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>
52 #include <libcxgb_cm.h>
56 static char *states[] = {
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=0)");
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 "Per-connection max ORD/IRD (default=32)");
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 "Enable tcp window scaling (default=1)");
102 static int peer2peer = 1;
103 module_param(peer2peer, int, 0644);
104 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
106 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
107 module_param(p2p_type, int, 0644);
108 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
109 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
111 static int ep_timeout_secs = 60;
112 module_param(ep_timeout_secs, int, 0644);
113 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
114 "in seconds (default=60)");
116 static int mpa_rev = 2;
117 module_param(mpa_rev, int, 0644);
118 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
119 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
120 " compliant (default=2)");
122 static int markers_enabled;
123 module_param(markers_enabled, int, 0644);
124 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
126 static int crc_enabled = 1;
127 module_param(crc_enabled, int, 0644);
128 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
130 static int rcv_win = 256 * 1024;
131 module_param(rcv_win, int, 0644);
132 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
134 static int snd_win = 128 * 1024;
135 module_param(snd_win, int, 0644);
136 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
138 static struct workqueue_struct *workq;
140 static struct sk_buff_head rxq;
142 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
143 static void ep_timeout(struct timer_list *t);
144 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
147 static LIST_HEAD(timeout_list);
148 static DEFINE_SPINLOCK(timeout_lock);
150 static void deref_cm_id(struct c4iw_ep_common *epc)
152 epc->cm_id->rem_ref(epc->cm_id);
154 set_bit(CM_ID_DEREFED, &epc->history);
157 static void ref_cm_id(struct c4iw_ep_common *epc)
159 set_bit(CM_ID_REFED, &epc->history);
160 epc->cm_id->add_ref(epc->cm_id);
163 static void deref_qp(struct c4iw_ep *ep)
165 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
166 clear_bit(QP_REFERENCED, &ep->com.flags);
167 set_bit(QP_DEREFED, &ep->com.history);
170 static void ref_qp(struct c4iw_ep *ep)
172 set_bit(QP_REFERENCED, &ep->com.flags);
173 set_bit(QP_REFED, &ep->com.history);
174 c4iw_qp_add_ref(&ep->com.qp->ibqp);
177 static void start_ep_timer(struct c4iw_ep *ep)
179 pr_debug("ep %p\n", ep);
180 if (timer_pending(&ep->timer)) {
181 pr_err("%s timer already started! ep %p\n",
185 clear_bit(TIMEOUT, &ep->com.flags);
186 c4iw_get_ep(&ep->com);
187 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
188 add_timer(&ep->timer);
191 static int stop_ep_timer(struct c4iw_ep *ep)
193 pr_debug("ep %p stopping\n", ep);
194 del_timer_sync(&ep->timer);
195 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
196 c4iw_put_ep(&ep->com);
202 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
203 struct l2t_entry *l2e)
207 if (c4iw_fatal_error(rdev)) {
209 pr_err("%s - device in error state - dropping\n", __func__);
212 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
215 else if (error == NET_XMIT_DROP)
217 return error < 0 ? error : 0;
220 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
224 if (c4iw_fatal_error(rdev)) {
226 pr_err("%s - device in error state - dropping\n", __func__);
229 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
232 return error < 0 ? error : 0;
235 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
237 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
239 skb = get_skb(skb, len, GFP_KERNEL);
243 cxgb_mk_tid_release(skb, len, hwtid, 0);
244 c4iw_ofld_send(rdev, skb);
248 static void set_emss(struct c4iw_ep *ep, u16 opt)
250 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
251 ((AF_INET == ep->com.remote_addr.ss_family) ?
252 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
253 sizeof(struct tcphdr);
255 if (TCPOPT_TSTAMP_G(opt))
256 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
260 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
261 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
262 pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss,
266 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
268 enum c4iw_ep_state state;
270 mutex_lock(&epc->mutex);
272 mutex_unlock(&epc->mutex);
276 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
281 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
283 mutex_lock(&epc->mutex);
284 pr_debug("%s -> %s\n", states[epc->state], states[new]);
285 __state_set(epc, new);
286 mutex_unlock(&epc->mutex);
290 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
296 len = roundup(sizeof(union cpl_wr_size), 16);
297 for (i = 0; i < size; i++) {
298 skb = alloc_skb(len, GFP_KERNEL);
301 skb_queue_tail(ep_skb_list, skb);
305 skb_queue_purge(ep_skb_list);
309 static void *alloc_ep(int size, gfp_t gfp)
311 struct c4iw_ep_common *epc;
313 epc = kzalloc(size, gfp);
315 epc->wr_waitp = c4iw_alloc_wr_wait(gfp);
316 if (!epc->wr_waitp) {
321 kref_init(&epc->kref);
322 mutex_init(&epc->mutex);
323 c4iw_init_wr_wait(epc->wr_waitp);
325 pr_debug("alloc ep %p\n", epc);
330 static void remove_ep_tid(struct c4iw_ep *ep)
334 xa_lock_irqsave(&ep->com.dev->hwtids, flags);
335 __xa_erase(&ep->com.dev->hwtids, ep->hwtid);
336 if (xa_empty(&ep->com.dev->hwtids))
337 wake_up(&ep->com.dev->wait);
338 xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
341 static int insert_ep_tid(struct c4iw_ep *ep)
346 xa_lock_irqsave(&ep->com.dev->hwtids, flags);
347 err = __xa_insert(&ep->com.dev->hwtids, ep->hwtid, ep, GFP_KERNEL);
348 xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
354 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
356 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
361 xa_lock_irqsave(&dev->hwtids, flags);
362 ep = xa_load(&dev->hwtids, tid);
364 c4iw_get_ep(&ep->com);
365 xa_unlock_irqrestore(&dev->hwtids, flags);
370 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
372 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
375 struct c4iw_listen_ep *ep;
378 xa_lock_irqsave(&dev->stids, flags);
379 ep = xa_load(&dev->stids, stid);
381 c4iw_get_ep(&ep->com);
382 xa_unlock_irqrestore(&dev->stids, flags);
386 void _c4iw_free_ep(struct kref *kref)
390 ep = container_of(kref, struct c4iw_ep, com.kref);
391 pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
392 if (test_bit(QP_REFERENCED, &ep->com.flags))
394 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
395 if (ep->com.remote_addr.ss_family == AF_INET6) {
396 struct sockaddr_in6 *sin6 =
397 (struct sockaddr_in6 *)
401 ep->com.dev->rdev.lldi.ports[0],
402 (const u32 *)&sin6->sin6_addr.s6_addr,
405 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
406 ep->com.local_addr.ss_family);
407 dst_release(ep->dst);
408 cxgb4_l2t_release(ep->l2t);
409 kfree_skb(ep->mpa_skb);
411 if (!skb_queue_empty(&ep->com.ep_skb_list))
412 skb_queue_purge(&ep->com.ep_skb_list);
413 c4iw_put_wr_wait(ep->com.wr_waitp);
417 static void release_ep_resources(struct c4iw_ep *ep)
419 set_bit(RELEASE_RESOURCES, &ep->com.flags);
422 * If we have a hwtid, then remove it from the idr table
423 * so lookups will no longer find this endpoint. Otherwise
424 * we have a race where one thread finds the ep ptr just
425 * before the other thread is freeing the ep memory.
429 c4iw_put_ep(&ep->com);
432 static int status2errno(int status)
437 case CPL_ERR_CONN_RESET:
439 case CPL_ERR_ARP_MISS:
440 return -EHOSTUNREACH;
441 case CPL_ERR_CONN_TIMEDOUT:
443 case CPL_ERR_TCAM_FULL:
445 case CPL_ERR_CONN_EXIST:
453 * Try and reuse skbs already allocated...
455 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
457 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
460 skb_reset_transport_header(skb);
462 skb = alloc_skb(len, gfp);
466 t4_set_arp_err_handler(skb, NULL, NULL);
470 static struct net_device *get_real_dev(struct net_device *egress_dev)
472 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
475 static void arp_failure_discard(void *handle, struct sk_buff *skb)
477 pr_err("ARP failure\n");
481 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
483 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
488 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
489 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
492 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
496 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 release_ep_resources(ep);
501 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
505 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
506 c4iw_put_ep(&ep->parent_ep->com);
507 release_ep_resources(ep);
512 * Fake up a special CPL opcode and call sched() so process_work() will call
513 * _put_ep_safe() in a safe context to free the ep resources. This is needed
514 * because ARP error handlers are called in an ATOMIC context, and
515 * _c4iw_free_ep() needs to block.
517 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
520 struct cpl_act_establish *rpl = cplhdr(skb);
522 /* Set our special ARP_FAILURE opcode */
523 rpl->ot.opcode = cpl;
526 * Save ep in the skb->cb area, after where sched() will save the dev
529 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
530 sched(ep->com.dev, skb);
533 /* Handle an ARP failure for an accept */
534 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
536 struct c4iw_ep *ep = handle;
538 pr_err("ARP failure during accept - tid %u - dropping connection\n",
541 __state_set(&ep->com, DEAD);
542 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
546 * Handle an ARP failure for an active open.
548 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
550 struct c4iw_ep *ep = handle;
552 pr_err("ARP failure during connect\n");
553 connect_reply_upcall(ep, -EHOSTUNREACH);
554 __state_set(&ep->com, DEAD);
555 if (ep->com.remote_addr.ss_family == AF_INET6) {
556 struct sockaddr_in6 *sin6 =
557 (struct sockaddr_in6 *)&ep->com.local_addr;
558 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
559 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
561 xa_erase_irq(&ep->com.dev->atids, ep->atid);
562 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
563 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
567 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
570 static void abort_arp_failure(void *handle, struct sk_buff *skb)
573 struct c4iw_ep *ep = handle;
574 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
575 struct cpl_abort_req *req = cplhdr(skb);
577 pr_debug("rdev %p\n", rdev);
578 req->cmd = CPL_ABORT_NO_RST;
580 ret = c4iw_ofld_send(rdev, skb);
582 __state_set(&ep->com, DEAD);
583 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
588 static int send_flowc(struct c4iw_ep *ep)
590 struct fw_flowc_wr *flowc;
591 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
592 u16 vlan = ep->l2t->vlan;
594 int flowclen, flowclen16;
599 if (vlan == CPL_L2T_VLAN_NONE)
604 flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
605 flowclen16 = DIV_ROUND_UP(flowclen, 16);
606 flowclen = flowclen16 * 16;
608 flowc = __skb_put(skb, flowclen);
609 memset(flowc, 0, flowclen);
611 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
612 FW_FLOWC_WR_NPARAMS_V(nparams));
613 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
614 FW_WR_FLOWID_V(ep->hwtid));
616 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
617 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
618 (ep->com.dev->rdev.lldi.pf));
619 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
620 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
621 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
622 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
623 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
624 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
625 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
626 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
627 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
628 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
629 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
630 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
631 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
632 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
633 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
634 flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
637 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
638 flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
639 flowc->mnemval[9].val = cpu_to_be32(pri);
642 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
643 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
646 static int send_halfclose(struct c4iw_ep *ep)
648 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
651 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
655 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
656 NULL, arp_failure_discard);
658 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
661 static void read_tcb(struct c4iw_ep *ep)
664 struct cpl_get_tcb *req;
665 int wrlen = roundup(sizeof(*req), 16);
667 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
671 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
672 req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
673 memset(req, 0, wrlen);
674 INIT_TP_WR(req, ep->hwtid);
675 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
676 req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
679 * keep a ref on the ep so the tcb is not unlocked before this
680 * cpl completes. The ref is released in read_tcb_rpl().
682 c4iw_get_ep(&ep->com);
683 if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
684 c4iw_put_ep(&ep->com);
687 static int send_abort_req(struct c4iw_ep *ep)
689 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
690 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
692 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
693 if (WARN_ON(!req_skb))
696 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
697 ep, abort_arp_failure);
699 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
702 static int send_abort(struct c4iw_ep *ep)
704 if (!ep->com.qp || !ep->com.qp->srq) {
708 set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
713 static int send_connect(struct c4iw_ep *ep)
715 struct cpl_act_open_req *req = NULL;
716 struct cpl_t5_act_open_req *t5req = NULL;
717 struct cpl_t6_act_open_req *t6req = NULL;
718 struct cpl_act_open_req6 *req6 = NULL;
719 struct cpl_t5_act_open_req6 *t5req6 = NULL;
720 struct cpl_t6_act_open_req6 *t6req6 = NULL;
724 unsigned int mtu_idx;
726 int win, sizev4, sizev6, wrlen;
727 struct sockaddr_in *la = (struct sockaddr_in *)
729 struct sockaddr_in *ra = (struct sockaddr_in *)
730 &ep->com.remote_addr;
731 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
733 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
734 &ep->com.remote_addr;
736 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
737 u32 isn = (get_random_u32() & ~7UL) - 1;
738 struct net_device *netdev;
741 netdev = ep->com.dev->rdev.lldi.ports[0];
743 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
745 sizev4 = sizeof(struct cpl_act_open_req);
746 sizev6 = sizeof(struct cpl_act_open_req6);
749 sizev4 = sizeof(struct cpl_t5_act_open_req);
750 sizev6 = sizeof(struct cpl_t5_act_open_req6);
753 sizev4 = sizeof(struct cpl_t6_act_open_req);
754 sizev6 = sizeof(struct cpl_t6_act_open_req6);
757 pr_err("T%d Chip is not supported\n",
758 CHELSIO_CHIP_VERSION(adapter_type));
762 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
763 roundup(sizev4, 16) :
766 pr_debug("ep %p atid %u\n", ep, ep->atid);
768 skb = get_skb(NULL, wrlen, GFP_KERNEL);
770 pr_err("%s - failed to alloc skb\n", __func__);
773 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
775 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
776 enable_tcp_timestamps,
777 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
778 wscale = cxgb_compute_wscale(rcv_win);
781 * Specify the largest window that will fit in opt0. The
782 * remainder will be specified in the rx_data_ack.
784 win = ep->rcv_win >> 10;
785 if (win > RCV_BUFSIZ_M)
788 opt0 = (nocong ? NO_CONG_F : 0) |
791 WND_SCALE_V(wscale) |
793 L2T_IDX_V(ep->l2t->idx) |
794 TX_CHAN_V(ep->tx_chan) |
795 SMAC_SEL_V(ep->smac_idx) |
796 DSCP_V(ep->tos >> 2) |
797 ULP_MODE_V(ULP_MODE_TCPDDP) |
799 opt2 = RX_CHANNEL_V(0) |
800 CCTRL_ECN_V(enable_ecn) |
801 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
802 if (enable_tcp_timestamps)
803 opt2 |= TSTAMPS_EN_F;
806 if (wscale && enable_tcp_window_scaling)
807 opt2 |= WND_SCALE_EN_F;
808 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
812 opt2 |= T5_OPT_2_VALID_F;
813 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
817 params = cxgb4_select_ntuple(netdev, ep->l2t);
819 if (ep->com.remote_addr.ss_family == AF_INET6)
820 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
821 (const u32 *)&la6->sin6_addr.s6_addr, 1);
823 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
825 if (ep->com.remote_addr.ss_family == AF_INET) {
826 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
828 req = skb_put(skb, wrlen);
832 t5req = skb_put(skb, wrlen);
833 INIT_TP_WR(t5req, 0);
834 req = (struct cpl_act_open_req *)t5req;
837 t6req = skb_put(skb, wrlen);
838 INIT_TP_WR(t6req, 0);
839 req = (struct cpl_act_open_req *)t6req;
840 t5req = (struct cpl_t5_act_open_req *)t6req;
843 pr_err("T%d Chip is not supported\n",
844 CHELSIO_CHIP_VERSION(adapter_type));
849 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
850 ((ep->rss_qid<<14) | ep->atid)));
851 req->local_port = la->sin_port;
852 req->peer_port = ra->sin_port;
853 req->local_ip = la->sin_addr.s_addr;
854 req->peer_ip = ra->sin_addr.s_addr;
855 req->opt0 = cpu_to_be64(opt0);
857 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
858 req->params = cpu_to_be32(params);
859 req->opt2 = cpu_to_be32(opt2);
861 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
863 cpu_to_be64(FILTER_TUPLE_V(params));
864 t5req->rsvd = cpu_to_be32(isn);
865 pr_debug("snd_isn %u\n", t5req->rsvd);
866 t5req->opt2 = cpu_to_be32(opt2);
869 cpu_to_be64(FILTER_TUPLE_V(params));
870 t6req->rsvd = cpu_to_be32(isn);
871 pr_debug("snd_isn %u\n", t6req->rsvd);
872 t6req->opt2 = cpu_to_be32(opt2);
876 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
878 req6 = skb_put(skb, wrlen);
882 t5req6 = skb_put(skb, wrlen);
883 INIT_TP_WR(t5req6, 0);
884 req6 = (struct cpl_act_open_req6 *)t5req6;
887 t6req6 = skb_put(skb, wrlen);
888 INIT_TP_WR(t6req6, 0);
889 req6 = (struct cpl_act_open_req6 *)t6req6;
890 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
893 pr_err("T%d Chip is not supported\n",
894 CHELSIO_CHIP_VERSION(adapter_type));
899 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
900 ((ep->rss_qid<<14)|ep->atid)));
901 req6->local_port = la6->sin6_port;
902 req6->peer_port = ra6->sin6_port;
903 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
904 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
905 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
906 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
907 req6->opt0 = cpu_to_be64(opt0);
909 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
910 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
912 req6->opt2 = cpu_to_be32(opt2);
914 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
916 cpu_to_be64(FILTER_TUPLE_V(params));
917 t5req6->rsvd = cpu_to_be32(isn);
918 pr_debug("snd_isn %u\n", t5req6->rsvd);
919 t5req6->opt2 = cpu_to_be32(opt2);
922 cpu_to_be64(FILTER_TUPLE_V(params));
923 t6req6->rsvd = cpu_to_be32(isn);
924 pr_debug("snd_isn %u\n", t6req6->rsvd);
925 t6req6->opt2 = cpu_to_be32(opt2);
931 set_bit(ACT_OPEN_REQ, &ep->com.history);
932 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
934 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
935 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
936 (const u32 *)&la6->sin6_addr.s6_addr, 1);
940 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
943 int mpalen, wrlen, ret;
944 struct fw_ofld_tx_data_wr *req;
945 struct mpa_message *mpa;
946 struct mpa_v2_conn_params mpa_v2_params;
948 pr_debug("ep %p tid %u pd_len %d\n",
949 ep, ep->hwtid, ep->plen);
951 mpalen = sizeof(*mpa) + ep->plen;
952 if (mpa_rev_to_use == 2)
953 mpalen += sizeof(struct mpa_v2_conn_params);
954 wrlen = roundup(mpalen + sizeof(*req), 16);
955 skb = get_skb(skb, wrlen, GFP_KERNEL);
957 connect_reply_upcall(ep, -ENOMEM);
960 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
962 req = skb_put_zero(skb, wrlen);
963 req->op_to_immdlen = cpu_to_be32(
964 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
966 FW_WR_IMMDLEN_V(mpalen));
967 req->flowid_len16 = cpu_to_be32(
968 FW_WR_FLOWID_V(ep->hwtid) |
969 FW_WR_LEN16_V(wrlen >> 4));
970 req->plen = cpu_to_be32(mpalen);
971 req->tunnel_to_proxy = cpu_to_be32(
972 FW_OFLD_TX_DATA_WR_FLUSH_F |
973 FW_OFLD_TX_DATA_WR_SHOVE_F);
975 mpa = (struct mpa_message *)(req + 1);
976 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
980 mpa->flags |= MPA_CRC;
981 if (markers_enabled) {
982 mpa->flags |= MPA_MARKERS;
983 ep->mpa_attr.recv_marker_enabled = 1;
985 ep->mpa_attr.recv_marker_enabled = 0;
987 if (mpa_rev_to_use == 2)
988 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
990 mpa->private_data_size = htons(ep->plen);
991 mpa->revision = mpa_rev_to_use;
992 if (mpa_rev_to_use == 1) {
993 ep->tried_with_mpa_v1 = 1;
994 ep->retry_with_mpa_v1 = 0;
997 if (mpa_rev_to_use == 2) {
998 mpa->private_data_size =
999 htons(ntohs(mpa->private_data_size) +
1000 sizeof(struct mpa_v2_conn_params));
1001 pr_debug("initiator ird %u ord %u\n", ep->ird,
1003 mpa_v2_params.ird = htons((u16)ep->ird);
1004 mpa_v2_params.ord = htons((u16)ep->ord);
1007 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1009 mpa_v2_params.ord |=
1010 htons(MPA_V2_RDMA_WRITE_RTR);
1011 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1012 mpa_v2_params.ord |=
1013 htons(MPA_V2_RDMA_READ_RTR);
1015 memcpy(mpa->private_data, &mpa_v2_params,
1016 sizeof(struct mpa_v2_conn_params));
1019 memcpy(mpa->private_data +
1020 sizeof(struct mpa_v2_conn_params),
1021 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1024 memcpy(mpa->private_data,
1025 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1028 * Reference the mpa skb. This ensures the data area
1029 * will remain in memory until the hw acks the tx.
1030 * Function fw4_ack() will deref it.
1033 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1035 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1039 __state_set(&ep->com, MPA_REQ_SENT);
1040 ep->mpa_attr.initiator = 1;
1041 ep->snd_seq += mpalen;
1045 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1048 struct fw_ofld_tx_data_wr *req;
1049 struct mpa_message *mpa;
1050 struct sk_buff *skb;
1051 struct mpa_v2_conn_params mpa_v2_params;
1053 pr_debug("ep %p tid %u pd_len %d\n",
1054 ep, ep->hwtid, ep->plen);
1056 mpalen = sizeof(*mpa) + plen;
1057 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1058 mpalen += sizeof(struct mpa_v2_conn_params);
1059 wrlen = roundup(mpalen + sizeof(*req), 16);
1061 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1063 pr_err("%s - cannot alloc skb!\n", __func__);
1066 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1068 req = skb_put_zero(skb, wrlen);
1069 req->op_to_immdlen = cpu_to_be32(
1070 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1072 FW_WR_IMMDLEN_V(mpalen));
1073 req->flowid_len16 = cpu_to_be32(
1074 FW_WR_FLOWID_V(ep->hwtid) |
1075 FW_WR_LEN16_V(wrlen >> 4));
1076 req->plen = cpu_to_be32(mpalen);
1077 req->tunnel_to_proxy = cpu_to_be32(
1078 FW_OFLD_TX_DATA_WR_FLUSH_F |
1079 FW_OFLD_TX_DATA_WR_SHOVE_F);
1081 mpa = (struct mpa_message *)(req + 1);
1082 memset(mpa, 0, sizeof(*mpa));
1083 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1084 mpa->flags = MPA_REJECT;
1085 mpa->revision = ep->mpa_attr.version;
1086 mpa->private_data_size = htons(plen);
1088 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1089 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1090 mpa->private_data_size =
1091 htons(ntohs(mpa->private_data_size) +
1092 sizeof(struct mpa_v2_conn_params));
1093 mpa_v2_params.ird = htons(((u16)ep->ird) |
1094 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1096 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1098 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1099 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1100 FW_RI_INIT_P2PTYPE_READ_REQ ?
1101 MPA_V2_RDMA_READ_RTR : 0) : 0));
1102 memcpy(mpa->private_data, &mpa_v2_params,
1103 sizeof(struct mpa_v2_conn_params));
1106 memcpy(mpa->private_data +
1107 sizeof(struct mpa_v2_conn_params), pdata, plen);
1110 memcpy(mpa->private_data, pdata, plen);
1113 * Reference the mpa skb again. This ensures the data area
1114 * will remain in memory until the hw acks the tx.
1115 * Function fw4_ack() will deref it.
1118 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1121 ep->snd_seq += mpalen;
1122 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1125 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1128 struct fw_ofld_tx_data_wr *req;
1129 struct mpa_message *mpa;
1130 struct sk_buff *skb;
1131 struct mpa_v2_conn_params mpa_v2_params;
1133 pr_debug("ep %p tid %u pd_len %d\n",
1134 ep, ep->hwtid, ep->plen);
1136 mpalen = sizeof(*mpa) + plen;
1137 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1138 mpalen += sizeof(struct mpa_v2_conn_params);
1139 wrlen = roundup(mpalen + sizeof(*req), 16);
1141 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1143 pr_err("%s - cannot alloc skb!\n", __func__);
1146 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1148 req = skb_put_zero(skb, wrlen);
1149 req->op_to_immdlen = cpu_to_be32(
1150 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1152 FW_WR_IMMDLEN_V(mpalen));
1153 req->flowid_len16 = cpu_to_be32(
1154 FW_WR_FLOWID_V(ep->hwtid) |
1155 FW_WR_LEN16_V(wrlen >> 4));
1156 req->plen = cpu_to_be32(mpalen);
1157 req->tunnel_to_proxy = cpu_to_be32(
1158 FW_OFLD_TX_DATA_WR_FLUSH_F |
1159 FW_OFLD_TX_DATA_WR_SHOVE_F);
1161 mpa = (struct mpa_message *)(req + 1);
1162 memset(mpa, 0, sizeof(*mpa));
1163 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1165 if (ep->mpa_attr.crc_enabled)
1166 mpa->flags |= MPA_CRC;
1167 if (ep->mpa_attr.recv_marker_enabled)
1168 mpa->flags |= MPA_MARKERS;
1169 mpa->revision = ep->mpa_attr.version;
1170 mpa->private_data_size = htons(plen);
1172 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1173 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1174 mpa->private_data_size =
1175 htons(ntohs(mpa->private_data_size) +
1176 sizeof(struct mpa_v2_conn_params));
1177 mpa_v2_params.ird = htons((u16)ep->ird);
1178 mpa_v2_params.ord = htons((u16)ep->ord);
1179 if (peer2peer && (ep->mpa_attr.p2p_type !=
1180 FW_RI_INIT_P2PTYPE_DISABLED)) {
1181 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1183 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1184 mpa_v2_params.ord |=
1185 htons(MPA_V2_RDMA_WRITE_RTR);
1186 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1187 mpa_v2_params.ord |=
1188 htons(MPA_V2_RDMA_READ_RTR);
1191 memcpy(mpa->private_data, &mpa_v2_params,
1192 sizeof(struct mpa_v2_conn_params));
1195 memcpy(mpa->private_data +
1196 sizeof(struct mpa_v2_conn_params), pdata, plen);
1199 memcpy(mpa->private_data, pdata, plen);
1202 * Reference the mpa skb. This ensures the data area
1203 * will remain in memory until the hw acks the tx.
1204 * Function fw4_ack() will deref it.
1207 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1209 __state_set(&ep->com, MPA_REP_SENT);
1210 ep->snd_seq += mpalen;
1211 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1214 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1217 struct cpl_act_establish *req = cplhdr(skb);
1218 unsigned short tcp_opt = ntohs(req->tcp_opt);
1219 unsigned int tid = GET_TID(req);
1220 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1221 struct tid_info *t = dev->rdev.lldi.tids;
1224 ep = lookup_atid(t, atid);
1228 pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1229 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1231 mutex_lock(&ep->com.mutex);
1232 dst_confirm(ep->dst);
1234 /* setup the hwtid for this connection */
1236 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1239 ep->snd_seq = be32_to_cpu(req->snd_isn);
1240 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1241 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1243 set_emss(ep, tcp_opt);
1245 /* dealloc the atid */
1246 xa_erase_irq(&ep->com.dev->atids, atid);
1247 cxgb4_free_atid(t, atid);
1248 set_bit(ACT_ESTAB, &ep->com.history);
1250 /* start MPA negotiation */
1251 ret = send_flowc(ep);
1254 if (ep->retry_with_mpa_v1)
1255 ret = send_mpa_req(ep, skb, 1);
1257 ret = send_mpa_req(ep, skb, mpa_rev);
1260 mutex_unlock(&ep->com.mutex);
1263 mutex_unlock(&ep->com.mutex);
1264 connect_reply_upcall(ep, -ENOMEM);
1265 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1269 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1271 struct iw_cm_event event;
1273 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1274 memset(&event, 0, sizeof(event));
1275 event.event = IW_CM_EVENT_CLOSE;
1276 event.status = status;
1277 if (ep->com.cm_id) {
1278 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1279 ep, ep->com.cm_id, ep->hwtid);
1280 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1281 deref_cm_id(&ep->com);
1282 set_bit(CLOSE_UPCALL, &ep->com.history);
1286 static void peer_close_upcall(struct c4iw_ep *ep)
1288 struct iw_cm_event event;
1290 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1291 memset(&event, 0, sizeof(event));
1292 event.event = IW_CM_EVENT_DISCONNECT;
1293 if (ep->com.cm_id) {
1294 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1295 ep, ep->com.cm_id, ep->hwtid);
1296 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1297 set_bit(DISCONN_UPCALL, &ep->com.history);
1301 static void peer_abort_upcall(struct c4iw_ep *ep)
1303 struct iw_cm_event event;
1305 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1306 memset(&event, 0, sizeof(event));
1307 event.event = IW_CM_EVENT_CLOSE;
1308 event.status = -ECONNRESET;
1309 if (ep->com.cm_id) {
1310 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1311 ep->com.cm_id, ep->hwtid);
1312 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1313 deref_cm_id(&ep->com);
1314 set_bit(ABORT_UPCALL, &ep->com.history);
1318 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1320 struct iw_cm_event event;
1322 pr_debug("ep %p tid %u status %d\n",
1323 ep, ep->hwtid, status);
1324 memset(&event, 0, sizeof(event));
1325 event.event = IW_CM_EVENT_CONNECT_REPLY;
1326 event.status = status;
1327 memcpy(&event.local_addr, &ep->com.local_addr,
1328 sizeof(ep->com.local_addr));
1329 memcpy(&event.remote_addr, &ep->com.remote_addr,
1330 sizeof(ep->com.remote_addr));
1332 if ((status == 0) || (status == -ECONNREFUSED)) {
1333 if (!ep->tried_with_mpa_v1) {
1334 /* this means MPA_v2 is used */
1335 event.ord = ep->ird;
1336 event.ird = ep->ord;
1337 event.private_data_len = ep->plen -
1338 sizeof(struct mpa_v2_conn_params);
1339 event.private_data = ep->mpa_pkt +
1340 sizeof(struct mpa_message) +
1341 sizeof(struct mpa_v2_conn_params);
1343 /* this means MPA_v1 is used */
1344 event.ord = cur_max_read_depth(ep->com.dev);
1345 event.ird = cur_max_read_depth(ep->com.dev);
1346 event.private_data_len = ep->plen;
1347 event.private_data = ep->mpa_pkt +
1348 sizeof(struct mpa_message);
1352 pr_debug("ep %p tid %u status %d\n", ep,
1354 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1355 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1358 deref_cm_id(&ep->com);
1361 static int connect_request_upcall(struct c4iw_ep *ep)
1363 struct iw_cm_event event;
1366 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1367 memset(&event, 0, sizeof(event));
1368 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1369 memcpy(&event.local_addr, &ep->com.local_addr,
1370 sizeof(ep->com.local_addr));
1371 memcpy(&event.remote_addr, &ep->com.remote_addr,
1372 sizeof(ep->com.remote_addr));
1373 event.provider_data = ep;
1374 if (!ep->tried_with_mpa_v1) {
1375 /* this means MPA_v2 is used */
1376 event.ord = ep->ord;
1377 event.ird = ep->ird;
1378 event.private_data_len = ep->plen -
1379 sizeof(struct mpa_v2_conn_params);
1380 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1381 sizeof(struct mpa_v2_conn_params);
1383 /* this means MPA_v1 is used. Send max supported */
1384 event.ord = cur_max_read_depth(ep->com.dev);
1385 event.ird = cur_max_read_depth(ep->com.dev);
1386 event.private_data_len = ep->plen;
1387 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1389 c4iw_get_ep(&ep->com);
1390 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1393 c4iw_put_ep(&ep->com);
1394 set_bit(CONNREQ_UPCALL, &ep->com.history);
1395 c4iw_put_ep(&ep->parent_ep->com);
1399 static void established_upcall(struct c4iw_ep *ep)
1401 struct iw_cm_event event;
1403 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1404 memset(&event, 0, sizeof(event));
1405 event.event = IW_CM_EVENT_ESTABLISHED;
1406 event.ird = ep->ord;
1407 event.ord = ep->ird;
1408 if (ep->com.cm_id) {
1409 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1410 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1411 set_bit(ESTAB_UPCALL, &ep->com.history);
1415 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1417 struct sk_buff *skb;
1418 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1421 pr_debug("ep %p tid %u credits %u\n",
1422 ep, ep->hwtid, credits);
1423 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1425 pr_err("update_rx_credits - cannot alloc skb!\n");
1430 * If we couldn't specify the entire rcv window at connection setup
1431 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1432 * then add the overage in to the credits returned.
1434 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1435 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1437 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1438 RX_DACK_MODE_V(dack_mode);
1440 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1443 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1447 #define RELAXED_IRD_NEGOTIATION 1
1450 * process_mpa_reply - process streaming mode MPA reply
1454 * 0 upon success indicating a connect request was delivered to the ULP
1455 * or the mpa request is incomplete but valid so far.
1457 * 1 if a failure requires the caller to close the connection.
1459 * 2 if a failure requires the caller to abort the connection.
1461 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1463 struct mpa_message *mpa;
1464 struct mpa_v2_conn_params *mpa_v2_params;
1466 u16 resp_ird, resp_ord;
1467 u8 rtr_mismatch = 0, insuff_ird = 0;
1468 struct c4iw_qp_attributes attrs;
1469 enum c4iw_qp_attr_mask mask;
1473 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1476 * If we get more than the supported amount of private data
1477 * then we must fail this connection.
1479 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1481 goto err_stop_timer;
1485 * copy the new data into our accumulation buffer.
1487 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1489 ep->mpa_pkt_len += skb->len;
1492 * if we don't even have the mpa message, then bail.
1494 if (ep->mpa_pkt_len < sizeof(*mpa))
1496 mpa = (struct mpa_message *) ep->mpa_pkt;
1498 /* Validate MPA header. */
1499 if (mpa->revision > mpa_rev) {
1500 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1501 __func__, mpa_rev, mpa->revision);
1503 goto err_stop_timer;
1505 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1507 goto err_stop_timer;
1510 plen = ntohs(mpa->private_data_size);
1513 * Fail if there's too much private data.
1515 if (plen > MPA_MAX_PRIVATE_DATA) {
1517 goto err_stop_timer;
1521 * If plen does not account for pkt size
1523 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1525 goto err_stop_timer;
1528 ep->plen = (u8) plen;
1531 * If we don't have all the pdata yet, then bail.
1532 * We'll continue process when more data arrives.
1534 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1537 if (mpa->flags & MPA_REJECT) {
1538 err = -ECONNREFUSED;
1539 goto err_stop_timer;
1543 * Stop mpa timer. If it expired, then
1544 * we ignore the MPA reply. process_timeout()
1545 * will abort the connection.
1547 if (stop_ep_timer(ep))
1551 * If we get here we have accumulated the entire mpa
1552 * start reply message including private data. And
1553 * the MPA header is valid.
1555 __state_set(&ep->com, FPDU_MODE);
1556 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1557 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1558 ep->mpa_attr.version = mpa->revision;
1559 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1561 if (mpa->revision == 2) {
1562 ep->mpa_attr.enhanced_rdma_conn =
1563 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1564 if (ep->mpa_attr.enhanced_rdma_conn) {
1565 mpa_v2_params = (struct mpa_v2_conn_params *)
1566 (ep->mpa_pkt + sizeof(*mpa));
1567 resp_ird = ntohs(mpa_v2_params->ird) &
1568 MPA_V2_IRD_ORD_MASK;
1569 resp_ord = ntohs(mpa_v2_params->ord) &
1570 MPA_V2_IRD_ORD_MASK;
1571 pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1572 resp_ird, resp_ord, ep->ird, ep->ord);
1575 * This is a double-check. Ideally, below checks are
1576 * not required since ird/ord stuff has been taken
1577 * care of in c4iw_accept_cr
1579 if (ep->ird < resp_ord) {
1580 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1581 ep->com.dev->rdev.lldi.max_ordird_qp)
1585 } else if (ep->ird > resp_ord) {
1588 if (ep->ord > resp_ird) {
1589 if (RELAXED_IRD_NEGOTIATION)
1600 if (ntohs(mpa_v2_params->ird) &
1601 MPA_V2_PEER2PEER_MODEL) {
1602 if (ntohs(mpa_v2_params->ord) &
1603 MPA_V2_RDMA_WRITE_RTR)
1604 ep->mpa_attr.p2p_type =
1605 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1606 else if (ntohs(mpa_v2_params->ord) &
1607 MPA_V2_RDMA_READ_RTR)
1608 ep->mpa_attr.p2p_type =
1609 FW_RI_INIT_P2PTYPE_READ_REQ;
1612 } else if (mpa->revision == 1)
1614 ep->mpa_attr.p2p_type = p2p_type;
1616 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1617 ep->mpa_attr.crc_enabled,
1618 ep->mpa_attr.recv_marker_enabled,
1619 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1620 ep->mpa_attr.p2p_type, p2p_type);
1623 * If responder's RTR does not match with that of initiator, assign
1624 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1625 * generated when moving QP to RTS state.
1626 * A TERM message will be sent after QP has moved to RTS state
1628 if ((ep->mpa_attr.version == 2) && peer2peer &&
1629 (ep->mpa_attr.p2p_type != p2p_type)) {
1630 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1634 attrs.mpa_attr = ep->mpa_attr;
1635 attrs.max_ird = ep->ird;
1636 attrs.max_ord = ep->ord;
1637 attrs.llp_stream_handle = ep;
1638 attrs.next_state = C4IW_QP_STATE_RTS;
1640 mask = C4IW_QP_ATTR_NEXT_STATE |
1641 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1642 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1644 /* bind QP and TID with INIT_WR */
1645 err = c4iw_modify_qp(ep->com.qp->rhp,
1646 ep->com.qp, mask, &attrs, 1);
1651 * If responder's RTR requirement did not match with what initiator
1652 * supports, generate TERM message
1655 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1656 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1657 attrs.ecode = MPA_NOMATCH_RTR;
1658 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1659 attrs.send_term = 1;
1660 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1661 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1668 * Generate TERM if initiator IRD is not sufficient for responder
1669 * provided ORD. Currently, we do the same behaviour even when
1670 * responder provided IRD is also not sufficient as regards to
1674 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1675 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1676 attrs.ecode = MPA_INSUFF_IRD;
1677 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1678 attrs.send_term = 1;
1679 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1680 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1691 connect_reply_upcall(ep, err);
1696 * process_mpa_request - process streaming mode MPA request
1700 * 0 upon success indicating a connect request was delivered to the ULP
1701 * or the mpa request is incomplete but valid so far.
1703 * 1 if a failure requires the caller to close the connection.
1705 * 2 if a failure requires the caller to abort the connection.
1707 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1709 struct mpa_message *mpa;
1710 struct mpa_v2_conn_params *mpa_v2_params;
1713 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1716 * If we get more than the supported amount of private data
1717 * then we must fail this connection.
1719 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1720 goto err_stop_timer;
1722 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1725 * Copy the new data into our accumulation buffer.
1727 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1729 ep->mpa_pkt_len += skb->len;
1732 * If we don't even have the mpa message, then bail.
1733 * We'll continue process when more data arrives.
1735 if (ep->mpa_pkt_len < sizeof(*mpa))
1738 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1739 mpa = (struct mpa_message *) ep->mpa_pkt;
1742 * Validate MPA Header.
1744 if (mpa->revision > mpa_rev) {
1745 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1746 __func__, mpa_rev, mpa->revision);
1747 goto err_stop_timer;
1750 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1751 goto err_stop_timer;
1753 plen = ntohs(mpa->private_data_size);
1756 * Fail if there's too much private data.
1758 if (plen > MPA_MAX_PRIVATE_DATA)
1759 goto err_stop_timer;
1762 * If plen does not account for pkt size
1764 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1765 goto err_stop_timer;
1766 ep->plen = (u8) plen;
1769 * If we don't have all the pdata yet, then bail.
1771 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1775 * If we get here we have accumulated the entire mpa
1776 * start reply message including private data.
1778 ep->mpa_attr.initiator = 0;
1779 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1780 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1781 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1782 ep->mpa_attr.version = mpa->revision;
1783 if (mpa->revision == 1)
1784 ep->tried_with_mpa_v1 = 1;
1785 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1787 if (mpa->revision == 2) {
1788 ep->mpa_attr.enhanced_rdma_conn =
1789 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1790 if (ep->mpa_attr.enhanced_rdma_conn) {
1791 mpa_v2_params = (struct mpa_v2_conn_params *)
1792 (ep->mpa_pkt + sizeof(*mpa));
1793 ep->ird = ntohs(mpa_v2_params->ird) &
1794 MPA_V2_IRD_ORD_MASK;
1795 ep->ird = min_t(u32, ep->ird,
1796 cur_max_read_depth(ep->com.dev));
1797 ep->ord = ntohs(mpa_v2_params->ord) &
1798 MPA_V2_IRD_ORD_MASK;
1799 ep->ord = min_t(u32, ep->ord,
1800 cur_max_read_depth(ep->com.dev));
1801 pr_debug("initiator ird %u ord %u\n",
1803 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1805 if (ntohs(mpa_v2_params->ord) &
1806 MPA_V2_RDMA_WRITE_RTR)
1807 ep->mpa_attr.p2p_type =
1808 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1809 else if (ntohs(mpa_v2_params->ord) &
1810 MPA_V2_RDMA_READ_RTR)
1811 ep->mpa_attr.p2p_type =
1812 FW_RI_INIT_P2PTYPE_READ_REQ;
1815 } else if (mpa->revision == 1)
1817 ep->mpa_attr.p2p_type = p2p_type;
1819 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1820 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1821 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1822 ep->mpa_attr.p2p_type);
1824 __state_set(&ep->com, MPA_REQ_RCVD);
1827 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1828 if (ep->parent_ep->com.state != DEAD) {
1829 if (connect_request_upcall(ep))
1830 goto err_unlock_parent;
1832 goto err_unlock_parent;
1834 mutex_unlock(&ep->parent_ep->com.mutex);
1838 mutex_unlock(&ep->parent_ep->com.mutex);
1841 (void)stop_ep_timer(ep);
1846 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1849 struct cpl_rx_data *hdr = cplhdr(skb);
1850 unsigned int dlen = ntohs(hdr->len);
1851 unsigned int tid = GET_TID(hdr);
1852 __u8 status = hdr->status;
1855 ep = get_ep_from_tid(dev, tid);
1858 pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1859 skb_pull(skb, sizeof(*hdr));
1860 skb_trim(skb, dlen);
1861 mutex_lock(&ep->com.mutex);
1863 switch (ep->com.state) {
1865 update_rx_credits(ep, dlen);
1866 ep->rcv_seq += dlen;
1867 disconnect = process_mpa_reply(ep, skb);
1870 update_rx_credits(ep, dlen);
1871 ep->rcv_seq += dlen;
1872 disconnect = process_mpa_request(ep, skb);
1875 struct c4iw_qp_attributes attrs;
1877 update_rx_credits(ep, dlen);
1879 pr_err("%s Unexpected streaming data." \
1880 " qpid %u ep %p state %d tid %u status %d\n",
1881 __func__, ep->com.qp->wq.sq.qid, ep,
1882 ep->com.state, ep->hwtid, status);
1883 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1884 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1885 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1892 mutex_unlock(&ep->com.mutex);
1894 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1895 c4iw_put_ep(&ep->com);
1899 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1901 enum chip_type adapter_type;
1903 adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1906 * If this TCB had a srq buffer cached, then we must complete
1907 * it. For user mode, that means saving the srqidx in the
1908 * user/kernel status page for this qp. For kernel mode, just
1909 * synthesize the CQE now.
1911 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1912 if (ep->com.qp->ibqp.uobject)
1913 t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1915 c4iw_flush_srqidx(ep->com.qp, srqidx);
1919 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1923 struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1925 unsigned int tid = GET_TID(rpl);
1927 ep = get_ep_from_tid(dev, tid);
1929 pr_warn("Abort rpl to freed endpoint\n");
1933 if (ep->com.qp && ep->com.qp->srq) {
1934 srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1935 complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1938 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1939 mutex_lock(&ep->com.mutex);
1940 switch (ep->com.state) {
1942 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1943 __state_set(&ep->com, DEAD);
1947 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1950 mutex_unlock(&ep->com.mutex);
1953 close_complete_upcall(ep, -ECONNRESET);
1954 release_ep_resources(ep);
1956 c4iw_put_ep(&ep->com);
1960 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1962 struct sk_buff *skb;
1963 struct fw_ofld_connection_wr *req;
1964 unsigned int mtu_idx;
1966 struct sockaddr_in *sin;
1969 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1973 req = __skb_put_zero(skb, sizeof(*req));
1974 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1975 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1976 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1977 ep->com.dev->rdev.lldi.ports[0],
1979 sin = (struct sockaddr_in *)&ep->com.local_addr;
1980 req->le.lport = sin->sin_port;
1981 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1982 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1983 req->le.pport = sin->sin_port;
1984 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1985 req->tcb.t_state_to_astid =
1986 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1987 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1988 req->tcb.cplrxdataack_cplpassacceptrpl =
1989 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1990 req->tcb.tx_max = (__force __be32) jiffies;
1991 req->tcb.rcv_adv = htons(1);
1992 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1993 enable_tcp_timestamps,
1994 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1995 wscale = cxgb_compute_wscale(rcv_win);
1998 * Specify the largest window that will fit in opt0. The
1999 * remainder will be specified in the rx_data_ack.
2001 win = ep->rcv_win >> 10;
2002 if (win > RCV_BUFSIZ_M)
2005 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2006 (nocong ? NO_CONG_F : 0) |
2009 WND_SCALE_V(wscale) |
2010 MSS_IDX_V(mtu_idx) |
2011 L2T_IDX_V(ep->l2t->idx) |
2012 TX_CHAN_V(ep->tx_chan) |
2013 SMAC_SEL_V(ep->smac_idx) |
2014 DSCP_V(ep->tos >> 2) |
2015 ULP_MODE_V(ULP_MODE_TCPDDP) |
2017 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2018 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2020 CCTRL_ECN_V(enable_ecn) |
2021 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2022 if (enable_tcp_timestamps)
2023 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2024 if (enable_tcp_sack)
2025 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2026 if (wscale && enable_tcp_window_scaling)
2027 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2028 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2029 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2030 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2031 set_bit(ACT_OFLD_CONN, &ep->com.history);
2032 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2036 * Some of the error codes above implicitly indicate that there is no TID
2037 * allocated with the result of an ACT_OPEN. We use this predicate to make
2040 static inline int act_open_has_tid(int status)
2042 return (status != CPL_ERR_TCAM_PARITY &&
2043 status != CPL_ERR_TCAM_MISS &&
2044 status != CPL_ERR_TCAM_FULL &&
2045 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2046 status != CPL_ERR_CONN_EXIST);
2049 static char *neg_adv_str(unsigned int status)
2052 case CPL_ERR_RTX_NEG_ADVICE:
2053 return "Retransmit timeout";
2054 case CPL_ERR_PERSIST_NEG_ADVICE:
2055 return "Persist timeout";
2056 case CPL_ERR_KEEPALV_NEG_ADVICE:
2057 return "Keepalive timeout";
2063 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2065 ep->snd_win = snd_win;
2066 ep->rcv_win = rcv_win;
2067 pr_debug("snd_win %d rcv_win %d\n",
2068 ep->snd_win, ep->rcv_win);
2071 #define ACT_OPEN_RETRY_COUNT 2
2073 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2074 struct dst_entry *dst, struct c4iw_dev *cdev,
2075 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2077 struct neighbour *n;
2079 struct net_device *pdev;
2081 n = dst_neigh_lookup(dst, peer_ip);
2087 if (n->dev->flags & IFF_LOOPBACK) {
2089 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2090 else if (IS_ENABLED(CONFIG_IPV6))
2091 for_each_netdev(&init_net, pdev) {
2092 if (ipv6_chk_addr(&init_net,
2093 (struct in6_addr *)peer_ip,
2104 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2105 n, pdev, rt_tos2priority(tos));
2110 ep->mtu = pdev->mtu;
2111 ep->tx_chan = cxgb4_port_chan(pdev);
2112 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2113 step = cdev->rdev.lldi.ntxq /
2114 cdev->rdev.lldi.nchan;
2115 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2116 step = cdev->rdev.lldi.nrxq /
2117 cdev->rdev.lldi.nchan;
2118 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2119 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2120 cxgb4_port_idx(pdev) * step];
2121 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2124 pdev = get_real_dev(n->dev);
2125 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2126 n, pdev, rt_tos2priority(tos));
2129 ep->mtu = dst_mtu(dst);
2130 ep->tx_chan = cxgb4_port_chan(pdev);
2131 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2132 step = cdev->rdev.lldi.ntxq /
2133 cdev->rdev.lldi.nchan;
2134 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2135 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2136 step = cdev->rdev.lldi.nrxq /
2137 cdev->rdev.lldi.nchan;
2138 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2139 cxgb4_port_idx(pdev) * step];
2140 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2143 ep->retry_with_mpa_v1 = 0;
2144 ep->tried_with_mpa_v1 = 0;
2156 static int c4iw_reconnect(struct c4iw_ep *ep)
2160 struct sockaddr_in *laddr = (struct sockaddr_in *)
2161 &ep->com.cm_id->m_local_addr;
2162 struct sockaddr_in *raddr = (struct sockaddr_in *)
2163 &ep->com.cm_id->m_remote_addr;
2164 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2165 &ep->com.cm_id->m_local_addr;
2166 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2167 &ep->com.cm_id->m_remote_addr;
2171 pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2172 c4iw_init_wr_wait(ep->com.wr_waitp);
2174 /* When MPA revision is different on nodes, the node with MPA_rev=2
2175 * tries to reconnect with MPA_rev 1 for the same EP through
2176 * c4iw_reconnect(), where the same EP is assigned with new tid for
2177 * further connection establishment. As we are using the same EP pointer
2178 * for reconnect, few skbs are used during the previous c4iw_connect(),
2179 * which leaves the EP with inadequate skbs for further
2180 * c4iw_reconnect(), Further causing a crash due to an empty
2181 * skb_list() during peer_abort(). Allocate skbs which is already used.
2183 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2184 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2190 * Allocate an active TID to initiate a TCP connection.
2192 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2193 if (ep->atid == -1) {
2194 pr_err("%s - cannot alloc atid\n", __func__);
2198 err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2203 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2204 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2205 laddr->sin_addr.s_addr,
2206 raddr->sin_addr.s_addr,
2208 raddr->sin_port, ep->com.cm_id->tos);
2210 ra = (__u8 *)&raddr->sin_addr;
2212 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2214 laddr6->sin6_addr.s6_addr,
2215 raddr6->sin6_addr.s6_addr,
2219 raddr6->sin6_scope_id);
2221 ra = (__u8 *)&raddr6->sin6_addr;
2224 pr_err("%s - cannot find route\n", __func__);
2225 err = -EHOSTUNREACH;
2228 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2229 ep->com.dev->rdev.lldi.adapter_type,
2230 ep->com.cm_id->tos);
2232 pr_err("%s - cannot alloc l2e\n", __func__);
2236 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2237 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2240 state_set(&ep->com, CONNECTING);
2241 ep->tos = ep->com.cm_id->tos;
2243 /* send connect request to rnic */
2244 err = send_connect(ep);
2248 cxgb4_l2t_release(ep->l2t);
2250 dst_release(ep->dst);
2252 xa_erase_irq(&ep->com.dev->atids, ep->atid);
2254 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2257 * remember to send notification to upper layer.
2258 * We are in here so the upper layer is not aware that this is
2259 * re-connect attempt and so, upper layer is still waiting for
2260 * response of 1st connect request.
2262 connect_reply_upcall(ep, -ECONNRESET);
2264 c4iw_put_ep(&ep->com);
2269 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2272 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2273 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2274 ntohl(rpl->atid_status)));
2275 struct tid_info *t = dev->rdev.lldi.tids;
2276 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2277 struct sockaddr_in *la;
2278 struct sockaddr_in *ra;
2279 struct sockaddr_in6 *la6;
2280 struct sockaddr_in6 *ra6;
2283 ep = lookup_atid(t, atid);
2287 la = (struct sockaddr_in *)&ep->com.local_addr;
2288 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2289 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2290 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2292 pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2293 status, status2errno(status));
2295 if (cxgb_is_neg_adv(status)) {
2296 pr_debug("Connection problems for atid %u status %u (%s)\n",
2297 atid, status, neg_adv_str(status));
2298 ep->stats.connect_neg_adv++;
2299 mutex_lock(&dev->rdev.stats.lock);
2300 dev->rdev.stats.neg_adv++;
2301 mutex_unlock(&dev->rdev.stats.lock);
2305 set_bit(ACT_OPEN_RPL, &ep->com.history);
2308 * Log interesting failures.
2311 case CPL_ERR_CONN_RESET:
2312 case CPL_ERR_CONN_TIMEDOUT:
2314 case CPL_ERR_TCAM_FULL:
2315 mutex_lock(&dev->rdev.stats.lock);
2316 dev->rdev.stats.tcam_full++;
2317 mutex_unlock(&dev->rdev.stats.lock);
2318 if (ep->com.local_addr.ss_family == AF_INET &&
2319 dev->rdev.lldi.enable_fw_ofld_conn) {
2320 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2321 ntohl(rpl->atid_status))));
2327 case CPL_ERR_CONN_EXIST:
2328 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2329 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2330 if (ep->com.remote_addr.ss_family == AF_INET6) {
2331 struct sockaddr_in6 *sin6 =
2332 (struct sockaddr_in6 *)
2333 &ep->com.local_addr;
2335 ep->com.dev->rdev.lldi.ports[0],
2337 &sin6->sin6_addr.s6_addr, 1);
2339 xa_erase_irq(&ep->com.dev->atids, atid);
2340 cxgb4_free_atid(t, atid);
2341 dst_release(ep->dst);
2342 cxgb4_l2t_release(ep->l2t);
2348 if (ep->com.local_addr.ss_family == AF_INET) {
2349 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2350 atid, status, status2errno(status),
2351 &la->sin_addr.s_addr, ntohs(la->sin_port),
2352 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2354 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2355 atid, status, status2errno(status),
2356 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2357 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2363 connect_reply_upcall(ep, status2errno(status));
2364 state_set(&ep->com, DEAD);
2366 if (ep->com.remote_addr.ss_family == AF_INET6) {
2367 struct sockaddr_in6 *sin6 =
2368 (struct sockaddr_in6 *)&ep->com.local_addr;
2369 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2370 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2372 if (status && act_open_has_tid(status))
2373 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2374 ep->com.local_addr.ss_family);
2376 xa_erase_irq(&ep->com.dev->atids, atid);
2377 cxgb4_free_atid(t, atid);
2378 dst_release(ep->dst);
2379 cxgb4_l2t_release(ep->l2t);
2380 c4iw_put_ep(&ep->com);
2385 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2387 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2388 unsigned int stid = GET_TID(rpl);
2389 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2392 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2395 pr_debug("ep %p status %d error %d\n", ep,
2396 rpl->status, status2errno(rpl->status));
2397 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2398 c4iw_put_ep(&ep->com);
2403 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2405 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2406 unsigned int stid = GET_TID(rpl);
2407 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2410 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2413 pr_debug("ep %p\n", ep);
2414 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2415 c4iw_put_ep(&ep->com);
2420 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2421 struct cpl_pass_accept_req *req)
2423 struct cpl_pass_accept_rpl *rpl;
2424 unsigned int mtu_idx;
2428 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2430 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2432 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2433 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2434 enable_tcp_timestamps && req->tcpopt.tstamp,
2435 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2436 wscale = cxgb_compute_wscale(rcv_win);
2439 * Specify the largest window that will fit in opt0. The
2440 * remainder will be specified in the rx_data_ack.
2442 win = ep->rcv_win >> 10;
2443 if (win > RCV_BUFSIZ_M)
2445 opt0 = (nocong ? NO_CONG_F : 0) |
2448 WND_SCALE_V(wscale) |
2449 MSS_IDX_V(mtu_idx) |
2450 L2T_IDX_V(ep->l2t->idx) |
2451 TX_CHAN_V(ep->tx_chan) |
2452 SMAC_SEL_V(ep->smac_idx) |
2453 DSCP_V(ep->tos >> 2) |
2454 ULP_MODE_V(ULP_MODE_TCPDDP) |
2456 opt2 = RX_CHANNEL_V(0) |
2457 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2459 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2460 opt2 |= TSTAMPS_EN_F;
2461 if (enable_tcp_sack && req->tcpopt.sack)
2463 if (wscale && enable_tcp_window_scaling)
2464 opt2 |= WND_SCALE_EN_F;
2466 const struct tcphdr *tcph;
2467 u32 hlen = ntohl(req->hdr_len);
2469 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2470 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2473 tcph = (const void *)(req + 1) +
2474 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2475 if (tcph->ece && tcph->cwr)
2476 opt2 |= CCTRL_ECN_V(1);
2479 if (!is_t4(adapter_type)) {
2480 u32 isn = (get_random_u32() & ~7UL) - 1;
2482 skb = get_skb(skb, roundup(sizeof(*rpl5), 16), GFP_KERNEL);
2483 rpl5 = __skb_put_zero(skb, roundup(sizeof(*rpl5), 16));
2485 INIT_TP_WR_CPL(rpl5, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2486 opt2 |= T5_OPT_2_VALID_F;
2487 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2491 rpl5->iss = cpu_to_be32(isn);
2492 pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2494 skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2495 rpl = __skb_put_zero(skb, sizeof(*rpl));
2496 INIT_TP_WR_CPL(rpl, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2499 rpl->opt0 = cpu_to_be64(opt0);
2500 rpl->opt2 = cpu_to_be32(opt2);
2501 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2502 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2504 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2507 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2509 pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2510 skb_trim(skb, sizeof(struct cpl_tid_release));
2511 release_tid(&dev->rdev, hwtid, skb);
2515 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2517 struct c4iw_ep *child_ep = NULL, *parent_ep;
2518 struct cpl_pass_accept_req *req = cplhdr(skb);
2519 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2520 struct tid_info *t = dev->rdev.lldi.tids;
2521 unsigned int hwtid = GET_TID(req);
2522 struct dst_entry *dst;
2523 __u8 local_ip[16], peer_ip[16];
2524 __be16 local_port, peer_port;
2525 struct sockaddr_in6 *sin6;
2527 u16 peer_mss = ntohs(req->tcpopt.mss);
2529 unsigned short hdrs;
2532 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2534 pr_err("%s connect request on invalid stid %d\n",
2539 if (state_read(&parent_ep->com) != LISTEN) {
2540 pr_err("%s - listening ep not in LISTEN\n", __func__);
2544 if (parent_ep->com.cm_id->tos_set)
2545 tos = parent_ep->com.cm_id->tos;
2547 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2549 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2550 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2552 /* Find output route */
2554 pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2556 local_ip, peer_ip, ntohs(local_port),
2557 ntohs(peer_port), peer_mss);
2558 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2559 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2560 local_port, peer_port, tos);
2562 pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2564 local_ip, peer_ip, ntohs(local_port),
2565 ntohs(peer_port), peer_mss);
2566 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2567 local_ip, peer_ip, local_port, peer_port,
2569 ((struct sockaddr_in6 *)
2570 &parent_ep->com.local_addr)->sin6_scope_id);
2573 pr_err("%s - failed to find dst entry!\n", __func__);
2577 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2579 pr_err("%s - failed to allocate ep entry!\n", __func__);
2584 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2585 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2587 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2593 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2594 sizeof(struct tcphdr) +
2595 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2596 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2597 child_ep->mtu = peer_mss + hdrs;
2599 skb_queue_head_init(&child_ep->com.ep_skb_list);
2600 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2603 state_set(&child_ep->com, CONNECTING);
2604 child_ep->com.dev = dev;
2605 child_ep->com.cm_id = NULL;
2608 struct sockaddr_in *sin = (struct sockaddr_in *)
2609 &child_ep->com.local_addr;
2611 sin->sin_family = AF_INET;
2612 sin->sin_port = local_port;
2613 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2615 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2616 sin->sin_family = AF_INET;
2617 sin->sin_port = ((struct sockaddr_in *)
2618 &parent_ep->com.local_addr)->sin_port;
2619 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2621 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2622 sin->sin_family = AF_INET;
2623 sin->sin_port = peer_port;
2624 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2626 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2627 sin6->sin6_family = PF_INET6;
2628 sin6->sin6_port = local_port;
2629 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2631 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2632 sin6->sin6_family = PF_INET6;
2633 sin6->sin6_port = ((struct sockaddr_in6 *)
2634 &parent_ep->com.local_addr)->sin6_port;
2635 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2637 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2638 sin6->sin6_family = PF_INET6;
2639 sin6->sin6_port = peer_port;
2640 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2643 c4iw_get_ep(&parent_ep->com);
2644 child_ep->parent_ep = parent_ep;
2645 child_ep->tos = tos;
2646 child_ep->dst = dst;
2647 child_ep->hwtid = hwtid;
2649 pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2650 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2652 timer_setup(&child_ep->timer, ep_timeout, 0);
2653 cxgb4_insert_tid(t, child_ep, hwtid,
2654 child_ep->com.local_addr.ss_family);
2655 insert_ep_tid(child_ep);
2656 if (accept_cr(child_ep, skb, req)) {
2657 c4iw_put_ep(&parent_ep->com);
2658 release_ep_resources(child_ep);
2660 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2663 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2664 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2665 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2669 c4iw_put_ep(&child_ep->com);
2671 reject_cr(dev, hwtid, skb);
2674 c4iw_put_ep(&parent_ep->com);
2678 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2681 struct cpl_pass_establish *req = cplhdr(skb);
2682 unsigned int tid = GET_TID(req);
2684 u16 tcp_opt = ntohs(req->tcp_opt);
2686 ep = get_ep_from_tid(dev, tid);
2690 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2691 ep->snd_seq = be32_to_cpu(req->snd_isn);
2692 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2693 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2695 pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2697 set_emss(ep, tcp_opt);
2699 dst_confirm(ep->dst);
2700 mutex_lock(&ep->com.mutex);
2701 ep->com.state = MPA_REQ_WAIT;
2703 set_bit(PASS_ESTAB, &ep->com.history);
2704 ret = send_flowc(ep);
2705 mutex_unlock(&ep->com.mutex);
2707 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2708 c4iw_put_ep(&ep->com);
2713 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2715 struct cpl_peer_close *hdr = cplhdr(skb);
2717 struct c4iw_qp_attributes attrs;
2720 unsigned int tid = GET_TID(hdr);
2723 ep = get_ep_from_tid(dev, tid);
2727 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2728 dst_confirm(ep->dst);
2730 set_bit(PEER_CLOSE, &ep->com.history);
2731 mutex_lock(&ep->com.mutex);
2732 switch (ep->com.state) {
2734 __state_set(&ep->com, CLOSING);
2737 __state_set(&ep->com, CLOSING);
2738 connect_reply_upcall(ep, -ECONNRESET);
2743 * We're gonna mark this puppy DEAD, but keep
2744 * the reference on it until the ULP accepts or
2745 * rejects the CR. Also wake up anyone waiting
2746 * in rdma connection migration (see c4iw_accept_cr()).
2748 __state_set(&ep->com, CLOSING);
2749 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2750 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2753 __state_set(&ep->com, CLOSING);
2754 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2755 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2759 __state_set(&ep->com, CLOSING);
2760 attrs.next_state = C4IW_QP_STATE_CLOSING;
2761 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2762 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2763 if (ret != -ECONNRESET) {
2764 peer_close_upcall(ep);
2772 __state_set(&ep->com, MORIBUND);
2776 (void)stop_ep_timer(ep);
2777 if (ep->com.cm_id && ep->com.qp) {
2778 attrs.next_state = C4IW_QP_STATE_IDLE;
2779 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2780 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2782 close_complete_upcall(ep, 0);
2783 __state_set(&ep->com, DEAD);
2791 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2793 mutex_unlock(&ep->com.mutex);
2795 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2797 release_ep_resources(ep);
2798 c4iw_put_ep(&ep->com);
2802 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2804 complete_cached_srq_buffers(ep, ep->srqe_idx);
2805 if (ep->com.cm_id && ep->com.qp) {
2806 struct c4iw_qp_attributes attrs;
2808 attrs.next_state = C4IW_QP_STATE_ERROR;
2809 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2810 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2812 peer_abort_upcall(ep);
2813 release_ep_resources(ep);
2814 c4iw_put_ep(&ep->com);
2817 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2819 struct cpl_abort_req_rss6 *req = cplhdr(skb);
2821 struct sk_buff *rpl_skb;
2822 struct c4iw_qp_attributes attrs;
2825 unsigned int tid = GET_TID(req);
2829 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2831 ep = get_ep_from_tid(dev, tid);
2835 status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2837 if (cxgb_is_neg_adv(status)) {
2838 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2839 ep->hwtid, status, neg_adv_str(status));
2840 ep->stats.abort_neg_adv++;
2841 mutex_lock(&dev->rdev.stats.lock);
2842 dev->rdev.stats.neg_adv++;
2843 mutex_unlock(&dev->rdev.stats.lock);
2847 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2849 set_bit(PEER_ABORT, &ep->com.history);
2852 * Wake up any threads in rdma_init() or rdma_fini().
2853 * However, this is not needed if com state is just
2856 if (ep->com.state != MPA_REQ_SENT)
2857 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2859 mutex_lock(&ep->com.mutex);
2860 switch (ep->com.state) {
2862 c4iw_put_ep(&ep->parent_ep->com);
2865 (void)stop_ep_timer(ep);
2868 (void)stop_ep_timer(ep);
2869 if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2870 (mpa_rev == 2 && ep->tried_with_mpa_v1))
2871 connect_reply_upcall(ep, -ECONNRESET);
2874 * we just don't send notification upwards because we
2875 * want to retry with mpa_v1 without upper layers even
2878 * do some housekeeping so as to re-initiate the
2881 pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2883 ep->retry_with_mpa_v1 = 1;
2895 if (ep->com.qp && ep->com.qp->srq) {
2896 srqidx = ABORT_RSS_SRQIDX_G(
2897 be32_to_cpu(req->srqidx_status));
2899 complete_cached_srq_buffers(ep, srqidx);
2901 /* Hold ep ref until finish_peer_abort() */
2902 c4iw_get_ep(&ep->com);
2903 __state_set(&ep->com, ABORTING);
2904 set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2911 if (ep->com.cm_id && ep->com.qp) {
2912 attrs.next_state = C4IW_QP_STATE_ERROR;
2913 ret = c4iw_modify_qp(ep->com.qp->rhp,
2914 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2917 pr_err("%s - qp <- error failed!\n", __func__);
2919 peer_abort_upcall(ep);
2924 pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2925 mutex_unlock(&ep->com.mutex);
2928 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2931 dst_confirm(ep->dst);
2932 if (ep->com.state != ABORTING) {
2933 __state_set(&ep->com, DEAD);
2934 /* we don't release if we want to retry with mpa_v1 */
2935 if (!ep->retry_with_mpa_v1)
2938 mutex_unlock(&ep->com.mutex);
2940 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2941 if (WARN_ON(!rpl_skb)) {
2946 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2948 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2951 release_ep_resources(ep);
2952 else if (ep->retry_with_mpa_v1) {
2953 if (ep->com.remote_addr.ss_family == AF_INET6) {
2954 struct sockaddr_in6 *sin6 =
2955 (struct sockaddr_in6 *)
2956 &ep->com.local_addr;
2958 ep->com.dev->rdev.lldi.ports[0],
2959 (const u32 *)&sin6->sin6_addr.s6_addr,
2962 xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2963 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2964 ep->com.local_addr.ss_family);
2965 dst_release(ep->dst);
2966 cxgb4_l2t_release(ep->l2t);
2971 c4iw_put_ep(&ep->com);
2972 /* Dereferencing ep, referenced in peer_abort_intr() */
2973 c4iw_put_ep(&ep->com);
2977 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2980 struct c4iw_qp_attributes attrs;
2981 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2983 unsigned int tid = GET_TID(rpl);
2985 ep = get_ep_from_tid(dev, tid);
2989 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2991 /* The cm_id may be null if we failed to connect */
2992 mutex_lock(&ep->com.mutex);
2993 set_bit(CLOSE_CON_RPL, &ep->com.history);
2994 switch (ep->com.state) {
2996 __state_set(&ep->com, MORIBUND);
2999 (void)stop_ep_timer(ep);
3000 if ((ep->com.cm_id) && (ep->com.qp)) {
3001 attrs.next_state = C4IW_QP_STATE_IDLE;
3002 c4iw_modify_qp(ep->com.qp->rhp,
3004 C4IW_QP_ATTR_NEXT_STATE,
3007 close_complete_upcall(ep, 0);
3008 __state_set(&ep->com, DEAD);
3015 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3018 mutex_unlock(&ep->com.mutex);
3020 release_ep_resources(ep);
3021 c4iw_put_ep(&ep->com);
3025 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3027 struct cpl_rdma_terminate *rpl = cplhdr(skb);
3028 unsigned int tid = GET_TID(rpl);
3030 struct c4iw_qp_attributes attrs;
3032 ep = get_ep_from_tid(dev, tid);
3036 pr_warn("TERM received tid %u qpid %u\n", tid,
3037 ep->com.qp->wq.sq.qid);
3038 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3039 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3040 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3043 /* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3044 * when entering the TERM state the RNIC MUST initiate a CLOSE.
3046 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3047 c4iw_put_ep(&ep->com);
3049 pr_warn("TERM received tid %u no ep/qp\n", tid);
3055 * Upcall from the adapter indicating data has been transmitted.
3056 * For us its just the single MPA request or reply. We can now free
3057 * the skb holding the mpa message.
3059 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3062 struct cpl_fw4_ack *hdr = cplhdr(skb);
3063 u8 credits = hdr->credits;
3064 unsigned int tid = GET_TID(hdr);
3067 ep = get_ep_from_tid(dev, tid);
3070 pr_debug("ep %p tid %u credits %u\n",
3071 ep, ep->hwtid, credits);
3073 pr_debug("0 credit ack ep %p tid %u state %u\n",
3074 ep, ep->hwtid, state_read(&ep->com));
3078 dst_confirm(ep->dst);
3080 pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3081 ep, ep->hwtid, state_read(&ep->com),
3082 ep->mpa_attr.initiator ? 1 : 0);
3083 mutex_lock(&ep->com.mutex);
3084 kfree_skb(ep->mpa_skb);
3086 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3088 mutex_unlock(&ep->com.mutex);
3091 c4iw_put_ep(&ep->com);
3095 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3098 struct c4iw_ep *ep = to_ep(cm_id);
3100 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3102 mutex_lock(&ep->com.mutex);
3103 if (ep->com.state != MPA_REQ_RCVD) {
3104 mutex_unlock(&ep->com.mutex);
3105 c4iw_put_ep(&ep->com);
3108 set_bit(ULP_REJECT, &ep->com.history);
3112 abort = send_mpa_reject(ep, pdata, pdata_len);
3113 mutex_unlock(&ep->com.mutex);
3116 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3117 c4iw_put_ep(&ep->com);
3121 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3124 struct c4iw_qp_attributes attrs;
3125 enum c4iw_qp_attr_mask mask;
3126 struct c4iw_ep *ep = to_ep(cm_id);
3127 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3128 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3131 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3133 mutex_lock(&ep->com.mutex);
3134 if (ep->com.state != MPA_REQ_RCVD) {
3144 set_bit(ULP_ACCEPT, &ep->com.history);
3145 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3146 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3151 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3152 if (conn_param->ord > ep->ird) {
3153 if (RELAXED_IRD_NEGOTIATION) {
3154 conn_param->ord = ep->ird;
3156 ep->ird = conn_param->ird;
3157 ep->ord = conn_param->ord;
3158 send_mpa_reject(ep, conn_param->private_data,
3159 conn_param->private_data_len);
3164 if (conn_param->ird < ep->ord) {
3165 if (RELAXED_IRD_NEGOTIATION &&
3166 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3167 conn_param->ird = ep->ord;
3174 ep->ird = conn_param->ird;
3175 ep->ord = conn_param->ord;
3177 if (ep->mpa_attr.version == 1) {
3178 if (peer2peer && ep->ird == 0)
3182 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3183 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3187 pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3189 ep->com.cm_id = cm_id;
3190 ref_cm_id(&ep->com);
3194 /* bind QP to EP and move to RTS */
3195 attrs.mpa_attr = ep->mpa_attr;
3196 attrs.max_ird = ep->ird;
3197 attrs.max_ord = ep->ord;
3198 attrs.llp_stream_handle = ep;
3199 attrs.next_state = C4IW_QP_STATE_RTS;
3201 /* bind QP and TID with INIT_WR */
3202 mask = C4IW_QP_ATTR_NEXT_STATE |
3203 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3204 C4IW_QP_ATTR_MPA_ATTR |
3205 C4IW_QP_ATTR_MAX_IRD |
3206 C4IW_QP_ATTR_MAX_ORD;
3208 err = c4iw_modify_qp(ep->com.qp->rhp,
3209 ep->com.qp, mask, &attrs, 1);
3211 goto err_deref_cm_id;
3213 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3214 err = send_mpa_reply(ep, conn_param->private_data,
3215 conn_param->private_data_len);
3217 goto err_deref_cm_id;
3219 __state_set(&ep->com, FPDU_MODE);
3220 established_upcall(ep);
3221 mutex_unlock(&ep->com.mutex);
3222 c4iw_put_ep(&ep->com);
3225 deref_cm_id(&ep->com);
3229 mutex_unlock(&ep->com.mutex);
3231 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3232 c4iw_put_ep(&ep->com);
3236 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3238 struct in_device *ind;
3240 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3241 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3242 const struct in_ifaddr *ifa;
3244 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3246 return -EADDRNOTAVAIL;
3248 in_dev_for_each_ifa_rcu(ifa, ind) {
3249 if (ifa->ifa_flags & IFA_F_SECONDARY)
3251 laddr->sin_addr.s_addr = ifa->ifa_address;
3252 raddr->sin_addr.s_addr = ifa->ifa_address;
3259 return found ? 0 : -EADDRNOTAVAIL;
3262 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3263 unsigned char banned_flags)
3265 struct inet6_dev *idev;
3266 int err = -EADDRNOTAVAIL;
3269 idev = __in6_dev_get(dev);
3271 struct inet6_ifaddr *ifp;
3273 read_lock_bh(&idev->lock);
3274 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3275 if (ifp->scope == IFA_LINK &&
3276 !(ifp->flags & banned_flags)) {
3277 memcpy(addr, &ifp->addr, 16);
3282 read_unlock_bh(&idev->lock);
3288 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3290 struct in6_addr addr;
3291 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3292 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3294 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3295 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3296 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3299 return -EADDRNOTAVAIL;
3302 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3304 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3307 struct sockaddr_in *laddr;
3308 struct sockaddr_in *raddr;
3309 struct sockaddr_in6 *laddr6;
3310 struct sockaddr_in6 *raddr6;
3314 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3315 (conn_param->ird > cur_max_read_depth(dev))) {
3319 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3321 pr_err("%s - cannot alloc ep\n", __func__);
3326 skb_queue_head_init(&ep->com.ep_skb_list);
3327 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3332 timer_setup(&ep->timer, ep_timeout, 0);
3333 ep->plen = conn_param->private_data_len;
3335 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3336 conn_param->private_data, ep->plen);
3337 ep->ird = conn_param->ird;
3338 ep->ord = conn_param->ord;
3340 if (peer2peer && ep->ord == 0)
3343 ep->com.cm_id = cm_id;
3344 ref_cm_id(&ep->com);
3345 cm_id->provider_data = ep;
3347 ep->com.qp = get_qhp(dev, conn_param->qpn);
3349 pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3354 pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3358 * Allocate an active TID to initiate a TCP connection.
3360 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3361 if (ep->atid == -1) {
3362 pr_err("%s - cannot alloc atid\n", __func__);
3366 err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3370 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3371 sizeof(ep->com.local_addr));
3372 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3373 sizeof(ep->com.remote_addr));
3375 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3376 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3377 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3378 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3380 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3382 ra = (__u8 *)&raddr->sin_addr;
3385 * Handle loopback requests to INADDR_ANY.
3387 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3388 err = pick_local_ipaddrs(dev, cm_id);
3394 pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3395 &laddr->sin_addr, ntohs(laddr->sin_port),
3396 ra, ntohs(raddr->sin_port));
3397 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3398 laddr->sin_addr.s_addr,
3399 raddr->sin_addr.s_addr,
3401 raddr->sin_port, cm_id->tos);
3404 ra = (__u8 *)&raddr6->sin6_addr;
3407 * Handle loopback requests to INADDR_ANY.
3409 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3410 err = pick_local_ip6addrs(dev, cm_id);
3416 pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3417 laddr6->sin6_addr.s6_addr,
3418 ntohs(laddr6->sin6_port),
3419 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3420 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3421 laddr6->sin6_addr.s6_addr,
3422 raddr6->sin6_addr.s6_addr,
3424 raddr6->sin6_port, cm_id->tos,
3425 raddr6->sin6_scope_id);
3428 pr_err("%s - cannot find route\n", __func__);
3429 err = -EHOSTUNREACH;
3433 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3434 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3436 pr_err("%s - cannot alloc l2e\n", __func__);
3440 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3441 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3444 state_set(&ep->com, CONNECTING);
3445 ep->tos = cm_id->tos;
3447 /* send connect request to rnic */
3448 err = send_connect(ep);
3452 cxgb4_l2t_release(ep->l2t);
3454 dst_release(ep->dst);
3456 xa_erase_irq(&ep->com.dev->atids, ep->atid);
3458 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3460 skb_queue_purge(&ep->com.ep_skb_list);
3461 deref_cm_id(&ep->com);
3463 c4iw_put_ep(&ep->com);
3468 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3471 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3472 &ep->com.local_addr;
3474 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3475 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3476 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3480 c4iw_init_wr_wait(ep->com.wr_waitp);
3481 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3482 ep->stid, &sin6->sin6_addr,
3484 ep->com.dev->rdev.lldi.rxq_ids[0]);
3486 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3490 err = net_xmit_errno(err);
3492 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3493 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3494 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3496 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3501 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3504 struct sockaddr_in *sin = (struct sockaddr_in *)
3505 &ep->com.local_addr;
3507 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3509 err = cxgb4_create_server_filter(
3510 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3511 sin->sin_addr.s_addr, sin->sin_port, 0,
3512 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3513 if (err == -EBUSY) {
3514 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3518 set_current_state(TASK_UNINTERRUPTIBLE);
3519 schedule_timeout(usecs_to_jiffies(100));
3521 } while (err == -EBUSY);
3523 c4iw_init_wr_wait(ep->com.wr_waitp);
3524 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3525 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3526 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3528 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3532 err = net_xmit_errno(err);
3535 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3537 &sin->sin_addr, ntohs(sin->sin_port));
3541 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3544 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3545 struct c4iw_listen_ep *ep;
3549 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3551 pr_err("%s - cannot alloc ep\n", __func__);
3555 skb_queue_head_init(&ep->com.ep_skb_list);
3556 pr_debug("ep %p\n", ep);
3557 ep->com.cm_id = cm_id;
3558 ref_cm_id(&ep->com);
3560 ep->backlog = backlog;
3561 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3562 sizeof(ep->com.local_addr));
3565 * Allocate a server TID.
3567 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3568 ep->com.local_addr.ss_family == AF_INET)
3569 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3570 cm_id->m_local_addr.ss_family, ep);
3572 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3573 cm_id->m_local_addr.ss_family, ep);
3575 if (ep->stid == -1) {
3576 pr_err("%s - cannot alloc stid\n", __func__);
3580 err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3584 state_set(&ep->com, LISTEN);
3585 if (ep->com.local_addr.ss_family == AF_INET)
3586 err = create_server4(dev, ep);
3588 err = create_server6(dev, ep);
3590 cm_id->provider_data = ep;
3593 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3595 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3596 ep->com.local_addr.ss_family);
3598 deref_cm_id(&ep->com);
3599 c4iw_put_ep(&ep->com);
3605 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3608 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3610 pr_debug("ep %p\n", ep);
3613 state_set(&ep->com, DEAD);
3614 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3615 ep->com.local_addr.ss_family == AF_INET) {
3616 err = cxgb4_remove_server_filter(
3617 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3618 ep->com.dev->rdev.lldi.rxq_ids[0], false);
3620 struct sockaddr_in6 *sin6;
3621 c4iw_init_wr_wait(ep->com.wr_waitp);
3622 err = cxgb4_remove_server(
3623 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3624 ep->com.dev->rdev.lldi.rxq_ids[0],
3625 ep->com.local_addr.ss_family == AF_INET6);
3628 err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3630 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3631 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3632 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3634 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3635 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3636 ep->com.local_addr.ss_family);
3638 deref_cm_id(&ep->com);
3639 c4iw_put_ep(&ep->com);
3643 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3648 struct c4iw_rdev *rdev;
3650 mutex_lock(&ep->com.mutex);
3652 pr_debug("ep %p state %s, abrupt %d\n", ep,
3653 states[ep->com.state], abrupt);
3656 * Ref the ep here in case we have fatal errors causing the
3657 * ep to be released and freed.
3659 c4iw_get_ep(&ep->com);
3661 rdev = &ep->com.dev->rdev;
3662 if (c4iw_fatal_error(rdev)) {
3664 close_complete_upcall(ep, -EIO);
3665 ep->com.state = DEAD;
3667 switch (ep->com.state) {
3676 ep->com.state = ABORTING;
3678 ep->com.state = CLOSING;
3681 * if we close before we see the fw4_ack() then we fix
3682 * up the timer state since we're reusing it.
3685 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3686 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3691 set_bit(CLOSE_SENT, &ep->com.flags);
3694 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3697 (void)stop_ep_timer(ep);
3698 ep->com.state = ABORTING;
3700 ep->com.state = MORIBUND;
3706 pr_debug("ignoring disconnect ep %p state %u\n",
3710 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3716 set_bit(EP_DISC_ABORT, &ep->com.history);
3717 ret = send_abort(ep);
3719 set_bit(EP_DISC_CLOSE, &ep->com.history);
3720 ret = send_halfclose(ep);
3723 set_bit(EP_DISC_FAIL, &ep->com.history);
3726 close_complete_upcall(ep, -EIO);
3729 struct c4iw_qp_attributes attrs;
3731 attrs.next_state = C4IW_QP_STATE_ERROR;
3732 ret = c4iw_modify_qp(ep->com.qp->rhp,
3734 C4IW_QP_ATTR_NEXT_STATE,
3737 pr_err("%s - qp <- error failed!\n",
3743 mutex_unlock(&ep->com.mutex);
3744 c4iw_put_ep(&ep->com);
3746 release_ep_resources(ep);
3750 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3751 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3754 int atid = be32_to_cpu(req->tid);
3756 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3757 (__force u32) req->tid);
3761 switch (req->retval) {
3763 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3764 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3765 send_fw_act_open_req(ep, atid);
3770 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3771 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3772 send_fw_act_open_req(ep, atid);
3777 pr_info("%s unexpected ofld conn wr retval %d\n",
3778 __func__, req->retval);
3781 pr_err("active ofld_connect_wr failure %d atid %d\n",
3783 mutex_lock(&dev->rdev.stats.lock);
3784 dev->rdev.stats.act_ofld_conn_fails++;
3785 mutex_unlock(&dev->rdev.stats.lock);
3786 connect_reply_upcall(ep, status2errno(req->retval));
3787 state_set(&ep->com, DEAD);
3788 if (ep->com.remote_addr.ss_family == AF_INET6) {
3789 struct sockaddr_in6 *sin6 =
3790 (struct sockaddr_in6 *)&ep->com.local_addr;
3791 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3792 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3794 xa_erase_irq(&dev->atids, atid);
3795 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3796 dst_release(ep->dst);
3797 cxgb4_l2t_release(ep->l2t);
3798 c4iw_put_ep(&ep->com);
3801 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3802 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3804 struct sk_buff *rpl_skb;
3805 struct cpl_pass_accept_req *cpl;
3808 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3810 pr_err("%s passive open failure %d\n", __func__, req->retval);
3811 mutex_lock(&dev->rdev.stats.lock);
3812 dev->rdev.stats.pas_ofld_conn_fails++;
3813 mutex_unlock(&dev->rdev.stats.lock);
3816 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3817 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3818 (__force u32) htonl(
3819 (__force u32) req->tid)));
3820 ret = pass_accept_req(dev, rpl_skb);
3827 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3829 u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3830 u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3834 t = (thi << shift) | (tlo >> shift);
3839 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3842 u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3846 v = (t >> shift) & mask;
3850 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3852 struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3853 __be64 *tcb = (__be64 *)(rpl + 1);
3854 unsigned int tid = GET_TID(rpl);
3859 ep = get_ep_from_tid(dev, tid);
3862 /* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3863 * determine if there's a rx PDU feedback event pending.
3865 * If that bit is set, it means we'll need to re-read the TCB's
3866 * rq_start value. The final value is the one present in a TCB
3867 * with the TF_RX_PDU_OUT bit cleared.
3870 t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3871 rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3873 c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3874 c4iw_put_ep(&ep->com); /* from read_tcb() */
3876 /* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3878 if (++ep->rx_pdu_out_cnt >= 2) {
3879 WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3886 ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3889 pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3891 if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3892 finish_peer_abort(dev, ep);
3893 else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3896 WARN_ONCE(1, "unexpected state!");
3901 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3903 struct cpl_fw6_msg *rpl = cplhdr(skb);
3904 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3906 switch (rpl->type) {
3908 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3910 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3911 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3912 switch (req->t_state) {
3914 active_ofld_conn_reply(dev, skb, req);
3917 passive_ofld_conn_reply(dev, skb, req);
3920 pr_err("%s unexpected ofld conn wr state %d\n",
3921 __func__, req->t_state);
3929 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3932 __be16 hdr_len, vlantag, len;
3934 int tcp_hdr_len, ip_hdr_len;
3936 struct cpl_rx_pkt *cpl = cplhdr(skb);
3937 struct cpl_pass_accept_req *req;
3938 struct tcp_options_received tmp_opt;
3939 struct c4iw_dev *dev;
3940 enum chip_type type;
3942 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3943 /* Store values from cpl_rx_pkt in temporary location. */
3944 vlantag = cpl->vlan;
3946 l2info = cpl->l2info;
3947 hdr_len = cpl->hdr_len;
3950 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3953 * We need to parse the TCP options from SYN packet.
3954 * to generate cpl_pass_accept_req.
3956 memset(&tmp_opt, 0, sizeof(tmp_opt));
3957 tcp_clear_options(&tmp_opt);
3958 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3960 req = __skb_push(skb, sizeof(*req));
3961 memset(req, 0, sizeof(*req));
3962 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3963 SYN_MAC_IDX_V(RX_MACIDX_G(
3964 be32_to_cpu(l2info))) |
3966 type = dev->rdev.lldi.adapter_type;
3967 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3968 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3970 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3971 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3972 eth_hdr_len = is_t4(type) ?
3973 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3974 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3975 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3976 IP_HDR_LEN_V(ip_hdr_len) |
3977 ETH_HDR_LEN_V(eth_hdr_len));
3978 } else { /* T6 and later */
3979 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3980 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3981 T6_IP_HDR_LEN_V(ip_hdr_len) |
3982 T6_ETH_HDR_LEN_V(eth_hdr_len));
3984 req->vlan = vlantag;
3986 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3987 PASS_OPEN_TOS_V(tos));
3988 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3989 if (tmp_opt.wscale_ok)
3990 req->tcpopt.wsf = tmp_opt.snd_wscale;
3991 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3992 if (tmp_opt.sack_ok)
3993 req->tcpopt.sack = 1;
3994 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3998 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3999 __be32 laddr, __be16 lport,
4000 __be32 raddr, __be16 rport,
4001 u32 rcv_isn, u32 filter, u16 window,
4002 u32 rss_qid, u8 port_id)
4004 struct sk_buff *req_skb;
4005 struct fw_ofld_connection_wr *req;
4006 struct cpl_pass_accept_req *cpl = cplhdr(skb);
4009 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4012 req = __skb_put_zero(req_skb, sizeof(*req));
4013 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4014 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4015 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4016 req->le.filter = (__force __be32) filter;
4017 req->le.lport = lport;
4018 req->le.pport = rport;
4019 req->le.u.ipv4.lip = laddr;
4020 req->le.u.ipv4.pip = raddr;
4021 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4022 req->tcb.rcv_adv = htons(window);
4023 req->tcb.t_state_to_astid =
4024 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4025 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4026 FW_OFLD_CONNECTION_WR_ASTID_V(
4027 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4030 * We store the qid in opt2 which will be used by the firmware
4031 * to send us the wr response.
4033 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4036 * We initialize the MSS index in TCB to 0xF.
4037 * So that when driver sends cpl_pass_accept_rpl
4038 * TCB picks up the correct value. If this was 0
4039 * TP will ignore any value > 0 for MSS index.
4041 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4042 req->cookie = (uintptr_t)skb;
4044 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4045 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4047 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4055 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4056 * messages when a filter is being used instead of server to
4057 * redirect a syn packet. When packets hit filter they are redirected
4058 * to the offload queue and driver tries to establish the connection
4059 * using firmware work request.
4061 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4064 unsigned int filter;
4065 struct ethhdr *eh = NULL;
4066 struct vlan_ethhdr *vlan_eh = NULL;
4068 struct tcphdr *tcph;
4069 struct rss_header *rss = (void *)skb->data;
4070 struct cpl_rx_pkt *cpl = (void *)skb->data;
4071 struct cpl_pass_accept_req *req = (void *)(rss + 1);
4072 struct l2t_entry *e;
4073 struct dst_entry *dst;
4074 struct c4iw_ep *lep = NULL;
4076 struct port_info *pi;
4077 struct net_device *pdev;
4078 u16 rss_qid, eth_hdr_len;
4080 struct neighbour *neigh;
4082 /* Drop all non-SYN packets */
4083 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4087 * Drop all packets which did not hit the filter.
4088 * Unlikely to happen.
4090 if (!(rss->filter_hit && rss->filter_tid))
4094 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4096 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4098 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4100 pr_warn("%s connect request on invalid stid %d\n",
4105 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4107 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4110 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4113 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4116 pr_err("T%d Chip is not supported\n",
4117 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4121 if (eth_hdr_len == ETH_HLEN) {
4122 eh = (struct ethhdr *)(req + 1);
4123 iph = (struct iphdr *)(eh + 1);
4125 vlan_eh = (struct vlan_ethhdr *)(req + 1);
4126 iph = (struct iphdr *)(vlan_eh + 1);
4127 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4130 if (iph->version != 0x4)
4133 tcph = (struct tcphdr *)(iph + 1);
4134 skb_set_network_header(skb, (void *)iph - (void *)rss);
4135 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4138 pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4139 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4140 ntohs(tcph->source), iph->tos);
4142 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4143 iph->daddr, iph->saddr, tcph->dest,
4144 tcph->source, iph->tos);
4146 pr_err("%s - failed to find dst entry!\n", __func__);
4149 neigh = dst_neigh_lookup_skb(dst, skb);
4152 pr_err("%s - failed to allocate neigh!\n", __func__);
4156 if (neigh->dev->flags & IFF_LOOPBACK) {
4157 pdev = ip_dev_find(&init_net, iph->daddr);
4159 pr_err("%s - failed to find device!\n", __func__);
4162 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4164 pi = (struct port_info *)netdev_priv(pdev);
4167 pdev = get_real_dev(neigh->dev);
4168 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4170 pi = (struct port_info *)netdev_priv(pdev);
4172 neigh_release(neigh);
4174 pr_err("%s - failed to allocate l2t entry!\n",
4179 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4180 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4181 window = (__force u16) htons((__force u16)tcph->window);
4183 /* Calcuate filter portion for LE region. */
4184 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4185 dev->rdev.lldi.ports[0],
4189 * Synthesize the cpl_pass_accept_req. We have everything except the
4190 * TID. Once firmware sends a reply with TID we update the TID field
4191 * in cpl and pass it through the regular cpl_pass_accept_req path.
4193 build_cpl_pass_accept_req(skb, stid, iph->tos);
4194 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4195 tcph->source, ntohl(tcph->seq), filter, window,
4196 rss_qid, pi->port_id);
4197 cxgb4_l2t_release(e);
4202 c4iw_put_ep(&lep->com);
4207 * These are the real handlers that are called from a
4210 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4211 [CPL_ACT_ESTABLISH] = act_establish,
4212 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4213 [CPL_RX_DATA] = rx_data,
4214 [CPL_ABORT_RPL_RSS] = abort_rpl,
4215 [CPL_ABORT_RPL] = abort_rpl,
4216 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4217 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4218 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4219 [CPL_PASS_ESTABLISH] = pass_establish,
4220 [CPL_PEER_CLOSE] = peer_close,
4221 [CPL_ABORT_REQ_RSS] = peer_abort,
4222 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4223 [CPL_RDMA_TERMINATE] = terminate,
4224 [CPL_FW4_ACK] = fw4_ack,
4225 [CPL_GET_TCB_RPL] = read_tcb_rpl,
4226 [CPL_FW6_MSG] = deferred_fw6_msg,
4227 [CPL_RX_PKT] = rx_pkt,
4228 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4229 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4232 static void process_timeout(struct c4iw_ep *ep)
4234 struct c4iw_qp_attributes attrs;
4237 mutex_lock(&ep->com.mutex);
4238 pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4239 set_bit(TIMEDOUT, &ep->com.history);
4240 switch (ep->com.state) {
4242 connect_reply_upcall(ep, -ETIMEDOUT);
4251 if (ep->com.cm_id && ep->com.qp) {
4252 attrs.next_state = C4IW_QP_STATE_ERROR;
4253 c4iw_modify_qp(ep->com.qp->rhp,
4254 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4257 close_complete_upcall(ep, -ETIMEDOUT);
4263 * These states are expected if the ep timed out at the same
4264 * time as another thread was calling stop_ep_timer().
4265 * So we silently do nothing for these states.
4270 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4271 __func__, ep, ep->hwtid, ep->com.state);
4274 mutex_unlock(&ep->com.mutex);
4276 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4277 c4iw_put_ep(&ep->com);
4280 static void process_timedout_eps(void)
4284 spin_lock_irq(&timeout_lock);
4285 while (!list_empty(&timeout_list)) {
4286 struct list_head *tmp;
4288 tmp = timeout_list.next;
4292 spin_unlock_irq(&timeout_lock);
4293 ep = list_entry(tmp, struct c4iw_ep, entry);
4294 process_timeout(ep);
4295 spin_lock_irq(&timeout_lock);
4297 spin_unlock_irq(&timeout_lock);
4300 static void process_work(struct work_struct *work)
4302 struct sk_buff *skb = NULL;
4303 struct c4iw_dev *dev;
4304 struct cpl_act_establish *rpl;
4305 unsigned int opcode;
4308 process_timedout_eps();
4309 while ((skb = skb_dequeue(&rxq))) {
4311 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4312 opcode = rpl->ot.opcode;
4314 if (opcode >= ARRAY_SIZE(work_handlers) ||
4315 !work_handlers[opcode]) {
4316 pr_err("No handler for opcode 0x%x.\n", opcode);
4319 ret = work_handlers[opcode](dev, skb);
4323 process_timedout_eps();
4327 static DECLARE_WORK(skb_work, process_work);
4329 static void ep_timeout(struct timer_list *t)
4331 struct c4iw_ep *ep = from_timer(ep, t, timer);
4334 spin_lock(&timeout_lock);
4335 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4337 * Only insert if it is not already on the list.
4339 if (!ep->entry.next) {
4340 list_add_tail(&ep->entry, &timeout_list);
4344 spin_unlock(&timeout_lock);
4346 queue_work(workq, &skb_work);
4350 * All the CM events are handled on a work queue to have a safe context.
4352 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4356 * Save dev in the skb->cb area.
4358 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4361 * Queue the skb and schedule the worker thread.
4363 skb_queue_tail(&rxq, skb);
4364 queue_work(workq, &skb_work);
4368 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4370 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4372 if (rpl->status != CPL_ERR_NONE) {
4373 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4374 rpl->status, GET_TID(rpl));
4380 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4382 struct cpl_fw6_msg *rpl = cplhdr(skb);
4383 struct c4iw_wr_wait *wr_waitp;
4386 pr_debug("type %u\n", rpl->type);
4388 switch (rpl->type) {
4389 case FW6_TYPE_WR_RPL:
4390 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4391 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4392 pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4394 c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4398 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4402 pr_err("%s unexpected fw6 msg type %u\n",
4403 __func__, rpl->type);
4410 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4412 struct cpl_abort_req_rss *req = cplhdr(skb);
4414 unsigned int tid = GET_TID(req);
4416 ep = get_ep_from_tid(dev, tid);
4417 /* This EP will be dereferenced in peer_abort() */
4419 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4423 if (cxgb_is_neg_adv(req->status)) {
4424 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4425 ep->hwtid, req->status,
4426 neg_adv_str(req->status));
4429 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4431 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4438 * Most upcalls from the T4 Core go to sched() to
4439 * schedule the processing on a work queue.
4441 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4442 [CPL_ACT_ESTABLISH] = sched,
4443 [CPL_ACT_OPEN_RPL] = sched,
4444 [CPL_RX_DATA] = sched,
4445 [CPL_ABORT_RPL_RSS] = sched,
4446 [CPL_ABORT_RPL] = sched,
4447 [CPL_PASS_OPEN_RPL] = sched,
4448 [CPL_CLOSE_LISTSRV_RPL] = sched,
4449 [CPL_PASS_ACCEPT_REQ] = sched,
4450 [CPL_PASS_ESTABLISH] = sched,
4451 [CPL_PEER_CLOSE] = sched,
4452 [CPL_CLOSE_CON_RPL] = sched,
4453 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4454 [CPL_RDMA_TERMINATE] = sched,
4455 [CPL_FW4_ACK] = sched,
4456 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4457 [CPL_GET_TCB_RPL] = sched,
4458 [CPL_FW6_MSG] = fw6_msg,
4459 [CPL_RX_PKT] = sched
4462 int __init c4iw_cm_init(void)
4464 skb_queue_head_init(&rxq);
4466 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4473 void c4iw_cm_term(void)
4475 WARN_ON(!list_empty(&timeout_list));
4476 destroy_workqueue(workq);