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 int 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);
187 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
188 struct l2t_entry *l2e)
192 if (c4iw_fatal_error(rdev)) {
194 PDBG("%s - device in error state - dropping\n", __func__);
197 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
200 return error < 0 ? error : 0;
203 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
207 if (c4iw_fatal_error(rdev)) {
209 PDBG("%s - device in error state - dropping\n", __func__);
212 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
215 return error < 0 ? error : 0;
218 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
220 struct cpl_tid_release *req;
222 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
225 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
226 INIT_TP_WR(req, hwtid);
227 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
228 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
229 c4iw_ofld_send(rdev, skb);
233 static void set_emss(struct c4iw_ep *ep, u16 opt)
235 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
237 if (GET_TCPOPT_TSTAMP(opt))
241 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
245 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
247 enum c4iw_ep_state state;
249 mutex_lock(&epc->mutex);
251 mutex_unlock(&epc->mutex);
255 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
260 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
262 mutex_lock(&epc->mutex);
263 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
264 __state_set(epc, new);
265 mutex_unlock(&epc->mutex);
269 static void *alloc_ep(int size, gfp_t gfp)
271 struct c4iw_ep_common *epc;
273 epc = kzalloc(size, gfp);
275 kref_init(&epc->kref);
276 mutex_init(&epc->mutex);
277 c4iw_init_wr_wait(&epc->wr_wait);
279 PDBG("%s alloc ep %p\n", __func__, epc);
283 void _c4iw_free_ep(struct kref *kref)
287 ep = container_of(kref, struct c4iw_ep, com.kref);
288 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
289 if (test_bit(QP_REFERENCED, &ep->com.flags))
291 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
292 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
293 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
294 dst_release(ep->dst);
295 cxgb4_l2t_release(ep->l2t);
300 static void release_ep_resources(struct c4iw_ep *ep)
302 set_bit(RELEASE_RESOURCES, &ep->com.flags);
303 c4iw_put_ep(&ep->com);
306 static int status2errno(int status)
311 case CPL_ERR_CONN_RESET:
313 case CPL_ERR_ARP_MISS:
314 return -EHOSTUNREACH;
315 case CPL_ERR_CONN_TIMEDOUT:
317 case CPL_ERR_TCAM_FULL:
319 case CPL_ERR_CONN_EXIST:
327 * Try and reuse skbs already allocated...
329 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
331 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
334 skb_reset_transport_header(skb);
336 skb = alloc_skb(len, gfp);
338 t4_set_arp_err_handler(skb, NULL, NULL);
342 static struct net_device *get_real_dev(struct net_device *egress_dev)
344 struct net_device *phys_dev = egress_dev;
345 if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
346 phys_dev = vlan_dev_real_dev(egress_dev);
350 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
354 egress_dev = get_real_dev(egress_dev);
355 for (i = 0; i < dev->rdev.lldi.nports; i++)
356 if (dev->rdev.lldi.ports[i] == egress_dev)
361 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
362 __u8 *peer_ip, __be16 local_port,
363 __be16 peer_port, u8 tos,
366 struct dst_entry *dst = NULL;
368 if (IS_ENABLED(CONFIG_IPV6)) {
371 memset(&fl6, 0, sizeof(fl6));
372 memcpy(&fl6.daddr, peer_ip, 16);
373 memcpy(&fl6.saddr, local_ip, 16);
374 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
375 fl6.flowi6_oif = sin6_scope_id;
376 dst = ip6_route_output(&init_net, NULL, &fl6);
379 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
380 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
390 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
391 __be32 peer_ip, __be16 local_port,
392 __be16 peer_port, u8 tos)
398 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
399 peer_port, local_port, IPPROTO_TCP,
403 n = dst_neigh_lookup(&rt->dst, &peer_ip);
406 if (!our_interface(dev, n->dev) &&
407 !(n->dev->flags & IFF_LOOPBACK)) {
408 dst_release(&rt->dst);
415 static void arp_failure_discard(void *handle, struct sk_buff *skb)
417 PDBG("%s c4iw_dev %p\n", __func__, handle);
422 * Handle an ARP failure for an active open.
424 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
426 printk(KERN_ERR MOD "ARP failure duing connect\n");
431 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
434 static void abort_arp_failure(void *handle, struct sk_buff *skb)
436 struct c4iw_rdev *rdev = handle;
437 struct cpl_abort_req *req = cplhdr(skb);
439 PDBG("%s rdev %p\n", __func__, rdev);
440 req->cmd = CPL_ABORT_NO_RST;
441 c4iw_ofld_send(rdev, skb);
444 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
446 unsigned int flowclen = 80;
447 struct fw_flowc_wr *flowc;
450 skb = get_skb(skb, flowclen, GFP_KERNEL);
451 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
453 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
454 FW_FLOWC_WR_NPARAMS(8));
455 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
456 16)) | FW_WR_FLOWID(ep->hwtid));
458 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
459 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
460 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
461 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
462 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
463 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
464 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
465 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
466 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
467 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
468 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
469 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
470 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
471 flowc->mnemval[6].val = cpu_to_be32(snd_win);
472 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
473 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
474 /* Pad WR to 16 byte boundary */
475 flowc->mnemval[8].mnemonic = 0;
476 flowc->mnemval[8].val = 0;
477 for (i = 0; i < 9; i++) {
478 flowc->mnemval[i].r4[0] = 0;
479 flowc->mnemval[i].r4[1] = 0;
480 flowc->mnemval[i].r4[2] = 0;
483 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
484 c4iw_ofld_send(&ep->com.dev->rdev, skb);
487 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
489 struct cpl_close_con_req *req;
491 int wrlen = roundup(sizeof *req, 16);
493 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
494 skb = get_skb(NULL, wrlen, gfp);
496 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
499 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
500 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
501 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
502 memset(req, 0, wrlen);
503 INIT_TP_WR(req, ep->hwtid);
504 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
506 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
509 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
511 struct cpl_abort_req *req;
512 int wrlen = roundup(sizeof *req, 16);
514 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
515 skb = get_skb(skb, wrlen, gfp);
517 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
521 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
522 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
523 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
524 memset(req, 0, wrlen);
525 INIT_TP_WR(req, ep->hwtid);
526 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
527 req->cmd = CPL_ABORT_SEND_RST;
528 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
531 static int send_connect(struct c4iw_ep *ep)
533 struct cpl_act_open_req *req;
534 struct cpl_t5_act_open_req *t5_req;
535 struct cpl_act_open_req6 *req6;
536 struct cpl_t5_act_open_req6 *t5_req6;
540 unsigned int mtu_idx;
543 int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
544 sizeof(struct cpl_act_open_req) :
545 sizeof(struct cpl_t5_act_open_req);
546 int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
547 sizeof(struct cpl_act_open_req6) :
548 sizeof(struct cpl_t5_act_open_req6);
549 struct sockaddr_in *la = (struct sockaddr_in *)&ep->com.local_addr;
550 struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
551 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
552 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
554 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
555 roundup(sizev4, 16) :
558 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
560 skb = get_skb(NULL, wrlen, GFP_KERNEL);
562 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
566 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
568 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
569 wscale = compute_wscale(rcv_win);
570 opt0 = (nocong ? NO_CONG(1) : 0) |
575 L2T_IDX(ep->l2t->idx) |
576 TX_CHAN(ep->tx_chan) |
577 SMAC_SEL(ep->smac_idx) |
579 ULP_MODE(ULP_MODE_TCPDDP) |
580 RCV_BUFSIZ(rcv_win>>10);
581 opt2 = RX_CHANNEL(0) |
582 CCTRL_ECN(enable_ecn) |
583 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
584 if (enable_tcp_timestamps)
585 opt2 |= TSTAMPS_EN(1);
588 if (wscale && enable_tcp_window_scaling)
589 opt2 |= WND_SCALE_EN(1);
590 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
592 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
593 if (ep->com.remote_addr.ss_family == AF_INET) {
594 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
596 OPCODE_TID(req) = cpu_to_be32(
597 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
598 ((ep->rss_qid << 14) | ep->atid)));
599 req->local_port = la->sin_port;
600 req->peer_port = ra->sin_port;
601 req->local_ip = la->sin_addr.s_addr;
602 req->peer_ip = ra->sin_addr.s_addr;
603 req->opt0 = cpu_to_be64(opt0);
604 req->params = cpu_to_be32(cxgb4_select_ntuple(
605 ep->com.dev->rdev.lldi.ports[0],
607 req->opt2 = cpu_to_be32(opt2);
609 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
612 OPCODE_TID(req6) = cpu_to_be32(
613 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
614 ((ep->rss_qid<<14)|ep->atid)));
615 req6->local_port = la6->sin6_port;
616 req6->peer_port = ra6->sin6_port;
617 req6->local_ip_hi = *((__be64 *)
618 (la6->sin6_addr.s6_addr));
619 req6->local_ip_lo = *((__be64 *)
620 (la6->sin6_addr.s6_addr + 8));
621 req6->peer_ip_hi = *((__be64 *)
622 (ra6->sin6_addr.s6_addr));
623 req6->peer_ip_lo = *((__be64 *)
624 (ra6->sin6_addr.s6_addr + 8));
625 req6->opt0 = cpu_to_be64(opt0);
626 req6->params = cpu_to_be32(cxgb4_select_ntuple(
627 ep->com.dev->rdev.lldi.ports[0],
629 req6->opt2 = cpu_to_be32(opt2);
632 if (ep->com.remote_addr.ss_family == AF_INET) {
633 t5_req = (struct cpl_t5_act_open_req *)
635 INIT_TP_WR(t5_req, 0);
636 OPCODE_TID(t5_req) = cpu_to_be32(
637 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
638 ((ep->rss_qid << 14) | ep->atid)));
639 t5_req->local_port = la->sin_port;
640 t5_req->peer_port = ra->sin_port;
641 t5_req->local_ip = la->sin_addr.s_addr;
642 t5_req->peer_ip = ra->sin_addr.s_addr;
643 t5_req->opt0 = cpu_to_be64(opt0);
644 t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
646 ep->com.dev->rdev.lldi.ports[0],
648 t5_req->opt2 = cpu_to_be32(opt2);
650 t5_req6 = (struct cpl_t5_act_open_req6 *)
652 INIT_TP_WR(t5_req6, 0);
653 OPCODE_TID(t5_req6) = cpu_to_be32(
654 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
655 ((ep->rss_qid<<14)|ep->atid)));
656 t5_req6->local_port = la6->sin6_port;
657 t5_req6->peer_port = ra6->sin6_port;
658 t5_req6->local_ip_hi = *((__be64 *)
659 (la6->sin6_addr.s6_addr));
660 t5_req6->local_ip_lo = *((__be64 *)
661 (la6->sin6_addr.s6_addr + 8));
662 t5_req6->peer_ip_hi = *((__be64 *)
663 (ra6->sin6_addr.s6_addr));
664 t5_req6->peer_ip_lo = *((__be64 *)
665 (ra6->sin6_addr.s6_addr + 8));
666 t5_req6->opt0 = cpu_to_be64(opt0);
667 t5_req6->params = (__force __be64)cpu_to_be32(
669 ep->com.dev->rdev.lldi.ports[0],
671 t5_req6->opt2 = cpu_to_be32(opt2);
675 set_bit(ACT_OPEN_REQ, &ep->com.history);
676 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
679 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
683 struct fw_ofld_tx_data_wr *req;
684 struct mpa_message *mpa;
685 struct mpa_v2_conn_params mpa_v2_params;
687 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
689 BUG_ON(skb_cloned(skb));
691 mpalen = sizeof(*mpa) + ep->plen;
692 if (mpa_rev_to_use == 2)
693 mpalen += sizeof(struct mpa_v2_conn_params);
694 wrlen = roundup(mpalen + sizeof *req, 16);
695 skb = get_skb(skb, wrlen, GFP_KERNEL);
697 connect_reply_upcall(ep, -ENOMEM);
700 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
702 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
703 memset(req, 0, wrlen);
704 req->op_to_immdlen = cpu_to_be32(
705 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
707 FW_WR_IMMDLEN(mpalen));
708 req->flowid_len16 = cpu_to_be32(
709 FW_WR_FLOWID(ep->hwtid) |
710 FW_WR_LEN16(wrlen >> 4));
711 req->plen = cpu_to_be32(mpalen);
712 req->tunnel_to_proxy = cpu_to_be32(
713 FW_OFLD_TX_DATA_WR_FLUSH(1) |
714 FW_OFLD_TX_DATA_WR_SHOVE(1));
716 mpa = (struct mpa_message *)(req + 1);
717 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
718 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
719 (markers_enabled ? MPA_MARKERS : 0) |
720 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
721 mpa->private_data_size = htons(ep->plen);
722 mpa->revision = mpa_rev_to_use;
723 if (mpa_rev_to_use == 1) {
724 ep->tried_with_mpa_v1 = 1;
725 ep->retry_with_mpa_v1 = 0;
728 if (mpa_rev_to_use == 2) {
729 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
730 sizeof (struct mpa_v2_conn_params));
731 mpa_v2_params.ird = htons((u16)ep->ird);
732 mpa_v2_params.ord = htons((u16)ep->ord);
735 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
736 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
738 htons(MPA_V2_RDMA_WRITE_RTR);
739 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
741 htons(MPA_V2_RDMA_READ_RTR);
743 memcpy(mpa->private_data, &mpa_v2_params,
744 sizeof(struct mpa_v2_conn_params));
747 memcpy(mpa->private_data +
748 sizeof(struct mpa_v2_conn_params),
749 ep->mpa_pkt + sizeof(*mpa), ep->plen);
752 memcpy(mpa->private_data,
753 ep->mpa_pkt + sizeof(*mpa), ep->plen);
756 * Reference the mpa skb. This ensures the data area
757 * will remain in memory until the hw acks the tx.
758 * Function fw4_ack() will deref it.
761 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
764 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
766 __state_set(&ep->com, MPA_REQ_SENT);
767 ep->mpa_attr.initiator = 1;
768 ep->snd_seq += mpalen;
772 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
775 struct fw_ofld_tx_data_wr *req;
776 struct mpa_message *mpa;
778 struct mpa_v2_conn_params mpa_v2_params;
780 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
782 mpalen = sizeof(*mpa) + plen;
783 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
784 mpalen += sizeof(struct mpa_v2_conn_params);
785 wrlen = roundup(mpalen + sizeof *req, 16);
787 skb = get_skb(NULL, wrlen, GFP_KERNEL);
789 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
792 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
794 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
795 memset(req, 0, wrlen);
796 req->op_to_immdlen = cpu_to_be32(
797 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
799 FW_WR_IMMDLEN(mpalen));
800 req->flowid_len16 = cpu_to_be32(
801 FW_WR_FLOWID(ep->hwtid) |
802 FW_WR_LEN16(wrlen >> 4));
803 req->plen = cpu_to_be32(mpalen);
804 req->tunnel_to_proxy = cpu_to_be32(
805 FW_OFLD_TX_DATA_WR_FLUSH(1) |
806 FW_OFLD_TX_DATA_WR_SHOVE(1));
808 mpa = (struct mpa_message *)(req + 1);
809 memset(mpa, 0, sizeof(*mpa));
810 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
811 mpa->flags = MPA_REJECT;
812 mpa->revision = ep->mpa_attr.version;
813 mpa->private_data_size = htons(plen);
815 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
816 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
817 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
818 sizeof (struct mpa_v2_conn_params));
819 mpa_v2_params.ird = htons(((u16)ep->ird) |
820 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
822 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
824 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
825 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
826 FW_RI_INIT_P2PTYPE_READ_REQ ?
827 MPA_V2_RDMA_READ_RTR : 0) : 0));
828 memcpy(mpa->private_data, &mpa_v2_params,
829 sizeof(struct mpa_v2_conn_params));
832 memcpy(mpa->private_data +
833 sizeof(struct mpa_v2_conn_params), pdata, plen);
836 memcpy(mpa->private_data, pdata, plen);
839 * Reference the mpa skb again. This ensures the data area
840 * will remain in memory until the hw acks the tx.
841 * Function fw4_ack() will deref it.
844 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
845 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
848 ep->snd_seq += mpalen;
849 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
852 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
855 struct fw_ofld_tx_data_wr *req;
856 struct mpa_message *mpa;
858 struct mpa_v2_conn_params mpa_v2_params;
860 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
862 mpalen = sizeof(*mpa) + plen;
863 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
864 mpalen += sizeof(struct mpa_v2_conn_params);
865 wrlen = roundup(mpalen + sizeof *req, 16);
867 skb = get_skb(NULL, wrlen, GFP_KERNEL);
869 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
872 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
874 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
875 memset(req, 0, wrlen);
876 req->op_to_immdlen = cpu_to_be32(
877 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
879 FW_WR_IMMDLEN(mpalen));
880 req->flowid_len16 = cpu_to_be32(
881 FW_WR_FLOWID(ep->hwtid) |
882 FW_WR_LEN16(wrlen >> 4));
883 req->plen = cpu_to_be32(mpalen);
884 req->tunnel_to_proxy = cpu_to_be32(
885 FW_OFLD_TX_DATA_WR_FLUSH(1) |
886 FW_OFLD_TX_DATA_WR_SHOVE(1));
888 mpa = (struct mpa_message *)(req + 1);
889 memset(mpa, 0, sizeof(*mpa));
890 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
891 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
892 (markers_enabled ? MPA_MARKERS : 0);
893 mpa->revision = ep->mpa_attr.version;
894 mpa->private_data_size = htons(plen);
896 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
897 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
898 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
899 sizeof (struct mpa_v2_conn_params));
900 mpa_v2_params.ird = htons((u16)ep->ird);
901 mpa_v2_params.ord = htons((u16)ep->ord);
902 if (peer2peer && (ep->mpa_attr.p2p_type !=
903 FW_RI_INIT_P2PTYPE_DISABLED)) {
904 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
906 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
908 htons(MPA_V2_RDMA_WRITE_RTR);
909 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
911 htons(MPA_V2_RDMA_READ_RTR);
914 memcpy(mpa->private_data, &mpa_v2_params,
915 sizeof(struct mpa_v2_conn_params));
918 memcpy(mpa->private_data +
919 sizeof(struct mpa_v2_conn_params), pdata, plen);
922 memcpy(mpa->private_data, pdata, plen);
925 * Reference the mpa skb. This ensures the data area
926 * will remain in memory until the hw acks the tx.
927 * Function fw4_ack() will deref it.
930 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
932 __state_set(&ep->com, MPA_REP_SENT);
933 ep->snd_seq += mpalen;
934 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
937 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
940 struct cpl_act_establish *req = cplhdr(skb);
941 unsigned int tid = GET_TID(req);
942 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
943 struct tid_info *t = dev->rdev.lldi.tids;
945 ep = lookup_atid(t, atid);
947 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
948 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
950 mutex_lock(&ep->com.mutex);
951 dst_confirm(ep->dst);
953 /* setup the hwtid for this connection */
955 cxgb4_insert_tid(t, ep, tid);
956 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
958 ep->snd_seq = be32_to_cpu(req->snd_isn);
959 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
961 set_emss(ep, ntohs(req->tcp_opt));
963 /* dealloc the atid */
964 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
965 cxgb4_free_atid(t, atid);
966 set_bit(ACT_ESTAB, &ep->com.history);
968 /* start MPA negotiation */
969 send_flowc(ep, NULL);
970 if (ep->retry_with_mpa_v1)
971 send_mpa_req(ep, skb, 1);
973 send_mpa_req(ep, skb, mpa_rev);
974 mutex_unlock(&ep->com.mutex);
978 static void close_complete_upcall(struct c4iw_ep *ep, int status)
980 struct iw_cm_event event;
982 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
983 memset(&event, 0, sizeof(event));
984 event.event = IW_CM_EVENT_CLOSE;
985 event.status = status;
987 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
988 ep, ep->com.cm_id, ep->hwtid);
989 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
990 ep->com.cm_id->rem_ref(ep->com.cm_id);
991 ep->com.cm_id = NULL;
992 set_bit(CLOSE_UPCALL, &ep->com.history);
996 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
998 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
999 state_set(&ep->com, ABORTING);
1000 set_bit(ABORT_CONN, &ep->com.history);
1001 return send_abort(ep, skb, gfp);
1004 static void peer_close_upcall(struct c4iw_ep *ep)
1006 struct iw_cm_event event;
1008 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1009 memset(&event, 0, sizeof(event));
1010 event.event = IW_CM_EVENT_DISCONNECT;
1011 if (ep->com.cm_id) {
1012 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1013 ep, ep->com.cm_id, ep->hwtid);
1014 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1015 set_bit(DISCONN_UPCALL, &ep->com.history);
1019 static void peer_abort_upcall(struct c4iw_ep *ep)
1021 struct iw_cm_event event;
1023 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1024 memset(&event, 0, sizeof(event));
1025 event.event = IW_CM_EVENT_CLOSE;
1026 event.status = -ECONNRESET;
1027 if (ep->com.cm_id) {
1028 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1029 ep->com.cm_id, ep->hwtid);
1030 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1031 ep->com.cm_id->rem_ref(ep->com.cm_id);
1032 ep->com.cm_id = NULL;
1033 set_bit(ABORT_UPCALL, &ep->com.history);
1037 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1039 struct iw_cm_event event;
1041 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1042 memset(&event, 0, sizeof(event));
1043 event.event = IW_CM_EVENT_CONNECT_REPLY;
1044 event.status = status;
1045 memcpy(&event.local_addr, &ep->com.local_addr,
1046 sizeof(ep->com.local_addr));
1047 memcpy(&event.remote_addr, &ep->com.remote_addr,
1048 sizeof(ep->com.remote_addr));
1050 if ((status == 0) || (status == -ECONNREFUSED)) {
1051 if (!ep->tried_with_mpa_v1) {
1052 /* this means MPA_v2 is used */
1053 event.private_data_len = ep->plen -
1054 sizeof(struct mpa_v2_conn_params);
1055 event.private_data = ep->mpa_pkt +
1056 sizeof(struct mpa_message) +
1057 sizeof(struct mpa_v2_conn_params);
1059 /* this means MPA_v1 is used */
1060 event.private_data_len = ep->plen;
1061 event.private_data = ep->mpa_pkt +
1062 sizeof(struct mpa_message);
1066 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1068 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1069 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1072 ep->com.cm_id->rem_ref(ep->com.cm_id);
1073 ep->com.cm_id = NULL;
1077 static int connect_request_upcall(struct c4iw_ep *ep)
1079 struct iw_cm_event event;
1082 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1083 memset(&event, 0, sizeof(event));
1084 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1085 memcpy(&event.local_addr, &ep->com.local_addr,
1086 sizeof(ep->com.local_addr));
1087 memcpy(&event.remote_addr, &ep->com.remote_addr,
1088 sizeof(ep->com.remote_addr));
1089 event.provider_data = ep;
1090 if (!ep->tried_with_mpa_v1) {
1091 /* this means MPA_v2 is used */
1092 event.ord = ep->ord;
1093 event.ird = ep->ird;
1094 event.private_data_len = ep->plen -
1095 sizeof(struct mpa_v2_conn_params);
1096 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1097 sizeof(struct mpa_v2_conn_params);
1099 /* this means MPA_v1 is used. Send max supported */
1100 event.ord = c4iw_max_read_depth;
1101 event.ird = c4iw_max_read_depth;
1102 event.private_data_len = ep->plen;
1103 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1105 c4iw_get_ep(&ep->com);
1106 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1109 c4iw_put_ep(&ep->com);
1110 set_bit(CONNREQ_UPCALL, &ep->com.history);
1111 c4iw_put_ep(&ep->parent_ep->com);
1115 static void established_upcall(struct c4iw_ep *ep)
1117 struct iw_cm_event event;
1119 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1120 memset(&event, 0, sizeof(event));
1121 event.event = IW_CM_EVENT_ESTABLISHED;
1122 event.ird = ep->ird;
1123 event.ord = ep->ord;
1124 if (ep->com.cm_id) {
1125 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1126 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1127 set_bit(ESTAB_UPCALL, &ep->com.history);
1131 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1133 struct cpl_rx_data_ack *req;
1134 struct sk_buff *skb;
1135 int wrlen = roundup(sizeof *req, 16);
1137 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1138 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1140 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1144 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1145 memset(req, 0, wrlen);
1146 INIT_TP_WR(req, ep->hwtid);
1147 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1149 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1151 V_RX_DACK_MODE(dack_mode));
1152 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1153 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1157 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1159 struct mpa_message *mpa;
1160 struct mpa_v2_conn_params *mpa_v2_params;
1162 u16 resp_ird, resp_ord;
1163 u8 rtr_mismatch = 0, insuff_ird = 0;
1164 struct c4iw_qp_attributes attrs;
1165 enum c4iw_qp_attr_mask mask;
1168 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1171 * Stop mpa timer. If it expired, then
1172 * we ignore the MPA reply. process_timeout()
1173 * will abort the connection.
1175 if (stop_ep_timer(ep))
1179 * If we get more than the supported amount of private data
1180 * then we must fail this connection.
1182 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1188 * copy the new data into our accumulation buffer.
1190 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1192 ep->mpa_pkt_len += skb->len;
1195 * if we don't even have the mpa message, then bail.
1197 if (ep->mpa_pkt_len < sizeof(*mpa))
1199 mpa = (struct mpa_message *) ep->mpa_pkt;
1201 /* Validate MPA header. */
1202 if (mpa->revision > mpa_rev) {
1203 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1204 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1208 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1213 plen = ntohs(mpa->private_data_size);
1216 * Fail if there's too much private data.
1218 if (plen > MPA_MAX_PRIVATE_DATA) {
1224 * If plen does not account for pkt size
1226 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1231 ep->plen = (u8) plen;
1234 * If we don't have all the pdata yet, then bail.
1235 * We'll continue process when more data arrives.
1237 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1240 if (mpa->flags & MPA_REJECT) {
1241 err = -ECONNREFUSED;
1246 * If we get here we have accumulated the entire mpa
1247 * start reply message including private data. And
1248 * the MPA header is valid.
1250 __state_set(&ep->com, FPDU_MODE);
1251 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1252 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1253 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1254 ep->mpa_attr.version = mpa->revision;
1255 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1257 if (mpa->revision == 2) {
1258 ep->mpa_attr.enhanced_rdma_conn =
1259 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1260 if (ep->mpa_attr.enhanced_rdma_conn) {
1261 mpa_v2_params = (struct mpa_v2_conn_params *)
1262 (ep->mpa_pkt + sizeof(*mpa));
1263 resp_ird = ntohs(mpa_v2_params->ird) &
1264 MPA_V2_IRD_ORD_MASK;
1265 resp_ord = ntohs(mpa_v2_params->ord) &
1266 MPA_V2_IRD_ORD_MASK;
1269 * This is a double-check. Ideally, below checks are
1270 * not required since ird/ord stuff has been taken
1271 * care of in c4iw_accept_cr
1273 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1280 if (ntohs(mpa_v2_params->ird) &
1281 MPA_V2_PEER2PEER_MODEL) {
1282 if (ntohs(mpa_v2_params->ord) &
1283 MPA_V2_RDMA_WRITE_RTR)
1284 ep->mpa_attr.p2p_type =
1285 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1286 else if (ntohs(mpa_v2_params->ord) &
1287 MPA_V2_RDMA_READ_RTR)
1288 ep->mpa_attr.p2p_type =
1289 FW_RI_INIT_P2PTYPE_READ_REQ;
1292 } else if (mpa->revision == 1)
1294 ep->mpa_attr.p2p_type = p2p_type;
1296 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1297 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1298 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1299 ep->mpa_attr.recv_marker_enabled,
1300 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1301 ep->mpa_attr.p2p_type, p2p_type);
1304 * If responder's RTR does not match with that of initiator, assign
1305 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1306 * generated when moving QP to RTS state.
1307 * A TERM message will be sent after QP has moved to RTS state
1309 if ((ep->mpa_attr.version == 2) && peer2peer &&
1310 (ep->mpa_attr.p2p_type != p2p_type)) {
1311 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1315 attrs.mpa_attr = ep->mpa_attr;
1316 attrs.max_ird = ep->ird;
1317 attrs.max_ord = ep->ord;
1318 attrs.llp_stream_handle = ep;
1319 attrs.next_state = C4IW_QP_STATE_RTS;
1321 mask = C4IW_QP_ATTR_NEXT_STATE |
1322 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1323 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1325 /* bind QP and TID with INIT_WR */
1326 err = c4iw_modify_qp(ep->com.qp->rhp,
1327 ep->com.qp, mask, &attrs, 1);
1332 * If responder's RTR requirement did not match with what initiator
1333 * supports, generate TERM message
1336 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1337 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1338 attrs.ecode = MPA_NOMATCH_RTR;
1339 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1340 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1341 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1347 * Generate TERM if initiator IRD is not sufficient for responder
1348 * provided ORD. Currently, we do the same behaviour even when
1349 * responder provided IRD is also not sufficient as regards to
1353 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1355 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1356 attrs.ecode = MPA_INSUFF_IRD;
1357 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1358 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1359 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1365 __state_set(&ep->com, ABORTING);
1366 send_abort(ep, skb, GFP_KERNEL);
1368 connect_reply_upcall(ep, err);
1372 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1374 struct mpa_message *mpa;
1375 struct mpa_v2_conn_params *mpa_v2_params;
1378 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1381 * If we get more than the supported amount of private data
1382 * then we must fail this connection.
1384 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1385 (void)stop_ep_timer(ep);
1386 abort_connection(ep, skb, GFP_KERNEL);
1390 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1393 * Copy the new data into our accumulation buffer.
1395 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1397 ep->mpa_pkt_len += skb->len;
1400 * If we don't even have the mpa message, then bail.
1401 * We'll continue process when more data arrives.
1403 if (ep->mpa_pkt_len < sizeof(*mpa))
1406 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1407 mpa = (struct mpa_message *) ep->mpa_pkt;
1410 * Validate MPA Header.
1412 if (mpa->revision > mpa_rev) {
1413 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1414 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1415 (void)stop_ep_timer(ep);
1416 abort_connection(ep, skb, GFP_KERNEL);
1420 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1421 (void)stop_ep_timer(ep);
1422 abort_connection(ep, skb, GFP_KERNEL);
1426 plen = ntohs(mpa->private_data_size);
1429 * Fail if there's too much private data.
1431 if (plen > MPA_MAX_PRIVATE_DATA) {
1432 (void)stop_ep_timer(ep);
1433 abort_connection(ep, skb, GFP_KERNEL);
1438 * If plen does not account for pkt size
1440 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1441 (void)stop_ep_timer(ep);
1442 abort_connection(ep, skb, GFP_KERNEL);
1445 ep->plen = (u8) plen;
1448 * If we don't have all the pdata yet, then bail.
1450 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1454 * If we get here we have accumulated the entire mpa
1455 * start reply message including private data.
1457 ep->mpa_attr.initiator = 0;
1458 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1459 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1460 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1461 ep->mpa_attr.version = mpa->revision;
1462 if (mpa->revision == 1)
1463 ep->tried_with_mpa_v1 = 1;
1464 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1466 if (mpa->revision == 2) {
1467 ep->mpa_attr.enhanced_rdma_conn =
1468 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1469 if (ep->mpa_attr.enhanced_rdma_conn) {
1470 mpa_v2_params = (struct mpa_v2_conn_params *)
1471 (ep->mpa_pkt + sizeof(*mpa));
1472 ep->ird = ntohs(mpa_v2_params->ird) &
1473 MPA_V2_IRD_ORD_MASK;
1474 ep->ord = ntohs(mpa_v2_params->ord) &
1475 MPA_V2_IRD_ORD_MASK;
1476 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1478 if (ntohs(mpa_v2_params->ord) &
1479 MPA_V2_RDMA_WRITE_RTR)
1480 ep->mpa_attr.p2p_type =
1481 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1482 else if (ntohs(mpa_v2_params->ord) &
1483 MPA_V2_RDMA_READ_RTR)
1484 ep->mpa_attr.p2p_type =
1485 FW_RI_INIT_P2PTYPE_READ_REQ;
1488 } else if (mpa->revision == 1)
1490 ep->mpa_attr.p2p_type = p2p_type;
1492 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1493 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1494 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1495 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1496 ep->mpa_attr.p2p_type);
1499 * If the endpoint timer already expired, then we ignore
1500 * the start request. process_timeout() will abort
1503 if (!stop_ep_timer(ep)) {
1504 __state_set(&ep->com, MPA_REQ_RCVD);
1507 mutex_lock(&ep->parent_ep->com.mutex);
1508 if (ep->parent_ep->com.state != DEAD) {
1509 if (connect_request_upcall(ep))
1510 abort_connection(ep, skb, GFP_KERNEL);
1512 abort_connection(ep, skb, GFP_KERNEL);
1514 mutex_unlock(&ep->parent_ep->com.mutex);
1519 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1522 struct cpl_rx_data *hdr = cplhdr(skb);
1523 unsigned int dlen = ntohs(hdr->len);
1524 unsigned int tid = GET_TID(hdr);
1525 struct tid_info *t = dev->rdev.lldi.tids;
1526 __u8 status = hdr->status;
1528 ep = lookup_tid(t, tid);
1531 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1532 skb_pull(skb, sizeof(*hdr));
1533 skb_trim(skb, dlen);
1534 mutex_lock(&ep->com.mutex);
1536 /* update RX credits */
1537 update_rx_credits(ep, dlen);
1539 switch (ep->com.state) {
1541 ep->rcv_seq += dlen;
1542 process_mpa_reply(ep, skb);
1545 ep->rcv_seq += dlen;
1546 process_mpa_request(ep, skb);
1549 struct c4iw_qp_attributes attrs;
1550 BUG_ON(!ep->com.qp);
1552 pr_err("%s Unexpected streaming data." \
1553 " qpid %u ep %p state %d tid %u status %d\n",
1554 __func__, ep->com.qp->wq.sq.qid, ep,
1555 ep->com.state, ep->hwtid, status);
1556 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1557 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1558 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1564 mutex_unlock(&ep->com.mutex);
1568 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1571 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1573 unsigned int tid = GET_TID(rpl);
1574 struct tid_info *t = dev->rdev.lldi.tids;
1576 ep = lookup_tid(t, tid);
1578 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1581 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1582 mutex_lock(&ep->com.mutex);
1583 switch (ep->com.state) {
1585 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1586 __state_set(&ep->com, DEAD);
1590 printk(KERN_ERR "%s ep %p state %d\n",
1591 __func__, ep, ep->com.state);
1594 mutex_unlock(&ep->com.mutex);
1597 release_ep_resources(ep);
1601 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1603 struct sk_buff *skb;
1604 struct fw_ofld_connection_wr *req;
1605 unsigned int mtu_idx;
1607 struct sockaddr_in *sin;
1609 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1610 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1611 memset(req, 0, sizeof(*req));
1612 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1613 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1614 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1615 ep->com.dev->rdev.lldi.ports[0],
1617 sin = (struct sockaddr_in *)&ep->com.local_addr;
1618 req->le.lport = sin->sin_port;
1619 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1620 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1621 req->le.pport = sin->sin_port;
1622 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1623 req->tcb.t_state_to_astid =
1624 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1625 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1626 req->tcb.cplrxdataack_cplpassacceptrpl =
1627 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1628 req->tcb.tx_max = (__force __be32) jiffies;
1629 req->tcb.rcv_adv = htons(1);
1630 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1631 wscale = compute_wscale(rcv_win);
1632 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1633 (nocong ? NO_CONG(1) : 0) |
1638 L2T_IDX(ep->l2t->idx) |
1639 TX_CHAN(ep->tx_chan) |
1640 SMAC_SEL(ep->smac_idx) |
1642 ULP_MODE(ULP_MODE_TCPDDP) |
1643 RCV_BUFSIZ(rcv_win >> 10));
1644 req->tcb.opt2 = (__force __be32) (PACE(1) |
1645 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1647 CCTRL_ECN(enable_ecn) |
1648 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1649 if (enable_tcp_timestamps)
1650 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1651 if (enable_tcp_sack)
1652 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1653 if (wscale && enable_tcp_window_scaling)
1654 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1655 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1656 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1657 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1658 set_bit(ACT_OFLD_CONN, &ep->com.history);
1659 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1663 * Return whether a failed active open has allocated a TID
1665 static inline int act_open_has_tid(int status)
1667 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1668 status != CPL_ERR_ARP_MISS;
1671 /* Returns whether a CPL status conveys negative advice.
1673 static int is_neg_adv(unsigned int status)
1675 return status == CPL_ERR_RTX_NEG_ADVICE ||
1676 status == CPL_ERR_PERSIST_NEG_ADVICE ||
1677 status == CPL_ERR_KEEPALV_NEG_ADVICE;
1680 #define ACT_OPEN_RETRY_COUNT 2
1682 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1683 struct dst_entry *dst, struct c4iw_dev *cdev,
1686 struct neighbour *n;
1688 struct net_device *pdev;
1690 n = dst_neigh_lookup(dst, peer_ip);
1696 if (n->dev->flags & IFF_LOOPBACK) {
1698 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1699 else if (IS_ENABLED(CONFIG_IPV6))
1700 for_each_netdev(&init_net, pdev) {
1701 if (ipv6_chk_addr(&init_net,
1702 (struct in6_addr *)peer_ip,
1713 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1717 ep->mtu = pdev->mtu;
1718 ep->tx_chan = cxgb4_port_chan(pdev);
1719 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1720 step = cdev->rdev.lldi.ntxq /
1721 cdev->rdev.lldi.nchan;
1722 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1723 step = cdev->rdev.lldi.nrxq /
1724 cdev->rdev.lldi.nchan;
1725 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1726 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1727 cxgb4_port_idx(pdev) * step];
1730 pdev = get_real_dev(n->dev);
1731 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1735 ep->mtu = dst_mtu(dst);
1736 ep->tx_chan = cxgb4_port_chan(n->dev);
1737 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1738 step = cdev->rdev.lldi.ntxq /
1739 cdev->rdev.lldi.nchan;
1740 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1741 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1742 step = cdev->rdev.lldi.nrxq /
1743 cdev->rdev.lldi.nchan;
1744 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1745 cxgb4_port_idx(n->dev) * step];
1748 ep->retry_with_mpa_v1 = 0;
1749 ep->tried_with_mpa_v1 = 0;
1761 static int c4iw_reconnect(struct c4iw_ep *ep)
1764 struct sockaddr_in *laddr = (struct sockaddr_in *)
1765 &ep->com.cm_id->local_addr;
1766 struct sockaddr_in *raddr = (struct sockaddr_in *)
1767 &ep->com.cm_id->remote_addr;
1768 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1769 &ep->com.cm_id->local_addr;
1770 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1771 &ep->com.cm_id->remote_addr;
1775 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1776 init_timer(&ep->timer);
1779 * Allocate an active TID to initiate a TCP connection.
1781 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1782 if (ep->atid == -1) {
1783 pr_err("%s - cannot alloc atid.\n", __func__);
1787 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1790 if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1791 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1792 raddr->sin_addr.s_addr, laddr->sin_port,
1793 raddr->sin_port, 0);
1795 ra = (__u8 *)&raddr->sin_addr;
1797 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1798 raddr6->sin6_addr.s6_addr,
1799 laddr6->sin6_port, raddr6->sin6_port, 0,
1800 raddr6->sin6_scope_id);
1802 ra = (__u8 *)&raddr6->sin6_addr;
1805 pr_err("%s - cannot find route.\n", __func__);
1806 err = -EHOSTUNREACH;
1809 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1811 pr_err("%s - cannot alloc l2e.\n", __func__);
1815 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1816 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1819 state_set(&ep->com, CONNECTING);
1822 /* send connect request to rnic */
1823 err = send_connect(ep);
1827 cxgb4_l2t_release(ep->l2t);
1829 dst_release(ep->dst);
1831 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1832 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1835 * remember to send notification to upper layer.
1836 * We are in here so the upper layer is not aware that this is
1837 * re-connect attempt and so, upper layer is still waiting for
1838 * response of 1st connect request.
1840 connect_reply_upcall(ep, -ECONNRESET);
1841 c4iw_put_ep(&ep->com);
1846 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1849 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1850 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1851 ntohl(rpl->atid_status)));
1852 struct tid_info *t = dev->rdev.lldi.tids;
1853 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1854 struct sockaddr_in *la;
1855 struct sockaddr_in *ra;
1856 struct sockaddr_in6 *la6;
1857 struct sockaddr_in6 *ra6;
1859 ep = lookup_atid(t, atid);
1860 la = (struct sockaddr_in *)&ep->com.local_addr;
1861 ra = (struct sockaddr_in *)&ep->com.remote_addr;
1862 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1863 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1865 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1866 status, status2errno(status));
1868 if (is_neg_adv(status)) {
1869 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1874 set_bit(ACT_OPEN_RPL, &ep->com.history);
1877 * Log interesting failures.
1880 case CPL_ERR_CONN_RESET:
1881 case CPL_ERR_CONN_TIMEDOUT:
1883 case CPL_ERR_TCAM_FULL:
1884 mutex_lock(&dev->rdev.stats.lock);
1885 dev->rdev.stats.tcam_full++;
1886 mutex_unlock(&dev->rdev.stats.lock);
1887 if (ep->com.local_addr.ss_family == AF_INET &&
1888 dev->rdev.lldi.enable_fw_ofld_conn) {
1889 send_fw_act_open_req(ep,
1890 GET_TID_TID(GET_AOPEN_ATID(
1891 ntohl(rpl->atid_status))));
1895 case CPL_ERR_CONN_EXIST:
1896 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1897 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1898 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1900 cxgb4_free_atid(t, atid);
1901 dst_release(ep->dst);
1902 cxgb4_l2t_release(ep->l2t);
1908 if (ep->com.local_addr.ss_family == AF_INET) {
1909 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1910 atid, status, status2errno(status),
1911 &la->sin_addr.s_addr, ntohs(la->sin_port),
1912 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
1914 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1915 atid, status, status2errno(status),
1916 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1917 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1922 connect_reply_upcall(ep, status2errno(status));
1923 state_set(&ep->com, DEAD);
1925 if (status && act_open_has_tid(status))
1926 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1928 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1929 cxgb4_free_atid(t, atid);
1930 dst_release(ep->dst);
1931 cxgb4_l2t_release(ep->l2t);
1932 c4iw_put_ep(&ep->com);
1937 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1939 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1940 struct tid_info *t = dev->rdev.lldi.tids;
1941 unsigned int stid = GET_TID(rpl);
1942 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1945 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1948 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1949 rpl->status, status2errno(rpl->status));
1950 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1956 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1958 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1959 struct tid_info *t = dev->rdev.lldi.tids;
1960 unsigned int stid = GET_TID(rpl);
1961 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1963 PDBG("%s ep %p\n", __func__, ep);
1964 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1968 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1969 struct cpl_pass_accept_req *req)
1971 struct cpl_pass_accept_rpl *rpl;
1972 unsigned int mtu_idx;
1977 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1978 BUG_ON(skb_cloned(skb));
1979 skb_trim(skb, sizeof(*rpl));
1981 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1982 wscale = compute_wscale(rcv_win);
1983 opt0 = (nocong ? NO_CONG(1) : 0) |
1988 L2T_IDX(ep->l2t->idx) |
1989 TX_CHAN(ep->tx_chan) |
1990 SMAC_SEL(ep->smac_idx) |
1991 DSCP(ep->tos >> 2) |
1992 ULP_MODE(ULP_MODE_TCPDDP) |
1993 RCV_BUFSIZ(rcv_win>>10);
1994 opt2 = RX_CHANNEL(0) |
1995 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1997 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1998 opt2 |= TSTAMPS_EN(1);
1999 if (enable_tcp_sack && req->tcpopt.sack)
2001 if (wscale && enable_tcp_window_scaling)
2002 opt2 |= WND_SCALE_EN(1);
2004 const struct tcphdr *tcph;
2005 u32 hlen = ntohl(req->hdr_len);
2007 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2009 if (tcph->ece && tcph->cwr)
2010 opt2 |= CCTRL_ECN(1);
2014 INIT_TP_WR(rpl, ep->hwtid);
2015 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2017 rpl->opt0 = cpu_to_be64(opt0);
2018 rpl->opt2 = cpu_to_be32(opt2);
2019 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2020 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2021 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2026 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2028 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2029 BUG_ON(skb_cloned(skb));
2030 skb_trim(skb, sizeof(struct cpl_tid_release));
2032 release_tid(&dev->rdev, hwtid, skb);
2036 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2037 __u8 *local_ip, __u8 *peer_ip,
2038 __be16 *local_port, __be16 *peer_port)
2040 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2041 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2042 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2043 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2044 struct tcphdr *tcp = (struct tcphdr *)
2045 ((u8 *)(req + 1) + eth_len + ip_len);
2047 if (ip->version == 4) {
2048 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2049 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2052 memcpy(peer_ip, &ip->saddr, 4);
2053 memcpy(local_ip, &ip->daddr, 4);
2055 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2056 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2059 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2060 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2062 *peer_port = tcp->source;
2063 *local_port = tcp->dest;
2068 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2070 struct c4iw_ep *child_ep = NULL, *parent_ep;
2071 struct cpl_pass_accept_req *req = cplhdr(skb);
2072 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2073 struct tid_info *t = dev->rdev.lldi.tids;
2074 unsigned int hwtid = GET_TID(req);
2075 struct dst_entry *dst;
2076 __u8 local_ip[16], peer_ip[16];
2077 __be16 local_port, peer_port;
2079 u16 peer_mss = ntohs(req->tcpopt.mss);
2082 parent_ep = lookup_stid(t, stid);
2084 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2088 if (state_read(&parent_ep->com) != LISTEN) {
2089 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2094 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2096 /* Find output route */
2098 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2099 , __func__, parent_ep, hwtid,
2100 local_ip, peer_ip, ntohs(local_port),
2101 ntohs(peer_port), peer_mss);
2102 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2103 local_port, peer_port,
2104 GET_POPEN_TOS(ntohl(req->tos_stid)));
2106 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2107 , __func__, parent_ep, hwtid,
2108 local_ip, peer_ip, ntohs(local_port),
2109 ntohs(peer_port), peer_mss);
2110 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2111 PASS_OPEN_TOS(ntohl(req->tos_stid)),
2112 ((struct sockaddr_in6 *)
2113 &parent_ep->com.local_addr)->sin6_scope_id);
2116 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2121 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2123 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2129 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2131 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2138 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2139 child_ep->mtu = peer_mss + 40;
2141 state_set(&child_ep->com, CONNECTING);
2142 child_ep->com.dev = dev;
2143 child_ep->com.cm_id = NULL;
2145 struct sockaddr_in *sin = (struct sockaddr_in *)
2146 &child_ep->com.local_addr;
2147 sin->sin_family = PF_INET;
2148 sin->sin_port = local_port;
2149 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2150 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2151 sin->sin_family = PF_INET;
2152 sin->sin_port = peer_port;
2153 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2155 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2156 &child_ep->com.local_addr;
2157 sin6->sin6_family = PF_INET6;
2158 sin6->sin6_port = local_port;
2159 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2160 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2161 sin6->sin6_family = PF_INET6;
2162 sin6->sin6_port = peer_port;
2163 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2165 c4iw_get_ep(&parent_ep->com);
2166 child_ep->parent_ep = parent_ep;
2167 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2168 child_ep->dst = dst;
2169 child_ep->hwtid = hwtid;
2171 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2172 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2174 init_timer(&child_ep->timer);
2175 cxgb4_insert_tid(t, child_ep, hwtid);
2176 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2177 accept_cr(child_ep, skb, req);
2178 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2181 reject_cr(dev, hwtid, skb);
2186 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2189 struct cpl_pass_establish *req = cplhdr(skb);
2190 struct tid_info *t = dev->rdev.lldi.tids;
2191 unsigned int tid = GET_TID(req);
2193 ep = lookup_tid(t, tid);
2194 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2195 ep->snd_seq = be32_to_cpu(req->snd_isn);
2196 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2198 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2199 ntohs(req->tcp_opt));
2201 set_emss(ep, ntohs(req->tcp_opt));
2203 dst_confirm(ep->dst);
2204 state_set(&ep->com, MPA_REQ_WAIT);
2206 send_flowc(ep, skb);
2207 set_bit(PASS_ESTAB, &ep->com.history);
2212 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2214 struct cpl_peer_close *hdr = cplhdr(skb);
2216 struct c4iw_qp_attributes attrs;
2219 struct tid_info *t = dev->rdev.lldi.tids;
2220 unsigned int tid = GET_TID(hdr);
2223 ep = lookup_tid(t, tid);
2224 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2225 dst_confirm(ep->dst);
2227 set_bit(PEER_CLOSE, &ep->com.history);
2228 mutex_lock(&ep->com.mutex);
2229 switch (ep->com.state) {
2231 __state_set(&ep->com, CLOSING);
2234 __state_set(&ep->com, CLOSING);
2235 connect_reply_upcall(ep, -ECONNRESET);
2240 * We're gonna mark this puppy DEAD, but keep
2241 * the reference on it until the ULP accepts or
2242 * rejects the CR. Also wake up anyone waiting
2243 * in rdma connection migration (see c4iw_accept_cr()).
2245 __state_set(&ep->com, CLOSING);
2246 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2247 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2250 __state_set(&ep->com, CLOSING);
2251 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2252 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2256 __state_set(&ep->com, CLOSING);
2257 attrs.next_state = C4IW_QP_STATE_CLOSING;
2258 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2259 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2260 if (ret != -ECONNRESET) {
2261 peer_close_upcall(ep);
2269 __state_set(&ep->com, MORIBUND);
2273 (void)stop_ep_timer(ep);
2274 if (ep->com.cm_id && ep->com.qp) {
2275 attrs.next_state = C4IW_QP_STATE_IDLE;
2276 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2277 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2279 close_complete_upcall(ep, 0);
2280 __state_set(&ep->com, DEAD);
2290 mutex_unlock(&ep->com.mutex);
2292 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2294 release_ep_resources(ep);
2298 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2300 struct cpl_abort_req_rss *req = cplhdr(skb);
2302 struct cpl_abort_rpl *rpl;
2303 struct sk_buff *rpl_skb;
2304 struct c4iw_qp_attributes attrs;
2307 struct tid_info *t = dev->rdev.lldi.tids;
2308 unsigned int tid = GET_TID(req);
2310 ep = lookup_tid(t, tid);
2311 if (is_neg_adv(req->status)) {
2312 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2316 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2318 set_bit(PEER_ABORT, &ep->com.history);
2321 * Wake up any threads in rdma_init() or rdma_fini().
2322 * However, this is not needed if com state is just
2325 if (ep->com.state != MPA_REQ_SENT)
2326 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2328 mutex_lock(&ep->com.mutex);
2329 switch (ep->com.state) {
2333 (void)stop_ep_timer(ep);
2336 (void)stop_ep_timer(ep);
2337 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2338 connect_reply_upcall(ep, -ECONNRESET);
2341 * we just don't send notification upwards because we
2342 * want to retry with mpa_v1 without upper layers even
2345 * do some housekeeping so as to re-initiate the
2348 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2350 ep->retry_with_mpa_v1 = 1;
2362 if (ep->com.cm_id && ep->com.qp) {
2363 attrs.next_state = C4IW_QP_STATE_ERROR;
2364 ret = c4iw_modify_qp(ep->com.qp->rhp,
2365 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2369 "%s - qp <- error failed!\n",
2372 peer_abort_upcall(ep);
2377 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2378 mutex_unlock(&ep->com.mutex);
2384 dst_confirm(ep->dst);
2385 if (ep->com.state != ABORTING) {
2386 __state_set(&ep->com, DEAD);
2387 /* we don't release if we want to retry with mpa_v1 */
2388 if (!ep->retry_with_mpa_v1)
2391 mutex_unlock(&ep->com.mutex);
2393 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2395 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2400 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2401 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2402 INIT_TP_WR(rpl, ep->hwtid);
2403 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2404 rpl->cmd = CPL_ABORT_NO_RST;
2405 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2408 release_ep_resources(ep);
2409 else if (ep->retry_with_mpa_v1) {
2410 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2411 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2412 dst_release(ep->dst);
2413 cxgb4_l2t_release(ep->l2t);
2420 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2423 struct c4iw_qp_attributes attrs;
2424 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2426 struct tid_info *t = dev->rdev.lldi.tids;
2427 unsigned int tid = GET_TID(rpl);
2429 ep = lookup_tid(t, tid);
2431 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2434 /* The cm_id may be null if we failed to connect */
2435 mutex_lock(&ep->com.mutex);
2436 switch (ep->com.state) {
2438 __state_set(&ep->com, MORIBUND);
2441 (void)stop_ep_timer(ep);
2442 if ((ep->com.cm_id) && (ep->com.qp)) {
2443 attrs.next_state = C4IW_QP_STATE_IDLE;
2444 c4iw_modify_qp(ep->com.qp->rhp,
2446 C4IW_QP_ATTR_NEXT_STATE,
2449 close_complete_upcall(ep, 0);
2450 __state_set(&ep->com, DEAD);
2460 mutex_unlock(&ep->com.mutex);
2462 release_ep_resources(ep);
2466 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2468 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2469 struct tid_info *t = dev->rdev.lldi.tids;
2470 unsigned int tid = GET_TID(rpl);
2472 struct c4iw_qp_attributes attrs;
2474 ep = lookup_tid(t, tid);
2477 if (ep && ep->com.qp) {
2478 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2479 ep->com.qp->wq.sq.qid);
2480 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2481 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2482 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2484 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2490 * Upcall from the adapter indicating data has been transmitted.
2491 * For us its just the single MPA request or reply. We can now free
2492 * the skb holding the mpa message.
2494 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2497 struct cpl_fw4_ack *hdr = cplhdr(skb);
2498 u8 credits = hdr->credits;
2499 unsigned int tid = GET_TID(hdr);
2500 struct tid_info *t = dev->rdev.lldi.tids;
2503 ep = lookup_tid(t, tid);
2504 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2506 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2507 __func__, ep, ep->hwtid, state_read(&ep->com));
2511 dst_confirm(ep->dst);
2513 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2514 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2515 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2516 kfree_skb(ep->mpa_skb);
2522 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2526 struct c4iw_ep *ep = to_ep(cm_id);
2527 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2529 mutex_lock(&ep->com.mutex);
2530 if (ep->com.state == DEAD) {
2531 mutex_unlock(&ep->com.mutex);
2532 c4iw_put_ep(&ep->com);
2535 set_bit(ULP_REJECT, &ep->com.history);
2536 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2538 abort_connection(ep, NULL, GFP_KERNEL);
2540 err = send_mpa_reject(ep, pdata, pdata_len);
2543 mutex_unlock(&ep->com.mutex);
2545 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2546 c4iw_put_ep(&ep->com);
2550 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2553 struct c4iw_qp_attributes attrs;
2554 enum c4iw_qp_attr_mask mask;
2555 struct c4iw_ep *ep = to_ep(cm_id);
2556 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2557 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2559 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2561 mutex_lock(&ep->com.mutex);
2562 if (ep->com.state == DEAD) {
2567 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2570 set_bit(ULP_ACCEPT, &ep->com.history);
2571 if ((conn_param->ord > c4iw_max_read_depth) ||
2572 (conn_param->ird > c4iw_max_read_depth)) {
2573 abort_connection(ep, NULL, GFP_KERNEL);
2578 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2579 if (conn_param->ord > ep->ird) {
2580 ep->ird = conn_param->ird;
2581 ep->ord = conn_param->ord;
2582 send_mpa_reject(ep, conn_param->private_data,
2583 conn_param->private_data_len);
2584 abort_connection(ep, NULL, GFP_KERNEL);
2588 if (conn_param->ird > ep->ord) {
2590 conn_param->ird = 1;
2592 abort_connection(ep, NULL, GFP_KERNEL);
2599 ep->ird = conn_param->ird;
2600 ep->ord = conn_param->ord;
2602 if (ep->mpa_attr.version != 2)
2603 if (peer2peer && ep->ird == 0)
2606 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2608 cm_id->add_ref(cm_id);
2609 ep->com.cm_id = cm_id;
2613 /* bind QP to EP and move to RTS */
2614 attrs.mpa_attr = ep->mpa_attr;
2615 attrs.max_ird = ep->ird;
2616 attrs.max_ord = ep->ord;
2617 attrs.llp_stream_handle = ep;
2618 attrs.next_state = C4IW_QP_STATE_RTS;
2620 /* bind QP and TID with INIT_WR */
2621 mask = C4IW_QP_ATTR_NEXT_STATE |
2622 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2623 C4IW_QP_ATTR_MPA_ATTR |
2624 C4IW_QP_ATTR_MAX_IRD |
2625 C4IW_QP_ATTR_MAX_ORD;
2627 err = c4iw_modify_qp(ep->com.qp->rhp,
2628 ep->com.qp, mask, &attrs, 1);
2631 err = send_mpa_reply(ep, conn_param->private_data,
2632 conn_param->private_data_len);
2636 __state_set(&ep->com, FPDU_MODE);
2637 established_upcall(ep);
2638 mutex_unlock(&ep->com.mutex);
2639 c4iw_put_ep(&ep->com);
2642 ep->com.cm_id = NULL;
2643 cm_id->rem_ref(cm_id);
2645 mutex_unlock(&ep->com.mutex);
2646 c4iw_put_ep(&ep->com);
2650 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2652 struct in_device *ind;
2654 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2655 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2657 ind = in_dev_get(dev->rdev.lldi.ports[0]);
2659 return -EADDRNOTAVAIL;
2660 for_primary_ifa(ind) {
2661 laddr->sin_addr.s_addr = ifa->ifa_address;
2662 raddr->sin_addr.s_addr = ifa->ifa_address;
2668 return found ? 0 : -EADDRNOTAVAIL;
2671 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2672 unsigned char banned_flags)
2674 struct inet6_dev *idev;
2675 int err = -EADDRNOTAVAIL;
2678 idev = __in6_dev_get(dev);
2680 struct inet6_ifaddr *ifp;
2682 read_lock_bh(&idev->lock);
2683 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2684 if (ifp->scope == IFA_LINK &&
2685 !(ifp->flags & banned_flags)) {
2686 memcpy(addr, &ifp->addr, 16);
2691 read_unlock_bh(&idev->lock);
2697 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2699 struct in6_addr uninitialized_var(addr);
2700 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2701 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2703 if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2704 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2705 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2708 return -EADDRNOTAVAIL;
2711 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2713 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2716 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2717 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2718 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2719 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2720 &cm_id->remote_addr;
2724 if ((conn_param->ord > c4iw_max_read_depth) ||
2725 (conn_param->ird > c4iw_max_read_depth)) {
2729 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2731 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2735 init_timer(&ep->timer);
2736 ep->plen = conn_param->private_data_len;
2738 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2739 conn_param->private_data, ep->plen);
2740 ep->ird = conn_param->ird;
2741 ep->ord = conn_param->ord;
2743 if (peer2peer && ep->ord == 0)
2746 cm_id->add_ref(cm_id);
2748 ep->com.cm_id = cm_id;
2749 ep->com.qp = get_qhp(dev, conn_param->qpn);
2751 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2756 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2760 * Allocate an active TID to initiate a TCP connection.
2762 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2763 if (ep->atid == -1) {
2764 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2768 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2770 if (cm_id->remote_addr.ss_family == AF_INET) {
2772 ra = (__u8 *)&raddr->sin_addr;
2775 * Handle loopback requests to INADDR_ANY.
2777 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2778 err = pick_local_ipaddrs(dev, cm_id);
2784 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2785 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2786 ra, ntohs(raddr->sin_port));
2787 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2788 raddr->sin_addr.s_addr, laddr->sin_port,
2789 raddr->sin_port, 0);
2792 ra = (__u8 *)&raddr6->sin6_addr;
2795 * Handle loopback requests to INADDR_ANY.
2797 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2798 err = pick_local_ip6addrs(dev, cm_id);
2804 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2805 __func__, laddr6->sin6_addr.s6_addr,
2806 ntohs(laddr6->sin6_port),
2807 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2808 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2809 raddr6->sin6_addr.s6_addr,
2810 laddr6->sin6_port, raddr6->sin6_port, 0,
2811 raddr6->sin6_scope_id);
2814 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2815 err = -EHOSTUNREACH;
2819 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2821 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2825 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2826 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2829 state_set(&ep->com, CONNECTING);
2831 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2832 sizeof(ep->com.local_addr));
2833 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2834 sizeof(ep->com.remote_addr));
2836 /* send connect request to rnic */
2837 err = send_connect(ep);
2841 cxgb4_l2t_release(ep->l2t);
2843 dst_release(ep->dst);
2845 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2846 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2848 cm_id->rem_ref(cm_id);
2849 c4iw_put_ep(&ep->com);
2854 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2857 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2859 c4iw_init_wr_wait(&ep->com.wr_wait);
2860 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2861 ep->stid, &sin6->sin6_addr,
2863 ep->com.dev->rdev.lldi.rxq_ids[0]);
2865 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2869 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2871 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2875 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2878 struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2880 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2882 err = cxgb4_create_server_filter(
2883 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2884 sin->sin_addr.s_addr, sin->sin_port, 0,
2885 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2886 if (err == -EBUSY) {
2887 set_current_state(TASK_UNINTERRUPTIBLE);
2888 schedule_timeout(usecs_to_jiffies(100));
2890 } while (err == -EBUSY);
2892 c4iw_init_wr_wait(&ep->com.wr_wait);
2893 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2894 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2895 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2897 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2902 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2904 &sin->sin_addr, ntohs(sin->sin_port));
2908 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2911 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2912 struct c4iw_listen_ep *ep;
2916 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2918 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2922 PDBG("%s ep %p\n", __func__, ep);
2923 cm_id->add_ref(cm_id);
2924 ep->com.cm_id = cm_id;
2926 ep->backlog = backlog;
2927 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2928 sizeof(ep->com.local_addr));
2931 * Allocate a server TID.
2933 if (dev->rdev.lldi.enable_fw_ofld_conn &&
2934 ep->com.local_addr.ss_family == AF_INET)
2935 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2936 cm_id->local_addr.ss_family, ep);
2938 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2939 cm_id->local_addr.ss_family, ep);
2941 if (ep->stid == -1) {
2942 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2946 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2947 state_set(&ep->com, LISTEN);
2948 if (ep->com.local_addr.ss_family == AF_INET)
2949 err = create_server4(dev, ep);
2951 err = create_server6(dev, ep);
2953 cm_id->provider_data = ep;
2956 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2957 ep->com.local_addr.ss_family);
2959 cm_id->rem_ref(cm_id);
2960 c4iw_put_ep(&ep->com);
2966 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2969 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2971 PDBG("%s ep %p\n", __func__, ep);
2974 state_set(&ep->com, DEAD);
2975 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2976 ep->com.local_addr.ss_family == AF_INET) {
2977 err = cxgb4_remove_server_filter(
2978 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2979 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2981 c4iw_init_wr_wait(&ep->com.wr_wait);
2982 err = cxgb4_remove_server(
2983 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2984 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2987 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2990 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2991 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2992 ep->com.local_addr.ss_family);
2994 cm_id->rem_ref(cm_id);
2995 c4iw_put_ep(&ep->com);
2999 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3004 struct c4iw_rdev *rdev;
3006 mutex_lock(&ep->com.mutex);
3008 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3009 states[ep->com.state], abrupt);
3011 rdev = &ep->com.dev->rdev;
3012 if (c4iw_fatal_error(rdev)) {
3014 close_complete_upcall(ep, -EIO);
3015 ep->com.state = DEAD;
3017 switch (ep->com.state) {
3025 ep->com.state = ABORTING;
3027 ep->com.state = CLOSING;
3030 set_bit(CLOSE_SENT, &ep->com.flags);
3033 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3036 (void)stop_ep_timer(ep);
3037 ep->com.state = ABORTING;
3039 ep->com.state = MORIBUND;
3045 PDBG("%s ignoring disconnect ep %p state %u\n",
3046 __func__, ep, ep->com.state);
3055 set_bit(EP_DISC_ABORT, &ep->com.history);
3056 close_complete_upcall(ep, -ECONNRESET);
3057 ret = send_abort(ep, NULL, gfp);
3059 set_bit(EP_DISC_CLOSE, &ep->com.history);
3060 ret = send_halfclose(ep, gfp);
3065 mutex_unlock(&ep->com.mutex);
3067 release_ep_resources(ep);
3071 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3072 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3075 int atid = be32_to_cpu(req->tid);
3077 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3078 (__force u32) req->tid);
3082 switch (req->retval) {
3084 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3085 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3086 send_fw_act_open_req(ep, atid);
3090 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3091 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3092 send_fw_act_open_req(ep, atid);
3097 pr_info("%s unexpected ofld conn wr retval %d\n",
3098 __func__, req->retval);
3101 pr_err("active ofld_connect_wr failure %d atid %d\n",
3103 mutex_lock(&dev->rdev.stats.lock);
3104 dev->rdev.stats.act_ofld_conn_fails++;
3105 mutex_unlock(&dev->rdev.stats.lock);
3106 connect_reply_upcall(ep, status2errno(req->retval));
3107 state_set(&ep->com, DEAD);
3108 remove_handle(dev, &dev->atid_idr, atid);
3109 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3110 dst_release(ep->dst);
3111 cxgb4_l2t_release(ep->l2t);
3112 c4iw_put_ep(&ep->com);
3115 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3116 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3118 struct sk_buff *rpl_skb;
3119 struct cpl_pass_accept_req *cpl;
3122 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3125 PDBG("%s passive open failure %d\n", __func__, req->retval);
3126 mutex_lock(&dev->rdev.stats.lock);
3127 dev->rdev.stats.pas_ofld_conn_fails++;
3128 mutex_unlock(&dev->rdev.stats.lock);
3131 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3132 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3133 (__force u32) htonl(
3134 (__force u32) req->tid)));
3135 ret = pass_accept_req(dev, rpl_skb);
3142 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3144 struct cpl_fw6_msg *rpl = cplhdr(skb);
3145 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3147 switch (rpl->type) {
3149 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3151 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3152 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3153 switch (req->t_state) {
3155 active_ofld_conn_reply(dev, skb, req);
3158 passive_ofld_conn_reply(dev, skb, req);
3161 pr_err("%s unexpected ofld conn wr state %d\n",
3162 __func__, req->t_state);
3170 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3173 u16 vlantag, len, hdr_len, eth_hdr_len;
3175 struct cpl_rx_pkt *cpl = cplhdr(skb);
3176 struct cpl_pass_accept_req *req;
3177 struct tcp_options_received tmp_opt;
3178 struct c4iw_dev *dev;
3180 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3181 /* Store values from cpl_rx_pkt in temporary location. */
3182 vlantag = (__force u16) cpl->vlan;
3183 len = (__force u16) cpl->len;
3184 l2info = (__force u32) cpl->l2info;
3185 hdr_len = (__force u16) cpl->hdr_len;
3188 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3191 * We need to parse the TCP options from SYN packet.
3192 * to generate cpl_pass_accept_req.
3194 memset(&tmp_opt, 0, sizeof(tmp_opt));
3195 tcp_clear_options(&tmp_opt);
3196 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3198 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3199 memset(req, 0, sizeof(*req));
3200 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3201 V_SYN_MAC_IDX(G_RX_MACIDX(
3202 (__force int) htonl(l2info))) |
3204 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3205 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3206 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3207 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3208 (__force int) htonl(l2info))) |
3209 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3210 (__force int) htons(hdr_len))) |
3211 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3212 (__force int) htons(hdr_len))) |
3213 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3214 req->vlan = (__force __be16) vlantag;
3215 req->len = (__force __be16) len;
3216 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3217 PASS_OPEN_TOS(tos));
3218 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3219 if (tmp_opt.wscale_ok)
3220 req->tcpopt.wsf = tmp_opt.snd_wscale;
3221 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3222 if (tmp_opt.sack_ok)
3223 req->tcpopt.sack = 1;
3224 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3228 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3229 __be32 laddr, __be16 lport,
3230 __be32 raddr, __be16 rport,
3231 u32 rcv_isn, u32 filter, u16 window,
3232 u32 rss_qid, u8 port_id)
3234 struct sk_buff *req_skb;
3235 struct fw_ofld_connection_wr *req;
3236 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3239 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3240 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3241 memset(req, 0, sizeof(*req));
3242 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3243 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3244 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3245 req->le.filter = (__force __be32) filter;
3246 req->le.lport = lport;
3247 req->le.pport = rport;
3248 req->le.u.ipv4.lip = laddr;
3249 req->le.u.ipv4.pip = raddr;
3250 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3251 req->tcb.rcv_adv = htons(window);
3252 req->tcb.t_state_to_astid =
3253 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3254 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3255 V_FW_OFLD_CONNECTION_WR_ASTID(
3256 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3259 * We store the qid in opt2 which will be used by the firmware
3260 * to send us the wr response.
3262 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3265 * We initialize the MSS index in TCB to 0xF.
3266 * So that when driver sends cpl_pass_accept_rpl
3267 * TCB picks up the correct value. If this was 0
3268 * TP will ignore any value > 0 for MSS index.
3270 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3271 req->cookie = (unsigned long)skb;
3273 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3274 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3276 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3284 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3285 * messages when a filter is being used instead of server to
3286 * redirect a syn packet. When packets hit filter they are redirected
3287 * to the offload queue and driver tries to establish the connection
3288 * using firmware work request.
3290 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3293 unsigned int filter;
3294 struct ethhdr *eh = NULL;
3295 struct vlan_ethhdr *vlan_eh = NULL;
3297 struct tcphdr *tcph;
3298 struct rss_header *rss = (void *)skb->data;
3299 struct cpl_rx_pkt *cpl = (void *)skb->data;
3300 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3301 struct l2t_entry *e;
3302 struct dst_entry *dst;
3303 struct c4iw_ep *lep;
3305 struct port_info *pi;
3306 struct net_device *pdev;
3307 u16 rss_qid, eth_hdr_len;
3310 struct neighbour *neigh;
3312 /* Drop all non-SYN packets */
3313 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3317 * Drop all packets which did not hit the filter.
3318 * Unlikely to happen.
3320 if (!(rss->filter_hit && rss->filter_tid))
3324 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3326 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3328 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3330 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3334 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3335 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3336 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3337 if (eth_hdr_len == ETH_HLEN) {
3338 eh = (struct ethhdr *)(req + 1);
3339 iph = (struct iphdr *)(eh + 1);
3341 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3342 iph = (struct iphdr *)(vlan_eh + 1);
3343 skb->vlan_tci = ntohs(cpl->vlan);
3346 if (iph->version != 0x4)
3349 tcph = (struct tcphdr *)(iph + 1);
3350 skb_set_network_header(skb, (void *)iph - (void *)rss);
3351 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3354 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3355 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3356 ntohs(tcph->source), iph->tos);
3358 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3361 pr_err("%s - failed to find dst entry!\n",
3365 neigh = dst_neigh_lookup_skb(dst, skb);
3368 pr_err("%s - failed to allocate neigh!\n",
3373 if (neigh->dev->flags & IFF_LOOPBACK) {
3374 pdev = ip_dev_find(&init_net, iph->daddr);
3375 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3377 pi = (struct port_info *)netdev_priv(pdev);
3378 tx_chan = cxgb4_port_chan(pdev);
3381 pdev = get_real_dev(neigh->dev);
3382 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3384 pi = (struct port_info *)netdev_priv(pdev);
3385 tx_chan = cxgb4_port_chan(pdev);
3387 neigh_release(neigh);
3389 pr_err("%s - failed to allocate l2t entry!\n",
3394 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3395 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3396 window = (__force u16) htons((__force u16)tcph->window);
3398 /* Calcuate filter portion for LE region. */
3399 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3400 dev->rdev.lldi.ports[0],
3404 * Synthesize the cpl_pass_accept_req. We have everything except the
3405 * TID. Once firmware sends a reply with TID we update the TID field
3406 * in cpl and pass it through the regular cpl_pass_accept_req path.
3408 build_cpl_pass_accept_req(skb, stid, iph->tos);
3409 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3410 tcph->source, ntohl(tcph->seq), filter, window,
3411 rss_qid, pi->port_id);
3412 cxgb4_l2t_release(e);
3420 * These are the real handlers that are called from a
3423 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3424 [CPL_ACT_ESTABLISH] = act_establish,
3425 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3426 [CPL_RX_DATA] = rx_data,
3427 [CPL_ABORT_RPL_RSS] = abort_rpl,
3428 [CPL_ABORT_RPL] = abort_rpl,
3429 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3430 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3431 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3432 [CPL_PASS_ESTABLISH] = pass_establish,
3433 [CPL_PEER_CLOSE] = peer_close,
3434 [CPL_ABORT_REQ_RSS] = peer_abort,
3435 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3436 [CPL_RDMA_TERMINATE] = terminate,
3437 [CPL_FW4_ACK] = fw4_ack,
3438 [CPL_FW6_MSG] = deferred_fw6_msg,
3439 [CPL_RX_PKT] = rx_pkt
3442 static void process_timeout(struct c4iw_ep *ep)
3444 struct c4iw_qp_attributes attrs;
3447 mutex_lock(&ep->com.mutex);
3448 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3450 set_bit(TIMEDOUT, &ep->com.history);
3451 switch (ep->com.state) {
3453 __state_set(&ep->com, ABORTING);
3454 connect_reply_upcall(ep, -ETIMEDOUT);
3457 __state_set(&ep->com, ABORTING);
3461 if (ep->com.cm_id && ep->com.qp) {
3462 attrs.next_state = C4IW_QP_STATE_ERROR;
3463 c4iw_modify_qp(ep->com.qp->rhp,
3464 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3467 __state_set(&ep->com, ABORTING);
3468 close_complete_upcall(ep, -ETIMEDOUT);
3474 * These states are expected if the ep timed out at the same
3475 * time as another thread was calling stop_ep_timer().
3476 * So we silently do nothing for these states.
3481 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3482 __func__, ep, ep->hwtid, ep->com.state);
3485 mutex_unlock(&ep->com.mutex);
3487 abort_connection(ep, NULL, GFP_KERNEL);
3488 c4iw_put_ep(&ep->com);
3491 static void process_timedout_eps(void)
3495 spin_lock_irq(&timeout_lock);
3496 while (!list_empty(&timeout_list)) {
3497 struct list_head *tmp;
3499 tmp = timeout_list.next;
3503 spin_unlock_irq(&timeout_lock);
3504 ep = list_entry(tmp, struct c4iw_ep, entry);
3505 process_timeout(ep);
3506 spin_lock_irq(&timeout_lock);
3508 spin_unlock_irq(&timeout_lock);
3511 static void process_work(struct work_struct *work)
3513 struct sk_buff *skb = NULL;
3514 struct c4iw_dev *dev;
3515 struct cpl_act_establish *rpl;
3516 unsigned int opcode;
3519 process_timedout_eps();
3520 while ((skb = skb_dequeue(&rxq))) {
3522 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3523 opcode = rpl->ot.opcode;
3525 BUG_ON(!work_handlers[opcode]);
3526 ret = work_handlers[opcode](dev, skb);
3529 process_timedout_eps();
3533 static DECLARE_WORK(skb_work, process_work);
3535 static void ep_timeout(unsigned long arg)
3537 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3540 spin_lock(&timeout_lock);
3541 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3543 * Only insert if it is not already on the list.
3545 if (!ep->entry.next) {
3546 list_add_tail(&ep->entry, &timeout_list);
3550 spin_unlock(&timeout_lock);
3552 queue_work(workq, &skb_work);
3556 * All the CM events are handled on a work queue to have a safe context.
3558 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3562 * Save dev in the skb->cb area.
3564 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3567 * Queue the skb and schedule the worker thread.
3569 skb_queue_tail(&rxq, skb);
3570 queue_work(workq, &skb_work);
3574 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3576 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3578 if (rpl->status != CPL_ERR_NONE) {
3579 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3580 "for tid %u\n", rpl->status, GET_TID(rpl));
3586 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3588 struct cpl_fw6_msg *rpl = cplhdr(skb);
3589 struct c4iw_wr_wait *wr_waitp;
3592 PDBG("%s type %u\n", __func__, rpl->type);
3594 switch (rpl->type) {
3595 case FW6_TYPE_WR_RPL:
3596 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3597 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3598 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3600 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3604 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3608 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3616 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3618 struct cpl_abort_req_rss *req = cplhdr(skb);
3620 struct tid_info *t = dev->rdev.lldi.tids;
3621 unsigned int tid = GET_TID(req);
3623 ep = lookup_tid(t, tid);
3625 printk(KERN_WARNING MOD
3626 "Abort on non-existent endpoint, tid %d\n", tid);
3630 if (is_neg_adv(req->status)) {
3631 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3636 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3640 * Wake up any threads in rdma_init() or rdma_fini().
3641 * However, if we are on MPAv2 and want to retry with MPAv1
3642 * then, don't wake up yet.
3644 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3645 if (ep->com.state != MPA_REQ_SENT)
3646 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3648 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3654 * Most upcalls from the T4 Core go to sched() to
3655 * schedule the processing on a work queue.
3657 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3658 [CPL_ACT_ESTABLISH] = sched,
3659 [CPL_ACT_OPEN_RPL] = sched,
3660 [CPL_RX_DATA] = sched,
3661 [CPL_ABORT_RPL_RSS] = sched,
3662 [CPL_ABORT_RPL] = sched,
3663 [CPL_PASS_OPEN_RPL] = sched,
3664 [CPL_CLOSE_LISTSRV_RPL] = sched,
3665 [CPL_PASS_ACCEPT_REQ] = sched,
3666 [CPL_PASS_ESTABLISH] = sched,
3667 [CPL_PEER_CLOSE] = sched,
3668 [CPL_CLOSE_CON_RPL] = sched,
3669 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3670 [CPL_RDMA_TERMINATE] = sched,
3671 [CPL_FW4_ACK] = sched,
3672 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3673 [CPL_FW6_MSG] = fw6_msg,
3674 [CPL_RX_PKT] = sched
3677 int __init c4iw_cm_init(void)
3679 spin_lock_init(&timeout_lock);
3680 skb_queue_head_init(&rxq);
3682 workq = create_singlethread_workqueue("iw_cxgb4");
3689 void __exit c4iw_cm_term(void)
3691 WARN_ON(!list_empty(&timeout_list));
3692 flush_workqueue(workq);
3693 destroy_workqueue(workq);