source: trunk/kernel/mm/vmm.c @ 8

/*
 * 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 <almos_config.h>
#include <hal_types.h>
#include <hal_special.h>
#include <hal_gpt.h>
#include <printk.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;

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

    // check UTILS zone size
    if( (CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + CONFIG_VMM_ENVS_SIZE ) >
        CONFIG_VMM_ELF_BASE )
    {
        printk("\n[PANIC] in %s : UTILS zone too small for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }
    
    // check max number of stacks slots
    if( CONFIG_THREAD_MAX_PER_CLUSTER > 32 )
    {
        printk("\n[PANIC] in %s : max number ot threads per cluster for a sihgle process"
               " cannot be larger than 32\n", __FUNCTION__ );
        hal_core_sleep();
    }

    // check STACK zone size
    if( (CONFIG_VMM_STACK_SIZE * CONFIG_THREAD_MAX_PER_CLUSTER) > 
        (CONFIG_VMM_VSPACE_SIZE - CONFIG_VMM_STACK_BASE) )
    {
        printk("\n[PANIC] in %s : STACK zone too small for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }

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

    // initialize local list of vsegs and radix-tree
        list_root_init( &vmm->vsegs_root );
    vmm->vsegs_nr = 0;
    error = grdxt_init( &vmm->grdxt,
                        CONFIG_VMM_GRDXT_W1,
                        CONFIG_VMM_GRDXT_W2,
                        CONFIG_VMM_GRDXT_W3 );
    if( error )
    {
        printk("\n[PANIC] in %s : cannot initialise radix tree for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }

    // 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 );
    if( vseg_kentry == NULL ) 
    {
        printk("\n[PANIC] in %s : cannot register kent vseg for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }
    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 );
    if( vseg_args == NULL ) 
    {
        printk("\n[PANIC] in %s : cannot register args vseg for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }
    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 );
    if( vseg_envs == NULL ) 
    {
        printk("\n[PANIC] in %s : cannot register envs vseg for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }
    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 );
    if( vseg_heap == NULL ) 
    {
        printk("\n[PANIC] in %s : cannot register heap vseg in for process %x\n",
               __FUNCTION__ , process->pid );
        hal_core_sleep();
    }
    vmm->heap_vpn_base = CONFIG_VMM_HEAP_BASE;

    // initialize generic page table
    error = hal_gpt_create( &vmm->gpt );
    if( error )
    {
        printk("PANIC in %s : cannot initialize page table\n", __FUNCTION__ );
        hal_core_sleep();
    }

    // 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] );

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

    hal_wbflush();

}  // end vmm_init()

///////////////////////////////////////
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;
}  // 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 alocated 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) ) 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_allocator()

//////////////////////////////////////////////
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;      // vseg first page 
    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_name[type] );
   
    // compute base, size, vpn_base, vpn_size, depending on 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
    {
        vpn_base = ARROUND_DOWN( base , CONFIG_PPM_PAGE_SIZE ) >> CONFIG_PPM_PAGE_SHIFT;
            vpn_size = ARROUND_UP( base + size , CONFIG_PPM_PAGE_SIZE ) >> CONFIG_PPM_PAGE_SHIFT;
    }

    // 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 , 0 , 0 );

    // 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;

}  // end vmm_create_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 );
    }

}  // end vmm_remove_vseg()


//////////////////////////////////////////
error_t vmm_map_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
    uint32_t type = vseg->type;
    if( (type != VSEG_TYPE_KCODE) && (type != VSEG_TYPE_KDATA) && (type != VSEG_TYPE_KDEV) )
    {
        printk("\n[PANIC] in %s : not a kernel vseg\n", __FUNCTION__ );
        hal_core_sleep();
    }

    // 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++ )
        {
            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( 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;
} // end vmm_map_vseg()

/////////////////////////////////////////
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 );
    }
} // end vmm_unmap_vseg()


/////////////////////////////////////////////
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
    {
        printk("\n[PANIC] in %s : resize not implemented yet\n", __FUNCTION__ );
        hal_core_sleep();
                error = 0;
    }
        else if( vseg->max == addr_max )                              // vseg must be resized
    {
        printk("\n[PANIC] in %s : resize not implemented yet\n", __FUNCTION__ );
        hal_core_sleep();
                error = 0;
    }
    else            // vseg cut in three regions => vseg must be resized & new vseg created
    {
        printk("\n[PANIC] in %s : resize not implemented yet\n", __FUNCTION__ );
        hal_core_sleep();
                error = 0;
    }

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

        return error;
} // end vmm_resize_vseg()

///////////////////////////////////////////
vseg_t * vmm_get_vseg( process_t * process,
                       intptr_t    vaddr )
{

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

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

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

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

    return vseg;

} // end vmm_get_vseg()

/////////////////////////////////////////
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;    // PPN from GPT entry
    uint32_t  attr;   // attributes from GPT entry
    error_t   error;

    // this function must be called by in the reference cluster 
    if( process->is_ref == false );
    {
        printk("\n[PANIC] in %s : not called in the reference cluster\n", __FUNCTION__ );
        hal_core_sleep();
    }

    // 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 unmapped => allocate one small physical page to map it
    if( (attr & GPT_MAPPED) == 0 )  
    {
        // get vseg pointer
        vseg = vmm_get_vseg( process , vpn<<CONFIG_PPM_PAGE_SHIFT );

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

        // select the target cluster for physical mapping
        uint32_t target_cxy;
        if( vseg->flags & VSEG_DISTRIB ) // depends on VPN LSB
        {
            uint32_t x_width = LOCAL_CLUSTER->x_width;
            uint32_t y_width = LOCAL_CLUSTER->y_width;
            target_cxy = vpn & ((1<<(x_width + y_width)) - 1);
        }
        else                             // defined in vseg descriptor
        {
            target_cxy = vseg->cxy;
        }

        // allocate memory for page fault
        kmem_req_t   req;
        page_t     * page;
        if( target_cxy == local_cxy )        // target cluster is the local cluster
        { 
            req.type  = KMEM_PAGE;
            req.size  = 0;
            req.flags = AF_NONE;
            page      = (page_t *)kmem_alloc( &req );

            error = ( page == NULL ) ? 1 : 0;
            ppn   = ppm_page2ppn( page );
        }
        else                                 // target cluster is not the local cluster
        {
            rpc_pmem_get_pages_client( target_cxy , 0 , &error , &ppn );
        }

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

        // defineg GPT attributes from vseg flags
        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;

        // set the missing PTE in local VMM
        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;
        }
    }

    *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
    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 );
    }

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

    // set the missing PTE in local VMM
    error = hal_gpt_set_pte( &vmm->gpt , vpn , attr , ppn );
    if( error )
    {
        printk("\n[ERROR] in %s : cannot register PTE / process = %x / vpn = %x\n",
               __FUNCTION__ , process->pid , vpn ); 
        return ENOMEM;
    }
 
    return 0;
}  // end vmm_handle_page_fault()

///////////////////////////////////////////
error_t vmm_v2p_translate( bool_t    ident,
                           void    * ptr,
                           paddr_t * paddr )
{
    uint32_t vaddr = (uint32_t)ptr;

    thread_t  * this    = CURRENT_THREAD;
    process_t * process = this->process;

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

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

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

    if( process->is_ref )   // calling process is reference process
    {
        error = vmm_get_pte( process, vpn , &attr , &ppn );
    }
    else                    // use a RPC
    {
        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 );
    }

    if( error )
    {
        printk("\n[ERROR] in %s : cannot get physical address for vaddr = %x\n",
               __FUNCTION__ , vaddr );
        return error;
    }
  
    // return paddr
    *paddr = (((paddr_t)ppn) << CONFIG_PPM_PAGE_SHIFT) | offset;
    return 0;
 
}  // end vmm_v2p_translate()




/*

///////////////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////////////
error_t vmm_inval_shared_page( vseg_t *vseg, vma_t vaddr, ppn_t ppn)
{
        pmm_page_info_t current;
        error_t err;

        error= pmm_get_page(&vseg->vmm->pmm, vaddr, &current);

        if((err) || (current.ppn != ppn))
                goto ended;

        current.ppn     = 0;
        current.attr    = 0;
        current.cluster = NULL; 

        error= pmm_set_page(&vseg->vmm->pmm, vaddr, &current);

ended:
        return err;
}

error_t vmm_update_shared_page( vseg_t *vseg, vma_t vaddr, ppn_t ppn)
{
        pmm_page_info_t current;
        error_t err;

        error= pmm_get_page(&vseg->vmm->pmm, vaddr, &current);

        if((err) || (current.attr != 0))
                goto ended;

        current.ppn     = ppn;
        current.attr    = vseg->vm_pgprot;
        current.cluster = NULL; // this function is called after invalidate one 

        error= pmm_set_page(&vseg->vmm->pmm, vaddr , &current);

ended:
        return err;
}

// Hypothesis: the vseg is shared-anon, mapper list is rdlocked, page is locked 
error_t vmm_migrate_shared_page_seq( vseg_t *vseg, struct page_s *page, struct page_s **new)
{
        register  vseg_t *reg;
        register struct process_s *process;
        register struct process_s *this_process;
        struct page_s *new_pg;
        struct list_entry *iter;
        kmem_req_t req;
        vma_t vaddr;
        ppn_t ppn;
        error_t err;

        vaddr     = (page->index << PMM_PAGE_SHIFT) + vseg->vm_start + vseg->vm_offset;
        ppn       = ppm_page2ppn(page);
        this_process = (new == NULL) ? NULL : current_process;
        iter      = &vseg->vm_shared_list;
        error      = ECANCELED;

        // Invalidate All 
        do
        {
                reg  = list_element(iter,  vseg_t, vm_shared_list);

                process = vmm_get_process(reg->vmm);

                if(process != this_process)
                {
                        error= vmm_inval_shared_page(reg, vaddr, ppn);

                        if(err) goto fail_inval;
                }

                assert(vseg->vm_mapper.m_home_cid == current_cid);
                iter = list_next(&vseg->vm_mapper.m_reg_root, iter);

        }while(iter != NULL);

        req.type  = KMEM_PAGE;
        req.size  = 0;
        req.excep_code = AF_USER;

        new_pg    = kmem_alloc(&req);
        *new      = new_pg;

        if(new_pg == NULL)
        {
                error= ENOMEM;
                goto fail_alloc;
        }

        page_copy(new_pg, page);

        page_lock(new_pg);

        new_pg->mapper = page->mapper;
        new_pg->index  = page->index;

        // TODO: do the complet job regading dirty page 
        if(PAGE_IS(page, PG_DIRTY))
                PAGE_SET(new_pg, PG_DIRTY);

        ppn  = ppm_page2ppn(new_pg);
        iter = &vseg->vm_shared_list;

        // Update All
        do
        {
                reg  = list_element(iter,  vseg_t, vm_shared_list);

                process = vmm_get_process(reg->vmm);

                if(process != this_process)
                        (void) vmm_update_shared_page(reg, vaddr, ppn);

                assert(vseg->vm_mapper.m_home_cid == current_cid);
                iter = list_next(&vseg->vm_mapper.m_reg_root, iter);


        }while(iter != NULL);

        page_unlock(new_pg);

fail_alloc:
fail_inval:
        return err;
}

//TODO: revisit all manipulation of the page->refcount
///////////////////////////////////////////////////////////////
static inline error_t vmm_do_migrate( vseg_t     * vseg, 
                                      pmm_page_info_t * pinfo,
                                      uint32_t          vaddr )
{
        kmem_req_t        req;
        pmm_page_info_t   current;
        page_t          * newpage;
        cluster_t       * cluster;
        thread_t        * this;
        error_t           err;
        ppn_t             ppn;
 
        assert( pinfo->ppn != 0 );

        ppn = pinfo->ppn;
        this = current_thread;
        newpage = NULL;
        cluster = current_cluster;
  
        current.attr = 0;
        current.ppn  = 0;

        error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &current);

        if(error|| (current.isAtomic == false) || 
              (current.ppn != ppn) || !(current.attr & PMM_MIGRATE))
        {
#if CONFIG_SHOW_SPURIOUS_PGFAULT
                printk(INFO, "%s: pid %d, tid %d, cpu %d, nothing to do for vaddr %x\n", 
                       __FUNCTION__, 
                       this->process->pid, 
                       this->info.order, 
                       cpu_get_id(),
                       vaddr);
#endif
                this->info.spurious_pgfault_cntr ++;
                pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
                pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
                return 0;
        }
  
        if(!ppn_is_local(ppn))
        {
                req.type  = KMEM_PAGE;
                req.size  = 0;
                req.excep_code = AF_PGFAULT;

                newpage = kmem_alloc(&req);

                if(newpage)
                {
                        newpage->mapper = NULL;//?
                        ppn_copy(ppm_page2ppn(newpage), ppn);

                        if(current.attr & PMM_COW)
                        {
                                current.attr |= PMM_WRITE;
                                current.attr &= ~(PMM_COW);
                        }

                        current.ppn = ppm_page2ppn(newpage);
                }
        }

        current.attr   |= PMM_PRESENT;
        current.attr   &= ~(PMM_MIGRATE);
        current.attr   &= ~(PMM_LOCKED);
        current.cluster = NULL;

        //also unlock the table entry
        error= pmm_set_page(&vseg->vmm->pmm, vaddr, &current);
        
        if(err)
        {
                // TODO: we should differ the kmem_free call 
                //page_unlock(page);
                (void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
                req.ptr = newpage;
                kmem_free(&req);
                return err;
        } 


        if(newpage)
        {
                ppn_refcount_down(ppn);
                current_thread->info.remote_pages_cntr ++;
#if CONFIG_SHOW_REMOTE_PGALLOC  
                printk(INFO, "%s: pid %d, tid %x, cpu %d, cid %d: got new remote page from cluster %d (vaddr %x)\n",
                       __FUNCTION__,
                       current_process->pid,
                       current_thread,
                       cpu_get_id(),
                       cluster->id,
                       newpage->cid,
                       vaddr);
#endif
        }

#if CONFIG_SHOW_VMMMGRT_MSG
        printk(INFO, "%s: pid %d, tid %d, cpu %d: Asked to migrate page (vaddr %x) from cluster %d to cluster %d, error%d\n",
               __FUNCTION__,
               current_process->pid,
               current_thread->info.order,
               cpu_get_id(),
               vaddr,
               ppn_ppn2cid(ppn),
               cluster->id,
               err);
#endif

        return err;
}

error_t vmm_do_cow( vseg_t *vseg, pmm_page_info_t *pinfo, uint32_t vaddr)
{
        register struct page_s *newpage;
        register struct page_s *page;
        register struct thread_s *this;
        register error_t err;
        register uint32_t count;
        register bool_t isCountDown;
        pmm_page_info_t old;
        pmm_page_info_t new;
        kmem_req_t req;

        this       = current_thread;
        old.attr  = 0;
        newpage    = NULL;
        isCountDown = true;

        vmm_dmsg(2,"%s: pid %d, tid %d, cpu %d, vaddr %x\n",
                 __FUNCTION__,
                 this->process->pid,
                 this->info.order,
                 cpu_get_id(),
                 vaddr);

  
        error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &old);

        //TODO: check this condition
        if(error|| (old.isAtomic == false) || !(old.attr & PMM_COW))
        {
#if CONFIG_SHOW_SPURIOUS_PGFAULT
                printk(INFO, "%s: pid %d, tid %d, cpu %d, nothing to do for vaddr %x\n", 
                       __FUNCTION__, 
                       this->process->pid, 
                       this->info.order, 
                       cpu_get_id(),
                       vaddr);
#endif
                this->info.spurious_pgfault_cntr ++;
                pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
                pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
                return err;
                //goto VMM_COW_END;
        }

        //if the ppn is local and the others (processus with wich we share the page) 
        //has done cow, then use the old.ppn directly
        if(ppn_is_local(old.ppn))
        {
                page = ppm_ppn2page(&current_cluster->ppm, old.ppn);

                if(page->mapper == NULL)
                {
                        count = page_refcount_get(page);
                        if(count == 1)
                        {
                                newpage = page;//don't copy the page. use it directly.
                                isCountDown = false;
                                vmm_dmsg(2, "%s: pid %d, tid %d, cpu %d, reuse same page for vaddr %x, pg_addr %x\n", 
                                         __FUNCTION__, 
                                         this->process->pid, 
                                         this->info.order, 
                                         cpu_get_id(), 
                                         vaddr, 
                                         ppm_page2addr(page));
                        }
                }
                //else: we need to do the cow even if it's local!

        }

        //else: alocate newpage and copy the data from the remote node
        //also defcount down the ppn
        if(newpage == NULL)
        {    
                req.type  = KMEM_PAGE;
                req.size  = 0;
                req.excep_code = AF_PGFAULT;

                if((newpage = kmem_alloc(&req)) == NULL)
                {
                        (void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
                        return ENOMEM;
                }       

                newpage->mapper = NULL;

                ppn_copy(ppm_page2ppn(newpage), old.ppn);
                assert(isCountDown);
                
                vmm_dmsg(2, 
                         "%s: pid %d, tid %d, cpu %d, newpage for vaddr %x, pg_addr %x\n", 
                         __FUNCTION__, 
                         this->process->pid, 
                         this->info.order, 
                         cpu_get_id(), 
                         vaddr, 
                         ppm_page2addr(newpage));

                if(newpage->cid != current_cid)
                        this->info.remote_pages_cntr ++;
        }

        new.attr    = vseg->vm_pgprot | PMM_WRITE;
        new.attr   &= ~(PMM_COW | PMM_MIGRATE);
        new.ppn     = ppm_page2ppn(newpage);
        new.cluster = NULL;

        //this also unlock the table entry (if no error)
        error= pmm_set_page(&vseg->vmm->pmm, vaddr, &new);

        if(err)
        {
                (void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
                req.ptr = newpage;
                kmem_free(&req);
                vmm_dmsg(3, "%s: ended [ error%d ]\n", __FUNCTION__, err);
                return err;
        }
        
        if(isCountDown) ppn_refcount_down(old.ppn);
        
        vmm_dmsg(2, "%s, pid %d, tid %d, cpu %d, COW ended [vaddr %x]\n", 
                 __FUNCTION__, 
                 this->process->pid,
                 this->info.order,
                 cpu_get_id(),
                 vaddr);
  
        return 0;
}


//refcount is taken on the file at mmap
static inline error_t vmm_do_mapped( vseg_t *vseg, uint32_t vaddr, uint32_t excep_code)
{
        ppn_t ppn;
        error_t err;
        uint32_t index;
        bool_t isDone;
        pmm_page_info_t info;
        pmm_page_info_t current;
        struct thread_s *this;

        this = current_thread;

        current.attr = 1;
        current.ppn  = 1;
        isDone       = false;

        error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &current);
        
        if(err) return err;

        if((current.isAtomic == false) || (current.attr != 0))
        {
#if CONFIG_SHOW_SPURIOUS_PGFAULT
                printk(INFO, "%s: pid %d, tid %d, cpu %d, nothing to do for vaddr %x\n", 
                       __FUNCTION__, 
                       this->process->pid, 
                       this->info.order, 
                       cpu_get_id(),
                       vaddr);
#endif
                this->info.spurious_pgfault_cntr ++;
                pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
                return 0;
        }

        index = ((vaddr - vseg->vm_start) + vseg->vm_offset) >> PMM_PAGE_SHIFT;

        //also hold a refcount!
        ppn = mapper_get_ppn(&vseg->vm_mapper, 
                               index, 
                               MAPPER_SYNC_OP);

        if(!ppn)
        {
                error= pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
                assert(!err); //FIXME: liberate the ppn ...
                return (VFS_FILE_IS_NULL(vseg->vm_file)) ? EIO : ENOMEM;
        }

        info.attr    = vseg->vm_pgprot;
        info.ppn     = ppn;
        info.cluster = NULL;

        //also unlock the page
        error= pmm_set_page(&vseg->vmm->pmm, vaddr, &info);

        assert(!err);//FIXME: liberate the ppn and unlock the table entry ...
        //error= pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);

        return err;
}

/////////////////////////////////////////////////////
static inline error_t vmm_do_aod( vseg_t *vseg, uint32_t vaddr)
{
        register error_t err;
        register struct page_s *page;
        register struct cluster_s *cluster;
        struct thread_s *this;
        pmm_page_info_t old;
        pmm_page_info_t new;
        kmem_req_t req;

        page      = NULL;
        old.attr  = 0;
        this      = current_thread;
  
        error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &old);

        if(err) return err;

        if(old.isAtomic == false)
        {
                this->info.spurious_pgfault_cntr ++;
                pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
                return 0;
        }

        req.type  = KMEM_PAGE;
        req.size  = 0;
        req.excep_code = AF_PGFAULT | AF_ZERO;

        if((page = kmem_alloc(&req)) == NULL)
        {
                (void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
                return ENOMEM;
        }

        page->mapper = NULL;

        new.attr    = vseg->vm_pgprot;
        new.ppn     = ppm_page2ppn(page);
        new.cluster = NULL;

        error= pmm_set_page(&vseg->vmm->pmm, vaddr, &new);
        
        if(err) goto fail_set_pg;

        cluster = current_cluster;

        if(page->cid != cluster->id)
                this->info.remote_pages_cntr ++;

        return 0;

fail_set_pg:
        (void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
        req.ptr = page;
        kmem_free(&req);

        vmm_dmsg(3, "%s: ended [ error%d ]\n", __FUNCTION__, err);
        return err;
}

VSEGION_PAGE_FAULT(vmm_default_pagefault)
{
        register struct thread_s *this;
        register error_t err;
        pmm_page_info_t info;

        if((error= pmm_get_page(&vseg->vmm->pmm, vaddr, &info)))
                return err;

        if((info.attr != 0) && (info.ppn != 0))
        {
                if((info.attr & PMM_COW) && pmm_except_isWrite(excep_code))
                {
                        error= vmm_do_cow(vseg, &info, vaddr);
                        return err;
                }

                if(info.attr & PMM_MIGRATE)
                        return vmm_do_migrate(vseg, &info, vaddr);

                if(info.attr & PMM_PRESENT)
                {
                        this = current_thread;

#if CONFIG_SHOW_SPURIOUS_PGFAULT
                        printk(WARNING, "WARNING: %s: pid %d, tid %d, cpu %d, excep_code %x but vaddr is valid %x, attr %x, ppn %x\n",
                               __FUNCTION__,
                               this->process->pid,
                               this->info.order,
                               cpu_get_id(),
                               excep_code,
                               vaddr,
                               info.attr,
                               info.ppn);
#endif

                        current_thread->info.spurious_pgfault_cntr ++;
                        pmm_tlb_flush_vaddr(vaddr, PMM_UNKNOWN);
                        return 0;
                }

                current_thread->info.spurious_pgfault_cntr ++;
                pmm_tlb_flush_vaddr(vaddr, PMM_UNKNOWN);
                return 0;
#if 0
#if CONFIG_SHOW_VMM_ERROR_MSG
                printk(ERROR, 
                       "ERROR: %s: pid %d, cpu %d, Unexpected page attributes configuration for vaddr %x, found: ppn %x, attr %x\n",
                       __FUNCTION__,
                       current_process->pid,
                       cpu_get_id(), 
                       vaddr, 
                       info.ppn,
                       info.attr);
#endif
    
                return EPERM;
#endif
        }

        if(!MAPPER_IS_NULL(vseg->vm_mapper))
                return vmm_do_mapped(vseg, vaddr, excep_code);

        return vmm_do_aod(vseg, vaddr);
}

*/