net: airoha: Add FLOW_CLS_STATS callback support

Introduce per-flow stats accounting to the flowtable hw offload in
the airoha_eth driver. Flow stats are split in the PPE and NPU modules:
- PPE: accounts for high 32bit of per-flow stats
- NPU: accounts for low 32bit of per-flow stats

FLOW_CLS_STATS can be enabled or disabled at compile time.

Signed-off-by: Lorenzo Bianconi <[e-mail]>
Reviewed-by: Simon Horman <[e-mail]>
Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Jakub Kicinski <[e-mail]>

drivers/net/ethernet/airoha/Kconfig

···
 	  This driver supports the gigabit ethernet MACs in the
 	  Airoha SoC family.
 
+config NET_AIROHA_FLOW_STATS
+	default y
+	bool "Airoha flow stats"
+	depends on NET_AIROHA && NET_AIROHA_NPU
+	help
+	  Enable Aiorha flowtable statistic counters.
+
 endif #NET_VENDOR_AIROHA

drivers/net/ethernet/airoha/airoha_eth.h

···
 #define PPE_NUM				2
 #define PPE1_SRAM_NUM_ENTRIES		(8 * 1024)
 #define PPE_SRAM_NUM_ENTRIES		(2 * PPE1_SRAM_NUM_ENTRIES)
+#ifdef CONFIG_NET_AIROHA_FLOW_STATS
+#define PPE1_STATS_NUM_ENTRIES		(4 * 1024)
+#else
+#define PPE1_STATS_NUM_ENTRIES		0
+#endif /* CONFIG_NET_AIROHA_FLOW_STATS */
+#define PPE_STATS_NUM_ENTRIES		(2 * PPE1_STATS_NUM_ENTRIES)
+#define PPE1_SRAM_NUM_DATA_ENTRIES	(PPE1_SRAM_NUM_ENTRIES - PPE1_STATS_NUM_ENTRIES)
+#define PPE_SRAM_NUM_DATA_ENTRIES	(2 * PPE1_SRAM_NUM_DATA_ENTRIES)
 #define PPE_DRAM_NUM_ENTRIES		(16 * 1024)
 #define PPE_NUM_ENTRIES			(PPE_SRAM_NUM_ENTRIES + PPE_DRAM_NUM_ENTRIES)
 #define PPE_HASH_MASK			(PPE_NUM_ENTRIES - 1)
···
 
 	u16 pppoe_id;
 	u16 src_mac_lo;
+
+	u32 meter;
 };
 
 #define AIROHA_FOE_IB1_UNBIND_PREBIND		BIT(24)
···
 #define AIROHA_FOE_DPI				BIT(7)
 #define AIROHA_FOE_TUNNEL			BIT(6)
 #define AIROHA_FOE_TUNNEL_ID			GENMASK(5, 0)
+
+#define AIROHA_FOE_TUNNEL_MTU			GENMASK(31, 16)
+#define AIROHA_FOE_ACNT_GRP3			GENMASK(15, 9)
+#define AIROHA_FOE_METER_GRP3			GENMASK(8, 5)
+#define AIROHA_FOE_METER_GRP2			GENMASK(4, 0)
 
 struct airoha_foe_bridge {
 	u32 dest_mac_hi;
···
 	u32 ib2;
 
 	struct airoha_foe_mac_info_common l2;
+
+	u32 meter;
 };
 
 struct airoha_foe_entry {
···
 		};
 		u8 data[PPE_ENTRY_SIZE];
 	};
+};
+
+struct airoha_foe_stats {
+	u32 bytes;
+	u32 packets;
+};
+
+struct airoha_foe_stats64 {
+	u64 bytes;
+	u64 packets;
 };
 
 struct airoha_flow_data {
···
 	struct hlist_node l2_subflow_node; /* PPE L2 subflow entry */
 	u32 hash;
 
+	struct airoha_foe_stats64 stats;
 	enum airoha_flow_entry_type type;
 
 	struct rhash_head node;
···
 	struct hlist_head *foe_flow;
 	u16 foe_check_time[PPE_NUM_ENTRIES];
 
+	struct airoha_foe_stats *foe_stats;
+	dma_addr_t foe_stats_dma;
+
 	struct dentry *debugfs_dir;
 };
 
···
 void airoha_ppe_deinit(struct airoha_eth *eth);
 struct airoha_foe_entry *airoha_ppe_foe_get_entry(struct airoha_ppe *ppe,
 						  u32 hash);
+void airoha_ppe_foe_entry_get_stats(struct airoha_ppe *ppe, u32 hash,
+				    struct airoha_foe_stats64 *stats);
 
 #ifdef CONFIG_DEBUG_FS
 int airoha_ppe_debugfs_init(struct airoha_ppe *ppe);

drivers/net/ethernet/airoha/airoha_npu.c

···
 #include <linux/of_reserved_mem.h>
 #include <linux/regmap.h>
 
+#include "airoha_eth.h"
 #include "airoha_npu.h"
 
 #define NPU_EN7581_FIRMWARE_DATA		"airoha/en7581_npu_data.bin"
···
 	PPE_FUNC_SET_WAIT_HWNAT_INIT,
 	PPE_FUNC_SET_WAIT_HWNAT_DEINIT,
 	PPE_FUNC_SET_WAIT_API,
+	PPE_FUNC_SET_WAIT_FLOW_STATS_SETUP,
 };
 
 enum {
···
 			u32 size;
 			u32 data;
 		} set_info;
+		struct {
+			u32 npu_stats_addr;
+			u32 foe_stats_addr;
+		} stats_info;
 	};
 };
 
···
 	return err;
 }
 
-struct airoha_npu *airoha_npu_get(struct device *dev)
+static int airoha_npu_stats_setup(struct airoha_npu *npu,
+				  dma_addr_t foe_stats_addr)
+{
+	int err, size = PPE_STATS_NUM_ENTRIES * sizeof(*npu->stats);
+	struct ppe_mbox_data *ppe_data;
+
+	if (!size) /* flow stats are disabled */
+		return 0;
+
+	ppe_data = kzalloc(sizeof(*ppe_data), GFP_ATOMIC);
+	if (!ppe_data)
+		return -ENOMEM;
+
+	ppe_data->func_type = NPU_OP_SET;
+	ppe_data->func_id = PPE_FUNC_SET_WAIT_FLOW_STATS_SETUP;
+	ppe_data->stats_info.foe_stats_addr = foe_stats_addr;
+
+	err = airoha_npu_send_msg(npu, NPU_FUNC_PPE, ppe_data,
+				  sizeof(*ppe_data));
+	if (err)
+		goto out;
+
+	npu->stats = devm_ioremap(npu->dev,
+				  ppe_data->stats_info.npu_stats_addr,
+				  size);
+	if (!npu->stats)
+		err = -ENOMEM;
+out:
+	kfree(ppe_data);
+
+	return err;
+}
+
+struct airoha_npu *airoha_npu_get(struct device *dev, dma_addr_t *stats_addr)
 {
 	struct platform_device *pdev;
 	struct device_node *np;
···
 			dev_name(dev));
 		npu = ERR_PTR(-EINVAL);
 		goto error_module_put;
+	}
+
+	if (stats_addr) {
+		int err;
+
+		err = airoha_npu_stats_setup(npu, *stats_addr);
+		if (err) {
+			dev_err(dev, "failed to allocate npu stats buffer\n");
+			npu = ERR_PTR(err);
+			goto error_module_put;
+		}
 	}
 
 	return npu;

drivers/net/ethernet/airoha/airoha_npu.h

···
 		struct work_struct wdt_work;
 	} cores[NPU_NUM_CORES];
 
+	struct airoha_foe_stats __iomem *stats;
+
 	struct {
 		int (*ppe_init)(struct airoha_npu *npu);
 		int (*ppe_deinit)(struct airoha_npu *npu);
···
 	} ops;
 };
 
-struct airoha_npu *airoha_npu_get(struct device *dev);
+struct airoha_npu *airoha_npu_get(struct device *dev, dma_addr_t *stats_addr);
 void airoha_npu_put(struct airoha_npu *npu);

drivers/net/ethernet/airoha/airoha_ppe.c

···
 
 	if (airoha_ppe2_is_enabled(eth)) {
 		sram_num_entries =
-			PPE_RAM_NUM_ENTRIES_SHIFT(PPE1_SRAM_NUM_ENTRIES);
+			PPE_RAM_NUM_ENTRIES_SHIFT(PPE1_SRAM_NUM_DATA_ENTRIES);
 		airoha_fe_rmw(eth, REG_PPE_TB_CFG(0),
 			      PPE_SRAM_TB_NUM_ENTRY_MASK |
 			      PPE_DRAM_TB_NUM_ENTRY_MASK,
···
 					 dram_num_entries));
 	} else {
 		sram_num_entries =
-			PPE_RAM_NUM_ENTRIES_SHIFT(PPE_SRAM_NUM_ENTRIES);
+			PPE_RAM_NUM_ENTRIES_SHIFT(PPE_SRAM_NUM_DATA_ENTRIES);
 		airoha_fe_rmw(eth, REG_PPE_TB_CFG(0),
 			      PPE_SRAM_TB_NUM_ENTRY_MASK |
 			      PPE_DRAM_TB_NUM_ENTRY_MASK,
···
 	return hash;
 }
 
+static u32 airoha_ppe_foe_get_flow_stats_index(struct airoha_ppe *ppe, u32 hash)
+{
+	if (!airoha_ppe2_is_enabled(ppe->eth))
+		return hash;
+
+	return hash >= PPE_STATS_NUM_ENTRIES ? hash - PPE1_STATS_NUM_ENTRIES
+					     : hash;
+}
+
+static void airoha_ppe_foe_flow_stat_entry_reset(struct airoha_ppe *ppe,
+						 struct airoha_npu *npu,
+						 int index)
+{
+	memset_io(&npu->stats[index], 0, sizeof(*npu->stats));
+	memset(&ppe->foe_stats[index], 0, sizeof(*ppe->foe_stats));
+}
+
+static void airoha_ppe_foe_flow_stats_reset(struct airoha_ppe *ppe,
+					    struct airoha_npu *npu)
+{
+	int i;
+
+	for (i = 0; i < PPE_STATS_NUM_ENTRIES; i++)
+		airoha_ppe_foe_flow_stat_entry_reset(ppe, npu, i);
+}
+
+static void airoha_ppe_foe_flow_stats_update(struct airoha_ppe *ppe,
+					     struct airoha_npu *npu,
+					     struct airoha_foe_entry *hwe,
+					     u32 hash)
+{
+	int type = FIELD_GET(AIROHA_FOE_IB1_BIND_PACKET_TYPE, hwe->ib1);
+	u32 index, pse_port, val, *data, *ib2, *meter;
+	u8 nbq;
+
+	index = airoha_ppe_foe_get_flow_stats_index(ppe, hash);
+	if (index >= PPE_STATS_NUM_ENTRIES)
+		return;
+
+	if (type == PPE_PKT_TYPE_BRIDGE) {
+		data = &hwe->bridge.data;
+		ib2 = &hwe->bridge.ib2;
+		meter = &hwe->bridge.l2.meter;
+	} else if (type >= PPE_PKT_TYPE_IPV6_ROUTE_3T) {
+		data = &hwe->ipv6.data;
+		ib2 = &hwe->ipv6.ib2;
+		meter = &hwe->ipv6.meter;
+	} else {
+		data = &hwe->ipv4.data;
+		ib2 = &hwe->ipv4.ib2;
+		meter = &hwe->ipv4.l2.meter;
+	}
+
+	airoha_ppe_foe_flow_stat_entry_reset(ppe, npu, index);
+
+	val = FIELD_GET(AIROHA_FOE_CHANNEL | AIROHA_FOE_QID, *data);
+	*data = (*data & ~AIROHA_FOE_ACTDP) |
+		FIELD_PREP(AIROHA_FOE_ACTDP, val);
+
+	val = *ib2 & (AIROHA_FOE_IB2_NBQ | AIROHA_FOE_IB2_PSE_PORT |
+		      AIROHA_FOE_IB2_PSE_QOS | AIROHA_FOE_IB2_FAST_PATH);
+	*meter |= FIELD_PREP(AIROHA_FOE_TUNNEL_MTU, val);
+
+	pse_port = FIELD_GET(AIROHA_FOE_IB2_PSE_PORT, *ib2);
+	nbq = pse_port == 1 ? 6 : 5;
+	*ib2 &= ~(AIROHA_FOE_IB2_NBQ | AIROHA_FOE_IB2_PSE_PORT |
+		  AIROHA_FOE_IB2_PSE_QOS);
+	*ib2 |= FIELD_PREP(AIROHA_FOE_IB2_PSE_PORT, 6) |
+		FIELD_PREP(AIROHA_FOE_IB2_NBQ, nbq);
+}
+
 struct airoha_foe_entry *airoha_ppe_foe_get_entry(struct airoha_ppe *ppe,
 						  u32 hash)
 {
···
 	struct airoha_foe_entry *hwe = ppe->foe + hash * sizeof(*hwe);
 	u32 ts = airoha_ppe_get_timestamp(ppe);
 	struct airoha_eth *eth = ppe->eth;
+	struct airoha_npu *npu;
+	int err = 0;
 
 	memcpy(&hwe->d, &e->d, sizeof(*hwe) - sizeof(hwe->ib1));
 	wmb();
···
 	e->ib1 |= FIELD_PREP(AIROHA_FOE_IB1_BIND_TIMESTAMP, ts);
 	hwe->ib1 = e->ib1;
 
+	rcu_read_lock();
+
+	npu = rcu_dereference(eth->npu);
+	if (!npu) {
+		err = -ENODEV;
+		goto unlock;
+	}
+
+	airoha_ppe_foe_flow_stats_update(ppe, npu, hwe, hash);
+
 	if (hash < PPE_SRAM_NUM_ENTRIES) {
 		dma_addr_t addr = ppe->foe_dma + hash * sizeof(*hwe);
 		bool ppe2 = airoha_ppe2_is_enabled(eth) &&
 			    hash >= PPE1_SRAM_NUM_ENTRIES;
-		struct airoha_npu *npu;
-		int err = -ENODEV;
 
-		rcu_read_lock();
-		npu = rcu_dereference(eth->npu);
-		if (npu)
-			err = npu->ops.ppe_foe_commit_entry(npu, addr,
-							    sizeof(*hwe), hash,
-							    ppe2);
-		rcu_read_unlock();
-
-		return err;
+		err = npu->ops.ppe_foe_commit_entry(npu, addr, sizeof(*hwe),
+						    hash, ppe2);
 	}
+unlock:
+	rcu_read_unlock();
 
-	return 0;
+	return err;
 }
 
 static void airoha_ppe_foe_remove_flow(struct airoha_ppe *ppe,
···
 		l2->common.etype = ETH_P_IPV6;
 
 	hwe.bridge.ib2 = e->data.bridge.ib2;
+	hwe.bridge.data = e->data.bridge.data;
 	airoha_ppe_foe_commit_entry(ppe, &hwe, hash);
 
 	return 0;
···
 	spin_unlock_bh(&ppe_lock);
 
 	return 0;
+}
+
+static int airoha_ppe_get_entry_idle_time(struct airoha_ppe *ppe, u32 ib1)
+{
+	u32 state = FIELD_GET(AIROHA_FOE_IB1_BIND_STATE, ib1);
+	u32 ts, ts_mask, now = airoha_ppe_get_timestamp(ppe);
+	int idle;
+
+	if (state == AIROHA_FOE_STATE_BIND) {
+		ts = FIELD_GET(AIROHA_FOE_IB1_BIND_TIMESTAMP, ib1);
+		ts_mask = AIROHA_FOE_IB1_BIND_TIMESTAMP;
+	} else {
+		ts = FIELD_GET(AIROHA_FOE_IB1_UNBIND_TIMESTAMP, ib1);
+		now = FIELD_GET(AIROHA_FOE_IB1_UNBIND_TIMESTAMP, now);
+		ts_mask = AIROHA_FOE_IB1_UNBIND_TIMESTAMP;
+	}
+	idle = now - ts;
+
+	return idle < 0 ? idle + ts_mask + 1 : idle;
+}
+
+static void
+airoha_ppe_foe_flow_l2_entry_update(struct airoha_ppe *ppe,
+				    struct airoha_flow_table_entry *e)
+{
+	int min_idle = airoha_ppe_get_entry_idle_time(ppe, e->data.ib1);
+	struct airoha_flow_table_entry *iter;
+	struct hlist_node *n;
+
+	lockdep_assert_held(&ppe_lock);
+
+	hlist_for_each_entry_safe(iter, n, &e->l2_flows, l2_subflow_node) {
+		struct airoha_foe_entry *hwe;
+		u32 ib1, state;
+		int idle;
+
+		hwe = airoha_ppe_foe_get_entry(ppe, iter->hash);
+		ib1 = READ_ONCE(hwe->ib1);
+
+		state = FIELD_GET(AIROHA_FOE_IB1_BIND_STATE, ib1);
+		if (state != AIROHA_FOE_STATE_BIND) {
+			iter->hash = 0xffff;
+			airoha_ppe_foe_remove_flow(ppe, iter);
+			continue;
+		}
+
+		idle = airoha_ppe_get_entry_idle_time(ppe, ib1);
+		if (idle >= min_idle)
+			continue;
+
+		min_idle = idle;
+		e->data.ib1 &= ~AIROHA_FOE_IB1_BIND_TIMESTAMP;
+		e->data.ib1 |= ib1 & AIROHA_FOE_IB1_BIND_TIMESTAMP;
+	}
+}
+
+static void airoha_ppe_foe_flow_entry_update(struct airoha_ppe *ppe,
+					     struct airoha_flow_table_entry *e)
+{
+	struct airoha_foe_entry *hwe_p, hwe = {};
+
+	spin_lock_bh(&ppe_lock);
+
+	if (e->type == FLOW_TYPE_L2) {
+		airoha_ppe_foe_flow_l2_entry_update(ppe, e);
+		goto unlock;
+	}
+
+	if (e->hash == 0xffff)
+		goto unlock;
+
+	hwe_p = airoha_ppe_foe_get_entry(ppe, e->hash);
+	if (!hwe_p)
+		goto unlock;
+
+	memcpy(&hwe, hwe_p, sizeof(*hwe_p));
+	if (!airoha_ppe_foe_compare_entry(e, &hwe)) {
+		e->hash = 0xffff;
+		goto unlock;
+	}
+
+	e->data.ib1 = hwe.ib1;
+unlock:
+	spin_unlock_bh(&ppe_lock);
+}
+
+static int airoha_ppe_entry_idle_time(struct airoha_ppe *ppe,
+				      struct airoha_flow_table_entry *e)
+{
+	airoha_ppe_foe_flow_entry_update(ppe, e);
+
+	return airoha_ppe_get_entry_idle_time(ppe, e->data.ib1);
 }
 
 static int airoha_ppe_flow_offload_replace(struct airoha_gdm_port *port,
···
 	return 0;
 }
 
+void airoha_ppe_foe_entry_get_stats(struct airoha_ppe *ppe, u32 hash,
+				    struct airoha_foe_stats64 *stats)
+{
+	u32 index = airoha_ppe_foe_get_flow_stats_index(ppe, hash);
+	struct airoha_eth *eth = ppe->eth;
+	struct airoha_npu *npu;
+
+	if (index >= PPE_STATS_NUM_ENTRIES)
+		return;
+
+	rcu_read_lock();
+
+	npu = rcu_dereference(eth->npu);
+	if (npu) {
+		u64 packets = ppe->foe_stats[index].packets;
+		u64 bytes = ppe->foe_stats[index].bytes;
+		struct airoha_foe_stats npu_stats;
+
+		memcpy_fromio(&npu_stats, &npu->stats[index],
+			      sizeof(*npu->stats));
+		stats->packets = packets << 32 | npu_stats.packets;
+		stats->bytes = bytes << 32 | npu_stats.bytes;
+	}
+
+	rcu_read_unlock();
+}
+
+static int airoha_ppe_flow_offload_stats(struct airoha_gdm_port *port,
+					 struct flow_cls_offload *f)
+{
+	struct airoha_eth *eth = port->qdma->eth;
+	struct airoha_flow_table_entry *e;
+	u32 idle;
+
+	e = rhashtable_lookup(&eth->flow_table, &f->cookie,
+			      airoha_flow_table_params);
+	if (!e)
+		return -ENOENT;
+
+	idle = airoha_ppe_entry_idle_time(eth->ppe, e);
+	f->stats.lastused = jiffies - idle * HZ;
+
+	if (e->hash != 0xffff) {
+		struct airoha_foe_stats64 stats = {};
+
+		airoha_ppe_foe_entry_get_stats(eth->ppe, e->hash, &stats);
+		f->stats.pkts += (stats.packets - e->stats.packets);
+		f->stats.bytes += (stats.bytes - e->stats.bytes);
+		e->stats = stats;
+	}
+
+	return 0;
+}
+
 static int airoha_ppe_flow_offload_cmd(struct airoha_gdm_port *port,
 				       struct flow_cls_offload *f)
 {
···
 		return airoha_ppe_flow_offload_replace(port, f);
 	case FLOW_CLS_DESTROY:
 		return airoha_ppe_flow_offload_destroy(port, f);
+	case FLOW_CLS_STATS:
+		return airoha_ppe_flow_offload_stats(port, f);
 	default:
 		break;
 	}
···
 
 static struct airoha_npu *airoha_ppe_npu_get(struct airoha_eth *eth)
 {
-	struct airoha_npu *npu = airoha_npu_get(eth->dev);
+	struct airoha_npu *npu = airoha_npu_get(eth->dev,
+						&eth->ppe->foe_stats_dma);
 
 	if (IS_ERR(npu)) {
 		request_module("airoha-npu");
-		npu = airoha_npu_get(eth->dev);
+		npu = airoha_npu_get(eth->dev, &eth->ppe->foe_stats_dma);
 	}
 
 	return npu;
···
 	err = airoha_ppe_flush_sram_entries(eth->ppe, npu);
 	if (err)
 		goto error_npu_put;
+
+	airoha_ppe_foe_flow_stats_reset(eth->ppe, npu);
 
 	rcu_assign_pointer(eth->npu, npu);
 	synchronize_rcu();
···
 				     GFP_KERNEL);
 	if (!ppe->foe_flow)
 		return -ENOMEM;
+
+	foe_size = PPE_STATS_NUM_ENTRIES * sizeof(*ppe->foe_stats);
+	if (foe_size) {
+		ppe->foe_stats = dmam_alloc_coherent(eth->dev, foe_size,
+						     &ppe->foe_stats_dma,
+						     GFP_KERNEL);
+		if (!ppe->foe_stats)
+			return -ENOMEM;
+	}
 
 	err = rhashtable_init(&eth->flow_table, &airoha_flow_table_params);
 	if (err)

drivers/net/ethernet/airoha/airoha_ppe_debugfs.c

···
 		u16 *src_port = NULL, *dest_port = NULL;
 		struct airoha_foe_mac_info_common *l2;
 		unsigned char h_source[ETH_ALEN] = {};
+		struct airoha_foe_stats64 stats = {};
 		unsigned char h_dest[ETH_ALEN];
 		struct airoha_foe_entry *hwe;
 		u32 type, state, ib2, data;
···
 				cpu_to_be16(hwe->ipv4.l2.src_mac_lo);
 		}
 
+		airoha_ppe_foe_entry_get_stats(ppe, i, &stats);
+
 		*((__be32 *)h_dest) = cpu_to_be32(l2->dest_mac_hi);
 		*((__be16 *)&h_dest[4]) = cpu_to_be16(l2->dest_mac_lo);
 		*((__be32 *)h_source) = cpu_to_be32(l2->src_mac_hi);
 
 		seq_printf(m, " eth=%pM->%pM etype=%04x data=%08x"
-			      " vlan=%d,%d ib1=%08x ib2=%08x\n",
+			      " vlan=%d,%d ib1=%08x ib2=%08x"
+			      " packets=%llu bytes=%llu\n",
 			   h_source, h_dest, l2->etype, data,
-			   l2->vlan1, l2->vlan2, hwe->ib1, ib2);
+			   l2->vlan1, l2->vlan2, hwe->ib1, ib2,
+			   stats.packets, stats.bytes);
 	}
 
 	return 0;