Ulrich Weigand wrote:
>
> > - (RvH) I changed host-linux.c so that it takes the amount of
> > quanta to run as an argument to ioctl(). Now it doesn't trap
> > back to the user application on ever quantum.
>
> Another point I just noticed: it doesn't even react on signals
> any more, so you can't stop the user process with Ctrl-C ;-)
> This is fixed by placing this check inside the loop:
>
> // Check for pending signals
> if (current->signal & ~current->blocked) {
> ret = -ERESTARTSYS;
> break;
> }
Okay. But the user application should be aware of signals:
on ctrl-c, it needs to tear down the vM, too, or we'll be stuck
with allocated data (is there no other way ? when I process is
killed, all its open files are released, no ? we should put
cleanup stuff in fmw_release()).
> Yet another point: the
>
> sti();
> soft_int(monitor_info.vector);
>
> creates a race condition, as the 'int' instruction is *not*
> the first instruction after the sti()! You'll need to place
> an sti assembly instruction inside the soft_int code, as the
> instruction immediately preceding the 'int'.
Okay.
> By the way, the soft_int self-modifying code is not particularly
> SMP-safe (but then again, the whole module isn't ...) ;-)
Hmmm... why not ? What can we do about that ?
Ramon
