Merge tag 'xfs-cil-scale-5.20' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs into xfs-5.20-mergeA

xfs: improve CIL scalability

This series aims to improve the scalability of XFS transaction
commits on large CPU count machines. My 32p machine hits contention
limits in xlog_cil_commit() at about 700,000 transaction commits a
section. It hits this at 16 thread workloads, and 32 thread
workloads go no faster and just burn CPU on the CIL spinlocks.

This patchset gets rid of spinlocks and global serialisation points
in the xlog_cil_commit() path. It does this by moving to a
combination of per-cpu counters, unordered per-cpu lists and
post-ordered per-cpu lists.

This results in transaction commit rates exceeding 1.4 million
commits/s under unlink certain workloads, and while the log lock
contention is largely gone there is still significant lock
contention in the VFS (dentry cache, inode cache and security layers)
at >600,000 transactions/s that still limit scalability.

The changes to the CIL accounting and behaviour, combined with the
structural changes to xlog_write() in prior patchsets make the
per-cpu restructuring possible and sane. This allows us to move to
precalculated reservation requirements that allow for reservation
stealing to be accounted across multiple CPUs accurately.

That is, instead of trying to account for continuation log opheaders
on a "growth" basis, we pre-calculate how many iclogs we'll need to
write out a maximally sized CIL checkpoint and steal that reserveD
that space one commit at a time until the CIL has a full
reservation. If we ever run a commit when we are already at the hard
limit (because post-throttling) we simply take an extra reservation
from each commit that is run when over the limit. Hence we don't
need to do space usage math in the fast path and so never need to
sum the per-cpu counters in this fast path.

Similarly, per-cpu lists have the problem of ordering - we can't
remove an item from a per-cpu list if we want to move it forward in
the CIL. We solve this problem by using an atomic counter to give
every commit a sequence number that is copied into the log items in
that transaction. Hence relogging items just overwrites the sequence
number in the log item, and does not move it in the per-cpu lists.
Once we reaggregate the per-cpu lists back into a single list in the
CIL push work, we can run it through list-sort() and reorder it back
into a globally ordered list. This costs a bit of CPU time, but now
that the CIL can run multiple works and pipelines properly, this is
not a limiting factor for performance. It does increase fsync
latency when the CIL is full, but workloads issuing large numbers of
fsync()s or sync transactions end up with very small CILs and so the
latency impact or sorting is not measurable for such workloads.

OVerall, this pushes the transaction commit bottleneck out to the
lockless reservation grant head updates. These atomic updates don't
start to be a limiting fact until > 1.5 million transactions/s are
being run, at which point the accounting functions start to show up
in profiles as the highest CPU users. Still, this series doubles
transaction throughput without increasing CPU usage before we get
to that cacheline contention breakdown point...
`
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>

* tag 'xfs-cil-scale-5.20' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs:
  xfs: expanding delayed logging design with background material
  xfs: xlog_sync() manually adjusts grant head space
  xfs: avoid cil push lock if possible
  xfs: move CIL ordering to the logvec chain
  xfs: convert log vector chain to use list heads
  xfs: convert CIL to unordered per cpu lists
  xfs: Add order IDs to log items in CIL
  xfs: convert CIL busy extents to per-cpu
  xfs: track CIL ticket reservation in percpu structure
  xfs: implement percpu cil space used calculation
  xfs: introduce per-cpu CIL tracking structure
  xfs: rework per-iclog header CIL reservation
  xfs: lift init CIL reservation out of xc_cil_lock
  xfs: use the CIL space used counter for emptiness checks

Author: Darrick J. Wong

Documentation/filesystems/xfs-delayed-logging-design.rst

@@ -57,7 +57,8 @@ xlog_grant_push_ail(
 STATIC void
 xlog_sync(
 	struct xlog		*log,
-	struct xlog_in_core	*iclog);
+	struct xlog_in_core	*iclog,
+	struct xlog_ticket	*ticket);
 #if defined(DEBUG)
 STATIC void
 xlog_verify_grant_tail(
@@ -567,7 +568,8 @@ xlog_state_shutdown_callbacks(
 int
 xlog_state_release_iclog(
 	struct xlog		*log,
-	struct xlog_in_core	*iclog)
+	struct xlog_in_core	*iclog,
+	struct xlog_ticket	*ticket)
 {
 	xfs_lsn_t		tail_lsn;
 	bool			last_ref;
@@ -614,7 +616,7 @@ xlog_state_release_iclog(
 	trace_xlog_iclog_syncing(iclog, _RET_IP_);
 
 	spin_unlock(&log->l_icloglock);
-	xlog_sync(log, iclog);
+	xlog_sync(log, iclog, ticket);
 	spin_lock(&log->l_icloglock);
 	return 0;
 }
@@ -881,7 +883,7 @@ xlog_force_iclog(
 	iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;
 	if (iclog->ic_state == XLOG_STATE_ACTIVE)
 		xlog_state_switch_iclogs(iclog->ic_log, iclog, 0);
-	return xlog_state_release_iclog(iclog->ic_log, iclog);
+	return xlog_state_release_iclog(iclog->ic_log, iclog, NULL);
 }
 
 /*
@@ -944,6 +946,8 @@ xlog_write_unmount_record(
 		.lv_niovecs = 1,
 		.lv_iovecp = &reg,
 	};
+	LIST_HEAD(lv_chain);
+	list_add(&vec.lv_list, &lv_chain);
 
 	BUILD_BUG_ON((sizeof(struct xlog_op_header) +
 		      sizeof(struct xfs_unmount_log_format)) !=
@@ -952,7 +956,7 @@ xlog_write_unmount_record(
 	/* account for space used by record data */
 	ticket->t_curr_res -= sizeof(unmount_rec);
 
-	return xlog_write(log, NULL, &vec, ticket, reg.i_len);
+	return xlog_write(log, NULL, &lv_chain, ticket, reg.i_len);
 }
 
 /*
@@ -2025,7 +2029,8 @@ xlog_calc_iclog_size(
 STATIC void
 xlog_sync(
 	struct xlog		*log,
-	struct xlog_in_core	*iclog)
+	struct xlog_in_core	*iclog,
+	struct xlog_ticket	*ticket)
 {
 	unsigned int		count;		/* byte count of bwrite */
 	unsigned int		roundoff;       /* roundoff to BB or stripe */
@@ -2037,12 +2042,20 @@ xlog_sync(
 
 	count = xlog_calc_iclog_size(log, iclog, &roundoff);
 
-	/* move grant heads by roundoff in sync */
-	xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
-	xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
+	/*
+	 * If we have a ticket, account for the roundoff via the ticket
+	 * reservation to avoid touching the hot grant heads needlessly.
+	 * Otherwise, we have to move grant heads directly.
+	 */
+	if (ticket) {
+		ticket->t_curr_res -= roundoff;
+	} else {
+		xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
+		xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
+	}
 
 	/* put cycle number in every block */
-	xlog_pack_data(log, iclog, roundoff); 
+	xlog_pack_data(log, iclog, roundoff);
 
 	/* real byte length */
 	size = iclog->ic_offset;
@@ -2275,7 +2288,7 @@ xlog_write_get_more_iclog_space(
 	spin_lock(&log->l_icloglock);
 	ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC);
 	xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
-	error = xlog_state_release_iclog(log, iclog);
+	error = xlog_state_release_iclog(log, iclog, ticket);
 	spin_unlock(&log->l_icloglock);
 	if (error)
 		return error;
@@ -2471,13 +2484,13 @@ int
 xlog_write(
 	struct xlog		*log,
 	struct xfs_cil_ctx	*ctx,
-	struct xfs_log_vec	*log_vector,
+	struct list_head	*lv_chain,
 	struct xlog_ticket	*ticket,
 	uint32_t		len)
 
 {
 	struct xlog_in_core	*iclog = NULL;
-	struct xfs_log_vec	*lv = log_vector;
+	struct xfs_log_vec	*lv;
 	uint32_t		record_cnt = 0;
 	uint32_t		data_cnt = 0;
 	int			error = 0;
@@ -2505,7 +2518,7 @@ xlog_write(
 	if (ctx)
 		xlog_cil_set_ctx_write_state(ctx, iclog);
 
-	while (lv) {
+	list_for_each_entry(lv, lv_chain, lv_list) {
 		/*
 		 * If the entire log vec does not fit in the iclog, punt it to
 		 * the partial copy loop which can handle this case.
@@ -2526,7 +2539,6 @@ xlog_write(
 			xlog_write_full(lv, ticket, iclog, &log_offset,
 					 &len, &record_cnt, &data_cnt);
 		}
-		lv = lv->lv_next;
 	}
 	ASSERT(len == 0);
 
@@ -2538,7 +2550,7 @@ xlog_write(
 	 */
 	spin_lock(&log->l_icloglock);
 	xlog_state_finish_copy(log, iclog, record_cnt, 0);
-	error = xlog_state_release_iclog(log, iclog);
+	error = xlog_state_release_iclog(log, iclog, ticket);
 	spin_unlock(&log->l_icloglock);
 
 	return error;
@@ -2958,7 +2970,7 @@ xlog_state_get_iclog_space(
 		 * reference to the iclog.
 		 */
 		if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1))
-			error = xlog_state_release_iclog(log, iclog);
+			error = xlog_state_release_iclog(log, iclog, ticket);
 		spin_unlock(&log->l_icloglock);
 		if (error)
 			return error;
@@ -3406,7 +3418,8 @@ xfs_log_ticket_get(
 static int
 xlog_calc_unit_res(
 	struct xlog		*log,
-	int			unit_bytes)
+	int			unit_bytes,
+	int			*niclogs)
 {
 	int			iclog_space;
 	uint			num_headers;
@@ -3486,6 +3499,8 @@ xlog_calc_unit_res(
 	/* roundoff padding for transaction data and one for commit record */
 	unit_bytes += 2 * log->l_iclog_roundoff;
 
+	if (niclogs)
+		*niclogs = num_headers;
 	return unit_bytes;
 }
 
@@ -3494,7 +3509,7 @@ xfs_log_calc_unit_res(
 	struct xfs_mount	*mp,
 	int			unit_bytes)
 {
-	return xlog_calc_unit_res(mp->m_log, unit_bytes);
+	return xlog_calc_unit_res(mp->m_log, unit_bytes, NULL);
 }
 
 /*
@@ -3512,7 +3527,7 @@ xlog_ticket_alloc(
 
 	tic = kmem_cache_zalloc(xfs_log_ticket_cache, GFP_NOFS | __GFP_NOFAIL);
 
-	unit_res = xlog_calc_unit_res(log, unit_bytes);
+	unit_res = xlog_calc_unit_res(log, unit_bytes, &tic->t_iclog_hdrs);
 
 	atomic_set(&tic->t_ref, 1);
 	tic->t_task		= current;

fs/xfs/xfs_log.c

@@ -9,7 +9,8 @@
 struct xfs_cil_ctx;
 
 struct xfs_log_vec {
-	struct xfs_log_vec	*lv_next;	/* next lv in build list */
+	struct list_head	lv_list;	/* CIL lv chain ptrs */
+	uint32_t		lv_order_id;	/* chain ordering info */
 	int			lv_niovecs;	/* number of iovecs in lv */
 	struct xfs_log_iovec	*lv_iovecp;	/* iovec array */
 	struct xfs_log_item	*lv_item;	/* owner */