gdb/gdbserver/linux-low.c

gdb/gdbserver/linux-low.c - gdb

Source code

/* Low level interface to ptrace, for the remote server for GDB.

   Copyright (C) 1995-2015 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 "server.h"

#include "linux-low.h"

#include "nat/linux-osdata.h"

#include "agent.h"



#include "nat/linux-nat.h"

#include "nat/linux-waitpid.h"

#include "gdb_wait.h"

#include <sys/ptrace.h>

#include "nat/linux-ptrace.h"

#include "nat/linux-procfs.h"

#include <signal.h>

#include <sys/ioctl.h>

#include <fcntl.h>

#include <unistd.h>

#include <sys/syscall.h>

#include <sched.h>

#include <ctype.h>

#include <pwd.h>

#include <sys/types.h>

#include <dirent.h>

#include <sys/stat.h>

#include <sys/vfs.h>

#include <sys/uio.h>

#include "filestuff.h"

#include "tracepoint.h"

#include "hostio.h"

#ifndef ELFMAG0

/* Don't include <linux/elf.h> here.  If it got included by gdb_proc_service.h

   then ELFMAG0 will have been defined.  If it didn't get included by

   gdb_proc_service.h then including it will likely introduce a duplicate

   definition of elf_fpregset_t.  */

#include <elf.h>

#endif



#ifndef SPUFS_MAGIC

‌#define SPUFS_MAGIC 0x23c9b64e

#endif



#ifdef HAVE_PERSONALITY

# include <sys/personality.h>

# if !HAVE_DECL_ADDR_NO_RANDOMIZE

‌#  define ADDR_NO_RANDOMIZE 0x0040000

# endif

#endif



#ifndef O_LARGEFILE

‌#define O_LARGEFILE 0

#endif



#ifndef W_STOPCODE

‌#define W_STOPCODE(sig) ((sig) << 8 | 0x7f)

#endif



/* This is the kernel's hard limit.  Not to be confused with

   SIGRTMIN.  */

#ifndef __SIGRTMIN

‌#define __SIGRTMIN 32

#endif



/* Some targets did not define these ptrace constants from the start,

   so gdbserver defines them locally here.  In the future, these may

   be removed after they are added to asm/ptrace.h.  */

#if !(defined(PT_TEXT_ADDR) \

      || defined(PT_DATA_ADDR) \

      || defined(PT_TEXT_END_ADDR))

#if defined(__mcoldfire__)

/* These are still undefined in 3.10 kernels.  */

‌#define PT_TEXT_ADDR 49*4

‌#define PT_DATA_ADDR 50*4

‌#define PT_TEXT_END_ADDR  51*4

/* BFIN already defines these since at least 2.6.32 kernels.  */

#elif defined(BFIN)

‌#define PT_TEXT_ADDR 220

‌#define PT_TEXT_END_ADDR 224

‌#define PT_DATA_ADDR 228

/* These are still undefined in 3.10 kernels.  */

#elif defined(__TMS320C6X__)

‌#define PT_TEXT_ADDR     (0x10000*4)

‌#define PT_DATA_ADDR     (0x10004*4)

‌#define PT_TEXT_END_ADDR (0x10008*4)

#endif

#endif



#ifdef HAVE_LINUX_BTRACE

# include "nat/linux-btrace.h"

#endif



#ifndef HAVE_ELF32_AUXV_T

/* Copied from glibc's elf.h.  */

typedef struct

{

  uint32_t a_type;                /* Entry type */

  union

    {

      uint32_t a_val;                /* Integer value */

      /* We use to have pointer elements added here.  We cannot do that,

         though, since it does not work when using 32-bit definitions

         on 64-bit platforms and vice versa.  */

    } a_un;

‌} Elf32_auxv_t;

#endif



#ifndef HAVE_ELF64_AUXV_T

/* Copied from glibc's elf.h.  */

typedef struct

{

  uint64_t a_type;                /* Entry type */

  union

    {

      uint64_t a_val;                /* Integer value */

      /* We use to have pointer elements added here.  We cannot do that,

         though, since it does not work when using 32-bit definitions

         on 64-bit platforms and vice versa.  */

    } a_un;

‌} Elf64_auxv_t;

#endif



/* A list of all unknown processes which receive stop signals.  Some

   other process will presumably claim each of these as forked

   children momentarily.  */



‌struct simple_pid_list

{

  /* The process ID.  */

  int pid;



  /* The status as reported by waitpid.  */

  int status;



  /* Next in chain.  */

  struct simple_pid_list *next;

};

‌struct simple_pid_list *stopped_pids;



/* Trivial list manipulation functions to keep track of a list of new

   stopped processes.  */



static void

‌add_to_pid_list (struct simple_pid_list **listp, int pid, int status)

{

  struct simple_pid_list *new_pid = xmalloc (sizeof (struct simple_pid_list));



  new_pid->pid = pid;

  new_pid->status = status;

  new_pid->next = *listp;

  *listp = new_pid;

}



static int

‌pull_pid_from_list (struct simple_pid_list **listp, int pid, int *statusp)

{

  struct simple_pid_list **p;



  for (p = listp; *p != NULL; p = &(*p)->next)

    if ((*p)->pid == pid)

      {

        struct simple_pid_list *next = (*p)->next;



        *statusp = (*p)->status;

        xfree (*p);

        *p = next;

        return 1;

      }

  return 0;

}



‌enum stopping_threads_kind

  {

    /* Not stopping threads presently.  */

    NOT_STOPPING_THREADS,



    /* Stopping threads.  */

    STOPPING_THREADS,



    /* Stopping and suspending threads.  */

    STOPPING_AND_SUSPENDING_THREADS

  };



/* This is set while stop_all_lwps is in effect.  */

‌enum stopping_threads_kind stopping_threads = NOT_STOPPING_THREADS;



/* FIXME make into a target method?  */

‌int using_threads = 1;



/* True if we're presently stabilizing threads (moving them out of

   jump pads).  */

‌static int stabilizing_threads;



static void linux_resume_one_lwp (struct lwp_info *lwp,

                                  int step, int signal, siginfo_t *info);

static void linux_resume (struct thread_resume *resume_info, size_t n);

static void stop_all_lwps (int suspend, struct lwp_info *except);

static void unstop_all_lwps (int unsuspend, struct lwp_info *except);

static int linux_wait_for_event_filtered (ptid_t wait_ptid, ptid_t filter_ptid,

                                          int *wstat, int options);

static int linux_wait_for_event (ptid_t ptid, int *wstat, int options);

static struct lwp_info *add_lwp (ptid_t ptid);

static int linux_stopped_by_watchpoint (void);

static void mark_lwp_dead (struct lwp_info *lwp, int wstat);

static void proceed_all_lwps (void);

static int finish_step_over (struct lwp_info *lwp);

static int kill_lwp (unsigned long lwpid, int signo);



/* When the event-loop is doing a step-over, this points at the thread

   being stepped.  */

‌ptid_t step_over_bkpt;



/* True if the low target can hardware single-step.  Such targets

   don't need a BREAKPOINT_REINSERT_ADDR callback.  */



static int

‌can_hardware_single_step (void)

{

  return (the_low_target.breakpoint_reinsert_addr == NULL);

}



/* True if the low target supports memory breakpoints.  If so, we'll

   have a GET_PC implementation.  */



static int

‌supports_breakpoints (void)

{

  return (the_low_target.get_pc != NULL);

}



/* Returns true if this target can support fast tracepoints.  This

   does not mean that the in-process agent has been loaded in the

   inferior.  */



static int

‌supports_fast_tracepoints (void)

{

  return the_low_target.install_fast_tracepoint_jump_pad != NULL;

}



/* True if LWP is stopped in its stepping range.  */



static int

‌lwp_in_step_range (struct lwp_info *lwp)

{

  CORE_ADDR pc = lwp->stop_pc;



  return (pc >= lwp->step_range_start && pc < lwp->step_range_end);

}



‌struct pending_signals

{

  int signal;

  siginfo_t info;

  struct pending_signals *prev;

};



/* The read/write ends of the pipe registered as waitable file in the

   event loop.  */

‌static int linux_event_pipe[2] = { -1, -1 };



/* True if we're currently in async mode.  */

‌#define target_is_async_p() (linux_event_pipe[0] != -1)



static void send_sigstop (struct lwp_info *lwp);

static void wait_for_sigstop (void);



/* Return non-zero if HEADER is a 64-bit ELF file.  */



static int

‌elf_64_header_p (const Elf64_Ehdr *header, unsigned int *machine)

{

  if (header->e_ident[EI_MAG0] == ELFMAG0

      && header->e_ident[EI_MAG1] == ELFMAG1

      && header->e_ident[EI_MAG2] == ELFMAG2

      && header->e_ident[EI_MAG3] == ELFMAG3)

    {

      *machine = header->e_machine;

      return header->e_ident[EI_CLASS] == ELFCLASS64;



    }

  *machine = EM_NONE;

  return -1;

}



/* Return non-zero if FILE is a 64-bit ELF file,

   zero if the file is not a 64-bit ELF file,

   and -1 if the file is not accessible or doesn't exist.  */



static int

‌elf_64_file_p (const char *file, unsigned int *machine)

{

  Elf64_Ehdr header;

  int fd;



  fd = open (file, O_RDONLY);

  if (fd < 0)

    return -1;



  if (read (fd, &header, sizeof (header)) != sizeof (header))

    {

      close (fd);

      return 0;

    }

  close (fd);



  return elf_64_header_p (&header, machine);

}



/* Accepts an integer PID; Returns true if the executable PID is

   running is a 64-bit ELF file..  */



int

‌linux_pid_exe_is_elf_64_file (int pid, unsigned int *machine)

{

  char file[PATH_MAX];



  sprintf (file, "/proc/%d/exe", pid);

  return elf_64_file_p (file, machine);

}



static void

‌delete_lwp (struct lwp_info *lwp)

{

  struct thread_info *thr = get_lwp_thread (lwp);



  if (debug_threads)

    debug_printf ("deleting %ld\n", lwpid_of (thr));



  remove_thread (thr);

  free (lwp->arch_private);

  free (lwp);

}



/* Add a process to the common process list, and set its private

   data.  */



static struct process_info *

‌linux_add_process (int pid, int attached)

{

  struct process_info *proc;



  proc = add_process (pid, attached);

  proc->private = xcalloc (1, sizeof (*proc->private));



  /* Set the arch when the first LWP stops.  */

  proc->private->new_inferior = 1;



  if (the_low_target.new_process != NULL)

    proc->private->arch_private = the_low_target.new_process ();



  return proc;

}



static CORE_ADDR get_pc (struct lwp_info *lwp);



/* Handle a GNU/Linux extended wait response.  If we see a clone

   event, we need to add the new LWP to our list (and not report the

   trap to higher layers).  */



static void

‌handle_extended_wait (struct lwp_info *event_child, int wstat)

{

  int event = linux_ptrace_get_extended_event (wstat);

  struct thread_info *event_thr = get_lwp_thread (event_child);

  struct lwp_info *new_lwp;



  if (event == PTRACE_EVENT_CLONE)

    {

      ptid_t ptid;

      unsigned long new_pid;

      int ret, status;



      ptrace (PTRACE_GETEVENTMSG, lwpid_of (event_thr), (PTRACE_TYPE_ARG3) 0,

              &new_pid);



      /* If we haven't already seen the new PID stop, wait for it now.  */

      if (!pull_pid_from_list (&stopped_pids, new_pid, &status))

        {

          /* The new child has a pending SIGSTOP.  We can't affect it until it

             hits the SIGSTOP, but we're already attached.  */



          ret = my_waitpid (new_pid, &status, __WALL);



          if (ret == -1)

            perror_with_name ("waiting for new child");

          else if (ret != new_pid)

            warning ("wait returned unexpected PID %d", ret);

          else if (!WIFSTOPPED (status))

            warning ("wait returned unexpected status 0x%x", status);

        }



      if (debug_threads)

        debug_printf ("HEW: Got clone event "

                      "from LWP %ld, new child is LWP %ld\n",

                      lwpid_of (event_thr), new_pid);



      ptid = ptid_build (pid_of (event_thr), new_pid, 0);

      new_lwp = add_lwp (ptid);



      /* Either we're going to immediately resume the new thread

         or leave it stopped.  linux_resume_one_lwp is a nop if it

         thinks the thread is currently running, so set this first

         before calling linux_resume_one_lwp.  */

      new_lwp->stopped = 1;



     /* If we're suspending all threads, leave this one suspended

        too.  */

      if (stopping_threads == STOPPING_AND_SUSPENDING_THREADS)

        new_lwp->suspended = 1;



      /* Normally we will get the pending SIGSTOP.  But in some cases

         we might get another signal delivered to the group first.

         If we do get another signal, be sure not to lose it.  */

      if (WSTOPSIG (status) == SIGSTOP)

        {

          if (stopping_threads == NOT_STOPPING_THREADS)

            linux_resume_one_lwp (new_lwp, 0, 0, NULL);

        }

      else

        {

          new_lwp->stop_expected = 1;



          if (stopping_threads != NOT_STOPPING_THREADS)

            {

              new_lwp->status_pending_p = 1;

              new_lwp->status_pending = status;

            }

          else

            /* Pass the signal on.  This is what GDB does - except

               shouldn't we really report it instead?  */

            linux_resume_one_lwp (new_lwp, 0, WSTOPSIG (status), NULL);

        }



      /* Always resume the current thread.  If we are stopping

         threads, it will have a pending SIGSTOP; we may as well

         collect it now.  */

      linux_resume_one_lwp (event_child, event_child->stepping, 0, NULL);

    }

}



/* Return the PC as read from the regcache of LWP, without any

   adjustment.  */



static CORE_ADDR

‌get_pc (struct lwp_info *lwp)

{

  struct thread_info *saved_thread;

  struct regcache *regcache;

  CORE_ADDR pc;



  if (the_low_target.get_pc == NULL)

    return 0;



  saved_thread = current_thread;

  current_thread = get_lwp_thread (lwp);



  regcache = get_thread_regcache (current_thread, 1);

  pc = (*the_low_target.get_pc) (regcache);



  if (debug_threads)

    debug_printf ("pc is 0x%lx\n", (long) pc);



  current_thread = saved_thread;

  return pc;

}



/* This function should only be called if LWP got a SIGTRAP.

   The SIGTRAP could mean several things.



   On i386, where decr_pc_after_break is non-zero:



   If we were single-stepping this process using PTRACE_SINGLESTEP, we

   will get only the one SIGTRAP.  The value of $eip will be the next

   instruction.  If the instruction we stepped over was a breakpoint,

   we need to decrement the PC.



   If we continue the process using PTRACE_CONT, we will get a

   SIGTRAP when we hit a breakpoint.  The value of $eip will be

   the instruction after the breakpoint (i.e. needs to be

   decremented).  If we report the SIGTRAP to GDB, we must also

   report the undecremented PC.  If the breakpoint is removed, we

   must resume at the decremented PC.



   On a non-decr_pc_after_break machine with hardware or kernel

   single-step:



   If we either single-step a breakpoint instruction, or continue and

   hit a breakpoint instruction, our PC will point at the breakpoint

   instruction.  */



static int

‌check_stopped_by_breakpoint (struct lwp_info *lwp)

{

  CORE_ADDR pc;

  CORE_ADDR sw_breakpoint_pc;

  struct thread_info *saved_thread;



  if (the_low_target.get_pc == NULL)

    return 0;



  pc = get_pc (lwp);

  sw_breakpoint_pc = pc - the_low_target.decr_pc_after_break;



  /* breakpoint_at reads from the current thread.  */

  saved_thread = current_thread;

  current_thread = get_lwp_thread (lwp);



  /* We may have just stepped a breakpoint instruction.  E.g., in

     non-stop mode, GDB first tells the thread A to step a range, and

     then the user inserts a breakpoint inside the range.  In that

     case, we need to report the breakpoint PC.  But, when we're

     trying to step past one of our own breakpoints, that happens to

     have been placed on top of a permanent breakpoint instruction, we

     shouldn't adjust the PC, otherwise the program would keep

     trapping the permanent breakpoint forever.  */

  if ((!lwp->stepping

       || (!ptid_equal (ptid_of (current_thread), step_over_bkpt)

           && lwp->stop_pc == sw_breakpoint_pc))

      && (*the_low_target.breakpoint_at) (sw_breakpoint_pc))

    {

      if (debug_threads)

        {

          struct thread_info *thr = get_lwp_thread (lwp);



          debug_printf ("CSBB: %s stopped by software breakpoint\n",

                        target_pid_to_str (ptid_of (thr)));

        }



      /* Back up the PC if necessary.  */

      if (pc != sw_breakpoint_pc)

        {

          struct regcache *regcache

            = get_thread_regcache (current_thread, 1);

          (*the_low_target.set_pc) (regcache, sw_breakpoint_pc);

        }



      lwp->stop_pc = sw_breakpoint_pc;

      lwp->stop_reason = LWP_STOPPED_BY_SW_BREAKPOINT;

      current_thread = saved_thread;

      return 1;

    }



  if (hardware_breakpoint_inserted_here (pc))

    {

      if (debug_threads)

        {

          struct thread_info *thr = get_lwp_thread (lwp);



          debug_printf ("CSBB: %s stopped by hardware breakpoint\n",

                        target_pid_to_str (ptid_of (thr)));

        }



      lwp->stop_pc = pc;

      lwp->stop_reason = LWP_STOPPED_BY_HW_BREAKPOINT;

      current_thread = saved_thread;

      return 1;

    }



  current_thread = saved_thread;

  return 0;

}



static struct lwp_info *

‌add_lwp (ptid_t ptid)

{

  struct lwp_info *lwp;



  lwp = (struct lwp_info *) xmalloc (sizeof (*lwp));

  memset (lwp, 0, sizeof (*lwp));



  if (the_low_target.new_thread != NULL)

    lwp->arch_private = the_low_target.new_thread ();



  lwp->thread = add_thread (ptid, lwp);



  return lwp;

}



/* Start an inferior process and returns its pid.

   ALLARGS is a vector of program-name and args. */



static int

‌linux_create_inferior (char *program, char **allargs)

{

#ifdef HAVE_PERSONALITY

  int personality_orig = 0, personality_set = 0;

#endif

  struct lwp_info *new_lwp;

  int pid;

  ptid_t ptid;



#ifdef HAVE_PERSONALITY

  if (disable_randomization)

    {

      errno = 0;

      personality_orig = personality (0xffffffff);

      if (errno == 0 && !(personality_orig & ADDR_NO_RANDOMIZE))

        {

          personality_set = 1;

          personality (personality_orig | ADDR_NO_RANDOMIZE);

        }

      if (errno != 0 || (personality_set

                         && !(personality (0xffffffff) & ADDR_NO_RANDOMIZE)))

        warning ("Error disabling address space randomization: %s",

                 strerror (errno));

    }

#endif



#if defined(__UCLIBC__) && defined(HAS_NOMMU)

  pid = vfork ();

#else

  pid = fork ();

#endif

  if (pid < 0)

    perror_with_name ("fork");



  if (pid == 0)

    {

      close_most_fds ();

      ptrace (PTRACE_TRACEME, 0, (PTRACE_TYPE_ARG3) 0, (PTRACE_TYPE_ARG4) 0);



#ifndef __ANDROID__ /* Bionic doesn't use SIGRTMIN the way glibc does.  */

      signal (__SIGRTMIN + 1, SIG_DFL);

#endif



      setpgid (0, 0);



      /* If gdbserver is connected to gdb via stdio, redirect the inferior's

         stdout to stderr so that inferior i/o doesn't corrupt the connection.

         Also, redirect stdin to /dev/null.  */

      if (remote_connection_is_stdio ())

        {

          close (0);

          open ("/dev/null", O_RDONLY);

          dup2 (2, 1);

          if (write (2, "stdin/stdout redirected\n",

                     sizeof ("stdin/stdout redirected\n") - 1) < 0)

            {

              /* Errors ignored.  */;

            }

        }



      execv (program, allargs);

      if (errno == ENOENT)

        execvp (program, allargs);



      fprintf (stderr, "Cannot exec %s: %s.\n", program,

               strerror (errno));

      fflush (stderr);

      _exit (0177);

    }



#ifdef HAVE_PERSONALITY

  if (personality_set)

    {

      errno = 0;

      personality (personality_orig);

      if (errno != 0)

        warning ("Error restoring address space randomization: %s",

                 strerror (errno));

    }

#endif



  linux_add_process (pid, 0);



  ptid = ptid_build (pid, pid, 0);

  new_lwp = add_lwp (ptid);

  new_lwp->must_set_ptrace_flags = 1;



  return pid;

}



/* Attach to an inferior process.  Returns 0 on success, ERRNO on

   error.  */



int

‌linux_attach_lwp (ptid_t ptid)

{

  struct lwp_info *new_lwp;

  int lwpid = ptid_get_lwp (ptid);



  if (ptrace (PTRACE_ATTACH, lwpid, (PTRACE_TYPE_ARG3) 0, (PTRACE_TYPE_ARG4) 0)

      != 0)

    return errno;



  new_lwp = add_lwp (ptid);



  /* We need to wait for SIGSTOP before being able to make the next

     ptrace call on this LWP.  */

  new_lwp->must_set_ptrace_flags = 1;



  if (linux_proc_pid_is_stopped (lwpid))

    {

      if (debug_threads)

        debug_printf ("Attached to a stopped process\n");



      /* The process is definitely stopped.  It is in a job control

         stop, unless the kernel predates the TASK_STOPPED /

         TASK_TRACED distinction, in which case it might be in a

         ptrace stop.  Make sure it is in a ptrace stop; from there we

         can kill it, signal it, et cetera.



         First make sure there is a pending SIGSTOP.  Since we are

         already attached, the process can not transition from stopped

         to running without a PTRACE_CONT; so we know this signal will

         go into the queue.  The SIGSTOP generated by PTRACE_ATTACH is

         probably already in the queue (unless this kernel is old

         enough to use TASK_STOPPED for ptrace stops); but since

         SIGSTOP is not an RT signal, it can only be queued once.  */

      kill_lwp (lwpid, SIGSTOP);



      /* Finally, resume the stopped process.  This will deliver the

         SIGSTOP (or a higher priority signal, just like normal

         PTRACE_ATTACH), which we'll catch later on.  */

      ptrace (PTRACE_CONT, lwpid, (PTRACE_TYPE_ARG3) 0, (PTRACE_TYPE_ARG4) 0);

    }



  /* The next time we wait for this LWP we'll see a SIGSTOP as PTRACE_ATTACH

     brings it to a halt.



     There are several cases to consider here:



     1) gdbserver has already attached to the process and is being notified

        of a new thread that is being created.

        In this case we should ignore that SIGSTOP and resume the

        process.  This is handled below by setting stop_expected = 1,

        and the fact that add_thread sets last_resume_kind ==

        resume_continue.



     2) This is the first thread (the process thread), and we're attaching

        to it via attach_inferior.

        In this case we want the process thread to stop.

        This is handled by having linux_attach set last_resume_kind ==

        resume_stop after we return.



        If the pid we are attaching to is also the tgid, we attach to and

        stop all the existing threads.  Otherwise, we attach to pid and

        ignore any other threads in the same group as this pid.



     3) GDB is connecting to gdbserver and is requesting an enumeration of all

        existing threads.

        In this case we want the thread to stop.

        FIXME: This case is currently not properly handled.

        We should wait for the SIGSTOP but don't.  Things work apparently

        because enough time passes between when we ptrace (ATTACH) and when

        gdb makes the next ptrace call on the thread.



     On the other hand, if we are currently trying to stop all threads, we

     should treat the new thread as if we had sent it a SIGSTOP.  This works

     because we are guaranteed that the add_lwp call above added us to the

     end of the list, and so the new thread has not yet reached

     wait_for_sigstop (but will).  */

  new_lwp->stop_expected = 1;



  return 0;

}



/* Callback for linux_proc_attach_tgid_threads.  Attach to PTID if not

   already attached.  Returns true if a new LWP is found, false

   otherwise.  */



static int

‌attach_proc_task_lwp_callback (ptid_t ptid)

{

  /* Is this a new thread?  */

  if (find_thread_ptid (ptid) == NULL)

    {

      int lwpid = ptid_get_lwp (ptid);

      int err;



      if (debug_threads)

        debug_printf ("Found new lwp %d\n", lwpid);



      err = linux_attach_lwp (ptid);



      /* Be quiet if we simply raced with the thread exiting.  EPERM

         is returned if the thread's task still exists, and is marked

         as exited or zombie, as well as other conditions, so in that

         case, confirm the status in /proc/PID/status.  */

      if (err == ESRCH

          || (err == EPERM && linux_proc_pid_is_gone (lwpid)))

        {

          if (debug_threads)

            {

              debug_printf ("Cannot attach to lwp %d: "

                            "thread is gone (%d: %s)\n",

                            lwpid, err, strerror (err));

            }

        }

      else if (err != 0)

        {

          warning (_("Cannot attach to lwp %d: %s"),

                   lwpid,

                   linux_ptrace_attach_fail_reason_string (ptid, err));

        }



      return 1;

    }

  return 0;

}



/* Attach to PID.  If PID is the tgid, attach to it and all

   of its threads.  */



static int

‌linux_attach (unsigned long pid)

{

  ptid_t ptid = ptid_build (pid, pid, 0);

  int err;



  /* Attach to PID.  We will check for other threads

     soon.  */

  err = linux_attach_lwp (ptid);

  if (err != 0)

    error ("Cannot attach to process %ld: %s",

           pid, linux_ptrace_attach_fail_reason_string (ptid, err));



  linux_add_process (pid, 1);



  if (!non_stop)

    {

      struct thread_info *thread;



     /* Don't ignore the initial SIGSTOP if we just attached to this

        process.  It will be collected by wait shortly.  */

      thread = find_thread_ptid (ptid_build (pid, pid, 0));

      thread->last_resume_kind = resume_stop;

    }



  /* We must attach to every LWP.  If /proc is mounted, use that to

     find them now.  On the one hand, the inferior may be using raw

     clone instead of using pthreads.  On the other hand, even if it

     is using pthreads, GDB may not be connected yet (thread_db needs

     to do symbol lookups, through qSymbol).  Also, thread_db walks

     structures in the inferior's address space to find the list of

     threads/LWPs, and those structures may well be corrupted.  Note

     that once thread_db is loaded, we'll still use it to list threads

     and associate pthread info with each LWP.  */

  linux_proc_attach_tgid_threads (pid, attach_proc_task_lwp_callback);

  return 0;

}



‌struct counter

{

  int pid;

  int count;

};



static int

‌second_thread_of_pid_p (struct inferior_list_entry *entry, void *args)

{

  struct counter *counter = args;



  if (ptid_get_pid (entry->id) == counter->pid)

    {

      if (++counter->count > 1)

        return 1;

    }



  return 0;

}



static int

‌last_thread_of_process_p (int pid)

{

  struct counter counter = { pid , 0 };



  return (find_inferior (&all_threads,

                         second_thread_of_pid_p, &counter) == NULL);

}



/* Kill LWP.  */



static void

‌linux_kill_one_lwp (struct lwp_info *lwp)

{

  struct thread_info *thr = get_lwp_thread (lwp);

  int pid = lwpid_of (thr);



  /* PTRACE_KILL is unreliable.  After stepping into a signal handler,

     there is no signal context, and ptrace(PTRACE_KILL) (or

     ptrace(PTRACE_CONT, SIGKILL), pretty much the same) acts like

     ptrace(CONT, pid, 0,0) and just resumes the tracee.  A better

     alternative is to kill with SIGKILL.  We only need one SIGKILL

     per process, not one for each thread.  But since we still support

     linuxthreads, and we also support debugging programs using raw

     clone without CLONE_THREAD, we send one for each thread.  For

     years, we used PTRACE_KILL only, so we're being a bit paranoid

     about some old kernels where PTRACE_KILL might work better

     (dubious if there are any such, but that's why it's paranoia), so

     we try SIGKILL first, PTRACE_KILL second, and so we're fine

     everywhere.  */



  errno = 0;

  kill_lwp (pid, SIGKILL);

  if (debug_threads)

    {

      int save_errno = errno;



      debug_printf ("LKL:  kill_lwp (SIGKILL) %s, 0, 0 (%s)\n",

                    target_pid_to_str (ptid_of (thr)),

                    save_errno ? strerror (save_errno) : "OK");

    }



  errno = 0;

  ptrace (PTRACE_KILL, pid, (PTRACE_TYPE_ARG3) 0, (PTRACE_TYPE_ARG4) 0);

  if (debug_threads)

    {

      int save_errno = errno;



      debug_printf ("LKL:  PTRACE_KILL %s, 0, 0 (%s)\n",

                    target_pid_to_str (ptid_of (thr)),

                    save_errno ? strerror (save_errno) : "OK");

    }

}



/* Kill LWP and wait for it to die.  */



static void

‌kill_wait_lwp (struct lwp_info *lwp)

{

  struct thread_info *thr = get_lwp_thread (lwp);

  int pid = ptid_get_pid (ptid_of (thr));

  int lwpid = ptid_get_lwp (ptid_of (thr));

  int wstat;

  int res;



  if (debug_threads)

    debug_printf ("kwl: killing lwp %d, for pid: %d\n", lwpid, pid);



  do

    {

      linux_kill_one_lwp (lwp);



      /* Make sure it died.  Notes:



         - The loop is most likely unnecessary.



         - We don't use linux_wait_for_event as that could delete lwps

           while we're iterating over them.  We're not interested in

           any pending status at this point, only in making sure all

           wait status on the kernel side are collected until the

           process is reaped.



         - We don't use __WALL here as the __WALL emulation relies on

           SIGCHLD, and killing a stopped process doesn't generate

           one, nor an exit status.

      */

      res = my_waitpid (lwpid, &wstat, 0);

      if (res == -1 && errno == ECHILD)

        res = my_waitpid (lwpid, &wstat, __WCLONE);

    } while (res > 0 && WIFSTOPPED (wstat));



  gdb_assert (res > 0);

}



/* Callback for `find_inferior'.  Kills an lwp of a given process,

   except the leader.  */



static int

‌kill_one_lwp_callback (struct inferior_list_entry *entry, void *args)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  int pid = * (int *) args;



  if (ptid_get_pid (entry->id) != pid)

    return 0;



  /* We avoid killing the first thread here, because of a Linux kernel (at

     least 2.6.0-test7 through 2.6.8-rc4) bug; if we kill the parent before

     the children get a chance to be reaped, it will remain a zombie

     forever.  */



  if (lwpid_of (thread) == pid)

    {

      if (debug_threads)

        debug_printf ("lkop: is last of process %s\n",

                      target_pid_to_str (entry->id));

      return 0;

    }



  kill_wait_lwp (lwp);

  return 0;

}



static int

‌linux_kill (int pid)

{

  struct process_info *process;

  struct lwp_info *lwp;



  process = find_process_pid (pid);

  if (process == NULL)

    return -1;



  /* If we're killing a running inferior, make sure it is stopped

     first, as PTRACE_KILL will not work otherwise.  */

  stop_all_lwps (0, NULL);



  find_inferior (&all_threads, kill_one_lwp_callback , &pid);



  /* See the comment in linux_kill_one_lwp.  We did not kill the first

     thread in the list, so do so now.  */

  lwp = find_lwp_pid (pid_to_ptid (pid));



  if (lwp == NULL)

    {

      if (debug_threads)

        debug_printf ("lk_1: cannot find lwp for pid: %d\n",

                      pid);

    }

  else

    kill_wait_lwp (lwp);



  the_target->mourn (process);



  /* Since we presently can only stop all lwps of all processes, we

     need to unstop lwps of other processes.  */

  unstop_all_lwps (0, NULL);

  return 0;

}



/* Get pending signal of THREAD, for detaching purposes.  This is the

   signal the thread last stopped for, which we need to deliver to the

   thread when detaching, otherwise, it'd be suppressed/lost.  */



static int

‌get_detach_signal (struct thread_info *thread)

{

  enum gdb_signal signo = GDB_SIGNAL_0;

  int status;

  struct lwp_info *lp = get_thread_lwp (thread);



  if (lp->status_pending_p)

    status = lp->status_pending;

  else

    {

      /* If the thread had been suspended by gdbserver, and it stopped

         cleanly, then it'll have stopped with SIGSTOP.  But we don't

         want to deliver that SIGSTOP.  */

      if (thread->last_status.kind != TARGET_WAITKIND_STOPPED

          || thread->last_status.value.sig == GDB_SIGNAL_0)

        return 0;



      /* Otherwise, we may need to deliver the signal we

         intercepted.  */

      status = lp->last_status;

    }



  if (!WIFSTOPPED (status))

    {

      if (debug_threads)

        debug_printf ("GPS: lwp %s hasn't stopped: no pending signal\n",

                      target_pid_to_str (ptid_of (thread)));

      return 0;

    }



  /* Extended wait statuses aren't real SIGTRAPs.  */

  if (WSTOPSIG (status) == SIGTRAP && linux_is_extended_waitstatus (status))

    {

      if (debug_threads)

        debug_printf ("GPS: lwp %s had stopped with extended "

                      "status: no pending signal\n",

                      target_pid_to_str (ptid_of (thread)));

      return 0;

    }



  signo = gdb_signal_from_host (WSTOPSIG (status));



  if (program_signals_p && !program_signals[signo])

    {

      if (debug_threads)

        debug_printf ("GPS: lwp %s had signal %s, but it is in nopass state\n",

                      target_pid_to_str (ptid_of (thread)),

                      gdb_signal_to_string (signo));

      return 0;

    }

  else if (!program_signals_p

           /* If we have no way to know which signals GDB does not

              want to have passed to the program, assume

              SIGTRAP/SIGINT, which is GDB's default.  */

           && (signo == GDB_SIGNAL_TRAP || signo == GDB_SIGNAL_INT))

    {

      if (debug_threads)

        debug_printf ("GPS: lwp %s had signal %s, "

                      "but we don't know if we should pass it. "

                      "Default to not.\n",

                      target_pid_to_str (ptid_of (thread)),

                      gdb_signal_to_string (signo));

      return 0;

    }

  else

    {

      if (debug_threads)

        debug_printf ("GPS: lwp %s has pending signal %s: delivering it.\n",

                      target_pid_to_str (ptid_of (thread)),

                      gdb_signal_to_string (signo));



      return WSTOPSIG (status);

    }

}



static int

‌linux_detach_one_lwp (struct inferior_list_entry *entry, void *args)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  int pid = * (int *) args;

  int sig;



  if (ptid_get_pid (entry->id) != pid)

    return 0;



  /* If there is a pending SIGSTOP, get rid of it.  */

  if (lwp->stop_expected)

    {

      if (debug_threads)

        debug_printf ("Sending SIGCONT to %s\n",

                      target_pid_to_str (ptid_of (thread)));



      kill_lwp (lwpid_of (thread), SIGCONT);

      lwp->stop_expected = 0;

    }



  /* Flush any pending changes to the process's registers.  */

  regcache_invalidate_thread (thread);



  /* Pass on any pending signal for this thread.  */

  sig = get_detach_signal (thread);



  /* Finally, let it resume.  */

  if (the_low_target.prepare_to_resume != NULL)

    the_low_target.prepare_to_resume (lwp);

  if (ptrace (PTRACE_DETACH, lwpid_of (thread), (PTRACE_TYPE_ARG3) 0,

              (PTRACE_TYPE_ARG4) (long) sig) < 0)

    error (_("Can't detach %s: %s"),

           target_pid_to_str (ptid_of (thread)),

           strerror (errno));



  delete_lwp (lwp);

  return 0;

}



static int

‌linux_detach (int pid)

{

  struct process_info *process;



  process = find_process_pid (pid);

  if (process == NULL)

    return -1;



  /* Stop all threads before detaching.  First, ptrace requires that

     the thread is stopped to sucessfully detach.  Second, thread_db

     may need to uninstall thread event breakpoints from memory, which

     only works with a stopped process anyway.  */

  stop_all_lwps (0, NULL);



#ifdef USE_THREAD_DB

  thread_db_detach (process);

#endif



  /* Stabilize threads (move out of jump pads).  */

  stabilize_threads ();



  find_inferior (&all_threads, linux_detach_one_lwp, &pid);



  the_target->mourn (process);



  /* Since we presently can only stop all lwps of all processes, we

     need to unstop lwps of other processes.  */

  unstop_all_lwps (0, NULL);

  return 0;

}



/* Remove all LWPs that belong to process PROC from the lwp list.  */



static int

‌delete_lwp_callback (struct inferior_list_entry *entry, void *proc)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  struct process_info *process = proc;



  if (pid_of (thread) == pid_of (process))

    delete_lwp (lwp);



  return 0;

}



static void

‌linux_mourn (struct process_info *process)

{

  struct process_info_private *priv;



#ifdef USE_THREAD_DB

  thread_db_mourn (process);

#endif



  find_inferior (&all_threads, delete_lwp_callback, process);



  /* Freeing all private data.  */

  priv = process->private;

  free (priv->arch_private);

  free (priv);

  process->private = NULL;



  remove_process (process);

}



static void

‌linux_join (int pid)

{

  int status, ret;



  do {

    ret = my_waitpid (pid, &status, 0);

    if (WIFEXITED (status) || WIFSIGNALED (status))

      break;

  } while (ret != -1 || errno != ECHILD);

}



/* Return nonzero if the given thread is still alive.  */

static int

‌linux_thread_alive (ptid_t ptid)

{

  struct lwp_info *lwp = find_lwp_pid (ptid);



  /* We assume we always know if a thread exits.  If a whole process

     exited but we still haven't been able to report it to GDB, we'll

     hold on to the last lwp of the dead process.  */

  if (lwp != NULL)

    return !lwp->dead;

  else

    return 0;

}



/* Return 1 if this lwp still has an interesting status pending.  If

   not (e.g., it had stopped for a breakpoint that is gone), return

   false.  */



static int

‌thread_still_has_status_pending_p (struct thread_info *thread)

{

  struct lwp_info *lp = get_thread_lwp (thread);



  if (!lp->status_pending_p)

    return 0;



  /* If we got a `vCont;t', but we haven't reported a stop yet, do

     report any status pending the LWP may have.  */

  if (thread->last_resume_kind == resume_stop

      && thread->last_status.kind != TARGET_WAITKIND_IGNORE)

    return 0;



  if (thread->last_resume_kind != resume_stop

      && (lp->stop_reason == LWP_STOPPED_BY_SW_BREAKPOINT

          || lp->stop_reason == LWP_STOPPED_BY_HW_BREAKPOINT))

    {

      struct thread_info *saved_thread;

      CORE_ADDR pc;

      int discard = 0;



      gdb_assert (lp->last_status != 0);



      pc = get_pc (lp);



      saved_thread = current_thread;

      current_thread = thread;



      if (pc != lp->stop_pc)

        {

          if (debug_threads)

            debug_printf ("PC of %ld changed\n",

                          lwpid_of (thread));

          discard = 1;

        }

      else if (lp->stop_reason == LWP_STOPPED_BY_SW_BREAKPOINT

               && !(*the_low_target.breakpoint_at) (pc))

        {

          if (debug_threads)

            debug_printf ("previous SW breakpoint of %ld gone\n",

                          lwpid_of (thread));

          discard = 1;

        }

      else if (lp->stop_reason == LWP_STOPPED_BY_HW_BREAKPOINT

               && !hardware_breakpoint_inserted_here (pc))

        {

          if (debug_threads)

            debug_printf ("previous HW breakpoint of %ld gone\n",

                          lwpid_of (thread));

          discard = 1;

        }



      current_thread = saved_thread;



      if (discard)

        {

          if (debug_threads)

            debug_printf ("discarding pending breakpoint status\n");

          lp->status_pending_p = 0;

          return 0;

        }

    }



  return 1;

}



/* Return 1 if this lwp has an interesting status pending.  */

static int

‌status_pending_p_callback (struct inferior_list_entry *entry, void *arg)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lp = get_thread_lwp (thread);

  ptid_t ptid = * (ptid_t *) arg;



  /* Check if we're only interested in events from a specific process

     or its lwps.  */

  if (!ptid_equal (minus_one_ptid, ptid)

      && ptid_get_pid (ptid) != ptid_get_pid (thread->entry.id))

    return 0;



  if (lp->status_pending_p

      && !thread_still_has_status_pending_p (thread))

    {

      linux_resume_one_lwp (lp, lp->stepping, GDB_SIGNAL_0, NULL);

      return 0;

    }



  return lp->status_pending_p;

}



static int

‌same_lwp (struct inferior_list_entry *entry, void *data)

{

  ptid_t ptid = *(ptid_t *) data;

  int lwp;



  if (ptid_get_lwp (ptid) != 0)

    lwp = ptid_get_lwp (ptid);

  else

    lwp = ptid_get_pid (ptid);



  if (ptid_get_lwp (entry->id) == lwp)

    return 1;



  return 0;

}



struct lwp_info *

‌find_lwp_pid (ptid_t ptid)

{

  struct inferior_list_entry *thread

    = find_inferior (&all_threads, same_lwp, &ptid);



  if (thread == NULL)

    return NULL;



  return get_thread_lwp ((struct thread_info *) thread);

}



/* Return the number of known LWPs in the tgid given by PID.  */



static int

‌num_lwps (int pid)

{

  struct inferior_list_entry *inf, *tmp;

  int count = 0;



  ALL_INFERIORS (&all_threads, inf, tmp)

    {

      if (ptid_get_pid (inf->id) == pid)

        count++;

    }



  return count;

}



/* Detect zombie thread group leaders, and "exit" them.  We can't reap

   their exits until all other threads in the group have exited.  */



static void

‌check_zombie_leaders (void)

{

  struct process_info *proc, *tmp;



  ALL_PROCESSES (proc, tmp)

    {

      pid_t leader_pid = pid_of (proc);

      struct lwp_info *leader_lp;



      leader_lp = find_lwp_pid (pid_to_ptid (leader_pid));



      if (debug_threads)

        debug_printf ("leader_pid=%d, leader_lp!=NULL=%d, "

                      "num_lwps=%d, zombie=%d\n",

                      leader_pid, leader_lp!= NULL, num_lwps (leader_pid),

                      linux_proc_pid_is_zombie (leader_pid));



      if (leader_lp != NULL

          /* Check if there are other threads in the group, as we may

             have raced with the inferior simply exiting.  */

          && !last_thread_of_process_p (leader_pid)

          && linux_proc_pid_is_zombie (leader_pid))

        {

          /* A leader zombie can mean one of two things:



             - It exited, and there's an exit status pending

             available, or only the leader exited (not the whole

             program).  In the latter case, we can't waitpid the

             leader's exit status until all other threads are gone.



             - There are 3 or more threads in the group, and a thread

             other than the leader exec'd.  On an exec, the Linux

             kernel destroys all other threads (except the execing

             one) in the thread group, and resets the execing thread's

             tid to the tgid.  No exit notification is sent for the

             execing thread -- from the ptracer's perspective, it

             appears as though the execing thread just vanishes.

             Until we reap all other threads except the leader and the

             execing thread, the leader will be zombie, and the

             execing thread will be in `D (disc sleep)'.  As soon as

             all other threads are reaped, the execing thread changes

             it's tid to the tgid, and the previous (zombie) leader

             vanishes, giving place to the "new" leader.  We could try

             distinguishing the exit and exec cases, by waiting once

             more, and seeing if something comes out, but it doesn't

             sound useful.  The previous leader _does_ go away, and

             we'll re-add the new one once we see the exec event

             (which is just the same as what would happen if the

             previous leader did exit voluntarily before some other

             thread execs).  */



          if (debug_threads)

            fprintf (stderr,

                     "CZL: Thread group leader %d zombie "

                     "(it exited, or another thread execd).\n",

                     leader_pid);



          delete_lwp (leader_lp);

        }

    }

}



/* Callback for `find_inferior'.  Returns the first LWP that is not

   stopped.  ARG is a PTID filter.  */



static int

‌not_stopped_callback (struct inferior_list_entry *entry, void *arg)

{

  struct thread_info *thr = (struct thread_info *) entry;

  struct lwp_info *lwp;

  ptid_t filter = *(ptid_t *) arg;



  if (!ptid_match (ptid_of (thr), filter))

    return 0;



  lwp = get_thread_lwp (thr);

  if (!lwp->stopped)

    return 1;



  return 0;

}



/* This function should only be called if the LWP got a SIGTRAP.



   Handle any tracepoint steps or hits.  Return true if a tracepoint

   event was handled, 0 otherwise.  */



static int

‌handle_tracepoints (struct lwp_info *lwp)

{

  struct thread_info *tinfo = get_lwp_thread (lwp);

  int tpoint_related_event = 0;



  gdb_assert (lwp->suspended == 0);



  /* If this tracepoint hit causes a tracing stop, we'll immediately

     uninsert tracepoints.  To do this, we temporarily pause all

     threads, unpatch away, and then unpause threads.  We need to make

     sure the unpausing doesn't resume LWP too.  */

  lwp->suspended++;



  /* And we need to be sure that any all-threads-stopping doesn't try

     to move threads out of the jump pads, as it could deadlock the

     inferior (LWP could be in the jump pad, maybe even holding the

     lock.)  */



  /* Do any necessary step collect actions.  */

  tpoint_related_event |= tracepoint_finished_step (tinfo, lwp->stop_pc);



  tpoint_related_event |= handle_tracepoint_bkpts (tinfo, lwp->stop_pc);



  /* See if we just hit a tracepoint and do its main collect

     actions.  */

  tpoint_related_event |= tracepoint_was_hit (tinfo, lwp->stop_pc);



  lwp->suspended--;



  gdb_assert (lwp->suspended == 0);

  gdb_assert (!stabilizing_threads || lwp->collecting_fast_tracepoint);



  if (tpoint_related_event)

    {

      if (debug_threads)

        debug_printf ("got a tracepoint event\n");

      return 1;

    }



  return 0;

}



/* Convenience wrapper.  Returns true if LWP is presently collecting a

   fast tracepoint.  */



static int

‌linux_fast_tracepoint_collecting (struct lwp_info *lwp,

                                  struct fast_tpoint_collect_status *status)

{

  CORE_ADDR thread_area;

  struct thread_info *thread = get_lwp_thread (lwp);



  if (the_low_target.get_thread_area == NULL)

    return 0;



  /* Get the thread area address.  This is used to recognize which

     thread is which when tracing with the in-process agent library.

     We don't read anything from the address, and treat it as opaque;

     it's the address itself that we assume is unique per-thread.  */

  if ((*the_low_target.get_thread_area) (lwpid_of (thread), &thread_area) == -1)

    return 0;



  return fast_tracepoint_collecting (thread_area, lwp->stop_pc, status);

}



/* The reason we resume in the caller, is because we want to be able

   to pass lwp->status_pending as WSTAT, and we need to clear

   status_pending_p before resuming, otherwise, linux_resume_one_lwp

   refuses to resume.  */



static int

‌maybe_move_out_of_jump_pad (struct lwp_info *lwp, int *wstat)

{

  struct thread_info *saved_thread;



  saved_thread = current_thread;

  current_thread = get_lwp_thread (lwp);



  if ((wstat == NULL

       || (WIFSTOPPED (*wstat) && WSTOPSIG (*wstat) != SIGTRAP))

      && supports_fast_tracepoints ()

      && agent_loaded_p ())

    {

      struct fast_tpoint_collect_status status;

      int r;



      if (debug_threads)

        debug_printf ("Checking whether LWP %ld needs to move out of the "

                      "jump pad.\n",

                      lwpid_of (current_thread));



      r = linux_fast_tracepoint_collecting (lwp, &status);



      if (wstat == NULL

          || (WSTOPSIG (*wstat) != SIGILL

              && WSTOPSIG (*wstat) != SIGFPE

              && WSTOPSIG (*wstat) != SIGSEGV

              && WSTOPSIG (*wstat) != SIGBUS))

        {

          lwp->collecting_fast_tracepoint = r;



          if (r != 0)

            {

              if (r == 1 && lwp->exit_jump_pad_bkpt == NULL)

                {

                  /* Haven't executed the original instruction yet.

                     Set breakpoint there, and wait till it's hit,

                     then single-step until exiting the jump pad.  */

                  lwp->exit_jump_pad_bkpt

                    = set_breakpoint_at (status.adjusted_insn_addr, NULL);

                }



              if (debug_threads)

                debug_printf ("Checking whether LWP %ld needs to move out of "

                              "the jump pad...it does\n",

                              lwpid_of (current_thread));

              current_thread = saved_thread;



              return 1;

            }

        }

      else

        {

          /* If we get a synchronous signal while collecting, *and*

             while executing the (relocated) original instruction,

             reset the PC to point at the tpoint address, before

             reporting to GDB.  Otherwise, it's an IPA lib bug: just

             report the signal to GDB, and pray for the best.  */



          lwp->collecting_fast_tracepoint = 0;



          if (r != 0

              && (status.adjusted_insn_addr <= lwp->stop_pc

                  && lwp->stop_pc < status.adjusted_insn_addr_end))

            {

              siginfo_t info;

              struct regcache *regcache;



              /* The si_addr on a few signals references the address

                 of the faulting instruction.  Adjust that as

                 well.  */

              if ((WSTOPSIG (*wstat) == SIGILL

                   || WSTOPSIG (*wstat) == SIGFPE

                   || WSTOPSIG (*wstat) == SIGBUS

                   || WSTOPSIG (*wstat) == SIGSEGV)

                  && ptrace (PTRACE_GETSIGINFO, lwpid_of (current_thread),

                             (PTRACE_TYPE_ARG3) 0, &info) == 0

                  /* Final check just to make sure we don't clobber

                     the siginfo of non-kernel-sent signals.  */

                  && (uintptr_t) info.si_addr == lwp->stop_pc)

                {

                  info.si_addr = (void *) (uintptr_t) status.tpoint_addr;

                  ptrace (PTRACE_SETSIGINFO, lwpid_of (current_thread),

                          (PTRACE_TYPE_ARG3) 0, &info);

                }



              regcache = get_thread_regcache (current_thread, 1);

              (*the_low_target.set_pc) (regcache, status.tpoint_addr);

              lwp->stop_pc = status.tpoint_addr;



              /* Cancel any fast tracepoint lock this thread was

                 holding.  */

              force_unlock_trace_buffer ();

            }



          if (lwp->exit_jump_pad_bkpt != NULL)

            {

              if (debug_threads)

                debug_printf ("Cancelling fast exit-jump-pad: removing bkpt. "

                              "stopping all threads momentarily.\n");



              stop_all_lwps (1, lwp);



              delete_breakpoint (lwp->exit_jump_pad_bkpt);

              lwp->exit_jump_pad_bkpt = NULL;



              unstop_all_lwps (1, lwp);



              gdb_assert (lwp->suspended >= 0);

            }

        }

    }



  if (debug_threads)

    debug_printf ("Checking whether LWP %ld needs to move out of the "

                  "jump pad...no\n",

                  lwpid_of (current_thread));



  current_thread = saved_thread;

  return 0;

}



/* Enqueue one signal in the "signals to report later when out of the

   jump pad" list.  */



static void

‌enqueue_one_deferred_signal (struct lwp_info *lwp, int *wstat)

{

  struct pending_signals *p_sig;

  struct thread_info *thread = get_lwp_thread (lwp);



  if (debug_threads)

    debug_printf ("Deferring signal %d for LWP %ld.\n",

                  WSTOPSIG (*wstat), lwpid_of (thread));



  if (debug_threads)

    {

      struct pending_signals *sig;



      for (sig = lwp->pending_signals_to_report;

           sig != NULL;

           sig = sig->prev)

        debug_printf ("   Already queued %d\n",

                      sig->signal);



      debug_printf ("   (no more currently queued signals)\n");

    }



  /* Don't enqueue non-RT signals if they are already in the deferred

     queue.  (SIGSTOP being the easiest signal to see ending up here

     twice)  */

  if (WSTOPSIG (*wstat) < __SIGRTMIN)

    {

      struct pending_signals *sig;



      for (sig = lwp->pending_signals_to_report;

           sig != NULL;

           sig = sig->prev)

        {

          if (sig->signal == WSTOPSIG (*wstat))

            {

              if (debug_threads)

                debug_printf ("Not requeuing already queued non-RT signal %d"

                              " for LWP %ld\n",

                              sig->signal,

                              lwpid_of (thread));

              return;

            }

        }

    }



  p_sig = xmalloc (sizeof (*p_sig));

  p_sig->prev = lwp->pending_signals_to_report;

  p_sig->signal = WSTOPSIG (*wstat);

  memset (&p_sig->info, 0, sizeof (siginfo_t));

  ptrace (PTRACE_GETSIGINFO, lwpid_of (thread), (PTRACE_TYPE_ARG3) 0,

          &p_sig->info);



  lwp->pending_signals_to_report = p_sig;

}



/* Dequeue one signal from the "signals to report later when out of

   the jump pad" list.  */



static int

‌dequeue_one_deferred_signal (struct lwp_info *lwp, int *wstat)

{

  struct thread_info *thread = get_lwp_thread (lwp);



  if (lwp->pending_signals_to_report != NULL)

    {

      struct pending_signals **p_sig;



      p_sig = &lwp->pending_signals_to_report;

      while ((*p_sig)->prev != NULL)

        p_sig = &(*p_sig)->prev;



      *wstat = W_STOPCODE ((*p_sig)->signal);

      if ((*p_sig)->info.si_signo != 0)

        ptrace (PTRACE_SETSIGINFO, lwpid_of (thread), (PTRACE_TYPE_ARG3) 0,

                &(*p_sig)->info);

      free (*p_sig);

      *p_sig = NULL;



      if (debug_threads)

        debug_printf ("Reporting deferred signal %d for LWP %ld.\n",

                      WSTOPSIG (*wstat), lwpid_of (thread));



      if (debug_threads)

        {

          struct pending_signals *sig;



          for (sig = lwp->pending_signals_to_report;

               sig != NULL;

               sig = sig->prev)

            debug_printf ("   Still queued %d\n",

                          sig->signal);



          debug_printf ("   (no more queued signals)\n");

        }



      return 1;

    }



  return 0;

}



/* Return true if the event in LP may be caused by breakpoint.  */



static int

‌wstatus_maybe_breakpoint (int wstatus)

{

  return (WIFSTOPPED (wstatus)

          && (WSTOPSIG (wstatus) == SIGTRAP

              /* SIGILL and SIGSEGV are also treated as traps in case a

                 breakpoint is inserted at the current PC.  */

              || WSTOPSIG (wstatus) == SIGILL

              || WSTOPSIG (wstatus) == SIGSEGV));

}



/* Fetch the possibly triggered data watchpoint info and store it in

   CHILD.



   On some archs, like x86, that use debug registers to set

   watchpoints, it's possible that the way to know which watched

   address trapped, is to check the register that is used to select

   which address to watch.  Problem is, between setting the watchpoint

   and reading back which data address trapped, the user may change

   the set of watchpoints, and, as a consequence, GDB changes the

   debug registers in the inferior.  To avoid reading back a stale

   stopped-data-address when that happens, we cache in LP the fact

   that a watchpoint trapped, and the corresponding data address, as

   soon as we see CHILD stop with a SIGTRAP.  If GDB changes the debug

   registers meanwhile, we have the cached data we can rely on.  */



static int

‌check_stopped_by_watchpoint (struct lwp_info *child)

{

  if (the_low_target.stopped_by_watchpoint != NULL)

    {

      struct thread_info *saved_thread;



      saved_thread = current_thread;

      current_thread = get_lwp_thread (child);



      if (the_low_target.stopped_by_watchpoint ())

        {

          child->stop_reason = LWP_STOPPED_BY_WATCHPOINT;



          if (the_low_target.stopped_data_address != NULL)

            child->stopped_data_address

              = the_low_target.stopped_data_address ();

          else

            child->stopped_data_address = 0;

        }



      current_thread = saved_thread;

    }



  return child->stop_reason == LWP_STOPPED_BY_WATCHPOINT;

}



/* Do low-level handling of the event, and check if we should go on

   and pass it to caller code.  Return the affected lwp if we are, or

   NULL otherwise.  */



static struct lwp_info *

‌linux_low_filter_event (int lwpid, int wstat)

{

  struct lwp_info *child;

  struct thread_info *thread;

  int have_stop_pc = 0;



  child = find_lwp_pid (pid_to_ptid (lwpid));



  /* If we didn't find a process, one of two things presumably happened:

     - A process we started and then detached from has exited.  Ignore it.

     - A process we are controlling has forked and the new child's stop

     was reported to us by the kernel.  Save its PID.  */

  if (child == NULL && WIFSTOPPED (wstat))

    {

      add_to_pid_list (&stopped_pids, lwpid, wstat);

      return NULL;

    }

  else if (child == NULL)

    return NULL;



  thread = get_lwp_thread (child);



  child->stopped = 1;



  child->last_status = wstat;



  /* Check if the thread has exited.  */

  if ((WIFEXITED (wstat) || WIFSIGNALED (wstat)))

    {

      if (debug_threads)

        debug_printf ("LLFE: %d exited.\n", lwpid);

      if (num_lwps (pid_of (thread)) > 1)

        {



          /* If there is at least one more LWP, then the exit signal was

             not the end of the debugged application and should be

             ignored.  */

          delete_lwp (child);

          return NULL;

        }

      else

        {

          /* This was the last lwp in the process.  Since events are

             serialized to GDB core, and we can't report this one

             right now, but GDB core and the other target layers will

             want to be notified about the exit code/signal, leave the

             status pending for the next time we're able to report

             it.  */

          mark_lwp_dead (child, wstat);

          return child;

        }

    }



  gdb_assert (WIFSTOPPED (wstat));



  if (WIFSTOPPED (wstat))

    {

      struct process_info *proc;



      /* Architecture-specific setup after inferior is running.  This

         needs to happen after we have attached to the inferior and it

         is stopped for the first time, but before we access any

         inferior registers.  */

      proc = find_process_pid (pid_of (thread));

      if (proc->private->new_inferior)

        {

          struct thread_info *saved_thread;



          saved_thread = current_thread;

          current_thread = thread;



          the_low_target.arch_setup ();



          current_thread = saved_thread;



          proc->private->new_inferior = 0;

        }

    }



  if (WIFSTOPPED (wstat) && child->must_set_ptrace_flags)

    {

      struct process_info *proc = find_process_pid (pid_of (thread));



      linux_enable_event_reporting (lwpid, proc->attached);

      child->must_set_ptrace_flags = 0;

    }



  /* Be careful to not overwrite stop_pc until

     check_stopped_by_breakpoint is called.  */

  if (WIFSTOPPED (wstat) && WSTOPSIG (wstat) == SIGTRAP

      && linux_is_extended_waitstatus (wstat))

    {

      child->stop_pc = get_pc (child);

      handle_extended_wait (child, wstat);

      return NULL;

    }



  if (WIFSTOPPED (wstat) && WSTOPSIG (wstat) == SIGTRAP

      && check_stopped_by_watchpoint (child))

    ;

  else if (WIFSTOPPED (wstat) && wstatus_maybe_breakpoint (wstat))

    {

      if (check_stopped_by_breakpoint (child))

        have_stop_pc = 1;

    }



  if (!have_stop_pc)

    child->stop_pc = get_pc (child);



  if (WIFSTOPPED (wstat) && WSTOPSIG (wstat) == SIGSTOP

      && child->stop_expected)

    {

      if (debug_threads)

        debug_printf ("Expected stop.\n");

      child->stop_expected = 0;



      if (thread->last_resume_kind == resume_stop)

        {

          /* We want to report the stop to the core.  Treat the

             SIGSTOP as a normal event.  */

        }

      else if (stopping_threads != NOT_STOPPING_THREADS)

        {

          /* Stopping threads.  We don't want this SIGSTOP to end up

             pending.  */

          return NULL;

        }

      else

        {

          /* Filter out the event.  */

          linux_resume_one_lwp (child, child->stepping, 0, NULL);

          return NULL;

        }

    }



  child->status_pending_p = 1;

  child->status_pending = wstat;

  return child;

}



/* Wait for an event from child(ren) WAIT_PTID, and return any that

   match FILTER_PTID (leaving others pending).  The PTIDs can be:

   minus_one_ptid, to specify any child; a pid PTID, specifying all

   lwps of a thread group; or a PTID representing a single lwp.  Store

   the stop status through the status pointer WSTAT.  OPTIONS is

   passed to the waitpid call.  Return 0 if no event was found and

   OPTIONS contains WNOHANG.  Return -1 if no unwaited-for children

   was found.  Return the PID of the stopped child otherwise.  */



static int

‌linux_wait_for_event_filtered (ptid_t wait_ptid, ptid_t filter_ptid,

                               int *wstatp, int options)

{

  struct thread_info *event_thread;

  struct lwp_info *event_child, *requested_child;

  sigset_t block_mask, prev_mask;



 retry:

  /* N.B. event_thread points to the thread_info struct that contains

     event_child.  Keep them in sync.  */

  event_thread = NULL;

  event_child = NULL;

  requested_child = NULL;



  /* Check for a lwp with a pending status.  */



  if (ptid_equal (filter_ptid, minus_one_ptid) || ptid_is_pid (filter_ptid))

    {

      event_thread = (struct thread_info *)

        find_inferior (&all_threads, status_pending_p_callback, &filter_ptid);

      if (event_thread != NULL)

        event_child = get_thread_lwp (event_thread);

      if (debug_threads && event_thread)

        debug_printf ("Got a pending child %ld\n", lwpid_of (event_thread));

    }

  else if (!ptid_equal (filter_ptid, null_ptid))

    {

      requested_child = find_lwp_pid (filter_ptid);



      if (stopping_threads == NOT_STOPPING_THREADS

          && requested_child->status_pending_p

          && requested_child->collecting_fast_tracepoint)

        {

          enqueue_one_deferred_signal (requested_child,

                                       &requested_child->status_pending);

          requested_child->status_pending_p = 0;

          requested_child->status_pending = 0;

          linux_resume_one_lwp (requested_child, 0, 0, NULL);

        }



      if (requested_child->suspended

          && requested_child->status_pending_p)

        {

          internal_error (__FILE__, __LINE__,

                          "requesting an event out of a"

                          " suspended child?");

        }



      if (requested_child->status_pending_p)

        {

          event_child = requested_child;

          event_thread = get_lwp_thread (event_child);

        }

    }



  if (event_child != NULL)

    {

      if (debug_threads)

        debug_printf ("Got an event from pending child %ld (%04x)\n",

                      lwpid_of (event_thread), event_child->status_pending);

      *wstatp = event_child->status_pending;

      event_child->status_pending_p = 0;

      event_child->status_pending = 0;

      current_thread = event_thread;

      return lwpid_of (event_thread);

    }



  /* But if we don't find a pending event, we'll have to wait.



     We only enter this loop if no process has a pending wait status.

     Thus any action taken in response to a wait status inside this

     loop is responding as soon as we detect the status, not after any

     pending events.  */



  /* Make sure SIGCHLD is blocked until the sigsuspend below.  Block

     all signals while here.  */

  sigfillset (&block_mask);

  sigprocmask (SIG_BLOCK, &block_mask, &prev_mask);



  /* Always pull all events out of the kernel.  We'll randomly select

     an event LWP out of all that have events, to prevent

     starvation.  */

  while (event_child == NULL)

    {

      pid_t ret = 0;



      /* Always use -1 and WNOHANG, due to couple of a kernel/ptrace

         quirks:



         - If the thread group leader exits while other threads in the

           thread group still exist, waitpid(TGID, ...) hangs.  That

           waitpid won't return an exit status until the other threads

           in the group are reaped.



         - When a non-leader thread execs, that thread just vanishes

           without reporting an exit (so we'd hang if we waited for it

           explicitly in that case).  The exec event is reported to

           the TGID pid (although we don't currently enable exec

           events).  */

      errno = 0;

      ret = my_waitpid (-1, wstatp, options | WNOHANG);



      if (debug_threads)

        debug_printf ("LWFE: waitpid(-1, ...) returned %d, %s\n",

                      ret, errno ? strerror (errno) : "ERRNO-OK");



      if (ret > 0)

        {

          if (debug_threads)

            {

              debug_printf ("LLW: waitpid %ld received %s\n",

                            (long) ret, status_to_str (*wstatp));

            }



          /* Filter all events.  IOW, leave all events pending.  We'll

             randomly select an event LWP out of all that have events

             below.  */

          linux_low_filter_event (ret, *wstatp);

          /* Retry until nothing comes out of waitpid.  A single

             SIGCHLD can indicate more than one child stopped.  */

          continue;

        }



      /* Now that we've pulled all events out of the kernel, check if

         there's any LWP with a status to report to the core.  */

      event_thread = (struct thread_info *)

        find_inferior (&all_threads, status_pending_p_callback, &filter_ptid);

      if (event_thread != NULL)

        {

          event_child = get_thread_lwp (event_thread);

          *wstatp = event_child->status_pending;

          event_child->status_pending_p = 0;

          event_child->status_pending = 0;

          break;

        }



      /* Check for zombie thread group leaders.  Those can't be reaped

         until all other threads in the thread group are.  */

      check_zombie_leaders ();



      /* If there are no resumed children left in the set of LWPs we

         want to wait for, bail.  We can't just block in

         waitpid/sigsuspend, because lwps might have been left stopped

         in trace-stop state, and we'd be stuck forever waiting for

         their status to change (which would only happen if we resumed

         them).  Even if WNOHANG is set, this return code is preferred

         over 0 (below), as it is more detailed.  */

      if ((find_inferior (&all_threads,

                          not_stopped_callback,

                          &wait_ptid) == NULL))

        {

          if (debug_threads)

            debug_printf ("LLW: exit (no unwaited-for LWP)\n");

          sigprocmask (SIG_SETMASK, &prev_mask, NULL);

          return -1;

        }



      /* No interesting event to report to the caller.  */

      if ((options & WNOHANG))

        {

          if (debug_threads)

            debug_printf ("WNOHANG set, no event found\n");



          sigprocmask (SIG_SETMASK, &prev_mask, NULL);

          return 0;

        }



      /* Block until we get an event reported with SIGCHLD.  */

      if (debug_threads)

        debug_printf ("sigsuspend'ing\n");



      sigsuspend (&prev_mask);

      sigprocmask (SIG_SETMASK, &prev_mask, NULL);

      goto retry;

    }



  sigprocmask (SIG_SETMASK, &prev_mask, NULL);



  current_thread = event_thread;



  /* Check for thread exit.  */

  if (! WIFSTOPPED (*wstatp))

    {

      gdb_assert (last_thread_of_process_p (pid_of (event_thread)));



      if (debug_threads)

        debug_printf ("LWP %d is the last lwp of process.  "

                      "Process %ld exiting.\n",

                      pid_of (event_thread), lwpid_of (event_thread));

      return lwpid_of (event_thread);

    }



  return lwpid_of (event_thread);

}



/* Wait for an event from child(ren) PTID.  PTIDs can be:

   minus_one_ptid, to specify any child; a pid PTID, specifying all

   lwps of a thread group; or a PTID representing a single lwp.  Store

   the stop status through the status pointer WSTAT.  OPTIONS is

   passed to the waitpid call.  Return 0 if no event was found and

   OPTIONS contains WNOHANG.  Return -1 if no unwaited-for children

   was found.  Return the PID of the stopped child otherwise.  */



static int

‌linux_wait_for_event (ptid_t ptid, int *wstatp, int options)

{

  return linux_wait_for_event_filtered (ptid, ptid, wstatp, options);

}



/* Count the LWP's that have had events.  */



static int

‌count_events_callback (struct inferior_list_entry *entry, void *data)

{

  struct thread_info *thread = (struct thread_info *) entry;

  int *count = data;



  gdb_assert (count != NULL);



  /* Count only resumed LWPs that have an event pending. */

  if (thread->last_status.kind == TARGET_WAITKIND_IGNORE

      && thread->last_resume_kind != resume_stop

      && thread->status_pending_p)

    (*count)++;



  return 0;

}



/* Select the LWP (if any) that is currently being single-stepped.  */



static int

‌select_singlestep_lwp_callback (struct inferior_list_entry *entry, void *data)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lp = get_thread_lwp (thread);



  if (thread->last_status.kind == TARGET_WAITKIND_IGNORE

      && thread->last_resume_kind == resume_step

      && lp->status_pending_p)

    return 1;

  else

    return 0;

}



/* Select the Nth LWP that has had a SIGTRAP event that should be

   reported to GDB.  */



static int

‌select_event_lwp_callback (struct inferior_list_entry *entry, void *data)

{

  struct thread_info *thread = (struct thread_info *) entry;

  int *selector = data;



  gdb_assert (selector != NULL);



  /* Select only resumed LWPs that have an event pending. */

  if (thread->last_resume_kind != resume_stop

      && thread->last_status.kind == TARGET_WAITKIND_IGNORE

      && thread->status_pending_p)

    if ((*selector)-- == 0)

      return 1;



  return 0;

}



/* Select one LWP out of those that have events pending.  */



static void

‌select_event_lwp (struct lwp_info **orig_lp)

{

  int num_events = 0;

  int random_selector;

  struct thread_info *event_thread = NULL;



  /* In all-stop, give preference to the LWP that is being

     single-stepped.  There will be at most one, and it's the LWP that

     the core is most interested in.  If we didn't do this, then we'd

     have to handle pending step SIGTRAPs somehow in case the core

     later continues the previously-stepped thread, otherwise we'd

     report the pending SIGTRAP, and the core, not having stepped the

     thread, wouldn't understand what the trap was for, and therefore

     would report it to the user as a random signal.  */

  if (!non_stop)

    {

      event_thread

        = (struct thread_info *) find_inferior (&all_threads,

                                                select_singlestep_lwp_callback,

                                                NULL);

      if (event_thread != NULL)

        {

          if (debug_threads)

            debug_printf ("SEL: Select single-step %s\n",

                          target_pid_to_str (ptid_of (event_thread)));

        }

    }

  if (event_thread == NULL)

    {

      /* No single-stepping LWP.  Select one at random, out of those

         which have had SIGTRAP events.  */



      /* First see how many SIGTRAP events we have.  */

      find_inferior (&all_threads, count_events_callback, &num_events);



      /* Now randomly pick a LWP out of those that have had a SIGTRAP.  */

      random_selector = (int)

        ((num_events * (double) rand ()) / (RAND_MAX + 1.0));



      if (debug_threads && num_events > 1)

        debug_printf ("SEL: Found %d SIGTRAP events, selecting #%d\n",

                      num_events, random_selector);



      event_thread

        = (struct thread_info *) find_inferior (&all_threads,

                                                select_event_lwp_callback,

                                                &random_selector);

    }



  if (event_thread != NULL)

    {

      struct lwp_info *event_lp = get_thread_lwp (event_thread);



      /* Switch the event LWP.  */

      *orig_lp = event_lp;

    }

}



/* Decrement the suspend count of an LWP.  */



static int

‌unsuspend_one_lwp (struct inferior_list_entry *entry, void *except)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  /* Ignore EXCEPT.  */

  if (lwp == except)

    return 0;



  lwp->suspended--;



  gdb_assert (lwp->suspended >= 0);

  return 0;

}



/* Decrement the suspend count of all LWPs, except EXCEPT, if non

   NULL.  */



static void

‌unsuspend_all_lwps (struct lwp_info *except)

{

  find_inferior (&all_threads, unsuspend_one_lwp, except);

}



static void move_out_of_jump_pad_callback (struct inferior_list_entry *entry);

static int stuck_in_jump_pad_callback (struct inferior_list_entry *entry,

                                       void *data);

static int lwp_running (struct inferior_list_entry *entry, void *data);

static ptid_t linux_wait_1 (ptid_t ptid,

                            struct target_waitstatus *ourstatus,

                            int target_options);



/* Stabilize threads (move out of jump pads).



   If a thread is midway collecting a fast tracepoint, we need to

   finish the collection and move it out of the jump pad before

   reporting the signal.



   This avoids recursion while collecting (when a signal arrives

   midway, and the signal handler itself collects), which would trash

   the trace buffer.  In case the user set a breakpoint in a signal

   handler, this avoids the backtrace showing the jump pad, etc..

   Most importantly, there are certain things we can't do safely if

   threads are stopped in a jump pad (or in its callee's).  For

   example:



     - starting a new trace run.  A thread still collecting the

   previous run, could trash the trace buffer when resumed.  The trace

   buffer control structures would have been reset but the thread had

   no way to tell.  The thread could even midway memcpy'ing to the

   buffer, which would mean that when resumed, it would clobber the

   trace buffer that had been set for a new run.



     - we can't rewrite/reuse the jump pads for new tracepoints

   safely.  Say you do tstart while a thread is stopped midway while

   collecting.  When the thread is later resumed, it finishes the

   collection, and returns to the jump pad, to execute the original

   instruction that was under the tracepoint jump at the time the

   older run had been started.  If the jump pad had been rewritten

   since for something else in the new run, the thread would now

   execute the wrong / random instructions.  */



static void

‌linux_stabilize_threads (void)

{

  struct thread_info *saved_thread;

  struct thread_info *thread_stuck;



  thread_stuck

    = (struct thread_info *) find_inferior (&all_threads,

                                            stuck_in_jump_pad_callback,

                                            NULL);

  if (thread_stuck != NULL)

    {

      if (debug_threads)

        debug_printf ("can't stabilize, LWP %ld is stuck in jump pad\n",

                      lwpid_of (thread_stuck));

      return;

    }



  saved_thread = current_thread;



  stabilizing_threads = 1;



  /* Kick 'em all.  */

  for_each_inferior (&all_threads, move_out_of_jump_pad_callback);



  /* Loop until all are stopped out of the jump pads.  */

  while (find_inferior (&all_threads, lwp_running, NULL) != NULL)

    {

      struct target_waitstatus ourstatus;

      struct lwp_info *lwp;

      int wstat;



      /* Note that we go through the full wait even loop.  While

         moving threads out of jump pad, we need to be able to step

         over internal breakpoints and such.  */

      linux_wait_1 (minus_one_ptid, &ourstatus, 0);



      if (ourstatus.kind == TARGET_WAITKIND_STOPPED)

        {

          lwp = get_thread_lwp (current_thread);



          /* Lock it.  */

          lwp->suspended++;



          if (ourstatus.value.sig != GDB_SIGNAL_0

              || current_thread->last_resume_kind == resume_stop)

            {

              wstat = W_STOPCODE (gdb_signal_to_host (ourstatus.value.sig));

              enqueue_one_deferred_signal (lwp, &wstat);

            }

        }

    }



  find_inferior (&all_threads, unsuspend_one_lwp, NULL);



  stabilizing_threads = 0;



  current_thread = saved_thread;



  if (debug_threads)

    {

      thread_stuck

        = (struct thread_info *) find_inferior (&all_threads,

                                                stuck_in_jump_pad_callback,

                                                NULL);

      if (thread_stuck != NULL)

        debug_printf ("couldn't stabilize, LWP %ld got stuck in jump pad\n",

                      lwpid_of (thread_stuck));

    }

}



static void async_file_mark (void);



/* Convenience function that is called when the kernel reports an

   event that is not passed out to GDB.  */



static ptid_t

‌ignore_event (struct target_waitstatus *ourstatus)

{

  /* If we got an event, there may still be others, as a single

     SIGCHLD can indicate more than one child stopped.  This forces

     another target_wait call.  */

  async_file_mark ();



  ourstatus->kind = TARGET_WAITKIND_IGNORE;

  return null_ptid;

}



/* Wait for process, returns status.  */



static ptid_t

‌linux_wait_1 (ptid_t ptid,

              struct target_waitstatus *ourstatus, int target_options)

{

  int w;

  struct lwp_info *event_child;

  int options;

  int pid;

  int step_over_finished;

  int bp_explains_trap;

  int maybe_internal_trap;

  int report_to_gdb;

  int trace_event;

  int in_step_range;



  if (debug_threads)

    {

      debug_enter ();

      debug_printf ("linux_wait_1: [%s]\n", target_pid_to_str (ptid));

    }



  /* Translate generic target options into linux options.  */

  options = __WALL;

  if (target_options & TARGET_WNOHANG)

    options |= WNOHANG;



  bp_explains_trap = 0;

  trace_event = 0;

  in_step_range = 0;

  ourstatus->kind = TARGET_WAITKIND_IGNORE;



  if (ptid_equal (step_over_bkpt, null_ptid))

    pid = linux_wait_for_event (ptid, &w, options);

  else

    {

      if (debug_threads)

        debug_printf ("step_over_bkpt set [%s], doing a blocking wait\n",

                      target_pid_to_str (step_over_bkpt));

      pid = linux_wait_for_event (step_over_bkpt, &w, options & ~WNOHANG);

    }



  if (pid == 0)

    {

      gdb_assert (target_options & TARGET_WNOHANG);



      if (debug_threads)

        {

          debug_printf ("linux_wait_1 ret = null_ptid, "

                        "TARGET_WAITKIND_IGNORE\n");

          debug_exit ();

        }



      ourstatus->kind = TARGET_WAITKIND_IGNORE;

      return null_ptid;

    }

  else if (pid == -1)

    {

      if (debug_threads)

        {

          debug_printf ("linux_wait_1 ret = null_ptid, "

                        "TARGET_WAITKIND_NO_RESUMED\n");

          debug_exit ();

        }



      ourstatus->kind = TARGET_WAITKIND_NO_RESUMED;

      return null_ptid;

    }



  event_child = get_thread_lwp (current_thread);



  /* linux_wait_for_event only returns an exit status for the last

     child of a process.  Report it.  */

  if (WIFEXITED (w) || WIFSIGNALED (w))

    {

      if (WIFEXITED (w))

        {

          ourstatus->kind = TARGET_WAITKIND_EXITED;

          ourstatus->value.integer = WEXITSTATUS (w);



          if (debug_threads)

            {

              debug_printf ("linux_wait_1 ret = %s, exited with "

                            "retcode %d\n",

                            target_pid_to_str (ptid_of (current_thread)),

                            WEXITSTATUS (w));

              debug_exit ();

            }

        }

      else

        {

          ourstatus->kind = TARGET_WAITKIND_SIGNALLED;

          ourstatus->value.sig = gdb_signal_from_host (WTERMSIG (w));



          if (debug_threads)

            {

              debug_printf ("linux_wait_1 ret = %s, terminated with "

                            "signal %d\n",

                            target_pid_to_str (ptid_of (current_thread)),

                            WTERMSIG (w));

              debug_exit ();

            }

        }



      return ptid_of (current_thread);

    }



  /* If this event was not handled before, and is not a SIGTRAP, we

     report it.  SIGILL and SIGSEGV are also treated as traps in case

     a breakpoint is inserted at the current PC.  If this target does

     not support internal breakpoints at all, we also report the

     SIGTRAP without further processing; it's of no concern to us.  */

  maybe_internal_trap

    = (supports_breakpoints ()

       && (WSTOPSIG (w) == SIGTRAP

           || ((WSTOPSIG (w) == SIGILL

                || WSTOPSIG (w) == SIGSEGV)

               && (*the_low_target.breakpoint_at) (event_child->stop_pc))));



  if (maybe_internal_trap)

    {

      /* Handle anything that requires bookkeeping before deciding to

         report the event or continue waiting.  */



      /* First check if we can explain the SIGTRAP with an internal

         breakpoint, or if we should possibly report the event to GDB.

         Do this before anything that may remove or insert a

         breakpoint.  */

      bp_explains_trap = breakpoint_inserted_here (event_child->stop_pc);



      /* We have a SIGTRAP, possibly a step-over dance has just

         finished.  If so, tweak the state machine accordingly,

         reinsert breakpoints and delete any reinsert (software

         single-step) breakpoints.  */

      step_over_finished = finish_step_over (event_child);



      /* Now invoke the callbacks of any internal breakpoints there.  */

      check_breakpoints (event_child->stop_pc);



      /* Handle tracepoint data collecting.  This may overflow the

         trace buffer, and cause a tracing stop, removing

         breakpoints.  */

      trace_event = handle_tracepoints (event_child);



      if (bp_explains_trap)

        {

          /* If we stepped or ran into an internal breakpoint, we've

             already handled it.  So next time we resume (from this

             PC), we should step over it.  */

          if (debug_threads)

            debug_printf ("Hit a gdbserver breakpoint.\n");



          if (breakpoint_here (event_child->stop_pc))

            event_child->need_step_over = 1;

        }

    }

  else

    {

      /* We have some other signal, possibly a step-over dance was in

         progress, and it should be cancelled too.  */

      step_over_finished = finish_step_over (event_child);

    }



  /* We have all the data we need.  Either report the event to GDB, or

     resume threads and keep waiting for more.  */



  /* If we're collecting a fast tracepoint, finish the collection and

     move out of the jump pad before delivering a signal.  See

     linux_stabilize_threads.  */



  if (WIFSTOPPED (w)

      && WSTOPSIG (w) != SIGTRAP

      && supports_fast_tracepoints ()

      && agent_loaded_p ())

    {

      if (debug_threads)

        debug_printf ("Got signal %d for LWP %ld.  Check if we need "

                      "to defer or adjust it.\n",

                      WSTOPSIG (w), lwpid_of (current_thread));



      /* Allow debugging the jump pad itself.  */

      if (current_thread->last_resume_kind != resume_step

          && maybe_move_out_of_jump_pad (event_child, &w))

        {

          enqueue_one_deferred_signal (event_child, &w);



          if (debug_threads)

            debug_printf ("Signal %d for LWP %ld deferred (in jump pad)\n",

                          WSTOPSIG (w), lwpid_of (current_thread));



          linux_resume_one_lwp (event_child, 0, 0, NULL);



          return ignore_event (ourstatus);

        }

    }



  if (event_child->collecting_fast_tracepoint)

    {

      if (debug_threads)

        debug_printf ("LWP %ld was trying to move out of the jump pad (%d). "

                      "Check if we're already there.\n",

                      lwpid_of (current_thread),

                      event_child->collecting_fast_tracepoint);



      trace_event = 1;



      event_child->collecting_fast_tracepoint

        = linux_fast_tracepoint_collecting (event_child, NULL);



      if (event_child->collecting_fast_tracepoint != 1)

        {

          /* No longer need this breakpoint.  */

          if (event_child->exit_jump_pad_bkpt != NULL)

            {

              if (debug_threads)

                debug_printf ("No longer need exit-jump-pad bkpt; removing it."

                              "stopping all threads momentarily.\n");



              /* Other running threads could hit this breakpoint.

                 We don't handle moribund locations like GDB does,

                 instead we always pause all threads when removing

                 breakpoints, so that any step-over or

                 decr_pc_after_break adjustment is always taken

                 care of while the breakpoint is still

                 inserted.  */

              stop_all_lwps (1, event_child);



              delete_breakpoint (event_child->exit_jump_pad_bkpt);

              event_child->exit_jump_pad_bkpt = NULL;



              unstop_all_lwps (1, event_child);



              gdb_assert (event_child->suspended >= 0);

            }

        }



      if (event_child->collecting_fast_tracepoint == 0)

        {

          if (debug_threads)

            debug_printf ("fast tracepoint finished "

                          "collecting successfully.\n");



          /* We may have a deferred signal to report.  */

          if (dequeue_one_deferred_signal (event_child, &w))

            {

              if (debug_threads)

                debug_printf ("dequeued one signal.\n");

            }

          else

            {

              if (debug_threads)

                debug_printf ("no deferred signals.\n");



              if (stabilizing_threads)

                {

                  ourstatus->kind = TARGET_WAITKIND_STOPPED;

                  ourstatus->value.sig = GDB_SIGNAL_0;



                  if (debug_threads)

                    {

                      debug_printf ("linux_wait_1 ret = %s, stopped "

                                    "while stabilizing threads\n",

                                    target_pid_to_str (ptid_of (current_thread)));

                      debug_exit ();

                    }



                  return ptid_of (current_thread);

                }

            }

        }

    }



  /* Check whether GDB would be interested in this event.  */



  /* If GDB is not interested in this signal, don't stop other

     threads, and don't report it to GDB.  Just resume the inferior

     right away.  We do this for threading-related signals as well as

     any that GDB specifically requested we ignore.  But never ignore

     SIGSTOP if we sent it ourselves, and do not ignore signals when

     stepping - they may require special handling to skip the signal

     handler.  */

  /* FIXME drow/2002-06-09: Get signal numbers from the inferior's

     thread library?  */

  if (WIFSTOPPED (w)

      && current_thread->last_resume_kind != resume_step

      && (

#if defined (USE_THREAD_DB) && !defined (__ANDROID__)

          (current_process ()->private->thread_db != NULL

           && (WSTOPSIG (w) == __SIGRTMIN

               || WSTOPSIG (w) == __SIGRTMIN + 1))

          ||

#endif

          (pass_signals[gdb_signal_from_host (WSTOPSIG (w))]

           && !(WSTOPSIG (w) == SIGSTOP

                && current_thread->last_resume_kind == resume_stop))))

    {

      siginfo_t info, *info_p;



      if (debug_threads)

        debug_printf ("Ignored signal %d for LWP %ld.\n",

                      WSTOPSIG (w), lwpid_of (current_thread));



      if (ptrace (PTRACE_GETSIGINFO, lwpid_of (current_thread),

                  (PTRACE_TYPE_ARG3) 0, &info) == 0)

        info_p = &info;

      else

        info_p = NULL;

      linux_resume_one_lwp (event_child, event_child->stepping,

                            WSTOPSIG (w), info_p);

      return ignore_event (ourstatus);

    }



  /* Note that all addresses are always "out of the step range" when

     there's no range to begin with.  */

  in_step_range = lwp_in_step_range (event_child);



  /* If GDB wanted this thread to single step, and the thread is out

     of the step range, we always want to report the SIGTRAP, and let

     GDB handle it.  Watchpoints should always be reported.  So should

     signals we can't explain.  A SIGTRAP we can't explain could be a

     GDB breakpoint --- we may or not support Z0 breakpoints.  If we

     do, we're be able to handle GDB breakpoints on top of internal

     breakpoints, by handling the internal breakpoint and still

     reporting the event to GDB.  If we don't, we're out of luck, GDB

     won't see the breakpoint hit.  */

  report_to_gdb = (!maybe_internal_trap

                   || (current_thread->last_resume_kind == resume_step

                       && !in_step_range)

                   || event_child->stop_reason == LWP_STOPPED_BY_WATCHPOINT

                   || (!step_over_finished && !in_step_range

                       && !bp_explains_trap && !trace_event)

                   || (gdb_breakpoint_here (event_child->stop_pc)

                       && gdb_condition_true_at_breakpoint (event_child->stop_pc)

                       && gdb_no_commands_at_breakpoint (event_child->stop_pc)));



  run_breakpoint_commands (event_child->stop_pc);



  /* We found no reason GDB would want us to stop.  We either hit one

     of our own breakpoints, or finished an internal step GDB

     shouldn't know about.  */

  if (!report_to_gdb)

    {

      if (debug_threads)

        {

          if (bp_explains_trap)

            debug_printf ("Hit a gdbserver breakpoint.\n");

          if (step_over_finished)

            debug_printf ("Step-over finished.\n");

          if (trace_event)

            debug_printf ("Tracepoint event.\n");

          if (lwp_in_step_range (event_child))

            debug_printf ("Range stepping pc 0x%s [0x%s, 0x%s).\n",

                          paddress (event_child->stop_pc),

                          paddress (event_child->step_range_start),

                          paddress (event_child->step_range_end));

        }



      /* We're not reporting this breakpoint to GDB, so apply the

         decr_pc_after_break adjustment to the inferior's regcache

         ourselves.  */



      if (the_low_target.set_pc != NULL)

        {

          struct regcache *regcache

            = get_thread_regcache (current_thread, 1);

          (*the_low_target.set_pc) (regcache, event_child->stop_pc);

        }



      /* We may have finished stepping over a breakpoint.  If so,

         we've stopped and suspended all LWPs momentarily except the

         stepping one.  This is where we resume them all again.  We're

         going to keep waiting, so use proceed, which handles stepping

         over the next breakpoint.  */

      if (debug_threads)

        debug_printf ("proceeding all threads.\n");



      if (step_over_finished)

        unsuspend_all_lwps (event_child);



      proceed_all_lwps ();

      return ignore_event (ourstatus);

    }



  if (debug_threads)

    {

      if (current_thread->last_resume_kind == resume_step)

        {

          if (event_child->step_range_start == event_child->step_range_end)

            debug_printf ("GDB wanted to single-step, reporting event.\n");

          else if (!lwp_in_step_range (event_child))

            debug_printf ("Out of step range, reporting event.\n");

        }

      if (event_child->stop_reason == LWP_STOPPED_BY_WATCHPOINT)

        debug_printf ("Stopped by watchpoint.\n");

      else if (gdb_breakpoint_here (event_child->stop_pc))

        debug_printf ("Stopped by GDB breakpoint.\n");

      if (debug_threads)

        debug_printf ("Hit a non-gdbserver trap event.\n");

    }



  /* Alright, we're going to report a stop.  */



  if (!stabilizing_threads)

    {

      /* In all-stop, stop all threads.  */

      if (!non_stop)

        stop_all_lwps (0, NULL);



      /* If we're not waiting for a specific LWP, choose an event LWP

         from among those that have had events.  Giving equal priority

         to all LWPs that have had events helps prevent

         starvation.  */

      if (ptid_equal (ptid, minus_one_ptid))

        {

          event_child->status_pending_p = 1;

          event_child->status_pending = w;



          select_event_lwp (&event_child);



          /* current_thread and event_child must stay in sync.  */

          current_thread = get_lwp_thread (event_child);



          event_child->status_pending_p = 0;

          w = event_child->status_pending;

        }



      if (step_over_finished)

        {

          if (!non_stop)

            {

              /* If we were doing a step-over, all other threads but

                 the stepping one had been paused in start_step_over,

                 with their suspend counts incremented.  We don't want

                 to do a full unstop/unpause, because we're in

                 all-stop mode (so we want threads stopped), but we

                 still need to unsuspend the other threads, to

                 decrement their `suspended' count back.  */

              unsuspend_all_lwps (event_child);

            }

          else

            {

              /* If we just finished a step-over, then all threads had

                 been momentarily paused.  In all-stop, that's fine,

                 we want threads stopped by now anyway.  In non-stop,

                 we need to re-resume threads that GDB wanted to be

                 running.  */

              unstop_all_lwps (1, event_child);

            }

        }



      /* Stabilize threads (move out of jump pads).  */

      if (!non_stop)

        stabilize_threads ();

    }

  else

    {

      /* If we just finished a step-over, then all threads had been

         momentarily paused.  In all-stop, that's fine, we want

         threads stopped by now anyway.  In non-stop, we need to

         re-resume threads that GDB wanted to be running.  */

      if (step_over_finished)

        unstop_all_lwps (1, event_child);

    }



  ourstatus->kind = TARGET_WAITKIND_STOPPED;



  /* Now that we've selected our final event LWP, un-adjust its PC if

     it was a software breakpoint.  */

  if (event_child->stop_reason == LWP_STOPPED_BY_SW_BREAKPOINT)

    {

      int decr_pc = the_low_target.decr_pc_after_break;



      if (decr_pc != 0)

        {

          struct regcache *regcache

            = get_thread_regcache (current_thread, 1);

          (*the_low_target.set_pc) (regcache, event_child->stop_pc + decr_pc);

        }

    }



  if (current_thread->last_resume_kind == resume_stop

      && WSTOPSIG (w) == SIGSTOP)

    {

      /* A thread that has been requested to stop by GDB with vCont;t,

         and it stopped cleanly, so report as SIG0.  The use of

         SIGSTOP is an implementation detail.  */

      ourstatus->value.sig = GDB_SIGNAL_0;

    }

  else if (current_thread->last_resume_kind == resume_stop

           && WSTOPSIG (w) != SIGSTOP)

    {

      /* A thread that has been requested to stop by GDB with vCont;t,

         but, it stopped for other reasons.  */

      ourstatus->value.sig = gdb_signal_from_host (WSTOPSIG (w));

    }

  else

    {

      ourstatus->value.sig = gdb_signal_from_host (WSTOPSIG (w));

    }



  gdb_assert (ptid_equal (step_over_bkpt, null_ptid));



  if (debug_threads)

    {

      debug_printf ("linux_wait_1 ret = %s, %d, %d\n",

                    target_pid_to_str (ptid_of (current_thread)),

                    ourstatus->kind, ourstatus->value.sig);

      debug_exit ();

    }



  return ptid_of (current_thread);

}



/* Get rid of any pending event in the pipe.  */

static void

‌async_file_flush (void)

{

  int ret;

  char buf;



  do

    ret = read (linux_event_pipe[0], &buf, 1);

  while (ret >= 0 || (ret == -1 && errno == EINTR));

}



/* Put something in the pipe, so the event loop wakes up.  */

static void

‌async_file_mark (void)

{

  int ret;



  async_file_flush ();



  do

    ret = write (linux_event_pipe[1], "+", 1);

  while (ret == 0 || (ret == -1 && errno == EINTR));



  /* Ignore EAGAIN.  If the pipe is full, the event loop will already

     be awakened anyway.  */

}



static ptid_t

‌linux_wait (ptid_t ptid,

            struct target_waitstatus *ourstatus, int target_options)

{

  ptid_t event_ptid;



  /* Flush the async file first.  */

  if (target_is_async_p ())

    async_file_flush ();



  do

    {

      event_ptid = linux_wait_1 (ptid, ourstatus, target_options);

    }

  while ((target_options & TARGET_WNOHANG) == 0

         && ptid_equal (event_ptid, null_ptid)

         && ourstatus->kind == TARGET_WAITKIND_IGNORE);



  /* If at least one stop was reported, there may be more.  A single

     SIGCHLD can signal more than one child stop.  */

  if (target_is_async_p ()

      && (target_options & TARGET_WNOHANG) != 0

      && !ptid_equal (event_ptid, null_ptid))

    async_file_mark ();



  return event_ptid;

}



/* Send a signal to an LWP.  */



static int

‌kill_lwp (unsigned long lwpid, int signo)

{

  /* Use tkill, if possible, in case we are using nptl threads.  If tkill

     fails, then we are not using nptl threads and we should be using kill.  */



#ifdef __NR_tkill

  {

    static int tkill_failed;



    if (!tkill_failed)

      {

        int ret;



        errno = 0;

        ret = syscall (__NR_tkill, lwpid, signo);

        if (errno != ENOSYS)

          return ret;

        tkill_failed = 1;

      }

  }

#endif



  return kill (lwpid, signo);

}



void

‌linux_stop_lwp (struct lwp_info *lwp)

{

  send_sigstop (lwp);

}



static void

‌send_sigstop (struct lwp_info *lwp)

{

  int pid;



  pid = lwpid_of (get_lwp_thread (lwp));



  /* If we already have a pending stop signal for this process, don't

     send another.  */

  if (lwp->stop_expected)

    {

      if (debug_threads)

        debug_printf ("Have pending sigstop for lwp %d\n", pid);



      return;

    }



  if (debug_threads)

    debug_printf ("Sending sigstop to lwp %d\n", pid);



  lwp->stop_expected = 1;

  kill_lwp (pid, SIGSTOP);

}



static int

‌send_sigstop_callback (struct inferior_list_entry *entry, void *except)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  /* Ignore EXCEPT.  */

  if (lwp == except)

    return 0;



  if (lwp->stopped)

    return 0;



  send_sigstop (lwp);

  return 0;

}



/* Increment the suspend count of an LWP, and stop it, if not stopped

   yet.  */

static int

‌suspend_and_send_sigstop_callback (struct inferior_list_entry *entry,

                                   void *except)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  /* Ignore EXCEPT.  */

  if (lwp == except)

    return 0;



  lwp->suspended++;



  return send_sigstop_callback (entry, except);

}



static void

‌mark_lwp_dead (struct lwp_info *lwp, int wstat)

{

  /* It's dead, really.  */

  lwp->dead = 1;



  /* Store the exit status for later.  */

  lwp->status_pending_p = 1;

  lwp->status_pending = wstat;



  /* Prevent trying to stop it.  */

  lwp->stopped = 1;



  /* No further stops are expected from a dead lwp.  */

  lwp->stop_expected = 0;

}



/* Wait for all children to stop for the SIGSTOPs we just queued.  */



static void

‌wait_for_sigstop (void)

{

  struct thread_info *saved_thread;

  ptid_t saved_tid;

  int wstat;

  int ret;



  saved_thread = current_thread;

  if (saved_thread != NULL)

    saved_tid = saved_thread->entry.id;

  else

    saved_tid = null_ptid; /* avoid bogus unused warning */



  if (debug_threads)

    debug_printf ("wait_for_sigstop: pulling events\n");



  /* Passing NULL_PTID as filter indicates we want all events to be

     left pending.  Eventually this returns when there are no

     unwaited-for children left.  */

  ret = linux_wait_for_event_filtered (minus_one_ptid, null_ptid,

                                       &wstat, __WALL);

  gdb_assert (ret == -1);



  if (saved_thread == NULL || linux_thread_alive (saved_tid))

    current_thread = saved_thread;

  else

    {

      if (debug_threads)

        debug_printf ("Previously current thread died.\n");



      if (non_stop)

        {

          /* We can't change the current inferior behind GDB's back,

             otherwise, a subsequent command may apply to the wrong

             process.  */

          current_thread = NULL;

        }

      else

        {

          /* Set a valid thread as current.  */

          set_desired_thread (0);

        }

    }

}



/* Returns true if LWP ENTRY is stopped in a jump pad, and we can't

   move it out, because we need to report the stop event to GDB.  For

   example, if the user puts a breakpoint in the jump pad, it's

   because she wants to debug it.  */



static int

‌stuck_in_jump_pad_callback (struct inferior_list_entry *entry, void *data)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  gdb_assert (lwp->suspended == 0);

  gdb_assert (lwp->stopped);



  /* Allow debugging the jump pad, gdb_collect, etc..  */

  return (supports_fast_tracepoints ()

          && agent_loaded_p ()

          && (gdb_breakpoint_here (lwp->stop_pc)

              || lwp->stop_reason == LWP_STOPPED_BY_WATCHPOINT

              || thread->last_resume_kind == resume_step)

          && linux_fast_tracepoint_collecting (lwp, NULL));

}



static void

‌move_out_of_jump_pad_callback (struct inferior_list_entry *entry)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  int *wstat;



  gdb_assert (lwp->suspended == 0);

  gdb_assert (lwp->stopped);



  wstat = lwp->status_pending_p ? &lwp->status_pending : NULL;



  /* Allow debugging the jump pad, gdb_collect, etc.  */

  if (!gdb_breakpoint_here (lwp->stop_pc)

      && lwp->stop_reason != LWP_STOPPED_BY_WATCHPOINT

      && thread->last_resume_kind != resume_step

      && maybe_move_out_of_jump_pad (lwp, wstat))

    {

      if (debug_threads)

        debug_printf ("LWP %ld needs stabilizing (in jump pad)\n",

                      lwpid_of (thread));



      if (wstat)

        {

          lwp->status_pending_p = 0;

          enqueue_one_deferred_signal (lwp, wstat);



          if (debug_threads)

            debug_printf ("Signal %d for LWP %ld deferred "

                          "(in jump pad)\n",

                          WSTOPSIG (*wstat), lwpid_of (thread));

        }



      linux_resume_one_lwp (lwp, 0, 0, NULL);

    }

  else

    lwp->suspended++;

}



static int

‌lwp_running (struct inferior_list_entry *entry, void *data)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  if (lwp->dead)

    return 0;

  if (lwp->stopped)

    return 0;

  return 1;

}



/* Stop all lwps that aren't stopped yet, except EXCEPT, if not NULL.

   If SUSPEND, then also increase the suspend count of every LWP,

   except EXCEPT.  */



static void

‌stop_all_lwps (int suspend, struct lwp_info *except)

{

  /* Should not be called recursively.  */

  gdb_assert (stopping_threads == NOT_STOPPING_THREADS);



  if (debug_threads)

    {

      debug_enter ();

      debug_printf ("stop_all_lwps (%s, except=%s)\n",

                    suspend ? "stop-and-suspend" : "stop",

                    except != NULL

                    ? target_pid_to_str (ptid_of (get_lwp_thread (except)))

                    : "none");

    }



  stopping_threads = (suspend

                      ? STOPPING_AND_SUSPENDING_THREADS

                      : STOPPING_THREADS);



  if (suspend)

    find_inferior (&all_threads, suspend_and_send_sigstop_callback, except);

  else

    find_inferior (&all_threads, send_sigstop_callback, except);

  wait_for_sigstop ();

  stopping_threads = NOT_STOPPING_THREADS;



  if (debug_threads)

    {

      debug_printf ("stop_all_lwps done, setting stopping_threads "

                    "back to !stopping\n");

      debug_exit ();

    }

}



/* Resume execution of the inferior process.

   If STEP is nonzero, single-step it.

   If SIGNAL is nonzero, give it that signal.  */



static void

‌linux_resume_one_lwp (struct lwp_info *lwp,

                      int step, int signal, siginfo_t *info)

{

  struct thread_info *thread = get_lwp_thread (lwp);

  struct thread_info *saved_thread;

  int fast_tp_collecting;



  if (lwp->stopped == 0)

    return;



  fast_tp_collecting = lwp->collecting_fast_tracepoint;



  gdb_assert (!stabilizing_threads || fast_tp_collecting);



  /* Cancel actions that rely on GDB not changing the PC (e.g., the

     user used the "jump" command, or "set $pc = foo").  */

  if (lwp->stop_pc != get_pc (lwp))

    {

      /* Collecting 'while-stepping' actions doesn't make sense

         anymore.  */

      release_while_stepping_state_list (thread);

    }



  /* If we have pending signals or status, and a new signal, enqueue the

     signal.  Also enqueue the signal if we are waiting to reinsert a

     breakpoint; it will be picked up again below.  */

  if (signal != 0

      && (lwp->status_pending_p

          || lwp->pending_signals != NULL

          || lwp->bp_reinsert != 0

          || fast_tp_collecting))

    {

      struct pending_signals *p_sig;

      p_sig = xmalloc (sizeof (*p_sig));

      p_sig->prev = lwp->pending_signals;

      p_sig->signal = signal;

      if (info == NULL)

        memset (&p_sig->info, 0, sizeof (siginfo_t));

      else

        memcpy (&p_sig->info, info, sizeof (siginfo_t));

      lwp->pending_signals = p_sig;

    }



  if (lwp->status_pending_p)

    {

      if (debug_threads)

        debug_printf ("Not resuming lwp %ld (%s, signal %d, stop %s);"

                      " has pending status\n",

                      lwpid_of (thread), step ? "step" : "continue", signal,

                      lwp->stop_expected ? "expected" : "not expected");

      return;

    }



  saved_thread = current_thread;

  current_thread = thread;



  if (debug_threads)

    debug_printf ("Resuming lwp %ld (%s, signal %d, stop %s)\n",

                  lwpid_of (thread), step ? "step" : "continue", signal,

                  lwp->stop_expected ? "expected" : "not expected");



  /* This bit needs some thinking about.  If we get a signal that

     we must report while a single-step reinsert is still pending,

     we often end up resuming the thread.  It might be better to

     (ew) allow a stack of pending events; then we could be sure that

     the reinsert happened right away and not lose any signals.



     Making this stack would also shrink the window in which breakpoints are

     uninserted (see comment in linux_wait_for_lwp) but not enough for

     complete correctness, so it won't solve that problem.  It may be

     worthwhile just to solve this one, however.  */

  if (lwp->bp_reinsert != 0)

    {

      if (debug_threads)

        debug_printf ("  pending reinsert at 0x%s\n",

                      paddress (lwp->bp_reinsert));



      if (can_hardware_single_step ())

        {

          if (fast_tp_collecting == 0)

            {

              if (step == 0)

                fprintf (stderr, "BAD - reinserting but not stepping.\n");

              if (lwp->suspended)

                fprintf (stderr, "BAD - reinserting and suspended(%d).\n",

                         lwp->suspended);

            }



          step = 1;

        }



      /* Postpone any pending signal.  It was enqueued above.  */

      signal = 0;

    }



  if (fast_tp_collecting == 1)

    {

      if (debug_threads)

        debug_printf ("lwp %ld wants to get out of fast tracepoint jump pad"

                      " (exit-jump-pad-bkpt)\n",

                      lwpid_of (thread));



      /* Postpone any pending signal.  It was enqueued above.  */

      signal = 0;

    }

  else if (fast_tp_collecting == 2)

    {

      if (debug_threads)

        debug_printf ("lwp %ld wants to get out of fast tracepoint jump pad"

                      " single-stepping\n",

                      lwpid_of (thread));



      if (can_hardware_single_step ())

        step = 1;

      else

        {

          internal_error (__FILE__, __LINE__,

                          "moving out of jump pad single-stepping"

                          " not implemented on this target");

        }



      /* Postpone any pending signal.  It was enqueued above.  */

      signal = 0;

    }



  /* If we have while-stepping actions in this thread set it stepping.

     If we have a signal to deliver, it may or may not be set to

     SIG_IGN, we don't know.  Assume so, and allow collecting

     while-stepping into a signal handler.  A possible smart thing to

     do would be to set an internal breakpoint at the signal return

     address, continue, and carry on catching this while-stepping

     action only when that breakpoint is hit.  A future

     enhancement.  */

  if (thread->while_stepping != NULL

      && can_hardware_single_step ())

    {

      if (debug_threads)

        debug_printf ("lwp %ld has a while-stepping action -> forcing step.\n",

                      lwpid_of (thread));

      step = 1;

    }



  if (the_low_target.get_pc != NULL)

    {

      struct regcache *regcache = get_thread_regcache (current_thread, 1);



      lwp->stop_pc = (*the_low_target.get_pc) (regcache);



      if (debug_threads)

        {

          debug_printf ("  %s from pc 0x%lx\n", step ? "step" : "continue",

                        (long) lwp->stop_pc);

        }

    }



  /* If we have pending signals, consume one unless we are trying to

     reinsert a breakpoint or we're trying to finish a fast tracepoint

     collect.  */

  if (lwp->pending_signals != NULL

      && lwp->bp_reinsert == 0

      && fast_tp_collecting == 0)

    {

      struct pending_signals **p_sig;



      p_sig = &lwp->pending_signals;

      while ((*p_sig)->prev != NULL)

        p_sig = &(*p_sig)->prev;



      signal = (*p_sig)->signal;

      if ((*p_sig)->info.si_signo != 0)

        ptrace (PTRACE_SETSIGINFO, lwpid_of (thread), (PTRACE_TYPE_ARG3) 0,

                &(*p_sig)->info);



      free (*p_sig);

      *p_sig = NULL;

    }



  if (the_low_target.prepare_to_resume != NULL)

    the_low_target.prepare_to_resume (lwp);



  regcache_invalidate_thread (thread);

  errno = 0;

  lwp->stopped = 0;

  lwp->stop_reason = LWP_STOPPED_BY_NO_REASON;

  lwp->stepping = step;

  ptrace (step ? PTRACE_SINGLESTEP : PTRACE_CONT, lwpid_of (thread),

          (PTRACE_TYPE_ARG3) 0,

          /* Coerce to a uintptr_t first to avoid potential gcc warning

             of coercing an 8 byte integer to a 4 byte pointer.  */

          (PTRACE_TYPE_ARG4) (uintptr_t) signal);



  current_thread = saved_thread;

  if (errno)

    {

      /* ESRCH from ptrace either means that the thread was already

         running (an error) or that it is gone (a race condition).  If

         it's gone, we will get a notification the next time we wait,

         so we can ignore the error.  We could differentiate these

         two, but it's tricky without waiting; the thread still exists

         as a zombie, so sending it signal 0 would succeed.  So just

         ignore ESRCH.  */

      if (errno == ESRCH)

        return;



      perror_with_name ("ptrace");

    }

}



‌struct thread_resume_array

{

  struct thread_resume *resume;

  size_t n;

};



/* This function is called once per thread via find_inferior.

   ARG is a pointer to a thread_resume_array struct.

   We look up the thread specified by ENTRY in ARG, and mark the thread

   with a pointer to the appropriate resume request.



   This algorithm is O(threads * resume elements), but resume elements

   is small (and will remain small at least until GDB supports thread

   suspension).  */



static int

‌linux_set_resume_request (struct inferior_list_entry *entry, void *arg)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  int ndx;

  struct thread_resume_array *r;



  r = arg;



  for (ndx = 0; ndx < r->n; ndx++)

    {

      ptid_t ptid = r->resume[ndx].thread;

      if (ptid_equal (ptid, minus_one_ptid)

          || ptid_equal (ptid, entry->id)

          /* Handle both 'pPID' and 'pPID.-1' as meaning 'all threads

             of PID'.  */

          || (ptid_get_pid (ptid) == pid_of (thread)

              && (ptid_is_pid (ptid)

                  || ptid_get_lwp (ptid) == -1)))

        {

          if (r->resume[ndx].kind == resume_stop

              && thread->last_resume_kind == resume_stop)

            {

              if (debug_threads)

                debug_printf ("already %s LWP %ld at GDB's request\n",

                              (thread->last_status.kind

                               == TARGET_WAITKIND_STOPPED)

                              ? "stopped"

                              : "stopping",

                              lwpid_of (thread));



              continue;

            }



          lwp->resume = &r->resume[ndx];

          thread->last_resume_kind = lwp->resume->kind;



          lwp->step_range_start = lwp->resume->step_range_start;

          lwp->step_range_end = lwp->resume->step_range_end;



          /* If we had a deferred signal to report, dequeue one now.

             This can happen if LWP gets more than one signal while

             trying to get out of a jump pad.  */

          if (lwp->stopped

              && !lwp->status_pending_p

              && dequeue_one_deferred_signal (lwp, &lwp->status_pending))

            {

              lwp->status_pending_p = 1;



              if (debug_threads)

                debug_printf ("Dequeueing deferred signal %d for LWP %ld, "

                              "leaving status pending.\n",

                              WSTOPSIG (lwp->status_pending),

                              lwpid_of (thread));

            }



          return 0;

        }

    }



  /* No resume action for this thread.  */

  lwp->resume = NULL;



  return 0;

}



/* find_inferior callback for linux_resume.

   Set *FLAG_P if this lwp has an interesting status pending.  */



static int

‌resume_status_pending_p (struct inferior_list_entry *entry, void *flag_p)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  /* LWPs which will not be resumed are not interesting, because

     we might not wait for them next time through linux_wait.  */

  if (lwp->resume == NULL)

    return 0;



  if (thread_still_has_status_pending_p (thread))

    * (int *) flag_p = 1;



  return 0;

}



/* Return 1 if this lwp that GDB wants running is stopped at an

   internal breakpoint that we need to step over.  It assumes that any

   required STOP_PC adjustment has already been propagated to the

   inferior's regcache.  */



static int

‌need_step_over_p (struct inferior_list_entry *entry, void *dummy)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  struct thread_info *saved_thread;

  CORE_ADDR pc;



  /* LWPs which will not be resumed are not interesting, because we

     might not wait for them next time through linux_wait.  */



  if (!lwp->stopped)

    {

      if (debug_threads)

        debug_printf ("Need step over [LWP %ld]? Ignoring, not stopped\n",

                      lwpid_of (thread));

      return 0;

    }



  if (thread->last_resume_kind == resume_stop)

    {

      if (debug_threads)

        debug_printf ("Need step over [LWP %ld]? Ignoring, should remain"

                      " stopped\n",

                      lwpid_of (thread));

      return 0;

    }



  gdb_assert (lwp->suspended >= 0);



  if (lwp->suspended)

    {

      if (debug_threads)

        debug_printf ("Need step over [LWP %ld]? Ignoring, suspended\n",

                      lwpid_of (thread));

      return 0;

    }



  if (!lwp->need_step_over)

    {

      if (debug_threads)

        debug_printf ("Need step over [LWP %ld]? No\n", lwpid_of (thread));

    }



  if (lwp->status_pending_p)

    {

      if (debug_threads)

        debug_printf ("Need step over [LWP %ld]? Ignoring, has pending"

                      " status.\n",

                      lwpid_of (thread));

      return 0;

    }



  /* Note: PC, not STOP_PC.  Either GDB has adjusted the PC already,

     or we have.  */

  pc = get_pc (lwp);



  /* If the PC has changed since we stopped, then don't do anything,

     and let the breakpoint/tracepoint be hit.  This happens if, for

     instance, GDB handled the decr_pc_after_break subtraction itself,

     GDB is OOL stepping this thread, or the user has issued a "jump"

     command, or poked thread's registers herself.  */

  if (pc != lwp->stop_pc)

    {

      if (debug_threads)

        debug_printf ("Need step over [LWP %ld]? Cancelling, PC was changed. "

                      "Old stop_pc was 0x%s, PC is now 0x%s\n",

                      lwpid_of (thread),

                      paddress (lwp->stop_pc), paddress (pc));



      lwp->need_step_over = 0;

      return 0;

    }



  saved_thread = current_thread;

  current_thread = thread;



  /* We can only step over breakpoints we know about.  */

  if (breakpoint_here (pc) || fast_tracepoint_jump_here (pc))

    {

      /* Don't step over a breakpoint that GDB expects to hit

         though.  If the condition is being evaluated on the target's side

         and it evaluate to false, step over this breakpoint as well.  */

      if (gdb_breakpoint_here (pc)

          && gdb_condition_true_at_breakpoint (pc)

          && gdb_no_commands_at_breakpoint (pc))

        {

          if (debug_threads)

            debug_printf ("Need step over [LWP %ld]? yes, but found"

                          " GDB breakpoint at 0x%s; skipping step over\n",

                          lwpid_of (thread), paddress (pc));



          current_thread = saved_thread;

          return 0;

        }

      else

        {

          if (debug_threads)

            debug_printf ("Need step over [LWP %ld]? yes, "

                          "found breakpoint at 0x%s\n",

                          lwpid_of (thread), paddress (pc));



          /* We've found an lwp that needs stepping over --- return 1 so

             that find_inferior stops looking.  */

          current_thread = saved_thread;



          /* If the step over is cancelled, this is set again.  */

          lwp->need_step_over = 0;

          return 1;

        }

    }



  current_thread = saved_thread;



  if (debug_threads)

    debug_printf ("Need step over [LWP %ld]? No, no breakpoint found"

                  " at 0x%s\n",

                  lwpid_of (thread), paddress (pc));



  return 0;

}



/* Start a step-over operation on LWP.  When LWP stopped at a

   breakpoint, to make progress, we need to remove the breakpoint out

   of the way.  If we let other threads run while we do that, they may

   pass by the breakpoint location and miss hitting it.  To avoid

   that, a step-over momentarily stops all threads while LWP is

   single-stepped while the breakpoint is temporarily uninserted from

   the inferior.  When the single-step finishes, we reinsert the

   breakpoint, and let all threads that are supposed to be running,

   run again.



   On targets that don't support hardware single-step, we don't

   currently support full software single-stepping.  Instead, we only

   support stepping over the thread event breakpoint, by asking the

   low target where to place a reinsert breakpoint.  Since this

   routine assumes the breakpoint being stepped over is a thread event

   breakpoint, it usually assumes the return address of the current

   function is a good enough place to set the reinsert breakpoint.  */



static int

‌start_step_over (struct lwp_info *lwp)

{

  struct thread_info *thread = get_lwp_thread (lwp);

  struct thread_info *saved_thread;

  CORE_ADDR pc;

  int step;



  if (debug_threads)

    debug_printf ("Starting step-over on LWP %ld.  Stopping all threads\n",

                  lwpid_of (thread));



  stop_all_lwps (1, lwp);

  gdb_assert (lwp->suspended == 0);



  if (debug_threads)

    debug_printf ("Done stopping all threads for step-over.\n");



  /* Note, we should always reach here with an already adjusted PC,

     either by GDB (if we're resuming due to GDB's request), or by our

     caller, if we just finished handling an internal breakpoint GDB

     shouldn't care about.  */

  pc = get_pc (lwp);



  saved_thread = current_thread;

  current_thread = thread;



  lwp->bp_reinsert = pc;

  uninsert_breakpoints_at (pc);

  uninsert_fast_tracepoint_jumps_at (pc);



  if (can_hardware_single_step ())

    {

      step = 1;

    }

  else

    {

      CORE_ADDR raddr = (*the_low_target.breakpoint_reinsert_addr) ();

      set_reinsert_breakpoint (raddr);

      step = 0;

    }



  current_thread = saved_thread;



  linux_resume_one_lwp (lwp, step, 0, NULL);



  /* Require next event from this LWP.  */

  step_over_bkpt = thread->entry.id;

  return 1;

}



/* Finish a step-over.  Reinsert the breakpoint we had uninserted in

   start_step_over, if still there, and delete any reinsert

   breakpoints we've set, on non hardware single-step targets.  */



static int

‌finish_step_over (struct lwp_info *lwp)

{

  if (lwp->bp_reinsert != 0)

    {

      if (debug_threads)

        debug_printf ("Finished step over.\n");



      /* Reinsert any breakpoint at LWP->BP_REINSERT.  Note that there

         may be no breakpoint to reinsert there by now.  */

      reinsert_breakpoints_at (lwp->bp_reinsert);

      reinsert_fast_tracepoint_jumps_at (lwp->bp_reinsert);



      lwp->bp_reinsert = 0;



      /* Delete any software-single-step reinsert breakpoints.  No

         longer needed.  We don't have to worry about other threads

         hitting this trap, and later not being able to explain it,

         because we were stepping over a breakpoint, and we hold all

         threads but LWP stopped while doing that.  */

      if (!can_hardware_single_step ())

        delete_reinsert_breakpoints ();



      step_over_bkpt = null_ptid;

      return 1;

    }

  else

    return 0;

}



/* This function is called once per thread.  We check the thread's resume

   request, which will tell us whether to resume, step, or leave the thread

   stopped; and what signal, if any, it should be sent.



   For threads which we aren't explicitly told otherwise, we preserve

   the stepping flag; this is used for stepping over gdbserver-placed

   breakpoints.



   If pending_flags was set in any thread, we queue any needed

   signals, since we won't actually resume.  We already have a pending

   event to report, so we don't need to preserve any step requests;

   they should be re-issued if necessary.  */



static int

‌linux_resume_one_thread (struct inferior_list_entry *entry, void *arg)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  int step;

  int leave_all_stopped = * (int *) arg;

  int leave_pending;



  if (lwp->resume == NULL)

    return 0;



  if (lwp->resume->kind == resume_stop)

    {

      if (debug_threads)

        debug_printf ("resume_stop request for LWP %ld\n", lwpid_of (thread));



      if (!lwp->stopped)

        {

          if (debug_threads)

            debug_printf ("stopping LWP %ld\n", lwpid_of (thread));



          /* Stop the thread, and wait for the event asynchronously,

             through the event loop.  */

          send_sigstop (lwp);

        }

      else

        {

          if (debug_threads)

            debug_printf ("already stopped LWP %ld\n",

                          lwpid_of (thread));



          /* The LWP may have been stopped in an internal event that

             was not meant to be notified back to GDB (e.g., gdbserver

             breakpoint), so we should be reporting a stop event in

             this case too.  */



          /* If the thread already has a pending SIGSTOP, this is a

             no-op.  Otherwise, something later will presumably resume

             the thread and this will cause it to cancel any pending

             operation, due to last_resume_kind == resume_stop.  If

             the thread already has a pending status to report, we

             will still report it the next time we wait - see

             status_pending_p_callback.  */



          /* If we already have a pending signal to report, then

             there's no need to queue a SIGSTOP, as this means we're

             midway through moving the LWP out of the jumppad, and we

             will report the pending signal as soon as that is

             finished.  */

          if (lwp->pending_signals_to_report == NULL)

            send_sigstop (lwp);

        }



      /* For stop requests, we're done.  */

      lwp->resume = NULL;

      thread->last_status.kind = TARGET_WAITKIND_IGNORE;

      return 0;

    }



  /* If this thread which is about to be resumed has a pending status,

     then don't resume any threads - we can just report the pending

     status.  Make sure to queue any signals that would otherwise be

     sent.  In all-stop mode, we do this decision based on if *any*

     thread has a pending status.  If there's a thread that needs the

     step-over-breakpoint dance, then don't resume any other thread

     but that particular one.  */

  leave_pending = (lwp->status_pending_p || leave_all_stopped);



  if (!leave_pending)

    {

      if (debug_threads)

        debug_printf ("resuming LWP %ld\n", lwpid_of (thread));



      step = (lwp->resume->kind == resume_step);

      linux_resume_one_lwp (lwp, step, lwp->resume->sig, NULL);

    }

  else

    {

      if (debug_threads)

        debug_printf ("leaving LWP %ld stopped\n", lwpid_of (thread));



      /* If we have a new signal, enqueue the signal.  */

      if (lwp->resume->sig != 0)

        {

          struct pending_signals *p_sig;

          p_sig = xmalloc (sizeof (*p_sig));

          p_sig->prev = lwp->pending_signals;

          p_sig->signal = lwp->resume->sig;

          memset (&p_sig->info, 0, sizeof (siginfo_t));



          /* If this is the same signal we were previously stopped by,

             make sure to queue its siginfo.  We can ignore the return

             value of ptrace; if it fails, we'll skip

             PTRACE_SETSIGINFO.  */

          if (WIFSTOPPED (lwp->last_status)

              && WSTOPSIG (lwp->last_status) == lwp->resume->sig)

            ptrace (PTRACE_GETSIGINFO, lwpid_of (thread), (PTRACE_TYPE_ARG3) 0,

                    &p_sig->info);



          lwp->pending_signals = p_sig;

        }

    }



  thread->last_status.kind = TARGET_WAITKIND_IGNORE;

  lwp->resume = NULL;

  return 0;

}



static void

‌linux_resume (struct thread_resume *resume_info, size_t n)

{

  struct thread_resume_array array = { resume_info, n };

  struct thread_info *need_step_over = NULL;

  int any_pending;

  int leave_all_stopped;



  if (debug_threads)

    {

      debug_enter ();

      debug_printf ("linux_resume:\n");

    }



  find_inferior (&all_threads, linux_set_resume_request, &array);



  /* If there is a thread which would otherwise be resumed, which has

     a pending status, then don't resume any threads - we can just

     report the pending status.  Make sure to queue any signals that

     would otherwise be sent.  In non-stop mode, we'll apply this

     logic to each thread individually.  We consume all pending events

     before considering to start a step-over (in all-stop).  */

  any_pending = 0;

  if (!non_stop)

    find_inferior (&all_threads, resume_status_pending_p, &any_pending);



  /* If there is a thread which would otherwise be resumed, which is

     stopped at a breakpoint that needs stepping over, then don't

     resume any threads - have it step over the breakpoint with all

     other threads stopped, then resume all threads again.  Make sure

     to queue any signals that would otherwise be delivered or

     queued.  */

  if (!any_pending && supports_breakpoints ())

    need_step_over

      = (struct thread_info *) find_inferior (&all_threads,

                                              need_step_over_p, NULL);



  leave_all_stopped = (need_step_over != NULL || any_pending);



  if (debug_threads)

    {

      if (need_step_over != NULL)

        debug_printf ("Not resuming all, need step over\n");

      else if (any_pending)

        debug_printf ("Not resuming, all-stop and found "

                      "an LWP with pending status\n");

      else

        debug_printf ("Resuming, no pending status or step over needed\n");

    }



  /* Even if we're leaving threads stopped, queue all signals we'd

     otherwise deliver.  */

  find_inferior (&all_threads, linux_resume_one_thread, &leave_all_stopped);



  if (need_step_over)

    start_step_over (get_thread_lwp (need_step_over));



  if (debug_threads)

    {

      debug_printf ("linux_resume done\n");

      debug_exit ();

    }

}



/* This function is called once per thread.  We check the thread's

   last resume request, which will tell us whether to resume, step, or

   leave the thread stopped.  Any signal the client requested to be

   delivered has already been enqueued at this point.



   If any thread that GDB wants running is stopped at an internal

   breakpoint that needs stepping over, we start a step-over operation

   on that particular thread, and leave all others stopped.  */



static int

‌proceed_one_lwp (struct inferior_list_entry *entry, void *except)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);

  int step;



  if (lwp == except)

    return 0;



  if (debug_threads)

    debug_printf ("proceed_one_lwp: lwp %ld\n", lwpid_of (thread));



  if (!lwp->stopped)

    {

      if (debug_threads)

        debug_printf ("   LWP %ld already running\n", lwpid_of (thread));

      return 0;

    }



  if (thread->last_resume_kind == resume_stop

      && thread->last_status.kind != TARGET_WAITKIND_IGNORE)

    {

      if (debug_threads)

        debug_printf ("   client wants LWP to remain %ld stopped\n",

                      lwpid_of (thread));

      return 0;

    }



  if (lwp->status_pending_p)

    {

      if (debug_threads)

        debug_printf ("   LWP %ld has pending status, leaving stopped\n",

                      lwpid_of (thread));

      return 0;

    }



  gdb_assert (lwp->suspended >= 0);



  if (lwp->suspended)

    {

      if (debug_threads)

        debug_printf ("   LWP %ld is suspended\n", lwpid_of (thread));

      return 0;

    }



  if (thread->last_resume_kind == resume_stop

      && lwp->pending_signals_to_report == NULL

      && lwp->collecting_fast_tracepoint == 0)

    {

      /* We haven't reported this LWP as stopped yet (otherwise, the

         last_status.kind check above would catch it, and we wouldn't

         reach here.  This LWP may have been momentarily paused by a

         stop_all_lwps call while handling for example, another LWP's

         step-over.  In that case, the pending expected SIGSTOP signal

         that was queued at vCont;t handling time will have already

         been consumed by wait_for_sigstop, and so we need to requeue

         another one here.  Note that if the LWP already has a SIGSTOP

         pending, this is a no-op.  */



      if (debug_threads)

        debug_printf ("Client wants LWP %ld to stop. "

                      "Making sure it has a SIGSTOP pending\n",

                      lwpid_of (thread));



      send_sigstop (lwp);

    }



  step = thread->last_resume_kind == resume_step;

  linux_resume_one_lwp (lwp, step, 0, NULL);

  return 0;

}



static int

‌unsuspend_and_proceed_one_lwp (struct inferior_list_entry *entry, void *except)

{

  struct thread_info *thread = (struct thread_info *) entry;

  struct lwp_info *lwp = get_thread_lwp (thread);



  if (lwp == except)

    return 0;



  lwp->suspended--;

  gdb_assert (lwp->suspended >= 0);



  return proceed_one_lwp (entry, except);

}



/* When we finish a step-over, set threads running again.  If there's

   another thread that may need a step-over, now's the time to start

   it.  Eventually, we'll move all threads past their breakpoints.  */



static void

‌proceed_all_lwps (void)

{

  struct thread_info *need_step_over;



  /* If there is a thread which would otherwise be resumed, which is

     stopped at a breakpoint that needs stepping over, then don't

     resume any threads - have it step over the breakpoint with all

     other threads stopped, then resume all threads again.  */



  if (supports_breakpoints ())

    {

      need_step_over

        = (struct thread_info *) find_inferior (&all_threads,

                                                need_step_over_p, NULL);



      if (need_step_over != NULL)

        {

          if (debug_threads)

            debug_printf ("proceed_all_lwps: found "

                          "thread %ld needing a step-over\n",

                          lwpid_of (need_step_over));



          start_step_over (get_thread_lwp (need_step_over));

          return;

        }

    }



  if (debug_threads)

    debug_printf ("Proceeding, no step-over needed\n");



  find_inferior (&all_threads, proceed_one_lwp, NULL);

}



/* Stopped LWPs that the client wanted to be running, that don't have

   pending statuses, are set to run again, except for EXCEPT, if not

   NULL.  This undoes a stop_all_lwps call.  */



static void

‌unstop_all_lwps (int unsuspend, struct lwp_info *except)

{

  if (debug_threads)

    {

      debug_enter ();

      if (except)

        debug_printf ("unstopping all lwps, except=(LWP %ld)\n",

                      lwpid_of (get_lwp_thread (except)));

      else

        debug_printf ("unstopping all lwps\n");

    }



  if (unsuspend)

    find_inferior (&all_threads, unsuspend_and_proceed_one_lwp, except);

  else

    find_inferior (&all_threads, proceed_one_lwp, except);



  if (debug_threads)

    {

      debug_printf ("unstop_all_lwps done\n");

      debug_exit ();

    }

}





#ifdef HAVE_LINUX_REGSETS



‌#define use_linux_regsets 1



/* Returns true if REGSET has been disabled.  */



static int

‌regset_disabled (struct regsets_info *info, struct regset_info *regset)

{

  return (info->disabled_regsets != NULL

          && info->disabled_regsets[regset - info->regsets]);

}



/* Disable REGSET.  */



static void

‌disable_regset (struct regsets_info *info, struct regset_info *regset)

{

  int dr_offset;



  dr_offset = regset - info->regsets;

  if (info->disabled_regsets == NULL)

    info->disabled_regsets = xcalloc (1, info->num_regsets);

  info->disabled_regsets[dr_offset] = 1;

}



static int

‌regsets_fetch_inferior_registers (struct regsets_info *regsets_info,

                                  struct regcache *regcache)

{

  struct regset_info *regset;

  int saw_general_regs = 0;

  int pid;

  struct iovec iov;



  pid = lwpid_of (current_thread);

  for (regset = regsets_info->regsets; regset->size >= 0; regset++)

    {

      void *buf, *data;

      int nt_type, res;



      if (regset->size == 0 || regset_disabled (regsets_info, regset))

        continue;



      buf = xmalloc (regset->size);



      nt_type = regset->nt_type;

      if (nt_type)

        {

          iov.iov_base = buf;

          iov.iov_len = regset->size;

          data = (void *) &iov;

        }

      else

        data = buf;



#ifndef __sparc__

      res = ptrace (regset->get_request, pid,

                    (PTRACE_TYPE_ARG3) (long) nt_type, data);

#else

      res = ptrace (regset->get_request, pid, data, nt_type);

#endif

      if (res < 0)

        {

          if (errno == EIO)

            {

              /* If we get EIO on a regset, do not try it again for

                 this process mode.  */

              disable_regset (regsets_info, regset);

            }

          else if (errno == ENODATA)

            {

              /* ENODATA may be returned if the regset is currently

                 not "active".  This can happen in normal operation,

                 so suppress the warning in this case.  */

            }

          else

            {

              char s[256];

              sprintf (s, "ptrace(regsets_fetch_inferior_registers) PID=%d",

                       pid);

              perror (s);

            }

        }

      else

        {

          if (regset->type == GENERAL_REGS)

            saw_general_regs = 1;

          regset->store_function (regcache, buf);

        }

      free (buf);

    }

  if (saw_general_regs)

    return 0;

  else

    return 1;

}



static int

‌regsets_store_inferior_registers (struct regsets_info *regsets_info,

                                  struct regcache *regcache)

{

  struct regset_info *regset;

  int saw_general_regs = 0;

  int pid;

  struct iovec iov;



  pid = lwpid_of (current_thread);

  for (regset = regsets_info->regsets; regset->size >= 0; regset++)

    {

      void *buf, *data;

      int nt_type, res;



      if (regset->size == 0 || regset_disabled (regsets_info, regset)

          || regset->fill_function == NULL)

        continue;



      buf = xmalloc (regset->size);



      /* First fill the buffer with the current register set contents,

         in case there are any items in the kernel's regset that are

         not in gdbserver's regcache.  */



      nt_type = regset->nt_type;

      if (nt_type)

        {

          iov.iov_base = buf;

          iov.iov_len = regset->size;

          data = (void *) &iov;

        }

      else

        data = buf;



#ifndef __sparc__

      res = ptrace (regset->get_request, pid,

                    (PTRACE_TYPE_ARG3) (long) nt_type, data);

#else

      res = ptrace (regset->get_request, pid, data, nt_type);

#endif



      if (res == 0)

        {

          /* Then overlay our cached registers on that.  */

          regset->fill_function (regcache, buf);



          /* Only now do we write the register set.  */

#ifndef __sparc__

          res = ptrace (regset->set_request, pid,

                        (PTRACE_TYPE_ARG3) (long) nt_type, data);

#else

          res = ptrace (regset->set_request, pid, data, nt_type);

#endif

        }



      if (res < 0)

        {

          if (errno == EIO)

            {

              /* If we get EIO on a regset, do not try it again for

                 this process mode.  */

              disable_regset (regsets_info, regset);

            }

          else if (errno == ESRCH)

            {

              /* At this point, ESRCH should mean the process is

                 already gone, in which case we simply ignore attempts

                 to change its registers.  See also the related

                 comment in linux_resume_one_lwp.  */

              free (buf);

              return 0;

            }

          else

            {

              perror ("Warning: ptrace(regsets_store_inferior_registers)");

            }

        }

      else if (regset->type == GENERAL_REGS)

        saw_general_regs = 1;

      free (buf);

    }

  if (saw_general_regs)

    return 0;

  else

    return 1;

}



#else /* !HAVE_LINUX_REGSETS */



‌#define use_linux_regsets 0

‌#define regsets_fetch_inferior_registers(regsets_info, regcache) 1

‌#define regsets_store_inferior_registers(regsets_info, regcache) 1



#endif



/* Return 1 if register REGNO is supported by one of the regset ptrace

   calls or 0 if it has to be transferred individually.  */



static int

‌linux_register_in_regsets (const struct regs_info *regs_info, int regno)

{

  unsigned char mask = 1 << (regno % 8);

  size_t index = regno / 8;



  return (use_linux_regsets

          && (regs_info->regset_bitmap == NULL

              || (regs_info->regset_bitmap[index] & mask) != 0));

}



#ifdef HAVE_LINUX_USRREGS



int

‌register_addr (const struct usrregs_info *usrregs, int regnum)

{

  int addr;



  if (regnum < 0 || regnum >= usrregs->num_regs)

    error ("Invalid register number %d.", regnum);



  addr = usrregs->regmap[regnum];



  return addr;

}



/* Fetch one register.  */

static void

‌fetch_register (const struct usrregs_info *usrregs,

                struct regcache *regcache, int regno)

{

  CORE_ADDR regaddr;

  int i, size;

  char *buf;

  int pid;



  if (regno >= usrregs->num_regs)

    return;

  if ((*the_low_target.cannot_fetch_register) (regno))

    return;



  regaddr = register_addr (usrregs, regno);

  if (regaddr == -1)

    return;



  size = ((register_size (regcache->tdesc, regno)

           + sizeof (PTRACE_XFER_TYPE) - 1)

          & -sizeof (PTRACE_XFER_TYPE));

  buf = alloca (size);



  pid = lwpid_of (current_thread);

  for (i = 0; i < size; i += sizeof (PTRACE_XFER_TYPE))

    {

      errno = 0;

      *(PTRACE_XFER_TYPE *) (buf + i) =

        ptrace (PTRACE_PEEKUSER, pid,

                /* Coerce to a uintptr_t first to avoid potential gcc warning

                   of coercing an 8 byte integer to a 4 byte pointer.  */

                (PTRACE_TYPE_ARG3) (uintptr_t) regaddr, (PTRACE_TYPE_ARG4) 0);

      regaddr += sizeof (PTRACE_XFER_TYPE);

      if (errno != 0)

        error ("reading register %d: %s", regno, strerror (errno));

    }



  if (the_low_target.supply_ptrace_register)

    the_low_target.supply_ptrace_register (regcache, regno, buf);

  else

    supply_register (regcache, regno, buf);

}



/* Store one register.  */

static void

‌store_register (const struct usrregs_info *usrregs,

                struct regcache *regcache, int regno)

{

  CORE_ADDR regaddr;

  int i, size;

  char *buf;

  int pid;



  if (regno >= usrregs->num_regs)

    return;

  if ((*the_low_target.cannot_store_register) (regno))

    return;



  regaddr = register_addr (usrregs, regno);

  if (regaddr == -1)

    return;



  size = ((register_size (regcache->tdesc, regno)

           + sizeof (PTRACE_XFER_TYPE) - 1)

          & -sizeof (PTRACE_XFER_TYPE));

  buf = alloca (size);

  memset (buf, 0, size);



  if (the_low_target.collect_ptrace_register)

    the_low_target.collect_ptrace_register (regcache, regno, buf);

  else

    collect_register (regcache, regno, buf);



  pid = lwpid_of (current_thread);

  for (i = 0; i < size; i += sizeof (PTRACE_XFER_TYPE))

    {

      errno = 0;

      ptrace (PTRACE_POKEUSER, pid,

            /* Coerce to a uintptr_t first to avoid potential gcc warning

               about coercing an 8 byte integer to a 4 byte pointer.  */

              (PTRACE_TYPE_ARG3) (uintptr_t) regaddr,

              (PTRACE_TYPE_ARG4) *(PTRACE_XFER_TYPE *) (buf + i));

      if (errno != 0)

        {

          /* At this point, ESRCH should mean the process is

             already gone, in which case we simply ignore attempts

             to change its registers.  See also the related

             comment in linux_resume_one_lwp.  */

          if (errno == ESRCH)

            return;



          if ((*the_low_target.cannot_store_register) (regno) == 0)

            error ("writing register %d: %s", regno, strerror (errno));

        }

      regaddr += sizeof (PTRACE_XFER_TYPE);

    }

}



/* Fetch all registers, or just one, from the child process.

   If REGNO is -1, do this for all registers, skipping any that are

   assumed to have been retrieved by regsets_fetch_inferior_registers,

   unless ALL is non-zero.

   Otherwise, REGNO specifies which register (so we can save time).  */

static void

‌usr_fetch_inferior_registers (const struct regs_info *regs_info,

                              struct regcache *regcache, int regno, int all)

{

  struct usrregs_info *usr = regs_info->usrregs;



  if (regno == -1)

    {

      for (regno = 0; regno < usr->num_regs; regno++)

        if (all || !linux_register_in_regsets (regs_info, regno))

          fetch_register (usr, regcache, regno);

    }

  else

    fetch_register (usr, regcache, regno);

}



/* Store our register values back into the inferior.

   If REGNO is -1, do this for all registers, skipping any that are

   assumed to have been saved by regsets_store_inferior_registers,

   unless ALL is non-zero.

   Otherwise, REGNO specifies which register (so we can save time).  */

static void

‌usr_store_inferior_registers (const struct regs_info *regs_info,

                              struct regcache *regcache, int regno, int all)

{

  struct usrregs_info *usr = regs_info->usrregs;



  if (regno == -1)

    {

      for (regno = 0; regno < usr->num_regs; regno++)

        if (all || !linux_register_in_regsets (regs_info, regno))

          store_register (usr, regcache, regno);

    }

  else

    store_register (usr, regcache, regno);

}



#else /* !HAVE_LINUX_USRREGS */



‌#define usr_fetch_inferior_registers(regs_info, regcache, regno, all) do {} while (0)

‌#define usr_store_inferior_registers(regs_info, regcache, regno, all) do {} while (0)



#endif





void

‌linux_fetch_registers (struct regcache *regcache, int regno)

{

  int use_regsets;

  int all = 0;

  const struct regs_info *regs_info = (*the_low_target.regs_info) ();



  if (regno == -1)

    {

      if (the_low_target.fetch_register != NULL

          && regs_info->usrregs != NULL)

        for (regno = 0; regno < regs_info->usrregs->num_regs; regno++)

          (*the_low_target.fetch_register) (regcache, regno);



      all = regsets_fetch_inferior_registers (regs_info->regsets_info, regcache);

      if (regs_info->usrregs != NULL)

        usr_fetch_inferior_registers (regs_info, regcache, -1, all);

    }

  else

    {

      if (the_low_target.fetch_register != NULL

          && (*the_low_target.fetch_register) (regcache, regno))

        return;



      use_regsets = linux_register_in_regsets (regs_info, regno);

      if (use_regsets)

        all = regsets_fetch_inferior_registers (regs_info->regsets_info,

                                                regcache);

      if ((!use_regsets || all) && regs_info->usrregs != NULL)

        usr_fetch_inferior_registers (regs_info, regcache, regno, 1);

    }

}



void

‌linux_store_registers (struct regcache *regcache, int regno)

{

  int use_regsets;

  int all = 0;

  const struct regs_info *regs_info = (*the_low_target.regs_info) ();



  if (regno == -1)

    {

      all = regsets_store_inferior_registers (regs_info->regsets_info,

                                              regcache);

      if (regs_info->usrregs != NULL)

        usr_store_inferior_registers (regs_info, regcache, regno, all);

    }

  else

    {

      use_regsets = linux_register_in_regsets (regs_info, regno);

      if (use_regsets)

        all = regsets_store_inferior_registers (regs_info->regsets_info,

                                                regcache);

      if ((!use_regsets || all) && regs_info->usrregs != NULL)

        usr_store_inferior_registers (regs_info, regcache, regno, 1);

    }

}





/* Copy LEN bytes from inferior's memory starting at MEMADDR

   to debugger memory starting at MYADDR.  */



static int

‌linux_read_memory (CORE_ADDR memaddr, unsigned char *myaddr, int len)

{

  int pid = lwpid_of (current_thread);

  register PTRACE_XFER_TYPE *buffer;

  register CORE_ADDR addr;

  register int count;

  char filename[64];

  register int i;

  int ret;

  int fd;



  /* Try using /proc.  Don't bother for one word.  */

  if (len >= 3 * sizeof (long))

    {

      int bytes;



      /* We could keep this file open and cache it - possibly one per

         thread.  That requires some juggling, but is even faster.  */

      sprintf (filename, "/proc/%d/mem", pid);

      fd = open (filename, O_RDONLY | O_LARGEFILE);

      if (fd == -1)

        goto no_proc;



      /* If pread64 is available, use it.  It's faster if the kernel

         supports it (only one syscall), and it's 64-bit safe even on

         32-bit platforms (for instance, SPARC debugging a SPARC64

         application).  */

#ifdef HAVE_PREAD64

      bytes = pread64 (fd, myaddr, len, memaddr);

#else

      bytes = -1;

      if (lseek (fd, memaddr, SEEK_SET) != -1)

        bytes = read (fd, myaddr, len);

#endif



      close (fd);

      if (bytes == len)

        return 0;



      /* Some data was read, we'll try to get the rest with ptrace.  */

      if (bytes > 0)

        {

          memaddr += bytes;

          myaddr += bytes;

          len -= bytes;

        }

    }



 no_proc:

  /* Round starting address down to longword boundary.  */

  addr = memaddr & -(CORE_ADDR) sizeof (PTRACE_XFER_TYPE);

  /* Round ending address up; get number of longwords that makes.  */

  count = ((((memaddr + len) - addr) + sizeof (PTRACE_XFER_TYPE) - 1)

           / sizeof (PTRACE_XFER_TYPE));

  /* Allocate buffer of that many longwords.  */

  buffer = (PTRACE_XFER_TYPE *) alloca (count * sizeof (PTRACE_XFER_TYPE));



  /* Read all the longwords */

  errno = 0;

  for (i = 0; i < count; i++, addr += sizeof (PTRACE_XFER_TYPE))

    {

      /* Coerce the 3rd arg to a uintptr_t first to avoid potential gcc warning

         about coercing an 8 byte integer to a 4 byte pointer.  */

      buffer[i] = ptrace (PTRACE_PEEKTEXT, pid,

                          (PTRACE_TYPE_ARG3) (uintptr_t) addr,

                          (PTRACE_TYPE_ARG4) 0);

      if (errno)

        break;

    }

  ret = errno;



  /* Copy appropriate bytes out of the buffer.  */

  if (i > 0)

    {

      i *= sizeof (PTRACE_XFER_TYPE);

      i -= memaddr & (sizeof (PTRACE_XFER_TYPE) - 1);

      memcpy (myaddr,

              (char *) buffer + (memaddr & (sizeof (PTRACE_XFER_TYPE) - 1)),

              i < len ? i : len);

    }



  return ret;

}



/* Copy LEN bytes of data from debugger memory at MYADDR to inferior's

   memory at MEMADDR.  On failure (cannot write to the inferior)

   returns the value of errno.  Always succeeds if LEN is zero.  */



static int

‌linux_write_memory (CORE_ADDR memaddr, const unsigned char *myaddr, int len)

{

  register int i;

  /* Round starting address down to longword boundary.  */

  register CORE_ADDR addr = memaddr & -(CORE_ADDR) sizeof (PTRACE_XFER_TYPE);

  /* Round ending address up; get number of longwords that makes.  */

  register int count

    = (((memaddr + len) - addr) + sizeof (PTRACE_XFER_TYPE) - 1)

    / sizeof (PTRACE_XFER_TYPE);



  /* Allocate buffer of that many longwords.  */

  register PTRACE_XFER_TYPE *buffer = (PTRACE_XFER_TYPE *)

    alloca (count * sizeof (PTRACE_XFER_TYPE));



  int pid = lwpid_of (current_thread);



  if (len == 0)

    {

      /* Zero length write always succeeds.  */

      return 0;

    }



  if (debug_threads)

    {

      /* Dump up to four bytes.  */

      unsigned int val = * (unsigned int *) myaddr;

      if (len == 1)

        val = val & 0xff;

      else if (len == 2)

        val = val & 0xffff;

      else if (len == 3)

        val = val & 0xffffff;

      debug_printf ("Writing %0*x to 0x%08lx\n", 2 * ((len < 4) ? len : 4),

                    val, (long)memaddr);

    }



  /* Fill start and end extra bytes of buffer with existing memory data.  */



  errno = 0;

  /* Coerce the 3rd arg to a uintptr_t first to avoid potential gcc warning

     about coercing an 8 byte integer to a 4 byte pointer.  */

  buffer[0] = ptrace (PTRACE_PEEKTEXT, pid,

                      (PTRACE_TYPE_ARG3) (uintptr_t) addr,

                      (PTRACE_TYPE_ARG4) 0);

  if (errno)

    return errno;



  if (count > 1)

    {

      errno = 0;

      buffer[count - 1]

        = ptrace (PTRACE_PEEKTEXT, pid,

                  /* Coerce to a uintptr_t first to avoid potential gcc warning

                     about coercing an 8 byte integer to a 4 byte pointer.  */

                  (PTRACE_TYPE_ARG3) (uintptr_t) (addr + (count - 1)

                                                  * sizeof (PTRACE_XFER_TYPE)),

                  (PTRACE_TYPE_ARG4) 0);

      if (errno)

        return errno;

    }



  /* Copy data to be written over corresponding part of buffer.  */



  memcpy ((char *) buffer + (memaddr & (sizeof (PTRACE_XFER_TYPE) - 1)),

          myaddr, len);



  /* Write the entire buffer.  */



  for (i = 0; i < count; i++, addr += sizeof (PTRACE_XFER_TYPE))

    {

      errno = 0;

      ptrace (PTRACE_POKETEXT, pid,

              /* Coerce to a uintptr_t first to avoid potential gcc warning

                 about coercing an 8 byte integer to a 4 byte pointer.  */

              (PTRACE_TYPE_ARG3) (uintptr_t) addr,

              (PTRACE_TYPE_ARG4) buffer[i]);

      if (errno)

        return errno;

    }



  return 0;

}



static void

‌linux_look_up_symbols (void)

{

#ifdef USE_THREAD_DB

  struct process_info *proc = current_process ();



  if (proc->private->thread_db != NULL)

    return;



  /* If the kernel supports tracing clones, then we don't need to

     use the magic thread event breakpoint to learn about

     threads.  */

  thread_db_init (!linux_supports_traceclone ());

#endif

}



static void

‌linux_request_interrupt (void)

{

  extern unsigned long signal_pid;



  /* Send a SIGINT to the process group.  This acts just like the user

     typed a ^C on the controlling terminal.  */

  kill (-signal_pid, SIGINT);

}



/* Copy LEN bytes from inferior's auxiliary vector starting at OFFSET

   to debugger memory starting at MYADDR.  */



static int

‌linux_read_auxv (CORE_ADDR offset, unsigned char *myaddr, unsigned int len)

{

  char filename[PATH_MAX];

  int fd, n;

  int pid = lwpid_of (current_thread);



  xsnprintf (filename, sizeof filename, "/proc/%d/auxv", pid);



  fd = open (filename, O_RDONLY);

  if (fd < 0)

    return -1;



  if (offset != (CORE_ADDR) 0

      && lseek (fd, (off_t) offset, SEEK_SET) != (off_t) offset)

    n = -1;

  else

    n = read (fd, myaddr, len);



  close (fd);



  return n;

}



/* These breakpoint and watchpoint related wrapper functions simply

   pass on the function call if the target has registered a

   corresponding function.  */



static int

‌linux_supports_z_point_type (char z_type)

{

  return (the_low_target.supports_z_point_type != NULL

          && the_low_target.supports_z_point_type (z_type));

}



static int

‌linux_insert_point (enum raw_bkpt_type type, CORE_ADDR addr,

                    int size, struct raw_breakpoint *bp)

{

  if (the_low_target.insert_point != NULL)

    return the_low_target.insert_point (type, addr, size, bp);

  else

    /* Unsupported (see target.h).  */

    return 1;

}



static int

‌linux_remove_point (enum raw_bkpt_type type, CORE_ADDR addr,

                    int size, struct raw_breakpoint *bp)

{

  if (the_low_target.remove_point != NULL)

    return the_low_target.remove_point (type, addr, size, bp);

  else

    /* Unsupported (see target.h).  */

    return 1;

}



static int

‌linux_stopped_by_watchpoint (void)

{

  struct lwp_info *lwp = get_thread_lwp (current_thread);



  return lwp->stop_reason == LWP_STOPPED_BY_WATCHPOINT;

}



static CORE_ADDR

‌linux_stopped_data_address (void)

{

  struct lwp_info *lwp = get_thread_lwp (current_thread);



  return lwp->stopped_data_address;

}



#if defined(__UCLIBC__) && defined(HAS_NOMMU)              \

    && defined(PT_TEXT_ADDR) && defined(PT_DATA_ADDR) \

    && defined(PT_TEXT_END_ADDR)



/* This is only used for targets that define PT_TEXT_ADDR,

   PT_DATA_ADDR and PT_TEXT_END_ADDR.  If those are not defined, supposedly

   the target has different ways of acquiring this information, like

   loadmaps.  */



/* Under uClinux, programs are loaded at non-zero offsets, which we need

   to tell gdb about.  */



static int

‌linux_read_offsets (CORE_ADDR *text_p, CORE_ADDR *data_p)

{

  unsigned long text, text_end, data;

  int pid = lwpid_of (get_thread_lwp (current_thread));



  errno = 0;



  text = ptrace (PTRACE_PEEKUSER, pid, (PTRACE_TYPE_ARG3) PT_TEXT_ADDR,

                 (PTRACE_TYPE_ARG4) 0);

  text_end = ptrace (PTRACE_PEEKUSER, pid, (PTRACE_TYPE_ARG3) PT_TEXT_END_ADDR,

                     (PTRACE_TYPE_ARG4) 0);

  data = ptrace (PTRACE_PEEKUSER, pid, (PTRACE_TYPE_ARG3) PT_DATA_ADDR,

                 (PTRACE_TYPE_ARG4) 0);



  if (errno == 0)

    {

      /* Both text and data offsets produced at compile-time (and so

         used by gdb) are relative to the beginning of the program,

         with the data segment immediately following the text segment.

         However, the actual runtime layout in memory may put the data

         somewhere else, so when we send gdb a data base-address, we

         use the real data base address and subtract the compile-time

         data base-address from it (which is just the length of the

         text segment).  BSS immediately follows data in both

         cases.  */

      *text_p = text;

      *data_p = data - (text_end - text);



      return 1;

    }

 return 0;

}

#endif



static int

‌linux_qxfer_osdata (const char *annex,

                    unsigned char *readbuf, unsigned const char *writebuf,

                    CORE_ADDR offset, int len)

{

  return linux_common_xfer_osdata (annex, readbuf, offset, len);

}



/* Convert a native/host siginfo object, into/from the siginfo in the

   layout of the inferiors' architecture.  */



static void

‌siginfo_fixup (siginfo_t *siginfo, void *inf_siginfo, int direction)

{

  int done = 0;



  if (the_low_target.siginfo_fixup != NULL)

    done = the_low_target.siginfo_fixup (siginfo, inf_siginfo, direction);



  /* If there was no callback, or the callback didn't do anything,

     then just do a straight memcpy.  */

  if (!done)

    {

      if (direction == 1)

        memcpy (siginfo, inf_siginfo, sizeof (siginfo_t));

      else

        memcpy (inf_siginfo, siginfo, sizeof (siginfo_t));

    }

}



static int

‌linux_xfer_siginfo (const char *annex, unsigned char *readbuf,

                    unsigned const char *writebuf, CORE_ADDR offset, int len)

{

  int pid;

  siginfo_t siginfo;

  char inf_siginfo[sizeof (siginfo_t)];



  if (current_thread == NULL)

    return -1;



  pid = lwpid_of (current_thread);



  if (debug_threads)

    debug_printf ("%s siginfo for lwp %d.\n",

                  readbuf != NULL ? "Reading" : "Writing",

                  pid);



  if (offset >= sizeof (siginfo))

    return -1;



  if (ptrace (PTRACE_GETSIGINFO, pid, (PTRACE_TYPE_ARG3) 0, &siginfo) != 0)

    return -1;



  /* When GDBSERVER is built as a 64-bit application, ptrace writes into

     SIGINFO an object with 64-bit layout.  Since debugging a 32-bit

     inferior with a 64-bit GDBSERVER should look the same as debugging it

     with a 32-bit GDBSERVER, we need to convert it.  */

  siginfo_fixup (&siginfo, inf_siginfo, 0);



  if (offset + len > sizeof (siginfo))

    len = sizeof (siginfo) - offset;



  if (readbuf != NULL)

    memcpy (readbuf, inf_siginfo + offset, len);

  else

    {

      memcpy (inf_siginfo + offset, writebuf, len);



      /* Convert back to ptrace layout before flushing it out.  */

      siginfo_fixup (&siginfo, inf_siginfo, 1);



      if (ptrace (PTRACE_SETSIGINFO, pid, (PTRACE_TYPE_ARG3) 0, &siginfo) != 0)

        return -1;

    }



  return len;

}



/* SIGCHLD handler that serves two purposes: In non-stop/async mode,

   so we notice when children change state; as the handler for the

   sigsuspend in my_waitpid.  */



static void

‌sigchld_handler (int signo)

{

  int old_errno = errno;



  if (debug_threads)

    {

      do

        {

          /* fprintf is not async-signal-safe, so call write

             directly.  */

          if (write (2, "sigchld_handler\n",

                     sizeof ("sigchld_handler\n") - 1) < 0)

            break; /* just ignore */

        } while (0);

    }



  if (target_is_async_p ())

    async_file_mark (); /* trigger a linux_wait */



  errno = old_errno;

}



static int

‌linux_supports_non_stop (void)

{

  return 1;

}



static int

‌linux_async (int enable)

{

  int previous = target_is_async_p ();



  if (debug_threads)

    debug_printf ("linux_async (%d), previous=%d\n",

                  enable, previous);



  if (previous != enable)

    {

      sigset_t mask;

      sigemptyset (&mask);

      sigaddset (&mask, SIGCHLD);



      sigprocmask (SIG_BLOCK, &mask, NULL);



      if (enable)

        {

          if (pipe (linux_event_pipe) == -1)

            {

              linux_event_pipe[0] = -1;

              linux_event_pipe[1] = -1;

              sigprocmask (SIG_UNBLOCK, &mask, NULL);



              warning ("creating event pipe failed.");

              return previous;

            }



          fcntl (linux_event_pipe[0], F_SETFL, O_NONBLOCK);

          fcntl (linux_event_pipe[1], F_SETFL, O_NONBLOCK);



          /* Register the event loop handler.  */

          add_file_handler (linux_event_pipe[0],

                            handle_target_event, NULL);



          /* Always trigger a linux_wait.  */

          async_file_mark ();

        }

      else

        {

          delete_file_handler (linux_event_pipe[0]);



          close (linux_event_pipe[0]);

          close (linux_event_pipe[1]);

          linux_event_pipe[0] = -1;

          linux_event_pipe[1] = -1;

        }



      sigprocmask (SIG_UNBLOCK, &mask, NULL);

    }



  return previous;

}



static int

‌linux_start_non_stop (int nonstop)

{

  /* Register or unregister from event-loop accordingly.  */

  linux_async (nonstop);



  if (target_is_async_p () != (nonstop != 0))

    return -1;



  return 0;

}



static int

‌linux_supports_multi_process (void)

{

  return 1;

}



static int

‌linux_supports_disable_randomization (void)

{

#ifdef HAVE_PERSONALITY

  return 1;

#else

  return 0;

#endif

}



static int

‌linux_supports_agent (void)

{

  return 1;

}



static int

‌linux_supports_range_stepping (void)

{

  if (*the_low_target.supports_range_stepping == NULL)

    return 0;



  return (*the_low_target.supports_range_stepping) ();

}



/* Enumerate spufs IDs for process PID.  */

static int

‌spu_enumerate_spu_ids (long pid, unsigned char *buf, CORE_ADDR offset, int len)

{

  int pos = 0;

  int written = 0;

  char path[128];

  DIR *dir;

  struct dirent *entry;



  sprintf (path, "/proc/%ld/fd", pid);

  dir = opendir (path);

  if (!dir)

    return -1;



  rewinddir (dir);

  while ((entry = readdir (dir)) != NULL)

    {

      struct stat st;

      struct statfs stfs;

      int fd;



      fd = atoi (entry->d_name);

      if (!fd)

        continue;



      sprintf (path, "/proc/%ld/fd/%d", pid, fd);

      if (stat (path, &st) != 0)

        continue;

      if (!S_ISDIR (st.st_mode))

        continue;



      if (statfs (path, &stfs) != 0)

        continue;

      if (stfs.f_type != SPUFS_MAGIC)

        continue;



      if (pos >= offset && pos + 4 <= offset + len)

        {

          *(unsigned int *)(buf + pos - offset) = fd;

          written += 4;

        }

      pos += 4;

    }



  closedir (dir);

  return written;

}



/* Implements the to_xfer_partial interface for the TARGET_OBJECT_SPU

   object type, using the /proc file system.  */

static int

‌linux_qxfer_spu (const char *annex, unsigned char *readbuf,

                 unsigned const char *writebuf,

                 CORE_ADDR offset, int len)

{

  long pid = lwpid_of (current_thread);

  char buf[128];

  int fd = 0;

  int ret = 0;



  if (!writebuf && !readbuf)

    return -1;



  if (!*annex)

    {

      if (!readbuf)

        return -1;

      else

        return spu_enumerate_spu_ids (pid, readbuf, offset, len);

    }



  sprintf (buf, "/proc/%ld/fd/%s", pid, annex);

  fd = open (buf, writebuf? O_WRONLY : O_RDONLY);

  if (fd <= 0)

    return -1;



  if (offset != 0

      && lseek (fd, (off_t) offset, SEEK_SET) != (off_t) offset)

    {

      close (fd);

      return 0;

    }



  if (writebuf)

    ret = write (fd, writebuf, (size_t) len);

  else

    ret = read (fd, readbuf, (size_t) len);



  close (fd);

  return ret;

}



#if defined PT_GETDSBT || defined PTRACE_GETFDPIC

‌struct target_loadseg

{

  /* Core address to which the segment is mapped.  */

  Elf32_Addr addr;

  /* VMA recorded in the program header.  */

  Elf32_Addr p_vaddr;

  /* Size of this segment in memory.  */

  Elf32_Word p_memsz;

};



# if defined PT_GETDSBT

‌struct target_loadmap

{

  /* Protocol version number, must be zero.  */

  Elf32_Word version;

  /* Pointer to the DSBT table, its size, and the DSBT index.  */

  unsigned *dsbt_table;

  unsigned dsbt_size, dsbt_index;

  /* Number of segments in this map.  */

  Elf32_Word nsegs;

  /* The actual memory map.  */

  struct target_loadseg segs[/*nsegs*/];

};

‌#  define LINUX_LOADMAP                PT_GETDSBT

‌#  define LINUX_LOADMAP_EXEC        PTRACE_GETDSBT_EXEC

‌#  define LINUX_LOADMAP_INTERP        PTRACE_GETDSBT_INTERP

# else

‌struct target_loadmap

{

  /* Protocol version number, must be zero.  */

  Elf32_Half version;

  /* Number of segments in this map.  */

  Elf32_Half nsegs;

  /* The actual memory map.  */

  struct target_loadseg segs[/*nsegs*/];

};

‌#  define LINUX_LOADMAP                PTRACE_GETFDPIC

‌#  define LINUX_LOADMAP_EXEC        PTRACE_GETFDPIC_EXEC

‌#  define LINUX_LOADMAP_INTERP        PTRACE_GETFDPIC_INTERP

# endif



static int

‌linux_read_loadmap (const char *annex, CORE_ADDR offset,

                    unsigned char *myaddr, unsigned int len)

{

  int pid = lwpid_of (current_thread);

  int addr = -1;

  struct target_loadmap *data = NULL;

  unsigned int actual_length, copy_length;



  if (strcmp (annex, "exec") == 0)

    addr = (int) LINUX_LOADMAP_EXEC;

  else if (strcmp (annex, "interp") == 0)

    addr = (int) LINUX_LOADMAP_INTERP;

  else

    return -1;



  if (ptrace (LINUX_LOADMAP, pid, addr, &data) != 0)

    return -1;



  if (data == NULL)

    return -1;



  actual_length = sizeof (struct target_loadmap)

    + sizeof (struct target_loadseg) * data->nsegs;



  if (offset < 0 || offset > actual_length)

    return -1;



  copy_length = actual_length - offset < len ? actual_length - offset : len;

  memcpy (myaddr, (char *) data + offset, copy_length);

  return copy_length;

}

#else

‌# define linux_read_loadmap NULL

#endif /* defined PT_GETDSBT || defined PTRACE_GETFDPIC */



static void

‌linux_process_qsupported (const char *query)

{

  if (the_low_target.process_qsupported != NULL)

    the_low_target.process_qsupported (query);

}



static int

‌linux_supports_tracepoints (void)

{

  if (*the_low_target.supports_tracepoints == NULL)

    return 0;



  return (*the_low_target.supports_tracepoints) ();

}



static CORE_ADDR

‌linux_read_pc (struct regcache *regcache)

{

  if (the_low_target.get_pc == NULL)

    return 0;



  return (*the_low_target.get_pc) (regcache);

}



static void

‌linux_write_pc (struct regcache *regcache, CORE_ADDR pc)

{

  gdb_assert (the_low_target.set_pc != NULL);



  (*the_low_target.set_pc) (regcache, pc);

}



static int

‌linux_thread_stopped (struct thread_info *thread)

{

  return get_thread_lwp (thread)->stopped;

}



/* This exposes stop-all-threads functionality to other modules.  */



static void

‌linux_pause_all (int freeze)

{

  stop_all_lwps (freeze, NULL);

}



/* This exposes unstop-all-threads functionality to other gdbserver

   modules.  */



static void

‌linux_unpause_all (int unfreeze)

{

  unstop_all_lwps (unfreeze, NULL);

}



static int

‌linux_prepare_to_access_memory (void)

{

  /* Neither ptrace nor /proc/PID/mem allow accessing memory through a

     running LWP.  */

  if (non_stop)

    linux_pause_all (1);

  return 0;

}



static void

‌linux_done_accessing_memory (void)

{

  /* Neither ptrace nor /proc/PID/mem allow accessing memory through a

     running LWP.  */

  if (non_stop)

    linux_unpause_all (1);

}



static int

‌linux_install_fast_tracepoint_jump_pad (CORE_ADDR tpoint, CORE_ADDR tpaddr,

                                        CORE_ADDR collector,

                                        CORE_ADDR lockaddr,

                                        ULONGEST orig_size,

                                        CORE_ADDR *jump_entry,

                                        CORE_ADDR *trampoline,

                                        ULONGEST *trampoline_size,

                                        unsigned char *jjump_pad_insn,

                                        ULONGEST *jjump_pad_insn_size,

                                        CORE_ADDR *adjusted_insn_addr,

                                        CORE_ADDR *adjusted_insn_addr_end,

                                        char *err)

{

  return (*the_low_target.install_fast_tracepoint_jump_pad)

    (tpoint, tpaddr, collector, lockaddr, orig_size,

     jump_entry, trampoline, trampoline_size,

     jjump_pad_insn, jjump_pad_insn_size,

     adjusted_insn_addr, adjusted_insn_addr_end,

     err);

}



static struct emit_ops *

‌linux_emit_ops (void)

{

  if (the_low_target.emit_ops != NULL)

    return (*the_low_target.emit_ops) ();

  else

    return NULL;

}



static int

‌linux_get_min_fast_tracepoint_insn_len (void)

{

  return (*the_low_target.get_min_fast_tracepoint_insn_len) ();

}



/* Extract &phdr and num_phdr in the inferior.  Return 0 on success.  */



static int

‌get_phdr_phnum_from_proc_auxv (const int pid, const int is_elf64,

                               CORE_ADDR *phdr_memaddr, int *num_phdr)

{

  char filename[PATH_MAX];

  int fd;

  const int auxv_size = is_elf64

    ? sizeof (Elf64_auxv_t) : sizeof (Elf32_auxv_t);

  char buf[sizeof (Elf64_auxv_t)];  /* The larger of the two.  */



  xsnprintf (filename, sizeof filename, "/proc/%d/auxv", pid);



  fd = open (filename, O_RDONLY);

  if (fd < 0)

    return 1;



  *phdr_memaddr = 0;

  *num_phdr = 0;

  while (read (fd, buf, auxv_size) == auxv_size

         && (*phdr_memaddr == 0 || *num_phdr == 0))

    {

      if (is_elf64)

        {

          Elf64_auxv_t *const aux = (Elf64_auxv_t *) buf;



          switch (aux->a_type)

            {

            case AT_PHDR:

              *phdr_memaddr = aux->a_un.a_val;

              break;

            case AT_PHNUM:

              *num_phdr = aux->a_un.a_val;

              break;

            }

        }

      else

        {

          Elf32_auxv_t *const aux = (Elf32_auxv_t *) buf;



          switch (aux->a_type)

            {

            case AT_PHDR:

              *phdr_memaddr = aux->a_un.a_val;

              break;

            case AT_PHNUM:

              *num_phdr = aux->a_un.a_val;

              break;

            }

        }

    }



  close (fd);



  if (*phdr_memaddr == 0 || *num_phdr == 0)

    {

      warning ("Unexpected missing AT_PHDR and/or AT_PHNUM: "

               "phdr_memaddr = %ld, phdr_num = %d",

               (long) *phdr_memaddr, *num_phdr);

      return 2;

    }



  return 0;

}



/* Return &_DYNAMIC (via PT_DYNAMIC) in the inferior, or 0 if not present.  */



static CORE_ADDR

‌get_dynamic (const int pid, const int is_elf64)

{

  CORE_ADDR phdr_memaddr, relocation;

  int num_phdr, i;

  unsigned char *phdr_buf;

  const int phdr_size = is_elf64 ? sizeof (Elf64_Phdr) : sizeof (Elf32_Phdr);



  if (get_phdr_phnum_from_proc_auxv (pid, is_elf64, &phdr_memaddr, &num_phdr))

    return 0;



  gdb_assert (num_phdr < 100);  /* Basic sanity check.  */

  phdr_buf = alloca (num_phdr * phdr_size);



  if (linux_read_memory (phdr_memaddr, phdr_buf, num_phdr * phdr_size))

    return 0;



  /* Compute relocation: it is expected to be 0 for "regular" executables,

     non-zero for PIE ones.  */

  relocation = -1;

  for (i = 0; relocation == -1 && i < num_phdr; i++)

    if (is_elf64)

      {

        Elf64_Phdr *const p = (Elf64_Phdr *) (phdr_buf + i * phdr_size);



        if (p->p_type == PT_PHDR)

          relocation = phdr_memaddr - p->p_vaddr;

      }

    else

      {

        Elf32_Phdr *const p = (Elf32_Phdr *) (phdr_buf + i * phdr_size);



        if (p->p_type == PT_PHDR)

          relocation = phdr_memaddr - p->p_vaddr;

      }



  if (relocation == -1)

    {

      /* PT_PHDR is optional, but necessary for PIE in general.  Fortunately

         any real world executables, including PIE executables, have always

         PT_PHDR present.  PT_PHDR is not present in some shared libraries or

         in fpc (Free Pascal 2.4) binaries but neither of those have a need for

         or present DT_DEBUG anyway (fpc binaries are statically linked).



         Therefore if there exists DT_DEBUG there is always also PT_PHDR.



         GDB could find RELOCATION also from AT_ENTRY - e_entry.  */



      return 0;

    }



  for (i = 0; i < num_phdr; i++)

    {

      if (is_elf64)

        {

          Elf64_Phdr *const p = (Elf64_Phdr *) (phdr_buf + i * phdr_size);



          if (p->p_type == PT_DYNAMIC)

            return p->p_vaddr + relocation;

        }

      else

        {

          Elf32_Phdr *const p = (Elf32_Phdr *) (phdr_buf + i * phdr_size);



          if (p->p_type == PT_DYNAMIC)

            return p->p_vaddr + relocation;

        }

    }



  return 0;

}



/* Return &_r_debug in the inferior, or -1 if not present.  Return value

   can be 0 if the inferior does not yet have the library list initialized.

   We look for DT_MIPS_RLD_MAP first.  MIPS executables use this instead of

   DT_DEBUG, although they sometimes contain an unused DT_DEBUG entry too.  */



static CORE_ADDR

‌get_r_debug (const int pid, const int is_elf64)

{

  CORE_ADDR dynamic_memaddr;

  const int dyn_size = is_elf64 ? sizeof (Elf64_Dyn) : sizeof (Elf32_Dyn);

  unsigned char buf[sizeof (Elf64_Dyn)];  /* The larger of the two.  */

  CORE_ADDR map = -1;



  dynamic_memaddr = get_dynamic (pid, is_elf64);

  if (dynamic_memaddr == 0)

    return map;



  while (linux_read_memory (dynamic_memaddr, buf, dyn_size) == 0)

    {

      if (is_elf64)

        {

          Elf64_Dyn *const dyn = (Elf64_Dyn *) buf;

#ifdef DT_MIPS_RLD_MAP

          union

            {

              Elf64_Xword map;

              unsigned char buf[sizeof (Elf64_Xword)];

            }

          rld_map;



          if (dyn->d_tag == DT_MIPS_RLD_MAP)

            {

              if (linux_read_memory (dyn->d_un.d_val,

                                     rld_map.buf, sizeof (rld_map.buf)) == 0)

                return rld_map.map;

              else

                break;

            }

#endif        /* DT_MIPS_RLD_MAP */



          if (dyn->d_tag == DT_DEBUG && map == -1)

            map = dyn->d_un.d_val;



          if (dyn->d_tag == DT_NULL)

            break;

        }

      else

        {

          Elf32_Dyn *const dyn = (Elf32_Dyn *) buf;

#ifdef DT_MIPS_RLD_MAP

          union

            {

              Elf32_Word map;

              unsigned char buf[sizeof (Elf32_Word)];

            }

          rld_map;



          if (dyn->d_tag == DT_MIPS_RLD_MAP)

            {

              if (linux_read_memory (dyn->d_un.d_val,

                                     rld_map.buf, sizeof (rld_map.buf)) == 0)

                return rld_map.map;

              else

                break;

            }

#endif        /* DT_MIPS_RLD_MAP */



          if (dyn->d_tag == DT_DEBUG && map == -1)

            map = dyn->d_un.d_val;



          if (dyn->d_tag == DT_NULL)

            break;

        }



      dynamic_memaddr += dyn_size;

    }



  return map;

}



/* Read one pointer from MEMADDR in the inferior.  */



static int

‌read_one_ptr (CORE_ADDR memaddr, CORE_ADDR *ptr, int ptr_size)

{

  int ret;



  /* Go through a union so this works on either big or little endian

     hosts, when the inferior's pointer size is smaller than the size

     of CORE_ADDR.  It is assumed the inferior's endianness is the

     same of the superior's.  */

  union

  {

    CORE_ADDR core_addr;

    unsigned int ui;

    unsigned char uc;

  } addr;



  ret = linux_read_memory (memaddr, &addr.uc, ptr_size);

  if (ret == 0)

    {

      if (ptr_size == sizeof (CORE_ADDR))

        *ptr = addr.core_addr;

      else if (ptr_size == sizeof (unsigned int))

        *ptr = addr.ui;

      else

        gdb_assert_not_reached ("unhandled pointer size");

    }

  return ret;

}



‌struct link_map_offsets

  {

    /* Offset and size of r_debug.r_version.  */

    int r_version_offset;



    /* Offset and size of r_debug.r_map.  */

    int r_map_offset;



    /* Offset to l_addr field in struct link_map.  */

    int l_addr_offset;



    /* Offset to l_name field in struct link_map.  */

    int l_name_offset;



    /* Offset to l_ld field in struct link_map.  */

    int l_ld_offset;



    /* Offset to l_next field in struct link_map.  */

    int l_next_offset;



    /* Offset to l_prev field in struct link_map.  */

    int l_prev_offset;

  };



/* Construct qXfer:libraries-svr4:read reply.  */



static int

‌linux_qxfer_libraries_svr4 (const char *annex, unsigned char *readbuf,

                            unsigned const char *writebuf,

                            CORE_ADDR offset, int len)

{

  char *document;

  unsigned document_len;

  struct process_info_private *const priv = current_process ()->private;

  char filename[PATH_MAX];

  int pid, is_elf64;



  static const struct link_map_offsets lmo_32bit_offsets =

    {

      0,     /* r_version offset. */

      4,     /* r_debug.r_map offset.  */

      0,     /* l_addr offset in link_map.  */

      4,     /* l_name offset in link_map.  */

      8,     /* l_ld offset in link_map.  */

      12,    /* l_next offset in link_map.  */

      16     /* l_prev offset in link_map.  */

    };



  static const struct link_map_offsets lmo_64bit_offsets =

    {

      0,     /* r_version offset. */

      8,     /* r_debug.r_map offset.  */

      0,     /* l_addr offset in link_map.  */

      8,     /* l_name offset in link_map.  */

      16,    /* l_ld offset in link_map.  */

      24,    /* l_next offset in link_map.  */

      32     /* l_prev offset in link_map.  */

    };

  const struct link_map_offsets *lmo;

  unsigned int machine;

  int ptr_size;

  CORE_ADDR lm_addr = 0, lm_prev = 0;

  int allocated = 1024;

  char *p;

  CORE_ADDR l_name, l_addr, l_ld, l_next, l_prev;

  int header_done = 0;



  if (writebuf != NULL)

    return -2;

  if (readbuf == NULL)

    return -1;



  pid = lwpid_of (current_thread);

  xsnprintf (filename, sizeof filename, "/proc/%d/exe", pid);

  is_elf64 = elf_64_file_p (filename, &machine);

  lmo = is_elf64 ? &lmo_64bit_offsets : &lmo_32bit_offsets;

  ptr_size = is_elf64 ? 8 : 4;



  while (annex[0] != '\0')

    {

      const char *sep;

      CORE_ADDR *addrp;

      int len;



      sep = strchr (annex, '=');

      if (sep == NULL)

        break;



      len = sep - annex;

      if (len == 5 && strncmp (annex, "start", 5) == 0)

        addrp = &lm_addr;

      else if (len == 4 && strncmp (annex, "prev", 4) == 0)

        addrp = &lm_prev;

      else

        {

          annex = strchr (sep, ';');

          if (annex == NULL)

            break;

          annex++;

          continue;

        }



      annex = decode_address_to_semicolon (addrp, sep + 1);

    }



  if (lm_addr == 0)

    {

      int r_version = 0;



      if (priv->r_debug == 0)

        priv->r_debug = get_r_debug (pid, is_elf64);



      /* We failed to find DT_DEBUG.  Such situation will not change

         for this inferior - do not retry it.  Report it to GDB as

         E01, see for the reasons at the GDB solib-svr4.c side.  */

      if (priv->r_debug == (CORE_ADDR) -1)

        return -1;



      if (priv->r_debug != 0)

        {

          if (linux_read_memory (priv->r_debug + lmo->r_version_offset,

                                 (unsigned char *) &r_version,

                                 sizeof (r_version)) != 0

              || r_version != 1)

            {

              warning ("unexpected r_debug version %d", r_version);

            }

          else if (read_one_ptr (priv->r_debug + lmo->r_map_offset,

                                 &lm_addr, ptr_size) != 0)

            {

              warning ("unable to read r_map from 0x%lx",

                       (long) priv->r_debug + lmo->r_map_offset);

            }

        }

    }



  document = xmalloc (allocated);

  strcpy (document, "<library-list-svr4 version=\"1.0\"");

  p = document + strlen (document);



  while (lm_addr

         && read_one_ptr (lm_addr + lmo->l_name_offset,

                          &l_name, ptr_size) == 0

         && read_one_ptr (lm_addr + lmo->l_addr_offset,

                          &l_addr, ptr_size) == 0

         && read_one_ptr (lm_addr + lmo->l_ld_offset,

                          &l_ld, ptr_size) == 0

         && read_one_ptr (lm_addr + lmo->l_prev_offset,

                          &l_prev, ptr_size) == 0

         && read_one_ptr (lm_addr + lmo->l_next_offset,

                          &l_next, ptr_size) == 0)

    {

      unsigned char libname[PATH_MAX];



      if (lm_prev != l_prev)

        {

          warning ("Corrupted shared library list: 0x%lx != 0x%lx",

                   (long) lm_prev, (long) l_prev);

          break;

        }



      /* Ignore the first entry even if it has valid name as the first entry

         corresponds to the main executable.  The first entry should not be

         skipped if the dynamic loader was loaded late by a static executable

         (see solib-svr4.c parameter ignore_first).  But in such case the main

         executable does not have PT_DYNAMIC present and this function already

         exited above due to failed get_r_debug.  */

      if (lm_prev == 0)

        {

          sprintf (p, " main-lm=\"0x%lx\"", (unsigned long) lm_addr);

          p = p + strlen (p);

        }

      else

        {

          /* Not checking for error because reading may stop before

             we've got PATH_MAX worth of characters.  */

          libname[0] = '\0';

          linux_read_memory (l_name, libname, sizeof (libname) - 1);

          libname[sizeof (libname) - 1] = '\0';

          if (libname[0] != '\0')

            {

              /* 6x the size for xml_escape_text below.  */

              size_t len = 6 * strlen ((char *) libname);

              char *name;



              if (!header_done)

                {

                  /* Terminate `<library-list-svr4'.  */

                  *p++ = '>';

                  header_done = 1;

                }



              while (allocated < p - document + len + 200)

                {

                  /* Expand to guarantee sufficient storage.  */

                  uintptr_t document_len = p - document;



                  document = xrealloc (document, 2 * allocated);

                  allocated *= 2;

                  p = document + document_len;

                }



              name = xml_escape_text ((char *) libname);

              p += sprintf (p, "<library name=\"%s\" lm=\"0x%lx\" "

                            "l_addr=\"0x%lx\" l_ld=\"0x%lx\"/>",

                            name, (unsigned long) lm_addr,

                            (unsigned long) l_addr, (unsigned long) l_ld);

              free (name);

            }

        }



      lm_prev = lm_addr;

      lm_addr = l_next;

    }



  if (!header_done)

    {

      /* Empty list; terminate `<library-list-svr4'.  */

      strcpy (p, "/>");

    }

  else

    strcpy (p, "</library-list-svr4>");



  document_len = strlen (document);

  if (offset < document_len)

    document_len -= offset;

  else

    document_len = 0;

  if (len > document_len)

    len = document_len;



  memcpy (readbuf, document + offset, len);

  xfree (document);



  return len;

}



#ifdef HAVE_LINUX_BTRACE



/* See to_enable_btrace target method.  */



static struct btrace_target_info *

‌linux_low_enable_btrace (ptid_t ptid)

{

  struct btrace_target_info *tinfo;



  tinfo = linux_enable_btrace (ptid);



  if (tinfo != NULL)

    {

      struct thread_info *thread = find_thread_ptid (ptid);

      struct regcache *regcache = get_thread_regcache (thread, 0);



      tinfo->ptr_bits = register_size (regcache->tdesc, 0) * 8;

    }



  return tinfo;

}



/* See to_disable_btrace target method.  */



static int

‌linux_low_disable_btrace (struct btrace_target_info *tinfo)

{

  enum btrace_error err;



  err = linux_disable_btrace (tinfo);

  return (err == BTRACE_ERR_NONE ? 0 : -1);

}



/* See to_read_btrace target method.  */



static int

‌linux_low_read_btrace (struct btrace_target_info *tinfo, struct buffer *buffer,

                       int type)

{

  VEC (btrace_block_s) *btrace;

  struct btrace_block *block;

  enum btrace_error err;

  int i;



  btrace = NULL;

  err = linux_read_btrace (&btrace, tinfo, type);

  if (err != BTRACE_ERR_NONE)

    {

      if (err == BTRACE_ERR_OVERFLOW)

        buffer_grow_str0 (buffer, "E.Overflow.");

      else

        buffer_grow_str0 (buffer, "E.Generic Error.");



      return -1;

    }



  buffer_grow_str (buffer, "<!DOCTYPE btrace SYSTEM \"btrace.dtd\">\n");

  buffer_grow_str (buffer, "<btrace version=\"1.0\">\n");



  for (i = 0; VEC_iterate (btrace_block_s, btrace, i, block); i++)

    buffer_xml_printf (buffer, "<block begin=\"0x%s\" end=\"0x%s\"/>\n",

                       paddress (block->begin), paddress (block->end));



  buffer_grow_str0 (buffer, "</btrace>\n");



  VEC_free (btrace_block_s, btrace);



  return 0;

}

#endif /* HAVE_LINUX_BTRACE */



‌static struct target_ops linux_target_ops = {

  linux_create_inferior,

  linux_attach,

  linux_kill,

  linux_detach,

  linux_mourn,

  linux_join,

  linux_thread_alive,

  linux_resume,

  linux_wait,

  linux_fetch_registers,

  linux_store_registers,

  linux_prepare_to_access_memory,

  linux_done_accessing_memory,

  linux_read_memory,

  linux_write_memory,

  linux_look_up_symbols,

  linux_request_interrupt,

  linux_read_auxv,

  linux_supports_z_point_type,

  linux_insert_point,

  linux_remove_point,

  linux_stopped_by_watchpoint,

  linux_stopped_data_address,

#if defined(__UCLIBC__) && defined(HAS_NOMMU)              \

    && defined(PT_TEXT_ADDR) && defined(PT_DATA_ADDR) \

    && defined(PT_TEXT_END_ADDR)

  linux_read_offsets,

#else

  NULL,

#endif

#ifdef USE_THREAD_DB

  thread_db_get_tls_address,

#else

  NULL,

#endif

  linux_qxfer_spu,

  hostio_last_error_from_errno,

  linux_qxfer_osdata,

  linux_xfer_siginfo,

  linux_supports_non_stop,

  linux_async,

  linux_start_non_stop,

  linux_supports_multi_process,

#ifdef USE_THREAD_DB

  thread_db_handle_monitor_command,

#else

  NULL,

#endif

  linux_common_core_of_thread,

  linux_read_loadmap,

  linux_process_qsupported,

  linux_supports_tracepoints,

  linux_read_pc,

  linux_write_pc,

  linux_thread_stopped,

  NULL,

  linux_pause_all,

  linux_unpause_all,

  linux_stabilize_threads,

  linux_install_fast_tracepoint_jump_pad,

  linux_emit_ops,

  linux_supports_disable_randomization,

  linux_get_min_fast_tracepoint_insn_len,

  linux_qxfer_libraries_svr4,

  linux_supports_agent,

#ifdef HAVE_LINUX_BTRACE

  linux_supports_btrace,

  linux_low_enable_btrace,

  linux_low_disable_btrace,

  linux_low_read_btrace,

#else

  NULL,

  NULL,

  NULL,

  NULL,

#endif

  linux_supports_range_stepping,

};



static void

‌linux_init_signals ()

{

  /* FIXME drow/2002-06-09: As above, we should check with LinuxThreads

     to find what the cancel signal actually is.  */

#ifndef __ANDROID__ /* Bionic doesn't use SIGRTMIN the way glibc does.  */

  signal (__SIGRTMIN+1, SIG_IGN);

#endif

}



#ifdef HAVE_LINUX_REGSETS

void

‌initialize_regsets_info (struct regsets_info *info)

{

  for (info->num_regsets = 0;

       info->regsets[info->num_regsets].size >= 0;

       info->num_regsets++)

    ;

}

#endif



void

‌initialize_low (void)

{

  struct sigaction sigchld_action;

  memset (&sigchld_action, 0, sizeof (sigchld_action));

  set_target_ops (&linux_target_ops);

  set_breakpoint_data (the_low_target.breakpoint,

                       the_low_target.breakpoint_len);

  linux_init_signals ();

  linux_ptrace_init_warnings ();



  sigchld_action.sa_handler = sigchld_handler;

  sigemptyset (&sigchld_action.sa_mask);

  sigchld_action.sa_flags = SA_RESTART;

  sigaction (SIGCHLD, &sigchld_action, NULL);



  initialize_low_arch ();

}
gdb/gdbserver/linux-low.c - gdb

Global variables defined

Data types defined

Functions defined

Macros defined

Source code
