Re: [Help-smalltalk] problems with gst-2.0.8 on NetBSD/i386 (1.5W)

[Greg A. Woods]

[ On Friday, November 15, 2002 at 22:45:43 (+0100), Bonzini wrote: ]
> Subject: Re: [Help-smalltalk] problems with gst-2.0.8 on NetBSD/i386 (1.5W)
>
> > Next I ran into a much bigger problem:
> > 
> > SMALLTALK_KERNEL="`cd ./kernel; pwd`" \
> > SMALLTALK_IMAGE="`pwd`" \
> >   ./gst -iQ dummy_file
> > gst: Bus error
> > gst: Error occurred while not in byte code interpreter!!
> > Bus error
> > gmake[2]: *** [gst.im] Error 138
> > 
> > 
> > I rebuilt with '-g' and here's what gdb says:
> 
> Can you try gst 2.0.7?  If it succeeds, try searching for MAP_NORESERVE
> in libgst/heap.c, and add "#undef MAP_NORESERVE" before the first match.

Yes, this does the trick.

MAP_NORESERVE is defined in NetBSD's <sys/mman.h>:

        #define MAP_NORESERVE    0x0040 /* Sun: don't reserve needed swap area 
*/

but it's not mentioned in the mmap(2) manual page.

I'm guessing the (-1) I was seeing as a pointer value was MAP_FAILED:

        /*
         * Error indicator returned by mmap(2)
         */
        #define MAP_FAILED      ((void *) -1)   /* mmap() failed */

I see that there is one very critical places where mmap()'s return code
isn't checked for errors (in libgst/heap.c @ line 433):

  /* Get two reference addresses which we will base ourselves on */
  mmapGuess = higher = ANON_MMAP (NULL, pagesize, true); 

and I'm guessing that's the initial cause of the SIGBUS....

-- 
                                                                Greg A. Woods

+1 416 218-0098;            <address@hidden>;           <address@hidden>
Planix, Inc. <address@hidden>; VE3TCP; Secrets of the Weird <address@hidden>