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)");
103 module_param(c4iw_debug, int, 0644);
104 MODULE_PARM_DESC(c4iw_debug, "obsolete");
106 static int peer2peer = 1;
107 module_param(peer2peer, int, 0644);
108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111 module_param(p2p_type, int, 0644);
112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
115 static int ep_timeout_secs = 60;
116 module_param(ep_timeout_secs, int, 0644);
117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 "in seconds (default=60)");
120 static int mpa_rev = 2;
121 module_param(mpa_rev, int, 0644);
122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
124 " compliant (default=2)");
126 static int markers_enabled;
127 module_param(markers_enabled, int, 0644);
128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
130 static int crc_enabled = 1;
131 module_param(crc_enabled, int, 0644);
132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
134 static int rcv_win = 256 * 1024;
135 module_param(rcv_win, int, 0644);
136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
138 static int snd_win = 128 * 1024;
139 module_param(snd_win, int, 0644);
140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
142 static struct workqueue_struct *workq;
144 static struct sk_buff_head rxq;
146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147 static void ep_timeout(unsigned long arg);
148 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
151 static LIST_HEAD(timeout_list);
152 static spinlock_t timeout_lock;
154 static void deref_cm_id(struct c4iw_ep_common *epc)
156 epc->cm_id->rem_ref(epc->cm_id);
158 set_bit(CM_ID_DEREFED, &epc->history);
161 static void ref_cm_id(struct c4iw_ep_common *epc)
163 set_bit(CM_ID_REFED, &epc->history);
164 epc->cm_id->add_ref(epc->cm_id);
167 static void deref_qp(struct c4iw_ep *ep)
169 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 clear_bit(QP_REFERENCED, &ep->com.flags);
171 set_bit(QP_DEREFED, &ep->com.history);
174 static void ref_qp(struct c4iw_ep *ep)
176 set_bit(QP_REFERENCED, &ep->com.flags);
177 set_bit(QP_REFED, &ep->com.history);
178 c4iw_qp_add_ref(&ep->com.qp->ibqp);
181 static void start_ep_timer(struct c4iw_ep *ep)
183 pr_debug("%s ep %p\n", __func__, ep);
184 if (timer_pending(&ep->timer)) {
185 pr_err("%s timer already started! ep %p\n",
189 clear_bit(TIMEOUT, &ep->com.flags);
190 c4iw_get_ep(&ep->com);
191 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 ep->timer.data = (unsigned long)ep;
193 ep->timer.function = ep_timeout;
194 add_timer(&ep->timer);
197 static int stop_ep_timer(struct c4iw_ep *ep)
199 pr_debug("%s ep %p stopping\n", __func__, ep);
200 del_timer_sync(&ep->timer);
201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
202 c4iw_put_ep(&ep->com);
208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 struct l2t_entry *l2e)
213 if (c4iw_fatal_error(rdev)) {
215 pr_debug("%s - device in error state - dropping\n", __func__);
218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
221 else if (error == NET_XMIT_DROP)
223 return error < 0 ? error : 0;
226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
230 if (c4iw_fatal_error(rdev)) {
232 pr_debug("%s - device in error state - dropping\n", __func__);
235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
238 return error < 0 ? error : 0;
241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
243 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
245 skb = get_skb(skb, len, GFP_KERNEL);
249 cxgb_mk_tid_release(skb, len, hwtid, 0);
250 c4iw_ofld_send(rdev, skb);
254 static void set_emss(struct c4iw_ep *ep, u16 opt)
256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
257 ((AF_INET == ep->com.remote_addr.ss_family) ?
258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
259 sizeof(struct tcphdr);
261 if (TCPOPT_TSTAMP_G(opt))
262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
266 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 pr_debug("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
274 enum c4iw_ep_state state;
276 mutex_lock(&epc->mutex);
278 mutex_unlock(&epc->mutex);
282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
289 mutex_lock(&epc->mutex);
290 pr_debug("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
291 __state_set(epc, new);
292 mutex_unlock(&epc->mutex);
296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
302 len = roundup(sizeof(union cpl_wr_size), 16);
303 for (i = 0; i < size; i++) {
304 skb = alloc_skb(len, GFP_KERNEL);
307 skb_queue_tail(ep_skb_list, skb);
311 skb_queue_purge(ep_skb_list);
315 static void *alloc_ep(int size, gfp_t gfp)
317 struct c4iw_ep_common *epc;
319 epc = kzalloc(size, gfp);
321 kref_init(&epc->kref);
322 mutex_init(&epc->mutex);
323 c4iw_init_wr_wait(&epc->wr_wait);
325 pr_debug("%s alloc ep %p\n", __func__, epc);
329 static void remove_ep_tid(struct c4iw_ep *ep)
333 spin_lock_irqsave(&ep->com.dev->lock, flags);
334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
335 if (idr_is_empty(&ep->com.dev->hwtid_idr))
336 wake_up(&ep->com.dev->wait);
337 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
340 static void insert_ep_tid(struct c4iw_ep *ep)
344 spin_lock_irqsave(&ep->com.dev->lock, flags);
345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
346 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357 spin_lock_irqsave(&dev->lock, flags);
358 ep = idr_find(&dev->hwtid_idr, tid);
360 c4iw_get_ep(&ep->com);
361 spin_unlock_irqrestore(&dev->lock, flags);
366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
371 struct c4iw_listen_ep *ep;
374 spin_lock_irqsave(&dev->lock, flags);
375 ep = idr_find(&dev->stid_idr, stid);
377 c4iw_get_ep(&ep->com);
378 spin_unlock_irqrestore(&dev->lock, flags);
382 void _c4iw_free_ep(struct kref *kref)
386 ep = container_of(kref, struct c4iw_ep, com.kref);
387 pr_debug("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
388 if (test_bit(QP_REFERENCED, &ep->com.flags))
390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
391 if (ep->com.remote_addr.ss_family == AF_INET6) {
392 struct sockaddr_in6 *sin6 =
393 (struct sockaddr_in6 *)
397 ep->com.dev->rdev.lldi.ports[0],
398 (const u32 *)&sin6->sin6_addr.s6_addr,
401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
402 ep->com.local_addr.ss_family);
403 dst_release(ep->dst);
404 cxgb4_l2t_release(ep->l2t);
406 kfree_skb(ep->mpa_skb);
408 if (!skb_queue_empty(&ep->com.ep_skb_list))
409 skb_queue_purge(&ep->com.ep_skb_list);
413 static void release_ep_resources(struct c4iw_ep *ep)
415 set_bit(RELEASE_RESOURCES, &ep->com.flags);
418 * If we have a hwtid, then remove it from the idr table
419 * so lookups will no longer find this endpoint. Otherwise
420 * we have a race where one thread finds the ep ptr just
421 * before the other thread is freeing the ep memory.
425 c4iw_put_ep(&ep->com);
428 static int status2errno(int status)
433 case CPL_ERR_CONN_RESET:
435 case CPL_ERR_ARP_MISS:
436 return -EHOSTUNREACH;
437 case CPL_ERR_CONN_TIMEDOUT:
439 case CPL_ERR_TCAM_FULL:
441 case CPL_ERR_CONN_EXIST:
449 * Try and reuse skbs already allocated...
451 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
453 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
456 skb_reset_transport_header(skb);
458 skb = alloc_skb(len, gfp);
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);
497 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
501 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
502 c4iw_put_ep(&ep->parent_ep->com);
503 release_ep_resources(ep);
508 * Fake up a special CPL opcode and call sched() so process_work() will call
509 * _put_ep_safe() in a safe context to free the ep resources. This is needed
510 * because ARP error handlers are called in an ATOMIC context, and
511 * _c4iw_free_ep() needs to block.
513 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
516 struct cpl_act_establish *rpl = cplhdr(skb);
518 /* Set our special ARP_FAILURE opcode */
519 rpl->ot.opcode = cpl;
522 * Save ep in the skb->cb area, after where sched() will save the dev
525 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
526 sched(ep->com.dev, skb);
529 /* Handle an ARP failure for an accept */
530 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
532 struct c4iw_ep *ep = handle;
534 pr_err("ARP failure during accept - tid %u - dropping connection\n",
537 __state_set(&ep->com, DEAD);
538 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
542 * Handle an ARP failure for an active open.
544 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
546 struct c4iw_ep *ep = handle;
548 pr_err("ARP failure during connect\n");
549 connect_reply_upcall(ep, -EHOSTUNREACH);
550 __state_set(&ep->com, DEAD);
551 if (ep->com.remote_addr.ss_family == AF_INET6) {
552 struct sockaddr_in6 *sin6 =
553 (struct sockaddr_in6 *)&ep->com.local_addr;
554 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
555 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
557 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
558 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
559 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
563 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
566 static void abort_arp_failure(void *handle, struct sk_buff *skb)
569 struct c4iw_ep *ep = handle;
570 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
571 struct cpl_abort_req *req = cplhdr(skb);
573 pr_debug("%s rdev %p\n", __func__, rdev);
574 req->cmd = CPL_ABORT_NO_RST;
576 ret = c4iw_ofld_send(rdev, skb);
578 __state_set(&ep->com, DEAD);
579 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
584 static int send_flowc(struct c4iw_ep *ep)
586 struct fw_flowc_wr *flowc;
587 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
589 u16 vlan = ep->l2t->vlan;
595 if (vlan == CPL_L2T_VLAN_NONE)
600 flowc = __skb_put(skb, FLOWC_LEN);
602 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
603 FW_FLOWC_WR_NPARAMS_V(nparams));
604 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
605 16)) | FW_WR_FLOWID_V(ep->hwtid));
607 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
608 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
609 (ep->com.dev->rdev.lldi.pf));
610 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
611 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
612 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
613 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
614 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
615 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
616 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
617 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
618 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
619 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
620 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
621 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
622 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
623 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
627 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
628 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
629 flowc->mnemval[8].val = cpu_to_be32(pri);
631 /* Pad WR to 16 byte boundary */
632 flowc->mnemval[8].mnemonic = 0;
633 flowc->mnemval[8].val = 0;
635 for (i = 0; i < 9; i++) {
636 flowc->mnemval[i].r4[0] = 0;
637 flowc->mnemval[i].r4[1] = 0;
638 flowc->mnemval[i].r4[2] = 0;
641 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
642 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
645 static int send_halfclose(struct c4iw_ep *ep)
647 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
648 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
650 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
654 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
655 NULL, arp_failure_discard);
657 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
660 static int send_abort(struct c4iw_ep *ep)
662 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
663 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
665 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
666 if (WARN_ON(!req_skb))
669 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
670 ep, abort_arp_failure);
672 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
675 static int send_connect(struct c4iw_ep *ep)
677 struct cpl_act_open_req *req = NULL;
678 struct cpl_t5_act_open_req *t5req = NULL;
679 struct cpl_t6_act_open_req *t6req = NULL;
680 struct cpl_act_open_req6 *req6 = NULL;
681 struct cpl_t5_act_open_req6 *t5req6 = NULL;
682 struct cpl_t6_act_open_req6 *t6req6 = NULL;
686 unsigned int mtu_idx;
688 int win, sizev4, sizev6, wrlen;
689 struct sockaddr_in *la = (struct sockaddr_in *)
691 struct sockaddr_in *ra = (struct sockaddr_in *)
692 &ep->com.remote_addr;
693 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
695 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
696 &ep->com.remote_addr;
698 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
699 u32 isn = (prandom_u32() & ~7UL) - 1;
700 struct net_device *netdev;
703 netdev = ep->com.dev->rdev.lldi.ports[0];
705 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
707 sizev4 = sizeof(struct cpl_act_open_req);
708 sizev6 = sizeof(struct cpl_act_open_req6);
711 sizev4 = sizeof(struct cpl_t5_act_open_req);
712 sizev6 = sizeof(struct cpl_t5_act_open_req6);
715 sizev4 = sizeof(struct cpl_t6_act_open_req);
716 sizev6 = sizeof(struct cpl_t6_act_open_req6);
719 pr_err("T%d Chip is not supported\n",
720 CHELSIO_CHIP_VERSION(adapter_type));
724 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
725 roundup(sizev4, 16) :
728 pr_debug("%s ep %p atid %u\n", __func__, ep, ep->atid);
730 skb = get_skb(NULL, wrlen, GFP_KERNEL);
732 pr_err("%s - failed to alloc skb\n", __func__);
735 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
737 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
738 enable_tcp_timestamps,
739 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
740 wscale = cxgb_compute_wscale(rcv_win);
743 * Specify the largest window that will fit in opt0. The
744 * remainder will be specified in the rx_data_ack.
746 win = ep->rcv_win >> 10;
747 if (win > RCV_BUFSIZ_M)
750 opt0 = (nocong ? NO_CONG_F : 0) |
753 WND_SCALE_V(wscale) |
755 L2T_IDX_V(ep->l2t->idx) |
756 TX_CHAN_V(ep->tx_chan) |
757 SMAC_SEL_V(ep->smac_idx) |
758 DSCP_V(ep->tos >> 2) |
759 ULP_MODE_V(ULP_MODE_TCPDDP) |
761 opt2 = RX_CHANNEL_V(0) |
762 CCTRL_ECN_V(enable_ecn) |
763 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
764 if (enable_tcp_timestamps)
765 opt2 |= TSTAMPS_EN_F;
768 if (wscale && enable_tcp_window_scaling)
769 opt2 |= WND_SCALE_EN_F;
770 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
774 opt2 |= T5_OPT_2_VALID_F;
775 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
779 params = cxgb4_select_ntuple(netdev, ep->l2t);
781 if (ep->com.remote_addr.ss_family == AF_INET6)
782 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
783 (const u32 *)&la6->sin6_addr.s6_addr, 1);
785 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
787 if (ep->com.remote_addr.ss_family == AF_INET) {
788 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
790 req = skb_put(skb, wrlen);
794 t5req = skb_put(skb, wrlen);
795 INIT_TP_WR(t5req, 0);
796 req = (struct cpl_act_open_req *)t5req;
799 t6req = skb_put(skb, wrlen);
800 INIT_TP_WR(t6req, 0);
801 req = (struct cpl_act_open_req *)t6req;
802 t5req = (struct cpl_t5_act_open_req *)t6req;
805 pr_err("T%d Chip is not supported\n",
806 CHELSIO_CHIP_VERSION(adapter_type));
811 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
812 ((ep->rss_qid<<14) | ep->atid)));
813 req->local_port = la->sin_port;
814 req->peer_port = ra->sin_port;
815 req->local_ip = la->sin_addr.s_addr;
816 req->peer_ip = ra->sin_addr.s_addr;
817 req->opt0 = cpu_to_be64(opt0);
819 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
820 req->params = cpu_to_be32(params);
821 req->opt2 = cpu_to_be32(opt2);
823 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
825 cpu_to_be64(FILTER_TUPLE_V(params));
826 t5req->rsvd = cpu_to_be32(isn);
827 pr_debug("%s snd_isn %u\n", __func__, t5req->rsvd);
828 t5req->opt2 = cpu_to_be32(opt2);
831 cpu_to_be64(FILTER_TUPLE_V(params));
832 t6req->rsvd = cpu_to_be32(isn);
833 pr_debug("%s snd_isn %u\n", __func__, t6req->rsvd);
834 t6req->opt2 = cpu_to_be32(opt2);
838 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
840 req6 = skb_put(skb, wrlen);
844 t5req6 = skb_put(skb, wrlen);
845 INIT_TP_WR(t5req6, 0);
846 req6 = (struct cpl_act_open_req6 *)t5req6;
849 t6req6 = skb_put(skb, wrlen);
850 INIT_TP_WR(t6req6, 0);
851 req6 = (struct cpl_act_open_req6 *)t6req6;
852 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
855 pr_err("T%d Chip is not supported\n",
856 CHELSIO_CHIP_VERSION(adapter_type));
861 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
862 ((ep->rss_qid<<14)|ep->atid)));
863 req6->local_port = la6->sin6_port;
864 req6->peer_port = ra6->sin6_port;
865 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
866 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
867 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
868 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
869 req6->opt0 = cpu_to_be64(opt0);
871 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
872 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
874 req6->opt2 = cpu_to_be32(opt2);
876 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
878 cpu_to_be64(FILTER_TUPLE_V(params));
879 t5req6->rsvd = cpu_to_be32(isn);
880 pr_debug("%s snd_isn %u\n", __func__, t5req6->rsvd);
881 t5req6->opt2 = cpu_to_be32(opt2);
884 cpu_to_be64(FILTER_TUPLE_V(params));
885 t6req6->rsvd = cpu_to_be32(isn);
886 pr_debug("%s snd_isn %u\n", __func__, t6req6->rsvd);
887 t6req6->opt2 = cpu_to_be32(opt2);
893 set_bit(ACT_OPEN_REQ, &ep->com.history);
894 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
896 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
897 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
898 (const u32 *)&la6->sin6_addr.s6_addr, 1);
902 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
905 int mpalen, wrlen, ret;
906 struct fw_ofld_tx_data_wr *req;
907 struct mpa_message *mpa;
908 struct mpa_v2_conn_params mpa_v2_params;
910 pr_debug("%s ep %p tid %u pd_len %d\n",
911 __func__, ep, ep->hwtid, ep->plen);
913 BUG_ON(skb_cloned(skb));
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("%s initiator ird %u ord %u\n", __func__, 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);
999 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1003 __state_set(&ep->com, MPA_REQ_SENT);
1004 ep->mpa_attr.initiator = 1;
1005 ep->snd_seq += mpalen;
1009 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1012 struct fw_ofld_tx_data_wr *req;
1013 struct mpa_message *mpa;
1014 struct sk_buff *skb;
1015 struct mpa_v2_conn_params mpa_v2_params;
1017 pr_debug("%s ep %p tid %u pd_len %d\n",
1018 __func__, ep, ep->hwtid, ep->plen);
1020 mpalen = sizeof(*mpa) + plen;
1021 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1022 mpalen += sizeof(struct mpa_v2_conn_params);
1023 wrlen = roundup(mpalen + sizeof *req, 16);
1025 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1027 pr_err("%s - cannot alloc skb!\n", __func__);
1030 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1032 req = skb_put_zero(skb, wrlen);
1033 req->op_to_immdlen = cpu_to_be32(
1034 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1036 FW_WR_IMMDLEN_V(mpalen));
1037 req->flowid_len16 = cpu_to_be32(
1038 FW_WR_FLOWID_V(ep->hwtid) |
1039 FW_WR_LEN16_V(wrlen >> 4));
1040 req->plen = cpu_to_be32(mpalen);
1041 req->tunnel_to_proxy = cpu_to_be32(
1042 FW_OFLD_TX_DATA_WR_FLUSH_F |
1043 FW_OFLD_TX_DATA_WR_SHOVE_F);
1045 mpa = (struct mpa_message *)(req + 1);
1046 memset(mpa, 0, sizeof(*mpa));
1047 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1048 mpa->flags = MPA_REJECT;
1049 mpa->revision = ep->mpa_attr.version;
1050 mpa->private_data_size = htons(plen);
1052 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1053 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1054 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1055 sizeof (struct mpa_v2_conn_params));
1056 mpa_v2_params.ird = htons(((u16)ep->ird) |
1057 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1059 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1061 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1062 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1063 FW_RI_INIT_P2PTYPE_READ_REQ ?
1064 MPA_V2_RDMA_READ_RTR : 0) : 0));
1065 memcpy(mpa->private_data, &mpa_v2_params,
1066 sizeof(struct mpa_v2_conn_params));
1069 memcpy(mpa->private_data +
1070 sizeof(struct mpa_v2_conn_params), pdata, plen);
1073 memcpy(mpa->private_data, pdata, plen);
1076 * Reference the mpa skb again. This ensures the data area
1077 * will remain in memory until the hw acks the tx.
1078 * Function fw4_ack() will deref it.
1081 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1082 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1083 BUG_ON(ep->mpa_skb);
1085 ep->snd_seq += mpalen;
1086 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1089 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1092 struct fw_ofld_tx_data_wr *req;
1093 struct mpa_message *mpa;
1094 struct sk_buff *skb;
1095 struct mpa_v2_conn_params mpa_v2_params;
1097 pr_debug("%s ep %p tid %u pd_len %d\n",
1098 __func__, ep, ep->hwtid, ep->plen);
1100 mpalen = sizeof(*mpa) + plen;
1101 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1102 mpalen += sizeof(struct mpa_v2_conn_params);
1103 wrlen = roundup(mpalen + sizeof *req, 16);
1105 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1107 pr_err("%s - cannot alloc skb!\n", __func__);
1110 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1112 req = skb_put_zero(skb, wrlen);
1113 req->op_to_immdlen = cpu_to_be32(
1114 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1116 FW_WR_IMMDLEN_V(mpalen));
1117 req->flowid_len16 = cpu_to_be32(
1118 FW_WR_FLOWID_V(ep->hwtid) |
1119 FW_WR_LEN16_V(wrlen >> 4));
1120 req->plen = cpu_to_be32(mpalen);
1121 req->tunnel_to_proxy = cpu_to_be32(
1122 FW_OFLD_TX_DATA_WR_FLUSH_F |
1123 FW_OFLD_TX_DATA_WR_SHOVE_F);
1125 mpa = (struct mpa_message *)(req + 1);
1126 memset(mpa, 0, sizeof(*mpa));
1127 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1129 if (ep->mpa_attr.crc_enabled)
1130 mpa->flags |= MPA_CRC;
1131 if (ep->mpa_attr.recv_marker_enabled)
1132 mpa->flags |= MPA_MARKERS;
1133 mpa->revision = ep->mpa_attr.version;
1134 mpa->private_data_size = htons(plen);
1136 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1137 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1138 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1139 sizeof (struct mpa_v2_conn_params));
1140 mpa_v2_params.ird = htons((u16)ep->ird);
1141 mpa_v2_params.ord = htons((u16)ep->ord);
1142 if (peer2peer && (ep->mpa_attr.p2p_type !=
1143 FW_RI_INIT_P2PTYPE_DISABLED)) {
1144 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1146 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1147 mpa_v2_params.ord |=
1148 htons(MPA_V2_RDMA_WRITE_RTR);
1149 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1150 mpa_v2_params.ord |=
1151 htons(MPA_V2_RDMA_READ_RTR);
1154 memcpy(mpa->private_data, &mpa_v2_params,
1155 sizeof(struct mpa_v2_conn_params));
1158 memcpy(mpa->private_data +
1159 sizeof(struct mpa_v2_conn_params), pdata, plen);
1162 memcpy(mpa->private_data, pdata, plen);
1165 * Reference the mpa skb. This ensures the data area
1166 * will remain in memory until the hw acks the tx.
1167 * Function fw4_ack() will deref it.
1170 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1172 __state_set(&ep->com, MPA_REP_SENT);
1173 ep->snd_seq += mpalen;
1174 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1177 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1180 struct cpl_act_establish *req = cplhdr(skb);
1181 unsigned int tid = GET_TID(req);
1182 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1183 struct tid_info *t = dev->rdev.lldi.tids;
1186 ep = lookup_atid(t, atid);
1188 pr_debug("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1189 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1191 mutex_lock(&ep->com.mutex);
1192 dst_confirm(ep->dst);
1194 /* setup the hwtid for this connection */
1196 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1199 ep->snd_seq = be32_to_cpu(req->snd_isn);
1200 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1202 set_emss(ep, ntohs(req->tcp_opt));
1204 /* dealloc the atid */
1205 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1206 cxgb4_free_atid(t, atid);
1207 set_bit(ACT_ESTAB, &ep->com.history);
1209 /* start MPA negotiation */
1210 ret = send_flowc(ep);
1213 if (ep->retry_with_mpa_v1)
1214 ret = send_mpa_req(ep, skb, 1);
1216 ret = send_mpa_req(ep, skb, mpa_rev);
1219 mutex_unlock(&ep->com.mutex);
1222 mutex_unlock(&ep->com.mutex);
1223 connect_reply_upcall(ep, -ENOMEM);
1224 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1228 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1230 struct iw_cm_event event;
1232 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1233 memset(&event, 0, sizeof(event));
1234 event.event = IW_CM_EVENT_CLOSE;
1235 event.status = status;
1236 if (ep->com.cm_id) {
1237 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1238 ep, ep->com.cm_id, ep->hwtid);
1239 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1240 deref_cm_id(&ep->com);
1241 set_bit(CLOSE_UPCALL, &ep->com.history);
1245 static void peer_close_upcall(struct c4iw_ep *ep)
1247 struct iw_cm_event event;
1249 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1250 memset(&event, 0, sizeof(event));
1251 event.event = IW_CM_EVENT_DISCONNECT;
1252 if (ep->com.cm_id) {
1253 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1254 ep, ep->com.cm_id, ep->hwtid);
1255 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1256 set_bit(DISCONN_UPCALL, &ep->com.history);
1260 static void peer_abort_upcall(struct c4iw_ep *ep)
1262 struct iw_cm_event event;
1264 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1265 memset(&event, 0, sizeof(event));
1266 event.event = IW_CM_EVENT_CLOSE;
1267 event.status = -ECONNRESET;
1268 if (ep->com.cm_id) {
1269 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1270 ep->com.cm_id, ep->hwtid);
1271 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1272 deref_cm_id(&ep->com);
1273 set_bit(ABORT_UPCALL, &ep->com.history);
1277 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1279 struct iw_cm_event event;
1281 pr_debug("%s ep %p tid %u status %d\n",
1282 __func__, ep, ep->hwtid, status);
1283 memset(&event, 0, sizeof(event));
1284 event.event = IW_CM_EVENT_CONNECT_REPLY;
1285 event.status = status;
1286 memcpy(&event.local_addr, &ep->com.local_addr,
1287 sizeof(ep->com.local_addr));
1288 memcpy(&event.remote_addr, &ep->com.remote_addr,
1289 sizeof(ep->com.remote_addr));
1291 if ((status == 0) || (status == -ECONNREFUSED)) {
1292 if (!ep->tried_with_mpa_v1) {
1293 /* this means MPA_v2 is used */
1294 event.ord = ep->ird;
1295 event.ird = ep->ord;
1296 event.private_data_len = ep->plen -
1297 sizeof(struct mpa_v2_conn_params);
1298 event.private_data = ep->mpa_pkt +
1299 sizeof(struct mpa_message) +
1300 sizeof(struct mpa_v2_conn_params);
1302 /* this means MPA_v1 is used */
1303 event.ord = cur_max_read_depth(ep->com.dev);
1304 event.ird = cur_max_read_depth(ep->com.dev);
1305 event.private_data_len = ep->plen;
1306 event.private_data = ep->mpa_pkt +
1307 sizeof(struct mpa_message);
1311 pr_debug("%s ep %p tid %u status %d\n", __func__, ep,
1313 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1314 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1317 deref_cm_id(&ep->com);
1320 static int connect_request_upcall(struct c4iw_ep *ep)
1322 struct iw_cm_event event;
1325 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1326 memset(&event, 0, sizeof(event));
1327 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1328 memcpy(&event.local_addr, &ep->com.local_addr,
1329 sizeof(ep->com.local_addr));
1330 memcpy(&event.remote_addr, &ep->com.remote_addr,
1331 sizeof(ep->com.remote_addr));
1332 event.provider_data = ep;
1333 if (!ep->tried_with_mpa_v1) {
1334 /* this means MPA_v2 is used */
1335 event.ord = ep->ord;
1336 event.ird = ep->ird;
1337 event.private_data_len = ep->plen -
1338 sizeof(struct mpa_v2_conn_params);
1339 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1340 sizeof(struct mpa_v2_conn_params);
1342 /* this means MPA_v1 is used. Send max supported */
1343 event.ord = cur_max_read_depth(ep->com.dev);
1344 event.ird = cur_max_read_depth(ep->com.dev);
1345 event.private_data_len = ep->plen;
1346 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1348 c4iw_get_ep(&ep->com);
1349 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1352 c4iw_put_ep(&ep->com);
1353 set_bit(CONNREQ_UPCALL, &ep->com.history);
1354 c4iw_put_ep(&ep->parent_ep->com);
1358 static void established_upcall(struct c4iw_ep *ep)
1360 struct iw_cm_event event;
1362 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1363 memset(&event, 0, sizeof(event));
1364 event.event = IW_CM_EVENT_ESTABLISHED;
1365 event.ird = ep->ord;
1366 event.ord = ep->ird;
1367 if (ep->com.cm_id) {
1368 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1369 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1370 set_bit(ESTAB_UPCALL, &ep->com.history);
1374 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1376 struct sk_buff *skb;
1377 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1380 pr_debug("%s ep %p tid %u credits %u\n",
1381 __func__, ep, ep->hwtid, credits);
1382 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1384 pr_err("update_rx_credits - cannot alloc skb!\n");
1389 * If we couldn't specify the entire rcv window at connection setup
1390 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1391 * then add the overage in to the credits returned.
1393 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1394 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1396 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1397 RX_DACK_MODE_V(dack_mode);
1399 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1402 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1406 #define RELAXED_IRD_NEGOTIATION 1
1409 * process_mpa_reply - process streaming mode MPA reply
1413 * 0 upon success indicating a connect request was delivered to the ULP
1414 * or the mpa request is incomplete but valid so far.
1416 * 1 if a failure requires the caller to close the connection.
1418 * 2 if a failure requires the caller to abort the connection.
1420 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1422 struct mpa_message *mpa;
1423 struct mpa_v2_conn_params *mpa_v2_params;
1425 u16 resp_ird, resp_ord;
1426 u8 rtr_mismatch = 0, insuff_ird = 0;
1427 struct c4iw_qp_attributes attrs;
1428 enum c4iw_qp_attr_mask mask;
1432 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1435 * If we get more than the supported amount of private data
1436 * then we must fail this connection.
1438 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1440 goto err_stop_timer;
1444 * copy the new data into our accumulation buffer.
1446 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1448 ep->mpa_pkt_len += skb->len;
1451 * if we don't even have the mpa message, then bail.
1453 if (ep->mpa_pkt_len < sizeof(*mpa))
1455 mpa = (struct mpa_message *) ep->mpa_pkt;
1457 /* Validate MPA header. */
1458 if (mpa->revision > mpa_rev) {
1459 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1460 __func__, mpa_rev, mpa->revision);
1462 goto err_stop_timer;
1464 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1466 goto err_stop_timer;
1469 plen = ntohs(mpa->private_data_size);
1472 * Fail if there's too much private data.
1474 if (plen > MPA_MAX_PRIVATE_DATA) {
1476 goto err_stop_timer;
1480 * If plen does not account for pkt size
1482 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1484 goto err_stop_timer;
1487 ep->plen = (u8) plen;
1490 * If we don't have all the pdata yet, then bail.
1491 * We'll continue process when more data arrives.
1493 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1496 if (mpa->flags & MPA_REJECT) {
1497 err = -ECONNREFUSED;
1498 goto err_stop_timer;
1502 * Stop mpa timer. If it expired, then
1503 * we ignore the MPA reply. process_timeout()
1504 * will abort the connection.
1506 if (stop_ep_timer(ep))
1510 * If we get here we have accumulated the entire mpa
1511 * start reply message including private data. And
1512 * the MPA header is valid.
1514 __state_set(&ep->com, FPDU_MODE);
1515 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1516 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1517 ep->mpa_attr.version = mpa->revision;
1518 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1520 if (mpa->revision == 2) {
1521 ep->mpa_attr.enhanced_rdma_conn =
1522 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1523 if (ep->mpa_attr.enhanced_rdma_conn) {
1524 mpa_v2_params = (struct mpa_v2_conn_params *)
1525 (ep->mpa_pkt + sizeof(*mpa));
1526 resp_ird = ntohs(mpa_v2_params->ird) &
1527 MPA_V2_IRD_ORD_MASK;
1528 resp_ord = ntohs(mpa_v2_params->ord) &
1529 MPA_V2_IRD_ORD_MASK;
1530 pr_debug("%s responder ird %u ord %u ep ird %u ord %u\n",
1532 resp_ird, resp_ord, ep->ird, ep->ord);
1535 * This is a double-check. Ideally, below checks are
1536 * not required since ird/ord stuff has been taken
1537 * care of in c4iw_accept_cr
1539 if (ep->ird < resp_ord) {
1540 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1541 ep->com.dev->rdev.lldi.max_ordird_qp)
1545 } else if (ep->ird > resp_ord) {
1548 if (ep->ord > resp_ird) {
1549 if (RELAXED_IRD_NEGOTIATION)
1560 if (ntohs(mpa_v2_params->ird) &
1561 MPA_V2_PEER2PEER_MODEL) {
1562 if (ntohs(mpa_v2_params->ord) &
1563 MPA_V2_RDMA_WRITE_RTR)
1564 ep->mpa_attr.p2p_type =
1565 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1566 else if (ntohs(mpa_v2_params->ord) &
1567 MPA_V2_RDMA_READ_RTR)
1568 ep->mpa_attr.p2p_type =
1569 FW_RI_INIT_P2PTYPE_READ_REQ;
1572 } else if (mpa->revision == 1)
1574 ep->mpa_attr.p2p_type = p2p_type;
1576 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1577 __func__, ep->mpa_attr.crc_enabled,
1578 ep->mpa_attr.recv_marker_enabled,
1579 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1580 ep->mpa_attr.p2p_type, p2p_type);
1583 * If responder's RTR does not match with that of initiator, assign
1584 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1585 * generated when moving QP to RTS state.
1586 * A TERM message will be sent after QP has moved to RTS state
1588 if ((ep->mpa_attr.version == 2) && peer2peer &&
1589 (ep->mpa_attr.p2p_type != p2p_type)) {
1590 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1594 attrs.mpa_attr = ep->mpa_attr;
1595 attrs.max_ird = ep->ird;
1596 attrs.max_ord = ep->ord;
1597 attrs.llp_stream_handle = ep;
1598 attrs.next_state = C4IW_QP_STATE_RTS;
1600 mask = C4IW_QP_ATTR_NEXT_STATE |
1601 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1602 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1604 /* bind QP and TID with INIT_WR */
1605 err = c4iw_modify_qp(ep->com.qp->rhp,
1606 ep->com.qp, mask, &attrs, 1);
1611 * If responder's RTR requirement did not match with what initiator
1612 * supports, generate TERM message
1615 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1616 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1617 attrs.ecode = MPA_NOMATCH_RTR;
1618 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1619 attrs.send_term = 1;
1620 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1621 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1628 * Generate TERM if initiator IRD is not sufficient for responder
1629 * provided ORD. Currently, we do the same behaviour even when
1630 * responder provided IRD is also not sufficient as regards to
1634 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1635 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1636 attrs.ecode = MPA_INSUFF_IRD;
1637 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1638 attrs.send_term = 1;
1639 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1640 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1651 connect_reply_upcall(ep, err);
1656 * process_mpa_request - process streaming mode MPA request
1660 * 0 upon success indicating a connect request was delivered to the ULP
1661 * or the mpa request is incomplete but valid so far.
1663 * 1 if a failure requires the caller to close the connection.
1665 * 2 if a failure requires the caller to abort the connection.
1667 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1669 struct mpa_message *mpa;
1670 struct mpa_v2_conn_params *mpa_v2_params;
1673 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1676 * If we get more than the supported amount of private data
1677 * then we must fail this connection.
1679 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1680 goto err_stop_timer;
1682 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1685 * Copy the new data into our accumulation buffer.
1687 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1689 ep->mpa_pkt_len += skb->len;
1692 * If we don't even have the mpa message, then bail.
1693 * We'll continue process when more data arrives.
1695 if (ep->mpa_pkt_len < sizeof(*mpa))
1698 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1699 mpa = (struct mpa_message *) ep->mpa_pkt;
1702 * Validate MPA Header.
1704 if (mpa->revision > mpa_rev) {
1705 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1706 __func__, mpa_rev, mpa->revision);
1707 goto err_stop_timer;
1710 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1711 goto err_stop_timer;
1713 plen = ntohs(mpa->private_data_size);
1716 * Fail if there's too much private data.
1718 if (plen > MPA_MAX_PRIVATE_DATA)
1719 goto err_stop_timer;
1722 * If plen does not account for pkt size
1724 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1725 goto err_stop_timer;
1726 ep->plen = (u8) plen;
1729 * If we don't have all the pdata yet, then bail.
1731 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1735 * If we get here we have accumulated the entire mpa
1736 * start reply message including private data.
1738 ep->mpa_attr.initiator = 0;
1739 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1740 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1741 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1742 ep->mpa_attr.version = mpa->revision;
1743 if (mpa->revision == 1)
1744 ep->tried_with_mpa_v1 = 1;
1745 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1747 if (mpa->revision == 2) {
1748 ep->mpa_attr.enhanced_rdma_conn =
1749 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1750 if (ep->mpa_attr.enhanced_rdma_conn) {
1751 mpa_v2_params = (struct mpa_v2_conn_params *)
1752 (ep->mpa_pkt + sizeof(*mpa));
1753 ep->ird = ntohs(mpa_v2_params->ird) &
1754 MPA_V2_IRD_ORD_MASK;
1755 ep->ird = min_t(u32, ep->ird,
1756 cur_max_read_depth(ep->com.dev));
1757 ep->ord = ntohs(mpa_v2_params->ord) &
1758 MPA_V2_IRD_ORD_MASK;
1759 ep->ord = min_t(u32, ep->ord,
1760 cur_max_read_depth(ep->com.dev));
1761 pr_debug("%s initiator ird %u ord %u\n",
1762 __func__, ep->ird, ep->ord);
1763 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1765 if (ntohs(mpa_v2_params->ord) &
1766 MPA_V2_RDMA_WRITE_RTR)
1767 ep->mpa_attr.p2p_type =
1768 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1769 else if (ntohs(mpa_v2_params->ord) &
1770 MPA_V2_RDMA_READ_RTR)
1771 ep->mpa_attr.p2p_type =
1772 FW_RI_INIT_P2PTYPE_READ_REQ;
1775 } else if (mpa->revision == 1)
1777 ep->mpa_attr.p2p_type = p2p_type;
1779 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1781 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1782 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1783 ep->mpa_attr.p2p_type);
1785 __state_set(&ep->com, MPA_REQ_RCVD);
1788 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1789 if (ep->parent_ep->com.state != DEAD) {
1790 if (connect_request_upcall(ep))
1791 goto err_unlock_parent;
1793 goto err_unlock_parent;
1795 mutex_unlock(&ep->parent_ep->com.mutex);
1799 mutex_unlock(&ep->parent_ep->com.mutex);
1802 (void)stop_ep_timer(ep);
1807 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1810 struct cpl_rx_data *hdr = cplhdr(skb);
1811 unsigned int dlen = ntohs(hdr->len);
1812 unsigned int tid = GET_TID(hdr);
1813 __u8 status = hdr->status;
1816 ep = get_ep_from_tid(dev, tid);
1819 pr_debug("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1820 skb_pull(skb, sizeof(*hdr));
1821 skb_trim(skb, dlen);
1822 mutex_lock(&ep->com.mutex);
1824 switch (ep->com.state) {
1826 update_rx_credits(ep, dlen);
1827 ep->rcv_seq += dlen;
1828 disconnect = process_mpa_reply(ep, skb);
1831 update_rx_credits(ep, dlen);
1832 ep->rcv_seq += dlen;
1833 disconnect = process_mpa_request(ep, skb);
1836 struct c4iw_qp_attributes attrs;
1838 update_rx_credits(ep, dlen);
1839 BUG_ON(!ep->com.qp);
1841 pr_err("%s Unexpected streaming data." \
1842 " qpid %u ep %p state %d tid %u status %d\n",
1843 __func__, ep->com.qp->wq.sq.qid, ep,
1844 ep->com.state, ep->hwtid, status);
1845 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1846 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1847 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1854 mutex_unlock(&ep->com.mutex);
1856 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1857 c4iw_put_ep(&ep->com);
1861 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1864 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1866 unsigned int tid = GET_TID(rpl);
1868 ep = get_ep_from_tid(dev, tid);
1870 pr_warn("Abort rpl to freed endpoint\n");
1873 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1874 mutex_lock(&ep->com.mutex);
1875 switch (ep->com.state) {
1877 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1878 __state_set(&ep->com, DEAD);
1882 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1885 mutex_unlock(&ep->com.mutex);
1888 close_complete_upcall(ep, -ECONNRESET);
1889 release_ep_resources(ep);
1891 c4iw_put_ep(&ep->com);
1895 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1897 struct sk_buff *skb;
1898 struct fw_ofld_connection_wr *req;
1899 unsigned int mtu_idx;
1901 struct sockaddr_in *sin;
1904 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1908 req = __skb_put_zero(skb, sizeof(*req));
1909 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1910 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1911 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1912 ep->com.dev->rdev.lldi.ports[0],
1914 sin = (struct sockaddr_in *)&ep->com.local_addr;
1915 req->le.lport = sin->sin_port;
1916 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1917 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1918 req->le.pport = sin->sin_port;
1919 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1920 req->tcb.t_state_to_astid =
1921 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1922 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1923 req->tcb.cplrxdataack_cplpassacceptrpl =
1924 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1925 req->tcb.tx_max = (__force __be32) jiffies;
1926 req->tcb.rcv_adv = htons(1);
1927 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1928 enable_tcp_timestamps,
1929 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1930 wscale = cxgb_compute_wscale(rcv_win);
1933 * Specify the largest window that will fit in opt0. The
1934 * remainder will be specified in the rx_data_ack.
1936 win = ep->rcv_win >> 10;
1937 if (win > RCV_BUFSIZ_M)
1940 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1941 (nocong ? NO_CONG_F : 0) |
1944 WND_SCALE_V(wscale) |
1945 MSS_IDX_V(mtu_idx) |
1946 L2T_IDX_V(ep->l2t->idx) |
1947 TX_CHAN_V(ep->tx_chan) |
1948 SMAC_SEL_V(ep->smac_idx) |
1949 DSCP_V(ep->tos >> 2) |
1950 ULP_MODE_V(ULP_MODE_TCPDDP) |
1952 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1953 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1955 CCTRL_ECN_V(enable_ecn) |
1956 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1957 if (enable_tcp_timestamps)
1958 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1959 if (enable_tcp_sack)
1960 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1961 if (wscale && enable_tcp_window_scaling)
1962 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1963 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1964 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1965 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1966 set_bit(ACT_OFLD_CONN, &ep->com.history);
1967 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1971 * Some of the error codes above implicitly indicate that there is no TID
1972 * allocated with the result of an ACT_OPEN. We use this predicate to make
1975 static inline int act_open_has_tid(int status)
1977 return (status != CPL_ERR_TCAM_PARITY &&
1978 status != CPL_ERR_TCAM_MISS &&
1979 status != CPL_ERR_TCAM_FULL &&
1980 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1981 status != CPL_ERR_CONN_EXIST);
1984 static char *neg_adv_str(unsigned int status)
1987 case CPL_ERR_RTX_NEG_ADVICE:
1988 return "Retransmit timeout";
1989 case CPL_ERR_PERSIST_NEG_ADVICE:
1990 return "Persist timeout";
1991 case CPL_ERR_KEEPALV_NEG_ADVICE:
1992 return "Keepalive timeout";
1998 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2000 ep->snd_win = snd_win;
2001 ep->rcv_win = rcv_win;
2002 pr_debug("%s snd_win %d rcv_win %d\n",
2003 __func__, ep->snd_win, ep->rcv_win);
2006 #define ACT_OPEN_RETRY_COUNT 2
2008 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2009 struct dst_entry *dst, struct c4iw_dev *cdev,
2010 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2012 struct neighbour *n;
2014 struct net_device *pdev;
2016 n = dst_neigh_lookup(dst, peer_ip);
2022 if (n->dev->flags & IFF_LOOPBACK) {
2024 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2025 else if (IS_ENABLED(CONFIG_IPV6))
2026 for_each_netdev(&init_net, pdev) {
2027 if (ipv6_chk_addr(&init_net,
2028 (struct in6_addr *)peer_ip,
2039 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2040 n, pdev, rt_tos2priority(tos));
2045 ep->mtu = pdev->mtu;
2046 ep->tx_chan = cxgb4_port_chan(pdev);
2047 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2048 cxgb4_port_viid(pdev));
2049 step = cdev->rdev.lldi.ntxq /
2050 cdev->rdev.lldi.nchan;
2051 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2052 step = cdev->rdev.lldi.nrxq /
2053 cdev->rdev.lldi.nchan;
2054 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2055 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2056 cxgb4_port_idx(pdev) * step];
2057 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2060 pdev = get_real_dev(n->dev);
2061 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2062 n, pdev, rt_tos2priority(tos));
2065 ep->mtu = dst_mtu(dst);
2066 ep->tx_chan = cxgb4_port_chan(pdev);
2067 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2068 cxgb4_port_viid(pdev));
2069 step = cdev->rdev.lldi.ntxq /
2070 cdev->rdev.lldi.nchan;
2071 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2072 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2073 step = cdev->rdev.lldi.nrxq /
2074 cdev->rdev.lldi.nchan;
2075 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2076 cxgb4_port_idx(pdev) * step];
2077 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2080 ep->retry_with_mpa_v1 = 0;
2081 ep->tried_with_mpa_v1 = 0;
2093 static int c4iw_reconnect(struct c4iw_ep *ep)
2097 struct sockaddr_in *laddr = (struct sockaddr_in *)
2098 &ep->com.cm_id->m_local_addr;
2099 struct sockaddr_in *raddr = (struct sockaddr_in *)
2100 &ep->com.cm_id->m_remote_addr;
2101 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2102 &ep->com.cm_id->m_local_addr;
2103 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2104 &ep->com.cm_id->m_remote_addr;
2108 pr_debug("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2109 init_timer(&ep->timer);
2110 c4iw_init_wr_wait(&ep->com.wr_wait);
2112 /* When MPA revision is different on nodes, the node with MPA_rev=2
2113 * tries to reconnect with MPA_rev 1 for the same EP through
2114 * c4iw_reconnect(), where the same EP is assigned with new tid for
2115 * further connection establishment. As we are using the same EP pointer
2116 * for reconnect, few skbs are used during the previous c4iw_connect(),
2117 * which leaves the EP with inadequate skbs for further
2118 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2119 * skb_list() during peer_abort(). Allocate skbs which is already used.
2121 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2122 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2128 * Allocate an active TID to initiate a TCP connection.
2130 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2131 if (ep->atid == -1) {
2132 pr_err("%s - cannot alloc atid\n", __func__);
2136 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2139 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2140 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2141 laddr->sin_addr.s_addr,
2142 raddr->sin_addr.s_addr,
2144 raddr->sin_port, ep->com.cm_id->tos);
2146 ra = (__u8 *)&raddr->sin_addr;
2148 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2150 laddr6->sin6_addr.s6_addr,
2151 raddr6->sin6_addr.s6_addr,
2155 raddr6->sin6_scope_id);
2157 ra = (__u8 *)&raddr6->sin6_addr;
2160 pr_err("%s - cannot find route\n", __func__);
2161 err = -EHOSTUNREACH;
2164 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2165 ep->com.dev->rdev.lldi.adapter_type,
2166 ep->com.cm_id->tos);
2168 pr_err("%s - cannot alloc l2e\n", __func__);
2172 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2173 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2176 state_set(&ep->com, CONNECTING);
2177 ep->tos = ep->com.cm_id->tos;
2179 /* send connect request to rnic */
2180 err = send_connect(ep);
2184 cxgb4_l2t_release(ep->l2t);
2186 dst_release(ep->dst);
2188 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2189 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2192 * remember to send notification to upper layer.
2193 * We are in here so the upper layer is not aware that this is
2194 * re-connect attempt and so, upper layer is still waiting for
2195 * response of 1st connect request.
2197 connect_reply_upcall(ep, -ECONNRESET);
2199 c4iw_put_ep(&ep->com);
2204 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2207 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2208 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2209 ntohl(rpl->atid_status)));
2210 struct tid_info *t = dev->rdev.lldi.tids;
2211 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2212 struct sockaddr_in *la;
2213 struct sockaddr_in *ra;
2214 struct sockaddr_in6 *la6;
2215 struct sockaddr_in6 *ra6;
2218 ep = lookup_atid(t, atid);
2219 la = (struct sockaddr_in *)&ep->com.local_addr;
2220 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2221 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2222 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2224 pr_debug("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2225 status, status2errno(status));
2227 if (cxgb_is_neg_adv(status)) {
2228 pr_debug("%s Connection problems for atid %u status %u (%s)\n",
2229 __func__, atid, status, neg_adv_str(status));
2230 ep->stats.connect_neg_adv++;
2231 mutex_lock(&dev->rdev.stats.lock);
2232 dev->rdev.stats.neg_adv++;
2233 mutex_unlock(&dev->rdev.stats.lock);
2237 set_bit(ACT_OPEN_RPL, &ep->com.history);
2240 * Log interesting failures.
2243 case CPL_ERR_CONN_RESET:
2244 case CPL_ERR_CONN_TIMEDOUT:
2246 case CPL_ERR_TCAM_FULL:
2247 mutex_lock(&dev->rdev.stats.lock);
2248 dev->rdev.stats.tcam_full++;
2249 mutex_unlock(&dev->rdev.stats.lock);
2250 if (ep->com.local_addr.ss_family == AF_INET &&
2251 dev->rdev.lldi.enable_fw_ofld_conn) {
2252 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2253 ntohl(rpl->atid_status))));
2259 case CPL_ERR_CONN_EXIST:
2260 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2261 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2262 if (ep->com.remote_addr.ss_family == AF_INET6) {
2263 struct sockaddr_in6 *sin6 =
2264 (struct sockaddr_in6 *)
2265 &ep->com.local_addr;
2267 ep->com.dev->rdev.lldi.ports[0],
2269 &sin6->sin6_addr.s6_addr, 1);
2271 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2273 cxgb4_free_atid(t, atid);
2274 dst_release(ep->dst);
2275 cxgb4_l2t_release(ep->l2t);
2281 if (ep->com.local_addr.ss_family == AF_INET) {
2282 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2283 atid, status, status2errno(status),
2284 &la->sin_addr.s_addr, ntohs(la->sin_port),
2285 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2287 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2288 atid, status, status2errno(status),
2289 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2290 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2296 connect_reply_upcall(ep, status2errno(status));
2297 state_set(&ep->com, DEAD);
2299 if (ep->com.remote_addr.ss_family == AF_INET6) {
2300 struct sockaddr_in6 *sin6 =
2301 (struct sockaddr_in6 *)&ep->com.local_addr;
2302 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2303 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2305 if (status && act_open_has_tid(status))
2306 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2307 ep->com.local_addr.ss_family);
2309 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2310 cxgb4_free_atid(t, atid);
2311 dst_release(ep->dst);
2312 cxgb4_l2t_release(ep->l2t);
2313 c4iw_put_ep(&ep->com);
2318 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2320 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2321 unsigned int stid = GET_TID(rpl);
2322 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2325 pr_debug("%s stid %d lookup failure!\n", __func__, stid);
2328 pr_debug("%s ep %p status %d error %d\n", __func__, ep,
2329 rpl->status, status2errno(rpl->status));
2330 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2331 c4iw_put_ep(&ep->com);
2336 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2338 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2339 unsigned int stid = GET_TID(rpl);
2340 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2343 pr_debug("%s stid %d lookup failure!\n", __func__, stid);
2346 pr_debug("%s ep %p\n", __func__, ep);
2347 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2348 c4iw_put_ep(&ep->com);
2353 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2354 struct cpl_pass_accept_req *req)
2356 struct cpl_pass_accept_rpl *rpl;
2357 unsigned int mtu_idx;
2361 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2363 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2365 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2366 BUG_ON(skb_cloned(skb));
2370 if (!is_t4(adapter_type)) {
2371 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2373 INIT_TP_WR(rpl5, ep->hwtid);
2375 skb_trim(skb, sizeof(*rpl));
2376 INIT_TP_WR(rpl, ep->hwtid);
2378 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2381 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2382 enable_tcp_timestamps && req->tcpopt.tstamp,
2383 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2384 wscale = cxgb_compute_wscale(rcv_win);
2387 * Specify the largest window that will fit in opt0. The
2388 * remainder will be specified in the rx_data_ack.
2390 win = ep->rcv_win >> 10;
2391 if (win > RCV_BUFSIZ_M)
2393 opt0 = (nocong ? NO_CONG_F : 0) |
2396 WND_SCALE_V(wscale) |
2397 MSS_IDX_V(mtu_idx) |
2398 L2T_IDX_V(ep->l2t->idx) |
2399 TX_CHAN_V(ep->tx_chan) |
2400 SMAC_SEL_V(ep->smac_idx) |
2401 DSCP_V(ep->tos >> 2) |
2402 ULP_MODE_V(ULP_MODE_TCPDDP) |
2404 opt2 = RX_CHANNEL_V(0) |
2405 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2407 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2408 opt2 |= TSTAMPS_EN_F;
2409 if (enable_tcp_sack && req->tcpopt.sack)
2411 if (wscale && enable_tcp_window_scaling)
2412 opt2 |= WND_SCALE_EN_F;
2414 const struct tcphdr *tcph;
2415 u32 hlen = ntohl(req->hdr_len);
2417 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2418 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2421 tcph = (const void *)(req + 1) +
2422 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2423 if (tcph->ece && tcph->cwr)
2424 opt2 |= CCTRL_ECN_V(1);
2426 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2427 u32 isn = (prandom_u32() & ~7UL) - 1;
2428 opt2 |= T5_OPT_2_VALID_F;
2429 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2432 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2435 rpl5->iss = cpu_to_be32(isn);
2436 pr_debug("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2439 rpl->opt0 = cpu_to_be64(opt0);
2440 rpl->opt2 = cpu_to_be32(opt2);
2441 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2442 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2444 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2447 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2449 pr_debug("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2450 BUG_ON(skb_cloned(skb));
2451 skb_trim(skb, sizeof(struct cpl_tid_release));
2452 release_tid(&dev->rdev, hwtid, skb);
2456 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2458 struct c4iw_ep *child_ep = NULL, *parent_ep;
2459 struct cpl_pass_accept_req *req = cplhdr(skb);
2460 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2461 struct tid_info *t = dev->rdev.lldi.tids;
2462 unsigned int hwtid = GET_TID(req);
2463 struct dst_entry *dst;
2464 __u8 local_ip[16], peer_ip[16];
2465 __be16 local_port, peer_port;
2466 struct sockaddr_in6 *sin6;
2468 u16 peer_mss = ntohs(req->tcpopt.mss);
2470 unsigned short hdrs;
2471 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2473 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2475 pr_debug("%s connect request on invalid stid %d\n",
2480 if (state_read(&parent_ep->com) != LISTEN) {
2481 pr_debug("%s - listening ep not in LISTEN\n", __func__);
2485 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2486 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2488 /* Find output route */
2490 pr_debug("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2491 , __func__, parent_ep, hwtid,
2492 local_ip, peer_ip, ntohs(local_port),
2493 ntohs(peer_port), peer_mss);
2494 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2495 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2496 local_port, peer_port, tos);
2498 pr_debug("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2499 , __func__, parent_ep, hwtid,
2500 local_ip, peer_ip, ntohs(local_port),
2501 ntohs(peer_port), peer_mss);
2502 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2503 local_ip, peer_ip, local_port, peer_port,
2504 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2505 ((struct sockaddr_in6 *)
2506 &parent_ep->com.local_addr)->sin6_scope_id);
2509 pr_err("%s - failed to find dst entry!\n", __func__);
2513 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2515 pr_err("%s - failed to allocate ep entry!\n", __func__);
2520 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2521 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2523 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2529 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2530 sizeof(struct tcphdr) +
2531 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2532 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2533 child_ep->mtu = peer_mss + hdrs;
2535 skb_queue_head_init(&child_ep->com.ep_skb_list);
2536 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2539 state_set(&child_ep->com, CONNECTING);
2540 child_ep->com.dev = dev;
2541 child_ep->com.cm_id = NULL;
2544 struct sockaddr_in *sin = (struct sockaddr_in *)
2545 &child_ep->com.local_addr;
2547 sin->sin_family = AF_INET;
2548 sin->sin_port = local_port;
2549 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2551 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2552 sin->sin_family = AF_INET;
2553 sin->sin_port = ((struct sockaddr_in *)
2554 &parent_ep->com.local_addr)->sin_port;
2555 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2557 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2558 sin->sin_family = AF_INET;
2559 sin->sin_port = peer_port;
2560 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2562 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2563 sin6->sin6_family = PF_INET6;
2564 sin6->sin6_port = local_port;
2565 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2567 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2568 sin6->sin6_family = PF_INET6;
2569 sin6->sin6_port = ((struct sockaddr_in6 *)
2570 &parent_ep->com.local_addr)->sin6_port;
2571 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2573 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2574 sin6->sin6_family = PF_INET6;
2575 sin6->sin6_port = peer_port;
2576 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2579 c4iw_get_ep(&parent_ep->com);
2580 child_ep->parent_ep = parent_ep;
2581 child_ep->tos = tos;
2582 child_ep->dst = dst;
2583 child_ep->hwtid = hwtid;
2585 pr_debug("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2586 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2588 init_timer(&child_ep->timer);
2589 cxgb4_insert_tid(t, child_ep, hwtid,
2590 child_ep->com.local_addr.ss_family);
2591 insert_ep_tid(child_ep);
2592 if (accept_cr(child_ep, skb, req)) {
2593 c4iw_put_ep(&parent_ep->com);
2594 release_ep_resources(child_ep);
2596 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2599 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2600 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2601 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2605 c4iw_put_ep(&child_ep->com);
2607 reject_cr(dev, hwtid, skb);
2610 c4iw_put_ep(&parent_ep->com);
2614 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2617 struct cpl_pass_establish *req = cplhdr(skb);
2618 unsigned int tid = GET_TID(req);
2621 ep = get_ep_from_tid(dev, tid);
2622 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2623 ep->snd_seq = be32_to_cpu(req->snd_isn);
2624 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2626 pr_debug("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2627 ntohs(req->tcp_opt));
2629 set_emss(ep, ntohs(req->tcp_opt));
2631 dst_confirm(ep->dst);
2632 mutex_lock(&ep->com.mutex);
2633 ep->com.state = MPA_REQ_WAIT;
2635 set_bit(PASS_ESTAB, &ep->com.history);
2636 ret = send_flowc(ep);
2637 mutex_unlock(&ep->com.mutex);
2639 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2640 c4iw_put_ep(&ep->com);
2645 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2647 struct cpl_peer_close *hdr = cplhdr(skb);
2649 struct c4iw_qp_attributes attrs;
2652 unsigned int tid = GET_TID(hdr);
2655 ep = get_ep_from_tid(dev, tid);
2659 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2660 dst_confirm(ep->dst);
2662 set_bit(PEER_CLOSE, &ep->com.history);
2663 mutex_lock(&ep->com.mutex);
2664 switch (ep->com.state) {
2666 __state_set(&ep->com, CLOSING);
2669 __state_set(&ep->com, CLOSING);
2670 connect_reply_upcall(ep, -ECONNRESET);
2675 * We're gonna mark this puppy DEAD, but keep
2676 * the reference on it until the ULP accepts or
2677 * rejects the CR. Also wake up anyone waiting
2678 * in rdma connection migration (see c4iw_accept_cr()).
2680 __state_set(&ep->com, CLOSING);
2681 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2682 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2685 __state_set(&ep->com, CLOSING);
2686 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2687 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2691 __state_set(&ep->com, CLOSING);
2692 attrs.next_state = C4IW_QP_STATE_CLOSING;
2693 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2694 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2695 if (ret != -ECONNRESET) {
2696 peer_close_upcall(ep);
2704 __state_set(&ep->com, MORIBUND);
2708 (void)stop_ep_timer(ep);
2709 if (ep->com.cm_id && ep->com.qp) {
2710 attrs.next_state = C4IW_QP_STATE_IDLE;
2711 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2712 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2714 close_complete_upcall(ep, 0);
2715 __state_set(&ep->com, DEAD);
2725 mutex_unlock(&ep->com.mutex);
2727 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2729 release_ep_resources(ep);
2730 c4iw_put_ep(&ep->com);
2734 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2736 struct cpl_abort_req_rss *req = cplhdr(skb);
2738 struct sk_buff *rpl_skb;
2739 struct c4iw_qp_attributes attrs;
2742 unsigned int tid = GET_TID(req);
2743 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2745 ep = get_ep_from_tid(dev, tid);
2749 if (cxgb_is_neg_adv(req->status)) {
2750 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
2751 __func__, ep->hwtid, req->status,
2752 neg_adv_str(req->status));
2753 ep->stats.abort_neg_adv++;
2754 mutex_lock(&dev->rdev.stats.lock);
2755 dev->rdev.stats.neg_adv++;
2756 mutex_unlock(&dev->rdev.stats.lock);
2759 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2761 set_bit(PEER_ABORT, &ep->com.history);
2764 * Wake up any threads in rdma_init() or rdma_fini().
2765 * However, this is not needed if com state is just
2768 if (ep->com.state != MPA_REQ_SENT)
2769 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2771 mutex_lock(&ep->com.mutex);
2772 switch (ep->com.state) {
2774 c4iw_put_ep(&ep->parent_ep->com);
2777 (void)stop_ep_timer(ep);
2780 (void)stop_ep_timer(ep);
2781 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2782 connect_reply_upcall(ep, -ECONNRESET);
2785 * we just don't send notification upwards because we
2786 * want to retry with mpa_v1 without upper layers even
2789 * do some housekeeping so as to re-initiate the
2792 pr_debug("%s: mpa_rev=%d. Retrying with mpav1\n",
2794 ep->retry_with_mpa_v1 = 1;
2806 if (ep->com.cm_id && ep->com.qp) {
2807 attrs.next_state = C4IW_QP_STATE_ERROR;
2808 ret = c4iw_modify_qp(ep->com.qp->rhp,
2809 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2812 pr_err("%s - qp <- error failed!\n", __func__);
2814 peer_abort_upcall(ep);
2819 pr_debug("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2820 mutex_unlock(&ep->com.mutex);
2826 dst_confirm(ep->dst);
2827 if (ep->com.state != ABORTING) {
2828 __state_set(&ep->com, DEAD);
2829 /* we don't release if we want to retry with mpa_v1 */
2830 if (!ep->retry_with_mpa_v1)
2833 mutex_unlock(&ep->com.mutex);
2835 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2836 if (WARN_ON(!rpl_skb)) {
2841 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2843 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2846 release_ep_resources(ep);
2847 else if (ep->retry_with_mpa_v1) {
2848 if (ep->com.remote_addr.ss_family == AF_INET6) {
2849 struct sockaddr_in6 *sin6 =
2850 (struct sockaddr_in6 *)
2851 &ep->com.local_addr;
2853 ep->com.dev->rdev.lldi.ports[0],
2854 (const u32 *)&sin6->sin6_addr.s6_addr,
2857 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2858 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2859 ep->com.local_addr.ss_family);
2860 dst_release(ep->dst);
2861 cxgb4_l2t_release(ep->l2t);
2866 c4iw_put_ep(&ep->com);
2867 /* Dereferencing ep, referenced in peer_abort_intr() */
2868 c4iw_put_ep(&ep->com);
2872 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2875 struct c4iw_qp_attributes attrs;
2876 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2878 unsigned int tid = GET_TID(rpl);
2880 ep = get_ep_from_tid(dev, tid);
2884 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2886 /* The cm_id may be null if we failed to connect */
2887 mutex_lock(&ep->com.mutex);
2888 set_bit(CLOSE_CON_RPL, &ep->com.history);
2889 switch (ep->com.state) {
2891 __state_set(&ep->com, MORIBUND);
2894 (void)stop_ep_timer(ep);
2895 if ((ep->com.cm_id) && (ep->com.qp)) {
2896 attrs.next_state = C4IW_QP_STATE_IDLE;
2897 c4iw_modify_qp(ep->com.qp->rhp,
2899 C4IW_QP_ATTR_NEXT_STATE,
2902 close_complete_upcall(ep, 0);
2903 __state_set(&ep->com, DEAD);
2913 mutex_unlock(&ep->com.mutex);
2915 release_ep_resources(ep);
2916 c4iw_put_ep(&ep->com);
2920 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2922 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2923 unsigned int tid = GET_TID(rpl);
2925 struct c4iw_qp_attributes attrs;
2927 ep = get_ep_from_tid(dev, tid);
2932 pr_warn("TERM received tid %u qpid %u\n", tid,
2933 ep->com.qp->wq.sq.qid);
2934 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2935 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2936 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2939 c4iw_put_ep(&ep->com);
2941 pr_warn("TERM received tid %u no ep/qp\n", tid);
2947 * Upcall from the adapter indicating data has been transmitted.
2948 * For us its just the single MPA request or reply. We can now free
2949 * the skb holding the mpa message.
2951 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2954 struct cpl_fw4_ack *hdr = cplhdr(skb);
2955 u8 credits = hdr->credits;
2956 unsigned int tid = GET_TID(hdr);
2959 ep = get_ep_from_tid(dev, tid);
2962 pr_debug("%s ep %p tid %u credits %u\n",
2963 __func__, ep, ep->hwtid, credits);
2965 pr_debug("%s 0 credit ack ep %p tid %u state %u\n",
2966 __func__, ep, ep->hwtid, state_read(&ep->com));
2970 dst_confirm(ep->dst);
2972 pr_debug("%s last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
2973 __func__, ep, ep->hwtid,
2974 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2975 mutex_lock(&ep->com.mutex);
2976 kfree_skb(ep->mpa_skb);
2978 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2980 mutex_unlock(&ep->com.mutex);
2983 c4iw_put_ep(&ep->com);
2987 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2990 struct c4iw_ep *ep = to_ep(cm_id);
2992 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2994 mutex_lock(&ep->com.mutex);
2995 if (ep->com.state != MPA_REQ_RCVD) {
2996 mutex_unlock(&ep->com.mutex);
2997 c4iw_put_ep(&ep->com);
3000 set_bit(ULP_REJECT, &ep->com.history);
3004 abort = send_mpa_reject(ep, pdata, pdata_len);
3005 mutex_unlock(&ep->com.mutex);
3008 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3009 c4iw_put_ep(&ep->com);
3013 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3016 struct c4iw_qp_attributes attrs;
3017 enum c4iw_qp_attr_mask mask;
3018 struct c4iw_ep *ep = to_ep(cm_id);
3019 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3020 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3023 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3025 mutex_lock(&ep->com.mutex);
3026 if (ep->com.state != MPA_REQ_RCVD) {
3033 set_bit(ULP_ACCEPT, &ep->com.history);
3034 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3035 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3040 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3041 if (conn_param->ord > ep->ird) {
3042 if (RELAXED_IRD_NEGOTIATION) {
3043 conn_param->ord = ep->ird;
3045 ep->ird = conn_param->ird;
3046 ep->ord = conn_param->ord;
3047 send_mpa_reject(ep, conn_param->private_data,
3048 conn_param->private_data_len);
3053 if (conn_param->ird < ep->ord) {
3054 if (RELAXED_IRD_NEGOTIATION &&
3055 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3056 conn_param->ird = ep->ord;
3063 ep->ird = conn_param->ird;
3064 ep->ord = conn_param->ord;
3066 if (ep->mpa_attr.version == 1) {
3067 if (peer2peer && ep->ird == 0)
3071 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3072 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3076 pr_debug("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3078 ep->com.cm_id = cm_id;
3079 ref_cm_id(&ep->com);
3083 /* bind QP to EP and move to RTS */
3084 attrs.mpa_attr = ep->mpa_attr;
3085 attrs.max_ird = ep->ird;
3086 attrs.max_ord = ep->ord;
3087 attrs.llp_stream_handle = ep;
3088 attrs.next_state = C4IW_QP_STATE_RTS;
3090 /* bind QP and TID with INIT_WR */
3091 mask = C4IW_QP_ATTR_NEXT_STATE |
3092 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3093 C4IW_QP_ATTR_MPA_ATTR |
3094 C4IW_QP_ATTR_MAX_IRD |
3095 C4IW_QP_ATTR_MAX_ORD;
3097 err = c4iw_modify_qp(ep->com.qp->rhp,
3098 ep->com.qp, mask, &attrs, 1);
3100 goto err_deref_cm_id;
3102 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3103 err = send_mpa_reply(ep, conn_param->private_data,
3104 conn_param->private_data_len);
3106 goto err_deref_cm_id;
3108 __state_set(&ep->com, FPDU_MODE);
3109 established_upcall(ep);
3110 mutex_unlock(&ep->com.mutex);
3111 c4iw_put_ep(&ep->com);
3114 deref_cm_id(&ep->com);
3118 mutex_unlock(&ep->com.mutex);
3120 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3121 c4iw_put_ep(&ep->com);
3125 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3127 struct in_device *ind;
3129 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3130 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3132 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3134 return -EADDRNOTAVAIL;
3135 for_primary_ifa(ind) {
3136 laddr->sin_addr.s_addr = ifa->ifa_address;
3137 raddr->sin_addr.s_addr = ifa->ifa_address;
3143 return found ? 0 : -EADDRNOTAVAIL;
3146 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3147 unsigned char banned_flags)
3149 struct inet6_dev *idev;
3150 int err = -EADDRNOTAVAIL;
3153 idev = __in6_dev_get(dev);
3155 struct inet6_ifaddr *ifp;
3157 read_lock_bh(&idev->lock);
3158 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3159 if (ifp->scope == IFA_LINK &&
3160 !(ifp->flags & banned_flags)) {
3161 memcpy(addr, &ifp->addr, 16);
3166 read_unlock_bh(&idev->lock);
3172 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3174 struct in6_addr addr;
3175 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3176 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3178 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3179 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3180 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3183 return -EADDRNOTAVAIL;
3186 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3188 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3191 struct sockaddr_in *laddr;
3192 struct sockaddr_in *raddr;
3193 struct sockaddr_in6 *laddr6;
3194 struct sockaddr_in6 *raddr6;
3198 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3199 (conn_param->ird > cur_max_read_depth(dev))) {
3203 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3205 pr_err("%s - cannot alloc ep\n", __func__);
3210 skb_queue_head_init(&ep->com.ep_skb_list);
3211 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3216 init_timer(&ep->timer);
3217 ep->plen = conn_param->private_data_len;
3219 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3220 conn_param->private_data, ep->plen);
3221 ep->ird = conn_param->ird;
3222 ep->ord = conn_param->ord;
3224 if (peer2peer && ep->ord == 0)
3227 ep->com.cm_id = cm_id;
3228 ref_cm_id(&ep->com);
3230 ep->com.qp = get_qhp(dev, conn_param->qpn);
3232 pr_debug("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3237 pr_debug("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3241 * Allocate an active TID to initiate a TCP connection.
3243 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3244 if (ep->atid == -1) {
3245 pr_err("%s - cannot alloc atid\n", __func__);
3249 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3251 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3252 sizeof(ep->com.local_addr));
3253 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3254 sizeof(ep->com.remote_addr));
3256 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3257 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3258 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3259 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3261 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3263 ra = (__u8 *)&raddr->sin_addr;
3266 * Handle loopback requests to INADDR_ANY.
3268 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3269 err = pick_local_ipaddrs(dev, cm_id);
3275 pr_debug("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3276 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3277 ra, ntohs(raddr->sin_port));
3278 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3279 laddr->sin_addr.s_addr,
3280 raddr->sin_addr.s_addr,
3282 raddr->sin_port, cm_id->tos);
3285 ra = (__u8 *)&raddr6->sin6_addr;
3288 * Handle loopback requests to INADDR_ANY.
3290 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3291 err = pick_local_ip6addrs(dev, cm_id);
3297 pr_debug("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3298 __func__, laddr6->sin6_addr.s6_addr,
3299 ntohs(laddr6->sin6_port),
3300 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3301 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3302 laddr6->sin6_addr.s6_addr,
3303 raddr6->sin6_addr.s6_addr,
3305 raddr6->sin6_port, cm_id->tos,
3306 raddr6->sin6_scope_id);
3309 pr_err("%s - cannot find route\n", __func__);
3310 err = -EHOSTUNREACH;
3314 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3315 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3317 pr_err("%s - cannot alloc l2e\n", __func__);
3321 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3322 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3325 state_set(&ep->com, CONNECTING);
3326 ep->tos = cm_id->tos;
3328 /* send connect request to rnic */
3329 err = send_connect(ep);
3333 cxgb4_l2t_release(ep->l2t);
3335 dst_release(ep->dst);
3337 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3338 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3340 skb_queue_purge(&ep->com.ep_skb_list);
3341 deref_cm_id(&ep->com);
3343 c4iw_put_ep(&ep->com);
3348 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3351 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3352 &ep->com.local_addr;
3354 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3355 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3356 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3360 c4iw_init_wr_wait(&ep->com.wr_wait);
3361 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3362 ep->stid, &sin6->sin6_addr,
3364 ep->com.dev->rdev.lldi.rxq_ids[0]);
3366 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3370 err = net_xmit_errno(err);
3372 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3373 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3374 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3376 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3381 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3384 struct sockaddr_in *sin = (struct sockaddr_in *)
3385 &ep->com.local_addr;
3387 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3389 err = cxgb4_create_server_filter(
3390 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3391 sin->sin_addr.s_addr, sin->sin_port, 0,
3392 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3393 if (err == -EBUSY) {
3394 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3398 set_current_state(TASK_UNINTERRUPTIBLE);
3399 schedule_timeout(usecs_to_jiffies(100));
3401 } while (err == -EBUSY);
3403 c4iw_init_wr_wait(&ep->com.wr_wait);
3404 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3405 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3406 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3408 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3412 err = net_xmit_errno(err);
3415 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3417 &sin->sin_addr, ntohs(sin->sin_port));
3421 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3424 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3425 struct c4iw_listen_ep *ep;
3429 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3431 pr_err("%s - cannot alloc ep\n", __func__);
3435 skb_queue_head_init(&ep->com.ep_skb_list);
3436 pr_debug("%s ep %p\n", __func__, ep);
3437 ep->com.cm_id = cm_id;
3438 ref_cm_id(&ep->com);
3440 ep->backlog = backlog;
3441 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3442 sizeof(ep->com.local_addr));
3445 * Allocate a server TID.
3447 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3448 ep->com.local_addr.ss_family == AF_INET)
3449 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3450 cm_id->m_local_addr.ss_family, ep);
3452 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3453 cm_id->m_local_addr.ss_family, ep);
3455 if (ep->stid == -1) {
3456 pr_err("%s - cannot alloc stid\n", __func__);
3460 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3462 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3463 sizeof(ep->com.local_addr));
3465 state_set(&ep->com, LISTEN);
3466 if (ep->com.local_addr.ss_family == AF_INET)
3467 err = create_server4(dev, ep);
3469 err = create_server6(dev, ep);
3471 cm_id->provider_data = ep;
3474 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3475 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3476 ep->com.local_addr.ss_family);
3478 deref_cm_id(&ep->com);
3479 c4iw_put_ep(&ep->com);
3485 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3488 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3490 pr_debug("%s ep %p\n", __func__, ep);
3493 state_set(&ep->com, DEAD);
3494 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3495 ep->com.local_addr.ss_family == AF_INET) {
3496 err = cxgb4_remove_server_filter(
3497 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3498 ep->com.dev->rdev.lldi.rxq_ids[0], false);
3500 struct sockaddr_in6 *sin6;
3501 c4iw_init_wr_wait(&ep->com.wr_wait);
3502 err = cxgb4_remove_server(
3503 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3504 ep->com.dev->rdev.lldi.rxq_ids[0],
3505 ep->com.local_addr.ss_family == AF_INET6);
3508 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3510 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3511 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3512 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3514 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3515 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3516 ep->com.local_addr.ss_family);
3518 deref_cm_id(&ep->com);
3519 c4iw_put_ep(&ep->com);
3523 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3528 struct c4iw_rdev *rdev;
3530 mutex_lock(&ep->com.mutex);
3532 pr_debug("%s ep %p state %s, abrupt %d\n", __func__, ep,
3533 states[ep->com.state], abrupt);
3536 * Ref the ep here in case we have fatal errors causing the
3537 * ep to be released and freed.
3539 c4iw_get_ep(&ep->com);
3541 rdev = &ep->com.dev->rdev;
3542 if (c4iw_fatal_error(rdev)) {
3544 close_complete_upcall(ep, -EIO);
3545 ep->com.state = DEAD;
3547 switch (ep->com.state) {
3556 ep->com.state = ABORTING;
3558 ep->com.state = CLOSING;
3561 * if we close before we see the fw4_ack() then we fix
3562 * up the timer state since we're reusing it.
3565 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3566 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3571 set_bit(CLOSE_SENT, &ep->com.flags);
3574 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3577 (void)stop_ep_timer(ep);
3578 ep->com.state = ABORTING;
3580 ep->com.state = MORIBUND;
3586 pr_debug("%s ignoring disconnect ep %p state %u\n",
3587 __func__, ep, ep->com.state);
3596 set_bit(EP_DISC_ABORT, &ep->com.history);
3597 ret = send_abort(ep);
3599 set_bit(EP_DISC_CLOSE, &ep->com.history);
3600 ret = send_halfclose(ep);
3603 set_bit(EP_DISC_FAIL, &ep->com.history);
3606 close_complete_upcall(ep, -EIO);
3609 struct c4iw_qp_attributes attrs;
3611 attrs.next_state = C4IW_QP_STATE_ERROR;
3612 ret = c4iw_modify_qp(ep->com.qp->rhp,
3614 C4IW_QP_ATTR_NEXT_STATE,
3617 pr_err("%s - qp <- error failed!\n",
3623 mutex_unlock(&ep->com.mutex);
3624 c4iw_put_ep(&ep->com);
3626 release_ep_resources(ep);
3630 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3631 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3634 int atid = be32_to_cpu(req->tid);
3636 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3637 (__force u32) req->tid);
3641 switch (req->retval) {
3643 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3644 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3645 send_fw_act_open_req(ep, atid);
3649 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3650 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3651 send_fw_act_open_req(ep, atid);
3656 pr_info("%s unexpected ofld conn wr retval %d\n",
3657 __func__, req->retval);
3660 pr_err("active ofld_connect_wr failure %d atid %d\n",
3662 mutex_lock(&dev->rdev.stats.lock);
3663 dev->rdev.stats.act_ofld_conn_fails++;
3664 mutex_unlock(&dev->rdev.stats.lock);
3665 connect_reply_upcall(ep, status2errno(req->retval));
3666 state_set(&ep->com, DEAD);
3667 if (ep->com.remote_addr.ss_family == AF_INET6) {
3668 struct sockaddr_in6 *sin6 =
3669 (struct sockaddr_in6 *)&ep->com.local_addr;
3670 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3671 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3673 remove_handle(dev, &dev->atid_idr, atid);
3674 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3675 dst_release(ep->dst);
3676 cxgb4_l2t_release(ep->l2t);
3677 c4iw_put_ep(&ep->com);
3680 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3681 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3683 struct sk_buff *rpl_skb;
3684 struct cpl_pass_accept_req *cpl;
3687 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3690 pr_debug("%s passive open failure %d\n", __func__, req->retval);
3691 mutex_lock(&dev->rdev.stats.lock);
3692 dev->rdev.stats.pas_ofld_conn_fails++;
3693 mutex_unlock(&dev->rdev.stats.lock);
3696 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3697 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3698 (__force u32) htonl(
3699 (__force u32) req->tid)));
3700 ret = pass_accept_req(dev, rpl_skb);
3707 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3709 struct cpl_fw6_msg *rpl = cplhdr(skb);
3710 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3712 switch (rpl->type) {
3714 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3716 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3717 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3718 switch (req->t_state) {
3720 active_ofld_conn_reply(dev, skb, req);
3723 passive_ofld_conn_reply(dev, skb, req);
3726 pr_err("%s unexpected ofld conn wr state %d\n",
3727 __func__, req->t_state);
3735 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3738 __be16 hdr_len, vlantag, len;
3740 int tcp_hdr_len, ip_hdr_len;
3742 struct cpl_rx_pkt *cpl = cplhdr(skb);
3743 struct cpl_pass_accept_req *req;
3744 struct tcp_options_received tmp_opt;
3745 struct c4iw_dev *dev;
3746 enum chip_type type;
3748 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3749 /* Store values from cpl_rx_pkt in temporary location. */
3750 vlantag = cpl->vlan;
3752 l2info = cpl->l2info;
3753 hdr_len = cpl->hdr_len;
3756 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3759 * We need to parse the TCP options from SYN packet.
3760 * to generate cpl_pass_accept_req.
3762 memset(&tmp_opt, 0, sizeof(tmp_opt));
3763 tcp_clear_options(&tmp_opt);
3764 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3766 req = __skb_push(skb, sizeof(*req));
3767 memset(req, 0, sizeof(*req));
3768 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3769 SYN_MAC_IDX_V(RX_MACIDX_G(
3770 be32_to_cpu(l2info))) |
3772 type = dev->rdev.lldi.adapter_type;
3773 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3774 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3776 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3777 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3778 eth_hdr_len = is_t4(type) ?
3779 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3780 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3781 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3782 IP_HDR_LEN_V(ip_hdr_len) |
3783 ETH_HDR_LEN_V(eth_hdr_len));
3784 } else { /* T6 and later */
3785 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3786 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3787 T6_IP_HDR_LEN_V(ip_hdr_len) |
3788 T6_ETH_HDR_LEN_V(eth_hdr_len));
3790 req->vlan = vlantag;
3792 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3793 PASS_OPEN_TOS_V(tos));
3794 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3795 if (tmp_opt.wscale_ok)
3796 req->tcpopt.wsf = tmp_opt.snd_wscale;
3797 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3798 if (tmp_opt.sack_ok)
3799 req->tcpopt.sack = 1;
3800 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3804 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3805 __be32 laddr, __be16 lport,
3806 __be32 raddr, __be16 rport,
3807 u32 rcv_isn, u32 filter, u16 window,
3808 u32 rss_qid, u8 port_id)
3810 struct sk_buff *req_skb;
3811 struct fw_ofld_connection_wr *req;
3812 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3815 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3818 req = __skb_put_zero(req_skb, sizeof(*req));
3819 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3820 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3821 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3822 req->le.filter = (__force __be32) filter;
3823 req->le.lport = lport;
3824 req->le.pport = rport;
3825 req->le.u.ipv4.lip = laddr;
3826 req->le.u.ipv4.pip = raddr;
3827 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3828 req->tcb.rcv_adv = htons(window);
3829 req->tcb.t_state_to_astid =
3830 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3831 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3832 FW_OFLD_CONNECTION_WR_ASTID_V(
3833 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3836 * We store the qid in opt2 which will be used by the firmware
3837 * to send us the wr response.
3839 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3842 * We initialize the MSS index in TCB to 0xF.
3843 * So that when driver sends cpl_pass_accept_rpl
3844 * TCB picks up the correct value. If this was 0
3845 * TP will ignore any value > 0 for MSS index.
3847 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3848 req->cookie = (uintptr_t)skb;
3850 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3851 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3853 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3861 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3862 * messages when a filter is being used instead of server to
3863 * redirect a syn packet. When packets hit filter they are redirected
3864 * to the offload queue and driver tries to establish the connection
3865 * using firmware work request.
3867 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3870 unsigned int filter;
3871 struct ethhdr *eh = NULL;
3872 struct vlan_ethhdr *vlan_eh = NULL;
3874 struct tcphdr *tcph;
3875 struct rss_header *rss = (void *)skb->data;
3876 struct cpl_rx_pkt *cpl = (void *)skb->data;
3877 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3878 struct l2t_entry *e;
3879 struct dst_entry *dst;
3880 struct c4iw_ep *lep = NULL;
3882 struct port_info *pi;
3883 struct net_device *pdev;
3884 u16 rss_qid, eth_hdr_len;
3887 struct neighbour *neigh;
3889 /* Drop all non-SYN packets */
3890 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3894 * Drop all packets which did not hit the filter.
3895 * Unlikely to happen.
3897 if (!(rss->filter_hit && rss->filter_tid))
3901 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3903 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3905 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3907 pr_debug("%s connect request on invalid stid %d\n",
3912 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3914 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3917 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3920 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3923 pr_err("T%d Chip is not supported\n",
3924 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3928 if (eth_hdr_len == ETH_HLEN) {
3929 eh = (struct ethhdr *)(req + 1);
3930 iph = (struct iphdr *)(eh + 1);
3932 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3933 iph = (struct iphdr *)(vlan_eh + 1);
3934 skb->vlan_tci = ntohs(cpl->vlan);
3937 if (iph->version != 0x4)
3940 tcph = (struct tcphdr *)(iph + 1);
3941 skb_set_network_header(skb, (void *)iph - (void *)rss);
3942 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3945 pr_debug("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3946 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3947 ntohs(tcph->source), iph->tos);
3949 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3950 iph->daddr, iph->saddr, tcph->dest,
3951 tcph->source, iph->tos);
3953 pr_err("%s - failed to find dst entry!\n",
3957 neigh = dst_neigh_lookup_skb(dst, skb);
3960 pr_err("%s - failed to allocate neigh!\n",
3965 if (neigh->dev->flags & IFF_LOOPBACK) {
3966 pdev = ip_dev_find(&init_net, iph->daddr);
3967 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3969 pi = (struct port_info *)netdev_priv(pdev);
3970 tx_chan = cxgb4_port_chan(pdev);
3973 pdev = get_real_dev(neigh->dev);
3974 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3976 pi = (struct port_info *)netdev_priv(pdev);
3977 tx_chan = cxgb4_port_chan(pdev);
3979 neigh_release(neigh);
3981 pr_err("%s - failed to allocate l2t entry!\n",
3986 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3987 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3988 window = (__force u16) htons((__force u16)tcph->window);
3990 /* Calcuate filter portion for LE region. */
3991 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3992 dev->rdev.lldi.ports[0],
3996 * Synthesize the cpl_pass_accept_req. We have everything except the
3997 * TID. Once firmware sends a reply with TID we update the TID field
3998 * in cpl and pass it through the regular cpl_pass_accept_req path.
4000 build_cpl_pass_accept_req(skb, stid, iph->tos);
4001 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4002 tcph->source, ntohl(tcph->seq), filter, window,
4003 rss_qid, pi->port_id);
4004 cxgb4_l2t_release(e);
4009 c4iw_put_ep(&lep->com);
4014 * These are the real handlers that are called from a
4017 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4018 [CPL_ACT_ESTABLISH] = act_establish,
4019 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4020 [CPL_RX_DATA] = rx_data,
4021 [CPL_ABORT_RPL_RSS] = abort_rpl,
4022 [CPL_ABORT_RPL] = abort_rpl,
4023 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4024 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4025 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4026 [CPL_PASS_ESTABLISH] = pass_establish,
4027 [CPL_PEER_CLOSE] = peer_close,
4028 [CPL_ABORT_REQ_RSS] = peer_abort,
4029 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4030 [CPL_RDMA_TERMINATE] = terminate,
4031 [CPL_FW4_ACK] = fw4_ack,
4032 [CPL_FW6_MSG] = deferred_fw6_msg,
4033 [CPL_RX_PKT] = rx_pkt,
4034 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4035 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4038 static void process_timeout(struct c4iw_ep *ep)
4040 struct c4iw_qp_attributes attrs;
4043 mutex_lock(&ep->com.mutex);
4044 pr_debug("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4046 set_bit(TIMEDOUT, &ep->com.history);
4047 switch (ep->com.state) {
4049 connect_reply_upcall(ep, -ETIMEDOUT);
4058 if (ep->com.cm_id && ep->com.qp) {
4059 attrs.next_state = C4IW_QP_STATE_ERROR;
4060 c4iw_modify_qp(ep->com.qp->rhp,
4061 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4064 close_complete_upcall(ep, -ETIMEDOUT);
4070 * These states are expected if the ep timed out at the same
4071 * time as another thread was calling stop_ep_timer().
4072 * So we silently do nothing for these states.
4077 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4078 __func__, ep, ep->hwtid, ep->com.state);
4081 mutex_unlock(&ep->com.mutex);
4083 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4084 c4iw_put_ep(&ep->com);
4087 static void process_timedout_eps(void)
4091 spin_lock_irq(&timeout_lock);
4092 while (!list_empty(&timeout_list)) {
4093 struct list_head *tmp;
4095 tmp = timeout_list.next;
4099 spin_unlock_irq(&timeout_lock);
4100 ep = list_entry(tmp, struct c4iw_ep, entry);
4101 process_timeout(ep);
4102 spin_lock_irq(&timeout_lock);
4104 spin_unlock_irq(&timeout_lock);
4107 static void process_work(struct work_struct *work)
4109 struct sk_buff *skb = NULL;
4110 struct c4iw_dev *dev;
4111 struct cpl_act_establish *rpl;
4112 unsigned int opcode;
4115 process_timedout_eps();
4116 while ((skb = skb_dequeue(&rxq))) {
4118 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4119 opcode = rpl->ot.opcode;
4121 BUG_ON(!work_handlers[opcode]);
4122 ret = work_handlers[opcode](dev, skb);
4125 process_timedout_eps();
4129 static DECLARE_WORK(skb_work, process_work);
4131 static void ep_timeout(unsigned long arg)
4133 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4136 spin_lock(&timeout_lock);
4137 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4139 * Only insert if it is not already on the list.
4141 if (!ep->entry.next) {
4142 list_add_tail(&ep->entry, &timeout_list);
4146 spin_unlock(&timeout_lock);
4148 queue_work(workq, &skb_work);
4152 * All the CM events are handled on a work queue to have a safe context.
4154 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4158 * Save dev in the skb->cb area.
4160 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4163 * Queue the skb and schedule the worker thread.
4165 skb_queue_tail(&rxq, skb);
4166 queue_work(workq, &skb_work);
4170 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4172 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4174 if (rpl->status != CPL_ERR_NONE) {
4175 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4176 rpl->status, GET_TID(rpl));
4182 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4184 struct cpl_fw6_msg *rpl = cplhdr(skb);
4185 struct c4iw_wr_wait *wr_waitp;
4188 pr_debug("%s type %u\n", __func__, rpl->type);
4190 switch (rpl->type) {
4191 case FW6_TYPE_WR_RPL:
4192 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4193 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4194 pr_debug("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4196 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4200 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4204 pr_err("%s unexpected fw6 msg type %u\n",
4205 __func__, rpl->type);
4212 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4214 struct cpl_abort_req_rss *req = cplhdr(skb);
4216 unsigned int tid = GET_TID(req);
4218 ep = get_ep_from_tid(dev, tid);
4219 /* This EP will be dereferenced in peer_abort() */
4221 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4225 if (cxgb_is_neg_adv(req->status)) {
4226 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
4227 __func__, ep->hwtid, req->status,
4228 neg_adv_str(req->status));
4231 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4234 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4241 * Most upcalls from the T4 Core go to sched() to
4242 * schedule the processing on a work queue.
4244 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4245 [CPL_ACT_ESTABLISH] = sched,
4246 [CPL_ACT_OPEN_RPL] = sched,
4247 [CPL_RX_DATA] = sched,
4248 [CPL_ABORT_RPL_RSS] = sched,
4249 [CPL_ABORT_RPL] = sched,
4250 [CPL_PASS_OPEN_RPL] = sched,
4251 [CPL_CLOSE_LISTSRV_RPL] = sched,
4252 [CPL_PASS_ACCEPT_REQ] = sched,
4253 [CPL_PASS_ESTABLISH] = sched,
4254 [CPL_PEER_CLOSE] = sched,
4255 [CPL_CLOSE_CON_RPL] = sched,
4256 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4257 [CPL_RDMA_TERMINATE] = sched,
4258 [CPL_FW4_ACK] = sched,
4259 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4260 [CPL_FW6_MSG] = fw6_msg,
4261 [CPL_RX_PKT] = sched
4264 int __init c4iw_cm_init(void)
4266 spin_lock_init(&timeout_lock);
4267 skb_queue_head_init(&rxq);
4269 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4276 void c4iw_cm_term(void)
4278 WARN_ON(!list_empty(&timeout_list));
4279 flush_workqueue(workq);
4280 destroy_workqueue(workq);