binutils-gdb/ld/emultempl/mmo.em
Alan Modra 8ce18f9cdf Avoid ubsan bug complaining about &p->field
I reckon it's quite OK to write &p->field in C when p might be NULL,
and lots of old C programmers probably agree with me.  However, ubsan
disagrees and so do some people I respect.  I suspect C++ influence is
to blame for the ubsan behaviour.  See
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92634.  So far no one has
educated me as to why I'm wrong to claim that there isn't anything in
the C standard to say that p->field is always (*p).field.  Note 79
doesn't quite do that because it doesn't cover null pointers.  If
there was such an equivalence then you could claim &p->field has a
null pointer reference when p is NULL, even though no C compiler would
ever dereference p.

Anyway, to silence ubsan I'm going to apply the following though I
prefer to avoid casts when possible.  And I'm using (void *)
deliberately because this is C, not C++!

	* ldlang.c (lang_output_section_find_by_flags): Don't use &p->field
	when p might be NULL.
	* ldelf.c (output_rel_find, ldelf_place_orphan): Likewise.
	(insert_os_after, lang_insert_orphan, lookup_name): Likewise.
	(strip_excluded_output_sections, lang_clear_os_map): Likewise.
	(lang_check, lang_for_each_input_file): Likewise.
	(lang_reset_memory_regions, find_replacements_insert_point): Likewise.
	(find_rescan_insertion, lang_propagate_lma_regions): Likewise.
	(lang_record_phdrs): Likewise.
	* emultempl/alphaelf.em (alpha_after_open): Likewise.
	* emultempl/mmo.em (mmo_place_orphan): Likewise.
	* emultempl/pe.em (gld_${EMULATION_NAME}_place_orphan): Likewise.
	* emultempl/pep.em (gld_${EMULATION_NAME}_place_orphan): Likewise.
	* emultempl/ppc32elf.em (ppc_after_check_relocs): Likewise.
	* emultempl/spuelf.em (spu_before_allocation): Likewise.
	(embedded_spu_file): Likewise.
2019-12-26 17:49:03 +10:30

238 lines
7.1 KiB
Plaintext
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# This shell script emits a C file. -*- C -*-
# Copyright (C) 2001-2019 Free Software Foundation, Inc.
#
# This file is part of the GNU Binutils.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
# MA 02110-1301, USA.
#
# This file is sourced from generic.em.
fragment <<EOF
/* Need to have this macro defined before mmix-elfnmmo, which uses the
name for the before_allocation function, defined in ldemul.c (for
the mmo "emulation") or in elf.em (for the elf64mmix
"emulation"). */
#define gldmmo_before_allocation before_allocation_default
/* We include this header *not* because we expect to handle ELF here
but because we use the map_segments function. But this is only to
get a weird testcase right; ld-mmix/bpo-22, forcing ELF to be
output from the mmo emulation: -m mmo --oformat elf64-mmix! */
#include "ldelfgen.h"
static void gld${EMULATION_NAME}_after_allocation (void);
EOF
source_em ${srcdir}/emultempl/elf-generic.em
source_em ${srcdir}/emultempl/mmix-elfnmmo.em
fragment <<EOF
/* Place an orphan section. We use this to put random SEC_CODE or
SEC_READONLY sections right after MMO_TEXT_SECTION_NAME. Much borrowed
from elf.em. */
static lang_output_section_statement_type *
mmo_place_orphan (asection *s,
const char *secname,
int constraint ATTRIBUTE_UNUSED)
{
static struct
{
flagword nonzero_flags;
struct orphan_save orphansave;
} holds[] =
{
{
SEC_CODE | SEC_READONLY,
{
MMO_TEXT_SECTION_NAME,
SEC_HAS_CONTENTS | SEC_ALLOC | SEC_LOAD | SEC_READONLY | SEC_CODE,
0, 0, 0, 0
}
},
{
SEC_LOAD | SEC_DATA,
{
MMO_DATA_SECTION_NAME,
SEC_HAS_CONTENTS | SEC_ALLOC | SEC_LOAD | SEC_DATA,
0, 0, 0, 0
}
},
{
SEC_ALLOC,
{
".bss",
SEC_ALLOC,
0, 0, 0, 0
}
}
};
struct orphan_save *place = NULL;
lang_output_section_statement_type *after;
lang_output_section_statement_type *os;
size_t i;
flagword flags;
asection *nexts;
/* We have nothing to say for anything other than a final link or
for sections that are excluded. */
if (bfd_link_relocatable (&link_info)
|| (s->flags & SEC_EXCLUDE) != 0)
return NULL;
os = lang_output_section_find (secname);
/* We have an output section by this name. Place the section inside it
(regardless of whether the linker script lists it as input). */
if (os != NULL)
{
lang_add_section (&os->children, s, NULL, os);
return os;
}
flags = s->flags;
if (!bfd_link_relocatable (&link_info))
{
nexts = s;
while ((nexts = bfd_get_next_section_by_name (nexts->owner, nexts))
!= NULL)
if (nexts->output_section == NULL
&& (nexts->flags & SEC_EXCLUDE) == 0
&& ((nexts->flags ^ flags) & (SEC_LOAD | SEC_ALLOC)) == 0
&& (nexts->owner->flags & DYNAMIC) == 0
&& !bfd_input_just_syms (nexts->owner))
flags = (((flags ^ SEC_READONLY) | (nexts->flags ^ SEC_READONLY))
^ SEC_READONLY);
}
/* Check for matching section type flags for sections we care about.
A section without contents can have SEC_LOAD == 0, but we still
want it attached to a sane section so the symbols appear as
expected. */
if ((flags & (SEC_ALLOC | SEC_READONLY)) != SEC_READONLY)
for (i = 0; i < sizeof (holds) / sizeof (holds[0]); i++)
if ((flags & holds[i].nonzero_flags) != 0)
{
place = &holds[i].orphansave;
if (place->os == NULL)
place->os = lang_output_section_find (place->name);
break;
}
if (place == NULL)
{
/* For other combinations, we have to give up, except we make
sure not to place the orphan section after the
linker-generated register section; that'd make it continue
the reg section and we never want that to happen for orphan
sections. */
lang_output_section_statement_type *before;
lang_output_section_statement_type *lookup;
static struct orphan_save hold_nonreg =
{
NULL,
SEC_READONLY,
0, 0, 0, 0
};
if (hold_nonreg.os == NULL)
{
before = lang_output_section_find (MMIX_REG_CONTENTS_SECTION_NAME);
/* If we have no such section, all fine; we don't care where
it's placed. */
if (before == NULL)
return NULL;
/* We have to find the oss before this one, so we can use that as
"after". */
for (lookup = (void *) lang_os_list.head;
lookup != NULL && lookup->next != before;
lookup = lookup->next)
;
hold_nonreg.os = lookup;
}
place = &hold_nonreg;
}
after = place->os;
if (after == NULL)
return NULL;
/* If there's an output section by *this* name, we'll use it, regardless
of actual section flags, in contrast to what's done in elf.em. */
os = lang_insert_orphan (s, secname, 0, after, place, NULL, NULL);
return os;
}
/* Remove the spurious settings of SEC_RELOC that make it to the output at
link time. We are as confused as elflink.h:elf_bfd_final_link, and
paper over the bug similarly. */
static void
mmo_wipe_sec_reloc_flag (bfd *abfd ATTRIBUTE_UNUSED, asection *sec,
void *ptr ATTRIBUTE_UNUSED)
{
bfd_set_section_flags (sec, bfd_section_flags (sec) & ~SEC_RELOC);
}
/* Iterate with bfd_map_over_sections over mmo_wipe_sec_reloc_flag... */
static void
gld${EMULATION_NAME}_after_allocation (void)
{
bfd_map_over_sections (link_info.output_bfd, mmo_wipe_sec_reloc_flag, NULL);
ldelf_map_segments (FALSE);
}
/* To get on-demand global register allocation right, we need to parse the
relocs, like what happens when linking to ELF. It needs to be done
before all input sections are supposed to be present. When linking to
ELF, it's done when reading symbols. When linking to mmo, we do it
when all input files are seen, which is equivalent. */
static void
mmo_after_open (void)
{
/* When there's a mismatch between the output format and the emulation
(using weird combinations like "-m mmo --oformat elf64-mmix" for
example), we'd count relocs twice because they'd also be counted
along the usual route for ELF-only linking, which would lead to an
internal accounting error. */
if (bfd_get_flavour (link_info.output_bfd) != bfd_target_elf_flavour)
{
LANG_FOR_EACH_INPUT_STATEMENT (is)
{
if (bfd_get_flavour (is->the_bfd) == bfd_target_elf_flavour
&& !_bfd_mmix_check_all_relocs (is->the_bfd, &link_info))
einfo (_("%X%P: internal problems scanning %pB after opening it"),
is->the_bfd);
}
}
after_open_default ();
}
EOF
LDEMUL_PLACE_ORPHAN=mmo_place_orphan
LDEMUL_AFTER_OPEN=mmo_after_open