[net-next,V2,3/6] net/mlx5: hw counters: Replace IDR+lists with xarray

Commit Message

Tariq Toukan Oct. 1, 2024, 10:37 a.m. UTC

From: Cosmin Ratiu <[e-mail]>

Previously, managing counters was a complicated affair involving an IDR,
a sorted double linked list, two single linked lists and a complex dance
between a non-periodic wq task and users adding/deleting counters.

Adding was done by inserting new counters into the IDR and into a single
linked list, leaving the wq to process the list and actually add the
counters into the double linked list, maintained sorted with the IDR.

Deleting involved adding the counter into another single linked list,
leaving the wq to actually unlink the counter from the other structures
and release it.

Dumping the counters is done with the bulk query API, which relies on
the counter list being sorted and unmutable during querying to
efficiently retrieve cached counter values.

Finally, the IDR data struct is deprecated.

This commit replaces all of that with an xarray.

Adding is now done directly, by using xa_lock.
Deleting is also done directly, under the xa_lock.

Querying is done from a periodic task running every sampling_interval
(default 1s) and uses the bulk query API for efficiency.
It works by iterating over the xarray:
- when a new bulk needs to be started, the bulk information is computed
  under the xa_lock.
- the xa iteration state is saved and the xa_lock dropped.
- the HW is queried for bulk counter values.
- the xa_lock is reacquired.
- counter caches with ids covered by the bulk response are updated.

Querying always requests the max bulk length, for simplicity.

Counters could be added/deleted while the HW is queried. This is safe,
as the HW API simply returns unknown values for counters not in HW, but
those values won't be accessed. Only counters present in xarray before
bulk query will actually read queried cache values.

This cuts down the size of mlx5_fc by 4 pointers (88->56 bytes), which
amounts to ~3MB / 100K counters.
But more importantly, this solves the wq spinlock congestion issue seen
happening on high-rate counter insertion+deletion.

Signed-off-by: Cosmin Ratiu <[e-mail]>
Signed-off-by: Tariq Toukan <[e-mail]>
---
 .../ethernet/mellanox/mlx5/core/fs_counters.c | 276 ++++++------------
 1 file changed, 82 insertions(+), 194 deletions(-)

Comments

Simon Horman Oct. 4, 2024, 8:58 a.m. UTC | #1

On Tue, Oct 01, 2024 at 01:37:06PM +0300, Tariq Toukan wrote:
> From: Cosmin Ratiu <[e-mail]>

...

> +/* 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;
> +	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. */

Hi Cosmin and Tariq,

I'm sorry if it is obvious, but I'm wondering if you could explain further
the relationship between the if block above, where bulk_query_time (and
bulk_base_id) is initialised and if block below, which is conditional on
bulk_query_time.

> +		if (time_after64(bulk_query_time, counter->cache.lastuse))
> +			update_counter_cache(counter->id - bulk_base_id, data,
> +					     &counter->cache);
>  	}
> +	xas_unlock(&xas);
>  }

...

Cosmin Ratiu Oct. 4, 2024, 9:32 a.m. UTC | #2

On Fri, 2024-10-04 at 09:58 +0100, Simon Horman wrote:
> On Tue, Oct 01, 2024 at 01:37:06PM +0300, Tariq Toukan wrote:
> > From: Cosmin Ratiu <[e-mail]>
> 
> ...
> 
> > +/* 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;
> > +	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. */
> 
> Hi Cosmin and Tariq,
> 
> I'm sorry if it is obvious, but I'm wondering if you could explain further
> the relationship between the if block above, where bulk_query_time (and
> bulk_base_id) is initialised and if block below, which is conditional on
> bulk_query_time.
> 
> > +		if (time_after64(bulk_query_time, counter->cache.lastuse))
> > +			update_counter_cache(counter->id - bulk_base_id, data,
> > +					     &counter->cache);
> >  	}
> > +	xas_unlock(&xas);
> >  }
> 
> ...

Hi Simon. Of course.

The first if (with 'unlikely') is the one that starts a bulk query.
The second if is the one that updates a counter's cached value with the
output from the bulk query. Bulks are usually ~32K counters, if I
remember correctly. In any case, a large number.

The first if sets up the bulk query params and executes it without the
lock held. During that time, counters could be added/removed. We don't
want to update counter values for counters added between when the bulk
query was executed and when the lock was reacquired. bulk_query_time
with jiffy granularity is used for that purpose. When a counter is
added, its 'cache.lastuse' is initialized to jiffies. Only counters
with ids between [bulk_base_id, last_bulk_id) added strictly before the
jiffy when bulk_query_time was set will be updated because the hw might
not have set newer counter values in the bulk result and values might
be garbage.

I also have this blurb in the commit description, but it is probably
lost in the wall of text:
"
Counters could be added/deleted while the HW is queried. This is safe,
as the HW API simply returns unknown values for counters not in HW, but
those values won't be accessed. Only counters present in xarray before
bulk query will actually read queried cache values.
"

There's also a comment bit in the "Synchronization notes" section:
 * - 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.

Hope this clears things out, let us know if you'd like something
improved.

Cosmin.

Simon Horman Oct. 4, 2024, 10:34 a.m. UTC | #3

On Fri, Oct 04, 2024 at 09:32:11AM +0000, Cosmin Ratiu wrote:
> On Fri, 2024-10-04 at 09:58 +0100, Simon Horman wrote:
> > On Tue, Oct 01, 2024 at 01:37:06PM +0300, Tariq Toukan wrote:
> > > From: Cosmin Ratiu <[e-mail]>
> > 
> > ...
> > 
> > > +/* 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;
> > > +	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. */
> > 
> > Hi Cosmin and Tariq,
> > 
> > I'm sorry if it is obvious, but I'm wondering if you could explain further
> > the relationship between the if block above, where bulk_query_time (and
> > bulk_base_id) is initialised and if block below, which is conditional on
> > bulk_query_time.
> > 
> > > +		if (time_after64(bulk_query_time, counter->cache.lastuse))
> > > +			update_counter_cache(counter->id - bulk_base_id, data,
> > > +					     &counter->cache);
> > >  	}
> > > +	xas_unlock(&xas);
> > >  }
> > 
> > ...
> 
> Hi Simon. Of course.
> 
> The first if (with 'unlikely') is the one that starts a bulk query.
> The second if is the one that updates a counter's cached value with the
> output from the bulk query. Bulks are usually ~32K counters, if I
> remember correctly. In any case, a large number.
> 
> The first if sets up the bulk query params and executes it without the
> lock held. During that time, counters could be added/removed. We don't
> want to update counter values for counters added between when the bulk
> query was executed and when the lock was reacquired. bulk_query_time
> with jiffy granularity is used for that purpose. When a counter is
> added, its 'cache.lastuse' is initialized to jiffies. Only counters
> with ids between [bulk_base_id, last_bulk_id) added strictly before the
> jiffy when bulk_query_time was set will be updated because the hw might
> not have set newer counter values in the bulk result and values might
> be garbage.
> 
> I also have this blurb in the commit description, but it is probably
> lost in the wall of text:
> "
> Counters could be added/deleted while the HW is queried. This is safe,
> as the HW API simply returns unknown values for counters not in HW, but
> those values won't be accessed. Only counters present in xarray before
> bulk query will actually read queried cache values.
> "

Thanks, I did see that, but for some reason I didn't relate
it to the question I asked.

> 
> There's also a comment bit in the "Synchronization notes" section:
>  * - 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.

But this one I had missed.

> 
> Hope this clears things out, let us know if you'd like something
> improved.

Yes, thank you. It is clear now :)

Patch

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c
index 9892895da9ee..05d9351ff577 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c
@@ -32,7 +32,6 @@ 
 
 #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"
@@ -51,21 +50,13 @@  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
-	 */
-	u64 lastpackets;
-	u64 lastbytes;
-
-	struct mlx5_fc_bulk *bulk;
 	u32 id;
 	bool aging;
-
+	struct mlx5_fc_bulk *bulk;
 	struct mlx5_fc_cache cache ____cacheline_aligned_in_smp;
+	/* last{packets,bytes} are used for calculating deltas since last reading. */
+	u64 lastpackets;
+	u64 lastbytes;
 };
 
 struct mlx5_fc_pool {
@@ -80,19 +71,14 @@  struct mlx5_fc_pool {
 };
 
 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 xarray counters;
 
 	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;
+	size_t num_counters;  /* Also protected by xarray->xa_lock. */
 	bool bulk_query_alloc_failed;
 	unsigned long next_bulk_query_alloc;
 	struct mlx5_fc_pool fc_pool;
@@ -103,78 +89,6 @@  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,
@@ -203,47 +117,64 @@  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;
+	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)
@@ -291,46 +222,19 @@  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;
-	/* 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++;
-	}
+	int num_counters;
 
-	llist_for_each_entry_safe(counter, tmp, dellist, dellist) {
-		mlx5_fc_stats_remove(dev, counter);
-
-		mlx5_fc_release(dev, counter);
-		fc_stats->num_counters--;
-	}
-
-	if (time_before(now, fc_stats->next_query) ||
-	    list_empty(&fc_stats->counters))
-		return;
+	queue_delayed_work(fc_stats->wq, &fc_stats->work, fc_stats->sampling_interval);
 
+	/* num_counters is only needed for determining whether to increase the buffer. */
+	xa_lock(&fc_stats->counters);
+	num_counters = fc_stats->num_counters;
+	xa_unlock(&fc_stats->counters);
 	if (fc_stats->bulk_query_len < get_max_bulk_query_len(dev) &&
-	    fc_stats->num_counters > get_init_bulk_query_len(dev))
+	    num_counters > get_init_bulk_query_len(dev))
 		mlx5_fc_stats_bulk_query_buf_realloc(dev, get_max_bulk_query_len(dev));
 
-	last = list_last_entry(&fc_stats->counters, struct mlx5_fc, list);
-
-	counter = list_first_entry(&fc_stats->counters, struct mlx5_fc,
-				   list);
-	if (counter)
-		mlx5_fc_stats_query_counter_range(dev, counter, last->id);
-
-	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)
@@ -374,7 +278,6 @@  struct mlx5_fc *mlx5_fc_create_ex(struct mlx5_core_dev *dev, bool aging)
 	if (IS_ERR(counter))
 		return counter;
 
-	INIT_LIST_HEAD(&counter->list);
 	counter->aging = aging;
 
 	if (aging) {
@@ -384,18 +287,15 @@  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);
+		xa_lock(&fc_stats->counters);
 
-		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) {
+			xa_unlock(&fc_stats->counters);
 			goto err_out_alloc;
-
-		llist_add(&counter->addlist, &fc_stats->addlist);
+		}
+		fc_stats->num_counters++;
+		xa_unlock(&fc_stats->counters);
 	}
 
 	return counter;
@@ -407,12 +307,7 @@  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_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;
+	return mlx5_fc_create_ex(dev, aging);
 }
 EXPORT_SYMBOL(mlx5_fc_create);
 
@@ -430,11 +325,11 @@  void mlx5_fc_destroy(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
 		return;
 
 	if (counter->aging) {
-		llist_add(&counter->dellist, &fc_stats->dellist);
-		mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
-		return;
+		xa_lock(&fc_stats->counters);
+		fc_stats->num_counters--;
+		__xa_erase(&fc_stats->counters, counter->id);
+		xa_unlock(&fc_stats->counters);
 	}
-
 	mlx5_fc_release(dev, counter);
 }
 EXPORT_SYMBOL(mlx5_fc_destroy);
@@ -448,11 +343,7 @@  int mlx5_init_fc_stats(struct mlx5_core_dev *dev)
 		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);
 
 	/* Allocate initial (small) bulk query buffer. */
 	mlx5_fc_stats_bulk_query_buf_realloc(dev, get_init_bulk_query_len(dev));
@@ -467,7 +358,7 @@  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:
@@ -480,23 +371,20 @@  int mlx5_init_fc_stats(struct mlx5_core_dev *dev)
 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 *counter;
-	struct mlx5_fc *tmp;
+	unsigned long id;
 
 	cancel_delayed_work_sync(&fc_stats->work);
 	destroy_workqueue(fc_stats->wq);
 	fc_stats->wq = NULL;
 
-	tmplist = llist_del_all(&fc_stats->addlist);
-	llist_for_each_entry_safe(counter, tmp, tmplist, addlist)
-		mlx5_fc_release(dev, counter);
-
-	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);
 	kvfree(fc_stats->bulk_query_out);
 	kfree(fc_stats);
 }