Re: [PATCH v3] f2fs: fix long latency due to discard during umount

From: Jaegeuk Kim
Date: Mon Apr 13 2020 - 12:52:48 EST

Next message: Arnaldo Carvalho de Melo: "[PATCH 05/26] capabilities: Introduce CAP_PERFMON to kernel and user space"
Previous message: Arnaldo Carvalho de Melo: "[PATCH 04/26] tools arch x86: Sync the msr-index.h copy with the kernel sources"
In reply to: Sahitya Tummala: "Re: [PATCH v3] f2fs: fix long latency due to discard during umount"
Next in thread: Jaegeuk Kim: "Re: [f2fs-dev] [PATCH v3] f2fs: fix long latency due to discard during umount"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 04/09, Sahitya Tummala wrote:
> On Wed, Apr 08, 2020 at 07:29:28PM -0700, Jaegeuk Kim wrote:
> > On 04/08, Sahitya Tummala wrote:
> > > Hi Jaegeuk,
> > >
> > > On Fri, Apr 03, 2020 at 10:19:43AM -0700, Jaegeuk Kim wrote:
> > > > On 04/01, Sahitya Tummala wrote:
> > > > > Hi Jaegeuk,
> > > > >
> > > > > On Tue, Mar 31, 2020 at 11:46:55AM -0700, Jaegeuk Kim wrote:
> > > > > > On 03/30, Sahitya Tummala wrote:
> > > > > > > F2FS already has a default timeout of 5 secs for discards that
> > > > > > > can be issued during umount, but it can take more than the 5 sec
> > > > > > > timeout if the underlying UFS device queue is already full and there
> > > > > > > are no more available free tags to be used. In that case, submit_bio()
> > > > > > > will wait for the already queued discard requests to complete to get
> > > > > > > a free tag, which can potentially take way more than 5 sec.
> > > > > > >
> > > > > > > Fix this by submitting the discard requests with REQ_NOWAIT
> > > > > > > flags during umount. This will return -EAGAIN for UFS queue/tag full
> > > > > > > scenario without waiting in the context of submit_bio(). The FS can
> > > > > > > then handle these requests by retrying again within the stipulated
> > > > > > > discard timeout period to avoid long latencies.
> > > > > >
> > > > > > Sorry, Sahitya, but, do we really need to do like this? How about just
> > > > > > controlling # of outstanding discarding bios in __issue_discard_cmd()?
> > > > >
> > > > > Do you mean something like this?
> > > > >
> > > > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > > > index 1a62b27..860dd43 100644
> > > > > --- a/fs/f2fs/segment.c
> > > > > +++ b/fs/f2fs/segment.c
> > > > > @@ -1099,7 +1099,7 @@ static void __init_discard_policy(struct f2fs_sb_info *sbi,
> > > > > } else if (discard_type == DPOLICY_FSTRIM) {
> > > > > dpolicy->io_aware = false;
> > > > > } else if (discard_type == DPOLICY_UMOUNT) {
> > > > > - dpolicy->max_requests = UINT_MAX;
> > > > > + dpolicy->max_requests = 30;
> > > >
> > > > Can we use max queue depth of the block device?
> > >
> > > I think it should be limited to 8 or 16 as Chao suggested, so that we can have
> > > better control on the given timeout value? Thoughts?
> >
> > Where is 8 or 16 coming from? What about SSD? Sorry, it's unclear to me.
>
> With this patch we now wait for a batch of discard requests in __issue_discard_cmd()
> with a upper timeout of 5 sec. So, I thought that having a smaller batch of
> discard requests would help us to avoid queuing more requests and end up waiting
> for more time in __wait_all_discard_cmd(). Today we have DEF_MAX_DISCARD_REQUEST
> as 8 for default max discard requests. If it too less, may be for umount
> we can use 16 instead. But the idea is to avoid more wait time.

Ok, then, it seems we can just remove this line to use dpolicy->max_requests,
and retry issuing all the discard commands like below. Later, I thnk it'd be
fine to tune the max_requests based on bd_queue->queue_depth.

>
> If you would like to propose to use max queue depth, we can get it from bdev as
> sbi->sb->s_bdev->bd_queue->queue_depth. Is it okay to use it in our FS? or you
> think it should be hardcoded to 32 (as most UFS/eMMC devices will have this as
> its queue depth)? Please let me know.
>
> Thanks,
>
> >
> > >
> > > Thanks,
> > >
> > > >
> > > > > dpolicy->io_aware = false;
> > > > > /* we need to issue all to keep CP_TRIMMED_FLAG */
> > > > > dpolicy->granularity = 1;
> > > > > @@ -1470,12 +1470,14 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > struct list_head *pend_list;
> > > > > struct discard_cmd *dc, *tmp;
> > > > > struct blk_plug plug;
> > > > > - int i, issued = 0;
> > > > > + int i, issued;
> > > > > bool io_interrupted = false;
> > > > >
> > > > > if (dpolicy->timeout != 0)
> > > > > f2fs_update_time(sbi, dpolicy->timeout);
> > > > >
> > > > > +retry:
> > > > > + issued = 0;
> > > > > for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
> > > > > if (dpolicy->timeout != 0 &&
> > > > > f2fs_time_over(sbi, dpolicy->timeout))
> > > > > @@ -1522,6 +1524,11 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > break;
> > > > > }
> > > > >
> > > > > + if (dpolicy->type == DPOLICY_UMOUNT && issued) {
> > > > > + __wait_all_discard_cmd(sbi, dpolicy);
> > > > > + goto retry;
> > > > > + }
> > > > > +
> > > > > if (!issued && io_interrupted)
> > > > > issued = -1;
> > > > >
> > > > > Thanks,
> > > > >
> > > > > >
> > > > > > >
> > > > > > > Signed-off-by: Sahitya Tummala <stummala@xxxxxxxxxxxxxx>
> > > > > > > ---
> > > > > > > v3:
> > > > > > > -Handle the regression reported by Chao with v2.
> > > > > > > -simplify the logic to split the dc with multiple bios incase any bio returns
> > > > > > > EAGAIN and retry those new dc within 5 sec timeout.
> > > > > > >
> > > > > > > fs/f2fs/segment.c | 65 +++++++++++++++++++++++++++++++++++++++++++------------
> > > > > > > 1 file changed, 51 insertions(+), 14 deletions(-)
> > > > > > >
> > > > > > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > > > > > index fb3e531..55d18c7 100644
> > > > > > > --- a/fs/f2fs/segment.c
> > > > > > > +++ b/fs/f2fs/segment.c
> > > > > > > @@ -1029,13 +1029,16 @@ static void f2fs_submit_discard_endio(struct bio *bio)
> > > > > > > struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
> > > > > > > unsigned long flags;
> > > > > > >
> > > > > > > - dc->error = blk_status_to_errno(bio->bi_status);
> > > > > > > -
> > > > > > > spin_lock_irqsave(&dc->lock, flags);
> > > > > > > + if (!dc->error)
> > > > > > > + dc->error = blk_status_to_errno(bio->bi_status);
> > > > > > > +
> > > > > > > dc->bio_ref--;
> > > > > > > - if (!dc->bio_ref && dc->state == D_SUBMIT) {
> > > > > > > - dc->state = D_DONE;
> > > > > > > - complete_all(&dc->wait);
> > > > > > > + if (!dc->bio_ref) {
> > > > > > > + if (dc->error || dc->state == D_SUBMIT) {
> > > > > > > + dc->state = D_DONE;
> > > > > > > + complete_all(&dc->wait);
> > > > > > > + }
> > > > > > > }
> > > > > > > spin_unlock_irqrestore(&dc->lock, flags);
> > > > > > > bio_put(bio);
> > > > > > > @@ -1124,10 +1127,13 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
> > > > > > > struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
> > > > > > > &(dcc->fstrim_list) : &(dcc->wait_list);
> > > > > > > - int flag = dpolicy->sync ? REQ_SYNC : 0;
> > > > > > > + int flag;
> > > > > > > block_t lstart, start, len, total_len;
> > > > > > > int err = 0;
> > > > > > >
> > > > > > > + flag = dpolicy->sync ? REQ_SYNC : 0;
> > > > > > > + flag |= dpolicy->type == DPOLICY_UMOUNT ? REQ_NOWAIT : 0;
> > > > > > > +
> > > > > > > if (dc->state != D_PREP)
> > > > > > > return 0;
> > > > > > >
> > > > > > > @@ -1192,10 +1198,6 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > dc->bio_ref++;
> > > > > > > spin_unlock_irqrestore(&dc->lock, flags);
> > > > > > >
> > > > > > > - atomic_inc(&dcc->queued_discard);
> > > > > > > - dc->queued++;
> > > > > > > - list_move_tail(&dc->list, wait_list);
> > > > > > > -
> > > > > > > /* sanity check on discard range */
> > > > > > > __check_sit_bitmap(sbi, lstart, lstart + len);
> > > > > > >
> > > > > > > @@ -1203,6 +1205,29 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > bio->bi_end_io = f2fs_submit_discard_endio;
> > > > > > > bio->bi_opf |= flag;
> > > > > > > submit_bio(bio);
> > > > > > > + if (flag & REQ_NOWAIT) {
> > > > > > > + if (dc->error == -EAGAIN) {
> > > > > > > + spin_lock_irqsave(&dc->lock, flags);
> > > > > > > + dc->len -= len;
> > > > > > > + if (!dc->len) {
> > > > > > > + dc->len = total_len;
> > > > > > > + dc->state = D_PREP;
> > > > > > > + reinit_completion(&dc->wait);
> > > > > > > + } else {
> > > > > > > + dcc->undiscard_blks -= total_len;
> > > > > > > + if (dc->state == D_PARTIAL)
> > > > > > > + dc->state = D_SUBMIT;
> > > > > > > + }
> > > > > > > + err = dc->error;
> > > > > > > + dc->error = 0;
> > > > > > > + spin_unlock_irqrestore(&dc->lock, flags);
> > > > > > > + break;
> > > > > > > + }
> > > > > > > + }
> > > > > > > +
> > > > > > > + atomic_inc(&dcc->queued_discard);
> > > > > > > + dc->queued++;
> > > > > > > + list_move_tail(&dc->list, wait_list);
> > > > > > >
> > > > > > > atomic_inc(&dcc->issued_discard);
> > > > > > >
> > > > > > > @@ -1214,8 +1239,9 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > len = total_len;
> > > > > > > }
> > > > > > >
> > > > > > > - if (!err && len)
> > > > > > > - __update_discard_tree_range(sbi, bdev, lstart, start, len);
> > > > > > > + if ((!err || err == -EAGAIN) && total_len && dc->start != start)
> > > > > > > + __update_discard_tree_range(sbi, bdev, lstart, start,
> > > > > > > + total_len);
> > > > > > > return err;
> > > > > > > }
> > > > > > >
> > > > > > > @@ -1470,12 +1496,15 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > struct list_head *pend_list;
> > > > > > > struct discard_cmd *dc, *tmp;
> > > > > > > struct blk_plug plug;
> > > > > > > - int i, issued = 0;
> > > > > > > + int i, err, issued = 0;
> > > > > > > bool io_interrupted = false;
> > > > > > > + bool retry;
> > > > > > >
> > > > > > > if (dpolicy->timeout != 0)
> > > > > > > f2fs_update_time(sbi, dpolicy->timeout);
> > > > > > >
> > > > > > > +retry:
> > > > > > > + retry = false;
> > > > > > > for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
> > > > > > > if (dpolicy->timeout != 0 &&
> > > > > > > f2fs_time_over(sbi, dpolicy->timeout))
> > > > > > > @@ -1509,7 +1538,12 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > break;
> > > > > > > }
> > > > > > >
> > > > > > > - __submit_discard_cmd(sbi, dpolicy, dc, &issued);
> > > > > > > + err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
> > > > > > > + if (err == -EAGAIN) {
> > > > > > > + congestion_wait(BLK_RW_ASYNC,
> > > > > > > + DEFAULT_IO_TIMEOUT);
> > > > > > > + retry = true;
> > > > > > > + }
> > > > > > >
> > > > > > > if (issued >= dpolicy->max_requests)
> > > > > > > break;
> > > > > > > @@ -1522,6 +1556,9 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> > > > > > > break;
> > > > > > > }
> > > > > > >
> > > > > > > + if (retry)
> > > > > > > + goto retry;
> > > > > > > +
> > > > > > > if (!issued && io_interrupted)
> > > > > > > issued = -1;
> > > > > > >
> > > > > > > --
> > > > > > > Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.
> > > > > > > Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.
> > > > >
> > > > > --
> > > > > --
> > > > > Sent by a consultant of the Qualcomm Innovation Center, Inc.
> > > > > The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.
> > >
> > > --
> > > --
> > > Sent by a consultant of the Qualcomm Innovation Center, Inc.
> > > The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.
>
> --
> --
> Sent by a consultant of the Qualcomm Innovation Center, Inc.
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

Next message: Arnaldo Carvalho de Melo: "[PATCH 05/26] capabilities: Introduce CAP_PERFMON to kernel and user space"
Previous message: Arnaldo Carvalho de Melo: "[PATCH 04/26] tools arch x86: Sync the msr-index.h copy with the kernel sources"
In reply to: Sahitya Tummala: "Re: [PATCH v3] f2fs: fix long latency due to discard during umount"
Next in thread: Jaegeuk Kim: "Re: [f2fs-dev] [PATCH v3] f2fs: fix long latency due to discard during umount"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]