[kaffe] CVS kaffe (dalibor): Fixed build on x86_64-linux

Gwenole Beauchesne gbeauchesne at mandrakesoft.com
Thu Jul 8 03:16:53 PDT 2004


On Wed, 7 Jul 2004, Dalibor Topic wrote:

> > Latest version that fully built and fully regtested was 1.1.4. Looking at 
> > the changes...
> 
> Thanks a lot for taking your time to look at them.

Well, I had to reschedule it up to this morning as higher priority items
came in the "dichotomy" process. ;-)

Narrowed to the following change:

2004-05-30  Guilhem Lavaux <guilhem at kaffe.org>

	* kaffe/kaffevm/systems/unix-jthreads/signal.c: As we
	don't need STACK_POINTER anymore for the detection we
	may relax the condition.

i.e. using the worse detectStackBoundaries() heuristic will no longer
hang. Does the SIGSEGV-based heuristic work on other 64-bit platforms?

Two possible solutions:
- provide a KAFFEMD_STACK{BASE,END} implementation
- fix the SIGSEGV based method

Hmm, you use attribute unused, which gcc really interprets it as unused 
nowadays, this optimizing out the load, hence the test.

> On a side note, Guilhem was wondering on IRC yesterday whether my
> approach of using struct sigcontext instead of ucontext (as
> sigcontextinfo.h from gnu libc does) was appropriate. So that might be a
> hint for an area to look for bugs.

Actually, I don't remember why I used sigcontext as I always used
ucontext_t in my other projects. The kernel sends an ucontext_t* as the 
3rd arg to a SIGINFO signal handler.

Do you know how to excercise SIGNAL_PC(), STACK_POINTER() in Kaffe?

Bye,
Gwenolé.




More information about the kaffe mailing list