source: soft/giet_vm/giet_kernel/sys_handler.c

///////////////////////////////////////////////////////////////////////////////////
// File     : sys_handler.c
// Date     : 01/04/2012
// Author   : alain greiner and joel porquet
// Copyright (c) UPMC-LIP6
///////////////////////////////////////////////////////////////////////////////////

#include <sys_handler.h>
#include <tty_driver.h>
#include <tim_driver.h>
#include <nic_driver.h>
#include <mmc_driver.h>
#include <mwr_driver.h>
#include <cma_driver.h>
#include <ctx_handler.h>
#include <fat32.h>
#include <elf-types.h>
#include <utils.h>
#include <kernel_malloc.h>
#include <tty0.h>
#include <vmem.h>
#include <hard_config.h>
#include <giet_config.h>
#include <mapping_info.h>
#include <irq_handler.h>
#include <io.h>

#if !defined(X_SIZE) 
# error: You must define X_SIZE in the hard_config.h file
#endif

#if !defined(Y_SIZE) 
# error: You must define Y_SIZE in the hard_config.h file
#endif

#if !defined(NB_PROCS_MAX)
# error: You must define NB_PROCS_MAX in the hard_config.h file
#endif

#if !defined(SEG_BOOT_MAPPING_BASE) 
# error: You must define SEG_BOOT_MAPPING_BASE in the hard_config.h file
#endif

#if !defined(NB_TTY_CHANNELS) 
# error: You must define NB_TTY_CHANNELS in the hard_config.h file
#endif

#if (NB_TTY_CHANNELS < 1)
# error: NB_TTY_CHANNELS cannot be smaller than 1!
#endif

#if !defined(NB_TIM_CHANNELS) 
# error: You must define NB_TIM_CHANNELS in the hard_config.h file
#endif

#if !defined(NB_NIC_CHANNELS) 
# error: You must define NB_NIC_CHANNELS in the hard_config.h file
#endif

#if !defined(NB_CMA_CHANNELS) 
# error: You must define NB_CMA_CHANNELS in the hard_config.h file
#endif

#if !defined(GIET_NO_HARD_CC)
# error: You must define GIET_NO_HARD_CC in the giet_config.h file
#endif

#if !defined ( GIET_NIC_MAC4 )
# error: You must define GIET_NIC_MAC4 in the giet_config.h file
#endif

#if !defined ( GIET_NIC_MAC2 )
# error: You must define GIET_NIC_MAC2 in the giet_config.h file
#endif

////////////////////////////////////////////////////////////////////////////
//        Extern variables
////////////////////////////////////////////////////////////////////////////

// allocated in tty0.c file.
extern sqt_lock_t _tty0_sqt_lock;

// allocated in mwr_driver.c file.
extern simple_lock_t  _coproc_lock[X_SIZE*Y_SIZE];
extern unsigned int   _coproc_type[X_SIZE*Y_SIZE];
extern unsigned int   _coproc_info[X_SIZE*Y_SIZE];
extern unsigned int   _coproc_mode[X_SIZE*Y_SIZE];
extern unsigned int   _coproc_error[X_SIZE*Y_SIZE];
extern unsigned int   _coproc_trdid[X_SIZE*Y_SIZE];

// allocated in tty_driver.c file.
extern tty_fifo_t   _tty_rx_fifo[NB_TTY_CHANNELS];

// allocated in kernel_init.c file
extern static_scheduler_t* _schedulers[X_SIZE][Y_SIZE][NB_PROCS_MAX]; 

// allocated in kernel_init.c file
extern unsigned long long  _ptabs_paddr[GIET_NB_VSPACE_MAX][X_SIZE][Y_SIZE]; 

////////////////////////////////////////////////////////////////////////////////
//     Allocator protecting exclusive access to FBF by a single application.
// - The number of users in a given application should be set by a single
//   thread using an _atomic_test_and_set().
// - The allocator is atomically decremented by each user thread when 
//   the thread exit.    
////////////////////////////////////////////////////////////////////////////////

__attribute__((section(".kdata")))
unsigned int _fbf_alloc = 0;

////////////////////////////////////////////////////////////////////////////////
//     Channel allocators for multi-channels peripherals
// - The array _***_channel_allocator[channel] defines the number of user
//   threads for a dynamically allocated channel of peripheral ***.
// - The array _***_channel_wti[channel] defines the WTI index and the 
//   processor coordinates for the processor receiving the channel WTI.
////////////////////////////////////////////////////////////////////////////////

#if NB_TTY_CHANNELS
__attribute__((section(".kdata")))
unsigned int _tty_channel_alloc[NB_TTY_CHANNELS] = {0};

__attribute__((section(".kdata")))
unsigned int _tty_channel_wti[NB_TTY_CHANNELS];
#endif

#if NB_TIM_CHANNELS
__attribute__((section(".kdata")))
unsigned int _tim_channel_alloc[NB_TIM_CHANNELS] = {0};

__attribute__((section(".kdata")))
unsigned int _tim_channel_wti[NB_TIM_CHANNELS];
#endif

#if NB_CMA_CHANNELS
__attribute__((section(".kdata")))
unsigned int _cma_channel_alloc[NB_CMA_CHANNELS] = {0};

__attribute__((section(".kdata")))
unsigned int _cma_channel_wti[NB_CMA_CHANNELS];
#endif

#if NB_NIC_CHANNELS
__attribute__((section(".kdata")))
unsigned int _nic_rx_channel_alloc[NB_NIC_CHANNELS] = {0};

__attribute__((section(".kdata")))
unsigned int _nic_rx_channel_wti[NB_NIC_CHANNELS];

__attribute__((section(".kdata")))
unsigned int _nic_tx_channel_alloc[NB_NIC_CHANNELS] = {0};

__attribute__((section(".kdata")))
unsigned int _nic_tx_channel_wti[NB_NIC_CHANNELS];
#endif

////////////////////////////////////////////////////////////////////////////
//     NIC_RX and NIC_TX kernel chbuf arrays 
////////////////////////////////////////////////////////////////////////////

__attribute__((section(".kdata")))
nic_chbuf_t  _nic_ker_rx_chbuf[NB_NIC_CHANNELS] __attribute__((aligned(64)));

__attribute__((section(".kdata")))
nic_chbuf_t  _nic_ker_tx_chbuf[NB_NIC_CHANNELS] __attribute__((aligned(64)));

////////////////////////////////////////////////////////////////////////////
//     FBF related chbuf
// The physical address of this chbuf is required for L2 cache sync.
////////////////////////////////////////////////////////////////////////////

__attribute__((section(".kdata")))
fbf_chbuf_t          _fbf_ker_chbuf  __attribute__((aligned(64)));

__attribute__((section(".kdata")))
unsigned long long   _fbf_chbuf_paddr;

////////////////////////////////////////////////////////////////////////////
//    Initialize the syscall vector with syscall handlers
// Note: This array must be synchronised with the define in file stdio.h
////////////////////////////////////////////////////////////////////////////

__attribute__((section(".kdata")))
const void * _syscall_vector[64] = 
{
    &_sys_proc_xyp,                  /* 0x00 */
    &_get_proctime,                  /* 0x01 */
    &_sys_procs_number,              /* 0x02 */
    &_sys_xy_from_ptr,               /* 0x03 */
    &_sys_ukn,                       /* 0x04 */
    &_sys_vseg_get_vbase,            /* 0x05 */
    &_sys_vseg_get_length,           /* 0x06 */
    &_sys_heap_info,                 /* 0x07 */
    &_sys_fbf_size,                  /* 0x08 */
    &_sys_fbf_alloc,                 /* 0x09 */ 

#if NB_CMA_CHANNELS
    &_sys_fbf_cma_alloc,             /* 0x0A */
    &_sys_fbf_cma_init_buf,          /* 0x0B */
    &_sys_fbf_cma_start,             /* 0x0C */
    &_sys_fbf_cma_display,           /* 0x0D */
    &_sys_fbf_cma_stop,              /* 0x0E */
    &_sys_fbf_cma_check,             /* 0x0F */
#else
    &_sys_ukn,                       /* 0x0A */
    &_sys_ukn,                       /* 0x0B */
    &_sys_ukn,                       /* 0x0C */
    &_sys_ukn,                       /* 0x0D */
    &_sys_ukn,                       /* 0x0E */
    &_sys_ukn,                       /* 0x0F */
#endif

    &_sys_applications_status,       /* 0x10 */
    &_sys_fbf_sync_write,            /* 0x11 */
    &_sys_fbf_sync_read,             /* 0x12 */
    &_sys_ukn,                       /* 0x13 */
    &_sys_tim_alloc,                 /* 0x14 */
    &_sys_tim_start,                 /* 0x15 */ 
    &_sys_tim_stop,                  /* 0x16 */
    &_sys_kill_application,          /* 0x17 */
    &_sys_exec_application,          /* 0x18 */   
    &_sys_ukn,                       /* 0x19 */
    &_sys_pthread_control,           /* 0x1A */
    &_sys_pthread_yield,             /* 0x1B */
    &_sys_pthread_kill,              /* 0x1C */
    &_sys_pthread_create,            /* 0x1D */
    &_sys_pthread_join,              /* 0x1E */
    &_sys_pthread_exit,              /* 0x1F */

    &_fat_open,                      /* 0x20 */
    &_sys_fat_read,                  /* 0x21 */
    &_sys_fat_write,                 /* 0x22 */
    &_fat_lseek,                     /* 0x23 */
    &_fat_file_info,                 /* 0x24 */
    &_fat_close,                     /* 0x25 */
    &_fat_remove,                    /* 0x26 */
    &_fat_rename,                    /* 0x27 */
    &_fat_mkdir,                     /* 0x28 */
    &_fat_opendir,                   /* 0x29 */
    &_fat_closedir,                  /* 0x2A */
    &_fat_readdir,                   /* 0x2B */
    &_sys_fat_pread,                 /* 0x2C */
    &_sys_fat_mmap,                  /* 0x2D */
    &_sys_fat_munmap,                /* 0x2E */
    &_sys_fat_dump,                  /* 0x2F */

#if NB_NIC_CHANNELS
    &_sys_nic_alloc,                 /* 0x30 */
    &_sys_nic_start,                 /* 0x31 */
    &_sys_nic_move,                  /* 0x32 */
    &_sys_nic_stop,                  /* 0x33 */
    &_sys_nic_stats,                 /* 0x34 */
    &_sys_nic_clear,                 /* 0x35 */ 
#else
    &_sys_ukn,                       /* 0x30 */
    &_sys_ukn,                       /* 0x31 */
    &_sys_ukn,                       /* 0x32 */
    &_sys_ukn,                       /* 0x33 */
    &_sys_ukn,                       /* 0x34 */
    &_sys_ukn,                       /* 0x35 */
#endif

    &_sys_tty_write,                 /* 0x36 */
    &_sys_tty_read,                  /* 0x37 */
    &_sys_tty_alloc,                 /* 0x38 */
    &_sys_ukn,                       /* 0x39 */
    &_sys_ukn,                       /* 0x3A */
    &_sys_coproc_completed,          /* 0x3B */
    &_sys_coproc_alloc,              /* 0x3C */
    &_sys_coproc_channel_init,       /* 0x3D */
    &_sys_coproc_run,                /* 0x3E */
    &_sys_coproc_release,            /* 0x3F */
};



///////////////////////////////////////////////////////////////////////////////////
//           File system related syscall handlers 
///////////////////////////////////////////////////////////////////////////////////

///////////////////////////////////////////////////////////////////////////////////
// This function is called by the _sys_fat_mmap() function.
// It implements a simplistic pages allocator from the MMAP vseg of the vspace
// identified by the <vspace_id> argument. The number of pages is defined by
// the <count> argument. 
// Allocated pages are never released, and the allocator is the "psegid" field
// in the vseg mapping, that is initialised to 0 by the boot code.
// In order to support concurrent system calls, the allocator must be
// atomically incremented.
///////////////////////////////////////////////////////////////////////////////////
// Returns the buffer vbase address in case of success.
// Returns 0 in case of error (no MMAP vseg or not enough space).
///////////////////////////////////////////////////////////////////////////////////
static unsigned int _mmap_pages_alloc( unsigned int  vspace_id,
                                       unsigned int  count )
{
    mapping_header_t*   header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t*   vspace  = _get_vspace_base( header );
    mapping_vseg_t*     vseg    = _get_vseg_base( header );

    // loop on the vsegs to find the MMAP vseg vbase, length, and offset
    unsigned int  vseg_id;
    unsigned int  vbase;    
    unsigned int  offset;   
    unsigned int  length;
    unsigned int  found = 0;
    for (vseg_id = vspace[vspace_id].vseg_offset;
         vseg_id < (vspace[vspace_id].vseg_offset + vspace[vspace_id].vsegs);
         vseg_id++) 
    {
        if ( vseg[vseg_id].type == VSEG_TYPE_MMAP ) 
        {
            offset = _atomic_increment( &vseg[vseg_id].psegid , count );
            vbase  = vseg[vseg_id].vbase;
            length = vseg[vseg_id].length;
            found  = 1;
            break;
        }
    }

    if ( (found == 0) || (((offset + count)<<12) > length) )
    {
        return 0;
    }
    else
    {
        return vbase + (offset<<12);
    }
}  // end _mmap_pages_alloc()


///////////////////////////////////////////////////////////////////////////////////
// This function implements the giet_fat_read() system call.
///////////////////////////////////////////////////////////////////////////////////
int _sys_fat_read( unsigned int fd_id,
                   unsigned int vaddr,
                   unsigned int count )
{
    return _fat_read( fd_id,
                      vaddr,
                      count,
                      0,                       // no paddr extension
                      0,                       // no forced offset
                      0 );                     // no special mode
}

////////////////////////////////////////////////////////////////
// This function implements the giet_fat_pread() system call.
////////////////////////////////////////////////////////////////
int _sys_fat_pread( unsigned int fd_id,
                    unsigned int vaddr,
                    unsigned int count,
                    unsigned int offset )
{
    return _fat_read( fd_id,
                      vaddr,
                      count,
                      0,                        // no paddr extension
                      offset,
                      FAT_FORCED_OFFSET );     
}

////////////////////////////////////////////////////////////////
// This function implements the giet_fat_write() system call.
////////////////////////////////////////////////////////////////
int _sys_fat_write( unsigned int fd_id,
                    unsigned int vaddr,
                    unsigned int count )
{
    return _fat_write( fd_id,
                       vaddr,
                       count,
                       0,                       // no paddr extension
                       0 );                     // no special mode
}

///////////////////////////////////////////////////////////////////////////////////
// This function implements the "giet_fat_mmap()" system call.
// It allocates <count> contiguous pages from the MMAP vseg of the calling vspace.
// It maps all these pages directly to the file_cache defined by <fd_id>.
// The <offset> in the file is a number of pages. 
// The <prot> argument defines the access modes MAP_PROT_WRITE and MAP_PROT_EXEC.
// In writable mode, the file size is extended to the (offset + count) pages.
// In non writable mode, the file size must be >= (offset + count) pages.
// It has the following limitations:
// - it does not support MAP_PRIVATE (no copy on write).
// - it does not support MAP_FIXED (no forced user buffer address).
// - it does not support MAP_ANONYMOUS (only file mapping).
// - the <offset> and <count> arguments must multiple of 4 Kbytes.
///////////////////////////////////////////////////////////////////////////////////
// Returns memory buffer vbase in case of success.
// Returns 0 on error.
///////////////////////////////////////////////////////////////////////////////////
int _sys_fat_mmap( unsigned int fd_id,
                   unsigned int count, 
                   unsigned int offset,
                   unsigned int prot )
{
    // takes the FAT lock and register it in thread context
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    _spin_lock_acquire( &_fat.fat_lock );
    _atomic_or( &psched->context[ltid].slot[CTX_LOCKS_ID] , LOCKS_MASK_FAT ); 

    // check fd_id overflow
    if ( fd_id >= GIET_OPEN_FILES_MAX )
    {
        _spin_lock_release( &_fat.fat_lock );
        _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FAT ); 

        _printf("\n[GIET ERROR] _sys_fat_mmap(): illegal file descriptor\n");
        return 0;
    }

    // check file open
    if ( _fat.fd[fd_id].allocated == 0 )
    {
        _spin_lock_release( &_fat.fat_lock );
        _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FAT ); 

        _printf("\n[GIET ERROR] _sys_fat_mmap(): file not open\n" );
        return 0;
    }

    // get access modes
    unsigned int writable   = prot & MAP_PROT_WRITE;
    unsigned int executable = prot & MAP_PROT_EXEC;

    // get inode pointer  
    fat_inode_t* inode  = _fat.fd[fd_id].inode;

    // check file writable
    if ( _fat.fd[fd_id].read_only && writable )
    {
        _spin_lock_release( &_fat.fat_lock );
        _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FAT ); 

        _printf("\n[GIET ERROR] _sys_fat_mmap(): file <%s> is read-only\n",
                inode->name );
        return 0;
    }

    // get vspace index and calling proc coordinates
    unsigned int vsid    = _get_context_slot( CTX_VSID_ID );

    // compute first and last cluster indexes
    unsigned int  first_cluster  = offset;
    unsigned int  last_cluster   = offset + count - 1; 

#if GIET_DEBUG_MMAP
unsigned int procid  = _get_procid();
unsigned int x_id    = procid >> (Y_WIDTH + P_WIDTH);
unsigned int y_id    = (procid >> P_WIDTH) & ((1<<Y_WIDTH)-1);
unsigned int p_id    = procid & ((1<<P_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_MMAP )
_printf("\n[DEBUG MMAP] _sys_fat_mmap() : P[%d,%d,%d] enters at cycle %d\n"
        " for file %s / size = %x / cluster = %x / cache = %x / desc[0] = %x\n"
        " first_cluster = %d / last_cluster = %d\n",
        x_id , y_id , p_id , _get_proctime() ,
        inode->name , inode->size , inode->cluster ,
        (unsigned int)inode->cache , (unsigned int)inode->cache->children[0] ,
        first_cluster , last_cluster );
#endif

    // get buffer vbase from MMAP vseg in calling vspace
    unsigned int buffer_vbase = _mmap_pages_alloc( vsid, count ); 
    if ( buffer_vbase == 0 ) 
    {
        _spin_lock_release( &_fat.fat_lock );
        _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FAT ); 

        _printf("\n[GIET ERROR] _sys_fat_mmap(): no space in MMAP vseg to map file %s\n",
                _fat.fd[fd_id].inode->name );
        return 0;
    }
  
    // set flags for all pages in buffer
    unsigned int flags = 0;
    flags |= PTE_C;                    // cachable
    flags |= PTE_U;                    // user access
    flags |= PTE_L;                    // local access (unused)
    flags |= PTE_R;                    // remote access (unused)
    if ( executable ) flags |= PTE_X;
    if ( writable   ) flags |= PTE_W;

    // loop on pages to be mapped
    unsigned int cid;
    unsigned int user_vaddr = buffer_vbase;
    for ( cid = first_cluster ; cid <= last_cluster ; cid++ )  
    {
        // get file_cache buffer vaddr
        unsigned char*     cache_vaddr;
        fat_cache_desc_t*  pdesc;
        if ( _get_file_cache_buffer( inode,
                                     cid,
                                     writable,
                                     &pdesc ) )
        {
            _spin_lock_release( &_fat.fat_lock );
            _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FAT ); 

            _printf("\n[FAT ERROR] _sys_fat_mmap(): cannot access file <%s>\n",
                    _fat.fd[fd_id].inode->name );
            return 0;
        }

        cache_vaddr = pdesc->buffer;
        
        // compute buffer paddr
        unsigned long long  cache_paddr;
        unsigned int        unused;
        cache_paddr = _v2p_translate( (unsigned int)cache_vaddr, &unused );

        // map the user_vaddr to cache_paddr 
        // in all pages tables defined for the vspace
        unsigned int x_cur;
        unsigned int y_cur;
        for  ( x_cur = 0 ; x_cur < X_SIZE ; x_cur++ )
        {
            for  ( y_cur = 0 ; y_cur < Y_SIZE ; y_cur++ )
            {
                if ( _ptabs_paddr[vsid][x_cur][y_cur] )  // Page table is defined
                {
                    _v2p_add_pte2( vsid,
                                   x_cur,
                                   y_cur,
                                   user_vaddr >> 12,
                                   flags,
                                   cache_paddr >> 12,
                                   0 );
                }
            }
        }

#if GIET_DEBUG_MMAP
if ( _get_proctime() > GIET_DEBUG_MMAP )
_printf("\n[DEBUG MMAP] _sys_fat_mmap() : P[%d,%d,%d] map cluster_id = %d\n"
        " user_vaddr = %x / cache_paddr = %l\n",
        x_id , y_id , p_id , cid , user_vaddr , cache_paddr );
#endif

        // increment user buffer vaddr
        user_vaddr += 4096;
    }

    // releases the FAT lock
    _spin_lock_release( &_fat.fat_lock );
    _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FAT ); 

#if GIET_DEBUG_MMAP
if ( _get_proctime() > GIET_DEBUG_MMAP )
_printf("\n[DEBUG MMAP] _sys_fat_mmap() : P[%d,%d,%d] returns buffer %x for file %s\n",
        x_id , y_id , p_id , buffer_vbase , inode->name );
#endif

   // returns pointer on mapped buffer in user space
    return buffer_vbase;

}  // end _sys_fat_mmap()


//////////////////////////////////////////////////////////////////
// This function implements the giet_fat_munmap() system call.
//////////////////////////////////////////////////////////////////
int _sys_fat_munmap( unsigned int vaddr,
                     unsigned int count ) 
{
    // get vspace index 
    unsigned int  vsid   = _get_context_slot( CTX_VSID_ID );

    // loop on pages
    unsigned int page;
    for ( page = 0 ; page < count ; page++ )
    {
        unsigned int vpn = (vaddr>>12) + page;

        // loop on all page tables defined for the vspace
        unsigned int x_cur;
        unsigned int y_cur;
        for  ( x_cur = 0 ; x_cur < X_SIZE ; x_cur++ )
        {
            for  ( y_cur = 0 ; y_cur < Y_SIZE ; y_cur++ )
            {
                if ( _ptabs_paddr[vsid][x_cur][y_cur] )  // Page table is defined
                {
                    _v2p_del_pte2( vsid , x_cur , y_cur , vpn );
                }
            }
        }
    }
    return 0;
}  // end _sys_fat_munmap()




///////////////////////////////////////////////////////////////////////////////////
// This service function is called by the _sys_fat_dump() kernel function.
// It displays on the calling thread terminal the content of a 512 bytes buffer
// defined by the <buf> pointer.
// The <string> , <cluster_id> and <block_id> arguments are only used 
// to build a meaningful title: <string> is an identifier, <cluster_id> is 
// the cluster index (in file or FAT), <block_id> is the block index (in cluster).
///////////////////////////////////////////////////////////////////////////////////
static void _dump_block( unsigned char* buf,
                         char*          string,
                         unsigned int   cluster_id,
                         unsigned int   block_id )
{
    unsigned int line;
    unsigned int word;

    _user_printf("\n---  %s : cluster_id = %d / block_id = %d ---\n",
            string , cluster_id , block_id );

    for ( line = 0 ; line < 16 ; line++ )
    {
        // display line index 
        _user_printf("%x : ", line );

        // display 8*4 bytes hexa per line
        for ( word=0 ; word<8 ; word++ )
        {
            unsigned int byte  = (line<<5) + (word<<2);
            unsigned int hexa  = (buf[byte  ]<<24) |
                                 (buf[byte+1]<<16) |
                                 (buf[byte+2]<< 8) |
                                 (buf[byte+3]      );
            _user_printf(" %X |", hexa );
        }
        _user_printf("\n");
    }
} // end _dump_block()  


///////////////////////////////////////////////////////////////////////////////////
// This function implements the giet_fat_dump() system call.
// It analyse the <type>, <pathname> and <sector_id> arguments to select
// a 512 bytes sector, and display the sector content on the calling thread TTY.
// The <type> argument can take the following values :
// - DUMP_BS   : boot sector
// - DUMP_FS   : fs_info sector
// - DUMP_FAT  : fat sector, identified by <sector_id>
// - DUMP_FILE : file sector, identified by <pathname> and <sector_id>
// - DUMP_DIR  : directory sector, identified by <pathname> and <sector_id>
// The <sector_id> argument defines the sector index in the file or in the FAT.
// For a file or a directory, or for the FAT itself, it uses the kernel File-Caches
// or Fat-Cache, and the Inode-Tree.
// For the boot sector, or fs_info sector, it access directly the block device,
// that is loaded in the FAT descriptor block-buffer.
///////////////////////////////////////////////////////////////////////////////////
// Returns 0 in case of success.
// Returns 1 in case of error.
///////////////////////////////////////////////////////////////////////////////////
int _sys_fat_dump( unsigned int type,
                   char*        pathname,
                   unsigned int sector_id )
{
    unsigned char* buf;                            // pointer on 512 bytes buffer
    unsigned int   cluster_id = sector_id >> 3;    // cluster index in cache
    unsigned int   block_id   = sector_id & 0x7;   // cluster index in cache
    unsigned int   fd_id;                          // file/dir descriptor index
    
    if( type == DUMP_BS )            // dump the boot sector
    {
        // access block_device
        buf = _fat.block_buffer; 
        if ( _fat_ioc_access( 1,                   // descheduling mode
                              1,                   // read 
                              0,                   // boot sector lba
                              (unsigned int)buf,
                              1 ) )                // one block                    
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot access device for BS\n");
            return 1;
        }
        _fat.block_buffer_lba = 0;

        // dump boot sector
        _dump_block( buf, "boot sector" , 0 , 0 );
    }
    else if ( type == DUMP_FS )     // dump the fs_info sector
    {
        // access block_device
        buf = _fat.block_buffer; 
        if ( _fat_ioc_access( 1,                  // descheduling mode
                              1,                  // read 
                              _fat.fs_info_lba,   // fs_info sector lba
                              (unsigned int)buf,
                              1 ) )               // one block                    
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot access device for FS\n");
            return 1;
        }
        _fat.block_buffer_lba = _fat.fs_info_lba;

        // dump fs-info sector
        _dump_block( buf, "fs_info sector" , 0 , 0 );
    }
    else if ( type == DUMP_FAT )    // dump a fat sector
    {
        // get pointer on the relevant buffer descriptor in Fat-Cache
        fat_cache_desc_t*  pdesc;
        if ( _get_fat_cache_buffer( cluster_id, &pdesc ) )         
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot access Fat-Cache\n");
            return 1;
        }
        buf = pdesc->buffer + (block_id<<9);

        // dump fat sector
        _dump_block( buf, "fat" , cluster_id , block_id );
    }
    else if ( type == DUMP_FILE )    // dump a file sector
    {
        // open file 
        fd_id = _fat_open( pathname , O_RDONLY );
        if ( fd_id < 0 )
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot open file <%s>\n", pathname );
            return 1;
        }
        fat_inode_t* inode = _fat.fd[fd_id].inode;

        // get pointer on the relevant buffer descriptor in File-Cache
        fat_cache_desc_t*  pdesc;
        if ( _get_file_cache_buffer( inode, cluster_id, 0, &pdesc ) )         
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot access file <%s>\n", pathname );
            return 1;
        }
        buf = pdesc->buffer + (block_id<<9);

        // dump file sector
        _dump_block( buf, pathname , cluster_id , block_id );

        // close file
        _fat_close( fd_id );
    }
    else if ( type == DUMP_DIR )    // dump a directory sector
    {
        // open directory 
        fd_id = _fat_opendir( pathname );
        if ( fd_id < 0 )
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot open <%s>\n", pathname );
            return 1;
        }
        fat_inode_t* inode = _fat.fd[fd_id].inode;

        // get pointer on the relevant buffer descriptor in File-Cache
        fat_cache_desc_t*  pdesc;
        if ( _get_file_cache_buffer( inode, cluster_id, 0, &pdesc ) )         
        {
            _printf("\n[FAT ERROR] _sys_fat_dump(): cannot access <%s>\n", pathname );
            return 1;
        }
        buf = pdesc->buffer + (block_id<<9);

        // dump directory sector
        _dump_block( buf, pathname , cluster_id , block_id );

        // close directory
        _fat_closedir( fd_id );
    }

    return 0;
}  // end sys_fat_dump



//////////////////////////////////////////////////////////////////////////////
//           Applications related syscall handlers 
//////////////////////////////////////////////////////////////////////////////

////////////////////////////////////////////////////////////////////////
// This function is called by the _sys_exec_application() function
// to reload all data segments contained in an application.elf file.
// File checking is minimal, because these segments have already
// been loaded by the boot code. 
////////////////////////////////////////////////////////////////////////
static unsigned int _load_writable_segments( mapping_vspace_t*  vspace )
{

#if GIET_DEBUG_EXEC  
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _load_writable_segments() at cycle %d\n"
        "P[%d,%d,%d] enters for %s\n",
        _get_proctime() , x , y , p , vspace->name );
#endif

    mapping_header_t*  header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vseg_t*    vseg    = _get_vseg_base(header);

    unsigned int vseg_id;        // vseg index in mapping
    char         buf[4096];      // buffer to store one cluster
    unsigned int fd    = 0;      // file descriptor
    unsigned int found = 0;

    // first scan on vsegs in vspace to find the .elf pathname
    for (vseg_id = vspace->vseg_offset;
         vseg_id < (vspace->vseg_offset + vspace->vsegs);
         vseg_id++)
    {
        if( vseg[vseg_id].type == VSEG_TYPE_ELF )
        {
            // open the .elf file associated to vspace
            fd = _fat_open( vseg[vseg_id].binpath , O_RDONLY );
            if ( fd < 0 ) return 1;

#if GIET_DEBUG_EXEC
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _load_writable_segments() at cycle %d\n"
        "P[%d,%d,%d] open %s / fd = %d\n", 
        _get_proctime() , x , y , p , vseg[vseg_id].binpath , fd );
#endif
            found = 1;
            break;
 
        }
    }
 
    // check .elf file found
    if ( found == 0 )
    {
        _printf("\n[GIET ERROR] _load_writable_segments() : .elf not found\n");
        return 1;
    }

    // load Elf-Header into buffer from .elf file
    if ( _fat_lseek( fd, 0, SEEK_SET ) < 0 )
    {
        _printf("\n[GIET ERROR] _load_writable_segments() : cannot seek\n");
        _fat_close( fd );
        return 1;
    }
    if ( _fat_read( fd, (unsigned int)buf, 4096, 0, 0, 0 ) < 0 )
    {
        _printf("\n[GIET ERROR] _load_writable_segments() : cannot read\n");
        _fat_close( fd );
        return 1;
    }

#if GIET_DEBUG_EXEC
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _load_writable_segments() at cycle %d\n"
        "P[%d,%d,%d] loaded Elf-Header\n",
        _get_proctime() , x , y , p );
#endif

    // get nsegments and Program-Header-Table offset from Elf-Header
    Elf32_Ehdr*  elf_header_ptr = (Elf32_Ehdr*)buf;
    unsigned int offset         = elf_header_ptr->e_phoff;
    unsigned int nsegments      = elf_header_ptr->e_phnum;

    // load Program-Header-Table from .elf file
    if ( _fat_lseek( fd, offset, SEEK_SET ) < 0 )
    {
        _fat_close( fd );
        return 1;
    }
    if ( _fat_read( fd, (unsigned int)buf, 4096, 0, 0, 0 ) < 0 )
    {
        _fat_close( fd );
        return 1;
    }

#if GIET_DEBUG_EXEC
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _load_writable_segments() at cycle %d\n"
        "P[%d,%d,%d] loaded Program-Header-Table\n",
        _get_proctime() , x , y , p );
#endif

    // set Program-Header-Table pointer
    Elf32_Phdr*  elf_pht_ptr = (Elf32_Phdr*)buf;

    // second scan on vsegs in vspace to load the seg_data segments :
    // - type == VSEG_TYPE_ELF
    // - non eXecutable
    for (vseg_id = vspace->vseg_offset;
         vseg_id < (vspace->vseg_offset + vspace->vsegs);
         vseg_id++)
    {
        if( (vseg[vseg_id].type == VSEG_TYPE_ELF) &&   // type ELF
            ((vseg[vseg_id].mode & 0x4) == 0) )        // non executable 
        {
            // get vbase and pbase
            paddr_t      pbase = vseg[vseg_id].pbase; 
            unsigned int vbase = vseg[vseg_id].vbase;

            // scan segments in Progam-Header-Table to find match
            // No match checking as the segment was previously found
            unsigned int seg;
            for (seg = 0 ; seg < nsegments ; seg++)
            {
                if ( (elf_pht_ptr[seg].p_type == PT_LOAD) &&    // loadable
                     (elf_pht_ptr[seg].p_flags & PF_W)    &&    // writable
                     (elf_pht_ptr[seg].p_vaddr == vbase) )      // matching
                {
                    // Get segment offset and size in .elf file
                    unsigned int seg_offset = elf_pht_ptr[seg].p_offset;
                    unsigned int seg_size   = elf_pht_ptr[seg].p_filesz;

                    // compute destination address and extension for _fat_read()
                    unsigned int dest   = (unsigned int)pbase;
                    unsigned int extend = (unsigned int)(pbase>>32);

                    // load the segment
                    if ( _fat_lseek( fd, seg_offset, SEEK_SET ) < 0 ) 
                    {
                        _fat_close( fd );
                        return 1;
                    }
                    if ( _fat_read( fd, dest, seg_size, extend, 0, FAT_PADDR_MODE ) < 0 )
                    {
                        _fat_close( fd );
                        return 1;
                    }
                }
            }

#if GIET_DEBUG_EXEC
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _load_writable_segments() at cycle %d\n"
        "P[%d,%d,%d] loaded segment %x\n",
        _get_proctime() , x , y , p , vbase );
#endif
        }
    }  // end loop on writable & loadable segments

    // close .elf file
    _fat_close( fd );

    return 0;
}  // end load_writable_segments()



///////////////////////////////////////
int _sys_exec_application( char* name )
{
    mapping_header_t * header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t * vspace  = _get_vspace_base(header);
    mapping_thread_t * thread  = _get_thread_base(header);
    mapping_vseg_t   * vseg    = _get_vseg_base(header);

    unsigned int vspace_id;
    unsigned int thread_id;

#if GIET_DEBUG_EXEC 
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_exec_application() at cycle %d\n"
        "P[%d,%d,%d] enters for vspace %s\n",
        _get_proctime() , x, y, p, name );
#endif

    unsigned int y_size = header->y_size;

    // scan vspaces to find matching vspace name
    for (vspace_id = 0 ; vspace_id < header->vspaces ; vspace_id++) 
    {
        if ( _strcmp( vspace[vspace_id].name, name ) == 0 )  // vspace found
        {

#if GIET_DEBUG_EXEC 
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_exec_application() at cycle %d\n"
        "P[%d,%d,%d] found vspace %s\n",
        _get_proctime() , x, y, p, name );
#endif
            // reload writable segments
            if ( _load_writable_segments( &vspace[vspace_id] ) )
            {
                _printf("[GIET ERROR] _sys_exec_application() : "
                        "can't load data segment for vspace %s\n", name );
                return SYSCALL_CANNOT_LOAD_DATA_SEGMENT;
            }
 
#if GIET_DEBUG_EXEC 
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_exec_application() at cycle %d\n"
        "P[%d,%d,%d] loaded all writable segments for vspace %s\n",
        _get_proctime() , x, y, p, name );
#endif
            // scan threads in vspace with three goals :
            // - check all threads desactivated
            // - re-initialise all threads contexts
            // - find main thread
            unsigned int        main_found  = 0;
            unsigned int        main_ltid   = 0;
            static_scheduler_t* main_psched = NULL;
            unsigned int        min         = vspace[vspace_id].thread_offset; 
            unsigned int        max         = min + vspace[vspace_id].threads; 
            for ( thread_id = min ; thread_id < max ; thread_id++ ) 
            {
                // get thread identifiers : [x,y,p,ltid]
                unsigned int cid   = thread[thread_id].clusterid;
                unsigned int x     = cid / y_size;
                unsigned int y     = cid % y_size;
                unsigned int p     = thread[thread_id].proclocid;
                unsigned int ltid  = thread[thread_id].ltid;
                unsigned int vsid  = thread[thread_id].stack_vseg_id;

                // get scheduler pointer
                static_scheduler_t* psched = _schedulers[x][y][p];

                // check thread non active
                if ( psched->context[ltid].slot[CTX_NORUN_ID] == 0 )  // runnable !!!
                {
                    _printf("\n[GIET ERROR] in _sys_exec_application() : "
                            "thread %s already active in vspace %s\n",
                            thread[thread_id].name, name );
                    return SYSCALL_THREAD_ALREADY_ACTIVE;
                }
                
                // initialise thread context
                unsigned int ctx_epc = psched->context[ltid].slot[CTX_ENTRY_ID];
                unsigned int ctx_sp  = vseg[vsid].vbase + vseg[vsid].length;
                unsigned int ctx_ra  = (unsigned int)&_ctx_eret;
                unsigned int ctx_sr  = GIET_SR_INIT_VALUE;

                psched->context[ltid].slot[CTX_EPC_ID] = ctx_epc;
                psched->context[ltid].slot[CTX_RA_ID]  = ctx_ra;
                psched->context[ltid].slot[CTX_SR_ID]  = ctx_sr;
                psched->context[ltid].slot[CTX_SP_ID]  = ctx_sp;

                // register information required to activate main thread
                // actual activation done when threads initialisation is completed
                if ( thread[thread_id].is_main ) 
                {
                    main_psched = psched;
                    main_ltid   = ltid;
                    main_found  = 1;
                }
 
#if GIET_DEBUG_EXEC 
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_exec_application() at cycle %d\n"
        "P[%d,%d,%d] initialise thread %s in vspace %s\n",
        _get_proctime() , x, y, p, thread[thread_id].name , name );
#endif
            }  // end loop on threads

            // activate main thread
            if ( main_found )
            {
                main_psched->context[main_ltid].slot[CTX_NORUN_ID] = 0;
            }
            else
            {
                _printf("\n[GIET ERROR] in _sys_exec_application() : "
                        "main not found in vspace %s\n", name );
                return SYSCALL_MAIN_NOT_FOUND;
            }
             
            _printf("\n[GIET WARNING] application %s launched : %d threads\n",
                    name , max-min );

            return SYSCALL_OK;
        }
    }  // end of loop on vspaces

    // vspace not found
    _printf("\n[GIET ERROR] in _sys_exec_application() : "
            "vspace %s not found\n", name );
    return SYSCALL_VSPACE_NOT_FOUND;

}  // end _sys_exec_application()
    

///////////////////////////////////////
int _sys_kill_application( char* name )
{
    mapping_header_t * header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t * vspace  = _get_vspace_base(header);
    mapping_thread_t * thread  = _get_thread_base(header);

    unsigned int vspace_id;
    unsigned int thread_id;

#if GIET_DEBUG_EXEC
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_kill_application() at cycle %d\n"
        "P[%d,%d,%d] enters for vspace %s\n",
        _get_proctime() , x , y , p , name );
#endif

    // shell cannot be killed
    if ( _strcmp( name , "shell" ) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_kill_application() : "
                "%s application cannot be killed\n", name );
        return SYSCALL_APPLI_CANNOT_BE_KILLED;
    }

    // scan vspaces to find matching vspace name
    for (vspace_id = 0; vspace_id < header->vspaces; vspace_id++) 
    {
        if ( _strcmp( vspace[vspace_id].name, name ) == 0 ) 
        {
            // scan threads to send KILL signal to all threads in vspace
            unsigned int y_size = header->y_size;
            unsigned int min    = vspace[vspace_id].thread_offset; 
            unsigned int max    = min + vspace[vspace_id].threads; 
            for ( thread_id = min ; thread_id < max ; thread_id++ ) 
            {
                unsigned int cid   = thread[thread_id].clusterid;
                unsigned int x     = cid / y_size;
                unsigned int y     = cid % y_size;
                unsigned int p     = thread[thread_id].proclocid;
                unsigned int ltid  = thread[thread_id].ltid;

                // get scheduler pointer for processor running the thread
                static_scheduler_t* psched  = (static_scheduler_t*)_schedulers[x][y][p];

                // set KILL signal bit
                _atomic_or( &psched->context[ltid].slot[CTX_SIGS_ID] , SIGS_MASK_KILL );
            } 

            _printf("\n[GIET WARNING] application %s killed / %d threads\n",
                    name , max-min );

            return SYSCALL_OK;
        }
    }  // en loop on vspaces

    _printf("\n[GIET ERROR] in _sys_kill_application() : "
            "application %s not found\n", name );
    return SYSCALL_VSPACE_NOT_FOUND;

}  // end _sys_kill_application()
    


//////////////////////////////////////////
int _sys_applications_status( char* name )
{
    mapping_header_t *  header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_thread_t *  thread  = _get_thread_base(header);
    mapping_vspace_t *  vspace  = _get_vspace_base(header);
    mapping_cluster_t * cluster = _get_cluster_base(header);

    unsigned int thread_id;   // thread index in mapping
    unsigned int vspace_id;   // vspace index in mapping

    // scan vspaces
    for( vspace_id = 0 ; vspace_id < header->vspaces ; vspace_id++ )
    {
        if ( (name == NULL) || (_strcmp(vspace[vspace_id].name , name ) == 0) )
        {
            _user_printf("\n--- vspace %s ---\n", vspace[vspace_id].name );

            // scan all threads in vspace
            unsigned int min = vspace[vspace_id].thread_offset ;
            unsigned int max = min + vspace[vspace_id].threads ;
            for ( thread_id = min ; thread_id < max ; thread_id++ )
            {
                unsigned int         clusterid = thread[thread_id].clusterid;
                unsigned int         p         = thread[thread_id].proclocid;
                unsigned int         x         = cluster[clusterid].x;
                unsigned int         y         = cluster[clusterid].y;
                unsigned int         ltid      = thread[thread_id].ltid;
                static_scheduler_t*  psched    = (static_scheduler_t*)_schedulers[x][y][p];
                unsigned int         norun     = psched->context[ltid].slot[CTX_NORUN_ID];
                unsigned int         tty       = psched->context[ltid].slot[CTX_TTY_ID];
                unsigned int         current   = psched->current;

                if ( current == ltid )
                _user_printf(" - thread %s / P[%d,%d,%d] / ltid = %d / "
                             "TTY = %d / norun = %x : running\n", 
                             thread[thread_id].name, x, y, p, ltid, tty, norun );
                else if ( norun == 0 )
                _user_printf(" - thread %s / P[%d,%d,%d] / ltid = %d / "
                             "TTY = %d / norun = %x : runable\n", 
                             thread[thread_id].name, x, y, p, ltid, tty, norun);
                else
                _user_printf(" - thread %s / P[%d,%d,%d] / ltid = %d / "
                             "TTY = %d / norun = %x : blocked\n", 
                             thread[thread_id].name, x, y, p, ltid, tty, norun);
            }
        }
    }
    _user_printf("\n");

    return SYSCALL_OK;
}  // end _sys_applications_status()



/////////////////////////////////////////////////////////////////////////////
//          Threads related syscall handlers
/////////////////////////////////////////////////////////////////////////////

////////////////////////////////////////////////
int _sys_pthread_create( unsigned int*  buffer,
                         void*          attr,
                         void*          function,
                         void*          arg )
{
    // attr argument not supported
    if ( attr != NULL )
    {
        _printf("\n[GIET ERROR] in _sys_pthread_create() : "
                "attr argument not supported\n" );
        
        return SYSCALL_PTHREAD_ARGUMENT_NOT_SUPPORTED;
    }

    // get pointers in mapping
    mapping_header_t*    header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_thread_t*    thread   = _get_thread_base(header);
    mapping_vspace_t*    vspace   = _get_vspace_base(header);
    mapping_cluster_t*   cluster  = _get_cluster_base(header);

    // get scheduler for processor running the calling thread
    static_scheduler_t* psched = (static_scheduler_t*)_get_sched();

    // get calling thread local index in scheduler
    unsigned int  current = psched->current;

    // get vspace index
    unsigned int  vspace_id = psched->context[current].slot[CTX_VSID_ID];

#if GIET_DEBUG_EXEC 
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_create() at cycle %d\n"
        "P[%d,%d,%d] enters for vspace %s / entry = %x\n",
        _get_proctime() , x , y , p , vspace[vspace_id].name , (unsigned int)function );
#endif

    unsigned int thread_id;        // searched thread : local index in mapping
    unsigned int clusterid;        // searched thread : cluster index
    unsigned int lpid;             // searched thread : processor local index
    unsigned int ltid;             // searched thread : scheduler thread index
    unsigned int cx;               // searched thread : X coordinate for searched thread
    unsigned int cy;               // searched thread : Y coordinate for searched thread
    unsigned int entry;            // searched thread : entry point
    unsigned int norun;            // searched thread : norun vector
    unsigned int trdid;            // searched thread : thread identifier

    // scan threads in vspace to find an inactive thread matching function
    unsigned int min   = vspace[vspace_id].thread_offset;
    unsigned int max   = min + vspace[vspace_id].threads;
    unsigned int found = 0;
    for ( thread_id = min ; (thread_id < max) && (found == 0) ; thread_id++ )
    {
        // get thread coordinates [cx,cy,lpid] and ltid from mapping
        ltid       = thread[thread_id].ltid;
        clusterid  = thread[thread_id].clusterid; 
        lpid       = thread[thread_id].proclocid;
        cx = cluster[clusterid].x;
        cy = cluster[clusterid].y;

        // get thread scheduler pointer
        psched = _schedulers[cx][cy][lpid];

        // get thread entry-point, norun-vector, and trdid from context
        entry = psched->context[ltid].slot[CTX_ENTRY_ID];
        norun = psched->context[ltid].slot[CTX_NORUN_ID];
        trdid = psched->context[ltid].slot[CTX_TRDID_ID];

        // check matching
        if ( ((unsigned int)function == entry ) && 
             (norun & NORUN_MASK_THREAD)  ) found = 1;

    }  // end loop on threads

    if ( found )  // one matching inactive thread has been found
    {
        // set argument value in thread context
        if ( arg != NULL ) psched->context[ltid].slot[CTX_A0_ID] = (unsigned int)arg;

        // activate thread
        psched->context[ltid].slot[CTX_NORUN_ID] = 0;

        // return launched thead global identifier
        *buffer   = trdid;
                
#if GIET_DEBUG_EXEC 
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_create() at cycle %d\n"
        "P[%d,%d,%d] exit : thread %x launched in vspace %s\n",
        _get_proctime() , x , y , p , trdid , vspace[vspace_id].name );
#endif
        return SYSCALL_OK;
    }
    else         // no matching thread found
    {
        _printf("\n[GIET ERROR] in _sys_pthread_create() : "
                "no matching thread for entry = %x in vspace %s\n",
                (unsigned int)function , vspace[vspace_id].name );
        
        return SYSCALL_THREAD_NOT_FOUND;
    }

} // end _sys_pthread_create()


///////////////////////////////////////////
int _sys_pthread_join( unsigned int  trdid,
                       void*         ptr )
{
    // ptr argument not supported
    if ( ptr != NULL )
    {
        _printf("\n[GIET ERROR] in _sys_pthread_join() : "
                "ptr argument not supported, must be NULL\n" );
        
        return SYSCALL_PTHREAD_ARGUMENT_NOT_SUPPORTED;
    }

    // get calling thread vspace
    unsigned int  caller_vspace = _get_context_slot( CTX_VSID_ID );
                
#if GIET_DEBUG_EXEC 
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_join() at cycle %d\n"
        "P[%d,%d,%d] enters for thread %x in vspace %d\n",
        _get_proctime() , x , y , p , trdid , caller_vspace );
#endif

    // get target thread indexes from trdid
    unsigned int cx   = (trdid>>24) & 0xFF;
    unsigned int cy   = (trdid>>16) & 0xFF;
    unsigned int lpid = (trdid>>8 ) & 0xFF;
    unsigned int ltid = (trdid    ) & 0xFF;

    // get target thread scheduler, vspace and registered trdid
    static_scheduler_t*  psched   = _schedulers[cx][cy][lpid];
    unsigned int target_vspace    = psched->context[ltid].slot[CTX_VSID_ID];
    unsigned int registered_trdid = psched->context[ltid].slot[CTX_TRDID_ID];

    // check trdid
    if ( trdid != registered_trdid )
    {
       _printf("\nerror in _sys_pthread_join() : "
               "trdid = %x / registered_trdid = %x\n",
               trdid , registered_trdid );

       return SYSCALL_UNCOHERENT_THREAD_CONTEXT;
    }
   
    // check calling thread and target thread in same vspace
    if ( caller_vspace != target_vspace )
    {
       _printf("\n[GIET ERROR] in _sys_pthread_join() : "
               " calling thread and target thread not in same vspace\n");

       return SYSCALL_NOT_IN_SAME_VSPACE;
    }

    // get target thread state
    volatile unsigned int* pnorun = &psched->context[ltid].slot[CTX_NORUN_ID];
    while (*pnorun == 0) _sys_pthread_yield();

#if GIET_DEBUG_EXEC 
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_join() at cycle %d\n"
        "P[%d,%d,%d] exit for thread %x in vspace %d\n",
        _get_proctime() , x , y , p , trdid , caller_vspace );
#endif

    return SYSCALL_OK;

}  // end _sys_pthread_join()
                       

////////////////////////////////////////
int _sys_pthread_kill( pthread_t  trdid,
                       int        signal )
{
    // get calling thread vspace
    unsigned int  caller_vspace = _get_context_slot( CTX_VSID_ID );

#if GIET_DEBUG_EXEC 
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_kill() at cycle %d\n"
        "P[%d,%d,%d] enters for thread %x in vspace %d\n",
        _get_proctime() , x , y , p , trdid , caller_vspace );
#endif


    // get and check target thread indexes from trdid
    unsigned int cx   = (trdid>>24) & 0xFF;
    unsigned int cy   = (trdid>>16) & 0xFF;
    unsigned int lpid = (trdid>>8 ) & 0xFF;
    unsigned int ltid = (trdid    ) & 0xFF;

    // get target thread scheduler, vspace and registered trdid
    static_scheduler_t*  psched       = _schedulers[cx][cy][lpid];
    unsigned int target_vspace        = psched->context[ltid].slot[CTX_VSID_ID];
    unsigned int registered_trdid = psched->context[ltid].slot[CTX_TRDID_ID];

    // check trdid
    if ( trdid != registered_trdid )
    {
       _printf("\n[GIET ERROR] in _sys_pthread_kill() : trdid = %x"
               " / registered_trdid = %x\n", trdid , registered_trdid );
       return SYSCALL_UNCOHERENT_THREAD_CONTEXT;
    }
   
    // check calling thread and target thread in same vspace
    if ( caller_vspace != target_vspace )
    {
       _printf("\n[GIET ERROR] in _sys_pthread_kill() : not in same vspace\n");
       return SYSCALL_NOT_IN_SAME_VSPACE;
    }

    // register KILL signal in target thread context if required
    if ( signal )
    {
        _atomic_or( &psched->context[ltid].slot[CTX_SIGS_ID] , SIGS_MASK_KILL );
    }

#if GIET_DEBUG_EXEC 
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_kill() at cycle %d\n"
        "P[%d,%d,%d] exit for thread %x in vspace %d\n",
        _get_proctime() , x , y , p , trdid , caller_vspace );
#endif

    return SYSCALL_OK;

}  // end _sys_pthread_kill()


/////////////////////////////////////
int _sys_pthread_exit( void* string ) 
{
    mapping_header_t * header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t * vspace  = _get_vspace_base(header);

    unsigned int ltid       = _get_context_slot(CTX_LTID_ID);
    unsigned int trdid      = _get_context_slot(CTX_TRDID_ID);
    unsigned int vsid       = _get_context_slot(CTX_VSID_ID);

    // print exit message
    if ( string != NULL )
    {
        _printf("\n[GIET WARNING] Exit thread %x in vspace %s\n"
                "               Cause : %s\n\n",
                     trdid , vspace[vsid].name , (char*) string );
    }
    
    // get scheduler pointer for calling thread
    static_scheduler_t*  psched = (static_scheduler_t*)_get_sched();

    // register KILL signal in calling thread context (suicid request)
    _atomic_or( &psched->context[ltid].slot[CTX_SIGS_ID] , SIGS_MASK_KILL );

    // deschedule calling thread
    unsigned int save_sr;  
    _it_disable( &save_sr );

    _ctx_switch();

    return SYSCALL_OK;

}  // end _sys_pthread_exit()

////////////////////////
int _sys_pthread_yield() 
{
    unsigned int save_sr;

    _it_disable( &save_sr );
    _ctx_switch();
    _it_restore( &save_sr );

    return SYSCALL_OK;
}

//////////////////////////////////////////////////
int _sys_pthread_control( unsigned  int  command,
                          char*     vspace_name,
                          char*     thread_name )
{

#if GIET_DEBUG_EXEC 
unsigned int gpid       = _get_procid();
unsigned int cluster_xy = gpid >> P_WIDTH;
unsigned int p          = gpid & ((1<<P_WIDTH)-1);
unsigned int x          = cluster_xy >> Y_WIDTH;
unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
if ( _get_proctime() > GIET_DEBUG_EXEC )
_printf("\n[DEBUG EXEC] _sys_pthread_control() at cycle %d\n"
        "P[%d,%d,%d] enter for vspace %s / thread %s / command = %d\n",
        _get_proctime() , x , y , p , vspace_name, thread_name, command );
#endif

    // get pointers in mapping
    mapping_header_t*    header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_thread_t*    thread   = _get_thread_base(header);
    mapping_vspace_t*    vspace   = _get_vspace_base(header);
    mapping_cluster_t*   cluster  = _get_cluster_base(header);

    unsigned int found;

    // search vspace name to get vspace index: vsid
    found = 0;
    unsigned int   vsid;
    for( vsid = 0 ; vsid < header->vspaces ; vsid++ )
    {
        if ( _strcmp( vspace[vsid].name, vspace_name ) == 0 )
        {
            found = 1;
            break;
        }
    }

    if ( found == 0 ) return SYSCALL_VSPACE_NOT_FOUND;

    // search thread name in vspace to get thread index: tid
    found = 0;
    unsigned int   tid;
    unsigned int   min = vspace[vsid].thread_offset;
    unsigned int   max = min + vspace[vsid].threads;
    for( tid = min ; tid < max ; tid++ )
    {
        if ( _strcmp( thread[tid].name, thread_name ) == 0 )
        {
            found = 1;
            break;
        }
    }

    if ( found == 0 ) return SYSCALL_THREAD_NOT_FOUND;

    // get target thread coordinates
    unsigned int cid  = thread[tid].clusterid;
    unsigned int cx   = cluster[cid].x;
    unsigned int cy   = cluster[cid].y;
    unsigned int cp   = thread[tid].proclocid;
    unsigned int ltid = thread[tid].ltid;

    // get target thread scheduler
    static_scheduler_t* psched = _schedulers[cx][cy][cp];

    // check trdid and vsid
    unsigned int trdid = cx<<24 | cy<<16 | cp<<8 | ltid;
    if ( (psched->context[ltid].slot[CTX_TRDID_ID] != trdid) ||
         (psched->context[ltid].slot[CTX_VSID_ID]  != vsid) )
    {
        return SYSCALL_UNCOHERENT_THREAD_CONTEXT;
    }

    // execute command
    if ( command == THREAD_CMD_PAUSE ) 
    {
        _atomic_or ( &psched->context[ltid].slot[CTX_NORUN_ID],  NORUN_MASK_THREAD );
        return SYSCALL_OK;
    }
    else if ( command == THREAD_CMD_RESUME )
    {
        _atomic_and( &psched->context[ltid].slot[CTX_NORUN_ID], ~NORUN_MASK_THREAD );
        return SYSCALL_OK;
    }
    else if ( command == THREAD_CMD_CONTEXT )
    {
        _user_printf( " - CTX_TRDID  = %x\n"
                      " - CTX_VSID   = %x\n"
                      " - CTX_EPC    = %x\n"
                      " - CTX_PTAB   = %x\n"
                      " - CTX_PTPR   = %x\n"
                      " - CTX_SR     = %x\n"
                      " - CTX_RA     = %x\n"
                      " - CTX_SP     = %x\n"
                      " - CTX_ENTRY  = %x\n"
                      " - CTX_NORUN  = %x\n"
                      " - CTX_SIGS   = %x\n"
                      " - CTX_LOCKS  = %x\n"
                      " - CTX_TTY    = %x\n"
                      " - CTX_NIC_RX = %x\n"
                      " - CTX_NIC_TX = %x\n"
                      " - CTX_CMA_RX = %x\n"
                      " - CTX_CMA_TX = %x\n"
                      " - CTX_CMA_FB = %x\n",
                      psched->context[ltid].slot[CTX_TRDID_ID], 
                      psched->context[ltid].slot[CTX_VSID_ID], 
                      psched->context[ltid].slot[CTX_EPC_ID], 
                      psched->context[ltid].slot[CTX_PTAB_ID], 
                      psched->context[ltid].slot[CTX_PTPR_ID], 
                      psched->context[ltid].slot[CTX_SR_ID], 
                      psched->context[ltid].slot[CTX_RA_ID], 
                      psched->context[ltid].slot[CTX_SP_ID], 
                      psched->context[ltid].slot[CTX_ENTRY_ID], 
                      psched->context[ltid].slot[CTX_NORUN_ID],
                      psched->context[ltid].slot[CTX_SIGS_ID],
                      psched->context[ltid].slot[CTX_LOCKS_ID],
                      psched->context[ltid].slot[CTX_TTY_ID],
                      psched->context[ltid].slot[CTX_NIC_RX_ID],
                      psched->context[ltid].slot[CTX_NIC_TX_ID],
                      psched->context[ltid].slot[CTX_CMA_RX_ID],
                      psched->context[ltid].slot[CTX_CMA_TX_ID],
                      psched->context[ltid].slot[CTX_CMA_FB_ID] );
        return SYSCALL_OK;
    }
    else
    {
        return SYSCALL_ILLEGAL_THREAD_COMMAND_TYPE;
    }

} // end _sys_pthread_control()




//////////////////////////////////////////////////////////////////////////////
//           Coprocessors related syscall handlers 
//////////////////////////////////////////////////////////////////////////////

//////////////////////////////////////////////////
int _sys_coproc_alloc( unsigned int   cluster_xy,
                       unsigned int   coproc_type,
                       unsigned int*  return_info )
{
    mapping_header_t  * header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_cluster_t * cluster = _get_cluster_base(header);
    mapping_periph_t  * periph  = _get_periph_base(header);

    // compute cluster coordinates and cluster index in mapping
    unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
    unsigned int x          = (cluster_xy>>Y_WIDTH) & ((1<<X_WIDTH)-1);
    unsigned int cluster_id = x * Y_SIZE + y;
 
    // search coprocessor in cluster
    mapping_periph_t*  found = NULL;
    unsigned int min   = cluster[cluster_id].periph_offset;
    unsigned int max   = min + cluster[cluster_id].periphs;
    unsigned int periph_id;
    for ( periph_id = min ; periph_id < max ; periph_id++ )
    {
        if ( (periph[periph_id].type == PERIPH_TYPE_MWR) &&
             (periph[periph_id].subtype == coproc_type) )
        {
            found = &periph[periph_id];
            break;
        }
    } 

    if ( found != NULL )
    {
        // get the lock 
        _simple_lock_acquire( &_coproc_lock[cluster_id] );

        // register coproc characteristics in kernel arrays
        _coproc_type[cluster_id] = coproc_type;
        _coproc_info[cluster_id] = (found->arg0 & 0xFF)     |
                                   (found->arg1 & 0xFF)<<8  |
                                   (found->arg2 & 0xFF)<<16 |
                                   (found->arg3 & 0xFF)<<24 ;

        // returns coprocessor info
        *return_info = _coproc_info[cluster_id];

#if GIET_DEBUG_COPROC
_printf("\n[DEBUG COPROC] _sys_coproc_alloc() at cycle %d\n"
        "type = %d in cluster[%d,%d] / coproc_info = %x\n",
        _get_proctime() , coproc_type, x , y , *return_info );
#endif
        return SYSCALL_OK;
    }
    else
    {
         _printf("\n[GIET_ERROR] in _sys_coproc_alloc(): "
                 "no coprocessor with type %d in cluster[%d,%d]\n",
                 coproc_type , x , y );

        return SYSCALL_COPROCESSOR_NOT_FOUND;
    }
}  // end _sys_coproc_alloc()

////////////////////////////////////////////////////////
int _sys_coproc_release( unsigned int cluster_xy,
                         unsigned int coproc_type )
{
    // compute cluster coordinates and cluster index 
    unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
    unsigned int x          = (cluster_xy>>Y_WIDTH) & ((1<<X_WIDTH)-1);
    unsigned int cluster_id = x * Y_SIZE + y;

    // check coprocessor type
    if ( _coproc_type[cluster_id] != coproc_type )
    {
         _printf("\n[GIET_ERROR] in _sys_coproc_release(): "
                 "no coprocessor of type %d allocated in cluster[%d,%d]\n", 
                 coproc_type, x , y );

         return SYSCALL_COPROCESSOR_NON_ALLOCATED;
    }

    unsigned int info       = _coproc_info[cluster_id];
    unsigned int nb_to      = info & 0xFF;
    unsigned int nb_from    = (info>>8) & 0xFF;
    unsigned int channel;

    // stops coprocessor and communication channels
    _mwr_set_coproc_register( cluster_xy , 0 , 0 );
    for ( channel = 0 ; channel < (nb_from + nb_to) ; channel++ )
    {
        _mwr_set_channel_register( cluster_xy , channel , MWR_CHANNEL_RUNNING , 0 );
    }

    // release coprocessor lock
    _simple_lock_release( &_coproc_lock[cluster_id] );

#if GIET_DEBUG_COPROC
_printf("\n[DEBUG COPROC] _sys_coproc_release() at cycle %d\n"
        "type = %d in cluster[%d,%d]\n",
        _get_proctime() , coproc_type , x , y );
#endif

    return SYSCALL_OK;
}  // end _sys_coproc_release()

////////////////////////////////////////////////////////////////
int _sys_coproc_channel_init( unsigned int            cluster_xy,
                              unsigned int            coproc_type,
                              unsigned int            channel,
                              giet_coproc_channel_t*  desc )
{
    // compute cluster coordinates and cluster index 
    unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
    unsigned int x          = (cluster_xy>>Y_WIDTH) & ((1<<X_WIDTH)-1);
    unsigned int cluster_id = x * Y_SIZE + y;

    // check coprocessor type
    if ( _coproc_type[cluster_id] != coproc_type )
    {
         _printf("\n[GIET_ERROR] in _sys_coproc_release(): "
                 "no coprocessor of type %d allocated in cluster[%d,%d]\n", 
                 coproc_type, x , y );

         return SYSCALL_COPROCESSOR_NON_ALLOCATED;
    }

    // check channel mode
    unsigned mode = desc->channel_mode;
    if ( (mode != MODE_MWMR) && 
         (mode != MODE_DMA_IRQ) && 
         (mode != MODE_DMA_NO_IRQ) )
    {
         _printf("\n[GIET_ERROR] in _sys_coproc_channel_init(): "
                 "illegal mode\n");

         return SYSCALL_COPROCESSOR_ILLEGAL_MODE;
    }

    // get memory buffer size
    unsigned int size = desc->buffer_size;
 
    // physical addresses
    unsigned long long buffer_paddr;
    unsigned int       buffer_lsb;
    unsigned int       buffer_msb;
    unsigned long long status_paddr = 0;
    unsigned int       status_lsb;
    unsigned int       status_msb;
    unsigned long long lock_paddr = 0;
    unsigned int       lock_lsb;
    unsigned int       lock_msb;

    unsigned int       flags;     // unused

    // compute memory buffer physical address
    buffer_paddr = _v2p_translate( desc->buffer_vaddr , &flags );
    buffer_lsb   = (unsigned int)buffer_paddr;
    buffer_msb   = (unsigned int)(buffer_paddr>>32); 

    // call MWMR_DMA driver
    _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_MODE, mode ); 
    _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_SIZE, size ); 
    _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_BUFFER_LSB, buffer_lsb ); 
    _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_BUFFER_MSB, buffer_msb ); 
                       
    if ( mode == MODE_MWMR )
    {
        // compute MWMR descriptor physical address
        status_paddr = _v2p_translate( desc->status_vaddr , &flags );
        status_lsb = (unsigned int)status_paddr;
        status_msb = (unsigned int)(status_paddr>>32); 

        // call MWMR_DMA driver
        _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_MWMR_LSB, status_lsb ); 
        _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_MWMR_MSB, status_msb ); 

        // compute lock physical address
        lock_paddr = _v2p_translate( desc->lock_vaddr , &flags );
        lock_lsb = (unsigned int)lock_paddr;
        lock_msb = (unsigned int)(lock_paddr>>32); 

        // call MWMR_DMA driver
        _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_LOCK_LSB, lock_lsb ); 
        _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_LOCK_MSB, lock_msb ); 
    }

#if GIET_DEBUG_COPROC
_printf("\n[DEBUG COPROC] _sys_coproc_channel_init() at cycle %d\n"
        "cluster[%d,%d] / channel = %d / mode = %d / buffer_size = %d\n"
        "buffer_vaddr = %x / status_vaddr = %x / lock_vaddr = %x\n"
        "buffer_paddr = %l / status_paddr = %l / lock_paddr = %l\n",
        _get_proctime() , x , y , channel , mode , size ,
        desc->buffer_vaddr, desc->status_vaddr, desc->lock_vaddr,
        buffer_paddr, status_paddr, lock_paddr );
#endif
        
    return SYSCALL_OK;
} // end _sys_coproc_channel_init()

//////////////////////////////////////////////
int _sys_coproc_run( unsigned int  cluster_xy,
                     unsigned int  coproc_type )
{
    // compute cluster coordinates and cluster index 
    unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
    unsigned int x          = (cluster_xy>>Y_WIDTH) & ((1<<X_WIDTH)-1);
    unsigned int cluster_id = x * Y_SIZE + y;

    // check coprocessor type
    if ( _coproc_type[cluster_id] != coproc_type )
    {
         _printf("\n[GIET_ERROR] in _sys_coproc_release(): "
                 "no coprocessor of type %d allocated in cluster[%d,%d]\n", 
                 coproc_type, x , y );

         return SYSCALL_COPROCESSOR_NON_ALLOCATED;
    }

    unsigned int info       = _coproc_info[cluster_id];
    unsigned int nb_to      = info & 0xFF;
    unsigned int nb_from    = (info>>8) & 0xFF;
    unsigned int mode       = 0xFFFFFFFF;
    unsigned int channel;

    // check channels modes, and register coprocessor running mode 
    for ( channel = 0 ; channel < (nb_from + nb_to) ; channel++ )
    {
        unsigned int temp;
        temp = _mwr_get_channel_register( cluster_xy , channel , MWR_CHANNEL_MODE );

        if ( mode == 0xFFFFFFFF ) 
        {
            mode = temp;
        }
        else if ( temp != mode )
        {
            _printf("\n[GIET_ERROR] in _sys_coproc_run(): "
                    "channels don't have same mode in coprocessor[%d,%d]\n", x , y );

            return SYSCALL_COPROCESSOR_ILLEGAL_MODE;
        }
    }
    _coproc_mode[cluster_id] = mode;

    // start all communication channels
    for ( channel = 0 ; channel < (nb_from + nb_to) ; channel++ )
    {
        _mwr_set_channel_register( cluster_xy , channel , MWR_CHANNEL_RUNNING , 1 );
    }

    //////////////////////////////////////////////////////////////////////////
    if ( (mode == MODE_MWMR) || (mode == MODE_DMA_NO_IRQ) )  // no descheduling
    {
        // start coprocessor
        _mwr_set_coproc_register( cluster_xy , 0 , 1 );

#if GIET_DEBUG_COPROC
if ( mode == MODE_MWMR )
_printf("\n[DEBUG COPROC] _sys_coproc_run() at cycle %d\n"
        "type = %d / cluster[%d,%d] / MODE_MWMR\n", 
        _get_proctime() , coproc_type , x , y );
else
_printf("\n[DEBUG COPROC] _sys_coproc_run() at cycle %d\n"
        "type = %d / cluster[%d,%d] / MODE_DMA_NO_IRQ\n", 
        _get_proctime() , coproc_type , x , y );
#endif

        return SYSCALL_OK;
    }
    ///////////////////////////////////////////////////////////////////////////
    else                                // mode == MODE_DMA_IRQ => descheduling
    {
        // register calling thread trdid
        unsigned int trdid = _get_thread_trdid();
        _coproc_trdid[cluster_id] = trdid;

        // enters critical section
        unsigned int save_sr;
        _it_disable( &save_sr ); 

        // set NORUN_MASK_COPROC bit
        static_scheduler_t* psched  = _get_sched();
        unsigned int        ltid    = _get_thread_ltid();
        unsigned int*       ptr     = &psched->context[ltid].slot[CTX_NORUN_ID];
        _atomic_or( ptr , NORUN_MASK_COPROC );

        // start coprocessor
        _mwr_set_coproc_register( cluster_xy , 0 , 1 );

#if GIET_DEBUG_COPROC
_printf("\n[DEBUG COPROC] _sys_coproc_run() at cycle %d\n"
        "thread %x starts coprocessor / type = %d / cluster[%d,%d] / MODE_DMA_IRQ\n", 
        _get_proctime() , trdid , coproc_type , x , y );
#endif

        // deschedule thread
        _ctx_switch(); 

#if GIET_DEBUG_COPROC
_printf("\n[DEBUG COPROC] _sys_coproc_run() at cycle %d\n"
        "thread %x resume after coprocessor[%d,%d] completion\n", 
        _get_proctime() , trdid , x , y );
#endif

        // restore SR
        _it_restore( &save_sr );

        // return error computed by mwr_isr()
        return _coproc_error[cluster_id];
    } 
} // end _sys_coproc_run()

////////////////////////////////////////////////////
int _sys_coproc_completed( unsigned int  cluster_xy,
                           unsigned int  coproc_type )
{
    // compute cluster coordinates and cluster index 
    unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
    unsigned int x          = (cluster_xy>>Y_WIDTH) & ((1<<X_WIDTH)-1);
    unsigned int cluster_id = x * Y_SIZE + y;

    // check coprocessor type
    if ( _coproc_type[cluster_id] != coproc_type )
    {
         _printf("\n[GIET_ERROR] in _sys_coproc_release(): "
                 "no coprocessor of type %d allocated in cluster[%d,%d]\n", 
                 coproc_type, x , y );

         return SYSCALL_COPROCESSOR_NON_ALLOCATED;
    }

    unsigned int mode = _coproc_mode[cluster_id];

    // analyse possible errors
    if ( mode == MODE_DMA_NO_IRQ )
    {
        unsigned int info       = _coproc_info[cluster_id];
        unsigned int nb_to      = info & 0xFF;
        unsigned int nb_from    = (info>>8) & 0xFF;
        unsigned int error      = 0;
        unsigned int channel;
        unsigned int status;

        // get status for all channels, and signal reported errors
        for ( channel = 0 ; channel < (nb_to +nb_from) ; channel++ )
        {
            do
            {
                status = _mwr_get_channel_register( cluster_xy, channel,
                                                    MWR_CHANNEL_STATUS );
                if ( status == MWR_CHANNEL_ERROR_DATA )
                {
                    _printf("\n[GIET_ERROR] in _sys_coproc_completed(): "
                            "channel %d / DATA_ERROR\n", channel );
                    error = 1;
                }
                else if ( status == MWR_CHANNEL_ERROR_LOCK )
                {
                    _printf("\n[GIET_ERROR] in _sys_coproc_completed()"
                            " / channel %d / LOCK_ERROR\n", channel );
                    error = 1;
                }
                else if ( status == MWR_CHANNEL_ERROR_DESC )
                {
                    _printf("\n[GIET_ERROR] in _sys_coproc_completed()"
                            " / channel %d / DESC_ERROR\n", channel );
                    error = 1;
                }
            } 
            while ( status == MWR_CHANNEL_BUSY );

            // reset channel
            _mwr_set_channel_register( cluster_xy, channel, MWR_CHANNEL_RUNNING , 0 ); 

        }  // end for channels

        if ( error )
        {
            return SYSCALL_COPROCESSOR_ILLEGAL_MODE;
        }
        else
        {

#if GIET_DEBUG_COPROC
_printf("\n[DEBUG COPROC] _sys_coproc_completed() at cycle %d\n"
        "coprocessor type = %d / cluster[%d,%d] completes operation for thread %d\n", 
        _get_proctime() , coproc_type , x , y , _get_thread_trdid() );
#endif
            return SYSCALL_OK;
        }
    }
    else  // mode == MODE_MWMR or MODE_DMA_IRQ
    {
        _printf("\n[GIET ERROR] in sys_coproc_completed(): "
                "coprocessor[%d,%d] is not running in MODE_DMA_NO_IRQ\n", x , y );

        return SYSCALL_COPROCESSOR_ILLEGAL_MODE;
    }
} // end _sys_coproc_completed()


//////////////////////////////////////////////////////////////////////////////
//             TTY related syscall handlers 
//////////////////////////////////////////////////////////////////////////////

/////////////////////////////////////////
int _sys_tty_alloc( unsigned int shared )
{
    unsigned int channel = 0;    // allocated TTY channel

    // get vsid for the calling thread
    unsigned int vsid  = _get_context_slot( CTX_VSID_ID );

    mapping_header_t  *header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t  *vspace   = _get_vspace_base(header);
    mapping_thread_t  *thread   = _get_thread_base(header);
    
    // compute number of users
    unsigned int users;
    if ( shared )  users = vspace[vsid].threads;
    else           users = 1;

#if NB_TTY_CHANNELS > 1
    // get trdid for the calling thread
    unsigned int trdid = _get_thread_trdid();

    // get a TTY channel 
    for ( channel = 0 ; channel < NB_TTY_CHANNELS ; channel++ )
    {
        unsigned int* palloc  = &_tty_channel_alloc[channel];

        if ( _atomic_test_and_set( palloc , users ) == 0 ) break;
    }
    if ( ( channel >= NB_TTY_CHANNELS ) )
    {
        _printf("\n[GIET_ERROR] in _sys_tty_alloc() : "
                "no TTY channel available for thread %x\n", trdid );
        return SYSCALL_NO_CHANNEL_AVAILABLE;
    }

    // initialise allocated TTY channel
    _tty_init( channel );

    // allocate a WTI mailbox to the calling proc if external IRQ
    unsigned int wti_id;
    if ( USE_PIC ) _ext_irq_alloc( ISR_TTY_RX , channel , &wti_id ); 

    // register wti_id and coordinates for processor receiving WTI
    unsigned int procid = _get_procid();
    unsigned int x      = procid >> (Y_WIDTH + P_WIDTH);
    unsigned int y      = (procid >> P_WIDTH) & ((1<<Y_WIDTH)-1);
    unsigned int p      = procid & ((1<<P_WIDTH)-1);
    _tty_channel_wti[channel] = x<<24 | y<<16 | p<<8 | wti_id;

#else

    unsigned int* palloc  = &_tty_channel_alloc[0];
    _atomic_increment( palloc, users );

#endif
   
    // update CTX_TTY_ID 
    if ( shared )         // for all threads in vspace
    {
        // scan threads in vspace
        unsigned int tid;
        for (tid = vspace[vsid].thread_offset;
             tid < (vspace[vsid].thread_offset + vspace[vsid].threads);
             tid++)
        {
            unsigned int y_size        = header->y_size;
            unsigned int cid           = thread[tid].clusterid;
            unsigned int x             = cid / y_size;
            unsigned int y             = cid % y_size;
            unsigned int p             = thread[tid].proclocid;
            unsigned int ltid          = thread[tid].ltid;
            static_scheduler_t* psched = (static_scheduler_t*)_schedulers[x][y][p];

            psched->context[ltid].slot[CTX_TTY_ID] = channel;
        }
    }
    else                  // for calling thread only
    {
        _set_context_slot( CTX_TTY_ID, channel );
    }

    return SYSCALL_OK;
}  // end _sys_tty_alloc()

//////////////////////
int _sys_tty_release()     // NOTE: not a syscall: used by _ctx_kill_thread()
{
    unsigned int channel = _get_context_slot( CTX_TTY_ID );

    if ( channel == -1 )
    {
        unsigned int trdid = _get_thread_trdid();
        _printf("\n[GIET_ERROR] in _sys_tty_release() : "
                "TTY channel already released for thread %x\n", trdid );

        return SYSCALL_CHANNEL_ALREADY_ALLOCATED;
    }

    // reset CTX_TTY_ID for the calling thread
    _set_context_slot( CTX_TTY_ID , 0xFFFFFFFF );

    // atomically decrement the _tty_channel_allocator[] array
    _atomic_increment( &_tty_channel_alloc[channel] , -1 );

    // release WTI mailbox if TTY channel no more used
    if ( USE_PIC  && (_tty_channel_alloc[channel] == 0) ) 
    {
        _ext_irq_release( ISR_TTY_RX , channel );
    }

    return SYSCALL_OK;
}  // end sys_tty_release()

////////////////////////////////////////
int _sys_tty_write( const char*  buffer,    
                    unsigned int length,    // number of characters
                    unsigned int channel)   // channel index 
{
    unsigned int  nwritten;

    // compute and check tty channel
    if( channel == 0xFFFFFFFF )  channel = _get_context_slot(CTX_TTY_ID);

    if( channel >= NB_TTY_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_tty_write() : "
                "no TTY channel allocated for thread %x\n", _get_thread_trdid() );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // write string to TTY channel
    for (nwritten = 0; nwritten < length; nwritten++) 
    {
        // check tty's status 
        if ( _tty_get_register( channel, TTY_STATUS ) & 0x2 )  break;

        // write one byte
        if (buffer[nwritten] == '\n') 
        {
            _tty_set_register( channel, TTY_WRITE, (unsigned int)'\r' );
        }
        _tty_set_register( channel, TTY_WRITE, (unsigned int)buffer[nwritten] );
    }
    
    return nwritten;
}

///////////////////////////////////////
int _sys_tty_read( char*        buffer, 
                   unsigned int length,    // unused
                   unsigned int channel)   // channel index
{
    // compute and check tty channel
    if( channel == 0xFFFFFFFF )  channel = _get_context_slot(CTX_TTY_ID);

    if( channel >= NB_TTY_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_tty_read() : "
                "no TTY channel allocated for thread %x\n", _get_thread_trdid() );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    unsigned int save_sr;
    unsigned int found = 0;

    // get pointer on TTY_RX FIFO
    tty_fifo_t*  fifo = &_tty_rx_fifo[channel];

    // try to read one character from FIFO
    // blocked in while loop until success
    while ( found == 0 )
    {
        if ( fifo->sts == 0)   // FIFO empty => deschedule
        {
            // enters critical section
             _it_disable( &save_sr );

            // set NORUN_MASK_TTY bit for calling thread
            static_scheduler_t* psched  = (static_scheduler_t*)_get_sched();
            unsigned int ltid = psched->current;
            _atomic_or( &psched->context[ltid].slot[CTX_NORUN_ID] , NORUN_MASK_TTY );

            // register descheduling thread trdid
            fifo->trdid = _get_thread_trdid();

             // deschedule calling thread
            _ctx_switch();

            // exit critical section
            _it_restore( &save_sr );
        }
        else                             // FIFO not empty => get one character
        {
            *buffer   = fifo->data[fifo->ptr];
            fifo->sts = fifo->sts - 1;
            fifo->ptr = (fifo->ptr + 1) % TTY_FIFO_DEPTH;
            found     = 1;
        }
    }

    return 1;
}



//////////////////////////////////////////////////////////////////////////////
//             TIMER related syscall handlers 
//////////////////////////////////////////////////////////////////////////////

////////////////////
int _sys_tim_alloc()
{

#if NB_TIM_CHANNELS

    unsigned int channel;    // allocated TIMER channel

    unsigned int trdid = _get_thread_trdid();

    // check no TIMER already allocated to calling thread
    if ( _get_context_slot( CTX_TIM_ID ) < NB_TIM_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_tim_alloc() : "
                "TIMER channel already allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_ALREADY_ALLOCATED;
    }

    // get a TIMER channel
    for ( channel = 0 ; channel < NB_TIM_CHANNELS ; channel++ )
    {
        unsigned int* palloc  = &_tim_channel_alloc[channel];

        if ( _atomic_test_and_set( palloc , 1 ) == 0 ) break;
    }
    if ( channel >= NB_TIM_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_tim_alloc() : "
                "no TIMER channel available for thread %x\n", trdid );

        return SYSCALL_NO_CHANNEL_AVAILABLE;
    }

    // allocate a WTI mailbox to the calling proc if external IRQ
    unsigned int wti_id;
    if ( USE_PIC ) _ext_irq_alloc( ISR_TIMER , channel , &wti_id ); 

    // register wti_id and coordinates for processor receiving WTI
    unsigned int procid = _get_procid();
    unsigned int x      = procid >> (Y_WIDTH + P_WIDTH);
    unsigned int y      = (procid >> P_WIDTH) & ((1<<Y_WIDTH)-1);
    unsigned int p      = procid & ((1<<P_WIDTH)-1);
    _tim_channel_wti[channel] = x<<24 | y<<16 | p<<8 | wti_id;
   
    // update CTX_TIM_ID in thread context
    _set_context_slot( CTX_TIM_ID, channel );

    return SYSCALL_OK;

#else

    _printf("\n[GIET ERROR] in _sys_tim_alloc(): NB_TIM_CHANNELS == 0\n");

    return SYSCALL_NO_CHANNEL_AVAILABLE;

#endif
}  // end _sys_tim_alloc()


//////////////////////
int _sys_tim_release()     // NOTE: not a syscall: used by _ctx_kill_thread()
{

#if NB_TIM_CHANNELS

    unsigned int channel = _get_context_slot( CTX_TIM_ID );

    if ( channel == -1 )
    {
        unsigned int trdid = _get_thread_trdid();
        _printf("\n[GIET_ERROR] in _sys_tim_release(): "
                "TIMER channel already released for thread %x\n", trdid );

        return SYSCALL_CHANNEL_ALREADY_ALLOCATED;
    }

    // reset CTX_TIM_ID for the calling thread
    _set_context_slot( CTX_TIM_ID , 0xFFFFFFFF );

    // reset the _tim_channel_alloc[] array
    _tim_channel_alloc[channel] = 0;

    // release WTI mailbox if TTY channel no more used
    if ( USE_PIC ) 
    {
        _ext_irq_release( PERIPH_TYPE_TIM , channel );
    }

    return SYSCALL_OK;

#else

    _printf("\n[GIET ERROR] in _sys_tim_release(): NB_TIM_CHANNELS = 0\n");

    return SYSCALL_NO_CHANNEL_AVAILABLE;

#endif
}  // end _sys_tim_release()

/////////////////////////////////////////
int _sys_tim_start( unsigned int period )
{

#if NB_TIM_CHANNELS

    // get timer index
    unsigned int channel = _get_context_slot( CTX_TIM_ID );

    if ( channel >= NB_TIM_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_tim_start(): not enough TIM channels\n");

        return SYSCALL_NO_CHANNEL_AVAILABLE;
    }

    // start timer
    _timer_start( channel, period );

    return SYSCALL_OK;

#else

    _printf("\n[GIET ERROR] in _sys_tim_start() : NB_TIM_CHANNELS = 0\n");

    return SYSCALL_NO_CHANNEL_AVAILABLE;

#endif
}

///////////////////
int _sys_tim_stop()
{

#if NB_TIM_CHANNELS

    // get timer index
    unsigned int channel = _get_context_slot( CTX_TIM_ID );

    if ( channel >= NB_TIM_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_tim_stop() : illegal timer index\n");

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // stop timer
    _timer_stop( channel );

    return SYSCALL_OK;

#else

    _printf("\n[GIET ERROR] in _sys_tim_stop() : NB_TIM_CHANNELS = 0\n");

    return SYSCALL_NO_CHANNEL_AVAILABLE;

#endif
}


//////////////////////////////////////////////////////////////////////////////
//             NIC related syscall handlers 
//////////////////////////////////////////////////////////////////////////////

#define NIC_CONTAINER_SIZE 4096

#if NB_NIC_CHANNELS && NB_CMA_CHANNELS

////////////////////////////////////////
int _sys_nic_alloc( unsigned int is_rx,
                    unsigned int xmax,
                    unsigned int ymax )
{
    mapping_header_t  *header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t  *vspace   = _get_vspace_base(header);
    mapping_thread_t  *thread   = _get_thread_base(header);

    // get calling thread trdid, vspace index, and number of threads
    unsigned int trdid = _get_thread_trdid();
    unsigned int vsid  = _get_context_slot( CTX_VSID_ID );
    unsigned int users = vspace[vsid].threads;

    // check xmax / ymax parameters
    if ( (xmax > X_SIZE) || (ymax > Y_SIZE) )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_alloc() "
                "xmax or ymax argument too large for thread %x\n", trdid );

        return SYSCALL_ILLEGAL_ARGUMENT;
    }

    ////////////////////////////////////////////////////////
    // Step 1: get and register CMA and NIC channel index //
    ////////////////////////////////////////////////////////

    unsigned int   nic_channel;
    unsigned int   cma_channel;
    unsigned int*  palloc;

    // get a NIC_RX or NIC_TX channel 
    for ( nic_channel = 0 ; nic_channel < NB_NIC_CHANNELS ; nic_channel++ )
    {
        if ( is_rx ) palloc = &_nic_rx_channel_alloc[nic_channel];
        else         palloc = &_nic_tx_channel_alloc[nic_channel];

        if ( _atomic_test_and_set( palloc , users ) == 0 ) break;
    }
    if ( (nic_channel >= NB_NIC_CHANNELS) )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_alloc() : "
                "no NIC channel available for thread %x\n", trdid );

        return SYSCALL_NO_CHANNEL_AVAILABLE;
    }

    // get a CMA channel 
    for ( cma_channel = 0 ; cma_channel < NB_CMA_CHANNELS ; cma_channel++ )
    {
        palloc = &_cma_channel_alloc[cma_channel];

        if ( _atomic_test_and_set( palloc , users ) == 0 ) break;
    }
    if ( cma_channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_alloc() : "
                "no CMA channel available for thread %x\n", trdid );
        if ( is_rx )  _nic_rx_channel_alloc[nic_channel] = 0;
        else          _nic_tx_channel_alloc[nic_channel] = 0;

        return SYSCALL_NO_CHANNEL_AVAILABLE;
    }

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] sys_nic_alloc() at cycle %d\n"
        "thread %d get nic_channel = %d / cma_channel = %d\n",
        _get_proctime() , trdid , nic_channel , cma_channel );
#endif

    // register nic_index and cma_index in all threads
    // contexts that are in the same vspace
    unsigned int tid;
    for (tid = vspace[vsid].thread_offset;
         tid < (vspace[vsid].thread_offset + vspace[vsid].threads);
         tid++)
    {
        unsigned int y_size        = header->y_size;
        unsigned int cid           = thread[tid].clusterid;
        unsigned int x             = cid / y_size;
        unsigned int y             = cid % y_size;
        unsigned int p             = thread[tid].proclocid;
        unsigned int ltid          = thread[tid].ltid;
        static_scheduler_t* psched = (static_scheduler_t*)_schedulers[x][y][p];

        if ( is_rx )
        {
            if ( (psched->context[ltid].slot[CTX_NIC_RX_ID] < NB_NIC_CHANNELS) ||
                 (psched->context[ltid].slot[CTX_CMA_RX_ID] < NB_CMA_CHANNELS) )
            {
                _printf("\n[GIET_ERROR] in _sys_nic_alloc() : "
                        "NIC_RX or CMA_RX channel already allocated for thread %x\n", trdid );
                _nic_rx_channel_alloc[nic_channel] = 0;
                _cma_channel_alloc[cma_channel]    = 0;

                return SYSCALL_CHANNEL_ALREADY_ALLOCATED;
            }
            else
            {
                psched->context[ltid].slot[CTX_NIC_RX_ID] = nic_channel;
                psched->context[ltid].slot[CTX_CMA_RX_ID] = cma_channel;
            }
        }
        else // is_tx
        {
            if ( (psched->context[ltid].slot[CTX_NIC_TX_ID] < NB_NIC_CHANNELS) ||
                 (psched->context[ltid].slot[CTX_CMA_TX_ID] < NB_CMA_CHANNELS) )
            {
                _printf("\n[GIET_ERROR] in _sys_nic_alloc() : "
                        "NIC_TX or CMA_TX channel already allocated for thread %x\n", trdid );
                _nic_tx_channel_alloc[nic_channel] = 0;
                _cma_channel_alloc[cma_channel]    = 0;

                return SYSCALL_CHANNEL_ALREADY_ALLOCATED;
            }
            else
            {
                psched->context[ltid].slot[CTX_NIC_TX_ID] = nic_channel;
                psched->context[ltid].slot[CTX_CMA_TX_ID] = cma_channel;
            }
        }
    }  // end loop on threads

    /////////////////////////////////////////////////////////////////////////////////
    // Step 2: loop on all the clusters                                            //
    // Allocate the kernel containers and status, compute the container and the    //
    // status physical addresses, fill and synchronize the kernel CHBUF descriptor //
    /////////////////////////////////////////////////////////////////////////////////

    // physical addresses to be registered in the CMA registers
    unsigned long long nic_chbuf_pbase;     // NIC chbuf physical address
    unsigned long long ker_chbuf_pbase;     // kernel chbuf physical address

    // allocate one kernel container and one status variable per cluster in the
    // (xmax / ymax) mesh
    unsigned int        cx;                 // cluster X coordinate
    unsigned int        cy;                 // cluster Y coordinate
    unsigned int        index;              // container index in chbuf
    unsigned int        vaddr;              // virtual address
    unsigned long long  cont_paddr;         // container physical address
    unsigned long long  sts_paddr;          // container status physical address

    unsigned int        flags;              // for _v2p_translate()

    for ( cx = 0 ; cx < xmax ; cx++ )
    {
        for ( cy = 0 ; cy < ymax ; cy++ )
        {
            // compute index in chbuf
            index = (cx * ymax) + cy; 

            // allocate the kernel container
            vaddr = (unsigned int)_remote_malloc( NIC_CONTAINER_SIZE, cx, cy );

            if ( vaddr == 0 )  // not enough kernel heap memory in cluster[cx,cy]
            {
                _printf("\n[GIET_ERROR] in _sys_nic_alloc() : "
                        "not enough kernel heap in cluster[%d,%d]\n", cx, cy );

                return SYSCALL_OUT_OF_KERNEL_HEAP_MEMORY;
            }

            // compute container physical address
            cont_paddr = _v2p_translate( vaddr , &flags );

            // checking container address alignment
            if ( cont_paddr & 0x3F )
            {
                _printf("\n[GIET ERROR] in _sys_nic_alloc() : "
                        "container address in cluster[%d,%d] not aligned\n", cx, cy);

                return SYSCALL_ADDRESS_NON_ALIGNED;
            }

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_alloc() at cycle %d\n"
        "thread %x allocates a container in cluster[%d,%d] / vaddr = %x / paddr = %l\n",
        -get_proctime() , trdid , cx , cy , vaddr, cont_paddr );
#endif

            // allocate the kernel container status
            // it occupies 64 bytes but only last bit is useful (1 for full and 0 for empty)
            vaddr = (unsigned int)_remote_malloc( 64, cx, cy );

            if ( vaddr == 0 )  // not enough kernel heap memory in cluster[cx,cy]
            {
                _printf("\n[GIET_ERROR] in _sys_nic_alloc() : "
                        "not enough kernel heap in cluster[%d,%d]\n", cx, cy );

                return SYSCALL_OUT_OF_KERNEL_HEAP_MEMORY;
            }

            // compute status physical address
            sts_paddr = _v2p_translate( vaddr , &flags );

            // checking status address alignment
            if ( sts_paddr & 0x3F )
            {
                _printf("\n[GIET ERROR] in _sys_nic_alloc() : "
                        "status address in cluster[%d,%d] not aligned\n", cx, cy);

                return SYSCALL_ADDRESS_NON_ALIGNED;
            }

            // initialize chbuf entry
            // The buffer descriptor has the following structure:
            // - the 26 LSB bits contain bits[6:31] of the buffer physical address
            // - the 26 following bits contain bits[6:31] of the physical address where the
            //   buffer status is located
            // - the 12 MSB bits contain the common address extension of the buffer and its
            //   status
            if ( is_rx )
                _nic_ker_rx_chbuf[nic_channel].buf_desc[index] =
                    (unsigned long long)
                    ((sts_paddr & 0xFFFFFFFFULL) >> 6) +
                    (((cont_paddr & 0xFFFFFFFFFFFULL) >> 6) << 26);
            else
                _nic_ker_tx_chbuf[nic_channel].buf_desc[index] =
                    (unsigned long long)
                    ((sts_paddr & 0xFFFFFFC0ULL) >> 6) +
                    (((cont_paddr & 0xFFFFFFFFFC0ULL) >> 6) << 26);

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_alloc() at cycle %d\n"
        "thread %x allocates a status in cluster[%d,%d] / vaddr = %x / paddr = %l\n"
        "   descriptor = %l\n",
        _get_proctime() , trdid , cx , cy , vaddr, sts_paddr,
        (unsigned long long)((sts_paddr & 0xFFFFFFFFULL) >> 6) + 
        (((cont_paddr & 0xFFFFFFFFFFFULL) >> 6) << 26) );
#endif
        }
    }

    // complete kernel chbuf initialisation
    if ( is_rx )
    {
        _nic_ker_rx_chbuf[nic_channel].xmax = xmax;
        _nic_ker_rx_chbuf[nic_channel].ymax = ymax;
    }
    else
    {
        _nic_ker_tx_chbuf[nic_channel].xmax = xmax;
        _nic_ker_tx_chbuf[nic_channel].ymax = ymax;
    }

    // compute the kernel chbuf descriptor physical address
    if ( is_rx ) vaddr = (unsigned int)( &_nic_ker_rx_chbuf[nic_channel] );
    else         vaddr = (unsigned int)( &_nic_ker_tx_chbuf[nic_channel] );

    ker_chbuf_pbase = _v2p_translate( vaddr , &flags );

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_alloc() at cycle %d\n"
        "thread %x initialise kernel chbuf / vaddr = %x / paddr = %l\n",
        _get_proctime() , trdid , vaddr , ker_chbuf_pbase );
#endif

    // sync the kernel chbuf in L2 after write in L2
    _mmc_sync( ker_chbuf_pbase, sizeof( nic_chbuf_t ) );

    ///////////////////////////////////////////////////////////////
    // Step 3: compute the NIC chbuf descriptor physical address //
    ///////////////////////////////////////////////////////////////

    unsigned int offset;
    if ( is_rx ) offset = 0x4100;
    else         offset = 0x4110;
    nic_chbuf_pbase = (((unsigned long long)((X_IO << Y_WIDTH) + Y_IO))<<32) |
                      (SEG_NIC_BASE + (nic_channel<<15) + offset);

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_alloc() at cycle %d\n"
        "thread %x get NIC chbuf paddr = %l\n",
        _get_proctime() , trdid , nic_chbuf_pbase );
#endif

    ////////////////////////////////////////////////////////////////////////////////
    // Step 4: initialize CMA registers defining the source & destination chbufs //
    ////////////////////////////////////////////////////////////////////////////////

    if ( is_rx )               // NIC to kernel
    {
        _cma_set_register( cma_channel, CHBUF_SRC_DESC , (unsigned int)(nic_chbuf_pbase) );
        _cma_set_register( cma_channel, CHBUF_SRC_EXT  , (unsigned int)(nic_chbuf_pbase>>32) );
        _cma_set_register( cma_channel, CHBUF_SRC_NBUFS, 2 );
        _cma_set_register( cma_channel, CHBUF_DST_DESC , (unsigned int)(ker_chbuf_pbase) );
        _cma_set_register( cma_channel, CHBUF_DST_EXT  , (unsigned int)(ker_chbuf_pbase>>32) );
        _cma_set_register( cma_channel, CHBUF_DST_NBUFS, xmax * ymax );
    }
    else                      // kernel to NIC
    {
        _cma_set_register( cma_channel, CHBUF_SRC_DESC , (unsigned int)(ker_chbuf_pbase) );
        _cma_set_register( cma_channel, CHBUF_SRC_EXT  , (unsigned int)(ker_chbuf_pbase>>32) );
        _cma_set_register( cma_channel, CHBUF_SRC_NBUFS, xmax * ymax );
        _cma_set_register( cma_channel, CHBUF_DST_DESC , (unsigned int)(nic_chbuf_pbase) );
        _cma_set_register( cma_channel, CHBUF_DST_EXT  , (unsigned int)(nic_chbuf_pbase>>32) );
        _cma_set_register( cma_channel, CHBUF_DST_NBUFS, 2 );
    }

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_alloc() at cycle %d\n"
        "thread %x exit\n", 
        _get_proctime() , trdid );
#endif

    return SYSCALL_OK;
} // end _sys_nic_alloc()


//////////////////////////////////////////
int _sys_nic_release( unsigned int is_rx )     // NOTE: not a syscall: used by _ctx_kill_thread()
{
    unsigned int trdid = _get_thread_trdid();

    unsigned int nic_channel;
    unsigned int cma_channel;
    
    // update the kernel tables 
    if ( is_rx )
    {
        nic_channel = _get_context_slot( CTX_NIC_RX_ID );
        cma_channel = _get_context_slot( CTX_CMA_RX_ID );

        if ( (nic_channel >= NB_NIC_CHANNELS) )
        {
            _printf("\n[GIET ERROR] in _sys_nic_release() : "
                    "NIC_RX channel already released for thread %x\n", trdid );

            return SYSCALL_CHANNEL_NON_ALLOCATED;
        }
        if ( (cma_channel >= NB_CMA_CHANNELS) )
        {
            _printf("\n[GIET ERROR] in _sys_nic_release() : "
                    "CMA_RX channel already released for thread %x\n", trdid );

            return SYSCALL_CHANNEL_NON_ALLOCATED;
        }

        // atomically decrement the NIC and CMA channel allocators
        _atomic_increment( &_nic_rx_channel_alloc[nic_channel] , -1 );
        _atomic_increment( &_cma_channel_alloc[cma_channel] , -1 );
    
        // stop the NIC and CMA peripherals channels if no more users
        if ( (_nic_rx_channel_alloc[nic_channel] == 0) &&
             (_cma_channel_alloc[cma_channel] == 0) )  _sys_nic_stop( 1 );

        // reset the calling thread context slots
        _set_context_slot( CTX_NIC_RX_ID , 0xFFFFFFFF );
        _set_context_slot( CTX_CMA_RX_ID , 0xFFFFFFFF );
    }         
    else
    {
        nic_channel = _get_context_slot( CTX_NIC_TX_ID );
        cma_channel = _get_context_slot( CTX_CMA_TX_ID );

        if ( (nic_channel >= NB_NIC_CHANNELS) )
        {
            _printf("\n[GIET ERROR] in _sys_nic_release() : "
                    "NIC_TX channel already released for thread %x\n", trdid );

            return SYSCALL_CHANNEL_NON_ALLOCATED;
        }
        if ( (cma_channel >= NB_CMA_CHANNELS) )
        {
            _printf("\n[GIET ERROR] in _sys_nic_release() : "
                    "CMA_TX channel already released for thread %x\n", trdid );

            return SYSCALL_CHANNEL_NON_ALLOCATED;
        }

        // atomically decrement the NIC and CMA channel allocators
        _atomic_increment( &_nic_tx_channel_alloc[nic_channel] , -1 );
        _atomic_increment( &_cma_channel_alloc[cma_channel] , -1 );
    
        // stop the NIC and CMA peripherals channels if no more users
        if ( (_nic_tx_channel_alloc[nic_channel] == 0) &&
             (_cma_channel_alloc[cma_channel] == 0) )  _sys_nic_stop( 0 );

        // reset the calling thread context slots
        _set_context_slot( CTX_NIC_TX_ID , 0xFFFFFFFF );
        _set_context_slot( CTX_CMA_TX_ID , 0xFFFFFFFF );
    }

    return SYSCALL_OK;
}  // end sys_nic_release()


////////////////////////////////////////
int _sys_nic_start( unsigned int is_rx )
{
    unsigned int trdid  = _get_context_slot( CTX_TRDID_ID );

    unsigned int nic_channel;
    unsigned int cma_channel;

    // get NIC channel index and CMA channel index from thread context
    if ( is_rx )
    {
        nic_channel = _get_context_slot( CTX_NIC_RX_ID );
        cma_channel = _get_context_slot( CTX_CMA_RX_ID );
    }
    else
    {
        nic_channel = _get_context_slot( CTX_NIC_TX_ID );
        cma_channel = _get_context_slot( CTX_CMA_TX_ID );
    }

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_start() at cycle %d\n"
        "thread %x enter / NIC channel = %d / CMA channel = %d\n",
        _get_proctime() , trdid , nic_channel, cma_channel );
#endif

    // check NIC and CMA channels index
    if ( nic_channel >= NB_NIC_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_start() : "
                "illegal NIC channel for thread %x\n", trdid );
        return -1111;
    }
    if ( cma_channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_start() : "
                "illegal CMA channel for thread %x\n", trdid );
        return -1111;
    }

    // start CMA transfer
    _cma_set_register( cma_channel, CHBUF_BUF_SIZE , NIC_CONTAINER_SIZE );
    _cma_set_register( cma_channel, CHBUF_PERIOD   , 0 );                  // OUT_OF_ORDER 
    _cma_set_register( cma_channel, CHBUF_RUN      , MODE_NORMAL );

    // activates NIC channel
    _nic_channel_start( nic_channel, is_rx, GIET_NIC_MAC4, GIET_NIC_MAC2 );

#if GIET_DEBUG_NIC
    _printf("\n[DEBUG NIC] _sys_nic_start() at cycle %d\n"
            "thread %d exit\n",
            _get_proctime() , trdid );
#endif

    return SYSCALL_OK;
}  // end _sys_nic_start()


//////////////////////////////////////
int _sys_nic_move( unsigned int is_rx,
                   void*        buffer )
{
    unsigned int trdid  = _get_context_slot( CTX_TRDID_ID );

    unsigned int channel;

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n",
        "thread %x enters\n",
        _get_proctime() , trdid );
#endif

    // get NIC channel index from thread context
    if ( is_rx )  channel = _get_context_slot( CTX_NIC_RX_ID );
    else          channel = _get_context_slot( CTX_NIC_TX_ID );

    // check NIC channel index
    if ( channel >= NB_NIC_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_move() : "
                "NIC channel non allocated for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // get kernel chbuf virtual address
    nic_chbuf_t* ker_chbuf;
    if ( is_rx )  ker_chbuf = &_nic_ker_rx_chbuf[channel];
    else          ker_chbuf = &_nic_ker_tx_chbuf[channel];

    // get xmax / ymax parameters
    unsigned int xmax = ker_chbuf->xmax;
    unsigned int ymax = ker_chbuf->ymax;

    // get cluster coordinates for the processor running the calling thread
    unsigned int  procid = _get_procid();
    unsigned int  cx     = procid >> (Y_WIDTH + P_WIDTH);
    unsigned int  cy     = (procid >> P_WIDTH) & ((1<<Y_WIDTH)-1);
    
    // check processor coordinates / (xmax,ymax)
    if ( (cx >= xmax) || (cy >= ymax) )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_move(): "
         "processor coordinates [%d,%d] larger than (xmax,ymax) = [%d,%d]\n",
         cx , cy , xmax , ymax );

        return SYSCALL_ILLEGAL_ARGUMENT;
    }
    
    unsigned long long usr_buf_paddr;       // user buffer physical address 
    unsigned long long ker_buf_paddr;       // kernel buffer physical address
    unsigned long long ker_sts_paddr;       // kernel buffer status physical address
    unsigned long long ker_buf_desc;        // kernel buffer descriptor
    unsigned int       ker_sts;             // kernel buffer status (full or empty)
    unsigned int       index;               // kernel buffer index in chbuf
    unsigned int       flags;               // for _v2P_translate

    // Compute user buffer physical address and check access rights
    usr_buf_paddr = _v2p_translate( (unsigned int)buffer , &flags );

    if ( (flags & PTE_U) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_nic_tx_move() : "
                "buffer address non user accessible\n");

        return SYSCALL_ADDRESS_NON_USER_ACCESSIBLE;
    }

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n"
        "thread %x get user buffer : paddr = %l\n",
        _get_proctime() , trdid , usr_buf_paddr );
#endif

    // compute buffer index, buffer descriptor paddr and buffer status paddr
    index = (ymax * cx) + cy;
    ker_buf_desc = ker_chbuf->buf_desc[index];
    ker_sts_paddr = ((ker_buf_desc & 0xFFF0000000000000ULL) >> 20) +
                    ((ker_buf_desc & 0x3FFFFFFULL) << 6);

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n"
        "thread %x get ker_buf_desc %d / paddr = %l\n",
        _get_proctime(), trdid , index , ker_buf_desc );
#endif

    // poll local kernel container status until success
    while ( 1 )
    {
        // inval buffer descriptor in L2 before read in L2
        _mmc_inval( ker_sts_paddr, 4 );
        ker_sts = _physical_read( ker_sts_paddr );

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n"
        "thread %x get status %d /  paddr = %l / status = %x\n",
        _get_proctime() , trdid , index , ker_sts_paddr, ker_sts );
#endif

        // test buffer status and break if found
        if ( ( is_rx != 0 ) && ( ker_sts == 0x1 ) ) break;
        if ( ( is_rx == 0 ) && ( ker_sts == 0 ) ) break;
    }

    // compute kernel buffer physical address
    ker_buf_paddr = (ker_buf_desc & 0xFFFFFFFFFC000000ULL) >> 20;
    
    // move one container
    if ( is_rx )              // RX transfer
    {
        // inval kernel buffer in L2 before read in L2
        _mmc_inval( ker_buf_paddr, NIC_CONTAINER_SIZE );

        // transfer data from kernel buffer to user buffer
        _physical_memcpy( usr_buf_paddr, 
                          ker_buf_paddr, 
                          NIC_CONTAINER_SIZE );
#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n"
        "thread %x transfer kernel buffer %l to user buffer %l\n",
        _get_proctime() , trdid , ker_buf_paddr , usr_buf_paddr );
#endif

    }
    else                      // TX transfer
    {
        // transfer data from user buffer to kernel buffer
        _physical_memcpy( ker_buf_paddr, 
                          usr_buf_paddr, 
                          NIC_CONTAINER_SIZE );

        // sync kernel buffer in L2 after write in L2
        _mmc_sync( ker_buf_paddr, NIC_CONTAINER_SIZE );

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n"
        "thread %x transfer user buffer %l to kernel buffer %l\n",
        _get_proctime() , trdid , usr_buf_paddr , ker_buf_paddr );
#endif

    }

    // update kernel chbuf status 
    if ( is_rx ) _physical_write ( ker_sts_paddr, 0 );
    else         _physical_write ( ker_sts_paddr, 0x1 );

    // sync kernel chbuf in L2 after write in L2
    _mmc_sync( ker_sts_paddr, 4 );

#if GIET_DEBUG_NIC
_printf("\n[DEBUG NIC] _sys_nic_move() at cycle %d\n"
        "thread %x exit\n",
        _get_proctime() , trdid );
#endif

    return SYSCALL_OK;
} // end _sys_nic_move()


///////////////////////////////////////
int _sys_nic_stop( unsigned int is_rx )
{
    unsigned int trdid  = _get_context_slot( CTX_TRDID_ID );

    unsigned int nic_channel;
    unsigned int cma_channel;

    // get NIC channel index and CMA channel index
    if ( is_rx )
    {
        nic_channel = _get_context_slot( CTX_NIC_RX_ID );
        cma_channel = _get_context_slot( CTX_CMA_RX_ID );
    }
    else
    {
        nic_channel = _get_context_slot( CTX_NIC_TX_ID );
        cma_channel = _get_context_slot( CTX_CMA_TX_ID );
    }

    // check NIC and CMA channels index
    if ( nic_channel >= NB_NIC_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_stop() : " 
                "NIC channel non allocated for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }
    if ( cma_channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_stop() : "
                "CMA channel non allocated for thread %x\n", trdid );
 
        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // desactivates the CMA channel
    _cma_set_register( cma_channel, CHBUF_RUN , MODE_IDLE );

    // wait until CMA channel IDLE
    unsigned int volatile status;
    do
    {
         status = _cma_get_register( cma_channel, CHBUF_STATUS );
    } 
    while ( status ); 

    // desactivates the NIC channel
    _nic_channel_stop( nic_channel, is_rx );

    return SYSCALL_OK;
}  // end _sys_nic_stop()

////////////////////////////////////////
int _sys_nic_clear( unsigned int is_rx )
{
    unsigned int trdid  = _get_context_slot( CTX_TRDID_ID );

    unsigned int channel;

    // get NIC channel 
    if ( is_rx )  channel = _get_context_slot( CTX_NIC_RX_ID );
    else          channel = _get_context_slot( CTX_NIC_TX_ID );

    if ( channel >= NB_NIC_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_clear() : "
                "NIC channel non allocated for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    if ( is_rx )
    {
        _nic_set_global_register( NIC_G_NPKT_RX_G2S_RECEIVED       , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DES_TOO_SMALL      , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DES_TOO_BIG        , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DES_MFIFO_FULL     , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DES_CRC_FAIL       , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DISPATCH_RECEIVED  , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DISPATCH_BROADCAST , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DISPATCH_DST_FAIL  , 0 );
        _nic_set_global_register( NIC_G_NPKT_RX_DISPATCH_CH_FULL   , 0 );
    } 
    else
    {
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_RECEIVED  , 0 );
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_TRANSMIT  , 0 );
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_TOO_BIG   , 0 );
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_TOO_SMALL , 0 );
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_SRC_FAIL  , 0 );
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_BYPASS    , 0 );
        _nic_set_global_register( NIC_G_NPKT_TX_DISPATCH_BROADCAST , 0 );
    }
    return SYSCALL_OK;
}  // en _sys_nic_clear()

////////////////////////////////////////
int _sys_nic_stats( unsigned int is_rx )
{
    unsigned int trdid  = _get_context_slot( CTX_TRDID_ID );

    unsigned int nic_channel;

    // get NIC channel 
    if ( is_rx )  nic_channel = _get_context_slot( CTX_NIC_RX_ID );
    else          nic_channel = _get_context_slot( CTX_NIC_TX_ID );

    if ( nic_channel >= NB_NIC_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_nic_stats() : "
                "NIC channel non allocated for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    if ( is_rx )
    {
        unsigned int received   = _nic_get_global_register( NIC_G_NPKT_RX_G2S_RECEIVED       );
        unsigned int too_small  = _nic_get_global_register( NIC_G_NPKT_RX_DES_TOO_SMALL      );
        unsigned int too_big    = _nic_get_global_register( NIC_G_NPKT_RX_DES_TOO_BIG        );
        unsigned int fifo_full  = _nic_get_global_register( NIC_G_NPKT_RX_DES_MFIFO_FULL     );
        unsigned int crc_fail   = _nic_get_global_register( NIC_G_NPKT_RX_DES_CRC_FAIL       );
        unsigned int broadcast  = _nic_get_global_register( NIC_G_NPKT_RX_DISPATCH_BROADCAST );
        unsigned int dst_fail   = _nic_get_global_register( NIC_G_NPKT_RX_DISPATCH_DST_FAIL  );
        unsigned int ch_full    = _nic_get_global_register( NIC_G_NPKT_RX_DISPATCH_CH_FULL   );

        _printf("\n### Network Controller RX Statistics ###\n"
                "- packets received : %d\n"
                "- too small        : %d\n"
                "- too big          : %d\n"
                "- fifo full        : %d\n" 
                "- crc fail         : %d\n" 
                "- dst mac fail     : %d\n" 
                "- channel full     : %d\n" 
                "- broadcast        : %d\n",
                received,
                too_small,
                too_big,
                fifo_full,
                crc_fail,
                dst_fail,
                ch_full,
                broadcast );
    } 
    else
    {
        unsigned int received   = _nic_get_global_register( NIC_G_NPKT_TX_DISPATCH_RECEIVED  );
        unsigned int too_big    = _nic_get_global_register( NIC_G_NPKT_TX_DISPATCH_TOO_BIG   );
        unsigned int too_small  = _nic_get_global_register( NIC_G_NPKT_TX_DISPATCH_TOO_SMALL );
        unsigned int src_fail   = _nic_get_global_register( NIC_G_NPKT_TX_DISPATCH_SRC_FAIL  );
        unsigned int bypass     = _nic_get_global_register( NIC_G_NPKT_TX_DISPATCH_BYPASS    );
        unsigned int broadcast  = _nic_get_global_register( NIC_G_NPKT_TX_DISPATCH_BROADCAST );

        _printf("\n### Network Controller TX Statistics ###\n"
                "- packets received : %d\n"
                "- too small        : %d\n"
                "- too big          : %d\n"
                "- src mac fail     : %d\n" 
                "- bypass           : %d\n" 
                "- broadcast        : %d\n",
                received,
                too_big,
                too_small,
                src_fail,
                bypass,
                broadcast );
    }
    return SYSCALL_OK;
}  // end _sys_nic_stats()

#endif // if NB_NIC_CHANNELS && NB_CMA_CHANNELS

/////////////////////////////////////////////////////////////////////////////////////////
//    FBF related syscall handlers
/////////////////////////////////////////////////////////////////////////////////////////

//////////////////////////////////////
int _sys_fbf_size( unsigned int* width,
                   unsigned int* height )
{
    if ( USE_FBF == 0 )
    {
        *width  = 0;
        *height = 0;
    }
    else
    {
        *width  = FBUF_X_SIZE;
        *height = FBUF_Y_SIZE;
    }

    return SYSCALL_OK;
}

////////////////////
int _sys_fbf_alloc()
{
    mapping_header_t  *header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t  *vspace   = _get_vspace_base(header);
    mapping_thread_t  *thread   = _get_thread_base(header);
    
    // compute number of users
    unsigned int   vsid  = _get_context_slot(CTX_VSID_ID);
    unsigned int   users = vspace[vsid].threads;

    // access FBF allocator 
    // register it in all threads contexts
    if ( _atomic_test_and_set( &_fbf_alloc , users ) == 0 )     // FBF available   
    {
        unsigned int   min   = vspace[vsid].thread_offset;
        unsigned int   max   = min + users;
        unsigned int   tid;
        for ( tid = min ; tid < max ; tid++ )
        {
            unsigned int y_size        = header->y_size;
            unsigned int cid           = thread[tid].clusterid;
            unsigned int x             = cid / y_size;
            unsigned int y             = cid % y_size;
            unsigned int p             = thread[tid].proclocid;
            unsigned int ltid          = thread[tid].ltid;
            static_scheduler_t* psched = (static_scheduler_t*)_schedulers[x][y][p];
            _atomic_or( &psched->context[ltid].slot[CTX_LOCKS_ID] , LOCKS_MASK_FBF ); 
        }
        return SYSCALL_OK;
    }
    else                                                       // FBF already allocated
    {
        return SYSCALL_SHARED_PERIPHERAL_BUSY;
    }
}

//////////////////////
int _sys_fbf_release()    // not a syscall: used by _ctx_kill_thread()
{
    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_release() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // decrement FBF allocator 
    // reset the calling thread context
    _atomic_increment( &_fbf_alloc , 0xFFFFFFFF );
    _atomic_and( &psched->context[ltid].slot[CTX_LOCKS_ID] , ~LOCKS_MASK_FBF ); 

    return SYSCALL_OK;   
}

/////////////////////////////////////////////
int _sys_fbf_sync_write( unsigned int offset,
                         void*        buffer,
                         unsigned int length )
{
    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_release() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    char* fbf_address = (char *)SEG_FBF_BASE + offset;
    memcpy( fbf_address, buffer, length);

    return SYSCALL_OK;
}

/////////////////////////////////////////////
int _sys_fbf_sync_read(  unsigned int offset,
                         void*        buffer,
                         unsigned int length )
{
    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_release() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    char* fbf_address = (char *)SEG_FBF_BASE + offset;
    memcpy( buffer, fbf_address, length);

    return SYSCALL_OK;
}

#if NB_CMA_CHANNELS

////////////////////////////////////////////
int _sys_fbf_cma_alloc( unsigned int nbufs )
{
    // compute trdid and vsid for the calling thread
    unsigned int vsid  = _get_context_slot( CTX_VSID_ID );
    unsigned int trdid = _get_thread_trdid();

    if ( _get_context_slot( CTX_CMA_FB_ID ) < NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_alloc() : "
                "CMA channel already allocated for thread %x\n", trdid );
        return SYSCALL_CHANNEL_ALREADY_ALLOCATED;
    }

    // compute number of threads in vspace from mapping
    mapping_header_t  *header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t  *vspace   = _get_vspace_base(header);
    mapping_thread_t  *thread   = _get_thread_base(header);
    unsigned int      first     = vspace[vsid].thread_offset;
    unsigned int      threads   = vspace[vsid].threads;

    // get a CMA channel
    unsigned int channel;
    for ( channel = 0 ; channel < NB_CMA_CHANNELS ; channel++ )
    {
        unsigned int*  palloc = &_cma_channel_alloc[channel];
        if ( _atomic_test_and_set( palloc , threads ) == 0 ) break;
    }

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_alloc() : no CMA channel available\n");
        return SYSCALL_NO_CHANNEL_AVAILABLE;
    }

    // check nbufs argument
    if ( nbufs > 256 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_alloc() : nbufs larger than 256\n");
        return SYSCALL_ILLEGAL_ARGUMENT;
    }

    // loop on all threads to register channel in thread contexts
    unsigned int      tid;
    for ( tid = first ; tid < (first + threads) ; tid++ )
    {
        unsigned int         y_size = header->y_size;
        unsigned int         cid    = thread[tid].clusterid;
        unsigned int         x      = cid / y_size;
        unsigned int         y      = cid % y_size;
        unsigned int         p      = thread[tid].proclocid;
        unsigned int         ltid   = thread[tid].ltid;
        static_scheduler_t* psched  = (static_scheduler_t*)_schedulers[x][y][p];
        psched->context[ltid].slot[CTX_CMA_FB_ID] = channel; 
    }

    unsigned int vaddr;
    unsigned int flags;

    // compute frame buffer physical addresses
    vaddr = (unsigned int)SEG_FBF_BASE;
    unsigned long long fbf_buf_paddr = _v2p_translate( vaddr , &flags );

    // initialize the FBF chbuf 
    // we don't register a status address in the fbf_desc, because
    // the CMA does not test the status for the frame buffer (no synchro) 
    _fbf_ker_chbuf.nbufs    = nbufs;
    _fbf_ker_chbuf.fbf_desc = (((fbf_buf_paddr & 0xFFFFFFFFFFFULL) >> 6 ) << 26);

    // register FBF chbuf physical address
    vaddr = (unsigned int)(&_fbf_ker_chbuf);
    _fbf_chbuf_paddr = _v2p_translate( vaddr , &flags );

#if GIET_DEBUG_FBF_CMA
_printf("\n[FBF_CMA DEBUG] _sys_fbf_cma_alloc()\n"
        " - channel               = %d\n"
        " - vaddr(_ker_fbf_chbuf) = %x\n"
        " - paddr(_ker_fbf_chbuf) = %l\n"
        " - nbufs                 = %d\n" 
        " - fbf_desc              = %l\n",
        channel , vaddr , _fbf_chbuf_paddr , nbufs , _fbf_ker_chbuf.fbf_desc );
#endif

    return SYSCALL_OK;
} // end sys_fbf_cma_alloc()

//////////////////////////
int _sys_fbf_cma_release()  // Not a syscall : used by _ctx_kill_thread()
{
    unsigned int channel = _get_context_slot( CTX_CMA_FB_ID );
    unsigned int trdid   = _get_thread_trdid();

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET_ERROR] in _sys_fbf_cma_release() : "
                "CMA_FB channel already released for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    if ( _cma_channel_alloc[channel] == 1 )  // the calling thread is the last user
    {
        // stop the CMA transfer
        _sys_fbf_cma_stop();

        // reset the CMA channel allocator
        _cma_channel_alloc[channel] = 0;
    }
    else                                     // not the last user
    {
        // atomically decrement the CMA channel allocator
        _atomic_increment( &_cma_channel_alloc[channel] , -1 );
    }

    // reset CTX_CMA_FB_ID slot in calling thread context
    _set_context_slot( CTX_CMA_FB_ID, 0xFFFFFFFF );

    return SYSCALL_OK;
} // end _sys_fbf_cma_release()

///////////////////////////////////////////////////
int _sys_fbf_cma_init_buf( unsigned int index,
                           void*        buf_vaddr, 
                           void*        sts_vaddr )
{
    unsigned int       vaddr;           // virtual address
    unsigned int       flags;           // for _v2p_translate()
    unsigned long long buf_paddr;       // user buffer physical address
    unsigned long long sts_paddr;       // user status physical address

    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    // check FBF allocated
    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // get channel index
    unsigned int channel = _get_context_slot( CTX_CMA_FB_ID );

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "CMA channel non allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

#if GIET_DEBUG_FBF_CMA
_printf("\n[FBF_CMA DEBUG] _sys_fbf_cma_init_buf()\n"
        " - channel     = %d / index = %d\n"
        " - buf vaddr   = %x\n"
        " - sts vaddr   = %x\n",
        channel, index,
        (unsigned int)buf_vaddr,
        (unsigned int)sts_vaddr );
#endif

    // checking index argument
    if ( index >= _fbf_ker_chbuf.nbufs )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "user buffer index too large %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // checking user buffer and status addresses alignment
    if ( ((unsigned int)buf_vaddr & 0x3F) || ((unsigned int)sts_vaddr & 0x3F) )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "user buffer or status not aligned for thread %x\n", trdid );

        return SYSCALL_ADDRESS_NON_ALIGNED;
    }

    // Compute user buffer and status physical addresses 
    vaddr = (unsigned int)buf_vaddr;
    buf_paddr = _v2p_translate( vaddr , &flags );
    if ((flags & PTE_U) == 0) 
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "buffer not in user space for thread %x\n", trdid );

        return SYSCALL_ADDRESS_NON_USER_ACCESSIBLE;
    }

    vaddr = (unsigned int)sts_vaddr;
    sts_paddr = _v2p_translate( vaddr , &flags );
    if ((flags & PTE_U) == 0) 
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "status not in user space for thread %x\n", trdid);

        return SYSCALL_ADDRESS_NON_USER_ACCESSIBLE;
    }

    // check user buffer and user status in same cluster
    if ( (buf_paddr & 0xFF00000000ULL) != (sts_paddr & 0xFF00000000ULL) ) 
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_init_buf() : "
                "user status and buffer not in same cluster for thread %x\n", trdid);

        return SYSCALL_ADDRESS_NON_USER_ACCESSIBLE;
    }

    // initialize _fbf_ker_chbuf.usr_desc[index]
    _fbf_ker_chbuf.usr_desc[index] = ((sts_paddr & 0xFFFFFFFFULL) >> 6) +
                                     (((buf_paddr & 0xFFFFFFFFFFULL) >> 6 ) << 26);

#if GIET_DEBUG_FBF_CMA
_printf(" - buf paddr   = %l\n"
        " - sts paddr   = %l\n"
        " - usr_desc[%d] = %l\n",
        buf_paddr,
        sts_paddr,
        index , _fbf_ker_chbuf.usr_desc[index] );
#endif

    return SYSCALL_OK;

} // end sys_fbf_cma_init_buf()

////////////////////////
int _sys_fbf_cma_start() 
{
    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    // check FBF allocated
    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_release() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // get CMA channel index
    unsigned int channel = _get_context_slot( CTX_CMA_FB_ID );

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _fbf_cma_start() : "
                "CMA channel non allocated\n");

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // check buffers initialization
    if ( _fbf_ker_chbuf.nbufs == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_start(): "
                "FBF chbuf not initialized for thread %x\n", trdid );

        return SYSCALL_MISSING_INITIALISATION;
    }

    // synchronize FBF chbuf that will be read by CMA peripheral
    if ( USE_IOB )
    {
        // SYNC request for fbf_chbuf descriptor
        _mmc_sync( _fbf_chbuf_paddr , sizeof( fbf_chbuf_t ) );
    }

    // start CMA transfer
    unsigned long long paddr = _fbf_chbuf_paddr;
    unsigned int dst_chbuf_paddr_lsb = (unsigned int)(paddr & 0xFFFFFFFF);
    unsigned int dst_chbuf_paddr_ext = (unsigned int)(paddr >> 32);
    unsigned int src_chbuf_paddr_lsb = dst_chbuf_paddr_lsb + 8;
    unsigned int src_chbuf_paddr_ext = dst_chbuf_paddr_ext;

#if GIET_DEBUG_FBF_CMA
_printf("\n[FBF_CMA DEBUG] _sys_fbf_cma_start()\n"
        " - src_chbuf_paddr_lsb = %x\n"
        " - src_chbuf_paddr_ext = %x\n"
        " - src_chbuf_nbufs     = %d\n"
        " - dst_chbuf_paddr_lsb = %x\n"
        " - dst_chbuf_paddr_ext = %x\n"
        " - dst_chbuf_nbufs     = 1 \n"
        " - buffer_size         = %d\n",
        src_chbuf_paddr_lsb,
        src_chbuf_paddr_ext,
        _fbf_ker_chbuf.nbufs,
        dst_chbuf_paddr_lsb,
        dst_chbuf_paddr_ext,
        FBUF_X_SIZE * FBUF_Y_SIZE );
#endif

    _cma_set_register( channel, CHBUF_SRC_DESC , src_chbuf_paddr_lsb );
    _cma_set_register( channel, CHBUF_SRC_EXT  , src_chbuf_paddr_ext );
    _cma_set_register( channel, CHBUF_SRC_NBUFS, _fbf_ker_chbuf.nbufs );
    _cma_set_register( channel, CHBUF_DST_DESC , dst_chbuf_paddr_lsb );
    _cma_set_register( channel, CHBUF_DST_EXT  , dst_chbuf_paddr_ext );
    _cma_set_register( channel, CHBUF_DST_NBUFS, 1 );
    _cma_set_register( channel, CHBUF_BUF_SIZE , FBUF_X_SIZE*FBUF_Y_SIZE );
    _cma_set_register( channel, CHBUF_PERIOD   , 1000 );
    _cma_set_register( channel, CHBUF_RUN      , MODE_NO_DST_SYNC );

    return SYSCALL_OK;

} // end _sys_fbf_cma_start()

////////////////////////////////////////////
int _sys_fbf_cma_check( unsigned int index )
{
    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    // check FBF allocated
    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_check() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // get channel index
    unsigned int channel = _get_context_slot( CTX_CMA_FB_ID );

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_check() : "
                "CMA channel non allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // check buffer index
    if ( index >= _fbf_ker_chbuf.nbufs )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_check() : "
                "buffer index too large for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // compute user buffer status physical addresses
    unsigned long long usr_sts_paddr;
    fbf_chbuf_t* pdesc = &_fbf_ker_chbuf;     
    usr_sts_paddr = ((pdesc->usr_desc[index] & 0xFFF0000000000000ULL) >> 20) +  
                    ((pdesc->usr_desc[index] & 0x3FFFFFFULL) << 6);          

#if GIET_DEBUG_FBF_CMA
_printf("\n[FBF_CMA DEBUG] enters _sys_fbf_cma_check()\n"
        " - cma channel      = %d\n"
        " - buffer index     = %d\n"
        " - usr_desc value   = %l\n"
        " - fbf_desc value   = %l\n"
        " - usr status paddr = %l\n",
        channel,
        index,
        _fbf_ker_chbuf.usr_desc[index],
        _fbf_ker_chbuf.fbf_desc,
        usr_sts_paddr );
#endif

    // waiting user buffer released by the CMA component)
    unsigned int full;
    do
    {  
        // INVAL L2 cache copy of user buffer status     
        // because it is modified in RAM by the CMA component 
        _mmc_inval( usr_sts_paddr , 4 );        

        full = _physical_read( usr_sts_paddr );
    }
    while ( full );

    return SYSCALL_OK;

}  // end _sys_fbf_cma_check()

//////////////////////////////////////////////
int _sys_fbf_cma_display( unsigned int index )
{
    // get calling thread scheduler, ltid and trdid
    static_scheduler_t*  psched = _get_sched();
    unsigned int         ltid   = _get_thread_ltid();
    unsigned int         trdid  = _get_thread_trdid();

    // check FBF allocated
    if ( (psched->context[ltid].slot[CTX_LOCKS_ID] & LOCKS_MASK_FBF) == 0 )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_display() : "
                "FBF not allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // get channel index
    unsigned int channel = _get_context_slot( CTX_CMA_FB_ID );

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_display() : "
                "CMA channel non allocated to thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // check buffer index
    if ( index >= _fbf_ker_chbuf.nbufs )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_display() : "
                "buffer index too large for thread %x\n", trdid );

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // compute user buffer and status physical addresses
    unsigned long long usr_sts_paddr;
    unsigned long long usr_buf_paddr;

    fbf_chbuf_t* pdesc = &_fbf_ker_chbuf;     

    usr_sts_paddr = ((pdesc->usr_desc[index] & 0xFFF0000000000000ULL) >> 20) +  
                    ((pdesc->usr_desc[index] & 0x3FFFFFFULL) << 6);          

    usr_buf_paddr = ((pdesc->usr_desc[index] & 0xFFFFFFFFFC000000ULL) >> 20); 

#if GIET_DEBUG_FBF_CMA
_printf("\n[FBF_CMA DEBUG] enters _sys_fbf_cma_display()\n"
        " - cma channel      = %d\n"
        " - buffer index     = %d\n"
        " - usr buffer paddr = %l\n"
        " - usr status paddr = %l\n",
        channel,
        index,
        usr_buf_paddr, 
        usr_sts_paddr ); 
#endif
        
    // SYNC request, because this buffer will be read from XRAM by the CMA component
    _mmc_sync( usr_buf_paddr , FBUF_X_SIZE * FBUF_Y_SIZE );

    // set user buffer status
    _physical_write( usr_sts_paddr, 0x1 );

    // SYNC request, because this status will be read from XRAM by the CMA component
    _mmc_sync( usr_sts_paddr, 4 );

    return SYSCALL_OK;

} // end _sys_fbf_cma_display()

///////////////////////
int _sys_fbf_cma_stop()
{
    // get channel index
    unsigned int channel = _get_context_slot( CTX_CMA_FB_ID );

    if ( channel >= NB_CMA_CHANNELS )
    {
        _printf("\n[GIET ERROR] in _sys_fbf_cma_stop() : CMA channel non allocated\n");

        return SYSCALL_CHANNEL_NON_ALLOCATED;
    }

    // Desactivate CMA channel
    _cma_set_register( channel, CHBUF_RUN, MODE_IDLE );

    return SYSCALL_OK;

} // end _sys_fbf_cma_stop()

#endif // if NB_CMA_CHANNELS


//////////////////////////////////////////////////////////////////////////////
//           Miscelaneous syscall handlers 
//////////////////////////////////////////////////////////////////////////////

///////////////
int _sys_ukn() 
{
    _printf("\n[GIET ERROR] Undefined System Call / EPC = %x\n", _get_epc() );

    return SYSCALL_UNDEFINED_SYSTEM_CALL;
}

////////////////////////////////////
int _sys_proc_xyp( unsigned int* x,
                   unsigned int* y,
                   unsigned int* p )
{
    unsigned int gpid = _get_procid();  // global processor index from CPO register

    *x = (gpid >> (Y_WIDTH + P_WIDTH)) & ((1<<X_WIDTH)-1);
    *y = (gpid >> P_WIDTH) & ((1<<Y_WIDTH)-1);
    *p = gpid & ((1<<P_WIDTH)-1);

    return SYSCALL_OK;
}

////////////////////////////////////////////
int _sys_procs_number( unsigned int* x_size,
                       unsigned int* y_size,
                       unsigned int* nprocs )
{
    mapping_header_t * header   = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_cluster_t * cluster = _get_cluster_base(header);

    unsigned int x;
    unsigned int y;
    unsigned int okmin = 1;
    unsigned int okmax = 1;

    // compute max values
    unsigned int xmax  = header->x_size;
    unsigned int ymax  = header->y_size;
    unsigned int procs = cluster[0].procs;

    // check the (ymax-1) lower rows
    for ( y = 0 ; y < ymax-1 ; y++ )
    {
        for ( x = 0 ; x < xmax ; x++ )
        {
            if (cluster[x*ymax+y].procs != procs ) okmin = 0;
        }
    }

    // check the upper row
    for ( x = 0 ; x < xmax ; x++ )
    {
        if (cluster[x*ymax+ymax-1].procs != procs ) okmax = 0;
    }

    // return values
    if ( okmin && okmax )
    {
        *x_size = xmax;
        *y_size = ymax;
        *nprocs = procs;
    }
    else if ( okmin )
    {
        *x_size = xmax;
        *y_size = ymax-1;
        *nprocs = procs;
    }
    else
    {
        *x_size = 0;
        *y_size = 0;
        *nprocs = 0;
    }
    return SYSCALL_OK;
}

///////////////////////////////////////////////////////
int _sys_vseg_get_vbase( char*             vspace_name, 
                         char*             vseg_name, 
                         unsigned int*     vbase ) 
{
    mapping_header_t * header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t * vspace = _get_vspace_base(header);
    mapping_vseg_t * vseg     = _get_vseg_base(header);

    unsigned int vspace_id;
    unsigned int vseg_id;

    // scan vspaces 
    for (vspace_id = 0; vspace_id < header->vspaces; vspace_id++) 
    {
        if (_strncmp( vspace[vspace_id].name, vspace_name, 31) == 0) 
        {
            // scan vsegs
            for (vseg_id = vspace[vspace_id].vseg_offset; 
                 vseg_id < (vspace[vspace_id].vseg_offset + vspace[vspace_id].vsegs); 
                 vseg_id++) 
            {
                if (_strncmp(vseg[vseg_id].name, vseg_name, 31) == 0) 
                {
                    *vbase = vseg[vseg_id].vbase;
                    return SYSCALL_OK;
                }
            } 
        }
    } 
    return SYSCALL_VSEG_NOT_FOUND;
}

/////////////////////////////////////////////////////////
int _sys_vseg_get_length( char*         vspace_name, 
                          char*         vseg_name,
                          unsigned int* length ) 
{
    mapping_header_t * header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_vspace_t * vspace = _get_vspace_base(header);
    mapping_vseg_t * vseg     = _get_vseg_base(header);

    unsigned int vspace_id;
    unsigned int vseg_id;

    // scan vspaces 
    for (vspace_id = 0; vspace_id < header->vspaces; vspace_id++) 
    {
        if (_strncmp( vspace[vspace_id].name, vspace_name, 31) == 0) 
        {
            // scan vsegs
            for (vseg_id = vspace[vspace_id].vseg_offset; 
                 vseg_id < (vspace[vspace_id].vseg_offset + vspace[vspace_id].vsegs); 
                 vseg_id++) 
            {
                if (_strncmp(vseg[vseg_id].name, vseg_name, 31) == 0) 
                {
                    *length = vseg[vseg_id].length;
                    return SYSCALL_OK;
                }
            } 
        }
    } 
    return SYSCALL_VSEG_NOT_FOUND;
}

////////////////////////////////////////
int _sys_xy_from_ptr( void*         ptr,
                      unsigned int* x,
                      unsigned int* y )
{
    unsigned int flags;
    unsigned long long paddr = _v2p_translate( (unsigned int)ptr , &flags );
    
    *x = (paddr>> (32 + Y_WIDTH)) & ((1 << X_WIDTH) - 1);
    *y = (paddr>>32) & ((1 << Y_WIDTH) - 1);

    return SYSCALL_OK;
}

/////////////////////////////////////////
int _sys_heap_info( unsigned int* vaddr, 
                    unsigned int* length,
                    unsigned int  x,
                    unsigned int  y ) 
{
    // checking parameters
    if ( (x >= X_SIZE) || (y >= Y_SIZE) )  
    {
        *vaddr  = 0;
        *length = 0;
        _printf("\n[GIET ERROR] in _sys_heap_info() : "
                "illegal (%d,%d) coordinates\n", x , y );
        return SYSCALL_ILLEGAL_CLUSTER_COORDINATES;
    }

    mapping_header_t * header  = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
    mapping_thread_t * thread  = _get_thread_base(header);
    mapping_vseg_t *   vseg    = _get_vseg_base(header);
    mapping_vspace_t * vspace  = _get_vspace_base(header);

    unsigned int thread_id;
    unsigned int vspace_id;
    unsigned int vseg_id = 0xFFFFFFFF;

    // get calling thread vspace index
    vspace_id = _get_context_slot(CTX_VSID_ID);

    // scan all threads in vspace to find one in clyster[x,y]
    unsigned int min = vspace[vspace_id].thread_offset ;
    unsigned int max = min + vspace[vspace_id].threads ;
    for ( thread_id = min ; thread_id < max ; thread_id++ )
    {
        if ( thread[thread_id].clusterid == (x * Y_SIZE + y) )
        {
            vseg_id = thread[thread_id].heap_vseg_id;
            break;
        }
    }

    // analysing the vseg_id
    if ( vseg_id != 0xFFFFFFFF ) 
    {
        *vaddr  = vseg[vseg_id].vbase;
        *length = vseg[vseg_id].length;
    }
    else 
    {
        *vaddr  = 0;
        *length = 0;
        _printf("error in _sys_heap_info() : no heap in cluster (%d,%d)\n", x , y );
    }
    return SYSCALL_OK;
}  // end _sys_heap_info()


// Local Variables:
// tab-width: 4
// c-basic-offset: 4
// c-file-offsets:((innamespace . 0)(inline-open . 0))
// indent-tabs-mode: nil
// End:
// vim: filetype=c:expandtab:shiftwidth=4:tabstop=4:softtabstop=4