https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99783
--- Comment #9 from shorne at gmail dot com ---
I was able to debug this.
Basically, the code to avoid overflow by masking the relocation value with
0xffff fails if the relocation value has the 16-bit sign bit set. I.e. 0x90e4
has 0x8000 set when we see the overflow message.
It seems the number of GOT relocations in libgeos as gone beyond this limit and
triggered this bug.
This is an oversite and needs a further patch to OpenRISC binutils. I will
think about how to patch it and post a patch in a few days
DEBUG:
x 1184 status = bfd_check_overflow (howto->complain_on_overflow,
x 1185 howto->bitsize,
x 1186 howto->rightshift,
x 1187 bfd_arch_bits_per_address
(input_bfd),
x 1188 value);
x 1189 value >>= howto->rightshift;
x 1190
x 1191 /* If we're overwriting the entire destination,
x 1192 then no need to read the current contents. */
x >1193 if (size == 0 || howto->dst_mask == N_ONES (size))
x 1194 x = 0;
x 1195 else
x 1196 {
x 1197 BFD_ASSERT (size == 4);
x 1198 x = bfd_get_32 (input_bfd, contents + offset);
x 1199 }
x 1200
x 1201 switch (howto->type)
x 1202 {
x 1203 case R_OR1K_SLO16:
x 1204 case R_OR1K_GOTOFF_SLO16:
x 1205 case R_OR1K_TLS_LE_SLO16:
native process 3648008 In: or1k_final_link_relocate L1193 PC:
0x43afb3
(gdb) n
or1k_final_link_relocate (howto=howto@entry=0x50fff0
<or1k_elf_howto_table+560>, input_bfd=input_bfd@entry=0x57ac90,
input_section=input_section@entry=0x593fb8,
contents=contents@entry=0xb3e17b0 "\234!\377\374\030`", offset=68,
value=37092) at elf32-or1k.c:1189
(gdb) p status
$3 = bfd_reloc_overflow
(gdb) p/x value
$4 = 0x90e4
On Fri, Dec 31, 2021 at 11:26:06AM +0000, giulio.benetti at benettiengineering
dot com wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99783
