binutils-gdb/ld/emultempl/aarch64elf.em
Marcus Shawcroft 4106101c44 [AArch64] Workaround for Cortex A53 erratum 843419
Some early revisions of the Cortex-A53 have an erratum (843419).  The
details of the erratum are quite complex and involve dynamic
conditions.  For the purposes of the workaround we have simplified the
static conditions to an ADRP in the last two instructions of a 4KByte
page, followed within four instructions by a load/store dependent on
the ADRP.

This patch adds support to conservatively scan for and workaround
Cortex A53 erratum 843419.  There are two different workaround
strategies used.  The first is to rewrite ADRP instructions which form
part of an erratum sequence with an ADR instruction.  In situations
where the ADR provides insufficient offset the dependent load or store
instruction from the sequence is moved to a stub section and branches
are inserted from the original sequence to the relocated instruction
and back again.

Stub section sizes are rounded up to a multiple of 4096 in order to
ensure that the act of inserting work around stubs does not create
more errata sequences.

Workaround stubs are always inserted into the stub section associated
with the input section containing the erratum sequence.  This ensures
that the fully relocated form of the veneered load store instruction
is available at the point in time when the stub section is written.
2015-04-01 13:16:38 +01:00

439 lines
13 KiB
Plaintext

# This shell script emits a C file. -*- C -*-
# Copyright (C) 2009-2015 Free Software Foundation, Inc.
# Contributed by ARM Ltd.
#
# 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; see the file COPYING3. If not,
# see <http://www.gnu.org/licenses/>.
#
# This file is sourced from elf32.em, and defines extra aarch64-elf
# specific routines.
#
fragment <<EOF
#include "ldctor.h"
#include "elf/aarch64.h"
static int no_enum_size_warning = 0;
static int no_wchar_size_warning = 0;
static int pic_veneer = 0;
static int fix_erratum_835769 = 0;
static int fix_erratum_843419 = 0;
static void
gld${EMULATION_NAME}_before_parse (void)
{
#ifndef TARGET_ /* I.e., if not generic. */
ldfile_set_output_arch ("`echo ${ARCH}`", bfd_arch_unknown);
#endif /* not TARGET_ */
input_flags.dynamic = ${DYNAMIC_LINK-TRUE};
config.has_shared = `if test -n "$GENERATE_SHLIB_SCRIPT" ; then echo TRUE ; else echo FALSE ; fi`;
config.separate_code = `if test "x${SEPARATE_CODE}" = xyes ; then echo TRUE ; else echo FALSE ; fi`;
}
static void
aarch64_elf_before_allocation (void)
{
/* We should be able to set the size of the interworking stub section. We
can't do it until later if we have dynamic sections, though. */
if (! elf_hash_table (&link_info)->dynamic_sections_created)
{
/* Here we rummage through the found bfds to collect information. */
LANG_FOR_EACH_INPUT_STATEMENT (is)
{
/* Initialise mapping tables for code/data. */
bfd_elf${ELFSIZE}_aarch64_init_maps (is->the_bfd);
}
}
/* Call the standard elf routine. */
gld${EMULATION_NAME}_before_allocation ();
}
/* Fake input file for stubs. */
static lang_input_statement_type *stub_file;
/* Whether we need to call gldarm_layout_sections_again. */
static int need_laying_out = 0;
/* Maximum size of a group of input sections that can be handled by
one stub section. A value of +/-1 indicates the bfd back-end
should use a suitable default size. */
static bfd_signed_vma group_size = 1;
struct hook_stub_info
{
lang_statement_list_type add;
asection *input_section;
};
/* Traverse the linker tree to find the spot where the stub goes. */
static bfd_boolean
hook_in_stub (struct hook_stub_info *info, lang_statement_union_type **lp)
{
lang_statement_union_type *l;
bfd_boolean ret;
for (; (l = *lp) != NULL; lp = &l->header.next)
{
switch (l->header.type)
{
case lang_constructors_statement_enum:
ret = hook_in_stub (info, &constructor_list.head);
if (ret)
return ret;
break;
case lang_output_section_statement_enum:
ret = hook_in_stub (info,
&l->output_section_statement.children.head);
if (ret)
return ret;
break;
case lang_wild_statement_enum:
ret = hook_in_stub (info, &l->wild_statement.children.head);
if (ret)
return ret;
break;
case lang_group_statement_enum:
ret = hook_in_stub (info, &l->group_statement.children.head);
if (ret)
return ret;
break;
case lang_input_section_enum:
if (l->input_section.section == info->input_section)
{
/* We've found our section. Insert the stub immediately
after its associated input section. */
*(info->add.tail) = l->header.next;
l->header.next = info->add.head;
return TRUE;
}
break;
case lang_data_statement_enum:
case lang_reloc_statement_enum:
case lang_object_symbols_statement_enum:
case lang_output_statement_enum:
case lang_target_statement_enum:
case lang_input_statement_enum:
case lang_assignment_statement_enum:
case lang_padding_statement_enum:
case lang_address_statement_enum:
case lang_fill_statement_enum:
break;
default:
FAIL ();
break;
}
}
return FALSE;
}
/* Call-back for elf${ELFSIZE}_aarch64_size_stubs. */
/* Create a new stub section, and arrange for it to be linked
immediately after INPUT_SECTION. */
static asection *
elf${ELFSIZE}_aarch64_add_stub_section (const char *stub_sec_name,
asection *input_section)
{
asection *stub_sec;
flagword flags;
asection *output_section;
lang_output_section_statement_type *os;
struct hook_stub_info info;
flags = (SEC_ALLOC | SEC_LOAD | SEC_READONLY | SEC_CODE
| SEC_HAS_CONTENTS | SEC_RELOC | SEC_IN_MEMORY | SEC_KEEP);
stub_sec = bfd_make_section_anyway_with_flags (stub_file->the_bfd,
stub_sec_name, flags);
if (stub_sec == NULL)
goto err_ret;
bfd_set_section_alignment (stub_file->the_bfd, stub_sec, 2);
output_section = input_section->output_section;
os = lang_output_section_get (output_section);
info.input_section = input_section;
lang_list_init (&info.add);
lang_add_section (&info.add, stub_sec, NULL, os);
if (info.add.head == NULL)
goto err_ret;
if (hook_in_stub (&info, &os->children.head))
return stub_sec;
err_ret:
einfo ("%X%P: can not make stub section: %E\n");
return NULL;
}
/* Another call-back for elf${ELFSIZE}_aarch64_size_stubs. */
static void
gldaarch64_layout_sections_again (void)
{
/* If we have changed sizes of the stub sections, then we need
to recalculate all the section offsets. This may mean we need to
add even more stubs. */
gld${EMULATION_NAME}_map_segments (TRUE);
need_laying_out = -1;
}
static void
build_section_lists (lang_statement_union_type *statement)
{
if (statement->header.type == lang_input_section_enum)
{
asection *i = statement->input_section.section;
if (!((lang_input_statement_type *) i->owner->usrdata)->flags.just_syms
&& (i->flags & SEC_EXCLUDE) == 0
&& i->output_section != NULL
&& i->output_section->owner == link_info.output_bfd)
elf${ELFSIZE}_aarch64_next_input_section (& link_info, i);
}
}
static void
gld${EMULATION_NAME}_after_allocation (void)
{
int ret;
/* bfd_elf32_discard_info just plays with debugging sections,
ie. doesn't affect any code, so we can delay resizing the
sections. It's likely we'll resize everything in the process of
adding stubs. */
ret = bfd_elf_discard_info (link_info.output_bfd, & link_info);
if (ret < 0)
{
einfo ("%X%P: .eh_frame/.stab edit: %E\n");
return;
}
else if (ret > 0)
need_laying_out = 1;
/* If generating a relocatable output file, then we don't
have to examine the relocs. */
if (stub_file != NULL && !link_info.relocatable)
{
ret = elf${ELFSIZE}_aarch64_setup_section_lists (link_info.output_bfd,
&link_info);
if (ret != 0)
{
if (ret < 0)
{
einfo ("%X%P: could not compute sections lists for stub generation: %E\n");
return;
}
lang_for_each_statement (build_section_lists);
/* Call into the BFD backend to do the real work. */
if (! elf${ELFSIZE}_aarch64_size_stubs (link_info.output_bfd,
stub_file->the_bfd,
& link_info,
group_size,
& elf${ELFSIZE}_aarch64_add_stub_section,
& gldaarch64_layout_sections_again))
{
einfo ("%X%P: cannot size stub section: %E\n");
return;
}
}
}
if (need_laying_out != -1)
gld${EMULATION_NAME}_map_segments (need_laying_out);
}
static void
gld${EMULATION_NAME}_finish (void)
{
if (! link_info.relocatable)
{
/* Now build the linker stubs. */
if (stub_file->the_bfd->sections != NULL)
{
if (! elf${ELFSIZE}_aarch64_build_stubs (& link_info))
einfo ("%X%P: can not build stubs: %E\n");
}
}
finish_default ();
}
/* This is a convenient point to tell BFD about target specific flags.
After the output has been created, but before inputs are read. */
static void
aarch64_elf_create_output_section_statements (void)
{
if (strstr (bfd_get_target (link_info.output_bfd), "aarch64") == NULL)
{
/* The arm backend needs special fields in the output hash structure.
These will only be created if the output format is an arm format,
hence we do not support linking and changing output formats at the
same time. Use a link followed by objcopy to change output formats. */
einfo ("%F%X%P: error: Cannot change output format whilst linking AArch64 binaries.\n");
return;
}
bfd_elf${ELFSIZE}_aarch64_set_options (link_info.output_bfd, &link_info,
no_enum_size_warning,
no_wchar_size_warning,
pic_veneer,
fix_erratum_835769, fix_erratum_843419);
stub_file = lang_add_input_file ("linker stubs",
lang_input_file_is_fake_enum,
NULL);
stub_file->the_bfd = bfd_create ("linker stubs", link_info.output_bfd);
if (stub_file->the_bfd == NULL
|| ! bfd_set_arch_mach (stub_file->the_bfd,
bfd_get_arch (link_info.output_bfd),
bfd_get_mach (link_info.output_bfd)))
{
einfo ("%X%P: can not create BFD %E\n");
return;
}
stub_file->the_bfd->flags |= BFD_LINKER_CREATED;
ldlang_add_file (stub_file);
}
/* Avoid processing the fake stub_file in vercheck, stat_needed and
check_needed routines. */
static void (*real_func) (lang_input_statement_type *);
static void aarch64_for_each_input_file_wrapper (lang_input_statement_type *l)
{
if (l != stub_file)
(*real_func) (l);
}
static void
aarch64_lang_for_each_input_file (void (*func) (lang_input_statement_type *))
{
real_func = func;
lang_for_each_input_file (&aarch64_for_each_input_file_wrapper);
}
#define lang_for_each_input_file aarch64_lang_for_each_input_file
EOF
# Define some shell vars to insert bits of code into the standard elf
# parse_args and list_options functions.
#
PARSE_AND_LIST_PROLOGUE='
#define OPTION_NO_ENUM_SIZE_WARNING 309
#define OPTION_PIC_VENEER 310
#define OPTION_STUBGROUP_SIZE 311
#define OPTION_NO_WCHAR_SIZE_WARNING 312
#define OPTION_FIX_ERRATUM_835769 313
#define OPTION_FIX_ERRATUM_843419 314
'
PARSE_AND_LIST_SHORTOPTS=p
PARSE_AND_LIST_LONGOPTS='
{ "no-pipeline-knowledge", no_argument, NULL, '\'p\''},
{ "no-enum-size-warning", no_argument, NULL, OPTION_NO_ENUM_SIZE_WARNING},
{ "pic-veneer", no_argument, NULL, OPTION_PIC_VENEER},
{ "stub-group-size", required_argument, NULL, OPTION_STUBGROUP_SIZE },
{ "no-wchar-size-warning", no_argument, NULL, OPTION_NO_WCHAR_SIZE_WARNING},
{ "fix-cortex-a53-835769", no_argument, NULL, OPTION_FIX_ERRATUM_835769},
{ "fix-cortex-a53-843419", no_argument, NULL, OPTION_FIX_ERRATUM_843419},
'
PARSE_AND_LIST_OPTIONS='
fprintf (file, _(" --no-enum-size-warning Don'\''t warn about objects with incompatible\n"
" enum sizes\n"));
fprintf (file, _(" --no-wchar-size-warning Don'\''t warn about objects with incompatible"
" wchar_t sizes\n"));
fprintf (file, _(" --pic-veneer Always generate PIC interworking veneers\n"));
fprintf (file, _("\
--stub-group-size=N Maximum size of a group of input sections that can be\n\
handled by one stub section. A negative value\n\
locates all stubs after their branches (with a\n\
group size of -N), while a positive value allows\n\
two groups of input sections, one before, and one\n\
after each stub section. Values of +/-1 indicate\n\
the linker should choose suitable defaults.\n"
));
fprintf (file, _(" --fix-cortex-a53-835769 Fix erratum 835769\n"));
fprintf (file, _(" --fix-cortex-a53-843419 Fix erratum 843419\n"));
'
PARSE_AND_LIST_ARGS_CASES='
case '\'p\'':
/* Only here for backwards compatibility. */
break;
case OPTION_NO_ENUM_SIZE_WARNING:
no_enum_size_warning = 1;
break;
case OPTION_NO_WCHAR_SIZE_WARNING:
no_wchar_size_warning = 1;
break;
case OPTION_PIC_VENEER:
pic_veneer = 1;
break;
case OPTION_FIX_ERRATUM_835769:
fix_erratum_835769 = 1;
break;
case OPTION_FIX_ERRATUM_843419:
fix_erratum_843419 = 1;
break;
case OPTION_STUBGROUP_SIZE:
{
const char *end;
group_size = bfd_scan_vma (optarg, &end, 0);
if (*end)
einfo (_("%P%F: invalid number `%s'\''\n"), optarg);
}
break;
'
# We have our own before_allocation etc. functions, but they call
# the standard routines, so give them a different name.
LDEMUL_BEFORE_ALLOCATION=aarch64_elf_before_allocation
LDEMUL_AFTER_ALLOCATION=gld${EMULATION_NAME}_after_allocation
LDEMUL_CREATE_OUTPUT_SECTION_STATEMENTS=aarch64_elf_create_output_section_statements
# Replace the elf before_parse function with our own.
LDEMUL_BEFORE_PARSE=gld"${EMULATION_NAME}"_before_parse
# Call the extra arm-elf function
LDEMUL_FINISH=gld${EMULATION_NAME}_finish