2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
52 static char *states[] = {
69 module_param(nocong, int, 0644);
70 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
72 static int enable_ecn;
73 module_param(enable_ecn, int, 0644);
74 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
76 static int dack_mode = 1;
77 module_param(dack_mode, int, 0644);
78 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
80 int c4iw_max_read_depth = 8;
81 module_param(c4iw_max_read_depth, int, 0644);
82 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
84 static int enable_tcp_timestamps;
85 module_param(enable_tcp_timestamps, int, 0644);
86 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
88 static int enable_tcp_sack;
89 module_param(enable_tcp_sack, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
92 static int enable_tcp_window_scaling = 1;
93 module_param(enable_tcp_window_scaling, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_window_scaling,
95 "Enable tcp window scaling (default=1)");
98 module_param(c4iw_debug, int, 0644);
99 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
101 static int peer2peer = 1;
102 module_param(peer2peer, int, 0644);
103 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
105 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
106 module_param(p2p_type, int, 0644);
107 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
108 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
110 static int ep_timeout_secs = 60;
111 module_param(ep_timeout_secs, int, 0644);
112 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
113 "in seconds (default=60)");
115 static int mpa_rev = 1;
116 module_param(mpa_rev, int, 0644);
117 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
118 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
119 " compliant (default=1)");
121 static int markers_enabled;
122 module_param(markers_enabled, int, 0644);
123 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
125 static int crc_enabled = 1;
126 module_param(crc_enabled, int, 0644);
127 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
129 static int rcv_win = 256 * 1024;
130 module_param(rcv_win, int, 0644);
131 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
133 static int snd_win = 128 * 1024;
134 module_param(snd_win, int, 0644);
135 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
137 static struct workqueue_struct *workq;
139 static struct sk_buff_head rxq;
141 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
142 static void ep_timeout(unsigned long arg);
143 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static LIST_HEAD(timeout_list);
146 static spinlock_t timeout_lock;
148 static void deref_qp(struct c4iw_ep *ep)
150 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
151 clear_bit(QP_REFERENCED, &ep->com.flags);
154 static void ref_qp(struct c4iw_ep *ep)
156 set_bit(QP_REFERENCED, &ep->com.flags);
157 c4iw_qp_add_ref(&ep->com.qp->ibqp);
160 static void start_ep_timer(struct c4iw_ep *ep)
162 PDBG("%s ep %p\n", __func__, ep);
163 if (timer_pending(&ep->timer)) {
164 pr_err("%s timer already started! ep %p\n",
168 clear_bit(TIMEOUT, &ep->com.flags);
169 c4iw_get_ep(&ep->com);
170 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
171 ep->timer.data = (unsigned long)ep;
172 ep->timer.function = ep_timeout;
173 add_timer(&ep->timer);
176 static void stop_ep_timer(struct c4iw_ep *ep)
178 PDBG("%s ep %p stopping\n", __func__, ep);
179 del_timer_sync(&ep->timer);
180 if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
181 c4iw_put_ep(&ep->com);
184 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
185 struct l2t_entry *l2e)
189 if (c4iw_fatal_error(rdev)) {
191 PDBG("%s - device in error state - dropping\n", __func__);
194 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
197 return error < 0 ? error : 0;
200 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
204 if (c4iw_fatal_error(rdev)) {
206 PDBG("%s - device in error state - dropping\n", __func__);
209 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
212 return error < 0 ? error : 0;
215 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
217 struct cpl_tid_release *req;
219 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
222 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
223 INIT_TP_WR(req, hwtid);
224 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
225 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
226 c4iw_ofld_send(rdev, skb);
230 static void set_emss(struct c4iw_ep *ep, u16 opt)
232 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
234 if (GET_TCPOPT_TSTAMP(opt))
238 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
242 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
244 enum c4iw_ep_state state;
246 mutex_lock(&epc->mutex);
248 mutex_unlock(&epc->mutex);
252 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
257 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
259 mutex_lock(&epc->mutex);
260 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
261 __state_set(epc, new);
262 mutex_unlock(&epc->mutex);
266 static void *alloc_ep(int size, gfp_t gfp)
268 struct c4iw_ep_common *epc;
270 epc = kzalloc(size, gfp);
272 kref_init(&epc->kref);
273 mutex_init(&epc->mutex);
274 c4iw_init_wr_wait(&epc->wr_wait);
276 PDBG("%s alloc ep %p\n", __func__, epc);
280 void _c4iw_free_ep(struct kref *kref)
284 ep = container_of(kref, struct c4iw_ep, com.kref);
285 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
286 if (test_bit(QP_REFERENCED, &ep->com.flags))
288 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
289 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
290 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
291 dst_release(ep->dst);
292 cxgb4_l2t_release(ep->l2t);
297 static void release_ep_resources(struct c4iw_ep *ep)
299 set_bit(RELEASE_RESOURCES, &ep->com.flags);
300 c4iw_put_ep(&ep->com);
303 static int status2errno(int status)
308 case CPL_ERR_CONN_RESET:
310 case CPL_ERR_ARP_MISS:
311 return -EHOSTUNREACH;
312 case CPL_ERR_CONN_TIMEDOUT:
314 case CPL_ERR_TCAM_FULL:
316 case CPL_ERR_CONN_EXIST:
324 * Try and reuse skbs already allocated...
326 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
328 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
331 skb_reset_transport_header(skb);
333 skb = alloc_skb(len, gfp);
335 t4_set_arp_err_handler(skb, NULL, NULL);
339 static struct net_device *get_real_dev(struct net_device *egress_dev)
341 struct net_device *phys_dev = egress_dev;
342 if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
343 phys_dev = vlan_dev_real_dev(egress_dev);
347 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
351 egress_dev = get_real_dev(egress_dev);
352 for (i = 0; i < dev->rdev.lldi.nports; i++)
353 if (dev->rdev.lldi.ports[i] == egress_dev)
358 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
359 __u8 *peer_ip, __be16 local_port,
360 __be16 peer_port, u8 tos,
363 struct dst_entry *dst = NULL;
365 if (IS_ENABLED(CONFIG_IPV6)) {
368 memset(&fl6, 0, sizeof(fl6));
369 memcpy(&fl6.daddr, peer_ip, 16);
370 memcpy(&fl6.saddr, local_ip, 16);
371 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
372 fl6.flowi6_oif = sin6_scope_id;
373 dst = ip6_route_output(&init_net, NULL, &fl6);
376 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
377 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
387 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
388 __be32 peer_ip, __be16 local_port,
389 __be16 peer_port, u8 tos)
395 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
396 peer_port, local_port, IPPROTO_TCP,
400 n = dst_neigh_lookup(&rt->dst, &peer_ip);
403 if (!our_interface(dev, n->dev) &&
404 !(n->dev->flags & IFF_LOOPBACK)) {
405 dst_release(&rt->dst);
412 static void arp_failure_discard(void *handle, struct sk_buff *skb)
414 PDBG("%s c4iw_dev %p\n", __func__, handle);
419 * Handle an ARP failure for an active open.
421 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
423 printk(KERN_ERR MOD "ARP failure duing connect\n");
428 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
431 static void abort_arp_failure(void *handle, struct sk_buff *skb)
433 struct c4iw_rdev *rdev = handle;
434 struct cpl_abort_req *req = cplhdr(skb);
436 PDBG("%s rdev %p\n", __func__, rdev);
437 req->cmd = CPL_ABORT_NO_RST;
438 c4iw_ofld_send(rdev, skb);
441 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
443 unsigned int flowclen = 80;
444 struct fw_flowc_wr *flowc;
447 skb = get_skb(skb, flowclen, GFP_KERNEL);
448 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
450 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
451 FW_FLOWC_WR_NPARAMS(8));
452 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
453 16)) | FW_WR_FLOWID(ep->hwtid));
455 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
456 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
457 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
458 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
459 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
460 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
461 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
462 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
463 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
464 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
465 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
466 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
467 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
468 flowc->mnemval[6].val = cpu_to_be32(snd_win);
469 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
470 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
471 /* Pad WR to 16 byte boundary */
472 flowc->mnemval[8].mnemonic = 0;
473 flowc->mnemval[8].val = 0;
474 for (i = 0; i < 9; i++) {
475 flowc->mnemval[i].r4[0] = 0;
476 flowc->mnemval[i].r4[1] = 0;
477 flowc->mnemval[i].r4[2] = 0;
480 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
481 c4iw_ofld_send(&ep->com.dev->rdev, skb);
484 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
486 struct cpl_close_con_req *req;
488 int wrlen = roundup(sizeof *req, 16);
490 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
491 skb = get_skb(NULL, wrlen, gfp);
493 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
496 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
497 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
498 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
499 memset(req, 0, wrlen);
500 INIT_TP_WR(req, ep->hwtid);
501 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
503 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
506 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
508 struct cpl_abort_req *req;
509 int wrlen = roundup(sizeof *req, 16);
511 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
512 skb = get_skb(skb, wrlen, gfp);
514 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
518 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
519 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
520 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
521 memset(req, 0, wrlen);
522 INIT_TP_WR(req, ep->hwtid);
523 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
524 req->cmd = CPL_ABORT_SEND_RST;
525 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
528 static int send_connect(struct c4iw_ep *ep)
530 struct cpl_act_open_req *req;
531 struct cpl_t5_act_open_req *t5_req;
532 struct cpl_act_open_req6 *req6;
533 struct cpl_t5_act_open_req6 *t5_req6;
537 unsigned int mtu_idx;
540 int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
541 sizeof(struct cpl_act_open_req) :
542 sizeof(struct cpl_t5_act_open_req);
543 int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
544 sizeof(struct cpl_act_open_req6) :
545 sizeof(struct cpl_t5_act_open_req6);
546 struct sockaddr_in *la = (struct sockaddr_in *)&ep->com.local_addr;
547 struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
548 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
549 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
551 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
552 roundup(sizev4, 16) :
555 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
557 skb = get_skb(NULL, wrlen, GFP_KERNEL);
559 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
563 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
565 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
566 wscale = compute_wscale(rcv_win);
567 opt0 = (nocong ? NO_CONG(1) : 0) |
572 L2T_IDX(ep->l2t->idx) |
573 TX_CHAN(ep->tx_chan) |
574 SMAC_SEL(ep->smac_idx) |
576 ULP_MODE(ULP_MODE_TCPDDP) |
577 RCV_BUFSIZ(rcv_win>>10);
578 opt2 = RX_CHANNEL(0) |
579 CCTRL_ECN(enable_ecn) |
580 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
581 if (enable_tcp_timestamps)
582 opt2 |= TSTAMPS_EN(1);
585 if (wscale && enable_tcp_window_scaling)
586 opt2 |= WND_SCALE_EN(1);
587 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
589 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
590 if (ep->com.remote_addr.ss_family == AF_INET) {
591 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
593 OPCODE_TID(req) = cpu_to_be32(
594 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
595 ((ep->rss_qid << 14) | ep->atid)));
596 req->local_port = la->sin_port;
597 req->peer_port = ra->sin_port;
598 req->local_ip = la->sin_addr.s_addr;
599 req->peer_ip = ra->sin_addr.s_addr;
600 req->opt0 = cpu_to_be64(opt0);
601 req->params = cpu_to_be32(cxgb4_select_ntuple(
602 ep->com.dev->rdev.lldi.ports[0],
604 req->opt2 = cpu_to_be32(opt2);
606 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
609 OPCODE_TID(req6) = cpu_to_be32(
610 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
611 ((ep->rss_qid<<14)|ep->atid)));
612 req6->local_port = la6->sin6_port;
613 req6->peer_port = ra6->sin6_port;
614 req6->local_ip_hi = *((__be64 *)
615 (la6->sin6_addr.s6_addr));
616 req6->local_ip_lo = *((__be64 *)
617 (la6->sin6_addr.s6_addr + 8));
618 req6->peer_ip_hi = *((__be64 *)
619 (ra6->sin6_addr.s6_addr));
620 req6->peer_ip_lo = *((__be64 *)
621 (ra6->sin6_addr.s6_addr + 8));
622 req6->opt0 = cpu_to_be64(opt0);
623 req6->params = cpu_to_be32(cxgb4_select_ntuple(
624 ep->com.dev->rdev.lldi.ports[0],
626 req6->opt2 = cpu_to_be32(opt2);
629 if (ep->com.remote_addr.ss_family == AF_INET) {
630 t5_req = (struct cpl_t5_act_open_req *)
632 INIT_TP_WR(t5_req, 0);
633 OPCODE_TID(t5_req) = cpu_to_be32(
634 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
635 ((ep->rss_qid << 14) | ep->atid)));
636 t5_req->local_port = la->sin_port;
637 t5_req->peer_port = ra->sin_port;
638 t5_req->local_ip = la->sin_addr.s_addr;
639 t5_req->peer_ip = ra->sin_addr.s_addr;
640 t5_req->opt0 = cpu_to_be64(opt0);
641 t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
643 ep->com.dev->rdev.lldi.ports[0],
645 t5_req->opt2 = cpu_to_be32(opt2);
647 t5_req6 = (struct cpl_t5_act_open_req6 *)
649 INIT_TP_WR(t5_req6, 0);
650 OPCODE_TID(t5_req6) = cpu_to_be32(
651 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
652 ((ep->rss_qid<<14)|ep->atid)));
653 t5_req6->local_port = la6->sin6_port;
654 t5_req6->peer_port = ra6->sin6_port;
655 t5_req6->local_ip_hi = *((__be64 *)
656 (la6->sin6_addr.s6_addr));
657 t5_req6->local_ip_lo = *((__be64 *)
658 (la6->sin6_addr.s6_addr + 8));
659 t5_req6->peer_ip_hi = *((__be64 *)
660 (ra6->sin6_addr.s6_addr));
661 t5_req6->peer_ip_lo = *((__be64 *)
662 (ra6->sin6_addr.s6_addr + 8));
663 t5_req6->opt0 = cpu_to_be64(opt0);
664 t5_req6->params = (__force __be64)cpu_to_be32(
666 ep->com.dev->rdev.lldi.ports[0],
668 t5_req6->opt2 = cpu_to_be32(opt2);
672 set_bit(ACT_OPEN_REQ, &ep->com.history);
673 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
676 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
680 struct fw_ofld_tx_data_wr *req;
681 struct mpa_message *mpa;
682 struct mpa_v2_conn_params mpa_v2_params;
684 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
686 BUG_ON(skb_cloned(skb));
688 mpalen = sizeof(*mpa) + ep->plen;
689 if (mpa_rev_to_use == 2)
690 mpalen += sizeof(struct mpa_v2_conn_params);
691 wrlen = roundup(mpalen + sizeof *req, 16);
692 skb = get_skb(skb, wrlen, GFP_KERNEL);
694 connect_reply_upcall(ep, -ENOMEM);
697 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
699 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
700 memset(req, 0, wrlen);
701 req->op_to_immdlen = cpu_to_be32(
702 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
704 FW_WR_IMMDLEN(mpalen));
705 req->flowid_len16 = cpu_to_be32(
706 FW_WR_FLOWID(ep->hwtid) |
707 FW_WR_LEN16(wrlen >> 4));
708 req->plen = cpu_to_be32(mpalen);
709 req->tunnel_to_proxy = cpu_to_be32(
710 FW_OFLD_TX_DATA_WR_FLUSH(1) |
711 FW_OFLD_TX_DATA_WR_SHOVE(1));
713 mpa = (struct mpa_message *)(req + 1);
714 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
715 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
716 (markers_enabled ? MPA_MARKERS : 0) |
717 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
718 mpa->private_data_size = htons(ep->plen);
719 mpa->revision = mpa_rev_to_use;
720 if (mpa_rev_to_use == 1) {
721 ep->tried_with_mpa_v1 = 1;
722 ep->retry_with_mpa_v1 = 0;
725 if (mpa_rev_to_use == 2) {
726 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
727 sizeof (struct mpa_v2_conn_params));
728 mpa_v2_params.ird = htons((u16)ep->ird);
729 mpa_v2_params.ord = htons((u16)ep->ord);
732 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
733 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
735 htons(MPA_V2_RDMA_WRITE_RTR);
736 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
738 htons(MPA_V2_RDMA_READ_RTR);
740 memcpy(mpa->private_data, &mpa_v2_params,
741 sizeof(struct mpa_v2_conn_params));
744 memcpy(mpa->private_data +
745 sizeof(struct mpa_v2_conn_params),
746 ep->mpa_pkt + sizeof(*mpa), ep->plen);
749 memcpy(mpa->private_data,
750 ep->mpa_pkt + sizeof(*mpa), ep->plen);
753 * Reference the mpa skb. This ensures the data area
754 * will remain in memory until the hw acks the tx.
755 * Function fw4_ack() will deref it.
758 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
761 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
763 __state_set(&ep->com, MPA_REQ_SENT);
764 ep->mpa_attr.initiator = 1;
765 ep->snd_seq += mpalen;
769 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
772 struct fw_ofld_tx_data_wr *req;
773 struct mpa_message *mpa;
775 struct mpa_v2_conn_params mpa_v2_params;
777 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
779 mpalen = sizeof(*mpa) + plen;
780 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
781 mpalen += sizeof(struct mpa_v2_conn_params);
782 wrlen = roundup(mpalen + sizeof *req, 16);
784 skb = get_skb(NULL, wrlen, GFP_KERNEL);
786 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
789 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
791 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
792 memset(req, 0, wrlen);
793 req->op_to_immdlen = cpu_to_be32(
794 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
796 FW_WR_IMMDLEN(mpalen));
797 req->flowid_len16 = cpu_to_be32(
798 FW_WR_FLOWID(ep->hwtid) |
799 FW_WR_LEN16(wrlen >> 4));
800 req->plen = cpu_to_be32(mpalen);
801 req->tunnel_to_proxy = cpu_to_be32(
802 FW_OFLD_TX_DATA_WR_FLUSH(1) |
803 FW_OFLD_TX_DATA_WR_SHOVE(1));
805 mpa = (struct mpa_message *)(req + 1);
806 memset(mpa, 0, sizeof(*mpa));
807 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
808 mpa->flags = MPA_REJECT;
809 mpa->revision = ep->mpa_attr.version;
810 mpa->private_data_size = htons(plen);
812 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
813 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
814 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
815 sizeof (struct mpa_v2_conn_params));
816 mpa_v2_params.ird = htons(((u16)ep->ird) |
817 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
819 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
821 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
822 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
823 FW_RI_INIT_P2PTYPE_READ_REQ ?
824 MPA_V2_RDMA_READ_RTR : 0) : 0));
825 memcpy(mpa->private_data, &mpa_v2_params,
826 sizeof(struct mpa_v2_conn_params));
829 memcpy(mpa->private_data +
830 sizeof(struct mpa_v2_conn_params), pdata, plen);
833 memcpy(mpa->private_data, pdata, plen);
836 * Reference the mpa skb again. This ensures the data area
837 * will remain in memory until the hw acks the tx.
838 * Function fw4_ack() will deref it.
841 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
842 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
845 ep->snd_seq += mpalen;
846 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
849 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
852 struct fw_ofld_tx_data_wr *req;
853 struct mpa_message *mpa;
855 struct mpa_v2_conn_params mpa_v2_params;
857 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
859 mpalen = sizeof(*mpa) + plen;
860 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
861 mpalen += sizeof(struct mpa_v2_conn_params);
862 wrlen = roundup(mpalen + sizeof *req, 16);
864 skb = get_skb(NULL, wrlen, GFP_KERNEL);
866 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
869 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
871 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
872 memset(req, 0, wrlen);
873 req->op_to_immdlen = cpu_to_be32(
874 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
876 FW_WR_IMMDLEN(mpalen));
877 req->flowid_len16 = cpu_to_be32(
878 FW_WR_FLOWID(ep->hwtid) |
879 FW_WR_LEN16(wrlen >> 4));
880 req->plen = cpu_to_be32(mpalen);
881 req->tunnel_to_proxy = cpu_to_be32(
882 FW_OFLD_TX_DATA_WR_FLUSH(1) |
883 FW_OFLD_TX_DATA_WR_SHOVE(1));
885 mpa = (struct mpa_message *)(req + 1);
886 memset(mpa, 0, sizeof(*mpa));
887 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
888 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
889 (markers_enabled ? MPA_MARKERS : 0);
890 mpa->revision = ep->mpa_attr.version;
891 mpa->private_data_size = htons(plen);
893 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
894 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
895 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
896 sizeof (struct mpa_v2_conn_params));
897 mpa_v2_params.ird = htons((u16)ep->ird);
898 mpa_v2_params.ord = htons((u16)ep->ord);
899 if (peer2peer && (ep->mpa_attr.p2p_type !=
900 FW_RI_INIT_P2PTYPE_DISABLED)) {
901 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
903 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
905 htons(MPA_V2_RDMA_WRITE_RTR);
906 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
908 htons(MPA_V2_RDMA_READ_RTR);
911 memcpy(mpa->private_data, &mpa_v2_params,
912 sizeof(struct mpa_v2_conn_params));
915 memcpy(mpa->private_data +
916 sizeof(struct mpa_v2_conn_params), pdata, plen);
919 memcpy(mpa->private_data, pdata, plen);
922 * Reference the mpa skb. This ensures the data area
923 * will remain in memory until the hw acks the tx.
924 * Function fw4_ack() will deref it.
927 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
929 __state_set(&ep->com, MPA_REP_SENT);
930 ep->snd_seq += mpalen;
931 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
934 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
937 struct cpl_act_establish *req = cplhdr(skb);
938 unsigned int tid = GET_TID(req);
939 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
940 struct tid_info *t = dev->rdev.lldi.tids;
942 ep = lookup_atid(t, atid);
944 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
945 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
947 mutex_lock(&ep->com.mutex);
948 dst_confirm(ep->dst);
950 /* setup the hwtid for this connection */
952 cxgb4_insert_tid(t, ep, tid);
953 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
955 ep->snd_seq = be32_to_cpu(req->snd_isn);
956 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
958 set_emss(ep, ntohs(req->tcp_opt));
960 /* dealloc the atid */
961 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
962 cxgb4_free_atid(t, atid);
963 set_bit(ACT_ESTAB, &ep->com.history);
965 /* start MPA negotiation */
966 send_flowc(ep, NULL);
967 if (ep->retry_with_mpa_v1)
968 send_mpa_req(ep, skb, 1);
970 send_mpa_req(ep, skb, mpa_rev);
971 mutex_unlock(&ep->com.mutex);
975 static void close_complete_upcall(struct c4iw_ep *ep, int status)
977 struct iw_cm_event event;
979 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
980 memset(&event, 0, sizeof(event));
981 event.event = IW_CM_EVENT_CLOSE;
982 event.status = status;
984 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
985 ep, ep->com.cm_id, ep->hwtid);
986 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
987 ep->com.cm_id->rem_ref(ep->com.cm_id);
988 ep->com.cm_id = NULL;
989 set_bit(CLOSE_UPCALL, &ep->com.history);
993 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
995 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
996 state_set(&ep->com, ABORTING);
997 set_bit(ABORT_CONN, &ep->com.history);
998 return send_abort(ep, skb, gfp);
1001 static void peer_close_upcall(struct c4iw_ep *ep)
1003 struct iw_cm_event event;
1005 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1006 memset(&event, 0, sizeof(event));
1007 event.event = IW_CM_EVENT_DISCONNECT;
1008 if (ep->com.cm_id) {
1009 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1010 ep, ep->com.cm_id, ep->hwtid);
1011 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1012 set_bit(DISCONN_UPCALL, &ep->com.history);
1016 static void peer_abort_upcall(struct c4iw_ep *ep)
1018 struct iw_cm_event event;
1020 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1021 memset(&event, 0, sizeof(event));
1022 event.event = IW_CM_EVENT_CLOSE;
1023 event.status = -ECONNRESET;
1024 if (ep->com.cm_id) {
1025 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1026 ep->com.cm_id, ep->hwtid);
1027 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1028 ep->com.cm_id->rem_ref(ep->com.cm_id);
1029 ep->com.cm_id = NULL;
1030 set_bit(ABORT_UPCALL, &ep->com.history);
1034 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1036 struct iw_cm_event event;
1038 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1039 memset(&event, 0, sizeof(event));
1040 event.event = IW_CM_EVENT_CONNECT_REPLY;
1041 event.status = status;
1042 memcpy(&event.local_addr, &ep->com.local_addr,
1043 sizeof(ep->com.local_addr));
1044 memcpy(&event.remote_addr, &ep->com.remote_addr,
1045 sizeof(ep->com.remote_addr));
1047 if ((status == 0) || (status == -ECONNREFUSED)) {
1048 if (!ep->tried_with_mpa_v1) {
1049 /* this means MPA_v2 is used */
1050 event.private_data_len = ep->plen -
1051 sizeof(struct mpa_v2_conn_params);
1052 event.private_data = ep->mpa_pkt +
1053 sizeof(struct mpa_message) +
1054 sizeof(struct mpa_v2_conn_params);
1056 /* this means MPA_v1 is used */
1057 event.private_data_len = ep->plen;
1058 event.private_data = ep->mpa_pkt +
1059 sizeof(struct mpa_message);
1063 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1065 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1066 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1069 ep->com.cm_id->rem_ref(ep->com.cm_id);
1070 ep->com.cm_id = NULL;
1074 static int connect_request_upcall(struct c4iw_ep *ep)
1076 struct iw_cm_event event;
1079 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1080 memset(&event, 0, sizeof(event));
1081 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1082 memcpy(&event.local_addr, &ep->com.local_addr,
1083 sizeof(ep->com.local_addr));
1084 memcpy(&event.remote_addr, &ep->com.remote_addr,
1085 sizeof(ep->com.remote_addr));
1086 event.provider_data = ep;
1087 if (!ep->tried_with_mpa_v1) {
1088 /* this means MPA_v2 is used */
1089 event.ord = ep->ord;
1090 event.ird = ep->ird;
1091 event.private_data_len = ep->plen -
1092 sizeof(struct mpa_v2_conn_params);
1093 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1094 sizeof(struct mpa_v2_conn_params);
1096 /* this means MPA_v1 is used. Send max supported */
1097 event.ord = c4iw_max_read_depth;
1098 event.ird = c4iw_max_read_depth;
1099 event.private_data_len = ep->plen;
1100 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1102 c4iw_get_ep(&ep->com);
1103 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1106 c4iw_put_ep(&ep->com);
1107 set_bit(CONNREQ_UPCALL, &ep->com.history);
1108 c4iw_put_ep(&ep->parent_ep->com);
1112 static void established_upcall(struct c4iw_ep *ep)
1114 struct iw_cm_event event;
1116 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1117 memset(&event, 0, sizeof(event));
1118 event.event = IW_CM_EVENT_ESTABLISHED;
1119 event.ird = ep->ird;
1120 event.ord = ep->ord;
1121 if (ep->com.cm_id) {
1122 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1123 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1124 set_bit(ESTAB_UPCALL, &ep->com.history);
1128 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1130 struct cpl_rx_data_ack *req;
1131 struct sk_buff *skb;
1132 int wrlen = roundup(sizeof *req, 16);
1134 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1135 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1137 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1141 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1142 memset(req, 0, wrlen);
1143 INIT_TP_WR(req, ep->hwtid);
1144 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1146 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1148 V_RX_DACK_MODE(dack_mode));
1149 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1150 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1154 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1156 struct mpa_message *mpa;
1157 struct mpa_v2_conn_params *mpa_v2_params;
1159 u16 resp_ird, resp_ord;
1160 u8 rtr_mismatch = 0, insuff_ird = 0;
1161 struct c4iw_qp_attributes attrs;
1162 enum c4iw_qp_attr_mask mask;
1165 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1168 * Stop mpa timer. If it expired, then the state has
1169 * changed and we bail since ep_timeout already aborted
1173 if (ep->com.state != MPA_REQ_SENT)
1177 * If we get more than the supported amount of private data
1178 * then we must fail this connection.
1180 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1186 * copy the new data into our accumulation buffer.
1188 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1190 ep->mpa_pkt_len += skb->len;
1193 * if we don't even have the mpa message, then bail.
1195 if (ep->mpa_pkt_len < sizeof(*mpa))
1197 mpa = (struct mpa_message *) ep->mpa_pkt;
1199 /* Validate MPA header. */
1200 if (mpa->revision > mpa_rev) {
1201 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1202 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1206 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1211 plen = ntohs(mpa->private_data_size);
1214 * Fail if there's too much private data.
1216 if (plen > MPA_MAX_PRIVATE_DATA) {
1222 * If plen does not account for pkt size
1224 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1229 ep->plen = (u8) plen;
1232 * If we don't have all the pdata yet, then bail.
1233 * We'll continue process when more data arrives.
1235 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1238 if (mpa->flags & MPA_REJECT) {
1239 err = -ECONNREFUSED;
1244 * If we get here we have accumulated the entire mpa
1245 * start reply message including private data. And
1246 * the MPA header is valid.
1248 __state_set(&ep->com, FPDU_MODE);
1249 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1250 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1251 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1252 ep->mpa_attr.version = mpa->revision;
1253 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1255 if (mpa->revision == 2) {
1256 ep->mpa_attr.enhanced_rdma_conn =
1257 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1258 if (ep->mpa_attr.enhanced_rdma_conn) {
1259 mpa_v2_params = (struct mpa_v2_conn_params *)
1260 (ep->mpa_pkt + sizeof(*mpa));
1261 resp_ird = ntohs(mpa_v2_params->ird) &
1262 MPA_V2_IRD_ORD_MASK;
1263 resp_ord = ntohs(mpa_v2_params->ord) &
1264 MPA_V2_IRD_ORD_MASK;
1267 * This is a double-check. Ideally, below checks are
1268 * not required since ird/ord stuff has been taken
1269 * care of in c4iw_accept_cr
1271 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1278 if (ntohs(mpa_v2_params->ird) &
1279 MPA_V2_PEER2PEER_MODEL) {
1280 if (ntohs(mpa_v2_params->ord) &
1281 MPA_V2_RDMA_WRITE_RTR)
1282 ep->mpa_attr.p2p_type =
1283 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1284 else if (ntohs(mpa_v2_params->ord) &
1285 MPA_V2_RDMA_READ_RTR)
1286 ep->mpa_attr.p2p_type =
1287 FW_RI_INIT_P2PTYPE_READ_REQ;
1290 } else if (mpa->revision == 1)
1292 ep->mpa_attr.p2p_type = p2p_type;
1294 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1295 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1296 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1297 ep->mpa_attr.recv_marker_enabled,
1298 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1299 ep->mpa_attr.p2p_type, p2p_type);
1302 * If responder's RTR does not match with that of initiator, assign
1303 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1304 * generated when moving QP to RTS state.
1305 * A TERM message will be sent after QP has moved to RTS state
1307 if ((ep->mpa_attr.version == 2) && peer2peer &&
1308 (ep->mpa_attr.p2p_type != p2p_type)) {
1309 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1313 attrs.mpa_attr = ep->mpa_attr;
1314 attrs.max_ird = ep->ird;
1315 attrs.max_ord = ep->ord;
1316 attrs.llp_stream_handle = ep;
1317 attrs.next_state = C4IW_QP_STATE_RTS;
1319 mask = C4IW_QP_ATTR_NEXT_STATE |
1320 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1321 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1323 /* bind QP and TID with INIT_WR */
1324 err = c4iw_modify_qp(ep->com.qp->rhp,
1325 ep->com.qp, mask, &attrs, 1);
1330 * If responder's RTR requirement did not match with what initiator
1331 * supports, generate TERM message
1334 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1335 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1336 attrs.ecode = MPA_NOMATCH_RTR;
1337 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1338 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1339 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1345 * Generate TERM if initiator IRD is not sufficient for responder
1346 * provided ORD. Currently, we do the same behaviour even when
1347 * responder provided IRD is also not sufficient as regards to
1351 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1353 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1354 attrs.ecode = MPA_INSUFF_IRD;
1355 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1356 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1357 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1363 __state_set(&ep->com, ABORTING);
1364 send_abort(ep, skb, GFP_KERNEL);
1366 connect_reply_upcall(ep, err);
1370 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1372 struct mpa_message *mpa;
1373 struct mpa_v2_conn_params *mpa_v2_params;
1376 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1378 if (ep->com.state != MPA_REQ_WAIT)
1382 * If we get more than the supported amount of private data
1383 * then we must fail this connection.
1385 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1387 abort_connection(ep, skb, GFP_KERNEL);
1391 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1394 * Copy the new data into our accumulation buffer.
1396 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1398 ep->mpa_pkt_len += skb->len;
1401 * If we don't even have the mpa message, then bail.
1402 * We'll continue process when more data arrives.
1404 if (ep->mpa_pkt_len < sizeof(*mpa))
1407 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1408 mpa = (struct mpa_message *) ep->mpa_pkt;
1411 * Validate MPA Header.
1413 if (mpa->revision > mpa_rev) {
1414 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1415 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1417 abort_connection(ep, skb, GFP_KERNEL);
1421 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1423 abort_connection(ep, skb, GFP_KERNEL);
1427 plen = ntohs(mpa->private_data_size);
1430 * Fail if there's too much private data.
1432 if (plen > MPA_MAX_PRIVATE_DATA) {
1434 abort_connection(ep, skb, GFP_KERNEL);
1439 * If plen does not account for pkt size
1441 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1443 abort_connection(ep, skb, GFP_KERNEL);
1446 ep->plen = (u8) plen;
1449 * If we don't have all the pdata yet, then bail.
1451 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1455 * If we get here we have accumulated the entire mpa
1456 * start reply message including private data.
1458 ep->mpa_attr.initiator = 0;
1459 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1460 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1461 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1462 ep->mpa_attr.version = mpa->revision;
1463 if (mpa->revision == 1)
1464 ep->tried_with_mpa_v1 = 1;
1465 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1467 if (mpa->revision == 2) {
1468 ep->mpa_attr.enhanced_rdma_conn =
1469 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1470 if (ep->mpa_attr.enhanced_rdma_conn) {
1471 mpa_v2_params = (struct mpa_v2_conn_params *)
1472 (ep->mpa_pkt + sizeof(*mpa));
1473 ep->ird = ntohs(mpa_v2_params->ird) &
1474 MPA_V2_IRD_ORD_MASK;
1475 ep->ord = ntohs(mpa_v2_params->ord) &
1476 MPA_V2_IRD_ORD_MASK;
1477 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1479 if (ntohs(mpa_v2_params->ord) &
1480 MPA_V2_RDMA_WRITE_RTR)
1481 ep->mpa_attr.p2p_type =
1482 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1483 else if (ntohs(mpa_v2_params->ord) &
1484 MPA_V2_RDMA_READ_RTR)
1485 ep->mpa_attr.p2p_type =
1486 FW_RI_INIT_P2PTYPE_READ_REQ;
1489 } else if (mpa->revision == 1)
1491 ep->mpa_attr.p2p_type = p2p_type;
1493 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1494 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1495 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1496 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1497 ep->mpa_attr.p2p_type);
1499 __state_set(&ep->com, MPA_REQ_RCVD);
1503 mutex_lock(&ep->parent_ep->com.mutex);
1504 if (ep->parent_ep->com.state != DEAD) {
1505 if (connect_request_upcall(ep))
1506 abort_connection(ep, skb, GFP_KERNEL);
1508 abort_connection(ep, skb, GFP_KERNEL);
1510 mutex_unlock(&ep->parent_ep->com.mutex);
1514 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1517 struct cpl_rx_data *hdr = cplhdr(skb);
1518 unsigned int dlen = ntohs(hdr->len);
1519 unsigned int tid = GET_TID(hdr);
1520 struct tid_info *t = dev->rdev.lldi.tids;
1521 __u8 status = hdr->status;
1523 ep = lookup_tid(t, tid);
1526 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1527 skb_pull(skb, sizeof(*hdr));
1528 skb_trim(skb, dlen);
1529 mutex_lock(&ep->com.mutex);
1531 /* update RX credits */
1532 update_rx_credits(ep, dlen);
1534 switch (ep->com.state) {
1536 ep->rcv_seq += dlen;
1537 process_mpa_reply(ep, skb);
1540 ep->rcv_seq += dlen;
1541 process_mpa_request(ep, skb);
1544 struct c4iw_qp_attributes attrs;
1545 BUG_ON(!ep->com.qp);
1547 pr_err("%s Unexpected streaming data." \
1548 " qpid %u ep %p state %d tid %u status %d\n",
1549 __func__, ep->com.qp->wq.sq.qid, ep,
1550 ep->com.state, ep->hwtid, status);
1551 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1552 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1553 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1559 mutex_unlock(&ep->com.mutex);
1563 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1566 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1568 unsigned int tid = GET_TID(rpl);
1569 struct tid_info *t = dev->rdev.lldi.tids;
1571 ep = lookup_tid(t, tid);
1573 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1576 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1577 mutex_lock(&ep->com.mutex);
1578 switch (ep->com.state) {
1580 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1581 __state_set(&ep->com, DEAD);
1585 printk(KERN_ERR "%s ep %p state %d\n",
1586 __func__, ep, ep->com.state);
1589 mutex_unlock(&ep->com.mutex);
1592 release_ep_resources(ep);
1596 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1598 struct sk_buff *skb;
1599 struct fw_ofld_connection_wr *req;
1600 unsigned int mtu_idx;
1602 struct sockaddr_in *sin;
1604 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1605 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1606 memset(req, 0, sizeof(*req));
1607 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1608 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1609 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1610 ep->com.dev->rdev.lldi.ports[0],
1612 sin = (struct sockaddr_in *)&ep->com.local_addr;
1613 req->le.lport = sin->sin_port;
1614 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1615 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1616 req->le.pport = sin->sin_port;
1617 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1618 req->tcb.t_state_to_astid =
1619 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1620 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1621 req->tcb.cplrxdataack_cplpassacceptrpl =
1622 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1623 req->tcb.tx_max = (__force __be32) jiffies;
1624 req->tcb.rcv_adv = htons(1);
1625 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1626 wscale = compute_wscale(rcv_win);
1627 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1628 (nocong ? NO_CONG(1) : 0) |
1633 L2T_IDX(ep->l2t->idx) |
1634 TX_CHAN(ep->tx_chan) |
1635 SMAC_SEL(ep->smac_idx) |
1637 ULP_MODE(ULP_MODE_TCPDDP) |
1638 RCV_BUFSIZ(rcv_win >> 10));
1639 req->tcb.opt2 = (__force __be32) (PACE(1) |
1640 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1642 CCTRL_ECN(enable_ecn) |
1643 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1644 if (enable_tcp_timestamps)
1645 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1646 if (enable_tcp_sack)
1647 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1648 if (wscale && enable_tcp_window_scaling)
1649 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1650 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1651 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1652 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1653 set_bit(ACT_OFLD_CONN, &ep->com.history);
1654 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1658 * Return whether a failed active open has allocated a TID
1660 static inline int act_open_has_tid(int status)
1662 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1663 status != CPL_ERR_ARP_MISS;
1666 /* Returns whether a CPL status conveys negative advice.
1668 static int is_neg_adv(unsigned int status)
1670 return status == CPL_ERR_RTX_NEG_ADVICE ||
1671 status == CPL_ERR_PERSIST_NEG_ADVICE ||
1672 status == CPL_ERR_KEEPALV_NEG_ADVICE;
1675 #define ACT_OPEN_RETRY_COUNT 2
1677 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1678 struct dst_entry *dst, struct c4iw_dev *cdev,
1681 struct neighbour *n;
1683 struct net_device *pdev;
1685 n = dst_neigh_lookup(dst, peer_ip);
1691 if (n->dev->flags & IFF_LOOPBACK) {
1693 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1694 else if (IS_ENABLED(CONFIG_IPV6))
1695 for_each_netdev(&init_net, pdev) {
1696 if (ipv6_chk_addr(&init_net,
1697 (struct in6_addr *)peer_ip,
1708 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1712 ep->mtu = pdev->mtu;
1713 ep->tx_chan = cxgb4_port_chan(pdev);
1714 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1715 step = cdev->rdev.lldi.ntxq /
1716 cdev->rdev.lldi.nchan;
1717 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1718 step = cdev->rdev.lldi.nrxq /
1719 cdev->rdev.lldi.nchan;
1720 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1721 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1722 cxgb4_port_idx(pdev) * step];
1725 pdev = get_real_dev(n->dev);
1726 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1730 ep->mtu = dst_mtu(dst);
1731 ep->tx_chan = cxgb4_port_chan(n->dev);
1732 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1733 step = cdev->rdev.lldi.ntxq /
1734 cdev->rdev.lldi.nchan;
1735 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1736 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1737 step = cdev->rdev.lldi.nrxq /
1738 cdev->rdev.lldi.nchan;
1739 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1740 cxgb4_port_idx(n->dev) * step];
1743 ep->retry_with_mpa_v1 = 0;
1744 ep->tried_with_mpa_v1 = 0;
1756 static int c4iw_reconnect(struct c4iw_ep *ep)
1759 struct sockaddr_in *laddr = (struct sockaddr_in *)
1760 &ep->com.cm_id->local_addr;
1761 struct sockaddr_in *raddr = (struct sockaddr_in *)
1762 &ep->com.cm_id->remote_addr;
1763 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1764 &ep->com.cm_id->local_addr;
1765 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1766 &ep->com.cm_id->remote_addr;
1770 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1771 init_timer(&ep->timer);
1774 * Allocate an active TID to initiate a TCP connection.
1776 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1777 if (ep->atid == -1) {
1778 pr_err("%s - cannot alloc atid.\n", __func__);
1782 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1785 if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1786 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1787 raddr->sin_addr.s_addr, laddr->sin_port,
1788 raddr->sin_port, 0);
1790 ra = (__u8 *)&raddr->sin_addr;
1792 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1793 raddr6->sin6_addr.s6_addr,
1794 laddr6->sin6_port, raddr6->sin6_port, 0,
1795 raddr6->sin6_scope_id);
1797 ra = (__u8 *)&raddr6->sin6_addr;
1800 pr_err("%s - cannot find route.\n", __func__);
1801 err = -EHOSTUNREACH;
1804 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1806 pr_err("%s - cannot alloc l2e.\n", __func__);
1810 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1811 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1814 state_set(&ep->com, CONNECTING);
1817 /* send connect request to rnic */
1818 err = send_connect(ep);
1822 cxgb4_l2t_release(ep->l2t);
1824 dst_release(ep->dst);
1826 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1827 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1830 * remember to send notification to upper layer.
1831 * We are in here so the upper layer is not aware that this is
1832 * re-connect attempt and so, upper layer is still waiting for
1833 * response of 1st connect request.
1835 connect_reply_upcall(ep, -ECONNRESET);
1836 c4iw_put_ep(&ep->com);
1841 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1844 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1845 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1846 ntohl(rpl->atid_status)));
1847 struct tid_info *t = dev->rdev.lldi.tids;
1848 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1849 struct sockaddr_in *la;
1850 struct sockaddr_in *ra;
1851 struct sockaddr_in6 *la6;
1852 struct sockaddr_in6 *ra6;
1854 ep = lookup_atid(t, atid);
1855 la = (struct sockaddr_in *)&ep->com.local_addr;
1856 ra = (struct sockaddr_in *)&ep->com.remote_addr;
1857 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1858 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1860 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1861 status, status2errno(status));
1863 if (is_neg_adv(status)) {
1864 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1869 set_bit(ACT_OPEN_RPL, &ep->com.history);
1872 * Log interesting failures.
1875 case CPL_ERR_CONN_RESET:
1876 case CPL_ERR_CONN_TIMEDOUT:
1878 case CPL_ERR_TCAM_FULL:
1879 mutex_lock(&dev->rdev.stats.lock);
1880 dev->rdev.stats.tcam_full++;
1881 mutex_unlock(&dev->rdev.stats.lock);
1882 if (ep->com.local_addr.ss_family == AF_INET &&
1883 dev->rdev.lldi.enable_fw_ofld_conn) {
1884 send_fw_act_open_req(ep,
1885 GET_TID_TID(GET_AOPEN_ATID(
1886 ntohl(rpl->atid_status))));
1890 case CPL_ERR_CONN_EXIST:
1891 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1892 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1893 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1895 cxgb4_free_atid(t, atid);
1896 dst_release(ep->dst);
1897 cxgb4_l2t_release(ep->l2t);
1903 if (ep->com.local_addr.ss_family == AF_INET) {
1904 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1905 atid, status, status2errno(status),
1906 &la->sin_addr.s_addr, ntohs(la->sin_port),
1907 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
1909 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1910 atid, status, status2errno(status),
1911 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1912 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1917 connect_reply_upcall(ep, status2errno(status));
1918 state_set(&ep->com, DEAD);
1920 if (status && act_open_has_tid(status))
1921 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1923 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1924 cxgb4_free_atid(t, atid);
1925 dst_release(ep->dst);
1926 cxgb4_l2t_release(ep->l2t);
1927 c4iw_put_ep(&ep->com);
1932 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1934 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1935 struct tid_info *t = dev->rdev.lldi.tids;
1936 unsigned int stid = GET_TID(rpl);
1937 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1940 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1943 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1944 rpl->status, status2errno(rpl->status));
1945 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1951 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1953 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1954 struct tid_info *t = dev->rdev.lldi.tids;
1955 unsigned int stid = GET_TID(rpl);
1956 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1958 PDBG("%s ep %p\n", __func__, ep);
1959 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1963 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1964 struct cpl_pass_accept_req *req)
1966 struct cpl_pass_accept_rpl *rpl;
1967 unsigned int mtu_idx;
1972 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1973 BUG_ON(skb_cloned(skb));
1974 skb_trim(skb, sizeof(*rpl));
1976 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1977 wscale = compute_wscale(rcv_win);
1978 opt0 = (nocong ? NO_CONG(1) : 0) |
1983 L2T_IDX(ep->l2t->idx) |
1984 TX_CHAN(ep->tx_chan) |
1985 SMAC_SEL(ep->smac_idx) |
1986 DSCP(ep->tos >> 2) |
1987 ULP_MODE(ULP_MODE_TCPDDP) |
1988 RCV_BUFSIZ(rcv_win>>10);
1989 opt2 = RX_CHANNEL(0) |
1990 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1992 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1993 opt2 |= TSTAMPS_EN(1);
1994 if (enable_tcp_sack && req->tcpopt.sack)
1996 if (wscale && enable_tcp_window_scaling)
1997 opt2 |= WND_SCALE_EN(1);
1999 const struct tcphdr *tcph;
2000 u32 hlen = ntohl(req->hdr_len);
2002 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2004 if (tcph->ece && tcph->cwr)
2005 opt2 |= CCTRL_ECN(1);
2009 INIT_TP_WR(rpl, ep->hwtid);
2010 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2012 rpl->opt0 = cpu_to_be64(opt0);
2013 rpl->opt2 = cpu_to_be32(opt2);
2014 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2015 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2016 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2021 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2023 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2024 BUG_ON(skb_cloned(skb));
2025 skb_trim(skb, sizeof(struct cpl_tid_release));
2027 release_tid(&dev->rdev, hwtid, skb);
2031 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2032 __u8 *local_ip, __u8 *peer_ip,
2033 __be16 *local_port, __be16 *peer_port)
2035 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2036 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2037 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2038 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2039 struct tcphdr *tcp = (struct tcphdr *)
2040 ((u8 *)(req + 1) + eth_len + ip_len);
2042 if (ip->version == 4) {
2043 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2044 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2047 memcpy(peer_ip, &ip->saddr, 4);
2048 memcpy(local_ip, &ip->daddr, 4);
2050 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2051 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2054 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2055 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2057 *peer_port = tcp->source;
2058 *local_port = tcp->dest;
2063 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2065 struct c4iw_ep *child_ep = NULL, *parent_ep;
2066 struct cpl_pass_accept_req *req = cplhdr(skb);
2067 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2068 struct tid_info *t = dev->rdev.lldi.tids;
2069 unsigned int hwtid = GET_TID(req);
2070 struct dst_entry *dst;
2071 __u8 local_ip[16], peer_ip[16];
2072 __be16 local_port, peer_port;
2074 u16 peer_mss = ntohs(req->tcpopt.mss);
2077 parent_ep = lookup_stid(t, stid);
2079 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2083 if (state_read(&parent_ep->com) != LISTEN) {
2084 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2089 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2091 /* Find output route */
2093 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2094 , __func__, parent_ep, hwtid,
2095 local_ip, peer_ip, ntohs(local_port),
2096 ntohs(peer_port), peer_mss);
2097 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2098 local_port, peer_port,
2099 GET_POPEN_TOS(ntohl(req->tos_stid)));
2101 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2102 , __func__, parent_ep, hwtid,
2103 local_ip, peer_ip, ntohs(local_port),
2104 ntohs(peer_port), peer_mss);
2105 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2106 PASS_OPEN_TOS(ntohl(req->tos_stid)),
2107 ((struct sockaddr_in6 *)
2108 &parent_ep->com.local_addr)->sin6_scope_id);
2111 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2116 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2118 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2124 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2126 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2133 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2134 child_ep->mtu = peer_mss + 40;
2136 state_set(&child_ep->com, CONNECTING);
2137 child_ep->com.dev = dev;
2138 child_ep->com.cm_id = NULL;
2140 struct sockaddr_in *sin = (struct sockaddr_in *)
2141 &child_ep->com.local_addr;
2142 sin->sin_family = PF_INET;
2143 sin->sin_port = local_port;
2144 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2145 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2146 sin->sin_family = PF_INET;
2147 sin->sin_port = peer_port;
2148 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2150 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2151 &child_ep->com.local_addr;
2152 sin6->sin6_family = PF_INET6;
2153 sin6->sin6_port = local_port;
2154 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2155 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2156 sin6->sin6_family = PF_INET6;
2157 sin6->sin6_port = peer_port;
2158 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2160 c4iw_get_ep(&parent_ep->com);
2161 child_ep->parent_ep = parent_ep;
2162 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2163 child_ep->dst = dst;
2164 child_ep->hwtid = hwtid;
2166 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2167 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2169 init_timer(&child_ep->timer);
2170 cxgb4_insert_tid(t, child_ep, hwtid);
2171 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2172 accept_cr(child_ep, skb, req);
2173 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2176 reject_cr(dev, hwtid, skb);
2181 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2184 struct cpl_pass_establish *req = cplhdr(skb);
2185 struct tid_info *t = dev->rdev.lldi.tids;
2186 unsigned int tid = GET_TID(req);
2188 ep = lookup_tid(t, tid);
2189 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2190 ep->snd_seq = be32_to_cpu(req->snd_isn);
2191 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2193 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2194 ntohs(req->tcp_opt));
2196 set_emss(ep, ntohs(req->tcp_opt));
2198 dst_confirm(ep->dst);
2199 state_set(&ep->com, MPA_REQ_WAIT);
2201 send_flowc(ep, skb);
2202 set_bit(PASS_ESTAB, &ep->com.history);
2207 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2209 struct cpl_peer_close *hdr = cplhdr(skb);
2211 struct c4iw_qp_attributes attrs;
2214 struct tid_info *t = dev->rdev.lldi.tids;
2215 unsigned int tid = GET_TID(hdr);
2218 ep = lookup_tid(t, tid);
2219 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2220 dst_confirm(ep->dst);
2222 set_bit(PEER_CLOSE, &ep->com.history);
2223 mutex_lock(&ep->com.mutex);
2224 switch (ep->com.state) {
2226 __state_set(&ep->com, CLOSING);
2229 __state_set(&ep->com, CLOSING);
2230 connect_reply_upcall(ep, -ECONNRESET);
2235 * We're gonna mark this puppy DEAD, but keep
2236 * the reference on it until the ULP accepts or
2237 * rejects the CR. Also wake up anyone waiting
2238 * in rdma connection migration (see c4iw_accept_cr()).
2240 __state_set(&ep->com, CLOSING);
2241 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2242 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2245 __state_set(&ep->com, CLOSING);
2246 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2247 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2251 __state_set(&ep->com, CLOSING);
2252 attrs.next_state = C4IW_QP_STATE_CLOSING;
2253 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2254 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2255 if (ret != -ECONNRESET) {
2256 peer_close_upcall(ep);
2264 __state_set(&ep->com, MORIBUND);
2269 if (ep->com.cm_id && ep->com.qp) {
2270 attrs.next_state = C4IW_QP_STATE_IDLE;
2271 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2272 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2274 close_complete_upcall(ep, 0);
2275 __state_set(&ep->com, DEAD);
2285 mutex_unlock(&ep->com.mutex);
2287 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2289 release_ep_resources(ep);
2293 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2295 struct cpl_abort_req_rss *req = cplhdr(skb);
2297 struct cpl_abort_rpl *rpl;
2298 struct sk_buff *rpl_skb;
2299 struct c4iw_qp_attributes attrs;
2302 struct tid_info *t = dev->rdev.lldi.tids;
2303 unsigned int tid = GET_TID(req);
2305 ep = lookup_tid(t, tid);
2306 if (is_neg_adv(req->status)) {
2307 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2311 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2313 set_bit(PEER_ABORT, &ep->com.history);
2316 * Wake up any threads in rdma_init() or rdma_fini().
2317 * However, this is not needed if com state is just
2320 if (ep->com.state != MPA_REQ_SENT)
2321 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2323 mutex_lock(&ep->com.mutex);
2324 switch (ep->com.state) {
2332 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2333 connect_reply_upcall(ep, -ECONNRESET);
2336 * we just don't send notification upwards because we
2337 * want to retry with mpa_v1 without upper layers even
2340 * do some housekeeping so as to re-initiate the
2343 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2345 ep->retry_with_mpa_v1 = 1;
2357 if (ep->com.cm_id && ep->com.qp) {
2358 attrs.next_state = C4IW_QP_STATE_ERROR;
2359 ret = c4iw_modify_qp(ep->com.qp->rhp,
2360 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2364 "%s - qp <- error failed!\n",
2367 peer_abort_upcall(ep);
2372 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2373 mutex_unlock(&ep->com.mutex);
2379 dst_confirm(ep->dst);
2380 if (ep->com.state != ABORTING) {
2381 __state_set(&ep->com, DEAD);
2382 /* we don't release if we want to retry with mpa_v1 */
2383 if (!ep->retry_with_mpa_v1)
2386 mutex_unlock(&ep->com.mutex);
2388 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2390 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2395 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2396 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2397 INIT_TP_WR(rpl, ep->hwtid);
2398 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2399 rpl->cmd = CPL_ABORT_NO_RST;
2400 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2403 release_ep_resources(ep);
2404 else if (ep->retry_with_mpa_v1) {
2405 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2406 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2407 dst_release(ep->dst);
2408 cxgb4_l2t_release(ep->l2t);
2415 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2418 struct c4iw_qp_attributes attrs;
2419 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2421 struct tid_info *t = dev->rdev.lldi.tids;
2422 unsigned int tid = GET_TID(rpl);
2424 ep = lookup_tid(t, tid);
2426 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2429 /* The cm_id may be null if we failed to connect */
2430 mutex_lock(&ep->com.mutex);
2431 switch (ep->com.state) {
2433 __state_set(&ep->com, MORIBUND);
2437 if ((ep->com.cm_id) && (ep->com.qp)) {
2438 attrs.next_state = C4IW_QP_STATE_IDLE;
2439 c4iw_modify_qp(ep->com.qp->rhp,
2441 C4IW_QP_ATTR_NEXT_STATE,
2444 close_complete_upcall(ep, 0);
2445 __state_set(&ep->com, DEAD);
2455 mutex_unlock(&ep->com.mutex);
2457 release_ep_resources(ep);
2461 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2463 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2464 struct tid_info *t = dev->rdev.lldi.tids;
2465 unsigned int tid = GET_TID(rpl);
2467 struct c4iw_qp_attributes attrs;
2469 ep = lookup_tid(t, tid);
2472 if (ep && ep->com.qp) {
2473 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2474 ep->com.qp->wq.sq.qid);
2475 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2476 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2477 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2479 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2485 * Upcall from the adapter indicating data has been transmitted.
2486 * For us its just the single MPA request or reply. We can now free
2487 * the skb holding the mpa message.
2489 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2492 struct cpl_fw4_ack *hdr = cplhdr(skb);
2493 u8 credits = hdr->credits;
2494 unsigned int tid = GET_TID(hdr);
2495 struct tid_info *t = dev->rdev.lldi.tids;
2498 ep = lookup_tid(t, tid);
2499 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2501 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2502 __func__, ep, ep->hwtid, state_read(&ep->com));
2506 dst_confirm(ep->dst);
2508 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2509 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2510 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2511 kfree_skb(ep->mpa_skb);
2517 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2521 struct c4iw_ep *ep = to_ep(cm_id);
2522 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2524 mutex_lock(&ep->com.mutex);
2525 if (ep->com.state == DEAD) {
2526 mutex_unlock(&ep->com.mutex);
2527 c4iw_put_ep(&ep->com);
2530 set_bit(ULP_REJECT, &ep->com.history);
2531 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2533 abort_connection(ep, NULL, GFP_KERNEL);
2535 err = send_mpa_reject(ep, pdata, pdata_len);
2538 mutex_unlock(&ep->com.mutex);
2540 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2541 c4iw_put_ep(&ep->com);
2545 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2548 struct c4iw_qp_attributes attrs;
2549 enum c4iw_qp_attr_mask mask;
2550 struct c4iw_ep *ep = to_ep(cm_id);
2551 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2552 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2554 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2556 mutex_lock(&ep->com.mutex);
2557 if (ep->com.state == DEAD) {
2562 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2565 set_bit(ULP_ACCEPT, &ep->com.history);
2566 if ((conn_param->ord > c4iw_max_read_depth) ||
2567 (conn_param->ird > c4iw_max_read_depth)) {
2568 abort_connection(ep, NULL, GFP_KERNEL);
2573 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2574 if (conn_param->ord > ep->ird) {
2575 ep->ird = conn_param->ird;
2576 ep->ord = conn_param->ord;
2577 send_mpa_reject(ep, conn_param->private_data,
2578 conn_param->private_data_len);
2579 abort_connection(ep, NULL, GFP_KERNEL);
2583 if (conn_param->ird > ep->ord) {
2585 conn_param->ird = 1;
2587 abort_connection(ep, NULL, GFP_KERNEL);
2594 ep->ird = conn_param->ird;
2595 ep->ord = conn_param->ord;
2597 if (ep->mpa_attr.version != 2)
2598 if (peer2peer && ep->ird == 0)
2601 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2603 cm_id->add_ref(cm_id);
2604 ep->com.cm_id = cm_id;
2608 /* bind QP to EP and move to RTS */
2609 attrs.mpa_attr = ep->mpa_attr;
2610 attrs.max_ird = ep->ird;
2611 attrs.max_ord = ep->ord;
2612 attrs.llp_stream_handle = ep;
2613 attrs.next_state = C4IW_QP_STATE_RTS;
2615 /* bind QP and TID with INIT_WR */
2616 mask = C4IW_QP_ATTR_NEXT_STATE |
2617 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2618 C4IW_QP_ATTR_MPA_ATTR |
2619 C4IW_QP_ATTR_MAX_IRD |
2620 C4IW_QP_ATTR_MAX_ORD;
2622 err = c4iw_modify_qp(ep->com.qp->rhp,
2623 ep->com.qp, mask, &attrs, 1);
2626 err = send_mpa_reply(ep, conn_param->private_data,
2627 conn_param->private_data_len);
2631 __state_set(&ep->com, FPDU_MODE);
2632 established_upcall(ep);
2633 mutex_unlock(&ep->com.mutex);
2634 c4iw_put_ep(&ep->com);
2637 ep->com.cm_id = NULL;
2638 cm_id->rem_ref(cm_id);
2640 mutex_unlock(&ep->com.mutex);
2641 c4iw_put_ep(&ep->com);
2645 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2647 struct in_device *ind;
2649 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2650 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2652 ind = in_dev_get(dev->rdev.lldi.ports[0]);
2654 return -EADDRNOTAVAIL;
2655 for_primary_ifa(ind) {
2656 laddr->sin_addr.s_addr = ifa->ifa_address;
2657 raddr->sin_addr.s_addr = ifa->ifa_address;
2663 return found ? 0 : -EADDRNOTAVAIL;
2666 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2667 unsigned char banned_flags)
2669 struct inet6_dev *idev;
2670 int err = -EADDRNOTAVAIL;
2673 idev = __in6_dev_get(dev);
2675 struct inet6_ifaddr *ifp;
2677 read_lock_bh(&idev->lock);
2678 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2679 if (ifp->scope == IFA_LINK &&
2680 !(ifp->flags & banned_flags)) {
2681 memcpy(addr, &ifp->addr, 16);
2686 read_unlock_bh(&idev->lock);
2692 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2694 struct in6_addr uninitialized_var(addr);
2695 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2696 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2698 if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2699 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2700 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2703 return -EADDRNOTAVAIL;
2706 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2708 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2711 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2712 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2713 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2714 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2715 &cm_id->remote_addr;
2719 if ((conn_param->ord > c4iw_max_read_depth) ||
2720 (conn_param->ird > c4iw_max_read_depth)) {
2724 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2726 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2730 init_timer(&ep->timer);
2731 ep->plen = conn_param->private_data_len;
2733 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2734 conn_param->private_data, ep->plen);
2735 ep->ird = conn_param->ird;
2736 ep->ord = conn_param->ord;
2738 if (peer2peer && ep->ord == 0)
2741 cm_id->add_ref(cm_id);
2743 ep->com.cm_id = cm_id;
2744 ep->com.qp = get_qhp(dev, conn_param->qpn);
2746 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2751 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2755 * Allocate an active TID to initiate a TCP connection.
2757 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2758 if (ep->atid == -1) {
2759 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2763 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2765 if (cm_id->remote_addr.ss_family == AF_INET) {
2767 ra = (__u8 *)&raddr->sin_addr;
2770 * Handle loopback requests to INADDR_ANY.
2772 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2773 err = pick_local_ipaddrs(dev, cm_id);
2779 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2780 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2781 ra, ntohs(raddr->sin_port));
2782 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2783 raddr->sin_addr.s_addr, laddr->sin_port,
2784 raddr->sin_port, 0);
2787 ra = (__u8 *)&raddr6->sin6_addr;
2790 * Handle loopback requests to INADDR_ANY.
2792 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2793 err = pick_local_ip6addrs(dev, cm_id);
2799 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2800 __func__, laddr6->sin6_addr.s6_addr,
2801 ntohs(laddr6->sin6_port),
2802 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2803 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2804 raddr6->sin6_addr.s6_addr,
2805 laddr6->sin6_port, raddr6->sin6_port, 0,
2806 raddr6->sin6_scope_id);
2809 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2810 err = -EHOSTUNREACH;
2814 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2816 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2820 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2821 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2824 state_set(&ep->com, CONNECTING);
2826 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2827 sizeof(ep->com.local_addr));
2828 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2829 sizeof(ep->com.remote_addr));
2831 /* send connect request to rnic */
2832 err = send_connect(ep);
2836 cxgb4_l2t_release(ep->l2t);
2838 dst_release(ep->dst);
2840 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2841 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2843 cm_id->rem_ref(cm_id);
2844 c4iw_put_ep(&ep->com);
2849 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2852 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2854 c4iw_init_wr_wait(&ep->com.wr_wait);
2855 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2856 ep->stid, &sin6->sin6_addr,
2858 ep->com.dev->rdev.lldi.rxq_ids[0]);
2860 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2864 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2866 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2870 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2873 struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2875 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2877 err = cxgb4_create_server_filter(
2878 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2879 sin->sin_addr.s_addr, sin->sin_port, 0,
2880 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2881 if (err == -EBUSY) {
2882 set_current_state(TASK_UNINTERRUPTIBLE);
2883 schedule_timeout(usecs_to_jiffies(100));
2885 } while (err == -EBUSY);
2887 c4iw_init_wr_wait(&ep->com.wr_wait);
2888 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2889 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2890 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2892 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2897 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2899 &sin->sin_addr, ntohs(sin->sin_port));
2903 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2906 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2907 struct c4iw_listen_ep *ep;
2911 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2913 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2917 PDBG("%s ep %p\n", __func__, ep);
2918 cm_id->add_ref(cm_id);
2919 ep->com.cm_id = cm_id;
2921 ep->backlog = backlog;
2922 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2923 sizeof(ep->com.local_addr));
2926 * Allocate a server TID.
2928 if (dev->rdev.lldi.enable_fw_ofld_conn &&
2929 ep->com.local_addr.ss_family == AF_INET)
2930 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2931 cm_id->local_addr.ss_family, ep);
2933 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2934 cm_id->local_addr.ss_family, ep);
2936 if (ep->stid == -1) {
2937 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2941 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2942 state_set(&ep->com, LISTEN);
2943 if (ep->com.local_addr.ss_family == AF_INET)
2944 err = create_server4(dev, ep);
2946 err = create_server6(dev, ep);
2948 cm_id->provider_data = ep;
2951 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2952 ep->com.local_addr.ss_family);
2954 cm_id->rem_ref(cm_id);
2955 c4iw_put_ep(&ep->com);
2961 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2964 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2966 PDBG("%s ep %p\n", __func__, ep);
2969 state_set(&ep->com, DEAD);
2970 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2971 ep->com.local_addr.ss_family == AF_INET) {
2972 err = cxgb4_remove_server_filter(
2973 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2974 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2976 c4iw_init_wr_wait(&ep->com.wr_wait);
2977 err = cxgb4_remove_server(
2978 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2979 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2982 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2985 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2986 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2987 ep->com.local_addr.ss_family);
2989 cm_id->rem_ref(cm_id);
2990 c4iw_put_ep(&ep->com);
2994 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2999 struct c4iw_rdev *rdev;
3001 mutex_lock(&ep->com.mutex);
3003 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3004 states[ep->com.state], abrupt);
3006 rdev = &ep->com.dev->rdev;
3007 if (c4iw_fatal_error(rdev)) {
3009 close_complete_upcall(ep, -EIO);
3010 ep->com.state = DEAD;
3012 switch (ep->com.state) {
3020 ep->com.state = ABORTING;
3022 ep->com.state = CLOSING;
3025 set_bit(CLOSE_SENT, &ep->com.flags);
3028 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3032 ep->com.state = ABORTING;
3034 ep->com.state = MORIBUND;
3040 PDBG("%s ignoring disconnect ep %p state %u\n",
3041 __func__, ep, ep->com.state);
3050 set_bit(EP_DISC_ABORT, &ep->com.history);
3051 close_complete_upcall(ep, -ECONNRESET);
3052 ret = send_abort(ep, NULL, gfp);
3054 set_bit(EP_DISC_CLOSE, &ep->com.history);
3055 ret = send_halfclose(ep, gfp);
3060 mutex_unlock(&ep->com.mutex);
3062 release_ep_resources(ep);
3066 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3067 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3070 int atid = be32_to_cpu(req->tid);
3072 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3073 (__force u32) req->tid);
3077 switch (req->retval) {
3079 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3080 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3081 send_fw_act_open_req(ep, atid);
3085 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3086 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3087 send_fw_act_open_req(ep, atid);
3092 pr_info("%s unexpected ofld conn wr retval %d\n",
3093 __func__, req->retval);
3096 pr_err("active ofld_connect_wr failure %d atid %d\n",
3098 mutex_lock(&dev->rdev.stats.lock);
3099 dev->rdev.stats.act_ofld_conn_fails++;
3100 mutex_unlock(&dev->rdev.stats.lock);
3101 connect_reply_upcall(ep, status2errno(req->retval));
3102 state_set(&ep->com, DEAD);
3103 remove_handle(dev, &dev->atid_idr, atid);
3104 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3105 dst_release(ep->dst);
3106 cxgb4_l2t_release(ep->l2t);
3107 c4iw_put_ep(&ep->com);
3110 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3111 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3113 struct sk_buff *rpl_skb;
3114 struct cpl_pass_accept_req *cpl;
3117 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3120 PDBG("%s passive open failure %d\n", __func__, req->retval);
3121 mutex_lock(&dev->rdev.stats.lock);
3122 dev->rdev.stats.pas_ofld_conn_fails++;
3123 mutex_unlock(&dev->rdev.stats.lock);
3126 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3127 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3128 (__force u32) htonl(
3129 (__force u32) req->tid)));
3130 ret = pass_accept_req(dev, rpl_skb);
3137 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3139 struct cpl_fw6_msg *rpl = cplhdr(skb);
3140 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3142 switch (rpl->type) {
3144 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3146 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3147 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3148 switch (req->t_state) {
3150 active_ofld_conn_reply(dev, skb, req);
3153 passive_ofld_conn_reply(dev, skb, req);
3156 pr_err("%s unexpected ofld conn wr state %d\n",
3157 __func__, req->t_state);
3165 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3168 u16 vlantag, len, hdr_len, eth_hdr_len;
3170 struct cpl_rx_pkt *cpl = cplhdr(skb);
3171 struct cpl_pass_accept_req *req;
3172 struct tcp_options_received tmp_opt;
3173 struct c4iw_dev *dev;
3175 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3176 /* Store values from cpl_rx_pkt in temporary location. */
3177 vlantag = (__force u16) cpl->vlan;
3178 len = (__force u16) cpl->len;
3179 l2info = (__force u32) cpl->l2info;
3180 hdr_len = (__force u16) cpl->hdr_len;
3183 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3186 * We need to parse the TCP options from SYN packet.
3187 * to generate cpl_pass_accept_req.
3189 memset(&tmp_opt, 0, sizeof(tmp_opt));
3190 tcp_clear_options(&tmp_opt);
3191 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3193 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3194 memset(req, 0, sizeof(*req));
3195 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3196 V_SYN_MAC_IDX(G_RX_MACIDX(
3197 (__force int) htonl(l2info))) |
3199 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3200 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3201 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3202 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3203 (__force int) htonl(l2info))) |
3204 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3205 (__force int) htons(hdr_len))) |
3206 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3207 (__force int) htons(hdr_len))) |
3208 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3209 req->vlan = (__force __be16) vlantag;
3210 req->len = (__force __be16) len;
3211 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3212 PASS_OPEN_TOS(tos));
3213 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3214 if (tmp_opt.wscale_ok)
3215 req->tcpopt.wsf = tmp_opt.snd_wscale;
3216 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3217 if (tmp_opt.sack_ok)
3218 req->tcpopt.sack = 1;
3219 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3223 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3224 __be32 laddr, __be16 lport,
3225 __be32 raddr, __be16 rport,
3226 u32 rcv_isn, u32 filter, u16 window,
3227 u32 rss_qid, u8 port_id)
3229 struct sk_buff *req_skb;
3230 struct fw_ofld_connection_wr *req;
3231 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3234 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3235 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3236 memset(req, 0, sizeof(*req));
3237 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3238 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3239 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3240 req->le.filter = (__force __be32) filter;
3241 req->le.lport = lport;
3242 req->le.pport = rport;
3243 req->le.u.ipv4.lip = laddr;
3244 req->le.u.ipv4.pip = raddr;
3245 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3246 req->tcb.rcv_adv = htons(window);
3247 req->tcb.t_state_to_astid =
3248 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3249 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3250 V_FW_OFLD_CONNECTION_WR_ASTID(
3251 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3254 * We store the qid in opt2 which will be used by the firmware
3255 * to send us the wr response.
3257 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3260 * We initialize the MSS index in TCB to 0xF.
3261 * So that when driver sends cpl_pass_accept_rpl
3262 * TCB picks up the correct value. If this was 0
3263 * TP will ignore any value > 0 for MSS index.
3265 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3266 req->cookie = (unsigned long)skb;
3268 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3269 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3271 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3279 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3280 * messages when a filter is being used instead of server to
3281 * redirect a syn packet. When packets hit filter they are redirected
3282 * to the offload queue and driver tries to establish the connection
3283 * using firmware work request.
3285 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3288 unsigned int filter;
3289 struct ethhdr *eh = NULL;
3290 struct vlan_ethhdr *vlan_eh = NULL;
3292 struct tcphdr *tcph;
3293 struct rss_header *rss = (void *)skb->data;
3294 struct cpl_rx_pkt *cpl = (void *)skb->data;
3295 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3296 struct l2t_entry *e;
3297 struct dst_entry *dst;
3298 struct c4iw_ep *lep;
3300 struct port_info *pi;
3301 struct net_device *pdev;
3302 u16 rss_qid, eth_hdr_len;
3305 struct neighbour *neigh;
3307 /* Drop all non-SYN packets */
3308 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3312 * Drop all packets which did not hit the filter.
3313 * Unlikely to happen.
3315 if (!(rss->filter_hit && rss->filter_tid))
3319 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3321 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3323 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3325 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3329 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3330 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3331 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3332 if (eth_hdr_len == ETH_HLEN) {
3333 eh = (struct ethhdr *)(req + 1);
3334 iph = (struct iphdr *)(eh + 1);
3336 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3337 iph = (struct iphdr *)(vlan_eh + 1);
3338 skb->vlan_tci = ntohs(cpl->vlan);
3341 if (iph->version != 0x4)
3344 tcph = (struct tcphdr *)(iph + 1);
3345 skb_set_network_header(skb, (void *)iph - (void *)rss);
3346 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3349 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3350 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3351 ntohs(tcph->source), iph->tos);
3353 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3356 pr_err("%s - failed to find dst entry!\n",
3360 neigh = dst_neigh_lookup_skb(dst, skb);
3363 pr_err("%s - failed to allocate neigh!\n",
3368 if (neigh->dev->flags & IFF_LOOPBACK) {
3369 pdev = ip_dev_find(&init_net, iph->daddr);
3370 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3372 pi = (struct port_info *)netdev_priv(pdev);
3373 tx_chan = cxgb4_port_chan(pdev);
3376 pdev = get_real_dev(neigh->dev);
3377 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3379 pi = (struct port_info *)netdev_priv(pdev);
3380 tx_chan = cxgb4_port_chan(pdev);
3382 neigh_release(neigh);
3384 pr_err("%s - failed to allocate l2t entry!\n",
3389 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3390 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3391 window = (__force u16) htons((__force u16)tcph->window);
3393 /* Calcuate filter portion for LE region. */
3394 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3395 dev->rdev.lldi.ports[0],
3399 * Synthesize the cpl_pass_accept_req. We have everything except the
3400 * TID. Once firmware sends a reply with TID we update the TID field
3401 * in cpl and pass it through the regular cpl_pass_accept_req path.
3403 build_cpl_pass_accept_req(skb, stid, iph->tos);
3404 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3405 tcph->source, ntohl(tcph->seq), filter, window,
3406 rss_qid, pi->port_id);
3407 cxgb4_l2t_release(e);
3415 * These are the real handlers that are called from a
3418 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3419 [CPL_ACT_ESTABLISH] = act_establish,
3420 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3421 [CPL_RX_DATA] = rx_data,
3422 [CPL_ABORT_RPL_RSS] = abort_rpl,
3423 [CPL_ABORT_RPL] = abort_rpl,
3424 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3425 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3426 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3427 [CPL_PASS_ESTABLISH] = pass_establish,
3428 [CPL_PEER_CLOSE] = peer_close,
3429 [CPL_ABORT_REQ_RSS] = peer_abort,
3430 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3431 [CPL_RDMA_TERMINATE] = terminate,
3432 [CPL_FW4_ACK] = fw4_ack,
3433 [CPL_FW6_MSG] = deferred_fw6_msg,
3434 [CPL_RX_PKT] = rx_pkt
3437 static void process_timeout(struct c4iw_ep *ep)
3439 struct c4iw_qp_attributes attrs;
3442 mutex_lock(&ep->com.mutex);
3443 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3445 set_bit(TIMEDOUT, &ep->com.history);
3446 switch (ep->com.state) {
3448 __state_set(&ep->com, ABORTING);
3449 connect_reply_upcall(ep, -ETIMEDOUT);
3452 __state_set(&ep->com, ABORTING);
3456 if (ep->com.cm_id && ep->com.qp) {
3457 attrs.next_state = C4IW_QP_STATE_ERROR;
3458 c4iw_modify_qp(ep->com.qp->rhp,
3459 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3462 __state_set(&ep->com, ABORTING);
3463 close_complete_upcall(ep, -ETIMEDOUT);
3466 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3467 __func__, ep, ep->hwtid, ep->com.state);
3470 mutex_unlock(&ep->com.mutex);
3472 abort_connection(ep, NULL, GFP_KERNEL);
3473 c4iw_put_ep(&ep->com);
3476 static void process_timedout_eps(void)
3480 spin_lock_irq(&timeout_lock);
3481 while (!list_empty(&timeout_list)) {
3482 struct list_head *tmp;
3484 tmp = timeout_list.next;
3486 spin_unlock_irq(&timeout_lock);
3487 ep = list_entry(tmp, struct c4iw_ep, entry);
3488 process_timeout(ep);
3489 spin_lock_irq(&timeout_lock);
3491 spin_unlock_irq(&timeout_lock);
3494 static void process_work(struct work_struct *work)
3496 struct sk_buff *skb = NULL;
3497 struct c4iw_dev *dev;
3498 struct cpl_act_establish *rpl;
3499 unsigned int opcode;
3502 while ((skb = skb_dequeue(&rxq))) {
3504 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3505 opcode = rpl->ot.opcode;
3507 BUG_ON(!work_handlers[opcode]);
3508 ret = work_handlers[opcode](dev, skb);
3512 process_timedout_eps();
3515 static DECLARE_WORK(skb_work, process_work);
3517 static void ep_timeout(unsigned long arg)
3519 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3522 spin_lock(&timeout_lock);
3523 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3524 list_add_tail(&ep->entry, &timeout_list);
3527 spin_unlock(&timeout_lock);
3529 queue_work(workq, &skb_work);
3533 * All the CM events are handled on a work queue to have a safe context.
3535 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3539 * Save dev in the skb->cb area.
3541 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3544 * Queue the skb and schedule the worker thread.
3546 skb_queue_tail(&rxq, skb);
3547 queue_work(workq, &skb_work);
3551 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3553 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3555 if (rpl->status != CPL_ERR_NONE) {
3556 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3557 "for tid %u\n", rpl->status, GET_TID(rpl));
3563 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3565 struct cpl_fw6_msg *rpl = cplhdr(skb);
3566 struct c4iw_wr_wait *wr_waitp;
3569 PDBG("%s type %u\n", __func__, rpl->type);
3571 switch (rpl->type) {
3572 case FW6_TYPE_WR_RPL:
3573 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3574 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3575 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3577 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3581 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3585 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3593 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3595 struct cpl_abort_req_rss *req = cplhdr(skb);
3597 struct tid_info *t = dev->rdev.lldi.tids;
3598 unsigned int tid = GET_TID(req);
3600 ep = lookup_tid(t, tid);
3602 printk(KERN_WARNING MOD
3603 "Abort on non-existent endpoint, tid %d\n", tid);
3607 if (is_neg_adv(req->status)) {
3608 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3613 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3617 * Wake up any threads in rdma_init() or rdma_fini().
3618 * However, if we are on MPAv2 and want to retry with MPAv1
3619 * then, don't wake up yet.
3621 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3622 if (ep->com.state != MPA_REQ_SENT)
3623 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3625 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3631 * Most upcalls from the T4 Core go to sched() to
3632 * schedule the processing on a work queue.
3634 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3635 [CPL_ACT_ESTABLISH] = sched,
3636 [CPL_ACT_OPEN_RPL] = sched,
3637 [CPL_RX_DATA] = sched,
3638 [CPL_ABORT_RPL_RSS] = sched,
3639 [CPL_ABORT_RPL] = sched,
3640 [CPL_PASS_OPEN_RPL] = sched,
3641 [CPL_CLOSE_LISTSRV_RPL] = sched,
3642 [CPL_PASS_ACCEPT_REQ] = sched,
3643 [CPL_PASS_ESTABLISH] = sched,
3644 [CPL_PEER_CLOSE] = sched,
3645 [CPL_CLOSE_CON_RPL] = sched,
3646 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3647 [CPL_RDMA_TERMINATE] = sched,
3648 [CPL_FW4_ACK] = sched,
3649 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3650 [CPL_FW6_MSG] = fw6_msg,
3651 [CPL_RX_PKT] = sched
3654 int __init c4iw_cm_init(void)
3656 spin_lock_init(&timeout_lock);
3657 skb_queue_head_init(&rxq);
3659 workq = create_singlethread_workqueue("iw_cxgb4");
3666 void __exit c4iw_cm_term(void)
3668 WARN_ON(!list_empty(&timeout_list));
3669 flush_workqueue(workq);
3670 destroy_workqueue(workq);