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;
768 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
771 struct fw_ofld_tx_data_wr *req;
772 struct mpa_message *mpa;
774 struct mpa_v2_conn_params mpa_v2_params;
776 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
778 mpalen = sizeof(*mpa) + plen;
779 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
780 mpalen += sizeof(struct mpa_v2_conn_params);
781 wrlen = roundup(mpalen + sizeof *req, 16);
783 skb = get_skb(NULL, wrlen, GFP_KERNEL);
785 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
788 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
790 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
791 memset(req, 0, wrlen);
792 req->op_to_immdlen = cpu_to_be32(
793 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
795 FW_WR_IMMDLEN(mpalen));
796 req->flowid_len16 = cpu_to_be32(
797 FW_WR_FLOWID(ep->hwtid) |
798 FW_WR_LEN16(wrlen >> 4));
799 req->plen = cpu_to_be32(mpalen);
800 req->tunnel_to_proxy = cpu_to_be32(
801 FW_OFLD_TX_DATA_WR_FLUSH(1) |
802 FW_OFLD_TX_DATA_WR_SHOVE(1));
804 mpa = (struct mpa_message *)(req + 1);
805 memset(mpa, 0, sizeof(*mpa));
806 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
807 mpa->flags = MPA_REJECT;
808 mpa->revision = ep->mpa_attr.version;
809 mpa->private_data_size = htons(plen);
811 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
812 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
813 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
814 sizeof (struct mpa_v2_conn_params));
815 mpa_v2_params.ird = htons(((u16)ep->ird) |
816 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
818 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
820 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
821 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
822 FW_RI_INIT_P2PTYPE_READ_REQ ?
823 MPA_V2_RDMA_READ_RTR : 0) : 0));
824 memcpy(mpa->private_data, &mpa_v2_params,
825 sizeof(struct mpa_v2_conn_params));
828 memcpy(mpa->private_data +
829 sizeof(struct mpa_v2_conn_params), pdata, plen);
832 memcpy(mpa->private_data, pdata, plen);
835 * Reference the mpa skb again. This ensures the data area
836 * will remain in memory until the hw acks the tx.
837 * Function fw4_ack() will deref it.
840 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
841 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
844 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
847 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
850 struct fw_ofld_tx_data_wr *req;
851 struct mpa_message *mpa;
853 struct mpa_v2_conn_params mpa_v2_params;
855 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
857 mpalen = sizeof(*mpa) + plen;
858 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
859 mpalen += sizeof(struct mpa_v2_conn_params);
860 wrlen = roundup(mpalen + sizeof *req, 16);
862 skb = get_skb(NULL, wrlen, GFP_KERNEL);
864 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
867 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
869 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
870 memset(req, 0, wrlen);
871 req->op_to_immdlen = cpu_to_be32(
872 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
874 FW_WR_IMMDLEN(mpalen));
875 req->flowid_len16 = cpu_to_be32(
876 FW_WR_FLOWID(ep->hwtid) |
877 FW_WR_LEN16(wrlen >> 4));
878 req->plen = cpu_to_be32(mpalen);
879 req->tunnel_to_proxy = cpu_to_be32(
880 FW_OFLD_TX_DATA_WR_FLUSH(1) |
881 FW_OFLD_TX_DATA_WR_SHOVE(1));
883 mpa = (struct mpa_message *)(req + 1);
884 memset(mpa, 0, sizeof(*mpa));
885 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
886 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
887 (markers_enabled ? MPA_MARKERS : 0);
888 mpa->revision = ep->mpa_attr.version;
889 mpa->private_data_size = htons(plen);
891 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
892 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
893 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
894 sizeof (struct mpa_v2_conn_params));
895 mpa_v2_params.ird = htons((u16)ep->ird);
896 mpa_v2_params.ord = htons((u16)ep->ord);
897 if (peer2peer && (ep->mpa_attr.p2p_type !=
898 FW_RI_INIT_P2PTYPE_DISABLED)) {
899 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
901 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
903 htons(MPA_V2_RDMA_WRITE_RTR);
904 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
906 htons(MPA_V2_RDMA_READ_RTR);
909 memcpy(mpa->private_data, &mpa_v2_params,
910 sizeof(struct mpa_v2_conn_params));
913 memcpy(mpa->private_data +
914 sizeof(struct mpa_v2_conn_params), pdata, plen);
917 memcpy(mpa->private_data, pdata, plen);
920 * Reference the mpa skb. This ensures the data area
921 * will remain in memory until the hw acks the tx.
922 * Function fw4_ack() will deref it.
925 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
927 state_set(&ep->com, MPA_REP_SENT);
928 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
931 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
934 struct cpl_act_establish *req = cplhdr(skb);
935 unsigned int tid = GET_TID(req);
936 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
937 struct tid_info *t = dev->rdev.lldi.tids;
939 ep = lookup_atid(t, atid);
941 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
942 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
944 dst_confirm(ep->dst);
946 /* setup the hwtid for this connection */
948 cxgb4_insert_tid(t, ep, tid);
949 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
951 ep->snd_seq = be32_to_cpu(req->snd_isn);
952 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
954 set_emss(ep, ntohs(req->tcp_opt));
956 /* dealloc the atid */
957 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
958 cxgb4_free_atid(t, atid);
959 set_bit(ACT_ESTAB, &ep->com.history);
961 /* start MPA negotiation */
962 send_flowc(ep, NULL);
963 if (ep->retry_with_mpa_v1)
964 send_mpa_req(ep, skb, 1);
966 send_mpa_req(ep, skb, mpa_rev);
971 static void close_complete_upcall(struct c4iw_ep *ep, int status)
973 struct iw_cm_event event;
975 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
976 memset(&event, 0, sizeof(event));
977 event.event = IW_CM_EVENT_CLOSE;
978 event.status = status;
980 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
981 ep, ep->com.cm_id, ep->hwtid);
982 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
983 ep->com.cm_id->rem_ref(ep->com.cm_id);
984 ep->com.cm_id = NULL;
985 set_bit(CLOSE_UPCALL, &ep->com.history);
989 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
991 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
992 state_set(&ep->com, ABORTING);
993 set_bit(ABORT_CONN, &ep->com.history);
994 return send_abort(ep, skb, gfp);
997 static void peer_close_upcall(struct c4iw_ep *ep)
999 struct iw_cm_event event;
1001 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1002 memset(&event, 0, sizeof(event));
1003 event.event = IW_CM_EVENT_DISCONNECT;
1004 if (ep->com.cm_id) {
1005 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1006 ep, ep->com.cm_id, ep->hwtid);
1007 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1008 set_bit(DISCONN_UPCALL, &ep->com.history);
1012 static void peer_abort_upcall(struct c4iw_ep *ep)
1014 struct iw_cm_event event;
1016 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1017 memset(&event, 0, sizeof(event));
1018 event.event = IW_CM_EVENT_CLOSE;
1019 event.status = -ECONNRESET;
1020 if (ep->com.cm_id) {
1021 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1022 ep->com.cm_id, ep->hwtid);
1023 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1024 ep->com.cm_id->rem_ref(ep->com.cm_id);
1025 ep->com.cm_id = NULL;
1026 set_bit(ABORT_UPCALL, &ep->com.history);
1030 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1032 struct iw_cm_event event;
1034 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1035 memset(&event, 0, sizeof(event));
1036 event.event = IW_CM_EVENT_CONNECT_REPLY;
1037 event.status = status;
1038 memcpy(&event.local_addr, &ep->com.local_addr,
1039 sizeof(ep->com.local_addr));
1040 memcpy(&event.remote_addr, &ep->com.remote_addr,
1041 sizeof(ep->com.remote_addr));
1043 if ((status == 0) || (status == -ECONNREFUSED)) {
1044 if (!ep->tried_with_mpa_v1) {
1045 /* this means MPA_v2 is used */
1046 event.private_data_len = ep->plen -
1047 sizeof(struct mpa_v2_conn_params);
1048 event.private_data = ep->mpa_pkt +
1049 sizeof(struct mpa_message) +
1050 sizeof(struct mpa_v2_conn_params);
1052 /* this means MPA_v1 is used */
1053 event.private_data_len = ep->plen;
1054 event.private_data = ep->mpa_pkt +
1055 sizeof(struct mpa_message);
1059 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1061 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1062 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1065 ep->com.cm_id->rem_ref(ep->com.cm_id);
1066 ep->com.cm_id = NULL;
1070 static int connect_request_upcall(struct c4iw_ep *ep)
1072 struct iw_cm_event event;
1075 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1076 memset(&event, 0, sizeof(event));
1077 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1078 memcpy(&event.local_addr, &ep->com.local_addr,
1079 sizeof(ep->com.local_addr));
1080 memcpy(&event.remote_addr, &ep->com.remote_addr,
1081 sizeof(ep->com.remote_addr));
1082 event.provider_data = ep;
1083 if (!ep->tried_with_mpa_v1) {
1084 /* this means MPA_v2 is used */
1085 event.ord = ep->ord;
1086 event.ird = ep->ird;
1087 event.private_data_len = ep->plen -
1088 sizeof(struct mpa_v2_conn_params);
1089 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1090 sizeof(struct mpa_v2_conn_params);
1092 /* this means MPA_v1 is used. Send max supported */
1093 event.ord = c4iw_max_read_depth;
1094 event.ird = c4iw_max_read_depth;
1095 event.private_data_len = ep->plen;
1096 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1098 c4iw_get_ep(&ep->com);
1099 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1102 c4iw_put_ep(&ep->com);
1103 set_bit(CONNREQ_UPCALL, &ep->com.history);
1104 c4iw_put_ep(&ep->parent_ep->com);
1108 static void established_upcall(struct c4iw_ep *ep)
1110 struct iw_cm_event event;
1112 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1113 memset(&event, 0, sizeof(event));
1114 event.event = IW_CM_EVENT_ESTABLISHED;
1115 event.ird = ep->ird;
1116 event.ord = ep->ord;
1117 if (ep->com.cm_id) {
1118 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1119 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1120 set_bit(ESTAB_UPCALL, &ep->com.history);
1124 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1126 struct cpl_rx_data_ack *req;
1127 struct sk_buff *skb;
1128 int wrlen = roundup(sizeof *req, 16);
1130 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1131 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1133 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1137 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1138 memset(req, 0, wrlen);
1139 INIT_TP_WR(req, ep->hwtid);
1140 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1142 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1144 V_RX_DACK_MODE(dack_mode));
1145 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1146 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1150 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1152 struct mpa_message *mpa;
1153 struct mpa_v2_conn_params *mpa_v2_params;
1155 u16 resp_ird, resp_ord;
1156 u8 rtr_mismatch = 0, insuff_ird = 0;
1157 struct c4iw_qp_attributes attrs;
1158 enum c4iw_qp_attr_mask mask;
1161 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1164 * Stop mpa timer. If it expired, then the state has
1165 * changed and we bail since ep_timeout already aborted
1169 if (state_read(&ep->com) != MPA_REQ_SENT)
1173 * If we get more than the supported amount of private data
1174 * then we must fail this connection.
1176 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1182 * copy the new data into our accumulation buffer.
1184 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1186 ep->mpa_pkt_len += skb->len;
1189 * if we don't even have the mpa message, then bail.
1191 if (ep->mpa_pkt_len < sizeof(*mpa))
1193 mpa = (struct mpa_message *) ep->mpa_pkt;
1195 /* Validate MPA header. */
1196 if (mpa->revision > mpa_rev) {
1197 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1198 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1202 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1207 plen = ntohs(mpa->private_data_size);
1210 * Fail if there's too much private data.
1212 if (plen > MPA_MAX_PRIVATE_DATA) {
1218 * If plen does not account for pkt size
1220 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1225 ep->plen = (u8) plen;
1228 * If we don't have all the pdata yet, then bail.
1229 * We'll continue process when more data arrives.
1231 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1234 if (mpa->flags & MPA_REJECT) {
1235 err = -ECONNREFUSED;
1240 * If we get here we have accumulated the entire mpa
1241 * start reply message including private data. And
1242 * the MPA header is valid.
1244 state_set(&ep->com, FPDU_MODE);
1245 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1246 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1247 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1248 ep->mpa_attr.version = mpa->revision;
1249 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1251 if (mpa->revision == 2) {
1252 ep->mpa_attr.enhanced_rdma_conn =
1253 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1254 if (ep->mpa_attr.enhanced_rdma_conn) {
1255 mpa_v2_params = (struct mpa_v2_conn_params *)
1256 (ep->mpa_pkt + sizeof(*mpa));
1257 resp_ird = ntohs(mpa_v2_params->ird) &
1258 MPA_V2_IRD_ORD_MASK;
1259 resp_ord = ntohs(mpa_v2_params->ord) &
1260 MPA_V2_IRD_ORD_MASK;
1263 * This is a double-check. Ideally, below checks are
1264 * not required since ird/ord stuff has been taken
1265 * care of in c4iw_accept_cr
1267 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1274 if (ntohs(mpa_v2_params->ird) &
1275 MPA_V2_PEER2PEER_MODEL) {
1276 if (ntohs(mpa_v2_params->ord) &
1277 MPA_V2_RDMA_WRITE_RTR)
1278 ep->mpa_attr.p2p_type =
1279 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1280 else if (ntohs(mpa_v2_params->ord) &
1281 MPA_V2_RDMA_READ_RTR)
1282 ep->mpa_attr.p2p_type =
1283 FW_RI_INIT_P2PTYPE_READ_REQ;
1286 } else if (mpa->revision == 1)
1288 ep->mpa_attr.p2p_type = p2p_type;
1290 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1291 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1292 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1293 ep->mpa_attr.recv_marker_enabled,
1294 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1295 ep->mpa_attr.p2p_type, p2p_type);
1298 * If responder's RTR does not match with that of initiator, assign
1299 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1300 * generated when moving QP to RTS state.
1301 * A TERM message will be sent after QP has moved to RTS state
1303 if ((ep->mpa_attr.version == 2) && peer2peer &&
1304 (ep->mpa_attr.p2p_type != p2p_type)) {
1305 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1309 attrs.mpa_attr = ep->mpa_attr;
1310 attrs.max_ird = ep->ird;
1311 attrs.max_ord = ep->ord;
1312 attrs.llp_stream_handle = ep;
1313 attrs.next_state = C4IW_QP_STATE_RTS;
1315 mask = C4IW_QP_ATTR_NEXT_STATE |
1316 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1317 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1319 /* bind QP and TID with INIT_WR */
1320 err = c4iw_modify_qp(ep->com.qp->rhp,
1321 ep->com.qp, mask, &attrs, 1);
1326 * If responder's RTR requirement did not match with what initiator
1327 * supports, generate TERM message
1330 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1331 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1332 attrs.ecode = MPA_NOMATCH_RTR;
1333 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1334 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1335 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1341 * Generate TERM if initiator IRD is not sufficient for responder
1342 * provided ORD. Currently, we do the same behaviour even when
1343 * responder provided IRD is also not sufficient as regards to
1347 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1349 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1350 attrs.ecode = MPA_INSUFF_IRD;
1351 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1352 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1353 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1359 state_set(&ep->com, ABORTING);
1360 send_abort(ep, skb, GFP_KERNEL);
1362 connect_reply_upcall(ep, err);
1366 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1368 struct mpa_message *mpa;
1369 struct mpa_v2_conn_params *mpa_v2_params;
1372 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1374 if (state_read(&ep->com) != MPA_REQ_WAIT)
1378 * If we get more than the supported amount of private data
1379 * then we must fail this connection.
1381 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1383 abort_connection(ep, skb, GFP_KERNEL);
1387 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1390 * Copy the new data into our accumulation buffer.
1392 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1394 ep->mpa_pkt_len += skb->len;
1397 * If we don't even have the mpa message, then bail.
1398 * We'll continue process when more data arrives.
1400 if (ep->mpa_pkt_len < sizeof(*mpa))
1403 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1404 mpa = (struct mpa_message *) ep->mpa_pkt;
1407 * Validate MPA Header.
1409 if (mpa->revision > mpa_rev) {
1410 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1411 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1413 abort_connection(ep, skb, GFP_KERNEL);
1417 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1419 abort_connection(ep, skb, GFP_KERNEL);
1423 plen = ntohs(mpa->private_data_size);
1426 * Fail if there's too much private data.
1428 if (plen > MPA_MAX_PRIVATE_DATA) {
1430 abort_connection(ep, skb, GFP_KERNEL);
1435 * If plen does not account for pkt size
1437 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1439 abort_connection(ep, skb, GFP_KERNEL);
1442 ep->plen = (u8) plen;
1445 * If we don't have all the pdata yet, then bail.
1447 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1451 * If we get here we have accumulated the entire mpa
1452 * start reply message including private data.
1454 ep->mpa_attr.initiator = 0;
1455 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1456 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1457 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1458 ep->mpa_attr.version = mpa->revision;
1459 if (mpa->revision == 1)
1460 ep->tried_with_mpa_v1 = 1;
1461 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1463 if (mpa->revision == 2) {
1464 ep->mpa_attr.enhanced_rdma_conn =
1465 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1466 if (ep->mpa_attr.enhanced_rdma_conn) {
1467 mpa_v2_params = (struct mpa_v2_conn_params *)
1468 (ep->mpa_pkt + sizeof(*mpa));
1469 ep->ird = ntohs(mpa_v2_params->ird) &
1470 MPA_V2_IRD_ORD_MASK;
1471 ep->ord = ntohs(mpa_v2_params->ord) &
1472 MPA_V2_IRD_ORD_MASK;
1473 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1475 if (ntohs(mpa_v2_params->ord) &
1476 MPA_V2_RDMA_WRITE_RTR)
1477 ep->mpa_attr.p2p_type =
1478 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1479 else if (ntohs(mpa_v2_params->ord) &
1480 MPA_V2_RDMA_READ_RTR)
1481 ep->mpa_attr.p2p_type =
1482 FW_RI_INIT_P2PTYPE_READ_REQ;
1485 } else if (mpa->revision == 1)
1487 ep->mpa_attr.p2p_type = p2p_type;
1489 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1490 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1491 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1492 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1493 ep->mpa_attr.p2p_type);
1495 state_set(&ep->com, MPA_REQ_RCVD);
1499 mutex_lock(&ep->parent_ep->com.mutex);
1500 if (ep->parent_ep->com.state != DEAD) {
1501 if (connect_request_upcall(ep))
1502 abort_connection(ep, skb, GFP_KERNEL);
1504 abort_connection(ep, skb, GFP_KERNEL);
1506 mutex_unlock(&ep->parent_ep->com.mutex);
1510 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1513 struct cpl_rx_data *hdr = cplhdr(skb);
1514 unsigned int dlen = ntohs(hdr->len);
1515 unsigned int tid = GET_TID(hdr);
1516 struct tid_info *t = dev->rdev.lldi.tids;
1517 __u8 status = hdr->status;
1519 ep = lookup_tid(t, tid);
1520 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1521 skb_pull(skb, sizeof(*hdr));
1522 skb_trim(skb, dlen);
1524 /* update RX credits */
1525 update_rx_credits(ep, dlen);
1527 switch (state_read(&ep->com)) {
1529 ep->rcv_seq += dlen;
1530 process_mpa_reply(ep, skb);
1533 ep->rcv_seq += dlen;
1534 process_mpa_request(ep, skb);
1537 struct c4iw_qp_attributes attrs;
1538 BUG_ON(!ep->com.qp);
1540 pr_err("%s Unexpected streaming data." \
1541 " qpid %u ep %p state %d tid %u status %d\n",
1542 __func__, ep->com.qp->wq.sq.qid, ep,
1543 state_read(&ep->com), ep->hwtid, status);
1544 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1545 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1546 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1555 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1558 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1560 unsigned int tid = GET_TID(rpl);
1561 struct tid_info *t = dev->rdev.lldi.tids;
1563 ep = lookup_tid(t, tid);
1565 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1568 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1569 mutex_lock(&ep->com.mutex);
1570 switch (ep->com.state) {
1572 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1573 __state_set(&ep->com, DEAD);
1577 printk(KERN_ERR "%s ep %p state %d\n",
1578 __func__, ep, ep->com.state);
1581 mutex_unlock(&ep->com.mutex);
1584 release_ep_resources(ep);
1588 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1590 struct sk_buff *skb;
1591 struct fw_ofld_connection_wr *req;
1592 unsigned int mtu_idx;
1594 struct sockaddr_in *sin;
1596 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1597 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1598 memset(req, 0, sizeof(*req));
1599 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1600 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1601 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1602 ep->com.dev->rdev.lldi.ports[0],
1604 sin = (struct sockaddr_in *)&ep->com.local_addr;
1605 req->le.lport = sin->sin_port;
1606 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1607 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1608 req->le.pport = sin->sin_port;
1609 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1610 req->tcb.t_state_to_astid =
1611 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1612 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1613 req->tcb.cplrxdataack_cplpassacceptrpl =
1614 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1615 req->tcb.tx_max = (__force __be32) jiffies;
1616 req->tcb.rcv_adv = htons(1);
1617 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1618 wscale = compute_wscale(rcv_win);
1619 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1620 (nocong ? NO_CONG(1) : 0) |
1625 L2T_IDX(ep->l2t->idx) |
1626 TX_CHAN(ep->tx_chan) |
1627 SMAC_SEL(ep->smac_idx) |
1629 ULP_MODE(ULP_MODE_TCPDDP) |
1630 RCV_BUFSIZ(rcv_win >> 10));
1631 req->tcb.opt2 = (__force __be32) (PACE(1) |
1632 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1634 CCTRL_ECN(enable_ecn) |
1635 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1636 if (enable_tcp_timestamps)
1637 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1638 if (enable_tcp_sack)
1639 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1640 if (wscale && enable_tcp_window_scaling)
1641 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1642 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1643 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1644 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1645 set_bit(ACT_OFLD_CONN, &ep->com.history);
1646 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1650 * Return whether a failed active open has allocated a TID
1652 static inline int act_open_has_tid(int status)
1654 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1655 status != CPL_ERR_ARP_MISS;
1658 #define ACT_OPEN_RETRY_COUNT 2
1660 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1661 struct dst_entry *dst, struct c4iw_dev *cdev,
1664 struct neighbour *n;
1666 struct net_device *pdev;
1668 n = dst_neigh_lookup(dst, peer_ip);
1674 if (n->dev->flags & IFF_LOOPBACK) {
1676 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1677 else if (IS_ENABLED(CONFIG_IPV6))
1678 for_each_netdev(&init_net, pdev) {
1679 if (ipv6_chk_addr(&init_net,
1680 (struct in6_addr *)peer_ip,
1691 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1695 ep->mtu = pdev->mtu;
1696 ep->tx_chan = cxgb4_port_chan(pdev);
1697 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1698 step = cdev->rdev.lldi.ntxq /
1699 cdev->rdev.lldi.nchan;
1700 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1701 step = cdev->rdev.lldi.nrxq /
1702 cdev->rdev.lldi.nchan;
1703 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1704 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1705 cxgb4_port_idx(pdev) * step];
1708 pdev = get_real_dev(n->dev);
1709 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1713 ep->mtu = dst_mtu(dst);
1714 ep->tx_chan = cxgb4_port_chan(n->dev);
1715 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1716 step = cdev->rdev.lldi.ntxq /
1717 cdev->rdev.lldi.nchan;
1718 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1719 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1720 step = cdev->rdev.lldi.nrxq /
1721 cdev->rdev.lldi.nchan;
1722 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1723 cxgb4_port_idx(n->dev) * step];
1726 ep->retry_with_mpa_v1 = 0;
1727 ep->tried_with_mpa_v1 = 0;
1739 static int c4iw_reconnect(struct c4iw_ep *ep)
1742 struct sockaddr_in *laddr = (struct sockaddr_in *)
1743 &ep->com.cm_id->local_addr;
1744 struct sockaddr_in *raddr = (struct sockaddr_in *)
1745 &ep->com.cm_id->remote_addr;
1746 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1747 &ep->com.cm_id->local_addr;
1748 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1749 &ep->com.cm_id->remote_addr;
1753 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1754 init_timer(&ep->timer);
1757 * Allocate an active TID to initiate a TCP connection.
1759 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1760 if (ep->atid == -1) {
1761 pr_err("%s - cannot alloc atid.\n", __func__);
1765 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1768 if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1769 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1770 raddr->sin_addr.s_addr, laddr->sin_port,
1771 raddr->sin_port, 0);
1773 ra = (__u8 *)&raddr->sin_addr;
1775 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1776 raddr6->sin6_addr.s6_addr,
1777 laddr6->sin6_port, raddr6->sin6_port, 0,
1778 raddr6->sin6_scope_id);
1780 ra = (__u8 *)&raddr6->sin6_addr;
1783 pr_err("%s - cannot find route.\n", __func__);
1784 err = -EHOSTUNREACH;
1787 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1789 pr_err("%s - cannot alloc l2e.\n", __func__);
1793 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1794 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1797 state_set(&ep->com, CONNECTING);
1800 /* send connect request to rnic */
1801 err = send_connect(ep);
1805 cxgb4_l2t_release(ep->l2t);
1807 dst_release(ep->dst);
1809 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1810 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1813 * remember to send notification to upper layer.
1814 * We are in here so the upper layer is not aware that this is
1815 * re-connect attempt and so, upper layer is still waiting for
1816 * response of 1st connect request.
1818 connect_reply_upcall(ep, -ECONNRESET);
1819 c4iw_put_ep(&ep->com);
1824 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1827 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1828 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1829 ntohl(rpl->atid_status)));
1830 struct tid_info *t = dev->rdev.lldi.tids;
1831 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1832 struct sockaddr_in *la;
1833 struct sockaddr_in *ra;
1834 struct sockaddr_in6 *la6;
1835 struct sockaddr_in6 *ra6;
1837 ep = lookup_atid(t, atid);
1838 la = (struct sockaddr_in *)&ep->com.local_addr;
1839 ra = (struct sockaddr_in *)&ep->com.remote_addr;
1840 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1841 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1843 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1844 status, status2errno(status));
1846 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1847 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1852 set_bit(ACT_OPEN_RPL, &ep->com.history);
1855 * Log interesting failures.
1858 case CPL_ERR_CONN_RESET:
1859 case CPL_ERR_CONN_TIMEDOUT:
1861 case CPL_ERR_TCAM_FULL:
1862 mutex_lock(&dev->rdev.stats.lock);
1863 dev->rdev.stats.tcam_full++;
1864 mutex_unlock(&dev->rdev.stats.lock);
1865 if (ep->com.local_addr.ss_family == AF_INET &&
1866 dev->rdev.lldi.enable_fw_ofld_conn) {
1867 send_fw_act_open_req(ep,
1868 GET_TID_TID(GET_AOPEN_ATID(
1869 ntohl(rpl->atid_status))));
1873 case CPL_ERR_CONN_EXIST:
1874 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1875 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1876 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1878 cxgb4_free_atid(t, atid);
1879 dst_release(ep->dst);
1880 cxgb4_l2t_release(ep->l2t);
1886 if (ep->com.local_addr.ss_family == AF_INET) {
1887 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1888 atid, status, status2errno(status),
1889 &la->sin_addr.s_addr, ntohs(la->sin_port),
1890 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
1892 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1893 atid, status, status2errno(status),
1894 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1895 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1900 connect_reply_upcall(ep, status2errno(status));
1901 state_set(&ep->com, DEAD);
1903 if (status && act_open_has_tid(status))
1904 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1906 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1907 cxgb4_free_atid(t, atid);
1908 dst_release(ep->dst);
1909 cxgb4_l2t_release(ep->l2t);
1910 c4iw_put_ep(&ep->com);
1915 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1917 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1918 struct tid_info *t = dev->rdev.lldi.tids;
1919 unsigned int stid = GET_TID(rpl);
1920 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1923 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1926 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1927 rpl->status, status2errno(rpl->status));
1928 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1934 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1936 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1937 struct tid_info *t = dev->rdev.lldi.tids;
1938 unsigned int stid = GET_TID(rpl);
1939 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1941 PDBG("%s ep %p\n", __func__, ep);
1942 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1946 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1947 struct cpl_pass_accept_req *req)
1949 struct cpl_pass_accept_rpl *rpl;
1950 unsigned int mtu_idx;
1955 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1956 BUG_ON(skb_cloned(skb));
1957 skb_trim(skb, sizeof(*rpl));
1959 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1960 wscale = compute_wscale(rcv_win);
1961 opt0 = (nocong ? NO_CONG(1) : 0) |
1966 L2T_IDX(ep->l2t->idx) |
1967 TX_CHAN(ep->tx_chan) |
1968 SMAC_SEL(ep->smac_idx) |
1969 DSCP(ep->tos >> 2) |
1970 ULP_MODE(ULP_MODE_TCPDDP) |
1971 RCV_BUFSIZ(rcv_win>>10);
1972 opt2 = RX_CHANNEL(0) |
1973 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1975 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1976 opt2 |= TSTAMPS_EN(1);
1977 if (enable_tcp_sack && req->tcpopt.sack)
1979 if (wscale && enable_tcp_window_scaling)
1980 opt2 |= WND_SCALE_EN(1);
1982 const struct tcphdr *tcph;
1983 u32 hlen = ntohl(req->hdr_len);
1985 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
1987 if (tcph->ece && tcph->cwr)
1988 opt2 |= CCTRL_ECN(1);
1992 INIT_TP_WR(rpl, ep->hwtid);
1993 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1995 rpl->opt0 = cpu_to_be64(opt0);
1996 rpl->opt2 = cpu_to_be32(opt2);
1997 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1998 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1999 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2004 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2006 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2007 BUG_ON(skb_cloned(skb));
2008 skb_trim(skb, sizeof(struct cpl_tid_release));
2010 release_tid(&dev->rdev, hwtid, skb);
2014 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2015 __u8 *local_ip, __u8 *peer_ip,
2016 __be16 *local_port, __be16 *peer_port)
2018 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2019 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2020 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2021 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2022 struct tcphdr *tcp = (struct tcphdr *)
2023 ((u8 *)(req + 1) + eth_len + ip_len);
2025 if (ip->version == 4) {
2026 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2027 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2030 memcpy(peer_ip, &ip->saddr, 4);
2031 memcpy(local_ip, &ip->daddr, 4);
2033 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2034 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2037 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2038 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2040 *peer_port = tcp->source;
2041 *local_port = tcp->dest;
2046 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2048 struct c4iw_ep *child_ep = NULL, *parent_ep;
2049 struct cpl_pass_accept_req *req = cplhdr(skb);
2050 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2051 struct tid_info *t = dev->rdev.lldi.tids;
2052 unsigned int hwtid = GET_TID(req);
2053 struct dst_entry *dst;
2054 __u8 local_ip[16], peer_ip[16];
2055 __be16 local_port, peer_port;
2057 u16 peer_mss = ntohs(req->tcpopt.mss);
2060 parent_ep = lookup_stid(t, stid);
2062 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2066 if (state_read(&parent_ep->com) != LISTEN) {
2067 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2072 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2074 /* Find output route */
2076 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2077 , __func__, parent_ep, hwtid,
2078 local_ip, peer_ip, ntohs(local_port),
2079 ntohs(peer_port), peer_mss);
2080 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2081 local_port, peer_port,
2082 GET_POPEN_TOS(ntohl(req->tos_stid)));
2084 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2085 , __func__, parent_ep, hwtid,
2086 local_ip, peer_ip, ntohs(local_port),
2087 ntohs(peer_port), peer_mss);
2088 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2089 PASS_OPEN_TOS(ntohl(req->tos_stid)),
2090 ((struct sockaddr_in6 *)
2091 &parent_ep->com.local_addr)->sin6_scope_id);
2094 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2099 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2101 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2107 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2109 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2116 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2117 child_ep->mtu = peer_mss + 40;
2119 state_set(&child_ep->com, CONNECTING);
2120 child_ep->com.dev = dev;
2121 child_ep->com.cm_id = NULL;
2123 struct sockaddr_in *sin = (struct sockaddr_in *)
2124 &child_ep->com.local_addr;
2125 sin->sin_family = PF_INET;
2126 sin->sin_port = local_port;
2127 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2128 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2129 sin->sin_family = PF_INET;
2130 sin->sin_port = peer_port;
2131 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2133 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2134 &child_ep->com.local_addr;
2135 sin6->sin6_family = PF_INET6;
2136 sin6->sin6_port = local_port;
2137 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2138 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2139 sin6->sin6_family = PF_INET6;
2140 sin6->sin6_port = peer_port;
2141 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2143 c4iw_get_ep(&parent_ep->com);
2144 child_ep->parent_ep = parent_ep;
2145 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2146 child_ep->dst = dst;
2147 child_ep->hwtid = hwtid;
2149 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2150 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2152 init_timer(&child_ep->timer);
2153 cxgb4_insert_tid(t, child_ep, hwtid);
2154 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2155 accept_cr(child_ep, skb, req);
2156 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2159 reject_cr(dev, hwtid, skb);
2164 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2167 struct cpl_pass_establish *req = cplhdr(skb);
2168 struct tid_info *t = dev->rdev.lldi.tids;
2169 unsigned int tid = GET_TID(req);
2171 ep = lookup_tid(t, tid);
2172 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2173 ep->snd_seq = be32_to_cpu(req->snd_isn);
2174 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2176 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2177 ntohs(req->tcp_opt));
2179 set_emss(ep, ntohs(req->tcp_opt));
2181 dst_confirm(ep->dst);
2182 state_set(&ep->com, MPA_REQ_WAIT);
2184 send_flowc(ep, skb);
2185 set_bit(PASS_ESTAB, &ep->com.history);
2190 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2192 struct cpl_peer_close *hdr = cplhdr(skb);
2194 struct c4iw_qp_attributes attrs;
2197 struct tid_info *t = dev->rdev.lldi.tids;
2198 unsigned int tid = GET_TID(hdr);
2201 ep = lookup_tid(t, tid);
2202 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2203 dst_confirm(ep->dst);
2205 set_bit(PEER_CLOSE, &ep->com.history);
2206 mutex_lock(&ep->com.mutex);
2207 switch (ep->com.state) {
2209 __state_set(&ep->com, CLOSING);
2212 __state_set(&ep->com, CLOSING);
2213 connect_reply_upcall(ep, -ECONNRESET);
2218 * We're gonna mark this puppy DEAD, but keep
2219 * the reference on it until the ULP accepts or
2220 * rejects the CR. Also wake up anyone waiting
2221 * in rdma connection migration (see c4iw_accept_cr()).
2223 __state_set(&ep->com, CLOSING);
2224 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2225 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2228 __state_set(&ep->com, CLOSING);
2229 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2230 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2234 __state_set(&ep->com, CLOSING);
2235 attrs.next_state = C4IW_QP_STATE_CLOSING;
2236 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2237 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2238 if (ret != -ECONNRESET) {
2239 peer_close_upcall(ep);
2247 __state_set(&ep->com, MORIBUND);
2252 if (ep->com.cm_id && ep->com.qp) {
2253 attrs.next_state = C4IW_QP_STATE_IDLE;
2254 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2255 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2257 close_complete_upcall(ep, 0);
2258 __state_set(&ep->com, DEAD);
2268 mutex_unlock(&ep->com.mutex);
2270 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2272 release_ep_resources(ep);
2277 * Returns whether an ABORT_REQ_RSS message is a negative advice.
2279 static int is_neg_adv_abort(unsigned int status)
2281 return status == CPL_ERR_RTX_NEG_ADVICE ||
2282 status == CPL_ERR_PERSIST_NEG_ADVICE;
2285 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2287 struct cpl_abort_req_rss *req = cplhdr(skb);
2289 struct cpl_abort_rpl *rpl;
2290 struct sk_buff *rpl_skb;
2291 struct c4iw_qp_attributes attrs;
2294 struct tid_info *t = dev->rdev.lldi.tids;
2295 unsigned int tid = GET_TID(req);
2297 ep = lookup_tid(t, tid);
2298 if (is_neg_adv_abort(req->status)) {
2299 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2303 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2305 set_bit(PEER_ABORT, &ep->com.history);
2308 * Wake up any threads in rdma_init() or rdma_fini().
2309 * However, this is not needed if com state is just
2312 if (ep->com.state != MPA_REQ_SENT)
2313 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2315 mutex_lock(&ep->com.mutex);
2316 switch (ep->com.state) {
2324 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2325 connect_reply_upcall(ep, -ECONNRESET);
2328 * we just don't send notification upwards because we
2329 * want to retry with mpa_v1 without upper layers even
2332 * do some housekeeping so as to re-initiate the
2335 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2337 ep->retry_with_mpa_v1 = 1;
2349 if (ep->com.cm_id && ep->com.qp) {
2350 attrs.next_state = C4IW_QP_STATE_ERROR;
2351 ret = c4iw_modify_qp(ep->com.qp->rhp,
2352 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2356 "%s - qp <- error failed!\n",
2359 peer_abort_upcall(ep);
2364 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2365 mutex_unlock(&ep->com.mutex);
2371 dst_confirm(ep->dst);
2372 if (ep->com.state != ABORTING) {
2373 __state_set(&ep->com, DEAD);
2374 /* we don't release if we want to retry with mpa_v1 */
2375 if (!ep->retry_with_mpa_v1)
2378 mutex_unlock(&ep->com.mutex);
2380 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2382 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2387 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2388 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2389 INIT_TP_WR(rpl, ep->hwtid);
2390 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2391 rpl->cmd = CPL_ABORT_NO_RST;
2392 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2395 release_ep_resources(ep);
2396 else if (ep->retry_with_mpa_v1) {
2397 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2398 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2399 dst_release(ep->dst);
2400 cxgb4_l2t_release(ep->l2t);
2407 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2410 struct c4iw_qp_attributes attrs;
2411 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2413 struct tid_info *t = dev->rdev.lldi.tids;
2414 unsigned int tid = GET_TID(rpl);
2416 ep = lookup_tid(t, tid);
2418 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2421 /* The cm_id may be null if we failed to connect */
2422 mutex_lock(&ep->com.mutex);
2423 switch (ep->com.state) {
2425 __state_set(&ep->com, MORIBUND);
2429 if ((ep->com.cm_id) && (ep->com.qp)) {
2430 attrs.next_state = C4IW_QP_STATE_IDLE;
2431 c4iw_modify_qp(ep->com.qp->rhp,
2433 C4IW_QP_ATTR_NEXT_STATE,
2436 close_complete_upcall(ep, 0);
2437 __state_set(&ep->com, DEAD);
2447 mutex_unlock(&ep->com.mutex);
2449 release_ep_resources(ep);
2453 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2455 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2456 struct tid_info *t = dev->rdev.lldi.tids;
2457 unsigned int tid = GET_TID(rpl);
2459 struct c4iw_qp_attributes attrs;
2461 ep = lookup_tid(t, tid);
2464 if (ep && ep->com.qp) {
2465 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2466 ep->com.qp->wq.sq.qid);
2467 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2468 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2469 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2471 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2477 * Upcall from the adapter indicating data has been transmitted.
2478 * For us its just the single MPA request or reply. We can now free
2479 * the skb holding the mpa message.
2481 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2484 struct cpl_fw4_ack *hdr = cplhdr(skb);
2485 u8 credits = hdr->credits;
2486 unsigned int tid = GET_TID(hdr);
2487 struct tid_info *t = dev->rdev.lldi.tids;
2490 ep = lookup_tid(t, tid);
2491 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2493 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2494 __func__, ep, ep->hwtid, state_read(&ep->com));
2498 dst_confirm(ep->dst);
2500 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2501 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2502 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2503 kfree_skb(ep->mpa_skb);
2509 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2512 struct c4iw_ep *ep = to_ep(cm_id);
2513 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2515 if (state_read(&ep->com) == DEAD) {
2516 c4iw_put_ep(&ep->com);
2519 set_bit(ULP_REJECT, &ep->com.history);
2520 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2522 abort_connection(ep, NULL, GFP_KERNEL);
2524 err = send_mpa_reject(ep, pdata, pdata_len);
2525 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2527 c4iw_put_ep(&ep->com);
2531 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2534 struct c4iw_qp_attributes attrs;
2535 enum c4iw_qp_attr_mask mask;
2536 struct c4iw_ep *ep = to_ep(cm_id);
2537 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2538 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2540 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2541 if (state_read(&ep->com) == DEAD) {
2546 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2549 set_bit(ULP_ACCEPT, &ep->com.history);
2550 if ((conn_param->ord > c4iw_max_read_depth) ||
2551 (conn_param->ird > c4iw_max_read_depth)) {
2552 abort_connection(ep, NULL, GFP_KERNEL);
2557 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2558 if (conn_param->ord > ep->ird) {
2559 ep->ird = conn_param->ird;
2560 ep->ord = conn_param->ord;
2561 send_mpa_reject(ep, conn_param->private_data,
2562 conn_param->private_data_len);
2563 abort_connection(ep, NULL, GFP_KERNEL);
2567 if (conn_param->ird > ep->ord) {
2569 conn_param->ird = 1;
2571 abort_connection(ep, NULL, GFP_KERNEL);
2578 ep->ird = conn_param->ird;
2579 ep->ord = conn_param->ord;
2581 if (ep->mpa_attr.version != 2)
2582 if (peer2peer && ep->ird == 0)
2585 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2587 cm_id->add_ref(cm_id);
2588 ep->com.cm_id = cm_id;
2592 /* bind QP to EP and move to RTS */
2593 attrs.mpa_attr = ep->mpa_attr;
2594 attrs.max_ird = ep->ird;
2595 attrs.max_ord = ep->ord;
2596 attrs.llp_stream_handle = ep;
2597 attrs.next_state = C4IW_QP_STATE_RTS;
2599 /* bind QP and TID with INIT_WR */
2600 mask = C4IW_QP_ATTR_NEXT_STATE |
2601 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2602 C4IW_QP_ATTR_MPA_ATTR |
2603 C4IW_QP_ATTR_MAX_IRD |
2604 C4IW_QP_ATTR_MAX_ORD;
2606 err = c4iw_modify_qp(ep->com.qp->rhp,
2607 ep->com.qp, mask, &attrs, 1);
2610 err = send_mpa_reply(ep, conn_param->private_data,
2611 conn_param->private_data_len);
2615 state_set(&ep->com, FPDU_MODE);
2616 established_upcall(ep);
2617 c4iw_put_ep(&ep->com);
2620 ep->com.cm_id = NULL;
2621 cm_id->rem_ref(cm_id);
2623 c4iw_put_ep(&ep->com);
2627 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2629 struct in_device *ind;
2631 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2632 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2634 ind = in_dev_get(dev->rdev.lldi.ports[0]);
2636 return -EADDRNOTAVAIL;
2637 for_primary_ifa(ind) {
2638 laddr->sin_addr.s_addr = ifa->ifa_address;
2639 raddr->sin_addr.s_addr = ifa->ifa_address;
2645 return found ? 0 : -EADDRNOTAVAIL;
2648 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2649 unsigned char banned_flags)
2651 struct inet6_dev *idev;
2652 int err = -EADDRNOTAVAIL;
2655 idev = __in6_dev_get(dev);
2657 struct inet6_ifaddr *ifp;
2659 read_lock_bh(&idev->lock);
2660 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2661 if (ifp->scope == IFA_LINK &&
2662 !(ifp->flags & banned_flags)) {
2663 memcpy(addr, &ifp->addr, 16);
2668 read_unlock_bh(&idev->lock);
2674 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2676 struct in6_addr uninitialized_var(addr);
2677 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2678 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2680 if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2681 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2682 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2685 return -EADDRNOTAVAIL;
2688 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2690 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2693 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2694 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2695 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2696 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2697 &cm_id->remote_addr;
2701 if ((conn_param->ord > c4iw_max_read_depth) ||
2702 (conn_param->ird > c4iw_max_read_depth)) {
2706 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2708 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2712 init_timer(&ep->timer);
2713 ep->plen = conn_param->private_data_len;
2715 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2716 conn_param->private_data, ep->plen);
2717 ep->ird = conn_param->ird;
2718 ep->ord = conn_param->ord;
2720 if (peer2peer && ep->ord == 0)
2723 cm_id->add_ref(cm_id);
2725 ep->com.cm_id = cm_id;
2726 ep->com.qp = get_qhp(dev, conn_param->qpn);
2728 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2733 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2737 * Allocate an active TID to initiate a TCP connection.
2739 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2740 if (ep->atid == -1) {
2741 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2745 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2747 if (cm_id->remote_addr.ss_family == AF_INET) {
2749 ra = (__u8 *)&raddr->sin_addr;
2752 * Handle loopback requests to INADDR_ANY.
2754 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2755 err = pick_local_ipaddrs(dev, cm_id);
2761 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2762 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2763 ra, ntohs(raddr->sin_port));
2764 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2765 raddr->sin_addr.s_addr, laddr->sin_port,
2766 raddr->sin_port, 0);
2769 ra = (__u8 *)&raddr6->sin6_addr;
2772 * Handle loopback requests to INADDR_ANY.
2774 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2775 err = pick_local_ip6addrs(dev, cm_id);
2781 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2782 __func__, laddr6->sin6_addr.s6_addr,
2783 ntohs(laddr6->sin6_port),
2784 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2785 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2786 raddr6->sin6_addr.s6_addr,
2787 laddr6->sin6_port, raddr6->sin6_port, 0,
2788 raddr6->sin6_scope_id);
2791 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2792 err = -EHOSTUNREACH;
2796 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2798 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2802 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2803 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2806 state_set(&ep->com, CONNECTING);
2808 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2809 sizeof(ep->com.local_addr));
2810 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2811 sizeof(ep->com.remote_addr));
2813 /* send connect request to rnic */
2814 err = send_connect(ep);
2818 cxgb4_l2t_release(ep->l2t);
2820 dst_release(ep->dst);
2822 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2823 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2825 cm_id->rem_ref(cm_id);
2826 c4iw_put_ep(&ep->com);
2831 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2834 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2836 c4iw_init_wr_wait(&ep->com.wr_wait);
2837 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2838 ep->stid, &sin6->sin6_addr,
2840 ep->com.dev->rdev.lldi.rxq_ids[0]);
2842 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2846 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2848 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2852 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2855 struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2857 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2859 err = cxgb4_create_server_filter(
2860 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2861 sin->sin_addr.s_addr, sin->sin_port, 0,
2862 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2863 if (err == -EBUSY) {
2864 set_current_state(TASK_UNINTERRUPTIBLE);
2865 schedule_timeout(usecs_to_jiffies(100));
2867 } while (err == -EBUSY);
2869 c4iw_init_wr_wait(&ep->com.wr_wait);
2870 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2871 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2872 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2874 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2879 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2881 &sin->sin_addr, ntohs(sin->sin_port));
2885 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2888 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2889 struct c4iw_listen_ep *ep;
2893 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2895 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2899 PDBG("%s ep %p\n", __func__, ep);
2900 cm_id->add_ref(cm_id);
2901 ep->com.cm_id = cm_id;
2903 ep->backlog = backlog;
2904 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2905 sizeof(ep->com.local_addr));
2908 * Allocate a server TID.
2910 if (dev->rdev.lldi.enable_fw_ofld_conn &&
2911 ep->com.local_addr.ss_family == AF_INET)
2912 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2913 cm_id->local_addr.ss_family, ep);
2915 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2916 cm_id->local_addr.ss_family, ep);
2918 if (ep->stid == -1) {
2919 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2923 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2924 state_set(&ep->com, LISTEN);
2925 if (ep->com.local_addr.ss_family == AF_INET)
2926 err = create_server4(dev, ep);
2928 err = create_server6(dev, ep);
2930 cm_id->provider_data = ep;
2933 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2934 ep->com.local_addr.ss_family);
2936 cm_id->rem_ref(cm_id);
2937 c4iw_put_ep(&ep->com);
2943 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2946 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2948 PDBG("%s ep %p\n", __func__, ep);
2951 state_set(&ep->com, DEAD);
2952 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2953 ep->com.local_addr.ss_family == AF_INET) {
2954 err = cxgb4_remove_server_filter(
2955 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2956 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2958 c4iw_init_wr_wait(&ep->com.wr_wait);
2959 err = cxgb4_remove_server(
2960 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2961 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2964 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2967 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2968 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2969 ep->com.local_addr.ss_family);
2971 cm_id->rem_ref(cm_id);
2972 c4iw_put_ep(&ep->com);
2976 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2981 struct c4iw_rdev *rdev;
2983 mutex_lock(&ep->com.mutex);
2985 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2986 states[ep->com.state], abrupt);
2988 rdev = &ep->com.dev->rdev;
2989 if (c4iw_fatal_error(rdev)) {
2991 close_complete_upcall(ep, -EIO);
2992 ep->com.state = DEAD;
2994 switch (ep->com.state) {
3002 ep->com.state = ABORTING;
3004 ep->com.state = CLOSING;
3007 set_bit(CLOSE_SENT, &ep->com.flags);
3010 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3014 ep->com.state = ABORTING;
3016 ep->com.state = MORIBUND;
3022 PDBG("%s ignoring disconnect ep %p state %u\n",
3023 __func__, ep, ep->com.state);
3032 set_bit(EP_DISC_ABORT, &ep->com.history);
3033 close_complete_upcall(ep, -ECONNRESET);
3034 ret = send_abort(ep, NULL, gfp);
3036 set_bit(EP_DISC_CLOSE, &ep->com.history);
3037 ret = send_halfclose(ep, gfp);
3042 mutex_unlock(&ep->com.mutex);
3044 release_ep_resources(ep);
3048 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3049 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3052 int atid = be32_to_cpu(req->tid);
3054 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3055 (__force u32) req->tid);
3059 switch (req->retval) {
3061 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3062 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3063 send_fw_act_open_req(ep, atid);
3067 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3068 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3069 send_fw_act_open_req(ep, atid);
3074 pr_info("%s unexpected ofld conn wr retval %d\n",
3075 __func__, req->retval);
3078 pr_err("active ofld_connect_wr failure %d atid %d\n",
3080 mutex_lock(&dev->rdev.stats.lock);
3081 dev->rdev.stats.act_ofld_conn_fails++;
3082 mutex_unlock(&dev->rdev.stats.lock);
3083 connect_reply_upcall(ep, status2errno(req->retval));
3084 state_set(&ep->com, DEAD);
3085 remove_handle(dev, &dev->atid_idr, atid);
3086 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3087 dst_release(ep->dst);
3088 cxgb4_l2t_release(ep->l2t);
3089 c4iw_put_ep(&ep->com);
3092 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3093 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3095 struct sk_buff *rpl_skb;
3096 struct cpl_pass_accept_req *cpl;
3099 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3102 PDBG("%s passive open failure %d\n", __func__, req->retval);
3103 mutex_lock(&dev->rdev.stats.lock);
3104 dev->rdev.stats.pas_ofld_conn_fails++;
3105 mutex_unlock(&dev->rdev.stats.lock);
3108 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3109 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3110 (__force u32) htonl(
3111 (__force u32) req->tid)));
3112 ret = pass_accept_req(dev, rpl_skb);
3119 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3121 struct cpl_fw6_msg *rpl = cplhdr(skb);
3122 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3124 switch (rpl->type) {
3126 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3128 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3129 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3130 switch (req->t_state) {
3132 active_ofld_conn_reply(dev, skb, req);
3135 passive_ofld_conn_reply(dev, skb, req);
3138 pr_err("%s unexpected ofld conn wr state %d\n",
3139 __func__, req->t_state);
3147 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3150 u16 vlantag, len, hdr_len, eth_hdr_len;
3152 struct cpl_rx_pkt *cpl = cplhdr(skb);
3153 struct cpl_pass_accept_req *req;
3154 struct tcp_options_received tmp_opt;
3155 struct c4iw_dev *dev;
3157 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3158 /* Store values from cpl_rx_pkt in temporary location. */
3159 vlantag = (__force u16) cpl->vlan;
3160 len = (__force u16) cpl->len;
3161 l2info = (__force u32) cpl->l2info;
3162 hdr_len = (__force u16) cpl->hdr_len;
3165 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3168 * We need to parse the TCP options from SYN packet.
3169 * to generate cpl_pass_accept_req.
3171 memset(&tmp_opt, 0, sizeof(tmp_opt));
3172 tcp_clear_options(&tmp_opt);
3173 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3175 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3176 memset(req, 0, sizeof(*req));
3177 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3178 V_SYN_MAC_IDX(G_RX_MACIDX(
3179 (__force int) htonl(l2info))) |
3181 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3182 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3183 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3184 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3185 (__force int) htonl(l2info))) |
3186 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3187 (__force int) htons(hdr_len))) |
3188 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3189 (__force int) htons(hdr_len))) |
3190 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3191 req->vlan = (__force __be16) vlantag;
3192 req->len = (__force __be16) len;
3193 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3194 PASS_OPEN_TOS(tos));
3195 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3196 if (tmp_opt.wscale_ok)
3197 req->tcpopt.wsf = tmp_opt.snd_wscale;
3198 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3199 if (tmp_opt.sack_ok)
3200 req->tcpopt.sack = 1;
3201 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3205 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3206 __be32 laddr, __be16 lport,
3207 __be32 raddr, __be16 rport,
3208 u32 rcv_isn, u32 filter, u16 window,
3209 u32 rss_qid, u8 port_id)
3211 struct sk_buff *req_skb;
3212 struct fw_ofld_connection_wr *req;
3213 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3216 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3217 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3218 memset(req, 0, sizeof(*req));
3219 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3220 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3221 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3222 req->le.filter = (__force __be32) filter;
3223 req->le.lport = lport;
3224 req->le.pport = rport;
3225 req->le.u.ipv4.lip = laddr;
3226 req->le.u.ipv4.pip = raddr;
3227 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3228 req->tcb.rcv_adv = htons(window);
3229 req->tcb.t_state_to_astid =
3230 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3231 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3232 V_FW_OFLD_CONNECTION_WR_ASTID(
3233 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3236 * We store the qid in opt2 which will be used by the firmware
3237 * to send us the wr response.
3239 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3242 * We initialize the MSS index in TCB to 0xF.
3243 * So that when driver sends cpl_pass_accept_rpl
3244 * TCB picks up the correct value. If this was 0
3245 * TP will ignore any value > 0 for MSS index.
3247 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3248 req->cookie = (unsigned long)skb;
3250 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3251 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3253 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3261 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3262 * messages when a filter is being used instead of server to
3263 * redirect a syn packet. When packets hit filter they are redirected
3264 * to the offload queue and driver tries to establish the connection
3265 * using firmware work request.
3267 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3270 unsigned int filter;
3271 struct ethhdr *eh = NULL;
3272 struct vlan_ethhdr *vlan_eh = NULL;
3274 struct tcphdr *tcph;
3275 struct rss_header *rss = (void *)skb->data;
3276 struct cpl_rx_pkt *cpl = (void *)skb->data;
3277 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3278 struct l2t_entry *e;
3279 struct dst_entry *dst;
3280 struct c4iw_ep *lep;
3282 struct port_info *pi;
3283 struct net_device *pdev;
3284 u16 rss_qid, eth_hdr_len;
3287 struct neighbour *neigh;
3289 /* Drop all non-SYN packets */
3290 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3294 * Drop all packets which did not hit the filter.
3295 * Unlikely to happen.
3297 if (!(rss->filter_hit && rss->filter_tid))
3301 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3303 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3305 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3307 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3311 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3312 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3313 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3314 if (eth_hdr_len == ETH_HLEN) {
3315 eh = (struct ethhdr *)(req + 1);
3316 iph = (struct iphdr *)(eh + 1);
3318 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3319 iph = (struct iphdr *)(vlan_eh + 1);
3320 skb->vlan_tci = ntohs(cpl->vlan);
3323 if (iph->version != 0x4)
3326 tcph = (struct tcphdr *)(iph + 1);
3327 skb_set_network_header(skb, (void *)iph - (void *)rss);
3328 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3331 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3332 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3333 ntohs(tcph->source), iph->tos);
3335 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3338 pr_err("%s - failed to find dst entry!\n",
3342 neigh = dst_neigh_lookup_skb(dst, skb);
3345 pr_err("%s - failed to allocate neigh!\n",
3350 if (neigh->dev->flags & IFF_LOOPBACK) {
3351 pdev = ip_dev_find(&init_net, iph->daddr);
3352 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3354 pi = (struct port_info *)netdev_priv(pdev);
3355 tx_chan = cxgb4_port_chan(pdev);
3358 pdev = get_real_dev(neigh->dev);
3359 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3361 pi = (struct port_info *)netdev_priv(pdev);
3362 tx_chan = cxgb4_port_chan(pdev);
3364 neigh_release(neigh);
3366 pr_err("%s - failed to allocate l2t entry!\n",
3371 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3372 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3373 window = (__force u16) htons((__force u16)tcph->window);
3375 /* Calcuate filter portion for LE region. */
3376 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3377 dev->rdev.lldi.ports[0],
3381 * Synthesize the cpl_pass_accept_req. We have everything except the
3382 * TID. Once firmware sends a reply with TID we update the TID field
3383 * in cpl and pass it through the regular cpl_pass_accept_req path.
3385 build_cpl_pass_accept_req(skb, stid, iph->tos);
3386 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3387 tcph->source, ntohl(tcph->seq), filter, window,
3388 rss_qid, pi->port_id);
3389 cxgb4_l2t_release(e);
3397 * These are the real handlers that are called from a
3400 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3401 [CPL_ACT_ESTABLISH] = act_establish,
3402 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3403 [CPL_RX_DATA] = rx_data,
3404 [CPL_ABORT_RPL_RSS] = abort_rpl,
3405 [CPL_ABORT_RPL] = abort_rpl,
3406 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3407 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3408 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3409 [CPL_PASS_ESTABLISH] = pass_establish,
3410 [CPL_PEER_CLOSE] = peer_close,
3411 [CPL_ABORT_REQ_RSS] = peer_abort,
3412 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3413 [CPL_RDMA_TERMINATE] = terminate,
3414 [CPL_FW4_ACK] = fw4_ack,
3415 [CPL_FW6_MSG] = deferred_fw6_msg,
3416 [CPL_RX_PKT] = rx_pkt
3419 static void process_timeout(struct c4iw_ep *ep)
3421 struct c4iw_qp_attributes attrs;
3424 mutex_lock(&ep->com.mutex);
3425 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3427 set_bit(TIMEDOUT, &ep->com.history);
3428 switch (ep->com.state) {
3430 __state_set(&ep->com, ABORTING);
3431 connect_reply_upcall(ep, -ETIMEDOUT);
3434 __state_set(&ep->com, ABORTING);
3438 if (ep->com.cm_id && ep->com.qp) {
3439 attrs.next_state = C4IW_QP_STATE_ERROR;
3440 c4iw_modify_qp(ep->com.qp->rhp,
3441 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3444 __state_set(&ep->com, ABORTING);
3445 close_complete_upcall(ep, -ETIMEDOUT);
3448 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3449 __func__, ep, ep->hwtid, ep->com.state);
3452 mutex_unlock(&ep->com.mutex);
3454 abort_connection(ep, NULL, GFP_KERNEL);
3455 c4iw_put_ep(&ep->com);
3458 static void process_timedout_eps(void)
3462 spin_lock_irq(&timeout_lock);
3463 while (!list_empty(&timeout_list)) {
3464 struct list_head *tmp;
3466 tmp = timeout_list.next;
3468 spin_unlock_irq(&timeout_lock);
3469 ep = list_entry(tmp, struct c4iw_ep, entry);
3470 process_timeout(ep);
3471 spin_lock_irq(&timeout_lock);
3473 spin_unlock_irq(&timeout_lock);
3476 static void process_work(struct work_struct *work)
3478 struct sk_buff *skb = NULL;
3479 struct c4iw_dev *dev;
3480 struct cpl_act_establish *rpl;
3481 unsigned int opcode;
3484 while ((skb = skb_dequeue(&rxq))) {
3486 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3487 opcode = rpl->ot.opcode;
3489 BUG_ON(!work_handlers[opcode]);
3490 ret = work_handlers[opcode](dev, skb);
3494 process_timedout_eps();
3497 static DECLARE_WORK(skb_work, process_work);
3499 static void ep_timeout(unsigned long arg)
3501 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3504 spin_lock(&timeout_lock);
3505 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3506 list_add_tail(&ep->entry, &timeout_list);
3509 spin_unlock(&timeout_lock);
3511 queue_work(workq, &skb_work);
3515 * All the CM events are handled on a work queue to have a safe context.
3517 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3521 * Save dev in the skb->cb area.
3523 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3526 * Queue the skb and schedule the worker thread.
3528 skb_queue_tail(&rxq, skb);
3529 queue_work(workq, &skb_work);
3533 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3535 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3537 if (rpl->status != CPL_ERR_NONE) {
3538 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3539 "for tid %u\n", rpl->status, GET_TID(rpl));
3545 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3547 struct cpl_fw6_msg *rpl = cplhdr(skb);
3548 struct c4iw_wr_wait *wr_waitp;
3551 PDBG("%s type %u\n", __func__, rpl->type);
3553 switch (rpl->type) {
3554 case FW6_TYPE_WR_RPL:
3555 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3556 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3557 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3559 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3563 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3567 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3575 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3577 struct cpl_abort_req_rss *req = cplhdr(skb);
3579 struct tid_info *t = dev->rdev.lldi.tids;
3580 unsigned int tid = GET_TID(req);
3582 ep = lookup_tid(t, tid);
3584 printk(KERN_WARNING MOD
3585 "Abort on non-existent endpoint, tid %d\n", tid);
3589 if (is_neg_adv_abort(req->status)) {
3590 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3595 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3599 * Wake up any threads in rdma_init() or rdma_fini().
3600 * However, if we are on MPAv2 and want to retry with MPAv1
3601 * then, don't wake up yet.
3603 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3604 if (ep->com.state != MPA_REQ_SENT)
3605 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3607 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3613 * Most upcalls from the T4 Core go to sched() to
3614 * schedule the processing on a work queue.
3616 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3617 [CPL_ACT_ESTABLISH] = sched,
3618 [CPL_ACT_OPEN_RPL] = sched,
3619 [CPL_RX_DATA] = sched,
3620 [CPL_ABORT_RPL_RSS] = sched,
3621 [CPL_ABORT_RPL] = sched,
3622 [CPL_PASS_OPEN_RPL] = sched,
3623 [CPL_CLOSE_LISTSRV_RPL] = sched,
3624 [CPL_PASS_ACCEPT_REQ] = sched,
3625 [CPL_PASS_ESTABLISH] = sched,
3626 [CPL_PEER_CLOSE] = sched,
3627 [CPL_CLOSE_CON_RPL] = sched,
3628 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3629 [CPL_RDMA_TERMINATE] = sched,
3630 [CPL_FW4_ACK] = sched,
3631 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3632 [CPL_FW6_MSG] = fw6_msg,
3633 [CPL_RX_PKT] = sched
3636 int __init c4iw_cm_init(void)
3638 spin_lock_init(&timeout_lock);
3639 skb_queue_head_init(&rxq);
3641 workq = create_singlethread_workqueue("iw_cxgb4");
3648 void __exit c4iw_cm_term(void)
3650 WARN_ON(!list_empty(&timeout_list));
3651 flush_workqueue(workq);
3652 destroy_workqueue(workq);