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 if (idr_is_empty(&ep->com.dev->hwtid_idr))
337 wake_up(&ep->com.dev->wait);
338 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
341 static void insert_ep_tid(struct c4iw_ep *ep)
345 spin_lock_irqsave(&ep->com.dev->lock, flags);
346 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
347 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
351 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
353 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
358 spin_lock_irqsave(&dev->lock, flags);
359 ep = idr_find(&dev->hwtid_idr, tid);
361 c4iw_get_ep(&ep->com);
362 spin_unlock_irqrestore(&dev->lock, flags);
367 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
369 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
372 struct c4iw_listen_ep *ep;
375 spin_lock_irqsave(&dev->lock, flags);
376 ep = idr_find(&dev->stid_idr, stid);
378 c4iw_get_ep(&ep->com);
379 spin_unlock_irqrestore(&dev->lock, flags);
383 void _c4iw_free_ep(struct kref *kref)
387 ep = container_of(kref, struct c4iw_ep, com.kref);
388 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
389 if (test_bit(QP_REFERENCED, &ep->com.flags))
391 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
392 if (ep->com.remote_addr.ss_family == AF_INET6) {
393 struct sockaddr_in6 *sin6 =
394 (struct sockaddr_in6 *)
398 ep->com.dev->rdev.lldi.ports[0],
399 (const u32 *)&sin6->sin6_addr.s6_addr,
402 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
403 dst_release(ep->dst);
404 cxgb4_l2t_release(ep->l2t);
406 kfree_skb(ep->mpa_skb);
408 if (!skb_queue_empty(&ep->com.ep_skb_list))
409 skb_queue_purge(&ep->com.ep_skb_list);
413 static void release_ep_resources(struct c4iw_ep *ep)
415 set_bit(RELEASE_RESOURCES, &ep->com.flags);
418 * If we have a hwtid, then remove it from the idr table
419 * so lookups will no longer find this endpoint. Otherwise
420 * we have a race where one thread finds the ep ptr just
421 * before the other thread is freeing the ep memory.
425 c4iw_put_ep(&ep->com);
428 static int status2errno(int status)
433 case CPL_ERR_CONN_RESET:
435 case CPL_ERR_ARP_MISS:
436 return -EHOSTUNREACH;
437 case CPL_ERR_CONN_TIMEDOUT:
439 case CPL_ERR_TCAM_FULL:
441 case CPL_ERR_CONN_EXIST:
449 * Try and reuse skbs already allocated...
451 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
453 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
456 skb_reset_transport_header(skb);
458 skb = alloc_skb(len, gfp);
460 t4_set_arp_err_handler(skb, NULL, NULL);
464 static struct net_device *get_real_dev(struct net_device *egress_dev)
466 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
469 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
473 egress_dev = get_real_dev(egress_dev);
474 for (i = 0; i < dev->rdev.lldi.nports; i++)
475 if (dev->rdev.lldi.ports[i] == egress_dev)
480 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
481 __u8 *peer_ip, __be16 local_port,
482 __be16 peer_port, u8 tos,
485 struct dst_entry *dst = NULL;
487 if (IS_ENABLED(CONFIG_IPV6)) {
490 memset(&fl6, 0, sizeof(fl6));
491 memcpy(&fl6.daddr, peer_ip, 16);
492 memcpy(&fl6.saddr, local_ip, 16);
493 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
494 fl6.flowi6_oif = sin6_scope_id;
495 dst = ip6_route_output(&init_net, NULL, &fl6);
498 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
499 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
509 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
510 __be32 peer_ip, __be16 local_port,
511 __be16 peer_port, u8 tos)
517 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
518 peer_port, local_port, IPPROTO_TCP,
522 n = dst_neigh_lookup(&rt->dst, &peer_ip);
525 if (!our_interface(dev, n->dev) &&
526 !(n->dev->flags & IFF_LOOPBACK)) {
528 dst_release(&rt->dst);
535 static void arp_failure_discard(void *handle, struct sk_buff *skb)
537 pr_err(MOD "ARP failure\n");
541 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
543 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
548 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
549 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
552 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
556 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
557 release_ep_resources(ep);
561 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
565 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
566 c4iw_put_ep(&ep->parent_ep->com);
567 release_ep_resources(ep);
572 * Fake up a special CPL opcode and call sched() so process_work() will call
573 * _put_ep_safe() in a safe context to free the ep resources. This is needed
574 * because ARP error handlers are called in an ATOMIC context, and
575 * _c4iw_free_ep() needs to block.
577 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
580 struct cpl_act_establish *rpl = cplhdr(skb);
582 /* Set our special ARP_FAILURE opcode */
583 rpl->ot.opcode = cpl;
586 * Save ep in the skb->cb area, after where sched() will save the dev
589 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
590 sched(ep->com.dev, skb);
593 /* Handle an ARP failure for an accept */
594 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
596 struct c4iw_ep *ep = handle;
598 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
601 __state_set(&ep->com, DEAD);
602 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
606 * Handle an ARP failure for an active open.
608 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
610 struct c4iw_ep *ep = handle;
612 printk(KERN_ERR MOD "ARP failure during connect\n");
613 connect_reply_upcall(ep, -EHOSTUNREACH);
614 __state_set(&ep->com, DEAD);
615 if (ep->com.remote_addr.ss_family == AF_INET6) {
616 struct sockaddr_in6 *sin6 =
617 (struct sockaddr_in6 *)&ep->com.local_addr;
618 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
619 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
621 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
622 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
623 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
627 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
630 static void abort_arp_failure(void *handle, struct sk_buff *skb)
633 struct c4iw_ep *ep = handle;
634 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
635 struct cpl_abort_req *req = cplhdr(skb);
637 PDBG("%s rdev %p\n", __func__, rdev);
638 req->cmd = CPL_ABORT_NO_RST;
639 ret = c4iw_ofld_send(rdev, skb);
641 __state_set(&ep->com, DEAD);
642 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
646 static int send_flowc(struct c4iw_ep *ep)
648 struct fw_flowc_wr *flowc;
649 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
651 u16 vlan = ep->l2t->vlan;
657 if (vlan == CPL_L2T_VLAN_NONE)
662 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
664 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
665 FW_FLOWC_WR_NPARAMS_V(nparams));
666 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
667 16)) | FW_WR_FLOWID_V(ep->hwtid));
669 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
670 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
671 (ep->com.dev->rdev.lldi.pf));
672 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
673 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
674 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
675 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
676 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
677 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
678 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
679 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
680 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
681 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
682 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
683 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
684 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
685 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
689 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
690 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
691 flowc->mnemval[8].val = cpu_to_be32(pri);
693 /* Pad WR to 16 byte boundary */
694 flowc->mnemval[8].mnemonic = 0;
695 flowc->mnemval[8].val = 0;
697 for (i = 0; i < 9; i++) {
698 flowc->mnemval[i].r4[0] = 0;
699 flowc->mnemval[i].r4[1] = 0;
700 flowc->mnemval[i].r4[2] = 0;
703 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
704 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
707 static int send_halfclose(struct c4iw_ep *ep)
709 struct cpl_close_con_req *req;
710 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
711 int wrlen = roundup(sizeof *req, 16);
713 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
717 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
718 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
719 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
720 memset(req, 0, wrlen);
721 INIT_TP_WR(req, ep->hwtid);
722 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
724 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
727 static int send_abort(struct c4iw_ep *ep)
729 struct cpl_abort_req *req;
730 int wrlen = roundup(sizeof *req, 16);
731 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
733 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
734 if (WARN_ON(!req_skb))
737 set_wr_txq(req_skb, CPL_PRIORITY_DATA, ep->txq_idx);
738 t4_set_arp_err_handler(req_skb, ep, abort_arp_failure);
739 req = (struct cpl_abort_req *)skb_put(req_skb, wrlen);
740 memset(req, 0, wrlen);
741 INIT_TP_WR(req, ep->hwtid);
742 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
743 req->cmd = CPL_ABORT_SEND_RST;
744 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
747 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
748 unsigned int *idx, int use_ts, int ipv6)
750 unsigned short hdr_size = (ipv6 ?
751 sizeof(struct ipv6hdr) :
752 sizeof(struct iphdr)) +
753 sizeof(struct tcphdr) +
755 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
756 unsigned short data_size = mtu - hdr_size;
758 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
761 static int send_connect(struct c4iw_ep *ep)
763 struct cpl_act_open_req *req = NULL;
764 struct cpl_t5_act_open_req *t5req = NULL;
765 struct cpl_t6_act_open_req *t6req = NULL;
766 struct cpl_act_open_req6 *req6 = NULL;
767 struct cpl_t5_act_open_req6 *t5req6 = NULL;
768 struct cpl_t6_act_open_req6 *t6req6 = NULL;
772 unsigned int mtu_idx;
774 int win, sizev4, sizev6, wrlen;
775 struct sockaddr_in *la = (struct sockaddr_in *)
777 struct sockaddr_in *ra = (struct sockaddr_in *)
778 &ep->com.remote_addr;
779 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
781 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
782 &ep->com.remote_addr;
784 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
785 u32 isn = (prandom_u32() & ~7UL) - 1;
787 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
789 sizev4 = sizeof(struct cpl_act_open_req);
790 sizev6 = sizeof(struct cpl_act_open_req6);
793 sizev4 = sizeof(struct cpl_t5_act_open_req);
794 sizev6 = sizeof(struct cpl_t5_act_open_req6);
797 sizev4 = sizeof(struct cpl_t6_act_open_req);
798 sizev6 = sizeof(struct cpl_t6_act_open_req6);
801 pr_err("T%d Chip is not supported\n",
802 CHELSIO_CHIP_VERSION(adapter_type));
806 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
807 roundup(sizev4, 16) :
810 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
812 skb = get_skb(NULL, wrlen, GFP_KERNEL);
814 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
818 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
820 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
821 enable_tcp_timestamps,
822 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
823 wscale = compute_wscale(rcv_win);
826 * Specify the largest window that will fit in opt0. The
827 * remainder will be specified in the rx_data_ack.
829 win = ep->rcv_win >> 10;
830 if (win > RCV_BUFSIZ_M)
833 opt0 = (nocong ? NO_CONG_F : 0) |
836 WND_SCALE_V(wscale) |
838 L2T_IDX_V(ep->l2t->idx) |
839 TX_CHAN_V(ep->tx_chan) |
840 SMAC_SEL_V(ep->smac_idx) |
841 DSCP_V(ep->tos >> 2) |
842 ULP_MODE_V(ULP_MODE_TCPDDP) |
844 opt2 = RX_CHANNEL_V(0) |
845 CCTRL_ECN_V(enable_ecn) |
846 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
847 if (enable_tcp_timestamps)
848 opt2 |= TSTAMPS_EN_F;
851 if (wscale && enable_tcp_window_scaling)
852 opt2 |= WND_SCALE_EN_F;
853 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
857 opt2 |= T5_OPT_2_VALID_F;
858 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
862 if (ep->com.remote_addr.ss_family == AF_INET6)
863 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
864 (const u32 *)&la6->sin6_addr.s6_addr, 1);
866 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
868 if (ep->com.remote_addr.ss_family == AF_INET) {
869 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
871 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
875 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
877 INIT_TP_WR(t5req, 0);
878 req = (struct cpl_act_open_req *)t5req;
881 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
883 INIT_TP_WR(t6req, 0);
884 req = (struct cpl_act_open_req *)t6req;
885 t5req = (struct cpl_t5_act_open_req *)t6req;
888 pr_err("T%d Chip is not supported\n",
889 CHELSIO_CHIP_VERSION(adapter_type));
894 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
895 ((ep->rss_qid<<14) | ep->atid)));
896 req->local_port = la->sin_port;
897 req->peer_port = ra->sin_port;
898 req->local_ip = la->sin_addr.s_addr;
899 req->peer_ip = ra->sin_addr.s_addr;
900 req->opt0 = cpu_to_be64(opt0);
902 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
903 req->params = cpu_to_be32(cxgb4_select_ntuple(
904 ep->com.dev->rdev.lldi.ports[0],
906 req->opt2 = cpu_to_be32(opt2);
908 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
910 ep->com.dev->rdev.lldi.ports[0],
912 t5req->rsvd = cpu_to_be32(isn);
913 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
914 t5req->opt2 = cpu_to_be32(opt2);
917 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
919 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
923 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
925 INIT_TP_WR(t5req6, 0);
926 req6 = (struct cpl_act_open_req6 *)t5req6;
929 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
931 INIT_TP_WR(t6req6, 0);
932 req6 = (struct cpl_act_open_req6 *)t6req6;
933 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
936 pr_err("T%d Chip is not supported\n",
937 CHELSIO_CHIP_VERSION(adapter_type));
942 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
943 ((ep->rss_qid<<14)|ep->atid)));
944 req6->local_port = la6->sin6_port;
945 req6->peer_port = ra6->sin6_port;
946 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
947 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
948 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
949 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
950 req6->opt0 = cpu_to_be64(opt0);
952 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
953 req6->params = cpu_to_be32(cxgb4_select_ntuple(
954 ep->com.dev->rdev.lldi.ports[0],
956 req6->opt2 = cpu_to_be32(opt2);
958 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
960 ep->com.dev->rdev.lldi.ports[0],
962 t5req6->rsvd = cpu_to_be32(isn);
963 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
964 t5req6->opt2 = cpu_to_be32(opt2);
968 set_bit(ACT_OPEN_REQ, &ep->com.history);
969 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
971 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
972 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
973 (const u32 *)&la6->sin6_addr.s6_addr, 1);
977 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
980 int mpalen, wrlen, ret;
981 struct fw_ofld_tx_data_wr *req;
982 struct mpa_message *mpa;
983 struct mpa_v2_conn_params mpa_v2_params;
985 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
987 BUG_ON(skb_cloned(skb));
989 mpalen = sizeof(*mpa) + ep->plen;
990 if (mpa_rev_to_use == 2)
991 mpalen += sizeof(struct mpa_v2_conn_params);
992 wrlen = roundup(mpalen + sizeof *req, 16);
993 skb = get_skb(skb, wrlen, GFP_KERNEL);
995 connect_reply_upcall(ep, -ENOMEM);
998 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1000 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1001 memset(req, 0, wrlen);
1002 req->op_to_immdlen = cpu_to_be32(
1003 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1005 FW_WR_IMMDLEN_V(mpalen));
1006 req->flowid_len16 = cpu_to_be32(
1007 FW_WR_FLOWID_V(ep->hwtid) |
1008 FW_WR_LEN16_V(wrlen >> 4));
1009 req->plen = cpu_to_be32(mpalen);
1010 req->tunnel_to_proxy = cpu_to_be32(
1011 FW_OFLD_TX_DATA_WR_FLUSH_F |
1012 FW_OFLD_TX_DATA_WR_SHOVE_F);
1014 mpa = (struct mpa_message *)(req + 1);
1015 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
1019 mpa->flags |= MPA_CRC;
1020 if (markers_enabled) {
1021 mpa->flags |= MPA_MARKERS;
1022 ep->mpa_attr.recv_marker_enabled = 1;
1024 ep->mpa_attr.recv_marker_enabled = 0;
1026 if (mpa_rev_to_use == 2)
1027 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1029 mpa->private_data_size = htons(ep->plen);
1030 mpa->revision = mpa_rev_to_use;
1031 if (mpa_rev_to_use == 1) {
1032 ep->tried_with_mpa_v1 = 1;
1033 ep->retry_with_mpa_v1 = 0;
1036 if (mpa_rev_to_use == 2) {
1037 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1038 sizeof (struct mpa_v2_conn_params));
1039 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1041 mpa_v2_params.ird = htons((u16)ep->ird);
1042 mpa_v2_params.ord = htons((u16)ep->ord);
1045 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1046 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1047 mpa_v2_params.ord |=
1048 htons(MPA_V2_RDMA_WRITE_RTR);
1049 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1050 mpa_v2_params.ord |=
1051 htons(MPA_V2_RDMA_READ_RTR);
1053 memcpy(mpa->private_data, &mpa_v2_params,
1054 sizeof(struct mpa_v2_conn_params));
1057 memcpy(mpa->private_data +
1058 sizeof(struct mpa_v2_conn_params),
1059 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1062 memcpy(mpa->private_data,
1063 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1066 * Reference the mpa skb. This ensures the data area
1067 * will remain in memory until the hw acks the tx.
1068 * Function fw4_ack() will deref it.
1071 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1072 BUG_ON(ep->mpa_skb);
1074 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1078 __state_set(&ep->com, MPA_REQ_SENT);
1079 ep->mpa_attr.initiator = 1;
1080 ep->snd_seq += mpalen;
1084 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1087 struct fw_ofld_tx_data_wr *req;
1088 struct mpa_message *mpa;
1089 struct sk_buff *skb;
1090 struct mpa_v2_conn_params mpa_v2_params;
1092 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1094 mpalen = sizeof(*mpa) + plen;
1095 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1096 mpalen += sizeof(struct mpa_v2_conn_params);
1097 wrlen = roundup(mpalen + sizeof *req, 16);
1099 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1101 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1104 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1106 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1107 memset(req, 0, wrlen);
1108 req->op_to_immdlen = cpu_to_be32(
1109 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1111 FW_WR_IMMDLEN_V(mpalen));
1112 req->flowid_len16 = cpu_to_be32(
1113 FW_WR_FLOWID_V(ep->hwtid) |
1114 FW_WR_LEN16_V(wrlen >> 4));
1115 req->plen = cpu_to_be32(mpalen);
1116 req->tunnel_to_proxy = cpu_to_be32(
1117 FW_OFLD_TX_DATA_WR_FLUSH_F |
1118 FW_OFLD_TX_DATA_WR_SHOVE_F);
1120 mpa = (struct mpa_message *)(req + 1);
1121 memset(mpa, 0, sizeof(*mpa));
1122 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1123 mpa->flags = MPA_REJECT;
1124 mpa->revision = ep->mpa_attr.version;
1125 mpa->private_data_size = htons(plen);
1127 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1128 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1129 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1130 sizeof (struct mpa_v2_conn_params));
1131 mpa_v2_params.ird = htons(((u16)ep->ird) |
1132 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1134 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1136 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1137 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1138 FW_RI_INIT_P2PTYPE_READ_REQ ?
1139 MPA_V2_RDMA_READ_RTR : 0) : 0));
1140 memcpy(mpa->private_data, &mpa_v2_params,
1141 sizeof(struct mpa_v2_conn_params));
1144 memcpy(mpa->private_data +
1145 sizeof(struct mpa_v2_conn_params), pdata, plen);
1148 memcpy(mpa->private_data, pdata, plen);
1151 * Reference the mpa skb again. This ensures the data area
1152 * will remain in memory until the hw acks the tx.
1153 * Function fw4_ack() will deref it.
1156 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1157 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1158 BUG_ON(ep->mpa_skb);
1160 ep->snd_seq += mpalen;
1161 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1164 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1167 struct fw_ofld_tx_data_wr *req;
1168 struct mpa_message *mpa;
1169 struct sk_buff *skb;
1170 struct mpa_v2_conn_params mpa_v2_params;
1172 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1174 mpalen = sizeof(*mpa) + plen;
1175 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1176 mpalen += sizeof(struct mpa_v2_conn_params);
1177 wrlen = roundup(mpalen + sizeof *req, 16);
1179 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1181 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1184 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1186 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1187 memset(req, 0, wrlen);
1188 req->op_to_immdlen = cpu_to_be32(
1189 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1191 FW_WR_IMMDLEN_V(mpalen));
1192 req->flowid_len16 = cpu_to_be32(
1193 FW_WR_FLOWID_V(ep->hwtid) |
1194 FW_WR_LEN16_V(wrlen >> 4));
1195 req->plen = cpu_to_be32(mpalen);
1196 req->tunnel_to_proxy = cpu_to_be32(
1197 FW_OFLD_TX_DATA_WR_FLUSH_F |
1198 FW_OFLD_TX_DATA_WR_SHOVE_F);
1200 mpa = (struct mpa_message *)(req + 1);
1201 memset(mpa, 0, sizeof(*mpa));
1202 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1204 if (ep->mpa_attr.crc_enabled)
1205 mpa->flags |= MPA_CRC;
1206 if (ep->mpa_attr.recv_marker_enabled)
1207 mpa->flags |= MPA_MARKERS;
1208 mpa->revision = ep->mpa_attr.version;
1209 mpa->private_data_size = htons(plen);
1211 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1212 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1213 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1214 sizeof (struct mpa_v2_conn_params));
1215 mpa_v2_params.ird = htons((u16)ep->ird);
1216 mpa_v2_params.ord = htons((u16)ep->ord);
1217 if (peer2peer && (ep->mpa_attr.p2p_type !=
1218 FW_RI_INIT_P2PTYPE_DISABLED)) {
1219 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1221 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1222 mpa_v2_params.ord |=
1223 htons(MPA_V2_RDMA_WRITE_RTR);
1224 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1225 mpa_v2_params.ord |=
1226 htons(MPA_V2_RDMA_READ_RTR);
1229 memcpy(mpa->private_data, &mpa_v2_params,
1230 sizeof(struct mpa_v2_conn_params));
1233 memcpy(mpa->private_data +
1234 sizeof(struct mpa_v2_conn_params), pdata, plen);
1237 memcpy(mpa->private_data, pdata, plen);
1240 * Reference the mpa skb. This ensures the data area
1241 * will remain in memory until the hw acks the tx.
1242 * Function fw4_ack() will deref it.
1245 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1247 __state_set(&ep->com, MPA_REP_SENT);
1248 ep->snd_seq += mpalen;
1249 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1252 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1255 struct cpl_act_establish *req = cplhdr(skb);
1256 unsigned int tid = GET_TID(req);
1257 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1258 struct tid_info *t = dev->rdev.lldi.tids;
1261 ep = lookup_atid(t, atid);
1263 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1264 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1266 mutex_lock(&ep->com.mutex);
1267 dst_confirm(ep->dst);
1269 /* setup the hwtid for this connection */
1271 cxgb4_insert_tid(t, ep, tid);
1274 ep->snd_seq = be32_to_cpu(req->snd_isn);
1275 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1277 set_emss(ep, ntohs(req->tcp_opt));
1279 /* dealloc the atid */
1280 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1281 cxgb4_free_atid(t, atid);
1282 set_bit(ACT_ESTAB, &ep->com.history);
1284 /* start MPA negotiation */
1285 ret = send_flowc(ep);
1288 if (ep->retry_with_mpa_v1)
1289 ret = send_mpa_req(ep, skb, 1);
1291 ret = send_mpa_req(ep, skb, mpa_rev);
1294 mutex_unlock(&ep->com.mutex);
1297 mutex_unlock(&ep->com.mutex);
1298 connect_reply_upcall(ep, -ENOMEM);
1299 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1303 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1305 struct iw_cm_event event;
1307 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1308 memset(&event, 0, sizeof(event));
1309 event.event = IW_CM_EVENT_CLOSE;
1310 event.status = status;
1311 if (ep->com.cm_id) {
1312 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1313 ep, ep->com.cm_id, ep->hwtid);
1314 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1315 deref_cm_id(&ep->com);
1316 set_bit(CLOSE_UPCALL, &ep->com.history);
1320 static void peer_close_upcall(struct c4iw_ep *ep)
1322 struct iw_cm_event event;
1324 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1325 memset(&event, 0, sizeof(event));
1326 event.event = IW_CM_EVENT_DISCONNECT;
1327 if (ep->com.cm_id) {
1328 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1329 ep, ep->com.cm_id, ep->hwtid);
1330 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1331 set_bit(DISCONN_UPCALL, &ep->com.history);
1335 static void peer_abort_upcall(struct c4iw_ep *ep)
1337 struct iw_cm_event event;
1339 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1340 memset(&event, 0, sizeof(event));
1341 event.event = IW_CM_EVENT_CLOSE;
1342 event.status = -ECONNRESET;
1343 if (ep->com.cm_id) {
1344 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1345 ep->com.cm_id, ep->hwtid);
1346 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1347 deref_cm_id(&ep->com);
1348 set_bit(ABORT_UPCALL, &ep->com.history);
1352 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1354 struct iw_cm_event event;
1356 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1357 memset(&event, 0, sizeof(event));
1358 event.event = IW_CM_EVENT_CONNECT_REPLY;
1359 event.status = status;
1360 memcpy(&event.local_addr, &ep->com.local_addr,
1361 sizeof(ep->com.local_addr));
1362 memcpy(&event.remote_addr, &ep->com.remote_addr,
1363 sizeof(ep->com.remote_addr));
1365 if ((status == 0) || (status == -ECONNREFUSED)) {
1366 if (!ep->tried_with_mpa_v1) {
1367 /* this means MPA_v2 is used */
1368 event.ord = ep->ird;
1369 event.ird = ep->ord;
1370 event.private_data_len = ep->plen -
1371 sizeof(struct mpa_v2_conn_params);
1372 event.private_data = ep->mpa_pkt +
1373 sizeof(struct mpa_message) +
1374 sizeof(struct mpa_v2_conn_params);
1376 /* this means MPA_v1 is used */
1377 event.ord = cur_max_read_depth(ep->com.dev);
1378 event.ird = cur_max_read_depth(ep->com.dev);
1379 event.private_data_len = ep->plen;
1380 event.private_data = ep->mpa_pkt +
1381 sizeof(struct mpa_message);
1385 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1387 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1388 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1391 deref_cm_id(&ep->com);
1394 static int connect_request_upcall(struct c4iw_ep *ep)
1396 struct iw_cm_event event;
1399 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1400 memset(&event, 0, sizeof(event));
1401 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1402 memcpy(&event.local_addr, &ep->com.local_addr,
1403 sizeof(ep->com.local_addr));
1404 memcpy(&event.remote_addr, &ep->com.remote_addr,
1405 sizeof(ep->com.remote_addr));
1406 event.provider_data = ep;
1407 if (!ep->tried_with_mpa_v1) {
1408 /* this means MPA_v2 is used */
1409 event.ord = ep->ord;
1410 event.ird = ep->ird;
1411 event.private_data_len = ep->plen -
1412 sizeof(struct mpa_v2_conn_params);
1413 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1414 sizeof(struct mpa_v2_conn_params);
1416 /* this means MPA_v1 is used. Send max supported */
1417 event.ord = cur_max_read_depth(ep->com.dev);
1418 event.ird = cur_max_read_depth(ep->com.dev);
1419 event.private_data_len = ep->plen;
1420 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1422 c4iw_get_ep(&ep->com);
1423 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1426 c4iw_put_ep(&ep->com);
1427 set_bit(CONNREQ_UPCALL, &ep->com.history);
1428 c4iw_put_ep(&ep->parent_ep->com);
1432 static void established_upcall(struct c4iw_ep *ep)
1434 struct iw_cm_event event;
1436 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1437 memset(&event, 0, sizeof(event));
1438 event.event = IW_CM_EVENT_ESTABLISHED;
1439 event.ird = ep->ord;
1440 event.ord = ep->ird;
1441 if (ep->com.cm_id) {
1442 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1443 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1444 set_bit(ESTAB_UPCALL, &ep->com.history);
1448 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1450 struct cpl_rx_data_ack *req;
1451 struct sk_buff *skb;
1452 int wrlen = roundup(sizeof *req, 16);
1454 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1455 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1457 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1462 * If we couldn't specify the entire rcv window at connection setup
1463 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1464 * then add the overage in to the credits returned.
1466 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1467 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1469 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1470 memset(req, 0, wrlen);
1471 INIT_TP_WR(req, ep->hwtid);
1472 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1474 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1476 RX_DACK_MODE_V(dack_mode));
1477 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1478 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1482 #define RELAXED_IRD_NEGOTIATION 1
1485 * process_mpa_reply - process streaming mode MPA reply
1489 * 0 upon success indicating a connect request was delivered to the ULP
1490 * or the mpa request is incomplete but valid so far.
1492 * 1 if a failure requires the caller to close the connection.
1494 * 2 if a failure requires the caller to abort the connection.
1496 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1498 struct mpa_message *mpa;
1499 struct mpa_v2_conn_params *mpa_v2_params;
1501 u16 resp_ird, resp_ord;
1502 u8 rtr_mismatch = 0, insuff_ird = 0;
1503 struct c4iw_qp_attributes attrs;
1504 enum c4iw_qp_attr_mask mask;
1508 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1511 * If we get more than the supported amount of private data
1512 * then we must fail this connection.
1514 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1516 goto err_stop_timer;
1520 * copy the new data into our accumulation buffer.
1522 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1524 ep->mpa_pkt_len += skb->len;
1527 * if we don't even have the mpa message, then bail.
1529 if (ep->mpa_pkt_len < sizeof(*mpa))
1531 mpa = (struct mpa_message *) ep->mpa_pkt;
1533 /* Validate MPA header. */
1534 if (mpa->revision > mpa_rev) {
1535 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1536 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1538 goto err_stop_timer;
1540 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1542 goto err_stop_timer;
1545 plen = ntohs(mpa->private_data_size);
1548 * Fail if there's too much private data.
1550 if (plen > MPA_MAX_PRIVATE_DATA) {
1552 goto err_stop_timer;
1556 * If plen does not account for pkt size
1558 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1560 goto err_stop_timer;
1563 ep->plen = (u8) plen;
1566 * If we don't have all the pdata yet, then bail.
1567 * We'll continue process when more data arrives.
1569 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1572 if (mpa->flags & MPA_REJECT) {
1573 err = -ECONNREFUSED;
1574 goto err_stop_timer;
1578 * Stop mpa timer. If it expired, then
1579 * we ignore the MPA reply. process_timeout()
1580 * will abort the connection.
1582 if (stop_ep_timer(ep))
1586 * If we get here we have accumulated the entire mpa
1587 * start reply message including private data. And
1588 * the MPA header is valid.
1590 __state_set(&ep->com, FPDU_MODE);
1591 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1592 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1593 ep->mpa_attr.version = mpa->revision;
1594 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1596 if (mpa->revision == 2) {
1597 ep->mpa_attr.enhanced_rdma_conn =
1598 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1599 if (ep->mpa_attr.enhanced_rdma_conn) {
1600 mpa_v2_params = (struct mpa_v2_conn_params *)
1601 (ep->mpa_pkt + sizeof(*mpa));
1602 resp_ird = ntohs(mpa_v2_params->ird) &
1603 MPA_V2_IRD_ORD_MASK;
1604 resp_ord = ntohs(mpa_v2_params->ord) &
1605 MPA_V2_IRD_ORD_MASK;
1606 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1607 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1610 * This is a double-check. Ideally, below checks are
1611 * not required since ird/ord stuff has been taken
1612 * care of in c4iw_accept_cr
1614 if (ep->ird < resp_ord) {
1615 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1616 ep->com.dev->rdev.lldi.max_ordird_qp)
1620 } else if (ep->ird > resp_ord) {
1623 if (ep->ord > resp_ird) {
1624 if (RELAXED_IRD_NEGOTIATION)
1635 if (ntohs(mpa_v2_params->ird) &
1636 MPA_V2_PEER2PEER_MODEL) {
1637 if (ntohs(mpa_v2_params->ord) &
1638 MPA_V2_RDMA_WRITE_RTR)
1639 ep->mpa_attr.p2p_type =
1640 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1641 else if (ntohs(mpa_v2_params->ord) &
1642 MPA_V2_RDMA_READ_RTR)
1643 ep->mpa_attr.p2p_type =
1644 FW_RI_INIT_P2PTYPE_READ_REQ;
1647 } else if (mpa->revision == 1)
1649 ep->mpa_attr.p2p_type = p2p_type;
1651 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1652 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1653 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1654 ep->mpa_attr.recv_marker_enabled,
1655 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1656 ep->mpa_attr.p2p_type, p2p_type);
1659 * If responder's RTR does not match with that of initiator, assign
1660 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1661 * generated when moving QP to RTS state.
1662 * A TERM message will be sent after QP has moved to RTS state
1664 if ((ep->mpa_attr.version == 2) && peer2peer &&
1665 (ep->mpa_attr.p2p_type != p2p_type)) {
1666 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1670 attrs.mpa_attr = ep->mpa_attr;
1671 attrs.max_ird = ep->ird;
1672 attrs.max_ord = ep->ord;
1673 attrs.llp_stream_handle = ep;
1674 attrs.next_state = C4IW_QP_STATE_RTS;
1676 mask = C4IW_QP_ATTR_NEXT_STATE |
1677 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1678 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1680 /* bind QP and TID with INIT_WR */
1681 err = c4iw_modify_qp(ep->com.qp->rhp,
1682 ep->com.qp, mask, &attrs, 1);
1687 * If responder's RTR requirement did not match with what initiator
1688 * supports, generate TERM message
1691 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1692 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1693 attrs.ecode = MPA_NOMATCH_RTR;
1694 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1695 attrs.send_term = 1;
1696 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1697 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1704 * Generate TERM if initiator IRD is not sufficient for responder
1705 * provided ORD. Currently, we do the same behaviour even when
1706 * responder provided IRD is also not sufficient as regards to
1710 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1712 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1713 attrs.ecode = MPA_INSUFF_IRD;
1714 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1715 attrs.send_term = 1;
1716 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1717 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1728 connect_reply_upcall(ep, err);
1733 * process_mpa_request - process streaming mode MPA request
1737 * 0 upon success indicating a connect request was delivered to the ULP
1738 * or the mpa request is incomplete but valid so far.
1740 * 1 if a failure requires the caller to close the connection.
1742 * 2 if a failure requires the caller to abort the connection.
1744 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1746 struct mpa_message *mpa;
1747 struct mpa_v2_conn_params *mpa_v2_params;
1750 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1753 * If we get more than the supported amount of private data
1754 * then we must fail this connection.
1756 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1757 goto err_stop_timer;
1759 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1762 * Copy the new data into our accumulation buffer.
1764 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1766 ep->mpa_pkt_len += skb->len;
1769 * If we don't even have the mpa message, then bail.
1770 * We'll continue process when more data arrives.
1772 if (ep->mpa_pkt_len < sizeof(*mpa))
1775 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1776 mpa = (struct mpa_message *) ep->mpa_pkt;
1779 * Validate MPA Header.
1781 if (mpa->revision > mpa_rev) {
1782 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1783 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1784 goto err_stop_timer;
1787 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1788 goto err_stop_timer;
1790 plen = ntohs(mpa->private_data_size);
1793 * Fail if there's too much private data.
1795 if (plen > MPA_MAX_PRIVATE_DATA)
1796 goto err_stop_timer;
1799 * If plen does not account for pkt size
1801 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1802 goto err_stop_timer;
1803 ep->plen = (u8) plen;
1806 * If we don't have all the pdata yet, then bail.
1808 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1812 * If we get here we have accumulated the entire mpa
1813 * start reply message including private data.
1815 ep->mpa_attr.initiator = 0;
1816 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1817 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1818 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1819 ep->mpa_attr.version = mpa->revision;
1820 if (mpa->revision == 1)
1821 ep->tried_with_mpa_v1 = 1;
1822 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1824 if (mpa->revision == 2) {
1825 ep->mpa_attr.enhanced_rdma_conn =
1826 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1827 if (ep->mpa_attr.enhanced_rdma_conn) {
1828 mpa_v2_params = (struct mpa_v2_conn_params *)
1829 (ep->mpa_pkt + sizeof(*mpa));
1830 ep->ird = ntohs(mpa_v2_params->ird) &
1831 MPA_V2_IRD_ORD_MASK;
1832 ep->ird = min_t(u32, ep->ird,
1833 cur_max_read_depth(ep->com.dev));
1834 ep->ord = ntohs(mpa_v2_params->ord) &
1835 MPA_V2_IRD_ORD_MASK;
1836 ep->ord = min_t(u32, ep->ord,
1837 cur_max_read_depth(ep->com.dev));
1838 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1840 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1842 if (ntohs(mpa_v2_params->ord) &
1843 MPA_V2_RDMA_WRITE_RTR)
1844 ep->mpa_attr.p2p_type =
1845 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1846 else if (ntohs(mpa_v2_params->ord) &
1847 MPA_V2_RDMA_READ_RTR)
1848 ep->mpa_attr.p2p_type =
1849 FW_RI_INIT_P2PTYPE_READ_REQ;
1852 } else if (mpa->revision == 1)
1854 ep->mpa_attr.p2p_type = p2p_type;
1856 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1857 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1858 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1859 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1860 ep->mpa_attr.p2p_type);
1862 __state_set(&ep->com, MPA_REQ_RCVD);
1865 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1866 if (ep->parent_ep->com.state != DEAD) {
1867 if (connect_request_upcall(ep))
1868 goto err_unlock_parent;
1870 goto err_unlock_parent;
1872 mutex_unlock(&ep->parent_ep->com.mutex);
1876 mutex_unlock(&ep->parent_ep->com.mutex);
1879 (void)stop_ep_timer(ep);
1884 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1887 struct cpl_rx_data *hdr = cplhdr(skb);
1888 unsigned int dlen = ntohs(hdr->len);
1889 unsigned int tid = GET_TID(hdr);
1890 __u8 status = hdr->status;
1893 ep = get_ep_from_tid(dev, tid);
1896 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1897 skb_pull(skb, sizeof(*hdr));
1898 skb_trim(skb, dlen);
1899 mutex_lock(&ep->com.mutex);
1901 /* update RX credits */
1902 update_rx_credits(ep, dlen);
1904 switch (ep->com.state) {
1906 ep->rcv_seq += dlen;
1907 disconnect = process_mpa_reply(ep, skb);
1910 ep->rcv_seq += dlen;
1911 disconnect = process_mpa_request(ep, skb);
1914 struct c4iw_qp_attributes attrs;
1915 BUG_ON(!ep->com.qp);
1917 pr_err("%s Unexpected streaming data." \
1918 " qpid %u ep %p state %d tid %u status %d\n",
1919 __func__, ep->com.qp->wq.sq.qid, ep,
1920 ep->com.state, ep->hwtid, status);
1921 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1922 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1923 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1930 mutex_unlock(&ep->com.mutex);
1932 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1933 c4iw_put_ep(&ep->com);
1937 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1940 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1942 unsigned int tid = GET_TID(rpl);
1944 ep = get_ep_from_tid(dev, tid);
1946 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1949 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1950 mutex_lock(&ep->com.mutex);
1951 switch (ep->com.state) {
1953 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1954 __state_set(&ep->com, DEAD);
1958 printk(KERN_ERR "%s ep %p state %d\n",
1959 __func__, ep, ep->com.state);
1962 mutex_unlock(&ep->com.mutex);
1965 release_ep_resources(ep);
1966 c4iw_put_ep(&ep->com);
1970 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1972 struct sk_buff *skb;
1973 struct fw_ofld_connection_wr *req;
1974 unsigned int mtu_idx;
1976 struct sockaddr_in *sin;
1979 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1980 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1981 memset(req, 0, sizeof(*req));
1982 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1983 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1984 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1985 ep->com.dev->rdev.lldi.ports[0],
1987 sin = (struct sockaddr_in *)&ep->com.local_addr;
1988 req->le.lport = sin->sin_port;
1989 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1990 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1991 req->le.pport = sin->sin_port;
1992 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1993 req->tcb.t_state_to_astid =
1994 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1995 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1996 req->tcb.cplrxdataack_cplpassacceptrpl =
1997 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1998 req->tcb.tx_max = (__force __be32) jiffies;
1999 req->tcb.rcv_adv = htons(1);
2000 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2001 enable_tcp_timestamps,
2002 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2003 wscale = compute_wscale(rcv_win);
2006 * Specify the largest window that will fit in opt0. The
2007 * remainder will be specified in the rx_data_ack.
2009 win = ep->rcv_win >> 10;
2010 if (win > RCV_BUFSIZ_M)
2013 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2014 (nocong ? NO_CONG_F : 0) |
2017 WND_SCALE_V(wscale) |
2018 MSS_IDX_V(mtu_idx) |
2019 L2T_IDX_V(ep->l2t->idx) |
2020 TX_CHAN_V(ep->tx_chan) |
2021 SMAC_SEL_V(ep->smac_idx) |
2022 DSCP_V(ep->tos >> 2) |
2023 ULP_MODE_V(ULP_MODE_TCPDDP) |
2025 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2026 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2028 CCTRL_ECN_V(enable_ecn) |
2029 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2030 if (enable_tcp_timestamps)
2031 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2032 if (enable_tcp_sack)
2033 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2034 if (wscale && enable_tcp_window_scaling)
2035 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2036 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2037 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2038 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2039 set_bit(ACT_OFLD_CONN, &ep->com.history);
2040 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2044 * Some of the error codes above implicitly indicate that there is no TID
2045 * allocated with the result of an ACT_OPEN. We use this predicate to make
2048 static inline int act_open_has_tid(int status)
2050 return (status != CPL_ERR_TCAM_PARITY &&
2051 status != CPL_ERR_TCAM_MISS &&
2052 status != CPL_ERR_TCAM_FULL &&
2053 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2054 status != CPL_ERR_CONN_EXIST);
2057 /* Returns whether a CPL status conveys negative advice.
2059 static int is_neg_adv(unsigned int status)
2061 return status == CPL_ERR_RTX_NEG_ADVICE ||
2062 status == CPL_ERR_PERSIST_NEG_ADVICE ||
2063 status == CPL_ERR_KEEPALV_NEG_ADVICE;
2066 static char *neg_adv_str(unsigned int status)
2069 case CPL_ERR_RTX_NEG_ADVICE:
2070 return "Retransmit timeout";
2071 case CPL_ERR_PERSIST_NEG_ADVICE:
2072 return "Persist timeout";
2073 case CPL_ERR_KEEPALV_NEG_ADVICE:
2074 return "Keepalive timeout";
2080 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2082 ep->snd_win = snd_win;
2083 ep->rcv_win = rcv_win;
2084 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2087 #define ACT_OPEN_RETRY_COUNT 2
2089 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2090 struct dst_entry *dst, struct c4iw_dev *cdev,
2091 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2093 struct neighbour *n;
2095 struct net_device *pdev;
2097 n = dst_neigh_lookup(dst, peer_ip);
2103 if (n->dev->flags & IFF_LOOPBACK) {
2105 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2106 else if (IS_ENABLED(CONFIG_IPV6))
2107 for_each_netdev(&init_net, pdev) {
2108 if (ipv6_chk_addr(&init_net,
2109 (struct in6_addr *)peer_ip,
2120 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2121 n, pdev, rt_tos2priority(tos));
2126 ep->mtu = pdev->mtu;
2127 ep->tx_chan = cxgb4_port_chan(pdev);
2128 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2129 cxgb4_port_viid(pdev));
2130 step = cdev->rdev.lldi.ntxq /
2131 cdev->rdev.lldi.nchan;
2132 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2133 step = cdev->rdev.lldi.nrxq /
2134 cdev->rdev.lldi.nchan;
2135 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2136 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2137 cxgb4_port_idx(pdev) * step];
2138 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2141 pdev = get_real_dev(n->dev);
2142 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2146 ep->mtu = dst_mtu(dst);
2147 ep->tx_chan = cxgb4_port_chan(pdev);
2148 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2149 cxgb4_port_viid(pdev));
2150 step = cdev->rdev.lldi.ntxq /
2151 cdev->rdev.lldi.nchan;
2152 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2153 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2154 step = cdev->rdev.lldi.nrxq /
2155 cdev->rdev.lldi.nchan;
2156 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2157 cxgb4_port_idx(pdev) * step];
2158 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2161 ep->retry_with_mpa_v1 = 0;
2162 ep->tried_with_mpa_v1 = 0;
2174 static int c4iw_reconnect(struct c4iw_ep *ep)
2178 struct sockaddr_in *laddr = (struct sockaddr_in *)
2179 &ep->com.cm_id->m_local_addr;
2180 struct sockaddr_in *raddr = (struct sockaddr_in *)
2181 &ep->com.cm_id->m_remote_addr;
2182 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2183 &ep->com.cm_id->m_local_addr;
2184 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2185 &ep->com.cm_id->m_remote_addr;
2189 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2190 init_timer(&ep->timer);
2191 c4iw_init_wr_wait(&ep->com.wr_wait);
2193 /* When MPA revision is different on nodes, the node with MPA_rev=2
2194 * tries to reconnect with MPA_rev 1 for the same EP through
2195 * c4iw_reconnect(), where the same EP is assigned with new tid for
2196 * further connection establishment. As we are using the same EP pointer
2197 * for reconnect, few skbs are used during the previous c4iw_connect(),
2198 * which leaves the EP with inadequate skbs for further
2199 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2200 * skb_list() during peer_abort(). Allocate skbs which is already used.
2202 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2203 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2209 * Allocate an active TID to initiate a TCP connection.
2211 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2212 if (ep->atid == -1) {
2213 pr_err("%s - cannot alloc atid.\n", __func__);
2217 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2220 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2221 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2222 raddr->sin_addr.s_addr, laddr->sin_port,
2223 raddr->sin_port, ep->com.cm_id->tos);
2225 ra = (__u8 *)&raddr->sin_addr;
2227 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2228 raddr6->sin6_addr.s6_addr,
2229 laddr6->sin6_port, raddr6->sin6_port, 0,
2230 raddr6->sin6_scope_id);
2232 ra = (__u8 *)&raddr6->sin6_addr;
2235 pr_err("%s - cannot find route.\n", __func__);
2236 err = -EHOSTUNREACH;
2239 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2240 ep->com.dev->rdev.lldi.adapter_type,
2241 ep->com.cm_id->tos);
2243 pr_err("%s - cannot alloc l2e.\n", __func__);
2247 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2248 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2251 state_set(&ep->com, CONNECTING);
2252 ep->tos = ep->com.cm_id->tos;
2254 /* send connect request to rnic */
2255 err = send_connect(ep);
2259 cxgb4_l2t_release(ep->l2t);
2261 dst_release(ep->dst);
2263 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2264 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2267 * remember to send notification to upper layer.
2268 * We are in here so the upper layer is not aware that this is
2269 * re-connect attempt and so, upper layer is still waiting for
2270 * response of 1st connect request.
2272 connect_reply_upcall(ep, -ECONNRESET);
2274 c4iw_put_ep(&ep->com);
2279 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2282 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2283 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2284 ntohl(rpl->atid_status)));
2285 struct tid_info *t = dev->rdev.lldi.tids;
2286 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2287 struct sockaddr_in *la;
2288 struct sockaddr_in *ra;
2289 struct sockaddr_in6 *la6;
2290 struct sockaddr_in6 *ra6;
2293 ep = lookup_atid(t, atid);
2294 la = (struct sockaddr_in *)&ep->com.local_addr;
2295 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2296 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2297 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2299 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2300 status, status2errno(status));
2302 if (is_neg_adv(status)) {
2303 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2304 __func__, atid, status, neg_adv_str(status));
2305 ep->stats.connect_neg_adv++;
2306 mutex_lock(&dev->rdev.stats.lock);
2307 dev->rdev.stats.neg_adv++;
2308 mutex_unlock(&dev->rdev.stats.lock);
2312 set_bit(ACT_OPEN_RPL, &ep->com.history);
2315 * Log interesting failures.
2318 case CPL_ERR_CONN_RESET:
2319 case CPL_ERR_CONN_TIMEDOUT:
2321 case CPL_ERR_TCAM_FULL:
2322 mutex_lock(&dev->rdev.stats.lock);
2323 dev->rdev.stats.tcam_full++;
2324 mutex_unlock(&dev->rdev.stats.lock);
2325 if (ep->com.local_addr.ss_family == AF_INET &&
2326 dev->rdev.lldi.enable_fw_ofld_conn) {
2327 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2328 ntohl(rpl->atid_status))));
2334 case CPL_ERR_CONN_EXIST:
2335 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2336 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2337 if (ep->com.remote_addr.ss_family == AF_INET6) {
2338 struct sockaddr_in6 *sin6 =
2339 (struct sockaddr_in6 *)
2340 &ep->com.local_addr;
2342 ep->com.dev->rdev.lldi.ports[0],
2344 &sin6->sin6_addr.s6_addr, 1);
2346 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2348 cxgb4_free_atid(t, atid);
2349 dst_release(ep->dst);
2350 cxgb4_l2t_release(ep->l2t);
2356 if (ep->com.local_addr.ss_family == AF_INET) {
2357 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2358 atid, status, status2errno(status),
2359 &la->sin_addr.s_addr, ntohs(la->sin_port),
2360 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2362 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2363 atid, status, status2errno(status),
2364 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2365 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2371 connect_reply_upcall(ep, status2errno(status));
2372 state_set(&ep->com, DEAD);
2374 if (ep->com.remote_addr.ss_family == AF_INET6) {
2375 struct sockaddr_in6 *sin6 =
2376 (struct sockaddr_in6 *)&ep->com.local_addr;
2377 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2378 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2380 if (status && act_open_has_tid(status))
2381 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2383 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2384 cxgb4_free_atid(t, atid);
2385 dst_release(ep->dst);
2386 cxgb4_l2t_release(ep->l2t);
2387 c4iw_put_ep(&ep->com);
2392 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2394 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2395 unsigned int stid = GET_TID(rpl);
2396 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2399 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2402 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2403 rpl->status, status2errno(rpl->status));
2404 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2405 c4iw_put_ep(&ep->com);
2410 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2412 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2413 unsigned int stid = GET_TID(rpl);
2414 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2416 PDBG("%s ep %p\n", __func__, ep);
2417 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2418 c4iw_put_ep(&ep->com);
2422 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2423 struct cpl_pass_accept_req *req)
2425 struct cpl_pass_accept_rpl *rpl;
2426 unsigned int mtu_idx;
2430 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2432 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2434 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2435 BUG_ON(skb_cloned(skb));
2439 if (!is_t4(adapter_type)) {
2440 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2442 INIT_TP_WR(rpl5, ep->hwtid);
2444 skb_trim(skb, sizeof(*rpl));
2445 INIT_TP_WR(rpl, ep->hwtid);
2447 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2450 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2451 enable_tcp_timestamps && req->tcpopt.tstamp,
2452 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2453 wscale = compute_wscale(rcv_win);
2456 * Specify the largest window that will fit in opt0. The
2457 * remainder will be specified in the rx_data_ack.
2459 win = ep->rcv_win >> 10;
2460 if (win > RCV_BUFSIZ_M)
2462 opt0 = (nocong ? NO_CONG_F : 0) |
2465 WND_SCALE_V(wscale) |
2466 MSS_IDX_V(mtu_idx) |
2467 L2T_IDX_V(ep->l2t->idx) |
2468 TX_CHAN_V(ep->tx_chan) |
2469 SMAC_SEL_V(ep->smac_idx) |
2470 DSCP_V(ep->tos >> 2) |
2471 ULP_MODE_V(ULP_MODE_TCPDDP) |
2473 opt2 = RX_CHANNEL_V(0) |
2474 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2476 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2477 opt2 |= TSTAMPS_EN_F;
2478 if (enable_tcp_sack && req->tcpopt.sack)
2480 if (wscale && enable_tcp_window_scaling)
2481 opt2 |= WND_SCALE_EN_F;
2483 const struct tcphdr *tcph;
2484 u32 hlen = ntohl(req->hdr_len);
2486 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2487 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2490 tcph = (const void *)(req + 1) +
2491 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2492 if (tcph->ece && tcph->cwr)
2493 opt2 |= CCTRL_ECN_V(1);
2495 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2496 u32 isn = (prandom_u32() & ~7UL) - 1;
2497 opt2 |= T5_OPT_2_VALID_F;
2498 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2501 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2504 rpl5->iss = cpu_to_be32(isn);
2505 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2508 rpl->opt0 = cpu_to_be64(opt0);
2509 rpl->opt2 = cpu_to_be32(opt2);
2510 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2511 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2513 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2516 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2518 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2519 BUG_ON(skb_cloned(skb));
2520 skb_trim(skb, sizeof(struct cpl_tid_release));
2521 release_tid(&dev->rdev, hwtid, skb);
2525 static void get_4tuple(struct cpl_pass_accept_req *req, enum chip_type type,
2526 int *iptype, __u8 *local_ip, __u8 *peer_ip,
2527 __be16 *local_port, __be16 *peer_port)
2529 int eth_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2530 ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2531 T6_ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2532 int ip_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2533 IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2534 T6_IP_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2535 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2536 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2537 struct tcphdr *tcp = (struct tcphdr *)
2538 ((u8 *)(req + 1) + eth_len + ip_len);
2540 if (ip->version == 4) {
2541 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2542 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2545 memcpy(peer_ip, &ip->saddr, 4);
2546 memcpy(local_ip, &ip->daddr, 4);
2548 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2549 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2552 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2553 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2555 *peer_port = tcp->source;
2556 *local_port = tcp->dest;
2561 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2563 struct c4iw_ep *child_ep = NULL, *parent_ep;
2564 struct cpl_pass_accept_req *req = cplhdr(skb);
2565 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2566 struct tid_info *t = dev->rdev.lldi.tids;
2567 unsigned int hwtid = GET_TID(req);
2568 struct dst_entry *dst;
2569 __u8 local_ip[16], peer_ip[16];
2570 __be16 local_port, peer_port;
2571 struct sockaddr_in6 *sin6;
2573 u16 peer_mss = ntohs(req->tcpopt.mss);
2575 unsigned short hdrs;
2576 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2578 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2580 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2584 if (state_read(&parent_ep->com) != LISTEN) {
2585 PDBG("%s - listening ep not in LISTEN\n", __func__);
2589 get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype,
2590 local_ip, peer_ip, &local_port, &peer_port);
2592 /* Find output route */
2594 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2595 , __func__, parent_ep, hwtid,
2596 local_ip, peer_ip, ntohs(local_port),
2597 ntohs(peer_port), peer_mss);
2598 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2599 local_port, peer_port,
2602 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2603 , __func__, parent_ep, hwtid,
2604 local_ip, peer_ip, ntohs(local_port),
2605 ntohs(peer_port), peer_mss);
2606 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2607 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2608 ((struct sockaddr_in6 *)
2609 &parent_ep->com.local_addr)->sin6_scope_id);
2612 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2617 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2619 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2625 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2626 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2628 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2635 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2636 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2637 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2638 child_ep->mtu = peer_mss + hdrs;
2640 skb_queue_head_init(&child_ep->com.ep_skb_list);
2641 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2644 state_set(&child_ep->com, CONNECTING);
2645 child_ep->com.dev = dev;
2646 child_ep->com.cm_id = NULL;
2649 struct sockaddr_in *sin = (struct sockaddr_in *)
2650 &child_ep->com.local_addr;
2652 sin->sin_family = PF_INET;
2653 sin->sin_port = local_port;
2654 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2656 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2657 sin->sin_family = PF_INET;
2658 sin->sin_port = ((struct sockaddr_in *)
2659 &parent_ep->com.local_addr)->sin_port;
2660 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2662 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2663 sin->sin_family = PF_INET;
2664 sin->sin_port = peer_port;
2665 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2667 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2668 sin6->sin6_family = PF_INET6;
2669 sin6->sin6_port = local_port;
2670 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2672 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2673 sin6->sin6_family = PF_INET6;
2674 sin6->sin6_port = ((struct sockaddr_in6 *)
2675 &parent_ep->com.local_addr)->sin6_port;
2676 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2678 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2679 sin6->sin6_family = PF_INET6;
2680 sin6->sin6_port = peer_port;
2681 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2684 c4iw_get_ep(&parent_ep->com);
2685 child_ep->parent_ep = parent_ep;
2686 child_ep->tos = tos;
2687 child_ep->dst = dst;
2688 child_ep->hwtid = hwtid;
2690 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2691 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2693 init_timer(&child_ep->timer);
2694 cxgb4_insert_tid(t, child_ep, hwtid);
2695 insert_ep_tid(child_ep);
2696 if (accept_cr(child_ep, skb, req)) {
2697 c4iw_put_ep(&parent_ep->com);
2698 release_ep_resources(child_ep);
2700 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2703 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2704 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2705 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2709 c4iw_put_ep(&child_ep->com);
2711 reject_cr(dev, hwtid, skb);
2713 c4iw_put_ep(&parent_ep->com);
2718 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2721 struct cpl_pass_establish *req = cplhdr(skb);
2722 unsigned int tid = GET_TID(req);
2725 ep = get_ep_from_tid(dev, tid);
2726 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2727 ep->snd_seq = be32_to_cpu(req->snd_isn);
2728 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2730 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2731 ntohs(req->tcp_opt));
2733 set_emss(ep, ntohs(req->tcp_opt));
2735 dst_confirm(ep->dst);
2736 mutex_lock(&ep->com.mutex);
2737 ep->com.state = MPA_REQ_WAIT;
2739 set_bit(PASS_ESTAB, &ep->com.history);
2740 ret = send_flowc(ep);
2741 mutex_unlock(&ep->com.mutex);
2743 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2744 c4iw_put_ep(&ep->com);
2749 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2751 struct cpl_peer_close *hdr = cplhdr(skb);
2753 struct c4iw_qp_attributes attrs;
2756 unsigned int tid = GET_TID(hdr);
2759 ep = get_ep_from_tid(dev, tid);
2763 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2764 dst_confirm(ep->dst);
2766 set_bit(PEER_CLOSE, &ep->com.history);
2767 mutex_lock(&ep->com.mutex);
2768 switch (ep->com.state) {
2770 __state_set(&ep->com, CLOSING);
2773 __state_set(&ep->com, CLOSING);
2774 connect_reply_upcall(ep, -ECONNRESET);
2779 * We're gonna mark this puppy DEAD, but keep
2780 * the reference on it until the ULP accepts or
2781 * rejects the CR. Also wake up anyone waiting
2782 * in rdma connection migration (see c4iw_accept_cr()).
2784 __state_set(&ep->com, CLOSING);
2785 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2786 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2789 __state_set(&ep->com, CLOSING);
2790 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2791 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2795 __state_set(&ep->com, CLOSING);
2796 attrs.next_state = C4IW_QP_STATE_CLOSING;
2797 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2798 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2799 if (ret != -ECONNRESET) {
2800 peer_close_upcall(ep);
2808 __state_set(&ep->com, MORIBUND);
2812 (void)stop_ep_timer(ep);
2813 if (ep->com.cm_id && ep->com.qp) {
2814 attrs.next_state = C4IW_QP_STATE_IDLE;
2815 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2816 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2818 close_complete_upcall(ep, 0);
2819 __state_set(&ep->com, DEAD);
2829 mutex_unlock(&ep->com.mutex);
2831 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2833 release_ep_resources(ep);
2834 c4iw_put_ep(&ep->com);
2838 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2840 struct cpl_abort_req_rss *req = cplhdr(skb);
2842 struct cpl_abort_rpl *rpl;
2843 struct sk_buff *rpl_skb;
2844 struct c4iw_qp_attributes attrs;
2847 unsigned int tid = GET_TID(req);
2849 ep = get_ep_from_tid(dev, tid);
2853 if (is_neg_adv(req->status)) {
2854 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2855 __func__, ep->hwtid, req->status,
2856 neg_adv_str(req->status));
2857 ep->stats.abort_neg_adv++;
2858 mutex_lock(&dev->rdev.stats.lock);
2859 dev->rdev.stats.neg_adv++;
2860 mutex_unlock(&dev->rdev.stats.lock);
2863 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2865 set_bit(PEER_ABORT, &ep->com.history);
2868 * Wake up any threads in rdma_init() or rdma_fini().
2869 * However, this is not needed if com state is just
2872 if (ep->com.state != MPA_REQ_SENT)
2873 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2875 mutex_lock(&ep->com.mutex);
2876 switch (ep->com.state) {
2878 c4iw_put_ep(&ep->parent_ep->com);
2881 (void)stop_ep_timer(ep);
2884 (void)stop_ep_timer(ep);
2885 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2886 connect_reply_upcall(ep, -ECONNRESET);
2889 * we just don't send notification upwards because we
2890 * want to retry with mpa_v1 without upper layers even
2893 * do some housekeeping so as to re-initiate the
2896 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2898 ep->retry_with_mpa_v1 = 1;
2910 if (ep->com.cm_id && ep->com.qp) {
2911 attrs.next_state = C4IW_QP_STATE_ERROR;
2912 ret = c4iw_modify_qp(ep->com.qp->rhp,
2913 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2917 "%s - qp <- error failed!\n",
2920 peer_abort_upcall(ep);
2925 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2926 mutex_unlock(&ep->com.mutex);
2932 dst_confirm(ep->dst);
2933 if (ep->com.state != ABORTING) {
2934 __state_set(&ep->com, DEAD);
2935 /* we don't release if we want to retry with mpa_v1 */
2936 if (!ep->retry_with_mpa_v1)
2939 mutex_unlock(&ep->com.mutex);
2941 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2942 if (WARN_ON(!rpl_skb)) {
2946 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2947 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2948 INIT_TP_WR(rpl, ep->hwtid);
2949 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2950 rpl->cmd = CPL_ABORT_NO_RST;
2951 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2954 release_ep_resources(ep);
2955 else if (ep->retry_with_mpa_v1) {
2956 if (ep->com.remote_addr.ss_family == AF_INET6) {
2957 struct sockaddr_in6 *sin6 =
2958 (struct sockaddr_in6 *)
2959 &ep->com.local_addr;
2961 ep->com.dev->rdev.lldi.ports[0],
2962 (const u32 *)&sin6->sin6_addr.s6_addr,
2965 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2966 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2967 dst_release(ep->dst);
2968 cxgb4_l2t_release(ep->l2t);
2973 c4iw_put_ep(&ep->com);
2974 /* Dereferencing ep, referenced in peer_abort_intr() */
2975 c4iw_put_ep(&ep->com);
2979 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2982 struct c4iw_qp_attributes attrs;
2983 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2985 unsigned int tid = GET_TID(rpl);
2987 ep = get_ep_from_tid(dev, tid);
2991 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2994 /* The cm_id may be null if we failed to connect */
2995 mutex_lock(&ep->com.mutex);
2996 set_bit(CLOSE_CON_RPL, &ep->com.history);
2997 switch (ep->com.state) {
2999 __state_set(&ep->com, MORIBUND);
3002 (void)stop_ep_timer(ep);
3003 if ((ep->com.cm_id) && (ep->com.qp)) {
3004 attrs.next_state = C4IW_QP_STATE_IDLE;
3005 c4iw_modify_qp(ep->com.qp->rhp,
3007 C4IW_QP_ATTR_NEXT_STATE,
3010 close_complete_upcall(ep, 0);
3011 __state_set(&ep->com, DEAD);
3021 mutex_unlock(&ep->com.mutex);
3023 release_ep_resources(ep);
3024 c4iw_put_ep(&ep->com);
3028 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3030 struct cpl_rdma_terminate *rpl = cplhdr(skb);
3031 unsigned int tid = GET_TID(rpl);
3033 struct c4iw_qp_attributes attrs;
3035 ep = get_ep_from_tid(dev, tid);
3038 if (ep && ep->com.qp) {
3039 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
3040 ep->com.qp->wq.sq.qid);
3041 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3042 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3043 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3045 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
3046 c4iw_put_ep(&ep->com);
3052 * Upcall from the adapter indicating data has been transmitted.
3053 * For us its just the single MPA request or reply. We can now free
3054 * the skb holding the mpa message.
3056 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3059 struct cpl_fw4_ack *hdr = cplhdr(skb);
3060 u8 credits = hdr->credits;
3061 unsigned int tid = GET_TID(hdr);
3064 ep = get_ep_from_tid(dev, tid);
3067 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
3069 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
3070 __func__, ep, ep->hwtid, state_read(&ep->com));
3074 dst_confirm(ep->dst);
3076 PDBG("%s last streaming msg ack ep %p tid %u state %u "
3077 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
3078 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
3079 mutex_lock(&ep->com.mutex);
3080 kfree_skb(ep->mpa_skb);
3082 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3084 mutex_unlock(&ep->com.mutex);
3087 c4iw_put_ep(&ep->com);
3091 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3094 struct c4iw_ep *ep = to_ep(cm_id);
3096 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3098 mutex_lock(&ep->com.mutex);
3099 if (ep->com.state != MPA_REQ_RCVD) {
3100 mutex_unlock(&ep->com.mutex);
3101 c4iw_put_ep(&ep->com);
3104 set_bit(ULP_REJECT, &ep->com.history);
3108 abort = send_mpa_reject(ep, pdata, pdata_len);
3109 mutex_unlock(&ep->com.mutex);
3112 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3113 c4iw_put_ep(&ep->com);
3117 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3120 struct c4iw_qp_attributes attrs;
3121 enum c4iw_qp_attr_mask mask;
3122 struct c4iw_ep *ep = to_ep(cm_id);
3123 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3124 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3127 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3129 mutex_lock(&ep->com.mutex);
3130 if (ep->com.state != MPA_REQ_RCVD) {
3137 set_bit(ULP_ACCEPT, &ep->com.history);
3138 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3139 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3144 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3145 if (conn_param->ord > ep->ird) {
3146 if (RELAXED_IRD_NEGOTIATION) {
3147 conn_param->ord = ep->ird;
3149 ep->ird = conn_param->ird;
3150 ep->ord = conn_param->ord;
3151 send_mpa_reject(ep, conn_param->private_data,
3152 conn_param->private_data_len);
3157 if (conn_param->ird < ep->ord) {
3158 if (RELAXED_IRD_NEGOTIATION &&
3159 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3160 conn_param->ird = ep->ord;
3167 ep->ird = conn_param->ird;
3168 ep->ord = conn_param->ord;
3170 if (ep->mpa_attr.version == 1) {
3171 if (peer2peer && ep->ird == 0)
3175 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3176 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3180 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3182 ep->com.cm_id = cm_id;
3183 ref_cm_id(&ep->com);
3187 /* bind QP to EP and move to RTS */
3188 attrs.mpa_attr = ep->mpa_attr;
3189 attrs.max_ird = ep->ird;
3190 attrs.max_ord = ep->ord;
3191 attrs.llp_stream_handle = ep;
3192 attrs.next_state = C4IW_QP_STATE_RTS;
3194 /* bind QP and TID with INIT_WR */
3195 mask = C4IW_QP_ATTR_NEXT_STATE |
3196 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3197 C4IW_QP_ATTR_MPA_ATTR |
3198 C4IW_QP_ATTR_MAX_IRD |
3199 C4IW_QP_ATTR_MAX_ORD;
3201 err = c4iw_modify_qp(ep->com.qp->rhp,
3202 ep->com.qp, mask, &attrs, 1);
3204 goto err_deref_cm_id;
3206 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3207 err = send_mpa_reply(ep, conn_param->private_data,
3208 conn_param->private_data_len);
3210 goto err_deref_cm_id;
3212 __state_set(&ep->com, FPDU_MODE);
3213 established_upcall(ep);
3214 mutex_unlock(&ep->com.mutex);
3215 c4iw_put_ep(&ep->com);
3218 deref_cm_id(&ep->com);
3222 mutex_unlock(&ep->com.mutex);
3224 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3225 c4iw_put_ep(&ep->com);
3229 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3231 struct in_device *ind;
3233 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3234 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3236 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3238 return -EADDRNOTAVAIL;
3239 for_primary_ifa(ind) {
3240 laddr->sin_addr.s_addr = ifa->ifa_address;
3241 raddr->sin_addr.s_addr = ifa->ifa_address;
3247 return found ? 0 : -EADDRNOTAVAIL;
3250 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3251 unsigned char banned_flags)
3253 struct inet6_dev *idev;
3254 int err = -EADDRNOTAVAIL;
3257 idev = __in6_dev_get(dev);
3259 struct inet6_ifaddr *ifp;
3261 read_lock_bh(&idev->lock);
3262 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3263 if (ifp->scope == IFA_LINK &&
3264 !(ifp->flags & banned_flags)) {
3265 memcpy(addr, &ifp->addr, 16);
3270 read_unlock_bh(&idev->lock);
3276 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3278 struct in6_addr uninitialized_var(addr);
3279 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3280 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3282 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3283 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3284 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3287 return -EADDRNOTAVAIL;
3290 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3292 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3295 struct sockaddr_in *laddr;
3296 struct sockaddr_in *raddr;
3297 struct sockaddr_in6 *laddr6;
3298 struct sockaddr_in6 *raddr6;
3302 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3303 (conn_param->ird > cur_max_read_depth(dev))) {
3307 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3309 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3314 skb_queue_head_init(&ep->com.ep_skb_list);
3315 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3320 init_timer(&ep->timer);
3321 ep->plen = conn_param->private_data_len;
3323 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3324 conn_param->private_data, ep->plen);
3325 ep->ird = conn_param->ird;
3326 ep->ord = conn_param->ord;
3328 if (peer2peer && ep->ord == 0)
3331 ep->com.cm_id = cm_id;
3332 ref_cm_id(&ep->com);
3334 ep->com.qp = get_qhp(dev, conn_param->qpn);
3336 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3341 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3345 * Allocate an active TID to initiate a TCP connection.
3347 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3348 if (ep->atid == -1) {
3349 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3353 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3355 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3356 sizeof(ep->com.local_addr));
3357 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3358 sizeof(ep->com.remote_addr));
3360 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3361 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3362 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3363 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3365 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3367 ra = (__u8 *)&raddr->sin_addr;
3370 * Handle loopback requests to INADDR_ANY.
3372 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3373 err = pick_local_ipaddrs(dev, cm_id);
3379 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3380 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3381 ra, ntohs(raddr->sin_port));
3382 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3383 raddr->sin_addr.s_addr, laddr->sin_port,
3384 raddr->sin_port, cm_id->tos);
3387 ra = (__u8 *)&raddr6->sin6_addr;
3390 * Handle loopback requests to INADDR_ANY.
3392 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3393 err = pick_local_ip6addrs(dev, cm_id);
3399 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3400 __func__, laddr6->sin6_addr.s6_addr,
3401 ntohs(laddr6->sin6_port),
3402 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3403 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3404 raddr6->sin6_addr.s6_addr,
3405 laddr6->sin6_port, raddr6->sin6_port, 0,
3406 raddr6->sin6_scope_id);
3409 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3410 err = -EHOSTUNREACH;
3414 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3415 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3417 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3421 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3422 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3425 state_set(&ep->com, CONNECTING);
3426 ep->tos = cm_id->tos;
3428 /* send connect request to rnic */
3429 err = send_connect(ep);
3433 cxgb4_l2t_release(ep->l2t);
3435 dst_release(ep->dst);
3437 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3438 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3440 skb_queue_purge(&ep->com.ep_skb_list);
3441 deref_cm_id(&ep->com);
3443 c4iw_put_ep(&ep->com);
3448 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3451 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3452 &ep->com.local_addr;
3454 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3455 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3456 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3460 c4iw_init_wr_wait(&ep->com.wr_wait);
3461 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3462 ep->stid, &sin6->sin6_addr,
3464 ep->com.dev->rdev.lldi.rxq_ids[0]);
3466 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3470 err = net_xmit_errno(err);
3472 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3473 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3474 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3476 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3481 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3484 struct sockaddr_in *sin = (struct sockaddr_in *)
3485 &ep->com.local_addr;
3487 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3489 err = cxgb4_create_server_filter(
3490 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3491 sin->sin_addr.s_addr, sin->sin_port, 0,
3492 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3493 if (err == -EBUSY) {
3494 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3498 set_current_state(TASK_UNINTERRUPTIBLE);
3499 schedule_timeout(usecs_to_jiffies(100));
3501 } while (err == -EBUSY);
3503 c4iw_init_wr_wait(&ep->com.wr_wait);
3504 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3505 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3506 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3508 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3512 err = net_xmit_errno(err);
3515 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3517 &sin->sin_addr, ntohs(sin->sin_port));
3521 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3524 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3525 struct c4iw_listen_ep *ep;
3529 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3531 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3535 skb_queue_head_init(&ep->com.ep_skb_list);
3536 PDBG("%s ep %p\n", __func__, ep);
3537 ep->com.cm_id = cm_id;
3538 ref_cm_id(&ep->com);
3540 ep->backlog = backlog;
3541 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3542 sizeof(ep->com.local_addr));
3545 * Allocate a server TID.
3547 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3548 ep->com.local_addr.ss_family == AF_INET)
3549 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3550 cm_id->m_local_addr.ss_family, ep);
3552 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3553 cm_id->m_local_addr.ss_family, ep);
3555 if (ep->stid == -1) {
3556 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3560 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3562 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3563 sizeof(ep->com.local_addr));
3565 state_set(&ep->com, LISTEN);
3566 if (ep->com.local_addr.ss_family == AF_INET)
3567 err = create_server4(dev, ep);
3569 err = create_server6(dev, ep);
3571 cm_id->provider_data = ep;
3575 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3576 ep->com.local_addr.ss_family);
3578 deref_cm_id(&ep->com);
3579 c4iw_put_ep(&ep->com);
3585 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3588 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3590 PDBG("%s ep %p\n", __func__, ep);
3593 state_set(&ep->com, DEAD);
3594 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3595 ep->com.local_addr.ss_family == AF_INET) {
3596 err = cxgb4_remove_server_filter(
3597 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3598 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3600 struct sockaddr_in6 *sin6;
3601 c4iw_init_wr_wait(&ep->com.wr_wait);
3602 err = cxgb4_remove_server(
3603 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3604 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3607 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3609 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3610 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3611 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3613 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3614 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3615 ep->com.local_addr.ss_family);
3617 deref_cm_id(&ep->com);
3618 c4iw_put_ep(&ep->com);
3622 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3627 struct c4iw_rdev *rdev;
3629 mutex_lock(&ep->com.mutex);
3631 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3632 states[ep->com.state], abrupt);
3635 * Ref the ep here in case we have fatal errors causing the
3636 * ep to be released and freed.
3638 c4iw_get_ep(&ep->com);
3640 rdev = &ep->com.dev->rdev;
3641 if (c4iw_fatal_error(rdev)) {
3643 close_complete_upcall(ep, -EIO);
3644 ep->com.state = DEAD;
3646 switch (ep->com.state) {
3655 ep->com.state = ABORTING;
3657 ep->com.state = CLOSING;
3660 * if we close before we see the fw4_ack() then we fix
3661 * up the timer state since we're reusing it.
3664 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3665 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3670 set_bit(CLOSE_SENT, &ep->com.flags);
3673 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3676 (void)stop_ep_timer(ep);
3677 ep->com.state = ABORTING;
3679 ep->com.state = MORIBUND;
3685 PDBG("%s ignoring disconnect ep %p state %u\n",
3686 __func__, ep, ep->com.state);
3695 set_bit(EP_DISC_ABORT, &ep->com.history);
3696 close_complete_upcall(ep, -ECONNRESET);
3697 ret = send_abort(ep);
3699 set_bit(EP_DISC_CLOSE, &ep->com.history);
3700 ret = send_halfclose(ep);
3703 set_bit(EP_DISC_FAIL, &ep->com.history);
3706 close_complete_upcall(ep, -EIO);
3709 struct c4iw_qp_attributes attrs;
3711 attrs.next_state = C4IW_QP_STATE_ERROR;
3712 ret = c4iw_modify_qp(ep->com.qp->rhp,
3714 C4IW_QP_ATTR_NEXT_STATE,
3718 "%s - qp <- error failed!\n",
3724 mutex_unlock(&ep->com.mutex);
3725 c4iw_put_ep(&ep->com);
3727 release_ep_resources(ep);
3731 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3732 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3735 int atid = be32_to_cpu(req->tid);
3737 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3738 (__force u32) req->tid);
3742 switch (req->retval) {
3744 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3745 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3746 send_fw_act_open_req(ep, atid);
3750 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3751 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3752 send_fw_act_open_req(ep, atid);
3757 pr_info("%s unexpected ofld conn wr retval %d\n",
3758 __func__, req->retval);
3761 pr_err("active ofld_connect_wr failure %d atid %d\n",
3763 mutex_lock(&dev->rdev.stats.lock);
3764 dev->rdev.stats.act_ofld_conn_fails++;
3765 mutex_unlock(&dev->rdev.stats.lock);
3766 connect_reply_upcall(ep, status2errno(req->retval));
3767 state_set(&ep->com, DEAD);
3768 if (ep->com.remote_addr.ss_family == AF_INET6) {
3769 struct sockaddr_in6 *sin6 =
3770 (struct sockaddr_in6 *)&ep->com.local_addr;
3771 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3772 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3774 remove_handle(dev, &dev->atid_idr, atid);
3775 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3776 dst_release(ep->dst);
3777 cxgb4_l2t_release(ep->l2t);
3778 c4iw_put_ep(&ep->com);
3781 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3782 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3784 struct sk_buff *rpl_skb;
3785 struct cpl_pass_accept_req *cpl;
3788 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3791 PDBG("%s passive open failure %d\n", __func__, req->retval);
3792 mutex_lock(&dev->rdev.stats.lock);
3793 dev->rdev.stats.pas_ofld_conn_fails++;
3794 mutex_unlock(&dev->rdev.stats.lock);
3797 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3798 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3799 (__force u32) htonl(
3800 (__force u32) req->tid)));
3801 ret = pass_accept_req(dev, rpl_skb);
3808 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3810 struct cpl_fw6_msg *rpl = cplhdr(skb);
3811 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3813 switch (rpl->type) {
3815 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3817 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3818 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3819 switch (req->t_state) {
3821 active_ofld_conn_reply(dev, skb, req);
3824 passive_ofld_conn_reply(dev, skb, req);
3827 pr_err("%s unexpected ofld conn wr state %d\n",
3828 __func__, req->t_state);
3836 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3839 __be16 hdr_len, vlantag, len;
3841 int tcp_hdr_len, ip_hdr_len;
3843 struct cpl_rx_pkt *cpl = cplhdr(skb);
3844 struct cpl_pass_accept_req *req;
3845 struct tcp_options_received tmp_opt;
3846 struct c4iw_dev *dev;
3847 enum chip_type type;
3849 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3850 /* Store values from cpl_rx_pkt in temporary location. */
3851 vlantag = cpl->vlan;
3853 l2info = cpl->l2info;
3854 hdr_len = cpl->hdr_len;
3857 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3860 * We need to parse the TCP options from SYN packet.
3861 * to generate cpl_pass_accept_req.
3863 memset(&tmp_opt, 0, sizeof(tmp_opt));
3864 tcp_clear_options(&tmp_opt);
3865 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3867 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3868 memset(req, 0, sizeof(*req));
3869 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3870 SYN_MAC_IDX_V(RX_MACIDX_G(
3871 be32_to_cpu(l2info))) |
3873 type = dev->rdev.lldi.adapter_type;
3874 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3875 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3877 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3878 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3879 eth_hdr_len = is_t4(type) ?
3880 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3881 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3882 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3883 IP_HDR_LEN_V(ip_hdr_len) |
3884 ETH_HDR_LEN_V(eth_hdr_len));
3885 } else { /* T6 and later */
3886 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3887 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3888 T6_IP_HDR_LEN_V(ip_hdr_len) |
3889 T6_ETH_HDR_LEN_V(eth_hdr_len));
3891 req->vlan = vlantag;
3893 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3894 PASS_OPEN_TOS_V(tos));
3895 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3896 if (tmp_opt.wscale_ok)
3897 req->tcpopt.wsf = tmp_opt.snd_wscale;
3898 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3899 if (tmp_opt.sack_ok)
3900 req->tcpopt.sack = 1;
3901 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3905 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3906 __be32 laddr, __be16 lport,
3907 __be32 raddr, __be16 rport,
3908 u32 rcv_isn, u32 filter, u16 window,
3909 u32 rss_qid, u8 port_id)
3911 struct sk_buff *req_skb;
3912 struct fw_ofld_connection_wr *req;
3913 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3916 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3917 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3918 memset(req, 0, sizeof(*req));
3919 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3920 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3921 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3922 req->le.filter = (__force __be32) filter;
3923 req->le.lport = lport;
3924 req->le.pport = rport;
3925 req->le.u.ipv4.lip = laddr;
3926 req->le.u.ipv4.pip = raddr;
3927 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3928 req->tcb.rcv_adv = htons(window);
3929 req->tcb.t_state_to_astid =
3930 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3931 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3932 FW_OFLD_CONNECTION_WR_ASTID_V(
3933 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3936 * We store the qid in opt2 which will be used by the firmware
3937 * to send us the wr response.
3939 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3942 * We initialize the MSS index in TCB to 0xF.
3943 * So that when driver sends cpl_pass_accept_rpl
3944 * TCB picks up the correct value. If this was 0
3945 * TP will ignore any value > 0 for MSS index.
3947 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3948 req->cookie = (uintptr_t)skb;
3950 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3951 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3953 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3961 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3962 * messages when a filter is being used instead of server to
3963 * redirect a syn packet. When packets hit filter they are redirected
3964 * to the offload queue and driver tries to establish the connection
3965 * using firmware work request.
3967 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3970 unsigned int filter;
3971 struct ethhdr *eh = NULL;
3972 struct vlan_ethhdr *vlan_eh = NULL;
3974 struct tcphdr *tcph;
3975 struct rss_header *rss = (void *)skb->data;
3976 struct cpl_rx_pkt *cpl = (void *)skb->data;
3977 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3978 struct l2t_entry *e;
3979 struct dst_entry *dst;
3980 struct c4iw_ep *lep = NULL;
3982 struct port_info *pi;
3983 struct net_device *pdev;
3984 u16 rss_qid, eth_hdr_len;
3987 struct neighbour *neigh;
3989 /* Drop all non-SYN packets */
3990 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3994 * Drop all packets which did not hit the filter.
3995 * Unlikely to happen.
3997 if (!(rss->filter_hit && rss->filter_tid))
4001 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4003 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4005 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4007 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
4011 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4013 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4016 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4019 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4022 pr_err("T%d Chip is not supported\n",
4023 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4027 if (eth_hdr_len == ETH_HLEN) {
4028 eh = (struct ethhdr *)(req + 1);
4029 iph = (struct iphdr *)(eh + 1);
4031 vlan_eh = (struct vlan_ethhdr *)(req + 1);
4032 iph = (struct iphdr *)(vlan_eh + 1);
4033 skb->vlan_tci = ntohs(cpl->vlan);
4036 if (iph->version != 0x4)
4039 tcph = (struct tcphdr *)(iph + 1);
4040 skb_set_network_header(skb, (void *)iph - (void *)rss);
4041 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4044 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
4045 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4046 ntohs(tcph->source), iph->tos);
4048 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
4051 pr_err("%s - failed to find dst entry!\n",
4055 neigh = dst_neigh_lookup_skb(dst, skb);
4058 pr_err("%s - failed to allocate neigh!\n",
4063 if (neigh->dev->flags & IFF_LOOPBACK) {
4064 pdev = ip_dev_find(&init_net, iph->daddr);
4065 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4067 pi = (struct port_info *)netdev_priv(pdev);
4068 tx_chan = cxgb4_port_chan(pdev);
4071 pdev = get_real_dev(neigh->dev);
4072 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4074 pi = (struct port_info *)netdev_priv(pdev);
4075 tx_chan = cxgb4_port_chan(pdev);
4077 neigh_release(neigh);
4079 pr_err("%s - failed to allocate l2t entry!\n",
4084 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4085 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4086 window = (__force u16) htons((__force u16)tcph->window);
4088 /* Calcuate filter portion for LE region. */
4089 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4090 dev->rdev.lldi.ports[0],
4094 * Synthesize the cpl_pass_accept_req. We have everything except the
4095 * TID. Once firmware sends a reply with TID we update the TID field
4096 * in cpl and pass it through the regular cpl_pass_accept_req path.
4098 build_cpl_pass_accept_req(skb, stid, iph->tos);
4099 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4100 tcph->source, ntohl(tcph->seq), filter, window,
4101 rss_qid, pi->port_id);
4102 cxgb4_l2t_release(e);
4107 c4iw_put_ep(&lep->com);
4112 * These are the real handlers that are called from a
4115 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4116 [CPL_ACT_ESTABLISH] = act_establish,
4117 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4118 [CPL_RX_DATA] = rx_data,
4119 [CPL_ABORT_RPL_RSS] = abort_rpl,
4120 [CPL_ABORT_RPL] = abort_rpl,
4121 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4122 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4123 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4124 [CPL_PASS_ESTABLISH] = pass_establish,
4125 [CPL_PEER_CLOSE] = peer_close,
4126 [CPL_ABORT_REQ_RSS] = peer_abort,
4127 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4128 [CPL_RDMA_TERMINATE] = terminate,
4129 [CPL_FW4_ACK] = fw4_ack,
4130 [CPL_FW6_MSG] = deferred_fw6_msg,
4131 [CPL_RX_PKT] = rx_pkt,
4132 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4133 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4136 static void process_timeout(struct c4iw_ep *ep)
4138 struct c4iw_qp_attributes attrs;
4141 mutex_lock(&ep->com.mutex);
4142 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4144 set_bit(TIMEDOUT, &ep->com.history);
4145 switch (ep->com.state) {
4147 connect_reply_upcall(ep, -ETIMEDOUT);
4156 if (ep->com.cm_id && ep->com.qp) {
4157 attrs.next_state = C4IW_QP_STATE_ERROR;
4158 c4iw_modify_qp(ep->com.qp->rhp,
4159 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4162 close_complete_upcall(ep, -ETIMEDOUT);
4168 * These states are expected if the ep timed out at the same
4169 * time as another thread was calling stop_ep_timer().
4170 * So we silently do nothing for these states.
4175 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4176 __func__, ep, ep->hwtid, ep->com.state);
4179 mutex_unlock(&ep->com.mutex);
4181 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4182 c4iw_put_ep(&ep->com);
4185 static void process_timedout_eps(void)
4189 spin_lock_irq(&timeout_lock);
4190 while (!list_empty(&timeout_list)) {
4191 struct list_head *tmp;
4193 tmp = timeout_list.next;
4197 spin_unlock_irq(&timeout_lock);
4198 ep = list_entry(tmp, struct c4iw_ep, entry);
4199 process_timeout(ep);
4200 spin_lock_irq(&timeout_lock);
4202 spin_unlock_irq(&timeout_lock);
4205 static void process_work(struct work_struct *work)
4207 struct sk_buff *skb = NULL;
4208 struct c4iw_dev *dev;
4209 struct cpl_act_establish *rpl;
4210 unsigned int opcode;
4213 process_timedout_eps();
4214 while ((skb = skb_dequeue(&rxq))) {
4216 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4217 opcode = rpl->ot.opcode;
4219 BUG_ON(!work_handlers[opcode]);
4220 ret = work_handlers[opcode](dev, skb);
4223 process_timedout_eps();
4227 static DECLARE_WORK(skb_work, process_work);
4229 static void ep_timeout(unsigned long arg)
4231 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4234 spin_lock(&timeout_lock);
4235 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4237 * Only insert if it is not already on the list.
4239 if (!ep->entry.next) {
4240 list_add_tail(&ep->entry, &timeout_list);
4244 spin_unlock(&timeout_lock);
4246 queue_work(workq, &skb_work);
4250 * All the CM events are handled on a work queue to have a safe context.
4252 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4256 * Save dev in the skb->cb area.
4258 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4261 * Queue the skb and schedule the worker thread.
4263 skb_queue_tail(&rxq, skb);
4264 queue_work(workq, &skb_work);
4268 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4270 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4272 if (rpl->status != CPL_ERR_NONE) {
4273 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4274 "for tid %u\n", rpl->status, GET_TID(rpl));
4280 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4282 struct cpl_fw6_msg *rpl = cplhdr(skb);
4283 struct c4iw_wr_wait *wr_waitp;
4286 PDBG("%s type %u\n", __func__, rpl->type);
4288 switch (rpl->type) {
4289 case FW6_TYPE_WR_RPL:
4290 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4291 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4292 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4294 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4298 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4302 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4310 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4312 struct cpl_abort_req_rss *req = cplhdr(skb);
4314 unsigned int tid = GET_TID(req);
4316 ep = get_ep_from_tid(dev, tid);
4317 /* This EP will be dereferenced in peer_abort() */
4319 printk(KERN_WARNING MOD
4320 "Abort on non-existent endpoint, tid %d\n", tid);
4324 if (is_neg_adv(req->status)) {
4325 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4326 __func__, ep->hwtid, req->status,
4327 neg_adv_str(req->status));
4330 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4333 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4340 * Most upcalls from the T4 Core go to sched() to
4341 * schedule the processing on a work queue.
4343 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4344 [CPL_ACT_ESTABLISH] = sched,
4345 [CPL_ACT_OPEN_RPL] = sched,
4346 [CPL_RX_DATA] = sched,
4347 [CPL_ABORT_RPL_RSS] = sched,
4348 [CPL_ABORT_RPL] = sched,
4349 [CPL_PASS_OPEN_RPL] = sched,
4350 [CPL_CLOSE_LISTSRV_RPL] = sched,
4351 [CPL_PASS_ACCEPT_REQ] = sched,
4352 [CPL_PASS_ESTABLISH] = sched,
4353 [CPL_PEER_CLOSE] = sched,
4354 [CPL_CLOSE_CON_RPL] = sched,
4355 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4356 [CPL_RDMA_TERMINATE] = sched,
4357 [CPL_FW4_ACK] = sched,
4358 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4359 [CPL_FW6_MSG] = fw6_msg,
4360 [CPL_RX_PKT] = sched
4363 int __init c4iw_cm_init(void)
4365 spin_lock_init(&timeout_lock);
4366 skb_queue_head_init(&rxq);
4368 workq = create_singlethread_workqueue("iw_cxgb4");
4375 void c4iw_cm_term(void)
4377 WARN_ON(!list_empty(&timeout_list));
4378 flush_workqueue(workq);
4379 destroy_workqueue(workq);