Looking at the ARM disassembler output, every comment seems to start
with a ';' character, so I assumed this was the correct character to
start an assembler comment.
I then spotted a couple of places where there was no ';', but instead,
just a '@' character. I thought that this was a case of a missing
';', and proposed a patch to add the missing ';' characters.
Turns out I was wrong, '@' is actually the ARM assembler comment
character, while ';' is the statement separator. Thus this:
nop ;@ comment
is two statements, the first is the 'nop' instruction, while the
second contains no instructions, just the '@ comment' comment text.
This:
nop @ comment
is a single 'nop' instruction followed by a comment. And finally,
this:
nop ; comment
is two statements, the first contains the 'nop' instruction, while the
second contains the instruction 'comment', which obviously isn't
actually an instruction at all.
Why this matters is that, in the next commit, I would like to add
libopcodes syntax styling support for ARM.
The question then is how should the disassembler style the three cases
above?
As '@' is the actual comment start character then clearly the '@' and
anything after it can be styled as a comment. But what about ';' in
the second example? Style as text? Style as a comment?
And the third example is even harder, what about the 'comment' text?
Style as an instruction mnemonic? Style as text? Style as a comment?
I think the only sensible answer is to move the disassembler to use
'@' consistently as its comment character, and remove all the uses of
';'.
Then, in the next commit, it's obvious what to do.
There's obviously a *lot* of tests that get updated by this commit,
the only actual code changes are in opcodes/arm-dis.c.
A user noticed that TYPE_CODE_FIXED_POINT was not exported by the gdb
Python layer. This patch fixes the bug, and prevents future
occurences of this type of bug.
Currently the resource checking is done if REPARSE is true. The hardware
watchpoint resource checking in update_watchpoint needs to be redone on
each call to function update_watchpoints as the value chain may have
changed. The number of hardware registers needed for a watchpoint can
change if the variable being watched changes. This situation occurs in
this test when watching variable **global_ptr_ptr. Initially when the
watch command is issued, only two addresses need to be watched as
**global_ptr_ptr has not yet been initialized. Once the value of
**global_ptr_ptr is initialized the locations to be tracked increase to
three addresses. However, update_watchpoints is not called again with
REPARSE set to 1 to force the resource checking to be redone. When the
test is run on Power 10, an internal gdb error occurs when the PowerPC
routine tries to setup the three hardware watchpoint address since the hw
only has two hardware watchpoint registers. The error occurs because the
resource checking was not redone in update_watchpoints after
**global_ptr_ptr changed.
The following descibes the situation in detail that occurs on Power 10 with
gdb running on the binary for gdb.base/watchpoint.c.
1 break func4
2 run
3 watch *global_ptr
4 next execute source code: buf[0] = 3;
5 next execute source code: global_ptr = buf;
6 next execute source code: buf[0] = 7;
7 delete 2 (delete watch *global_ptr)
8 watch **global_ptr_ptr
9 next execute source code: buf[1] = 5;
10 next global_ptr_ptr = &global_ptr;
11 next buf[0] = 9;
In step 8, the the watch **global_ptr_prt command calls update_watchpoint
in breakpoint.c with REPARSE set to 1. The function update_watchpoint
calls can_use_hardware_watchpoint to see if there are enough
resources available to add the watchpoint since REPARSE is set to 1. At
this point, **global_ptr_ptr has not been initialized so only two addresses
are watched. The val_chain contains the address for **global_ptr_ptr and 0
since **global_ptr_ptr has not been initialized. The update_watchpoint
updates the breakpoint list as follows:
breakpoint 0
loc 0: b->address = 0x100009c0
breakpoint 1
loc 1: b->address = 0x7ffff7f838a0
breakpoint 2
loc 2: b->address = 0x7ffff7b7fc54
breakpoint 3
loc 3: b->address = 0x7ffff7a5788c
breakpoint 4
loc 4: b->address = 0x0 <-- location pointed to by global_ptr_ptr
loc 5: b->address = 0x100200b8 <-- global_ptr_ptr watchpoint
breakpoint 5
loc 6: b->address = 0x7ffff7b7fc54
In step 10, the next command executes the source code
global_ptr_ptr = &global_ptr. This changes the set of locations to be
watched for the watchpoint **global_ptr_prt. The list of addresses for the
breakpoint consist of the address for global_ptr_prt, global_ptr and buf.
The breakpoint list gets updated by update_watchpoint as follows:
breakpoint 0
loc 0: b->address = 0x100009c0
breakpoint 1
loc 1: b->address = 0x7ffff7f838a0
breakpoint 2
loc 2: b->address = 0x7ffff7b7fc54
breakpoint 3
loc 3: b->address = 0x7ffff7a5788c
breakpoint 4
loc 4: b->address = 0x10020050 buf
loc 5: b->address = 0x100200b0 watch *global_ptr
loc 6: b->address = 0x100200b8 watch **global_ptr_ptr
breakpoint 5
loc 7: b->address = 0x7ffff7b7fc54
breakpoint 6
However, the hardware resource checking was not redone because
update_breakpoint was called with REPARSE equal to 0.
Step 11, execute the third next command. The function
ppc_linux_nat_target::low_prepare_to_resume() attempts a ptrace
call to setup each of the three address for breakpoint 4. The slot
value returned for the third ptrace call is -1 indicating an error
because there are only two hardware watchpoint registers available on
Power 10.
This patch removes just the statement "if (reparse)" in function
update_watchpoint to force the resources to be rechecked on every call to
the function. This ensures that any changes to the val_chain resulting
in needing more resources then available will be caught.
The patch has been tested on Power 8, Power 10 and X86-64. Note the patch
has no effect on Power 9 since hardware watchpoint support is disabled on
that processor.
Test gdb.reverse/pipe-reverse.exp fails on Power 10 running the fedora 36
distro. The gdb record error message is:
Process record and replay target doesn't support syscall number 317.
System call 317 on PowerPC maps to the pipe2 system call.
This patch adds support for the missing pipe2 system call for PowerPC.
Patch fixes the test failure in gdb.reverse/pipe-reverse.exp.
The patch has been tested on Power 10 with no regression failures.
Earlier tidying still missed an opportunity: There's no need for the
"anyimm" static variable. Instead of using it in the loop to mask
"allowed" (which is necessary to satisfy operand_type_or()'s assertions)
simply use "mask", requiring it to be calculated first. That way the
post-loop masking by "mask" ahead of the operand_type_all_zero() can be
dropped.
When reading/writing arbitrary data to the system's memory, the unsigned
char pointer type doesn't make that much sense. Switch it to void so we
align a bit with standard C library read/write functions, and to avoid
having to sprinkle casts everywhere.
Silence GCC 12 warning on tc-i386.c:
gas/config/tc-i386.c: In function ‘md_assemble’:
gas/config/tc-i386.c:5039:16: error: too many arguments for format [-Werror=format-extra-args]
5039 | as_warn (_("only support RIP-relative address"), i.tm.name);
* config/tc-i386.c (md_assemble): Print mnemonic in RIP-relative
warning.
* estsuite/gas/i386/x86-64-prefetchi-warn.l: Updated.
This changes gdbarch to use the "predefault" to initialize its members
inline. This required changing a couple of the Value instantiations
to avoid a use of "gdbarch" during initialization, but on the whole I
think this is better -- it removes a hidden ordering dependency.
PR c++/29243 points out that "info func" on a certain C++ executable
will cause an infinite loop in gdb.
I tracked this down to a bug introduced by commit 6b5a7bc76 ("Handle
member pointers directly in generic_value_print"). Before this
commit, the C++ code to print a member pointer would wind up calling
value_print_scalar_formatted; but afterward it simply calls
generic_value_print and gets into a loop.
This patch restores the previous behavior and adds a regression test.
Clang attempts to minimize the size of the debug-info by not adding
complete information about types that aren't constructed in a given
file. Specifically, this meant that there was minimal information about
class B in the test gdb.cp/converts.exp. To fix this, we just need to
construct any object of type B in that file.
Approved-By: Andrew Burgess <aburgess@redhat.com>
When running gdb.cp/ptype-flags.exp using Clang, we get an unexpected
failure when printing the type of a class with an internal typedef. This
happens because Clang doesn't add accessibility information for typedefs
inside classes (see https://github.com/llvm/llvm-project/issues/57608
for more info). To help with Clang testing, an XFAIL was added to this
test.
Ref: https://gcc.gnu.org/wiki/DebugFissionDWP?action=recall&rev=3
Fuzzers have found a weakness in the code stashing pool section
entries. With random nonsensical values in the index entries (rather
than each index pointing to its own set distinct from other sets),
it's possible to overflow the space allocated, losing the NULL
terminator. Without a terminator, find_section_in_set can run off the
end of the shndx_pool buffer. Fix this by scanning the pool directly.
binutils/
* dwarf.c (add_shndx_to_cu_tu_entry): Delete range check.
(end_cu_tu_entry): Likewise.
(process_cu_tu_index): Fill shndx_pool by directly scanning
pool, rather than indirectly from index entries.
Clang generates a warning if there is a function declaration/definition
with zero arguments. Such declarations/definitions without a prototype (an
argument list) are deprecated forms of indefinite arguments
("-Wdeprecated-non-prototype"). On the default configuration, it causes a
build failure (unless "--disable-werror" is specified).
But there is another issue. This function declaration in sim/sh/interp.c
is completely redundant. This commit just removes that declaration.
The variable "list" is only initialized when arg1 > 0 and when arg1 == 0,
an uninitialized value is passed to translate_endian_h2t function.
Although this behavior is harmless, this commit adds initialization to avoid
a GCC warning ("-Wmaybe-uninitialized").
Clang generates a warning if an argument is passed to a function without
prototype (zero arguments, even without (void)). Such calls are deprecated
forms of indefinite arguments passing ("-Wdeprecated-non-prototype").
On the default configuration, it (somehow) doesn't cause a build failure but
a warning is generated.
But because the cause is the same as the issue the author fixed in
"sim/erc32: Use int32_t as event callback argument", it would be better to
fix it now to prevent problems in the future.
To fix the issue, this commit makes struct irqcall to use int32_t as a
callback (callback) argument of an IRQ.
Clang generates a warning if an argument is passed to a function without
prototype (zero arguments, even without (void)). Such calls are deprecated
forms of indefinite arguments passing ("-Wdeprecated-non-prototype").
On the default configuration, it causes a build failure (unless
"--disable-werror" is specified).
To fix that, this commit makes struct evcell to use int32_t as a callback
(cfunc) argument of an event. int32_t is chosen because "event" function
accepts "int32_t arg".
Clang generates a warning if there is a function declaration/definition
with zero arguments. Such declarations/definitions without a prototype (an
argument list) are deprecated forms of indefinite arguments
("-Wdeprecated-non-prototype"). On the default configuration, it causes a
build failure (unless "--disable-werror" is specified).
This commit replaces () with (void) to avoid this warning.
When running test-case gdb.server/multi-ui-errors.exp on target board
remote-gdbserver-on-localhost.exp, I run into:
...
(gdb) PASS: gdb.server/multi-ui-errors.exp: connect to gdbserver
continue^M
Continuing.^M
PASS: gdb.server/multi-ui-errors.exp: continue - extra UI
Remote debugging from host ::1, port 35466^M
FAIL: gdb.server/multi-ui-errors.exp: ensure inferior is running
...
The problem is that the target board uses ssh -T, which fails to guarantee
that output from the inferior will be available.
Fix this by copying proc ${board}_spawn from local-remote-host.exp, which
ensures using ssh -t. [ It would be nice to define an ssh base board to
get rid of the copies, but I'm not addressing that in this commit. ]
Likewise for target board remote-stdio-gdbserver.exp.
Tested on x86_64-linux.
With test-case gdb.server/multi-ui-errors.exp and host board
local-remote-host-notty, I run into:
...
(gdb) PASS: gdb.server/multi-ui-errors.exp: interact with GDB's main UI
Executing on target: kill -9 29666 (timeout = 300)
builtin_spawn -ignore SIGHUP kill -9 29666^M
echo^M
Remote connection closed^M
(gdb) (gdb) FAIL: gdb.server/multi-ui-errors.exp: \
main UI, prompt after gdbserver dies (timeout)
...
In contrast, with local-remote-host (so, everything the same but editing off):
...
(gdb) PASS: gdb.server/multi-ui-errors.exp: interact with GDB's main UI
Executing on target: kill -9 31245 (timeout = 300)
builtin_spawn -ignore SIGHUP kill -9 31245^M
Remote connection closed^M
(gdb) echo^M
(gdb) PASS: gdb.server/multi-ui-errors.exp: main UI, prompt after gdbserver dies
...
The test-case issues a kill, which results in a "Remote connection closed"
message and a prompt.
The problem is that the prompt is not consumed, so the subsequent echo may be
issued before that prompt, which causes a mismatch when matching the result
of the echo.
Fix this by consuming the "Remote connection closed" message and prompt.
Tested on x86_64-linux.
With test-case gdb.server/multi-ui-errors.exp we see:
...
(gdb) PASS: multi-ui-errors.exp: main UI, prompt after gdbserver dies
continue^M
Continuing.^M
echo^M
(gdb) PASS: multi-ui-errors.exp: extra UI, prompt after gdbserver dies
...
The continue is issued earlier in the test-case, but the output has not been
consumed, which makes it show up much later.
Consume the continue output asap, to make it clear when the continue is issued:
...
(gdb) PASS: gdb.server/multi-ui-errors.exp: connect to gdbserver
continue^M
Continuing.^M
PASS: gdb.server/multi-ui-errors.exp: continue - extra UI
...
Tested on x86_64-linux.
The usage for board remote-stdio-gdbserver.exp is advertised as:
...
# bash$ make check RUNTESTFLAGS="--target_board=remote-stdio-gdbserver \
# REMOTE_USERNAME=... REMOTE_HOSTNAME=... REMOTE_PORTNUM=... \
# [REMOTE_TMPDIR=${remote_dir}] [GDBSERVER=${remote_gdbserver}]"
...
but when adding REMOTE_PORTNUM=22, I run into:
...
Running stop-reply-no-thread-multi.exp ...
ERROR: tcl error sourcing stop-reply-no-thread-multi.exp.
ERROR: couldn't execute "/usr/bin/ssh -p22": no such file or directory
while executing
"builtin_spawn {/usr/bin/ssh -p22} -l vries localhost {/usr/bin/gdbserver \
--once localhost:2346 \
/home/vries/gdb_versions/devel/build/gdb/testsuite/outp..."
...
Fix this by simply removing REMOTE_PORTNUM.
Tested on x86_64-linux.
Because of a Libiberty hack, getopt on GNU libc (2.25 or earlier) is
currently unusable on sim, causing a regression on CentOS 7.
This is caused as follows:
1. If HAVE_DECL_GETOPT is defined (getopt declaration with known prototype
is detected while configuration), a declaration of getopt in
"include/getopt.h" is suppressed.
The author started to define HAVE_DECL_GETOPT in sim with the commit
340aa4f6872c ("sim: Check known getopt definition existence").
2. GNU libc (2.25 or earlier)'s <unistd.h> includes <getopt.h> with a
special purpose macro defined to declare only getopt function but due
to include path (not tested while configuration), it causes <unistd.h>
to include Libiberty's "include/getopt.h".
3. If both 1. and 2. are satisfied, despite that <unistd.h> tries to
declare getopt by including <getopt.h>, "include/getopt.h" does not do
so, causing getopt function undeclared.
Getting rid of "include/getopt.h" (e.g. renaming this header file) is the
best solution to avoid hacking but as a short-term solution, this commit
replaces getopt with getopt_long under sim/.
And do the sanity check in a way that can't overflow.
* pef.c (bfd_pef_parse_function_stubs): Sanity check header
imported_library_count and total_imported_symbol_count before
allocating memory.
objcopy copy_object may omit the call to bfd_copy_private_bfd_data for
various conditions deemed non-fatal, in which case obj_som_exec_data
will be NULL for the output file.
* som.c (som_finish_writing): Don't dereference NULL
obj_som_exec_data.
Before figuring out the suppress rule of mapping symbol with architecture
(changed back to $x), always generate them at the start of the sections.
gas/
* config/tc-riscv.c (need_arch_map_symbol): Removed.
(riscv_mapping_state): Updated.
(riscv_check_mapping_symbols): Updated.
* testsuite/gas/riscv/mapping-non-arch.d: Removed.
* testsuite/gas/riscv/mapping-non-arch.s: Likewise.
We don't support instructions longer than 64-bits yet. Still, we can
modify validate_riscv_insn function to prevent unexpected behavior by
limiting the "length" of an instruction to 64-bit (or less).
gas/ChangeLog:
* config/tc-riscv.c (validate_riscv_insn): Fix function
description comment based on current spec. Limit instruction
length up to 64-bit for now. Make sure that required_bits does
not corrupt even if unsigned long long is longer than 64-bit.
Some versions of gcc warn by default about shadowed outer-scope
declarations. This affects frag_align_code, which is declared in
frags.h. Rename the offending function parameter. While there also
switch to using true/false at the function call sites.
Commit 40f1a1a4564b ("RISC-V: Output mapping symbols with ISA string.")
caused a build failure on GCC 12 as follows:
make[3]: Entering directory '$(builddir)/gas'
CC config/tc-riscv.o
In file included from $(srcdir)/gas/config/tc-riscv.c:23:
$(srcdir)/gas/as.h: In function ‘make_mapping_symbol’:
$(srcdir)/gas/as.h:123:15: error: ‘buff’ may be used uninitialized [-Werror=maybe-uninitialized]
123 | #define xfree free
| ^~~~
$(srcdir)/gas/config/tc-riscv.c:487:9: note: ‘buff’ was declared here
487 | char *buff;
| ^~~~
cc1: all warnings being treated as errors
make[3]: *** [Makefile:1425: config/tc-riscv.o] Error 1
This is caused by a false positive of "maybe uninitialized" variable
detection (-Wmaybe-uninitialized). To avoid this error, this commit
initializes the local variable buff to NULL first in all cases.
gas/ChangeLog:
* config/tc-riscv.c (make_mapping_symbol): Initialize variable
buff with NULL to avoid build failure caused by a GCC's false
positive of maybe uninitialized variable detection.
In gdb/nat/linux-btrace.c:btrace_this_cpu() we initialize the cpu
structure given to the libipt btrace decoder.
We only consider the extended model field for family 0x6 and forget about
family 0xf and we don't consider the extended family field. Fix it.
