Instead of waiting for each discard request keep the CIL context alive until all of them are done, at which point we can tear it down completly and remove the busy extents from the rbtree. Signed-off-by: Christoph Hellwig Index: xfs/fs/xfs/linux-2.6/xfs_discard.c =================================================================== --- xfs.orig/fs/xfs/linux-2.6/xfs_discard.c 2011-05-03 19:43:13.467745055 +0200 +++ xfs/fs/xfs/linux-2.6/xfs_discard.c 2011-05-03 19:43:14.514406051 +0200 @@ -30,6 +30,7 @@ #include "xfs_inode.h" #include "xfs_alloc.h" #include "xfs_error.h" +#include "xfs_log_priv.h" #include "xfs_discard.h" #include "xfs_trace.h" @@ -192,31 +193,88 @@ xfs_ioc_trim( return 0; } -int -xfs_discard_extents( - struct xfs_mount *mp, - struct list_head *list) +STATIC void +xfs_discard_end_io( + struct bio *bio, + int err) { - struct xfs_busy_extent *busyp; - int error = 0; + struct xfs_cil_ctx *ctx = bio->bi_private; - list_for_each_entry(busyp, list, list) { - trace_xfs_discard_extent(mp, busyp->agno, busyp->bno, - busyp->length); - - error = -blkdev_issue_discard(mp->m_ddev_targp->bt_bdev, - XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno), - XFS_FSB_TO_BB(mp, busyp->length), - GFP_NOFS, 0); - if (error && error != EOPNOTSUPP) { - xfs_info(mp, - "discard failed for extent [0x%llu,%u], error %d", - (unsigned long long)busyp->bno, - busyp->length, - error); - return error; + if (err && err != EOPNOTSUPP) { + xfs_info(ctx->cil->xc_log->l_mp, + "discard failed at sector 0x%llu, error %d", + (unsigned long long)bio->bi_sector, err); + } + + if (atomic_dec_and_test(&ctx->ref)) + queue_work(xfs_discard_workqueue, &ctx->work); + bio_put(bio); +} + +STATIC int +xfs_issue_discard( + struct xfs_cil_ctx *ctx, + struct xfs_busy_extent *busyp) +{ + struct xfs_mount *mp = ctx->cil->xc_log->l_mp; + struct block_device *bdev = mp->m_ddev_targp->bt_bdev; + struct request_queue *q = bdev_get_queue(bdev); + unsigned int max_discard_sectors; + struct bio *bio; + sector_t sector; + sector_t nr_sects; + + if (!blk_queue_discard(q)) + return -EOPNOTSUPP; + + trace_xfs_discard_extent(mp, busyp->agno, busyp->bno, busyp->length); + + sector = XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno); + nr_sects = XFS_FSB_TO_BB(mp, busyp->length); + + /* + * Ensure that max_discard_sectors is of the proper granularity + */ + max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9); + if (q->limits.discard_granularity) { + unsigned int disc_sects = q->limits.discard_granularity >> 9; + + max_discard_sectors &= ~(disc_sects - 1); + } + + while (nr_sects) { + bio = bio_alloc(GFP_NOFS, 1); + if (!bio) + return -ENOMEM; + + bio->bi_sector = sector; + bio->bi_end_io = xfs_discard_end_io; + bio->bi_bdev = bdev; + bio->bi_private = ctx; + + if (nr_sects > max_discard_sectors) { + bio->bi_size = max_discard_sectors << 9; + nr_sects -= max_discard_sectors; + sector += max_discard_sectors; + } else { + bio->bi_size = nr_sects << 9; + nr_sects = 0; } + + atomic_inc(&ctx->ref); + submit_bio(REQ_WRITE | REQ_DISCARD, bio); } return 0; } + +int +xfs_discard_extents( + struct xfs_cil_ctx *ctx) +{ + struct xfs_busy_extent *busyp; + + list_for_each_entry(busyp, &ctx->busy_extents, list) + xfs_issue_discard(ctx, busyp); + return 0; +} Index: xfs/fs/xfs/linux-2.6/xfs_buf.c =================================================================== --- xfs.orig/fs/xfs/linux-2.6/xfs_buf.c 2011-05-03 19:43:07.164445869 +0200 +++ xfs/fs/xfs/linux-2.6/xfs_buf.c 2011-05-03 19:43:14.517739367 +0200 @@ -48,6 +48,7 @@ STATIC void xfs_buf_delwri_queue(xfs_buf static struct workqueue_struct *xfslogd_workqueue; struct workqueue_struct *xfsdatad_workqueue; struct workqueue_struct *xfsconvertd_workqueue; +struct workqueue_struct *xfs_discard_workqueue; #ifdef XFS_BUF_LOCK_TRACKING # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) @@ -1785,6 +1786,7 @@ xfs_flush_buftarg( LIST_HEAD(wait_list); struct blk_plug plug; + xfs_buf_runall_queues(xfs_discard_workqueue); xfs_buf_runall_queues(xfsconvertd_workqueue); xfs_buf_runall_queues(xfsdatad_workqueue); xfs_buf_runall_queues(xfslogd_workqueue); @@ -1848,8 +1850,15 @@ xfs_buf_init(void) if (!xfsconvertd_workqueue) goto out_destroy_xfsdatad_workqueue; + xfs_discard_workqueue = alloc_workqueue("xfs_discard", + WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); + if (!xfs_discard_workqueue) + goto out_destroy_xfsconvertd_workqueue; + return 0; + out_destroy_xfsconvertd_workqueue: + destroy_workqueue(xfsconvertd_workqueue); out_destroy_xfsdatad_workqueue: destroy_workqueue(xfsdatad_workqueue); out_destroy_xfslogd_workqueue: @@ -1863,6 +1872,7 @@ xfs_buf_init(void) void xfs_buf_terminate(void) { + destroy_workqueue(xfs_discard_workqueue); destroy_workqueue(xfsconvertd_workqueue); destroy_workqueue(xfsdatad_workqueue); destroy_workqueue(xfslogd_workqueue); Index: xfs/fs/xfs/linux-2.6/xfs_buf.h =================================================================== --- xfs.orig/fs/xfs/linux-2.6/xfs_buf.h 2011-05-03 19:43:07.181112445 +0200 +++ xfs/fs/xfs/linux-2.6/xfs_buf.h 2011-05-03 19:43:14.517739367 +0200 @@ -348,4 +348,6 @@ extern struct list_head *xfs_get_buftarg #define xfs_binval(buftarg) xfs_flush_buftarg(buftarg, 1) #define XFS_bflush(buftarg) xfs_flush_buftarg(buftarg, 1) +extern struct workqueue_struct *xfs_discard_workqueue; + #endif /* __XFS_BUF_H__ */ Index: xfs/fs/xfs/xfs_alloc.c =================================================================== --- xfs.orig/fs/xfs/xfs_alloc.c 2011-05-03 19:43:13.967742345 +0200 +++ xfs/fs/xfs/xfs_alloc.c 2011-05-03 19:43:14.517739367 +0200 @@ -1084,6 +1084,7 @@ restart: if (!forced++) { trace_xfs_alloc_near_busy(args); xfs_log_force(args->mp, XFS_LOG_SYNC); + flush_workqueue(xfs_discard_workqueue); goto restart; } @@ -1243,8 +1244,10 @@ restart: xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); trace_xfs_alloc_size_busy(args); - if (!forced++) + if (!forced++) { xfs_log_force(args->mp, XFS_LOG_SYNC); + flush_workqueue(xfs_discard_workqueue); + } goto restart; } } @@ -1314,6 +1317,7 @@ restart: xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); trace_xfs_alloc_size_busy(args); xfs_log_force(args->mp, XFS_LOG_SYNC); + flush_workqueue(xfs_discard_workqueue); goto restart; } goto out_nominleft; @@ -2612,13 +2616,13 @@ xfs_alloc_busy_update_extent( xfs_agblock_t bend = bbno + busyp->length; /* - * This extent is currently beeing discard. Give the thread - * performing the discard a chance to mark the extent unbusy - * and retry. + * This extent is currently beeing discard. Flush the discard + * completion queue and retry the search. */ if (busyp->flags & XFS_ALLOC_BUSY_DISCARDED) { spin_unlock(&pag->pagb_lock); - delay(1); + flush_workqueue(xfs_discard_workqueue); + trace_xfs_alloc_busy_discarded(mp, pag->pag_agno, fbno, flen); spin_lock(&pag->pagb_lock); return false; } Index: xfs/fs/xfs/linux-2.6/xfs_trace.h =================================================================== --- xfs.orig/fs/xfs/linux-2.6/xfs_trace.h 2011-05-03 19:43:07.194445707 +0200 +++ xfs/fs/xfs/linux-2.6/xfs_trace.h 2011-05-03 19:43:14.521072682 +0200 @@ -1183,6 +1183,7 @@ DEFINE_BUSY_EVENT(xfs_alloc_busy_enomem) DEFINE_BUSY_EVENT(xfs_alloc_busy_force); DEFINE_BUSY_EVENT(xfs_alloc_busy_reuse); DEFINE_BUSY_EVENT(xfs_alloc_busy_clear); +DEFINE_BUSY_EVENT(xfs_alloc_busy_discarded); TRACE_EVENT(xfs_alloc_busy_trim, TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, Index: xfs/fs/xfs/linux-2.6/xfs_discard.h =================================================================== --- xfs.orig/fs/xfs/linux-2.6/xfs_discard.h 2011-05-03 19:43:13.467745055 +0200 +++ xfs/fs/xfs/linux-2.6/xfs_discard.h 2011-05-03 19:43:14.521072682 +0200 @@ -2,9 +2,9 @@ #define XFS_DISCARD_H 1 struct fstrim_range; -struct list_head; +struct xfs_cil_ctx; extern int xfs_ioc_trim(struct xfs_mount *, struct fstrim_range __user *); -extern int xfs_discard_extents(struct xfs_mount *, struct list_head *); +extern int xfs_discard_extents(struct xfs_cil_ctx *); #endif /* XFS_DISCARD_H */ Index: xfs/fs/xfs/xfs_log_cil.c =================================================================== --- xfs.orig/fs/xfs/xfs_log_cil.c 2011-05-03 19:43:14.214407676 +0200 +++ xfs/fs/xfs/xfs_log_cil.c 2011-05-03 19:43:14.521072682 +0200 @@ -415,7 +415,7 @@ xlog_cil_committed( if (!list_empty(&ctx->busy_extents)) { ASSERT(mp->m_flags & XFS_MOUNT_DISCARD); - xfs_discard_extents(mp, &ctx->busy_extents); + xfs_discard_extents(ctx); } if (atomic_dec_and_test(&ctx->ref)) _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs