21308 - Linker error in layout/build with SGI/IRIX 6.5

Status: VERIFIED FIXED

(Core :: Layout, defect, P3)

Description

[Chris Dolan]

I'm building with M11, so this *might* be out of date, but I checked
lxr.mozilla.org and I see no obvious changes.

When doing make libs in layout/build, the SGI linker generates MANY errors
trying to make libraptorhtml.so

--------------------------------------------------------------------------
c++ -elf -shared -o libraptorhtml.so    nsLayoutDLF.o nsLayoutFactory.o
nsLayoutModule.o    -Wl,-all ../../dist/lib/libraptorevents_s.a
../../dist/lib/libraptorhtmlbase_s.a ../../dist/lib/libnglhtmlcon_s.a
../../dist/lib/libraptorhtmldoc_s.a ../../dist/lib/libraptorhtmlforms_s.a
../../dist/lib/libraptorhtmlstyle_s.a ../../dist/lib/libraptorhtmltable_s.a
../../dist/lib/libraptorxmlcontent_s.a ../../dist/lib/libraptorxmldoc_s.a
../../dist/lib/libraptorxulcontent_s.a ../../dist/lib/libraptorxulbase_s.a
../../dist/lib/libraptorlayout_s.a  -Wl,-none ../../dist/lib/libneckoutil_s.a
../../dist/lib/libtimer_s.a -L../../dist/bin -lplds3 -lplc3 -lnspr3 -lpthread
-lxpcom -lraptorgfx -ljsdom -lmozjs
ld32: WARNING 84 : /usr/local/lib/gcc-lib/mips-sgi-irix6.5/2.95/libstdc++.a is
not used for resolving any symbol.
ld32: WARNING 127: Two shared objects with the same soname,
/usr/lib32/mips3/libm.so and /usr/lib32/libm.so, have been been linked. This is
probably due to a missing -L specification. Ignoring the latter.
ld32: INFO    171: Multigot invoked. Gp relative region broken up into 2
separate regions.
ld32: ERROR   104: GOT page/offset relocation out of range:
../../dist/lib/libraptorxmlcontent_s.a(nsXMLNamedNodeMap.o) section: .rela.text
relocation offset: 0x8a0.
Please use -LD_LAYOUT:lgot_buffer=20. If that doesn't work see the
gp_overflow(5) manpage.
--------------------------------------------------------------------------

My solution is to add "-Wl,-LD_LAYOUT:lgot_buffer=30" to the CFLAGS, which makes
it finish with just warnings, not errors (I got the 20 by trial and error).  So,
it compiles, but I haven't got mozilla to run due to other issues, so this may
just clean up the symptom, not the source of the problem...

I'd recommend (maybe) adding to the layout/build/Makefile something like:

ifeq ($(OS_ARCH),IRIX)
  CFLAGS += -Wl,-LD_LAYOUT:lgot_buffer=30
endif

Comment 1

[troy]

I don't know who this should be assigned to. It looks like a build problem under
SGI/IRIX

Comment 2

[rickg]

Hey Jim -- is this the kind of thing I can bribe you into taking care of for me?
I'd be very greatful -- and owe you a favor.

Comment 3

[Jim Dunn]

Unfortunately I no even less about Irix than I do about AIX.
I have asked briano his thoughts.

Comment 4

[Jim Dunn]

Jonathon Longman wrote:

I did a man gp_overflow as suggested in the output, which suggested checking 
question 22 in man dso.  This particularly looks like a
simple fix. 

from "man dso" on Irix 6.5: 

     22. What should I do about Global Offset Table (GOT) overflow?  GOT 
     overflow has occured if you receive messages from the linker saying 
     GP-relative sections overflow by 0x??? bytes, GOT overflow, or GOT 
     unreachable. 

     To fix this situation, perform one of more of the following steps: 

     * Recompile with the -TENV:large_got=ON option on your compiler 
       command line.  Only the large input file that is causing overflow 
       needs to be recompiled with this option. 

     * Break the large input file.o into two or more smaller files. 

     * Use the -m option on the ld(1) command to obtian a link map.  This 
       map indicates large objects that you can recompile with -G0 or some 
       other small -G value. 

       Data objects affected by the -Gnum option are numeric literals, 
       addreses (including those generated by the compiler), all C/C++ 
       static veriables, and, if the -static option is in effect, all 
       Fortran local variables.  For more information on the -Gnum option, 
       see your compiler command line. 

Comment 5

[Jim Dunn]

unfortunately at this poing I can't even get this far, due
to the compiler problem 28717

Comment 6

[aman]

Hi-

I was able to compile M16 on Irix using native compilers
which don't give the lgot problem. I also had to use
classic NSPR instead of pthreads (running with pthreads
enabled NSPR dumped core on pthread initialization). Another
change I had to do was not use the -none/-Wl,-none flag
to CC/ld when trying to produce the dso's; using this flag
resulted in a dso which didn't export all the symbols from
the archives intended to be contained in the dso; thereby,
I didn't provide any OS/standard libraries to resolve undefined
symbols in producing the dso itself (leaving them to be
resolved when producing the executables themselves).

Having created the dist/bin/mozilla-bin executable, running
it does some registery stuff and then returns with
"bus error (core dumped)" and unfortunately doesn't create
a core file at all. The registery stuff is done only for
the first time a newly produced executable is run.

I know its not enough information to debug this by.. but
I'm not familiar with mozilla code/structure etc. and doing
this entirely out of personal interest. If there are any
tests - running which would help people debug this, I might
be willing to run them, as and when I can spare time.

:a

Comment 7

[Jim Dunn]

Rich I am going to reassign to you since you are looking at this
much more than I am...

Comment 8

[John Vandenberg]

Trying to get the gcc build of Mozilla working on IRIX, I come across this 
again with the latest source, in content/build .  

---
gmake[2]: Entering directory 
`/projects/sise/mozilla/devel/workpits/moz/latest_gcc/workarea/content/build'
rm -f libgkcontent.so
g++  -fno-rtti -fno-exceptions -Wall -pedantic -Wno-long-long  -DDEBUG -
DDEBUG_johnv -DTRACING -g -elf -shared -o libgkcontent.so  nsContentDLF.o 
nsContentFactory.
o nsContentModule.o nsContentHTTPStartup.o    -Wl,-
all ../../dist/lib/libgkconevents_s.a ../../dist/lib/libgkconhtmlcon_s.a ../../d
ist/lib/libgkconhtmldoc_s.a ../.
./dist/lib/libgkconhtmlstyle_s.a ../../dist/lib/libgkconxmlcon_s.a ../../dist/li
b/libgkconxmldoc_s.a ../../dist/lib/libgkconxsldoc_s.a ../../dist/lib/libgkconxu
lco
n_s.a ../../dist/lib/libgkconxuldoc_s.a ../../dist/lib/libgkconxultmpl_s.a ../..
/dist/lib/libgkconxbl_s.a ../../dist/lib/libgkconbase_s.a ../../dist/lib/libgkco
nsh
ared_s.a  -Wl,-none -L../../dist/bin -ljsdom -lgkgfx -L../../dist/bin -lxpcom -
L../../dist/bin -
L/projects/sise/mozilla/devel/workpits/moz/latest_gcc/workarea/dist
/lib -lplds4 -lplc4 -lnspr4 -lpthread  -lmozjs   -lsocket -ldl -lm    
ld32: WARNING 84 : /usr/lib32/libsocket.so is not used for resolving any symbol.
ld32: WARNING 84 : /usr/lib32/libdl.so is not used for resolving any symbol.
ld32: WARNING 84 : /usr/freeware/lib/gcc-lib/mips-sgi-
irix6.2/2.95.2/libstdc++.a is not used for resolving any symbol.
ld32: WARNING 47 : This module (../../dist/lib/libgkconxuldoc_s.a
(nsXULDocument.o) .text) contains branch instruction(s) 
        that might degrade performance on an older version (rev. 2.2) R4000 
processor.
ld32: INFO    171: Multigot invoked. Gp relative region broken up into 2 
separate regions.
ld32: ERROR   104: GOT page/offset relocation out of 
range: ../../dist/lib/libgkconxuldoc_s.a(nsXULDocument.o) section: .rela.text 
relocation offset: 0xc6c.
Please use -LD_LAYOUT:lgot_buffer=20. If that doesn't work see the gp_overflow
(5) manpage.
ld32: ERROR   104: GOT page/offset relocation out of 
range: ../../dist/lib/libgkconxuldoc_s.a(nsXULDocument.o) section: .rela.text 
relocation offset: 0xdbc.
Please use -LD_LAYOUT:lgot_buffer=20. If that doesn't work see the gp_overflow
(5) manpage.
...
--- 
Using the same workaround, I had to add "-Wl,-LD_LAYOUT:lgot_buffer=50" .  As 
IRIX now has a MIPSpro build, the previous Makefile change is not sufficient, 
as this CFLAG should only be added if OS_ARCH = IRIX , and compiler is GNU.  I 
am not sure how the latter would be tested.

Comment 9

[John Vandenberg]

Looking for how to add to LDFLAGS (I presume this is a better candidate than 
CFLAGS) when OS is IRIX and compiler is g++, I have found a define 'GNU_CXX' in 
config/rules.mk , used in conjunction with SunOS.

http://lxr.mozilla.org/seamonkey/source/config/rules.mk#244

Tested this with gcc and MIPSpro, both now work.

Comment 10

[John Vandenberg]

Attachment: Patch to link using g++

Comment 11

[cls]

r=cls

Comment 12

[Asa Dotzler [:asa]]

a= [e-mail] for checkin to 0.9.1

Comment 13

[Richard Hess]

working on getting an ETA for this.  I'll update ASAP...

:-l

Comment 14

[cls]

Patch has been checked in.  Marking fixed.

Comment 15

[Chris Petersen]

Marking verified per last comments.