/*
* urcu-static.h
*
- * Userspace RCU header, to be included only in LGPL-compatible code.
+ * Userspace RCU header.
*
- * Copyright February 2009 - Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
+ * TO BE INCLUDED ONLY IN LGPL-COMPATIBLE CODE. See urcu.h for linking
+ * dynamically with the userspace rcu library.
*
- * Credits for Paul E. McKenney <paulmck@linux.vnet.ibm.com>
- * for inspiration coming from the Linux kernel RCU and rcu-preempt.
+ * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
+ * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
*
- * Distributed under LGPLv2.1
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
* IBM's contributions to this file may be relicensed under LGPLv2 or later.
*/
#include <stdlib.h>
#include <pthread.h>
+#include <sched.h>
#include <compiler.h>
#include <arch.h>
* Inserts memory barriers on architectures that require them (currently only
* Alpha) and documents which pointers are protected by RCU.
*
+ * The compiler memory barrier in LOAD_SHARED() ensures that value-speculative
+ * optimizations (e.g. VSS: Value Speculation Scheduling) does not perform the
+ * data read before the pointer read by speculating the value of the pointer.
+ * Correct ordering is ensured because the pointer is read as a volatile access.
+ * This acts as a global side-effect operation, which forbids reordering of
+ * dependent memory operations. Note that such concern about dependency-breaking
+ * optimizations will eventually be taken care of by the "memory_order_consume"
+ * addition to forthcoming C++ standard.
+ *
* Should match rcu_assign_pointer() or rcu_xchg_pointer().
*/
*/
#define KICK_READER_LOOPS 10000
+/*
+ * Active attempts to check for reader Q.S. before calling sched_yield().
+ */
+#define RCU_QS_ACTIVE_ATTEMPTS 100
+
+#ifdef DEBUG_RCU
+#define rcu_assert(args...) assert(args)
+#else
+#define rcu_assert(args...)
+#endif
+
#ifdef DEBUG_YIELD
#include <sched.h>
#include <time.h>
#define YIELD_READ (1 << 0)
#define YIELD_WRITE (1 << 1)
-/* Updates without DEBUG_FULL_MB are much slower. Account this in the delay */
-#ifdef DEBUG_FULL_MB
+/*
+ * Updates without URCU_MB are much slower. Account this in
+ * the delay.
+ */
+#ifdef URCU_MB
/* maximum sleep delay, in us */
#define MAX_SLEEP 50
#else
}
#endif
-#ifdef DEBUG_FULL_MB
+#ifdef URCU_MB
static inline void reader_barrier()
{
smp_mb();
/* Use the amount of bits equal to half of the architecture long size */
#define RCU_GP_CTR_BIT (1UL << (sizeof(long) << 2))
#define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_BIT - 1)
+#define RCU_GP_ONGOING (RCU_GP_CTR_BIT << 1)
/*
* Global quiescent period counter with low-order bits unused.
*/
extern long urcu_gp_ctr;
-extern long __thread urcu_active_readers;
+struct urcu_reader_status {
+ long active_readers;
+ long gp_waiting;
+};
+
+extern struct urcu_reader_status __thread urcu_reader_status;
static inline int rcu_old_gp_ongoing(long *value)
{
static inline void _rcu_read_lock(void)
{
- long tmp;
+ long tmp, gp_ctr;
- tmp = urcu_active_readers;
+ tmp = urcu_reader_status.active_readers;
/* urcu_gp_ctr = RCU_GP_COUNT | (~RCU_GP_CTR_BIT or RCU_GP_CTR_BIT) */
- /*
- * The data dependency "read urcu_gp_ctr, write urcu_active_readers",
- * serializes those two memory operations. The memory barrier in the
- * signal handler ensures we receive the proper memory commit barriers
- * required by _STORE_SHARED and _LOAD_SHARED whenever communication
- * with the writer is needed.
- */
- if (likely(!(tmp & RCU_GP_CTR_NEST_MASK)))
- _STORE_SHARED(urcu_active_readers, _LOAD_SHARED(urcu_gp_ctr));
- else
- _STORE_SHARED(urcu_active_readers, tmp + RCU_GP_COUNT);
- /*
- * Increment active readers count before accessing the pointer.
- * See force_mb_all_threads().
- */
- reader_barrier();
+ if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
+ /*
+ * volatile accesses can be reordered and optimized when within
+ * the same statement.
+ */
+ if (unlikely((gp_ctr = _LOAD_SHARED(urcu_gp_ctr))
+ & RCU_GP_ONGOING) &&
+ unlikely(LOAD_SHARED(urcu_reader_status.gp_waiting))) {
+ sched_yield();
+ gp_ctr = _LOAD_SHARED(urcu_gp_ctr);
+ }
+ _STORE_SHARED(urcu_reader_status.active_readers, gp_ctr);
+ /*
+ * Set active readers count for outermost nesting level before
+ * accessing the pointer. See force_mb_all_threads().
+ */
+ reader_barrier();
+ } else {
+ _STORE_SHARED(urcu_reader_status.active_readers,
+ tmp + RCU_GP_COUNT);
+ }
}
static inline void _rcu_read_unlock(void)
{
- reader_barrier();
/*
* Finish using rcu before decrementing the pointer.
* See force_mb_all_threads().
+ * Formally only needed for outermost nesting level, but leave barrier
+ * in place for nested unlocks to remove a branch from the common case
+ * (no nesting).
*/
- _STORE_SHARED(urcu_active_readers, urcu_active_readers - RCU_GP_COUNT);
+ reader_barrier();
+ _STORE_SHARED(urcu_reader_status.active_readers,
+ urcu_reader_status.active_readers - RCU_GP_COUNT);
}
/**
STORE_SHARED(p, v); \
})
+/**
+ * _rcu_cmpxchg_pointer - same as rcu_assign_pointer, but tests if the pointer
+ * is as expected by "old". If succeeds, returns the previous pointer to the
+ * data structure, which can be safely freed after waiting for a quiescent state
+ * using synchronize_rcu(). If fails (unexpected value), returns old (which
+ * should not be freed !).
+ */
+
+#define _rcu_cmpxchg_pointer(p, old, _new) \
+ ({ \
+ if (!__builtin_constant_p(_new) || \
+ ((_new) != NULL)) \
+ wmb(); \
+ cmpxchg(p, old, _new); \
+ })
+
/**
* _rcu_xchg_pointer - same as rcu_assign_pointer, but returns the previous
- * pointer to the data structure, which can be safely freed after waitin for a
+ * pointer to the data structure, which can be safely freed after waiting for a
* quiescent state using synchronize_rcu().
*/