Re: Test DWARF2_UNWIND_INFO value for sjlj exceptions

[Rainer Orth <ro at TechFak dot Uni-Bielefeld dot DE>]

Richard Henderson writes:

> In this case you should fall all the way back to the Tru64 solution,
> where we emit .byte stuff into the .data section, and have collect2
> pull together all the fragments.

Ok, I'm following this path.

> Briefly, it looks like TARGET_ASM_EH_FRAME_SECTION should be able to
> switch for you on demand.  Feel free to break up default_eh_frame_section
> into separate fragments so that you can call back into the bit that
> does the data_section emission.

So far, it's coming along nicely, with one exception: I've rediscovering
the problem which prompted me to go for sjlj exceptions initially.
Compiling the O32 libgcc produces assembler errors.  When I compile
_divdi3.o like this:

./xgcc -B./ -B/.vol/gcc/share/mips-sgi-irix6.5/bin/ -isystem /.vol/gcc/share/mips-sgi-irix6.5/include -isystem /.vol/gcc/share/mips-sgi-irix6.5/sys-include -L/.vol/gcc/obj/gcc-3.4-20030916/6.5-cc+o32/gcc/../ld -O2  -DIN_GCC    -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -isystem ./include   -g  -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED  -I. -I. -I/.vol/gcc/src/gcc/gcc -I/.vol/gcc/src/gcc/gcc/. -I/.vol/gcc/src/gcc/gcc/../include   -mabi=32 -DL_divdi3 -c /.vol/gcc/src/gcc/gcc/libgcc2.c -fexceptions -fnon-call-exceptions -o libgcc/mabi=32/_divdi3.o -save-temps -fverbose-asm -dA

I get the following end of _divdi3.s:

$LFE3:
 #APP
	.data
	.align	2
	.globl	_GLOBAL__F___divdi3
_GLOBAL__F___divdi3:
$Lframe1:
	.align 0
	.word	$LECIE1-$LSCIE1	 # Length of Common Information Entry
$LSCIE1:
	.align 0
	.word	0x0	 # CIE Identifier Tag
	.byte	0x1	 # CIE Version
	.ascii "\0"	 # CIE Augmentation
	.byte	0x1	 # uleb128 0x1; CIE Code Alignment Factor
	.byte	0x4	 # sleb128 4; CIE Data Alignment Factor
	.byte	0x40	 # CIE RA Column
	.byte	0xc	 # DW_CFA_def_cfa
	.byte	0x1d	 # uleb128 0x1d
	.byte	0x0	 # uleb128 0x0
	.align	2
$LECIE1:
$LSFDE1:
	.align 0
	.word	$LEFDE1-$LASFDE1	 # FDE Length
$LASFDE1:
	.align 0
	.word	$LASFDE1-$Lframe1	 # FDE CIE offset
	.align 0
	.word	$LFB3	 # FDE initial location
	.align 0
	.word	$LFE3-$LFB3	 # FDE address range
	.byte	0x4	 # DW_CFA_advance_loc4
	.align 0
	.word	$LCFI0-$LFB3
	.byte	0xe	 # DW_CFA_def_cfa_offset
	.byte	0x8	 # uleb128 0x8
	.byte	0x4	 # DW_CFA_advance_loc4
	.align 0
	.word	$LCFI2-$LCFI0
	.byte	0x11	 # DW_CFA_offset_extended_sf
	.byte	0x10	 # uleb128 0x10
	.byte	0x7e	 # sleb128 -2
	.byte	0x11	 # DW_CFA_offset_extended_sf
	.byte	0x11	 # uleb128 0x11
	.byte	0x7f	 # sleb128 -1
	.align	2
$LEFDE1:
	.align 0
	.word	0x0	 # End of Table
	.align	0
 #NO_APP

Unfortunately, as -32 -mips2 complains:

as0: Error: _divdi3.s, line 880:undefined symbol in expression
      .word $LECIE1-$LSCIE1  
as0: Error: _divdi3.s, line 896:undefined symbol in expression
      .word $LEFDE1-$LASFDE1  
as0: Error: _divdi3.s, line 896:Symbol must have absolute value: $LECIE1
      .word $LEFDE1-$LECIE1  
as0: Error: _divdi3.s, line 899:Symbol must have absolute value: $Lframe1
      .word $LASFDE1-$Lframe1  
as0: Error: _divdi3.s, line 903:Symbol must have absolute value: $LFB3
      .word $LFE3-$LFB3  
as0: Error: _divdi3.s, line 906:Symbol must have absolute value: $LFB3
      .word $LCFI0-$LFB3
as0: Error: _divdi3.s, line 911:Symbol must have absolute value: $LCFI0
      .word $LCFI2-$LCFI0

I.e. none of the label differences can be computed.  I'll see if I can come
up with a different solution with some implementation of
ASM_OUTPUT_DWARF_DELTA instead of the generic dw2_asm_output_delta in
dwarf2asm.c.

	Rainer