This patch improves the error reporting when handling SystemTap SDT
probes. "Handling", in this case, mostly means "parsing".
On gdb/probe.h, only trivial changes on functions' comments in order
to explicitly mention that some of them can throw exceptions. This is
just to make the API a bit more clear.
On gdb/stap-probe.c, I have s/internal_error/error/ on two functions
that are responsible for parsing specific bits of the probes'
arguments: stap_get_opcode and stap_get_expected_argument_type. It is
not correct to call internal_error on such situations because it is
not really GDB's fault if the probes have malformed arguments. I also
improved the error reported on stap_get_expected_argument_type by also
including the probe name on it.
Aside from that, and perhaps most importantly, I added a check on
stap_get_arg to make sure that we don't try to extract an argument
from a probe that has no arguments. This check issues an
internal_error, because it really means that GDB is doing something it
shouldn't.
Although it can be considered almost trivial, and despite the fact
that I am the maintainer for this part of the code, I am posting this
patch for review. I will wait a few days, and if nobody has anything
to say, I will go ahead and push it.
gdb/ChangeLog:
2015-09-01 Sergio Durigan Junior <sergiodj@redhat.com>
* probe.h (struct probe_ops) <get_probe_argument_count,
evaluate_probe_argument, enable_probe, disable_probe>: Mention in
the comment that the function can throw an exception.
(get_probe_argument_count): Likewise.
(evaluate_probe_argument): Likewise.
* stap-probe.c (stap_get_opcode): Call error instead of
internal_error.
(stap_get_expected_argument_type): Likewise. Add argument
'probe'. Improve error message by mentioning the probe's name.
(stap_parse_probe_arguments): Adjust call to
stap_get_expected_argument_type.
(stap_get_arg): Add comment. Assert that 'probe->args_parsed' is
not zero. Call internal_error if GDB requests an argument but the
probe has no arguments.
Before this change, trying to complete an expression ending with an
ambiguous function name (i.e. for which there are multiple matches)
would display a menu with a prompt for the user to pick one. For
instance:
(gdb) p func<tab>Multiple matches for func
[0] cancel
[1] pack2.func at pack2.adb:5
[2] pack.func at pack.adb:5
>
This is not user friendly and actually triggered a segmentation fault
after the user did pick one. It is not clear whether the segmentation
fault needs a separate fix, but this is the only known case which
exhibits it at the moment, and this case must be fixed itself.
The problem lies in ada-lang.c (ada_resolve_function): when we got
multiple matches, we should not display the menu if we are in completion
mode. This patch adjusts the corresponding condition accordingly.
gdb/ChangeLog:
* ada-lang.c (ada_resolve_function): Do not ask the user what
match to use when in completion mode.
gdb/testsuite/ChangeLog:
* gdb.ada/complete.exp: Add "pck.ambiguous_func" to the relevant
expected outputs. Add two testcases for completing ambiguous
functions.
* gdb.ada/complete/aux_pck.adb: New file.
* gdb.ada/complete/aux_pck.ads: New file.
* gdb.ada/complete/foo.adb: Pull Aux_Pck and call the two
Ambiguous_Func functions.
* gdb.ada/complete/pck.ads: Add an Ambiguous_Func function.
* gdb.ada/complete/pck.adb: Likewise.
Tested on x86_64-linux, no regression.
We use its PLT entry to initialize function pointer at run-time. If
there is no other usage for the PLT entry, we can generate run-time
function pointer relocations in read-write section, which can be
resolved by dynamic linker, to initialize function pointers. It avoids
the extra indirect branch overhead in PLT.
bfd/
PR ld/18900
* elf32-i386.c (elf_i386_link_hash_entry): Add
func_pointer_refcount.
(elf_i386_link_hash_newfunc): Clear func_pointer_refcount.
(elf_i386_get_local_sym_hash): Likewise.
(elf_i386_copy_indirect_symbol): Also copy
func_pointer_refcount.
(elf_i386_check_relocs): Increment func_pointer_refcount.
(elf_i386_gc_sweep_hook): Decrement func_pointer_refcount.
(elf_i386_allocate_dynrelocs): Don't create the PLT entry if
there are only function pointer relocations which can be
resolved at run-time. Keep dynanamic relocations for run-time
function pointer initialization.
(elf_i386_relocate_section): Copy dynamic function pointer
relocations.
* elf64-x86-64.c (elf_x86_64_link_hash_entry): Add
func_pointer_refcount.
(elf_x86_64_link_hash_newfunc): Clear func_pointer_refcount.
(elf_x86_64_get_local_sym_hash): Likewise.
(elf_x86_64_copy_indirect_symbol): Also copy
func_pointer_refcount.
(elf_x86_64_check_relocs): Increment func_pointer_refcount.
(elf_x86_64_gc_sweep_hook): Decrement func_pointer_refcount.
(elf_x86_64_allocate_dynrelocs): Don't create the PLT entry if
there are only function pointer relocations which can be
resolved at run-time. Keep dynanamic relocations for run-time
function pointer initialization.
(elf_x86_64_relocate_section): Copy dynamic function pointer
relocations.
ld/testsuite/
PR ld/18900
* ld-i386/i386.exp: Run tests for PR ld/18900.
* ld-x86-64/x86-64.exp: Likewise.
* ld-i386/pr18900.out: New file.
* ld-i386/pr18900a.c: Likewise.
* ld-i386/pr18900a.c: Likewise.
* ld-i386/pr18900a.rd: Likewise.
* ld-i386/pr18900b.c: Likewise.
* ld-i386/pr18900b.rd: Likewise.
* ld-i386/pr18900c.c: Likewise.
* ld-x86-64/pr18900.out: Likewise.
* ld-x86-64/pr18900a.c: Likewise.
* ld-x86-64/pr18900a.rd: Likewise.
* ld-x86-64/pr18900b.c: Likewise.
* ld-x86-64/pr18900b.rd: Likewise.
* ld-x86-64/pr18900c.c: Likewise.
* ld-x86-64/mpx3.dd: Updated.
OPD lookup goes awry when .opd has been shuffled to remove unused
functions.
* elf64-ppc.c (ppc64_elf_maybe_function_sym): Adjust symbol value
if .opd section has been edited.
Don't emit long branch or plt branch stubs to save/restore functions.
Copy them instead. The problem is that plt branch stubs currently
trash r12, one of the parameters to some of the save/restore
functions, and there is no free register available to use instead of
r12.
6f20ed8a is prerequisite for this patch.
PR 18878
* elf64-ppc.c (ARRAY_SIZE): Define. Use throughout.
(enum ppc_stub_type): Add ppc_stub_save_res.
(struct map_stub): Add "next" and "needs_save_res".
(struct ppc_link_hash_entry): Add "save_res" flag.
(struct ppc_link_hash_table): Add "group".
(sfpr_define): Add stub_sec param. Define symbol in stub_sec if
stub_sec is non-null. Set "save_res".
(save_res_funcs): Make file scope, rename from funcs. Adjust uses.
(ppc64_elf_adjust_dynamic_symbol): Prohibit plt call to save_res syms.
(ppc_build_one_stub): Handle ppc_stub_save_res.
(ppc_size_one_stub): Set stub type to ppc_size_one_stub on finding
stub for linker defined save_res sym.
(group_sections): Init new fields of struct map_stub.
(ppc64_elf_size_stubs): Reserve space for save/restore func copy.
(ppc64_elf_build_stubs): Copy save/restore funcs to groups. Emit
alias syms too.
(ppc64_elf_relocate_section): Set destination for ppc_stub_save_res.
Sometimes these stubs don't need to change the low 16-bits of r2, so in
that case omit a useless addi r2,r2,0 insn. Also, change the get_r2off
error return from 0 to -1 since 0 is a valid return for ELFv2 -R objects.
* elf64-ppc.c (get_r2off): Return -1 on error.
(ppc_build_one_stub): Adjust for get_r2off change. Don't emit
addi r2,r2,0 on r2off stubs when the low 16-bit delta is zero.
(ppc_size_one_stub): Corresponding size changes for r2off stubs.
Add condition in test for -R objects.
Two organizational changes to the array of additional info kept for
sections.
1) Move group info into a per-group allocated struct, in preparation
for future changes that need per-group accounting.
2) Expand the array to include output sections, which simplifies
sizing and removes the need for a separate output section array.
* section.c (section_id): Make file scope.
(bfd_get_next_section_id): New function.
* elf64-ppc.c (struct map_stub): Remove toc_off field. Move decl.
(struct ppc_stub_hash_entry): Delete stub_sec and id_sec. Add
group. Update all uses.
(struct ppc_link_hash_table): Delete top_id, top_index, and
input_list. Add sec_info_arr_size. Rename stub_group to
sec_info, and make group info indirect. Update stub_group refs
throughout file.
(ppc_add_stub): Don't look for stub_sec on link_sec stub_group
entry.
(ppc_build_one_stub): Delete FIXME.
(ppc64_elf_setup_section_lists): Size htab->sec_info for all
sections, not just input sections. Don't create htab->input_list.
(ppc64_elf_next_input_section): Update to use sec_info union as
list pointer.
(PREV_SEC): Delete.
(group_sections): Pass "info" param rather than "htab". Iterate
over output sections rather than input_list. Use sec_info union
as list pointers. Alloc atruct map_stub, and return fail status.
* bfd-in2.h: Regenerate.
Small clean up to make a local variable const and remove a cast of NULL.
gdb/ChangeLog:
* tui/tui-data.c (tui_win_name): Make local variable const, remove
cast of NULL.
Use XSHAL_ABI value provided by xtensa-config.h to correctly initialize
xtensa_tdep.call_abi
This fixes calls to functions from GDB that otherwise fail with the
following assertion in call0 configuration:
gdb/regcache.c:602: internal-error: regcache_raw_read: Assertion
`regnum >= 0 && regnum < regcache->descr->nr_raw_registers' failed.
2015-08-31 Max Filippov <jcmvbkbc@gmail.com>
gdb/
* xtensa-tdep.h (XTENSA_GDBARCH_TDEP_INSTANTIATE): Initialize
call_abi using XSHAL_ABI macro.
These int vars are really unsigned, so make them so.
* section.c (struct bfd_section): Make "id" and "index" unsigned.
* coff-rs6000.c (_bfd_xcoff_sizeof_headers): Adjust local var to suit.
* elf.c (elf_map_symbols): Likewise.
* elf64-ppc.c (sym_exists_at): Make "id" param unsigned.
(struct ppc_link_hash_table): Make "top_id" and "top_index" unsigned.
(ppc64_elf_setup_section_lists): Ditto for local vars.
* elf32-arm.c: Similarly to elf64-ppc.c.
* elf32-avr.c: Likewise.
* elf32-hppa.c: Likewise.
* elf32-m68hc1x.c: Likewise.
* elf32-metag.c: Likewise.
* elf32-nios2.c: Likewise.
* elfnn-aarch64.c: Likewise.
* simple.c (struct saved_offsets): Make "section_count" unsigned.
* bfd-in2.h: Regenerate.
Since the type whose name is being set is now being allocated on the
gdbarch obstack, we should allocate its TYPE_NAME on the obstack too.
This reduces the number of individual valgrind warnings for the command
"gdb gdb" from ~300 to ~150.
Tested on x86_64-unknown-linux-gnu.
gdb/ChangeLog:
* gdbarch.h (gdbarch_obstack_strdup): Declare.
* gdbarch.c (gdbarch_obstack_strdup): Define.
* gdbtypes.c (arch_type): Use it.
For the command "gdb gdb" valgrind currently reports 100s of individual
memory leaks, 500 of which originate solely out of the function
alloc_type_arch. This function allocates a "struct type" associated
with the given gdbarch using malloc but apparently the types allocated
by this function are never freed.
This patch fixes these leaks by making the function alloc_type_arch
allocate these gdbarch-associated types on the gdbarch obstack instead
of on the general heap. Since, from what I can tell, the types
allocated by this function are all fundamental "wired-in" types, such
types would not benefit from more granular memory management anyway.
They would likely live as long as the gdbarch is alive so allocating
them on the gdbarch obstack makes sense.
With this patch, the number of individual vargrind warnings emitted for
the command "gdb gdb" drops from ~800 to ~300.
Tested on x86_64-unknown-linux-gnu.
gdb/ChangeLog:
* gdbtypes.c (alloc_type_arch): Allocate the type on the given
gdbarch obstack instead of on the heap. Update commentary
accordingly.
solib_ops are installed as a property of the inferior gdbarch,
so they need to be systematically looked up via that architecture,
not some objfile architecture.
ChangeLog:
Revert:
2014-11-06 Doug Evans <xdje42@gmail.com>
* solib.c (solib_global_lookup): Fetch arch from objfile,
not target_gdbarch.
When debugging Cell/B.E. code, the cross-architecture frame unwinding
works by accessing data structures refered to by a thread-local variable
in the inferior. While accessing this variable via minimal symbol,
code would use BMSYMBOL_VALUE_ADDRESS to determine the offset of the
variable in the thread-local storage block.
This is incorrect, since it adds any relocation offset of the shared
library defining the TLS variable. That offset would be OK when
accessing the initial copy present in the .tbss section, but it must
not be applied to the offset of the variable in the TLS block at
runtime. Depending on whether or not the libspe2.so library was
prelinked, access to the variable would fail due to the incorrectly
relocated offset.
ChangeLog:
* ppc-linux-tdep.c (ppc_linux_spe_context_lookup): Do not
attempt to relocate a TLS variable offset.
With recent changes to inferior handling, parse_spufs_run needs to be
more careful in assumptions it makes. In particular, this patch:
- Bails out early if the current inferior has not yet been registered
(e.g. during fork procession) to avoid assertion failures in register
cache code.
- Sets inferior_ptid to the current ptid while calling target_read_memory
to make sure the correct process is accessed if parse_spufs_run is
called early when inferior_ptid has not yet been switched by the caller.
ChangeLog:
* spu-multiarch.c (parse_spufs_run): Bail out if inferior is not
registered yet. Set inferior_ptid while calling target_read_memory.
The Linux target and gdbserver now check the siginfo si_code
reported on a SIGTRAP to detect whether the trap indicates
a software breakpoint was hit.
Unfortunately, on Cell/B.E., the kernel uses an si_code value
of TRAP_BRKPT when a SW breakpoint was hit in PowerPC code,
but a si_code value of SI_KERNEL when a SW breakpoint was
hit in SPU code.
This patch updates Linux target and gdbserver to accept both
si_code values to indicate SW breakpoint on PowerPC.
ChangeLog:
* nat/linux-ptrace.h (GDB_ARCH_TRAP_BRKPT): Replace by ...
(GDB_ARCH_IS_TRAP_BRKPT): ... this. Add __powerpc__ case.
* linux-nat.c (check_stopped_by_breakpoint): Use
GDB_ARCH_IS_TRAP_BRKPT instead of GDB_ARCH_TRAP_BRKPT.
gdbserver/ChangeLog:
* linux-low.c (check_stopped_by_breakpoint): Use
GDB_ARCH_IS_TRAP_BRKPT instead of GDB_ARCH_TRAP_BRKPT.
Since we are no longer using thread events by default in linux-thread-db,
the find_new_threads_once routine contains an assertion that it should
never be called on a live inferior unless using thread events:
gdb_assert (!target_has_execution || thread_db_use_events ());
However, there is a code path from thread_db_get_thread_local_address
that will in fact call find_new_threads_once in some scenarios. In
particular, this is currently always triggered when starting up any
Cell/B.E. combined exeuctable.
To fix this, this patch removes the call to thread_db_find_new_threads_1
when the current thread was not yet detected. In its place, we now just
call thread_from_lwp to detect this one thread if necessary.
ChangeLog:
* linux-thread-db.c (thread_db_get_thread_local_address): If the
thread was not yet discovered, use thread_from_lwp instead of
calling thread_db_find_new_threads_1.
The gdb.cell testcases use the predicate skip_cell_tests defined in
lib/cell.exp to determine whether Cell/B.E. test cases ought to be
run. This tests verifies that we have a toolchain that supports
generating combined Cell/B.E. binaries, and that the target machine
actually is a Cell/B.E.
In order to do so, a small test program is built and run (under the
debugger). Any failure is taken as a sign that we don't have a
Cell/B.E. machine and the tests are to be skipped.
This has the unfortunate effect that a serious bug in GDB that causes
internal compiler errors even on the trivial test program does not
lead to any failures in the testsuite, since now all gdb.cell test
are simply skipped.
This patch changes skip_cell_tests to at least report UNRESOLVED
in cases where execution of the test program fails in unexpected
ways.
testsuite/ChangeLog:
* lib/cell.exp (skip_cell_tests): Report UNRESOLVED on unexpected
failures to run the test program under GDB.
Some of the TC_START_LABEL implementations need to adjust the end of
the symbol, when a colon doesn't mean a label definition. That means
they need access to nul_char both the restore the NUL location (it may
be a quote rather than a colon) and to store the new nul_char. Others
need adjusting to step over a potential trailing quote.
PR gas/18581
* config/tc-aarch64.h (TC_START_LABEL): Redefine.
* config/tc-arm.c (tc_start_label_without_colon): Delete params.
Use input_line_pointer directly.
* config/tc-arm.h (TC_START_LABEL): Redefine.
(TC_START_LABEL_WITHOUT_COLON): Redefine.
(tc_start_label_without_colon): Update prototype.
* config/tc-bfin.c (bfin_start_label): Delete ptr param. Check
for NUL instead.
* config/tc-bfin.h (bfin_start_label): Update prototype.
(TC_START_LABEL): Redefine.
* config/tc-d30v.h (TC_START_LABEL): Redefine.
* config/tc-fr30.c (restore_colon): Rewrite.
(fr30_is_colon_insn): Add nul_char param. Return int. Bump
i_l_p over quote. Update restore_colon calls.
* config/tc-fr30.h (TC_START_LABEL): Redefine.
(fr30_is_colon_insn): Update prototype.
* config/tc-m32c.c (restore_colon, m32c_is_colon_insn): As above.
* config/tc-m32c.h (TC_START_LABEL): Redefine.
(m32c_is_colon_insn): Update prototype.
* config/tc-m32r.h (TC_START_LABEL): Redefine.
* config/tc-mep.h (TC_START_LABEL): Redefine.
* config/tc-nds32.h (TC_START_LABEL): Redefine.
* config/tc-tic54x.c (tic54x_start_label): Replace params with
nul_char and next_char. Step over trailing quote.
* config/tc-tic54x.h (TC_START_LABEL_WITHOUT_COLON): Redefine.
(tic54x_start_label): Update prototype.
* read.c (TC_START_LABEL): Redefine. Update invocation.
(TC_START_LABEL_WITHOUT_COLON): Update invocation.
* config/tc-nios2.c (s_nios2_set): Save initial input_line_pointer
and restore if calling s_set. Don't restore delim again.
These two statements were inverted by mistake in commit "Replace some
xmalloc-family functions with XNEW-family ones". It obviously doesn't
make sense to have them in this order, which is why I am pushing this as
obvious.
gdb/ChangeLog:
* m88k-tdep.c (m88k_analyze_prologue): Fix inverted allocation
statements.
I think it would be a good idea to document that gdb now has (basic)
support to read/write memory on architectures with non-8-bits memory.
Hopefully somebody will see it and say "Hey! We can now (more easily)
port GDB to our strange DSP that has 32-bits-addressable memory!" and do
it.
gdb/ChangeLog:
* NEWS: Document support for non-8-bits addressable memory.
Building in C++ mode shows:
int write_inferior_memory (CORE_ADDR memaddr, const unsigned char *myaddr,
^
src/gdb/gdbserver/proc-service.c:93:64: error: invalid conversion from ‘int’ to ‘ps_err_e’ [-fpermissive]
return write_inferior_memory ((unsigned long) addr, buf, size);
^
It only works today by accident, write_inferior_memory does not return
a ps_err_e.
gdb/gdbserver/ChangeLog:
2015-08-27 Pedro Alves <palves@redhat.com>
* proc-service.c (ps_pdwrite): Return PS_ERR/PS_OK explicily.
Since we're using sighandler_t, nothing else refers to RETSIGTYPE in
gdb.
(Actually, given gdb/remote.c has been assuming signal handlers return
void for a long time, we could have gotten get rid of this even
without gnulib's sighandler_t.)
gdb/ChangeLog:
2015-08-27 Pedro Alves <palves@redhat.com>
* configure.ac: Remove AC_TYPE_SIGNAL call.
* configure, config.in: Regenerate.
This fixes 14 build errors like these in C++ mode:
src/gdb/extension.c: In function ‘void install_sigint_handler(const signal_handler*)’:
src/gdb/extension.c:698:41: error: invalid conversion from ‘void (*)()’ to ‘__sighandler_t {aka void (*)(int)}’ [-fpermissive]
signal (SIGINT, handler_state->handler);
^
In file included from build-gnulib/import/signal.h:52:0,
from ../../src/gdb/extension.c:24:
/usr/include/signal.h:102:23: error: initializing argument 2 of ‘void (* signal(int, __sighandler_t))(int)’ [-fpermissive]
extern __sighandler_t signal (int __sig, __sighandler_t __handler)
^
Instead of this everywhere:
- RETSIGTYPE (*handle_sigint_for_compare) () = handle_sigint;
+ RETSIGTYPE (*handle_sigint_for_compare) (int) = handle_sigint;
Use sighandler_t (a GNU extension). That's OK to use unconditionaly
because gnulib's signal.h replacement makes sure that it is available.
gdb/ChangeLog:
2015-08-27 Pedro Alves <palves@redhat.com>
* cp-support.c (gdb_demangle): Use sighandler_t. Remove cast.
* extension-priv.h: Include signal.h.
(struct signal_handler) <handler>: Change type to sighandler_t.
* extension.c (install_gdb_sigint_handler): Use sighandler_t.
* inflow.c (sigint_ours, sigquit_ours): Change type to
sighandler_t.
(child_terminal_inferior): Remove casts.
(child_terminal_ours_1, new_tty): Use sighandler_t. Remove casts.
(osig): Change type to sighandler_t.
* nto-procfs.c (ofunc): Change type to sighandler_t.
(procfs_wait): Remove casts.
* remote-m32r-sdi.c (m32r_wait, m32r_load): Use sighandler_t.
* remote-sim.c (gdbsim_wait): Use sighandler_t.
* utils.c (wait_to_die_with_timeout): Use sighandler_t.
This gives us a signal.h replacement that makes sure the sighandler_t
typedef (a GNU extension) is always available. A follow up patch will
make use of this.
gdb/ChangeLog:
2015-08-27 Pedro Alves <palves@redhat.com>
* gnulib/update-gnulib.sh (IMPORTED_GNULIB_MODULES): Add signal-h.
* gnulib/aclocal.m4: Renegerate.
* gnulib/config.in: Renegerate.
* gnulib/configure: Renegerate.
* gnulib/import/Makefile.am: Update.
* gnulib/import/Makefile.in: Regenerate.
* gnulib/import/m4/gnulib-cache.m4: Update.
* gnulib/import/m4/gnulib-comp.m4: Update.
* gnulib/import/m4/signal_h.m4: New file.
* gnulib/import/signal.in.h: New file.
The remote packet buffer size is currently capped to 16384 mostly for
historical reasons, related to use of alloca. Stop using alloca and
remove the limitation.
Tested on x86_64 Fedora 20.
gdb/ChangeLog:
2015-08-27 Pedro Alves <palves@redhat.com>
* remote.c (DEFAULT_MAX_MEMORY_PACKET_SIZE)
(MIN_MEMORY_PACKET_SIZE): New.
(MAX_REMOTE_PACKET_SIZE, MIN_REMOTE_PACKET_SIZE): Delete.
(get_memory_packet_size): Adjust. No longer limit the max packet
size.
(set_memory_packet_size): Adjust, and remove dead code.
(remote_check_symbols): Use xmalloc and a cleanup instead of
alloca.
(remote_packet_size): No longer cap the packet size.
(putpkt_binary): Use xmalloc and a cleanup instead of alloca.
This patch improves one of the compile error messages by mentioning the
language.
Before - No compiler support for this language.
After - No compiler support for language <language>.
gdb/ChangeLog:
2015-08-26 Luis Machado <lgustavo@codesourcery.com>
* compile/compile.c (compile_to_object): Mention language in
error message.
Due to the lack of debug information in the binary, GDB is unable to figure
out what language is being used. This may be a problem when doing remote
debugging and the binary stops at the entry point containing asm code.
In this case GDB will switch to asm as current language and will not switch
back to C when it reaches main, which in turn causes the compile feature check
to malfunction.
This is solved by forcing the language to C after reaching main.
gdb/testsuite/ChangeLog:
2015-08-26 Luis Machado <lgustavo@codesourcery.com>
* gdb.compile/compile-ifunc.exp (with_test_prefix): Force language
to C.
We currently set attach_flag when attaching to a process, so we should
make sure to unset it when forking a new process. Otherwise attach_flag
would remain set after forking, if the previous process associated with
the inferior was attached to.
gdb/ChangeLog:
* target.c (target_pre_inferior): Unset attach_flag.
gdb/testsuite/ChangeLog:
* gdb.base/run-after-attach.exp: New test file.
* gdb.base/run-after-attach.c: New test file.
