On Mon, Aug 02, 1999 at 07:11:42PM +0200, Andy Polyakov wrote:
> Hi again! As far as I can see you're not subscribed to the
> [EMAIL PROTECTED] list so you probably haven't seen following
> message posted couple of hours ago:

This is correct. Thanks Andy for passing this message on. To all other
who wishes to respond, please add [EMAIL PROTECTED] to the
Cc:-list.
 
>> From: Lutz Jaenicke <[EMAIL PROTECTED]>
>> > Do you really mean that when building openssl from object code
>> > at the size of 5 MB it actually needs over 65 MB to be linked??  What's
>> > the normal size of this binary?  >65MB?
>> 
>> at +O4 the optimization is done during the link step. The compilation is
>> fast, as it just parses the .c files and all the hard work is left
>> for the link stage.
>> During the link stage, the optimization is done on the complete target
>> executable at once and the HP-UX ANSI compiler is a real resource killer
>> then (in both time and memory requirements).

I've now changed the maxdsiz to 128MB (considering to increase it even
more), and I've patched the C/Ansi C compiler with patch PHSS_17507 (see
more on this further down). Then I did the linking of openssl and timed
it. Afterward I rebuild the whole thingy and watched the linking process
with top.

In both cases it used approx. 18 min, and the last 5-10 minutes it
actually had reserved 106 MB and the working set was 99,9 KB!  I find this
extreme... :-)  But it managed to finish.  So a big thank you to Andy
and Lutz!

As a sidenote when I rebuild everything it wasn't able to link without
error messages. After some tracking I found that having the option '-c'
to the final building of openssl is erroneuos and makes it do nothing.
Here is an excerpt from the man pages: 

     -c        Suppress the link edit phase of the compilation, and
               force an object (.o) file to be produced for each .c
               file even if only one program is compiled.  Object
               files produced from C programs must be linked before
               being executed.

The error stated itself as 'doing the link operation' in no time, and
not producing a binary... ;-/  

> > > I've already tried removing _all_ optimization. It still goes over the
> > > top...
> > Hmm, I don't have a machine with maxdsiz at 64MB. Are you sure that you
> > "recompiled" the .c to .o? Otherwise the linker will still take the information
> > stored in the ".o"s and will try to "+O4" it during the finale link stage.

I did a total rebuild, but it didn't help. Now I know why! Openssl needs
in excess of 100 MB to link on hpux 10.20...

> > And, please patch up your compiler to the latest patchlevel. You can
> > download the patches for free and I really had problems at high optimization
> > levels that went away with the latest patchlevels.

As stated before I applied PHSS_17507. This is a bundle and especially
PHSS_13421 seem interesting, here is a bit of the description of this
patch: 
  PHSS_13421:
      1) Memory leak problem in compiler optimizer
      2) compiled at +O2 or above, compiler aborts
         with  internal panic 6904
(And my original message included this 6904 error message! )

> > This one only for you, will summarize my efforts later on...
> But please do anyway! Any input is very much appreciated.

I hope someone will find this information useful. As to how to change
the maxdsiz kernel parameter I used the easy way of using sam, changing
kernel parameters. Followed by a build kernel and reboot.

I've not figured out where to remove the '-c', but I strongly suggest it
(along with adding my patch for locating the perl interpreter using
environment variables).

Regards,
Even Holen
-- 
<><   [EMAIL PROTECTED], ansatt i ITEA-drift   :-)
______________________________________________________________________
OpenSSL Project                                 http://www.openssl.org
Development Mailing List                       [EMAIL PROTECTED]
Automated List Manager                           [EMAIL PROTECTED]

Reply via email to