From: "Darrick J. Wong" <djwong@kernel.org>
To: Omar Sandoval <osandov@osandov.com>
Cc: linux-xfs@vger.kernel.org, kernel-team@fb.com,
Prashant Nema <pnema@fb.com>
Subject: Re: [PATCH 2/6] xfs: invert the realtime summary cache
Date: Wed, 12 Jul 2023 15:40:01 -0700 [thread overview]
Message-ID: <20230712224001.GV108251@frogsfrogsfrogs> (raw)
In-Reply-To: <e3ae5bfc7cd4b640e83a25f001169d4ae50d797a.1687296675.git.osandov@osandov.com>
On Tue, Jun 20, 2023 at 02:32:12PM -0700, Omar Sandoval wrote:
> From: Omar Sandoval <osandov@fb.com>
>
> In commit 355e3532132b ("xfs: cache minimum realtime summary level"), I
> added a cache of the minimum level of the realtime summary that has any
> free extents. However, it turns out that the _maximum_ level is more
> useful for upcoming optimizations, and basically equivalent for the
> existing usage. So, let's change the meaning of the cache to be the
> maximum level + 1, or 0 if there are no free extents.
Hmm. If I'm reading xfs_rtmodify_summary_int right, m_rsum_cache[b] now
tells us the maximum log2(length) of the free extents starting in
rtbitmap block b?
IOWs, let's say the cache contents are:
{2, 3, 2, 15, 8}
Someone asks for a 400rtx (realtime extent) allocation, so we want to
find a free space of at least magnitude floor(log2(400)) == 8.
The cache tells us that there aren't any free extents longer than 2^1
blocks in rtbitmap blocks 0 and 2; longer than 2^2 blocks in rtbmp block
1; longer than 2^7 blocks in rtbmp block 4; nor longer than 2^14 blocks
in rtbmp block 3?
From the cache contents, we should therefore examine rtbitmap block 3.
If the cache contents were instead:
{2, 3, 2, 8, 8}
Then we instead might scan rtbitmap blocks 3 and 4 for the longest
allocation that we can get? Looking back at the original commit, that
seems to make more sense to me...
> Signed-off-by: Omar Sandoval <osandov@fb.com>
> ---
> fs/xfs/libxfs/xfs_rtbitmap.c | 6 +++---
> fs/xfs/xfs_mount.h | 6 +++---
> fs/xfs/xfs_rtalloc.c | 31 +++++++++++++++++++------------
> 3 files changed, 25 insertions(+), 18 deletions(-)
>
> diff --git a/fs/xfs/libxfs/xfs_rtbitmap.c b/fs/xfs/libxfs/xfs_rtbitmap.c
> index 1a832c9a412f..d9493f64adfc 100644
> --- a/fs/xfs/libxfs/xfs_rtbitmap.c
> +++ b/fs/xfs/libxfs/xfs_rtbitmap.c
> @@ -503,10 +503,10 @@ xfs_rtmodify_summary_int(
>
> *sp += delta;
> if (mp->m_rsum_cache) {
> - if (*sp == 0 && log == mp->m_rsum_cache[bbno])
> - mp->m_rsum_cache[bbno]++;
> - if (*sp != 0 && log < mp->m_rsum_cache[bbno])
> + if (*sp == 0 && log + 1 == mp->m_rsum_cache[bbno])
> mp->m_rsum_cache[bbno] = log;
> + if (*sp != 0 && log >= mp->m_rsum_cache[bbno])
> + mp->m_rsum_cache[bbno] = log + 1;
> }
> xfs_trans_log_buf(tp, bp, first, first + sizeof(*sp) - 1);
> }
> diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
> index 6c09f89534d3..964541c36730 100644
> --- a/fs/xfs/xfs_mount.h
> +++ b/fs/xfs/xfs_mount.h
> @@ -103,9 +103,9 @@ typedef struct xfs_mount {
>
> /*
> * Optional cache of rt summary level per bitmap block with the
> - * invariant that m_rsum_cache[bbno] <= the minimum i for which
> - * rsum[i][bbno] != 0. Reads and writes are serialized by the rsumip
> - * inode lock.
> + * invariant that m_rsum_cache[bbno] > the maximum i for which
> + * rsum[i][bbno] != 0, or 0 if rsum[i][bbno] == 0 for all i.
> + * Reads and writes are serialized by the rsumip inode lock.
> */
> uint8_t *m_rsum_cache;
> struct xfs_mru_cache *m_filestream; /* per-mount filestream data */
> diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c
> index 61ef13286654..d3c76532d20e 100644
> --- a/fs/xfs/xfs_rtalloc.c
> +++ b/fs/xfs/xfs_rtalloc.c
> @@ -56,14 +56,19 @@ xfs_rtany_summary(
> int log; /* loop counter, log2 of ext. size */
> xfs_suminfo_t sum; /* summary data */
>
> - /* There are no extents at levels < m_rsum_cache[bbno]. */
> - if (mp->m_rsum_cache && low < mp->m_rsum_cache[bbno])
> - low = mp->m_rsum_cache[bbno];
> + /* There are no extents at levels >= m_rsum_cache[bbno]. */
> + if (mp->m_rsum_cache) {
> + high = min(high, mp->m_rsum_cache[bbno] - 1);
> + if (low > high) {
> + *stat = 0;
> + return 0;
> + }
> + }
>
> /*
> * Loop over logs of extent sizes.
> */
> - for (log = low; log <= high; log++) {
> + for (log = high; log >= low; log--) {
> /*
> * Get one summary datum.
> */
> @@ -84,9 +89,9 @@ xfs_rtany_summary(
> */
> *stat = 0;
> out:
> - /* There were no extents at levels < log. */
> - if (mp->m_rsum_cache && log > mp->m_rsum_cache[bbno])
> - mp->m_rsum_cache[bbno] = log;
> + /* There were no extents at levels > log. */
> + if (mp->m_rsum_cache && log + 1 < mp->m_rsum_cache[bbno])
> + mp->m_rsum_cache[bbno] = log + 1;
> return 0;
> }
>
> @@ -878,12 +883,14 @@ xfs_alloc_rsum_cache(
> xfs_extlen_t rbmblocks) /* number of rt bitmap blocks */
> {
> /*
> - * The rsum cache is initialized to all zeroes, which is trivially a
> - * lower bound on the minimum level with any free extents. We can
> - * continue without the cache if it couldn't be allocated.
> + * The rsum cache is initialized to the maximum value, which is
> + * trivially an upper bound on the maximum level with any free extents.
> + * We can continue without the cache if it couldn't be allocated.
> */
> - mp->m_rsum_cache = kvzalloc(rbmblocks, GFP_KERNEL);
> - if (!mp->m_rsum_cache)
> + mp->m_rsum_cache = kvmalloc(rbmblocks, GFP_KERNEL);
> + if (mp->m_rsum_cache)
> + memset(mp->m_rsum_cache, -1, rbmblocks);
> + else
> xfs_warn(mp, "could not allocate realtime summary cache");
> }
>
> --
> 2.41.0
>
next prev parent reply other threads:[~2023-07-12 22:40 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-06-20 21:32 [PATCH 0/6] xfs: CPU usage optimizations for realtime allocator Omar Sandoval
2023-06-20 21:32 ` [PATCH 1/6] xfs: cache last bitmap block in " Omar Sandoval
2023-07-12 18:29 ` Darrick J. Wong
2023-07-17 18:18 ` Omar Sandoval
2023-08-01 22:48 ` Darrick J. Wong
2023-06-20 21:32 ` [PATCH 2/6] xfs: invert the realtime summary cache Omar Sandoval
2023-07-12 22:40 ` Darrick J. Wong [this message]
2023-07-17 19:54 ` Omar Sandoval
2023-08-01 23:17 ` Darrick J. Wong
2023-06-20 21:32 ` [PATCH 3/6] xfs: return maximum free size from xfs_rtany_summary() Omar Sandoval
2023-07-12 22:44 ` Darrick J. Wong
2023-06-20 21:32 ` [PATCH 4/6] xfs: limit maxlen based on available space in xfs_rtallocate_extent_near() Omar Sandoval
2023-07-12 23:01 ` Darrick J. Wong
2023-07-17 20:33 ` Omar Sandoval
2023-06-20 21:32 ` [PATCH 5/6] xfs: don't try redundant allocations " Omar Sandoval
2023-07-12 23:34 ` Darrick J. Wong
2023-07-17 21:06 ` Omar Sandoval
2023-07-31 20:58 ` Omar Sandoval
2023-08-01 23:00 ` Darrick J. Wong
2023-06-20 21:32 ` [PATCH 6/6] xfs: don't look for end of extent further than necessary " Omar Sandoval
2023-08-01 23:40 ` Darrick J. Wong
2023-07-06 21:39 ` [PATCH 0/6] xfs: CPU usage optimizations for realtime allocator Omar Sandoval
2023-07-07 0:36 ` Dave Chinner
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=20230712224001.GV108251@frogsfrogsfrogs \
--to=djwong@kernel.org \
--cc=kernel-team@fb.com \
--cc=linux-xfs@vger.kernel.org \
--cc=osandov@osandov.com \
--cc=pnema@fb.com \
/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