This test tests passing arguments made of exactly two single-quotes
('') or a single newline character through the --args argument of GDB.
For some reason, GDB adds some extra single quotes when transmitting the
arguments to GDBserver. This produces some FAILs when testing with the
native-extended-gdbserver board:
FAIL: gdb.base/args.exp: argv[2] for one empty (with single quotes)
FAIL: gdb.base/args.exp: argv[2] for two empty (with single quotes)
FAIL: gdb.base/args.exp: argv[3] for two empty (with single quotes)
FAIL: gdb.base/args.exp: argv[2] for one newline
FAIL: gdb.base/args.exp: argv[2] for two newlines
FAIL: gdb.base/args.exp: argv[3] for two newlines
This is documented as PR 27989. Add some appropriate KFAILs.
gdb/testsuite/ChangeLog:
* gdb.base/args.exp: Check target, KFAIL if remote.
(args_test): Add parameter and use it.
Change-Id: I49225d1c7df7ebaba480ebdd596df80f8fbf62f0
All tests in this file append to GDBFLAGS instead of overwriting it,
except the last two. I noticed because when testing with the
native-extended-remote board, it removes the "set sysroot" argument, and
it causes the test to be very long to run, due to big glibc debug info
being read through the remote target.
I think this oddity is due to a race condition between these two
commits:
[1] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=c22261528c50f7760dd6a2e29314662b377eebb4
[2] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=6b8ce727297b1e40738e50f83a75881b290fe6a6
The first one added the two tests. The second one changes the test to
append to GDBFLAGS instead of overwriting it. But the second one was
probably written before the first one was it, so missed the new tests.
Change those two tests to be like the others.
gdb/testsuite/ChangeLog:
* gdb.base/args.exp: Use $old_gdbflags in all tests.
Change-Id: I531276125ecb70e80f52adbd320ebb85b0c8eba0
Use save_vars instead of manually saving/restoring. This ensures that
if anything throws an error, GDBFLAGS will be correctly restored.
Remove the global GDBFLAGS declaration at the top, it's not necessary.
gdb/testsuite/ChangeLog:
* gdb.base/args.exp: Use save_vars.
Change-Id: I3a45e4fc1635ec0212de2415040f91eecaf4a057
Currently, when the focus is on the command window, we disable the
keypad. This means that when the command window has the focus, keys
such as up/down/home/end etc. are not processed by curses, and their
escape sequences go straight to readline.
A side effect of disabling keypad mode is that wgetch no longer
processes mouse escape sequences, with the end result being the mouse
doesn't work, and worse, the raw mouse escape sequences are printed on
the terminal.
This commit makes the TUI command window support the mouse as well, by
always enabling the keypad, and then to avoid losing support for
up/down browsing the command history, home/end/left/right moving the
cursor position, etc., we forward those keys as raw escape sequences
to readline. Note we don't make an effort to pass down to readline
all keys returned by curses, only the common ones that readline
understands by default. Given users can specify their own readline
bindings (inputrc file, bind utility), this approach is good in
practice, though not 100% transparent or perfect.
Note that the patch makes it so that CTLC-L is always passed to
readline even if the command window does not have the focus. It was
simpler to implement that way, and it just seems correct to me. I
don't know of a reason we shouldn't do that.
The patch improves the TUI behavior in a related way. Now we can pass
special keys to readline irrespective of which window has the focus.
First, we try to dispatch the key to a window, via
tui_displatch_ctrl_char. If the key is dispatched, then we don't pass
it to readline. E.g., pressing "up" when you have the source window
in focus results in scrolling the source window, and nothing else. If
however, you press ctrl-del instead, that results in killing the next
word in the command window, no matter which window has has focus.
Before, it would only work if you had the command window in focus.
Similarly, ctrl-left/ctrl-right to move between words, etc.
Similarly, the previous spot where we handled mouse events was
incorrect. It was never reached if the window with focus can't
scroll, which is the case for the command window. Mouse scrolling
affects the window under the mouse cursor, not the window in focus.
We now always try to dispatch mouse events.
One last bit in the patch -- now if we don't recognize the non-8-bit
curses key, then we don't pass it down to readline at all. Before
that would result in bogus characters in the input line.
gdb/ChangeLog:
yyyy-mm-dd Pedro Alves <pedro@palves.net>
* tui/tui-io.c (tui_dispatch_mouse_event): New, factored out from
...
(tui_dispatch_ctrl_char): ... this. Move CTRL-L handling to
tui_getc_1.
(cur_seq, start_sequence): New.
(tui_getc_1): Pass key escape sequences for curses control keys to
readline. Handle mouse and ctrl-l here.
(tui_resize_all): Disable/reenable the keypad if the command
window has the focus too.
* tui/tui-win.c (tui_set_focus_command): Don't change keypad
setting.
* tui/tui.c (tui_rl_other_window): Don't change keypad setting.
Change-Id: Ie0a7d849943cfb47f4a6589e1c73341563740fa9
AIX ld expects the section contents for relocations BFD_RELOC_PPC_TLSM
or BFD_RELOC_PPC64_TLSM to be zero.
Actually, it seems to be the case for all the TLS relocations
generated by AIX assembly but only these two are mandatory.
* config/tc-ppc.c (md_apply_fix): Adjust addend to nullify
section contents for BFD_RELOC_PPC_TLSM or
BFD_RELOC_PPC64_TLSM.
* testsuite/gas/ppc/xcoff-tlsm-32.d: New test.
* testsuite/gas/ppc/xcoff-tlsm-64.d: New test.
* testsuite/gas/ppc/xcoff-tlsm.s: New test.
* testsuite/gas/ppc/aix.exp: Run new tests.
Use INLINE2 instead of INLINE to fix builds when -O0 are used -- the
latter define is omitted at -O0 levels while the former is always
set to inline. These helper funcs are used by defines in here but
the defines aren't always called.
The sim-basics.h is too big and includes too many things. This leads
to some arch's sim-main.h having circular loop issues with defs, and
makes it hard to separate out common objects from arch-specific defs.
By splitting up sim-basics.h and killing off sim-main.h, it'll make
it easier to separate out the two.
Start with splitting out sim/callback.h.
Use GDB's silent-rules.mk to make some rules silent by default. These
rules cover most of what is built in sim/.
gdb/ChangeLog:
* silent-rules.mk (ECHO_CCLD, ECHO_AR, ECHO_RANLIB): New.
sim/ChangeLog:
* common/Make-common.in (COMPILE, libsim.a, run$(EXEEXT),
gentmap.o, gentmap): Make rules silent.
Change-Id: Idf9ba5beaee10c7c614859ace5fbdcd1de0287db
The documentation for 'record function-call-history' mentions lines instead
of functions when talking about the number of functions printed, as currently
there is only one line printed per function. Yet the code actually handles
this on function granularity not on line basis. Future patches will
extend the number of lines printed per function. This also makes it
consistent with the 'record instruction-history' command, where multiple
lines can be printed per instruction.
gdb/doc/ChangeLog:
2021-06-16 Felix Willgerodt <felix.willgerodt@intel.com>
* gdb.texinfo (Process Record and Replay): Stop mentioning lines
for "record function-call-history" and
"set record function-call-history-size".
This is another attempt at fixing the problem described in commit 4cf88725da
"[gdb/symtab] Fix infinite recursion in dwarf2_cu::get_builder()", which was
reverted in commit 3db19b2d72.
First off, some context.
A DWARF CU can be viewed as a symbol table: toplevel children of a CU DIE
represent symbol table entries for that CU. Furthermore, there is a
hierarchy: a symbol table entry such as a function itself has a symbol table
containing parameters and local variables.
The dwarf reader maintains a notion of current symbol table (that is: the
symbol table a new symbol needs to be entered into) in dwarf2_cu member
list_in_scope.
A problem then presents itself when reading inter-CU references:
- a new symbol read from a CU B needs to be entered into the symbol table of
another CU A.
- the notion of current symbol table is tracked on a per-CU basis.
This is addressed in inherit_abstract_dies by temporarily overwriting the
list_in_scope for CU B with the one for CU A.
The current symbol table is one aspect of the current dwarf reader context
that is tracked, but there are more, f.i. ones that are tracked via the
dwarf2_cu member m_builder, f.i. m_builder->m_local_using_directives.
A similar problem exists in relation to inter-CU references, but a different
solution was chosen:
- to keep track of an ancestor field in dwarf2_cu, which is updated
when traversing inter-CU references, and
- to use the ancestor field in dwarf2_cu::get_builder to return the m_builder
in scope.
There is no actual concept of a CU having an ancestor, it just marks the most
recent CU from which a CU was inter-CU-referenced. Consequently, when
following inter-CU references from a CU A to another CU B and back to CU A,
the ancestors form a cycle, which causes dwarf2_cu::get_builder to hang or
segfault, as reported in PR26327.
ISTM that the ancestor implementation is confusing and fragile, and should
go. Furthermore, it seems that keeping track of the m_builder in scope can be
handled simply with a per-objfile variable.
Fix the hang / segfault by:
- keeping track of the m_builder in scope using a new variable
per_obj->sym_cu, and
- using it in dwarf2_cu::get_builder.
Tested on x86_64-linux (openSUSE Leap 15.2), no regressions for config:
- using default gcc version 7.5.0
(with 5 unexpected FAILs)
- gcc 10.3.0 and target board
unix/-flto/-O0/-flto-partition=none/-ffat-lto-objects
(with 1000 unexpected FAILs)
gdb/ChangeLog:
2021-06-16 Tom de Vries <tdevries@suse.de>
PR symtab/26327
* dwarf2/cu.h (dwarf2_cu::ancestor): Remove.
(dwarf2_cu::get_builder): Declare and move ...
* dwarf2/cu.c (dwarf2_cu::get_builder): ... here. Use sym_cu instead
of ancestor. Assert return value is non-null.
* dwarf2/read.c (read_file_scope): Set per_objfile->sym_cu.
(follow_die_offset, follow_die_sig_1): Remove setting of ancestor.
(dwarf2_per_objfile): Add sym_cu field.
Unlike for .dc.? the parsing here failed to skip the colon before
calling hex_float(). To avoid both variants of parsing going out of sync
again, introduce a helper used by both.
While the logic in fixup_segment() so far was off by 1 for fixups
dealing with quantities without known signedness, thus failing to report
an overflow when e.g. a byte-sized resulting value is -0x100, the logic
in emit_expr_with_reloc() reported an overflow even on large negative
values when the respective positive ones would not be warned
about, and when fixup_segment() wouldn't do so either. Such diagnostics
all ought to follow a similar pattern of what value range is acceptable.
(If expressions' X_unsigned was reliably set, emit_expr_with_reloc()'s
checking might make sense to tighten based on that flag.)
Note that with commit 80aab57939 ("Changes to let cons handle bignums
like general expressions") having added handling of nbytes >
sizeof(valueT) in the O_constant case, converting to O_big, the setting
to zero of "hibit" had become dead. With "hibit" no longer used, this
code now gets dropped altogether. But additionally a respective know()
gets inserted.
The compiler doesn't like passing non-constant strings to printf
functions, so tweak the code to always pass one in. This code is
a little more verbose, but it's probably the same performance.
The macro usage is a bit ugly, but maybe less than copying &
pasting the extended conditional format logic.
This model uses unsigned char buffers, but this temporary pointer is
declared as signed. Switch it to unsigned since it's just a temporary
variable to hold the new pointer.
Rather than re-invent endian defines, as well as maintain our own list
of OS & arch-specific includes, punt all that logic in favor of the bfd
ones already set up and maintained elsewhere. We already rely on the
bfd library, so leveraging the endian aspect should be fine.
This was done for all the other ports years ago, so catch ppc up.
The --enable-sim-hostendian flag was purely so people had an escape route
for when cross-compiling. This is because historically, AC_C_BIGENDIAN
did not work in those cases. That was fixed a while ago though, so we can
require that macro everywhere now and simplify a good bit of code.
This was done for all the other ports years ago, so catch ppc up.
All of the settings in here are handled by the common top-level
config.h, so drop the individual arch-config.h files entirely.
This will also help guarantee that we don't add any new arch
specific defines that would affect common code which will help
with the effort of unifying them.
The testsuite uses the output to determine whether BFD64 is in effect.
--x32 is supported for ELF only; don't advertise it for PE/COFF. --64 is
also supported for Mach-O; advertise it. Adjust the testsuite's BFD64
check accordingly.
Also replace "code" by "object", since it's the object format that the
options primarily control. It's also _initial_ code bitness, but this
can be changed by directives.
First of all eliminate the disp{16,32,32s} static variables, which are
used solely for setting a temporary variable in build_modrm_byte(). The
effect there can be had without use of such a temporary and without
operand_type_or(), by just setting the single bit each that needs
setting.
Then use operand_type_and_not(..., anydisp) when all dispNN bits want
clearing together.
Applying a mask with all bits set (or its inverse, with hence all bits
clear) won't alter the result (or won't trigger the warning). Re-arrange
the code to eliminate two more of the somewhat odd (2 << width_minus_1)
constructs.
Certain disp values may trigger "... shortened to ..." warnings when
equivalent imm ones don't. In some of the cases there are also
differences (for non-64-bit code) between BFD64 and !BFD64 builds. The
resulting encodings (i.e. use [or not] of the shorter disp8 / imm8
forms) are also different in some cases. Make this handling consistent.
Note that using equivalent 16-bit mode displacements / immediates
continues to expose entirely different behavior (see the disp-imm-16
testcase added by an earlier patch). This may want to be the subject of
further changes, but it'll then quickly become obvious that e.g. keying
use of extend_to_32bit_address() to non-64-bit mode isn't appropriate
either: Once we allow wrapping operands, we would better do so
consistently, in which case all of this would need to become dependent
upon address or operand size instead of mode.
In case there is something which gets shortened (perhaps only on a BFD64
build targeting a 32-bit binary), seeing the full original value is
often helpful to understand what's actually going wrong. Of course for
non-64-bit binaries we better wouldn't be seeing such warnings at all,
as they're often indicative of a behavioral difference between BFD64 and
!BFD64 builds.
Prior to "gas: drop sprint_value()", which introduced the use of
bfd_sprintf_vma(), the output had other shortcomings.
Just like e.g. 0x10000 triggers a warning for size 2, -0x10000 ought to
as well.
Note that some of the encodings produced aren't ones one would expect,
and hence the generated code is not being checked for in the new
testcases.
The description of e68c3d59ac ("x86: better respect quotes in
parse_operands()") wrongly states:
"In i386_att_operand(), which needs adjustment to remain in sync, besides
respecting double quotes now, also change the logic such that we don't
count parentheses anymore: Finding any opening or closing parenthesis or
any double quote means we're done, because the subsequent parsing code
wouldn't accept (extra) instances of these anyway."
I didn't pay attention to the possibility of the scale factor being
specified as an expression, which may contain parentheses. Thanks to
Martin for pointing this out. Restore prior behavior or matching
parentheses (backwards), while giving the variable a more suitable name.
Note that this simple and immediate fix is not ging to be enough: This
expression could itself involve quoted symbols. However, to address this
backwards parsing needs to be done away with altogether here (such that
parentheses which are part of such a quoted symbol name can also
properly be accounted for), which is going to be a more intrusive
change.
I finally found time to teach readelf to identify PIEs in the file
header display and program header display. So in place of
"DYN (Shared object file)" which isn't completely true, show
"DYN (Position-Independent Executable file)".
It requires a little bit of untangling code in readelf due to
process_program_headers setting up dynamic_addr and dynamic_size,
needed to scan .dynamic for the DT_FLAGS_1 entry, and
process_program_headers itself wanting to display the file type in
some cases. At first I modified process_program_header using a
"probe" parameter similar to get_section_headers in order to inhibit
output, but decided it was cleaner to separate out
locate_dynamic_sections.
binutils/
* readelf.c (locate_dynamic_section, is_pie): New functions.
(get_file_type): Replace e_type parameter with filedata. Call
is_pie for ET_DYN. Update all callers.
(process_program_headers): Use local variables dynamic_addr and
dynamic_size, updating filedata on exit from function. Set
dynamic_size of 1 to indicate no dynamic section or segment.
Update tests of dynamic_size throughout.
* testsuite/binutils-all/x86-64/pr27708.dump: Update expected output.
ld/
* testsuite/ld-pie/vaddr-0.d: Update expected output.
gdb/
* testsuite/lib/gdb.exp (exec_is_pie): Match new PIE readelf output.
Add conditional logic around fcntl.h F_{G,S}ETFL usage to fix builds
on systems that don't have it (e.g. Windows). The code is only used
to save & restore limited terminal stdin state.
