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)");
80 static int dack_mode = 1;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
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 spinlock_t 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);
502 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
506 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
507 c4iw_put_ep(&ep->parent_ep->com);
508 release_ep_resources(ep);
514 * Fake up a special CPL opcode and call sched() so process_work() will call
515 * _put_ep_safe() in a safe context to free the ep resources. This is needed
516 * because ARP error handlers are called in an ATOMIC context, and
517 * _c4iw_free_ep() needs to block.
519 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
522 struct cpl_act_establish *rpl = cplhdr(skb);
524 /* Set our special ARP_FAILURE opcode */
525 rpl->ot.opcode = cpl;
528 * Save ep in the skb->cb area, after where sched() will save the dev
531 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
532 sched(ep->com.dev, skb);
535 /* Handle an ARP failure for an accept */
536 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
538 struct c4iw_ep *ep = handle;
540 pr_err("ARP failure during accept - tid %u - dropping connection\n",
543 __state_set(&ep->com, DEAD);
544 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
548 * Handle an ARP failure for an active open.
550 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
552 struct c4iw_ep *ep = handle;
554 pr_err("ARP failure during connect\n");
555 connect_reply_upcall(ep, -EHOSTUNREACH);
556 __state_set(&ep->com, DEAD);
557 if (ep->com.remote_addr.ss_family == AF_INET6) {
558 struct sockaddr_in6 *sin6 =
559 (struct sockaddr_in6 *)&ep->com.local_addr;
560 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
561 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
563 xa_erase_irq(&ep->com.dev->atids, ep->atid);
564 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
565 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
569 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
572 static void abort_arp_failure(void *handle, struct sk_buff *skb)
575 struct c4iw_ep *ep = handle;
576 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
577 struct cpl_abort_req *req = cplhdr(skb);
579 pr_debug("rdev %p\n", rdev);
580 req->cmd = CPL_ABORT_NO_RST;
582 ret = c4iw_ofld_send(rdev, skb);
584 __state_set(&ep->com, DEAD);
585 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
590 static int send_flowc(struct c4iw_ep *ep)
592 struct fw_flowc_wr *flowc;
593 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
594 u16 vlan = ep->l2t->vlan;
596 int flowclen, flowclen16;
601 if (vlan == CPL_L2T_VLAN_NONE)
606 flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
607 flowclen16 = DIV_ROUND_UP(flowclen, 16);
608 flowclen = flowclen16 * 16;
610 flowc = __skb_put(skb, flowclen);
611 memset(flowc, 0, flowclen);
613 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
614 FW_FLOWC_WR_NPARAMS_V(nparams));
615 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
616 FW_WR_FLOWID_V(ep->hwtid));
618 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
619 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
620 (ep->com.dev->rdev.lldi.pf));
621 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
622 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
623 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
624 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
625 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
626 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
627 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
628 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
629 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
630 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
631 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
632 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
633 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
634 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
635 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
636 flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
639 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
640 flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
641 flowc->mnemval[9].val = cpu_to_be32(pri);
644 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
645 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
648 static int send_halfclose(struct c4iw_ep *ep)
650 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
651 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
653 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
657 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
658 NULL, arp_failure_discard);
660 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
663 static void read_tcb(struct c4iw_ep *ep)
666 struct cpl_get_tcb *req;
667 int wrlen = roundup(sizeof(*req), 16);
669 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
673 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
674 req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
675 memset(req, 0, wrlen);
676 INIT_TP_WR(req, ep->hwtid);
677 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
678 req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
681 * keep a ref on the ep so the tcb is not unlocked before this
682 * cpl completes. The ref is released in read_tcb_rpl().
684 c4iw_get_ep(&ep->com);
685 if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
686 c4iw_put_ep(&ep->com);
689 static int send_abort_req(struct c4iw_ep *ep)
691 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
692 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
694 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
695 if (WARN_ON(!req_skb))
698 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
699 ep, abort_arp_failure);
701 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
704 static int send_abort(struct c4iw_ep *ep)
706 if (!ep->com.qp || !ep->com.qp->srq) {
710 set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
715 static int send_connect(struct c4iw_ep *ep)
717 struct cpl_act_open_req *req = NULL;
718 struct cpl_t5_act_open_req *t5req = NULL;
719 struct cpl_t6_act_open_req *t6req = NULL;
720 struct cpl_act_open_req6 *req6 = NULL;
721 struct cpl_t5_act_open_req6 *t5req6 = NULL;
722 struct cpl_t6_act_open_req6 *t6req6 = NULL;
726 unsigned int mtu_idx;
728 int win, sizev4, sizev6, wrlen;
729 struct sockaddr_in *la = (struct sockaddr_in *)
731 struct sockaddr_in *ra = (struct sockaddr_in *)
732 &ep->com.remote_addr;
733 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
735 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
736 &ep->com.remote_addr;
738 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
739 u32 isn = (prandom_u32() & ~7UL) - 1;
740 struct net_device *netdev;
743 netdev = ep->com.dev->rdev.lldi.ports[0];
745 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
747 sizev4 = sizeof(struct cpl_act_open_req);
748 sizev6 = sizeof(struct cpl_act_open_req6);
751 sizev4 = sizeof(struct cpl_t5_act_open_req);
752 sizev6 = sizeof(struct cpl_t5_act_open_req6);
755 sizev4 = sizeof(struct cpl_t6_act_open_req);
756 sizev6 = sizeof(struct cpl_t6_act_open_req6);
759 pr_err("T%d Chip is not supported\n",
760 CHELSIO_CHIP_VERSION(adapter_type));
764 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
765 roundup(sizev4, 16) :
768 pr_debug("ep %p atid %u\n", ep, ep->atid);
770 skb = get_skb(NULL, wrlen, GFP_KERNEL);
772 pr_err("%s - failed to alloc skb\n", __func__);
775 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
777 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
778 enable_tcp_timestamps,
779 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
780 wscale = cxgb_compute_wscale(rcv_win);
783 * Specify the largest window that will fit in opt0. The
784 * remainder will be specified in the rx_data_ack.
786 win = ep->rcv_win >> 10;
787 if (win > RCV_BUFSIZ_M)
790 opt0 = (nocong ? NO_CONG_F : 0) |
793 WND_SCALE_V(wscale) |
795 L2T_IDX_V(ep->l2t->idx) |
796 TX_CHAN_V(ep->tx_chan) |
797 SMAC_SEL_V(ep->smac_idx) |
798 DSCP_V(ep->tos >> 2) |
799 ULP_MODE_V(ULP_MODE_TCPDDP) |
801 opt2 = RX_CHANNEL_V(0) |
802 CCTRL_ECN_V(enable_ecn) |
803 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
804 if (enable_tcp_timestamps)
805 opt2 |= TSTAMPS_EN_F;
808 if (wscale && enable_tcp_window_scaling)
809 opt2 |= WND_SCALE_EN_F;
810 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
814 opt2 |= T5_OPT_2_VALID_F;
815 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
819 params = cxgb4_select_ntuple(netdev, ep->l2t);
821 if (ep->com.remote_addr.ss_family == AF_INET6)
822 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
823 (const u32 *)&la6->sin6_addr.s6_addr, 1);
825 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
827 if (ep->com.remote_addr.ss_family == AF_INET) {
828 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
830 req = skb_put(skb, wrlen);
834 t5req = skb_put(skb, wrlen);
835 INIT_TP_WR(t5req, 0);
836 req = (struct cpl_act_open_req *)t5req;
839 t6req = skb_put(skb, wrlen);
840 INIT_TP_WR(t6req, 0);
841 req = (struct cpl_act_open_req *)t6req;
842 t5req = (struct cpl_t5_act_open_req *)t6req;
845 pr_err("T%d Chip is not supported\n",
846 CHELSIO_CHIP_VERSION(adapter_type));
851 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
852 ((ep->rss_qid<<14) | ep->atid)));
853 req->local_port = la->sin_port;
854 req->peer_port = ra->sin_port;
855 req->local_ip = la->sin_addr.s_addr;
856 req->peer_ip = ra->sin_addr.s_addr;
857 req->opt0 = cpu_to_be64(opt0);
859 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
860 req->params = cpu_to_be32(params);
861 req->opt2 = cpu_to_be32(opt2);
863 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
865 cpu_to_be64(FILTER_TUPLE_V(params));
866 t5req->rsvd = cpu_to_be32(isn);
867 pr_debug("snd_isn %u\n", t5req->rsvd);
868 t5req->opt2 = cpu_to_be32(opt2);
871 cpu_to_be64(FILTER_TUPLE_V(params));
872 t6req->rsvd = cpu_to_be32(isn);
873 pr_debug("snd_isn %u\n", t6req->rsvd);
874 t6req->opt2 = cpu_to_be32(opt2);
878 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
880 req6 = skb_put(skb, wrlen);
884 t5req6 = skb_put(skb, wrlen);
885 INIT_TP_WR(t5req6, 0);
886 req6 = (struct cpl_act_open_req6 *)t5req6;
889 t6req6 = skb_put(skb, wrlen);
890 INIT_TP_WR(t6req6, 0);
891 req6 = (struct cpl_act_open_req6 *)t6req6;
892 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
895 pr_err("T%d Chip is not supported\n",
896 CHELSIO_CHIP_VERSION(adapter_type));
901 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
902 ((ep->rss_qid<<14)|ep->atid)));
903 req6->local_port = la6->sin6_port;
904 req6->peer_port = ra6->sin6_port;
905 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
906 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
907 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
908 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
909 req6->opt0 = cpu_to_be64(opt0);
911 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
912 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
914 req6->opt2 = cpu_to_be32(opt2);
916 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
918 cpu_to_be64(FILTER_TUPLE_V(params));
919 t5req6->rsvd = cpu_to_be32(isn);
920 pr_debug("snd_isn %u\n", t5req6->rsvd);
921 t5req6->opt2 = cpu_to_be32(opt2);
924 cpu_to_be64(FILTER_TUPLE_V(params));
925 t6req6->rsvd = cpu_to_be32(isn);
926 pr_debug("snd_isn %u\n", t6req6->rsvd);
927 t6req6->opt2 = cpu_to_be32(opt2);
933 set_bit(ACT_OPEN_REQ, &ep->com.history);
934 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
936 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
937 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
938 (const u32 *)&la6->sin6_addr.s6_addr, 1);
942 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
945 int mpalen, wrlen, ret;
946 struct fw_ofld_tx_data_wr *req;
947 struct mpa_message *mpa;
948 struct mpa_v2_conn_params mpa_v2_params;
950 pr_debug("ep %p tid %u pd_len %d\n",
951 ep, ep->hwtid, ep->plen);
953 mpalen = sizeof(*mpa) + ep->plen;
954 if (mpa_rev_to_use == 2)
955 mpalen += sizeof(struct mpa_v2_conn_params);
956 wrlen = roundup(mpalen + sizeof *req, 16);
957 skb = get_skb(skb, wrlen, GFP_KERNEL);
959 connect_reply_upcall(ep, -ENOMEM);
962 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
964 req = skb_put_zero(skb, wrlen);
965 req->op_to_immdlen = cpu_to_be32(
966 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
968 FW_WR_IMMDLEN_V(mpalen));
969 req->flowid_len16 = cpu_to_be32(
970 FW_WR_FLOWID_V(ep->hwtid) |
971 FW_WR_LEN16_V(wrlen >> 4));
972 req->plen = cpu_to_be32(mpalen);
973 req->tunnel_to_proxy = cpu_to_be32(
974 FW_OFLD_TX_DATA_WR_FLUSH_F |
975 FW_OFLD_TX_DATA_WR_SHOVE_F);
977 mpa = (struct mpa_message *)(req + 1);
978 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
982 mpa->flags |= MPA_CRC;
983 if (markers_enabled) {
984 mpa->flags |= MPA_MARKERS;
985 ep->mpa_attr.recv_marker_enabled = 1;
987 ep->mpa_attr.recv_marker_enabled = 0;
989 if (mpa_rev_to_use == 2)
990 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
992 mpa->private_data_size = htons(ep->plen);
993 mpa->revision = mpa_rev_to_use;
994 if (mpa_rev_to_use == 1) {
995 ep->tried_with_mpa_v1 = 1;
996 ep->retry_with_mpa_v1 = 0;
999 if (mpa_rev_to_use == 2) {
1000 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1001 sizeof (struct mpa_v2_conn_params));
1002 pr_debug("initiator ird %u ord %u\n", ep->ird,
1004 mpa_v2_params.ird = htons((u16)ep->ird);
1005 mpa_v2_params.ord = htons((u16)ep->ord);
1008 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1009 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1010 mpa_v2_params.ord |=
1011 htons(MPA_V2_RDMA_WRITE_RTR);
1012 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1013 mpa_v2_params.ord |=
1014 htons(MPA_V2_RDMA_READ_RTR);
1016 memcpy(mpa->private_data, &mpa_v2_params,
1017 sizeof(struct mpa_v2_conn_params));
1020 memcpy(mpa->private_data +
1021 sizeof(struct mpa_v2_conn_params),
1022 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1025 memcpy(mpa->private_data,
1026 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1029 * Reference the mpa skb. This ensures the data area
1030 * will remain in memory until the hw acks the tx.
1031 * Function fw4_ack() will deref it.
1034 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1036 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1040 __state_set(&ep->com, MPA_REQ_SENT);
1041 ep->mpa_attr.initiator = 1;
1042 ep->snd_seq += mpalen;
1046 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1049 struct fw_ofld_tx_data_wr *req;
1050 struct mpa_message *mpa;
1051 struct sk_buff *skb;
1052 struct mpa_v2_conn_params mpa_v2_params;
1054 pr_debug("ep %p tid %u pd_len %d\n",
1055 ep, ep->hwtid, ep->plen);
1057 mpalen = sizeof(*mpa) + plen;
1058 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1059 mpalen += sizeof(struct mpa_v2_conn_params);
1060 wrlen = roundup(mpalen + sizeof *req, 16);
1062 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1064 pr_err("%s - cannot alloc skb!\n", __func__);
1067 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1069 req = skb_put_zero(skb, wrlen);
1070 req->op_to_immdlen = cpu_to_be32(
1071 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1073 FW_WR_IMMDLEN_V(mpalen));
1074 req->flowid_len16 = cpu_to_be32(
1075 FW_WR_FLOWID_V(ep->hwtid) |
1076 FW_WR_LEN16_V(wrlen >> 4));
1077 req->plen = cpu_to_be32(mpalen);
1078 req->tunnel_to_proxy = cpu_to_be32(
1079 FW_OFLD_TX_DATA_WR_FLUSH_F |
1080 FW_OFLD_TX_DATA_WR_SHOVE_F);
1082 mpa = (struct mpa_message *)(req + 1);
1083 memset(mpa, 0, sizeof(*mpa));
1084 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1085 mpa->flags = MPA_REJECT;
1086 mpa->revision = ep->mpa_attr.version;
1087 mpa->private_data_size = htons(plen);
1089 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1090 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1091 mpa->private_data_size = 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 = htons(ntohs(mpa->private_data_size) +
1175 sizeof (struct mpa_v2_conn_params));
1176 mpa_v2_params.ird = htons((u16)ep->ird);
1177 mpa_v2_params.ord = htons((u16)ep->ord);
1178 if (peer2peer && (ep->mpa_attr.p2p_type !=
1179 FW_RI_INIT_P2PTYPE_DISABLED)) {
1180 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1182 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1183 mpa_v2_params.ord |=
1184 htons(MPA_V2_RDMA_WRITE_RTR);
1185 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1186 mpa_v2_params.ord |=
1187 htons(MPA_V2_RDMA_READ_RTR);
1190 memcpy(mpa->private_data, &mpa_v2_params,
1191 sizeof(struct mpa_v2_conn_params));
1194 memcpy(mpa->private_data +
1195 sizeof(struct mpa_v2_conn_params), pdata, plen);
1198 memcpy(mpa->private_data, pdata, plen);
1201 * Reference the mpa skb. This ensures the data area
1202 * will remain in memory until the hw acks the tx.
1203 * Function fw4_ack() will deref it.
1206 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1208 __state_set(&ep->com, MPA_REP_SENT);
1209 ep->snd_seq += mpalen;
1210 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1213 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1216 struct cpl_act_establish *req = cplhdr(skb);
1217 unsigned short tcp_opt = ntohs(req->tcp_opt);
1218 unsigned int tid = GET_TID(req);
1219 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1220 struct tid_info *t = dev->rdev.lldi.tids;
1223 ep = lookup_atid(t, atid);
1225 pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1226 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1228 mutex_lock(&ep->com.mutex);
1229 dst_confirm(ep->dst);
1231 /* setup the hwtid for this connection */
1233 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1236 ep->snd_seq = be32_to_cpu(req->snd_isn);
1237 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1238 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1240 set_emss(ep, tcp_opt);
1242 /* dealloc the atid */
1243 xa_erase_irq(&ep->com.dev->atids, atid);
1244 cxgb4_free_atid(t, atid);
1245 set_bit(ACT_ESTAB, &ep->com.history);
1247 /* start MPA negotiation */
1248 ret = send_flowc(ep);
1251 if (ep->retry_with_mpa_v1)
1252 ret = send_mpa_req(ep, skb, 1);
1254 ret = send_mpa_req(ep, skb, mpa_rev);
1257 mutex_unlock(&ep->com.mutex);
1260 mutex_unlock(&ep->com.mutex);
1261 connect_reply_upcall(ep, -ENOMEM);
1262 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1266 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1268 struct iw_cm_event event;
1270 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1271 memset(&event, 0, sizeof(event));
1272 event.event = IW_CM_EVENT_CLOSE;
1273 event.status = status;
1274 if (ep->com.cm_id) {
1275 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1276 ep, ep->com.cm_id, ep->hwtid);
1277 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1278 deref_cm_id(&ep->com);
1279 set_bit(CLOSE_UPCALL, &ep->com.history);
1283 static void peer_close_upcall(struct c4iw_ep *ep)
1285 struct iw_cm_event event;
1287 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1288 memset(&event, 0, sizeof(event));
1289 event.event = IW_CM_EVENT_DISCONNECT;
1290 if (ep->com.cm_id) {
1291 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1292 ep, ep->com.cm_id, ep->hwtid);
1293 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1294 set_bit(DISCONN_UPCALL, &ep->com.history);
1298 static void peer_abort_upcall(struct c4iw_ep *ep)
1300 struct iw_cm_event event;
1302 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1303 memset(&event, 0, sizeof(event));
1304 event.event = IW_CM_EVENT_CLOSE;
1305 event.status = -ECONNRESET;
1306 if (ep->com.cm_id) {
1307 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1308 ep->com.cm_id, ep->hwtid);
1309 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1310 deref_cm_id(&ep->com);
1311 set_bit(ABORT_UPCALL, &ep->com.history);
1315 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1317 struct iw_cm_event event;
1319 pr_debug("ep %p tid %u status %d\n",
1320 ep, ep->hwtid, status);
1321 memset(&event, 0, sizeof(event));
1322 event.event = IW_CM_EVENT_CONNECT_REPLY;
1323 event.status = status;
1324 memcpy(&event.local_addr, &ep->com.local_addr,
1325 sizeof(ep->com.local_addr));
1326 memcpy(&event.remote_addr, &ep->com.remote_addr,
1327 sizeof(ep->com.remote_addr));
1329 if ((status == 0) || (status == -ECONNREFUSED)) {
1330 if (!ep->tried_with_mpa_v1) {
1331 /* this means MPA_v2 is used */
1332 event.ord = ep->ird;
1333 event.ird = ep->ord;
1334 event.private_data_len = ep->plen -
1335 sizeof(struct mpa_v2_conn_params);
1336 event.private_data = ep->mpa_pkt +
1337 sizeof(struct mpa_message) +
1338 sizeof(struct mpa_v2_conn_params);
1340 /* this means MPA_v1 is used */
1341 event.ord = cur_max_read_depth(ep->com.dev);
1342 event.ird = cur_max_read_depth(ep->com.dev);
1343 event.private_data_len = ep->plen;
1344 event.private_data = ep->mpa_pkt +
1345 sizeof(struct mpa_message);
1349 pr_debug("ep %p tid %u status %d\n", ep,
1351 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1352 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1355 deref_cm_id(&ep->com);
1358 static int connect_request_upcall(struct c4iw_ep *ep)
1360 struct iw_cm_event event;
1363 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1364 memset(&event, 0, sizeof(event));
1365 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1366 memcpy(&event.local_addr, &ep->com.local_addr,
1367 sizeof(ep->com.local_addr));
1368 memcpy(&event.remote_addr, &ep->com.remote_addr,
1369 sizeof(ep->com.remote_addr));
1370 event.provider_data = ep;
1371 if (!ep->tried_with_mpa_v1) {
1372 /* this means MPA_v2 is used */
1373 event.ord = ep->ord;
1374 event.ird = ep->ird;
1375 event.private_data_len = ep->plen -
1376 sizeof(struct mpa_v2_conn_params);
1377 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1378 sizeof(struct mpa_v2_conn_params);
1380 /* this means MPA_v1 is used. Send max supported */
1381 event.ord = cur_max_read_depth(ep->com.dev);
1382 event.ird = cur_max_read_depth(ep->com.dev);
1383 event.private_data_len = ep->plen;
1384 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1386 c4iw_get_ep(&ep->com);
1387 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1390 c4iw_put_ep(&ep->com);
1391 set_bit(CONNREQ_UPCALL, &ep->com.history);
1392 c4iw_put_ep(&ep->parent_ep->com);
1396 static void established_upcall(struct c4iw_ep *ep)
1398 struct iw_cm_event event;
1400 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1401 memset(&event, 0, sizeof(event));
1402 event.event = IW_CM_EVENT_ESTABLISHED;
1403 event.ird = ep->ord;
1404 event.ord = ep->ird;
1405 if (ep->com.cm_id) {
1406 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1407 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1408 set_bit(ESTAB_UPCALL, &ep->com.history);
1412 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1414 struct sk_buff *skb;
1415 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1418 pr_debug("ep %p tid %u credits %u\n",
1419 ep, ep->hwtid, credits);
1420 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1422 pr_err("update_rx_credits - cannot alloc skb!\n");
1427 * If we couldn't specify the entire rcv window at connection setup
1428 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1429 * then add the overage in to the credits returned.
1431 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1432 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1434 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1435 RX_DACK_MODE_V(dack_mode);
1437 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1440 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1444 #define RELAXED_IRD_NEGOTIATION 1
1447 * process_mpa_reply - process streaming mode MPA reply
1451 * 0 upon success indicating a connect request was delivered to the ULP
1452 * or the mpa request is incomplete but valid so far.
1454 * 1 if a failure requires the caller to close the connection.
1456 * 2 if a failure requires the caller to abort the connection.
1458 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1460 struct mpa_message *mpa;
1461 struct mpa_v2_conn_params *mpa_v2_params;
1463 u16 resp_ird, resp_ord;
1464 u8 rtr_mismatch = 0, insuff_ird = 0;
1465 struct c4iw_qp_attributes attrs;
1466 enum c4iw_qp_attr_mask mask;
1470 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1473 * If we get more than the supported amount of private data
1474 * then we must fail this connection.
1476 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1478 goto err_stop_timer;
1482 * copy the new data into our accumulation buffer.
1484 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1486 ep->mpa_pkt_len += skb->len;
1489 * if we don't even have the mpa message, then bail.
1491 if (ep->mpa_pkt_len < sizeof(*mpa))
1493 mpa = (struct mpa_message *) ep->mpa_pkt;
1495 /* Validate MPA header. */
1496 if (mpa->revision > mpa_rev) {
1497 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1498 __func__, mpa_rev, mpa->revision);
1500 goto err_stop_timer;
1502 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1504 goto err_stop_timer;
1507 plen = ntohs(mpa->private_data_size);
1510 * Fail if there's too much private data.
1512 if (plen > MPA_MAX_PRIVATE_DATA) {
1514 goto err_stop_timer;
1518 * If plen does not account for pkt size
1520 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522 goto err_stop_timer;
1525 ep->plen = (u8) plen;
1528 * If we don't have all the pdata yet, then bail.
1529 * We'll continue process when more data arrives.
1531 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1534 if (mpa->flags & MPA_REJECT) {
1535 err = -ECONNREFUSED;
1536 goto err_stop_timer;
1540 * Stop mpa timer. If it expired, then
1541 * we ignore the MPA reply. process_timeout()
1542 * will abort the connection.
1544 if (stop_ep_timer(ep))
1548 * If we get here we have accumulated the entire mpa
1549 * start reply message including private data. And
1550 * the MPA header is valid.
1552 __state_set(&ep->com, FPDU_MODE);
1553 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1554 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1555 ep->mpa_attr.version = mpa->revision;
1556 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1558 if (mpa->revision == 2) {
1559 ep->mpa_attr.enhanced_rdma_conn =
1560 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1561 if (ep->mpa_attr.enhanced_rdma_conn) {
1562 mpa_v2_params = (struct mpa_v2_conn_params *)
1563 (ep->mpa_pkt + sizeof(*mpa));
1564 resp_ird = ntohs(mpa_v2_params->ird) &
1565 MPA_V2_IRD_ORD_MASK;
1566 resp_ord = ntohs(mpa_v2_params->ord) &
1567 MPA_V2_IRD_ORD_MASK;
1568 pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1569 resp_ird, resp_ord, ep->ird, ep->ord);
1572 * This is a double-check. Ideally, below checks are
1573 * not required since ird/ord stuff has been taken
1574 * care of in c4iw_accept_cr
1576 if (ep->ird < resp_ord) {
1577 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1578 ep->com.dev->rdev.lldi.max_ordird_qp)
1582 } else if (ep->ird > resp_ord) {
1585 if (ep->ord > resp_ird) {
1586 if (RELAXED_IRD_NEGOTIATION)
1597 if (ntohs(mpa_v2_params->ird) &
1598 MPA_V2_PEER2PEER_MODEL) {
1599 if (ntohs(mpa_v2_params->ord) &
1600 MPA_V2_RDMA_WRITE_RTR)
1601 ep->mpa_attr.p2p_type =
1602 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1603 else if (ntohs(mpa_v2_params->ord) &
1604 MPA_V2_RDMA_READ_RTR)
1605 ep->mpa_attr.p2p_type =
1606 FW_RI_INIT_P2PTYPE_READ_REQ;
1609 } else if (mpa->revision == 1)
1611 ep->mpa_attr.p2p_type = p2p_type;
1613 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1614 ep->mpa_attr.crc_enabled,
1615 ep->mpa_attr.recv_marker_enabled,
1616 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1617 ep->mpa_attr.p2p_type, p2p_type);
1620 * If responder's RTR does not match with that of initiator, assign
1621 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1622 * generated when moving QP to RTS state.
1623 * A TERM message will be sent after QP has moved to RTS state
1625 if ((ep->mpa_attr.version == 2) && peer2peer &&
1626 (ep->mpa_attr.p2p_type != p2p_type)) {
1627 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1631 attrs.mpa_attr = ep->mpa_attr;
1632 attrs.max_ird = ep->ird;
1633 attrs.max_ord = ep->ord;
1634 attrs.llp_stream_handle = ep;
1635 attrs.next_state = C4IW_QP_STATE_RTS;
1637 mask = C4IW_QP_ATTR_NEXT_STATE |
1638 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1639 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1641 /* bind QP and TID with INIT_WR */
1642 err = c4iw_modify_qp(ep->com.qp->rhp,
1643 ep->com.qp, mask, &attrs, 1);
1648 * If responder's RTR requirement did not match with what initiator
1649 * supports, generate TERM message
1652 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1653 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1654 attrs.ecode = MPA_NOMATCH_RTR;
1655 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1656 attrs.send_term = 1;
1657 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1658 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1665 * Generate TERM if initiator IRD is not sufficient for responder
1666 * provided ORD. Currently, we do the same behaviour even when
1667 * responder provided IRD is also not sufficient as regards to
1671 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1672 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1673 attrs.ecode = MPA_INSUFF_IRD;
1674 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1675 attrs.send_term = 1;
1676 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1677 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1688 connect_reply_upcall(ep, err);
1693 * process_mpa_request - process streaming mode MPA request
1697 * 0 upon success indicating a connect request was delivered to the ULP
1698 * or the mpa request is incomplete but valid so far.
1700 * 1 if a failure requires the caller to close the connection.
1702 * 2 if a failure requires the caller to abort the connection.
1704 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1706 struct mpa_message *mpa;
1707 struct mpa_v2_conn_params *mpa_v2_params;
1710 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1713 * If we get more than the supported amount of private data
1714 * then we must fail this connection.
1716 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1717 goto err_stop_timer;
1719 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1722 * Copy the new data into our accumulation buffer.
1724 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1726 ep->mpa_pkt_len += skb->len;
1729 * If we don't even have the mpa message, then bail.
1730 * We'll continue process when more data arrives.
1732 if (ep->mpa_pkt_len < sizeof(*mpa))
1735 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1736 mpa = (struct mpa_message *) ep->mpa_pkt;
1739 * Validate MPA Header.
1741 if (mpa->revision > mpa_rev) {
1742 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1743 __func__, mpa_rev, mpa->revision);
1744 goto err_stop_timer;
1747 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1748 goto err_stop_timer;
1750 plen = ntohs(mpa->private_data_size);
1753 * Fail if there's too much private data.
1755 if (plen > MPA_MAX_PRIVATE_DATA)
1756 goto err_stop_timer;
1759 * If plen does not account for pkt size
1761 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1762 goto err_stop_timer;
1763 ep->plen = (u8) plen;
1766 * If we don't have all the pdata yet, then bail.
1768 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1772 * If we get here we have accumulated the entire mpa
1773 * start reply message including private data.
1775 ep->mpa_attr.initiator = 0;
1776 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1777 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1778 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1779 ep->mpa_attr.version = mpa->revision;
1780 if (mpa->revision == 1)
1781 ep->tried_with_mpa_v1 = 1;
1782 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1784 if (mpa->revision == 2) {
1785 ep->mpa_attr.enhanced_rdma_conn =
1786 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1787 if (ep->mpa_attr.enhanced_rdma_conn) {
1788 mpa_v2_params = (struct mpa_v2_conn_params *)
1789 (ep->mpa_pkt + sizeof(*mpa));
1790 ep->ird = ntohs(mpa_v2_params->ird) &
1791 MPA_V2_IRD_ORD_MASK;
1792 ep->ird = min_t(u32, ep->ird,
1793 cur_max_read_depth(ep->com.dev));
1794 ep->ord = ntohs(mpa_v2_params->ord) &
1795 MPA_V2_IRD_ORD_MASK;
1796 ep->ord = min_t(u32, ep->ord,
1797 cur_max_read_depth(ep->com.dev));
1798 pr_debug("initiator ird %u ord %u\n",
1800 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1802 if (ntohs(mpa_v2_params->ord) &
1803 MPA_V2_RDMA_WRITE_RTR)
1804 ep->mpa_attr.p2p_type =
1805 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1806 else if (ntohs(mpa_v2_params->ord) &
1807 MPA_V2_RDMA_READ_RTR)
1808 ep->mpa_attr.p2p_type =
1809 FW_RI_INIT_P2PTYPE_READ_REQ;
1812 } else if (mpa->revision == 1)
1814 ep->mpa_attr.p2p_type = p2p_type;
1816 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1817 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1818 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1819 ep->mpa_attr.p2p_type);
1821 __state_set(&ep->com, MPA_REQ_RCVD);
1824 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1825 if (ep->parent_ep->com.state != DEAD) {
1826 if (connect_request_upcall(ep))
1827 goto err_unlock_parent;
1829 goto err_unlock_parent;
1831 mutex_unlock(&ep->parent_ep->com.mutex);
1835 mutex_unlock(&ep->parent_ep->com.mutex);
1838 (void)stop_ep_timer(ep);
1843 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1846 struct cpl_rx_data *hdr = cplhdr(skb);
1847 unsigned int dlen = ntohs(hdr->len);
1848 unsigned int tid = GET_TID(hdr);
1849 __u8 status = hdr->status;
1852 ep = get_ep_from_tid(dev, tid);
1855 pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1856 skb_pull(skb, sizeof(*hdr));
1857 skb_trim(skb, dlen);
1858 mutex_lock(&ep->com.mutex);
1860 switch (ep->com.state) {
1862 update_rx_credits(ep, dlen);
1863 ep->rcv_seq += dlen;
1864 disconnect = process_mpa_reply(ep, skb);
1867 update_rx_credits(ep, dlen);
1868 ep->rcv_seq += dlen;
1869 disconnect = process_mpa_request(ep, skb);
1872 struct c4iw_qp_attributes attrs;
1874 update_rx_credits(ep, dlen);
1876 pr_err("%s Unexpected streaming data." \
1877 " qpid %u ep %p state %d tid %u status %d\n",
1878 __func__, ep->com.qp->wq.sq.qid, ep,
1879 ep->com.state, ep->hwtid, status);
1880 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1881 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1882 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1889 mutex_unlock(&ep->com.mutex);
1891 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1892 c4iw_put_ep(&ep->com);
1896 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1898 enum chip_type adapter_type;
1900 adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1903 * If this TCB had a srq buffer cached, then we must complete
1904 * it. For user mode, that means saving the srqidx in the
1905 * user/kernel status page for this qp. For kernel mode, just
1906 * synthesize the CQE now.
1908 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1909 if (ep->com.qp->ibqp.uobject)
1910 t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1912 c4iw_flush_srqidx(ep->com.qp, srqidx);
1916 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1920 struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1922 unsigned int tid = GET_TID(rpl);
1924 ep = get_ep_from_tid(dev, tid);
1926 pr_warn("Abort rpl to freed endpoint\n");
1930 if (ep->com.qp && ep->com.qp->srq) {
1931 srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1932 complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1935 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1936 mutex_lock(&ep->com.mutex);
1937 switch (ep->com.state) {
1939 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1940 __state_set(&ep->com, DEAD);
1944 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1947 mutex_unlock(&ep->com.mutex);
1950 close_complete_upcall(ep, -ECONNRESET);
1951 release_ep_resources(ep);
1953 c4iw_put_ep(&ep->com);
1957 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1959 struct sk_buff *skb;
1960 struct fw_ofld_connection_wr *req;
1961 unsigned int mtu_idx;
1963 struct sockaddr_in *sin;
1966 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1967 req = __skb_put_zero(skb, sizeof(*req));
1968 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1969 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1970 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1971 ep->com.dev->rdev.lldi.ports[0],
1973 sin = (struct sockaddr_in *)&ep->com.local_addr;
1974 req->le.lport = sin->sin_port;
1975 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1976 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1977 req->le.pport = sin->sin_port;
1978 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1979 req->tcb.t_state_to_astid =
1980 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1981 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1982 req->tcb.cplrxdataack_cplpassacceptrpl =
1983 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1984 req->tcb.tx_max = (__force __be32) jiffies;
1985 req->tcb.rcv_adv = htons(1);
1986 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1987 enable_tcp_timestamps,
1988 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1989 wscale = cxgb_compute_wscale(rcv_win);
1992 * Specify the largest window that will fit in opt0. The
1993 * remainder will be specified in the rx_data_ack.
1995 win = ep->rcv_win >> 10;
1996 if (win > RCV_BUFSIZ_M)
1999 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2000 (nocong ? NO_CONG_F : 0) |
2003 WND_SCALE_V(wscale) |
2004 MSS_IDX_V(mtu_idx) |
2005 L2T_IDX_V(ep->l2t->idx) |
2006 TX_CHAN_V(ep->tx_chan) |
2007 SMAC_SEL_V(ep->smac_idx) |
2008 DSCP_V(ep->tos >> 2) |
2009 ULP_MODE_V(ULP_MODE_TCPDDP) |
2011 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2012 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2014 CCTRL_ECN_V(enable_ecn) |
2015 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2016 if (enable_tcp_timestamps)
2017 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2018 if (enable_tcp_sack)
2019 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2020 if (wscale && enable_tcp_window_scaling)
2021 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2022 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2023 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2024 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2025 set_bit(ACT_OFLD_CONN, &ep->com.history);
2026 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2030 * Some of the error codes above implicitly indicate that there is no TID
2031 * allocated with the result of an ACT_OPEN. We use this predicate to make
2034 static inline int act_open_has_tid(int status)
2036 return (status != CPL_ERR_TCAM_PARITY &&
2037 status != CPL_ERR_TCAM_MISS &&
2038 status != CPL_ERR_TCAM_FULL &&
2039 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2040 status != CPL_ERR_CONN_EXIST);
2043 static char *neg_adv_str(unsigned int status)
2046 case CPL_ERR_RTX_NEG_ADVICE:
2047 return "Retransmit timeout";
2048 case CPL_ERR_PERSIST_NEG_ADVICE:
2049 return "Persist timeout";
2050 case CPL_ERR_KEEPALV_NEG_ADVICE:
2051 return "Keepalive timeout";
2057 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2059 ep->snd_win = snd_win;
2060 ep->rcv_win = rcv_win;
2061 pr_debug("snd_win %d rcv_win %d\n",
2062 ep->snd_win, ep->rcv_win);
2065 #define ACT_OPEN_RETRY_COUNT 2
2067 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2068 struct dst_entry *dst, struct c4iw_dev *cdev,
2069 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2071 struct neighbour *n;
2073 struct net_device *pdev;
2075 n = dst_neigh_lookup(dst, peer_ip);
2081 if (n->dev->flags & IFF_LOOPBACK) {
2083 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2084 else if (IS_ENABLED(CONFIG_IPV6))
2085 for_each_netdev(&init_net, pdev) {
2086 if (ipv6_chk_addr(&init_net,
2087 (struct in6_addr *)peer_ip,
2098 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2099 n, pdev, rt_tos2priority(tos));
2104 ep->mtu = pdev->mtu;
2105 ep->tx_chan = cxgb4_port_chan(pdev);
2106 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2107 step = cdev->rdev.lldi.ntxq /
2108 cdev->rdev.lldi.nchan;
2109 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2110 step = cdev->rdev.lldi.nrxq /
2111 cdev->rdev.lldi.nchan;
2112 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2113 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2114 cxgb4_port_idx(pdev) * step];
2115 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2118 pdev = get_real_dev(n->dev);
2119 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2120 n, pdev, rt_tos2priority(tos));
2123 ep->mtu = dst_mtu(dst);
2124 ep->tx_chan = cxgb4_port_chan(pdev);
2125 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2126 step = cdev->rdev.lldi.ntxq /
2127 cdev->rdev.lldi.nchan;
2128 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2129 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2130 step = cdev->rdev.lldi.nrxq /
2131 cdev->rdev.lldi.nchan;
2132 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2133 cxgb4_port_idx(pdev) * step];
2134 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2137 ep->retry_with_mpa_v1 = 0;
2138 ep->tried_with_mpa_v1 = 0;
2150 static int c4iw_reconnect(struct c4iw_ep *ep)
2154 struct sockaddr_in *laddr = (struct sockaddr_in *)
2155 &ep->com.cm_id->m_local_addr;
2156 struct sockaddr_in *raddr = (struct sockaddr_in *)
2157 &ep->com.cm_id->m_remote_addr;
2158 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2159 &ep->com.cm_id->m_local_addr;
2160 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2161 &ep->com.cm_id->m_remote_addr;
2165 pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2166 c4iw_init_wr_wait(ep->com.wr_waitp);
2168 /* When MPA revision is different on nodes, the node with MPA_rev=2
2169 * tries to reconnect with MPA_rev 1 for the same EP through
2170 * c4iw_reconnect(), where the same EP is assigned with new tid for
2171 * further connection establishment. As we are using the same EP pointer
2172 * for reconnect, few skbs are used during the previous c4iw_connect(),
2173 * which leaves the EP with inadequate skbs for further
2174 * c4iw_reconnect(), Further causing a crash due to an empty
2175 * skb_list() during peer_abort(). Allocate skbs which is already used.
2177 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2178 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2184 * Allocate an active TID to initiate a TCP connection.
2186 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2187 if (ep->atid == -1) {
2188 pr_err("%s - cannot alloc atid\n", __func__);
2192 err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2197 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2198 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2199 laddr->sin_addr.s_addr,
2200 raddr->sin_addr.s_addr,
2202 raddr->sin_port, ep->com.cm_id->tos);
2204 ra = (__u8 *)&raddr->sin_addr;
2206 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2208 laddr6->sin6_addr.s6_addr,
2209 raddr6->sin6_addr.s6_addr,
2213 raddr6->sin6_scope_id);
2215 ra = (__u8 *)&raddr6->sin6_addr;
2218 pr_err("%s - cannot find route\n", __func__);
2219 err = -EHOSTUNREACH;
2222 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2223 ep->com.dev->rdev.lldi.adapter_type,
2224 ep->com.cm_id->tos);
2226 pr_err("%s - cannot alloc l2e\n", __func__);
2230 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2231 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2234 state_set(&ep->com, CONNECTING);
2235 ep->tos = ep->com.cm_id->tos;
2237 /* send connect request to rnic */
2238 err = send_connect(ep);
2242 cxgb4_l2t_release(ep->l2t);
2244 dst_release(ep->dst);
2246 xa_erase_irq(&ep->com.dev->atids, ep->atid);
2248 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2251 * remember to send notification to upper layer.
2252 * We are in here so the upper layer is not aware that this is
2253 * re-connect attempt and so, upper layer is still waiting for
2254 * response of 1st connect request.
2256 connect_reply_upcall(ep, -ECONNRESET);
2258 c4iw_put_ep(&ep->com);
2263 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2266 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2267 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2268 ntohl(rpl->atid_status)));
2269 struct tid_info *t = dev->rdev.lldi.tids;
2270 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2271 struct sockaddr_in *la;
2272 struct sockaddr_in *ra;
2273 struct sockaddr_in6 *la6;
2274 struct sockaddr_in6 *ra6;
2277 ep = lookup_atid(t, atid);
2278 la = (struct sockaddr_in *)&ep->com.local_addr;
2279 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2280 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2281 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2283 pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2284 status, status2errno(status));
2286 if (cxgb_is_neg_adv(status)) {
2287 pr_debug("Connection problems for atid %u status %u (%s)\n",
2288 atid, status, neg_adv_str(status));
2289 ep->stats.connect_neg_adv++;
2290 mutex_lock(&dev->rdev.stats.lock);
2291 dev->rdev.stats.neg_adv++;
2292 mutex_unlock(&dev->rdev.stats.lock);
2296 set_bit(ACT_OPEN_RPL, &ep->com.history);
2299 * Log interesting failures.
2302 case CPL_ERR_CONN_RESET:
2303 case CPL_ERR_CONN_TIMEDOUT:
2305 case CPL_ERR_TCAM_FULL:
2306 mutex_lock(&dev->rdev.stats.lock);
2307 dev->rdev.stats.tcam_full++;
2308 mutex_unlock(&dev->rdev.stats.lock);
2309 if (ep->com.local_addr.ss_family == AF_INET &&
2310 dev->rdev.lldi.enable_fw_ofld_conn) {
2311 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2312 ntohl(rpl->atid_status))));
2318 case CPL_ERR_CONN_EXIST:
2319 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2320 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2321 if (ep->com.remote_addr.ss_family == AF_INET6) {
2322 struct sockaddr_in6 *sin6 =
2323 (struct sockaddr_in6 *)
2324 &ep->com.local_addr;
2326 ep->com.dev->rdev.lldi.ports[0],
2328 &sin6->sin6_addr.s6_addr, 1);
2330 xa_erase_irq(&ep->com.dev->atids, atid);
2331 cxgb4_free_atid(t, atid);
2332 dst_release(ep->dst);
2333 cxgb4_l2t_release(ep->l2t);
2339 if (ep->com.local_addr.ss_family == AF_INET) {
2340 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2341 atid, status, status2errno(status),
2342 &la->sin_addr.s_addr, ntohs(la->sin_port),
2343 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2345 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2346 atid, status, status2errno(status),
2347 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2348 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2354 connect_reply_upcall(ep, status2errno(status));
2355 state_set(&ep->com, DEAD);
2357 if (ep->com.remote_addr.ss_family == AF_INET6) {
2358 struct sockaddr_in6 *sin6 =
2359 (struct sockaddr_in6 *)&ep->com.local_addr;
2360 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2361 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2363 if (status && act_open_has_tid(status))
2364 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2365 ep->com.local_addr.ss_family);
2367 xa_erase_irq(&ep->com.dev->atids, atid);
2368 cxgb4_free_atid(t, atid);
2369 dst_release(ep->dst);
2370 cxgb4_l2t_release(ep->l2t);
2371 c4iw_put_ep(&ep->com);
2376 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2378 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2379 unsigned int stid = GET_TID(rpl);
2380 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2383 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2386 pr_debug("ep %p status %d error %d\n", ep,
2387 rpl->status, status2errno(rpl->status));
2388 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2389 c4iw_put_ep(&ep->com);
2394 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2396 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2397 unsigned int stid = GET_TID(rpl);
2398 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2401 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2404 pr_debug("ep %p\n", ep);
2405 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2406 c4iw_put_ep(&ep->com);
2411 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2412 struct cpl_pass_accept_req *req)
2414 struct cpl_pass_accept_rpl *rpl;
2415 unsigned int mtu_idx;
2419 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2421 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2423 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2427 if (!is_t4(adapter_type)) {
2428 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2430 INIT_TP_WR(rpl5, ep->hwtid);
2432 skb_trim(skb, sizeof(*rpl));
2433 INIT_TP_WR(rpl, ep->hwtid);
2435 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2438 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2439 enable_tcp_timestamps && req->tcpopt.tstamp,
2440 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2441 wscale = cxgb_compute_wscale(rcv_win);
2444 * Specify the largest window that will fit in opt0. The
2445 * remainder will be specified in the rx_data_ack.
2447 win = ep->rcv_win >> 10;
2448 if (win > RCV_BUFSIZ_M)
2450 opt0 = (nocong ? NO_CONG_F : 0) |
2453 WND_SCALE_V(wscale) |
2454 MSS_IDX_V(mtu_idx) |
2455 L2T_IDX_V(ep->l2t->idx) |
2456 TX_CHAN_V(ep->tx_chan) |
2457 SMAC_SEL_V(ep->smac_idx) |
2458 DSCP_V(ep->tos >> 2) |
2459 ULP_MODE_V(ULP_MODE_TCPDDP) |
2461 opt2 = RX_CHANNEL_V(0) |
2462 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2464 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2465 opt2 |= TSTAMPS_EN_F;
2466 if (enable_tcp_sack && req->tcpopt.sack)
2468 if (wscale && enable_tcp_window_scaling)
2469 opt2 |= WND_SCALE_EN_F;
2471 const struct tcphdr *tcph;
2472 u32 hlen = ntohl(req->hdr_len);
2474 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2475 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2478 tcph = (const void *)(req + 1) +
2479 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2480 if (tcph->ece && tcph->cwr)
2481 opt2 |= CCTRL_ECN_V(1);
2483 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2484 u32 isn = (prandom_u32() & ~7UL) - 1;
2485 opt2 |= T5_OPT_2_VALID_F;
2486 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2489 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2492 rpl5->iss = cpu_to_be32(isn);
2493 pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2496 rpl->opt0 = cpu_to_be64(opt0);
2497 rpl->opt2 = cpu_to_be32(opt2);
2498 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2499 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2501 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2504 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2506 pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2507 skb_trim(skb, sizeof(struct cpl_tid_release));
2508 release_tid(&dev->rdev, hwtid, skb);
2512 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2514 struct c4iw_ep *child_ep = NULL, *parent_ep;
2515 struct cpl_pass_accept_req *req = cplhdr(skb);
2516 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2517 struct tid_info *t = dev->rdev.lldi.tids;
2518 unsigned int hwtid = GET_TID(req);
2519 struct dst_entry *dst;
2520 __u8 local_ip[16], peer_ip[16];
2521 __be16 local_port, peer_port;
2522 struct sockaddr_in6 *sin6;
2524 u16 peer_mss = ntohs(req->tcpopt.mss);
2526 unsigned short hdrs;
2529 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2531 pr_err("%s connect request on invalid stid %d\n",
2536 if (state_read(&parent_ep->com) != LISTEN) {
2537 pr_err("%s - listening ep not in LISTEN\n", __func__);
2541 if (parent_ep->com.cm_id->tos_set)
2542 tos = parent_ep->com.cm_id->tos;
2544 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2546 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2547 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2549 /* Find output route */
2551 pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2553 local_ip, peer_ip, ntohs(local_port),
2554 ntohs(peer_port), peer_mss);
2555 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2556 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2557 local_port, peer_port, tos);
2559 pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2561 local_ip, peer_ip, ntohs(local_port),
2562 ntohs(peer_port), peer_mss);
2563 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2564 local_ip, peer_ip, local_port, peer_port,
2566 ((struct sockaddr_in6 *)
2567 &parent_ep->com.local_addr)->sin6_scope_id);
2570 pr_err("%s - failed to find dst entry!\n", __func__);
2574 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2576 pr_err("%s - failed to allocate ep entry!\n", __func__);
2581 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2582 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2584 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2590 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2591 sizeof(struct tcphdr) +
2592 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2593 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2594 child_ep->mtu = peer_mss + hdrs;
2596 skb_queue_head_init(&child_ep->com.ep_skb_list);
2597 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2600 state_set(&child_ep->com, CONNECTING);
2601 child_ep->com.dev = dev;
2602 child_ep->com.cm_id = NULL;
2605 struct sockaddr_in *sin = (struct sockaddr_in *)
2606 &child_ep->com.local_addr;
2608 sin->sin_family = AF_INET;
2609 sin->sin_port = local_port;
2610 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2612 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2613 sin->sin_family = AF_INET;
2614 sin->sin_port = ((struct sockaddr_in *)
2615 &parent_ep->com.local_addr)->sin_port;
2616 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2618 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2619 sin->sin_family = AF_INET;
2620 sin->sin_port = peer_port;
2621 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2623 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2624 sin6->sin6_family = PF_INET6;
2625 sin6->sin6_port = local_port;
2626 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2628 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2629 sin6->sin6_family = PF_INET6;
2630 sin6->sin6_port = ((struct sockaddr_in6 *)
2631 &parent_ep->com.local_addr)->sin6_port;
2632 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2634 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2635 sin6->sin6_family = PF_INET6;
2636 sin6->sin6_port = peer_port;
2637 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2640 c4iw_get_ep(&parent_ep->com);
2641 child_ep->parent_ep = parent_ep;
2642 child_ep->tos = tos;
2643 child_ep->dst = dst;
2644 child_ep->hwtid = hwtid;
2646 pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2647 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2649 timer_setup(&child_ep->timer, ep_timeout, 0);
2650 cxgb4_insert_tid(t, child_ep, hwtid,
2651 child_ep->com.local_addr.ss_family);
2652 insert_ep_tid(child_ep);
2653 if (accept_cr(child_ep, skb, req)) {
2654 c4iw_put_ep(&parent_ep->com);
2655 release_ep_resources(child_ep);
2657 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2660 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2661 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2662 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2666 c4iw_put_ep(&child_ep->com);
2668 reject_cr(dev, hwtid, skb);
2671 c4iw_put_ep(&parent_ep->com);
2675 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2678 struct cpl_pass_establish *req = cplhdr(skb);
2679 unsigned int tid = GET_TID(req);
2681 u16 tcp_opt = ntohs(req->tcp_opt);
2683 ep = get_ep_from_tid(dev, tid);
2684 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2685 ep->snd_seq = be32_to_cpu(req->snd_isn);
2686 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2687 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2689 pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2691 set_emss(ep, tcp_opt);
2693 dst_confirm(ep->dst);
2694 mutex_lock(&ep->com.mutex);
2695 ep->com.state = MPA_REQ_WAIT;
2697 set_bit(PASS_ESTAB, &ep->com.history);
2698 ret = send_flowc(ep);
2699 mutex_unlock(&ep->com.mutex);
2701 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2702 c4iw_put_ep(&ep->com);
2707 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2709 struct cpl_peer_close *hdr = cplhdr(skb);
2711 struct c4iw_qp_attributes attrs;
2714 unsigned int tid = GET_TID(hdr);
2717 ep = get_ep_from_tid(dev, tid);
2721 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2722 dst_confirm(ep->dst);
2724 set_bit(PEER_CLOSE, &ep->com.history);
2725 mutex_lock(&ep->com.mutex);
2726 switch (ep->com.state) {
2728 __state_set(&ep->com, CLOSING);
2731 __state_set(&ep->com, CLOSING);
2732 connect_reply_upcall(ep, -ECONNRESET);
2737 * We're gonna mark this puppy DEAD, but keep
2738 * the reference on it until the ULP accepts or
2739 * rejects the CR. Also wake up anyone waiting
2740 * in rdma connection migration (see c4iw_accept_cr()).
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);
2747 __state_set(&ep->com, CLOSING);
2748 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2749 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2753 __state_set(&ep->com, CLOSING);
2754 attrs.next_state = C4IW_QP_STATE_CLOSING;
2755 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2756 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2757 if (ret != -ECONNRESET) {
2758 peer_close_upcall(ep);
2766 __state_set(&ep->com, MORIBUND);
2770 (void)stop_ep_timer(ep);
2771 if (ep->com.cm_id && ep->com.qp) {
2772 attrs.next_state = C4IW_QP_STATE_IDLE;
2773 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2774 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2776 close_complete_upcall(ep, 0);
2777 __state_set(&ep->com, DEAD);
2785 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2787 mutex_unlock(&ep->com.mutex);
2789 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2791 release_ep_resources(ep);
2792 c4iw_put_ep(&ep->com);
2796 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2798 complete_cached_srq_buffers(ep, ep->srqe_idx);
2799 if (ep->com.cm_id && ep->com.qp) {
2800 struct c4iw_qp_attributes attrs;
2802 attrs.next_state = C4IW_QP_STATE_ERROR;
2803 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2804 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2806 peer_abort_upcall(ep);
2807 release_ep_resources(ep);
2808 c4iw_put_ep(&ep->com);
2811 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2813 struct cpl_abort_req_rss6 *req = cplhdr(skb);
2815 struct sk_buff *rpl_skb;
2816 struct c4iw_qp_attributes attrs;
2819 unsigned int tid = GET_TID(req);
2823 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2825 ep = get_ep_from_tid(dev, tid);
2829 status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2831 if (cxgb_is_neg_adv(status)) {
2832 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2833 ep->hwtid, status, neg_adv_str(status));
2834 ep->stats.abort_neg_adv++;
2835 mutex_lock(&dev->rdev.stats.lock);
2836 dev->rdev.stats.neg_adv++;
2837 mutex_unlock(&dev->rdev.stats.lock);
2841 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2843 set_bit(PEER_ABORT, &ep->com.history);
2846 * Wake up any threads in rdma_init() or rdma_fini().
2847 * However, this is not needed if com state is just
2850 if (ep->com.state != MPA_REQ_SENT)
2851 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2853 mutex_lock(&ep->com.mutex);
2854 switch (ep->com.state) {
2856 c4iw_put_ep(&ep->parent_ep->com);
2859 (void)stop_ep_timer(ep);
2862 (void)stop_ep_timer(ep);
2863 if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2864 (mpa_rev == 2 && ep->tried_with_mpa_v1))
2865 connect_reply_upcall(ep, -ECONNRESET);
2868 * we just don't send notification upwards because we
2869 * want to retry with mpa_v1 without upper layers even
2872 * do some housekeeping so as to re-initiate the
2875 pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2877 ep->retry_with_mpa_v1 = 1;
2889 if (ep->com.qp && ep->com.qp->srq) {
2890 srqidx = ABORT_RSS_SRQIDX_G(
2891 be32_to_cpu(req->srqidx_status));
2893 complete_cached_srq_buffers(ep,
2894 req->srqidx_status);
2896 /* Hold ep ref until finish_peer_abort() */
2897 c4iw_get_ep(&ep->com);
2898 __state_set(&ep->com, ABORTING);
2899 set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2906 if (ep->com.cm_id && ep->com.qp) {
2907 attrs.next_state = C4IW_QP_STATE_ERROR;
2908 ret = c4iw_modify_qp(ep->com.qp->rhp,
2909 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2912 pr_err("%s - qp <- error failed!\n", __func__);
2914 peer_abort_upcall(ep);
2919 pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2920 mutex_unlock(&ep->com.mutex);
2923 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2926 dst_confirm(ep->dst);
2927 if (ep->com.state != ABORTING) {
2928 __state_set(&ep->com, DEAD);
2929 /* we don't release if we want to retry with mpa_v1 */
2930 if (!ep->retry_with_mpa_v1)
2933 mutex_unlock(&ep->com.mutex);
2935 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2936 if (WARN_ON(!rpl_skb)) {
2941 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2943 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2946 release_ep_resources(ep);
2947 else if (ep->retry_with_mpa_v1) {
2948 if (ep->com.remote_addr.ss_family == AF_INET6) {
2949 struct sockaddr_in6 *sin6 =
2950 (struct sockaddr_in6 *)
2951 &ep->com.local_addr;
2953 ep->com.dev->rdev.lldi.ports[0],
2954 (const u32 *)&sin6->sin6_addr.s6_addr,
2957 xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2958 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2959 ep->com.local_addr.ss_family);
2960 dst_release(ep->dst);
2961 cxgb4_l2t_release(ep->l2t);
2966 c4iw_put_ep(&ep->com);
2967 /* Dereferencing ep, referenced in peer_abort_intr() */
2968 c4iw_put_ep(&ep->com);
2972 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2975 struct c4iw_qp_attributes attrs;
2976 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2978 unsigned int tid = GET_TID(rpl);
2980 ep = get_ep_from_tid(dev, tid);
2984 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2986 /* The cm_id may be null if we failed to connect */
2987 mutex_lock(&ep->com.mutex);
2988 set_bit(CLOSE_CON_RPL, &ep->com.history);
2989 switch (ep->com.state) {
2991 __state_set(&ep->com, MORIBUND);
2994 (void)stop_ep_timer(ep);
2995 if ((ep->com.cm_id) && (ep->com.qp)) {
2996 attrs.next_state = C4IW_QP_STATE_IDLE;
2997 c4iw_modify_qp(ep->com.qp->rhp,
2999 C4IW_QP_ATTR_NEXT_STATE,
3002 close_complete_upcall(ep, 0);
3003 __state_set(&ep->com, DEAD);
3010 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3013 mutex_unlock(&ep->com.mutex);
3015 release_ep_resources(ep);
3016 c4iw_put_ep(&ep->com);
3020 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3022 struct cpl_rdma_terminate *rpl = cplhdr(skb);
3023 unsigned int tid = GET_TID(rpl);
3025 struct c4iw_qp_attributes attrs;
3027 ep = get_ep_from_tid(dev, tid);
3031 pr_warn("TERM received tid %u qpid %u\n", tid,
3032 ep->com.qp->wq.sq.qid);
3033 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3034 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3035 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3038 c4iw_put_ep(&ep->com);
3040 pr_warn("TERM received tid %u no ep/qp\n", tid);
3046 * Upcall from the adapter indicating data has been transmitted.
3047 * For us its just the single MPA request or reply. We can now free
3048 * the skb holding the mpa message.
3050 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3053 struct cpl_fw4_ack *hdr = cplhdr(skb);
3054 u8 credits = hdr->credits;
3055 unsigned int tid = GET_TID(hdr);
3058 ep = get_ep_from_tid(dev, tid);
3061 pr_debug("ep %p tid %u credits %u\n",
3062 ep, ep->hwtid, credits);
3064 pr_debug("0 credit ack ep %p tid %u state %u\n",
3065 ep, ep->hwtid, state_read(&ep->com));
3069 dst_confirm(ep->dst);
3071 pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3072 ep, ep->hwtid, state_read(&ep->com),
3073 ep->mpa_attr.initiator ? 1 : 0);
3074 mutex_lock(&ep->com.mutex);
3075 kfree_skb(ep->mpa_skb);
3077 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3079 mutex_unlock(&ep->com.mutex);
3082 c4iw_put_ep(&ep->com);
3086 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3089 struct c4iw_ep *ep = to_ep(cm_id);
3091 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3093 mutex_lock(&ep->com.mutex);
3094 if (ep->com.state != MPA_REQ_RCVD) {
3095 mutex_unlock(&ep->com.mutex);
3096 c4iw_put_ep(&ep->com);
3099 set_bit(ULP_REJECT, &ep->com.history);
3103 abort = send_mpa_reject(ep, pdata, pdata_len);
3104 mutex_unlock(&ep->com.mutex);
3107 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3108 c4iw_put_ep(&ep->com);
3112 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3115 struct c4iw_qp_attributes attrs;
3116 enum c4iw_qp_attr_mask mask;
3117 struct c4iw_ep *ep = to_ep(cm_id);
3118 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3119 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3122 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3124 mutex_lock(&ep->com.mutex);
3125 if (ep->com.state != MPA_REQ_RCVD) {
3135 set_bit(ULP_ACCEPT, &ep->com.history);
3136 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3137 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3142 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3143 if (conn_param->ord > ep->ird) {
3144 if (RELAXED_IRD_NEGOTIATION) {
3145 conn_param->ord = ep->ird;
3147 ep->ird = conn_param->ird;
3148 ep->ord = conn_param->ord;
3149 send_mpa_reject(ep, conn_param->private_data,
3150 conn_param->private_data_len);
3155 if (conn_param->ird < ep->ord) {
3156 if (RELAXED_IRD_NEGOTIATION &&
3157 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3158 conn_param->ird = ep->ord;
3165 ep->ird = conn_param->ird;
3166 ep->ord = conn_param->ord;
3168 if (ep->mpa_attr.version == 1) {
3169 if (peer2peer && ep->ird == 0)
3173 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3174 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3178 pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3180 ep->com.cm_id = cm_id;
3181 ref_cm_id(&ep->com);
3185 /* bind QP to EP and move to RTS */
3186 attrs.mpa_attr = ep->mpa_attr;
3187 attrs.max_ird = ep->ird;
3188 attrs.max_ord = ep->ord;
3189 attrs.llp_stream_handle = ep;
3190 attrs.next_state = C4IW_QP_STATE_RTS;
3192 /* bind QP and TID with INIT_WR */
3193 mask = C4IW_QP_ATTR_NEXT_STATE |
3194 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3195 C4IW_QP_ATTR_MPA_ATTR |
3196 C4IW_QP_ATTR_MAX_IRD |
3197 C4IW_QP_ATTR_MAX_ORD;
3199 err = c4iw_modify_qp(ep->com.qp->rhp,
3200 ep->com.qp, mask, &attrs, 1);
3202 goto err_deref_cm_id;
3204 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3205 err = send_mpa_reply(ep, conn_param->private_data,
3206 conn_param->private_data_len);
3208 goto err_deref_cm_id;
3210 __state_set(&ep->com, FPDU_MODE);
3211 established_upcall(ep);
3212 mutex_unlock(&ep->com.mutex);
3213 c4iw_put_ep(&ep->com);
3216 deref_cm_id(&ep->com);
3220 mutex_unlock(&ep->com.mutex);
3222 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3223 c4iw_put_ep(&ep->com);
3227 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3229 struct in_device *ind;
3231 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3232 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3234 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3236 return -EADDRNOTAVAIL;
3237 for_primary_ifa(ind) {
3238 laddr->sin_addr.s_addr = ifa->ifa_address;
3239 raddr->sin_addr.s_addr = ifa->ifa_address;
3245 return found ? 0 : -EADDRNOTAVAIL;
3248 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3249 unsigned char banned_flags)
3251 struct inet6_dev *idev;
3252 int err = -EADDRNOTAVAIL;
3255 idev = __in6_dev_get(dev);
3257 struct inet6_ifaddr *ifp;
3259 read_lock_bh(&idev->lock);
3260 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3261 if (ifp->scope == IFA_LINK &&
3262 !(ifp->flags & banned_flags)) {
3263 memcpy(addr, &ifp->addr, 16);
3268 read_unlock_bh(&idev->lock);
3274 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3276 struct in6_addr uninitialized_var(addr);
3277 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3278 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3280 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3281 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3282 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3285 return -EADDRNOTAVAIL;
3288 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3290 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3293 struct sockaddr_in *laddr;
3294 struct sockaddr_in *raddr;
3295 struct sockaddr_in6 *laddr6;
3296 struct sockaddr_in6 *raddr6;
3300 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3301 (conn_param->ird > cur_max_read_depth(dev))) {
3305 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3307 pr_err("%s - cannot alloc ep\n", __func__);
3312 skb_queue_head_init(&ep->com.ep_skb_list);
3313 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3318 timer_setup(&ep->timer, ep_timeout, 0);
3319 ep->plen = conn_param->private_data_len;
3321 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3322 conn_param->private_data, ep->plen);
3323 ep->ird = conn_param->ird;
3324 ep->ord = conn_param->ord;
3326 if (peer2peer && ep->ord == 0)
3329 ep->com.cm_id = cm_id;
3330 ref_cm_id(&ep->com);
3331 cm_id->provider_data = ep;
3333 ep->com.qp = get_qhp(dev, conn_param->qpn);
3335 pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3340 pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3344 * Allocate an active TID to initiate a TCP connection.
3346 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3347 if (ep->atid == -1) {
3348 pr_err("%s - cannot alloc atid\n", __func__);
3352 err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3356 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3357 sizeof(ep->com.local_addr));
3358 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3359 sizeof(ep->com.remote_addr));
3361 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3362 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3363 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3364 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3366 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3368 ra = (__u8 *)&raddr->sin_addr;
3371 * Handle loopback requests to INADDR_ANY.
3373 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3374 err = pick_local_ipaddrs(dev, cm_id);
3380 pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3381 &laddr->sin_addr, ntohs(laddr->sin_port),
3382 ra, ntohs(raddr->sin_port));
3383 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3384 laddr->sin_addr.s_addr,
3385 raddr->sin_addr.s_addr,
3387 raddr->sin_port, cm_id->tos);
3390 ra = (__u8 *)&raddr6->sin6_addr;
3393 * Handle loopback requests to INADDR_ANY.
3395 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3396 err = pick_local_ip6addrs(dev, cm_id);
3402 pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3403 laddr6->sin6_addr.s6_addr,
3404 ntohs(laddr6->sin6_port),
3405 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3406 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3407 laddr6->sin6_addr.s6_addr,
3408 raddr6->sin6_addr.s6_addr,
3410 raddr6->sin6_port, cm_id->tos,
3411 raddr6->sin6_scope_id);
3414 pr_err("%s - cannot find route\n", __func__);
3415 err = -EHOSTUNREACH;
3419 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3420 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3422 pr_err("%s - cannot alloc l2e\n", __func__);
3426 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3427 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3430 state_set(&ep->com, CONNECTING);
3431 ep->tos = cm_id->tos;
3433 /* send connect request to rnic */
3434 err = send_connect(ep);
3438 cxgb4_l2t_release(ep->l2t);
3440 dst_release(ep->dst);
3442 xa_erase_irq(&ep->com.dev->atids, ep->atid);
3444 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3446 skb_queue_purge(&ep->com.ep_skb_list);
3447 deref_cm_id(&ep->com);
3449 c4iw_put_ep(&ep->com);
3454 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3457 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3458 &ep->com.local_addr;
3460 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3461 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3462 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3466 c4iw_init_wr_wait(ep->com.wr_waitp);
3467 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3468 ep->stid, &sin6->sin6_addr,
3470 ep->com.dev->rdev.lldi.rxq_ids[0]);
3472 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3476 err = net_xmit_errno(err);
3478 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3479 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3480 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3482 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3487 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3490 struct sockaddr_in *sin = (struct sockaddr_in *)
3491 &ep->com.local_addr;
3493 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3495 err = cxgb4_create_server_filter(
3496 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3497 sin->sin_addr.s_addr, sin->sin_port, 0,
3498 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3499 if (err == -EBUSY) {
3500 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3504 set_current_state(TASK_UNINTERRUPTIBLE);
3505 schedule_timeout(usecs_to_jiffies(100));
3507 } while (err == -EBUSY);
3509 c4iw_init_wr_wait(ep->com.wr_waitp);
3510 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3511 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3512 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3514 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3518 err = net_xmit_errno(err);
3521 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3523 &sin->sin_addr, ntohs(sin->sin_port));
3527 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3530 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3531 struct c4iw_listen_ep *ep;
3535 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3537 pr_err("%s - cannot alloc ep\n", __func__);
3541 skb_queue_head_init(&ep->com.ep_skb_list);
3542 pr_debug("ep %p\n", ep);
3543 ep->com.cm_id = cm_id;
3544 ref_cm_id(&ep->com);
3546 ep->backlog = backlog;
3547 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3548 sizeof(ep->com.local_addr));
3551 * Allocate a server TID.
3553 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3554 ep->com.local_addr.ss_family == AF_INET)
3555 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3556 cm_id->m_local_addr.ss_family, ep);
3558 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3559 cm_id->m_local_addr.ss_family, ep);
3561 if (ep->stid == -1) {
3562 pr_err("%s - cannot alloc stid\n", __func__);
3566 err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3570 state_set(&ep->com, LISTEN);
3571 if (ep->com.local_addr.ss_family == AF_INET)
3572 err = create_server4(dev, ep);
3574 err = create_server6(dev, ep);
3576 cm_id->provider_data = ep;
3579 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3581 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3582 ep->com.local_addr.ss_family);
3584 deref_cm_id(&ep->com);
3585 c4iw_put_ep(&ep->com);
3591 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3594 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3596 pr_debug("ep %p\n", ep);
3599 state_set(&ep->com, DEAD);
3600 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3601 ep->com.local_addr.ss_family == AF_INET) {
3602 err = cxgb4_remove_server_filter(
3603 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3604 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3606 struct sockaddr_in6 *sin6;
3607 c4iw_init_wr_wait(ep->com.wr_waitp);
3608 err = cxgb4_remove_server(
3609 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3610 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3613 err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3615 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3616 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3617 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3619 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3620 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3621 ep->com.local_addr.ss_family);
3623 deref_cm_id(&ep->com);
3624 c4iw_put_ep(&ep->com);
3628 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3633 struct c4iw_rdev *rdev;
3635 mutex_lock(&ep->com.mutex);
3637 pr_debug("ep %p state %s, abrupt %d\n", ep,
3638 states[ep->com.state], abrupt);
3641 * Ref the ep here in case we have fatal errors causing the
3642 * ep to be released and freed.
3644 c4iw_get_ep(&ep->com);
3646 rdev = &ep->com.dev->rdev;
3647 if (c4iw_fatal_error(rdev)) {
3649 close_complete_upcall(ep, -EIO);
3650 ep->com.state = DEAD;
3652 switch (ep->com.state) {
3661 ep->com.state = ABORTING;
3663 ep->com.state = CLOSING;
3666 * if we close before we see the fw4_ack() then we fix
3667 * up the timer state since we're reusing it.
3670 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3671 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3676 set_bit(CLOSE_SENT, &ep->com.flags);
3679 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3682 (void)stop_ep_timer(ep);
3683 ep->com.state = ABORTING;
3685 ep->com.state = MORIBUND;
3691 pr_debug("ignoring disconnect ep %p state %u\n",
3695 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3701 set_bit(EP_DISC_ABORT, &ep->com.history);
3702 ret = send_abort(ep);
3704 set_bit(EP_DISC_CLOSE, &ep->com.history);
3705 ret = send_halfclose(ep);
3708 set_bit(EP_DISC_FAIL, &ep->com.history);
3711 close_complete_upcall(ep, -EIO);
3714 struct c4iw_qp_attributes attrs;
3716 attrs.next_state = C4IW_QP_STATE_ERROR;
3717 ret = c4iw_modify_qp(ep->com.qp->rhp,
3719 C4IW_QP_ATTR_NEXT_STATE,
3722 pr_err("%s - qp <- error failed!\n",
3728 mutex_unlock(&ep->com.mutex);
3729 c4iw_put_ep(&ep->com);
3731 release_ep_resources(ep);
3735 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3736 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3739 int atid = be32_to_cpu(req->tid);
3741 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3742 (__force u32) req->tid);
3746 switch (req->retval) {
3748 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3749 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3750 send_fw_act_open_req(ep, atid);
3755 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3756 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3757 send_fw_act_open_req(ep, atid);
3762 pr_info("%s unexpected ofld conn wr retval %d\n",
3763 __func__, req->retval);
3766 pr_err("active ofld_connect_wr failure %d atid %d\n",
3768 mutex_lock(&dev->rdev.stats.lock);
3769 dev->rdev.stats.act_ofld_conn_fails++;
3770 mutex_unlock(&dev->rdev.stats.lock);
3771 connect_reply_upcall(ep, status2errno(req->retval));
3772 state_set(&ep->com, DEAD);
3773 if (ep->com.remote_addr.ss_family == AF_INET6) {
3774 struct sockaddr_in6 *sin6 =
3775 (struct sockaddr_in6 *)&ep->com.local_addr;
3776 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3777 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3779 xa_erase_irq(&dev->atids, atid);
3780 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3781 dst_release(ep->dst);
3782 cxgb4_l2t_release(ep->l2t);
3783 c4iw_put_ep(&ep->com);
3786 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3787 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3789 struct sk_buff *rpl_skb;
3790 struct cpl_pass_accept_req *cpl;
3793 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3795 pr_err("%s passive open failure %d\n", __func__, req->retval);
3796 mutex_lock(&dev->rdev.stats.lock);
3797 dev->rdev.stats.pas_ofld_conn_fails++;
3798 mutex_unlock(&dev->rdev.stats.lock);
3801 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3802 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3803 (__force u32) htonl(
3804 (__force u32) req->tid)));
3805 ret = pass_accept_req(dev, rpl_skb);
3812 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3814 u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3815 u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3819 t = (thi << shift) | (tlo >> shift);
3824 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3827 u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3831 v = (t >> shift) & mask;
3835 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3837 struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3838 __be64 *tcb = (__be64 *)(rpl + 1);
3839 unsigned int tid = GET_TID(rpl);
3844 ep = get_ep_from_tid(dev, tid);
3847 /* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3848 * determine if there's a rx PDU feedback event pending.
3850 * If that bit is set, it means we'll need to re-read the TCB's
3851 * rq_start value. The final value is the one present in a TCB
3852 * with the TF_RX_PDU_OUT bit cleared.
3855 t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3856 rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3858 c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3859 c4iw_put_ep(&ep->com); /* from read_tcb() */
3861 /* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3863 if (++ep->rx_pdu_out_cnt >= 2) {
3864 WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3871 ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_W,
3874 pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3876 if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3877 finish_peer_abort(dev, ep);
3878 else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3881 WARN_ONCE(1, "unexpected state!");
3886 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3888 struct cpl_fw6_msg *rpl = cplhdr(skb);
3889 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3891 switch (rpl->type) {
3893 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3895 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3896 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3897 switch (req->t_state) {
3899 active_ofld_conn_reply(dev, skb, req);
3902 passive_ofld_conn_reply(dev, skb, req);
3905 pr_err("%s unexpected ofld conn wr state %d\n",
3906 __func__, req->t_state);
3914 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3917 __be16 hdr_len, vlantag, len;
3919 int tcp_hdr_len, ip_hdr_len;
3921 struct cpl_rx_pkt *cpl = cplhdr(skb);
3922 struct cpl_pass_accept_req *req;
3923 struct tcp_options_received tmp_opt;
3924 struct c4iw_dev *dev;
3925 enum chip_type type;
3927 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3928 /* Store values from cpl_rx_pkt in temporary location. */
3929 vlantag = cpl->vlan;
3931 l2info = cpl->l2info;
3932 hdr_len = cpl->hdr_len;
3935 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3938 * We need to parse the TCP options from SYN packet.
3939 * to generate cpl_pass_accept_req.
3941 memset(&tmp_opt, 0, sizeof(tmp_opt));
3942 tcp_clear_options(&tmp_opt);
3943 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3945 req = __skb_push(skb, sizeof(*req));
3946 memset(req, 0, sizeof(*req));
3947 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3948 SYN_MAC_IDX_V(RX_MACIDX_G(
3949 be32_to_cpu(l2info))) |
3951 type = dev->rdev.lldi.adapter_type;
3952 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3953 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3955 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3956 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3957 eth_hdr_len = is_t4(type) ?
3958 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3959 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3960 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3961 IP_HDR_LEN_V(ip_hdr_len) |
3962 ETH_HDR_LEN_V(eth_hdr_len));
3963 } else { /* T6 and later */
3964 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3965 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3966 T6_IP_HDR_LEN_V(ip_hdr_len) |
3967 T6_ETH_HDR_LEN_V(eth_hdr_len));
3969 req->vlan = vlantag;
3971 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3972 PASS_OPEN_TOS_V(tos));
3973 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3974 if (tmp_opt.wscale_ok)
3975 req->tcpopt.wsf = tmp_opt.snd_wscale;
3976 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3977 if (tmp_opt.sack_ok)
3978 req->tcpopt.sack = 1;
3979 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3983 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3984 __be32 laddr, __be16 lport,
3985 __be32 raddr, __be16 rport,
3986 u32 rcv_isn, u32 filter, u16 window,
3987 u32 rss_qid, u8 port_id)
3989 struct sk_buff *req_skb;
3990 struct fw_ofld_connection_wr *req;
3991 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3994 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3997 req = __skb_put_zero(req_skb, sizeof(*req));
3998 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3999 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4000 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4001 req->le.filter = (__force __be32) filter;
4002 req->le.lport = lport;
4003 req->le.pport = rport;
4004 req->le.u.ipv4.lip = laddr;
4005 req->le.u.ipv4.pip = raddr;
4006 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4007 req->tcb.rcv_adv = htons(window);
4008 req->tcb.t_state_to_astid =
4009 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4010 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4011 FW_OFLD_CONNECTION_WR_ASTID_V(
4012 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4015 * We store the qid in opt2 which will be used by the firmware
4016 * to send us the wr response.
4018 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4021 * We initialize the MSS index in TCB to 0xF.
4022 * So that when driver sends cpl_pass_accept_rpl
4023 * TCB picks up the correct value. If this was 0
4024 * TP will ignore any value > 0 for MSS index.
4026 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4027 req->cookie = (uintptr_t)skb;
4029 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4030 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4032 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4040 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4041 * messages when a filter is being used instead of server to
4042 * redirect a syn packet. When packets hit filter they are redirected
4043 * to the offload queue and driver tries to establish the connection
4044 * using firmware work request.
4046 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4049 unsigned int filter;
4050 struct ethhdr *eh = NULL;
4051 struct vlan_ethhdr *vlan_eh = NULL;
4053 struct tcphdr *tcph;
4054 struct rss_header *rss = (void *)skb->data;
4055 struct cpl_rx_pkt *cpl = (void *)skb->data;
4056 struct cpl_pass_accept_req *req = (void *)(rss + 1);
4057 struct l2t_entry *e;
4058 struct dst_entry *dst;
4059 struct c4iw_ep *lep = NULL;
4061 struct port_info *pi;
4062 struct net_device *pdev;
4063 u16 rss_qid, eth_hdr_len;
4065 struct neighbour *neigh;
4067 /* Drop all non-SYN packets */
4068 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4072 * Drop all packets which did not hit the filter.
4073 * Unlikely to happen.
4075 if (!(rss->filter_hit && rss->filter_tid))
4079 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4081 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4083 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4085 pr_warn("%s connect request on invalid stid %d\n",
4090 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4092 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4095 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4098 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4101 pr_err("T%d Chip is not supported\n",
4102 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4106 if (eth_hdr_len == ETH_HLEN) {
4107 eh = (struct ethhdr *)(req + 1);
4108 iph = (struct iphdr *)(eh + 1);
4110 vlan_eh = (struct vlan_ethhdr *)(req + 1);
4111 iph = (struct iphdr *)(vlan_eh + 1);
4112 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4115 if (iph->version != 0x4)
4118 tcph = (struct tcphdr *)(iph + 1);
4119 skb_set_network_header(skb, (void *)iph - (void *)rss);
4120 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4123 pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4124 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4125 ntohs(tcph->source), iph->tos);
4127 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4128 iph->daddr, iph->saddr, tcph->dest,
4129 tcph->source, iph->tos);
4131 pr_err("%s - failed to find dst entry!\n", __func__);
4134 neigh = dst_neigh_lookup_skb(dst, skb);
4137 pr_err("%s - failed to allocate neigh!\n", __func__);
4141 if (neigh->dev->flags & IFF_LOOPBACK) {
4142 pdev = ip_dev_find(&init_net, iph->daddr);
4143 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4145 pi = (struct port_info *)netdev_priv(pdev);
4148 pdev = get_real_dev(neigh->dev);
4149 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4151 pi = (struct port_info *)netdev_priv(pdev);
4153 neigh_release(neigh);
4155 pr_err("%s - failed to allocate l2t entry!\n",
4160 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4161 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4162 window = (__force u16) htons((__force u16)tcph->window);
4164 /* Calcuate filter portion for LE region. */
4165 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4166 dev->rdev.lldi.ports[0],
4170 * Synthesize the cpl_pass_accept_req. We have everything except the
4171 * TID. Once firmware sends a reply with TID we update the TID field
4172 * in cpl and pass it through the regular cpl_pass_accept_req path.
4174 build_cpl_pass_accept_req(skb, stid, iph->tos);
4175 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4176 tcph->source, ntohl(tcph->seq), filter, window,
4177 rss_qid, pi->port_id);
4178 cxgb4_l2t_release(e);
4183 c4iw_put_ep(&lep->com);
4188 * These are the real handlers that are called from a
4191 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4192 [CPL_ACT_ESTABLISH] = act_establish,
4193 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4194 [CPL_RX_DATA] = rx_data,
4195 [CPL_ABORT_RPL_RSS] = abort_rpl,
4196 [CPL_ABORT_RPL] = abort_rpl,
4197 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4198 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4199 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4200 [CPL_PASS_ESTABLISH] = pass_establish,
4201 [CPL_PEER_CLOSE] = peer_close,
4202 [CPL_ABORT_REQ_RSS] = peer_abort,
4203 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4204 [CPL_RDMA_TERMINATE] = terminate,
4205 [CPL_FW4_ACK] = fw4_ack,
4206 [CPL_GET_TCB_RPL] = read_tcb_rpl,
4207 [CPL_FW6_MSG] = deferred_fw6_msg,
4208 [CPL_RX_PKT] = rx_pkt,
4209 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4210 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4213 static void process_timeout(struct c4iw_ep *ep)
4215 struct c4iw_qp_attributes attrs;
4218 mutex_lock(&ep->com.mutex);
4219 pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4220 set_bit(TIMEDOUT, &ep->com.history);
4221 switch (ep->com.state) {
4223 connect_reply_upcall(ep, -ETIMEDOUT);
4232 if (ep->com.cm_id && ep->com.qp) {
4233 attrs.next_state = C4IW_QP_STATE_ERROR;
4234 c4iw_modify_qp(ep->com.qp->rhp,
4235 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4238 close_complete_upcall(ep, -ETIMEDOUT);
4244 * These states are expected if the ep timed out at the same
4245 * time as another thread was calling stop_ep_timer().
4246 * So we silently do nothing for these states.
4251 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4252 __func__, ep, ep->hwtid, ep->com.state);
4255 mutex_unlock(&ep->com.mutex);
4257 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4258 c4iw_put_ep(&ep->com);
4261 static void process_timedout_eps(void)
4265 spin_lock_irq(&timeout_lock);
4266 while (!list_empty(&timeout_list)) {
4267 struct list_head *tmp;
4269 tmp = timeout_list.next;
4273 spin_unlock_irq(&timeout_lock);
4274 ep = list_entry(tmp, struct c4iw_ep, entry);
4275 process_timeout(ep);
4276 spin_lock_irq(&timeout_lock);
4278 spin_unlock_irq(&timeout_lock);
4281 static void process_work(struct work_struct *work)
4283 struct sk_buff *skb = NULL;
4284 struct c4iw_dev *dev;
4285 struct cpl_act_establish *rpl;
4286 unsigned int opcode;
4289 process_timedout_eps();
4290 while ((skb = skb_dequeue(&rxq))) {
4292 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4293 opcode = rpl->ot.opcode;
4295 if (opcode >= ARRAY_SIZE(work_handlers) ||
4296 !work_handlers[opcode]) {
4297 pr_err("No handler for opcode 0x%x.\n", opcode);
4300 ret = work_handlers[opcode](dev, skb);
4304 process_timedout_eps();
4308 static DECLARE_WORK(skb_work, process_work);
4310 static void ep_timeout(struct timer_list *t)
4312 struct c4iw_ep *ep = from_timer(ep, t, timer);
4315 spin_lock(&timeout_lock);
4316 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4318 * Only insert if it is not already on the list.
4320 if (!ep->entry.next) {
4321 list_add_tail(&ep->entry, &timeout_list);
4325 spin_unlock(&timeout_lock);
4327 queue_work(workq, &skb_work);
4331 * All the CM events are handled on a work queue to have a safe context.
4333 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4337 * Save dev in the skb->cb area.
4339 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4342 * Queue the skb and schedule the worker thread.
4344 skb_queue_tail(&rxq, skb);
4345 queue_work(workq, &skb_work);
4349 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4351 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4353 if (rpl->status != CPL_ERR_NONE) {
4354 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4355 rpl->status, GET_TID(rpl));
4361 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4363 struct cpl_fw6_msg *rpl = cplhdr(skb);
4364 struct c4iw_wr_wait *wr_waitp;
4367 pr_debug("type %u\n", rpl->type);
4369 switch (rpl->type) {
4370 case FW6_TYPE_WR_RPL:
4371 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4372 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4373 pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4375 c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4379 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4383 pr_err("%s unexpected fw6 msg type %u\n",
4384 __func__, rpl->type);
4391 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4393 struct cpl_abort_req_rss *req = cplhdr(skb);
4395 unsigned int tid = GET_TID(req);
4397 ep = get_ep_from_tid(dev, tid);
4398 /* This EP will be dereferenced in peer_abort() */
4400 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4404 if (cxgb_is_neg_adv(req->status)) {
4405 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4406 ep->hwtid, req->status,
4407 neg_adv_str(req->status));
4410 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4412 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4419 * Most upcalls from the T4 Core go to sched() to
4420 * schedule the processing on a work queue.
4422 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4423 [CPL_ACT_ESTABLISH] = sched,
4424 [CPL_ACT_OPEN_RPL] = sched,
4425 [CPL_RX_DATA] = sched,
4426 [CPL_ABORT_RPL_RSS] = sched,
4427 [CPL_ABORT_RPL] = sched,
4428 [CPL_PASS_OPEN_RPL] = sched,
4429 [CPL_CLOSE_LISTSRV_RPL] = sched,
4430 [CPL_PASS_ACCEPT_REQ] = sched,
4431 [CPL_PASS_ESTABLISH] = sched,
4432 [CPL_PEER_CLOSE] = sched,
4433 [CPL_CLOSE_CON_RPL] = sched,
4434 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4435 [CPL_RDMA_TERMINATE] = sched,
4436 [CPL_FW4_ACK] = sched,
4437 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4438 [CPL_GET_TCB_RPL] = sched,
4439 [CPL_FW6_MSG] = fw6_msg,
4440 [CPL_RX_PKT] = sched
4443 int __init c4iw_cm_init(void)
4445 spin_lock_init(&timeout_lock);
4446 skb_queue_head_init(&rxq);
4448 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4455 void c4iw_cm_term(void)
4457 WARN_ON(!list_empty(&timeout_list));
4458 flush_workqueue(workq);
4459 destroy_workqueue(workq);