Re: [lwip-users] Odd byte packets in LWIP

[Konrad, Guido]

I also faced this problem with the c6900.c driver code  from Leon. Our HW interface

allows only 16 bit writes. I implemented a quick and dirty solution which makes copies

if there is an odd number of characters in one buffer. (lwip 0.5.3). Attached you will

find my output function. Perhaps it is in some kind usefull even if it kills performance.

If there is a better solution please let me know.

 

-- Guido

 

static err_t cs8900_output(struct cs8900if *cs8900if, struct pbuf *p)
{
    struct pbuf *q;
    u32_t i;
    u16_t *ptr, val;
    u8_t  *ptr8, buf[TCP_MSS];
    u32_t len;
    int tries = 0, valHasValue = 0;
    INT8U  ucErr;
   
    OSMutexPend( pDeviceMut, 0, &ucErr );
   
    // exit if link has failed
    PACKETPP = CS_PP_LINESTATUS;
    if ((PPDATA & 0x0080U/*LinkOK*/) == 0) {
        (void)OSMutexPost( pDeviceMut );
        return ERR_MEM; // TODO: find a correct error code
    }
   
    /* transmit command */
    TXCMD = 0x00C9U;
    TXLENGTH = p->tot_len;
   
    PACKETPP = CS_PP_BUSSTATUS;
    // not ready for transmission and still within 100 retries?
    while(((PPDATA & 0x0100U/*Rdy4TxNOW*/) == 0) && (tries++ < 100))
    {
        // throw away the last committed received frame
        PACKETPP = CS_PP_RXCFG;
        PPDATA = (0x0003U | 0x0040U/*Skip_1*/ | 0x0100U/*RxOKiE*/);
        PACKETPP = CS_PP_BUSSTATUS;
        /* cs8900if->dropped++; CHECK: we do not know if we actually will drop a frame here */
    }
    // ready to transmit?
    if((PPDATA & 0x0100U/*Rdy4TxNOW*/) != 0)
    {
        for(q = p; q != NULL; q = q->next)
        {
            DEBUGF(ETH_DEBUG, ("cs8900_output: pbuf @%p len %d\n", q, q->len));
           
            /* Send the data from the pbuf to the interface, one pbuf at a
            time. The size of the data in each pbuf is kept in the ->len
            variable. */
            /* Check alignement */
            if( (((int)q->payload) & 0x01 ) == 0x01 ) {
                DEBUGF(ETH_DEBUG, ("cs8900_output: buffer not aligned\n")); 
            }
            if( !q->len ) continue;
           
            if( valHasValue ) {
                ptr8 = (u8_t *)q->payload;
                val |= (*ptr8)<<8;
                len = --(q->len);
                for( i = 1; i <= len; i++ ) {
                    buf[i-1] = ptr8[i];
                }
                ptr = (u16_t *)buf;
                len &= (u32_t)0xFFFFFFFE;
                valHasValue = 2;
                /* Transmit value */
                RXTXREG = val;
            } else {
                ptr = (u16_t *)q->payload;
                len = q->len;
                len &= (u32_t)0xFFFFFFFE;
            }
            for(i = 0; i < len; i += 2) {
                RXTXREG = *ptr++;
            }
                      
            if( len != q->len ) {
                val = (u16_t)(*(u8_t*)ptr);
                if( valHasValue == 2 ) {
                    ++(q->len);
                }
                valHasValue = 1;
            } else {
                if( valHasValue == 2 ) {
                    ++(q->len);
                    valHasValue = 0;
                }           
            }
        }
        if( valHasValue ) {
            RXTXREG = val;
        }
    } else {
        DEBUGF(ETH_DEBUG, ("cs8900_output: device not ready to send pbuf @%p len %d\n", q, q->len)); 
    }
   
    (void)OSMutexPost( pDeviceMut );
   
    return ERR_OK;
}

-----Ursprüngliche Nachricht-----
Von: Nishen Naidu [mailto:address@hidden
Gesendet: Dienstag, 11. November 2003 07:29
An: Mailing list for lwIP users
Betreff: [lwip-users] Odd byte packets in LWIP

Hi

 

I'm having a problem sending out packets of odd length on certain occasions.

Does lwip work well with odd length packets?

I read the following comment in some lwip sample code:

/*

 

Important!!! LwIP data alignment issue:
   As of now LwIP code does not handle odd byte length packets properly. So round packets to even byte
   length by adding space characters (' '). This is true only for processors that cannot handle
   mis-aligned memory access.

 

*/

is this correct??

 

Thanks

Nishen