net/mlx5: Introduce bulk reading of flow counters

This commit utilize the ability of ConnectX-4 to bulk read flow counters.
Few bulk counter queries could be done instead of issuing thousands of
firmware commands per second to get statistics of all flows set to HW,
such as those programmed when we offload tc filters.

Counters are stored sorted by hardware id, and queried in blocks (id +
number of counters).

Due to hardware requirement, start of block and number of counters in a
block must be four aligned.

Reviewed-by: Or Gerlitz <[email protected]>
Signed-off-by: Amir Vadai <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c

@@ -413,3 +413,70 @@ int mlx5_cmd_fc_query(struct mlx5_core_dev *dev, u16 id,
 
 	return 0;
 }
+
+struct mlx5_cmd_fc_bulk {
+	u16 id;
+	int num;
+	int outlen;
+	u32 out[0];
+};
+
+struct mlx5_cmd_fc_bulk *
+mlx5_cmd_fc_bulk_alloc(struct mlx5_core_dev *dev, u16 id, int num)
+{
+	struct mlx5_cmd_fc_bulk *b;
+	int outlen = sizeof(*b) +
+		MLX5_ST_SZ_BYTES(query_flow_counter_out) +
+		MLX5_ST_SZ_BYTES(traffic_counter) * num;
+
+	b = kzalloc(outlen, GFP_KERNEL);
+	if (!b)
+		return NULL;
+
+	b->id = id;
+	b->num = num;
+	b->outlen = outlen;
+
+	return b;
+}
+
+void mlx5_cmd_fc_bulk_free(struct mlx5_cmd_fc_bulk *b)
+{
+	kfree(b);
+}
+
+int
+mlx5_cmd_fc_bulk_query(struct mlx5_core_dev *dev, struct mlx5_cmd_fc_bulk *b)
+{
+	u32 in[MLX5_ST_SZ_DW(query_flow_counter_in)];
+
+	memset(in, 0, sizeof(in));
+
+	MLX5_SET(query_flow_counter_in, in, opcode,
+		 MLX5_CMD_OP_QUERY_FLOW_COUNTER);
+	MLX5_SET(query_flow_counter_in, in, op_mod, 0);
+	MLX5_SET(query_flow_counter_in, in, flow_counter_id, b->id);
+	MLX5_SET(query_flow_counter_in, in, num_of_counters, b->num);
+
+	return mlx5_cmd_exec_check_status(dev, in, sizeof(in),
+					  b->out, b->outlen);
+}
+
+void mlx5_cmd_fc_bulk_get(struct mlx5_core_dev *dev,
+			  struct mlx5_cmd_fc_bulk *b, u16 id,
+			  u64 *packets, u64 *bytes)
+{
+	int index = id - b->id;
+	void *stats;
+
+	if (index < 0 || index >= b->num) {
+		mlx5_core_warn(dev, "Flow counter id (0x%x) out of range (0x%x..0x%x). Counter ignored.\n",
+			       id, b->id, b->id + b->num - 1);
+		return;
+	}
+
+	stats = MLX5_ADDR_OF(query_flow_counter_out, b->out,
+			     flow_statistics[index]);
+	*packets = MLX5_GET64(traffic_counter, stats, packets);
+	*bytes = MLX5_GET64(traffic_counter, stats, octets);
+}

drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.h

@@ -76,4 +76,16 @@ int mlx5_cmd_fc_alloc(struct mlx5_core_dev *dev, u16 *id);
 int mlx5_cmd_fc_free(struct mlx5_core_dev *dev, u16 id);
 int mlx5_cmd_fc_query(struct mlx5_core_dev *dev, u16 id,
 		      u64 *packets, u64 *bytes);
+
+struct mlx5_cmd_fc_bulk;
+
+struct mlx5_cmd_fc_bulk *
+mlx5_cmd_fc_bulk_alloc(struct mlx5_core_dev *dev, u16 id, int num);
+void mlx5_cmd_fc_bulk_free(struct mlx5_cmd_fc_bulk *b);
+int
+mlx5_cmd_fc_bulk_query(struct mlx5_core_dev *dev, struct mlx5_cmd_fc_bulk *b);
+void mlx5_cmd_fc_bulk_get(struct mlx5_core_dev *dev,
+			  struct mlx5_cmd_fc_bulk *b, u16 id,
+			  u64 *packets, u64 *bytes);
+
 #endif

drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c

@@ -90,16 +90,66 @@ static void mlx5_fc_stats_insert(struct rb_root *root, struct mlx5_fc *counter)
 	rb_insert_color(&counter->node, root);
 }
 
+static struct rb_node *mlx5_fc_stats_query(struct mlx5_core_dev *dev,
+					   struct mlx5_fc *first,
+					   u16 last_id)
+{
+	struct mlx5_cmd_fc_bulk *b;
+	struct rb_node *node = NULL;
+	u16 afirst_id;
+	int num;
+	int err;
+	int max_bulk = 1 << MLX5_CAP_GEN(dev, log_max_flow_counter_bulk);
+
+	/* first id must be aligned to 4 when using bulk query */
+	afirst_id = first->id & ~0x3;
+
+	/* number of counters to query inc. the last counter */
+	num = ALIGN(last_id - afirst_id + 1, 4);
+	if (num > max_bulk) {
+		num = max_bulk;
+		last_id = afirst_id + num - 1;
+	}
+
+	b = mlx5_cmd_fc_bulk_alloc(dev, afirst_id, num);
+	if (!b) {
+		mlx5_core_err(dev, "Error allocating resources for bulk query\n");
+		return NULL;
+	}
+
+	err = mlx5_cmd_fc_bulk_query(dev, b);
+	if (err) {
+		mlx5_core_err(dev, "Error doing bulk query: %d\n", err);
+		goto out;
+	}
+
+	for (node = &first->node; node; node = rb_next(node)) {
+		struct mlx5_fc *counter = rb_entry(node, struct mlx5_fc, node);
+		struct mlx5_fc_cache *c = &counter->cache;
+
+		if (counter->id > last_id)
+			break;
+
+		mlx5_cmd_fc_bulk_get(dev, b,
+				     counter->id, &c->packets, &c->bytes);
+	}
+
+out:
+	mlx5_cmd_fc_bulk_free(b);
+
+	return node;
+}
+
 static void mlx5_fc_stats_work(struct work_struct *work)
 {
 	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;
 	unsigned long now = jiffies;
-	struct mlx5_fc *counter;
+	struct mlx5_fc *counter = NULL;
+	struct mlx5_fc *last = NULL;
 	struct rb_node *node;
 	LIST_HEAD(tmplist);
-	int err = 0;
 
 	spin_lock(&fc_stats->addlist_lock);
 
@@ -115,12 +165,7 @@ static void mlx5_fc_stats_work(struct work_struct *work)
 
 	node = rb_first(&fc_stats->counters);
 	while (node) {
-		struct mlx5_fc_cache *c;
-		u64 packets;
-		u64 bytes;
-
 		counter = rb_entry(node, struct mlx5_fc, node);
-		c = &counter->cache;
 
 		node = rb_next(node);
 
@@ -133,26 +178,20 @@ static void mlx5_fc_stats_work(struct work_struct *work)
 			continue;
 		}
 
-		if (time_before(now, fc_stats->next_query))
-			continue;
+		last = counter;
+	}
 
-		err = mlx5_cmd_fc_query(dev, counter->id, &packets, &bytes);
-		if (err) {
-			pr_err("Error querying stats for counter id %d\n",
-			       counter->id);
-			continue;
-		}
+	if (time_before(now, fc_stats->next_query) || !last)
+		return;
 
-		if (packets == c->packets)
-			continue;
+	node = rb_first(&fc_stats->counters);
+	while (node) {
+		counter = rb_entry(node, struct mlx5_fc, node);
 
-		c->lastuse = jiffies;
-		c->packets = packets;
-		c->bytes   = bytes;
+		node = mlx5_fc_stats_query(dev, counter, last->id);
 	}
 
-	if (time_after_eq(now, fc_stats->next_query))
-		fc_stats->next_query = now + MLX5_FC_STATS_PERIOD;
+	fc_stats->next_query = now + MLX5_FC_STATS_PERIOD;
 }
 
 struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging)

include/linux/mlx5/mlx5_ifc.h

@@ -893,7 +893,10 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 	u8         reserved_at_330[0xb];
 	u8         log_max_xrcd[0x5];
 
-	u8         reserved_at_340[0x20];
+	u8         reserved_at_340[0x8];
+	u8         log_max_flow_counter_bulk[0x8];
+	u8         max_flow_counter[0x10];
+
 
 	u8         reserved_at_360[0x3];
 	u8         log_max_rq[0x5];
@@ -980,7 +983,8 @@ struct mlx5_ifc_dest_format_struct_bits {
 };
 
 struct mlx5_ifc_flow_counter_list_bits {
-	u8         reserved_at_0[0x10];
+	u8         clear[0x1];
+	u8         num_of_counters[0xf];
 	u8         flow_counter_id[0x10];
 
 	u8         reserved_at_20[0x20];