Age | Commit message (Collapse) | Author | Files | Lines |
|
This commit adds the signed register move (movs) instructions
introduced in the BPF ISA version 4, including opcodes and assembler
tests.
Tested in bpf-unknown-none.
include/ChangeLog:
2023-07-21 Jose E. Marchesi <jose.marchesi@oracle.com>
* opcode/bpf.h (BPF_OFFSET16_MOVS8): Define.
(BPF_OFFSET16_MOVS16): Likewise.
(BPF_OFFSET16_MOVS32): Likewise.
(enum bpf_insn_id): Add entries for MOVS{8,16,32}R and
MOVS32{8,16,32}R.
opcodes/ChangeLog:
2023-07-21 Jose E. Marchesi <jose.marchesi@oracle.com>
* bpf-opc.c (bpf_opcodes): Add entries for MOVS{8,16,32}R and
MOVS32{8,16,32}R instructions. and MOVS32I instructions.
gas/ChangeLog:
2023-07-21 Jose E. Marchesi <jose.marchesi@oracle.com>
* testsuite/gas/bpf/alu.s: Test movs instructions.
* testsuite/gas/bpf/alu-pseudoc.s: Likewise.
* testsuite/gas/bpf/alu32.s: Likewise for movs32 instruction.
* testsuite/gas/bpf/alu32-pseudoc.s: Likewise.
* testsuite/gas/bpf/alu.d: Add expected results.
* testsuite/gas/bpf/alu32.d: Likewise.
* testsuite/gas/bpf/alu-be.d: Likewise.
* testsuite/gas/bpf/alu32-be.d: Likewise.
* testsuite/gas/bpf/alu-pseudoc.d: Likewise.
* testsuite/gas/bpf/alu32-pseudoc.d: Likewise.
* testsuite/gas/bpf/alu-be-pseudoc.d: Likewise.
* testsuite/gas/bpf/alu32-be-pseudoc.d: Likewise.
|
|
In a review, Eli pointed out that @findex is redundant when used with
@defun. This patch removes all such uses from python.texi, plus a
couple uses before @defvar that are also unnecessary.
Approved-By: Eli Zaretskii <eliz@gnu.org>
|
|
I noticed that a doc string py-type.c says "an signed".
This patch corrects it to "a signed".
|
|
Ada 2022 adds the "target name symbol", which can be used on the right
hand side of an assignment to refer to the left hand side. This
allows for convenient updates. This patch implements this for gdb's
Ada expression parser.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
|
|
The DAP disassemble command lets the client return the underlying
bytes of the instruction in an implementation-defined format. This
patch updates gdb to return this, and simply uses a hex string of the
bytes as the format.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
|
|
I ran across this very old code in gdb's Ada support. After a bit of
archaeology, we couldn't determine what bug this might have been
working around. It is no longer needed, so this patch removes it.
As this is entirely Ada-specific and was reviewed and tested at
AdaCore, I'm checking it in.
|
|
This was breaking --enable-targets=all builds.
opcodes/ChangeLog:
2023-07-21 Jose E. Marchesi <jose.marchesi@oracle.com>
* Makefile.am (TARGET64_LIBOPCODES_CFILES): Add missing bpf-dis.c
* Makefile.in: Regenerate.
|
|
The BPF port in binutils has been rewritten (commit
d218e7fedc74d67837d2134120917f4ac877454c) in order to not be longer
based on CGEN. Please see that commit log for more information.
This patch updates the BPF simulator accordingly. The new
implementation is much simpler and it is based on the new BPF opcodes.
Tested with target bpf-unknown-none with both 64-bit little-endian
host and 32-bit little-endian host.
Note that I have not tested in a big-endian host yet. I will do so
once this lands upstream so I can use the GCC compiler farm.
|
|
CGEN is cool, but the BPF architecture is simply too bizarre for it.
The weird way of BPF to handle endianness in instruction encoding, the
weird C-like alternative assembly syntax, the weird abuse of
multi-byte (or infra-byte) instruction fields as opcodes, the unusual
presence of opcodes beyond the first 32-bits of some instructions, are
all examples of what makes it a PITA to continue using CGEN for this
port. The bpf.cpu file is becoming so complex and so nested with
p-macros that it is very difficult to read, and quite challenging to
update. Also, every time we are forced to change something in CGEN to
accommodate BPF requirements (which is often) we have to do extensive
testing to make sure we do not break any other target using CGEN.
This is getting un-maintenable.
So I have decided to bite the bullet and revamp/rewrite the port so it
no longer uses CGEN. Overall, this involved:
* To remove the cpu/bpf.{cpu,opc} descriptions.
* To remove the CGEN generated files.
* To replace the CGEN generated opcodes table with a new hand-written
opcodes table for BPF.
* To replace the CGEN generated disassembler wih a new disassembler
that uses the new opcodes.
* To replace the CGEN generated assembler with a new assembler that uses the
new opcodes.
* To replace the CGEN generated simulator with a new simulator that uses the
new opcodes. [This is pushed in GDB in another patch.]
* To adapt the build systems to the new situation.
Additionally, this patch introduces some extensions and improvements:
* A new BPF relocation BPF_RELOC_BPF_DISP16 plus corresponding ELF
relocation R_BPF_GNU_64_16 are added to the BPF BFD port. These
relocations are used for section-relative 16-bit offsets used in
load/store instructions.
* The disassembler now has support for the "pseudo-c" assembly syntax of
BPF. What dialect to use when disassembling is controlled by a command
line option.
* The disassembler now has support for dumping instruction immediates in
either octal, hexadecimal or decimal. The used output base is controlled
by a new command-line option.
* The GAS BPF test suite has been re-structured and expanded in order to
test the disassembler pseudoc syntax support. Minor bugs have been also
fixed there. The assembler generic tests that were disabled for bpf-*-*
targets due to the previous implementation of pseudoc syntax are now
re-enabled. Additional tests have been added to test the new features of
the assembler. .dump files are no longer used.
* The linker BPF test suite has been adapted to the command line options
used by the new disassembler.
The result is very satisfactory. This patchs adds 3448 lines of code
and removes 10542 lines of code.
Tested in:
* Target bpf-unknown-none with 64-bit little-endian host and 32-bit
little-endian host.
* Target x86-64-linux-gnu with --enable-targets=all
Note that I have not tested in a big-endian host yet. I will do so
once this lands upstream so I can use the GCC compiler farm.
I have not included ChangeLog entries in this patch: these would be
massive and not very useful, considering this is pretty much a rewrite
of the port. I beg the indulgence of the global maintainers.
|
|
ROCm programs can load a high number of compute kernels on GPU devices,
especially if lazy code-object loading have been disabled. Each code
object containing such program is loaded once for each device available,
and each instance is reported by GDB as an individual shared library.
We came across situations where the number of shared libraries opened by
GDB gets higher than the allowed number of opened files for the process.
Increasing the opened files limit works around the problem, but there is a
better way this patch proposes to follow.
Under the hood, the GPU code objects are embedded inside the host
application binary and shared library binaries. GDB currently opens the
underlying file once for each shared library it sees. That means that
the same file is re-opened every time a code object is loaded on a GPU.
This patch proposes to only open each underlying file once. This is
done by implementing a reference counting mechanism so the underlying
file is opened when the underlying file first needs to be opened, and
closed when the last BFD using the underlying file is closed.
On a program where GDB used to open about 1500 files to load all shared
libraries, this patch makes it so only 54 opened file descriptors are
needed.
I have tested this patch on downstream ROCgdb's full testsuite and
upstream GDB testsuite with no regression.
Approved-By: Pedro Alves <pedro@palves.net>
|
|
Bring disassembly back in line with what the assembler accepts, thus
also making it self-consistent (with, in particular selector load/store
insns). While there further add D to all affected insns except ARPL
(where S is used, matching LAR/LSL), to also behave correctly in suffix-
always mode.
While there also hook up the Intel variant of the LKGS test.
|
|
For whatever reason in c9f5b96bdab0 ("x86: correct handling of LAR and
LSL") I didn't realize that we can easily use Sv instead of going
through mod_table[]. Redo this aspect of that change.
|
|
Consider the test-case:
...
$ cat main.c
int main (void) { return 0; }
$ cat static-optimized-out.c
static int aaa;
...
compiled like this:
...
$ gcc-12 static-optimized-out.c main.c -g -O2 -flto
...
There's a difference in behaviour depending on symtab expansion state:
...
$ gdb -q -batch a.out -ex "print aaa"
No symbol "aaa" in current context.
$ gdb -q -batch a.out -ex "maint expand-symtab" -ex "print aaa"
$1 = <optimized out>
...
The reason for the difference is that the optimized out variable aaa:
...
<1><104>: Abbrev Number: 2 (DW_TAG_variable)
<105> DW_AT_name : aaa
<109> DW_AT_decl_file : 1
<10a> DW_AT_decl_line : 18
<10b> DW_AT_decl_column : 12
<10c> DW_AT_type : <0x110>
...
is not added to the cooked index because of this clause in abbrev_table::read:
...
else if (!has_location && !has_specification_or_origin && !has_external
&& cur_abbrev->tag == DW_TAG_variable)
cur_abbrev->interesting = false;
...
Fix this inconsistency by making sure that the optimized out variable is added
to the cooked index.
Regression tested on x86_64-linux.
Add two test-cases, a C test-case gdb.opt/static-optimized-out.exp and a dwarf
assembly test-case gdb.dwarf2/static-optimized-out.exp.
Tested gdb.opt/static-optimized-out.exp with gcc-8 to gcc-12, for which we now
consistently get:
...
(gdb) print aaa^M
$1 = <optimized out>^M
...
and with gcc 7.5.0 and clang 13.0.1, for which we still consistently get:
...
(gdb) print aaa^M
No symbol "aaa" in current context.^M
...
due to missing debug info for the variable.
PR symtab/30656
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30656
Approved-By: Tom Tromey <tom@tromey.com>
|
|
In test-case gdb.tui/long-prompt.exp, with a prompt of 40 chars, the same size
as the terminal width, we get a superfluous newline at line 19:
...
16 (gdb) set prompt 123456789A123456789B123
17 456789C123456789>
18 123456789A123456789B123456789C123456789>
19
20 123456789A123456789B123456789C123456789>
21 set prompt (gdb)
22 (gdb)
...
as well as a superfluous repetition of the prompt at line 20 once we type the
's' starting "set prompt".
I traced the superfluous newline back to readline's readline_internal_setup,
that does:
...
/* If we're not echoing, we still want to at least print a prompt, because
rl_redisplay will not do it for us. If the calling application has a
custom redisplay function, though, let that function handle it. */
if (_rl_echoing_p == 0 && rl_redisplay_function == rl_redisplay)
...
else
{
if (rl_prompt && rl_already_prompted)
rl_on_new_line_with_prompt ();
else
rl_on_new_line ();
(*rl_redisplay_function) ();
...
and then we hit the case that calls rl_on_new_line_with_prompt, which does:
...
/* If the prompt length is a multiple of real_screenwidth, we don't know
whether the cursor is at the end of the last line, or already at the
beginning of the next line. Output a newline just to be safe. */
if (l > 0 && (l % real_screenwidth) == 0)
_rl_output_some_chars ("\n", 1);
...
This doesn't look like a readline bug, because the behaviour matches the
comment.
[ And the fact that the output of the newline doesn't happen in the scope of
tui_redisplay_readline means it doesn't get the prompt wrap detection
treatment, causing start_line to be incorrect, which causes the superfluous
repetition of the prompt. ]
I looked at ways to work around this, and managed by switching off
rl_already_prompted, which we set to 1 in tui_rl_startup_hook:
...
/* Readline hook to redisplay ourself the gdb prompt.
In the SingleKey mode, the prompt is not printed so that
the command window is cleaner. It will be displayed if
we temporarily leave the SingleKey mode. */
static int
tui_rl_startup_hook (void)
{
rl_already_prompted = 1;
if (tui_current_key_mode != TUI_COMMAND_MODE
&& !gdb_in_secondary_prompt_p (current_ui))
tui_set_key_mode (TUI_SINGLE_KEY_MODE);
tui_redisplay_readline ();
return 0;
}
...
Then I started looking at why rl_already_prompted is set to 1.
The use case for rl_already_prompted seems to be:
- app (application, the readline user) outputs prompt,
- app sets rl_already_prompted to 1, and
- app calls readline, which calls rl_on_new_line_with_prompt, which figures
out how long the prompt is, and sets a few readline variables accordingly,
which can be used in the following call to rl_redisplay_function.
AFAICT, TUI does not fit this pattern. It does not output an initial prompt,
rather it writes the prompt in every rl_redisplay_function. It doesn't use
the variables set by rl_on_new_line_with_prompt, instead it figures stuff out
by itself.
Fix this by removing the rl_already_prompted setting.
Also remove the call to tui_redisplay_readline, it's not necessary, the
function is called anyway.
Tested on x86_64-linux, no regressions.
|
|
|
|
* elfxx-mips.c (_bfd_mips_elf_symbol_processing): Don't treat
__gnu_lto_slim as SHN_MIPS_SCOMMON.
(_bfd_mips_elf_add_symbol_hook): Likewise.
|
|
|
|
In the current code, when a register is fetched, the entire regset
are fetched via ptrace, but only this register status is updated in
regcache, it needs to fetch the same regset through ptrace again if
another register in this regset is fetched later, this is obviously
unnecessary. It is proper to update the status of the entire regset
in regcache when fetching a register via ptrace.
Signed-off-by: Hui Li <lihui@loongson.cn>
Reviewed-By: Tom Tromey <tom@tromey.com>
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
|
|
Andrew reported that the previous change to gdb.Inferior.read_memory &
friends introducing scoped_restore_current_inferior_for_memory broke
gdb.dap/stop-at-main.exp. This is also reported as PR dap/30644.
The root of the problem is that all the methods that now use
scoped_restore_current_inferior_for_memory cause GDB to crash with a
failed assert if they are run on an inferior that is not yet started.
E.g.:
(gdb) python i = gdb.selected_inferior ()
(gdb) python i.read_memory (4,4)
gdb/thread.c:626: internal-error: any_thread_of_inferior: Assertion `inf->pid != 0' failed.
This patch fixes the problem by removing
scoped_restore_current_inferior_for_memory's ctor ptid parameter and
the any_thread_of_inferior calls completely, and making
scoped_restore_current_inferior_for_memory switch inferior_ptid to a
pid ptid.
I was a little worried that some port might be assuming inferior_ptid
points at a thread in the xfer_partial memory access routines. We
know that anything that supports forks must not assume that, due to
how detach_breakpoints works. I looked at a number of xfer_partial
implementations, and didn't see anything that is looking at
inferior_ptid in a way that would misbehave. I'm thinking that we
could go forward with this and just fix ports if they break.
While on some ports like on AMD GPU we have thread-specific address
spaces, and so when accessing memory for those address spaces, we must
have the right thread context (via inferior_ptid) selected, in
Inferior.read_memory, we only have the inferior to work with, so this
API as is can't be used to access thread-specific address spaces.
IOW, it can only be used to access the global address space that is
visible to both the CPU and the GPUs.
In proc-service.c:ps_xfer_memory, the other spot using
scoped_restore_current_inferior_for_memory, we're always accessing
per-inferior memory.
If we end up using scoped_restore_current_inferior_for_memory later to
set up the context to read memory from a specific thread, then we can
add an alternative ctor that takes a thread_info pointer, and make
inferior_ptid point to the thread, for example.
New test added to gdb.python/py-inferior.exp, exercising
Inferior.read_memory without execution.
No regressions on native and extended-gdbserver x86_64 GNU/Linux.
Reviewed-By: Tom Tromey <tom@tromey.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30644
Change-Id: I11309c5ddbbb51a4594cf63c21b3858bfd9aed19
|
|
|
|
Current implementation of amd_dbgapi_target::detach (inferior *, int)
does the following:
remove_breakpoints_inf (current_inferior ());
detach_amd_dbgapi (inf);
beneath ()->detach (inf, from_tty);
I find that using a mix of `current_inferior ()` and `inf` disturbing.
At this point, we know that both are the same (target_detach does assert
that `inf == current_inferior ()` before calling target_ops::detach).
To improve consistency, this patch replaces `current_inferior ()` with
`inf` in amd_dbgapi_target::detach.
Change-Id: I01b7ba2e661c25839438354b509d7abbddb7c5ed
Approved-By: Pedro Alves <pedro@palves.net>
|
|
With a gdb build with -O2 -flto=auto using gcc 7.5.0, I run into:
...
(gdb) ptype global_c^M
^M
Thread 1 "xgdb" hit Breakpoint 3, \
_Z12c_print_typeP4typePKcP7ui_fileii8languagePK18type_print_options () at \
gdb/c-typeprint.c:175^M
175 {^M
(outer-gdb) FAIL: gdb.gdb/python-helper.exp: hit breakpoint in outer gdb again
...
This is a problem with the debug info, which marks the CU containing the
function declaration as C rather than C++. This is fixed in gcc 8 and later.
Work around this compiler problem by allowing the mangled name.
Tested on x86_64-linux.
PR testsuite/30648
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30648
|
|
It is possible to build huge binaries on powerpc64, where 32-bit
addresses in debug info are insufficient to descibe locations in the
binary. Help out the user, and only warn about debug overflows.
* powerpc.cc (Target_powerpc::Relocate::relocate): Warn on
relocation overflows in debug info.
|
|
Separate out some of the defines from the block handling windows
support, so they don't get lost. Delete an unused variable.
|
|
Only the xcoff and pe extensions were enabled. Build the lot, and fix
some more printf format problems when the host is 32-bit.
* configure.ac (od_vectors): Set up for --enable-targets=all.
* configure: Regenerate.
* od-elf32_avr.c (elf32_avr_dump_mem_usage): Correct format
specifier vs. arg mismatch.
(elf32_avr_dump_avr_prop): Likewise.
|
|
* config/tc-d10v.c (find_opcode): Correct format specifier vs.
arg mismatch.
* config/tc-m68hc11.c (fixup8, fixup16, fixup24, fixup8_xg): Likewise.
* config/tc-vax.c (md_assemble): Likewise.
* config/tc-xtensa.c (dump_litpools): Likewise.
* config/tc-z80.c (emit_data_val, emit_byte): Likewise.
|
|
|
|
|
|
|
|
This patch support Zcb extension, contains new compressed instructions,
some instructions depend on other existed extension, like 'zba', 'zbb'
and 'zmmul'. Zcb also imply Zca extension to enable the compressing
features.
Co-Authored by: Charlie Keaney <charlie.keaney@embecosm.com>
Co-Authored by: Mary Bennett <mary.bennett@embecosm.com>
Co-Authored by: Nandni Jamnadas <nandni.jamnadas@embecosm.com>
Co-Authored by: Sinan Lin <sinan.lin@linux.alibaba.com>
Co-Authored by: Simon Cook <simon.cook@embecosm.com>
Co-Authored by: Shihua Liao <shihua@iscas.ac.cn>
Co-Authored by: Yulong Shi <yulong@iscas.ac.cn>
bfd/ChangeLog:
* elfxx-riscv.c (riscv_multi_subset_supports): New extension.
(riscv_multi_subset_supports_ext): Ditto.
gas/ChangeLog:
* config/tc-riscv.c (validate_riscv_insn): New operators.
(riscv_ip): Ditto.
* testsuite/gas/riscv/zcb.d: New test.
* testsuite/gas/riscv/zcb.s: New test.
include/ChangeLog:
* opcode/riscv-opc.h (MATCH_C_LBU): New opcode.
(MASK_C_LBU): New mask.
(MATCH_C_LHU): New opcode.
(MASK_C_LHU): New mask.
(MATCH_C_LH): New opcode.
(MASK_C_LH): New mask.
(MATCH_C_SB): New opcode.
(MASK_C_SB): New mask.
(MATCH_C_SH): New opcode.
(MASK_C_SH): New mask.
(MATCH_C_ZEXT_B): New opcode.
(MASK_C_ZEXT_B): New mask.
(MATCH_C_SEXT_B): New opcode.
(MASK_C_SEXT_B): New mask.
(MATCH_C_ZEXT_H): New opcode.
(MASK_C_ZEXT_H): New mask.
(MATCH_C_SEXT_H): New opcode.
(MASK_C_SEXT_H): New mask.
(MATCH_C_ZEXT_W): New opcode.
(MASK_C_ZEXT_W): New mask.
(MATCH_C_NOT): New opcode.
(MASK_C_NOT): New mask.
(MATCH_C_MUL): New opcode.
(MASK_C_MUL): New mask.
(DECLARE_INSN): New opcode.
* opcode/riscv.h (EXTRACT_ZCB_BYTE_UIMM): New inline func.
(EXTRACT_ZCB_HALFWORD_UIMM): Ditto.
(ENCODE_ZCB_BYTE_UIMM): Ditto.
(ENCODE_ZCB_HALFWORD_UIMM): Ditto.
(VALID_ZCB_BYTE_UIMM): Ditto.
(VALID_ZCB_HALFWORD_UIMM): Ditto.
(enum riscv_insn_class): New extension class.
opcodes/ChangeLog:
* riscv-dis.c (print_insn_args): New operators.
* riscv-opc.c: New instructions.
|
|
This patch add Zca/f/d extensions support, since all ZC*
extensions will imply Zca extension, just enabled compress
feature when Zca extension is available.
Co-Authored by: Charlie Keaney <charlie.keaney@embecosm.com>
Co-Authored by: Mary Bennett <mary.bennett@embecosm.com>
Co-Authored by: Nandni Jamnadas <nandni.jamnadas@embecosm.com>
Co-Authored by: Sinan Lin <sinan.lin@linux.alibaba.com>
Co-Authored by: Simon Cook <simon.cook@embecosm.com>
Co-Authored by: Shihua Liao <shihua@iscas.ac.cn>
Co-Authored by: Yulong Shi <yulong@iscas.ac.cn>
bfd/ChangeLog:
* elfxx-riscv.c (riscv_multi_subset_supports): New extensions.
(riscv_multi_subset_supports_ext): Ditto.
gas/ChangeLog:
* config/tc-riscv.c (riscv_set_arch): Extend compress check.
* testsuite/gas/riscv/zca.d: New test.
* testsuite/gas/riscv/zca.s: New test.
* testsuite/gas/riscv/zcd.d: New test.
* testsuite/gas/riscv/zcd.s: New test.
* testsuite/gas/riscv/zcf.d: New test.
* testsuite/gas/riscv/zcf.s: New test.
|
|
|
|
child_terminal_init_with_pgrp is declared but not defined. This patch
removes the declaration. Tested by grep and rebuilding.
|
|
this requires also changes in ldgram.y and ldexp.c, unlike
accepting '^' only. But let's do this anyway, if only for
symmetry.
|
|
While working on some of the recent patches relating to vfork handling
I found this debug output very helpful, I'd like to propose adding
this into GDB.
With debug turned off there should be no user visible changes after
this commit.
|
|
While working on gdb.base/foll-vfork.exp I noticed that there are
several random 'sleep' calls throughout the test.
The comment suggests these are to allow for output from a vforked
child to arrive, but in each case the test is about to close and
restart GDB, so I don't see how random output from a child process
could impact testing.
I removed the sleep calls and couldn't reproduce any failures from
this test, I left the test running for a couple of hours, and tried
loading my machine, and the test seems fine with these removed.
I've left this as a separate commit so that if, in the future, someone
can show that these are required, it will be easy to revert this one
patch and bring them back.
There should be no change in what is tested after this commit.
|
|
This commit provides tests for all of the bugs fixed in the previous
four commits, this is achieved by expanding gdb.base/foll-vfork.exp to
run with different configurations:
* target-non-stop on/off
* non-stop on/off
* schedule-multiple on/off
We don't test with schedule-multiple on if we are using a remote
target, this is due to bug gdb/30574. I've added a link to that bug
in this commit, but all this commit does is expose that bug, there's
no fixes here.
Some of the bugs fixed in the previous commits are very timing
dependent, as such, they don't always show up. I've had more success
when running this test on a very loaded machine -- I usually run ~8
copies of the test in parallel, then the bugs would normally show up
pretty quickly.
Other than running the test in more configurations, I've not made any
changes to what is actually being tested, other than in one place
where, when testing with non-stop mode, GDB stops in a different
inferior, as such I had to add a new 'inferior 2' call, this can be
found in vfork_relations_in_info_inferiors.
I have cleaned things up a little, for example, making use of
proc_with_prefix to remove some with_test_prefix calls.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30574
|
|
Like the last few commit, this fixes yet another vfork related issue.
Like the commit titled:
gdb: don't restart vfork parent while waiting for child to finish
which addressed a case in linux-nat where we would try to resume a
vfork parent, this commit addresses a very similar case, but this time
occurring in infrun.c. Just like with that previous commit, this bug
results in the assert:
x86-linux-dregs.c:146: internal-error: x86_linux_update_debug_registers: Assertion `lwp_is_stopped (lwp)' failed.
In this case the issue occurs when target-non-stop is on, but non-stop
is off, and again, schedule-multiple is on. As with the previous
commit, GDB is in follow-fork-mode child. If you have not done so, it
is worth reading the earlier commit as many of the problems leading to
the failure are the same, they just appear in a different part of GDB.
Here are the steps leading to the assertion failure:
1. The user performs a 'next' over a vfork, GDB stop in the vfork
child,
2. As we are planning to follow the child GDB sets the vfork_parent
and vfork_child member variables in the two inferiors, the
thread_waiting_for_vfork_done member is left as nullptr, that member
is only used when GDB is planning to follow the parent inferior,
3. The user does 'continue', our expectation is that the vfork child
should resume, and once that process has exited or execd, GDB should
detach from the vfork parent. As a result of the 'continue' GDB
eventually enters the proceed function,
4. In proceed we selected a ptid_t to resume, because
schedule-multiple is on we select minus_one_ptid (see
user_visible_resume_ptid),
5. As GDB is running in all-stop on top of non-stop mode, in the
proceed function we iterate over all threads that match the resume
ptid, which turns out to be all threads, and call
proceed_resume_thread_checked. One of the threads we iterate over
is the vfork parent thread,
6. As the thread passed to proceed_resume_thread_checked doesn't
match any of the early return conditions, GDB will set the thread
resumed,
7. As we are resuming one thread at a time, this thread is seen by
the lower layers (e.g. linux-nat) as the "event thread", which means
we don't apply any of the checks, e.g. is this thread a
vfork parent, instead we assume that GDB core knows what it's doing,
and linux-nat will resume the thread, we have now incorrectly set
running the vfork parent thread when this thread should be waiting
for the vfork child to complete,
8. Back in the proceed function GDB continues to iterate over all
threads, and now (correctly) resumes the vfork child thread,
8. As the vfork child is still alive the kernel holds the vfork
parent stopped,
9. Eventually the child performs its exec and GDB is sent and EXECD
event. However, because the parent is resumed, as soon as the child
performs its exec the vfork parent also sends a VFORK_DONE event to
GDB,
10. Depending on timing both of these events might seem to arrive in
GDB at the same time. Normally GDB expects to see the EXECD or
EXITED/SIGNALED event from the vfork child before getting the
VFORK_DONE in the parent. We know this because it is as a result of
the EXECD/EXITED/SIGNALED that GDB detaches from the parent (see
handle_vfork_child_exec_or_exit for details). Further the comment
in target/waitstatus.h on TARGET_WAITKIND_VFORK_DONE indicates that
when we remain attached to the child (not the parent) we should not
expect to see a VFORK_DONE,
11. If both events arrive at the same time then GDB will randomly
choose one event to handle first, in some cases this will be the
VFORK_DONE. As described above, upon seeing a VFORK_DONE GDB
expects that (a) the vfork child has finished, however, in this case
this is not completely true, the child has finished, but GDB has not
processed the event associated with the completion yet, and (b) upon
seeing a VFORK_DONE GDB assumes we are remaining attached to the
parent, and so resumes the parent process,
12. GDB now handles the EXECD event. In our case we are detaching
from the parent, so GDB calls target_detach (see
handle_vfork_child_exec_or_exit),
13. While this has been going on the vfork parent is executing, and
might even exit,
14. In linux_nat_target::detach the first thing we do is stop all
threads in the process we're detaching from, the result of the stop
request will be cached on the lwp_info object,
15. In our case the vfork parent has exited though, so when GDB
waits for the thread, instead of a stop due to signal, we instead
get a thread exited status,
16. Later in the detach process we try to resume the threads just
prior to making the ptrace call to actually detach (see
detach_one_lwp), as part of the process to resume a thread we try to
touch some registers within the thread, and before doing this GDB
asserts that the thread is stopped,
17. An exited thread is not classified as stopped, and so the assert
triggers!
Just like with the earlier commit, the fix is to spot the vfork parent
status of the thread, and not resume such threads. Where the earlier
commit fixed this in linux-nat, in this case I think the fix should
live in infrun.c, in proceed_resume_thread_checked. This function
already has a similar check to not resume the vfork parent in the case
where we are planning to follow the vfork parent, I propose adding a
similar case that checks for the vfork parent when we plan to follow
the vfork child.
This new check will mean that at step #6 above GDB doesn't try to
resume the vfork parent thread, which prevents the VFORK_DONE from
ever arriving. Once GDB sees the EXECD/EXITED/SIGNALLED event from
the vfork child GDB will detach from the parent.
There's no test included in this commit. In a subsequent commit I
will expand gdb.base/foll-vfork.exp which is when this bug would be
exposed.
If you do want to reproduce this failure then you will for certainly
need to run the gdb.base/foll-vfork.exp test in a loop as the failures
are all very timing sensitive. I've found that running multiple
copies in parallel makes the failure more likely to appear, I usually
run ~6 copies in parallel and expect to see a failure after within
10mins.
|
|
Split the thread resuming code from proceed into new function
proceed_resume_thread_checked.
Co-Authored-By: Christina Schimpe <christina.schimpe@intel.com>
|
|
This commit fixes a bug introduced by this commit:
commit d8bbae6ea080249c05ca90b1f8640fde48a18301
Date: Fri Jan 14 15:40:59 2022 -0500
gdb: fix handling of vfork by multi-threaded program (follow-fork-mode=parent, detach-on-fork=on)
The problem can be seen in this GDB session:
$ gdb -q
(gdb) set non-stop on
(gdb) file ./gdb/testsuite/outputs/gdb.base/foll-vfork/foll-vfork
Reading symbols from ./gdb/testsuite/outputs/gdb.base/foll-vfork/foll-vfork...
(gdb) tcatch vfork
Catchpoint 1 (vfork)
(gdb) run
Starting program: /tmp/gdb/testsuite/outputs/gdb.base/foll-vfork/foll-vfork
Temporary catchpoint 1 (vforked process 1375914), 0x00007ffff7d5043c in vfork () from /lib64/libc.so.6
(gdb) bt
#0 0x00007ffff7d5043c in vfork () from /lib64/libc.so.6
#1 0x00000000004011af in main (argc=1, argv=0x7fffffffad88) at .../gdb/testsuite/gdb.base/foll-vfork.c:32
(gdb) finish
Run till exit from #0 0x00007ffff7d5043c in vfork () from /lib64/libc.so.6
[Detaching after vfork from child process 1375914]
No unwaited-for children left.
(gdb)
Notice the "No unwaited-for children left." error. This is incorrect,
given where we are stopped there's no reason why we shouldn't be able
to use "finish" to return to the main frame.
When the inferior is stopped as a result of the 'tcatch vfork', the
inferior is in the process of performing the vfork, that is, GDB has
seen the VFORKED event, but has not yet attached to the new child
process, nor has the child process been resumed.
However, GDB has seen the VFORKED, and, as we are going to follow the
parent process, the inferior for the vfork parent will have its
thread_waiting_for_vfork_done member variable set, this will point to
the one and only thread that makes up the vfork parent process.
When the "finish" command is used GDB eventually ends up in the
proceed function (in infrun.c), in here we pass through all the
function until we eventually encounter this 'else if' condition:
else if (!cur_thr->resumed ()
&& !thread_is_in_step_over_chain (cur_thr)
/* In non-stop, forbid resuming a thread if some other thread of
that inferior is waiting for a vfork-done event (this means
breakpoints are out for this inferior). */
&& !(non_stop
&& cur_thr->inf->thread_waiting_for_vfork_done != nullptr))
{
The first two of these conditions will both be true, the thread is not
already resumed, and is not in the step-over chain, however, the third
condition, this one:
&& !(non_stop
&& cur_thr->inf->thread_waiting_for_vfork_done != nullptr))
is false, and this prevents the thread we are trying to finish from
being resumed. This condition is false because (a) non_stop is true,
and (b) cur_thr->inf->thread_waiting_for_vfork_done is not
nullptr (see above for why).
Now, if we check the comment embedded within the condition it says:
/* In non-stop, forbid resuming a thread if some other thread of
that inferior is waiting for a vfork-done event (this means
breakpoints are out for this inferior). */
And this makes sense, if we have a vfork parent with two thread, and
one thread has performed a vfork, then we shouldn't try to resume the
second thread.
However, if we are trying to resume the thread that actually performed
a vfork, then this is fine. If we never resume the vfork parent then
we'll never get a VFORK_DONE event, and so the vfork will never
complete.
Thus, the condition should actually be:
&& !(non_stop
&& cur_thr->inf->thread_waiting_for_vfork_done != nullptr
&& cur_thr->inf->thread_waiting_for_vfork_done != cur_thr))
This extra check will allow the vfork parent thread to resume, but
prevent any other thread in the vfork parent process from resuming.
This is the same condition that already exists in the all-stop on a
non-stop-target block earlier in the proceed function.
My actual fix is slightly different to the above, first, I've chosen
to use a nested 'if' check instead of extending the original 'else if'
check, this makes it easier to write a longer comment explaining
what's going on, and second, instead of checking 'non_stop' I've
switched to checking 'target_is_non_stop_p'. In this context this is
effectively the same thing, a previous 'else if' block in proceed
already handles '!non_stop && target_is_non_stop_p ()', so by the time
we get here, if 'target_is_non_stop_p ()' then we must be running in
non_stop mode.
Both of these tweaks will make the next patch easier, which is a
refactor to merge two parts of the proceed function, so this nested
'if' block is not going to exist for long.
For testing, there is no test included with this commit. The test was
exposed when using a modified version of the gdb.base/foll-vfork.exp
test script, however, there are other bugs that are exposed when using
the modified test script. These bugs will be addressed in subsequent
commits, and then I'll add the updated gdb.base/foll-vfork.exp.
If you wish to reproduce this failure then grab the updates to
gdb.base/foll-vfork.exp from the later commit and run this test, the
failure is always reproducible.
|
|
While working on a later patch, which changes gdb.base/foll-vfork.exp,
I noticed that sometimes I would hit this assert:
x86_linux_update_debug_registers: Assertion `lwp_is_stopped (lwp)' failed.
I eventually tracked it down to a combination of schedule-multiple
mode being on, target-non-stop being off, follow-fork-mode being set
to child, and some bad timing. The failing case is pretty simple, a
single threaded application performs a vfork, the child process then
execs some other application while the parent process (once the vfork
child has completed its exec) just exits. As best I understand
things, here's what happens when things go wrong:
1. The parent process performs a vfork, GDB sees the VFORKED event
and creates an inferior and thread for the vfork child,
2. GDB resumes the vfork child process. As schedule-multiple is on
and target-non-stop is off, this is translated into a request to
start all processes (see user_visible_resume_ptid),
3. In the linux-nat layer we spot that one of the threads we are
about to start is a vfork parent, and so don't start that
thread (see resume_lwp), the vfork child thread is resumed,
4. GDB waits for the next event, eventually entering
linux_nat_target::wait, which in turn calls linux_nat_wait_1,
5. In linux_nat_wait_1 we eventually call
resume_stopped_resumed_lwps, this should restart threads that have
stopped but don't actually have anything interesting to report.
6. Unfortunately, resume_stopped_resumed_lwps doesn't check for
vfork parents like resume_lwp does, so at this point the vfork
parent is resumed. This feels like the start of the bug, and this
is where I'm proposing to fix things, but, resuming the vfork parent
isn't the worst thing in the world because....
7. As the vfork child is still alive the kernel holds the vfork
parent stopped,
8. Eventually the child performs its exec and GDB is sent and EXECD
event. However, because the parent is resumed, as soon as the child
performs its exec the vfork parent also sends a VFORK_DONE event to
GDB,
9. Depending on timing both of these events might seem to arrive in
GDB at the same time. Normally GDB expects to see the EXECD or
EXITED/SIGNALED event from the vfork child before getting the
VFORK_DONE in the parent. We know this because it is as a result of
the EXECD/EXITED/SIGNALED that GDB detaches from the parent (see
handle_vfork_child_exec_or_exit for details). Further the comment
in target/waitstatus.h on TARGET_WAITKIND_VFORK_DONE indicates that
when we remain attached to the child (not the parent) we should not
expect to see a VFORK_DONE,
10. If both events arrive at the same time then GDB will randomly
choose one event to handle first, in some cases this will be the
VFORK_DONE. As described above, upon seeing a VFORK_DONE GDB
expects that (a) the vfork child has finished, however, in this case
this is not completely true, the child has finished, but GDB has not
processed the event associated with the completion yet, and (b) upon
seeing a VFORK_DONE GDB assumes we are remaining attached to the
parent, and so resumes the parent process,
11. GDB now handles the EXECD event. In our case we are detaching
from the parent, so GDB calls target_detach (see
handle_vfork_child_exec_or_exit),
12. While this has been going on the vfork parent is executing, and
might even exit,
13. In linux_nat_target::detach the first thing we do is stop all
threads in the process we're detaching from, the result of the stop
request will be cached on the lwp_info object,
14. In our case the vfork parent has exited though, so when GDB
waits for the thread, instead of a stop due to signal, we instead
get a thread exited status,
15. Later in the detach process we try to resume the threads just
prior to making the ptrace call to actually detach (see
detach_one_lwp), as part of the process to resume a thread we try to
touch some registers within the thread, and before doing this GDB
asserts that the thread is stopped,
16. An exited thread is not classified as stopped, and so the assert
triggers!
So there's two bugs I see here. The first, and most critical one here
is in step #6. I think that resume_stopped_resumed_lwps should not
resume a vfork parent, just like resume_lwp doesn't resume a vfork
parent.
With this change in place the vfork parent will remain stopped in step
instead GDB will only see the EXECD/EXITED/SIGNALLED event. The
problems in #9 and #10 are therefore skipped and we arrive at #11,
handling the EXECD event. As the parent is still stopped #12 doesn't
apply, and in #13 when we try to stop the process we will see that it
is already stopped, there's no risk of the vfork parent exiting before
we get to this point. And finally, in #15 we are safe to poke the
process registers because it will not have exited by this point.
However, I did mention two bugs.
The second bug I've not yet managed to actually trigger, but I'm
convinced it must exist: if we forget vforks for a moment, in step #13
above, when linux_nat_target::detach is called, we first try to stop
all threads in the process GDB is detaching from. If we imagine a
multi-threaded inferior with many threads, and GDB running in non-stop
mode, then, if the user tries to detach there is a chance that thread
could exit just as linux_nat_target::detach is entered, in which case
we should be able to trigger the same assert.
But, like I said, I've not (yet) managed to trigger this second bug,
and even if I could, the fix would not belong in this commit, so I'm
pointing this out just for completeness.
There's no test included in this commit. In a couple of commits time
I will expand gdb.base/foll-vfork.exp which is when this bug would be
exposed. Unfortunately there are at least two other bugs (separate
from the ones discussed above) that need fixing first, these will be
fixed in the next commits before the gdb.base/foll-vfork.exp test is
expanded.
If you do want to reproduce this failure then you will for certainly
need to run the gdb.base/foll-vfork.exp test in a loop as the failures
are all very timing sensitive. I've found that running multiple
copies in parallel makes the failure more likely to appear, I usually
run ~6 copies in parallel and expect to see a failure after within
10mins.
|
|
For *reasons* I was looking at gdb.base/foll-vfork.exp. This test
script has a proc 'setup_gdb' that could (potentially) fail. The
setup_gdb proc is called from many places and I, initially, didn't
think that we were checking if setup_gdb had failed or not.
My confusion was because I didn't understand what this tcl construct
did:
return -code return
this will actually act as a return in the context of a proc's caller,
effectively returning two levels of the call stack. Neat (I guess).
So it turns out my worries were misplaced, everywhere setup_gdb is
called, if setup_gdb fails then we will (magically) return.
However, I did spot one place where we managed to confuse ourselves
with our cleverness.
In check_vfork_catchpoints, this proc is called to check that GDB is
able to catch vforks or not. This proc is called early in the test
script, and the intention is that, should this proc fail, we'll mark
the whole test script as unsupported and then move onto the next test
script.
However, check_vfork_catchpoints also uses setup_gdb, and so, if that
call to setup_gdb fails we'll end up returning immediately from
check_vfork_catchpoints, and then continue with the test of _this_
test script, which is not correct.
To fix this I see two choices, one is remove the use of 'return -code
return' from setup_gdb, however, this would require every use of
setup_gdb to then be placed inside:
if { ![setup_gdb] } {
return
}
Or, I can wrap the one call to setup_gdb in check_vfork_catchpoints
and check its return code.
I chose the second option as this is the smaller code change.
There should be no change in what is tested after this commit.
|
|
|
|
Many of the reloc error messages have already been converted from
using %C to using %H in ld.bfd, to print section+offset as well as
file/line/function. This catches a few remaining, and changes gold to
do the same.
PR 10957
bfd/
* elf32-sh.c (sh_elf_relocate_section): Use %H in error messages.
gold/
* object.cc (Relocate_info::location): Always report section+offset.
* testsuite/debug_msg.sh: Adjust to suit.
* testsuite/x32_overflow_pc32.sh: Likewise.
* testsuite/x86_64_overflow_pc32.sh: Likewise.
ld/
* emultempl/pe.em (read_addend): Use %H in error message.
* emultempl/pep.em (read_addend): Likewise.
* ldcref.c (check_reloc_refs): Likewise.
* ldmain.c (warning_find_reloc, undefined_symbol): Likewise.
* pe-dll.c (pe_create_import_fixup): Likewise.
* testsuite/ld-cris/undef2.d: Adjust expected output to suit.
* testsuite/ld-cris/undef3.d: Likewise.
* testsuite/ld-elf/shared.exp: Likewise.
* testsuite/ld-i386/compressed1.d: Likewise.
* testsuite/ld-ia64/line.exp: Likewise.
* testsuite/ld-plugin/lto.exp: Likewise.
* testsuite/ld-undefined/undefined.exp: Likewise.
* testsuite/ld-x86-64/compressed1.d: Likewise.
* testsuite/ld-x86-64/line.exp: Likewise.
* testsuite/ld-x86-64/pr27587.err: Likewise.
|
|
This patch is aimed at making __bss_start properly aligned with the
first of any bss-style sections following. Most of the work here
involves keeping track of the last output section seen when processing
the linker script.
You can almost align __bss_start properly by using
${RELOCATING+. = ALIGN(${DATA_SDATA-${NO_SMALL_DATA-ALIGNOF(.${SBSS_NAME}) != 0 ? ALIGNOF(.${SBSS_NAME}) : }}${BSS_PLT+ALIGNOF(.plt) != 0 ? ALIGNOF(.plt) : }ALIGNOF(.${BSS_NAME}));}
and changing every place that defines NO_SMALL_DATA to use " ", but
having two .plt sections (marked SPECIAL) on some backends foils that
idea. The problem is that you only want to pick up the following
.plt, not the one preceeding __bss_start in the data segment if the
backend decides that is the proper .plt section.
Perhaps that could be fixed too, but I decided instead to extend the
linker script a little. THIS_SECTION and PREV_SECTION could easily be
added too.
* ld.texi (ALIGNOF, SIZEOF): Update and mention NEXT_SECTION.
* ldexp.c (output_section_find): New function.
(fold_name <ALIGNOF, SIZEOF>): Use output_section_find.
(exp_fold_tree): Add os parameter. Adjust all calls.
(exp_fold_tree_no_dot, exp_get_vma, exp_get_power): Likewise.
* ldexp.h (struct ldexp_control): Add last_os.
(exp_fold_tree, exp_fold_tree_no_dot): Update prototypes.
(exp_get_vma, exp_get_power): Likewise.
* ldlang.c: Pass last output section to expression folder
calls throughout file.
(open_input_bfds): Add os parameter to track last os seen.
(lang_size_sections_1): Rename output_section_statement param
to current_os. Track last os.
(lang_do_assignments_1): Track last os.
* scripttempl/arclinux.sc: Align to ALIGNOF NEXT_SECTION
before defining __bss_start.
* scripttempl/elf.sc: Likewise.
* scripttempl/elf64bpf.sc: Likewise.
* scripttempl/elf64hppa.sc: Likewise.
* scripttempl/elf_chaos.sc: Likewise.
* scripttempl/elfarc.sc: Likewise.
* scripttempl/elfd10v.sc: Likewise.
* scripttempl/elfxtensa.sc: Likewise.
* scripttempl/epiphany_4x4.sc: Likewise.
* scripttempl/iq2000.sc: Likewise.
* scripttempl/mep.sc: Likewise.
* scripttempl/nds32elf.sc: Likewise.
* scripttempl/xstormy16.sc: Likewise.
* testsuite/ld-x86-64/pe-x86-64-5.od: Update expected __bss_start.
* testsuite/ld-x86-64/pe-x86-64-5.rd: Likewise.
|
|
gdb.testsuite/gdb-caching-proc-consistency.exp
With test-case gdb.testsuite/gdb-caching-proc-consistency.exp we run into:
...
ERROR: no fileid for xerxes
Couldn't send help target native to GDB.
UNRESOLVED: <exp>: have_native_target: initial: help target native
...
Fix this by handling have_native_target in
gdb.testsuite/gdb-caching-proc-consistency.exp.
Tested on x86_64-linux.
|
|
|
|
This commit builds on this earlier work:
commit 9fe01a376b2fb096e4836e985ba316ce9dc02399
Date: Thu Jun 29 11:26:55 2023 -0600
Update TUI window title when changed
and makes tui_win_info::title private, renaming to m_title at the same
time. There's a new tui_win_info::title() member function to provide
read-only access to the title.
There should be no user visible changes after this commit.
Approved-By: Tom Tromey <tom@tromey.com>
|
|
After the commit:
commit 6647f05df023b63bbe056e9167e9e234172fa2ca
Date: Tue Jan 24 18:13:38 2023 +0100
gdb: defer warnings when loading separate debug files
It was pointed out[1] that the warnings being deferred and then later
emitted lacked styling. The warnings lacked styling before the above
commit, but it was suggested that the filenames in these warnings
should be styled, and this commit does this.
There were a couple of previous attempts[2][3][4] to solve this
problem, but these all tried to extend the mechanism introduced in the
above commit, the deferred warnings were placed directly into a
std::vector, but now we tried to, when appropriate, style these
warnings. The review feedback that this approach looked too complex.
So instead, this revision adds a new helper class 'deferred_warnings'
which can be used to collect a set of deferred warnings, and then emit
these deferred warnings later, if needed. This helper class hides the
complexity, so at the point the deferred warning is created no extra
logic is required.
The deferred_warnings class will style the deferred warnings only if
gdb_stderr supports styling. GDB's warnings are sent to gdb_stderr,
so this should ensure we only style when expected.
There was also review feedback[5] that all of the warnings should be
bundled into a single string_file, this has not been done. I feel
pretty strongly that separate warnings should be emitted using
separate "warning" calls. If we do end up with multiple warnings in
this case they aren't really related, one will be about looking up
debug via .gnu_debuglink, while the other will be about build-id based
lookup. So I'd really rather keep the warnings separate.
[1] https://inbox.sourceware.org/gdb-patches/87edr9pcku.fsf@tromey.com/
[2] https://inbox.sourceware.org/gdb-patches/20230216195604.2685177-1-ahajkova@redhat.com/
[3] https://inbox.sourceware.org/gdb-patches/20230217123547.2737612-1-ahajkova@redhat.com/
[4] https://inbox.sourceware.org/gdb-patches/20230320145638.1202335-1-ahajkova@redhat.com/
[5] https://inbox.sourceware.org/gdb-patches/87o7nh1g8h.fsf@tromey.com/
Co-Authored-By: Alexandra Hájková <ahajkova@redhat.com>
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
When running test-case gdb.dwarf2/forward-spec.exp with check-read1 we run
into:
...
parent: ((cooked_index_entry *) 0xFAIL: <exp>: v has a parent
7fdc1c002ed0) [ns]^M
...
The problem is using regexps containing '.' to avoid escaping, which makes
them too generic.
Fix this by eliminating the '.' from the regexps.
Tested on x86_64-linux.
|