When the symbol referenced by a GOT reloc is an ifunc, we can't
optimise away the GOT indirection. Well, we can, but only if a global
entry stub is created with the ifunc symbol redefined to the stub.
But that results in slower code and an indirection via the PLT so
there isn't much to like about that solution.
* elf64-ppc.c (ppc64_elf_edit_toc): Exclude ifunc from GOT
optimisation.
(ppc64_elf_relocate_section): Likewise.
PR 25823
* peXXigen.c (_bfd_XXi_swap_sym_in <C_SECTION>): Don't use a
pointer into strings that may be freed for section name, always
allocate a new string.
This patch extends the relocation and line number counters for
coff-go32 and coff-go32-exe to 32 bits. As I understand it works the
same as for PE-COFF:
If the number of relocations in an object file exceeds 65534, the
NRELOC field is set to 65535 and the actual number of relocations is
stored in the VADDR field of the first relocation entry.
Executable files have no relocations, and thus the NRELOC field is
repurposed to extend NLNNO to 32-bits.
bfd * coff-go32.c (COFF_GO32, IMAGE_SCN_LNK_NRELOC_OVFL)
(coff_SWAP_scnhdr_in, coff_SWAP_scnhdr_out): Define.
(_bfd_go32_swap_scnhdr_in, _bfd_go32_swap_scnhdr_out)
(_bfd_go32_mkobject): New functions.
* coff-stgo32.c (IMAGE_SCN_LNK_NRELOC_OVFL)
(coff_SWAP_scnhdr_in, coff_SWAP_scnhdr_out): Define.
(go32exe_mkobject): Call _bfd_go32_mkobject.
* coffcode.h (COFF_WITH_EXTENDED_RELOC_COUNTER): Define.
(coff_set_alignment_hook): Define function for COFF_GO32_EXE
and COFF_GO32.
(coff_write_relocs): Enable extended reloc counter code if
COFF_WITH_EXTENDED_RELOC_COUNTER is defined. Test for obj_go32.
(coff_write_object_contents): Likewise. Pad section headers
for COFF_GO32 and COFF_GO32EXE. Use bfd_coff_swap_scnhdr_out
instead of coff_swap_scnhdr_out.
* cofflink.c (_bfd_coff_final_link): Test also for obj_go32 to
enable extended reloc counter.
* coffswap.h: (coff_swap_scnhdr_in, coff_swap_scnhdr_out):
Declare with ATTRIBUTE_UNUSED.
* libcoff-in.h: (struct coff_tdata): New field go32.
(obj_go32): Define.
* libcoff.h: Regenerate.
Several ld plugin tests currently FAIL on 32-bit Solaris/SPARC:
FAIL: load plugin with source
FAIL: plugin claimfile lost symbol with source
FAIL: plugin claimfile replace symbol with source
FAIL: plugin claimfile resolve symbol with source
FAIL: plugin claimfile replace file with source
FAIL: plugin set symbol visibility with source
FAIL: plugin ignore lib with source
FAIL: plugin claimfile replace lib with source
FAIL: plugin 2 with source lib
FAIL: load plugin 2 with source
FAIL: load plugin 2 with source and -r
FAIL: plugin 3 with source lib
FAIL: load plugin 3 with source
FAIL: load plugin 3 with source and -r
FAIL: PR ld/20070
all of them in the same way:
./ld-new: BFD (GNU Binutils) 2.34.50.20200328 internal error, aborting at /vol/src/gnu/binutils/hg/master/git/bfd/elf32-sparc.c:154 in sparc_final_write_processing
This happens when bfd_get_mach returns 0 when abfd refers to a source
file:
$11 = {
filename = 0x28c358 "/vol/src/gnu/binutils/hg/master/local/ld/testsuite/ld-plugin/func.c (symbol from plugin)", xvec = 0x24ed6c <sparc_elf32_sol2_vec>,
[...]
While I could find no specification what abfd's are allowed/expected in
*_final_write_processing, I could find no other target that behaved the
same. And indeed ignoring the 0 case fixes the failures. The code now
errors for other values. 64-bit SPARC is not affected because it doesn't
have a specific implementation of elf_backend_final_write_processing.
Tested on sparc-sun-solaris2.11.
2020-04-07 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
Nick Clifton <nickc@redhat.com>
* elf32-sparc.c (sparc_final_write_processing): Fix whitespace.
<0>: Ignore.
<default>: Error rather than abort.
Remove zero-sized relocation section from a section group since it has
been removed from the output.
PR ld/25767
* elf.c (_bfd_elf_fixup_group_sections): Remove zero-sized
relocation section from section group.
The stub size in GO32 executables has historically been fixed at 2048
bytes, due to hardcoded limitations in bfd. Recent patches to djgpp by
Stas Sergeev (CC'd) have pushed the stub right up to this limit, so if
any further expansion is desired, this must first be patched in bfd.
This series includes three patches: The first changes the meaning of
the bfd->origin field slightly, so that target code can use this to
specify an offset into the file where the actual bfd is located.
The second patch then uses this to enable support for variable-sized
stubs in the coff-go32-exe format.
The final patch is only a cleanup, it normalizes function and variable
names in coff-stgo32.c so that they all begin with the same prefix.
bfd * bfdio.c (bfd_bread, bfd_tell, bfd_seek, bfd_mmap): Always add
bfd->origin to file offset.
* bfdwin.c (bfd_get_file_window): Likewise.
* bfd.c: Clarify the use of the bfd->origin field.
* bfd-in2.h: Regenerate.
* coff-i386.c: Don't include go32exe.h. Allow overriding
coff_write_object_contents via COFF_WRITE_CONTENTS.
* coff-stgo32.c (go32exe_cleanup, go32exe_mkobject)
(go32exe_write_object_contents): New functions.
(go32exe_temp_stub, go32exe_temp_stub_size): New static globals.
(COFF_WRITE_CONTENTS, GO32EXE_DEFAULT_STUB_SIZE): Define.
(create_go32_stub): Remove check for 2k size limit. Read stub
from go32exe_temp_stub if present.
(go32_stubbed_coff_bfd_copy_private_bfd_data): Allocate and
copy variable-length stub.
(go32_check_format): Read stub to go32exe_temp_stub, set
origin, return go32exe_cleanup.
(adjust_filehdr_in_post, adjust_filehdr_out_pre)
(adjust_filehdr_out_post, adjust_scnhdr_in_post)
(adjust_scnhdr_out_pre, adjust_scnhdr_out_post)
(adjust_aux_in_post, adjust_aux_out_pre, adjust_aux_out_post):
Remove functions and their associated #defines.
* coffcode.h (coff_mkobject_hook): Remove stub copying code.
* libcoff-in.h: (struct coff_tdata): New field stub_size.
Rename field go32stub to stub.
* libcoff.h: Regenerate.
* coff-stgo32.c (go32_check_format): Rename to...
(go32exe_check_format): ...this.
(go32_stubbed_coff_bfd_copy_private_bfd_data): Rename to...
(go32exe_copy_private_bfd_data): ...this.
(stub_bytes): Rename to...
(go32exe_default_stub): ...this.
(create_go32_stub): Rename to...
(go32exe_create_stub): ...this.
* coff-stgo32.c (go32exe_copy_private_bfd_data): Avoid realloc
when possible.
include * coff/go32exe.h: Remove file.
* coff/internal.h (struct internal_filehdr): Remove field
go32stub. Remove flag F_GO32STUB.
Since value of non-preemptible absolute symbol (SHN_ABS) won't change,
only relocations, which can be resolved as absolute value + addend, and
GOTPCREL relocations, where absolute value + addend is stored in the GOT
slot, against non-preemptible absolute symbol are allowed in PIE and
shared library.
Also convert load relocation to R_386_32, R_X86_64_32S or R_X86_64_32 for
relocation against non-preemptible absolute symbol. Don't convert to
R_X86_64_32S nor R_X86_64_32 for non-preemptible absolute symbol if they
overflow.
bfd/
PR ld/25749
PR ld/25754
* elf32-i386.c (elf_i386_convert_load_reloc): Convert load
relocation to R_386_32 for relocation against non-preemptible
absolute symbol.
(elf_i386_check_relocs): Call _bfd_elf_x86_valid_reloc_p. Don't
allocate dynamic relocation for non-preemptible absolute symbol.
(elf_i386_relocate_section): Pass sec to
GENERATE_DYNAMIC_RELOCATION_P.
* elf64-x86-64.c (R_X86_64_converted_reloc_bit): Moved.
(elf_x86_64_convert_load_reloc): Covert load relocation to
R_X86_64_32S or R_X86_64_32 for relocation against non-preemptible
absolute symbol. Don't convert to R_X86_64_32S nor R_X86_64_32
for non-preemptible absolute symbol if they overflow.
(elf_x86_64_check_relocs): Call _bfd_elf_x86_valid_reloc_p. Set
tls_type for GOT slot to GOT_ABS for non-preemptible absolute
symbol. Don't allocate dynamic relocation for non-preemptible
absolute symbol.
(elf_x86_64_relocate_section): Don't generate relative relocation
for GOTPCREL relocations aganst local absolute symbol. Pass sec
to GENERATE_DYNAMIC_RELOCATION_P.
* elfxx-x86.c (elf_x86_allocate_dynrelocs): No dynamic relocation
against non-preemptible absolute symbol.
(_bfd_elf_x86_valid_reloc_p): New function.
(_bfd_x86_elf_size_dynamic_sections): No dynamic relocation for
GOT_ABS GOT slot.
* elfxx-x86.h (GENERATE_DYNAMIC_RELOCATION_P): Add an SEC
argument. Don't generate dynamic relocation against
non-preemptible absolute symbol.
(ABS_SYMBOL_P): New.
(GENERATE_RELATIVE_RELOC_P): Don't generate relative relocation
against non-preemptible absolute symbol.
(GOT_ABS): New.
(R_X86_64_converted_reloc_bit): New. Moved from elf64-x86-64.c.
(_bfd_elf_x86_valid_reloc_p): New.
ld/
PR ld/25749
PR ld/25754
* testsuite/ld-elf/linux-x86.exp: Run ld/25749 tests.
* testsuite/ld-elf/pr25749-1.c: New file.
* testsuite/ld-elf/pr25749-1a.c: Likewise.
* testsuite/ld-elf/pr25749-1b.c: Likewise.
* testsuite/ld-elf/pr25749-1b.err: Likewise.
* testsuite/ld-elf/pr25749-1c.c: Likewise.
* testsuite/ld-elf/pr25749-1d.c: Likewise.
* testsuite/ld-elf/pr25749-2.c: Likewise.
* testsuite/ld-elf/pr25749-2a.s: Likewise.
* testsuite/ld-elf/pr25749-2b.s: Likewise.
* testsuite/ld-elf/pr25749.rd: Likewise.
* testsuite/ld-elf/pr25754-1a.c: Likewise.
* testsuite/ld-elf/pr25754-1b.s: Likewise.
* testsuite/ld-elf/pr25754-2a.c: Likewise.
* testsuite/ld-elf/pr25754-2b.err: Likewise.
* testsuite/ld-elf/pr25754-2b.s: Likewise.
* testsuite/ld-elf/pr25754-3a.c: Likewise.
* testsuite/ld-elf/pr25754-3b.s: Likewise.
* testsuite/ld-elf/pr25754-4a.c: Likewise.
* testsuite/ld-elf/pr25754-4b.s: Likewise.
* testsuite/ld-elf/pr25754-4c.s: Likewise.
* testsuite/ld-elf/pr25754-5a.c: Likewise.
* testsuite/ld-elf/pr25754-5b.s: Likewise.
* testsuite/ld-elf/pr25754-5c.s: Likewise.
* testsuite/ld-elf/pr25754-6a.c: Likewise.
* testsuite/ld-elf/pr25754-6b.s: Likewise.
* testsuite/ld-x86-64/pr19609-6a.d: Don't expect linker error.
When you have a Thumb only PLT then the address in the GOT for PLT0 needs to
have the Thumb bit set since the instruction used in PLTn to get there is
`ldr.w pc` which is an inter-working instruction:
the PLT sequence in question is
00000120 <foo@plt>:
120: f240 0c98 movw ip, #152 ; 0x98
124: f2c0 0c01 movt ip, #1
128: 44fc add ip, pc
12a: f8dc f000 ldr.w pc, [ip]
12e: e7fc b.n 12a <foo@plt+0xa>
Disassembly of section .text:
00000130 <bar>:
130: b580 push {r7, lr}
132: af00 add r7, sp, #0
134: f7ff fff4 bl 120 <foo@plt>
and previously the linker would generate
Hex dump of section '.got':
...
0x000101b8 40010100 00000000 00000000 10010000 @...............
Which would make it jump and transition out of thumb mode and crash since you
only have thumb mode on such cores.
Now it correctly generates
Hex dump of section '.got':
...
0x000101b8 40010100 00000000 00000000 11010000 @...............
Thanks to Amol for testing patch and to rgujju for reporting it.
bfd/ChangeLog:
PR ld/16017
* elf32-arm.c (elf32_arm_populate_plt_entry): Set LSB of the PLT0
address in the GOT if in thumb only mode.
ld/ChangeLog:
PR ld/16017
* testsuite/ld-arm/arm-elf.exp (thumb-plt-got): New.
* testsuite/ld-arm/thumb-plt-got.d: New test.
When I previously changed these offsets I had incorrectly used an offset of -2
for this Thumb2 PLT. Unfortunately because we had no tests for this PLT I had
missed that the result was incorrect.
This patch fixes the offset to PC .-4 so that it correctly addresses the
previous instruction and adds a test for this PLT stub.
bfd/ChangeLog:
* elf32-arm.c (elf32_thumb2_plt_entry): Fix PC-rel offset.
ld/ChangeLog:
* testsuite/ld-arm/arm-elf.exp (thumb-plt): New.
* testsuite/ld-arm/thumb-plt.d: New test.
* testsuite/ld-arm/thumb-plt.s: New test.
* mmo.c (mmo_scan): Create .text section only when needed, not
from the start.
For the test-case at hand, the .data section is created and output
first by the linker, but the mmo input-reader mmo_scan always creates
a .text section. Since sections are output in the order in which
they're created, it's output first, breaking the assumption that
obcopy without options (or with -p) creates output identical to its
input. The point of creating it at the top of mmo_scan is a trivial
default assignment for the current section variable "sec". Instead we
now defer the default, creating it only when needed and sec is NULL.
* vms-alpha.c (image_write): Check bounds for sections without
contents too. Error on non-zero write to section without
contents.
(_bfd_vms_slurp_etir): Check return of image_write* functions.
PR binutils/pr25662
bfd * libcoff-in.h (struct pe_tdata): Rename the insert_timestamp
field to timestamp and make it an integer.
* libcoff.h: Regenerate.
* peXXigen.c (_bfd_XXi_only_swap_filehdr_out): Test the timestamp
field in the pe_data structure rather than the insert_timestamp
field.
binutils* objcopy.c (copy_object): When copying PE format files set the
timestamp field in the pe_data structure if the preserve_dates
flag is set.
* testsuite/binutils-all/objcopy.exp (objcopy_test) Use
--preserve-dates in place of the -p option, in order to make its
effect more obvious.
ld * emultempl/pe.em (after_open): Replace initialisation of the
insert_timestamp field in the pe_data structure with an
initialisation of the timestamp field.
* emultemp/pep.em: Likewise.
* pe-dll.c (fill_edata): Use the timestamp field in the pe_data
structure instead of the insert_timestamp field.
I doubt anyone will want to create more than 16M debug location
entries. If there is no bound the object format allows for 32-bit
indices and of course fuzzers find that and attempt allocation of up
to a 16G byte array. The patch also fixes potential integer overflows
in calculating the array size.
* vms-alpha.c (dst_define_location): Limit size of dst_ptr_offsets
array.
(_bfd_vms_slurp_object_records): Rename "err" to "ok".
PRINT_DEBUG_RELOC_INFO_BEFORE() macro prints bunch of parameters
for debugging purposes. Due to a seemingly copy/paste mistake,
the "input_section->vma" is printed under the field name
"symbol_section->vma". This commit fixes that.
This fix is a courtesy of xiangzhai.
* elf32-arc.c (PRINT_DEBUG_RELOC_INFO_BEFORE): Use the
correct field name in the output string.
