A previous patch made ld fail early on Thumb-only where branch_type is
ST_BRANCH_UNKNOWN.
However, this fails erroneously when the target is undefweak: in that
case the branch should be replaced by a branch to the next instruction
(or nop.w on thumb2). This patch accepts this case and restores the
previous behaviour in such cases.
This was reported by failures in the GCC testsuite, where we fail to
link executables because __deregister_frame_info is undefweak:
(__deregister_frame_info): Unknown destination type (ARM/Thumb) in ...crtbegin.o
crtbegin.o: in function `__do_global_dtors_aux':
crtstuff.c:(.text+0x52): dangerous relocation: unsupported relocation
Since linker calls bfd_plugin_object_p, which calls ld_plugin_object_p,
only for command-line input objects, pass true to ld_plugin_object_p so
that the same input IR file won't be included twice if the new LTO hook,
LDPT_REGISTER_CLAIM_FILE_HOOK_V2 isn't used.
PR ld/32153
* plugin.c (bfd_plugin_object_p): Pass true to ld_plugin_object_p.
Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
LoongArch: Not append rela for absolute symbol
Use la.global to get absolute symbol like la.abs.
la.global put address of a global symbol into a got
entry and append a rela for it, which will be used
to relocate by dynamic linker. Dynamic linker should
not relocate for got entry of absolute symbol as it
stores symval not symbol's address.
LoongArch: Add macros to get opcode and register of instructions appropriately
Currently, we get opcode of an instruction by manipulate the binary with
it's mask, it's a bit of a pain. Now a macro is defined to do this and a
macro to get the RD and RJ registers which is applicable to most instructions
of LoongArch are added.
This patch initializes the "op" variable in skip_cfa_op() function
of bfd/elf-eh-frame.c to "0" at its declaration point to avoid the
"maybe-uninitialized" warning.
Building binutils on a system with GCC version 13.2.0 and a configure
command that sets the optimization level to "-Og" leads to a build
failure because of a warning being treated as an error:
---------------------------------------------------------------------
$ ./configure CFLAGS="-Og"
$ make
...
CC elf-eh-frame.lo
/src/gdb/bfd/elf-eh-frame.c: In function 'skip_cfa_op':
/src/gdb/bfd/elf-eh-frame.c:354:33: error: 'op' may be used
uninitialized [-Werror=maybe-uninitialized]
354 | switch (op & 0xc0 ? op & 0xc0 : op)
| ~~~~~~~~~~~~~~~~~~~~~~^~~~
/src/gdb/bfd/elf-eh-frame.c:348:12: note: 'op' was declared here
348 | bfd_byte op;
| ^~
cc1: all warnings being treated as errors
...
---------------------------------------------------------------------
The relevant code snippet related to this warning looks like:
---------------------------------------------------------------------
static inline bool
read_byte (bfd_byte **iter, bfd_byte *end, unsigned char *result)
{
if (*iter >= end)
return false;
*result = *((*iter)++);
return true;
}
static bool
skip_cfa_op (bfd_byte **iter, bfd_byte *end,...)
{
bfd_byte op;
if (!read_byte (iter, end, &op))
return false;
switch (op & 0xc0 ? op & 0xc0 : op)
...
}
---------------------------------------------------------------------
This warning probably happens because "-Og" results in GCC not
inlining the "read_byte()" function. Therefore, GCC treats its
invocation inside "skip_cfa_op()" like a black box and that ends
in the aforementioned warning.
Acknowledgement:
Lancelot Six -- for coming with the idea behind this fix.
Jan Beulich -- for reviewing.
bfd/ChangeLog:
* elf-eh-frame.c (skip_cfa_op): Initialize the "op" variable.
We don't really want to align the last section's size to object
alignment (when that section may itself not be aligned as much), we want
image size to be a multiple thereof.
Commit "b67a17aa7c0c478a" modified the logic of allocating dynamic
relocation space for TLS GD/IE, but only modified the logic of
generation dynamic relocations for TLS GD/IE in ABI v2.00. When
linking an object file of ABI v1.00 with bfd ld of ABI v2.00, it
will cause an assertion failure.
Modified the dynamic relocation generation logic of TLS GD/IE
in ABI v1.00 to be consistent with ABI v2.00.
We recently fixed a bug in libgcc
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115360)
where a symbol was missing a %function .type decoration.
This meant the linker would silently pick the wrong type of 'farcall
stub', involving Arm-mode instructions on Thumb-only CPUs.
This patch emits an error instead, and warns in some other cases, to
encourage users to add the missing '.type foo,%function' directive.
In practice: in arm_type_of_stub() we no longer try to infer which
stub to use if the destination is of unknown type and the CPU is
Thumb-only; so we won't lie to elf32_arm_size_stubs() which does not
check branch_type.
If branch_type is ST_BRANCH_TO_ARM but the CPU is Thumb-only, we now
convert it to ST_BRANCH_TO_THUMB only if the destination is of
absolute type. This is to support the case where the destination of
the branch is defined by the linker script (see thumb-b-lks-sym.s and
thumb-bl-lks-sym.s testcases for instance).
The motivating case is covered by the new farcall-missing-type
testcase, where we now emit an error message. We do not emit an error
when branch_type is ST_BRANCH_UNKNOWN and the CPU supports Arm-mode: a
lot of legacy code (e.g. newlib's crt0.S) lacks the corresponding
'.type foo, %function' directives and even a (too verbose) warning
could be perceived as a nuisance.
Existing testcases where such a warning would trigger:
arm-static-app.s (app_func, app_func2)
arm-rel32.s (foo)
arm-app.s (app_func)
rel32-reject.s () main)
fix-arm1176.s (func_to_branch_to)
but this list is not exhaustive.
For the sake of clarity, the patch replaces occurrences of
sym.st_target_internal = 0; with
sym.st_target_internal = ST_BRANCH_TO_ARM;
enum arm_st_branch_type is defined in include/elf/arm.h, and relies on
ST_BRANCH_TO_ARM==0, as sym.st_target_internal is also initialized to
0 in other target-independent parts of BFD code. (For instance,
swapping the ST_BRANCH_TO_ARM and ST_BRANCH_TO_THUMB entries in the
enum definition leads to 'interesting' results...)
Regarding the testsuite:
* new expected warning for thumb-b-lks-sym and thumb-bl-lks-sym
* new testcase farcall-missing-type to check the new error case
* attr-merge-arch-2b.s, branch-futures (and bfs-1.s) updated to avoid
a diagnostic
Tested on arm-eabi and arm-pe with no regression.
MSVC 2022 is more pedantic than MSVC 2019 when it comes to loading PDB
files in readonly mode, and was rejecting PDB files generated by binutils
because of their invalid free-space bitmaps. It's unknown what would
have happened if you tried to use MS tools to modify a PDB created by
binutils, but it probably would have led to a corrupted file.
This patch fixes pdb_write_bitmap so we generate files that MSVC will accept.
Specifically there were three things we were doing wrong:
- We weren't including the superblock (block 0)
- We were setting bits in bytes backwards (MSB to LSB, rather than LSB to MSB)
- We should have been marking the contents of stream 0 as free. This is
because, as the comment says, it's intended to be used for the
directory for the previous write, to allow atomic updates.
R_386_TLS_IE is used only in
movl foo@indntpoff, %eax
movl foo@indntpoff, %reg
addl foo@indntpoff, %reg
R_386_TLS_DESC_CALL and R_X86_64_TLSDESC_CALL are used only in
call *x@tlscall(%[er]ax)
* elf32-i386.c (elf_i386_check_tls_transition): Use foo@indntpoff
in comments for R_386_TLS_IE check.
(elf_i386_tls_transition): Use @tlscall in comments for
R_386_TLS_DESC_CALL check.
* elf64-x86-64.c (elf_x86_64_tls_transition): Use @tlscall in
comments for R_X86_64_TLSDESC_CALL check.
Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
TLS descriptor call,
call *x@tlsdesc(%rax)
or
call *x@tlsdesc(%eax)
calls _dl_tlsdesc_return which expects that RAX/EAX points to the TLS
descriptor. Update x86 linker to issue an error with or without TLS
transition.
bfd/
PR ld/32123
* elf32-i386.c (elf_i386_check_tls_transition): Move
R_386_TLS_DESC_CALL to ...
(elf_i386_tls_transition): Here.
* elf64-x86-64.c (elf_x86_64_check_tls_transition): Move.
R_X86_64_TLSDESC_CALL check to ...
(elf_x86_64_tls_transition): Here.
ld/
PR ld/32123
* testsuite/ld-i386/i386.exp: Run tlsgdesc3.
* testsuite/ld-i386/tlsgdesc3.d: New file.
* testsuite/ld-x86-64/tlsdesc5.d: Likewise.
* testsuite/ld-x86-64/x86-64.exp: Run tlsdesc5.
Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
Since R_386_TLS_DESC_CALL can only be used with
call *variable@TLSCALL(%eax)
and R_X86_64_TLSDESC_CALL can only be used with
call *variable@TLSCALL(%rax)
update TLS transition error report to display the expected register in
indirect CALL.
bfd/
PR ld/32017
* elfxx-x86.c (_bfd_x86_elf_link_hash_table_create): Initialize
the ax_register field.
(_bfd_x86_elf_link_report_tls_transition_error): Report the
expected register in elf_x86_tls_error_indirect_call error.
* elfxx-x86.h (elf_x86_link_hash_table): Add ax_register.
ld/
PR ld/32017
* testsuite/ld-i386/tlsgdesc2.d: Updated.
* testsuite/ld-i386/tlsgdesc2.s: Change jmp to call via ECX.
* testsuite/ld-x86-64/tlsdesc4.d: Updated.
* testsuite/ld-x86-64/tlsdesc4.s: Change jmp to call via RCX.
Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
