More non-bugs flagged by ubsan, unless you happen to be compiling for
a 1's complement host.
cpu/
* frv.cpu (f-u12): Multiply rather than left shift signed values.
(f-label16, f-label24): Likewise.
opcodes/
* frv-ibld.c: Regenerate.
When the command "info registers" (same as "info registers general"),
is issued, _all_ the registers from a tdesc XML are printed. This
includes the registers with empty register groups (set as "") which
are supposed to be only printed by "info registers all" (or "info
all-registers").
This bug got introduced after all the overhauls that the
tdesc_register_in_reggroup_p() went through. You can see that the
logic of tdesc_register_in_reggroup_p() did NOT remain the same after
all those changes:
git difftool c9c895b9666..HEAD -- gdb/target-descriptions.c
With the current implementation, when the reg->group is an empty
string, this function returns -1, while in the working revision
(c9c895b966), it returned 0. This patch makes sure that the 0 is
returned again.
The old implementation of tdesc_register_in_reggroup_p() returned
-1 when "reggroup" was set to "all_reggroups" at line 4 below:
1 tdesc_register_reggroup_p (...)
2 {
3 ...
4 ret = tdesc_register_in_reggroup_p (gdbarch, regno, reggroup);
5 if (ret != -1)
6 return ret;
7
8 return default_register_reggroup_p (gdbarch, regno, reggroup);
9 }
As a result, the execution continued at line 8 and the
default_register_reggroup_p(..., reggroup=all_reggroups) would
return 1. However, with the current implementation of
tdesc_register_in_reggroup_p() that allows checking against any
arbitrary group name, it returns 0 when comparing the "reg->group"
against the string "all" which is the group name for "all_reggroups".
I have added a special check to cover this case and
"info all-registers" works as expected.
gdb/ChangeLog:
* target-descriptions.c (tdesc_register_in_reggroup_p): Return 0
when reg->group is empty and reggroup is not.
Change-Id: I9eaf9d7fb36410ed5684ae652fe4756b1b2e61a3
There's already existing logic to handle this on other targets, so
this patch just makes nios2 use it.
2020-01-31 Sandra Loosemore <sandra@codesourcery.com>
bfd/
* elf-eh-frame.c (_bfd_elf_write_section_eh_frame): DW_EH_PE_datarel
encodings are relative to the GOT on nios2, too.
The nios2 ABI documentation lists %gotoff as assembler syntax for the
R_NIOS2_GOTOFF relocation, used to represent a 32-bit GOT-relative offset
in data sections. This was previously unimplemented in GAS.
2020-01-31 Sandra Loosemore <sandra@codesourcery.com>
gas/
* config/tc-nios2.c (nios2_cons): Handle %gotoff as well as
%tls_ldo.
We noticed +mve was not enabling DSP instructions as it should, reported in PR
25472.
The MVE architecture extension for Armv8.1-M Mainline implies DSP extensions.
This patch reflects that in the '+mve' command line option.
gas/ChangeLog:
2020-01-31 Andre Vieira <andre.simoesdiasvieira@arm.com>
PR gas/25472
* config/tc-arm.c (armv8m_main_ext_table): Refactored +dsp adding.
(armv8_1m_main_ext_table): Refactored +dsp adding and enabled dsp for
+mve.
* testsuite/gas/arm/mve_dsp.d: New test.
This patch adds support for assembly instructions vldmia, vldmdb, vstmia
and vstmdb in MVE. This instructions are already supported for Armv8-M
Floating-point Extension.
gas/ChangeLog:
2020-01-31 Srinath Parvathaneni <srinath.parvathaneni@arm.com>
* config/tc-arm.c (fldmias): Moved inside "THUMB_VARIANT & arm_ext_v6t2"
to support VLDMIA instruction for MVE.
(fldmdbs): Moved inside "THUMB_VARIANT & arm_ext_v6t2" to support VLDMDB
instruction for MVE.
(fstmias): Moved inside "THUMB_VARIANT & arm_ext_v6t2" to support VSTMIA
instruction for MVE.
(fstmdbs): Moved inside "THUMB_VARIANT & arm_ext_v6t2" to support VSTMDB
instruction for MVE.
* testsuite/gas/arm/mve-ldst.d: New test.
* testsuite/gas/arm/mve-ldst.s: Likewise.
bfcvt converts a .S input to a .H output, so any predicated movprfx
needs to operate on .S rather than .H. In common with SVE2 narrowing
top operations, bfcvtnt doesn't accept movprfx.
2020-01-31 Richard Sandiford <richard.sandiford@arm.com>
opcodes/
* aarch64-tbl.h (aarch64_opcode): Set C_MAX_ELEM for SVE bfcvt.
Remove C_SCAN_MOVPRFX for SVE bfcvtnt.
gas/
* testsuite/gas/aarch64/sve-bfloat-movprfx.s: Use .h rather than
.s for the movprfx.
* testsuite/gas/aarch64/sve-bfloat-movprfx.d: Update accordingly.
* testsuite/gas/aarch64/sve-movprfx_28.d,
* testsuite/gas/aarch64/sve-movprfx_28.l,
* testsuite/gas/aarch64/sve-movprfx_28.s: New test.
ravenscar-thread.c needed a change to adapt to multi-target:
ravenscar_thread_target::mourn_inferior called the mourn_inferior
method on the target beneat -- but when the target beneath was the
remote target, this resulted in the ravenscar target being deleted.
Switching the order of the calls to unpush_target and the beneath's
mourn_inferior fixes this problem.
gdb/ChangeLog
2020-01-31 Tom Tromey <tromey@adacore.com>
* ravenscar-thread.c (ravenscar_thread_target::mourn_inferior):
Call beneath target's mourn_inferior after unpushing.
Change-Id: Ia80380515c403adc40505a6b3420c9cb35754370
In TUI mode, if the disassembly output for the program is less than
one screen long, then currently if the user scrolls down until on the
last assembly instruction is displayed and then tries to scroll up
using Page-Up, the display doesn't update - they are stuck viewing the
last line.
If the user tries to scroll up using the Up-Arrow, then the display
scrolls normally.
What is happening is on the Page-Up we ask GDB to scroll backward the
same number of lines as the height of the TUI ASM window. The back
scanner, which looks for a good place to start disassembling, fails to
find a starting address which will provide the requested number of new
lines before we get back to the original starting address (which is
not surprising, our whole program contains less than a screen height
of instructions), as a result the back scanner gives up and returns
the original starting address.
When we scroll with Up-Arrow we only ask the back scanner to find 1
new instruction, which it manages to do, so this scroll works.
The solution here is, when we fail to find enough instructions, to
return the lowest address we did manage to find. This will ensure we
jump to the lowest possible address in the disassembly output.
gdb/ChangeLog:
PR tui/9765
* tui/tui-disasm.c (tui_find_disassembly_address): If we don't
have enough lines to fill the screen, still return the lowest
address we found.
gdb/testsuite/ChangeLog:
PR tui/9765
* gdb.tui/tui-layout-asm-short-prog.S: New file.
* gdb.tui/tui-layout-asm-short-prog.exp: New file.
Change-Id: I6a6a7972c68a0559e9717fd8d82870b669a40af3
GDB has some commands ('+', '-', '<', and '>') for scrolling the SRC
and ASM TUI windows from the CMD window, however the help text for
these commands lists the arguments in the wrong order.
This commit updates the help text to match how GDB actually works, and
also extends the text to describe what the arguments mean, and what
the defaults are.
There should be no change in GDBs functionality after this commit.
gdb/ChangeLog:
* tui/tui-win.c (_initialize_tui_win): Update help text for '+',
'-', '<', and '>' commands.
Change-Id: Ib2624891de1f4ba983838822206304e4c3ed982e
This patch removes the leak of Nick's source directory into bfd.pot,
and emits #line for some generated files so that those files aren't
referenced by comments in the .pot file. You can see both of these
effects in the following diff. I've also removed use of an
unnecessary temp file in the make rules.
@@ -92,10 +92,8 @@ msgstr ""
#: elf64-nfp.c:238 elf64-ppc.c:1014 elf64-ppc.c:1349 elf64-ppc.c:1358
#: elf64-s390.c:328 elf64-s390.c:378 elf64-x86-64.c:285 elfn32-mips.c:3786
#: elfxx-ia64.c:324 elfxx-riscv.c:955 elfxx-sparc.c:589 elfxx-sparc.c:639
-#: elfxx-tilegx.c:912 elfxx-tilegx.c:952
-#: /work/sources/binutils/current/bfd/elfnn-aarch64.c:2215
-#: /work/sources/binutils/current/bfd/elfnn-aarch64.c:2313 elf32-ia64.c:214
-#: elf32-ia64.c:3862 elf64-ia64.c:214 elf64-ia64.c:3862
+#: elfxx-tilegx.c:912 elfxx-tilegx.c:952 elfnn-aarch64.c:2215
+#: elfnn-aarch64.c:2313 elfnn-ia64.c:214 elfnn-ia64.c:3862
#, c-format
msgid "%pB: unsupported relocation type %#x"
msgstr ""
* Makefile.am (elf32-target.h, elf64-target.h): Don't use a temp
file. Use $< and $@ in rules.
(elf32-aarch64.c, elf64-aarch64.c): Likewise.
(elf32-ia64.c, elf64-ia64.c): Likewise.
(elf32-riscv.c, elf64-riscv.c): Likewise.
(peigen.c, pepigen.c, pex64igen.c): Likewise.
(elf32-aarch64.c, elf64-aarch64.c): Don't emit $srcdir on #line.
(elf32-riscv.c, elf64-riscv.c): Likewise, and use $(SED).
(elf32-ia64.c, elf64-ia64.c): Do emit #line.
(peigen.c, pepigen.c, pex64igen.c): Likewise.
* Makefile.in: Regenerate.
We alloc, seek and read using section sizes in object files. Fuzzed
objects can have silly sizes, but that's OK if the system supports
memory over-commit. The read fails because we hit EOF and that
usually results in a graceful exit.
But if we memset before the read then the invalid size results in
attempting to write to a huge number of memory pages, and an eventual
Out Of Memory after probably swapping like crazy. So don't memset.
There really isn't a need to clear the section contents anyway. All
bytes are written with a good object file by the read and following
loop converting section index in target order to ELF section header
pointer, and the only untidy bytes are the 4 bytes past the group
flags when pointers are 8 bytes. Those don't matter but the patch
clears them for anyone poking around in a debugger. On error paths
it's as good to free section contents as it is to clear them.
Noticed when looking at PR4110 fourth test case.
PR 4110
* elf.c (setup_group): Don't clear entire section contents,
just the padding after group flags. Release alloc'd memory
after a seek or read failure.
Comparison of i.tm.base_opcode against particular but not sufficiently
specific values needs to be accompanied by other qualification. Exclude
VEX and alike encodings here, and also exclude all forms of prefixes
explicitly specified in the opcodes table. While using @GOT with such
insns may not be very useful, it also isn't with e.g. ADC and SBB, yet
these get explicitly listed in comments as supported.
Commit 9e7028aa1e ("PowerPC64 __tls_get_addr_desc") introduced a use
of @option which apparently newer makeinfo tolerates, but older ones
reject. Drop the unnecessary (a per all other uses of @option) blank.
More nonsense fixing "bugs" with left shifts of signed values. Yes,
the C standard does say this is undefined (and right shifts of signed
values are implementation defined BTW) but in practice there is no
problem with current machines. 1's complement is a thing of the past.
cpu/
* m32c.cpu (f-src32-rn-unprefixed-QI): Shift before inverting.
(f-src32-rn-prefixed-QI, f-dst32-rn-unprefixed-QI): Likewise.
(f-dst32-rn-prefixed-QI): Likewise.
(f-dsp-32-s32): Mask before shifting left.
(f-dsp-48-u32, f-dsp-48-s32): Likewise.
(f-bitbase32-16-s11-unprefixed): Multiply signed field rather than
shifting left.
(f-bitbase32-24-s11-prefixed, f-bitbase32-24-s19-prefixed): Likewise.
(h-gr-SI): Mask before shifting.
opcodes/
* m32c-ibld.c: Regenerate.
These are produced by MSVC when the '/Brepro' flag is used.
To quote from the PE specification [1]:
"The presence of an entry of type IMAGE_DEBUG_TYPE_REPRO indicates the
PE file is built in a way to achieve determinism or reproducibility. If
the input does not change, the output PE file is guaranteed to be
bit-for-bit identical no matter when or where the PE is produced.
Various date/time stamp fields in the PE file are filled with part or
all the bits from a calculated hash value that uses PE file content as
input, and therefore no longer represent the actual date and time when a
PE file or related specific data within the PE is produced. The raw data
of this debug entry may be empty, or may contain a calculated hash value
preceded by a four-byte value that represents the hash value length."
[1] https://docs.microsoft.com/en-us/windows/win32/debug/pe-format
bfd/ChangeLog:
2020-01-16 Jon Turney <jon.turney@dronecode.org.uk>
* peXXigen.c (pe_is_repro): New function.
(_bfd_XX_print_private_bfd_data_common): Note timestamp is
actually a build hash if PE_IMAGE_DEBUG_TYPE_REPRO is present.
IMAGE_DEBUG_TYPE_REPRO is defined in the latest version of the PE
specification [1]. The others are defined in Windows SDK headers and/or
reported by DUMPBIN.
[1] https://docs.microsoft.com/en-us/windows/win32/debug/pe-format
bfd/ChangeLog:
2020-01-16 Jon Turney <jon.turney@dronecode.org.uk>
* peXXigen.c (debug_type_names): Add names for new debug data type
values.
include/ChangeLog:
2020-01-16 Jon Turney <jon.turney@dronecode.org.uk>
* coff/internal.h (PE_IMAGE_DEBUG_TYPE_VC_FEATURE)
(PE_IMAGE_DEBUG_TYPE_POGO, PE_IMAGE_DEBUG_TYPE_ILTCG)
(PE_IMAGE_DEBUG_TYPE_MPX, PE_IMAGE_DEBUG_TYPE_REPRO): Add.
Use a separate iteration variable for inner loop (😊). This
generally prevented any debug directory entries after a
IMAGE_DEBUG_TYPE_CODEVIEW entry from being reported.
Don't leak the memory allocated for the section containing the debug
directory.
bfd/ChangeLog:
2020-01-16 Jon Turney <jon.turney@dronecode.org.uk>
* peXXigen.c (pe_print_debugdata): Fix the iteration variable for
inner loop. Fix a memory leak.
This patch fixes the neg/neg32 BPF instructions, which have K (=0)
instead of X (=1) in their header source bit, despite operating on
registes.
cpu/ChangeLog:
2020-01-30 Jose E. Marchesi <jose.marchesi@oracle.com>
* bpf.cpu (define-alu-insn-un): The unary BPF instructions
(neg and neg32) use OP_SRC_K even if they operate only in
registers.
opcodes/ChangeLog:
2020-01-30 Jose E. Marchesi <jose.marchesi@oracle.com>
* bpf-opc.c: Regenerate.
gas/ChangeLog:
2020-01-30 Jose E. Marchesi <jose.marchesi@oracle.com>
* testsuite/gas/bpf/alu.d: Update expected opcode for `neg'.
* testsuite/gas/bpf/alu-be.d: Likewise.
* testsuite/gas/bpf/alu32.d: Likewise for `neg32'.
* testsuite/gas/bpf/alu32-be.d: Likewise.
While vendors agree about default operand size (64 bits) and hence
unavilability of a 32-bit form, AMD honors a 16-bit operand size
override (0x66) while Intel doesn't.
Other than near returns these default to 32-bit operand size, and hence
it isn't really unlikely that 64-bit forms are meant. Hence these should
have disambiguating suffixes. In Intel mode, however, don't error in
these cases unconditionally - MASM accepts these without suffix _and_
without warning.
- 64-bit CALL permitting just a single operand size doesn't need it.
- FLDENV et al should never have had it.
It remains suspicious that a number of 64-bit only insns continue to
have the attribute, despite this being intended for .code16gcc handling
only.
The patch also fixes a case where libopcodes built for a 64-bit
bfd_vma may print different results to libopcodes built for a 32-bit
bfd_vma.
* tic4x-dis.c (tic4x_dp): Make unsigned.
While testing a GCC 10 build of our git HEAD, Sergio noticed an error
triggered by -Werror-stringop on
infcmd.c:construct_inferior_arguments. One of the things the function
does is calculate the length of the string that will hold the
inferior's arguments. GCC warns us that 'length' can be 0, which can
lead to undesired behaviour:
../../gdb/infcmd.c: In function 'char* construct_inferior_arguments(int, char**)':
../../gdb/infcmd.c:369:17: error: writing 1 byte into a region of size 0 [-Werror=stringop-overflow=]
369 | result[0] = '\0';
| ~~~~~~~~~~^~~~~~
../../gdb/infcmd.c:368:33: note: at offset 0 to an object with size 0 allocated by 'xmalloc' here
368 | result = (char *) xmalloc (length);
| ~~~~~~~~^~~~~~~~
The solution here is to assert that 'argc' is greater than 0 on entry,
which makes GCC understand that the loops always run at least once,
and thus 'length' is always > 0.
Tested by rebuilding.
gdb/ChangeLog:
2020-01-29 Pedro Alves <palves@redhat.com>
Sergio Durigan Junior <sergiodj@redhat.com>
* infcmd.c (construct_inferior_arguments): Assert that
'argc' is greater than 0.
Change-Id: Ide8407cbedcb4921de1843a6a15bbcb7676c7d26
The move of gdbsupport to the top level directory requires a small
change to src-release.sh's "getver" function, which is responsible for
determining the version string that will be appended to the release
tarball: now the create-version.sh script lives under ./gdbsupport,
and not under gdb/gdbsupport anymore.
This patch unbreaks the snapshot generation, which hasn't been working
since January 14th.
ChangeLog:
2020-01-29 Sergio Durigan Junior <sergiodj@redhat.com>
* src-release.sh (getver): Look for gdbsupport's
create-version.sh script at the current directory if tool is
"gdb".
Change-Id: Id3b8bed6583a1aaa120c07009366f6c94a62d5db
An error in commit 42cd72aa02 caused
srv_tgtobj to be overwritten and linux-ppc-low.o to be missed when
linking gdbserver for Linux on PowerPC. This patch fixes the error.
gdb/gdbserver/ChangeLog:
2020-01-29 Pedro Franco de Carvalho <pedromfc@linux.ibm.com>
* configure.srv (powerpc*-*-linux*): Use srv_tgtobj in second
assignment instead of srv_linux_obj.
New in v5:
- Use gdb_test_name for gdb_test_multiple.
- Use gdb_assert.
- Verify count matches the expected sigtraps exactly.
New in v4:
- Fix formatting nit in gdb/testsuite/gdb.arch/aarch64-brk-patterns.c.
New in v3:
- Minor formatting and code cleanups.
- Added count check to validate number of brk SIGTRAP's.
- Moved count to SIGTRAP check conditional block.
This test exercises the previous patch's code and makes sure GDB can
properly get a SIGTRAP from various brk instruction patterns.
GDB needs to be able to see the program exiting normally. If GDB doesn't
support the additional brk instructions, we will see timeouts.
We bail out with the first timeout since we won't be able to step through
the program breakpoint anyway, so it is no use carrying on.
gdb/testsuite/ChangeLog:
2020-01-29 Luis Machado <luis.machado@linaro.org>
* gdb.arch/aarch64-brk-patterns.c: New source file.
* gdb.arch/aarch64-brk-patterns.exp: New test.
New in v3:
- Code cleanups based on reviews.
New in v2:
- Fixed misc problems based on reviews.
- Switched to using gdbarch_program_breakpoint_here_p as opposed to
gdbarch_insn_is_breakpoint.
- Fixed matching of brk instructions. Previously the mask was incorrect, which
was showing up as a few failures in the testsuite. Now it is clean.
- New testcase (separate patch).
- Moved program_breakpoint_here () to arch-utils.c and made it the default
implementation of gdbarch_program_breakpoint_here_p.
--
It was reported to me that program breakpoints (permanent ones inserted into
the code itself) other than the one GDB uses for AArch64 (0xd4200000) do not
generate visible stops when continuing, and GDB will continue spinning
infinitely.
This happens because GDB, upon hitting one of those program breakpoints, thinks
the SIGTRAP came from a delayed breakpoint hit...
(gdb) x/i $pc
=> 0x4005c0 <problem_function>: brk #0x90f
(gdb) c
Continuing.
infrun: clear_proceed_status_thread (process 14198)
infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT)
infrun: proceed: resuming process 14198
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 14198] at 0x4005c0
infrun: infrun_async(1)
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 14198.14198.0 [process 14198],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: handle_inferior_event status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: stop_pc = 0x4005c0
infrun: delayed software breakpoint trap, ignoring
infrun: no stepping, continue
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 14198] at 0x4005c0
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 14198.14198.0 [process 14198],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: handle_inferior_event status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: stop_pc = 0x4005c0
infrun: delayed software breakpoint trap, ignoring
infrun: no stepping, continue
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 14198] at 0x4005c0
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 14198.14198.0 [process 14198],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: handle_inferior_event status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: stop_pc = 0x4005c0
infrun: delayed software breakpoint trap, ignoring
infrun: no stepping, continue
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 14198] at 0x4005c0
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 14198.14198.0 [process 14198],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: handle_inferior_event status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: stop_pc = 0x4005c0
infrun: delayed software breakpoint trap, ignoring
infrun: no stepping, continue
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 14198] at 0x4005c0
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 14198.14198.0 [process 14198],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
...
... which is not the case.
If the program breakpoint is one GDB recognizes, then it will stop when it
hits it.
(gdb) x/i $pc
=> 0x4005c0 <problem_function>: brk #0x0
(gdb) c
Continuing.
infrun: clear_proceed_status_thread (process 14193)
infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT)
infrun: proceed: resuming process 14193
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 14193] at 0x4005c0
infrun: infrun_async(1)
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 14193.14193.0 [process 14193],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: handle_inferior_event status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: stop_pc = 0x4005c0
infrun: random signal (GDB_SIGNAL_TRAP)
infrun: stop_waiting
infrun: stop_all_threads
infrun: stop_all_threads, pass=0, iterations=0
infrun: process 14193 not executing
infrun: stop_all_threads, pass=1, iterations=1
infrun: process 14193 not executing
infrun: stop_all_threads done
Program received signal SIGTRAP, Trace/breakpoint trap.
problem_function () at brk_0.c:7
7 asm("brk %0\n\t" ::"n"(0x0));
infrun: infrun_async(0)
Otherwise GDB will keep trying to resume the inferior and will keep
seeing the SIGTRAP's, without stopping.
To the user it appears GDB has gone into an infinite loop, interruptible only
by Ctrl-C.
Also, windbg seems to use a different variation of AArch64 breakpoint compared
to GDB. This causes problems when debugging Windows on ARM binaries, when
program breakpoints are being used.
The proposed patch creates a new gdbarch method (gdbarch_program_breakpoint_here_p)
that tells GDB whether the underlying instruction is a breakpoint instruction
or not.
This is more general than only checking for the instruction GDB uses as
breakpoint.
The existing logic is still preserved for targets that do not implement this
new gdbarch method.
The end result is like so:
(gdb) x/i $pc
=> 0x4005c0 <problem_function>: brk #0x90f
(gdb) c
Continuing.
infrun: clear_proceed_status_thread (process 16417)
infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT)
infrun: proceed: resuming process 16417
infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 16417] at 0x4005c0
infrun: infrun_async(1)
infrun: prepare_to_wait
infrun: target_wait (-1.0.0, status) =
infrun: 16417.16417.0 [process 16417],
infrun: status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: handle_inferior_event status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: stop_pc = 0x4005c0
infrun: random signal (GDB_SIGNAL_TRAP)
infrun: stop_waiting
infrun: stop_all_threads
infrun: stop_all_threads, pass=0, iterations=0
infrun: process 16417 not executing
infrun: stop_all_threads, pass=1, iterations=1
infrun: process 16417 not executing
infrun: stop_all_threads done
Program received signal SIGTRAP, Trace/breakpoint trap.
problem_function () at brk.c:7
7 asm("brk %0\n\t" ::"n"(0x900 + 0xf));
infrun: infrun_async(0)
gdb/ChangeLog:
2020-01-29 Luis Machado <luis.machado@linaro.org>
* aarch64-tdep.c (BRK_INSN_MASK): Define to 0xffe0001f.
(BRK_INSN_MASK): Define to 0xd4200000.
(aarch64_program_breakpoint_here_p): New function.
(aarch64_gdbarch_init): Set gdbarch_program_breakpoint_here_p hook.
* arch-utils.c (default_program_breakpoint_here_p): Moved from
breakpoint.c.
* arch-utils.h (default_program_breakpoint_here_p): Moved from
breakpoint.h
* breakpoint.c (bp_loc_is_permanent): Changed return type to bool and
call gdbarch_program_breakpoint_here_p.
(program_breakpoint_here): Moved to arch-utils.c, renamed to
default_program_breakpoint_here_p, changed return type to bool and
simplified.
* breakpoint.h (program_breakpoint_here): Moved prototype to
arch-utils.h, renamed to default_program_breakpoint_here_p and changed
return type to bool.
* gdbarch.c: Regenerate.
* gdbarch.h: Regenerate.
* gdbarch.sh (program_breakpoint_here_p): New method.
* infrun.c (handle_signal_stop): Call
gdbarch_program_breakpoint_here_p.
There exist expected failures in the pass-by-ref.exp and
pass-by-ref-2.exp tests based on the GCC and Clang version.
* GCC version <= 6 and Clang do not emit DW_AT_deleted and
DW_AT_defaulted.
* Clang version >= 7 emits DW_AT_calling_convention, which helps the
debugger make the right calling convention decision in some cases
despite lacking the 'defaulted' and 'deleted' attributes.
Mark the related tests as XFAIL based on the compiler version.
Tested on X86_64 using GCC 5.5.0, 6.5.0, 7.4.0, 8.3.0, 9.2.1;
and Clang 5.0.1, 6.0.0, 7.0.0, 8.0.0, 9.0.1, 10.0.0.
gdb/testsuite/ChangeLog:
2020-01-29 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* gdb.cp/pass-by-ref-2.exp: Mark some tests as XFAIL based on the
GCC/Clang version.
* gdb.cp/pass-by-ref.exp: Ditto.
Change-Id: I1d8440aa438049f7c4da7f4f76f201c48550f1e4
I ran into:
...
Thread 3.1 "watchpoint-fork" hit Breakpoint 3, marker () at \
watchpoint-fork-mt.c:42^M
42 }^M
(gdb) parent2: 1945^M
FAIL: gdb.threads/watchpoint-fork.exp: child: multithreaded: breakpoint (A) \
after the second fork (timeout)
...
The problem is that the FAILing gdb_test expects '(gdb) ' to be the last thing
printed, but the inferior prints something after that.
A similar FAIL is described in the sources in watchpoint-fork-parent.c:
...
printf ("child%d: %d\n", nr, (int) getpid ());
/* Delay to get both the "child%d" and "parent%d" message printed
without a race breaking expect by its endless wait on `$gdb_prompt$':
Breakpoint 3, marker () at watchpoint-fork.c:33
33 }
(gdb) parent2: 14223 */
i = sleep (1);
...
I noticed that while the executables print output, the output is not verified in
the test-case, so it's merely debug output.
Fix this by:
- guarding the prints in the executables (as well as related
sleep and setbuf calls) with #if DEBUG, and
- compiling by default with DEBUG=0.
gdb/testsuite/ChangeLog:
2020-01-29 Tom de Vries <tdevries@suse.de>
* gdb.threads/watchpoint-fork-child.c: Guard prints with #if DEBUG.
* gdb.threads/watchpoint-fork-mt.c: Same.
* gdb.threads/watchpoint-fork-parent.c: Same.
* gdb.threads/watchpoint-fork-st.c: Same.
* gdb.threads/watchpoint-fork.exp: Compile with DEBUG=0.
Change-Id: I63efd4c7771f96b5f5cd87ef2ab36795484ae2be
The address was written as a long value, but long is always a 32bit value
on Windows, which lead to truncated addresses.
The solution was to use paddress instead.
gdb/gdbserver/ChangeLog:
2020-01-28 Hannes Domani <ssbssa@yahoo.de>
* server.c (handle_qxfer_libraries): Write segment-address with
paddress.
Maciej reported a problem found by his RISC-V gdbserver port.
warning: while parsing target description (at line 4): Target description specified unknown architecture "riscv:rv64id"
warning: Could not load XML target description; ignoring
We only have two arches defined, riscv:rv32 and riscv:rv64. Both bfd and
gdb are creating arch strings that have extension letters added to the base
architecture. The bfd_default_scan function requires an exact match, so
these strings fail to map to a bfd_arch. I think we should ignore the
extension letters in a RISC-V specific scan function.
bfd/
* cpu-riscv.c (riscv_scan): New.
(N): Change bfd_default_scan to riscv_scan.
Change-Id: I096476705e1da5cb8934c5005b1eed2a8989f7a7
New in v3:
- Verify if the syscall number matches what is expected for the target.
- Used gdb_assert for one more check.
New in v2:
- Set initial values to -1 instead of 0.
- Rewrote RE to prevent unexpected matching when parsing one character at a
time.
- Used gdb_assert for an additional check.
- Validated with check-read1
There are a couple problems with this test.
First
--
gdb.base/step-over-syscall.exp records the address of a syscall instruction
within fork/vfork/clone functions and also the address of the instruction
after that syscall instruction.
It uses these couples addresses to make sure we stepped over a syscall
instruction (fork/vfork/clone events) correctly.
The way the test fetches the addresses of the instructions is by stepi-ing
its way through the fork/vfork/clone functions until it finds a match for
a syscall. Then it stepi's once again to get the address of the next
instruction.
This assumes that stepi-ing over a syscall is working correctly and landing
in the right PC. This is not the case for AArch64/Linux, where we're
landing a couple instructions after the syscall in some cases.
The following patch lets the test execute as before, but adds a new instruction
address check using the x command as opposed to stepi.
I didn't want to change how the test works since we may also be
interested in checking if stepi-ing over the syscall under different
conditions (displaced stepping on/off) yields the same results. I don't
feel strongly about this, so i'm OK with changing how we compare PC's for
the entire test if folks decide it is reasonable.
Second
--
FAIL: gdb.base/step-over-syscall.exp: vfork: displaced=off: continue to vfork (3rd time) (the program exited)
FAIL: gdb.base/step-over-syscall.exp: vfork: displaced=off: continue to syscall insn vfork (the program is no longer running)
FAIL: gdb.base/step-over-syscall.exp: vfork: displaced=off: single step over vfork (the program is no longer running)
Depending on the glibc version we may have different code generated for the
fork/vfork/clone functions.
I ran into the situation where vfork for newer glibc's on AArch64/Linux is
very short, so "break vfork" will put a breakpoint right at the syscall
instruction, which is something the testcase isn't expecting (a off-by-1
of sorts).
The patch adds extra code to handle this case. If the test detects we're
already sitting at a syscall instruction, it records the address and moves
on to record the address after that particular instruction.
Another measure is to "break *$syscall" instead of "break $syscall". That
guarantees we're stopping at the first instruction of the syscall function,
if it ever happens that the syscall instruction is the first instruction of
those functions.
With these changes i can fix some failures for aarch64-linux-gnu and also
expose the problems i've reported here:
https://sourceware.org/ml/gdb-patches/2019-12/msg01071.html
These tests now fail for aarch64-linux-gnu (patch for this is going through
reviews):
FAIL: gdb.base/step-over-syscall.exp: vfork: displaced=off: pc after stepi matches insn addr after syscall
FAIL: gdb.base/step-over-syscall.exp: vfork: displaced=on: pc after stepi matches insn addr after syscall
gdb/testsuite/ChangeLog:
2020-01-27 Luis Machado <luis.machado@linaro.org>
* gdb.base/step-over-syscall.exp (setup): Check if we're already
sitting at a syscall instruction when we hit the syscall function's
breakpoint.
Check PC against one obtained with the x command.
Validate syscall number.
(step_over_syscall): Don't continue to the syscall instruction if
we're already there.