Add new files:
gdb/arch/csky.c
gdb/arch/csky.h
gdb/features/cskyv2-linux.c
gdbserver/linux-csky-low.cc
1. In gdb/arch/csky.c file, add function "csky_create_target_description()"
for csky_target::low_arch_setup(). later, it can be used for csky native gdb.
2. In gdb/features/cskyv2-linux.c file, create target_tdesc for csky, include
gprs, pc, hi, lo, float, vector and float control registers.
3. In gdbserver/linux-csky-low.cc file, using PTRACE_GET/SET_RGESET to
get/set registers. The main data structures in asm/ptrace.h are:
struct pt_regs {
unsigned long tls;
unsigned long lr;
unsigned long pc;
unsigned long sr;
unsigned long usp;
/*
* a0, a1, a2, a3:
* r0, r1, r2, r3
*/
unsigned long orig_a0;
unsigned long a0;
unsigned long a1;
unsigned long a2;
unsigned long a3;
/*
* r4 ~ r13
*/
unsigned long regs[10];
/* r16 ~ r30 */
unsigned long exregs[15];
unsigned long rhi;
unsigned long rlo;
unsigned long dcsr;
};
struct user_fp {
unsigned long vr[96];
unsigned long fcr;
unsigned long fesr;
unsigned long fid;
unsigned long reserved;
};
I went through all the uses of dynamic_cast<> in gdb, looking for ones
that could be replaced with checked_static_cast. This patch is the
result. Regression tested on x86-64 Fedora 34.
clang doesn't add encoding to the name of complex variables, only says
that the type name is complex, making the relevant tests fail.
This patch adds the xfails to the tests that expect the variable name to
include it.
When running gdb.base/call-ar-st.exp against Clang, we see one FAIL,
like so:
print_all_arrays (array_i=<main.integer_array>, array_c=<main.char_array> "ZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZa
ZaZaZaZaZaZaZaZaZaZaZaZa", array_f=<main.float_array>, array_d=<main.double_array>) at ../../../src/gdb/testsuite/gdb.base/call-ar-st.c:274
274 print_int_array(array_i); /* -step1- */
(gdb) FAIL: gdb.base/call-ar-st.exp: step inside print_all_arrays
With GCC we instead see:
print_all_arrays (array_i=<integer_array>, array_c=<char_array> "ZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZaZa", array_f=<float_array>, array_d=<double_array>) at /home/pedro/gdb/build/gdb/testsuite/../../../src/gdb/testsuite/gdb.base/call-ar-st.c:274
274 print_int_array(array_i); /* -step1- */
(gdb) PASS: gdb.base/call-ar-st.exp: step inside print_all_arrays
The difference is that with Clang we get:
array_i=<main.integer_array>, ...
instead of
array_i = <integer_array>, ...
These symbols are local static variables, and "main" is the name of
the function they are defined in. GCC instead appends a sequence
number to the linkage name:
$ nm -A call-ar-st.gcc | grep integer_
call-ar-st/call-ar-st:00000000000061a0 b integer_array.3968
$ nm -A call-ar-st.clang | grep integer_
call-ar-st:00000000004061a0 b main.integer_array
This commit changes the testcase to accept both outputs, as they are
functionally identical.
Co-Authored-By: Pedro Alves <pedro@palves.net>
Change-Id: Iaf2ccdb9d5996e0268ed12f595a6e04b368bfcb4
The test specifically mentions that it doesn't care where the program
stops, however it was still testing for a specific location. The clang
compiler emits different line information for epilogue, so GDB reports a
different stopping location, depending on the used compiler. With this
patch the test works even with clang.
Clang organizes the variables differently to gcc in the original version
of this code, leading to the following differences when testing
p (int*) &dataglobal + 1
gcc:
$16 = (int *) 0x404034 <datalocal>
clang:
$16 = (int *) 0x404034 <dataglobal8>
However, since the important part of this test doesn't seem to be which
symbol is linked, but rather if GDB is correctly increasing the
address. This test was changed to actually measure address changes,
instead of assuming the ordering and naming of symbols.
Co-Authored-By: Andrew Burgess <aburgess@redhat.com>
When running selftest run_on_main_thread and pressing ^C, we can run into:
...
Running selftest run_on_main_thread.
terminate called without an active exception
Fatal signal: Aborted
...
The selftest function looks like this:
...
static void
run_tests ()
{
std::thread thread;
done = false;
{
gdb::block_signals blocker;
thread = std::thread (set_done);
}
while (!done && gdb_do_one_event () >= 0)
;
/* Actually the test will just hang, but we want to test
something. */
SELF_CHECK (done);
thread.join ();
}
...
The error message we see is due to the destructor of thread being called while
thread is joinable.
This is supposed to be taken care of by thread.join (), but the ^C prevents
that one from being called, while the destructor is still called.
Fix this by ensuring thread.join () is called (if indeed required) before the
destructor using SCOPE_EXIT.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29549
In commit cd919f5533 ("[gdb/testsuite] Fix
gdb.dwarf2/dw2-dir-file-name.exp"), I made gdb.dwarf2/dw2-dir-file-name.exp
independent of prologue analyzers, using this change:
...
- gdb_breakpoint $func
+ gdb_breakpoint *$func
...
That however caused a regression on ppc64le. For PowerPC, as described in the
ELFv2 ABI, a function can have a global and local entry point.
Setting a breakpoint on *$func effectively creates a breakpoint for the global
entry point, so if the function is entered through the local entry point, the
breakpoint doesn't trigger.
Fix this by reverting commit cd919f5533, and setting the breakpoint on
${func}_label instead.
Tested on x86_64-linux and ppc64le-linux.
When running test-case gdb.dwarf2/dw2-dir-file-name.exp with clang, we run
into:
...
(gdb) break *compdir_missing__ldir_missing__file_basename^M
Breakpoint 2 at 0x400580^M
(gdb) continue^M
Continuing.^M
^M
Breakpoint 2, 0x0000000000400580 in \
compdir_missing.ldir_missing.file_basename ()^M
(gdb) FAIL: gdb.dwarf2/dw2-dir-file-name.exp: \
compdir_missing__ldir_missing__file_basename: continue to breakpoint: \
compdir_missing__ldir_missing__file_basename
...
The problem is that the test-case uses labels outside functions, which is know
to cause problem with clang, as documented in the comment for proc
function_range.
Fix this by using get_func_info instead.
Tested on x86_64-linux, with both gcc 7.5.0 and clang 13.0.0.
Currently, the test-case contained in this patch fails:
...
(gdb) p (int) foo ()^M
Invalid cast.^M
(gdb) FAIL: gdb.dwarf2/dw2-unspecified-type.exp: p (int) foo ()
...
because DW_TAG_unspecified_type is translated as void.
There's some code in read_unspecified_type that marks the type as stub, but
that's only active for ada:
...
if (cu->lang () == language_ada)
type->set_is_stub (true);
...
Fix this by:
- marking the type as a stub for all languages, and
- handling the stub return type case in call_function_by_hand_dummy.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29558
This test generates 48 failures on Power 9 when testing with HW watchpoints
enabled. Note HW watchpoint support is disabled on Power 9 due to a HW bug.
The skip_hw_watchpoint_tests proc must be used to correctly determine
if the processor supports HW watchpoints.
This patch replaces the [target_info exists gdb,no_hardware_watchpoints]
with the skip_hw_watchpoint_tests check.
This patch was tested on Power 9, Power 10 and X86-64 with no regressions.
On aarch64-linux, with gcc 7.5.0, we run into:
...
(gdb) frame^M
#0 callee.increment (val=99.0, val@entry=9.18340949e-41, msg=...) at \
callee.adb:21^M
21 if Val > 200.0 then^M
(gdb) FAIL: gdb.ada/O2_float_param.exp: scenario=all: frame
...
The problem is a GCC bug, filed as "PR98148 - [AArch64] Wrong location
expression for function entry values" (
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98148 ).
Xfail the test for aarch64 and gcc 7.
Tested on x86_64-linux and aarch64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29418
On aarch64-linux, I run into:
...
Breakpoint 2, pck.inspect (obj=0x430eb0 \
<system.pool_global.global_pool_object>, <objL>=0) at pck.adb:17^M
17 procedure Inspect (Obj: access Top_T'Class) is^M
(gdb) FAIL: gdb.ada/access_tagged_param.exp: continue
...
while on x86_64-linux, I see:
...
Breakpoint 2, pck.inspect (obj=0x62b2a0, <objL>=2) at pck.adb:19^M
19 null;^M
(gdb) PASS: gdb.ada/access_tagged_param.exp: continue
...
Note the different line numbers, 17 vs 19.
The difference comes from the gdbarch_skip_prologue implementation.
The amd64_skip_prologue implementation doesn't use gcc line numbers, and falls
back to the architecture-specific prologue analyzer, which correctly skips
past the prologue, to address 0x4022f7:
...
00000000004022ec <pck__inspect>:
4022ec: 55 push %rbp
4022ed: 48 89 e5 mov %rsp,%rbp
4022f0: 48 89 7d f8 mov %rdi,-0x8(%rbp)
4022f4: 89 75 f4 mov %esi,-0xc(%rbp)
4022f7: 90 nop
4022f8: 90 nop
4022f9: 5d pop %rbp
4022fa: c3 ret
...
The aarch64_skip_prologue implementation does use gcc line numbers, which are:
...
File name Line number Starting address View Stmt
pck.adb 17 0x402580 x
pck.adb 17 0x402580 1 x
pck.adb 19 0x40258c x
pck.adb 20 0x402590 x
...
and which are represented like this internally in gdb:
...
INDEX LINE ADDRESS IS-STMT PROLOGUE-END
0 17 0x0000000000402580 Y
1 17 0x0000000000402580 Y
2 19 0x000000000040258c Y
3 20 0x0000000000402590 Y
4 END 0x00000000004025a0 Y
...
The second entry is interpreted as end-of-prologue, so 0x402580 is used, while
the actual end of the prologue is at 0x40258c:
...
0000000000402580 <pck__inspect>:
402580: d10043ff sub sp, sp, #0x10
402584: f90007e0 str x0, [sp, #8]
402588: b90007e1 str w1, [sp, #4]
40258c: d503201f nop
402590: d503201f nop
402594: 910043ff add sp, sp, #0x10
402598: d65f03c0 ret
40259c: d503201f nop
...
Note that the architecture-specific prologue analyzer would have gotten this
right:
...
(gdb) p /x aarch64_analyze_prologue (gdbarch, pc, pc + 128, 0)
$2 = 0x40258c
...
Fix the FAIL by making the test-case more robust against problems in prologue
skipping, by setting the breakpoint on line 19 instead.
Likewise in a few similar test-cases.
Tested on x86_64-linux and aarch64-linux.
v2:
- Add 32-bit Arm instruction selftest
- Refactored abstract memory reader into abstract instruction reader
- Adjusted code to use templated type and to use host endianness as
opposed to target endianness.
The arm record tests handle 16-bit and 32-bit thumb instructions, but the
code is laid out in a way that handles the 32-bit thumb instructions as
two 16-bit parts.
This is fine, but it is prone to host-endianness issues given how the two
16-bit parts are stored and how they are accessed later on. Arm is
little-endian by default, so running this test with a GDB built with
--enable-targets=all and on a big endian host will run into the following:
Running selftest arm-record.
Process record and replay target doesn't support syscall number -2036195
Process record does not support instruction 0x7f70ee1d at address 0x0.
Self test failed: self-test failed at ../../binutils-gdb/gdb/arm-tdep.c:14482
It turns out the abstract memory reader class is more generic than it needs to
be, and we can simplify the code a bit by assuming we have a simple instruction
reader that only reads up to 4 bytes, which is the length of a 32-bit
instruction.
Instead of returning a bool, we return instead the instruction that has been
read. This way we avoid having to deal with the endianness conversion, and use
the host endianness instead. The Arm selftests can be executed on non-Arm
hosts.
While at it, Tom suggested adding a 32-bit Arm instruction selftest to increase
the coverage of the selftests.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29432
Co-authored-by: Tom de Vries <tdevries@suse.de>
On openSUSE Tumbleweed (using glibc 2.36), I run into:
...
(gdb) print /d (int) munmap (4198400, 4096)^M
Invalid cast.^M
(gdb) FAIL: gdb.base/break-main-file-remove-fail.exp: cmdline: \
get integer valueof "(int) munmap (4198400, 4096)"
...
The problem is that after starting the executable, the symbol has type
"void (*) (void)":
...
(gdb) p munmap
$1 = {<text variable, no debug info>} 0x401030 <munmap@plt>
(gdb) start
...
(gdb) p munmap
$2 = {void (void)} 0x7ffff7feb9a0 <__GI_munmap>
...
which causes the "Invalid cast" error.
Looking at the debug info for glibc for symbol __GI_munmap:
...
<0><189683>: Abbrev Number: 1 (DW_TAG_compile_unit)
<189691> DW_AT_name : ../sysdeps/unix/syscall-template.S
<189699> DW_AT_producer : GNU AS 2.39.0
<1><1896ae>: Abbrev Number: 2 (DW_TAG_subprogram)
<1896af> DW_AT_name : __GI___munmap
<1896b3> DW_AT_external : 1
<1896b4> DW_AT_low_pc : 0x10cad0
<1896bc> DW_AT_high_pc : 37
...
that's probably caused by this bit (or similar bits for other munmap aliases).
This is fixed in gas on trunk by commit 5578fbf672 ("GAS: Add a return type
tag to DWARF DIEs generated for function symbols").
Work around this (for say gas 2.39) by explicitly specifying the prototype for
munmap.
Likewise for getpid in a couple of other test-cases.
Tested on x86_64-linux.
This is paired with "opcodes: Add non-enum disassembler options".
There is a portable mechanism for disassembler options and used on some
architectures:
- ARC
- Arm
- MIPS
- PowerPC
- RISC-V
- S/390
However, it only supports following forms:
- [NAME]
- [NAME]=[ENUM_VALUE]
Valid values for [ENUM_VALUE] must be predefined in
disasm_option_arg_t.values. For instance, for -M cpu=[CPU] in ARC
architecture, opcodes/arc-dis.c builds valid CPU model list from
include/elf/arc-cpu.def.
In this commit, it adds following format:
- [NAME]=[ARBITRARY_VALUE] (cannot contain "," though)
This is identified by NULL value of disasm_option_arg_t.values
(normally, this is a non-NULL pointer to a NULL-terminated list).
gdb/ChangeLog:
* gdb/disasm.c (set_disassembler_options): Add support for
non-enum disassembler options.
(show_disassembler_options_sfunc): Likewise.
When running test-case gdb.cp/cpexprs-debug-types.exp on target board
cc-with-debug-names/gdb:debug_flags=-gdwarf-5, we get an executable with
a .debug_names section, but no .debug_types section. For dwarf-5, the TUs
are no longer put in a separate unit, but instead they're put in the
.debug_info section.
When loading the executable, the .debug_names section is silently ignored
because of this check in dwarf2_read_debug_names:
...
if (map->tu_count != 0)
{
/* We can only handle a single .debug_types when we have an
index. */
if (per_bfd->types.size () != 1)
return false;
...
which triggers because per_bfd->types.size () == 0.
The intention of the check is to make sure we don't have more that one
.debug_types section, as can happen in a object file (see PR12984):
...
$ grep "\.debug_types" 11.s
.section .debug_types,"G",@progbits,wt.75c042c23a9a07ee,comdat
.section .debug_types,"G",@progbits,wt.c59c413bf50a4607,comdat
...
Fix this by:
- changing the check condition to "per_bfd->types.size () > 1", and
- handling per_bfd->types.size () == 0.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29385
Add test-case gdb.dwarf2/debug-names-bad-cu-index.exp, a regression test for
commit 2fe9a3c41f ("[gdb/symtab] Fix bad compile unit index complaint").
Tested on x86_64-linux.
Add a test-case gdb.dwarf2/debug-names-tu.exp, that uses the dwarf assembler
to specify a .debug_names index with the TU list referring to a TU from the
.debug_types section.
This is intended to produce something similar to:
...
$ gcc -g -fdebug-types-section ~/hello.c -gdwarf-4
$ gdb-add-index -dwarf-5 a.out
...
Tested on x86_64-linux.
When running the gdb/configure script on ubuntu 22.04 with
python-3.10.4, I see:
checking for python... no
checking for python3... /usr/bin/python3
[...]/gdb/python/python-config.py:7: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils import sysconfig
[...]/gdb/python/python-config.py:7: DeprecationWarning: The distutils.sysconfig module is deprecated, use sysconfig instead
from distutils import sysconfig
[...]/gdb/python/python-config.py:7: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils import sysconfig
[...]/gdb/python/python-config.py:7: DeprecationWarning: The distutils.sysconfig module is deprecated, use sysconfig instead
from distutils import sysconfig
[...]/gdb/python/python-config.py:7: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils import sysconfig
[...]/gdb/python/python-config.py:7: DeprecationWarning: The distutils.sysconfig module is deprecated, use sysconfig instead
from distutils import sysconfig
checking for python... yes
The distutils module is deprecated as per the PEP 632[1] and will be
removed in python-3.12.
This patch migrates gdb/python/python-config.py from distutils.sysconfig
to the sysconfig module[2].
The sysconfig module has has been introduced in the standard library in
python 3.2. Given that support for python < 3.2 has been removed by
edae3fd660: "gdb/python: remove Python 2 support", this patch does not
need to support both implementations for backward compatibility.
Tested on ubuntu-22.04 and ubuntu 20.04.
[1] https://peps.python.org/pep-0632/
[2] https://docs.python.org/3/library/sysconfig.html
Change-Id: Id0df2baf3ee6ce68bd01c236b829ab4c0a4526f6
PR mi/10347 points out that using interpreter-exec inside of a
"define" command will crash gdb. The bug here is that
gdb_setup_readline doesn't check for the case where instream==nullptr.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=10347
PR mi/15811 points out that "source"ing a file that uses
interpreter-exec will put gdb in a weird state, where the CLI stops
working. The bug is that tui_interp::suspend does not unregister the
event file descriptor.
The test case is from Andrew Burgess.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=15811
gdb_setup_readline makes new streams and assigns to the various stream
members of struct ui. However, these assignments cause the previous
values to leak. As far as I can, this code is simply unnecessary and
can be removed -- with the exception of the assignment to gdb_stdtarg,
which is not initialized anywhere else.
A few spots setting some gdb output stream variables have a "for
moment" comment. These comments aren't useful and I think the moment
has passed -- these are permanent now.
This changes gdb so that, if ui::input_fd is set to -1, then it will
not be registered with the event loop. This is useful for the DAP
support code I wrote, but as it turns out to also be useful to
Insight, it seems best to check it in separately.
First, some background on the RISC-V registers fflags, frm, and fcsr.
These three registers all relate to the floating-point status and
control mechanism on RISC-V. The fcsr is the floatint-point control
status register, and consists of two parts, the flags (bits 0 to 4)
and the rounding-mode (bits 5 to 7).
The fcsr register is just one of many control/status registers (or
CSRs) available on RISC-V. The fflags and frm registers are also
CSRs. These CSRs are aliases for the relevant parts of the fcsr
register. So fflags is an alias for bits 0 to 4 of fcsr, and frm is
an alias for bits 5 to 7 of fcsr.
This means that a user can change the floating-point rounding mode
either, by writing a complete new value into fcsr, or by writing just
the rounding mode into frm.
How this impacts on GDB is like this: a target description could,
legitimately include all three registers, fcsr, fflags, and frm. The
QEMU target currently does this, and this makes sense. The target is
emulating the complete system, and has all three CSRs available, so
why not tell GDB about this.
In contrast, the RISC-V native Linux target only has access to the
fcsr. This is because the ptrace data structure that the kernel uses
for reading and writing floating point state only contains a copy of
the fcsr, after all, this one field really contains both the fflags
and frm fields, so why carry around duplicate data.
So, we might expect that the target description for the RISC-V native
Linux GDB would only contain the fcsr register. Unfortunately, this
is not the case. The RISC-V native Linux target uses GDB's builtin
target descriptions by calling riscv_lookup_target_description, this
will then add an fpu feature from gdb/features/riscv, either
32bit-fpu.xml or 64bit-fpu.xml. The problem, is that these features
include an entry for fcsr, fflags, and frm. This means that GDB
expects the target to handle reading and writing these registers. And
the RISC-V native Linux target currently doesn't.
In riscv_linux_nat_target::store_registers and
riscv_linux_nat_target::fetch_registers only the fcsr register is
handled, this means that, for RISC-V native Linux, the fflags and frm
registers always show up as <unavailable> - they are present in the
target description, but the target doesn't know how to access the
registers.
A final complication relating to these floating pointer CSRs is which
target description feature the registers appear in.
These registers are CSRs, so it would seem sensible that these
registers should appear in the CSR target description feature.
However, when I first added RISC-V target description support, I was
using a RISC-V simulator that didn't support any CSRs other than the
floating point related ones. This simulator bundled all the float
related CSRs into the fpu target feature. This didn't feel completely
unreasonable to me, and so I had GDB check for these registers in
either target feature.
In this commit I make some changes relating to how GDB handles the
three floating point CSR:
1. Remove fflags and frm from 32bit-fpu.xml and 64bit-fpu.xml. This
means that the default RISC-V target description (which RISC-V native
FreeBSD), and the target descriptions created for RISC-V native Linux,
will not include these registers. There's nothing stopping some other
target (e.g. QEMU) from continuing to include all three of these CSRs,
the code in riscv-tdep.c continues to check for all three of these
registers, and will handle them correctly if they are present.
2. If a target supplied fcsr, but does not supply fflags and/or frm,
then RISC-V GDB will now create two pseudo registers in order to
emulate the two missing CSRs. These new pseudo-registers do the
obvious thing of just reading and writing the fcsr register.
3. With the new pseudo-registers we can no longer make use of the GDB
register numbers RISCV_CSR_FFLAGS_REGNUM and RISCV_CSR_FRM_REGNUM.
These will be the numbers used if the target supplies the registers in
its target description, but, if GDB falls back to using
pseudo-registers, then new, unique numbers will be used. To handle
this I've added riscv_gdbarch_tdep::fflags_regnum and
riscv_gdbarch_tdep::frm_regnum, I've then updated the RISC-V code to
compare against these fields.
When adding the pseudo-register support, it is important that the
pseudo-register numbers are calculated after the call to
tdesc_use_registers. This is because we don't know the total number
of physical registers until after this call, and the psuedo-register
numbers must follow on from the real (target supplied) registers.
I've updated some tests to include more testing of the fflags and frm
registers, as well as adding a new test.
This commit adds a new function to the target description API within
GDB. This new function is not used in this commit, but will be used
in the next commit, I'm splitting it out into a separate patch for
easier review.
What I want to do in the next commit is check to see if a target
description supplied a particular register, however, the register in
question could appear in one of two possible features.
The new function allows me to ask the tdesc_arch_data whether a
register was found and assigned a particular GDB register number once
all of the features have been checked. I think this is a much simpler
solution than adding code such that, while checking each feature, I
spot if the register I'm processing is the one I care about.
No tests here as the new code is not used, but this code will be
exercised in the next commit.
On RISC-V the FCSR (float control/status register) is split into two
parts, FFLAGS (the flags) and FRM (the rounding mode). Both of these
two fields are part of the FCSR register, but can also be accessed as
separate registers in their own right. And so, we have three separate
registers, $fflags, $frm, and $fcsr, with the last of these being the
combination of the first two.
Here's how the bits of FCSR are split between FRM and FFLAGS:
,--------- FFLAGS
|---|
76543210 <----- FCSR
|-|
'--------------FRM
Here's how GDB currently displays these registers:
(gdb) info registers $fflags $frm $fcsr
fflags 0x0 RD:0 NV:0 DZ:0 OF:0 UF:0 NX:0
frm 0x0 FRM:0 [RNE (round to nearest; ties to even)]
fcsr 0x0 RD:0 NV:0 DZ:0 OF:0 UF:0 NX:0 FRM:0 [RNE (round to nearest; ties to even)]
Notice the 'RD' field which is present in both $fflags and $fcsr.
This field contains the value of the FRM field, which makes sense when
displaying the $fcsr, but makes no sense when displaying $fflags, as
the $fflags doesn't include the FRM field.
Additionally, the $fcsr already includes an FRM field, so the
information in 'RD' is duplicated. Consider this:
(gdb) set $frm = 0x3
(gdb) info registers $fflags $frm $fcsr │
fflags 0x0 RD:0 NV:0 DZ:0 OF:0 UF:0 NX:0
frm 0x3 FRM:3 [RUP (Round up towards +INF)]
fcsr 0x60 RD:3 NV:0 DZ:0 OF:0 UF:0 NX:0 FRM:3 [RUP (Round up towards +INF)]
See how the 'RD' field in $fflags still displays 0, while the 'RD' and
'FRM' fields in $fcsr show the same information.
The first change I propose in this commit is to remove the 'RD'
field. After this change the output now looks like this:
(gdb) info registers $fflags $frm $fcsr
fflags 0x0 NV:0 DZ:0 OF:0 UF:0 NX:0
frm 0x0 FRM:0 [RNE (round to nearest; ties to even)]
fcsr 0x0 NV:0 DZ:0 OF:0 UF:0 NX:0 FRM:0 [RNE (round to nearest; ties to even)]
Next, I spotted that the text that goes along with the 'FRM' field was
not wrapped in the i18n markers for internationalisation, so I added
those.
Next, I spotted that:
(gdb) set $frm=0x7
(gdb) info registers $fflags $frm $fcsr
fflags 0x0 RD:0 NV:0 DZ:0 OF:0 UF:0 NX:0
frm 0x7 FRM:3 [RUP (Round up towards +INF)]
fcsr 0xe0 RD:7 NV:0 DZ:0 OF:0 UF:0 NX:0 FRM:3 [RUP (Round up towards +INF)]
Notice that despite being a 3-bit field, FRM masks to 2-bits.
Checking the manual I can see that the FRM field is 3-bits, and is
defined for all 8 values. That GDB masks to 2-bits is just a bug I
think, so I've fixed this.
Finally, the 'FRM' text for value 0x7 is wrong. Currently we use the
text 'dynamic rounding mode' for value 0x7. However, this is not
really correct.
A RISC-V instruction can either encode the rounding mode within the
instruction, or a RISC-V instruction can choose to use a global,
dynamic rounding mode.
So, for the rounding-mode field of an _instruction_ the value 0x7
indicates "dynamic round mode", the instruction should defer to the
rounding mode held in the FRM field of the $fcsr.
But it makes no sense for the FRM of $fcsr to itself be set to
0x7 (dynamic rounding mode), and indeed, section 11.2, "Floating-Point
Control and Status Register" of the RISC-V manual, says that a value
of 0x7 in the $fcsr FRM field is invalid, and if an instruction has
_its_ round-mode set to dynamic, and the FRM field is also set to 0x7,
then an illegal instruction exception is raised.
And so, I propose changing the text for value 0x7 of the FRM field to
be "INVALID[7] (Dynamic rounding mode)". We already use the text
"INVALID[5]" and "INVALID[6]" for the two other invalid fields,
however, I think adding the extra "Dynamic round mode" hint might be
helpful.
I've added a new test that uses 'info registers' to check what GDB
prints for the three registers related to this patch. There is one
slight oddity with this test - for the fflags and frm registers, the
test accepts both the "normal" output (as described above), but also
allows these registers to be reported as '<unavailable>'.
The reason why I accept <unavailable> is that currently, the RISC-V,
native Linux target advertises these registers in its target
description, but then doesn't support reading or writing of these
registers, this results in the registers being reported as
unavailable.
A later patch in this series will address this issue, and will remove
this check for <unavailable>.
The following GDB behavior was also reported as a GDB bug in
https://sourceware.org/bugzilla/show_bug.cgi?id=28396
I will reiterate the problem a bit and give some more information here.
This patch closes the above mentioned bug.
The DWARF 5 standard 2.23 'Template Parameters' reads:
A template type parameter is represented by a debugging information
entry with the tag DW_TAG_template_type_parameter. A template value
parameter is represented by a debugging information entry with the tag
DW_TAG_template_value_parameter. The actual template parameter entries
appear in the same order as the corresponding template formal
parameter declarations in the source progam.
A type or value parameter entry may have a DW_AT_name attribute, whose
value is a null-terminated string containing the name of the
corresponding formal parameter.
So the DW_AT_name attribute for DW_TAG_template_type_parameter and
DW_TAG_template_value_parameter is optional.
Within GDB, creating a new symbol from some read DIE usually requires the
presence of a DW_AT_name for the DIE (an exception here is the case of
unnamed namespaces or the existence of a linkage name).
This patch makes the presence of the DW_AT_name for template value/type
tags optional, similar to the unnamed namespaces.
For unnamed namespaces dwarf2_name simply returns the constant string
CP_ANONYMOUS_NAMESPACE_STR '(anonymous namespace)'. For template tags a
case was added to the switch statement calling the
unnamed_template_tag_name helper. Within the scope of parent which
the template parameter is a child of, the helper counts the position
of the template tag within the unnamed template tags and returns
'<unnamedNUMBER>' where NUMBER is its position. This way we end up with
unique names within the respective scope of the function/class/struct
(these are the only currenltly supported template kinds within GDB and
usually the compilers) where we discovered the template tags in.
While I do not know of a way to bring GCC to emit template tags without
names there is one for clang/icpx. Consider the following example
template<typename A, typename B, typename C>
class Foo {};
template<typename, typename B, typename>
class Foo;
int main () {
Foo<double, int, float> f;
return 0;
}
The forward declaration for 'Foo' with the missing template type names
'A' and 'C' makes clang emit a bunch of template tags without names:
...
<2><43>: Abbrev Number: 3 (DW_TAG_variable)
<44> DW_AT_location : 2 byte block: 91 78 (DW_OP_fbreg: -8)
<47> DW_AT_name : (indirect string, offset: 0x63): f
<4b> DW_AT_decl_file : 1
<4c> DW_AT_decl_line : 8
<4d> DW_AT_type : <0x59>
...
<1><59>: Abbrev Number: 5 (DW_TAG_class_type)
<5a> DW_AT_calling_convention: 5 (pass by value)
<5b> DW_AT_name : (indirect string, offset: 0x74): Foo<double, int, float>
<5f> DW_AT_byte_size : 1
<60> DW_AT_decl_file : 1
<61> DW_AT_decl_line : 2
<2><62>: Abbrev Number: 6 (DW_TAG_template_type_param)
<63> DW_AT_type : <0x76>
<2><67>: Abbrev Number: 7 (DW_TAG_template_type_param)
<68> DW_AT_type : <0x52>
<6c> DW_AT_name : (indirect string, offset: 0x6c): B
<2><70>: Abbrev Number: 6 (DW_TAG_template_type_param)
<71> DW_AT_type : <0x7d>
...
Befor this patch, GDB would not create any symbols for the read template
tag DIEs and thus lose knowledge about them. Breaking at the return
statement and printing f's type would read
(gdb) ptype f
type = class Foo<double, int, float> [with B = int] {
<no data fields>
}
After this patch GDB does generate symbols from the DWARF (with their
artificial names:
(gdb) ptype f
type = class Foo<double, int, float> [with <unnamed0> = double, B = int,
<unnamed1> = float] {
<no data fields>
}
The same principle theoretically applies to template functions. Also
here, GDB would not record unnamed template TAGs but I know of no visual
way to trigger and test this changed behavior. Template functions do
not emit a '[with...]' list and their name generation also does not
suffer from template tags without names. GDB does not check whether or
not a template tag has a name in 'dwarf2_compute_name' and thus, the
names of the template functions are created independently of whether or
not the template TAGs have a DW_TAT_name attribute. A testcase has
been added in the gdb.dwarf2 for template classes and structs.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28396
When writing a dwarf testcase for some C++ code I wanted to use the
MACRO_AT_range which in turn uses the function_range proc in dwarf.exp
to extract the bounds of 'main'.
However, the macro failed as GDB prints the C++ 'main' with its
arguments as 'main(int, char**)' or 'main()'.
The reason for this is that in read.c::dwarf2_compute_name we call
c_type_print_args on C++ functions and append their arguments to the
function name. This happens to all C++ functions, but is only visible
when the function doesn't have a linkage name.
An example might make this more clear. Given the following code
>> cat c.cpp
int foo (int a, float b)
{
return 0;
}
int main (int argc, char **argv)
{
return 0;
}
which is legal in both languages, C and C++, and compiling it with
e.g. clang or gcc will make the disassemble command look like:
>> clang --version
clang version 10.0.0-4ubuntu1
...
>> clang -O0 -g ./c.cpp
>> gdb -q ./a.out -ex "start"
...
(gdb) disassemble main
Dump of assembler code for function main(int, char**):
0x0000000000401120 <+0>: push %rbp
0x0000000000401121 <+1>: mov %rsp,%rbp
...
0x0000000000401135 <+21>: ret
End of assembler dump.
(gdb) disassemble foo
Dump of assembler code for function _Z3fooif:
0x0000000000401110 <+0>: push %rbp
0x0000000000401111 <+1>: mov %rsp,%rbp
...
0x000000000040111f <+15>: ret
End of assembler dump.
Note, that main is emitted with its arguments while for foo the linkage
name is being printed, as also visible in its DWARF:
>> objdump ./a.out --dwarf=info | grep "foo" -A3 -B3
<2b> DW_AT_low_pc : 0x401110
<33> DW_AT_high_pc : 0x10
<37> DW_AT_frame_base : 1 byte block: 56 (DW_OP_reg6 (rbp))
<39> DW_AT_linkage_name: (indirect string, offset: 0x39): _Z3fooif
<3d> DW_AT_name : (indirect string, offset: 0x42): foo
<41> DW_AT_decl_file : 1
<42> DW_AT_decl_line : 1
<43> DW_AT_type : <0x9a>
Now, let's rename the C++ file and compile it as C:
>> mv c.cpp c.c
>> clang -O0 -g ./c.c
>> gdb -q ./a.out -ex "start'
...
(gdb) disassemble main
Dump of assembler code for function main:
0x0000000000401120 <+0>: push %rbp
0x0000000000401121 <+1>: mov %rsp,%rbp
...
0x0000000000401135 <+21>: ret
End of assembler dump.
(gdb) disassemble foo
Dump of assembler code for function foo:
0x0000000000401110 <+0>: push %rbp
0x0000000000401111 <+1>: mov %rsp,%rbp
...
0x000000000040111f <+15>: ret
End of assembler dump.
Note, for foo we did not get a linkage name emitted in DWARF, so
it is printed by its name:
>> objdump --dwarf=info ./a.out | grep foo -A3 -B3
<2b> DW_AT_low_pc : 0x401110
<33> DW_AT_high_pc : 0x10
<37> DW_AT_frame_base : 1 byte block: 56 (DW_OP_reg6 (rbp))
<39> DW_AT_name : (indirect string, offset: 0x37): foo
<3d> DW_AT_decl_file : 1
<3e> DW_AT_decl_line : 1
<3f> DW_AT_prototyped : 1
To make the macro and proc work with C++ as well, an optional argument
list was added to the regex matching the function name in the
disassemble command in function_range. This does not change any used
behavior as currently, there exists no C++ test using the proc
function_range.
Signed-off-by: Nils-Christian Kempke <nils-christian.kempke@intel.com>
The call to debuginfod_debuginfo_query in elf_symfile_read is given
objfile->original_name as the filename to print when downloading the
objfile's debuginfo.
In some cases original_name is prefixed with gdb's working directory
even though the objfile is not located in the working directory. This
causes debuginfod to display the wrong path of the objfile during a download.
Fix this by using the objfile's bfd filename instead.
GDB overwrites Python's sys.stdout and sys.stderr, but does not
properly implement the 'flush' method -- it only ever will flush
stdout. This patch fixes the bug. I couldn't find a straightforward
way to write a test for this.
The print_one_detail_ranged_breakpoint has been renamed to
ranged_breakpoint::print_one_detail in this commit:
commit ec45bb676c
Date: Sat Jan 15 16:34:51 2022 -0700
Convert ranged breakpoints to vtable ops
So their comments should be updated as well.
When running the included test-case, we run into:
...
(gdb) break _start^M
read.h:309: internal-error: set_length: \
Assertion `m_length == length' failed.^M
...
The problem is that while there are two CUs:
...
$ readelf -wi debug-names-missing-cu | grep @
Compilation Unit @ offset 0x0:
Compilation Unit @ offset 0x2d:
...
the CU table in the .debug_names section only contains the first one:
...
CU table:
[ 0] 0x0
...
The incomplete CU table makes create_cus_from_debug_names_list set the size of
the CU at 0x0 to the actual size of both CUs combined.
This eventually leads to the assert, when we read the actual size from the CU
header.
While having an incomplete CU table in a .debug_names section is incorrect,
we need a better failure mode than asserting.
The easiest way to fix this is to set the length to 0 (meaning: unkown) in
create_cus_from_debug_names_list.
This makes the failure mode to accept the incomplete CU table, but to ignore
the missing CU.
It would be nice to instead reject the .debug_names index, and build a
complete CU list, but the point where we find this out is well after
dwarf2_initialize_objfile, so it looks rather intrusive to restart at that
point.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29453
