On 01/05, Chao Yu wrote:
> On 2017/1/4 17:29, Chao Yu wrote:
> > On 2016/12/31 2:51, Jaegeuk Kim wrote:
> >> This patch relaxes async discard commands to avoid waiting its end_io
> >> during
> >> checkpoint.
> >> Instead of waiting them during checkpoint, it will be done when actually
> >> reusing
> >> them.
> >>
> >> Test on initial partition of nvme drive.
> >>
> >> # time fstrim /mnt/test
> >>
> >> Before : 6.158s
> >> After : 4.822s
> >>
> >> Signed-off-by: Jaegeuk Kim <jaeg...@kernel.org>
> >
> > Reviewed-by: Chao Yu <yuch...@huawei.com>
> >
> > One comment below,
>
> I still have a comment on this patch.
>
> >> -void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
> >> +/* This should be covered by global mutex, &sit_i->sentry_lock */
> >> +void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, unsigned int segno)
> >> {
> >> struct list_head *wait_list = &(SM_I(sbi)->wait_list);
> >> struct bio_entry *be, *tmp;
> >> @@ -646,7 +650,15 @@ void f2fs_wait_all_discard_bio(struct f2fs_sb_info
> >> *sbi)
> >> struct bio *bio = be->bio;
> >> int err;
> >>
> >> - wait_for_completion_io(&be->event);
> >> + if (!completion_done(&be->event)) {
> >> + if ((be->start_segno >= segno &&
> >> + be->end_segno <= segno) ||
> >
> > segno >= be->start_segno && segno < be->end_segno ?
>
> Can you check this?
The be->start_segno and be->end_segno are assigned by:
unsigned int start_segno = GET_SEGNO(sbi, blkstart);
unsigned int end_segno = GET_SEGNO(sbi, blkstart + blklen);
in __f2fs_issue_discard_async().
So, the problem comes when, for example, blkstart = 0 and blklen = 512.
I should have got the end_segno by:
unsigned int end_segno = GET_SEGNO(sbi, blkstart + blklen - 1);
Thanks,
