[PATCH v2 4/8] xfsdump: simplify qlock ordinal bitmap

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



The qlock abstraction manages an array of ordinal bitmaps, one bitmap
for each thread. The bitmap indicates which locks a thread is holding
and is used to verify that the locks are obtained in the correct order.

There is no need to use an array to store the bitmaps, and in fact
adding entries to the array is broken because qlock_thrdcnt is not
protected by a lock. A simpler approach is to create a per-thread bitmap
using thread local storage.

With this change, there is no need for each new thread to register with
the qlock abstraction, so qlock_thrdinit() goes away. Remove
qlock_init() since it can be statically initialized, and that means
there's no need for a flag (qlock_inited) indicating that the module has
been initialized.  Also there's no longer a need to track or limit the
number of threads that the qlock abstraction can handle.

Signed-off-by: Bill Kendall <wkendall@xxxxxxx>
Reviewed-by: Christoph Hellwig <hch@xxxxxx>
---
 common/cldmgr.c |    4 -
 common/main.c   |    7 --
 common/qlock.c  |  191 ++++---------------------------------------------------
 common/qlock.h  |   11 +---
 4 files changed, 14 insertions(+), 199 deletions(-)

diff --git a/common/cldmgr.c b/common/cldmgr.c
index 7784a15..d327bab 100644
--- a/common/cldmgr.c
+++ b/common/cldmgr.c
@@ -213,12 +213,8 @@ cldmgr_entry( void *arg1 )
 {
 	cld_t *cldp = ( cld_t * )arg1;
 	pid_t pid = getpid( );
-	/* REFERENCED */
-	bool_t ok;
 
 	cldp->c_pid = pid;
-	ok = qlock_thrdinit( );
-	ASSERT( ok );
 	if ( ( intgen_t )( cldp->c_streamix ) >= 0 ) {
 		stream_register( pid, ( intgen_t )cldp->c_streamix );
 	}
diff --git a/common/main.c b/common/main.c
index be8a921..25c0838 100644
--- a/common/main.c
+++ b/common/main.c
@@ -357,13 +357,6 @@ main( int argc, char *argv[] )
 		miniroot = BOOL_TRUE;
 	}
 
-	/* initialize the spinlock allocator
-	 */
-	ok = qlock_init( );
-	if ( ! ok ) {
-		return mlog_exit(EXIT_ERROR, RV_INIT);
-	}
-
 	/* initialize message logging (stage 2) - allocate the message lock
 	 */
 	ok = mlog_init2( );
diff --git a/common/qlock.c b/common/qlock.c
index adaa7dd..ae8466d 100644
--- a/common/qlock.c
+++ b/common/qlock.c
@@ -37,14 +37,6 @@ typedef struct  qlock qlock_t;
 	/* internal qlock
 	 */
 
-#define QLOCK_THRDCNTMAX			256
-	/* arbitrary limit on number of threads supported
-	 */
-
-static size_t qlock_thrdcnt;
-	/* how many threads have checked in
-	 */
-
 typedef size_t ordmap_t;
 	/* bitmap of ordinals. used to track what ordinals have
 	 * been allocated.
@@ -58,12 +50,7 @@ static ordmap_t qlock_ordalloced;
 	/* to enforce allocation of only one lock to each ordinal value
 	 */
 
-struct thrddesc {
-	pthread_t td_tid;
-	ordmap_t td_ordmap;
-};
-typedef struct thrddesc thrddesc_t;
-static thrddesc_t qlock_thrddesc[ QLOCK_THRDCNTMAX ];
+static __thread ordmap_t thread_ordmap;
 	/* holds the ordmap for each thread
 	 */
 
@@ -83,69 +70,12 @@ static thrddesc_t qlock_thrddesc[ QLOCK_THRDCNTMAX ];
 	/* checks if any bits less than ord are set in the ordmap
 	 */
 
-/* REFERENCED */
-static bool_t qlock_inited = BOOL_FALSE;
-	/* to sanity check initialization
-	 */
-
-/* forward declarations
- */
-static void qlock_ordmap_add( pthread_t tid );
-static ordmap_t *qlock_ordmapp_get( pthread_t tid );
-static ix_t qlock_thrdix_get( pthread_t tid );
-
-bool_t
-qlock_init( void )
-{
-	/* sanity checks
-	 */
-	ASSERT( ! qlock_inited );
-
-	/* initially no threads checked in
-	 */
-	qlock_thrdcnt = 0;
-
-	/* initially no ordinals allocated
-	 */
-	qlock_ordalloced = 0;
-
-	/* now say we are initialized
-	 */
-	qlock_inited = BOOL_TRUE;
-
-	/* add the parent thread to the thread list
-	 */
-	if ( ! qlock_thrdinit( )) {
-		qlock_inited = BOOL_FALSE;
-		return BOOL_FALSE;
-	}
-
-	return BOOL_TRUE;
-}
-
-bool_t
-qlock_thrdinit( void )
-{
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
-	/* add thread to ordmap list
-	 */
-	qlock_ordmap_add( pthread_self() );
-
-	return BOOL_TRUE;
-}
 
 qlockh_t
 qlock_alloc( ix_t ord )
 {
 	qlock_t *qlockp;
 
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	/* verify the ordinal is not already taken, and mark as taken
 	 */
 	ASSERT( ! QLOCK_ORDMAP_GET( qlock_ordalloced, ord ));
@@ -172,48 +102,34 @@ qlock_lock( qlockh_t qlockh )
 {
 	qlock_t *qlockp = ( qlock_t * )qlockh;
 	pthread_t tid;
-	ix_t thrdix;
-	ordmap_t *ordmapp;
 	/* REFERENCED */
 	intgen_t rval;
 	
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
-	/* get the caller's tid and thread index
+	/* get the caller's tid
 	 */
 	tid = pthread_self();
 
-	thrdix = qlock_thrdix_get( tid );
-
-	/* get the ordmap for this thread
-	 */
-	ordmapp = qlock_ordmapp_get( tid );
-
 	/* assert that this lock not already held by this thread
 	 */
-	if ( QLOCK_ORDMAP_GET( *ordmapp, qlockp->ql_ord )) {
+	if ( QLOCK_ORDMAP_GET( thread_ordmap, qlockp->ql_ord )) {
 		mlog( MLOG_NORMAL | MLOG_WARNING | MLOG_NOLOCK,
-		      _("lock already held: thrd %d tid %lu ord %d map %x\n"),
-		      thrdix,
+		      _("lock already held: tid %lu ord %d map %x\n"),
 		      tid,
 		      qlockp->ql_ord,
-		      *ordmapp );
+		      thread_ordmap );
 	}
-	ASSERT( ! QLOCK_ORDMAP_GET( *ordmapp, qlockp->ql_ord ));
+	ASSERT( ! QLOCK_ORDMAP_GET( thread_ordmap, qlockp->ql_ord ));
 
 	/* assert that no locks with a lesser ordinal are held by this thread
 	 */
-	if ( QLOCK_ORDMAP_CHK( *ordmapp, qlockp->ql_ord )) {
+	if ( QLOCK_ORDMAP_CHK( thread_ordmap, qlockp->ql_ord )) {
 		mlog( MLOG_NORMAL | MLOG_WARNING | MLOG_NOLOCK,
-		      _("lock ordinal violation: thrd %d tid %lu ord %d map %x\n"),
-		      thrdix,
+		      _("lock ordinal violation: tid %lu ord %d map %x\n"),
 		      tid,
 		      qlockp->ql_ord,
-		      *ordmapp );
+		      thread_ordmap );
 	}
-	ASSERT( ! QLOCK_ORDMAP_CHK( *ordmapp, qlockp->ql_ord ));
+	ASSERT( ! QLOCK_ORDMAP_CHK( thread_ordmap, qlockp->ql_ord ));
 
 	/* acquire the lock
 	 */
@@ -222,32 +138,23 @@ qlock_lock( qlockh_t qlockh )
 
 	/* add ordinal to this threads ordmap
 	 */
-	QLOCK_ORDMAP_SET( *ordmapp, qlockp->ql_ord );
+	QLOCK_ORDMAP_SET( thread_ordmap, qlockp->ql_ord );
 }
 
 void
 qlock_unlock( qlockh_t qlockh )
 {
 	qlock_t *qlockp = ( qlock_t * )qlockh;
-	ordmap_t *ordmapp;
 	/* REFERENCED */
 	intgen_t rval;
 	
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
-	/* get the ordmap for this thread
-	 */
-	ordmapp = qlock_ordmapp_get( pthread_self() );
-
 	/* verify lock is held by this thread
 	 */
-	ASSERT( QLOCK_ORDMAP_GET( *ordmapp, qlockp->ql_ord ));
+	ASSERT( QLOCK_ORDMAP_GET( thread_ordmap, qlockp->ql_ord ));
 
 	/* clear lock's ord from thread's ord map
 	 */
-	QLOCK_ORDMAP_CLR( *ordmapp, qlockp->ql_ord );
+	QLOCK_ORDMAP_CLR( thread_ordmap, qlockp->ql_ord );
 	
 	/* release the lock
 	 */
@@ -261,10 +168,6 @@ qsem_alloc( ix_t cnt )
 	sem_t *semp;
 	intgen_t rval;
 
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	/* allocate a semaphore
 	 */
 	semp = ( sem_t * )calloc( 1, sizeof( sem_t ));
@@ -284,10 +187,6 @@ qsem_free( qsemh_t qsemh )
 	sem_t *semp = ( sem_t * )qsemh;
 	intgen_t rval;
 
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	/* destroy the mutex and condition
 	 */
 	rval = sem_destroy( semp );
@@ -304,10 +203,6 @@ qsemP( qsemh_t qsemh )
 	sem_t *semp = ( sem_t * )qsemh;
 	intgen_t rval;
 	
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	/* "P" the semaphore
 	 */
 	rval = sem_wait( semp );
@@ -320,10 +215,6 @@ qsemV( qsemh_t qsemh )
 	sem_t *semp = ( sem_t * )qsemh;
 	intgen_t rval;
 	
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	/* "V" the semaphore
 	 */
 	rval = sem_post( semp );
@@ -337,10 +228,6 @@ qsemPwouldblock( qsemh_t qsemh )
 	int count;
 	intgen_t rval;
 
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	rval = sem_getvalue( semp, &count );
 	ASSERT( !rval );
 
@@ -354,60 +241,8 @@ qsemPavail( qsemh_t qsemh )
 	int count;
 	intgen_t rval;
 
-	/* sanity checks
-	 */
-	ASSERT( qlock_inited );
-
 	rval = sem_getvalue( semp, &count );
 	ASSERT( !rval );
 
 	return count < 0 ? 0 : count;
 }
-
-/* internal ordinal map abstraction
- */
-static void
-qlock_ordmap_add( pthread_t tid )
-{
-	ASSERT( qlock_thrdcnt < QLOCK_THRDCNTMAX );
-	qlock_thrddesc[ qlock_thrdcnt ].td_tid = tid;
-	qlock_thrddesc[ qlock_thrdcnt ].td_ordmap = 0;
-	qlock_thrdcnt++;
-}
-
-static thrddesc_t *
-qlock_thrddesc_get( pthread_t tid )
-{
-	thrddesc_t *p;
-	thrddesc_t *endp;
-
-	for ( p = &qlock_thrddesc[ 0 ],
-	      endp = &qlock_thrddesc[ qlock_thrdcnt ]
-	      ;
-	      p < endp
-	      ;
-	      p++ ) {
-		if ( pthread_equal( p->td_tid, tid ) ) {
-			return p;
-		}
-	}
-
-	return 0;
-}
-
-static ordmap_t *
-qlock_ordmapp_get( pthread_t tid )
-{
-	thrddesc_t *p;
-	p = qlock_thrddesc_get( tid );
-	return &p->td_ordmap;
-}
-
-static ix_t
-qlock_thrdix_get( pthread_t tid )
-{
-	thrddesc_t *p;
-	p = qlock_thrddesc_get( tid );
-	ASSERT( p >= &qlock_thrddesc[ 0 ] );
-	return ( ix_t )( p - &qlock_thrddesc[ 0 ] );
-}
diff --git a/common/qlock.h b/common/qlock.h
index ae411bb..6c2dd18 100644
--- a/common/qlock.h
+++ b/common/qlock.h
@@ -21,7 +21,7 @@
 /* qlock - quick locks abstraction
  *
  * threads may allocate quick locks using qlock_alloc, and free them with
- * qlock_free. the abstraction is initialized with qlock_init.
+ * qlock_free.
  *
  * deadlock detection is accomplished by giving an ordinal number to each
  * lock allocated, and record all locks held by each thread. locks may not
@@ -48,15 +48,6 @@ typedef void *qlockh_t;
 	/* opaque handle
 	 */
 
-extern bool_t qlock_init( void );
-	/* called by main to initialize abstraction. returns FALSE if
-	 * utility should abort.
-	 */
-
-extern bool_t qlock_thrdinit( void );
-	/* called by each thread to prepare it for participation
-	 */
-
 extern qlockh_t qlock_alloc( ix_t ord );
 	/* allocates a qlock with the specified ordinal. returns
 	 * NULL if lock can't be allocated.
-- 
1.7.0.4

_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs


[Index of Archives]     [Linux XFS Devel]     [Linux Filesystem Development]     [Filesystem Testing]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux