PR target/19938
bfd * elf-bbfd.h (struct elf_backend_data): New field:
elf_strtab_flags.
New field: elf_backend_set_special_section_info_and_link
* elfxx-target.h (elf_backend_strtab_flags): Define if not already
defined.
(elf_backend_set_special_section_info_and_link): Define if not
already defined.
(elfNN_bed): Use elf_backend_set_special_section_info_and_link and
elf_backend_strtab_flags macros to initialise fields in structure.
* elf.c (_bfd_elf_make_section_from_shdr): Check for SHF_STRINGS
being set even if SHF_MERGE is not set.
(elf_fake_sections): Likewise.
(section_match): New function. Matches two ELF sections based
upon fixed characteristics.
(find_link): New function. Locates a section in a BFD that
matches a section in a different BFD.
(_bfd_elf_copy_private_bfd_data): Copy the sh_info and sh_link
fields of reserved sections.
(bfd_elf_compute_section_file_positions): Set the flags for the
.shstrtab section based upon the elf_strtab_flags field in the
elf_backend_data structure.
(swap_out_syms): Likewise for the .strtab section.
* elflink.c (bfd_elf_final_link): Set the flags for the
.strtab section based upon the elf_strtab_flags field in the
elf_backend_data structure.
* elf32-i386.c (elf32_i386_set_special_info_link): New function.
(elf_backend_strtab_flags): Set to SHF_STRINGS for Solaris
targets.
(elf_backend_set_special_section_info_and_link): Define for
Solaris targets.
* elf32-sparc.c: Likewise.
* elf64-x86-64.c: Likewise.
binutils* testsuite/binutils-all/i386/compressed-1b.d: Allow for the
string sections possibly having the SHF_STRINGS flag bit set.
* testsuite/binutils-all/i386/compressed-1c.d: Likewise.
* testsuite/binutils-all/readelf.s: Likewise.
* testsuite/binutils-all/readelf.s-64: Likewise.
* testsuite/binutils-all/x86-64/compressed-1b.d: Likewise.
* testsuite/binutils-all/x86-64/compressed-1c.d: Likewise.
gas * testsuite/gas/i386/ilp32/x86-64-unwind.d: Allow for the string
sections possibly having the SHF_STRINGS flag bit set.
* testsuite/gas/i386/x86-64-unwind.d: Likewise.
This test exercises the scenarios where we attempt to connect GDB to GDBserver
in standard remote mode, query the symbol file path, attempt to open said
symbol file on GDB's end and fail, causing the connection to drop abruptly.
Regression-tested on x86-64/Ubuntu.
With an unpatched GDB we should see this:
FAIL: gdb.server/connect-with-no-symbol-file.exp: sysroot=: action=permission: connection to GDBserver succeeded (the program is no longer running)
FAIL: gdb.server/connect-with-no-symbol-file.exp: sysroot=: action=delete: connection to GDBserver succeeded (the program is no longer running)
FAIL: gdb.server/connect-with-no-symbol-file.exp: sysroot=target:: action=permission: connection to GDBserver succeeded (the program is no longer running)
FAIL: gdb.server/connect-with-no-symbol-file.exp: sysroot=target:: action=delete: connection to GDBserver succeeded (the program is no longer running)
A patched GDB should have full passes.
gdb/testsuite/ChangeLog:
2016-04-13 Luis Machado <lgustavo@codesourcery.com>
* gdb.server/connect-with-no-symbol-file.c: New file.
* gdb.server/connect-with-no-symbol-file.exp: New file.
When we attempt to debug a process using GDBserver in standard remote mode
without a symbol file on GDB's end, we may run into an issue where GDB cuts
the connection attempt short due to an error. The error is caused by not
being able to open a symbol file, like so:
--
(gdb) set sysroot
(gdb) tar rem :2345
Remote debugging using :2345
/proc/23769/exe: Permission denied.
(gdb) i r
The program has no registers now.
(gdb)
It should've been like this:
(gdb) set sysroot
(gdb) tar rem :2345
Remote debugging using :2345
warning: /tmp/symbol-file: Permission denied.
0xf7ddb2d0 in ?? ()
(gdb) i r
eax 0x0 0
ecx 0x0 0
edx 0x0 0
ebx 0x0 0
esp 0xffffdfa0 0xffffdfa0
ebp 0x0 0x0
esi 0x0 0
edi 0x0 0
eip 0xf7ddb2d0 0xf7ddb2d0
eflags 0x200 [ IF ]
cs 0x33 51
ss 0x2b 43
ds 0x0 0
es 0x0 0
fs 0x0 0
gs 0x0 0
(gdb)
This is caused by a couple of function calls within exec_file_locate_attach
that can potentially throw errors.
The following patch guards both exec_file_attach and symbol_file_add_main to
prevent the errors from disrupting the connection process.
There was also a case where native GDB tripped on this problem, but it was
mostly fixed by bf74e428bc.
Regression-tested on x86-64/Ubuntu.
gdb/ChangeLog:
2016-04-13 Luis Machado <lgustavo@codesourcery.com>
* exec.c (exec_file_locate_attach): Guard a couple functions
that can throw errors.
(exception_print_same): New helper function.
This patch fixes the documentation for the zero_ext bytecode description.
It removes parts that seemed like a copy/paste from ext, since zero_ext
zeros the bits to the left.
gdb/doc/ChangeLog:
* agentexpr.texi (zero_ext): Fix zero_ext description.
gas/
2016-04-12 Claudiu Zissulescu <claziss@synopsys.com>
* config/tc-arc.c (mach_type_specified_p): Change type to
bfd_boolean.
(arc_option): Set private flags when parsing cpu pseudo-op.
(md_parse_option): Set mach_type_specified_p to TRUE.
This patch fixes the current comment in gdb_remote_download, which is
false (the "except if that's already where it is" part). It also
improves it, by explaining why pass TOFILE through standard_output_file,
even it is an absolute path.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (gdb_remote_download): Fix and extend comment.
PR target/19983
* readelf.c (get_solaris_section_type): New function: Returns the
name of Solaris specific section types.
(get_solaris_dynamic_type): New function: Return the name of
Solaris specific dynamic types.
(get_dynamic_type): Use get_solaris_dynamic_type.
(get_section_type_name): Use get_solaris_section_type.
(get_solaris_symbol_visibility): New function: Returns Solaris
specific symbol visibilities.
(print_dynamic_symbol): Use get_solaris_symbol_visibility.
(process_symbol_table): Likewise.
In some cases a variable could be left uninitialised and then an attempt
made to read this variable, resulting in a tcl error. This commit
initialises the variable in all cases.
ld/ChangeLog:
* testsuite/lib/ld-lib.exp (run_dump_test): Initialise
check_ld(terminal).
gdbserver-base.exp is used as the base for both native-gdbserver.exp and
native-extended-gdbserver.exp. (Despite its name, it should really be
considered as a "local-gdbserver-base", as it's not really appropriate to
implement a remote gdbserver board.)
Currently, the _download procedure is implemented as a no-op (it returns
the source file path). Because of the SONAME change, The fast
tracepoint tests now require the executable and the IPA
(libinproctrace.so) to be located in the same directory (see [1]). When
using the native-gdbserver board, because _download returns the original
file path, the executable does not end up in the same directory as the
library, and it fails to execute.
In more general terms, with the recent changes, the testsuite now
assumes that when it does
${board}_download <source path 1> <destination path 1>
${board}_download <source path 2> <destination path 2>
where the destination paths are relative (generally just the file name),
both files will end up in the same base directory. That assumption does
not hold for the current implementation in gdbserver-base.exp.
The proper fix would be to make native-gdbserver non-remote, so that
gdb_remote_download would not call DejaGnu's remote_download (see [2]).
We could then get rid of ${board}_download in gdbserver-base.exp.
However, that will likely take some time to complete. In the mean time,
in order to make the fast tracepoint tests pass, we can simply copy the
file to the standard output directory. Basically, it just mimics what
gdb_remote_download would do if the board wasn't flagged as remote.
Note that I missed these failures originally because I had a
libinproctrace.so in /usr/local/lib. So, even though libinproctrace.so
wasn't copied to the test output directory, it did find the one in
/usr/local/lib. It would be nice to find a way to protect against this,
as it could easily happen again...
Regtested with unix, native-gdbserver and native-extended-gdbserver, and
didn't see anything notable, except the ftrace tests now passing for
native-gdbserver.
[1] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=commit;h=6e774b13c3b81ac2599812adf058796948ce7e95
[2] https://sourceware.org/ml/gdb-patches/2016-04/msg00112.html
gdb/testsuite/ChangeLog:
* boards/gdbserver-base.exp (${board}_download): Copy source file to
standard output directory.
PR target/19937
opcode * v850-opc.c (v850_opcodes): Correct masks for long versions of
the LD.B and LD.BU instructions.
gas * testsuite/gas/v850/pr19937.s: New test.
* testsuite/gas/v850/pr19937.d: New test control file.
* testsuite/gas/v850/basic.exp: Run the new test.
This patch fixes the following failure:
FAIL: gdb.trace/trace-condition.exp: ftrace: -(21 << 1) == -42: check 10
frames were collected.
This was due to aarch64_emit_sub using the wrong order in its operands, so the
operation would end up being 42 - 0 rather than 0 - 42.
This patch also fixes the order of aarch64_emit_add for clarity.
The test case for emit_sub is fixed so that the proper order of
the operands is needed for the test to pass.
Tested on aarch64-native-extended-gdbserver.
Note: trace-condition.exp was broken a bit so I had to modify it to run
the test. A fix is coming for that in another patch.
gdb/gdbserver/ChangeLog:
* linux-aarch64-low.c (aarch64_emit_add): Switch x1 and x0.
(aarch64_emit_sub): Likewise.
gdb/testsuite/ChangeLog:
* gdb.trace/trace-condition.exp (foreach): Fix emit_sub testcase.
Reverse debugging against a remote target that does reverse debugging
itself (with the bs/bc packets) always trips on:
(gdb) target remote localhost:...
(gdb) reverse-stepi
../../gdb/target.c:602: internal-error: default_execution_direction: to_execution_direction must be implemented for reverse async
I missed adding a to_execution_direction method to remote.c in commit
3223143295 (Adds target_execution_direction to make record targets
support async mode), GDB 7.4 time. Later, GDB 7.8 switched to
target-async on by default, making the regression user-visible by
default too.
Fix is simply to add the missing to_execution_direction implementation
to target remote.
Tested by Andi Kleen against Simics.
gdb/ChangeLog:
2016-04-13 Pedro Alves <palves@redhat.com>
PR remote/19840
* remote.c (struct remote_state) <last_resume_exec_dir>: New
field.
(new_remote_state): Default last_resume_exec_dir to EXEC_FORWARD.
(remote_open_1): Reset last_resume_exec_dir to EXEC_FORWARD.
(remote_resume): Store the last execution direction.
(remote_execution_direction): New function.
(init_remote_ops): Install it as to_execution_direction target_ops
method.
On systems with a newer version of GCC the gdb.btrace/instruction_history.exp
test fails to build like this:
Running .../gdb.btrace/instruction_history.exp ...
gdb compile failed, .../gdb.btrace/instruction_history.c:
In function 'main': .../gdb.btrace/instruction_history.c:24:3: warning:
implicit declaration of function 'loop' [-Wimplicit-function-declaration]
loop ();
^
Declare loop to fix it.
testsuite/
* gdb.btrace/instruction_history.c (loop): Add declaration.
Since compiler may pass --as-needed to ld by default, link .o file
before .so file in x86-64 tests.
PR ld/19774
* testsuite/ld-x86-64/x86-64.exp: Link tmpdir/pr17689b.o before
tmpdir/pr17689.so, fix gotpcrel1 test and add more --as-needed
tests.
Weak symbols can be preempted at link time so always choose the longer
sequence in branch relaxation, according to the relaxation level chosen,
so that any symbol finally used as the branch target is reachable.
2016-04-13 Maciej W. Rozycki <macro@imgtec.com>
Andrew Bennett <andrew.bennett@imgtec.com>
gas/
* config/tc-mips.c (relaxed_branch_length): Use the long
sequence where the target is a weak symbol.
(relaxed_micromips_32bit_branch_length): Likewise.
(relaxed_micromips_16bit_branch_length): Likewise.
* testsuite/gas/mips/branch-weak-1.d: New test.
* testsuite/gas/mips/branch-weak-2.d: New test.
* testsuite/gas/mips/branch-weak-3.d: New test.
* testsuite/gas/mips/branch-weak-4.d: New test.
* testsuite/gas/mips/branch-weak-5.d: New test.
* testsuite/gas/mips/branch-weak.l: New stderr output.
* testsuite/gas/mips/branch-weak.s: New test source.
* testsuite/gas/mips/mips.exp: Run the new tests.
Where branch relaxation is enabled emit the long sequence for branches
whose distance cannot be determined, i.e. to symbols that are undefined
or in a different segment. These symbols are only resolved at link time
and therefore the longer sequence ensures the branch target is in range,
which cannot be guaranteed with a direct branch.
This is the opposite to the current implementation, originally proposed
here: <https://sourceware.org/ml/binutils/2002-09/msg00218.html>. The
proposal was then extensively discussed before the final version was
posted here: <https://sourceware.org/ml/binutils/2002-10/msg00191.html>
and eventually committed:
commit 4a6a3df43d
Author: Alexandre Oliva <aoliva@redhat.com>
Date: Sat Oct 12 05:23:33 2002 +0000
The case considered here was not commented in the review however and the
original version remains. With branch relaxation enabled it makes more
sense to do it consistently, so that all code impure with respect to
branch distances can be linked. Direct branches are still produced for
the cases concerned where branch relaxation is disabled, which is the
default.
gas/
* config/tc-mips.c (relaxed_branch_length): Use the long
sequence where the distance cannot be determined.
(relaxed_micromips_32bit_branch_length): Likewise.
* testsuite/gas/mips/branch-extern-1.d: New test.
* testsuite/gas/mips/branch-extern-2.d: New test.
* testsuite/gas/mips/branch-extern-3.d: New test.
* testsuite/gas/mips/branch-extern-4.d: New test.
* testsuite/gas/mips/branch-extern.l: New stderr output.
* testsuite/gas/mips/branch-extern.s: New test source.
* testsuite/gas/mips/branch-section-1.d: New test.
* testsuite/gas/mips/branch-section-2.d: New test.
* testsuite/gas/mips/branch-section-3.d: New test.
* testsuite/gas/mips/branch-section-4.d: New test.
* testsuite/gas/mips/branch-section.l: New stderr output.
* testsuite/gas/mips/branch-section.s: New test source.
* testsuite/gas/mips/mips.exp: Run the new tests.
This obvious patch replaces "ond" wiht "cond" as the test prefix for
conditional tests.
gdb/testsuite/ChangeLog:
* gdb.trace/ftrace.exp (proc): Change test prefix from "ond" to "cond".
Now that we don't ever throw GDB exceptions from signal handlers [1],
we can switch back to having TRY/CATCH implemented in terms of C++
try/catch instead of sigjmp/longjmp.
[1] - https://sourceware.org/ml/gdb-patches/2016-03/msg00351.html
Tested on x86_64 Fedora 23, native and gdbserver.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* common/common-exceptions.h (GDB_XCPT_TRY): Update comment.
[__cplusplus] (GDB_XCPT): Define as GDB_XCPT_TRY.
Now that we don't ever throw GDB exceptions from signal handlers [1],
we can switch to have TRY/CATCH implemented in terms of plain
setjmp/longjmp instead of sigsetjmp/siglongjmp.
In https://sourceware.org/ml/gdb-patches/2015-02/msg00114.html, Yichun
Zhang mentions a 11%/14%+ speedup in his GDB python scripts with a
patch that did something similar to only a specific set of TRY/CATCH
calls.
[1] - https://sourceware.org/ml/gdb-patches/2016-03/msg00351.html
Tested on x86_64 Fedora 23, native and gdbserver.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* common/common-exceptions.c (struct catcher) <buf>: Now a
'jmp_buf' instead of SIGJMP_BUF.
(exceptions_state_mc_init): Change return type to 'jmp_buf'.
(throw_exception): Use longjmp instead of SIGLONGJMP.
* common/common-exceptions.h: Include <setjmp.h> instead of
"gdb_setjmp.h".
(exceptions_state_mc_init): Change return type to 'jmp_buf'.
[GDB_XCPT == GDB_XCPT_SJMP] (TRY): Use setjmp instead of
SIGSETJMP.
* cp-support.c: Include "gdb_setjmp.h".
This finally gets rid of immediate_quit (and surrounding
infrustruture), as nothing sets it anymore.
gdb_call_async_signal_handler was only necessary in order to handle
immediate_quit. We can just call mark_async_signal_handler directly
on all hosts now.
In turn, we can clean up mingw-hdep.c's gdb_select a bit, as
sigint_event / sigint_handler is no longer needed.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* defs.h: Update comments on SIGINT handling.
(immediate_quit): Delete declaration.
* event-loop.c (call_async_signal_handler): Delete.
* event-loop.h (call_async_signal_handler): Delete declaration.
(mark_async_signal_handler): Update comments.
(gdb_call_async_signal_handler): Delete declaration.
* event-top.c (handle_sigint): Call mark_async_signal_handler
instead of gdb_call_async_signal_handler.
* exceptions.c (prepare_to_throw_exception): Remove reference to
immediate_quit.
(exception_fprintf): Remove comments about immediate_quit.
* mingw-hdep.c (sigint_event, sigint_handler): Delete.
(gdb_select): Don't wait on sigint_event.
(gdb_call_async_signal_handler): Delete.
(_initialize_mingw_hdep): Delete.
* posix-hdep.c (gdb_call_async_signal_handler): Delete.
* utils.c (immediate_quit): Delete.
remote.c is the last user of immediate_quit. It's relied on to
immediately break the initial remote connection sync up, if the user
does Ctrl-C, assuming that was because the target isn't responding.
At that stage, since the connection isn't synced yet, disconnecting is
the only safe thing to do. This commit reworks that, to not rely on
throwing from the SIGINT signal handler.
So, this commit:
- Introduces the concept of a "quit handler". This is used to
override what does the QUIT macro do when the quit flag is set.
- Makes the "struct serial" reachar / write code call QUIT in the
partial read/write loops, so the current quit handler is invoked
whenever a serial->read_prim / serial->write_prim returns EINTR.
- Makes the "struct serial" reachar / write code call
interruptible_select instead of gdb_select, so that QUITs are
detected in a race-free manner.
- Stops remote.c from setting immediate_quit during the initial
connection.
- Instead, we install a custom quit handler whenever we're calling
into the serial code. This custom quit handler knows to immediately
throw a quit when we're in the initial connection setup, and
otherwise defer handling the quit/Ctrl-C request to later, when
we're safely out of a packet command/response sequence. This also
is what is now responsible for handling "double Ctrl-C because
target connection is stuck/wedged."
- remote.c no longer installs a specialized SIGINT handlers, and
instead re-uses the quit flag. Since we want to rely on the QUIT
macro, the SIGINT handler must also set the quit. And the easiest
is just to not install custom SIGINT handler in remote.c. Let the
standard SIGINT handler do its job of setting the quit flag.
Centralizing SIGINT handlers seems like a good thing to me, anyway.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* defs.h (quit_handler_ftype, quit_handler)
(make_cleanup_override_quit_handler, default_quit_handler): New.
(QUIT): Adjust comments.
* event-top.c (default_quit_handler): New function.
(quit_handler): New global.
(struct quit_handler_cleanup_data): New.
(restore_quit_handler, restore_quit_handler_dtor)
(make_cleanup_override_quit_handler): New.
(async_request_quit): Call QUIT.
* remote.c (struct remote_state) <got_ctrlc_during_io>: New field.
(async_sigint_remote_twice_token, async_sigint_remote_token):
Delete.
(remote_close): Update comments.
(remote_start_remote): Don't set immediate_quit. Set starting_up
earlier.
(remote_serial_quit_handler, remote_unpush_and_throw): New
functions.
(remote_open_1): Clear got_ctrlc_during_io. Set
remote_async_terminal_ours_p unconditionally.
(async_initialize_sigint_signal_handler)
(async_handle_remote_sigint, async_handle_remote_sigint_twice)
(remote_check_pending_interrupt, async_remote_interrupt)
(async_remote_interrupt_twice)
(async_cleanup_sigint_signal_handler, ofunc)
(sync_remote_interrupt, sync_remote_interrupt_twice): Delete.
(remote_terminal_inferior, remote_terminal_ours): Remove async
checks.
(remote_wait_as): Don't install a SIGINT handler in sync mode.
(readchar, remote_serial_write): Override the quit handler with
remote_serial_quit_handler.
(getpkt_or_notif_sane_1): Don't call QUIT.
(initialize_remote_ops): Don't install
remote_check_pending_interrupt.
(_initialize_remote): Don't create async_sigint_remote_token and
async_sigint_remote_twice_token.
* ser-base.c (ser_base_wait_for): Call QUIT and use
interruptible_select.
(ser_base_write): Call QUIT.
* ser-go32.c (dos_readchar, dos_write): Call QUIT.
* ser-unix.c (wait_for): Don't use VTIME. Always take the
gdb_select path, but call QUIT and interruptible_select.
* utils.c (maybe_quit): Call the current quit handler. Don't call
target_check_pending_interrupt.
(defaulted_query, prompt_for_continue): Override the quit handler
with the default quit handler.
The MI code only does output, so leave raw/cooked mode alone, as well
as the SIGINT handler. Restore terminal settings after output, while
at it. Also, a couple events missed calling target_terminal_ours
before output, even.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* mi/mi-interp.c (mi_new_thread): Put
target_terminal_ours_for_output in effect while outputting.
(mi_thread_exit): Use target_terminal_ours_for_output instead of
target_terminal_ours.
(mi_record_changed, mi_inferior_added, mi_inferior_appeared)
(mi_inferior_exit, mi_inferior_removed, mi_traceframe_changed)
(mi_tsv_created, mi_tsv_deleted, mi_tsv_modified)
(mi_breakpoint_created, mi_breakpoint_deleted)
(mi_breakpoint_modified, mi_solib_loaded, mi_solib_unloaded)
(mi_command_param_changed, mi_memory_changed)
(report_initial_inferior): Use target_terminal_ours_for_output
instead of target_terminal_ours. Restore terminal settings.
* mi/mi-main.c (mi_execute_command): Use
target_terminal_ours_for_output instead of target_terminal_ours.
Restore terminal settings.
Any time a caller calls query & friends / prompt_for_continue without
ensuring that gdb owns the terminal for input is a bug. So do that in
defaulted_query / prompt_for_continue directly instead.
An example of a case where we currently miss calling
target_terminal_ours is internal_error. Ever since defaulted_query
was made to use gdb_readline_callback, there's no way to answer the
internal error query if the internal error happens while the target is
has the terminal:
(gdb) c
Continuing.
.../src/gdb/linux-nat.c:1676: internal-error: linux_nat_resume: Assertion `dummy_counter < 10' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n) _
Entering 'y' or 'n' does not work, GDB does not respond.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
PR gdb/19828
* gnu-nat.c (inf_validate_task_sc): Don't call
target_terminal_ours / target_terminal_inferior around query.
* i386-tdep.c (i386_record_lea_modrm, i386_process_record): Don't
call target_terminal_ours / target_terminal_inferior around
yquery.
* linux-record.c (record_linux_system_call): Don't call
target_terminal_ours / target_terminal_inferior around yquery.
* nto-procfs.c (interrupt_query): Don't call target_terminal_ours
/ target_terminal_inferior around query.
* record-full.c (record_full_check_insn_num): Remove
'set_terminal' parameter. Don't call target_terminal_ours /
target_terminal_inferior around query.
(record_full_message, record_full_registers_change)
(record_full_xfer_partial): Adjust.
* remote.c (interrupt_query): Don't call target_terminal_ours /
target_terminal_inferior around query.
* utils.c (defaulted_query): Install cleanup to restore target
terminal. Put target_terminal_ours_for_output in effect while
defaulted producing, and target_terminal_ours in in effect while
handling input.
(prompt_for_continue): Install cleanup to restore target terminal.
Put target_terminal_ours in in effect while handling input.
Some of the error paths in these functions leak.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* utils.c (defaulted_query, prompt_for_continue): Free temporary
strings with cleanups, instead of xfree.
We're only doing output here, so leave raw/cooked mode alone, as well
as the SIGINT handler.
And restore terminal settings, while at it.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* utils.c (vwarning, internal_vproblem): Use
make_cleanup_restore_target_terminal and
target_terminal_ours_for_output.
We're only doing output here, so leave raw/cooked mode alone, as well
as the SIGINT handler.
No need to restore terminal settings, we'll set inferior modes on the
following resume.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* infcmd.c (post_create_inferior, prepare_one_step): Use
target_terminal_ours_for_output instead of target_terminal_ours.
We're only doing output here, so leave raw/cooked mode alone, as well
as the SIGINT handler.
Restore terminal settings after output, while at it.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* exceptions.c (print_flush): Use target_terminal_ours_for_output
instead of target_terminal_ours, and restore target terminal with
a cleanup.
We're only doing output here, so leave raw/cooked mode alone, as well
as the SIGINT handler.
Restore terminal settings after output, while at it.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* cp-support.c (gdb_demangle): Use target_terminal_ours_for_output
instead of target_terminal_ours, and restore target terminal with
a cleanup.
A couple wrong things here
- We should not use target_terminal_ours when all we want is output.
We should use target_terminal_ours_for_output instead, which
preserves raw/cooked terminal modes, and SIGINT forwarding.
- Most importantly, relying on stderr output immediately preceding
the error/exception print isn't correct. The exception could be
caught and handled, for example; MI frontends won't display the
stderr part in an error dialog box. Etc.
This commit introduces a type_as_string helper that allows building a
full error string including type info.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* ada-lang.c (type_as_string, type_as_string_and_cleanup): New
functions.
(ada_lookup_struct_elt_type): Use type_as_string_and_cleanup.
- If serial->write_prim returns EINTR, ser_bas_write returns it to the
caller. This just looks wrong to me -- part of the output may have
already been sent, and there's no way for the caller to know that,
and thus no way for a caller to handle a partial write correctly.
- While ser-unix.c:ser_unix_read_prim retries on EINTR,
ser-tcp.c:net_read_prim does not.
This commit moves EINTR handling to the ser_base_write and
ser_base_readchar level, so all serial backends (at least those that
use it) end up handling EINTR consistently.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* ser-base.c (fd_event): Retry read_prim on EINTR.
(do_ser_base_readchar): Retry read_prim on EINTR.
(ser_base_write): Retry write_prim on EINTR.
* ser-unix.c (ser_unix_read_prim): Don't retry on EINTR here.
(ser_unix_write_prim): Remove comment.
If the user presses Ctrl-C immediately before target_terminal_inferior
is called and the target is resumed, instead of after, the Ctrl-C ends
up pending in the quit flag until the target next stops.
remote.c has this bit to handle this:
if (!target_is_async_p ())
{
ofunc = signal (SIGINT, sync_remote_interrupt);
/* If the user hit C-c before this packet, or between packets,
pretend that it was hit right here. */
if (check_quit_flag ())
sync_remote_interrupt (SIGINT);
}
But that's only reachable if async is off, while async is on by
default nowadays. It's also obviously not reacheable on native
targets.
This patch generalizes that to all targets.
We can't remove that remote.c bit yet, until we get rid of the sync
SIGINT handler though. That'll be done later in the series.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* remote.c (remote_pass_ctrlc): New function.
(init_remote_ops): Install it.
* target.c (target_terminal_inferior): Pass pending Ctrl-C to the
target.
(target_pass_ctrlc, default_target_pass_ctrlc): New functions.
* target.h (struct target_ops) <to_pass_ctrlc>: New method.
(target_pass_ctrlc, default_target_pass_ctrlc): New declarations.
* target-delegates.c: Regenerate.
In non-stop mode, "interrupt" results in a "stop with no signal",
while in all-stop mode, it results in a remote interrupt request /
stop with SIGINT. This is currently implemented in both the Linux and
remote target backends. Move it to the core code instead, making
target_interrupt specifically always about "Interrupting as if with
Ctrl-C", just like it is documented.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* infcmd.c (interrupt_target_1): Call target_stop is in non-stop
mode.
* linux-nat.c (linux_nat_interrupt): Delete.
(linux_nat_add_target): Don't install linux_nat_interrupt.
* remote.c (remote_interrupt_ns): Change return type to void.
Throw error if interrupting the target is not supported.
(remote_interrupt): Don't call the remote_stop_ns/remote_stop_as.
This call seems pointless. For instance, a SIGINT handler is only
installed later on. And if wasn't, I can't see why we'd want to lose
a Ctrl-C request.
Getting rid of this allows getting rid of clear_quit_flag.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* main.c (captured_main): Don't clear the quit flag.
I think this is reminiscent of the time when a longjmp would always
jump to the top level. Nowaways code that throw exceptions other than
a quit, which may even be caught and handled without reaching the top
level. Certainly such exceptions shouldn't clear an interrupt
request...
(We also need to get rid of prepare_to_throw_exception in order to be
able to just do "throw ex;" in C++.)
One could argue that we should clear the quit flag when we throw a
quit from the SIGINT handler, when immediate_quit is in effect, to
handle a race, here:
immediate_quit++;
QUIT;
... that's the usual pattern code must use when enabling
immediate_quit. The QUIT is there to catch the case of Ctrl-C having
already been pressed before immediate_quit was enabled. However, this
can happen:
immediate_quit++;
<< Ctrl-C pressed here too.
QUIT;
And in that case, if the quit flag was already set, it'll stay set
even after throwing a quit from the SIGINT handler. The end result is
a double quit. But OTOH, the user did press Ctrl-C two times. Since
I'm getting rid of immediate_quit, I'm not bothering with this.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* exceptions.c (prepare_to_throw_exception): Don't clear the quit
flag.
This just looks totally wrong to me, for completetly discarding a
user-requested Ctrl-C. I can't think of why we'd want do this here.
Actually, I digged the history, and found out that this has been here
since at least 7b4ac7e1ed (gdb-2.4, the initial revision, 1988), at
a time were we had a top level setjmp/longjmp, long before that got
wrapped in throw_exception and friends, and this code was in an
explicit loop, with the quit_flag cleared on every iteration, before
executing a command...
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* event-top.c (command_handler): Don't call clear_quit_flag.
Obviously not necessary since check_quit_flag clears the flag as side
effect.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* remote-sim.c (gdb_os_poll_quit): Don't call clear_quit_flag.
* remote.c (remote_wait_as): Don't call clear_quit_flag.
Now that we have an abstract for wakeable events, use it instead of a
(heavier) serial pipe.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* python/python.c: Include "ser-event.h".
(gdbpy_event_fds): Delete.
(gdbpy_serial_event): New.
(gdbpy_run_events): Change prototype. Use serial_event_clear
instead of serial_readchar.
(gdbpy_post_event): Use serial_event_set instead of serial_write.
(gdbpy_initialize_events): Use make_serial_event instead of
serial_pipe.
We have places where we call a blocking gdb_select expecting that a
Ctrl-C will unblock it. However, if the Ctrl-C is pressed just before
gdb_select, the SIGINT handler runs before gdb_select, and thus
gdb_select won't return.
For example gdb_readline_no_editing:
QUIT;
/* Wait until at least one byte of data is available. Control-C
can interrupt gdb_select, but not fgetc. */
FD_ZERO (&readfds);
FD_SET (fd, &readfds);
if (gdb_select (fd + 1, &readfds, NULL, NULL, NULL) == -1)
and stdio_file_read:
/* For the benefit of Windows, call gdb_select before reading from
the file. Wait until at least one byte of data is available.
Control-C can interrupt gdb_select, but not read. */
{
fd_set readfds;
FD_ZERO (&readfds);
FD_SET (stdio->fd, &readfds);
if (gdb_select (stdio->fd + 1, &readfds, NULL, NULL, NULL) == -1)
return -1;
}
return read (stdio->fd, buf, length_buf);
This is a race classically fixed with either the self-pipe trick, or
by blocking SIGINT and then using pselect instead of select.
Blocking SIGINT most of the time would mean that check_quit_flag (and
thus QUIT) would need to do a syscall every time it is called, which
sounds best avoided, since QUIT is called in many loops. Thus we take
the self-pipe trick route (wrapped in a serial event).
Instead of having all places that need this manually add an extra file
descriptor to the set of gdb_select's watched file descriptors, we
introduce a wrapper, interruptible_select, that does that.
The Windows version of gdb_select actually does not suffer from this,
because mingw-hdep.c:gdb_call_async_signal_handler sets a Windows
event that gdb_select always waits on. So this patch can be seen as
generalization of that technique. We can't remove that extra event
from mingw-hdep.c until we get rid of immediate_quit though.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* defs.h: Extend QUIT-related comments to mention
interruptible_select.
(quit_serial_event_set, quit_serial_event_clear): Declare.
* event-top.c: Include "ser-event.h" and "gdb_select.h".
(quit_serial_event): New global.
(async_init_signals): Make quit_serial_event.
(quit_serial_event_set, quit_serial_event_clear)
(quit_serial_event_fd, interruptible_select): New functions.
* extension.c (set_quit_flag): Set the quit serial event.
(check_quit_flag): Clear the quit serial event.
* gdb_select.h (interruptible_select): New declaration.
* guile/scm-ports.c (ioscm_input_waiting): Use
interruptible_select instead of gdb_select.
* top.c (gdb_readline_no_editing): Likewise.
* ui-file.c (stdio_file_read): Likewise.
GDB's core signal handling suffers from a classical signal handler /
mainline code race:
int
gdb_do_one_event (void)
{
...
/* First let's see if there are any asynchronous signal handlers
that are ready. These would be the result of invoking any of the
signal handlers. */
if (invoke_async_signal_handlers ())
return 1;
...
/* Block waiting for a new event. (...). */
if (gdb_wait_for_event (1) < 0)
return -1;
...
}
If a signal is delivered while gdb is blocked in the poll/select
inside gdb_wait_for_event, then the select/poll breaks with EINTR,
we'll loop back around and call invoke_async_signal_handlers.
However, if the signal handler runs between
invoke_async_signal_handlers and gdb_wait_for_event,
gdb_wait_for_event will block, until the next unrelated event...
The fix is to a struct serial_event, and register it in the set of
files that select/poll in gdb_wait_for_event waits on. The signal
handlers that defer work to invoke_async_signal_handlers call
mark_async_signal_handler, which is adjusted to also set the new
serial event in addition to setting a flag, and is thus now is
garanteed to immediately unblock the next gdb_select/poll call, up
until invoke_async_signal_handlers is called and the event is cleared.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* event-loop.c: Include "ser-event.h".
(async_signal_handlers_serial_event): New global.
(async_signals_handler, initialize_async_signal_handlers): New
functions.
(mark_async_signal_handler): Set
async_signal_handlers_serial_event.
(invoke_async_signal_handlers): Clear
async_signal_handlers_serial_event.
* event-top.c (async_init_signals): Call
initialize_async_signal_handlers.
This patch adds a new "event" struct serial type, that is an
abstraction specifically for waking up blocking waits/selects,
implemented on top of a pipe on POSIX, and on top of a native Windows
event (CreateEvent, etc.) on Windows.
This will be used to plug signal handler / mainline code races.
For example, GDB can indefinitely delay handling a quit request if the
user presses Ctrl-C between the last QUIT call and the next (blocking)
gdb_select call in the event loop:
QUIT;
<<< press ctrl-c here and end up blocked in gdb_select
indefinitely.
gdb_select (...); // whoops, SIGINT was already handled, no EINTR.
A global alone (either the quit flag, or the "ready" flag of the async
signal handlers in the event loop) is not sufficient.
To plug races such as these on POSIX systems, we have to register some
waitable file descriptor in the set of files gdb_select waits on, and
write to it from the signal handler. This is classically a pipe, and
the pattern called the self-pipe trick. On Linux, it could be a more
efficient eventfd instead, but I'm sticking with a pipe for
simplifity, as we need it for portability anyway.
(Alternatively, we could use pselect/ppoll, and block signals until
the pselect. The latter is not a design I think GDB could use,
because we want the QUIT macro to be super cheap, as it is used in
loops. Plus, Windows.)
This is a "struct serial" because Windows's gdb_select relies on that.
Windows's gdb_select, our "select" replacement, knows how to wait on
all kinds of handles (regular files, pipes, sockets, console, etc.)
unlike the native Windows "select" function, which can only wait on
sockets. Each file descriptor for a "serial" type that is not
normally waitable with WaitForMultipleObjects must have a
corresponding struct serial instance. gdb_select then internally
looks up the struct serial instance that wraps each file descriptor,
and asks it for the corresponding Windows waitable handle.
We could use serial_pipe() to create a "struct serial"-wrapped pipe
that is usable everywhere, including Windows. That's what currently
python/python.c uses for cross-thread posting of events.
However, serial_write and serial_readchar are not designed to be
async-signal-safe on POSIX hosts. It's easier to bypass those when
setting/clearing the event source.
And writing and a serial pipe is a bit heavy weight on Windows.
gdb_select requires an extra thread to wait on the pipe and several
Windows events, when a single manual-reset Windows event, with no
extra thread is sufficient.
The intended usage is simply:
- Call make_serial_event to create a serial event object.
- From the signal handler call serial_event_set to set the event.
- From mainline code, have select/poll wait for serial_event_fd(), in
addition to whatever other files you're about to wait for.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* Makefile.in (SFILES): Add ser-event.c.
(HFILES_NO_SRCDIR): Add ser-event.h.
(COMMON_OBS): Add ser-event.o.
* ser-event.c, ser-event.h: New files.
* serial.c (new_serial): New function, factored out from
(serial_fdopen_ops): ... this.
(serial_open_ops_1): New function, factored out from
(serial_open): ... this.
(serial_open_ops): New function.
* serial.h (struct serial): Forware declare.
(serial_open_ops): New declaration.
Not used by anything.
gdb/ChangeLog:
2016-04-12 Pedro Alves <palves@redhat.com>
* serial.c (serial_open, serial_fdopen_ops, do_serial_close):
Remove references to name.
* serial.h (struct serial) <name>: Delete.
