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, "Enable debug logging (default=0)");
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 PDBG("%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 PDBG("%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 PDBG("%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 PDBG("%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 struct cpl_tid_release *req;
245 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
248 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
249 INIT_TP_WR(req, hwtid);
250 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
251 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
252 c4iw_ofld_send(rdev, skb);
256 static void set_emss(struct c4iw_ep *ep, u16 opt)
258 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
259 ((AF_INET == ep->com.remote_addr.ss_family) ?
260 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
261 sizeof(struct tcphdr);
263 if (TCPOPT_TSTAMP_G(opt))
264 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
268 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
269 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
270 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
274 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
276 enum c4iw_ep_state state;
278 mutex_lock(&epc->mutex);
280 mutex_unlock(&epc->mutex);
284 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
289 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
291 mutex_lock(&epc->mutex);
292 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
293 __state_set(epc, new);
294 mutex_unlock(&epc->mutex);
298 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
304 len = roundup(sizeof(union cpl_wr_size), 16);
305 for (i = 0; i < size; i++) {
306 skb = alloc_skb(len, GFP_KERNEL);
309 skb_queue_tail(ep_skb_list, skb);
313 skb_queue_purge(ep_skb_list);
317 static void *alloc_ep(int size, gfp_t gfp)
319 struct c4iw_ep_common *epc;
321 epc = kzalloc(size, gfp);
323 kref_init(&epc->kref);
324 mutex_init(&epc->mutex);
325 c4iw_init_wr_wait(&epc->wr_wait);
327 PDBG("%s alloc ep %p\n", __func__, epc);
331 static void remove_ep_tid(struct c4iw_ep *ep)
335 spin_lock_irqsave(&ep->com.dev->lock, flags);
336 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
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 PDBG("%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 dst_release(ep->dst);
403 cxgb4_l2t_release(ep->l2t);
405 kfree_skb(ep->mpa_skb);
407 if (!skb_queue_empty(&ep->com.ep_skb_list))
408 skb_queue_purge(&ep->com.ep_skb_list);
412 static void release_ep_resources(struct c4iw_ep *ep)
414 set_bit(RELEASE_RESOURCES, &ep->com.flags);
417 * If we have a hwtid, then remove it from the idr table
418 * so lookups will no longer find this endpoint. Otherwise
419 * we have a race where one thread finds the ep ptr just
420 * before the other thread is freeing the ep memory.
424 c4iw_put_ep(&ep->com);
427 static int status2errno(int status)
432 case CPL_ERR_CONN_RESET:
434 case CPL_ERR_ARP_MISS:
435 return -EHOSTUNREACH;
436 case CPL_ERR_CONN_TIMEDOUT:
438 case CPL_ERR_TCAM_FULL:
440 case CPL_ERR_CONN_EXIST:
448 * Try and reuse skbs already allocated...
450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
452 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
455 skb_reset_transport_header(skb);
457 skb = alloc_skb(len, gfp);
459 t4_set_arp_err_handler(skb, NULL, NULL);
463 static struct net_device *get_real_dev(struct net_device *egress_dev)
465 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
468 static void arp_failure_discard(void *handle, struct sk_buff *skb)
470 pr_err(MOD "ARP failure\n");
474 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
476 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
481 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
482 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
485 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
489 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
490 release_ep_resources(ep);
494 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
498 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
499 c4iw_put_ep(&ep->parent_ep->com);
500 release_ep_resources(ep);
505 * Fake up a special CPL opcode and call sched() so process_work() will call
506 * _put_ep_safe() in a safe context to free the ep resources. This is needed
507 * because ARP error handlers are called in an ATOMIC context, and
508 * _c4iw_free_ep() needs to block.
510 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
513 struct cpl_act_establish *rpl = cplhdr(skb);
515 /* Set our special ARP_FAILURE opcode */
516 rpl->ot.opcode = cpl;
519 * Save ep in the skb->cb area, after where sched() will save the dev
522 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
523 sched(ep->com.dev, skb);
526 /* Handle an ARP failure for an accept */
527 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
529 struct c4iw_ep *ep = handle;
531 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
534 __state_set(&ep->com, DEAD);
535 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
539 * Handle an ARP failure for an active open.
541 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
543 struct c4iw_ep *ep = handle;
545 printk(KERN_ERR MOD "ARP failure during connect\n");
546 connect_reply_upcall(ep, -EHOSTUNREACH);
547 __state_set(&ep->com, DEAD);
548 if (ep->com.remote_addr.ss_family == AF_INET6) {
549 struct sockaddr_in6 *sin6 =
550 (struct sockaddr_in6 *)&ep->com.local_addr;
551 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
552 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
554 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
555 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
556 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
560 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
563 static void abort_arp_failure(void *handle, struct sk_buff *skb)
566 struct c4iw_ep *ep = handle;
567 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
568 struct cpl_abort_req *req = cplhdr(skb);
570 PDBG("%s rdev %p\n", __func__, rdev);
571 req->cmd = CPL_ABORT_NO_RST;
572 ret = c4iw_ofld_send(rdev, skb);
574 __state_set(&ep->com, DEAD);
575 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
579 static int send_flowc(struct c4iw_ep *ep)
581 struct fw_flowc_wr *flowc;
582 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
584 u16 vlan = ep->l2t->vlan;
590 if (vlan == CPL_L2T_VLAN_NONE)
595 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
597 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
598 FW_FLOWC_WR_NPARAMS_V(nparams));
599 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
600 16)) | FW_WR_FLOWID_V(ep->hwtid));
602 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
603 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
604 (ep->com.dev->rdev.lldi.pf));
605 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
606 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
607 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
608 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
609 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
610 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
611 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
612 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
613 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
614 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
615 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
616 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
617 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
618 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
622 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
623 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
624 flowc->mnemval[8].val = cpu_to_be32(pri);
626 /* Pad WR to 16 byte boundary */
627 flowc->mnemval[8].mnemonic = 0;
628 flowc->mnemval[8].val = 0;
630 for (i = 0; i < 9; i++) {
631 flowc->mnemval[i].r4[0] = 0;
632 flowc->mnemval[i].r4[1] = 0;
633 flowc->mnemval[i].r4[2] = 0;
636 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
637 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
640 static int send_halfclose(struct c4iw_ep *ep)
642 struct cpl_close_con_req *req;
643 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
644 int wrlen = roundup(sizeof *req, 16);
646 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
650 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
651 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
652 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
653 memset(req, 0, wrlen);
654 INIT_TP_WR(req, ep->hwtid);
655 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
657 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
660 static int send_abort(struct c4iw_ep *ep)
662 struct cpl_abort_req *req;
663 int wrlen = roundup(sizeof *req, 16);
664 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
666 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
667 if (WARN_ON(!req_skb))
670 set_wr_txq(req_skb, CPL_PRIORITY_DATA, ep->txq_idx);
671 t4_set_arp_err_handler(req_skb, ep, abort_arp_failure);
672 req = (struct cpl_abort_req *)skb_put(req_skb, wrlen);
673 memset(req, 0, wrlen);
674 INIT_TP_WR(req, ep->hwtid);
675 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
676 req->cmd = CPL_ABORT_SEND_RST;
677 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
680 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
681 unsigned int *idx, int use_ts, int ipv6)
683 unsigned short hdr_size = (ipv6 ?
684 sizeof(struct ipv6hdr) :
685 sizeof(struct iphdr)) +
686 sizeof(struct tcphdr) +
688 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
689 unsigned short data_size = mtu - hdr_size;
691 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
694 static int send_connect(struct c4iw_ep *ep)
696 struct cpl_act_open_req *req = NULL;
697 struct cpl_t5_act_open_req *t5req = NULL;
698 struct cpl_t6_act_open_req *t6req = NULL;
699 struct cpl_act_open_req6 *req6 = NULL;
700 struct cpl_t5_act_open_req6 *t5req6 = NULL;
701 struct cpl_t6_act_open_req6 *t6req6 = NULL;
705 unsigned int mtu_idx;
707 int win, sizev4, sizev6, wrlen;
708 struct sockaddr_in *la = (struct sockaddr_in *)
710 struct sockaddr_in *ra = (struct sockaddr_in *)
711 &ep->com.remote_addr;
712 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
714 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
715 &ep->com.remote_addr;
717 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
718 u32 isn = (prandom_u32() & ~7UL) - 1;
720 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
722 sizev4 = sizeof(struct cpl_act_open_req);
723 sizev6 = sizeof(struct cpl_act_open_req6);
726 sizev4 = sizeof(struct cpl_t5_act_open_req);
727 sizev6 = sizeof(struct cpl_t5_act_open_req6);
730 sizev4 = sizeof(struct cpl_t6_act_open_req);
731 sizev6 = sizeof(struct cpl_t6_act_open_req6);
734 pr_err("T%d Chip is not supported\n",
735 CHELSIO_CHIP_VERSION(adapter_type));
739 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
740 roundup(sizev4, 16) :
743 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
745 skb = get_skb(NULL, wrlen, GFP_KERNEL);
747 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
751 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
753 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
754 enable_tcp_timestamps,
755 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
756 wscale = compute_wscale(rcv_win);
759 * Specify the largest window that will fit in opt0. The
760 * remainder will be specified in the rx_data_ack.
762 win = ep->rcv_win >> 10;
763 if (win > RCV_BUFSIZ_M)
766 opt0 = (nocong ? NO_CONG_F : 0) |
769 WND_SCALE_V(wscale) |
771 L2T_IDX_V(ep->l2t->idx) |
772 TX_CHAN_V(ep->tx_chan) |
773 SMAC_SEL_V(ep->smac_idx) |
774 DSCP_V(ep->tos >> 2) |
775 ULP_MODE_V(ULP_MODE_TCPDDP) |
777 opt2 = RX_CHANNEL_V(0) |
778 CCTRL_ECN_V(enable_ecn) |
779 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
780 if (enable_tcp_timestamps)
781 opt2 |= TSTAMPS_EN_F;
784 if (wscale && enable_tcp_window_scaling)
785 opt2 |= WND_SCALE_EN_F;
786 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
790 opt2 |= T5_OPT_2_VALID_F;
791 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
795 if (ep->com.remote_addr.ss_family == AF_INET6)
796 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
797 (const u32 *)&la6->sin6_addr.s6_addr, 1);
799 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
801 if (ep->com.remote_addr.ss_family == AF_INET) {
802 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
804 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
808 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
810 INIT_TP_WR(t5req, 0);
811 req = (struct cpl_act_open_req *)t5req;
814 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
816 INIT_TP_WR(t6req, 0);
817 req = (struct cpl_act_open_req *)t6req;
818 t5req = (struct cpl_t5_act_open_req *)t6req;
821 pr_err("T%d Chip is not supported\n",
822 CHELSIO_CHIP_VERSION(adapter_type));
827 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
828 ((ep->rss_qid<<14) | ep->atid)));
829 req->local_port = la->sin_port;
830 req->peer_port = ra->sin_port;
831 req->local_ip = la->sin_addr.s_addr;
832 req->peer_ip = ra->sin_addr.s_addr;
833 req->opt0 = cpu_to_be64(opt0);
835 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
836 req->params = cpu_to_be32(cxgb4_select_ntuple(
837 ep->com.dev->rdev.lldi.ports[0],
839 req->opt2 = cpu_to_be32(opt2);
841 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
843 ep->com.dev->rdev.lldi.ports[0],
845 t5req->rsvd = cpu_to_be32(isn);
846 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
847 t5req->opt2 = cpu_to_be32(opt2);
850 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
852 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
856 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
858 INIT_TP_WR(t5req6, 0);
859 req6 = (struct cpl_act_open_req6 *)t5req6;
862 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
864 INIT_TP_WR(t6req6, 0);
865 req6 = (struct cpl_act_open_req6 *)t6req6;
866 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
869 pr_err("T%d Chip is not supported\n",
870 CHELSIO_CHIP_VERSION(adapter_type));
875 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
876 ((ep->rss_qid<<14)|ep->atid)));
877 req6->local_port = la6->sin6_port;
878 req6->peer_port = ra6->sin6_port;
879 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
880 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
881 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
882 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
883 req6->opt0 = cpu_to_be64(opt0);
885 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
886 req6->params = cpu_to_be32(cxgb4_select_ntuple(
887 ep->com.dev->rdev.lldi.ports[0],
889 req6->opt2 = cpu_to_be32(opt2);
891 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
893 ep->com.dev->rdev.lldi.ports[0],
895 t5req6->rsvd = cpu_to_be32(isn);
896 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
897 t5req6->opt2 = cpu_to_be32(opt2);
901 set_bit(ACT_OPEN_REQ, &ep->com.history);
902 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
904 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
905 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
906 (const u32 *)&la6->sin6_addr.s6_addr, 1);
910 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
913 int mpalen, wrlen, ret;
914 struct fw_ofld_tx_data_wr *req;
915 struct mpa_message *mpa;
916 struct mpa_v2_conn_params mpa_v2_params;
918 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
920 BUG_ON(skb_cloned(skb));
922 mpalen = sizeof(*mpa) + ep->plen;
923 if (mpa_rev_to_use == 2)
924 mpalen += sizeof(struct mpa_v2_conn_params);
925 wrlen = roundup(mpalen + sizeof *req, 16);
926 skb = get_skb(skb, wrlen, GFP_KERNEL);
928 connect_reply_upcall(ep, -ENOMEM);
931 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
933 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
934 memset(req, 0, wrlen);
935 req->op_to_immdlen = cpu_to_be32(
936 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
938 FW_WR_IMMDLEN_V(mpalen));
939 req->flowid_len16 = cpu_to_be32(
940 FW_WR_FLOWID_V(ep->hwtid) |
941 FW_WR_LEN16_V(wrlen >> 4));
942 req->plen = cpu_to_be32(mpalen);
943 req->tunnel_to_proxy = cpu_to_be32(
944 FW_OFLD_TX_DATA_WR_FLUSH_F |
945 FW_OFLD_TX_DATA_WR_SHOVE_F);
947 mpa = (struct mpa_message *)(req + 1);
948 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
952 mpa->flags |= MPA_CRC;
953 if (markers_enabled) {
954 mpa->flags |= MPA_MARKERS;
955 ep->mpa_attr.recv_marker_enabled = 1;
957 ep->mpa_attr.recv_marker_enabled = 0;
959 if (mpa_rev_to_use == 2)
960 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
962 mpa->private_data_size = htons(ep->plen);
963 mpa->revision = mpa_rev_to_use;
964 if (mpa_rev_to_use == 1) {
965 ep->tried_with_mpa_v1 = 1;
966 ep->retry_with_mpa_v1 = 0;
969 if (mpa_rev_to_use == 2) {
970 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
971 sizeof (struct mpa_v2_conn_params));
972 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
974 mpa_v2_params.ird = htons((u16)ep->ird);
975 mpa_v2_params.ord = htons((u16)ep->ord);
978 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
979 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
981 htons(MPA_V2_RDMA_WRITE_RTR);
982 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
984 htons(MPA_V2_RDMA_READ_RTR);
986 memcpy(mpa->private_data, &mpa_v2_params,
987 sizeof(struct mpa_v2_conn_params));
990 memcpy(mpa->private_data +
991 sizeof(struct mpa_v2_conn_params),
992 ep->mpa_pkt + sizeof(*mpa), ep->plen);
995 memcpy(mpa->private_data,
996 ep->mpa_pkt + sizeof(*mpa), ep->plen);
999 * Reference the mpa skb. This ensures the data area
1000 * will remain in memory until the hw acks the tx.
1001 * Function fw4_ack() will deref it.
1004 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1005 BUG_ON(ep->mpa_skb);
1007 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1011 __state_set(&ep->com, MPA_REQ_SENT);
1012 ep->mpa_attr.initiator = 1;
1013 ep->snd_seq += mpalen;
1017 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1020 struct fw_ofld_tx_data_wr *req;
1021 struct mpa_message *mpa;
1022 struct sk_buff *skb;
1023 struct mpa_v2_conn_params mpa_v2_params;
1025 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1027 mpalen = sizeof(*mpa) + plen;
1028 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1029 mpalen += sizeof(struct mpa_v2_conn_params);
1030 wrlen = roundup(mpalen + sizeof *req, 16);
1032 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1034 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1037 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1039 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1040 memset(req, 0, wrlen);
1041 req->op_to_immdlen = cpu_to_be32(
1042 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1044 FW_WR_IMMDLEN_V(mpalen));
1045 req->flowid_len16 = cpu_to_be32(
1046 FW_WR_FLOWID_V(ep->hwtid) |
1047 FW_WR_LEN16_V(wrlen >> 4));
1048 req->plen = cpu_to_be32(mpalen);
1049 req->tunnel_to_proxy = cpu_to_be32(
1050 FW_OFLD_TX_DATA_WR_FLUSH_F |
1051 FW_OFLD_TX_DATA_WR_SHOVE_F);
1053 mpa = (struct mpa_message *)(req + 1);
1054 memset(mpa, 0, sizeof(*mpa));
1055 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1056 mpa->flags = MPA_REJECT;
1057 mpa->revision = ep->mpa_attr.version;
1058 mpa->private_data_size = htons(plen);
1060 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1061 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1062 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1063 sizeof (struct mpa_v2_conn_params));
1064 mpa_v2_params.ird = htons(((u16)ep->ird) |
1065 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1067 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1069 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1070 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1071 FW_RI_INIT_P2PTYPE_READ_REQ ?
1072 MPA_V2_RDMA_READ_RTR : 0) : 0));
1073 memcpy(mpa->private_data, &mpa_v2_params,
1074 sizeof(struct mpa_v2_conn_params));
1077 memcpy(mpa->private_data +
1078 sizeof(struct mpa_v2_conn_params), pdata, plen);
1081 memcpy(mpa->private_data, pdata, plen);
1084 * Reference the mpa skb again. This ensures the data area
1085 * will remain in memory until the hw acks the tx.
1086 * Function fw4_ack() will deref it.
1089 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1090 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1091 BUG_ON(ep->mpa_skb);
1093 ep->snd_seq += mpalen;
1094 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1097 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1100 struct fw_ofld_tx_data_wr *req;
1101 struct mpa_message *mpa;
1102 struct sk_buff *skb;
1103 struct mpa_v2_conn_params mpa_v2_params;
1105 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1107 mpalen = sizeof(*mpa) + plen;
1108 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1109 mpalen += sizeof(struct mpa_v2_conn_params);
1110 wrlen = roundup(mpalen + sizeof *req, 16);
1112 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1114 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1117 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1120 memset(req, 0, wrlen);
1121 req->op_to_immdlen = cpu_to_be32(
1122 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1124 FW_WR_IMMDLEN_V(mpalen));
1125 req->flowid_len16 = cpu_to_be32(
1126 FW_WR_FLOWID_V(ep->hwtid) |
1127 FW_WR_LEN16_V(wrlen >> 4));
1128 req->plen = cpu_to_be32(mpalen);
1129 req->tunnel_to_proxy = cpu_to_be32(
1130 FW_OFLD_TX_DATA_WR_FLUSH_F |
1131 FW_OFLD_TX_DATA_WR_SHOVE_F);
1133 mpa = (struct mpa_message *)(req + 1);
1134 memset(mpa, 0, sizeof(*mpa));
1135 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1137 if (ep->mpa_attr.crc_enabled)
1138 mpa->flags |= MPA_CRC;
1139 if (ep->mpa_attr.recv_marker_enabled)
1140 mpa->flags |= MPA_MARKERS;
1141 mpa->revision = ep->mpa_attr.version;
1142 mpa->private_data_size = htons(plen);
1144 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1145 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1146 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1147 sizeof (struct mpa_v2_conn_params));
1148 mpa_v2_params.ird = htons((u16)ep->ird);
1149 mpa_v2_params.ord = htons((u16)ep->ord);
1150 if (peer2peer && (ep->mpa_attr.p2p_type !=
1151 FW_RI_INIT_P2PTYPE_DISABLED)) {
1152 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1154 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1155 mpa_v2_params.ord |=
1156 htons(MPA_V2_RDMA_WRITE_RTR);
1157 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1158 mpa_v2_params.ord |=
1159 htons(MPA_V2_RDMA_READ_RTR);
1162 memcpy(mpa->private_data, &mpa_v2_params,
1163 sizeof(struct mpa_v2_conn_params));
1166 memcpy(mpa->private_data +
1167 sizeof(struct mpa_v2_conn_params), pdata, plen);
1170 memcpy(mpa->private_data, pdata, plen);
1173 * Reference the mpa skb. This ensures the data area
1174 * will remain in memory until the hw acks the tx.
1175 * Function fw4_ack() will deref it.
1178 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1180 __state_set(&ep->com, MPA_REP_SENT);
1181 ep->snd_seq += mpalen;
1182 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1185 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1188 struct cpl_act_establish *req = cplhdr(skb);
1189 unsigned int tid = GET_TID(req);
1190 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1191 struct tid_info *t = dev->rdev.lldi.tids;
1194 ep = lookup_atid(t, atid);
1196 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1197 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1199 mutex_lock(&ep->com.mutex);
1200 dst_confirm(ep->dst);
1202 /* setup the hwtid for this connection */
1204 cxgb4_insert_tid(t, ep, tid);
1207 ep->snd_seq = be32_to_cpu(req->snd_isn);
1208 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1210 set_emss(ep, ntohs(req->tcp_opt));
1212 /* dealloc the atid */
1213 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1214 cxgb4_free_atid(t, atid);
1215 set_bit(ACT_ESTAB, &ep->com.history);
1217 /* start MPA negotiation */
1218 ret = send_flowc(ep);
1221 if (ep->retry_with_mpa_v1)
1222 ret = send_mpa_req(ep, skb, 1);
1224 ret = send_mpa_req(ep, skb, mpa_rev);
1227 mutex_unlock(&ep->com.mutex);
1230 mutex_unlock(&ep->com.mutex);
1231 connect_reply_upcall(ep, -ENOMEM);
1232 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1236 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1238 struct iw_cm_event event;
1240 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1241 memset(&event, 0, sizeof(event));
1242 event.event = IW_CM_EVENT_CLOSE;
1243 event.status = status;
1244 if (ep->com.cm_id) {
1245 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1246 ep, ep->com.cm_id, ep->hwtid);
1247 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1248 deref_cm_id(&ep->com);
1249 set_bit(CLOSE_UPCALL, &ep->com.history);
1253 static void peer_close_upcall(struct c4iw_ep *ep)
1255 struct iw_cm_event event;
1257 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1258 memset(&event, 0, sizeof(event));
1259 event.event = IW_CM_EVENT_DISCONNECT;
1260 if (ep->com.cm_id) {
1261 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1262 ep, ep->com.cm_id, ep->hwtid);
1263 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1264 set_bit(DISCONN_UPCALL, &ep->com.history);
1268 static void peer_abort_upcall(struct c4iw_ep *ep)
1270 struct iw_cm_event event;
1272 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1273 memset(&event, 0, sizeof(event));
1274 event.event = IW_CM_EVENT_CLOSE;
1275 event.status = -ECONNRESET;
1276 if (ep->com.cm_id) {
1277 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1278 ep->com.cm_id, ep->hwtid);
1279 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1280 deref_cm_id(&ep->com);
1281 set_bit(ABORT_UPCALL, &ep->com.history);
1285 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1287 struct iw_cm_event event;
1289 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1290 memset(&event, 0, sizeof(event));
1291 event.event = IW_CM_EVENT_CONNECT_REPLY;
1292 event.status = status;
1293 memcpy(&event.local_addr, &ep->com.local_addr,
1294 sizeof(ep->com.local_addr));
1295 memcpy(&event.remote_addr, &ep->com.remote_addr,
1296 sizeof(ep->com.remote_addr));
1298 if ((status == 0) || (status == -ECONNREFUSED)) {
1299 if (!ep->tried_with_mpa_v1) {
1300 /* this means MPA_v2 is used */
1301 event.ord = ep->ird;
1302 event.ird = ep->ord;
1303 event.private_data_len = ep->plen -
1304 sizeof(struct mpa_v2_conn_params);
1305 event.private_data = ep->mpa_pkt +
1306 sizeof(struct mpa_message) +
1307 sizeof(struct mpa_v2_conn_params);
1309 /* this means MPA_v1 is used */
1310 event.ord = cur_max_read_depth(ep->com.dev);
1311 event.ird = cur_max_read_depth(ep->com.dev);
1312 event.private_data_len = ep->plen;
1313 event.private_data = ep->mpa_pkt +
1314 sizeof(struct mpa_message);
1318 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1320 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1321 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1324 deref_cm_id(&ep->com);
1327 static int connect_request_upcall(struct c4iw_ep *ep)
1329 struct iw_cm_event event;
1332 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1333 memset(&event, 0, sizeof(event));
1334 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1335 memcpy(&event.local_addr, &ep->com.local_addr,
1336 sizeof(ep->com.local_addr));
1337 memcpy(&event.remote_addr, &ep->com.remote_addr,
1338 sizeof(ep->com.remote_addr));
1339 event.provider_data = ep;
1340 if (!ep->tried_with_mpa_v1) {
1341 /* this means MPA_v2 is used */
1342 event.ord = ep->ord;
1343 event.ird = ep->ird;
1344 event.private_data_len = ep->plen -
1345 sizeof(struct mpa_v2_conn_params);
1346 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1347 sizeof(struct mpa_v2_conn_params);
1349 /* this means MPA_v1 is used. Send max supported */
1350 event.ord = cur_max_read_depth(ep->com.dev);
1351 event.ird = cur_max_read_depth(ep->com.dev);
1352 event.private_data_len = ep->plen;
1353 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1355 c4iw_get_ep(&ep->com);
1356 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1359 c4iw_put_ep(&ep->com);
1360 set_bit(CONNREQ_UPCALL, &ep->com.history);
1361 c4iw_put_ep(&ep->parent_ep->com);
1365 static void established_upcall(struct c4iw_ep *ep)
1367 struct iw_cm_event event;
1369 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1370 memset(&event, 0, sizeof(event));
1371 event.event = IW_CM_EVENT_ESTABLISHED;
1372 event.ird = ep->ord;
1373 event.ord = ep->ird;
1374 if (ep->com.cm_id) {
1375 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1376 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1377 set_bit(ESTAB_UPCALL, &ep->com.history);
1381 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1383 struct cpl_rx_data_ack *req;
1384 struct sk_buff *skb;
1385 int wrlen = roundup(sizeof *req, 16);
1387 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1388 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1390 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1395 * If we couldn't specify the entire rcv window at connection setup
1396 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1397 * then add the overage in to the credits returned.
1399 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1400 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1402 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1403 memset(req, 0, wrlen);
1404 INIT_TP_WR(req, ep->hwtid);
1405 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1407 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1409 RX_DACK_MODE_V(dack_mode));
1410 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1411 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1415 #define RELAXED_IRD_NEGOTIATION 1
1418 * process_mpa_reply - process streaming mode MPA reply
1422 * 0 upon success indicating a connect request was delivered to the ULP
1423 * or the mpa request is incomplete but valid so far.
1425 * 1 if a failure requires the caller to close the connection.
1427 * 2 if a failure requires the caller to abort the connection.
1429 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1431 struct mpa_message *mpa;
1432 struct mpa_v2_conn_params *mpa_v2_params;
1434 u16 resp_ird, resp_ord;
1435 u8 rtr_mismatch = 0, insuff_ird = 0;
1436 struct c4iw_qp_attributes attrs;
1437 enum c4iw_qp_attr_mask mask;
1441 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1444 * If we get more than the supported amount of private data
1445 * then we must fail this connection.
1447 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1449 goto err_stop_timer;
1453 * copy the new data into our accumulation buffer.
1455 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1457 ep->mpa_pkt_len += skb->len;
1460 * if we don't even have the mpa message, then bail.
1462 if (ep->mpa_pkt_len < sizeof(*mpa))
1464 mpa = (struct mpa_message *) ep->mpa_pkt;
1466 /* Validate MPA header. */
1467 if (mpa->revision > mpa_rev) {
1468 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1469 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1471 goto err_stop_timer;
1473 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1475 goto err_stop_timer;
1478 plen = ntohs(mpa->private_data_size);
1481 * Fail if there's too much private data.
1483 if (plen > MPA_MAX_PRIVATE_DATA) {
1485 goto err_stop_timer;
1489 * If plen does not account for pkt size
1491 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1493 goto err_stop_timer;
1496 ep->plen = (u8) plen;
1499 * If we don't have all the pdata yet, then bail.
1500 * We'll continue process when more data arrives.
1502 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1505 if (mpa->flags & MPA_REJECT) {
1506 err = -ECONNREFUSED;
1507 goto err_stop_timer;
1511 * Stop mpa timer. If it expired, then
1512 * we ignore the MPA reply. process_timeout()
1513 * will abort the connection.
1515 if (stop_ep_timer(ep))
1519 * If we get here we have accumulated the entire mpa
1520 * start reply message including private data. And
1521 * the MPA header is valid.
1523 __state_set(&ep->com, FPDU_MODE);
1524 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1525 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1526 ep->mpa_attr.version = mpa->revision;
1527 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1529 if (mpa->revision == 2) {
1530 ep->mpa_attr.enhanced_rdma_conn =
1531 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1532 if (ep->mpa_attr.enhanced_rdma_conn) {
1533 mpa_v2_params = (struct mpa_v2_conn_params *)
1534 (ep->mpa_pkt + sizeof(*mpa));
1535 resp_ird = ntohs(mpa_v2_params->ird) &
1536 MPA_V2_IRD_ORD_MASK;
1537 resp_ord = ntohs(mpa_v2_params->ord) &
1538 MPA_V2_IRD_ORD_MASK;
1539 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1540 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1543 * This is a double-check. Ideally, below checks are
1544 * not required since ird/ord stuff has been taken
1545 * care of in c4iw_accept_cr
1547 if (ep->ird < resp_ord) {
1548 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1549 ep->com.dev->rdev.lldi.max_ordird_qp)
1553 } else if (ep->ird > resp_ord) {
1556 if (ep->ord > resp_ird) {
1557 if (RELAXED_IRD_NEGOTIATION)
1568 if (ntohs(mpa_v2_params->ird) &
1569 MPA_V2_PEER2PEER_MODEL) {
1570 if (ntohs(mpa_v2_params->ord) &
1571 MPA_V2_RDMA_WRITE_RTR)
1572 ep->mpa_attr.p2p_type =
1573 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1574 else if (ntohs(mpa_v2_params->ord) &
1575 MPA_V2_RDMA_READ_RTR)
1576 ep->mpa_attr.p2p_type =
1577 FW_RI_INIT_P2PTYPE_READ_REQ;
1580 } else if (mpa->revision == 1)
1582 ep->mpa_attr.p2p_type = p2p_type;
1584 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1585 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1586 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1587 ep->mpa_attr.recv_marker_enabled,
1588 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1589 ep->mpa_attr.p2p_type, p2p_type);
1592 * If responder's RTR does not match with that of initiator, assign
1593 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1594 * generated when moving QP to RTS state.
1595 * A TERM message will be sent after QP has moved to RTS state
1597 if ((ep->mpa_attr.version == 2) && peer2peer &&
1598 (ep->mpa_attr.p2p_type != p2p_type)) {
1599 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1603 attrs.mpa_attr = ep->mpa_attr;
1604 attrs.max_ird = ep->ird;
1605 attrs.max_ord = ep->ord;
1606 attrs.llp_stream_handle = ep;
1607 attrs.next_state = C4IW_QP_STATE_RTS;
1609 mask = C4IW_QP_ATTR_NEXT_STATE |
1610 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1611 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1613 /* bind QP and TID with INIT_WR */
1614 err = c4iw_modify_qp(ep->com.qp->rhp,
1615 ep->com.qp, mask, &attrs, 1);
1620 * If responder's RTR requirement did not match with what initiator
1621 * supports, generate TERM message
1624 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1625 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1626 attrs.ecode = MPA_NOMATCH_RTR;
1627 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1628 attrs.send_term = 1;
1629 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1630 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1637 * Generate TERM if initiator IRD is not sufficient for responder
1638 * provided ORD. Currently, we do the same behaviour even when
1639 * responder provided IRD is also not sufficient as regards to
1643 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1645 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1646 attrs.ecode = MPA_INSUFF_IRD;
1647 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1648 attrs.send_term = 1;
1649 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1650 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1661 connect_reply_upcall(ep, err);
1666 * process_mpa_request - process streaming mode MPA request
1670 * 0 upon success indicating a connect request was delivered to the ULP
1671 * or the mpa request is incomplete but valid so far.
1673 * 1 if a failure requires the caller to close the connection.
1675 * 2 if a failure requires the caller to abort the connection.
1677 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1679 struct mpa_message *mpa;
1680 struct mpa_v2_conn_params *mpa_v2_params;
1683 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1686 * If we get more than the supported amount of private data
1687 * then we must fail this connection.
1689 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1690 goto err_stop_timer;
1692 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1695 * Copy the new data into our accumulation buffer.
1697 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1699 ep->mpa_pkt_len += skb->len;
1702 * If we don't even have the mpa message, then bail.
1703 * We'll continue process when more data arrives.
1705 if (ep->mpa_pkt_len < sizeof(*mpa))
1708 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1709 mpa = (struct mpa_message *) ep->mpa_pkt;
1712 * Validate MPA Header.
1714 if (mpa->revision > mpa_rev) {
1715 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1716 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1717 goto err_stop_timer;
1720 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1721 goto err_stop_timer;
1723 plen = ntohs(mpa->private_data_size);
1726 * Fail if there's too much private data.
1728 if (plen > MPA_MAX_PRIVATE_DATA)
1729 goto err_stop_timer;
1732 * If plen does not account for pkt size
1734 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1735 goto err_stop_timer;
1736 ep->plen = (u8) plen;
1739 * If we don't have all the pdata yet, then bail.
1741 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1745 * If we get here we have accumulated the entire mpa
1746 * start reply message including private data.
1748 ep->mpa_attr.initiator = 0;
1749 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1750 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1751 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1752 ep->mpa_attr.version = mpa->revision;
1753 if (mpa->revision == 1)
1754 ep->tried_with_mpa_v1 = 1;
1755 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1757 if (mpa->revision == 2) {
1758 ep->mpa_attr.enhanced_rdma_conn =
1759 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1760 if (ep->mpa_attr.enhanced_rdma_conn) {
1761 mpa_v2_params = (struct mpa_v2_conn_params *)
1762 (ep->mpa_pkt + sizeof(*mpa));
1763 ep->ird = ntohs(mpa_v2_params->ird) &
1764 MPA_V2_IRD_ORD_MASK;
1765 ep->ird = min_t(u32, ep->ird,
1766 cur_max_read_depth(ep->com.dev));
1767 ep->ord = ntohs(mpa_v2_params->ord) &
1768 MPA_V2_IRD_ORD_MASK;
1769 ep->ord = min_t(u32, ep->ord,
1770 cur_max_read_depth(ep->com.dev));
1771 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1773 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1775 if (ntohs(mpa_v2_params->ord) &
1776 MPA_V2_RDMA_WRITE_RTR)
1777 ep->mpa_attr.p2p_type =
1778 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1779 else if (ntohs(mpa_v2_params->ord) &
1780 MPA_V2_RDMA_READ_RTR)
1781 ep->mpa_attr.p2p_type =
1782 FW_RI_INIT_P2PTYPE_READ_REQ;
1785 } else if (mpa->revision == 1)
1787 ep->mpa_attr.p2p_type = p2p_type;
1789 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1790 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1791 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1792 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1793 ep->mpa_attr.p2p_type);
1795 __state_set(&ep->com, MPA_REQ_RCVD);
1798 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1799 if (ep->parent_ep->com.state != DEAD) {
1800 if (connect_request_upcall(ep))
1801 goto err_unlock_parent;
1803 goto err_unlock_parent;
1805 mutex_unlock(&ep->parent_ep->com.mutex);
1809 mutex_unlock(&ep->parent_ep->com.mutex);
1812 (void)stop_ep_timer(ep);
1817 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1820 struct cpl_rx_data *hdr = cplhdr(skb);
1821 unsigned int dlen = ntohs(hdr->len);
1822 unsigned int tid = GET_TID(hdr);
1823 __u8 status = hdr->status;
1826 ep = get_ep_from_tid(dev, tid);
1829 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1830 skb_pull(skb, sizeof(*hdr));
1831 skb_trim(skb, dlen);
1832 mutex_lock(&ep->com.mutex);
1834 /* update RX credits */
1835 update_rx_credits(ep, dlen);
1837 switch (ep->com.state) {
1839 ep->rcv_seq += dlen;
1840 disconnect = process_mpa_reply(ep, skb);
1843 ep->rcv_seq += dlen;
1844 disconnect = process_mpa_request(ep, skb);
1847 struct c4iw_qp_attributes attrs;
1848 BUG_ON(!ep->com.qp);
1850 pr_err("%s Unexpected streaming data." \
1851 " qpid %u ep %p state %d tid %u status %d\n",
1852 __func__, ep->com.qp->wq.sq.qid, ep,
1853 ep->com.state, ep->hwtid, status);
1854 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1855 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1856 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1863 mutex_unlock(&ep->com.mutex);
1865 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1866 c4iw_put_ep(&ep->com);
1870 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1873 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1875 unsigned int tid = GET_TID(rpl);
1877 ep = get_ep_from_tid(dev, tid);
1879 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1882 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1883 mutex_lock(&ep->com.mutex);
1884 switch (ep->com.state) {
1886 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1887 __state_set(&ep->com, DEAD);
1891 printk(KERN_ERR "%s ep %p state %d\n",
1892 __func__, ep, ep->com.state);
1895 mutex_unlock(&ep->com.mutex);
1898 release_ep_resources(ep);
1899 c4iw_put_ep(&ep->com);
1903 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1905 struct sk_buff *skb;
1906 struct fw_ofld_connection_wr *req;
1907 unsigned int mtu_idx;
1909 struct sockaddr_in *sin;
1912 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1913 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1914 memset(req, 0, sizeof(*req));
1915 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1916 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1917 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1918 ep->com.dev->rdev.lldi.ports[0],
1920 sin = (struct sockaddr_in *)&ep->com.local_addr;
1921 req->le.lport = sin->sin_port;
1922 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1923 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1924 req->le.pport = sin->sin_port;
1925 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1926 req->tcb.t_state_to_astid =
1927 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1928 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1929 req->tcb.cplrxdataack_cplpassacceptrpl =
1930 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1931 req->tcb.tx_max = (__force __be32) jiffies;
1932 req->tcb.rcv_adv = htons(1);
1933 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1934 enable_tcp_timestamps,
1935 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
1936 wscale = compute_wscale(rcv_win);
1939 * Specify the largest window that will fit in opt0. The
1940 * remainder will be specified in the rx_data_ack.
1942 win = ep->rcv_win >> 10;
1943 if (win > RCV_BUFSIZ_M)
1946 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1947 (nocong ? NO_CONG_F : 0) |
1950 WND_SCALE_V(wscale) |
1951 MSS_IDX_V(mtu_idx) |
1952 L2T_IDX_V(ep->l2t->idx) |
1953 TX_CHAN_V(ep->tx_chan) |
1954 SMAC_SEL_V(ep->smac_idx) |
1955 DSCP_V(ep->tos >> 2) |
1956 ULP_MODE_V(ULP_MODE_TCPDDP) |
1958 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1959 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1961 CCTRL_ECN_V(enable_ecn) |
1962 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1963 if (enable_tcp_timestamps)
1964 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1965 if (enable_tcp_sack)
1966 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1967 if (wscale && enable_tcp_window_scaling)
1968 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1969 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1970 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1971 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1972 set_bit(ACT_OFLD_CONN, &ep->com.history);
1973 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1977 * Some of the error codes above implicitly indicate that there is no TID
1978 * allocated with the result of an ACT_OPEN. We use this predicate to make
1981 static inline int act_open_has_tid(int status)
1983 return (status != CPL_ERR_TCAM_PARITY &&
1984 status != CPL_ERR_TCAM_MISS &&
1985 status != CPL_ERR_TCAM_FULL &&
1986 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1987 status != CPL_ERR_CONN_EXIST);
1990 /* Returns whether a CPL status conveys negative advice.
1992 static int is_neg_adv(unsigned int status)
1994 return status == CPL_ERR_RTX_NEG_ADVICE ||
1995 status == CPL_ERR_PERSIST_NEG_ADVICE ||
1996 status == CPL_ERR_KEEPALV_NEG_ADVICE;
1999 static char *neg_adv_str(unsigned int status)
2002 case CPL_ERR_RTX_NEG_ADVICE:
2003 return "Retransmit timeout";
2004 case CPL_ERR_PERSIST_NEG_ADVICE:
2005 return "Persist timeout";
2006 case CPL_ERR_KEEPALV_NEG_ADVICE:
2007 return "Keepalive timeout";
2013 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2015 ep->snd_win = snd_win;
2016 ep->rcv_win = rcv_win;
2017 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2020 #define ACT_OPEN_RETRY_COUNT 2
2022 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2023 struct dst_entry *dst, struct c4iw_dev *cdev,
2024 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2026 struct neighbour *n;
2028 struct net_device *pdev;
2030 n = dst_neigh_lookup(dst, peer_ip);
2036 if (n->dev->flags & IFF_LOOPBACK) {
2038 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2039 else if (IS_ENABLED(CONFIG_IPV6))
2040 for_each_netdev(&init_net, pdev) {
2041 if (ipv6_chk_addr(&init_net,
2042 (struct in6_addr *)peer_ip,
2053 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2054 n, pdev, rt_tos2priority(tos));
2057 ep->mtu = pdev->mtu;
2058 ep->tx_chan = cxgb4_port_chan(pdev);
2059 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2060 cxgb4_port_viid(pdev));
2061 step = cdev->rdev.lldi.ntxq /
2062 cdev->rdev.lldi.nchan;
2063 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2064 step = cdev->rdev.lldi.nrxq /
2065 cdev->rdev.lldi.nchan;
2066 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2067 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2068 cxgb4_port_idx(pdev) * step];
2069 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2072 pdev = get_real_dev(n->dev);
2073 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2077 ep->mtu = dst_mtu(dst);
2078 ep->tx_chan = cxgb4_port_chan(pdev);
2079 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2080 cxgb4_port_viid(pdev));
2081 step = cdev->rdev.lldi.ntxq /
2082 cdev->rdev.lldi.nchan;
2083 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2084 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2085 step = cdev->rdev.lldi.nrxq /
2086 cdev->rdev.lldi.nchan;
2087 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2088 cxgb4_port_idx(pdev) * step];
2089 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2092 ep->retry_with_mpa_v1 = 0;
2093 ep->tried_with_mpa_v1 = 0;
2105 static int c4iw_reconnect(struct c4iw_ep *ep)
2109 struct sockaddr_in *laddr = (struct sockaddr_in *)
2110 &ep->com.cm_id->m_local_addr;
2111 struct sockaddr_in *raddr = (struct sockaddr_in *)
2112 &ep->com.cm_id->m_remote_addr;
2113 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2114 &ep->com.cm_id->m_local_addr;
2115 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2116 &ep->com.cm_id->m_remote_addr;
2120 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2121 init_timer(&ep->timer);
2122 c4iw_init_wr_wait(&ep->com.wr_wait);
2124 /* When MPA revision is different on nodes, the node with MPA_rev=2
2125 * tries to reconnect with MPA_rev 1 for the same EP through
2126 * c4iw_reconnect(), where the same EP is assigned with new tid for
2127 * further connection establishment. As we are using the same EP pointer
2128 * for reconnect, few skbs are used during the previous c4iw_connect(),
2129 * which leaves the EP with inadequate skbs for further
2130 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2131 * skb_list() during peer_abort(). Allocate skbs which is already used.
2133 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2134 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2140 * Allocate an active TID to initiate a TCP connection.
2142 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2143 if (ep->atid == -1) {
2144 pr_err("%s - cannot alloc atid.\n", __func__);
2148 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2151 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2152 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2153 laddr->sin_addr.s_addr,
2154 raddr->sin_addr.s_addr,
2156 raddr->sin_port, ep->com.cm_id->tos);
2158 ra = (__u8 *)&raddr->sin_addr;
2160 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2162 laddr6->sin6_addr.s6_addr,
2163 raddr6->sin6_addr.s6_addr,
2165 raddr6->sin6_port, 0,
2166 raddr6->sin6_scope_id);
2168 ra = (__u8 *)&raddr6->sin6_addr;
2171 pr_err("%s - cannot find route.\n", __func__);
2172 err = -EHOSTUNREACH;
2175 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2176 ep->com.dev->rdev.lldi.adapter_type,
2177 ep->com.cm_id->tos);
2179 pr_err("%s - cannot alloc l2e.\n", __func__);
2183 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2184 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2187 state_set(&ep->com, CONNECTING);
2188 ep->tos = ep->com.cm_id->tos;
2190 /* send connect request to rnic */
2191 err = send_connect(ep);
2195 cxgb4_l2t_release(ep->l2t);
2197 dst_release(ep->dst);
2199 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2200 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2203 * remember to send notification to upper layer.
2204 * We are in here so the upper layer is not aware that this is
2205 * re-connect attempt and so, upper layer is still waiting for
2206 * response of 1st connect request.
2208 connect_reply_upcall(ep, -ECONNRESET);
2210 c4iw_put_ep(&ep->com);
2215 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2218 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2219 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2220 ntohl(rpl->atid_status)));
2221 struct tid_info *t = dev->rdev.lldi.tids;
2222 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2223 struct sockaddr_in *la;
2224 struct sockaddr_in *ra;
2225 struct sockaddr_in6 *la6;
2226 struct sockaddr_in6 *ra6;
2229 ep = lookup_atid(t, atid);
2230 la = (struct sockaddr_in *)&ep->com.local_addr;
2231 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2232 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2233 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2235 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2236 status, status2errno(status));
2238 if (is_neg_adv(status)) {
2239 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2240 __func__, atid, status, neg_adv_str(status));
2241 ep->stats.connect_neg_adv++;
2242 mutex_lock(&dev->rdev.stats.lock);
2243 dev->rdev.stats.neg_adv++;
2244 mutex_unlock(&dev->rdev.stats.lock);
2248 set_bit(ACT_OPEN_RPL, &ep->com.history);
2251 * Log interesting failures.
2254 case CPL_ERR_CONN_RESET:
2255 case CPL_ERR_CONN_TIMEDOUT:
2257 case CPL_ERR_TCAM_FULL:
2258 mutex_lock(&dev->rdev.stats.lock);
2259 dev->rdev.stats.tcam_full++;
2260 mutex_unlock(&dev->rdev.stats.lock);
2261 if (ep->com.local_addr.ss_family == AF_INET &&
2262 dev->rdev.lldi.enable_fw_ofld_conn) {
2263 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2264 ntohl(rpl->atid_status))));
2270 case CPL_ERR_CONN_EXIST:
2271 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2272 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2273 if (ep->com.remote_addr.ss_family == AF_INET6) {
2274 struct sockaddr_in6 *sin6 =
2275 (struct sockaddr_in6 *)
2276 &ep->com.local_addr;
2278 ep->com.dev->rdev.lldi.ports[0],
2280 &sin6->sin6_addr.s6_addr, 1);
2282 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2284 cxgb4_free_atid(t, atid);
2285 dst_release(ep->dst);
2286 cxgb4_l2t_release(ep->l2t);
2292 if (ep->com.local_addr.ss_family == AF_INET) {
2293 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2294 atid, status, status2errno(status),
2295 &la->sin_addr.s_addr, ntohs(la->sin_port),
2296 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2298 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2299 atid, status, status2errno(status),
2300 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2301 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2307 connect_reply_upcall(ep, status2errno(status));
2308 state_set(&ep->com, DEAD);
2310 if (ep->com.remote_addr.ss_family == AF_INET6) {
2311 struct sockaddr_in6 *sin6 =
2312 (struct sockaddr_in6 *)&ep->com.local_addr;
2313 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2314 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2316 if (status && act_open_has_tid(status))
2317 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2319 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2320 cxgb4_free_atid(t, atid);
2321 dst_release(ep->dst);
2322 cxgb4_l2t_release(ep->l2t);
2323 c4iw_put_ep(&ep->com);
2328 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2330 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2331 unsigned int stid = GET_TID(rpl);
2332 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2335 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2338 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2339 rpl->status, status2errno(rpl->status));
2340 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2341 c4iw_put_ep(&ep->com);
2346 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2348 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2349 unsigned int stid = GET_TID(rpl);
2350 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2352 PDBG("%s ep %p\n", __func__, ep);
2353 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2354 c4iw_put_ep(&ep->com);
2358 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2359 struct cpl_pass_accept_req *req)
2361 struct cpl_pass_accept_rpl *rpl;
2362 unsigned int mtu_idx;
2366 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2368 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2370 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2371 BUG_ON(skb_cloned(skb));
2375 if (!is_t4(adapter_type)) {
2376 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2378 INIT_TP_WR(rpl5, ep->hwtid);
2380 skb_trim(skb, sizeof(*rpl));
2381 INIT_TP_WR(rpl, ep->hwtid);
2383 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2386 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2387 enable_tcp_timestamps && req->tcpopt.tstamp,
2388 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2389 wscale = compute_wscale(rcv_win);
2392 * Specify the largest window that will fit in opt0. The
2393 * remainder will be specified in the rx_data_ack.
2395 win = ep->rcv_win >> 10;
2396 if (win > RCV_BUFSIZ_M)
2398 opt0 = (nocong ? NO_CONG_F : 0) |
2401 WND_SCALE_V(wscale) |
2402 MSS_IDX_V(mtu_idx) |
2403 L2T_IDX_V(ep->l2t->idx) |
2404 TX_CHAN_V(ep->tx_chan) |
2405 SMAC_SEL_V(ep->smac_idx) |
2406 DSCP_V(ep->tos >> 2) |
2407 ULP_MODE_V(ULP_MODE_TCPDDP) |
2409 opt2 = RX_CHANNEL_V(0) |
2410 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2412 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2413 opt2 |= TSTAMPS_EN_F;
2414 if (enable_tcp_sack && req->tcpopt.sack)
2416 if (wscale && enable_tcp_window_scaling)
2417 opt2 |= WND_SCALE_EN_F;
2419 const struct tcphdr *tcph;
2420 u32 hlen = ntohl(req->hdr_len);
2422 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2423 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2426 tcph = (const void *)(req + 1) +
2427 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2428 if (tcph->ece && tcph->cwr)
2429 opt2 |= CCTRL_ECN_V(1);
2431 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2432 u32 isn = (prandom_u32() & ~7UL) - 1;
2433 opt2 |= T5_OPT_2_VALID_F;
2434 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2437 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2440 rpl5->iss = cpu_to_be32(isn);
2441 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2444 rpl->opt0 = cpu_to_be64(opt0);
2445 rpl->opt2 = cpu_to_be32(opt2);
2446 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2447 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2449 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2452 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2454 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2455 BUG_ON(skb_cloned(skb));
2456 skb_trim(skb, sizeof(struct cpl_tid_release));
2457 release_tid(&dev->rdev, hwtid, skb);
2461 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2463 struct c4iw_ep *child_ep = NULL, *parent_ep;
2464 struct cpl_pass_accept_req *req = cplhdr(skb);
2465 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2466 struct tid_info *t = dev->rdev.lldi.tids;
2467 unsigned int hwtid = GET_TID(req);
2468 struct dst_entry *dst;
2469 __u8 local_ip[16], peer_ip[16];
2470 __be16 local_port, peer_port;
2471 struct sockaddr_in6 *sin6;
2473 u16 peer_mss = ntohs(req->tcpopt.mss);
2475 unsigned short hdrs;
2476 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2478 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2480 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2484 if (state_read(&parent_ep->com) != LISTEN) {
2485 PDBG("%s - listening ep not in LISTEN\n", __func__);
2489 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2490 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2492 /* Find output route */
2494 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2495 , __func__, parent_ep, hwtid,
2496 local_ip, peer_ip, ntohs(local_port),
2497 ntohs(peer_port), peer_mss);
2498 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2499 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2500 local_port, peer_port, tos);
2502 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2503 , __func__, parent_ep, hwtid,
2504 local_ip, peer_ip, ntohs(local_port),
2505 ntohs(peer_port), peer_mss);
2506 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2507 local_ip, peer_ip, local_port, peer_port,
2508 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2509 ((struct sockaddr_in6 *)
2510 &parent_ep->com.local_addr)->sin6_scope_id);
2513 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2518 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2520 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2526 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2527 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2529 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2536 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2537 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2538 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2539 child_ep->mtu = peer_mss + hdrs;
2541 skb_queue_head_init(&child_ep->com.ep_skb_list);
2542 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2545 state_set(&child_ep->com, CONNECTING);
2546 child_ep->com.dev = dev;
2547 child_ep->com.cm_id = NULL;
2550 struct sockaddr_in *sin = (struct sockaddr_in *)
2551 &child_ep->com.local_addr;
2553 sin->sin_family = PF_INET;
2554 sin->sin_port = local_port;
2555 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2557 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2558 sin->sin_family = PF_INET;
2559 sin->sin_port = ((struct sockaddr_in *)
2560 &parent_ep->com.local_addr)->sin_port;
2561 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2563 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2564 sin->sin_family = PF_INET;
2565 sin->sin_port = peer_port;
2566 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2568 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2569 sin6->sin6_family = PF_INET6;
2570 sin6->sin6_port = local_port;
2571 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2573 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2574 sin6->sin6_family = PF_INET6;
2575 sin6->sin6_port = ((struct sockaddr_in6 *)
2576 &parent_ep->com.local_addr)->sin6_port;
2577 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2579 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2580 sin6->sin6_family = PF_INET6;
2581 sin6->sin6_port = peer_port;
2582 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2585 c4iw_get_ep(&parent_ep->com);
2586 child_ep->parent_ep = parent_ep;
2587 child_ep->tos = tos;
2588 child_ep->dst = dst;
2589 child_ep->hwtid = hwtid;
2591 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2592 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2594 init_timer(&child_ep->timer);
2595 cxgb4_insert_tid(t, child_ep, hwtid);
2596 insert_ep_tid(child_ep);
2597 if (accept_cr(child_ep, skb, req)) {
2598 c4iw_put_ep(&parent_ep->com);
2599 release_ep_resources(child_ep);
2601 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2604 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2605 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2606 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2610 c4iw_put_ep(&child_ep->com);
2612 reject_cr(dev, hwtid, skb);
2614 c4iw_put_ep(&parent_ep->com);
2619 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2622 struct cpl_pass_establish *req = cplhdr(skb);
2623 unsigned int tid = GET_TID(req);
2626 ep = get_ep_from_tid(dev, tid);
2627 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2628 ep->snd_seq = be32_to_cpu(req->snd_isn);
2629 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2631 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2632 ntohs(req->tcp_opt));
2634 set_emss(ep, ntohs(req->tcp_opt));
2636 dst_confirm(ep->dst);
2637 mutex_lock(&ep->com.mutex);
2638 ep->com.state = MPA_REQ_WAIT;
2640 set_bit(PASS_ESTAB, &ep->com.history);
2641 ret = send_flowc(ep);
2642 mutex_unlock(&ep->com.mutex);
2644 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2645 c4iw_put_ep(&ep->com);
2650 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2652 struct cpl_peer_close *hdr = cplhdr(skb);
2654 struct c4iw_qp_attributes attrs;
2657 unsigned int tid = GET_TID(hdr);
2660 ep = get_ep_from_tid(dev, tid);
2664 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2665 dst_confirm(ep->dst);
2667 set_bit(PEER_CLOSE, &ep->com.history);
2668 mutex_lock(&ep->com.mutex);
2669 switch (ep->com.state) {
2671 __state_set(&ep->com, CLOSING);
2674 __state_set(&ep->com, CLOSING);
2675 connect_reply_upcall(ep, -ECONNRESET);
2680 * We're gonna mark this puppy DEAD, but keep
2681 * the reference on it until the ULP accepts or
2682 * rejects the CR. Also wake up anyone waiting
2683 * in rdma connection migration (see c4iw_accept_cr()).
2685 __state_set(&ep->com, CLOSING);
2686 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2687 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2690 __state_set(&ep->com, CLOSING);
2691 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2692 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2696 __state_set(&ep->com, CLOSING);
2697 attrs.next_state = C4IW_QP_STATE_CLOSING;
2698 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2699 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2700 if (ret != -ECONNRESET) {
2701 peer_close_upcall(ep);
2709 __state_set(&ep->com, MORIBUND);
2713 (void)stop_ep_timer(ep);
2714 if (ep->com.cm_id && ep->com.qp) {
2715 attrs.next_state = C4IW_QP_STATE_IDLE;
2716 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2717 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2719 close_complete_upcall(ep, 0);
2720 __state_set(&ep->com, DEAD);
2730 mutex_unlock(&ep->com.mutex);
2732 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2734 release_ep_resources(ep);
2735 c4iw_put_ep(&ep->com);
2739 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2741 struct cpl_abort_req_rss *req = cplhdr(skb);
2743 struct cpl_abort_rpl *rpl;
2744 struct sk_buff *rpl_skb;
2745 struct c4iw_qp_attributes attrs;
2748 unsigned int tid = GET_TID(req);
2750 ep = get_ep_from_tid(dev, tid);
2754 if (is_neg_adv(req->status)) {
2755 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2756 __func__, ep->hwtid, req->status,
2757 neg_adv_str(req->status));
2758 ep->stats.abort_neg_adv++;
2759 mutex_lock(&dev->rdev.stats.lock);
2760 dev->rdev.stats.neg_adv++;
2761 mutex_unlock(&dev->rdev.stats.lock);
2764 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2766 set_bit(PEER_ABORT, &ep->com.history);
2769 * Wake up any threads in rdma_init() or rdma_fini().
2770 * However, this is not needed if com state is just
2773 if (ep->com.state != MPA_REQ_SENT)
2774 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2776 mutex_lock(&ep->com.mutex);
2777 switch (ep->com.state) {
2779 c4iw_put_ep(&ep->parent_ep->com);
2782 (void)stop_ep_timer(ep);
2785 (void)stop_ep_timer(ep);
2786 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2787 connect_reply_upcall(ep, -ECONNRESET);
2790 * we just don't send notification upwards because we
2791 * want to retry with mpa_v1 without upper layers even
2794 * do some housekeeping so as to re-initiate the
2797 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2799 ep->retry_with_mpa_v1 = 1;
2811 if (ep->com.cm_id && ep->com.qp) {
2812 attrs.next_state = C4IW_QP_STATE_ERROR;
2813 ret = c4iw_modify_qp(ep->com.qp->rhp,
2814 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2818 "%s - qp <- error failed!\n",
2821 peer_abort_upcall(ep);
2826 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2827 mutex_unlock(&ep->com.mutex);
2833 dst_confirm(ep->dst);
2834 if (ep->com.state != ABORTING) {
2835 __state_set(&ep->com, DEAD);
2836 /* we don't release if we want to retry with mpa_v1 */
2837 if (!ep->retry_with_mpa_v1)
2840 mutex_unlock(&ep->com.mutex);
2842 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2843 if (WARN_ON(!rpl_skb)) {
2847 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2848 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2849 INIT_TP_WR(rpl, ep->hwtid);
2850 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2851 rpl->cmd = CPL_ABORT_NO_RST;
2852 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2855 release_ep_resources(ep);
2856 else if (ep->retry_with_mpa_v1) {
2857 if (ep->com.remote_addr.ss_family == AF_INET6) {
2858 struct sockaddr_in6 *sin6 =
2859 (struct sockaddr_in6 *)
2860 &ep->com.local_addr;
2862 ep->com.dev->rdev.lldi.ports[0],
2863 (const u32 *)&sin6->sin6_addr.s6_addr,
2866 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2867 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2868 dst_release(ep->dst);
2869 cxgb4_l2t_release(ep->l2t);
2874 c4iw_put_ep(&ep->com);
2875 /* Dereferencing ep, referenced in peer_abort_intr() */
2876 c4iw_put_ep(&ep->com);
2880 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2883 struct c4iw_qp_attributes attrs;
2884 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2886 unsigned int tid = GET_TID(rpl);
2888 ep = get_ep_from_tid(dev, tid);
2892 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2895 /* The cm_id may be null if we failed to connect */
2896 mutex_lock(&ep->com.mutex);
2897 set_bit(CLOSE_CON_RPL, &ep->com.history);
2898 switch (ep->com.state) {
2900 __state_set(&ep->com, MORIBUND);
2903 (void)stop_ep_timer(ep);
2904 if ((ep->com.cm_id) && (ep->com.qp)) {
2905 attrs.next_state = C4IW_QP_STATE_IDLE;
2906 c4iw_modify_qp(ep->com.qp->rhp,
2908 C4IW_QP_ATTR_NEXT_STATE,
2911 close_complete_upcall(ep, 0);
2912 __state_set(&ep->com, DEAD);
2922 mutex_unlock(&ep->com.mutex);
2924 release_ep_resources(ep);
2925 c4iw_put_ep(&ep->com);
2929 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2931 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2932 unsigned int tid = GET_TID(rpl);
2934 struct c4iw_qp_attributes attrs;
2936 ep = get_ep_from_tid(dev, tid);
2939 if (ep && ep->com.qp) {
2940 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2941 ep->com.qp->wq.sq.qid);
2942 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2943 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2944 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2946 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2947 c4iw_put_ep(&ep->com);
2953 * Upcall from the adapter indicating data has been transmitted.
2954 * For us its just the single MPA request or reply. We can now free
2955 * the skb holding the mpa message.
2957 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2960 struct cpl_fw4_ack *hdr = cplhdr(skb);
2961 u8 credits = hdr->credits;
2962 unsigned int tid = GET_TID(hdr);
2965 ep = get_ep_from_tid(dev, tid);
2968 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2970 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2971 __func__, ep, ep->hwtid, state_read(&ep->com));
2975 dst_confirm(ep->dst);
2977 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2978 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2979 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2980 mutex_lock(&ep->com.mutex);
2981 kfree_skb(ep->mpa_skb);
2983 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2985 mutex_unlock(&ep->com.mutex);
2988 c4iw_put_ep(&ep->com);
2992 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2995 struct c4iw_ep *ep = to_ep(cm_id);
2997 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2999 mutex_lock(&ep->com.mutex);
3000 if (ep->com.state != MPA_REQ_RCVD) {
3001 mutex_unlock(&ep->com.mutex);
3002 c4iw_put_ep(&ep->com);
3005 set_bit(ULP_REJECT, &ep->com.history);
3009 abort = send_mpa_reject(ep, pdata, pdata_len);
3010 mutex_unlock(&ep->com.mutex);
3013 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3014 c4iw_put_ep(&ep->com);
3018 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3021 struct c4iw_qp_attributes attrs;
3022 enum c4iw_qp_attr_mask mask;
3023 struct c4iw_ep *ep = to_ep(cm_id);
3024 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3025 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3028 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3030 mutex_lock(&ep->com.mutex);
3031 if (ep->com.state != MPA_REQ_RCVD) {
3038 set_bit(ULP_ACCEPT, &ep->com.history);
3039 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3040 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3045 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3046 if (conn_param->ord > ep->ird) {
3047 if (RELAXED_IRD_NEGOTIATION) {
3048 conn_param->ord = ep->ird;
3050 ep->ird = conn_param->ird;
3051 ep->ord = conn_param->ord;
3052 send_mpa_reject(ep, conn_param->private_data,
3053 conn_param->private_data_len);
3058 if (conn_param->ird < ep->ord) {
3059 if (RELAXED_IRD_NEGOTIATION &&
3060 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3061 conn_param->ird = ep->ord;
3068 ep->ird = conn_param->ird;
3069 ep->ord = conn_param->ord;
3071 if (ep->mpa_attr.version == 1) {
3072 if (peer2peer && ep->ird == 0)
3076 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3077 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3081 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3083 ep->com.cm_id = cm_id;
3084 ref_cm_id(&ep->com);
3088 /* bind QP to EP and move to RTS */
3089 attrs.mpa_attr = ep->mpa_attr;
3090 attrs.max_ird = ep->ird;
3091 attrs.max_ord = ep->ord;
3092 attrs.llp_stream_handle = ep;
3093 attrs.next_state = C4IW_QP_STATE_RTS;
3095 /* bind QP and TID with INIT_WR */
3096 mask = C4IW_QP_ATTR_NEXT_STATE |
3097 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3098 C4IW_QP_ATTR_MPA_ATTR |
3099 C4IW_QP_ATTR_MAX_IRD |
3100 C4IW_QP_ATTR_MAX_ORD;
3102 err = c4iw_modify_qp(ep->com.qp->rhp,
3103 ep->com.qp, mask, &attrs, 1);
3105 goto err_deref_cm_id;
3107 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3108 err = send_mpa_reply(ep, conn_param->private_data,
3109 conn_param->private_data_len);
3111 goto err_deref_cm_id;
3113 __state_set(&ep->com, FPDU_MODE);
3114 established_upcall(ep);
3115 mutex_unlock(&ep->com.mutex);
3116 c4iw_put_ep(&ep->com);
3119 deref_cm_id(&ep->com);
3123 mutex_unlock(&ep->com.mutex);
3125 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3126 c4iw_put_ep(&ep->com);
3130 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3132 struct in_device *ind;
3134 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3135 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3137 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3139 return -EADDRNOTAVAIL;
3140 for_primary_ifa(ind) {
3141 laddr->sin_addr.s_addr = ifa->ifa_address;
3142 raddr->sin_addr.s_addr = ifa->ifa_address;
3148 return found ? 0 : -EADDRNOTAVAIL;
3151 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3152 unsigned char banned_flags)
3154 struct inet6_dev *idev;
3155 int err = -EADDRNOTAVAIL;
3158 idev = __in6_dev_get(dev);
3160 struct inet6_ifaddr *ifp;
3162 read_lock_bh(&idev->lock);
3163 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3164 if (ifp->scope == IFA_LINK &&
3165 !(ifp->flags & banned_flags)) {
3166 memcpy(addr, &ifp->addr, 16);
3171 read_unlock_bh(&idev->lock);
3177 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3179 struct in6_addr uninitialized_var(addr);
3180 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3181 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3183 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3184 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3185 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3188 return -EADDRNOTAVAIL;
3191 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3193 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3196 struct sockaddr_in *laddr;
3197 struct sockaddr_in *raddr;
3198 struct sockaddr_in6 *laddr6;
3199 struct sockaddr_in6 *raddr6;
3203 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3204 (conn_param->ird > cur_max_read_depth(dev))) {
3208 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3210 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3215 skb_queue_head_init(&ep->com.ep_skb_list);
3216 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3221 init_timer(&ep->timer);
3222 ep->plen = conn_param->private_data_len;
3224 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3225 conn_param->private_data, ep->plen);
3226 ep->ird = conn_param->ird;
3227 ep->ord = conn_param->ord;
3229 if (peer2peer && ep->ord == 0)
3232 ep->com.cm_id = cm_id;
3233 ref_cm_id(&ep->com);
3235 ep->com.qp = get_qhp(dev, conn_param->qpn);
3237 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3242 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3246 * Allocate an active TID to initiate a TCP connection.
3248 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3249 if (ep->atid == -1) {
3250 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3254 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3256 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3257 sizeof(ep->com.local_addr));
3258 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3259 sizeof(ep->com.remote_addr));
3261 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3262 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3263 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3264 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3266 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3268 ra = (__u8 *)&raddr->sin_addr;
3271 * Handle loopback requests to INADDR_ANY.
3273 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3274 err = pick_local_ipaddrs(dev, cm_id);
3280 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3281 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3282 ra, ntohs(raddr->sin_port));
3283 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3284 laddr->sin_addr.s_addr,
3285 raddr->sin_addr.s_addr,
3287 raddr->sin_port, cm_id->tos);
3290 ra = (__u8 *)&raddr6->sin6_addr;
3293 * Handle loopback requests to INADDR_ANY.
3295 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3296 err = pick_local_ip6addrs(dev, cm_id);
3302 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3303 __func__, laddr6->sin6_addr.s6_addr,
3304 ntohs(laddr6->sin6_port),
3305 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3306 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3307 laddr6->sin6_addr.s6_addr,
3308 raddr6->sin6_addr.s6_addr,
3310 raddr6->sin6_port, 0,
3311 raddr6->sin6_scope_id);
3314 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3315 err = -EHOSTUNREACH;
3319 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3320 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3322 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3326 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3327 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3330 state_set(&ep->com, CONNECTING);
3331 ep->tos = cm_id->tos;
3333 /* send connect request to rnic */
3334 err = send_connect(ep);
3338 cxgb4_l2t_release(ep->l2t);
3340 dst_release(ep->dst);
3342 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3343 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3345 skb_queue_purge(&ep->com.ep_skb_list);
3346 deref_cm_id(&ep->com);
3348 c4iw_put_ep(&ep->com);
3353 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3356 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3357 &ep->com.local_addr;
3359 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3360 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3361 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3365 c4iw_init_wr_wait(&ep->com.wr_wait);
3366 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3367 ep->stid, &sin6->sin6_addr,
3369 ep->com.dev->rdev.lldi.rxq_ids[0]);
3371 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3375 err = net_xmit_errno(err);
3377 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3378 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3379 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3381 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3386 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3389 struct sockaddr_in *sin = (struct sockaddr_in *)
3390 &ep->com.local_addr;
3392 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3394 err = cxgb4_create_server_filter(
3395 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3396 sin->sin_addr.s_addr, sin->sin_port, 0,
3397 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3398 if (err == -EBUSY) {
3399 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3403 set_current_state(TASK_UNINTERRUPTIBLE);
3404 schedule_timeout(usecs_to_jiffies(100));
3406 } while (err == -EBUSY);
3408 c4iw_init_wr_wait(&ep->com.wr_wait);
3409 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3410 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3411 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3413 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3417 err = net_xmit_errno(err);
3420 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3422 &sin->sin_addr, ntohs(sin->sin_port));
3426 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3429 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3430 struct c4iw_listen_ep *ep;
3434 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3436 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3440 skb_queue_head_init(&ep->com.ep_skb_list);
3441 PDBG("%s ep %p\n", __func__, ep);
3442 ep->com.cm_id = cm_id;
3443 ref_cm_id(&ep->com);
3445 ep->backlog = backlog;
3446 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3447 sizeof(ep->com.local_addr));
3450 * Allocate a server TID.
3452 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3453 ep->com.local_addr.ss_family == AF_INET)
3454 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3455 cm_id->m_local_addr.ss_family, ep);
3457 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3458 cm_id->m_local_addr.ss_family, ep);
3460 if (ep->stid == -1) {
3461 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3465 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3467 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3468 sizeof(ep->com.local_addr));
3470 state_set(&ep->com, LISTEN);
3471 if (ep->com.local_addr.ss_family == AF_INET)
3472 err = create_server4(dev, ep);
3474 err = create_server6(dev, ep);
3476 cm_id->provider_data = ep;
3480 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3481 ep->com.local_addr.ss_family);
3483 deref_cm_id(&ep->com);
3484 c4iw_put_ep(&ep->com);
3490 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3493 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3495 PDBG("%s ep %p\n", __func__, ep);
3498 state_set(&ep->com, DEAD);
3499 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3500 ep->com.local_addr.ss_family == AF_INET) {
3501 err = cxgb4_remove_server_filter(
3502 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3503 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3505 struct sockaddr_in6 *sin6;
3506 c4iw_init_wr_wait(&ep->com.wr_wait);
3507 err = cxgb4_remove_server(
3508 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3509 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3512 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3514 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3515 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3516 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3518 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3519 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3520 ep->com.local_addr.ss_family);
3522 deref_cm_id(&ep->com);
3523 c4iw_put_ep(&ep->com);
3527 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3532 struct c4iw_rdev *rdev;
3534 mutex_lock(&ep->com.mutex);
3536 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3537 states[ep->com.state], abrupt);
3540 * Ref the ep here in case we have fatal errors causing the
3541 * ep to be released and freed.
3543 c4iw_get_ep(&ep->com);
3545 rdev = &ep->com.dev->rdev;
3546 if (c4iw_fatal_error(rdev)) {
3548 close_complete_upcall(ep, -EIO);
3549 ep->com.state = DEAD;
3551 switch (ep->com.state) {
3560 ep->com.state = ABORTING;
3562 ep->com.state = CLOSING;
3565 * if we close before we see the fw4_ack() then we fix
3566 * up the timer state since we're reusing it.
3569 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3570 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3575 set_bit(CLOSE_SENT, &ep->com.flags);
3578 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3581 (void)stop_ep_timer(ep);
3582 ep->com.state = ABORTING;
3584 ep->com.state = MORIBUND;
3590 PDBG("%s ignoring disconnect ep %p state %u\n",
3591 __func__, ep, ep->com.state);
3600 set_bit(EP_DISC_ABORT, &ep->com.history);
3601 close_complete_upcall(ep, -ECONNRESET);
3602 ret = send_abort(ep);
3604 set_bit(EP_DISC_CLOSE, &ep->com.history);
3605 ret = send_halfclose(ep);
3608 set_bit(EP_DISC_FAIL, &ep->com.history);
3611 close_complete_upcall(ep, -EIO);
3614 struct c4iw_qp_attributes attrs;
3616 attrs.next_state = C4IW_QP_STATE_ERROR;
3617 ret = c4iw_modify_qp(ep->com.qp->rhp,
3619 C4IW_QP_ATTR_NEXT_STATE,
3623 "%s - qp <- error failed!\n",
3629 mutex_unlock(&ep->com.mutex);
3630 c4iw_put_ep(&ep->com);
3632 release_ep_resources(ep);
3636 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3637 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3640 int atid = be32_to_cpu(req->tid);
3642 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3643 (__force u32) req->tid);
3647 switch (req->retval) {
3649 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3650 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3651 send_fw_act_open_req(ep, atid);
3655 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3656 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3657 send_fw_act_open_req(ep, atid);
3662 pr_info("%s unexpected ofld conn wr retval %d\n",
3663 __func__, req->retval);
3666 pr_err("active ofld_connect_wr failure %d atid %d\n",
3668 mutex_lock(&dev->rdev.stats.lock);
3669 dev->rdev.stats.act_ofld_conn_fails++;
3670 mutex_unlock(&dev->rdev.stats.lock);
3671 connect_reply_upcall(ep, status2errno(req->retval));
3672 state_set(&ep->com, DEAD);
3673 if (ep->com.remote_addr.ss_family == AF_INET6) {
3674 struct sockaddr_in6 *sin6 =
3675 (struct sockaddr_in6 *)&ep->com.local_addr;
3676 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3677 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3679 remove_handle(dev, &dev->atid_idr, atid);
3680 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3681 dst_release(ep->dst);
3682 cxgb4_l2t_release(ep->l2t);
3683 c4iw_put_ep(&ep->com);
3686 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3687 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3689 struct sk_buff *rpl_skb;
3690 struct cpl_pass_accept_req *cpl;
3693 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3696 PDBG("%s passive open failure %d\n", __func__, req->retval);
3697 mutex_lock(&dev->rdev.stats.lock);
3698 dev->rdev.stats.pas_ofld_conn_fails++;
3699 mutex_unlock(&dev->rdev.stats.lock);
3702 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3703 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3704 (__force u32) htonl(
3705 (__force u32) req->tid)));
3706 ret = pass_accept_req(dev, rpl_skb);
3713 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3715 struct cpl_fw6_msg *rpl = cplhdr(skb);
3716 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3718 switch (rpl->type) {
3720 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3722 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3723 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3724 switch (req->t_state) {
3726 active_ofld_conn_reply(dev, skb, req);
3729 passive_ofld_conn_reply(dev, skb, req);
3732 pr_err("%s unexpected ofld conn wr state %d\n",
3733 __func__, req->t_state);
3741 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3744 __be16 hdr_len, vlantag, len;
3746 int tcp_hdr_len, ip_hdr_len;
3748 struct cpl_rx_pkt *cpl = cplhdr(skb);
3749 struct cpl_pass_accept_req *req;
3750 struct tcp_options_received tmp_opt;
3751 struct c4iw_dev *dev;
3752 enum chip_type type;
3754 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3755 /* Store values from cpl_rx_pkt in temporary location. */
3756 vlantag = cpl->vlan;
3758 l2info = cpl->l2info;
3759 hdr_len = cpl->hdr_len;
3762 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3765 * We need to parse the TCP options from SYN packet.
3766 * to generate cpl_pass_accept_req.
3768 memset(&tmp_opt, 0, sizeof(tmp_opt));
3769 tcp_clear_options(&tmp_opt);
3770 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3772 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3773 memset(req, 0, sizeof(*req));
3774 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3775 SYN_MAC_IDX_V(RX_MACIDX_G(
3776 be32_to_cpu(l2info))) |
3778 type = dev->rdev.lldi.adapter_type;
3779 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3780 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3782 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3783 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3784 eth_hdr_len = is_t4(type) ?
3785 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3786 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3787 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3788 IP_HDR_LEN_V(ip_hdr_len) |
3789 ETH_HDR_LEN_V(eth_hdr_len));
3790 } else { /* T6 and later */
3791 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3792 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3793 T6_IP_HDR_LEN_V(ip_hdr_len) |
3794 T6_ETH_HDR_LEN_V(eth_hdr_len));
3796 req->vlan = vlantag;
3798 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3799 PASS_OPEN_TOS_V(tos));
3800 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3801 if (tmp_opt.wscale_ok)
3802 req->tcpopt.wsf = tmp_opt.snd_wscale;
3803 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3804 if (tmp_opt.sack_ok)
3805 req->tcpopt.sack = 1;
3806 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3810 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3811 __be32 laddr, __be16 lport,
3812 __be32 raddr, __be16 rport,
3813 u32 rcv_isn, u32 filter, u16 window,
3814 u32 rss_qid, u8 port_id)
3816 struct sk_buff *req_skb;
3817 struct fw_ofld_connection_wr *req;
3818 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3821 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3822 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3823 memset(req, 0, sizeof(*req));
3824 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3825 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3826 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3827 req->le.filter = (__force __be32) filter;
3828 req->le.lport = lport;
3829 req->le.pport = rport;
3830 req->le.u.ipv4.lip = laddr;
3831 req->le.u.ipv4.pip = raddr;
3832 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3833 req->tcb.rcv_adv = htons(window);
3834 req->tcb.t_state_to_astid =
3835 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3836 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3837 FW_OFLD_CONNECTION_WR_ASTID_V(
3838 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3841 * We store the qid in opt2 which will be used by the firmware
3842 * to send us the wr response.
3844 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3847 * We initialize the MSS index in TCB to 0xF.
3848 * So that when driver sends cpl_pass_accept_rpl
3849 * TCB picks up the correct value. If this was 0
3850 * TP will ignore any value > 0 for MSS index.
3852 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3853 req->cookie = (uintptr_t)skb;
3855 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3856 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3858 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3866 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3867 * messages when a filter is being used instead of server to
3868 * redirect a syn packet. When packets hit filter they are redirected
3869 * to the offload queue and driver tries to establish the connection
3870 * using firmware work request.
3872 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3875 unsigned int filter;
3876 struct ethhdr *eh = NULL;
3877 struct vlan_ethhdr *vlan_eh = NULL;
3879 struct tcphdr *tcph;
3880 struct rss_header *rss = (void *)skb->data;
3881 struct cpl_rx_pkt *cpl = (void *)skb->data;
3882 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3883 struct l2t_entry *e;
3884 struct dst_entry *dst;
3885 struct c4iw_ep *lep = NULL;
3887 struct port_info *pi;
3888 struct net_device *pdev;
3889 u16 rss_qid, eth_hdr_len;
3892 struct neighbour *neigh;
3894 /* Drop all non-SYN packets */
3895 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3899 * Drop all packets which did not hit the filter.
3900 * Unlikely to happen.
3902 if (!(rss->filter_hit && rss->filter_tid))
3906 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3908 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3910 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3912 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3916 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3918 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3921 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3924 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3927 pr_err("T%d Chip is not supported\n",
3928 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3932 if (eth_hdr_len == ETH_HLEN) {
3933 eh = (struct ethhdr *)(req + 1);
3934 iph = (struct iphdr *)(eh + 1);
3936 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3937 iph = (struct iphdr *)(vlan_eh + 1);
3938 skb->vlan_tci = ntohs(cpl->vlan);
3941 if (iph->version != 0x4)
3944 tcph = (struct tcphdr *)(iph + 1);
3945 skb_set_network_header(skb, (void *)iph - (void *)rss);
3946 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3949 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3950 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3951 ntohs(tcph->source), iph->tos);
3953 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3954 iph->daddr, iph->saddr, tcph->dest,
3955 tcph->source, iph->tos);
3957 pr_err("%s - failed to find dst entry!\n",
3961 neigh = dst_neigh_lookup_skb(dst, skb);
3964 pr_err("%s - failed to allocate neigh!\n",
3969 if (neigh->dev->flags & IFF_LOOPBACK) {
3970 pdev = ip_dev_find(&init_net, iph->daddr);
3971 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3973 pi = (struct port_info *)netdev_priv(pdev);
3974 tx_chan = cxgb4_port_chan(pdev);
3977 pdev = get_real_dev(neigh->dev);
3978 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3980 pi = (struct port_info *)netdev_priv(pdev);
3981 tx_chan = cxgb4_port_chan(pdev);
3983 neigh_release(neigh);
3985 pr_err("%s - failed to allocate l2t entry!\n",
3990 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3991 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3992 window = (__force u16) htons((__force u16)tcph->window);
3994 /* Calcuate filter portion for LE region. */
3995 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3996 dev->rdev.lldi.ports[0],
4000 * Synthesize the cpl_pass_accept_req. We have everything except the
4001 * TID. Once firmware sends a reply with TID we update the TID field
4002 * in cpl and pass it through the regular cpl_pass_accept_req path.
4004 build_cpl_pass_accept_req(skb, stid, iph->tos);
4005 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4006 tcph->source, ntohl(tcph->seq), filter, window,
4007 rss_qid, pi->port_id);
4008 cxgb4_l2t_release(e);
4013 c4iw_put_ep(&lep->com);
4018 * These are the real handlers that are called from a
4021 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4022 [CPL_ACT_ESTABLISH] = act_establish,
4023 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4024 [CPL_RX_DATA] = rx_data,
4025 [CPL_ABORT_RPL_RSS] = abort_rpl,
4026 [CPL_ABORT_RPL] = abort_rpl,
4027 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4028 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4029 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4030 [CPL_PASS_ESTABLISH] = pass_establish,
4031 [CPL_PEER_CLOSE] = peer_close,
4032 [CPL_ABORT_REQ_RSS] = peer_abort,
4033 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4034 [CPL_RDMA_TERMINATE] = terminate,
4035 [CPL_FW4_ACK] = fw4_ack,
4036 [CPL_FW6_MSG] = deferred_fw6_msg,
4037 [CPL_RX_PKT] = rx_pkt,
4038 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4039 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4042 static void process_timeout(struct c4iw_ep *ep)
4044 struct c4iw_qp_attributes attrs;
4047 mutex_lock(&ep->com.mutex);
4048 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4050 set_bit(TIMEDOUT, &ep->com.history);
4051 switch (ep->com.state) {
4053 connect_reply_upcall(ep, -ETIMEDOUT);
4062 if (ep->com.cm_id && ep->com.qp) {
4063 attrs.next_state = C4IW_QP_STATE_ERROR;
4064 c4iw_modify_qp(ep->com.qp->rhp,
4065 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4068 close_complete_upcall(ep, -ETIMEDOUT);
4074 * These states are expected if the ep timed out at the same
4075 * time as another thread was calling stop_ep_timer().
4076 * So we silently do nothing for these states.
4081 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4082 __func__, ep, ep->hwtid, ep->com.state);
4085 mutex_unlock(&ep->com.mutex);
4087 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4088 c4iw_put_ep(&ep->com);
4091 static void process_timedout_eps(void)
4095 spin_lock_irq(&timeout_lock);
4096 while (!list_empty(&timeout_list)) {
4097 struct list_head *tmp;
4099 tmp = timeout_list.next;
4103 spin_unlock_irq(&timeout_lock);
4104 ep = list_entry(tmp, struct c4iw_ep, entry);
4105 process_timeout(ep);
4106 spin_lock_irq(&timeout_lock);
4108 spin_unlock_irq(&timeout_lock);
4111 static void process_work(struct work_struct *work)
4113 struct sk_buff *skb = NULL;
4114 struct c4iw_dev *dev;
4115 struct cpl_act_establish *rpl;
4116 unsigned int opcode;
4119 process_timedout_eps();
4120 while ((skb = skb_dequeue(&rxq))) {
4122 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4123 opcode = rpl->ot.opcode;
4125 BUG_ON(!work_handlers[opcode]);
4126 ret = work_handlers[opcode](dev, skb);
4129 process_timedout_eps();
4133 static DECLARE_WORK(skb_work, process_work);
4135 static void ep_timeout(unsigned long arg)
4137 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4140 spin_lock(&timeout_lock);
4141 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4143 * Only insert if it is not already on the list.
4145 if (!ep->entry.next) {
4146 list_add_tail(&ep->entry, &timeout_list);
4150 spin_unlock(&timeout_lock);
4152 queue_work(workq, &skb_work);
4156 * All the CM events are handled on a work queue to have a safe context.
4158 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4162 * Save dev in the skb->cb area.
4164 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4167 * Queue the skb and schedule the worker thread.
4169 skb_queue_tail(&rxq, skb);
4170 queue_work(workq, &skb_work);
4174 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4176 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4178 if (rpl->status != CPL_ERR_NONE) {
4179 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4180 "for tid %u\n", rpl->status, GET_TID(rpl));
4186 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4188 struct cpl_fw6_msg *rpl = cplhdr(skb);
4189 struct c4iw_wr_wait *wr_waitp;
4192 PDBG("%s type %u\n", __func__, rpl->type);
4194 switch (rpl->type) {
4195 case FW6_TYPE_WR_RPL:
4196 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4197 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4198 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4200 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4204 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4208 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4216 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4218 struct cpl_abort_req_rss *req = cplhdr(skb);
4220 unsigned int tid = GET_TID(req);
4222 ep = get_ep_from_tid(dev, tid);
4223 /* This EP will be dereferenced in peer_abort() */
4225 printk(KERN_WARNING MOD
4226 "Abort on non-existent endpoint, tid %d\n", tid);
4230 if (is_neg_adv(req->status)) {
4231 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4232 __func__, ep->hwtid, req->status,
4233 neg_adv_str(req->status));
4236 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4239 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4246 * Most upcalls from the T4 Core go to sched() to
4247 * schedule the processing on a work queue.
4249 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4250 [CPL_ACT_ESTABLISH] = sched,
4251 [CPL_ACT_OPEN_RPL] = sched,
4252 [CPL_RX_DATA] = sched,
4253 [CPL_ABORT_RPL_RSS] = sched,
4254 [CPL_ABORT_RPL] = sched,
4255 [CPL_PASS_OPEN_RPL] = sched,
4256 [CPL_CLOSE_LISTSRV_RPL] = sched,
4257 [CPL_PASS_ACCEPT_REQ] = sched,
4258 [CPL_PASS_ESTABLISH] = sched,
4259 [CPL_PEER_CLOSE] = sched,
4260 [CPL_CLOSE_CON_RPL] = sched,
4261 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4262 [CPL_RDMA_TERMINATE] = sched,
4263 [CPL_FW4_ACK] = sched,
4264 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4265 [CPL_FW6_MSG] = fw6_msg,
4266 [CPL_RX_PKT] = sched
4269 int __init c4iw_cm_init(void)
4271 spin_lock_init(&timeout_lock);
4272 skb_queue_head_init(&rxq);
4274 workq = create_singlethread_workqueue("iw_cxgb4");
4281 void c4iw_cm_term(void)
4283 WARN_ON(!list_empty(&timeout_list));
4284 flush_workqueue(workq);
4285 destroy_workqueue(workq);