On Tue, Mar 12, 2024 at 07:11:01PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@xxxxxxxxxx>
>
> Amortize the cost of indirect calls further by loading a batch of
> records into a new btree block instead of one record per ->get_record
> call. On a rmap btree with 3.9 million records, this reduces the
> runtime of xfs_btree_bload by 3% for xfsprogs. For the upcoming online
> repair functionality, this will reduce runtime by 6% when spectre
> mitigations are enabled in the kernel.
>
> Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx>
Reviewed-by: Bill O'Donnell <bodonnel@xxxxxxxxxx>
> ---
> repair/agbtree.c | 161 ++++++++++++++++++++++++++++++------------------------
> 1 file changed, 90 insertions(+), 71 deletions(-)
>
>
> diff --git a/repair/agbtree.c b/repair/agbtree.c
> index 981d8e340bf2..e014e216e0a5 100644
> --- a/repair/agbtree.c
> +++ b/repair/agbtree.c
> @@ -220,15 +220,19 @@ get_bnobt_records(
> struct bt_rebuild *btr = priv;
> struct xfs_alloc_rec_incore *arec = &cur->bc_rec.a;
> union xfs_btree_rec *block_rec;
> + unsigned int loaded;
>
> - btr->bno_rec = get_bno_rec(cur, btr->bno_rec);
> - arec->ar_startblock = btr->bno_rec->ex_startblock;
> - arec->ar_blockcount = btr->bno_rec->ex_blockcount;
> - btr->freeblks += btr->bno_rec->ex_blockcount;
> + for (loaded = 0; loaded < nr_wanted; loaded++, idx++) {
> + btr->bno_rec = get_bno_rec(cur, btr->bno_rec);
> + arec->ar_startblock = btr->bno_rec->ex_startblock;
> + arec->ar_blockcount = btr->bno_rec->ex_blockcount;
> + btr->freeblks += btr->bno_rec->ex_blockcount;
>
> - block_rec = libxfs_btree_rec_addr(cur, idx, block);
> - cur->bc_ops->init_rec_from_cur(cur, block_rec);
> - return 1;
> + block_rec = libxfs_btree_rec_addr(cur, idx, block);
> + cur->bc_ops->init_rec_from_cur(cur, block_rec);
> + }
> +
> + return loaded;
> }
>
> void
> @@ -388,65 +392,72 @@ get_inobt_records(
> {
> struct bt_rebuild *btr = priv;
> struct xfs_inobt_rec_incore *irec = &cur->bc_rec.i;
> - struct ino_tree_node *ino_rec;
> - union xfs_btree_rec *block_rec;
> - int inocnt = 0;
> - int finocnt = 0;
> - int k;
> -
> - btr->ino_rec = ino_rec = get_ino_rec(cur, btr->ino_rec);
> -
> - /* Transform the incore record into an on-disk record. */
> - irec->ir_startino = ino_rec->ino_startnum;
> - irec->ir_free = ino_rec->ir_free;
> -
> - for (k = 0; k < sizeof(xfs_inofree_t) * NBBY; k++) {
> - ASSERT(is_inode_confirmed(ino_rec, k));
> -
> - if (is_inode_sparse(ino_rec, k))
> - continue;
> - if (is_inode_free(ino_rec, k))
> - finocnt++;
> - inocnt++;
> - }
> + unsigned int loaded = 0;
> +
> + while (loaded < nr_wanted) {
> + struct ino_tree_node *ino_rec;
> + union xfs_btree_rec *block_rec;
> + int inocnt = 0;
> + int finocnt = 0;
> + int k;
> +
> + btr->ino_rec = ino_rec = get_ino_rec(cur, btr->ino_rec);
>
> - irec->ir_count = inocnt;
> - irec->ir_freecount = finocnt;
> -
> - if (xfs_has_sparseinodes(cur->bc_mp)) {
> - uint64_t sparse;
> - int spmask;
> - uint16_t holemask;
> -
> - /*
> - * Convert the 64-bit in-core sparse inode state to the
> - * 16-bit on-disk holemask.
> - */
> - holemask = 0;
> - spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> - sparse = ino_rec->ir_sparse;
> - for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> - if (sparse & spmask) {
> - ASSERT((sparse & spmask) == spmask);
> - holemask |= (1 << k);
> - } else
> - ASSERT((sparse & spmask) == 0);
> - sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> + /* Transform the incore record into an on-disk record. */
> + irec->ir_startino = ino_rec->ino_startnum;
> + irec->ir_free = ino_rec->ir_free;
> +
> + for (k = 0; k < sizeof(xfs_inofree_t) * NBBY; k++) {
> + ASSERT(is_inode_confirmed(ino_rec, k));
> +
> + if (is_inode_sparse(ino_rec, k))
> + continue;
> + if (is_inode_free(ino_rec, k))
> + finocnt++;
> + inocnt++;
> }
>
> - irec->ir_holemask = holemask;
> - } else {
> - irec->ir_holemask = 0;
> - }
> + irec->ir_count = inocnt;
> + irec->ir_freecount = finocnt;
>
> - if (btr->first_agino == NULLAGINO)
> - btr->first_agino = ino_rec->ino_startnum;
> - btr->freecount += finocnt;
> - btr->count += inocnt;
> + if (xfs_has_sparseinodes(cur->bc_mp)) {
> + uint64_t sparse;
> + int spmask;
> + uint16_t holemask;
> +
> + /*
> + * Convert the 64-bit in-core sparse inode state to the
> + * 16-bit on-disk holemask.
> + */
> + holemask = 0;
> + spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> + sparse = ino_rec->ir_sparse;
> + for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> + if (sparse & spmask) {
> + ASSERT((sparse & spmask) == spmask);
> + holemask |= (1 << k);
> + } else
> + ASSERT((sparse & spmask) == 0);
> + sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> + }
> +
> + irec->ir_holemask = holemask;
> + } else {
> + irec->ir_holemask = 0;
> + }
> +
> + if (btr->first_agino == NULLAGINO)
> + btr->first_agino = ino_rec->ino_startnum;
> + btr->freecount += finocnt;
> + btr->count += inocnt;
> +
> + block_rec = libxfs_btree_rec_addr(cur, idx, block);
> + cur->bc_ops->init_rec_from_cur(cur, block_rec);
> + loaded++;
> + idx++;
> + }
>
> - block_rec = libxfs_btree_rec_addr(cur, idx, block);
> - cur->bc_ops->init_rec_from_cur(cur, block_rec);
> - return 1;
> + return loaded;
> }
>
> /* Initialize both inode btree cursors as needed. */
> @@ -585,13 +596,17 @@ get_rmapbt_records(
> struct xfs_rmap_irec *rec;
> struct bt_rebuild *btr = priv;
> union xfs_btree_rec *block_rec;
> + unsigned int loaded;
>
> - rec = pop_slab_cursor(btr->slab_cursor);
> - memcpy(&cur->bc_rec.r, rec, sizeof(struct xfs_rmap_irec));
> + for (loaded = 0; loaded < nr_wanted; loaded++, idx++) {
> + rec = pop_slab_cursor(btr->slab_cursor);
> + memcpy(&cur->bc_rec.r, rec, sizeof(struct xfs_rmap_irec));
>
> - block_rec = libxfs_btree_rec_addr(cur, idx, block);
> - cur->bc_ops->init_rec_from_cur(cur, block_rec);
> - return 1;
> + block_rec = libxfs_btree_rec_addr(cur, idx, block);
> + cur->bc_ops->init_rec_from_cur(cur, block_rec);
> + }
> +
> + return loaded;
> }
>
> /* Set up the rmap rebuild parameters. */
> @@ -663,13 +678,17 @@ get_refcountbt_records(
> struct xfs_refcount_irec *rec;
> struct bt_rebuild *btr = priv;
> union xfs_btree_rec *block_rec;
> + unsigned int loaded;
>
> - rec = pop_slab_cursor(btr->slab_cursor);
> - memcpy(&cur->bc_rec.rc, rec, sizeof(struct xfs_refcount_irec));
> + for (loaded = 0; loaded < nr_wanted; loaded++, idx++) {
> + rec = pop_slab_cursor(btr->slab_cursor);
> + memcpy(&cur->bc_rec.rc, rec, sizeof(struct xfs_refcount_irec));
>
> - block_rec = libxfs_btree_rec_addr(cur, idx, block);
> - cur->bc_ops->init_rec_from_cur(cur, block_rec);
> - return 1;
> + block_rec = libxfs_btree_rec_addr(cur, idx, block);
> + cur->bc_ops->init_rec_from_cur(cur, block_rec);
> + }
> +
> + return loaded;
> }
>
> /* Set up the refcount rebuild parameters. */
>
>
