Changed to sourceforge tdb code. This includes spinlocks (so we now have

a --with-spinlocks option to configure, this does mean the on-disk tdb
format has changed, so 2.2alphaX sites will need to re-create their
tdb's. The upside is no more tdb fragmentation and a +5% on netbench.
Swings and roundabouts....
Jeremy.
(This used to be commit 9dea7b7c257db487f8ced7dad3fce92fba03ea91)

1 files changed, 403 insertions, 0 deletions

diff --git a/source3/tdb/spinlock.c b/source3/tdb/spinlock.c
new file mode 100644
index 00000000000..731188a1530
--- /dev/null
+++ b/source3/tdb/spinlock.c
@@ -0,0 +1,403 @@
+#if STANDALONE
+#if HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <time.h>
+#include "tdb.h"
+#include "spinlock.h"
+
+#define DEBUG
+#else
+#include "includes.h"
+#endif
+
+#ifdef USE_SPINLOCKS
+
+/*
+ * ARCH SPECIFIC
+ */
+
+#if defined(SPARC_SPINLOCKS)
+
+static inline int __spin_trylock(spinlock_t *lock)
+{
+	unsigned int result;
+
+	asm volatile("ldstub    [%1], %0"
+		: "=r" (result)
+		: "r" (lock)
+		: "memory");
+
+	return (result == 0) ? 0 : EBUSY;
+}
+
+static inline void __spin_unlock(spinlock_t *lock)
+{
+	*lock = 0;
+}
+
+static inline void __spin_lock_init(spinlock_t *lock)
+{
+	*lock = 0;
+}
+
+static inline int __spin_is_locked(spinlock_t *lock)
+{
+	return (*lock != 0);
+}
+
+#elif defined(POWERPC_SPINLOCKS) 
+
+static inline int __spin_trylock(spinlock_t *lock)
+{
+	int result;
+
+	__asm__ __volatile__ (
+		"	eieio;"
+		"0:	lwarx %0,0,%1;"
+		"	cmpwi 0,%0,0;"
+		"	bne- 1f;"
+		"	stwcx. %2,0,%1;"
+		"	bne- 0b;"
+		"	sync;"
+		"1:"
+                : "=&r"(result)
+                : "r"(lock), "r"(1)
+                : "cr0", "memory");
+
+	return (result == 0) ? 0 : EBUSY;
+}
+
+static inline void __spin_unlock(spinlock_t *lock)
+{
+	asm volatile("sync");
+	*lock = 0;
+}
+
+static inline void __spin_lock_init(spinlock_t *lock)
+{
+	*lock = 0;
+}
+
+static inline int __spin_is_locked(spinlock_t *lock)
+{
+	return (*lock != 0);
+}
+
+#elif defined(INTEL_SPINLOCKS) 
+
+static inline int __spin_trylock(spinlock_t *lock)
+{
+	int oldval;
+
+	asm volatile("xchgl %0,%1"
+		: "=r" (oldval), "=m" (*lock)
+		: "0" (0));
+	return oldval > 0 ? 0 : EBUSY;
+}
+
+static inline void __spin_unlock(spinlock_t *lock)
+{
+	*lock = 1;
+}
+
+static inline void __spin_lock_init(spinlock_t *lock)
+{
+	*lock = 1;
+}
+
+static inline int __spin_is_locked(spinlock_t *lock)
+{
+	return (*lock != 1);
+}
+
+#elif defined(MIPS_SPINLOCKS) 
+
+static inline unsigned int load_linked(unsigned long addr)
+{
+	unsigned int res;
+
+	__asm__ __volatile__("ll\t%0,(%1)"
+		: "=r" (res)
+		: "r" (addr));
+
+	return res;
+}
+
+static inline unsigned int store_conditional(unsigned long addr, unsigned int value)
+{
+	unsigned int res;
+
+	__asm__ __volatile__("sc\t%0,(%2)"
+		: "=r" (res)
+		: "0" (value), "r" (addr));
+	return res;
+}
+
+static inline int __spin_trylock(spinlock_t *lock)
+{
+	unsigned int mw;
+
+	do {
+		mw = load_linked(lock);
+		if (mw) 
+			return EBUSY;
+	} while (!store_conditional(lock, 1));
+
+	return 0;
+}
+
+static inline void __spin_unlock(spinlock_t *lock)
+{
+	*lock = 0;
+}
+
+static inline void __spin_lock_init(spinlock_t *lock)
+{
+	*lock = 0;
+}
+
+static inline int __spin_is_locked(spinlock_t *lock)
+{
+	return (*lock != 0);
+}
+
+#else
+#error Need to implement spinlock code in spinlock.c
+#endif
+
+/*
+ * OS SPECIFIC
+ */
+
+static void yield_cpu(void)
+{
+	struct timespec tm;
+
+#ifdef USE_SCHED_YIELD
+	sched_yield();
+#else
+	/* Linux will busy loop for delays < 2ms on real time tasks */
+	tm.tv_sec = 0;
+	tm.tv_nsec = 2000000L + 1;
+	nanosleep(&tm, NULL);
+#endif
+}
+
+static int this_is_smp(void)
+{
+	return 0;
+}
+
+/*
+ * GENERIC
+ */
+
+static int smp_machine = 0;
+
+static inline void __spin_lock(spinlock_t *lock)
+{
+	int ntries = 0;
+
+	while(__spin_trylock(lock)) {
+		while(__spin_is_locked(lock)) {
+			if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
+				continue;
+			yield_cpu();
+		}
+	}
+}
+
+static void __read_lock(rwlock_t *rwlock)
+{
+	int ntries = 0;
+
+	while(1) {
+		__spin_lock(&rwlock->lock);
+
+		if (!(rwlock->count & RWLOCK_BIAS)) {
+			rwlock->count++;
+			__spin_unlock(&rwlock->lock);
+			return;
+		}
+	
+		__spin_unlock(&rwlock->lock);
+
+		while(rwlock->count & RWLOCK_BIAS) {
+			if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
+				continue;
+			yield_cpu();
+		}
+	}
+}
+
+static void __write_lock(rwlock_t *rwlock)
+{
+	int ntries = 0;
+
+	while(1) {
+		__spin_lock(&rwlock->lock);
+
+		if (rwlock->count == 0) {
+			rwlock->count |= RWLOCK_BIAS;
+			__spin_unlock(&rwlock->lock);
+			return;
+		}
+
+		__spin_unlock(&rwlock->lock);
+
+		while(rwlock->count != 0) {
+			if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
+				continue;
+			yield_cpu();
+		}
+	}
+}
+
+static void __write_unlock(rwlock_t *rwlock)
+{
+	__spin_lock(&rwlock->lock);
+
+#ifdef DEBUG
+	if (!(rwlock->count & RWLOCK_BIAS))
+		fprintf(stderr, "bug: write_unlock\n");
+#endif
+
+	rwlock->count &= ~RWLOCK_BIAS;
+	__spin_unlock(&rwlock->lock);
+}
+
+static void __read_unlock(rwlock_t *rwlock)
+{
+	__spin_lock(&rwlock->lock);
+
+#ifdef DEBUG
+	if (!rwlock->count)
+		fprintf(stderr, "bug: read_unlock\n");
+
+	if (rwlock->count & RWLOCK_BIAS)
+		fprintf(stderr, "bug: read_unlock\n");
+#endif
+
+	rwlock->count--;
+	__spin_unlock(&rwlock->lock);
+}
+
+/* TDB SPECIFIC */
+
+/* lock a list in the database. list -1 is the alloc list */
+int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type)
+{
+	rwlock_t *rwlocks;
+
+	if (!tdb->map_ptr) return -1;
+	rwlocks = (rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
+
+	switch(rw_type) {
+	case F_RDLCK:
+		__read_lock(&rwlocks[list+1]);
+		break;
+
+	case F_WRLCK:
+		__write_lock(&rwlocks[list+1]);
+		break;
+
+	default:
+		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+	}
+	return 0;
+}
+
+/* unlock the database. */
+int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type)
+{
+	rwlock_t *rwlocks;
+
+	if (!tdb->map_ptr) return -1;
+	rwlocks = (rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
+
+	switch(rw_type) {
+	case F_RDLCK:
+		__read_unlock(&rwlocks[list+1]);
+		break;
+
+	case F_WRLCK:
+		__write_unlock(&rwlocks[list+1]);
+		break;
+
+	default:
+		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+	}
+
+	return 0;
+}
+
+int tdb_create_rwlocks(int fd, unsigned int hash_size)
+{
+	unsigned size, i;
+	rwlock_t *rwlocks;
+
+	size = (hash_size + 1) * sizeof(rwlock_t);
+	rwlocks = malloc(size);
+	if (!rwlocks)
+		return -1;
+
+	for(i = 0; i < hash_size+1; i++) {
+		__spin_lock_init(&rwlocks[i].lock);
+		rwlocks[i].count = 0;
+	}
+
+	/* Write it out (appending to end) */
+	if (write(fd, rwlocks, size) != size) {
+		free(rwlocks);
+		return -1;
+	}
+	smp_machine = this_is_smp();
+	free(rwlocks);
+	return 0;
+}
+
+int tdb_clear_spinlocks(TDB_CONTEXT *tdb)
+{
+	rwlock_t *rwlocks;
+	unsigned i;
+
+	if (tdb->header.rwlocks == 0) return 0;
+	if (!tdb->map_ptr) return -1;
+
+	/* We're mmapped here */
+	rwlocks = (rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
+	for(i = 0; i < tdb->header.hash_size+1; i++) {
+		__spin_lock_init(&rwlocks[i].lock);
+		rwlocks[i].count = 0;
+	}
+	return 0;
+}
+#else
+int tdb_create_rwlocks(int fd, unsigned int hash_size) { return 0; }
+int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type) { return -1; }
+int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type) { return -1; }
+
+/* Non-spinlock version: remove spinlock pointer */
+int tdb_clear_spinlocks(TDB_CONTEXT *tdb)
+{
+	tdb_off off = (tdb_off)((char *)&tdb->header.rwlocks
+				- (char *)&tdb->header);
+
+	tdb->header.rwlocks = 0;
+	if (lseek(tdb->fd, off, SEEK_SET) != off
+	    || write(tdb->fd, (void *)&tdb->header.rwlocks,
+		     sizeof(tdb->header.rwlocks)) 
+	    != sizeof(tdb->header.rwlocks))
+		return -1;
+	return 0;
+}
+#endif