Re: [Gcl-devel] Support for little-endian ppc64

[Camm Maguire]

Greetings!

I think both patches are essentially identical, and work correctly, and
that this mprotect permission failure is unrelated, and stems from some
security setting.  If the reloc was wrong, you'd get an illegal
instruction error, or random corruption.  In the snippet below, it seems
as if many loads had proceeded correctly before this point.  This seems
to indicate that there is some permission problem in the particular
range being loaded at the moment, 0x14e0000.

Please let me know if you find that this is not the case.  If it is the
case, then it would appear gcl itself does not need to be changed.

I had access to a big endian ppc64, and can probably get one again if
needed, but I never needed the reloc in question, and never had mprotect
permission problems (Debian).  If you think it is profitable for me to
try to reproduce/test this, please let me know your compiler
flags/version that outputs this reloc, and I'll see if I can regain
access. 


Take care,

Jerry James <address@hidden> writes:

> On Mon, Sep 29, 2014 at 9:27 AM, Camm Maguire <address@hidden> wrote:
>
>     Greetings!  I have a slightly adapted version checked into
>     Version_2_6_12pre.  If you have a moment to verify that it works for
>     you, that would be wonderful.
>
> Sadly, no, that leads to this on big endian ppc64:
>
> Compiling GCL_PCL_LOW...
> Compiling /builddir/build/BUILD/gcl/unixport/../pcl/gcl_pcl_low.lisp.
> End of Pass 1.  
> End of Pass 2.  
> OPTIMIZE levels: Safety=1 (No runtime error checking), Space=0, Speed=3
> Finished compiling /builddir/build/BUILD/gcl/unixport/../pcl/gcl_pcl_low.o.
> Loading binary of GCL_PCL_LOW...
> Loading /builddir/build/BUILD/gcl/unixport/../pcl/gcl_pcl_low.o
> mprotect failure: 0x14e0000 65536 7: Permission denied
> Error: The assertion !clear_protect_memory(memory) on line 580 of sfaslelf.c 
> in function fasload failed
>
> Even more sadly, the patch I proposed worked for me when I did a test build, 
> but then failed in the same way when a member of Fedora's PPC team did
> a build later.  However, little endian ppc64 seems to work happily with both 
> versions of the patch.  So we're close, but not quite right somehow. 
> I am failing to understand what isn't quite right about either version of 
> this patch. :-(
> --
> Jerry James
> http://www.jamezone.org/
>

-- 
Camm Maguire                                        address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah