| |
Synchronization objects include mutexes, condition variables, semaphores, and reader-writer locks. In the same way that thread operations use a thread handle of type td_thrhandle_t, operations on synchronization objects use a synchronization object handle of type td_synchandle_t.
The controlling process obtains synchronization object handles either by calling the function td_ta_sync_iter() to obtain handles for all synchronization objects of the target process that are known to the libthread_db library of interfaces, or by mapping the
address of a synchronization object in the address space of the target process to a handle by calling td_ta_map_addr2sync(3THR).
Not all synchronization objects that a process uses may be known to the libthread_db library and returned by td_ta_sync_iter(3THR). A synchronization object
is known to libthread_db only if it has been the target of a synchronization primitive in the process (such as mutex_lock(), described on the mutex_init(3THR)
manual page) after td_ta_new(3THR) has been called to attach to the process and td_ta_sync_tracking_enable() has been called to enable synchronization object tracking.
The td_ta_sync_tracking_enable() function turns synchronization object tracking on or off for the process identified by ta_p, depending on whether on_off is 0 (off) or non-zero (on).
The td_sync_get_info() function fills in the td_syncinfo_t structure *si_p with values for the synchronization object identified by sh_p. The td_syncinfo_t structure contains
the following fields:
-
td_thragent_t *si_ta_p
- The internal process handle identifying the target process through which this synchronization object handle was obtained. Synchronization objects may be process-private or process-shared. In
the latter case, the same synchronization object may have multiple handles, one for each target process's "view" of the synchronization object.
-
psaddr_t si_sv_addr
- The address of the synchronization object in this target process's address space.
-
td_sync_type_e si_type
- The type of the synchronization variable: mutex, condition variable, semaphore, or readers-writer lock.
-
int si_shared_type
- If the USYNC_PROCESS or USYNC_PROCESS_ROBUST bit is set in si_shared_type, this synchronization object is process-shared, otherwise
it is process-private.
-
td_sync_flags_t si_flags
- Flags dependent on the type of the synchronization object.
-
int si_state.sema_count
- Semaphores only. The current value of the semaphore
-
int si_state.nreaders
- Readers-writer locks only. The number of readers currently holding the lock, or -1, if a writer is currently holding the lock.
-
int si_state.mutex_locked
- For mutexes only. Non-zero if and only if the mutex is currently locked.
-
int si_size
- The size of the synchronization object.
-
uint8_t si_has_waiters
- Non-zero if and only if at least one thread is blocked on this synchronization object.
-
uint8_t si_is_wlocked
- For reader-writer locks only. The value is non-zero if and only if this lock is held by a writer.
-
uint8_t si_rcount
-
PTHREAD_MUTEX_RECURSIVE mutexes only. If the mutex is held, the recursion count.
-
uint8_t si_prioceiling
-
PTHREAD_PRIO_PROTECT protocol mutexes only. The priority ceiling.
-
td_thrhandle_t si_owner
- Mutexes and readers-writer locks only. This is the thread holding the mutex, or the write lock, if this is a reader-writer lock. The value is NULL if no one
holds the mutex or write-lock.
-
pid_t si_ownerpid
- Mutexes only. For a locked process-shared mutex, this is the process-ID of the process containing the owning thread.
The td_sync_get_stats() function fills in the td_syncstats_t structure *ss_p with values for the synchronization object identified by sh_p. The td_syncstats_t structure contains an
embedded td_syncinfo_t structure that is filled in as described above for td_sync_get_info(). In addition, usage statistics gathered since td_ta_sync_tracking_enable() was called to enable synchronization object tracking are returned in the ss_un.mutex, ss_un.cond, ss_un.rwlock, or ss_un.sema members of the td_syncstats_t structure, depending on the type of the synchronization object.
The td_sync_setstate function modifies the state of synchronization object si_p, depending on the synchronization object type. For mutexes, td_sync_setstate is unlocked if the value is 0. Otherwise it is locked. For
semaphores, the semaphore's count is set to the value. For reader-writer locks, the reader count set to the value if value is >0. The count is set to write-locked if value is -1. It is set to unlocked if the value is 0. Setting the state
of a synchronization object from a libthread_db interface may cause the synchronization object's semantics to be violated from the point of view of the threads in the target process. For example, if a thread holds a mutex, and td_sync_setstate is used to set the mutex
to unlocked, then a different thread will also be able to subsequently acquire the same mutex.
The td_sync_waiters function iterates over the set of thread handles of threads blocked on sh_p. The callback function cb is called once for each such thread handle, and is passed the thread handle and cb_data_p.
If the callback function returns a non-zero value, iteration is terminated early. See td_ta_thr_iter(3THR).
|
