aboutsummaryrefslogtreecommitdiff
path: root/gdb/testsuite/gdb.multi/multi-re-run-1.c
diff options
context:
space:
mode:
authorPedro Alves <palves@redhat.com>2020-01-24 18:46:20 +0000
committerPedro Alves <palves@redhat.com>2020-01-24 18:46:20 +0000
commit53af73bf5e9bad42a76bcc47cdf44d91bbbc4eb7 (patch)
treefc819dc2834b14507fc16fb2db0e17df6bf350db /gdb/testsuite/gdb.multi/multi-re-run-1.c
parentbdaed379e9c19ff38a00c9e210e4f7d56f7c5880 (diff)
downloadgdb-53af73bf5e9bad42a76bcc47cdf44d91bbbc4eb7.zip
gdb-53af73bf5e9bad42a76bcc47cdf44d91bbbc4eb7.tar.gz
gdb-53af73bf5e9bad42a76bcc47cdf44d91bbbc4eb7.tar.bz2
Fix re-runs of a second inferior (PR gdb/25410)
This fixes a latent bug exposed by the multi-target patch (5b6d1e4fa "Multi-target support), and then fixes two other latent bugs exposed by fixing that first latent bug. The symptom described in the bug report is that starting a first inferior, then trying to run a second (multi-threaded) inferior twice, causes libthread_db to fail to load, along with other erratic behavior: (gdb) run Starting program: /tmp/foo warning: td_ta_new failed: generic error Going a bit deeply, I found that if the two inferiors have different symbols, we can see that just after inferior 2 exits, we are left with inferior 2 selected, which is correct, but the symbols in scope belong to inferior 1, which is obviously incorrect... This problem is that there's a path in scoped_restore_current_thread::restore() that switches to no thread selected, and switches the current inferior, but leaves the current program space as is, resulting in leaving the program space pointing to the wrong program space (the one of the other inferior). This was happening after handling TARGET_WAITKIND_NO_RESUMED, which is an event that triggers after TARGET_WAITKIND_EXITED for the previous inferior exit. Subsequent symbol lookups find the symbols of the wrong inferior. The fix is to use switch_to_inferior_no_thread in that problem spot. This function was recently added along with the multi-target work exactly for these situations. As for testing, this patch adds a new testcase that tests symbol printing just after inferior exit, which exercises the root cause of the problem more directly. And then, to cover the use case described in the bug too, it also exercises the lithread_db.so mis-loading, by using TLS printing as a proxy for being sure that threaded debugging was activated sucessfully. The testcase fails without the fix like this, for the "print symbol just after exit" bits: ... [Inferior 1 (process 8719) exited normally] (gdb) PASS: gdb.multi/multi-re-run.exp: re_run_inf=1: iter=1: continue until exit print re_run_var_1 No symbol "re_run_var_1" in current context. (gdb) FAIL: gdb.multi/multi-re-run.exp: re_run_inf=1: iter=1: print re_run_var_1 ... And like this for the "libthread_db.so loading" bits: (gdb) run Starting program: /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.multi/multi-re-run/multi-re-run warning: td_ta_new failed: generic error [New LWP 27001] Thread 1.1 "multi-re-run" hit Breakpoint 3, all_started () at /home/pedro/gdb/binutils-gdb/build/../src/gdb/testsuite/gdb.multi/multi-re-run.c:44 44 } (gdb) PASS: gdb.multi/multi-re-run.exp: re_run_inf=1: iter=2: running to all_started in runto print tls_var Cannot find thread-local storage for LWP 27000, executable file /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.multi/multi-re-run/multi-re-run: Cannot find thread-local variables on this target (gdb) FAIL: gdb.multi/multi-re-run.exp: re_run_inf=1: iter=2: print tls_var As mentioned, that fix above goes on to expose a couple other latent bugs. This commit fixes those as well. The first latent bug exposed is in infrun.c:handle_vfork_child_exec_or_exit. The current code is leaving inf->pspace == NULL while calling clone_program_space. The idea was to make it so that the breakpoints module doesn't use this inferior's pspace to set breakpoints. With that, any scoped_restore_current_thread use from within clone_program_space tries to restore a NULL program space, which hits an assertion: Attaching after Thread 0x7ffff74b8700 (LWP 27276) vfork to child process 27277] [New inferior 2 (process 27277)] [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib64/libthread_db.so.1". /home/pedro/gdb/binutils-gdb/build/../src/gdb/progspace.c:243: internal-error: void set_current_program_space(program_space*): Assertion `pspace != NULL' faile d. A problem internal to GDB has been detected, further debugging may prove unreliable. Quit this debugging session? (y or n) FAIL: gdb.threads/vfork-follow-child-exit.exp: detach-on-fork=off: continue (GDB internal error) That NULL pspace idea was legitimate, but it's no longer necessary, since commit b2e586e850db ("Defer breakpoint reset when cloning progspace for fork child"). So the fix is to just set the inferior's program space earlier. The other latent bug exposed is in exec.c. When exec_close is called from the program_space destructor, it is purposedly called with a current program space that is not the current inferior's program space. The problem is that the multi-target work added some code to remove_target_sections that loops over all inferiors, and uses scoped_restore_current_thread to save/restore the previous thread/inferior/frame state. This makes it so that exec_close returns with the current program space set to the current inferior's program space, which is exactly what we did not want. Then the program_space destructor continues into free_all_objfiles, but it is now running that method on the wrong program space, resulting in: Reading symbols from /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.threads/fork-plus-threads/fork-plus-threads... Reading symbols from /usr/lib/debug/usr/lib64/libpthread-2.26.so.debug... Reading symbols from /usr/lib/debug/usr/lib64/libm-2.26.so.debug... Reading symbols from /usr/lib/debug/usr/lib64/libc-2.26.so.debug... Reading symbols from /usr/lib/debug/usr/lib64/ld-2.26.so.debug... [Inferior 3 (process 9583) exited normally] /home/pedro/gdb/binutils-gdb/build/../src/gdb/progspace.c:170: internal-error: void program_space::free_all_objfiles(): Assertion `so->objfile == NULL' failed. A problem internal to GDB has been detected, further debugging may prove unreliable. Quit this debugging session? (y or n) FAIL: gdb.threads/fork-plus-threads.exp: detach-on-fork=off: inferior 1 exited (GDB internal error) The fix is to use scoped_restore_current_pspace_and_thread instead of scoped_restore_current_thread. gdb/ChangeLog: 2020-01-24 Pedro Alves <palves@redhat.com> PR gdb/25410 * thread.c (scoped_restore_current_thread::restore): Use switch_to_inferior_no_thread. * exec.c: Include "progspace-and-thread.h". (add_target_sections, remove_target_sections): scoped_restore_current_pspace_and_thread instead of scoped_restore_current_thread. * infrun.c (handle_vfork_child_exec_or_exit): Assign the pspace and aspace to the inferior before calling clone_program_space. Remove stale comment. gdb/testsuite/ChangeLog: 2020-01-24 Pedro Alves <palves@redhat.com> PR gdb/25410 * gdb.multi/multi-re-run-1.c: New. * gdb.multi/multi-re-run-2.c: New. * gdb.multi/multi-re-run.exp: New.
Diffstat (limited to 'gdb/testsuite/gdb.multi/multi-re-run-1.c')
-rw-r--r--gdb/testsuite/gdb.multi/multi-re-run-1.c61
1 files changed, 61 insertions, 0 deletions
diff --git a/gdb/testsuite/gdb.multi/multi-re-run-1.c b/gdb/testsuite/gdb.multi/multi-re-run-1.c
new file mode 100644
index 0000000..91b0dbc
--- /dev/null
+++ b/gdb/testsuite/gdb.multi/multi-re-run-1.c
@@ -0,0 +1,61 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright 2020 Free Software Foundation, Inc.
+
+ 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 <stdlib.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <limits.h>
+#include <string.h>
+#include <pthread.h>
+
+int re_run_var_1 = 1;
+
+#define NUM_THREADS 1
+
+__thread int tls_var = 1;
+
+static pthread_barrier_t barrier;
+
+static void *
+thread_start (void *arg)
+{
+ pthread_barrier_wait (&barrier);
+
+ while (1)
+ sleep (1);
+ return NULL;
+}
+
+static void
+all_started (void)
+{
+}
+
+int
+main (int argc, char ** argv)
+{
+ pthread_t thread;
+ int len;
+
+ pthread_barrier_init (&barrier, NULL, NUM_THREADS + 1);
+ pthread_create (&thread, NULL, thread_start, NULL);
+
+ pthread_barrier_wait (&barrier);
+ all_started ();
+
+ return 0;
+}