exp-drd/drd_main.c

/*
  This file is part of drd, a data race detector.

  Copyright (C) 2006-2008 Bart Van Assche
  bart.vanassche@gmail.com

  This program is free software; you can redistribute it and/or
  modify it under the terms of the GNU General Public License as
  published by the Free Software Foundation; either version 2 of the
  License, or (at your option) any later version.

  This program is distributed in the hope that it will be useful, but
  WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with this program; if not, write to the Free Software
  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  02111-1307, USA.

  The GNU General Public License is contained in the file COPYING.
*/


#include "drd_barrier.h"
#include "drd_clientreq.h"
#include "drd_clientobj.h"
#include "drd_cond.h"
#include "drd_error.h"
#include "drd_malloc_wrappers.h"
#include "drd_mutex.h"
#include "drd_rwlock.h"
#include "drd_segment.h"
#include "drd_semaphore.h"
#include "drd_suppression.h"
#include "drd_thread.h"
#include "drd_thread_bitmap.h"
#include "drd_track.h"
#include "drd_vc.h"
#include "libvex_guest_offsets.h"
#include "priv_drd_clientreq.h"
#include "pub_drd_bitmap.h"
#include "pub_tool_vki.h"         // Must be included before pub_tool_libcproc
#include "pub_tool_basics.h"
#include "pub_tool_debuginfo.h"   // VG_(describe_IP)()
#include "pub_tool_libcassert.h"  // tl_assert()
#include "pub_tool_libcbase.h"    // VG_(strcmp)
#include "pub_tool_libcprint.h"   // VG_(printf)
#include "pub_tool_libcproc.h"
#include "pub_tool_machine.h"
#include "pub_tool_mallocfree.h"  // VG_(malloc)(), VG_(free)()
#include "pub_tool_options.h"     // command line options
#include "pub_tool_replacemalloc.h"
#include "pub_tool_threadstate.h" // VG_(get_running_tid)()
#include "pub_tool_tooliface.h"


/* Include several source files here in order to allow the compiler to */
/* do more inlining.                                                   */
#include "drd_bitmap.c"
#include "drd_segment.c"
#include "drd_thread.c"
#include "drd_vc.c"



// Function declarations.

static void drd_start_client_code(const ThreadId tid, const ULong bbs_done);


// Local variables.

static Bool s_drd_check_stack_accesses = False;
static Bool s_drd_print_stats          = False;
static Bool s_drd_trace_fork_join      = False;
static Bool s_drd_var_info             = False;
static Bool s_show_stack_usage         = False;


//
// Implement the needs_command_line_options for drd.
//

static Bool drd_process_cmd_line_option(Char* arg)
{
  int exclusive_threshold_ms = -1;
  int segment_merging        = -1;
  int shared_threshold_ms    = -1;
  int show_confl_seg         = -1;
  int trace_barrier          = -1;
  int trace_clientobj        = -1;
  int trace_cond             = -1;
  int trace_csw              = -1;
  int trace_conflict_set     = -1;
  int trace_mutex            = -1;
  int trace_rwlock           = -1;
  int trace_segment          = -1;
  int trace_semaphore        = -1;
  int trace_suppression      = -1;
  Char* trace_address        = 0;

  VG_BOOL_CLO     (arg, "--check-stack-var",     s_drd_check_stack_accesses)
  else VG_BOOL_CLO(arg, "--drd-stats",           s_drd_print_stats)
  else VG_BOOL_CLO(arg, "--segment-merging",     segment_merging)
  else VG_BOOL_CLO(arg, "--show-confl-seg",      show_confl_seg)
  else VG_BOOL_CLO(arg, "--show-stack-usage",    s_show_stack_usage)
  else VG_BOOL_CLO(arg, "--trace-barrier",       trace_barrier)
  else VG_BOOL_CLO(arg, "--trace-clientobj",     trace_clientobj)
  else VG_BOOL_CLO(arg, "--trace-cond",          trace_cond)
  else VG_BOOL_CLO(arg, "--trace-csw",           trace_csw)
  else VG_BOOL_CLO(arg, "--trace-conflict-set",  trace_conflict_set)
  else VG_BOOL_CLO(arg, "--trace-fork-join",     s_drd_trace_fork_join)
  else VG_BOOL_CLO(arg, "--trace-mutex",         trace_mutex)
  else VG_BOOL_CLO(arg, "--trace-rwlock",        trace_rwlock)
  else VG_BOOL_CLO(arg, "--trace-segment",       trace_segment)
  else VG_BOOL_CLO(arg, "--trace-semaphore",     trace_semaphore)
  else VG_BOOL_CLO(arg, "--trace-suppr",         trace_suppression)
  else VG_BOOL_CLO(arg, "--var-info",            s_drd_var_info)
  else VG_NUM_CLO (arg, "--exclusive-threshold", exclusive_threshold_ms)
  else VG_NUM_CLO (arg, "--shared-threshold",    shared_threshold_ms)
  else VG_STR_CLO (arg, "--trace-addr",          trace_address)
  else
    return VG_(replacement_malloc_process_cmd_line_option)(arg);

  if (exclusive_threshold_ms != -1)
  {
    mutex_set_lock_threshold(exclusive_threshold_ms);
    rwlock_set_exclusive_threshold(exclusive_threshold_ms);
  }
  if (shared_threshold_ms != -1)
  {
    rwlock_set_shared_threshold(shared_threshold_ms);
  }
  if (segment_merging != -1)
    thread_set_segment_merging(segment_merging);
  if (show_confl_seg != -1)
    set_show_conflicting_segments(show_confl_seg);
  if (trace_address)
  {
    const Addr addr = VG_(strtoll16)(trace_address, 0);
    drd_start_tracing_address_range(addr, addr + 1);
  }
  if (trace_barrier != -1)
    barrier_set_trace(trace_barrier);
  if (trace_clientobj != -1)
    clientobj_set_trace(trace_clientobj);
  if (trace_cond != -1)
    cond_set_trace(trace_cond);
  if (trace_csw != -1)
    thread_trace_context_switches(trace_csw);
  if (trace_conflict_set != -1)
    thread_trace_conflict_set(trace_conflict_set);
  if (trace_mutex != -1)
    mutex_set_trace(trace_mutex);
  if (trace_rwlock != -1)
    rwlock_set_trace(trace_rwlock);
  if (trace_segment != -1)
    sg_set_trace(trace_segment);
  if (trace_semaphore != -1)
    semaphore_set_trace(trace_semaphore);
  if (trace_suppression != -1)
    suppression_set_trace(trace_suppression);

  return True;
}

static void drd_print_usage(void)
{
  VG_(printf)(
"    --check-stack-var=yes|no  Whether or not to report data races on\n"
"                              stack variables [no].\n"
"    --exclusive-threshold=<n> Print an error message if any mutex or\n"
"        writer lock is held longer than the specified time (in milliseconds).\n"
"    --segment-merging=yes|no  Controls segment merging [yes].\n"
"        Segment merging is an algorithm to limit memory usage of the\n"
"        data race detection algorithm. Disabling segment merging may\n"
"        improve the accuracy of the so-called 'other segments' displayed\n"
"        in race reports but can also trigger an out of memory error.\n"
"    --shared-threshold=<n>    Print an error message if a reader lock\n"
"        is held longer than the specified time (in milliseconds).\n"
"    --show-confl-seg=yes|no   Show conflicting segments in race reports [yes].\n"
"    --show-stack-usage=yes|no Print stack usage at thread exit time [no].\n"
"    --var-info=yes|no         Display the names of global, static and\n"
"        stack variables when a race is reported on such a variable. This\n"
"        information is by default not displayed since for big programs\n"
"        reading in all debug information at once may cause an out of\n"
"        memory error [no].\n"
"\n"
"  exp-drd options for monitoring process behavior:\n"
"    --trace-addr=<address>    Trace all load and store activity for the.\n"
"                              specified address [off].\n"
"    --trace-barrier=yes|no    Trace all barrier activity [no].\n"
"    --trace-cond=yes|no       Trace all condition variable activity [no].\n"
"    --trace-fork-join=yes|no  Trace all thread fork/join activity [no].\n"
"    --trace-mutex=yes|no      Trace all mutex activity [no].\n"
"    --trace-rwlock=yes|no     Trace all reader-writer lock activity[no].\n"
"    --trace-semaphore=yes|no  Trace all semaphore activity [no].\n"
              );
   VG_(replacement_malloc_print_usage)();
}

static void drd_print_debug_usage(void)
{  
  VG_(printf)(
"    --drd-stats=yes|no        Print statistics about DRD activity [no].\n"
"    --trace-clientobj=yes|no  Trace all client object activity [no].\n"
"    --trace-csw=yes|no        Trace all scheduler context switches [no].\n"
"    --trace-conflict-set=yes|no Trace all conflict set updates [no].\n"
"    --trace-segment=yes|no    Trace segment actions [no].\n"
"    --trace-suppr=yes|no      Trace all address suppression actions [no].\n"
              );
   VG_(replacement_malloc_print_debug_usage)();
}


//
// Implements the thread-related core callbacks.
//

static void drd_trace_mem_access(const Addr addr, const SizeT size,
                                 const BmAccessTypeT access_type)
{
  if (drd_is_any_traced(addr, addr + size))
  {
    char vc[80];
    vc_snprint(vc, sizeof(vc), thread_get_vc(thread_get_running_tid()));
    VG_(message)(Vg_UserMsg,
                 "%s 0x%lx size %ld (vg %d / drd %d / vc %s)",
                 access_type == eLoad
                 ? "load "
                 : access_type == eStore
                 ? "store"
                 : access_type == eStart
                 ? "start"
                 : access_type == eEnd
                 ? "end  "
                 : "????",
                 addr,
                 size,
                 VG_(get_running_tid)(),
                 thread_get_running_tid(),
                 vc);
    VG_(get_and_pp_StackTrace)(VG_(get_running_tid)(),
                               VG_(clo_backtrace_size));
    tl_assert(DrdThreadIdToVgThreadId(thread_get_running_tid())
              == VG_(get_running_tid)());
  }
}

static VG_REGPARM(2) void drd_trace_mem_load(const Addr addr, const SizeT size)
{
  return drd_trace_mem_access(addr, size, eLoad);
}

static VG_REGPARM(2) void drd_trace_mem_store(const Addr addr,const SizeT size)
{
  return drd_trace_mem_access(addr, size, eStore);
}

static void drd_report_race(const Addr addr, const SizeT size,
                            const BmAccessTypeT access_type)
{
  DataRaceErrInfo drei;

  drei.tid  = thread_get_running_tid();
  drei.addr = addr;
  drei.size = size;
  drei.access_type = access_type;
  VG_(maybe_record_error)(VG_(get_running_tid)(),
                          DataRaceErr,
                          VG_(get_IP)(VG_(get_running_tid)()),
                          "Conflicting accesses",
                          &drei);
}

static VG_REGPARM(2) void drd_trace_load(Addr addr, SizeT size)
{
#ifdef ENABLE_DRD_CONSISTENCY_CHECKS
  /* The assert below has been commented out because of performance reasons.*/
  tl_assert(thread_get_running_tid()
            == VgThreadIdToDrdThreadId(VG_(get_running_tid())));
#endif

  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_load_triggers_conflict(addr, addr + size)
      && ! drd_is_suppressed(addr, addr + size))
  {
    drd_report_race(addr, size, eLoad);
  }
}

static VG_REGPARM(1) void drd_trace_load_1(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_load_1_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 1))
  {
    drd_report_race(addr, 1, eLoad);
  }
}

static VG_REGPARM(1) void drd_trace_load_2(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_load_2_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 2))
  {
    drd_report_race(addr, 2, eLoad);
  }
}

static VG_REGPARM(1) void drd_trace_load_4(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_load_4_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 4))
  {
    drd_report_race(addr, 4, eLoad);
  }
}

static VG_REGPARM(1) void drd_trace_load_8(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_load_8_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 8))
  {
    drd_report_race(addr, 8, eLoad);
  }
}

static
VG_REGPARM(2) void drd_trace_store(Addr addr, SizeT size)
{
#ifdef ENABLE_DRD_CONSISTENCY_CHECKS
  /* The assert below has been commented out because of performance reasons.*/
  tl_assert(thread_get_running_tid()
            == VgThreadIdToDrdThreadId(VG_(get_running_tid())));
#endif

  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_store_triggers_conflict(addr, addr + size)
      && ! drd_is_suppressed(addr, addr + size))
  {
    drd_report_race(addr, size, eStore);
  }
}

static VG_REGPARM(1) void drd_trace_store_1(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_store_1_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 1))
  {
    drd_report_race(addr, 1, eStore);
  }
}

static VG_REGPARM(1) void drd_trace_store_2(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_store_2_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 2))
  {
    drd_report_race(addr, 2, eStore);
  }
}

static VG_REGPARM(1) void drd_trace_store_4(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_store_4_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 4))
  {
    drd_report_race(addr, 4, eStore);
  }
}

static VG_REGPARM(1) void drd_trace_store_8(Addr addr)
{
  if (running_thread_is_recording()
      && (s_drd_check_stack_accesses || ! thread_address_on_stack(addr))
      && bm_access_store_8_triggers_conflict(addr)
      && ! drd_is_suppressed(addr, addr + 8))
  {
    drd_report_race(addr, 8, eStore);
  }
}

static void drd_pre_mem_read(const CorePart part,
                             const ThreadId tid,
                             Char* const s,
                             const Addr a,
                             const SizeT size)
{
  if (size > 0)
  {
    drd_trace_load(a, size);
  }
}

static void drd_pre_mem_read_asciiz(const CorePart part,
                                    const ThreadId tid,
                                    Char* const s,
                                    const Addr a)
{
  const char* p = (void*)a;
  SizeT size = 0;

  /* Note: the expression '*p' reads client memory and may crash if the */
  /* client provided an invalid pointer !                               */
  while (*p)
  {
    p++;
    size++;
  }
  // To do: find out what a reasonable upper limit on 'size' is.
  tl_assert(size < 4096);
  if (size > 0)
  {
    drd_trace_load(a, size);
  }
}

static void drd_post_mem_write(const CorePart part,
                               const ThreadId tid,
                               const Addr a,
                               const SizeT size)
{
  thread_set_vg_running_tid(VG_(get_running_tid)());
  if (size > 0)
  {
    drd_trace_store(a, size);
  }
}

static __inline__
void drd_start_using_mem(const Addr a1, const SizeT len)
{
  tl_assert(a1 < a1 + len);

  if (UNLIKELY(drd_any_address_is_traced()))
  {
    drd_trace_mem_access(a1, len, eStart);
  }
}

static void drd_start_using_mem_w_ecu(const Addr a1,
                                      const SizeT len,
                                      UInt ec_uniq)
{
  drd_start_using_mem(a1, len);
}

static void drd_start_using_mem_w_tid(const Addr a1,
                                      const SizeT len,
                                      ThreadId tid)
{
  drd_start_using_mem(a1, len);
}

static __inline__
void drd_stop_using_mem(const Addr a1, const SizeT len,
                        const Bool is_stack_mem)
{
  const Addr a2 = a1 + len;

  tl_assert(a1 < a2);

  if (UNLIKELY(drd_any_address_is_traced()))
  {
    drd_trace_mem_access(a1, len, eEnd);
  }
  if (! is_stack_mem || s_drd_check_stack_accesses)
  {
    thread_stop_using_mem(a1, a2);
    clientobj_stop_using_mem(a1, a2);
    drd_suppression_stop_using_mem(a1, a2);
  }
}

static __inline__
void drd_stop_using_nonstack_mem(const Addr a1, const SizeT len)
{
  drd_stop_using_mem(a1, len, False);
}

/** Suppress data race reports on all addresses contained in .plt and
 *  .got.plt sections inside the address range [ a, a + len [. The data in
 *  these sections is modified by _dl_relocate_object() every time a function
 *  in a shared library is called for the first time. Since the first call
 *  to a function in a shared library can happen from a multithreaded context,
 *  such calls can cause conflicting accesses. See also Ulrich Drepper's
 *  paper "How to Write Shared Libraries" for more information about relocation
 *  (http://people.redhat.com/drepper/dsohowto.pdf).
 */
static void suppress_relocation_conflicts(const Addr a, const SizeT len)
{
  const DebugInfo* di;

#if 0
  VG_(printf)("Evaluating range @ 0x%lx size %ld\n", a, len);
#endif

  for (di = VG_(next_seginfo)(0); di; di = VG_(next_seginfo)(di))
  {
    Addr  avma;
    SizeT size;

    avma = VG_(seginfo_get_plt_avma)(di);
    size = VG_(seginfo_get_plt_size)(di);
    if (a <= avma && avma + size <= a + len)
    {
#if 0
      VG_(printf)("Suppressing .plt @ 0x%lx size %ld\n", avma, size);
#endif
      tl_assert(VG_(seginfo_sect_kind)(NULL, 0, avma) == Vg_SectPLT);
      drd_start_suppression(avma, avma + size, ".plt");
    }

    avma = VG_(seginfo_get_gotplt_avma)(di);
    size = VG_(seginfo_get_gotplt_size)(di);
    if (a <= avma && avma + size <= a + len)
    {
#if 0
      VG_(printf)("Suppressing .got.plt @ 0x%lx size %ld\n", avma, size);
#endif
      tl_assert(VG_(seginfo_sect_kind)(NULL, 0, avma) == Vg_SectGOTPLT);
      drd_start_suppression(avma, avma + size, ".gotplt");
    }
  }
}

static
void drd_start_using_mem_w_perms(const Addr a, const SizeT len,
                                 const Bool rr, const Bool ww, const Bool xx)
{
  thread_set_vg_running_tid(VG_(get_running_tid)());

  drd_start_using_mem(a, len);

  suppress_relocation_conflicts(a, len);
}

/* Called by the core when the stack of a thread grows, to indicate that */
/* the addresses in range [ a, a + len [ may now be used by the client.  */
/* Assumption: stacks grow downward.                                     */
static __inline__
void drd_start_using_mem_stack(const Addr a, const SizeT len)
{
  thread_set_stack_min(thread_get_running_tid(), a - VG_STACK_REDZONE_SZB);
  drd_start_using_mem(a - VG_STACK_REDZONE_SZB, 
                      len + VG_STACK_REDZONE_SZB);
}

/* Called by the core when the stack of a thread shrinks, to indicate that */
/* the addresses [ a, a + len [ are no longer accessible for the client.   */
/* Assumption: stacks grow downward.                                       */
static __inline__
void drd_stop_using_mem_stack(const Addr a, const SizeT len)
{
  thread_set_stack_min(thread_get_running_tid(),
                       a + len - VG_STACK_REDZONE_SZB);
  drd_stop_using_mem(a - VG_STACK_REDZONE_SZB, len + VG_STACK_REDZONE_SZB,
                     True);
}

static void drd_start_using_mem_stack_signal(
               const Addr a, const SizeT len,
               ThreadId tid_for_whom_the_signal_frame_is_being_constructed)
{
  thread_set_vg_running_tid(VG_(get_running_tid)());
  drd_start_using_mem(a, len);
}

static void drd_stop_using_mem_stack_signal(Addr a, SizeT len)
{
  drd_stop_using_mem(a, len, True);
}

static
void drd_pre_thread_create(const ThreadId creator, const ThreadId created)
{
  const DrdThreadId drd_creator = VgThreadIdToDrdThreadId(creator);
  tl_assert(created != VG_INVALID_THREADID);
  thread_pre_create(drd_creator, created);
  if (IsValidDrdThreadId(drd_creator))
  {
    thread_new_segment(drd_creator);
  }
  if (s_drd_trace_fork_join)
  {
    VG_(message)(Vg_DebugMsg,
                 "drd_pre_thread_create creator = %d/%d, created = %d",
                 creator, drd_creator, created);
  }
}

/* Called by Valgrind's core before any loads or stores are performed on */
/* the context of thread "created". At startup, this function is called  */
/* with arguments (0,1).                                                 */
static
void drd_post_thread_create(const ThreadId vg_created)
{
  DrdThreadId drd_created;

  tl_assert(vg_created != VG_INVALID_THREADID);

  drd_created = thread_post_create(vg_created);
  if (s_drd_trace_fork_join)
  {
    VG_(message)(Vg_DebugMsg,
                 "drd_post_thread_create created = %d/%d",
                 vg_created, drd_created);
  }
  if (! s_drd_check_stack_accesses)
  {
    drd_start_suppression(thread_get_stack_max(drd_created)
                          - thread_get_stack_size(drd_created),
                          thread_get_stack_max(drd_created),
                          "stack");
  }
}

/* Process VG_USERREQ__POST_THREAD_JOIN. This client request is invoked just */
/* after thread drd_joiner joined thread drd_joinee.                         */
void drd_post_thread_join(DrdThreadId drd_joiner, DrdThreadId drd_joinee)
{
  tl_assert(IsValidDrdThreadId(drd_joiner));
  tl_assert(IsValidDrdThreadId(drd_joinee));
  thread_new_segment(drd_joinee);
  thread_combine_vc(drd_joiner, drd_joinee);
  thread_new_segment(drd_joiner);

  if (s_drd_trace_fork_join)
  {
    const ThreadId joiner = DrdThreadIdToVgThreadId(drd_joiner);
    const ThreadId joinee = DrdThreadIdToVgThreadId(drd_joinee);
    const unsigned msg_size = 256;
    char* msg;

    msg = VG_(malloc)(msg_size);
    tl_assert(msg);
    VG_(snprintf)(msg, msg_size,
                  "drd_post_thread_join joiner = %d/%d, joinee = %d/%d",
                  joiner, drd_joiner, joinee, drd_joinee);
    if (joiner)
    {
      VG_(snprintf)(msg + VG_(strlen)(msg), msg_size - VG_(strlen)(msg),
                    ", new vc: ");
      vc_snprint(msg + VG_(strlen)(msg), msg_size - VG_(strlen)(msg),
                 thread_get_vc(drd_joiner));
    }
    VG_(message)(Vg_DebugMsg, msg);
    VG_(free)(msg);
  }

  if (! s_drd_check_stack_accesses)
  {
    drd_finish_suppression(thread_get_stack_max(drd_joinee)
                           - thread_get_stack_size(drd_joinee),
                           thread_get_stack_max(drd_joinee));
  }
  thread_delete(drd_joinee);
  mutex_thread_delete(drd_joinee);
  cond_thread_delete(drd_joinee);
  semaphore_thread_delete(drd_joinee);
  barrier_thread_delete(drd_joinee);
}


/* Called after a thread has performed its last memory access. */
static void drd_thread_finished(ThreadId vg_tid)
{
  DrdThreadId drd_tid;

  tl_assert(VG_(get_running_tid)() == vg_tid);

  drd_tid = VgThreadIdToDrdThreadId(vg_tid);
  if (s_drd_trace_fork_join)
  {
    VG_(message)(Vg_DebugMsg,
                 "drd_thread_finished tid = %d/%d%s",
                 vg_tid,
                 drd_tid,
                 thread_get_joinable(drd_tid)
                 ? ""
                 : " (which is a detached thread)");
  }
  if (s_show_stack_usage)
  {
    const SizeT stack_size = thread_get_stack_size(drd_tid);
    const SizeT used_stack
      = thread_get_stack_max(drd_tid) - thread_get_stack_min_min(drd_tid);
    VG_(message)(Vg_UserMsg,
                 "thread %d/%d%s finished and used %ld bytes out of %ld"
                 " on its stack. Margin: %ld bytes.",
                 vg_tid,
                 drd_tid,
                 thread_get_joinable(drd_tid)
                 ? ""
                 : " (which is a detached thread)",
                 used_stack,
                 stack_size,
                 stack_size - used_stack);

  }
  drd_stop_using_mem(thread_get_stack_min(drd_tid),
                     thread_get_stack_max(drd_tid)
                     - thread_get_stack_min(drd_tid),
                     True);
  thread_stop_recording(drd_tid);
  thread_finished(drd_tid);
}

void drd_pre_mutex_init(const Addr mutex, const MutexT mutex_type)
{
  mutex_init(mutex, mutex_type);
}

void drd_post_mutex_destroy(const Addr mutex, const MutexT mutex_type)
{
  mutex_post_destroy(mutex);
}

void drd_pre_mutex_lock(const Addr mutex, const MutexT mutex_type,
                        const Bool trylock)
{
  mutex_pre_lock(mutex, mutex_type, trylock);
}

void drd_post_mutex_lock(const Addr mutex, const Bool took_lock)
{
  mutex_post_lock(mutex, took_lock, False);
}

void drd_pre_mutex_unlock(const Addr mutex, const MutexT mutex_type)
{
  mutex_unlock(mutex, mutex_type);
}

void drd_pre_cond_init(Addr cond)
{
  cond_pre_init(cond);
}

void drd_post_cond_destroy(Addr cond)
{
  cond_post_destroy(cond);
}

void drd_semaphore_init(const Addr semaphore,
                        const Word pshared, const Word value)
{
  semaphore_init(semaphore, pshared, value);
}

void drd_semaphore_destroy(const Addr semaphore)
{
  semaphore_destroy(semaphore);
}

void drd_semaphore_pre_wait(const DrdThreadId tid, const Addr semaphore)
{
  semaphore_pre_wait(semaphore);
}

void drd_semaphore_post_wait(const DrdThreadId tid, const Addr semaphore,
                             const Bool waited)
{
  semaphore_post_wait(tid, semaphore, waited);
}

void drd_semaphore_pre_post(const DrdThreadId tid, const Addr semaphore)
{
  semaphore_pre_post(tid, semaphore);
}

void drd_semaphore_post_post(const DrdThreadId tid, const Addr semaphore,
                             const Bool waited)
{
  semaphore_post_post(tid, semaphore, waited);
}


void drd_barrier_init(const Addr barrier,
                      const BarrierT barrier_type, const Word count,
                      const Bool reinitialization)
{
  barrier_init(barrier, barrier_type, count, reinitialization);
}

void drd_barrier_destroy(const Addr barrier, const BarrierT barrier_type)
{
  barrier_destroy(barrier, barrier_type);
}

void drd_barrier_pre_wait(const DrdThreadId tid, const Addr barrier,
                          const BarrierT barrier_type)
{
  barrier_pre_wait(tid, barrier, barrier_type);
}

void drd_barrier_post_wait(const DrdThreadId tid, const Addr barrier,
                           const BarrierT barrier_type, const Bool waited)
{
  barrier_post_wait(tid, barrier, barrier_type, waited);
}


//
// Implementation of the tool interface.
//

static
void drd_post_clo_init(void)
{
#  if defined(VGP_x86_linux) || defined(VGP_amd64_linux)
  /* fine */
#  elif defined(VGP_ppc32_linux) || defined(VGP_ppc64_linux)
  VG_(printf)(
"\nWARNING: support for PowerPC-specific atomic instructions like lwarx and\n"
"stwcx is not yet complete. As a result, false positives will be reported on\n"
"code that uses these instructions. This will happen e.g. when printf() is\n"
"called from more than one thread.\n\n");
#  else
  VG_(printf)("\nWARNING: DRD has only been tested on x86-linux and amd64-linux.\n\n");
#  endif

  if (s_drd_var_info)
  {
    VG_(needs_var_info)();
  }
}

#if defined(VGA_x86)
#define STACK_POINTER_OFFSET OFFSET_x86_ESP
#elif defined(VGA_amd64)
#define STACK_POINTER_OFFSET OFFSET_amd64_RSP
#elif defined(VGA_ppc32)
#define STACK_POINTER_OFFSET ((OFFSET_ppc32_GPR0 + OFFSET_ppc32_GPR2) / 2)
#elif defined(VGA_ppc64)
#define STACK_POINTER_OFFSET ((OFFSET_ppc64_GPR0 + OFFSET_ppc64_GPR2) / 2)
#else
#error Unknown architecture.
#endif


/** Return true if and only if addr_expr matches the pattern (SP) or
 *  <offset>(SP).
 */
static Bool is_stack_access(IRSB* const bb, IRExpr* const addr_expr)
{
  Bool result = False;

  if (addr_expr->tag == Iex_RdTmp)
  {
    int i;
    for (i = 0; i < bb->stmts_size; i++)
    {
      if (bb->stmts[i]
          && bb->stmts[i]->tag == Ist_WrTmp
          && bb->stmts[i]->Ist.WrTmp.tmp == addr_expr->Iex.RdTmp.tmp)
      {
        IRExpr* e = bb->stmts[i]->Ist.WrTmp.data;
        if (e->tag == Iex_Get && e->Iex.Get.offset == STACK_POINTER_OFFSET)
        {
          result = True;
        }

        //ppIRExpr(e);
        //VG_(printf)(" (%s)\n", result ? "True" : "False");
        break;
      }
    }
  }
  return result;
}

static void instrument_load(IRSB* const bb,
                            IRExpr* const addr_expr,
                            const HWord size)
{
  IRExpr* size_expr;
  IRExpr** argv;
  IRDirty* di;

  if (UNLIKELY(drd_any_address_is_traced()))
  {
    addStmtToIRSB(bb,
		  IRStmt_Dirty(
		    unsafeIRDirty_0_N(/*regparms*/2,
				      "drd_trace_load",
				      VG_(fnptr_to_fnentry)
				      (drd_trace_mem_load),
				      mkIRExprVec_2(addr_expr,
						    mkIRExpr_HWord(size)))));
  }

  if (! s_drd_check_stack_accesses && is_stack_access(bb, addr_expr))
    return;

  switch (size)
  {
  case 1:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_load_1",
                           VG_(fnptr_to_fnentry)(drd_trace_load_1),
                           argv);
    break;
  case 2:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_load_2",
                           VG_(fnptr_to_fnentry)(drd_trace_load_2),
                           argv);
    break;
  case 4:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_load_4",
                           VG_(fnptr_to_fnentry)(drd_trace_load_4),
                           argv);
    break;
  case 8:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_load_8",
                           VG_(fnptr_to_fnentry)(drd_trace_load_8),
                           argv);
    break;
  default:
    size_expr = mkIRExpr_HWord(size);
    argv = mkIRExprVec_2(addr_expr, size_expr);
    di = unsafeIRDirty_0_N(/*regparms*/2,
                           "drd_trace_load",
                           VG_(fnptr_to_fnentry)(drd_trace_load),
                           argv);
    break;
  }
  addStmtToIRSB(bb, IRStmt_Dirty(di));
}

static void instrument_store(IRSB* const bb,
                             IRExpr* const addr_expr,
                             const HWord size)
{
  IRExpr* size_expr;
  IRExpr** argv;
  IRDirty* di;

  if (UNLIKELY(drd_any_address_is_traced()))
  {
    addStmtToIRSB(bb,
		  IRStmt_Dirty(
		    unsafeIRDirty_0_N(/*regparms*/2,
				      "drd_trace_store",
				      VG_(fnptr_to_fnentry)
				      (drd_trace_mem_store),
				      mkIRExprVec_2(addr_expr,
						    mkIRExpr_HWord(size)))));
  }

  if (! s_drd_check_stack_accesses && is_stack_access(bb, addr_expr))
    return;

  switch (size)
  {
  case 1:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_store_1",
                           VG_(fnptr_to_fnentry)(drd_trace_store_1),
                           argv);
    break;
  case 2:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_store_2",
                           VG_(fnptr_to_fnentry)(drd_trace_store_2),
                           argv);
    break;
  case 4:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_store_4",
                           VG_(fnptr_to_fnentry)(drd_trace_store_4),
                           argv);
    break;
  case 8:
    argv = mkIRExprVec_1(addr_expr);
    di = unsafeIRDirty_0_N(/*regparms*/1,
                           "drd_trace_store_8",
                           VG_(fnptr_to_fnentry)(drd_trace_store_8),
                           argv);
    break;
  default:
    size_expr = mkIRExpr_HWord(size);
    argv = mkIRExprVec_2(addr_expr, size_expr);
    di = unsafeIRDirty_0_N(/*regparms*/2,
                           "drd_trace_store",
                           VG_(fnptr_to_fnentry)(drd_trace_store),
                           argv);
    break;
  }
  addStmtToIRSB(bb, IRStmt_Dirty(di));
}

static
IRSB* drd_instrument(VgCallbackClosure* const closure,
                     IRSB* const bb_in,
                     VexGuestLayout* const layout,
                     VexGuestExtents* const vge, 
                     IRType const gWordTy,
                     IRType const hWordTy)
{
  IRDirty* di;
  Int      i;
  IRSB*    bb;
  IRExpr** argv;
  Bool     instrument = True;
  Bool     bus_locked = False;

  /* Set up BB */
  bb           = emptyIRSB();
  bb->tyenv    = deepCopyIRTypeEnv(bb_in->tyenv);
  bb->next     = deepCopyIRExpr(bb_in->next);
  bb->jumpkind = bb_in->jumpkind;

  for (i = 0; i < bb_in->stmts_used; i++)
  {
    IRStmt* const st = bb_in->stmts[i];
    tl_assert(st);
    if (st->tag == Ist_NoOp)
      continue;

    switch (st->tag)
    {
    case Ist_MBE:
      switch (st->Ist.MBE.event)
      {
      case Imbe_Fence:
        break; /* not interesting */
      case Imbe_BusLock:
        tl_assert(! bus_locked);
        bus_locked = True;
        break;
      case Imbe_BusUnlock:
        tl_assert(bus_locked);
        bus_locked = False;
        break;
      default:
        tl_assert(0);
      }
      addStmtToIRSB(bb, st);
      break;

    case Ist_Store:
      if (instrument && ! bus_locked)
      {
        instrument_store(bb,
                         st->Ist.Store.addr,
                         sizeofIRType(typeOfIRExpr(bb->tyenv,
                                                   st->Ist.Store.data)));
      }
      addStmtToIRSB(bb, st);
      break;

    case Ist_WrTmp:
      if (instrument)
      {
        const IRExpr* const data = st->Ist.WrTmp.data;
        if (data->tag == Iex_Load)
        {
          instrument_load(bb,
                          data->Iex.Load.addr,
                          sizeofIRType(data->Iex.Load.ty));
        }
      }
      addStmtToIRSB(bb, st);
      break;

    case Ist_Dirty:
      if (instrument)
      {
        IRDirty* d = st->Ist.Dirty.details;
        IREffect const mFx = d->mFx;
        switch (mFx) {
        case Ifx_None:
          break;
        case Ifx_Read:
        case Ifx_Write:
        case Ifx_Modify:
          tl_assert(d->mAddr);
          tl_assert(d->mSize > 0);
          argv = mkIRExprVec_2(d->mAddr, mkIRExpr_HWord(d->mSize));
          if (mFx == Ifx_Read || mFx == Ifx_Modify) {
            di = unsafeIRDirty_0_N(
                                   /*regparms*/2,
                                   "drd_trace_load",
                                   VG_(fnptr_to_fnentry)(drd_trace_load),
                                   argv);
            addStmtToIRSB(bb, IRStmt_Dirty(di));
          }
          if ((mFx == Ifx_Write || mFx == Ifx_Modify)
              && ! bus_locked)
          {
            di = unsafeIRDirty_0_N(
                                   /*regparms*/2,
                                   "drd_trace_store",
                                   VG_(fnptr_to_fnentry)(drd_trace_store),
                                   argv);
            addStmtToIRSB(bb, IRStmt_Dirty(di));
          }
          break;
        default:
          tl_assert(0);
        }
      }
      addStmtToIRSB(bb, st);
      break;

    default:
      addStmtToIRSB(bb, st);
      break;
    }
  }

  tl_assert(! bus_locked);

  return bb;
}

static void drd_start_client_code(const ThreadId tid, const ULong bbs_done)
{
  tl_assert(tid == VG_(get_running_tid)());
  thread_set_vg_running_tid(tid);
}

static
void drd_fini(Int exitcode)
{
  // thread_print_all();
  if (VG_(clo_verbosity) > 1 || s_drd_print_stats)
  {
    ULong update_conflict_set_count;
    ULong dsnsc;
    ULong dscvc;

    update_conflict_set_count
      = thread_get_update_conflict_set_count(&dsnsc, &dscvc);

    VG_(message)(Vg_UserMsg,
                 "   thread: %lld context switches"
                 " / %lld updates of the conflict set",
                 thread_get_context_switch_count(),
                 update_conflict_set_count);
    VG_(message)(Vg_UserMsg,
                 "           (%lld new sg + %lld combine vc + %lld csw).",
                 dsnsc,
                 dscvc,
                 update_conflict_set_count - dsnsc - dscvc);
    VG_(message)(Vg_UserMsg,
                 " segments: created %lld segments, max %lld alive,"
                 " %lld discard points.",
                 sg_get_created_segments_count(),
                 sg_get_max_alive_segments_count(),
                 thread_get_discard_ordered_segments_count());
    VG_(message)(Vg_UserMsg,
                 "           (%lld m, %lld rw, %lld s, %lld b)",
                 get_mutex_segment_creation_count(),
                 get_rwlock_segment_creation_count(),
                 get_semaphore_segment_creation_count(),
                 get_barrier_segment_creation_count());
    VG_(message)(Vg_UserMsg,
                 "  bitmaps: %lld level 1 / %lld level 2 bitmap refs",
                 bm_get_bitmap_creation_count(),
                 bm_get_bitmap2_node_creation_count());
    VG_(message)(Vg_UserMsg,
                 "           and %lld level 2 bitmaps were allocated.",
                 bm_get_bitmap2_creation_count());
    VG_(message)(Vg_UserMsg,
                 "    mutex: %lld non-recursive lock/unlock events.",
                 get_mutex_lock_count());
    drd_print_malloc_stats();
  }
}

static
void drd_pre_clo_init(void)
{
  // Basic tool stuff.

  VG_(details_name)            ("exp-drd");
  VG_(details_version)         (NULL);
  VG_(details_description)     ("a data race detector");
  VG_(details_copyright_author)("Copyright (C) 2006-2008, and GNU GPL'd,"
                                " by Bart Van Assche.");
  VG_(details_bug_reports_to)  (VG_BUGS_TO);

  VG_(basic_tool_funcs)        (drd_post_clo_init,
                                drd_instrument,
                                drd_fini);

  // Command line stuff.
  VG_(needs_command_line_options)(drd_process_cmd_line_option,
                                  drd_print_usage,
                                  drd_print_debug_usage);

  // Error handling.
  drd_register_error_handlers();

  // Core event tracking.
  VG_(track_pre_mem_read)         (drd_pre_mem_read);
  VG_(track_pre_mem_read_asciiz)  (drd_pre_mem_read_asciiz);
  VG_(track_post_mem_write)       (drd_post_mem_write);
  VG_(track_new_mem_brk)          (drd_start_using_mem_w_tid);
  VG_(track_new_mem_mmap)         (drd_start_using_mem_w_perms);
  VG_(track_new_mem_stack)        (drd_start_using_mem_stack);
  VG_(track_new_mem_stack_signal) (drd_start_using_mem_stack_signal);
  VG_(track_new_mem_startup)      (drd_start_using_mem_w_perms);
  VG_(track_die_mem_brk)          (drd_stop_using_nonstack_mem);
  VG_(track_die_mem_munmap)       (drd_stop_using_nonstack_mem);
  VG_(track_die_mem_stack)        (drd_stop_using_mem_stack);
  VG_(track_die_mem_stack_signal) (drd_stop_using_mem_stack_signal);
  VG_(track_start_client_code)    (drd_start_client_code);
  VG_(track_pre_thread_ll_create) (drd_pre_thread_create);
  VG_(track_pre_thread_first_insn)(drd_post_thread_create);
  VG_(track_pre_thread_ll_exit)   (drd_thread_finished);

  // Other stuff.
  drd_register_malloc_wrappers(drd_start_using_mem_w_ecu,
                               drd_stop_using_nonstack_mem);

  drd_clientreq_init();

  drd_suppression_init();

  clientobj_init();
}


VG_DETERMINE_INTERFACE_VERSION(drd_pre_clo_init)