Re: return CERROR result in a0 too (was: More toolchain issues)

["Michael L. Hitch" <mhitch%lightning.msu.montana.edu@localhost>]

On Wed, 7 Oct 2009, John Klos wrote:

> It seems that the -current sets I'm runnning have the change, but the problem 
> still exists, even though some symptoms are fixed.

  What problem still exists?  The part about Franks program all looks like 
it working 'properly'

> Running Frank's program without unlimiting gives me:
>
> ...
> 88 MB
> 89 MB
> 90 MB
> 91 MB
> 92 MB
> 93 MB
> Segmentation fault (core dumped)
>
> When the program gets to 93 megs, it chews on CPU for quite a while before it 
> dumps core:

  Franks program attempts to write a byte to the address returned by 
malloc(), but does not check for a NULL return, and will always cause a 
segment violation.

  93 MB would be about right with the default data size.  This is because 
jemalloc() will begin allocating from the memory following the shared 
libraries up to the beginning of the user stack, which accounts for around 
64 MB.  It then starts allocating data space up to the 32MB default limit
before returning NULL.

  I'm not certain exactly what it might be doing once it faults, but it 
will seit there for a while running while it's dumping memory.

> When I unlimit, it runs to 185 megs or so before it starts hitting swap.

  Again, it's using the ~64MB space up to the user stack, then allocates 
from the 128MB data limit.  At some point it's going to run out of 
physical memory and begin swapping and will eventually segment fault and 
dump core.

> However, pkgsrc/devel/pcre still fails, and I have to wonder why the process 
> limits aren't honored...

  I was able to build pkgsrc/devel/pcre on my A4000 with a CyberStorm, so 
I'm not sure what this part of the problem might be.  [My build did have 
-o2 -m68060 options set though.]

  Process limits only account for the data area used by brk(), but not 
mmap().  [Do you account for all shared library space as coming from the 
processes data limit?]

--
Michael L. Hitch                        mhitch%montana.edu@localhost
Computer Consultant
Information Technology Center
Montana State University        Bozeman, MT     USA