RE:(2) [PATCH v2] f2fs: add async reset zone command support
From: Daejun Park
Date: Wed Apr 26 2023 - 03:03:34 EST
Thank you for your review. I have responded in the comments below.
> On 2023/4/25 16:10, Daejun Park wrote:
> > Changelog:
> >
> > v1 -> v2
> > Changed to apply the optional async reset write pointer by default.
> >
> >
> > This patch enables submit reset zone command asynchornously. It helps
> > decrease average latency of write IOs in high utilization scenario by
> > faster checkpointing.
> >
> > Signed-off-by: Daejun Park <daejun7.park@xxxxxxxxxxx>
> > ---
> > fs/f2fs/segment.c | 95 ++++++++++++++++++++++++++++++++++---
> > include/trace/events/f2fs.h | 18 ++++++-
> > 2 files changed, 105 insertions(+), 8 deletions(-)
> >
> > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > index 82430f80c5da..27074a6eaf20 100644
> > --- a/fs/f2fs/segment.c
> > +++ b/fs/f2fs/segment.c
> > @@ -1195,6 +1195,46 @@ static void __init_discard_policy(struct f2fs_sb_info *sbi,
> > static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
> > struct block_device *bdev, block_t lstart,
> > block_t start, block_t len);
> > +
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > +static int __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
> > + struct discard_cmd *dc, blk_opf_t flag,
> > + struct list_head *wait_list,
> > + unsigned int *issued)
> > +{
> > + struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
> > + struct block_device *bdev = dc->bdev;
> > + struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
> > + unsigned long flags;
> > +
> > + trace_f2fs_issue_reset_zone(bdev, SECTOR_FROM_BLOCK(dc->di.start));
>
> trace_f2fs_issue_reset_zone(bdev, dc->di.start); ?
I will change it.
>
> > +
> > + spin_lock_irqsave(&dc->lock, flags);
> > + dc->state = D_SUBMIT;
> > + dc->bio_ref++;
> > + spin_unlock_irqrestore(&dc->lock, flags);
> > +
> > + if (issued)
> > + (*issued)++;
> > +
> > + atomic_inc(&dcc->queued_discard);
> > + dc->queued++;
> > + list_move_tail(&dc->list, wait_list);
> > +
> > + /* sanity check on discard range */
> > + __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
> > +
> > + bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
> > + bio->bi_private = dc;
> > + bio->bi_end_io = f2fs_submit_discard_endio;
> > + submit_bio(bio);
> > +
> > + atomic_inc(&dcc->issued_discard);
> > +
> > + return 0;
>
> Unneeded return value?
I will change return type of this function as void.
> > +}
> > +#endif
> > +
> > /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
> > static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > struct discard_policy *dpolicy,
> > @@ -1216,6 +1256,11 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
> > return 0;
> >
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > + if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
> > + return __submit_zone_reset_cmd(sbi, dc, flag, wait_list, issued);
> > +#endif
> > +
> > trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
> >
> > lstart = dc->di.lstart;
> > @@ -1460,21 +1505,42 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
> > }
> > }
> >
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > +static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
> > + struct block_device *bdev, block_t blkstart, block_t blklen)
>
> block_t lblkstart,
>
> > +{
> > + block_t lblkstart = blkstart;
>
> block_t blkstart = lblkstart;
I refactored this function as follows.
static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
struct block_device *bdev, block_t blkstart, block_t lblkstart,
block_t blklen)
{
trace_f2fs_queue_reset_zone(bdev, blkstart);
mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
}
>
> > +
> > + if (f2fs_is_multi_device(sbi)) {
> > + int devi = f2fs_target_device_index(sbi, blkstart);
> > +
> > + blkstart -= FDEV(devi).start_blk;
> > + }
> > +
> > + trace_f2fs_queue_reset_zone(bdev, blkstart);
> > +
> > + mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
> > + __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
> > + mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
> > +}
> > +#endif
> > +
> > static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
> > struct block_device *bdev, block_t blkstart, block_t blklen)
> > {
> > block_t lblkstart = blkstart;
> >
> > + trace_f2fs_queue_discard(bdev, blkstart, blklen);
>
> If device doesn't support discard, it's better not show above tracepoint?
This issue was caused by a mistake during the merge process. I have fixed it.
> > +
> > if (!f2fs_bdev_support_discard(bdev))
> > return;
> >
> > - trace_f2fs_queue_discard(bdev, blkstart, blklen);
> > -
> > if (f2fs_is_multi_device(sbi)) {
> > int devi = f2fs_target_device_index(sbi, blkstart);
> >
> > blkstart -= FDEV(devi).start_blk;
> > }
> > +
> > mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
> > __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
> > mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
> > @@ -1725,7 +1791,18 @@ static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
> > dc = __lookup_discard_cmd(sbi, blkaddr);
> > if (dc) {
> > if (dc->state == D_PREP) {
> > - __punch_discard_cmd(sbi, dc, blkaddr);
> > + if (f2fs_sb_has_blkzoned(sbi) &&
> > + bdev_is_zoned(dc->bdev)) {
> > + struct discard_policy dpolicy;
> > + /* force submit zone reset */
> > + __init_discard_policy(sbi, &dpolicy,
> > + DPOLICY_FORCE, 1);
> > + __submit_discard_cmd(sbi, &dpolicy, dc, NULL);
>
> Can we call __submit_zone_reset_cmd() directly?
Sure. I refactored as follows.
if (dc) {
if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
if (dc->state == D_PREP)
__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
&dcc->wait_list, NULL);
dc->ref++;
need_wait = true;
} else {
if (dc->state == D_PREP) {
__punch_discard_cmd(sbi, dc, blkaddr);
} else {
dc->ref++;
need_wait = true;
}
}
}
Thanks,
>
> Thanks,
>
> > + dc->ref++;
> > + need_wait = true;
> > + } else {
> > + __punch_discard_cmd(sbi, dc, blkaddr);
> > + }
> > } else {
> > dc->ref++;
> > need_wait = true;
> > @@ -1875,9 +1952,15 @@ static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
> > blkstart, blklen);
> > return -EIO;
> > }
> > - trace_f2fs_issue_reset_zone(bdev, blkstart);
> > - return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
> > - sector, nr_sects, GFP_NOFS);
> > +
> > + if (unlikely(!is_sbi_flag_set(sbi, SBI_POR_DOING))) {
> > + trace_f2fs_issue_reset_zone(bdev, blkstart);
> > + return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
> > + sector, nr_sects, GFP_NOFS);
> > + }
> > +
> > + __queue_zone_reset_cmd(sbi, bdev, lblkstart, blklen);
> > + return 0;
> > }
> >
> > /* For conventional zones, use regular discard if supported */
> > diff --git a/include/trace/events/f2fs.h b/include/trace/events/f2fs.h
> > index 99cbc5949e3c..ee1477de8324 100644
> > --- a/include/trace/events/f2fs.h
> > +++ b/include/trace/events/f2fs.h
> > @@ -1512,7 +1512,7 @@ DEFINE_EVENT(f2fs_discard, f2fs_remove_discard,
> > TP_ARGS(dev, blkstart, blklen)
> > );
> >
> > -TRACE_EVENT(f2fs_issue_reset_zone,
> > +DECLARE_EVENT_CLASS(f2fs_reset_zone,
> >
> > TP_PROTO(struct block_device *dev, block_t blkstart),
> >
> > @@ -1528,11 +1528,25 @@ TRACE_EVENT(f2fs_issue_reset_zone,
> > __entry->blkstart = blkstart;
> > ),
> >
> > - TP_printk("dev = (%d,%d), reset zone at block = 0x%llx",
> > + TP_printk("dev = (%d,%d), zone at block = 0x%llx",
> > show_dev(__entry->dev),
> > (unsigned long long)__entry->blkstart)
> > );
> >
> > +DEFINE_EVENT(f2fs_reset_zone, f2fs_queue_reset_zone,
> > +
> > + TP_PROTO(struct block_device *dev, block_t blkstart),
> > +
> > + TP_ARGS(dev, blkstart)
> > +);
> > +
> > +DEFINE_EVENT(f2fs_reset_zone, f2fs_issue_reset_zone,
> > +
> > + TP_PROTO(struct block_device *dev, block_t blkstart),
> > +
> > + TP_ARGS(dev, blkstart)
> > +);
> > +
> > TRACE_EVENT(f2fs_issue_flush,
> >
> > TP_PROTO(struct block_device *dev, unsigned int nobarrier,