libjava failed to build on ia64

libjava failed to build on ia64

[H. J. Lu]

As of Tue May 13 14:38:59 PDT 2003, gcc mainline failed in libjava:

gnu/java/locale/.libs/LocaleInformation_en_PH.o(.text+0x52): In function
`gnu::java::locale::LocaleInformation_en_PH::<clinit>()':
/net/gnu/export/gnu/src/gcc/gcc/libjava/gnu/java/locale/LocaleInformation_en_PH.java:47:
relocation truncated to fit: GPREL22 .text
collect2: ld returned 1 exit status
make[5]: *** [libgcj.la] Error 1

I am not sure if it is the right usage of LTOFF22X

0000000000000050
<_ZN3gnu4java6locale23LocaleInformation_en_PH18__U3c_clinit__U3e_Ev>:
  50:   08 30 ae 50 80 05       [MMI]       alloc r70=ar.pfs,43,40,0
                        52: LTOFF22X    .data.rel.local
  56:   c0 00 32 7c 46 00                   adds r12=-192,r12
  5c:   04 08 00 90                         addl r32=0,r1
  60:   09 08 02 02 00 24       [MMI]       addl r65=0,r1
                        60: LTOFF22X
_ZN3gnu4java6locale23LocaleInformation_en_PH16decimalSeparatorE
                        61: LTOFF22X
_ZN3gnu4java6locale23LocaleInformation_en_PH17groupingSeparatorE
                        62: LTOFF22X
_ZN3gnu4java6locale23LocaleInformation_en_PH12numberFormatE
  66:   20 04 04 00 48 60                   addl r66=0,r1
  6c:   08 08 00 90                         addl r67=0,r1;;


H.J.

Re: libjava failed to build on ia64

[Richard Henderson]

On Tue, May 13, 2003 at 04:36:47PM -0700, H. J. Lu wrote:
> I am not sure if it is the right usage of LTOFF22X

It is.  You're allowed to use it with any operand whatsoever.


r~

PATCH: Fix the relax finalize pass

[H. J. Lu]

[-- Attachment #1: Type: text/plain, Size: 453 bytes --]

On Wed, May 14, 2003 at 12:53:30AM -0700, Richard Henderson wrote:
> On Tue, May 13, 2003 at 04:36:47PM -0700, H. J. Lu wrote:
> > I am not sure if it is the right usage of LTOFF22X
> 
> It is.  You're allowed to use it with any operand whatsoever.
> 
> 

The problem is linker keeps changing data segment address after the
relax finalize pass starts. It makes the GP calculation in the relax
finalize pass invalid. This patch tries to avoid it.


H.J.

[-- Attachment #2: ld-final.patch --]
[-- Type: text/plain, Size: 11106 bytes --]

2003-05-14  H.J. Lu <hongjiu.lu@intel.com>

	* ldlang.c (lang_size_sections_1): Take one more argument to
	indicate if the relax finalize pass is needed.
	(lang_size_sections): Updated.
	(lang_process): Likewise.
	* ldlang.h (lang_size_sections_1): Likewise.
	* pe-dll.c (pe_dll_fill_sections): Likewise.
	(pe_exe_fill_sections): Likewise.
	* emultempl/elf32.em (gld${EMULATION_NAME}_finish): Likewise.
	* emultempl/hppaelf.em (hppaelf_layout_sections_again): Likewise.
	* emultempl/ppc64elf.em (ppc_before_allocation): Likewise.
	(ppc_layout_sections_again): Likewise.

	* ldlang.c (lang_size_sections): Don't adjust data segment
	address after the relax finalize pass starts.
	(lang_process): Perform the relax finalize pass only when
	needed. Finalize addresses before the relax finalize pass.

--- ld/emultempl/elf32.em.final	2003-05-09 08:13:31.000000000 -0700
+++ ld/emultempl/elf32.em	2003-05-14 15:21:48.000000000 -0700
@@ -1430,7 +1430,8 @@ gld${EMULATION_NAME}_finish ()
 
       /* Resize the sections.  */
       lang_size_sections (stat_ptr->head, abs_output_section,
-			  &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+			  &stat_ptr->head, 0, (bfd_vma) 0, NULL,
+			  NULL, TRUE);
 
       /* Redo special stuff.  */
       ldemul_after_allocation ();
--- ld/emultempl/hppaelf.em.final	2003-02-28 12:53:34.000000000 -0800
+++ ld/emultempl/hppaelf.em	2003-05-14 15:21:57.000000000 -0700
@@ -231,7 +231,8 @@ hppaelf_layout_sections_again ()
 
   /* Resize the sections.  */
   lang_size_sections (stat_ptr->head, abs_output_section,
-		      &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+		      &stat_ptr->head, 0, (bfd_vma) 0, NULL, NULL,
+		      TRUE);
 
   /* Redo special stuff.  */
   ldemul_after_allocation ();
--- ld/emultempl/ppc64elf.em.final	2003-02-28 12:53:38.000000000 -0800
+++ ld/emultempl/ppc64elf.em	2003-05-14 15:22:16.000000000 -0700
@@ -118,7 +118,8 @@ ppc_before_allocation ()
 	  /* Size the sections.  This is premature, but we want to know the
 	     TLS segment layout so that certain optimizations can be done.  */
 	  lang_size_sections (stat_ptr->head, abs_output_section,
-			      &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+			      &stat_ptr->head, 0, (bfd_vma) 0,
+			      NULL, NULL, TRUE);
 
 	  if (!ppc64_elf_tls_optimize (output_bfd, &link_info))
 	    {
@@ -273,7 +274,8 @@ ppc_layout_sections_again ()
 
   /* Resize the sections.  */
   lang_size_sections (stat_ptr->head, abs_output_section,
-		      &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+		      &stat_ptr->head, 0, (bfd_vma) 0,
+		      NULL, NULL, TRUE);
 
   /* Recalculate TOC base.  */
   ldemul_after_allocation ();
--- ld/ldlang.c.final	2003-05-13 08:45:34.000000000 -0700
+++ ld/ldlang.c	2003-05-14 15:36:47.000000000 -0700
@@ -199,9 +199,10 @@ static void os_region_check
   PARAMS ((lang_output_section_statement_type *,
 	   struct memory_region_struct *, etree_type *, bfd_vma));
 static bfd_vma lang_size_sections_1
-  PARAMS ((lang_statement_union_type *, lang_output_section_statement_type *,
-	   lang_statement_union_type **, fill_type *, bfd_vma, bfd_boolean *,
-	   bfd_boolean));
+  PARAMS ((lang_statement_union_type *,
+	   lang_output_section_statement_type *,
+	   lang_statement_union_type **, fill_type *, bfd_vma,
+	   bfd_boolean *, bfd_boolean *, bfd_boolean));
 typedef void (*callback_t)
   PARAMS ((lang_wild_statement_type *, struct wildcard_list *, asection *,
 	   lang_input_statement_type *, PTR));
@@ -2978,14 +2979,15 @@ os_region_check (os, region, tree, base)
 /* Set the sizes for all the output sections.  */
 
 static bfd_vma
-lang_size_sections_1 (s, output_section_statement, prev, fill, dot, relax,
-		      check_regions)
+lang_size_sections_1 (s, output_section_statement, prev, fill, dot,
+		      relax, need_finalize_relax, check_regions)
      lang_statement_union_type *s;
      lang_output_section_statement_type *output_section_statement;
      lang_statement_union_type **prev;
      fill_type *fill;
      bfd_vma dot;
      bfd_boolean *relax;
+     bfd_boolean *need_finalize_relax;
      bfd_boolean check_regions;
 {
   unsigned opb = bfd_arch_mach_octets_per_byte (ldfile_output_architecture,
@@ -3123,8 +3125,10 @@ lang_size_sections_1 (s, output_section_
 		os->bfd_section->output_offset = 0;
 	      }
 
-	    lang_size_sections_1 (os->children.head, os, &os->children.head,
-				  os->fill, dot, relax, check_regions);
+	    lang_size_sections_1 (os->children.head, os,
+				  &os->children.head, os->fill, dot,
+				  relax, need_finalize_relax,
+				  check_regions);
 
 	    /* Put the section within the requested block size, or
 	       align at the block boundary.  */
@@ -3193,7 +3197,9 @@ lang_size_sections_1 (s, output_section_
 	  dot = lang_size_sections_1 (constructor_list.head,
 				      output_section_statement,
 				      &s->wild_statement.children.head,
-				      fill, dot, relax, check_regions);
+				      fill, dot, relax,
+				      need_finalize_relax,
+				      check_regions);
 	  break;
 
 	case lang_data_statement_enum:
@@ -3257,7 +3263,9 @@ lang_size_sections_1 (s, output_section_
 	  dot = lang_size_sections_1 (s->wild_statement.children.head,
 				      output_section_statement,
 				      &s->wild_statement.children.head,
-				      fill, dot, relax, check_regions);
+				      fill, dot, relax,
+				      need_finalize_relax,
+				      check_regions);
 
 	  break;
 
@@ -3286,6 +3294,7 @@ lang_size_sections_1 (s, output_section_
 		  einfo (_("%P%F: can't relax section: %E\n"));
 		if (again)
 		  *relax = TRUE;
+		*need_finalize_relax |= i->need_finalize_relax;
 	      }
 	    dot = size_input_section (prev, output_section_statement,
 				      output_section_statement->fill, dot);
@@ -3355,7 +3364,9 @@ lang_size_sections_1 (s, output_section_
 	  dot = lang_size_sections_1 (s->group_statement.children.head,
 				      output_section_statement,
 				      &s->group_statement.children.head,
-				      fill, dot, relax, check_regions);
+				      fill, dot, relax,
+				      need_finalize_relax,
+				      check_regions);
 	  break;
 
 	default:
@@ -3372,22 +3383,28 @@ lang_size_sections_1 (s, output_section_
 }
 
 bfd_vma
-lang_size_sections (s, output_section_statement, prev, fill, dot, relax,
-		    check_regions)
+lang_size_sections (s, output_section_statement, prev, fill, dot,
+		    relax, need_finalize_relax, check_regions)
      lang_statement_union_type *s;
      lang_output_section_statement_type *output_section_statement;
      lang_statement_union_type **prev;
      fill_type *fill;
      bfd_vma dot;
      bfd_boolean *relax;
+     bfd_boolean *need_finalize_relax;
      bfd_boolean check_regions;
 {
   bfd_vma result;
 
   exp_data_seg.phase = exp_dataseg_none;
-  result = lang_size_sections_1 (s, output_section_statement, prev, fill,
-				 dot, relax, check_regions);
-  if (exp_data_seg.phase == exp_dataseg_end_seen)
+  result = lang_size_sections_1 (s, output_section_statement, prev,
+				 fill, dot, relax, need_finalize_relax,
+				 check_regions);
+
+  /* We can't change addresses after the relax finalize pass starts
+     since the relaxation may assume addresses won't be changes.  */
+  if (!link_info.relax_finalizing
+      && exp_data_seg.phase == exp_dataseg_end_seen)
     {
       /* If DATA_SEGMENT_ALIGN DATA_SEGMENT_END pair was seen, check whether
 	 a page could be saved in the data segment.  */
@@ -3401,8 +3418,10 @@ lang_size_sections (s, output_section_st
 	  && first + last <= exp_data_seg.pagesize)
 	{
 	  exp_data_seg.phase = exp_dataseg_adjust;
-	  result = lang_size_sections_1 (s, output_section_statement, prev,
-					 fill, dot, relax, check_regions);
+	  result = lang_size_sections_1 (s, output_section_statement,
+					 prev, fill, dot, relax,
+					 need_finalize_relax,
+					 check_regions);
 	}
     }
 
@@ -4378,7 +4397,7 @@ lang_process ()
   /* Size up the sections.  */
   lang_size_sections (statement_list.head,
 		      abs_output_section,
-		      &statement_list.head, 0, (bfd_vma) 0, NULL,
+		      &statement_list.head, 0, (bfd_vma) 0, NULL, NULL,
 		      command_line.relax ? FALSE : TRUE);
 
   /* Now run around and relax if we can.  */
@@ -4386,6 +4405,7 @@ lang_process ()
     {
       /* Keep relaxing until bfd_relax_section gives up.  */
       bfd_boolean relax_again;
+      bfd_boolean need_finalize_relax;
 
       do
 	{
@@ -4408,14 +4428,29 @@ lang_process ()
 	  lang_size_sections (statement_list.head,
 			      abs_output_section,
 			      &statement_list.head, 0, (bfd_vma) 0,
-			      &relax_again, FALSE);
+			      &relax_again, &need_finalize_relax,
+			      FALSE);
 
 	  /* If the normal relax is done and the relax finalize pass
 	     is not performed yet, we perform another relax pass.  */
-	  if (!relax_again && !link_info.relax_finalizing)
+	  if (!relax_again
+	      && need_finalize_relax
+	      && !link_info.relax_finalizing)
 	    {
 	      link_info.relax_finalizing = TRUE;
 	      relax_again = TRUE;
+	      if (exp_data_seg.phase == exp_dataseg_adjust)
+		{
+		  /* Make sure addresses are finalized.  */
+		  lang_reset_memory_regions ();
+		  lang_do_assignments (statement_list.head,
+				       abs_output_section,
+				       (fill_type *) 0, (bfd_vma) 0);
+		  lang_size_sections (statement_list.head,
+				      abs_output_section,
+				      &statement_list.head, 0,
+				      (bfd_vma) 0, NULL, NULL, TRUE);
+		}
 	    }
 	}
       while (relax_again);
@@ -4428,7 +4463,7 @@ lang_process ()
       lang_size_sections (statement_list.head,
 			  abs_output_section,
 			  & statement_list.head, 0, (bfd_vma) 0, 
-			  NULL, TRUE);
+			  NULL, NULL, TRUE);
     }
 
   /* See if anything special should be done now we know how big
--- ld/ldlang.h.final	2003-03-04 11:11:00.000000000 -0800
+++ ld/ldlang.h	2003-05-14 15:09:58.000000000 -0700
@@ -482,7 +482,9 @@ extern bfd_vma lang_size_sections
   PARAMS ((lang_statement_union_type *s,
 	   lang_output_section_statement_type *output_section_statement,
 	   lang_statement_union_type **prev, fill_type *fill,
-	   bfd_vma dot, bfd_boolean *relax, bfd_boolean check_regions));
+	   bfd_vma dot, bfd_boolean *relax,
+	   bfd_boolean *need_finalize_relax,
+	   bfd_boolean check_regions));
 extern void lang_enter_group
   PARAMS ((void));
 extern void lang_leave_group
--- ld/pe-dll.c.final	2003-04-02 13:42:02.000000000 -0800
+++ ld/pe-dll.c	2003-05-14 15:10:50.000000000 -0700
@@ -2703,7 +2703,8 @@ pe_dll_fill_sections (abfd, info)
 
       /* Resize the sections.  */
       lang_size_sections (stat_ptr->head, abs_output_section,
-			  &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+			  &stat_ptr->head, 0, (bfd_vma) 0,
+			  NULL, NULL, TRUE);
 
       /* Redo special stuff.  */
       ldemul_after_allocation ();
@@ -2738,7 +2739,8 @@ pe_exe_fill_sections (abfd, info)
 
       /* Resize the sections.  */
       lang_size_sections (stat_ptr->head, abs_output_section,
-			  &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+			  &stat_ptr->head, 0, (bfd_vma) 0,
+			  NULL, NULL, TRUE);
 
       /* Redo special stuff.  */
       ldemul_after_allocation ();

Re: PATCH: Fix the relax finalize pass

[Richard Henderson]

On Wed, May 14, 2003 at 03:53:46PM -0700, H. J. Lu wrote:
> 	* ldlang.c (lang_size_sections_1): Take one more argument to
> 	indicate if the relax finalize pass is needed.
> 	(lang_size_sections): Updated.
> 	(lang_process): Likewise.
> 	* ldlang.h (lang_size_sections_1): Likewise.
> 	* pe-dll.c (pe_dll_fill_sections): Likewise.
> 	(pe_exe_fill_sections): Likewise.
> 	* emultempl/elf32.em (gld${EMULATION_NAME}_finish): Likewise.
> 	* emultempl/hppaelf.em (hppaelf_layout_sections_again): Likewise.
> 	* emultempl/ppc64elf.em (ppc_before_allocation): Likewise.
> 	(ppc_layout_sections_again): Likewise.
> 
> 	* ldlang.c (lang_size_sections): Don't adjust data segment
> 	address after the relax finalize pass starts.
> 	(lang_process): Perform the relax finalize pass only when
> 	needed. Finalize addresses before the relax finalize pass.

Ug.  This is becoming gross.

I think the proper way to handle this is to break up the relax
pass in bfd so that the backend gets more control.

  (1) We need a "beginning of new pass" hook that is called
      after sizes are guessed and addresses are assigned.
      This is where we'd assign a new GP if needed.

      We'd like for this to be able to allocate a data structure
      private to the backend, so that the pieces and passes can
      communicate.

  (2) Existing relax hook operates on each section as usual.
      It records information into the private data structure
      as needed.

  (3) An "end of pass" hook that is called after all sections
      are processed.  Using data collected from sections it
      may decide to re-layout the GOT, or adjust dynamic 
      relocations or whatever.  Currently we do this kind of
      thing for every section, which wastes time.

      If this "end of pass" hook returns "no more iterations",
      then it is also the case that the private data has been
      deallocated.

Operation for ia64 would indeed pass through two phases, one
in which we expand code by adding trampolines, and one in which
we shrink data by eliminating got entries.  But all this would
be controlled by the backend, and all lang_size_sections knows
is that it's been told to continue around the relaxation loop.

As a side benefit, ia64 would get to keep track of the
trampolines that it has added between sections and iterations,
so that they can be re-used when they're in range.


r~

Re: PATCH: Fix the relax finalize pass

[H. J. Lu]

On Wed, May 14, 2003 at 06:01:25PM -0700, Richard Henderson wrote:
> On Wed, May 14, 2003 at 03:53:46PM -0700, H. J. Lu wrote:
> > 	* ldlang.c (lang_size_sections_1): Take one more argument to
> > 	indicate if the relax finalize pass is needed.
> > 	(lang_size_sections): Updated.
> > 	(lang_process): Likewise.
> > 	* ldlang.h (lang_size_sections_1): Likewise.
> > 	* pe-dll.c (pe_dll_fill_sections): Likewise.
> > 	(pe_exe_fill_sections): Likewise.
> > 	* emultempl/elf32.em (gld${EMULATION_NAME}_finish): Likewise.
> > 	* emultempl/hppaelf.em (hppaelf_layout_sections_again): Likewise.
> > 	* emultempl/ppc64elf.em (ppc_before_allocation): Likewise.
> > 	(ppc_layout_sections_again): Likewise.
> > 
> > 	* ldlang.c (lang_size_sections): Don't adjust data segment
> > 	address after the relax finalize pass starts.
> > 	(lang_process): Perform the relax finalize pass only when
> > 	needed. Finalize addresses before the relax finalize pass.
> 
> Ug.  This is becoming gross.
> 
> I think the proper way to handle this is to break up the relax
> pass in bfd so that the backend gets more control.
> 
>   (1) We need a "beginning of new pass" hook that is called
>       after sizes are guessed and addresses are assigned.
>       This is where we'd assign a new GP if needed.
> 

The problem is

. = DATA_SEGMENT_ALIGN (0x10000, 0x4000);

used in elf64_ia64.xs and the corresponding code in lang_size_sections

  exp_data_seg.phase = exp_dataseg_none;
  result = lang_size_sections_1 (s, output_section_statement, prev, fill,
                                 dot, relax, check_regions);
  if (exp_data_seg.phase == exp_dataseg_end_seen)
    {
      /* If DATA_SEGMENT_ALIGN DATA_SEGMENT_END pair was seen, check whether
         a page could be saved in the data segment.  */
      bfd_vma first, last;
 
      first = -exp_data_seg.base & (exp_data_seg.pagesize - 1);
      last = exp_data_seg.end & (exp_data_seg.pagesize - 1);
      if (first && last
          && ((exp_data_seg.base & ~(exp_data_seg.pagesize - 1)) 
              != (exp_data_seg.end & ~(exp_data_seg.pagesize - 1)))
          && first + last <= exp_data_seg.pagesize)
        {
          exp_data_seg.phase = exp_dataseg_adjust;
          result = lang_size_sections_1 (s, output_section_statement, prev,
                                         fill, dot, relax, check_regions);
        }
    }

That means every time when you start a relax pass, ld will try to
squeeze one page out of the data segment. Unless you tell it not to
do it,  when the final pass starts, any section size change in the
final pass may lead to data segment address change due to the call
to lang_do_assignments after the final relax pass.


>       We'd like for this to be able to allocate a data structure
>       private to the backend, so that the pieces and passes can
>       communicate.
> 
>   (2) Existing relax hook operates on each section as usual.
>       It records information into the private data structure
>       as needed.
> 
>   (3) An "end of pass" hook that is called after all sections
>       are processed.  Using data collected from sections it
>       may decide to re-layout the GOT, or adjust dynamic 
>       relocations or whatever.  Currently we do this kind of
>       thing for every section, which wastes time.


The current code has

 if ((sec->flags & SEC_RELOC) == 0
      || sec->reloc_count == 0
      || (link_info->relax_finalizing
          && sec->need_finalize_relax == 0))
    return TRUE;

If need_finalize_relax is 0, it will return immediately.

> 
>       If this "end of pass" hook returns "no more iterations",
>       then it is also the case that the private data has been
>       deallocated.
> 
> Operation for ia64 would indeed pass through two phases, one
> in which we expand code by adding trampolines, and one in which
> we shrink data by eliminating got entries.  But all this would
> be controlled by the backend, and all lang_size_sections knows
> is that it's been told to continue around the relaxation loop.
> 
> As a side benefit, ia64 would get to keep track of the
> trampolines that it has added between sections and iterations,
> so that they can be re-used when they're in range.
> 

The only thing my current patch doesn't do is sharing trampolines.
The current linker doesn't work too well on ia64. My current patch
is an improvement. I can spend time to investigate a better solution.
But I'd like to see a working linker in the meantime.

Another simple kludge is to disable the

  if (exp_data_seg.phase == exp_dataseg_end_seen)
    {
      /* If DATA_SEGMENT_ALIGN DATA_SEGMENT_END pair was seen, check whether
         a page could be saved in the data segment.  */
	...
    }

block. We can put a FIXME there.



H.J.

Re: PATCH: Fix the relax finalize pass

[Richard Henderson]

On Wed, May 14, 2003 at 10:24:51PM -0700, H. J. Lu wrote:
> ... when the final pass starts, any section size change in the
> final pass may lead to data segment address change due to the call
> to lang_do_assignments after the final relax pass.

Seems to me then you shouldn't be calling into the bfd backend
if that's the case.  So after bfd says it's done, you go one
more round to finalize the data section start.

> Another simple kludge is to disable the
> 
>   if (exp_data_seg.phase == exp_dataseg_end_seen)
>     {
>       /* If DATA_SEGMENT_ALIGN DATA_SEGMENT_END pair was seen, check whether
>          a page could be saved in the data segment.  */
> 	...
>     }
> 
> block. We can put a FIXME there.

Maybe that would be better.  It'd mean less kruft to have
to back out of the main linker.


r~

Re: PATCH: Fix the relax finalize pass

[H. J. Lu]

On Wed, May 14, 2003 at 10:37:48PM -0700, Richard Henderson wrote:
> > Another simple kludge is to disable the
> > 
> >   if (exp_data_seg.phase == exp_dataseg_end_seen)
> >     {
> >       /* If DATA_SEGMENT_ALIGN DATA_SEGMENT_END pair was seen, check whether
> >          a page could be saved in the data segment.  */
> > 	...
> >     }
> > 
> > block. We can put a FIXME there.
> 
> Maybe that would be better.  It'd mean less kruft to have
> to back out of the main linker.
> 

I just posted one.


H.J.