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 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
591 opt2 |= T5_OPT_2_VALID;
592 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
594 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
596 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
597 if (ep->com.remote_addr.ss_family == AF_INET) {
598 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
600 OPCODE_TID(req) = cpu_to_be32(
601 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
602 ((ep->rss_qid << 14) | ep->atid)));
603 req->local_port = la->sin_port;
604 req->peer_port = ra->sin_port;
605 req->local_ip = la->sin_addr.s_addr;
606 req->peer_ip = ra->sin_addr.s_addr;
607 req->opt0 = cpu_to_be64(opt0);
608 req->params = cpu_to_be32(cxgb4_select_ntuple(
609 ep->com.dev->rdev.lldi.ports[0],
611 req->opt2 = cpu_to_be32(opt2);
613 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
616 OPCODE_TID(req6) = cpu_to_be32(
617 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
618 ((ep->rss_qid<<14)|ep->atid)));
619 req6->local_port = la6->sin6_port;
620 req6->peer_port = ra6->sin6_port;
621 req6->local_ip_hi = *((__be64 *)
622 (la6->sin6_addr.s6_addr));
623 req6->local_ip_lo = *((__be64 *)
624 (la6->sin6_addr.s6_addr + 8));
625 req6->peer_ip_hi = *((__be64 *)
626 (ra6->sin6_addr.s6_addr));
627 req6->peer_ip_lo = *((__be64 *)
628 (ra6->sin6_addr.s6_addr + 8));
629 req6->opt0 = cpu_to_be64(opt0);
630 req6->params = cpu_to_be32(cxgb4_select_ntuple(
631 ep->com.dev->rdev.lldi.ports[0],
633 req6->opt2 = cpu_to_be32(opt2);
636 if (ep->com.remote_addr.ss_family == AF_INET) {
637 t5_req = (struct cpl_t5_act_open_req *)
639 INIT_TP_WR(t5_req, 0);
640 OPCODE_TID(t5_req) = cpu_to_be32(
641 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
642 ((ep->rss_qid << 14) | ep->atid)));
643 t5_req->local_port = la->sin_port;
644 t5_req->peer_port = ra->sin_port;
645 t5_req->local_ip = la->sin_addr.s_addr;
646 t5_req->peer_ip = ra->sin_addr.s_addr;
647 t5_req->opt0 = cpu_to_be64(opt0);
648 t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
650 ep->com.dev->rdev.lldi.ports[0],
652 t5_req->opt2 = cpu_to_be32(opt2);
654 t5_req6 = (struct cpl_t5_act_open_req6 *)
656 INIT_TP_WR(t5_req6, 0);
657 OPCODE_TID(t5_req6) = cpu_to_be32(
658 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
659 ((ep->rss_qid<<14)|ep->atid)));
660 t5_req6->local_port = la6->sin6_port;
661 t5_req6->peer_port = ra6->sin6_port;
662 t5_req6->local_ip_hi = *((__be64 *)
663 (la6->sin6_addr.s6_addr));
664 t5_req6->local_ip_lo = *((__be64 *)
665 (la6->sin6_addr.s6_addr + 8));
666 t5_req6->peer_ip_hi = *((__be64 *)
667 (ra6->sin6_addr.s6_addr));
668 t5_req6->peer_ip_lo = *((__be64 *)
669 (ra6->sin6_addr.s6_addr + 8));
670 t5_req6->opt0 = cpu_to_be64(opt0);
671 t5_req6->params = (__force __be64)cpu_to_be32(
673 ep->com.dev->rdev.lldi.ports[0],
675 t5_req6->opt2 = cpu_to_be32(opt2);
679 set_bit(ACT_OPEN_REQ, &ep->com.history);
680 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
683 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
687 struct fw_ofld_tx_data_wr *req;
688 struct mpa_message *mpa;
689 struct mpa_v2_conn_params mpa_v2_params;
691 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
693 BUG_ON(skb_cloned(skb));
695 mpalen = sizeof(*mpa) + ep->plen;
696 if (mpa_rev_to_use == 2)
697 mpalen += sizeof(struct mpa_v2_conn_params);
698 wrlen = roundup(mpalen + sizeof *req, 16);
699 skb = get_skb(skb, wrlen, GFP_KERNEL);
701 connect_reply_upcall(ep, -ENOMEM);
704 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
706 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
707 memset(req, 0, wrlen);
708 req->op_to_immdlen = cpu_to_be32(
709 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
711 FW_WR_IMMDLEN(mpalen));
712 req->flowid_len16 = cpu_to_be32(
713 FW_WR_FLOWID(ep->hwtid) |
714 FW_WR_LEN16(wrlen >> 4));
715 req->plen = cpu_to_be32(mpalen);
716 req->tunnel_to_proxy = cpu_to_be32(
717 FW_OFLD_TX_DATA_WR_FLUSH(1) |
718 FW_OFLD_TX_DATA_WR_SHOVE(1));
720 mpa = (struct mpa_message *)(req + 1);
721 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
722 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
723 (markers_enabled ? MPA_MARKERS : 0) |
724 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
725 mpa->private_data_size = htons(ep->plen);
726 mpa->revision = mpa_rev_to_use;
727 if (mpa_rev_to_use == 1) {
728 ep->tried_with_mpa_v1 = 1;
729 ep->retry_with_mpa_v1 = 0;
732 if (mpa_rev_to_use == 2) {
733 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
734 sizeof (struct mpa_v2_conn_params));
735 mpa_v2_params.ird = htons((u16)ep->ird);
736 mpa_v2_params.ord = htons((u16)ep->ord);
739 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
740 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
742 htons(MPA_V2_RDMA_WRITE_RTR);
743 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
745 htons(MPA_V2_RDMA_READ_RTR);
747 memcpy(mpa->private_data, &mpa_v2_params,
748 sizeof(struct mpa_v2_conn_params));
751 memcpy(mpa->private_data +
752 sizeof(struct mpa_v2_conn_params),
753 ep->mpa_pkt + sizeof(*mpa), ep->plen);
756 memcpy(mpa->private_data,
757 ep->mpa_pkt + sizeof(*mpa), ep->plen);
760 * Reference the mpa skb. This ensures the data area
761 * will remain in memory until the hw acks the tx.
762 * Function fw4_ack() will deref it.
765 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
768 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
770 __state_set(&ep->com, MPA_REQ_SENT);
771 ep->mpa_attr.initiator = 1;
772 ep->snd_seq += mpalen;
776 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
779 struct fw_ofld_tx_data_wr *req;
780 struct mpa_message *mpa;
782 struct mpa_v2_conn_params mpa_v2_params;
784 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
786 mpalen = sizeof(*mpa) + plen;
787 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
788 mpalen += sizeof(struct mpa_v2_conn_params);
789 wrlen = roundup(mpalen + sizeof *req, 16);
791 skb = get_skb(NULL, wrlen, GFP_KERNEL);
793 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
796 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
798 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
799 memset(req, 0, wrlen);
800 req->op_to_immdlen = cpu_to_be32(
801 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
803 FW_WR_IMMDLEN(mpalen));
804 req->flowid_len16 = cpu_to_be32(
805 FW_WR_FLOWID(ep->hwtid) |
806 FW_WR_LEN16(wrlen >> 4));
807 req->plen = cpu_to_be32(mpalen);
808 req->tunnel_to_proxy = cpu_to_be32(
809 FW_OFLD_TX_DATA_WR_FLUSH(1) |
810 FW_OFLD_TX_DATA_WR_SHOVE(1));
812 mpa = (struct mpa_message *)(req + 1);
813 memset(mpa, 0, sizeof(*mpa));
814 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
815 mpa->flags = MPA_REJECT;
816 mpa->revision = ep->mpa_attr.version;
817 mpa->private_data_size = htons(plen);
819 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
820 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
821 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
822 sizeof (struct mpa_v2_conn_params));
823 mpa_v2_params.ird = htons(((u16)ep->ird) |
824 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
826 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
828 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
829 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
830 FW_RI_INIT_P2PTYPE_READ_REQ ?
831 MPA_V2_RDMA_READ_RTR : 0) : 0));
832 memcpy(mpa->private_data, &mpa_v2_params,
833 sizeof(struct mpa_v2_conn_params));
836 memcpy(mpa->private_data +
837 sizeof(struct mpa_v2_conn_params), pdata, plen);
840 memcpy(mpa->private_data, pdata, plen);
843 * Reference the mpa skb again. This ensures the data area
844 * will remain in memory until the hw acks the tx.
845 * Function fw4_ack() will deref it.
848 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
849 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
852 ep->snd_seq += mpalen;
853 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
856 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
859 struct fw_ofld_tx_data_wr *req;
860 struct mpa_message *mpa;
862 struct mpa_v2_conn_params mpa_v2_params;
864 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
866 mpalen = sizeof(*mpa) + plen;
867 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
868 mpalen += sizeof(struct mpa_v2_conn_params);
869 wrlen = roundup(mpalen + sizeof *req, 16);
871 skb = get_skb(NULL, wrlen, GFP_KERNEL);
873 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
876 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
878 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
879 memset(req, 0, wrlen);
880 req->op_to_immdlen = cpu_to_be32(
881 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
883 FW_WR_IMMDLEN(mpalen));
884 req->flowid_len16 = cpu_to_be32(
885 FW_WR_FLOWID(ep->hwtid) |
886 FW_WR_LEN16(wrlen >> 4));
887 req->plen = cpu_to_be32(mpalen);
888 req->tunnel_to_proxy = cpu_to_be32(
889 FW_OFLD_TX_DATA_WR_FLUSH(1) |
890 FW_OFLD_TX_DATA_WR_SHOVE(1));
892 mpa = (struct mpa_message *)(req + 1);
893 memset(mpa, 0, sizeof(*mpa));
894 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
895 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
896 (markers_enabled ? MPA_MARKERS : 0);
897 mpa->revision = ep->mpa_attr.version;
898 mpa->private_data_size = htons(plen);
900 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
901 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
902 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
903 sizeof (struct mpa_v2_conn_params));
904 mpa_v2_params.ird = htons((u16)ep->ird);
905 mpa_v2_params.ord = htons((u16)ep->ord);
906 if (peer2peer && (ep->mpa_attr.p2p_type !=
907 FW_RI_INIT_P2PTYPE_DISABLED)) {
908 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
910 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
912 htons(MPA_V2_RDMA_WRITE_RTR);
913 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
915 htons(MPA_V2_RDMA_READ_RTR);
918 memcpy(mpa->private_data, &mpa_v2_params,
919 sizeof(struct mpa_v2_conn_params));
922 memcpy(mpa->private_data +
923 sizeof(struct mpa_v2_conn_params), pdata, plen);
926 memcpy(mpa->private_data, pdata, plen);
929 * Reference the mpa skb. This ensures the data area
930 * will remain in memory until the hw acks the tx.
931 * Function fw4_ack() will deref it.
934 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
936 __state_set(&ep->com, MPA_REP_SENT);
937 ep->snd_seq += mpalen;
938 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
941 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
944 struct cpl_act_establish *req = cplhdr(skb);
945 unsigned int tid = GET_TID(req);
946 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
947 struct tid_info *t = dev->rdev.lldi.tids;
949 ep = lookup_atid(t, atid);
951 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
952 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
954 mutex_lock(&ep->com.mutex);
955 dst_confirm(ep->dst);
957 /* setup the hwtid for this connection */
959 cxgb4_insert_tid(t, ep, tid);
960 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
962 ep->snd_seq = be32_to_cpu(req->snd_isn);
963 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
965 set_emss(ep, ntohs(req->tcp_opt));
967 /* dealloc the atid */
968 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
969 cxgb4_free_atid(t, atid);
970 set_bit(ACT_ESTAB, &ep->com.history);
972 /* start MPA negotiation */
973 send_flowc(ep, NULL);
974 if (ep->retry_with_mpa_v1)
975 send_mpa_req(ep, skb, 1);
977 send_mpa_req(ep, skb, mpa_rev);
978 mutex_unlock(&ep->com.mutex);
982 static void close_complete_upcall(struct c4iw_ep *ep, int status)
984 struct iw_cm_event event;
986 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
987 memset(&event, 0, sizeof(event));
988 event.event = IW_CM_EVENT_CLOSE;
989 event.status = status;
991 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
992 ep, ep->com.cm_id, ep->hwtid);
993 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
994 ep->com.cm_id->rem_ref(ep->com.cm_id);
995 ep->com.cm_id = NULL;
996 set_bit(CLOSE_UPCALL, &ep->com.history);
1000 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1002 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1003 __state_set(&ep->com, ABORTING);
1004 set_bit(ABORT_CONN, &ep->com.history);
1005 return send_abort(ep, skb, gfp);
1008 static void peer_close_upcall(struct c4iw_ep *ep)
1010 struct iw_cm_event event;
1012 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1013 memset(&event, 0, sizeof(event));
1014 event.event = IW_CM_EVENT_DISCONNECT;
1015 if (ep->com.cm_id) {
1016 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1017 ep, ep->com.cm_id, ep->hwtid);
1018 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1019 set_bit(DISCONN_UPCALL, &ep->com.history);
1023 static void peer_abort_upcall(struct c4iw_ep *ep)
1025 struct iw_cm_event event;
1027 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1028 memset(&event, 0, sizeof(event));
1029 event.event = IW_CM_EVENT_CLOSE;
1030 event.status = -ECONNRESET;
1031 if (ep->com.cm_id) {
1032 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1033 ep->com.cm_id, ep->hwtid);
1034 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1035 ep->com.cm_id->rem_ref(ep->com.cm_id);
1036 ep->com.cm_id = NULL;
1037 set_bit(ABORT_UPCALL, &ep->com.history);
1041 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1043 struct iw_cm_event event;
1045 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1046 memset(&event, 0, sizeof(event));
1047 event.event = IW_CM_EVENT_CONNECT_REPLY;
1048 event.status = status;
1049 memcpy(&event.local_addr, &ep->com.local_addr,
1050 sizeof(ep->com.local_addr));
1051 memcpy(&event.remote_addr, &ep->com.remote_addr,
1052 sizeof(ep->com.remote_addr));
1054 if ((status == 0) || (status == -ECONNREFUSED)) {
1055 if (!ep->tried_with_mpa_v1) {
1056 /* this means MPA_v2 is used */
1057 event.private_data_len = ep->plen -
1058 sizeof(struct mpa_v2_conn_params);
1059 event.private_data = ep->mpa_pkt +
1060 sizeof(struct mpa_message) +
1061 sizeof(struct mpa_v2_conn_params);
1063 /* this means MPA_v1 is used */
1064 event.private_data_len = ep->plen;
1065 event.private_data = ep->mpa_pkt +
1066 sizeof(struct mpa_message);
1070 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1072 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1073 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1076 ep->com.cm_id->rem_ref(ep->com.cm_id);
1077 ep->com.cm_id = NULL;
1081 static int connect_request_upcall(struct c4iw_ep *ep)
1083 struct iw_cm_event event;
1086 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1087 memset(&event, 0, sizeof(event));
1088 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1089 memcpy(&event.local_addr, &ep->com.local_addr,
1090 sizeof(ep->com.local_addr));
1091 memcpy(&event.remote_addr, &ep->com.remote_addr,
1092 sizeof(ep->com.remote_addr));
1093 event.provider_data = ep;
1094 if (!ep->tried_with_mpa_v1) {
1095 /* this means MPA_v2 is used */
1096 event.ord = ep->ord;
1097 event.ird = ep->ird;
1098 event.private_data_len = ep->plen -
1099 sizeof(struct mpa_v2_conn_params);
1100 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1101 sizeof(struct mpa_v2_conn_params);
1103 /* this means MPA_v1 is used. Send max supported */
1104 event.ord = c4iw_max_read_depth;
1105 event.ird = c4iw_max_read_depth;
1106 event.private_data_len = ep->plen;
1107 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1109 c4iw_get_ep(&ep->com);
1110 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1113 c4iw_put_ep(&ep->com);
1114 set_bit(CONNREQ_UPCALL, &ep->com.history);
1115 c4iw_put_ep(&ep->parent_ep->com);
1119 static void established_upcall(struct c4iw_ep *ep)
1121 struct iw_cm_event event;
1123 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1124 memset(&event, 0, sizeof(event));
1125 event.event = IW_CM_EVENT_ESTABLISHED;
1126 event.ird = ep->ird;
1127 event.ord = ep->ord;
1128 if (ep->com.cm_id) {
1129 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1130 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1131 set_bit(ESTAB_UPCALL, &ep->com.history);
1135 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1137 struct cpl_rx_data_ack *req;
1138 struct sk_buff *skb;
1139 int wrlen = roundup(sizeof *req, 16);
1141 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1142 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1144 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1148 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1149 memset(req, 0, wrlen);
1150 INIT_TP_WR(req, ep->hwtid);
1151 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1153 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1155 V_RX_DACK_MODE(dack_mode));
1156 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1157 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1161 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1163 struct mpa_message *mpa;
1164 struct mpa_v2_conn_params *mpa_v2_params;
1166 u16 resp_ird, resp_ord;
1167 u8 rtr_mismatch = 0, insuff_ird = 0;
1168 struct c4iw_qp_attributes attrs;
1169 enum c4iw_qp_attr_mask mask;
1173 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1176 * Stop mpa timer. If it expired, then
1177 * we ignore the MPA reply. process_timeout()
1178 * will abort the connection.
1180 if (stop_ep_timer(ep))
1184 * If we get more than the supported amount of private data
1185 * then we must fail this connection.
1187 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1193 * copy the new data into our accumulation buffer.
1195 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1197 ep->mpa_pkt_len += skb->len;
1200 * if we don't even have the mpa message, then bail.
1202 if (ep->mpa_pkt_len < sizeof(*mpa))
1204 mpa = (struct mpa_message *) ep->mpa_pkt;
1206 /* Validate MPA header. */
1207 if (mpa->revision > mpa_rev) {
1208 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1209 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1213 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1218 plen = ntohs(mpa->private_data_size);
1221 * Fail if there's too much private data.
1223 if (plen > MPA_MAX_PRIVATE_DATA) {
1229 * If plen does not account for pkt size
1231 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1236 ep->plen = (u8) plen;
1239 * If we don't have all the pdata yet, then bail.
1240 * We'll continue process when more data arrives.
1242 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1245 if (mpa->flags & MPA_REJECT) {
1246 err = -ECONNREFUSED;
1251 * If we get here we have accumulated the entire mpa
1252 * start reply message including private data. And
1253 * the MPA header is valid.
1255 __state_set(&ep->com, FPDU_MODE);
1256 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1257 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1258 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1259 ep->mpa_attr.version = mpa->revision;
1260 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1262 if (mpa->revision == 2) {
1263 ep->mpa_attr.enhanced_rdma_conn =
1264 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1265 if (ep->mpa_attr.enhanced_rdma_conn) {
1266 mpa_v2_params = (struct mpa_v2_conn_params *)
1267 (ep->mpa_pkt + sizeof(*mpa));
1268 resp_ird = ntohs(mpa_v2_params->ird) &
1269 MPA_V2_IRD_ORD_MASK;
1270 resp_ord = ntohs(mpa_v2_params->ord) &
1271 MPA_V2_IRD_ORD_MASK;
1274 * This is a double-check. Ideally, below checks are
1275 * not required since ird/ord stuff has been taken
1276 * care of in c4iw_accept_cr
1278 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1285 if (ntohs(mpa_v2_params->ird) &
1286 MPA_V2_PEER2PEER_MODEL) {
1287 if (ntohs(mpa_v2_params->ord) &
1288 MPA_V2_RDMA_WRITE_RTR)
1289 ep->mpa_attr.p2p_type =
1290 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1291 else if (ntohs(mpa_v2_params->ord) &
1292 MPA_V2_RDMA_READ_RTR)
1293 ep->mpa_attr.p2p_type =
1294 FW_RI_INIT_P2PTYPE_READ_REQ;
1297 } else if (mpa->revision == 1)
1299 ep->mpa_attr.p2p_type = p2p_type;
1301 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1302 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1303 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1304 ep->mpa_attr.recv_marker_enabled,
1305 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1306 ep->mpa_attr.p2p_type, p2p_type);
1309 * If responder's RTR does not match with that of initiator, assign
1310 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1311 * generated when moving QP to RTS state.
1312 * A TERM message will be sent after QP has moved to RTS state
1314 if ((ep->mpa_attr.version == 2) && peer2peer &&
1315 (ep->mpa_attr.p2p_type != p2p_type)) {
1316 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1320 attrs.mpa_attr = ep->mpa_attr;
1321 attrs.max_ird = ep->ird;
1322 attrs.max_ord = ep->ord;
1323 attrs.llp_stream_handle = ep;
1324 attrs.next_state = C4IW_QP_STATE_RTS;
1326 mask = C4IW_QP_ATTR_NEXT_STATE |
1327 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1328 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1330 /* bind QP and TID with INIT_WR */
1331 err = c4iw_modify_qp(ep->com.qp->rhp,
1332 ep->com.qp, mask, &attrs, 1);
1337 * If responder's RTR requirement did not match with what initiator
1338 * supports, generate TERM message
1341 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1342 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1343 attrs.ecode = MPA_NOMATCH_RTR;
1344 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1345 attrs.send_term = 1;
1346 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1347 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1354 * Generate TERM if initiator IRD is not sufficient for responder
1355 * provided ORD. Currently, we do the same behaviour even when
1356 * responder provided IRD is also not sufficient as regards to
1360 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1362 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1363 attrs.ecode = MPA_INSUFF_IRD;
1364 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1365 attrs.send_term = 1;
1366 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1367 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1374 __state_set(&ep->com, ABORTING);
1375 send_abort(ep, skb, GFP_KERNEL);
1377 connect_reply_upcall(ep, err);
1381 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1383 struct mpa_message *mpa;
1384 struct mpa_v2_conn_params *mpa_v2_params;
1387 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1390 * If we get more than the supported amount of private data
1391 * then we must fail this connection.
1393 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1394 (void)stop_ep_timer(ep);
1395 abort_connection(ep, skb, GFP_KERNEL);
1399 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1402 * Copy the new data into our accumulation buffer.
1404 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1406 ep->mpa_pkt_len += skb->len;
1409 * If we don't even have the mpa message, then bail.
1410 * We'll continue process when more data arrives.
1412 if (ep->mpa_pkt_len < sizeof(*mpa))
1415 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1416 mpa = (struct mpa_message *) ep->mpa_pkt;
1419 * Validate MPA Header.
1421 if (mpa->revision > mpa_rev) {
1422 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1423 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1424 (void)stop_ep_timer(ep);
1425 abort_connection(ep, skb, GFP_KERNEL);
1429 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1430 (void)stop_ep_timer(ep);
1431 abort_connection(ep, skb, GFP_KERNEL);
1435 plen = ntohs(mpa->private_data_size);
1438 * Fail if there's too much private data.
1440 if (plen > MPA_MAX_PRIVATE_DATA) {
1441 (void)stop_ep_timer(ep);
1442 abort_connection(ep, skb, GFP_KERNEL);
1447 * If plen does not account for pkt size
1449 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1450 (void)stop_ep_timer(ep);
1451 abort_connection(ep, skb, GFP_KERNEL);
1454 ep->plen = (u8) plen;
1457 * If we don't have all the pdata yet, then bail.
1459 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1463 * If we get here we have accumulated the entire mpa
1464 * start reply message including private data.
1466 ep->mpa_attr.initiator = 0;
1467 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1468 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1469 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1470 ep->mpa_attr.version = mpa->revision;
1471 if (mpa->revision == 1)
1472 ep->tried_with_mpa_v1 = 1;
1473 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1475 if (mpa->revision == 2) {
1476 ep->mpa_attr.enhanced_rdma_conn =
1477 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1478 if (ep->mpa_attr.enhanced_rdma_conn) {
1479 mpa_v2_params = (struct mpa_v2_conn_params *)
1480 (ep->mpa_pkt + sizeof(*mpa));
1481 ep->ird = ntohs(mpa_v2_params->ird) &
1482 MPA_V2_IRD_ORD_MASK;
1483 ep->ord = ntohs(mpa_v2_params->ord) &
1484 MPA_V2_IRD_ORD_MASK;
1485 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1487 if (ntohs(mpa_v2_params->ord) &
1488 MPA_V2_RDMA_WRITE_RTR)
1489 ep->mpa_attr.p2p_type =
1490 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1491 else if (ntohs(mpa_v2_params->ord) &
1492 MPA_V2_RDMA_READ_RTR)
1493 ep->mpa_attr.p2p_type =
1494 FW_RI_INIT_P2PTYPE_READ_REQ;
1497 } else if (mpa->revision == 1)
1499 ep->mpa_attr.p2p_type = p2p_type;
1501 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1502 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1503 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1504 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1505 ep->mpa_attr.p2p_type);
1508 * If the endpoint timer already expired, then we ignore
1509 * the start request. process_timeout() will abort
1512 if (!stop_ep_timer(ep)) {
1513 __state_set(&ep->com, MPA_REQ_RCVD);
1516 mutex_lock(&ep->parent_ep->com.mutex);
1517 if (ep->parent_ep->com.state != DEAD) {
1518 if (connect_request_upcall(ep))
1519 abort_connection(ep, skb, GFP_KERNEL);
1521 abort_connection(ep, skb, GFP_KERNEL);
1523 mutex_unlock(&ep->parent_ep->com.mutex);
1528 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1531 struct cpl_rx_data *hdr = cplhdr(skb);
1532 unsigned int dlen = ntohs(hdr->len);
1533 unsigned int tid = GET_TID(hdr);
1534 struct tid_info *t = dev->rdev.lldi.tids;
1535 __u8 status = hdr->status;
1538 ep = lookup_tid(t, tid);
1541 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1542 skb_pull(skb, sizeof(*hdr));
1543 skb_trim(skb, dlen);
1544 mutex_lock(&ep->com.mutex);
1546 /* update RX credits */
1547 update_rx_credits(ep, dlen);
1549 switch (ep->com.state) {
1551 ep->rcv_seq += dlen;
1552 disconnect = process_mpa_reply(ep, skb);
1555 ep->rcv_seq += dlen;
1556 process_mpa_request(ep, skb);
1559 struct c4iw_qp_attributes attrs;
1560 BUG_ON(!ep->com.qp);
1562 pr_err("%s Unexpected streaming data." \
1563 " qpid %u ep %p state %d tid %u status %d\n",
1564 __func__, ep->com.qp->wq.sq.qid, ep,
1565 ep->com.state, ep->hwtid, status);
1566 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1567 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1568 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1575 mutex_unlock(&ep->com.mutex);
1577 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1581 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1584 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1586 unsigned int tid = GET_TID(rpl);
1587 struct tid_info *t = dev->rdev.lldi.tids;
1589 ep = lookup_tid(t, tid);
1591 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1594 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1595 mutex_lock(&ep->com.mutex);
1596 switch (ep->com.state) {
1598 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1599 __state_set(&ep->com, DEAD);
1603 printk(KERN_ERR "%s ep %p state %d\n",
1604 __func__, ep, ep->com.state);
1607 mutex_unlock(&ep->com.mutex);
1610 release_ep_resources(ep);
1614 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1616 struct sk_buff *skb;
1617 struct fw_ofld_connection_wr *req;
1618 unsigned int mtu_idx;
1620 struct sockaddr_in *sin;
1622 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1623 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1624 memset(req, 0, sizeof(*req));
1625 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1626 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1627 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1628 ep->com.dev->rdev.lldi.ports[0],
1630 sin = (struct sockaddr_in *)&ep->com.local_addr;
1631 req->le.lport = sin->sin_port;
1632 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1633 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1634 req->le.pport = sin->sin_port;
1635 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1636 req->tcb.t_state_to_astid =
1637 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1638 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1639 req->tcb.cplrxdataack_cplpassacceptrpl =
1640 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1641 req->tcb.tx_max = (__force __be32) jiffies;
1642 req->tcb.rcv_adv = htons(1);
1643 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1644 wscale = compute_wscale(rcv_win);
1645 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1646 (nocong ? NO_CONG(1) : 0) |
1651 L2T_IDX(ep->l2t->idx) |
1652 TX_CHAN(ep->tx_chan) |
1653 SMAC_SEL(ep->smac_idx) |
1655 ULP_MODE(ULP_MODE_TCPDDP) |
1656 RCV_BUFSIZ(rcv_win >> 10));
1657 req->tcb.opt2 = (__force __be32) (PACE(1) |
1658 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1660 CCTRL_ECN(enable_ecn) |
1661 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1662 if (enable_tcp_timestamps)
1663 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1664 if (enable_tcp_sack)
1665 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1666 if (wscale && enable_tcp_window_scaling)
1667 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1668 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1669 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1670 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1671 set_bit(ACT_OFLD_CONN, &ep->com.history);
1672 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1676 * Return whether a failed active open has allocated a TID
1678 static inline int act_open_has_tid(int status)
1680 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1681 status != CPL_ERR_ARP_MISS;
1684 /* Returns whether a CPL status conveys negative advice.
1686 static int is_neg_adv(unsigned int status)
1688 return status == CPL_ERR_RTX_NEG_ADVICE ||
1689 status == CPL_ERR_PERSIST_NEG_ADVICE ||
1690 status == CPL_ERR_KEEPALV_NEG_ADVICE;
1693 #define ACT_OPEN_RETRY_COUNT 2
1695 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1696 struct dst_entry *dst, struct c4iw_dev *cdev,
1699 struct neighbour *n;
1701 struct net_device *pdev;
1703 n = dst_neigh_lookup(dst, peer_ip);
1709 if (n->dev->flags & IFF_LOOPBACK) {
1711 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1712 else if (IS_ENABLED(CONFIG_IPV6))
1713 for_each_netdev(&init_net, pdev) {
1714 if (ipv6_chk_addr(&init_net,
1715 (struct in6_addr *)peer_ip,
1726 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1730 ep->mtu = pdev->mtu;
1731 ep->tx_chan = cxgb4_port_chan(pdev);
1732 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1733 step = cdev->rdev.lldi.ntxq /
1734 cdev->rdev.lldi.nchan;
1735 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1736 step = cdev->rdev.lldi.nrxq /
1737 cdev->rdev.lldi.nchan;
1738 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1739 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1740 cxgb4_port_idx(pdev) * step];
1743 pdev = get_real_dev(n->dev);
1744 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1748 ep->mtu = dst_mtu(dst);
1749 ep->tx_chan = cxgb4_port_chan(n->dev);
1750 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1751 step = cdev->rdev.lldi.ntxq /
1752 cdev->rdev.lldi.nchan;
1753 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1754 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1755 step = cdev->rdev.lldi.nrxq /
1756 cdev->rdev.lldi.nchan;
1757 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1758 cxgb4_port_idx(n->dev) * step];
1761 ep->retry_with_mpa_v1 = 0;
1762 ep->tried_with_mpa_v1 = 0;
1774 static int c4iw_reconnect(struct c4iw_ep *ep)
1777 struct sockaddr_in *laddr = (struct sockaddr_in *)
1778 &ep->com.cm_id->local_addr;
1779 struct sockaddr_in *raddr = (struct sockaddr_in *)
1780 &ep->com.cm_id->remote_addr;
1781 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1782 &ep->com.cm_id->local_addr;
1783 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1784 &ep->com.cm_id->remote_addr;
1788 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1789 init_timer(&ep->timer);
1792 * Allocate an active TID to initiate a TCP connection.
1794 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1795 if (ep->atid == -1) {
1796 pr_err("%s - cannot alloc atid.\n", __func__);
1800 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1803 if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1804 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1805 raddr->sin_addr.s_addr, laddr->sin_port,
1806 raddr->sin_port, 0);
1808 ra = (__u8 *)&raddr->sin_addr;
1810 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1811 raddr6->sin6_addr.s6_addr,
1812 laddr6->sin6_port, raddr6->sin6_port, 0,
1813 raddr6->sin6_scope_id);
1815 ra = (__u8 *)&raddr6->sin6_addr;
1818 pr_err("%s - cannot find route.\n", __func__);
1819 err = -EHOSTUNREACH;
1822 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1824 pr_err("%s - cannot alloc l2e.\n", __func__);
1828 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1829 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1832 state_set(&ep->com, CONNECTING);
1835 /* send connect request to rnic */
1836 err = send_connect(ep);
1840 cxgb4_l2t_release(ep->l2t);
1842 dst_release(ep->dst);
1844 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1845 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1848 * remember to send notification to upper layer.
1849 * We are in here so the upper layer is not aware that this is
1850 * re-connect attempt and so, upper layer is still waiting for
1851 * response of 1st connect request.
1853 connect_reply_upcall(ep, -ECONNRESET);
1854 c4iw_put_ep(&ep->com);
1859 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1862 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1863 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1864 ntohl(rpl->atid_status)));
1865 struct tid_info *t = dev->rdev.lldi.tids;
1866 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1867 struct sockaddr_in *la;
1868 struct sockaddr_in *ra;
1869 struct sockaddr_in6 *la6;
1870 struct sockaddr_in6 *ra6;
1872 ep = lookup_atid(t, atid);
1873 la = (struct sockaddr_in *)&ep->com.local_addr;
1874 ra = (struct sockaddr_in *)&ep->com.remote_addr;
1875 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1876 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1878 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1879 status, status2errno(status));
1881 if (is_neg_adv(status)) {
1882 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1887 set_bit(ACT_OPEN_RPL, &ep->com.history);
1890 * Log interesting failures.
1893 case CPL_ERR_CONN_RESET:
1894 case CPL_ERR_CONN_TIMEDOUT:
1896 case CPL_ERR_TCAM_FULL:
1897 mutex_lock(&dev->rdev.stats.lock);
1898 dev->rdev.stats.tcam_full++;
1899 mutex_unlock(&dev->rdev.stats.lock);
1900 if (ep->com.local_addr.ss_family == AF_INET &&
1901 dev->rdev.lldi.enable_fw_ofld_conn) {
1902 send_fw_act_open_req(ep,
1903 GET_TID_TID(GET_AOPEN_ATID(
1904 ntohl(rpl->atid_status))));
1908 case CPL_ERR_CONN_EXIST:
1909 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1910 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1911 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1913 cxgb4_free_atid(t, atid);
1914 dst_release(ep->dst);
1915 cxgb4_l2t_release(ep->l2t);
1921 if (ep->com.local_addr.ss_family == AF_INET) {
1922 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1923 atid, status, status2errno(status),
1924 &la->sin_addr.s_addr, ntohs(la->sin_port),
1925 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
1927 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1928 atid, status, status2errno(status),
1929 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1930 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1935 connect_reply_upcall(ep, status2errno(status));
1936 state_set(&ep->com, DEAD);
1938 if (status && act_open_has_tid(status))
1939 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1941 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1942 cxgb4_free_atid(t, atid);
1943 dst_release(ep->dst);
1944 cxgb4_l2t_release(ep->l2t);
1945 c4iw_put_ep(&ep->com);
1950 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1952 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1953 struct tid_info *t = dev->rdev.lldi.tids;
1954 unsigned int stid = GET_TID(rpl);
1955 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1958 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1961 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1962 rpl->status, status2errno(rpl->status));
1963 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1969 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1971 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1972 struct tid_info *t = dev->rdev.lldi.tids;
1973 unsigned int stid = GET_TID(rpl);
1974 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1976 PDBG("%s ep %p\n", __func__, ep);
1977 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1981 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1982 struct cpl_pass_accept_req *req)
1984 struct cpl_pass_accept_rpl *rpl;
1985 unsigned int mtu_idx;
1990 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1991 BUG_ON(skb_cloned(skb));
1992 skb_trim(skb, sizeof(*rpl));
1994 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1995 wscale = compute_wscale(rcv_win);
1996 opt0 = (nocong ? NO_CONG(1) : 0) |
2001 L2T_IDX(ep->l2t->idx) |
2002 TX_CHAN(ep->tx_chan) |
2003 SMAC_SEL(ep->smac_idx) |
2004 DSCP(ep->tos >> 2) |
2005 ULP_MODE(ULP_MODE_TCPDDP) |
2006 RCV_BUFSIZ(rcv_win>>10);
2007 opt2 = RX_CHANNEL(0) |
2008 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
2010 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2011 opt2 |= TSTAMPS_EN(1);
2012 if (enable_tcp_sack && req->tcpopt.sack)
2014 if (wscale && enable_tcp_window_scaling)
2015 opt2 |= WND_SCALE_EN(1);
2017 const struct tcphdr *tcph;
2018 u32 hlen = ntohl(req->hdr_len);
2020 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2022 if (tcph->ece && tcph->cwr)
2023 opt2 |= CCTRL_ECN(1);
2025 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
2026 opt2 |= T5_OPT_2_VALID;
2027 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
2031 INIT_TP_WR(rpl, ep->hwtid);
2032 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2034 rpl->opt0 = cpu_to_be64(opt0);
2035 rpl->opt2 = cpu_to_be32(opt2);
2036 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2037 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2038 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2043 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2045 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2046 BUG_ON(skb_cloned(skb));
2047 skb_trim(skb, sizeof(struct cpl_tid_release));
2049 release_tid(&dev->rdev, hwtid, skb);
2053 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2054 __u8 *local_ip, __u8 *peer_ip,
2055 __be16 *local_port, __be16 *peer_port)
2057 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2058 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2059 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2060 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2061 struct tcphdr *tcp = (struct tcphdr *)
2062 ((u8 *)(req + 1) + eth_len + ip_len);
2064 if (ip->version == 4) {
2065 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2066 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2069 memcpy(peer_ip, &ip->saddr, 4);
2070 memcpy(local_ip, &ip->daddr, 4);
2072 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2073 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2076 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2077 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2079 *peer_port = tcp->source;
2080 *local_port = tcp->dest;
2085 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2087 struct c4iw_ep *child_ep = NULL, *parent_ep;
2088 struct cpl_pass_accept_req *req = cplhdr(skb);
2089 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2090 struct tid_info *t = dev->rdev.lldi.tids;
2091 unsigned int hwtid = GET_TID(req);
2092 struct dst_entry *dst;
2093 __u8 local_ip[16], peer_ip[16];
2094 __be16 local_port, peer_port;
2096 u16 peer_mss = ntohs(req->tcpopt.mss);
2099 parent_ep = lookup_stid(t, stid);
2101 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2105 if (state_read(&parent_ep->com) != LISTEN) {
2106 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2111 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2113 /* Find output route */
2115 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2116 , __func__, parent_ep, hwtid,
2117 local_ip, peer_ip, ntohs(local_port),
2118 ntohs(peer_port), peer_mss);
2119 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2120 local_port, peer_port,
2121 GET_POPEN_TOS(ntohl(req->tos_stid)));
2123 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2124 , __func__, parent_ep, hwtid,
2125 local_ip, peer_ip, ntohs(local_port),
2126 ntohs(peer_port), peer_mss);
2127 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2128 PASS_OPEN_TOS(ntohl(req->tos_stid)),
2129 ((struct sockaddr_in6 *)
2130 &parent_ep->com.local_addr)->sin6_scope_id);
2133 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2138 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2140 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2146 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2148 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2155 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2156 child_ep->mtu = peer_mss + 40;
2158 state_set(&child_ep->com, CONNECTING);
2159 child_ep->com.dev = dev;
2160 child_ep->com.cm_id = NULL;
2162 struct sockaddr_in *sin = (struct sockaddr_in *)
2163 &child_ep->com.local_addr;
2164 sin->sin_family = PF_INET;
2165 sin->sin_port = local_port;
2166 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2167 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2168 sin->sin_family = PF_INET;
2169 sin->sin_port = peer_port;
2170 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2172 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2173 &child_ep->com.local_addr;
2174 sin6->sin6_family = PF_INET6;
2175 sin6->sin6_port = local_port;
2176 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2177 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2178 sin6->sin6_family = PF_INET6;
2179 sin6->sin6_port = peer_port;
2180 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2182 c4iw_get_ep(&parent_ep->com);
2183 child_ep->parent_ep = parent_ep;
2184 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2185 child_ep->dst = dst;
2186 child_ep->hwtid = hwtid;
2188 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2189 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2191 init_timer(&child_ep->timer);
2192 cxgb4_insert_tid(t, child_ep, hwtid);
2193 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2194 accept_cr(child_ep, skb, req);
2195 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2198 reject_cr(dev, hwtid, skb);
2203 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2206 struct cpl_pass_establish *req = cplhdr(skb);
2207 struct tid_info *t = dev->rdev.lldi.tids;
2208 unsigned int tid = GET_TID(req);
2210 ep = lookup_tid(t, tid);
2211 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2212 ep->snd_seq = be32_to_cpu(req->snd_isn);
2213 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2215 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2216 ntohs(req->tcp_opt));
2218 set_emss(ep, ntohs(req->tcp_opt));
2220 dst_confirm(ep->dst);
2221 state_set(&ep->com, MPA_REQ_WAIT);
2223 send_flowc(ep, skb);
2224 set_bit(PASS_ESTAB, &ep->com.history);
2229 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2231 struct cpl_peer_close *hdr = cplhdr(skb);
2233 struct c4iw_qp_attributes attrs;
2236 struct tid_info *t = dev->rdev.lldi.tids;
2237 unsigned int tid = GET_TID(hdr);
2240 ep = lookup_tid(t, tid);
2241 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2242 dst_confirm(ep->dst);
2244 set_bit(PEER_CLOSE, &ep->com.history);
2245 mutex_lock(&ep->com.mutex);
2246 switch (ep->com.state) {
2248 __state_set(&ep->com, CLOSING);
2251 __state_set(&ep->com, CLOSING);
2252 connect_reply_upcall(ep, -ECONNRESET);
2257 * We're gonna mark this puppy DEAD, but keep
2258 * the reference on it until the ULP accepts or
2259 * rejects the CR. Also wake up anyone waiting
2260 * in rdma connection migration (see c4iw_accept_cr()).
2262 __state_set(&ep->com, CLOSING);
2263 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2264 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2267 __state_set(&ep->com, CLOSING);
2268 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2269 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2273 __state_set(&ep->com, CLOSING);
2274 attrs.next_state = C4IW_QP_STATE_CLOSING;
2275 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2276 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2277 if (ret != -ECONNRESET) {
2278 peer_close_upcall(ep);
2286 __state_set(&ep->com, MORIBUND);
2290 (void)stop_ep_timer(ep);
2291 if (ep->com.cm_id && ep->com.qp) {
2292 attrs.next_state = C4IW_QP_STATE_IDLE;
2293 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2294 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2296 close_complete_upcall(ep, 0);
2297 __state_set(&ep->com, DEAD);
2307 mutex_unlock(&ep->com.mutex);
2309 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2311 release_ep_resources(ep);
2315 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2317 struct cpl_abort_req_rss *req = cplhdr(skb);
2319 struct cpl_abort_rpl *rpl;
2320 struct sk_buff *rpl_skb;
2321 struct c4iw_qp_attributes attrs;
2324 struct tid_info *t = dev->rdev.lldi.tids;
2325 unsigned int tid = GET_TID(req);
2327 ep = lookup_tid(t, tid);
2328 if (is_neg_adv(req->status)) {
2329 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2333 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2335 set_bit(PEER_ABORT, &ep->com.history);
2338 * Wake up any threads in rdma_init() or rdma_fini().
2339 * However, this is not needed if com state is just
2342 if (ep->com.state != MPA_REQ_SENT)
2343 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2345 mutex_lock(&ep->com.mutex);
2346 switch (ep->com.state) {
2350 (void)stop_ep_timer(ep);
2353 (void)stop_ep_timer(ep);
2354 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2355 connect_reply_upcall(ep, -ECONNRESET);
2358 * we just don't send notification upwards because we
2359 * want to retry with mpa_v1 without upper layers even
2362 * do some housekeeping so as to re-initiate the
2365 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2367 ep->retry_with_mpa_v1 = 1;
2379 if (ep->com.cm_id && ep->com.qp) {
2380 attrs.next_state = C4IW_QP_STATE_ERROR;
2381 ret = c4iw_modify_qp(ep->com.qp->rhp,
2382 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2386 "%s - qp <- error failed!\n",
2389 peer_abort_upcall(ep);
2394 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2395 mutex_unlock(&ep->com.mutex);
2401 dst_confirm(ep->dst);
2402 if (ep->com.state != ABORTING) {
2403 __state_set(&ep->com, DEAD);
2404 /* we don't release if we want to retry with mpa_v1 */
2405 if (!ep->retry_with_mpa_v1)
2408 mutex_unlock(&ep->com.mutex);
2410 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2412 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2417 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2418 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2419 INIT_TP_WR(rpl, ep->hwtid);
2420 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2421 rpl->cmd = CPL_ABORT_NO_RST;
2422 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2425 release_ep_resources(ep);
2426 else if (ep->retry_with_mpa_v1) {
2427 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2428 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2429 dst_release(ep->dst);
2430 cxgb4_l2t_release(ep->l2t);
2437 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2440 struct c4iw_qp_attributes attrs;
2441 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2443 struct tid_info *t = dev->rdev.lldi.tids;
2444 unsigned int tid = GET_TID(rpl);
2446 ep = lookup_tid(t, tid);
2448 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2451 /* The cm_id may be null if we failed to connect */
2452 mutex_lock(&ep->com.mutex);
2453 switch (ep->com.state) {
2455 __state_set(&ep->com, MORIBUND);
2458 (void)stop_ep_timer(ep);
2459 if ((ep->com.cm_id) && (ep->com.qp)) {
2460 attrs.next_state = C4IW_QP_STATE_IDLE;
2461 c4iw_modify_qp(ep->com.qp->rhp,
2463 C4IW_QP_ATTR_NEXT_STATE,
2466 close_complete_upcall(ep, 0);
2467 __state_set(&ep->com, DEAD);
2477 mutex_unlock(&ep->com.mutex);
2479 release_ep_resources(ep);
2483 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2485 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2486 struct tid_info *t = dev->rdev.lldi.tids;
2487 unsigned int tid = GET_TID(rpl);
2489 struct c4iw_qp_attributes attrs;
2491 ep = lookup_tid(t, tid);
2494 if (ep && ep->com.qp) {
2495 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2496 ep->com.qp->wq.sq.qid);
2497 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2498 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2499 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2501 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2507 * Upcall from the adapter indicating data has been transmitted.
2508 * For us its just the single MPA request or reply. We can now free
2509 * the skb holding the mpa message.
2511 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2514 struct cpl_fw4_ack *hdr = cplhdr(skb);
2515 u8 credits = hdr->credits;
2516 unsigned int tid = GET_TID(hdr);
2517 struct tid_info *t = dev->rdev.lldi.tids;
2520 ep = lookup_tid(t, tid);
2521 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2523 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2524 __func__, ep, ep->hwtid, state_read(&ep->com));
2528 dst_confirm(ep->dst);
2530 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2531 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2532 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2533 kfree_skb(ep->mpa_skb);
2539 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2543 struct c4iw_ep *ep = to_ep(cm_id);
2544 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2546 mutex_lock(&ep->com.mutex);
2547 if (ep->com.state == DEAD) {
2548 mutex_unlock(&ep->com.mutex);
2549 c4iw_put_ep(&ep->com);
2552 set_bit(ULP_REJECT, &ep->com.history);
2553 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2555 abort_connection(ep, NULL, GFP_KERNEL);
2557 err = send_mpa_reject(ep, pdata, pdata_len);
2560 mutex_unlock(&ep->com.mutex);
2562 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2563 c4iw_put_ep(&ep->com);
2567 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2570 struct c4iw_qp_attributes attrs;
2571 enum c4iw_qp_attr_mask mask;
2572 struct c4iw_ep *ep = to_ep(cm_id);
2573 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2574 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2576 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2578 mutex_lock(&ep->com.mutex);
2579 if (ep->com.state == DEAD) {
2584 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2587 set_bit(ULP_ACCEPT, &ep->com.history);
2588 if ((conn_param->ord > c4iw_max_read_depth) ||
2589 (conn_param->ird > c4iw_max_read_depth)) {
2590 abort_connection(ep, NULL, GFP_KERNEL);
2595 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2596 if (conn_param->ord > ep->ird) {
2597 ep->ird = conn_param->ird;
2598 ep->ord = conn_param->ord;
2599 send_mpa_reject(ep, conn_param->private_data,
2600 conn_param->private_data_len);
2601 abort_connection(ep, NULL, GFP_KERNEL);
2605 if (conn_param->ird > ep->ord) {
2607 conn_param->ird = 1;
2609 abort_connection(ep, NULL, GFP_KERNEL);
2616 ep->ird = conn_param->ird;
2617 ep->ord = conn_param->ord;
2619 if (ep->mpa_attr.version != 2)
2620 if (peer2peer && ep->ird == 0)
2623 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2625 cm_id->add_ref(cm_id);
2626 ep->com.cm_id = cm_id;
2630 /* bind QP to EP and move to RTS */
2631 attrs.mpa_attr = ep->mpa_attr;
2632 attrs.max_ird = ep->ird;
2633 attrs.max_ord = ep->ord;
2634 attrs.llp_stream_handle = ep;
2635 attrs.next_state = C4IW_QP_STATE_RTS;
2637 /* bind QP and TID with INIT_WR */
2638 mask = C4IW_QP_ATTR_NEXT_STATE |
2639 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2640 C4IW_QP_ATTR_MPA_ATTR |
2641 C4IW_QP_ATTR_MAX_IRD |
2642 C4IW_QP_ATTR_MAX_ORD;
2644 err = c4iw_modify_qp(ep->com.qp->rhp,
2645 ep->com.qp, mask, &attrs, 1);
2648 err = send_mpa_reply(ep, conn_param->private_data,
2649 conn_param->private_data_len);
2653 __state_set(&ep->com, FPDU_MODE);
2654 established_upcall(ep);
2655 mutex_unlock(&ep->com.mutex);
2656 c4iw_put_ep(&ep->com);
2659 ep->com.cm_id = NULL;
2660 cm_id->rem_ref(cm_id);
2662 mutex_unlock(&ep->com.mutex);
2663 c4iw_put_ep(&ep->com);
2667 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2669 struct in_device *ind;
2671 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2672 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2674 ind = in_dev_get(dev->rdev.lldi.ports[0]);
2676 return -EADDRNOTAVAIL;
2677 for_primary_ifa(ind) {
2678 laddr->sin_addr.s_addr = ifa->ifa_address;
2679 raddr->sin_addr.s_addr = ifa->ifa_address;
2685 return found ? 0 : -EADDRNOTAVAIL;
2688 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2689 unsigned char banned_flags)
2691 struct inet6_dev *idev;
2692 int err = -EADDRNOTAVAIL;
2695 idev = __in6_dev_get(dev);
2697 struct inet6_ifaddr *ifp;
2699 read_lock_bh(&idev->lock);
2700 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2701 if (ifp->scope == IFA_LINK &&
2702 !(ifp->flags & banned_flags)) {
2703 memcpy(addr, &ifp->addr, 16);
2708 read_unlock_bh(&idev->lock);
2714 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2716 struct in6_addr uninitialized_var(addr);
2717 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2718 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2720 if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2721 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2722 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2725 return -EADDRNOTAVAIL;
2728 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2730 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2733 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2734 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2735 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2736 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2737 &cm_id->remote_addr;
2741 if ((conn_param->ord > c4iw_max_read_depth) ||
2742 (conn_param->ird > c4iw_max_read_depth)) {
2746 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2748 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2752 init_timer(&ep->timer);
2753 ep->plen = conn_param->private_data_len;
2755 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2756 conn_param->private_data, ep->plen);
2757 ep->ird = conn_param->ird;
2758 ep->ord = conn_param->ord;
2760 if (peer2peer && ep->ord == 0)
2763 cm_id->add_ref(cm_id);
2765 ep->com.cm_id = cm_id;
2766 ep->com.qp = get_qhp(dev, conn_param->qpn);
2768 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2773 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2777 * Allocate an active TID to initiate a TCP connection.
2779 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2780 if (ep->atid == -1) {
2781 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2785 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2787 if (cm_id->remote_addr.ss_family == AF_INET) {
2789 ra = (__u8 *)&raddr->sin_addr;
2792 * Handle loopback requests to INADDR_ANY.
2794 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2795 err = pick_local_ipaddrs(dev, cm_id);
2801 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2802 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2803 ra, ntohs(raddr->sin_port));
2804 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2805 raddr->sin_addr.s_addr, laddr->sin_port,
2806 raddr->sin_port, 0);
2809 ra = (__u8 *)&raddr6->sin6_addr;
2812 * Handle loopback requests to INADDR_ANY.
2814 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2815 err = pick_local_ip6addrs(dev, cm_id);
2821 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2822 __func__, laddr6->sin6_addr.s6_addr,
2823 ntohs(laddr6->sin6_port),
2824 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2825 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2826 raddr6->sin6_addr.s6_addr,
2827 laddr6->sin6_port, raddr6->sin6_port, 0,
2828 raddr6->sin6_scope_id);
2831 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2832 err = -EHOSTUNREACH;
2836 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2838 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2842 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2843 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2846 state_set(&ep->com, CONNECTING);
2848 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2849 sizeof(ep->com.local_addr));
2850 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2851 sizeof(ep->com.remote_addr));
2853 /* send connect request to rnic */
2854 err = send_connect(ep);
2858 cxgb4_l2t_release(ep->l2t);
2860 dst_release(ep->dst);
2862 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2863 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2865 cm_id->rem_ref(cm_id);
2866 c4iw_put_ep(&ep->com);
2871 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2874 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2876 c4iw_init_wr_wait(&ep->com.wr_wait);
2877 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2878 ep->stid, &sin6->sin6_addr,
2880 ep->com.dev->rdev.lldi.rxq_ids[0]);
2882 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2886 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2888 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2892 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2895 struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2897 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2899 err = cxgb4_create_server_filter(
2900 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2901 sin->sin_addr.s_addr, sin->sin_port, 0,
2902 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2903 if (err == -EBUSY) {
2904 set_current_state(TASK_UNINTERRUPTIBLE);
2905 schedule_timeout(usecs_to_jiffies(100));
2907 } while (err == -EBUSY);
2909 c4iw_init_wr_wait(&ep->com.wr_wait);
2910 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2911 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2912 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2914 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2919 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2921 &sin->sin_addr, ntohs(sin->sin_port));
2925 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2928 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2929 struct c4iw_listen_ep *ep;
2933 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2935 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2939 PDBG("%s ep %p\n", __func__, ep);
2940 cm_id->add_ref(cm_id);
2941 ep->com.cm_id = cm_id;
2943 ep->backlog = backlog;
2944 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2945 sizeof(ep->com.local_addr));
2948 * Allocate a server TID.
2950 if (dev->rdev.lldi.enable_fw_ofld_conn &&
2951 ep->com.local_addr.ss_family == AF_INET)
2952 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2953 cm_id->local_addr.ss_family, ep);
2955 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2956 cm_id->local_addr.ss_family, ep);
2958 if (ep->stid == -1) {
2959 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2963 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2964 state_set(&ep->com, LISTEN);
2965 if (ep->com.local_addr.ss_family == AF_INET)
2966 err = create_server4(dev, ep);
2968 err = create_server6(dev, ep);
2970 cm_id->provider_data = ep;
2973 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2974 ep->com.local_addr.ss_family);
2976 cm_id->rem_ref(cm_id);
2977 c4iw_put_ep(&ep->com);
2983 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2986 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2988 PDBG("%s ep %p\n", __func__, ep);
2991 state_set(&ep->com, DEAD);
2992 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2993 ep->com.local_addr.ss_family == AF_INET) {
2994 err = cxgb4_remove_server_filter(
2995 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2996 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2998 c4iw_init_wr_wait(&ep->com.wr_wait);
2999 err = cxgb4_remove_server(
3000 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3001 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3004 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3007 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3008 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3009 ep->com.local_addr.ss_family);
3011 cm_id->rem_ref(cm_id);
3012 c4iw_put_ep(&ep->com);
3016 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3021 struct c4iw_rdev *rdev;
3023 mutex_lock(&ep->com.mutex);
3025 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3026 states[ep->com.state], abrupt);
3028 rdev = &ep->com.dev->rdev;
3029 if (c4iw_fatal_error(rdev)) {
3031 close_complete_upcall(ep, -EIO);
3032 ep->com.state = DEAD;
3034 switch (ep->com.state) {
3042 ep->com.state = ABORTING;
3044 ep->com.state = CLOSING;
3047 set_bit(CLOSE_SENT, &ep->com.flags);
3050 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3053 (void)stop_ep_timer(ep);
3054 ep->com.state = ABORTING;
3056 ep->com.state = MORIBUND;
3062 PDBG("%s ignoring disconnect ep %p state %u\n",
3063 __func__, ep, ep->com.state);
3072 set_bit(EP_DISC_ABORT, &ep->com.history);
3073 close_complete_upcall(ep, -ECONNRESET);
3074 ret = send_abort(ep, NULL, gfp);
3076 set_bit(EP_DISC_CLOSE, &ep->com.history);
3077 ret = send_halfclose(ep, gfp);
3082 mutex_unlock(&ep->com.mutex);
3084 release_ep_resources(ep);
3088 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3089 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3092 int atid = be32_to_cpu(req->tid);
3094 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3095 (__force u32) req->tid);
3099 switch (req->retval) {
3101 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3102 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3103 send_fw_act_open_req(ep, atid);
3107 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3108 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3109 send_fw_act_open_req(ep, atid);
3114 pr_info("%s unexpected ofld conn wr retval %d\n",
3115 __func__, req->retval);
3118 pr_err("active ofld_connect_wr failure %d atid %d\n",
3120 mutex_lock(&dev->rdev.stats.lock);
3121 dev->rdev.stats.act_ofld_conn_fails++;
3122 mutex_unlock(&dev->rdev.stats.lock);
3123 connect_reply_upcall(ep, status2errno(req->retval));
3124 state_set(&ep->com, DEAD);
3125 remove_handle(dev, &dev->atid_idr, atid);
3126 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3127 dst_release(ep->dst);
3128 cxgb4_l2t_release(ep->l2t);
3129 c4iw_put_ep(&ep->com);
3132 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3133 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3135 struct sk_buff *rpl_skb;
3136 struct cpl_pass_accept_req *cpl;
3139 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3142 PDBG("%s passive open failure %d\n", __func__, req->retval);
3143 mutex_lock(&dev->rdev.stats.lock);
3144 dev->rdev.stats.pas_ofld_conn_fails++;
3145 mutex_unlock(&dev->rdev.stats.lock);
3148 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3149 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3150 (__force u32) htonl(
3151 (__force u32) req->tid)));
3152 ret = pass_accept_req(dev, rpl_skb);
3159 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3161 struct cpl_fw6_msg *rpl = cplhdr(skb);
3162 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3164 switch (rpl->type) {
3166 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3168 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3169 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3170 switch (req->t_state) {
3172 active_ofld_conn_reply(dev, skb, req);
3175 passive_ofld_conn_reply(dev, skb, req);
3178 pr_err("%s unexpected ofld conn wr state %d\n",
3179 __func__, req->t_state);
3187 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3190 u16 vlantag, len, hdr_len, eth_hdr_len;
3192 struct cpl_rx_pkt *cpl = cplhdr(skb);
3193 struct cpl_pass_accept_req *req;
3194 struct tcp_options_received tmp_opt;
3195 struct c4iw_dev *dev;
3197 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3198 /* Store values from cpl_rx_pkt in temporary location. */
3199 vlantag = (__force u16) cpl->vlan;
3200 len = (__force u16) cpl->len;
3201 l2info = (__force u32) cpl->l2info;
3202 hdr_len = (__force u16) cpl->hdr_len;
3205 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3208 * We need to parse the TCP options from SYN packet.
3209 * to generate cpl_pass_accept_req.
3211 memset(&tmp_opt, 0, sizeof(tmp_opt));
3212 tcp_clear_options(&tmp_opt);
3213 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3215 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3216 memset(req, 0, sizeof(*req));
3217 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3218 V_SYN_MAC_IDX(G_RX_MACIDX(
3219 (__force int) htonl(l2info))) |
3221 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3222 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3223 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3224 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3225 (__force int) htonl(l2info))) |
3226 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3227 (__force int) htons(hdr_len))) |
3228 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3229 (__force int) htons(hdr_len))) |
3230 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3231 req->vlan = (__force __be16) vlantag;
3232 req->len = (__force __be16) len;
3233 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3234 PASS_OPEN_TOS(tos));
3235 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3236 if (tmp_opt.wscale_ok)
3237 req->tcpopt.wsf = tmp_opt.snd_wscale;
3238 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3239 if (tmp_opt.sack_ok)
3240 req->tcpopt.sack = 1;
3241 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3245 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3246 __be32 laddr, __be16 lport,
3247 __be32 raddr, __be16 rport,
3248 u32 rcv_isn, u32 filter, u16 window,
3249 u32 rss_qid, u8 port_id)
3251 struct sk_buff *req_skb;
3252 struct fw_ofld_connection_wr *req;
3253 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3256 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3257 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3258 memset(req, 0, sizeof(*req));
3259 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3260 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3261 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3262 req->le.filter = (__force __be32) filter;
3263 req->le.lport = lport;
3264 req->le.pport = rport;
3265 req->le.u.ipv4.lip = laddr;
3266 req->le.u.ipv4.pip = raddr;
3267 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3268 req->tcb.rcv_adv = htons(window);
3269 req->tcb.t_state_to_astid =
3270 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3271 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3272 V_FW_OFLD_CONNECTION_WR_ASTID(
3273 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3276 * We store the qid in opt2 which will be used by the firmware
3277 * to send us the wr response.
3279 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3282 * We initialize the MSS index in TCB to 0xF.
3283 * So that when driver sends cpl_pass_accept_rpl
3284 * TCB picks up the correct value. If this was 0
3285 * TP will ignore any value > 0 for MSS index.
3287 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3288 req->cookie = (unsigned long)skb;
3290 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3291 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3293 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3301 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3302 * messages when a filter is being used instead of server to
3303 * redirect a syn packet. When packets hit filter they are redirected
3304 * to the offload queue and driver tries to establish the connection
3305 * using firmware work request.
3307 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3310 unsigned int filter;
3311 struct ethhdr *eh = NULL;
3312 struct vlan_ethhdr *vlan_eh = NULL;
3314 struct tcphdr *tcph;
3315 struct rss_header *rss = (void *)skb->data;
3316 struct cpl_rx_pkt *cpl = (void *)skb->data;
3317 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3318 struct l2t_entry *e;
3319 struct dst_entry *dst;
3320 struct c4iw_ep *lep;
3322 struct port_info *pi;
3323 struct net_device *pdev;
3324 u16 rss_qid, eth_hdr_len;
3327 struct neighbour *neigh;
3329 /* Drop all non-SYN packets */
3330 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3334 * Drop all packets which did not hit the filter.
3335 * Unlikely to happen.
3337 if (!(rss->filter_hit && rss->filter_tid))
3341 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3343 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3345 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3347 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3351 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3352 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3353 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3354 if (eth_hdr_len == ETH_HLEN) {
3355 eh = (struct ethhdr *)(req + 1);
3356 iph = (struct iphdr *)(eh + 1);
3358 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3359 iph = (struct iphdr *)(vlan_eh + 1);
3360 skb->vlan_tci = ntohs(cpl->vlan);
3363 if (iph->version != 0x4)
3366 tcph = (struct tcphdr *)(iph + 1);
3367 skb_set_network_header(skb, (void *)iph - (void *)rss);
3368 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3371 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3372 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3373 ntohs(tcph->source), iph->tos);
3375 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3378 pr_err("%s - failed to find dst entry!\n",
3382 neigh = dst_neigh_lookup_skb(dst, skb);
3385 pr_err("%s - failed to allocate neigh!\n",
3390 if (neigh->dev->flags & IFF_LOOPBACK) {
3391 pdev = ip_dev_find(&init_net, iph->daddr);
3392 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3394 pi = (struct port_info *)netdev_priv(pdev);
3395 tx_chan = cxgb4_port_chan(pdev);
3398 pdev = get_real_dev(neigh->dev);
3399 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3401 pi = (struct port_info *)netdev_priv(pdev);
3402 tx_chan = cxgb4_port_chan(pdev);
3404 neigh_release(neigh);
3406 pr_err("%s - failed to allocate l2t entry!\n",
3411 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3412 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3413 window = (__force u16) htons((__force u16)tcph->window);
3415 /* Calcuate filter portion for LE region. */
3416 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3417 dev->rdev.lldi.ports[0],
3421 * Synthesize the cpl_pass_accept_req. We have everything except the
3422 * TID. Once firmware sends a reply with TID we update the TID field
3423 * in cpl and pass it through the regular cpl_pass_accept_req path.
3425 build_cpl_pass_accept_req(skb, stid, iph->tos);
3426 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3427 tcph->source, ntohl(tcph->seq), filter, window,
3428 rss_qid, pi->port_id);
3429 cxgb4_l2t_release(e);
3437 * These are the real handlers that are called from a
3440 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3441 [CPL_ACT_ESTABLISH] = act_establish,
3442 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3443 [CPL_RX_DATA] = rx_data,
3444 [CPL_ABORT_RPL_RSS] = abort_rpl,
3445 [CPL_ABORT_RPL] = abort_rpl,
3446 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3447 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3448 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3449 [CPL_PASS_ESTABLISH] = pass_establish,
3450 [CPL_PEER_CLOSE] = peer_close,
3451 [CPL_ABORT_REQ_RSS] = peer_abort,
3452 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3453 [CPL_RDMA_TERMINATE] = terminate,
3454 [CPL_FW4_ACK] = fw4_ack,
3455 [CPL_FW6_MSG] = deferred_fw6_msg,
3456 [CPL_RX_PKT] = rx_pkt
3459 static void process_timeout(struct c4iw_ep *ep)
3461 struct c4iw_qp_attributes attrs;
3464 mutex_lock(&ep->com.mutex);
3465 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3467 set_bit(TIMEDOUT, &ep->com.history);
3468 switch (ep->com.state) {
3470 __state_set(&ep->com, ABORTING);
3471 connect_reply_upcall(ep, -ETIMEDOUT);
3474 __state_set(&ep->com, ABORTING);
3478 if (ep->com.cm_id && ep->com.qp) {
3479 attrs.next_state = C4IW_QP_STATE_ERROR;
3480 c4iw_modify_qp(ep->com.qp->rhp,
3481 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3484 __state_set(&ep->com, ABORTING);
3485 close_complete_upcall(ep, -ETIMEDOUT);
3491 * These states are expected if the ep timed out at the same
3492 * time as another thread was calling stop_ep_timer().
3493 * So we silently do nothing for these states.
3498 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3499 __func__, ep, ep->hwtid, ep->com.state);
3503 abort_connection(ep, NULL, GFP_KERNEL);
3504 mutex_unlock(&ep->com.mutex);
3505 c4iw_put_ep(&ep->com);
3508 static void process_timedout_eps(void)
3512 spin_lock_irq(&timeout_lock);
3513 while (!list_empty(&timeout_list)) {
3514 struct list_head *tmp;
3516 tmp = timeout_list.next;
3520 spin_unlock_irq(&timeout_lock);
3521 ep = list_entry(tmp, struct c4iw_ep, entry);
3522 process_timeout(ep);
3523 spin_lock_irq(&timeout_lock);
3525 spin_unlock_irq(&timeout_lock);
3528 static void process_work(struct work_struct *work)
3530 struct sk_buff *skb = NULL;
3531 struct c4iw_dev *dev;
3532 struct cpl_act_establish *rpl;
3533 unsigned int opcode;
3536 process_timedout_eps();
3537 while ((skb = skb_dequeue(&rxq))) {
3539 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3540 opcode = rpl->ot.opcode;
3542 BUG_ON(!work_handlers[opcode]);
3543 ret = work_handlers[opcode](dev, skb);
3546 process_timedout_eps();
3550 static DECLARE_WORK(skb_work, process_work);
3552 static void ep_timeout(unsigned long arg)
3554 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3557 spin_lock(&timeout_lock);
3558 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3560 * Only insert if it is not already on the list.
3562 if (!ep->entry.next) {
3563 list_add_tail(&ep->entry, &timeout_list);
3567 spin_unlock(&timeout_lock);
3569 queue_work(workq, &skb_work);
3573 * All the CM events are handled on a work queue to have a safe context.
3575 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3579 * Save dev in the skb->cb area.
3581 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3584 * Queue the skb and schedule the worker thread.
3586 skb_queue_tail(&rxq, skb);
3587 queue_work(workq, &skb_work);
3591 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3593 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3595 if (rpl->status != CPL_ERR_NONE) {
3596 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3597 "for tid %u\n", rpl->status, GET_TID(rpl));
3603 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3605 struct cpl_fw6_msg *rpl = cplhdr(skb);
3606 struct c4iw_wr_wait *wr_waitp;
3609 PDBG("%s type %u\n", __func__, rpl->type);
3611 switch (rpl->type) {
3612 case FW6_TYPE_WR_RPL:
3613 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3614 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3615 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3617 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3621 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3625 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3633 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3635 struct cpl_abort_req_rss *req = cplhdr(skb);
3637 struct tid_info *t = dev->rdev.lldi.tids;
3638 unsigned int tid = GET_TID(req);
3640 ep = lookup_tid(t, tid);
3642 printk(KERN_WARNING MOD
3643 "Abort on non-existent endpoint, tid %d\n", tid);
3647 if (is_neg_adv(req->status)) {
3648 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3653 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3657 * Wake up any threads in rdma_init() or rdma_fini().
3658 * However, if we are on MPAv2 and want to retry with MPAv1
3659 * then, don't wake up yet.
3661 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3662 if (ep->com.state != MPA_REQ_SENT)
3663 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3665 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3671 * Most upcalls from the T4 Core go to sched() to
3672 * schedule the processing on a work queue.
3674 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3675 [CPL_ACT_ESTABLISH] = sched,
3676 [CPL_ACT_OPEN_RPL] = sched,
3677 [CPL_RX_DATA] = sched,
3678 [CPL_ABORT_RPL_RSS] = sched,
3679 [CPL_ABORT_RPL] = sched,
3680 [CPL_PASS_OPEN_RPL] = sched,
3681 [CPL_CLOSE_LISTSRV_RPL] = sched,
3682 [CPL_PASS_ACCEPT_REQ] = sched,
3683 [CPL_PASS_ESTABLISH] = sched,
3684 [CPL_PEER_CLOSE] = sched,
3685 [CPL_CLOSE_CON_RPL] = sched,
3686 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3687 [CPL_RDMA_TERMINATE] = sched,
3688 [CPL_FW4_ACK] = sched,
3689 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3690 [CPL_FW6_MSG] = fw6_msg,
3691 [CPL_RX_PKT] = sched
3694 int __init c4iw_cm_init(void)
3696 spin_lock_init(&timeout_lock);
3697 skb_queue_head_init(&rxq);
3699 workq = create_singlethread_workqueue("iw_cxgb4");
3706 void __exit c4iw_cm_term(void)
3708 WARN_ON(!list_empty(&timeout_list));
3709 flush_workqueue(workq);
3710 destroy_workqueue(workq);