
int shared_region_trace_level = SHARED_REGION_TRACE_ERROR_LVL;

/* this lock protects all the shared region data structures */
lck_grp_t *vm_shared_region_lck_grp;
lck_mtx_t vm_shared_region_lock;

#define vm_shared_region_lock() lck_mtx_lock(&vm_shared_region_lock)
#define vm_shared_region_unlock() lck_mtx_unlock(&vm_shared_region_lock)
#define vm_shared_region_sleep(event, interruptible)                    \
        lck_mtx_sleep(&vm_shared_region_lock,                           \
                      LCK_SLEEP_DEFAULT,                                \
                      (event_t) (event),                                \
                      (interruptible))

/* the list of currently available shared regions (one per environment) */
queue_head_t    vm_shared_region_queue;

static void vm_shared_region_reference_locked(vm_shared_region_t shared_region);
static vm_shared_region_t vm_shared_region_create(
        void                    *root_dir,
        cpu_type_t              cputype,
        boolean_t               is_64bit);
static void vm_shared_region_destroy(vm_shared_region_t shared_region);

/*
 * Initialize the module...
 */
void
vm_shared_region_init(void)
{
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> init\n"));

        vm_shared_region_lck_grp = lck_grp_alloc_init("vm shared region",
                                                      LCK_GRP_ATTR_NULL);
        lck_mtx_init(&vm_shared_region_lock,
                     vm_shared_region_lck_grp,
                     LCK_ATTR_NULL);

        queue_init(&vm_shared_region_queue);

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: <- init\n"));
}

/*
 * Retrieve a task's shared region and grab an extra reference to 
 * make sure it doesn't disappear while the caller is using it. 
 * The caller is responsible for consuming that extra reference if
 * necessary.
 */
vm_shared_region_t
vm_shared_region_get(
        task_t          task)
{
        vm_shared_region_t      shared_region;

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> get(%p)\n",
                 task));

        task_lock(task);
        vm_shared_region_lock();
        shared_region = task->shared_region;
        if (shared_region) {
                assert(shared_region->sr_ref_count > 0);
                vm_shared_region_reference_locked(shared_region);
        }
        vm_shared_region_unlock();
        task_unlock(task);

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: get(%p) <- %p\n",
                 task, shared_region));

        return shared_region;
}

/*
 * Get the base address of the shared region.
 * That's the address at which it needs to be mapped in the process's address
 * space.
 * No need to lock since this data is set when the shared region is
 * created and is never modified after that.  The caller must hold an extra
 * reference on the shared region to prevent it from being destroyed.
 */
mach_vm_offset_t
vm_shared_region_base_address(
        vm_shared_region_t      shared_region)
{
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> base_address(%p)\n",
                 shared_region));
        assert(shared_region->sr_ref_count > 1);
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: base_address(%p) <- 0x%llx\n",
                 shared_region, (long long)shared_region->sr_base_address));
        return shared_region->sr_base_address;
}

/*
 * Get the size of the shared region.
 * That's the size that needs to be mapped in the process's address
 * space.
 * No need to lock since this data is set when the shared region is
 * created and is never modified after that.  The caller must hold an extra
 * reference on the shared region to prevent it from being destroyed.
 */
mach_vm_size_t
vm_shared_region_size(
        vm_shared_region_t      shared_region)
{
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> size(%p)\n",
                 shared_region));
        assert(shared_region->sr_ref_count > 1);
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: size(%p) <- 0x%llx\n",
                 shared_region, (long long)shared_region->sr_size));
        return shared_region->sr_size;
}

/*
 * Get the memory entry of the shared region.
 * That's the "memory object" that needs to be mapped in the process's address
 * space.
 * No need to lock since this data is set when the shared region is
 * created and is never modified after that.  The caller must hold an extra
 * reference on the shared region to prevent it from being destroyed.
 */
ipc_port_t
vm_shared_region_mem_entry(
        vm_shared_region_t      shared_region)
{
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> mem_entry(%p)\n",
                 shared_region));
        assert(shared_region->sr_ref_count > 1);
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: mem_entry(%p) <- %p\n",
                 shared_region, shared_region->sr_mem_entry));
        return shared_region->sr_mem_entry;
}

/*
 * Set the shared region the process should use.
 * A NULL new shared region means that we just want to release the old
 * shared region.
 * The caller should already have an extra reference on the new shared region
 * (if any).  We release a reference on the old shared region (if any).
 */
void
vm_shared_region_set(
        task_t                  task,
        vm_shared_region_t      new_shared_region)
{
        vm_shared_region_t      old_shared_region;

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> set(%p, %p)\n",
                 task, new_shared_region));

        task_lock(task);
        vm_shared_region_lock();

        old_shared_region = task->shared_region;
        if (new_shared_region) {
                assert(new_shared_region->sr_ref_count > 0);
        }

        task->shared_region = new_shared_region;

        vm_shared_region_unlock();
        task_unlock(task);

        if (old_shared_region) {
                assert(old_shared_region->sr_ref_count > 0);
                vm_shared_region_deallocate(old_shared_region);
        }

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: set(%p) <- old=%p new=%p\n",
                 task, old_shared_region, new_shared_region));
}

/*
 * Lookup up the shared region for the desired environment.
 * If none is found, create a new (empty) one.
 * Grab an extra reference on the returned shared region, to make sure
 * it doesn't get destroyed before the caller is done with it.  The caller
 * is responsible for consuming that extra reference if necessary.
 */
vm_shared_region_t
vm_shared_region_lookup(
        void            *root_dir,
        cpu_type_t      cputype,
        boolean_t       is_64bit)
{
        vm_shared_region_t      shared_region;
        vm_shared_region_t      new_shared_region;

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> lookup(root=%p,cpu=%d,64bit=%d)\n",
                 root_dir, cputype, is_64bit));

        shared_region = NULL;
        new_shared_region = NULL;

        vm_shared_region_lock();
        for (;;) {
                queue_iterate(&vm_shared_region_queue,
                              shared_region,
                              vm_shared_region_t,
                              sr_q) {
                        assert(shared_region->sr_ref_count > 0);
                        if (shared_region->sr_cpu_type == cputype &&
                            shared_region->sr_root_dir == root_dir &&
                            shared_region->sr_64bit == is_64bit) {
                                /* found a match ! */
                                vm_shared_region_reference_locked(shared_region);
                                goto done;
                        }
                }
                if (new_shared_region == NULL) {
                        /* no match: create a new one */
                        vm_shared_region_unlock();
                        new_shared_region = vm_shared_region_create(root_dir,
                                                                    cputype,
                                                                    is_64bit);
                        /* do the lookup again, in case we lost a race */
                        vm_shared_region_lock();
                        continue;
                }
                /* still no match: use our new one */
                shared_region = new_shared_region;
                new_shared_region = NULL;
                queue_enter(&vm_shared_region_queue,
                            shared_region,
                            vm_shared_region_t,
                            sr_q);
                break;
        }

done:
        vm_shared_region_unlock();

        if (new_shared_region) {
                /*
                 * We lost a race with someone else to create a new shared
                 * region for that environment.  Get rid of our unused one.
                 */
                assert(new_shared_region->sr_ref_count == 1);
                new_shared_region->sr_ref_count--;
                vm_shared_region_destroy(new_shared_region);
                new_shared_region = NULL;
        }

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: lookup(root=%p,cpu=%d,64bit=%d) <- %p\n",
                 root_dir, cputype, is_64bit, shared_region));

        assert(shared_region->sr_ref_count > 0);
        return shared_region;
}

/*
 * Take an extra reference on a shared region.
 * The vm_shared_region_lock should already be held by the caller.
 */
static void
vm_shared_region_reference_locked(
        vm_shared_region_t      shared_region)
{
#if DEBUG
        lck_mtx_assert(&vm_shared_region_lock, LCK_MTX_ASSERT_OWNED);
#endif

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> reference_locked(%p)\n",
                 shared_region));
        assert(shared_region->sr_ref_count > 0);
        shared_region->sr_ref_count++;
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: reference_locked(%p) <- %d\n",
                 shared_region, shared_region->sr_ref_count));
}

/*
 * Release a reference on the shared region.
 * Destroy it if there are no references left.
 */
void
vm_shared_region_deallocate(
        vm_shared_region_t      shared_region)
{
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> deallocate(%p)\n",
                 shared_region));

        vm_shared_region_lock();
        
        assert(shared_region->sr_ref_count > 0);

        if (shared_region->sr_root_dir == NULL) {
                /*
                 * Local (i.e. based on the boot volume) shared regions
                 * can persist or not based on the "shared_region_persistence"
                 * sysctl.
                 * Make sure that this one complies.
                 */
                if (shared_region_persistence &&
                    !shared_region->sr_persists) {
                        /* make this one persistent */
                        shared_region->sr_ref_count++;
                        shared_region->sr_persists = TRUE;
                } else if (!shared_region_persistence &&
                           shared_region->sr_persists) {
                        /* make this one no longer persistent */
                        assert(shared_region->sr_ref_count > 1);
                        shared_region->sr_ref_count--;
                        shared_region->sr_persists = FALSE;
                }
        }

        assert(shared_region->sr_ref_count > 0);
        shared_region->sr_ref_count--;
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: deallocate(%p): ref now %d\n",
                 shared_region, shared_region->sr_ref_count));

        if (shared_region->sr_ref_count == 0) {
                assert(! shared_region->sr_mapping_in_progress);
                /* remove it from the queue first, so no one can find it... */
                queue_remove(&vm_shared_region_queue,
                             shared_region,
                             vm_shared_region_t,
                             sr_q);
                vm_shared_region_unlock();
                /* ... and destroy it */
                vm_shared_region_destroy(shared_region);
                shared_region = NULL;
        } else {
                vm_shared_region_unlock();
        }

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: deallocate(%p) <-\n",
                 shared_region));
}

/*
 * Create a new (empty) shared region for a new environment.
 */
static vm_shared_region_t
vm_shared_region_create(
        void                    *root_dir,
        cpu_type_t              cputype,
        boolean_t               is_64bit)
{
        kern_return_t           kr;
        vm_named_entry_t        mem_entry;
        ipc_port_t              mem_entry_port;
        vm_shared_region_t      shared_region;
        vm_map_t                sub_map;
        mach_vm_offset_t        base_address, pmap_nesting_start;
        mach_vm_size_t          size, pmap_nesting_size;

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> create(root=%p,cpu=%d,64bit=%d)\n",
                 root_dir, cputype, is_64bit));

        base_address = 0;
        size = 0;
        mem_entry = NULL;
        mem_entry_port = IPC_PORT_NULL;
        sub_map = VM_MAP_NULL;

        /* create a new shared region structure... */
        shared_region = kalloc(sizeof (*shared_region));
        if (shared_region == NULL) {
                SHARED_REGION_TRACE_ERROR(
                        ("shared_region: create: couldn't allocate\n"));
                goto done;
        }

        /* figure out the correct settings for the desired environment */
        if (is_64bit) {
                switch (cputype) {
                case CPU_TYPE_I386:
                        base_address = SHARED_REGION_BASE_X86_64;
                        size = SHARED_REGION_SIZE_X86_64;
                        pmap_nesting_start = SHARED_REGION_NESTING_BASE_X86_64;
                        pmap_nesting_size = SHARED_REGION_NESTING_SIZE_X86_64;
                        break;
                case CPU_TYPE_POWERPC:
                        base_address = SHARED_REGION_BASE_PPC64;
                        size = SHARED_REGION_SIZE_PPC64;
                        pmap_nesting_start = SHARED_REGION_NESTING_BASE_PPC64;
                        pmap_nesting_size = SHARED_REGION_NESTING_SIZE_PPC64;
                        break;
                default:
                        SHARED_REGION_TRACE_ERROR(
                                ("shared_region: create: unknown cpu type %d\n",
                                 cputype));
                        kfree(shared_region, sizeof (*shared_region));
                        shared_region = NULL;
                        goto done;
                }
        } else {
                switch (cputype) {
                case CPU_TYPE_I386:
                        base_address = SHARED_REGION_BASE_I386;
                        size = SHARED_REGION_SIZE_I386;
                        pmap_nesting_start = SHARED_REGION_NESTING_BASE_I386;
                        pmap_nesting_size = SHARED_REGION_NESTING_SIZE_I386;
                        break;
                case CPU_TYPE_POWERPC:
                        base_address = SHARED_REGION_BASE_PPC;
                        size = SHARED_REGION_SIZE_PPC;
                        pmap_nesting_start = SHARED_REGION_NESTING_BASE_PPC;
                        pmap_nesting_size = SHARED_REGION_NESTING_SIZE_PPC;
                        break;
#ifdef CPU_TYPE_ARM
                case CPU_TYPE_ARM:
                        base_address = SHARED_REGION_BASE_ARM;
                        size = SHARED_REGION_SIZE_ARM;
                        pmap_nesting_start = SHARED_REGION_NESTING_BASE_ARM;
                        pmap_nesting_size = SHARED_REGION_NESTING_SIZE_ARM;
                        break;
#endif /* CPU_TYPE_ARM */
                default:
                        SHARED_REGION_TRACE_ERROR(
                                ("shared_region: create: unknown cpu type %d\n",
                                 cputype));
                        kfree(shared_region, sizeof (*shared_region));
                        shared_region = NULL;
                        goto done;
                        
                }
        }

        /* create a memory entry structure and a Mach port handle */
        kr = mach_memory_entry_allocate(&mem_entry,
                                        &mem_entry_port);
        if (kr != KERN_SUCCESS) {
                kfree(shared_region, sizeof (*shared_region));
                shared_region = NULL;
                SHARED_REGION_TRACE_ERROR(
                        ("shared_region: create: "
                         "couldn't allocate mem_entry\n"));
                goto done;
        }

        /* create a VM sub map and its pmap */
        sub_map = vm_map_create(pmap_create(0, is_64bit),
                                0, size,
                                TRUE);
        if (sub_map == VM_MAP_NULL) {
                ipc_port_release_send(mem_entry_port);
                kfree(shared_region, sizeof (*shared_region));
                shared_region = NULL;
                SHARED_REGION_TRACE_ERROR(
                        ("shared_region: create: "
                         "couldn't allocate map\n"));
                goto done;
        }

        /* make the memory entry point to the VM sub map */
        mem_entry->is_sub_map = TRUE;
        mem_entry->backing.map = sub_map;
        mem_entry->size = size;
        mem_entry->protection = VM_PROT_ALL;

        /* make the shared region point at the memory entry */
        shared_region->sr_mem_entry = mem_entry_port;

        /* fill in the shared region's environment and settings */
        shared_region->sr_base_address = base_address;
        shared_region->sr_size = size;
        shared_region->sr_pmap_nesting_start = pmap_nesting_start;
        shared_region->sr_pmap_nesting_size = pmap_nesting_size;
        shared_region->sr_cpu_type = cputype;
        shared_region->sr_64bit = is_64bit;
        shared_region->sr_root_dir = root_dir;

        queue_init(&shared_region->sr_q);
        shared_region->sr_mapping_in_progress = FALSE;
        shared_region->sr_persists = FALSE;
        shared_region->sr_first_mapping = (mach_vm_offset_t) -1;

        /* grab a reference for the caller */
        shared_region->sr_ref_count = 1;

done:
        if (shared_region) {
                SHARED_REGION_TRACE_INFO(
                        ("shared_region: create(root=%p,cpu=%d,64bit=%d,"
                         "base=0x%llx,size=0x%llx) <- "
                         "%p mem=(%p,%p) map=%p pmap=%p\n",
                         root_dir, cputype, is_64bit, (long long)base_address,
                         (long long)size, shared_region,
                         mem_entry_port, mem_entry, sub_map, sub_map->pmap));
        } else {
                SHARED_REGION_TRACE_INFO(
                        ("shared_region: create(root=%p,cpu=%d,64bit=%d,"
                         "base=0x%llx,size=0x%llx) <- NULL",
                         root_dir, cputype, is_64bit, (long long)base_address,
                         (long long)size));
        }
        return shared_region;
}

/*
 * Destroy a now-unused shared region.
 * The shared region is no longer in the queue and can not be looked up.
 */
static void
vm_shared_region_destroy(
        vm_shared_region_t      shared_region)
{
        vm_named_entry_t        mem_entry;
        vm_map_t                map;

        SHARED_REGION_TRACE_INFO(
                ("shared_region: -> destroy(%p) (root=%p,cpu=%d,64bit=%d)\n",
                 shared_region,
                 shared_region->sr_root_dir,
                 shared_region->sr_cpu_type,
                 shared_region->sr_64bit));

        assert(shared_region->sr_ref_count == 0);
        assert(!shared_region->sr_persists);

        mem_entry = (vm_named_entry_t) shared_region->sr_mem_entry->ip_kobject;
        assert(mem_entry->is_sub_map);
        assert(!mem_entry->internal);
        assert(!mem_entry->is_pager);
        map = mem_entry->backing.map;

        /*
         * Clean up the pmap first.  The virtual addresses that were
         * entered in this possibly "nested" pmap may have different values
         * than the VM map's min and max offsets, if the VM sub map was
         * mapped at a non-zero offset in the processes' main VM maps, which
         * is usually the case, so the clean-up we do in vm_map_destroy() would
         * not be enough.
         */
        if (map->pmap) {
                pmap_remove(map->pmap,
                            shared_region->sr_base_address,
                            (shared_region->sr_base_address +
                             shared_region->sr_size));
        }

        /*
         * Release our (one and only) handle on the memory entry.
         * This will generate a no-senders notification, which will be processed
         * by ipc_kobject_notify(), which will release the one and only
         * reference on the memory entry and cause it to be destroyed, along
         * with the VM sub map and its pmap.
         */
        mach_memory_entry_port_release(shared_region->sr_mem_entry);
        mem_entry = NULL;
        shared_region->sr_mem_entry = IPC_PORT_NULL;

        /* release the shared region structure... */
        kfree(shared_region, sizeof (*shared_region));
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: destroy(%p) <-\n",
                 shared_region));
        shared_region = NULL;

}

/*
 * Gets the address of the first (in time) mapping in the shared region.
 */
kern_return_t
vm_shared_region_start_address(
        vm_shared_region_t      shared_region,
        mach_vm_offset_t        *start_address)
{
        kern_return_t           kr;
        mach_vm_offset_t        sr_base_address;
        mach_vm_offset_t        sr_first_mapping;

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> start_address(%p)\n",
                 shared_region));
        assert(shared_region->sr_ref_count > 1);

        vm_shared_region_lock();

        /*
         * Wait if there's another thread establishing a mapping
         * in this shared region right when we're looking at it.
         * We want a consistent view of the map...
         */
        while (shared_region->sr_mapping_in_progress) {
                /* wait for our turn... */
                assert(shared_region->sr_ref_count > 1);
                vm_shared_region_sleep(&shared_region->sr_mapping_in_progress,
                                       THREAD_UNINT);
        }
        assert(! shared_region->sr_mapping_in_progress);
        assert(shared_region->sr_ref_count > 1);
        
        sr_base_address = shared_region->sr_base_address;
        sr_first_mapping = shared_region->sr_first_mapping;

        if (sr_first_mapping == (mach_vm_offset_t) -1) {
                /* shared region is empty */
                kr = KERN_INVALID_ADDRESS;
        } else {
                kr = KERN_SUCCESS;
                *start_address = sr_base_address + sr_first_mapping;
        }

        vm_shared_region_unlock();
        
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: start_address(%p) <- 0x%llx\n",
                 shared_region, (long long)shared_region->sr_base_address));

        return kr;
}
/*
 * Establish some mappings of a file in the shared region.
 * This is used by "dyld" via the shared_region_map_np() system call
 * to populate the shared region with the appropriate shared cache.
 *
 * One could also call it several times to incrementally load several
 * libraries, as long as they do not overlap.  
 * It will return KERN_SUCCESS if the mappings were successfully established
 * or if they were already established identically by another process.
 */
kern_return_t
vm_shared_region_map_file(
        vm_shared_region_t              shared_region,
        unsigned int                    mappings_count,
        struct shared_file_mapping_np   *mappings,
        memory_object_control_t         file_control,
        memory_object_size_t            file_size,
        void                            *root_dir)
{
        kern_return_t           kr;
        vm_object_t             file_object;
        ipc_port_t              sr_handle;
        vm_named_entry_t        sr_mem_entry;
        vm_map_t                sr_map;
        mach_vm_offset_t        sr_base_address;
        unsigned int            i;
        mach_port_t             map_port;
        mach_vm_offset_t        target_address;
        vm_object_t             object;
        vm_object_size_t        obj_size;


        kr = KERN_SUCCESS;

        vm_shared_region_lock();
        assert(shared_region->sr_ref_count > 1);

        if (shared_region->sr_root_dir != root_dir) {
                /*
                 * This shared region doesn't match the current root
                 * directory of this process.  Deny the mapping to
                 * avoid tainting the shared region with something that 
                 * doesn't quite belong into it.
                 */
                vm_shared_region_unlock();
                kr = KERN_PROTECTION_FAILURE;
                goto done;
        }

        /*
         * Make sure we handle only one mapping at a time in a given
         * shared region, to avoid race conditions.  This should not
         * happen frequently...
         */
        while (shared_region->sr_mapping_in_progress) {
                /* wait for our turn... */
                vm_shared_region_sleep(&shared_region->sr_mapping_in_progress,
                                       THREAD_UNINT);
        }
        assert(! shared_region->sr_mapping_in_progress);
        assert(shared_region->sr_ref_count > 1);
        /* let others know we're working in this shared region */
        shared_region->sr_mapping_in_progress = TRUE;

        vm_shared_region_unlock();

        /* no need to lock because this data is never modified... */
        sr_handle = shared_region->sr_mem_entry;
        sr_mem_entry = (vm_named_entry_t) sr_handle->ip_kobject;
        sr_map = sr_mem_entry->backing.map;
        sr_base_address = shared_region->sr_base_address;

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> map(%p,%d,%p,%p,0x%llx)\n",
                 shared_region, mappings_count, mappings,
                 file_control, file_size));

        /* get the VM object associated with the file to be mapped */
        file_object = memory_object_control_to_vm_object(file_control);

        /* establish the mappings */
        for (i = 0; i < mappings_count; i++) {
                SHARED_REGION_TRACE_INFO(
                        ("shared_region: mapping[%d]: "
                         "address:0x%016llx size:0x%016llx offset:0x%016llx "
                         "maxprot:0x%x prot:0x%x\n",
                         i,
                         (long long)mappings[i].sfm_address,
                         (long long)mappings[i].sfm_size,
                         (long long)mappings[i].sfm_file_offset,
                         mappings[i].sfm_max_prot,
                         mappings[i].sfm_init_prot));

                if (mappings[i].sfm_init_prot & VM_PROT_ZF) {
                        /* zero-filled memory */
                        map_port = MACH_PORT_NULL;
                } else {
                        /* file-backed memory */
                        map_port = (ipc_port_t) file_object->pager;
                }

                /* mapping's address is relative to the shared region base */
                target_address =
                        mappings[i].sfm_address - sr_base_address;

                /* establish that mapping, OK if it's "already" there */
                if (map_port == MACH_PORT_NULL) {
                        /*
                         * We want to map some anonymous memory in a
                         * shared region.
                         * We have to create the VM object now, so that it
                         * can be mapped "copy-on-write".
                         */
                        obj_size = vm_map_round_page(mappings[i].sfm_size);
                        object = vm_object_allocate(obj_size);
                        if (object == VM_OBJECT_NULL) {
                                kr = KERN_RESOURCE_SHORTAGE;
                        } else {
                                kr = vm_map_enter(
                                        sr_map,
                                        &target_address,
                                        vm_map_round_page(mappings[i].sfm_size),
                                        0,
                                        VM_FLAGS_FIXED | VM_FLAGS_ALREADY,
                                        object,
                                        0,
                                        TRUE,
                                        mappings[i].sfm_init_prot & VM_PROT_ALL,
                                        mappings[i].sfm_max_prot & VM_PROT_ALL,
                                        VM_INHERIT_DEFAULT);
                        }
                } else {
                        object = VM_OBJECT_NULL; /* no anonymous memory here */
                        kr = vm_map_enter_mem_object(
                                sr_map,
                                &target_address,
                                vm_map_round_page(mappings[i].sfm_size),
                                0,
                                VM_FLAGS_FIXED | VM_FLAGS_ALREADY,
                                map_port,
                                mappings[i].sfm_file_offset,
                                TRUE,
                                mappings[i].sfm_init_prot & VM_PROT_ALL,
                                mappings[i].sfm_max_prot & VM_PROT_ALL,
                                VM_INHERIT_DEFAULT);
                }

                if (kr != KERN_SUCCESS) {
                        if (map_port == MACH_PORT_NULL) {
                                /*
                                 * Get rid of the VM object we just created
                                 * but failed to map.
                                 */
                                vm_object_deallocate(object);
                                object = VM_OBJECT_NULL;
                        }
                        if (kr == KERN_MEMORY_PRESENT) {
                                /*
                                 * This exact mapping was already there:
                                 * that's fine.
                                 */
                                SHARED_REGION_TRACE_INFO(
                                        ("shared_region: mapping[%d]: "
                                         "address:0x%016llx size:0x%016llx "
                                         "offset:0x%016llx "
                                         "maxprot:0x%x prot:0x%x "
                                         "already mapped...\n",
                                         i,
                                         (long long)mappings[i].sfm_address,
                                         (long long)mappings[i].sfm_size,
                                         (long long)mappings[i].sfm_file_offset,
                                         mappings[i].sfm_max_prot,
                                         mappings[i].sfm_init_prot));
                                /*
                                 * We didn't establish this mapping ourselves;
                                 * let's reset its size, so that we do not
                                 * attempt to undo it if an error occurs later.
                                 */
                                mappings[i].sfm_size = 0;
                                kr = KERN_SUCCESS;
                        } else {
                                unsigned int j;

                                /* this mapping failed ! */
                                SHARED_REGION_TRACE_ERROR(
                                        ("shared_region: mapping[%d]: "
                                         "address:0x%016llx size:0x%016llx "
                                         "offset:0x%016llx "
                                         "maxprot:0x%x prot:0x%x failed 0x%x\n",
                                         i,
                                         (long long)mappings[i].sfm_address,
                                         (long long)mappings[i].sfm_size,
                                         (long long)mappings[i].sfm_file_offset,
                                         mappings[i].sfm_max_prot,
                                         mappings[i].sfm_init_prot,
                                         kr));

                                /*
                                 * Undo the mappings we've established so far.
                                 */
                                for (j = 0; j < i; j++) {
                                        kern_return_t kr2;

                                        if (mappings[j].sfm_size == 0) {
                                                /*
                                                 * We didn't establish this
                                                 * mapping, so nothing to undo.
                                                 */
                                                continue;
                                        }
                                        SHARED_REGION_TRACE_INFO(
                                                ("shared_region: mapping[%d]: "
                                                 "address:0x%016llx "
                                                 "size:0x%016llx "
                                                 "offset:0x%016llx "
                                                 "maxprot:0x%x prot:0x%x: "
                                                 "undoing...\n",
                                                 j,
                                                 (long long)mappings[j].sfm_address,
                                                 (long long)mappings[j].sfm_size,
                                                 (long long)mappings[j].sfm_file_offset,
                                                 mappings[j].sfm_max_prot,
                                                 mappings[j].sfm_init_prot));
                                        kr2 = mach_vm_deallocate(
                                                sr_map,
                                                (mappings[j].sfm_address -
                                                 sr_base_address),
                                                mappings[j].sfm_size);
                                        assert(kr2 == KERN_SUCCESS);
                                }

                                break;
                        }

                }

                /*
                 * Record the first (chronologically) mapping in
                 * this shared region.
                 * We're protected by "sr_mapping_in_progress" here,
                 * so no need to lock "shared_region".
                 */
                if (shared_region->sr_first_mapping == (mach_vm_offset_t) -1) {
                        shared_region->sr_first_mapping = target_address;
                }
        }

        vm_shared_region_lock();
        assert(shared_region->sr_ref_count > 1);
        assert(shared_region->sr_mapping_in_progress);
        /* we're done working on that shared region */
        shared_region->sr_mapping_in_progress = FALSE;
        thread_wakeup((event_t) &shared_region->sr_mapping_in_progress);
        vm_shared_region_unlock();

done:
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: map(%p,%d,%p,%p,0x%llx) <- 0x%x \n",
                 shared_region, mappings_count, mappings,
                 file_control, file_size, kr));
        return kr;
}

/*
 * Enter the appropriate shared region into "map" for "task".
 * This involves looking up the shared region (and possibly creating a new
 * one) for the desired environment, then mapping the VM sub map into the
 * task's VM "map", with the appropriate level of pmap-nesting.
 */
kern_return_t
vm_shared_region_enter(
        struct _vm_map          *map,
        struct task             *task,
        void                    *fsroot,
        cpu_type_t              cpu)
{
        kern_return_t           kr;
        vm_shared_region_t      shared_region;
        vm_map_offset_t         sr_address, sr_offset, target_address;
        vm_map_size_t           sr_size, mapping_size;
        vm_map_offset_t         sr_pmap_nesting_start;
        vm_map_size_t           sr_pmap_nesting_size;
        ipc_port_t              sr_handle;
        boolean_t               is_64bit;

        is_64bit = task_has_64BitAddr(task);

        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: -> "
                 "enter(map=%p,task=%p,root=%p,cpu=%d,64bit=%d)\n",
                 map, task, fsroot, cpu, is_64bit));

        /* lookup (create if needed) the shared region for this environment */
        shared_region = vm_shared_region_lookup(fsroot, cpu, is_64bit);
        if (shared_region == NULL) {
                /* this should not happen ! */
                SHARED_REGION_TRACE_ERROR(
                        ("shared_region: -> "
                         "enter(map=%p,task=%p,root=%p,cpu=%d,64bit=%d): "
                         "lookup failed !\n",
                         map, task, fsroot, cpu, is_64bit));
                //panic("shared_region_enter: lookup failed\n");
                return KERN_FAILURE;
        }
        
        /* let the task use that shared region */
        vm_shared_region_set(task, shared_region);

        kr = KERN_SUCCESS;
        /* no need to lock since this data is never modified */
        sr_address = shared_region->sr_base_address;
        sr_size = shared_region->sr_size;
        sr_handle = shared_region->sr_mem_entry;
        sr_pmap_nesting_start = shared_region->sr_pmap_nesting_start;
        sr_pmap_nesting_size = shared_region->sr_pmap_nesting_size;

        /*
         * Start mapping the shared region's VM sub map into the task's VM map.
         */
        sr_offset = 0;

        if (sr_pmap_nesting_start > sr_address) {
                /* we need to map a range without pmap-nesting first */
                target_address = sr_address;
                mapping_size = sr_pmap_nesting_start - sr_address;
                kr = vm_map_enter_mem_object(
                        map,
                        &target_address,
                        mapping_size,
                        0,
                        VM_FLAGS_FIXED,
                        sr_handle,
                        sr_offset,
                        TRUE,
                        VM_PROT_READ,
                        VM_PROT_ALL,
                        VM_INHERIT_SHARE);
                if (kr != KERN_SUCCESS) {
                        SHARED_REGION_TRACE_ERROR(
                                ("shared_region: enter(%p,%p,%p,%d,%d): "
                                 "vm_map_enter(0x%llx,0x%llx,%p) error 0x%x\n",
                                 map, task, fsroot, cpu, is_64bit,
                                 (long long)target_address,
                                 (long long)mapping_size, sr_handle, kr));
                        goto done;
                }
                SHARED_REGION_TRACE_DEBUG(
                        ("shared_region: enter(%p,%p,%p,%d,%d): "
                         "vm_map_enter(0x%llx,0x%llx,%p) error 0x%x\n",
                         map, task, fsroot, cpu, is_64bit,
                         (long long)target_address, (long long)mapping_size,
                         sr_handle, kr));
                sr_offset += mapping_size;
                sr_size -= mapping_size;
        }
        /*
         * We may need to map several pmap-nested portions, due to platform
         * specific restrictions on pmap nesting.
         * The pmap-nesting is triggered by the "VM_MEMORY_SHARED_PMAP" alias...
         */
        for (;
             sr_pmap_nesting_size > 0;
             sr_offset += mapping_size,
                     sr_size -= mapping_size,
                     sr_pmap_nesting_size -= mapping_size) {
                target_address = sr_address + sr_offset;
                mapping_size = sr_pmap_nesting_size;
                if (mapping_size > pmap_nesting_size_max) {
                        mapping_size = (vm_map_offset_t) pmap_nesting_size_max;
                }
                kr = vm_map_enter_mem_object(
                        map,
                        &target_address,
                        mapping_size,
                        0,
                        (VM_FLAGS_FIXED | VM_MAKE_TAG(VM_MEMORY_SHARED_PMAP)),
                        sr_handle,
                        sr_offset,
                        TRUE,
                        VM_PROT_READ,
                        VM_PROT_ALL,
                        VM_INHERIT_SHARE);
                if (kr != KERN_SUCCESS) {
                        SHARED_REGION_TRACE_ERROR(
                                ("shared_region: enter(%p,%p,%p,%d,%d): "
                                 "vm_map_enter(0x%llx,0x%llx,%p) error 0x%x\n",
                                 map, task, fsroot, cpu, is_64bit,
                                 (long long)target_address,
                                 (long long)mapping_size, sr_handle, kr));
                        goto done;
                }
                SHARED_REGION_TRACE_DEBUG(
                        ("shared_region: enter(%p,%p,%p,%d,%d): "
                         "nested vm_map_enter(0x%llx,0x%llx,%p) error 0x%x\n",
                         map, task, fsroot, cpu, is_64bit,
                         (long long)target_address, (long long)mapping_size,
                         sr_handle, kr));
        }
        if (sr_size > 0) {
                /* and there's some left to be mapped without pmap-nesting */
                target_address = sr_address + sr_offset;
                mapping_size = sr_size;
                kr = vm_map_enter_mem_object(
                        map,
                        &target_address,
                        mapping_size,
                        0,
                        VM_FLAGS_FIXED,
                        sr_handle,
                        sr_offset,
                        TRUE,
                        VM_PROT_READ,
                        VM_PROT_ALL,
                        VM_INHERIT_SHARE);
                if (kr != KERN_SUCCESS) {
                        SHARED_REGION_TRACE_ERROR(
                                ("shared_region: enter(%p,%p,%p,%d,%d): "
                                 "vm_map_enter(0x%llx,0x%llx,%p) error 0x%x\n",
                                 map, task, fsroot, cpu, is_64bit,
                                 (long long)target_address,
                                 (long long)mapping_size, sr_handle, kr));
                        goto done;
                }
                SHARED_REGION_TRACE_DEBUG(
                        ("shared_region: enter(%p,%p,%p,%d,%d): "
                         "vm_map_enter(0x%llx,0x%llx,%p) error 0x%x\n",
                         map, task, fsroot, cpu, is_64bit,
                         (long long)target_address, (long long)mapping_size,
                         sr_handle, kr));
                sr_offset += mapping_size;
                sr_size -= mapping_size;
        }
        assert(sr_size == 0);

done:
        SHARED_REGION_TRACE_DEBUG(
                ("shared_region: enter(%p,%p,%p,%d,%d) <- 0x%x\n",
                 map, task, fsroot, cpu, is_64bit, kr));
        return kr;
}

/******************************************************************************/
/* Comm page support                                                          */
/******************************************************************************/

ipc_port_t commpage32_handle = IPC_PORT_NULL;
ipc_port_t commpage64_handle = IPC_PORT_NULL;
vm_named_entry_t commpage32_entry = NULL;
vm_named_entry_t commpage64_entry = NULL;
vm_map_t commpage32_map = VM_MAP_NULL;
vm_map_t commpage64_map = VM_MAP_NULL;

/*
 * Create a memory entry, VM submap and pmap for one commpage.
 */
static void
_vm_commpage_init(
        ipc_port_t      *handlep,
        vm_map_size_t   size)
{
        kern_return_t           kr;
        vm_named_entry_t        mem_entry;
        vm_map_t                new_map;

        SHARED_REGION_TRACE_DEBUG(
                ("commpage: -> _init(0x%llx)\n",
                 (long long)size));

        kr = mach_memory_entry_allocate(&mem_entry,
                                        handlep);
        if (kr != KERN_SUCCESS) {
                panic("_vm_commpage_init: could not allocate mem_entry");
        }
        new_map = vm_map_create(pmap_create(0, FALSE), 0, size, TRUE);
        if (new_map == VM_MAP_NULL) {
                panic("_vm_commpage_init: could not allocate VM map");
        }
        mem_entry->backing.map = new_map;
        mem_entry->internal = TRUE;
        mem_entry->is_sub_map = TRUE;
        mem_entry->offset = 0;
        mem_entry->protection = VM_PROT_ALL;
        mem_entry->size = size;

        SHARED_REGION_TRACE_DEBUG(
                ("commpage: _init(0x%llx) <- %p\n",
                 (long long)size, *handlep));
}

/*
 * Initialize the comm pages at boot time.
 */
void
vm_commpage_init(void)
{
        SHARED_REGION_TRACE_DEBUG(
                ("commpage: -> init()\n"));

        /* create the 32-bit comm page */
        _vm_commpage_init(&commpage32_handle, _COMM_PAGE32_AREA_LENGTH);
        commpage32_entry = (vm_named_entry_t) commpage32_handle->ip_kobject;
        commpage32_map = commpage32_entry->backing.map;

        /* XXX if (cpu_is_64bit_capable()) ? */
        /* create the 64-bit comm page */
        _vm_commpage_init(&commpage64_handle, _COMM_PAGE64_AREA_LENGTH);
        commpage64_entry = (vm_named_entry_t) commpage64_handle->ip_kobject;
        commpage64_map = commpage64_entry->backing.map;

        /* populate them according to this specific platform */
        commpage_populate();

        SHARED_REGION_TRACE_DEBUG(
                ("commpage: init() <-\n"));
}

/*
 * Enter the appropriate comm page into the task's address space.
 * This is called at exec() time via vm_map_exec().
 */
kern_return_t
vm_commpage_enter(
        vm_map_t        map,
        task_t          task)
{
        ipc_port_t              commpage_handle;
        vm_map_offset_t         commpage_address, objc_address;
        vm_map_size_t           commpage_size, objc_size;
        int                     vm_flags;
        kern_return_t           kr;

        SHARED_REGION_TRACE_DEBUG(
                ("commpage: -> enter(%p,%p)\n",
                 map, task));

        /* the comm page is likely to be beyond the actual end of the VM map */
        vm_flags = VM_FLAGS_FIXED | VM_FLAGS_BEYOND_MAX;

        /* select the appropriate comm page for this task */
        assert(! (task_has_64BitAddr(task) ^ vm_map_is_64bit(map)));
        if (task_has_64BitAddr(task)) {
#ifdef __ppc__
                /*
                 * PPC51: ppc64 is limited to 51-bit addresses.
                 * Memory above that limit is handled specially at the
                 * pmap level, so do not interfere.
                 */
                vm_flags |= VM_FLAGS_NO_PMAP_CHECK;
#endif /* __ppc__ */
                commpage_handle = commpage64_handle;
                commpage_address = (vm_map_offset_t) _COMM_PAGE64_BASE_ADDRESS;
                commpage_size = _COMM_PAGE64_AREA_LENGTH;
                objc_size = _COMM_PAGE64_OBJC_SIZE;
                objc_address = _COMM_PAGE64_OBJC_BASE;
        } else {
                commpage_handle = commpage32_handle;
                commpage_address =
                        (vm_map_offset_t)(unsigned) _COMM_PAGE32_BASE_ADDRESS;
                commpage_size = _COMM_PAGE32_AREA_LENGTH;
                objc_size = _COMM_PAGE32_OBJC_SIZE;
                objc_address = _COMM_PAGE32_OBJC_BASE;
        }

        if ((commpage_address & (pmap_nesting_size_min - 1)) == 0 &&
            (commpage_size & (pmap_nesting_size_min - 1)) == 0) {
                /* the commpage is properly aligned or sized for pmap-nesting */
                vm_flags |= VM_MAKE_TAG(VM_MEMORY_SHARED_PMAP);
        }

        /* map the comm page in the task's address space */
        assert(commpage_handle != IPC_PORT_NULL);
        kr = vm_map_enter_mem_object(
                map,
                &commpage_address,
                commpage_size,
                0,
                vm_flags,
                commpage_handle,
                0,
                FALSE,
                VM_PROT_READ|VM_PROT_EXECUTE,
                VM_PROT_READ|VM_PROT_EXECUTE,
                VM_INHERIT_SHARE);
        if (kr != KERN_SUCCESS) {
                SHARED_REGION_TRACE_ERROR(
                        ("commpage: enter(%p,0x%llx,0x%llx) "
                         "commpage %p mapping failed 0x%x\n",
                         map, (long long)commpage_address,
                         (long long)commpage_size, commpage_handle, kr));
        }

        /*
         * Since we're here, we also pre-allocate some virtual space for the
         * Objective-C run-time, if needed...
         */
        if (objc_size != 0) {
                kr = vm_map_enter_mem_object(
                        map,
                        &objc_address,
                        objc_size,
                        0,
                        VM_FLAGS_FIXED | VM_FLAGS_BEYOND_MAX,
                        IPC_PORT_NULL,
                        0,
                        FALSE,
                        VM_PROT_ALL,
                        VM_PROT_ALL,
                        VM_INHERIT_DEFAULT);
                if (kr != KERN_SUCCESS) {
                        SHARED_REGION_TRACE_ERROR(
                                ("commpage: enter(%p,0x%llx,0x%llx) "
                                 "objc mapping failed 0x%x\n",
                                 map, (long long)objc_address,
                                 (long long)objc_size, kr));
                }
        }

        SHARED_REGION_TRACE_DEBUG(
                ("commpage: enter(%p,%p) <- 0x%x\n",
                 map, task, kr));
        return kr;
}