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 spin_lock_irqsave(&ep->com.dev->lock, flags);
335 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
336 if (idr_is_empty(&ep->com.dev->hwtid_idr))
337 wake_up(&ep->com.dev->wait);
338 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
341 static void insert_ep_tid(struct c4iw_ep *ep)
345 spin_lock_irqsave(&ep->com.dev->lock, flags);
346 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
347 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
351 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
353 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
358 spin_lock_irqsave(&dev->lock, flags);
359 ep = idr_find(&dev->hwtid_idr, tid);
361 c4iw_get_ep(&ep->com);
362 spin_unlock_irqrestore(&dev->lock, flags);
367 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
369 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
372 struct c4iw_listen_ep *ep;
375 spin_lock_irqsave(&dev->lock, flags);
376 ep = idr_find(&dev->stid_idr, stid);
378 c4iw_get_ep(&ep->com);
379 spin_unlock_irqrestore(&dev->lock, flags);
383 void _c4iw_free_ep(struct kref *kref)
387 ep = container_of(kref, struct c4iw_ep, com.kref);
388 pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
389 if (test_bit(QP_REFERENCED, &ep->com.flags))
391 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
392 if (ep->com.remote_addr.ss_family == AF_INET6) {
393 struct sockaddr_in6 *sin6 =
394 (struct sockaddr_in6 *)
398 ep->com.dev->rdev.lldi.ports[0],
399 (const u32 *)&sin6->sin6_addr.s6_addr,
402 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
403 ep->com.local_addr.ss_family);
404 dst_release(ep->dst);
405 cxgb4_l2t_release(ep->l2t);
407 kfree_skb(ep->mpa_skb);
409 if (!skb_queue_empty(&ep->com.ep_skb_list))
410 skb_queue_purge(&ep->com.ep_skb_list);
411 c4iw_put_wr_wait(ep->com.wr_waitp);
415 static void release_ep_resources(struct c4iw_ep *ep)
417 set_bit(RELEASE_RESOURCES, &ep->com.flags);
420 * If we have a hwtid, then remove it from the idr table
421 * so lookups will no longer find this endpoint. Otherwise
422 * we have a race where one thread finds the ep ptr just
423 * before the other thread is freeing the ep memory.
427 c4iw_put_ep(&ep->com);
430 static int status2errno(int status)
435 case CPL_ERR_CONN_RESET:
437 case CPL_ERR_ARP_MISS:
438 return -EHOSTUNREACH;
439 case CPL_ERR_CONN_TIMEDOUT:
441 case CPL_ERR_TCAM_FULL:
443 case CPL_ERR_CONN_EXIST:
451 * Try and reuse skbs already allocated...
453 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
455 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
458 skb_reset_transport_header(skb);
460 skb = alloc_skb(len, gfp);
462 t4_set_arp_err_handler(skb, NULL, NULL);
466 static struct net_device *get_real_dev(struct net_device *egress_dev)
468 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
471 static void arp_failure_discard(void *handle, struct sk_buff *skb)
473 pr_err("ARP failure\n");
477 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
479 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
484 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
485 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
488 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
492 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
493 release_ep_resources(ep);
498 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
502 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
503 c4iw_put_ep(&ep->parent_ep->com);
504 release_ep_resources(ep);
510 * Fake up a special CPL opcode and call sched() so process_work() will call
511 * _put_ep_safe() in a safe context to free the ep resources. This is needed
512 * because ARP error handlers are called in an ATOMIC context, and
513 * _c4iw_free_ep() needs to block.
515 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
518 struct cpl_act_establish *rpl = cplhdr(skb);
520 /* Set our special ARP_FAILURE opcode */
521 rpl->ot.opcode = cpl;
524 * Save ep in the skb->cb area, after where sched() will save the dev
527 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
528 sched(ep->com.dev, skb);
531 /* Handle an ARP failure for an accept */
532 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
534 struct c4iw_ep *ep = handle;
536 pr_err("ARP failure during accept - tid %u - dropping connection\n",
539 __state_set(&ep->com, DEAD);
540 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
544 * Handle an ARP failure for an active open.
546 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
548 struct c4iw_ep *ep = handle;
550 pr_err("ARP failure during connect\n");
551 connect_reply_upcall(ep, -EHOSTUNREACH);
552 __state_set(&ep->com, DEAD);
553 if (ep->com.remote_addr.ss_family == AF_INET6) {
554 struct sockaddr_in6 *sin6 =
555 (struct sockaddr_in6 *)&ep->com.local_addr;
556 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
557 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
559 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
560 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
561 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
565 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
568 static void abort_arp_failure(void *handle, struct sk_buff *skb)
571 struct c4iw_ep *ep = handle;
572 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
573 struct cpl_abort_req *req = cplhdr(skb);
575 pr_debug("rdev %p\n", rdev);
576 req->cmd = CPL_ABORT_NO_RST;
578 ret = c4iw_ofld_send(rdev, skb);
580 __state_set(&ep->com, DEAD);
581 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
586 static int send_flowc(struct c4iw_ep *ep)
588 struct fw_flowc_wr *flowc;
589 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
591 u16 vlan = ep->l2t->vlan;
597 if (vlan == CPL_L2T_VLAN_NONE)
602 flowc = __skb_put(skb, FLOWC_LEN);
604 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
605 FW_FLOWC_WR_NPARAMS_V(nparams));
606 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
607 16)) | FW_WR_FLOWID_V(ep->hwtid));
609 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
610 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
611 (ep->com.dev->rdev.lldi.pf));
612 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
613 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
614 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
615 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
616 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
617 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
618 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
619 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
620 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
621 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
622 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
623 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
624 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
625 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
629 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
630 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
631 flowc->mnemval[8].val = cpu_to_be32(pri);
633 /* Pad WR to 16 byte boundary */
634 flowc->mnemval[8].mnemonic = 0;
635 flowc->mnemval[8].val = 0;
637 for (i = 0; i < 9; i++) {
638 flowc->mnemval[i].r4[0] = 0;
639 flowc->mnemval[i].r4[1] = 0;
640 flowc->mnemval[i].r4[2] = 0;
643 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
644 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
647 static int send_halfclose(struct c4iw_ep *ep)
649 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
650 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
652 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
656 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
657 NULL, arp_failure_discard);
659 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
662 static int send_abort(struct c4iw_ep *ep)
664 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
665 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
667 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
668 if (WARN_ON(!req_skb))
671 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
672 ep, abort_arp_failure);
674 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
677 static int send_connect(struct c4iw_ep *ep)
679 struct cpl_act_open_req *req = NULL;
680 struct cpl_t5_act_open_req *t5req = NULL;
681 struct cpl_t6_act_open_req *t6req = NULL;
682 struct cpl_act_open_req6 *req6 = NULL;
683 struct cpl_t5_act_open_req6 *t5req6 = NULL;
684 struct cpl_t6_act_open_req6 *t6req6 = NULL;
688 unsigned int mtu_idx;
690 int win, sizev4, sizev6, wrlen;
691 struct sockaddr_in *la = (struct sockaddr_in *)
693 struct sockaddr_in *ra = (struct sockaddr_in *)
694 &ep->com.remote_addr;
695 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
697 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
698 &ep->com.remote_addr;
700 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
701 u32 isn = (prandom_u32() & ~7UL) - 1;
702 struct net_device *netdev;
705 netdev = ep->com.dev->rdev.lldi.ports[0];
707 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
709 sizev4 = sizeof(struct cpl_act_open_req);
710 sizev6 = sizeof(struct cpl_act_open_req6);
713 sizev4 = sizeof(struct cpl_t5_act_open_req);
714 sizev6 = sizeof(struct cpl_t5_act_open_req6);
717 sizev4 = sizeof(struct cpl_t6_act_open_req);
718 sizev6 = sizeof(struct cpl_t6_act_open_req6);
721 pr_err("T%d Chip is not supported\n",
722 CHELSIO_CHIP_VERSION(adapter_type));
726 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
727 roundup(sizev4, 16) :
730 pr_debug("ep %p atid %u\n", ep, ep->atid);
732 skb = get_skb(NULL, wrlen, GFP_KERNEL);
734 pr_err("%s - failed to alloc skb\n", __func__);
737 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
739 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
740 enable_tcp_timestamps,
741 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
742 wscale = cxgb_compute_wscale(rcv_win);
745 * Specify the largest window that will fit in opt0. The
746 * remainder will be specified in the rx_data_ack.
748 win = ep->rcv_win >> 10;
749 if (win > RCV_BUFSIZ_M)
752 opt0 = (nocong ? NO_CONG_F : 0) |
755 WND_SCALE_V(wscale) |
757 L2T_IDX_V(ep->l2t->idx) |
758 TX_CHAN_V(ep->tx_chan) |
759 SMAC_SEL_V(ep->smac_idx) |
760 DSCP_V(ep->tos >> 2) |
761 ULP_MODE_V(ULP_MODE_TCPDDP) |
763 opt2 = RX_CHANNEL_V(0) |
764 CCTRL_ECN_V(enable_ecn) |
765 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
766 if (enable_tcp_timestamps)
767 opt2 |= TSTAMPS_EN_F;
770 if (wscale && enable_tcp_window_scaling)
771 opt2 |= WND_SCALE_EN_F;
772 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
776 opt2 |= T5_OPT_2_VALID_F;
777 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
781 params = cxgb4_select_ntuple(netdev, ep->l2t);
783 if (ep->com.remote_addr.ss_family == AF_INET6)
784 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
785 (const u32 *)&la6->sin6_addr.s6_addr, 1);
787 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
789 if (ep->com.remote_addr.ss_family == AF_INET) {
790 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
792 req = skb_put(skb, wrlen);
796 t5req = skb_put(skb, wrlen);
797 INIT_TP_WR(t5req, 0);
798 req = (struct cpl_act_open_req *)t5req;
801 t6req = skb_put(skb, wrlen);
802 INIT_TP_WR(t6req, 0);
803 req = (struct cpl_act_open_req *)t6req;
804 t5req = (struct cpl_t5_act_open_req *)t6req;
807 pr_err("T%d Chip is not supported\n",
808 CHELSIO_CHIP_VERSION(adapter_type));
813 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
814 ((ep->rss_qid<<14) | ep->atid)));
815 req->local_port = la->sin_port;
816 req->peer_port = ra->sin_port;
817 req->local_ip = la->sin_addr.s_addr;
818 req->peer_ip = ra->sin_addr.s_addr;
819 req->opt0 = cpu_to_be64(opt0);
821 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
822 req->params = cpu_to_be32(params);
823 req->opt2 = cpu_to_be32(opt2);
825 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
827 cpu_to_be64(FILTER_TUPLE_V(params));
828 t5req->rsvd = cpu_to_be32(isn);
829 pr_debug("snd_isn %u\n", t5req->rsvd);
830 t5req->opt2 = cpu_to_be32(opt2);
833 cpu_to_be64(FILTER_TUPLE_V(params));
834 t6req->rsvd = cpu_to_be32(isn);
835 pr_debug("snd_isn %u\n", t6req->rsvd);
836 t6req->opt2 = cpu_to_be32(opt2);
840 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
842 req6 = skb_put(skb, wrlen);
846 t5req6 = skb_put(skb, wrlen);
847 INIT_TP_WR(t5req6, 0);
848 req6 = (struct cpl_act_open_req6 *)t5req6;
851 t6req6 = skb_put(skb, wrlen);
852 INIT_TP_WR(t6req6, 0);
853 req6 = (struct cpl_act_open_req6 *)t6req6;
854 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
857 pr_err("T%d Chip is not supported\n",
858 CHELSIO_CHIP_VERSION(adapter_type));
863 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
864 ((ep->rss_qid<<14)|ep->atid)));
865 req6->local_port = la6->sin6_port;
866 req6->peer_port = ra6->sin6_port;
867 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
868 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
869 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
870 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
871 req6->opt0 = cpu_to_be64(opt0);
873 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
874 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
876 req6->opt2 = cpu_to_be32(opt2);
878 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
880 cpu_to_be64(FILTER_TUPLE_V(params));
881 t5req6->rsvd = cpu_to_be32(isn);
882 pr_debug("snd_isn %u\n", t5req6->rsvd);
883 t5req6->opt2 = cpu_to_be32(opt2);
886 cpu_to_be64(FILTER_TUPLE_V(params));
887 t6req6->rsvd = cpu_to_be32(isn);
888 pr_debug("snd_isn %u\n", t6req6->rsvd);
889 t6req6->opt2 = cpu_to_be32(opt2);
895 set_bit(ACT_OPEN_REQ, &ep->com.history);
896 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
898 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
899 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
900 (const u32 *)&la6->sin6_addr.s6_addr, 1);
904 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
907 int mpalen, wrlen, ret;
908 struct fw_ofld_tx_data_wr *req;
909 struct mpa_message *mpa;
910 struct mpa_v2_conn_params mpa_v2_params;
912 pr_debug("ep %p tid %u pd_len %d\n",
913 ep, ep->hwtid, ep->plen);
915 mpalen = sizeof(*mpa) + ep->plen;
916 if (mpa_rev_to_use == 2)
917 mpalen += sizeof(struct mpa_v2_conn_params);
918 wrlen = roundup(mpalen + sizeof *req, 16);
919 skb = get_skb(skb, wrlen, GFP_KERNEL);
921 connect_reply_upcall(ep, -ENOMEM);
924 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
926 req = skb_put_zero(skb, wrlen);
927 req->op_to_immdlen = cpu_to_be32(
928 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
930 FW_WR_IMMDLEN_V(mpalen));
931 req->flowid_len16 = cpu_to_be32(
932 FW_WR_FLOWID_V(ep->hwtid) |
933 FW_WR_LEN16_V(wrlen >> 4));
934 req->plen = cpu_to_be32(mpalen);
935 req->tunnel_to_proxy = cpu_to_be32(
936 FW_OFLD_TX_DATA_WR_FLUSH_F |
937 FW_OFLD_TX_DATA_WR_SHOVE_F);
939 mpa = (struct mpa_message *)(req + 1);
940 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
944 mpa->flags |= MPA_CRC;
945 if (markers_enabled) {
946 mpa->flags |= MPA_MARKERS;
947 ep->mpa_attr.recv_marker_enabled = 1;
949 ep->mpa_attr.recv_marker_enabled = 0;
951 if (mpa_rev_to_use == 2)
952 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
954 mpa->private_data_size = htons(ep->plen);
955 mpa->revision = mpa_rev_to_use;
956 if (mpa_rev_to_use == 1) {
957 ep->tried_with_mpa_v1 = 1;
958 ep->retry_with_mpa_v1 = 0;
961 if (mpa_rev_to_use == 2) {
962 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
963 sizeof (struct mpa_v2_conn_params));
964 pr_debug("initiator ird %u ord %u\n", ep->ird,
966 mpa_v2_params.ird = htons((u16)ep->ird);
967 mpa_v2_params.ord = htons((u16)ep->ord);
970 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
971 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
973 htons(MPA_V2_RDMA_WRITE_RTR);
974 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
976 htons(MPA_V2_RDMA_READ_RTR);
978 memcpy(mpa->private_data, &mpa_v2_params,
979 sizeof(struct mpa_v2_conn_params));
982 memcpy(mpa->private_data +
983 sizeof(struct mpa_v2_conn_params),
984 ep->mpa_pkt + sizeof(*mpa), ep->plen);
987 memcpy(mpa->private_data,
988 ep->mpa_pkt + sizeof(*mpa), ep->plen);
991 * Reference the mpa skb. This ensures the data area
992 * will remain in memory until the hw acks the tx.
993 * Function fw4_ack() will deref it.
996 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
998 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1002 __state_set(&ep->com, MPA_REQ_SENT);
1003 ep->mpa_attr.initiator = 1;
1004 ep->snd_seq += mpalen;
1008 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1011 struct fw_ofld_tx_data_wr *req;
1012 struct mpa_message *mpa;
1013 struct sk_buff *skb;
1014 struct mpa_v2_conn_params mpa_v2_params;
1016 pr_debug("ep %p tid %u pd_len %d\n",
1017 ep, ep->hwtid, ep->plen);
1019 mpalen = sizeof(*mpa) + plen;
1020 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1021 mpalen += sizeof(struct mpa_v2_conn_params);
1022 wrlen = roundup(mpalen + sizeof *req, 16);
1024 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1026 pr_err("%s - cannot alloc skb!\n", __func__);
1029 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1031 req = skb_put_zero(skb, wrlen);
1032 req->op_to_immdlen = cpu_to_be32(
1033 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1035 FW_WR_IMMDLEN_V(mpalen));
1036 req->flowid_len16 = cpu_to_be32(
1037 FW_WR_FLOWID_V(ep->hwtid) |
1038 FW_WR_LEN16_V(wrlen >> 4));
1039 req->plen = cpu_to_be32(mpalen);
1040 req->tunnel_to_proxy = cpu_to_be32(
1041 FW_OFLD_TX_DATA_WR_FLUSH_F |
1042 FW_OFLD_TX_DATA_WR_SHOVE_F);
1044 mpa = (struct mpa_message *)(req + 1);
1045 memset(mpa, 0, sizeof(*mpa));
1046 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1047 mpa->flags = MPA_REJECT;
1048 mpa->revision = ep->mpa_attr.version;
1049 mpa->private_data_size = htons(plen);
1051 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1052 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1053 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1054 sizeof (struct mpa_v2_conn_params));
1055 mpa_v2_params.ird = htons(((u16)ep->ird) |
1056 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1058 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1060 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1061 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1062 FW_RI_INIT_P2PTYPE_READ_REQ ?
1063 MPA_V2_RDMA_READ_RTR : 0) : 0));
1064 memcpy(mpa->private_data, &mpa_v2_params,
1065 sizeof(struct mpa_v2_conn_params));
1068 memcpy(mpa->private_data +
1069 sizeof(struct mpa_v2_conn_params), pdata, plen);
1072 memcpy(mpa->private_data, pdata, plen);
1075 * Reference the mpa skb again. This ensures the data area
1076 * will remain in memory until the hw acks the tx.
1077 * Function fw4_ack() will deref it.
1080 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1081 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1083 ep->snd_seq += mpalen;
1084 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1087 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1090 struct fw_ofld_tx_data_wr *req;
1091 struct mpa_message *mpa;
1092 struct sk_buff *skb;
1093 struct mpa_v2_conn_params mpa_v2_params;
1095 pr_debug("ep %p tid %u pd_len %d\n",
1096 ep, ep->hwtid, ep->plen);
1098 mpalen = sizeof(*mpa) + plen;
1099 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1100 mpalen += sizeof(struct mpa_v2_conn_params);
1101 wrlen = roundup(mpalen + sizeof *req, 16);
1103 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1105 pr_err("%s - cannot alloc skb!\n", __func__);
1108 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1110 req = skb_put_zero(skb, wrlen);
1111 req->op_to_immdlen = cpu_to_be32(
1112 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1114 FW_WR_IMMDLEN_V(mpalen));
1115 req->flowid_len16 = cpu_to_be32(
1116 FW_WR_FLOWID_V(ep->hwtid) |
1117 FW_WR_LEN16_V(wrlen >> 4));
1118 req->plen = cpu_to_be32(mpalen);
1119 req->tunnel_to_proxy = cpu_to_be32(
1120 FW_OFLD_TX_DATA_WR_FLUSH_F |
1121 FW_OFLD_TX_DATA_WR_SHOVE_F);
1123 mpa = (struct mpa_message *)(req + 1);
1124 memset(mpa, 0, sizeof(*mpa));
1125 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1127 if (ep->mpa_attr.crc_enabled)
1128 mpa->flags |= MPA_CRC;
1129 if (ep->mpa_attr.recv_marker_enabled)
1130 mpa->flags |= MPA_MARKERS;
1131 mpa->revision = ep->mpa_attr.version;
1132 mpa->private_data_size = htons(plen);
1134 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1135 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1136 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1137 sizeof (struct mpa_v2_conn_params));
1138 mpa_v2_params.ird = htons((u16)ep->ird);
1139 mpa_v2_params.ord = htons((u16)ep->ord);
1140 if (peer2peer && (ep->mpa_attr.p2p_type !=
1141 FW_RI_INIT_P2PTYPE_DISABLED)) {
1142 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1144 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1145 mpa_v2_params.ord |=
1146 htons(MPA_V2_RDMA_WRITE_RTR);
1147 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1148 mpa_v2_params.ord |=
1149 htons(MPA_V2_RDMA_READ_RTR);
1152 memcpy(mpa->private_data, &mpa_v2_params,
1153 sizeof(struct mpa_v2_conn_params));
1156 memcpy(mpa->private_data +
1157 sizeof(struct mpa_v2_conn_params), pdata, plen);
1160 memcpy(mpa->private_data, pdata, plen);
1163 * Reference the mpa skb. This ensures the data area
1164 * will remain in memory until the hw acks the tx.
1165 * Function fw4_ack() will deref it.
1168 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1170 __state_set(&ep->com, MPA_REP_SENT);
1171 ep->snd_seq += mpalen;
1172 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1175 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1178 struct cpl_act_establish *req = cplhdr(skb);
1179 unsigned int tid = GET_TID(req);
1180 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1181 struct tid_info *t = dev->rdev.lldi.tids;
1184 ep = lookup_atid(t, atid);
1186 pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1187 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1189 mutex_lock(&ep->com.mutex);
1190 dst_confirm(ep->dst);
1192 /* setup the hwtid for this connection */
1194 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1197 ep->snd_seq = be32_to_cpu(req->snd_isn);
1198 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1200 set_emss(ep, ntohs(req->tcp_opt));
1202 /* dealloc the atid */
1203 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1204 cxgb4_free_atid(t, atid);
1205 set_bit(ACT_ESTAB, &ep->com.history);
1207 /* start MPA negotiation */
1208 ret = send_flowc(ep);
1211 if (ep->retry_with_mpa_v1)
1212 ret = send_mpa_req(ep, skb, 1);
1214 ret = send_mpa_req(ep, skb, mpa_rev);
1217 mutex_unlock(&ep->com.mutex);
1220 mutex_unlock(&ep->com.mutex);
1221 connect_reply_upcall(ep, -ENOMEM);
1222 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1226 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1228 struct iw_cm_event event;
1230 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1231 memset(&event, 0, sizeof(event));
1232 event.event = IW_CM_EVENT_CLOSE;
1233 event.status = status;
1234 if (ep->com.cm_id) {
1235 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1236 ep, ep->com.cm_id, ep->hwtid);
1237 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1238 deref_cm_id(&ep->com);
1239 set_bit(CLOSE_UPCALL, &ep->com.history);
1243 static void peer_close_upcall(struct c4iw_ep *ep)
1245 struct iw_cm_event event;
1247 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1248 memset(&event, 0, sizeof(event));
1249 event.event = IW_CM_EVENT_DISCONNECT;
1250 if (ep->com.cm_id) {
1251 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1252 ep, ep->com.cm_id, ep->hwtid);
1253 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1254 set_bit(DISCONN_UPCALL, &ep->com.history);
1258 static void peer_abort_upcall(struct c4iw_ep *ep)
1260 struct iw_cm_event event;
1262 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1263 memset(&event, 0, sizeof(event));
1264 event.event = IW_CM_EVENT_CLOSE;
1265 event.status = -ECONNRESET;
1266 if (ep->com.cm_id) {
1267 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1268 ep->com.cm_id, ep->hwtid);
1269 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1270 deref_cm_id(&ep->com);
1271 set_bit(ABORT_UPCALL, &ep->com.history);
1275 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1277 struct iw_cm_event event;
1279 pr_debug("ep %p tid %u status %d\n",
1280 ep, ep->hwtid, status);
1281 memset(&event, 0, sizeof(event));
1282 event.event = IW_CM_EVENT_CONNECT_REPLY;
1283 event.status = status;
1284 memcpy(&event.local_addr, &ep->com.local_addr,
1285 sizeof(ep->com.local_addr));
1286 memcpy(&event.remote_addr, &ep->com.remote_addr,
1287 sizeof(ep->com.remote_addr));
1289 if ((status == 0) || (status == -ECONNREFUSED)) {
1290 if (!ep->tried_with_mpa_v1) {
1291 /* this means MPA_v2 is used */
1292 event.ord = ep->ird;
1293 event.ird = ep->ord;
1294 event.private_data_len = ep->plen -
1295 sizeof(struct mpa_v2_conn_params);
1296 event.private_data = ep->mpa_pkt +
1297 sizeof(struct mpa_message) +
1298 sizeof(struct mpa_v2_conn_params);
1300 /* this means MPA_v1 is used */
1301 event.ord = cur_max_read_depth(ep->com.dev);
1302 event.ird = cur_max_read_depth(ep->com.dev);
1303 event.private_data_len = ep->plen;
1304 event.private_data = ep->mpa_pkt +
1305 sizeof(struct mpa_message);
1309 pr_debug("ep %p tid %u status %d\n", ep,
1311 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1312 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1315 deref_cm_id(&ep->com);
1318 static int connect_request_upcall(struct c4iw_ep *ep)
1320 struct iw_cm_event event;
1323 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1324 memset(&event, 0, sizeof(event));
1325 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1326 memcpy(&event.local_addr, &ep->com.local_addr,
1327 sizeof(ep->com.local_addr));
1328 memcpy(&event.remote_addr, &ep->com.remote_addr,
1329 sizeof(ep->com.remote_addr));
1330 event.provider_data = ep;
1331 if (!ep->tried_with_mpa_v1) {
1332 /* this means MPA_v2 is used */
1333 event.ord = ep->ord;
1334 event.ird = ep->ird;
1335 event.private_data_len = ep->plen -
1336 sizeof(struct mpa_v2_conn_params);
1337 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1338 sizeof(struct mpa_v2_conn_params);
1340 /* this means MPA_v1 is used. Send max supported */
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 + sizeof(struct mpa_message);
1346 c4iw_get_ep(&ep->com);
1347 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1350 c4iw_put_ep(&ep->com);
1351 set_bit(CONNREQ_UPCALL, &ep->com.history);
1352 c4iw_put_ep(&ep->parent_ep->com);
1356 static void established_upcall(struct c4iw_ep *ep)
1358 struct iw_cm_event event;
1360 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1361 memset(&event, 0, sizeof(event));
1362 event.event = IW_CM_EVENT_ESTABLISHED;
1363 event.ird = ep->ord;
1364 event.ord = ep->ird;
1365 if (ep->com.cm_id) {
1366 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1367 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1368 set_bit(ESTAB_UPCALL, &ep->com.history);
1372 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1374 struct sk_buff *skb;
1375 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1378 pr_debug("ep %p tid %u credits %u\n",
1379 ep, ep->hwtid, credits);
1380 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1382 pr_err("update_rx_credits - cannot alloc skb!\n");
1387 * If we couldn't specify the entire rcv window at connection setup
1388 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1389 * then add the overage in to the credits returned.
1391 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1392 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1394 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1395 RX_DACK_MODE_V(dack_mode);
1397 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1400 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1404 #define RELAXED_IRD_NEGOTIATION 1
1407 * process_mpa_reply - process streaming mode MPA reply
1411 * 0 upon success indicating a connect request was delivered to the ULP
1412 * or the mpa request is incomplete but valid so far.
1414 * 1 if a failure requires the caller to close the connection.
1416 * 2 if a failure requires the caller to abort the connection.
1418 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1420 struct mpa_message *mpa;
1421 struct mpa_v2_conn_params *mpa_v2_params;
1423 u16 resp_ird, resp_ord;
1424 u8 rtr_mismatch = 0, insuff_ird = 0;
1425 struct c4iw_qp_attributes attrs;
1426 enum c4iw_qp_attr_mask mask;
1430 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1433 * If we get more than the supported amount of private data
1434 * then we must fail this connection.
1436 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1438 goto err_stop_timer;
1442 * copy the new data into our accumulation buffer.
1444 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1446 ep->mpa_pkt_len += skb->len;
1449 * if we don't even have the mpa message, then bail.
1451 if (ep->mpa_pkt_len < sizeof(*mpa))
1453 mpa = (struct mpa_message *) ep->mpa_pkt;
1455 /* Validate MPA header. */
1456 if (mpa->revision > mpa_rev) {
1457 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1458 __func__, mpa_rev, mpa->revision);
1460 goto err_stop_timer;
1462 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1464 goto err_stop_timer;
1467 plen = ntohs(mpa->private_data_size);
1470 * Fail if there's too much private data.
1472 if (plen > MPA_MAX_PRIVATE_DATA) {
1474 goto err_stop_timer;
1478 * If plen does not account for pkt size
1480 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1482 goto err_stop_timer;
1485 ep->plen = (u8) plen;
1488 * If we don't have all the pdata yet, then bail.
1489 * We'll continue process when more data arrives.
1491 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1494 if (mpa->flags & MPA_REJECT) {
1495 err = -ECONNREFUSED;
1496 goto err_stop_timer;
1500 * Stop mpa timer. If it expired, then
1501 * we ignore the MPA reply. process_timeout()
1502 * will abort the connection.
1504 if (stop_ep_timer(ep))
1508 * If we get here we have accumulated the entire mpa
1509 * start reply message including private data. And
1510 * the MPA header is valid.
1512 __state_set(&ep->com, FPDU_MODE);
1513 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1514 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1515 ep->mpa_attr.version = mpa->revision;
1516 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1518 if (mpa->revision == 2) {
1519 ep->mpa_attr.enhanced_rdma_conn =
1520 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1521 if (ep->mpa_attr.enhanced_rdma_conn) {
1522 mpa_v2_params = (struct mpa_v2_conn_params *)
1523 (ep->mpa_pkt + sizeof(*mpa));
1524 resp_ird = ntohs(mpa_v2_params->ird) &
1525 MPA_V2_IRD_ORD_MASK;
1526 resp_ord = ntohs(mpa_v2_params->ord) &
1527 MPA_V2_IRD_ORD_MASK;
1528 pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1529 resp_ird, resp_ord, ep->ird, ep->ord);
1532 * This is a double-check. Ideally, below checks are
1533 * not required since ird/ord stuff has been taken
1534 * care of in c4iw_accept_cr
1536 if (ep->ird < resp_ord) {
1537 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1538 ep->com.dev->rdev.lldi.max_ordird_qp)
1542 } else if (ep->ird > resp_ord) {
1545 if (ep->ord > resp_ird) {
1546 if (RELAXED_IRD_NEGOTIATION)
1557 if (ntohs(mpa_v2_params->ird) &
1558 MPA_V2_PEER2PEER_MODEL) {
1559 if (ntohs(mpa_v2_params->ord) &
1560 MPA_V2_RDMA_WRITE_RTR)
1561 ep->mpa_attr.p2p_type =
1562 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1563 else if (ntohs(mpa_v2_params->ord) &
1564 MPA_V2_RDMA_READ_RTR)
1565 ep->mpa_attr.p2p_type =
1566 FW_RI_INIT_P2PTYPE_READ_REQ;
1569 } else if (mpa->revision == 1)
1571 ep->mpa_attr.p2p_type = p2p_type;
1573 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1574 ep->mpa_attr.crc_enabled,
1575 ep->mpa_attr.recv_marker_enabled,
1576 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1577 ep->mpa_attr.p2p_type, p2p_type);
1580 * If responder's RTR does not match with that of initiator, assign
1581 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1582 * generated when moving QP to RTS state.
1583 * A TERM message will be sent after QP has moved to RTS state
1585 if ((ep->mpa_attr.version == 2) && peer2peer &&
1586 (ep->mpa_attr.p2p_type != p2p_type)) {
1587 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1591 attrs.mpa_attr = ep->mpa_attr;
1592 attrs.max_ird = ep->ird;
1593 attrs.max_ord = ep->ord;
1594 attrs.llp_stream_handle = ep;
1595 attrs.next_state = C4IW_QP_STATE_RTS;
1597 mask = C4IW_QP_ATTR_NEXT_STATE |
1598 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1599 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1601 /* bind QP and TID with INIT_WR */
1602 err = c4iw_modify_qp(ep->com.qp->rhp,
1603 ep->com.qp, mask, &attrs, 1);
1608 * If responder's RTR requirement did not match with what initiator
1609 * supports, generate TERM message
1612 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1613 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1614 attrs.ecode = MPA_NOMATCH_RTR;
1615 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1616 attrs.send_term = 1;
1617 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1618 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1625 * Generate TERM if initiator IRD is not sufficient for responder
1626 * provided ORD. Currently, we do the same behaviour even when
1627 * responder provided IRD is also not sufficient as regards to
1631 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1632 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1633 attrs.ecode = MPA_INSUFF_IRD;
1634 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1635 attrs.send_term = 1;
1636 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1637 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1648 connect_reply_upcall(ep, err);
1653 * process_mpa_request - process streaming mode MPA request
1657 * 0 upon success indicating a connect request was delivered to the ULP
1658 * or the mpa request is incomplete but valid so far.
1660 * 1 if a failure requires the caller to close the connection.
1662 * 2 if a failure requires the caller to abort the connection.
1664 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1666 struct mpa_message *mpa;
1667 struct mpa_v2_conn_params *mpa_v2_params;
1670 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1673 * If we get more than the supported amount of private data
1674 * then we must fail this connection.
1676 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1677 goto err_stop_timer;
1679 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1682 * Copy the new data into our accumulation buffer.
1684 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1686 ep->mpa_pkt_len += skb->len;
1689 * If we don't even have the mpa message, then bail.
1690 * We'll continue process when more data arrives.
1692 if (ep->mpa_pkt_len < sizeof(*mpa))
1695 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1696 mpa = (struct mpa_message *) ep->mpa_pkt;
1699 * Validate MPA Header.
1701 if (mpa->revision > mpa_rev) {
1702 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1703 __func__, mpa_rev, mpa->revision);
1704 goto err_stop_timer;
1707 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1708 goto err_stop_timer;
1710 plen = ntohs(mpa->private_data_size);
1713 * Fail if there's too much private data.
1715 if (plen > MPA_MAX_PRIVATE_DATA)
1716 goto err_stop_timer;
1719 * If plen does not account for pkt size
1721 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1722 goto err_stop_timer;
1723 ep->plen = (u8) plen;
1726 * If we don't have all the pdata yet, then bail.
1728 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1732 * If we get here we have accumulated the entire mpa
1733 * start reply message including private data.
1735 ep->mpa_attr.initiator = 0;
1736 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1737 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1738 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1739 ep->mpa_attr.version = mpa->revision;
1740 if (mpa->revision == 1)
1741 ep->tried_with_mpa_v1 = 1;
1742 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1744 if (mpa->revision == 2) {
1745 ep->mpa_attr.enhanced_rdma_conn =
1746 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1747 if (ep->mpa_attr.enhanced_rdma_conn) {
1748 mpa_v2_params = (struct mpa_v2_conn_params *)
1749 (ep->mpa_pkt + sizeof(*mpa));
1750 ep->ird = ntohs(mpa_v2_params->ird) &
1751 MPA_V2_IRD_ORD_MASK;
1752 ep->ird = min_t(u32, ep->ird,
1753 cur_max_read_depth(ep->com.dev));
1754 ep->ord = ntohs(mpa_v2_params->ord) &
1755 MPA_V2_IRD_ORD_MASK;
1756 ep->ord = min_t(u32, ep->ord,
1757 cur_max_read_depth(ep->com.dev));
1758 pr_debug("initiator ird %u ord %u\n",
1760 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1762 if (ntohs(mpa_v2_params->ord) &
1763 MPA_V2_RDMA_WRITE_RTR)
1764 ep->mpa_attr.p2p_type =
1765 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1766 else if (ntohs(mpa_v2_params->ord) &
1767 MPA_V2_RDMA_READ_RTR)
1768 ep->mpa_attr.p2p_type =
1769 FW_RI_INIT_P2PTYPE_READ_REQ;
1772 } else if (mpa->revision == 1)
1774 ep->mpa_attr.p2p_type = p2p_type;
1776 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1777 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1778 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1779 ep->mpa_attr.p2p_type);
1781 __state_set(&ep->com, MPA_REQ_RCVD);
1784 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1785 if (ep->parent_ep->com.state != DEAD) {
1786 if (connect_request_upcall(ep))
1787 goto err_unlock_parent;
1789 goto err_unlock_parent;
1791 mutex_unlock(&ep->parent_ep->com.mutex);
1795 mutex_unlock(&ep->parent_ep->com.mutex);
1798 (void)stop_ep_timer(ep);
1803 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1806 struct cpl_rx_data *hdr = cplhdr(skb);
1807 unsigned int dlen = ntohs(hdr->len);
1808 unsigned int tid = GET_TID(hdr);
1809 __u8 status = hdr->status;
1812 ep = get_ep_from_tid(dev, tid);
1815 pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1816 skb_pull(skb, sizeof(*hdr));
1817 skb_trim(skb, dlen);
1818 mutex_lock(&ep->com.mutex);
1820 switch (ep->com.state) {
1822 update_rx_credits(ep, dlen);
1823 ep->rcv_seq += dlen;
1824 disconnect = process_mpa_reply(ep, skb);
1827 update_rx_credits(ep, dlen);
1828 ep->rcv_seq += dlen;
1829 disconnect = process_mpa_request(ep, skb);
1832 struct c4iw_qp_attributes attrs;
1834 update_rx_credits(ep, dlen);
1836 pr_err("%s Unexpected streaming data." \
1837 " qpid %u ep %p state %d tid %u status %d\n",
1838 __func__, ep->com.qp->wq.sq.qid, ep,
1839 ep->com.state, ep->hwtid, status);
1840 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1841 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1842 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1849 mutex_unlock(&ep->com.mutex);
1851 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1852 c4iw_put_ep(&ep->com);
1856 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx_status)
1858 enum chip_type adapter_type;
1862 adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1863 status = ABORT_RSS_STATUS_G(be32_to_cpu(srqidx_status));
1864 srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(srqidx_status));
1867 * If this TCB had a srq buffer cached, then we must complete
1868 * it. For user mode, that means saving the srqidx in the
1869 * user/kernel status page for this qp. For kernel mode, just
1870 * synthesize the CQE now.
1872 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1873 if (ep->com.qp->ibqp.uobject)
1874 t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1876 c4iw_flush_srqidx(ep->com.qp, srqidx);
1880 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1883 struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1885 unsigned int tid = GET_TID(rpl);
1887 ep = get_ep_from_tid(dev, tid);
1889 pr_warn("Abort rpl to freed endpoint\n");
1893 complete_cached_srq_buffers(ep, rpl->srqidx_status);
1895 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1896 mutex_lock(&ep->com.mutex);
1897 switch (ep->com.state) {
1899 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1900 __state_set(&ep->com, DEAD);
1904 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1907 mutex_unlock(&ep->com.mutex);
1910 release_ep_resources(ep);
1911 c4iw_put_ep(&ep->com);
1915 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1917 struct sk_buff *skb;
1918 struct fw_ofld_connection_wr *req;
1919 unsigned int mtu_idx;
1921 struct sockaddr_in *sin;
1924 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1925 req = __skb_put_zero(skb, sizeof(*req));
1926 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1927 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1928 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1929 ep->com.dev->rdev.lldi.ports[0],
1931 sin = (struct sockaddr_in *)&ep->com.local_addr;
1932 req->le.lport = sin->sin_port;
1933 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1934 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1935 req->le.pport = sin->sin_port;
1936 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1937 req->tcb.t_state_to_astid =
1938 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1939 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1940 req->tcb.cplrxdataack_cplpassacceptrpl =
1941 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1942 req->tcb.tx_max = (__force __be32) jiffies;
1943 req->tcb.rcv_adv = htons(1);
1944 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1945 enable_tcp_timestamps,
1946 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1947 wscale = cxgb_compute_wscale(rcv_win);
1950 * Specify the largest window that will fit in opt0. The
1951 * remainder will be specified in the rx_data_ack.
1953 win = ep->rcv_win >> 10;
1954 if (win > RCV_BUFSIZ_M)
1957 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1958 (nocong ? NO_CONG_F : 0) |
1961 WND_SCALE_V(wscale) |
1962 MSS_IDX_V(mtu_idx) |
1963 L2T_IDX_V(ep->l2t->idx) |
1964 TX_CHAN_V(ep->tx_chan) |
1965 SMAC_SEL_V(ep->smac_idx) |
1966 DSCP_V(ep->tos >> 2) |
1967 ULP_MODE_V(ULP_MODE_TCPDDP) |
1969 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1970 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1972 CCTRL_ECN_V(enable_ecn) |
1973 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1974 if (enable_tcp_timestamps)
1975 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1976 if (enable_tcp_sack)
1977 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1978 if (wscale && enable_tcp_window_scaling)
1979 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1980 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1981 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1982 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1983 set_bit(ACT_OFLD_CONN, &ep->com.history);
1984 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1988 * Some of the error codes above implicitly indicate that there is no TID
1989 * allocated with the result of an ACT_OPEN. We use this predicate to make
1992 static inline int act_open_has_tid(int status)
1994 return (status != CPL_ERR_TCAM_PARITY &&
1995 status != CPL_ERR_TCAM_MISS &&
1996 status != CPL_ERR_TCAM_FULL &&
1997 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1998 status != CPL_ERR_CONN_EXIST);
2001 static char *neg_adv_str(unsigned int status)
2004 case CPL_ERR_RTX_NEG_ADVICE:
2005 return "Retransmit timeout";
2006 case CPL_ERR_PERSIST_NEG_ADVICE:
2007 return "Persist timeout";
2008 case CPL_ERR_KEEPALV_NEG_ADVICE:
2009 return "Keepalive timeout";
2015 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2017 ep->snd_win = snd_win;
2018 ep->rcv_win = rcv_win;
2019 pr_debug("snd_win %d rcv_win %d\n",
2020 ep->snd_win, ep->rcv_win);
2023 #define ACT_OPEN_RETRY_COUNT 2
2025 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2026 struct dst_entry *dst, struct c4iw_dev *cdev,
2027 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2029 struct neighbour *n;
2031 struct net_device *pdev;
2033 n = dst_neigh_lookup(dst, peer_ip);
2039 if (n->dev->flags & IFF_LOOPBACK) {
2041 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2042 else if (IS_ENABLED(CONFIG_IPV6))
2043 for_each_netdev(&init_net, pdev) {
2044 if (ipv6_chk_addr(&init_net,
2045 (struct in6_addr *)peer_ip,
2056 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2057 n, pdev, rt_tos2priority(tos));
2062 ep->mtu = pdev->mtu;
2063 ep->tx_chan = cxgb4_port_chan(pdev);
2064 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2065 cxgb4_port_viid(pdev));
2066 step = cdev->rdev.lldi.ntxq /
2067 cdev->rdev.lldi.nchan;
2068 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2069 step = cdev->rdev.lldi.nrxq /
2070 cdev->rdev.lldi.nchan;
2071 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2072 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2073 cxgb4_port_idx(pdev) * step];
2074 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2077 pdev = get_real_dev(n->dev);
2078 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2082 ep->mtu = dst_mtu(dst);
2083 ep->tx_chan = cxgb4_port_chan(pdev);
2084 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2085 cxgb4_port_viid(pdev));
2086 step = cdev->rdev.lldi.ntxq /
2087 cdev->rdev.lldi.nchan;
2088 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2089 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2090 step = cdev->rdev.lldi.nrxq /
2091 cdev->rdev.lldi.nchan;
2092 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2093 cxgb4_port_idx(pdev) * step];
2094 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2097 ep->retry_with_mpa_v1 = 0;
2098 ep->tried_with_mpa_v1 = 0;
2110 static int c4iw_reconnect(struct c4iw_ep *ep)
2114 struct sockaddr_in *laddr = (struct sockaddr_in *)
2115 &ep->com.cm_id->m_local_addr;
2116 struct sockaddr_in *raddr = (struct sockaddr_in *)
2117 &ep->com.cm_id->m_remote_addr;
2118 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2119 &ep->com.cm_id->m_local_addr;
2120 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2121 &ep->com.cm_id->m_remote_addr;
2125 pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2126 c4iw_init_wr_wait(ep->com.wr_waitp);
2128 /* When MPA revision is different on nodes, the node with MPA_rev=2
2129 * tries to reconnect with MPA_rev 1 for the same EP through
2130 * c4iw_reconnect(), where the same EP is assigned with new tid for
2131 * further connection establishment. As we are using the same EP pointer
2132 * for reconnect, few skbs are used during the previous c4iw_connect(),
2133 * which leaves the EP with inadequate skbs for further
2134 * c4iw_reconnect(), Further causing a crash due to an empty
2135 * skb_list() during peer_abort(). Allocate skbs which is already used.
2137 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2138 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2144 * Allocate an active TID to initiate a TCP connection.
2146 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2147 if (ep->atid == -1) {
2148 pr_err("%s - cannot alloc atid\n", __func__);
2152 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2155 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2156 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2157 laddr->sin_addr.s_addr,
2158 raddr->sin_addr.s_addr,
2160 raddr->sin_port, ep->com.cm_id->tos);
2162 ra = (__u8 *)&raddr->sin_addr;
2164 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2166 laddr6->sin6_addr.s6_addr,
2167 raddr6->sin6_addr.s6_addr,
2169 raddr6->sin6_port, 0,
2170 raddr6->sin6_scope_id);
2172 ra = (__u8 *)&raddr6->sin6_addr;
2175 pr_err("%s - cannot find route\n", __func__);
2176 err = -EHOSTUNREACH;
2179 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2180 ep->com.dev->rdev.lldi.adapter_type,
2181 ep->com.cm_id->tos);
2183 pr_err("%s - cannot alloc l2e\n", __func__);
2187 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2188 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2191 state_set(&ep->com, CONNECTING);
2192 ep->tos = ep->com.cm_id->tos;
2194 /* send connect request to rnic */
2195 err = send_connect(ep);
2199 cxgb4_l2t_release(ep->l2t);
2201 dst_release(ep->dst);
2203 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2204 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2207 * remember to send notification to upper layer.
2208 * We are in here so the upper layer is not aware that this is
2209 * re-connect attempt and so, upper layer is still waiting for
2210 * response of 1st connect request.
2212 connect_reply_upcall(ep, -ECONNRESET);
2214 c4iw_put_ep(&ep->com);
2219 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2222 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2223 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2224 ntohl(rpl->atid_status)));
2225 struct tid_info *t = dev->rdev.lldi.tids;
2226 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2227 struct sockaddr_in *la;
2228 struct sockaddr_in *ra;
2229 struct sockaddr_in6 *la6;
2230 struct sockaddr_in6 *ra6;
2233 ep = lookup_atid(t, atid);
2234 la = (struct sockaddr_in *)&ep->com.local_addr;
2235 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2236 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2237 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2239 pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2240 status, status2errno(status));
2242 if (cxgb_is_neg_adv(status)) {
2243 pr_debug("Connection problems for atid %u status %u (%s)\n",
2244 atid, status, neg_adv_str(status));
2245 ep->stats.connect_neg_adv++;
2246 mutex_lock(&dev->rdev.stats.lock);
2247 dev->rdev.stats.neg_adv++;
2248 mutex_unlock(&dev->rdev.stats.lock);
2252 set_bit(ACT_OPEN_RPL, &ep->com.history);
2255 * Log interesting failures.
2258 case CPL_ERR_CONN_RESET:
2259 case CPL_ERR_CONN_TIMEDOUT:
2261 case CPL_ERR_TCAM_FULL:
2262 mutex_lock(&dev->rdev.stats.lock);
2263 dev->rdev.stats.tcam_full++;
2264 mutex_unlock(&dev->rdev.stats.lock);
2265 if (ep->com.local_addr.ss_family == AF_INET &&
2266 dev->rdev.lldi.enable_fw_ofld_conn) {
2267 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2268 ntohl(rpl->atid_status))));
2274 case CPL_ERR_CONN_EXIST:
2275 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2276 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2277 if (ep->com.remote_addr.ss_family == AF_INET6) {
2278 struct sockaddr_in6 *sin6 =
2279 (struct sockaddr_in6 *)
2280 &ep->com.local_addr;
2282 ep->com.dev->rdev.lldi.ports[0],
2284 &sin6->sin6_addr.s6_addr, 1);
2286 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2288 cxgb4_free_atid(t, atid);
2289 dst_release(ep->dst);
2290 cxgb4_l2t_release(ep->l2t);
2296 if (ep->com.local_addr.ss_family == AF_INET) {
2297 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2298 atid, status, status2errno(status),
2299 &la->sin_addr.s_addr, ntohs(la->sin_port),
2300 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2302 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2303 atid, status, status2errno(status),
2304 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2305 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2311 connect_reply_upcall(ep, status2errno(status));
2312 state_set(&ep->com, DEAD);
2314 if (ep->com.remote_addr.ss_family == AF_INET6) {
2315 struct sockaddr_in6 *sin6 =
2316 (struct sockaddr_in6 *)&ep->com.local_addr;
2317 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2318 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2320 if (status && act_open_has_tid(status))
2321 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2322 ep->com.local_addr.ss_family);
2324 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2325 cxgb4_free_atid(t, atid);
2326 dst_release(ep->dst);
2327 cxgb4_l2t_release(ep->l2t);
2328 c4iw_put_ep(&ep->com);
2333 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2335 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2336 unsigned int stid = GET_TID(rpl);
2337 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2340 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2343 pr_debug("ep %p status %d error %d\n", ep,
2344 rpl->status, status2errno(rpl->status));
2345 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2346 c4iw_put_ep(&ep->com);
2351 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2353 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2354 unsigned int stid = GET_TID(rpl);
2355 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2358 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2361 pr_debug("ep %p\n", ep);
2362 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2363 c4iw_put_ep(&ep->com);
2368 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2369 struct cpl_pass_accept_req *req)
2371 struct cpl_pass_accept_rpl *rpl;
2372 unsigned int mtu_idx;
2376 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2378 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2380 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2384 if (!is_t4(adapter_type)) {
2385 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2387 INIT_TP_WR(rpl5, ep->hwtid);
2389 skb_trim(skb, sizeof(*rpl));
2390 INIT_TP_WR(rpl, ep->hwtid);
2392 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2395 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2396 enable_tcp_timestamps && req->tcpopt.tstamp,
2397 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2398 wscale = cxgb_compute_wscale(rcv_win);
2401 * Specify the largest window that will fit in opt0. The
2402 * remainder will be specified in the rx_data_ack.
2404 win = ep->rcv_win >> 10;
2405 if (win > RCV_BUFSIZ_M)
2407 opt0 = (nocong ? NO_CONG_F : 0) |
2410 WND_SCALE_V(wscale) |
2411 MSS_IDX_V(mtu_idx) |
2412 L2T_IDX_V(ep->l2t->idx) |
2413 TX_CHAN_V(ep->tx_chan) |
2414 SMAC_SEL_V(ep->smac_idx) |
2415 DSCP_V(ep->tos >> 2) |
2416 ULP_MODE_V(ULP_MODE_TCPDDP) |
2418 opt2 = RX_CHANNEL_V(0) |
2419 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2421 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2422 opt2 |= TSTAMPS_EN_F;
2423 if (enable_tcp_sack && req->tcpopt.sack)
2425 if (wscale && enable_tcp_window_scaling)
2426 opt2 |= WND_SCALE_EN_F;
2428 const struct tcphdr *tcph;
2429 u32 hlen = ntohl(req->hdr_len);
2431 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2432 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2435 tcph = (const void *)(req + 1) +
2436 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2437 if (tcph->ece && tcph->cwr)
2438 opt2 |= CCTRL_ECN_V(1);
2440 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2441 u32 isn = (prandom_u32() & ~7UL) - 1;
2442 opt2 |= T5_OPT_2_VALID_F;
2443 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2446 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2449 rpl5->iss = cpu_to_be32(isn);
2450 pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2453 rpl->opt0 = cpu_to_be64(opt0);
2454 rpl->opt2 = cpu_to_be32(opt2);
2455 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2456 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2458 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2461 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2463 pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2464 skb_trim(skb, sizeof(struct cpl_tid_release));
2465 release_tid(&dev->rdev, hwtid, skb);
2469 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2471 struct c4iw_ep *child_ep = NULL, *parent_ep;
2472 struct cpl_pass_accept_req *req = cplhdr(skb);
2473 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2474 struct tid_info *t = dev->rdev.lldi.tids;
2475 unsigned int hwtid = GET_TID(req);
2476 struct dst_entry *dst;
2477 __u8 local_ip[16], peer_ip[16];
2478 __be16 local_port, peer_port;
2479 struct sockaddr_in6 *sin6;
2481 u16 peer_mss = ntohs(req->tcpopt.mss);
2483 unsigned short hdrs;
2484 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2486 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2488 pr_err("%s connect request on invalid stid %d\n",
2493 if (state_read(&parent_ep->com) != LISTEN) {
2494 pr_err("%s - listening ep not in LISTEN\n", __func__);
2498 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2499 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2501 /* Find output route */
2503 pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2505 local_ip, peer_ip, ntohs(local_port),
2506 ntohs(peer_port), peer_mss);
2507 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2508 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2509 local_port, peer_port, tos);
2511 pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2513 local_ip, peer_ip, ntohs(local_port),
2514 ntohs(peer_port), peer_mss);
2515 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2516 local_ip, peer_ip, local_port, peer_port,
2517 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2518 ((struct sockaddr_in6 *)
2519 &parent_ep->com.local_addr)->sin6_scope_id);
2522 pr_err("%s - failed to find dst entry!\n", __func__);
2526 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2528 pr_err("%s - failed to allocate ep entry!\n", __func__);
2533 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2534 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2536 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2542 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2543 sizeof(struct tcphdr) +
2544 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2545 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2546 child_ep->mtu = peer_mss + hdrs;
2548 skb_queue_head_init(&child_ep->com.ep_skb_list);
2549 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2552 state_set(&child_ep->com, CONNECTING);
2553 child_ep->com.dev = dev;
2554 child_ep->com.cm_id = NULL;
2557 struct sockaddr_in *sin = (struct sockaddr_in *)
2558 &child_ep->com.local_addr;
2560 sin->sin_family = AF_INET;
2561 sin->sin_port = local_port;
2562 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2564 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2565 sin->sin_family = AF_INET;
2566 sin->sin_port = ((struct sockaddr_in *)
2567 &parent_ep->com.local_addr)->sin_port;
2568 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2570 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2571 sin->sin_family = AF_INET;
2572 sin->sin_port = peer_port;
2573 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2575 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2576 sin6->sin6_family = PF_INET6;
2577 sin6->sin6_port = local_port;
2578 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2580 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2581 sin6->sin6_family = PF_INET6;
2582 sin6->sin6_port = ((struct sockaddr_in6 *)
2583 &parent_ep->com.local_addr)->sin6_port;
2584 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2586 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2587 sin6->sin6_family = PF_INET6;
2588 sin6->sin6_port = peer_port;
2589 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2592 c4iw_get_ep(&parent_ep->com);
2593 child_ep->parent_ep = parent_ep;
2594 child_ep->tos = tos;
2595 child_ep->dst = dst;
2596 child_ep->hwtid = hwtid;
2598 pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2599 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2601 timer_setup(&child_ep->timer, ep_timeout, 0);
2602 cxgb4_insert_tid(t, child_ep, hwtid,
2603 child_ep->com.local_addr.ss_family);
2604 insert_ep_tid(child_ep);
2605 if (accept_cr(child_ep, skb, req)) {
2606 c4iw_put_ep(&parent_ep->com);
2607 release_ep_resources(child_ep);
2609 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2612 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2613 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2614 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2618 c4iw_put_ep(&child_ep->com);
2620 reject_cr(dev, hwtid, skb);
2623 c4iw_put_ep(&parent_ep->com);
2627 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2630 struct cpl_pass_establish *req = cplhdr(skb);
2631 unsigned int tid = GET_TID(req);
2634 ep = get_ep_from_tid(dev, tid);
2635 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2636 ep->snd_seq = be32_to_cpu(req->snd_isn);
2637 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2639 pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid,
2640 ntohs(req->tcp_opt));
2642 set_emss(ep, ntohs(req->tcp_opt));
2644 dst_confirm(ep->dst);
2645 mutex_lock(&ep->com.mutex);
2646 ep->com.state = MPA_REQ_WAIT;
2648 set_bit(PASS_ESTAB, &ep->com.history);
2649 ret = send_flowc(ep);
2650 mutex_unlock(&ep->com.mutex);
2652 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2653 c4iw_put_ep(&ep->com);
2658 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2660 struct cpl_peer_close *hdr = cplhdr(skb);
2662 struct c4iw_qp_attributes attrs;
2665 unsigned int tid = GET_TID(hdr);
2668 ep = get_ep_from_tid(dev, tid);
2672 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2673 dst_confirm(ep->dst);
2675 set_bit(PEER_CLOSE, &ep->com.history);
2676 mutex_lock(&ep->com.mutex);
2677 switch (ep->com.state) {
2679 __state_set(&ep->com, CLOSING);
2682 __state_set(&ep->com, CLOSING);
2683 connect_reply_upcall(ep, -ECONNRESET);
2688 * We're gonna mark this puppy DEAD, but keep
2689 * the reference on it until the ULP accepts or
2690 * rejects the CR. Also wake up anyone waiting
2691 * in rdma connection migration (see c4iw_accept_cr()).
2693 __state_set(&ep->com, CLOSING);
2694 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2695 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2698 __state_set(&ep->com, CLOSING);
2699 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2700 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2704 __state_set(&ep->com, CLOSING);
2705 attrs.next_state = C4IW_QP_STATE_CLOSING;
2706 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2707 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2708 if (ret != -ECONNRESET) {
2709 peer_close_upcall(ep);
2717 __state_set(&ep->com, MORIBUND);
2721 (void)stop_ep_timer(ep);
2722 if (ep->com.cm_id && ep->com.qp) {
2723 attrs.next_state = C4IW_QP_STATE_IDLE;
2724 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2725 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2727 close_complete_upcall(ep, 0);
2728 __state_set(&ep->com, DEAD);
2736 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2738 mutex_unlock(&ep->com.mutex);
2740 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2742 release_ep_resources(ep);
2743 c4iw_put_ep(&ep->com);
2747 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2749 struct cpl_abort_req_rss6 *req = cplhdr(skb);
2751 struct sk_buff *rpl_skb;
2752 struct c4iw_qp_attributes attrs;
2755 unsigned int tid = GET_TID(req);
2758 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2760 ep = get_ep_from_tid(dev, tid);
2764 status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2766 if (cxgb_is_neg_adv(status)) {
2767 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2768 ep->hwtid, status, neg_adv_str(status));
2769 ep->stats.abort_neg_adv++;
2770 mutex_lock(&dev->rdev.stats.lock);
2771 dev->rdev.stats.neg_adv++;
2772 mutex_unlock(&dev->rdev.stats.lock);
2776 complete_cached_srq_buffers(ep, req->srqidx_status);
2778 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2780 set_bit(PEER_ABORT, &ep->com.history);
2783 * Wake up any threads in rdma_init() or rdma_fini().
2784 * However, this is not needed if com state is just
2787 if (ep->com.state != MPA_REQ_SENT)
2788 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2790 mutex_lock(&ep->com.mutex);
2791 switch (ep->com.state) {
2793 c4iw_put_ep(&ep->parent_ep->com);
2796 (void)stop_ep_timer(ep);
2799 (void)stop_ep_timer(ep);
2800 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2801 connect_reply_upcall(ep, -ECONNRESET);
2804 * we just don't send notification upwards because we
2805 * want to retry with mpa_v1 without upper layers even
2808 * do some housekeeping so as to re-initiate the
2811 pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2813 ep->retry_with_mpa_v1 = 1;
2825 if (ep->com.cm_id && ep->com.qp) {
2826 attrs.next_state = C4IW_QP_STATE_ERROR;
2827 ret = c4iw_modify_qp(ep->com.qp->rhp,
2828 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2831 pr_err("%s - qp <- error failed!\n", __func__);
2833 peer_abort_upcall(ep);
2838 pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2839 mutex_unlock(&ep->com.mutex);
2842 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2845 dst_confirm(ep->dst);
2846 if (ep->com.state != ABORTING) {
2847 __state_set(&ep->com, DEAD);
2848 /* we don't release if we want to retry with mpa_v1 */
2849 if (!ep->retry_with_mpa_v1)
2852 mutex_unlock(&ep->com.mutex);
2854 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2855 if (WARN_ON(!rpl_skb)) {
2860 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2862 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2865 release_ep_resources(ep);
2866 else if (ep->retry_with_mpa_v1) {
2867 if (ep->com.remote_addr.ss_family == AF_INET6) {
2868 struct sockaddr_in6 *sin6 =
2869 (struct sockaddr_in6 *)
2870 &ep->com.local_addr;
2872 ep->com.dev->rdev.lldi.ports[0],
2873 (const u32 *)&sin6->sin6_addr.s6_addr,
2876 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2877 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2878 ep->com.local_addr.ss_family);
2879 dst_release(ep->dst);
2880 cxgb4_l2t_release(ep->l2t);
2885 c4iw_put_ep(&ep->com);
2886 /* Dereferencing ep, referenced in peer_abort_intr() */
2887 c4iw_put_ep(&ep->com);
2891 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2894 struct c4iw_qp_attributes attrs;
2895 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2897 unsigned int tid = GET_TID(rpl);
2899 ep = get_ep_from_tid(dev, tid);
2903 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2905 /* The cm_id may be null if we failed to connect */
2906 mutex_lock(&ep->com.mutex);
2907 set_bit(CLOSE_CON_RPL, &ep->com.history);
2908 switch (ep->com.state) {
2910 __state_set(&ep->com, MORIBUND);
2913 (void)stop_ep_timer(ep);
2914 if ((ep->com.cm_id) && (ep->com.qp)) {
2915 attrs.next_state = C4IW_QP_STATE_IDLE;
2916 c4iw_modify_qp(ep->com.qp->rhp,
2918 C4IW_QP_ATTR_NEXT_STATE,
2921 close_complete_upcall(ep, 0);
2922 __state_set(&ep->com, DEAD);
2929 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2932 mutex_unlock(&ep->com.mutex);
2934 release_ep_resources(ep);
2935 c4iw_put_ep(&ep->com);
2939 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2941 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2942 unsigned int tid = GET_TID(rpl);
2944 struct c4iw_qp_attributes attrs;
2946 ep = get_ep_from_tid(dev, tid);
2948 if (ep && ep->com.qp) {
2949 pr_warn("TERM received tid %u qpid %u\n",
2950 tid, ep->com.qp->wq.sq.qid);
2951 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2952 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2953 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2955 pr_warn("TERM received tid %u no ep/qp\n", tid);
2956 c4iw_put_ep(&ep->com);
2962 * Upcall from the adapter indicating data has been transmitted.
2963 * For us its just the single MPA request or reply. We can now free
2964 * the skb holding the mpa message.
2966 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2969 struct cpl_fw4_ack *hdr = cplhdr(skb);
2970 u8 credits = hdr->credits;
2971 unsigned int tid = GET_TID(hdr);
2974 ep = get_ep_from_tid(dev, tid);
2977 pr_debug("ep %p tid %u credits %u\n",
2978 ep, ep->hwtid, credits);
2980 pr_debug("0 credit ack ep %p tid %u state %u\n",
2981 ep, ep->hwtid, state_read(&ep->com));
2985 dst_confirm(ep->dst);
2987 pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
2988 ep, ep->hwtid, state_read(&ep->com),
2989 ep->mpa_attr.initiator ? 1 : 0);
2990 mutex_lock(&ep->com.mutex);
2991 kfree_skb(ep->mpa_skb);
2993 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2995 mutex_unlock(&ep->com.mutex);
2998 c4iw_put_ep(&ep->com);
3002 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3005 struct c4iw_ep *ep = to_ep(cm_id);
3007 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3009 mutex_lock(&ep->com.mutex);
3010 if (ep->com.state != MPA_REQ_RCVD) {
3011 mutex_unlock(&ep->com.mutex);
3012 c4iw_put_ep(&ep->com);
3015 set_bit(ULP_REJECT, &ep->com.history);
3019 abort = send_mpa_reject(ep, pdata, pdata_len);
3020 mutex_unlock(&ep->com.mutex);
3023 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3024 c4iw_put_ep(&ep->com);
3028 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3031 struct c4iw_qp_attributes attrs;
3032 enum c4iw_qp_attr_mask mask;
3033 struct c4iw_ep *ep = to_ep(cm_id);
3034 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3035 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3038 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3040 mutex_lock(&ep->com.mutex);
3041 if (ep->com.state != MPA_REQ_RCVD) {
3051 set_bit(ULP_ACCEPT, &ep->com.history);
3052 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3053 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3058 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3059 if (conn_param->ord > ep->ird) {
3060 if (RELAXED_IRD_NEGOTIATION) {
3061 conn_param->ord = ep->ird;
3063 ep->ird = conn_param->ird;
3064 ep->ord = conn_param->ord;
3065 send_mpa_reject(ep, conn_param->private_data,
3066 conn_param->private_data_len);
3071 if (conn_param->ird < ep->ord) {
3072 if (RELAXED_IRD_NEGOTIATION &&
3073 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3074 conn_param->ird = ep->ord;
3081 ep->ird = conn_param->ird;
3082 ep->ord = conn_param->ord;
3084 if (ep->mpa_attr.version == 1) {
3085 if (peer2peer && ep->ird == 0)
3089 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3090 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3094 pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3096 ep->com.cm_id = cm_id;
3097 ref_cm_id(&ep->com);
3101 /* bind QP to EP and move to RTS */
3102 attrs.mpa_attr = ep->mpa_attr;
3103 attrs.max_ird = ep->ird;
3104 attrs.max_ord = ep->ord;
3105 attrs.llp_stream_handle = ep;
3106 attrs.next_state = C4IW_QP_STATE_RTS;
3108 /* bind QP and TID with INIT_WR */
3109 mask = C4IW_QP_ATTR_NEXT_STATE |
3110 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3111 C4IW_QP_ATTR_MPA_ATTR |
3112 C4IW_QP_ATTR_MAX_IRD |
3113 C4IW_QP_ATTR_MAX_ORD;
3115 err = c4iw_modify_qp(ep->com.qp->rhp,
3116 ep->com.qp, mask, &attrs, 1);
3118 goto err_deref_cm_id;
3120 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3121 err = send_mpa_reply(ep, conn_param->private_data,
3122 conn_param->private_data_len);
3124 goto err_deref_cm_id;
3126 __state_set(&ep->com, FPDU_MODE);
3127 established_upcall(ep);
3128 mutex_unlock(&ep->com.mutex);
3129 c4iw_put_ep(&ep->com);
3132 deref_cm_id(&ep->com);
3136 mutex_unlock(&ep->com.mutex);
3138 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3139 c4iw_put_ep(&ep->com);
3143 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3145 struct in_device *ind;
3147 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3148 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3150 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3152 return -EADDRNOTAVAIL;
3153 for_primary_ifa(ind) {
3154 laddr->sin_addr.s_addr = ifa->ifa_address;
3155 raddr->sin_addr.s_addr = ifa->ifa_address;
3161 return found ? 0 : -EADDRNOTAVAIL;
3164 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3165 unsigned char banned_flags)
3167 struct inet6_dev *idev;
3168 int err = -EADDRNOTAVAIL;
3171 idev = __in6_dev_get(dev);
3173 struct inet6_ifaddr *ifp;
3175 read_lock_bh(&idev->lock);
3176 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3177 if (ifp->scope == IFA_LINK &&
3178 !(ifp->flags & banned_flags)) {
3179 memcpy(addr, &ifp->addr, 16);
3184 read_unlock_bh(&idev->lock);
3190 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3192 struct in6_addr uninitialized_var(addr);
3193 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3194 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3196 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3197 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3198 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3201 return -EADDRNOTAVAIL;
3204 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3206 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3209 struct sockaddr_in *laddr;
3210 struct sockaddr_in *raddr;
3211 struct sockaddr_in6 *laddr6;
3212 struct sockaddr_in6 *raddr6;
3216 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3217 (conn_param->ird > cur_max_read_depth(dev))) {
3221 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3223 pr_err("%s - cannot alloc ep\n", __func__);
3228 skb_queue_head_init(&ep->com.ep_skb_list);
3229 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3234 timer_setup(&ep->timer, ep_timeout, 0);
3235 ep->plen = conn_param->private_data_len;
3237 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3238 conn_param->private_data, ep->plen);
3239 ep->ird = conn_param->ird;
3240 ep->ord = conn_param->ord;
3242 if (peer2peer && ep->ord == 0)
3245 ep->com.cm_id = cm_id;
3246 ref_cm_id(&ep->com);
3247 cm_id->provider_data = ep;
3249 ep->com.qp = get_qhp(dev, conn_param->qpn);
3251 pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3256 pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3260 * Allocate an active TID to initiate a TCP connection.
3262 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3263 if (ep->atid == -1) {
3264 pr_err("%s - cannot alloc atid\n", __func__);
3268 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3270 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3271 sizeof(ep->com.local_addr));
3272 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3273 sizeof(ep->com.remote_addr));
3275 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3276 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3277 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3278 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3280 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3282 ra = (__u8 *)&raddr->sin_addr;
3285 * Handle loopback requests to INADDR_ANY.
3287 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3288 err = pick_local_ipaddrs(dev, cm_id);
3294 pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3295 &laddr->sin_addr, ntohs(laddr->sin_port),
3296 ra, ntohs(raddr->sin_port));
3297 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3298 laddr->sin_addr.s_addr,
3299 raddr->sin_addr.s_addr,
3301 raddr->sin_port, cm_id->tos);
3304 ra = (__u8 *)&raddr6->sin6_addr;
3307 * Handle loopback requests to INADDR_ANY.
3309 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3310 err = pick_local_ip6addrs(dev, cm_id);
3316 pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3317 laddr6->sin6_addr.s6_addr,
3318 ntohs(laddr6->sin6_port),
3319 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3320 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3321 laddr6->sin6_addr.s6_addr,
3322 raddr6->sin6_addr.s6_addr,
3324 raddr6->sin6_port, 0,
3325 raddr6->sin6_scope_id);
3328 pr_err("%s - cannot find route\n", __func__);
3329 err = -EHOSTUNREACH;
3333 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3334 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3336 pr_err("%s - cannot alloc l2e\n", __func__);
3340 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3341 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3344 state_set(&ep->com, CONNECTING);
3345 ep->tos = cm_id->tos;
3347 /* send connect request to rnic */
3348 err = send_connect(ep);
3352 cxgb4_l2t_release(ep->l2t);
3354 dst_release(ep->dst);
3356 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3357 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3359 skb_queue_purge(&ep->com.ep_skb_list);
3360 deref_cm_id(&ep->com);
3362 c4iw_put_ep(&ep->com);
3367 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3370 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3371 &ep->com.local_addr;
3373 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3374 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3375 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3379 c4iw_init_wr_wait(ep->com.wr_waitp);
3380 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3381 ep->stid, &sin6->sin6_addr,
3383 ep->com.dev->rdev.lldi.rxq_ids[0]);
3385 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3389 err = net_xmit_errno(err);
3391 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3392 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3393 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3395 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3400 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3403 struct sockaddr_in *sin = (struct sockaddr_in *)
3404 &ep->com.local_addr;
3406 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3408 err = cxgb4_create_server_filter(
3409 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3410 sin->sin_addr.s_addr, sin->sin_port, 0,
3411 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3412 if (err == -EBUSY) {
3413 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3417 set_current_state(TASK_UNINTERRUPTIBLE);
3418 schedule_timeout(usecs_to_jiffies(100));
3420 } while (err == -EBUSY);
3422 c4iw_init_wr_wait(ep->com.wr_waitp);
3423 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3424 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3425 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3427 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3431 err = net_xmit_errno(err);
3434 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3436 &sin->sin_addr, ntohs(sin->sin_port));
3440 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3443 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3444 struct c4iw_listen_ep *ep;
3448 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3450 pr_err("%s - cannot alloc ep\n", __func__);
3454 skb_queue_head_init(&ep->com.ep_skb_list);
3455 pr_debug("ep %p\n", ep);
3456 ep->com.cm_id = cm_id;
3457 ref_cm_id(&ep->com);
3459 ep->backlog = backlog;
3460 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3461 sizeof(ep->com.local_addr));
3464 * Allocate a server TID.
3466 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3467 ep->com.local_addr.ss_family == AF_INET)
3468 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3469 cm_id->m_local_addr.ss_family, ep);
3471 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3472 cm_id->m_local_addr.ss_family, ep);
3474 if (ep->stid == -1) {
3475 pr_err("%s - cannot alloc stid\n", __func__);
3479 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3481 state_set(&ep->com, LISTEN);
3482 if (ep->com.local_addr.ss_family == AF_INET)
3483 err = create_server4(dev, ep);
3485 err = create_server6(dev, ep);
3487 cm_id->provider_data = ep;
3490 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3491 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3492 ep->com.local_addr.ss_family);
3494 deref_cm_id(&ep->com);
3495 c4iw_put_ep(&ep->com);
3501 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3504 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3506 pr_debug("ep %p\n", ep);
3509 state_set(&ep->com, DEAD);
3510 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3511 ep->com.local_addr.ss_family == AF_INET) {
3512 err = cxgb4_remove_server_filter(
3513 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3514 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3516 struct sockaddr_in6 *sin6;
3517 c4iw_init_wr_wait(ep->com.wr_waitp);
3518 err = cxgb4_remove_server(
3519 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3520 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3523 err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3525 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3526 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3527 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3529 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3530 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3531 ep->com.local_addr.ss_family);
3533 deref_cm_id(&ep->com);
3534 c4iw_put_ep(&ep->com);
3538 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3543 struct c4iw_rdev *rdev;
3545 mutex_lock(&ep->com.mutex);
3547 pr_debug("ep %p state %s, abrupt %d\n", ep,
3548 states[ep->com.state], abrupt);
3551 * Ref the ep here in case we have fatal errors causing the
3552 * ep to be released and freed.
3554 c4iw_get_ep(&ep->com);
3556 rdev = &ep->com.dev->rdev;
3557 if (c4iw_fatal_error(rdev)) {
3559 close_complete_upcall(ep, -EIO);
3560 ep->com.state = DEAD;
3562 switch (ep->com.state) {
3571 ep->com.state = ABORTING;
3573 ep->com.state = CLOSING;
3576 * if we close before we see the fw4_ack() then we fix
3577 * up the timer state since we're reusing it.
3580 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3581 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3586 set_bit(CLOSE_SENT, &ep->com.flags);
3589 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3592 (void)stop_ep_timer(ep);
3593 ep->com.state = ABORTING;
3595 ep->com.state = MORIBUND;
3601 pr_debug("ignoring disconnect ep %p state %u\n",
3605 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3611 set_bit(EP_DISC_ABORT, &ep->com.history);
3612 close_complete_upcall(ep, -ECONNRESET);
3613 ret = send_abort(ep);
3615 set_bit(EP_DISC_CLOSE, &ep->com.history);
3616 ret = send_halfclose(ep);
3619 set_bit(EP_DISC_FAIL, &ep->com.history);
3622 close_complete_upcall(ep, -EIO);
3625 struct c4iw_qp_attributes attrs;
3627 attrs.next_state = C4IW_QP_STATE_ERROR;
3628 ret = c4iw_modify_qp(ep->com.qp->rhp,
3630 C4IW_QP_ATTR_NEXT_STATE,
3633 pr_err("%s - qp <- error failed!\n",
3639 mutex_unlock(&ep->com.mutex);
3640 c4iw_put_ep(&ep->com);
3642 release_ep_resources(ep);
3646 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3647 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3650 int atid = be32_to_cpu(req->tid);
3652 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3653 (__force u32) req->tid);
3657 switch (req->retval) {
3659 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3660 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3661 send_fw_act_open_req(ep, atid);
3666 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3667 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3668 send_fw_act_open_req(ep, atid);
3673 pr_info("%s unexpected ofld conn wr retval %d\n",
3674 __func__, req->retval);
3677 pr_err("active ofld_connect_wr failure %d atid %d\n",
3679 mutex_lock(&dev->rdev.stats.lock);
3680 dev->rdev.stats.act_ofld_conn_fails++;
3681 mutex_unlock(&dev->rdev.stats.lock);
3682 connect_reply_upcall(ep, status2errno(req->retval));
3683 state_set(&ep->com, DEAD);
3684 if (ep->com.remote_addr.ss_family == AF_INET6) {
3685 struct sockaddr_in6 *sin6 =
3686 (struct sockaddr_in6 *)&ep->com.local_addr;
3687 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3688 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3690 remove_handle(dev, &dev->atid_idr, atid);
3691 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3692 dst_release(ep->dst);
3693 cxgb4_l2t_release(ep->l2t);
3694 c4iw_put_ep(&ep->com);
3697 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3698 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3700 struct sk_buff *rpl_skb;
3701 struct cpl_pass_accept_req *cpl;
3704 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3706 pr_err("%s passive open failure %d\n", __func__, req->retval);
3707 mutex_lock(&dev->rdev.stats.lock);
3708 dev->rdev.stats.pas_ofld_conn_fails++;
3709 mutex_unlock(&dev->rdev.stats.lock);
3712 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3713 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3714 (__force u32) htonl(
3715 (__force u32) req->tid)));
3716 ret = pass_accept_req(dev, rpl_skb);
3723 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3725 struct cpl_fw6_msg *rpl = cplhdr(skb);
3726 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3728 switch (rpl->type) {
3730 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3732 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3733 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3734 switch (req->t_state) {
3736 active_ofld_conn_reply(dev, skb, req);
3739 passive_ofld_conn_reply(dev, skb, req);
3742 pr_err("%s unexpected ofld conn wr state %d\n",
3743 __func__, req->t_state);
3751 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3754 __be16 hdr_len, vlantag, len;
3756 int tcp_hdr_len, ip_hdr_len;
3758 struct cpl_rx_pkt *cpl = cplhdr(skb);
3759 struct cpl_pass_accept_req *req;
3760 struct tcp_options_received tmp_opt;
3761 struct c4iw_dev *dev;
3762 enum chip_type type;
3764 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3765 /* Store values from cpl_rx_pkt in temporary location. */
3766 vlantag = cpl->vlan;
3768 l2info = cpl->l2info;
3769 hdr_len = cpl->hdr_len;
3772 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3775 * We need to parse the TCP options from SYN packet.
3776 * to generate cpl_pass_accept_req.
3778 memset(&tmp_opt, 0, sizeof(tmp_opt));
3779 tcp_clear_options(&tmp_opt);
3780 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3782 req = __skb_push(skb, sizeof(*req));
3783 memset(req, 0, sizeof(*req));
3784 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3785 SYN_MAC_IDX_V(RX_MACIDX_G(
3786 be32_to_cpu(l2info))) |
3788 type = dev->rdev.lldi.adapter_type;
3789 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3790 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3792 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3793 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3794 eth_hdr_len = is_t4(type) ?
3795 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3796 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3797 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3798 IP_HDR_LEN_V(ip_hdr_len) |
3799 ETH_HDR_LEN_V(eth_hdr_len));
3800 } else { /* T6 and later */
3801 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3802 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3803 T6_IP_HDR_LEN_V(ip_hdr_len) |
3804 T6_ETH_HDR_LEN_V(eth_hdr_len));
3806 req->vlan = vlantag;
3808 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3809 PASS_OPEN_TOS_V(tos));
3810 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3811 if (tmp_opt.wscale_ok)
3812 req->tcpopt.wsf = tmp_opt.snd_wscale;
3813 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3814 if (tmp_opt.sack_ok)
3815 req->tcpopt.sack = 1;
3816 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3820 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3821 __be32 laddr, __be16 lport,
3822 __be32 raddr, __be16 rport,
3823 u32 rcv_isn, u32 filter, u16 window,
3824 u32 rss_qid, u8 port_id)
3826 struct sk_buff *req_skb;
3827 struct fw_ofld_connection_wr *req;
3828 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3831 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3834 req = __skb_put_zero(req_skb, sizeof(*req));
3835 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3836 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3837 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3838 req->le.filter = (__force __be32) filter;
3839 req->le.lport = lport;
3840 req->le.pport = rport;
3841 req->le.u.ipv4.lip = laddr;
3842 req->le.u.ipv4.pip = raddr;
3843 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3844 req->tcb.rcv_adv = htons(window);
3845 req->tcb.t_state_to_astid =
3846 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3847 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3848 FW_OFLD_CONNECTION_WR_ASTID_V(
3849 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3852 * We store the qid in opt2 which will be used by the firmware
3853 * to send us the wr response.
3855 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3858 * We initialize the MSS index in TCB to 0xF.
3859 * So that when driver sends cpl_pass_accept_rpl
3860 * TCB picks up the correct value. If this was 0
3861 * TP will ignore any value > 0 for MSS index.
3863 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3864 req->cookie = (uintptr_t)skb;
3866 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3867 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3869 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3877 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3878 * messages when a filter is being used instead of server to
3879 * redirect a syn packet. When packets hit filter they are redirected
3880 * to the offload queue and driver tries to establish the connection
3881 * using firmware work request.
3883 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3886 unsigned int filter;
3887 struct ethhdr *eh = NULL;
3888 struct vlan_ethhdr *vlan_eh = NULL;
3890 struct tcphdr *tcph;
3891 struct rss_header *rss = (void *)skb->data;
3892 struct cpl_rx_pkt *cpl = (void *)skb->data;
3893 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3894 struct l2t_entry *e;
3895 struct dst_entry *dst;
3896 struct c4iw_ep *lep = NULL;
3898 struct port_info *pi;
3899 struct net_device *pdev;
3900 u16 rss_qid, eth_hdr_len;
3902 struct neighbour *neigh;
3904 /* Drop all non-SYN packets */
3905 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3909 * Drop all packets which did not hit the filter.
3910 * Unlikely to happen.
3912 if (!(rss->filter_hit && rss->filter_tid))
3916 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3918 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3920 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3922 pr_warn("%s connect request on invalid stid %d\n",
3927 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3929 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3932 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3935 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3938 pr_err("T%d Chip is not supported\n",
3939 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3943 if (eth_hdr_len == ETH_HLEN) {
3944 eh = (struct ethhdr *)(req + 1);
3945 iph = (struct iphdr *)(eh + 1);
3947 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3948 iph = (struct iphdr *)(vlan_eh + 1);
3949 skb->vlan_tci = ntohs(cpl->vlan);
3952 if (iph->version != 0x4)
3955 tcph = (struct tcphdr *)(iph + 1);
3956 skb_set_network_header(skb, (void *)iph - (void *)rss);
3957 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3960 pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
3961 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3962 ntohs(tcph->source), iph->tos);
3964 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3965 iph->daddr, iph->saddr, tcph->dest,
3966 tcph->source, iph->tos);
3968 pr_err("%s - failed to find dst entry!\n", __func__);
3971 neigh = dst_neigh_lookup_skb(dst, skb);
3974 pr_err("%s - failed to allocate neigh!\n", __func__);
3978 if (neigh->dev->flags & IFF_LOOPBACK) {
3979 pdev = ip_dev_find(&init_net, iph->daddr);
3980 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3982 pi = (struct port_info *)netdev_priv(pdev);
3985 pdev = get_real_dev(neigh->dev);
3986 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3988 pi = (struct port_info *)netdev_priv(pdev);
3990 neigh_release(neigh);
3992 pr_err("%s - failed to allocate l2t entry!\n",
3997 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3998 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3999 window = (__force u16) htons((__force u16)tcph->window);
4001 /* Calcuate filter portion for LE region. */
4002 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4003 dev->rdev.lldi.ports[0],
4007 * Synthesize the cpl_pass_accept_req. We have everything except the
4008 * TID. Once firmware sends a reply with TID we update the TID field
4009 * in cpl and pass it through the regular cpl_pass_accept_req path.
4011 build_cpl_pass_accept_req(skb, stid, iph->tos);
4012 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4013 tcph->source, ntohl(tcph->seq), filter, window,
4014 rss_qid, pi->port_id);
4015 cxgb4_l2t_release(e);
4020 c4iw_put_ep(&lep->com);
4025 * These are the real handlers that are called from a
4028 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4029 [CPL_ACT_ESTABLISH] = act_establish,
4030 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4031 [CPL_RX_DATA] = rx_data,
4032 [CPL_ABORT_RPL_RSS] = abort_rpl,
4033 [CPL_ABORT_RPL] = abort_rpl,
4034 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4035 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4036 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4037 [CPL_PASS_ESTABLISH] = pass_establish,
4038 [CPL_PEER_CLOSE] = peer_close,
4039 [CPL_ABORT_REQ_RSS] = peer_abort,
4040 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4041 [CPL_RDMA_TERMINATE] = terminate,
4042 [CPL_FW4_ACK] = fw4_ack,
4043 [CPL_FW6_MSG] = deferred_fw6_msg,
4044 [CPL_RX_PKT] = rx_pkt,
4045 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4046 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4049 static void process_timeout(struct c4iw_ep *ep)
4051 struct c4iw_qp_attributes attrs;
4054 mutex_lock(&ep->com.mutex);
4055 pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4056 set_bit(TIMEDOUT, &ep->com.history);
4057 switch (ep->com.state) {
4059 connect_reply_upcall(ep, -ETIMEDOUT);
4068 if (ep->com.cm_id && ep->com.qp) {
4069 attrs.next_state = C4IW_QP_STATE_ERROR;
4070 c4iw_modify_qp(ep->com.qp->rhp,
4071 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4074 close_complete_upcall(ep, -ETIMEDOUT);
4080 * These states are expected if the ep timed out at the same
4081 * time as another thread was calling stop_ep_timer().
4082 * So we silently do nothing for these states.
4087 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4088 __func__, ep, ep->hwtid, ep->com.state);
4091 mutex_unlock(&ep->com.mutex);
4093 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4094 c4iw_put_ep(&ep->com);
4097 static void process_timedout_eps(void)
4101 spin_lock_irq(&timeout_lock);
4102 while (!list_empty(&timeout_list)) {
4103 struct list_head *tmp;
4105 tmp = timeout_list.next;
4109 spin_unlock_irq(&timeout_lock);
4110 ep = list_entry(tmp, struct c4iw_ep, entry);
4111 process_timeout(ep);
4112 spin_lock_irq(&timeout_lock);
4114 spin_unlock_irq(&timeout_lock);
4117 static void process_work(struct work_struct *work)
4119 struct sk_buff *skb = NULL;
4120 struct c4iw_dev *dev;
4121 struct cpl_act_establish *rpl;
4122 unsigned int opcode;
4125 process_timedout_eps();
4126 while ((skb = skb_dequeue(&rxq))) {
4128 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4129 opcode = rpl->ot.opcode;
4131 if (opcode >= ARRAY_SIZE(work_handlers) ||
4132 !work_handlers[opcode]) {
4133 pr_err("No handler for opcode 0x%x.\n", opcode);
4136 ret = work_handlers[opcode](dev, skb);
4140 process_timedout_eps();
4144 static DECLARE_WORK(skb_work, process_work);
4146 static void ep_timeout(struct timer_list *t)
4148 struct c4iw_ep *ep = from_timer(ep, t, timer);
4151 spin_lock(&timeout_lock);
4152 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4154 * Only insert if it is not already on the list.
4156 if (!ep->entry.next) {
4157 list_add_tail(&ep->entry, &timeout_list);
4161 spin_unlock(&timeout_lock);
4163 queue_work(workq, &skb_work);
4167 * All the CM events are handled on a work queue to have a safe context.
4169 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4173 * Save dev in the skb->cb area.
4175 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4178 * Queue the skb and schedule the worker thread.
4180 skb_queue_tail(&rxq, skb);
4181 queue_work(workq, &skb_work);
4185 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4187 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4189 if (rpl->status != CPL_ERR_NONE) {
4190 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4191 rpl->status, GET_TID(rpl));
4197 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4199 struct cpl_fw6_msg *rpl = cplhdr(skb);
4200 struct c4iw_wr_wait *wr_waitp;
4203 pr_debug("type %u\n", rpl->type);
4205 switch (rpl->type) {
4206 case FW6_TYPE_WR_RPL:
4207 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4208 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4209 pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4211 c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4215 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4219 pr_err("%s unexpected fw6 msg type %u\n",
4220 __func__, rpl->type);
4227 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4229 struct cpl_abort_req_rss *req = cplhdr(skb);
4231 unsigned int tid = GET_TID(req);
4233 ep = get_ep_from_tid(dev, tid);
4234 /* This EP will be dereferenced in peer_abort() */
4236 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4240 if (cxgb_is_neg_adv(req->status)) {
4241 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4242 ep->hwtid, req->status,
4243 neg_adv_str(req->status));
4246 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4248 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4255 * Most upcalls from the T4 Core go to sched() to
4256 * schedule the processing on a work queue.
4258 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4259 [CPL_ACT_ESTABLISH] = sched,
4260 [CPL_ACT_OPEN_RPL] = sched,
4261 [CPL_RX_DATA] = sched,
4262 [CPL_ABORT_RPL_RSS] = sched,
4263 [CPL_ABORT_RPL] = sched,
4264 [CPL_PASS_OPEN_RPL] = sched,
4265 [CPL_CLOSE_LISTSRV_RPL] = sched,
4266 [CPL_PASS_ACCEPT_REQ] = sched,
4267 [CPL_PASS_ESTABLISH] = sched,
4268 [CPL_PEER_CLOSE] = sched,
4269 [CPL_CLOSE_CON_RPL] = sched,
4270 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4271 [CPL_RDMA_TERMINATE] = sched,
4272 [CPL_FW4_ACK] = sched,
4273 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4274 [CPL_FW6_MSG] = fw6_msg,
4275 [CPL_RX_PKT] = sched
4278 int __init c4iw_cm_init(void)
4280 spin_lock_init(&timeout_lock);
4281 skb_queue_head_init(&rxq);
4283 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4290 void c4iw_cm_term(void)
4292 WARN_ON(!list_empty(&timeout_list));
4293 flush_workqueue(workq);
4294 destroy_workqueue(workq);