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);
1226 pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1227 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1229 mutex_lock(&ep->com.mutex);
1230 dst_confirm(ep->dst);
1232 /* setup the hwtid for this connection */
1234 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1237 ep->snd_seq = be32_to_cpu(req->snd_isn);
1238 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1239 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1241 set_emss(ep, tcp_opt);
1243 /* dealloc the atid */
1244 xa_erase_irq(&ep->com.dev->atids, atid);
1245 cxgb4_free_atid(t, atid);
1246 set_bit(ACT_ESTAB, &ep->com.history);
1248 /* start MPA negotiation */
1249 ret = send_flowc(ep);
1252 if (ep->retry_with_mpa_v1)
1253 ret = send_mpa_req(ep, skb, 1);
1255 ret = send_mpa_req(ep, skb, mpa_rev);
1258 mutex_unlock(&ep->com.mutex);
1261 mutex_unlock(&ep->com.mutex);
1262 connect_reply_upcall(ep, -ENOMEM);
1263 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1267 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1269 struct iw_cm_event event;
1271 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1272 memset(&event, 0, sizeof(event));
1273 event.event = IW_CM_EVENT_CLOSE;
1274 event.status = status;
1275 if (ep->com.cm_id) {
1276 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1277 ep, ep->com.cm_id, ep->hwtid);
1278 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1279 deref_cm_id(&ep->com);
1280 set_bit(CLOSE_UPCALL, &ep->com.history);
1284 static void peer_close_upcall(struct c4iw_ep *ep)
1286 struct iw_cm_event event;
1288 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1289 memset(&event, 0, sizeof(event));
1290 event.event = IW_CM_EVENT_DISCONNECT;
1291 if (ep->com.cm_id) {
1292 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1293 ep, ep->com.cm_id, ep->hwtid);
1294 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1295 set_bit(DISCONN_UPCALL, &ep->com.history);
1299 static void peer_abort_upcall(struct c4iw_ep *ep)
1301 struct iw_cm_event event;
1303 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1304 memset(&event, 0, sizeof(event));
1305 event.event = IW_CM_EVENT_CLOSE;
1306 event.status = -ECONNRESET;
1307 if (ep->com.cm_id) {
1308 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1309 ep->com.cm_id, ep->hwtid);
1310 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1311 deref_cm_id(&ep->com);
1312 set_bit(ABORT_UPCALL, &ep->com.history);
1316 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1318 struct iw_cm_event event;
1320 pr_debug("ep %p tid %u status %d\n",
1321 ep, ep->hwtid, status);
1322 memset(&event, 0, sizeof(event));
1323 event.event = IW_CM_EVENT_CONNECT_REPLY;
1324 event.status = status;
1325 memcpy(&event.local_addr, &ep->com.local_addr,
1326 sizeof(ep->com.local_addr));
1327 memcpy(&event.remote_addr, &ep->com.remote_addr,
1328 sizeof(ep->com.remote_addr));
1330 if ((status == 0) || (status == -ECONNREFUSED)) {
1331 if (!ep->tried_with_mpa_v1) {
1332 /* this means MPA_v2 is used */
1333 event.ord = ep->ird;
1334 event.ird = ep->ord;
1335 event.private_data_len = ep->plen -
1336 sizeof(struct mpa_v2_conn_params);
1337 event.private_data = ep->mpa_pkt +
1338 sizeof(struct mpa_message) +
1339 sizeof(struct mpa_v2_conn_params);
1341 /* this means MPA_v1 is used */
1342 event.ord = cur_max_read_depth(ep->com.dev);
1343 event.ird = cur_max_read_depth(ep->com.dev);
1344 event.private_data_len = ep->plen;
1345 event.private_data = ep->mpa_pkt +
1346 sizeof(struct mpa_message);
1350 pr_debug("ep %p tid %u status %d\n", ep,
1352 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1353 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1356 deref_cm_id(&ep->com);
1359 static int connect_request_upcall(struct c4iw_ep *ep)
1361 struct iw_cm_event event;
1364 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1365 memset(&event, 0, sizeof(event));
1366 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1367 memcpy(&event.local_addr, &ep->com.local_addr,
1368 sizeof(ep->com.local_addr));
1369 memcpy(&event.remote_addr, &ep->com.remote_addr,
1370 sizeof(ep->com.remote_addr));
1371 event.provider_data = ep;
1372 if (!ep->tried_with_mpa_v1) {
1373 /* this means MPA_v2 is used */
1374 event.ord = ep->ord;
1375 event.ird = ep->ird;
1376 event.private_data_len = ep->plen -
1377 sizeof(struct mpa_v2_conn_params);
1378 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1379 sizeof(struct mpa_v2_conn_params);
1381 /* this means MPA_v1 is used. Send max supported */
1382 event.ord = cur_max_read_depth(ep->com.dev);
1383 event.ird = cur_max_read_depth(ep->com.dev);
1384 event.private_data_len = ep->plen;
1385 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1387 c4iw_get_ep(&ep->com);
1388 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1391 c4iw_put_ep(&ep->com);
1392 set_bit(CONNREQ_UPCALL, &ep->com.history);
1393 c4iw_put_ep(&ep->parent_ep->com);
1397 static void established_upcall(struct c4iw_ep *ep)
1399 struct iw_cm_event event;
1401 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1402 memset(&event, 0, sizeof(event));
1403 event.event = IW_CM_EVENT_ESTABLISHED;
1404 event.ird = ep->ord;
1405 event.ord = ep->ird;
1406 if (ep->com.cm_id) {
1407 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1408 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1409 set_bit(ESTAB_UPCALL, &ep->com.history);
1413 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1415 struct sk_buff *skb;
1416 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1419 pr_debug("ep %p tid %u credits %u\n",
1420 ep, ep->hwtid, credits);
1421 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1423 pr_err("update_rx_credits - cannot alloc skb!\n");
1428 * If we couldn't specify the entire rcv window at connection setup
1429 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1430 * then add the overage in to the credits returned.
1432 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1433 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1435 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1436 RX_DACK_MODE_V(dack_mode);
1438 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1441 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1445 #define RELAXED_IRD_NEGOTIATION 1
1448 * process_mpa_reply - process streaming mode MPA reply
1452 * 0 upon success indicating a connect request was delivered to the ULP
1453 * or the mpa request is incomplete but valid so far.
1455 * 1 if a failure requires the caller to close the connection.
1457 * 2 if a failure requires the caller to abort the connection.
1459 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1461 struct mpa_message *mpa;
1462 struct mpa_v2_conn_params *mpa_v2_params;
1464 u16 resp_ird, resp_ord;
1465 u8 rtr_mismatch = 0, insuff_ird = 0;
1466 struct c4iw_qp_attributes attrs;
1467 enum c4iw_qp_attr_mask mask;
1471 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1474 * If we get more than the supported amount of private data
1475 * then we must fail this connection.
1477 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1479 goto err_stop_timer;
1483 * copy the new data into our accumulation buffer.
1485 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1487 ep->mpa_pkt_len += skb->len;
1490 * if we don't even have the mpa message, then bail.
1492 if (ep->mpa_pkt_len < sizeof(*mpa))
1494 mpa = (struct mpa_message *) ep->mpa_pkt;
1496 /* Validate MPA header. */
1497 if (mpa->revision > mpa_rev) {
1498 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1499 __func__, mpa_rev, mpa->revision);
1501 goto err_stop_timer;
1503 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1505 goto err_stop_timer;
1508 plen = ntohs(mpa->private_data_size);
1511 * Fail if there's too much private data.
1513 if (plen > MPA_MAX_PRIVATE_DATA) {
1515 goto err_stop_timer;
1519 * If plen does not account for pkt size
1521 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1523 goto err_stop_timer;
1526 ep->plen = (u8) plen;
1529 * If we don't have all the pdata yet, then bail.
1530 * We'll continue process when more data arrives.
1532 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1535 if (mpa->flags & MPA_REJECT) {
1536 err = -ECONNREFUSED;
1537 goto err_stop_timer;
1541 * Stop mpa timer. If it expired, then
1542 * we ignore the MPA reply. process_timeout()
1543 * will abort the connection.
1545 if (stop_ep_timer(ep))
1549 * If we get here we have accumulated the entire mpa
1550 * start reply message including private data. And
1551 * the MPA header is valid.
1553 __state_set(&ep->com, FPDU_MODE);
1554 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1555 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1556 ep->mpa_attr.version = mpa->revision;
1557 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1559 if (mpa->revision == 2) {
1560 ep->mpa_attr.enhanced_rdma_conn =
1561 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1562 if (ep->mpa_attr.enhanced_rdma_conn) {
1563 mpa_v2_params = (struct mpa_v2_conn_params *)
1564 (ep->mpa_pkt + sizeof(*mpa));
1565 resp_ird = ntohs(mpa_v2_params->ird) &
1566 MPA_V2_IRD_ORD_MASK;
1567 resp_ord = ntohs(mpa_v2_params->ord) &
1568 MPA_V2_IRD_ORD_MASK;
1569 pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1570 resp_ird, resp_ord, ep->ird, ep->ord);
1573 * This is a double-check. Ideally, below checks are
1574 * not required since ird/ord stuff has been taken
1575 * care of in c4iw_accept_cr
1577 if (ep->ird < resp_ord) {
1578 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1579 ep->com.dev->rdev.lldi.max_ordird_qp)
1583 } else if (ep->ird > resp_ord) {
1586 if (ep->ord > resp_ird) {
1587 if (RELAXED_IRD_NEGOTIATION)
1598 if (ntohs(mpa_v2_params->ird) &
1599 MPA_V2_PEER2PEER_MODEL) {
1600 if (ntohs(mpa_v2_params->ord) &
1601 MPA_V2_RDMA_WRITE_RTR)
1602 ep->mpa_attr.p2p_type =
1603 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1604 else if (ntohs(mpa_v2_params->ord) &
1605 MPA_V2_RDMA_READ_RTR)
1606 ep->mpa_attr.p2p_type =
1607 FW_RI_INIT_P2PTYPE_READ_REQ;
1610 } else if (mpa->revision == 1)
1612 ep->mpa_attr.p2p_type = p2p_type;
1614 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1615 ep->mpa_attr.crc_enabled,
1616 ep->mpa_attr.recv_marker_enabled,
1617 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1618 ep->mpa_attr.p2p_type, p2p_type);
1621 * If responder's RTR does not match with that of initiator, assign
1622 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1623 * generated when moving QP to RTS state.
1624 * A TERM message will be sent after QP has moved to RTS state
1626 if ((ep->mpa_attr.version == 2) && peer2peer &&
1627 (ep->mpa_attr.p2p_type != p2p_type)) {
1628 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1632 attrs.mpa_attr = ep->mpa_attr;
1633 attrs.max_ird = ep->ird;
1634 attrs.max_ord = ep->ord;
1635 attrs.llp_stream_handle = ep;
1636 attrs.next_state = C4IW_QP_STATE_RTS;
1638 mask = C4IW_QP_ATTR_NEXT_STATE |
1639 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1640 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1642 /* bind QP and TID with INIT_WR */
1643 err = c4iw_modify_qp(ep->com.qp->rhp,
1644 ep->com.qp, mask, &attrs, 1);
1649 * If responder's RTR requirement did not match with what initiator
1650 * supports, generate TERM message
1653 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1654 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1655 attrs.ecode = MPA_NOMATCH_RTR;
1656 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1657 attrs.send_term = 1;
1658 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1659 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1666 * Generate TERM if initiator IRD is not sufficient for responder
1667 * provided ORD. Currently, we do the same behaviour even when
1668 * responder provided IRD is also not sufficient as regards to
1672 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1673 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1674 attrs.ecode = MPA_INSUFF_IRD;
1675 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1676 attrs.send_term = 1;
1677 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1678 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1689 connect_reply_upcall(ep, err);
1694 * process_mpa_request - process streaming mode MPA request
1698 * 0 upon success indicating a connect request was delivered to the ULP
1699 * or the mpa request is incomplete but valid so far.
1701 * 1 if a failure requires the caller to close the connection.
1703 * 2 if a failure requires the caller to abort the connection.
1705 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1707 struct mpa_message *mpa;
1708 struct mpa_v2_conn_params *mpa_v2_params;
1711 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1714 * If we get more than the supported amount of private data
1715 * then we must fail this connection.
1717 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1718 goto err_stop_timer;
1720 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1723 * Copy the new data into our accumulation buffer.
1725 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1727 ep->mpa_pkt_len += skb->len;
1730 * If we don't even have the mpa message, then bail.
1731 * We'll continue process when more data arrives.
1733 if (ep->mpa_pkt_len < sizeof(*mpa))
1736 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1737 mpa = (struct mpa_message *) ep->mpa_pkt;
1740 * Validate MPA Header.
1742 if (mpa->revision > mpa_rev) {
1743 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1744 __func__, mpa_rev, mpa->revision);
1745 goto err_stop_timer;
1748 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1749 goto err_stop_timer;
1751 plen = ntohs(mpa->private_data_size);
1754 * Fail if there's too much private data.
1756 if (plen > MPA_MAX_PRIVATE_DATA)
1757 goto err_stop_timer;
1760 * If plen does not account for pkt size
1762 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1763 goto err_stop_timer;
1764 ep->plen = (u8) plen;
1767 * If we don't have all the pdata yet, then bail.
1769 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1773 * If we get here we have accumulated the entire mpa
1774 * start reply message including private data.
1776 ep->mpa_attr.initiator = 0;
1777 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1778 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1779 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1780 ep->mpa_attr.version = mpa->revision;
1781 if (mpa->revision == 1)
1782 ep->tried_with_mpa_v1 = 1;
1783 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1785 if (mpa->revision == 2) {
1786 ep->mpa_attr.enhanced_rdma_conn =
1787 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1788 if (ep->mpa_attr.enhanced_rdma_conn) {
1789 mpa_v2_params = (struct mpa_v2_conn_params *)
1790 (ep->mpa_pkt + sizeof(*mpa));
1791 ep->ird = ntohs(mpa_v2_params->ird) &
1792 MPA_V2_IRD_ORD_MASK;
1793 ep->ird = min_t(u32, ep->ird,
1794 cur_max_read_depth(ep->com.dev));
1795 ep->ord = ntohs(mpa_v2_params->ord) &
1796 MPA_V2_IRD_ORD_MASK;
1797 ep->ord = min_t(u32, ep->ord,
1798 cur_max_read_depth(ep->com.dev));
1799 pr_debug("initiator ird %u ord %u\n",
1801 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1803 if (ntohs(mpa_v2_params->ord) &
1804 MPA_V2_RDMA_WRITE_RTR)
1805 ep->mpa_attr.p2p_type =
1806 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1807 else if (ntohs(mpa_v2_params->ord) &
1808 MPA_V2_RDMA_READ_RTR)
1809 ep->mpa_attr.p2p_type =
1810 FW_RI_INIT_P2PTYPE_READ_REQ;
1813 } else if (mpa->revision == 1)
1815 ep->mpa_attr.p2p_type = p2p_type;
1817 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1818 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1819 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1820 ep->mpa_attr.p2p_type);
1822 __state_set(&ep->com, MPA_REQ_RCVD);
1825 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1826 if (ep->parent_ep->com.state != DEAD) {
1827 if (connect_request_upcall(ep))
1828 goto err_unlock_parent;
1830 goto err_unlock_parent;
1832 mutex_unlock(&ep->parent_ep->com.mutex);
1836 mutex_unlock(&ep->parent_ep->com.mutex);
1839 (void)stop_ep_timer(ep);
1844 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1847 struct cpl_rx_data *hdr = cplhdr(skb);
1848 unsigned int dlen = ntohs(hdr->len);
1849 unsigned int tid = GET_TID(hdr);
1850 __u8 status = hdr->status;
1853 ep = get_ep_from_tid(dev, tid);
1856 pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1857 skb_pull(skb, sizeof(*hdr));
1858 skb_trim(skb, dlen);
1859 mutex_lock(&ep->com.mutex);
1861 switch (ep->com.state) {
1863 update_rx_credits(ep, dlen);
1864 ep->rcv_seq += dlen;
1865 disconnect = process_mpa_reply(ep, skb);
1868 update_rx_credits(ep, dlen);
1869 ep->rcv_seq += dlen;
1870 disconnect = process_mpa_request(ep, skb);
1873 struct c4iw_qp_attributes attrs;
1875 update_rx_credits(ep, dlen);
1877 pr_err("%s Unexpected streaming data." \
1878 " qpid %u ep %p state %d tid %u status %d\n",
1879 __func__, ep->com.qp->wq.sq.qid, ep,
1880 ep->com.state, ep->hwtid, status);
1881 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1882 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1883 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1890 mutex_unlock(&ep->com.mutex);
1892 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1893 c4iw_put_ep(&ep->com);
1897 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1899 enum chip_type adapter_type;
1901 adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1904 * If this TCB had a srq buffer cached, then we must complete
1905 * it. For user mode, that means saving the srqidx in the
1906 * user/kernel status page for this qp. For kernel mode, just
1907 * synthesize the CQE now.
1909 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1910 if (ep->com.qp->ibqp.uobject)
1911 t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1913 c4iw_flush_srqidx(ep->com.qp, srqidx);
1917 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1921 struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1923 unsigned int tid = GET_TID(rpl);
1925 ep = get_ep_from_tid(dev, tid);
1927 pr_warn("Abort rpl to freed endpoint\n");
1931 if (ep->com.qp && ep->com.qp->srq) {
1932 srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1933 complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1936 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1937 mutex_lock(&ep->com.mutex);
1938 switch (ep->com.state) {
1940 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1941 __state_set(&ep->com, DEAD);
1945 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1948 mutex_unlock(&ep->com.mutex);
1951 close_complete_upcall(ep, -ECONNRESET);
1952 release_ep_resources(ep);
1954 c4iw_put_ep(&ep->com);
1958 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1960 struct sk_buff *skb;
1961 struct fw_ofld_connection_wr *req;
1962 unsigned int mtu_idx;
1964 struct sockaddr_in *sin;
1967 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1968 req = __skb_put_zero(skb, sizeof(*req));
1969 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1970 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1971 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1972 ep->com.dev->rdev.lldi.ports[0],
1974 sin = (struct sockaddr_in *)&ep->com.local_addr;
1975 req->le.lport = sin->sin_port;
1976 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1977 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1978 req->le.pport = sin->sin_port;
1979 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1980 req->tcb.t_state_to_astid =
1981 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1982 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1983 req->tcb.cplrxdataack_cplpassacceptrpl =
1984 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1985 req->tcb.tx_max = (__force __be32) jiffies;
1986 req->tcb.rcv_adv = htons(1);
1987 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1988 enable_tcp_timestamps,
1989 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1990 wscale = cxgb_compute_wscale(rcv_win);
1993 * Specify the largest window that will fit in opt0. The
1994 * remainder will be specified in the rx_data_ack.
1996 win = ep->rcv_win >> 10;
1997 if (win > RCV_BUFSIZ_M)
2000 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2001 (nocong ? NO_CONG_F : 0) |
2004 WND_SCALE_V(wscale) |
2005 MSS_IDX_V(mtu_idx) |
2006 L2T_IDX_V(ep->l2t->idx) |
2007 TX_CHAN_V(ep->tx_chan) |
2008 SMAC_SEL_V(ep->smac_idx) |
2009 DSCP_V(ep->tos >> 2) |
2010 ULP_MODE_V(ULP_MODE_TCPDDP) |
2012 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2013 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2015 CCTRL_ECN_V(enable_ecn) |
2016 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2017 if (enable_tcp_timestamps)
2018 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2019 if (enable_tcp_sack)
2020 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2021 if (wscale && enable_tcp_window_scaling)
2022 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2023 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2024 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2025 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2026 set_bit(ACT_OFLD_CONN, &ep->com.history);
2027 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2031 * Some of the error codes above implicitly indicate that there is no TID
2032 * allocated with the result of an ACT_OPEN. We use this predicate to make
2035 static inline int act_open_has_tid(int status)
2037 return (status != CPL_ERR_TCAM_PARITY &&
2038 status != CPL_ERR_TCAM_MISS &&
2039 status != CPL_ERR_TCAM_FULL &&
2040 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2041 status != CPL_ERR_CONN_EXIST);
2044 static char *neg_adv_str(unsigned int status)
2047 case CPL_ERR_RTX_NEG_ADVICE:
2048 return "Retransmit timeout";
2049 case CPL_ERR_PERSIST_NEG_ADVICE:
2050 return "Persist timeout";
2051 case CPL_ERR_KEEPALV_NEG_ADVICE:
2052 return "Keepalive timeout";
2058 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2060 ep->snd_win = snd_win;
2061 ep->rcv_win = rcv_win;
2062 pr_debug("snd_win %d rcv_win %d\n",
2063 ep->snd_win, ep->rcv_win);
2066 #define ACT_OPEN_RETRY_COUNT 2
2068 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2069 struct dst_entry *dst, struct c4iw_dev *cdev,
2070 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2072 struct neighbour *n;
2074 struct net_device *pdev;
2076 n = dst_neigh_lookup(dst, peer_ip);
2082 if (n->dev->flags & IFF_LOOPBACK) {
2084 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2085 else if (IS_ENABLED(CONFIG_IPV6))
2086 for_each_netdev(&init_net, pdev) {
2087 if (ipv6_chk_addr(&init_net,
2088 (struct in6_addr *)peer_ip,
2099 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2100 n, pdev, rt_tos2priority(tos));
2105 ep->mtu = pdev->mtu;
2106 ep->tx_chan = cxgb4_port_chan(pdev);
2107 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2108 step = cdev->rdev.lldi.ntxq /
2109 cdev->rdev.lldi.nchan;
2110 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2111 step = cdev->rdev.lldi.nrxq /
2112 cdev->rdev.lldi.nchan;
2113 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2114 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2115 cxgb4_port_idx(pdev) * step];
2116 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2119 pdev = get_real_dev(n->dev);
2120 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2121 n, pdev, rt_tos2priority(tos));
2124 ep->mtu = dst_mtu(dst);
2125 ep->tx_chan = cxgb4_port_chan(pdev);
2126 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2127 step = cdev->rdev.lldi.ntxq /
2128 cdev->rdev.lldi.nchan;
2129 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2130 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2131 step = cdev->rdev.lldi.nrxq /
2132 cdev->rdev.lldi.nchan;
2133 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2134 cxgb4_port_idx(pdev) * step];
2135 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2138 ep->retry_with_mpa_v1 = 0;
2139 ep->tried_with_mpa_v1 = 0;
2151 static int c4iw_reconnect(struct c4iw_ep *ep)
2155 struct sockaddr_in *laddr = (struct sockaddr_in *)
2156 &ep->com.cm_id->m_local_addr;
2157 struct sockaddr_in *raddr = (struct sockaddr_in *)
2158 &ep->com.cm_id->m_remote_addr;
2159 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2160 &ep->com.cm_id->m_local_addr;
2161 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2162 &ep->com.cm_id->m_remote_addr;
2166 pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2167 c4iw_init_wr_wait(ep->com.wr_waitp);
2169 /* When MPA revision is different on nodes, the node with MPA_rev=2
2170 * tries to reconnect with MPA_rev 1 for the same EP through
2171 * c4iw_reconnect(), where the same EP is assigned with new tid for
2172 * further connection establishment. As we are using the same EP pointer
2173 * for reconnect, few skbs are used during the previous c4iw_connect(),
2174 * which leaves the EP with inadequate skbs for further
2175 * c4iw_reconnect(), Further causing a crash due to an empty
2176 * skb_list() during peer_abort(). Allocate skbs which is already used.
2178 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2179 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2185 * Allocate an active TID to initiate a TCP connection.
2187 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2188 if (ep->atid == -1) {
2189 pr_err("%s - cannot alloc atid\n", __func__);
2193 err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2198 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2199 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2200 laddr->sin_addr.s_addr,
2201 raddr->sin_addr.s_addr,
2203 raddr->sin_port, ep->com.cm_id->tos);
2205 ra = (__u8 *)&raddr->sin_addr;
2207 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2209 laddr6->sin6_addr.s6_addr,
2210 raddr6->sin6_addr.s6_addr,
2214 raddr6->sin6_scope_id);
2216 ra = (__u8 *)&raddr6->sin6_addr;
2219 pr_err("%s - cannot find route\n", __func__);
2220 err = -EHOSTUNREACH;
2223 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2224 ep->com.dev->rdev.lldi.adapter_type,
2225 ep->com.cm_id->tos);
2227 pr_err("%s - cannot alloc l2e\n", __func__);
2231 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2232 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2235 state_set(&ep->com, CONNECTING);
2236 ep->tos = ep->com.cm_id->tos;
2238 /* send connect request to rnic */
2239 err = send_connect(ep);
2243 cxgb4_l2t_release(ep->l2t);
2245 dst_release(ep->dst);
2247 xa_erase_irq(&ep->com.dev->atids, ep->atid);
2249 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2252 * remember to send notification to upper layer.
2253 * We are in here so the upper layer is not aware that this is
2254 * re-connect attempt and so, upper layer is still waiting for
2255 * response of 1st connect request.
2257 connect_reply_upcall(ep, -ECONNRESET);
2259 c4iw_put_ep(&ep->com);
2264 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2267 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2268 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2269 ntohl(rpl->atid_status)));
2270 struct tid_info *t = dev->rdev.lldi.tids;
2271 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2272 struct sockaddr_in *la;
2273 struct sockaddr_in *ra;
2274 struct sockaddr_in6 *la6;
2275 struct sockaddr_in6 *ra6;
2278 ep = lookup_atid(t, atid);
2279 la = (struct sockaddr_in *)&ep->com.local_addr;
2280 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2281 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2282 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2284 pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2285 status, status2errno(status));
2287 if (cxgb_is_neg_adv(status)) {
2288 pr_debug("Connection problems for atid %u status %u (%s)\n",
2289 atid, status, neg_adv_str(status));
2290 ep->stats.connect_neg_adv++;
2291 mutex_lock(&dev->rdev.stats.lock);
2292 dev->rdev.stats.neg_adv++;
2293 mutex_unlock(&dev->rdev.stats.lock);
2297 set_bit(ACT_OPEN_RPL, &ep->com.history);
2300 * Log interesting failures.
2303 case CPL_ERR_CONN_RESET:
2304 case CPL_ERR_CONN_TIMEDOUT:
2306 case CPL_ERR_TCAM_FULL:
2307 mutex_lock(&dev->rdev.stats.lock);
2308 dev->rdev.stats.tcam_full++;
2309 mutex_unlock(&dev->rdev.stats.lock);
2310 if (ep->com.local_addr.ss_family == AF_INET &&
2311 dev->rdev.lldi.enable_fw_ofld_conn) {
2312 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2313 ntohl(rpl->atid_status))));
2319 case CPL_ERR_CONN_EXIST:
2320 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2321 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2322 if (ep->com.remote_addr.ss_family == AF_INET6) {
2323 struct sockaddr_in6 *sin6 =
2324 (struct sockaddr_in6 *)
2325 &ep->com.local_addr;
2327 ep->com.dev->rdev.lldi.ports[0],
2329 &sin6->sin6_addr.s6_addr, 1);
2331 xa_erase_irq(&ep->com.dev->atids, atid);
2332 cxgb4_free_atid(t, atid);
2333 dst_release(ep->dst);
2334 cxgb4_l2t_release(ep->l2t);
2340 if (ep->com.local_addr.ss_family == AF_INET) {
2341 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2342 atid, status, status2errno(status),
2343 &la->sin_addr.s_addr, ntohs(la->sin_port),
2344 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2346 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2347 atid, status, status2errno(status),
2348 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2349 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2355 connect_reply_upcall(ep, status2errno(status));
2356 state_set(&ep->com, DEAD);
2358 if (ep->com.remote_addr.ss_family == AF_INET6) {
2359 struct sockaddr_in6 *sin6 =
2360 (struct sockaddr_in6 *)&ep->com.local_addr;
2361 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2362 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2364 if (status && act_open_has_tid(status))
2365 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2366 ep->com.local_addr.ss_family);
2368 xa_erase_irq(&ep->com.dev->atids, atid);
2369 cxgb4_free_atid(t, atid);
2370 dst_release(ep->dst);
2371 cxgb4_l2t_release(ep->l2t);
2372 c4iw_put_ep(&ep->com);
2377 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2379 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2380 unsigned int stid = GET_TID(rpl);
2381 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2384 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2387 pr_debug("ep %p status %d error %d\n", ep,
2388 rpl->status, status2errno(rpl->status));
2389 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2390 c4iw_put_ep(&ep->com);
2395 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2397 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2398 unsigned int stid = GET_TID(rpl);
2399 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2402 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2405 pr_debug("ep %p\n", ep);
2406 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2407 c4iw_put_ep(&ep->com);
2412 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2413 struct cpl_pass_accept_req *req)
2415 struct cpl_pass_accept_rpl *rpl;
2416 unsigned int mtu_idx;
2420 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2422 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2424 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2425 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2426 enable_tcp_timestamps && req->tcpopt.tstamp,
2427 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2428 wscale = cxgb_compute_wscale(rcv_win);
2431 * Specify the largest window that will fit in opt0. The
2432 * remainder will be specified in the rx_data_ack.
2434 win = ep->rcv_win >> 10;
2435 if (win > RCV_BUFSIZ_M)
2437 opt0 = (nocong ? NO_CONG_F : 0) |
2440 WND_SCALE_V(wscale) |
2441 MSS_IDX_V(mtu_idx) |
2442 L2T_IDX_V(ep->l2t->idx) |
2443 TX_CHAN_V(ep->tx_chan) |
2444 SMAC_SEL_V(ep->smac_idx) |
2445 DSCP_V(ep->tos >> 2) |
2446 ULP_MODE_V(ULP_MODE_TCPDDP) |
2448 opt2 = RX_CHANNEL_V(0) |
2449 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2451 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2452 opt2 |= TSTAMPS_EN_F;
2453 if (enable_tcp_sack && req->tcpopt.sack)
2455 if (wscale && enable_tcp_window_scaling)
2456 opt2 |= WND_SCALE_EN_F;
2458 const struct tcphdr *tcph;
2459 u32 hlen = ntohl(req->hdr_len);
2461 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2462 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2465 tcph = (const void *)(req + 1) +
2466 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2467 if (tcph->ece && tcph->cwr)
2468 opt2 |= CCTRL_ECN_V(1);
2471 if (!is_t4(adapter_type)) {
2472 u32 isn = (get_random_u32() & ~7UL) - 1;
2474 skb = get_skb(skb, roundup(sizeof(*rpl5), 16), GFP_KERNEL);
2475 rpl5 = __skb_put_zero(skb, roundup(sizeof(*rpl5), 16));
2477 INIT_TP_WR_CPL(rpl5, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2478 opt2 |= T5_OPT_2_VALID_F;
2479 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2483 rpl5->iss = cpu_to_be32(isn);
2484 pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2486 skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2487 rpl = __skb_put_zero(skb, sizeof(*rpl));
2488 INIT_TP_WR_CPL(rpl, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2491 rpl->opt0 = cpu_to_be64(opt0);
2492 rpl->opt2 = cpu_to_be32(opt2);
2493 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2494 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2496 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2499 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2501 pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2502 skb_trim(skb, sizeof(struct cpl_tid_release));
2503 release_tid(&dev->rdev, hwtid, skb);
2507 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2509 struct c4iw_ep *child_ep = NULL, *parent_ep;
2510 struct cpl_pass_accept_req *req = cplhdr(skb);
2511 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2512 struct tid_info *t = dev->rdev.lldi.tids;
2513 unsigned int hwtid = GET_TID(req);
2514 struct dst_entry *dst;
2515 __u8 local_ip[16], peer_ip[16];
2516 __be16 local_port, peer_port;
2517 struct sockaddr_in6 *sin6;
2519 u16 peer_mss = ntohs(req->tcpopt.mss);
2521 unsigned short hdrs;
2524 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2526 pr_err("%s connect request on invalid stid %d\n",
2531 if (state_read(&parent_ep->com) != LISTEN) {
2532 pr_err("%s - listening ep not in LISTEN\n", __func__);
2536 if (parent_ep->com.cm_id->tos_set)
2537 tos = parent_ep->com.cm_id->tos;
2539 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2541 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2542 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2544 /* Find output route */
2546 pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2548 local_ip, peer_ip, ntohs(local_port),
2549 ntohs(peer_port), peer_mss);
2550 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2551 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2552 local_port, peer_port, tos);
2554 pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 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_route6(&dev->rdev.lldi, get_real_dev,
2559 local_ip, peer_ip, local_port, peer_port,
2561 ((struct sockaddr_in6 *)
2562 &parent_ep->com.local_addr)->sin6_scope_id);
2565 pr_err("%s - failed to find dst entry!\n", __func__);
2569 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2571 pr_err("%s - failed to allocate ep entry!\n", __func__);
2576 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2577 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2579 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2585 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2586 sizeof(struct tcphdr) +
2587 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2588 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2589 child_ep->mtu = peer_mss + hdrs;
2591 skb_queue_head_init(&child_ep->com.ep_skb_list);
2592 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2595 state_set(&child_ep->com, CONNECTING);
2596 child_ep->com.dev = dev;
2597 child_ep->com.cm_id = NULL;
2600 struct sockaddr_in *sin = (struct sockaddr_in *)
2601 &child_ep->com.local_addr;
2603 sin->sin_family = AF_INET;
2604 sin->sin_port = local_port;
2605 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2607 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2608 sin->sin_family = AF_INET;
2609 sin->sin_port = ((struct sockaddr_in *)
2610 &parent_ep->com.local_addr)->sin_port;
2611 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2613 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2614 sin->sin_family = AF_INET;
2615 sin->sin_port = peer_port;
2616 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2618 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2619 sin6->sin6_family = PF_INET6;
2620 sin6->sin6_port = local_port;
2621 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2623 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2624 sin6->sin6_family = PF_INET6;
2625 sin6->sin6_port = ((struct sockaddr_in6 *)
2626 &parent_ep->com.local_addr)->sin6_port;
2627 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2629 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2630 sin6->sin6_family = PF_INET6;
2631 sin6->sin6_port = peer_port;
2632 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2635 c4iw_get_ep(&parent_ep->com);
2636 child_ep->parent_ep = parent_ep;
2637 child_ep->tos = tos;
2638 child_ep->dst = dst;
2639 child_ep->hwtid = hwtid;
2641 pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2642 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2644 timer_setup(&child_ep->timer, ep_timeout, 0);
2645 cxgb4_insert_tid(t, child_ep, hwtid,
2646 child_ep->com.local_addr.ss_family);
2647 insert_ep_tid(child_ep);
2648 if (accept_cr(child_ep, skb, req)) {
2649 c4iw_put_ep(&parent_ep->com);
2650 release_ep_resources(child_ep);
2652 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2655 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2656 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2657 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2661 c4iw_put_ep(&child_ep->com);
2663 reject_cr(dev, hwtid, skb);
2666 c4iw_put_ep(&parent_ep->com);
2670 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2673 struct cpl_pass_establish *req = cplhdr(skb);
2674 unsigned int tid = GET_TID(req);
2676 u16 tcp_opt = ntohs(req->tcp_opt);
2678 ep = get_ep_from_tid(dev, tid);
2679 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2680 ep->snd_seq = be32_to_cpu(req->snd_isn);
2681 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2682 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2684 pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2686 set_emss(ep, tcp_opt);
2688 dst_confirm(ep->dst);
2689 mutex_lock(&ep->com.mutex);
2690 ep->com.state = MPA_REQ_WAIT;
2692 set_bit(PASS_ESTAB, &ep->com.history);
2693 ret = send_flowc(ep);
2694 mutex_unlock(&ep->com.mutex);
2696 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2697 c4iw_put_ep(&ep->com);
2702 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2704 struct cpl_peer_close *hdr = cplhdr(skb);
2706 struct c4iw_qp_attributes attrs;
2709 unsigned int tid = GET_TID(hdr);
2712 ep = get_ep_from_tid(dev, tid);
2716 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2717 dst_confirm(ep->dst);
2719 set_bit(PEER_CLOSE, &ep->com.history);
2720 mutex_lock(&ep->com.mutex);
2721 switch (ep->com.state) {
2723 __state_set(&ep->com, CLOSING);
2726 __state_set(&ep->com, CLOSING);
2727 connect_reply_upcall(ep, -ECONNRESET);
2732 * We're gonna mark this puppy DEAD, but keep
2733 * the reference on it until the ULP accepts or
2734 * rejects the CR. Also wake up anyone waiting
2735 * in rdma connection migration (see c4iw_accept_cr()).
2737 __state_set(&ep->com, CLOSING);
2738 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2739 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2742 __state_set(&ep->com, CLOSING);
2743 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2744 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2748 __state_set(&ep->com, CLOSING);
2749 attrs.next_state = C4IW_QP_STATE_CLOSING;
2750 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2751 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2752 if (ret != -ECONNRESET) {
2753 peer_close_upcall(ep);
2761 __state_set(&ep->com, MORIBUND);
2765 (void)stop_ep_timer(ep);
2766 if (ep->com.cm_id && ep->com.qp) {
2767 attrs.next_state = C4IW_QP_STATE_IDLE;
2768 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2769 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2771 close_complete_upcall(ep, 0);
2772 __state_set(&ep->com, DEAD);
2780 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2782 mutex_unlock(&ep->com.mutex);
2784 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2786 release_ep_resources(ep);
2787 c4iw_put_ep(&ep->com);
2791 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2793 complete_cached_srq_buffers(ep, ep->srqe_idx);
2794 if (ep->com.cm_id && ep->com.qp) {
2795 struct c4iw_qp_attributes attrs;
2797 attrs.next_state = C4IW_QP_STATE_ERROR;
2798 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2799 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2801 peer_abort_upcall(ep);
2802 release_ep_resources(ep);
2803 c4iw_put_ep(&ep->com);
2806 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2808 struct cpl_abort_req_rss6 *req = cplhdr(skb);
2810 struct sk_buff *rpl_skb;
2811 struct c4iw_qp_attributes attrs;
2814 unsigned int tid = GET_TID(req);
2818 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2820 ep = get_ep_from_tid(dev, tid);
2824 status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2826 if (cxgb_is_neg_adv(status)) {
2827 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2828 ep->hwtid, status, neg_adv_str(status));
2829 ep->stats.abort_neg_adv++;
2830 mutex_lock(&dev->rdev.stats.lock);
2831 dev->rdev.stats.neg_adv++;
2832 mutex_unlock(&dev->rdev.stats.lock);
2836 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2838 set_bit(PEER_ABORT, &ep->com.history);
2841 * Wake up any threads in rdma_init() or rdma_fini().
2842 * However, this is not needed if com state is just
2845 if (ep->com.state != MPA_REQ_SENT)
2846 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2848 mutex_lock(&ep->com.mutex);
2849 switch (ep->com.state) {
2851 c4iw_put_ep(&ep->parent_ep->com);
2854 (void)stop_ep_timer(ep);
2857 (void)stop_ep_timer(ep);
2858 if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2859 (mpa_rev == 2 && ep->tried_with_mpa_v1))
2860 connect_reply_upcall(ep, -ECONNRESET);
2863 * we just don't send notification upwards because we
2864 * want to retry with mpa_v1 without upper layers even
2867 * do some housekeeping so as to re-initiate the
2870 pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2872 ep->retry_with_mpa_v1 = 1;
2884 if (ep->com.qp && ep->com.qp->srq) {
2885 srqidx = ABORT_RSS_SRQIDX_G(
2886 be32_to_cpu(req->srqidx_status));
2888 complete_cached_srq_buffers(ep, srqidx);
2890 /* Hold ep ref until finish_peer_abort() */
2891 c4iw_get_ep(&ep->com);
2892 __state_set(&ep->com, ABORTING);
2893 set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2900 if (ep->com.cm_id && ep->com.qp) {
2901 attrs.next_state = C4IW_QP_STATE_ERROR;
2902 ret = c4iw_modify_qp(ep->com.qp->rhp,
2903 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2906 pr_err("%s - qp <- error failed!\n", __func__);
2908 peer_abort_upcall(ep);
2913 pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2914 mutex_unlock(&ep->com.mutex);
2917 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2920 dst_confirm(ep->dst);
2921 if (ep->com.state != ABORTING) {
2922 __state_set(&ep->com, DEAD);
2923 /* we don't release if we want to retry with mpa_v1 */
2924 if (!ep->retry_with_mpa_v1)
2927 mutex_unlock(&ep->com.mutex);
2929 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2930 if (WARN_ON(!rpl_skb)) {
2935 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2937 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2940 release_ep_resources(ep);
2941 else if (ep->retry_with_mpa_v1) {
2942 if (ep->com.remote_addr.ss_family == AF_INET6) {
2943 struct sockaddr_in6 *sin6 =
2944 (struct sockaddr_in6 *)
2945 &ep->com.local_addr;
2947 ep->com.dev->rdev.lldi.ports[0],
2948 (const u32 *)&sin6->sin6_addr.s6_addr,
2951 xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2952 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2953 ep->com.local_addr.ss_family);
2954 dst_release(ep->dst);
2955 cxgb4_l2t_release(ep->l2t);
2960 c4iw_put_ep(&ep->com);
2961 /* Dereferencing ep, referenced in peer_abort_intr() */
2962 c4iw_put_ep(&ep->com);
2966 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2969 struct c4iw_qp_attributes attrs;
2970 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2972 unsigned int tid = GET_TID(rpl);
2974 ep = get_ep_from_tid(dev, tid);
2978 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2980 /* The cm_id may be null if we failed to connect */
2981 mutex_lock(&ep->com.mutex);
2982 set_bit(CLOSE_CON_RPL, &ep->com.history);
2983 switch (ep->com.state) {
2985 __state_set(&ep->com, MORIBUND);
2988 (void)stop_ep_timer(ep);
2989 if ((ep->com.cm_id) && (ep->com.qp)) {
2990 attrs.next_state = C4IW_QP_STATE_IDLE;
2991 c4iw_modify_qp(ep->com.qp->rhp,
2993 C4IW_QP_ATTR_NEXT_STATE,
2996 close_complete_upcall(ep, 0);
2997 __state_set(&ep->com, DEAD);
3004 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3007 mutex_unlock(&ep->com.mutex);
3009 release_ep_resources(ep);
3010 c4iw_put_ep(&ep->com);
3014 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3016 struct cpl_rdma_terminate *rpl = cplhdr(skb);
3017 unsigned int tid = GET_TID(rpl);
3019 struct c4iw_qp_attributes attrs;
3021 ep = get_ep_from_tid(dev, tid);
3025 pr_warn("TERM received tid %u qpid %u\n", tid,
3026 ep->com.qp->wq.sq.qid);
3027 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3028 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3029 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3032 /* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3033 * when entering the TERM state the RNIC MUST initiate a CLOSE.
3035 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3036 c4iw_put_ep(&ep->com);
3038 pr_warn("TERM received tid %u no ep/qp\n", tid);
3044 * Upcall from the adapter indicating data has been transmitted.
3045 * For us its just the single MPA request or reply. We can now free
3046 * the skb holding the mpa message.
3048 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3051 struct cpl_fw4_ack *hdr = cplhdr(skb);
3052 u8 credits = hdr->credits;
3053 unsigned int tid = GET_TID(hdr);
3056 ep = get_ep_from_tid(dev, tid);
3059 pr_debug("ep %p tid %u credits %u\n",
3060 ep, ep->hwtid, credits);
3062 pr_debug("0 credit ack ep %p tid %u state %u\n",
3063 ep, ep->hwtid, state_read(&ep->com));
3067 dst_confirm(ep->dst);
3069 pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3070 ep, ep->hwtid, state_read(&ep->com),
3071 ep->mpa_attr.initiator ? 1 : 0);
3072 mutex_lock(&ep->com.mutex);
3073 kfree_skb(ep->mpa_skb);
3075 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3077 mutex_unlock(&ep->com.mutex);
3080 c4iw_put_ep(&ep->com);
3084 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3087 struct c4iw_ep *ep = to_ep(cm_id);
3089 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3091 mutex_lock(&ep->com.mutex);
3092 if (ep->com.state != MPA_REQ_RCVD) {
3093 mutex_unlock(&ep->com.mutex);
3094 c4iw_put_ep(&ep->com);
3097 set_bit(ULP_REJECT, &ep->com.history);
3101 abort = send_mpa_reject(ep, pdata, pdata_len);
3102 mutex_unlock(&ep->com.mutex);
3105 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3106 c4iw_put_ep(&ep->com);
3110 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3113 struct c4iw_qp_attributes attrs;
3114 enum c4iw_qp_attr_mask mask;
3115 struct c4iw_ep *ep = to_ep(cm_id);
3116 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3117 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3120 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3122 mutex_lock(&ep->com.mutex);
3123 if (ep->com.state != MPA_REQ_RCVD) {
3133 set_bit(ULP_ACCEPT, &ep->com.history);
3134 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3135 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3140 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3141 if (conn_param->ord > ep->ird) {
3142 if (RELAXED_IRD_NEGOTIATION) {
3143 conn_param->ord = ep->ird;
3145 ep->ird = conn_param->ird;
3146 ep->ord = conn_param->ord;
3147 send_mpa_reject(ep, conn_param->private_data,
3148 conn_param->private_data_len);
3153 if (conn_param->ird < ep->ord) {
3154 if (RELAXED_IRD_NEGOTIATION &&
3155 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3156 conn_param->ird = ep->ord;
3163 ep->ird = conn_param->ird;
3164 ep->ord = conn_param->ord;
3166 if (ep->mpa_attr.version == 1) {
3167 if (peer2peer && ep->ird == 0)
3171 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3172 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3176 pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3178 ep->com.cm_id = cm_id;
3179 ref_cm_id(&ep->com);
3183 /* bind QP to EP and move to RTS */
3184 attrs.mpa_attr = ep->mpa_attr;
3185 attrs.max_ird = ep->ird;
3186 attrs.max_ord = ep->ord;
3187 attrs.llp_stream_handle = ep;
3188 attrs.next_state = C4IW_QP_STATE_RTS;
3190 /* bind QP and TID with INIT_WR */
3191 mask = C4IW_QP_ATTR_NEXT_STATE |
3192 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3193 C4IW_QP_ATTR_MPA_ATTR |
3194 C4IW_QP_ATTR_MAX_IRD |
3195 C4IW_QP_ATTR_MAX_ORD;
3197 err = c4iw_modify_qp(ep->com.qp->rhp,
3198 ep->com.qp, mask, &attrs, 1);
3200 goto err_deref_cm_id;
3202 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3203 err = send_mpa_reply(ep, conn_param->private_data,
3204 conn_param->private_data_len);
3206 goto err_deref_cm_id;
3208 __state_set(&ep->com, FPDU_MODE);
3209 established_upcall(ep);
3210 mutex_unlock(&ep->com.mutex);
3211 c4iw_put_ep(&ep->com);
3214 deref_cm_id(&ep->com);
3218 mutex_unlock(&ep->com.mutex);
3220 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3221 c4iw_put_ep(&ep->com);
3225 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3227 struct in_device *ind;
3229 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3230 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3231 const struct in_ifaddr *ifa;
3233 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3235 return -EADDRNOTAVAIL;
3237 in_dev_for_each_ifa_rcu(ifa, ind) {
3238 if (ifa->ifa_flags & IFA_F_SECONDARY)
3240 laddr->sin_addr.s_addr = ifa->ifa_address;
3241 raddr->sin_addr.s_addr = ifa->ifa_address;
3248 return found ? 0 : -EADDRNOTAVAIL;
3251 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3252 unsigned char banned_flags)
3254 struct inet6_dev *idev;
3255 int err = -EADDRNOTAVAIL;
3258 idev = __in6_dev_get(dev);
3260 struct inet6_ifaddr *ifp;
3262 read_lock_bh(&idev->lock);
3263 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3264 if (ifp->scope == IFA_LINK &&
3265 !(ifp->flags & banned_flags)) {
3266 memcpy(addr, &ifp->addr, 16);
3271 read_unlock_bh(&idev->lock);
3277 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3279 struct in6_addr addr;
3280 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3281 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3283 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3284 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3285 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3288 return -EADDRNOTAVAIL;
3291 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3293 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3296 struct sockaddr_in *laddr;
3297 struct sockaddr_in *raddr;
3298 struct sockaddr_in6 *laddr6;
3299 struct sockaddr_in6 *raddr6;
3303 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3304 (conn_param->ird > cur_max_read_depth(dev))) {
3308 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3310 pr_err("%s - cannot alloc ep\n", __func__);
3315 skb_queue_head_init(&ep->com.ep_skb_list);
3316 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3321 timer_setup(&ep->timer, ep_timeout, 0);
3322 ep->plen = conn_param->private_data_len;
3324 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3325 conn_param->private_data, ep->plen);
3326 ep->ird = conn_param->ird;
3327 ep->ord = conn_param->ord;
3329 if (peer2peer && ep->ord == 0)
3332 ep->com.cm_id = cm_id;
3333 ref_cm_id(&ep->com);
3334 cm_id->provider_data = ep;
3336 ep->com.qp = get_qhp(dev, conn_param->qpn);
3338 pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3343 pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3347 * Allocate an active TID to initiate a TCP connection.
3349 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3350 if (ep->atid == -1) {
3351 pr_err("%s - cannot alloc atid\n", __func__);
3355 err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3359 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3360 sizeof(ep->com.local_addr));
3361 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3362 sizeof(ep->com.remote_addr));
3364 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3365 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3366 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3367 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3369 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3371 ra = (__u8 *)&raddr->sin_addr;
3374 * Handle loopback requests to INADDR_ANY.
3376 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3377 err = pick_local_ipaddrs(dev, cm_id);
3383 pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3384 &laddr->sin_addr, ntohs(laddr->sin_port),
3385 ra, ntohs(raddr->sin_port));
3386 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3387 laddr->sin_addr.s_addr,
3388 raddr->sin_addr.s_addr,
3390 raddr->sin_port, cm_id->tos);
3393 ra = (__u8 *)&raddr6->sin6_addr;
3396 * Handle loopback requests to INADDR_ANY.
3398 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3399 err = pick_local_ip6addrs(dev, cm_id);
3405 pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3406 laddr6->sin6_addr.s6_addr,
3407 ntohs(laddr6->sin6_port),
3408 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3409 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3410 laddr6->sin6_addr.s6_addr,
3411 raddr6->sin6_addr.s6_addr,
3413 raddr6->sin6_port, cm_id->tos,
3414 raddr6->sin6_scope_id);
3417 pr_err("%s - cannot find route\n", __func__);
3418 err = -EHOSTUNREACH;
3422 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3423 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3425 pr_err("%s - cannot alloc l2e\n", __func__);
3429 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3430 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3433 state_set(&ep->com, CONNECTING);
3434 ep->tos = cm_id->tos;
3436 /* send connect request to rnic */
3437 err = send_connect(ep);
3441 cxgb4_l2t_release(ep->l2t);
3443 dst_release(ep->dst);
3445 xa_erase_irq(&ep->com.dev->atids, ep->atid);
3447 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3449 skb_queue_purge(&ep->com.ep_skb_list);
3450 deref_cm_id(&ep->com);
3452 c4iw_put_ep(&ep->com);
3457 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3460 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3461 &ep->com.local_addr;
3463 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3464 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3465 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3469 c4iw_init_wr_wait(ep->com.wr_waitp);
3470 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3471 ep->stid, &sin6->sin6_addr,
3473 ep->com.dev->rdev.lldi.rxq_ids[0]);
3475 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3479 err = net_xmit_errno(err);
3481 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3482 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3483 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3485 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3490 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3493 struct sockaddr_in *sin = (struct sockaddr_in *)
3494 &ep->com.local_addr;
3496 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3498 err = cxgb4_create_server_filter(
3499 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3500 sin->sin_addr.s_addr, sin->sin_port, 0,
3501 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3502 if (err == -EBUSY) {
3503 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3507 set_current_state(TASK_UNINTERRUPTIBLE);
3508 schedule_timeout(usecs_to_jiffies(100));
3510 } while (err == -EBUSY);
3512 c4iw_init_wr_wait(ep->com.wr_waitp);
3513 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3514 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3515 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3517 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3521 err = net_xmit_errno(err);
3524 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3526 &sin->sin_addr, ntohs(sin->sin_port));
3530 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3533 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3534 struct c4iw_listen_ep *ep;
3538 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3540 pr_err("%s - cannot alloc ep\n", __func__);
3544 skb_queue_head_init(&ep->com.ep_skb_list);
3545 pr_debug("ep %p\n", ep);
3546 ep->com.cm_id = cm_id;
3547 ref_cm_id(&ep->com);
3549 ep->backlog = backlog;
3550 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3551 sizeof(ep->com.local_addr));
3554 * Allocate a server TID.
3556 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3557 ep->com.local_addr.ss_family == AF_INET)
3558 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3559 cm_id->m_local_addr.ss_family, ep);
3561 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3562 cm_id->m_local_addr.ss_family, ep);
3564 if (ep->stid == -1) {
3565 pr_err("%s - cannot alloc stid\n", __func__);
3569 err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3573 state_set(&ep->com, LISTEN);
3574 if (ep->com.local_addr.ss_family == AF_INET)
3575 err = create_server4(dev, ep);
3577 err = create_server6(dev, ep);
3579 cm_id->provider_data = ep;
3582 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3584 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3585 ep->com.local_addr.ss_family);
3587 deref_cm_id(&ep->com);
3588 c4iw_put_ep(&ep->com);
3594 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3597 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3599 pr_debug("ep %p\n", ep);
3602 state_set(&ep->com, DEAD);
3603 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3604 ep->com.local_addr.ss_family == AF_INET) {
3605 err = cxgb4_remove_server_filter(
3606 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3607 ep->com.dev->rdev.lldi.rxq_ids[0], false);
3609 struct sockaddr_in6 *sin6;
3610 c4iw_init_wr_wait(ep->com.wr_waitp);
3611 err = cxgb4_remove_server(
3612 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3613 ep->com.dev->rdev.lldi.rxq_ids[0],
3614 ep->com.local_addr.ss_family == AF_INET6);
3617 err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3619 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3620 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3621 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3623 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3624 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3625 ep->com.local_addr.ss_family);
3627 deref_cm_id(&ep->com);
3628 c4iw_put_ep(&ep->com);
3632 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3637 struct c4iw_rdev *rdev;
3639 mutex_lock(&ep->com.mutex);
3641 pr_debug("ep %p state %s, abrupt %d\n", ep,
3642 states[ep->com.state], abrupt);
3645 * Ref the ep here in case we have fatal errors causing the
3646 * ep to be released and freed.
3648 c4iw_get_ep(&ep->com);
3650 rdev = &ep->com.dev->rdev;
3651 if (c4iw_fatal_error(rdev)) {
3653 close_complete_upcall(ep, -EIO);
3654 ep->com.state = DEAD;
3656 switch (ep->com.state) {
3665 ep->com.state = ABORTING;
3667 ep->com.state = CLOSING;
3670 * if we close before we see the fw4_ack() then we fix
3671 * up the timer state since we're reusing it.
3674 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3675 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3680 set_bit(CLOSE_SENT, &ep->com.flags);
3683 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3686 (void)stop_ep_timer(ep);
3687 ep->com.state = ABORTING;
3689 ep->com.state = MORIBUND;
3695 pr_debug("ignoring disconnect ep %p state %u\n",
3699 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3705 set_bit(EP_DISC_ABORT, &ep->com.history);
3706 ret = send_abort(ep);
3708 set_bit(EP_DISC_CLOSE, &ep->com.history);
3709 ret = send_halfclose(ep);
3712 set_bit(EP_DISC_FAIL, &ep->com.history);
3715 close_complete_upcall(ep, -EIO);
3718 struct c4iw_qp_attributes attrs;
3720 attrs.next_state = C4IW_QP_STATE_ERROR;
3721 ret = c4iw_modify_qp(ep->com.qp->rhp,
3723 C4IW_QP_ATTR_NEXT_STATE,
3726 pr_err("%s - qp <- error failed!\n",
3732 mutex_unlock(&ep->com.mutex);
3733 c4iw_put_ep(&ep->com);
3735 release_ep_resources(ep);
3739 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3740 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3743 int atid = be32_to_cpu(req->tid);
3745 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3746 (__force u32) req->tid);
3750 switch (req->retval) {
3752 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3753 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3754 send_fw_act_open_req(ep, atid);
3759 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3760 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3761 send_fw_act_open_req(ep, atid);
3766 pr_info("%s unexpected ofld conn wr retval %d\n",
3767 __func__, req->retval);
3770 pr_err("active ofld_connect_wr failure %d atid %d\n",
3772 mutex_lock(&dev->rdev.stats.lock);
3773 dev->rdev.stats.act_ofld_conn_fails++;
3774 mutex_unlock(&dev->rdev.stats.lock);
3775 connect_reply_upcall(ep, status2errno(req->retval));
3776 state_set(&ep->com, DEAD);
3777 if (ep->com.remote_addr.ss_family == AF_INET6) {
3778 struct sockaddr_in6 *sin6 =
3779 (struct sockaddr_in6 *)&ep->com.local_addr;
3780 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3781 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3783 xa_erase_irq(&dev->atids, atid);
3784 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3785 dst_release(ep->dst);
3786 cxgb4_l2t_release(ep->l2t);
3787 c4iw_put_ep(&ep->com);
3790 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3791 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3793 struct sk_buff *rpl_skb;
3794 struct cpl_pass_accept_req *cpl;
3797 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3799 pr_err("%s passive open failure %d\n", __func__, req->retval);
3800 mutex_lock(&dev->rdev.stats.lock);
3801 dev->rdev.stats.pas_ofld_conn_fails++;
3802 mutex_unlock(&dev->rdev.stats.lock);
3805 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3806 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3807 (__force u32) htonl(
3808 (__force u32) req->tid)));
3809 ret = pass_accept_req(dev, rpl_skb);
3816 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3818 u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3819 u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3823 t = (thi << shift) | (tlo >> shift);
3828 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3831 u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3835 v = (t >> shift) & mask;
3839 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3841 struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3842 __be64 *tcb = (__be64 *)(rpl + 1);
3843 unsigned int tid = GET_TID(rpl);
3848 ep = get_ep_from_tid(dev, tid);
3851 /* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3852 * determine if there's a rx PDU feedback event pending.
3854 * If that bit is set, it means we'll need to re-read the TCB's
3855 * rq_start value. The final value is the one present in a TCB
3856 * with the TF_RX_PDU_OUT bit cleared.
3859 t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3860 rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3862 c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3863 c4iw_put_ep(&ep->com); /* from read_tcb() */
3865 /* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3867 if (++ep->rx_pdu_out_cnt >= 2) {
3868 WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3875 ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3878 pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3880 if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3881 finish_peer_abort(dev, ep);
3882 else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3885 WARN_ONCE(1, "unexpected state!");
3890 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3892 struct cpl_fw6_msg *rpl = cplhdr(skb);
3893 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3895 switch (rpl->type) {
3897 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3899 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3900 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3901 switch (req->t_state) {
3903 active_ofld_conn_reply(dev, skb, req);
3906 passive_ofld_conn_reply(dev, skb, req);
3909 pr_err("%s unexpected ofld conn wr state %d\n",
3910 __func__, req->t_state);
3918 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3921 __be16 hdr_len, vlantag, len;
3923 int tcp_hdr_len, ip_hdr_len;
3925 struct cpl_rx_pkt *cpl = cplhdr(skb);
3926 struct cpl_pass_accept_req *req;
3927 struct tcp_options_received tmp_opt;
3928 struct c4iw_dev *dev;
3929 enum chip_type type;
3931 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3932 /* Store values from cpl_rx_pkt in temporary location. */
3933 vlantag = cpl->vlan;
3935 l2info = cpl->l2info;
3936 hdr_len = cpl->hdr_len;
3939 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3942 * We need to parse the TCP options from SYN packet.
3943 * to generate cpl_pass_accept_req.
3945 memset(&tmp_opt, 0, sizeof(tmp_opt));
3946 tcp_clear_options(&tmp_opt);
3947 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3949 req = __skb_push(skb, sizeof(*req));
3950 memset(req, 0, sizeof(*req));
3951 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3952 SYN_MAC_IDX_V(RX_MACIDX_G(
3953 be32_to_cpu(l2info))) |
3955 type = dev->rdev.lldi.adapter_type;
3956 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3957 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3959 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3960 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3961 eth_hdr_len = is_t4(type) ?
3962 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3963 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3964 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3965 IP_HDR_LEN_V(ip_hdr_len) |
3966 ETH_HDR_LEN_V(eth_hdr_len));
3967 } else { /* T6 and later */
3968 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3969 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3970 T6_IP_HDR_LEN_V(ip_hdr_len) |
3971 T6_ETH_HDR_LEN_V(eth_hdr_len));
3973 req->vlan = vlantag;
3975 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3976 PASS_OPEN_TOS_V(tos));
3977 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3978 if (tmp_opt.wscale_ok)
3979 req->tcpopt.wsf = tmp_opt.snd_wscale;
3980 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3981 if (tmp_opt.sack_ok)
3982 req->tcpopt.sack = 1;
3983 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3987 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3988 __be32 laddr, __be16 lport,
3989 __be32 raddr, __be16 rport,
3990 u32 rcv_isn, u32 filter, u16 window,
3991 u32 rss_qid, u8 port_id)
3993 struct sk_buff *req_skb;
3994 struct fw_ofld_connection_wr *req;
3995 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3998 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4001 req = __skb_put_zero(req_skb, sizeof(*req));
4002 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4003 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4004 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4005 req->le.filter = (__force __be32) filter;
4006 req->le.lport = lport;
4007 req->le.pport = rport;
4008 req->le.u.ipv4.lip = laddr;
4009 req->le.u.ipv4.pip = raddr;
4010 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4011 req->tcb.rcv_adv = htons(window);
4012 req->tcb.t_state_to_astid =
4013 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4014 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4015 FW_OFLD_CONNECTION_WR_ASTID_V(
4016 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4019 * We store the qid in opt2 which will be used by the firmware
4020 * to send us the wr response.
4022 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4025 * We initialize the MSS index in TCB to 0xF.
4026 * So that when driver sends cpl_pass_accept_rpl
4027 * TCB picks up the correct value. If this was 0
4028 * TP will ignore any value > 0 for MSS index.
4030 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4031 req->cookie = (uintptr_t)skb;
4033 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4034 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4036 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4044 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4045 * messages when a filter is being used instead of server to
4046 * redirect a syn packet. When packets hit filter they are redirected
4047 * to the offload queue and driver tries to establish the connection
4048 * using firmware work request.
4050 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4053 unsigned int filter;
4054 struct ethhdr *eh = NULL;
4055 struct vlan_ethhdr *vlan_eh = NULL;
4057 struct tcphdr *tcph;
4058 struct rss_header *rss = (void *)skb->data;
4059 struct cpl_rx_pkt *cpl = (void *)skb->data;
4060 struct cpl_pass_accept_req *req = (void *)(rss + 1);
4061 struct l2t_entry *e;
4062 struct dst_entry *dst;
4063 struct c4iw_ep *lep = NULL;
4065 struct port_info *pi;
4066 struct net_device *pdev;
4067 u16 rss_qid, eth_hdr_len;
4069 struct neighbour *neigh;
4071 /* Drop all non-SYN packets */
4072 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4076 * Drop all packets which did not hit the filter.
4077 * Unlikely to happen.
4079 if (!(rss->filter_hit && rss->filter_tid))
4083 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4085 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4087 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4089 pr_warn("%s connect request on invalid stid %d\n",
4094 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4096 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4099 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4102 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4105 pr_err("T%d Chip is not supported\n",
4106 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4110 if (eth_hdr_len == ETH_HLEN) {
4111 eh = (struct ethhdr *)(req + 1);
4112 iph = (struct iphdr *)(eh + 1);
4114 vlan_eh = (struct vlan_ethhdr *)(req + 1);
4115 iph = (struct iphdr *)(vlan_eh + 1);
4116 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4119 if (iph->version != 0x4)
4122 tcph = (struct tcphdr *)(iph + 1);
4123 skb_set_network_header(skb, (void *)iph - (void *)rss);
4124 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4127 pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4128 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4129 ntohs(tcph->source), iph->tos);
4131 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4132 iph->daddr, iph->saddr, tcph->dest,
4133 tcph->source, iph->tos);
4135 pr_err("%s - failed to find dst entry!\n", __func__);
4138 neigh = dst_neigh_lookup_skb(dst, skb);
4141 pr_err("%s - failed to allocate neigh!\n", __func__);
4145 if (neigh->dev->flags & IFF_LOOPBACK) {
4146 pdev = ip_dev_find(&init_net, iph->daddr);
4147 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4149 pi = (struct port_info *)netdev_priv(pdev);
4152 pdev = get_real_dev(neigh->dev);
4153 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4155 pi = (struct port_info *)netdev_priv(pdev);
4157 neigh_release(neigh);
4159 pr_err("%s - failed to allocate l2t entry!\n",
4164 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4165 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4166 window = (__force u16) htons((__force u16)tcph->window);
4168 /* Calcuate filter portion for LE region. */
4169 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4170 dev->rdev.lldi.ports[0],
4174 * Synthesize the cpl_pass_accept_req. We have everything except the
4175 * TID. Once firmware sends a reply with TID we update the TID field
4176 * in cpl and pass it through the regular cpl_pass_accept_req path.
4178 build_cpl_pass_accept_req(skb, stid, iph->tos);
4179 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4180 tcph->source, ntohl(tcph->seq), filter, window,
4181 rss_qid, pi->port_id);
4182 cxgb4_l2t_release(e);
4187 c4iw_put_ep(&lep->com);
4192 * These are the real handlers that are called from a
4195 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4196 [CPL_ACT_ESTABLISH] = act_establish,
4197 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4198 [CPL_RX_DATA] = rx_data,
4199 [CPL_ABORT_RPL_RSS] = abort_rpl,
4200 [CPL_ABORT_RPL] = abort_rpl,
4201 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4202 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4203 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4204 [CPL_PASS_ESTABLISH] = pass_establish,
4205 [CPL_PEER_CLOSE] = peer_close,
4206 [CPL_ABORT_REQ_RSS] = peer_abort,
4207 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4208 [CPL_RDMA_TERMINATE] = terminate,
4209 [CPL_FW4_ACK] = fw4_ack,
4210 [CPL_GET_TCB_RPL] = read_tcb_rpl,
4211 [CPL_FW6_MSG] = deferred_fw6_msg,
4212 [CPL_RX_PKT] = rx_pkt,
4213 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4214 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4217 static void process_timeout(struct c4iw_ep *ep)
4219 struct c4iw_qp_attributes attrs;
4222 mutex_lock(&ep->com.mutex);
4223 pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4224 set_bit(TIMEDOUT, &ep->com.history);
4225 switch (ep->com.state) {
4227 connect_reply_upcall(ep, -ETIMEDOUT);
4236 if (ep->com.cm_id && ep->com.qp) {
4237 attrs.next_state = C4IW_QP_STATE_ERROR;
4238 c4iw_modify_qp(ep->com.qp->rhp,
4239 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4242 close_complete_upcall(ep, -ETIMEDOUT);
4248 * These states are expected if the ep timed out at the same
4249 * time as another thread was calling stop_ep_timer().
4250 * So we silently do nothing for these states.
4255 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4256 __func__, ep, ep->hwtid, ep->com.state);
4259 mutex_unlock(&ep->com.mutex);
4261 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4262 c4iw_put_ep(&ep->com);
4265 static void process_timedout_eps(void)
4269 spin_lock_irq(&timeout_lock);
4270 while (!list_empty(&timeout_list)) {
4271 struct list_head *tmp;
4273 tmp = timeout_list.next;
4277 spin_unlock_irq(&timeout_lock);
4278 ep = list_entry(tmp, struct c4iw_ep, entry);
4279 process_timeout(ep);
4280 spin_lock_irq(&timeout_lock);
4282 spin_unlock_irq(&timeout_lock);
4285 static void process_work(struct work_struct *work)
4287 struct sk_buff *skb = NULL;
4288 struct c4iw_dev *dev;
4289 struct cpl_act_establish *rpl;
4290 unsigned int opcode;
4293 process_timedout_eps();
4294 while ((skb = skb_dequeue(&rxq))) {
4296 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4297 opcode = rpl->ot.opcode;
4299 if (opcode >= ARRAY_SIZE(work_handlers) ||
4300 !work_handlers[opcode]) {
4301 pr_err("No handler for opcode 0x%x.\n", opcode);
4304 ret = work_handlers[opcode](dev, skb);
4308 process_timedout_eps();
4312 static DECLARE_WORK(skb_work, process_work);
4314 static void ep_timeout(struct timer_list *t)
4316 struct c4iw_ep *ep = from_timer(ep, t, timer);
4319 spin_lock(&timeout_lock);
4320 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4322 * Only insert if it is not already on the list.
4324 if (!ep->entry.next) {
4325 list_add_tail(&ep->entry, &timeout_list);
4329 spin_unlock(&timeout_lock);
4331 queue_work(workq, &skb_work);
4335 * All the CM events are handled on a work queue to have a safe context.
4337 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4341 * Save dev in the skb->cb area.
4343 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4346 * Queue the skb and schedule the worker thread.
4348 skb_queue_tail(&rxq, skb);
4349 queue_work(workq, &skb_work);
4353 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4355 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4357 if (rpl->status != CPL_ERR_NONE) {
4358 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4359 rpl->status, GET_TID(rpl));
4365 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4367 struct cpl_fw6_msg *rpl = cplhdr(skb);
4368 struct c4iw_wr_wait *wr_waitp;
4371 pr_debug("type %u\n", rpl->type);
4373 switch (rpl->type) {
4374 case FW6_TYPE_WR_RPL:
4375 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4376 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4377 pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4379 c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4383 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4387 pr_err("%s unexpected fw6 msg type %u\n",
4388 __func__, rpl->type);
4395 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4397 struct cpl_abort_req_rss *req = cplhdr(skb);
4399 unsigned int tid = GET_TID(req);
4401 ep = get_ep_from_tid(dev, tid);
4402 /* This EP will be dereferenced in peer_abort() */
4404 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4408 if (cxgb_is_neg_adv(req->status)) {
4409 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4410 ep->hwtid, req->status,
4411 neg_adv_str(req->status));
4414 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4416 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4423 * Most upcalls from the T4 Core go to sched() to
4424 * schedule the processing on a work queue.
4426 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4427 [CPL_ACT_ESTABLISH] = sched,
4428 [CPL_ACT_OPEN_RPL] = sched,
4429 [CPL_RX_DATA] = sched,
4430 [CPL_ABORT_RPL_RSS] = sched,
4431 [CPL_ABORT_RPL] = sched,
4432 [CPL_PASS_OPEN_RPL] = sched,
4433 [CPL_CLOSE_LISTSRV_RPL] = sched,
4434 [CPL_PASS_ACCEPT_REQ] = sched,
4435 [CPL_PASS_ESTABLISH] = sched,
4436 [CPL_PEER_CLOSE] = sched,
4437 [CPL_CLOSE_CON_RPL] = sched,
4438 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4439 [CPL_RDMA_TERMINATE] = sched,
4440 [CPL_FW4_ACK] = sched,
4441 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4442 [CPL_GET_TCB_RPL] = sched,
4443 [CPL_FW6_MSG] = fw6_msg,
4444 [CPL_RX_PKT] = sched
4447 int __init c4iw_cm_init(void)
4449 skb_queue_head_init(&rxq);
4451 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4458 void c4iw_cm_term(void)
4460 WARN_ON(!list_empty(&timeout_list));
4461 destroy_workqueue(workq);