source: trunk/kernel/kern/kernel_init.c @ 648

/*
 * kernel_init.c - kernel parallel initialization
 *
 * Authors :  Mohamed Lamine Karaoui (2015)
 *            Alain Greiner  (2016,2017,2018,2019)
 *
 * Copyright (c) Sorbonne Universites
 *
 * This file is part of ALMOS-MKH.
 *
 * ALMOS-MKH is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2.0 of the License.
 *
 * ALMOS-MKH is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <kernel_config.h>
#include <errno.h>
#include <hal_kernel_types.h>
#include <hal_special.h>
#include <hal_context.h>
#include <hal_irqmask.h>
#include <hal_macros.h>
#include <hal_ppm.h>
#include <barrier.h>
#include <xbarrier.h>
#include <remote_fifo.h>
#include <core.h>
#include <list.h>
#include <xlist.h>
#include <xhtab.h>
#include <thread.h>
#include <scheduler.h>
#include <kmem.h>
#include <cluster.h>
#include <string.h>
#include <memcpy.h>
#include <ppm.h>
#include <page.h>
#include <chdev.h>
#include <boot_info.h>
#include <dqdt.h>
#include <dev_mmc.h>
#include <dev_dma.h>
#include <dev_iob.h>
#include <dev_ioc.h>
#include <dev_txt.h>
#include <dev_pic.h>
#include <printk.h>
#include <vfs.h>
#include <devfs.h>
#include <mapper.h>

///////////////////////////////////////////////////////////////////////////////////////////
// All the following global variables are replicated in all clusters.
// They are initialised by the kernel_init() function.
//
// WARNING : The section names have been defined to control the base addresses of the
// boot_info structure and the idle thread descriptors, through the kernel.ld script:
// - the boot_info structure is built by the bootloader, and used by kernel_init.
//   it must be the first object in the kdata segment.
// - the array of idle threads descriptors must be placed on the first page boundary after
//   the boot_info structure in the kdata segment.
///////////////////////////////////////////////////////////////////////////////////////////

// This variable defines the local boot_info structure
__attribute__((section(".kinfo")))
boot_info_t          boot_info;

// This variable defines the "idle" threads descriptors array
__attribute__((section(".kidle")))
char                 idle_threads[CONFIG_THREAD_DESC_SIZE *
                                   CONFIG_MAX_LOCAL_CORES]   CONFIG_PPM_PAGE_ALIGNED;

// This variable defines the local cluster manager
__attribute__((section(".kdata")))
cluster_t            cluster_manager                         CONFIG_CACHE_LINE_ALIGNED;

// This variable defines the TXT_TX[0] chdev
__attribute__((section(".kdata")))
chdev_t              txt0_tx_chdev                           CONFIG_CACHE_LINE_ALIGNED;

// This variable defines the TXT_RX[0] chdev
__attribute__((section(".kdata")))
chdev_t              txt0_rx_chdev                           CONFIG_CACHE_LINE_ALIGNED;

// This variables define the kernel process0 descriptor
__attribute__((section(".kdata")))
process_t            process_zero                            CONFIG_CACHE_LINE_ALIGNED;

// This variable defines a set of extended pointers on the distributed chdevs
__attribute__((section(".kdata")))
chdev_directory_t    chdev_dir                               CONFIG_CACHE_LINE_ALIGNED;

// This variable contains the input IRQ indexes for the IOPIC controller
__attribute__((section(".kdata")))
iopic_input_t        iopic_input                             CONFIG_CACHE_LINE_ALIGNED;

// This variable contains the input IRQ indexes for the LAPIC controller
__attribute__((section(".kdata")))
lapic_input_t        lapic_input                             CONFIG_CACHE_LINE_ALIGNED;

// This variable defines the local cluster identifier
__attribute__((section(".kdata")))
cxy_t                local_cxy                               CONFIG_CACHE_LINE_ALIGNED;

// This variable is used for core[0] cores synchronisation in kernel_init()
__attribute__((section(".kdata")))
xbarrier_t           global_barrier                          CONFIG_CACHE_LINE_ALIGNED;

// This variable is used for local cores synchronisation in kernel_init()
__attribute__((section(".kdata")))
barrier_t            local_barrier                           CONFIG_CACHE_LINE_ALIGNED;

// This variable defines the array of supported File System contexts
__attribute__((section(".kdata")))
vfs_ctx_t            fs_context[FS_TYPES_NR]                 CONFIG_CACHE_LINE_ALIGNED;

// This array is used for debug, and describes the kernel locks usage,
// It must be kept consistent with the defines in kernel_config.h file.
__attribute__((section(".kdata")))
char * lock_type_str[] =
{
    "unused_0",              //  0

    "CLUSTER_KCM",           //  1
    "SCHED_STATE",           //  2
    "VMM_STACK",             //  3
    "VMM_MMAP",              //  4
    "VFS_CTX",               //  5
    "KCM_STATE",             //  6
    "KHM_STATE",             //  7
    "HTAB_STATE",            //  8

    "PPM_FREE",              //  9
    "THREAD_JOIN",           // 10
    "XHTAB_STATE",           // 11
    "CHDEV_QUEUE",           // 12
    "CHDEV_TXT0",            // 13
    "CHDEV_TXTLIST",         // 14
    "PAGE_STATE",            // 15
    "MUTEX_STATE",           // 16
    "CONDVAR_STATE",         // 17
    "SEM_STATE",             // 18
    "PROCESS_CWD",           // 19
    "BARRIER_STATE",         // 20

    "CLUSTER_PREFTBL",       // 21

    "PPM_DIRTY",             // 22
    "CLUSTER_LOCALS",        // 23
    "CLUSTER_COPIES",        // 24
    "PROCESS_CHILDREN",      // 25
    "PROCESS_USERSYNC",      // 26
    "PROCESS_FDARRAY",       // 27
    "PROCESS_DIR",           // 28
    "VMM_VSL",               // 29

    "PROCESS_THTBL",         // 30

    "MAPPER_STATE",          // 31
    "VFS_SIZE",              // 32
    "VFS_FILE",              // 33
    "VFS_MAIN",              // 34
    "FATFS_FAT",             // 35
};        

// debug variables to analyse the sys_read() and sys_write() syscalls timing

#if DEBUG_SYS_READ
uint32_t   enter_sys_read;
uint32_t   exit_sys_read;

uint32_t   enter_devfs_read;
uint32_t   exit_devfs_read;

uint32_t   enter_txt_read;
uint32_t   exit_txt_read;

uint32_t   enter_chdev_cmd_read;
uint32_t   exit_chdev_cmd_read;

uint32_t   enter_chdev_server_read;
uint32_t   exit_chdev_server_read;

uint32_t   enter_tty_cmd_read;
uint32_t   exit_tty_cmd_read;

uint32_t   enter_tty_isr_read;
uint32_t   exit_tty_isr_read;
#endif

// these debug variables are used to analyse the sys_write() syscall timing

#if DEBUG_SYS_WRITE    
uint32_t   enter_sys_write;
uint32_t   exit_sys_write;

uint32_t   enter_devfs_write;
uint32_t   exit_devfs_write;

uint32_t   enter_txt_write;
uint32_t   exit_txt_write;

uint32_t   enter_chdev_cmd_write;
uint32_t   exit_chdev_cmd_write;

uint32_t   enter_chdev_server_write;
uint32_t   exit_chdev_server_write;

uint32_t   enter_tty_cmd_write;
uint32_t   exit_tty_cmd_write;

uint32_t   enter_tty_isr_write;
uint32_t   exit_tty_isr_write;
#endif

// intrumentation variables : cumulated costs per syscall type in cluster

#if CONFIG_INSTRUMENTATION_SYSCALLS
__attribute__((section(".kdata")))
uint32_t   syscalls_cumul_cost[SYSCALLS_NR];

__attribute__((section(".kdata")))
uint32_t   syscalls_occurences[SYSCALLS_NR];
#endif

///////////////////////////////////////////////////////////////////////////////////////////
// This function displays the ALMOS_MKH banner.
///////////////////////////////////////////////////////////////////////////////////////////
static void print_banner( uint32_t nclusters , uint32_t ncores )
{
    printk("\n"
           "                    _        __    __     _____     ______         __    __    _   __   _     _   \n"
           "          /\\       | |      |  \\  /  |   / ___ \\   / _____|       |  \\  /  |  | | / /  | |   | |  \n"
           "         /  \\      | |      |   \\/   |  | /   \\ | | /             |   \\/   |  | |/ /   | |   | |  \n"
           "        / /\\ \\     | |      | |\\  /| |  | |   | | | |_____   ___  | |\\  /| |  |   /    | |___| |  \n"
           "       / /__\\ \\    | |      | | \\/ | |  | |   | | \\_____  \\ |___| | | \\/ | |  |   \\    |  ___  |  \n"
           "      / ______ \\   | |      | |    | |  | |   | |       | |       | |    | |  | |\\ \\   | |   | |  \n"
           "     / /      \\ \\  | |____  | |    | |  | \\___/ |  _____/ |       | |    | |  | | \\ \\  | |   | |  \n"
           "    /_/        \\_\\ |______| |_|    |_|   \\_____/  |______/        |_|    |_|  |_|  \\_\\ |_|   |_|  \n"
           "\n\n\t\t Advanced Locality Management Operating System / Multi Kernel Hybrid\n"
           "\n\n\t\t %s / %d cluster(s) / %d core(s) per cluster\n\n",
           CONFIG_VERSION , nclusters , ncores );
}


///////////////////////////////////////////////////////////////////////////////////////////
// This function initializes the TXT_TX[0] and TXT_RX[0] chdev descriptors, implementing
// the "kernel terminal", shared by all kernel instances for debug messages. 
// These chdev are implemented as global variables (replicated in all clusters),
// because this terminal is used before the kmem allocator initialisation, but only 
// the chdevs in cluster 0 are registered in the "chdev_dir" directory.
// As this TXT0 chdev supports only the TXT_SYNC_WRITE command, we don't create
// a server thread, we don't allocate a WTI, and we don't initialize the waiting queue.
// Note: The TXT_RX[0] chdev is created, but is not used by ALMOS-MKH (september 2018).
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on the local boot-info structure.
///////////////////////////////////////////////////////////////////////////////////////////
static void __attribute__ ((noinline)) txt0_device_init( boot_info_t * info )
{
    boot_device_t * dev_tbl;         // pointer on array of devices in boot_info
    uint32_t        dev_nr;          // actual number of devices in this cluster
    xptr_t          base;            // remote pointer on segment base
    uint32_t        func;            // device functional index
    uint32_t        impl;            // device implementation index
    uint32_t        i;               // device index in dev_tbl
    uint32_t        x;               // X cluster coordinate
    uint32_t        y;               // Y cluster coordinate

    // get number of peripherals and base of devices array from boot_info
    dev_nr      = info->ext_dev_nr;
    dev_tbl     = info->ext_dev;

    // loop on external peripherals to find TXT device
    for( i = 0 ; i < dev_nr ; i++ )
    {
        base        = dev_tbl[i].base;
        func        = FUNC_FROM_TYPE( dev_tbl[i].type );
        impl        = IMPL_FROM_TYPE( dev_tbl[i].type );

        if (func == DEV_FUNC_TXT )
        {
            // initialize TXT_TX[0] chdev
            txt0_tx_chdev.func    = func;
            txt0_tx_chdev.impl    = impl;
            txt0_tx_chdev.channel = 0;
            txt0_tx_chdev.base    = base;
            txt0_tx_chdev.is_rx   = false;
            remote_busylock_init( XPTR( local_cxy , &txt0_tx_chdev.wait_lock ),
                                  LOCK_CHDEV_TXT0 );
            
            // initialize TXT_RX[0] chdev
            txt0_rx_chdev.func    = func;
            txt0_rx_chdev.impl    = impl;
            txt0_rx_chdev.channel = 0;
            txt0_rx_chdev.base    = base;
            txt0_rx_chdev.is_rx   = true;
            remote_busylock_init( XPTR( local_cxy , &txt0_rx_chdev.wait_lock ),
                                  LOCK_CHDEV_TXT0 );
            
            // make TXT specific initialisations 
            dev_txt_init( &txt0_tx_chdev );                 
            dev_txt_init( &txt0_rx_chdev );

            // register TXT_TX[0] & TXT_RX[0] in chdev_dir[x][y]
            // for all valid clusters             
            for( x = 0 ; x < info->x_size ; x++ )
            {
                for( y = 0 ; y < info->y_size ; y++ )
                {
                    cxy_t cxy = HAL_CXY_FROM_XY( x , y );

                    if( cluster_is_active( cxy ) )
                    {
                        hal_remote_s64( XPTR( cxy , &chdev_dir.txt_tx[0] ) ,
                                        XPTR( local_cxy , &txt0_tx_chdev ) );
                        hal_remote_s64( XPTR( cxy , &chdev_dir.txt_rx[0] ) ,
                                        XPTR( local_cxy , &txt0_rx_chdev ) );
                    }
                }
            }

            hal_fence();
        }
        } // end loop on devices
}  // end txt0_device_init()

///////////////////////////////////////////////////////////////////////////////////////////
// This function allocates memory and initializes the chdev descriptors for the internal
// peripherals contained in the local cluster, other than the LAPIC, as specified by 
// the boot_info, including the linking with the driver for the specified implementation.
// The relevant entries in all copies of the devices directory are initialised.
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on the local boot-info structure.
///////////////////////////////////////////////////////////////////////////////////////////
static void __attribute__ ((noinline)) internal_devices_init( boot_info_t * info )
{
    boot_device_t * dev_tbl;         // pointer on array of internaldevices in boot_info
        uint32_t        dev_nr;          // actual number of devices in this cluster
        xptr_t          base;            // remote pointer on segment base
    uint32_t        func;            // device functionnal index
    uint32_t        impl;            // device implementation index
        uint32_t        i;               // device index in dev_tbl
        uint32_t        x;               // X cluster coordinate
        uint32_t        y;               // Y cluster coordinate
        uint32_t        channels;        // number of channels
        uint32_t        channel;         // channel index
        chdev_t       * chdev_ptr;       // local pointer on created chdev

    // get number of internal peripherals and base from boot_info
        dev_nr  = info->int_dev_nr;
    dev_tbl = info->int_dev;

    // loop on internal peripherals
        for( i = 0 ; i < dev_nr ; i++ )
        {
        base        = dev_tbl[i].base;
        channels    = dev_tbl[i].channels;
        func        = FUNC_FROM_TYPE( dev_tbl[i].type );
        impl        = IMPL_FROM_TYPE( dev_tbl[i].type );
 
        //////////////////////////
        if( func == DEV_FUNC_MMC )  
        {

            // check channels
            if( channels != 1 )
            {
                printk("\n[PANIC] in %s : MMC device must be single channel\n",
                __FUNCTION__ );
                hal_core_sleep();
            }

            // create chdev in local cluster
            chdev_ptr = chdev_create( func,
                                      impl,
                                      0,          // channel
                                      false,      // direction
                                      base );

            // check memory
            if( chdev_ptr == NULL )
            {
                printk("\n[PANIC] in %s : cannot create MMC chdev\n",
                __FUNCTION__ );
                hal_core_sleep();
            }
            
            // make MMC specific initialisation
            dev_mmc_init( chdev_ptr );

            // set the MMC field in all chdev_dir[x][y] structures
            for( x = 0 ; x < info->x_size ; x++ )
            {
                for( y = 0 ; y < info->y_size ; y++ )
                {
                    cxy_t cxy = HAL_CXY_FROM_XY( x , y );

                    if( cluster_is_active( cxy ) )
                    {
                        hal_remote_s64( XPTR( cxy , &chdev_dir.mmc[local_cxy] ), 
                                        XPTR( local_cxy , chdev_ptr ) );
                    }
                }
            }

#if( DEBUG_KERNEL_INIT & 0x1 )
if( hal_time_stamp() > DEBUG_KERNEL_INIT )
printk("\n[%s] : created MMC in cluster %x / chdev = %x\n",
__FUNCTION__ , local_cxy , chdev_ptr );
#endif
        }
        ///////////////////////////////
        else if( func == DEV_FUNC_DMA )
        {
            // create one chdev per channel in local cluster
            for( channel = 0 ; channel < channels ; channel++ )
            {   
                // create chdev[channel] in local cluster
                chdev_ptr = chdev_create( func,
                                          impl,
                                          channel,
                                          false,     // direction
                                          base );

                // check memory
                if( chdev_ptr == NULL )
                {
                    printk("\n[PANIC] in %s : cannot create DMA chdev\n",
                    __FUNCTION__ );
                    hal_core_sleep();
                }
            
                // make DMA specific initialisation
                dev_dma_init( chdev_ptr );     

                // initialize only the DMA[channel] field in the local chdev_dir[x][y]
                // structure because the DMA device is not remotely accessible.
                chdev_dir.dma[channel] = XPTR( local_cxy , chdev_ptr );

#if( DEBUG_KERNEL_INIT & 0x1 )
if( hal_time_stamp() > DEBUG_KERNEL_INIT )
printk("\n[%s] : created DMA[%d] in cluster %x / chdev = %x\n",
__FUNCTION__ , channel , local_cxy , chdev_ptr );
#endif
            }
        }
    }
}  // end internal_devices_init()

///////////////////////////////////////////////////////////////////////////////////////////
// This function allocates memory and initializes the chdev descriptors for the  
// external (shared) peripherals other than the IOPIC, as specified by the boot_info.
// This includes the dynamic linking with the driver for the specified implementation.
// These chdev descriptors are distributed on all clusters, using a modulo on a global 
// index, identically computed in all clusters.
// This function is executed in all clusters by the core[0] core, that computes a global index
// for all external chdevs. Each core[0] core creates only the chdevs that must be placed in 
// the local cluster, because the global index matches the local index. 
// The relevant entries in all copies of the devices directory are initialised.
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on the local boot-info structure.
///////////////////////////////////////////////////////////////////////////////////////////
static void external_devices_init( boot_info_t * info )
{
    boot_device_t * dev_tbl;         // pointer on array of external devices in boot_info
        uint32_t        dev_nr;          // actual number of external devices 
        xptr_t          base;            // remote pointer on segment base
    uint32_t        func;            // device functionnal index
    uint32_t        impl;            // device implementation index
        uint32_t        i;               // device index in dev_tbl
        uint32_t        x;               // X cluster coordinate
        uint32_t        y;               // Y cluster coordinate
        uint32_t        channels;        // number of channels
        uint32_t        channel;         // channel index
        uint32_t        directions;      // number of directions (1 or 2)
        uint32_t        rx;              // direction index (0 or 1)
    chdev_t       * chdev;           // local pointer on one channel_device descriptor
    uint32_t        ext_chdev_gid;   // global index of external chdev

    // get number of peripherals and base of devices array from boot_info
    dev_nr      = info->ext_dev_nr;
    dev_tbl     = info->ext_dev;

    // initializes global index (PIC is already placed in cluster 0
    ext_chdev_gid = 1;

    // loop on external peripherals
    for( i = 0 ; i < dev_nr ; i++ )
    {
        base     = dev_tbl[i].base;
        channels = dev_tbl[i].channels;
        func     = FUNC_FROM_TYPE( dev_tbl[i].type );
        impl     = IMPL_FROM_TYPE( dev_tbl[i].type );

        // There is one chdev per direction for NIC and for TXT
        if((func == DEV_FUNC_NIC) || (func == DEV_FUNC_TXT)) directions = 2;
        else                                                 directions = 1;

        // do nothing for ROM, that does not require a device descriptor.
        if( func == DEV_FUNC_ROM ) continue;

        // do nothing for PIC, that is already initialized
        if( func == DEV_FUNC_PIC ) continue;

        // check PIC device initialized
        if( chdev_dir.pic == XPTR_NULL )
        {
            printk("\n[PANIC] in %s : PIC device must be initialized first\n",
            __FUNCTION__ );
            hal_core_sleep();
        }

        // check external device functionnal type
        if( (func != DEV_FUNC_IOB) && (func != DEV_FUNC_IOC) && (func != DEV_FUNC_TXT) &&
            (func != DEV_FUNC_NIC) && (func != DEV_FUNC_FBF) )
        {
            printk("\n[PANIC] in %s : undefined peripheral type\n",
            __FUNCTION__ );
            hal_core_sleep();
        }

        // loops on channels
        for( channel = 0 ; channel < channels ; channel++ )
        {
            // loop on directions
            for( rx = 0 ; rx < directions ; rx++ )
            {
                // skip TXT0 that has already been initialized
                if( (func == DEV_FUNC_TXT) && (channel == 0) ) continue;

                // all kernel instances compute the target cluster for all chdevs,
                // computing the global index ext_chdev_gid[func,channel,direction]
                cxy_t target_cxy;
                while( 1 )
                {
                    uint32_t offset     = ext_chdev_gid % ( info->x_size * info->y_size );
                    uint32_t x          = offset / info->y_size;
                    uint32_t y          = offset % info->y_size;

                    target_cxy = HAL_CXY_FROM_XY( x , y );

                    // exit loop if target cluster is active
                    if( cluster_is_active( target_cxy ) ) break;
                
                    // increment global index otherwise
                    ext_chdev_gid++;
                }

                // allocate and initialize a local chdev
                // when local cluster matches target cluster
                if( target_cxy == local_cxy )
                {

#if( DEBUG_KERNEL_INIT & 0x3 )
if( hal_time_stamp() > DEBUG_KERNEL_INIT )
printk("\n[%s] : found chdev %s / channel = %d / rx = %d / cluster %x\n",
__FUNCTION__ , chdev_func_str( func ), channel , rx , local_cxy );
#endif
                    chdev = chdev_create( func,
                                          impl,
                                          channel,
                                          rx,          // direction
                                          base );

                    if( chdev == NULL )
                    {
                        printk("\n[PANIC] in %s : cannot allocate chdev\n",
                        __FUNCTION__ );
                        hal_core_sleep();
                    }

                    // make device type specific initialisation
                    if     ( func == DEV_FUNC_IOB ) dev_iob_init( chdev );
                    else if( func == DEV_FUNC_IOC ) dev_ioc_init( chdev );
                    else if( func == DEV_FUNC_TXT ) dev_txt_init( chdev );
                    else if( func == DEV_FUNC_NIC ) dev_nic_init( chdev );
                    else if( func == DEV_FUNC_FBF ) dev_fbf_init( chdev );

                    // all external (shared) devices are remotely accessible
                    // initialize the replicated chdev_dir[x][y] structures
                    // defining the extended pointers on chdev descriptors
                    xptr_t * entry = NULL;

                    if(func==DEV_FUNC_IOB             ) entry  = &chdev_dir.iob;
                    if(func==DEV_FUNC_IOC             ) entry  = &chdev_dir.ioc[channel];
                    if(func==DEV_FUNC_FBF             ) entry  = &chdev_dir.fbf[channel];
                    if((func==DEV_FUNC_TXT) && (rx==0)) entry  = &chdev_dir.txt_tx[channel];
                    if((func==DEV_FUNC_TXT) && (rx==1)) entry  = &chdev_dir.txt_rx[channel];
                    if((func==DEV_FUNC_NIC) && (rx==0)) entry  = &chdev_dir.nic_tx[channel];
                    if((func==DEV_FUNC_NIC) && (rx==1)) entry  = &chdev_dir.nic_rx[channel];

                    for( x = 0 ; x < info->x_size ; x++ )
                    {
                        for( y = 0 ; y < info->y_size ; y++ )
                        {
                            cxy_t cxy = HAL_CXY_FROM_XY( x , y );

                            if( cluster_is_active( cxy ) && ( entry != NULL ) )
                            {
                                hal_remote_s64( XPTR( cxy , entry ),
                                                XPTR( local_cxy , chdev ) );
                            }
                        }
                    }

#if( DEBUG_KERNEL_INIT & 0x3 )
if( hal_time_stamp() > DEBUG_KERNEL_INIT )
printk("\n[%s] : created chdev %s / channel = %d / rx = %d / cluster %x / chdev = %x\n",
__FUNCTION__ , chdev_func_str( func ), channel , rx , local_cxy , chdev );
#endif
                }  // end if match

                // increment chdev global index (matching or not)
                ext_chdev_gid++;

            } // end loop on directions
        }  // end loop on channels
        } // end loop on devices
}  // end external_devices_init()

///////////////////////////////////////////////////////////////////////////////////////////
// This function is called by core[0] in cluster 0 to allocate memory and initialize the PIC
// device, namely the informations attached to the external IOPIC controller, that
// must be replicated in all clusters (struct iopic_input).
// This initialisation must be done before other devices initialisation because the IRQ
// routing infrastructure is required for both internal and external devices init.
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on the local boot-info structure.
///////////////////////////////////////////////////////////////////////////////////////////
static void __attribute__ ((noinline)) iopic_init( boot_info_t * info )
{
    boot_device_t * dev_tbl;         // pointer on boot_info external devices array
        uint32_t        dev_nr;          // actual number of external devices
        xptr_t          base;            // remote pointer on segment base
    uint32_t        func;            // device functionnal index
    uint32_t        impl;            // device implementation index
        uint32_t        i;               // device index in dev_tbl
    uint32_t        x;               // cluster X coordinate
    uint32_t        y;               // cluster Y coordinate
    bool_t          found;           // IOPIC found
        chdev_t       * chdev;           // pointer on PIC chdev descriptor

    // get number of external peripherals and base of array from boot_info
        dev_nr      = info->ext_dev_nr;
    dev_tbl     = info->ext_dev;

    // avoid GCC warning
    base        = XPTR_NULL;
    impl        = 0;

    // loop on external peripherals to get the IOPIC  
        for( i = 0 , found = false ; i < dev_nr ; i++ )
        {
        func = FUNC_FROM_TYPE( dev_tbl[i].type );

        if( func == DEV_FUNC_PIC )
        {
            base     = dev_tbl[i].base;
            impl     = IMPL_FROM_TYPE( dev_tbl[i].type );
            found    = true;
            break;
        }
    }

    // check PIC existence
    if( found == false )
    {
        printk("\n[PANIC] in %s : PIC device not found\n",
        __FUNCTION__ );
        hal_core_sleep();
    }

    // allocate and initialize the PIC chdev in cluster 0
    chdev = chdev_create( DEV_FUNC_PIC,
                          impl,
                          0,      // channel
                          0,      // direction,
                          base );

    // check memory
    if( chdev == NULL )
    {
        printk("\n[PANIC] in %s : no memory for PIC chdev\n",
        __FUNCTION__ );
        hal_core_sleep();
    }

    // make PIC device type specific initialisation
    dev_pic_init( chdev );

    // register, in all clusters, the extended pointer
    // on PIC chdev in "chdev_dir" array
    xptr_t * entry = &chdev_dir.pic;    
                
    for( x = 0 ; x < info->x_size ; x++ )
    {
        for( y = 0 ; y < info->y_size ; y++ )
        {
            cxy_t cxy = HAL_CXY_FROM_XY( x , y );

            if( cluster_is_active( cxy ) )
            {
                hal_remote_s64( XPTR( cxy , entry ) , 
                                XPTR( local_cxy , chdev ) );
            }
        }
    }

    // initialize, in all clusters, the "iopic_input" structure
    // defining how external IRQs are connected to IOPIC 

    // register default value for unused inputs 
    for( x = 0 ; x < info->x_size ; x++ )
    {
        for( y = 0 ; y < info->y_size ; y++ )
        {
            cxy_t cxy = HAL_CXY_FROM_XY( x , y );

            if( cluster_is_active( cxy ) )
            {
                hal_remote_memset( XPTR( cxy , &iopic_input ), 
                                   0xFF , sizeof(iopic_input_t) );
            }
        }
    }

    // register input IRQ index for valid inputs 
    uint32_t   id;             // input IRQ index
    uint8_t    valid;          // input IRQ is connected
    uint32_t   type;           // source device type
    uint8_t    channel;        // source device channel
    uint8_t    is_rx;          // source device direction
    uint32_t * ptr = NULL;     // local pointer on one field in iopic_input stucture

    for( id = 0 ; id < CONFIG_MAX_EXTERNAL_IRQS ; id++ )
    {
        valid   = dev_tbl[i].irq[id].valid;
        type    = dev_tbl[i].irq[id].dev_type;
        channel = dev_tbl[i].irq[id].channel;
        is_rx   = dev_tbl[i].irq[id].is_rx;
        func    = FUNC_FROM_TYPE( type );

        // get pointer on relevant field in iopic_input
        if( valid )
        {
            if     ( func == DEV_FUNC_IOC )                 ptr = &iopic_input.ioc[channel]; 
            else if((func == DEV_FUNC_TXT) && (is_rx == 0)) ptr = &iopic_input.txt_tx[channel];
            else if((func == DEV_FUNC_TXT) && (is_rx != 0)) ptr = &iopic_input.txt_rx[channel];
            else if((func == DEV_FUNC_NIC) && (is_rx == 0)) ptr = &iopic_input.nic_tx[channel];
            else if((func == DEV_FUNC_NIC) && (is_rx != 0)) ptr = &iopic_input.nic_rx[channel];
            else if( func == DEV_FUNC_IOB )                 ptr = &iopic_input.iob;
            else
            {
                printk("\n[PANIC] in %s : illegal source device for IOPIC input\n",
                __FUNCTION__ );
                hal_core_sleep();
            }

            // set one entry in all "iopic_input" structures
            for( x = 0 ; x < info->x_size ; x++ )
            {
                for( y = 0 ; y < info->y_size ; y++ )
                {
                    cxy_t cxy = HAL_CXY_FROM_XY( x , y );

                    if( cluster_is_active( cxy ) )
                    {
                        hal_remote_s64( XPTR( cxy , ptr ) , id ); 
                    }
                }
            }
        }
    } 

#if( DEBUG_KERNEL_INIT & 0x1 )
if( hal_time_stamp() > DEBUG_KERNEL_INIT )
{
    printk("\n[%s] created PIC chdev in cluster %x at cycle %d\n",
    __FUNCTION__ , local_cxy , (uint32_t)hal_time_stamp() );
    dev_pic_inputs_display();
}
#endif
    
}  // end iopic_init()

///////////////////////////////////////////////////////////////////////////////////////////
// This function is called by all core[0]s in all cluster to complete the PIC device
// initialisation, namely the informations attached to the LAPIC controller.
// This initialisation must be done after the IOPIC initialisation, but before other 
// devices initialisation because the IRQ routing infrastructure is required for both
// internal and external devices initialisation.
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on the local boot-info structure.
///////////////////////////////////////////////////////////////////////////////////////////
static void __attribute__ ((noinline)) lapic_init( boot_info_t * info )
{
    boot_device_t * dev_tbl;      // pointer on boot_info internal devices array
    uint32_t        dev_nr;       // number of internal devices
    uint32_t        i;            // device index in dev_tbl
        xptr_t          base;         // remote pointer on segment base
    uint32_t        func;         // device functionnal type in boot_info
    bool_t          found;        // LAPIC found

    // get number of internal peripherals and base 
        dev_nr      = info->int_dev_nr;
    dev_tbl     = info->int_dev;

    // loop on internal peripherals to get the lapic device 
        for( i = 0 , found = false ; i < dev_nr ; i++ )
        {
        func = FUNC_FROM_TYPE( dev_tbl[i].type );

        if( func == DEV_FUNC_ICU )
        {
            base     = dev_tbl[i].base;
            found    = true;
            break;
        }
    }

    // if the LAPIC controller is not defined in the boot_info,
    // we simply don't initialize the PIC extensions in the kernel, 
    // making the assumption that the LAPIC related informations 
    // are hidden in the hardware specific PIC driver. 
    if( found )
    {
        // initialise the PIC extensions for
        // the core descriptor and core manager extensions
        dev_pic_extend_init( (uint32_t *)GET_PTR( base ) );

        // initialize the "lapic_input" structure
        // defining how internal IRQs are connected to LAPIC 
        uint32_t        id;
        uint8_t         valid;
        uint8_t         channel;
        uint32_t        func;

        for( id = 0 ; id < CONFIG_MAX_INTERNAL_IRQS ; id++ )
        {
            valid    = dev_tbl[i].irq[id].valid;
            func     = FUNC_FROM_TYPE( dev_tbl[i].irq[id].dev_type );
            channel  = dev_tbl[i].irq[id].channel;

            if( valid ) // only valid local IRQs are registered
            {
                if     ( func == DEV_FUNC_MMC ) lapic_input.mmc = id;
                else if( func == DEV_FUNC_DMA ) lapic_input.dma[channel] = id;
                else
                {
                    printk("\n[PANIC] in %s : illegal source device for LAPIC input\n",
                    __FUNCTION__ );
                    hal_core_sleep();
                }
            }
        }
    }
}  // end lapic_init()

///////////////////////////////////////////////////////////////////////////////////////////
// This static function returns the identifiers of the calling core.
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on boot_info structure.
// @ lid     : [out] core local index in cluster.
// @ cxy     : [out] cluster identifier.
// @ lid     : [out] core global identifier (hardware).
// @ return 0 if success / return EINVAL if not found.
///////////////////////////////////////////////////////////////////////////////////////////
static error_t __attribute__ ((noinline)) get_core_identifiers( boot_info_t * info,
                                                                lid_t       * lid,
                                                                cxy_t       * cxy,
                                                                gid_t       * gid )
{
    uint32_t   i;
    gid_t      global_id;

    // get global identifier from hardware register
    global_id = hal_get_gid();

    // makes an associative search in boot_info to get (cxy,lid) from global_id
    for( i = 0 ; i < info->cores_nr ; i++ )
    {
        if( global_id == info->core[i].gid )
        {
            *lid = info->core[i].lid;
            *cxy = info->core[i].cxy;
            *gid = global_id;
            return 0;
        }
    }
    return EINVAL;
}





/////////////////////////////////
// kleenex debug function
/////////////////////////////////
void display_fat( uint32_t step )
{
    fatfs_ctx_t * fatfs_ctx = fs_context[FS_TYPE_FATFS].extend;
    if( fatfs_ctx != NULL ) 
    {
        printk("\n[%s] step %d at cycle %d\n", __FUNCTION__, step, (uint32_t)hal_get_cycles() );
        xptr_t     mapper_xp = fatfs_ctx->fat_mapper_xp;
        mapper_display_page( mapper_xp , 0 , 128 );
    }
    else
    {
        printk("\n[%s] step %d : fatfs context not initialized\n", __FUNCTION__, step );
    }
}





///////////////////////////////////////////////////////////////////////////////////////////
// This function is the entry point for the kernel initialisation.
// It is executed by all cores in all clusters, but only core[0] initializes
// the shared resources such as the cluster manager, or the local peripherals.
// To comply with the multi-kernels paradigm, it accesses only local cluster memory, using
// only information contained in the local boot_info_t structure, set by the bootloader.
// Only core[0] in cluster 0 print the log messages.
///////////////////////////////////////////////////////////////////////////////////////////
// @ info    : pointer on the local boot-info structure.
///////////////////////////////////////////////////////////////////////////////////////////
void kernel_init( boot_info_t * info )
{
    lid_t        core_lid = -1;             // running core local index
    cxy_t        core_cxy = -1;             // running core cluster identifier
    gid_t        core_gid;                  // running core hardware identifier 
    cluster_t  * cluster;                   // pointer on local cluster manager
    core_t     * core;                      // pointer on running core descriptor
    thread_t   * thread;                    // pointer on idle thread descriptor

    xptr_t       vfs_root_inode_xp;         // extended pointer on VFS root inode
    xptr_t       devfs_dev_inode_xp;        // extended pointer on DEVFS dev inode    
    xptr_t       devfs_external_inode_xp;   // extended pointer on DEVFS external inode       
    xptr_t       devfs_internal_inode_xp;   // extended pointer on DEVFS internal inode       

    error_t      error;
    reg_t        status;                    // running core status register

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 1 : Each core get its core identifier from boot_info, and makes 
    //          a partial initialisation of its private idle thread descriptor.
    //          core[0] initializes the "local_cxy" global variable.
    //          core[0] in cluster[0] initializes the TXT0 chdev for log messages.
    /////////////////////////////////////////////////////////////////////////////////

    error = get_core_identifiers( info,
                                  &core_lid,
                                  &core_cxy,
                                  &core_gid );

    // core[0] initialize cluster identifier
    if( core_lid == 0 ) local_cxy = info->cxy;

    // each core gets a pointer on its private idle thread descriptor
    thread = (thread_t *)( idle_threads + (core_lid * CONFIG_THREAD_DESC_SIZE) );

    // each core registers this thread pointer in hardware register
    hal_set_current_thread( thread );

    // each core register core descriptor pointer in idle thread descriptor
    thread->core = &LOCAL_CLUSTER->core_tbl[core_lid];

    // each core initializes the idle thread locks counters
    thread->busylocks = 0;

#if DEBUG_BUSYLOCK
    // each core initialise the idle thread list of busylocks
    xlist_root_init( XPTR( local_cxy , &thread->busylocks_root ) );
#endif

    // core[0] initializes cluster info
    if( core_lid == 0 ) cluster_info_init( info );

    // core[0] in cluster[0] initialises TXT0 chdev descriptor
    if( (core_lid == 0) && (core_cxy == 0) ) txt0_device_init( info );

    // all cores check identifiers
    if( error )
    {
        printk("\n[PANIC] in %s : illegal core : gid %x / cxy %x / lid %d",
        __FUNCTION__, core_lid, core_cxy, core_lid );
        hal_core_sleep();
    }

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 1 : TXT0 initialized / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 2 : core[0] initializes the cluster manager,
    //          including the physical memory allocators. 
    /////////////////////////////////////////////////////////////////////////////////

    // core[0] initialises DQDT (only core[0] in cluster 0 build the quad-tree)
    if( core_lid == 0 ) dqdt_init();
    
    // core[0] initialize other cluster manager complex structures
    if( core_lid == 0 )
    {
        error = cluster_manager_init( info );

        if( error )
        {
             printk("\n[PANIC] in %s : cannot initialize cluster manager in cluster %x\n",
             __FUNCTION__, local_cxy );
             hal_core_sleep();
        }
    }

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 2 : cluster manager initialized / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 3 : all cores initialize the idle thread descriptor.
    //          core[0] initializes the process_zero descriptor,
    //          including the kernel VMM (both GPT and VSL) 
    /////////////////////////////////////////////////////////////////////////////////

    // all cores get pointer on local cluster manager & core descriptor
    cluster = &cluster_manager;
    core    = &cluster->core_tbl[core_lid];

    // all cores update the register(s) defining the kernel
    // entry points for interrupts, exceptions and syscalls,
    // this must be done before VFS initialisation, because 
    // kernel_init() uses RPCs requiring IPIs...
    hal_set_kentry();

    // all cores initialize the idle thread descriptor
    thread_idle_init( thread,
                      THREAD_IDLE,
                      &thread_idle_func,
                      NULL,
                      core_lid );

    // core[0] initializes the process_zero descriptor,
    if( core_lid == 0 ) process_zero_create( &process_zero , info );

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 3 : kernel processs initialized / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 4 : all cores initialize their private MMU 
    //          core[0] in cluster 0 initializes the IOPIC device.
    /////////////////////////////////////////////////////////////////////////////////

    // all cores initialise their MMU
    hal_mmu_init( &process_zero.vmm.gpt );

    // core[0] in cluster[0] initializes the PIC chdev,
    if( (core_lid == 0) && (local_cxy == 0) ) iopic_init( info );
    
    ////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    ////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 4 : MMU and IOPIC initialized / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

    ////////////////////////////////////////////////////////////////////////////////
    // STEP 5 : core[0] initialize the distibuted LAPIC descriptor.
    //          core[0] initialize the internal chdev descriptors
    //          core[0] initialize the local external chdev descriptors
    ////////////////////////////////////////////////////////////////////////////////

    // all core[0]s initialize their local LAPIC extension,
    if( core_lid == 0 ) lapic_init( info );

    // core[0] scan the internal (private) peripherals,
    // and allocates memory for the corresponding chdev descriptors.
    if( core_lid == 0 ) internal_devices_init( info );
        

    // All core[0]s contribute to initialise external peripheral chdev descriptors.
    // Each core[0][cxy] scan the set of external (shared) peripherals (but the TXT0),
    // and allocates memory for the chdev descriptors that must be placed
    // on the (cxy) cluster according to the global index value.

    if( core_lid == 0 ) external_devices_init( info );

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 5 : chdevs initialised / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

#if( DEBUG_KERNEL_INIT & 1 )
if( (core_lid ==  0) & (local_cxy == 0) ) 
chdev_dir_display();
#endif
    
    /////////////////////////////////////////////////////////////////////////////////
    // STEP 6 : all cores enable IPI (Inter Procesor Interrupt),
    //          all cores unblock the idle thread, and register it in scheduler.
    //          core[0] in cluster[0] creates the VFS root inode.
    //          It access the boot device to initialize the file system context.
    /////////////////////////////////////////////////////////////////////////////////

    // All cores enable IPI 
    dev_pic_enable_ipi();
    hal_enable_irq( &status );

    // all cores unblock the idle thread, and register it in scheduler
    thread_unblock( XPTR( local_cxy , thread ) , THREAD_BLOCKED_GLOBAL );
    core->scheduler.idle = thread;

    // core[O] in cluster[0] creates the VFS root
    if( (core_lid ==  0) && (local_cxy == 0 ) ) 
    {
        vfs_root_inode_xp = XPTR_NULL;

        // Only FATFS is supported yet,
        // other File System can be introduced here
        if( CONFIG_VFS_ROOT_IS_FATFS )
        {
            // 1. allocate memory for FATFS context in cluster 0
            fatfs_ctx_t * fatfs_ctx = fatfs_ctx_alloc();

            if( fatfs_ctx == NULL )
            {
                printk("\n[PANIC] in %s : cannot create FATFS context in cluster 0\n",
                __FUNCTION__ );
                hal_core_sleep();
            }

            // 2. access boot device to initialize FATFS context
            fatfs_ctx_init( fatfs_ctx );

            // 3. get various informations from FATFS context
            uint32_t root_dir_cluster = fatfs_ctx->root_dir_cluster;
            uint32_t cluster_size     = fatfs_ctx->bytes_per_sector * 
                                        fatfs_ctx->sectors_per_cluster;
            uint32_t total_clusters   = fatfs_ctx->fat_sectors_count << 7;
 
            // 4. create VFS root inode in cluster 0
            error = vfs_inode_create( FS_TYPE_FATFS,                       // fs_type
                                      0,                                   // attr
                                      0,                                   // rights
                                      0,                                   // uid
                                      0,                                   // gid
                                      &vfs_root_inode_xp );                // return
            if( error )
            {
                printk("\n[PANIC] in %s : cannot create VFS root inode in cluster 0\n",
                __FUNCTION__ );
                hal_core_sleep();
            }

            // 5. update FATFS root inode "type" and "extend" fields  
            cxy_t         vfs_root_cxy = GET_CXY( vfs_root_inode_xp );
            vfs_inode_t * vfs_root_ptr = GET_PTR( vfs_root_inode_xp );
            hal_remote_s32( XPTR( vfs_root_cxy , &vfs_root_ptr->type ), INODE_TYPE_DIR );
            hal_remote_spt( XPTR( vfs_root_cxy , &vfs_root_ptr->extend ), 
                            (void*)(intptr_t)root_dir_cluster );

            // 6. initialize the generic VFS context for FATFS 
            vfs_ctx_init( FS_TYPE_FATFS,                               // fs type
                          0,                                           // attributes: unused 
                              total_clusters,                              // number of clusters
                              cluster_size,                                // bytes
                              vfs_root_inode_xp,                           // VFS root
                          fatfs_ctx );                                 // extend
        }
        else
        {
            printk("\n[PANIC] in %s : unsupported VFS type in cluster 0\n",
            __FUNCTION__ );
            hal_core_sleep();
        }

        // create the <.> and <..> dentries in VFS root directory
        // the VFS root parent inode is the VFS root inode itself 
        vfs_add_special_dentries( vfs_root_inode_xp,
                                  vfs_root_inode_xp );

        // register VFS root inode in process_zero descriptor of cluster 0
        process_zero.vfs_root_xp = vfs_root_inode_xp;
        process_zero.cwd_xp      = vfs_root_inode_xp;
    }

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 6 : VFS root (%x,%x) in cluster 0 / cycle %d\n",
__FUNCTION__, GET_CXY(process_zero.vfs_root_xp),
GET_PTR(process_zero.vfs_root_xp), (uint32_t)hal_get_cycles() );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 7 : In all other clusters than cluster[0], the core[0] allocates memory
    //          for the selected FS context, and initialise the local FS context and 
    //          the local VFS context from values stored in cluster 0. 
    //          They get the VFS root inode extended pointer from cluster 0.
    /////////////////////////////////////////////////////////////////////////////////

    if( (core_lid ==  0) && (local_cxy != 0) ) 
    {
        // File System must be FATFS in this implementation,
        // but other File System can be introduced here
        if( CONFIG_VFS_ROOT_IS_FATFS )
        {
            // 1. allocate memory for local FATFS context
            fatfs_ctx_t * local_fatfs_ctx = fatfs_ctx_alloc();

            // check memory
            if( local_fatfs_ctx == NULL )
            {
                printk("\n[PANIC] in %s : cannot create FATFS context in cluster %x\n",
                __FUNCTION__ , local_cxy );
                hal_core_sleep();
            }

            // 2. get local pointer on VFS context for FATFS
            vfs_ctx_t   * vfs_ctx = &fs_context[FS_TYPE_FATFS];

            // 3. get local pointer on FATFS context in cluster 0 
            fatfs_ctx_t * remote_fatfs_ctx = hal_remote_lpt( XPTR( 0 , &vfs_ctx->extend ) );

            // 4. copy FATFS context from cluster 0 to local cluster
            hal_remote_memcpy( XPTR( local_cxy , local_fatfs_ctx ), 
                               XPTR( 0 ,         remote_fatfs_ctx ), sizeof(fatfs_ctx_t) );

            // 5. copy VFS context from cluster 0 to local cluster
            hal_remote_memcpy( XPTR( local_cxy , vfs_ctx ), 
                               XPTR( 0 ,         vfs_ctx ), sizeof(vfs_ctx_t) );

            // 6. update extend field in local copy of VFS context
            vfs_ctx->extend = local_fatfs_ctx;

            if( ((fatfs_ctx_t *)vfs_ctx->extend)->sectors_per_cluster != 8 )
            {
                printk("\n[PANIC] in %s : illegal FATFS context in cluster %x\n",
                __FUNCTION__ , local_cxy );
                hal_core_sleep();
            }
        }

        // get extended pointer on VFS root inode from cluster 0
        vfs_root_inode_xp = hal_remote_l64( XPTR( 0 , &process_zero.vfs_root_xp ) );

        // update local process_zero descriptor
        process_zero.vfs_root_xp = vfs_root_inode_xp;
        process_zero.cwd_xp      = vfs_root_inode_xp;
    }

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 1) ) 
printk("\n[%s] exit barrier 7 : VFS root (%x,%x) in cluster 1 / cycle %d\n",
__FUNCTION__, GET_CXY(process_zero.vfs_root_xp),
GET_PTR(process_zero.vfs_root_xp), (uint32_t)hal_get_cycles() );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 8 : core[0] in cluster 0 makes the global DEVFS initialisation:
    //          It initializes the DEVFS context, and creates the DEVFS
    //          "dev" and "external" inodes in cluster 0.
    /////////////////////////////////////////////////////////////////////////////////

    if( (core_lid ==  0) && (local_cxy == 0) ) 
    {
        // 1. allocate memory for DEVFS context extension in cluster 0
        devfs_ctx_t * devfs_ctx = devfs_ctx_alloc();

        if( devfs_ctx == NULL )
        {
            printk("\n[PANIC] in %s : cannot create DEVFS context in cluster 0\n",
            __FUNCTION__ , local_cxy );
            hal_core_sleep();
        }

        // 2. initialize the DEVFS entry in the vfs_context[] array 
        vfs_ctx_init( FS_TYPE_DEVFS,                                // fs type
                      0,                                            // attributes: unused
                          0,                                            // total_clusters: unused
                          0,                                            // cluster_size: unused 
                          vfs_root_inode_xp,                            // VFS root
                      devfs_ctx );                                  // extend

        // 3. create "dev" and "external" inodes (directories)
        devfs_global_init( process_zero.vfs_root_xp,
                           &devfs_dev_inode_xp,
                           &devfs_external_inode_xp );

        // 4. initializes DEVFS context extension
        devfs_ctx_init( devfs_ctx,
                        devfs_dev_inode_xp,
                        devfs_external_inode_xp );
    }    

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 8 : DEVFS root initialized in cluster 0 / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 9 : In all clusters in parallel, core[0] completes DEVFS initialization.
    //          Each core[0] get the "dev" and "external" extended pointers from
    //          values stored in cluster(0), creates the DEVFS "internal" directory, 
    //          and creates the pseudo-files for all chdevs in local cluster.
    /////////////////////////////////////////////////////////////////////////////////

    if( core_lid == 0 )
    {
        // get extended pointer on "extend" field of VFS context for DEVFS in cluster 0
        xptr_t  extend_xp = XPTR( 0 , &fs_context[FS_TYPE_DEVFS].extend );

        // get pointer on DEVFS context in cluster 0
        devfs_ctx_t * devfs_ctx = hal_remote_lpt( extend_xp );
        
        devfs_dev_inode_xp      = hal_remote_l64( XPTR( 0 , &devfs_ctx->dev_inode_xp ) );
        devfs_external_inode_xp = hal_remote_l64( XPTR( 0 , &devfs_ctx->external_inode_xp ) );

        // populate DEVFS in all clusters
        devfs_local_init( devfs_dev_inode_xp,
                          devfs_external_inode_xp,
                          &devfs_internal_inode_xp );
    }

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ), 
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 9 : DEVFS initialized in cluster 0 / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

#if( DEBUG_KERNEL_INIT & 1 )
if( (core_lid ==  0) & (local_cxy == 0) ) 
vfs_display( vfs_root_inode_xp );
#endif

    /////////////////////////////////////////////////////////////////////////////////
    // STEP 10 : core[0] in cluster 0 creates the first user process (process_init).
    //           This include the first user process VMM (GPT and VSL) creation.
    //           Finally, it prints the ALMOS-MKH banner.
    /////////////////////////////////////////////////////////////////////////////////

    if( (core_lid == 0) && (local_cxy == 0) ) 
    {
       process_init_create();
    }

#if DEBUG_KERNEL_INIT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n[%s] exit barrier 10 : process_init created in cluster 0 / cycle %d\n",
__FUNCTION__, (uint32_t)hal_get_cycles() );
#endif

    if( (core_lid == 0) && (local_cxy == 0) ) 
    {
        print_banner( (info->x_size * info->y_size) , info->cores_nr );
    }

#if CONFIG_INSTRUMENTATION_FOOTPRINT
if( (core_lid ==  0) & (local_cxy == 0) ) 
printk("\n\n***** memory fooprint for main kernel objects\n\n"
                   " - thread descriptor  : %d bytes\n"
                   " - process descriptor : %d bytes\n"
                   " - cluster manager    : %d bytes\n"
                   " - chdev descriptor   : %d bytes\n"
                   " - core descriptor    : %d bytes\n"
                   " - scheduler          : %d bytes\n"
                   " - rpc fifo           : %d bytes\n"
                   " - page descriptor    : %d bytes\n"
                   " - mapper descriptor  : %d bytes\n"
                   " - vseg descriptor    : %d bytes\n"
                   " - ppm manager        : %d bytes\n"
                   " - kcm manager        : %d bytes\n"
                   " - khm manager        : %d bytes\n"
                   " - vmm manager        : %d bytes\n"
                   " - vfs inode          : %d bytes\n"
                   " - vfs dentry         : %d bytes\n"
                   " - vfs file           : %d bytes\n"
                   " - vfs context        : %d bytes\n"
                   " - xhtab root         : %d bytes\n"
                   " - list item          : %d bytes\n"
                   " - xlist item         : %d bytes\n"
                   " - busylock           : %d bytes\n"
                   " - remote busylock    : %d bytes\n"
                   " - queuelock          : %d bytes\n"
                   " - remote queuelock   : %d bytes\n"
                   " - rwlock             : %d bytes\n"
                   " - remote rwlock      : %d bytes\n",
                   sizeof( thread_t           ),
                   sizeof( process_t          ),
                   sizeof( cluster_t          ),
                   sizeof( chdev_t            ),
                   sizeof( core_t             ),
                   sizeof( scheduler_t        ),
                   sizeof( remote_fifo_t      ),
                   sizeof( page_t             ),
                   sizeof( mapper_t           ),
                   sizeof( vseg_t             ),
                   sizeof( ppm_t              ),
                   sizeof( kcm_t              ),
                   sizeof( khm_t              ),
                   sizeof( vmm_t              ),
                   sizeof( vfs_inode_t        ),
                   sizeof( vfs_dentry_t       ),
                   sizeof( vfs_file_t         ),
                   sizeof( vfs_ctx_t          ),
                   sizeof( xhtab_t            ),
                   sizeof( list_entry_t       ),
                   sizeof( xlist_entry_t      ),
                   sizeof( busylock_t         ),
                   sizeof( remote_busylock_t  ),
                   sizeof( queuelock_t        ),
                   sizeof( remote_queuelock_t ),
                   sizeof( rwlock_t           ),
                   sizeof( remote_rwlock_t    ));
#endif

    // each core activates its private TICK IRQ
    dev_pic_enable_timer( CONFIG_SCHED_TICK_MS_PERIOD );

    /////////////////////////////////////////////////////////////////////////////////
    if( core_lid == 0 ) xbarrier_wait( XPTR( 0 , &global_barrier ),
                                        (info->x_size * info->y_size) );
    barrier_wait( &local_barrier , info->cores_nr );
    /////////////////////////////////////////////////////////////////////////////////

#if DEBUG_KERNEL_INIT 
thread_t * this = CURRENT_THREAD;
printk("\n[%s] : thread[%x,%x] on core[%x,%d] jumps to thread_idle_func() / cycle %d\n",
__FUNCTION__ , this->process->pid, this->trdid,
local_cxy, core_lid, (uint32_t)hal_get_cycles() );
#endif

    // each core jump to thread_idle_func
    thread_idle_func();

}  // end kernel_init()