Re: [lwip-users] Possible alignment bug in 1.4.0?

[Mason]

Krzysztof Wesołowski wrote:

> I have encountered some HardFaults in rcurrent 1.4.0 version, file
> etharp.c code, line 416.
> 
> code (stripped from asserts):
> static err_t
> etharp_send_ip(struct netif *netif, struct pbuf *p, struct eth_addr
> *src, struct eth_addr *dst)
> {
>   struct eth_hdr *ethhdr = (struct eth_hdr *)p->payload;
>   ETHADDR32_COPY(&ethhdr->dest, dst);
>   ETHADDR16_COPY(&ethhdr->src, src);
> 
>   ethhdr->type = PP_HTONS(ETHTYPE_IP);
>   return netif->linkoutput(netif, p);
> }
> 
> Assumes that &ethhdr->dest is 32bit aligned, which is true only for
> ETH_PAD_SIZE==0.

Did you define your own ETHADDR32_COPY macro?
Or your own SMEMCPY macro?
(memcpy is required to work for any alignment.)

/** MEMCPY-like macro to copy to/from struct eth_addr's that are local variables
 * or known to be 32-bit aligned within the protocol header. */
#ifndef ETHADDR32_COPY
#define ETHADDR32_COPY(src, dst)  SMEMCPY(src, dst, ETHARP_HWADDR_LEN)
#endif

/** MEMCPY-like macro to copy to/from struct eth_addr's that are no local
 * variables and known to be 16-bit aligned within the protocol header. */
#ifndef ETHADDR16_COPY
#define ETHADDR16_COPY(src, dst)  SMEMCPY(src, dst, ETHARP_HWADDR_LEN)
#endif

/**
 * SMEMCPY: override this with care! Some compilers (e.g. gcc) can inline a
 * call to memcpy() if the length is known at compile time and is small.
 */
#ifndef SMEMCPY
#define SMEMCPY(dst,src,len)            memcpy(dst,src,len)
#endif

In principle, you're right. If one defines ETH_PAD_SIZE to 2
to align the IP header to 32 bits, then the first field of the
Ethernet header (MAC dest) cannot be aligned to 32 bits.

-- 
Regards.