cxgb4/chcr: Add ipv6 support and statistics
authorRohit Maheshwari <rohitm@chelsio.com>
Sat, 7 Mar 2020 14:36:08 +0000 (20:06 +0530)
committerDavid S. Miller <davem@davemloft.net>
Mon, 9 Mar 2020 04:16:23 +0000 (21:16 -0700)
Adding ipv6 support and ktls related statistics.

v1->v2:
- added blank lines at 2 places.

v3->v4:
- Replaced atomic_t with atomic64_t
- added few necessary stat counters.

Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/crypto/chelsio/chcr_ktls.c
drivers/crypto/chelsio/chcr_ktls.h
drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c
drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h

index 5dff444..f0c3834 100644 (file)
@@ -3,6 +3,7 @@
 
 #ifdef CONFIG_CHELSIO_TLS_DEVICE
 #include "chcr_ktls.h"
+#include "clip_tbl.h"
 
 static int chcr_init_tcb_fields(struct chcr_ktls_info *tx_info);
 /*
@@ -153,8 +154,10 @@ static int chcr_ktls_update_connection_state(struct chcr_ktls_info *tx_info,
                /* FALLTHRU */
        case KTLS_CONN_SET_TCB_RPL:
                /* Check if l2t state is valid, then move to ready state. */
-               if (cxgb4_check_l2t_valid(tx_info->l2te))
+               if (cxgb4_check_l2t_valid(tx_info->l2te)) {
                        tx_info->connection_state = KTLS_CONN_TX_READY;
+                       atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_ctx);
+               }
                break;
 
        case KTLS_CONN_TX_READY:
@@ -219,6 +222,56 @@ static int chcr_ktls_act_open_req(struct sock *sk,
        return cxgb4_l2t_send(tx_info->netdev, skb, tx_info->l2te);
 }
 
+/*
+ * chcr_ktls_act_open_req6: creates TCB entry for ipv6 connection.
+ * @sk - tcp socket.
+ * @tx_info - driver specific tls info.
+ * @atid - connection active tid.
+ * return - send success/failure.
+ */
+static int chcr_ktls_act_open_req6(struct sock *sk,
+                                  struct chcr_ktls_info *tx_info,
+                                  int atid)
+{
+       struct inet_sock *inet = inet_sk(sk);
+       struct cpl_t6_act_open_req6 *cpl6;
+       struct cpl_act_open_req6 *cpl;
+       struct sk_buff *skb;
+       unsigned int len;
+       int qid_atid;
+       u64 options;
+
+       len = sizeof(*cpl6);
+       skb = alloc_skb(len, GFP_KERNEL);
+       if (unlikely(!skb))
+               return -ENOMEM;
+       /* mark it a control pkt */
+       set_wr_txq(skb, CPL_PRIORITY_CONTROL, tx_info->port_id);
+
+       cpl6 = __skb_put_zero(skb, len);
+       cpl = (struct cpl_act_open_req6 *)cpl6;
+       INIT_TP_WR(cpl6, 0);
+       qid_atid = TID_QID_V(tx_info->rx_qid) | TID_TID_V(atid);
+       OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, qid_atid));
+       cpl->local_port = inet->inet_sport;
+       cpl->peer_port = inet->inet_dport;
+       cpl->local_ip_hi = *(__be64 *)&sk->sk_v6_rcv_saddr.in6_u.u6_addr8[0];
+       cpl->local_ip_lo = *(__be64 *)&sk->sk_v6_rcv_saddr.in6_u.u6_addr8[8];
+       cpl->peer_ip_hi = *(__be64 *)&sk->sk_v6_daddr.in6_u.u6_addr8[0];
+       cpl->peer_ip_lo = *(__be64 *)&sk->sk_v6_daddr.in6_u.u6_addr8[8];
+
+       /* first 64 bit option field. */
+       options = TCAM_BYPASS_F | ULP_MODE_V(ULP_MODE_NONE) | NON_OFFLOAD_F |
+                 SMAC_SEL_V(tx_info->smt_idx) | TX_CHAN_V(tx_info->tx_chan);
+       cpl->opt0 = cpu_to_be64(options);
+       /* next 64 bit option field. */
+       options =
+               TX_QUEUE_V(tx_info->adap->params.tp.tx_modq[tx_info->tx_chan]);
+       cpl->opt2 = htonl(options);
+
+       return cxgb4_l2t_send(tx_info->netdev, skb, tx_info->l2te);
+}
+
 /*
  * chcr_setup_connection:  create a TCB entry so that TP will form tcp packets.
  * @sk - tcp socket.
@@ -245,7 +298,13 @@ static int chcr_setup_connection(struct sock *sk,
                ret = chcr_ktls_act_open_req(sk, tx_info, atid);
        } else {
                tx_info->ip_family = AF_INET6;
-               ret = -EOPNOTSUPP;
+               ret =
+               cxgb4_clip_get(tx_info->netdev,
+                              (const u32 *)&sk->sk_v6_rcv_saddr.in6_u.u6_addr8,
+                              1);
+               if (ret)
+                       goto out;
+               ret = chcr_ktls_act_open_req6(sk, tx_info, atid);
        }
 
        /* if return type is NET_XMIT_CN, msg will be sent but delayed, mark ret
@@ -322,23 +381,35 @@ static void chcr_ktls_dev_del(struct net_device *netdev,
        struct chcr_ktls_ofld_ctx_tx *tx_ctx =
                                chcr_get_ktls_tx_context(tls_ctx);
        struct chcr_ktls_info *tx_info = tx_ctx->chcr_info;
+       struct sock *sk;
 
        if (!tx_info)
                return;
+       sk = tx_info->sk;
 
        spin_lock(&tx_info->lock);
        tx_info->connection_state = KTLS_CONN_CLOSED;
        spin_unlock(&tx_info->lock);
 
+       /* clear l2t entry */
        if (tx_info->l2te)
                cxgb4_l2t_release(tx_info->l2te);
 
+       /* clear clip entry */
+       if (tx_info->ip_family == AF_INET6)
+               cxgb4_clip_release(netdev,
+                                  (const u32 *)&sk->sk_v6_daddr.in6_u.u6_addr8,
+                                  1);
+
+       /* clear tid */
        if (tx_info->tid != -1) {
                /* clear tcb state and then release tid */
                chcr_ktls_mark_tcb_close(tx_info);
                cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
                                 tx_info->tid, tx_info->ip_family);
        }
+
+       atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_connection_close);
        kvfree(tx_info);
        tx_ctx->chcr_info = NULL;
 }
@@ -424,7 +495,7 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
             ipv6_addr_type(&sk->sk_v6_daddr) == IPV6_ADDR_MAPPED)) {
                memcpy(daaddr, &sk->sk_daddr, 4);
        } else {
-               goto out2;
+               memcpy(daaddr, sk->sk_v6_daddr.in6_u.u6_addr8, 16);
        }
 
        /* get the l2t index */
@@ -458,10 +529,12 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
        if (ret)
                goto out2;
 
+       atomic64_inc(&adap->chcr_stats.ktls_tx_connection_open);
        return 0;
 out2:
        kvfree(tx_info);
 out:
+       atomic64_inc(&adap->chcr_stats.ktls_tx_connection_fail);
        return ret;
 }
 
@@ -729,6 +802,7 @@ static int chcr_ktls_xmit_tcb_cpls(struct chcr_ktls_info *tx_info,
                                                 TCB_SND_UNA_RAW_V
                                                 (TCB_SND_UNA_RAW_M),
                                                 TCB_SND_UNA_RAW_V(0), 0);
+               atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_ooo);
                cpl++;
        }
        /* update ack */
@@ -1152,6 +1226,7 @@ static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
 
        chcr_txq_advance(&q->q, ndesc);
        cxgb4_ring_tx_db(adap, &q->q, ndesc);
+       atomic64_inc(&adap->chcr_stats.ktls_tx_send_records);
 
        return 0;
 }
@@ -1562,6 +1637,7 @@ static int chcr_end_part_handler(struct chcr_ktls_info *tx_info,
        /* check if it is a complete record */
        if (tls_end_offset == record->len) {
                nskb = skb;
+               atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_complete_pkts);
        } else {
                dev_kfree_skb_any(skb);
 
@@ -1579,6 +1655,7 @@ static int chcr_end_part_handler(struct chcr_ktls_info *tx_info,
                 */
                if (chcr_ktls_update_snd_una(tx_info, q))
                        goto out;
+               atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_end_pkts);
        }
 
        if (chcr_ktls_xmit_wr_complete(nskb, tx_info, q, tcp_seq,
@@ -1649,6 +1726,7 @@ static int chcr_short_record_handler(struct chcr_ktls_info *tx_info,
                /* free the last trimmed portion */
                dev_kfree_skb_any(skb);
                skb = tmp_skb;
+               atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_trimmed_pkts);
        }
        data_len = skb->data_len;
        /* check if the middle record's start point is 16 byte aligned. CTR
@@ -1720,6 +1798,7 @@ static int chcr_short_record_handler(struct chcr_ktls_info *tx_info,
                 */
                if (chcr_ktls_update_snd_una(tx_info, q))
                        goto out;
+               atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_middle_pkts);
        } else {
                /* Else means, its a partial first part of the record. Check if
                 * its only the header, don't need to send for encryption then.
@@ -1734,6 +1813,7 @@ static int chcr_short_record_handler(struct chcr_ktls_info *tx_info,
                        }
                        return 0;
                }
+               atomic64_inc(&tx_info->adap->chcr_stats.ktls_tx_start_pkts);
        }
 
        if (chcr_ktls_xmit_wr_short(skb, tx_info, q, tcp_seq, tcp_push_no_fin,
@@ -1755,6 +1835,7 @@ int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
        struct tcphdr *th = tcp_hdr(skb);
        int data_len, qidx, ret = 0, mss;
        struct tls_record_info *record;
+       struct chcr_stats_debug *stats;
        struct chcr_ktls_info *tx_info;
        u32 tls_end_offset, tcp_seq;
        struct tls_context *tls_ctx;
@@ -1800,6 +1881,8 @@ int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
                return NETDEV_TX_BUSY;
 
        adap = tx_info->adap;
+       stats = &adap->chcr_stats;
+
        qidx = skb->queue_mapping;
        q = &adap->sge.ethtxq[qidx + tx_info->first_qset];
        cxgb4_reclaim_completed_tx(adap, &q->q, true);
@@ -1829,6 +1912,7 @@ int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
         * part of the record is received. Incase of partial end part of record,
         * we will send the complete record again.
         */
+
        do {
                int i;
 
@@ -1843,11 +1927,13 @@ int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
                 */
                if (unlikely(!record)) {
                        spin_unlock_irqrestore(&tx_ctx->base.lock, flags);
+                       atomic64_inc(&stats->ktls_tx_drop_no_sync_data);
                        goto out;
                }
 
                if (unlikely(tls_record_is_start_marker(record))) {
                        spin_unlock_irqrestore(&tx_ctx->base.lock, flags);
+                       atomic64_inc(&stats->ktls_tx_skip_no_sync_data);
                        goto out;
                }
 
@@ -1918,6 +2004,10 @@ clear_ref:
        } while (data_len > 0);
 
        tx_info->prev_seq = ntohl(th->seq) + skb->data_len;
+
+       atomic64_inc(&stats->ktls_tx_encrypted_packets);
+       atomic64_add(skb->data_len, &stats->ktls_tx_encrypted_bytes);
+
        /* tcp finish is set, send a separate tcp msg including all the options
         * as well.
         */
index 9ffb8cc..5a7ae2c 100644 (file)
@@ -11,6 +11,7 @@
 #include "t4_tcb.h"
 #include "l2t.h"
 #include "chcr_common.h"
+#include "cxgb4_uld.h"
 
 #define CHCR_TCB_STATE_CLOSED  0
 #define CHCR_KTLS_KEY_CTX_LEN  16
index fe883cb..ebed99f 100644 (file)
@@ -3409,6 +3409,41 @@ static int chcr_stats_show(struct seq_file *seq, void *v)
                   atomic_read(&adap->chcr_stats.tls_pdu_rx));
        seq_printf(seq, "TLS Keys (DDR) Count: %10u\n",
                   atomic_read(&adap->chcr_stats.tls_key));
+#ifdef CONFIG_CHELSIO_TLS_DEVICE
+       seq_puts(seq, "\nChelsio KTLS Crypto Accelerator Stats\n");
+       seq_printf(seq, "Tx HW offload contexts added:     %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_ctx));
+       seq_printf(seq, "Tx connection created:            %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_connection_open));
+       seq_printf(seq, "Tx connection failed:             %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_connection_fail));
+       seq_printf(seq, "Tx connection closed:             %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_connection_close));
+       seq_printf(seq, "Packets passed for encryption :   %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_encrypted_packets));
+       seq_printf(seq, "Bytes passed for encryption :     %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_encrypted_bytes));
+       seq_printf(seq, "Tx records send:                  %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_send_records));
+       seq_printf(seq, "Tx partial start of records:      %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_start_pkts));
+       seq_printf(seq, "Tx partial middle of records:     %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_middle_pkts));
+       seq_printf(seq, "Tx partial end of record:         %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_end_pkts));
+       seq_printf(seq, "Tx complete records:              %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_complete_pkts));
+       seq_printf(seq, "TX trim pkts :                    %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_trimmed_pkts));
+       seq_printf(seq, "Tx out of order packets:          %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_ooo));
+       seq_printf(seq, "Tx drop pkts before HW offload:   %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_skip_no_sync_data));
+       seq_printf(seq, "Tx drop not synced packets:       %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_drop_no_sync_data));
+       seq_printf(seq, "Tx drop bypass req:               %20llu\n",
+                  atomic64_read(&adap->chcr_stats.ktls_tx_drop_bypass_req));
+#endif
 
        return 0;
 }
index d9d27bc..03b9bdc 100644 (file)
@@ -357,6 +357,26 @@ struct chcr_stats_debug {
        atomic_t tls_pdu_tx;
        atomic_t tls_pdu_rx;
        atomic_t tls_key;
+#ifdef CONFIG_CHELSIO_TLS_DEVICE
+       atomic64_t ktls_tx_connection_open;
+       atomic64_t ktls_tx_connection_fail;
+       atomic64_t ktls_tx_connection_close;
+       atomic64_t ktls_tx_send_records;
+       atomic64_t ktls_tx_end_pkts;
+       atomic64_t ktls_tx_start_pkts;
+       atomic64_t ktls_tx_middle_pkts;
+       atomic64_t ktls_tx_retransmit_pkts;
+       atomic64_t ktls_tx_complete_pkts;
+       atomic64_t ktls_tx_trimmed_pkts;
+       atomic64_t ktls_tx_encrypted_packets;
+       atomic64_t ktls_tx_encrypted_bytes;
+       atomic64_t ktls_tx_ctx;
+       atomic64_t ktls_tx_ooo;
+       atomic64_t ktls_tx_skip_no_sync_data;
+       atomic64_t ktls_tx_drop_no_sync_data;
+       atomic64_t ktls_tx_drop_bypass_req;
+
+#endif
 };
 
 #define OCQ_WIN_OFFSET(pdev, vres) \