Replace rs->multicast->mutex with a dedicated mutex for the API.

Use PTHREAD_MUTEX_RECURSIVE and change LOCK() to always call rig_lock().
Still needs a reliable way to determine availability.
pull/1675/head
George Baltz N3GB 2025-02-28 05:29:44 -05:00
rodzic 439bd96b5d
commit 0147da5a3c
2 zmienionych plików z 32 dodań i 4 usunięć

Wyświetl plik

@ -2891,6 +2891,7 @@ struct rig_state {
struct timespec freq_event_elapsed;
int freq_skip; /*!< allow frequency skip for gpredict RX/TX freq set */
client_t client;
pthread_mutex_t api_mutex; // Lock for any API entry
// New rig_state items go before this line ============================================
};

Wyświetl plik

@ -171,11 +171,21 @@ const char hamlib_copyright[231] = /* hamlib 1.2 ABI specifies 231 bytes */
#define ICOM_EXCEPTIONS (rig->caps->rig_model == RIG_MODEL_IC9700 || rig->caps->rig_model == RIG_MODEL_IC9100 || rig->caps->rig_model == RIG_MODEL_IC910)
// If the OS/library supports it, use a recursive mutex for the main lock.
// This eliminates depth races, and guards against multiple app threads, too.
// Set define to 0 to use depth-based locking. It should be deduced from the
// environment, but I can't find a fine-grained enough parameter. Should be
// OK on any POSIX-2017 or later system.
#define USE_RECURSIVE_MUTEX 1
#if USE_RECURSIVE_MUTEX
#define LOCK(n) rig_lock(rig,n)
#else
// The LOCK macro is for the primary thread calling the rig functions
// For a separate thread use rig_lock directly
// The purpose here is to avoid deadlock during recursion
// Any other thread should grab the mutex itself via rig_lock
#define LOCK(n) if (STATE(rig)->depth == 1) { rig_debug(RIG_DEBUG_CACHE, "%s: %s\n", n?"lock":"unlock", __func__); rig_lock(rig,n); }
#endif
MUTEX(morse_mutex);
@ -891,8 +901,20 @@ RIG *HAMLIB_API rig_init(rig_model_t rig_model)
// we have to copy rs to rig->state_deprecated for DLL backwards compatibility
memcpy(&rig->state_deprecated, rs, sizeof(rig->state_deprecated));
// Set up lock for any API entry point
// If available, use a recursive mutex. Else, fall back on the
// depth count.
pthread_mutexattr_t api_attr;
pthread_mutexattr_init(&api_attr);
#if USE_RECURSIVE_MUTEX
pthread_mutexattr_settype(&api_attr, PTHREAD_MUTEX_RECURSIVE);
HAMLIB_TRACE;
#endif
pthread_mutex_init(&rs->api_mutex, &api_attr);
pthread_mutexattr_destroy(&api_attr);
/*
* let the backend a chance to setup his private data
* Give the backend a chance to setup his private data
* This must be done only once defaults are setup,
* so the backend init can override rig_state.
*/
@ -1858,7 +1880,10 @@ int HAMLIB_API rig_cleanup(RIG *rig)
rig->caps->rig_cleanup(rig);
}
//TODO Release and null any allocated port structures
//pthread_mutex_destroy(&STATE(rig)->api_mutex);
//TODO Release and null any allocated data -
// state, ports, cache, etc.
free(rig);
@ -8365,6 +8390,7 @@ void rig_lock(RIG *rig, int lock)
struct rig_state *rs = STATE(rig);
#if 0
if (rs->multicast == NULL)
{
rig_debug(RIG_DEBUG_BUG, "%s: locking skipped, lock = %d\n", __func__, lock);
@ -8376,16 +8402,17 @@ void rig_lock(RIG *rig, int lock)
rs->multicast->mutex = initializer;
rs->multicast->mutex_initialized = 1;
}
#endif
if (lock)
{
pthread_mutex_lock(&rs->multicast->mutex);
pthread_mutex_lock(&rs->api_mutex);
rig_debug(RIG_DEBUG_VERBOSE, "%s: client lock engaged\n", __func__);
}
else
{
rig_debug(RIG_DEBUG_VERBOSE, "%s: client lock disengaged\n", __func__);
pthread_mutex_unlock(&rs->multicast->mutex);
pthread_mutex_unlock(&rs->api_mutex);
}
#endif