[Bug ld/30422] New: or1k relocation truncated to fit: R_OR1K_GOT16 even when using -mcmodel=large

[gcc at scarsita dot it]

https://sourceware.org/bugzilla/show_bug.cgi?id=30422

            Bug ID: 30422
           Summary: or1k relocation truncated to fit: R_OR1K_GOT16 even
                    when using -mcmodel=large
           Product: binutils
           Version: 2.40
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: gcc at scarsita dot it
  Target Milestone: ---

Created attachment 14863
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14863&action=edit
Patch to pre-scan the presence of R_OR1K_GOT_AHI16 to detect -mcmodel=large
usage

While compiling ghostscript shared library with -mcmodel=large on or1k, there
is still the GOT overflow error:

./soobj/gxdevndi.o: in function `gx_devn_reduce_colored_halftone':
gxdevndi.c:(.text+0x7c): relocation truncated to fit: R_OR1K_GOT16 against
symbol `fc_color_quo' defined in .data.rel.ro.local section in
./soobj/gxdevndi.o
./soobj/gxblend.o: in function `art_blend_pixel_8_inline':
gxblend.c:(.text+0x1954): relocation truncated to fit: R_OR1K_GOT16 against
symbol `art_blend_sq_diff_8' defined in .rodata section in ./soobj/gxblend.o

This is because in some case gcc place R_OR1K_GOT_AHI16 *after* R_OR1K_GOT16,
so ld does not detected the overflow handling:

gxdevndi.o:     file format elf32-or1k

RELOCATION RECORDS FOR [.text]:
OFFSET   TYPE              VALUE
0000004c R_OR1K_GOTPC_HI16  _GLOBAL_OFFSET_TABLE_-0x00000004
00000050 R_OR1K_GOTPC_LO16  _GLOBAL_OFFSET_TABLE_
0000007c R_OR1K_GOT16      fc_color_quo
000000dc R_OR1K_GOT_AHI16  fc_color_quo
0000014c R_OR1K_GOT_AHI16  gx_dc_type_pure
0000015c R_OR1K_GOT16      gx_dc_type_pure
00000260 R_OR1K_GOT_AHI16  gx_dc_type_ht_binary
00000270 R_OR1K_GOT16      gx_dc_type_ht_binary
000002d8 R_OR1K_GOT_AHI16  fc_color_quo
000002e8 R_OR1K_GOT16      fc_color_quo
[...]

gxblend.o:     file format elf32-or1k

RELOCATION RECORDS FOR [.text]:
OFFSET   TYPE              VALUE
00001530 R_OR1K_GOTPC_HI16  _GLOBAL_OFFSET_TABLE_-0x00000004
00001534 R_OR1K_GOTPC_LO16  _GLOBAL_OFFSET_TABLE_
000017ac R_OR1K_GOTOFF_AHI16  .rodata.str1.1
000017b8 R_OR1K_PLT26      memcpy
000018f0 R_OR1K_GOTOFF_LO16  .rodata.str1.1
000018f4 R_OR1K_PLT26      dprintf_file_and_line
000018fc R_OR1K_GOTOFF_AHI16  .rodata.str1.1+0x00000011
00001908 R_OR1K_GOTOFF_LO16  .rodata.str1.1+0x00000011
0000190c R_OR1K_PLT26      errprintf_nomem
00001954 R_OR1K_GOT16      art_blend_sq_diff_8
000019ac R_OR1K_GOT_AHI16  art_blend_soft_light_8
000019b0 R_OR1K_GOT_AHI16  art_blend_sq_diff_8
000019d4 R_OR1K_GOT16      art_blend_soft_light_8
[...]


One solution could be pre-scanning the presence of R_OR1K_GOT_AHI16, like in
the attached patch (tested, and it works).

-- 
You are receiving this mail because:
You are on the CC list for the bug.