OpenE2K
/
binutils-gdb
12
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

* elfxx-ia64.c (struct elfNN_ia64_dyn_sym_info): Add want_gotx; 

(elfNN_ia64_check_relocs): Set it.
        (allocate_global_data_got): Check it.
        (allocate_local_got): Likewise.
        (allocate_dynrel_entries): Likewise.
        (elfNN_ia64_relax_ldxmov): New.
        (elfNN_ia64_relax_section): Handle LTOFF22X, LDXMOV.
        (elfNN_ia64_choose_gp): Split out from ...
        (elfNN_ia64_final_link): ... here.
This commit is contained in:
Richard Henderson 2003-02-28 00:22:46 +00:00
parent e03e15dbd3
commit 2c4c2bc0a5
2 changed files with 375 additions and 232 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
bfd/ChangeLog
View File

@ -1,3 +1,15 @@
2003-02-27 Richard Henderson <rth@redhat.com>
* elfxx-ia64.c (struct elfNN_ia64_dyn_sym_info): Add want_gotx;
(elfNN_ia64_check_relocs): Set it.
(allocate_global_data_got): Check it.
(allocate_local_got): Likewise.
(allocate_dynrel_entries): Likewise.
(elfNN_ia64_relax_ldxmov): New.
(elfNN_ia64_relax_section): Handle LTOFF22X, LDXMOV.
(elfNN_ia64_choose_gp): Split out from ...
(elfNN_ia64_final_link): ... here.
2003-02-27 Andrew Cagney <cagney@redhat.com>
* bfd.c (struct bfd): Rename "struct _bfd".

595
bfd/elfxx-ia64.c
View File

@ -102,6 +102,7 @@ struct elfNN_ia64_dyn_sym_info
/* TRUE for the different kinds of linker data we want created. */
unsigned want_got : 1;
unsigned want_gotx : 1;
unsigned want_fptr : 1;
unsigned want_ltoff_fptr : 1;
unsigned want_plt : 1;
@ -154,6 +155,12 @@ struct elfNN_ia64_link_hash_table
struct elfNN_ia64_local_hash_table loc_hash_table;
};
struct elfNN_ia64_allocate_data
{
struct bfd_link_info *info;
bfd_size_type ofs;
};
#define elfNN_ia64_hash_table(p) \
((struct elfNN_ia64_link_hash_table *) ((p)->hash))
@ -169,6 +176,8 @@ static void elfNN_ia64_info_to_howto
static bfd_boolean elfNN_ia64_relax_section
PARAMS((bfd *abfd, asection *sec, struct bfd_link_info *link_info,
bfd_boolean *again));
static void elfNN_ia64_relax_ldxmov
PARAMS((bfd *abfd, bfd_byte *contents, bfd_vma off));
static bfd_boolean is_unwind_section_name
PARAMS ((bfd *abfd, const char *));
static bfd_boolean elfNN_ia64_section_from_shdr
@ -299,6 +308,8 @@ static bfd_vma elfNN_ia64_dtprel_base
PARAMS ((struct bfd_link_info *info));
static int elfNN_ia64_unwind_entry_compare
PARAMS ((const PTR, const PTR));
static bfd_boolean elfNN_ia64_choose_gp
PARAMS ((bfd *abfd, struct bfd_link_info *info));
static bfd_boolean elfNN_ia64_final_link
PARAMS ((bfd *abfd, struct bfd_link_info *info));
static bfd_boolean elfNN_ia64_relocate_section
@ -671,11 +682,7 @@ static const bfd_byte oor_ip[48] =
};
#endif
/* These functions do relaxation for IA-64 ELF.
This is primarily to support branches to targets out of range;
relaxation of R_IA64_LTOFF22X and R_IA64_LDXMOV is handled in
relocate_section directly. */
/* These functions do relaxation for IA-64 ELF. */
static bfd_boolean
elfNN_ia64_relax_section (abfd, sec, link_info, again)
@ -701,6 +708,8 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
struct one_fixup *fixups = NULL;
bfd_boolean changed_contents = FALSE;
bfd_boolean changed_relocs = FALSE;
bfd_boolean changed_got = FALSE;
bfd_vma gp = 0;
/* Assume we're not going to change any sizes, and we'll only need
one pass. */
@ -728,24 +737,6 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
ia64_info = elfNN_ia64_hash_table (link_info);
irelend = internal_relocs + sec->reloc_count;
for (irel = internal_relocs; irel < irelend; irel++)
{
unsigned long r_type = ELFNN_R_TYPE (irel->r_info);
if (r_type == R_IA64_PCREL21B
|| r_type == R_IA64_PCREL21BI
|| r_type == R_IA64_PCREL21M
|| r_type == R_IA64_PCREL21F)
break;
}
/* No branch-type relocations. */
if (irel == irelend)
{
if (elf_section_data (sec)->relocs != internal_relocs)
free (internal_relocs);
return TRUE;
}
/* Get the section contents. */
if (elf_section_data (sec)->this_hdr.contents != NULL)
contents = elf_section_data (sec)->this_hdr.contents;
@ -760,19 +751,33 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
goto error_return;
}
for (; irel < irelend; irel++)
for (irel = internal_relocs; irel < irelend; irel++)
{
unsigned long r_type = ELFNN_R_TYPE (irel->r_info);
bfd_vma symaddr, reladdr, trampoff, toff, roff;
asection *tsec;
struct one_fixup *f;
bfd_size_type amt;
bfd_boolean is_branch;
struct elfNN_ia64_dyn_sym_info *dyn_i;
if (r_type != R_IA64_PCREL21B
&& r_type != R_IA64_PCREL21BI
&& r_type != R_IA64_PCREL21M
&& r_type != R_IA64_PCREL21F)
continue;
switch (r_type)
{
case R_IA64_PCREL21B:
case R_IA64_PCREL21BI:
case R_IA64_PCREL21M:
case R_IA64_PCREL21F:
is_branch = TRUE;
break;
case R_IA64_LTOFF22X:
case R_IA64_LDXMOV:
is_branch = FALSE;
break;
default:
continue;
}
/* Get the value of the symbol referred to by the reloc. */
if (ELFNN_R_SYM (irel->r_info) < symtab_hdr->sh_info)
@ -805,12 +810,12 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
tsec = bfd_section_from_elf_index (abfd, isym->st_shndx);
toff = isym->st_value;
dyn_i = get_dyn_sym_info (ia64_info, NULL, abfd, irel, FALSE);
}
else
{
unsigned long indx;
struct elf_link_hash_entry *h;
struct elfNN_ia64_dyn_sym_info *dyn_i;
indx = ELFNN_R_SYM (irel->r_info) - symtab_hdr->sh_info;
h = elf_sym_hashes (abfd)[indx];
@ -824,7 +829,7 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
/* For branches to dynamic symbols, we're interested instead
in a branch to the PLT entry. */
if (dyn_i && dyn_i->want_plt2)
if (is_branch && dyn_i && dyn_i->want_plt2)
{
/* Internal branches shouldn't be sent to the PLT.
Leave this for now and we'll give an error later. */
@ -834,6 +839,11 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
tsec = ia64_info->plt_sec;
toff = dyn_i->plt2_offset;
}
/* Can't do anything else with dynamic symbols. */
else if (elfNN_ia64_dynamic_symbol_p (h, link_info))
continue;
else
{
/* We can't do anthing with undefined symbols. */
@ -852,103 +862,152 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
+ irel->r_addend);
roff = irel->r_offset;
reladdr = (sec->output_section->vma
+ sec->output_offset
+ roff) & (bfd_vma) -4;
/* If the branch is in range, no need to do anything. */
if ((bfd_signed_vma) (symaddr - reladdr) >= -0x1000000
&& (bfd_signed_vma) (symaddr - reladdr) <= 0x0FFFFF0)
continue;
/* If the branch and target are in the same section, you've
got one honking big section and we can't help you. You'll
get an error message later. */
if (tsec == sec)
continue;
/* Look for an existing fixup to this address. */
for (f = fixups; f ; f = f->next)
if (f->tsec == tsec && f->toff == toff)
break;
if (f == NULL)
if (is_branch)
{
/* Two alternatives: If it's a branch to a PLT entry, we can
make a copy of the FULL_PLT entry. Otherwise, we'll have
to use a `brl' insn to get where we're going. */
reladdr = (sec->output_section->vma
+ sec->output_offset
+ roff) & (bfd_vma) -4;
size_t size;
/* If the branch is in range, no need to do anything. */
if ((bfd_signed_vma) (symaddr - reladdr) >= -0x1000000
&& (bfd_signed_vma) (symaddr - reladdr) <= 0x0FFFFF0)
continue;
if (tsec == ia64_info->plt_sec)
size = sizeof (plt_full_entry);
/* If the branch and target are in the same section, you've
got one honking big section and we can't help you. You'll
get an error message later. */
if (tsec == sec)
continue;
/* Look for an existing fixup to this address. */
for (f = fixups; f ; f = f->next)
if (f->tsec == tsec && f->toff == toff)
break;
if (f == NULL)
{
/* Two alternatives: If it's a branch to a PLT entry, we can
make a copy of the FULL_PLT entry. Otherwise, we'll have
to use a `brl' insn to get where we're going. */
size_t size;
if (tsec == ia64_info->plt_sec)
size = sizeof (plt_full_entry);
else
{
#ifdef USE_BRL
size = sizeof (oor_brl);
#else
size = sizeof (oor_ip);
#endif
}
/* Resize the current section to make room for the new branch. */
trampoff = (sec->_cooked_size + 15) & (bfd_vma) -16;
amt = trampoff + size;
contents = (bfd_byte *) bfd_realloc (contents, amt);
if (contents == NULL)
goto error_return;
sec->_cooked_size = amt;
if (tsec == ia64_info->plt_sec)
{
memcpy (contents + trampoff, plt_full_entry, size);
/* Hijack the old relocation for use as the PLTOFF reloc. */
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_PLTOFF22);
irel->r_offset = trampoff;
}
else
{
#ifdef USE_BRL
memcpy (contents + trampoff, oor_brl, size);
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_PCREL60B);
irel->r_offset = trampoff + 2;
#else
memcpy (contents + trampoff, oor_ip, size);
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_PCREL64I);
irel->r_addend -= 16;
irel->r_offset = trampoff + 2;
#endif
}
/* Record the fixup so we don't do it again this section. */
f = (struct one_fixup *)
bfd_malloc ((bfd_size_type) sizeof (*f));
f->next = fixups;
f->tsec = tsec;
f->toff = toff;
f->trampoff = trampoff;
fixups = f;
}
else
{
#ifdef USE_BRL
size = sizeof (oor_brl);
#else
size = sizeof (oor_ip);
#endif
/* Nop out the reloc, since we're finalizing things here. */
irel->r_info = ELFNN_R_INFO (0, R_IA64_NONE);
}
/* Resize the current section to make room for the new branch. */
trampoff = (sec->_cooked_size + 15) & (bfd_vma) -16;
amt = trampoff + size;
contents = (bfd_byte *) bfd_realloc (contents, amt);
if (contents == NULL)
/* Fix up the existing branch to hit the trampoline. Hope like
hell this doesn't overflow too. */
if (elfNN_ia64_install_value (abfd, contents + roff,
f->trampoff - (roff & (bfd_vma) -4),
r_type) != bfd_reloc_ok)
goto error_return;
sec->_cooked_size = amt;
if (tsec == ia64_info->plt_sec)
{
memcpy (contents + trampoff, plt_full_entry, size);
/* Hijack the old relocation for use as the PLTOFF reloc. */
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_PLTOFF22);
irel->r_offset = trampoff;
}
else
{
#ifdef USE_BRL
memcpy (contents + trampoff, oor_brl, size);
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_PCREL60B);
irel->r_offset = trampoff + 2;
#else
memcpy (contents + trampoff, oor_ip, size);
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_PCREL64I);
irel->r_addend -= 16;
irel->r_offset = trampoff + 2;
#endif
}
/* Record the fixup so we don't do it again this section. */
f = (struct one_fixup *) bfd_malloc ((bfd_size_type) sizeof (*f));
f->next = fixups;
f->tsec = tsec;
f->toff = toff;
f->trampoff = trampoff;
fixups = f;
changed_contents = TRUE;
changed_relocs = TRUE;
}
else
{
/* Nop out the reloc, since we're finalizing things here. */
irel->r_info = ELFNN_R_INFO (0, R_IA64_NONE);
/* Fetch the gp. */
if (gp == 0)
{
bfd *obfd = sec->output_section->owner;
gp = _bfd_get_gp_value (obfd);
if (gp == 0)
{
if (!elfNN_ia64_choose_gp (obfd, link_info))
goto error_return;
gp = _bfd_get_gp_value (obfd);
}
}
/* If the data is out of range, do nothing. */
if ((bfd_signed_vma) (symaddr - gp) >= 0x400000
||(bfd_signed_vma) (symaddr - gp) < -0x400000)
continue;
if (r_type == R_IA64_LTOFF22X)
{
irel->r_info = ELFNN_R_INFO (ELFNN_R_SYM (irel->r_info),
R_IA64_GPREL22);
changed_relocs = TRUE;
if (dyn_i->want_gotx)
{
dyn_i->want_gotx = 0;
changed_got |= !dyn_i->want_got;
}
}
else
{
elfNN_ia64_relax_ldxmov (abfd, contents, roff);
irel->r_info = ELFNN_R_INFO (0, R_IA64_NONE);
changed_contents = TRUE;
changed_relocs = TRUE;
}
}
/* Fix up the existing branch to hit the trampoline. Hope like
hell this doesn't overflow too. */
if (elfNN_ia64_install_value (abfd, contents + roff,
f->trampoff - (roff & (bfd_vma) -4),
r_type) != bfd_reloc_ok)
goto error_return;
changed_contents = TRUE;
changed_relocs = TRUE;
}
/* ??? If we created fixups, this may push the code segment large
enough that the data segment moves, which will change the GP.
Reset the GP so that we re-calculate next round. We need to
do this at the _beginning_ of the next round; now will not do. */
/* Clean up and go home. */
while (fixups)
{
@ -989,6 +1048,21 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
elf_section_data (sec)->relocs = internal_relocs;
}
if (changed_got)
{
struct elfNN_ia64_allocate_data data;
data.info = link_info;
data.ofs = 0;
elfNN_ia64_dyn_sym_traverse (ia64_info, allocate_global_data_got, &data);
elfNN_ia64_dyn_sym_traverse (ia64_info, allocate_global_fptr_got, &data);
elfNN_ia64_dyn_sym_traverse (ia64_info, allocate_local_got, &data);
ia64_info->got_sec->_raw_size = data.ofs;
ia64_info->got_sec->_cooked_size = data.ofs;
/* ??? Resize .rela.got too. */
}
*again = changed_contents || changed_relocs;
return TRUE;
@ -1003,6 +1077,39 @@ elfNN_ia64_relax_section (abfd, sec, link_info, again)
free (internal_relocs);
return FALSE;
}
static void
elfNN_ia64_relax_ldxmov (abfd, contents, off)
bfd *abfd;
bfd_byte *contents;
bfd_vma off;
{
int shift, r1, r3;
bfd_vma dword, insn;
switch ((int)off & 0x3)
{
case 0: shift = 5; break;
case 1: shift = 14; off += 3; break;
case 2: shift = 23; off += 6; break;
case 3:
abort ();
}
dword = bfd_get_64 (abfd, contents + off);
insn = (dword >> shift) & 0x1ffffffffffLL;
r1 = (insn >> 6) & 127;
r3 = (insn >> 20) & 127;
if (r1 == r3)
insn = 0x8000000; /* nop */
else
insn = (insn & 0x7f01fff) | 0x10800000000LL; /* (qp) mov r1 = r3 */
dword &= ~(0x1ffffffffffLL << shift);
dword |= (insn << shift);
bfd_put_64 (abfd, dword, contents + off);
}
/* Return TRUE if NAME is an unwind table section name. */
@ -2125,15 +2232,16 @@ elfNN_ia64_check_relocs (abfd, info, sec, relocs)
{
enum {
NEED_GOT = 1,
NEED_FPTR = 2,
NEED_PLTOFF = 4,
NEED_MIN_PLT = 8,
NEED_FULL_PLT = 16,
NEED_DYNREL = 32,
NEED_LTOFF_FPTR = 64,
NEED_TPREL = 128,
NEED_DTPMOD = 256,
NEED_DTPREL = 512
NEED_GOTX = 2,
NEED_FPTR = 4,
NEED_PLTOFF = 8,
NEED_MIN_PLT = 16,
NEED_FULL_PLT = 32,
NEED_DYNREL = 64,
NEED_LTOFF_FPTR = 128,
NEED_TPREL = 256,
NEED_DTPMOD = 512,
NEED_DTPREL = 1024
};
struct elf_link_hash_entry *h = NULL;
@ -2230,11 +2338,14 @@ elfNN_ia64_check_relocs (abfd, info, sec, relocs)
break;
case R_IA64_LTOFF22:
case R_IA64_LTOFF22X:
case R_IA64_LTOFF64I:
need_entry = NEED_GOT;
break;
case R_IA64_LTOFF22X:
need_entry = NEED_GOTX;
break;
case R_IA64_PLTOFF22:
case R_IA64_PLTOFF64I:
case R_IA64_PLTOFF64MSB:
@ -2316,7 +2427,8 @@ elfNN_ia64_check_relocs (abfd, info, sec, relocs)
dyn_i->h = h;
/* Create what's needed. */
if (need_entry & (NEED_GOT | NEED_TPREL | NEED_DTPMOD | NEED_DTPREL))
if (need_entry & (NEED_GOT | NEED_GOTX | NEED_TPREL
| NEED_DTPMOD | NEED_DTPREL))
{
if (!got)
{
@ -2326,6 +2438,8 @@ elfNN_ia64_check_relocs (abfd, info, sec, relocs)
}
if (need_entry & NEED_GOT)
dyn_i->want_got = 1;
if (need_entry & NEED_GOTX)
dyn_i->want_gotx = 1;
if (need_entry & NEED_TPREL)
dyn_i->want_tprel = 1;
if (need_entry & NEED_DTPMOD)
@ -2385,12 +2499,6 @@ elfNN_ia64_check_relocs (abfd, info, sec, relocs)
return TRUE;
}
struct elfNN_ia64_allocate_data
{
struct bfd_link_info *info;
bfd_size_type ofs;
};
/* For cleanliness, and potentially faster dynamic loading, allocate
external GOT entries first. */
@ -2401,7 +2509,7 @@ allocate_global_data_got (dyn_i, data)
{
struct elfNN_ia64_allocate_data *x = (struct elfNN_ia64_allocate_data *)data;
if (dyn_i->want_got
if ((dyn_i->want_got || dyn_i->want_gotx)
&& ! dyn_i->want_fptr
&& (elfNN_ia64_dynamic_symbol_p (dyn_i->h, x->info)
|| (elfNN_ia64_aix_vec (x->info->hash->creator)
@ -2473,7 +2581,7 @@ allocate_local_got (dyn_i, data)
{
struct elfNN_ia64_allocate_data *x = (struct elfNN_ia64_allocate_data *)data;
if (dyn_i->want_got
if ((dyn_i->want_got || dyn_i->want_gotx)
&& ! (elfNN_ia64_dynamic_symbol_p (dyn_i->h, x->info)
|| elfNN_ia64_aix_vec (x->info->hash->creator)))
{
@ -2699,7 +2807,7 @@ allocate_dynrel_entries (dyn_i, data)
/* Take care of the GOT and PLT relocations. */
if (((dynamic_symbol || shared) && dyn_i->want_got)
if (((dynamic_symbol || shared) && (dyn_i->want_got || dyn_i->want_gotx))
|| (dyn_i->want_ltoff_fptr && dyn_i->h && dyn_i->h->dynindx != -1))
ia64_info->rel_got_sec->_raw_size += sizeof (ElfNN_External_Rela);
if ((dynamic_symbol || shared) && dyn_i->want_tprel)
@ -3516,6 +3624,123 @@ elfNN_ia64_unwind_entry_compare (a, b)
return (av < bv ? -1 : av > bv ? 1 : 0);
}
/* Make sure we've got ourselves a nice fat __gp value. */
static bfd_boolean
elfNN_ia64_choose_gp (abfd, info)
bfd *abfd;
struct bfd_link_info *info;
{
bfd_vma min_vma = (bfd_vma) -1, max_vma = 0;
bfd_vma min_short_vma = min_vma, max_short_vma = 0;
struct elf_link_hash_entry *gp;
bfd_vma gp_val;
asection *os;
struct elfNN_ia64_link_hash_table *ia64_info;
ia64_info = elfNN_ia64_hash_table (info);
/* Find the min and max vma of all sections marked short. Also collect
min and max vma of any type, for use in selecting a nice gp. */
for (os = abfd->sections; os ; os = os->next)
{
bfd_vma lo, hi;
if ((os->flags & SEC_ALLOC) == 0)
continue;
lo = os->vma;
hi = os->vma + os->_raw_size;
if (hi < lo)
hi = (bfd_vma) -1;
if (min_vma > lo)
min_vma = lo;
if (max_vma < hi)
max_vma = hi;
if (os->flags & SEC_SMALL_DATA)
{
if (min_short_vma > lo)
min_short_vma = lo;
if (max_short_vma < hi)
max_short_vma = hi;
}
}
/* See if the user wants to force a value. */
gp = elf_link_hash_lookup (elf_hash_table (info), "__gp", FALSE,
FALSE, FALSE);
if (gp
&& (gp->root.type == bfd_link_hash_defined
|| gp->root.type == bfd_link_hash_defweak))
{
asection *gp_sec = gp->root.u.def.section;
gp_val = (gp->root.u.def.value
+ gp_sec->output_section->vma
+ gp_sec->output_offset);
}
else
{
/* Pick a sensible value. */
asection *got_sec = ia64_info->got_sec;
/* Start with just the address of the .got. */
if (got_sec)
gp_val = got_sec->output_section->vma;
else if (max_short_vma != 0)
gp_val = min_short_vma;
else
gp_val = min_vma;
/* If it is possible to address the entire image, but we
don't with the choice above, adjust. */
if (max_vma - min_vma < 0x400000
&& max_vma - gp_val <= 0x200000
&& gp_val - min_vma > 0x200000)
gp_val = min_vma + 0x200000;
else if (max_short_vma != 0)
{
/* If we don't cover all the short data, adjust. */
if (max_short_vma - gp_val >= 0x200000)
gp_val = min_short_vma + 0x200000;
/* If we're addressing stuff past the end, adjust back. */
if (gp_val > max_vma)
gp_val = max_vma - 0x200000 + 8;
}
}
/* Validate whether all SHF_IA_64_SHORT sections are within
range of the chosen GP. */
if (max_short_vma != 0)
{
if (max_short_vma - min_short_vma >= 0x400000)
{
(*_bfd_error_handler)
(_("%s: short data segment overflowed (0x%lx >= 0x400000)"),
bfd_get_filename (abfd),
(unsigned long) (max_short_vma - min_short_vma));
return FALSE;
}
else if ((gp_val > min_short_vma
&& gp_val - min_short_vma > 0x200000)
|| (gp_val < max_short_vma
&& max_short_vma - gp_val >= 0x200000))
{
(*_bfd_error_handler)
(_("%s: __gp does not cover short data segment"),
bfd_get_filename (abfd));
return FALSE;
}
}
_bfd_set_gp_value (abfd, gp_val);
return TRUE;
}
static bfd_boolean
elfNN_ia64_final_link (abfd, info)
bfd *abfd;
@ -3529,112 +3754,18 @@ elfNN_ia64_final_link (abfd, info)
/* Make sure we've got ourselves a nice fat __gp value. */
if (!info->relocateable)
{
bfd_vma min_vma = (bfd_vma) -1, max_vma = 0;
bfd_vma min_short_vma = min_vma, max_short_vma = 0;
bfd_vma gp_val = _bfd_get_gp_value (abfd);
struct elf_link_hash_entry *gp;
bfd_vma gp_val;
asection *os;
/* Find the min and max vma of all sections marked short. Also
collect min and max vma of any type, for use in selecting a
nice gp. */
for (os = abfd->sections; os ; os = os->next)
if (gp_val == 0)
{
bfd_vma lo, hi;
if ((os->flags & SEC_ALLOC) == 0)
continue;
lo = os->vma;
hi = os->vma + os->_raw_size;
if (hi < lo)
hi = (bfd_vma) -1;
if (min_vma > lo)
min_vma = lo;
if (max_vma < hi)
max_vma = hi;
if (os->flags & SEC_SMALL_DATA)
{
if (min_short_vma > lo)
min_short_vma = lo;
if (max_short_vma < hi)
max_short_vma = hi;
}
if (! elfNN_ia64_choose_gp (abfd, info))
return FALSE;
gp_val = _bfd_get_gp_value (abfd);
}
/* See if the user wants to force a value. */
gp = elf_link_hash_lookup (elf_hash_table (info), "__gp", FALSE,
FALSE, FALSE);
if (gp
&& (gp->root.type == bfd_link_hash_defined
|| gp->root.type == bfd_link_hash_defweak))
{
asection *gp_sec = gp->root.u.def.section;
gp_val = (gp->root.u.def.value
+ gp_sec->output_section->vma
+ gp_sec->output_offset);
}
else
{
/* Pick a sensible value. */
asection *got_sec = ia64_info->got_sec;
/* Start with just the address of the .got. */
if (got_sec)
gp_val = got_sec->output_section->vma;
else if (max_short_vma != 0)
gp_val = min_short_vma;
else
gp_val = min_vma;
/* If it is possible to address the entire image, but we
don't with the choice above, adjust. */
if (max_vma - min_vma < 0x400000
&& max_vma - gp_val <= 0x200000
&& gp_val - min_vma > 0x200000)
gp_val = min_vma + 0x200000;
else if (max_short_vma != 0)
{
/* If we don't cover all the short data, adjust. */
if (max_short_vma - gp_val >= 0x200000)
gp_val = min_short_vma + 0x200000;
/* If we're addressing stuff past the end, adjust back. */
if (gp_val > max_vma)
gp_val = max_vma - 0x200000 + 8;
}
}
/* Validate whether all SHF_IA_64_SHORT sections are within
range of the chosen GP. */
if (max_short_vma != 0)
{
if (max_short_vma - min_short_vma >= 0x400000)
{
(*_bfd_error_handler)
(_("%s: short data segment overflowed (0x%lx >= 0x400000)"),
bfd_get_filename (abfd),
(unsigned long) (max_short_vma - min_short_vma));
return FALSE;
}
else if ((gp_val > min_short_vma
&& gp_val - min_short_vma > 0x200000)
|| (gp_val < max_short_vma
&& max_short_vma - gp_val >= 0x200000))
{
(*_bfd_error_handler)
(_("%s: __gp does not cover short data segment"),
bfd_get_filename (abfd));
return FALSE;
}
}
_bfd_set_gp_value (abfd, gp_val);
FALSE, FALSE);
if (gp)
{
gp->root.type = bfd_link_hash_defined;