From: David Chinner <dgc@sgi.com>
To: xfs-dev <xfs-dev@sgi.com>
Cc: xfs-oss <xfs@oss.sgi.com>
Subject: [patch] Use atomics for iclog reference counting
Date: Mon, 21 Jan 2008 16:30:21 +1100 [thread overview]
Message-ID: <20080121053021.GH155259@sgi.com> (raw)
Now that we update the log tail LSN less frequently on
transaction completion, we pass the contention straight to
the global block stat lock (l_iclog_lock) during transaction
completion.
We currently have to take this lock to decrement the iclog
reference count. there is a reference count on each iclog,
so we need to take þhe global lock for all refcount changes.
When large numbers of processes are all doing small trnasctions,
the iclog reference counts will be quite high, and the state change
that absolutely requires the l_iclog_lock is the except rather than
the norm.
Change the reference counting on the iclogs to use atomic_inc/dec
so that we can use atomic_dec_and_lock during transaction completion
and avoid the need for grabbing the l_iclog_lock for every reference
count decrement except the one that matters - the last.
Signed-off-by: Dave Chinner <dgc@sgi.com>
---
fs/xfs/xfs_log.c | 36 +++++++++++++++++++++---------------
fs/xfs/xfs_log_priv.h | 2 +-
fs/xfs/xfsidbg.c | 2 +-
3 files changed, 23 insertions(+), 17 deletions(-)
Index: 2.6.x-xfs-new/fs/xfs/xfs_log.c
===================================================================
--- 2.6.x-xfs-new.orig/fs/xfs/xfs_log.c 2008-01-21 16:16:51.804146394 +1100
+++ 2.6.x-xfs-new/fs/xfs/xfs_log.c 2008-01-21 16:23:35.369691221 +1100
@@ -675,7 +675,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
- iclog->ic_refcnt++;
+ atomic_inc(&iclog->ic_refcnt);
spin_unlock(&log->l_icloglock);
xlog_state_want_sync(log, iclog);
(void) xlog_state_release_iclog(log, iclog);
@@ -713,7 +713,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
*/
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
- iclog->ic_refcnt++;
+ atomic_inc(&iclog->ic_refcnt);
spin_unlock(&log->l_icloglock);
xlog_state_want_sync(log, iclog);
@@ -1405,7 +1405,7 @@ xlog_sync(xlog_t *log,
int v2 = XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb);
XFS_STATS_INC(xs_log_writes);
- ASSERT(iclog->ic_refcnt == 0);
+ ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
/* Add for LR header */
count_init = log->l_iclog_hsize + iclog->ic_offset;
@@ -2311,7 +2311,7 @@ xlog_state_done_syncing(
ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
iclog->ic_state == XLOG_STATE_IOERROR);
- ASSERT(iclog->ic_refcnt == 0);
+ ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
@@ -2393,7 +2393,7 @@ restart:
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
head = &iclog->ic_header;
- iclog->ic_refcnt++; /* prevents sync */
+ atomic_inc(&iclog->ic_refcnt); /* prevents sync */
log_offset = iclog->ic_offset;
/* On the 1st write to an iclog, figure out lsn. This works
@@ -2425,12 +2425,12 @@ restart:
xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
/* If I'm the only one writing to this iclog, sync it to disk */
- if (iclog->ic_refcnt == 1) {
+ if (atomic_read(&iclog->ic_refcnt) == 1) {
spin_unlock(&log->l_icloglock);
if ((error = xlog_state_release_iclog(log, iclog)))
return error;
} else {
- iclog->ic_refcnt--;
+ atomic_dec(&iclog->ic_refcnt);
spin_unlock(&log->l_icloglock);
}
goto restart;
@@ -2821,18 +2821,23 @@ xlog_state_release_iclog(
{
int sync = 0; /* do we sync? */
- spin_lock(&log->l_icloglock);
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
- ASSERT(iclog->ic_refcnt > 0);
+ ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
+ if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
+ return 0;
+
+ if (iclog->ic_state & XLOG_STATE_IOERROR) {
+ spin_unlock(&log->l_icloglock);
+ return XFS_ERROR(EIO);
+ }
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_WANT_SYNC);
- if (--iclog->ic_refcnt == 0 &&
- iclog->ic_state == XLOG_STATE_WANT_SYNC) {
+ if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
/* update tail before writing to iclog */
xlog_assign_tail_lsn(log->l_mp);
sync++;
@@ -2952,7 +2957,8 @@ xlog_state_sync_all(xlog_t *log, uint fl
* previous iclog and go to sleep.
*/
if (iclog->ic_state == XLOG_STATE_DIRTY ||
- (iclog->ic_refcnt == 0 && iclog->ic_offset == 0)) {
+ (atomic_read(&iclog->ic_refcnt) == 0
+ && iclog->ic_offset == 0)) {
iclog = iclog->ic_prev;
if (iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)
@@ -2960,14 +2966,14 @@ xlog_state_sync_all(xlog_t *log, uint fl
else
goto maybe_sleep;
} else {
- if (iclog->ic_refcnt == 0) {
+ if (atomic_read(&iclog->ic_refcnt) == 0) {
/* We are the only one with access to this
* iclog. Flush it out now. There should
* be a roundoff of zero to show that someone
* has already taken care of the roundoff from
* the previous sync.
*/
- iclog->ic_refcnt++;
+ atomic_inc(&iclog->ic_refcnt);
lsn = be64_to_cpu(iclog->ic_header.h_lsn);
xlog_state_switch_iclogs(log, iclog, 0);
spin_unlock(&log->l_icloglock);
@@ -3099,7 +3105,7 @@ try_again:
already_slept = 1;
goto try_again;
} else {
- iclog->ic_refcnt++;
+ atomic_inc(&iclog->ic_refcnt);
xlog_state_switch_iclogs(log, iclog, 0);
spin_unlock(&log->l_icloglock);
if (xlog_state_release_iclog(log, iclog))
Index: 2.6.x-xfs-new/fs/xfs/xfs_log_priv.h
===================================================================
--- 2.6.x-xfs-new.orig/fs/xfs/xfs_log_priv.h 2008-01-21 16:06:27.127557437 +1100
+++ 2.6.x-xfs-new/fs/xfs/xfs_log_priv.h 2008-01-21 16:23:35.369691221 +1100
@@ -339,7 +339,7 @@ typedef struct xlog_iclog_fields {
#endif
int ic_size;
int ic_offset;
- int ic_refcnt;
+ atomic_t ic_refcnt;
int ic_bwritecnt;
ushort_t ic_state;
char *ic_datap; /* pointer to iclog data */
Index: 2.6.x-xfs-new/fs/xfs/xfsidbg.c
===================================================================
--- 2.6.x-xfs-new.orig/fs/xfs/xfsidbg.c 2008-01-21 16:06:27.127557437 +1100
+++ 2.6.x-xfs-new/fs/xfs/xfsidbg.c 2008-01-21 16:23:35.385689220 +1100
@@ -5633,7 +5633,7 @@ xfsidbg_xiclog(xlog_in_core_t *iclog)
#else
NULL,
#endif
- iclog->ic_refcnt, iclog->ic_bwritecnt);
+ atomic_read(&iclog->ic_refcnt), iclog->ic_bwritecnt);
if (iclog->ic_state & XLOG_STATE_ALL)
printflags(iclog->ic_state, ic_flags, " state:");
else
next reply other threads:[~2008-01-21 5:30 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2008-01-21 5:30 David Chinner [this message]
2008-01-23 5:13 ` [patch] Use atomics for iclog reference counting Timothy Shimmin
2008-02-14 23:47 ` David Chinner
2008-02-15 0:24 ` David Chinner
2008-02-15 3:34 ` Timothy Shimmin
2008-02-15 4:27 ` David Chinner
2008-02-15 5:05 ` Timothy Shimmin
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=20080121053021.GH155259@sgi.com \
--to=dgc@sgi.com \
--cc=xfs-dev@sgi.com \
--cc=xfs@oss.sgi.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