[XFS] clean up use of run_task_queue in xfs

SGI Modid: 2.5.x-xfs:slinx:133254a

fs/xfs/pagebuf/page_buf.c

@@ -1124,7 +1124,7 @@ _pagebuf_wait_unpin(
 		if (atomic_read(&PBP(pb)->pb_pin_count) == 0) {
 			break;
 		}
-		blk_run_queues();
+		pagebuf_run_task_queue(pb);
 		schedule();
 	}
 	remove_wait_queue(&PBP(pb)->pb_waiters, &wait);
@@ -1162,7 +1162,8 @@ pagebuf_iodone_work(
 
 void
 pagebuf_iodone(
-	page_buf_t		*pb)
+	page_buf_t		*pb,
+	int			schedule)
 {
 	pb->pb_flags &= ~(PBF_READ | PBF_WRITE);
 	if (pb->pb_error == 0) {
@@ -1172,8 +1173,12 @@ pagebuf_iodone(
 	PB_TRACE(pb, PB_TRACE_REC(done), pb->pb_iodone);
 
 	if ((pb->pb_iodone) || (pb->pb_flags & PBF_ASYNC)) {
+		if (schedule) {
 			INIT_WORK(&pb->pb_iodone_work, pagebuf_iodone_work, pb);
 			queue_work(pagebuf_workqueue, &pb->pb_iodone_work);
+		} else {
+			pagebuf_iodone_work(pb);
+		}
 	} else {
 		up(&pb->pb_iodonesema);
 	}
@@ -1291,9 +1296,9 @@ bio_end_io_pagebuf(
 		}
 	}
 
-	if (atomic_dec_and_test(&PBP(pb)->pb_io_remaining)) {
+	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
 		pb->pb_locked = 0;
-		pagebuf_iodone(pb);
+		pagebuf_iodone(pb, 1);
 	}
 
 	bio_put(bio);
@@ -1345,7 +1350,7 @@ pagebuf_iorequest(			/* start real I/O		*/
 	 * completion callout which happens before we have started
 	 * all the I/O from calling iodone too early
 	 */
-	atomic_set(&PBP(pb)->pb_io_remaining, 1);
+	atomic_set(&pb->pb_io_remaining, 1);
 
 	/* Special code path for reading a sub page size pagebuf in --
 	 * we populate up the whole page, and hence the other metadata
@@ -1369,7 +1374,7 @@ pagebuf_iorequest(			/* start real I/O		*/
 		bvec->bv_len = PAGE_CACHE_SIZE;
 		bvec->bv_offset = 0;
 
-		atomic_inc(&PBP(pb)->pb_io_remaining);
+		atomic_inc(&pb->pb_io_remaining);
 		submit_bio(READ, bio);
 
 		goto io_submitted;
@@ -1402,7 +1407,7 @@ pagebuf_iorequest(			/* start real I/O		*/
 	map_i = 0;
 
 next_chunk:
-	atomic_inc(&PBP(pb)->pb_io_remaining);
+	atomic_inc(&pb->pb_io_remaining);
 	nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
 	if (nr_pages > total_nr_pages)
 		nr_pages = total_nr_pages;
@@ -1444,10 +1449,8 @@ pagebuf_iorequest(			/* start real I/O		*/
 
 io_submitted:
 
-	if (atomic_dec_and_test(&PBP(pb)->pb_io_remaining) == 1) {
-		pagebuf_iodone(pb);
-	} else if ((pb->pb_flags & (PBF_SYNC|PBF_ASYNC)) == PBF_SYNC)  {
-		blk_run_queues();
+	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
+		pagebuf_iodone(pb, 0);
 	}
 
 	return status < 0 ? status : 0;
@@ -1465,7 +1468,7 @@ pagebuf_iowait(
 	page_buf_t		*pb)
 {
 	PB_TRACE(pb, PB_TRACE_REC(iowait), 0);
-	blk_run_queues();
+	pagebuf_run_task_queue(pb);
 	down(&pb->pb_iodonesema);
 	PB_TRACE(pb, PB_TRACE_REC(iowaited), (int)pb->pb_error);
 	return pb->pb_error;
@@ -1709,10 +1712,10 @@ pagebuf_daemon(
 			__pagebuf_iorequest(pb);
 		}
 
-		if (count)
-			blk_run_queues();
 		if (as_list_len > 0)
 			purge_addresses();
+		if (count)
+			pagebuf_run_task_queue(NULL);
 
 		force_flush = 0;
 	} while (pb_daemon->active == 1);
@@ -1783,7 +1786,7 @@ pagebuf_delwri_flush(
 
 	spin_unlock(&pb_daemon->pb_delwrite_lock);
 
-	blk_run_queues();
+	pagebuf_run_task_queue(NULL);
 
 	if (pinptr)
 		*pinptr = pincount;

fs/xfs/pagebuf/page_buf.h

@@ -207,6 +207,7 @@ typedef struct page_buf_s {
 	size_t			pb_count_desired; /* desired transfer size */
 	void			*pb_addr;	/* virtual address of buffer */
 	struct work_struct	pb_iodone_work;
+	atomic_t		pb_io_remaining;/* #outstanding I/O requests */
 	page_buf_iodone_t	pb_iodone;	/* I/O completion function */
 	page_buf_relse_t	pb_relse;	/* releasing function */
 	page_buf_bdstrat_t	pb_strat;	/* pre-write function */
@@ -306,7 +307,9 @@ static inline int pagebuf_geterror(page_buf_t *pb)
 }
 
 extern void pagebuf_iodone(		/* mark buffer I/O complete	*/
-		page_buf_t *);		/* buffer to mark		*/
+		page_buf_t *,		/* buffer to mark		*/
+		int);			/* run completion locally, or in
+					 * a helper thread. 		*/
 
 extern void pagebuf_ioerror(		/* mark buffer in error (or not) */
 		page_buf_t *,		/* buffer to mark		*/
@@ -375,6 +378,14 @@ static __inline__ int __pagebuf_iorequest(page_buf_t *pb)
 	return pagebuf_iorequest(pb);
 }
 
+static __inline__ void pagebuf_run_task_queue(page_buf_t *pb)
+{
+	if (pb && (atomic_read(&pb->pb_io_remaining) == 0))
+		return;
+
+	blk_run_queues();
+}
+
 extern struct workqueue_struct *pagebuf_workqueue;
 
 #endif /* __PAGE_BUF_H__ */

fs/xfs/pagebuf/page_buf_internal.h

@@ -52,7 +52,6 @@ typedef struct page_buf_private_s {
 	page_buf_t		pb_common;	/* public part of structure */
 	struct semaphore	pb_sema;	/* semaphore for lockables  */
 	unsigned long		pb_flushtime;	/* time to flush pagebuf    */
-	atomic_t		pb_io_remaining;/* #outstanding I/O requests */
 	atomic_t		pb_pin_count;	/* pin count		    */
 	wait_queue_head_t	pb_waiters;	/* unpin waiters	    */
 #ifdef PAGEBUF_LOCK_TRACKING

fs/xfs/pagebuf/page_buf_locking.c

@@ -113,8 +113,7 @@ pagebuf_lock(
 	ASSERT(pb->pb_flags & _PBF_LOCKABLE);
 
 	PB_TRACE(pb, PB_TRACE_REC(lock), 0);
-	if (atomic_read(&PBP(pb)->pb_io_remaining))
-		blk_run_queues();
+	pagebuf_run_task_queue(pb);
 	down(&PBP(pb)->pb_sema);
 	PB_SET_OWNER(pb);
 	PB_TRACE(pb, PB_TRACE_REC(locked), 0);

fs/xfs/xfs_buf.h

@@ -210,7 +210,7 @@ static inline int	xfs_bawrite(void *mp, page_buf_t *bp)
 	bp->pb_strat = xfs_bdstrat_cb;
 	xfs_buf_undelay(bp);
 	if ((ret = pagebuf_iostart(bp, PBF_WRITE | PBF_ASYNC)) == 0)
-		blk_run_queues();
+		pagebuf_run_task_queue(bp);
 	return ret;
 }
 
@@ -236,7 +236,7 @@ static inline void	xfs_buf_relse(page_buf_t *bp)
 
 
 #define xfs_biodone(pb)		    \
-	    pagebuf_iodone(pb)
+	    pagebuf_iodone(pb, 0)
 
 #define xfs_incore(buftarg,blkno,len,lockit) \
 	    pagebuf_find(buftarg, blkno ,len, lockit)
@@ -265,7 +265,7 @@ static inline int	XFS_bwrite(page_buf_t *pb)
 		error = pagebuf_iowait(pb);
 		xfs_buf_relse(pb);
 	} else {
-		blk_run_queues();
+		pagebuf_run_task_queue(pb);
 		error = 0;
 	}
 

fs/xfs/xfsidbg.c

@@ -1824,7 +1824,8 @@ kdbm_pb(int argc, const char **argv, const char **envp, struct pt_regs *regs)
 		   bp.pb_common.pb_bn,
 		   (unsigned long) bp.pb_common.pb_count_desired);
 	kdb_printf("  pb_io_remaining %d pb_error %u\n",
-		   bp.pb_io_remaining.counter, bp.pb_common.pb_error);
+		   bp.pb_common.pb_io_remaining.counter,
+		   bp.pb_common.pb_error);
 	kdb_printf("  pb_page_count %u pb_offset 0x%x pb_pages 0x%p\n",
 		bp.pb_common.pb_page_count, bp.pb_common.pb_offset,
 		bp.pb_common.pb_pages);