/*
 * vmm.c - virtual memory manager related operations interface.
 *
 * Authors   Ghassan Almaless (2008,2009,2010,2011, 2012)
 *           Mohamed Lamine Karaoui (2015)
 *           Alain Greiner (2016)
 *
 * Copyright (c) UPMC Sorbonne Universites
 *
 * This file is part of ALMOS-MKH.
 *
 * ALMOS-MKH is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2.0 of the License.
 *
 * ALMOS-MKH is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <kernel_config.h>
#include <hal_types.h>
#include <hal_special.h>
#include <hal_gpt.h>
#include <printk.h>
#include <memcpy.h>
#include <rwlock.h>
#include <list.h>
#include <bits.h>
#include <process.h>
#include <thread.h>
#include <vseg.h>
#include <cluster.h>
#include <scheduler.h>
#include <vfs.h>
#include <mapper.h>
#include <page.h>
#include <kmem.h>
#include <vmm.h>

//////////////////////////////////////////////////////////////////////////////////
//   Extern global variables
//////////////////////////////////////////////////////////////////////////////////

extern  process_t  process_zero;   // defined in cluster.c file


////////////////////////////////////
void vmm_init( process_t * process )
{
    error_t   error;
    vseg_t  * vseg_kentry;
    vseg_t  * vseg_args;
    vseg_t  * vseg_envs;
    vseg_t  * vseg_heap;
    intptr_t  base;
    intptr_t  size;

    vmm_dmsg("\n[INFO] %s : enter for process %x\n", __FUNCTION__ , process->pid );

    // get pointer on VMM
    vmm_t   * vmm = &process->vmm;

    assert( ((CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + CONFIG_VMM_ENVS_SIZE) 
            <= CONFIG_VMM_ELF_BASE) , __FUNCTION__ , "UTILS zone too small\n" );

    assert( (CONFIG_THREAD_MAX_PER_CLUSTER <= 32) , __FUNCTION__ ,
            "no more than 32 threads per cluster for a single process\n");

    assert( ((CONFIG_VMM_STACK_SIZE * CONFIG_THREAD_MAX_PER_CLUSTER) <=
             (CONFIG_VMM_VSPACE_SIZE - CONFIG_VMM_STACK_BASE)) , __FUNCTION__ ,
             "STACK zone too small\n");

    // initialize the rwlock protecting the vsegs list
	rwlock_init( &vmm->vsegs_lock );

    // initialize local list of vsegs and radix-tree
    vmm->vsegs_nr = 0;
	list_root_init( &vmm->vsegs_root );
    error = grdxt_init( &vmm->grdxt,
                        CONFIG_VMM_GRDXT_W1,
                        CONFIG_VMM_GRDXT_W2,
                        CONFIG_VMM_GRDXT_W3 );
 
    assert( (error == 0) , __FUNCTION__ , "cannot initialize radix tree\n" );

    // register kentry vseg in VMM
    base = 1 << CONFIG_PPM_PAGE_SHIFT;
    size = CONFIG_VMM_KENTRY_SIZE << CONFIG_PPM_PAGE_SHIFT;
    vseg_kentry = vmm_create_vseg( process , base , size , VSEG_TYPE_CODE );

    assert( (vseg_kentry != NULL) , __FUNCTION__ , "cannot register kentry vseg\n" );

    vmm->kent_vpn_base = 1;

    // register the args vseg in VMM
    base = (CONFIG_VMM_KENTRY_SIZE + 1 )<<CONFIG_PPM_PAGE_SHIFT;
    size = CONFIG_VMM_ARGS_SIZE << CONFIG_PPM_PAGE_SHIFT;
    vseg_args = vmm_create_vseg( process , base , size , VSEG_TYPE_DATA );

    assert( (vseg_args != NULL) , __FUNCTION__ , "cannot register args vseg\n" );

    vmm->args_vpn_base = CONFIG_VMM_KENTRY_SIZE + 1;

    // register the envs vseg in VMM
    base = (CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + 1 )<<CONFIG_PPM_PAGE_SHIFT;
    size = CONFIG_VMM_ENVS_SIZE << CONFIG_PPM_PAGE_SHIFT;
    vseg_envs = vmm_create_vseg( process , base , size , VSEG_TYPE_DATA );

    assert( (vseg_envs != NULL) , __FUNCTION__ , "cannot register envs vseg\n" );

    vmm->envs_vpn_base = CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + 1;

    // register the heap vseg in VMM
    base = CONFIG_VMM_HEAP_BASE << CONFIG_PPM_PAGE_SHIFT;
    size = (CONFIG_VMM_MMAP_BASE-CONFIG_VMM_HEAP_BASE) << CONFIG_PPM_PAGE_SHIFT;
    vseg_heap = vmm_create_vseg( process , base , size , VSEG_TYPE_HEAP );

    assert( (vseg_heap != NULL) , __FUNCTION__ , "cannot register heap vseg\n" );

    vmm->heap_vpn_base = CONFIG_VMM_HEAP_BASE;

    // initialize generic page table
    error = hal_gpt_create( &vmm->gpt );

    assert( (error == 0) , __FUNCTION__ , "cannot initialize page table\n");

    // initialize STACK allocator
    vmm->stack_mgr.bitmap   = 0;
    vmm->stack_mgr.vpn_base = CONFIG_VMM_STACK_BASE;

    // initialize MMAP allocator
    vmm->mmap_mgr.vpn_base        = CONFIG_VMM_MMAP_BASE;
    vmm->mmap_mgr.vpn_size        = CONFIG_VMM_STACK_BASE - CONFIG_VMM_MMAP_BASE;
    vmm->mmap_mgr.first_free_vpn  = CONFIG_VMM_MMAP_BASE;
    uint32_t i;
    for( i = 0 ; i < 32 ; i++ ) list_root_init( &vmm->mmap_mgr.zombi_list[i] );

    // initialize instrumentation counters
	vmm->pgfault_nr          = 0;
	vmm->u_err_nr            = 0;
	vmm->m_err_nr            = 0;

    hal_fence();

    vmm_dmsg("\n[INFO] %s : exit for process %x\n", __FUNCTION__ , process->pid );

}  // end vmm_init()

//////////////////////////////////////////
error_t vmm_copy( process_t * dst_process,
                  process_t * src_process )
{
    error_t error;

    vmm_t * src_vmm = &src_process->vmm;
    vmm_t * dst_vmm = &dst_process->vmm;

    // take the src_vmm vsegs_lock
    rwlock_wr_lock( &src_vmm->vsegs_lock );

    // initialize dst_vmm vsegs_lock
    rwlock_init( &dst_vmm->vsegs_lock );

    // initialize the dst_vmm vsegs list and the radix tree
    dst_vmm->vsegs_nr = 0;
    list_root_init( &dst_vmm->vsegs_root );
    error = grdxt_init( &dst_vmm->grdxt,
                        CONFIG_VMM_GRDXT_W1,
                        CONFIG_VMM_GRDXT_W2,
                        CONFIG_VMM_GRDXT_W3 );
    if( error )
    {
        printk("\n[ERROR] in %s : cannot initialize radix tree for process %x\n",
               __FUNCTION__ , dst_process->pid );
        return ENOMEM;
    }

    // loop on src_vmm list of vsegs to create
    // and register vsegs copies in dst_vmm
    list_entry_t * iter;
    vseg_t       * src_vseg;
    vseg_t       * dst_vseg;
    LIST_FOREACH( &src_vmm->vsegs_root , iter )
    {
        // get pointer on current src_vseg
        src_vseg = LIST_ELEMENT( iter , vseg_t , list );

        // allocate memory for a new dst_vseg
        dst_vseg = vseg_alloc();

        if( dst_vseg == NULL )
        {
            // release all allocated vsegs
            LIST_FOREACH( &dst_vmm->vsegs_root , iter )
            {
                dst_vseg = LIST_ELEMENT( iter , vseg_t , list );
                vseg_free( dst_vseg );
            }
            return ENOMEM;
        }

        // copy src_vseg to dst_vseg
        vseg_init_from_ref( dst_vseg , XPTR( local_cxy , src_vseg ) );

        // register dst_vseg in dst_vmm
        vseg_attach( dst_vmm , dst_vseg );
    }

    // release the src_vmm vsegs_lock
    rwlock_wr_unlock( &src_vmm->vsegs_lock );

    // initialize generic page table
    error = hal_gpt_create( &dst_vmm->gpt );

    if( error )
    {
        printk("\n[ERROR] in %s : cannot initialize page table\n", __FUNCTION__ );
        return ENOMEM;
    }

    // initialize STACK allocator
    dst_vmm->stack_mgr.bitmap   = 0;
    dst_vmm->stack_mgr.vpn_base = CONFIG_VMM_STACK_BASE;

    // initialize MMAP allocator
    dst_vmm->mmap_mgr.vpn_base        = CONFIG_VMM_MMAP_BASE;
    dst_vmm->mmap_mgr.vpn_size        = CONFIG_VMM_STACK_BASE - CONFIG_VMM_MMAP_BASE;
    dst_vmm->mmap_mgr.first_free_vpn  = CONFIG_VMM_MMAP_BASE;
    uint32_t i;
    for( i = 0 ; i < 32 ; i++ ) list_root_init( &dst_vmm->mmap_mgr.zombi_list[i] );

    // initialize instrumentation counters
	dst_vmm->pgfault_nr    = 0;
	dst_vmm->u_err_nr      = 0;
	dst_vmm->m_err_nr      = 0;

    // copy base addresses
    dst_vmm->kent_vpn_base = src_vmm->kent_vpn_base;
    dst_vmm->args_vpn_base = src_vmm->args_vpn_base;
    dst_vmm->envs_vpn_base = src_vmm->envs_vpn_base;
    dst_vmm->heap_vpn_base = src_vmm->heap_vpn_base;
    dst_vmm->code_vpn_base = src_vmm->code_vpn_base;
    dst_vmm->data_vpn_base = src_vmm->data_vpn_base;

    dst_vmm->entry_point   = src_vmm->entry_point;

    // HEAP TODO : new heap for child ???
    dst_vmm->heap_vseg     = src_vmm->heap_vseg;

    // initialize generic page table
    error = hal_gpt_create( &dst_vmm->gpt );

    if( error )
    {
        printk("\n[ERROR] in %s : cannot initialize page table\n", __FUNCTION__ );
        return ENOMEM;
    }

    // copy GPT content from src_vmm to dst_vmm, activating "Copy-On-Write"
    // TODO register Copy-On_Write in page descriptors
    bool_t cow = true;
    hal_gpt_copy( &dst_vmm->gpt , &src_vmm->gpt , cow );

    hal_fence();

    return 0;

}  // vmm_copy()

///////////////////////////////////////
void vmm_destroy( process_t * process )
{
	vseg_t * vseg;

    // get pointer on VMM
    vmm_t  * vmm = &process->vmm;

    // get lock protecting vseg list
	rwlock_wr_lock( &vmm->vsegs_lock );

    // remove all vsegs registered in vmm
	while( !list_is_empty( &vmm->vsegs_root ) )
	{
		vseg = LIST_FIRST( &vmm->vsegs_root ,  vseg_t , list );
		vseg_detach( vmm , vseg );
        vseg_free( vseg );
	}

    // delete vsegs radix_tree
    grdxt_destroy( &vmm->grdxt );

    // release lock
	rwlock_wr_unlock(&vmm->vsegs_lock);

    // remove all vsegs from zombi_lists in MMAP allocator
    uint32_t i;
    for( i = 0 ; i<32 ; i++ )
    {
	    while( !list_is_empty( &vmm->mmap_mgr.zombi_list[i] ) )
	    {
		    vseg = LIST_FIRST( &vmm->mmap_mgr.zombi_list[i] , vseg_t , list );
		    vseg_detach( vmm , vseg );
            vseg_free( vseg );
	    }
    }

    // release memory allocated to the local page table
    hal_gpt_destroy( &vmm->gpt );

}  // end vmm_destroy()

/////////////////////////////////////////////////
vseg_t * vmm_check_conflict( process_t * process,
                             vpn_t       vpn_base,
                             vpn_t       vpn_size )
{
    vmm_t        * vmm = &process->vmm;
	vseg_t       * vseg;
    list_entry_t * iter;

    // scan the list of registered vsegs
	LIST_FOREACH( &vmm->vsegs_root , iter )
	{
		vseg = LIST_ELEMENT( iter , vseg_t , list );

		if( ((vpn_base + vpn_size) > vseg->vpn_base) &&
             (vpn_base < (vseg->vpn_base + vseg->vpn_size)) ) return vseg;
	}
    return NULL;

}  // end vmm_check_conflict()

////////////////////////////////////////////////////////////////////////////////////////////
// This static function is called by the vmm_create_vseg() function, and implements
// the VMM stack_vseg specific allocator.
////////////////////////////////////////////////////////////////////////////////////////////
// @ vmm      : pointer on VMM.
// @ vpn_base : (return value) first allocated page
// @ vpn_size : (return value) number of allocated pages
////////////////////////////////////////////////////////////////////////////////////////////
static error_t vmm_stack_alloc( vmm_t * vmm,
                                vpn_t * vpn_base,
                                vpn_t * vpn_size )
{
    // get stack allocator pointer
    stack_mgr_t * mgr = &vmm->stack_mgr;

    // get lock on stack allocator
    spinlock_lock( &mgr->lock );

    // get first free slot index in bitmap
    int32_t index = bitmap_ffc( &mgr->bitmap , 4 );
    if( (index < 0) || (index > 31) )
    {
        spinlock_unlock( &mgr->lock );
        return ENOMEM;
    }

    // update bitmap
    bitmap_set( &mgr->bitmap , index );

    // release lock on stack allocator
    spinlock_unlock( &mgr->lock );

    // returns vpn_base, vpn_size (one page non allocated)
    *vpn_base = mgr->vpn_base + index * CONFIG_VMM_STACK_SIZE + 1;
    *vpn_size = CONFIG_VMM_STACK_SIZE - 1;
    return 0;

} // end vmm_stack_alloc()

////////////////////////////////////////////////////////////////////////////////////////////
// This static function is called by the vmm_create_vseg() function, and implements
// the VMM MMAP specific allocator.
////////////////////////////////////////////////////////////////////////////////////////////
// @ vmm      : [in] pointer on VMM.
// @ npages   : [in] requested number of pages.
// @ vpn_base : [out] first allocated page.
// @ vpn_size : [out] actual number of allocated pages.
////////////////////////////////////////////////////////////////////////////////////////////
static error_t vmm_mmap_alloc( vmm_t * vmm,
                               vpn_t   npages,
                               vpn_t * vpn_base,
                               vpn_t * vpn_size )
{
    uint32_t   index;
    vseg_t   * vseg;
    vpn_t      base;
    vpn_t      size;
    vpn_t      free;

    // mmap vseg size must be power of 2
    // compute actual size and index in zombi_list array
    size  = POW2_ROUNDUP( npages );
    index = bits_log2( size );

    // get mmap allocator pointer
    mmap_mgr_t * mgr = &vmm->mmap_mgr;

    // get lock on mmap allocator
    spinlock_lock( &mgr->lock );

    // get vseg from zombi_list or from mmap zone
    if( list_is_empty( &mgr->zombi_list[index] ) )     // from mmap zone
    {
        // check overflow
        free = mgr->first_free_vpn;
        if( (free + size) > mgr->vpn_size ) return ENOMEM;

        // update STACK allocator
        mgr->first_free_vpn += size;

        // compute base
        base = free;
    }
    else                                             // from zombi_list
    {
        // get pointer on zombi vseg from zombi_list
        vseg = LIST_FIRST( &mgr->zombi_list[index] , vseg_t , list );

        // remove vseg from free-list
        list_unlink( &vseg->list );

        // compute base
        base = vseg->vpn_base;
    }

    // release lock on mmap allocator
    spinlock_unlock( &mgr->lock );

    // returns vpn_base, vpn_size
    *vpn_base = base;
    *vpn_size = size;
    return 0;

}  // end vmm_mmap_alloc()

//////////////////////////////////////////////
vseg_t * vmm_create_vseg( process_t * process,
                          intptr_t    base,
	                      intptr_t    size,
	                      uint32_t    type )
{
    vseg_t     * vseg;          // created vseg pointer
    vpn_t        vpn_base;      // first page index
    vpn_t        vpn_size;      // number of pages
	error_t      error;

    // get pointer on VMM
	vmm_t * vmm = &process->vmm;

	vmm_dmsg("\n[INFO] %s : enter for process %x / base = %x / size = %x / type = %s\n",
		     __FUNCTION__ , process->pid , base , size , vseg_type_str(type) );

    // compute base, size, vpn_base, vpn_size, depending on vseg type
    // we use the VMM specific allocators for STACK and MMAP vsegs
    if( type == VSEG_TYPE_STACK )
    {
        // get vpn_base and vpn_size from STACK allocator
        error = vmm_stack_alloc( vmm , &vpn_base , &vpn_size );
        if( error )
        {
            printk("\n[ERROR] in %s : no vspace for stack vseg / process %x in cluster %x\n",
                   __FUNCTION__ , process->pid , local_cxy );
            return NULL;
        }

        // compute vseg base and size from vpn_base and vpn_size
        base = vpn_base << CONFIG_PPM_PAGE_SHIFT;
        size = vpn_size << CONFIG_PPM_PAGE_SHIFT;
    }
    else if( (type == VSEG_TYPE_ANON) ||
             (type == VSEG_TYPE_FILE) ||
             (type == VSEG_TYPE_REMOTE) )
    {
        // get vpn_base and vpn_size from MMAP allocator
        vpn_t npages = size >> CONFIG_PPM_PAGE_SHIFT;
        error = vmm_mmap_alloc( vmm , npages , &vpn_base , &vpn_size );
        if( error )
        {
            printk("\n[ERROR] in %s : no vspace for mmap vseg / process %x in cluster %x\n",
                   __FUNCTION__ , process->pid , local_cxy );
            return NULL;
        }

        // compute vseg base and size from vpn_base and vpn_size
        base = vpn_base << CONFIG_PPM_PAGE_SHIFT;
        size = vpn_size << CONFIG_PPM_PAGE_SHIFT;
    }
    else
    {
        uint32_t vpn_min = base >> CONFIG_PPM_PAGE_SHIFT;
        uint32_t vpn_max = (base + size - 1) >> CONFIG_PPM_PAGE_SHIFT;

        vpn_base = vpn_min;
	    vpn_size = vpn_max - vpn_min + 1;
    }

    // check collisions
    vseg = vmm_check_conflict( process , vpn_base , vpn_size );
    if( vseg != NULL )
    {
        printk("\n[ERROR] in %s for process %x : new vseg [vpn_base = %x / vpn_size = %x]\n"
               "  overlap existing vseg [vpn_base = %x / vpn_size = %x]\n",
               __FUNCTION__ , process->pid, vpn_base, vpn_size,
               vseg->vpn_base, vseg->vpn_size );
        return NULL;
    }

    // allocate physical memory for vseg descriptor
	vseg = vseg_alloc();
	if( vseg == NULL )
	{
	    printk("\n[ERROR] in %s for process %x : cannot allocate memory for vseg\n",
             __FUNCTION__ , process->pid );
        return NULL;
	}

    // initialize vseg descriptor
	vseg_init( vseg , base, size , vpn_base , vpn_size , type , local_cxy );

    // update "heap_vseg" in VMM
	process->vmm.heap_vseg = vseg;

    // attach vseg to vmm
	rwlock_wr_lock( &vmm->vsegs_lock );
	vseg_attach( vmm , vseg );
	rwlock_wr_unlock( &vmm->vsegs_lock );

	vmm_dmsg("\n[INFO] %s : exit for process %x / vseg [%x, %x] has been mapped\n",
		     __FUNCTION__ , process->pid , vseg->min , vseg->max );

	return vseg;
}

/////////////////////////////////////
void vmm_remove_vseg( vseg_t * vseg )
{
    // get pointers on calling process and VMM
    thread_t   * this    = CURRENT_THREAD;
    process_t  * process = this->process;
    vmm_t      * vmm     = &this->process->vmm;
    uint32_t     type    = vseg->type;

    // detach vseg from VMM
	rwlock_wr_lock( &vmm->vsegs_lock );
    vseg_detach( &process->vmm , vseg );
	rwlock_wr_unlock( &vmm->vsegs_lock );

    // release the stack slot to VMM stack allocator if STACK type
    if( type == VSEG_TYPE_STACK )
    {
        // get pointer on stack allocator
        stack_mgr_t * mgr = &vmm->stack_mgr;

        // compute slot index
        uint32_t index = ((vseg->vpn_base - mgr->vpn_base - 1) / CONFIG_VMM_STACK_SIZE);

        // update stacks_bitmap
        spinlock_lock( &mgr->lock );
        bitmap_clear( &mgr->bitmap , index );
        spinlock_unlock( &mgr->lock );
    }

    // release the vseg to VMM mmap allocator if MMAP type
    if( (type == VSEG_TYPE_ANON) || (type == VSEG_TYPE_FILE) || (type == VSEG_TYPE_REMOTE) )
    {
        // get pointer on mmap allocator
        mmap_mgr_t * mgr = &vmm->mmap_mgr;

        // compute zombi_list index
        uint32_t index = bits_log2( vseg->vpn_size );

        // update zombi_list
        spinlock_lock( &mgr->lock );
        list_add_first( &mgr->zombi_list[index] , &vseg->list );
        spinlock_unlock( &mgr->lock );
    }

    // release physical memory allocated for vseg descriptor if no MMAP type
    if( (type != VSEG_TYPE_ANON) && (type != VSEG_TYPE_FILE) && (type != VSEG_TYPE_REMOTE) )
    {
        vseg_free( vseg );
    }
}

//////////////////////////////////////////////
error_t vmm_map_kernel_vseg( vseg_t    * vseg,
                             uint32_t    attr )
{
    vpn_t       vpn;        // VPN of PTE to be set
    vpn_t       vpn_min;    // VPN of first PTE to be set
    vpn_t       vpn_max;    // VPN of last PTE to be set (excluded)
	ppn_t       ppn;        // PPN of allocated physical page
	uint32_t    order;      // ln( number of small pages for one single PTE )
	page_t    * page;
    error_t     error;

    // check vseg type : must be a kernel vseg
    uint32_t type = vseg->type;
    assert( ((type==VSEG_TYPE_KCODE) || (type==VSEG_TYPE_KDATA) || (type==VSEG_TYPE_KDEV)),
            __FUNCTION__ , "not a kernel vseg\n" );

    // get pointer on page table
    gpt_t * gpt = &process_zero.vmm.gpt;

    // define number of small pages per PTE
	if( attr & GPT_SMALL ) order = 0;   // 1 small page
	else                   order = 9;   // 512 small pages

    // loop on pages in vseg
    vpn_min = vseg->vpn_base;
    vpn_max = vpn_min + vseg->vpn_size;
	for( vpn = vpn_min ; vpn < vpn_max ; vpn++ )
	{
        // allocate a physical page from local PPM
	    kmem_req_t req;
	    req.type  = KMEM_PAGE;
	    req.size  = order;
	    req.flags = AF_KERNEL | AF_ZERO;
	    page      = (page_t *)kmem_alloc( &req );
		if( page == NULL )
        {
            printk("\n[ERROR] in %s : cannot allocate physical memory\n", __FUNCTION__ );
            return ENOMEM;
        }

        // set page table entry
        ppn = ppm_page2ppn( XPTR( local_cxy , page ) );
        error = hal_gpt_set_pte( gpt , vpn , ppn , attr );
		if( error )
        {
            printk("\n[ERROR] in %s : cannot register PPE\n", __FUNCTION__ );
            return ENOMEM;
        }
	}

	return 0;
}

/////////////////////////////////////////
void vmm_unmap_vseg( process_t * process,
                     vseg_t    * vseg )
{
    vpn_t       vpn;        // VPN of current PTE
    vpn_t       vpn_min;    // VPN of first PTE
    vpn_t       vpn_max;    // VPN of last PTE (excluded)

    // get pointer on process page table
    gpt_t     * gpt = &process->vmm.gpt;

    // loop on pages in vseg
    vpn_min = vseg->vpn_base;
    vpn_max = vpn_min + vseg->vpn_size;
	for( vpn = vpn_min ; vpn < vpn_max ; vpn++ )
    {
        hal_gpt_reset_pte( gpt , vpn );
    }
}

/////////////////////////////////////////////
error_t vmm_resize_vseg( process_t * process,
                         intptr_t    base,
                         intptr_t    size )
{
	error_t error;

    // get pointer on process VMM
    vmm_t * vmm = &process->vmm;

    intptr_t addr_min = base;
	intptr_t addr_max = base + size;
    uint32_t shift    = CONFIG_PPM_PAGE_SHIFT;

    // get pointer on vseg
	vseg_t * vseg = grdxt_lookup( &vmm->grdxt , (uint32_t)(base >> shift) );

	if( vseg == NULL)  return EINVAL;

    // get VMM lock protecting vsegs list
	rwlock_wr_lock( &vmm->vsegs_lock );

	if( (vseg->min > addr_min) || (vseg->max < addr_max) )   // region not included in vseg
    {
        error = EINVAL;
    }
	else if( (vseg->min == addr_min) && (vseg->max == addr_max) ) // vseg must be removed
    {
        vmm_remove_vseg( vseg );
        error = 0;
    }
	else if( vseg->min == addr_min )                              // vseg must be resized
    {
        panic("resize not implemented yet");
		error = 0;
    }
	else if( vseg->max == addr_max )                              // vseg must be resized
    {
        panic("resize not implemented yet");
		error = 0;
    }
    else            // vseg cut in three regions => vseg must be resized & new vseg created
    {
        panic("resize not implemented yet");
		error = 0;
    }

    // release VMM lock
	rwlock_wr_unlock( &vmm->vsegs_lock );

	return error;
}

///////////////////////////////////////////
error_t  vmm_get_vseg( process_t * process,
                       intptr_t    vaddr,
                       vseg_t   ** found_vseg )
{
    vmm_t    * vmm;
    vseg_t   * vseg;

    // get pointer on process VMM
    vmm = &process->vmm;

    // get lock protecting the vseg list
    rwlock_rd_lock( &vmm->vsegs_lock );

    // get pointer on vseg from local radix tree
	vseg = grdxt_lookup( &vmm->grdxt, (uint32_t)(vaddr >> CONFIG_PPM_PAGE_SHIFT) );

    // release the lock
    rwlock_rd_unlock( &vmm->vsegs_lock );

    if( vseg == NULL )   // vseg not found in local cluster => try to get it from ref
	{
        // get extended pointer on reference process
        xptr_t ref_xp = process->ref_xp;

        // get cluster and local pointer on reference process 
        cxy_t       ref_cxy = GET_CXY( ref_xp );
        process_t * ref_ptr = (process_t *)GET_PTR( ref_xp );

        if( local_cxy == ref_cxy )  return -1;   // local cluster is the reference

        // get extended pointer on reference vseg
        xptr_t   vseg_xp;
        error_t  error;
        rpc_vmm_get_vseg_client( ref_cxy , ref_ptr , vaddr , &vseg_xp , &error );
            
        if( error )   return -1;       // vseg not found => illegal user vaddr 
        
        // allocate a vseg in local cluster
        vseg = vseg_alloc();

        if( vseg == NULL ) panic("no memory for vseg copy in cluster %x", local_cxy );

        // initialise local vseg from reference
        vseg_init_from_ref( vseg , vseg_xp );

        // register local vseg in local VMM
        error = vseg_attach( &process->vmm , vseg );

        if( error ) panic("no memory for vseg registration in cluster %x", local_cxy );
    }   
    
    // success
    *found_vseg = vseg;
    return 0;

}  // end vmm_get_vseg()

////////////////////////////////////////
error_t vmm_get_one_ppn( vseg_t * vseg,
                         vpn_t    vpn,
                         ppn_t  * ppn )
{
    error_t    error;
    cxy_t      page_cxy;          // physical page cluster
    page_t   * page_ptr;          // local pointer on physical page descriptor

    uint32_t   type      = vseg->type;
    xptr_t     mapper_xp = vseg->mapper_xp;
    uint32_t   flags     = vseg->flags;

    // get mapper cluster and local pointer
    cxy_t      mapper_cxy = GET_CXY( mapper_xp );
    mapper_t * mapper_ptr = (mapper_t *)GET_PTR( mapper_xp );

    // FILE type : simply get the physical page from the file mapper
    if( type == VSEG_TYPE_FILE )
    {
        // compute index in file mapper
        uint32_t index = vpn - vseg->vpn_base;

        // get page descriptor from mapper
        if( mapper_cxy == local_cxy )             // mapper is local
        {
            page_ptr = mapper_get_page( mapper_ptr , index );
        }
        else                                      // mapper is remote
        {
            rpc_mapper_get_page_client( mapper_cxy , mapper_ptr , index , &page_ptr );
        }

        if ( page_ptr == NULL ) return EINVAL;

        page_cxy = mapper_cxy;
    }

    // all other types : allocate a physical page from target cluster,
    else
    {
        // get target cluster for physical page
        if( flags & VSEG_DISTRIB ) // depends on VPN LSB
        {
            uint32_t x_width = LOCAL_CLUSTER->x_width;
            uint32_t y_width = LOCAL_CLUSTER->y_width;
            page_cxy = vpn & ((1<<(x_width + y_width)) - 1);
        }
        else                       // defined in vseg descriptor
        {
            page_cxy = vseg->cxy;
        }

        // allocate a physical page in target cluster
        kmem_req_t   req;
        if( page_cxy == local_cxy )    // target cluster is the local cluster
        {
            req.type  = KMEM_PAGE;
            req.size  = 0;
            req.flags = AF_NONE;
            page_ptr  = (page_t *)kmem_alloc( &req );
        }
        else                           // target cluster is not the local cluster
        {
            rpc_pmem_get_pages_client( page_cxy , 0 , &page_ptr );
        }

        if( page_ptr == NULL ) return ENOMEM;

        // initialise page from .elf file mapper for DATA and CODE types
        if( (type == VSEG_TYPE_CODE) || (type == VSEG_TYPE_DATA) )
        {
            // compute missing page index in vseg
            vpn_t page_index = vpn - vseg->vpn_base;

            // compute missing page offset in .elf file
            intptr_t page_offset = vseg->file_offset + 
                                   (page_index << CONFIG_PPM_PAGE_SHIFT);

            // compute extended pointer on page first byte 
            xptr_t base_xp = ppm_page2base( XPTR( page_cxy , page_ptr ) );

            // file_size can be smaller than vseg_size for BSS
            intptr_t file_size = vseg->file_size;

            if( file_size < page_offset )                                // fully in  BSS 
            {
                if( page_cxy == local_cxy )
                {
                    memset( GET_PTR( base_xp ) , 0 , CONFIG_PPM_PAGE_SIZE );
                }
                else
                {
                   hal_remote_memset( base_xp , 0 , CONFIG_PPM_PAGE_SIZE );       
                }
            }
            else if( file_size >= (page_offset + CONFIG_PPM_PAGE_SIZE) )  // fully in  mapper
            {
                if( mapper_cxy == local_cxy ) 
                {
                    error = mapper_move_kernel( mapper_ptr,
                                                true,             // to_buffer
                                                page_offset,
                                                base_xp,
                                                CONFIG_PPM_PAGE_SIZE ); 
                }
                else 
                {
                    rpc_mapper_move_buffer_client( mapper_cxy,
                                                   mapper_ptr,
                                                   true,         // to buffer
                                                   false,        // kernel buffer
                                                   page_offset,
                                                   (uint64_t)base_xp,
                                                   CONFIG_PPM_PAGE_SIZE,
                                                   &error );
                }
                if( error ) return EINVAL;
            }
            else  // in mapper : from page_offset -> (file_size - page_offset)
                  // in BSS    : from file_size   -> (page_offset + page_size)
            {
                // initialize mapper part
                if( mapper_cxy == local_cxy )
                {
                    error = mapper_move_kernel( mapper_ptr,
                                                true,         // to buffer
                                                page_offset,
                                                base_xp,
                                                file_size - page_offset ); 
                }
                else                                
                {
                    rpc_mapper_move_buffer_client( mapper_cxy,
                                                   mapper_ptr,
                                                   true,         // to buffer
                                                   false,        // kernel buffer
                                                   page_offset,
                                                   (uint64_t)base_xp,
                                                   file_size - page_offset, 
                                                   &error );
                }
                if( error ) return EINVAL;

                // initialize BSS part
                if( page_cxy == local_cxy )
                {
                    memset( GET_PTR( base_xp ) + file_size - page_offset , 0 , 
                            page_offset + CONFIG_PPM_PAGE_SIZE - file_size );
                }
                else
                {
                   hal_remote_memset( base_xp + file_size - page_offset , 0 , 
                                      page_offset + CONFIG_PPM_PAGE_SIZE - file_size );
                }
            }    
        }  // end initialisation for CODE or DATA types   
    } 

    // return ppn
    *ppn = ppm_page2ppn( XPTR( page_cxy , page_ptr ) );
    return 0;

}  // end vmm_get_one_ppn()

/////////////////////////////////////////
error_t vmm_get_pte( process_t * process,
                     vpn_t       vpn,
                     uint32_t  * ret_attr,
                     ppn_t     * ret_ppn )
{
    vseg_t  * vseg;   // pointer on vseg containing VPN
    ppn_t     ppn;    // physical page number
    uint32_t  attr;   // attributes from GPT entry
    error_t   error;

    // this function must be called by a thread running in the reference cluster
    assert( (GET_CXY( process->ref_xp ) == local_cxy ) , __FUNCTION__ ,
             " not called in the reference cluster\n" );

    // get VMM pointer
    vmm_t * vmm = &process->vmm;

    // access GPT to get PTE attributes and PPN
    hal_gpt_get_pte( &vmm->gpt , vpn , &attr , &ppn );

    // if PTE is unmapped
    // 1) get VSEG containing the missing VPN
    // 2) get & initialize physical page (depending on vseg type),
    // 3) register the PTE in reference GPT 
    if( (attr & GPT_MAPPED) == 0 )
    {
        // 1. get vseg pointer
        error = vmm_get_vseg( process , vpn<<CONFIG_PPM_PAGE_SHIFT , &vseg );

        if( error )
        {
            printk("\n[ERROR] in %s : out of segment / process = %x / vpn = %x\n",
                   __FUNCTION__ , process->pid , vpn );
            return EINVAL;
        }

        // 2. get physical page number, depending on vseg type
        error = vmm_get_one_ppn( vseg , vpn , &ppn );

        if( error )
        {
            printk("\n[ERROR] in %s : cannot allocate memory / process = %x / vpn = %x\n",
                   __FUNCTION__ , process->pid , vpn );
            return error;
        }

        // 3. define attributes from vseg flags and register in GPT
        attr = GPT_MAPPED | GPT_SMALL;
        if( vseg->flags & VSEG_USER  ) attr |= GPT_USER;
        if( vseg->flags & VSEG_WRITE ) attr |= GPT_WRITABLE;
        if( vseg->flags & VSEG_EXEC  ) attr |= GPT_EXECUTABLE;
        if( vseg->flags & VSEG_CACHE ) attr |= GPT_CACHABLE;

        error = hal_gpt_set_pte( &vmm->gpt , vpn , ppn , attr );

        if( error )
        {
            printk("\n[ERROR] in %s : cannot register PTE / process = %x / vpn = %x\n",
                   __FUNCTION__ , process->pid , vpn );
            return ENOMEM;
        }
    }  // end new PTE

    *ret_ppn  = ppn;
    *ret_attr = attr;
    return 0;

}  // end vmm_get_pte()

///////////////////////////////////////////////////
error_t vmm_handle_page_fault( process_t * process,
                               vseg_t    * vseg,
                               vpn_t       vpn )
{
    uint32_t         attr;          // missing page attributes
    ppn_t            ppn;           // missing page PPN
    error_t          error;         // return value

    // get local VMM pointer
	vmm_t * vmm = &process->vmm;

    // get reference process cluster and local pointer
    cxy_t       ref_cxy = GET_CXY( process->ref_xp );
    process_t * ref_ptr = (process_t *)GET_PTR( process->ref_xp );

    // get missing PTE attributes and PPN from reference cluster
    if( local_cxy != ref_cxy )   // local cluster is not the reference cluster
    {
        rpc_vmm_get_pte_client( ref_cxy , ref_ptr , vpn , &attr , &ppn , &error );
    }
    else                              // local cluster is the reference cluster
    {
        error = vmm_get_pte( process , vpn , &attr , &ppn );
    }

    return error;

}  // end vmm_handle_page_fault()


///////////////////////////////////////////
error_t vmm_v2p_translate( bool_t    ident,
                           void    * ptr,
                           paddr_t * paddr )
{
    process_t * process = CURRENT_THREAD->process;

    if( ident )  // identity mapping
    {
        *paddr = (paddr_t)PADDR( local_cxy , (lpa_t)ptr );
        return 0;
    }

    // access page table
    error_t  error;
    vpn_t    vpn;
    uint32_t attr;
    ppn_t    ppn;
    uint32_t offset;

    vpn    = (vpn_t)( (intptr_t)ptr >> CONFIG_PPM_PAGE_SHIFT );
    offset = (uint32_t)( ((intptr_t)ptr) & CONFIG_PPM_PAGE_MASK );

    if( local_cxy == GET_CXY( process->ref_xp) ) // calling process is reference process
    {
        error = vmm_get_pte( process, vpn , &attr , &ppn );
    }
    else                                         // calling process is not reference process
    {
        cxy_t       ref_cxy = GET_CXY( process->ref_xp );
        process_t * ref_ptr = (process_t *)GET_PTR( process->ref_xp );
        rpc_vmm_get_pte_client( ref_cxy , ref_ptr , vpn , &attr , &ppn , &error );
    }

    // set paddr
    *paddr = (((paddr_t)ppn) << CONFIG_PPM_PAGE_SHIFT) | offset;

    return error;

}  // end vmm_v2p_translate()