mirror of
https://sourceware.org/git/binutils-gdb.git
synced 2024-12-09 04:21:49 +08:00
f3c469b95b
With multi-target, each inferior now has its own target connection. The problem in switch_to_program_space_and_thread is that in the current state GDB switches to "no thread" and also sets the program space but because the inferior is not switched, potentially an incorrect target remains selected. Here is a sample scenario that exploits this flow: On terminal 1, start a gdbserver on a program named foo: $ gdbserver :1234 ./foo On terminal 2, start gdb on a program named bar. Suppose foo and bar are compiled from foo.c and bar.c. They are completely separate. So, bar.c:2 has no meaning for foo. $ gdb -q ./bar Reading symbols from ./bar... (gdb) add-inferior [New inferior 2] Added inferior 2 (gdb) inferior 2 [Switching to inferior 2 [<null>] (<noexec>)] (gdb) target remote :1234 ... (gdb) set debug remote 2 (gdb) break bar.c:2 Sending packet: $Hgp0.0#ad...Packet received: OK Sending packet: $m5fa,12#f8...Packet received: E01 Sending packet: $m5fa,1#c6...Packet received: E01 Sending packet: $m5fb,3#c9...Packet received: E01 Sending packet: $m5fe,1#ca...Packet received: E01 Breakpoint 1 at 0x5fe: file bar.c, line 2. (gdb) Here we have an unnecessary sending of the packets to the gdbserver. With this fix in progspace-and-thread.c, we'll get this: (gdb) break bar.c:2 Breakpoint 1 at 0x5fe: file bar.c, line 2. (gdb) Now there is no sending of the packets to gdbserver. The changes around clear_symtab_users calls are necessary because otherwise we regress gdb.base/step-over-exit.exp, hitting the new assertion in switch_to_program_space_and_thread. The problem is, a forked child terminates, and when GDB decides to auto-purge that inferior, GDB tries to switch to the pspace of that no-longer-existing inferior. The root of the problem is within the program_space destructor: program_space::~program_space () { ... set_current_program_space (this); # (1) ... breakpoint_program_space_exit (this); # (2) ... free_all_objfiles (); # (3) ... } We get here from delete_inferior -> delete_program_space. So we're deleting an inferior, and the inferior to be deleted is no longer in the inferior list. At (2), we've deleted all the breakpoints and locations for the program space being deleted. The crash happens while doing a breakpoint re-set, called by clear_symtab_users at the tail end of (3). That is, while recreating breakpoints for the current program space, which is the program space we're tearing down. During breakpoint re-set, we try to switch to the new location's pspace (the current pspace set in (1), so the pspace we're tearing down) with switch_to_program_space_and_thread, and that hits the failed assertion. It's the fact that we recreate breakpoints in the program_space destructor that is the latent bug here. Just don't do that, and we don't end up in the crash situation. My first approach to fix this added a symfile_add_flags parameter to program_space::free_all_objfiles, and then passed that down to clear_symtab_users. The program_space dtor would then pass down SYMFILE_DEFER_BP_RESET to free_all_objfiles. I couldn't help feeling that adding that parameter to free_all_objfiles looked a little awkward, so I settled on something a little different -- hoist the clear_symtab_users call to the callers. There are only two callers. I felt that that didn't look as odd, particularly since remove_symbol_file_command also does: objf->unlink (); clear_symtab_users (0); I.e., objfile deletion is already separate from calling clear_symtab_users in some places. gdb/ChangeLog: 2020-01-10 Aleksandar Paunovic <aleksandar.paunovic@intel.com> Pedro Alves <palves@redhat.com> * progspace-and-thread.c (switch_to_program_space_and_thread): Assert there's an inferior for PSPACE. Use switch_to_inferior_no_thread to switch the inferior too. * progspace.c (program_space::~program_space): Call clear_symtab_users here, with SYMFILE_DEFER_BP_RESET. (program_space::free_all_objfiles): Don't call clear_symtab_users here. * symfile.c (symbol_file_clear): Call clear_symtab_users here. gdb/testsuite/ChangeLog: 2020-01-10 Pedro Alves <palves@redhat.com> * gdb.server/bkpt-other-inferior.exp: New file.
45 lines
1.2 KiB
C
45 lines
1.2 KiB
C
/* Copyright (C) 2009-2020 Free Software Foundation, Inc.
|
|
|
|
This file is part of GDB.
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
it under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation; either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>. */
|
|
|
|
#include "defs.h"
|
|
#include "progspace-and-thread.h"
|
|
#include "inferior.h"
|
|
|
|
/* See progspace-and-thread.h */
|
|
|
|
void
|
|
switch_to_program_space_and_thread (program_space *pspace)
|
|
{
|
|
inferior *inf = find_inferior_for_program_space (pspace);
|
|
gdb_assert (inf != nullptr);
|
|
|
|
if (inf->pid != 0)
|
|
{
|
|
thread_info *tp = any_live_thread_of_inferior (inf);
|
|
|
|
if (tp != NULL)
|
|
{
|
|
switch_to_thread (tp);
|
|
/* Switching thread switches pspace implicitly. We're
|
|
done. */
|
|
return;
|
|
}
|
|
}
|
|
|
|
switch_to_inferior_no_thread (inf);
|
|
}
|