On 11/8/21 7:49 PM, Warner Losh wrote:
> > /* code path for having successfully executed the syscall */
> > + ADJUST_SYSCALL_RETCODE
> > ret
> >
> > 0:
>
> Not sure about this, really. Is it really that much cleaner to insert this than create
> separate 10-line files, with the adjustment included?
...
> The adjustments have all been 3 lines (gmail seems to hate my formatting):
>
> +#define ADJUST_SYSCALL_RETCODE \
> + jnb 2f; \
> + neg %rax; \
> + 2:
>
> which is significantly easier to maintain than having to monitor these files for changes
> and copying over the changes that happen.
...
> The other alternative I considered was having a #ifdef __FreeBSD__ .. #endif in all those
> files, but I thought that even more intrusive.
Actually, the ifdef sounds surprisingly attractive to me. Is it ENOCOFFEE?
What I find awkward about ADJUST_SYSCALL_RETCODE is that when you're looking at the
definition, you have no reference to the context, and vice versa. Not that it can't be
worked out, but it seems like the same amount of code either way, and clearer when it's
together.
We've already split the host cpu apart, which is the major point of ifdeffery, so it
doesn't seem like we'll wind up with a large amount of ifdefs here; we're not likely to
see mynewos-user wanting to share this code any time soon.
I feel sufficiently fuzzy on this to solicit other opinions though.
I'm OK with #ifdef. The macro hides the ifdef in a general way... but there's
always a problem generalizing from one example...
If nobody has a better idea, I'll respin with the #ifdefs in a day or two. That would also
give good sharing and be a couple of steps down the road towards getting a series of
signals patches queued up...
Warner