Changeset 657 for trunk/kernel/devices/dev_nic.c

Timestamp:

Mar 18, 2020, 11:16:59 PM (4 years ago)

Author:

alain

Message:

Introduce remote_buf.c/.h & socket.c/.h files.
Update dev_nic.c/.h files.

File:

• trunk/kernel/devices/dev_nic.c (modified) (3 diffs)

trunk/kernel/devices/dev_nic.c

@@ -2 +2 @@
  * dev_nic.c - NIC (Network Controler) generic device API implementation.
  * 
- * Author  Alain Greiner    (2016,2017,2018,2019)
+ * Author  Alain Greiner    (2016,2017,2018,2019,2020)
  *
  * Copyright (c) UPMC Sorbonne Universites
@@ -24 +24 @@
 #include <hal_kernel_types.h>
 #include <hal_special.h>
+#include <hal_uspace.h>
+#include <remote_buf.h>
 #include <printk.h>
 #include <chdev.h>
 #include <thread.h>
+#include <socket.h>
 #include <hal_drivers.h>
 #include <dev_nic.h>
+#include <vfs.h>
+#include <shared_socket.h>
 
 /////////////////////////////////////////////////////////////////////////////////////////
-// Extern global variables
+//                 Extern global variables
 /////////////////////////////////////////////////////////////////////////////////////////
 
 extern chdev_directory_t  chdev_dir;         // allocated in kernel_init.c
+
+////////////////////////////////////////////////////////////////////////////////////////////
+// This static function is used by the dev_nic_rx_handle_tcp() & dev_nic_tx_handle_tcp()
+// functions to check acceptability of a given sequence number. It returns true when 
+// the <seq> argument is contained in a wrap-around window defined by the <min> and <max>
+// arguments. The window wrap-around when (min > max).
+////////////////////////////////////////////////////////////////////////////////////////////
+// @ seq   : [in] value to be checked.
+// @ min   : [in] first  base.
+// @ max   : [in] window size.
+////////////////////////////////////////////////////////////////////////////////////////////
+static inline bool_t is_in_window( uint32_t seq,
+                                   uint32_t min,
+                                   uint32_t max )
+{
+    if( max >= min )    // no wrap_around => only one window [min,max]
+    {
+        return( (seq >= min) && (seq <= max) ); 
+    }
+    else                // window wrap-around => two windows [min,0xFFFFFFFF] and [0,max]
+    {
+        return( (seq <= max) || (seq >= min) );
+    }
+}  
+
+////////////////////////////////////////////////////////////////////////////////////////////
+// this static function compute a channel index in range [0,nic_channelx[ from
+// a remote IP address and remote port.
+// TODO this function should be provided by the NIC driver.
+////////////////////////////////////////////////////////////////////////////////////////////
+// @ addr     : [in] IP address.
+// @ port     : [in] TCP/UDP port.
+////////////////////////////////////////////////////////////////////////////////////////////
+static inline uint32_t dev_nic_channel_index( uint32_t addr,
+                                              uint16_t port )
+{
+    // get number of NIC channels
+    uint32_t nic_channels = LOCAL_CLUSTER->nb_nic_channels;
+ 
+    // compute NIC channel index
+    return ( ((addr     ) & 0xFF) ^
+             ((addr > 8 ) & 0xFF) ^
+             ((addr > 16) & 0xFF) ^
+             ((addr > 24) & 0xFF) ^
+             ((port     ) & 0xFF) ^
+             ((port > 8 ) & 0xFF) ) % nic_channels;
+}
+
+////////////////////////////////////////////////////////////////////////////////////////
+// This static function computes the checksum for an IP packet header.
+// The "checksum" field itself is not taken into account for this computation.
+////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer      : [in] pointer on IP packet header (20 bytes)
+// @ return the checksum value on 16 bits
+////////////////////////////////////////////////////////////////////////////////////////
+uint16_t dev_nic_ip_checksum( uint8_t  * buffer )
+{
+    uint32_t   i;           
+    uint32_t   cs;      // 32 bits accumulator
+    uint16_t * buf;     
+    
+    buf = (uint16_t *)buffer;
+
+    // compute checksum 
+    for( i = 0 , cs = 0 ; i < 10 ; i++ )
+    {
+        if( i != 5 )  cs += buf[i];
+    }
+
+    // one's complement
+    return ~cs;
+}
+
+////////////////////////////////////////////////////////////////////////////////////////
+// This static function computes the checksum for an UDP packet defined by 
+// the <buffer> and <size> arguments.
+////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer      : [in] pointer on UDP packet base.
+// @ size        : [in] number of bytes in this packet (including header).
+// @ return the checksum value on 16 bits
+////////////////////////////////////////////////////////////////////////////////////////
+uint16_t dev_nic_udp_checksum( uint8_t  * buffer,
+                               uint32_t   size )
+{
+    uint32_t   i;           
+    uint32_t   carry;
+    uint32_t   cs;      // 32 bits accumulator
+    uint16_t * buf;     
+    uint32_t   max;     // number of uint16_t in packet
+    
+    // compute max & buf
+    buf = (uint16_t *)buffer;
+    max = size >> 1;
+
+    // extend buffer[] if required
+    if( size & 1 )
+    {
+        max++;
+        buffer[size] = 0;
+    }
+
+    // compute checksum for UDP packet
+    for( i = 0 , cs = 0 ; i < size ; i++ )  cs += buf[i];
+
+    // handle carry
+    carry = (cs >> 16);
+    if( carry ) 
+    {
+        cs += carry; 
+        carry = (cs >> 16);
+        if( carry ) cs += carry;
+    } 
+
+    // one's complement
+    return ~cs;
+}
+
+////////////////////////////////////////////////////////////////////////////////////////
+// This static function computes the checksum for a TCP segment defined by
+// the <buffer> and <size> arguments.
+// It includes the pseudo header defined by the <src_ip_addr>, <dst_ip_addr>,
+// <size> arguments, and by the TCP_PROTOCOL code.
+////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer      : [in] pointer on TCP segment base.
+// @ size        : [in] number of bytes in this segment (including header).
+// @ src_ip_addr : [in] source IP address (pseudo header)
+// @ dst_ip_addr : [in] destination IP address (pseudo header)
+// @ return the checksum value on 16 bits
+////////////////////////////////////////////////////////////////////////////////////////
+uint16_t dev_nic_tcp_checksum( uint8_t  * buffer,
+                               uint32_t   size,
+                               uint32_t   src_ip_addr,
+                               uint32_t   dst_ip_addr )
+{
+    uint32_t   i;           
+    uint32_t   carry;
+    uint32_t   cs;      // 32 bits accumulator
+    uint16_t * buf;
+    uint32_t   max;     // number of uint16_t in segment
+
+    // compute max & buf
+    buf = (uint16_t *)buffer;
+    max = size >> 1;
+
+    // extend buffer[] if required
+    if( size & 1 )
+    {
+        max++;
+        buffer[size] = 0;
+    }
+
+    // compute checksum for TCP segment
+    for( i = 0 , cs = 0 ; i < size ; i++ )  cs += buf[i];
+
+    // complete checksum for pseudo-header
+    cs += src_ip_addr;
+    cs += dst_ip_addr;
+    cs += PROTOCOL_TCP;
+    cs += size;
+
+    // handle carry
+    carry = (cs >> 16);
+    if( carry ) 
+    {
+        cs += carry; 
+        carry = (cs >> 16);
+        if( carry ) cs += carry;
+    } 
+
+    // one's complement
+    return ~cs;
+}
 
 //////////////////////////////////
@@ -80 +257 @@
 }  // end dev_nic_init()
 
+
+/////////////////////////////////////////////////////////////////////////////////////////
+//                 Functions implementing the SOCKET related syscalls
+/////////////////////////////////////////////////////////////////////////////////////////
+
+//////////////////////////////////////
+int dev_nic_socket( uint32_t   domain,
+                    uint32_t   type )
+{
+    uint32_t    fdid;
+    socket_t  * socket;
+    error_t     error;
+
+    // allocate memory for the file descriptor and for the socket
+    error = socket_create( local_cxy,
+                           domain,
+                           type,
+                           &socket,    // unused here
+                           &fdid );
+
+    if( error ) return -1;
+    return fdid;
+}
+
+////////////////////////////////
+int dev_nic_bind( uint32_t fdid,
+                  uint32_t addr,
+                  uint16_t port )
+{
+    vfs_inode_type_t    type;
+    socket_t          * socket;
+    uint32_t            state;
+
+    thread_t  * this    = CURRENT_THREAD;
+    process_t * process = this->process;
+
+    // get pointers on file descriptor
+    xptr_t       file_xp  = process_fd_get_xptr( process , fdid );
+    vfs_file_t * file_ptr = GET_PTR( file_xp );
+    cxy_t        file_cxy = GET_CXY( file_xp );
+
+    // check file_xp
+    if( file_xp == XPTR_NULL )
+    {
+        printk("\n[ERROR] in %s : undefined fdid %d",
+        __FUNCTION__, fdid );
+        return -1;
+    }
+
+    type   = hal_remote_l32( XPTR( file_cxy , &file_ptr->type ) );
+    socket = hal_remote_lpt( XPTR( file_cxy , &file_ptr->socket ) );
+
+    // check file descriptor type
+    if( type != INODE_TYPE_SOCK )
+    {
+        printk("\n[ERROR] in %s : illegal file type %s",
+        __FUNCTION__, vfs_inode_type_str( type ) );
+        return -1;
+    }
+
+    state = (type == SOCK_STREAM) ? TCP_STATE_BOUND : UDP_STATE_BOUND;
+
+    // update the socket descriptor
+    hal_remote_s32( XPTR( file_cxy , &socket->local_addr ) , addr );
+    hal_remote_s32( XPTR( file_cxy , &socket->local_port ) , port );
+    hal_remote_s32( XPTR( file_cxy , &socket->state      ) , state );
+
+    return 0;
+
+}  // end dev_nic_bind()
+
+//////////////////////////////////
+int dev_nic_listen( uint32_t fdid,
+                    uint32_t max_pending )
+{
+    xptr_t              file_xp;
+    vfs_file_t        * file_ptr;
+    cxy_t               file_cxy;
+    vfs_inode_type_t    file_type;
+    socket_t          * socket_ptr;
+    uint32_t            socket_type;
+    uint32_t            socket_state;
+
+    thread_t  * this    = CURRENT_THREAD;
+    process_t * process = this->process;
+
+    if( max_pending != 0 )
+    {
+        printk("\n[WARNING] in %s : max_pending argument non supported\n",
+        __FUNCTION__ );
+    }
+
+    // get pointers on file descriptor
+    file_xp  = process_fd_get_xptr( process , fdid );
+    file_ptr = GET_PTR( file_xp );
+    file_cxy = GET_CXY( file_xp );
+
+    // check file_xp
+    if( file_xp == XPTR_NULL )
+    {
+        printk("\n[ERROR] in %s : undefined fdid %d",
+        __FUNCTION__, fdid );
+        return -1;
+    }
+
+    file_type  = hal_remote_l32( XPTR( file_cxy , &file_ptr->type ) );
+    socket_ptr = hal_remote_lpt( XPTR( file_cxy , &file_ptr->socket ) );
+
+    // check file descriptor type
+    if( file_type != INODE_TYPE_SOCK )
+    {
+        printk("\n[ERROR] in %s : illegal file type %s",
+        __FUNCTION__, vfs_inode_type_str(file_type) );
+        return -1;
+    }
+
+    // get socket type and state
+    socket_type  = hal_remote_l32( XPTR( file_cxy , &socket_ptr->type )); 
+    socket_state = hal_remote_l32( XPTR( file_cxy , &socket_ptr->state )); 
+
+    // check socket type 
+    if( socket_type != SOCK_STREAM )
+    {
+        printk("\n[ERROR] in %s : illegal socket type",
+        __FUNCTION__ );
+        return -1;
+    }
+    
+    // check socket state
+    if( socket_state != TCP_STATE_BOUND )
+    {
+        printk("\n[ERROR] in %s : illegal socket state %s",
+        __FUNCTION__, socket_state_str(socket_state) );
+        return -1;
+    }
+    
+    // update socket.state
+    hal_remote_s32( XPTR( file_cxy , &socket_ptr->state ) , TCP_STATE_LISTEN );
+
+    return 0;
+
+}  // end dev_nic_listen()
+
 ///////////////////////////////////
-error_t dev_nic_read( pkd_t * pkd )
-{
-    error_t  error;
-
-    // get pointers on this NIC-RX kernel thread 
-    thread_t * thread_ptr = CURRENT_THREAD;
-    xptr_t     thread_xp  = XPTR( local_cxy , thread_ptr );
-
-    // get local pointer on core running this kernel thead
-    core_t * core = thread_ptr->core;
-
-// check thread can yield
-assert( (thread_ptr->busylocks == 0),
-"cannot yield : busylocks = %d\n", thread_ptr->busylocks );
+int dev_nic_connect( uint32_t fdid,
+                     uint32_t remote_addr,
+                     uint16_t remote_port )
+{
+    vfs_inode_type_t    file_type;
+    socket_t          * socket;
+    uint32_t            socket_state;     // socket state
+    uint32_t            socket_type;      // socket type  
+    uint32_t            local_addr;       // local IP address
+    uint32_t            local_port;       // local port
+    xptr_t              tx_server_xp;     // extended pointer on TX server thread
+    thread_t          * tx_server_ptr;    // local pointer on TX server thread
+
+    thread_t  * this    = CURRENT_THREAD;
+    process_t * process = this->process;
+
+    // get pointers on file descriptor
+    xptr_t       file_xp  = process_fd_get_xptr( process , fdid );
+    vfs_file_t * file_ptr = GET_PTR( file_xp );
+    cxy_t        file_cxy = GET_CXY( file_xp );
+
+    // check file_xp
+    if( file_xp == XPTR_NULL )
+    {
+        printk("\n[ERROR] in %s : undefined fdid %d",
+        __FUNCTION__, fdid );
+        return -1;
+    }
+
+    file_type = hal_remote_l32( XPTR( file_cxy , &file_ptr->type ) );
+    socket    = hal_remote_lpt( XPTR( file_cxy , &file_ptr->socket ) );
+
+    // check file descriptor type
+    if( file_type != INODE_TYPE_SOCK )
+    {
+        printk("\n[ERROR] in %s : illegal file type %s",
+        __FUNCTION__, vfs_inode_type_str( file_type ) );
+        return -1;
+    }
+
+    // get relevant socket infos
+    socket_type   = hal_remote_l32( XPTR( file_cxy , &socket->type ) );
+    socket_state  = hal_remote_l32( XPTR( file_cxy , &socket->state ) );
+    local_addr    = hal_remote_l32( XPTR( file_cxy , &socket->local_addr ) );
+    local_port    = hal_remote_l32( XPTR( file_cxy , &socket->local_port ) );
+
+    if( socket_type == SOCK_DGRAM )       // UDP
+    {
+        if( socket_state != UDP_STATE_BOUND )
+        {
+            printk("\n[ERROR] in %s : illegal socket statea %s for CONNECT",
+            __FUNCTION__, socket_state_str(socket_state) );
+            return -1;
+        }
+    }
+    else if( socket_type == SOCK_STREAM )  // TCP
+    {
+        if( socket_state != TCP_STATE_BOUND )
+        {
+            printk("\n[ERROR] in %s : illegal socket state %s for CONNECT",
+            __FUNCTION__, socket_state_str(socket_state) );
+            return -1;
+        }
+    }
+    else
+    {
+        printk("\n[ERROR] in %s : illegal socket type %d for CONNECT",
+        __FUNCTION__, socket_type );
+        return -1;
+    }
+
+    // compute nic_channel index from remote_addr and remote_port
+    uint32_t nic_channel = dev_nic_channel_index( remote_addr , remote_port );
+
+    // link new socket to chdev servers
+    socket_link_to_servers( XPTR( file_cxy , socket ),
+                            nic_channel );
+
+    // update the socket descriptor
+    hal_remote_s32( XPTR( file_cxy , &socket->remote_addr ) , remote_addr  );
+    hal_remote_s32( XPTR( file_cxy , &socket->remote_port ) , remote_port  );
+    hal_remote_s32( XPTR( file_cxy , &socket->nic_channel ) , nic_channel  );
+
+    // the actual connection mechanism depends on socket type
+    // UDP : client thread directly updates the local socket state
+    // TCP : client thread request TX server thread to start the 3 steps handshake
+
+    if( socket_type == SOCK_DGRAM )  // UDP
+    {
+        // directly update the local socket state
+        hal_remote_s32( XPTR( file_cxy , &socket->state ) , UDP_STATE_CONNECT );
+    }
+    else                             // TCP
+    {
+        // get pointers on NIC_TX[index] chdev
+        xptr_t    tx_chdev_xp  = chdev_dir.nic_tx[nic_channel];
+        chdev_t * tx_chdev_ptr = GET_PTR( tx_chdev_xp );
+        cxy_t     tx_chdev_cxy = GET_CXY( tx_chdev_xp );
+
+        // get pointers on NIC_TX[channel] server thread
+        tx_server_ptr = hal_remote_lpt( XPTR( tx_chdev_cxy , &tx_chdev_ptr->server ));
+        tx_server_xp  = XPTR( tx_chdev_cxy , tx_server_ptr );
+
+        // register command arguments in socket descriptor
+        hal_remote_s64( XPTR( file_cxy , &socket->tx_cmd ),
+                        SOCKET_TX_CONNECT );
+
+        // update the "tx_client" field in socket descriptor
+        hal_remote_s64( XPTR( file_cxy , &socket->tx_client ),
+                        XPTR( local_cxy , this ) );
+
+        // unblock NIC_TX server thread
+        thread_unblock( tx_server_xp , THREAD_BLOCKED_CLIENT );
+ 
+        // block on THREAD_BLOCKED_IO condition and deschedules
+        thread_block( XPTR( local_cxy , this ) , THREAD_BLOCKED_IO );
+        sched_yield( "blocked in connect" );
+
+        // reset the "tx_client" field in socket descriptor
+        hal_remote_s64( XPTR( file_cxy , &socket->tx_client ),
+                        XPTR_NULL );
+    }
+
+    return 0;
+
+}  // end dev_nic_connect()
+
+////////////////////////////////////
+int dev_nic_accept( uint32_t   fdid,
+                    uint32_t * remote_addr,
+                    uint16_t * remote_port )
+{
+    xptr_t              file_xp;             // extended pointer on remote file
+    vfs_file_t        * file_ptr;
+    cxy_t               file_cxy;
+    vfs_inode_type_t    file_type;           // file descriptor type
+    socket_t          * socket;              // local pointer on remote waiting socket
+    uint32_t            socket_type;         // waiting socket type    
+    uint32_t            socket_state;        // waiting socket state
+    uint32_t            socket_domain;       // waiting socket domain
+    uint32_t            socket_local_addr;   // waiting socket local IP address
+    uint32_t            socket_local_port;   // waiting socket local port
+    xptr_t              crqq_xp;             // extended pointer on socket.crqq queue
+    socket_t          * new_socket;          // local pointer on new socket
+    uint32_t            new_fdid;            // new socket file descriptor index
+    sockaddr_t          new_sockaddr;        // one request in crqq queue
+    uint32_t            new_remote_addr;     // new socket remote IP address
+    uint32_t            new_remote_port;     // new socket remote port
+    error_t             error;
+
+    thread_t  * this    = CURRENT_THREAD;
+    process_t * process = this->process;
+
+    // get pointers on file descriptor
+    file_xp  = process_fd_get_xptr( process , fdid );
+    file_ptr = GET_PTR( file_xp );
+    file_cxy = GET_CXY( file_xp );
+
+    // check file_xp
+    if( file_xp == XPTR_NULL )
+    {
+        printk("\n[ERROR] in %s : undefined fdid %d",
+        __FUNCTION__, fdid );
+        return -1;
+    }
+
+    file_type = hal_remote_l32( XPTR( file_cxy , &file_ptr->type ) );
+    socket    = hal_remote_lpt( XPTR( file_cxy , &file_ptr->socket ) );
+
+    // check file descriptor type
+    if( file_type != INODE_TYPE_SOCK )
+    {
+        printk("\n[ERROR] in %s : illegal file type %s / thread[%x,%x]\n",
+        __FUNCTION__, vfs_inode_type_str(file_type), process->pid, this->trdid );
+        return -1;
+    }
+
+    // get socket type, domain, state, local_addr and local_port
+    socket_type       = hal_remote_l32( XPTR( file_cxy , &socket->type )); 
+    socket_state      = hal_remote_l32( XPTR( file_cxy , &socket->state )); 
+    socket_domain     = hal_remote_l32( XPTR( file_cxy , &socket->domain )); 
+    socket_local_addr = hal_remote_l32( XPTR( file_cxy , &socket->local_addr )); 
+    socket_local_port = hal_remote_l32( XPTR( file_cxy , &socket->local_port )); 
+
+    // check socket type 
+    if( socket_type != SOCK_STREAM )
+    {
+        printk("\n[ERROR] in %s : illegal socket type / thread[%x,%x]\n",
+        __FUNCTION__, process->pid , this->trdid );
+        return -1;
+    }
+    
+    // check socket state
+    if( socket_state != TCP_STATE_LISTEN ) 
+    {
+        printk("\n[ERROR] in %s : illegal socket state %s / thread[%x,%x]\n",
+        __FUNCTION__, socket_state_str(socket_state), process->pid, this->trdid );
+        return -1;
+    }
+    
+    // select a cluster for the new socket
+    cxy_t new_cxy = cluster_random_select();
+
+    // allocate memory for the new socket descriptor 
+    error = socket_create( new_cxy,
+                           socket_domain,
+                           socket_type,
+                           &new_socket,
+                           &new_fdid );
+    if( error )
+    {
+        printk("\n[ERROR] in %s : cannot allocate new socket / thread[%x,%x]\n",
+        __FUNCTION__, process->pid, this->trdid );
+        return -1;
+    }
+   
+    // build extended pointer on socket.crqq
+    crqq_xp  = XPTR( file_cxy , &socket->crqq );
+
+    // blocks and deschedules if requests queue empty
+    if( remote_buf_status( crqq_xp ) == 0 )
+    {
+        thread_block( XPTR( local_cxy , this ) , THREAD_BLOCKED_IO );
+        sched_yield( "socket.crqq queue empty");
+    }
+        
+    // extract first request from the socket.crqq queue
+    remote_buf_get_to_kernel( crqq_xp,
+                              (uint8_t *)(&new_sockaddr),
+                              sizeof(sockaddr_t) );
+
+    new_remote_addr = new_sockaddr.s_addr;
+    new_remote_port = new_sockaddr.s_port;
+
+    // compute NIC channel index from remote_addr and remote_port
+    uint32_t nic_channel = dev_nic_channel_index( new_remote_addr , new_remote_port );
+
+    // update new socket descriptor 
+    new_socket->local_addr  = hal_remote_l32(XPTR( file_cxy , &socket->local_addr ));
+    new_socket->local_port  = hal_remote_l32(XPTR( file_cxy , &socket->local_port ));
+    new_socket->remote_addr = new_remote_addr;
+    new_socket->remote_port = new_remote_port;
+    new_socket->nic_channel = nic_channel;
+
+    // link new socket to chdev servers
+    socket_link_to_servers( XPTR( new_cxy , new_socket ),
+                            nic_channel );
+    // return success
+    *remote_addr = new_remote_addr;
+    *remote_port = new_remote_port;
+
+    return new_fdid;
+
+}  // end dev_nic_accept()
+
+////////////////////////////////////////////////////////////////////////////////////////
+// This static and blocking function is called by the four functions :
+// dev_nic_send() / dev_nic_recv() / dev_nic_sendto() / dev_nic_recvfrom().
+////////////////////////////////////////////////////////////////////////////////////////
+// Implementation note
+// The behavior is very different for SEND & RECV :
+// - For a SEND, the client thread checks that there is no TX command registered
+//   in the socket. It registers the command arguments in the socket descriptor
+//   (tx_client, tx_cmd, tx_buf, tx_len). Then the client thread unblocks the 
+//   TX server thread from the BLOCKED_CLIENT condition, blocks itself on the
+//   BLOCKED_IO condition, and deschedules. It is unblocked by the TX server thread
+//   when the last byte has been sent (for UDP) or acknowledged (for TCP).
+//   When the client thread resumes, it reset the command in socket, and returns.
+// - For a RECV, the client thread checks that there is no RX command registered
+//   in the socket. It registers itself in socket (rx_client). It checks the status
+//   of the receive buffer. It the rx_buf is empty, it blocks on the BLOCKED_IO
+//   condition, and deschedules. It is unblocked by the RX server thread when an UDP
+//   packet or TCP segment has been writen in the rx_buf. When it resumes, it moves
+//   the available data from the rx_buf to the user buffer, reset its registration
+//   in socket (reset the rx_buf for an UDP socket), and returns.
+////////////////////////////////////////////////////////////////////////////////////////
+int dev_nic_register_cmd( bool_t     is_send,
+                          uint32_t   fdid,
+                          uint8_t  * u_buf,
+                          uint32_t   length,
+                          bool_t     explicit,
+                          uint32_t   explicit_addr,
+                          uint32_t   explicit_port )
+{
+    vfs_inode_type_t    file_type;       // file descriptor type
+    socket_t          * socket_ptr;      // local pointer on socket descriptor
+    uint32_t            socket_state;    // current socket state
+    uint32_t            socket_type;     // socket type (UDP/TCP)
+    uint32_t            nic_channel;     // NIC channel for this socket
+    xptr_t              socket_lock_xp;  // extended pointer on socket lock
+    xptr_t              file_xp;         // extended pointer on file descriptor
+    vfs_file_t        * file_ptr;
+    cxy_t               file_cxy;
+    xptr_t              chdev_xp;        // extended pointer on NIC_TX[channel] chdev
+    chdev_t           * chdev_ptr;
+    cxy_t               chdev_cxy;
+    uint32_t            remote_addr;
+    uint32_t            remote_port;
+    uint32_t            status;          // number of bytes in rx_buf
+    int32_t             moved_bytes;     // total number of moved bytes (fot return)
+    xptr_t              server_xp;       // extended pointer on NIC_TX / NIC_RX server thread
+    thread_t          * server_ptr;      // local pointer on NIC_TX / NIC_RX server thread
+
+    thread_t  * this    = CURRENT_THREAD;
+    process_t * process = this->process;
+
+    // get pointers on file descriptor identifying the socket
+    file_xp  = process_fd_get_xptr( process , fdid );
+    file_ptr = GET_PTR( file_xp );
+    file_cxy = GET_CXY( file_xp );
+
+    if( file_xp == XPTR_NULL )
+    {
+        printk("\n[ERROR] in %s : undefined fdid %d / thread%x,%x]\n",
+        __FUNCTION__, fdid , process->pid, this->trdid );
+        return -1;
+    }
+ 
+    // get file type and socket pointer
+    file_type  = hal_remote_l32( XPTR( file_cxy , &file_ptr->type ) );
+
+    // get local pointer on socket
+    socket_ptr = hal_remote_lpt( XPTR( file_cxy , &file_ptr->socket ) );
+
+    // check file descriptor type
+    if( file_type != INODE_TYPE_SOCK )
+    {
+        printk("\n[ERROR] in %s : illegal file type %s / fdid %d / thread%x,%x]\n",
+        __FUNCTION__, vfs_inode_type_str(file_type), fdid, process->pid, this->trdid );
+        return -1;
+    }
+
+    // build extended pointer on file lock protecting socket 
+    socket_lock_xp = XPTR( file_cxy , &file_ptr->lock );
+
+    // take the socket lock
+    remote_rwlock_wr_acquire( socket_lock_xp );
+
+    // get socket type, state, and channel
+    socket_type  = hal_remote_l32( XPTR( file_cxy , &socket_ptr->type ));
+    socket_state = hal_remote_l32( XPTR( file_cxy , &socket_ptr->state ));
+    nic_channel  = hal_remote_l32( XPTR( file_cxy , &socket_ptr->nic_channel ));
+
+    // check socket state / type
+    if( socket_type == SOCK_STREAM )     // TCP socket
+    {
+        if( socket_state != TCP_STATE_ESTAB )
+        {
+            printk("\n[ERROR] in %s : illegal SEND/RECV for state %s / thread%x,%x]\n",
+            __FUNCTION__, socket_state_str(socket_state), process->pid, this->trdid );
+            return -1;
+        }
+
+        if( explicit )
+        {
+            // get remote IP address and type from socket descriptor
+            remote_addr = hal_remote_l32( XPTR( file_cxy , &socket_ptr->remote_addr ));
+            remote_port = hal_remote_l32( XPTR( file_cxy , &socket_ptr->remote_port ));
+
+            if( (remote_addr != explicit_addr) || (remote_port != explicit_port) )
+            {
+                printk("\n[ERROR] in %s : wrong expliciy access / thread%x,%x]\n",
+                __FUNCTION__, process->pid, this->trdid );
+                return -1;
+            }
+        }
+    }
+    else                              // UDP socket
+    {
+        if( explicit )
+        {
+            if( socket_state == UDP_STATE_UNBOUND )
+            {
+                printk("\n[ERROR] in %s : illegal SEND/RECV for state %s / thread%x,%x]\n",
+                __FUNCTION__, socket_state_str(socket_state), process->pid, this->trdid );
+                return -1;
+            }
+
+            // update remote IP address and port into socket descriptor
+            hal_remote_s32( XPTR( file_cxy , &socket_ptr->remote_addr ), explicit_addr );
+            hal_remote_s32( XPTR( file_cxy , &socket_ptr->remote_port ), explicit_port );
+        }
+        else
+        {
+            if( socket_state != UDP_STATE_CONNECT )
+            {
+                printk("\n[ERROR] in %s : illegal SEND/RECV for state %s / thread%x,%x]\n",
+                __FUNCTION__, socket_state_str(socket_state), process->pid, this->trdid );
+                return -1;
+            }
+        }
+    }
+
+    ///////////////////////////////////////////////////////
+    if( is_send )                       // SEND command
+    {
+        // build extended pointer on socket "tx_client"
+        xptr_t client_xp = XPTR( file_cxy , &socket_ptr->tx_client );
+
+        // check no previous SEND command
+        xptr_t client = hal_remote_l64( client_xp );
+
+        if( client != XPTR_NULL )  // release socket lock and return error
+        {
+            // release socket lock
+            remote_rwlock_wr_release( socket_lock_xp );
+                    
+            // get previous thread cluster & local pointer
+            cxy_t      prev_cxy = GET_CXY( client );
+            thread_t * prev_ptr = GET_PTR( client );
+
+            // get previous command type and trdid 
+            uint32_t prev_cmd = hal_remote_l32( XPTR( prev_cxy , &prev_ptr->nic_cmd.type )); 
+            uint32_t prev_tid = hal_remote_l32( XPTR( prev_cxy , &prev_ptr->trdid ));
+
+            printk("\n[ERROR] in %s : previous command %s for thread %x / thread%x,%x]\n",
+            __FUNCTION__, socket_cmd_str(prev_cmd), prev_tid,
+            process->pid, this->trdid );
+
+            return -1;
+        }
+
+        // client thread registers in socket descriptor
+        hal_remote_s64( client_xp , XPTR( local_cxy , this ) );
+        hal_remote_s32( XPTR( file_cxy , &socket_ptr->tx_cmd  ) , SOCKET_TX_SEND );
+        hal_remote_spt( XPTR( file_cxy , &socket_ptr->tx_buf  ) , u_buf );
+        hal_remote_s32( XPTR( file_cxy , &socket_ptr->tx_len  ) , length );
+        hal_remote_s32( XPTR( file_cxy , &socket_ptr->tx_todo ) , length );
+
+        // release socket lock
+        remote_rwlock_wr_release( socket_lock_xp );
+                    
+        // get pointers on relevant chdev
+        chdev_xp  = chdev_dir.nic_tx[nic_channel];
+        chdev_ptr = GET_PTR( chdev_xp );
+        chdev_cxy = GET_CXY( chdev_xp );
+
+        // get pointers on NIC_TX[channel] server thread
+        server_ptr = hal_remote_lpt( XPTR( chdev_cxy , &chdev_ptr->server ));
+        server_xp  = XPTR( chdev_cxy , server_ptr );
+
+        // unblocks the NIC_TX server thread
+        thread_unblock( server_xp , THREAD_BLOCKED_CLIENT );
+
+        // client thread blocks itself and deschedules
+        thread_block( XPTR( local_cxy , this ) , THREAD_BLOCKED_IO );
+        sched_yield( "blocked in nic_io" );
+
+        // take the socket lock when unblocked
+        remote_rwlock_wr_acquire( socket_lock_xp );
+
+        // unlink client thread from socket 
+        hal_remote_s64( client_xp , XPTR_NULL );
+                    
+        // release socket lock
+        remote_rwlock_wr_release( socket_lock_xp );
+                    
+        // exit waiting loop and return
+        return length;
+
+    }  // end SEND
+
+    ////////////////////////////////////////////////////////
+    else                                 // RECV command
+    {
+        // build extended pointers on socket "rx_client"
+        xptr_t client_xp = XPTR( file_cxy , &socket_ptr->rx_client );
+
+        // check no previous RECV command
+        xptr_t client = hal_remote_l64( client_xp );
+
+        if( client != XPTR_NULL )  // release socket lock and return error
+        {
+            // release socket lock
+            remote_rwlock_wr_release( socket_lock_xp );
+                    
+            // get previous thread cluster & local pointer
+            cxy_t      prev_cxy = GET_CXY( client );
+            thread_t * prev_ptr = GET_PTR( client );
+
+            // get previous command type and trdid 
+            uint32_t prev_cmd = hal_remote_l32( XPTR( prev_cxy , &prev_ptr->nic_cmd.type )); 
+            uint32_t prev_tid = hal_remote_l32( XPTR( prev_cxy , &prev_ptr->trdid ));
+
+            printk("\n[ERROR] in %s : previous command %s for thread %x / thread%x,%x]\n",
+            __FUNCTION__, socket_cmd_str(prev_cmd), prev_tid, process->pid, this->trdid );
+            return -1;
+        }
+
+        // build extended pointer on "rx_buf"
+        xptr_t rx_buf_xp = XPTR( file_cxy , &socket_ptr->rx_buf );
+
+        // get rx_buf status from socket
+        status = remote_buf_status( rx_buf_xp );
+
+        if( status == 0 )    // rx_buf empty => blocks and deschedules
+        { 
+            // release socket lock
+            remote_rwlock_wr_release( socket_lock_xp );
+                 
+            // client thread blocks itself and deschedules
+            thread_block( XPTR( local_cxy , this ) , THREAD_BLOCKED_IO );
+            sched_yield( "blocked in nic_io" );
+
+            // take socket lock
+            remote_rwlock_wr_release( socket_lock_xp );
+        }
+
+        // number of moved bytes cannot be larger than u_buf size
+        moved_bytes = ( length < status ) ? length : status;
+
+        // move data from kernel rx_buf to user u_buf
+        remote_buf_get_to_user( rx_buf_xp,
+                                 u_buf,
+                                 moved_bytes );
+
+        // reset rx_buf for an UDP socket
+        if( socket_type == SOCK_DGRAM ) remote_buf_reset( rx_buf_xp );
+
+        // unlink client thread from socket 
+        hal_remote_s64( client_xp , XPTR_NULL );
+                    
+        // release socket lock
+        remote_rwlock_wr_release( socket_lock_xp );
+                    
+        // exit waiting loop and return
+        return moved_bytes;
+
+    }  // end SEND  
+
+} // end dev_nic_register_cmd()
+
+
+///////////////////////////////////
+int dev_nic_send( uint32_t    fdid,
+                  uint8_t   * u_buf,
+                  uint32_t    length )
+{
+#if DEBUG_DEV_NIC_TX
+thread_t  * this    = CURRENT_THREAD;
+process_t * process = this->process;
+trdid_t     trdid   = this->trdid;
+pid_t       pid     = process->pid;
+uint32_t cycle = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_TX < cycle )
+printk("[%s] thread[%x,%x] enters : fdid %d / buf %x / length %d / cycle %d\n",
+__FUNCTION__, pid, trdid, fdid, u_buf, length, cycle );
+#endif
+
+    error_t error = dev_nic_register_cmd( true,           // SEND
+                                          fdid,
+                                          u_buf,
+                                          length,
+                                          false, 0, 0 );  // no explicit remote socket
+#if DEBUG_DEV_NIC_TX
+cycle = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_TX < cycle )
+printk("[%s] thread[%x,%x] exit : fdid %d / cycle %d\n",
+__FUNCTION__, pid, trdid, cycle );
+#endif
+
+    return error;
+
+}  // end dev_nic_send()
+
+///////////////////////////////////
+int dev_nic_recv( uint32_t    fdid,
+                  uint8_t   * u_buf,
+                  uint32_t    length )
+{
+#if DEBUG_DEV_NIC_RX
+thread_t  * this    = CURRENT_THREAD;
+process_t * process = this->process;
+trdid_t     trdid   = this->trdid;
+pid_t       pid     = process->pid;
+uint32_t    cycle   = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_RX < cycle )
+printk("[%s] thread[%x,%x] enters : fdid %d / buf %x / length %d / cycle %d\n",
+__FUNCTION__, pid, trdid, fdid, u_buf, length, cycle );
+#endif
+
+    error_t error = dev_nic_register_cmd( false,          // RECV
+                                          fdid,
+                                          u_buf,
+                                          length,
+                                          false, 0, 0 );  // no explicit remote socket
+#if DEBUG_DEV_NIC_RX
+cycle = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_RX < cycle )
+printk("[%s] thread[%x,%x] exit : fdid %d / cycle %d\n",
+__FUNCTION__, pid, trdid, cycle );
+#endif
+
+    return error;
+
+} // end dev_nic_recv()
+
+/////////////////////////////////////
+int dev_nic_sendto( uint32_t    fdid,
+                    uint8_t   * u_buf,
+                    uint32_t    length,
+                    uint32_t    remote_addr,
+                    uint32_t    remote_port )
+{
+#if DEBUG_DEV_NIC_TX
+thread_t  * this    = CURRENT_THREAD;
+process_t * process = this->process;
+trdid_t     trdid   = this->trdid;
+pid_t       pid     = process->pid;
+uint32_t cycle = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_TX < cycle )
+printk("[%s] thread[%x,%x] enters : fdid %d / buf %x / length %d / cycle %d\n",
+__FUNCTION__, pid, trdid, fdid, u_buf, length, cycle );
+#endif
+
+    error_t error = dev_nic_register_cmd( true,          // SEND
+                                          fdid,
+                                          u_buf,
+                                          length,
+                                          true,          // explicit remote socket
+                                          remote_addr,
+                                          remote_port );
+#if DEBUG_DEV_NIC_TX
+cycle = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_TX < cycle )
+printk("[%s] thread[%x,%x] exit : fdid %d / cycle %d\n",
+__FUNCTION__, pid, trdid, cycle );
+#endif
+
+    return error;
+
+}  // end dev_nic_sendto()
+
+///////////////////////////////////////
+int dev_nic_recvfrom( uint32_t    fdid,
+                      uint8_t   * u_buf,
+                      uint32_t    length,
+                      uint32_t    remote_addr,
+                      uint32_t    remote_port )
+{
+#if DEBUG_DEV_NIC_RX
+thread_t  * this    = CURRENT_THREAD;
+process_t * process = this->process;
+trdid_t     trdid   = this->trdid;
+pid_t       pid     = process->pid;
+uint32_t    cycle   = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_RX < cycle )
+printk("[%s] thread[%x,%x] enters : fdid %d / buf %x / length %d / cycle %d\n",
+__FUNCTION__, pid, trdid, fdid, u_buf, length, cycle );
+#endif
+
+    error_t error = dev_nic_register_cmd( false,         // RECV
+                                          fdid,
+                                          u_buf,
+                                          length,
+                                          true,          // explicit remote socket
+                                          remote_addr,
+                                          remote_port );
+#if DEBUG_DEV_NIC_RX
+cycle = (uint32_t)hal_get_cycle();
+if (DEBUG_DEV_NIC_RX < cycle )
+printk("[%s] thread[%x,%x] exit : fdid %d / cycle %d\n",
+__FUNCTION__, pid, trdid, cycle );
+#endif
+
+    return error;
+
+}  // end dev_nic_recvfrom()
+
+
+
+
+
+
+
+
+///////////////////////////////////////////////////////////////////////////////////////////
+//               Functions called by the NIC_RX server thread
+///////////////////////////////////////////////////////////////////////////////////////////
+
+/////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the NIC_RX[channel] server thread to register
+// a send request defined by the <flags> argument in the R2T queue  specified by 
+// the <queue_xp> argument.
+/////////////////////////////////////////////////////////////////////////////////////////
+// @ queue_xp   : [in] extended pointer on the R2T qeue descriptor.
+// @ flags      : [in] flags to be set in the TCP segment.
+/////////////////////////////////////////////////////////////////////////////////////////
+static void dev_nic_rx_put_r2t_request( xptr_t    queue_xp,
+                                        uint32_t  flags )
+{
+    while( 1 )
+    {
+        error_t error = remote_buf_put_from_kernel( queue_xp,
+                                                    (uint8_t *)(&flags),
+                                                    1 );
+
+        if( error )  sched_yield( "waiting R2T queue" );
+        else         break;
+    }
+
+}  // end dev_nic_rx_put_r2t_request()
+  
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_rx_server() function.
+// It calls directly the NIC driver (with the READABLE command) and returns the status
+// of the NIC_RX queue identified by the <chdev> argument.
+// in the <readable> buffer.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ chdev     : [in]  local pointer on NIC_TX chdev.
+// @ readable  : [out] zero if queue empty.  
+// @ returns 0 if success / returns -1 if failure in accessing NIC device.
+///////////////////////////////////////////////////////////////////////////////////////////
+error_t dev_nic_rx_queue_readable( chdev_t  * chdev,
+                                   uint32_t * readable )
+{
+    thread_t * this = CURRENT_THREAD;
+
+    // initialize NIC_READABLE command in thread descriptor
+    this->nic_cmd.dev_xp = XPTR( local_cxy , chdev );
+    this->nic_cmd.type   = NIC_CMD_READABLE;
+
+    // call driver to test readable
+    chdev->cmd( XPTR( local_cxy , this ) );
+
+    // return status
+    *readable = this->nic_cmd.status;
+
+    // return error
+    return this->nic_cmd.error;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_rx_server() function.
+// It moves one Ethernet packet from the NIC_RX_QUEUE identified the <chdev> argument,
+// to the 2K bytes kernel buffer identified by the <buffer> argument. The actual
+// Ethernet packet length is returned in the <length> argument.
+// It calls directly the NIC driver with the READ command, without registering in the
+// waiting queue, because only the NIC_RX server thread can access this NIC_RX_QUEUE.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ chdev   : [in]  local pointer on NIC_TX chdev.
+// @ buffer  : [in]  local pointer on destination kernel buffer.
+// @ length  : [out] Ethernet packet size in bytes.
+// @ returns 0 if success / returns -1 if failure in accessing NIC device.
+///////////////////////////////////////////////////////////////////////////////////////////
+error_t dev_nic_rx_move_packet( chdev_t  * chdev,
+                                uint8_t  * k_buf,
+                                uint32_t * length )
+{
+    thread_t * this = CURRENT_THREAD;
 
 #if DEBUG_DEV_NIC_RX
 uint32_t cycle = (uint32_t)hal_get_cycles();
 if( DEBUG_DEV_NIC_RX < cycle )
-printk("\n[DBG] %s : thread %x enters for packet %x in cluster %x\n", 
-__FUNCTION__ , thread_ptr , pkd , local_cxy );
+printk("\n[%s] thread[%x,%x] enters / cycle %d\n", 
+__FUNCTION__, this->process->pid, this->trdid, cycle );
 #endif
 
-    // get pointer on NIC-RX chdev descriptor
-    uint32_t   channel = thread_ptr->chdev->channel; 
-    xptr_t     dev_xp  = chdev_dir.nic_rx[channel];
-    cxy_t      dev_cxy = GET_CXY( dev_xp );
-    chdev_t  * dev_ptr = (chdev_t *)GET_PTR( dev_xp );
-
-    assert( (dev_xp != XPTR_NULL) , "undefined NIC chdev descriptor" );
-
-    assert( (dev_cxy == local_cxy) , " chdev must be local" );
-
-    // initialize command in thread descriptor
-    thread_ptr->nic_cmd.dev_xp = dev_xp;
-
-    // call driver to test readable
-    thread_ptr->nic_cmd.cmd = NIC_CMD_READABLE;
-    dev_ptr->cmd( thread_xp );
+    // initialize NIC_READ command in thread descriptor
+    this->nic_cmd.type    = NIC_CMD_READ;
+    this->nic_cmd.buffer  = k_buf;
+
+    // call NIC driver  
+    chdev->cmd( XPTR( local_cxy , this ) );
+
+    // returns packet length   
+    *length = this->nic_cmd.length;
 
     // check error
-    error = thread_ptr->nic_cmd.error;
-    if( error ) return error;
-
-    // block and deschedule if queue non readable
-    if( thread_ptr->nic_cmd.status == false )  
-    {
-        // enable NIC-RX IRQ 
-        dev_pic_enable_irq( core->lid , dev_xp );
-
-        // block client thread on THREAD_BLOCKED_IO
-        thread_block( XPTR( local_cxy , thread_ptr ) , THREAD_BLOCKED_IO );
-
-        // deschedule client thread
-        sched_yield("client blocked on I/O");
-
-        // disable NIC-RX IRQ 
-        dev_pic_disable_irq( core->lid , dev_xp );
-    }
-
-    // call driver for actual read 
-    thread_ptr->nic_cmd.cmd     = NIC_CMD_READ;
-    thread_ptr->nic_cmd.buffer  = pkd->buffer;
-    dev_ptr->cmd( thread_xp );
-
-    // check error
-    error = thread_ptr->nic_cmd.error;
-    if( error ) return error;
-
-    // returns packet length   
-    pkd->length = thread_ptr->nic_cmd.length;
+    if( this->nic_cmd.error ) 
+    {
 
 #if DEBUG_DEV_NIC_RX
 cycle = (uint32_t)hal_get_cycles();
 if( DEBUG_DEV_NIC_RX < cycle )
-printk("\n[DBG] %s : thread %x exit for packet %x in cluster %x\n", 
-__FUNCTION__ , thread_ptr , pkd , local_cxy );
+printk("\n[%s] thread[%x,%x] exit / ERROR in NIC_RX / cycle %d\n", 
+__FUNCTION__, this->process->pid, this->trdid, cycle );
 #endif
 
+        return -1;
+    }
+    else
+    {
+
+#if DEBUG_DEV_NIC_RX
+cycle = (uint32_t)hal_get_cycles();
+if( DEBUG_DEV_NIC_RX < cycle )
+printk("\n[%s] thread[%x,%x] exit / SUCCESS / cycle %d\n", 
+__FUNCTION__, this->process->pid, this->trdid, cycle );
+#endif
+
+        return 0;
+    }
+
+}   // end dev_nic_rx_move_packet()
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_rx_server() function.
+// It analyses an Ethernet frame contained in the kernel buffer defined
+// by the <buffer> argument, and returns in the  <ip_length> argument the length
+// of the IP packet contained in the Ethernet packet payload.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer     : [in] pointer on a received Ethernet packet
+// @ ip_length  : [out] length of IP packet (in bytes).
+// @ return 0 if success / return -1 if illegal packet length. 
+///////////////////////////////////////////////////////////////////////////////////////////
+static error_t dev_nic_rx_check_eth( uint8_t  * buffer,
+                                     uint32_t * ip_length )
+{
+    uint32_t length = ((uint32_t)buffer[12] << 8) | (uint32_t)buffer[13];
+
+    *ip_length = length;
+
     return 0;
-
-}   // end dev_nic_read()
-
-
-////////////////////////////////////
-error_t dev_nic_write( pkd_t * pkd )
-{
-    error_t error;
-
-    // get pointers on the NIC-TX kernel tread 
-    thread_t * thread_ptr = CURRENT_THREAD;
-    xptr_t     thread_xp  = XPTR( local_cxy , thread_ptr );
-
-    // get local pointer on core running this kernel thead
-    core_t * core = thread_ptr->core;
+}
+   
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function analyses the IP packet contained in the kernel buffer
+// defined by the <buffer> argument, and returns in the <ip_src_addr>, <ip_dst_addr>, 
+// <header_length> and <protocol> arguments the informations contained in the IP header.
+// It checks the IP packet length versus the value contained in Ethernet header.
+// It checks the IP header checksum.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer          : [in] pointer on the IP packet.
+// @ expected_length : [in] expected IP packet length (from Ethernet header). 
+// @ ip_src_addr     : [out] source IP address.
+// @ ip_dst_addr     : [out] destination IP address.
+// @ protocol        : [out] transport protocol type.
+// @ return 0 if success / return -1 if illegal packet. 
+///////////////////////////////////////////////////////////////////////////////////////////
+static error_t dev_nic_rx_check_ip( uint8_t  * buffer,
+                                    uint32_t   expected_length, 
+                                    uint32_t * ip_src_addr, 
+                                    uint32_t * ip_dst_addr,
+                                    uint32_t * trsp_protocol )
+{
+    uint32_t length = ((uint32_t)buffer[2] << 8) | (uint32_t)buffer[3];
+
+    // discard packet if eth_length != ip_length
+    if( length != expected_length )
+    {
+
+#if DEBUG_NIC_DEV
+thread_t * this = CURRENT_THREAD;
+printk("\n[%s] thread[%x,%x] enters : length (%d) != expected_length (%d)\n",
+__FUNCTION__, this->process->pid, this->trdid, length, expected_length );
+#endif
+
+        return -1;
+    }
+
+    // compute IP header checksum 
+    uint32_t received_cs = (uint32_t)dev_nic_ip_checksum( buffer );
+
+    // extract IP header checksum
+    uint32_t computed_cs = ((uint32_t)buffer[10] << 8) | ((uint32_t)buffer[11]);
+
+    // discard packet if bad checksum
+    if( received_cs != computed_cs )
+    {
+
+#if DEBUG_NIC_DEV
+thread_t * this = CURRENT_THREAD;
+printk("\n[%s] thread[%x,%x] computed checksum (%d) != received checksum (%d)\n",
+__FUNCTION__, this->process->pid, this->trdid, computed_cs, received_cs );
+#endif
+
+        return -1;
+    }
+
+
+    *ip_src_addr = ((uint32_t)buffer[12] << 24) |
+                   ((uint32_t)buffer[13] << 16) |
+                   ((uint32_t)buffer[14] <<  8) |
+                   ((uint32_t)buffer[15]      ) ;
+
+    *ip_dst_addr = ((uint32_t)buffer[16] << 24) |
+                   ((uint32_t)buffer[17] << 16) |
+                   ((uint32_t)buffer[18] <<  8) |
+                   ((uint32_t)buffer[19]      ) ;
+
+    *trsp_protocol = (uint32_t)buffer[9];
+   
+    return 0;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function analyses the UDP packet contained in the kernel buffer
+// defined by the <k_buf> and <k_length> arguments.
+// It checks the UDP checksum, and discard corrupted packets.
+// It scans the list of sockets attached to the NIC_RX chdev to find a matching socket,
+// and discard the received packet if no UDP socket found.
+// Finally, it copies the payload to the socket "rx_buf", as long as the packet payload
+// is not larger than the rx_buf. 
+// It set the "rx_valid" flip-flop, and unblock the client thread when the last expected
+// byte has been received.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ chdev         : [in] local pointer on local NIC_RX chdev descriptor.
+// @ k_buf         : [in] pointer on the UDP packet in local kernel buffer.
+// @ k_length      : [in] number of bytes in buffer (including UDP header).
+// @ pkt_src_addr  : [in] source IP address (from IP packet header).
+// @ pkt_dst_addr  : [in] destination IP address (from IP packet header).
+///////////////////////////////////////////////////////////////////////////////////////////
+static void dev_nic_rx_handle_udp_packet( chdev_t  * chdev,
+                                          uint8_t  * k_buf,
+                                          uint32_t   k_length,
+                                          uint32_t   pkt_src_addr,
+                                          uint32_t   pkt_dst_addr )
+{
+    xptr_t     root_xp;           // extended pointer on attached sockets list root 
+    xptr_t     lock_xp;           // extended pointer on chdev lock 
+    xptr_t     iter_xp;           // iterator on socket list
+    xptr_t     socket_xp;         // extended pointer on socket descriptor
+    cxy_t      socket_cxy;
+    socket_t * socket_ptr;
+    uint32_t   socket_type;       // socket type
+    uint32_t   socket_state;      // socket state
+    uint32_t   local_addr;        // local IP address from socket
+    uint32_t   local_port;        // local port from socket
+    uint32_t   remote_addr;       // remote IP address from socket
+    uint32_t   remote_port;       // remote port from socket
+    bool_t     match_socket;      // matching socket found
+    uint16_t   checksum;          // computed checksum
+    uint16_t   pkt_checksum;      // received checksum
+    xptr_t     socket_rbuf_xp;    // extended pointer on socket rx_buf
+    xptr_t     socket_lock_xp;    // extended pointer on socket lock
+    xptr_t     socket_client_xp;  // extended pointer on socket rx_client field
+    xptr_t     client_xp;         // extended pointer on client thread descriptor
+    uint32_t   payload;           // number of bytes in payload
+    uint32_t   status;            // number of bytes in rx_buf
+    uint32_t   space;             // number of free slots in rx_buf
+    uint32_t   moved_bytes;       // number of bytes actually moved to rx_buf
+
+    // build extended pointers on list of sockets attached to NIC_RX chdev
+    root_xp = XPTR( local_cxy , &chdev->wait_root );
+    lock_xp = XPTR( local_cxy , &chdev->wait_lock );
+
+    // compute UDP packet checksum 
+    checksum = dev_nic_udp_checksum( k_buf , k_length );
+
+    // get checksum from received packet header 
+    pkt_checksum = ((uint16_t)k_buf[6] << 8) | (uint16_t)k_buf[7];
+
+    // discard corrupted packet  
+    if( pkt_checksum != checksum ) return;
+    
+    // get src_port and dst_port from UDP header
+    uint32_t pkt_src_port = ((uint32_t)k_buf[0] << 8) | (uint32_t)k_buf[1];
+    uint32_t pkt_dst_port = ((uint32_t)k_buf[2] << 8) | (uint32_t)k_buf[3];
+
+    // discard unexpected packet
+    if( xlist_is_empty( root_xp ) ) return;
+ 
+    // take the tock protecting the sockets list
+    remote_busylock_acquire( lock_xp );
+
+    match_socket = false;
+
+    // scan sockets list to find a match
+    XLIST_FOREACH( root_xp , iter_xp )
+    {
+        // get socket cluster and local pointer
+        socket_xp  = XLIST_ELEMENT( iter_xp , socket_t , rx_list );
+        socket_ptr = GET_PTR( socket_xp );
+        socket_cxy = GET_CXY( socket_xp );
+
+        // get socket type 
+        socket_type  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->type ));
+        socket_state = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->state ));
+                
+        // skip TCP socket
+        if( socket_type == SOCK_STREAM ) continue;
+
+        // get relevant info from socket descriptor
+        local_addr  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->local_addr )); 
+        remote_addr = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->remote_addr )); 
+        local_port  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->local_port )); 
+        remote_port = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->remote_port )); 
+
+        // compute matching
+        bool_t local_match  = (local_addr  == pkt_dst_addr) && 
+                              (local_port  == pkt_dst_port);
+
+        bool_t remote_match = (remote_addr == pkt_src_addr) &&
+                              (remote_port == pkt_src_port);
+
+        if (socket_state == UDP_STATE_CONNECT ) match_socket = local_match && remote_match;
+        else                                    match_socket = local_match;
+
+        // exit loop when socket found
+        if( match_socket ) break;
+    }
+
+    // release the lock protecting the sockets list
+    remote_busylock_release( lock_xp );
+
+    // discard unexpected packet
+    if( match_socket == false ) return;
+   
+    // build extended pointers on various socket fields
+    socket_rbuf_xp   = XPTR( socket_cxy , &socket_ptr->rx_buf );
+    socket_lock_xp   = XPTR( socket_cxy , &socket_ptr->lock ); 
+    socket_client_xp = XPTR( socket_cxy , &socket_ptr->rx_client );
+
+    // take the lock protecting the socket 
+    remote_rwlock_wr_acquire( socket_lock_xp );
+
+    // get status & space from rx_buf
+    status = remote_buf_status( socket_rbuf_xp );
+    space  = NIC_RX_BUF_SIZE - status;
+
+    // get client thread 
+    client_xp  = hal_remote_l64( socket_client_xp );
+
+    // get number of bytes in payload
+    payload = k_length - UDP_HEAD_LEN;
+
+    // compute number of bytes to move : min (space , seg_payload)
+    moved_bytes = ( space < payload ) ? space : payload;
+
+    // move payload from kernel buffer to socket rx_buf
+    remote_buf_put_from_kernel( socket_rbuf_xp,
+                                 k_buf + UDP_HEAD_LEN,
+                                 moved_bytes );
+
+    // unblock client thread if registered
+    if( client_xp != XPTR_NULL )
+    {
+        thread_unblock( client_xp , THREAD_BLOCKED_IO );
+    }
+
+    // release the lock protecting the socket
+    remote_rwlock_wr_release( socket_lock_xp );
+
+}  // end dev_nic_rx_handle_udp_packet()
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_rx_server() function to handle one RX
+// TCP segment contained in a kernel buffer defined by the <k_buf> & <k_length> arguments.
+// It the received segment doesn't match an existing local socket, or is corrupted,
+// this faulty segment is discarded. 
+///////////////////////////////////////////////////////////////////////////////////////////
+// Implementation note:
+// 1) It checks the TCP checksum, and discard the corrupted segment.
+// 2) It scans the list of sockets attached to the RX chdev, to find the socket
+//    matching the TCP segment header, and discards the segment if no socket found.
+// 3) When a socket has been found, it takes the lock protecting the socket state,
+//    because the socket is accessed by both the NIC_TX and NIC_RX server threads.
+// 4) Depending on the socket state, it handle the received segment, including the
+//    SYN, FIN, ACK and RST flags. It updates the socket state when required, moves
+//    data to the rx_buf when possible, and registers requests to the TX server 
+//    thread in the R2T queue attached to the socket, to insert control flags in the
+//    TX stream, as required.
+// 5) Finally, it releases the lock protecting the socke and returns.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ chdev         : [in] local pointer on local NIC_RX chdev descriptor.
+// @ k_buf         : [in] pointer on the TCP packet in local kernel buffer.
+// @ k_length      : [in] number of bytes in buffer (including TCP header).
+// @ seg_src_addr  : [in] source IP address (from IP packet header).
+// @ seg_dst_addr  : [in] destination IP address (from IP packet header).
+///////////////////////////////////////////////////////////////////////////////////////////
+static void dev_nic_rx_handle_tcp_segment( chdev_t  * chdev,
+                                           uint8_t  * k_buf,
+                                           uint32_t   k_length,
+                                           uint32_t   seg_src_addr,
+                                           uint32_t   seg_dst_addr )
+{
+    xptr_t     root_xp;           // extended pointer on attached sockets list root 
+    xptr_t     lock_xp;           // extended pointer on chdev lock 
+    xptr_t     iter_xp;           // iterator for these queues
+    bool_t     match_socket;      // true if socket found
+    xptr_t     socket_xp;         // extended pointer on matching socket descriptor
+    cxy_t      socket_cxy;
+    socket_t * socket_ptr;
+    uint32_t   local_addr;        // local IP address from socket
+    uint32_t   local_port;        // local port from socket
+    uint32_t   remote_addr;       // remote IP address from socket
+    uint32_t   remote_port;       // remote port from socket
+    uint32_t   socket_state;      // socket state
+    uint32_t   socket_type;       // socket type
+    uint32_t   socket_tx_nxt;    // next byte to send in TX stream
+    uint32_t   socket_tx_una;    // first unacknowledged byte in TX stream
+    uint32_t   socket_rx_nxt;    // next expected byte in RX stream
+    uint32_t   socket_rx_wnd;    // current window value in RX stream
+    xptr_t     socket_lock_xp;    // extended pointer on lock protecting socket state
+    xptr_t     socket_rx_buf_xp;  // extended pointer on socket rx_buf
+    xptr_t     socket_r2tq_xp;    // extended pointer on socket r2t queue
+    xptr_t     socket_client_xp;  // extended pointer on socket rx_client thread
+    uint16_t   checksum;          // computed TCP segment chechsum
+
+    // build extended pointer on xlist of all sockets attached to NIC_RX chdev
+    root_xp = XPTR( local_cxy , &chdev->wait_root );
+    lock_xp = XPTR( local_cxy , &chdev->wait_lock );
+
+    // get relevant infos from TCP segment header
+    uint32_t seg_src_port = ((uint32_t)k_buf[0] << 8) | (uint32_t)k_buf[1];
+    uint32_t seg_dst_port = ((uint32_t)k_buf[2] << 8) | (uint32_t)k_buf[3];
+
+    uint32_t seg_seq_num  = ((uint32_t)k_buf[4]  << 24) | 
+                            ((uint32_t)k_buf[5]  << 16) |
+                            ((uint32_t)k_buf[6]  <<  8) |
+                            ((uint32_t)k_buf[7]       );
+
+    uint32_t seg_ack_num  = ((uint32_t)k_buf[8]  << 24) | 
+                            ((uint32_t)k_buf[9]  << 16) |
+                            ((uint32_t)k_buf[10] <<  8) |
+                            ((uint32_t)k_buf[11]      );
+
+    uint8_t  seg_hlen     = k_buf[12] >> 2;       // TCP header length in bytes
+ 
+    uint8_t  seg_flags    = k_buf[13];
+
+    bool_t   seg_ack_set  = ((seg_flags & TCP_FLAG_ACK) != 0);
+    bool_t   seg_syn_set  = ((seg_flags & TCP_FLAG_SYN) != 0);
+    bool_t   seg_fin_set  = ((seg_flags & TCP_FLAG_FIN) != 0);
+    bool_t   seg_rst_set  = ((seg_flags & TCP_FLAG_RST) != 0);
+
+    uint16_t seg_window   = ((uint32_t)k_buf[14] << 8) | (uint32_t)k_buf[15];
+
+    uint16_t seg_checksum = ((uint32_t)k_buf[16] << 8) | (uint32_t)k_buf[17];
+
+    uint32_t seg_payload  = k_length - seg_hlen;  // number of bytes in payload
+
+    // 1. compute TCP checksum
+    checksum = dev_nic_tcp_checksum( k_buf,
+                                     k_length,
+                                     seg_src_addr,
+                                     seg_dst_addr );
+
+    // discard segment if corrupted 
+    if( seg_checksum != checksum ) return;
+    
+    match_socket = false;
+
+    // take the lock protecting the list of sockets
+    remote_busylock_acquire( lock_xp );
+
+    // 2. scan list of sockets to find a matching socket
+    XLIST_FOREACH( root_xp , iter_xp )
+    {
+        // get socket cluster and local pointer
+        socket_xp  = XLIST_ELEMENT( iter_xp , socket_t , rx_list );
+        socket_ptr = GET_PTR( socket_xp );
+        socket_cxy = GET_CXY( socket_xp );
+
+        // get socket type and state
+        socket_type  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->type ));
+        socket_state = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->state ));
+                
+        // skip UDP socket
+        if( socket_type == SOCK_DGRAM ) continue;
+
+        // get relevant socket infos for matching
+        local_addr   = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->local_addr )); 
+        remote_addr  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->remote_addr )); 
+        local_port   = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->local_port )); 
+        remote_port  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->remote_port )); 
+                
+        // compute matching condition
+        // (in LISTEN state, remote_port and remote_addr can be unspecified)
+        if( socket_state == TCP_STATE_LISTEN )
+        {
+            match_socket = (local_addr  == seg_dst_addr) && 
+                           (local_port  == seg_dst_port) ;
+        }
+        else
+        {
+            match_socket = (local_addr  == seg_dst_addr) && 
+                           (local_port  == seg_dst_port) &&
+                           (remote_addr == seg_src_addr) &&
+                           (remote_port == seg_src_port) ;
+        }
+
+        // exit loop if matching
+        if( match_socket ) break;
+
+    }  // end loop on sockets
+
+    // release the lock protecting the list of sockets
+    remote_busylock_release( lock_xp );
+
+    // discard segment if no matching socket found
+    if( match_socket == false ) return;
+
+    // From here the actions depend on both the socket state,
+    // and the received segment flags
+    // - update socket state,
+    // - move data to rx_buf,
+    // - make a R2T request when required
+
+    // build extended pointers on various socket fields
+    socket_lock_xp    = XPTR( socket_cxy , &socket_ptr->lock );
+    socket_rx_buf_xp  = XPTR( socket_cxy , &socket_ptr->rx_buf );
+    socket_r2tq_xp    = XPTR( socket_cxy , &socket_ptr->r2tq );
+    socket_client_xp  = XPTR( socket_cxy , &socket_ptr->rx_client );
+
+    // 3. take the lock protecting the matching socket
+    remote_rwlock_wr_acquire( socket_lock_xp );
+
+    // get relevant socket infos from socket descriptor
+    socket_state   = hal_remote_l32(XPTR( socket_cxy , &socket_ptr->state ));
+    socket_rx_nxt = hal_remote_l32(XPTR( socket_cxy , &socket_ptr->rx_nxt )); 
+    socket_rx_wnd = hal_remote_l32(XPTR( socket_cxy , &socket_ptr->rx_wnd )); 
+    socket_tx_una = hal_remote_l32(XPTR( socket_cxy , &socket_ptr->tx_una )); 
+    socket_tx_nxt = hal_remote_l32(XPTR( socket_cxy , &socket_ptr->tx_nxt )); 
+
+    switch( socket_state )
+    {
+        //////////////////////
+        case TCP_STATE_LISTEN:
+        {
+            // [1] discard segment if RST flag
+            if( seg_rst_set )  return;
+
+            // [2] send a RST & discard segment if ACK flag
+            if( seg_ack_set )
+            {
+                    // set socket.tx_nxt to seg_ack_num
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                    seg_ack_num );
+
+                    // make RST request to R2T queue
+                    dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                TCP_FLAG_RST );
+                    // discard segment
+                    break;
+                }
+
+                // [3] handle SYN flag
+                if( seg_syn_set )
+                {
+                    // set socket.rx_nxt to seg_seq_num + 1
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_nxt ),
+                                    seg_seq_num + 1 );
+
+                    // set socket.tx_nxt to ISS
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                    TCP_ISS );
+
+                    // set socket.rx_irs to seg_seq_num 
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_irs ),
+                                    seg_seq_num + 1 );
+
+                    // make SYN.ACK request to R2T queue
+                    dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                TCP_FLAG_SYN | TCP_FLAG_ACK );
+                     
+                    // set socket.tx_nxt to ISS + 1
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                    TCP_ISS + 1 );
+
+                    // set socket.tx_una to ISS
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_una ),
+                                    TCP_ISS ); 
+            
+                    // update socket.state
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_SYN_RCVD );
+
+                    // update socket.remote_addr 
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->remote_addr ),
+                                    seg_src_addr );
+
+                    // update socket.remote_port 
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->remote_port ),
+                                    seg_src_port );
+                }                
+                break;
+            }
+            ////////////////////////
+            case TCP_STATE_SYN_SENT:
+            {
+                // [1] check ACK flag 
+                if( seg_ack_set )
+                {
+                    if( seg_ack_num != TCP_ISS + 1 )  // ACK not acceptable
+                    {
+                        // discard segment if RST
+                        if( seg_rst_set ) break;
+
+                        // set socket.tx_nxt to seg_ack_num
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                        seg_ack_num );
+
+                        // make an RST request to R2T queue
+                        dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                    TCP_FLAG_RST );
+                        // discard segment
+                        break;
+                    }
+                }
+
+                // [2] check RST flag
+                if( seg_rst_set )
+                {
+                    // TODO signal "error: connection reset" to user
+
+                    // update socket.state
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_BOUND );
+
+                    // discard segment
+                    break;
+                }
+
+                // [3] handle SYN flag when (no ACK or acceptable ACK, and no RST)
+                if( seg_syn_set )
+                {
+                    // TODO Ne faut-il pas tester seg_seq_num ?
+
+                    if( seg_ack_set )  // received both SYN and ACK
+                    { 
+                        // set socket.rx_nxt to seg_seq_num + 1 
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_nxt ),
+                                        seg_seq_num + 1 );
+
+                        // set socket.tx_una to seg_ack_num 
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_una ),
+                                        seg_ack_num );
+
+                        // set socket.rx_irs to seg_seq_num
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_irs ),
+                                        seg_seq_num );
+
+                        // update socket.state
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                        TCP_STATE_ESTAB );
+
+                        // make an ACK request to R2T queue
+                        dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                    TCP_FLAG_ACK );
+                    }
+                    else               // received SYN without ACK
+                    {
+                        // update socket.state
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                        TCP_STATE_SYN_RCVD );
+
+                        // set socket.tx_nxt to ISS
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                        TCP_ISS );     
+
+                        // make a SYN.ACK request to R2T queue
+                        dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                    TCP_FLAG_SYN | TCP_FLAG_ACK );
+                    }
+                }
+                break;
+            }
+            ////////////////////////
+            case TCP_STATE_SYN_RCVD:
+            case TCP_STATE_ESTAB:
+            case TCP_STATE_FIN_WAIT1:
+            case TCP_STATE_FIN_WAIT2:
+            case TCP_STATE_CLOSE_WAIT:
+            case TCP_STATE_CLOSING:
+            case TCP_STATE_LAST_ACK:
+            case TCP_STATE_TIME_WAIT:
+            {
+                // [1] check sequence number 
+
+                // compute min & max acceptable sequence numbers
+                uint32_t seq_min  = socket_rx_nxt;
+                uint32_t seq_max  = socket_rx_nxt + socket_rx_wnd - 1;
+
+                // compute sequence number for last byte in segment 
+                uint32_t seg_seq_last = seg_seq_num + seg_payload - 1;
+
+                if( (seg_seq_num != socket_rx_nxt) ||     // out_of_order 
+                    (is_in_window( seg_seq_last, 
+                                   seq_min,
+                                   seq_max ) == false) )  // out_of_window
+                {
+                    // discard segment
+                    return;
+                } 
+
+                // [2] handle RST flag
+
+                if( seg_rst_set )
+                {
+                     if( socket_state == TCP_STATE_SYN_RCVD )
+                     {
+                         // TODO unblock all clients threads with "reset" responses
+                     }
+                     else if( (socket_state == TCP_STATE_ESTAB     ) ||
+                              (socket_state == TCP_STATE_FIN_WAIT1 ) ||
+                              (socket_state == TCP_STATE_FIN_WAIT2 ) ||
+                              (socket_state == TCP_STATE_CLOSE_WAIT) )
+                     {
+                         // TODO all pending send & received commands 
+                         // must receive "reset" responses
+
+                         // TODO destroy the socket
+                     }
+                     else  // all other states
+                     {
+                             
+
+                }
+
+                // [3] handle security & precedence TODO ... someday
+
+                // [4] handle SYN flag
+
+                if( seg_syn_set )        // received SYN
+                {
+                    // TODO signal error to user
+
+                    // make an RST request to R2T queue 
+                    dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                TCP_FLAG_RST );
+                    // update socket state
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_BOUND );
+                }
+
+                // [5] handle  ACK flag 
+
+                if( seg_ack_set == false ) 
+                {
+                    // discard segment when ACK not set
+                    break;
+                }
+                else if( socket_state == TCP_STATE_SYN_RCVD )
+                {
+                    if( is_in_window( seg_ack_num , socket_tx_una , socket_tx_nxt ) ) 
+                    {
+                        // update socket.state to ESTAB
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                              TCP_STATE_ESTAB );
+                    }
+                    else   // unacceptable ACK
+                    {
+                        // set socket.tx_nxt to seg_ack_num 
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_nxt ),
+                                        seg_ack_num );
+
+                        // make an RST request to R2T queue 
+                        dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                    TCP_FLAG_RST );
+                    }
+                }
+                else if( (socket_state == TCP_STATE_ESTAB)      ||
+                         (socket_state == TCP_STATE_FIN_WAIT1)  ||
+                         (socket_state == TCP_STATE_FIN_WAIT1)  ||
+                         (socket_state == TCP_STATE_CLOSE_WAIT) ||
+                         (socket_state == TCP_STATE_CLOSING)    )
+                {
+                    if( is_in_window( seg_ack_num + 1 , socket_tx_una , socket_tx_nxt ) ) 
+                    {
+                        // update socket.tx_una
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_una ),
+                                        seg_ack_num );
+
+                        // update socket.tx_wnd  
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_wnd ),
+                                        seg_window );
+                    }
+                    else   // unacceptable ACK
+                    {
+                        // discard segment
+                        break;
+                    }
+                
+                    // specific for FIN_WAIT1 
+                    if( socket_state == TCP_STATE_FIN_WAIT1 )
+                    {
+                        if( seg_fin_set )
+                        {
+                            // update socket.state
+                            hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                                  TCP_STATE_FIN_WAIT2 );
+                        }
+                    }
+
+                    // specific for CLOSING
+                    if( socket_state == TCP_STATE_CLOSING )
+                    {
+                        if( seg_ack_set )
+                        {
+                            // update socket.state
+                            hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                                  TCP_STATE_TIME_WAIT );
+                        }
+                        else
+                        {
+                            // discard segment
+                            break;
+                        }
+                    }
+                }
+                else if( socket_state == TCP_STATE_LAST_ACK )
+                {
+                    if( seg_ack_set )
+                    {
+                        // update socket.state
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                              TCP_STATE_TIME_WAIT );
+                    }
+                    
+                }
+
+                // [6] handle URG flag  TODO ... someday
+
+                // [7] Move DATA to rx_buf and unblock client thread
+
+                if( seg_payload )
+                {
+                    if( (socket_state == TCP_STATE_ESTAB)     ||
+                        (socket_state == TCP_STATE_FIN_WAIT1) ||
+                        (socket_state == TCP_STATE_FIN_WAIT2) )
+                    {
+                        // get number of bytes already stored in rx_buf
+                        uint32_t status = remote_buf_status( socket_rx_buf_xp );
+
+                        // compute empty space in rx_buf
+                        uint32_t space = NIC_RX_BUF_SIZE - status;
+
+                        // compute number of bytes to move : min (space , seg_payload)
+                        uint32_t nbytes = ( space < seg_payload ) ? space : seg_payload;
+
+                        // move payload from k_buf to rx_buf
+                        remote_buf_put_from_kernel( socket_rx_buf_xp, 
+                                                    k_buf + seg_hlen,
+                                                    nbytes );
+                        // update socket.rx_nxt 
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_nxt ),
+                                              socket_rx_nxt + nbytes );
+
+                        // update socket.rx_wnd 
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_wnd ),
+                                        socket_rx_wnd - nbytes ); 
+
+                        // make an ACK request to R2T queue
+                        dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                    TCP_FLAG_ACK );
+
+                        // get extended pointer on rx_client thread
+                        xptr_t client_xp = hal_remote_l64( socket_client_xp );
+                        
+                        // unblock client thread 
+                        if( client_xp != XPTR_NULL )
+                        {
+                            thread_unblock( client_xp , THREAD_BLOCKED_IO );
+                        }
+                    }
+                }
+
+                // [8] handle FIN flag 
+
+                if( seg_fin_set )
+                {
+                    if( (socket_state == TCP_STATE_UNBOUND) ||
+                        (socket_state == TCP_STATE_BOUND)   ||
+                        (socket_state == TCP_STATE_LISTEN)  ||
+                        (socket_state == TCP_STATE_SYN_SENT) ) 
+                    {
+                        // discard segment
+                        break;
+                    }
+                    else // all other states
+                    {
+                        // TODO signal "connection closing"
+
+                        // make an ACK request to R2T queue
+                        dev_nic_rx_put_r2t_request( socket_r2tq_xp,
+                                                    TCP_FLAG_ACK );
+
+                        // increment socket.rx_nxt
+                        hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_nxt ),
+                                        socket_rx_nxt + 1 ); 
+
+                        if( (socket_state == TCP_STATE_SYN_RCVD) ||
+                            (socket_state == TCP_STATE_ESTAB) )
+                        {
+                            // update socket.state
+                            hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                            TCP_STATE_TIME_WAIT );
+                        }
+                        else if( socket_state == TCP_STATE_FIN_WAIT1 )
+                        { 
+                            if( seg_ack_set ) 
+                            {
+                                // TODO start "time-wait" timer / turn off others timers
+
+                                // update socket.state
+                                hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                                TCP_STATE_TIME_WAIT );
+                            }
+                            else
+                            {
+                                // update socket.state
+                                hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                                TCP_STATE_CLOSING );
+                            }
+                        }
+                        else if( socket_state == TCP_STATE_FIN_WAIT2 )
+                        {
+                            // TODO start "time-wait" timer / turn off other timers
+
+                            // update socket.state
+                            hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                            TCP_STATE_TIME_WAIT );
+                        }
+                        else if( socket_state == TCP_STATE_TIME_WAIT )
+                        {
+                            // TODO restart "time_wait" timer
+                        }
+                    }
+                }  // end if FIN 
+            }  // end case sockets synchronized
+        }  // end switch socket state
+
+        // release the lock protecting socket
+        remote_rwlock_wr_acquire( socket_lock_xp );
+
+    }  // end socket found
+
+}  // end dev_nic_rx_handle_tcp_segment()
+
+
+/////////////////////////////////////////
+void dev_nic_rx_server( chdev_t * chdev )
+{
+    uint8_t       k_buf[2048];          // kernel buffer for one ETH/IP/UDP packet
+                                  
+    uint32_t      pkt_src_addr;         // packet source IP address 
+    uint32_t      pkt_dst_addr;         // packet destination IP address
+    uint32_t      trsp_protocol;        // transport protocol (TCP / UDP)
+    uint32_t      eth_length;           // size of Ethernet packet (bytes)
+    uint32_t      ip_length;            // size of IP packet in bytes
+    uint32_t      nic_queue_readable;   // NIC_RX queue non empty when true
+    error_t       error;
+
+    thread_t * this = CURRENT_THREAD;
+
+// check chdev direction and type
+assert( (chdev->func == DEV_FUNC_NIC) && (chdev->is_rx == true) ,
+"illegal chdev type or direction" );
 
 // check thread can yield
-assert( (thread_ptr->busylocks == 0),
-"cannot yield : busylocks = %d\n", thread_ptr->busylocks );
+assert( (this->busylocks == 0),
+"cannot yield : busylocks = %d\n", this->busylocks );
+
+    while( 1 )
+    {
+        // check NIC_RX_QUEUE readable
+        error = dev_nic_rx_queue_readable( chdev,
+                                           &nic_queue_readable );
+        if( error )
+        {
+            printk("\n[PANIC] in %s : cannot access NIC_TX[%d] queue\n",
+            __FUNCTION__, chdev->channel );
+        }
+    
+        if( nic_queue_readable ) // NIC_TX_QUEUE non empty
+        {
+            // moves one Ethernet packet to kernel buffer
+            error = dev_nic_rx_move_packet( chdev,
+                                            k_buf,
+                                            &eth_length );
+            if( error )
+            {
+                printk("\n[PANIC] in %s : cannot read the NIC_TX[%d] queue\n",
+                __FUNCTION__, chdev->channel );
+            }
+
+            // analyse the ETH header
+            error = dev_nic_rx_check_eth( k_buf,
+                                          &ip_length );
+
+            // discard packet if error reported by Ethernet layer
+            if( error ) continue; 
+
+            // analyse the IP header
+            error = dev_nic_rx_check_ip( k_buf + ETH_HEAD_LEN,
+                                         ip_length, 
+                                         &pkt_src_addr, 
+                                         &pkt_dst_addr,
+                                         &trsp_protocol );
+
+            // discard packet if error reported by IP layer
+            if( error ) continue; 
+
+            // call relevant transport protocol
+            if( trsp_protocol == PROTOCOL_UDP )   
+            { 
+                dev_nic_rx_handle_udp_packet( chdev,
+                                              k_buf + ETH_HEAD_LEN + IP_HEAD_LEN,
+                                              ip_length - IP_HEAD_LEN,
+                                              pkt_src_addr,
+                                              pkt_dst_addr );
+            }
+            else if ( trsp_protocol == PROTOCOL_TCP)
+            {
+                dev_nic_rx_handle_tcp_segment( chdev,
+                                               k_buf + ETH_HEAD_LEN + IP_HEAD_LEN,
+                                               ip_length - IP_HEAD_LEN,
+                                               pkt_src_addr,
+                                               pkt_dst_addr );
+            }
+        }
+        else     // block and deschedule if NIC_RX_QUEUE empty
+        {
+            thread_block( XPTR( local_cxy , this ) , THREAD_BLOCKED_ISR );
+            sched_yield( "waiting RX client" );
+        }
+
+    } // end of while loop
+
+}  // end dev_nic_rx_server()
+
+
+
+
+
+
+
+
+
+
+///////////////////////////////////////////////////////////////////////////////////////////
+//              Functions used by the NIC_TX server thread
+///////////////////////////////////////////////////////////////////////////////////////////
+
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// These static functions are called by the NIC_TX server thread to report the 
+// completion  (success or error) of a TX command.
+// - it print an error message in case of error.
+// - it updates the "tx_error"  field in socket descriptor.
+// - it unblocks the client thread.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ socket_xp    : [in] extended pointer on socket
+// @ cmd_type     : [in] SOCKET_TX_CONNECT / SOCKET_TX_SEND / SOCKET_TX_CLOSE
+// @ socket_state : [in] current socket state
+///////////////////////////////////////////////////////////////////////////////////////////
+static void dev_nic_tx_report_error( xptr_t    socket_xp,
+                                     uint32_t  cmd_type,
+                                     uint32_t  socket_state )
+{
+    printk("\n[ERROR] in %s : command %s in %s state\n",
+    __FUNCTION__, socket_cmd_str(cmd_type), socket_state_str(socket_state) );
+
+    // get socket thread cluster and local pointer
+    socket_t * socket_ptr = GET_PTR( socket_xp );
+    cxy_t      socket_cxy = GET_CXY( socket_xp );
+
+    // set tx_error field in socket descriptor
+    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_error ) , 1 );
+
+    // get extended point on client thread
+    xptr_t client_xp = hal_remote_l64( XPTR( socket_cxy , &socket_ptr->tx_client ));
+
+    // unblock the client thread
+    thread_unblock( client_xp , THREAD_BLOCKED_IO );
+}
+
+////////////////////////////////////////////////////////////
+static void dev_nic_tx_report_success( xptr_t    socket_xp )
+{
+    // get socket thread cluster and local pointer
+    socket_t * socket_ptr = GET_PTR( socket_xp );
+    cxy_t      socket_cxy = GET_CXY( socket_xp );
+
+    // set tx_error field in socket descriptor
+    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_error ) , 0 );
+
+    // get extended point on client thread
+    xptr_t client_xp = hal_remote_l64( XPTR( socket_cxy , &socket_ptr->tx_client ));
+
+    // unblock the client thread
+    thread_unblock( client_xp , THREAD_BLOCKED_IO );
+}
+
+
+
+
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function.
+// It calls directly the NIC driver (WRITABLE command) and returns the status
+// of the NIC_TX queue identified by the <chdev> argument.
+// in the <writable> buffer.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ chdev     : [in]  local pointer on NIC_TX chdev.
+// @ length    : [in]  packet length in bytes.
+// @ writable  : [out] zero if queue full.  
+// @ returns 0 if success / returns -1 if failure in accessing NIC device.
+///////////////////////////////////////////////////////////////////////////////////////////
+error_t dev_nic_tx_queue_writable( chdev_t  * chdev,
+                                   uint32_t   length,
+                                   uint32_t * writable )
+{
+    thread_t * this = CURRENT_THREAD;
+
+    // initialize READABLE command in thread descriptor
+    this->nic_cmd.dev_xp = XPTR( local_cxy , chdev );
+    this->nic_cmd.type   = NIC_CMD_WRITABLE;
+    this->nic_cmd.length = length;
+
+    // call driver to test writable
+    chdev->cmd( XPTR( local_cxy , this ) );
+
+    // return status
+    *writable = this->nic_cmd.status;
+
+    // return error
+    return this->nic_cmd.error;
+
+}  // end dev_nic_tx_queue_writable
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function.
+// It moves one ETH/IP/UDP packet from the kernel buffer identified by the <buffer> and 
+// <length> arguments to the NIC_TX_QUEUE identified the <chdev> argument.
+// It calls directly the NIC driver, without registering in a waiting queue, because 
+// only this NIC_TX server thread can access this NIC_TX_QUEUE.
+// 1) It checks NIC_TX_QUEUE status in a while loop, using the NIC_CMD_WRITABLE command.
+//    As long as the queue is not writable, it blocks and deschedules. It is re-activated 
+//    by the NIC-TX ISR as soon as the queue changes status.
+// 2) When the queue is writable, it put the ETH/IP/UDP packet into the NIC_TX_QUEUE,
+//   using the driver NIC_CMD_WRITE command.
+// Both commands are successively registered in this NIC-TX server thread descriptor
+// to be passed to the driver. 
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ chdev   : [in] local pointer on NIC_TX chdev.
+// @ buffer  : [in] pointer on a local kernel buffer (2K bytes).
+// @ length  : [in] actual Ethernet packet length in bytes.
+///////////////////////////////////////////////////////////////////////////////////////////
+void dev_nic_tx_move_packet( chdev_t  * chdev,
+                             uint8_t  * buffer,
+                             uint32_t   length )
+{
+    error_t    error;
+    uint32_t   writable;
+
+    thread_t * this = CURRENT_THREAD;
+
+    // get extended pointers on server tread and chdev 
+    xptr_t     thread_xp = XPTR( local_cxy , this );
+    xptr_t     chdev_xp  = XPTR( local_cxy , chdev );
+
+    // get local pointer on core running this server thead
+    core_t * core = this->core;
+
+// check thread can yield
+assert( (this->busylocks == 0),
+"cannot yield : busylocks = %d\n", this->busylocks );
 
 #if DEBUG_DEV_NIC_RX
 uint32_t cycle = (uint32_t)hal_get_cycles();
 if( DEBUG_DEV_NIC_RX < cycle )
-printk("\n[DBG] %s : thread %x enters for packet %x in cluster %x\n", 
-__FUNCTION__ , thread_ptr , pkd , local_cxy );
+printk("\n[%s] thread[%x,%x] enters for packet %x / cycle %d\n", 
+__FUNCTION__, this->process->pid, this->trdid, pkd, cycle );
 #endif
 
-    // get pointer on NIC-TX chdev descriptor
-    uint32_t   channel = thread_ptr->chdev->channel; 
-    xptr_t     dev_xp  = chdev_dir.nic_tx[channel];
-    cxy_t      dev_cxy = GET_CXY( dev_xp );
-    chdev_t  * dev_ptr = (chdev_t *)GET_PTR( dev_xp );
-
-    assert( (dev_xp != XPTR_NULL) , "undefined NIC chdev descriptor" );
-
-    assert( (dev_cxy == local_cxy) , " chdev must be local" );
-
-    // initialize command in thread descriptor
-    thread_ptr->nic_cmd.dev_xp = dev_xp;
-
-    // call driver to test writable
-    thread_ptr->nic_cmd.cmd = NIC_CMD_WRITABLE;
-    dev_ptr->cmd( thread_xp );
-
-    // check error
-    error = thread_ptr->nic_cmd.error;
-    if( error ) return error;
-
-    // block and deschedule if queue non writable
-    if( thread_ptr->nic_cmd.status == false )  
-    {
-        // enable NIC-TX IRQ 
-        dev_pic_enable_irq( core->lid ,dev_xp );
-
-        // block client thread on THREAD_BLOCKED I/O condition
-        thread_block( XPTR( local_cxy , thread_ptr ) , THREAD_BLOCKED_IO );
-
-        // deschedule client thread
-        sched_yield("client blocked on I/O");
-
-        // disable NIC-TX IRQ 
-        dev_pic_disable_irq( core->lid , dev_xp );
-    }
-
-    // call driver for actual write
-    thread_ptr->nic_cmd.cmd    = NIC_CMD_WRITE;
-    thread_ptr->nic_cmd.buffer = pkd->buffer;
-    thread_ptr->nic_cmd.length = pkd->length;
-    dev_ptr->cmd( thread_xp );
-
-    // check error
-    error = thread_ptr->nic_cmd.error;
-    if( error ) return error;
+    // check NIC_TX_QUEUE writable
+    while( 1 )
+    {
+        error = dev_nic_tx_queue_writable( chdev,
+                                           length,
+                                           &writable );
+        if( error )
+        {
+            printk("\n[PANIC] in %s : cannot access NIC_TX queue\n", __FUNCTION__ );
+            return;
+        }
+            
+        if( writable == 0 )  // block & deschedule if non writable
+        {
+            // enable NIC-TX IRQ 
+            dev_pic_enable_irq( core->lid , chdev_xp );
+
+            // block TX server thread
+            thread_block( thread_xp , THREAD_BLOCKED_ISR );
+
+            // deschedule TX server thread
+            sched_yield("client blocked on NIC_TX queue full");
+
+            // disable NIC-TX IRQ 
+            dev_pic_disable_irq( core->lid , chdev_xp );
+        }
+        else                // exit loop if writable
+        {
+            break;
+        }
+    }
+
+    // initialize WRITE command in server thread descriptor
+    this->nic_cmd.dev_xp = chdev_xp;
+    this->nic_cmd.type   = NIC_CMD_WRITE;
+    this->nic_cmd.buffer = buffer;
+    this->nic_cmd.length = length;
+
+    // call driver to move packet
+    chdev->cmd( thread_xp );
 
 #if DEBUG_DEV_NIC_RX
 cycle = (uint32_t)hal_get_cycles();
 if( DEBUG_DEV_NIC_RX < cycle )
-printk("\n[DBG] %s : thread %x exit for packet %x in cluster %x\n", 
-__FUNCTION__ , thread_ptr , pkd , local_cxy );
+printk("\n[%s] thread[%x,%x] exit for packet %x\n", 
+__FUNCTION__ , this->process->pid, this->trdid , pkd );
 #endif
 
-    return 0;
-}  // end dev_nic_write()
-
-
-
+    return;
+
+}  // end dev_nic_tx_move_packet()
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function to build an UDP
+// header in the kernel buffer defined by the <k_buf> arguement,  as specified by the 
+// <socket_xp> argument. The <length> argument defines the number of bytes in payload.
+// It set the "src_port", "dst_port", "total_length" and "checksum" fields in UDP header.
+// The payload must be previouly loaded in the pernel buffer.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ k_buf      : [in]  pointer on first byte of UDP header in kernel buffer.
+// @ socket_xp  : [in]  extended pointer on socket.
+// @ length     : [in]  number of bytes in payload.
+///////////////////////////////////////////////////////////////////////////////////////////
+void dev_nic_tx_build_udp_header( uint8_t  * k_buf,
+                                  xptr_t     socket_xp,
+                                  uint32_t   length )
+{
+    uint16_t   checksum;        // checksum value
+    uint32_t   total_length;    // total UDP packet length
+    uint32_t   local_addr;      // local IP address
+    uint32_t   remote_addr;     // remote IP address
+    uint32_t   local_port;      // local port
+    uint32_t   remote_port;     // remote port
+
+    // get socket cluster an local pointer
+    socket_t * socket_ptr = GET_PTR( socket_xp );
+    cxy_t      socket_cxy = GET_CXY( socket_xp );
+
+    // get relevant infos from socket
+    local_addr  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->local_addr )); 
+    remote_addr = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->remote_addr )); 
+    local_port  = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->local_port )); 
+    remote_port = hal_remote_l32(XPTR(socket_cxy , &socket_ptr->remote_port )); 
+
+    // compute UDP packet total length
+    total_length = length + UDP_HEAD_LEN;
+
+    // set src_port and dst_port in header
+    k_buf[0] = local_port >> 8; 
+    k_buf[1] = local_port; 
+    k_buf[2] = remote_port >> 8; 
+    k_buf[3] = remote_port; 
+
+    // set packet length in header
+    k_buf[4] = total_length >> 8;
+    k_buf[5] = total_length;
+    
+    // compute UDP packet checksum
+    checksum = dev_nic_udp_checksum( k_buf , total_length );
+
+    // set checksum
+    k_buf[6] = checksum >> 8;
+    k_buf[7] = checksum;
+
+}  // end dev_nic_tx_build_udp_header()
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function. 
+// It builds a TCP header in the kernel buffer defined by the <k_buf> argument.
+// The payload must have been previouly registered in this buffer.
+// The "local_addr", "local_port", "remote_addr", "remote_port", seq_num", "ack_num",
+// and "window" fields are obtained from the <socket_xp> argument.
+// The <length> argument defines the number of bytes in payload, and the <flags> argument
+// defines the flags to be set in TCP header.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ k_buf      : [in]  pointer on first byte of TCP header in kernel buffer.
+// @ length     : [in]  number of bytes in payload.
+// @ socket_xp  : [in]  extended pointer on socket.
+// @ flags      : [in]  flags to be set in TCP header.
+///////////////////////////////////////////////////////////////////////////////////////////
+void dev_nic_tx_build_tcp_header( uint8_t  * k_buf,
+                                  uint32_t   length,
+                                  xptr_t     socket_xp,
+                                  uint8_t    flags )
+{
+    uint16_t   checksum;        // global segment checksum
+    uint32_t   total_length;    // total UDP packet length
+    uint32_t   src_addr;        // local IP address
+    uint32_t   dst_addr;        // remote IP address
+    uint16_t   src_port;        // local port
+    uint16_t   dst_port;        // remote port
+    uint32_t   seq_num;         // first byte of segment in TX stream
+    uint32_t   ack_num;         // next expected byte in RX stream
+    uint16_t   window;          // window of accepted segments in RX stream
+
+    // get socket cluster an local pointer
+    socket_t * sock_ptr = GET_PTR( socket_xp );
+    cxy_t      sock_cxy = GET_CXY( socket_xp );
+
+    // get relevant infos from socket
+    src_addr = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->local_addr )); 
+    dst_addr = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->remote_addr )); 
+    src_port = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->local_port )); 
+    dst_port = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->remote_port )); 
+    seq_num  = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->tx_nxt ));
+    ack_num  = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->rx_nxt ));
+    window   = hal_remote_l32(XPTR( sock_cxy , &sock_ptr->rx_wnd ));
+
+    // compute TCP segment total length
+    total_length = length + TCP_HEAD_LEN;
+
+    // set "src_port" and "dst_port" 
+    k_buf[0]  = src_port >> 8; 
+    k_buf[1]  = src_port; 
+    k_buf[2]  = dst_port >> 8; 
+    k_buf[3]  = dst_port; 
+
+    // set "seq_num"  
+    k_buf[4]  = seq_num >> 24;
+    k_buf[5]  = seq_num >> 16;
+    k_buf[6]  = seq_num >>  8;
+    k_buf[7]  = seq_num;
+
+    // set "ack_num"  
+    k_buf[8]  = ack_num >> 24;
+    k_buf[9]  = ack_num >> 16;
+    k_buf[10] = ack_num >>  8;
+    k_buf[11] = ack_num;
+
+    // set "hlen"
+    k_buf[12] = 5;
+
+    // set "flags"
+    k_buf[13] = flags & 0x3F;
+
+    // set "window"
+    k_buf[14] = window >> 8;
+    k_buf[15] = window;
+
+    // reset "checksum"
+    k_buf[16] = 0;
+    k_buf[17] = 0;
+    
+    // set "urgent_ptr"
+    k_buf[18] = 0;
+    k_buf[19] = 0;
+  
+    // compute TCP segment checksum
+    checksum = dev_nic_tcp_checksum( k_buf,
+                                     total_length,
+                                     src_addr,
+                                     dst_addr );
+    // set "checksum"
+    k_buf[16] = checksum >> 8;
+    k_buf[17] = checksum;
+
+}  // end dev_nic_tx_build_tcp_header()
+
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function.
+// It builds the IP header in the 20 first bytes of <buffer>.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer     : pointer on first byte of IP header in kernel buffer
+// @ src_addr   : source IP address.
+// @ dst_addr   : destination IP address.
+// @ length     : number of bytes in IP packet payload.
+///////////////////////////////////////////////////////////////////////////////////////////
+void dev_nic_tx_build_ip_header( uint8_t * buffer,
+                                 uint32_t  src_addr,
+                                 uint32_t  dst_addr,
+                                 uint16_t  length )
+{
+    uint16_t   hcs;
+
+    uint16_t   total = length + IP_HEAD_LEN;
+
+    buffer[0]  = 0x45;         // IPV4 / IHL = 20 bytes  
+    buffer[1]  = 0;            // DSCP / ECN
+    buffer[2]  = total >> 8;  
+    buffer[3]  = total;  
+
+    buffer[4]  = 0x40;         // Don't Fragment
+    buffer[5]  = 0; 
+    buffer[6]  = 0;
+    buffer[7]  = 0; 
+
+    buffer[8]  = 0xFF;         // TTL
+    buffer[9]  = 0x11;         // UDP protocol 
+    
+    buffer[12] = src_addr >> 24; 
+    buffer[13] = src_addr >> 16; 
+    buffer[14] = src_addr >> 8; 
+    buffer[15] = src_addr; 
+
+    buffer[16] = dst_addr >> 24; 
+    buffer[17] = dst_addr >> 16; 
+    buffer[18] = dst_addr >> 8;
+    buffer[19] = dst_addr; 
+
+    // compute IP header checksum
+    hcs = dev_nic_ip_checksum( buffer ); 
+
+    // set checksum 
+    buffer[10] = hcs >> 8;
+    buffer[11] = hcs; 
+
+}  // end dev_nic_tx_build_ip_header
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function.
+// It builds the Ethernet header in the 14 first bytes of <buffer>.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ buffer     : pointer on first byte of Ethernet header in kernel buffer
+// @ src_mac_54 : two MSB bytes in source MAC address.
+// @ src_mac_32 : two MED bytes in source MAC address.
+// @ src_mac_10 : two LSB bytes in source MAC address.
+// @ dst_mac_54 : two MSB bytes in destination MAC address.
+// @ dst_mac_32 : two MED bytes in destination MAC address.
+// @ dst_mac_10 : two LSB bytes in destination MAC address.
+// @ length     : number of bytes in Ethernet frame payload.
+///////////////////////////////////////////////////////////////////////////////////////////
+void dev_nic_tx_build_eth_header( uint8_t * buffer,
+                                  uint16_t  src_mac_54,
+                                  uint16_t  src_mac_32,
+                                  uint16_t  src_mac_10,
+                                  uint16_t  dst_mac_54,
+                                  uint16_t  dst_mac_32,
+                                  uint16_t  dst_mac_10,
+                                  uint32_t  length )
+{
+    buffer[0]  = dst_mac_54 >> 8;
+    buffer[1]  = dst_mac_54;
+    buffer[2]  = dst_mac_32 >> 8;
+    buffer[3]  = dst_mac_32;
+    buffer[4]  = dst_mac_10 >> 8;
+    buffer[5]  = dst_mac_10;
+
+    buffer[6]  = src_mac_54 >> 8;
+    buffer[7]  = src_mac_54;
+    buffer[8]  = src_mac_32 >> 8;
+    buffer[9]  = src_mac_32;
+    buffer[10] = src_mac_10 >> 8;
+    buffer[11] = src_mac_10;
+
+    buffer[12] = length >> 8;
+    buffer[13] = length;
+
+}  // end dev_nic_tx_build_eth_header()
+
+///////////////////////////////////////////////////////////////////////////////////////////
+// This static function is called by the dev_nic_tx_server() function to handle one 
+// TX command, or one R2T request, registered in the socket identified by the <socket_xp>
+// argument. If there is one valid command, or if the R2T queue is non empty (for a TCP
+// socket), it builds an ETH/IP/UDP packet (or a ETH/IP/TCP segment), in the buffer 
+// defined by the  <k_buf> argument, and registers it in the NIC_TX queue defined by the 
+// <chdev> argument. The supported commands are SOCKET_SEND/SOCKET_CONNECT/SOCKET_CLOSE.
+// It unblocks the client thread when the command is completed.
+///////////////////////////////////////////////////////////////////////////////////////////
+// When there is a packet to send, it makes the following actions:
+// 1) it takes the lock protecting the socket state.
+// 2) it get the command arguments from client thread descriptor.
+// 3) it build an UDP packet or a TCP segment, depending on both the command type, and
+//    the socket state, updates the socket state, and unblocks the client thread.
+// 4) it release the lock protecting the socket.
+// 5) it build the IP header.
+// 6) it build the ETH header.
+// 7) it copies the packet in the NIC_TX queue.
+///////////////////////////////////////////////////////////////////////////////////////////
+// @ socket_xp   : [in] extended pointer on client socket.  
+// @ k_buf       : [in] local pointer on kernel buffer (2 Kbytes).
+// @ chdev       : [in] local pointer on NIC_RX chdev.
+///////////////////////////////////////////////////////////////////////////////////////////
+static void dev_nic_tx_handle_one_cmd( xptr_t    socket_xp,
+                                       uint8_t * k_buf,
+                                       chdev_t * chdev )
+{
+    socket_t  * socket_ptr;
+    cxy_t       socket_cxy;
+    xptr_t      client_xp;       // extended pointer on client thread  
+    thread_t  * client_ptr;
+    cxy_t       client_cxy;
+    sock_cmd_t  cmd;             // NIC command type
+    uint8_t   * buf;             // pointer on user buffer
+    uint32_t    len;             // user buffer length
+    uint32_t    todo;            // number of bytes not yet sent
+    uint32_t    socket_type;     // socket type (UDP/TCP)
+    uint32_t    socket_state;    // socket state        
+    xptr_t      socket_lock_xp;  // extended pointer on socket lock
+    xptr_t      socket_r2tq_xp;  // extended pointer on R2T queue
+    uint32_t    src_ip_addr;     // source IP address
+    uint32_t    dst_ip_addr;     // destination IP address
+    uint32_t    tx_una;          // next byte to be sent
+    uint32_t    tx_nxt;          // first unacknowledged byte
+    uint32_t    nbytes;          // number of bytes in UDP/TCP packet payload
+    uint8_t   * k_base;          // pointer UDP/TCP packet in kernel buffer
+    uint32_t    trsp_length;     // length of TCP/UDP packet
+    uint8_t     r2t_flags;       // flags defined by one R2T queue request
+    bool_t      do_send;         // build & send a packet when true
+  
+    // get socket cluster and local pointer
+    socket_cxy = GET_CXY( socket_xp );
+    socket_ptr = GET_PTR( socket_xp );
+    
+    // build extended pointer on socket lock and r2t queue
+    socket_lock_xp = XPTR( socket_cxy , &socket_ptr->lock );
+    socket_r2tq_xp = XPTR( socket_cxy , &socket_ptr->r2tq );
+
+    // 1. take lock protecting this socket
+    remote_rwlock_wr_acquire( socket_lock_xp );
+
+    // get pointers on TX client thread from socket
+    client_xp = hal_remote_l64( XPTR( socket_cxy , &socket_ptr->tx_client ));
+    client_cxy = GET_CXY( client_xp );
+    client_ptr = GET_PTR( client_xp );
+
+    // check valid command 
+    if( client_xp != XPTR_NULL )   // valid command found
+    {
+        // 2. get command arguments from socket
+        cmd  = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->tx_cmd ));
+        buf  = hal_remote_lpt( XPTR(socket_cxy , &socket_ptr->tx_buf ));
+        len  = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->tx_len ));
+        todo = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->tx_todo ));
+        
+        // get socket type and state
+        socket_type  = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->type ));
+        socket_state = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->state ));
+
+        // 3. UDP : build UDP packet and update UDP socket state
+        if( socket_type == SOCK_DGRAM )        
+        {
+            if( socket_state == UDP_STATE_UNBOUND )
+            {
+                // report illegal command
+                dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                do_send = false;
+            }
+            else  // BOUND or CONNECT state
+            {
+                if( cmd == SOCKET_TX_SEND ) 
+                {
+                    // compute payload length
+                    nbytes = ( PAYLOAD_MAX_LEN < todo ) ? PAYLOAD_MAX_LEN : todo;
+
+                    // compute UDP packet base in kernel buffer
+                    k_base = k_buf + ETH_HEAD_LEN + IP_HEAD_LEN;
+
+                    // move payload to kernel buffer
+                    hal_copy_from_uspace( XPTR(local_cxy , k_base + UDP_HEAD_LEN ),
+                                          buf + (len - todo),
+                                          nbytes );
+                    // build UDP header 
+                    dev_nic_tx_build_udp_header( k_base,
+                                                 socket_xp,
+                                                 nbytes );
+
+                    // update "tx_todo" in socket descriptor 
+                    hal_remote_s32( XPTR(socket_cxy , socket_ptr->tx_todo),
+                                    todo - nbytes );
+
+                    // unblock client thread when SEND command completed
+                    if( nbytes == todo )
+                    {
+                        dev_nic_tx_report_success( socket_xp );
+                    }
+
+                    do_send = true;
+                }
+                else
+                {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                    do_send = false;
+                }
+            }
+
+            // compute transport packet length
+            trsp_length = UDP_HEAD_LEN + nbytes;
+
+        }  // end UDP
+
+        // 3. TCP : build TCP segment and update TCP socket state
+        if( socket_type == SOCK_STREAM ) 
+        {
+            // extract one request from TCP socket R2T queue if queue non empty
+            if( remote_buf_status( socket_r2tq_xp ) )
+            {
+                remote_buf_get_to_kernel( socket_r2tq_xp , &r2t_flags , 1 );
+            }
+            else
+            {
+                r2t_flags = 0;
+            }
+
+            /////////////////////////////////////
+            if( socket_state == TCP_STATE_ESTAB )    // connected TCP socket
+            {
+                if( cmd == SOCKET_TX_SEND )          
+                {
+                    // get  "tx_nxt", and "tx_una" from socket descriptor
+                    tx_nxt = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->tx_nxt ));
+                    tx_una = hal_remote_l32( XPTR(socket_cxy , &socket_ptr->tx_una ));
+
+                    // compute actual payload length
+                    nbytes = ( PAYLOAD_MAX_LEN < todo ) ? PAYLOAD_MAX_LEN : todo;
+
+                    // compute TCP segment base in kernel buffer
+                    k_base = k_buf + ETH_HEAD_LEN + IP_HEAD_LEN;
+
+                    // move payload to kernel buffer
+                    hal_copy_from_uspace( XPTR( local_cxy , k_base + TCP_HEAD_LEN ),
+                                          buf + (len - todo),
+                                          nbytes );
+
+                    // build TCP header 
+                    dev_nic_tx_build_tcp_header( k_base,
+                                                 socket_xp,
+                                                 nbytes,                       // payload
+                                                 TCP_FLAG_ACK | r2t_flags );   // flags
+
+                    // update "tx_todo" in socket descriptor
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_todo ),
+                                    todo - nbytes );
+
+                    // update "tx_nxt" in socket descriptor
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                    tx_nxt + nbytes );
+
+                    // unblock client thread when SEND command completed
+                    if( (todo == 0) && (tx_nxt == tx_una) ) 
+                    {
+                        dev_nic_tx_report_success( socket_xp );
+                    }
+
+                    do_send = true;
+                }
+                else if( cmd == SOCKET_TX_CLOSE )          
+                {
+                    // build TCP FIN segment
+                    dev_nic_tx_build_tcp_header( k_base,
+                                                 socket_xp,
+                                                 0,                            // payload
+                                                 TCP_FLAG_FIN | r2t_flags );   // flags
+                    // update socket state
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_FIN_WAIT1 );
+
+                    do_send = true;
+                }
+                else  // cmd == CONNECT
+                {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp , cmd , socket_state );
+
+                    do_send = false;
+                }
+            }
+            //////////////////////////////////////////
+            else if( socket_state == TCP_STATE_BOUND )  // unconnected TCP socket
+            {
+                if ( cmd == SOCKET_TX_CONNECT )  
+                {
+                    // set socket.tx_nxt
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_nxt ),
+                                     TCP_ISS  );
+                     
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_nxt ), 0 );
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->rx_wnd ),
+                                    NIC_RX_BUF_SIZE);
+                                     
+                    // build TCP SYN segment
+                    dev_nic_tx_build_tcp_header( k_base,
+                                                  socket_xp,
+                                                  0,                // payload
+                                                  TCP_FLAG_SYN );   // flags
+                    // update socket state
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                     TCP_STATE_SYN_SENT );
+
+                    do_send = true;
+                }
+                else   // cmd == SEND / CLOSE
+                {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                    do_send = false;
+                }
+            }
+            ///////////////////////////////////////////
+            else if( socket_state == TCP_STATE_LISTEN )  // server wait connect
+            {
+                if( cmd == SOCKET_TX_CONNECT )          
+                {
+                    // update socket.state
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_SYN_SENT );
+                    
+                    // set socket.tx_una  
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_una ),
+                                    TCP_ISS );
+
+                    // set socket.tx_nxt 
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->tx_una ),
+                                    TCP_ISS + 1 );
+
+                    // build TCP SYN segment
+                    dev_nic_tx_build_tcp_header( k_base,
+                                                 socket_xp,
+                                                 0,                // payload
+                                                 TCP_FLAG_SYN );   // flags
+                    do_send = true;
+                }
+                else  // cmd == CLOSE / SEND
+                {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                    do_send = false;
+                }
+            }
+            /////////////////////////////////////////////
+            else if( socket_state == TCP_STATE_SYN_RCVD )  // socket wait ACK
+            {
+                if( cmd == SOCKET_TX_CLOSE )          
+                {
+                    // build TCP FIN segment
+                    dev_nic_tx_build_tcp_header( k_base,
+                                                 socket_xp,
+                                                 0,                // payload
+                                                 TCP_FLAG_FIN );   // flags
+                    // update socket state 
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_FIN_WAIT1 );
+
+                    do_send = true;
+                }
+                else  // SEND / CONNECT
+                {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                    do_send = false;
+                }
+            }
+            ////////////////////////////////////////////////
+            else if( socket_state == TCP_STATE_CLOSE_WAIT )  // wait local close()
+            {
+                if( cmd == SOCKET_TX_CLOSE )          
+                {
+                    // build TCP FIN segment
+                    dev_nic_tx_build_tcp_header( k_base,
+                                                 socket_xp,
+                                                 0,                // payload
+                                                 TCP_FLAG_FIN );   // flags
+                    // update socket state 
+                    hal_remote_s32( XPTR( socket_cxy , &socket_ptr->state ),
+                                    TCP_STATE_LAST_ACK );
+
+                    do_send = true;
+                }
+                else  // SEND / CONNECT
+                {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                    do_send = false;
+                }
+            }
+            ////
+            else
+            {
+                    // report illegal command
+                    dev_nic_tx_report_error( socket_xp, cmd, socket_state );
+
+                    do_send = false;
+            }
+
+            // compute TCP segment length
+            trsp_length = TCP_HEAD_LEN + nbytes;
+        }
+    }  
+    else                        // no valid command found
+    {
+        if( socket_type == SOCK_DGRAM )    //  UDP socket
+        {
+            do_send = false;
+        }
+        else                               //  TCP socket
+        {
+            if( remote_buf_status( socket_r2tq_xp ) == 0 )  // R2T queue empty
+            {
+                do_send = false;
+            }
+            else                                // pending request in R2T queue
+            {
+                // get one request from R2T queue 
+                remote_buf_get_to_kernel( socket_r2tq_xp , &r2t_flags , 1 );
+
+                // build TCP header for an empty segment
+                dev_nic_tx_build_tcp_header( k_base,
+                                             socket_xp,
+                                             0,             // payload
+                                             r2t_flags );   // flags
+                do_send = true;
+            }          
+        }
+    }
+
+    // 4. release the lock protecting the socket
+    remote_rwlock_wr_release( socket_lock_xp );
+
+    // return if no packet to send
+    if( do_send == false ) return;
+
+    // 5. build IP header
+    dev_nic_tx_build_ip_header( k_buf + ETH_HEAD_LEN,
+                                src_ip_addr,
+                                dst_ip_addr,
+                                IP_HEAD_LEN + trsp_length );
+
+    // 6. build ETH header
+    dev_nic_tx_build_eth_header( k_buf,
+                                 (uint16_t)SRC_MAC_54,
+                                 (uint16_t)SRC_MAC_32,
+                                 (uint16_t)SRC_MAC_10,
+                                 (uint16_t)DST_MAC_54,
+                                 (uint16_t)DST_MAC_32,
+                                 (uint16_t)DST_MAC_10,
+                                 ETH_HEAD_LEN + IP_HEAD_LEN + trsp_length );
+
+    // 7. move packet to NIC_TX queue
+    dev_nic_tx_move_packet( chdev,
+                            k_buf,
+                            ETH_HEAD_LEN + IP_HEAD_LEN + trsp_length );
+
+}  // end dev_nic_tx_handle_one_cmd()
+
+/////////////////////////////////////////
+void dev_nic_tx_server( chdev_t * chdev )
+{
+    uint8_t       k_buf[NIC_KERNEL_BUF_SIZE];  // buffer for one packet
+
+    xptr_t        root_xp;         // extended pointer on clients list root
+    xptr_t        lock_xp;         // extended pointer on lock protecting this list
+    xptr_t        socket_xp;       // extended pointer on on client socket 
+    socket_t    * socket_ptr;
+    cxy_t         socket_cxy;
+    xptr_t        entry_xp;        // extended pointer on socket tx_list entry
+
+    thread_t * this = CURRENT_THREAD;
+
+// check chdev direction and type
+assert( (chdev->func == DEV_FUNC_NIC) && (chdev->is_rx == false) ,
+"illegal chdev type or direction" );
+
+// check thread can yield
+assert( (this->busylocks == 0),
+"cannot yield : busylocks = %d\n", this->busylocks );
+
+    // build extended pointer on client sockets lock & root 
+    lock_xp = XPTR( local_cxy , &chdev->wait_lock );
+    root_xp = XPTR( local_cxy , &chdev->wait_root );
+
+    while( 1 )  // TX server infinite loop
+    {
+        // take the lock protecting the client sockets queue
+        remote_busylock_acquire( lock_xp );
+
+        /////////////// block and deschedule if no clients
+        if( xlist_is_empty( root_xp ) == false )
+        {
+            // release the lock protecting the TX client sockets queue
+            remote_busylock_release( lock_xp );
+ 
+            // block and deschedule
+            thread_block( XPTR( local_cxy , this ) , THREAD_BLOCKED_CLIENT );
+            sched_yield( "waiting client" );
+        }
+        //////////////  
+        else
+        {
+            // get first client socket
+            socket_xp  = XLIST_FIRST( root_xp , socket_t , tx_list );
+            socket_cxy = GET_CXY( socket_xp );
+            socket_ptr = GET_PTR( socket_xp );
+ 
+            // build extended pointer on socket xlist_entry
+            entry_xp = XPTR( socket_cxy , &socket_ptr->tx_list );
+
+            // remove this socket from the waiting queue
+            xlist_unlink( entry_xp );
+
+            // release the lock protecting the client sockets queue
+            remote_busylock_release( lock_xp );
+
+            // handle this TX client
+            dev_nic_tx_handle_one_cmd( socket_xp,
+                                       k_buf,
+                                       chdev ); 
+
+            // take the lock protecting the client sockets queue
+            remote_busylock_acquire( lock_xp );
+
+            // add this socket in last position of queue
+            xlist_add_last( root_xp , entry_xp ); 
+
+            // release the lock protecting the client sockets queue
+            remote_busylock_release( lock_xp );
+        }
+    }   // end while
+}  // end dev_nic_tx_server()
+
+