Home | History | Annotate | Download | only in vm
      1 /*
      2  * Copyright (C) 2008 The Android Open Source Project
      3  *
      4  * Licensed under the Apache License, Version 2.0 (the "License");
      5  * you may not use this file except in compliance with the License.
      6  * You may obtain a copy of the License at
      7  *
      8  *      http://www.apache.org/licenses/LICENSE-2.0
      9  *
     10  * Unless required by applicable law or agreed to in writing, software
     11  * distributed under the License is distributed on an "AS IS" BASIS,
     12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     13  * See the License for the specific language governing permissions and
     14  * limitations under the License.
     15  */
     16 
     17 #include "Dalvik.h"
     18 
     19 #include <fcntl.h>
     20 #include <stdlib.h>
     21 #include <unistd.h>
     22 #include <pthread.h>
     23 #include <time.h>
     24 #include <errno.h>
     25 
     26 /*
     27  * Every Object has a monitor associated with it, but not every Object is
     28  * actually locked.  Even the ones that are locked do not need a
     29  * full-fledged monitor until a) there is actual contention or b) wait()
     30  * is called on the Object.
     31  *
     32  * For Dalvik, we have implemented a scheme similar to the one described
     33  * in Bacon et al.'s "Thin locks: featherweight synchronization for Java"
     34  * (ACM 1998).  Things are even easier for us, though, because we have
     35  * a full 32 bits to work with.
     36  *
     37  * The two states of an Object's lock are referred to as "thin" and
     38  * "fat".  A lock may transition from the "thin" state to the "fat"
     39  * state and this transition is referred to as inflation.  Once a lock
     40  * has been inflated it remains in the "fat" state indefinitely.
     41  *
     42  * The lock value itself is stored in Object.lock.  The LSB of the
     43  * lock encodes its state.  When cleared, the lock is in the "thin"
     44  * state and its bits are formatted as follows:
     45  *
     46  *    [31 ---- 19] [18 ---- 3] [2 ---- 1] [0]
     47  *     lock count   thread id  hash state  0
     48  *
     49  * When set, the lock is in the "fat" state and its bits are formatted
     50  * as follows:
     51  *
     52  *    [31 ---- 3] [2 ---- 1] [0]
     53  *      pointer   hash state  1
     54  *
     55  * For an in-depth description of the mechanics of thin-vs-fat locking,
     56  * read the paper referred to above.
     57  */
     58 
     59 /*
     60  * Monitors provide:
     61  *  - mutually exclusive access to resources
     62  *  - a way for multiple threads to wait for notification
     63  *
     64  * In effect, they fill the role of both mutexes and condition variables.
     65  *
     66  * Only one thread can own the monitor at any time.  There may be several
     67  * threads waiting on it (the wait call unlocks it).  One or more waiting
     68  * threads may be getting interrupted or notified at any given time.
     69  *
     70  * TODO: the various members of monitor are not SMP-safe.
     71  */
     72 struct Monitor {
     73     Thread*     owner;          /* which thread currently owns the lock? */
     74     int         lockCount;      /* owner's recursive lock depth */
     75     Object*     obj;            /* what object are we part of [debug only] */
     76 
     77     Thread*     waitSet;	/* threads currently waiting on this monitor */
     78 
     79     pthread_mutex_t lock;
     80 
     81     Monitor*    next;
     82 
     83     /*
     84      * Who last acquired this monitor, when lock sampling is enabled.
     85      * Even when enabled, ownerMethod may be NULL.
     86      */
     87     const Method* ownerMethod;
     88     u4 ownerPc;
     89 };
     90 
     91 
     92 /*
     93  * Create and initialize a monitor.
     94  */
     95 Monitor* dvmCreateMonitor(Object* obj)
     96 {
     97     Monitor* mon;
     98 
     99     mon = (Monitor*) calloc(1, sizeof(Monitor));
    100     if (mon == NULL) {
    101         ALOGE("Unable to allocate monitor");
    102         dvmAbort();
    103     }
    104     if (((u4)mon & 7) != 0) {
    105         ALOGE("Misaligned monitor: %p", mon);
    106         dvmAbort();
    107     }
    108     mon->obj = obj;
    109     dvmInitMutex(&mon->lock);
    110 
    111     /* replace the head of the list with the new monitor */
    112     do {
    113         mon->next = gDvm.monitorList;
    114     } while (android_atomic_release_cas((int32_t)mon->next, (int32_t)mon,
    115             (int32_t*)(void*)&gDvm.monitorList) != 0);
    116 
    117     return mon;
    118 }
    119 
    120 /*
    121  * Free the monitor list.  Only used when shutting the VM down.
    122  */
    123 void dvmFreeMonitorList()
    124 {
    125     Monitor* mon;
    126     Monitor* nextMon;
    127 
    128     mon = gDvm.monitorList;
    129     while (mon != NULL) {
    130         nextMon = mon->next;
    131         free(mon);
    132         mon = nextMon;
    133     }
    134 }
    135 
    136 /*
    137  * Get the object that a monitor is part of.
    138  */
    139 Object* dvmGetMonitorObject(Monitor* mon)
    140 {
    141     if (mon == NULL)
    142         return NULL;
    143     else
    144         return mon->obj;
    145 }
    146 
    147 /*
    148  * Returns the thread id of the thread owning the given lock.
    149  */
    150 static u4 lockOwner(Object* obj)
    151 {
    152     Thread *owner;
    153     u4 lock;
    154 
    155     assert(obj != NULL);
    156     /*
    157      * Since we're reading the lock value multiple times, latch it so
    158      * that it doesn't change out from under us if we get preempted.
    159      */
    160     lock = obj->lock;
    161     if (LW_SHAPE(lock) == LW_SHAPE_THIN) {
    162         return LW_LOCK_OWNER(lock);
    163     } else {
    164         owner = LW_MONITOR(lock)->owner;
    165         return owner ? owner->threadId : 0;
    166     }
    167 }
    168 
    169 /*
    170  * Get the thread that holds the lock on the specified object.  The
    171  * object may be unlocked, thin-locked, or fat-locked.
    172  *
    173  * The caller must lock the thread list before calling here.
    174  */
    175 Thread* dvmGetObjectLockHolder(Object* obj)
    176 {
    177     u4 threadId = lockOwner(obj);
    178 
    179     if (threadId == 0)
    180         return NULL;
    181     return dvmGetThreadByThreadId(threadId);
    182 }
    183 
    184 /*
    185  * Checks whether the given thread holds the given
    186  * objects's lock.
    187  */
    188 bool dvmHoldsLock(Thread* thread, Object* obj)
    189 {
    190     if (thread == NULL || obj == NULL) {
    191         return false;
    192     } else {
    193         return thread->threadId == lockOwner(obj);
    194     }
    195 }
    196 
    197 /*
    198  * Free the monitor associated with an object and make the object's lock
    199  * thin again.  This is called during garbage collection.
    200  */
    201 static void freeMonitor(Monitor *mon)
    202 {
    203     assert(mon != NULL);
    204     assert(mon->obj != NULL);
    205     assert(LW_SHAPE(mon->obj->lock) == LW_SHAPE_FAT);
    206 
    207     /* This lock is associated with an object
    208      * that's being swept.  The only possible way
    209      * anyone could be holding this lock would be
    210      * if some JNI code locked but didn't unlock
    211      * the object, in which case we've got some bad
    212      * native code somewhere.
    213      */
    214     assert(pthread_mutex_trylock(&mon->lock) == 0);
    215     assert(pthread_mutex_unlock(&mon->lock) == 0);
    216     dvmDestroyMutex(&mon->lock);
    217     free(mon);
    218 }
    219 
    220 /*
    221  * Frees monitor objects belonging to unmarked objects.
    222  */
    223 void dvmSweepMonitorList(Monitor** mon, int (*isUnmarkedObject)(void*))
    224 {
    225     Monitor handle;
    226     Monitor *prev, *curr;
    227     Object *obj;
    228 
    229     assert(mon != NULL);
    230     assert(isUnmarkedObject != NULL);
    231     prev = &handle;
    232     prev->next = curr = *mon;
    233     while (curr != NULL) {
    234         obj = curr->obj;
    235         if (obj != NULL && (*isUnmarkedObject)(obj) != 0) {
    236             prev->next = curr->next;
    237             freeMonitor(curr);
    238             curr = prev->next;
    239         } else {
    240             prev = curr;
    241             curr = curr->next;
    242         }
    243     }
    244     *mon = handle.next;
    245 }
    246 
    247 static char *logWriteInt(char *dst, int value)
    248 {
    249     *dst++ = EVENT_TYPE_INT;
    250     set4LE((u1 *)dst, value);
    251     return dst + 4;
    252 }
    253 
    254 static char *logWriteString(char *dst, const char *value, size_t len)
    255 {
    256     *dst++ = EVENT_TYPE_STRING;
    257     len = len < 32 ? len : 32;
    258     set4LE((u1 *)dst, len);
    259     dst += 4;
    260     memcpy(dst, value, len);
    261     return dst + len;
    262 }
    263 
    264 #define EVENT_LOG_TAG_dvm_lock_sample 20003
    265 
    266 static void logContentionEvent(Thread *self, u4 waitMs, u4 samplePercent,
    267                                const char *ownerFileName, u4 ownerLineNumber)
    268 {
    269     const StackSaveArea *saveArea;
    270     const Method *meth;
    271     u4 relativePc;
    272     char eventBuffer[174];
    273     const char *fileName;
    274     char procName[33];
    275     char *cp;
    276     size_t len;
    277     int fd;
    278 
    279     saveArea = SAVEAREA_FROM_FP(self->interpSave.curFrame);
    280     meth = saveArea->method;
    281     cp = eventBuffer;
    282 
    283     /* Emit the event list length, 1 byte. */
    284     *cp++ = 9;
    285 
    286     /* Emit the process name, <= 37 bytes. */
    287     fd = open("/proc/self/cmdline", O_RDONLY);
    288     memset(procName, 0, sizeof(procName));
    289     read(fd, procName, sizeof(procName) - 1);
    290     close(fd);
    291     len = strlen(procName);
    292     cp = logWriteString(cp, procName, len);
    293 
    294     /* Emit the sensitive thread ("main thread") status, 5 bytes. */
    295     bool isSensitive = false;
    296     if (gDvm.isSensitiveThreadHook != NULL) {
    297         isSensitive = gDvm.isSensitiveThreadHook();
    298     }
    299     cp = logWriteInt(cp, isSensitive);
    300 
    301     /* Emit self thread name string, <= 37 bytes. */
    302     std::string selfName = dvmGetThreadName(self);
    303     cp = logWriteString(cp, selfName.c_str(), selfName.size());
    304 
    305     /* Emit the wait time, 5 bytes. */
    306     cp = logWriteInt(cp, waitMs);
    307 
    308     /* Emit the source code file name, <= 37 bytes. */
    309     fileName = dvmGetMethodSourceFile(meth);
    310     if (fileName == NULL) fileName = "";
    311     cp = logWriteString(cp, fileName, strlen(fileName));
    312 
    313     /* Emit the source code line number, 5 bytes. */
    314     relativePc = saveArea->xtra.currentPc - saveArea->method->insns;
    315     cp = logWriteInt(cp, dvmLineNumFromPC(meth, relativePc));
    316 
    317     /* Emit the lock owner source code file name, <= 37 bytes. */
    318     if (ownerFileName == NULL) {
    319         ownerFileName = "";
    320     } else if (strcmp(fileName, ownerFileName) == 0) {
    321         /* Common case, so save on log space. */
    322         ownerFileName = "-";
    323     }
    324     cp = logWriteString(cp, ownerFileName, strlen(ownerFileName));
    325 
    326     /* Emit the source code line number, 5 bytes. */
    327     cp = logWriteInt(cp, ownerLineNumber);
    328 
    329     /* Emit the sample percentage, 5 bytes. */
    330     cp = logWriteInt(cp, samplePercent);
    331 
    332     assert((size_t)(cp - eventBuffer) <= sizeof(eventBuffer));
    333     android_btWriteLog(EVENT_LOG_TAG_dvm_lock_sample,
    334                        EVENT_TYPE_LIST,
    335                        eventBuffer,
    336                        (size_t)(cp - eventBuffer));
    337 }
    338 
    339 /*
    340  * Lock a monitor.
    341  */
    342 static void lockMonitor(Thread* self, Monitor* mon)
    343 {
    344     ThreadStatus oldStatus;
    345     u4 waitThreshold, samplePercent;
    346     u8 waitStart, waitEnd, waitMs;
    347 
    348     if (mon->owner == self) {
    349         mon->lockCount++;
    350         return;
    351     }
    352     if (dvmTryLockMutex(&mon->lock) != 0) {
    353         oldStatus = dvmChangeStatus(self, THREAD_MONITOR);
    354         waitThreshold = gDvm.lockProfThreshold;
    355         if (waitThreshold) {
    356             waitStart = dvmGetRelativeTimeUsec();
    357         }
    358 
    359         const Method* currentOwnerMethod = mon->ownerMethod;
    360         u4 currentOwnerPc = mon->ownerPc;
    361 
    362         dvmLockMutex(&mon->lock);
    363         if (waitThreshold) {
    364             waitEnd = dvmGetRelativeTimeUsec();
    365         }
    366         dvmChangeStatus(self, oldStatus);
    367         if (waitThreshold) {
    368             waitMs = (waitEnd - waitStart) / 1000;
    369             if (waitMs >= waitThreshold) {
    370                 samplePercent = 100;
    371             } else {
    372                 samplePercent = 100 * waitMs / waitThreshold;
    373             }
    374             if (samplePercent != 0 && ((u4)rand() % 100 < samplePercent)) {
    375                 const char* currentOwnerFileName = "no_method";
    376                 u4 currentOwnerLineNumber = 0;
    377                 if (currentOwnerMethod != NULL) {
    378                     currentOwnerFileName = dvmGetMethodSourceFile(currentOwnerMethod);
    379                     if (currentOwnerFileName == NULL) {
    380                         currentOwnerFileName = "no_method_file";
    381                     }
    382                     currentOwnerLineNumber = dvmLineNumFromPC(currentOwnerMethod, currentOwnerPc);
    383                 }
    384                 logContentionEvent(self, waitMs, samplePercent,
    385                                    currentOwnerFileName, currentOwnerLineNumber);
    386             }
    387         }
    388     }
    389     mon->owner = self;
    390     assert(mon->lockCount == 0);
    391 
    392     // When debugging, save the current monitor holder for future
    393     // acquisition failures to use in sampled logging.
    394     if (gDvm.lockProfThreshold > 0) {
    395         mon->ownerMethod = NULL;
    396         mon->ownerPc = 0;
    397         if (self->interpSave.curFrame == NULL) {
    398             return;
    399         }
    400         const StackSaveArea* saveArea = SAVEAREA_FROM_FP(self->interpSave.curFrame);
    401         if (saveArea == NULL) {
    402             return;
    403         }
    404         mon->ownerMethod = saveArea->method;
    405         mon->ownerPc = (saveArea->xtra.currentPc - saveArea->method->insns);
    406     }
    407 }
    408 
    409 /*
    410  * Try to lock a monitor.
    411  *
    412  * Returns "true" on success.
    413  */
    414 #ifdef WITH_COPYING_GC
    415 static bool tryLockMonitor(Thread* self, Monitor* mon)
    416 {
    417     if (mon->owner == self) {
    418         mon->lockCount++;
    419         return true;
    420     } else {
    421         if (dvmTryLockMutex(&mon->lock) == 0) {
    422             mon->owner = self;
    423             assert(mon->lockCount == 0);
    424             return true;
    425         } else {
    426             return false;
    427         }
    428     }
    429 }
    430 #endif
    431 
    432 /*
    433  * Unlock a monitor.
    434  *
    435  * Returns true if the unlock succeeded.
    436  * If the unlock failed, an exception will be pending.
    437  */
    438 static bool unlockMonitor(Thread* self, Monitor* mon)
    439 {
    440     assert(self != NULL);
    441     assert(mon != NULL);
    442     if (mon->owner == self) {
    443         /*
    444          * We own the monitor, so nobody else can be in here.
    445          */
    446         if (mon->lockCount == 0) {
    447             mon->owner = NULL;
    448             mon->ownerMethod = NULL;
    449             mon->ownerPc = 0;
    450             dvmUnlockMutex(&mon->lock);
    451         } else {
    452             mon->lockCount--;
    453         }
    454     } else {
    455         /*
    456          * We don't own this, so we're not allowed to unlock it.
    457          * The JNI spec says that we should throw IllegalMonitorStateException
    458          * in this case.
    459          */
    460         dvmThrowIllegalMonitorStateException("unlock of unowned monitor");
    461         return false;
    462     }
    463     return true;
    464 }
    465 
    466 /*
    467  * Checks the wait set for circular structure.  Returns 0 if the list
    468  * is not circular.  Otherwise, returns 1.  Used only by asserts.
    469  */
    470 #ifndef NDEBUG
    471 static int waitSetCheck(Monitor *mon)
    472 {
    473     Thread *fast, *slow;
    474     size_t n;
    475 
    476     assert(mon != NULL);
    477     fast = slow = mon->waitSet;
    478     n = 0;
    479     for (;;) {
    480         if (fast == NULL) return 0;
    481         if (fast->waitNext == NULL) return 0;
    482         if (fast == slow && n > 0) return 1;
    483         n += 2;
    484         fast = fast->waitNext->waitNext;
    485         slow = slow->waitNext;
    486     }
    487 }
    488 #endif
    489 
    490 /*
    491  * Links a thread into a monitor's wait set.  The monitor lock must be
    492  * held by the caller of this routine.
    493  */
    494 static void waitSetAppend(Monitor *mon, Thread *thread)
    495 {
    496     Thread *elt;
    497 
    498     assert(mon != NULL);
    499     assert(mon->owner == dvmThreadSelf());
    500     assert(thread != NULL);
    501     assert(thread->waitNext == NULL);
    502     assert(waitSetCheck(mon) == 0);
    503     if (mon->waitSet == NULL) {
    504         mon->waitSet = thread;
    505         return;
    506     }
    507     elt = mon->waitSet;
    508     while (elt->waitNext != NULL) {
    509         elt = elt->waitNext;
    510     }
    511     elt->waitNext = thread;
    512 }
    513 
    514 /*
    515  * Unlinks a thread from a monitor's wait set.  The monitor lock must
    516  * be held by the caller of this routine.
    517  */
    518 static void waitSetRemove(Monitor *mon, Thread *thread)
    519 {
    520     Thread *elt;
    521 
    522     assert(mon != NULL);
    523     assert(mon->owner == dvmThreadSelf());
    524     assert(thread != NULL);
    525     assert(waitSetCheck(mon) == 0);
    526     if (mon->waitSet == NULL) {
    527         return;
    528     }
    529     if (mon->waitSet == thread) {
    530         mon->waitSet = thread->waitNext;
    531         thread->waitNext = NULL;
    532         return;
    533     }
    534     elt = mon->waitSet;
    535     while (elt->waitNext != NULL) {
    536         if (elt->waitNext == thread) {
    537             elt->waitNext = thread->waitNext;
    538             thread->waitNext = NULL;
    539             return;
    540         }
    541         elt = elt->waitNext;
    542     }
    543 }
    544 
    545 /*
    546  * Converts the given relative waiting time into an absolute time.
    547  */
    548 static void absoluteTime(s8 msec, s4 nsec, struct timespec *ts)
    549 {
    550     s8 endSec;
    551 
    552 #ifdef HAVE_TIMEDWAIT_MONOTONIC
    553     clock_gettime(CLOCK_MONOTONIC, ts);
    554 #else
    555     {
    556         struct timeval tv;
    557         gettimeofday(&tv, NULL);
    558         ts->tv_sec = tv.tv_sec;
    559         ts->tv_nsec = tv.tv_usec * 1000;
    560     }
    561 #endif
    562     endSec = ts->tv_sec + msec / 1000;
    563     if (endSec >= 0x7fffffff) {
    564         ALOGV("NOTE: end time exceeds epoch");
    565         endSec = 0x7ffffffe;
    566     }
    567     ts->tv_sec = endSec;
    568     ts->tv_nsec = (ts->tv_nsec + (msec % 1000) * 1000000) + nsec;
    569 
    570     /* catch rollover */
    571     if (ts->tv_nsec >= 1000000000L) {
    572         ts->tv_sec++;
    573         ts->tv_nsec -= 1000000000L;
    574     }
    575 }
    576 
    577 int dvmRelativeCondWait(pthread_cond_t* cond, pthread_mutex_t* mutex,
    578                         s8 msec, s4 nsec)
    579 {
    580     int ret;
    581     struct timespec ts;
    582     absoluteTime(msec, nsec, &ts);
    583 #if defined(HAVE_TIMEDWAIT_MONOTONIC)
    584     ret = pthread_cond_timedwait_monotonic(cond, mutex, &ts);
    585 #else
    586     ret = pthread_cond_timedwait(cond, mutex, &ts);
    587 #endif
    588     assert(ret == 0 || ret == ETIMEDOUT);
    589     return ret;
    590 }
    591 
    592 /*
    593  * Wait on a monitor until timeout, interrupt, or notification.  Used for
    594  * Object.wait() and (somewhat indirectly) Thread.sleep() and Thread.join().
    595  *
    596  * If another thread calls Thread.interrupt(), we throw InterruptedException
    597  * and return immediately if one of the following are true:
    598  *  - blocked in wait(), wait(long), or wait(long, int) methods of Object
    599  *  - blocked in join(), join(long), or join(long, int) methods of Thread
    600  *  - blocked in sleep(long), or sleep(long, int) methods of Thread
    601  * Otherwise, we set the "interrupted" flag.
    602  *
    603  * Checks to make sure that "nsec" is in the range 0-999999
    604  * (i.e. fractions of a millisecond) and throws the appropriate
    605  * exception if it isn't.
    606  *
    607  * The spec allows "spurious wakeups", and recommends that all code using
    608  * Object.wait() do so in a loop.  This appears to derive from concerns
    609  * about pthread_cond_wait() on multiprocessor systems.  Some commentary
    610  * on the web casts doubt on whether these can/should occur.
    611  *
    612  * Since we're allowed to wake up "early", we clamp extremely long durations
    613  * to return at the end of the 32-bit time epoch.
    614  */
    615 static void waitMonitor(Thread* self, Monitor* mon, s8 msec, s4 nsec,
    616     bool interruptShouldThrow)
    617 {
    618     struct timespec ts;
    619     bool wasInterrupted = false;
    620     bool timed;
    621     int ret;
    622 
    623     assert(self != NULL);
    624     assert(mon != NULL);
    625 
    626     /* Make sure that we hold the lock. */
    627     if (mon->owner != self) {
    628         dvmThrowIllegalMonitorStateException(
    629             "object not locked by thread before wait()");
    630         return;
    631     }
    632 
    633     /*
    634      * Enforce the timeout range.
    635      */
    636     if (msec < 0 || nsec < 0 || nsec > 999999) {
    637         dvmThrowIllegalArgumentException("timeout arguments out of range");
    638         return;
    639     }
    640 
    641     /*
    642      * Compute absolute wakeup time, if necessary.
    643      */
    644     if (msec == 0 && nsec == 0) {
    645         timed = false;
    646     } else {
    647         absoluteTime(msec, nsec, &ts);
    648         timed = true;
    649     }
    650 
    651     /*
    652      * Add ourselves to the set of threads waiting on this monitor, and
    653      * release our hold.  We need to let it go even if we're a few levels
    654      * deep in a recursive lock, and we need to restore that later.
    655      *
    656      * We append to the wait set ahead of clearing the count and owner
    657      * fields so the subroutine can check that the calling thread owns
    658      * the monitor.  Aside from that, the order of member updates is
    659      * not order sensitive as we hold the pthread mutex.
    660      */
    661     waitSetAppend(mon, self);
    662     int prevLockCount = mon->lockCount;
    663     mon->lockCount = 0;
    664     mon->owner = NULL;
    665 
    666     const Method* savedMethod = mon->ownerMethod;
    667     u4 savedPc = mon->ownerPc;
    668     mon->ownerMethod = NULL;
    669     mon->ownerPc = 0;
    670 
    671     /*
    672      * Update thread status.  If the GC wakes up, it'll ignore us, knowing
    673      * that we won't touch any references in this state, and we'll check
    674      * our suspend mode before we transition out.
    675      */
    676     if (timed)
    677         dvmChangeStatus(self, THREAD_TIMED_WAIT);
    678     else
    679         dvmChangeStatus(self, THREAD_WAIT);
    680 
    681     dvmLockMutex(&self->waitMutex);
    682 
    683     /*
    684      * Set waitMonitor to the monitor object we will be waiting on.
    685      * When waitMonitor is non-NULL a notifying or interrupting thread
    686      * must signal the thread's waitCond to wake it up.
    687      */
    688     assert(self->waitMonitor == NULL);
    689     self->waitMonitor = mon;
    690 
    691     /*
    692      * Handle the case where the thread was interrupted before we called
    693      * wait().
    694      */
    695     if (self->interrupted) {
    696         wasInterrupted = true;
    697         self->waitMonitor = NULL;
    698         dvmUnlockMutex(&self->waitMutex);
    699         goto done;
    700     }
    701 
    702     /*
    703      * Release the monitor lock and wait for a notification or
    704      * a timeout to occur.
    705      */
    706     dvmUnlockMutex(&mon->lock);
    707 
    708     if (!timed) {
    709         ret = pthread_cond_wait(&self->waitCond, &self->waitMutex);
    710         assert(ret == 0);
    711     } else {
    712 #ifdef HAVE_TIMEDWAIT_MONOTONIC
    713         ret = pthread_cond_timedwait_monotonic(&self->waitCond, &self->waitMutex, &ts);
    714 #else
    715         ret = pthread_cond_timedwait(&self->waitCond, &self->waitMutex, &ts);
    716 #endif
    717         assert(ret == 0 || ret == ETIMEDOUT);
    718     }
    719     if (self->interrupted) {
    720         wasInterrupted = true;
    721     }
    722 
    723     self->interrupted = false;
    724     self->waitMonitor = NULL;
    725 
    726     dvmUnlockMutex(&self->waitMutex);
    727 
    728     /* Reacquire the monitor lock. */
    729     lockMonitor(self, mon);
    730 
    731 done:
    732     /*
    733      * We remove our thread from wait set after restoring the count
    734      * and owner fields so the subroutine can check that the calling
    735      * thread owns the monitor. Aside from that, the order of member
    736      * updates is not order sensitive as we hold the pthread mutex.
    737      */
    738     mon->owner = self;
    739     mon->lockCount = prevLockCount;
    740     mon->ownerMethod = savedMethod;
    741     mon->ownerPc = savedPc;
    742     waitSetRemove(mon, self);
    743 
    744     /* set self->status back to THREAD_RUNNING, and self-suspend if needed */
    745     dvmChangeStatus(self, THREAD_RUNNING);
    746 
    747     if (wasInterrupted) {
    748         /*
    749          * We were interrupted while waiting, or somebody interrupted an
    750          * un-interruptible thread earlier and we're bailing out immediately.
    751          *
    752          * The doc sayeth: "The interrupted status of the current thread is
    753          * cleared when this exception is thrown."
    754          */
    755         self->interrupted = false;
    756         if (interruptShouldThrow) {
    757             dvmThrowInterruptedException(NULL);
    758         }
    759     }
    760 }
    761 
    762 /*
    763  * Notify one thread waiting on this monitor.
    764  */
    765 static void notifyMonitor(Thread* self, Monitor* mon)
    766 {
    767     Thread* thread;
    768 
    769     assert(self != NULL);
    770     assert(mon != NULL);
    771 
    772     /* Make sure that we hold the lock. */
    773     if (mon->owner != self) {
    774         dvmThrowIllegalMonitorStateException(
    775             "object not locked by thread before notify()");
    776         return;
    777     }
    778     /* Signal the first waiting thread in the wait set. */
    779     while (mon->waitSet != NULL) {
    780         thread = mon->waitSet;
    781         mon->waitSet = thread->waitNext;
    782         thread->waitNext = NULL;
    783         dvmLockMutex(&thread->waitMutex);
    784         /* Check to see if the thread is still waiting. */
    785         if (thread->waitMonitor != NULL) {
    786             pthread_cond_signal(&thread->waitCond);
    787             dvmUnlockMutex(&thread->waitMutex);
    788             return;
    789         }
    790         dvmUnlockMutex(&thread->waitMutex);
    791     }
    792 }
    793 
    794 /*
    795  * Notify all threads waiting on this monitor.
    796  */
    797 static void notifyAllMonitor(Thread* self, Monitor* mon)
    798 {
    799     Thread* thread;
    800 
    801     assert(self != NULL);
    802     assert(mon != NULL);
    803 
    804     /* Make sure that we hold the lock. */
    805     if (mon->owner != self) {
    806         dvmThrowIllegalMonitorStateException(
    807             "object not locked by thread before notifyAll()");
    808         return;
    809     }
    810     /* Signal all threads in the wait set. */
    811     while (mon->waitSet != NULL) {
    812         thread = mon->waitSet;
    813         mon->waitSet = thread->waitNext;
    814         thread->waitNext = NULL;
    815         dvmLockMutex(&thread->waitMutex);
    816         /* Check to see if the thread is still waiting. */
    817         if (thread->waitMonitor != NULL) {
    818             pthread_cond_signal(&thread->waitCond);
    819         }
    820         dvmUnlockMutex(&thread->waitMutex);
    821     }
    822 }
    823 
    824 /*
    825  * Changes the shape of a monitor from thin to fat, preserving the
    826  * internal lock state.  The calling thread must own the lock.
    827  */
    828 static void inflateMonitor(Thread *self, Object *obj)
    829 {
    830     Monitor *mon;
    831     u4 thin;
    832 
    833     assert(self != NULL);
    834     assert(obj != NULL);
    835     assert(LW_SHAPE(obj->lock) == LW_SHAPE_THIN);
    836     assert(LW_LOCK_OWNER(obj->lock) == self->threadId);
    837     /* Allocate and acquire a new monitor. */
    838     mon = dvmCreateMonitor(obj);
    839     lockMonitor(self, mon);
    840     /* Propagate the lock state. */
    841     thin = obj->lock;
    842     mon->lockCount = LW_LOCK_COUNT(thin);
    843     thin &= LW_HASH_STATE_MASK << LW_HASH_STATE_SHIFT;
    844     thin |= (u4)mon | LW_SHAPE_FAT;
    845     /* Publish the updated lock word. */
    846     android_atomic_release_store(thin, (int32_t *)&obj->lock);
    847 }
    848 
    849 /*
    850  * Implements monitorenter for "synchronized" stuff.
    851  *
    852  * This does not fail or throw an exception (unless deadlock prediction
    853  * is enabled and set to "err" mode).
    854  */
    855 void dvmLockObject(Thread* self, Object *obj)
    856 {
    857     volatile u4 *thinp;
    858     ThreadStatus oldStatus;
    859     struct timespec tm;
    860     long sleepDelayNs;
    861     long minSleepDelayNs = 1000000;  /* 1 millisecond */
    862     long maxSleepDelayNs = 1000000000;  /* 1 second */
    863     u4 thin, newThin, threadId;
    864 
    865     assert(self != NULL);
    866     assert(obj != NULL);
    867     threadId = self->threadId;
    868     thinp = &obj->lock;
    869 retry:
    870     thin = *thinp;
    871     if (LW_SHAPE(thin) == LW_SHAPE_THIN) {
    872         /*
    873          * The lock is a thin lock.  The owner field is used to
    874          * determine the acquire method, ordered by cost.
    875          */
    876         if (LW_LOCK_OWNER(thin) == threadId) {
    877             /*
    878              * The calling thread owns the lock.  Increment the
    879              * value of the recursion count field.
    880              */
    881             obj->lock += 1 << LW_LOCK_COUNT_SHIFT;
    882             if (LW_LOCK_COUNT(obj->lock) == LW_LOCK_COUNT_MASK) {
    883                 /*
    884                  * The reacquisition limit has been reached.  Inflate
    885                  * the lock so the next acquire will not overflow the
    886                  * recursion count field.
    887                  */
    888                 inflateMonitor(self, obj);
    889             }
    890         } else if (LW_LOCK_OWNER(thin) == 0) {
    891             /*
    892              * The lock is unowned.  Install the thread id of the
    893              * calling thread into the owner field.  This is the
    894              * common case.  In performance critical code the JIT
    895              * will have tried this before calling out to the VM.
    896              */
    897             newThin = thin | (threadId << LW_LOCK_OWNER_SHIFT);
    898             if (android_atomic_acquire_cas(thin, newThin,
    899                     (int32_t*)thinp) != 0) {
    900                 /*
    901                  * The acquire failed.  Try again.
    902                  */
    903                 goto retry;
    904             }
    905         } else {
    906             ALOGV("(%d) spin on lock %p: %#x (%#x) %#x",
    907                  threadId, &obj->lock, 0, *thinp, thin);
    908             /*
    909              * The lock is owned by another thread.  Notify the VM
    910              * that we are about to wait.
    911              */
    912             oldStatus = dvmChangeStatus(self, THREAD_MONITOR);
    913             /*
    914              * Spin until the thin lock is released or inflated.
    915              */
    916             sleepDelayNs = 0;
    917             for (;;) {
    918                 thin = *thinp;
    919                 /*
    920                  * Check the shape of the lock word.  Another thread
    921                  * may have inflated the lock while we were waiting.
    922                  */
    923                 if (LW_SHAPE(thin) == LW_SHAPE_THIN) {
    924                     if (LW_LOCK_OWNER(thin) == 0) {
    925                         /*
    926                          * The lock has been released.  Install the
    927                          * thread id of the calling thread into the
    928                          * owner field.
    929                          */
    930                         newThin = thin | (threadId << LW_LOCK_OWNER_SHIFT);
    931                         if (android_atomic_acquire_cas(thin, newThin,
    932                                 (int32_t *)thinp) == 0) {
    933                             /*
    934                              * The acquire succeed.  Break out of the
    935                              * loop and proceed to inflate the lock.
    936                              */
    937                             break;
    938                         }
    939                     } else {
    940                         /*
    941                          * The lock has not been released.  Yield so
    942                          * the owning thread can run.
    943                          */
    944                         if (sleepDelayNs == 0) {
    945                             sched_yield();
    946                             sleepDelayNs = minSleepDelayNs;
    947                         } else {
    948                             tm.tv_sec = 0;
    949                             tm.tv_nsec = sleepDelayNs;
    950                             nanosleep(&tm, NULL);
    951                             /*
    952                              * Prepare the next delay value.  Wrap to
    953                              * avoid once a second polls for eternity.
    954                              */
    955                             if (sleepDelayNs < maxSleepDelayNs / 2) {
    956                                 sleepDelayNs *= 2;
    957                             } else {
    958                                 sleepDelayNs = minSleepDelayNs;
    959                             }
    960                         }
    961                     }
    962                 } else {
    963                     /*
    964                      * The thin lock was inflated by another thread.
    965                      * Let the VM know we are no longer waiting and
    966                      * try again.
    967                      */
    968                     ALOGV("(%d) lock %p surprise-fattened",
    969                              threadId, &obj->lock);
    970                     dvmChangeStatus(self, oldStatus);
    971                     goto retry;
    972                 }
    973             }
    974             ALOGV("(%d) spin on lock done %p: %#x (%#x) %#x",
    975                  threadId, &obj->lock, 0, *thinp, thin);
    976             /*
    977              * We have acquired the thin lock.  Let the VM know that
    978              * we are no longer waiting.
    979              */
    980             dvmChangeStatus(self, oldStatus);
    981             /*
    982              * Fatten the lock.
    983              */
    984             inflateMonitor(self, obj);
    985             ALOGV("(%d) lock %p fattened", threadId, &obj->lock);
    986         }
    987     } else {
    988         /*
    989          * The lock is a fat lock.
    990          */
    991         assert(LW_MONITOR(obj->lock) != NULL);
    992         lockMonitor(self, LW_MONITOR(obj->lock));
    993     }
    994 }
    995 
    996 /*
    997  * Implements monitorexit for "synchronized" stuff.
    998  *
    999  * On failure, throws an exception and returns "false".
   1000  */
   1001 bool dvmUnlockObject(Thread* self, Object *obj)
   1002 {
   1003     u4 thin;
   1004 
   1005     assert(self != NULL);
   1006     assert(self->status == THREAD_RUNNING);
   1007     assert(obj != NULL);
   1008     /*
   1009      * Cache the lock word as its value can change while we are
   1010      * examining its state.
   1011      */
   1012     thin = *(volatile u4 *)&obj->lock;
   1013     if (LW_SHAPE(thin) == LW_SHAPE_THIN) {
   1014         /*
   1015          * The lock is thin.  We must ensure that the lock is owned
   1016          * by the given thread before unlocking it.
   1017          */
   1018         if (LW_LOCK_OWNER(thin) == self->threadId) {
   1019             /*
   1020              * We are the lock owner.  It is safe to update the lock
   1021              * without CAS as lock ownership guards the lock itself.
   1022              */
   1023             if (LW_LOCK_COUNT(thin) == 0) {
   1024                 /*
   1025                  * The lock was not recursively acquired, the common
   1026                  * case.  Unlock by clearing all bits except for the
   1027                  * hash state.
   1028                  */
   1029                 thin &= (LW_HASH_STATE_MASK << LW_HASH_STATE_SHIFT);
   1030                 android_atomic_release_store(thin, (int32_t*)&obj->lock);
   1031             } else {
   1032                 /*
   1033                  * The object was recursively acquired.  Decrement the
   1034                  * lock recursion count field.
   1035                  */
   1036                 obj->lock -= 1 << LW_LOCK_COUNT_SHIFT;
   1037             }
   1038         } else {
   1039             /*
   1040              * We do not own the lock.  The JVM spec requires that we
   1041              * throw an exception in this case.
   1042              */
   1043             dvmThrowIllegalMonitorStateException("unlock of unowned monitor");
   1044             return false;
   1045         }
   1046     } else {
   1047         /*
   1048          * The lock is fat.  We must check to see if unlockMonitor has
   1049          * raised any exceptions before continuing.
   1050          */
   1051         assert(LW_MONITOR(obj->lock) != NULL);
   1052         if (!unlockMonitor(self, LW_MONITOR(obj->lock))) {
   1053             /*
   1054              * An exception has been raised.  Do not fall through.
   1055              */
   1056             return false;
   1057         }
   1058     }
   1059     return true;
   1060 }
   1061 
   1062 /*
   1063  * Object.wait().  Also called for class init.
   1064  */
   1065 void dvmObjectWait(Thread* self, Object *obj, s8 msec, s4 nsec,
   1066     bool interruptShouldThrow)
   1067 {
   1068     Monitor* mon;
   1069     u4 thin = *(volatile u4 *)&obj->lock;
   1070 
   1071     /* If the lock is still thin, we need to fatten it.
   1072      */
   1073     if (LW_SHAPE(thin) == LW_SHAPE_THIN) {
   1074         /* Make sure that 'self' holds the lock.
   1075          */
   1076         if (LW_LOCK_OWNER(thin) != self->threadId) {
   1077             dvmThrowIllegalMonitorStateException(
   1078                 "object not locked by thread before wait()");
   1079             return;
   1080         }
   1081 
   1082         /* This thread holds the lock.  We need to fatten the lock
   1083          * so 'self' can block on it.  Don't update the object lock
   1084          * field yet, because 'self' needs to acquire the lock before
   1085          * any other thread gets a chance.
   1086          */
   1087         inflateMonitor(self, obj);
   1088         ALOGV("(%d) lock %p fattened by wait()", self->threadId, &obj->lock);
   1089     }
   1090     mon = LW_MONITOR(obj->lock);
   1091     waitMonitor(self, mon, msec, nsec, interruptShouldThrow);
   1092 }
   1093 
   1094 /*
   1095  * Object.notify().
   1096  */
   1097 void dvmObjectNotify(Thread* self, Object *obj)
   1098 {
   1099     u4 thin = *(volatile u4 *)&obj->lock;
   1100 
   1101     /* If the lock is still thin, there aren't any waiters;
   1102      * waiting on an object forces lock fattening.
   1103      */
   1104     if (LW_SHAPE(thin) == LW_SHAPE_THIN) {
   1105         /* Make sure that 'self' holds the lock.
   1106          */
   1107         if (LW_LOCK_OWNER(thin) != self->threadId) {
   1108             dvmThrowIllegalMonitorStateException(
   1109                 "object not locked by thread before notify()");
   1110             return;
   1111         }
   1112 
   1113         /* no-op;  there are no waiters to notify.
   1114          */
   1115     } else {
   1116         /* It's a fat lock.
   1117          */
   1118         notifyMonitor(self, LW_MONITOR(thin));
   1119     }
   1120 }
   1121 
   1122 /*
   1123  * Object.notifyAll().
   1124  */
   1125 void dvmObjectNotifyAll(Thread* self, Object *obj)
   1126 {
   1127     u4 thin = *(volatile u4 *)&obj->lock;
   1128 
   1129     /* If the lock is still thin, there aren't any waiters;
   1130      * waiting on an object forces lock fattening.
   1131      */
   1132     if (LW_SHAPE(thin) == LW_SHAPE_THIN) {
   1133         /* Make sure that 'self' holds the lock.
   1134          */
   1135         if (LW_LOCK_OWNER(thin) != self->threadId) {
   1136             dvmThrowIllegalMonitorStateException(
   1137                 "object not locked by thread before notifyAll()");
   1138             return;
   1139         }
   1140 
   1141         /* no-op;  there are no waiters to notify.
   1142          */
   1143     } else {
   1144         /* It's a fat lock.
   1145          */
   1146         notifyAllMonitor(self, LW_MONITOR(thin));
   1147     }
   1148 }
   1149 
   1150 /*
   1151  * This implements java.lang.Thread.sleep(long msec, int nsec).
   1152  *
   1153  * The sleep is interruptible by other threads, which means we can't just
   1154  * plop into an OS sleep call.  (We probably could if we wanted to send
   1155  * signals around and rely on EINTR, but that's inefficient and relies
   1156  * on native code respecting our signal mask.)
   1157  *
   1158  * We have to do all of this stuff for Object.wait() as well, so it's
   1159  * easiest to just sleep on a private Monitor.
   1160  *
   1161  * It appears that we want sleep(0,0) to go through the motions of sleeping
   1162  * for a very short duration, rather than just returning.
   1163  */
   1164 void dvmThreadSleep(u8 msec, u4 nsec)
   1165 {
   1166     Thread* self = dvmThreadSelf();
   1167     Monitor* mon = gDvm.threadSleepMon;
   1168 
   1169     /* sleep(0,0) wakes up immediately, wait(0,0) means wait forever; adjust */
   1170     if (msec == 0 && nsec == 0)
   1171         nsec++;
   1172 
   1173     lockMonitor(self, mon);
   1174     waitMonitor(self, mon, msec, nsec, true);
   1175     unlockMonitor(self, mon);
   1176 }
   1177 
   1178 /*
   1179  * Implement java.lang.Thread.interrupt().
   1180  */
   1181 void dvmThreadInterrupt(Thread* thread)
   1182 {
   1183     assert(thread != NULL);
   1184 
   1185     dvmLockMutex(&thread->waitMutex);
   1186 
   1187     /*
   1188      * If the interrupted flag is already set no additional action is
   1189      * required.
   1190      */
   1191     if (thread->interrupted == true) {
   1192         dvmUnlockMutex(&thread->waitMutex);
   1193         return;
   1194     }
   1195 
   1196     /*
   1197      * Raise the "interrupted" flag.  This will cause it to bail early out
   1198      * of the next wait() attempt, if it's not currently waiting on
   1199      * something.
   1200      */
   1201     thread->interrupted = true;
   1202 
   1203     /*
   1204      * Is the thread waiting?
   1205      *
   1206      * Note that fat vs. thin doesn't matter here;  waitMonitor
   1207      * is only set when a thread actually waits on a monitor,
   1208      * which implies that the monitor has already been fattened.
   1209      */
   1210     if (thread->waitMonitor != NULL) {
   1211         pthread_cond_signal(&thread->waitCond);
   1212     }
   1213 
   1214     dvmUnlockMutex(&thread->waitMutex);
   1215 }
   1216 
   1217 #ifndef WITH_COPYING_GC
   1218 u4 dvmIdentityHashCode(Object *obj)
   1219 {
   1220     return (u4)obj;
   1221 }
   1222 #else
   1223 /*
   1224  * Returns the identity hash code of the given object.
   1225  */
   1226 u4 dvmIdentityHashCode(Object *obj)
   1227 {
   1228     Thread *self, *thread;
   1229     volatile u4 *lw;
   1230     size_t size;
   1231     u4 lock, owner, hashState;
   1232 
   1233     if (obj == NULL) {
   1234         /*
   1235          * Null is defined to have an identity hash code of 0.
   1236          */
   1237         return 0;
   1238     }
   1239     lw = &obj->lock;
   1240 retry:
   1241     hashState = LW_HASH_STATE(*lw);
   1242     if (hashState == LW_HASH_STATE_HASHED) {
   1243         /*
   1244          * The object has been hashed but has not had its hash code
   1245          * relocated by the garbage collector.  Use the raw object
   1246          * address.
   1247          */
   1248         return (u4)obj >> 3;
   1249     } else if (hashState == LW_HASH_STATE_HASHED_AND_MOVED) {
   1250         /*
   1251          * The object has been hashed and its hash code has been
   1252          * relocated by the collector.  Use the value of the naturally
   1253          * aligned word following the instance data.
   1254          */
   1255         assert(!dvmIsClassObject(obj));
   1256         if (IS_CLASS_FLAG_SET(obj->clazz, CLASS_ISARRAY)) {
   1257             size = dvmArrayObjectSize((ArrayObject *)obj);
   1258             size = (size + 2) & ~2;
   1259         } else {
   1260             size = obj->clazz->objectSize;
   1261         }
   1262         return *(u4 *)(((char *)obj) + size);
   1263     } else if (hashState == LW_HASH_STATE_UNHASHED) {
   1264         /*
   1265          * The object has never been hashed.  Change the hash state to
   1266          * hashed and use the raw object address.
   1267          */
   1268         self = dvmThreadSelf();
   1269         if (self->threadId == lockOwner(obj)) {
   1270             /*
   1271              * We already own the lock so we can update the hash state
   1272              * directly.
   1273              */
   1274             *lw |= (LW_HASH_STATE_HASHED << LW_HASH_STATE_SHIFT);
   1275             return (u4)obj >> 3;
   1276         }
   1277         /*
   1278          * We do not own the lock.  Try acquiring the lock.  Should
   1279          * this fail, we must suspend the owning thread.
   1280          */
   1281         if (LW_SHAPE(*lw) == LW_SHAPE_THIN) {
   1282             /*
   1283              * If the lock is thin assume it is unowned.  We simulate
   1284              * an acquire, update, and release with a single CAS.
   1285              */
   1286             lock = (LW_HASH_STATE_HASHED << LW_HASH_STATE_SHIFT);
   1287             if (android_atomic_acquire_cas(
   1288                                 0,
   1289                                 (int32_t)lock,
   1290                                 (int32_t *)lw) == 0) {
   1291                 /*
   1292                  * A new lockword has been installed with a hash state
   1293                  * of hashed.  Use the raw object address.
   1294                  */
   1295                 return (u4)obj >> 3;
   1296             }
   1297         } else {
   1298             if (tryLockMonitor(self, LW_MONITOR(*lw))) {
   1299                 /*
   1300                  * The monitor lock has been acquired.  Change the
   1301                  * hash state to hashed and use the raw object
   1302                  * address.
   1303                  */
   1304                 *lw |= (LW_HASH_STATE_HASHED << LW_HASH_STATE_SHIFT);
   1305                 unlockMonitor(self, LW_MONITOR(*lw));
   1306                 return (u4)obj >> 3;
   1307             }
   1308         }
   1309         /*
   1310          * At this point we have failed to acquire the lock.  We must
   1311          * identify the owning thread and suspend it.
   1312          */
   1313         dvmLockThreadList(self);
   1314         /*
   1315          * Cache the lock word as its value can change between
   1316          * determining its shape and retrieving its owner.
   1317          */
   1318         lock = *lw;
   1319         if (LW_SHAPE(lock) == LW_SHAPE_THIN) {
   1320             /*
   1321              * Find the thread with the corresponding thread id.
   1322              */
   1323             owner = LW_LOCK_OWNER(lock);
   1324             assert(owner != self->threadId);
   1325             /*
   1326              * If the lock has no owner do not bother scanning the
   1327              * thread list and fall through to the failure handler.
   1328              */
   1329             thread = owner ? gDvm.threadList : NULL;
   1330             while (thread != NULL) {
   1331                 if (thread->threadId == owner) {
   1332                     break;
   1333                 }
   1334                 thread = thread->next;
   1335             }
   1336         } else {
   1337             thread = LW_MONITOR(lock)->owner;
   1338         }
   1339         /*
   1340          * If thread is NULL the object has been released since the
   1341          * thread list lock was acquired.  Try again.
   1342          */
   1343         if (thread == NULL) {
   1344             dvmUnlockThreadList();
   1345             goto retry;
   1346         }
   1347         /*
   1348          * Wait for the owning thread to suspend.
   1349          */
   1350         dvmSuspendThread(thread);
   1351         if (dvmHoldsLock(thread, obj)) {
   1352             /*
   1353              * The owning thread has been suspended.  We can safely
   1354              * change the hash state to hashed.
   1355              */
   1356             *lw |= (LW_HASH_STATE_HASHED << LW_HASH_STATE_SHIFT);
   1357             dvmResumeThread(thread);
   1358             dvmUnlockThreadList();
   1359             return (u4)obj >> 3;
   1360         }
   1361         /*
   1362          * The wrong thread has been suspended.  Try again.
   1363          */
   1364         dvmResumeThread(thread);
   1365         dvmUnlockThreadList();
   1366         goto retry;
   1367     }
   1368     ALOGE("object %p has an unknown hash state %#x", obj, hashState);
   1369     dvmDumpThread(dvmThreadSelf(), false);
   1370     dvmAbort();
   1371     return 0;  /* Quiet the compiler. */
   1372 }
   1373 #endif  /* WITH_COPYING_GC */
   1374