[PR 17] GRETH: Fixed autonego timeout overflow problem

[Ralf Corsepius]

On 03/20/2012 08:51 AM, Daniel Hellstrom wrote:
> Hello Ralf,
>
> The problem was first reported by Joris, please see BUG 1465.
>
> - msecs =
> (tnow.tv_sec-tstart.tv_sec)*1000+(tnow.tv_usec-tstart.tv_usec)/1000;
> + msecs =
> (int)(tnow.tv_sec-tstart.tv_sec)*1000+((int)tnow.tv_usec-(int)tstart.tv_usec)/1000;
>
>
> The seconds part is no problem. If (tstart.tv_usec > tnow.tv_usec) will
> be negative, but since the variables are unsigned a underflow will
> result in a very long msecs timeout.

Do I understand this correctly, your problem is msecs being unsigned,
which means it can't take a negative time, the case which would happen
if (tstart > tnow)?

In this case, I'd propose to introduce a signed temporary variable to 
take the substraction's result and to special-case negative substraction 
results later on.

Rough outline:

    long _msecs = 
(tnow.tv_sec-tstart.tv_sec)*1000+(tnow.tv_usec-tstart.tv_usec)/1000;

    if (_msecs >= 0 ) {
      msecs = _msecs;
    } else {
      msecs = 0; /* or whatever might be suitable */
    }

Ralf