At the cost of an extra field in the symbol table hash entries, this
simplification to the relocate_section dynamic reloc test should help
maintainability.
* elf64-ppc.c (struct ppc_link_hash_entry): Add weakref.
(ppc64_elf_copy_indirect_symbol): Set weakref. Don't merge
dyn_relocs for weakdefs.
(alias_readonly_dynrelocs): New function.
(ppc64_elf_adjust_dynamic_symbol): Use alias_readonly_dynrelocs.
(ppc64_elf_relocate_section): Simplify condition under which
dyn_relocs are emitted.
This patch fixes quite a lot of confusion in allocate_dynrelocs over
ifuncs. Function descriptors make ELFv1 quite different to ELFv2.
PR 20472
* elf64-ppc.c (ppc64_elf_before_check_relocs): Tweak abiversion test.
(readonly_dynrelocs): Comment fix.
(global_entry_stub): New function.
(ppc64_elf_adjust_dynamic_symbol): Tweak abiversion test. Match
ELFv2 code deciding on dynamic relocs vs. global entry stubs to
that in size_global_entry_stubs, handling ifunc too. Delete dead
weak sym code.
(allocate_dynrelocs): Ensure dyn_relocs field is cleared when no
dyn_relocs are needed. Correct handling of ifunc dyn_relocs.
Tidy ELIMINATE_COPY_RELOCS code, only setting dynindx for
undefweak syms. Expand and correct comments.
(size_global_entry_stubs): Ensure symbol is defined.
(ppc64_elf_relocate_section): Match condition under which
dyn_relocs are emitted to that in allocate_dynrelocs.
gdb/ChangeLog:
* MAINTAINERS (Write After Approval): Add "Carl Love".
gdb/testsuite/ChangeLog:
* gdb.arch/powerpc-power.s: Add new Power9 instruction tests
and sync up the test with tests in gas/testsuite/gas/ppc.
* gdb.arch/powerpc-power.exp: Likewise.
The GDB testsuite reports 5 test failures on Power 7 instructions.
Additionally the ppc test is missing the new Power 9 instructions as
well as a large number of older instructions. Additionally, some
instruction names have changed or been deleted. This patch
fixes the test failures and completely updates the test to make it
consistent with the supported Power 9 instructions listed in:
gas/testsuite/gas/ppc/power7.d
gas/testsuite/gas/ppc/power8.d
gas/testsuite/gas/ppc/power9.d
gas/testsuite/gas/ppc/altivec.d
gas/testsuite/gas/ppc/altivec2.d
gas/testsuite/gas/ppc/altivec3.d
gas/testsuite/gas/ppc/vsx.d
gas/testsuite/gas/ppc/vsx2.d
gas/testsuite/gas/ppc/vsx3.d
-----------------------------------------------------
gdb/testsuite/ChangeLog
2016-08-18 Carl Love <cel@us.ibm.com>
* gdb.arch/powerpc-power.s: Add new Power9 instruction tests
and sync up the test with tests in gas/testsuite/gas/ppc.
* gdb.arch/powerpc-power.exp: Likewise.
The patch fixes the record support of Hardware Transactional Memory
instructions on Power. It also solves a large number of unexpected failures
from gdb.reverse testcases sigall-precsave.exp and sigall-reverse.exp that
occur on distros which glibc uses HTM instructions.
gdb/ChangeLog
2016-08-18 Edjunior Barbosa Machado <emachado@linux.vnet.ibm.com>
* rs6000-tdep.c (ppc_process_record_op31): Handle HTM instructions.
This error message should not contain the word symbol:
(gdb) remove-inferiors 1
Warning: Can not remove current symbol inferior 1.
gdb/ChangeLog:
* inferior.c (remove_inferior_command): Fix error message.
gdb/testsuite/ChangeLog:
* gdb.multi/remove-inferiors.exp (test_remove_inferiors): Fix
expected error message.
I noticed that the remove-inferiors command was not tested, and as I am
doing some changes related to the user selection, I want to make sure I
don't break it. For example, I want to make sure it's not possible to
remove the current inferior.
gdb/testsuite/ChangeLog:
* gdb.multi/remove-inferiors.exp: New file.
* gdb.multi/remove-inferiors.c: New file.
Build the test objects with the in-tree assembler. Also fix some cascading
error messages caused by not resetting the skip_call_tls_get_addr_ flag
after printing the error.
gold/
* i386.cc (Target_i386): Reset skip_call_tls_get_addr_ after printing
error message.
* testsuite/Makefile.am (pr20216a): Add missing dependencies.
(pr20308a): Add -Bgcctestdir/ to compile rules.
* testsuite/Makefile.in: Regenerate.
The common sim tracing code already handles loading and tracking of
symbols from the target program so that it can show symbol info in
trace/disassembly calls. Once we touch up the trace code and add a
few API callbacks, ports don't need to do loading and searching of
symbol tables themselves anymore.
GLIBC BZ#20311 introduced a change to install proc_service.h so that gdb
didn't have to use the version it embeds in gdb_proc_service.h. The
embedded version is guarded by HAVE_PROC_SERVICE_H and
gdb_proc_service.h has a number other of includes and definitions, all
of which are uncondional except for an include for gregset.h. This is
only included if HAVE_PROC_SERIVCE_H is not defined.
This causes a build failure when cross compiling gdb with the latest
glibc because type definitions in gregset are used independently of
HAVE_PROC_SERIVCE_H. In particular, they are used in gdb_proc_service.h
when PRFPREGSET_T_BROKEN is set.
The error messages on the failure are
----
binutils-gdb/gdb/gdb_proc_service.h:173:9: error: ‘gdb_fpregset_t’ does
not name a type; did you mean ‘elf_fpregset_t’?
typedef gdb_fpregset_t gdb_prfpregset_t;
^~~~~~~~~~~~~~
elf_fpregset_t
binutils-gdb/gdb/gdb_proc_service.h:173:9: error: ‘gdb_fpregset_t’ does
not name a type; did you mean ‘elf_fpregset_t’?
typedef gdb_fpregset_t gdb_prfpregset_t;
^~~~~~~~~~~~~~
elf_fpregset_t
binutils-gdb/gdb/proc-service.c:218:15: error: ‘gdb_prfpregset_t’ does
not name a type; did you mean ‘gdb_fpregset_t’?
const gdb_prfpregset_t *fpregset)
^~~~~~~~~~~~~~~~
gdb_fpregset_t
----
This patch moves the include for gregset.h to before the code guarded by
HAVE_PROC_SERIVCE_H, so that it is always included. This is enough to
fix the build.
2016-08-15 Matthew Wahab <matthew.wahab@arm.com>
PR gdb/20457
* gdb_proc_service.h: Add an include of gregset.h
[!HAVE_PROC_SERVICE_H]: Remove the include of gregset.h.
I build GDB with -fsanitize=address, and see the error in tests,
(gdb) PASS: gdb.linespec/ls-errs.exp: lang=C++: break 3 foo
break -line 3 foo^M
=================================================================^M
==4401==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x603000047487 at pc 0x819d8e bp 0x7fff4e4e6bb0 sp 0x7fff4e4e6ba8^M
READ of size 1 at 0x603000047487 thread T0^[[1m^[[0m^M
#0 0x819d8d in explicit_location_lex_one /home/yao/SourceCode/gnu/gdb/git/gdb/location.c:502^M
#1 0x81a185 in string_to_explicit_location(char const**, language_defn const*, int) /home/yao/SourceCode/gnu/gdb/git/gdb/location.c:556^M
#2 0x81ac10 in string_to_event_location(char**, language_defn const*) /home/yao/SourceCode/gnu/gdb/git/gdb/location.c:687^
the code in question is:
> /* Special case: C++ operator,. */
> if (language->la_language == language_cplus
> && strncmp (*inp, "operator", 8) <--- [1]
> && (*inp)[9] == ',')
> (*inp) += 9;
> ++(*inp);
The error is caused by the access to (*inp)[9] if 9 is out of its bounds.
However [1] looks odd to me, because if strncmp returns true (non-zero),
the following check "(*inp)[9] == ','" makes no sense any more. I
suspect it was a typo in the code we meant to "strncmp () == 0". Another
problem in the code above is that if *inp is "operator,", we first
increment *inp by 9, and then increment it by one again, which is wrong
to me. We should only increment *inp by 8 to skip "operator", and go
back to the loop header to decide where we stop.
gdb:
2016-08-15 Yao Qi <yao.qi@linaro.org>
* location.c (explicit_location_lex_one): Compare the return
value of strncmp with zero. Don't check (*inp)[9]. Increment
*inp by 8.
A few funcs are only used locally, so mark them static to avoid warnings
due to -Wmissing-prototypes.
Some funcs cast the return value wrong, so drop them (and let void * just
work by default).
Update some prototypes to be new style.
The current machs.h mixes common enums with Blackfin-specific defines.
This causes us troubles with header inclusion order such that we can't
drop the old SIM_CPU typedef (which is duplicated in common code). By
splitting the two up, we can unwind this dependency chain, and drop the
old typedef. It also fixes building with older gcc versions.
* readelf.c (process_symbol_table): Generate a warning if a local
symbol is found at and offste greater than or equal to the sh_info
field of it's section header.
* simulator.c (aarch64_step): Revert pervious delta.
(aarch64_run): Call sim_events_tick after each
instruction is simulated, and if necessary call
sim_events_process.
* simulator.h: Revert previous delta.
I see the following warning when running signals-state-child.exp.
gdb/testsuite/gdb.base/signals-state-child.c:77:4: warning: too many arguments for format [-Wformat-extra-args]
fprintf (out, "sigaction={sa_handler=", i);
^
this patch is to remove the argument from fprintf.
gdb/testsuite:
2016-08-12 Yao Qi <yao.qi@linaro.org>
* gdb.base/signals-state-child.c (main): Remove "i" from fprintf's
argument list.
* testsuite/ld-undefined/weak-fundef.s: New.
* testsuite/ld-undefined/weak-undef.t: Don't specify filename.
* testsuite/ld-undefined/weak-undef.exp: Run new tests. Rearrange
much of old code. Use is_elf_format to select targets.
* interp.c (sim_create_inferior): Allow for being called with a
NULL abfd parameter. If a bfd is provided, initialise the sim
with that start address.
* simulator.c (HALT_NYI): Just print out the numeric value of the
instruction when not tracing.
(aarch64_step): Change from static to global.
* simulator.h: Add a prototype for aarch64_step().
Fixes, on NIOS GNU/Linux:
In file included from
/scratch/mbilal/nois-lite/src/gdb-trunk/gdb/gdbserver/../nat/linux-ptrace.c:26:0:
/scratch/mbilal/nois-lite/src/gdb-trunk/gdb/gdbserver/../gregset.h:27:23:
error: unknown type name 'gregset_t'
#define GDB_GREGSET_T gregset_t
^
Fix this by including sys/procfs.h directly. We shouldn't really be
including a gdb-only header in a gdb/nat/ file, anyway. Whoops.
gdb/ChangeLog:
2016-08-11 Pedro Alves <palves@redhat.com>
PR gdb/20413
* nat/linux-ptrace.c: Include <sys/procfs.h> instead of
"gregset.h".
PR ld/20436
* testsuite/lib/ld-lib.exp (check_gcc_plugin_enabled): When not
testing remotely, check to see if target compiler is installed
before trying to run it.
parse_aarch64_imm_float was accepting -0.0 even though that's not
a valid immediate for any instruction. The FPIMM0 caller rejected
it, but the FPIMM one would silently treat it as -2.0.
This patch rejects -0.0 and adds testcases to illegal.[sd].
Before the patch, the final error emitted for illegal.s was:
Error: cannot do 16-byte relocation
which was matched by:
[^:]*:569: Error: .*
The error was reported against the last line of the file rather than
the instruction that required the reloc. Adding more instructions
meant that the line number also changed.
Reporting against the wrong line isn't good from a QoI perspective
but isn't what I'm trying to fix here. Until it's fixed, I thought
it would be better to adjust the match to be against an end-of-file
comment rather than against whatever the last instruction happens to be.
gas/
* config/tc-aarch64.c (parse_aarch64_imm_float): Reject -0.0.
* testsuite/gas/aarch64/illegal.s, testsuite/gas/aarch64/illegal.l:
Add tests for -0.0. Add an end-of-file comment.
Undefined weak functions, like __gmon_start__, were not being made
dynamic or emitting plt call code. While the behaviour of undefined
weak symbols is not defined in the ELF standard, the intention on
powerpc64 was to make it possible to link without a definition of such
symbols and at run time behave the same as if a definition was found
at link time in a shared library.
* elf64-ppc.c (ppc64_elf_adjust_dynamic_symbol): Don't exit with
non_got_ref true in any case where we could have generated dynbss
copies but decide not to do so.
Complement commit b75d42bce5 ("Fix mips segfault on GOT access of
absolute symbol") and unmark the `__ehdr_start' test 2 as known to fail
for the n32 and n64 ABIs, as with the change referred in place the
causing issue has been properly addressed and consequently the tests do
not fail anymore and neither are supposed to.
ld/
PR ld/15428
* testsuite/ld-mips-elf/mips-elf.exp: Un-KFAIL `__ehdr_start'
test 2.
Right after a fork is detected, we detach breakpoints from the child
(detach_breakpoints), which calls into target_remove_breakpoint with
inferior_ptid pointing at the child process, but leaves the breakpoint
marked inserted (in the parent).
The problem is that record-full.c always deletes all knowledge of the
breakpoint. Then when we later really delete the breakpoint from the
parent, we fail the assertion, since the breakpoint is unexpectedly
not found in the record-full.c breakpoint table.
The fix is simply to not forget about the breakpoint if we're
detaching it from a fork child.
gdb/ChangeLog:
2016-08-10 Pedro Alves <palves@redhat.com>
PR gdb/19187
* record-full.c (record_full_remove_breakpoint): Don't remove the
breakpoint from the record_full_breakpoints VEC if we're detaching
the breakpoint from a fork child.
gdb/testsuite/ChangeLog:
2016-08-10 Pedro Alves <palves@redhat.com>
PR gdb/19187
* gdb.reverse/waitpid-reverse.exp: Add comment and remove
setup_kfails.
Makes the code more obvious.
gdb/ChangeLog:
2016-08-10 Pedro Alves <palves@redhat.com>
PR gdb/19187
* breakpoint.c (insertion_state_t): Delete.
(enum remove_bp_reason): New.
(detach_breakpoints, remove_breakpoint_1, remove_breakpoint):
Adjust to use enum remove_bp_reason instead of insertion_state_t.
All callers pass mark_uninserted, so there's no need for the 'is'
parameter.
gdb/ChangeLog:
2016-08-10 Pedro Alves <palves@redhat.com>
PR gdb/19187
* breakpoint.c (remove_breakpoint): Remove 'is' parameter and
always pass mark_uninserted to remove_breakpoint_1.
(insert_breakpoint_locations, remove_breakpoints)
(remove_breakpoints_pid, update_global_location_list): Update
callers.
Repeat `PIC and non-PIC test 1' checks for microMIPS LA25 stubs,
covering code generation and stub symbol annotation.
ld/
* testsuite/ld-mips-elf/pic-and-nonpic-1-micromips-rel.dd: New
test.
* testsuite/ld-mips-elf/pic-and-nonpic-1-micromips-rel.nd: New
test.
* testsuite/ld-mips-elf/pic-and-nonpic-1-micromips.dd: New test.
* testsuite/ld-mips-elf/pic-and-nonpic-1-micromips.nd: New test.
* testsuite/ld-mips-elf/pic-and-nonpic-1a-micromips.s: New test
source.
* testsuite/ld-mips-elf/pic-and-nonpic-1b-micromips.s: New test
source.
* testsuite/ld-mips-elf/mips-elf.exp: Run the new tests.
For the case where a function which requires an LA25 stub is at the
beginning of a section we use a short sequence comprised of a LUI/ADDIU
instruction pair only and prepended to the associated function rather
than using a trailing jump to reach the function. This works by
checking for the offset into section of the function symbol being 0.
This is however never the case for microMIPS function symbols, which
have the ISA bit set. Consequently the short LA25 sequence is never
produced for microMIPS functions, like with the following example:
$ cat la25a.s
.abicalls
.global f1
.ent f1
f1:
.set noreorder
.cpload $25
.set reorder
.option pic0
jal f2
.option pic2
jr $31
.end f1
.global f2
.ent f2
f2:
jr $31
.end f2
$ cat la25b.s
.abicalls
.option pic0
.global __start
.ent __start
__start:
jal f1
jal f2
.end __start
$ as -mmicromips -32 -EB -o la25a.o la25a.s
$ as -mmicromips -32 -EB -o la25b.o la25b.s
$ ld -melf32btsmip -o la25 la25a.o la25b.o
$ objdump -d la25
la25: file format elf32-tradbigmips
Disassembly of section .text:
004000d0 <.pic.f2>:
4000d0: 41b9 0040 lui t9,0x40
4000d4: d420 0083 j 400106 <f2>
4000d8: 3339 0107 addiu t9,t9,263
4000dc: 0000 0000 nop
004000e0 <.pic.f1>:
4000e0: 41b9 0040 lui t9,0x40
4000e4: d420 0078 j 4000f0 <f1>
4000e8: 3339 00f1 addiu t9,t9,241
4000ec: 0000 0000 nop
004000f0 <f1>:
4000f0: 41bc 0002 lui gp,0x2
4000f4: 339c 801f addiu gp,gp,-32737
4000f8: 033c e150 addu gp,gp,t9
4000fc: f420 0083 jal 400106 <f2>
400100: 0000 0000 nop
400104: 45bf jrc ra
00400106 <f2>:
400106: 45bf jrc ra
...
00400110 <__start>:
400110: f420 0070 jal 4000e0 <.pic.f1>
400114: 0000 0000 nop
400118: f420 0068 jal 4000d0 <.pic.f2>
40011c: 0000 0000 nop
$
where `.pic.f1' could omit the trailing jump and the filler NOP and just
fall through to `f1'.
Correct the problem by masking out the ISA bit from microMIPS functions,
which fixes the earlier example:
$ objdump -d la25
la25: file format elf32-tradbigmips
Disassembly of section .text:
004000d0 <.pic.f2>:
4000d0: 41b9 0040 lui t9,0x40
4000d4: d420 0083 j 400106 <f2>
4000d8: 3339 0107 addiu t9,t9,263
...
004000e8 <.pic.f1>:
4000e8: 41b9 0040 lui t9,0x40
4000ec: 3339 00f1 addiu t9,t9,241
004000f0 <f1>:
4000f0: 41bc 0002 lui gp,0x2
4000f4: 339c 801f addiu gp,gp,-32737
4000f8: 033c e150 addu gp,gp,t9
4000fc: f420 0083 jal 400106 <f2>
400100: 0000 0000 nop
400104: 45bf jrc ra
00400106 <f2>:
400106: 45bf jrc ra
...
00400110 <__start>:
400110: f420 0074 jal 4000e8 <.pic.f1>
400114: 0000 0000 nop
400118: f420 0068 jal 4000d0 <.pic.f2>
40011c: 0000 0000 nop
$
There is no need to do anything for MIPS16 functions, because if any
LA25 stub has been generated for such a function, then it is only
required for an associated call thunk only, which is regular MIPS code
and the address of which, with the ISA bit clear, is returned by
`mips_elf_get_la25_target'.
This problem has been there since the beginning of microMIPS support:
commit df58fc944d
Author: Richard Sandiford <rdsandiford@googlemail.com>
Date: Sun Jul 24 14:20:15 2011 +0000
<https://sourceware.org/ml/binutils/2011-07/msg00198.html>, ("MIPS:
microMIPS ASE support").
bfd/
* elfxx-mips.c (mips_elf_add_la25_stub): Clear the ISA bit of
the stub address retrieved if associated with a microMIPS
function.
Nick Clifton
Alan Modra
GDB Administrator
Carl E. Love
Carl Love
Edjunior Barbosa Machado
Simon Marchi
Cary Coutant
Mike Frysinger
Matthew Wahab
Yao Qi
Roland McGrath
Pedro Alves
Richard Sandiford
Maciej W. Rozycki