From: "Darrick J. Wong" <djwong@kernel.org>
To: Christoph Hellwig <hch@lst.de>
Cc: Chandan Babu R <chandan.babu@oracle.com>, linux-xfs@vger.kernel.org
Subject: Re: [PATCH 20/22] xfs: simplify and optimize the RT allocation fallback cascade
Date: Mon, 18 Dec 2023 14:17:32 -0800 [thread overview]
Message-ID: <20231218221732.GV361584@frogsfrogsfrogs> (raw)
In-Reply-To: <20231218045738.711465-21-hch@lst.de>
On Mon, Dec 18, 2023 at 05:57:36AM +0100, Christoph Hellwig wrote:
> There are currently multiple levels of fall back if an RT allocation
> can not be satisfied:
>
> 1) xfs_rtallocate_extent extends the minlen and reduces the maxlen due
> to the extent size hint. If that can't be done, it return -ENOSPC
> and let's xfs_bmap_rtalloc retry, which then not only drops the
> extent size hint based alignment, but also the minlen adjustment
> 2) if xfs_rtallocate_extent gets -ENOSPC from the underlying functions,
> it only drops the extent size hint based alignment and retries
> 3) if that still does not succeed, xfs_rtallocate_extent drops the
> extent size hint (which is a complex no-op at this point) and the
> minlen using the same code as (1) above
> 4) if that still doesn't success and the caller wanted an allocation
> near a blkno, drop that blkno hint.
>
> The handling in 1 is rather inefficient as we could just drop the
> alignment and continue, and 2/3 interact in really weird ways due to
> the duplicate policy.
>
> Move aligning the min and maxlen out of xfs_rtallocate_extent and into
> a helper called directly by xfs_bmap_rtalloc. This allows just
> continuing with the allocation if we have to drop the alignment instead
> of going through the retry loop and also dropping the perfectly usable
> minlen adjustment that didn't cause the problem, and then just use
> a single retry that drops both the minlen and alignment requirement
> when we really are out of space, thus consolidating cases (2) and (3)
> above.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
Looks good now,
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
--D
> ---
> fs/xfs/xfs_rtalloc.c | 58 ++++++++++++++++++++++++++------------------
> 1 file changed, 35 insertions(+), 23 deletions(-)
>
> diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c
> index bac8eacd628c29..8a09e42b2dcdcc 100644
> --- a/fs/xfs/xfs_rtalloc.c
> +++ b/fs/xfs/xfs_rtalloc.c
> @@ -1089,21 +1089,6 @@ xfs_rtallocate_extent(
> ASSERT(xfs_isilocked(args.mp->m_rbmip, XFS_ILOCK_EXCL));
> ASSERT(minlen > 0 && minlen <= maxlen);
>
> - /*
> - * If prod is set then figure out what to do to minlen and maxlen.
> - */
> - if (prod > 1) {
> - xfs_rtxlen_t i;
> -
> - if ((i = maxlen % prod))
> - maxlen -= i;
> - if ((i = minlen % prod))
> - minlen += prod - i;
> - if (maxlen < minlen)
> - return -ENOSPC;
> - }
> -
> -retry:
> if (start == 0) {
> error = xfs_rtallocate_extent_size(&args, minlen,
> maxlen, len, prod, rtx);
> @@ -1112,13 +1097,8 @@ xfs_rtallocate_extent(
> maxlen, len, prod, rtx);
> }
> xfs_rtbuf_cache_relse(&args);
> - if (error) {
> - if (error == -ENOSPC && prod > 1) {
> - prod = 1;
> - goto retry;
> - }
> + if (error)
> return error;
> - }
>
> /*
> * If it worked, update the superblock.
> @@ -1349,6 +1329,35 @@ xfs_rtpick_extent(
> return 0;
> }
>
> +static void
> +xfs_rtalloc_align_minmax(
> + xfs_rtxlen_t *raminlen,
> + xfs_rtxlen_t *ramaxlen,
> + xfs_rtxlen_t *prod)
> +{
> + xfs_rtxlen_t newmaxlen = *ramaxlen;
> + xfs_rtxlen_t newminlen = *raminlen;
> + xfs_rtxlen_t slack;
> +
> + slack = newmaxlen % *prod;
> + if (slack)
> + newmaxlen -= slack;
> + slack = newminlen % *prod;
> + if (slack)
> + newminlen += *prod - slack;
> +
> + /*
> + * If adjusting for extent size hint alignment produces an invalid
> + * min/max len combination, go ahead without it.
> + */
> + if (newmaxlen < newminlen) {
> + *prod = 1;
> + return;
> + }
> + *ramaxlen = newmaxlen;
> + *raminlen = newminlen;
> +}
> +
> int
> xfs_bmap_rtalloc(
> struct xfs_bmalloca *ap)
> @@ -1431,10 +1440,13 @@ xfs_bmap_rtalloc(
> * perfectly aligned, otherwise it will just get us in trouble.
> */
> div_u64_rem(ap->offset, align, &mod);
> - if (mod || ap->length % align)
> + if (mod || ap->length % align) {
> prod = 1;
> - else
> + } else {
> prod = xfs_extlen_to_rtxlen(mp, align);
> + if (prod > 1)
> + xfs_rtalloc_align_minmax(&raminlen, &ralen, &prod);
> + }
>
> error = xfs_rtallocate_extent(ap->tp, start, raminlen, ralen, &ralen,
> ap->wasdel, prod, &rtx);
> --
> 2.39.2
>
>
next prev parent reply other threads:[~2023-12-18 22:17 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-12-18 4:57 RT allocator tidy ups v2 Christoph Hellwig
2023-12-18 4:57 ` [PATCH 01/22] xfs: consider minlen sized extents in xfs_rtallocate_extent_block Christoph Hellwig
2023-12-18 4:57 ` [PATCH 02/22] xfs: turn the xfs_trans_mod_dquot_byino stub into an inline function Christoph Hellwig
2023-12-18 4:57 ` [PATCH 03/22] xfs: remove the xfs_alloc_arg argument to xfs_bmap_btalloc_accounting Christoph Hellwig
2023-12-18 4:57 ` [PATCH 04/22] xfs: also use xfs_bmap_btalloc_accounting for RT allocations Christoph Hellwig
2023-12-18 4:57 ` [PATCH 05/22] xfs: move xfs_bmap_rtalloc to xfs_rtalloc.c Christoph Hellwig
2023-12-18 4:57 ` [PATCH 06/22] xfs: return -ENOSPC from xfs_rtallocate_* Christoph Hellwig
2023-12-18 4:57 ` [PATCH 07/22] xfs: reflow the tail end of xfs_bmap_rtalloc Christoph Hellwig
2023-12-18 4:57 ` [PATCH 08/22] xfs: indicate if xfs_bmap_adjacent changed ap->blkno Christoph Hellwig
2023-12-18 4:57 ` [PATCH 09/22] xfs: cleanup picking the start extent hint in xfs_bmap_rtalloc Christoph Hellwig
2023-12-18 4:57 ` [PATCH 10/22] xfs: move xfs_rtget_summary to xfs_rtbitmap.c Christoph Hellwig
2023-12-18 4:57 ` [PATCH 11/22] xfs: split xfs_rtmodify_summary_int Christoph Hellwig
2023-12-18 4:57 ` [PATCH 12/22] xfs: invert a check in xfs_rtallocate_extent_block Christoph Hellwig
2023-12-18 17:50 ` Darrick J. Wong
2023-12-18 4:57 ` [PATCH 13/22] xfs: reflow the tail end of xfs_rtallocate_extent_block Christoph Hellwig
2023-12-18 17:51 ` Darrick J. Wong
2023-12-18 4:57 ` [PATCH 14/22] xfs: merge the calls to xfs_rtallocate_range in xfs_rtallocate_block Christoph Hellwig
2023-12-18 17:52 ` Darrick J. Wong
2023-12-18 4:57 ` [PATCH 15/22] xfs: tidy up xfs_rtallocate_extent_exact Christoph Hellwig
2023-12-18 4:57 ` [PATCH 16/22] xfs: factor out a xfs_rtalloc_sumlevel helper Christoph Hellwig
2023-12-18 4:57 ` [PATCH 17/22] xfs: remove rt-wrappers from xfs_format.h Christoph Hellwig
2023-12-18 4:57 ` [PATCH 18/22] xfs: remove XFS_RTMIN/XFS_RTMAX Christoph Hellwig
2023-12-18 4:57 ` [PATCH 19/22] xfs: reorder the minlen and prod calculations in xfs_bmap_rtalloc Christoph Hellwig
2023-12-18 4:57 ` [PATCH 20/22] xfs: simplify and optimize the RT allocation fallback cascade Christoph Hellwig
2023-12-18 22:17 ` Darrick J. Wong [this message]
2023-12-18 4:57 ` [PATCH 21/22] xfs: fold xfs_rtallocate_extent into xfs_bmap_rtalloc Christoph Hellwig
2023-12-18 4:57 ` [PATCH 22/22] xfs: rename xfs_bmap_rtalloc to xfs_rtallocate_extent Christoph Hellwig
2023-12-18 22:24 ` Darrick J. Wong
2023-12-19 4:17 ` Christoph Hellwig
2023-12-19 4:51 ` Darrick J. Wong
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20231218221732.GV361584@frogsfrogsfrogs \
--to=djwong@kernel.org \
--cc=chandan.babu@oracle.com \
--cc=hch@lst.de \
--cc=linux-xfs@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox