When running test-case gdb.python/py-mi-cmd.exp on openSUSE Leap 42.3 with
python 3.4, I occasionally run into:
...
Expecting: ^(-pycmd dct[^M
]+)?(\^done,result={hello="world",times="42"}[^M
]+[(]gdb[)] ^M
[ ]*)
-pycmd dct^M
^done,result={times="42",hello="world"}^M
(gdb) ^M
FAIL: gdb.python/py-mi-cmd.exp: -pycmd dct (unexpected output)
...
The problem is that the data type used here in py-mi-cmd.py:
...
elif argv[0] == "dct":
return {"result": {"hello": "world", "times": 42}}
...
is a dictionary, and only starting version 3.6 are dictionaries insertion
ordered, so using PyDict_Next in serialize_mi_result doesn't guarantee a
fixed order.
Fix this by allowing the alternative order.
Tested on x86_64-linux.
Some ravenscar runtimes implement lazy FPU handling. On these
runtimes, the FPU is only initialized when a task tries to use it.
Furthermore, the FP registers aren't automatically saved on a task
switch -- instead, the save is deferred until the new task tries to
use the FPU. Furthermore, each task's context area has a flag
indicating whether the FPU has been initialized for this task.
This patch teaches GDB to understand this implementation. When
fetching or storing registers, GDB now checks to see whether the live
FP registers should be used. If not, the task's saved FP registers
will be used if the task has caused FPU initialization.
Currently only AArch64 uses this code. bb-runtimes implements this
for ARM as well, but GDB doesn't yet have an arm-ravenscar-thread.c.
Currently, the ravenscar-thread implementation for each architecture
is written by hand. However, these are actually written by
copy-paste. It seems better to switch to a table-driven approach.
The previous code also fetched all registers whenever any register was
requested. This is corrected in the new implementation.
We found a few bugs in aarch64-ravenscar-thread.c.
First, some of the register offsets were incorrect. The "bb-runtimes"
file for this runtime had the wrong offsets in comments, which GDB
took to be correct. However, those comments didn't account for
alignment. This patch adjusts the offsets.
Next, the "FPU Saved field" is not a register -- it is an
implementation detail of the runtime. This is removed.
Finally, I think the FP registers are actually named V0-V31, and the
"Q" names are pseudo-registers. This patch fixes the comment.
Implement LoongArch/Linux support, including XML target description
handling based on features determined, GPR regset support, and software
breakpoint handling.
In the Linux kernel code of LoongArch, ptrace implements PTRACE_POKEUSR
and PTRACE_PEEKUSR in the arch_ptrace function, so srv_linux_usrregs is
set to yes.
With this patch on LoongArch:
$ make check-gdb TESTS="gdb.server/server-connect.exp"
[...]
# of expected passes 18
[...]
Signed-off-by: Youling Tang <tangyouling@loongson.cn>
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
With gcc-12 and target board unix/-m32, we run into:
...
(gdb) ^M
Expecting: ^(-var-create A_String_Access \* A_String_Access[^M
]+)?(\^done,name="A_String_Access",numchild="1",.*[^M
]+[(]gdb[)] ^M
[ ]*)
-var-create A_String_Access * A_String_Access^M
^error,msg="Value out of range."^M
(gdb) ^M
FAIL: gdb.ada/mi_var_access.exp: Create varobj (unexpected output)
...
What happens is easier to understand if we take things out of the mi context:
...
$ gdb -q -batch \
outputs/gdb.ada/mi_var_access/mi_access \
-ex "b mi_access.adb:19" \
-ex run \
-ex "p A_String_Access"
...
Breakpoint 1, mi_access () at mi_access.adb:19
19 A_String : String (3 .. 5) := "345"; -- STOP
$1 = (pck.string_access) <error reading variable: Value out of range.>
...
while with target board unix we have instead:
...
$1 = (pck.string_access) 0x431b40 <ada_main.sec_default_sized_stacks>
...
The var-create command samples the value of the variable at a location where
the variable is not yet initialized, and with target board unix we
accidentally hit a valid address, but with target board unix/-m32 that's not
the case.
Fix the FAIL by accepting the error message.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28464
On Thu, Jun 09, 2022 at 08:59:37AM -0700, John Baldwin wrote:
> On 6/9/22 1:58 AM, Tom de Vries via Gdb-patches wrote:
> > Hi,
> >
> > With an --enable-targets=all build and target board unix/-m32 I run into a
> > FAIL in test-case gdb.base/corefile.exp:
> > ...
> > (gdb) file outputs/gdb.base/corefile/corefile^M
> > Reading symbols from outputs/gdb.base/corefile/corefile...^M
> > (gdb) core-file outputs/gdb.base/corefile/corefile.core^M
> > warning: core file may not match specified executable file.^M
> > [New LWP 12011]^M
> > Core was generated by `outputs/gdb.base/corefile/co'.^M
> > Program terminated with signal SIGABRT, Aborted.^M
> > (gdb) FAIL: gdb.base/corefile.exp: core-file warning-free
> > ...
> >
> > The warning is there because of this mismatch between core and exec:
> > ...
> > (gdb) p core_bfd->xvec
> > $3 = (const struct bfd_target *) 0x20112a0 <i386_elf32_fbsd_vec>
> > (gdb) p exec_bfd->xvec
> > $4 = (const struct bfd_target *) 0x2010b00 <i386_elf32_vec>
> > ...
> >
> > In the exec case, the detected architecture is i386_elf32_vec because this bit
> > of code in elfcode.h:elf_object_p():
> > ...
> > if (ebd->elf_machine_code != EM_NONE
> > && i_ehdrp->e_ident[EI_OSABI] != ebd->elf_osabi
> > && ebd->elf_osabi != ELFOSABI_NONE)
> > goto got_wrong_format_error;
> > ...
> > prevents i386_elf32_fbsd from matching.
> >
> > Fix the core matching by copying that code to elfcore.h:elf_core_file_p().
> >
> > Tested on x86_64-linux.
> >
> > Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29227
> >
> > Any comments?
Looks good.
> Looking at elfcore.h, it seems to have not gotten changes made to elfcode.h over
> time and is a bit rotted. I suspect that all of changes made in commit 0aabe54e62
> that added these lines in elfcode.h (along with several other changes) need to
> be applied to this function in elfcore.h, not just adding these lines.
Yes, the commit 0aabe54e62 changes likely should go in too. I'm a
little wary of adding all the sanity checks to elf_core_file_p since
that might result in some core files not being recognised at all. For
example, despite the FIXME I'd guess leaving out the EI_VERSION check
was deliberate. The following seems reasonable to me. Please test.
Starting with (future) Clang 15 (since
https://reviews.llvm.org/D120989), Clang emits the DWARF information
of global alias variables as DW_TAG_imported_declaration. However,
GDB does not handle it. It incorrectly always reads this tag as
C++/Fortran imported declaration (type alias, namespace alias and
Fortran module). This commit adds support to handle this tag as an
alias variable.
This change fixes the failures in the gdb.base/symbol-alias.exp
testcase with current git Clang. This testcase is also updated to
test nested (recursive) aliases.
MIPS should not be using BFD_RELOC_16 for its R_MIPS_16 relocation,
since R_MIPS_16 specifies a 16-bit field in a 32-bit word.
BFD_RELOC_16, emitted by generic code to handle fixups on 16-bit data
directives, expects fixups to operate on the whole of a 16-bit word.
This patch corrects the problem by using BFD_RELOC_MIPS_16, a new bfd
reloc that is used to generate R_MIPS_16. BFD_RELOC_16 is handled in
md_apply_fix for cases where the fixup can be applied at assembly
time. Like BFD_RELOC_8, BFD_RELOC_16 now has no corresponding object
file relocation, and thus .half, .hword, .short and .dc.w must be
resolved at assembly time. BFD_RELOC_MIPS_REL16 is removed by this
patch since it isn't used.
PR 3243
PR 26542
* reloc.c (BFD_RELOC_MIPS_16): Rename from BFD_RELOC_MIPS_REL16.
* elf32-mips.c (mips_reloc_map): Map BFD_RELOC_MIPS_16 to R_MIPS_16.
* elf64-mips.c (mips_reloc_map): Likewise, delete BFD_RELOC_MIPS_REL16.
* elfn32-mips.c (mips_reloc_map): Likewise.
* libbfd.h: Regenerate.
* bfd-in2.h: Regenerate.
gas/
* config/tc-mips.c (append_insn): Handle BFD_RELOC_MIPS_16.
(macro_build): Likewise.
(mips_percent_op <%half>): Generate BFD_RELOC_MIPS_16.
(md_apply_fix): Handle BFD_RELOC_16 and BFD_RELOC_MIPS_16 when fx_done.
ld/
* testsuite/ld-mips-elf/reloc-local-overflow.d,
* testsuite/ld-mips-elf/reloc-local-overflow.s: Rewrite.
If the howto is actually used, an all-zero dst_mask will result in
unchanged section contents on attempting to apply R_MIPS_16.
* elfn32-mips.c (elf_mips_howto_table_rela <R_MIPS_16>): Correct
dst_mask.
Make it a little more obvious that remap_debug_filename returns an
allocated string (that should be freed) by returning a char * rather
than const char *. Free a few missed cases in dwarf2dbg.c, and free
other memory allocated in dwarf2dbg.c. Also remove static
initialisation of variables and initialise in dwarf2_init instead,
in order to ensure gas state is saner for oss-fuzz.
* remap.c (remap_debug_filename): Remove const from return.
* as.h (remap_debug_filename): Update prototype.
* config/obj-elf.c (obj_elf_ident): Simplify free of
remap_debug_filename output.
* stabs.c (stabs_generate_asm_file): Likewise.
* dwarf2dbg.c (dirs, dirs_in_use, dirs_allocated, current): Don't
initialise statically..
(dwarf2_init): ..do so here, along with most other static vars.
(assign_file_to_slot): Don't set files_allocated until we
succeed in allocating memory.
(purge_generated_debug): Add bool param, free more stuff if true.
(dwarf2_directive_filename): Adjust purge_generated_debug call.
(process_entries): Don't free line_entry here..
(dwarf2_cleanup): ..do so here instead, new function.
(dwarf2_finish): Call dwarf2_cleanup. When chaining together
subseg line entries, unhook entries from old subseg list.
(dwarf2_directive_loc): Free remap_debug_filename string.
(out_dir_and_file_list): Likewise.
(out_debug_str): Likewise.
When running test-case gdb.reverse/test_ioctl_TCSETSW.exp with glibc debuginfo
installed, I run into:
...
(gdb) PASS: gdb.reverse/test_ioctl_TCSETSW.exp: at TCSETSW call
step^M
__tcsetattr (fd=0, optional_actions=1, termios_p=0x7fffffffcf50) at \
../sysdeps/unix/sysv/linux/tcsetattr.c:45^M
45 {^M
(gdb) FAIL: gdb.reverse/test_ioctl_TCSETSW.exp: handle TCSETSW
...
The problem is that the step is expected to step over the call to tcsetattr,
but due to glibc debuginfo being installed, we step into the call.
Fix this by using next instead of step.
Tested on x86_64-linux.
With --enable-targets=all we have:
...
$ gdb -q -batch -ex "maint selftest"
...
Running selftest regcache::cooked_read_test::m68hc11.
warning: No frame soft register found in the symbol table.
Stack backtrace will not work.
Running selftest regcache::cooked_read_test::m68hc12.
warning: No frame soft register found in the symbol table.
Stack backtrace will not work.
Running selftest regcache::cooked_read_test::m68hc12:HCS12.
warning: No frame soft register found in the symbol table.
Stack backtrace will not work.
...
Likewise for regcache::cooked_write_test.
The warning has no use in the selftest context.
Fix this by skipping the specific selftests.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29224
We can't put a breakpoint in the middle of a ll/sc atomic sequence,
so look for the end of the sequence and put the breakpoint there.
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Results in configure output like:
```
checking for X... no
/var/tmp/portage/sys-devel/gdb-12.1/work/gdb-12.1/gdb/configure: 18837: test: yes: unexpected operator
checking whether to use babeltrace... auto
```
... when /bin/sh is provided by a POSIX-compliant shell, like dash,
instead of bash.
Registers in CSKY architecture included:
1. 32 gprs
2. 16 ars (alternative gprs used for quick interrupt)
3. hi, lo, pc
4. fr0~fr31, fcsr, fid, fesr
5. vr0~vr15
6. ((32 banks) * 32) cr regs (max 32 banks, 32 control regs a bank)
For register names:
Except over control registers, other registers, like gprs, hi, lo ...
are fixed names. Among the 32*32 control registers, some used registers
will have fixed names, others will have a default name "cpxcry". 'x'
refers to bank, y refers index in the bank(a control register in bank
4 with index 14 will has a default name cp4cr14).
For register numbers in GDB:
We assign a fixed number to each register in GDB, like:
r0~r31 with 0~31
hi, lo with 36, 37
fpu/vpu with 40~71
...
described in function csky_get_supported_register_by_index().
Function csky_get_supported_tdesc_registers_count():
To calculate the total number of registers that GDB can analyze,
including those with fixed names and those with default register names.
Function csky_get_supported_register_by_index():
To find a supported struct csky_supported_tdesc_register, return a
struct include name with regnum via index.
Arrays csky_supported_tdesc_feature_names[]:
Include all supported feature names in tdesc-xmls.
We use the information described above to load the register description
file of the target from the stub. When loading, do a little check that
whether the register description file contains SP, LR and PC.
On openSUSE Leap 42.3 with python 3.4, I run into:
...
(gdb) python import pygments^M
Traceback (most recent call last):^M
File "<string>", line 1, in <module>^M
ImportError: No module named 'pygments'^M
Error while executing Python code.^M
(gdb) FAIL: gdb.base/style.exp: python import pygments
ERROR: unexpected output from python import
...
because gdb_py_module_available doesn't handle the single quotes around the
module name in the ImportError.
Fix this by allowing the single quotes.
Tested on x86_64-linux.
With its movement to the stack, and with the subsequent desire to
initialize the entire instr_info instances, this has become doubly
inefficient. Individual users have better knowledge of how big a buffer
they need, and in a number of cases going through an intermediate buffer
can be avoided altogether.
Having got confirmation that it wasn't intentional to print memory
operand displacements with inconsistent style, print_displacement() is
now using dis_style_address_offset consistently (eliminating the need
for callers to pass in a style).
While touching print_operand_value() also convert its "hex" parameter to
bool. And while altering (and moving) oappend_immediate(), fold
oappend_maybe_intel_with_style() into its only remaining caller. Finally
where doing adjustments, use snprintf() in favor of sprintf().
By changing the values used for "artificial" prefix values,
all_prefixes[] can be shrunk to array of unsigned char. All that
additionally needs adjusting is the printing of possible apparently
standalone prefixes when recovering from longjmp(): Simply check
whether any prefixes were successfully decoded, to avoid converting
opcode bytes matching the "artificial" values to prefix mnemonics.
Similarly by re-arranging the bits assigned to PREFIX_* mask values
we can fit all segment register masks in a byte and hence shrink
active_seg_prefix to unsigned char.
Somewhat similarly with last_*_prefix representing offsets into the
opcode being disassembled, signed char is sufficient to hold all possible
values.
Commit 39fb369834 ("opcodes: Make i386-dis.c thread-safe") introduced
a lot of uninitialized data. Alan has in particular observed ubsan
taking issue with the loop inverting the order of operands, where
op_riprel[] - an array of bool - can hold values other than 0 or 1.
Move instantiation of struct instr_info into print_insn() (thus having
just a single central point), and make use of C99 dedicated initializers
to fill fields right in the initializer where possible. This way all
fields not explicitly initialized will be zero-filled, which in turn
allows dropping of some other explicit initialization later in the
function or in ckprefix(). Additionally this removes a lot of
indirection, as all "ins->info" uses can simply become "info".
Make one further arrangement though, to limit the amount of data needing
(zero)initializing on every invocation: Convert the op_out structure
member to just an array of pointers, with the actual arrays living
inside print_insn() (and, as befoe, having just their 1st char filled
with nul).
While there, instead of adjusting print_insn()'s forward declaration,
arrange for no such declaration to be needed in the first place.
Mark pointed out that my recent addrmap C++-ficiation changes caused a
regression in the self-tests. This patch fixes the problem by
updating this test not to allocate the mutable addrmap on an obstack.
While working on addrmaps, I noticed that psymtab_addrmap is no longer
needed now. It was introduced in ancient times as an optimization for
DWARF, but no other symbol reader was ever updated to use it. Now
that DWARF does not use psymtabs, it can be deleted.
Mutable addrmaps currently require an obstack. This was probably done
to avoid having to call splay_tree_delete, but examination of the code
shows that all mutable obstacks have a limited lifetime -- now it's
simple to treat them as ordinary C++ objects, in some cases
stack-allocating them, and have a destructor to make the needed call.
This patch implements this change.
This is a simply C++-ification of the basics of addrmap: it uses
virtual methods rather than a table of function pointers, and it
changes the concrete implementations to be subclasses.
Prior to c6ca3dab dropping support for Cygwin 1.5, __USEWIDE was not
defined for Cygwin 1.5. After that, it's always defined if __CYGWIN__
is, so remove __USEWIDE conditionals inside __CYGWIN__ conditionals.
With the registry rewrite series, on Fedora 34, I started seeing this
error in xcoffread.c:
../../binutils-gdb/gdb/xcoffread.c: In function ‘void read_xcoff_symtab(objfile*, legacy_psymtab*)’:
../../binutils-gdb/gdb/xcoffread.c:948:25: error: ‘main_aux’ is used uninitialized [-Werror=uninitialized]
948 | union internal_auxent fcn_aux_saved = main_aux;
| ^~~~~~~~~~~~~
../../binutils-gdb/gdb/xcoffread.c:933:25: note: ‘main_aux’ declared here
933 | union internal_auxent main_aux;
| ^~~~~~~~
I don't know why this error started suddenly... that seems weird,
because it's not obviously related to the changes I made.
Looking into it, it seems this line was intended to avoid a similar
warning -- but since 'main_aux' is uninitialized at the point where it
is used, this fix was incomplete.
This patch avoids the warning by initializing using "{}". I'm
checking this in.
The if statement in case gdb_sys_ioctl in function
record_linux_system_call in file gdb/linux-record.c is as follows:
if (tmpulongest == tdep->ioctl_FIOCLEX
|| tmpulongest == tdep->ioctl_FIONCLEX
....
|| tmpulongest == tdep->ioctl_TCSETSW
...
}
The PowerPC ioctl value for ioctl_TCSETW is 0x802c7415. The variable
ioctl_TCSETW is defined in gdb/linux-record.h as an int. The TCSETW value
has the MSB set to one so it is a negative integer. The comparison of the
unsigned long value tmpulongest to a negative integer value for
ioctl_TCSETSW fails.
This patch changes the declarations for the ioctl_* values in struct
linux_record_tdep to unsigned long to fix the comparisons between
tmpulongest and the tdep->ioctl_* values.
An additional test gdb.reverse/test_ioctl_TCSETSW.exp is added to verify
the gdb record_linux_system_call() if statement for the ioctl TCSETSW
succeeds.
This patch has been tested on Power 10 and Intel with no test failures.
Some of the ioctl numbers are based on the size of kernel termios structure.
Currently the PowerPC GDB definitions are "hard coded" into the ioctl
number.
The current PowerPC values for TCGETS, TCSETS, TCSETSW and TCSETSF are
defined in gdb/ppc-linux-tdep.c as:
record_tdep->ioctl_TCGETS = 0x403c7413;
record_tdep->ioctl_TCSETS = 0x803c7414;
record_tdep->ioctl_TCSETSW = 0x803c7415;
record_tdep->ioctl_TCSETSF = 0x803c7416;
Where the termios structure size is in hex digits [5:4] as 0x3c.
The definition for the PowerPC termios structure is given in:
arch/powerpc/include/uapi/asm/termbits.h
The size of the termios data structure in this file is 0x2c not 0x3c.
This patch changes the hex digits for the size of the PowerPC termios size
in the ioctl values for TCGETS, TCSETS, TCSETSW and TCSETSF to 0x2c.
This patch also changes the hard coding to generate the number based on a
it easier to update the ioctl numbers.
Since pretty much forever the get_compiler_info function has included
these lines:
# Most compilers will evaluate comparisons and other boolean
# operations to 0 or 1.
uplevel \#0 { set true 1 }
uplevel \#0 { set false 0 }
These define global variables true (to 1) and false (to 0).
It seems odd to me that these globals are defined in
get_compiler_info, I guess maybe the original thinking was that if a
compiler had different true/false values then we would detect it there
and define true/false differently.
I don't think we should be bundling this logic into get_compiler_info,
it seems weird to me that in order to use $true/$false a user needs to
first call get_compiler_info.
It would be better I think if each test script that wants these
variables just defined them itself, if in the future we did need
different true/false values based on compiler version then we'd just
do:
if { [test_compiler_info "some_pattern"] } {
# Defined true/false one way...
} else {
# Defined true/false another way...
}
But given the current true/false definitions have been in place since
at least 1999, I suspect this will not be needed any time soon.
Given that the definitions of true/false are so simple, right now my
suggestion is just to define them in each test script that wants
them (there's not that many). If we ever did need more complex logic
then we can always add a function in gdb.exp that sets up these
globals, but that seems overkill for now.
There should be no change in what is tested after this commit.
This variable is useful when exercising AArch64 multi-arch support (debugging
32-bit AArch32 executables).
Unfortunately it isn't well documented. This patch adds information about it
and explains how to use it.
With gcc-12, I get for test-case gdb.base/vla-struct-fields.exp:
...
(gdb) print inner_vla_struct_object_size == sizeof(inner_vla_struct_object)^M
$7 = 1^M
(gdb) XPASS: gdb.base/vla-struct-fields.exp: size of inner_vla_struct_object
...
Fix this by limiting the xfailing to gcc-11 and earlier. Also, limit the
xfailing to the equality test.
Tested on x86_64-linux.
On openSUSE Tumbleweed with gcc-12, I run into a timeout:
...
(gdb) print value^M
Multiple matches for value^M
[0] cancel^M
[1] ada.strings.maps.value (<ref> ada.strings.maps.character_mapping; \
character) return character at a-strmap.adb:599^M
[2] pck.value at src/gdb/testsuite/gdb.ada/ghost/pck.ads:17^M
[3] system.object_reader.value (<ref> system.object_reader.object_symbol) \
return system.object_reader.uint64 at s-objrea.adb:2279^M
[4] system.traceback.symbolic.value (system.address) return string at \
s-trasym.adb:200^M
> FAIL: gdb.ada/ghost.exp: print value (timeout)
print ghost_value^M
Argument must be choice number^M
(gdb) FAIL: gdb.ada/ghost.exp: print ghost_value
...
Fix this by prefixing value (as well as the other printed values) with the
package name:
...
(gdb) print pck.value^M
...
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29055
