/* * $Id$ * * Author: Till Straumann , 3/2002 */ /* provide locking for the global environment 'environ' */ #if HAVE_CONFIG_H #include "config.h" #endif #include #include #include /* * NOTES: * - although it looks like a classical multiple-readers / single writer (MRSW) * locking problem, we still use a single lock for the following reasons: * 1) newlib has no provision / hook for calling different locking routines * from setenv/putenv and getenv, respectively. * 2) MRSW involves calling several semaphore-primitives, even in the most * likely case of a first-reader's access. This probably takes more CPU * time than just waiting until another reader is done; environment * access is fast. * - the lock implementation must allow nesting (same thread may call * lock-lock-unlock-unlock). * - NEWLIB-1.8.2 has an ugly BUG: if environ is NULL, _findenv_r() bails * out leaving the lock held :-( * */ static rtems_id envLock=0; static void __rtems_envlock_init(void) { extern char **environ; rtems_status_code rc; if (envLock) /* already initialized */ return; assert(environ && "MUST have non-NULL 'environ' due to newlib bug"); rc = rtems_semaphore_create( rtems_build_name('E','N','V','S'), 1, RTEMS_BINARY_SEMAPHORE | RTEMS_PRIORITY | RTEMS_INHERIT_PRIORITY, 0, &envLock); if (RTEMS_SUCCESSFUL!=rc) rtems_fatal_error_occurred(rc); } void __env_lock(struct _reent *r) { /* Do lazy init */ if (!envLock) __rtems_envlock_init(); /* * Must not use a semaphore before pre-tasking hook is called. * - it will corrupt memory :-( */ if (_Thread_Executing) rtems_semaphore_obtain(envLock, RTEMS_WAIT, RTEMS_NO_TIMEOUT); } void __env_unlock(struct _reent *r) { /* * Must not use a semaphore before pre-tasking hook is called. * - it will corrupt memory :-( */ if (_Thread_Executing) rtems_semaphore_release(envLock); }