Re: [lwip-devel] Port on 16 bits processor: problem in ip4_addr_netmask_valid() function

[Giuseppe Modugno]

On 17 Feb 2011 at 13:58, Simon Goldschmidt wrote:

> "Giuseppe Modugno" <address@hidden> wrote:

> > [...]

> > The operation 1U << 31 can't be well performed on a 16 bits

> > processor. It should be (u32_t)1U << 31. What do you think?

>

> I guess it should rather be '1UL' instead of '(u32_t)1U' (which should also work...). Does 1UL yield to a 32-bit constant for your controller?

Yes, you're right. It's better '1UL'.

> BTW: This should be fixed for the 1.4.0 release, since it's really a small fix.

>

> Also, please let us know if you run into any other problems regarding the 16-bit processor, as there haven't been many people active on the lists using 16-bit processors lately.

I have found another issue in ip_addr.c. Consider the following instruction:

  if ((IPH_OFFSET(iphdr) & PP_HTONS(IP_OFFMASK | IP_MF)) != 0) {

In def.h:

  #define PP_HTONS(x) ((((x) & 0xff) << 8) | (((x) & 0xff00) >> 8))

In ip.h:

  #define IP_OFFMASK 0x1fff   /* mask for fragmenting bits */

  #define IP_MF 0x2000        /* more fragments flag */

I think IP_OFFMASK and IP_MF (and others) should be declared unsigned ('U' suffix), otherwise the left shifting by 8 positions can could wrong for a signed int.

What do you think?