MIPS GOT overflow in gcc 3.2.

MIPS GOT overflow in gcc 3.2.

[H. J. Lu]

On Mon, Jul 01, 2002 at 06:29:24PM -0700, H. J. Lu wrote:
> 
> libjava failed to build. I got so many
> 
> .libs/prims.o: In function `_Jv_ThrowBadArrayIndex':
> /home/hjl/work/gnu/src/gcc/gcc/libjava/prims.cc:313: relocation truncated to
> fit: R_MIPS_GOT16 java::lang::ArrayIndexOutOfBoundsException::class$
> .libs/prims.o: In function `_Jv_ThisExecutable(char const*)':
> /home/hjl/work/gnu/src/gcc/gcc/libjava/prims.cc:722: relocation truncated to
> fit: R_MIPS_CALL16 strcpy
> .libs/jni.o: In function `mangled_name':
> /home/hjl/work/gnu/src/gcc/gcc/libjava/gcj/cni.h:70: relocation truncated to
> fit: R_MIPS_CALL16 java::lang::Class::getName()
> .libs/jni.o: In function `mangled_name':
> :50: relocation truncated to fit: R_MIPS_CALL16 java::lang::Class::getName()
> 
> I don't know if it is due to GOT overflow or something else. gcc 3.1.1
> builds fine.
> 

For gcc 3.1.1, I got

  [17] .got              PROGBITS        0068b8a0 64b8a0 00ff90 04 WAp  0   0
  16

0x00ff90 is very close to 64K. I thought it would only happen to KDE.



H.J.

Re: MIPS GOT overflow in gcc 3.2.

[Eric Christopher]

> For gcc 3.1.1, I got
> 
>   [17] .got              PROGBITS        0068b8a0 64b8a0 00ff90 04 WAp  0   0
>   16
> 
> 0x00ff90 is very close to 64K. I thought it would only happen to KDE.

AFAIK it happens to mozilla as well.

Guh.

-eric

-- 
I will not carve gods

Re: MIPS GOT overflow in gcc 3.2.

[Eric Christopher]

> AFAIK it happens to mozilla as well.
> 
> Guh.

There are a few different possible solutions, one is to do the
-fPIC/-fpic split, another is to copy to SGI multigot, I'm sure there
are other solutions as well...

-eric

-- 
I will not carve gods

Re: MIPS GOT overflow in gcc 3.2.

[H. J. Lu]

On Mon, Jul 01, 2002 at 08:09:59PM -0700, Eric Christopher wrote:
> 
> > AFAIK it happens to mozilla as well.
> > 
> > Guh.
> 
> There are a few different possible solutions, one is to do the
> -fPIC/-fpic split, another is to copy to SGI multigot, I'm sure there
> are other solutions as well...
> 

I am testing this patch now. It will take a day to verify it.


H.J.
----
2002-07-02  H.J. Lu <hjl@gnu.org>

	* ltcf-gcj.sh (ac_cv_prog_cc_pic): Add "-Wa,-xgot" for
	Linux/mips.

--- ltcf-gcj.sh.xgot	Mon Sep 10 08:39:17 2001
+++ ltcf-gcj.sh	Mon Jul  1 23:49:06 2002
@@ -639,6 +639,14 @@ fi
 	 ac_cv_prog_cc_pic=-Kconform_pic
       fi
       ;;
+    linux*) 
+      ac_cv_prog_cc_pic='-fPIC'
+      case $host_cpu in
+      mips*)  
+	ac_cv_prog_cc_pic="$ac_cv_prog_cc_pic -Wa,-xgot"
+	;;      
+      esac
+      ;;      
     *)
       ac_cv_prog_cc_pic='-fPIC'
       ;;

RFC: Use -Wa,-xgot for Linux/mips (Re: MIPS GOT overflow in gcc 3.2.)

[H. J. Lu]

[-- Attachment #1: Type: text/plain, Size: 431 bytes --]

On Mon, Jul 01, 2002 at 08:09:59PM -0700, Eric Christopher wrote:
> 
> > AFAIK it happens to mozilla as well.
> > 
> > Guh.
> 
> There are a few different possible solutions, one is to do the
> -fPIC/-fpic split, another is to copy to SGI multigot, I'm sure there
> are other solutions as well...
> 

I am enclosing a kludge here. With that, I got

http://gcc.gnu.org/ml/gcc-testresults/2002-07/msg00062.html

Any comments?


H.J.

[-- Attachment #2: libjava-xgot.patch --]
[-- Type: text/plain, Size: 1429 bytes --]

2002-07-02  H.J. Lu <hjl@gnu.org>

	* ltcf-c.sh (ac_cv_prog_cc_pic): Add "-Wa,-xgot" for
	Linux/mips.
	* ltcf-cxx.sh (ac_cv_prog_cc_pic): Likewise.
	* ltcf-gcj.sh (ac_cv_prog_cc_pic): Likewise.

--- gcc/ltcf-c.sh.xgot	Mon Sep 10 08:39:17 2001
+++ gcc/ltcf-c.sh	Tue Jul  2 00:06:54 2002
@@ -672,6 +672,14 @@ else
 	 ac_cv_prog_cc_pic=-Kconform_pic
       fi
       ;;
+    linux*) 
+      ac_cv_prog_cc_pic='-fPIC'
+      case "$host_cpu" in
+      mips*)  
+	ac_cv_prog_cc_pic="$ac_cv_prog_cc_pic -Wa,-xgot"
+	;;      
+      esac
+      ;;      
     *)
       ac_cv_prog_cc_pic='-fPIC'
       ;;
--- gcc/ltcf-cxx.sh.xgot	Mon Sep 10 08:39:17 2001
+++ gcc/ltcf-cxx.sh	Tue Jul  2 13:40:41 2002
@@ -707,6 +707,14 @@ if test "$with_gcc" = yes; then
       ac_cv_prog_cc_pic=-Kconform_pic
     fi
     ;;
+  linux*) 
+    ac_cv_prog_cc_pic='-fPIC'
+    case "$host_cpu" in
+    mips*)  
+      ac_cv_prog_cc_pic="$ac_cv_prog_cc_pic -Wa,-xgot"
+      ;;      
+    esac
+    ;;      
   *)
     ac_cv_prog_cc_pic='-fPIC'
     ;;
--- gcc/ltcf-gcj.sh.xgot	Mon Sep 10 08:39:17 2001
+++ gcc/ltcf-gcj.sh	Mon Jul  1 23:49:06 2002
@@ -639,6 +639,14 @@ fi
 	 ac_cv_prog_cc_pic=-Kconform_pic
       fi
       ;;
+    linux*) 
+      ac_cv_prog_cc_pic='-fPIC'
+      case "$host_cpu" in
+      mips*)  
+	ac_cv_prog_cc_pic="$ac_cv_prog_cc_pic -Wa,-xgot"
+	;;      
+      esac
+      ;;      
     *)
       ac_cv_prog_cc_pic='-fPIC'
       ;;

Re: RFC: Use -Wa,-xgot for Linux/mips (Re: MIPS GOT overflow in gcc 3.2.)

[Daniel Jacobowitz]

On Wed, Jul 03, 2002 at 09:35:18AM -0700, H. J. Lu wrote:
> On Mon, Jul 01, 2002 at 08:09:59PM -0700, Eric Christopher wrote:
> > 
> > > AFAIK it happens to mozilla as well.
> > > 
> > > Guh.
> > 
> > There are a few different possible solutions, one is to do the
> > -fPIC/-fpic split, another is to copy to SGI multigot, I'm sure there
> > are other solutions as well...
> > 
> 
> I am enclosing a kludge here. With that, I got
> 
> http://gcc.gnu.org/ml/gcc-testresults/2002-07/msg00062.html
> 
> Any comments?

Did you see my message about mixing GOT models?  I'd need Ralf or Eric to
confirm this, but I'm pretty sure there's a lot of luck involved in
your workaround.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: RFC: Use -Wa,-xgot for Linux/mips (Re: MIPS GOT overflow in gcc 3.2.)

[H. J. Lu]

On Wed, Jul 03, 2002 at 12:40:39PM -0400, Daniel Jacobowitz wrote:
> On Wed, Jul 03, 2002 at 09:35:18AM -0700, H. J. Lu wrote:
> > On Mon, Jul 01, 2002 at 08:09:59PM -0700, Eric Christopher wrote:
> > > 
> > > > AFAIK it happens to mozilla as well.
> > > > 
> > > > Guh.
> > > 
> > > There are a few different possible solutions, one is to do the
> > > -fPIC/-fpic split, another is to copy to SGI multigot, I'm sure there
> > > are other solutions as well...
> > > 
> > 
> > I am enclosing a kludge here. With that, I got
> > 
> > http://gcc.gnu.org/ml/gcc-testresults/2002-07/msg00062.html
> > 
> > Any comments?
> 
> Did you see my message about mixing GOT models?  I'd need Ralf or Eric to

A testcase?


H.J.

Re: RFC: Use -Wa,-xgot for Linux/mips (Re: MIPS GOT overflow in gcc 3.2.)

[Ralf Baechle]

On Wed, Jul 03, 2002 at 12:40:39PM -0400, Daniel Jacobowitz wrote:

> Did you see my message about mixing GOT models?  I'd need Ralf or Eric to
> confirm this, but I'm pretty sure there's a lot of luck involved in
> your workaround.

I've not seen your posting but yes, due to other static objects involved
into linking there definately is luck involved.

  Ralf