Merge branch 'net-mlx5-hw-counters-refactor'

Tariq Toukan says:

====================
net/mlx5: hw counters refactor

This is a patchset re-post, see:
https://lore.kernel.org/20240815054656.2210494-7-tariqt@nvidia.com

In this patchset, Cosmin refactors hw counters and solves perf scaling
issue.

Series generated against:
commit c824deb1a897 ("cxgb4: clip_tbl: Fix spelling mistake "wont" -> "won't"")

HW counters are central to mlx5 driver operations. They are hardware
objects created and used alongside most steering operations, and queried
from a variety of places. Most counters are queried in bulk from a
periodic task in fs_counters.c.

Counter performance is important and as such, a variety of improvements
have been done over the years. Currently, counters are allocated from
pools, which are bulk allocated to amortize the cost of firmware
commands. Counters are managed through an IDR, a doubly linked list and
two atomic single linked lists. Adding/removing counters is a complex
dance between user contexts requesting it and the mlx5_fc_stats_work
task which does most of the work.

Under high load (e.g. from connection tracking flow insertion/deletion),
the counter code becomes a bottleneck, as seen on flame graphs. Whenever
a counter is deleted, it gets added to a list and the wq task is
scheduled to run immediately to actually delete it. This is done via
mod_delayed_work which uses an internal spinlock. In some tests, waiting
for this spinlock took up to 66% of all samples.

This series refactors the counter code to use a more straight-forward
approach, avoiding the mod_delayed_work problem and making the code
easier to understand. For that:

- patch #1 moves counters data structs to a more appropriate place.
- patch #2 simplifies the bulk query allocation scheme by using vmalloc.
- patch #3 replaces the IDR+3 lists with an xarray. This is the main
  patch of the series, solving the spinlock congestion issue.
- patch #4 removes an unnecessary cacheline alignment causing a lot of
  memory to be wasted.
- patches #5 and #6 are small cleanups enabled by the refactoring.
====================

Link: https://patch.msgid.link/20241001103709.58127-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

4 files changed, 147 insertions, 278 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
index dcfccaaa8d91..4877a9d86807 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
@@ -1026,7 +1026,7 @@ mlx5_tc_ct_counter_create(struct mlx5_tc_ct_priv *ct_priv)
 		return ERR_PTR(-ENOMEM);
 
 	counter->is_shared = false;
-	counter->counter = mlx5_fc_create_ex(ct_priv->dev, true);
+	counter->counter = mlx5_fc_create(ct_priv->dev, true);
 	if (IS_ERR(counter->counter)) {
 		ct_dbg("Failed to create counter for ct entry");
 		ret = PTR_ERR(counter->counter);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c
index 0c26d707eed2..62d0c689796b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c
@@ -32,13 +32,11 @@
 
 #include <linux/mlx5/driver.h>
 #include <linux/mlx5/fs.h>
-#include <linux/rbtree.h>
 #include "mlx5_core.h"
 #include "fs_core.h"
 #include "fs_cmd.h"
 
 #define MLX5_FC_STATS_PERIOD msecs_to_jiffies(1000)
-#define MLX5_FC_BULK_QUERY_ALLOC_PERIOD msecs_to_jiffies(180 * 1000)
 /* Max number of counters to query in bulk read is 32K */
 #define MLX5_SW_MAX_COUNTERS_BULK BIT(15)
 #define MLX5_INIT_COUNTERS_BULK 8
@@ -52,21 +50,37 @@ struct mlx5_fc_cache {
 };
 
 struct mlx5_fc {
-	struct list_head list;
-	struct llist_node addlist;
-	struct llist_node dellist;
-
-	/* last{packets,bytes} members are used when calculating the delta since
-	 * last reading
-	 */
+	u32 id;
+	bool aging;
+	struct mlx5_fc_bulk *bulk;
+	struct mlx5_fc_cache cache;
+	/* last{packets,bytes} are used for calculating deltas since last reading. */
 	u64 lastpackets;
 	u64 lastbytes;
+};
 
-	struct mlx5_fc_bulk *bulk;
-	u32 id;
-	bool aging;
+struct mlx5_fc_pool {
+	struct mlx5_core_dev *dev;
+	struct mutex pool_lock; /* protects pool lists */
+	struct list_head fully_used;
+	struct list_head partially_used;
+	struct list_head unused;
+	int available_fcs;
+	int used_fcs;
+	int threshold;
+};
 
-	struct mlx5_fc_cache cache ____cacheline_aligned_in_smp;
+struct mlx5_fc_stats {
+	struct xarray counters;
+
+	struct workqueue_struct *wq;
+	struct delayed_work work;
+	unsigned long sampling_interval; /* jiffies */
+	u32 *bulk_query_out;
+	int bulk_query_len;
+	bool bulk_query_alloc_failed;
+	unsigned long next_bulk_query_alloc;
+	struct mlx5_fc_pool fc_pool;
 };
 
 static void mlx5_fc_pool_init(struct mlx5_fc_pool *fc_pool, struct mlx5_core_dev *dev);
@@ -74,78 +88,6 @@ static void mlx5_fc_pool_cleanup(struct mlx5_fc_pool *fc_pool);
 static struct mlx5_fc *mlx5_fc_pool_acquire_counter(struct mlx5_fc_pool *fc_pool);
 static void mlx5_fc_pool_release_counter(struct mlx5_fc_pool *fc_pool, struct mlx5_fc *fc);
 
-/* locking scheme:
- *
- * It is the responsibility of the user to prevent concurrent calls or bad
- * ordering to mlx5_fc_create(), mlx5_fc_destroy() and accessing a reference
- * to struct mlx5_fc.
- * e.g en_tc.c is protected by RTNL lock of its caller, and will never call a
- * dump (access to struct mlx5_fc) after a counter is destroyed.
- *
- * access to counter list:
- * - create (user context)
- *   - mlx5_fc_create() only adds to an addlist to be used by
- *     mlx5_fc_stats_work(). addlist is a lockless single linked list
- *     that doesn't require any additional synchronization when adding single
- *     node.
- *   - spawn thread to do the actual destroy
- *
- * - destroy (user context)
- *   - add a counter to lockless dellist
- *   - spawn thread to do the actual del
- *
- * - dump (user context)
- *   user should not call dump after destroy
- *
- * - query (single thread workqueue context)
- *   destroy/dump - no conflict (see destroy)
- *   query/dump - packets and bytes might be inconsistent (since update is not
- *                atomic)
- *   query/create - no conflict (see create)
- *   since every create/destroy spawn the work, only after necessary time has
- *   elapsed, the thread will actually query the hardware.
- */
-
-static struct list_head *mlx5_fc_counters_lookup_next(struct mlx5_core_dev *dev,
-						      u32 id)
-{
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-	unsigned long next_id = (unsigned long)id + 1;
-	struct mlx5_fc *counter;
-	unsigned long tmp;
-
-	rcu_read_lock();
-	/* skip counters that are in idr, but not yet in counters list */
-	idr_for_each_entry_continue_ul(&fc_stats->counters_idr,
-				       counter, tmp, next_id) {
-		if (!list_empty(&counter->list))
-			break;
-	}
-	rcu_read_unlock();
-
-	return counter ? &counter->list : &fc_stats->counters;
-}
-
-static void mlx5_fc_stats_insert(struct mlx5_core_dev *dev,
-				 struct mlx5_fc *counter)
-{
-	struct list_head *next = mlx5_fc_counters_lookup_next(dev, counter->id);
-
-	list_add_tail(&counter->list, next);
-}
-
-static void mlx5_fc_stats_remove(struct mlx5_core_dev *dev,
-				 struct mlx5_fc *counter)
-{
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-
-	list_del(&counter->list);
-
-	spin_lock(&fc_stats->counters_idr_lock);
-	WARN_ON(!idr_remove(&fc_stats->counters_idr, counter->id));
-	spin_unlock(&fc_stats->counters_idr_lock);
-}
-
 static int get_init_bulk_query_len(struct mlx5_core_dev *dev)
 {
 	return min_t(int, MLX5_INIT_COUNTERS_BULK,
@@ -174,47 +116,64 @@ static void update_counter_cache(int index, u32 *bulk_raw_data,
 	cache->lastuse = jiffies;
 }
 
-static void mlx5_fc_stats_query_counter_range(struct mlx5_core_dev *dev,
-					      struct mlx5_fc *first,
-					      u32 last_id)
+/* Synchronization notes
+ *
+ * Access to counter array:
+ * - create - mlx5_fc_create() (user context)
+ *   - inserts the counter into the xarray.
+ *
+ * - destroy - mlx5_fc_destroy() (user context)
+ *   - erases the counter from the xarray and releases it.
+ *
+ * - query mlx5_fc_query(), mlx5_fc_query_cached{,_raw}() (user context)
+ *   - user should not access a counter after destroy.
+ *
+ * - bulk query (single thread workqueue context)
+ *   - create: query relies on 'lastuse' to avoid updating counters added
+ *             around the same time as the current bulk cmd.
+ *   - destroy: destroyed counters will not be accessed, even if they are
+ *              destroyed during a bulk query command.
+ */
+static void mlx5_fc_stats_query_all_counters(struct mlx5_core_dev *dev)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-	bool query_more_counters = (first->id <= last_id);
-	int cur_bulk_len = fc_stats->bulk_query_len;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
+	u32 bulk_len = fc_stats->bulk_query_len;
+	XA_STATE(xas, &fc_stats->counters, 0);
 	u32 *data = fc_stats->bulk_query_out;
-	struct mlx5_fc *counter = first;
+	struct mlx5_fc *counter;
+	u32 last_bulk_id = 0;
+	u64 bulk_query_time;
 	u32 bulk_base_id;
-	int bulk_len;
 	int err;
 
-	while (query_more_counters) {
-		/* first id must be aligned to 4 when using bulk query */
-		bulk_base_id = counter->id & ~0x3;
-
-		/* number of counters to query inc. the last counter */
-		bulk_len = min_t(int, cur_bulk_len,
-				 ALIGN(last_id - bulk_base_id + 1, 4));
-
-		err = mlx5_cmd_fc_bulk_query(dev, bulk_base_id, bulk_len,
-					     data);
-		if (err) {
-			mlx5_core_err(dev, "Error doing bulk query: %d\n", err);
-			return;
-		}
-		query_more_counters = false;
-
-		list_for_each_entry_from(counter, &fc_stats->counters, list) {
-			int counter_index = counter->id - bulk_base_id;
-			struct mlx5_fc_cache *cache = &counter->cache;
-
-			if (counter->id >= bulk_base_id + bulk_len) {
-				query_more_counters = true;
-				break;
+	xas_lock(&xas);
+	xas_for_each(&xas, counter, U32_MAX) {
+		if (xas_retry(&xas, counter))
+			continue;
+		if (unlikely(counter->id >= last_bulk_id)) {
+			/* Start new bulk query. */
+			/* First id must be aligned to 4 when using bulk query. */
+			bulk_base_id = counter->id & ~0x3;
+			last_bulk_id = bulk_base_id + bulk_len;
+			/* The lock is released while querying the hw and reacquired after. */
+			xas_unlock(&xas);
+			/* The same id needs to be processed again in the next loop iteration. */
+			xas_reset(&xas);
+			bulk_query_time = jiffies;
+			err = mlx5_cmd_fc_bulk_query(dev, bulk_base_id, bulk_len, data);
+			if (err) {
+				mlx5_core_err(dev, "Error doing bulk query: %d\n", err);
+				return;
 			}
-
-			update_counter_cache(counter_index, data, cache);
+			xas_lock(&xas);
+			continue;
 		}
+		/* Do not update counters added after bulk query was started. */
+		if (time_after64(bulk_query_time, counter->cache.lastuse))
+			update_counter_cache(counter->id - bulk_base_id, data,
+					     &counter->cache);
 	}
+	xas_unlock(&xas);
 }
 
 static void mlx5_fc_free(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
@@ -225,7 +184,7 @@ static void mlx5_fc_free(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
 
 static void mlx5_fc_release(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 
 	if (counter->bulk)
 		mlx5_fc_pool_release_counter(&fc_stats->fc_pool, counter);
@@ -233,85 +192,55 @@ static void mlx5_fc_release(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
 		mlx5_fc_free(dev, counter);
 }
 
-static void mlx5_fc_stats_bulk_query_size_increase(struct mlx5_core_dev *dev)
+static void mlx5_fc_stats_bulk_query_buf_realloc(struct mlx5_core_dev *dev,
+						 int bulk_query_len)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-	int max_bulk_len = get_max_bulk_query_len(dev);
-	unsigned long now = jiffies;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 	u32 *bulk_query_out_tmp;
-	int max_out_len;
-
-	if (fc_stats->bulk_query_alloc_failed &&
-	    time_before(now, fc_stats->next_bulk_query_alloc))
-		return;
+	int out_len;
 
-	max_out_len = mlx5_cmd_fc_get_bulk_query_out_len(max_bulk_len);
-	bulk_query_out_tmp = kzalloc(max_out_len, GFP_KERNEL);
+	out_len = mlx5_cmd_fc_get_bulk_query_out_len(bulk_query_len);
+	bulk_query_out_tmp = kvzalloc(out_len, GFP_KERNEL);
 	if (!bulk_query_out_tmp) {
 		mlx5_core_warn_once(dev,
-				    "Can't increase flow counters bulk query buffer size, insufficient memory, bulk_size(%d)\n",
-				    max_bulk_len);
-		fc_stats->bulk_query_alloc_failed = true;
-		fc_stats->next_bulk_query_alloc =
-			now + MLX5_FC_BULK_QUERY_ALLOC_PERIOD;
+				    "Can't increase flow counters bulk query buffer size, alloc failed, bulk_query_len(%d)\n",
+				    bulk_query_len);
 		return;
 	}
 
-	kfree(fc_stats->bulk_query_out);
+	kvfree(fc_stats->bulk_query_out);
 	fc_stats->bulk_query_out = bulk_query_out_tmp;
-	fc_stats->bulk_query_len = max_bulk_len;
-	if (fc_stats->bulk_query_alloc_failed) {
-		mlx5_core_info(dev,
-			       "Flow counters bulk query buffer size increased, bulk_size(%d)\n",
-			       max_bulk_len);
-		fc_stats->bulk_query_alloc_failed = false;
-	}
+	fc_stats->bulk_query_len = bulk_query_len;
+	mlx5_core_info(dev,
+		       "Flow counters bulk query buffer size increased, bulk_query_len(%d)\n",
+		       bulk_query_len);
 }
 
-static void mlx5_fc_stats_work(struct work_struct *work)
+static int mlx5_fc_num_counters(struct mlx5_fc_stats *fc_stats)
 {
-	struct mlx5_core_dev *dev = container_of(work, struct mlx5_core_dev,
-						 priv.fc_stats.work.work);
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-	/* Take dellist first to ensure that counters cannot be deleted before
-	 * they are inserted.
-	 */
-	struct llist_node *dellist = llist_del_all(&fc_stats->dellist);
-	struct llist_node *addlist = llist_del_all(&fc_stats->addlist);
-	struct mlx5_fc *counter = NULL, *last = NULL, *tmp;
-	unsigned long now = jiffies;
-
-	if (addlist || !list_empty(&fc_stats->counters))
-		queue_delayed_work(fc_stats->wq, &fc_stats->work,
-				   fc_stats->sampling_interval);
-
-	llist_for_each_entry(counter, addlist, addlist) {
-		mlx5_fc_stats_insert(dev, counter);
-		fc_stats->num_counters++;
-	}
-
-	llist_for_each_entry_safe(counter, tmp, dellist, dellist) {
-		mlx5_fc_stats_remove(dev, counter);
+	struct mlx5_fc *counter;
+	int num_counters = 0;
+	unsigned long id;
 
-		mlx5_fc_release(dev, counter);
-		fc_stats->num_counters--;
-	}
+	xa_for_each(&fc_stats->counters, id, counter)
+		num_counters++;
+	return num_counters;
+}
 
-	if (fc_stats->bulk_query_len < get_max_bulk_query_len(dev) &&
-	    fc_stats->num_counters > get_init_bulk_query_len(dev))
-		mlx5_fc_stats_bulk_query_size_increase(dev);
+static void mlx5_fc_stats_work(struct work_struct *work)
+{
+	struct mlx5_fc_stats *fc_stats = container_of(work, struct mlx5_fc_stats,
+						      work.work);
+	struct mlx5_core_dev *dev = fc_stats->fc_pool.dev;
 
-	if (time_before(now, fc_stats->next_query) ||
-	    list_empty(&fc_stats->counters))
-		return;
-	last = list_last_entry(&fc_stats->counters, struct mlx5_fc, list);
+	queue_delayed_work(fc_stats->wq, &fc_stats->work, fc_stats->sampling_interval);
 
-	counter = list_first_entry(&fc_stats->counters, struct mlx5_fc,
-				   list);
-	if (counter)
-		mlx5_fc_stats_query_counter_range(dev, counter, last->id);
+	/* Grow the bulk query buffer to max if not maxed and enough counters are present. */
+	if (unlikely(fc_stats->bulk_query_len < get_max_bulk_query_len(dev) &&
+		     mlx5_fc_num_counters(fc_stats) > get_init_bulk_query_len(dev)))
+		mlx5_fc_stats_bulk_query_buf_realloc(dev, get_max_bulk_query_len(dev));
 
-	fc_stats->next_query = now + fc_stats->sampling_interval;
+	mlx5_fc_stats_query_all_counters(dev);
 }
 
 static struct mlx5_fc *mlx5_fc_single_alloc(struct mlx5_core_dev *dev)
@@ -334,7 +263,7 @@ static struct mlx5_fc *mlx5_fc_single_alloc(struct mlx5_core_dev *dev)
 
 static struct mlx5_fc *mlx5_fc_acquire(struct mlx5_core_dev *dev, bool aging)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 	struct mlx5_fc *counter;
 
 	if (aging && MLX5_CAP_GEN(dev, flow_counter_bulk_alloc) != 0) {
@@ -346,16 +275,15 @@ static struct mlx5_fc *mlx5_fc_acquire(struct mlx5_core_dev *dev, bool aging)
 	return mlx5_fc_single_alloc(dev);
 }
 
-struct mlx5_fc *mlx5_fc_create_ex(struct mlx5_core_dev *dev, bool aging)
+struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging)
 {
 	struct mlx5_fc *counter = mlx5_fc_acquire(dev, aging);
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 	int err;
 
 	if (IS_ERR(counter))
 		return counter;
 
-	INIT_LIST_HEAD(&counter->list);
 	counter->aging = aging;
 
 	if (aging) {
@@ -365,18 +293,9 @@ struct mlx5_fc *mlx5_fc_create_ex(struct mlx5_core_dev *dev, bool aging)
 		counter->lastbytes = counter->cache.bytes;
 		counter->lastpackets = counter->cache.packets;
 
-		idr_preload(GFP_KERNEL);
-		spin_lock(&fc_stats->counters_idr_lock);
-
-		err = idr_alloc_u32(&fc_stats->counters_idr, counter, &id, id,
-				    GFP_NOWAIT);
-
-		spin_unlock(&fc_stats->counters_idr_lock);
-		idr_preload_end();
-		if (err)
+		err = xa_err(xa_store(&fc_stats->counters, id, counter, GFP_KERNEL));
+		if (err != 0)
 			goto err_out_alloc;
-
-		llist_add(&counter->addlist, &fc_stats->addlist);
 	}
 
 	return counter;
@@ -385,16 +304,6 @@ err_out_alloc:
 	mlx5_fc_release(dev, counter);
 	return ERR_PTR(err);
 }
-
-struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging)
-{
-	struct mlx5_fc *counter = mlx5_fc_create_ex(dev, aging);
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-
-	if (aging)
-		mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
-	return counter;
-}
 EXPORT_SYMBOL(mlx5_fc_create);
 
 u32 mlx5_fc_id(struct mlx5_fc *counter)
@@ -405,39 +314,32 @@ EXPORT_SYMBOL(mlx5_fc_id);
 
 void mlx5_fc_destroy(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 
 	if (!counter)
 		return;
 
-	if (counter->aging) {
-		llist_add(&counter->dellist, &fc_stats->dellist);
-		mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
-		return;
-	}
-
+	if (counter->aging)
+		xa_erase(&fc_stats->counters, counter->id);
 	mlx5_fc_release(dev, counter);
 }
 EXPORT_SYMBOL(mlx5_fc_destroy);
 
 int mlx5_init_fc_stats(struct mlx5_core_dev *dev)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-	int init_bulk_len;
-	int init_out_len;
+	struct mlx5_fc_stats *fc_stats;
+
+	fc_stats = kzalloc(sizeof(*fc_stats), GFP_KERNEL);
+	if (!fc_stats)
+		return -ENOMEM;
+	dev->priv.fc_stats = fc_stats;
 
-	spin_lock_init(&fc_stats->counters_idr_lock);
-	idr_init(&fc_stats->counters_idr);
-	INIT_LIST_HEAD(&fc_stats->counters);
-	init_llist_head(&fc_stats->addlist);
-	init_llist_head(&fc_stats->dellist);
+	xa_init(&fc_stats->counters);
 
-	init_bulk_len = get_init_bulk_query_len(dev);
-	init_out_len = mlx5_cmd_fc_get_bulk_query_out_len(init_bulk_len);
-	fc_stats->bulk_query_out = kzalloc(init_out_len, GFP_KERNEL);
+	/* Allocate initial (small) bulk query buffer. */
+	mlx5_fc_stats_bulk_query_buf_realloc(dev, get_init_bulk_query_len(dev));
 	if (!fc_stats->bulk_query_out)
-		return -ENOMEM;
-	fc_stats->bulk_query_len = init_bulk_len;
+		goto err_bulk;
 
 	fc_stats->wq = create_singlethread_workqueue("mlx5_fc");
 	if (!fc_stats->wq)
@@ -447,34 +349,35 @@ int mlx5_init_fc_stats(struct mlx5_core_dev *dev)
 	INIT_DELAYED_WORK(&fc_stats->work, mlx5_fc_stats_work);
 
 	mlx5_fc_pool_init(&fc_stats->fc_pool, dev);
+	queue_delayed_work(fc_stats->wq, &fc_stats->work, MLX5_FC_STATS_PERIOD);
 	return 0;
 
 err_wq_create:
-	kfree(fc_stats->bulk_query_out);
+	kvfree(fc_stats->bulk_query_out);
+err_bulk:
+	kfree(fc_stats);
 	return -ENOMEM;
 }
 
 void mlx5_cleanup_fc_stats(struct mlx5_core_dev *dev)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
-	struct llist_node *tmplist;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 	struct mlx5_fc *counter;
-	struct mlx5_fc *tmp;
-
-	cancel_delayed_work_sync(&dev->priv.fc_stats.work);
-	destroy_workqueue(dev->priv.fc_stats.wq);
-	dev->priv.fc_stats.wq = NULL;
+	unsigned long id;
 
-	tmplist = llist_del_all(&fc_stats->addlist);
-	llist_for_each_entry_safe(counter, tmp, tmplist, addlist)
-		mlx5_fc_release(dev, counter);
+	cancel_delayed_work_sync(&fc_stats->work);
+	destroy_workqueue(fc_stats->wq);
+	fc_stats->wq = NULL;
 
-	list_for_each_entry_safe(counter, tmp, &fc_stats->counters, list)
+	xa_for_each(&fc_stats->counters, id, counter) {
+		xa_erase(&fc_stats->counters, id);
 		mlx5_fc_release(dev, counter);
+	}
+	xa_destroy(&fc_stats->counters);
 
 	mlx5_fc_pool_cleanup(&fc_stats->fc_pool);
-	idr_destroy(&fc_stats->counters_idr);
-	kfree(fc_stats->bulk_query_out);
+	kvfree(fc_stats->bulk_query_out);
+	kfree(fc_stats);
 }
 
 int mlx5_fc_query(struct mlx5_core_dev *dev, struct mlx5_fc *counter,
@@ -518,7 +421,7 @@ void mlx5_fc_queue_stats_work(struct mlx5_core_dev *dev,
 			      struct delayed_work *dwork,
 			      unsigned long delay)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 
 	queue_delayed_work(fc_stats->wq, dwork, delay);
 }
@@ -526,7 +429,7 @@ void mlx5_fc_queue_stats_work(struct mlx5_core_dev *dev,
 void mlx5_fc_update_sampling_interval(struct mlx5_core_dev *dev,
 				      unsigned long interval)
 {
-	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
+	struct mlx5_fc_stats *fc_stats = dev->priv.fc_stats;
 
 	fc_stats->sampling_interval = min_t(unsigned long, interval,
 					    fc_stats->sampling_interval);
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index e23c692a34c7..fc7e6153b73d 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -45,7 +45,6 @@
 #include <linux/workqueue.h>
 #include <linux/mempool.h>
 #include <linux/interrupt.h>
-#include <linux/idr.h>
 #include <linux/notifier.h>
 #include <linux/refcount.h>
 #include <linux/auxiliary_bus.h>
@@ -474,36 +473,6 @@ struct mlx5_core_sriov {
 	u16			max_ec_vfs;
 };
 
-struct mlx5_fc_pool {
-	struct mlx5_core_dev *dev;
-	struct mutex pool_lock; /* protects pool lists */
-	struct list_head fully_used;
-	struct list_head partially_used;
-	struct list_head unused;
-	int available_fcs;
-	int used_fcs;
-	int threshold;
-};
-
-struct mlx5_fc_stats {
-	spinlock_t counters_idr_lock; /* protects counters_idr */
-	struct idr counters_idr;
-	struct list_head counters;
-	struct llist_head addlist;
-	struct llist_head dellist;
-
-	struct workqueue_struct *wq;
-	struct delayed_work work;
-	unsigned long next_query;
-	unsigned long sampling_interval; /* jiffies */
-	u32 *bulk_query_out;
-	int bulk_query_len;
-	size_t num_counters;
-	bool bulk_query_alloc_failed;
-	unsigned long next_bulk_query_alloc;
-	struct mlx5_fc_pool fc_pool;
-};
-
 struct mlx5_events;
 struct mlx5_mpfs;
 struct mlx5_eswitch;
@@ -630,7 +599,7 @@ struct mlx5_priv {
 	struct mlx5_devcom_comp_dev *hca_devcom_comp;
 	struct mlx5_fw_reset	*fw_reset;
 	struct mlx5_core_roce	roce;
-	struct mlx5_fc_stats		fc_stats;
+	struct mlx5_fc_stats		*fc_stats;
 	struct mlx5_rl_table            rl_table;
 	struct mlx5_ft_pool		*ft_pool;
 
diff --git a/include/linux/mlx5/fs.h b/include/linux/mlx5/fs.h
index b744e554f014..438db888bde0 100644
--- a/include/linux/mlx5/fs.h
+++ b/include/linux/mlx5/fs.h
@@ -298,9 +298,6 @@ int mlx5_modify_rule_destination(struct mlx5_flow_handle *handler,
 
 struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging);
 
-/* As mlx5_fc_create() but doesn't queue stats refresh thread. */
-struct mlx5_fc *mlx5_fc_create_ex(struct mlx5_core_dev *dev, bool aging);
-
 void mlx5_fc_destroy(struct mlx5_core_dev *dev, struct mlx5_fc *counter);
 u64 mlx5_fc_query_lastuse(struct mlx5_fc *counter);
 void mlx5_fc_query_cached(struct mlx5_fc *counter,