summaryrefslogtreecommitdiff
path: root/gdb/linux-thread.c
diff options
context:
space:
mode:
authorStan Shebs <shebs@apple.com>1999-09-08 23:58:56 +0000
committerStan Shebs <shebs@apple.com>1999-09-08 23:58:56 +0000
commit23d3706dfc7311a7e9a1a30dafaec5640b24000e (patch)
tree6f7eca0359aa5998c3758a492b2c3cfa986236cf /gdb/linux-thread.c
parent6343bc02175740a12359fa28ab86d83bd08f6ca6 (diff)
downloadgdb-23d3706dfc7311a7e9a1a30dafaec5640b24000e.tar.gz
import gdb-1999-09-08 snapshot
Diffstat (limited to 'gdb/linux-thread.c')
-rw-r--r--gdb/linux-thread.c1637
1 files changed, 1637 insertions, 0 deletions
diff --git a/gdb/linux-thread.c b/gdb/linux-thread.c
new file mode 100644
index 00000000000..9f0a80713b4
--- /dev/null
+++ b/gdb/linux-thread.c
@@ -0,0 +1,1637 @@
+/* Low level interface for debugging GNU/Linux threads for GDB,
+ the GNU debugger.
+ Copyright 1998, 1999 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 2 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, write to the Free Software
+Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
+
+/* This module implements the debugging interface of the linuxthreads package
+ of the glibc. This package implements a simple clone()-based implementation
+ of Posix threads for Linux. To use this module, be sure that you have at
+ least the version of the linuxthreads package that holds the support of
+ GDB (currently 0.8 included in the glibc-2.0.7).
+
+ Right now, the linuxthreads package does not care of priority scheduling,
+ so, neither this module does; In particular, the threads are resumed
+ in any order, which could lead to different scheduling than the one
+ happening when GDB does not control the execution.
+
+ The latest point is that ptrace(PT_ATTACH, ...) is intrusive in Linux:
+ When a process is attached, then the attaching process becomes the current
+ parent of the attached process, and the old parent has lost this child.
+ If the old parent does a wait[...](), then this child is no longer
+ considered by the kernel as a child of the old parent, thus leading to
+ results of the call different when the child is attached and when it's not.
+
+ A fix has been submitted to the Linux community to solve this problem,
+ which consequences are not visible to the application itself, but on the
+ process which may wait() for the completion of the application (mostly,
+ it may consider that the application no longer exists (errno == ECHILD),
+ although it does, and thus being unable to get the exit status and resource
+ usage of the child. If by chance, it is able to wait() for the application
+ after it has died (by receiving first a SIGCHILD, and then doing a wait(),
+ then the exit status and resource usage may be wrong, because the
+ linuxthreads package heavily relies on wait() synchronization to keep
+ them correct. */
+
+#include <sys/types.h> /* for pid_t */
+#include <sys/ptrace.h> /* for PT_* flags */
+#include <sys/wait.h> /* for WUNTRACED and __WCLONE flags */
+#include <signal.h> /* for struct sigaction and NSIG */
+#include <sys/utsname.h>
+
+#include "defs.h"
+#include "target.h"
+#include "inferior.h"
+#include "gdbcore.h"
+#include "gdbthread.h"
+#include "wait.h"
+#include "gdbcmd.h"
+#include "breakpoint.h"
+
+#ifndef PT_ATTACH
+#define PT_ATTACH PTRACE_ATTACH
+#endif
+#ifndef PT_KILL
+#define PT_KILL PTRACE_KILL
+#endif
+#ifndef PT_READ_U
+#define PT_READ_U PTRACE_PEEKUSR
+#endif
+
+#ifdef NSIG
+#define LINUXTHREAD_NSIG NSIG
+#else
+#ifdef _NSIG
+#define LINUXTHREAD_NSIG _NSIG
+#endif
+#endif
+
+extern int child_suppress_run; /* make inftarg.c non-runnable */
+struct target_ops linuxthreads_ops; /* Forward declaration */
+extern struct target_ops child_ops; /* target vector for inftarg.c */
+
+static CORE_ADDR linuxthreads_handles; /* array of linuxthreads handles */
+static CORE_ADDR linuxthreads_manager; /* pid of linuxthreads manager thread */
+static CORE_ADDR linuxthreads_initial; /* pid of linuxthreads initial thread */
+static CORE_ADDR linuxthreads_debug; /* linuxthreads internal debug flag */
+static CORE_ADDR linuxthreads_num; /* number of valid handle entries */
+
+static int linuxthreads_max; /* Maximum number of linuxthreads.
+ Zero if this executable doesn't use
+ threads, or wasn't linked with a
+ debugger-friendly version of the
+ linuxthreads library. */
+
+static int linuxthreads_sizeof_handle; /* size of a linuxthreads handle */
+static int linuxthreads_offset_descr; /* h_descr offset of the linuxthreads
+ handle */
+static int linuxthreads_offset_pid; /* p_pid offset of the linuxthreads
+ descr */
+
+static int linuxthreads_manager_pid; /* manager pid */
+static int linuxthreads_initial_pid; /* initial pid */
+
+/* These variables form a bag of threads with interesting status. If
+ wait_thread (PID) finds that PID stopped for some interesting
+ reason (i.e. anything other than stopped with SIGSTOP), then it
+ records its status in this queue. linuxthreads_wait and
+ linuxthreads_find_trap extract processes from here. */
+static int *linuxthreads_wait_pid; /* wait array of pid */
+static int *linuxthreads_wait_status; /* wait array of status */
+static int linuxthreads_wait_last; /* index of last valid elt in
+ linuxthreads_wait_{pid,status} */
+
+static sigset_t linuxthreads_wait_mask; /* sigset with SIGCHLD */
+
+static int linuxthreads_step_pid; /* current stepped pid */
+static int linuxthreads_step_signo; /* current stepped target signal */
+static int linuxthreads_exit_status; /* exit status of initial thread */
+
+static int linuxthreads_inferior_pid; /* temporary internal inferior pid */
+static int linuxthreads_breakpoint_pid; /* last pid that hit a breakpoint */
+static int linuxthreads_attach_pending; /* attach command without wait */
+
+static int linuxthreads_breakpoints_inserted; /* any breakpoints inserted */
+
+/* LinuxThreads uses certain signals for communication between
+ processes; we need to tell GDB to pass them through silently to the
+ inferior. The LinuxThreads library has global variables we can
+ read containing the relevant signal numbers, but since the signal
+ numbers are chosen at run-time, those variables aren't initialized
+ until the shared library's constructors have had a chance to run. */
+
+struct linuxthreads_signal {
+
+ /* The name of the LinuxThreads library variable that contains
+ the signal number. */
+ char *var;
+
+ /* True if this variable must exist for us to debug properly. */
+ int required;
+
+ /* The variable's address in the inferior, or zero if the
+ LinuxThreads library hasn't been loaded into this inferior yet. */
+ CORE_ADDR addr;
+
+ /* The signal number, or zero if we don't know yet (either because
+ we haven't found the variable, or it hasn't been initialized).
+ This is an actual target signal number that you could pass to
+ `kill', not a GDB signal number. */
+ int signal;
+
+ /* GDB's original settings for `stop' and `print' for this signal.
+ We restore them when the user selects a different executable.
+ Invariant: if sig->signal != 0, then sig->{stop,print} contain
+ the original settings. */
+ int stop, print;
+};
+
+struct linuxthreads_signal linuxthreads_sig_restart = {
+ "__pthread_sig_restart", 1, 0, 0, 0
+};
+struct linuxthreads_signal linuxthreads_sig_cancel = {
+ "__pthread_sig_cancel", 1, 0, 0, 0
+};
+struct linuxthreads_signal linuxthreads_sig_debug = {
+ "__pthread_sig_debug", 0, 0, 0, 0
+};
+
+/* A table of breakpoint locations, one per PID. */
+static struct linuxthreads_breakpoint {
+ CORE_ADDR pc; /* PC of breakpoint */
+ int pid; /* pid of breakpoint */
+ int step; /* whether the pc has been reached after sstep */
+} *linuxthreads_breakpoint_zombie; /* Zombie breakpoints array */
+static int linuxthreads_breakpoint_last; /* Last zombie breakpoint */
+
+/* linuxthreads_{insert,remove}_breakpoint pass the breakpoint address
+ to {insert,remove}_breakpoint via this variable, since
+ iterate_active_threads doesn't provide any way to pass values
+ through to the worker function. */
+static CORE_ADDR linuxthreads_breakpoint_addr;
+
+#define REMOVE_BREAKPOINT_ZOMBIE(_i) \
+{ \
+ if ((_i) < linuxthreads_breakpoint_last) \
+ linuxthreads_breakpoint_zombie[(_i)] = \
+ linuxthreads_breakpoint_zombie[linuxthreads_breakpoint_last]; \
+ linuxthreads_breakpoint_last--; \
+}
+
+
+
+#ifndef PTRACE_XFER_TYPE
+#define PTRACE_XFER_TYPE int
+#endif
+/* Check to see if the given thread is alive. */
+static int
+linuxthreads_thread_alive (pid)
+ int pid;
+{
+ errno = 0;
+ return ptrace (PT_READ_U, pid, (PTRACE_ARG3_TYPE)0, 0) >= 0 || errno == 0;
+}
+
+/* On detach(), find a SIGTRAP status. If stop is non-zero, find a
+ SIGSTOP one, too.
+
+ Make sure PID is ready to run, and free of interference from our
+ efforts to debug it (e.g., pending SIGSTOP or SIGTRAP signals). If
+ STOP is zero, just look for a SIGTRAP. If STOP is non-zero, look
+ for a SIGSTOP, too. Return non-zero if PID is alive and ready to
+ run; return zero if PID is dead.
+
+ PID may or may not be stopped at the moment, and we may or may not
+ have waited for it already. We check the linuxthreads_wait bag in
+ case we've already got a status for it. We may possibly wait for
+ it ourselves.
+
+ PID may have signals waiting to be delivered. If they're caused by
+ our efforts to debug it, accept them with wait, but don't pass them
+ through to PID. Do pass all other signals through. */
+static int
+linuxthreads_find_trap (pid, stop)
+ int pid;
+ int stop;
+{
+ int i;
+ int rpid;
+ int status;
+ int found_stop = 0;
+ int found_trap = 0;
+
+ /* PID may have any number of signals pending. The kernel will
+ report each of them to us via wait, and then it's up to us to
+ pass them along to the process via ptrace, if we so choose.
+
+ We need to paw through the whole set until we've found a SIGTRAP
+ (or a SIGSTOP, if `stop' is set). We don't pass the SIGTRAP (or
+ SIGSTOP) through, but we do re-send all the others, so PID will
+ receive them when we resume it. */
+ int *wstatus = alloca (LINUXTHREAD_NSIG * sizeof (int));
+ int last = 0;
+
+ /* Look at the pending status */
+ for (i = linuxthreads_wait_last; i >= 0; i--)
+ if (linuxthreads_wait_pid[i] == pid)
+ {
+ status = linuxthreads_wait_status[i];
+
+ /* Delete the i'th member of the table. Since the table is
+ unordered, we can do this simply by copying the table's
+ last element to the i'th position, and shrinking the table
+ by one element. */
+ if (i < linuxthreads_wait_last)
+ {
+ linuxthreads_wait_status[i] =
+ linuxthreads_wait_status[linuxthreads_wait_last];
+ linuxthreads_wait_pid[i] =
+ linuxthreads_wait_pid[linuxthreads_wait_last];
+ }
+ linuxthreads_wait_last--;
+
+ if (!WIFSTOPPED(status)) /* Thread has died */
+ return 0;
+
+ if (WSTOPSIG(status) == SIGTRAP)
+ {
+ if (stop)
+ found_trap = 1;
+ else
+ return 1;
+ }
+ else if (WSTOPSIG(status) == SIGSTOP)
+ {
+ if (stop)
+ found_stop = 1;
+ }
+ else
+ {
+ wstatus[0] = status;
+ last = 1;
+ }
+
+ break;
+ }
+
+ if (stop)
+ {
+ /* Make sure that we'll find what we're looking for. */
+ if (!found_trap)
+ kill (pid, SIGTRAP);
+ if (!found_stop)
+ kill (pid, SIGSTOP);
+ }
+
+ /* Catch all status until SIGTRAP and optionally SIGSTOP show up. */
+ for (;;)
+ {
+ child_resume (pid, 1, TARGET_SIGNAL_0);
+
+ for (;;)
+ {
+ rpid = waitpid (pid, &status, __WCLONE);
+ if (rpid > 0)
+ break;
+ if (errno == EINTR)
+ continue;
+
+ /* There are a few reasons the wait call above may have
+ failed. If the thread manager dies, its children get
+ reparented, and this interferes with GDB waiting for
+ them, in some cases. Another possibility is that the
+ initial thread was not cloned, so calling wait with
+ __WCLONE won't find it. I think neither of these should
+ occur in modern Linux kernels --- they don't seem to in
+ 2.0.36. */
+ rpid = waitpid (pid, &status, 0);
+ if (rpid > 0)
+ break;
+ if (errno != EINTR)
+ perror_with_name ("waitpid");
+ }
+
+ if (!WIFSTOPPED(status)) /* Thread has died */
+ return 0;
+
+ if (WSTOPSIG(status) == SIGTRAP)
+ if (!stop || found_stop)
+ break;
+ else
+ found_trap = 1;
+ else if (WSTOPSIG(status) != SIGSTOP)
+ wstatus[last++] = status;
+ else if (stop)
+ if (found_trap)
+ break;
+ else
+ found_stop = 1;
+ }
+
+ /* Resend any other signals we noticed to the thread, to be received
+ when we continue it. */
+ while (--last >= 0)
+ kill (pid, WSTOPSIG(wstatus[last]));
+
+ return 1;
+}
+
+/* Cleanup stub for save_inferior_pid. */
+static void
+restore_inferior_pid (arg)
+ void *arg;
+{
+ int pid = (int) arg;
+ inferior_pid = pid;
+}
+
+/* Register a cleanup to restore the value of inferior_pid. */
+static struct cleanup *
+save_inferior_pid ()
+{
+ return make_cleanup (restore_inferior_pid, (void *) inferior_pid);
+}
+
+static void
+sigchld_handler (signo)
+ int signo;
+{
+ /* This handler is used to get an EINTR while doing waitpid()
+ when an event is received */
+}
+
+/* Have we already collected a wait status for PID in the
+ linuxthreads_wait bag? */
+static int
+linuxthreads_pending_status (pid)
+ int pid;
+{
+ int i;
+ for (i = linuxthreads_wait_last; i >= 0; i--)
+ if (linuxthreads_wait_pid[i] == pid)
+ return 1;
+ return 0;
+}
+
+
+/* Internal linuxthreads signal management */
+
+/* Check in OBJFILE for the variable that holds the number for signal SIG.
+ We assume that we've already found other LinuxThreads-ish variables
+ in OBJFILE, so we complain if it's required, but not there.
+ Return true iff things are okay. */
+static int
+find_signal_var (sig, objfile)
+ struct linuxthreads_signal *sig;
+ struct objfile *objfile;
+{
+ struct minimal_symbol *ms = lookup_minimal_symbol (sig->var, NULL, objfile);
+
+ if (! ms)
+ {
+ if (sig->required)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ sig->var);
+ return 0;
+ }
+ else
+ {
+ sig->addr = 0;
+ return 1;
+ }
+ }
+
+ sig->addr = SYMBOL_VALUE_ADDRESS (ms);
+
+ return 1;
+}
+
+static int
+find_all_signal_vars (objfile)
+ struct objfile *objfile;
+{
+ return ( find_signal_var (&linuxthreads_sig_restart, objfile)
+ && find_signal_var (&linuxthreads_sig_cancel, objfile)
+ && find_signal_var (&linuxthreads_sig_debug, objfile));
+}
+
+/* A struct complaint isn't appropriate here. */
+static int complained_cannot_determine_thread_signal_number = 0;
+
+/* Check to see if the variable holding the signal number for SIG has
+ been initialized yet. If it has, tell GDB to pass that signal
+ through to the inferior silently. */
+static void
+check_signal_number (sig)
+ struct linuxthreads_signal *sig;
+{
+ int num;
+
+ if (sig->signal)
+ /* We already know this signal number. */
+ return;
+
+ if (! sig->addr)
+ /* We don't know the variable's address yet. */
+ return;
+
+ if (target_read_memory (sig->addr, (char *)&num, sizeof (num))
+ != 0)
+ {
+ /* If this happens once, it'll probably happen for all the
+ signals, so only complain once. */
+ if (! complained_cannot_determine_thread_signal_number)
+ warning ("Cannot determine thread signal number; "
+ "GDB may report spurious signals.");
+ complained_cannot_determine_thread_signal_number = 1;
+ return;
+ }
+
+ if (num == 0)
+ /* It hasn't been initialized yet. */
+ return;
+
+ /* We know sig->signal was zero, and is becoming non-zero, so it's
+ okay to sample GDB's original settings. */
+ sig->signal = num;
+ sig->stop = signal_stop_update (target_signal_from_host (num), 0);
+ sig->print = signal_print_update (target_signal_from_host (num), 0);
+}
+
+
+static void
+check_all_signal_numbers ()
+{
+ /* If this isn't a LinuxThreads program, quit early. */
+ if (! linuxthreads_max)
+ return;
+
+ check_signal_number (&linuxthreads_sig_restart);
+ check_signal_number (&linuxthreads_sig_cancel);
+ check_signal_number (&linuxthreads_sig_debug);
+
+ /* handle linuxthread exit */
+ if (linuxthreads_sig_debug.signal
+ || linuxthreads_sig_restart.signal)
+ {
+ struct sigaction sact;
+
+ sact.sa_handler = sigchld_handler;
+ sigemptyset(&sact.sa_mask);
+ sact.sa_flags = 0;
+ if (linuxthreads_sig_debug.signal > 0)
+ sigaction(linuxthreads_sig_cancel.signal, &sact, NULL);
+ else
+ sigaction(linuxthreads_sig_restart.signal, &sact, NULL);
+ }
+}
+
+
+/* Restore GDB's original settings for SIG.
+ This should only be called when we're no longer sure if we're
+ talking to an executable that uses LinuxThreads, so we clear the
+ signal number and variable address too. */
+static void
+restore_signal (sig)
+ struct linuxthreads_signal *sig;
+{
+ if (! sig->signal)
+ return;
+
+ /* We know sig->signal was non-zero, and is becoming zero, so it's
+ okay to restore GDB's original settings. */
+ signal_stop_update (target_signal_from_host (sig->signal), sig->stop);
+ signal_print_update (target_signal_from_host (sig->signal), sig->print);
+
+ sig->signal = 0;
+ sig->addr = 0;
+}
+
+
+/* Restore GDB's original settings for all LinuxThreads signals.
+ This should only be called when we're no longer sure if we're
+ talking to an executable that uses LinuxThreads, so we clear the
+ signal number and variable address too. */
+static void
+restore_all_signals ()
+{
+ restore_signal (&linuxthreads_sig_restart);
+ restore_signal (&linuxthreads_sig_cancel);
+ restore_signal (&linuxthreads_sig_debug);
+
+ /* If it happens again, we should complain again. */
+ complained_cannot_determine_thread_signal_number = 0;
+}
+
+
+
+
+/* Apply FUNC to the pid of each active thread. This consults the
+ inferior's handle table to find active threads.
+
+ If ALL is non-zero, process all threads.
+ If ALL is zero, skip threads with pending status. */
+static void
+iterate_active_threads (func, all)
+ void (*func)(int);
+ int all;
+{
+ CORE_ADDR descr;
+ int pid;
+ int i;
+ int num;
+
+ read_memory (linuxthreads_num, (char *)&num, sizeof (int));
+
+ for (i = 0; i < linuxthreads_max && num > 0; i++)
+ {
+ read_memory (linuxthreads_handles +
+ linuxthreads_sizeof_handle * i + linuxthreads_offset_descr,
+ (char *)&descr, sizeof (void *));
+ if (descr)
+ {
+ num--;
+ read_memory (descr + linuxthreads_offset_pid,
+ (char *)&pid, sizeof (pid_t));
+ if (pid > 0 && pid != linuxthreads_manager_pid
+ && (all || (!linuxthreads_pending_status (pid))))
+ (*func)(pid);
+ }
+ }
+
+}
+
+/* Insert a thread breakpoint at linuxthreads_breakpoint_addr.
+ This is the worker function for linuxthreads_insert_breakpoint,
+ which passes it to iterate_active_threads. */
+static void
+insert_breakpoint (pid)
+ int pid;
+{
+ int j;
+
+ /* Remove (if any) the positive zombie breakpoint. */
+ for (j = linuxthreads_breakpoint_last; j >= 0; j--)
+ if (linuxthreads_breakpoint_zombie[j].pid == pid)
+ {
+ if ((linuxthreads_breakpoint_zombie[j].pc - DECR_PC_AFTER_BREAK
+ == linuxthreads_breakpoint_addr)
+ && !linuxthreads_breakpoint_zombie[j].step)
+ REMOVE_BREAKPOINT_ZOMBIE(j);
+ break;
+ }
+}
+
+/* Note that we're about to remove a thread breakpoint at
+ linuxthreads_breakpoint_addr.
+
+ This is the worker function for linuxthreads_remove_breakpoint,
+ which passes it to iterate_active_threads. The actual work of
+ overwriting the breakpoint instruction is done by
+ child_ops.to_remove_breakpoint; here, we simply create a zombie
+ breakpoint if the thread's PC is pointing at the breakpoint being
+ removed. */
+static void
+remove_breakpoint (pid)
+ int pid;
+{
+ int j;
+
+ /* Insert a positive zombie breakpoint (if needed). */
+ for (j = 0; j <= linuxthreads_breakpoint_last; j++)
+ if (linuxthreads_breakpoint_zombie[j].pid == pid)
+ break;
+
+ if (in_thread_list (pid) && linuxthreads_thread_alive (pid))
+ {
+ CORE_ADDR pc = read_pc_pid (pid);
+ if (linuxthreads_breakpoint_addr == pc - DECR_PC_AFTER_BREAK
+ && j > linuxthreads_breakpoint_last)
+ {
+ linuxthreads_breakpoint_zombie[j].pid = pid;
+ linuxthreads_breakpoint_zombie[j].pc = pc;
+ linuxthreads_breakpoint_zombie[j].step = 0;
+ linuxthreads_breakpoint_last++;
+ }
+ }
+}
+
+/* Kill a thread */
+static void
+kill_thread (pid)
+ int pid;
+{
+ if (in_thread_list (pid))
+ ptrace (PT_KILL, pid, (PTRACE_ARG3_TYPE) 0, 0);
+ else
+ kill (pid, SIGKILL);
+}
+
+/* Resume a thread */
+static void
+resume_thread (pid)
+ int pid;
+{
+ if (pid != inferior_pid
+ && in_thread_list (pid)
+ && linuxthreads_thread_alive (pid))
+ if (pid == linuxthreads_step_pid)
+ child_resume (pid, 1, linuxthreads_step_signo);
+ else
+ child_resume (pid, 0, TARGET_SIGNAL_0);
+}
+
+/* Detach a thread */
+static void
+detach_thread (pid)
+ int pid;
+{
+ if (in_thread_list (pid) && linuxthreads_thread_alive (pid))
+ {
+ /* Remove pending SIGTRAP and SIGSTOP */
+ linuxthreads_find_trap (pid, 1);
+
+ inferior_pid = pid;
+ detach (TARGET_SIGNAL_0);
+ inferior_pid = linuxthreads_manager_pid;
+ }
+}
+
+/* Stop a thread */
+static void
+stop_thread (pid)
+ int pid;
+{
+ if (pid != inferior_pid)
+ if (in_thread_list (pid))
+ kill (pid, SIGSTOP);
+ else if (ptrace (PT_ATTACH, pid, (PTRACE_ARG3_TYPE) 0, 0) == 0)
+ {
+ if (!linuxthreads_attach_pending)
+ printf_unfiltered ("[New %s]\n", target_pid_to_str (pid));
+ add_thread (pid);
+ if (linuxthreads_sig_debug.signal)
+ /* After a new thread in glibc 2.1 signals gdb its existence,
+ it suspends itself and wait for linuxthreads_sig_restart,
+ now we can wake up it. */
+ kill (pid, linuxthreads_sig_restart.signal);
+ }
+ else
+ perror_with_name ("ptrace in stop_thread");
+}
+
+/* Wait for a thread */
+static void
+wait_thread (pid)
+ int pid;
+{
+ int status;
+ int rpid;
+
+ if (pid != inferior_pid && in_thread_list (pid))
+ {
+ for (;;)
+ {
+ /* Get first pid status. */
+ rpid = waitpid(pid, &status, __WCLONE);
+ if (rpid > 0)
+ break;
+ if (errno == EINTR)
+ continue;
+
+ /* There are two reasons this might have failed:
+
+ 1) PID is the initial thread, which wasn't cloned, so
+ passing the __WCLONE flag to waitpid prevented us from
+ finding it.
+
+ 2) The manager thread is the parent of all but the
+ initial thread; if it dies, the children will all be
+ reparented to init, which will wait for them. This means
+ our call to waitpid won't find them.
+
+ Actually, based on a casual look at the 2.0.36 kernel
+ code, I don't think either of these cases happen. But I
+ don't have things set up for remotely debugging the
+ kernel, so I'm not sure. And perhaps older kernels
+ didn't work. */
+ rpid = waitpid(pid, &status, 0);
+ if (rpid > 0)
+ break;
+ if (errno != EINTR && linuxthreads_thread_alive (pid))
+ perror_with_name ("waitpid");
+
+ /* the thread is dead. */
+ return;
+ }
+ if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGSTOP)
+ {
+ linuxthreads_wait_pid[++linuxthreads_wait_last] = pid;
+ linuxthreads_wait_status[linuxthreads_wait_last] = status;
+ }
+ }
+}
+
+/* Walk through the linuxthreads handles in order to detect all
+ threads and stop them */
+static void
+update_stop_threads (test_pid)
+ int test_pid;
+{
+ struct cleanup *old_chain = NULL;
+
+ check_all_signal_numbers ();
+
+ if (linuxthreads_manager_pid == 0)
+ {
+ if (linuxthreads_manager)
+ {
+ if (test_pid > 0 && test_pid != inferior_pid)
+ {
+ old_chain = save_inferior_pid ();
+ inferior_pid = test_pid;
+ }
+ read_memory (linuxthreads_manager,
+ (char *)&linuxthreads_manager_pid, sizeof (pid_t));
+ }
+ if (linuxthreads_initial)
+ {
+ if (test_pid > 0 && test_pid != inferior_pid)
+ {
+ old_chain = save_inferior_pid ();
+ inferior_pid = test_pid;
+ }
+ read_memory(linuxthreads_initial,
+ (char *)&linuxthreads_initial_pid, sizeof (pid_t));
+ }
+ }
+
+ if (linuxthreads_manager_pid != 0)
+ {
+ if (old_chain == NULL && test_pid > 0 &&
+ test_pid != inferior_pid && linuxthreads_thread_alive (test_pid))
+ {
+ old_chain = save_inferior_pid ();
+ inferior_pid = test_pid;
+ }
+
+ if (linuxthreads_thread_alive (inferior_pid))
+ {
+ if (test_pid > 0)
+ {
+ if (test_pid != linuxthreads_manager_pid
+ && !linuxthreads_pending_status (linuxthreads_manager_pid))
+ {
+ stop_thread (linuxthreads_manager_pid);
+ wait_thread (linuxthreads_manager_pid);
+ }
+ if (!in_thread_list (test_pid))
+ {
+ if (!linuxthreads_attach_pending)
+ printf_unfiltered ("[New %s]\n",
+ target_pid_to_str (test_pid));
+ add_thread (test_pid);
+ if (linuxthreads_sig_debug.signal
+ && inferior_pid == test_pid)
+ /* After a new thread in glibc 2.1 signals gdb its
+ existence, it suspends itself and wait for
+ linuxthreads_sig_restart, now we can wake up
+ it. */
+ kill (test_pid, linuxthreads_sig_restart.signal);
+ }
+ }
+ iterate_active_threads (stop_thread, 0);
+ iterate_active_threads (wait_thread, 0);
+ }
+ }
+
+ if (old_chain != NULL)
+ do_cleanups (old_chain);
+}
+
+/* This routine is called whenever a new symbol table is read in, or when all
+ symbol tables are removed. libpthread can only be initialized when it
+ finds the right variables in libpthread.so. Since it's a shared library,
+ those variables don't show up until the library gets mapped and the symbol
+ table is read in. */
+
+void
+linuxthreads_new_objfile (objfile)
+ struct objfile *objfile;
+{
+ struct minimal_symbol *ms;
+
+ if (!objfile)
+ {
+ /* We're starting an entirely new executable, so we can no
+ longer be sure that it uses LinuxThreads. Restore the signal
+ flags to their original states. */
+ restore_all_signals ();
+
+ /* Indicate that we don't know anything's address any more. */
+ linuxthreads_max = 0;
+
+ return;
+ }
+
+ /* If we've already found our variables in another objfile, don't
+ bother looking for them again. */
+ if (linuxthreads_max)
+ return;
+
+ if (! lookup_minimal_symbol ("__pthread_initial_thread", NULL, objfile))
+ /* This object file isn't the pthreads library. */
+ return;
+
+ if ((ms = lookup_minimal_symbol ("__pthread_threads_debug",
+ NULL, objfile)) == NULL)
+ {
+ /* The debugging-aware libpthreads is not present in this objfile */
+ warning ("\
+This program seems to use POSIX threads, but the thread library used\n\
+does not support debugging. This may make using GDB difficult. Don't\n\
+set breakpoints or single-step through code that might be executed by\n\
+any thread other than the main thread.");
+ return;
+ }
+ linuxthreads_debug = SYMBOL_VALUE_ADDRESS (ms);
+
+ /* Read internal structures configuration */
+ if ((ms = lookup_minimal_symbol ("__pthread_sizeof_handle",
+ NULL, objfile)) == NULL
+ || target_read_memory (SYMBOL_VALUE_ADDRESS (ms),
+ (char *)&linuxthreads_sizeof_handle,
+ sizeof (linuxthreads_sizeof_handle)) != 0)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_sizeof_handle");
+ return;
+ }
+
+ if ((ms = lookup_minimal_symbol ("__pthread_offsetof_descr",
+ NULL, objfile)) == NULL
+ || target_read_memory (SYMBOL_VALUE_ADDRESS (ms),
+ (char *)&linuxthreads_offset_descr,
+ sizeof (linuxthreads_offset_descr)) != 0)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_offsetof_descr");
+ return;
+ }
+
+ if ((ms = lookup_minimal_symbol ("__pthread_offsetof_pid",
+ NULL, objfile)) == NULL
+ || target_read_memory (SYMBOL_VALUE_ADDRESS (ms),
+ (char *)&linuxthreads_offset_pid,
+ sizeof (linuxthreads_offset_pid)) != 0)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_offsetof_pid");
+ return;
+ }
+
+ if (! find_all_signal_vars (objfile))
+ return;
+
+ /* Read adresses of internal structures to access */
+ if ((ms = lookup_minimal_symbol ("__pthread_handles",
+ NULL, objfile)) == NULL)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_handles");
+ return;
+ }
+ linuxthreads_handles = SYMBOL_VALUE_ADDRESS (ms);
+
+ if ((ms = lookup_minimal_symbol ("__pthread_handles_num",
+ NULL, objfile)) == NULL)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_handles_num");
+ return;
+ }
+ linuxthreads_num = SYMBOL_VALUE_ADDRESS (ms);
+
+ if ((ms = lookup_minimal_symbol ("__pthread_manager_thread",
+ NULL, objfile)) == NULL)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_manager_thread");
+ return;
+ }
+ linuxthreads_manager = SYMBOL_VALUE_ADDRESS (ms) + linuxthreads_offset_pid;
+
+ if ((ms = lookup_minimal_symbol ("__pthread_initial_thread",
+ NULL, objfile)) == NULL)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_initial_thread");
+ return;
+ }
+ linuxthreads_initial = SYMBOL_VALUE_ADDRESS (ms) + linuxthreads_offset_pid;
+
+ /* Search for this last, so it won't be set to a non-zero value unless
+ we successfully found all the symbols above. */
+ if ((ms = lookup_minimal_symbol ("__pthread_threads_max",
+ NULL, objfile)) == NULL
+ || target_read_memory (SYMBOL_VALUE_ADDRESS (ms),
+ (char *)&linuxthreads_max,
+ sizeof (linuxthreads_max)) != 0)
+ {
+ fprintf_unfiltered (gdb_stderr,
+ "Unable to find linuxthreads symbol \"%s\"\n",
+ "__pthread_threads_max");
+ return;
+ }
+
+ /* Allocate gdb internal structures */
+ linuxthreads_wait_pid =
+ (int *) xmalloc (sizeof (int) * (linuxthreads_max + 1));
+ linuxthreads_wait_status =
+ (int *) xmalloc (sizeof (int) * (linuxthreads_max + 1));
+ linuxthreads_breakpoint_zombie = (struct linuxthreads_breakpoint *)
+ xmalloc (sizeof (struct linuxthreads_breakpoint) * (linuxthreads_max + 1));
+
+ if (inferior_pid && !linuxthreads_attach_pending)
+ {
+ int on = 1;
+ target_write_memory (linuxthreads_debug, (char *)&on, sizeof (on));
+ linuxthreads_attach_pending = 1;
+ update_stop_threads (inferior_pid);
+ linuxthreads_attach_pending = 0;
+ }
+}
+
+/* If we have switched threads from a one that stopped at breakpoint,
+ return 1 otherwise 0. */
+
+int
+linuxthreads_prepare_to_proceed (step)
+ int step;
+{
+ if (!linuxthreads_max
+ || !linuxthreads_manager_pid
+ || !linuxthreads_breakpoint_pid
+ || !breakpoint_here_p (read_pc_pid (linuxthreads_breakpoint_pid)))
+ return 0;
+
+ if (step)
+ {
+ /* Mark the current inferior as single stepping process. */
+ linuxthreads_step_pid = inferior_pid;
+ }
+
+ linuxthreads_inferior_pid = linuxthreads_breakpoint_pid;
+ return linuxthreads_breakpoint_pid;
+}
+
+/* Convert a pid to printable form. */
+
+char *
+linuxthreads_pid_to_str (pid)
+ int pid;
+{
+ static char buf[100];
+
+ sprintf (buf, "%s %d%s", linuxthreads_max ? "Thread" : "Pid", pid,
+ (pid == linuxthreads_manager_pid) ? " (manager thread)"
+ : (pid == linuxthreads_initial_pid) ? " (initial thread)"
+ : "");
+
+ return buf;
+}
+
+/* Attach to process PID, then initialize for debugging it
+ and wait for the trace-trap that results from attaching. */
+
+static void
+linuxthreads_attach (args, from_tty)
+ char *args;
+ int from_tty;
+{
+ if (!args)
+ error_no_arg ("process-id to attach");
+
+ push_target (&linuxthreads_ops);
+ linuxthreads_breakpoints_inserted = 1;
+ linuxthreads_breakpoint_last = -1;
+ linuxthreads_wait_last = -1;
+ linuxthreads_exit_status = __W_STOPCODE(0);
+
+ child_ops.to_attach (args, from_tty);
+
+ if (linuxthreads_max)
+ linuxthreads_attach_pending = 1;
+}
+
+/* Take a program previously attached to and detaches it.
+ The program resumes execution and will no longer stop
+ on signals, etc. We'd better not have left any breakpoints
+ in the program or it'll die when it hits one. For this
+ to work, it may be necessary for the process to have been
+ previously attached. It *might* work if the program was
+ started via the normal ptrace (PTRACE_TRACEME). */
+
+static void
+linuxthreads_detach (args, from_tty)
+ char *args;
+ int from_tty;
+{
+ if (linuxthreads_max)
+ {
+ int i;
+ int pid;
+ int off = 0;
+ target_write_memory (linuxthreads_debug, (char *)&off, sizeof (off));
+
+ /* Walk through linuxthreads array in order to detach known threads. */
+ if (linuxthreads_manager_pid != 0)
+ {
+ /* Get rid of all positive zombie breakpoints. */
+ for (i = 0; i <= linuxthreads_breakpoint_last; i++)
+ {
+ if (linuxthreads_breakpoint_zombie[i].step)
+ continue;
+
+ pid = linuxthreads_breakpoint_zombie[i].pid;
+ if (!linuxthreads_thread_alive (pid))
+ continue;
+
+ if (linuxthreads_breakpoint_zombie[i].pc != read_pc_pid (pid))
+ continue;
+
+ /* Continue in STEP mode until the thread pc has moved or
+ until SIGTRAP is found on the same PC. */
+ if (linuxthreads_find_trap (pid, 0)
+ && linuxthreads_breakpoint_zombie[i].pc == read_pc_pid (pid))
+ write_pc_pid (linuxthreads_breakpoint_zombie[i].pc
+ - DECR_PC_AFTER_BREAK, pid);
+ }
+
+ /* Detach thread after thread. */
+ inferior_pid = linuxthreads_manager_pid;
+ iterate_active_threads (detach_thread, 1);
+
+ /* Remove pending SIGTRAP and SIGSTOP */
+ linuxthreads_find_trap (inferior_pid, 1);
+
+ linuxthreads_wait_last = -1;
+ linuxthreads_exit_status = __W_STOPCODE(0);
+ }
+
+ linuxthreads_inferior_pid = 0;
+ linuxthreads_breakpoint_pid = 0;
+ linuxthreads_step_pid = 0;
+ linuxthreads_step_signo = TARGET_SIGNAL_0;
+ linuxthreads_manager_pid = 0;
+ linuxthreads_initial_pid = 0;
+ linuxthreads_attach_pending = 0;
+ init_thread_list (); /* Destroy thread info */
+ }
+
+ child_ops.to_detach (args, from_tty);
+
+ unpush_target (&linuxthreads_ops);
+}
+
+/* Resume execution of process PID. If STEP is nozero, then
+ just single step it. If SIGNAL is nonzero, restart it with that
+ signal activated. */
+
+static void
+linuxthreads_resume (pid, step, signo)
+ int pid;
+ int step;
+ enum target_signal signo;
+{
+ if (!linuxthreads_max || stop_soon_quietly || linuxthreads_manager_pid == 0)
+ child_ops.to_resume (pid, step, signo);
+ else
+ {
+ int rpid;
+ if (linuxthreads_inferior_pid)
+ {
+ /* Prepare resume of the last thread that hit a breakpoint */
+ linuxthreads_breakpoints_inserted = 0;
+ rpid = linuxthreads_inferior_pid;
+ linuxthreads_step_signo = signo;
+ }
+ else
+ {
+ struct cleanup *old_chain = NULL;
+ int i;
+
+ if (pid < 0)
+ {
+ linuxthreads_step_pid = step ? inferior_pid : 0;
+ linuxthreads_step_signo = signo;
+ rpid = inferior_pid;
+ }
+ else
+ rpid = pid;
+
+ if (pid < 0 || !step)
+ {
+ linuxthreads_breakpoints_inserted = 1;
+
+ /* Walk through linuxthreads array in order to resume threads */
+ if (pid >= 0 && inferior_pid != pid)
+ {
+ old_chain = save_inferior_pid ();
+ inferior_pid = pid;
+ }
+
+ iterate_active_threads (resume_thread, 0);
+ if (linuxthreads_manager_pid != inferior_pid
+ && !linuxthreads_pending_status (linuxthreads_manager_pid))
+ resume_thread (linuxthreads_manager_pid);
+ }
+ else
+ linuxthreads_breakpoints_inserted = 0;
+
+ /* Deal with zombie breakpoint */
+ for (i = 0; i <= linuxthreads_breakpoint_last; i++)
+ if (linuxthreads_breakpoint_zombie[i].pid == rpid)
+ {
+ if (linuxthreads_breakpoint_zombie[i].pc != read_pc_pid (rpid))
+ {
+ /* The current pc is out of zombie breakpoint. */
+ REMOVE_BREAKPOINT_ZOMBIE(i);
+ }
+ break;
+ }
+
+ if (old_chain != NULL)
+ do_cleanups (old_chain);
+ }
+
+ /* Resume initial thread. */
+ if (!linuxthreads_pending_status (rpid))
+ child_ops.to_resume (rpid, step, signo);
+ }
+}
+
+/* Wait for any threads to stop. We may have to convert PID from a thread id
+ to a LWP id, and vice versa on the way out. */
+
+static int
+linuxthreads_wait (pid, ourstatus)
+ int pid;
+ struct target_waitstatus *ourstatus;
+{
+ int status;
+ int rpid;
+ int i;
+ int last;
+ int *wstatus;
+
+ if (linuxthreads_max && !linuxthreads_breakpoints_inserted)
+ wstatus = alloca (LINUXTHREAD_NSIG * sizeof (int));
+
+ /* See if the inferior has chosen values for its signals yet. By
+ checking for them here, we can be sure we've updated GDB's signal
+ handling table before the inferior ever gets one of them. (Well,
+ before we notice, anyway.) */
+ check_all_signal_numbers ();
+
+ for (;;)
+ {
+ if (!linuxthreads_max)
+ rpid = 0;
+ else if (!linuxthreads_breakpoints_inserted)
+ {
+ if (linuxthreads_inferior_pid)
+ pid = linuxthreads_inferior_pid;
+ else if (pid < 0)
+ pid = inferior_pid;
+ last = rpid = 0;
+ }
+ else if (pid < 0 && linuxthreads_wait_last >= 0)
+ {
+ status = linuxthreads_wait_status[linuxthreads_wait_last];
+ rpid = linuxthreads_wait_pid[linuxthreads_wait_last--];
+ }
+ else if (pid > 0 && linuxthreads_pending_status (pid))
+ {
+ for (i = linuxthreads_wait_last; i >= 0; i--)
+ if (linuxthreads_wait_pid[i] == pid)
+ break;
+ if (i < 0)
+ rpid = 0;
+ else
+ {
+ status = linuxthreads_wait_status[i];
+ rpid = pid;
+ if (i < linuxthreads_wait_last)
+ {
+ linuxthreads_wait_status[i] =
+ linuxthreads_wait_status[linuxthreads_wait_last];
+ linuxthreads_wait_pid[i] =
+ linuxthreads_wait_pid[linuxthreads_wait_last];
+ }
+ linuxthreads_wait_last--;
+ }
+ }
+ else
+ rpid = 0;
+
+ if (rpid == 0)
+ {
+ int save_errno;
+ sigset_t omask;
+
+ set_sigint_trap(); /* Causes SIGINT to be passed on to the
+ attached process. */
+ set_sigio_trap ();
+
+ sigprocmask(SIG_BLOCK, &linuxthreads_wait_mask, &omask);
+ for (;;)
+ {
+ rpid = waitpid (pid, &status, __WCLONE | WNOHANG);
+ if (rpid > 0)
+ break;
+ if (rpid == 0)
+ save_errno = 0;
+ else if (errno != EINTR)
+ save_errno = errno;
+ else
+ continue;
+
+ rpid = waitpid (pid, &status, WNOHANG);
+ if (rpid > 0)
+ break;
+ if (rpid < 0)
+ if (errno == EINTR)
+ continue;
+ else if (save_errno != 0)
+ break;
+
+ sigsuspend(&omask);
+ }
+ sigprocmask(SIG_SETMASK, &omask, NULL);
+
+ save_errno = errno;
+ clear_sigio_trap ();
+
+ clear_sigint_trap();
+
+ if (rpid == -1)
+ {
+ if (WIFEXITED(linuxthreads_exit_status))
+ {
+ store_waitstatus (ourstatus, linuxthreads_exit_status);
+ return inferior_pid;
+ }
+ else
+ {
+ fprintf_unfiltered
+ (gdb_stderr, "Child process unexpectedly missing: %s.\n",
+ safe_strerror (save_errno));
+ /* Claim it exited with unknown signal. */
+ ourstatus->kind = TARGET_WAITKIND_SIGNALLED;
+ ourstatus->value.sig = TARGET_SIGNAL_UNKNOWN;
+ return -1;
+ }
+ }
+
+ /* Signals arrive in any order. So get all signals until SIGTRAP
+ and resend previous ones to be held after. */
+ if (linuxthreads_max
+ && !linuxthreads_breakpoints_inserted
+ && WIFSTOPPED(status))
+ if (WSTOPSIG(status) == SIGTRAP)
+ {
+ while (--last >= 0)
+ kill (rpid, WSTOPSIG(wstatus[last]));
+
+ /* insert negative zombie breakpoint */
+ for (i = 0; i <= linuxthreads_breakpoint_last; i++)
+ if (linuxthreads_breakpoint_zombie[i].pid == rpid)
+ break;
+ if (i > linuxthreads_breakpoint_last)
+ {
+ linuxthreads_breakpoint_zombie[i].pid = rpid;
+ linuxthreads_breakpoint_last++;
+ }
+ linuxthreads_breakpoint_zombie[i].pc = read_pc_pid (rpid);
+ linuxthreads_breakpoint_zombie[i].step = 1;
+ }
+ else
+ {
+ if (WSTOPSIG(status) != SIGSTOP)
+ {
+ for (i = 0; i < last; i++)
+ if (wstatus[i] == status)
+ break;
+ if (i >= last)
+ wstatus[last++] = status;
+ }
+ child_resume (rpid, 1, TARGET_SIGNAL_0);
+ continue;
+ }
+ if (linuxthreads_inferior_pid)
+ linuxthreads_inferior_pid = 0;
+ }
+
+ if (linuxthreads_max && !stop_soon_quietly)
+ {
+ if (linuxthreads_max
+ && WIFSTOPPED(status)
+ && WSTOPSIG(status) == SIGSTOP)
+ {
+ /* Skip SIGSTOP signals. */
+ if (!linuxthreads_pending_status (rpid))
+ if (linuxthreads_step_pid == rpid)
+ child_resume (rpid, 1, linuxthreads_step_signo);
+ else
+ child_resume (rpid, 0, TARGET_SIGNAL_0);
+ continue;
+ }
+
+ /* Do no report exit status of cloned threads. */
+ if (WIFEXITED(status))
+ {
+ if (rpid == linuxthreads_initial_pid)
+ linuxthreads_exit_status = status;
+
+ /* Remove any zombie breakpoint. */
+ for (i = 0; i <= linuxthreads_breakpoint_last; i++)
+ if (linuxthreads_breakpoint_zombie[i].pid == rpid)
+ {
+ REMOVE_BREAKPOINT_ZOMBIE(i);
+ break;
+ }
+ if (pid > 0)
+ pid = -1;
+ continue;
+ }
+
+ /* Deal with zombie breakpoint */
+ for (i = 0; i <= linuxthreads_breakpoint_last; i++)
+ if (linuxthreads_breakpoint_zombie[i].pid == rpid)
+ break;
+
+ if (i <= linuxthreads_breakpoint_last)
+ {
+ /* There is a potential zombie breakpoint */
+ if (WIFEXITED(status)
+ || linuxthreads_breakpoint_zombie[i].pc != read_pc_pid (rpid))
+ {
+ /* The current pc is out of zombie breakpoint. */
+ REMOVE_BREAKPOINT_ZOMBIE(i);
+ }
+ else if (!linuxthreads_breakpoint_zombie[i].step
+ && WIFSTOPPED(status) && WSTOPSIG(status) == SIGTRAP)
+ {
+ /* This is a real one ==> decrement PC and restart. */
+ write_pc_pid (linuxthreads_breakpoint_zombie[i].pc
+ - DECR_PC_AFTER_BREAK, rpid);
+ if (linuxthreads_step_pid == rpid)
+ child_resume (rpid, 1, linuxthreads_step_signo);
+ else
+ child_resume (rpid, 0, TARGET_SIGNAL_0);
+ continue;
+ }
+ }
+
+ /* Walk through linuxthreads array in order to stop them */
+ if (linuxthreads_breakpoints_inserted)
+ update_stop_threads (rpid);
+
+ }
+ else if (rpid != inferior_pid)
+ continue;
+
+ store_waitstatus (ourstatus, status);
+
+ if (linuxthreads_attach_pending && !stop_soon_quietly)
+ {
+ int on = 1;
+ target_write_memory (linuxthreads_debug, (char *)&on, sizeof (on));
+ update_stop_threads (rpid);
+ linuxthreads_attach_pending = 0;
+ }
+
+ if (linuxthreads_breakpoints_inserted
+ && WIFSTOPPED(status)
+ && WSTOPSIG(status) == SIGTRAP)
+ linuxthreads_breakpoint_pid = rpid;
+ else if (linuxthreads_breakpoint_pid)
+ linuxthreads_breakpoint_pid = 0;
+
+ return rpid;
+ }
+}
+
+/* Fork an inferior process, and start debugging it with ptrace. */
+
+static void
+linuxthreads_create_inferior (exec_file, allargs, env)
+ char *exec_file;
+ char *allargs;
+ char **env;
+{
+ if (!exec_file && !exec_bfd)
+ {
+ error ("No executable file specified.\n\
+Use the \"file\" or \"exec-file\" command.");
+ return;
+ }
+
+ push_target (&linuxthreads_ops);
+ linuxthreads_breakpoints_inserted = 1;
+ linuxthreads_breakpoint_last = -1;
+ linuxthreads_wait_last = -1;
+ linuxthreads_exit_status = __W_STOPCODE(0);
+
+ if (linuxthreads_max)
+ linuxthreads_attach_pending = 1;
+
+ child_ops.to_create_inferior (exec_file, allargs, env);
+}
+
+/* Clean up after the inferior dies. */
+
+static void
+linuxthreads_mourn_inferior ()
+{
+ if (linuxthreads_max)
+ {
+ int off = 0;
+ target_write_memory (linuxthreads_debug, (char *)&off, sizeof (off));
+
+ linuxthreads_inferior_pid = 0;
+ linuxthreads_breakpoint_pid = 0;
+ linuxthreads_step_pid = 0;
+ linuxthreads_step_signo = TARGET_SIGNAL_0;
+ linuxthreads_manager_pid = 0;
+ linuxthreads_initial_pid = 0;
+ linuxthreads_attach_pending = 0;
+ init_thread_list(); /* Destroy thread info */
+ }
+
+ child_ops.to_mourn_inferior ();
+
+ unpush_target (&linuxthreads_ops);
+}
+
+/* Kill the inferior process */
+
+static void
+linuxthreads_kill ()
+{
+ int rpid;
+ int status;
+
+ if (inferior_pid == 0)
+ return;
+
+ if (linuxthreads_max && linuxthreads_manager_pid != 0)
+ {
+ /* Remove all threads status. */
+ inferior_pid = linuxthreads_manager_pid;
+ iterate_active_threads (kill_thread, 1);
+ }
+
+ kill_thread (inferior_pid);
+
+#if 0
+ /* doing_quit_force solves a real problem, but I think a properly
+ placed call to catch_errors would do the trick much more cleanly. */
+ if (doing_quit_force >= 0)
+ {
+ if (linuxthreads_max && linuxthreads_manager_pid != 0)
+ {
+ /* Wait for thread to complete */
+ while ((rpid = waitpid (-1, &status, __WCLONE)) > 0)
+ if (!WIFEXITED(status))
+ kill_thread (rpid);
+
+ while ((rpid = waitpid (-1, &status, 0)) > 0)
+ if (!WIFEXITED(status))
+ kill_thread (rpid);
+ }
+ else
+ while ((rpid = waitpid (inferior_pid, &status, 0)) > 0)
+ if (!WIFEXITED(status))
+ ptrace (PT_KILL, inferior_pid, (PTRACE_ARG3_TYPE) 0, 0);
+ }
+#endif
+
+ /* Wait for all threads. */
+ do
+ rpid = waitpid (-1, &status, __WCLONE | WNOHANG);
+ while (rpid > 0 || errno == EINTR);
+
+ do
+ rpid = waitpid (-1, &status, WNOHANG);
+ while (rpid > 0 || errno == EINTR);
+
+ linuxthreads_mourn_inferior ();
+}
+
+/* Insert a breakpoint */
+
+static int
+linuxthreads_insert_breakpoint (addr, contents_cache)
+ CORE_ADDR addr;
+ char *contents_cache;
+{
+ if (linuxthreads_max && linuxthreads_manager_pid != 0)
+ {
+ linuxthreads_breakpoint_addr = addr;
+ iterate_active_threads (insert_breakpoint, 1);
+ insert_breakpoint (linuxthreads_manager_pid);
+ }
+
+ return child_ops.to_insert_breakpoint (addr, contents_cache);
+}
+
+/* Remove a breakpoint */
+
+static int
+linuxthreads_remove_breakpoint (addr, contents_cache)
+ CORE_ADDR addr;
+ char *contents_cache;
+{
+ if (linuxthreads_max && linuxthreads_manager_pid != 0)
+ {
+ linuxthreads_breakpoint_addr = addr;
+ iterate_active_threads (remove_breakpoint, 1);
+ remove_breakpoint (linuxthreads_manager_pid);
+ }
+
+ return child_ops.to_remove_breakpoint (addr, contents_cache);
+}
+
+/* Mark our target-struct as eligible for stray "run" and "attach" commands. */
+
+static int
+linuxthreads_can_run ()
+{
+ return child_suppress_run;
+}
+
+static void
+init_linuxthreads_ops ()
+{
+ linuxthreads_ops.to_shortname = "linuxthreads";
+ linuxthreads_ops.to_longname = "LINUX threads and pthread.";
+ linuxthreads_ops.to_doc = "LINUX threads and pthread support.";
+ linuxthreads_ops.to_attach = linuxthreads_attach;
+ linuxthreads_ops.to_detach = linuxthreads_detach;
+ linuxthreads_ops.to_resume = linuxthreads_resume;
+ linuxthreads_ops.to_wait = linuxthreads_wait;
+ linuxthreads_ops.to_kill = linuxthreads_kill;
+ linuxthreads_ops.to_can_run = linuxthreads_can_run;
+ linuxthreads_ops.to_stratum = thread_stratum;
+ linuxthreads_ops.to_insert_breakpoint = linuxthreads_insert_breakpoint;
+ linuxthreads_ops.to_remove_breakpoint = linuxthreads_remove_breakpoint;
+ linuxthreads_ops.to_create_inferior = linuxthreads_create_inferior;
+ linuxthreads_ops.to_mourn_inferior = linuxthreads_mourn_inferior;
+ linuxthreads_ops.to_thread_alive = linuxthreads_thread_alive;
+ linuxthreads_ops.to_magic = OPS_MAGIC;
+}
+
+void
+_initialize_linuxthreads ()
+{
+ struct sigaction sact;
+
+ init_linuxthreads_ops ();
+ add_target (&linuxthreads_ops);
+ child_suppress_run = 1;
+
+ /* Attach SIGCHLD handler */
+ sact.sa_handler = sigchld_handler;
+ sigemptyset (&sact.sa_mask);
+ sact.sa_flags = 0;
+ sigaction (SIGCHLD, &sact, NULL);
+
+ /* initialize SIGCHLD mask */
+ sigemptyset (&linuxthreads_wait_mask);
+ sigaddset (&linuxthreads_wait_mask, SIGCHLD);
+}