2 * Copyright (c) 2009-2014 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>
50 #include <rdma/ib_addr.h>
55 static char *states[] = {
72 module_param(nocong, int, 0644);
73 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75 static int enable_ecn;
76 module_param(enable_ecn, int, 0644);
77 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79 static int dack_mode = 1;
80 module_param(dack_mode, int, 0644);
81 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
83 uint c4iw_max_read_depth = 32;
84 module_param(c4iw_max_read_depth, int, 0644);
85 MODULE_PARM_DESC(c4iw_max_read_depth,
86 "Per-connection max ORD/IRD (default=32)");
88 static int enable_tcp_timestamps;
89 module_param(enable_tcp_timestamps, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92 static int enable_tcp_sack;
93 module_param(enable_tcp_sack, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96 static int enable_tcp_window_scaling = 1;
97 module_param(enable_tcp_window_scaling, int, 0644);
98 MODULE_PARM_DESC(enable_tcp_window_scaling,
99 "Enable tcp window scaling (default=1)");
102 module_param(c4iw_debug, int, 0644);
103 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
105 static int peer2peer = 1;
106 module_param(peer2peer, int, 0644);
107 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
109 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
110 module_param(p2p_type, int, 0644);
111 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
112 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
114 static int ep_timeout_secs = 60;
115 module_param(ep_timeout_secs, int, 0644);
116 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
117 "in seconds (default=60)");
119 static int mpa_rev = 2;
120 module_param(mpa_rev, int, 0644);
121 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
122 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
123 " compliant (default=2)");
125 static int markers_enabled;
126 module_param(markers_enabled, int, 0644);
127 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
129 static int crc_enabled = 1;
130 module_param(crc_enabled, int, 0644);
131 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
133 static int rcv_win = 256 * 1024;
134 module_param(rcv_win, int, 0644);
135 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
137 static int snd_win = 128 * 1024;
138 module_param(snd_win, int, 0644);
139 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
141 static struct workqueue_struct *workq;
143 static struct sk_buff_head rxq;
145 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
146 static void ep_timeout(unsigned long arg);
147 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
148 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
150 static LIST_HEAD(timeout_list);
151 static spinlock_t timeout_lock;
153 static void deref_cm_id(struct c4iw_ep_common *epc)
155 epc->cm_id->rem_ref(epc->cm_id);
157 set_bit(CM_ID_DEREFED, &epc->history);
160 static void ref_cm_id(struct c4iw_ep_common *epc)
162 set_bit(CM_ID_REFED, &epc->history);
163 epc->cm_id->add_ref(epc->cm_id);
166 static void deref_qp(struct c4iw_ep *ep)
168 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
169 clear_bit(QP_REFERENCED, &ep->com.flags);
170 set_bit(QP_DEREFED, &ep->com.history);
173 static void ref_qp(struct c4iw_ep *ep)
175 set_bit(QP_REFERENCED, &ep->com.flags);
176 set_bit(QP_REFED, &ep->com.history);
177 c4iw_qp_add_ref(&ep->com.qp->ibqp);
180 static void start_ep_timer(struct c4iw_ep *ep)
182 PDBG("%s ep %p\n", __func__, ep);
183 if (timer_pending(&ep->timer)) {
184 pr_err("%s timer already started! ep %p\n",
188 clear_bit(TIMEOUT, &ep->com.flags);
189 c4iw_get_ep(&ep->com);
190 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
191 ep->timer.data = (unsigned long)ep;
192 ep->timer.function = ep_timeout;
193 add_timer(&ep->timer);
196 static int stop_ep_timer(struct c4iw_ep *ep)
198 PDBG("%s ep %p stopping\n", __func__, ep);
199 del_timer_sync(&ep->timer);
200 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
201 c4iw_put_ep(&ep->com);
207 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
208 struct l2t_entry *l2e)
212 if (c4iw_fatal_error(rdev)) {
214 PDBG("%s - device in error state - dropping\n", __func__);
217 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
220 else if (error == NET_XMIT_DROP)
222 return error < 0 ? error : 0;
225 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
229 if (c4iw_fatal_error(rdev)) {
231 PDBG("%s - device in error state - dropping\n", __func__);
234 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
237 return error < 0 ? error : 0;
240 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
242 struct cpl_tid_release *req;
244 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
247 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
248 INIT_TP_WR(req, hwtid);
249 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
250 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
251 c4iw_ofld_send(rdev, skb);
255 static void set_emss(struct c4iw_ep *ep, u16 opt)
257 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
258 ((AF_INET == ep->com.remote_addr.ss_family) ?
259 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
260 sizeof(struct tcphdr);
262 if (TCPOPT_TSTAMP_G(opt))
263 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
267 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
268 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
269 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
273 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
275 enum c4iw_ep_state state;
277 mutex_lock(&epc->mutex);
279 mutex_unlock(&epc->mutex);
283 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
288 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
290 mutex_lock(&epc->mutex);
291 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
292 __state_set(epc, new);
293 mutex_unlock(&epc->mutex);
297 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
303 len = roundup(sizeof(union cpl_wr_size), 16);
304 for (i = 0; i < size; i++) {
305 skb = alloc_skb(len, GFP_KERNEL);
308 skb_queue_tail(ep_skb_list, skb);
312 skb_queue_purge(ep_skb_list);
316 static void *alloc_ep(int size, gfp_t gfp)
318 struct c4iw_ep_common *epc;
320 epc = kzalloc(size, gfp);
322 kref_init(&epc->kref);
323 mutex_init(&epc->mutex);
324 c4iw_init_wr_wait(&epc->wr_wait);
326 PDBG("%s alloc ep %p\n", __func__, epc);
330 static void remove_ep_tid(struct c4iw_ep *ep)
334 spin_lock_irqsave(&ep->com.dev->lock, flags);
335 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
336 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
339 static void insert_ep_tid(struct c4iw_ep *ep)
343 spin_lock_irqsave(&ep->com.dev->lock, flags);
344 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
345 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
349 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
351 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
356 spin_lock_irqsave(&dev->lock, flags);
357 ep = idr_find(&dev->hwtid_idr, tid);
359 c4iw_get_ep(&ep->com);
360 spin_unlock_irqrestore(&dev->lock, flags);
365 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
367 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
370 struct c4iw_listen_ep *ep;
373 spin_lock_irqsave(&dev->lock, flags);
374 ep = idr_find(&dev->stid_idr, stid);
376 c4iw_get_ep(&ep->com);
377 spin_unlock_irqrestore(&dev->lock, flags);
381 void _c4iw_free_ep(struct kref *kref)
385 ep = container_of(kref, struct c4iw_ep, com.kref);
386 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
387 if (test_bit(QP_REFERENCED, &ep->com.flags))
389 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
390 if (ep->com.remote_addr.ss_family == AF_INET6) {
391 struct sockaddr_in6 *sin6 =
392 (struct sockaddr_in6 *)
396 ep->com.dev->rdev.lldi.ports[0],
397 (const u32 *)&sin6->sin6_addr.s6_addr,
400 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
401 dst_release(ep->dst);
402 cxgb4_l2t_release(ep->l2t);
404 kfree_skb(ep->mpa_skb);
406 if (!skb_queue_empty(&ep->com.ep_skb_list))
407 skb_queue_purge(&ep->com.ep_skb_list);
411 static void release_ep_resources(struct c4iw_ep *ep)
413 set_bit(RELEASE_RESOURCES, &ep->com.flags);
416 * If we have a hwtid, then remove it from the idr table
417 * so lookups will no longer find this endpoint. Otherwise
418 * we have a race where one thread finds the ep ptr just
419 * before the other thread is freeing the ep memory.
423 c4iw_put_ep(&ep->com);
426 static int status2errno(int status)
431 case CPL_ERR_CONN_RESET:
433 case CPL_ERR_ARP_MISS:
434 return -EHOSTUNREACH;
435 case CPL_ERR_CONN_TIMEDOUT:
437 case CPL_ERR_TCAM_FULL:
439 case CPL_ERR_CONN_EXIST:
447 * Try and reuse skbs already allocated...
449 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
451 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
454 skb_reset_transport_header(skb);
456 skb = alloc_skb(len, gfp);
458 t4_set_arp_err_handler(skb, NULL, NULL);
462 static struct net_device *get_real_dev(struct net_device *egress_dev)
464 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
467 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
471 egress_dev = get_real_dev(egress_dev);
472 for (i = 0; i < dev->rdev.lldi.nports; i++)
473 if (dev->rdev.lldi.ports[i] == egress_dev)
478 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
479 __u8 *peer_ip, __be16 local_port,
480 __be16 peer_port, u8 tos,
483 struct dst_entry *dst = NULL;
485 if (IS_ENABLED(CONFIG_IPV6)) {
488 memset(&fl6, 0, sizeof(fl6));
489 memcpy(&fl6.daddr, peer_ip, 16);
490 memcpy(&fl6.saddr, local_ip, 16);
491 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
492 fl6.flowi6_oif = sin6_scope_id;
493 dst = ip6_route_output(&init_net, NULL, &fl6);
496 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
497 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
507 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
508 __be32 peer_ip, __be16 local_port,
509 __be16 peer_port, u8 tos)
515 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
516 peer_port, local_port, IPPROTO_TCP,
520 n = dst_neigh_lookup(&rt->dst, &peer_ip);
523 if (!our_interface(dev, n->dev) &&
524 !(n->dev->flags & IFF_LOOPBACK)) {
526 dst_release(&rt->dst);
533 static void arp_failure_discard(void *handle, struct sk_buff *skb)
535 pr_err(MOD "ARP failure\n");
539 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
541 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
546 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
547 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
550 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
554 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
555 release_ep_resources(ep);
559 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
563 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
564 c4iw_put_ep(&ep->parent_ep->com);
565 release_ep_resources(ep);
570 * Fake up a special CPL opcode and call sched() so process_work() will call
571 * _put_ep_safe() in a safe context to free the ep resources. This is needed
572 * because ARP error handlers are called in an ATOMIC context, and
573 * _c4iw_free_ep() needs to block.
575 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
578 struct cpl_act_establish *rpl = cplhdr(skb);
580 /* Set our special ARP_FAILURE opcode */
581 rpl->ot.opcode = cpl;
584 * Save ep in the skb->cb area, after where sched() will save the dev
587 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
588 sched(ep->com.dev, skb);
591 /* Handle an ARP failure for an accept */
592 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
594 struct c4iw_ep *ep = handle;
596 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
599 __state_set(&ep->com, DEAD);
600 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
604 * Handle an ARP failure for an active open.
606 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
608 struct c4iw_ep *ep = handle;
610 printk(KERN_ERR MOD "ARP failure during connect\n");
611 connect_reply_upcall(ep, -EHOSTUNREACH);
612 __state_set(&ep->com, DEAD);
613 if (ep->com.remote_addr.ss_family == AF_INET6) {
614 struct sockaddr_in6 *sin6 =
615 (struct sockaddr_in6 *)&ep->com.local_addr;
616 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
617 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
619 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
620 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
621 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
625 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
628 static void abort_arp_failure(void *handle, struct sk_buff *skb)
631 struct c4iw_ep *ep = handle;
632 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
633 struct cpl_abort_req *req = cplhdr(skb);
635 PDBG("%s rdev %p\n", __func__, rdev);
636 req->cmd = CPL_ABORT_NO_RST;
637 ret = c4iw_ofld_send(rdev, skb);
639 __state_set(&ep->com, DEAD);
640 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
644 static int send_flowc(struct c4iw_ep *ep)
646 struct fw_flowc_wr *flowc;
647 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 u16 vlan = ep->l2t->vlan;
655 if (vlan == CPL_L2T_VLAN_NONE)
660 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
662 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
663 FW_FLOWC_WR_NPARAMS_V(nparams));
664 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
665 16)) | FW_WR_FLOWID_V(ep->hwtid));
667 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
668 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
669 (ep->com.dev->rdev.lldi.pf));
670 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
671 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
672 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
673 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
674 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
675 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
676 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
677 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
678 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
679 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
680 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
681 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
682 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
683 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
687 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
688 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
689 flowc->mnemval[8].val = cpu_to_be32(pri);
691 /* Pad WR to 16 byte boundary */
692 flowc->mnemval[8].mnemonic = 0;
693 flowc->mnemval[8].val = 0;
695 for (i = 0; i < 9; i++) {
696 flowc->mnemval[i].r4[0] = 0;
697 flowc->mnemval[i].r4[1] = 0;
698 flowc->mnemval[i].r4[2] = 0;
701 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
702 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
705 static int send_halfclose(struct c4iw_ep *ep)
707 struct cpl_close_con_req *req;
708 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
709 int wrlen = roundup(sizeof *req, 16);
711 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
715 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
716 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
717 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
718 memset(req, 0, wrlen);
719 INIT_TP_WR(req, ep->hwtid);
720 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
722 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
725 static int send_abort(struct c4iw_ep *ep)
727 struct cpl_abort_req *req;
728 int wrlen = roundup(sizeof *req, 16);
729 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
731 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
732 if (WARN_ON(!req_skb))
735 set_wr_txq(req_skb, CPL_PRIORITY_DATA, ep->txq_idx);
736 t4_set_arp_err_handler(req_skb, ep, abort_arp_failure);
737 req = (struct cpl_abort_req *)skb_put(req_skb, wrlen);
738 memset(req, 0, wrlen);
739 INIT_TP_WR(req, ep->hwtid);
740 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
741 req->cmd = CPL_ABORT_SEND_RST;
742 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
745 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
746 unsigned int *idx, int use_ts, int ipv6)
748 unsigned short hdr_size = (ipv6 ?
749 sizeof(struct ipv6hdr) :
750 sizeof(struct iphdr)) +
751 sizeof(struct tcphdr) +
753 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
754 unsigned short data_size = mtu - hdr_size;
756 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
759 static int send_connect(struct c4iw_ep *ep)
761 struct cpl_act_open_req *req = NULL;
762 struct cpl_t5_act_open_req *t5req = NULL;
763 struct cpl_t6_act_open_req *t6req = NULL;
764 struct cpl_act_open_req6 *req6 = NULL;
765 struct cpl_t5_act_open_req6 *t5req6 = NULL;
766 struct cpl_t6_act_open_req6 *t6req6 = NULL;
770 unsigned int mtu_idx;
772 int win, sizev4, sizev6, wrlen;
773 struct sockaddr_in *la = (struct sockaddr_in *)
775 struct sockaddr_in *ra = (struct sockaddr_in *)
776 &ep->com.remote_addr;
777 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
779 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
780 &ep->com.remote_addr;
782 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
783 u32 isn = (prandom_u32() & ~7UL) - 1;
785 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
787 sizev4 = sizeof(struct cpl_act_open_req);
788 sizev6 = sizeof(struct cpl_act_open_req6);
791 sizev4 = sizeof(struct cpl_t5_act_open_req);
792 sizev6 = sizeof(struct cpl_t5_act_open_req6);
795 sizev4 = sizeof(struct cpl_t6_act_open_req);
796 sizev6 = sizeof(struct cpl_t6_act_open_req6);
799 pr_err("T%d Chip is not supported\n",
800 CHELSIO_CHIP_VERSION(adapter_type));
804 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
805 roundup(sizev4, 16) :
808 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
810 skb = get_skb(NULL, wrlen, GFP_KERNEL);
812 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
816 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
818 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
819 enable_tcp_timestamps,
820 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
821 wscale = compute_wscale(rcv_win);
824 * Specify the largest window that will fit in opt0. The
825 * remainder will be specified in the rx_data_ack.
827 win = ep->rcv_win >> 10;
828 if (win > RCV_BUFSIZ_M)
831 opt0 = (nocong ? NO_CONG_F : 0) |
834 WND_SCALE_V(wscale) |
836 L2T_IDX_V(ep->l2t->idx) |
837 TX_CHAN_V(ep->tx_chan) |
838 SMAC_SEL_V(ep->smac_idx) |
839 DSCP_V(ep->tos >> 2) |
840 ULP_MODE_V(ULP_MODE_TCPDDP) |
842 opt2 = RX_CHANNEL_V(0) |
843 CCTRL_ECN_V(enable_ecn) |
844 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
845 if (enable_tcp_timestamps)
846 opt2 |= TSTAMPS_EN_F;
849 if (wscale && enable_tcp_window_scaling)
850 opt2 |= WND_SCALE_EN_F;
851 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
855 opt2 |= T5_OPT_2_VALID_F;
856 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
860 if (ep->com.remote_addr.ss_family == AF_INET6)
861 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
862 (const u32 *)&la6->sin6_addr.s6_addr, 1);
864 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
866 if (ep->com.remote_addr.ss_family == AF_INET) {
867 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
869 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
873 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
875 INIT_TP_WR(t5req, 0);
876 req = (struct cpl_act_open_req *)t5req;
879 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
881 INIT_TP_WR(t6req, 0);
882 req = (struct cpl_act_open_req *)t6req;
883 t5req = (struct cpl_t5_act_open_req *)t6req;
886 pr_err("T%d Chip is not supported\n",
887 CHELSIO_CHIP_VERSION(adapter_type));
892 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
893 ((ep->rss_qid<<14) | ep->atid)));
894 req->local_port = la->sin_port;
895 req->peer_port = ra->sin_port;
896 req->local_ip = la->sin_addr.s_addr;
897 req->peer_ip = ra->sin_addr.s_addr;
898 req->opt0 = cpu_to_be64(opt0);
900 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
901 req->params = cpu_to_be32(cxgb4_select_ntuple(
902 ep->com.dev->rdev.lldi.ports[0],
904 req->opt2 = cpu_to_be32(opt2);
906 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
908 ep->com.dev->rdev.lldi.ports[0],
910 t5req->rsvd = cpu_to_be32(isn);
911 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
912 t5req->opt2 = cpu_to_be32(opt2);
915 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
917 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
921 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
923 INIT_TP_WR(t5req6, 0);
924 req6 = (struct cpl_act_open_req6 *)t5req6;
927 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
929 INIT_TP_WR(t6req6, 0);
930 req6 = (struct cpl_act_open_req6 *)t6req6;
931 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
934 pr_err("T%d Chip is not supported\n",
935 CHELSIO_CHIP_VERSION(adapter_type));
940 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
941 ((ep->rss_qid<<14)|ep->atid)));
942 req6->local_port = la6->sin6_port;
943 req6->peer_port = ra6->sin6_port;
944 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
945 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
946 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
947 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
948 req6->opt0 = cpu_to_be64(opt0);
950 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
951 req6->params = cpu_to_be32(cxgb4_select_ntuple(
952 ep->com.dev->rdev.lldi.ports[0],
954 req6->opt2 = cpu_to_be32(opt2);
956 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
958 ep->com.dev->rdev.lldi.ports[0],
960 t5req6->rsvd = cpu_to_be32(isn);
961 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
962 t5req6->opt2 = cpu_to_be32(opt2);
966 set_bit(ACT_OPEN_REQ, &ep->com.history);
967 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
969 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
970 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
971 (const u32 *)&la6->sin6_addr.s6_addr, 1);
975 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
978 int mpalen, wrlen, ret;
979 struct fw_ofld_tx_data_wr *req;
980 struct mpa_message *mpa;
981 struct mpa_v2_conn_params mpa_v2_params;
983 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
985 BUG_ON(skb_cloned(skb));
987 mpalen = sizeof(*mpa) + ep->plen;
988 if (mpa_rev_to_use == 2)
989 mpalen += sizeof(struct mpa_v2_conn_params);
990 wrlen = roundup(mpalen + sizeof *req, 16);
991 skb = get_skb(skb, wrlen, GFP_KERNEL);
993 connect_reply_upcall(ep, -ENOMEM);
996 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
998 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
999 memset(req, 0, wrlen);
1000 req->op_to_immdlen = cpu_to_be32(
1001 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1003 FW_WR_IMMDLEN_V(mpalen));
1004 req->flowid_len16 = cpu_to_be32(
1005 FW_WR_FLOWID_V(ep->hwtid) |
1006 FW_WR_LEN16_V(wrlen >> 4));
1007 req->plen = cpu_to_be32(mpalen);
1008 req->tunnel_to_proxy = cpu_to_be32(
1009 FW_OFLD_TX_DATA_WR_FLUSH_F |
1010 FW_OFLD_TX_DATA_WR_SHOVE_F);
1012 mpa = (struct mpa_message *)(req + 1);
1013 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
1017 mpa->flags |= MPA_CRC;
1018 if (markers_enabled) {
1019 mpa->flags |= MPA_MARKERS;
1020 ep->mpa_attr.recv_marker_enabled = 1;
1022 ep->mpa_attr.recv_marker_enabled = 0;
1024 if (mpa_rev_to_use == 2)
1025 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1027 mpa->private_data_size = htons(ep->plen);
1028 mpa->revision = mpa_rev_to_use;
1029 if (mpa_rev_to_use == 1) {
1030 ep->tried_with_mpa_v1 = 1;
1031 ep->retry_with_mpa_v1 = 0;
1034 if (mpa_rev_to_use == 2) {
1035 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1036 sizeof (struct mpa_v2_conn_params));
1037 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1039 mpa_v2_params.ird = htons((u16)ep->ird);
1040 mpa_v2_params.ord = htons((u16)ep->ord);
1043 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1044 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1045 mpa_v2_params.ord |=
1046 htons(MPA_V2_RDMA_WRITE_RTR);
1047 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1048 mpa_v2_params.ord |=
1049 htons(MPA_V2_RDMA_READ_RTR);
1051 memcpy(mpa->private_data, &mpa_v2_params,
1052 sizeof(struct mpa_v2_conn_params));
1055 memcpy(mpa->private_data +
1056 sizeof(struct mpa_v2_conn_params),
1057 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1060 memcpy(mpa->private_data,
1061 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1064 * Reference the mpa skb. This ensures the data area
1065 * will remain in memory until the hw acks the tx.
1066 * Function fw4_ack() will deref it.
1069 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1070 BUG_ON(ep->mpa_skb);
1072 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1076 __state_set(&ep->com, MPA_REQ_SENT);
1077 ep->mpa_attr.initiator = 1;
1078 ep->snd_seq += mpalen;
1082 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1085 struct fw_ofld_tx_data_wr *req;
1086 struct mpa_message *mpa;
1087 struct sk_buff *skb;
1088 struct mpa_v2_conn_params mpa_v2_params;
1090 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1092 mpalen = sizeof(*mpa) + plen;
1093 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1094 mpalen += sizeof(struct mpa_v2_conn_params);
1095 wrlen = roundup(mpalen + sizeof *req, 16);
1097 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1099 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1102 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1104 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1105 memset(req, 0, wrlen);
1106 req->op_to_immdlen = cpu_to_be32(
1107 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1109 FW_WR_IMMDLEN_V(mpalen));
1110 req->flowid_len16 = cpu_to_be32(
1111 FW_WR_FLOWID_V(ep->hwtid) |
1112 FW_WR_LEN16_V(wrlen >> 4));
1113 req->plen = cpu_to_be32(mpalen);
1114 req->tunnel_to_proxy = cpu_to_be32(
1115 FW_OFLD_TX_DATA_WR_FLUSH_F |
1116 FW_OFLD_TX_DATA_WR_SHOVE_F);
1118 mpa = (struct mpa_message *)(req + 1);
1119 memset(mpa, 0, sizeof(*mpa));
1120 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1121 mpa->flags = MPA_REJECT;
1122 mpa->revision = ep->mpa_attr.version;
1123 mpa->private_data_size = htons(plen);
1125 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1126 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1127 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1128 sizeof (struct mpa_v2_conn_params));
1129 mpa_v2_params.ird = htons(((u16)ep->ird) |
1130 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1132 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1134 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1135 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1136 FW_RI_INIT_P2PTYPE_READ_REQ ?
1137 MPA_V2_RDMA_READ_RTR : 0) : 0));
1138 memcpy(mpa->private_data, &mpa_v2_params,
1139 sizeof(struct mpa_v2_conn_params));
1142 memcpy(mpa->private_data +
1143 sizeof(struct mpa_v2_conn_params), pdata, plen);
1146 memcpy(mpa->private_data, pdata, plen);
1149 * Reference the mpa skb again. This ensures the data area
1150 * will remain in memory until the hw acks the tx.
1151 * Function fw4_ack() will deref it.
1154 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1155 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1156 BUG_ON(ep->mpa_skb);
1158 ep->snd_seq += mpalen;
1159 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1162 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1165 struct fw_ofld_tx_data_wr *req;
1166 struct mpa_message *mpa;
1167 struct sk_buff *skb;
1168 struct mpa_v2_conn_params mpa_v2_params;
1170 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1172 mpalen = sizeof(*mpa) + plen;
1173 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1174 mpalen += sizeof(struct mpa_v2_conn_params);
1175 wrlen = roundup(mpalen + sizeof *req, 16);
1177 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1179 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1182 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1184 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1185 memset(req, 0, wrlen);
1186 req->op_to_immdlen = cpu_to_be32(
1187 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1189 FW_WR_IMMDLEN_V(mpalen));
1190 req->flowid_len16 = cpu_to_be32(
1191 FW_WR_FLOWID_V(ep->hwtid) |
1192 FW_WR_LEN16_V(wrlen >> 4));
1193 req->plen = cpu_to_be32(mpalen);
1194 req->tunnel_to_proxy = cpu_to_be32(
1195 FW_OFLD_TX_DATA_WR_FLUSH_F |
1196 FW_OFLD_TX_DATA_WR_SHOVE_F);
1198 mpa = (struct mpa_message *)(req + 1);
1199 memset(mpa, 0, sizeof(*mpa));
1200 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1202 if (ep->mpa_attr.crc_enabled)
1203 mpa->flags |= MPA_CRC;
1204 if (ep->mpa_attr.recv_marker_enabled)
1205 mpa->flags |= MPA_MARKERS;
1206 mpa->revision = ep->mpa_attr.version;
1207 mpa->private_data_size = htons(plen);
1209 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1210 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1211 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1212 sizeof (struct mpa_v2_conn_params));
1213 mpa_v2_params.ird = htons((u16)ep->ird);
1214 mpa_v2_params.ord = htons((u16)ep->ord);
1215 if (peer2peer && (ep->mpa_attr.p2p_type !=
1216 FW_RI_INIT_P2PTYPE_DISABLED)) {
1217 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1219 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1220 mpa_v2_params.ord |=
1221 htons(MPA_V2_RDMA_WRITE_RTR);
1222 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1223 mpa_v2_params.ord |=
1224 htons(MPA_V2_RDMA_READ_RTR);
1227 memcpy(mpa->private_data, &mpa_v2_params,
1228 sizeof(struct mpa_v2_conn_params));
1231 memcpy(mpa->private_data +
1232 sizeof(struct mpa_v2_conn_params), pdata, plen);
1235 memcpy(mpa->private_data, pdata, plen);
1238 * Reference the mpa skb. This ensures the data area
1239 * will remain in memory until the hw acks the tx.
1240 * Function fw4_ack() will deref it.
1243 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1245 __state_set(&ep->com, MPA_REP_SENT);
1246 ep->snd_seq += mpalen;
1247 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1250 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1253 struct cpl_act_establish *req = cplhdr(skb);
1254 unsigned int tid = GET_TID(req);
1255 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1256 struct tid_info *t = dev->rdev.lldi.tids;
1259 ep = lookup_atid(t, atid);
1261 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1262 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1264 mutex_lock(&ep->com.mutex);
1265 dst_confirm(ep->dst);
1267 /* setup the hwtid for this connection */
1269 cxgb4_insert_tid(t, ep, tid);
1272 ep->snd_seq = be32_to_cpu(req->snd_isn);
1273 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1275 set_emss(ep, ntohs(req->tcp_opt));
1277 /* dealloc the atid */
1278 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1279 cxgb4_free_atid(t, atid);
1280 set_bit(ACT_ESTAB, &ep->com.history);
1282 /* start MPA negotiation */
1283 ret = send_flowc(ep);
1286 if (ep->retry_with_mpa_v1)
1287 ret = send_mpa_req(ep, skb, 1);
1289 ret = send_mpa_req(ep, skb, mpa_rev);
1292 mutex_unlock(&ep->com.mutex);
1295 mutex_unlock(&ep->com.mutex);
1296 connect_reply_upcall(ep, -ENOMEM);
1297 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1301 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1303 struct iw_cm_event event;
1305 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1306 memset(&event, 0, sizeof(event));
1307 event.event = IW_CM_EVENT_CLOSE;
1308 event.status = status;
1309 if (ep->com.cm_id) {
1310 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1311 ep, ep->com.cm_id, ep->hwtid);
1312 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1313 deref_cm_id(&ep->com);
1314 set_bit(CLOSE_UPCALL, &ep->com.history);
1318 static void peer_close_upcall(struct c4iw_ep *ep)
1320 struct iw_cm_event event;
1322 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1323 memset(&event, 0, sizeof(event));
1324 event.event = IW_CM_EVENT_DISCONNECT;
1325 if (ep->com.cm_id) {
1326 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1327 ep, ep->com.cm_id, ep->hwtid);
1328 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1329 set_bit(DISCONN_UPCALL, &ep->com.history);
1333 static void peer_abort_upcall(struct c4iw_ep *ep)
1335 struct iw_cm_event event;
1337 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1338 memset(&event, 0, sizeof(event));
1339 event.event = IW_CM_EVENT_CLOSE;
1340 event.status = -ECONNRESET;
1341 if (ep->com.cm_id) {
1342 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1343 ep->com.cm_id, ep->hwtid);
1344 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1345 deref_cm_id(&ep->com);
1346 set_bit(ABORT_UPCALL, &ep->com.history);
1350 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1352 struct iw_cm_event event;
1354 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1355 memset(&event, 0, sizeof(event));
1356 event.event = IW_CM_EVENT_CONNECT_REPLY;
1357 event.status = status;
1358 memcpy(&event.local_addr, &ep->com.local_addr,
1359 sizeof(ep->com.local_addr));
1360 memcpy(&event.remote_addr, &ep->com.remote_addr,
1361 sizeof(ep->com.remote_addr));
1363 if ((status == 0) || (status == -ECONNREFUSED)) {
1364 if (!ep->tried_with_mpa_v1) {
1365 /* this means MPA_v2 is used */
1366 event.ord = ep->ird;
1367 event.ird = ep->ord;
1368 event.private_data_len = ep->plen -
1369 sizeof(struct mpa_v2_conn_params);
1370 event.private_data = ep->mpa_pkt +
1371 sizeof(struct mpa_message) +
1372 sizeof(struct mpa_v2_conn_params);
1374 /* this means MPA_v1 is used */
1375 event.ord = cur_max_read_depth(ep->com.dev);
1376 event.ird = cur_max_read_depth(ep->com.dev);
1377 event.private_data_len = ep->plen;
1378 event.private_data = ep->mpa_pkt +
1379 sizeof(struct mpa_message);
1383 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1385 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1386 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1389 deref_cm_id(&ep->com);
1392 static int connect_request_upcall(struct c4iw_ep *ep)
1394 struct iw_cm_event event;
1397 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1398 memset(&event, 0, sizeof(event));
1399 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1400 memcpy(&event.local_addr, &ep->com.local_addr,
1401 sizeof(ep->com.local_addr));
1402 memcpy(&event.remote_addr, &ep->com.remote_addr,
1403 sizeof(ep->com.remote_addr));
1404 event.provider_data = ep;
1405 if (!ep->tried_with_mpa_v1) {
1406 /* this means MPA_v2 is used */
1407 event.ord = ep->ord;
1408 event.ird = ep->ird;
1409 event.private_data_len = ep->plen -
1410 sizeof(struct mpa_v2_conn_params);
1411 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1412 sizeof(struct mpa_v2_conn_params);
1414 /* this means MPA_v1 is used. Send max supported */
1415 event.ord = cur_max_read_depth(ep->com.dev);
1416 event.ird = cur_max_read_depth(ep->com.dev);
1417 event.private_data_len = ep->plen;
1418 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1420 c4iw_get_ep(&ep->com);
1421 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1424 c4iw_put_ep(&ep->com);
1425 set_bit(CONNREQ_UPCALL, &ep->com.history);
1426 c4iw_put_ep(&ep->parent_ep->com);
1430 static void established_upcall(struct c4iw_ep *ep)
1432 struct iw_cm_event event;
1434 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1435 memset(&event, 0, sizeof(event));
1436 event.event = IW_CM_EVENT_ESTABLISHED;
1437 event.ird = ep->ord;
1438 event.ord = ep->ird;
1439 if (ep->com.cm_id) {
1440 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1441 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1442 set_bit(ESTAB_UPCALL, &ep->com.history);
1446 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1448 struct cpl_rx_data_ack *req;
1449 struct sk_buff *skb;
1450 int wrlen = roundup(sizeof *req, 16);
1452 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1453 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1455 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1460 * If we couldn't specify the entire rcv window at connection setup
1461 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1462 * then add the overage in to the credits returned.
1464 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1465 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1467 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1468 memset(req, 0, wrlen);
1469 INIT_TP_WR(req, ep->hwtid);
1470 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1472 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1474 RX_DACK_MODE_V(dack_mode));
1475 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1476 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1480 #define RELAXED_IRD_NEGOTIATION 1
1483 * process_mpa_reply - process streaming mode MPA reply
1487 * 0 upon success indicating a connect request was delivered to the ULP
1488 * or the mpa request is incomplete but valid so far.
1490 * 1 if a failure requires the caller to close the connection.
1492 * 2 if a failure requires the caller to abort the connection.
1494 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1496 struct mpa_message *mpa;
1497 struct mpa_v2_conn_params *mpa_v2_params;
1499 u16 resp_ird, resp_ord;
1500 u8 rtr_mismatch = 0, insuff_ird = 0;
1501 struct c4iw_qp_attributes attrs;
1502 enum c4iw_qp_attr_mask mask;
1506 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1509 * If we get more than the supported amount of private data
1510 * then we must fail this connection.
1512 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1514 goto err_stop_timer;
1518 * copy the new data into our accumulation buffer.
1520 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1522 ep->mpa_pkt_len += skb->len;
1525 * if we don't even have the mpa message, then bail.
1527 if (ep->mpa_pkt_len < sizeof(*mpa))
1529 mpa = (struct mpa_message *) ep->mpa_pkt;
1531 /* Validate MPA header. */
1532 if (mpa->revision > mpa_rev) {
1533 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1534 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1536 goto err_stop_timer;
1538 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1540 goto err_stop_timer;
1543 plen = ntohs(mpa->private_data_size);
1546 * Fail if there's too much private data.
1548 if (plen > MPA_MAX_PRIVATE_DATA) {
1550 goto err_stop_timer;
1554 * If plen does not account for pkt size
1556 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1558 goto err_stop_timer;
1561 ep->plen = (u8) plen;
1564 * If we don't have all the pdata yet, then bail.
1565 * We'll continue process when more data arrives.
1567 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1570 if (mpa->flags & MPA_REJECT) {
1571 err = -ECONNREFUSED;
1572 goto err_stop_timer;
1576 * Stop mpa timer. If it expired, then
1577 * we ignore the MPA reply. process_timeout()
1578 * will abort the connection.
1580 if (stop_ep_timer(ep))
1584 * If we get here we have accumulated the entire mpa
1585 * start reply message including private data. And
1586 * the MPA header is valid.
1588 __state_set(&ep->com, FPDU_MODE);
1589 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1590 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1591 ep->mpa_attr.version = mpa->revision;
1592 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1594 if (mpa->revision == 2) {
1595 ep->mpa_attr.enhanced_rdma_conn =
1596 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1597 if (ep->mpa_attr.enhanced_rdma_conn) {
1598 mpa_v2_params = (struct mpa_v2_conn_params *)
1599 (ep->mpa_pkt + sizeof(*mpa));
1600 resp_ird = ntohs(mpa_v2_params->ird) &
1601 MPA_V2_IRD_ORD_MASK;
1602 resp_ord = ntohs(mpa_v2_params->ord) &
1603 MPA_V2_IRD_ORD_MASK;
1604 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1605 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1608 * This is a double-check. Ideally, below checks are
1609 * not required since ird/ord stuff has been taken
1610 * care of in c4iw_accept_cr
1612 if (ep->ird < resp_ord) {
1613 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1614 ep->com.dev->rdev.lldi.max_ordird_qp)
1618 } else if (ep->ird > resp_ord) {
1621 if (ep->ord > resp_ird) {
1622 if (RELAXED_IRD_NEGOTIATION)
1633 if (ntohs(mpa_v2_params->ird) &
1634 MPA_V2_PEER2PEER_MODEL) {
1635 if (ntohs(mpa_v2_params->ord) &
1636 MPA_V2_RDMA_WRITE_RTR)
1637 ep->mpa_attr.p2p_type =
1638 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1639 else if (ntohs(mpa_v2_params->ord) &
1640 MPA_V2_RDMA_READ_RTR)
1641 ep->mpa_attr.p2p_type =
1642 FW_RI_INIT_P2PTYPE_READ_REQ;
1645 } else if (mpa->revision == 1)
1647 ep->mpa_attr.p2p_type = p2p_type;
1649 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1650 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1651 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1652 ep->mpa_attr.recv_marker_enabled,
1653 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1654 ep->mpa_attr.p2p_type, p2p_type);
1657 * If responder's RTR does not match with that of initiator, assign
1658 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1659 * generated when moving QP to RTS state.
1660 * A TERM message will be sent after QP has moved to RTS state
1662 if ((ep->mpa_attr.version == 2) && peer2peer &&
1663 (ep->mpa_attr.p2p_type != p2p_type)) {
1664 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1668 attrs.mpa_attr = ep->mpa_attr;
1669 attrs.max_ird = ep->ird;
1670 attrs.max_ord = ep->ord;
1671 attrs.llp_stream_handle = ep;
1672 attrs.next_state = C4IW_QP_STATE_RTS;
1674 mask = C4IW_QP_ATTR_NEXT_STATE |
1675 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1676 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1678 /* bind QP and TID with INIT_WR */
1679 err = c4iw_modify_qp(ep->com.qp->rhp,
1680 ep->com.qp, mask, &attrs, 1);
1685 * If responder's RTR requirement did not match with what initiator
1686 * supports, generate TERM message
1689 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1690 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1691 attrs.ecode = MPA_NOMATCH_RTR;
1692 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1693 attrs.send_term = 1;
1694 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1695 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1702 * Generate TERM if initiator IRD is not sufficient for responder
1703 * provided ORD. Currently, we do the same behaviour even when
1704 * responder provided IRD is also not sufficient as regards to
1708 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1710 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1711 attrs.ecode = MPA_INSUFF_IRD;
1712 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1713 attrs.send_term = 1;
1714 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1715 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1726 connect_reply_upcall(ep, err);
1731 * process_mpa_request - process streaming mode MPA request
1735 * 0 upon success indicating a connect request was delivered to the ULP
1736 * or the mpa request is incomplete but valid so far.
1738 * 1 if a failure requires the caller to close the connection.
1740 * 2 if a failure requires the caller to abort the connection.
1742 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1744 struct mpa_message *mpa;
1745 struct mpa_v2_conn_params *mpa_v2_params;
1748 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1751 * If we get more than the supported amount of private data
1752 * then we must fail this connection.
1754 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1755 goto err_stop_timer;
1757 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1760 * Copy the new data into our accumulation buffer.
1762 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1764 ep->mpa_pkt_len += skb->len;
1767 * If we don't even have the mpa message, then bail.
1768 * We'll continue process when more data arrives.
1770 if (ep->mpa_pkt_len < sizeof(*mpa))
1773 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1774 mpa = (struct mpa_message *) ep->mpa_pkt;
1777 * Validate MPA Header.
1779 if (mpa->revision > mpa_rev) {
1780 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1781 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1782 goto err_stop_timer;
1785 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1786 goto err_stop_timer;
1788 plen = ntohs(mpa->private_data_size);
1791 * Fail if there's too much private data.
1793 if (plen > MPA_MAX_PRIVATE_DATA)
1794 goto err_stop_timer;
1797 * If plen does not account for pkt size
1799 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1800 goto err_stop_timer;
1801 ep->plen = (u8) plen;
1804 * If we don't have all the pdata yet, then bail.
1806 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1810 * If we get here we have accumulated the entire mpa
1811 * start reply message including private data.
1813 ep->mpa_attr.initiator = 0;
1814 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1815 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1816 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1817 ep->mpa_attr.version = mpa->revision;
1818 if (mpa->revision == 1)
1819 ep->tried_with_mpa_v1 = 1;
1820 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1822 if (mpa->revision == 2) {
1823 ep->mpa_attr.enhanced_rdma_conn =
1824 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1825 if (ep->mpa_attr.enhanced_rdma_conn) {
1826 mpa_v2_params = (struct mpa_v2_conn_params *)
1827 (ep->mpa_pkt + sizeof(*mpa));
1828 ep->ird = ntohs(mpa_v2_params->ird) &
1829 MPA_V2_IRD_ORD_MASK;
1830 ep->ird = min_t(u32, ep->ird,
1831 cur_max_read_depth(ep->com.dev));
1832 ep->ord = ntohs(mpa_v2_params->ord) &
1833 MPA_V2_IRD_ORD_MASK;
1834 ep->ord = min_t(u32, ep->ord,
1835 cur_max_read_depth(ep->com.dev));
1836 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1838 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1840 if (ntohs(mpa_v2_params->ord) &
1841 MPA_V2_RDMA_WRITE_RTR)
1842 ep->mpa_attr.p2p_type =
1843 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1844 else if (ntohs(mpa_v2_params->ord) &
1845 MPA_V2_RDMA_READ_RTR)
1846 ep->mpa_attr.p2p_type =
1847 FW_RI_INIT_P2PTYPE_READ_REQ;
1850 } else if (mpa->revision == 1)
1852 ep->mpa_attr.p2p_type = p2p_type;
1854 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1855 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1856 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1857 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1858 ep->mpa_attr.p2p_type);
1860 __state_set(&ep->com, MPA_REQ_RCVD);
1863 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1864 if (ep->parent_ep->com.state != DEAD) {
1865 if (connect_request_upcall(ep))
1866 goto err_unlock_parent;
1868 goto err_unlock_parent;
1870 mutex_unlock(&ep->parent_ep->com.mutex);
1874 mutex_unlock(&ep->parent_ep->com.mutex);
1877 (void)stop_ep_timer(ep);
1882 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1885 struct cpl_rx_data *hdr = cplhdr(skb);
1886 unsigned int dlen = ntohs(hdr->len);
1887 unsigned int tid = GET_TID(hdr);
1888 __u8 status = hdr->status;
1891 ep = get_ep_from_tid(dev, tid);
1894 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1895 skb_pull(skb, sizeof(*hdr));
1896 skb_trim(skb, dlen);
1897 mutex_lock(&ep->com.mutex);
1899 /* update RX credits */
1900 update_rx_credits(ep, dlen);
1902 switch (ep->com.state) {
1904 ep->rcv_seq += dlen;
1905 disconnect = process_mpa_reply(ep, skb);
1908 ep->rcv_seq += dlen;
1909 disconnect = process_mpa_request(ep, skb);
1912 struct c4iw_qp_attributes attrs;
1913 BUG_ON(!ep->com.qp);
1915 pr_err("%s Unexpected streaming data." \
1916 " qpid %u ep %p state %d tid %u status %d\n",
1917 __func__, ep->com.qp->wq.sq.qid, ep,
1918 ep->com.state, ep->hwtid, status);
1919 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1920 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1921 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1928 mutex_unlock(&ep->com.mutex);
1930 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1931 c4iw_put_ep(&ep->com);
1935 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1938 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1940 unsigned int tid = GET_TID(rpl);
1942 ep = get_ep_from_tid(dev, tid);
1944 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1947 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1948 mutex_lock(&ep->com.mutex);
1949 switch (ep->com.state) {
1951 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1952 __state_set(&ep->com, DEAD);
1956 printk(KERN_ERR "%s ep %p state %d\n",
1957 __func__, ep, ep->com.state);
1960 mutex_unlock(&ep->com.mutex);
1963 release_ep_resources(ep);
1964 c4iw_put_ep(&ep->com);
1968 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1970 struct sk_buff *skb;
1971 struct fw_ofld_connection_wr *req;
1972 unsigned int mtu_idx;
1974 struct sockaddr_in *sin;
1977 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1978 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1979 memset(req, 0, sizeof(*req));
1980 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1981 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1982 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1983 ep->com.dev->rdev.lldi.ports[0],
1985 sin = (struct sockaddr_in *)&ep->com.local_addr;
1986 req->le.lport = sin->sin_port;
1987 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1988 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1989 req->le.pport = sin->sin_port;
1990 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1991 req->tcb.t_state_to_astid =
1992 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1993 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1994 req->tcb.cplrxdataack_cplpassacceptrpl =
1995 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1996 req->tcb.tx_max = (__force __be32) jiffies;
1997 req->tcb.rcv_adv = htons(1);
1998 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1999 enable_tcp_timestamps,
2000 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2001 wscale = compute_wscale(rcv_win);
2004 * Specify the largest window that will fit in opt0. The
2005 * remainder will be specified in the rx_data_ack.
2007 win = ep->rcv_win >> 10;
2008 if (win > RCV_BUFSIZ_M)
2011 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2012 (nocong ? NO_CONG_F : 0) |
2015 WND_SCALE_V(wscale) |
2016 MSS_IDX_V(mtu_idx) |
2017 L2T_IDX_V(ep->l2t->idx) |
2018 TX_CHAN_V(ep->tx_chan) |
2019 SMAC_SEL_V(ep->smac_idx) |
2020 DSCP_V(ep->tos >> 2) |
2021 ULP_MODE_V(ULP_MODE_TCPDDP) |
2023 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2024 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2026 CCTRL_ECN_V(enable_ecn) |
2027 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2028 if (enable_tcp_timestamps)
2029 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2030 if (enable_tcp_sack)
2031 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2032 if (wscale && enable_tcp_window_scaling)
2033 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2034 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2035 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2036 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2037 set_bit(ACT_OFLD_CONN, &ep->com.history);
2038 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2042 * Some of the error codes above implicitly indicate that there is no TID
2043 * allocated with the result of an ACT_OPEN. We use this predicate to make
2046 static inline int act_open_has_tid(int status)
2048 return (status != CPL_ERR_TCAM_PARITY &&
2049 status != CPL_ERR_TCAM_MISS &&
2050 status != CPL_ERR_TCAM_FULL &&
2051 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2052 status != CPL_ERR_CONN_EXIST);
2055 /* Returns whether a CPL status conveys negative advice.
2057 static int is_neg_adv(unsigned int status)
2059 return status == CPL_ERR_RTX_NEG_ADVICE ||
2060 status == CPL_ERR_PERSIST_NEG_ADVICE ||
2061 status == CPL_ERR_KEEPALV_NEG_ADVICE;
2064 static char *neg_adv_str(unsigned int status)
2067 case CPL_ERR_RTX_NEG_ADVICE:
2068 return "Retransmit timeout";
2069 case CPL_ERR_PERSIST_NEG_ADVICE:
2070 return "Persist timeout";
2071 case CPL_ERR_KEEPALV_NEG_ADVICE:
2072 return "Keepalive timeout";
2078 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2080 ep->snd_win = snd_win;
2081 ep->rcv_win = rcv_win;
2082 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2085 #define ACT_OPEN_RETRY_COUNT 2
2087 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2088 struct dst_entry *dst, struct c4iw_dev *cdev,
2089 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2091 struct neighbour *n;
2093 struct net_device *pdev;
2095 n = dst_neigh_lookup(dst, peer_ip);
2101 if (n->dev->flags & IFF_LOOPBACK) {
2103 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2104 else if (IS_ENABLED(CONFIG_IPV6))
2105 for_each_netdev(&init_net, pdev) {
2106 if (ipv6_chk_addr(&init_net,
2107 (struct in6_addr *)peer_ip,
2118 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2119 n, pdev, rt_tos2priority(tos));
2122 ep->mtu = pdev->mtu;
2123 ep->tx_chan = cxgb4_port_chan(pdev);
2124 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2125 cxgb4_port_viid(pdev));
2126 step = cdev->rdev.lldi.ntxq /
2127 cdev->rdev.lldi.nchan;
2128 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2129 step = cdev->rdev.lldi.nrxq /
2130 cdev->rdev.lldi.nchan;
2131 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2132 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2133 cxgb4_port_idx(pdev) * step];
2134 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2137 pdev = get_real_dev(n->dev);
2138 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2142 ep->mtu = dst_mtu(dst);
2143 ep->tx_chan = cxgb4_port_chan(pdev);
2144 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2145 cxgb4_port_viid(pdev));
2146 step = cdev->rdev.lldi.ntxq /
2147 cdev->rdev.lldi.nchan;
2148 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2149 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2150 step = cdev->rdev.lldi.nrxq /
2151 cdev->rdev.lldi.nchan;
2152 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2153 cxgb4_port_idx(pdev) * step];
2154 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2157 ep->retry_with_mpa_v1 = 0;
2158 ep->tried_with_mpa_v1 = 0;
2170 static int c4iw_reconnect(struct c4iw_ep *ep)
2174 struct sockaddr_in *laddr = (struct sockaddr_in *)
2175 &ep->com.cm_id->m_local_addr;
2176 struct sockaddr_in *raddr = (struct sockaddr_in *)
2177 &ep->com.cm_id->m_remote_addr;
2178 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2179 &ep->com.cm_id->m_local_addr;
2180 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2181 &ep->com.cm_id->m_remote_addr;
2185 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2186 init_timer(&ep->timer);
2187 c4iw_init_wr_wait(&ep->com.wr_wait);
2189 /* When MPA revision is different on nodes, the node with MPA_rev=2
2190 * tries to reconnect with MPA_rev 1 for the same EP through
2191 * c4iw_reconnect(), where the same EP is assigned with new tid for
2192 * further connection establishment. As we are using the same EP pointer
2193 * for reconnect, few skbs are used during the previous c4iw_connect(),
2194 * which leaves the EP with inadequate skbs for further
2195 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2196 * skb_list() during peer_abort(). Allocate skbs which is already used.
2198 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2199 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2205 * Allocate an active TID to initiate a TCP connection.
2207 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2208 if (ep->atid == -1) {
2209 pr_err("%s - cannot alloc atid.\n", __func__);
2213 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2216 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2217 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2218 raddr->sin_addr.s_addr, laddr->sin_port,
2219 raddr->sin_port, ep->com.cm_id->tos);
2221 ra = (__u8 *)&raddr->sin_addr;
2223 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2224 raddr6->sin6_addr.s6_addr,
2225 laddr6->sin6_port, raddr6->sin6_port, 0,
2226 raddr6->sin6_scope_id);
2228 ra = (__u8 *)&raddr6->sin6_addr;
2231 pr_err("%s - cannot find route.\n", __func__);
2232 err = -EHOSTUNREACH;
2235 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2236 ep->com.dev->rdev.lldi.adapter_type,
2237 ep->com.cm_id->tos);
2239 pr_err("%s - cannot alloc l2e.\n", __func__);
2243 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2244 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2247 state_set(&ep->com, CONNECTING);
2248 ep->tos = ep->com.cm_id->tos;
2250 /* send connect request to rnic */
2251 err = send_connect(ep);
2255 cxgb4_l2t_release(ep->l2t);
2257 dst_release(ep->dst);
2259 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2260 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2263 * remember to send notification to upper layer.
2264 * We are in here so the upper layer is not aware that this is
2265 * re-connect attempt and so, upper layer is still waiting for
2266 * response of 1st connect request.
2268 connect_reply_upcall(ep, -ECONNRESET);
2270 c4iw_put_ep(&ep->com);
2275 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2278 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2279 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2280 ntohl(rpl->atid_status)));
2281 struct tid_info *t = dev->rdev.lldi.tids;
2282 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2283 struct sockaddr_in *la;
2284 struct sockaddr_in *ra;
2285 struct sockaddr_in6 *la6;
2286 struct sockaddr_in6 *ra6;
2289 ep = lookup_atid(t, atid);
2290 la = (struct sockaddr_in *)&ep->com.local_addr;
2291 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2292 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2293 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2295 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2296 status, status2errno(status));
2298 if (is_neg_adv(status)) {
2299 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2300 __func__, atid, status, neg_adv_str(status));
2301 ep->stats.connect_neg_adv++;
2302 mutex_lock(&dev->rdev.stats.lock);
2303 dev->rdev.stats.neg_adv++;
2304 mutex_unlock(&dev->rdev.stats.lock);
2308 set_bit(ACT_OPEN_RPL, &ep->com.history);
2311 * Log interesting failures.
2314 case CPL_ERR_CONN_RESET:
2315 case CPL_ERR_CONN_TIMEDOUT:
2317 case CPL_ERR_TCAM_FULL:
2318 mutex_lock(&dev->rdev.stats.lock);
2319 dev->rdev.stats.tcam_full++;
2320 mutex_unlock(&dev->rdev.stats.lock);
2321 if (ep->com.local_addr.ss_family == AF_INET &&
2322 dev->rdev.lldi.enable_fw_ofld_conn) {
2323 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2324 ntohl(rpl->atid_status))));
2330 case CPL_ERR_CONN_EXIST:
2331 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2332 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2333 if (ep->com.remote_addr.ss_family == AF_INET6) {
2334 struct sockaddr_in6 *sin6 =
2335 (struct sockaddr_in6 *)
2336 &ep->com.local_addr;
2338 ep->com.dev->rdev.lldi.ports[0],
2340 &sin6->sin6_addr.s6_addr, 1);
2342 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2344 cxgb4_free_atid(t, atid);
2345 dst_release(ep->dst);
2346 cxgb4_l2t_release(ep->l2t);
2352 if (ep->com.local_addr.ss_family == AF_INET) {
2353 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2354 atid, status, status2errno(status),
2355 &la->sin_addr.s_addr, ntohs(la->sin_port),
2356 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2358 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2359 atid, status, status2errno(status),
2360 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2361 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2367 connect_reply_upcall(ep, status2errno(status));
2368 state_set(&ep->com, DEAD);
2370 if (ep->com.remote_addr.ss_family == AF_INET6) {
2371 struct sockaddr_in6 *sin6 =
2372 (struct sockaddr_in6 *)&ep->com.local_addr;
2373 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2374 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2376 if (status && act_open_has_tid(status))
2377 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2379 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2380 cxgb4_free_atid(t, atid);
2381 dst_release(ep->dst);
2382 cxgb4_l2t_release(ep->l2t);
2383 c4iw_put_ep(&ep->com);
2388 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2390 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2391 unsigned int stid = GET_TID(rpl);
2392 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2395 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2398 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2399 rpl->status, status2errno(rpl->status));
2400 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2401 c4iw_put_ep(&ep->com);
2406 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2408 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2409 unsigned int stid = GET_TID(rpl);
2410 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2412 PDBG("%s ep %p\n", __func__, ep);
2413 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2414 c4iw_put_ep(&ep->com);
2418 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2419 struct cpl_pass_accept_req *req)
2421 struct cpl_pass_accept_rpl *rpl;
2422 unsigned int mtu_idx;
2426 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2428 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2430 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2431 BUG_ON(skb_cloned(skb));
2435 if (!is_t4(adapter_type)) {
2436 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2438 INIT_TP_WR(rpl5, ep->hwtid);
2440 skb_trim(skb, sizeof(*rpl));
2441 INIT_TP_WR(rpl, ep->hwtid);
2443 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2446 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2447 enable_tcp_timestamps && req->tcpopt.tstamp,
2448 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2449 wscale = compute_wscale(rcv_win);
2452 * Specify the largest window that will fit in opt0. The
2453 * remainder will be specified in the rx_data_ack.
2455 win = ep->rcv_win >> 10;
2456 if (win > RCV_BUFSIZ_M)
2458 opt0 = (nocong ? NO_CONG_F : 0) |
2461 WND_SCALE_V(wscale) |
2462 MSS_IDX_V(mtu_idx) |
2463 L2T_IDX_V(ep->l2t->idx) |
2464 TX_CHAN_V(ep->tx_chan) |
2465 SMAC_SEL_V(ep->smac_idx) |
2466 DSCP_V(ep->tos >> 2) |
2467 ULP_MODE_V(ULP_MODE_TCPDDP) |
2469 opt2 = RX_CHANNEL_V(0) |
2470 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2472 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2473 opt2 |= TSTAMPS_EN_F;
2474 if (enable_tcp_sack && req->tcpopt.sack)
2476 if (wscale && enable_tcp_window_scaling)
2477 opt2 |= WND_SCALE_EN_F;
2479 const struct tcphdr *tcph;
2480 u32 hlen = ntohl(req->hdr_len);
2482 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2483 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2486 tcph = (const void *)(req + 1) +
2487 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2488 if (tcph->ece && tcph->cwr)
2489 opt2 |= CCTRL_ECN_V(1);
2491 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2492 u32 isn = (prandom_u32() & ~7UL) - 1;
2493 opt2 |= T5_OPT_2_VALID_F;
2494 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2497 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2500 rpl5->iss = cpu_to_be32(isn);
2501 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2504 rpl->opt0 = cpu_to_be64(opt0);
2505 rpl->opt2 = cpu_to_be32(opt2);
2506 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2507 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2509 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2512 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2514 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2515 BUG_ON(skb_cloned(skb));
2516 skb_trim(skb, sizeof(struct cpl_tid_release));
2517 release_tid(&dev->rdev, hwtid, skb);
2521 static void get_4tuple(struct cpl_pass_accept_req *req, enum chip_type type,
2522 int *iptype, __u8 *local_ip, __u8 *peer_ip,
2523 __be16 *local_port, __be16 *peer_port)
2525 int eth_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2526 ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2527 T6_ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2528 int ip_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2529 IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2530 T6_IP_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2531 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2532 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2533 struct tcphdr *tcp = (struct tcphdr *)
2534 ((u8 *)(req + 1) + eth_len + ip_len);
2536 if (ip->version == 4) {
2537 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2538 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2541 memcpy(peer_ip, &ip->saddr, 4);
2542 memcpy(local_ip, &ip->daddr, 4);
2544 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2545 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2548 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2549 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2551 *peer_port = tcp->source;
2552 *local_port = tcp->dest;
2557 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2559 struct c4iw_ep *child_ep = NULL, *parent_ep;
2560 struct cpl_pass_accept_req *req = cplhdr(skb);
2561 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2562 struct tid_info *t = dev->rdev.lldi.tids;
2563 unsigned int hwtid = GET_TID(req);
2564 struct dst_entry *dst;
2565 __u8 local_ip[16], peer_ip[16];
2566 __be16 local_port, peer_port;
2567 struct sockaddr_in6 *sin6;
2569 u16 peer_mss = ntohs(req->tcpopt.mss);
2571 unsigned short hdrs;
2572 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2574 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2576 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2580 if (state_read(&parent_ep->com) != LISTEN) {
2581 PDBG("%s - listening ep not in LISTEN\n", __func__);
2585 get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype,
2586 local_ip, peer_ip, &local_port, &peer_port);
2588 /* Find output route */
2590 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2591 , __func__, parent_ep, hwtid,
2592 local_ip, peer_ip, ntohs(local_port),
2593 ntohs(peer_port), peer_mss);
2594 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2595 local_port, peer_port,
2598 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2599 , __func__, parent_ep, hwtid,
2600 local_ip, peer_ip, ntohs(local_port),
2601 ntohs(peer_port), peer_mss);
2602 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2603 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2604 ((struct sockaddr_in6 *)
2605 &parent_ep->com.local_addr)->sin6_scope_id);
2608 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2613 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2615 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2621 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2622 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2624 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2631 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2632 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2633 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2634 child_ep->mtu = peer_mss + hdrs;
2636 skb_queue_head_init(&child_ep->com.ep_skb_list);
2637 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2640 state_set(&child_ep->com, CONNECTING);
2641 child_ep->com.dev = dev;
2642 child_ep->com.cm_id = NULL;
2645 struct sockaddr_in *sin = (struct sockaddr_in *)
2646 &child_ep->com.local_addr;
2648 sin->sin_family = PF_INET;
2649 sin->sin_port = local_port;
2650 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2652 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2653 sin->sin_family = PF_INET;
2654 sin->sin_port = ((struct sockaddr_in *)
2655 &parent_ep->com.local_addr)->sin_port;
2656 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2658 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2659 sin->sin_family = PF_INET;
2660 sin->sin_port = peer_port;
2661 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2663 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2664 sin6->sin6_family = PF_INET6;
2665 sin6->sin6_port = local_port;
2666 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2668 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2669 sin6->sin6_family = PF_INET6;
2670 sin6->sin6_port = ((struct sockaddr_in6 *)
2671 &parent_ep->com.local_addr)->sin6_port;
2672 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2674 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2675 sin6->sin6_family = PF_INET6;
2676 sin6->sin6_port = peer_port;
2677 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2680 c4iw_get_ep(&parent_ep->com);
2681 child_ep->parent_ep = parent_ep;
2682 child_ep->tos = tos;
2683 child_ep->dst = dst;
2684 child_ep->hwtid = hwtid;
2686 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2687 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2689 init_timer(&child_ep->timer);
2690 cxgb4_insert_tid(t, child_ep, hwtid);
2691 insert_ep_tid(child_ep);
2692 if (accept_cr(child_ep, skb, req)) {
2693 c4iw_put_ep(&parent_ep->com);
2694 release_ep_resources(child_ep);
2696 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2699 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2700 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2701 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2705 c4iw_put_ep(&child_ep->com);
2707 reject_cr(dev, hwtid, skb);
2709 c4iw_put_ep(&parent_ep->com);
2714 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2717 struct cpl_pass_establish *req = cplhdr(skb);
2718 unsigned int tid = GET_TID(req);
2721 ep = get_ep_from_tid(dev, tid);
2722 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2723 ep->snd_seq = be32_to_cpu(req->snd_isn);
2724 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2726 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2727 ntohs(req->tcp_opt));
2729 set_emss(ep, ntohs(req->tcp_opt));
2731 dst_confirm(ep->dst);
2732 mutex_lock(&ep->com.mutex);
2733 ep->com.state = MPA_REQ_WAIT;
2735 set_bit(PASS_ESTAB, &ep->com.history);
2736 ret = send_flowc(ep);
2737 mutex_unlock(&ep->com.mutex);
2739 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2740 c4iw_put_ep(&ep->com);
2745 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2747 struct cpl_peer_close *hdr = cplhdr(skb);
2749 struct c4iw_qp_attributes attrs;
2752 unsigned int tid = GET_TID(hdr);
2755 ep = get_ep_from_tid(dev, tid);
2759 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2760 dst_confirm(ep->dst);
2762 set_bit(PEER_CLOSE, &ep->com.history);
2763 mutex_lock(&ep->com.mutex);
2764 switch (ep->com.state) {
2766 __state_set(&ep->com, CLOSING);
2769 __state_set(&ep->com, CLOSING);
2770 connect_reply_upcall(ep, -ECONNRESET);
2775 * We're gonna mark this puppy DEAD, but keep
2776 * the reference on it until the ULP accepts or
2777 * rejects the CR. Also wake up anyone waiting
2778 * in rdma connection migration (see c4iw_accept_cr()).
2780 __state_set(&ep->com, CLOSING);
2781 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2782 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2785 __state_set(&ep->com, CLOSING);
2786 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2787 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2791 __state_set(&ep->com, CLOSING);
2792 attrs.next_state = C4IW_QP_STATE_CLOSING;
2793 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2794 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2795 if (ret != -ECONNRESET) {
2796 peer_close_upcall(ep);
2804 __state_set(&ep->com, MORIBUND);
2808 (void)stop_ep_timer(ep);
2809 if (ep->com.cm_id && ep->com.qp) {
2810 attrs.next_state = C4IW_QP_STATE_IDLE;
2811 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2812 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2814 close_complete_upcall(ep, 0);
2815 __state_set(&ep->com, DEAD);
2825 mutex_unlock(&ep->com.mutex);
2827 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2829 release_ep_resources(ep);
2830 c4iw_put_ep(&ep->com);
2834 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2836 struct cpl_abort_req_rss *req = cplhdr(skb);
2838 struct cpl_abort_rpl *rpl;
2839 struct sk_buff *rpl_skb;
2840 struct c4iw_qp_attributes attrs;
2843 unsigned int tid = GET_TID(req);
2845 ep = get_ep_from_tid(dev, tid);
2849 if (is_neg_adv(req->status)) {
2850 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2851 __func__, ep->hwtid, req->status,
2852 neg_adv_str(req->status));
2853 ep->stats.abort_neg_adv++;
2854 mutex_lock(&dev->rdev.stats.lock);
2855 dev->rdev.stats.neg_adv++;
2856 mutex_unlock(&dev->rdev.stats.lock);
2859 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2861 set_bit(PEER_ABORT, &ep->com.history);
2864 * Wake up any threads in rdma_init() or rdma_fini().
2865 * However, this is not needed if com state is just
2868 if (ep->com.state != MPA_REQ_SENT)
2869 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2871 mutex_lock(&ep->com.mutex);
2872 switch (ep->com.state) {
2874 c4iw_put_ep(&ep->parent_ep->com);
2877 (void)stop_ep_timer(ep);
2880 (void)stop_ep_timer(ep);
2881 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2882 connect_reply_upcall(ep, -ECONNRESET);
2885 * we just don't send notification upwards because we
2886 * want to retry with mpa_v1 without upper layers even
2889 * do some housekeeping so as to re-initiate the
2892 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2894 ep->retry_with_mpa_v1 = 1;
2906 if (ep->com.cm_id && ep->com.qp) {
2907 attrs.next_state = C4IW_QP_STATE_ERROR;
2908 ret = c4iw_modify_qp(ep->com.qp->rhp,
2909 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2913 "%s - qp <- error failed!\n",
2916 peer_abort_upcall(ep);
2921 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2922 mutex_unlock(&ep->com.mutex);
2928 dst_confirm(ep->dst);
2929 if (ep->com.state != ABORTING) {
2930 __state_set(&ep->com, DEAD);
2931 /* we don't release if we want to retry with mpa_v1 */
2932 if (!ep->retry_with_mpa_v1)
2935 mutex_unlock(&ep->com.mutex);
2937 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2938 if (WARN_ON(!rpl_skb)) {
2942 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2943 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2944 INIT_TP_WR(rpl, ep->hwtid);
2945 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2946 rpl->cmd = CPL_ABORT_NO_RST;
2947 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2950 release_ep_resources(ep);
2951 else if (ep->retry_with_mpa_v1) {
2952 if (ep->com.remote_addr.ss_family == AF_INET6) {
2953 struct sockaddr_in6 *sin6 =
2954 (struct sockaddr_in6 *)
2955 &ep->com.local_addr;
2957 ep->com.dev->rdev.lldi.ports[0],
2958 (const u32 *)&sin6->sin6_addr.s6_addr,
2961 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2962 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2963 dst_release(ep->dst);
2964 cxgb4_l2t_release(ep->l2t);
2969 c4iw_put_ep(&ep->com);
2970 /* Dereferencing ep, referenced in peer_abort_intr() */
2971 c4iw_put_ep(&ep->com);
2975 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2978 struct c4iw_qp_attributes attrs;
2979 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2981 unsigned int tid = GET_TID(rpl);
2983 ep = get_ep_from_tid(dev, tid);
2987 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2990 /* The cm_id may be null if we failed to connect */
2991 mutex_lock(&ep->com.mutex);
2992 set_bit(CLOSE_CON_RPL, &ep->com.history);
2993 switch (ep->com.state) {
2995 __state_set(&ep->com, MORIBUND);
2998 (void)stop_ep_timer(ep);
2999 if ((ep->com.cm_id) && (ep->com.qp)) {
3000 attrs.next_state = C4IW_QP_STATE_IDLE;
3001 c4iw_modify_qp(ep->com.qp->rhp,
3003 C4IW_QP_ATTR_NEXT_STATE,
3006 close_complete_upcall(ep, 0);
3007 __state_set(&ep->com, DEAD);
3017 mutex_unlock(&ep->com.mutex);
3019 release_ep_resources(ep);
3020 c4iw_put_ep(&ep->com);
3024 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3026 struct cpl_rdma_terminate *rpl = cplhdr(skb);
3027 unsigned int tid = GET_TID(rpl);
3029 struct c4iw_qp_attributes attrs;
3031 ep = get_ep_from_tid(dev, tid);
3034 if (ep && ep->com.qp) {
3035 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
3036 ep->com.qp->wq.sq.qid);
3037 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3038 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3039 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3041 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
3042 c4iw_put_ep(&ep->com);
3048 * Upcall from the adapter indicating data has been transmitted.
3049 * For us its just the single MPA request or reply. We can now free
3050 * the skb holding the mpa message.
3052 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3055 struct cpl_fw4_ack *hdr = cplhdr(skb);
3056 u8 credits = hdr->credits;
3057 unsigned int tid = GET_TID(hdr);
3060 ep = get_ep_from_tid(dev, tid);
3063 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
3065 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
3066 __func__, ep, ep->hwtid, state_read(&ep->com));
3070 dst_confirm(ep->dst);
3072 PDBG("%s last streaming msg ack ep %p tid %u state %u "
3073 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
3074 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
3075 mutex_lock(&ep->com.mutex);
3076 kfree_skb(ep->mpa_skb);
3078 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3080 mutex_unlock(&ep->com.mutex);
3083 c4iw_put_ep(&ep->com);
3087 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3090 struct c4iw_ep *ep = to_ep(cm_id);
3092 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3094 mutex_lock(&ep->com.mutex);
3095 if (ep->com.state != MPA_REQ_RCVD) {
3096 mutex_unlock(&ep->com.mutex);
3097 c4iw_put_ep(&ep->com);
3100 set_bit(ULP_REJECT, &ep->com.history);
3104 abort = send_mpa_reject(ep, pdata, pdata_len);
3105 mutex_unlock(&ep->com.mutex);
3108 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3109 c4iw_put_ep(&ep->com);
3113 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3116 struct c4iw_qp_attributes attrs;
3117 enum c4iw_qp_attr_mask mask;
3118 struct c4iw_ep *ep = to_ep(cm_id);
3119 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3120 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3123 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3125 mutex_lock(&ep->com.mutex);
3126 if (ep->com.state != MPA_REQ_RCVD) {
3133 set_bit(ULP_ACCEPT, &ep->com.history);
3134 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3135 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3140 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3141 if (conn_param->ord > ep->ird) {
3142 if (RELAXED_IRD_NEGOTIATION) {
3145 ep->ird = conn_param->ird;
3146 ep->ord = conn_param->ord;
3147 send_mpa_reject(ep, conn_param->private_data,
3148 conn_param->private_data_len);
3153 if (conn_param->ird < ep->ord) {
3154 if (RELAXED_IRD_NEGOTIATION &&
3155 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3156 conn_param->ird = ep->ord;
3163 ep->ird = conn_param->ird;
3164 ep->ord = conn_param->ord;
3166 if (ep->mpa_attr.version == 1) {
3167 if (peer2peer && ep->ird == 0)
3171 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3172 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3176 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3178 ep->com.cm_id = cm_id;
3179 ref_cm_id(&ep->com);
3183 /* bind QP to EP and move to RTS */
3184 attrs.mpa_attr = ep->mpa_attr;
3185 attrs.max_ird = ep->ird;
3186 attrs.max_ord = ep->ord;
3187 attrs.llp_stream_handle = ep;
3188 attrs.next_state = C4IW_QP_STATE_RTS;
3190 /* bind QP and TID with INIT_WR */
3191 mask = C4IW_QP_ATTR_NEXT_STATE |
3192 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3193 C4IW_QP_ATTR_MPA_ATTR |
3194 C4IW_QP_ATTR_MAX_IRD |
3195 C4IW_QP_ATTR_MAX_ORD;
3197 err = c4iw_modify_qp(ep->com.qp->rhp,
3198 ep->com.qp, mask, &attrs, 1);
3200 goto err_deref_cm_id;
3202 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3203 err = send_mpa_reply(ep, conn_param->private_data,
3204 conn_param->private_data_len);
3206 goto err_deref_cm_id;
3208 __state_set(&ep->com, FPDU_MODE);
3209 established_upcall(ep);
3210 mutex_unlock(&ep->com.mutex);
3211 c4iw_put_ep(&ep->com);
3214 deref_cm_id(&ep->com);
3218 mutex_unlock(&ep->com.mutex);
3220 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3221 c4iw_put_ep(&ep->com);
3225 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3227 struct in_device *ind;
3229 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3230 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3232 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3234 return -EADDRNOTAVAIL;
3235 for_primary_ifa(ind) {
3236 laddr->sin_addr.s_addr = ifa->ifa_address;
3237 raddr->sin_addr.s_addr = ifa->ifa_address;
3243 return found ? 0 : -EADDRNOTAVAIL;
3246 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3247 unsigned char banned_flags)
3249 struct inet6_dev *idev;
3250 int err = -EADDRNOTAVAIL;
3253 idev = __in6_dev_get(dev);
3255 struct inet6_ifaddr *ifp;
3257 read_lock_bh(&idev->lock);
3258 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3259 if (ifp->scope == IFA_LINK &&
3260 !(ifp->flags & banned_flags)) {
3261 memcpy(addr, &ifp->addr, 16);
3266 read_unlock_bh(&idev->lock);
3272 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3274 struct in6_addr uninitialized_var(addr);
3275 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3276 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3278 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3279 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3280 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3283 return -EADDRNOTAVAIL;
3286 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3288 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3291 struct sockaddr_in *laddr;
3292 struct sockaddr_in *raddr;
3293 struct sockaddr_in6 *laddr6;
3294 struct sockaddr_in6 *raddr6;
3298 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3299 (conn_param->ird > cur_max_read_depth(dev))) {
3303 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3305 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3310 skb_queue_head_init(&ep->com.ep_skb_list);
3311 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3316 init_timer(&ep->timer);
3317 ep->plen = conn_param->private_data_len;
3319 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3320 conn_param->private_data, ep->plen);
3321 ep->ird = conn_param->ird;
3322 ep->ord = conn_param->ord;
3324 if (peer2peer && ep->ord == 0)
3327 ep->com.cm_id = cm_id;
3328 ref_cm_id(&ep->com);
3330 ep->com.qp = get_qhp(dev, conn_param->qpn);
3332 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3337 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3341 * Allocate an active TID to initiate a TCP connection.
3343 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3344 if (ep->atid == -1) {
3345 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3349 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3351 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3352 sizeof(ep->com.local_addr));
3353 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3354 sizeof(ep->com.remote_addr));
3356 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3357 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3358 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3359 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3361 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3363 ra = (__u8 *)&raddr->sin_addr;
3366 * Handle loopback requests to INADDR_ANY.
3368 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3369 err = pick_local_ipaddrs(dev, cm_id);
3375 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3376 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3377 ra, ntohs(raddr->sin_port));
3378 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3379 raddr->sin_addr.s_addr, laddr->sin_port,
3380 raddr->sin_port, cm_id->tos);
3383 ra = (__u8 *)&raddr6->sin6_addr;
3386 * Handle loopback requests to INADDR_ANY.
3388 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3389 err = pick_local_ip6addrs(dev, cm_id);
3395 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3396 __func__, laddr6->sin6_addr.s6_addr,
3397 ntohs(laddr6->sin6_port),
3398 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3399 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3400 raddr6->sin6_addr.s6_addr,
3401 laddr6->sin6_port, raddr6->sin6_port, 0,
3402 raddr6->sin6_scope_id);
3405 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3406 err = -EHOSTUNREACH;
3410 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3411 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3413 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3417 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3418 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3421 state_set(&ep->com, CONNECTING);
3422 ep->tos = cm_id->tos;
3424 /* send connect request to rnic */
3425 err = send_connect(ep);
3429 cxgb4_l2t_release(ep->l2t);
3431 dst_release(ep->dst);
3433 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3434 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3436 skb_queue_purge(&ep->com.ep_skb_list);
3437 deref_cm_id(&ep->com);
3439 c4iw_put_ep(&ep->com);
3444 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3447 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3448 &ep->com.local_addr;
3450 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3451 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3452 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3456 c4iw_init_wr_wait(&ep->com.wr_wait);
3457 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3458 ep->stid, &sin6->sin6_addr,
3460 ep->com.dev->rdev.lldi.rxq_ids[0]);
3462 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3466 err = net_xmit_errno(err);
3468 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3469 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3470 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3472 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3477 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3480 struct sockaddr_in *sin = (struct sockaddr_in *)
3481 &ep->com.local_addr;
3483 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3485 err = cxgb4_create_server_filter(
3486 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3487 sin->sin_addr.s_addr, sin->sin_port, 0,
3488 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3489 if (err == -EBUSY) {
3490 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3494 set_current_state(TASK_UNINTERRUPTIBLE);
3495 schedule_timeout(usecs_to_jiffies(100));
3497 } while (err == -EBUSY);
3499 c4iw_init_wr_wait(&ep->com.wr_wait);
3500 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3501 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3502 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3504 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3508 err = net_xmit_errno(err);
3511 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3513 &sin->sin_addr, ntohs(sin->sin_port));
3517 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3520 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3521 struct c4iw_listen_ep *ep;
3525 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3527 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3531 skb_queue_head_init(&ep->com.ep_skb_list);
3532 PDBG("%s ep %p\n", __func__, ep);
3533 ep->com.cm_id = cm_id;
3534 ref_cm_id(&ep->com);
3536 ep->backlog = backlog;
3537 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3538 sizeof(ep->com.local_addr));
3541 * Allocate a server TID.
3543 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3544 ep->com.local_addr.ss_family == AF_INET)
3545 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3546 cm_id->m_local_addr.ss_family, ep);
3548 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3549 cm_id->m_local_addr.ss_family, ep);
3551 if (ep->stid == -1) {
3552 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3556 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3558 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3559 sizeof(ep->com.local_addr));
3561 state_set(&ep->com, LISTEN);
3562 if (ep->com.local_addr.ss_family == AF_INET)
3563 err = create_server4(dev, ep);
3565 err = create_server6(dev, ep);
3567 cm_id->provider_data = ep;
3571 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3572 ep->com.local_addr.ss_family);
3574 deref_cm_id(&ep->com);
3575 c4iw_put_ep(&ep->com);
3581 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3584 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3586 PDBG("%s ep %p\n", __func__, ep);
3589 state_set(&ep->com, DEAD);
3590 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3591 ep->com.local_addr.ss_family == AF_INET) {
3592 err = cxgb4_remove_server_filter(
3593 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3594 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3596 struct sockaddr_in6 *sin6;
3597 c4iw_init_wr_wait(&ep->com.wr_wait);
3598 err = cxgb4_remove_server(
3599 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3600 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3603 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3605 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3606 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3607 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3609 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3610 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3611 ep->com.local_addr.ss_family);
3613 deref_cm_id(&ep->com);
3614 c4iw_put_ep(&ep->com);
3618 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3623 struct c4iw_rdev *rdev;
3625 mutex_lock(&ep->com.mutex);
3627 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3628 states[ep->com.state], abrupt);
3631 * Ref the ep here in case we have fatal errors causing the
3632 * ep to be released and freed.
3634 c4iw_get_ep(&ep->com);
3636 rdev = &ep->com.dev->rdev;
3637 if (c4iw_fatal_error(rdev)) {
3639 close_complete_upcall(ep, -EIO);
3640 ep->com.state = DEAD;
3642 switch (ep->com.state) {
3651 ep->com.state = ABORTING;
3653 ep->com.state = CLOSING;
3656 * if we close before we see the fw4_ack() then we fix
3657 * up the timer state since we're reusing it.
3660 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3661 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3666 set_bit(CLOSE_SENT, &ep->com.flags);
3669 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3672 (void)stop_ep_timer(ep);
3673 ep->com.state = ABORTING;
3675 ep->com.state = MORIBUND;
3681 PDBG("%s ignoring disconnect ep %p state %u\n",
3682 __func__, ep, ep->com.state);
3691 set_bit(EP_DISC_ABORT, &ep->com.history);
3692 close_complete_upcall(ep, -ECONNRESET);
3693 ret = send_abort(ep);
3695 set_bit(EP_DISC_CLOSE, &ep->com.history);
3696 ret = send_halfclose(ep);
3699 set_bit(EP_DISC_FAIL, &ep->com.history);
3702 close_complete_upcall(ep, -EIO);
3705 struct c4iw_qp_attributes attrs;
3707 attrs.next_state = C4IW_QP_STATE_ERROR;
3708 ret = c4iw_modify_qp(ep->com.qp->rhp,
3710 C4IW_QP_ATTR_NEXT_STATE,
3714 "%s - qp <- error failed!\n",
3720 mutex_unlock(&ep->com.mutex);
3721 c4iw_put_ep(&ep->com);
3723 release_ep_resources(ep);
3727 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3728 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3731 int atid = be32_to_cpu(req->tid);
3733 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3734 (__force u32) req->tid);
3738 switch (req->retval) {
3740 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3741 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3742 send_fw_act_open_req(ep, atid);
3746 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3747 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3748 send_fw_act_open_req(ep, atid);
3753 pr_info("%s unexpected ofld conn wr retval %d\n",
3754 __func__, req->retval);
3757 pr_err("active ofld_connect_wr failure %d atid %d\n",
3759 mutex_lock(&dev->rdev.stats.lock);
3760 dev->rdev.stats.act_ofld_conn_fails++;
3761 mutex_unlock(&dev->rdev.stats.lock);
3762 connect_reply_upcall(ep, status2errno(req->retval));
3763 state_set(&ep->com, DEAD);
3764 if (ep->com.remote_addr.ss_family == AF_INET6) {
3765 struct sockaddr_in6 *sin6 =
3766 (struct sockaddr_in6 *)&ep->com.local_addr;
3767 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3768 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3770 remove_handle(dev, &dev->atid_idr, atid);
3771 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3772 dst_release(ep->dst);
3773 cxgb4_l2t_release(ep->l2t);
3774 c4iw_put_ep(&ep->com);
3777 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3778 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3780 struct sk_buff *rpl_skb;
3781 struct cpl_pass_accept_req *cpl;
3784 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3787 PDBG("%s passive open failure %d\n", __func__, req->retval);
3788 mutex_lock(&dev->rdev.stats.lock);
3789 dev->rdev.stats.pas_ofld_conn_fails++;
3790 mutex_unlock(&dev->rdev.stats.lock);
3793 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3794 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3795 (__force u32) htonl(
3796 (__force u32) req->tid)));
3797 ret = pass_accept_req(dev, rpl_skb);
3804 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3806 struct cpl_fw6_msg *rpl = cplhdr(skb);
3807 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3809 switch (rpl->type) {
3811 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3813 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3814 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3815 switch (req->t_state) {
3817 active_ofld_conn_reply(dev, skb, req);
3820 passive_ofld_conn_reply(dev, skb, req);
3823 pr_err("%s unexpected ofld conn wr state %d\n",
3824 __func__, req->t_state);
3832 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3835 __be16 hdr_len, vlantag, len;
3837 int tcp_hdr_len, ip_hdr_len;
3839 struct cpl_rx_pkt *cpl = cplhdr(skb);
3840 struct cpl_pass_accept_req *req;
3841 struct tcp_options_received tmp_opt;
3842 struct c4iw_dev *dev;
3843 enum chip_type type;
3845 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3846 /* Store values from cpl_rx_pkt in temporary location. */
3847 vlantag = cpl->vlan;
3849 l2info = cpl->l2info;
3850 hdr_len = cpl->hdr_len;
3853 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3856 * We need to parse the TCP options from SYN packet.
3857 * to generate cpl_pass_accept_req.
3859 memset(&tmp_opt, 0, sizeof(tmp_opt));
3860 tcp_clear_options(&tmp_opt);
3861 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3863 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3864 memset(req, 0, sizeof(*req));
3865 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3866 SYN_MAC_IDX_V(RX_MACIDX_G(
3867 be32_to_cpu(l2info))) |
3869 type = dev->rdev.lldi.adapter_type;
3870 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3871 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3873 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3874 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3875 eth_hdr_len = is_t4(type) ?
3876 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3877 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3878 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3879 IP_HDR_LEN_V(ip_hdr_len) |
3880 ETH_HDR_LEN_V(eth_hdr_len));
3881 } else { /* T6 and later */
3882 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3883 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3884 T6_IP_HDR_LEN_V(ip_hdr_len) |
3885 T6_ETH_HDR_LEN_V(eth_hdr_len));
3887 req->vlan = vlantag;
3889 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3890 PASS_OPEN_TOS_V(tos));
3891 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3892 if (tmp_opt.wscale_ok)
3893 req->tcpopt.wsf = tmp_opt.snd_wscale;
3894 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3895 if (tmp_opt.sack_ok)
3896 req->tcpopt.sack = 1;
3897 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3901 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3902 __be32 laddr, __be16 lport,
3903 __be32 raddr, __be16 rport,
3904 u32 rcv_isn, u32 filter, u16 window,
3905 u32 rss_qid, u8 port_id)
3907 struct sk_buff *req_skb;
3908 struct fw_ofld_connection_wr *req;
3909 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3912 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3913 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3914 memset(req, 0, sizeof(*req));
3915 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3916 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3917 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3918 req->le.filter = (__force __be32) filter;
3919 req->le.lport = lport;
3920 req->le.pport = rport;
3921 req->le.u.ipv4.lip = laddr;
3922 req->le.u.ipv4.pip = raddr;
3923 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3924 req->tcb.rcv_adv = htons(window);
3925 req->tcb.t_state_to_astid =
3926 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3927 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3928 FW_OFLD_CONNECTION_WR_ASTID_V(
3929 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3932 * We store the qid in opt2 which will be used by the firmware
3933 * to send us the wr response.
3935 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3938 * We initialize the MSS index in TCB to 0xF.
3939 * So that when driver sends cpl_pass_accept_rpl
3940 * TCB picks up the correct value. If this was 0
3941 * TP will ignore any value > 0 for MSS index.
3943 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3944 req->cookie = (uintptr_t)skb;
3946 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3947 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3949 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3957 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3958 * messages when a filter is being used instead of server to
3959 * redirect a syn packet. When packets hit filter they are redirected
3960 * to the offload queue and driver tries to establish the connection
3961 * using firmware work request.
3963 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3966 unsigned int filter;
3967 struct ethhdr *eh = NULL;
3968 struct vlan_ethhdr *vlan_eh = NULL;
3970 struct tcphdr *tcph;
3971 struct rss_header *rss = (void *)skb->data;
3972 struct cpl_rx_pkt *cpl = (void *)skb->data;
3973 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3974 struct l2t_entry *e;
3975 struct dst_entry *dst;
3976 struct c4iw_ep *lep = NULL;
3978 struct port_info *pi;
3979 struct net_device *pdev;
3980 u16 rss_qid, eth_hdr_len;
3983 struct neighbour *neigh;
3985 /* Drop all non-SYN packets */
3986 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3990 * Drop all packets which did not hit the filter.
3991 * Unlikely to happen.
3993 if (!(rss->filter_hit && rss->filter_tid))
3997 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3999 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4001 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4003 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
4007 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4009 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4012 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4015 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4018 pr_err("T%d Chip is not supported\n",
4019 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4023 if (eth_hdr_len == ETH_HLEN) {
4024 eh = (struct ethhdr *)(req + 1);
4025 iph = (struct iphdr *)(eh + 1);
4027 vlan_eh = (struct vlan_ethhdr *)(req + 1);
4028 iph = (struct iphdr *)(vlan_eh + 1);
4029 skb->vlan_tci = ntohs(cpl->vlan);
4032 if (iph->version != 0x4)
4035 tcph = (struct tcphdr *)(iph + 1);
4036 skb_set_network_header(skb, (void *)iph - (void *)rss);
4037 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4040 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
4041 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4042 ntohs(tcph->source), iph->tos);
4044 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
4047 pr_err("%s - failed to find dst entry!\n",
4051 neigh = dst_neigh_lookup_skb(dst, skb);
4054 pr_err("%s - failed to allocate neigh!\n",
4059 if (neigh->dev->flags & IFF_LOOPBACK) {
4060 pdev = ip_dev_find(&init_net, iph->daddr);
4061 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4063 pi = (struct port_info *)netdev_priv(pdev);
4064 tx_chan = cxgb4_port_chan(pdev);
4067 pdev = get_real_dev(neigh->dev);
4068 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4070 pi = (struct port_info *)netdev_priv(pdev);
4071 tx_chan = cxgb4_port_chan(pdev);
4073 neigh_release(neigh);
4075 pr_err("%s - failed to allocate l2t entry!\n",
4080 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4081 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4082 window = (__force u16) htons((__force u16)tcph->window);
4084 /* Calcuate filter portion for LE region. */
4085 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4086 dev->rdev.lldi.ports[0],
4090 * Synthesize the cpl_pass_accept_req. We have everything except the
4091 * TID. Once firmware sends a reply with TID we update the TID field
4092 * in cpl and pass it through the regular cpl_pass_accept_req path.
4094 build_cpl_pass_accept_req(skb, stid, iph->tos);
4095 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4096 tcph->source, ntohl(tcph->seq), filter, window,
4097 rss_qid, pi->port_id);
4098 cxgb4_l2t_release(e);
4103 c4iw_put_ep(&lep->com);
4108 * These are the real handlers that are called from a
4111 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4112 [CPL_ACT_ESTABLISH] = act_establish,
4113 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4114 [CPL_RX_DATA] = rx_data,
4115 [CPL_ABORT_RPL_RSS] = abort_rpl,
4116 [CPL_ABORT_RPL] = abort_rpl,
4117 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4118 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4119 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4120 [CPL_PASS_ESTABLISH] = pass_establish,
4121 [CPL_PEER_CLOSE] = peer_close,
4122 [CPL_ABORT_REQ_RSS] = peer_abort,
4123 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4124 [CPL_RDMA_TERMINATE] = terminate,
4125 [CPL_FW4_ACK] = fw4_ack,
4126 [CPL_FW6_MSG] = deferred_fw6_msg,
4127 [CPL_RX_PKT] = rx_pkt,
4128 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4129 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4132 static void process_timeout(struct c4iw_ep *ep)
4134 struct c4iw_qp_attributes attrs;
4137 mutex_lock(&ep->com.mutex);
4138 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4140 set_bit(TIMEDOUT, &ep->com.history);
4141 switch (ep->com.state) {
4143 connect_reply_upcall(ep, -ETIMEDOUT);
4152 if (ep->com.cm_id && ep->com.qp) {
4153 attrs.next_state = C4IW_QP_STATE_ERROR;
4154 c4iw_modify_qp(ep->com.qp->rhp,
4155 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4158 close_complete_upcall(ep, -ETIMEDOUT);
4164 * These states are expected if the ep timed out at the same
4165 * time as another thread was calling stop_ep_timer().
4166 * So we silently do nothing for these states.
4171 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4172 __func__, ep, ep->hwtid, ep->com.state);
4175 mutex_unlock(&ep->com.mutex);
4177 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4178 c4iw_put_ep(&ep->com);
4181 static void process_timedout_eps(void)
4185 spin_lock_irq(&timeout_lock);
4186 while (!list_empty(&timeout_list)) {
4187 struct list_head *tmp;
4189 tmp = timeout_list.next;
4193 spin_unlock_irq(&timeout_lock);
4194 ep = list_entry(tmp, struct c4iw_ep, entry);
4195 process_timeout(ep);
4196 spin_lock_irq(&timeout_lock);
4198 spin_unlock_irq(&timeout_lock);
4201 static void process_work(struct work_struct *work)
4203 struct sk_buff *skb = NULL;
4204 struct c4iw_dev *dev;
4205 struct cpl_act_establish *rpl;
4206 unsigned int opcode;
4209 process_timedout_eps();
4210 while ((skb = skb_dequeue(&rxq))) {
4212 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4213 opcode = rpl->ot.opcode;
4215 BUG_ON(!work_handlers[opcode]);
4216 ret = work_handlers[opcode](dev, skb);
4219 process_timedout_eps();
4223 static DECLARE_WORK(skb_work, process_work);
4225 static void ep_timeout(unsigned long arg)
4227 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4230 spin_lock(&timeout_lock);
4231 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4233 * Only insert if it is not already on the list.
4235 if (!ep->entry.next) {
4236 list_add_tail(&ep->entry, &timeout_list);
4240 spin_unlock(&timeout_lock);
4242 queue_work(workq, &skb_work);
4246 * All the CM events are handled on a work queue to have a safe context.
4248 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4252 * Save dev in the skb->cb area.
4254 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4257 * Queue the skb and schedule the worker thread.
4259 skb_queue_tail(&rxq, skb);
4260 queue_work(workq, &skb_work);
4264 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4266 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4268 if (rpl->status != CPL_ERR_NONE) {
4269 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4270 "for tid %u\n", rpl->status, GET_TID(rpl));
4276 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4278 struct cpl_fw6_msg *rpl = cplhdr(skb);
4279 struct c4iw_wr_wait *wr_waitp;
4282 PDBG("%s type %u\n", __func__, rpl->type);
4284 switch (rpl->type) {
4285 case FW6_TYPE_WR_RPL:
4286 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4287 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4288 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4290 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4294 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4298 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4306 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4308 struct cpl_abort_req_rss *req = cplhdr(skb);
4310 unsigned int tid = GET_TID(req);
4312 ep = get_ep_from_tid(dev, tid);
4313 /* This EP will be dereferenced in peer_abort() */
4315 printk(KERN_WARNING MOD
4316 "Abort on non-existent endpoint, tid %d\n", tid);
4320 if (is_neg_adv(req->status)) {
4321 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4322 __func__, ep->hwtid, req->status,
4323 neg_adv_str(req->status));
4326 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4329 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4336 * Most upcalls from the T4 Core go to sched() to
4337 * schedule the processing on a work queue.
4339 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4340 [CPL_ACT_ESTABLISH] = sched,
4341 [CPL_ACT_OPEN_RPL] = sched,
4342 [CPL_RX_DATA] = sched,
4343 [CPL_ABORT_RPL_RSS] = sched,
4344 [CPL_ABORT_RPL] = sched,
4345 [CPL_PASS_OPEN_RPL] = sched,
4346 [CPL_CLOSE_LISTSRV_RPL] = sched,
4347 [CPL_PASS_ACCEPT_REQ] = sched,
4348 [CPL_PASS_ESTABLISH] = sched,
4349 [CPL_PEER_CLOSE] = sched,
4350 [CPL_CLOSE_CON_RPL] = sched,
4351 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4352 [CPL_RDMA_TERMINATE] = sched,
4353 [CPL_FW4_ACK] = sched,
4354 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4355 [CPL_FW6_MSG] = fw6_msg,
4356 [CPL_RX_PKT] = sched
4359 int __init c4iw_cm_init(void)
4361 spin_lock_init(&timeout_lock);
4362 skb_queue_head_init(&rxq);
4364 workq = create_singlethread_workqueue("iw_cxgb4");
4371 void c4iw_cm_term(void)
4373 WARN_ON(!list_empty(&timeout_list));
4374 flush_workqueue(workq);
4375 destroy_workqueue(workq);